Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/f2fs/node.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8#include <linux/fs.h>
9#include <linux/f2fs_fs.h>
10#include <linux/mpage.h>
11#include <linux/sched/mm.h>
12#include <linux/blkdev.h>
13#include <linux/pagevec.h>
14#include <linux/swap.h>
15
16#include "f2fs.h"
17#include "node.h"
18#include "segment.h"
19#include "xattr.h"
20#include "iostat.h"
21#include <trace/events/f2fs.h>
22
23#define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24
25static struct kmem_cache *nat_entry_slab;
26static struct kmem_cache *free_nid_slab;
27static struct kmem_cache *nat_entry_set_slab;
28static struct kmem_cache *fsync_node_entry_slab;
29
30/*
31 * Check whether the given nid is within node id range.
32 */
33int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34{
35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 set_sbi_flag(sbi, SBI_NEED_FSCK);
37 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 __func__, nid);
39 f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
40 return -EFSCORRUPTED;
41 }
42 return 0;
43}
44
45bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46{
47 struct f2fs_nm_info *nm_i = NM_I(sbi);
48 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
49 struct sysinfo val;
50 unsigned long avail_ram;
51 unsigned long mem_size = 0;
52 bool res = false;
53
54 if (!nm_i)
55 return true;
56
57 si_meminfo(&val);
58
59 /* only uses low memory */
60 avail_ram = val.totalram - val.totalhigh;
61
62 /*
63 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
64 */
65 if (type == FREE_NIDS) {
66 mem_size = (nm_i->nid_cnt[FREE_NID] *
67 sizeof(struct free_nid)) >> PAGE_SHIFT;
68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 } else if (type == NAT_ENTRIES) {
70 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
71 sizeof(struct nat_entry)) >> PAGE_SHIFT;
72 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
73 if (excess_cached_nats(sbi))
74 res = false;
75 } else if (type == DIRTY_DENTS) {
76 if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 return false;
78 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
79 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
80 } else if (type == INO_ENTRIES) {
81 int i;
82
83 for (i = 0; i < MAX_INO_ENTRY; i++)
84 mem_size += sbi->im[i].ino_num *
85 sizeof(struct ino_entry);
86 mem_size >>= PAGE_SHIFT;
87 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 } else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
89 enum extent_type etype = type == READ_EXTENT_CACHE ?
90 EX_READ : EX_BLOCK_AGE;
91 struct extent_tree_info *eti = &sbi->extent_tree[etype];
92
93 mem_size = (atomic_read(&eti->total_ext_tree) *
94 sizeof(struct extent_tree) +
95 atomic_read(&eti->total_ext_node) *
96 sizeof(struct extent_node)) >> PAGE_SHIFT;
97 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
98 } else if (type == DISCARD_CACHE) {
99 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
100 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
101 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
102 } else if (type == COMPRESS_PAGE) {
103#ifdef CONFIG_F2FS_FS_COMPRESSION
104 unsigned long free_ram = val.freeram;
105
106 /*
107 * free memory is lower than watermark or cached page count
108 * exceed threshold, deny caching compress page.
109 */
110 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
111 (COMPRESS_MAPPING(sbi)->nrpages <
112 free_ram * sbi->compress_percent / 100);
113#else
114 res = false;
115#endif
116 } else {
117 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
118 return true;
119 }
120 return res;
121}
122
123static void clear_node_page_dirty(struct page *page)
124{
125 if (PageDirty(page)) {
126 f2fs_clear_page_cache_dirty_tag(page);
127 clear_page_dirty_for_io(page);
128 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
129 }
130 ClearPageUptodate(page);
131}
132
133static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134{
135 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
136}
137
138static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
139{
140 struct page *src_page;
141 struct page *dst_page;
142 pgoff_t dst_off;
143 void *src_addr;
144 void *dst_addr;
145 struct f2fs_nm_info *nm_i = NM_I(sbi);
146
147 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
148
149 /* get current nat block page with lock */
150 src_page = get_current_nat_page(sbi, nid);
151 if (IS_ERR(src_page))
152 return src_page;
153 dst_page = f2fs_grab_meta_page(sbi, dst_off);
154 f2fs_bug_on(sbi, PageDirty(src_page));
155
156 src_addr = page_address(src_page);
157 dst_addr = page_address(dst_page);
158 memcpy(dst_addr, src_addr, PAGE_SIZE);
159 set_page_dirty(dst_page);
160 f2fs_put_page(src_page, 1);
161
162 set_to_next_nat(nm_i, nid);
163
164 return dst_page;
165}
166
167static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
168 nid_t nid, bool no_fail)
169{
170 struct nat_entry *new;
171
172 new = f2fs_kmem_cache_alloc(nat_entry_slab,
173 GFP_F2FS_ZERO, no_fail, sbi);
174 if (new) {
175 nat_set_nid(new, nid);
176 nat_reset_flag(new);
177 }
178 return new;
179}
180
181static void __free_nat_entry(struct nat_entry *e)
182{
183 kmem_cache_free(nat_entry_slab, e);
184}
185
186/* must be locked by nat_tree_lock */
187static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
188 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
189{
190 if (no_fail)
191 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
192 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
193 return NULL;
194
195 if (raw_ne)
196 node_info_from_raw_nat(&ne->ni, raw_ne);
197
198 spin_lock(&nm_i->nat_list_lock);
199 list_add_tail(&ne->list, &nm_i->nat_entries);
200 spin_unlock(&nm_i->nat_list_lock);
201
202 nm_i->nat_cnt[TOTAL_NAT]++;
203 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
204 return ne;
205}
206
207static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
208{
209 struct nat_entry *ne;
210
211 ne = radix_tree_lookup(&nm_i->nat_root, n);
212
213 /* for recent accessed nat entry, move it to tail of lru list */
214 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
215 spin_lock(&nm_i->nat_list_lock);
216 if (!list_empty(&ne->list))
217 list_move_tail(&ne->list, &nm_i->nat_entries);
218 spin_unlock(&nm_i->nat_list_lock);
219 }
220
221 return ne;
222}
223
224static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
225 nid_t start, unsigned int nr, struct nat_entry **ep)
226{
227 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
228}
229
230static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
231{
232 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
233 nm_i->nat_cnt[TOTAL_NAT]--;
234 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
235 __free_nat_entry(e);
236}
237
238static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
239 struct nat_entry *ne)
240{
241 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
242 struct nat_entry_set *head;
243
244 head = radix_tree_lookup(&nm_i->nat_set_root, set);
245 if (!head) {
246 head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
247 GFP_NOFS, true, NULL);
248
249 INIT_LIST_HEAD(&head->entry_list);
250 INIT_LIST_HEAD(&head->set_list);
251 head->set = set;
252 head->entry_cnt = 0;
253 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
254 }
255 return head;
256}
257
258static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
259 struct nat_entry *ne)
260{
261 struct nat_entry_set *head;
262 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
263
264 if (!new_ne)
265 head = __grab_nat_entry_set(nm_i, ne);
266
267 /*
268 * update entry_cnt in below condition:
269 * 1. update NEW_ADDR to valid block address;
270 * 2. update old block address to new one;
271 */
272 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
273 !get_nat_flag(ne, IS_DIRTY)))
274 head->entry_cnt++;
275
276 set_nat_flag(ne, IS_PREALLOC, new_ne);
277
278 if (get_nat_flag(ne, IS_DIRTY))
279 goto refresh_list;
280
281 nm_i->nat_cnt[DIRTY_NAT]++;
282 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
283 set_nat_flag(ne, IS_DIRTY, true);
284refresh_list:
285 spin_lock(&nm_i->nat_list_lock);
286 if (new_ne)
287 list_del_init(&ne->list);
288 else
289 list_move_tail(&ne->list, &head->entry_list);
290 spin_unlock(&nm_i->nat_list_lock);
291}
292
293static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
294 struct nat_entry_set *set, struct nat_entry *ne)
295{
296 spin_lock(&nm_i->nat_list_lock);
297 list_move_tail(&ne->list, &nm_i->nat_entries);
298 spin_unlock(&nm_i->nat_list_lock);
299
300 set_nat_flag(ne, IS_DIRTY, false);
301 set->entry_cnt--;
302 nm_i->nat_cnt[DIRTY_NAT]--;
303 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
304}
305
306static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
307 nid_t start, unsigned int nr, struct nat_entry_set **ep)
308{
309 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
310 start, nr);
311}
312
313bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
314{
315 return NODE_MAPPING(sbi) == page->mapping &&
316 IS_DNODE(page) && is_cold_node(page);
317}
318
319void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
320{
321 spin_lock_init(&sbi->fsync_node_lock);
322 INIT_LIST_HEAD(&sbi->fsync_node_list);
323 sbi->fsync_seg_id = 0;
324 sbi->fsync_node_num = 0;
325}
326
327static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
328 struct page *page)
329{
330 struct fsync_node_entry *fn;
331 unsigned long flags;
332 unsigned int seq_id;
333
334 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
335 GFP_NOFS, true, NULL);
336
337 get_page(page);
338 fn->page = page;
339 INIT_LIST_HEAD(&fn->list);
340
341 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
342 list_add_tail(&fn->list, &sbi->fsync_node_list);
343 fn->seq_id = sbi->fsync_seg_id++;
344 seq_id = fn->seq_id;
345 sbi->fsync_node_num++;
346 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347
348 return seq_id;
349}
350
351void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
352{
353 struct fsync_node_entry *fn;
354 unsigned long flags;
355
356 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
357 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
358 if (fn->page == page) {
359 list_del(&fn->list);
360 sbi->fsync_node_num--;
361 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
362 kmem_cache_free(fsync_node_entry_slab, fn);
363 put_page(page);
364 return;
365 }
366 }
367 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
368 f2fs_bug_on(sbi, 1);
369}
370
371void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
372{
373 unsigned long flags;
374
375 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
376 sbi->fsync_seg_id = 0;
377 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
378}
379
380int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
381{
382 struct f2fs_nm_info *nm_i = NM_I(sbi);
383 struct nat_entry *e;
384 bool need = false;
385
386 f2fs_down_read(&nm_i->nat_tree_lock);
387 e = __lookup_nat_cache(nm_i, nid);
388 if (e) {
389 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
390 !get_nat_flag(e, HAS_FSYNCED_INODE))
391 need = true;
392 }
393 f2fs_up_read(&nm_i->nat_tree_lock);
394 return need;
395}
396
397bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
398{
399 struct f2fs_nm_info *nm_i = NM_I(sbi);
400 struct nat_entry *e;
401 bool is_cp = true;
402
403 f2fs_down_read(&nm_i->nat_tree_lock);
404 e = __lookup_nat_cache(nm_i, nid);
405 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
406 is_cp = false;
407 f2fs_up_read(&nm_i->nat_tree_lock);
408 return is_cp;
409}
410
411bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
412{
413 struct f2fs_nm_info *nm_i = NM_I(sbi);
414 struct nat_entry *e;
415 bool need_update = true;
416
417 f2fs_down_read(&nm_i->nat_tree_lock);
418 e = __lookup_nat_cache(nm_i, ino);
419 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
420 (get_nat_flag(e, IS_CHECKPOINTED) ||
421 get_nat_flag(e, HAS_FSYNCED_INODE)))
422 need_update = false;
423 f2fs_up_read(&nm_i->nat_tree_lock);
424 return need_update;
425}
426
427/* must be locked by nat_tree_lock */
428static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
429 struct f2fs_nat_entry *ne)
430{
431 struct f2fs_nm_info *nm_i = NM_I(sbi);
432 struct nat_entry *new, *e;
433
434 /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
435 if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
436 return;
437
438 new = __alloc_nat_entry(sbi, nid, false);
439 if (!new)
440 return;
441
442 f2fs_down_write(&nm_i->nat_tree_lock);
443 e = __lookup_nat_cache(nm_i, nid);
444 if (!e)
445 e = __init_nat_entry(nm_i, new, ne, false);
446 else
447 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
448 nat_get_blkaddr(e) !=
449 le32_to_cpu(ne->block_addr) ||
450 nat_get_version(e) != ne->version);
451 f2fs_up_write(&nm_i->nat_tree_lock);
452 if (e != new)
453 __free_nat_entry(new);
454}
455
456static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
457 block_t new_blkaddr, bool fsync_done)
458{
459 struct f2fs_nm_info *nm_i = NM_I(sbi);
460 struct nat_entry *e;
461 struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
462
463 f2fs_down_write(&nm_i->nat_tree_lock);
464 e = __lookup_nat_cache(nm_i, ni->nid);
465 if (!e) {
466 e = __init_nat_entry(nm_i, new, NULL, true);
467 copy_node_info(&e->ni, ni);
468 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
469 } else if (new_blkaddr == NEW_ADDR) {
470 /*
471 * when nid is reallocated,
472 * previous nat entry can be remained in nat cache.
473 * So, reinitialize it with new information.
474 */
475 copy_node_info(&e->ni, ni);
476 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
477 }
478 /* let's free early to reduce memory consumption */
479 if (e != new)
480 __free_nat_entry(new);
481
482 /* sanity check */
483 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
484 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
485 new_blkaddr == NULL_ADDR);
486 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
487 new_blkaddr == NEW_ADDR);
488 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
489 new_blkaddr == NEW_ADDR);
490
491 /* increment version no as node is removed */
492 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
493 unsigned char version = nat_get_version(e);
494
495 nat_set_version(e, inc_node_version(version));
496 }
497
498 /* change address */
499 nat_set_blkaddr(e, new_blkaddr);
500 if (!__is_valid_data_blkaddr(new_blkaddr))
501 set_nat_flag(e, IS_CHECKPOINTED, false);
502 __set_nat_cache_dirty(nm_i, e);
503
504 /* update fsync_mark if its inode nat entry is still alive */
505 if (ni->nid != ni->ino)
506 e = __lookup_nat_cache(nm_i, ni->ino);
507 if (e) {
508 if (fsync_done && ni->nid == ni->ino)
509 set_nat_flag(e, HAS_FSYNCED_INODE, true);
510 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
511 }
512 f2fs_up_write(&nm_i->nat_tree_lock);
513}
514
515int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
516{
517 struct f2fs_nm_info *nm_i = NM_I(sbi);
518 int nr = nr_shrink;
519
520 if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
521 return 0;
522
523 spin_lock(&nm_i->nat_list_lock);
524 while (nr_shrink) {
525 struct nat_entry *ne;
526
527 if (list_empty(&nm_i->nat_entries))
528 break;
529
530 ne = list_first_entry(&nm_i->nat_entries,
531 struct nat_entry, list);
532 list_del(&ne->list);
533 spin_unlock(&nm_i->nat_list_lock);
534
535 __del_from_nat_cache(nm_i, ne);
536 nr_shrink--;
537
538 spin_lock(&nm_i->nat_list_lock);
539 }
540 spin_unlock(&nm_i->nat_list_lock);
541
542 f2fs_up_write(&nm_i->nat_tree_lock);
543 return nr - nr_shrink;
544}
545
546int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
547 struct node_info *ni, bool checkpoint_context)
548{
549 struct f2fs_nm_info *nm_i = NM_I(sbi);
550 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
551 struct f2fs_journal *journal = curseg->journal;
552 nid_t start_nid = START_NID(nid);
553 struct f2fs_nat_block *nat_blk;
554 struct page *page = NULL;
555 struct f2fs_nat_entry ne;
556 struct nat_entry *e;
557 pgoff_t index;
558 block_t blkaddr;
559 int i;
560
561 ni->nid = nid;
562retry:
563 /* Check nat cache */
564 f2fs_down_read(&nm_i->nat_tree_lock);
565 e = __lookup_nat_cache(nm_i, nid);
566 if (e) {
567 ni->ino = nat_get_ino(e);
568 ni->blk_addr = nat_get_blkaddr(e);
569 ni->version = nat_get_version(e);
570 f2fs_up_read(&nm_i->nat_tree_lock);
571 return 0;
572 }
573
574 /*
575 * Check current segment summary by trying to grab journal_rwsem first.
576 * This sem is on the critical path on the checkpoint requiring the above
577 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
578 * while not bothering checkpoint.
579 */
580 if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
581 down_read(&curseg->journal_rwsem);
582 } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
583 !down_read_trylock(&curseg->journal_rwsem)) {
584 f2fs_up_read(&nm_i->nat_tree_lock);
585 goto retry;
586 }
587
588 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
589 if (i >= 0) {
590 ne = nat_in_journal(journal, i);
591 node_info_from_raw_nat(ni, &ne);
592 }
593 up_read(&curseg->journal_rwsem);
594 if (i >= 0) {
595 f2fs_up_read(&nm_i->nat_tree_lock);
596 goto cache;
597 }
598
599 /* Fill node_info from nat page */
600 index = current_nat_addr(sbi, nid);
601 f2fs_up_read(&nm_i->nat_tree_lock);
602
603 page = f2fs_get_meta_page(sbi, index);
604 if (IS_ERR(page))
605 return PTR_ERR(page);
606
607 nat_blk = (struct f2fs_nat_block *)page_address(page);
608 ne = nat_blk->entries[nid - start_nid];
609 node_info_from_raw_nat(ni, &ne);
610 f2fs_put_page(page, 1);
611cache:
612 blkaddr = le32_to_cpu(ne.block_addr);
613 if (__is_valid_data_blkaddr(blkaddr) &&
614 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
615 return -EFAULT;
616
617 /* cache nat entry */
618 cache_nat_entry(sbi, nid, &ne);
619 return 0;
620}
621
622/*
623 * readahead MAX_RA_NODE number of node pages.
624 */
625static void f2fs_ra_node_pages(struct page *parent, int start, int n)
626{
627 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
628 struct blk_plug plug;
629 int i, end;
630 nid_t nid;
631
632 blk_start_plug(&plug);
633
634 /* Then, try readahead for siblings of the desired node */
635 end = start + n;
636 end = min(end, NIDS_PER_BLOCK);
637 for (i = start; i < end; i++) {
638 nid = get_nid(parent, i, false);
639 f2fs_ra_node_page(sbi, nid);
640 }
641
642 blk_finish_plug(&plug);
643}
644
645pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
646{
647 const long direct_index = ADDRS_PER_INODE(dn->inode);
648 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
649 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
650 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
651 int cur_level = dn->cur_level;
652 int max_level = dn->max_level;
653 pgoff_t base = 0;
654
655 if (!dn->max_level)
656 return pgofs + 1;
657
658 while (max_level-- > cur_level)
659 skipped_unit *= NIDS_PER_BLOCK;
660
661 switch (dn->max_level) {
662 case 3:
663 base += 2 * indirect_blks;
664 fallthrough;
665 case 2:
666 base += 2 * direct_blks;
667 fallthrough;
668 case 1:
669 base += direct_index;
670 break;
671 default:
672 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
673 }
674
675 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
676}
677
678/*
679 * The maximum depth is four.
680 * Offset[0] will have raw inode offset.
681 */
682static int get_node_path(struct inode *inode, long block,
683 int offset[4], unsigned int noffset[4])
684{
685 const long direct_index = ADDRS_PER_INODE(inode);
686 const long direct_blks = ADDRS_PER_BLOCK(inode);
687 const long dptrs_per_blk = NIDS_PER_BLOCK;
688 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
689 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
690 int n = 0;
691 int level = 0;
692
693 noffset[0] = 0;
694
695 if (block < direct_index) {
696 offset[n] = block;
697 goto got;
698 }
699 block -= direct_index;
700 if (block < direct_blks) {
701 offset[n++] = NODE_DIR1_BLOCK;
702 noffset[n] = 1;
703 offset[n] = block;
704 level = 1;
705 goto got;
706 }
707 block -= direct_blks;
708 if (block < direct_blks) {
709 offset[n++] = NODE_DIR2_BLOCK;
710 noffset[n] = 2;
711 offset[n] = block;
712 level = 1;
713 goto got;
714 }
715 block -= direct_blks;
716 if (block < indirect_blks) {
717 offset[n++] = NODE_IND1_BLOCK;
718 noffset[n] = 3;
719 offset[n++] = block / direct_blks;
720 noffset[n] = 4 + offset[n - 1];
721 offset[n] = block % direct_blks;
722 level = 2;
723 goto got;
724 }
725 block -= indirect_blks;
726 if (block < indirect_blks) {
727 offset[n++] = NODE_IND2_BLOCK;
728 noffset[n] = 4 + dptrs_per_blk;
729 offset[n++] = block / direct_blks;
730 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
731 offset[n] = block % direct_blks;
732 level = 2;
733 goto got;
734 }
735 block -= indirect_blks;
736 if (block < dindirect_blks) {
737 offset[n++] = NODE_DIND_BLOCK;
738 noffset[n] = 5 + (dptrs_per_blk * 2);
739 offset[n++] = block / indirect_blks;
740 noffset[n] = 6 + (dptrs_per_blk * 2) +
741 offset[n - 1] * (dptrs_per_blk + 1);
742 offset[n++] = (block / direct_blks) % dptrs_per_blk;
743 noffset[n] = 7 + (dptrs_per_blk * 2) +
744 offset[n - 2] * (dptrs_per_blk + 1) +
745 offset[n - 1];
746 offset[n] = block % direct_blks;
747 level = 3;
748 goto got;
749 } else {
750 return -E2BIG;
751 }
752got:
753 return level;
754}
755
756/*
757 * Caller should call f2fs_put_dnode(dn).
