Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/inode.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/inode.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
18 *
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
20 */
21
22#include <linux/fs.h>
23#include <linux/mount.h>
24#include <linux/time.h>
25#include <linux/highuid.h>
26#include <linux/pagemap.h>
27#include <linux/dax.h>
28#include <linux/quotaops.h>
29#include <linux/string.h>
30#include <linux/buffer_head.h>
31#include <linux/writeback.h>
32#include <linux/pagevec.h>
33#include <linux/mpage.h>
34#include <linux/namei.h>
35#include <linux/uio.h>
36#include <linux/bio.h>
37#include <linux/workqueue.h>
38#include <linux/kernel.h>
39#include <linux/printk.h>
40#include <linux/slab.h>
41#include <linux/bitops.h>
42#include <linux/iomap.h>
43#include <linux/iversion.h>
44
45#include "ext4_jbd2.h"
46#include "xattr.h"
47#include "acl.h"
48#include "truncate.h"
49
50#include <trace/events/ext4.h>
51
52static void ext4_journalled_zero_new_buffers(handle_t *handle,
53 struct inode *inode,
54 struct folio *folio,
55 unsigned from, unsigned to);
56
57static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
58 struct ext4_inode_info *ei)
59{
60 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
61 __u32 csum;
62 __u16 dummy_csum = 0;
63 int offset = offsetof(struct ext4_inode, i_checksum_lo);
64 unsigned int csum_size = sizeof(dummy_csum);
65
66 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
67 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
68 offset += csum_size;
69 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
70 EXT4_GOOD_OLD_INODE_SIZE - offset);
71
72 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
73 offset = offsetof(struct ext4_inode, i_checksum_hi);
74 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
75 EXT4_GOOD_OLD_INODE_SIZE,
76 offset - EXT4_GOOD_OLD_INODE_SIZE);
77 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
78 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
79 csum_size);
80 offset += csum_size;
81 }
82 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
83 EXT4_INODE_SIZE(inode->i_sb) - offset);
84 }
85
86 return csum;
87}
88
89static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
90 struct ext4_inode_info *ei)
91{
92 __u32 provided, calculated;
93
94 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
95 cpu_to_le32(EXT4_OS_LINUX) ||
96 !ext4_has_metadata_csum(inode->i_sb))
97 return 1;
98
99 provided = le16_to_cpu(raw->i_checksum_lo);
100 calculated = ext4_inode_csum(inode, raw, ei);
101 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
102 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
103 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
104 else
105 calculated &= 0xFFFF;
106
107 return provided == calculated;
108}
109
110void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
111 struct ext4_inode_info *ei)
112{
113 __u32 csum;
114
115 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
116 cpu_to_le32(EXT4_OS_LINUX) ||
117 !ext4_has_metadata_csum(inode->i_sb))
118 return;
119
120 csum = ext4_inode_csum(inode, raw, ei);
121 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
122 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
123 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
124 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
125}
126
127static inline int ext4_begin_ordered_truncate(struct inode *inode,
128 loff_t new_size)
129{
130 trace_ext4_begin_ordered_truncate(inode, new_size);
131 /*
132 * If jinode is zero, then we never opened the file for
133 * writing, so there's no need to call
134 * jbd2_journal_begin_ordered_truncate() since there's no
135 * outstanding writes we need to flush.
136 */
137 if (!EXT4_I(inode)->jinode)
138 return 0;
139 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
140 EXT4_I(inode)->jinode,
141 new_size);
142}
143
144static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
145 int pextents);
146
147/*
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 */
151int ext4_inode_is_fast_symlink(struct inode *inode)
152{
153 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
156
157 if (ext4_has_inline_data(inode))
158 return 0;
159
160 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 }
162 return S_ISLNK(inode->i_mode) && inode->i_size &&
163 (inode->i_size < EXT4_N_BLOCKS * 4);
164}
165
166/*
167 * Called at the last iput() if i_nlink is zero.
168 */
169void ext4_evict_inode(struct inode *inode)
170{
171 handle_t *handle;
172 int err;
173 /*
174 * Credits for final inode cleanup and freeing:
175 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
176 * (xattr block freeing), bitmap, group descriptor (inode freeing)
177 */
178 int extra_credits = 6;
179 struct ext4_xattr_inode_array *ea_inode_array = NULL;
180 bool freeze_protected = false;
181
182 trace_ext4_evict_inode(inode);
183
184 if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
185 ext4_evict_ea_inode(inode);
186 if (inode->i_nlink) {
187 truncate_inode_pages_final(&inode->i_data);
188
189 goto no_delete;
190 }
191
192 if (is_bad_inode(inode))
193 goto no_delete;
194 dquot_initialize(inode);
195
196 if (ext4_should_order_data(inode))
197 ext4_begin_ordered_truncate(inode, 0);
198 truncate_inode_pages_final(&inode->i_data);
199
200 /*
201 * For inodes with journalled data, transaction commit could have
202 * dirtied the inode. And for inodes with dioread_nolock, unwritten
203 * extents converting worker could merge extents and also have dirtied
204 * the inode. Flush worker is ignoring it because of I_FREEING flag but
205 * we still need to remove the inode from the writeback lists.
206 */
207 if (!list_empty_careful(&inode->i_io_list))
208 inode_io_list_del(inode);
209
210 /*
211 * Protect us against freezing - iput() caller didn't have to have any
212 * protection against it. When we are in a running transaction though,
213 * we are already protected against freezing and we cannot grab further
214 * protection due to lock ordering constraints.
215 */
216 if (!ext4_journal_current_handle()) {
217 sb_start_intwrite(inode->i_sb);
218 freeze_protected = true;
219 }
220
221 if (!IS_NOQUOTA(inode))
222 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
223
224 /*
225 * Block bitmap, group descriptor, and inode are accounted in both
226 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
227 */
228 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
229 ext4_blocks_for_truncate(inode) + extra_credits - 3);
230 if (IS_ERR(handle)) {
231 ext4_std_error(inode->i_sb, PTR_ERR(handle));
232 /*
233 * If we're going to skip the normal cleanup, we still need to
234 * make sure that the in-core orphan linked list is properly
235 * cleaned up.
236 */
237 ext4_orphan_del(NULL, inode);
238 if (freeze_protected)
239 sb_end_intwrite(inode->i_sb);
240 goto no_delete;
241 }
242
243 if (IS_SYNC(inode))
244 ext4_handle_sync(handle);
245
246 /*
247 * Set inode->i_size to 0 before calling ext4_truncate(). We need
248 * special handling of symlinks here because i_size is used to
249 * determine whether ext4_inode_info->i_data contains symlink data or
250 * block mappings. Setting i_size to 0 will remove its fast symlink
251 * status. Erase i_data so that it becomes a valid empty block map.
252 */
253 if (ext4_inode_is_fast_symlink(inode))
254 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
255 inode->i_size = 0;
256 err = ext4_mark_inode_dirty(handle, inode);
257 if (err) {
258 ext4_warning(inode->i_sb,
259 "couldn't mark inode dirty (err %d)", err);
260 goto stop_handle;
261 }
262 if (inode->i_blocks) {
263 err = ext4_truncate(inode);
264 if (err) {
265 ext4_error_err(inode->i_sb, -err,
266 "couldn't truncate inode %lu (err %d)",
267 inode->i_ino, err);
268 goto stop_handle;
269 }
270 }
271
272 /* Remove xattr references. */
273 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
274 extra_credits);
275 if (err) {
276 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
277stop_handle:
278 ext4_journal_stop(handle);
279 ext4_orphan_del(NULL, inode);
280 if (freeze_protected)
281 sb_end_intwrite(inode->i_sb);
282 ext4_xattr_inode_array_free(ea_inode_array);
283 goto no_delete;
284 }
285
286 /*
287 * Kill off the orphan record which ext4_truncate created.
288 * AKPM: I think this can be inside the above `if'.
289 * Note that ext4_orphan_del() has to be able to cope with the
290 * deletion of a non-existent orphan - this is because we don't
291 * know if ext4_truncate() actually created an orphan record.
292 * (Well, we could do this if we need to, but heck - it works)
293 */
294 ext4_orphan_del(handle, inode);
295 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
296
297 /*
298 * One subtle ordering requirement: if anything has gone wrong
299 * (transaction abort, IO errors, whatever), then we can still
300 * do these next steps (the fs will already have been marked as
301 * having errors), but we can't free the inode if the mark_dirty
302 * fails.
303 */
304 if (ext4_mark_inode_dirty(handle, inode))
305 /* If that failed, just do the required in-core inode clear. */
306 ext4_clear_inode(inode);
307 else
308 ext4_free_inode(handle, inode);
309 ext4_journal_stop(handle);
310 if (freeze_protected)
311 sb_end_intwrite(inode->i_sb);
312 ext4_xattr_inode_array_free(ea_inode_array);
313 return;
314no_delete:
315 /*
316 * Check out some where else accidentally dirty the evicting inode,
317 * which may probably cause inode use-after-free issues later.
318 */
319 WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
320
321 if (!list_empty(&EXT4_I(inode)->i_fc_list))
322 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
323 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
324}
325
326#ifdef CONFIG_QUOTA
327qsize_t *ext4_get_reserved_space(struct inode *inode)
328{
329 return &EXT4_I(inode)->i_reserved_quota;
330}
331#endif
332
333/*
334 * Called with i_data_sem down, which is important since we can call
335 * ext4_discard_preallocations() from here.
336 */
337void ext4_da_update_reserve_space(struct inode *inode,
338 int used, int quota_claim)
339{
340 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
341 struct ext4_inode_info *ei = EXT4_I(inode);
342
343 spin_lock(&ei->i_block_reservation_lock);
344 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
345 if (unlikely(used > ei->i_reserved_data_blocks)) {
346 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
347 "with only %d reserved data blocks",
348 __func__, inode->i_ino, used,
349 ei->i_reserved_data_blocks);
350 WARN_ON(1);
351 used = ei->i_reserved_data_blocks;
352 }
353
354 /* Update per-inode reservations */
355 ei->i_reserved_data_blocks -= used;
356 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
357
358 spin_unlock(&ei->i_block_reservation_lock);
359
360 /* Update quota subsystem for data blocks */
361 if (quota_claim)
362 dquot_claim_block(inode, EXT4_C2B(sbi, used));
363 else {
364 /*
365 * We did fallocate with an offset that is already delayed
366 * allocated. So on delayed allocated writeback we should
367 * not re-claim the quota for fallocated blocks.
368 */
369 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
370 }
371
372 /*
373 * If we have done all the pending block allocations and if
374 * there aren't any writers on the inode, we can discard the
375 * inode's preallocations.
376 */
377 if ((ei->i_reserved_data_blocks == 0) &&
378 !inode_is_open_for_write(inode))
379 ext4_discard_preallocations(inode);
380}
381
382static int __check_block_validity(struct inode *inode, const char *func,
383 unsigned int line,
384 struct ext4_map_blocks *map)
385{
386 if (ext4_has_feature_journal(inode->i_sb) &&
387 (inode->i_ino ==
388 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
389 return 0;
390 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
391 ext4_error_inode(inode, func, line, map->m_pblk,
392 "lblock %lu mapped to illegal pblock %llu "
393 "(length %d)", (unsigned long) map->m_lblk,
394 map->m_pblk, map->m_len);
395 return -EFSCORRUPTED;
396 }
397 return 0;
398}
399
400int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
401 ext4_lblk_t len)
402{
403 int ret;
404
405 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
406 return fscrypt_zeroout_range(inode, lblk, pblk, len);
407
408 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
409 if (ret > 0)
410 ret = 0;
411
412 return ret;
413}
414
415#define check_block_validity(inode, map) \
416 __check_block_validity((inode), __func__, __LINE__, (map))
417
418#ifdef ES_AGGRESSIVE_TEST
419static void ext4_map_blocks_es_recheck(handle_t *handle,
420 struct inode *inode,
421 struct ext4_map_blocks *es_map,
422 struct ext4_map_blocks *map,
423 int flags)
424{
425 int retval;
426
427 map->m_flags = 0;
428 /*
429 * There is a race window that the result is not the same.
430 * e.g. xfstests #223 when dioread_nolock enables. The reason
431 * is that we lookup a block mapping in extent status tree with
432 * out taking i_data_sem. So at the time the unwritten extent
433 * could be converted.
434 */
435 down_read(&EXT4_I(inode)->i_data_sem);
436 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
437 retval = ext4_ext_map_blocks(handle, inode, map, 0);
438 } else {
439 retval = ext4_ind_map_blocks(handle, inode, map, 0);
440 }
441 up_read((&EXT4_I(inode)->i_data_sem));
442
443 /*
444 * We don't check m_len because extent will be collpased in status
445 * tree. So the m_len might not equal.
446 */
447 if (es_map->m_lblk != map->m_lblk ||
448 es_map->m_flags != map->m_flags ||
449 es_map->m_pblk != map->m_pblk) {
450 printk("ES cache assertion failed for inode: %lu "
451 "es_cached ex [%d/%d/%llu/%x] != "
452 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
453 inode->i_ino, es_map->m_lblk, es_map->m_len,
454 es_map->m_pblk, es_map->m_flags, map->m_lblk,
455 map->m_len, map->m_pblk, map->m_flags,
456 retval, flags);
457 }
458}
459#endif /* ES_AGGRESSIVE_TEST */
460
461static int ext4_map_query_blocks(handle_t *handle, struct inode *inode,
462 struct ext4_map_blocks *map)
463{
464 unsigned int status;
465 int retval;
466
467 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
468 retval = ext4_ext_map_blocks(handle, inode, map, 0);
469 else
470 retval = ext4_ind_map_blocks(handle, inode, map, 0);
471
472 if (retval <= 0)
473 return retval;
474
475 if (unlikely(retval != map->m_len)) {
476 ext4_warning(inode->i_sb,
477 "ES len assertion failed for inode "
478 "%lu: retval %d != map->m_len %d",
479 inode->i_ino, retval, map->m_len);
480 WARN_ON(1);
481 }
482
483 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
484 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
485 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
486 map->m_pblk, status, false);
487 return retval;
488}
489
490static int ext4_map_create_blocks(handle_t *handle, struct inode *inode,
491 struct ext4_map_blocks *map, int flags)
492{
493 struct extent_status es;
494 unsigned int status;
495 int err, retval = 0;
496
497 /*
498 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE
499 * indicates that the blocks and quotas has already been
500 * checked when the data was copied into the page cache.
501 */
502 if (map->m_flags & EXT4_MAP_DELAYED)
503 flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
504
505 /*
506 * Here we clear m_flags because after allocating an new extent,
507 * it will be set again.
508 */
509 map->m_flags &= ~EXT4_MAP_FLAGS;
510
511 /*
512 * We need to check for EXT4 here because migrate could have
513 * changed the inode type in between.
514 */
515 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
516 retval = ext4_ext_map_blocks(handle, inode, map, flags);
517 } else {
518 retval = ext4_ind_map_blocks(handle, inode, map, flags);
519
520 /*
521 * We allocated new blocks which will result in i_data's
522 * format changing. Force the migrate to fail by clearing
523 * migrate flags.
524 */
525 if (retval > 0 && map->m_flags & EXT4_MAP_NEW)
526 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
527 }
528 if (retval <= 0)
529 return retval;
530
531 if (unlikely(retval != map->m_len)) {
532 ext4_warning(inode->i_sb,
533 "ES len assertion failed for inode %lu: "
534 "retval %d != map->m_len %d",
535 inode->i_ino, retval, map->m_len);
536 WARN_ON(1);
537 }
538
539 /*
540 * We have to zeroout blocks before inserting them into extent
541 * status tree. Otherwise someone could look them up there and
542 * use them before they are really zeroed. We also have to
543 * unmap metadata before zeroing as otherwise writeback can
544 * overwrite zeros with stale data from block device.
545 */
546 if (flags & EXT4_GET_BLOCKS_ZERO &&
547 map->m_flags & EXT4_MAP_MAPPED && map->m_flags & EXT4_MAP_NEW) {
548 err = ext4_issue_zeroout(inode, map->m_lblk, map->m_pblk,
549 map->m_len);
550 if (err)
551 return err;
552 }
553
554 /*
555 * If the extent has been zeroed out, we don't need to update
556 * extent status tree.
557 */
558 if (flags & EXT4_GET_BLOCKS_PRE_IO &&
559 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
560 if (ext4_es_is_written(&es))
561 return retval;
562 }
563
564 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
565 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
566 ext4_es_insert_extent(inode, map->m_lblk, map->m_len, map->m_pblk,
567 status, flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE);
568
569 return retval;
570}
571
572/*
573 * The ext4_map_blocks() function tries to look up the requested blocks,
574 * and returns if the blocks are already mapped.
575 *
576 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
577 * and store the allocated blocks in the result buffer head and mark it
578 * mapped.
579 *
580 * If file type is extents based, it will call ext4_ext_map_blocks(),
581 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
582 * based files
583 *
584 * On success, it returns the number of blocks being mapped or allocated.
585 * If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are
586 * pre-allocated and unwritten, the resulting @map is marked as unwritten.
587 * If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped.
588 *
589 * It returns 0 if plain look up failed (blocks have not been allocated), in
590 * that case, @map is returned as unmapped but we still do fill map->m_len to
591 * indicate the length of a hole starting at map->m_lblk.
592 *
593 * It returns the error in case of allocation failure.
594 */
595int ext4_map_blocks(handle_t *handle, struct inode *inode,
596 struct ext4_map_blocks *map, int flags)
597{
598 struct extent_status es;
599 int retval;
600 int ret = 0;
601#ifdef ES_AGGRESSIVE_TEST
602 struct ext4_map_blocks orig_map;
603
604 memcpy(&orig_map, map, sizeof(*map));
605#endif
606
607 map->m_flags = 0;
608 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
609 flags, map->m_len, (unsigned long) map->m_lblk);
610
611 /*
612 * ext4_map_blocks returns an int, and m_len is an unsigned int
613 */
614 if (unlikely(map->m_len > INT_MAX))
615 map->m_len = INT_MAX;
616
617 /* We can handle the block number less than EXT_MAX_BLOCKS */
618 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
619 return -EFSCORRUPTED;
620
621 /* Lookup extent status tree firstly */
622 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
623 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
624 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
625 map->m_pblk = ext4_es_pblock(&es) +
626 map->m_lblk - es.es_lblk;
627 map->m_flags |= ext4_es_is_written(&es) ?
628 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
629 retval = es.es_len - (map->m_lblk - es.es_lblk);
630 if (retval > map->m_len)
631 retval = map->m_len;
632 map->m_len = retval;
633 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
634 map->m_pblk = 0;
635 map->m_flags |= ext4_es_is_delayed(&es) ?
636 EXT4_MAP_DELAYED : 0;
637 retval = es.es_len - (map->m_lblk - es.es_lblk);
638 if (retval > map->m_len)
639 retval = map->m_len;
640 map->m_len = retval;
641 retval = 0;
642 } else {
643 BUG();
644 }
645
646 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
647 return retval;
648#ifdef ES_AGGRESSIVE_TEST
649 ext4_map_blocks_es_recheck(handle, inode, map,
650 &orig_map, flags);
651#endif
652 goto found;
653 }
654 /*
655 * In the query cache no-wait mode, nothing we can do more if we
656 * cannot find extent in the cache.
657 */
658 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
659 return 0;
660
661 /*
662 * Try to see if we can get the block without requesting a new
663 * file system block.
664 */
665 down_read(&EXT4_I(inode)->i_data_sem);
666 retval = ext4_map_query_blocks(handle, inode, map);
667 up_read((&EXT4_I(inode)->i_data_sem));
668
669found:
670 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
671 ret = check_block_validity(inode, map);
672 if (ret != 0)
673 return ret;
674 }
675
676 /* If it is only a block(s) look up */
677 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
678 return retval;
679
680 /*
681 * Returns if the blocks have already allocated
682 *
683 * Note that if blocks have been preallocated
684 * ext4_ext_map_blocks() returns with buffer head unmapped
685 */
686 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
687 /*
688 * If we need to convert extent to unwritten
689 * we continue and do the actual work in
690 * ext4_ext_map_blocks()
691 */
692 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
693 return retval;
694
695 /*
696 * New blocks allocate and/or writing to unwritten extent
697 * will possibly result in updating i_data, so we take
698 * the write lock of i_data_sem, and call get_block()
699 * with create == 1 flag.
700 */
701 down_write(&EXT4_I(inode)->i_data_sem);
702 retval = ext4_map_create_blocks(handle, inode, map, flags);
703 up_write((&EXT4_I(inode)->i_data_sem));
704 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
705 ret = check_block_validity(inode, map);
706 if (ret != 0)
707 return ret;
708
709 /*
710 * Inodes with freshly allocated blocks where contents will be
711 * visible after transaction commit must be on transaction's
712 * ordered data list.
713 */
714 if (map->m_flags & EXT4_MAP_NEW &&
715 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
716 !(flags & EXT4_GET_BLOCKS_ZERO) &&
717 !ext4_is_quota_file(inode) &&
718 ext4_should_order_data(inode)) {
719 loff_t start_byte =
720 (loff_t)map->m_lblk << inode->i_blkbits;
721 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
722
723 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
724 ret = ext4_jbd2_inode_add_wait(handle, inode,
725 start_byte, length);
726 else
727 ret = ext4_jbd2_inode_add_write(handle, inode,
728 start_byte, length);
729 if (ret)
730 return ret;
731 }
732 }
733 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
734 map->m_flags & EXT4_MAP_MAPPED))
735 ext4_fc_track_range(handle, inode, map->m_lblk,
736 map->m_lblk + map->m_len - 1);
737 if (retval < 0)
738 ext_debug(inode, "failed with err %d\n", retval);
739 return retval;
740}
741
742/*
743 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
744 * we have to be careful as someone else may be manipulating b_state as well.
745 */
746static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
747{
748 unsigned long old_state;
749 unsigned long new_state;
750
751 flags &= EXT4_MAP_FLAGS;
752
753 /* Dummy buffer_head? Set non-atomically. */
754 if (!bh->b_page) {
755 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
756 return;
757 }
758 /*
759 * Someone else may be modifying b_state. Be careful! This is ugly but
760 * once we get rid of using bh as a container for mapping information
761 * to pass to / from get_block functions, this can go away.
762 */
763 old_state = READ_ONCE(bh->b_state);
764 do {
765 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
766 } while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
767}
768
769static int _ext4_get_block(struct inode *inode, sector_t iblock,
770 struct buffer_head *bh, int flags)
771{
772 struct ext4_map_blocks map;
773 int ret = 0;
774
775 if (ext4_has_inline_data(inode))
776 return -ERANGE;
777
778 map.m_lblk = iblock;
779 map.m_len = bh->b_size >> inode->i_blkbits;
780
781 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
782 flags);
783 if (ret > 0) {
784 map_bh(bh, inode->i_sb, map.m_pblk);
785 ext4_update_bh_state(bh, map.m_flags);
786 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
787 ret = 0;
788 } else if (ret == 0) {
789 /* hole case, need to fill in bh->b_size */
790 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
791 }
792 return ret;
793}
794
795int ext4_get_block(struct inode *inode, sector_t iblock,
796 struct buffer_head *bh, int create)
797{
798 return _ext4_get_block(inode, iblock, bh,
799 create ? EXT4_GET_BLOCKS_CREATE : 0);
800}
801
802/*
803 * Get block function used when preparing for buffered write if we require
804 * creating an unwritten extent if blocks haven't been allocated. The extent
805 * will be converted to written after the IO is complete.
806 */
807int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
808 struct buffer_head *bh_result, int create)
809{
810 int ret = 0;
811
812 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
813 inode->i_ino, create);
814 ret = _ext4_get_block(inode, iblock, bh_result,
815 EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
816
817 /*
818 * If the buffer is marked unwritten, mark it as new to make sure it is
819 * zeroed out correctly in case of partial writes. Otherwise, there is
820 * a chance of stale data getting exposed.
821 */
822 if (ret == 0 && buffer_unwritten(bh_result))
823 set_buffer_new(bh_result);
824
825 return ret;
826}
827
828/* Maximum number of blocks we map for direct IO at once. */
829#define DIO_MAX_BLOCKS 4096
830
831/*
832 * `handle' can be NULL if create is zero
833 */
834struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
835 ext4_lblk_t block, int map_flags)
836{
837 struct ext4_map_blocks map;
838 struct buffer_head *bh;
839 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
840 bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
841 int err;
842
843 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
844 || handle != NULL || create == 0);
845 ASSERT(create == 0 || !nowait);
846
847 map.m_lblk = block;
848 map.m_len = 1;
849 err = ext4_map_blocks(handle, inode, &map, map_flags);
850
851 if (err == 0)
852 return create ? ERR_PTR(-ENOSPC) : NULL;
853 if (err < 0)
854 return ERR_PTR(err);
855
856 if (nowait)
857 return sb_find_get_block(inode->i_sb, map.m_pblk);
858
859 /*
860 * Since bh could introduce extra ref count such as referred by
861 * journal_head etc. Try to avoid using __GFP_MOVABLE here
862 * as it may fail the migration when journal_head remains.
863 */
864 bh = getblk_unmovable(inode->i_sb->s_bdev, map.m_pblk,
865 inode->i_sb->s_blocksize);
866
867 if (unlikely(!bh))
868 return ERR_PTR(-ENOMEM);
869 if (map.m_flags & EXT4_MAP_NEW) {
870 ASSERT(create != 0);
871 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
872 || (handle != NULL));
873
874 /*
875 * Now that we do not always journal data, we should
876 * keep in mind whether this should always journal the
877 * new buffer as metadata. For now, regular file
878 * writes use ext4_get_block instead, so it's not a
879 * problem.
880 */
881 lock_buffer(bh);
882 BUFFER_TRACE(bh, "call get_create_access");
883 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
884 EXT4_JTR_NONE);
885 if (unlikely(err)) {
886 unlock_buffer(bh);
887 goto errout;
888 }
889 if (!buffer_uptodate(bh)) {
890 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
891 set_buffer_uptodate(bh);
892 }
893 unlock_buffer(bh);
894 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
895 err = ext4_handle_dirty_metadata(handle, inode, bh);
896 if (unlikely(err))
897 goto errout;
898 } else
899 BUFFER_TRACE(bh, "not a new buffer");
900 return bh;
901errout:
902 brelse(bh);
903 return ERR_PTR(err);
904}
905
906struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
907 ext4_lblk_t block, int map_flags)
908{
909 struct buffer_head *bh;
910 int ret;
911
912 bh = ext4_getblk(handle, inode, block, map_flags);
913 if (IS_ERR(bh))
914 return bh;
915 if (!bh || ext4_buffer_uptodate(bh))
916 return bh;
917
918 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
919 if (ret) {
920 put_bh(bh);
921 return ERR_PTR(ret);
922 }
923 return bh;
924}
925
926/* Read a contiguous batch of blocks. */
927int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
928 bool wait, struct buffer_head **bhs)
929{
930 int i, err;
931
932 for (i = 0; i < bh_count; i++) {
933 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
934 if (IS_ERR(bhs[i])) {
935 err = PTR_ERR(bhs[i]);
936 bh_count = i;
937 goto out_brelse;
938 }
939 }
940
941 for (i = 0; i < bh_count; i++)
942 /* Note that NULL bhs[i] is valid because of holes. */
943 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
944 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
945
946 if (!wait)
947 return 0;
948
949 for (i = 0; i < bh_count; i++)
950 if (bhs[i])
951 wait_on_buffer(bhs[i]);
952
953 for (i = 0; i < bh_count; i++) {
954 if (bhs[i] && !buffer_uptodate(bhs[i])) {
955 err = -EIO;
956 goto out_brelse;
957 }
958 }
959 return 0;
960
961out_brelse:
962 for (i = 0; i < bh_count; i++) {
963 brelse(bhs[i]);
964 bhs[i] = NULL;
965 }
966 return err;
967}
968
969int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
970 struct buffer_head *head,
971 unsigned from,
972 unsigned to,
973 int *partial,
974 int (*fn)(handle_t *handle, struct inode *inode,
975 struct buffer_head *bh))
976{
977 struct buffer_head *bh;
978 unsigned block_start, block_end;
979 unsigned blocksize = head->b_size;
980 int err, ret = 0;
981 struct buffer_head *next;
982
983 for (bh = head, block_start = 0;
984 ret == 0 && (bh != head || !block_start);
985 block_start = block_end, bh = next) {
986 next = bh->b_this_page;
987 block_end = block_start + blocksize;
988 if (block_end <= from || block_start >= to) {
989 if (partial && !buffer_uptodate(bh))
990 *partial = 1;
991 continue;
992 }
993 err = (*fn)(handle, inode, bh);
994 if (!ret)
995 ret = err;
996 }
997 return ret;
998}
999
1000/*
1001 * Helper for handling dirtying of journalled data. We also mark the folio as
1002 * dirty so that writeback code knows about this page (and inode) contains
1003 * dirty data. ext4_writepages() then commits appropriate transaction to
1004 * make data stable.
1005 */
1006static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
1007{
1008 folio_mark_dirty(bh->b_folio);
1009 return ext4_handle_dirty_metadata(handle, NULL, bh);
1010}
1011
1012int do_journal_get_write_access(handle_t *handle, struct inode *inode,
1013 struct buffer_head *bh)
1014{
1015 if (!buffer_mapped(bh) || buffer_freed(bh))
1016 return 0;
1017 BUFFER_TRACE(bh, "get write access");
1018 return ext4_journal_get_write_access(handle, inode->i_sb, bh,
1019 EXT4_JTR_NONE);
1020}
1021
1022int ext4_block_write_begin(handle_t *handle, struct folio *folio,
1023 loff_t pos, unsigned len,
1024 get_block_t *get_block)
1025{
1026 unsigned from = pos & (PAGE_SIZE - 1);
1027 unsigned to = from + len;
1028 struct inode *inode = folio->mapping->host;
1029 unsigned block_start, block_end;
1030 sector_t block;
1031 int err = 0;
1032 unsigned blocksize = inode->i_sb->s_blocksize;
1033 unsigned bbits;
1034 struct buffer_head *bh, *head, *wait[2];
1035 int nr_wait = 0;
1036 int i;
1037 bool should_journal_data = ext4_should_journal_data(inode);
1038
1039 BUG_ON(!folio_test_locked(folio));
1040 BUG_ON(from > PAGE_SIZE);
1041 BUG_ON(to > PAGE_SIZE);
1042 BUG_ON(from > to);
1043
1044 head = folio_buffers(folio);
1045 if (!head)
1046 head = create_empty_buffers(folio, blocksize, 0);
1047 bbits = ilog2(blocksize);
1048 block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1049
1050 for (bh = head, block_start = 0; bh != head || !block_start;
1051 block++, block_start = block_end, bh = bh->b_this_page) {
1052 block_end = block_start + blocksize;
1053 if (block_end <= from || block_start >= to) {
1054 if (folio_test_uptodate(folio)) {
1055 set_buffer_uptodate(bh);
1056 }
1057 continue;
1058 }
1059 if (buffer_new(bh))
1060 clear_buffer_new(bh);
1061 if (!buffer_mapped(bh)) {
1062 WARN_ON(bh->b_size != blocksize);
1063 err = get_block(inode, block, bh, 1);
1064 if (err)
1065 break;
1066 if (buffer_new(bh)) {
1067 /*
1068 * We may be zeroing partial buffers or all new
1069 * buffers in case of failure. Prepare JBD2 for
1070 * that.
1071 */
1072 if (should_journal_data)
1073 do_journal_get_write_access(handle,
1074 inode, bh);
1075 if (folio_test_uptodate(folio)) {
1076 /*
1077 * Unlike __block_write_begin() we leave
1078 * dirtying of new uptodate buffers to
1079 * ->write_end() time or
1080 * folio_zero_new_buffers().
1081 */
1082 set_buffer_uptodate(bh);
1083 continue;
1084 }
1085 if (block_end > to || block_start < from)
1086 folio_zero_segments(folio, to,
1087 block_end,
1088 block_start, from);
1089 continue;
1090 }
1091 }
1092 if (folio_test_uptodate(folio)) {
1093 set_buffer_uptodate(bh);
1094 continue;
1095 }
1096 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1097 !buffer_unwritten(bh) &&
1098 (block_start < from || block_end > to)) {
1099 ext4_read_bh_lock(bh, 0, false);
1100 wait[nr_wait++] = bh;
1101 }
1102 }
1103 /*
1104 * If we issued read requests, let them complete.
1105 */
1106 for (i = 0; i < nr_wait; i++) {
1107 wait_on_buffer(wait[i]);
1108 if (!buffer_uptodate(wait[i]))
1109 err = -EIO;
1110 }
1111 if (unlikely(err)) {
1112 if (should_journal_data)
1113 ext4_journalled_zero_new_buffers(handle, inode, folio,
1114 from, to);
1115 else
1116 folio_zero_new_buffers(folio, from, to);
1117 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1118 for (i = 0; i < nr_wait; i++) {
1119 int err2;
1120
1121 err2 = fscrypt_decrypt_pagecache_blocks(folio,
1122 blocksize, bh_offset(wait[i]));
1123 if (err2) {
1124 clear_buffer_uptodate(wait[i]);
1125 err = err2;
1126 }
1127 }
1128 }
1129
1130 return err;
1131}
1132
1133/*
1134 * To preserve ordering, it is essential that the hole instantiation and
1135 * the data write be encapsulated in a single transaction. We cannot
1136 * close off a transaction and start a new one between the ext4_get_block()
1137 * and the ext4_write_end(). So doing the jbd2_journal_start at the start of
1138 * ext4_write_begin() is the right place.
1139 */
1140static int ext4_write_begin(struct file *file, struct address_space *mapping,
1141 loff_t pos, unsigned len,
1142 struct folio **foliop, void **fsdata)
1143{
1144 struct inode *inode = mapping->host;
1145 int ret, needed_blocks;
1146 handle_t *handle;
1147 int retries = 0;
1148 struct folio *folio;
1149 pgoff_t index;
1150 unsigned from, to;
1151
1152 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
1153 return -EIO;
1154
1155 trace_ext4_write_begin(inode, pos, len);
1156 /*
1157 * Reserve one block more for addition to orphan list in case
1158 * we allocate blocks but write fails for some reason
1159 */
1160 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1161 index = pos >> PAGE_SHIFT;
1162 from = pos & (PAGE_SIZE - 1);
1163 to = from + len;
1164
1165 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1166 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1167 foliop);
1168 if (ret < 0)
1169 return ret;
1170 if (ret == 1)
1171 return 0;
1172 }
1173
1174 /*
1175 * __filemap_get_folio() can take a long time if the
1176 * system is thrashing due to memory pressure, or if the folio
1177 * is being written back. So grab it first before we start
1178 * the transaction handle. This also allows us to allocate
1179 * the folio (if needed) without using GFP_NOFS.
1180 */
1181retry_grab:
1182 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
1183 mapping_gfp_mask(mapping));
1184 if (IS_ERR(folio))
1185 return PTR_ERR(folio);
1186 /*
1187 * The same as page allocation, we prealloc buffer heads before
1188 * starting the handle.
1189 */
1190 if (!folio_buffers(folio))
1191 create_empty_buffers(folio, inode->i_sb->s_blocksize, 0);
1192
1193 folio_unlock(folio);
1194
1195retry_journal:
1196 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1197 if (IS_ERR(handle)) {
1198 folio_put(folio);
1199 return PTR_ERR(handle);
1200 }
1201
1202 folio_lock(folio);
1203 if (folio->mapping != mapping) {
1204 /* The folio got truncated from under us */
1205 folio_unlock(folio);
1206 folio_put(folio);
1207 ext4_journal_stop(handle);
1208 goto retry_grab;
1209 }
1210 /* In case writeback began while the folio was unlocked */
1211 folio_wait_stable(folio);
1212
1213 if (ext4_should_dioread_nolock(inode))
1214 ret = ext4_block_write_begin(handle, folio, pos, len,
1215 ext4_get_block_unwritten);
1216 else
1217 ret = ext4_block_write_begin(handle, folio, pos, len,
1218 ext4_get_block);
1219 if (!ret && ext4_should_journal_data(inode)) {
1220 ret = ext4_walk_page_buffers(handle, inode,
1221 folio_buffers(folio), from, to,
1222 NULL, do_journal_get_write_access);
1223 }
1224
1225 if (ret) {
1226 bool extended = (pos + len > inode->i_size) &&
1227 !ext4_verity_in_progress(inode);
1228
1229 folio_unlock(folio);
1230 /*
1231 * ext4_block_write_begin may have instantiated a few blocks
1232 * outside i_size. Trim these off again. Don't need
1233 * i_size_read because we hold i_rwsem.
1234 *
1235 * Add inode to orphan list in case we crash before
1236 * truncate finishes
1237 */
1238 if (extended && ext4_can_truncate(inode))
1239 ext4_orphan_add(handle, inode);
1240
1241 ext4_journal_stop(handle);
1242 if (extended) {
1243 ext4_truncate_failed_write(inode);
1244 /*
1245 * If truncate failed early the inode might
1246 * still be on the orphan list; we need to
1247 * make sure the inode is removed from the
1248 * orphan list in that case.
1249 */
1250 if (inode->i_nlink)
1251 ext4_orphan_del(NULL, inode);
1252 }
1253
1254 if (ret == -ENOSPC &&
1255 ext4_should_retry_alloc(inode->i_sb, &retries))
1256 goto retry_journal;
1257 folio_put(folio);
1258 return ret;
1259 }
1260 *foliop = folio;
1261 return ret;
1262}
1263
1264/* For write_end() in data=journal mode */
1265static int write_end_fn(handle_t *handle, struct inode *inode,
1266 struct buffer_head *bh)
1267{
1268 int ret;
1269 if (!buffer_mapped(bh) || buffer_freed(bh))
1270 return 0;
1271 set_buffer_uptodate(bh);
1272 ret = ext4_dirty_journalled_data(handle, bh);
1273 clear_buffer_meta(bh);
1274 clear_buffer_prio(bh);
1275 return ret;
1276}
1277
1278/*
1279 * We need to pick up the new inode size which generic_commit_write gave us
1280 * `file' can be NULL - eg, when called from page_symlink().
1281 *
1282 * ext4 never places buffers on inode->i_mapping->i_private_list. metadata
1283 * buffers are managed internally.
1284 */
1285static int ext4_write_end(struct file *file,
1286 struct address_space *mapping,
1287 loff_t pos, unsigned len, unsigned copied,
1288 struct folio *folio, void *fsdata)
1289{
1290 handle_t *handle = ext4_journal_current_handle();
1291 struct inode *inode = mapping->host;
1292 loff_t old_size = inode->i_size;
1293 int ret = 0, ret2;
1294 int i_size_changed = 0;
1295 bool verity = ext4_verity_in_progress(inode);
1296
1297 trace_ext4_write_end(inode, pos, len, copied);
1298
1299 if (ext4_has_inline_data(inode) &&
1300 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1301 return ext4_write_inline_data_end(inode, pos, len, copied,
1302 folio);
1303
1304 copied = block_write_end(file, mapping, pos, len, copied, folio, fsdata);
1305 /*
1306 * it's important to update i_size while still holding folio lock:
1307 * page writeout could otherwise come in and zero beyond i_size.
1308 *
1309 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1310 * blocks are being written past EOF, so skip the i_size update.
1311 */
1312 if (!verity)
1313 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1314 folio_unlock(folio);
1315 folio_put(folio);
1316
1317 if (old_size < pos && !verity) {
1318 pagecache_isize_extended(inode, old_size, pos);
1319 ext4_zero_partial_blocks(handle, inode, old_size, pos - old_size);
1320 }
1321 /*
1322 * Don't mark the inode dirty under folio lock. First, it unnecessarily
1323 * makes the holding time of folio lock longer. Second, it forces lock
1324 * ordering of folio lock and transaction start for journaling
1325 * filesystems.
1326 */
1327 if (i_size_changed)
1328 ret = ext4_mark_inode_dirty(handle, inode);
1329
1330 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1331 /* if we have allocated more blocks and copied
1332 * less. We will have blocks allocated outside
1333 * inode->i_size. So truncate them
1334 */
1335 ext4_orphan_add(handle, inode);
1336
1337 ret2 = ext4_journal_stop(handle);
1338 if (!ret)
1339 ret = ret2;
1340
1341 if (pos + len > inode->i_size && !verity) {
1342 ext4_truncate_failed_write(inode);
1343 /*
1344 * If truncate failed early the inode might still be
1345 * on the orphan list; we need to make sure the inode
1346 * is removed from the orphan list in that case.
1347 */
1348 if (inode->i_nlink)
1349 ext4_orphan_del(NULL, inode);
1350 }
1351
1352 return ret ? ret : copied;
1353}
1354
1355/*
1356 * This is a private version of folio_zero_new_buffers() which doesn't
1357 * set the buffer to be dirty, since in data=journalled mode we need
1358 * to call ext4_dirty_journalled_data() instead.
1359 */
1360static void ext4_journalled_zero_new_buffers(handle_t *handle,
1361 struct inode *inode,
1362 struct folio *folio,
1363 unsigned from, unsigned to)
1364{
1365 unsigned int block_start = 0, block_end;
1366 struct buffer_head *head, *bh;
1367
1368 bh = head = folio_buffers(folio);
1369 do {
1370 block_end = block_start + bh->b_size;
1371 if (buffer_new(bh)) {
1372 if (block_end > from && block_start < to) {
1373 if (!folio_test_uptodate(folio)) {
1374 unsigned start, size;
1375
1376 start = max(from, block_start);
1377 size = min(to, block_end) - start;
1378
1379 folio_zero_range(folio, start, size);
1380 }
1381 clear_buffer_new(bh);
1382 write_end_fn(handle, inode, bh);
1383 }
1384 }
1385 block_start = block_end;
1386 bh = bh->b_this_page;
1387 } while (bh != head);
1388}
1389
1390static int ext4_journalled_write_end(struct file *file,
1391 struct address_space *mapping,
1392 loff_t pos, unsigned len, unsigned copied,
1393 struct folio *folio, void *fsdata)
1394{
1395 handle_t *handle = ext4_journal_current_handle();
1396 struct inode *inode = mapping->host;
1397 loff_t old_size = inode->i_size;
1398 int ret = 0, ret2;
1399 int partial = 0;
1400 unsigned from, to;
1401 int size_changed = 0;
1402 bool verity = ext4_verity_in_progress(inode);
1403
1404 trace_ext4_journalled_write_end(inode, pos, len, copied);
1405 from = pos & (PAGE_SIZE - 1);
1406 to = from + len;
1407
1408 BUG_ON(!ext4_handle_valid(handle));
1409
1410 if (ext4_has_inline_data(inode))
1411 return ext4_write_inline_data_end(inode, pos, len, copied,
1412 folio);
1413
1414 if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1415 copied = 0;
1416 ext4_journalled_zero_new_buffers(handle, inode, folio,
1417 from, to);
1418 } else {
1419 if (unlikely(copied < len))
1420 ext4_journalled_zero_new_buffers(handle, inode, folio,
1421 from + copied, to);
1422 ret = ext4_walk_page_buffers(handle, inode,
1423 folio_buffers(folio),
1424 from, from + copied, &partial,
1425 write_end_fn);
1426 if (!partial)
1427 folio_mark_uptodate(folio);
1428 }
1429 if (!verity)
1430 size_changed = ext4_update_inode_size(inode, pos + copied);
1431 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1432 folio_unlock(folio);
1433 folio_put(folio);
1434
1435 if (old_size < pos && !verity) {
1436 pagecache_isize_extended(inode, old_size, pos);
1437 ext4_zero_partial_blocks(handle, inode, old_size, pos - old_size);
1438 }
1439
1440 if (size_changed) {
1441 ret2 = ext4_mark_inode_dirty(handle, inode);
1442 if (!ret)
1443 ret = ret2;
1444 }
1445
1446 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1447 /* if we have allocated more blocks and copied
1448 * less. We will have blocks allocated outside
1449 * inode->i_size. So truncate them
1450 */
1451 ext4_orphan_add(handle, inode);
1452
1453 ret2 = ext4_journal_stop(handle);
1454 if (!ret)
1455 ret = ret2;
1456 if (pos + len > inode->i_size && !verity) {
1457 ext4_truncate_failed_write(inode);
1458 /*
1459 * If truncate failed early the inode might still be
1460 * on the orphan list; we need to make sure the inode
1461 * is removed from the orphan list in that case.
1462 */
1463 if (inode->i_nlink)
1464 ext4_orphan_del(NULL, inode);
1465 }
1466
1467 return ret ? ret : copied;
1468}
1469
1470/*
1471 * Reserve space for 'nr_resv' clusters
1472 */
1473static int ext4_da_reserve_space(struct inode *inode, int nr_resv)
1474{
1475 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1476 struct ext4_inode_info *ei = EXT4_I(inode);
1477 int ret;
1478
1479 /*
1480 * We will charge metadata quota at writeout time; this saves
1481 * us from metadata over-estimation, though we may go over by
1482 * a small amount in the end. Here we just reserve for data.
1483 */
1484 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, nr_resv));
1485 if (ret)
1486 return ret;
1487
1488 spin_lock(&ei->i_block_reservation_lock);
1489 if (ext4_claim_free_clusters(sbi, nr_resv, 0)) {
1490 spin_unlock(&ei->i_block_reservation_lock);
1491 dquot_release_reservation_block(inode, EXT4_C2B(sbi, nr_resv));
1492 return -ENOSPC;
1493 }
1494 ei->i_reserved_data_blocks += nr_resv;
1495 trace_ext4_da_reserve_space(inode, nr_resv);
1496 spin_unlock(&ei->i_block_reservation_lock);
1497
1498 return 0; /* success */
1499}
1500
1501void ext4_da_release_space(struct inode *inode, int to_free)
1502{
1503 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1504 struct ext4_inode_info *ei = EXT4_I(inode);
1505
1506 if (!to_free)
1507 return; /* Nothing to release, exit */
1508
1509 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1510
1511 trace_ext4_da_release_space(inode, to_free);
1512 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1513 /*
1514 * if there aren't enough reserved blocks, then the
1515 * counter is messed up somewhere. Since this
1516 * function is called from invalidate page, it's
1517 * harmless to return without any action.
1518 */
1519 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1520 "ino %lu, to_free %d with only %d reserved "
1521 "data blocks", inode->i_ino, to_free,
1522 ei->i_reserved_data_blocks);
1523 WARN_ON(1);
1524 to_free = ei->i_reserved_data_blocks;
1525 }
1526 ei->i_reserved_data_blocks -= to_free;
1527
1528 /* update fs dirty data blocks counter */
1529 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1530
1531 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1532
1533 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1534}
1535
1536/*
1537 * Delayed allocation stuff
1538 */
1539
1540struct mpage_da_data {
1541 /* These are input fields for ext4_do_writepages() */
1542 struct inode *inode;
1543 struct writeback_control *wbc;
1544 unsigned int can_map:1; /* Can writepages call map blocks? */
1545
1546 /* These are internal state of ext4_do_writepages() */
1547 pgoff_t first_page; /* The first page to write */
1548 pgoff_t next_page; /* Current page to examine */
1549 pgoff_t last_page; /* Last page to examine */
1550 /*
1551 * Extent to map - this can be after first_page because that can be
1552 * fully mapped. We somewhat abuse m_flags to store whether the extent
1553 * is delalloc or unwritten.
1554 */
1555 struct ext4_map_blocks map;
1556 struct ext4_io_submit io_submit; /* IO submission data */
1557 unsigned int do_map:1;
1558 unsigned int scanned_until_end:1;
1559 unsigned int journalled_more_data:1;
1560};
1561
1562static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1563 bool invalidate)
1564{
1565 unsigned nr, i;
1566 pgoff_t index, end;
1567 struct folio_batch fbatch;
1568 struct inode *inode = mpd->inode;
1569 struct address_space *mapping = inode->i_mapping;
1570
1571 /* This is necessary when next_page == 0. */
1572 if (mpd->first_page >= mpd->next_page)
1573 return;
1574
1575 mpd->scanned_until_end = 0;
1576 index = mpd->first_page;
1577 end = mpd->next_page - 1;
1578 if (invalidate) {
1579 ext4_lblk_t start, last;
1580 start = index << (PAGE_SHIFT - inode->i_blkbits);
1581 last = end << (PAGE_SHIFT - inode->i_blkbits);
1582
1583 /*
1584 * avoid racing with extent status tree scans made by
1585 * ext4_insert_delayed_block()
1586 */
1587 down_write(&EXT4_I(inode)->i_data_sem);
1588 ext4_es_remove_extent(inode, start, last - start + 1);
1589 up_write(&EXT4_I(inode)->i_data_sem);
1590 }
1591
1592 folio_batch_init(&fbatch);
1593 while (index <= end) {
1594 nr = filemap_get_folios(mapping, &index, end, &fbatch);
1595 if (nr == 0)
1596 break;
1597 for (i = 0; i < nr; i++) {
1598 struct folio *folio = fbatch.folios[i];
1599
1600 if (folio->index < mpd->first_page)
1601 continue;
1602 if (folio_next_index(folio) - 1 > end)
1603 continue;
1604 BUG_ON(!folio_test_locked(folio));
1605 BUG_ON(folio_test_writeback(folio));
1606 if (invalidate) {
1607 if (folio_mapped(folio))
1608 folio_clear_dirty_for_io(folio);
1609 block_invalidate_folio(folio, 0,
1610 folio_size(folio));
1611 folio_clear_uptodate(folio);
1612 }
1613 folio_unlock(folio);
1614 }
1615 folio_batch_release(&fbatch);
1616 }
1617}
1618
1619static void ext4_print_free_blocks(struct inode *inode)
1620{
1621 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1622 struct super_block *sb = inode->i_sb;
1623 struct ext4_inode_info *ei = EXT4_I(inode);
1624
1625 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1626 EXT4_C2B(EXT4_SB(inode->i_sb),
1627 ext4_count_free_clusters(sb)));
1628 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1629 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1630 (long long) EXT4_C2B(EXT4_SB(sb),
1631 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1632 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1633 (long long) EXT4_C2B(EXT4_SB(sb),
1634 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1635 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1636 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1637 ei->i_reserved_data_blocks);
1638 return;
1639}
1640
1641/*
1642 * Check whether the cluster containing lblk has been allocated or has
1643 * delalloc reservation.
