Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
5 * All Rights Reserved.
6 */
7#include "xfs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_inode.h"
14#include "xfs_trans.h"
15#include "xfs_iomap.h"
16#include "xfs_trace.h"
17#include "xfs_bmap.h"
18#include "xfs_bmap_util.h"
19#include "xfs_reflink.h"
20#include "xfs_errortag.h"
21#include "xfs_error.h"
22
23struct xfs_writepage_ctx {
24 struct iomap_writepage_ctx ctx;
25 unsigned int data_seq;
26 unsigned int cow_seq;
27};
28
29static inline struct xfs_writepage_ctx *
30XFS_WPC(struct iomap_writepage_ctx *ctx)
31{
32 return container_of(ctx, struct xfs_writepage_ctx, ctx);
33}
34
35/*
36 * Fast and loose check if this write could update the on-disk inode size.
37 */
38static inline bool xfs_ioend_is_append(struct iomap_ioend *ioend)
39{
40 return ioend->io_offset + ioend->io_size >
41 XFS_I(ioend->io_inode)->i_disk_size;
42}
43
44/*
45 * Update on-disk file size now that data has been written to disk.
46 */
47int
48xfs_setfilesize(
49 struct xfs_inode *ip,
50 xfs_off_t offset,
51 size_t size)
52{
53 struct xfs_mount *mp = ip->i_mount;
54 struct xfs_trans *tp;
55 xfs_fsize_t isize;
56 int error;
57
58 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
59 if (error)
60 return error;
61
62 xfs_ilock(ip, XFS_ILOCK_EXCL);
63 isize = xfs_new_eof(ip, offset + size);
64 if (!isize) {
65 xfs_iunlock(ip, XFS_ILOCK_EXCL);
66 xfs_trans_cancel(tp);
67 return 0;
68 }
69
70 trace_xfs_setfilesize(ip, offset, size);
71
72 ip->i_disk_size = isize;
73 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
74 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
75
76 return xfs_trans_commit(tp);
77}
78
79/*
80 * IO write completion.
81 */
82STATIC void
83xfs_end_ioend(
84 struct iomap_ioend *ioend)
85{
86 struct xfs_inode *ip = XFS_I(ioend->io_inode);
87 struct xfs_mount *mp = ip->i_mount;
88 xfs_off_t offset = ioend->io_offset;
89 size_t size = ioend->io_size;
90 unsigned int nofs_flag;
91 int error;
92
93 /*
94 * We can allocate memory here while doing writeback on behalf of
95 * memory reclaim. To avoid memory allocation deadlocks set the
96 * task-wide nofs context for the following operations.
97 */
98 nofs_flag = memalloc_nofs_save();
99
100 /*
101 * Just clean up the in-memory structures if the fs has been shut down.
102 */
103 if (xfs_is_shutdown(mp)) {
104 error = -EIO;
105 goto done;
106 }
107
108 /*
109 * Clean up all COW blocks and underlying data fork delalloc blocks on
110 * I/O error. The delalloc punch is required because this ioend was
111 * mapped to blocks in the COW fork and the associated pages are no
112 * longer dirty. If we don't remove delalloc blocks here, they become
113 * stale and can corrupt free space accounting on unmount.
114 */
115 error = blk_status_to_errno(ioend->io_bio->bi_status);
116 if (unlikely(error)) {
117 if (ioend->io_flags & IOMAP_F_SHARED) {
118 xfs_reflink_cancel_cow_range(ip, offset, size, true);
119 xfs_bmap_punch_delalloc_range(ip, offset,
120 offset + size);
121 }
122 goto done;
123 }
124
125 /*
126 * Success: commit the COW or unwritten blocks if needed.
127 */
128 if (ioend->io_flags & IOMAP_F_SHARED)
129 error = xfs_reflink_end_cow(ip, offset, size);
130 else if (ioend->io_type == IOMAP_UNWRITTEN)
131 error = xfs_iomap_write_unwritten(ip, offset, size, false);
132
133 if (!error && xfs_ioend_is_append(ioend))
134 error = xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
135done:
136 iomap_finish_ioends(ioend, error);
137 memalloc_nofs_restore(nofs_flag);
138}
139
140/*
141 * Finish all pending IO completions that require transactional modifications.
142 *
143 * We try to merge physical and logically contiguous ioends before completion to
144 * minimise the number of transactions we need to perform during IO completion.
145 * Both unwritten extent conversion and COW remapping need to iterate and modify
146 * one physical extent at a time, so we gain nothing by merging physically
147 * discontiguous extents here.
148 *
149 * The ioend chain length that we can be processing here is largely unbound in
150 * length and we may have to perform significant amounts of work on each ioend
151 * to complete it. Hence we have to be careful about holding the CPU for too
152 * long in this loop.
153 */
154void
155xfs_end_io(
156 struct work_struct *work)
157{
158 struct xfs_inode *ip =
159 container_of(work, struct xfs_inode, i_ioend_work);
160 struct iomap_ioend *ioend;
161 struct list_head tmp;
162 unsigned long flags;
163
164 spin_lock_irqsave(&ip->i_ioend_lock, flags);
165 list_replace_init(&ip->i_ioend_list, &tmp);
166 spin_unlock_irqrestore(&ip->i_ioend_lock, flags);
167
168 iomap_sort_ioends(&tmp);
169 while ((ioend = list_first_entry_or_null(&tmp, struct iomap_ioend,
170 io_list))) {
171 list_del_init(&ioend->io_list);
172 iomap_ioend_try_merge(ioend, &tmp);
173 xfs_end_ioend(ioend);
174 cond_resched();
175 }
176}
177
178STATIC void
179xfs_end_bio(
180 struct bio *bio)
181{
182 struct iomap_ioend *ioend = bio->bi_private;
183 struct xfs_inode *ip = XFS_I(ioend->io_inode);
184 unsigned long flags;
185
186 spin_lock_irqsave(&ip->i_ioend_lock, flags);
187 if (list_empty(&ip->i_ioend_list))
188 WARN_ON_ONCE(!queue_work(ip->i_mount->m_unwritten_workqueue,
189 &ip->i_ioend_work));
190 list_add_tail(&ioend->io_list, &ip->i_ioend_list);
191 spin_unlock_irqrestore(&ip->i_ioend_lock, flags);
192}
193
194/*
195 * Fast revalidation of the cached writeback mapping. Return true if the current
196 * mapping is valid, false otherwise.
197 */
198static bool
199xfs_imap_valid(
200 struct iomap_writepage_ctx *wpc,
201 struct xfs_inode *ip,
202 loff_t offset)
203{
204 if (offset < wpc->iomap.offset ||
205 offset >= wpc->iomap.offset + wpc->iomap.length)
206 return false;
207 /*
208 * If this is a COW mapping, it is sufficient to check that the mapping
209 * covers the offset. Be careful to check this first because the caller
210 * can revalidate a COW mapping without updating the data seqno.
211 */
212 if (wpc->iomap.flags & IOMAP_F_SHARED)
213 return true;
214
215 /*
216 * This is not a COW mapping. Check the sequence number of the data fork
217 * because concurrent changes could have invalidated the extent. Check
218 * the COW fork because concurrent changes since the last time we
219 * checked (and found nothing at this offset) could have added
220 * overlapping blocks.
221 */
222 if (XFS_WPC(wpc)->data_seq != READ_ONCE(ip->i_df.if_seq)) {
223 trace_xfs_wb_data_iomap_invalid(ip, &wpc->iomap,
224 XFS_WPC(wpc)->data_seq, XFS_DATA_FORK);
225 return false;
226 }
227 if (xfs_inode_has_cow_data(ip) &&
228 XFS_WPC(wpc)->cow_seq != READ_ONCE(ip->i_cowfp->if_seq)) {
229 trace_xfs_wb_cow_iomap_invalid(ip, &wpc->iomap,
230 XFS_WPC(wpc)->cow_seq, XFS_COW_FORK);
231 return false;
232 }
233 return true;
234}
235
236/*
237 * Pass in a dellalloc extent and convert it to real extents, return the real
238 * extent that maps offset_fsb in wpc->iomap.
239 *
240 * The current page is held locked so nothing could have removed the block
241 * backing offset_fsb, although it could have moved from the COW to the data
242 * fork by another thread.