758 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
759 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
760 */
761int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
762{
763 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
764 struct page *npage[4];
765 struct page *parent = NULL;
766 int offset[4];
767 unsigned int noffset[4];
768 nid_t nids[4];
769 int level, i = 0;
770 int err = 0;
771
772 level = get_node_path(dn->inode, index, offset, noffset);
773 if (level < 0)
774 return level;
775
776 nids[0] = dn->inode->i_ino;
777 npage[0] = dn->inode_page;
778
779 if (!npage[0]) {
780 npage[0] = f2fs_get_node_page(sbi, nids[0]);
781 if (IS_ERR(npage[0]))
782 return PTR_ERR(npage[0]);
783 }
784
785 /* if inline_data is set, should not report any block indices */
786 if (f2fs_has_inline_data(dn->inode) && index) {
787 err = -ENOENT;
788 f2fs_put_page(npage[0], 1);
789 goto release_out;
790 }
791
792 parent = npage[0];
793 if (level != 0)
794 nids[1] = get_nid(parent, offset[0], true);
795 dn->inode_page = npage[0];
796 dn->inode_page_locked = true;
797
798 /* get indirect or direct nodes */
799 for (i = 1; i <= level; i++) {
800 bool done = false;
801
802 if (!nids[i] && mode == ALLOC_NODE) {
803 /* alloc new node */
804 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
805 err = -ENOSPC;
806 goto release_pages;
807 }
808
809 dn->nid = nids[i];
810 npage[i] = f2fs_new_node_page(dn, noffset[i]);
811 if (IS_ERR(npage[i])) {
812 f2fs_alloc_nid_failed(sbi, nids[i]);
813 err = PTR_ERR(npage[i]);
814 goto release_pages;
815 }
816
817 set_nid(parent, offset[i - 1], nids[i], i == 1);
818 f2fs_alloc_nid_done(sbi, nids[i]);
819 done = true;
820 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
821 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
822 if (IS_ERR(npage[i])) {
823 err = PTR_ERR(npage[i]);
824 goto release_pages;
825 }
826 done = true;
827 }
828 if (i == 1) {
829 dn->inode_page_locked = false;
830 unlock_page(parent);
831 } else {
832 f2fs_put_page(parent, 1);
833 }
834
835 if (!done) {
836 npage[i] = f2fs_get_node_page(sbi, nids[i]);
837 if (IS_ERR(npage[i])) {
838 err = PTR_ERR(npage[i]);
839 f2fs_put_page(npage[0], 0);
840 goto release_out;
841 }
842 }
843 if (i < level) {
844 parent = npage[i];
845 nids[i + 1] = get_nid(parent, offset[i], false);
846 }
847 }
848 dn->nid = nids[level];
849 dn->ofs_in_node = offset[level];
850 dn->node_page = npage[level];
851 dn->data_blkaddr = f2fs_data_blkaddr(dn);
852
853 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
854 f2fs_sb_has_readonly(sbi)) {
855 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
856 block_t blkaddr;
857
858 if (!c_len)
859 goto out;
860
861 blkaddr = f2fs_data_blkaddr(dn);
862 if (blkaddr == COMPRESS_ADDR)
863 blkaddr = data_blkaddr(dn->inode, dn->node_page,
864 dn->ofs_in_node + 1);
865
866 f2fs_update_read_extent_tree_range_compressed(dn->inode,
867 index, blkaddr,
868 F2FS_I(dn->inode)->i_cluster_size,
869 c_len);
870 }
871out:
872 return 0;
873
874release_pages:
875 f2fs_put_page(parent, 1);
876 if (i > 1)
877 f2fs_put_page(npage[0], 0);
878release_out:
879 dn->inode_page = NULL;
880 dn->node_page = NULL;
881 if (err == -ENOENT) {
882 dn->cur_level = i;
883 dn->max_level = level;
884 dn->ofs_in_node = offset[level];
885 }
886 return err;
887}
888
889static int truncate_node(struct dnode_of_data *dn)
890{
891 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
892 struct node_info ni;
893 int err;
894 pgoff_t index;
895
896 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
897 if (err)
898 return err;
899
900 /* Deallocate node address */
901 f2fs_invalidate_blocks(sbi, ni.blk_addr);
902 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
903 set_node_addr(sbi, &ni, NULL_ADDR, false);
904
905 if (dn->nid == dn->inode->i_ino) {
906 f2fs_remove_orphan_inode(sbi, dn->nid);
907 dec_valid_inode_count(sbi);
908 f2fs_inode_synced(dn->inode);
909 }
910
911 clear_node_page_dirty(dn->node_page);
912 set_sbi_flag(sbi, SBI_IS_DIRTY);
913
914 index = dn->node_page->index;
915 f2fs_put_page(dn->node_page, 1);
916
917 invalidate_mapping_pages(NODE_MAPPING(sbi),
918 index, index);
919
920 dn->node_page = NULL;
921 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
922
923 return 0;
924}
925
926static int truncate_dnode(struct dnode_of_data *dn)
927{
928 struct page *page;
929 int err;
930
931 if (dn->nid == 0)
932 return 1;
933
934 /* get direct node */
935 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
936 if (PTR_ERR(page) == -ENOENT)
937 return 1;
938 else if (IS_ERR(page))
939 return PTR_ERR(page);
940
941 /* Make dnode_of_data for parameter */
942 dn->node_page = page;
943 dn->ofs_in_node = 0;
944 f2fs_truncate_data_blocks(dn);
945 err = truncate_node(dn);
946 if (err)
947 return err;
948
949 return 1;
950}
951
952static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
953 int ofs, int depth)
954{
955 struct dnode_of_data rdn = *dn;
956 struct page *page;
957 struct f2fs_node *rn;
958 nid_t child_nid;
959 unsigned int child_nofs;
960 int freed = 0;
961 int i, ret;
962
963 if (dn->nid == 0)
964 return NIDS_PER_BLOCK + 1;
965
966 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
967
968 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
969 if (IS_ERR(page)) {
970 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
971 return PTR_ERR(page);
972 }
973
974 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
975
976 rn = F2FS_NODE(page);
977 if (depth < 3) {
978 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
979 child_nid = le32_to_cpu(rn->in.nid[i]);
980 if (child_nid == 0)
981 continue;
982 rdn.nid = child_nid;
983 ret = truncate_dnode(&rdn);
984 if (ret < 0)
985 goto out_err;
986 if (set_nid(page, i, 0, false))
987 dn->node_changed = true;
988 }
989 } else {
990 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
991 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
992 child_nid = le32_to_cpu(rn->in.nid[i]);
993 if (child_nid == 0) {
994 child_nofs += NIDS_PER_BLOCK + 1;
995 continue;
996 }
997 rdn.nid = child_nid;
998 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
999 if (ret == (NIDS_PER_BLOCK + 1)) {
1000 if (set_nid(page, i, 0, false))
1001 dn->node_changed = true;
1002 child_nofs += ret;
1003 } else if (ret < 0 && ret != -ENOENT) {
1004 goto out_err;
1005 }
1006 }
1007 freed = child_nofs;
1008 }
1009
1010 if (!ofs) {
1011 /* remove current indirect node */
1012 dn->node_page = page;
1013 ret = truncate_node(dn);
1014 if (ret)
1015 goto out_err;
1016 freed++;
1017 } else {
1018 f2fs_put_page(page, 1);
1019 }
1020 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1021 return freed;
1022
1023out_err:
1024 f2fs_put_page(page, 1);
1025 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1026 return ret;
1027}
1028
1029static int truncate_partial_nodes(struct dnode_of_data *dn,
1030 struct f2fs_inode *ri, int *offset, int depth)
1031{
1032 struct page *pages[2];
1033 nid_t nid[3];
1034 nid_t child_nid;
1035 int err = 0;
1036 int i;
1037 int idx = depth - 2;
1038
1039 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1040 if (!nid[0])
1041 return 0;
1042
1043 /* get indirect nodes in the path */
1044 for (i = 0; i < idx + 1; i++) {
1045 /* reference count'll be increased */
1046 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1047 if (IS_ERR(pages[i])) {
1048 err = PTR_ERR(pages[i]);
1049 idx = i - 1;
1050 goto fail;
1051 }
1052 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1053 }
1054
1055 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1056
1057 /* free direct nodes linked to a partial indirect node */
1058 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1059 child_nid = get_nid(pages[idx], i, false);
1060 if (!child_nid)
1061 continue;
1062 dn->nid = child_nid;
1063 err = truncate_dnode(dn);
1064 if (err < 0)
1065 goto fail;
1066 if (set_nid(pages[idx], i, 0, false))
1067 dn->node_changed = true;
1068 }
1069
1070 if (offset[idx + 1] == 0) {
1071 dn->node_page = pages[idx];
1072 dn->nid = nid[idx];
1073 err = truncate_node(dn);
1074 if (err)
1075 goto fail;
1076 } else {
1077 f2fs_put_page(pages[idx], 1);
1078 }
1079 offset[idx]++;
1080 offset[idx + 1] = 0;
1081 idx--;
1082fail:
1083 for (i = idx; i >= 0; i--)
1084 f2fs_put_page(pages[i], 1);
1085
1086 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1087
1088 return err;
1089}
1090
1091/*
1092 * All the block addresses of data and nodes should be nullified.
1093 */
1094int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1095{
1096 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1097 int err = 0, cont = 1;
1098 int level, offset[4], noffset[4];
1099 unsigned int nofs = 0;
1100 struct f2fs_inode *ri;
1101 struct dnode_of_data dn;
1102 struct page *page;
1103
1104 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1105
1106 level = get_node_path(inode, from, offset, noffset);
1107 if (level < 0) {
1108 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1109 return level;
1110 }
1111
1112 page = f2fs_get_node_page(sbi, inode->i_ino);
1113 if (IS_ERR(page)) {
1114 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1115 return PTR_ERR(page);
1116 }
1117
1118 set_new_dnode(&dn, inode, page, NULL, 0);
1119 unlock_page(page);
1120
1121 ri = F2FS_INODE(page);
1122 switch (level) {
1123 case 0:
1124 case 1:
1125 nofs = noffset[1];
1126 break;
1127 case 2:
1128 nofs = noffset[1];
1129 if (!offset[level - 1])
1130 goto skip_partial;
1131 err = truncate_partial_nodes(&dn, ri, offset, level);
1132 if (err < 0 && err != -ENOENT)
1133 goto fail;
1134 nofs += 1 + NIDS_PER_BLOCK;
1135 break;
1136 case 3:
1137 nofs = 5 + 2 * NIDS_PER_BLOCK;
1138 if (!offset[level - 1])
1139 goto skip_partial;
1140 err = truncate_partial_nodes(&dn, ri, offset, level);
1141 if (err < 0 && err != -ENOENT)
1142 goto fail;
1143 break;
1144 default:
1145 BUG();
1146 }
1147
1148skip_partial:
1149 while (cont) {
1150 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1151 switch (offset[0]) {
1152 case NODE_DIR1_BLOCK:
1153 case NODE_DIR2_BLOCK:
1154 err = truncate_dnode(&dn);
1155 break;
1156
1157 case NODE_IND1_BLOCK:
1158 case NODE_IND2_BLOCK:
1159 err = truncate_nodes(&dn, nofs, offset[1], 2);
1160 break;
1161
1162 case NODE_DIND_BLOCK:
1163 err = truncate_nodes(&dn, nofs, offset[1], 3);
1164 cont = 0;
1165 break;
1166
1167 default:
1168 BUG();
1169 }
1170 if (err < 0 && err != -ENOENT)
1171 goto fail;
1172 if (offset[1] == 0 &&
1173 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1174 lock_page(page);
1175 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1176 f2fs_wait_on_page_writeback(page, NODE, true, true);
1177 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1178 set_page_dirty(page);
1179 unlock_page(page);
1180 }
1181 offset[1] = 0;
1182 offset[0]++;
1183 nofs += err;
1184 }
1185fail:
1186 f2fs_put_page(page, 0);
1187 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1188 return err > 0 ? 0 : err;
1189}
1190
1191/* caller must lock inode page */
1192int f2fs_truncate_xattr_node(struct inode *inode)
1193{
1194 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1195 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1196 struct dnode_of_data dn;
1197 struct page *npage;
1198 int err;
1199
1200 if (!nid)
1201 return 0;
1202
1203 npage = f2fs_get_node_page(sbi, nid);
1204 if (IS_ERR(npage))
1205 return PTR_ERR(npage);
1206
1207 set_new_dnode(&dn, inode, NULL, npage, nid);
1208 err = truncate_node(&dn);
1209 if (err) {
1210 f2fs_put_page(npage, 1);
1211 return err;
1212 }
1213
1214 f2fs_i_xnid_write(inode, 0);
1215
1216 return 0;
1217}
1218
1219/*
1220 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1221 * f2fs_unlock_op().
1222 */
1223int f2fs_remove_inode_page(struct inode *inode)
1224{
1225 struct dnode_of_data dn;
1226 int err;
1227
1228 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1229 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1230 if (err)
1231 return err;
1232
1233 err = f2fs_truncate_xattr_node(inode);
1234 if (err) {
1235 f2fs_put_dnode(&dn);
1236 return err;
1237 }
1238
1239 /* remove potential inline_data blocks */
1240 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1241 S_ISLNK(inode->i_mode))
1242 f2fs_truncate_data_blocks_range(&dn, 1);
1243
1244 /* 0 is possible, after f2fs_new_inode() has failed */
1245 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1246 f2fs_put_dnode(&dn);
1247 return -EIO;
1248 }
1249
1250 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1251 f2fs_warn(F2FS_I_SB(inode),
1252 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1253 inode->i_ino, (unsigned long long)inode->i_blocks);
1254 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1255 }
1256
1257 /* will put inode & node pages */
1258 err = truncate_node(&dn);
1259 if (err) {
1260 f2fs_put_dnode(&dn);
1261 return err;
1262 }
1263 return 0;
1264}
1265
1266struct page *f2fs_new_inode_page(struct inode *inode)
1267{
1268 struct dnode_of_data dn;
1269
1270 /* allocate inode page for new inode */
1271 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1272
1273 /* caller should f2fs_put_page(page, 1); */
1274 return f2fs_new_node_page(&dn, 0);
1275}
1276
1277struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1278{
1279 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1280 struct node_info new_ni;
1281 struct page *page;
1282 int err;
1283
1284 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1285 return ERR_PTR(-EPERM);
1286
1287 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1288 if (!page)
1289 return ERR_PTR(-ENOMEM);
1290
1291 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1292 goto fail;
1293
1294#ifdef CONFIG_F2FS_CHECK_FS
1295 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1296 if (err) {
1297 dec_valid_node_count(sbi, dn->inode, !ofs);
1298 goto fail;
1299 }
1300 if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1301 err = -EFSCORRUPTED;
1302 set_sbi_flag(sbi, SBI_NEED_FSCK);
1303 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1304 goto fail;
1305 }
1306#endif
1307 new_ni.nid = dn->nid;
1308 new_ni.ino = dn->inode->i_ino;
1309 new_ni.blk_addr = NULL_ADDR;
1310 new_ni.flag = 0;
1311 new_ni.version = 0;
1312 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1313
1314 f2fs_wait_on_page_writeback(page, NODE, true, true);
1315 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1316 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1317 if (!PageUptodate(page))
1318 SetPageUptodate(page);
1319 if (set_page_dirty(page))
1320 dn->node_changed = true;
1321
1322 if (f2fs_has_xattr_block(ofs))
1323 f2fs_i_xnid_write(dn->inode, dn->nid);
1324
1325 if (ofs == 0)
1326 inc_valid_inode_count(sbi);
1327 return page;
1328
1329fail:
1330 clear_node_page_dirty(page);
1331 f2fs_put_page(page, 1);
1332 return ERR_PTR(err);
1333}
1334
1335/*
1336 * Caller should do after getting the following values.