1644 *
1645 * Returns 0 if the cluster doesn't have either, 1 if it has delalloc
1646 * reservation, 2 if it's already been allocated, negative error code on
1647 * failure.
1648 */
1649static int ext4_clu_alloc_state(struct inode *inode, ext4_lblk_t lblk)
1650{
1651 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1652 int ret;
1653
1654 /* Has delalloc reservation? */
1655 if (ext4_es_scan_clu(inode, &ext4_es_is_delayed, lblk))
1656 return 1;
1657
1658 /* Already been allocated? */
1659 if (ext4_es_scan_clu(inode, &ext4_es_is_mapped, lblk))
1660 return 2;
1661 ret = ext4_clu_mapped(inode, EXT4_B2C(sbi, lblk));
1662 if (ret < 0)
1663 return ret;
1664 if (ret > 0)
1665 return 2;
1666
1667 return 0;
1668}
1669
1670/*
1671 * ext4_insert_delayed_blocks - adds a multiple delayed blocks to the extents
1672 * status tree, incrementing the reserved
1673 * cluster/block count or making pending
1674 * reservations where needed
1675 *
1676 * @inode - file containing the newly added block
1677 * @lblk - start logical block to be added
1678 * @len - length of blocks to be added
1679 *
1680 * Returns 0 on success, negative error code on failure.
1681 */
1682static int ext4_insert_delayed_blocks(struct inode *inode, ext4_lblk_t lblk,
1683 ext4_lblk_t len)
1684{
1685 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1686 int ret;
1687 bool lclu_allocated = false;
1688 bool end_allocated = false;
1689 ext4_lblk_t resv_clu;
1690 ext4_lblk_t end = lblk + len - 1;
1691
1692 /*
1693 * If the cluster containing lblk or end is shared with a delayed,
1694 * written, or unwritten extent in a bigalloc file system, it's
1695 * already been accounted for and does not need to be reserved.
1696 * A pending reservation must be made for the cluster if it's
1697 * shared with a written or unwritten extent and doesn't already
1698 * have one. Written and unwritten extents can be purged from the
1699 * extents status tree if the system is under memory pressure, so
1700 * it's necessary to examine the extent tree if a search of the
1701 * extents status tree doesn't get a match.
1702 */
1703 if (sbi->s_cluster_ratio == 1) {
1704 ret = ext4_da_reserve_space(inode, len);
1705 if (ret != 0) /* ENOSPC */
1706 return ret;
1707 } else { /* bigalloc */
1708 resv_clu = EXT4_B2C(sbi, end) - EXT4_B2C(sbi, lblk) + 1;
1709
1710 ret = ext4_clu_alloc_state(inode, lblk);
1711 if (ret < 0)
1712 return ret;
1713 if (ret > 0) {
1714 resv_clu--;
1715 lclu_allocated = (ret == 2);
1716 }
1717
1718 if (EXT4_B2C(sbi, lblk) != EXT4_B2C(sbi, end)) {
1719 ret = ext4_clu_alloc_state(inode, end);
1720 if (ret < 0)
1721 return ret;
1722 if (ret > 0) {
1723 resv_clu--;
1724 end_allocated = (ret == 2);
1725 }
1726 }
1727
1728 if (resv_clu) {
1729 ret = ext4_da_reserve_space(inode, resv_clu);
1730 if (ret != 0) /* ENOSPC */
1731 return ret;
1732 }
1733 }
1734
1735 ext4_es_insert_delayed_extent(inode, lblk, len, lclu_allocated,
1736 end_allocated);
1737 return 0;
1738}
1739
1740/*
1741 * Looks up the requested blocks and sets the delalloc extent map.
1742 * First try to look up for the extent entry that contains the requested
1743 * blocks in the extent status tree without i_data_sem, then try to look
1744 * up for the ondisk extent mapping with i_data_sem in read mode,
1745 * finally hold i_data_sem in write mode, looks up again and add a
1746 * delalloc extent entry if it still couldn't find any extent. Pass out
1747 * the mapped extent through @map and return 0 on success.
1748 */
1749static int ext4_da_map_blocks(struct inode *inode, struct ext4_map_blocks *map)
1750{
1751 struct extent_status es;
1752 int retval;
1753#ifdef ES_AGGRESSIVE_TEST
1754 struct ext4_map_blocks orig_map;
1755
1756 memcpy(&orig_map, map, sizeof(*map));
1757#endif
1758
1759 map->m_flags = 0;
1760 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1761 (unsigned long) map->m_lblk);
1762
1763 /* Lookup extent status tree firstly */
1764 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
1765 map->m_len = min_t(unsigned int, map->m_len,
1766 es.es_len - (map->m_lblk - es.es_lblk));
1767
1768 if (ext4_es_is_hole(&es))
1769 goto add_delayed;
1770
1771found:
1772 /*
1773 * Delayed extent could be allocated by fallocate.
1774 * So we need to check it.
1775 */
1776 if (ext4_es_is_delayed(&es)) {
1777 map->m_flags |= EXT4_MAP_DELAYED;
1778 return 0;
1779 }
1780
1781 map->m_pblk = ext4_es_pblock(&es) + map->m_lblk - es.es_lblk;
1782 if (ext4_es_is_written(&es))
1783 map->m_flags |= EXT4_MAP_MAPPED;
1784 else if (ext4_es_is_unwritten(&es))
1785 map->m_flags |= EXT4_MAP_UNWRITTEN;
1786 else
1787 BUG();
1788
1789#ifdef ES_AGGRESSIVE_TEST
1790 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1791#endif
1792 return 0;
1793 }
1794
1795 /*
1796 * Try to see if we can get the block without requesting a new
1797 * file system block.
1798 */
1799 down_read(&EXT4_I(inode)->i_data_sem);
1800 if (ext4_has_inline_data(inode))
1801 retval = 0;
1802 else
1803 retval = ext4_map_query_blocks(NULL, inode, map);
1804 up_read(&EXT4_I(inode)->i_data_sem);
1805 if (retval)
1806 return retval < 0 ? retval : 0;
1807
1808add_delayed:
1809 down_write(&EXT4_I(inode)->i_data_sem);
1810 /*
1811 * Page fault path (ext4_page_mkwrite does not take i_rwsem)
1812 * and fallocate path (no folio lock) can race. Make sure we
1813 * lookup the extent status tree here again while i_data_sem
1814 * is held in write mode, before inserting a new da entry in
1815 * the extent status tree.
1816 */
1817 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
1818 map->m_len = min_t(unsigned int, map->m_len,
1819 es.es_len - (map->m_lblk - es.es_lblk));
1820
1821 if (!ext4_es_is_hole(&es)) {
1822 up_write(&EXT4_I(inode)->i_data_sem);
1823 goto found;
1824 }
1825 } else if (!ext4_has_inline_data(inode)) {
1826 retval = ext4_map_query_blocks(NULL, inode, map);
1827 if (retval) {
1828 up_write(&EXT4_I(inode)->i_data_sem);
1829 return retval < 0 ? retval : 0;
1830 }
1831 }
1832
1833 map->m_flags |= EXT4_MAP_DELAYED;
1834 retval = ext4_insert_delayed_blocks(inode, map->m_lblk, map->m_len);
1835 up_write(&EXT4_I(inode)->i_data_sem);
1836
1837 return retval;
1838}
1839
1840/*
1841 * This is a special get_block_t callback which is used by
1842 * ext4_da_write_begin(). It will either return mapped block or
1843 * reserve space for a single block.
1844 *
1845 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1846 * We also have b_blocknr = -1 and b_bdev initialized properly
1847 *
1848 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1849 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1850 * initialized properly.
1851 */
1852int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1853 struct buffer_head *bh, int create)
1854{
1855 struct ext4_map_blocks map;
1856 sector_t invalid_block = ~((sector_t) 0xffff);
1857 int ret = 0;
1858
1859 BUG_ON(create == 0);
1860 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1861
1862 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1863 invalid_block = ~0;
1864
1865 map.m_lblk = iblock;
1866 map.m_len = 1;
1867
1868 /*
1869 * first, we need to know whether the block is allocated already
1870 * preallocated blocks are unmapped but should treated
1871 * the same as allocated blocks.
1872 */
1873 ret = ext4_da_map_blocks(inode, &map);
1874 if (ret < 0)
1875 return ret;
1876
1877 if (map.m_flags & EXT4_MAP_DELAYED) {
1878 map_bh(bh, inode->i_sb, invalid_block);
1879 set_buffer_new(bh);
1880 set_buffer_delay(bh);
1881 return 0;
1882 }
1883
1884 map_bh(bh, inode->i_sb, map.m_pblk);
1885 ext4_update_bh_state(bh, map.m_flags);
1886
1887 if (buffer_unwritten(bh)) {
1888 /* A delayed write to unwritten bh should be marked
1889 * new and mapped. Mapped ensures that we don't do
1890 * get_block multiple times when we write to the same
1891 * offset and new ensures that we do proper zero out
1892 * for partial write.
1893 */
1894 set_buffer_new(bh);
1895 set_buffer_mapped(bh);
1896 }
1897 return 0;
1898}
1899
1900static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
1901{
1902 mpd->first_page += folio_nr_pages(folio);
1903 folio_unlock(folio);
1904}
1905
1906static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
1907{
1908 size_t len;
1909 loff_t size;
1910 int err;
1911
1912 BUG_ON(folio->index != mpd->first_page);
1913 folio_clear_dirty_for_io(folio);
1914 /*
1915 * We have to be very careful here! Nothing protects writeback path
1916 * against i_size changes and the page can be writeably mapped into
1917 * page tables. So an application can be growing i_size and writing
1918 * data through mmap while writeback runs. folio_clear_dirty_for_io()
1919 * write-protects our page in page tables and the page cannot get
1920 * written to again until we release folio lock. So only after
1921 * folio_clear_dirty_for_io() we are safe to sample i_size for
1922 * ext4_bio_write_folio() to zero-out tail of the written page. We rely
1923 * on the barrier provided by folio_test_clear_dirty() in
1924 * folio_clear_dirty_for_io() to make sure i_size is really sampled only
1925 * after page tables are updated.
1926 */
1927 size = i_size_read(mpd->inode);
1928 len = folio_size(folio);
1929 if (folio_pos(folio) + len > size &&
1930 !ext4_verity_in_progress(mpd->inode))
1931 len = size & (len - 1);
1932 err = ext4_bio_write_folio(&mpd->io_submit, folio, len);
1933 if (!err)
1934 mpd->wbc->nr_to_write--;
1935
1936 return err;
1937}
1938
1939#define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
1940
1941/*
1942 * mballoc gives us at most this number of blocks...
1943 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1944 * The rest of mballoc seems to handle chunks up to full group size.
1945 */
1946#define MAX_WRITEPAGES_EXTENT_LEN 2048
1947
1948/*
1949 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1950 *
1951 * @mpd - extent of blocks
1952 * @lblk - logical number of the block in the file
1953 * @bh - buffer head we want to add to the extent
1954 *
1955 * The function is used to collect contig. blocks in the same state. If the
1956 * buffer doesn't require mapping for writeback and we haven't started the
1957 * extent of buffers to map yet, the function returns 'true' immediately - the
1958 * caller can write the buffer right away. Otherwise the function returns true
1959 * if the block has been added to the extent, false if the block couldn't be
1960 * added.
1961 */
1962static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1963 struct buffer_head *bh)
1964{
1965 struct ext4_map_blocks *map = &mpd->map;
1966
1967 /* Buffer that doesn't need mapping for writeback? */
1968 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1969 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1970 /* So far no extent to map => we write the buffer right away */
1971 if (map->m_len == 0)
1972 return true;
1973 return false;
1974 }
1975
1976 /* First block in the extent? */
1977 if (map->m_len == 0) {
1978 /* We cannot map unless handle is started... */
1979 if (!mpd->do_map)
1980 return false;
1981 map->m_lblk = lblk;
1982 map->m_len = 1;
1983 map->m_flags = bh->b_state & BH_FLAGS;
1984 return true;
1985 }
1986
1987 /* Don't go larger than mballoc is willing to allocate */
1988 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1989 return false;
1990
1991 /* Can we merge the block to our big extent? */
1992 if (lblk == map->m_lblk + map->m_len &&
1993 (bh->b_state & BH_FLAGS) == map->m_flags) {
1994 map->m_len++;
1995 return true;
1996 }
1997 return false;
1998}
1999
2000/*
2001 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2002 *
2003 * @mpd - extent of blocks for mapping
2004 * @head - the first buffer in the page
2005 * @bh - buffer we should start processing from
2006 * @lblk - logical number of the block in the file corresponding to @bh
2007 *
2008 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2009 * the page for IO if all buffers in this page were mapped and there's no
2010 * accumulated extent of buffers to map or add buffers in the page to the
2011 * extent of buffers to map. The function returns 1 if the caller can continue
2012 * by processing the next page, 0 if it should stop adding buffers to the
2013 * extent to map because we cannot extend it anymore. It can also return value
2014 * < 0 in case of error during IO submission.
2015 */
2016static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2017 struct buffer_head *head,
2018 struct buffer_head *bh,
2019 ext4_lblk_t lblk)
2020{
2021 struct inode *inode = mpd->inode;
2022 int err;
2023 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2024 >> inode->i_blkbits;
2025
2026 if (ext4_verity_in_progress(inode))
2027 blocks = EXT_MAX_BLOCKS;
2028
2029 do {
2030 BUG_ON(buffer_locked(bh));
2031
2032 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2033 /* Found extent to map? */
2034 if (mpd->map.m_len)
2035 return 0;
2036 /* Buffer needs mapping and handle is not started? */
2037 if (!mpd->do_map)
2038 return 0;
2039 /* Everything mapped so far and we hit EOF */
2040 break;
2041 }
2042 } while (lblk++, (bh = bh->b_this_page) != head);
2043 /* So far everything mapped? Submit the page for IO. */
2044 if (mpd->map.m_len == 0) {
2045 err = mpage_submit_folio(mpd, head->b_folio);
2046 if (err < 0)
2047 return err;
2048 mpage_folio_done(mpd, head->b_folio);
2049 }
2050 if (lblk >= blocks) {
2051 mpd->scanned_until_end = 1;
2052 return 0;
2053 }
2054 return 1;
2055}
2056
2057/*
2058 * mpage_process_folio - update folio buffers corresponding to changed extent
2059 * and may submit fully mapped page for IO
2060 * @mpd: description of extent to map, on return next extent to map
2061 * @folio: Contains these buffers.
2062 * @m_lblk: logical block mapping.
2063 * @m_pblk: corresponding physical mapping.
2064 * @map_bh: determines on return whether this page requires any further
2065 * mapping or not.
2066 *
2067 * Scan given folio buffers corresponding to changed extent and update buffer
2068 * state according to new extent state.
2069 * We map delalloc buffers to their physical location, clear unwritten bits.
2070 * If the given folio is not fully mapped, we update @mpd to the next extent in
2071 * the given folio that needs mapping & return @map_bh as true.
2072 */
2073static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2074 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2075 bool *map_bh)
2076{
2077 struct buffer_head *head, *bh;
2078 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2079 ext4_lblk_t lblk = *m_lblk;
2080 ext4_fsblk_t pblock = *m_pblk;
2081 int err = 0;
2082 int blkbits = mpd->inode->i_blkbits;
2083 ssize_t io_end_size = 0;
2084 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2085
2086 bh = head = folio_buffers(folio);
2087 do {
2088 if (lblk < mpd->map.m_lblk)
2089 continue;
2090 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2091 /*
2092 * Buffer after end of mapped extent.
2093 * Find next buffer in the folio to map.
2094 */
2095 mpd->map.m_len = 0;
2096 mpd->map.m_flags = 0;
2097 io_end_vec->size += io_end_size;
2098
2099 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2100 if (err > 0)
2101 err = 0;
2102 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2103 io_end_vec = ext4_alloc_io_end_vec(io_end);
2104 if (IS_ERR(io_end_vec)) {
2105 err = PTR_ERR(io_end_vec);
2106 goto out;
2107 }
2108 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2109 }
2110 *map_bh = true;
2111 goto out;
2112 }
2113 if (buffer_delay(bh)) {
2114 clear_buffer_delay(bh);
2115 bh->b_blocknr = pblock++;
2116 }
2117 clear_buffer_unwritten(bh);
2118 io_end_size += (1 << blkbits);
2119 } while (lblk++, (bh = bh->b_this_page) != head);
2120
2121 io_end_vec->size += io_end_size;
2122 *map_bh = false;
2123out:
2124 *m_lblk = lblk;
2125 *m_pblk = pblock;
2126 return err;
2127}
2128
2129/*
2130 * mpage_map_buffers - update buffers corresponding to changed extent and
2131 * submit fully mapped pages for IO
2132 *
2133 * @mpd - description of extent to map, on return next extent to map
2134 *
2135 * Scan buffers corresponding to changed extent (we expect corresponding pages
2136 * to be already locked) and update buffer state according to new extent state.
2137 * We map delalloc buffers to their physical location, clear unwritten bits,
2138 * and mark buffers as uninit when we perform writes to unwritten extents
2139 * and do extent conversion after IO is finished. If the last page is not fully
2140 * mapped, we update @map to the next extent in the last page that needs
2141 * mapping. Otherwise we submit the page for IO.
2142 */
2143static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2144{
2145 struct folio_batch fbatch;
2146 unsigned nr, i;
2147 struct inode *inode = mpd->inode;
2148 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2149 pgoff_t start, end;
2150 ext4_lblk_t lblk;
2151 ext4_fsblk_t pblock;
2152 int err;
2153 bool map_bh = false;
2154
2155 start = mpd->map.m_lblk >> bpp_bits;
2156 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2157 lblk = start << bpp_bits;
2158 pblock = mpd->map.m_pblk;
2159
2160 folio_batch_init(&fbatch);
2161 while (start <= end) {
2162 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2163 if (nr == 0)
2164 break;
2165 for (i = 0; i < nr; i++) {
2166 struct folio *folio = fbatch.folios[i];
2167
2168 err = mpage_process_folio(mpd, folio, &lblk, &pblock,
2169 &map_bh);
2170 /*
2171 * If map_bh is true, means page may require further bh
2172 * mapping, or maybe the page was submitted for IO.
2173 * So we return to call further extent mapping.
2174 */
2175 if (err < 0 || map_bh)
2176 goto out;
2177 /* Page fully mapped - let IO run! */
2178 err = mpage_submit_folio(mpd, folio);
2179 if (err < 0)
2180 goto out;
2181 mpage_folio_done(mpd, folio);
2182 }
2183 folio_batch_release(&fbatch);
2184 }
2185 /* Extent fully mapped and matches with page boundary. We are done. */
2186 mpd->map.m_len = 0;
2187 mpd->map.m_flags = 0;
2188 return 0;
2189out:
2190 folio_batch_release(&fbatch);
2191 return err;
2192}
2193
2194static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2195{
2196 struct inode *inode = mpd->inode;
2197 struct ext4_map_blocks *map = &mpd->map;
2198 int get_blocks_flags;
2199 int err, dioread_nolock;
2200
2201 trace_ext4_da_write_pages_extent(inode, map);
2202 /*
2203 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2204 * to convert an unwritten extent to be initialized (in the case
2205 * where we have written into one or more preallocated blocks). It is
2206 * possible that we're going to need more metadata blocks than
2207 * previously reserved. However we must not fail because we're in
2208 * writeback and there is nothing we can do about it so it might result
2209 * in data loss. So use reserved blocks to allocate metadata if
2210 * possible.
2211 */
2212 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2213 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2214 EXT4_GET_BLOCKS_IO_SUBMIT;
2215 dioread_nolock = ext4_should_dioread_nolock(inode);
2216 if (dioread_nolock)
2217 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2218
2219 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2220 if (err < 0)
2221 return err;
2222 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2223 if (!mpd->io_submit.io_end->handle &&
2224 ext4_handle_valid(handle)) {
2225 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2226 handle->h_rsv_handle = NULL;
2227 }
2228 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2229 }
2230
2231 BUG_ON(map->m_len == 0);
2232 return 0;
2233}
2234
2235/*
2236 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2237 * mpd->len and submit pages underlying it for IO
2238 *
2239 * @handle - handle for journal operations
2240 * @mpd - extent to map
2241 * @give_up_on_write - we set this to true iff there is a fatal error and there
2242 * is no hope of writing the data. The caller should discard
2243 * dirty pages to avoid infinite loops.
2244 *
2245 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2246 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2247 * them to initialized or split the described range from larger unwritten
2248 * extent. Note that we need not map all the described range since allocation
2249 * can return less blocks or the range is covered by more unwritten extents. We
2250 * cannot map more because we are limited by reserved transaction credits. On
2251 * the other hand we always make sure that the last touched page is fully
2252 * mapped so that it can be written out (and thus forward progress is
2253 * guaranteed). After mapping we submit all mapped pages for IO.
2254 */
2255static int mpage_map_and_submit_extent(handle_t *handle,
2256 struct mpage_da_data *mpd,
2257 bool *give_up_on_write)
2258{
2259 struct inode *inode = mpd->inode;
2260 struct ext4_map_blocks *map = &mpd->map;
2261 int err;
2262 loff_t disksize;
2263 int progress = 0;
2264 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2265 struct ext4_io_end_vec *io_end_vec;
2266
2267 io_end_vec = ext4_alloc_io_end_vec(io_end);
2268 if (IS_ERR(io_end_vec))
2269 return PTR_ERR(io_end_vec);
2270 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2271 do {
2272 err = mpage_map_one_extent(handle, mpd);
2273 if (err < 0) {
2274 struct super_block *sb = inode->i_sb;
2275
2276 if (ext4_forced_shutdown(sb))
2277 goto invalidate_dirty_pages;
2278 /*
2279 * Let the uper layers retry transient errors.
2280 * In the case of ENOSPC, if ext4_count_free_blocks()
2281 * is non-zero, a commit should free up blocks.
2282 */
2283 if ((err == -ENOMEM) ||
2284 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2285 if (progress)
2286 goto update_disksize;
2287 return err;
2288 }
2289 ext4_msg(sb, KERN_CRIT,
2290 "Delayed block allocation failed for "
2291 "inode %lu at logical offset %llu with"
2292 " max blocks %u with error %d",
2293 inode->i_ino,
2294 (unsigned long long)map->m_lblk,
2295 (unsigned)map->m_len, -err);
2296 ext4_msg(sb, KERN_CRIT,
2297 "This should not happen!! Data will "
2298 "be lost\n");
2299 if (err == -ENOSPC)
2300 ext4_print_free_blocks(inode);
2301 invalidate_dirty_pages:
2302 *give_up_on_write = true;
2303 return err;
2304 }
2305 progress = 1;
2306 /*
2307 * Update buffer state, submit mapped pages, and get us new
2308 * extent to map
2309 */
2310 err = mpage_map_and_submit_buffers(mpd);
2311 if (err < 0)
2312 goto update_disksize;
2313 } while (map->m_len);
2314
2315update_disksize:
2316 /*
2317 * Update on-disk size after IO is submitted. Races with
2318 * truncate are avoided by checking i_size under i_data_sem.
2319 */
2320 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2321 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2322 int err2;
2323 loff_t i_size;
2324
2325 down_write(&EXT4_I(inode)->i_data_sem);
2326 i_size = i_size_read(inode);
2327 if (disksize > i_size)
2328 disksize = i_size;
2329 if (disksize > EXT4_I(inode)->i_disksize)
2330 EXT4_I(inode)->i_disksize = disksize;
2331 up_write(&EXT4_I(inode)->i_data_sem);
2332 err2 = ext4_mark_inode_dirty(handle, inode);
2333 if (err2) {
2334 ext4_error_err(inode->i_sb, -err2,
2335 "Failed to mark inode %lu dirty",
2336 inode->i_ino);
2337 }
2338 if (!err)
2339 err = err2;
2340 }
2341 return err;
2342}
2343
2344/*
2345 * Calculate the total number of credits to reserve for one writepages
2346 * iteration. This is called from ext4_writepages(). We map an extent of
2347 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2348 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2349 * bpp - 1 blocks in bpp different extents.
2350 */
2351static int ext4_da_writepages_trans_blocks(struct inode *inode)
2352{
2353 int bpp = ext4_journal_blocks_per_page(inode);
2354
2355 return ext4_meta_trans_blocks(inode,
2356 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2357}
2358
2359static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio,
2360 size_t len)
2361{
2362 struct buffer_head *page_bufs = folio_buffers(folio);
2363 struct inode *inode = folio->mapping->host;
2364 int ret, err;
2365
2366 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2367 NULL, do_journal_get_write_access);
2368 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2369 NULL, write_end_fn);
2370 if (ret == 0)
2371 ret = err;
2372 err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len);
2373 if (ret == 0)
2374 ret = err;
2375 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2376
2377 return ret;
2378}
2379
2380static int mpage_journal_page_buffers(handle_t *handle,
2381 struct mpage_da_data *mpd,
2382 struct folio *folio)
2383{
2384 struct inode *inode = mpd->inode;
2385 loff_t size = i_size_read(inode);
2386 size_t len = folio_size(folio);
2387
2388 folio_clear_checked(folio);
2389 mpd->wbc->nr_to_write--;
2390
2391 if (folio_pos(folio) + len > size &&
2392 !ext4_verity_in_progress(inode))
2393 len = size & (len - 1);
2394
2395 return ext4_journal_folio_buffers(handle, folio, len);
2396}
2397
2398/*
2399 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2400 * needing mapping, submit mapped pages
2401 *
2402 * @mpd - where to look for pages
2403 *
2404 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2405 * IO immediately. If we cannot map blocks, we submit just already mapped
2406 * buffers in the page for IO and keep page dirty. When we can map blocks and
2407 * we find a page which isn't mapped we start accumulating extent of buffers
2408 * underlying these pages that needs mapping (formed by either delayed or
2409 * unwritten buffers). We also lock the pages containing these buffers. The
2410 * extent found is returned in @mpd structure (starting at mpd->lblk with
2411 * length mpd->len blocks).
2412 *
2413 * Note that this function can attach bios to one io_end structure which are
2414 * neither logically nor physically contiguous. Although it may seem as an
2415 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2416 * case as we need to track IO to all buffers underlying a page in one io_end.
2417 */
2418static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2419{
2420 struct address_space *mapping = mpd->inode->i_mapping;
2421 struct folio_batch fbatch;
2422 unsigned int nr_folios;
2423 pgoff_t index = mpd->first_page;
2424 pgoff_t end = mpd->last_page;
2425 xa_mark_t tag;
2426 int i, err = 0;
2427 int blkbits = mpd->inode->i_blkbits;
2428 ext4_lblk_t lblk;
2429 struct buffer_head *head;
2430 handle_t *handle = NULL;
2431 int bpp = ext4_journal_blocks_per_page(mpd->inode);
2432
2433 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2434 tag = PAGECACHE_TAG_TOWRITE;
2435 else
2436 tag = PAGECACHE_TAG_DIRTY;
2437
2438 mpd->map.m_len = 0;
2439 mpd->next_page = index;
2440 if (ext4_should_journal_data(mpd->inode)) {
2441 handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2442 bpp);
2443 if (IS_ERR(handle))
2444 return PTR_ERR(handle);
2445 }
2446 folio_batch_init(&fbatch);
2447 while (index <= end) {
2448 nr_folios = filemap_get_folios_tag(mapping, &index, end,
2449 tag, &fbatch);
2450 if (nr_folios == 0)
2451 break;
2452
2453 for (i = 0; i < nr_folios; i++) {
2454 struct folio *folio = fbatch.folios[i];
2455
2456 /*
2457 * Accumulated enough dirty pages? This doesn't apply
2458 * to WB_SYNC_ALL mode. For integrity sync we have to
2459 * keep going because someone may be concurrently
2460 * dirtying pages, and we might have synced a lot of
2461 * newly appeared dirty pages, but have not synced all
2462 * of the old dirty pages.
2463 */
2464 if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2465 mpd->wbc->nr_to_write <=
2466 mpd->map.m_len >> (PAGE_SHIFT - blkbits))
2467 goto out;
2468
2469 /* If we can't merge this page, we are done. */
2470 if (mpd->map.m_len > 0 && mpd->next_page != folio->index)
2471 goto out;
2472
2473 if (handle) {
2474 err = ext4_journal_ensure_credits(handle, bpp,
2475 0);
2476 if (err < 0)
2477 goto out;
2478 }
2479
2480 folio_lock(folio);
2481 /*
2482 * If the page is no longer dirty, or its mapping no
2483 * longer corresponds to inode we are writing (which
2484 * means it has been truncated or invalidated), or the
2485 * page is already under writeback and we are not doing
2486 * a data integrity writeback, skip the page
2487 */
2488 if (!folio_test_dirty(folio) ||
2489 (folio_test_writeback(folio) &&
2490 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2491 unlikely(folio->mapping != mapping)) {
2492 folio_unlock(folio);
2493 continue;
2494 }
2495
2496 folio_wait_writeback(folio);
2497 BUG_ON(folio_test_writeback(folio));
2498
2499 /*
2500 * Should never happen but for buggy code in
2501 * other subsystems that call
2502 * set_page_dirty() without properly warning
2503 * the file system first. See [1] for more
2504 * information.
2505 *
2506 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2507 */
2508 if (!folio_buffers(folio)) {
2509 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2510 folio_clear_dirty(folio);
2511 folio_unlock(folio);
2512 continue;
2513 }
2514
2515 if (mpd->map.m_len == 0)
2516 mpd->first_page = folio->index;
2517 mpd->next_page = folio_next_index(folio);
2518 /*
2519 * Writeout when we cannot modify metadata is simple.
2520 * Just submit the page. For data=journal mode we
2521 * first handle writeout of the page for checkpoint and
2522 * only after that handle delayed page dirtying. This
2523 * makes sure current data is checkpointed to the final
2524 * location before possibly journalling it again which
2525 * is desirable when the page is frequently dirtied
2526 * through a pin.
2527 */
2528 if (!mpd->can_map) {
2529 err = mpage_submit_folio(mpd, folio);
2530 if (err < 0)
2531 goto out;
2532 /* Pending dirtying of journalled data? */
2533 if (folio_test_checked(folio)) {
2534 err = mpage_journal_page_buffers(handle,
2535 mpd, folio);
2536 if (err < 0)
2537 goto out;
2538 mpd->journalled_more_data = 1;
2539 }
2540 mpage_folio_done(mpd, folio);
2541 } else {
2542 /* Add all dirty buffers to mpd */
2543 lblk = ((ext4_lblk_t)folio->index) <<
2544 (PAGE_SHIFT - blkbits);
2545 head = folio_buffers(folio);
2546 err = mpage_process_page_bufs(mpd, head, head,
2547 lblk);
2548 if (err <= 0)
2549 goto out;
2550 err = 0;
2551 }
2552 }
2553 folio_batch_release(&fbatch);
2554 cond_resched();
2555 }
2556 mpd->scanned_until_end = 1;
2557 if (handle)
2558 ext4_journal_stop(handle);
2559 return 0;
2560out:
2561 folio_batch_release(&fbatch);
2562 if (handle)
2563 ext4_journal_stop(handle);
2564 return err;
2565}
2566
2567static int ext4_do_writepages(struct mpage_da_data *mpd)
2568{
2569 struct writeback_control *wbc = mpd->wbc;
2570 pgoff_t writeback_index = 0;
2571 long nr_to_write = wbc->nr_to_write;
2572 int range_whole = 0;
2573 int cycled = 1;
2574 handle_t *handle = NULL;
2575 struct inode *inode = mpd->inode;
2576 struct address_space *mapping = inode->i_mapping;
2577 int needed_blocks, rsv_blocks = 0, ret = 0;
2578 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2579 struct blk_plug plug;
2580 bool give_up_on_write = false;
2581
2582 trace_ext4_writepages(inode, wbc);
2583
2584 /*
2585 * No pages to write? This is mainly a kludge to avoid starting
2586 * a transaction for special inodes like journal inode on last iput()
2587 * because that could violate lock ordering on umount
2588 */
2589 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2590 goto out_writepages;
2591
2592 /*
2593 * If the filesystem has aborted, it is read-only, so return
2594 * right away instead of dumping stack traces later on that
2595 * will obscure the real source of the problem. We test
2596 * fs shutdown state instead of sb->s_flag's SB_RDONLY because
2597 * the latter could be true if the filesystem is mounted
2598 * read-only, and in that case, ext4_writepages should
2599 * *never* be called, so if that ever happens, we would want
2600 * the stack trace.
2601 */
2602 if (unlikely(ext4_forced_shutdown(mapping->host->i_sb))) {
2603 ret = -EROFS;
2604 goto out_writepages;
2605 }
2606
2607 /*
2608 * If we have inline data and arrive here, it means that
2609 * we will soon create the block for the 1st page, so
2610 * we'd better clear the inline data here.
2611 */
2612 if (ext4_has_inline_data(inode)) {
2613 /* Just inode will be modified... */
2614 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2615 if (IS_ERR(handle)) {
2616 ret = PTR_ERR(handle);
2617 goto out_writepages;
2618 }
2619 BUG_ON(ext4_test_inode_state(inode,
2620 EXT4_STATE_MAY_INLINE_DATA));
2621 ext4_destroy_inline_data(handle, inode);
2622 ext4_journal_stop(handle);
2623 }
2624
2625 /*
2626 * data=journal mode does not do delalloc so we just need to writeout /
2627 * journal already mapped buffers. On the other hand we need to commit
2628 * transaction to make data stable. We expect all the data to be
2629 * already in the journal (the only exception are DMA pinned pages
2630 * dirtied behind our back) so we commit transaction here and run the
2631 * writeback loop to checkpoint them. The checkpointing is not actually
2632 * necessary to make data persistent *but* quite a few places (extent
2633 * shifting operations, fsverity, ...) depend on being able to drop
2634 * pagecache pages after calling filemap_write_and_wait() and for that
2635 * checkpointing needs to happen.
2636 */
2637 if (ext4_should_journal_data(inode)) {
2638 mpd->can_map = 0;
2639 if (wbc->sync_mode == WB_SYNC_ALL)
2640 ext4_fc_commit(sbi->s_journal,
2641 EXT4_I(inode)->i_datasync_tid);
2642 }
2643 mpd->journalled_more_data = 0;
2644
2645 if (ext4_should_dioread_nolock(inode)) {
2646 /*
2647 * We may need to convert up to one extent per block in
2648 * the page and we may dirty the inode.
2649 */
2650 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2651 PAGE_SIZE >> inode->i_blkbits);
2652 }
2653
2654 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2655 range_whole = 1;
2656
2657 if (wbc->range_cyclic) {
2658 writeback_index = mapping->writeback_index;
2659 if (writeback_index)
2660 cycled = 0;
2661 mpd->first_page = writeback_index;
2662 mpd->last_page = -1;
2663 } else {
2664 mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2665 mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2666 }
2667
2668 ext4_io_submit_init(&mpd->io_submit, wbc);
2669retry:
2670 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2671 tag_pages_for_writeback(mapping, mpd->first_page,
2672 mpd->last_page);
2673 blk_start_plug(&plug);
2674
2675 /*
2676 * First writeback pages that don't need mapping - we can avoid
2677 * starting a transaction unnecessarily and also avoid being blocked
2678 * in the block layer on device congestion while having transaction
2679 * started.
2680 */
2681 mpd->do_map = 0;
2682 mpd->scanned_until_end = 0;
2683 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2684 if (!mpd->io_submit.io_end) {
2685 ret = -ENOMEM;
2686 goto unplug;
2687 }
2688 ret = mpage_prepare_extent_to_map(mpd);
2689 /* Unlock pages we didn't use */
2690 mpage_release_unused_pages(mpd, false);
2691 /* Submit prepared bio */
2692 ext4_io_submit(&mpd->io_submit);
2693 ext4_put_io_end_defer(mpd->io_submit.io_end);
2694 mpd->io_submit.io_end = NULL;
2695 if (ret < 0)
2696 goto unplug;
2697
2698 while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2699 /* For each extent of pages we use new io_end */
2700 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2701 if (!mpd->io_submit.io_end) {
2702 ret = -ENOMEM;
2703 break;
2704 }
2705
2706 WARN_ON_ONCE(!mpd->can_map);
2707 /*
2708 * We have two constraints: We find one extent to map and we
2709 * must always write out whole page (makes a difference when
2710 * blocksize < pagesize) so that we don't block on IO when we
2711 * try to write out the rest of the page. Journalled mode is
2712 * not supported by delalloc.
2713 */
2714 BUG_ON(ext4_should_journal_data(inode));
2715 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2716
2717 /* start a new transaction */
2718 handle = ext4_journal_start_with_reserve(inode,
2719 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2720 if (IS_ERR(handle)) {
2721 ret = PTR_ERR(handle);
2722 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2723 "%ld pages, ino %lu; err %d", __func__,
2724 wbc->nr_to_write, inode->i_ino, ret);
2725 /* Release allocated io_end */
2726 ext4_put_io_end(mpd->io_submit.io_end);
2727 mpd->io_submit.io_end = NULL;
2728 break;
2729 }
2730 mpd->do_map = 1;
2731
2732 trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2733 ret = mpage_prepare_extent_to_map(mpd);
2734 if (!ret && mpd->map.m_len)
2735 ret = mpage_map_and_submit_extent(handle, mpd,
2736 &give_up_on_write);
2737 /*
2738 * Caution: If the handle is synchronous,
2739 * ext4_journal_stop() can wait for transaction commit
2740 * to finish which may depend on writeback of pages to
2741 * complete or on page lock to be released. In that
2742 * case, we have to wait until after we have
2743 * submitted all the IO, released page locks we hold,
2744 * and dropped io_end reference (for extent conversion
2745 * to be able to complete) before stopping the handle.
2746 */
2747 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2748 ext4_journal_stop(handle);
2749 handle = NULL;
2750 mpd->do_map = 0;
2751 }
2752 /* Unlock pages we didn't use */
2753 mpage_release_unused_pages(mpd, give_up_on_write);
2754 /* Submit prepared bio */
2755 ext4_io_submit(&mpd->io_submit);
2756
2757 /*
2758 * Drop our io_end reference we got from init. We have
2759 * to be careful and use deferred io_end finishing if
2760 * we are still holding the transaction as we can
2761 * release the last reference to io_end which may end
2762 * up doing unwritten extent conversion.
2763 */
2764 if (handle) {
2765 ext4_put_io_end_defer(mpd->io_submit.io_end);
2766 ext4_journal_stop(handle);
2767 } else
2768 ext4_put_io_end(mpd->io_submit.io_end);
2769 mpd->io_submit.io_end = NULL;
2770
2771 if (ret == -ENOSPC && sbi->s_journal) {
2772 /*
2773 * Commit the transaction which would
2774 * free blocks released in the transaction
2775 * and try again
2776 */
2777 jbd2_journal_force_commit_nested(sbi->s_journal);
2778 ret = 0;
2779 continue;
2780 }
2781 /* Fatal error - ENOMEM, EIO... */
2782 if (ret)
2783 break;
2784 }
2785unplug:
2786 blk_finish_plug(&plug);
2787 if (!ret && !cycled && wbc->nr_to_write > 0) {
2788 cycled = 1;
2789 mpd->last_page = writeback_index - 1;
2790 mpd->first_page = 0;
2791 goto retry;
2792 }
2793
2794 /* Update index */
2795 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2796 /*
2797 * Set the writeback_index so that range_cyclic
2798 * mode will write it back later
2799 */
2800 mapping->writeback_index = mpd->first_page;
2801
2802out_writepages:
2803 trace_ext4_writepages_result(inode, wbc, ret,
2804 nr_to_write - wbc->nr_to_write);
2805 return ret;
2806}
2807
2808static int ext4_writepages(struct address_space *mapping,
2809 struct writeback_control *wbc)
2810{
2811 struct super_block *sb = mapping->host->i_sb;
2812 struct mpage_da_data mpd = {
2813 .inode = mapping->host,
2814 .wbc = wbc,
2815 .can_map = 1,
2816 };
2817 int ret;
2818 int alloc_ctx;
2819
2820 if (unlikely(ext4_forced_shutdown(sb)))
2821 return -EIO;
2822
2823 alloc_ctx = ext4_writepages_down_read(sb);
2824 ret = ext4_do_writepages(&mpd);
2825 /*
2826 * For data=journal writeback we could have come across pages marked
2827 * for delayed dirtying (PageChecked) which were just added to the
2828 * running transaction. Try once more to get them to stable storage.
2829 */
2830 if (!ret && mpd.journalled_more_data)
2831 ret = ext4_do_writepages(&mpd);
2832 ext4_writepages_up_read(sb, alloc_ctx);
2833
2834 return ret;
2835}
2836
2837int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2838{
2839 struct writeback_control wbc = {
2840 .sync_mode = WB_SYNC_ALL,
2841 .nr_to_write = LONG_MAX,
2842 .range_start = jinode->i_dirty_start,
2843 .range_end = jinode->i_dirty_end,
2844 };
2845 struct mpage_da_data mpd = {
2846 .inode = jinode->i_vfs_inode,
2847 .wbc = &wbc,
2848 .can_map = 0,
2849 };
2850 return ext4_do_writepages(&mpd);
2851}
2852
2853static int ext4_dax_writepages(struct address_space *mapping,
2854 struct writeback_control *wbc)
2855{
2856 int ret;
2857 long nr_to_write = wbc->nr_to_write;
2858 struct inode *inode = mapping->host;
2859 int alloc_ctx;
2860
2861 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2862 return -EIO;
2863
2864 alloc_ctx = ext4_writepages_down_read(inode->i_sb);
2865 trace_ext4_writepages(inode, wbc);
2866
2867 ret = dax_writeback_mapping_range(mapping,
2868 EXT4_SB(inode->i_sb)->s_daxdev, wbc);
2869 trace_ext4_writepages_result(inode, wbc, ret,
2870 nr_to_write - wbc->nr_to_write);
2871 ext4_writepages_up_read(inode->i_sb, alloc_ctx);
2872 return ret;
2873}
2874
2875static int ext4_nonda_switch(struct super_block *sb)
2876{
2877 s64 free_clusters, dirty_clusters;
2878 struct ext4_sb_info *sbi = EXT4_SB(sb);
2879
2880 /*
2881 * switch to non delalloc mode if we are running low
2882 * on free block. The free block accounting via percpu
2883 * counters can get slightly wrong with percpu_counter_batch getting
2884 * accumulated on each CPU without updating global counters
2885 * Delalloc need an accurate free block accounting. So switch
2886 * to non delalloc when we are near to error range.
2887 */
2888 free_clusters =
2889 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2890 dirty_clusters =
2891 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2892 /*
2893 * Start pushing delalloc when 1/2 of free blocks are dirty.
2894 */
2895 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2896 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2897
2898 if (2 * free_clusters < 3 * dirty_clusters ||
2899 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2900 /*
2901 * free block count is less than 150% of dirty blocks
2902 * or free blocks is less than watermark
2903 */
2904 return 1;
2905 }
2906 return 0;
2907}
2908
2909static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2910 loff_t pos, unsigned len,
2911 struct folio **foliop, void **fsdata)
2912{
2913 int ret, retries = 0;
2914 struct folio *folio;
2915 pgoff_t index;
2916 struct inode *inode = mapping->host;
2917
2918 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2919 return -EIO;
2920
2921 index = pos >> PAGE_SHIFT;
2922
2923 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2924 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2925 return ext4_write_begin(file, mapping, pos,
2926 len, foliop, fsdata);
2927 }
2928 *fsdata = (void *)0;
2929 trace_ext4_da_write_begin(inode, pos, len);
2930
2931 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2932 ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
2933 foliop, fsdata);
2934 if (ret < 0)
2935 return ret;
2936 if (ret == 1)
2937 return 0;
2938 }
2939
2940retry:
2941 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
2942 mapping_gfp_mask(mapping));
2943 if (IS_ERR(folio))
2944 return PTR_ERR(folio);
2945
2946 ret = ext4_block_write_begin(NULL, folio, pos, len,
2947 ext4_da_get_block_prep);
2948 if (ret < 0) {
2949 folio_unlock(folio);
2950 folio_put(folio);
2951 /*
2952 * block_write_begin may have instantiated a few blocks
2953 * outside i_size. Trim these off again. Don't need
2954 * i_size_read because we hold inode lock.
2955 */
2956 if (pos + len > inode->i_size)
2957 ext4_truncate_failed_write(inode);
2958
2959 if (ret == -ENOSPC &&
2960 ext4_should_retry_alloc(inode->i_sb, &retries))
2961 goto retry;
2962 return ret;
2963 }
2964
2965 *foliop = folio;
2966 return ret;
2967}
2968
2969/*
2970 * Check if we should update i_disksize
2971 * when write to the end of file but not require block allocation
2972 */
2973static int ext4_da_should_update_i_disksize(struct folio *folio,
2974 unsigned long offset)
2975{
2976 struct buffer_head *bh;
2977 struct inode *inode = folio->mapping->host;
2978 unsigned int idx;
2979 int i;
2980
2981 bh = folio_buffers(folio);
2982 idx = offset >> inode->i_blkbits;
2983
2984 for (i = 0; i < idx; i++)
2985 bh = bh->b_this_page;
2986
2987 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2988 return 0;
2989 return 1;
2990}
2991
2992static int ext4_da_do_write_end(struct address_space *mapping,
2993 loff_t pos, unsigned len, unsigned copied,
2994 struct folio *folio)
2995{
2996 struct inode *inode = mapping->host;
2997 loff_t old_size = inode->i_size;
2998 bool disksize_changed = false;
2999 loff_t new_i_size, zero_len = 0;
3000 handle_t *handle;
3001
3002 if (unlikely(!folio_buffers(folio))) {
3003 folio_unlock(folio);
3004 folio_put(folio);
3005 return -EIO;
3006 }
3007 /*
3008 * block_write_end() will mark the inode as dirty with I_DIRTY_PAGES
3009 * flag, which all that's needed to trigger page writeback.
3010 */
3011 copied = block_write_end(NULL, mapping, pos, len, copied,
3012 folio, NULL);
3013 new_i_size = pos + copied;
3014
3015 /*
3016 * It's important to update i_size while still holding folio lock,
3017 * because folio writeout could otherwise come in and zero beyond
3018 * i_size.
3019 *
3020 * Since we are holding inode lock, we are sure i_disksize <=
3021 * i_size. We also know that if i_disksize < i_size, there are
3022 * delalloc writes pending in the range up to i_size. If the end of
3023 * the current write is <= i_size, there's no need to touch
3024 * i_disksize since writeback will push i_disksize up to i_size
3025 * eventually. If the end of the current write is > i_size and
3026 * inside an allocated block which ext4_da_should_update_i_disksize()
3027 * checked, we need to update i_disksize here as certain
3028 * ext4_writepages() paths not allocating blocks and update i_disksize.
3029 */
3030 if (new_i_size > inode->i_size) {
3031 unsigned long end;
3032
3033 i_size_write(inode, new_i_size);
3034 end = (new_i_size - 1) & (PAGE_SIZE - 1);
3035 if (copied && ext4_da_should_update_i_disksize(folio, end)) {
3036 ext4_update_i_disksize(inode, new_i_size);
3037 disksize_changed = true;
3038 }
3039 }
3040
3041 folio_unlock(folio);
3042 folio_put(folio);
3043
3044 if (pos > old_size) {
3045 pagecache_isize_extended(inode, old_size, pos);
3046 zero_len = pos - old_size;
3047 }
3048
3049 if (!disksize_changed && !zero_len)
3050 return copied;
3051
3052 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3053 if (IS_ERR(handle))
3054 return PTR_ERR(handle);
3055 if (zero_len)
3056 ext4_zero_partial_blocks(handle, inode, old_size, zero_len);
3057 ext4_mark_inode_dirty(handle, inode);
3058 ext4_journal_stop(handle);
3059
3060 return copied;
3061}
3062
3063static int ext4_da_write_end(struct file *file,
3064 struct address_space *mapping,
3065 loff_t pos, unsigned len, unsigned copied,
3066 struct folio *folio, void *fsdata)
3067{
3068 struct inode *inode = mapping->host;
3069 int write_mode = (int)(unsigned long)fsdata;
3070
3071 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3072 return ext4_write_end(file, mapping, pos,
3073 len, copied, folio, fsdata);
3074
3075 trace_ext4_da_write_end(inode, pos, len, copied);
3076
3077 if (write_mode != CONVERT_INLINE_DATA &&
3078 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3079 ext4_has_inline_data(inode))
3080 return ext4_write_inline_data_end(inode, pos, len, copied,
3081 folio);
3082
3083 if (unlikely(copied < len) && !folio_test_uptodate(folio))
3084 copied = 0;
3085
3086 return ext4_da_do_write_end(mapping, pos, len, copied, folio);
3087}
3088
3089/*
3090 * Force all delayed allocation blocks to be allocated for a given inode.
3091 */
3092int ext4_alloc_da_blocks(struct inode *inode)
3093{
3094 trace_ext4_alloc_da_blocks(inode);
3095
3096 if (!EXT4_I(inode)->i_reserved_data_blocks)
3097 return 0;
3098
3099 /*
3100 * We do something simple for now. The filemap_flush() will
3101 * also start triggering a write of the data blocks, which is
3102 * not strictly speaking necessary (and for users of
3103 * laptop_mode, not even desirable). However, to do otherwise
3104 * would require replicating code paths in:
3105 *
3106 * ext4_writepages() ->
3107 * write_cache_pages() ---> (via passed in callback function)
3108 * __mpage_da_writepage() -->
3109 * mpage_add_bh_to_extent()
3110 * mpage_da_map_blocks()
3111 *
3112 * The problem is that write_cache_pages(), located in
3113 * mm/page-writeback.c, marks pages clean in preparation for
3114 * doing I/O, which is not desirable if we're not planning on
3115 * doing I/O at all.