243 */
244static int
245xfs_convert_blocks(
246 struct iomap_writepage_ctx *wpc,
247 struct xfs_inode *ip,
248 int whichfork,
249 loff_t offset)
250{
251 int error;
252 unsigned *seq;
253
254 if (whichfork == XFS_COW_FORK)
255 seq = &XFS_WPC(wpc)->cow_seq;
256 else
257 seq = &XFS_WPC(wpc)->data_seq;
258
259 /*
260 * Attempt to allocate whatever delalloc extent currently backs offset
261 * and put the result into wpc->iomap. Allocate in a loop because it
262 * may take several attempts to allocate real blocks for a contiguous
263 * delalloc extent if free space is sufficiently fragmented.
264 */
265 do {
266 error = xfs_bmapi_convert_delalloc(ip, whichfork, offset,
267 &wpc->iomap, seq);
268 if (error)
269 return error;
270 } while (wpc->iomap.offset + wpc->iomap.length <= offset);
271
272 return 0;
273}
274
275static int
276xfs_map_blocks(
277 struct iomap_writepage_ctx *wpc,
278 struct inode *inode,
279 loff_t offset)
280{
281 struct xfs_inode *ip = XFS_I(inode);
282 struct xfs_mount *mp = ip->i_mount;
283 ssize_t count = i_blocksize(inode);
284 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
285 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + count);
286 xfs_fileoff_t cow_fsb;
287 int whichfork;
288 struct xfs_bmbt_irec imap;
289 struct xfs_iext_cursor icur;
290 int retries = 0;
291 int error = 0;
292
293 if (xfs_is_shutdown(mp))
294 return -EIO;
295
296 XFS_ERRORTAG_DELAY(mp, XFS_ERRTAG_WB_DELAY_MS);
297
298 /*
299 * COW fork blocks can overlap data fork blocks even if the blocks
300 * aren't shared. COW I/O always takes precedent, so we must always
301 * check for overlap on reflink inodes unless the mapping is already a
302 * COW one, or the COW fork hasn't changed from the last time we looked
303 * at it.
304 *
305 * It's safe to check the COW fork if_seq here without the ILOCK because
306 * we've indirectly protected against concurrent updates: writeback has
307 * the page locked, which prevents concurrent invalidations by reflink
308 * and directio and prevents concurrent buffered writes to the same
309 * page. Changes to if_seq always happen under i_lock, which protects
310 * against concurrent updates and provides a memory barrier on the way
311 * out that ensures that we always see the current value.
312 */
313 if (xfs_imap_valid(wpc, ip, offset))
314 return 0;
315
316 /*
317 * If we don't have a valid map, now it's time to get a new one for this
318 * offset. This will convert delayed allocations (including COW ones)
319 * into real extents. If we return without a valid map, it means we
320 * landed in a hole and we skip the block.
321 */
322retry:
323 cow_fsb = NULLFILEOFF;
324 whichfork = XFS_DATA_FORK;
325 xfs_ilock(ip, XFS_ILOCK_SHARED);
326 ASSERT(!xfs_need_iread_extents(&ip->i_df));
327
328 /*
329 * Check if this is offset is covered by a COW extents, and if yes use
330 * it directly instead of looking up anything in the data fork.
331 */
332 if (xfs_inode_has_cow_data(ip) &&
333 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &imap))
334 cow_fsb = imap.br_startoff;
335 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
336 XFS_WPC(wpc)->cow_seq = READ_ONCE(ip->i_cowfp->if_seq);
337 xfs_iunlock(ip, XFS_ILOCK_SHARED);
338
339 whichfork = XFS_COW_FORK;
340 goto allocate_blocks;
341 }
342
343 /*
344 * No COW extent overlap. Revalidate now that we may have updated
345 * ->cow_seq. If the data mapping is still valid, we're done.
346 */
347 if (xfs_imap_valid(wpc, ip, offset)) {
348 xfs_iunlock(ip, XFS_ILOCK_SHARED);
349 return 0;
350 }
351
352 /*
353 * If we don't have a valid map, now it's time to get a new one for this
354 * offset. This will convert delayed allocations (including COW ones)
355 * into real extents.
356 */
357 if (!xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap))
358 imap.br_startoff = end_fsb; /* fake a hole past EOF */
359 XFS_WPC(wpc)->data_seq = READ_ONCE(ip->i_df.if_seq);
360 xfs_iunlock(ip, XFS_ILOCK_SHARED);
361
362 /* landed in a hole or beyond EOF? */
363 if (imap.br_startoff > offset_fsb) {
364 imap.br_blockcount = imap.br_startoff - offset_fsb;
365 imap.br_startoff = offset_fsb;
366 imap.br_startblock = HOLESTARTBLOCK;
367 imap.br_state = XFS_EXT_NORM;
368 }
369
370 /*
371 * Truncate to the next COW extent if there is one. This is the only
372 * opportunity to do this because we can skip COW fork lookups for the
373 * subsequent blocks in the mapping; however, the requirement to treat
374 * the COW range separately remains.
375 */
376 if (cow_fsb != NULLFILEOFF &&
377 cow_fsb < imap.br_startoff + imap.br_blockcount)
378 imap.br_blockcount = cow_fsb - imap.br_startoff;
379
380 /* got a delalloc extent? */
381 if (imap.br_startblock != HOLESTARTBLOCK &&
382 isnullstartblock(imap.br_startblock))
383 goto allocate_blocks;
384
385 xfs_bmbt_to_iomap(ip, &wpc->iomap, &imap, 0, 0, XFS_WPC(wpc)->data_seq);
386 trace_xfs_map_blocks_found(ip, offset, count, whichfork, &imap);
387 return 0;
388allocate_blocks:
389 error = xfs_convert_blocks(wpc, ip, whichfork, offset);
390 if (error) {
391 /*
392 * If we failed to find the extent in the COW fork we might have
393 * raced with a COW to data fork conversion or truncate.
394 * Restart the lookup to catch the extent in the data fork for
395 * the former case, but prevent additional retries to avoid
396 * looping forever for the latter case.
397 */
398 if (error == -EAGAIN && whichfork == XFS_COW_FORK && !retries++)
399 goto retry;
400 ASSERT(error != -EAGAIN);
401 return error;
402 }
403
404 /*
405 * Due to merging the return real extent might be larger than the
406 * original delalloc one. Trim the return extent to the next COW
407 * boundary again to force a re-lookup.
408 */
409 if (whichfork != XFS_COW_FORK && cow_fsb != NULLFILEOFF) {
410 loff_t cow_offset = XFS_FSB_TO_B(mp, cow_fsb);
411
412 if (cow_offset < wpc->iomap.offset + wpc->iomap.length)
413 wpc->iomap.length = cow_offset - wpc->iomap.offset;
414 }
415
416 ASSERT(wpc->iomap.offset <= offset);
417 ASSERT(wpc->iomap.offset + wpc->iomap.length > offset);
418 trace_xfs_map_blocks_alloc(ip, offset, count, whichfork, &imap);
419 return 0;
420}
421
422static int
423xfs_prepare_ioend(
424 struct iomap_ioend *ioend,
425 int status)
426{
427 unsigned int nofs_flag;
428
429 /*
430 * We can allocate memory here while doing writeback on behalf of
431 * memory reclaim. To avoid memory allocation deadlocks set the
432 * task-wide nofs context for the following operations.
433 */
434 nofs_flag = memalloc_nofs_save();
435
436 /* Convert CoW extents to regular */
437 if (!status && (ioend->io_flags & IOMAP_F_SHARED)) {
438 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
439 ioend->io_offset, ioend->io_size);
440 }
441
442 memalloc_nofs_restore(nofs_flag);
443
444 /* send ioends that might require a transaction to the completion wq */
445 if (xfs_ioend_is_append(ioend) || ioend->io_type == IOMAP_UNWRITTEN ||
446 (ioend->io_flags & IOMAP_F_SHARED))
447 ioend->io_bio->bi_end_io = xfs_end_bio;
448 return status;
449}
450
451/*
452 * If the page has delalloc blocks on it, we need to punch them out before we
453 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
454 * inode that can trip up a later direct I/O read operation on the same region.
455 *
456 * We prevent this by truncating away the delalloc regions on the page. Because
457 * they are delalloc, we can do this without needing a transaction. Indeed - if
458 * we get ENOSPC errors, we have to be able to do this truncation without a
459 * transaction as there is no space left for block reservation (typically why we
460 * see a ENOSPC in writeback).