1337 * 0: f2fs_put_page(page, 0)
1338 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1339 */
1340static int read_node_page(struct page *page, blk_opf_t op_flags)
1341{
1342 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1343 struct node_info ni;
1344 struct f2fs_io_info fio = {
1345 .sbi = sbi,
1346 .type = NODE,
1347 .op = REQ_OP_READ,
1348 .op_flags = op_flags,
1349 .page = page,
1350 .encrypted_page = NULL,
1351 };
1352 int err;
1353
1354 if (PageUptodate(page)) {
1355 if (!f2fs_inode_chksum_verify(sbi, page)) {
1356 ClearPageUptodate(page);
1357 return -EFSBADCRC;
1358 }
1359 return LOCKED_PAGE;
1360 }
1361
1362 err = f2fs_get_node_info(sbi, page->index, &ni, false);
1363 if (err)
1364 return err;
1365
1366 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1367 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1368 ClearPageUptodate(page);
1369 return -ENOENT;
1370 }
1371
1372 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1373
1374 err = f2fs_submit_page_bio(&fio);
1375
1376 if (!err)
1377 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1378
1379 return err;
1380}
1381
1382/*
1383 * Readahead a node page
1384 */
1385void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1386{
1387 struct page *apage;
1388 int err;
1389
1390 if (!nid)
1391 return;
1392 if (f2fs_check_nid_range(sbi, nid))
1393 return;
1394
1395 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1396 if (apage)
1397 return;
1398
1399 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1400 if (!apage)
1401 return;
1402
1403 err = read_node_page(apage, REQ_RAHEAD);
1404 f2fs_put_page(apage, err ? 1 : 0);
1405}
1406
1407static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1408 struct page *parent, int start)
1409{
1410 struct page *page;
1411 int err;
1412
1413 if (!nid)
1414 return ERR_PTR(-ENOENT);
1415 if (f2fs_check_nid_range(sbi, nid))
1416 return ERR_PTR(-EINVAL);
1417repeat:
1418 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1419 if (!page)
1420 return ERR_PTR(-ENOMEM);
1421
1422 err = read_node_page(page, 0);
1423 if (err < 0) {
1424 goto out_put_err;
1425 } else if (err == LOCKED_PAGE) {
1426 err = 0;
1427 goto page_hit;
1428 }
1429
1430 if (parent)
1431 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1432
1433 lock_page(page);
1434
1435 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1436 f2fs_put_page(page, 1);
1437 goto repeat;
1438 }
1439
1440 if (unlikely(!PageUptodate(page))) {
1441 err = -EIO;
1442 goto out_err;
1443 }
1444
1445 if (!f2fs_inode_chksum_verify(sbi, page)) {
1446 err = -EFSBADCRC;
1447 goto out_err;
1448 }
1449page_hit:
1450 if (likely(nid == nid_of_node(page)))
1451 return page;
1452
1453 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1454 nid, nid_of_node(page), ino_of_node(page),
1455 ofs_of_node(page), cpver_of_node(page),
1456 next_blkaddr_of_node(page));
1457 set_sbi_flag(sbi, SBI_NEED_FSCK);
1458 err = -EINVAL;
1459out_err:
1460 ClearPageUptodate(page);
1461out_put_err:
1462 /* ENOENT comes from read_node_page which is not an error. */
1463 if (err != -ENOENT)
1464 f2fs_handle_page_eio(sbi, page->index, NODE);
1465 f2fs_put_page(page, 1);
1466 return ERR_PTR(err);
1467}
1468
1469struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1470{
1471 return __get_node_page(sbi, nid, NULL, 0);
1472}
1473
1474struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1475{
1476 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1477 nid_t nid = get_nid(parent, start, false);
1478
1479 return __get_node_page(sbi, nid, parent, start);
1480}
1481
1482static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1483{
1484 struct inode *inode;
1485 struct page *page;
1486 int ret;
1487
1488 /* should flush inline_data before evict_inode */
1489 inode = ilookup(sbi->sb, ino);
1490 if (!inode)
1491 return;
1492
1493 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1494 FGP_LOCK|FGP_NOWAIT, 0);
1495 if (!page)
1496 goto iput_out;
1497
1498 if (!PageUptodate(page))
1499 goto page_out;
1500
1501 if (!PageDirty(page))
1502 goto page_out;
1503
1504 if (!clear_page_dirty_for_io(page))
1505 goto page_out;
1506
1507 ret = f2fs_write_inline_data(inode, page);
1508 inode_dec_dirty_pages(inode);
1509 f2fs_remove_dirty_inode(inode);
1510 if (ret)
1511 set_page_dirty(page);
1512page_out:
1513 f2fs_put_page(page, 1);
1514iput_out:
1515 iput(inode);
1516}
1517
1518static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1519{
1520 pgoff_t index;
1521 struct pagevec pvec;
1522 struct page *last_page = NULL;
1523 int nr_pages;
1524
1525 pagevec_init(&pvec);
1526 index = 0;
1527
1528 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1529 PAGECACHE_TAG_DIRTY))) {
1530 int i;
1531
1532 for (i = 0; i < nr_pages; i++) {
1533 struct page *page = pvec.pages[i];
1534
1535 if (unlikely(f2fs_cp_error(sbi))) {
1536 f2fs_put_page(last_page, 0);
1537 pagevec_release(&pvec);
1538 return ERR_PTR(-EIO);
1539 }
1540
1541 if (!IS_DNODE(page) || !is_cold_node(page))
1542 continue;
1543 if (ino_of_node(page) != ino)
1544 continue;
1545
1546 lock_page(page);
1547
1548 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1549continue_unlock:
1550 unlock_page(page);
1551 continue;
1552 }
1553 if (ino_of_node(page) != ino)
1554 goto continue_unlock;
1555
1556 if (!PageDirty(page)) {
1557 /* someone wrote it for us */
1558 goto continue_unlock;
1559 }
1560
1561 if (last_page)
1562 f2fs_put_page(last_page, 0);
1563
1564 get_page(page);
1565 last_page = page;
1566 unlock_page(page);
1567 }
1568 pagevec_release(&pvec);
1569 cond_resched();
1570 }
1571 return last_page;
1572}
1573
1574static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1575 struct writeback_control *wbc, bool do_balance,
1576 enum iostat_type io_type, unsigned int *seq_id)
1577{
1578 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1579 nid_t nid;
1580 struct node_info ni;
1581 struct f2fs_io_info fio = {
1582 .sbi = sbi,
1583 .ino = ino_of_node(page),
1584 .type = NODE,
1585 .op = REQ_OP_WRITE,
1586 .op_flags = wbc_to_write_flags(wbc),
1587 .page = page,
1588 .encrypted_page = NULL,
1589 .submitted = false,
1590 .io_type = io_type,
1591 .io_wbc = wbc,
1592 };
1593 unsigned int seq;
1594
1595 trace_f2fs_writepage(page, NODE);
1596
1597 if (unlikely(f2fs_cp_error(sbi))) {
1598 ClearPageUptodate(page);
1599 dec_page_count(sbi, F2FS_DIRTY_NODES);
1600 unlock_page(page);
1601 return 0;
1602 }
1603
1604 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1605 goto redirty_out;
1606
1607 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1608 wbc->sync_mode == WB_SYNC_NONE &&
1609 IS_DNODE(page) && is_cold_node(page))
1610 goto redirty_out;
1611
1612 /* get old block addr of this node page */
1613 nid = nid_of_node(page);
1614 f2fs_bug_on(sbi, page->index != nid);
1615
1616 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1617 goto redirty_out;
1618
1619 if (wbc->for_reclaim) {
1620 if (!f2fs_down_read_trylock(&sbi->node_write))
1621 goto redirty_out;
1622 } else {
1623 f2fs_down_read(&sbi->node_write);
1624 }
1625
1626 /* This page is already truncated */
1627 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1628 ClearPageUptodate(page);
1629 dec_page_count(sbi, F2FS_DIRTY_NODES);
1630 f2fs_up_read(&sbi->node_write);
1631 unlock_page(page);
1632 return 0;
1633 }
1634
1635 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1636 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1637 DATA_GENERIC_ENHANCE)) {
1638 f2fs_up_read(&sbi->node_write);
1639 goto redirty_out;
1640 }
1641
1642 if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1643 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1644
1645 /* should add to global list before clearing PAGECACHE status */
1646 if (f2fs_in_warm_node_list(sbi, page)) {
1647 seq = f2fs_add_fsync_node_entry(sbi, page);
1648 if (seq_id)
1649 *seq_id = seq;
1650 }
1651
1652 set_page_writeback(page);
1653 ClearPageError(page);
1654
1655 fio.old_blkaddr = ni.blk_addr;
1656 f2fs_do_write_node_page(nid, &fio);
1657 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1658 dec_page_count(sbi, F2FS_DIRTY_NODES);
1659 f2fs_up_read(&sbi->node_write);
1660
1661 if (wbc->for_reclaim) {
1662 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1663 submitted = NULL;
1664 }
1665
1666 unlock_page(page);
1667
1668 if (unlikely(f2fs_cp_error(sbi))) {
1669 f2fs_submit_merged_write(sbi, NODE);
1670 submitted = NULL;
1671 }
1672 if (submitted)
1673 *submitted = fio.submitted;
1674
1675 if (do_balance)
1676 f2fs_balance_fs(sbi, false);
1677 return 0;
1678
1679redirty_out:
1680 redirty_page_for_writepage(wbc, page);
1681 return AOP_WRITEPAGE_ACTIVATE;
1682}
1683
1684int f2fs_move_node_page(struct page *node_page, int gc_type)
1685{
1686 int err = 0;
1687
1688 if (gc_type == FG_GC) {
1689 struct writeback_control wbc = {
1690 .sync_mode = WB_SYNC_ALL,
1691 .nr_to_write = 1,
1692 .for_reclaim = 0,
1693 };
1694
1695 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1696
1697 set_page_dirty(node_page);
1698
1699 if (!clear_page_dirty_for_io(node_page)) {
1700 err = -EAGAIN;
1701 goto out_page;
1702 }
1703
1704 if (__write_node_page(node_page, false, NULL,
1705 &wbc, false, FS_GC_NODE_IO, NULL)) {
1706 err = -EAGAIN;
1707 unlock_page(node_page);
1708 }
1709 goto release_page;
1710 } else {
1711 /* set page dirty and write it */
1712 if (!PageWriteback(node_page))
1713 set_page_dirty(node_page);
1714 }
1715out_page:
1716 unlock_page(node_page);
1717release_page:
1718 f2fs_put_page(node_page, 0);
1719 return err;
1720}
1721
1722static int f2fs_write_node_page(struct page *page,
1723 struct writeback_control *wbc)
1724{
1725 return __write_node_page(page, false, NULL, wbc, false,
1726 FS_NODE_IO, NULL);
1727}
1728
1729int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1730 struct writeback_control *wbc, bool atomic,
1731 unsigned int *seq_id)
1732{
1733 pgoff_t index;
1734 struct pagevec pvec;
1735 int ret = 0;
1736 struct page *last_page = NULL;
1737 bool marked = false;
1738 nid_t ino = inode->i_ino;
1739 int nr_pages;
1740 int nwritten = 0;
1741
1742 if (atomic) {
1743 last_page = last_fsync_dnode(sbi, ino);
1744 if (IS_ERR_OR_NULL(last_page))
1745 return PTR_ERR_OR_ZERO(last_page);
1746 }
1747retry:
1748 pagevec_init(&pvec);
1749 index = 0;
1750
1751 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1752 PAGECACHE_TAG_DIRTY))) {
1753 int i;
1754
1755 for (i = 0; i < nr_pages; i++) {
1756 struct page *page = pvec.pages[i];
1757 bool submitted = false;
1758
1759 if (unlikely(f2fs_cp_error(sbi))) {
1760 f2fs_put_page(last_page, 0);
1761 pagevec_release(&pvec);
1762 ret = -EIO;
1763 goto out;
1764 }
1765
1766 if (!IS_DNODE(page) || !is_cold_node(page))
1767 continue;
1768 if (ino_of_node(page) != ino)
1769 continue;
1770
1771 lock_page(page);
1772
1773 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1774continue_unlock:
1775 unlock_page(page);
1776 continue;
1777 }
1778 if (ino_of_node(page) != ino)
1779 goto continue_unlock;
1780
1781 if (!PageDirty(page) && page != last_page) {
1782 /* someone wrote it for us */
1783 goto continue_unlock;
1784 }
1785
1786 f2fs_wait_on_page_writeback(page, NODE, true, true);
1787
1788 set_fsync_mark(page, 0);
1789 set_dentry_mark(page, 0);
1790
1791 if (!atomic || page == last_page) {
1792 set_fsync_mark(page, 1);
1793 percpu_counter_inc(&sbi->rf_node_block_count);
1794 if (IS_INODE(page)) {
1795 if (is_inode_flag_set(inode,
1796 FI_DIRTY_INODE))
1797 f2fs_update_inode(inode, page);
1798 set_dentry_mark(page,
1799 f2fs_need_dentry_mark(sbi, ino));
1800 }
1801 /* may be written by other thread */
1802 if (!PageDirty(page))
1803 set_page_dirty(page);
1804 }
1805
1806 if (!clear_page_dirty_for_io(page))
1807 goto continue_unlock;
1808
1809 ret = __write_node_page(page, atomic &&
1810 page == last_page,
1811 &submitted, wbc, true,
1812 FS_NODE_IO, seq_id);
1813 if (ret) {
1814 unlock_page(page);
1815 f2fs_put_page(last_page, 0);
1816 break;
1817 } else if (submitted) {
1818 nwritten++;
1819 }
1820
1821 if (page == last_page) {
1822 f2fs_put_page(page, 0);
1823 marked = true;
1824 break;
1825 }
1826 }
1827 pagevec_release(&pvec);
1828 cond_resched();
1829
1830 if (ret || marked)
1831 break;
1832 }
1833 if (!ret && atomic && !marked) {
1834 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1835 ino, last_page->index);
1836 lock_page(last_page);
1837 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1838 set_page_dirty(last_page);
1839 unlock_page(last_page);
1840 goto retry;
1841 }
1842out:
1843 if (nwritten)
1844 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1845 return ret ? -EIO : 0;
1846}
1847
1848static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1849{
1850 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1851 bool clean;
1852
1853 if (inode->i_ino != ino)
1854 return 0;
1855
1856 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1857 return 0;
1858
1859 spin_lock(&sbi->inode_lock[DIRTY_META]);
1860 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1861 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1862
1863 if (clean)
1864 return 0;
1865
1866 inode = igrab(inode);
1867 if (!inode)
1868 return 0;
1869 return 1;
1870}
1871
1872static bool flush_dirty_inode(struct page *page)
1873{
1874 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1875 struct inode *inode;
1876 nid_t ino = ino_of_node(page);
1877
1878 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1879 if (!inode)
1880 return false;
1881
1882 f2fs_update_inode(inode, page);
1883 unlock_page(page);
1884
1885 iput(inode);
1886 return true;
1887}
1888
1889void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1890{
1891 pgoff_t index = 0;
1892 struct pagevec pvec;
1893 int nr_pages;
1894
1895 pagevec_init(&pvec);
1896
1897 while ((nr_pages = pagevec_lookup_tag(&pvec,
1898 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1899 int i;
1900
1901 for (i = 0; i < nr_pages; i++) {
1902 struct page *page = pvec.pages[i];
1903
1904 if (!IS_DNODE(page))
1905 continue;
1906
1907 lock_page(page);
1908
1909 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1910continue_unlock:
1911 unlock_page(page);
1912 continue;
1913 }
1914
1915 if (!PageDirty(page)) {
1916 /* someone wrote it for us */
1917 goto continue_unlock;
1918 }
1919
1920 /* flush inline_data, if it's async context. */
1921 if (page_private_inline(page)) {
1922 clear_page_private_inline(page);
1923 unlock_page(page);
1924 flush_inline_data(sbi, ino_of_node(page));
1925 continue;
1926 }
1927 unlock_page(page);
1928 }
1929 pagevec_release(&pvec);
1930 cond_resched();
1931 }
1932}
1933
1934int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1935 struct writeback_control *wbc,
1936 bool do_balance, enum iostat_type io_type)
1937{
1938 pgoff_t index;
1939 struct pagevec pvec;
1940 int step = 0;
1941 int nwritten = 0;
1942 int ret = 0;
1943 int nr_pages, done = 0;
1944
1945 pagevec_init(&pvec);
1946
1947next_step:
1948 index = 0;
1949
1950 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1951 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1952 int i;
1953
1954 for (i = 0; i < nr_pages; i++) {
1955 struct page *page = pvec.pages[i];
1956 bool submitted = false;
1957
1958 /* give a priority to WB_SYNC threads */
1959 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1960 wbc->sync_mode == WB_SYNC_NONE) {
1961 done = 1;
1962 break;
1963 }
1964
1965 /*
1966 * flushing sequence with step:
1967 * 0. indirect nodes
1968 * 1. dentry dnodes
1969 * 2. file dnodes
1970 */
1971 if (step == 0 && IS_DNODE(page))
1972 continue;
1973 if (step == 1 && (!IS_DNODE(page) ||
1974 is_cold_node(page)))
1975 continue;
1976 if (step == 2 && (!IS_DNODE(page) ||
1977 !is_cold_node(page)))
1978 continue;
1979lock_node:
1980 if (wbc->sync_mode == WB_SYNC_ALL)
1981 lock_page(page);
1982 else if (!trylock_page(page))
1983 continue;
1984
1985 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1986continue_unlock:
1987 unlock_page(page);
1988 continue;
1989 }
1990
1991 if (!PageDirty(page)) {
1992 /* someone wrote it for us */
1993 goto continue_unlock;
1994 }
1995
1996 /* flush inline_data/inode, if it's async context. */
1997 if (!do_balance)
1998 goto write_node;
1999
2000 /* flush inline_data */
2001 if (page_private_inline(page)) {
2002 clear_page_private_inline(page);
2003 unlock_page(page);
2004 flush_inline_data(sbi, ino_of_node(page));
2005 goto lock_node;
2006 }
2007
2008 /* flush dirty inode */
2009 if (IS_INODE(page) && flush_dirty_inode(page))
2010 goto lock_node;
2011write_node:
2012 f2fs_wait_on_page_writeback(page, NODE, true, true);
2013
2014 if (!clear_page_dirty_for_io(page))
2015 goto continue_unlock;
2016
2017 set_fsync_mark(page, 0);
2018 set_dentry_mark(page, 0);
2019
2020 ret = __write_node_page(page, false, &submitted,
2021 wbc, do_balance, io_type, NULL);
2022 if (ret)
2023 unlock_page(page);
2024 else if (submitted)
2025 nwritten++;
2026
2027 if (--wbc->nr_to_write == 0)
2028 break;
2029 }
2030 pagevec_release(&pvec);
2031 cond_resched();
2032
2033 if (wbc->nr_to_write == 0) {
2034 step = 2;
2035 break;
2036 }
2037 }
2038
2039 if (step < 2) {
2040 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2041 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2042 goto out;
2043 step++;
2044 goto next_step;
2045 }
2046out:
2047 if (nwritten)
2048 f2fs_submit_merged_write(sbi, NODE);
2049
2050 if (unlikely(f2fs_cp_error(sbi)))
2051 return -EIO;
2052 return ret;
2053}
2054
2055int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2056 unsigned int seq_id)
2057{
2058 struct fsync_node_entry *fn;
2059 struct page *page;
2060 struct list_head *head = &sbi->fsync_node_list;
2061 unsigned long flags;
2062 unsigned int cur_seq_id = 0;
2063 int ret2, ret = 0;
2064
2065 while (seq_id && cur_seq_id < seq_id) {
2066 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2067 if (list_empty(head)) {
2068 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2069 break;
2070 }
2071 fn = list_first_entry(head, struct fsync_node_entry, list);
2072 if (fn->seq_id > seq_id) {
2073 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2074 break;
2075 }
2076 cur_seq_id = fn->seq_id;
2077 page = fn->page;
2078 get_page(page);
2079 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2080
2081 f2fs_wait_on_page_writeback(page, NODE, true, false);
2082 if (TestClearPageError(page))
2083 ret = -EIO;
2084
2085 put_page(page);
2086
2087 if (ret)
2088 break;
2089 }
2090
2091 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2092 if (!ret)
2093 ret = ret2;
2094
2095 return ret;
2096}
2097
2098static int f2fs_write_node_pages(struct address_space *mapping,
2099 struct writeback_control *wbc)
2100{
2101 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2102 struct blk_plug plug;
2103 long diff;
2104
2105 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2106 goto skip_write;
2107
2108 /* balancing f2fs's metadata in background */
2109 f2fs_balance_fs_bg(sbi, true);
2110
2111 /* collect a number of dirty node pages and write together */
2112 if (wbc->sync_mode != WB_SYNC_ALL &&
2113 get_pages(sbi, F2FS_DIRTY_NODES) <
2114 nr_pages_to_skip(sbi, NODE))
2115 goto skip_write;
2116
2117 if (wbc->sync_mode == WB_SYNC_ALL)
2118 atomic_inc(&sbi->wb_sync_req[NODE]);
2119 else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2120 /* to avoid potential deadlock */
2121 if (current->plug)
2122 blk_finish_plug(current->plug);
2123 goto skip_write;
2124 }
2125
2126 trace_f2fs_writepages(mapping->host, wbc, NODE);
2127
2128 diff = nr_pages_to_write(sbi, NODE, wbc);
2129 blk_start_plug(&plug);
2130 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2131 blk_finish_plug(&plug);
2132 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2133
2134 if (wbc->sync_mode == WB_SYNC_ALL)
2135 atomic_dec(&sbi->wb_sync_req[NODE]);
2136 return 0;
2137
2138skip_write:
2139 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2140 trace_f2fs_writepages(mapping->host, wbc, NODE);
2141 return 0;
2142}
2143
2144static bool f2fs_dirty_node_folio(struct address_space *mapping,
2145 struct folio *folio)
2146{
2147 trace_f2fs_set_page_dirty(&folio->page, NODE);
2148
2149 if (!folio_test_uptodate(folio))
2150 folio_mark_uptodate(folio);
2151#ifdef CONFIG_F2FS_CHECK_FS