3116 *
3117 * We could call write_cache_pages(), and then redirty all of
3118 * the pages by calling redirty_page_for_writepage() but that
3119 * would be ugly in the extreme. So instead we would need to
3120 * replicate parts of the code in the above functions,
3121 * simplifying them because we wouldn't actually intend to
3122 * write out the pages, but rather only collect contiguous
3123 * logical block extents, call the multi-block allocator, and
3124 * then update the buffer heads with the block allocations.
3125 *
3126 * For now, though, we'll cheat by calling filemap_flush(),
3127 * which will map the blocks, and start the I/O, but not
3128 * actually wait for the I/O to complete.
3129 */
3130 return filemap_flush(inode->i_mapping);
3131}
3132
3133/*
3134 * bmap() is special. It gets used by applications such as lilo and by
3135 * the swapper to find the on-disk block of a specific piece of data.
3136 *
3137 * Naturally, this is dangerous if the block concerned is still in the
3138 * journal. If somebody makes a swapfile on an ext4 data-journaling
3139 * filesystem and enables swap, then they may get a nasty shock when the
3140 * data getting swapped to that swapfile suddenly gets overwritten by
3141 * the original zero's written out previously to the journal and
3142 * awaiting writeback in the kernel's buffer cache.
3143 *
3144 * So, if we see any bmap calls here on a modified, data-journaled file,
3145 * take extra steps to flush any blocks which might be in the cache.
3146 */
3147static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3148{
3149 struct inode *inode = mapping->host;
3150 sector_t ret = 0;
3151
3152 inode_lock_shared(inode);
3153 /*
3154 * We can get here for an inline file via the FIBMAP ioctl
3155 */
3156 if (ext4_has_inline_data(inode))
3157 goto out;
3158
3159 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3160 (test_opt(inode->i_sb, DELALLOC) ||
3161 ext4_should_journal_data(inode))) {
3162 /*
3163 * With delalloc or journalled data we want to sync the file so
3164 * that we can make sure we allocate blocks for file and data
3165 * is in place for the user to see it
3166 */
3167 filemap_write_and_wait(mapping);
3168 }
3169
3170 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3171
3172out:
3173 inode_unlock_shared(inode);
3174 return ret;
3175}
3176
3177static int ext4_read_folio(struct file *file, struct folio *folio)
3178{
3179 int ret = -EAGAIN;
3180 struct inode *inode = folio->mapping->host;
3181
3182 trace_ext4_read_folio(inode, folio);
3183
3184 if (ext4_has_inline_data(inode))
3185 ret = ext4_readpage_inline(inode, folio);
3186
3187 if (ret == -EAGAIN)
3188 return ext4_mpage_readpages(inode, NULL, folio);
3189
3190 return ret;
3191}
3192
3193static void ext4_readahead(struct readahead_control *rac)
3194{
3195 struct inode *inode = rac->mapping->host;
3196
3197 /* If the file has inline data, no need to do readahead. */
3198 if (ext4_has_inline_data(inode))
3199 return;
3200
3201 ext4_mpage_readpages(inode, rac, NULL);
3202}
3203
3204static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3205 size_t length)
3206{
3207 trace_ext4_invalidate_folio(folio, offset, length);
3208
3209 /* No journalling happens on data buffers when this function is used */
3210 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3211
3212 block_invalidate_folio(folio, offset, length);
3213}
3214
3215static int __ext4_journalled_invalidate_folio(struct folio *folio,
3216 size_t offset, size_t length)
3217{
3218 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3219
3220 trace_ext4_journalled_invalidate_folio(folio, offset, length);
3221
3222 /*
3223 * If it's a full truncate we just forget about the pending dirtying
3224 */
3225 if (offset == 0 && length == folio_size(folio))
3226 folio_clear_checked(folio);
3227
3228 return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3229}
3230
3231/* Wrapper for aops... */
3232static void ext4_journalled_invalidate_folio(struct folio *folio,
3233 size_t offset,
3234 size_t length)
3235{
3236 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3237}
3238
3239static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3240{
3241 struct inode *inode = folio->mapping->host;
3242 journal_t *journal = EXT4_JOURNAL(inode);
3243
3244 trace_ext4_release_folio(inode, folio);
3245
3246 /* Page has dirty journalled data -> cannot release */
3247 if (folio_test_checked(folio))
3248 return false;
3249 if (journal)
3250 return jbd2_journal_try_to_free_buffers(journal, folio);
3251 else
3252 return try_to_free_buffers(folio);
3253}
3254
3255static bool ext4_inode_datasync_dirty(struct inode *inode)
3256{
3257 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3258
3259 if (journal) {
3260 if (jbd2_transaction_committed(journal,
3261 EXT4_I(inode)->i_datasync_tid))
3262 return false;
3263 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3264 return !list_empty(&EXT4_I(inode)->i_fc_list);
3265 return true;
3266 }
3267
3268 /* Any metadata buffers to write? */
3269 if (!list_empty(&inode->i_mapping->i_private_list))
3270 return true;
3271 return inode->i_state & I_DIRTY_DATASYNC;
3272}
3273
3274static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3275 struct ext4_map_blocks *map, loff_t offset,
3276 loff_t length, unsigned int flags)
3277{
3278 u8 blkbits = inode->i_blkbits;
3279
3280 /*
3281 * Writes that span EOF might trigger an I/O size update on completion,
3282 * so consider them to be dirty for the purpose of O_DSYNC, even if
3283 * there is no other metadata changes being made or are pending.
3284 */
3285 iomap->flags = 0;
3286 if (ext4_inode_datasync_dirty(inode) ||
3287 offset + length > i_size_read(inode))
3288 iomap->flags |= IOMAP_F_DIRTY;
3289
3290 if (map->m_flags & EXT4_MAP_NEW)
3291 iomap->flags |= IOMAP_F_NEW;
3292
3293 if (flags & IOMAP_DAX)
3294 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3295 else
3296 iomap->bdev = inode->i_sb->s_bdev;
3297 iomap->offset = (u64) map->m_lblk << blkbits;
3298 iomap->length = (u64) map->m_len << blkbits;
3299
3300 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3301 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3302 iomap->flags |= IOMAP_F_MERGED;
3303
3304 /*
3305 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3306 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3307 * set. In order for any allocated unwritten extents to be converted
3308 * into written extents correctly within the ->end_io() handler, we
3309 * need to ensure that the iomap->type is set appropriately. Hence, the
3310 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3311 * been set first.
3312 */
3313 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3314 iomap->type = IOMAP_UNWRITTEN;
3315 iomap->addr = (u64) map->m_pblk << blkbits;
3316 if (flags & IOMAP_DAX)
3317 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3318 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3319 iomap->type = IOMAP_MAPPED;
3320 iomap->addr = (u64) map->m_pblk << blkbits;
3321 if (flags & IOMAP_DAX)
3322 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3323 } else if (map->m_flags & EXT4_MAP_DELAYED) {
3324 iomap->type = IOMAP_DELALLOC;
3325 iomap->addr = IOMAP_NULL_ADDR;
3326 } else {
3327 iomap->type = IOMAP_HOLE;
3328 iomap->addr = IOMAP_NULL_ADDR;
3329 }
3330}
3331
3332static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3333 unsigned int flags)
3334{
3335 handle_t *handle;
3336 u8 blkbits = inode->i_blkbits;
3337 int ret, dio_credits, m_flags = 0, retries = 0;
3338
3339 /*
3340 * Trim the mapping request to the maximum value that we can map at
3341 * once for direct I/O.
3342 */
3343 if (map->m_len > DIO_MAX_BLOCKS)
3344 map->m_len = DIO_MAX_BLOCKS;
3345 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3346
3347retry:
3348 /*
3349 * Either we allocate blocks and then don't get an unwritten extent, so
3350 * in that case we have reserved enough credits. Or, the blocks are
3351 * already allocated and unwritten. In that case, the extent conversion
3352 * fits into the credits as well.
3353 */
3354 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3355 if (IS_ERR(handle))
3356 return PTR_ERR(handle);
3357
3358 /*
3359 * DAX and direct I/O are the only two operations that are currently
3360 * supported with IOMAP_WRITE.
3361 */
3362 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3363 if (flags & IOMAP_DAX)
3364 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3365 /*
3366 * We use i_size instead of i_disksize here because delalloc writeback
3367 * can complete at any point during the I/O and subsequently push the
3368 * i_disksize out to i_size. This could be beyond where direct I/O is
3369 * happening and thus expose allocated blocks to direct I/O reads.
3370 */
3371 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3372 m_flags = EXT4_GET_BLOCKS_CREATE;
3373 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3374 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3375
3376 ret = ext4_map_blocks(handle, inode, map, m_flags);
3377
3378 /*
3379 * We cannot fill holes in indirect tree based inodes as that could
3380 * expose stale data in the case of a crash. Use the magic error code
3381 * to fallback to buffered I/O.
3382 */
3383 if (!m_flags && !ret)
3384 ret = -ENOTBLK;
3385
3386 ext4_journal_stop(handle);
3387 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3388 goto retry;
3389
3390 return ret;
3391}
3392
3393
3394static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3395 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3396{
3397 int ret;
3398 struct ext4_map_blocks map;
3399 u8 blkbits = inode->i_blkbits;
3400
3401 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3402 return -EINVAL;
3403
3404 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3405 return -ERANGE;
3406
3407 /*
3408 * Calculate the first and last logical blocks respectively.
3409 */
3410 map.m_lblk = offset >> blkbits;
3411 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3412 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3413
3414 if (flags & IOMAP_WRITE) {
3415 /*
3416 * We check here if the blocks are already allocated, then we
3417 * don't need to start a journal txn and we can directly return
3418 * the mapping information. This could boost performance
3419 * especially in multi-threaded overwrite requests.
3420 */
3421 if (offset + length <= i_size_read(inode)) {
3422 ret = ext4_map_blocks(NULL, inode, &map, 0);
3423 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3424 goto out;
3425 }
3426 ret = ext4_iomap_alloc(inode, &map, flags);
3427 } else {
3428 ret = ext4_map_blocks(NULL, inode, &map, 0);
3429 }
3430
3431 if (ret < 0)
3432 return ret;
3433out:
3434 /*
3435 * When inline encryption is enabled, sometimes I/O to an encrypted file
3436 * has to be broken up to guarantee DUN contiguity. Handle this by
3437 * limiting the length of the mapping returned.
3438 */
3439 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3440
3441 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3442
3443 return 0;
3444}
3445
3446static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3447 loff_t length, unsigned flags, struct iomap *iomap,
3448 struct iomap *srcmap)
3449{
3450 int ret;
3451
3452 /*
3453 * Even for writes we don't need to allocate blocks, so just pretend
3454 * we are reading to save overhead of starting a transaction.
3455 */
3456 flags &= ~IOMAP_WRITE;
3457 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3458 WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
3459 return ret;
3460}
3461
3462static inline bool ext4_want_directio_fallback(unsigned flags, ssize_t written)
3463{
3464 /* must be a directio to fall back to buffered */
3465 if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) !=
3466 (IOMAP_WRITE | IOMAP_DIRECT))
3467 return false;
3468
3469 /* atomic writes are all-or-nothing */
3470 if (flags & IOMAP_ATOMIC)
3471 return false;
3472
3473 /* can only try again if we wrote nothing */
3474 return written == 0;
3475}
3476
3477static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3478 ssize_t written, unsigned flags, struct iomap *iomap)
3479{
3480 /*
3481 * Check to see whether an error occurred while writing out the data to
3482 * the allocated blocks. If so, return the magic error code for
3483 * non-atomic write so that we fallback to buffered I/O and attempt to
3484 * complete the remainder of the I/O.
3485 * For non-atomic writes, any blocks that may have been
3486 * allocated in preparation for the direct I/O will be reused during
3487 * buffered I/O. For atomic write, we never fallback to buffered-io.
3488 */
3489 if (ext4_want_directio_fallback(flags, written))
3490 return -ENOTBLK;
3491
3492 return 0;
3493}
3494
3495const struct iomap_ops ext4_iomap_ops = {
3496 .iomap_begin = ext4_iomap_begin,
3497 .iomap_end = ext4_iomap_end,
3498};
3499
3500const struct iomap_ops ext4_iomap_overwrite_ops = {
3501 .iomap_begin = ext4_iomap_overwrite_begin,
3502 .iomap_end = ext4_iomap_end,
3503};
3504
3505static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3506 loff_t length, unsigned int flags,
3507 struct iomap *iomap, struct iomap *srcmap)
3508{
3509 int ret;
3510 struct ext4_map_blocks map;
3511 u8 blkbits = inode->i_blkbits;
3512
3513 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3514 return -EINVAL;
3515
3516 if (ext4_has_inline_data(inode)) {
3517 ret = ext4_inline_data_iomap(inode, iomap);
3518 if (ret != -EAGAIN) {
3519 if (ret == 0 && offset >= iomap->length)
3520 ret = -ENOENT;
3521 return ret;
3522 }
3523 }
3524
3525 /*
3526 * Calculate the first and last logical block respectively.
3527 */
3528 map.m_lblk = offset >> blkbits;
3529 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3530 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3531
3532 /*
3533 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3534 * So handle it here itself instead of querying ext4_map_blocks().
3535 * Since ext4_map_blocks() will warn about it and will return
3536 * -EIO error.
3537 */
3538 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3539 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3540
3541 if (offset >= sbi->s_bitmap_maxbytes) {
3542 map.m_flags = 0;
3543 goto set_iomap;
3544 }
3545 }
3546
3547 ret = ext4_map_blocks(NULL, inode, &map, 0);
3548 if (ret < 0)
3549 return ret;
3550set_iomap:
3551 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3552
3553 return 0;
3554}
3555
3556const struct iomap_ops ext4_iomap_report_ops = {
3557 .iomap_begin = ext4_iomap_begin_report,
3558};
3559
3560/*
3561 * For data=journal mode, folio should be marked dirty only when it was
3562 * writeably mapped. When that happens, it was already attached to the
3563 * transaction and marked as jbddirty (we take care of this in
3564 * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3565 * so we should have nothing to do here, except for the case when someone
3566 * had the page pinned and dirtied the page through this pin (e.g. by doing
3567 * direct IO to it). In that case we'd need to attach buffers here to the
3568 * transaction but we cannot due to lock ordering. We cannot just dirty the
3569 * folio and leave attached buffers clean, because the buffers' dirty state is
3570 * "definitive". We cannot just set the buffers dirty or jbddirty because all
3571 * the journalling code will explode. So what we do is to mark the folio
3572 * "pending dirty" and next time ext4_writepages() is called, attach buffers
3573 * to the transaction appropriately.
3574 */
3575static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3576 struct folio *folio)
3577{
3578 WARN_ON_ONCE(!folio_buffers(folio));
3579 if (folio_maybe_dma_pinned(folio))
3580 folio_set_checked(folio);
3581 return filemap_dirty_folio(mapping, folio);
3582}
3583
3584static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3585{
3586 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3587 WARN_ON_ONCE(!folio_buffers(folio));
3588 return block_dirty_folio(mapping, folio);
3589}
3590
3591static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3592 struct file *file, sector_t *span)
3593{
3594 return iomap_swapfile_activate(sis, file, span,
3595 &ext4_iomap_report_ops);
3596}
3597
3598static const struct address_space_operations ext4_aops = {
3599 .read_folio = ext4_read_folio,
3600 .readahead = ext4_readahead,
3601 .writepages = ext4_writepages,
3602 .write_begin = ext4_write_begin,
3603 .write_end = ext4_write_end,
3604 .dirty_folio = ext4_dirty_folio,
3605 .bmap = ext4_bmap,
3606 .invalidate_folio = ext4_invalidate_folio,
3607 .release_folio = ext4_release_folio,
3608 .migrate_folio = buffer_migrate_folio,
3609 .is_partially_uptodate = block_is_partially_uptodate,
3610 .error_remove_folio = generic_error_remove_folio,
3611 .swap_activate = ext4_iomap_swap_activate,
3612};
3613
3614static const struct address_space_operations ext4_journalled_aops = {
3615 .read_folio = ext4_read_folio,
3616 .readahead = ext4_readahead,
3617 .writepages = ext4_writepages,
3618 .write_begin = ext4_write_begin,
3619 .write_end = ext4_journalled_write_end,
3620 .dirty_folio = ext4_journalled_dirty_folio,
3621 .bmap = ext4_bmap,
3622 .invalidate_folio = ext4_journalled_invalidate_folio,
3623 .release_folio = ext4_release_folio,
3624 .migrate_folio = buffer_migrate_folio_norefs,
3625 .is_partially_uptodate = block_is_partially_uptodate,
3626 .error_remove_folio = generic_error_remove_folio,
3627 .swap_activate = ext4_iomap_swap_activate,
3628};
3629
3630static const struct address_space_operations ext4_da_aops = {
3631 .read_folio = ext4_read_folio,
3632 .readahead = ext4_readahead,
3633 .writepages = ext4_writepages,
3634 .write_begin = ext4_da_write_begin,
3635 .write_end = ext4_da_write_end,
3636 .dirty_folio = ext4_dirty_folio,
3637 .bmap = ext4_bmap,
3638 .invalidate_folio = ext4_invalidate_folio,
3639 .release_folio = ext4_release_folio,
3640 .migrate_folio = buffer_migrate_folio,
3641 .is_partially_uptodate = block_is_partially_uptodate,
3642 .error_remove_folio = generic_error_remove_folio,
3643 .swap_activate = ext4_iomap_swap_activate,
3644};
3645
3646static const struct address_space_operations ext4_dax_aops = {
3647 .writepages = ext4_dax_writepages,
3648 .dirty_folio = noop_dirty_folio,
3649 .bmap = ext4_bmap,
3650 .swap_activate = ext4_iomap_swap_activate,
3651};
3652
3653void ext4_set_aops(struct inode *inode)
3654{
3655 switch (ext4_inode_journal_mode(inode)) {
3656 case EXT4_INODE_ORDERED_DATA_MODE:
3657 case EXT4_INODE_WRITEBACK_DATA_MODE:
3658 break;
3659 case EXT4_INODE_JOURNAL_DATA_MODE:
3660 inode->i_mapping->a_ops = &ext4_journalled_aops;
3661 return;
3662 default:
3663 BUG();
3664 }
3665 if (IS_DAX(inode))
3666 inode->i_mapping->a_ops = &ext4_dax_aops;
3667 else if (test_opt(inode->i_sb, DELALLOC))
3668 inode->i_mapping->a_ops = &ext4_da_aops;
3669 else
3670 inode->i_mapping->a_ops = &ext4_aops;
3671}
3672
3673/*
3674 * Here we can't skip an unwritten buffer even though it usually reads zero
3675 * because it might have data in pagecache (eg, if called from ext4_zero_range,
3676 * ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a
3677 * racing writeback can come later and flush the stale pagecache to disk.
3678 */
3679static int __ext4_block_zero_page_range(handle_t *handle,
3680 struct address_space *mapping, loff_t from, loff_t length)
3681{
3682 ext4_fsblk_t index = from >> PAGE_SHIFT;
3683 unsigned offset = from & (PAGE_SIZE-1);
3684 unsigned blocksize, pos;
3685 ext4_lblk_t iblock;
3686 struct inode *inode = mapping->host;
3687 struct buffer_head *bh;
3688 struct folio *folio;
3689 int err = 0;
3690
3691 folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT,
3692 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3693 mapping_gfp_constraint(mapping, ~__GFP_FS));
3694 if (IS_ERR(folio))
3695 return PTR_ERR(folio);
3696
3697 blocksize = inode->i_sb->s_blocksize;
3698
3699 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3700
3701 bh = folio_buffers(folio);
3702 if (!bh)
3703 bh = create_empty_buffers(folio, blocksize, 0);
3704
3705 /* Find the buffer that contains "offset" */
3706 pos = blocksize;
3707 while (offset >= pos) {
3708 bh = bh->b_this_page;
3709 iblock++;
3710 pos += blocksize;
3711 }
3712 if (buffer_freed(bh)) {
3713 BUFFER_TRACE(bh, "freed: skip");
3714 goto unlock;
3715 }
3716 if (!buffer_mapped(bh)) {
3717 BUFFER_TRACE(bh, "unmapped");
3718 ext4_get_block(inode, iblock, bh, 0);
3719 /* unmapped? It's a hole - nothing to do */
3720 if (!buffer_mapped(bh)) {
3721 BUFFER_TRACE(bh, "still unmapped");
3722 goto unlock;
3723 }
3724 }
3725
3726 /* Ok, it's mapped. Make sure it's up-to-date */
3727 if (folio_test_uptodate(folio))
3728 set_buffer_uptodate(bh);
3729
3730 if (!buffer_uptodate(bh)) {
3731 err = ext4_read_bh_lock(bh, 0, true);
3732 if (err)
3733 goto unlock;
3734 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3735 /* We expect the key to be set. */
3736 BUG_ON(!fscrypt_has_encryption_key(inode));
3737 err = fscrypt_decrypt_pagecache_blocks(folio,
3738 blocksize,
3739 bh_offset(bh));
3740 if (err) {
3741 clear_buffer_uptodate(bh);
3742 goto unlock;
3743 }
3744 }
3745 }
3746 if (ext4_should_journal_data(inode)) {
3747 BUFFER_TRACE(bh, "get write access");
3748 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3749 EXT4_JTR_NONE);
3750 if (err)
3751 goto unlock;
3752 }
3753 folio_zero_range(folio, offset, length);
3754 BUFFER_TRACE(bh, "zeroed end of block");
3755
3756 if (ext4_should_journal_data(inode)) {
3757 err = ext4_dirty_journalled_data(handle, bh);
3758 } else {
3759 err = 0;
3760 mark_buffer_dirty(bh);
3761 if (ext4_should_order_data(inode))
3762 err = ext4_jbd2_inode_add_write(handle, inode, from,
3763 length);
3764 }
3765
3766unlock:
3767 folio_unlock(folio);
3768 folio_put(folio);
3769 return err;
3770}
3771
3772/*
3773 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3774 * starting from file offset 'from'. The range to be zero'd must
3775 * be contained with in one block. If the specified range exceeds
3776 * the end of the block it will be shortened to end of the block
3777 * that corresponds to 'from'
3778 */
3779static int ext4_block_zero_page_range(handle_t *handle,
3780 struct address_space *mapping, loff_t from, loff_t length)
3781{
3782 struct inode *inode = mapping->host;
3783 unsigned offset = from & (PAGE_SIZE-1);
3784 unsigned blocksize = inode->i_sb->s_blocksize;
3785 unsigned max = blocksize - (offset & (blocksize - 1));
3786
3787 /*
3788 * correct length if it does not fall between
3789 * 'from' and the end of the block
3790 */
3791 if (length > max || length < 0)
3792 length = max;
3793
3794 if (IS_DAX(inode)) {
3795 return dax_zero_range(inode, from, length, NULL,
3796 &ext4_iomap_ops);
3797 }
3798 return __ext4_block_zero_page_range(handle, mapping, from, length);
3799}
3800
3801/*
3802 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3803 * up to the end of the block which corresponds to `from'.
3804 * This required during truncate. We need to physically zero the tail end
3805 * of that block so it doesn't yield old data if the file is later grown.
3806 */
3807static int ext4_block_truncate_page(handle_t *handle,
3808 struct address_space *mapping, loff_t from)
3809{
3810 unsigned offset = from & (PAGE_SIZE-1);
3811 unsigned length;
3812 unsigned blocksize;
3813 struct inode *inode = mapping->host;
3814
3815 /* If we are processing an encrypted inode during orphan list handling */
3816 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3817 return 0;
3818
3819 blocksize = inode->i_sb->s_blocksize;
3820 length = blocksize - (offset & (blocksize - 1));
3821
3822 return ext4_block_zero_page_range(handle, mapping, from, length);
3823}
3824
3825int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3826 loff_t lstart, loff_t length)
3827{
3828 struct super_block *sb = inode->i_sb;
3829 struct address_space *mapping = inode->i_mapping;
3830 unsigned partial_start, partial_end;
3831 ext4_fsblk_t start, end;
3832 loff_t byte_end = (lstart + length - 1);
3833 int err = 0;
3834
3835 partial_start = lstart & (sb->s_blocksize - 1);
3836 partial_end = byte_end & (sb->s_blocksize - 1);
3837
3838 start = lstart >> sb->s_blocksize_bits;
3839 end = byte_end >> sb->s_blocksize_bits;
3840
3841 /* Handle partial zero within the single block */
3842 if (start == end &&
3843 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3844 err = ext4_block_zero_page_range(handle, mapping,
3845 lstart, length);
3846 return err;
3847 }
3848 /* Handle partial zero out on the start of the range */
3849 if (partial_start) {
3850 err = ext4_block_zero_page_range(handle, mapping,
3851 lstart, sb->s_blocksize);
3852 if (err)
3853 return err;
3854 }
3855 /* Handle partial zero out on the end of the range */
3856 if (partial_end != sb->s_blocksize - 1)
3857 err = ext4_block_zero_page_range(handle, mapping,
3858 byte_end - partial_end,
3859 partial_end + 1);
3860 return err;
3861}
3862
3863int ext4_can_truncate(struct inode *inode)
3864{
3865 if (S_ISREG(inode->i_mode))
3866 return 1;
3867 if (S_ISDIR(inode->i_mode))
3868 return 1;
3869 if (S_ISLNK(inode->i_mode))
3870 return !ext4_inode_is_fast_symlink(inode);
3871 return 0;
3872}
3873
3874/*
3875 * We have to make sure i_disksize gets properly updated before we truncate
3876 * page cache due to hole punching or zero range. Otherwise i_disksize update
3877 * can get lost as it may have been postponed to submission of writeback but
3878 * that will never happen after we truncate page cache.
3879 */
3880int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3881 loff_t len)
3882{
3883 handle_t *handle;
3884 int ret;
3885
3886 loff_t size = i_size_read(inode);
3887
3888 WARN_ON(!inode_is_locked(inode));
3889 if (offset > size || offset + len < size)
3890 return 0;
3891
3892 if (EXT4_I(inode)->i_disksize >= size)
3893 return 0;
3894
3895 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3896 if (IS_ERR(handle))
3897 return PTR_ERR(handle);
3898 ext4_update_i_disksize(inode, size);
3899 ret = ext4_mark_inode_dirty(handle, inode);
3900 ext4_journal_stop(handle);
3901
3902 return ret;
3903}
3904
3905static void ext4_wait_dax_page(struct inode *inode)
3906{
3907 filemap_invalidate_unlock(inode->i_mapping);
3908 schedule();
3909 filemap_invalidate_lock(inode->i_mapping);
3910}
3911
3912int ext4_break_layouts(struct inode *inode)
3913{
3914 struct page *page;
3915 int error;
3916
3917 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3918 return -EINVAL;
3919
3920 do {
3921 page = dax_layout_busy_page(inode->i_mapping);
3922 if (!page)
3923 return 0;
3924
3925 error = ___wait_var_event(&page->_refcount,
3926 atomic_read(&page->_refcount) == 1,
3927 TASK_INTERRUPTIBLE, 0, 0,
3928 ext4_wait_dax_page(inode));
3929 } while (error == 0);
3930
3931 return error;
3932}
3933
3934/*
3935 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3936 * associated with the given offset and length
3937 *
3938 * @inode: File inode
3939 * @offset: The offset where the hole will begin
3940 * @len: The length of the hole
3941 *
3942 * Returns: 0 on success or negative on failure
3943 */
3944
3945int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3946{
3947 struct inode *inode = file_inode(file);
3948 struct super_block *sb = inode->i_sb;
3949 ext4_lblk_t first_block, stop_block;
3950 struct address_space *mapping = inode->i_mapping;
3951 loff_t first_block_offset, last_block_offset, max_length;
3952 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3953 handle_t *handle;
3954 unsigned int credits;
3955 int ret = 0, ret2 = 0;
3956
3957 trace_ext4_punch_hole(inode, offset, length, 0);
3958
3959 /*
3960 * Write out all dirty pages to avoid race conditions
3961 * Then release them.
3962 */
3963 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3964 ret = filemap_write_and_wait_range(mapping, offset,
3965 offset + length - 1);
3966 if (ret)
3967 return ret;
3968 }
3969
3970 inode_lock(inode);
3971
3972 /* No need to punch hole beyond i_size */
3973 if (offset >= inode->i_size)
3974 goto out_mutex;
3975
3976 /*
3977 * If the hole extends beyond i_size, set the hole
3978 * to end after the page that contains i_size
3979 */
3980 if (offset + length > inode->i_size) {
3981 length = inode->i_size +
3982 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3983 offset;
3984 }
3985
3986 /*
3987 * For punch hole the length + offset needs to be within one block
3988 * before last range. Adjust the length if it goes beyond that limit.
3989 */
3990 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
3991 if (offset + length > max_length)
3992 length = max_length - offset;
3993
3994 if (offset & (sb->s_blocksize - 1) ||
3995 (offset + length) & (sb->s_blocksize - 1)) {
3996 /*
3997 * Attach jinode to inode for jbd2 if we do any zeroing of
3998 * partial block
3999 */
4000 ret = ext4_inode_attach_jinode(inode);
4001 if (ret < 0)
4002 goto out_mutex;
4003
4004 }
4005
4006 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4007 inode_dio_wait(inode);
4008
4009 ret = file_modified(file);
4010 if (ret)
4011 goto out_mutex;
4012
4013 /*
4014 * Prevent page faults from reinstantiating pages we have released from
4015 * page cache.
4016 */
4017 filemap_invalidate_lock(mapping);
4018
4019 ret = ext4_break_layouts(inode);
4020 if (ret)
4021 goto out_dio;
4022
4023 first_block_offset = round_up(offset, sb->s_blocksize);
4024 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4025
4026 /* Now release the pages and zero block aligned part of pages*/
4027 if (last_block_offset > first_block_offset) {
4028 ret = ext4_update_disksize_before_punch(inode, offset, length);
4029 if (ret)
4030 goto out_dio;
4031 truncate_pagecache_range(inode, first_block_offset,
4032 last_block_offset);
4033 }
4034
4035 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4036 credits = ext4_writepage_trans_blocks(inode);
4037 else
4038 credits = ext4_blocks_for_truncate(inode);
4039 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4040 if (IS_ERR(handle)) {
4041 ret = PTR_ERR(handle);
4042 ext4_std_error(sb, ret);
4043 goto out_dio;
4044 }
4045
4046 ret = ext4_zero_partial_blocks(handle, inode, offset,
4047 length);
4048 if (ret)
4049 goto out_stop;
4050
4051 first_block = (offset + sb->s_blocksize - 1) >>
4052 EXT4_BLOCK_SIZE_BITS(sb);
4053 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4054
4055 /* If there are blocks to remove, do it */
4056 if (stop_block > first_block) {
4057 ext4_lblk_t hole_len = stop_block - first_block;
4058
4059 down_write(&EXT4_I(inode)->i_data_sem);
4060 ext4_discard_preallocations(inode);
4061
4062 ext4_es_remove_extent(inode, first_block, hole_len);
4063
4064 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4065 ret = ext4_ext_remove_space(inode, first_block,
4066 stop_block - 1);
4067 else
4068 ret = ext4_ind_remove_space(handle, inode, first_block,
4069 stop_block);
4070
4071 ext4_es_insert_extent(inode, first_block, hole_len, ~0,
4072 EXTENT_STATUS_HOLE, 0);
4073 up_write(&EXT4_I(inode)->i_data_sem);
4074 }
4075 ext4_fc_track_range(handle, inode, first_block, stop_block);
4076 if (IS_SYNC(inode))
4077 ext4_handle_sync(handle);
4078
4079 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4080 ret2 = ext4_mark_inode_dirty(handle, inode);
4081 if (unlikely(ret2))
4082 ret = ret2;
4083 if (ret >= 0)
4084 ext4_update_inode_fsync_trans(handle, inode, 1);
4085out_stop:
4086 ext4_journal_stop(handle);
4087out_dio:
4088 filemap_invalidate_unlock(mapping);
4089out_mutex:
4090 inode_unlock(inode);
4091 return ret;
4092}
4093
4094int ext4_inode_attach_jinode(struct inode *inode)
4095{
4096 struct ext4_inode_info *ei = EXT4_I(inode);
4097 struct jbd2_inode *jinode;
4098
4099 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4100 return 0;
4101
4102 jinode = jbd2_alloc_inode(GFP_KERNEL);
4103 spin_lock(&inode->i_lock);
4104 if (!ei->jinode) {
4105 if (!jinode) {
4106 spin_unlock(&inode->i_lock);
4107 return -ENOMEM;
4108 }
4109 ei->jinode = jinode;
4110 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4111 jinode = NULL;
4112 }
4113 spin_unlock(&inode->i_lock);
4114 if (unlikely(jinode != NULL))
4115 jbd2_free_inode(jinode);
4116 return 0;
4117}
4118
4119/*
4120 * ext4_truncate()
4121 *
4122 * We block out ext4_get_block() block instantiations across the entire
4123 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4124 * simultaneously on behalf of the same inode.
4125 *
4126 * As we work through the truncate and commit bits of it to the journal there
4127 * is one core, guiding principle: the file's tree must always be consistent on
4128 * disk. We must be able to restart the truncate after a crash.
4129 *
4130 * The file's tree may be transiently inconsistent in memory (although it
4131 * probably isn't), but whenever we close off and commit a journal transaction,
4132 * the contents of (the filesystem + the journal) must be consistent and
4133 * restartable. It's pretty simple, really: bottom up, right to left (although
4134 * left-to-right works OK too).
4135 *
4136 * Note that at recovery time, journal replay occurs *before* the restart of
4137 * truncate against the orphan inode list.
4138 *
4139 * The committed inode has the new, desired i_size (which is the same as
4140 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4141 * that this inode's truncate did not complete and it will again call
4142 * ext4_truncate() to have another go. So there will be instantiated blocks
4143 * to the right of the truncation point in a crashed ext4 filesystem. But
4144 * that's fine - as long as they are linked from the inode, the post-crash
4145 * ext4_truncate() run will find them and release them.
4146 */
4147int ext4_truncate(struct inode *inode)
4148{
4149 struct ext4_inode_info *ei = EXT4_I(inode);
4150 unsigned int credits;
4151 int err = 0, err2;
4152 handle_t *handle;
4153 struct address_space *mapping = inode->i_mapping;
4154
4155 /*
4156 * There is a possibility that we're either freeing the inode
4157 * or it's a completely new inode. In those cases we might not
4158 * have i_rwsem locked because it's not necessary.
4159 */
4160 if (!(inode->i_state & (I_NEW|I_FREEING)))
4161 WARN_ON(!inode_is_locked(inode));
4162 trace_ext4_truncate_enter(inode);
4163
4164 if (!ext4_can_truncate(inode))
4165 goto out_trace;
4166
4167 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4168 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4169
4170 if (ext4_has_inline_data(inode)) {
4171 int has_inline = 1;
4172
4173 err = ext4_inline_data_truncate(inode, &has_inline);
4174 if (err || has_inline)
4175 goto out_trace;
4176 }
4177
4178 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4179 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4180 err = ext4_inode_attach_jinode(inode);
4181 if (err)
4182 goto out_trace;
4183 }
4184
4185 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4186 credits = ext4_writepage_trans_blocks(inode);
4187 else
4188 credits = ext4_blocks_for_truncate(inode);
4189
4190 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4191 if (IS_ERR(handle)) {
4192 err = PTR_ERR(handle);
4193 goto out_trace;
4194 }
4195
4196 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4197 ext4_block_truncate_page(handle, mapping, inode->i_size);
4198
4199 /*
4200 * We add the inode to the orphan list, so that if this
4201 * truncate spans multiple transactions, and we crash, we will
4202 * resume the truncate when the filesystem recovers. It also
4203 * marks the inode dirty, to catch the new size.
4204 *
4205 * Implication: the file must always be in a sane, consistent
4206 * truncatable state while each transaction commits.
4207 */
4208 err = ext4_orphan_add(handle, inode);
4209 if (err)
4210 goto out_stop;
4211
4212 down_write(&EXT4_I(inode)->i_data_sem);
4213
4214 ext4_discard_preallocations(inode);
4215
4216 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4217 err = ext4_ext_truncate(handle, inode);
4218 else
4219 ext4_ind_truncate(handle, inode);
4220
4221 up_write(&ei->i_data_sem);
4222 if (err)
4223 goto out_stop;
4224
4225 if (IS_SYNC(inode))
4226 ext4_handle_sync(handle);
4227
4228out_stop:
4229 /*
4230 * If this was a simple ftruncate() and the file will remain alive,
4231 * then we need to clear up the orphan record which we created above.
4232 * However, if this was a real unlink then we were called by
4233 * ext4_evict_inode(), and we allow that function to clean up the
4234 * orphan info for us.
4235 */
4236 if (inode->i_nlink)
4237 ext4_orphan_del(handle, inode);
4238
4239 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4240 err2 = ext4_mark_inode_dirty(handle, inode);
4241 if (unlikely(err2 && !err))
4242 err = err2;
4243 ext4_journal_stop(handle);
4244
4245out_trace:
4246 trace_ext4_truncate_exit(inode);
4247 return err;
4248}
4249
4250static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4251{
4252 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4253 return inode_peek_iversion_raw(inode);
4254 else
4255 return inode_peek_iversion(inode);
4256}
4257
4258static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4259 struct ext4_inode_info *ei)
4260{
4261 struct inode *inode = &(ei->vfs_inode);
4262 u64 i_blocks = READ_ONCE(inode->i_blocks);
4263 struct super_block *sb = inode->i_sb;
4264
4265 if (i_blocks <= ~0U) {
4266 /*
4267 * i_blocks can be represented in a 32 bit variable
4268 * as multiple of 512 bytes
4269 */
4270 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4271 raw_inode->i_blocks_high = 0;
4272 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4273 return 0;
4274 }
4275
4276 /*
4277 * This should never happen since sb->s_maxbytes should not have
4278 * allowed this, sb->s_maxbytes was set according to the huge_file
4279 * feature in ext4_fill_super().
4280 */
4281 if (!ext4_has_feature_huge_file(sb))
4282 return -EFSCORRUPTED;
4283
4284 if (i_blocks <= 0xffffffffffffULL) {
4285 /*
4286 * i_blocks can be represented in a 48 bit variable
4287 * as multiple of 512 bytes
4288 */
4289 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4290 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4291 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4292 } else {
4293 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4294 /* i_block is stored in file system block size */
4295 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4296 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4297 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4298 }
4299 return 0;
4300}
4301
4302static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4303{
4304 struct ext4_inode_info *ei = EXT4_I(inode);
4305 uid_t i_uid;
4306 gid_t i_gid;
4307 projid_t i_projid;
4308 int block;
4309 int err;
4310
4311 err = ext4_inode_blocks_set(raw_inode, ei);
4312
4313 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4314 i_uid = i_uid_read(inode);
4315 i_gid = i_gid_read(inode);
4316 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4317 if (!(test_opt(inode->i_sb, NO_UID32))) {
4318 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4319 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4320 /*
4321 * Fix up interoperability with old kernels. Otherwise,
4322 * old inodes get re-used with the upper 16 bits of the
4323 * uid/gid intact.
4324 */
4325 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4326 raw_inode->i_uid_high = 0;
4327 raw_inode->i_gid_high = 0;
4328 } else {
4329 raw_inode->i_uid_high =
4330 cpu_to_le16(high_16_bits(i_uid));
4331 raw_inode->i_gid_high =
4332 cpu_to_le16(high_16_bits(i_gid));
4333 }
4334 } else {
4335 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4336 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4337 raw_inode->i_uid_high = 0;
4338 raw_inode->i_gid_high = 0;
4339 }
4340 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4341
4342 EXT4_INODE_SET_CTIME(inode, raw_inode);
4343 EXT4_INODE_SET_MTIME(inode, raw_inode);
4344 EXT4_INODE_SET_ATIME(inode, raw_inode);
4345 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4346
4347 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4348 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4349 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4350 raw_inode->i_file_acl_high =
4351 cpu_to_le16(ei->i_file_acl >> 32);
4352 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4353 ext4_isize_set(raw_inode, ei->i_disksize);
4354
4355 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4356 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4357 if (old_valid_dev(inode->i_rdev)) {
4358 raw_inode->i_block[0] =
4359 cpu_to_le32(old_encode_dev(inode->i_rdev));
4360 raw_inode->i_block[1] = 0;
4361 } else {
4362 raw_inode->i_block[0] = 0;
4363 raw_inode->i_block[1] =
4364 cpu_to_le32(new_encode_dev(inode->i_rdev));
4365 raw_inode->i_block[2] = 0;
4366 }
4367 } else if (!ext4_has_inline_data(inode)) {
4368 for (block = 0; block < EXT4_N_BLOCKS; block++)
4369 raw_inode->i_block[block] = ei->i_data[block];
4370 }
4371
4372 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4373 u64 ivers = ext4_inode_peek_iversion(inode);
4374
4375 raw_inode->i_disk_version = cpu_to_le32(ivers);
4376 if (ei->i_extra_isize) {
4377 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4378 raw_inode->i_version_hi =
4379 cpu_to_le32(ivers >> 32);
4380 raw_inode->i_extra_isize =
4381 cpu_to_le16(ei->i_extra_isize);
4382 }
4383 }
4384
4385 if (i_projid != EXT4_DEF_PROJID &&
4386 !ext4_has_feature_project(inode->i_sb))
4387 err = err ?: -EFSCORRUPTED;
4388
4389 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4390 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4391 raw_inode->i_projid = cpu_to_le32(i_projid);
4392
4393 ext4_inode_csum_set(inode, raw_inode, ei);
4394 return err;
4395}
4396
4397/*
4398 * ext4_get_inode_loc returns with an extra refcount against the inode's
4399 * underlying buffer_head on success. If we pass 'inode' and it does not
4400 * have in-inode xattr, we have all inode data in memory that is needed
4401 * to recreate the on-disk version of this inode.
4402 */
4403static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4404 struct inode *inode, struct ext4_iloc *iloc,
4405 ext4_fsblk_t *ret_block)
4406{
4407 struct ext4_group_desc *gdp;
4408 struct buffer_head *bh;
4409 ext4_fsblk_t block;
4410 struct blk_plug plug;
4411 int inodes_per_block, inode_offset;
4412
4413 iloc->bh = NULL;
4414 if (ino < EXT4_ROOT_INO ||
4415 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4416 return -EFSCORRUPTED;
4417
4418 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4419 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4420 if (!gdp)
4421 return -EIO;
4422
4423 /*
4424 * Figure out the offset within the block group inode table
4425 */
4426 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4427 inode_offset = ((ino - 1) %
4428 EXT4_INODES_PER_GROUP(sb));
4429 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4430
4431 block = ext4_inode_table(sb, gdp);
4432 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4433 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4434 ext4_error(sb, "Invalid inode table block %llu in "
4435 "block_group %u", block, iloc->block_group);
4436 return -EFSCORRUPTED;
4437 }
4438 block += (inode_offset / inodes_per_block);
4439
4440 bh = sb_getblk(sb, block);
4441 if (unlikely(!bh))
4442 return -ENOMEM;
4443 if (ext4_buffer_uptodate(bh))
4444 goto has_buffer;
4445
4446 lock_buffer(bh);
4447 if (ext4_buffer_uptodate(bh)) {
4448 /* Someone brought it uptodate while we waited */
4449 unlock_buffer(bh);
4450 goto has_buffer;
4451 }
4452
4453 /*
4454 * If we have all information of the inode in memory and this
4455 * is the only valid inode in the block, we need not read the
4456 * block.
4457 */
4458 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4459 struct buffer_head *bitmap_bh;
4460 int i, start;
4461
4462 start = inode_offset & ~(inodes_per_block - 1);
4463
4464 /* Is the inode bitmap in cache? */
4465 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4466 if (unlikely(!bitmap_bh))
4467 goto make_io;
4468
4469 /*
4470 * If the inode bitmap isn't in cache then the
4471 * optimisation may end up performing two reads instead
4472 * of one, so skip it.
4473 */
4474 if (!buffer_uptodate(bitmap_bh)) {
4475 brelse(bitmap_bh);
4476 goto make_io;
4477 }
4478 for (i = start; i < start + inodes_per_block; i++) {
4479 if (i == inode_offset)
4480 continue;
4481 if (ext4_test_bit(i, bitmap_bh->b_data))
4482 break;
4483 }
4484 brelse(bitmap_bh);
4485 if (i == start + inodes_per_block) {
4486 struct ext4_inode *raw_inode =
4487 (struct ext4_inode *) (bh->b_data + iloc->offset);
4488
4489 /* all other inodes are free, so skip I/O */
4490 memset(bh->b_data, 0, bh->b_size);
4491 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4492 ext4_fill_raw_inode(inode, raw_inode);
4493 set_buffer_uptodate(bh);
4494 unlock_buffer(bh);
4495 goto has_buffer;
4496 }
4497 }
4498
4499make_io:
4500 /*
4501 * If we need to do any I/O, try to pre-readahead extra
4502 * blocks from the inode table.
4503 */
4504 blk_start_plug(&plug);
4505 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4506 ext4_fsblk_t b, end, table;
4507 unsigned num;
4508 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4509
4510 table = ext4_inode_table(sb, gdp);
4511 /* s_inode_readahead_blks is always a power of 2 */
4512 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4513 if (table > b)
4514 b = table;
4515 end = b + ra_blks;
4516 num = EXT4_INODES_PER_GROUP(sb);
4517 if (ext4_has_group_desc_csum(sb))
4518 num -= ext4_itable_unused_count(sb, gdp);
4519 table += num / inodes_per_block;
4520 if (end > table)
4521 end = table;
4522 while (b <= end)
4523 ext4_sb_breadahead_unmovable(sb, b++);
4524 }
4525
4526 /*
4527 * There are other valid inodes in the buffer, this inode
4528 * has in-inode xattrs, or we don't have this inode in memory.
4529 * Read the block from disk.
4530 */
4531 trace_ext4_load_inode(sb, ino);
4532 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL,
4533 ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO));
4534 blk_finish_plug(&plug);
4535 wait_on_buffer(bh);
4536 if (!buffer_uptodate(bh)) {
4537 if (ret_block)
4538 *ret_block = block;
4539 brelse(bh);
4540 return -EIO;
4541 }
4542has_buffer:
4543 iloc->bh = bh;
4544 return 0;
4545}
4546
4547static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4548 struct ext4_iloc *iloc)
4549{
4550 ext4_fsblk_t err_blk = 0;
4551 int ret;
4552
4553 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4554 &err_blk);
4555
4556 if (ret == -EIO)
4557 ext4_error_inode_block(inode, err_blk, EIO,
4558 "unable to read itable block");
4559
4560 return ret;
4561}
4562
4563int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4564{
4565 ext4_fsblk_t err_blk = 0;
4566 int ret;
4567
4568 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4569 &err_blk);
4570
4571 if (ret == -EIO)
4572 ext4_error_inode_block(inode, err_blk, EIO,
4573 "unable to read itable block");
4574
4575 return ret;
4576}
4577
4578
4579int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4580 struct ext4_iloc *iloc)
4581{
4582 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4583}
4584
4585static bool ext4_should_enable_dax(struct inode *inode)
4586{
4587 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4588
4589 if (test_opt2(inode->i_sb, DAX_NEVER))
4590 return false;
4591 if (!S_ISREG(inode->i_mode))
4592 return false;
4593 if (ext4_should_journal_data(inode))
4594 return false;
4595 if (ext4_has_inline_data(inode))
4596 return false;
4597 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4598 return false;
4599 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4600 return false;
4601 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4602 return false;
4603 if (test_opt(inode->i_sb, DAX_ALWAYS))
4604 return true;
4605
4606 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4607}
4608
4609void ext4_set_inode_flags(struct inode *inode, bool init)
4610{
4611 unsigned int flags = EXT4_I(inode)->i_flags;
4612 unsigned int new_fl = 0;
4613
4614 WARN_ON_ONCE(IS_DAX(inode) && init);
4615
4616 if (flags & EXT4_SYNC_FL)
4617 new_fl |= S_SYNC;
4618 if (flags & EXT4_APPEND_FL)
4619 new_fl |= S_APPEND;
4620 if (flags & EXT4_IMMUTABLE_FL)
4621 new_fl |= S_IMMUTABLE;
4622 if (flags & EXT4_NOATIME_FL)
4623 new_fl |= S_NOATIME;
4624 if (flags & EXT4_DIRSYNC_FL)
4625 new_fl |= S_DIRSYNC;
4626
4627 /* Because of the way inode_set_flags() works we must preserve S_DAX
4628 * here if already set. */
4629 new_fl |= (inode->i_flags & S_DAX);
4630 if (init && ext4_should_enable_dax(inode))
4631 new_fl |= S_DAX;
4632
4633 if (flags & EXT4_ENCRYPT_FL)
4634 new_fl |= S_ENCRYPTED;
4635 if (flags & EXT4_CASEFOLD_FL)
4636 new_fl |= S_CASEFOLD;
4637 if (flags & EXT4_VERITY_FL)
4638 new_fl |= S_VERITY;
4639 inode_set_flags(inode, new_fl,
4640 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4641 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4642}
4643
4644static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4645 struct ext4_inode_info *ei)
4646{
4647 blkcnt_t i_blocks ;
4648 struct inode *inode = &(ei->vfs_inode);
4649 struct super_block *sb = inode->i_sb;
4650
4651 if (ext4_has_feature_huge_file(sb)) {
4652 /* we are using combined 48 bit field */
4653 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4654 le32_to_cpu(raw_inode->i_blocks_lo);
4655 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4656 /* i_blocks represent file system block size */
4657 return i_blocks << (inode->i_blkbits - 9);
4658 } else {
4659 return i_blocks;
4660 }
4661 } else {
4662 return le32_to_cpu(raw_inode->i_blocks_lo);
4663 }
4664}
4665
4666static inline int ext4_iget_extra_inode(struct inode *inode,
4667 struct ext4_inode *raw_inode,
4668 struct ext4_inode_info *ei)
4669{
4670 __le32 *magic = (void *)raw_inode +
4671 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4672
4673 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4674 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4675 int err;
4676
4677 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4678 err = ext4_find_inline_data_nolock(inode);
4679 if (!err && ext4_has_inline_data(inode))
4680 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4681 return err;
4682 } else
4683 EXT4_I(inode)->i_inline_off = 0;
4684 return 0;
4685}
4686
4687int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4688{
4689 if (!ext4_has_feature_project(inode->i_sb))
4690 return -EOPNOTSUPP;
4691 *projid = EXT4_I(inode)->i_projid;
4692 return 0;
4693}
4694
4695/*
4696 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4697 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4698 * set.