461 */
462static void
463xfs_discard_folio(
464 struct folio *folio,
465 loff_t pos)
466{
467 struct xfs_inode *ip = XFS_I(folio->mapping->host);
468 struct xfs_mount *mp = ip->i_mount;
469 int error;
470
471 if (xfs_is_shutdown(mp))
472 return;
473
474 xfs_alert_ratelimited(mp,
475 "page discard on page "PTR_FMT", inode 0x%llx, pos %llu.",
476 folio, ip->i_ino, pos);
477
478 error = xfs_bmap_punch_delalloc_range(ip, pos,
479 round_up(pos, folio_size(folio)));
480
481 if (error && !xfs_is_shutdown(mp))
482 xfs_alert(mp, "page discard unable to remove delalloc mapping.");
483}
484
485static const struct iomap_writeback_ops xfs_writeback_ops = {
486 .map_blocks = xfs_map_blocks,
487 .prepare_ioend = xfs_prepare_ioend,
488 .discard_folio = xfs_discard_folio,
489};
490
491STATIC int
492xfs_vm_writepages(
493 struct address_space *mapping,
494 struct writeback_control *wbc)
495{
496 struct xfs_writepage_ctx wpc = { };
497
498 /*
499 * Writing back data in a transaction context can result in recursive
500 * transactions. This is bad, so issue a warning and get out of here.
501 */
502 if (WARN_ON_ONCE(current->journal_info))
503 return 0;
504
505 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
506 return iomap_writepages(mapping, wbc, &wpc.ctx, &xfs_writeback_ops);
507}
508
509STATIC int
510xfs_dax_writepages(
511 struct address_space *mapping,
512 struct writeback_control *wbc)
513{
514 struct xfs_inode *ip = XFS_I(mapping->host);
515
516 xfs_iflags_clear(ip, XFS_ITRUNCATED);
517 return dax_writeback_mapping_range(mapping,
518 xfs_inode_buftarg(ip)->bt_daxdev, wbc);
519}
520
521STATIC sector_t
522xfs_vm_bmap(
523 struct address_space *mapping,
524 sector_t block)
525{
526 struct xfs_inode *ip = XFS_I(mapping->host);
527
528 trace_xfs_vm_bmap(ip);
529
530 /*
531 * The swap code (ab-)uses ->bmap to get a block mapping and then
532 * bypasses the file system for actual I/O. We really can't allow
533 * that on reflinks inodes, so we have to skip out here. And yes,
534 * 0 is the magic code for a bmap error.
535 *
536 * Since we don't pass back blockdev info, we can't return bmap
537 * information for rt files either.
538 */
539 if (xfs_is_cow_inode(ip) || XFS_IS_REALTIME_INODE(ip))
540 return 0;
541 return iomap_bmap(mapping, block, &xfs_read_iomap_ops);
542}
543
544STATIC int
545xfs_vm_read_folio(
546 struct file *unused,
547 struct folio *folio)
548{
549 return iomap_read_folio(folio, &xfs_read_iomap_ops);
550}
551
552STATIC void
553xfs_vm_readahead(
554 struct readahead_control *rac)
555{
556 iomap_readahead(rac, &xfs_read_iomap_ops);
557}
558
559static int
560xfs_iomap_swapfile_activate(
561 struct swap_info_struct *sis,
562 struct file *swap_file,
563 sector_t *span)
564{
565 sis->bdev = xfs_inode_buftarg(XFS_I(file_inode(swap_file)))->bt_bdev;
566 return iomap_swapfile_activate(sis, swap_file, span,
567 &xfs_read_iomap_ops);
568}
569
570const struct address_space_operations xfs_address_space_operations = {
571 .read_folio = xfs_vm_read_folio,
572 .readahead = xfs_vm_readahead,
573 .writepages = xfs_vm_writepages,
574 .dirty_folio = filemap_dirty_folio,
575 .release_folio = iomap_release_folio,
576 .invalidate_folio = iomap_invalidate_folio,
577 .bmap = xfs_vm_bmap,
578 .direct_IO = noop_direct_IO,
579 .migrate_folio = filemap_migrate_folio,
580 .is_partially_uptodate = iomap_is_partially_uptodate,
581 .error_remove_page = generic_error_remove_page,
582 .swap_activate = xfs_iomap_swapfile_activate,
583};
584
585const struct address_space_operations xfs_dax_aops = {
586 .writepages = xfs_dax_writepages,
587 .direct_IO = noop_direct_IO,
588 .dirty_folio = noop_dirty_folio,
589 .swap_activate = xfs_iomap_swapfile_activate,
590};
1/*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_shared.h"
20#include "xfs_format.h"
21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
23#include "xfs_mount.h"
24#include "xfs_inode.h"
25#include "xfs_trans.h"
26#include "xfs_inode_item.h"
27#include "xfs_alloc.h"
28#include "xfs_error.h"
29#include "xfs_iomap.h"
30#include "xfs_trace.h"
31#include "xfs_bmap.h"
32#include "xfs_bmap_util.h"
33#include "xfs_bmap_btree.h"
34#include "xfs_reflink.h"
35#include <linux/gfp.h>
36#include <linux/mpage.h>
37#include <linux/pagevec.h>
38#include <linux/writeback.h>
39
40/*
41 * structure owned by writepages passed to individual writepage calls
42 */
43struct xfs_writepage_ctx {
44 struct xfs_bmbt_irec imap;
45 bool imap_valid;
46 unsigned int io_type;
47 struct xfs_ioend *ioend;
48 sector_t last_block;
49};
50
51void
52xfs_count_page_state(
53 struct page *page,
54 int *delalloc,
55 int *unwritten)
56{
57 struct buffer_head *bh, *head;
58
59 *delalloc = *unwritten = 0;
60
61 bh = head = page_buffers(page);
62 do {
63 if (buffer_unwritten(bh))
64 (*unwritten) = 1;
65 else if (buffer_delay(bh))
66 (*delalloc) = 1;
67 } while ((bh = bh->b_this_page) != head);
68}
69
70struct block_device *
71xfs_find_bdev_for_inode(
72 struct inode *inode)
73{
74 struct xfs_inode *ip = XFS_I(inode);
75 struct xfs_mount *mp = ip->i_mount;
76
77 if (XFS_IS_REALTIME_INODE(ip))
78 return mp->m_rtdev_targp->bt_bdev;
79 else
80 return mp->m_ddev_targp->bt_bdev;
81}
82
83/*
84 * We're now finished for good with this page. Update the page state via the
85 * associated buffer_heads, paying attention to the start and end offsets that
86 * we need to process on the page.
87 *
88 * Landmine Warning: bh->b_end_io() will call end_page_writeback() on the last
89 * buffer in the IO. Once it does this, it is unsafe to access the bufferhead or
90 * the page at all, as we may be racing with memory reclaim and it can free both
91 * the bufferhead chain and the page as it will see the page as clean and
92 * unused.
93 */
94static void
95xfs_finish_page_writeback(
96 struct inode *inode,
97 struct bio_vec *bvec,
98 int error)
99{
100 unsigned int end = bvec->bv_offset + bvec->bv_len - 1;
101 struct buffer_head *head, *bh, *next;
102 unsigned int off = 0;
103 unsigned int bsize;
104
105 ASSERT(bvec->bv_offset < PAGE_SIZE);
106 ASSERT((bvec->bv_offset & ((1 << inode->i_blkbits) - 1)) == 0);
107 ASSERT(end < PAGE_SIZE);
108 ASSERT((bvec->bv_len & ((1 << inode->i_blkbits) - 1)) == 0);
109
110 bh = head = page_buffers(bvec->bv_page);
111
112 bsize = bh->b_size;
113 do {
114 next = bh->b_this_page;
115 if (off < bvec->bv_offset)
116 goto next_bh;
117 if (off > end)
118 break;
119 bh->b_end_io(bh, !error);
120next_bh:
121 off += bsize;
122 } while ((bh = next) != head);
123}
124
125/*
126 * We're now finished for good with this ioend structure. Update the page
127 * state, release holds on bios, and finally free up memory. Do not use the
128 * ioend after this.
129 */
130STATIC void
131xfs_destroy_ioend(
132 struct xfs_ioend *ioend,
133 int error)
134{
135 struct inode *inode = ioend->io_inode;
136 struct bio *last = ioend->io_bio;
137 struct bio *bio, *next;
138
139 for (bio = &ioend->io_inline_bio; bio; bio = next) {
140 struct bio_vec *bvec;
141 int i;
142
143 /*
144 * For the last bio, bi_private points to the ioend, so we
145 * need to explicitly end the iteration here.