2152 if (IS_INODE(&folio->page))
2153 f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2154#endif
2155 if (filemap_dirty_folio(mapping, folio)) {
2156 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2157 set_page_private_reference(&folio->page);
2158 return true;
2159 }
2160 return false;
2161}
2162
2163/*
2164 * Structure of the f2fs node operations
2165 */
2166const struct address_space_operations f2fs_node_aops = {
2167 .writepage = f2fs_write_node_page,
2168 .writepages = f2fs_write_node_pages,
2169 .dirty_folio = f2fs_dirty_node_folio,
2170 .invalidate_folio = f2fs_invalidate_folio,
2171 .release_folio = f2fs_release_folio,
2172 .migrate_folio = filemap_migrate_folio,
2173};
2174
2175static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2176 nid_t n)
2177{
2178 return radix_tree_lookup(&nm_i->free_nid_root, n);
2179}
2180
2181static int __insert_free_nid(struct f2fs_sb_info *sbi,
2182 struct free_nid *i)
2183{
2184 struct f2fs_nm_info *nm_i = NM_I(sbi);
2185 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2186
2187 if (err)
2188 return err;
2189
2190 nm_i->nid_cnt[FREE_NID]++;
2191 list_add_tail(&i->list, &nm_i->free_nid_list);
2192 return 0;
2193}
2194
2195static void __remove_free_nid(struct f2fs_sb_info *sbi,
2196 struct free_nid *i, enum nid_state state)
2197{
2198 struct f2fs_nm_info *nm_i = NM_I(sbi);
2199
2200 f2fs_bug_on(sbi, state != i->state);
2201 nm_i->nid_cnt[state]--;
2202 if (state == FREE_NID)
2203 list_del(&i->list);
2204 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2205}
2206
2207static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2208 enum nid_state org_state, enum nid_state dst_state)
2209{
2210 struct f2fs_nm_info *nm_i = NM_I(sbi);
2211
2212 f2fs_bug_on(sbi, org_state != i->state);
2213 i->state = dst_state;
2214 nm_i->nid_cnt[org_state]--;
2215 nm_i->nid_cnt[dst_state]++;
2216
2217 switch (dst_state) {
2218 case PREALLOC_NID:
2219 list_del(&i->list);
2220 break;
2221 case FREE_NID:
2222 list_add_tail(&i->list, &nm_i->free_nid_list);
2223 break;
2224 default:
2225 BUG_ON(1);
2226 }
2227}
2228
2229bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2230{
2231 struct f2fs_nm_info *nm_i = NM_I(sbi);
2232 unsigned int i;
2233 bool ret = true;
2234
2235 f2fs_down_read(&nm_i->nat_tree_lock);
2236 for (i = 0; i < nm_i->nat_blocks; i++) {
2237 if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2238 ret = false;
2239 break;
2240 }
2241 }
2242 f2fs_up_read(&nm_i->nat_tree_lock);
2243
2244 return ret;
2245}
2246
2247static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2248 bool set, bool build)
2249{
2250 struct f2fs_nm_info *nm_i = NM_I(sbi);
2251 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2252 unsigned int nid_ofs = nid - START_NID(nid);
2253
2254 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2255 return;
2256
2257 if (set) {
2258 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2259 return;
2260 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2261 nm_i->free_nid_count[nat_ofs]++;
2262 } else {
2263 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2264 return;
2265 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2266 if (!build)
2267 nm_i->free_nid_count[nat_ofs]--;
2268 }
2269}
2270
2271/* return if the nid is recognized as free */
2272static bool add_free_nid(struct f2fs_sb_info *sbi,
2273 nid_t nid, bool build, bool update)
2274{
2275 struct f2fs_nm_info *nm_i = NM_I(sbi);
2276 struct free_nid *i, *e;
2277 struct nat_entry *ne;
2278 int err = -EINVAL;
2279 bool ret = false;
2280
2281 /* 0 nid should not be used */
2282 if (unlikely(nid == 0))
2283 return false;
2284
2285 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2286 return false;
2287
2288 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2289 i->nid = nid;
2290 i->state = FREE_NID;
2291
2292 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2293
2294 spin_lock(&nm_i->nid_list_lock);
2295
2296 if (build) {
2297 /*
2298 * Thread A Thread B
2299 * - f2fs_create
2300 * - f2fs_new_inode
2301 * - f2fs_alloc_nid
2302 * - __insert_nid_to_list(PREALLOC_NID)
2303 * - f2fs_balance_fs_bg
2304 * - f2fs_build_free_nids
2305 * - __f2fs_build_free_nids
2306 * - scan_nat_page
2307 * - add_free_nid
2308 * - __lookup_nat_cache
2309 * - f2fs_add_link
2310 * - f2fs_init_inode_metadata
2311 * - f2fs_new_inode_page
2312 * - f2fs_new_node_page
2313 * - set_node_addr
2314 * - f2fs_alloc_nid_done
2315 * - __remove_nid_from_list(PREALLOC_NID)
2316 * - __insert_nid_to_list(FREE_NID)
2317 */
2318 ne = __lookup_nat_cache(nm_i, nid);
2319 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2320 nat_get_blkaddr(ne) != NULL_ADDR))
2321 goto err_out;
2322
2323 e = __lookup_free_nid_list(nm_i, nid);
2324 if (e) {
2325 if (e->state == FREE_NID)
2326 ret = true;
2327 goto err_out;
2328 }
2329 }
2330 ret = true;
2331 err = __insert_free_nid(sbi, i);
2332err_out:
2333 if (update) {
2334 update_free_nid_bitmap(sbi, nid, ret, build);
2335 if (!build)
2336 nm_i->available_nids++;
2337 }
2338 spin_unlock(&nm_i->nid_list_lock);
2339 radix_tree_preload_end();
2340
2341 if (err)
2342 kmem_cache_free(free_nid_slab, i);
2343 return ret;
2344}
2345
2346static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2347{
2348 struct f2fs_nm_info *nm_i = NM_I(sbi);
2349 struct free_nid *i;
2350 bool need_free = false;
2351
2352 spin_lock(&nm_i->nid_list_lock);
2353 i = __lookup_free_nid_list(nm_i, nid);
2354 if (i && i->state == FREE_NID) {
2355 __remove_free_nid(sbi, i, FREE_NID);
2356 need_free = true;
2357 }
2358 spin_unlock(&nm_i->nid_list_lock);
2359
2360 if (need_free)
2361 kmem_cache_free(free_nid_slab, i);
2362}
2363
2364static int scan_nat_page(struct f2fs_sb_info *sbi,
2365 struct page *nat_page, nid_t start_nid)
2366{
2367 struct f2fs_nm_info *nm_i = NM_I(sbi);
2368 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2369 block_t blk_addr;
2370 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2371 int i;
2372
2373 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2374
2375 i = start_nid % NAT_ENTRY_PER_BLOCK;
2376
2377 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2378 if (unlikely(start_nid >= nm_i->max_nid))
2379 break;
2380
2381 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2382
2383 if (blk_addr == NEW_ADDR)
2384 return -EINVAL;
2385
2386 if (blk_addr == NULL_ADDR) {
2387 add_free_nid(sbi, start_nid, true, true);
2388 } else {
2389 spin_lock(&NM_I(sbi)->nid_list_lock);
2390 update_free_nid_bitmap(sbi, start_nid, false, true);
2391 spin_unlock(&NM_I(sbi)->nid_list_lock);
2392 }
2393 }
2394
2395 return 0;
2396}
2397
2398static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2399{
2400 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2401 struct f2fs_journal *journal = curseg->journal;
2402 int i;
2403
2404 down_read(&curseg->journal_rwsem);
2405 for (i = 0; i < nats_in_cursum(journal); i++) {
2406 block_t addr;
2407 nid_t nid;
2408
2409 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2410 nid = le32_to_cpu(nid_in_journal(journal, i));
2411 if (addr == NULL_ADDR)
2412 add_free_nid(sbi, nid, true, false);
2413 else
2414 remove_free_nid(sbi, nid);
2415 }
2416 up_read(&curseg->journal_rwsem);
2417}
2418
2419static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2420{
2421 struct f2fs_nm_info *nm_i = NM_I(sbi);
2422 unsigned int i, idx;
2423 nid_t nid;
2424
2425 f2fs_down_read(&nm_i->nat_tree_lock);
2426
2427 for (i = 0; i < nm_i->nat_blocks; i++) {
2428 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2429 continue;
2430 if (!nm_i->free_nid_count[i])
2431 continue;
2432 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2433 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2434 NAT_ENTRY_PER_BLOCK, idx);
2435 if (idx >= NAT_ENTRY_PER_BLOCK)
2436 break;
2437
2438 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2439 add_free_nid(sbi, nid, true, false);
2440
2441 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2442 goto out;
2443 }
2444 }
2445out:
2446 scan_curseg_cache(sbi);
2447
2448 f2fs_up_read(&nm_i->nat_tree_lock);
2449}
2450
2451static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2452 bool sync, bool mount)
2453{
2454 struct f2fs_nm_info *nm_i = NM_I(sbi);
2455 int i = 0, ret;
2456 nid_t nid = nm_i->next_scan_nid;
2457
2458 if (unlikely(nid >= nm_i->max_nid))
2459 nid = 0;
2460
2461 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2462 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2463
2464 /* Enough entries */
2465 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2466 return 0;
2467
2468 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2469 return 0;
2470
2471 if (!mount) {
2472 /* try to find free nids in free_nid_bitmap */
2473 scan_free_nid_bits(sbi);
2474
2475 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2476 return 0;
2477 }
2478
2479 /* readahead nat pages to be scanned */
2480 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2481 META_NAT, true);
2482
2483 f2fs_down_read(&nm_i->nat_tree_lock);
2484
2485 while (1) {
2486 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2487 nm_i->nat_block_bitmap)) {
2488 struct page *page = get_current_nat_page(sbi, nid);
2489
2490 if (IS_ERR(page)) {
2491 ret = PTR_ERR(page);
2492 } else {
2493 ret = scan_nat_page(sbi, page, nid);
2494 f2fs_put_page(page, 1);
2495 }
2496
2497 if (ret) {
2498 f2fs_up_read(&nm_i->nat_tree_lock);
2499 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2500 return ret;
2501 }
2502 }
2503
2504 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2505 if (unlikely(nid >= nm_i->max_nid))
2506 nid = 0;
2507
2508 if (++i >= FREE_NID_PAGES)
2509 break;
2510 }
2511
2512 /* go to the next free nat pages to find free nids abundantly */
2513 nm_i->next_scan_nid = nid;
2514
2515 /* find free nids from current sum_pages */
2516 scan_curseg_cache(sbi);
2517
2518 f2fs_up_read(&nm_i->nat_tree_lock);
2519
2520 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2521 nm_i->ra_nid_pages, META_NAT, false);
2522
2523 return 0;
2524}
2525
2526int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2527{
2528 int ret;
2529
2530 mutex_lock(&NM_I(sbi)->build_lock);
2531 ret = __f2fs_build_free_nids(sbi, sync, mount);
2532 mutex_unlock(&NM_I(sbi)->build_lock);
2533
2534 return ret;
2535}
2536
2537/*
2538 * If this function returns success, caller can obtain a new nid
2539 * from second parameter of this function.
2540 * The returned nid could be used ino as well as nid when inode is created.
2541 */
2542bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2543{
2544 struct f2fs_nm_info *nm_i = NM_I(sbi);
2545 struct free_nid *i = NULL;
2546retry:
2547 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2548 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2549 return false;
2550 }
2551
2552 spin_lock(&nm_i->nid_list_lock);
2553
2554 if (unlikely(nm_i->available_nids == 0)) {
2555 spin_unlock(&nm_i->nid_list_lock);
2556 return false;
2557 }
2558
2559 /* We should not use stale free nids created by f2fs_build_free_nids */
2560 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2561 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2562 i = list_first_entry(&nm_i->free_nid_list,
2563 struct free_nid, list);
2564 *nid = i->nid;
2565
2566 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2567 nm_i->available_nids--;
2568
2569 update_free_nid_bitmap(sbi, *nid, false, false);
2570
2571 spin_unlock(&nm_i->nid_list_lock);
2572 return true;
2573 }
2574 spin_unlock(&nm_i->nid_list_lock);
2575
2576 /* Let's scan nat pages and its caches to get free nids */
2577 if (!f2fs_build_free_nids(sbi, true, false))
2578 goto retry;
2579 return false;
2580}
2581
2582/*
2583 * f2fs_alloc_nid() should be called prior to this function.
2584 */
2585void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2586{
2587 struct f2fs_nm_info *nm_i = NM_I(sbi);
2588 struct free_nid *i;
2589
2590 spin_lock(&nm_i->nid_list_lock);
2591 i = __lookup_free_nid_list(nm_i, nid);
2592 f2fs_bug_on(sbi, !i);
2593 __remove_free_nid(sbi, i, PREALLOC_NID);
2594 spin_unlock(&nm_i->nid_list_lock);
2595
2596 kmem_cache_free(free_nid_slab, i);
2597}
2598
2599/*
2600 * f2fs_alloc_nid() should be called prior to this function.
2601 */
2602void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2603{
2604 struct f2fs_nm_info *nm_i = NM_I(sbi);
2605 struct free_nid *i;
2606 bool need_free = false;
2607
2608 if (!nid)
2609 return;
2610
2611 spin_lock(&nm_i->nid_list_lock);
2612 i = __lookup_free_nid_list(nm_i, nid);
2613 f2fs_bug_on(sbi, !i);
2614
2615 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2616 __remove_free_nid(sbi, i, PREALLOC_NID);
2617 need_free = true;
2618 } else {
2619 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2620 }
2621
2622 nm_i->available_nids++;
2623
2624 update_free_nid_bitmap(sbi, nid, true, false);
2625
2626 spin_unlock(&nm_i->nid_list_lock);
2627
2628 if (need_free)
2629 kmem_cache_free(free_nid_slab, i);
2630}
2631
2632int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2633{
2634 struct f2fs_nm_info *nm_i = NM_I(sbi);
2635 int nr = nr_shrink;
2636
2637 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2638 return 0;
2639
2640 if (!mutex_trylock(&nm_i->build_lock))
2641 return 0;
2642
2643 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2644 struct free_nid *i, *next;
2645 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2646
2647 spin_lock(&nm_i->nid_list_lock);
2648 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2649 if (!nr_shrink || !batch ||
2650 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2651 break;
2652 __remove_free_nid(sbi, i, FREE_NID);
2653 kmem_cache_free(free_nid_slab, i);
2654 nr_shrink--;
2655 batch--;
2656 }
2657 spin_unlock(&nm_i->nid_list_lock);
2658 }
2659
2660 mutex_unlock(&nm_i->build_lock);
2661
2662 return nr - nr_shrink;
2663}
2664
2665int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2666{
2667 void *src_addr, *dst_addr;
2668 size_t inline_size;
2669 struct page *ipage;
2670 struct f2fs_inode *ri;
2671
2672 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2673 if (IS_ERR(ipage))
2674 return PTR_ERR(ipage);
2675
2676 ri = F2FS_INODE(page);
2677 if (ri->i_inline & F2FS_INLINE_XATTR) {
2678 if (!f2fs_has_inline_xattr(inode)) {
2679 set_inode_flag(inode, FI_INLINE_XATTR);
2680 stat_inc_inline_xattr(inode);
2681 }
2682 } else {
2683 if (f2fs_has_inline_xattr(inode)) {
2684 stat_dec_inline_xattr(inode);
2685 clear_inode_flag(inode, FI_INLINE_XATTR);
2686 }
2687 goto update_inode;
2688 }
2689
2690 dst_addr = inline_xattr_addr(inode, ipage);
2691 src_addr = inline_xattr_addr(inode, page);
2692 inline_size = inline_xattr_size(inode);
2693
2694 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2695 memcpy(dst_addr, src_addr, inline_size);
2696update_inode:
2697 f2fs_update_inode(inode, ipage);
2698 f2fs_put_page(ipage, 1);
2699 return 0;
2700}
2701
2702int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2703{
2704 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2705 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2706 nid_t new_xnid;
2707 struct dnode_of_data dn;
2708 struct node_info ni;
2709 struct page *xpage;
2710 int err;
2711
2712 if (!prev_xnid)
2713 goto recover_xnid;
2714
2715 /* 1: invalidate the previous xattr nid */
2716 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2717 if (err)
2718 return err;
2719
2720 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2721 dec_valid_node_count(sbi, inode, false);
2722 set_node_addr(sbi, &ni, NULL_ADDR, false);
2723
2724recover_xnid:
2725 /* 2: update xattr nid in inode */
2726 if (!f2fs_alloc_nid(sbi, &new_xnid))
2727 return -ENOSPC;
2728
2729 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2730 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2731 if (IS_ERR(xpage)) {
2732 f2fs_alloc_nid_failed(sbi, new_xnid);
2733 return PTR_ERR(xpage);
2734 }
2735
2736 f2fs_alloc_nid_done(sbi, new_xnid);
2737 f2fs_update_inode_page(inode);
2738
2739 /* 3: update and set xattr node page dirty */
2740 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2741
2742 set_page_dirty(xpage);
2743 f2fs_put_page(xpage, 1);
2744
2745 return 0;
2746}
2747
2748int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2749{
2750 struct f2fs_inode *src, *dst;
2751 nid_t ino = ino_of_node(page);
2752 struct node_info old_ni, new_ni;
2753 struct page *ipage;
2754 int err;
2755
2756 err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2757 if (err)
2758 return err;
2759
2760 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2761 return -EINVAL;
2762retry:
2763 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2764 if (!ipage) {
2765 memalloc_retry_wait(GFP_NOFS);
2766 goto retry;
2767 }
2768
2769 /* Should not use this inode from free nid list */
2770 remove_free_nid(sbi, ino);
2771
2772 if (!PageUptodate(ipage))
2773 SetPageUptodate(ipage);
2774 fill_node_footer(ipage, ino, ino, 0, true);
2775 set_cold_node(ipage, false);
2776
2777 src = F2FS_INODE(page);
2778 dst = F2FS_INODE(ipage);
2779
2780 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2781 dst->i_size = 0;
2782 dst->i_blocks = cpu_to_le64(1);
2783 dst->i_links = cpu_to_le32(1);
2784 dst->i_xattr_nid = 0;
2785 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2786 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2787 dst->i_extra_isize = src->i_extra_isize;
2788
2789 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2790 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2791 i_inline_xattr_size))
2792 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2793
2794 if (f2fs_sb_has_project_quota(sbi) &&
2795 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2796 i_projid))
2797 dst->i_projid = src->i_projid;
2798
2799 if (f2fs_sb_has_inode_crtime(sbi) &&
2800 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2801 i_crtime_nsec)) {
2802 dst->i_crtime = src->i_crtime;
2803 dst->i_crtime_nsec = src->i_crtime_nsec;
2804 }
2805 }
2806
2807 new_ni = old_ni;
2808 new_ni.ino = ino;
2809
2810 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2811 WARN_ON(1);
2812 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2813 inc_valid_inode_count(sbi);
2814 set_page_dirty(ipage);
2815 f2fs_put_page(ipage, 1);
2816 return 0;
2817}
2818
2819int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2820 unsigned int segno, struct f2fs_summary_block *sum)
2821{
2822 struct f2fs_node *rn;
2823 struct f2fs_summary *sum_entry;
2824 block_t addr;
2825 int i, idx, last_offset, nrpages;
2826
2827 /* scan the node segment */
2828 last_offset = sbi->blocks_per_seg;
2829 addr = START_BLOCK(sbi, segno);
2830 sum_entry = &sum->entries[0];
2831
2832 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2833 nrpages = bio_max_segs(last_offset - i);
2834
2835 /* readahead node pages */
2836 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2837
2838 for (idx = addr; idx < addr + nrpages; idx++) {
2839 struct page *page = f2fs_get_tmp_page(sbi, idx);
2840
2841 if (IS_ERR(page))
2842 return PTR_ERR(page);
2843
2844 rn = F2FS_NODE(page);
2845 sum_entry->nid = rn->footer.nid;
2846 sum_entry->version = 0;
2847 sum_entry->ofs_in_node = 0;
2848 sum_entry++;
2849 f2fs_put_page(page, 1);
2850 }
2851
2852 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2853 addr + nrpages);
2854 }
2855 return 0;
2856}
2857
2858static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2859{
2860 struct f2fs_nm_info *nm_i = NM_I(sbi);
2861 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2862 struct f2fs_journal *journal = curseg->journal;
2863 int i;
2864
2865 down_write(&curseg->journal_rwsem);
2866 for (i = 0; i < nats_in_cursum(journal); i++) {
2867 struct nat_entry *ne;
2868 struct f2fs_nat_entry raw_ne;
2869 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2870
2871 if (f2fs_check_nid_range(sbi, nid))
2872 continue;
2873
2874 raw_ne = nat_in_journal(journal, i);
2875
2876 ne = __lookup_nat_cache(nm_i, nid);
2877 if (!ne) {
2878 ne = __alloc_nat_entry(sbi, nid, true);
2879 __init_nat_entry(nm_i, ne, &raw_ne, true);
2880 }
2881
2882 /*
2883 * if a free nat in journal has not been used after last
2884 * checkpoint, we should remove it from available nids,
2885 * since later we will add it again.