4699 */
4700static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4701{
4702 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4703 inode_set_iversion_raw(inode, val);
4704 else
4705 inode_set_iversion_queried(inode, val);
4706}
4707
4708static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags)
4709
4710{
4711 if (flags & EXT4_IGET_EA_INODE) {
4712 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4713 return "missing EA_INODE flag";
4714 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4715 EXT4_I(inode)->i_file_acl)
4716 return "ea_inode with extended attributes";
4717 } else {
4718 if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4719 return "unexpected EA_INODE flag";
4720 }
4721 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD))
4722 return "unexpected bad inode w/o EXT4_IGET_BAD";
4723 return NULL;
4724}
4725
4726struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4727 ext4_iget_flags flags, const char *function,
4728 unsigned int line)
4729{
4730 struct ext4_iloc iloc;
4731 struct ext4_inode *raw_inode;
4732 struct ext4_inode_info *ei;
4733 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4734 struct inode *inode;
4735 const char *err_str;
4736 journal_t *journal = EXT4_SB(sb)->s_journal;
4737 long ret;
4738 loff_t size;
4739 int block;
4740 uid_t i_uid;
4741 gid_t i_gid;
4742 projid_t i_projid;
4743
4744 if ((!(flags & EXT4_IGET_SPECIAL) &&
4745 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4746 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4747 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4748 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4749 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4750 (ino < EXT4_ROOT_INO) ||
4751 (ino > le32_to_cpu(es->s_inodes_count))) {
4752 if (flags & EXT4_IGET_HANDLE)
4753 return ERR_PTR(-ESTALE);
4754 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4755 "inode #%lu: comm %s: iget: illegal inode #",
4756 ino, current->comm);
4757 return ERR_PTR(-EFSCORRUPTED);
4758 }
4759
4760 inode = iget_locked(sb, ino);
4761 if (!inode)
4762 return ERR_PTR(-ENOMEM);
4763 if (!(inode->i_state & I_NEW)) {
4764 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4765 ext4_error_inode(inode, function, line, 0, err_str);
4766 iput(inode);
4767 return ERR_PTR(-EFSCORRUPTED);
4768 }
4769 return inode;
4770 }
4771
4772 ei = EXT4_I(inode);
4773 iloc.bh = NULL;
4774
4775 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4776 if (ret < 0)
4777 goto bad_inode;
4778 raw_inode = ext4_raw_inode(&iloc);
4779
4780 if ((flags & EXT4_IGET_HANDLE) &&
4781 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4782 ret = -ESTALE;
4783 goto bad_inode;
4784 }
4785
4786 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4787 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4788 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4789 EXT4_INODE_SIZE(inode->i_sb) ||
4790 (ei->i_extra_isize & 3)) {
4791 ext4_error_inode(inode, function, line, 0,
4792 "iget: bad extra_isize %u "
4793 "(inode size %u)",
4794 ei->i_extra_isize,
4795 EXT4_INODE_SIZE(inode->i_sb));
4796 ret = -EFSCORRUPTED;
4797 goto bad_inode;
4798 }
4799 } else
4800 ei->i_extra_isize = 0;
4801
4802 /* Precompute checksum seed for inode metadata */
4803 if (ext4_has_metadata_csum(sb)) {
4804 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4805 __u32 csum;
4806 __le32 inum = cpu_to_le32(inode->i_ino);
4807 __le32 gen = raw_inode->i_generation;
4808 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4809 sizeof(inum));
4810 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4811 sizeof(gen));
4812 }
4813
4814 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4815 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4816 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4817 ext4_error_inode_err(inode, function, line, 0,
4818 EFSBADCRC, "iget: checksum invalid");
4819 ret = -EFSBADCRC;
4820 goto bad_inode;
4821 }
4822
4823 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4824 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4825 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4826 if (ext4_has_feature_project(sb) &&
4827 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4828 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4829 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4830 else
4831 i_projid = EXT4_DEF_PROJID;
4832
4833 if (!(test_opt(inode->i_sb, NO_UID32))) {
4834 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4835 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4836 }
4837 i_uid_write(inode, i_uid);
4838 i_gid_write(inode, i_gid);
4839 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4840 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4841
4842 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4843 ei->i_inline_off = 0;
4844 ei->i_dir_start_lookup = 0;
4845 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4846 /* We now have enough fields to check if the inode was active or not.
4847 * This is needed because nfsd might try to access dead inodes
4848 * the test is that same one that e2fsck uses
4849 * NeilBrown 1999oct15
4850 */
4851 if (inode->i_nlink == 0) {
4852 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4853 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4854 ino != EXT4_BOOT_LOADER_INO) {
4855 /* this inode is deleted or unallocated */
4856 if (flags & EXT4_IGET_SPECIAL) {
4857 ext4_error_inode(inode, function, line, 0,
4858 "iget: special inode unallocated");
4859 ret = -EFSCORRUPTED;
4860 } else
4861 ret = -ESTALE;
4862 goto bad_inode;
4863 }
4864 /* The only unlinked inodes we let through here have
4865 * valid i_mode and are being read by the orphan
4866 * recovery code: that's fine, we're about to complete
4867 * the process of deleting those.
4868 * OR it is the EXT4_BOOT_LOADER_INO which is
4869 * not initialized on a new filesystem. */
4870 }
4871 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4872 ext4_set_inode_flags(inode, true);
4873 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4874 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4875 if (ext4_has_feature_64bit(sb))
4876 ei->i_file_acl |=
4877 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4878 inode->i_size = ext4_isize(sb, raw_inode);
4879 if ((size = i_size_read(inode)) < 0) {
4880 ext4_error_inode(inode, function, line, 0,
4881 "iget: bad i_size value: %lld", size);
4882 ret = -EFSCORRUPTED;
4883 goto bad_inode;
4884 }
4885 /*
4886 * If dir_index is not enabled but there's dir with INDEX flag set,
4887 * we'd normally treat htree data as empty space. But with metadata
4888 * checksumming that corrupts checksums so forbid that.
4889 */
4890 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4891 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4892 ext4_error_inode(inode, function, line, 0,
4893 "iget: Dir with htree data on filesystem without dir_index feature.");
4894 ret = -EFSCORRUPTED;
4895 goto bad_inode;
4896 }
4897 ei->i_disksize = inode->i_size;
4898#ifdef CONFIG_QUOTA
4899 ei->i_reserved_quota = 0;
4900#endif
4901 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4902 ei->i_block_group = iloc.block_group;
4903 ei->i_last_alloc_group = ~0;
4904 /*
4905 * NOTE! The in-memory inode i_data array is in little-endian order
4906 * even on big-endian machines: we do NOT byteswap the block numbers!
4907 */
4908 for (block = 0; block < EXT4_N_BLOCKS; block++)
4909 ei->i_data[block] = raw_inode->i_block[block];
4910 INIT_LIST_HEAD(&ei->i_orphan);
4911 ext4_fc_init_inode(&ei->vfs_inode);
4912
4913 /*
4914 * Set transaction id's of transactions that have to be committed
4915 * to finish f[data]sync. We set them to currently running transaction
4916 * as we cannot be sure that the inode or some of its metadata isn't
4917 * part of the transaction - the inode could have been reclaimed and
4918 * now it is reread from disk.
4919 */
4920 if (journal) {
4921 transaction_t *transaction;
4922 tid_t tid;
4923
4924 read_lock(&journal->j_state_lock);
4925 if (journal->j_running_transaction)
4926 transaction = journal->j_running_transaction;
4927 else
4928 transaction = journal->j_committing_transaction;
4929 if (transaction)
4930 tid = transaction->t_tid;
4931 else
4932 tid = journal->j_commit_sequence;
4933 read_unlock(&journal->j_state_lock);
4934 ei->i_sync_tid = tid;
4935 ei->i_datasync_tid = tid;
4936 }
4937
4938 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4939 if (ei->i_extra_isize == 0) {
4940 /* The extra space is currently unused. Use it. */
4941 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4942 ei->i_extra_isize = sizeof(struct ext4_inode) -
4943 EXT4_GOOD_OLD_INODE_SIZE;
4944 } else {
4945 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4946 if (ret)
4947 goto bad_inode;
4948 }
4949 }
4950
4951 EXT4_INODE_GET_CTIME(inode, raw_inode);
4952 EXT4_INODE_GET_ATIME(inode, raw_inode);
4953 EXT4_INODE_GET_MTIME(inode, raw_inode);
4954 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4955
4956 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4957 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4958
4959 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4960 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4961 ivers |=
4962 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4963 }
4964 ext4_inode_set_iversion_queried(inode, ivers);
4965 }
4966
4967 ret = 0;
4968 if (ei->i_file_acl &&
4969 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4970 ext4_error_inode(inode, function, line, 0,
4971 "iget: bad extended attribute block %llu",
4972 ei->i_file_acl);
4973 ret = -EFSCORRUPTED;
4974 goto bad_inode;
4975 } else if (!ext4_has_inline_data(inode)) {
4976 /* validate the block references in the inode */
4977 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4978 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4979 (S_ISLNK(inode->i_mode) &&
4980 !ext4_inode_is_fast_symlink(inode)))) {
4981 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4982 ret = ext4_ext_check_inode(inode);
4983 else
4984 ret = ext4_ind_check_inode(inode);
4985 }
4986 }
4987 if (ret)
4988 goto bad_inode;
4989
4990 if (S_ISREG(inode->i_mode)) {
4991 inode->i_op = &ext4_file_inode_operations;
4992 inode->i_fop = &ext4_file_operations;
4993 ext4_set_aops(inode);
4994 } else if (S_ISDIR(inode->i_mode)) {
4995 inode->i_op = &ext4_dir_inode_operations;
4996 inode->i_fop = &ext4_dir_operations;
4997 } else if (S_ISLNK(inode->i_mode)) {
4998 /* VFS does not allow setting these so must be corruption */
4999 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5000 ext4_error_inode(inode, function, line, 0,
5001 "iget: immutable or append flags "
5002 "not allowed on symlinks");
5003 ret = -EFSCORRUPTED;
5004 goto bad_inode;
5005 }
5006 if (IS_ENCRYPTED(inode)) {
5007 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5008 } else if (ext4_inode_is_fast_symlink(inode)) {
5009 inode->i_link = (char *)ei->i_data;
5010 inode->i_op = &ext4_fast_symlink_inode_operations;
5011 nd_terminate_link(ei->i_data, inode->i_size,
5012 sizeof(ei->i_data) - 1);
5013 } else {
5014 inode->i_op = &ext4_symlink_inode_operations;
5015 }
5016 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5017 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5018 inode->i_op = &ext4_special_inode_operations;
5019 if (raw_inode->i_block[0])
5020 init_special_inode(inode, inode->i_mode,
5021 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5022 else
5023 init_special_inode(inode, inode->i_mode,
5024 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5025 } else if (ino == EXT4_BOOT_LOADER_INO) {
5026 make_bad_inode(inode);
5027 } else {
5028 ret = -EFSCORRUPTED;
5029 ext4_error_inode(inode, function, line, 0,
5030 "iget: bogus i_mode (%o)", inode->i_mode);
5031 goto bad_inode;
5032 }
5033 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) {
5034 ext4_error_inode(inode, function, line, 0,
5035 "casefold flag without casefold feature");
5036 ret = -EFSCORRUPTED;
5037 goto bad_inode;
5038 }
5039 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
5040 ext4_error_inode(inode, function, line, 0, err_str);
5041 ret = -EFSCORRUPTED;
5042 goto bad_inode;
5043 }
5044
5045 brelse(iloc.bh);
5046 unlock_new_inode(inode);
5047 return inode;
5048
5049bad_inode:
5050 brelse(iloc.bh);
5051 iget_failed(inode);
5052 return ERR_PTR(ret);
5053}
5054
5055static void __ext4_update_other_inode_time(struct super_block *sb,
5056 unsigned long orig_ino,
5057 unsigned long ino,
5058 struct ext4_inode *raw_inode)
5059{
5060 struct inode *inode;
5061
5062 inode = find_inode_by_ino_rcu(sb, ino);
5063 if (!inode)
5064 return;
5065
5066 if (!inode_is_dirtytime_only(inode))
5067 return;
5068
5069 spin_lock(&inode->i_lock);
5070 if (inode_is_dirtytime_only(inode)) {
5071 struct ext4_inode_info *ei = EXT4_I(inode);
5072
5073 inode->i_state &= ~I_DIRTY_TIME;
5074 spin_unlock(&inode->i_lock);
5075
5076 spin_lock(&ei->i_raw_lock);
5077 EXT4_INODE_SET_CTIME(inode, raw_inode);
5078 EXT4_INODE_SET_MTIME(inode, raw_inode);
5079 EXT4_INODE_SET_ATIME(inode, raw_inode);
5080 ext4_inode_csum_set(inode, raw_inode, ei);
5081 spin_unlock(&ei->i_raw_lock);
5082 trace_ext4_other_inode_update_time(inode, orig_ino);
5083 return;
5084 }
5085 spin_unlock(&inode->i_lock);
5086}
5087
5088/*
5089 * Opportunistically update the other time fields for other inodes in
5090 * the same inode table block.
5091 */
5092static void ext4_update_other_inodes_time(struct super_block *sb,
5093 unsigned long orig_ino, char *buf)
5094{
5095 unsigned long ino;
5096 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5097 int inode_size = EXT4_INODE_SIZE(sb);
5098
5099 /*
5100 * Calculate the first inode in the inode table block. Inode
5101 * numbers are one-based. That is, the first inode in a block
5102 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5103 */
5104 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5105 rcu_read_lock();
5106 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5107 if (ino == orig_ino)
5108 continue;
5109 __ext4_update_other_inode_time(sb, orig_ino, ino,
5110 (struct ext4_inode *)buf);
5111 }
5112 rcu_read_unlock();
5113}
5114
5115/*
5116 * Post the struct inode info into an on-disk inode location in the
5117 * buffer-cache. This gobbles the caller's reference to the
5118 * buffer_head in the inode location struct.
5119 *
5120 * The caller must have write access to iloc->bh.
5121 */
5122static int ext4_do_update_inode(handle_t *handle,
5123 struct inode *inode,
5124 struct ext4_iloc *iloc)
5125{
5126 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5127 struct ext4_inode_info *ei = EXT4_I(inode);
5128 struct buffer_head *bh = iloc->bh;
5129 struct super_block *sb = inode->i_sb;
5130 int err;
5131 int need_datasync = 0, set_large_file = 0;
5132
5133 spin_lock(&ei->i_raw_lock);
5134
5135 /*
5136 * For fields not tracked in the in-memory inode, initialise them
5137 * to zero for new inodes.
5138 */
5139 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5140 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5141
5142 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5143 need_datasync = 1;
5144 if (ei->i_disksize > 0x7fffffffULL) {
5145 if (!ext4_has_feature_large_file(sb) ||
5146 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5147 set_large_file = 1;
5148 }
5149
5150 err = ext4_fill_raw_inode(inode, raw_inode);
5151 spin_unlock(&ei->i_raw_lock);
5152 if (err) {
5153 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5154 goto out_brelse;
5155 }
5156
5157 if (inode->i_sb->s_flags & SB_LAZYTIME)
5158 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5159 bh->b_data);
5160
5161 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5162 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5163 if (err)
5164 goto out_error;
5165 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5166 if (set_large_file) {
5167 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5168 err = ext4_journal_get_write_access(handle, sb,
5169 EXT4_SB(sb)->s_sbh,
5170 EXT4_JTR_NONE);
5171 if (err)
5172 goto out_error;
5173 lock_buffer(EXT4_SB(sb)->s_sbh);
5174 ext4_set_feature_large_file(sb);
5175 ext4_superblock_csum_set(sb);
5176 unlock_buffer(EXT4_SB(sb)->s_sbh);
5177 ext4_handle_sync(handle);
5178 err = ext4_handle_dirty_metadata(handle, NULL,
5179 EXT4_SB(sb)->s_sbh);
5180 }
5181 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5182out_error:
5183 ext4_std_error(inode->i_sb, err);
5184out_brelse:
5185 brelse(bh);
5186 return err;
5187}
5188
5189/*
5190 * ext4_write_inode()
5191 *
5192 * We are called from a few places:
5193 *
5194 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5195 * Here, there will be no transaction running. We wait for any running
5196 * transaction to commit.
5197 *
5198 * - Within flush work (sys_sync(), kupdate and such).
5199 * We wait on commit, if told to.
5200 *
5201 * - Within iput_final() -> write_inode_now()
5202 * We wait on commit, if told to.
5203 *
5204 * In all cases it is actually safe for us to return without doing anything,
5205 * because the inode has been copied into a raw inode buffer in
5206 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5207 * writeback.
5208 *
5209 * Note that we are absolutely dependent upon all inode dirtiers doing the
5210 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5211 * which we are interested.
5212 *
5213 * It would be a bug for them to not do this. The code:
5214 *
5215 * mark_inode_dirty(inode)
5216 * stuff();
5217 * inode->i_size = expr;
5218 *
5219 * is in error because write_inode() could occur while `stuff()' is running,
5220 * and the new i_size will be lost. Plus the inode will no longer be on the
5221 * superblock's dirty inode list.
5222 */
5223int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5224{
5225 int err;
5226
5227 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5228 return 0;
5229
5230 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5231 return -EIO;
5232
5233 if (EXT4_SB(inode->i_sb)->s_journal) {
5234 if (ext4_journal_current_handle()) {
5235 ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5236 dump_stack();
5237 return -EIO;
5238 }
5239
5240 /*
5241 * No need to force transaction in WB_SYNC_NONE mode. Also
5242 * ext4_sync_fs() will force the commit after everything is
5243 * written.
5244 */
5245 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5246 return 0;
5247
5248 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5249 EXT4_I(inode)->i_sync_tid);
5250 } else {
5251 struct ext4_iloc iloc;
5252
5253 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5254 if (err)
5255 return err;
5256 /*
5257 * sync(2) will flush the whole buffer cache. No need to do
5258 * it here separately for each inode.
5259 */
5260 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5261 sync_dirty_buffer(iloc.bh);
5262 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5263 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5264 "IO error syncing inode");
5265 err = -EIO;
5266 }
5267 brelse(iloc.bh);
5268 }
5269 return err;
5270}
5271
5272/*
5273 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5274 * buffers that are attached to a folio straddling i_size and are undergoing
5275 * commit. In that case we have to wait for commit to finish and try again.
5276 */
5277static void ext4_wait_for_tail_page_commit(struct inode *inode)
5278{
5279 unsigned offset;
5280 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5281 tid_t commit_tid;
5282 int ret;
5283 bool has_transaction;
5284
5285 offset = inode->i_size & (PAGE_SIZE - 1);
5286 /*
5287 * If the folio is fully truncated, we don't need to wait for any commit
5288 * (and we even should not as __ext4_journalled_invalidate_folio() may
5289 * strip all buffers from the folio but keep the folio dirty which can then
5290 * confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5291 * buffers). Also we don't need to wait for any commit if all buffers in
5292 * the folio remain valid. This is most beneficial for the common case of
5293 * blocksize == PAGESIZE.
5294 */
5295 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5296 return;
5297 while (1) {
5298 struct folio *folio = filemap_lock_folio(inode->i_mapping,
5299 inode->i_size >> PAGE_SHIFT);
5300 if (IS_ERR(folio))
5301 return;
5302 ret = __ext4_journalled_invalidate_folio(folio, offset,
5303 folio_size(folio) - offset);
5304 folio_unlock(folio);
5305 folio_put(folio);
5306 if (ret != -EBUSY)
5307 return;
5308 has_transaction = false;
5309 read_lock(&journal->j_state_lock);
5310 if (journal->j_committing_transaction) {
5311 commit_tid = journal->j_committing_transaction->t_tid;
5312 has_transaction = true;
5313 }
5314 read_unlock(&journal->j_state_lock);
5315 if (has_transaction)
5316 jbd2_log_wait_commit(journal, commit_tid);
5317 }
5318}
5319
5320/*
5321 * ext4_setattr()
5322 *
5323 * Called from notify_change.
5324 *
5325 * We want to trap VFS attempts to truncate the file as soon as
5326 * possible. In particular, we want to make sure that when the VFS
5327 * shrinks i_size, we put the inode on the orphan list and modify
5328 * i_disksize immediately, so that during the subsequent flushing of
5329 * dirty pages and freeing of disk blocks, we can guarantee that any
5330 * commit will leave the blocks being flushed in an unused state on
5331 * disk. (On recovery, the inode will get truncated and the blocks will
5332 * be freed, so we have a strong guarantee that no future commit will
5333 * leave these blocks visible to the user.)
5334 *
5335 * Another thing we have to assure is that if we are in ordered mode
5336 * and inode is still attached to the committing transaction, we must
5337 * we start writeout of all the dirty pages which are being truncated.
5338 * This way we are sure that all the data written in the previous
5339 * transaction are already on disk (truncate waits for pages under
5340 * writeback).
5341 *
5342 * Called with inode->i_rwsem down.
5343 */
5344int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5345 struct iattr *attr)
5346{
5347 struct inode *inode = d_inode(dentry);
5348 int error, rc = 0;
5349 int orphan = 0;
5350 const unsigned int ia_valid = attr->ia_valid;
5351 bool inc_ivers = true;
5352
5353 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5354 return -EIO;
5355
5356 if (unlikely(IS_IMMUTABLE(inode)))
5357 return -EPERM;
5358
5359 if (unlikely(IS_APPEND(inode) &&
5360 (ia_valid & (ATTR_MODE | ATTR_UID |
5361 ATTR_GID | ATTR_TIMES_SET))))
5362 return -EPERM;
5363
5364 error = setattr_prepare(idmap, dentry, attr);
5365 if (error)
5366 return error;
5367
5368 error = fscrypt_prepare_setattr(dentry, attr);
5369 if (error)
5370 return error;
5371
5372 error = fsverity_prepare_setattr(dentry, attr);
5373 if (error)
5374 return error;
5375
5376 if (is_quota_modification(idmap, inode, attr)) {
5377 error = dquot_initialize(inode);
5378 if (error)
5379 return error;
5380 }
5381
5382 if (i_uid_needs_update(idmap, attr, inode) ||
5383 i_gid_needs_update(idmap, attr, inode)) {
5384 handle_t *handle;
5385
5386 /* (user+group)*(old+new) structure, inode write (sb,
5387 * inode block, ? - but truncate inode update has it) */
5388 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5389 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5390 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5391 if (IS_ERR(handle)) {
5392 error = PTR_ERR(handle);
5393 goto err_out;
5394 }
5395
5396 /* dquot_transfer() calls back ext4_get_inode_usage() which
5397 * counts xattr inode references.
5398 */
5399 down_read(&EXT4_I(inode)->xattr_sem);
5400 error = dquot_transfer(idmap, inode, attr);
5401 up_read(&EXT4_I(inode)->xattr_sem);
5402
5403 if (error) {
5404 ext4_journal_stop(handle);
5405 return error;
5406 }
5407 /* Update corresponding info in inode so that everything is in
5408 * one transaction */
5409 i_uid_update(idmap, attr, inode);
5410 i_gid_update(idmap, attr, inode);
5411 error = ext4_mark_inode_dirty(handle, inode);
5412 ext4_journal_stop(handle);
5413 if (unlikely(error)) {
5414 return error;
5415 }
5416 }
5417
5418 if (attr->ia_valid & ATTR_SIZE) {
5419 handle_t *handle;
5420 loff_t oldsize = inode->i_size;
5421 loff_t old_disksize;
5422 int shrink = (attr->ia_size < inode->i_size);
5423
5424 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5425 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5426
5427 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5428 return -EFBIG;
5429 }
5430 }
5431 if (!S_ISREG(inode->i_mode)) {
5432 return -EINVAL;
5433 }
5434
5435 if (attr->ia_size == inode->i_size)
5436 inc_ivers = false;
5437
5438 if (shrink) {
5439 if (ext4_should_order_data(inode)) {
5440 error = ext4_begin_ordered_truncate(inode,
5441 attr->ia_size);
5442 if (error)
5443 goto err_out;
5444 }
5445 /*
5446 * Blocks are going to be removed from the inode. Wait
5447 * for dio in flight.
5448 */
5449 inode_dio_wait(inode);
5450 }
5451
5452 filemap_invalidate_lock(inode->i_mapping);
5453
5454 rc = ext4_break_layouts(inode);
5455 if (rc) {
5456 filemap_invalidate_unlock(inode->i_mapping);
5457 goto err_out;
5458 }
5459
5460 if (attr->ia_size != inode->i_size) {
5461 /* attach jbd2 jinode for EOF folio tail zeroing */
5462 if (attr->ia_size & (inode->i_sb->s_blocksize - 1) ||
5463 oldsize & (inode->i_sb->s_blocksize - 1)) {
5464 error = ext4_inode_attach_jinode(inode);
5465 if (error)
5466 goto err_out;
5467 }
5468
5469 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5470 if (IS_ERR(handle)) {
5471 error = PTR_ERR(handle);
5472 goto out_mmap_sem;
5473 }
5474 if (ext4_handle_valid(handle) && shrink) {
5475 error = ext4_orphan_add(handle, inode);
5476 orphan = 1;
5477 }
5478 /*
5479 * Update c/mtime and tail zero the EOF folio on
5480 * truncate up. ext4_truncate() handles the shrink case
5481 * below.
5482 */
5483 if (!shrink) {
5484 inode_set_mtime_to_ts(inode,
5485 inode_set_ctime_current(inode));
5486 if (oldsize & (inode->i_sb->s_blocksize - 1))
5487 ext4_block_truncate_page(handle,
5488 inode->i_mapping, oldsize);
5489 }
5490
5491 if (shrink)
5492 ext4_fc_track_range(handle, inode,
5493 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5494 inode->i_sb->s_blocksize_bits,
5495 EXT_MAX_BLOCKS - 1);
5496 else
5497 ext4_fc_track_range(
5498 handle, inode,
5499 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5500 inode->i_sb->s_blocksize_bits,
5501 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5502 inode->i_sb->s_blocksize_bits);
5503
5504 down_write(&EXT4_I(inode)->i_data_sem);
5505 old_disksize = EXT4_I(inode)->i_disksize;
5506 EXT4_I(inode)->i_disksize = attr->ia_size;
5507 rc = ext4_mark_inode_dirty(handle, inode);
5508 if (!error)
5509 error = rc;
5510 /*
5511 * We have to update i_size under i_data_sem together
5512 * with i_disksize to avoid races with writeback code
5513 * running ext4_wb_update_i_disksize().
5514 */
5515 if (!error)
5516 i_size_write(inode, attr->ia_size);
5517 else
5518 EXT4_I(inode)->i_disksize = old_disksize;
5519 up_write(&EXT4_I(inode)->i_data_sem);
5520 ext4_journal_stop(handle);
5521 if (error)
5522 goto out_mmap_sem;
5523 if (!shrink) {
5524 pagecache_isize_extended(inode, oldsize,
5525 inode->i_size);
5526 } else if (ext4_should_journal_data(inode)) {
5527 ext4_wait_for_tail_page_commit(inode);
5528 }
5529 }
5530
5531 /*
5532 * Truncate pagecache after we've waited for commit
5533 * in data=journal mode to make pages freeable.
5534 */
5535 truncate_pagecache(inode, inode->i_size);
5536 /*
5537 * Call ext4_truncate() even if i_size didn't change to
5538 * truncate possible preallocated blocks.
5539 */
5540 if (attr->ia_size <= oldsize) {
5541 rc = ext4_truncate(inode);
5542 if (rc)
5543 error = rc;
5544 }
5545out_mmap_sem:
5546 filemap_invalidate_unlock(inode->i_mapping);
5547 }
5548
5549 if (!error) {
5550 if (inc_ivers)
5551 inode_inc_iversion(inode);
5552 setattr_copy(idmap, inode, attr);
5553 mark_inode_dirty(inode);
5554 }
5555
5556 /*
5557 * If the call to ext4_truncate failed to get a transaction handle at
5558 * all, we need to clean up the in-core orphan list manually.
5559 */
5560 if (orphan && inode->i_nlink)
5561 ext4_orphan_del(NULL, inode);
5562
5563 if (!error && (ia_valid & ATTR_MODE))
5564 rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
5565
5566err_out:
5567 if (error)
5568 ext4_std_error(inode->i_sb, error);
5569 if (!error)
5570 error = rc;
5571 return error;
5572}
5573
5574u32 ext4_dio_alignment(struct inode *inode)
5575{
5576 if (fsverity_active(inode))
5577 return 0;
5578 if (ext4_should_journal_data(inode))
5579 return 0;
5580 if (ext4_has_inline_data(inode))
5581 return 0;
5582 if (IS_ENCRYPTED(inode)) {
5583 if (!fscrypt_dio_supported(inode))
5584 return 0;
5585 return i_blocksize(inode);
5586 }
5587 return 1; /* use the iomap defaults */
5588}
5589
5590int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
5591 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5592{
5593 struct inode *inode = d_inode(path->dentry);
5594 struct ext4_inode *raw_inode;
5595 struct ext4_inode_info *ei = EXT4_I(inode);
5596 unsigned int flags;
5597
5598 if ((request_mask & STATX_BTIME) &&
5599 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5600 stat->result_mask |= STATX_BTIME;
5601 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5602 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5603 }
5604
5605 /*
5606 * Return the DIO alignment restrictions if requested. We only return
5607 * this information when requested, since on encrypted files it might
5608 * take a fair bit of work to get if the file wasn't opened recently.
5609 */
5610 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5611 u32 dio_align = ext4_dio_alignment(inode);
5612
5613 stat->result_mask |= STATX_DIOALIGN;
5614 if (dio_align == 1) {
5615 struct block_device *bdev = inode->i_sb->s_bdev;
5616
5617 /* iomap defaults */
5618 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5619 stat->dio_offset_align = bdev_logical_block_size(bdev);
5620 } else {
5621 stat->dio_mem_align = dio_align;
5622 stat->dio_offset_align = dio_align;
5623 }
5624 }
5625
5626 if ((request_mask & STATX_WRITE_ATOMIC) && S_ISREG(inode->i_mode)) {
5627 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5628 unsigned int awu_min = 0, awu_max = 0;
5629
5630 if (ext4_inode_can_atomic_write(inode)) {
5631 awu_min = sbi->s_awu_min;
5632 awu_max = sbi->s_awu_max;
5633 }
5634
5635 generic_fill_statx_atomic_writes(stat, awu_min, awu_max);
5636 }
5637
5638 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5639 if (flags & EXT4_APPEND_FL)
5640 stat->attributes |= STATX_ATTR_APPEND;
5641 if (flags & EXT4_COMPR_FL)
5642 stat->attributes |= STATX_ATTR_COMPRESSED;
5643 if (flags & EXT4_ENCRYPT_FL)
5644 stat->attributes |= STATX_ATTR_ENCRYPTED;
5645 if (flags & EXT4_IMMUTABLE_FL)
5646 stat->attributes |= STATX_ATTR_IMMUTABLE;
5647 if (flags & EXT4_NODUMP_FL)
5648 stat->attributes |= STATX_ATTR_NODUMP;
5649 if (flags & EXT4_VERITY_FL)
5650 stat->attributes |= STATX_ATTR_VERITY;
5651
5652 stat->attributes_mask |= (STATX_ATTR_APPEND |
5653 STATX_ATTR_COMPRESSED |
5654 STATX_ATTR_ENCRYPTED |
5655 STATX_ATTR_IMMUTABLE |
5656 STATX_ATTR_NODUMP |
5657 STATX_ATTR_VERITY);
5658
5659 generic_fillattr(idmap, request_mask, inode, stat);
5660 return 0;
5661}
5662
5663int ext4_file_getattr(struct mnt_idmap *idmap,
5664 const struct path *path, struct kstat *stat,
5665 u32 request_mask, unsigned int query_flags)
5666{
5667 struct inode *inode = d_inode(path->dentry);
5668 u64 delalloc_blocks;
5669
5670 ext4_getattr(idmap, path, stat, request_mask, query_flags);
5671
5672 /*
5673 * If there is inline data in the inode, the inode will normally not
5674 * have data blocks allocated (it may have an external xattr block).
5675 * Report at least one sector for such files, so tools like tar, rsync,
5676 * others don't incorrectly think the file is completely sparse.
5677 */
5678 if (unlikely(ext4_has_inline_data(inode)))
5679 stat->blocks += (stat->size + 511) >> 9;
5680
5681 /*
5682 * We can't update i_blocks if the block allocation is delayed
5683 * otherwise in the case of system crash before the real block
5684 * allocation is done, we will have i_blocks inconsistent with
5685 * on-disk file blocks.
5686 * We always keep i_blocks updated together with real
5687 * allocation. But to not confuse with user, stat
5688 * will return the blocks that include the delayed allocation
5689 * blocks for this file.
5690 */
5691 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5692 EXT4_I(inode)->i_reserved_data_blocks);
5693 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5694 return 0;
5695}
5696
5697static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5698 int pextents)
5699{
5700 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5701 return ext4_ind_trans_blocks(inode, lblocks);
5702 return ext4_ext_index_trans_blocks(inode, pextents);
5703}
5704
5705/*
5706 * Account for index blocks, block groups bitmaps and block group
5707 * descriptor blocks if modify datablocks and index blocks
5708 * worse case, the indexs blocks spread over different block groups
5709 *
5710 * If datablocks are discontiguous, they are possible to spread over
5711 * different block groups too. If they are contiguous, with flexbg,
5712 * they could still across block group boundary.
5713 *
5714 * Also account for superblock, inode, quota and xattr blocks
5715 */
5716static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5717 int pextents)
5718{
5719 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5720 int gdpblocks;
5721 int idxblocks;
5722 int ret;
5723
5724 /*
5725 * How many index blocks need to touch to map @lblocks logical blocks
5726 * to @pextents physical extents?
5727 */
5728 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5729
5730 ret = idxblocks;
5731
5732 /*
5733 * Now let's see how many group bitmaps and group descriptors need
5734 * to account
5735 */
5736 groups = idxblocks + pextents;
5737 gdpblocks = groups;
5738 if (groups > ngroups)
5739 groups = ngroups;
5740 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5741 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5742
5743 /* bitmaps and block group descriptor blocks */
5744 ret += groups + gdpblocks;
5745
5746 /* Blocks for super block, inode, quota and xattr blocks */
5747 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5748
5749 return ret;
5750}
5751
5752/*
5753 * Calculate the total number of credits to reserve to fit
5754 * the modification of a single pages into a single transaction,
5755 * which may include multiple chunks of block allocations.
5756 *
5757 * This could be called via ext4_write_begin()
5758 *
5759 * We need to consider the worse case, when
5760 * one new block per extent.
5761 */
5762int ext4_writepage_trans_blocks(struct inode *inode)
5763{
5764 int bpp = ext4_journal_blocks_per_page(inode);
5765 int ret;
5766
5767 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5768
5769 /* Account for data blocks for journalled mode */
5770 if (ext4_should_journal_data(inode))
5771 ret += bpp;
5772 return ret;
5773}
5774
5775/*
5776 * Calculate the journal credits for a chunk of data modification.
5777 *
5778 * This is called from DIO, fallocate or whoever calling
5779 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5780 *
5781 * journal buffers for data blocks are not included here, as DIO
5782 * and fallocate do no need to journal data buffers.
5783 */
5784int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5785{
5786 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5787}
5788
5789/*
5790 * The caller must have previously called ext4_reserve_inode_write().
5791 * Give this, we know that the caller already has write access to iloc->bh.
5792 */
5793int ext4_mark_iloc_dirty(handle_t *handle,
5794 struct inode *inode, struct ext4_iloc *iloc)
5795{
5796 int err = 0;
5797
5798 if (unlikely(ext4_forced_shutdown(inode->i_sb))) {
5799 put_bh(iloc->bh);
5800 return -EIO;
5801 }
5802 ext4_fc_track_inode(handle, inode);
5803
5804 /* the do_update_inode consumes one bh->b_count */
5805 get_bh(iloc->bh);
5806
5807 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5808 err = ext4_do_update_inode(handle, inode, iloc);
5809 put_bh(iloc->bh);
5810 return err;
5811}
5812
5813/*
5814 * On success, We end up with an outstanding reference count against
5815 * iloc->bh. This _must_ be cleaned up later.
5816 */
5817
5818int
5819ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5820 struct ext4_iloc *iloc)
5821{
5822 int err;
5823
5824 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5825 return -EIO;
5826
5827 err = ext4_get_inode_loc(inode, iloc);
5828 if (!err) {
5829 BUFFER_TRACE(iloc->bh, "get_write_access");
5830 err = ext4_journal_get_write_access(handle, inode->i_sb,
5831 iloc->bh, EXT4_JTR_NONE);
5832 if (err) {
5833 brelse(iloc->bh);
5834 iloc->bh = NULL;
5835 }
5836 }
5837 ext4_std_error(inode->i_sb, err);
5838 return err;
5839}
5840
5841static int __ext4_expand_extra_isize(struct inode *inode,
5842 unsigned int new_extra_isize,
5843 struct ext4_iloc *iloc,
5844 handle_t *handle, int *no_expand)
5845{
5846 struct ext4_inode *raw_inode;
5847 struct ext4_xattr_ibody_header *header;
5848 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5849 struct ext4_inode_info *ei = EXT4_I(inode);
5850 int error;
5851
5852 /* this was checked at iget time, but double check for good measure */
5853 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5854 (ei->i_extra_isize & 3)) {
5855 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5856 ei->i_extra_isize,
5857 EXT4_INODE_SIZE(inode->i_sb));
5858 return -EFSCORRUPTED;
5859 }
5860 if ((new_extra_isize < ei->i_extra_isize) ||
5861 (new_extra_isize < 4) ||
5862 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5863 return -EINVAL; /* Should never happen */
5864
5865 raw_inode = ext4_raw_inode(iloc);
5866
5867 header = IHDR(inode, raw_inode);
5868
5869 /* No extended attributes present */
5870 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5871 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5872 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5873 EXT4_I(inode)->i_extra_isize, 0,
5874 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5875 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5876 return 0;
5877 }
5878
5879 /*
5880 * We may need to allocate external xattr block so we need quotas
5881 * initialized. Here we can be called with various locks held so we
5882 * cannot affort to initialize quotas ourselves. So just bail.
5883 */
5884 if (dquot_initialize_needed(inode))
5885 return -EAGAIN;
5886
5887 /* try to expand with EAs present */
5888 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5889 raw_inode, handle);
5890 if (error) {
5891 /*
5892 * Inode size expansion failed; don't try again
5893 */
5894 *no_expand = 1;
5895 }
5896
5897 return error;
5898}
5899
5900/*
5901 * Expand an inode by new_extra_isize bytes.
5902 * Returns 0 on success or negative error number on failure.
5903 */
5904static int ext4_try_to_expand_extra_isize(struct inode *inode,
5905 unsigned int new_extra_isize,
5906 struct ext4_iloc iloc,
5907 handle_t *handle)
5908{
5909 int no_expand;
5910 int error;
5911
5912 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5913 return -EOVERFLOW;
5914
5915 /*
5916 * In nojournal mode, we can immediately attempt to expand
5917 * the inode. When journaled, we first need to obtain extra
5918 * buffer credits since we may write into the EA block
5919 * with this same handle. If journal_extend fails, then it will
5920 * only result in a minor loss of functionality for that inode.
5921 * If this is felt to be critical, then e2fsck should be run to
5922 * force a large enough s_min_extra_isize.
5923 */
5924 if (ext4_journal_extend(handle,
5925 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5926 return -ENOSPC;
5927
5928 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5929 return -EBUSY;
5930
5931 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5932 handle, &no_expand);
5933 ext4_write_unlock_xattr(inode, &no_expand);
5934
5935 return error;
5936}
5937
5938int ext4_expand_extra_isize(struct inode *inode,
5939 unsigned int new_extra_isize,
5940 struct ext4_iloc *iloc)
5941{
5942 handle_t *handle;
5943 int no_expand;
5944 int error, rc;
5945
5946 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5947 brelse(iloc->bh);
5948 return -EOVERFLOW;
5949 }
5950
5951 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5952 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5953 if (IS_ERR(handle)) {
5954 error = PTR_ERR(handle);
5955 brelse(iloc->bh);
5956 return error;
5957 }
5958
5959 ext4_write_lock_xattr(inode, &no_expand);
5960
5961 BUFFER_TRACE(iloc->bh, "get_write_access");
5962 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5963 EXT4_JTR_NONE);
5964 if (error) {
5965 brelse(iloc->bh);
5966 goto out_unlock;
5967 }
5968
5969 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5970 handle, &no_expand);
5971
5972 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5973 if (!error)
5974 error = rc;
5975
5976out_unlock:
5977 ext4_write_unlock_xattr(inode, &no_expand);
5978 ext4_journal_stop(handle);
5979 return error;
5980}
5981
5982/*
5983 * What we do here is to mark the in-core inode as clean with respect to inode
5984 * dirtiness (it may still be data-dirty).
5985 * This means that the in-core inode may be reaped by prune_icache
5986 * without having to perform any I/O. This is a very good thing,
5987 * because *any* task may call prune_icache - even ones which
5988 * have a transaction open against a different journal.
5989 *
5990 * Is this cheating? Not really. Sure, we haven't written the
5991 * inode out, but prune_icache isn't a user-visible syncing function.
5992 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5993 * we start and wait on commits.
5994 */
5995int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5996 const char *func, unsigned int line)
5997{
5998 struct ext4_iloc iloc;
5999 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6000 int err;
6001
6002 might_sleep();
6003 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6004 err = ext4_reserve_inode_write(handle, inode, &iloc);
6005 if (err)
6006 goto out;
6007
6008 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6009 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6010 iloc, handle);
6011
6012 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
6013out:
6014 if (unlikely(err))
6015 ext4_error_inode_err(inode, func, line, 0, err,
6016 "mark_inode_dirty error");
6017 return err;
6018}
6019
6020/*
6021 * ext4_dirty_inode() is called from __mark_inode_dirty()
6022 *
6023 * We're really interested in the case where a file is being extended.
6024 * i_size has been changed by generic_commit_write() and we thus need
6025 * to include the updated inode in the current transaction.
6026 *
6027 * Also, dquot_alloc_block() will always dirty the inode when blocks
6028 * are allocated to the file.
6029 *
6030 * If the inode is marked synchronous, we don't honour that here - doing
6031 * so would cause a commit on atime updates, which we don't bother doing.
6032 * We handle synchronous inodes at the highest possible level.
6033 */
6034void ext4_dirty_inode(struct inode *inode, int flags)
6035{
6036 handle_t *handle;
6037
6038 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6039 if (IS_ERR(handle))
6040 return;
6041 ext4_mark_inode_dirty(handle, inode);
6042 ext4_journal_stop(handle);
6043}
6044
6045int ext4_change_inode_journal_flag(struct inode *inode, int val)
6046{
6047 journal_t *journal;
6048 handle_t *handle;
6049 int err;
6050 int alloc_ctx;
6051
6052 /*
6053 * We have to be very careful here: changing a data block's
6054 * journaling status dynamically is dangerous. If we write a
6055 * data block to the journal, change the status and then delete
6056 * that block, we risk forgetting to revoke the old log record
6057 * from the journal and so a subsequent replay can corrupt data.
6058 * So, first we make sure that the journal is empty and that
6059 * nobody is changing anything.
6060 */
6061
6062 journal = EXT4_JOURNAL(inode);
6063 if (!journal)
6064 return 0;
6065 if (is_journal_aborted(journal))
6066 return -EROFS;
6067
6068 /* Wait for all existing dio workers */
6069 inode_dio_wait(inode);
6070
6071 /*
6072 * Before flushing the journal and switching inode's aops, we have
6073 * to flush all dirty data the inode has. There can be outstanding
6074 * delayed allocations, there can be unwritten extents created by
6075 * fallocate or buffered writes in dioread_nolock mode covered by
6076 * dirty data which can be converted only after flushing the dirty
6077 * data (and journalled aops don't know how to handle these cases).
6078 */
6079 if (val) {
6080 filemap_invalidate_lock(inode->i_mapping);
6081 err = filemap_write_and_wait(inode->i_mapping);
6082 if (err < 0) {
6083 filemap_invalidate_unlock(inode->i_mapping);
6084 return err;
6085 }
6086 }
6087
6088 alloc_ctx = ext4_writepages_down_write(inode->i_sb);
6089 jbd2_journal_lock_updates(journal);
6090
6091 /*
6092 * OK, there are no updates running now, and all cached data is
6093 * synced to disk. We are now in a completely consistent state
6094 * which doesn't have anything in the journal, and we know that
6095 * no filesystem updates are running, so it is safe to modify
6096 * the inode's in-core data-journaling state flag now.
6097 */
6098
6099 if (val)
6100 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6101 else {
6102 err = jbd2_journal_flush(journal, 0);
6103 if (err < 0) {
6104 jbd2_journal_unlock_updates(journal);
6105 ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6106 return err;
6107 }
6108 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6109 }
6110 ext4_set_aops(inode);
6111
6112 jbd2_journal_unlock_updates(journal);
6113 ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6114
6115 if (val)
6116 filemap_invalidate_unlock(inode->i_mapping);
6117
6118 /* Finally we can mark the inode as dirty. */
6119
6120 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6121 if (IS_ERR(handle))
6122 return PTR_ERR(handle);
6123
6124 ext4_fc_mark_ineligible(inode->i_sb,
6125 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6126 err = ext4_mark_inode_dirty(handle, inode);
6127 ext4_handle_sync(handle);
6128 ext4_journal_stop(handle);
6129 ext4_std_error(inode->i_sb, err);
6130
6131 return err;
6132}
6133
6134static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6135 struct buffer_head *bh)
6136{
6137 return !buffer_mapped(bh);
6138}
6139
6140vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6141{
6142 struct vm_area_struct *vma = vmf->vma;
6143 struct folio *folio = page_folio(vmf->page);
6144 loff_t size;
6145 unsigned long len;
6146 int err;
6147 vm_fault_t ret;
6148 struct file *file = vma->vm_file;
6149 struct inode *inode = file_inode(file);
6150 struct address_space *mapping = inode->i_mapping;
6151 handle_t *handle;
6152 get_block_t *get_block;
6153 int retries = 0;
6154
6155 if (unlikely(IS_IMMUTABLE(inode)))
6156 return VM_FAULT_SIGBUS;
6157
6158 sb_start_pagefault(inode->i_sb);
6159 file_update_time(vma->vm_file);
6160
6161 filemap_invalidate_lock_shared(mapping);
6162
6163 err = ext4_convert_inline_data(inode);
6164 if (err)
6165 goto out_ret;
6166
6167 /*
6168 * On data journalling we skip straight to the transaction handle:
6169 * there's no delalloc; page truncated will be checked later; the
6170 * early return w/ all buffers mapped (calculates size/len) can't
6171 * be used; and there's no dioread_nolock, so only ext4_get_block.
6172 */
6173 if (ext4_should_journal_data(inode))
6174 goto retry_alloc;
6175
6176 /* Delalloc case is easy... */
6177 if (test_opt(inode->i_sb, DELALLOC) &&
6178 !ext4_nonda_switch(inode->i_sb)) {
6179 do {
6180 err = block_page_mkwrite(vma, vmf,
6181 ext4_da_get_block_prep);
6182 } while (err == -ENOSPC &&
6183 ext4_should_retry_alloc(inode->i_sb, &retries));
6184 goto out_ret;
6185 }
6186
6187 folio_lock(folio);
6188 size = i_size_read(inode);
6189 /* Page got truncated from under us? */
6190 if (folio->mapping != mapping || folio_pos(folio) > size) {
6191 folio_unlock(folio);
6192 ret = VM_FAULT_NOPAGE;
6193 goto out;
6194 }
6195
6196 len = folio_size(folio);
6197 if (folio_pos(folio) + len > size)
6198 len = size - folio_pos(folio);
6199 /*
6200 * Return if we have all the buffers mapped. This avoids the need to do
6201 * journal_start/journal_stop which can block and take a long time
6202 *
6203 * This cannot be done for data journalling, as we have to add the
6204 * inode to the transaction's list to writeprotect pages on commit.
6205 */
6206 if (folio_buffers(folio)) {
6207 if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6208 0, len, NULL,
6209 ext4_bh_unmapped)) {
6210 /* Wait so that we don't change page under IO */
6211 folio_wait_stable(folio);
6212 ret = VM_FAULT_LOCKED;
6213 goto out;
6214 }
6215 }
6216 folio_unlock(folio);
6217 /* OK, we need to fill the hole... */
6218 if (ext4_should_dioread_nolock(inode))
6219 get_block = ext4_get_block_unwritten;
6220 else
6221 get_block = ext4_get_block;
6222retry_alloc:
6223 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6224 ext4_writepage_trans_blocks(inode));
6225 if (IS_ERR(handle)) {
6226 ret = VM_FAULT_SIGBUS;
6227 goto out;
6228 }
6229 /*
6230 * Data journalling can't use block_page_mkwrite() because it
6231 * will set_buffer_dirty() before do_journal_get_write_access()
6232 * thus might hit warning messages for dirty metadata buffers.