146 */
147 if (bio == last)
148 next = NULL;
149 else
150 next = bio->bi_private;
151
152 /* walk each page on bio, ending page IO on them */
153 bio_for_each_segment_all(bvec, bio, i)
154 xfs_finish_page_writeback(inode, bvec, error);
155
156 bio_put(bio);
157 }
158}
159
160/*
161 * Fast and loose check if this write could update the on-disk inode size.
162 */
163static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
164{
165 return ioend->io_offset + ioend->io_size >
166 XFS_I(ioend->io_inode)->i_d.di_size;
167}
168
169STATIC int
170xfs_setfilesize_trans_alloc(
171 struct xfs_ioend *ioend)
172{
173 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
174 struct xfs_trans *tp;
175 int error;
176
177 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
178 if (error)
179 return error;
180
181 ioend->io_append_trans = tp;
182
183 /*
184 * We may pass freeze protection with a transaction. So tell lockdep
185 * we released it.
186 */
187 __sb_writers_release(ioend->io_inode->i_sb, SB_FREEZE_FS);
188 /*
189 * We hand off the transaction to the completion thread now, so
190 * clear the flag here.
191 */
192 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
193 return 0;
194}
195
196/*
197 * Update on-disk file size now that data has been written to disk.
198 */
199STATIC int
200__xfs_setfilesize(
201 struct xfs_inode *ip,
202 struct xfs_trans *tp,
203 xfs_off_t offset,
204 size_t size)
205{
206 xfs_fsize_t isize;
207
208 xfs_ilock(ip, XFS_ILOCK_EXCL);
209 isize = xfs_new_eof(ip, offset + size);
210 if (!isize) {
211 xfs_iunlock(ip, XFS_ILOCK_EXCL);
212 xfs_trans_cancel(tp);
213 return 0;
214 }
215
216 trace_xfs_setfilesize(ip, offset, size);
217
218 ip->i_d.di_size = isize;
219 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
220 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
221
222 return xfs_trans_commit(tp);
223}
224
225int
226xfs_setfilesize(
227 struct xfs_inode *ip,
228 xfs_off_t offset,
229 size_t size)
230{
231 struct xfs_mount *mp = ip->i_mount;
232 struct xfs_trans *tp;
233 int error;
234
235 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp);
236 if (error)
237 return error;
238
239 return __xfs_setfilesize(ip, tp, offset, size);
240}
241
242STATIC int
243xfs_setfilesize_ioend(
244 struct xfs_ioend *ioend,
245 int error)
246{
247 struct xfs_inode *ip = XFS_I(ioend->io_inode);
248 struct xfs_trans *tp = ioend->io_append_trans;
249
250 /*
251 * The transaction may have been allocated in the I/O submission thread,
252 * thus we need to mark ourselves as being in a transaction manually.
253 * Similarly for freeze protection.
254 */
255 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
256 __sb_writers_acquired(VFS_I(ip)->i_sb, SB_FREEZE_FS);
257
258 /* we abort the update if there was an IO error */
259 if (error) {
260 xfs_trans_cancel(tp);
261 return error;
262 }
263
264 return __xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
265}
266
267/*
268 * IO write completion.
269 */
270STATIC void
271xfs_end_io(
272 struct work_struct *work)
273{
274 struct xfs_ioend *ioend =
275 container_of(work, struct xfs_ioend, io_work);
276 struct xfs_inode *ip = XFS_I(ioend->io_inode);
277 xfs_off_t offset = ioend->io_offset;
278 size_t size = ioend->io_size;
279 int error = ioend->io_bio->bi_error;
280
281 /*
282 * Just clean up the in-memory strutures if the fs has been shut down.
283 */
284 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
285 error = -EIO;
286 goto done;
287 }
288
289 /*
290 * Clean up any COW blocks on an I/O error.
291 */
292 if (unlikely(error)) {
293 switch (ioend->io_type) {
294 case XFS_IO_COW:
295 xfs_reflink_cancel_cow_range(ip, offset, size, true);
296 break;
297 }
298
299 goto done;
300 }
301
302 /*
303 * Success: commit the COW or unwritten blocks if needed.
304 */
305 switch (ioend->io_type) {
306 case XFS_IO_COW:
307 error = xfs_reflink_end_cow(ip, offset, size);
308 break;
309 case XFS_IO_UNWRITTEN:
310 error = xfs_iomap_write_unwritten(ip, offset, size);
311 break;
312 default:
313 ASSERT(!xfs_ioend_is_append(ioend) || ioend->io_append_trans);
314 break;
315 }
316
317done:
318 if (ioend->io_append_trans)
319 error = xfs_setfilesize_ioend(ioend, error);
320 xfs_destroy_ioend(ioend, error);
321}
322
323STATIC void
324xfs_end_bio(
325 struct bio *bio)
326{
327 struct xfs_ioend *ioend = bio->bi_private;
328 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
329
330 if (ioend->io_type == XFS_IO_UNWRITTEN || ioend->io_type == XFS_IO_COW)
331 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
332 else if (ioend->io_append_trans)
333 queue_work(mp->m_data_workqueue, &ioend->io_work);
334 else
335 xfs_destroy_ioend(ioend, bio->bi_error);
336}
337
338STATIC int
339xfs_map_blocks(
340 struct inode *inode,
341 loff_t offset,
342 struct xfs_bmbt_irec *imap,
343 int type)
344{
345 struct xfs_inode *ip = XFS_I(inode);
346 struct xfs_mount *mp = ip->i_mount;
347 ssize_t count = 1 << inode->i_blkbits;
348 xfs_fileoff_t offset_fsb, end_fsb;
349 int error = 0;
350 int bmapi_flags = XFS_BMAPI_ENTIRE;
351 int nimaps = 1;
352
353 if (XFS_FORCED_SHUTDOWN(mp))
354 return -EIO;
355
356 ASSERT(type != XFS_IO_COW);
357 if (type == XFS_IO_UNWRITTEN)
358 bmapi_flags |= XFS_BMAPI_IGSTATE;
359
360 xfs_ilock(ip, XFS_ILOCK_SHARED);
361 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
362 (ip->i_df.if_flags & XFS_IFEXTENTS));
363 ASSERT(offset <= mp->m_super->s_maxbytes);
364
365 if (offset + count > mp->m_super->s_maxbytes)
366 count = mp->m_super->s_maxbytes - offset;
367 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
368 offset_fsb = XFS_B_TO_FSBT(mp, offset);
369 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
370 imap, &nimaps, bmapi_flags);
371 /*
372 * Truncate an overwrite extent if there's a pending CoW
373 * reservation before the end of this extent. This forces us
374 * to come back to writepage to take care of the CoW.
375 */
376 if (nimaps && type == XFS_IO_OVERWRITE)
377 xfs_reflink_trim_irec_to_next_cow(ip, offset_fsb, imap);
378 xfs_iunlock(ip, XFS_ILOCK_SHARED);
379
380 if (error)
381 return error;
382
383 if (type == XFS_IO_DELALLOC &&
384 (!nimaps || isnullstartblock(imap->br_startblock))) {
385 error = xfs_iomap_write_allocate(ip, XFS_DATA_FORK, offset,
386 imap);
387 if (!error)
388 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
389 return error;
390 }
391
392#ifdef DEBUG
393 if (type == XFS_IO_UNWRITTEN) {
394 ASSERT(nimaps);
395 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
396 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
397 }
398#endif
399 if (nimaps)
400 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
401 return 0;
402}
403
404STATIC bool
405xfs_imap_valid(
406 struct inode *inode,
407 struct xfs_bmbt_irec *imap,
408 xfs_off_t offset)
409{
410 offset >>= inode->i_blkbits;
411
412 return offset >= imap->br_startoff &&
413 offset < imap->br_startoff + imap->br_blockcount;
414}
415
416STATIC void
417xfs_start_buffer_writeback(
418 struct buffer_head *bh)
419{
420 ASSERT(buffer_mapped(bh));
421 ASSERT(buffer_locked(bh));
422 ASSERT(!buffer_delay(bh));
423 ASSERT(!buffer_unwritten(bh));
424
425 mark_buffer_async_write(bh);
426 set_buffer_uptodate(bh);
427 clear_buffer_dirty(bh);
428}
429
430STATIC void
431xfs_start_page_writeback(
432 struct page *page,
433 int clear_dirty)
434{
435 ASSERT(PageLocked(page));
436 ASSERT(!PageWriteback(page));
437
438 /*
439 * if the page was not fully cleaned, we need to ensure that the higher
440 * layers come back to it correctly. That means we need to keep the page
441 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
442 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
443 * write this page in this writeback sweep will be made.