2886 */
2887 if (!get_nat_flag(ne, IS_DIRTY) &&
2888 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2889 spin_lock(&nm_i->nid_list_lock);
2890 nm_i->available_nids--;
2891 spin_unlock(&nm_i->nid_list_lock);
2892 }
2893
2894 __set_nat_cache_dirty(nm_i, ne);
2895 }
2896 update_nats_in_cursum(journal, -i);
2897 up_write(&curseg->journal_rwsem);
2898}
2899
2900static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2901 struct list_head *head, int max)
2902{
2903 struct nat_entry_set *cur;
2904
2905 if (nes->entry_cnt >= max)
2906 goto add_out;
2907
2908 list_for_each_entry(cur, head, set_list) {
2909 if (cur->entry_cnt >= nes->entry_cnt) {
2910 list_add(&nes->set_list, cur->set_list.prev);
2911 return;
2912 }
2913 }
2914add_out:
2915 list_add_tail(&nes->set_list, head);
2916}
2917
2918static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2919 unsigned int valid)
2920{
2921 if (valid == 0) {
2922 __set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2923 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2924 return;
2925 }
2926
2927 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2928 if (valid == NAT_ENTRY_PER_BLOCK)
2929 __set_bit_le(nat_ofs, nm_i->full_nat_bits);
2930 else
2931 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2932}
2933
2934static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2935 struct page *page)
2936{
2937 struct f2fs_nm_info *nm_i = NM_I(sbi);
2938 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2939 struct f2fs_nat_block *nat_blk = page_address(page);
2940 int valid = 0;
2941 int i = 0;
2942
2943 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2944 return;
2945
2946 if (nat_index == 0) {
2947 valid = 1;
2948 i = 1;
2949 }
2950 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2951 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2952 valid++;
2953 }
2954
2955 __update_nat_bits(nm_i, nat_index, valid);
2956}
2957
2958void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2959{
2960 struct f2fs_nm_info *nm_i = NM_I(sbi);
2961 unsigned int nat_ofs;
2962
2963 f2fs_down_read(&nm_i->nat_tree_lock);
2964
2965 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2966 unsigned int valid = 0, nid_ofs = 0;
2967
2968 /* handle nid zero due to it should never be used */
2969 if (unlikely(nat_ofs == 0)) {
2970 valid = 1;
2971 nid_ofs = 1;
2972 }
2973
2974 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2975 if (!test_bit_le(nid_ofs,
2976 nm_i->free_nid_bitmap[nat_ofs]))
2977 valid++;
2978 }
2979
2980 __update_nat_bits(nm_i, nat_ofs, valid);
2981 }
2982
2983 f2fs_up_read(&nm_i->nat_tree_lock);
2984}
2985
2986static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2987 struct nat_entry_set *set, struct cp_control *cpc)
2988{
2989 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2990 struct f2fs_journal *journal = curseg->journal;
2991 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2992 bool to_journal = true;
2993 struct f2fs_nat_block *nat_blk;
2994 struct nat_entry *ne, *cur;
2995 struct page *page = NULL;
2996
2997 /*
2998 * there are two steps to flush nat entries:
2999 * #1, flush nat entries to journal in current hot data summary block.
3000 * #2, flush nat entries to nat page.
3001 */
3002 if ((cpc->reason & CP_UMOUNT) ||
3003 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3004 to_journal = false;
3005
3006 if (to_journal) {
3007 down_write(&curseg->journal_rwsem);
3008 } else {
3009 page = get_next_nat_page(sbi, start_nid);
3010 if (IS_ERR(page))
3011 return PTR_ERR(page);
3012
3013 nat_blk = page_address(page);
3014 f2fs_bug_on(sbi, !nat_blk);
3015 }
3016
3017 /* flush dirty nats in nat entry set */
3018 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3019 struct f2fs_nat_entry *raw_ne;
3020 nid_t nid = nat_get_nid(ne);
3021 int offset;
3022
3023 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3024
3025 if (to_journal) {
3026 offset = f2fs_lookup_journal_in_cursum(journal,
3027 NAT_JOURNAL, nid, 1);
3028 f2fs_bug_on(sbi, offset < 0);
3029 raw_ne = &nat_in_journal(journal, offset);
3030 nid_in_journal(journal, offset) = cpu_to_le32(nid);
3031 } else {
3032 raw_ne = &nat_blk->entries[nid - start_nid];
3033 }
3034 raw_nat_from_node_info(raw_ne, &ne->ni);
3035 nat_reset_flag(ne);
3036 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3037 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3038 add_free_nid(sbi, nid, false, true);
3039 } else {
3040 spin_lock(&NM_I(sbi)->nid_list_lock);
3041 update_free_nid_bitmap(sbi, nid, false, false);
3042 spin_unlock(&NM_I(sbi)->nid_list_lock);
3043 }
3044 }
3045
3046 if (to_journal) {
3047 up_write(&curseg->journal_rwsem);
3048 } else {
3049 update_nat_bits(sbi, start_nid, page);
3050 f2fs_put_page(page, 1);
3051 }
3052
3053 /* Allow dirty nats by node block allocation in write_begin */
3054 if (!set->entry_cnt) {
3055 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3056 kmem_cache_free(nat_entry_set_slab, set);
3057 }
3058 return 0;
3059}
3060
3061/*
3062 * This function is called during the checkpointing process.
3063 */
3064int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3065{
3066 struct f2fs_nm_info *nm_i = NM_I(sbi);
3067 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3068 struct f2fs_journal *journal = curseg->journal;
3069 struct nat_entry_set *setvec[SETVEC_SIZE];
3070 struct nat_entry_set *set, *tmp;
3071 unsigned int found;
3072 nid_t set_idx = 0;
3073 LIST_HEAD(sets);
3074 int err = 0;
3075
3076 /*
3077 * during unmount, let's flush nat_bits before checking
3078 * nat_cnt[DIRTY_NAT].
3079 */
3080 if (cpc->reason & CP_UMOUNT) {
3081 f2fs_down_write(&nm_i->nat_tree_lock);
3082 remove_nats_in_journal(sbi);
3083 f2fs_up_write(&nm_i->nat_tree_lock);
3084 }
3085
3086 if (!nm_i->nat_cnt[DIRTY_NAT])
3087 return 0;
3088
3089 f2fs_down_write(&nm_i->nat_tree_lock);
3090
3091 /*
3092 * if there are no enough space in journal to store dirty nat
3093 * entries, remove all entries from journal and merge them
3094 * into nat entry set.
3095 */
3096 if (cpc->reason & CP_UMOUNT ||
3097 !__has_cursum_space(journal,
3098 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3099 remove_nats_in_journal(sbi);
3100
3101 while ((found = __gang_lookup_nat_set(nm_i,
3102 set_idx, SETVEC_SIZE, setvec))) {
3103 unsigned idx;
3104
3105 set_idx = setvec[found - 1]->set + 1;
3106 for (idx = 0; idx < found; idx++)
3107 __adjust_nat_entry_set(setvec[idx], &sets,
3108 MAX_NAT_JENTRIES(journal));
3109 }
3110
3111 /* flush dirty nats in nat entry set */
3112 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3113 err = __flush_nat_entry_set(sbi, set, cpc);
3114 if (err)
3115 break;
3116 }
3117
3118 f2fs_up_write(&nm_i->nat_tree_lock);
3119 /* Allow dirty nats by node block allocation in write_begin */
3120
3121 return err;
3122}
3123
3124static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3125{
3126 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3127 struct f2fs_nm_info *nm_i = NM_I(sbi);
3128 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3129 unsigned int i;
3130 __u64 cp_ver = cur_cp_version(ckpt);
3131 block_t nat_bits_addr;
3132
3133 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3134 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3135 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3136 if (!nm_i->nat_bits)
3137 return -ENOMEM;
3138
3139 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3140 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3141
3142 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3143 return 0;
3144
3145 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3146 nm_i->nat_bits_blocks;
3147 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3148 struct page *page;
3149
3150 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3151 if (IS_ERR(page))
3152 return PTR_ERR(page);
3153
3154 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3155 page_address(page), F2FS_BLKSIZE);
3156 f2fs_put_page(page, 1);
3157 }
3158
3159 cp_ver |= (cur_cp_crc(ckpt) << 32);
3160 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3161 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3162 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3163 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3164 return 0;
3165 }
3166
3167 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3168 return 0;
3169}
3170
3171static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3172{
3173 struct f2fs_nm_info *nm_i = NM_I(sbi);
3174 unsigned int i = 0;
3175 nid_t nid, last_nid;
3176
3177 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3178 return;
3179
3180 for (i = 0; i < nm_i->nat_blocks; i++) {
3181 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3182 if (i >= nm_i->nat_blocks)
3183 break;
3184
3185 __set_bit_le(i, nm_i->nat_block_bitmap);
3186
3187 nid = i * NAT_ENTRY_PER_BLOCK;
3188 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3189
3190 spin_lock(&NM_I(sbi)->nid_list_lock);
3191 for (; nid < last_nid; nid++)
3192 update_free_nid_bitmap(sbi, nid, true, true);
3193 spin_unlock(&NM_I(sbi)->nid_list_lock);
3194 }
3195
3196 for (i = 0; i < nm_i->nat_blocks; i++) {
3197 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3198 if (i >= nm_i->nat_blocks)
3199 break;
3200
3201 __set_bit_le(i, nm_i->nat_block_bitmap);
3202 }
3203}
3204
3205static int init_node_manager(struct f2fs_sb_info *sbi)
3206{
3207 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3208 struct f2fs_nm_info *nm_i = NM_I(sbi);
3209 unsigned char *version_bitmap;
3210 unsigned int nat_segs;
3211 int err;
3212
3213 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3214
3215 /* segment_count_nat includes pair segment so divide to 2. */
3216 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3217 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3218 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3219
3220 /* not used nids: 0, node, meta, (and root counted as valid node) */
3221 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3222 F2FS_RESERVED_NODE_NUM;
3223 nm_i->nid_cnt[FREE_NID] = 0;
3224 nm_i->nid_cnt[PREALLOC_NID] = 0;
3225 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3226 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3227 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3228 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3229
3230 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3231 INIT_LIST_HEAD(&nm_i->free_nid_list);
3232 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3233 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3234 INIT_LIST_HEAD(&nm_i->nat_entries);
3235 spin_lock_init(&nm_i->nat_list_lock);
3236
3237 mutex_init(&nm_i->build_lock);
3238 spin_lock_init(&nm_i->nid_list_lock);
3239 init_f2fs_rwsem(&nm_i->nat_tree_lock);
3240
3241 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3242 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3243 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3244 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3245 GFP_KERNEL);
3246 if (!nm_i->nat_bitmap)
3247 return -ENOMEM;
3248
3249 err = __get_nat_bitmaps(sbi);
3250 if (err)
3251 return err;
3252
3253#ifdef CONFIG_F2FS_CHECK_FS
3254 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3255 GFP_KERNEL);
3256 if (!nm_i->nat_bitmap_mir)
3257 return -ENOMEM;
3258#endif
3259
3260 return 0;
3261}
3262
3263static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3264{
3265 struct f2fs_nm_info *nm_i = NM_I(sbi);
3266 int i;
3267
3268 nm_i->free_nid_bitmap =
3269 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3270 nm_i->nat_blocks),
3271 GFP_KERNEL);
3272 if (!nm_i->free_nid_bitmap)
3273 return -ENOMEM;
3274
3275 for (i = 0; i < nm_i->nat_blocks; i++) {
3276 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3277 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3278 if (!nm_i->free_nid_bitmap[i])
3279 return -ENOMEM;
3280 }
3281
3282 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3283 GFP_KERNEL);
3284 if (!nm_i->nat_block_bitmap)
3285 return -ENOMEM;
3286
3287 nm_i->free_nid_count =
3288 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3289 nm_i->nat_blocks),
3290 GFP_KERNEL);
3291 if (!nm_i->free_nid_count)
3292 return -ENOMEM;
3293 return 0;
3294}
3295
3296int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3297{
3298 int err;
3299
3300 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3301 GFP_KERNEL);
3302 if (!sbi->nm_info)
3303 return -ENOMEM;
3304
3305 err = init_node_manager(sbi);
3306 if (err)
3307 return err;
3308
3309 err = init_free_nid_cache(sbi);
3310 if (err)
3311 return err;
3312
3313 /* load free nid status from nat_bits table */
3314 load_free_nid_bitmap(sbi);
3315
3316 return f2fs_build_free_nids(sbi, true, true);
3317}
3318
3319void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3320{
3321 struct f2fs_nm_info *nm_i = NM_I(sbi);
3322 struct free_nid *i, *next_i;
3323 struct nat_entry *natvec[NATVEC_SIZE];
3324 struct nat_entry_set *setvec[SETVEC_SIZE];
3325 nid_t nid = 0;
3326 unsigned int found;
3327
3328 if (!nm_i)
3329 return;
3330
3331 /* destroy free nid list */
3332 spin_lock(&nm_i->nid_list_lock);
3333 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3334 __remove_free_nid(sbi, i, FREE_NID);
3335 spin_unlock(&nm_i->nid_list_lock);
3336 kmem_cache_free(free_nid_slab, i);
3337 spin_lock(&nm_i->nid_list_lock);
3338 }
3339 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3340 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3341 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3342 spin_unlock(&nm_i->nid_list_lock);
3343
3344 /* destroy nat cache */
3345 f2fs_down_write(&nm_i->nat_tree_lock);
3346 while ((found = __gang_lookup_nat_cache(nm_i,
3347 nid, NATVEC_SIZE, natvec))) {
3348 unsigned idx;
3349
3350 nid = nat_get_nid(natvec[found - 1]) + 1;
3351 for (idx = 0; idx < found; idx++) {
3352 spin_lock(&nm_i->nat_list_lock);
3353 list_del(&natvec[idx]->list);
3354 spin_unlock(&nm_i->nat_list_lock);
3355
3356 __del_from_nat_cache(nm_i, natvec[idx]);
3357 }
3358 }
3359 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3360
3361 /* destroy nat set cache */
3362 nid = 0;
3363 while ((found = __gang_lookup_nat_set(nm_i,
3364 nid, SETVEC_SIZE, setvec))) {
3365 unsigned idx;
3366
3367 nid = setvec[found - 1]->set + 1;
3368 for (idx = 0; idx < found; idx++) {
3369 /* entry_cnt is not zero, when cp_error was occurred */
3370 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3371 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3372 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3373 }
3374 }
3375 f2fs_up_write(&nm_i->nat_tree_lock);
3376
3377 kvfree(nm_i->nat_block_bitmap);
3378 if (nm_i->free_nid_bitmap) {
3379 int i;
3380
3381 for (i = 0; i < nm_i->nat_blocks; i++)
3382 kvfree(nm_i->free_nid_bitmap[i]);
3383 kvfree(nm_i->free_nid_bitmap);
3384 }
3385 kvfree(nm_i->free_nid_count);
3386
3387 kvfree(nm_i->nat_bitmap);
3388 kvfree(nm_i->nat_bits);
3389#ifdef CONFIG_F2FS_CHECK_FS
3390 kvfree(nm_i->nat_bitmap_mir);
3391#endif
3392 sbi->nm_info = NULL;
3393 kfree(nm_i);
3394}
3395
3396int __init f2fs_create_node_manager_caches(void)
3397{
3398 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3399 sizeof(struct nat_entry));
3400 if (!nat_entry_slab)
3401 goto fail;
3402
3403 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3404 sizeof(struct free_nid));
3405 if (!free_nid_slab)
3406 goto destroy_nat_entry;
3407
3408 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3409 sizeof(struct nat_entry_set));
3410 if (!nat_entry_set_slab)
3411 goto destroy_free_nid;
3412
3413 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3414 sizeof(struct fsync_node_entry));
3415 if (!fsync_node_entry_slab)
3416 goto destroy_nat_entry_set;
3417 return 0;
3418
3419destroy_nat_entry_set:
3420 kmem_cache_destroy(nat_entry_set_slab);
3421destroy_free_nid:
3422 kmem_cache_destroy(free_nid_slab);
3423destroy_nat_entry:
3424 kmem_cache_destroy(nat_entry_slab);
3425fail:
3426 return -ENOMEM;
3427}
3428
3429void f2fs_destroy_node_manager_caches(void)
3430{
3431 kmem_cache_destroy(fsync_node_entry_slab);
3432 kmem_cache_destroy(nat_entry_set_slab);
3433 kmem_cache_destroy(free_nid_slab);
3434 kmem_cache_destroy(nat_entry_slab);
3435}
1/*
2 * fs/f2fs/node.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/fs.h>
12#include <linux/f2fs_fs.h>
13#include <linux/mpage.h>
14#include <linux/backing-dev.h>
15#include <linux/blkdev.h>
16#include <linux/pagevec.h>
17#include <linux/swap.h>
18
19#include "f2fs.h"
20#include "node.h"
21#include "segment.h"
22#include "trace.h"
23#include <trace/events/f2fs.h>
24
25#define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26
27static struct kmem_cache *nat_entry_slab;
28static struct kmem_cache *free_nid_slab;
29static struct kmem_cache *nat_entry_set_slab;
30
31bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32{
33 struct f2fs_nm_info *nm_i = NM_I(sbi);
34 struct sysinfo val;
35 unsigned long avail_ram;
36 unsigned long mem_size = 0;
37 bool res = false;
38
39 si_meminfo(&val);
40
41 /* only uses low memory */
42 avail_ram = val.totalram - val.totalhigh;
43
44 /*
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
46 */
47 if (type == FREE_NIDS) {
48 mem_size = (nm_i->nid_cnt[FREE_NID_LIST] *
49 sizeof(struct free_nid)) >> PAGE_SHIFT;
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 PAGE_SHIFT;
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 if (excess_cached_nats(sbi))
56 res = false;
57 } else if (type == DIRTY_DENTS) {
58 if (sbi->sb->s_bdi->wb.dirty_exceeded)
59 return false;
60 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
61 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
62 } else if (type == INO_ENTRIES) {
63 int i;
64
65 for (i = 0; i <= UPDATE_INO; i++)
66 mem_size += (sbi->im[i].ino_num *
67 sizeof(struct ino_entry)) >> PAGE_SHIFT;
68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
69 } else if (type == EXTENT_CACHE) {
70 mem_size = (atomic_read(&sbi->total_ext_tree) *
71 sizeof(struct extent_tree) +
72 atomic_read(&sbi->total_ext_node) *
73 sizeof(struct extent_node)) >> PAGE_SHIFT;
74 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
75 } else {
76 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
77 return true;
78 }
79 return res;
80}
81
82static void clear_node_page_dirty(struct page *page)
83{
84 struct address_space *mapping = page->mapping;
85 unsigned int long flags;
86
87 if (PageDirty(page)) {
88 spin_lock_irqsave(&mapping->tree_lock, flags);
89 radix_tree_tag_clear(&mapping->page_tree,
90 page_index(page),
91 PAGECACHE_TAG_DIRTY);
92 spin_unlock_irqrestore(&mapping->tree_lock, flags);
93
94 clear_page_dirty_for_io(page);
95 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
96 }
97 ClearPageUptodate(page);
98}
99
100static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
101{
102 pgoff_t index = current_nat_addr(sbi, nid);
103 return get_meta_page(sbi, index);
104}
105
106static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
107{
108 struct page *src_page;
109 struct page *dst_page;
110 pgoff_t src_off;
111 pgoff_t dst_off;
112 void *src_addr;
113 void *dst_addr;
114 struct f2fs_nm_info *nm_i = NM_I(sbi);
115
116 src_off = current_nat_addr(sbi, nid);
117 dst_off = next_nat_addr(sbi, src_off);
118
119 /* get current nat block page with lock */
120 src_page = get_meta_page(sbi, src_off);
121 dst_page = grab_meta_page(sbi, dst_off);
122 f2fs_bug_on(sbi, PageDirty(src_page));
123
124 src_addr = page_address(src_page);
125 dst_addr = page_address(dst_page);
126 memcpy(dst_addr, src_addr, PAGE_SIZE);
127 set_page_dirty(dst_page);
128 f2fs_put_page(src_page, 1);
129
130 set_to_next_nat(nm_i, nid);
131
132 return dst_page;
133}
134
135static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
136{
137 return radix_tree_lookup(&nm_i->nat_root, n);
138}
139
140static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
141 nid_t start, unsigned int nr, struct nat_entry **ep)
142{
143 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
144}
145
146static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
147{
148 list_del(&e->list);
149 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
150 nm_i->nat_cnt--;
151 kmem_cache_free(nat_entry_slab, e);
152}
153
154static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
155 struct nat_entry *ne)
156{
157 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
158 struct nat_entry_set *head;
159
160 if (get_nat_flag(ne, IS_DIRTY))
161 return;
162
163 head = radix_tree_lookup(&nm_i->nat_set_root, set);
164 if (!head) {
165 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
166
167 INIT_LIST_HEAD(&head->entry_list);
168 INIT_LIST_HEAD(&head->set_list);
169 head->set = set;
170 head->entry_cnt = 0;
171 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
172 }
173 list_move_tail(&ne->list, &head->entry_list);
174 nm_i->dirty_nat_cnt++;
175 head->entry_cnt++;
176 set_nat_flag(ne, IS_DIRTY, true);
177}
178
179static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
180 struct nat_entry *ne)
181{
182 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
183 struct nat_entry_set *head;
184
185 head = radix_tree_lookup(&nm_i->nat_set_root, set);
186 if (head) {
187 list_move_tail(&ne->list, &nm_i->nat_entries);
188 set_nat_flag(ne, IS_DIRTY, false);
189 head->entry_cnt--;
190 nm_i->dirty_nat_cnt--;
191 }
192}
193
194static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
195 nid_t start, unsigned int nr, struct nat_entry_set **ep)
196{
197 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
198 start, nr);
199}
200
201int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
202{
203 struct f2fs_nm_info *nm_i = NM_I(sbi);
204 struct nat_entry *e;
205 bool need = false;
206
207 down_read(&nm_i->nat_tree_lock);
208 e = __lookup_nat_cache(nm_i, nid);
209 if (e) {
210 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
211 !get_nat_flag(e, HAS_FSYNCED_INODE))
212 need = true;
213 }
214 up_read(&nm_i->nat_tree_lock);
215 return need;
216}
217
218bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
219{
220 struct f2fs_nm_info *nm_i = NM_I(sbi);
221 struct nat_entry *e;
222 bool is_cp = true;
223
224 down_read(&nm_i->nat_tree_lock);
225 e = __lookup_nat_cache(nm_i, nid);
226 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
227 is_cp = false;
228 up_read(&nm_i->nat_tree_lock);
229 return is_cp;
230}
231
232bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
233{
234 struct f2fs_nm_info *nm_i = NM_I(sbi);
235 struct nat_entry *e;
236 bool need_update = true;
237
238 down_read(&nm_i->nat_tree_lock);
239 e = __lookup_nat_cache(nm_i, ino);
240 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
241 (get_nat_flag(e, IS_CHECKPOINTED) ||
242 get_nat_flag(e, HAS_FSYNCED_INODE)))
243 need_update = false;
244 up_read(&nm_i->nat_tree_lock);
245 return need_update;
246}
247
248static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
249{
250 struct nat_entry *new;
251
252 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
253 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
254 memset(new, 0, sizeof(struct nat_entry));
255 nat_set_nid(new, nid);
256 nat_reset_flag(new);
257 list_add_tail(&new->list, &nm_i->nat_entries);
258 nm_i->nat_cnt++;
259 return new;
260}
261
262static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
263 struct f2fs_nat_entry *ne)
264{
265 struct f2fs_nm_info *nm_i = NM_I(sbi);
266 struct nat_entry *e;
267
268 e = __lookup_nat_cache(nm_i, nid);
269 if (!e) {
270 e = grab_nat_entry(nm_i, nid);
271 node_info_from_raw_nat(&e->ni, ne);
272 } else {
273 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
274 nat_get_blkaddr(e) !=
275 le32_to_cpu(ne->block_addr) ||
276 nat_get_version(e) != ne->version);
277 }
278}
279
280static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
281 block_t new_blkaddr, bool fsync_done)
282{
283 struct f2fs_nm_info *nm_i = NM_I(sbi);
284 struct nat_entry *e;
285
286 down_write(&nm_i->nat_tree_lock);
287 e = __lookup_nat_cache(nm_i, ni->nid);
288 if (!e) {
289 e = grab_nat_entry(nm_i, ni->nid);
290 copy_node_info(&e->ni, ni);
291 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
292 } else if (new_blkaddr == NEW_ADDR) {
293 /*
294 * when nid is reallocated,
295 * previous nat entry can be remained in nat cache.