6233 */
6234 if (!ext4_should_journal_data(inode)) {
6235 err = block_page_mkwrite(vma, vmf, get_block);
6236 } else {
6237 folio_lock(folio);
6238 size = i_size_read(inode);
6239 /* Page got truncated from under us? */
6240 if (folio->mapping != mapping || folio_pos(folio) > size) {
6241 ret = VM_FAULT_NOPAGE;
6242 goto out_error;
6243 }
6244
6245 len = folio_size(folio);
6246 if (folio_pos(folio) + len > size)
6247 len = size - folio_pos(folio);
6248
6249 err = ext4_block_write_begin(handle, folio, 0, len,
6250 ext4_get_block);
6251 if (!err) {
6252 ret = VM_FAULT_SIGBUS;
6253 if (ext4_journal_folio_buffers(handle, folio, len))
6254 goto out_error;
6255 } else {
6256 folio_unlock(folio);
6257 }
6258 }
6259 ext4_journal_stop(handle);
6260 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6261 goto retry_alloc;
6262out_ret:
6263 ret = vmf_fs_error(err);
6264out:
6265 filemap_invalidate_unlock_shared(mapping);
6266 sb_end_pagefault(inode->i_sb);
6267 return ret;
6268out_error:
6269 folio_unlock(folio);
6270 ext4_journal_stop(handle);
6271 goto out;
6272}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/inode.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/inode.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
18 *
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
20 */
21
22#include <linux/fs.h>
23#include <linux/time.h>
24#include <linux/highuid.h>
25#include <linux/pagemap.h>
26#include <linux/dax.h>
27#include <linux/quotaops.h>
28#include <linux/string.h>
29#include <linux/buffer_head.h>
30#include <linux/writeback.h>
31#include <linux/pagevec.h>
32#include <linux/mpage.h>
33#include <linux/namei.h>
34#include <linux/uio.h>
35#include <linux/bio.h>
36#include <linux/workqueue.h>
37#include <linux/kernel.h>
38#include <linux/printk.h>
39#include <linux/slab.h>
40#include <linux/bitops.h>
41#include <linux/iomap.h>
42#include <linux/iversion.h>
43
44#include "ext4_jbd2.h"
45#include "xattr.h"
46#include "acl.h"
47#include "truncate.h"
48
49#include <trace/events/ext4.h>
50
51#define MPAGE_DA_EXTENT_TAIL 0x01
52
53static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
54 struct ext4_inode_info *ei)
55{
56 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
57 __u32 csum;
58 __u16 dummy_csum = 0;
59 int offset = offsetof(struct ext4_inode, i_checksum_lo);
60 unsigned int csum_size = sizeof(dummy_csum);
61
62 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
63 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
64 offset += csum_size;
65 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
66 EXT4_GOOD_OLD_INODE_SIZE - offset);
67
68 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
69 offset = offsetof(struct ext4_inode, i_checksum_hi);
70 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
71 EXT4_GOOD_OLD_INODE_SIZE,
72 offset - EXT4_GOOD_OLD_INODE_SIZE);
73 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
74 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
75 csum_size);
76 offset += csum_size;
77 }
78 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
79 EXT4_INODE_SIZE(inode->i_sb) - offset);
80 }
81
82 return csum;
83}
84
85static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
86 struct ext4_inode_info *ei)
87{
88 __u32 provided, calculated;
89
90 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
91 cpu_to_le32(EXT4_OS_LINUX) ||
92 !ext4_has_metadata_csum(inode->i_sb))
93 return 1;
94
95 provided = le16_to_cpu(raw->i_checksum_lo);
96 calculated = ext4_inode_csum(inode, raw, ei);
97 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
98 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
99 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
100 else
101 calculated &= 0xFFFF;
102
103 return provided == calculated;
104}
105
106static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
107 struct ext4_inode_info *ei)
108{
109 __u32 csum;
110
111 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
112 cpu_to_le32(EXT4_OS_LINUX) ||
113 !ext4_has_metadata_csum(inode->i_sb))
114 return;
115
116 csum = ext4_inode_csum(inode, raw, ei);
117 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
118 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
119 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
120 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
121}
122
123static inline int ext4_begin_ordered_truncate(struct inode *inode,
124 loff_t new_size)
125{
126 trace_ext4_begin_ordered_truncate(inode, new_size);
127 /*
128 * If jinode is zero, then we never opened the file for
129 * writing, so there's no need to call
130 * jbd2_journal_begin_ordered_truncate() since there's no
131 * outstanding writes we need to flush.
132 */
133 if (!EXT4_I(inode)->jinode)
134 return 0;
135 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
136 EXT4_I(inode)->jinode,
137 new_size);
138}
139
140static void ext4_invalidatepage(struct page *page, unsigned int offset,
141 unsigned int length);
142static int __ext4_journalled_writepage(struct page *page, unsigned int len);
143static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
144static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
145 int pextents);
146
147/*
148 * Test whether an inode is a fast symlink.
149 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 */
151int ext4_inode_is_fast_symlink(struct inode *inode)
152{
153 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
154 int ea_blocks = EXT4_I(inode)->i_file_acl ?
155 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
156
157 if (ext4_has_inline_data(inode))
158 return 0;
159
160 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
161 }
162 return S_ISLNK(inode->i_mode) && inode->i_size &&
163 (inode->i_size < EXT4_N_BLOCKS * 4);
164}
165
166/*
167 * Restart the transaction associated with *handle. This does a commit,
168 * so before we call here everything must be consistently dirtied against
169 * this transaction.
170 */
171int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
172 int nblocks)
173{
174 int ret;
175
176 /*
177 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
178 * moment, get_block can be called only for blocks inside i_size since
179 * page cache has been already dropped and writes are blocked by
180 * i_mutex. So we can safely drop the i_data_sem here.
181 */
182 BUG_ON(EXT4_JOURNAL(inode) == NULL);
183 jbd_debug(2, "restarting handle %p\n", handle);
184 up_write(&EXT4_I(inode)->i_data_sem);
185 ret = ext4_journal_restart(handle, nblocks);
186 down_write(&EXT4_I(inode)->i_data_sem);
187 ext4_discard_preallocations(inode);
188
189 return ret;
190}
191
192/*
193 * Called at the last iput() if i_nlink is zero.
194 */
195void ext4_evict_inode(struct inode *inode)
196{
197 handle_t *handle;
198 int err;
199 int extra_credits = 3;
200 struct ext4_xattr_inode_array *ea_inode_array = NULL;
201
202 trace_ext4_evict_inode(inode);
203
204 if (inode->i_nlink) {
205 /*
206 * When journalling data dirty buffers are tracked only in the
207 * journal. So although mm thinks everything is clean and
208 * ready for reaping the inode might still have some pages to
209 * write in the running transaction or waiting to be
210 * checkpointed. Thus calling jbd2_journal_invalidatepage()
211 * (via truncate_inode_pages()) to discard these buffers can
212 * cause data loss. Also even if we did not discard these
213 * buffers, we would have no way to find them after the inode
214 * is reaped and thus user could see stale data if he tries to
215 * read them before the transaction is checkpointed. So be
216 * careful and force everything to disk here... We use
217 * ei->i_datasync_tid to store the newest transaction
218 * containing inode's data.
219 *
220 * Note that directories do not have this problem because they
221 * don't use page cache.
222 */
223 if (inode->i_ino != EXT4_JOURNAL_INO &&
224 ext4_should_journal_data(inode) &&
225 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
226 inode->i_data.nrpages) {
227 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
228 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
229
230 jbd2_complete_transaction(journal, commit_tid);
231 filemap_write_and_wait(&inode->i_data);
232 }
233 truncate_inode_pages_final(&inode->i_data);
234
235 goto no_delete;
236 }
237
238 if (is_bad_inode(inode))
239 goto no_delete;
240 dquot_initialize(inode);
241
242 if (ext4_should_order_data(inode))
243 ext4_begin_ordered_truncate(inode, 0);
244 truncate_inode_pages_final(&inode->i_data);
245
246 /*
247 * Protect us against freezing - iput() caller didn't have to have any
248 * protection against it
249 */
250 sb_start_intwrite(inode->i_sb);
251
252 if (!IS_NOQUOTA(inode))
253 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
254
255 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
256 ext4_blocks_for_truncate(inode)+extra_credits);
257 if (IS_ERR(handle)) {
258 ext4_std_error(inode->i_sb, PTR_ERR(handle));
259 /*
260 * If we're going to skip the normal cleanup, we still need to
261 * make sure that the in-core orphan linked list is properly
262 * cleaned up.
263 */
264 ext4_orphan_del(NULL, inode);
265 sb_end_intwrite(inode->i_sb);
266 goto no_delete;
267 }
268
269 if (IS_SYNC(inode))
270 ext4_handle_sync(handle);
271
272 /*
273 * Set inode->i_size to 0 before calling ext4_truncate(). We need
274 * special handling of symlinks here because i_size is used to
275 * determine whether ext4_inode_info->i_data contains symlink data or
276 * block mappings. Setting i_size to 0 will remove its fast symlink
277 * status. Erase i_data so that it becomes a valid empty block map.
278 */
279 if (ext4_inode_is_fast_symlink(inode))
280 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
281 inode->i_size = 0;
282 err = ext4_mark_inode_dirty(handle, inode);
283 if (err) {
284 ext4_warning(inode->i_sb,
285 "couldn't mark inode dirty (err %d)", err);
286 goto stop_handle;
287 }
288 if (inode->i_blocks) {
289 err = ext4_truncate(inode);
290 if (err) {
291 ext4_error(inode->i_sb,
292 "couldn't truncate inode %lu (err %d)",
293 inode->i_ino, err);
294 goto stop_handle;
295 }
296 }
297
298 /* Remove xattr references. */
299 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
300 extra_credits);
301 if (err) {
302 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
303stop_handle:
304 ext4_journal_stop(handle);
305 ext4_orphan_del(NULL, inode);
306 sb_end_intwrite(inode->i_sb);
307 ext4_xattr_inode_array_free(ea_inode_array);
308 goto no_delete;
309 }
310
311 /*
312 * Kill off the orphan record which ext4_truncate created.
313 * AKPM: I think this can be inside the above `if'.
314 * Note that ext4_orphan_del() has to be able to cope with the
315 * deletion of a non-existent orphan - this is because we don't
316 * know if ext4_truncate() actually created an orphan record.
317 * (Well, we could do this if we need to, but heck - it works)
318 */
319 ext4_orphan_del(handle, inode);
320 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
321
322 /*
323 * One subtle ordering requirement: if anything has gone wrong
324 * (transaction abort, IO errors, whatever), then we can still
325 * do these next steps (the fs will already have been marked as
326 * having errors), but we can't free the inode if the mark_dirty
327 * fails.
328 */
329 if (ext4_mark_inode_dirty(handle, inode))
330 /* If that failed, just do the required in-core inode clear. */
331 ext4_clear_inode(inode);
332 else
333 ext4_free_inode(handle, inode);
334 ext4_journal_stop(handle);
335 sb_end_intwrite(inode->i_sb);
336 ext4_xattr_inode_array_free(ea_inode_array);
337 return;
338no_delete:
339 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
340}
341
342#ifdef CONFIG_QUOTA
343qsize_t *ext4_get_reserved_space(struct inode *inode)
344{
345 return &EXT4_I(inode)->i_reserved_quota;
346}
347#endif
348
349/*
350 * Called with i_data_sem down, which is important since we can call
351 * ext4_discard_preallocations() from here.
352 */
353void ext4_da_update_reserve_space(struct inode *inode,
354 int used, int quota_claim)
355{
356 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
357 struct ext4_inode_info *ei = EXT4_I(inode);
358
359 spin_lock(&ei->i_block_reservation_lock);
360 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
361 if (unlikely(used > ei->i_reserved_data_blocks)) {
362 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
363 "with only %d reserved data blocks",
364 __func__, inode->i_ino, used,
365 ei->i_reserved_data_blocks);
366 WARN_ON(1);
367 used = ei->i_reserved_data_blocks;
368 }
369
370 /* Update per-inode reservations */
371 ei->i_reserved_data_blocks -= used;
372 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
373
374 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
375
376 /* Update quota subsystem for data blocks */
377 if (quota_claim)
378 dquot_claim_block(inode, EXT4_C2B(sbi, used));
379 else {
380 /*
381 * We did fallocate with an offset that is already delayed
382 * allocated. So on delayed allocated writeback we should
383 * not re-claim the quota for fallocated blocks.
384 */
385 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
386 }
387
388 /*
389 * If we have done all the pending block allocations and if
390 * there aren't any writers on the inode, we can discard the
391 * inode's preallocations.
392 */
393 if ((ei->i_reserved_data_blocks == 0) &&
394 !inode_is_open_for_write(inode))
395 ext4_discard_preallocations(inode);
396}
397
398static int __check_block_validity(struct inode *inode, const char *func,
399 unsigned int line,
400 struct ext4_map_blocks *map)
401{
402 if (ext4_has_feature_journal(inode->i_sb) &&
403 (inode->i_ino ==
404 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
405 return 0;
406 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
407 map->m_len)) {
408 ext4_error_inode(inode, func, line, map->m_pblk,
409 "lblock %lu mapped to illegal pblock %llu "
410 "(length %d)", (unsigned long) map->m_lblk,
411 map->m_pblk, map->m_len);
412 return -EFSCORRUPTED;
413 }
414 return 0;
415}
416
417int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
418 ext4_lblk_t len)
419{
420 int ret;
421
422 if (IS_ENCRYPTED(inode))
423 return fscrypt_zeroout_range(inode, lblk, pblk, len);
424
425 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
426 if (ret > 0)
427 ret = 0;
428
429 return ret;
430}
431
432#define check_block_validity(inode, map) \
433 __check_block_validity((inode), __func__, __LINE__, (map))
434
435#ifdef ES_AGGRESSIVE_TEST
436static void ext4_map_blocks_es_recheck(handle_t *handle,
437 struct inode *inode,
438 struct ext4_map_blocks *es_map,
439 struct ext4_map_blocks *map,
440 int flags)
441{
442 int retval;
443
444 map->m_flags = 0;
445 /*
446 * There is a race window that the result is not the same.
447 * e.g. xfstests #223 when dioread_nolock enables. The reason
448 * is that we lookup a block mapping in extent status tree with
449 * out taking i_data_sem. So at the time the unwritten extent
450 * could be converted.
451 */
452 down_read(&EXT4_I(inode)->i_data_sem);
453 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
454 retval = ext4_ext_map_blocks(handle, inode, map, flags &
455 EXT4_GET_BLOCKS_KEEP_SIZE);
456 } else {
457 retval = ext4_ind_map_blocks(handle, inode, map, flags &
458 EXT4_GET_BLOCKS_KEEP_SIZE);
459 }
460 up_read((&EXT4_I(inode)->i_data_sem));
461
462 /*
463 * We don't check m_len because extent will be collpased in status
464 * tree. So the m_len might not equal.
465 */
466 if (es_map->m_lblk != map->m_lblk ||
467 es_map->m_flags != map->m_flags ||
468 es_map->m_pblk != map->m_pblk) {
469 printk("ES cache assertion failed for inode: %lu "
470 "es_cached ex [%d/%d/%llu/%x] != "
471 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472 inode->i_ino, es_map->m_lblk, es_map->m_len,
473 es_map->m_pblk, es_map->m_flags, map->m_lblk,
474 map->m_len, map->m_pblk, map->m_flags,
475 retval, flags);
476 }
477}
478#endif /* ES_AGGRESSIVE_TEST */
479
480/*
481 * The ext4_map_blocks() function tries to look up the requested blocks,
482 * and returns if the blocks are already mapped.
483 *
484 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485 * and store the allocated blocks in the result buffer head and mark it
486 * mapped.
487 *
488 * If file type is extents based, it will call ext4_ext_map_blocks(),
489 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
490 * based files
491 *
492 * On success, it returns the number of blocks being mapped or allocated. if
493 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
494 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
495 *
496 * It returns 0 if plain look up failed (blocks have not been allocated), in
497 * that case, @map is returned as unmapped but we still do fill map->m_len to
498 * indicate the length of a hole starting at map->m_lblk.
499 *
500 * It returns the error in case of allocation failure.
501 */
502int ext4_map_blocks(handle_t *handle, struct inode *inode,
503 struct ext4_map_blocks *map, int flags)
504{
505 struct extent_status es;
506 int retval;
507 int ret = 0;
508#ifdef ES_AGGRESSIVE_TEST
509 struct ext4_map_blocks orig_map;
510
511 memcpy(&orig_map, map, sizeof(*map));
512#endif
513
514 map->m_flags = 0;
515 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
516 "logical block %lu\n", inode->i_ino, flags, map->m_len,
517 (unsigned long) map->m_lblk);
518
519 /*
520 * ext4_map_blocks returns an int, and m_len is an unsigned int
521 */
522 if (unlikely(map->m_len > INT_MAX))
523 map->m_len = INT_MAX;
524
525 /* We can handle the block number less than EXT_MAX_BLOCKS */
526 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
527 return -EFSCORRUPTED;
528
529 /* Lookup extent status tree firstly */
530 if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
531 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
532 map->m_pblk = ext4_es_pblock(&es) +
533 map->m_lblk - es.es_lblk;
534 map->m_flags |= ext4_es_is_written(&es) ?
535 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
536 retval = es.es_len - (map->m_lblk - es.es_lblk);
537 if (retval > map->m_len)
538 retval = map->m_len;
539 map->m_len = retval;
540 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
541 map->m_pblk = 0;
542 retval = es.es_len - (map->m_lblk - es.es_lblk);
543 if (retval > map->m_len)
544 retval = map->m_len;
545 map->m_len = retval;
546 retval = 0;
547 } else {
548 BUG();
549 }
550#ifdef ES_AGGRESSIVE_TEST
551 ext4_map_blocks_es_recheck(handle, inode, map,
552 &orig_map, flags);
553#endif
554 goto found;
555 }
556
557 /*
558 * Try to see if we can get the block without requesting a new
559 * file system block.
560 */
561 down_read(&EXT4_I(inode)->i_data_sem);
562 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
563 retval = ext4_ext_map_blocks(handle, inode, map, flags &
564 EXT4_GET_BLOCKS_KEEP_SIZE);
565 } else {
566 retval = ext4_ind_map_blocks(handle, inode, map, flags &
567 EXT4_GET_BLOCKS_KEEP_SIZE);
568 }
569 if (retval > 0) {
570 unsigned int status;
571
572 if (unlikely(retval != map->m_len)) {
573 ext4_warning(inode->i_sb,
574 "ES len assertion failed for inode "
575 "%lu: retval %d != map->m_len %d",
576 inode->i_ino, retval, map->m_len);
577 WARN_ON(1);
578 }
579
580 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
581 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
582 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
583 !(status & EXTENT_STATUS_WRITTEN) &&
584 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
585 map->m_lblk + map->m_len - 1))
586 status |= EXTENT_STATUS_DELAYED;
587 ret = ext4_es_insert_extent(inode, map->m_lblk,
588 map->m_len, map->m_pblk, status);
589 if (ret < 0)
590 retval = ret;
591 }
592 up_read((&EXT4_I(inode)->i_data_sem));
593
594found:
595 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
596 ret = check_block_validity(inode, map);
597 if (ret != 0)
598 return ret;
599 }
600
601 /* If it is only a block(s) look up */
602 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
603 return retval;
604
605 /*
606 * Returns if the blocks have already allocated
607 *
608 * Note that if blocks have been preallocated
609 * ext4_ext_get_block() returns the create = 0
610 * with buffer head unmapped.
611 */
612 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
613 /*
614 * If we need to convert extent to unwritten
615 * we continue and do the actual work in
616 * ext4_ext_map_blocks()
617 */
618 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
619 return retval;
620
621 /*
622 * Here we clear m_flags because after allocating an new extent,
623 * it will be set again.
624 */
625 map->m_flags &= ~EXT4_MAP_FLAGS;
626
627 /*
628 * New blocks allocate and/or writing to unwritten extent
629 * will possibly result in updating i_data, so we take
630 * the write lock of i_data_sem, and call get_block()
631 * with create == 1 flag.
632 */
633 down_write(&EXT4_I(inode)->i_data_sem);
634
635 /*
636 * We need to check for EXT4 here because migrate
637 * could have changed the inode type in between
638 */
639 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
640 retval = ext4_ext_map_blocks(handle, inode, map, flags);
641 } else {
642 retval = ext4_ind_map_blocks(handle, inode, map, flags);
643
644 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
645 /*
646 * We allocated new blocks which will result in
647 * i_data's format changing. Force the migrate
648 * to fail by clearing migrate flags
649 */
650 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
651 }
652
653 /*
654 * Update reserved blocks/metadata blocks after successful
655 * block allocation which had been deferred till now. We don't
656 * support fallocate for non extent files. So we can update
657 * reserve space here.
658 */
659 if ((retval > 0) &&
660 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
661 ext4_da_update_reserve_space(inode, retval, 1);
662 }
663
664 if (retval > 0) {
665 unsigned int status;
666
667 if (unlikely(retval != map->m_len)) {
668 ext4_warning(inode->i_sb,
669 "ES len assertion failed for inode "
670 "%lu: retval %d != map->m_len %d",
671 inode->i_ino, retval, map->m_len);
672 WARN_ON(1);
673 }
674
675 /*
676 * We have to zeroout blocks before inserting them into extent
677 * status tree. Otherwise someone could look them up there and
678 * use them before they are really zeroed. We also have to
679 * unmap metadata before zeroing as otherwise writeback can
680 * overwrite zeros with stale data from block device.
681 */
682 if (flags & EXT4_GET_BLOCKS_ZERO &&
683 map->m_flags & EXT4_MAP_MAPPED &&
684 map->m_flags & EXT4_MAP_NEW) {
685 ret = ext4_issue_zeroout(inode, map->m_lblk,
686 map->m_pblk, map->m_len);
687 if (ret) {
688 retval = ret;
689 goto out_sem;
690 }
691 }
692
693 /*
694 * If the extent has been zeroed out, we don't need to update
695 * extent status tree.
696 */
697 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
698 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
699 if (ext4_es_is_written(&es))
700 goto out_sem;
701 }
702 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
703 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
704 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
705 !(status & EXTENT_STATUS_WRITTEN) &&
706 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
707 map->m_lblk + map->m_len - 1))
708 status |= EXTENT_STATUS_DELAYED;
709 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
710 map->m_pblk, status);
711 if (ret < 0) {
712 retval = ret;
713 goto out_sem;
714 }
715 }
716
717out_sem:
718 up_write((&EXT4_I(inode)->i_data_sem));
719 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
720 ret = check_block_validity(inode, map);
721 if (ret != 0)
722 return ret;
723
724 /*
725 * Inodes with freshly allocated blocks where contents will be
726 * visible after transaction commit must be on transaction's
727 * ordered data list.
728 */
729 if (map->m_flags & EXT4_MAP_NEW &&
730 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
731 !(flags & EXT4_GET_BLOCKS_ZERO) &&
732 !ext4_is_quota_file(inode) &&
733 ext4_should_order_data(inode)) {
734 loff_t start_byte =
735 (loff_t)map->m_lblk << inode->i_blkbits;
736 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
737
738 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
739 ret = ext4_jbd2_inode_add_wait(handle, inode,
740 start_byte, length);
741 else
742 ret = ext4_jbd2_inode_add_write(handle, inode,
743 start_byte, length);
744 if (ret)
745 return ret;
746 }
747 }
748 return retval;
749}
750
751/*
752 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
753 * we have to be careful as someone else may be manipulating b_state as well.
754 */
755static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
756{
757 unsigned long old_state;
758 unsigned long new_state;
759
760 flags &= EXT4_MAP_FLAGS;
761
762 /* Dummy buffer_head? Set non-atomically. */
763 if (!bh->b_page) {
764 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
765 return;
766 }
767 /*
768 * Someone else may be modifying b_state. Be careful! This is ugly but
769 * once we get rid of using bh as a container for mapping information
770 * to pass to / from get_block functions, this can go away.
771 */
772 do {
773 old_state = READ_ONCE(bh->b_state);
774 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
775 } while (unlikely(
776 cmpxchg(&bh->b_state, old_state, new_state) != old_state));
777}
778
779static int _ext4_get_block(struct inode *inode, sector_t iblock,
780 struct buffer_head *bh, int flags)
781{
782 struct ext4_map_blocks map;
783 int ret = 0;
784
785 if (ext4_has_inline_data(inode))
786 return -ERANGE;
787
788 map.m_lblk = iblock;
789 map.m_len = bh->b_size >> inode->i_blkbits;
790
791 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
792 flags);
793 if (ret > 0) {
794 map_bh(bh, inode->i_sb, map.m_pblk);
795 ext4_update_bh_state(bh, map.m_flags);
796 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
797 ret = 0;
798 } else if (ret == 0) {
799 /* hole case, need to fill in bh->b_size */
800 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
801 }
802 return ret;
803}
804
805int ext4_get_block(struct inode *inode, sector_t iblock,
806 struct buffer_head *bh, int create)
807{
808 return _ext4_get_block(inode, iblock, bh,
809 create ? EXT4_GET_BLOCKS_CREATE : 0);
810}
811
812/*
813 * Get block function used when preparing for buffered write if we require
814 * creating an unwritten extent if blocks haven't been allocated. The extent
815 * will be converted to written after the IO is complete.
816 */
817int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
818 struct buffer_head *bh_result, int create)
819{
820 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
821 inode->i_ino, create);
822 return _ext4_get_block(inode, iblock, bh_result,
823 EXT4_GET_BLOCKS_IO_CREATE_EXT);
824}
825
826/* Maximum number of blocks we map for direct IO at once. */
827#define DIO_MAX_BLOCKS 4096
828
829/*
830 * Get blocks function for the cases that need to start a transaction -
831 * generally difference cases of direct IO and DAX IO. It also handles retries
832 * in case of ENOSPC.
833 */
834static int ext4_get_block_trans(struct inode *inode, sector_t iblock,
835 struct buffer_head *bh_result, int flags)
836{
837 int dio_credits;
838 handle_t *handle;
839 int retries = 0;
840 int ret;
841
842 /* Trim mapping request to maximum we can map at once for DIO */
843 if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS)
844 bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits;
845 dio_credits = ext4_chunk_trans_blocks(inode,
846 bh_result->b_size >> inode->i_blkbits);
847retry:
848 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
849 if (IS_ERR(handle))
850 return PTR_ERR(handle);
851
852 ret = _ext4_get_block(inode, iblock, bh_result, flags);
853 ext4_journal_stop(handle);
854
855 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
856 goto retry;
857 return ret;
858}
859
860/* Get block function for DIO reads and writes to inodes without extents */
861int ext4_dio_get_block(struct inode *inode, sector_t iblock,
862 struct buffer_head *bh, int create)
863{
864 /* We don't expect handle for direct IO */
865 WARN_ON_ONCE(ext4_journal_current_handle());
866
867 if (!create)
868 return _ext4_get_block(inode, iblock, bh, 0);
869 return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE);
870}
871
872/*
873 * Get block function for AIO DIO writes when we create unwritten extent if
874 * blocks are not allocated yet. The extent will be converted to written
875 * after IO is complete.
876 */
877static int ext4_dio_get_block_unwritten_async(struct inode *inode,
878 sector_t iblock, struct buffer_head *bh_result, int create)
879{
880 int ret;
881
882 /* We don't expect handle for direct IO */
883 WARN_ON_ONCE(ext4_journal_current_handle());
884
885 ret = ext4_get_block_trans(inode, iblock, bh_result,
886 EXT4_GET_BLOCKS_IO_CREATE_EXT);
887
888 /*
889 * When doing DIO using unwritten extents, we need io_end to convert
890 * unwritten extents to written on IO completion. We allocate io_end
891 * once we spot unwritten extent and store it in b_private. Generic
892 * DIO code keeps b_private set and furthermore passes the value to
893 * our completion callback in 'private' argument.
894 */
895 if (!ret && buffer_unwritten(bh_result)) {
896 if (!bh_result->b_private) {
897 ext4_io_end_t *io_end;
898
899 io_end = ext4_init_io_end(inode, GFP_KERNEL);
900 if (!io_end)
901 return -ENOMEM;
902 bh_result->b_private = io_end;
903 ext4_set_io_unwritten_flag(inode, io_end);
904 }
905 set_buffer_defer_completion(bh_result);
906 }
907
908 return ret;
909}
910
911/*
912 * Get block function for non-AIO DIO writes when we create unwritten extent if
913 * blocks are not allocated yet. The extent will be converted to written
914 * after IO is complete by ext4_direct_IO_write().
915 */
916static int ext4_dio_get_block_unwritten_sync(struct inode *inode,
917 sector_t iblock, struct buffer_head *bh_result, int create)
918{
919 int ret;
920
921 /* We don't expect handle for direct IO */
922 WARN_ON_ONCE(ext4_journal_current_handle());
923
924 ret = ext4_get_block_trans(inode, iblock, bh_result,
925 EXT4_GET_BLOCKS_IO_CREATE_EXT);
926
927 /*
928 * Mark inode as having pending DIO writes to unwritten extents.
929 * ext4_direct_IO_write() checks this flag and converts extents to
930 * written.
931 */
932 if (!ret && buffer_unwritten(bh_result))
933 ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
934
935 return ret;
936}
937
938static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock,
939 struct buffer_head *bh_result, int create)
940{
941 int ret;
942
943 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
944 inode->i_ino, create);
945 /* We don't expect handle for direct IO */
946 WARN_ON_ONCE(ext4_journal_current_handle());
947
948 ret = _ext4_get_block(inode, iblock, bh_result, 0);
949 /*
950 * Blocks should have been preallocated! ext4_file_write_iter() checks
951 * that.
952 */
953 WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result));
954
955 return ret;
956}
957
958
959/*
960 * `handle' can be NULL if create is zero
961 */
962struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
963 ext4_lblk_t block, int map_flags)
964{
965 struct ext4_map_blocks map;
966 struct buffer_head *bh;
967 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
968 int err;
969
970 J_ASSERT(handle != NULL || create == 0);
971
972 map.m_lblk = block;
973 map.m_len = 1;
974 err = ext4_map_blocks(handle, inode, &map, map_flags);
975
976 if (err == 0)
977 return create ? ERR_PTR(-ENOSPC) : NULL;
978 if (err < 0)
979 return ERR_PTR(err);
980
981 bh = sb_getblk(inode->i_sb, map.m_pblk);
982 if (unlikely(!bh))
983 return ERR_PTR(-ENOMEM);
984 if (map.m_flags & EXT4_MAP_NEW) {
985 J_ASSERT(create != 0);
986 J_ASSERT(handle != NULL);
987
988 /*
989 * Now that we do not always journal data, we should
990 * keep in mind whether this should always journal the
991 * new buffer as metadata. For now, regular file
992 * writes use ext4_get_block instead, so it's not a
993 * problem.
994 */
995 lock_buffer(bh);
996 BUFFER_TRACE(bh, "call get_create_access");
997 err = ext4_journal_get_create_access(handle, bh);
998 if (unlikely(err)) {
999 unlock_buffer(bh);
1000 goto errout;
1001 }
1002 if (!buffer_uptodate(bh)) {
1003 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
1004 set_buffer_uptodate(bh);
1005 }
1006 unlock_buffer(bh);
1007 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1008 err = ext4_handle_dirty_metadata(handle, inode, bh);
1009 if (unlikely(err))
1010 goto errout;
1011 } else
1012 BUFFER_TRACE(bh, "not a new buffer");
1013 return bh;
1014errout:
1015 brelse(bh);
1016 return ERR_PTR(err);
1017}
1018
1019struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1020 ext4_lblk_t block, int map_flags)
1021{
1022 struct buffer_head *bh;
1023
1024 bh = ext4_getblk(handle, inode, block, map_flags);
1025 if (IS_ERR(bh))
1026 return bh;
1027 if (!bh || ext4_buffer_uptodate(bh))
1028 return bh;
1029 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
1030 wait_on_buffer(bh);
1031 if (buffer_uptodate(bh))
1032 return bh;
1033 put_bh(bh);
1034 return ERR_PTR(-EIO);
1035}
1036
1037/* Read a contiguous batch of blocks. */
1038int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
1039 bool wait, struct buffer_head **bhs)
1040{
1041 int i, err;
1042
1043 for (i = 0; i < bh_count; i++) {
1044 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
1045 if (IS_ERR(bhs[i])) {
1046 err = PTR_ERR(bhs[i]);
1047 bh_count = i;
1048 goto out_brelse;
1049 }
1050 }
1051
1052 for (i = 0; i < bh_count; i++)
1053 /* Note that NULL bhs[i] is valid because of holes. */
1054 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
1055 ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
1056 &bhs[i]);
1057
1058 if (!wait)
1059 return 0;
1060
1061 for (i = 0; i < bh_count; i++)
1062 if (bhs[i])
1063 wait_on_buffer(bhs[i]);
1064
1065 for (i = 0; i < bh_count; i++) {
1066 if (bhs[i] && !buffer_uptodate(bhs[i])) {
1067 err = -EIO;
1068 goto out_brelse;
1069 }
1070 }
1071 return 0;
1072
1073out_brelse:
1074 for (i = 0; i < bh_count; i++) {
1075 brelse(bhs[i]);
1076 bhs[i] = NULL;
1077 }
1078 return err;
1079}
1080
1081int ext4_walk_page_buffers(handle_t *handle,
1082 struct buffer_head *head,
1083 unsigned from,
1084 unsigned to,
1085 int *partial,
1086 int (*fn)(handle_t *handle,
1087 struct buffer_head *bh))
1088{
1089 struct buffer_head *bh;
1090 unsigned block_start, block_end;
1091 unsigned blocksize = head->b_size;
1092 int err, ret = 0;
1093 struct buffer_head *next;
1094
1095 for (bh = head, block_start = 0;
1096 ret == 0 && (bh != head || !block_start);
1097 block_start = block_end, bh = next) {
1098 next = bh->b_this_page;
1099 block_end = block_start + blocksize;
1100 if (block_end <= from || block_start >= to) {
1101 if (partial && !buffer_uptodate(bh))
1102 *partial = 1;
1103 continue;
1104 }
1105 err = (*fn)(handle, bh);
1106 if (!ret)
1107 ret = err;
1108 }
1109 return ret;
1110}
1111
1112/*
1113 * To preserve ordering, it is essential that the hole instantiation and
1114 * the data write be encapsulated in a single transaction. We cannot
1115 * close off a transaction and start a new one between the ext4_get_block()
1116 * and the commit_write(). So doing the jbd2_journal_start at the start of
1117 * prepare_write() is the right place.
1118 *
1119 * Also, this function can nest inside ext4_writepage(). In that case, we
1120 * *know* that ext4_writepage() has generated enough buffer credits to do the
1121 * whole page. So we won't block on the journal in that case, which is good,
1122 * because the caller may be PF_MEMALLOC.
1123 *
1124 * By accident, ext4 can be reentered when a transaction is open via
1125 * quota file writes. If we were to commit the transaction while thus
1126 * reentered, there can be a deadlock - we would be holding a quota
1127 * lock, and the commit would never complete if another thread had a
1128 * transaction open and was blocking on the quota lock - a ranking
1129 * violation.
1130 *
1131 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1132 * will _not_ run commit under these circumstances because handle->h_ref
1133 * is elevated. We'll still have enough credits for the tiny quotafile
1134 * write.
1135 */
1136int do_journal_get_write_access(handle_t *handle,
1137 struct buffer_head *bh)
1138{
1139 int dirty = buffer_dirty(bh);
1140 int ret;
1141
1142 if (!buffer_mapped(bh) || buffer_freed(bh))
1143 return 0;
1144 /*
1145 * __block_write_begin() could have dirtied some buffers. Clean
1146 * the dirty bit as jbd2_journal_get_write_access() could complain
1147 * otherwise about fs integrity issues. Setting of the dirty bit
1148 * by __block_write_begin() isn't a real problem here as we clear
1149 * the bit before releasing a page lock and thus writeback cannot
1150 * ever write the buffer.
1151 */
1152 if (dirty)
1153 clear_buffer_dirty(bh);
1154 BUFFER_TRACE(bh, "get write access");
1155 ret = ext4_journal_get_write_access(handle, bh);
1156 if (!ret && dirty)
1157 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1158 return ret;
1159}
1160
1161#ifdef CONFIG_FS_ENCRYPTION
1162static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1163 get_block_t *get_block)
1164{
1165 unsigned from = pos & (PAGE_SIZE - 1);
1166 unsigned to = from + len;
1167 struct inode *inode = page->mapping->host;
1168 unsigned block_start, block_end;
1169 sector_t block;
1170 int err = 0;
1171 unsigned blocksize = inode->i_sb->s_blocksize;
1172 unsigned bbits;
1173 struct buffer_head *bh, *head, *wait[2];
1174 int nr_wait = 0;
1175 int i;
1176
1177 BUG_ON(!PageLocked(page));
1178 BUG_ON(from > PAGE_SIZE);
1179 BUG_ON(to > PAGE_SIZE);
1180 BUG_ON(from > to);
1181
1182 if (!page_has_buffers(page))
1183 create_empty_buffers(page, blocksize, 0);
1184 head = page_buffers(page);
1185 bbits = ilog2(blocksize);
1186 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1187
1188 for (bh = head, block_start = 0; bh != head || !block_start;
1189 block++, block_start = block_end, bh = bh->b_this_page) {
1190 block_end = block_start + blocksize;
1191 if (block_end <= from || block_start >= to) {
1192 if (PageUptodate(page)) {
1193 if (!buffer_uptodate(bh))
1194 set_buffer_uptodate(bh);
1195 }
1196 continue;
1197 }
1198 if (buffer_new(bh))
1199 clear_buffer_new(bh);
1200 if (!buffer_mapped(bh)) {
1201 WARN_ON(bh->b_size != blocksize);
1202 err = get_block(inode, block, bh, 1);
1203 if (err)
1204 break;
1205 if (buffer_new(bh)) {
1206 if (PageUptodate(page)) {
1207 clear_buffer_new(bh);
1208 set_buffer_uptodate(bh);
1209 mark_buffer_dirty(bh);
1210 continue;
1211 }
1212 if (block_end > to || block_start < from)
1213 zero_user_segments(page, to, block_end,
1214 block_start, from);
1215 continue;
1216 }
1217 }
1218 if (PageUptodate(page)) {
1219 if (!buffer_uptodate(bh))
1220 set_buffer_uptodate(bh);
1221 continue;
1222 }
1223 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1224 !buffer_unwritten(bh) &&
1225 (block_start < from || block_end > to)) {
1226 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1227 wait[nr_wait++] = bh;
1228 }
1229 }
1230 /*
1231 * If we issued read requests, let them complete.
1232 */
1233 for (i = 0; i < nr_wait; i++) {
1234 wait_on_buffer(wait[i]);
1235 if (!buffer_uptodate(wait[i]))
1236 err = -EIO;
1237 }
1238 if (unlikely(err)) {
1239 page_zero_new_buffers(page, from, to);
1240 } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
1241 for (i = 0; i < nr_wait; i++) {
1242 int err2;
1243
1244 err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1245 bh_offset(wait[i]));
1246 if (err2) {
1247 clear_buffer_uptodate(wait[i]);
1248 err = err2;
1249 }
1250 }
1251 }
1252
1253 return err;
1254}
1255#endif
1256
1257static int ext4_write_begin(struct file *file, struct address_space *mapping,
1258 loff_t pos, unsigned len, unsigned flags,
1259 struct page **pagep, void **fsdata)
1260{
1261 struct inode *inode = mapping->host;
1262 int ret, needed_blocks;
1263 handle_t *handle;
1264 int retries = 0;
1265 struct page *page;
1266 pgoff_t index;
1267 unsigned from, to;
1268
1269 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1270 return -EIO;
1271
1272 trace_ext4_write_begin(inode, pos, len, flags);
1273 /*
1274 * Reserve one block more for addition to orphan list in case
1275 * we allocate blocks but write fails for some reason
1276 */
1277 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1278 index = pos >> PAGE_SHIFT;
1279 from = pos & (PAGE_SIZE - 1);
1280 to = from + len;
1281
1282 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1283 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1284 flags, pagep);
1285 if (ret < 0)
1286 return ret;
1287 if (ret == 1)
1288 return 0;
1289 }
1290
1291 /*
1292 * grab_cache_page_write_begin() can take a long time if the
1293 * system is thrashing due to memory pressure, or if the page
1294 * is being written back. So grab it first before we start
1295 * the transaction handle. This also allows us to allocate
1296 * the page (if needed) without using GFP_NOFS.
1297 */
1298retry_grab:
1299 page = grab_cache_page_write_begin(mapping, index, flags);
1300 if (!page)
1301 return -ENOMEM;
1302 unlock_page(page);
1303
1304retry_journal:
1305 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1306 if (IS_ERR(handle)) {
1307 put_page(page);
1308 return PTR_ERR(handle);
1309 }
1310
1311 lock_page(page);
1312 if (page->mapping != mapping) {
1313 /* The page got truncated from under us */
1314 unlock_page(page);
1315 put_page(page);
1316 ext4_journal_stop(handle);
1317 goto retry_grab;
1318 }
1319 /* In case writeback began while the page was unlocked */
1320 wait_for_stable_page(page);
1321
1322#ifdef CONFIG_FS_ENCRYPTION
1323 if (ext4_should_dioread_nolock(inode))
1324 ret = ext4_block_write_begin(page, pos, len,
1325 ext4_get_block_unwritten);
1326 else
1327 ret = ext4_block_write_begin(page, pos, len,
1328 ext4_get_block);
1329#else
1330 if (ext4_should_dioread_nolock(inode))
1331 ret = __block_write_begin(page, pos, len,
1332 ext4_get_block_unwritten);
1333 else
1334 ret = __block_write_begin(page, pos, len, ext4_get_block);
1335#endif
1336 if (!ret && ext4_should_journal_data(inode)) {
1337 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1338 from, to, NULL,
1339 do_journal_get_write_access);
1340 }
1341
1342 if (ret) {
1343 bool extended = (pos + len > inode->i_size) &&
1344 !ext4_verity_in_progress(inode);
1345
1346 unlock_page(page);
1347 /*
1348 * __block_write_begin may have instantiated a few blocks
1349 * outside i_size. Trim these off again. Don't need
1350 * i_size_read because we hold i_mutex.
1351 *
1352 * Add inode to orphan list in case we crash before
1353 * truncate finishes
1354 */
1355 if (extended && ext4_can_truncate(inode))
1356 ext4_orphan_add(handle, inode);
1357
1358 ext4_journal_stop(handle);
1359 if (extended) {
1360 ext4_truncate_failed_write(inode);
1361 /*
1362 * If truncate failed early the inode might
1363 * still be on the orphan list; we need to
1364 * make sure the inode is removed from the
1365 * orphan list in that case.
1366 */
1367 if (inode->i_nlink)
1368 ext4_orphan_del(NULL, inode);
1369 }
1370
1371 if (ret == -ENOSPC &&
1372 ext4_should_retry_alloc(inode->i_sb, &retries))
1373 goto retry_journal;
1374 put_page(page);
1375 return ret;
1376 }
1377 *pagep = page;
1378 return ret;
1379}
1380
1381/* For write_end() in data=journal mode */
1382static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1383{
1384 int ret;
1385 if (!buffer_mapped(bh) || buffer_freed(bh))
1386 return 0;
1387 set_buffer_uptodate(bh);
1388 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1389 clear_buffer_meta(bh);
1390 clear_buffer_prio(bh);
1391 return ret;
1392}
1393
1394/*
1395 * We need to pick up the new inode size which generic_commit_write gave us
1396 * `file' can be NULL - eg, when called from page_symlink().
1397 *
1398 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1399 * buffers are managed internally.
1400 */
1401static int ext4_write_end(struct file *file,
1402 struct address_space *mapping,
1403 loff_t pos, unsigned len, unsigned copied,
1404 struct page *page, void *fsdata)
1405{
1406 handle_t *handle = ext4_journal_current_handle();
1407 struct inode *inode = mapping->host;
1408 loff_t old_size = inode->i_size;
1409 int ret = 0, ret2;
1410 int i_size_changed = 0;
1411 int inline_data = ext4_has_inline_data(inode);
1412 bool verity = ext4_verity_in_progress(inode);
1413
1414 trace_ext4_write_end(inode, pos, len, copied);
1415 if (inline_data) {
1416 ret = ext4_write_inline_data_end(inode, pos, len,
1417 copied, page);
1418 if (ret < 0) {
1419 unlock_page(page);
1420 put_page(page);
1421 goto errout;
1422 }
1423 copied = ret;
1424 } else
1425 copied = block_write_end(file, mapping, pos,
1426 len, copied, page, fsdata);
1427 /*
1428 * it's important to update i_size while still holding page lock:
1429 * page writeout could otherwise come in and zero beyond i_size.
1430 *
1431 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1432 * blocks are being written past EOF, so skip the i_size update.
1433 */
1434 if (!verity)
1435 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1436 unlock_page(page);
1437 put_page(page);
1438
1439 if (old_size < pos && !verity)
1440 pagecache_isize_extended(inode, old_size, pos);
1441 /*
1442 * Don't mark the inode dirty under page lock. First, it unnecessarily
1443 * makes the holding time of page lock longer. Second, it forces lock
1444 * ordering of page lock and transaction start for journaling
1445 * filesystems.
1446 */
1447 if (i_size_changed || inline_data)
1448 ext4_mark_inode_dirty(handle, inode);
1449
1450 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1451 /* if we have allocated more blocks and copied
1452 * less. We will have blocks allocated outside
1453 * inode->i_size. So truncate them
1454 */
1455 ext4_orphan_add(handle, inode);
1456errout:
1457 ret2 = ext4_journal_stop(handle);
1458 if (!ret)
1459 ret = ret2;
1460
1461 if (pos + len > inode->i_size && !verity) {
1462 ext4_truncate_failed_write(inode);
1463 /*
1464 * If truncate failed early the inode might still be
1465 * on the orphan list; we need to make sure the inode
1466 * is removed from the orphan list in that case.
1467 */
1468 if (inode->i_nlink)
1469 ext4_orphan_del(NULL, inode);
1470 }
1471
1472 return ret ? ret : copied;
1473}
1474
1475/*
1476 * This is a private version of page_zero_new_buffers() which doesn't
1477 * set the buffer to be dirty, since in data=journalled mode we need
1478 * to call ext4_handle_dirty_metadata() instead.
1479 */
1480static void ext4_journalled_zero_new_buffers(handle_t *handle,
1481 struct page *page,
1482 unsigned from, unsigned to)
1483{
1484 unsigned int block_start = 0, block_end;
1485 struct buffer_head *head, *bh;
1486
1487 bh = head = page_buffers(page);
1488 do {
1489 block_end = block_start + bh->b_size;
1490 if (buffer_new(bh)) {
1491 if (block_end > from && block_start < to) {
1492 if (!PageUptodate(page)) {
1493 unsigned start, size;
1494
1495 start = max(from, block_start);
1496 size = min(to, block_end) - start;
1497
1498 zero_user(page, start, size);
1499 write_end_fn(handle, bh);
1500 }
1501 clear_buffer_new(bh);
1502 }
1503 }
1504 block_start = block_end;
1505 bh = bh->b_this_page;
1506 } while (bh != head);
1507}
1508
1509static int ext4_journalled_write_end(struct file *file,
1510 struct address_space *mapping,
1511 loff_t pos, unsigned len, unsigned copied,
1512 struct page *page, void *fsdata)
1513{
1514 handle_t *handle = ext4_journal_current_handle();
1515 struct inode *inode = mapping->host;
1516 loff_t old_size = inode->i_size;
1517 int ret = 0, ret2;
1518 int partial = 0;
1519 unsigned from, to;
1520 int size_changed = 0;
1521 int inline_data = ext4_has_inline_data(inode);
1522 bool verity = ext4_verity_in_progress(inode);
1523
1524 trace_ext4_journalled_write_end(inode, pos, len, copied);
1525 from = pos & (PAGE_SIZE - 1);
1526 to = from + len;
1527
1528 BUG_ON(!ext4_handle_valid(handle));
1529
1530 if (inline_data) {
1531 ret = ext4_write_inline_data_end(inode, pos, len,
1532 copied, page);
1533 if (ret < 0) {
1534 unlock_page(page);
1535 put_page(page);
1536 goto errout;
1537 }
1538 copied = ret;
1539 } else if (unlikely(copied < len) && !PageUptodate(page)) {
1540 copied = 0;
1541 ext4_journalled_zero_new_buffers(handle, page, from, to);
1542 } else {
1543 if (unlikely(copied < len))
1544 ext4_journalled_zero_new_buffers(handle, page,
1545 from + copied, to);
1546 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1547 from + copied, &partial,
1548 write_end_fn);
1549 if (!partial)
1550 SetPageUptodate(page);
1551 }
1552 if (!verity)
1553 size_changed = ext4_update_inode_size(inode, pos + copied);
1554 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1555 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1556 unlock_page(page);
1557 put_page(page);
1558
1559 if (old_size < pos && !verity)
1560 pagecache_isize_extended(inode, old_size, pos);
1561
1562 if (size_changed || inline_data) {
1563 ret2 = ext4_mark_inode_dirty(handle, inode);
1564 if (!ret)
1565 ret = ret2;
1566 }
1567
1568 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1569 /* if we have allocated more blocks and copied
1570 * less. We will have blocks allocated outside
1571 * inode->i_size. So truncate them
1572 */
1573 ext4_orphan_add(handle, inode);
1574
1575errout:
1576 ret2 = ext4_journal_stop(handle);
1577 if (!ret)
1578 ret = ret2;
1579 if (pos + len > inode->i_size && !verity) {
1580 ext4_truncate_failed_write(inode);
1581 /*
1582 * If truncate failed early the inode might still be
1583 * on the orphan list; we need to make sure the inode
1584 * is removed from the orphan list in that case.
1585 */
1586 if (inode->i_nlink)
1587 ext4_orphan_del(NULL, inode);
1588 }
1589
1590 return ret ? ret : copied;
1591}
1592
1593/*
1594 * Reserve space for a single cluster
1595 */
1596static int ext4_da_reserve_space(struct inode *inode)
1597{
1598 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1599 struct ext4_inode_info *ei = EXT4_I(inode);
1600 int ret;
1601
1602 /*
1603 * We will charge metadata quota at writeout time; this saves
1604 * us from metadata over-estimation, though we may go over by
1605 * a small amount in the end. Here we just reserve for data.
1606 */
1607 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1608 if (ret)
1609 return ret;
1610
1611 spin_lock(&ei->i_block_reservation_lock);
1612 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1613 spin_unlock(&ei->i_block_reservation_lock);
1614 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1615 return -ENOSPC;
1616 }
1617 ei->i_reserved_data_blocks++;
1618 trace_ext4_da_reserve_space(inode);
1619 spin_unlock(&ei->i_block_reservation_lock);
1620
1621 return 0; /* success */
1622}
1623
1624void ext4_da_release_space(struct inode *inode, int to_free)
1625{
1626 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1627 struct ext4_inode_info *ei = EXT4_I(inode);
1628
1629 if (!to_free)
1630 return; /* Nothing to release, exit */
1631
1632 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1633
1634 trace_ext4_da_release_space(inode, to_free);
1635 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1636 /*
1637 * if there aren't enough reserved blocks, then the
1638 * counter is messed up somewhere. Since this
1639 * function is called from invalidate page, it's
1640 * harmless to return without any action.