444 */
445 if (clear_dirty) {
446 clear_page_dirty_for_io(page);
447 set_page_writeback(page);
448 } else
449 set_page_writeback_keepwrite(page);
450
451 unlock_page(page);
452}
453
454static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
455{
456 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
457}
458
459/*
460 * Submit the bio for an ioend. We are passed an ioend with a bio attached to
461 * it, and we submit that bio. The ioend may be used for multiple bio
462 * submissions, so we only want to allocate an append transaction for the ioend
463 * once. In the case of multiple bio submission, each bio will take an IO
464 * reference to the ioend to ensure that the ioend completion is only done once
465 * all bios have been submitted and the ioend is really done.
466 *
467 * If @fail is non-zero, it means that we have a situation where some part of
468 * the submission process has failed after we have marked paged for writeback
469 * and unlocked them. In this situation, we need to fail the bio and ioend
470 * rather than submit it to IO. This typically only happens on a filesystem
471 * shutdown.
472 */
473STATIC int
474xfs_submit_ioend(
475 struct writeback_control *wbc,
476 struct xfs_ioend *ioend,
477 int status)
478{
479 /* Convert CoW extents to regular */
480 if (!status && ioend->io_type == XFS_IO_COW) {
481 status = xfs_reflink_convert_cow(XFS_I(ioend->io_inode),
482 ioend->io_offset, ioend->io_size);
483 }
484
485 /* Reserve log space if we might write beyond the on-disk inode size. */
486 if (!status &&
487 ioend->io_type != XFS_IO_UNWRITTEN &&
488 xfs_ioend_is_append(ioend) &&
489 !ioend->io_append_trans)
490 status = xfs_setfilesize_trans_alloc(ioend);
491
492 ioend->io_bio->bi_private = ioend;
493 ioend->io_bio->bi_end_io = xfs_end_bio;
494 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
495
496 /*
497 * If we are failing the IO now, just mark the ioend with an
498 * error and finish it. This will run IO completion immediately
499 * as there is only one reference to the ioend at this point in
500 * time.
501 */
502 if (status) {
503 ioend->io_bio->bi_error = status;
504 bio_endio(ioend->io_bio);
505 return status;
506 }
507
508 submit_bio(ioend->io_bio);
509 return 0;
510}
511
512static void
513xfs_init_bio_from_bh(
514 struct bio *bio,
515 struct buffer_head *bh)
516{
517 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
518 bio->bi_bdev = bh->b_bdev;
519}
520
521static struct xfs_ioend *
522xfs_alloc_ioend(
523 struct inode *inode,
524 unsigned int type,
525 xfs_off_t offset,
526 struct buffer_head *bh)
527{
528 struct xfs_ioend *ioend;
529 struct bio *bio;
530
531 bio = bio_alloc_bioset(GFP_NOFS, BIO_MAX_PAGES, xfs_ioend_bioset);
532 xfs_init_bio_from_bh(bio, bh);
533
534 ioend = container_of(bio, struct xfs_ioend, io_inline_bio);
535 INIT_LIST_HEAD(&ioend->io_list);
536 ioend->io_type = type;
537 ioend->io_inode = inode;
538 ioend->io_size = 0;
539 ioend->io_offset = offset;
540 INIT_WORK(&ioend->io_work, xfs_end_io);
541 ioend->io_append_trans = NULL;
542 ioend->io_bio = bio;
543 return ioend;
544}
545
546/*
547 * Allocate a new bio, and chain the old bio to the new one.
548 *
549 * Note that we have to do perform the chaining in this unintuitive order
550 * so that the bi_private linkage is set up in the right direction for the
551 * traversal in xfs_destroy_ioend().
552 */
553static void
554xfs_chain_bio(
555 struct xfs_ioend *ioend,
556 struct writeback_control *wbc,
557 struct buffer_head *bh)
558{
559 struct bio *new;
560
561 new = bio_alloc(GFP_NOFS, BIO_MAX_PAGES);
562 xfs_init_bio_from_bh(new, bh);
563
564 bio_chain(ioend->io_bio, new);
565 bio_get(ioend->io_bio); /* for xfs_destroy_ioend */
566 ioend->io_bio->bi_opf = REQ_OP_WRITE | wbc_to_write_flags(wbc);
567 submit_bio(ioend->io_bio);
568 ioend->io_bio = new;
569}
570
571/*
572 * Test to see if we've been building up a completion structure for
573 * earlier buffers -- if so, we try to append to this ioend if we
574 * can, otherwise we finish off any current ioend and start another.
575 * Return the ioend we finished off so that the caller can submit it
576 * once it has finished processing the dirty page.
577 */
578STATIC void
579xfs_add_to_ioend(
580 struct inode *inode,
581 struct buffer_head *bh,
582 xfs_off_t offset,
583 struct xfs_writepage_ctx *wpc,
584 struct writeback_control *wbc,
585 struct list_head *iolist)
586{
587 if (!wpc->ioend || wpc->io_type != wpc->ioend->io_type ||
588 bh->b_blocknr != wpc->last_block + 1 ||
589 offset != wpc->ioend->io_offset + wpc->ioend->io_size) {
590 if (wpc->ioend)
591 list_add(&wpc->ioend->io_list, iolist);
592 wpc->ioend = xfs_alloc_ioend(inode, wpc->io_type, offset, bh);
593 }
594
595 /*
596 * If the buffer doesn't fit into the bio we need to allocate a new
597 * one. This shouldn't happen more than once for a given buffer.
598 */
599 while (xfs_bio_add_buffer(wpc->ioend->io_bio, bh) != bh->b_size)
600 xfs_chain_bio(wpc->ioend, wbc, bh);
601
602 wpc->ioend->io_size += bh->b_size;
603 wpc->last_block = bh->b_blocknr;
604 xfs_start_buffer_writeback(bh);
605}
606
607STATIC void
608xfs_map_buffer(
609 struct inode *inode,
610 struct buffer_head *bh,
611 struct xfs_bmbt_irec *imap,
612 xfs_off_t offset)
613{
614 sector_t bn;
615 struct xfs_mount *m = XFS_I(inode)->i_mount;
616 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
617 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
618
619 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
620 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
621
622 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
623 ((offset - iomap_offset) >> inode->i_blkbits);
624
625 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
626
627 bh->b_blocknr = bn;
628 set_buffer_mapped(bh);
629}
630
631STATIC void
632xfs_map_at_offset(
633 struct inode *inode,
634 struct buffer_head *bh,
635 struct xfs_bmbt_irec *imap,
636 xfs_off_t offset)
637{
638 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
639 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
640
641 xfs_map_buffer(inode, bh, imap, offset);
642 set_buffer_mapped(bh);
643 clear_buffer_delay(bh);
644 clear_buffer_unwritten(bh);
645}
646
647/*
648 * Test if a given page contains at least one buffer of a given @type.
649 * If @check_all_buffers is true, then we walk all the buffers in the page to
650 * try to find one of the type passed in. If it is not set, then the caller only
651 * needs to check the first buffer on the page for a match.
652 */
653STATIC bool
654xfs_check_page_type(
655 struct page *page,
656 unsigned int type,
657 bool check_all_buffers)
658{
659 struct buffer_head *bh;
660 struct buffer_head *head;
661
662 if (PageWriteback(page))
663 return false;
664 if (!page->mapping)
665 return false;
666 if (!page_has_buffers(page))
667 return false;
668
669 bh = head = page_buffers(page);
670 do {
671 if (buffer_unwritten(bh)) {
672 if (type == XFS_IO_UNWRITTEN)
673 return true;
674 } else if (buffer_delay(bh)) {
675 if (type == XFS_IO_DELALLOC)
676 return true;
677 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
678 if (type == XFS_IO_OVERWRITE)
679 return true;
680 }
681
682 /* If we are only checking the first buffer, we are done now. */
683 if (!check_all_buffers)
684 break;
685 } while ((bh = bh->b_this_page) != head);
686
687 return false;
688}
689
690STATIC void
691xfs_vm_invalidatepage(
692 struct page *page,
693 unsigned int offset,
694 unsigned int length)
695{
696 trace_xfs_invalidatepage(page->mapping->host, page, offset,
697 length);
698 block_invalidatepage(page, offset, length);
699}
700
701/*
702 * If the page has delalloc buffers on it, we need to punch them out before we
703 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
704 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
705 * is done on that same region - the delalloc extent is returned when none is
706 * supposed to be there.