296 * So, reinitialize it with new information.
297 */
298 copy_node_info(&e->ni, ni);
299 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
300 }
301
302 /* sanity check */
303 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
304 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
305 new_blkaddr == NULL_ADDR);
306 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
307 new_blkaddr == NEW_ADDR);
308 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
309 nat_get_blkaddr(e) != NULL_ADDR &&
310 new_blkaddr == NEW_ADDR);
311
312 /* increment version no as node is removed */
313 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
314 unsigned char version = nat_get_version(e);
315 nat_set_version(e, inc_node_version(version));
316
317 /* in order to reuse the nid */
318 if (nm_i->next_scan_nid > ni->nid)
319 nm_i->next_scan_nid = ni->nid;
320 }
321
322 /* change address */
323 nat_set_blkaddr(e, new_blkaddr);
324 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
325 set_nat_flag(e, IS_CHECKPOINTED, false);
326 __set_nat_cache_dirty(nm_i, e);
327
328 /* update fsync_mark if its inode nat entry is still alive */
329 if (ni->nid != ni->ino)
330 e = __lookup_nat_cache(nm_i, ni->ino);
331 if (e) {
332 if (fsync_done && ni->nid == ni->ino)
333 set_nat_flag(e, HAS_FSYNCED_INODE, true);
334 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
335 }
336 up_write(&nm_i->nat_tree_lock);
337}
338
339int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
340{
341 struct f2fs_nm_info *nm_i = NM_I(sbi);
342 int nr = nr_shrink;
343
344 if (!down_write_trylock(&nm_i->nat_tree_lock))
345 return 0;
346
347 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
348 struct nat_entry *ne;
349 ne = list_first_entry(&nm_i->nat_entries,
350 struct nat_entry, list);
351 __del_from_nat_cache(nm_i, ne);
352 nr_shrink--;
353 }
354 up_write(&nm_i->nat_tree_lock);
355 return nr - nr_shrink;
356}
357
358/*
359 * This function always returns success
360 */
361void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
362{
363 struct f2fs_nm_info *nm_i = NM_I(sbi);
364 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
365 struct f2fs_journal *journal = curseg->journal;
366 nid_t start_nid = START_NID(nid);
367 struct f2fs_nat_block *nat_blk;
368 struct page *page = NULL;
369 struct f2fs_nat_entry ne;
370 struct nat_entry *e;
371 int i;
372
373 ni->nid = nid;
374
375 /* Check nat cache */
376 down_read(&nm_i->nat_tree_lock);
377 e = __lookup_nat_cache(nm_i, nid);
378 if (e) {
379 ni->ino = nat_get_ino(e);
380 ni->blk_addr = nat_get_blkaddr(e);
381 ni->version = nat_get_version(e);
382 up_read(&nm_i->nat_tree_lock);
383 return;
384 }
385
386 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
387
388 /* Check current segment summary */
389 down_read(&curseg->journal_rwsem);
390 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
391 if (i >= 0) {
392 ne = nat_in_journal(journal, i);
393 node_info_from_raw_nat(ni, &ne);
394 }
395 up_read(&curseg->journal_rwsem);
396 if (i >= 0)
397 goto cache;
398
399 /* Fill node_info from nat page */
400 page = get_current_nat_page(sbi, start_nid);
401 nat_blk = (struct f2fs_nat_block *)page_address(page);
402 ne = nat_blk->entries[nid - start_nid];
403 node_info_from_raw_nat(ni, &ne);
404 f2fs_put_page(page, 1);
405cache:
406 up_read(&nm_i->nat_tree_lock);
407 /* cache nat entry */
408 down_write(&nm_i->nat_tree_lock);
409 cache_nat_entry(sbi, nid, &ne);
410 up_write(&nm_i->nat_tree_lock);
411}
412
413/*
414 * readahead MAX_RA_NODE number of node pages.
415 */
416static void ra_node_pages(struct page *parent, int start, int n)
417{
418 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
419 struct blk_plug plug;
420 int i, end;
421 nid_t nid;
422
423 blk_start_plug(&plug);
424
425 /* Then, try readahead for siblings of the desired node */
426 end = start + n;
427 end = min(end, NIDS_PER_BLOCK);
428 for (i = start; i < end; i++) {
429 nid = get_nid(parent, i, false);
430 ra_node_page(sbi, nid);
431 }
432
433 blk_finish_plug(&plug);
434}
435
436pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
437{
438 const long direct_index = ADDRS_PER_INODE(dn->inode);
439 const long direct_blks = ADDRS_PER_BLOCK;
440 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
441 unsigned int skipped_unit = ADDRS_PER_BLOCK;
442 int cur_level = dn->cur_level;
443 int max_level = dn->max_level;
444 pgoff_t base = 0;
445
446 if (!dn->max_level)
447 return pgofs + 1;
448
449 while (max_level-- > cur_level)
450 skipped_unit *= NIDS_PER_BLOCK;
451
452 switch (dn->max_level) {
453 case 3:
454 base += 2 * indirect_blks;
455 case 2:
456 base += 2 * direct_blks;
457 case 1:
458 base += direct_index;
459 break;
460 default:
461 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
462 }
463
464 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
465}
466
467/*
468 * The maximum depth is four.
469 * Offset[0] will have raw inode offset.
470 */
471static int get_node_path(struct inode *inode, long block,
472 int offset[4], unsigned int noffset[4])
473{
474 const long direct_index = ADDRS_PER_INODE(inode);
475 const long direct_blks = ADDRS_PER_BLOCK;
476 const long dptrs_per_blk = NIDS_PER_BLOCK;
477 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
478 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
479 int n = 0;
480 int level = 0;
481
482 noffset[0] = 0;
483
484 if (block < direct_index) {
485 offset[n] = block;
486 goto got;
487 }
488 block -= direct_index;
489 if (block < direct_blks) {
490 offset[n++] = NODE_DIR1_BLOCK;
491 noffset[n] = 1;
492 offset[n] = block;
493 level = 1;
494 goto got;
495 }
496 block -= direct_blks;
497 if (block < direct_blks) {
498 offset[n++] = NODE_DIR2_BLOCK;
499 noffset[n] = 2;
500 offset[n] = block;
501 level = 1;
502 goto got;
503 }
504 block -= direct_blks;
505 if (block < indirect_blks) {
506 offset[n++] = NODE_IND1_BLOCK;
507 noffset[n] = 3;
508 offset[n++] = block / direct_blks;
509 noffset[n] = 4 + offset[n - 1];
510 offset[n] = block % direct_blks;
511 level = 2;
512 goto got;
513 }
514 block -= indirect_blks;
515 if (block < indirect_blks) {
516 offset[n++] = NODE_IND2_BLOCK;
517 noffset[n] = 4 + dptrs_per_blk;
518 offset[n++] = block / direct_blks;
519 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
520 offset[n] = block % direct_blks;
521 level = 2;
522 goto got;
523 }
524 block -= indirect_blks;
525 if (block < dindirect_blks) {
526 offset[n++] = NODE_DIND_BLOCK;
527 noffset[n] = 5 + (dptrs_per_blk * 2);
528 offset[n++] = block / indirect_blks;
529 noffset[n] = 6 + (dptrs_per_blk * 2) +
530 offset[n - 1] * (dptrs_per_blk + 1);
531 offset[n++] = (block / direct_blks) % dptrs_per_blk;
532 noffset[n] = 7 + (dptrs_per_blk * 2) +
533 offset[n - 2] * (dptrs_per_blk + 1) +
534 offset[n - 1];
535 offset[n] = block % direct_blks;
536 level = 3;
537 goto got;
538 } else {
539 BUG();
540 }
541got:
542 return level;
543}
544
545/*
546 * Caller should call f2fs_put_dnode(dn).
547 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
548 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
549 * In the case of RDONLY_NODE, we don't need to care about mutex.
550 */
551int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
552{
553 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
554 struct page *npage[4];
555 struct page *parent = NULL;
556 int offset[4];
557 unsigned int noffset[4];
558 nid_t nids[4];
559 int level, i = 0;
560 int err = 0;
561
562 level = get_node_path(dn->inode, index, offset, noffset);
563
564 nids[0] = dn->inode->i_ino;
565 npage[0] = dn->inode_page;
566
567 if (!npage[0]) {
568 npage[0] = get_node_page(sbi, nids[0]);
569 if (IS_ERR(npage[0]))
570 return PTR_ERR(npage[0]);
571 }
572
573 /* if inline_data is set, should not report any block indices */
574 if (f2fs_has_inline_data(dn->inode) && index) {
575 err = -ENOENT;
576 f2fs_put_page(npage[0], 1);
577 goto release_out;
578 }
579
580 parent = npage[0];
581 if (level != 0)
582 nids[1] = get_nid(parent, offset[0], true);
583 dn->inode_page = npage[0];
584 dn->inode_page_locked = true;
585
586 /* get indirect or direct nodes */
587 for (i = 1; i <= level; i++) {
588 bool done = false;
589
590 if (!nids[i] && mode == ALLOC_NODE) {
591 /* alloc new node */
592 if (!alloc_nid(sbi, &(nids[i]))) {
593 err = -ENOSPC;
594 goto release_pages;
595 }
596
597 dn->nid = nids[i];
598 npage[i] = new_node_page(dn, noffset[i], NULL);
599 if (IS_ERR(npage[i])) {
600 alloc_nid_failed(sbi, nids[i]);
601 err = PTR_ERR(npage[i]);
602 goto release_pages;
603 }
604
605 set_nid(parent, offset[i - 1], nids[i], i == 1);
606 alloc_nid_done(sbi, nids[i]);
607 done = true;
608 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
609 npage[i] = get_node_page_ra(parent, offset[i - 1]);
610 if (IS_ERR(npage[i])) {
611 err = PTR_ERR(npage[i]);
612 goto release_pages;
613 }
614 done = true;
615 }
616 if (i == 1) {
617 dn->inode_page_locked = false;
618 unlock_page(parent);
619 } else {
620 f2fs_put_page(parent, 1);
621 }
622
623 if (!done) {
624 npage[i] = get_node_page(sbi, nids[i]);
625 if (IS_ERR(npage[i])) {
626 err = PTR_ERR(npage[i]);
627 f2fs_put_page(npage[0], 0);
628 goto release_out;
629 }
630 }
631 if (i < level) {
632 parent = npage[i];
633 nids[i + 1] = get_nid(parent, offset[i], false);
634 }
635 }
636 dn->nid = nids[level];
637 dn->ofs_in_node = offset[level];
638 dn->node_page = npage[level];
639 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
640 return 0;
641
642release_pages:
643 f2fs_put_page(parent, 1);
644 if (i > 1)
645 f2fs_put_page(npage[0], 0);
646release_out:
647 dn->inode_page = NULL;
648 dn->node_page = NULL;
649 if (err == -ENOENT) {
650 dn->cur_level = i;
651 dn->max_level = level;
652 dn->ofs_in_node = offset[level];
653 }
654 return err;
655}
656
657static void truncate_node(struct dnode_of_data *dn)
658{
659 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
660 struct node_info ni;
661
662 get_node_info(sbi, dn->nid, &ni);
663 if (dn->inode->i_blocks == 0) {
664 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
665 goto invalidate;
666 }
667 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
668
669 /* Deallocate node address */
670 invalidate_blocks(sbi, ni.blk_addr);
671 dec_valid_node_count(sbi, dn->inode);
672 set_node_addr(sbi, &ni, NULL_ADDR, false);
673
674 if (dn->nid == dn->inode->i_ino) {
675 remove_orphan_inode(sbi, dn->nid);
676 dec_valid_inode_count(sbi);
677 f2fs_inode_synced(dn->inode);
678 }
679invalidate:
680 clear_node_page_dirty(dn->node_page);
681 set_sbi_flag(sbi, SBI_IS_DIRTY);
682
683 f2fs_put_page(dn->node_page, 1);
684
685 invalidate_mapping_pages(NODE_MAPPING(sbi),
686 dn->node_page->index, dn->node_page->index);
687
688 dn->node_page = NULL;
689 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
690}
691
692static int truncate_dnode(struct dnode_of_data *dn)
693{
694 struct page *page;
695
696 if (dn->nid == 0)
697 return 1;
698
699 /* get direct node */
700 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
701 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
702 return 1;
703 else if (IS_ERR(page))
704 return PTR_ERR(page);
705
706 /* Make dnode_of_data for parameter */
707 dn->node_page = page;
708 dn->ofs_in_node = 0;
709 truncate_data_blocks(dn);
710 truncate_node(dn);
711 return 1;
712}
713
714static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
715 int ofs, int depth)
716{
717 struct dnode_of_data rdn = *dn;
718 struct page *page;
719 struct f2fs_node *rn;
720 nid_t child_nid;
721 unsigned int child_nofs;
722 int freed = 0;
723 int i, ret;
724
725 if (dn->nid == 0)
726 return NIDS_PER_BLOCK + 1;
727
728 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
729
730 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
731 if (IS_ERR(page)) {
732 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
733 return PTR_ERR(page);
734 }
735
736 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
737
738 rn = F2FS_NODE(page);
739 if (depth < 3) {
740 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
741 child_nid = le32_to_cpu(rn->in.nid[i]);
742 if (child_nid == 0)
743 continue;
744 rdn.nid = child_nid;
745 ret = truncate_dnode(&rdn);
746 if (ret < 0)
747 goto out_err;
748 if (set_nid(page, i, 0, false))
749 dn->node_changed = true;
750 }
751 } else {
752 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
753 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
754 child_nid = le32_to_cpu(rn->in.nid[i]);
755 if (child_nid == 0) {
756 child_nofs += NIDS_PER_BLOCK + 1;
757 continue;
758 }
759 rdn.nid = child_nid;
760 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
761 if (ret == (NIDS_PER_BLOCK + 1)) {
762 if (set_nid(page, i, 0, false))
763 dn->node_changed = true;
764 child_nofs += ret;
765 } else if (ret < 0 && ret != -ENOENT) {
766 goto out_err;
767 }
768 }
769 freed = child_nofs;
770 }
771
772 if (!ofs) {
773 /* remove current indirect node */
774 dn->node_page = page;
775 truncate_node(dn);
776 freed++;
777 } else {
778 f2fs_put_page(page, 1);
779 }
780 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
781 return freed;
782
783out_err:
784 f2fs_put_page(page, 1);
785 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
786 return ret;
787}
788
789static int truncate_partial_nodes(struct dnode_of_data *dn,
790 struct f2fs_inode *ri, int *offset, int depth)
791{
792 struct page *pages[2];
793 nid_t nid[3];
794 nid_t child_nid;
795 int err = 0;
796 int i;
797 int idx = depth - 2;
798
799 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
800 if (!nid[0])
801 return 0;
802
803 /* get indirect nodes in the path */
804 for (i = 0; i < idx + 1; i++) {
805 /* reference count'll be increased */
806 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
807 if (IS_ERR(pages[i])) {
808 err = PTR_ERR(pages[i]);
809 idx = i - 1;
810 goto fail;
811 }
812 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
813 }
814
815 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
816
817 /* free direct nodes linked to a partial indirect node */
818 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
819 child_nid = get_nid(pages[idx], i, false);
820 if (!child_nid)
821 continue;
822 dn->nid = child_nid;
823 err = truncate_dnode(dn);
824 if (err < 0)
825 goto fail;
826 if (set_nid(pages[idx], i, 0, false))
827 dn->node_changed = true;
828 }
829
830 if (offset[idx + 1] == 0) {
831 dn->node_page = pages[idx];
832 dn->nid = nid[idx];
833 truncate_node(dn);
834 } else {
835 f2fs_put_page(pages[idx], 1);
836 }
837 offset[idx]++;
838 offset[idx + 1] = 0;
839 idx--;
840fail:
841 for (i = idx; i >= 0; i--)
842 f2fs_put_page(pages[i], 1);
843
844 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
845
846 return err;
847}
848
849/*
850 * All the block addresses of data and nodes should be nullified.