1641 */
1642 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1643 "ino %lu, to_free %d with only %d reserved "
1644 "data blocks", inode->i_ino, to_free,
1645 ei->i_reserved_data_blocks);
1646 WARN_ON(1);
1647 to_free = ei->i_reserved_data_blocks;
1648 }
1649 ei->i_reserved_data_blocks -= to_free;
1650
1651 /* update fs dirty data blocks counter */
1652 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1653
1654 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1655
1656 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1657}
1658
1659/*
1660 * Delayed allocation stuff
1661 */
1662
1663struct mpage_da_data {
1664 struct inode *inode;
1665 struct writeback_control *wbc;
1666
1667 pgoff_t first_page; /* The first page to write */
1668 pgoff_t next_page; /* Current page to examine */
1669 pgoff_t last_page; /* Last page to examine */
1670 /*
1671 * Extent to map - this can be after first_page because that can be
1672 * fully mapped. We somewhat abuse m_flags to store whether the extent
1673 * is delalloc or unwritten.
1674 */
1675 struct ext4_map_blocks map;
1676 struct ext4_io_submit io_submit; /* IO submission data */
1677 unsigned int do_map:1;
1678};
1679
1680static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1681 bool invalidate)
1682{
1683 int nr_pages, i;
1684 pgoff_t index, end;
1685 struct pagevec pvec;
1686 struct inode *inode = mpd->inode;
1687 struct address_space *mapping = inode->i_mapping;
1688
1689 /* This is necessary when next_page == 0. */
1690 if (mpd->first_page >= mpd->next_page)
1691 return;
1692
1693 index = mpd->first_page;
1694 end = mpd->next_page - 1;
1695 if (invalidate) {
1696 ext4_lblk_t start, last;
1697 start = index << (PAGE_SHIFT - inode->i_blkbits);
1698 last = end << (PAGE_SHIFT - inode->i_blkbits);
1699 ext4_es_remove_extent(inode, start, last - start + 1);
1700 }
1701
1702 pagevec_init(&pvec);
1703 while (index <= end) {
1704 nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1705 if (nr_pages == 0)
1706 break;
1707 for (i = 0; i < nr_pages; i++) {
1708 struct page *page = pvec.pages[i];
1709
1710 BUG_ON(!PageLocked(page));
1711 BUG_ON(PageWriteback(page));
1712 if (invalidate) {
1713 if (page_mapped(page))
1714 clear_page_dirty_for_io(page);
1715 block_invalidatepage(page, 0, PAGE_SIZE);
1716 ClearPageUptodate(page);
1717 }
1718 unlock_page(page);
1719 }
1720 pagevec_release(&pvec);
1721 }
1722}
1723
1724static void ext4_print_free_blocks(struct inode *inode)
1725{
1726 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1727 struct super_block *sb = inode->i_sb;
1728 struct ext4_inode_info *ei = EXT4_I(inode);
1729
1730 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1731 EXT4_C2B(EXT4_SB(inode->i_sb),
1732 ext4_count_free_clusters(sb)));
1733 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1734 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1735 (long long) EXT4_C2B(EXT4_SB(sb),
1736 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1737 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1738 (long long) EXT4_C2B(EXT4_SB(sb),
1739 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1740 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1741 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1742 ei->i_reserved_data_blocks);
1743 return;
1744}
1745
1746static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1747{
1748 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1749}
1750
1751/*
1752 * ext4_insert_delayed_block - adds a delayed block to the extents status
1753 * tree, incrementing the reserved cluster/block
1754 * count or making a pending reservation
1755 * where needed
1756 *
1757 * @inode - file containing the newly added block
1758 * @lblk - logical block to be added
1759 *
1760 * Returns 0 on success, negative error code on failure.
1761 */
1762static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1763{
1764 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1765 int ret;
1766 bool allocated = false;
1767
1768 /*
1769 * If the cluster containing lblk is shared with a delayed,
1770 * written, or unwritten extent in a bigalloc file system, it's
1771 * already been accounted for and does not need to be reserved.
1772 * A pending reservation must be made for the cluster if it's
1773 * shared with a written or unwritten extent and doesn't already
1774 * have one. Written and unwritten extents can be purged from the
1775 * extents status tree if the system is under memory pressure, so
1776 * it's necessary to examine the extent tree if a search of the
1777 * extents status tree doesn't get a match.
1778 */
1779 if (sbi->s_cluster_ratio == 1) {
1780 ret = ext4_da_reserve_space(inode);
1781 if (ret != 0) /* ENOSPC */
1782 goto errout;
1783 } else { /* bigalloc */
1784 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1785 if (!ext4_es_scan_clu(inode,
1786 &ext4_es_is_mapped, lblk)) {
1787 ret = ext4_clu_mapped(inode,
1788 EXT4_B2C(sbi, lblk));
1789 if (ret < 0)
1790 goto errout;
1791 if (ret == 0) {
1792 ret = ext4_da_reserve_space(inode);
1793 if (ret != 0) /* ENOSPC */
1794 goto errout;
1795 } else {
1796 allocated = true;
1797 }
1798 } else {
1799 allocated = true;
1800 }
1801 }
1802 }
1803
1804 ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1805
1806errout:
1807 return ret;
1808}
1809
1810/*
1811 * This function is grabs code from the very beginning of
1812 * ext4_map_blocks, but assumes that the caller is from delayed write
1813 * time. This function looks up the requested blocks and sets the
1814 * buffer delay bit under the protection of i_data_sem.
1815 */
1816static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1817 struct ext4_map_blocks *map,
1818 struct buffer_head *bh)
1819{
1820 struct extent_status es;
1821 int retval;
1822 sector_t invalid_block = ~((sector_t) 0xffff);
1823#ifdef ES_AGGRESSIVE_TEST
1824 struct ext4_map_blocks orig_map;
1825
1826 memcpy(&orig_map, map, sizeof(*map));
1827#endif
1828
1829 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1830 invalid_block = ~0;
1831
1832 map->m_flags = 0;
1833 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1834 "logical block %lu\n", inode->i_ino, map->m_len,
1835 (unsigned long) map->m_lblk);
1836
1837 /* Lookup extent status tree firstly */
1838 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1839 if (ext4_es_is_hole(&es)) {
1840 retval = 0;
1841 down_read(&EXT4_I(inode)->i_data_sem);
1842 goto add_delayed;
1843 }
1844
1845 /*
1846 * Delayed extent could be allocated by fallocate.
1847 * So we need to check it.
1848 */
1849 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1850 map_bh(bh, inode->i_sb, invalid_block);
1851 set_buffer_new(bh);
1852 set_buffer_delay(bh);
1853 return 0;
1854 }
1855
1856 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1857 retval = es.es_len - (iblock - es.es_lblk);
1858 if (retval > map->m_len)
1859 retval = map->m_len;
1860 map->m_len = retval;
1861 if (ext4_es_is_written(&es))
1862 map->m_flags |= EXT4_MAP_MAPPED;
1863 else if (ext4_es_is_unwritten(&es))
1864 map->m_flags |= EXT4_MAP_UNWRITTEN;
1865 else
1866 BUG();
1867
1868#ifdef ES_AGGRESSIVE_TEST
1869 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1870#endif
1871 return retval;
1872 }
1873
1874 /*
1875 * Try to see if we can get the block without requesting a new
1876 * file system block.
1877 */
1878 down_read(&EXT4_I(inode)->i_data_sem);
1879 if (ext4_has_inline_data(inode))
1880 retval = 0;
1881 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1882 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1883 else
1884 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1885
1886add_delayed:
1887 if (retval == 0) {
1888 int ret;
1889
1890 /*
1891 * XXX: __block_prepare_write() unmaps passed block,
1892 * is it OK?
1893 */
1894
1895 ret = ext4_insert_delayed_block(inode, map->m_lblk);
1896 if (ret != 0) {
1897 retval = ret;
1898 goto out_unlock;
1899 }
1900
1901 map_bh(bh, inode->i_sb, invalid_block);
1902 set_buffer_new(bh);
1903 set_buffer_delay(bh);
1904 } else if (retval > 0) {
1905 int ret;
1906 unsigned int status;
1907
1908 if (unlikely(retval != map->m_len)) {
1909 ext4_warning(inode->i_sb,
1910 "ES len assertion failed for inode "
1911 "%lu: retval %d != map->m_len %d",
1912 inode->i_ino, retval, map->m_len);
1913 WARN_ON(1);
1914 }
1915
1916 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1917 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1918 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1919 map->m_pblk, status);
1920 if (ret != 0)
1921 retval = ret;
1922 }
1923
1924out_unlock:
1925 up_read((&EXT4_I(inode)->i_data_sem));
1926
1927 return retval;
1928}
1929
1930/*
1931 * This is a special get_block_t callback which is used by
1932 * ext4_da_write_begin(). It will either return mapped block or
1933 * reserve space for a single block.
1934 *
1935 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1936 * We also have b_blocknr = -1 and b_bdev initialized properly
1937 *
1938 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1939 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1940 * initialized properly.
1941 */
1942int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1943 struct buffer_head *bh, int create)
1944{
1945 struct ext4_map_blocks map;
1946 int ret = 0;
1947
1948 BUG_ON(create == 0);
1949 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1950
1951 map.m_lblk = iblock;
1952 map.m_len = 1;
1953
1954 /*
1955 * first, we need to know whether the block is allocated already
1956 * preallocated blocks are unmapped but should treated
1957 * the same as allocated blocks.
1958 */
1959 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1960 if (ret <= 0)
1961 return ret;
1962
1963 map_bh(bh, inode->i_sb, map.m_pblk);
1964 ext4_update_bh_state(bh, map.m_flags);
1965
1966 if (buffer_unwritten(bh)) {
1967 /* A delayed write to unwritten bh should be marked
1968 * new and mapped. Mapped ensures that we don't do
1969 * get_block multiple times when we write to the same
1970 * offset and new ensures that we do proper zero out
1971 * for partial write.
1972 */
1973 set_buffer_new(bh);
1974 set_buffer_mapped(bh);
1975 }
1976 return 0;
1977}
1978
1979static int bget_one(handle_t *handle, struct buffer_head *bh)
1980{
1981 get_bh(bh);
1982 return 0;
1983}
1984
1985static int bput_one(handle_t *handle, struct buffer_head *bh)
1986{
1987 put_bh(bh);
1988 return 0;
1989}
1990
1991static int __ext4_journalled_writepage(struct page *page,
1992 unsigned int len)
1993{
1994 struct address_space *mapping = page->mapping;
1995 struct inode *inode = mapping->host;
1996 struct buffer_head *page_bufs = NULL;
1997 handle_t *handle = NULL;
1998 int ret = 0, err = 0;
1999 int inline_data = ext4_has_inline_data(inode);
2000 struct buffer_head *inode_bh = NULL;
2001
2002 ClearPageChecked(page);
2003
2004 if (inline_data) {
2005 BUG_ON(page->index != 0);
2006 BUG_ON(len > ext4_get_max_inline_size(inode));
2007 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
2008 if (inode_bh == NULL)
2009 goto out;
2010 } else {
2011 page_bufs = page_buffers(page);
2012 if (!page_bufs) {
2013 BUG();
2014 goto out;
2015 }
2016 ext4_walk_page_buffers(handle, page_bufs, 0, len,
2017 NULL, bget_one);
2018 }
2019 /*
2020 * We need to release the page lock before we start the
2021 * journal, so grab a reference so the page won't disappear
2022 * out from under us.
2023 */
2024 get_page(page);
2025 unlock_page(page);
2026
2027 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2028 ext4_writepage_trans_blocks(inode));
2029 if (IS_ERR(handle)) {
2030 ret = PTR_ERR(handle);
2031 put_page(page);
2032 goto out_no_pagelock;
2033 }
2034 BUG_ON(!ext4_handle_valid(handle));
2035
2036 lock_page(page);
2037 put_page(page);
2038 if (page->mapping != mapping) {
2039 /* The page got truncated from under us */
2040 ext4_journal_stop(handle);
2041 ret = 0;
2042 goto out;
2043 }
2044
2045 if (inline_data) {
2046 ret = ext4_mark_inode_dirty(handle, inode);
2047 } else {
2048 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2049 do_journal_get_write_access);
2050
2051 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
2052 write_end_fn);
2053 }
2054 if (ret == 0)
2055 ret = err;
2056 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2057 err = ext4_journal_stop(handle);
2058 if (!ret)
2059 ret = err;
2060
2061 if (!ext4_has_inline_data(inode))
2062 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
2063 NULL, bput_one);
2064 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2065out:
2066 unlock_page(page);
2067out_no_pagelock:
2068 brelse(inode_bh);
2069 return ret;
2070}
2071
2072/*
2073 * Note that we don't need to start a transaction unless we're journaling data
2074 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2075 * need to file the inode to the transaction's list in ordered mode because if
2076 * we are writing back data added by write(), the inode is already there and if
2077 * we are writing back data modified via mmap(), no one guarantees in which
2078 * transaction the data will hit the disk. In case we are journaling data, we
2079 * cannot start transaction directly because transaction start ranks above page
2080 * lock so we have to do some magic.
2081 *
2082 * This function can get called via...
2083 * - ext4_writepages after taking page lock (have journal handle)
2084 * - journal_submit_inode_data_buffers (no journal handle)
2085 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2086 * - grab_page_cache when doing write_begin (have journal handle)
2087 *
2088 * We don't do any block allocation in this function. If we have page with
2089 * multiple blocks we need to write those buffer_heads that are mapped. This
2090 * is important for mmaped based write. So if we do with blocksize 1K
2091 * truncate(f, 1024);
2092 * a = mmap(f, 0, 4096);
2093 * a[0] = 'a';
2094 * truncate(f, 4096);
2095 * we have in the page first buffer_head mapped via page_mkwrite call back
2096 * but other buffer_heads would be unmapped but dirty (dirty done via the
2097 * do_wp_page). So writepage should write the first block. If we modify
2098 * the mmap area beyond 1024 we will again get a page_fault and the
2099 * page_mkwrite callback will do the block allocation and mark the
2100 * buffer_heads mapped.
2101 *
2102 * We redirty the page if we have any buffer_heads that is either delay or
2103 * unwritten in the page.
2104 *
2105 * We can get recursively called as show below.
2106 *
2107 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2108 * ext4_writepage()
2109 *
2110 * But since we don't do any block allocation we should not deadlock.
2111 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2112 */
2113static int ext4_writepage(struct page *page,
2114 struct writeback_control *wbc)
2115{
2116 int ret = 0;
2117 loff_t size;
2118 unsigned int len;
2119 struct buffer_head *page_bufs = NULL;
2120 struct inode *inode = page->mapping->host;
2121 struct ext4_io_submit io_submit;
2122 bool keep_towrite = false;
2123
2124 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2125 ext4_invalidatepage(page, 0, PAGE_SIZE);
2126 unlock_page(page);
2127 return -EIO;
2128 }
2129
2130 trace_ext4_writepage(page);
2131 size = i_size_read(inode);
2132 if (page->index == size >> PAGE_SHIFT &&
2133 !ext4_verity_in_progress(inode))
2134 len = size & ~PAGE_MASK;
2135 else
2136 len = PAGE_SIZE;
2137
2138 page_bufs = page_buffers(page);
2139 /*
2140 * We cannot do block allocation or other extent handling in this
2141 * function. If there are buffers needing that, we have to redirty
2142 * the page. But we may reach here when we do a journal commit via
2143 * journal_submit_inode_data_buffers() and in that case we must write
2144 * allocated buffers to achieve data=ordered mode guarantees.
2145 *
2146 * Also, if there is only one buffer per page (the fs block
2147 * size == the page size), if one buffer needs block
2148 * allocation or needs to modify the extent tree to clear the
2149 * unwritten flag, we know that the page can't be written at
2150 * all, so we might as well refuse the write immediately.
2151 * Unfortunately if the block size != page size, we can't as
2152 * easily detect this case using ext4_walk_page_buffers(), but
2153 * for the extremely common case, this is an optimization that
2154 * skips a useless round trip through ext4_bio_write_page().
2155 */
2156 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2157 ext4_bh_delay_or_unwritten)) {
2158 redirty_page_for_writepage(wbc, page);
2159 if ((current->flags & PF_MEMALLOC) ||
2160 (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2161 /*
2162 * For memory cleaning there's no point in writing only
2163 * some buffers. So just bail out. Warn if we came here
2164 * from direct reclaim.
2165 */
2166 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2167 == PF_MEMALLOC);
2168 unlock_page(page);
2169 return 0;
2170 }
2171 keep_towrite = true;
2172 }
2173
2174 if (PageChecked(page) && ext4_should_journal_data(inode))
2175 /*
2176 * It's mmapped pagecache. Add buffers and journal it. There
2177 * doesn't seem much point in redirtying the page here.
2178 */
2179 return __ext4_journalled_writepage(page, len);
2180
2181 ext4_io_submit_init(&io_submit, wbc);
2182 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2183 if (!io_submit.io_end) {
2184 redirty_page_for_writepage(wbc, page);
2185 unlock_page(page);
2186 return -ENOMEM;
2187 }
2188 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2189 ext4_io_submit(&io_submit);
2190 /* Drop io_end reference we got from init */
2191 ext4_put_io_end_defer(io_submit.io_end);
2192 return ret;
2193}
2194
2195static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2196{
2197 int len;
2198 loff_t size;
2199 int err;
2200
2201 BUG_ON(page->index != mpd->first_page);
2202 clear_page_dirty_for_io(page);
2203 /*
2204 * We have to be very careful here! Nothing protects writeback path
2205 * against i_size changes and the page can be writeably mapped into
2206 * page tables. So an application can be growing i_size and writing
2207 * data through mmap while writeback runs. clear_page_dirty_for_io()
2208 * write-protects our page in page tables and the page cannot get
2209 * written to again until we release page lock. So only after
2210 * clear_page_dirty_for_io() we are safe to sample i_size for
2211 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2212 * on the barrier provided by TestClearPageDirty in
2213 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2214 * after page tables are updated.
2215 */
2216 size = i_size_read(mpd->inode);
2217 if (page->index == size >> PAGE_SHIFT &&
2218 !ext4_verity_in_progress(mpd->inode))
2219 len = size & ~PAGE_MASK;
2220 else
2221 len = PAGE_SIZE;
2222 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2223 if (!err)
2224 mpd->wbc->nr_to_write--;
2225 mpd->first_page++;
2226
2227 return err;
2228}
2229
2230#define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2231
2232/*
2233 * mballoc gives us at most this number of blocks...
2234 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2235 * The rest of mballoc seems to handle chunks up to full group size.
2236 */
2237#define MAX_WRITEPAGES_EXTENT_LEN 2048
2238
2239/*
2240 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2241 *
2242 * @mpd - extent of blocks
2243 * @lblk - logical number of the block in the file
2244 * @bh - buffer head we want to add to the extent
2245 *
2246 * The function is used to collect contig. blocks in the same state. If the
2247 * buffer doesn't require mapping for writeback and we haven't started the
2248 * extent of buffers to map yet, the function returns 'true' immediately - the
2249 * caller can write the buffer right away. Otherwise the function returns true
2250 * if the block has been added to the extent, false if the block couldn't be
2251 * added.
2252 */
2253static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2254 struct buffer_head *bh)
2255{
2256 struct ext4_map_blocks *map = &mpd->map;
2257
2258 /* Buffer that doesn't need mapping for writeback? */
2259 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2260 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2261 /* So far no extent to map => we write the buffer right away */
2262 if (map->m_len == 0)
2263 return true;
2264 return false;
2265 }
2266
2267 /* First block in the extent? */
2268 if (map->m_len == 0) {
2269 /* We cannot map unless handle is started... */
2270 if (!mpd->do_map)
2271 return false;
2272 map->m_lblk = lblk;
2273 map->m_len = 1;
2274 map->m_flags = bh->b_state & BH_FLAGS;
2275 return true;
2276 }
2277
2278 /* Don't go larger than mballoc is willing to allocate */
2279 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2280 return false;
2281
2282 /* Can we merge the block to our big extent? */
2283 if (lblk == map->m_lblk + map->m_len &&
2284 (bh->b_state & BH_FLAGS) == map->m_flags) {
2285 map->m_len++;
2286 return true;
2287 }
2288 return false;
2289}
2290
2291/*
2292 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2293 *
2294 * @mpd - extent of blocks for mapping
2295 * @head - the first buffer in the page
2296 * @bh - buffer we should start processing from
2297 * @lblk - logical number of the block in the file corresponding to @bh
2298 *
2299 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2300 * the page for IO if all buffers in this page were mapped and there's no
2301 * accumulated extent of buffers to map or add buffers in the page to the
2302 * extent of buffers to map. The function returns 1 if the caller can continue
2303 * by processing the next page, 0 if it should stop adding buffers to the
2304 * extent to map because we cannot extend it anymore. It can also return value
2305 * < 0 in case of error during IO submission.
2306 */
2307static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2308 struct buffer_head *head,
2309 struct buffer_head *bh,
2310 ext4_lblk_t lblk)
2311{
2312 struct inode *inode = mpd->inode;
2313 int err;
2314 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2315 >> inode->i_blkbits;
2316
2317 if (ext4_verity_in_progress(inode))
2318 blocks = EXT_MAX_BLOCKS;
2319
2320 do {
2321 BUG_ON(buffer_locked(bh));
2322
2323 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2324 /* Found extent to map? */
2325 if (mpd->map.m_len)
2326 return 0;
2327 /* Buffer needs mapping and handle is not started? */
2328 if (!mpd->do_map)
2329 return 0;
2330 /* Everything mapped so far and we hit EOF */
2331 break;
2332 }
2333 } while (lblk++, (bh = bh->b_this_page) != head);
2334 /* So far everything mapped? Submit the page for IO. */
2335 if (mpd->map.m_len == 0) {
2336 err = mpage_submit_page(mpd, head->b_page);
2337 if (err < 0)
2338 return err;
2339 }
2340 return lblk < blocks;
2341}
2342
2343/*
2344 * mpage_map_buffers - update buffers corresponding to changed extent and
2345 * submit fully mapped pages for IO
2346 *
2347 * @mpd - description of extent to map, on return next extent to map
2348 *
2349 * Scan buffers corresponding to changed extent (we expect corresponding pages
2350 * to be already locked) and update buffer state according to new extent state.
2351 * We map delalloc buffers to their physical location, clear unwritten bits,
2352 * and mark buffers as uninit when we perform writes to unwritten extents
2353 * and do extent conversion after IO is finished. If the last page is not fully
2354 * mapped, we update @map to the next extent in the last page that needs
2355 * mapping. Otherwise we submit the page for IO.
2356 */
2357static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2358{
2359 struct pagevec pvec;
2360 int nr_pages, i;
2361 struct inode *inode = mpd->inode;
2362 struct buffer_head *head, *bh;
2363 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2364 pgoff_t start, end;
2365 ext4_lblk_t lblk;
2366 sector_t pblock;
2367 int err;
2368
2369 start = mpd->map.m_lblk >> bpp_bits;
2370 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2371 lblk = start << bpp_bits;
2372 pblock = mpd->map.m_pblk;
2373
2374 pagevec_init(&pvec);
2375 while (start <= end) {
2376 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2377 &start, end);
2378 if (nr_pages == 0)
2379 break;
2380 for (i = 0; i < nr_pages; i++) {
2381 struct page *page = pvec.pages[i];
2382
2383 bh = head = page_buffers(page);
2384 do {
2385 if (lblk < mpd->map.m_lblk)
2386 continue;
2387 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2388 /*
2389 * Buffer after end of mapped extent.
2390 * Find next buffer in the page to map.
2391 */
2392 mpd->map.m_len = 0;
2393 mpd->map.m_flags = 0;
2394 /*
2395 * FIXME: If dioread_nolock supports
2396 * blocksize < pagesize, we need to make
2397 * sure we add size mapped so far to
2398 * io_end->size as the following call
2399 * can submit the page for IO.
2400 */
2401 err = mpage_process_page_bufs(mpd, head,
2402 bh, lblk);
2403 pagevec_release(&pvec);
2404 if (err > 0)
2405 err = 0;
2406 return err;
2407 }
2408 if (buffer_delay(bh)) {
2409 clear_buffer_delay(bh);
2410 bh->b_blocknr = pblock++;
2411 }
2412 clear_buffer_unwritten(bh);
2413 } while (lblk++, (bh = bh->b_this_page) != head);
2414
2415 /*
2416 * FIXME: This is going to break if dioread_nolock
2417 * supports blocksize < pagesize as we will try to
2418 * convert potentially unmapped parts of inode.
2419 */
2420 mpd->io_submit.io_end->size += PAGE_SIZE;
2421 /* Page fully mapped - let IO run! */
2422 err = mpage_submit_page(mpd, page);
2423 if (err < 0) {
2424 pagevec_release(&pvec);
2425 return err;
2426 }
2427 }
2428 pagevec_release(&pvec);
2429 }
2430 /* Extent fully mapped and matches with page boundary. We are done. */
2431 mpd->map.m_len = 0;
2432 mpd->map.m_flags = 0;
2433 return 0;
2434}
2435
2436static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2437{
2438 struct inode *inode = mpd->inode;
2439 struct ext4_map_blocks *map = &mpd->map;
2440 int get_blocks_flags;
2441 int err, dioread_nolock;
2442
2443 trace_ext4_da_write_pages_extent(inode, map);
2444 /*
2445 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2446 * to convert an unwritten extent to be initialized (in the case
2447 * where we have written into one or more preallocated blocks). It is
2448 * possible that we're going to need more metadata blocks than
2449 * previously reserved. However we must not fail because we're in
2450 * writeback and there is nothing we can do about it so it might result
2451 * in data loss. So use reserved blocks to allocate metadata if
2452 * possible.
2453 *
2454 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2455 * the blocks in question are delalloc blocks. This indicates
2456 * that the blocks and quotas has already been checked when
2457 * the data was copied into the page cache.
2458 */
2459 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2460 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2461 EXT4_GET_BLOCKS_IO_SUBMIT;
2462 dioread_nolock = ext4_should_dioread_nolock(inode);
2463 if (dioread_nolock)
2464 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2465 if (map->m_flags & (1 << BH_Delay))
2466 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2467
2468 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2469 if (err < 0)
2470 return err;
2471 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2472 if (!mpd->io_submit.io_end->handle &&
2473 ext4_handle_valid(handle)) {
2474 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2475 handle->h_rsv_handle = NULL;
2476 }
2477 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2478 }
2479
2480 BUG_ON(map->m_len == 0);
2481 return 0;
2482}
2483
2484/*
2485 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2486 * mpd->len and submit pages underlying it for IO
2487 *
2488 * @handle - handle for journal operations
2489 * @mpd - extent to map
2490 * @give_up_on_write - we set this to true iff there is a fatal error and there
2491 * is no hope of writing the data. The caller should discard
2492 * dirty pages to avoid infinite loops.
2493 *
2494 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2495 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2496 * them to initialized or split the described range from larger unwritten
2497 * extent. Note that we need not map all the described range since allocation
2498 * can return less blocks or the range is covered by more unwritten extents. We
2499 * cannot map more because we are limited by reserved transaction credits. On
2500 * the other hand we always make sure that the last touched page is fully
2501 * mapped so that it can be written out (and thus forward progress is
2502 * guaranteed). After mapping we submit all mapped pages for IO.
2503 */
2504static int mpage_map_and_submit_extent(handle_t *handle,
2505 struct mpage_da_data *mpd,
2506 bool *give_up_on_write)
2507{
2508 struct inode *inode = mpd->inode;
2509 struct ext4_map_blocks *map = &mpd->map;
2510 int err;
2511 loff_t disksize;
2512 int progress = 0;
2513
2514 mpd->io_submit.io_end->offset =
2515 ((loff_t)map->m_lblk) << inode->i_blkbits;
2516 do {
2517 err = mpage_map_one_extent(handle, mpd);
2518 if (err < 0) {
2519 struct super_block *sb = inode->i_sb;
2520
2521 if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2522 EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2523 goto invalidate_dirty_pages;
2524 /*
2525 * Let the uper layers retry transient errors.
2526 * In the case of ENOSPC, if ext4_count_free_blocks()
2527 * is non-zero, a commit should free up blocks.
2528 */
2529 if ((err == -ENOMEM) ||
2530 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2531 if (progress)
2532 goto update_disksize;
2533 return err;
2534 }
2535 ext4_msg(sb, KERN_CRIT,
2536 "Delayed block allocation failed for "
2537 "inode %lu at logical offset %llu with"
2538 " max blocks %u with error %d",
2539 inode->i_ino,
2540 (unsigned long long)map->m_lblk,
2541 (unsigned)map->m_len, -err);
2542 ext4_msg(sb, KERN_CRIT,
2543 "This should not happen!! Data will "
2544 "be lost\n");
2545 if (err == -ENOSPC)
2546 ext4_print_free_blocks(inode);
2547 invalidate_dirty_pages:
2548 *give_up_on_write = true;
2549 return err;
2550 }
2551 progress = 1;
2552 /*
2553 * Update buffer state, submit mapped pages, and get us new
2554 * extent to map
2555 */
2556 err = mpage_map_and_submit_buffers(mpd);
2557 if (err < 0)
2558 goto update_disksize;
2559 } while (map->m_len);
2560
2561update_disksize:
2562 /*
2563 * Update on-disk size after IO is submitted. Races with
2564 * truncate are avoided by checking i_size under i_data_sem.
2565 */
2566 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2567 if (disksize > EXT4_I(inode)->i_disksize) {
2568 int err2;
2569 loff_t i_size;
2570
2571 down_write(&EXT4_I(inode)->i_data_sem);
2572 i_size = i_size_read(inode);
2573 if (disksize > i_size)
2574 disksize = i_size;
2575 if (disksize > EXT4_I(inode)->i_disksize)
2576 EXT4_I(inode)->i_disksize = disksize;
2577 up_write(&EXT4_I(inode)->i_data_sem);
2578 err2 = ext4_mark_inode_dirty(handle, inode);
2579 if (err2)
2580 ext4_error(inode->i_sb,
2581 "Failed to mark inode %lu dirty",
2582 inode->i_ino);
2583 if (!err)
2584 err = err2;
2585 }
2586 return err;
2587}
2588
2589/*
2590 * Calculate the total number of credits to reserve for one writepages
2591 * iteration. This is called from ext4_writepages(). We map an extent of
2592 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2593 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2594 * bpp - 1 blocks in bpp different extents.
2595 */
2596static int ext4_da_writepages_trans_blocks(struct inode *inode)
2597{
2598 int bpp = ext4_journal_blocks_per_page(inode);
2599
2600 return ext4_meta_trans_blocks(inode,
2601 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2602}
2603
2604/*
2605 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2606 * and underlying extent to map
2607 *
2608 * @mpd - where to look for pages
2609 *
2610 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2611 * IO immediately. When we find a page which isn't mapped we start accumulating
2612 * extent of buffers underlying these pages that needs mapping (formed by
2613 * either delayed or unwritten buffers). We also lock the pages containing
2614 * these buffers. The extent found is returned in @mpd structure (starting at
2615 * mpd->lblk with length mpd->len blocks).
2616 *
2617 * Note that this function can attach bios to one io_end structure which are
2618 * neither logically nor physically contiguous. Although it may seem as an
2619 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2620 * case as we need to track IO to all buffers underlying a page in one io_end.
2621 */
2622static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2623{
2624 struct address_space *mapping = mpd->inode->i_mapping;
2625 struct pagevec pvec;
2626 unsigned int nr_pages;
2627 long left = mpd->wbc->nr_to_write;
2628 pgoff_t index = mpd->first_page;
2629 pgoff_t end = mpd->last_page;
2630 xa_mark_t tag;
2631 int i, err = 0;
2632 int blkbits = mpd->inode->i_blkbits;
2633 ext4_lblk_t lblk;
2634 struct buffer_head *head;
2635
2636 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2637 tag = PAGECACHE_TAG_TOWRITE;
2638 else
2639 tag = PAGECACHE_TAG_DIRTY;
2640
2641 pagevec_init(&pvec);
2642 mpd->map.m_len = 0;
2643 mpd->next_page = index;
2644 while (index <= end) {
2645 nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2646 tag);
2647 if (nr_pages == 0)
2648 goto out;
2649
2650 for (i = 0; i < nr_pages; i++) {
2651 struct page *page = pvec.pages[i];
2652
2653 /*
2654 * Accumulated enough dirty pages? This doesn't apply
2655 * to WB_SYNC_ALL mode. For integrity sync we have to
2656 * keep going because someone may be concurrently
2657 * dirtying pages, and we might have synced a lot of
2658 * newly appeared dirty pages, but have not synced all
2659 * of the old dirty pages.
2660 */
2661 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2662 goto out;
2663
2664 /* If we can't merge this page, we are done. */
2665 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2666 goto out;
2667
2668 lock_page(page);
2669 /*
2670 * If the page is no longer dirty, or its mapping no
2671 * longer corresponds to inode we are writing (which
2672 * means it has been truncated or invalidated), or the
2673 * page is already under writeback and we are not doing
2674 * a data integrity writeback, skip the page
2675 */
2676 if (!PageDirty(page) ||
2677 (PageWriteback(page) &&
2678 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2679 unlikely(page->mapping != mapping)) {
2680 unlock_page(page);
2681 continue;
2682 }
2683
2684 wait_on_page_writeback(page);
2685 BUG_ON(PageWriteback(page));
2686
2687 if (mpd->map.m_len == 0)
2688 mpd->first_page = page->index;
2689 mpd->next_page = page->index + 1;
2690 /* Add all dirty buffers to mpd */
2691 lblk = ((ext4_lblk_t)page->index) <<
2692 (PAGE_SHIFT - blkbits);
2693 head = page_buffers(page);
2694 err = mpage_process_page_bufs(mpd, head, head, lblk);
2695 if (err <= 0)
2696 goto out;
2697 err = 0;
2698 left--;
2699 }
2700 pagevec_release(&pvec);
2701 cond_resched();
2702 }
2703 return 0;
2704out:
2705 pagevec_release(&pvec);
2706 return err;
2707}
2708
2709static int ext4_writepages(struct address_space *mapping,
2710 struct writeback_control *wbc)
2711{
2712 pgoff_t writeback_index = 0;
2713 long nr_to_write = wbc->nr_to_write;
2714 int range_whole = 0;
2715 int cycled = 1;
2716 handle_t *handle = NULL;
2717 struct mpage_da_data mpd;
2718 struct inode *inode = mapping->host;
2719 int needed_blocks, rsv_blocks = 0, ret = 0;
2720 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2721 bool done;
2722 struct blk_plug plug;
2723 bool give_up_on_write = false;
2724
2725 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2726 return -EIO;
2727
2728 percpu_down_read(&sbi->s_journal_flag_rwsem);
2729 trace_ext4_writepages(inode, wbc);
2730
2731 /*
2732 * No pages to write? This is mainly a kludge to avoid starting
2733 * a transaction for special inodes like journal inode on last iput()
2734 * because that could violate lock ordering on umount
2735 */
2736 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2737 goto out_writepages;
2738
2739 if (ext4_should_journal_data(inode)) {
2740 ret = generic_writepages(mapping, wbc);
2741 goto out_writepages;
2742 }
2743
2744 /*
2745 * If the filesystem has aborted, it is read-only, so return
2746 * right away instead of dumping stack traces later on that
2747 * will obscure the real source of the problem. We test
2748 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2749 * the latter could be true if the filesystem is mounted
2750 * read-only, and in that case, ext4_writepages should
2751 * *never* be called, so if that ever happens, we would want
2752 * the stack trace.
2753 */
2754 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2755 sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2756 ret = -EROFS;
2757 goto out_writepages;
2758 }
2759
2760 /*
2761 * If we have inline data and arrive here, it means that
2762 * we will soon create the block for the 1st page, so
2763 * we'd better clear the inline data here.
2764 */
2765 if (ext4_has_inline_data(inode)) {
2766 /* Just inode will be modified... */
2767 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2768 if (IS_ERR(handle)) {
2769 ret = PTR_ERR(handle);
2770 goto out_writepages;
2771 }
2772 BUG_ON(ext4_test_inode_state(inode,
2773 EXT4_STATE_MAY_INLINE_DATA));
2774 ext4_destroy_inline_data(handle, inode);
2775 ext4_journal_stop(handle);
2776 }
2777
2778 if (ext4_should_dioread_nolock(inode)) {
2779 /*
2780 * We may need to convert up to one extent per block in
2781 * the page and we may dirty the inode.
2782 */
2783 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2784 PAGE_SIZE >> inode->i_blkbits);
2785 }
2786
2787 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2788 range_whole = 1;
2789
2790 if (wbc->range_cyclic) {
2791 writeback_index = mapping->writeback_index;
2792 if (writeback_index)
2793 cycled = 0;
2794 mpd.first_page = writeback_index;
2795 mpd.last_page = -1;
2796 } else {
2797 mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2798 mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2799 }
2800
2801 mpd.inode = inode;
2802 mpd.wbc = wbc;
2803 ext4_io_submit_init(&mpd.io_submit, wbc);
2804retry:
2805 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2806 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2807 done = false;
2808 blk_start_plug(&plug);
2809
2810 /*
2811 * First writeback pages that don't need mapping - we can avoid
2812 * starting a transaction unnecessarily and also avoid being blocked
2813 * in the block layer on device congestion while having transaction
2814 * started.
2815 */
2816 mpd.do_map = 0;
2817 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2818 if (!mpd.io_submit.io_end) {
2819 ret = -ENOMEM;
2820 goto unplug;
2821 }
2822 ret = mpage_prepare_extent_to_map(&mpd);
2823 /* Unlock pages we didn't use */
2824 mpage_release_unused_pages(&mpd, false);
2825 /* Submit prepared bio */
2826 ext4_io_submit(&mpd.io_submit);
2827 ext4_put_io_end_defer(mpd.io_submit.io_end);
2828 mpd.io_submit.io_end = NULL;
2829 if (ret < 0)
2830 goto unplug;
2831
2832 while (!done && mpd.first_page <= mpd.last_page) {
2833 /* For each extent of pages we use new io_end */
2834 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2835 if (!mpd.io_submit.io_end) {
2836 ret = -ENOMEM;
2837 break;
2838 }
2839
2840 /*
2841 * We have two constraints: We find one extent to map and we
2842 * must always write out whole page (makes a difference when
2843 * blocksize < pagesize) so that we don't block on IO when we
2844 * try to write out the rest of the page. Journalled mode is
2845 * not supported by delalloc.
2846 */
2847 BUG_ON(ext4_should_journal_data(inode));
2848 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2849
2850 /* start a new transaction */
2851 handle = ext4_journal_start_with_reserve(inode,
2852 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2853 if (IS_ERR(handle)) {
2854 ret = PTR_ERR(handle);
2855 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2856 "%ld pages, ino %lu; err %d", __func__,
2857 wbc->nr_to_write, inode->i_ino, ret);
2858 /* Release allocated io_end */
2859 ext4_put_io_end(mpd.io_submit.io_end);
2860 mpd.io_submit.io_end = NULL;
2861 break;
2862 }
2863 mpd.do_map = 1;
2864
2865 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2866 ret = mpage_prepare_extent_to_map(&mpd);
2867 if (!ret) {
2868 if (mpd.map.m_len)
2869 ret = mpage_map_and_submit_extent(handle, &mpd,
2870 &give_up_on_write);
2871 else {
2872 /*
2873 * We scanned the whole range (or exhausted
2874 * nr_to_write), submitted what was mapped and
2875 * didn't find anything needing mapping. We are
2876 * done.
2877 */
2878 done = true;
2879 }
2880 }
2881 /*
2882 * Caution: If the handle is synchronous,
2883 * ext4_journal_stop() can wait for transaction commit
2884 * to finish which may depend on writeback of pages to
2885 * complete or on page lock to be released. In that
2886 * case, we have to wait until after after we have
2887 * submitted all the IO, released page locks we hold,
2888 * and dropped io_end reference (for extent conversion
2889 * to be able to complete) before stopping the handle.
2890 */
2891 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2892 ext4_journal_stop(handle);
2893 handle = NULL;
2894 mpd.do_map = 0;
2895 }
2896 /* Unlock pages we didn't use */
2897 mpage_release_unused_pages(&mpd, give_up_on_write);
2898 /* Submit prepared bio */
2899 ext4_io_submit(&mpd.io_submit);
2900
2901 /*
2902 * Drop our io_end reference we got from init. We have
2903 * to be careful and use deferred io_end finishing if
2904 * we are still holding the transaction as we can
2905 * release the last reference to io_end which may end
2906 * up doing unwritten extent conversion.
2907 */
2908 if (handle) {
2909 ext4_put_io_end_defer(mpd.io_submit.io_end);
2910 ext4_journal_stop(handle);
2911 } else
2912 ext4_put_io_end(mpd.io_submit.io_end);
2913 mpd.io_submit.io_end = NULL;
2914
2915 if (ret == -ENOSPC && sbi->s_journal) {
2916 /*
2917 * Commit the transaction which would
2918 * free blocks released in the transaction
2919 * and try again
2920 */
2921 jbd2_journal_force_commit_nested(sbi->s_journal);
2922 ret = 0;
2923 continue;
2924 }
2925 /* Fatal error - ENOMEM, EIO... */
2926 if (ret)
2927 break;
2928 }
2929unplug:
2930 blk_finish_plug(&plug);
2931 if (!ret && !cycled && wbc->nr_to_write > 0) {
2932 cycled = 1;
2933 mpd.last_page = writeback_index - 1;
2934 mpd.first_page = 0;
2935 goto retry;
2936 }
2937
2938 /* Update index */
2939 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2940 /*
2941 * Set the writeback_index so that range_cyclic
2942 * mode will write it back later
2943 */
2944 mapping->writeback_index = mpd.first_page;
2945
2946out_writepages:
2947 trace_ext4_writepages_result(inode, wbc, ret,
2948 nr_to_write - wbc->nr_to_write);
2949 percpu_up_read(&sbi->s_journal_flag_rwsem);
2950 return ret;
2951}
2952
2953static int ext4_dax_writepages(struct address_space *mapping,
2954 struct writeback_control *wbc)
2955{
2956 int ret;
2957 long nr_to_write = wbc->nr_to_write;
2958 struct inode *inode = mapping->host;
2959 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2960
2961 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2962 return -EIO;
2963
2964 percpu_down_read(&sbi->s_journal_flag_rwsem);
2965 trace_ext4_writepages(inode, wbc);
2966
2967 ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
2968 trace_ext4_writepages_result(inode, wbc, ret,
2969 nr_to_write - wbc->nr_to_write);
2970 percpu_up_read(&sbi->s_journal_flag_rwsem);
2971 return ret;
2972}
2973
2974static int ext4_nonda_switch(struct super_block *sb)
2975{
2976 s64 free_clusters, dirty_clusters;
2977 struct ext4_sb_info *sbi = EXT4_SB(sb);
2978
2979 /*
2980 * switch to non delalloc mode if we are running low
2981 * on free block. The free block accounting via percpu
2982 * counters can get slightly wrong with percpu_counter_batch getting
2983 * accumulated on each CPU without updating global counters
2984 * Delalloc need an accurate free block accounting. So switch
2985 * to non delalloc when we are near to error range.
2986 */
2987 free_clusters =
2988 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2989 dirty_clusters =
2990 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2991 /*
2992 * Start pushing delalloc when 1/2 of free blocks are dirty.
2993 */
2994 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2995 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2996
2997 if (2 * free_clusters < 3 * dirty_clusters ||
2998 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2999 /*
3000 * free block count is less than 150% of dirty blocks
3001 * or free blocks is less than watermark
3002 */
3003 return 1;
3004 }
3005 return 0;
3006}
3007
3008/* We always reserve for an inode update; the superblock could be there too */
3009static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
3010{
3011 if (likely(ext4_has_feature_large_file(inode->i_sb)))
3012 return 1;
3013
3014 if (pos + len <= 0x7fffffffULL)
3015 return 1;
3016
3017 /* We might need to update the superblock to set LARGE_FILE */
3018 return 2;
3019}
3020
3021static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
3022 loff_t pos, unsigned len, unsigned flags,
3023 struct page **pagep, void **fsdata)
3024{
3025 int ret, retries = 0;
3026 struct page *page;
3027 pgoff_t index;
3028 struct inode *inode = mapping->host;
3029 handle_t *handle;
3030
3031 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
3032 return -EIO;
3033
3034 index = pos >> PAGE_SHIFT;
3035
3036 if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
3037 ext4_verity_in_progress(inode)) {
3038 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
3039 return ext4_write_begin(file, mapping, pos,
3040 len, flags, pagep, fsdata);
3041 }
3042 *fsdata = (void *)0;
3043 trace_ext4_da_write_begin(inode, pos, len, flags);
3044
3045 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
3046 ret = ext4_da_write_inline_data_begin(mapping, inode,
3047 pos, len, flags,
3048 pagep, fsdata);
3049 if (ret < 0)
3050 return ret;
3051 if (ret == 1)
3052 return 0;
3053 }
3054
3055 /*
3056 * grab_cache_page_write_begin() can take a long time if the
3057 * system is thrashing due to memory pressure, or if the page
3058 * is being written back. So grab it first before we start
3059 * the transaction handle. This also allows us to allocate
3060 * the page (if needed) without using GFP_NOFS.
3061 */
3062retry_grab:
3063 page = grab_cache_page_write_begin(mapping, index, flags);
3064 if (!page)
3065 return -ENOMEM;
3066 unlock_page(page);
3067
3068 /*
3069 * With delayed allocation, we don't log the i_disksize update
3070 * if there is delayed block allocation. But we still need
3071 * to journalling the i_disksize update if writes to the end
3072 * of file which has an already mapped buffer.
3073 */
3074retry_journal:
3075 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
3076 ext4_da_write_credits(inode, pos, len));
3077 if (IS_ERR(handle)) {
3078 put_page(page);
3079 return PTR_ERR(handle);
3080 }
3081
3082 lock_page(page);
3083 if (page->mapping != mapping) {
3084 /* The page got truncated from under us */
3085 unlock_page(page);
3086 put_page(page);
3087 ext4_journal_stop(handle);
3088 goto retry_grab;
3089 }
3090 /* In case writeback began while the page was unlocked */
3091 wait_for_stable_page(page);
3092
3093#ifdef CONFIG_FS_ENCRYPTION
3094 ret = ext4_block_write_begin(page, pos, len,
3095 ext4_da_get_block_prep);
3096#else
3097 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3098#endif
3099 if (ret < 0) {
3100 unlock_page(page);
3101 ext4_journal_stop(handle);
3102 /*
3103 * block_write_begin may have instantiated a few blocks
3104 * outside i_size. Trim these off again. Don't need
3105 * i_size_read because we hold i_mutex.
3106 */
3107 if (pos + len > inode->i_size)
3108 ext4_truncate_failed_write(inode);
3109
3110 if (ret == -ENOSPC &&
3111 ext4_should_retry_alloc(inode->i_sb, &retries))
3112 goto retry_journal;
3113
3114 put_page(page);
3115 return ret;
3116 }
3117
3118 *pagep = page;
3119 return ret;
3120}
3121
3122/*
3123 * Check if we should update i_disksize
3124 * when write to the end of file but not require block allocation
3125 */
3126static int ext4_da_should_update_i_disksize(struct page *page,
3127 unsigned long offset)
3128{
3129 struct buffer_head *bh;
3130 struct inode *inode = page->mapping->host;
3131 unsigned int idx;
3132 int i;
3133
3134 bh = page_buffers(page);
3135 idx = offset >> inode->i_blkbits;
3136
3137 for (i = 0; i < idx; i++)
3138 bh = bh->b_this_page;
3139
3140 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3141 return 0;
3142 return 1;
3143}
3144
3145static int ext4_da_write_end(struct file *file,
3146 struct address_space *mapping,
3147 loff_t pos, unsigned len, unsigned copied,
3148 struct page *page, void *fsdata)
3149{
3150 struct inode *inode = mapping->host;
3151 int ret = 0, ret2;
3152 handle_t *handle = ext4_journal_current_handle();
3153 loff_t new_i_size;
3154 unsigned long start, end;
3155 int write_mode = (int)(unsigned long)fsdata;
3156
3157 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3158 return ext4_write_end(file, mapping, pos,
3159 len, copied, page, fsdata);
3160
3161 trace_ext4_da_write_end(inode, pos, len, copied);
3162 start = pos & (PAGE_SIZE - 1);
3163 end = start + copied - 1;
3164
3165 /*
3166 * generic_write_end() will run mark_inode_dirty() if i_size
3167 * changes. So let's piggyback the i_disksize mark_inode_dirty
3168 * into that.
3169 */
3170 new_i_size = pos + copied;
3171 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3172 if (ext4_has_inline_data(inode) ||
3173 ext4_da_should_update_i_disksize(page, end)) {
3174 ext4_update_i_disksize(inode, new_i_size);
3175 /* We need to mark inode dirty even if
3176 * new_i_size is less that inode->i_size
3177 * bu greater than i_disksize.(hint delalloc)
3178 */
3179 ext4_mark_inode_dirty(handle, inode);
3180 }
3181 }
3182
3183 if (write_mode != CONVERT_INLINE_DATA &&
3184 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3185 ext4_has_inline_data(inode))
3186 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3187 page);
3188 else
3189 ret2 = generic_write_end(file, mapping, pos, len, copied,
3190 page, fsdata);
3191
3192 copied = ret2;
3193 if (ret2 < 0)
3194 ret = ret2;
3195 ret2 = ext4_journal_stop(handle);
3196 if (!ret)
3197 ret = ret2;
3198
3199 return ret ? ret : copied;
3200}
3201
3202/*
3203 * Force all delayed allocation blocks to be allocated for a given inode.