707 *
708 * We prevent this by truncating away the delalloc regions on the page before
709 * invalidating it. Because they are delalloc, we can do this without needing a
710 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
711 * truncation without a transaction as there is no space left for block
712 * reservation (typically why we see a ENOSPC in writeback).
713 *
714 * This is not a performance critical path, so for now just do the punching a
715 * buffer head at a time.
716 */
717STATIC void
718xfs_aops_discard_page(
719 struct page *page)
720{
721 struct inode *inode = page->mapping->host;
722 struct xfs_inode *ip = XFS_I(inode);
723 struct buffer_head *bh, *head;
724 loff_t offset = page_offset(page);
725
726 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
727 goto out_invalidate;
728
729 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
730 goto out_invalidate;
731
732 xfs_alert(ip->i_mount,
733 "page discard on page %p, inode 0x%llx, offset %llu.",
734 page, ip->i_ino, offset);
735
736 xfs_ilock(ip, XFS_ILOCK_EXCL);
737 bh = head = page_buffers(page);
738 do {
739 int error;
740 xfs_fileoff_t start_fsb;
741
742 if (!buffer_delay(bh))
743 goto next_buffer;
744
745 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
746 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
747 if (error) {
748 /* something screwed, just bail */
749 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
750 xfs_alert(ip->i_mount,
751 "page discard unable to remove delalloc mapping.");
752 }
753 break;
754 }
755next_buffer:
756 offset += 1 << inode->i_blkbits;
757
758 } while ((bh = bh->b_this_page) != head);
759
760 xfs_iunlock(ip, XFS_ILOCK_EXCL);
761out_invalidate:
762 xfs_vm_invalidatepage(page, 0, PAGE_SIZE);
763 return;
764}
765
766static int
767xfs_map_cow(
768 struct xfs_writepage_ctx *wpc,
769 struct inode *inode,
770 loff_t offset,
771 unsigned int *new_type)
772{
773 struct xfs_inode *ip = XFS_I(inode);
774 struct xfs_bmbt_irec imap;
775 bool is_cow = false;
776 int error;
777
778 /*
779 * If we already have a valid COW mapping keep using it.
780 */
781 if (wpc->io_type == XFS_IO_COW) {
782 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap, offset);
783 if (wpc->imap_valid) {
784 *new_type = XFS_IO_COW;
785 return 0;
786 }
787 }
788
789 /*
790 * Else we need to check if there is a COW mapping at this offset.
791 */
792 xfs_ilock(ip, XFS_ILOCK_SHARED);
793 is_cow = xfs_reflink_find_cow_mapping(ip, offset, &imap);
794 xfs_iunlock(ip, XFS_ILOCK_SHARED);
795
796 if (!is_cow)
797 return 0;
798
799 /*
800 * And if the COW mapping has a delayed extent here we need to
801 * allocate real space for it now.
802 */
803 if (isnullstartblock(imap.br_startblock)) {
804 error = xfs_iomap_write_allocate(ip, XFS_COW_FORK, offset,
805 &imap);
806 if (error)
807 return error;
808 }
809
810 wpc->io_type = *new_type = XFS_IO_COW;
811 wpc->imap_valid = true;
812 wpc->imap = imap;
813 return 0;
814}
815
816/*
817 * We implement an immediate ioend submission policy here to avoid needing to
818 * chain multiple ioends and hence nest mempool allocations which can violate
819 * forward progress guarantees we need to provide. The current ioend we are
820 * adding buffers to is cached on the writepage context, and if the new buffer
821 * does not append to the cached ioend it will create a new ioend and cache that
822 * instead.
823 *
824 * If a new ioend is created and cached, the old ioend is returned and queued
825 * locally for submission once the entire page is processed or an error has been
826 * detected. While ioends are submitted immediately after they are completed,
827 * batching optimisations are provided by higher level block plugging.
828 *
829 * At the end of a writeback pass, there will be a cached ioend remaining on the
830 * writepage context that the caller will need to submit.
831 */
832static int
833xfs_writepage_map(
834 struct xfs_writepage_ctx *wpc,
835 struct writeback_control *wbc,
836 struct inode *inode,
837 struct page *page,
838 loff_t offset,
839 __uint64_t end_offset)
840{
841 LIST_HEAD(submit_list);
842 struct xfs_ioend *ioend, *next;
843 struct buffer_head *bh, *head;
844 ssize_t len = 1 << inode->i_blkbits;
845 int error = 0;
846 int count = 0;
847 int uptodate = 1;
848 unsigned int new_type;
849
850 bh = head = page_buffers(page);
851 offset = page_offset(page);
852 do {
853 if (offset >= end_offset)
854 break;
855 if (!buffer_uptodate(bh))
856 uptodate = 0;
857
858 /*
859 * set_page_dirty dirties all buffers in a page, independent
860 * of their state. The dirty state however is entirely
861 * meaningless for holes (!mapped && uptodate), so skip
862 * buffers covering holes here.
863 */
864 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
865 wpc->imap_valid = false;
866 continue;
867 }
868
869 if (buffer_unwritten(bh))
870 new_type = XFS_IO_UNWRITTEN;
871 else if (buffer_delay(bh))
872 new_type = XFS_IO_DELALLOC;
873 else if (buffer_uptodate(bh))
874 new_type = XFS_IO_OVERWRITE;
875 else {
876 if (PageUptodate(page))
877 ASSERT(buffer_mapped(bh));
878 /*
879 * This buffer is not uptodate and will not be
880 * written to disk. Ensure that we will put any
881 * subsequent writeable buffers into a new
882 * ioend.
883 */
884 wpc->imap_valid = false;
885 continue;
886 }
887
888 if (xfs_is_reflink_inode(XFS_I(inode))) {
889 error = xfs_map_cow(wpc, inode, offset, &new_type);
890 if (error)
891 goto out;
892 }
893
894 if (wpc->io_type != new_type) {
895 wpc->io_type = new_type;
896 wpc->imap_valid = false;
897 }
898
899 if (wpc->imap_valid)
900 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
901 offset);
902 if (!wpc->imap_valid) {
903 error = xfs_map_blocks(inode, offset, &wpc->imap,
904 wpc->io_type);
905 if (error)
906 goto out;
907 wpc->imap_valid = xfs_imap_valid(inode, &wpc->imap,
908 offset);
909 }
910 if (wpc->imap_valid) {
911 lock_buffer(bh);
912 if (wpc->io_type != XFS_IO_OVERWRITE)
913 xfs_map_at_offset(inode, bh, &wpc->imap, offset);
914 xfs_add_to_ioend(inode, bh, offset, wpc, wbc, &submit_list);
915 count++;
916 }
917
918 } while (offset += len, ((bh = bh->b_this_page) != head));
919
920 if (uptodate && bh == head)
921 SetPageUptodate(page);
922
923 ASSERT(wpc->ioend || list_empty(&submit_list));
924
925out:
926 /*
927 * On error, we have to fail the ioend here because we have locked
928 * buffers in the ioend. If we don't do this, we'll deadlock
929 * invalidating the page as that tries to lock the buffers on the page.
930 * Also, because we may have set pages under writeback, we have to make
931 * sure we run IO completion to mark the error state of the IO
932 * appropriately, so we can't cancel the ioend directly here. That means
933 * we have to mark this page as under writeback if we included any
934 * buffers from it in the ioend chain so that completion treats it
935 * correctly.
936 *
937 * If we didn't include the page in the ioend, the on error we can
938 * simply discard and unlock it as there are no other users of the page
939 * or it's buffers right now. The caller will still need to trigger
940 * submission of outstanding ioends on the writepage context so they are
941 * treated correctly on error.
942 */
943 if (count) {
944 xfs_start_page_writeback(page, !error);
945
946 /*
947 * Preserve the original error if there was one, otherwise catch
948 * submission errors here and propagate into subsequent ioend
949 * submissions.