851 */
852int truncate_inode_blocks(struct inode *inode, pgoff_t from)
853{
854 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
855 int err = 0, cont = 1;
856 int level, offset[4], noffset[4];
857 unsigned int nofs = 0;
858 struct f2fs_inode *ri;
859 struct dnode_of_data dn;
860 struct page *page;
861
862 trace_f2fs_truncate_inode_blocks_enter(inode, from);
863
864 level = get_node_path(inode, from, offset, noffset);
865
866 page = get_node_page(sbi, inode->i_ino);
867 if (IS_ERR(page)) {
868 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
869 return PTR_ERR(page);
870 }
871
872 set_new_dnode(&dn, inode, page, NULL, 0);
873 unlock_page(page);
874
875 ri = F2FS_INODE(page);
876 switch (level) {
877 case 0:
878 case 1:
879 nofs = noffset[1];
880 break;
881 case 2:
882 nofs = noffset[1];
883 if (!offset[level - 1])
884 goto skip_partial;
885 err = truncate_partial_nodes(&dn, ri, offset, level);
886 if (err < 0 && err != -ENOENT)
887 goto fail;
888 nofs += 1 + NIDS_PER_BLOCK;
889 break;
890 case 3:
891 nofs = 5 + 2 * NIDS_PER_BLOCK;
892 if (!offset[level - 1])
893 goto skip_partial;
894 err = truncate_partial_nodes(&dn, ri, offset, level);
895 if (err < 0 && err != -ENOENT)
896 goto fail;
897 break;
898 default:
899 BUG();
900 }
901
902skip_partial:
903 while (cont) {
904 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
905 switch (offset[0]) {
906 case NODE_DIR1_BLOCK:
907 case NODE_DIR2_BLOCK:
908 err = truncate_dnode(&dn);
909 break;
910
911 case NODE_IND1_BLOCK:
912 case NODE_IND2_BLOCK:
913 err = truncate_nodes(&dn, nofs, offset[1], 2);
914 break;
915
916 case NODE_DIND_BLOCK:
917 err = truncate_nodes(&dn, nofs, offset[1], 3);
918 cont = 0;
919 break;
920
921 default:
922 BUG();
923 }
924 if (err < 0 && err != -ENOENT)
925 goto fail;
926 if (offset[1] == 0 &&
927 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
928 lock_page(page);
929 BUG_ON(page->mapping != NODE_MAPPING(sbi));
930 f2fs_wait_on_page_writeback(page, NODE, true);
931 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
932 set_page_dirty(page);
933 unlock_page(page);
934 }
935 offset[1] = 0;
936 offset[0]++;
937 nofs += err;
938 }
939fail:
940 f2fs_put_page(page, 0);
941 trace_f2fs_truncate_inode_blocks_exit(inode, err);
942 return err > 0 ? 0 : err;
943}
944
945int truncate_xattr_node(struct inode *inode, struct page *page)
946{
947 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
948 nid_t nid = F2FS_I(inode)->i_xattr_nid;
949 struct dnode_of_data dn;
950 struct page *npage;
951
952 if (!nid)
953 return 0;
954
955 npage = get_node_page(sbi, nid);
956 if (IS_ERR(npage))
957 return PTR_ERR(npage);
958
959 f2fs_i_xnid_write(inode, 0);
960
961 /* need to do checkpoint during fsync */
962 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
963
964 set_new_dnode(&dn, inode, page, npage, nid);
965
966 if (page)
967 dn.inode_page_locked = true;
968 truncate_node(&dn);
969 return 0;
970}
971
972/*
973 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
974 * f2fs_unlock_op().
975 */
976int remove_inode_page(struct inode *inode)
977{
978 struct dnode_of_data dn;
979 int err;
980
981 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
982 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
983 if (err)
984 return err;
985
986 err = truncate_xattr_node(inode, dn.inode_page);
987 if (err) {
988 f2fs_put_dnode(&dn);
989 return err;
990 }
991
992 /* remove potential inline_data blocks */
993 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
994 S_ISLNK(inode->i_mode))
995 truncate_data_blocks_range(&dn, 1);
996
997 /* 0 is possible, after f2fs_new_inode() has failed */
998 f2fs_bug_on(F2FS_I_SB(inode),
999 inode->i_blocks != 0 && inode->i_blocks != 1);
1000
1001 /* will put inode & node pages */
1002 truncate_node(&dn);
1003 return 0;
1004}
1005
1006struct page *new_inode_page(struct inode *inode)
1007{
1008 struct dnode_of_data dn;
1009
1010 /* allocate inode page for new inode */
1011 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1012
1013 /* caller should f2fs_put_page(page, 1); */
1014 return new_node_page(&dn, 0, NULL);
1015}
1016
1017struct page *new_node_page(struct dnode_of_data *dn,
1018 unsigned int ofs, struct page *ipage)
1019{
1020 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1021 struct node_info old_ni, new_ni;
1022 struct page *page;
1023 int err;
1024
1025 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1026 return ERR_PTR(-EPERM);
1027
1028 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1029 if (!page)
1030 return ERR_PTR(-ENOMEM);
1031
1032 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1033 err = -ENOSPC;
1034 goto fail;
1035 }
1036
1037 get_node_info(sbi, dn->nid, &old_ni);
1038
1039 /* Reinitialize old_ni with new node page */
1040 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1041 new_ni = old_ni;
1042 new_ni.ino = dn->inode->i_ino;
1043 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1044
1045 f2fs_wait_on_page_writeback(page, NODE, true);
1046 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1047 set_cold_node(dn->inode, page);
1048 if (!PageUptodate(page))
1049 SetPageUptodate(page);
1050 if (set_page_dirty(page))
1051 dn->node_changed = true;
1052
1053 if (f2fs_has_xattr_block(ofs))
1054 f2fs_i_xnid_write(dn->inode, dn->nid);
1055
1056 if (ofs == 0)
1057 inc_valid_inode_count(sbi);
1058 return page;
1059
1060fail:
1061 clear_node_page_dirty(page);
1062 f2fs_put_page(page, 1);
1063 return ERR_PTR(err);
1064}
1065
1066/*
1067 * Caller should do after getting the following values.
1068 * 0: f2fs_put_page(page, 0)
1069 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1070 */
1071static int read_node_page(struct page *page, int op_flags)
1072{
1073 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1074 struct node_info ni;
1075 struct f2fs_io_info fio = {
1076 .sbi = sbi,
1077 .type = NODE,
1078 .op = REQ_OP_READ,
1079 .op_flags = op_flags,
1080 .page = page,
1081 .encrypted_page = NULL,
1082 };
1083
1084 if (PageUptodate(page))
1085 return LOCKED_PAGE;
1086
1087 get_node_info(sbi, page->index, &ni);
1088
1089 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1090 ClearPageUptodate(page);
1091 return -ENOENT;
1092 }
1093
1094 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1095 return f2fs_submit_page_bio(&fio);
1096}
1097
1098/*
1099 * Readahead a node page
1100 */
1101void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1102{
1103 struct page *apage;
1104 int err;
1105
1106 if (!nid)
1107 return;
1108 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1109
1110 rcu_read_lock();
1111 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1112 rcu_read_unlock();
1113 if (apage)
1114 return;
1115
1116 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1117 if (!apage)
1118 return;
1119
1120 err = read_node_page(apage, REQ_RAHEAD);
1121 f2fs_put_page(apage, err ? 1 : 0);
1122}
1123
1124static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1125 struct page *parent, int start)
1126{
1127 struct page *page;
1128 int err;
1129
1130 if (!nid)
1131 return ERR_PTR(-ENOENT);
1132 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1133repeat:
1134 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1135 if (!page)
1136 return ERR_PTR(-ENOMEM);
1137
1138 err = read_node_page(page, 0);
1139 if (err < 0) {
1140 f2fs_put_page(page, 1);
1141 return ERR_PTR(err);
1142 } else if (err == LOCKED_PAGE) {
1143 goto page_hit;
1144 }
1145
1146 if (parent)
1147 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1148
1149 lock_page(page);
1150
1151 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1152 f2fs_put_page(page, 1);
1153 goto repeat;
1154 }
1155
1156 if (unlikely(!PageUptodate(page)))
1157 goto out_err;
1158page_hit:
1159 if(unlikely(nid != nid_of_node(page))) {
1160 f2fs_bug_on(sbi, 1);
1161 ClearPageUptodate(page);
1162out_err:
1163 f2fs_put_page(page, 1);
1164 return ERR_PTR(-EIO);
1165 }
1166 return page;
1167}
1168
1169struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1170{
1171 return __get_node_page(sbi, nid, NULL, 0);
1172}
1173
1174struct page *get_node_page_ra(struct page *parent, int start)
1175{
1176 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1177 nid_t nid = get_nid(parent, start, false);
1178
1179 return __get_node_page(sbi, nid, parent, start);
1180}
1181
1182static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1183{
1184 struct inode *inode;
1185 struct page *page;
1186 int ret;
1187
1188 /* should flush inline_data before evict_inode */
1189 inode = ilookup(sbi->sb, ino);
1190 if (!inode)
1191 return;
1192
1193 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1194 if (!page)
1195 goto iput_out;
1196
1197 if (!PageUptodate(page))
1198 goto page_out;
1199
1200 if (!PageDirty(page))
1201 goto page_out;
1202
1203 if (!clear_page_dirty_for_io(page))
1204 goto page_out;
1205
1206 ret = f2fs_write_inline_data(inode, page);
1207 inode_dec_dirty_pages(inode);
1208 remove_dirty_inode(inode);
1209 if (ret)
1210 set_page_dirty(page);
1211page_out:
1212 f2fs_put_page(page, 1);
1213iput_out:
1214 iput(inode);
1215}
1216
1217void move_node_page(struct page *node_page, int gc_type)
1218{
1219 if (gc_type == FG_GC) {
1220 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1221 struct writeback_control wbc = {
1222 .sync_mode = WB_SYNC_ALL,
1223 .nr_to_write = 1,
1224 .for_reclaim = 0,
1225 };
1226
1227 set_page_dirty(node_page);
1228 f2fs_wait_on_page_writeback(node_page, NODE, true);
1229
1230 f2fs_bug_on(sbi, PageWriteback(node_page));
1231 if (!clear_page_dirty_for_io(node_page))
1232 goto out_page;
1233
1234 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1235 unlock_page(node_page);
1236 goto release_page;
1237 } else {
1238 /* set page dirty and write it */
1239 if (!PageWriteback(node_page))
1240 set_page_dirty(node_page);
1241 }
1242out_page:
1243 unlock_page(node_page);
1244release_page:
1245 f2fs_put_page(node_page, 0);
1246}
1247
1248static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1249{
1250 pgoff_t index, end;
1251 struct pagevec pvec;
1252 struct page *last_page = NULL;
1253
1254 pagevec_init(&pvec, 0);
1255 index = 0;
1256 end = ULONG_MAX;
1257
1258 while (index <= end) {
1259 int i, nr_pages;
1260 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1261 PAGECACHE_TAG_DIRTY,
1262 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1263 if (nr_pages == 0)
1264 break;
1265
1266 for (i = 0; i < nr_pages; i++) {
1267 struct page *page = pvec.pages[i];
1268
1269 if (unlikely(f2fs_cp_error(sbi))) {
1270 f2fs_put_page(last_page, 0);
1271 pagevec_release(&pvec);
1272 return ERR_PTR(-EIO);
1273 }
1274
1275 if (!IS_DNODE(page) || !is_cold_node(page))
1276 continue;
1277 if (ino_of_node(page) != ino)
1278 continue;
1279
1280 lock_page(page);
1281
1282 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1283continue_unlock:
1284 unlock_page(page);
1285 continue;
1286 }
1287 if (ino_of_node(page) != ino)
1288 goto continue_unlock;
1289
1290 if (!PageDirty(page)) {
1291 /* someone wrote it for us */
1292 goto continue_unlock;
1293 }
1294
1295 if (last_page)
1296 f2fs_put_page(last_page, 0);
1297
1298 get_page(page);
1299 last_page = page;
1300 unlock_page(page);
1301 }
1302 pagevec_release(&pvec);
1303 cond_resched();
1304 }
1305 return last_page;
1306}
1307
1308int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1309 struct writeback_control *wbc, bool atomic)
1310{
1311 pgoff_t index, end;
1312 struct pagevec pvec;
1313 int ret = 0;
1314 struct page *last_page = NULL;
1315 bool marked = false;
1316 nid_t ino = inode->i_ino;
1317 int nwritten = 0;
1318
1319 if (atomic) {
1320 last_page = last_fsync_dnode(sbi, ino);
1321 if (IS_ERR_OR_NULL(last_page))
1322 return PTR_ERR_OR_ZERO(last_page);
1323 }
1324retry:
1325 pagevec_init(&pvec, 0);
1326 index = 0;
1327 end = ULONG_MAX;
1328
1329 while (index <= end) {
1330 int i, nr_pages;
1331 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1332 PAGECACHE_TAG_DIRTY,
1333 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1334 if (nr_pages == 0)
1335 break;
1336
1337 for (i = 0; i < nr_pages; i++) {
1338 struct page *page = pvec.pages[i];
1339
1340 if (unlikely(f2fs_cp_error(sbi))) {
1341 f2fs_put_page(last_page, 0);
1342 pagevec_release(&pvec);
1343 ret = -EIO;
1344 goto out;
1345 }
1346
1347 if (!IS_DNODE(page) || !is_cold_node(page))
1348 continue;
1349 if (ino_of_node(page) != ino)
1350 continue;
1351
1352 lock_page(page);
1353
1354 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1355continue_unlock:
1356 unlock_page(page);
1357 continue;
1358 }
1359 if (ino_of_node(page) != ino)
1360 goto continue_unlock;
1361
1362 if (!PageDirty(page) && page != last_page) {
1363 /* someone wrote it for us */
1364 goto continue_unlock;
1365 }
1366
1367 f2fs_wait_on_page_writeback(page, NODE, true);
1368 BUG_ON(PageWriteback(page));
1369
1370 if (!atomic || page == last_page) {
1371 set_fsync_mark(page, 1);
1372 if (IS_INODE(page)) {
1373 if (is_inode_flag_set(inode,
1374 FI_DIRTY_INODE))
1375 update_inode(inode, page);
1376 set_dentry_mark(page,
1377 need_dentry_mark(sbi, ino));
1378 }
1379 /* may be written by other thread */
1380 if (!PageDirty(page))
1381 set_page_dirty(page);
1382 }
1383
1384 if (!clear_page_dirty_for_io(page))
1385 goto continue_unlock;
1386
1387 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1388 if (ret) {
1389 unlock_page(page);
1390 f2fs_put_page(last_page, 0);
1391 break;
1392 } else {
1393 nwritten++;
1394 }
1395
1396 if (page == last_page) {
1397 f2fs_put_page(page, 0);
1398 marked = true;
1399 break;
1400 }
1401 }
1402 pagevec_release(&pvec);
1403 cond_resched();
1404
1405 if (ret || marked)
1406 break;
1407 }
1408 if (!ret && atomic && !marked) {
1409 f2fs_msg(sbi->sb, KERN_DEBUG,
1410 "Retry to write fsync mark: ino=%u, idx=%lx",
1411 ino, last_page->index);
1412 lock_page(last_page);
1413 f2fs_wait_on_page_writeback(last_page, NODE, true);
1414 set_page_dirty(last_page);
1415 unlock_page(last_page);
1416 goto retry;
1417 }
1418out:
1419 if (nwritten)
1420 f2fs_submit_merged_bio_cond(sbi, NULL, NULL, ino, NODE, WRITE);
1421 return ret ? -EIO: 0;
1422}
1423
1424int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1425{
1426 pgoff_t index, end;
1427 struct pagevec pvec;
1428 int step = 0;
1429 int nwritten = 0;
1430 int ret = 0;
1431
1432 pagevec_init(&pvec, 0);
1433
1434next_step:
1435 index = 0;
1436 end = ULONG_MAX;
1437
1438 while (index <= end) {
1439 int i, nr_pages;
1440 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1441 PAGECACHE_TAG_DIRTY,
1442 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1443 if (nr_pages == 0)
1444 break;
1445
1446 for (i = 0; i < nr_pages; i++) {
1447 struct page *page = pvec.pages[i];
1448
1449 if (unlikely(f2fs_cp_error(sbi))) {
1450 pagevec_release(&pvec);
1451 ret = -EIO;
1452 goto out;
1453 }
1454
1455 /*
1456 * flushing sequence with step:
1457 * 0. indirect nodes
1458 * 1. dentry dnodes
1459 * 2. file dnodes
1460 */
1461 if (step == 0 && IS_DNODE(page))
1462 continue;
1463 if (step == 1 && (!IS_DNODE(page) ||
1464 is_cold_node(page)))
1465 continue;
1466 if (step == 2 && (!IS_DNODE(page) ||
1467 !is_cold_node(page)))
1468 continue;
1469lock_node:
1470 if (!trylock_page(page))
1471 continue;
1472
1473 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1474continue_unlock:
1475 unlock_page(page);
1476 continue;
1477 }
1478
1479 if (!PageDirty(page)) {
1480 /* someone wrote it for us */
1481 goto continue_unlock;
1482 }
1483
1484 /* flush inline_data */
1485 if (is_inline_node(page)) {
1486 clear_inline_node(page);
1487 unlock_page(page);
1488 flush_inline_data(sbi, ino_of_node(page));
1489 goto lock_node;
1490 }
1491
1492 f2fs_wait_on_page_writeback(page, NODE, true);
1493
1494 BUG_ON(PageWriteback(page));
1495 if (!clear_page_dirty_for_io(page))
1496 goto continue_unlock;
1497
1498 set_fsync_mark(page, 0);
1499 set_dentry_mark(page, 0);
1500
1501 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1502 unlock_page(page);
1503 else
1504 nwritten++;
1505
1506 if (--wbc->nr_to_write == 0)
1507 break;
1508 }
1509 pagevec_release(&pvec);
1510 cond_resched();
1511
1512 if (wbc->nr_to_write == 0) {
1513 step = 2;
1514 break;
1515 }
1516 }
1517
1518 if (step < 2) {
1519 step++;
1520 goto next_step;
1521 }
1522out:
1523 if (nwritten)
1524 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1525 return ret;
1526}
1527
1528int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1529{
1530 pgoff_t index = 0, end = ULONG_MAX;
1531 struct pagevec pvec;
1532 int ret2, ret = 0;
1533
1534 pagevec_init(&pvec, 0);
1535
1536 while (index <= end) {
1537 int i, nr_pages;
1538 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1539 PAGECACHE_TAG_WRITEBACK,
1540 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1541 if (nr_pages == 0)
1542 break;
1543
1544 for (i = 0; i < nr_pages; i++) {
1545 struct page *page = pvec.pages[i];
1546
1547 /* until radix tree lookup accepts end_index */
1548 if (unlikely(page->index > end))
1549 continue;
1550
1551 if (ino && ino_of_node(page) == ino) {
1552 f2fs_wait_on_page_writeback(page, NODE, true);
1553 if (TestClearPageError(page))
1554 ret = -EIO;
1555 }
1556 }
1557 pagevec_release(&pvec);
1558 cond_resched();
1559 }
1560
1561 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1562 if (!ret)
1563 ret = ret2;
1564 return ret;
1565}
1566
1567static int f2fs_write_node_page(struct page *page,
1568 struct writeback_control *wbc)
1569{
1570 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1571 nid_t nid;
1572 struct node_info ni;
1573 struct f2fs_io_info fio = {
1574 .sbi = sbi,
1575 .type = NODE,
1576 .op = REQ_OP_WRITE,
1577 .op_flags = wbc_to_write_flags(wbc),
1578 .page = page,
1579 .encrypted_page = NULL,
1580 };
1581
1582 trace_f2fs_writepage(page, NODE);
1583
1584 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1585 goto redirty_out;
1586 if (unlikely(f2fs_cp_error(sbi)))
1587 goto redirty_out;
1588
1589 /* get old block addr of this node page */
1590 nid = nid_of_node(page);
1591 f2fs_bug_on(sbi, page->index != nid);
1592
1593 if (wbc->for_reclaim) {
1594 if (!down_read_trylock(&sbi->node_write))
1595 goto redirty_out;
1596 } else {
1597 down_read(&sbi->node_write);
1598 }
1599
1600 get_node_info(sbi, nid, &ni);
1601
1602 /* This page is already truncated */
1603 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1604 ClearPageUptodate(page);
1605 dec_page_count(sbi, F2FS_DIRTY_NODES);
1606 up_read(&sbi->node_write);
1607 unlock_page(page);
1608 return 0;
1609 }
1610
1611 set_page_writeback(page);
1612 fio.old_blkaddr = ni.blk_addr;
1613 write_node_page(nid, &fio);