3204 */
3205int ext4_alloc_da_blocks(struct inode *inode)
3206{
3207 trace_ext4_alloc_da_blocks(inode);
3208
3209 if (!EXT4_I(inode)->i_reserved_data_blocks)
3210 return 0;
3211
3212 /*
3213 * We do something simple for now. The filemap_flush() will
3214 * also start triggering a write of the data blocks, which is
3215 * not strictly speaking necessary (and for users of
3216 * laptop_mode, not even desirable). However, to do otherwise
3217 * would require replicating code paths in:
3218 *
3219 * ext4_writepages() ->
3220 * write_cache_pages() ---> (via passed in callback function)
3221 * __mpage_da_writepage() -->
3222 * mpage_add_bh_to_extent()
3223 * mpage_da_map_blocks()
3224 *
3225 * The problem is that write_cache_pages(), located in
3226 * mm/page-writeback.c, marks pages clean in preparation for
3227 * doing I/O, which is not desirable if we're not planning on
3228 * doing I/O at all.
3229 *
3230 * We could call write_cache_pages(), and then redirty all of
3231 * the pages by calling redirty_page_for_writepage() but that
3232 * would be ugly in the extreme. So instead we would need to
3233 * replicate parts of the code in the above functions,
3234 * simplifying them because we wouldn't actually intend to
3235 * write out the pages, but rather only collect contiguous
3236 * logical block extents, call the multi-block allocator, and
3237 * then update the buffer heads with the block allocations.
3238 *
3239 * For now, though, we'll cheat by calling filemap_flush(),
3240 * which will map the blocks, and start the I/O, but not
3241 * actually wait for the I/O to complete.
3242 */
3243 return filemap_flush(inode->i_mapping);
3244}
3245
3246/*
3247 * bmap() is special. It gets used by applications such as lilo and by
3248 * the swapper to find the on-disk block of a specific piece of data.
3249 *
3250 * Naturally, this is dangerous if the block concerned is still in the
3251 * journal. If somebody makes a swapfile on an ext4 data-journaling
3252 * filesystem and enables swap, then they may get a nasty shock when the
3253 * data getting swapped to that swapfile suddenly gets overwritten by
3254 * the original zero's written out previously to the journal and
3255 * awaiting writeback in the kernel's buffer cache.
3256 *
3257 * So, if we see any bmap calls here on a modified, data-journaled file,
3258 * take extra steps to flush any blocks which might be in the cache.
3259 */
3260static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3261{
3262 struct inode *inode = mapping->host;
3263 journal_t *journal;
3264 int err;
3265
3266 /*
3267 * We can get here for an inline file via the FIBMAP ioctl
3268 */
3269 if (ext4_has_inline_data(inode))
3270 return 0;
3271
3272 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3273 test_opt(inode->i_sb, DELALLOC)) {
3274 /*
3275 * With delalloc we want to sync the file
3276 * so that we can make sure we allocate
3277 * blocks for file
3278 */
3279 filemap_write_and_wait(mapping);
3280 }
3281
3282 if (EXT4_JOURNAL(inode) &&
3283 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3284 /*
3285 * This is a REALLY heavyweight approach, but the use of
3286 * bmap on dirty files is expected to be extremely rare:
3287 * only if we run lilo or swapon on a freshly made file
3288 * do we expect this to happen.
3289 *
3290 * (bmap requires CAP_SYS_RAWIO so this does not
3291 * represent an unprivileged user DOS attack --- we'd be
3292 * in trouble if mortal users could trigger this path at
3293 * will.)
3294 *
3295 * NB. EXT4_STATE_JDATA is not set on files other than
3296 * regular files. If somebody wants to bmap a directory
3297 * or symlink and gets confused because the buffer
3298 * hasn't yet been flushed to disk, they deserve
3299 * everything they get.
3300 */
3301
3302 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3303 journal = EXT4_JOURNAL(inode);
3304 jbd2_journal_lock_updates(journal);
3305 err = jbd2_journal_flush(journal);
3306 jbd2_journal_unlock_updates(journal);
3307
3308 if (err)
3309 return 0;
3310 }
3311
3312 return generic_block_bmap(mapping, block, ext4_get_block);
3313}
3314
3315static int ext4_readpage(struct file *file, struct page *page)
3316{
3317 int ret = -EAGAIN;
3318 struct inode *inode = page->mapping->host;
3319
3320 trace_ext4_readpage(page);
3321
3322 if (ext4_has_inline_data(inode))
3323 ret = ext4_readpage_inline(inode, page);
3324
3325 if (ret == -EAGAIN)
3326 return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3327 false);
3328
3329 return ret;
3330}
3331
3332static int
3333ext4_readpages(struct file *file, struct address_space *mapping,
3334 struct list_head *pages, unsigned nr_pages)
3335{
3336 struct inode *inode = mapping->host;
3337
3338 /* If the file has inline data, no need to do readpages. */
3339 if (ext4_has_inline_data(inode))
3340 return 0;
3341
3342 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3343}
3344
3345static void ext4_invalidatepage(struct page *page, unsigned int offset,
3346 unsigned int length)
3347{
3348 trace_ext4_invalidatepage(page, offset, length);
3349
3350 /* No journalling happens on data buffers when this function is used */
3351 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3352
3353 block_invalidatepage(page, offset, length);
3354}
3355
3356static int __ext4_journalled_invalidatepage(struct page *page,
3357 unsigned int offset,
3358 unsigned int length)
3359{
3360 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3361
3362 trace_ext4_journalled_invalidatepage(page, offset, length);
3363
3364 /*
3365 * If it's a full truncate we just forget about the pending dirtying
3366 */
3367 if (offset == 0 && length == PAGE_SIZE)
3368 ClearPageChecked(page);
3369
3370 return jbd2_journal_invalidatepage(journal, page, offset, length);
3371}
3372
3373/* Wrapper for aops... */
3374static void ext4_journalled_invalidatepage(struct page *page,
3375 unsigned int offset,
3376 unsigned int length)
3377{
3378 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3379}
3380
3381static int ext4_releasepage(struct page *page, gfp_t wait)
3382{
3383 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3384
3385 trace_ext4_releasepage(page);
3386
3387 /* Page has dirty journalled data -> cannot release */
3388 if (PageChecked(page))
3389 return 0;
3390 if (journal)
3391 return jbd2_journal_try_to_free_buffers(journal, page, wait);
3392 else
3393 return try_to_free_buffers(page);
3394}
3395
3396static bool ext4_inode_datasync_dirty(struct inode *inode)
3397{
3398 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3399
3400 if (journal)
3401 return !jbd2_transaction_committed(journal,
3402 EXT4_I(inode)->i_datasync_tid);
3403 /* Any metadata buffers to write? */
3404 if (!list_empty(&inode->i_mapping->private_list))
3405 return true;
3406 return inode->i_state & I_DIRTY_DATASYNC;
3407}
3408
3409static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3410 unsigned flags, struct iomap *iomap)
3411{
3412 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3413 unsigned int blkbits = inode->i_blkbits;
3414 unsigned long first_block, last_block;
3415 struct ext4_map_blocks map;
3416 bool delalloc = false;
3417 int ret;
3418
3419 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3420 return -EINVAL;
3421 first_block = offset >> blkbits;
3422 last_block = min_t(loff_t, (offset + length - 1) >> blkbits,
3423 EXT4_MAX_LOGICAL_BLOCK);
3424
3425 if (flags & IOMAP_REPORT) {
3426 if (ext4_has_inline_data(inode)) {
3427 ret = ext4_inline_data_iomap(inode, iomap);
3428 if (ret != -EAGAIN) {
3429 if (ret == 0 && offset >= iomap->length)
3430 ret = -ENOENT;
3431 return ret;
3432 }
3433 }
3434 } else {
3435 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3436 return -ERANGE;
3437 }
3438
3439 map.m_lblk = first_block;
3440 map.m_len = last_block - first_block + 1;
3441
3442 if (flags & IOMAP_REPORT) {
3443 ret = ext4_map_blocks(NULL, inode, &map, 0);
3444 if (ret < 0)
3445 return ret;
3446
3447 if (ret == 0) {
3448 ext4_lblk_t end = map.m_lblk + map.m_len - 1;
3449 struct extent_status es;
3450
3451 ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3452 map.m_lblk, end, &es);
3453
3454 if (!es.es_len || es.es_lblk > end) {
3455 /* entire range is a hole */
3456 } else if (es.es_lblk > map.m_lblk) {
3457 /* range starts with a hole */
3458 map.m_len = es.es_lblk - map.m_lblk;
3459 } else {
3460 ext4_lblk_t offs = 0;
3461
3462 if (es.es_lblk < map.m_lblk)
3463 offs = map.m_lblk - es.es_lblk;
3464 map.m_lblk = es.es_lblk + offs;
3465 map.m_len = es.es_len - offs;
3466 delalloc = true;
3467 }
3468 }
3469 } else if (flags & IOMAP_WRITE) {
3470 int dio_credits;
3471 handle_t *handle;
3472 int retries = 0;
3473
3474 /* Trim mapping request to maximum we can map at once for DIO */
3475 if (map.m_len > DIO_MAX_BLOCKS)
3476 map.m_len = DIO_MAX_BLOCKS;
3477 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
3478retry:
3479 /*
3480 * Either we allocate blocks and then we don't get unwritten
3481 * extent so we have reserved enough credits, or the blocks
3482 * are already allocated and unwritten and in that case
3483 * extent conversion fits in the credits as well.
3484 */
3485 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
3486 dio_credits);
3487 if (IS_ERR(handle))
3488 return PTR_ERR(handle);
3489
3490 ret = ext4_map_blocks(handle, inode, &map,
3491 EXT4_GET_BLOCKS_CREATE_ZERO);
3492 if (ret < 0) {
3493 ext4_journal_stop(handle);
3494 if (ret == -ENOSPC &&
3495 ext4_should_retry_alloc(inode->i_sb, &retries))
3496 goto retry;
3497 return ret;
3498 }
3499
3500 /*
3501 * If we added blocks beyond i_size, we need to make sure they
3502 * will get truncated if we crash before updating i_size in
3503 * ext4_iomap_end(). For faults we don't need to do that (and
3504 * even cannot because for orphan list operations inode_lock is
3505 * required) - if we happen to instantiate block beyond i_size,
3506 * it is because we race with truncate which has already added
3507 * the inode to the orphan list.
3508 */
3509 if (!(flags & IOMAP_FAULT) && first_block + map.m_len >
3510 (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) {
3511 int err;
3512
3513 err = ext4_orphan_add(handle, inode);
3514 if (err < 0) {
3515 ext4_journal_stop(handle);
3516 return err;
3517 }
3518 }
3519 ext4_journal_stop(handle);
3520 } else {
3521 ret = ext4_map_blocks(NULL, inode, &map, 0);
3522 if (ret < 0)
3523 return ret;
3524 }
3525
3526 iomap->flags = 0;
3527 if (ext4_inode_datasync_dirty(inode))
3528 iomap->flags |= IOMAP_F_DIRTY;
3529 iomap->bdev = inode->i_sb->s_bdev;
3530 iomap->dax_dev = sbi->s_daxdev;
3531 iomap->offset = (u64)first_block << blkbits;
3532 iomap->length = (u64)map.m_len << blkbits;
3533
3534 if (ret == 0) {
3535 iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE;
3536 iomap->addr = IOMAP_NULL_ADDR;
3537 } else {
3538 if (map.m_flags & EXT4_MAP_MAPPED) {
3539 iomap->type = IOMAP_MAPPED;
3540 } else if (map.m_flags & EXT4_MAP_UNWRITTEN) {
3541 iomap->type = IOMAP_UNWRITTEN;
3542 } else {
3543 WARN_ON_ONCE(1);
3544 return -EIO;
3545 }
3546 iomap->addr = (u64)map.m_pblk << blkbits;
3547 }
3548
3549 if (map.m_flags & EXT4_MAP_NEW)
3550 iomap->flags |= IOMAP_F_NEW;
3551
3552 return 0;
3553}
3554
3555static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3556 ssize_t written, unsigned flags, struct iomap *iomap)
3557{
3558 int ret = 0;
3559 handle_t *handle;
3560 int blkbits = inode->i_blkbits;
3561 bool truncate = false;
3562
3563 if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT))
3564 return 0;
3565
3566 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3567 if (IS_ERR(handle)) {
3568 ret = PTR_ERR(handle);
3569 goto orphan_del;
3570 }
3571 if (ext4_update_inode_size(inode, offset + written))
3572 ext4_mark_inode_dirty(handle, inode);
3573 /*
3574 * We may need to truncate allocated but not written blocks beyond EOF.
3575 */
3576 if (iomap->offset + iomap->length >
3577 ALIGN(inode->i_size, 1 << blkbits)) {
3578 ext4_lblk_t written_blk, end_blk;
3579
3580 written_blk = (offset + written) >> blkbits;
3581 end_blk = (offset + length) >> blkbits;
3582 if (written_blk < end_blk && ext4_can_truncate(inode))
3583 truncate = true;
3584 }
3585 /*
3586 * Remove inode from orphan list if we were extending a inode and
3587 * everything went fine.
3588 */
3589 if (!truncate && inode->i_nlink &&
3590 !list_empty(&EXT4_I(inode)->i_orphan))
3591 ext4_orphan_del(handle, inode);
3592 ext4_journal_stop(handle);
3593 if (truncate) {
3594 ext4_truncate_failed_write(inode);
3595orphan_del:
3596 /*
3597 * If truncate failed early the inode might still be on the
3598 * orphan list; we need to make sure the inode is removed from
3599 * the orphan list in that case.
3600 */
3601 if (inode->i_nlink)
3602 ext4_orphan_del(NULL, inode);
3603 }
3604 return ret;
3605}
3606
3607const struct iomap_ops ext4_iomap_ops = {
3608 .iomap_begin = ext4_iomap_begin,
3609 .iomap_end = ext4_iomap_end,
3610};
3611
3612static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
3613 ssize_t size, void *private)
3614{
3615 ext4_io_end_t *io_end = private;
3616
3617 /* if not async direct IO just return */
3618 if (!io_end)
3619 return 0;
3620
3621 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3622 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3623 io_end, io_end->inode->i_ino, iocb, offset, size);
3624
3625 /*
3626 * Error during AIO DIO. We cannot convert unwritten extents as the
3627 * data was not written. Just clear the unwritten flag and drop io_end.
3628 */
3629 if (size <= 0) {
3630 ext4_clear_io_unwritten_flag(io_end);
3631 size = 0;
3632 }
3633 io_end->offset = offset;
3634 io_end->size = size;
3635 ext4_put_io_end(io_end);
3636
3637 return 0;
3638}
3639
3640/*
3641 * Handling of direct IO writes.
3642 *
3643 * For ext4 extent files, ext4 will do direct-io write even to holes,
3644 * preallocated extents, and those write extend the file, no need to
3645 * fall back to buffered IO.
3646 *
3647 * For holes, we fallocate those blocks, mark them as unwritten
3648 * If those blocks were preallocated, we mark sure they are split, but
3649 * still keep the range to write as unwritten.
3650 *
3651 * The unwritten extents will be converted to written when DIO is completed.
3652 * For async direct IO, since the IO may still pending when return, we
3653 * set up an end_io call back function, which will do the conversion
3654 * when async direct IO completed.
3655 *
3656 * If the O_DIRECT write will extend the file then add this inode to the
3657 * orphan list. So recovery will truncate it back to the original size
3658 * if the machine crashes during the write.
3659 *
3660 */
3661static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter)
3662{
3663 struct file *file = iocb->ki_filp;
3664 struct inode *inode = file->f_mapping->host;
3665 struct ext4_inode_info *ei = EXT4_I(inode);
3666 ssize_t ret;
3667 loff_t offset = iocb->ki_pos;
3668 size_t count = iov_iter_count(iter);
3669 int overwrite = 0;
3670 get_block_t *get_block_func = NULL;
3671 int dio_flags = 0;
3672 loff_t final_size = offset + count;
3673 int orphan = 0;
3674 handle_t *handle;
3675
3676 if (final_size > inode->i_size || final_size > ei->i_disksize) {
3677 /* Credits for sb + inode write */
3678 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3679 if (IS_ERR(handle)) {
3680 ret = PTR_ERR(handle);
3681 goto out;
3682 }
3683 ret = ext4_orphan_add(handle, inode);
3684 if (ret) {
3685 ext4_journal_stop(handle);
3686 goto out;
3687 }
3688 orphan = 1;
3689 ext4_update_i_disksize(inode, inode->i_size);
3690 ext4_journal_stop(handle);
3691 }
3692
3693 BUG_ON(iocb->private == NULL);
3694
3695 /*
3696 * Make all waiters for direct IO properly wait also for extent
3697 * conversion. This also disallows race between truncate() and
3698 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3699 */
3700 inode_dio_begin(inode);
3701
3702 /* If we do a overwrite dio, i_mutex locking can be released */
3703 overwrite = *((int *)iocb->private);
3704
3705 if (overwrite)
3706 inode_unlock(inode);
3707
3708 /*
3709 * For extent mapped files we could direct write to holes and fallocate.
3710 *
3711 * Allocated blocks to fill the hole are marked as unwritten to prevent
3712 * parallel buffered read to expose the stale data before DIO complete
3713 * the data IO.
3714 *
3715 * As to previously fallocated extents, ext4 get_block will just simply
3716 * mark the buffer mapped but still keep the extents unwritten.
3717 *
3718 * For non AIO case, we will convert those unwritten extents to written
3719 * after return back from blockdev_direct_IO. That way we save us from
3720 * allocating io_end structure and also the overhead of offloading
3721 * the extent convertion to a workqueue.
3722 *
3723 * For async DIO, the conversion needs to be deferred when the
3724 * IO is completed. The ext4 end_io callback function will be
3725 * called to take care of the conversion work. Here for async
3726 * case, we allocate an io_end structure to hook to the iocb.
3727 */
3728 iocb->private = NULL;
3729 if (overwrite)
3730 get_block_func = ext4_dio_get_block_overwrite;
3731 else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) ||
3732 round_down(offset, i_blocksize(inode)) >= inode->i_size) {
3733 get_block_func = ext4_dio_get_block;
3734 dio_flags = DIO_LOCKING | DIO_SKIP_HOLES;
3735 } else if (is_sync_kiocb(iocb)) {
3736 get_block_func = ext4_dio_get_block_unwritten_sync;
3737 dio_flags = DIO_LOCKING;
3738 } else {
3739 get_block_func = ext4_dio_get_block_unwritten_async;
3740 dio_flags = DIO_LOCKING;
3741 }
3742 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
3743 get_block_func, ext4_end_io_dio, NULL,
3744 dio_flags);
3745
3746 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
3747 EXT4_STATE_DIO_UNWRITTEN)) {
3748 int err;
3749 /*
3750 * for non AIO case, since the IO is already
3751 * completed, we could do the conversion right here
3752 */
3753 err = ext4_convert_unwritten_extents(NULL, inode,
3754 offset, ret);
3755 if (err < 0)
3756 ret = err;
3757 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3758 }
3759
3760 inode_dio_end(inode);
3761 /* take i_mutex locking again if we do a ovewrite dio */
3762 if (overwrite)
3763 inode_lock(inode);
3764
3765 if (ret < 0 && final_size > inode->i_size)
3766 ext4_truncate_failed_write(inode);
3767
3768 /* Handle extending of i_size after direct IO write */
3769 if (orphan) {
3770 int err;
3771
3772 /* Credits for sb + inode write */
3773 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
3774 if (IS_ERR(handle)) {
3775 /*
3776 * We wrote the data but cannot extend
3777 * i_size. Bail out. In async io case, we do
3778 * not return error here because we have
3779 * already submmitted the corresponding
3780 * bio. Returning error here makes the caller
3781 * think that this IO is done and failed
3782 * resulting in race with bio's completion
3783 * handler.
3784 */
3785 if (!ret)
3786 ret = PTR_ERR(handle);
3787 if (inode->i_nlink)
3788 ext4_orphan_del(NULL, inode);
3789
3790 goto out;
3791 }
3792 if (inode->i_nlink)
3793 ext4_orphan_del(handle, inode);
3794 if (ret > 0) {
3795 loff_t end = offset + ret;
3796 if (end > inode->i_size || end > ei->i_disksize) {
3797 ext4_update_i_disksize(inode, end);
3798 if (end > inode->i_size)
3799 i_size_write(inode, end);
3800 /*
3801 * We're going to return a positive `ret'
3802 * here due to non-zero-length I/O, so there's
3803 * no way of reporting error returns from
3804 * ext4_mark_inode_dirty() to userspace. So
3805 * ignore it.
3806 */
3807 ext4_mark_inode_dirty(handle, inode);
3808 }
3809 }
3810 err = ext4_journal_stop(handle);
3811 if (ret == 0)
3812 ret = err;
3813 }
3814out:
3815 return ret;
3816}
3817
3818static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter)
3819{
3820 struct address_space *mapping = iocb->ki_filp->f_mapping;
3821 struct inode *inode = mapping->host;
3822 size_t count = iov_iter_count(iter);
3823 ssize_t ret;
3824
3825 /*
3826 * Shared inode_lock is enough for us - it protects against concurrent
3827 * writes & truncates and since we take care of writing back page cache,
3828 * we are protected against page writeback as well.
3829 */
3830 inode_lock_shared(inode);
3831 ret = filemap_write_and_wait_range(mapping, iocb->ki_pos,
3832 iocb->ki_pos + count - 1);
3833 if (ret)
3834 goto out_unlock;
3835 ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
3836 iter, ext4_dio_get_block, NULL, NULL, 0);
3837out_unlock:
3838 inode_unlock_shared(inode);
3839 return ret;
3840}
3841
3842static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
3843{
3844 struct file *file = iocb->ki_filp;
3845 struct inode *inode = file->f_mapping->host;
3846 size_t count = iov_iter_count(iter);
3847 loff_t offset = iocb->ki_pos;
3848 ssize_t ret;
3849
3850#ifdef CONFIG_FS_ENCRYPTION
3851 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
3852 return 0;
3853#endif
3854 if (fsverity_active(inode))
3855 return 0;
3856
3857 /*
3858 * If we are doing data journalling we don't support O_DIRECT
3859 */
3860 if (ext4_should_journal_data(inode))
3861 return 0;
3862
3863 /* Let buffer I/O handle the inline data case. */
3864 if (ext4_has_inline_data(inode))
3865 return 0;
3866
3867 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
3868 if (iov_iter_rw(iter) == READ)
3869 ret = ext4_direct_IO_read(iocb, iter);
3870 else
3871 ret = ext4_direct_IO_write(iocb, iter);
3872 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
3873 return ret;
3874}
3875
3876/*
3877 * Pages can be marked dirty completely asynchronously from ext4's journalling
3878 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3879 * much here because ->set_page_dirty is called under VFS locks. The page is
3880 * not necessarily locked.
3881 *
3882 * We cannot just dirty the page and leave attached buffers clean, because the
3883 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3884 * or jbddirty because all the journalling code will explode.
3885 *
3886 * So what we do is to mark the page "pending dirty" and next time writepage
3887 * is called, propagate that into the buffers appropriately.
3888 */
3889static int ext4_journalled_set_page_dirty(struct page *page)
3890{
3891 SetPageChecked(page);
3892 return __set_page_dirty_nobuffers(page);
3893}
3894
3895static int ext4_set_page_dirty(struct page *page)
3896{
3897 WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3898 WARN_ON_ONCE(!page_has_buffers(page));
3899 return __set_page_dirty_buffers(page);
3900}
3901
3902static const struct address_space_operations ext4_aops = {
3903 .readpage = ext4_readpage,
3904 .readpages = ext4_readpages,
3905 .writepage = ext4_writepage,
3906 .writepages = ext4_writepages,
3907 .write_begin = ext4_write_begin,
3908 .write_end = ext4_write_end,
3909 .set_page_dirty = ext4_set_page_dirty,
3910 .bmap = ext4_bmap,
3911 .invalidatepage = ext4_invalidatepage,
3912 .releasepage = ext4_releasepage,
3913 .direct_IO = ext4_direct_IO,
3914 .migratepage = buffer_migrate_page,
3915 .is_partially_uptodate = block_is_partially_uptodate,
3916 .error_remove_page = generic_error_remove_page,
3917};
3918
3919static const struct address_space_operations ext4_journalled_aops = {
3920 .readpage = ext4_readpage,
3921 .readpages = ext4_readpages,
3922 .writepage = ext4_writepage,
3923 .writepages = ext4_writepages,
3924 .write_begin = ext4_write_begin,
3925 .write_end = ext4_journalled_write_end,
3926 .set_page_dirty = ext4_journalled_set_page_dirty,
3927 .bmap = ext4_bmap,
3928 .invalidatepage = ext4_journalled_invalidatepage,
3929 .releasepage = ext4_releasepage,
3930 .direct_IO = ext4_direct_IO,
3931 .is_partially_uptodate = block_is_partially_uptodate,
3932 .error_remove_page = generic_error_remove_page,
3933};
3934
3935static const struct address_space_operations ext4_da_aops = {
3936 .readpage = ext4_readpage,
3937 .readpages = ext4_readpages,
3938 .writepage = ext4_writepage,
3939 .writepages = ext4_writepages,
3940 .write_begin = ext4_da_write_begin,
3941 .write_end = ext4_da_write_end,
3942 .set_page_dirty = ext4_set_page_dirty,
3943 .bmap = ext4_bmap,
3944 .invalidatepage = ext4_invalidatepage,
3945 .releasepage = ext4_releasepage,
3946 .direct_IO = ext4_direct_IO,
3947 .migratepage = buffer_migrate_page,
3948 .is_partially_uptodate = block_is_partially_uptodate,
3949 .error_remove_page = generic_error_remove_page,
3950};
3951
3952static const struct address_space_operations ext4_dax_aops = {
3953 .writepages = ext4_dax_writepages,
3954 .direct_IO = noop_direct_IO,
3955 .set_page_dirty = noop_set_page_dirty,
3956 .bmap = ext4_bmap,
3957 .invalidatepage = noop_invalidatepage,
3958};
3959
3960void ext4_set_aops(struct inode *inode)
3961{
3962 switch (ext4_inode_journal_mode(inode)) {
3963 case EXT4_INODE_ORDERED_DATA_MODE:
3964 case EXT4_INODE_WRITEBACK_DATA_MODE:
3965 break;
3966 case EXT4_INODE_JOURNAL_DATA_MODE:
3967 inode->i_mapping->a_ops = &ext4_journalled_aops;
3968 return;
3969 default:
3970 BUG();
3971 }
3972 if (IS_DAX(inode))
3973 inode->i_mapping->a_ops = &ext4_dax_aops;
3974 else if (test_opt(inode->i_sb, DELALLOC))
3975 inode->i_mapping->a_ops = &ext4_da_aops;
3976 else
3977 inode->i_mapping->a_ops = &ext4_aops;
3978}
3979
3980static int __ext4_block_zero_page_range(handle_t *handle,
3981 struct address_space *mapping, loff_t from, loff_t length)
3982{
3983 ext4_fsblk_t index = from >> PAGE_SHIFT;
3984 unsigned offset = from & (PAGE_SIZE-1);
3985 unsigned blocksize, pos;
3986 ext4_lblk_t iblock;
3987 struct inode *inode = mapping->host;
3988 struct buffer_head *bh;
3989 struct page *page;
3990 int err = 0;
3991
3992 page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3993 mapping_gfp_constraint(mapping, ~__GFP_FS));
3994 if (!page)
3995 return -ENOMEM;
3996
3997 blocksize = inode->i_sb->s_blocksize;
3998
3999 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
4000
4001 if (!page_has_buffers(page))
4002 create_empty_buffers(page, blocksize, 0);
4003
4004 /* Find the buffer that contains "offset" */
4005 bh = page_buffers(page);
4006 pos = blocksize;
4007 while (offset >= pos) {
4008 bh = bh->b_this_page;
4009 iblock++;
4010 pos += blocksize;
4011 }
4012 if (buffer_freed(bh)) {
4013 BUFFER_TRACE(bh, "freed: skip");
4014 goto unlock;
4015 }
4016 if (!buffer_mapped(bh)) {
4017 BUFFER_TRACE(bh, "unmapped");
4018 ext4_get_block(inode, iblock, bh, 0);
4019 /* unmapped? It's a hole - nothing to do */
4020 if (!buffer_mapped(bh)) {
4021 BUFFER_TRACE(bh, "still unmapped");
4022 goto unlock;
4023 }
4024 }
4025
4026 /* Ok, it's mapped. Make sure it's up-to-date */
4027 if (PageUptodate(page))
4028 set_buffer_uptodate(bh);
4029
4030 if (!buffer_uptodate(bh)) {
4031 err = -EIO;
4032 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
4033 wait_on_buffer(bh);
4034 /* Uhhuh. Read error. Complain and punt. */
4035 if (!buffer_uptodate(bh))
4036 goto unlock;
4037 if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
4038 /* We expect the key to be set. */
4039 BUG_ON(!fscrypt_has_encryption_key(inode));
4040 WARN_ON_ONCE(fscrypt_decrypt_pagecache_blocks(
4041 page, blocksize, bh_offset(bh)));
4042 }
4043 }
4044 if (ext4_should_journal_data(inode)) {
4045 BUFFER_TRACE(bh, "get write access");
4046 err = ext4_journal_get_write_access(handle, bh);
4047 if (err)
4048 goto unlock;
4049 }
4050 zero_user(page, offset, length);
4051 BUFFER_TRACE(bh, "zeroed end of block");
4052
4053 if (ext4_should_journal_data(inode)) {
4054 err = ext4_handle_dirty_metadata(handle, inode, bh);
4055 } else {
4056 err = 0;
4057 mark_buffer_dirty(bh);
4058 if (ext4_should_order_data(inode))
4059 err = ext4_jbd2_inode_add_write(handle, inode, from,
4060 length);
4061 }
4062
4063unlock:
4064 unlock_page(page);
4065 put_page(page);
4066 return err;
4067}
4068
4069/*
4070 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4071 * starting from file offset 'from'. The range to be zero'd must
4072 * be contained with in one block. If the specified range exceeds
4073 * the end of the block it will be shortened to end of the block
4074 * that cooresponds to 'from'
4075 */
4076static int ext4_block_zero_page_range(handle_t *handle,
4077 struct address_space *mapping, loff_t from, loff_t length)
4078{
4079 struct inode *inode = mapping->host;
4080 unsigned offset = from & (PAGE_SIZE-1);
4081 unsigned blocksize = inode->i_sb->s_blocksize;
4082 unsigned max = blocksize - (offset & (blocksize - 1));
4083
4084 /*
4085 * correct length if it does not fall between
4086 * 'from' and the end of the block
4087 */
4088 if (length > max || length < 0)
4089 length = max;
4090
4091 if (IS_DAX(inode)) {
4092 return iomap_zero_range(inode, from, length, NULL,
4093 &ext4_iomap_ops);
4094 }
4095 return __ext4_block_zero_page_range(handle, mapping, from, length);
4096}
4097
4098/*
4099 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4100 * up to the end of the block which corresponds to `from'.
4101 * This required during truncate. We need to physically zero the tail end
4102 * of that block so it doesn't yield old data if the file is later grown.
4103 */
4104static int ext4_block_truncate_page(handle_t *handle,
4105 struct address_space *mapping, loff_t from)
4106{
4107 unsigned offset = from & (PAGE_SIZE-1);
4108 unsigned length;
4109 unsigned blocksize;
4110 struct inode *inode = mapping->host;
4111
4112 /* If we are processing an encrypted inode during orphan list handling */
4113 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
4114 return 0;
4115
4116 blocksize = inode->i_sb->s_blocksize;
4117 length = blocksize - (offset & (blocksize - 1));
4118
4119 return ext4_block_zero_page_range(handle, mapping, from, length);
4120}
4121
4122int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
4123 loff_t lstart, loff_t length)
4124{
4125 struct super_block *sb = inode->i_sb;
4126 struct address_space *mapping = inode->i_mapping;
4127 unsigned partial_start, partial_end;
4128 ext4_fsblk_t start, end;
4129 loff_t byte_end = (lstart + length - 1);
4130 int err = 0;
4131
4132 partial_start = lstart & (sb->s_blocksize - 1);
4133 partial_end = byte_end & (sb->s_blocksize - 1);
4134
4135 start = lstart >> sb->s_blocksize_bits;
4136 end = byte_end >> sb->s_blocksize_bits;
4137
4138 /* Handle partial zero within the single block */
4139 if (start == end &&
4140 (partial_start || (partial_end != sb->s_blocksize - 1))) {
4141 err = ext4_block_zero_page_range(handle, mapping,
4142 lstart, length);
4143 return err;
4144 }
4145 /* Handle partial zero out on the start of the range */
4146 if (partial_start) {
4147 err = ext4_block_zero_page_range(handle, mapping,
4148 lstart, sb->s_blocksize);
4149 if (err)
4150 return err;
4151 }
4152 /* Handle partial zero out on the end of the range */
4153 if (partial_end != sb->s_blocksize - 1)
4154 err = ext4_block_zero_page_range(handle, mapping,
4155 byte_end - partial_end,
4156 partial_end + 1);
4157 return err;
4158}
4159
4160int ext4_can_truncate(struct inode *inode)
4161{
4162 if (S_ISREG(inode->i_mode))
4163 return 1;
4164 if (S_ISDIR(inode->i_mode))
4165 return 1;
4166 if (S_ISLNK(inode->i_mode))
4167 return !ext4_inode_is_fast_symlink(inode);
4168 return 0;
4169}
4170
4171/*
4172 * We have to make sure i_disksize gets properly updated before we truncate
4173 * page cache due to hole punching or zero range. Otherwise i_disksize update
4174 * can get lost as it may have been postponed to submission of writeback but
4175 * that will never happen after we truncate page cache.
4176 */
4177int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
4178 loff_t len)
4179{
4180 handle_t *handle;
4181 loff_t size = i_size_read(inode);
4182
4183 WARN_ON(!inode_is_locked(inode));
4184 if (offset > size || offset + len < size)
4185 return 0;
4186
4187 if (EXT4_I(inode)->i_disksize >= size)
4188 return 0;
4189
4190 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
4191 if (IS_ERR(handle))
4192 return PTR_ERR(handle);
4193 ext4_update_i_disksize(inode, size);
4194 ext4_mark_inode_dirty(handle, inode);
4195 ext4_journal_stop(handle);
4196
4197 return 0;
4198}
4199
4200static void ext4_wait_dax_page(struct ext4_inode_info *ei)
4201{
4202 up_write(&ei->i_mmap_sem);
4203 schedule();
4204 down_write(&ei->i_mmap_sem);
4205}
4206
4207int ext4_break_layouts(struct inode *inode)
4208{
4209 struct ext4_inode_info *ei = EXT4_I(inode);
4210 struct page *page;
4211 int error;
4212
4213 if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
4214 return -EINVAL;
4215
4216 do {
4217 page = dax_layout_busy_page(inode->i_mapping);
4218 if (!page)
4219 return 0;
4220
4221 error = ___wait_var_event(&page->_refcount,
4222 atomic_read(&page->_refcount) == 1,
4223 TASK_INTERRUPTIBLE, 0, 0,
4224 ext4_wait_dax_page(ei));
4225 } while (error == 0);
4226
4227 return error;
4228}
4229
4230/*
4231 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4232 * associated with the given offset and length
4233 *
4234 * @inode: File inode
4235 * @offset: The offset where the hole will begin
4236 * @len: The length of the hole
4237 *
4238 * Returns: 0 on success or negative on failure
4239 */
4240
4241int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
4242{
4243 struct super_block *sb = inode->i_sb;
4244 ext4_lblk_t first_block, stop_block;
4245 struct address_space *mapping = inode->i_mapping;
4246 loff_t first_block_offset, last_block_offset;
4247 handle_t *handle;
4248 unsigned int credits;
4249 int ret = 0;
4250
4251 if (!S_ISREG(inode->i_mode))
4252 return -EOPNOTSUPP;
4253
4254 trace_ext4_punch_hole(inode, offset, length, 0);
4255
4256 ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4257 if (ext4_has_inline_data(inode)) {
4258 down_write(&EXT4_I(inode)->i_mmap_sem);
4259 ret = ext4_convert_inline_data(inode);
4260 up_write(&EXT4_I(inode)->i_mmap_sem);
4261 if (ret)
4262 return ret;
4263 }
4264
4265 /*
4266 * Write out all dirty pages to avoid race conditions
4267 * Then release them.
4268 */
4269 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4270 ret = filemap_write_and_wait_range(mapping, offset,
4271 offset + length - 1);
4272 if (ret)
4273 return ret;
4274 }
4275
4276 inode_lock(inode);
4277
4278 /* No need to punch hole beyond i_size */
4279 if (offset >= inode->i_size)
4280 goto out_mutex;
4281
4282 /*
4283 * If the hole extends beyond i_size, set the hole
4284 * to end after the page that contains i_size
4285 */
4286 if (offset + length > inode->i_size) {
4287 length = inode->i_size +
4288 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4289 offset;
4290 }
4291
4292 if (offset & (sb->s_blocksize - 1) ||
4293 (offset + length) & (sb->s_blocksize - 1)) {
4294 /*
4295 * Attach jinode to inode for jbd2 if we do any zeroing of
4296 * partial block
4297 */
4298 ret = ext4_inode_attach_jinode(inode);
4299 if (ret < 0)
4300 goto out_mutex;
4301
4302 }
4303
4304 /* Wait all existing dio workers, newcomers will block on i_mutex */
4305 inode_dio_wait(inode);
4306
4307 /*
4308 * Prevent page faults from reinstantiating pages we have released from
4309 * page cache.
4310 */
4311 down_write(&EXT4_I(inode)->i_mmap_sem);
4312
4313 ret = ext4_break_layouts(inode);
4314 if (ret)
4315 goto out_dio;
4316
4317 first_block_offset = round_up(offset, sb->s_blocksize);
4318 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4319
4320 /* Now release the pages and zero block aligned part of pages*/
4321 if (last_block_offset > first_block_offset) {
4322 ret = ext4_update_disksize_before_punch(inode, offset, length);
4323 if (ret)
4324 goto out_dio;
4325 truncate_pagecache_range(inode, first_block_offset,
4326 last_block_offset);
4327 }
4328
4329 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4330 credits = ext4_writepage_trans_blocks(inode);
4331 else
4332 credits = ext4_blocks_for_truncate(inode);
4333 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4334 if (IS_ERR(handle)) {
4335 ret = PTR_ERR(handle);
4336 ext4_std_error(sb, ret);
4337 goto out_dio;
4338 }
4339
4340 ret = ext4_zero_partial_blocks(handle, inode, offset,
4341 length);
4342 if (ret)
4343 goto out_stop;
4344
4345 first_block = (offset + sb->s_blocksize - 1) >>
4346 EXT4_BLOCK_SIZE_BITS(sb);
4347 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4348
4349 /* If there are blocks to remove, do it */
4350 if (stop_block > first_block) {
4351
4352 down_write(&EXT4_I(inode)->i_data_sem);
4353 ext4_discard_preallocations(inode);
4354
4355 ret = ext4_es_remove_extent(inode, first_block,
4356 stop_block - first_block);
4357 if (ret) {
4358 up_write(&EXT4_I(inode)->i_data_sem);
4359 goto out_stop;
4360 }
4361
4362 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4363 ret = ext4_ext_remove_space(inode, first_block,
4364 stop_block - 1);
4365 else
4366 ret = ext4_ind_remove_space(handle, inode, first_block,
4367 stop_block);
4368
4369 up_write(&EXT4_I(inode)->i_data_sem);
4370 }
4371 if (IS_SYNC(inode))
4372 ext4_handle_sync(handle);
4373
4374 inode->i_mtime = inode->i_ctime = current_time(inode);
4375 ext4_mark_inode_dirty(handle, inode);
4376 if (ret >= 0)
4377 ext4_update_inode_fsync_trans(handle, inode, 1);
4378out_stop:
4379 ext4_journal_stop(handle);
4380out_dio:
4381 up_write(&EXT4_I(inode)->i_mmap_sem);
4382out_mutex:
4383 inode_unlock(inode);
4384 return ret;
4385}
4386
4387int ext4_inode_attach_jinode(struct inode *inode)
4388{
4389 struct ext4_inode_info *ei = EXT4_I(inode);
4390 struct jbd2_inode *jinode;
4391
4392 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4393 return 0;
4394
4395 jinode = jbd2_alloc_inode(GFP_KERNEL);
4396 spin_lock(&inode->i_lock);
4397 if (!ei->jinode) {
4398 if (!jinode) {
4399 spin_unlock(&inode->i_lock);
4400 return -ENOMEM;
4401 }
4402 ei->jinode = jinode;
4403 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4404 jinode = NULL;
4405 }
4406 spin_unlock(&inode->i_lock);
4407 if (unlikely(jinode != NULL))
4408 jbd2_free_inode(jinode);
4409 return 0;
4410}
4411
4412/*
4413 * ext4_truncate()
4414 *
4415 * We block out ext4_get_block() block instantiations across the entire
4416 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4417 * simultaneously on behalf of the same inode.
4418 *
4419 * As we work through the truncate and commit bits of it to the journal there
4420 * is one core, guiding principle: the file's tree must always be consistent on
4421 * disk. We must be able to restart the truncate after a crash.
4422 *
4423 * The file's tree may be transiently inconsistent in memory (although it
4424 * probably isn't), but whenever we close off and commit a journal transaction,
4425 * the contents of (the filesystem + the journal) must be consistent and
4426 * restartable. It's pretty simple, really: bottom up, right to left (although
4427 * left-to-right works OK too).
4428 *
4429 * Note that at recovery time, journal replay occurs *before* the restart of
4430 * truncate against the orphan inode list.
4431 *
4432 * The committed inode has the new, desired i_size (which is the same as
4433 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4434 * that this inode's truncate did not complete and it will again call
4435 * ext4_truncate() to have another go. So there will be instantiated blocks
4436 * to the right of the truncation point in a crashed ext4 filesystem. But
4437 * that's fine - as long as they are linked from the inode, the post-crash
4438 * ext4_truncate() run will find them and release them.
4439 */
4440int ext4_truncate(struct inode *inode)
4441{
4442 struct ext4_inode_info *ei = EXT4_I(inode);
4443 unsigned int credits;
4444 int err = 0;
4445 handle_t *handle;
4446 struct address_space *mapping = inode->i_mapping;
4447
4448 /*
4449 * There is a possibility that we're either freeing the inode
4450 * or it's a completely new inode. In those cases we might not
4451 * have i_mutex locked because it's not necessary.
4452 */
4453 if (!(inode->i_state & (I_NEW|I_FREEING)))
4454 WARN_ON(!inode_is_locked(inode));
4455 trace_ext4_truncate_enter(inode);
4456
4457 if (!ext4_can_truncate(inode))
4458 return 0;
4459
4460 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4461
4462 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4463 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4464
4465 if (ext4_has_inline_data(inode)) {
4466 int has_inline = 1;
4467
4468 err = ext4_inline_data_truncate(inode, &has_inline);
4469 if (err)
4470 return err;
4471 if (has_inline)
4472 return 0;
4473 }
4474
4475 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4476 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4477 if (ext4_inode_attach_jinode(inode) < 0)
4478 return 0;
4479 }
4480
4481 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4482 credits = ext4_writepage_trans_blocks(inode);
4483 else
4484 credits = ext4_blocks_for_truncate(inode);
4485
4486 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4487 if (IS_ERR(handle))
4488 return PTR_ERR(handle);
4489
4490 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4491 ext4_block_truncate_page(handle, mapping, inode->i_size);
4492
4493 /*
4494 * We add the inode to the orphan list, so that if this
4495 * truncate spans multiple transactions, and we crash, we will
4496 * resume the truncate when the filesystem recovers. It also
4497 * marks the inode dirty, to catch the new size.
4498 *
4499 * Implication: the file must always be in a sane, consistent
4500 * truncatable state while each transaction commits.
4501 */
4502 err = ext4_orphan_add(handle, inode);
4503 if (err)
4504 goto out_stop;
4505
4506 down_write(&EXT4_I(inode)->i_data_sem);
4507
4508 ext4_discard_preallocations(inode);
4509
4510 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4511 err = ext4_ext_truncate(handle, inode);
4512 else
4513 ext4_ind_truncate(handle, inode);
4514
4515 up_write(&ei->i_data_sem);
4516 if (err)
4517 goto out_stop;
4518
4519 if (IS_SYNC(inode))
4520 ext4_handle_sync(handle);
4521
4522out_stop:
4523 /*
4524 * If this was a simple ftruncate() and the file will remain alive,
4525 * then we need to clear up the orphan record which we created above.
4526 * However, if this was a real unlink then we were called by
4527 * ext4_evict_inode(), and we allow that function to clean up the
4528 * orphan info for us.
4529 */
4530 if (inode->i_nlink)
4531 ext4_orphan_del(handle, inode);
4532
4533 inode->i_mtime = inode->i_ctime = current_time(inode);
4534 ext4_mark_inode_dirty(handle, inode);
4535 ext4_journal_stop(handle);
4536
4537 trace_ext4_truncate_exit(inode);
4538 return err;
4539}
4540
4541/*
4542 * ext4_get_inode_loc returns with an extra refcount against the inode's
4543 * underlying buffer_head on success. If 'in_mem' is true, we have all
4544 * data in memory that is needed to recreate the on-disk version of this
4545 * inode.
4546 */
4547static int __ext4_get_inode_loc(struct inode *inode,
4548 struct ext4_iloc *iloc, int in_mem)
4549{
4550 struct ext4_group_desc *gdp;
4551 struct buffer_head *bh;
4552 struct super_block *sb = inode->i_sb;
4553 ext4_fsblk_t block;
4554 struct blk_plug plug;
4555 int inodes_per_block, inode_offset;
4556
4557 iloc->bh = NULL;
4558 if (inode->i_ino < EXT4_ROOT_INO ||
4559 inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4560 return -EFSCORRUPTED;
4561
4562 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4563 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4564 if (!gdp)
4565 return -EIO;
4566
4567 /*
4568 * Figure out the offset within the block group inode table
4569 */
4570 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4571 inode_offset = ((inode->i_ino - 1) %
4572 EXT4_INODES_PER_GROUP(sb));
4573 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4574 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4575
4576 bh = sb_getblk(sb, block);
4577 if (unlikely(!bh))
4578 return -ENOMEM;
4579 if (!buffer_uptodate(bh)) {
4580 lock_buffer(bh);
4581
4582 /*
4583 * If the buffer has the write error flag, we have failed
4584 * to write out another inode in the same block. In this
4585 * case, we don't have to read the block because we may
4586 * read the old inode data successfully.
4587 */
4588 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4589 set_buffer_uptodate(bh);
4590
4591 if (buffer_uptodate(bh)) {
4592 /* someone brought it uptodate while we waited */
4593 unlock_buffer(bh);
4594 goto has_buffer;
4595 }
4596
4597 /*
4598 * If we have all information of the inode in memory and this
4599 * is the only valid inode in the block, we need not read the
4600 * block.
4601 */
4602 if (in_mem) {
4603 struct buffer_head *bitmap_bh;
4604 int i, start;
4605
4606 start = inode_offset & ~(inodes_per_block - 1);
4607
4608 /* Is the inode bitmap in cache? */
4609 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4610 if (unlikely(!bitmap_bh))
4611 goto make_io;
4612
4613 /*
4614 * If the inode bitmap isn't in cache then the
4615 * optimisation may end up performing two reads instead
4616 * of one, so skip it.
4617 */
4618 if (!buffer_uptodate(bitmap_bh)) {
4619 brelse(bitmap_bh);
4620 goto make_io;
4621 }
4622 for (i = start; i < start + inodes_per_block; i++) {
4623 if (i == inode_offset)
4624 continue;
4625 if (ext4_test_bit(i, bitmap_bh->b_data))
4626 break;
4627 }
4628 brelse(bitmap_bh);
4629 if (i == start + inodes_per_block) {
4630 /* all other inodes are free, so skip I/O */
4631 memset(bh->b_data, 0, bh->b_size);
4632 set_buffer_uptodate(bh);
4633 unlock_buffer(bh);
4634 goto has_buffer;
4635 }
4636 }
4637
4638make_io:
4639 /*
4640 * If we need to do any I/O, try to pre-readahead extra
4641 * blocks from the inode table.
4642 */
4643 blk_start_plug(&plug);
4644 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4645 ext4_fsblk_t b, end, table;
4646 unsigned num;
4647 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4648
4649 table = ext4_inode_table(sb, gdp);
4650 /* s_inode_readahead_blks is always a power of 2 */
4651 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4652 if (table > b)
4653 b = table;
4654 end = b + ra_blks;
4655 num = EXT4_INODES_PER_GROUP(sb);
4656 if (ext4_has_group_desc_csum(sb))
4657 num -= ext4_itable_unused_count(sb, gdp);
4658 table += num / inodes_per_block;
4659 if (end > table)
4660 end = table;
4661 while (b <= end)
4662 sb_breadahead(sb, b++);
4663 }
4664
4665 /*
4666 * There are other valid inodes in the buffer, this inode
4667 * has in-inode xattrs, or we don't have this inode in memory.
4668 * Read the block from disk.