950 */
951 list_for_each_entry_safe(ioend, next, &submit_list, io_list) {
952 int error2;
953
954 list_del_init(&ioend->io_list);
955 error2 = xfs_submit_ioend(wbc, ioend, error);
956 if (error2 && !error)
957 error = error2;
958 }
959 } else if (error) {
960 xfs_aops_discard_page(page);
961 ClearPageUptodate(page);
962 unlock_page(page);
963 } else {
964 /*
965 * We can end up here with no error and nothing to write if we
966 * race with a partial page truncate on a sub-page block sized
967 * filesystem. In that case we need to mark the page clean.
968 */
969 xfs_start_page_writeback(page, 1);
970 end_page_writeback(page);
971 }
972
973 mapping_set_error(page->mapping, error);
974 return error;
975}
976
977/*
978 * Write out a dirty page.
979 *
980 * For delalloc space on the page we need to allocate space and flush it.
981 * For unwritten space on the page we need to start the conversion to
982 * regular allocated space.
983 * For any other dirty buffer heads on the page we should flush them.
984 */
985STATIC int
986xfs_do_writepage(
987 struct page *page,
988 struct writeback_control *wbc,
989 void *data)
990{
991 struct xfs_writepage_ctx *wpc = data;
992 struct inode *inode = page->mapping->host;
993 loff_t offset;
994 __uint64_t end_offset;
995 pgoff_t end_index;
996
997 trace_xfs_writepage(inode, page, 0, 0);
998
999 ASSERT(page_has_buffers(page));
1000
1001 /*
1002 * Refuse to write the page out if we are called from reclaim context.
1003 *
1004 * This avoids stack overflows when called from deeply used stacks in
1005 * random callers for direct reclaim or memcg reclaim. We explicitly
1006 * allow reclaim from kswapd as the stack usage there is relatively low.
1007 *
1008 * This should never happen except in the case of a VM regression so
1009 * warn about it.
1010 */
1011 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
1012 PF_MEMALLOC))
1013 goto redirty;
1014
1015 /*
1016 * Given that we do not allow direct reclaim to call us, we should
1017 * never be called while in a filesystem transaction.
1018 */
1019 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
1020 goto redirty;
1021
1022 /*
1023 * Is this page beyond the end of the file?
1024 *
1025 * The page index is less than the end_index, adjust the end_offset
1026 * to the highest offset that this page should represent.
1027 * -----------------------------------------------------
1028 * | file mapping | <EOF> |
1029 * -----------------------------------------------------
1030 * | Page ... | Page N-2 | Page N-1 | Page N | |
1031 * ^--------------------------------^----------|--------
1032 * | desired writeback range | see else |
1033 * ---------------------------------^------------------|
1034 */
1035 offset = i_size_read(inode);
1036 end_index = offset >> PAGE_SHIFT;
1037 if (page->index < end_index)
1038 end_offset = (xfs_off_t)(page->index + 1) << PAGE_SHIFT;
1039 else {
1040 /*
1041 * Check whether the page to write out is beyond or straddles
1042 * i_size or not.
1043 * -------------------------------------------------------
1044 * | file mapping | <EOF> |
1045 * -------------------------------------------------------
1046 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1047 * ^--------------------------------^-----------|---------
1048 * | | Straddles |
1049 * ---------------------------------^-----------|--------|
1050 */
1051 unsigned offset_into_page = offset & (PAGE_SIZE - 1);
1052
1053 /*
1054 * Skip the page if it is fully outside i_size, e.g. due to a
1055 * truncate operation that is in progress. We must redirty the
1056 * page so that reclaim stops reclaiming it. Otherwise
1057 * xfs_vm_releasepage() is called on it and gets confused.
1058 *
1059 * Note that the end_index is unsigned long, it would overflow
1060 * if the given offset is greater than 16TB on 32-bit system
1061 * and if we do check the page is fully outside i_size or not
1062 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1063 * will be evaluated to 0. Hence this page will be redirtied
1064 * and be written out repeatedly which would result in an
1065 * infinite loop, the user program that perform this operation
1066 * will hang. Instead, we can verify this situation by checking
1067 * if the page to write is totally beyond the i_size or if it's
1068 * offset is just equal to the EOF.
1069 */
1070 if (page->index > end_index ||
1071 (page->index == end_index && offset_into_page == 0))
1072 goto redirty;
1073
1074 /*
1075 * The page straddles i_size. It must be zeroed out on each
1076 * and every writepage invocation because it may be mmapped.
1077 * "A file is mapped in multiples of the page size. For a file
1078 * that is not a multiple of the page size, the remaining
1079 * memory is zeroed when mapped, and writes to that region are
1080 * not written out to the file."
1081 */
1082 zero_user_segment(page, offset_into_page, PAGE_SIZE);
1083
1084 /* Adjust the end_offset to the end of file */
1085 end_offset = offset;
1086 }
1087
1088 return xfs_writepage_map(wpc, wbc, inode, page, offset, end_offset);
1089
1090redirty:
1091 redirty_page_for_writepage(wbc, page);
1092 unlock_page(page);
1093 return 0;
1094}
1095
1096STATIC int
1097xfs_vm_writepage(
1098 struct page *page,
1099 struct writeback_control *wbc)
1100{
1101 struct xfs_writepage_ctx wpc = {
1102 .io_type = XFS_IO_INVALID,
1103 };
1104 int ret;
1105
1106 ret = xfs_do_writepage(page, wbc, &wpc);
1107 if (wpc.ioend)
1108 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1109 return ret;
1110}
1111
1112STATIC int
1113xfs_vm_writepages(
1114 struct address_space *mapping,
1115 struct writeback_control *wbc)
1116{
1117 struct xfs_writepage_ctx wpc = {
1118 .io_type = XFS_IO_INVALID,
1119 };
1120 int ret;
1121
1122 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
1123 if (dax_mapping(mapping))
1124 return dax_writeback_mapping_range(mapping,
1125 xfs_find_bdev_for_inode(mapping->host), wbc);
1126
1127 ret = write_cache_pages(mapping, wbc, xfs_do_writepage, &wpc);
1128 if (wpc.ioend)
1129 ret = xfs_submit_ioend(wbc, wpc.ioend, ret);
1130 return ret;
1131}
1132
1133/*
1134 * Called to move a page into cleanable state - and from there
1135 * to be released. The page should already be clean. We always
1136 * have buffer heads in this call.
1137 *
1138 * Returns 1 if the page is ok to release, 0 otherwise.
1139 */
1140STATIC int
1141xfs_vm_releasepage(
1142 struct page *page,
1143 gfp_t gfp_mask)
1144{
1145 int delalloc, unwritten;
1146
1147 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1148
1149 /*
1150 * mm accommodates an old ext3 case where clean pages might not have had
1151 * the dirty bit cleared. Thus, it can send actual dirty pages to
1152 * ->releasepage() via shrink_active_list(). Conversely,
1153 * block_invalidatepage() can send pages that are still marked dirty
1154 * but otherwise have invalidated buffers.
1155 *
1156 * We want to release the latter to avoid unnecessary buildup of the
1157 * LRU, skip the former and warn if we've left any lingering
1158 * delalloc/unwritten buffers on clean pages. Skip pages with delalloc
1159 * or unwritten buffers and warn if the page is not dirty. Otherwise
1160 * try to release the buffers.
1161 */
1162 xfs_count_page_state(page, &delalloc, &unwritten);
1163
1164 if (delalloc) {
1165 WARN_ON_ONCE(!PageDirty(page));
1166 return 0;
1167 }
1168 if (unwritten) {
1169 WARN_ON_ONCE(!PageDirty(page));
1170 return 0;
1171 }
1172
1173 return try_to_free_buffers(page);
1174}
1175
1176/*
1177 * If this is O_DIRECT or the mpage code calling tell them how large the mapping
1178 * is, so that we can avoid repeated get_blocks calls.
1179 *
1180 * If the mapping spans EOF, then we have to break the mapping up as the mapping
1181 * for blocks beyond EOF must be marked new so that sub block regions can be
1182 * correctly zeroed. We can't do this for mappings within EOF unless the mapping
1183 * was just allocated or is unwritten, otherwise the callers would overwrite
1184 * existing data with zeros. Hence we have to split the mapping into a range up
1185 * to and including EOF, and a second mapping for beyond EOF.
1186 */
1187static void
1188xfs_map_trim_size(
1189 struct inode *inode,
1190 sector_t iblock,
1191 struct buffer_head *bh_result,
1192 struct xfs_bmbt_irec *imap,
1193 xfs_off_t offset,
1194 ssize_t size)
1195{
1196 xfs_off_t mapping_size;
1197
1198 mapping_size = imap->br_startoff + imap->br_blockcount - iblock;
1199 mapping_size <<= inode->i_blkbits;
1200
1201 ASSERT(mapping_size > 0);
1202 if (mapping_size > size)
1203 mapping_size = size;
1204 if (offset < i_size_read(inode) &&
1205 offset + mapping_size >= i_size_read(inode)) {
1206 /* limit mapping to block that spans EOF */
1207 mapping_size = roundup_64(i_size_read(inode) - offset,
1208 1 << inode->i_blkbits);
1209 }
1210 if (mapping_size > LONG_MAX)
1211 mapping_size = LONG_MAX;
1212
1213 bh_result->b_size = mapping_size;
1214}
1215
1216static int
1217xfs_get_blocks(
1218 struct inode *inode,
1219 sector_t iblock,
1220 struct buffer_head *bh_result,
1221 int create)
1222{
1223 struct xfs_inode *ip = XFS_I(inode);
1224 struct xfs_mount *mp = ip->i_mount;
1225 xfs_fileoff_t offset_fsb, end_fsb;
1226 int error = 0;
1227 int lockmode = 0;
1228 struct xfs_bmbt_irec imap;
1229 int nimaps = 1;
1230 xfs_off_t offset;
1231 ssize_t size;
1232
1233 BUG_ON(create);
1234
1235 if (XFS_FORCED_SHUTDOWN(mp))
1236 return -EIO;
1237
1238 offset = (xfs_off_t)iblock << inode->i_blkbits;
1239 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1240 size = bh_result->b_size;
1241
1242 if (offset >= i_size_read(inode))
1243 return 0;
1244
1245 /*
1246 * Direct I/O is usually done on preallocated files, so try getting
1247 * a block mapping without an exclusive lock first.
1248 */
1249 lockmode = xfs_ilock_data_map_shared(ip);
1250
1251 ASSERT(offset <= mp->m_super->s_maxbytes);
1252 if (offset + size > mp->m_super->s_maxbytes)
1253 size = mp->m_super->s_maxbytes - offset;
1254 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1255 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1256
1257 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1258 &imap, &nimaps, XFS_BMAPI_ENTIRE);
1259 if (error)
1260 goto out_unlock;
1261
1262 if (nimaps) {
1263 trace_xfs_get_blocks_found(ip, offset, size,
1264 ISUNWRITTEN(&imap) ? XFS_IO_UNWRITTEN
1265 : XFS_IO_OVERWRITE, &imap);
1266 xfs_iunlock(ip, lockmode);
1267 } else {
1268 trace_xfs_get_blocks_notfound(ip, offset, size);
1269 goto out_unlock;
1270 }
1271
1272 /* trim mapping down to size requested */
1273 xfs_map_trim_size(inode, iblock, bh_result, &imap, offset, size);
1274
1275 /*
1276 * For unwritten extents do not report a disk address in the buffered
1277 * read case (treat as if we're reading into a hole).
1278 */
1279 if (imap.br_startblock != HOLESTARTBLOCK &&
1280 imap.br_startblock != DELAYSTARTBLOCK &&
1281 !ISUNWRITTEN(&imap))
1282 xfs_map_buffer(inode, bh_result, &imap, offset);
1283
1284 /*
1285 * If this is a realtime file, data may be on a different device.
1286 * to that pointed to from the buffer_head b_bdev currently.
1287 */
1288 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
1289 return 0;
1290
1291out_unlock:
1292 xfs_iunlock(ip, lockmode);
1293 return error;
1294}
1295
1296STATIC ssize_t
1297xfs_vm_direct_IO(
1298 struct kiocb *iocb,
1299 struct iov_iter *iter)
1300{
1301 /*
1302 * We just need the method present so that open/fcntl allow direct I/O.
1303 */
1304 return -EINVAL;
1305}
1306
1307STATIC sector_t
1308xfs_vm_bmap(
1309 struct address_space *mapping,
1310 sector_t block)
1311{
1312 struct inode *inode = (struct inode *)mapping->host;
1313 struct xfs_inode *ip = XFS_I(inode);
1314
1315 trace_xfs_vm_bmap(XFS_I(inode));
1316
1317 /*
1318 * The swap code (ab-)uses ->bmap to get a block mapping and then
1319 * bypasseѕ the file system for actual I/O. We really can't allow
1320 * that on reflinks inodes, so we have to skip out here. And yes,
1321 * 0 is the magic code for a bmap error..
1322 */
1323 if (xfs_is_reflink_inode(ip))
1324 return 0;
1325
1326 filemap_write_and_wait(mapping);
1327 return generic_block_bmap(mapping, block, xfs_get_blocks);
1328}
1329
1330STATIC int
1331xfs_vm_readpage(
1332 struct file *unused,
1333 struct page *page)
1334{
1335 trace_xfs_vm_readpage(page->mapping->host, 1);
1336 return mpage_readpage(page, xfs_get_blocks);
1337}
1338
1339STATIC int
1340xfs_vm_readpages(
1341 struct file *unused,
1342 struct address_space *mapping,
1343 struct list_head *pages,
1344 unsigned nr_pages)
1345{
1346 trace_xfs_vm_readpages(mapping->host, nr_pages);
1347 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
1348}
1349
1350/*
1351 * This is basically a copy of __set_page_dirty_buffers() with one
1352 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1353 * dirty, we'll never be able to clean them because we don't write buffers
1354 * beyond EOF, and that means we can't invalidate pages that span EOF
1355 * that have been marked dirty. Further, the dirty state can leak into
1356 * the file interior if the file is extended, resulting in all sorts of
1357 * bad things happening as the state does not match the underlying data.
1358 *
1359 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1360 * this only exist because of bufferheads and how the generic code manages them.
1361 */
1362STATIC int
1363xfs_vm_set_page_dirty(
1364 struct page *page)
1365{
1366 struct address_space *mapping = page->mapping;
1367 struct inode *inode = mapping->host;
1368 loff_t end_offset;
1369 loff_t offset;
1370 int newly_dirty;
1371
1372 if (unlikely(!mapping))
1373 return !TestSetPageDirty(page);
1374
1375 end_offset = i_size_read(inode);
1376 offset = page_offset(page);
1377
1378 spin_lock(&mapping->private_lock);
1379 if (page_has_buffers(page)) {
1380 struct buffer_head *head = page_buffers(page);
1381 struct buffer_head *bh = head;
1382
1383 do {
1384 if (offset < end_offset)
1385 set_buffer_dirty(bh);
1386 bh = bh->b_this_page;
1387 offset += 1 << inode->i_blkbits;
1388 } while (bh != head);
1389 }
1390 /*
1391 * Lock out page->mem_cgroup migration to keep PageDirty
1392 * synchronized with per-memcg dirty page counters.
1393 */
1394 lock_page_memcg(page);
1395 newly_dirty = !TestSetPageDirty(page);
1396 spin_unlock(&mapping->private_lock);
1397
1398 if (newly_dirty) {
1399 /* sigh - __set_page_dirty() is static, so copy it here, too */
1400 unsigned long flags;
1401
1402 spin_lock_irqsave(&mapping->tree_lock, flags);
1403 if (page->mapping) { /* Race with truncate? */
1404 WARN_ON_ONCE(!PageUptodate(page));
1405 account_page_dirtied(page, mapping);
1406 radix_tree_tag_set(&mapping->page_tree,
1407 page_index(page), PAGECACHE_TAG_DIRTY);
1408 }
1409 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1410 }
1411 unlock_page_memcg(page);
1412 if (newly_dirty)
1413 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1414 return newly_dirty;
1415}
1416
1417const struct address_space_operations xfs_address_space_operations = {
1418 .readpage = xfs_vm_readpage,
1419 .readpages = xfs_vm_readpages,
1420 .writepage = xfs_vm_writepage,
1421 .writepages = xfs_vm_writepages,
1422 .set_page_dirty = xfs_vm_set_page_dirty,
1423 .releasepage = xfs_vm_releasepage,
1424 .invalidatepage = xfs_vm_invalidatepage,
1425 .bmap = xfs_vm_bmap,
1426 .direct_IO = xfs_vm_direct_IO,
1427 .migratepage = buffer_migrate_page,
1428 .is_partially_uptodate = block_is_partially_uptodate,
1429 .error_remove_page = generic_error_remove_page,
1430};