4669 */
4670 trace_ext4_load_inode(inode);
4671 get_bh(bh);
4672 bh->b_end_io = end_buffer_read_sync;
4673 submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4674 blk_finish_plug(&plug);
4675 wait_on_buffer(bh);
4676 if (!buffer_uptodate(bh)) {
4677 EXT4_ERROR_INODE_BLOCK(inode, block,
4678 "unable to read itable block");
4679 brelse(bh);
4680 return -EIO;
4681 }
4682 }
4683has_buffer:
4684 iloc->bh = bh;
4685 return 0;
4686}
4687
4688int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4689{
4690 /* We have all inode data except xattrs in memory here. */
4691 return __ext4_get_inode_loc(inode, iloc,
4692 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4693}
4694
4695static bool ext4_should_use_dax(struct inode *inode)
4696{
4697 if (!test_opt(inode->i_sb, DAX))
4698 return false;
4699 if (!S_ISREG(inode->i_mode))
4700 return false;
4701 if (ext4_should_journal_data(inode))
4702 return false;
4703 if (ext4_has_inline_data(inode))
4704 return false;
4705 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4706 return false;
4707 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4708 return false;
4709 return true;
4710}
4711
4712void ext4_set_inode_flags(struct inode *inode)
4713{
4714 unsigned int flags = EXT4_I(inode)->i_flags;
4715 unsigned int new_fl = 0;
4716
4717 if (flags & EXT4_SYNC_FL)
4718 new_fl |= S_SYNC;
4719 if (flags & EXT4_APPEND_FL)
4720 new_fl |= S_APPEND;
4721 if (flags & EXT4_IMMUTABLE_FL)
4722 new_fl |= S_IMMUTABLE;
4723 if (flags & EXT4_NOATIME_FL)
4724 new_fl |= S_NOATIME;
4725 if (flags & EXT4_DIRSYNC_FL)
4726 new_fl |= S_DIRSYNC;
4727 if (ext4_should_use_dax(inode))
4728 new_fl |= S_DAX;
4729 if (flags & EXT4_ENCRYPT_FL)
4730 new_fl |= S_ENCRYPTED;
4731 if (flags & EXT4_CASEFOLD_FL)
4732 new_fl |= S_CASEFOLD;
4733 if (flags & EXT4_VERITY_FL)
4734 new_fl |= S_VERITY;
4735 inode_set_flags(inode, new_fl,
4736 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4737 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4738}
4739
4740static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4741 struct ext4_inode_info *ei)
4742{
4743 blkcnt_t i_blocks ;
4744 struct inode *inode = &(ei->vfs_inode);
4745 struct super_block *sb = inode->i_sb;
4746
4747 if (ext4_has_feature_huge_file(sb)) {
4748 /* we are using combined 48 bit field */
4749 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4750 le32_to_cpu(raw_inode->i_blocks_lo);
4751 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4752 /* i_blocks represent file system block size */
4753 return i_blocks << (inode->i_blkbits - 9);
4754 } else {
4755 return i_blocks;
4756 }
4757 } else {
4758 return le32_to_cpu(raw_inode->i_blocks_lo);
4759 }
4760}
4761
4762static inline int ext4_iget_extra_inode(struct inode *inode,
4763 struct ext4_inode *raw_inode,
4764 struct ext4_inode_info *ei)
4765{
4766 __le32 *magic = (void *)raw_inode +
4767 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4768
4769 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4770 EXT4_INODE_SIZE(inode->i_sb) &&
4771 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4772 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4773 return ext4_find_inline_data_nolock(inode);
4774 } else
4775 EXT4_I(inode)->i_inline_off = 0;
4776 return 0;
4777}
4778
4779int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4780{
4781 if (!ext4_has_feature_project(inode->i_sb))
4782 return -EOPNOTSUPP;
4783 *projid = EXT4_I(inode)->i_projid;
4784 return 0;
4785}
4786
4787/*
4788 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4789 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4790 * set.
4791 */
4792static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4793{
4794 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4795 inode_set_iversion_raw(inode, val);
4796 else
4797 inode_set_iversion_queried(inode, val);
4798}
4799static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4800{
4801 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4802 return inode_peek_iversion_raw(inode);
4803 else
4804 return inode_peek_iversion(inode);
4805}
4806
4807struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4808 ext4_iget_flags flags, const char *function,
4809 unsigned int line)
4810{
4811 struct ext4_iloc iloc;
4812 struct ext4_inode *raw_inode;
4813 struct ext4_inode_info *ei;
4814 struct inode *inode;
4815 journal_t *journal = EXT4_SB(sb)->s_journal;
4816 long ret;
4817 loff_t size;
4818 int block;
4819 uid_t i_uid;
4820 gid_t i_gid;
4821 projid_t i_projid;
4822
4823 if ((!(flags & EXT4_IGET_SPECIAL) &&
4824 (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4825 (ino < EXT4_ROOT_INO) ||
4826 (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4827 if (flags & EXT4_IGET_HANDLE)
4828 return ERR_PTR(-ESTALE);
4829 __ext4_error(sb, function, line,
4830 "inode #%lu: comm %s: iget: illegal inode #",
4831 ino, current->comm);
4832 return ERR_PTR(-EFSCORRUPTED);
4833 }
4834
4835 inode = iget_locked(sb, ino);
4836 if (!inode)
4837 return ERR_PTR(-ENOMEM);
4838 if (!(inode->i_state & I_NEW))
4839 return inode;
4840
4841 ei = EXT4_I(inode);
4842 iloc.bh = NULL;
4843
4844 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4845 if (ret < 0)
4846 goto bad_inode;
4847 raw_inode = ext4_raw_inode(&iloc);
4848
4849 if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4850 ext4_error_inode(inode, function, line, 0,
4851 "iget: root inode unallocated");
4852 ret = -EFSCORRUPTED;
4853 goto bad_inode;
4854 }
4855
4856 if ((flags & EXT4_IGET_HANDLE) &&
4857 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4858 ret = -ESTALE;
4859 goto bad_inode;
4860 }
4861
4862 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4863 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4864 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4865 EXT4_INODE_SIZE(inode->i_sb) ||
4866 (ei->i_extra_isize & 3)) {
4867 ext4_error_inode(inode, function, line, 0,
4868 "iget: bad extra_isize %u "
4869 "(inode size %u)",
4870 ei->i_extra_isize,
4871 EXT4_INODE_SIZE(inode->i_sb));
4872 ret = -EFSCORRUPTED;
4873 goto bad_inode;
4874 }
4875 } else
4876 ei->i_extra_isize = 0;
4877
4878 /* Precompute checksum seed for inode metadata */
4879 if (ext4_has_metadata_csum(sb)) {
4880 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4881 __u32 csum;
4882 __le32 inum = cpu_to_le32(inode->i_ino);
4883 __le32 gen = raw_inode->i_generation;
4884 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4885 sizeof(inum));
4886 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4887 sizeof(gen));
4888 }
4889
4890 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4891 ext4_error_inode(inode, function, line, 0,
4892 "iget: checksum invalid");
4893 ret = -EFSBADCRC;
4894 goto bad_inode;
4895 }
4896
4897 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4898 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4899 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4900 if (ext4_has_feature_project(sb) &&
4901 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4902 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4903 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4904 else
4905 i_projid = EXT4_DEF_PROJID;
4906
4907 if (!(test_opt(inode->i_sb, NO_UID32))) {
4908 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4909 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4910 }
4911 i_uid_write(inode, i_uid);
4912 i_gid_write(inode, i_gid);
4913 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4914 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4915
4916 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4917 ei->i_inline_off = 0;
4918 ei->i_dir_start_lookup = 0;
4919 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4920 /* We now have enough fields to check if the inode was active or not.
4921 * This is needed because nfsd might try to access dead inodes
4922 * the test is that same one that e2fsck uses
4923 * NeilBrown 1999oct15
4924 */
4925 if (inode->i_nlink == 0) {
4926 if ((inode->i_mode == 0 ||
4927 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4928 ino != EXT4_BOOT_LOADER_INO) {
4929 /* this inode is deleted */
4930 ret = -ESTALE;
4931 goto bad_inode;
4932 }
4933 /* The only unlinked inodes we let through here have
4934 * valid i_mode and are being read by the orphan
4935 * recovery code: that's fine, we're about to complete
4936 * the process of deleting those.
4937 * OR it is the EXT4_BOOT_LOADER_INO which is
4938 * not initialized on a new filesystem. */
4939 }
4940 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4941 ext4_set_inode_flags(inode);
4942 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4943 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4944 if (ext4_has_feature_64bit(sb))
4945 ei->i_file_acl |=
4946 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4947 inode->i_size = ext4_isize(sb, raw_inode);
4948 if ((size = i_size_read(inode)) < 0) {
4949 ext4_error_inode(inode, function, line, 0,
4950 "iget: bad i_size value: %lld", size);
4951 ret = -EFSCORRUPTED;
4952 goto bad_inode;
4953 }
4954 ei->i_disksize = inode->i_size;
4955#ifdef CONFIG_QUOTA
4956 ei->i_reserved_quota = 0;
4957#endif
4958 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4959 ei->i_block_group = iloc.block_group;
4960 ei->i_last_alloc_group = ~0;
4961 /*
4962 * NOTE! The in-memory inode i_data array is in little-endian order
4963 * even on big-endian machines: we do NOT byteswap the block numbers!
4964 */
4965 for (block = 0; block < EXT4_N_BLOCKS; block++)
4966 ei->i_data[block] = raw_inode->i_block[block];
4967 INIT_LIST_HEAD(&ei->i_orphan);
4968
4969 /*
4970 * Set transaction id's of transactions that have to be committed
4971 * to finish f[data]sync. We set them to currently running transaction
4972 * as we cannot be sure that the inode or some of its metadata isn't
4973 * part of the transaction - the inode could have been reclaimed and
4974 * now it is reread from disk.
4975 */
4976 if (journal) {
4977 transaction_t *transaction;
4978 tid_t tid;
4979
4980 read_lock(&journal->j_state_lock);
4981 if (journal->j_running_transaction)
4982 transaction = journal->j_running_transaction;
4983 else
4984 transaction = journal->j_committing_transaction;
4985 if (transaction)
4986 tid = transaction->t_tid;
4987 else
4988 tid = journal->j_commit_sequence;
4989 read_unlock(&journal->j_state_lock);
4990 ei->i_sync_tid = tid;
4991 ei->i_datasync_tid = tid;
4992 }
4993
4994 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4995 if (ei->i_extra_isize == 0) {
4996 /* The extra space is currently unused. Use it. */
4997 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4998 ei->i_extra_isize = sizeof(struct ext4_inode) -
4999 EXT4_GOOD_OLD_INODE_SIZE;
5000 } else {
5001 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
5002 if (ret)
5003 goto bad_inode;
5004 }
5005 }
5006
5007 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
5008 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
5009 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
5010 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
5011
5012 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5013 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
5014
5015 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
5016 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5017 ivers |=
5018 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
5019 }
5020 ext4_inode_set_iversion_queried(inode, ivers);
5021 }
5022
5023 ret = 0;
5024 if (ei->i_file_acl &&
5025 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
5026 ext4_error_inode(inode, function, line, 0,
5027 "iget: bad extended attribute block %llu",
5028 ei->i_file_acl);
5029 ret = -EFSCORRUPTED;
5030 goto bad_inode;
5031 } else if (!ext4_has_inline_data(inode)) {
5032 /* validate the block references in the inode */
5033 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
5034 (S_ISLNK(inode->i_mode) &&
5035 !ext4_inode_is_fast_symlink(inode))) {
5036 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
5037 ret = ext4_ext_check_inode(inode);
5038 else
5039 ret = ext4_ind_check_inode(inode);
5040 }
5041 }
5042 if (ret)
5043 goto bad_inode;
5044
5045 if (S_ISREG(inode->i_mode)) {
5046 inode->i_op = &ext4_file_inode_operations;
5047 inode->i_fop = &ext4_file_operations;
5048 ext4_set_aops(inode);
5049 } else if (S_ISDIR(inode->i_mode)) {
5050 inode->i_op = &ext4_dir_inode_operations;
5051 inode->i_fop = &ext4_dir_operations;
5052 } else if (S_ISLNK(inode->i_mode)) {
5053 /* VFS does not allow setting these so must be corruption */
5054 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
5055 ext4_error_inode(inode, function, line, 0,
5056 "iget: immutable or append flags "
5057 "not allowed on symlinks");
5058 ret = -EFSCORRUPTED;
5059 goto bad_inode;
5060 }
5061 if (IS_ENCRYPTED(inode)) {
5062 inode->i_op = &ext4_encrypted_symlink_inode_operations;
5063 ext4_set_aops(inode);
5064 } else if (ext4_inode_is_fast_symlink(inode)) {
5065 inode->i_link = (char *)ei->i_data;
5066 inode->i_op = &ext4_fast_symlink_inode_operations;
5067 nd_terminate_link(ei->i_data, inode->i_size,
5068 sizeof(ei->i_data) - 1);
5069 } else {
5070 inode->i_op = &ext4_symlink_inode_operations;
5071 ext4_set_aops(inode);
5072 }
5073 inode_nohighmem(inode);
5074 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
5075 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
5076 inode->i_op = &ext4_special_inode_operations;
5077 if (raw_inode->i_block[0])
5078 init_special_inode(inode, inode->i_mode,
5079 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
5080 else
5081 init_special_inode(inode, inode->i_mode,
5082 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
5083 } else if (ino == EXT4_BOOT_LOADER_INO) {
5084 make_bad_inode(inode);
5085 } else {
5086 ret = -EFSCORRUPTED;
5087 ext4_error_inode(inode, function, line, 0,
5088 "iget: bogus i_mode (%o)", inode->i_mode);
5089 goto bad_inode;
5090 }
5091 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
5092 ext4_error_inode(inode, function, line, 0,
5093 "casefold flag without casefold feature");
5094 brelse(iloc.bh);
5095
5096 unlock_new_inode(inode);
5097 return inode;
5098
5099bad_inode:
5100 brelse(iloc.bh);
5101 iget_failed(inode);
5102 return ERR_PTR(ret);
5103}
5104
5105static int ext4_inode_blocks_set(handle_t *handle,
5106 struct ext4_inode *raw_inode,
5107 struct ext4_inode_info *ei)
5108{
5109 struct inode *inode = &(ei->vfs_inode);
5110 u64 i_blocks = inode->i_blocks;
5111 struct super_block *sb = inode->i_sb;
5112
5113 if (i_blocks <= ~0U) {
5114 /*
5115 * i_blocks can be represented in a 32 bit variable
5116 * as multiple of 512 bytes
5117 */
5118 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5119 raw_inode->i_blocks_high = 0;
5120 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5121 return 0;
5122 }
5123 if (!ext4_has_feature_huge_file(sb))
5124 return -EFBIG;
5125
5126 if (i_blocks <= 0xffffffffffffULL) {
5127 /*
5128 * i_blocks can be represented in a 48 bit variable
5129 * as multiple of 512 bytes
5130 */
5131 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5132 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5133 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5134 } else {
5135 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
5136 /* i_block is stored in file system block size */
5137 i_blocks = i_blocks >> (inode->i_blkbits - 9);
5138 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
5139 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
5140 }
5141 return 0;
5142}
5143
5144struct other_inode {
5145 unsigned long orig_ino;
5146 struct ext4_inode *raw_inode;
5147};
5148
5149static int other_inode_match(struct inode * inode, unsigned long ino,
5150 void *data)
5151{
5152 struct other_inode *oi = (struct other_inode *) data;
5153
5154 if ((inode->i_ino != ino) ||
5155 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5156 I_DIRTY_INODE)) ||
5157 ((inode->i_state & I_DIRTY_TIME) == 0))
5158 return 0;
5159 spin_lock(&inode->i_lock);
5160 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
5161 I_DIRTY_INODE)) == 0) &&
5162 (inode->i_state & I_DIRTY_TIME)) {
5163 struct ext4_inode_info *ei = EXT4_I(inode);
5164
5165 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
5166 spin_unlock(&inode->i_lock);
5167
5168 spin_lock(&ei->i_raw_lock);
5169 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
5170 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
5171 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
5172 ext4_inode_csum_set(inode, oi->raw_inode, ei);
5173 spin_unlock(&ei->i_raw_lock);
5174 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
5175 return -1;
5176 }
5177 spin_unlock(&inode->i_lock);
5178 return -1;
5179}
5180
5181/*
5182 * Opportunistically update the other time fields for other inodes in
5183 * the same inode table block.
5184 */
5185static void ext4_update_other_inodes_time(struct super_block *sb,
5186 unsigned long orig_ino, char *buf)
5187{
5188 struct other_inode oi;
5189 unsigned long ino;
5190 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5191 int inode_size = EXT4_INODE_SIZE(sb);
5192
5193 oi.orig_ino = orig_ino;
5194 /*
5195 * Calculate the first inode in the inode table block. Inode
5196 * numbers are one-based. That is, the first inode in a block
5197 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5198 */
5199 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5200 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5201 if (ino == orig_ino)
5202 continue;
5203 oi.raw_inode = (struct ext4_inode *) buf;
5204 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
5205 }
5206}
5207
5208/*
5209 * Post the struct inode info into an on-disk inode location in the
5210 * buffer-cache. This gobbles the caller's reference to the
5211 * buffer_head in the inode location struct.
5212 *
5213 * The caller must have write access to iloc->bh.
5214 */
5215static int ext4_do_update_inode(handle_t *handle,
5216 struct inode *inode,
5217 struct ext4_iloc *iloc)
5218{
5219 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5220 struct ext4_inode_info *ei = EXT4_I(inode);
5221 struct buffer_head *bh = iloc->bh;
5222 struct super_block *sb = inode->i_sb;
5223 int err = 0, rc, block;
5224 int need_datasync = 0, set_large_file = 0;
5225 uid_t i_uid;
5226 gid_t i_gid;
5227 projid_t i_projid;
5228
5229 spin_lock(&ei->i_raw_lock);
5230
5231 /* For fields not tracked in the in-memory inode,
5232 * initialise them to zero for new inodes. */
5233 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5234 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5235
5236 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5237 i_uid = i_uid_read(inode);
5238 i_gid = i_gid_read(inode);
5239 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5240 if (!(test_opt(inode->i_sb, NO_UID32))) {
5241 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5242 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5243/*
5244 * Fix up interoperability with old kernels. Otherwise, old inodes get
5245 * re-used with the upper 16 bits of the uid/gid intact
5246 */
5247 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5248 raw_inode->i_uid_high = 0;
5249 raw_inode->i_gid_high = 0;
5250 } else {
5251 raw_inode->i_uid_high =
5252 cpu_to_le16(high_16_bits(i_uid));
5253 raw_inode->i_gid_high =
5254 cpu_to_le16(high_16_bits(i_gid));
5255 }
5256 } else {
5257 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5258 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5259 raw_inode->i_uid_high = 0;
5260 raw_inode->i_gid_high = 0;
5261 }
5262 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5263
5264 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5265 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5266 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5267 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5268
5269 err = ext4_inode_blocks_set(handle, raw_inode, ei);
5270 if (err) {
5271 spin_unlock(&ei->i_raw_lock);
5272 goto out_brelse;
5273 }
5274 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5275 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5276 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5277 raw_inode->i_file_acl_high =
5278 cpu_to_le16(ei->i_file_acl >> 32);
5279 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5280 if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
5281 ext4_isize_set(raw_inode, ei->i_disksize);
5282 need_datasync = 1;
5283 }
5284 if (ei->i_disksize > 0x7fffffffULL) {
5285 if (!ext4_has_feature_large_file(sb) ||
5286 EXT4_SB(sb)->s_es->s_rev_level ==
5287 cpu_to_le32(EXT4_GOOD_OLD_REV))
5288 set_large_file = 1;
5289 }
5290 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5291 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5292 if (old_valid_dev(inode->i_rdev)) {
5293 raw_inode->i_block[0] =
5294 cpu_to_le32(old_encode_dev(inode->i_rdev));
5295 raw_inode->i_block[1] = 0;
5296 } else {
5297 raw_inode->i_block[0] = 0;
5298 raw_inode->i_block[1] =
5299 cpu_to_le32(new_encode_dev(inode->i_rdev));
5300 raw_inode->i_block[2] = 0;
5301 }
5302 } else if (!ext4_has_inline_data(inode)) {
5303 for (block = 0; block < EXT4_N_BLOCKS; block++)
5304 raw_inode->i_block[block] = ei->i_data[block];
5305 }
5306
5307 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5308 u64 ivers = ext4_inode_peek_iversion(inode);
5309
5310 raw_inode->i_disk_version = cpu_to_le32(ivers);
5311 if (ei->i_extra_isize) {
5312 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5313 raw_inode->i_version_hi =
5314 cpu_to_le32(ivers >> 32);
5315 raw_inode->i_extra_isize =
5316 cpu_to_le16(ei->i_extra_isize);
5317 }
5318 }
5319
5320 BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5321 i_projid != EXT4_DEF_PROJID);
5322
5323 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5324 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5325 raw_inode->i_projid = cpu_to_le32(i_projid);
5326
5327 ext4_inode_csum_set(inode, raw_inode, ei);
5328 spin_unlock(&ei->i_raw_lock);
5329 if (inode->i_sb->s_flags & SB_LAZYTIME)
5330 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5331 bh->b_data);
5332
5333 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5334 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5335 if (!err)
5336 err = rc;
5337 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5338 if (set_large_file) {
5339 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5340 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5341 if (err)
5342 goto out_brelse;
5343 ext4_set_feature_large_file(sb);
5344 ext4_handle_sync(handle);
5345 err = ext4_handle_dirty_super(handle, sb);
5346 }
5347 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5348out_brelse:
5349 brelse(bh);
5350 ext4_std_error(inode->i_sb, err);
5351 return err;
5352}
5353
5354/*
5355 * ext4_write_inode()
5356 *
5357 * We are called from a few places:
5358 *
5359 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5360 * Here, there will be no transaction running. We wait for any running
5361 * transaction to commit.
5362 *
5363 * - Within flush work (sys_sync(), kupdate and such).
5364 * We wait on commit, if told to.
5365 *
5366 * - Within iput_final() -> write_inode_now()
5367 * We wait on commit, if told to.
5368 *
5369 * In all cases it is actually safe for us to return without doing anything,
5370 * because the inode has been copied into a raw inode buffer in
5371 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5372 * writeback.
5373 *
5374 * Note that we are absolutely dependent upon all inode dirtiers doing the
5375 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5376 * which we are interested.
5377 *
5378 * It would be a bug for them to not do this. The code:
5379 *
5380 * mark_inode_dirty(inode)
5381 * stuff();
5382 * inode->i_size = expr;
5383 *
5384 * is in error because write_inode() could occur while `stuff()' is running,
5385 * and the new i_size will be lost. Plus the inode will no longer be on the
5386 * superblock's dirty inode list.
5387 */
5388int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5389{
5390 int err;
5391
5392 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5393 sb_rdonly(inode->i_sb))
5394 return 0;
5395
5396 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5397 return -EIO;
5398
5399 if (EXT4_SB(inode->i_sb)->s_journal) {
5400 if (ext4_journal_current_handle()) {
5401 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5402 dump_stack();
5403 return -EIO;
5404 }
5405
5406 /*
5407 * No need to force transaction in WB_SYNC_NONE mode. Also
5408 * ext4_sync_fs() will force the commit after everything is
5409 * written.
5410 */
5411 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5412 return 0;
5413
5414 err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5415 EXT4_I(inode)->i_sync_tid);
5416 } else {
5417 struct ext4_iloc iloc;
5418
5419 err = __ext4_get_inode_loc(inode, &iloc, 0);
5420 if (err)
5421 return err;
5422 /*
5423 * sync(2) will flush the whole buffer cache. No need to do
5424 * it here separately for each inode.
5425 */
5426 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5427 sync_dirty_buffer(iloc.bh);
5428 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5429 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5430 "IO error syncing inode");
5431 err = -EIO;
5432 }
5433 brelse(iloc.bh);
5434 }
5435 return err;
5436}
5437
5438/*
5439 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5440 * buffers that are attached to a page stradding i_size and are undergoing
5441 * commit. In that case we have to wait for commit to finish and try again.
5442 */
5443static void ext4_wait_for_tail_page_commit(struct inode *inode)
5444{
5445 struct page *page;
5446 unsigned offset;
5447 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5448 tid_t commit_tid = 0;
5449 int ret;
5450
5451 offset = inode->i_size & (PAGE_SIZE - 1);
5452 /*
5453 * All buffers in the last page remain valid? Then there's nothing to
5454 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5455 * blocksize case
5456 */
5457 if (offset > PAGE_SIZE - i_blocksize(inode))
5458 return;
5459 while (1) {
5460 page = find_lock_page(inode->i_mapping,
5461 inode->i_size >> PAGE_SHIFT);
5462 if (!page)
5463 return;
5464 ret = __ext4_journalled_invalidatepage(page, offset,
5465 PAGE_SIZE - offset);
5466 unlock_page(page);
5467 put_page(page);
5468 if (ret != -EBUSY)
5469 return;
5470 commit_tid = 0;
5471 read_lock(&journal->j_state_lock);
5472 if (journal->j_committing_transaction)
5473 commit_tid = journal->j_committing_transaction->t_tid;
5474 read_unlock(&journal->j_state_lock);
5475 if (commit_tid)
5476 jbd2_log_wait_commit(journal, commit_tid);
5477 }
5478}
5479
5480/*
5481 * ext4_setattr()
5482 *
5483 * Called from notify_change.
5484 *
5485 * We want to trap VFS attempts to truncate the file as soon as
5486 * possible. In particular, we want to make sure that when the VFS
5487 * shrinks i_size, we put the inode on the orphan list and modify
5488 * i_disksize immediately, so that during the subsequent flushing of
5489 * dirty pages and freeing of disk blocks, we can guarantee that any
5490 * commit will leave the blocks being flushed in an unused state on
5491 * disk. (On recovery, the inode will get truncated and the blocks will
5492 * be freed, so we have a strong guarantee that no future commit will
5493 * leave these blocks visible to the user.)
5494 *
5495 * Another thing we have to assure is that if we are in ordered mode
5496 * and inode is still attached to the committing transaction, we must
5497 * we start writeout of all the dirty pages which are being truncated.
5498 * This way we are sure that all the data written in the previous
5499 * transaction are already on disk (truncate waits for pages under
5500 * writeback).
5501 *
5502 * Called with inode->i_mutex down.
5503 */
5504int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5505{
5506 struct inode *inode = d_inode(dentry);
5507 int error, rc = 0;
5508 int orphan = 0;
5509 const unsigned int ia_valid = attr->ia_valid;
5510
5511 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5512 return -EIO;
5513
5514 if (unlikely(IS_IMMUTABLE(inode)))
5515 return -EPERM;
5516
5517 if (unlikely(IS_APPEND(inode) &&
5518 (ia_valid & (ATTR_MODE | ATTR_UID |
5519 ATTR_GID | ATTR_TIMES_SET))))
5520 return -EPERM;
5521
5522 error = setattr_prepare(dentry, attr);
5523 if (error)
5524 return error;
5525
5526 error = fscrypt_prepare_setattr(dentry, attr);
5527 if (error)
5528 return error;
5529
5530 error = fsverity_prepare_setattr(dentry, attr);
5531 if (error)
5532 return error;
5533
5534 if (is_quota_modification(inode, attr)) {
5535 error = dquot_initialize(inode);
5536 if (error)
5537 return error;
5538 }
5539 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5540 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5541 handle_t *handle;
5542
5543 /* (user+group)*(old+new) structure, inode write (sb,
5544 * inode block, ? - but truncate inode update has it) */
5545 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5546 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5547 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5548 if (IS_ERR(handle)) {
5549 error = PTR_ERR(handle);
5550 goto err_out;
5551 }
5552
5553 /* dquot_transfer() calls back ext4_get_inode_usage() which
5554 * counts xattr inode references.
5555 */
5556 down_read(&EXT4_I(inode)->xattr_sem);
5557 error = dquot_transfer(inode, attr);
5558 up_read(&EXT4_I(inode)->xattr_sem);
5559
5560 if (error) {
5561 ext4_journal_stop(handle);
5562 return error;
5563 }
5564 /* Update corresponding info in inode so that everything is in
5565 * one transaction */
5566 if (attr->ia_valid & ATTR_UID)
5567 inode->i_uid = attr->ia_uid;
5568 if (attr->ia_valid & ATTR_GID)
5569 inode->i_gid = attr->ia_gid;
5570 error = ext4_mark_inode_dirty(handle, inode);
5571 ext4_journal_stop(handle);
5572 }
5573
5574 if (attr->ia_valid & ATTR_SIZE) {
5575 handle_t *handle;
5576 loff_t oldsize = inode->i_size;
5577 int shrink = (attr->ia_size < inode->i_size);
5578
5579 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5580 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5581
5582 if (attr->ia_size > sbi->s_bitmap_maxbytes)
5583 return -EFBIG;
5584 }
5585 if (!S_ISREG(inode->i_mode))
5586 return -EINVAL;
5587
5588 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5589 inode_inc_iversion(inode);
5590
5591 if (shrink) {
5592 if (ext4_should_order_data(inode)) {
5593 error = ext4_begin_ordered_truncate(inode,
5594 attr->ia_size);
5595 if (error)
5596 goto err_out;
5597 }
5598 /*
5599 * Blocks are going to be removed from the inode. Wait
5600 * for dio in flight.
5601 */
5602 inode_dio_wait(inode);
5603 }
5604
5605 down_write(&EXT4_I(inode)->i_mmap_sem);
5606
5607 rc = ext4_break_layouts(inode);
5608 if (rc) {
5609 up_write(&EXT4_I(inode)->i_mmap_sem);
5610 return rc;
5611 }
5612
5613 if (attr->ia_size != inode->i_size) {
5614 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5615 if (IS_ERR(handle)) {
5616 error = PTR_ERR(handle);
5617 goto out_mmap_sem;
5618 }
5619 if (ext4_handle_valid(handle) && shrink) {
5620 error = ext4_orphan_add(handle, inode);
5621 orphan = 1;
5622 }
5623 /*
5624 * Update c/mtime on truncate up, ext4_truncate() will
5625 * update c/mtime in shrink case below
5626 */
5627 if (!shrink) {
5628 inode->i_mtime = current_time(inode);
5629 inode->i_ctime = inode->i_mtime;
5630 }
5631 down_write(&EXT4_I(inode)->i_data_sem);
5632 EXT4_I(inode)->i_disksize = attr->ia_size;
5633 rc = ext4_mark_inode_dirty(handle, inode);
5634 if (!error)
5635 error = rc;
5636 /*
5637 * We have to update i_size under i_data_sem together
5638 * with i_disksize to avoid races with writeback code
5639 * running ext4_wb_update_i_disksize().
5640 */
5641 if (!error)
5642 i_size_write(inode, attr->ia_size);
5643 up_write(&EXT4_I(inode)->i_data_sem);
5644 ext4_journal_stop(handle);
5645 if (error)
5646 goto out_mmap_sem;
5647 if (!shrink) {
5648 pagecache_isize_extended(inode, oldsize,
5649 inode->i_size);
5650 } else if (ext4_should_journal_data(inode)) {
5651 ext4_wait_for_tail_page_commit(inode);
5652 }
5653 }
5654
5655 /*
5656 * Truncate pagecache after we've waited for commit
5657 * in data=journal mode to make pages freeable.
5658 */
5659 truncate_pagecache(inode, inode->i_size);
5660 /*
5661 * Call ext4_truncate() even if i_size didn't change to
5662 * truncate possible preallocated blocks.
5663 */
5664 if (attr->ia_size <= oldsize) {
5665 rc = ext4_truncate(inode);
5666 if (rc)
5667 error = rc;
5668 }
5669out_mmap_sem:
5670 up_write(&EXT4_I(inode)->i_mmap_sem);
5671 }
5672
5673 if (!error) {
5674 setattr_copy(inode, attr);
5675 mark_inode_dirty(inode);
5676 }
5677
5678 /*
5679 * If the call to ext4_truncate failed to get a transaction handle at
5680 * all, we need to clean up the in-core orphan list manually.
5681 */
5682 if (orphan && inode->i_nlink)
5683 ext4_orphan_del(NULL, inode);
5684
5685 if (!error && (ia_valid & ATTR_MODE))
5686 rc = posix_acl_chmod(inode, inode->i_mode);
5687
5688err_out:
5689 ext4_std_error(inode->i_sb, error);
5690 if (!error)
5691 error = rc;
5692 return error;
5693}
5694
5695int ext4_getattr(const struct path *path, struct kstat *stat,
5696 u32 request_mask, unsigned int query_flags)
5697{
5698 struct inode *inode = d_inode(path->dentry);
5699 struct ext4_inode *raw_inode;
5700 struct ext4_inode_info *ei = EXT4_I(inode);
5701 unsigned int flags;
5702
5703 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5704 stat->result_mask |= STATX_BTIME;
5705 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5706 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5707 }
5708
5709 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5710 if (flags & EXT4_APPEND_FL)
5711 stat->attributes |= STATX_ATTR_APPEND;
5712 if (flags & EXT4_COMPR_FL)
5713 stat->attributes |= STATX_ATTR_COMPRESSED;
5714 if (flags & EXT4_ENCRYPT_FL)
5715 stat->attributes |= STATX_ATTR_ENCRYPTED;
5716 if (flags & EXT4_IMMUTABLE_FL)
5717 stat->attributes |= STATX_ATTR_IMMUTABLE;
5718 if (flags & EXT4_NODUMP_FL)
5719 stat->attributes |= STATX_ATTR_NODUMP;
5720
5721 stat->attributes_mask |= (STATX_ATTR_APPEND |
5722 STATX_ATTR_COMPRESSED |
5723 STATX_ATTR_ENCRYPTED |
5724 STATX_ATTR_IMMUTABLE |
5725 STATX_ATTR_NODUMP);
5726
5727 generic_fillattr(inode, stat);
5728 return 0;
5729}
5730
5731int ext4_file_getattr(const struct path *path, struct kstat *stat,
5732 u32 request_mask, unsigned int query_flags)
5733{
5734 struct inode *inode = d_inode(path->dentry);
5735 u64 delalloc_blocks;
5736
5737 ext4_getattr(path, stat, request_mask, query_flags);
5738
5739 /*
5740 * If there is inline data in the inode, the inode will normally not
5741 * have data blocks allocated (it may have an external xattr block).
5742 * Report at least one sector for such files, so tools like tar, rsync,
5743 * others don't incorrectly think the file is completely sparse.
5744 */
5745 if (unlikely(ext4_has_inline_data(inode)))
5746 stat->blocks += (stat->size + 511) >> 9;
5747
5748 /*
5749 * We can't update i_blocks if the block allocation is delayed
5750 * otherwise in the case of system crash before the real block
5751 * allocation is done, we will have i_blocks inconsistent with
5752 * on-disk file blocks.
5753 * We always keep i_blocks updated together with real
5754 * allocation. But to not confuse with user, stat
5755 * will return the blocks that include the delayed allocation
5756 * blocks for this file.
5757 */
5758 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5759 EXT4_I(inode)->i_reserved_data_blocks);
5760 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5761 return 0;
5762}
5763
5764static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5765 int pextents)
5766{
5767 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5768 return ext4_ind_trans_blocks(inode, lblocks);
5769 return ext4_ext_index_trans_blocks(inode, pextents);
5770}
5771
5772/*
5773 * Account for index blocks, block groups bitmaps and block group
5774 * descriptor blocks if modify datablocks and index blocks
5775 * worse case, the indexs blocks spread over different block groups
5776 *
5777 * If datablocks are discontiguous, they are possible to spread over
5778 * different block groups too. If they are contiguous, with flexbg,
5779 * they could still across block group boundary.
5780 *
5781 * Also account for superblock, inode, quota and xattr blocks
5782 */
5783static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5784 int pextents)
5785{
5786 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5787 int gdpblocks;
5788 int idxblocks;
5789 int ret = 0;
5790
5791 /*
5792 * How many index blocks need to touch to map @lblocks logical blocks
5793 * to @pextents physical extents?
5794 */
5795 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5796
5797 ret = idxblocks;
5798
5799 /*
5800 * Now let's see how many group bitmaps and group descriptors need
5801 * to account
5802 */
5803 groups = idxblocks + pextents;
5804 gdpblocks = groups;
5805 if (groups > ngroups)
5806 groups = ngroups;
5807 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5808 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5809
5810 /* bitmaps and block group descriptor blocks */
5811 ret += groups + gdpblocks;
5812
5813 /* Blocks for super block, inode, quota and xattr blocks */
5814 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5815
5816 return ret;
5817}
5818
5819/*
5820 * Calculate the total number of credits to reserve to fit
5821 * the modification of a single pages into a single transaction,
5822 * which may include multiple chunks of block allocations.
5823 *
5824 * This could be called via ext4_write_begin()
5825 *
5826 * We need to consider the worse case, when
5827 * one new block per extent.
5828 */
5829int ext4_writepage_trans_blocks(struct inode *inode)
5830{
5831 int bpp = ext4_journal_blocks_per_page(inode);
5832 int ret;
5833
5834 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5835
5836 /* Account for data blocks for journalled mode */
5837 if (ext4_should_journal_data(inode))
5838 ret += bpp;
5839 return ret;
5840}
5841
5842/*
5843 * Calculate the journal credits for a chunk of data modification.
5844 *
5845 * This is called from DIO, fallocate or whoever calling
5846 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5847 *
5848 * journal buffers for data blocks are not included here, as DIO
5849 * and fallocate do no need to journal data buffers.
5850 */
5851int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5852{
5853 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5854}
5855
5856/*
5857 * The caller must have previously called ext4_reserve_inode_write().
5858 * Give this, we know that the caller already has write access to iloc->bh.
5859 */
5860int ext4_mark_iloc_dirty(handle_t *handle,
5861 struct inode *inode, struct ext4_iloc *iloc)
5862{
5863 int err = 0;
5864
5865 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5866 put_bh(iloc->bh);
5867 return -EIO;
5868 }
5869 if (IS_I_VERSION(inode))
5870 inode_inc_iversion(inode);
5871
5872 /* the do_update_inode consumes one bh->b_count */
5873 get_bh(iloc->bh);
5874
5875 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5876 err = ext4_do_update_inode(handle, inode, iloc);
5877 put_bh(iloc->bh);
5878 return err;
5879}
5880
5881/*
5882 * On success, We end up with an outstanding reference count against
5883 * iloc->bh. This _must_ be cleaned up later.
5884 */
5885
5886int
5887ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5888 struct ext4_iloc *iloc)
5889{
5890 int err;
5891
5892 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5893 return -EIO;
5894
5895 err = ext4_get_inode_loc(inode, iloc);
5896 if (!err) {
5897 BUFFER_TRACE(iloc->bh, "get_write_access");
5898 err = ext4_journal_get_write_access(handle, iloc->bh);
5899 if (err) {
5900 brelse(iloc->bh);
5901 iloc->bh = NULL;
5902 }
5903 }
5904 ext4_std_error(inode->i_sb, err);
5905 return err;
5906}
5907
5908static int __ext4_expand_extra_isize(struct inode *inode,
5909 unsigned int new_extra_isize,
5910 struct ext4_iloc *iloc,
5911 handle_t *handle, int *no_expand)
5912{
5913 struct ext4_inode *raw_inode;
5914 struct ext4_xattr_ibody_header *header;
5915 int error;
5916
5917 raw_inode = ext4_raw_inode(iloc);
5918
5919 header = IHDR(inode, raw_inode);
5920
5921 /* No extended attributes present */
5922 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5923 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5924 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5925 EXT4_I(inode)->i_extra_isize, 0,
5926 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5927 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5928 return 0;
5929 }
5930
5931 /* try to expand with EAs present */
5932 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5933 raw_inode, handle);
5934 if (error) {
5935 /*
5936 * Inode size expansion failed; don't try again
5937 */
5938 *no_expand = 1;
5939 }
5940
5941 return error;
5942}
5943
5944/*
5945 * Expand an inode by new_extra_isize bytes.
5946 * Returns 0 on success or negative error number on failure.
5947 */
5948static int ext4_try_to_expand_extra_isize(struct inode *inode,
5949 unsigned int new_extra_isize,
5950 struct ext4_iloc iloc,
5951 handle_t *handle)
5952{
5953 int no_expand;
5954 int error;
5955
5956 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5957 return -EOVERFLOW;
5958
5959 /*
5960 * In nojournal mode, we can immediately attempt to expand
5961 * the inode. When journaled, we first need to obtain extra
5962 * buffer credits since we may write into the EA block
5963 * with this same handle. If journal_extend fails, then it will
5964 * only result in a minor loss of functionality for that inode.
5965 * If this is felt to be critical, then e2fsck should be run to
5966 * force a large enough s_min_extra_isize.
5967 */
5968 if (ext4_handle_valid(handle) &&
5969 jbd2_journal_extend(handle,
5970 EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0)
5971 return -ENOSPC;
5972
5973 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5974 return -EBUSY;
5975
5976 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5977 handle, &no_expand);
5978 ext4_write_unlock_xattr(inode, &no_expand);
5979
5980 return error;
5981}
5982
5983int ext4_expand_extra_isize(struct inode *inode,
5984 unsigned int new_extra_isize,
5985 struct ext4_iloc *iloc)
5986{
5987 handle_t *handle;
5988 int no_expand;
5989 int error, rc;
5990
5991 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5992 brelse(iloc->bh);
5993 return -EOVERFLOW;
5994 }
5995
5996 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5997 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5998 if (IS_ERR(handle)) {
5999 error = PTR_ERR(handle);
6000 brelse(iloc->bh);
6001 return error;
6002 }
6003
6004 ext4_write_lock_xattr(inode, &no_expand);
6005
6006 BUFFER_TRACE(iloc->bh, "get_write_access");
6007 error = ext4_journal_get_write_access(handle, iloc->bh);
6008 if (error) {
6009 brelse(iloc->bh);
6010 goto out_stop;
6011 }
6012
6013 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
6014 handle, &no_expand);
6015
6016 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
6017 if (!error)
6018 error = rc;
6019
6020 ext4_write_unlock_xattr(inode, &no_expand);
6021out_stop:
6022 ext4_journal_stop(handle);
6023 return error;
6024}
6025
6026/*
6027 * What we do here is to mark the in-core inode as clean with respect to inode
6028 * dirtiness (it may still be data-dirty).
6029 * This means that the in-core inode may be reaped by prune_icache
6030 * without having to perform any I/O. This is a very good thing,
6031 * because *any* task may call prune_icache - even ones which
6032 * have a transaction open against a different journal.
6033 *
6034 * Is this cheating? Not really. Sure, we haven't written the
6035 * inode out, but prune_icache isn't a user-visible syncing function.
6036 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
6037 * we start and wait on commits.
6038 */
6039int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
6040{
6041 struct ext4_iloc iloc;
6042 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6043 int err;
6044
6045 might_sleep();
6046 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
6047 err = ext4_reserve_inode_write(handle, inode, &iloc);
6048 if (err)
6049 return err;
6050
6051 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
6052 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
6053 iloc, handle);
6054
6055 return ext4_mark_iloc_dirty(handle, inode, &iloc);
6056}
6057
6058/*
6059 * ext4_dirty_inode() is called from __mark_inode_dirty()
6060 *
6061 * We're really interested in the case where a file is being extended.
6062 * i_size has been changed by generic_commit_write() and we thus need
6063 * to include the updated inode in the current transaction.
6064 *
6065 * Also, dquot_alloc_block() will always dirty the inode when blocks
6066 * are allocated to the file.
6067 *
6068 * If the inode is marked synchronous, we don't honour that here - doing
6069 * so would cause a commit on atime updates, which we don't bother doing.
6070 * We handle synchronous inodes at the highest possible level.
6071 *
6072 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
6073 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
6074 * to copy into the on-disk inode structure are the timestamp files.
6075 */
6076void ext4_dirty_inode(struct inode *inode, int flags)
6077{
6078 handle_t *handle;
6079
6080 if (flags == I_DIRTY_TIME)
6081 return;
6082 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
6083 if (IS_ERR(handle))
6084 goto out;
6085
6086 ext4_mark_inode_dirty(handle, inode);
6087
6088 ext4_journal_stop(handle);
6089out:
6090 return;
6091}
6092
6093int ext4_change_inode_journal_flag(struct inode *inode, int val)
6094{
6095 journal_t *journal;
6096 handle_t *handle;
6097 int err;
6098 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
6099
6100 /*
6101 * We have to be very careful here: changing a data block's
6102 * journaling status dynamically is dangerous. If we write a
6103 * data block to the journal, change the status and then delete
6104 * that block, we risk forgetting to revoke the old log record
6105 * from the journal and so a subsequent replay can corrupt data.
6106 * So, first we make sure that the journal is empty and that
6107 * nobody is changing anything.
6108 */
6109
6110 journal = EXT4_JOURNAL(inode);
6111 if (!journal)
6112 return 0;
6113 if (is_journal_aborted(journal))
6114 return -EROFS;
6115
6116 /* Wait for all existing dio workers */
6117 inode_dio_wait(inode);
6118
6119 /*
6120 * Before flushing the journal and switching inode's aops, we have
6121 * to flush all dirty data the inode has. There can be outstanding
6122 * delayed allocations, there can be unwritten extents created by
6123 * fallocate or buffered writes in dioread_nolock mode covered by
6124 * dirty data which can be converted only after flushing the dirty
6125 * data (and journalled aops don't know how to handle these cases).
6126 */
6127 if (val) {
6128 down_write(&EXT4_I(inode)->i_mmap_sem);
6129 err = filemap_write_and_wait(inode->i_mapping);
6130 if (err < 0) {
6131 up_write(&EXT4_I(inode)->i_mmap_sem);
6132 return err;
6133 }
6134 }
6135
6136 percpu_down_write(&sbi->s_journal_flag_rwsem);
6137 jbd2_journal_lock_updates(journal);
6138
6139 /*
6140 * OK, there are no updates running now, and all cached data is
6141 * synced to disk. We are now in a completely consistent state
6142 * which doesn't have anything in the journal, and we know that
6143 * no filesystem updates are running, so it is safe to modify
6144 * the inode's in-core data-journaling state flag now.
6145 */
6146
6147 if (val)
6148 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6149 else {
6150 err = jbd2_journal_flush(journal);
6151 if (err < 0) {
6152 jbd2_journal_unlock_updates(journal);
6153 percpu_up_write(&sbi->s_journal_flag_rwsem);
6154 return err;
6155 }
6156 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6157 }
6158 ext4_set_aops(inode);
6159
6160 jbd2_journal_unlock_updates(journal);
6161 percpu_up_write(&sbi->s_journal_flag_rwsem);
6162
6163 if (val)
6164 up_write(&EXT4_I(inode)->i_mmap_sem);
6165
6166 /* Finally we can mark the inode as dirty. */
6167
6168 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6169 if (IS_ERR(handle))
6170 return PTR_ERR(handle);
6171
6172 err = ext4_mark_inode_dirty(handle, inode);
6173 ext4_handle_sync(handle);
6174 ext4_journal_stop(handle);
6175 ext4_std_error(inode->i_sb, err);
6176
6177 return err;
6178}
6179
6180static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6181{
6182 return !buffer_mapped(bh);
6183}
6184
6185vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6186{
6187 struct vm_area_struct *vma = vmf->vma;
6188 struct page *page = vmf->page;
6189 loff_t size;
6190 unsigned long len;
6191 int err;
6192 vm_fault_t ret;
6193 struct file *file = vma->vm_file;
6194 struct inode *inode = file_inode(file);
6195 struct address_space *mapping = inode->i_mapping;
6196 handle_t *handle;
6197 get_block_t *get_block;
6198 int retries = 0;
6199
6200 if (unlikely(IS_IMMUTABLE(inode)))
6201 return VM_FAULT_SIGBUS;
6202
6203 sb_start_pagefault(inode->i_sb);
6204 file_update_time(vma->vm_file);
6205
6206 down_read(&EXT4_I(inode)->i_mmap_sem);
6207
6208 err = ext4_convert_inline_data(inode);
6209 if (err)
6210 goto out_ret;
6211
6212 /* Delalloc case is easy... */
6213 if (test_opt(inode->i_sb, DELALLOC) &&
6214 !ext4_should_journal_data(inode) &&
6215 !ext4_nonda_switch(inode->i_sb)) {
6216 do {
6217 err = block_page_mkwrite(vma, vmf,
6218 ext4_da_get_block_prep);
6219 } while (err == -ENOSPC &&
6220 ext4_should_retry_alloc(inode->i_sb, &retries));
6221 goto out_ret;
6222 }
6223
6224 lock_page(page);
6225 size = i_size_read(inode);
6226 /* Page got truncated from under us? */
6227 if (page->mapping != mapping || page_offset(page) > size) {
6228 unlock_page(page);
6229 ret = VM_FAULT_NOPAGE;
6230 goto out;
6231 }
6232
6233 if (page->index == size >> PAGE_SHIFT)
6234 len = size & ~PAGE_MASK;
6235 else
6236 len = PAGE_SIZE;
6237 /*
6238 * Return if we have all the buffers mapped. This avoids the need to do
6239 * journal_start/journal_stop which can block and take a long time
6240 */
6241 if (page_has_buffers(page)) {
6242 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6243 0, len, NULL,
6244 ext4_bh_unmapped)) {
6245 /* Wait so that we don't change page under IO */
6246 wait_for_stable_page(page);
6247 ret = VM_FAULT_LOCKED;
6248 goto out;
6249 }
6250 }
6251 unlock_page(page);
6252 /* OK, we need to fill the hole... */
6253 if (ext4_should_dioread_nolock(inode))
6254 get_block = ext4_get_block_unwritten;
6255 else
6256 get_block = ext4_get_block;
6257retry_alloc:
6258 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6259 ext4_writepage_trans_blocks(inode));
6260 if (IS_ERR(handle)) {
6261 ret = VM_FAULT_SIGBUS;
6262 goto out;
6263 }
6264 err = block_page_mkwrite(vma, vmf, get_block);
6265 if (!err && ext4_should_journal_data(inode)) {
6266 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
6267 PAGE_SIZE, NULL, do_journal_get_write_access)) {
6268 unlock_page(page);
6269 ret = VM_FAULT_SIGBUS;
6270 ext4_journal_stop(handle);
6271 goto out;
6272 }
6273 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6274 }
6275 ext4_journal_stop(handle);
6276 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6277 goto retry_alloc;
6278out_ret:
6279 ret = block_page_mkwrite_return(err);
6280out:
6281 up_read(&EXT4_I(inode)->i_mmap_sem);
6282 sb_end_pagefault(inode->i_sb);
6283 return ret;
6284}
6285
6286vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6287{
6288 struct inode *inode = file_inode(vmf->vma->vm_file);
6289 vm_fault_t ret;
6290
6291 down_read(&EXT4_I(inode)->i_mmap_sem);
6292 ret = filemap_fault(vmf);
6293 up_read(&EXT4_I(inode)->i_mmap_sem);
6294
6295 return ret;
6296}