1614 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1615 dec_page_count(sbi, F2FS_DIRTY_NODES);
1616 up_read(&sbi->node_write);
1617
1618 if (wbc->for_reclaim)
1619 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1620
1621 unlock_page(page);
1622
1623 if (unlikely(f2fs_cp_error(sbi)))
1624 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1625
1626 return 0;
1627
1628redirty_out:
1629 redirty_page_for_writepage(wbc, page);
1630 return AOP_WRITEPAGE_ACTIVATE;
1631}
1632
1633static int f2fs_write_node_pages(struct address_space *mapping,
1634 struct writeback_control *wbc)
1635{
1636 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1637 struct blk_plug plug;
1638 long diff;
1639
1640 /* balancing f2fs's metadata in background */
1641 f2fs_balance_fs_bg(sbi);
1642
1643 /* collect a number of dirty node pages and write together */
1644 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1645 goto skip_write;
1646
1647 trace_f2fs_writepages(mapping->host, wbc, NODE);
1648
1649 diff = nr_pages_to_write(sbi, NODE, wbc);
1650 wbc->sync_mode = WB_SYNC_NONE;
1651 blk_start_plug(&plug);
1652 sync_node_pages(sbi, wbc);
1653 blk_finish_plug(&plug);
1654 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1655 return 0;
1656
1657skip_write:
1658 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1659 trace_f2fs_writepages(mapping->host, wbc, NODE);
1660 return 0;
1661}
1662
1663static int f2fs_set_node_page_dirty(struct page *page)
1664{
1665 trace_f2fs_set_page_dirty(page, NODE);
1666
1667 if (!PageUptodate(page))
1668 SetPageUptodate(page);
1669 if (!PageDirty(page)) {
1670 f2fs_set_page_dirty_nobuffers(page);
1671 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1672 SetPagePrivate(page);
1673 f2fs_trace_pid(page);
1674 return 1;
1675 }
1676 return 0;
1677}
1678
1679/*
1680 * Structure of the f2fs node operations
1681 */
1682const struct address_space_operations f2fs_node_aops = {
1683 .writepage = f2fs_write_node_page,
1684 .writepages = f2fs_write_node_pages,
1685 .set_page_dirty = f2fs_set_node_page_dirty,
1686 .invalidatepage = f2fs_invalidate_page,
1687 .releasepage = f2fs_release_page,
1688#ifdef CONFIG_MIGRATION
1689 .migratepage = f2fs_migrate_page,
1690#endif
1691};
1692
1693static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1694 nid_t n)
1695{
1696 return radix_tree_lookup(&nm_i->free_nid_root, n);
1697}
1698
1699static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1700 struct free_nid *i, enum nid_list list, bool new)
1701{
1702 struct f2fs_nm_info *nm_i = NM_I(sbi);
1703
1704 if (new) {
1705 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1706 if (err)
1707 return err;
1708 }
1709
1710 f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1711 i->state != NID_ALLOC);
1712 nm_i->nid_cnt[list]++;
1713 list_add_tail(&i->list, &nm_i->nid_list[list]);
1714 return 0;
1715}
1716
1717static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1718 struct free_nid *i, enum nid_list list, bool reuse)
1719{
1720 struct f2fs_nm_info *nm_i = NM_I(sbi);
1721
1722 f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1723 i->state != NID_ALLOC);
1724 nm_i->nid_cnt[list]--;
1725 list_del(&i->list);
1726 if (!reuse)
1727 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1728}
1729
1730static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1731{
1732 struct f2fs_nm_info *nm_i = NM_I(sbi);
1733 struct free_nid *i;
1734 struct nat_entry *ne;
1735 int err;
1736
1737 /* 0 nid should not be used */
1738 if (unlikely(nid == 0))
1739 return 0;
1740
1741 if (build) {
1742 /* do not add allocated nids */
1743 ne = __lookup_nat_cache(nm_i, nid);
1744 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1745 nat_get_blkaddr(ne) != NULL_ADDR))
1746 return 0;
1747 }
1748
1749 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1750 i->nid = nid;
1751 i->state = NID_NEW;
1752
1753 if (radix_tree_preload(GFP_NOFS)) {
1754 kmem_cache_free(free_nid_slab, i);
1755 return 0;
1756 }
1757
1758 spin_lock(&nm_i->nid_list_lock);
1759 err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1760 spin_unlock(&nm_i->nid_list_lock);
1761 radix_tree_preload_end();
1762 if (err) {
1763 kmem_cache_free(free_nid_slab, i);
1764 return 0;
1765 }
1766 return 1;
1767}
1768
1769static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1770{
1771 struct f2fs_nm_info *nm_i = NM_I(sbi);
1772 struct free_nid *i;
1773 bool need_free = false;
1774
1775 spin_lock(&nm_i->nid_list_lock);
1776 i = __lookup_free_nid_list(nm_i, nid);
1777 if (i && i->state == NID_NEW) {
1778 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1779 need_free = true;
1780 }
1781 spin_unlock(&nm_i->nid_list_lock);
1782
1783 if (need_free)
1784 kmem_cache_free(free_nid_slab, i);
1785}
1786
1787static void scan_nat_page(struct f2fs_sb_info *sbi,
1788 struct page *nat_page, nid_t start_nid)
1789{
1790 struct f2fs_nm_info *nm_i = NM_I(sbi);
1791 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1792 block_t blk_addr;
1793 int i;
1794
1795 i = start_nid % NAT_ENTRY_PER_BLOCK;
1796
1797 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1798
1799 if (unlikely(start_nid >= nm_i->max_nid))
1800 break;
1801
1802 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1803 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1804 if (blk_addr == NULL_ADDR)
1805 add_free_nid(sbi, start_nid, true);
1806 }
1807}
1808
1809static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync)
1810{
1811 struct f2fs_nm_info *nm_i = NM_I(sbi);
1812 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1813 struct f2fs_journal *journal = curseg->journal;
1814 int i = 0;
1815 nid_t nid = nm_i->next_scan_nid;
1816
1817 /* Enough entries */
1818 if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1819 return;
1820
1821 if (!sync && !available_free_memory(sbi, FREE_NIDS))
1822 return;
1823
1824 /* readahead nat pages to be scanned */
1825 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1826 META_NAT, true);
1827
1828 down_read(&nm_i->nat_tree_lock);
1829
1830 while (1) {
1831 struct page *page = get_current_nat_page(sbi, nid);
1832
1833 scan_nat_page(sbi, page, nid);
1834 f2fs_put_page(page, 1);
1835
1836 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1837 if (unlikely(nid >= nm_i->max_nid))
1838 nid = 0;
1839
1840 if (++i >= FREE_NID_PAGES)
1841 break;
1842 }
1843
1844 /* go to the next free nat pages to find free nids abundantly */
1845 nm_i->next_scan_nid = nid;
1846
1847 /* find free nids from current sum_pages */
1848 down_read(&curseg->journal_rwsem);
1849 for (i = 0; i < nats_in_cursum(journal); i++) {
1850 block_t addr;
1851
1852 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1853 nid = le32_to_cpu(nid_in_journal(journal, i));
1854 if (addr == NULL_ADDR)
1855 add_free_nid(sbi, nid, true);
1856 else
1857 remove_free_nid(sbi, nid);
1858 }
1859 up_read(&curseg->journal_rwsem);
1860 up_read(&nm_i->nat_tree_lock);
1861
1862 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1863 nm_i->ra_nid_pages, META_NAT, false);
1864}
1865
1866void build_free_nids(struct f2fs_sb_info *sbi, bool sync)
1867{
1868 mutex_lock(&NM_I(sbi)->build_lock);
1869 __build_free_nids(sbi, sync);
1870 mutex_unlock(&NM_I(sbi)->build_lock);
1871}
1872
1873/*
1874 * If this function returns success, caller can obtain a new nid
1875 * from second parameter of this function.
1876 * The returned nid could be used ino as well as nid when inode is created.
1877 */
1878bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1879{
1880 struct f2fs_nm_info *nm_i = NM_I(sbi);
1881 struct free_nid *i = NULL;
1882retry:
1883#ifdef CONFIG_F2FS_FAULT_INJECTION
1884 if (time_to_inject(sbi, FAULT_ALLOC_NID))
1885 return false;
1886#endif
1887 spin_lock(&nm_i->nid_list_lock);
1888
1889 if (unlikely(nm_i->available_nids == 0)) {
1890 spin_unlock(&nm_i->nid_list_lock);
1891 return false;
1892 }
1893
1894 /* We should not use stale free nids created by build_free_nids */
1895 if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
1896 f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
1897 i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
1898 struct free_nid, list);
1899 *nid = i->nid;
1900
1901 __remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
1902 i->state = NID_ALLOC;
1903 __insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
1904 nm_i->available_nids--;
1905 spin_unlock(&nm_i->nid_list_lock);
1906 return true;
1907 }
1908 spin_unlock(&nm_i->nid_list_lock);
1909
1910 /* Let's scan nat pages and its caches to get free nids */
1911 build_free_nids(sbi, true);
1912 goto retry;
1913}
1914
1915/*
1916 * alloc_nid() should be called prior to this function.
1917 */
1918void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1919{
1920 struct f2fs_nm_info *nm_i = NM_I(sbi);
1921 struct free_nid *i;
1922
1923 spin_lock(&nm_i->nid_list_lock);
1924 i = __lookup_free_nid_list(nm_i, nid);
1925 f2fs_bug_on(sbi, !i);
1926 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
1927 spin_unlock(&nm_i->nid_list_lock);
1928
1929 kmem_cache_free(free_nid_slab, i);
1930}
1931
1932/*
1933 * alloc_nid() should be called prior to this function.
1934 */
1935void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1936{
1937 struct f2fs_nm_info *nm_i = NM_I(sbi);
1938 struct free_nid *i;
1939 bool need_free = false;
1940
1941 if (!nid)
1942 return;
1943
1944 spin_lock(&nm_i->nid_list_lock);
1945 i = __lookup_free_nid_list(nm_i, nid);
1946 f2fs_bug_on(sbi, !i);
1947
1948 if (!available_free_memory(sbi, FREE_NIDS)) {
1949 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
1950 need_free = true;
1951 } else {
1952 __remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
1953 i->state = NID_NEW;
1954 __insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
1955 }
1956
1957 nm_i->available_nids++;
1958
1959 spin_unlock(&nm_i->nid_list_lock);
1960
1961 if (need_free)
1962 kmem_cache_free(free_nid_slab, i);
1963}
1964
1965int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1966{
1967 struct f2fs_nm_info *nm_i = NM_I(sbi);
1968 struct free_nid *i, *next;
1969 int nr = nr_shrink;
1970
1971 if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
1972 return 0;
1973
1974 if (!mutex_trylock(&nm_i->build_lock))
1975 return 0;
1976
1977 spin_lock(&nm_i->nid_list_lock);
1978 list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
1979 list) {
1980 if (nr_shrink <= 0 ||
1981 nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
1982 break;
1983
1984 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1985 kmem_cache_free(free_nid_slab, i);
1986 nr_shrink--;
1987 }
1988 spin_unlock(&nm_i->nid_list_lock);
1989 mutex_unlock(&nm_i->build_lock);
1990
1991 return nr - nr_shrink;
1992}
1993
1994void recover_inline_xattr(struct inode *inode, struct page *page)
1995{
1996 void *src_addr, *dst_addr;
1997 size_t inline_size;
1998 struct page *ipage;
1999 struct f2fs_inode *ri;
2000
2001 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2002 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2003
2004 ri = F2FS_INODE(page);
2005 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2006 clear_inode_flag(inode, FI_INLINE_XATTR);
2007 goto update_inode;
2008 }
2009
2010 dst_addr = inline_xattr_addr(ipage);
2011 src_addr = inline_xattr_addr(page);
2012 inline_size = inline_xattr_size(inode);
2013
2014 f2fs_wait_on_page_writeback(ipage, NODE, true);
2015 memcpy(dst_addr, src_addr, inline_size);
2016update_inode:
2017 update_inode(inode, ipage);
2018 f2fs_put_page(ipage, 1);
2019}
2020
2021void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2022{
2023 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2024 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2025 nid_t new_xnid = nid_of_node(page);
2026 struct node_info ni;
2027
2028 /* 1: invalidate the previous xattr nid */
2029 if (!prev_xnid)
2030 goto recover_xnid;
2031
2032 /* Deallocate node address */
2033 get_node_info(sbi, prev_xnid, &ni);
2034 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2035 invalidate_blocks(sbi, ni.blk_addr);
2036 dec_valid_node_count(sbi, inode);
2037 set_node_addr(sbi, &ni, NULL_ADDR, false);
2038
2039recover_xnid:
2040 /* 2: allocate new xattr nid */
2041 if (unlikely(!inc_valid_node_count(sbi, inode)))
2042 f2fs_bug_on(sbi, 1);
2043
2044 remove_free_nid(sbi, new_xnid);
2045 get_node_info(sbi, new_xnid, &ni);
2046 ni.ino = inode->i_ino;
2047 set_node_addr(sbi, &ni, NEW_ADDR, false);
2048 f2fs_i_xnid_write(inode, new_xnid);
2049
2050 /* 3: update xattr blkaddr */
2051 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2052 set_node_addr(sbi, &ni, blkaddr, false);
2053}
2054
2055int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2056{
2057 struct f2fs_inode *src, *dst;
2058 nid_t ino = ino_of_node(page);
2059 struct node_info old_ni, new_ni;
2060 struct page *ipage;
2061
2062 get_node_info(sbi, ino, &old_ni);
2063
2064 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2065 return -EINVAL;
2066retry:
2067 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2068 if (!ipage) {
2069 congestion_wait(BLK_RW_ASYNC, HZ/50);
2070 goto retry;
2071 }
2072
2073 /* Should not use this inode from free nid list */
2074 remove_free_nid(sbi, ino);
2075
2076 if (!PageUptodate(ipage))
2077 SetPageUptodate(ipage);
2078 fill_node_footer(ipage, ino, ino, 0, true);
2079
2080 src = F2FS_INODE(page);
2081 dst = F2FS_INODE(ipage);
2082
2083 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2084 dst->i_size = 0;
2085 dst->i_blocks = cpu_to_le64(1);
2086 dst->i_links = cpu_to_le32(1);
2087 dst->i_xattr_nid = 0;
2088 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2089
2090 new_ni = old_ni;
2091 new_ni.ino = ino;
2092
2093 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2094 WARN_ON(1);
2095 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2096 inc_valid_inode_count(sbi);
2097 set_page_dirty(ipage);
2098 f2fs_put_page(ipage, 1);
2099 return 0;
2100}
2101
2102int restore_node_summary(struct f2fs_sb_info *sbi,
2103 unsigned int segno, struct f2fs_summary_block *sum)
2104{
2105 struct f2fs_node *rn;
2106 struct f2fs_summary *sum_entry;
2107 block_t addr;
2108 int i, idx, last_offset, nrpages;
2109
2110 /* scan the node segment */
2111 last_offset = sbi->blocks_per_seg;
2112 addr = START_BLOCK(sbi, segno);
2113 sum_entry = &sum->entries[0];
2114
2115 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2116 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2117
2118 /* readahead node pages */
2119 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2120
2121 for (idx = addr; idx < addr + nrpages; idx++) {
2122 struct page *page = get_tmp_page(sbi, idx);
2123
2124 rn = F2FS_NODE(page);
2125 sum_entry->nid = rn->footer.nid;
2126 sum_entry->version = 0;
2127 sum_entry->ofs_in_node = 0;
2128 sum_entry++;
2129 f2fs_put_page(page, 1);
2130 }
2131
2132 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2133 addr + nrpages);
2134 }
2135 return 0;
2136}
2137
2138static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2139{
2140 struct f2fs_nm_info *nm_i = NM_I(sbi);
2141 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2142 struct f2fs_journal *journal = curseg->journal;
2143 int i;
2144
2145 down_write(&curseg->journal_rwsem);
2146 for (i = 0; i < nats_in_cursum(journal); i++) {
2147 struct nat_entry *ne;
2148 struct f2fs_nat_entry raw_ne;
2149 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2150
2151 raw_ne = nat_in_journal(journal, i);
2152
2153 ne = __lookup_nat_cache(nm_i, nid);
2154 if (!ne) {
2155 ne = grab_nat_entry(nm_i, nid);
2156 node_info_from_raw_nat(&ne->ni, &raw_ne);
2157 }
2158
2159 /*
2160 * if a free nat in journal has not been used after last
2161 * checkpoint, we should remove it from available nids,
2162 * since later we will add it again.
2163 */
2164 if (!get_nat_flag(ne, IS_DIRTY) &&
2165 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2166 spin_lock(&nm_i->nid_list_lock);
2167 nm_i->available_nids--;
2168 spin_unlock(&nm_i->nid_list_lock);
2169 }
2170
2171 __set_nat_cache_dirty(nm_i, ne);
2172 }
2173 update_nats_in_cursum(journal, -i);
2174 up_write(&curseg->journal_rwsem);
2175}
2176
2177static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2178 struct list_head *head, int max)
2179{
2180 struct nat_entry_set *cur;
2181
2182 if (nes->entry_cnt >= max)
2183 goto add_out;
2184
2185 list_for_each_entry(cur, head, set_list) {
2186 if (cur->entry_cnt >= nes->entry_cnt) {
2187 list_add(&nes->set_list, cur->set_list.prev);
2188 return;
2189 }
2190 }
2191add_out:
2192 list_add_tail(&nes->set_list, head);
2193}
2194
2195static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2196 struct nat_entry_set *set)
2197{
2198 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2199 struct f2fs_journal *journal = curseg->journal;
2200 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2201 bool to_journal = true;
2202 struct f2fs_nat_block *nat_blk;
2203 struct nat_entry *ne, *cur;
2204 struct page *page = NULL;
2205
2206 /*
2207 * there are two steps to flush nat entries:
2208 * #1, flush nat entries to journal in current hot data summary block.
2209 * #2, flush nat entries to nat page.
2210 */
2211 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2212 to_journal = false;
2213
2214 if (to_journal) {
2215 down_write(&curseg->journal_rwsem);
2216 } else {
2217 page = get_next_nat_page(sbi, start_nid);
2218 nat_blk = page_address(page);
2219 f2fs_bug_on(sbi, !nat_blk);
2220 }
2221
2222 /* flush dirty nats in nat entry set */
2223 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2224 struct f2fs_nat_entry *raw_ne;
2225 nid_t nid = nat_get_nid(ne);
2226 int offset;
2227
2228 if (nat_get_blkaddr(ne) == NEW_ADDR)
2229 continue;
2230
2231 if (to_journal) {
2232 offset = lookup_journal_in_cursum(journal,
2233 NAT_JOURNAL, nid, 1);
2234 f2fs_bug_on(sbi, offset < 0);
2235 raw_ne = &nat_in_journal(journal, offset);
2236 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2237 } else {
2238 raw_ne = &nat_blk->entries[nid - start_nid];
2239 }
2240 raw_nat_from_node_info(raw_ne, &ne->ni);
2241 nat_reset_flag(ne);
2242 __clear_nat_cache_dirty(NM_I(sbi), ne);
2243 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2244 add_free_nid(sbi, nid, false);
2245 spin_lock(&NM_I(sbi)->nid_list_lock);
2246 NM_I(sbi)->available_nids++;
2247 spin_unlock(&NM_I(sbi)->nid_list_lock);
2248 }
2249 }
2250
2251 if (to_journal)
2252 up_write(&curseg->journal_rwsem);
2253 else
2254 f2fs_put_page(page, 1);
2255
2256 f2fs_bug_on(sbi, set->entry_cnt);
2257
2258 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2259 kmem_cache_free(nat_entry_set_slab, set);
2260}
2261
2262/*
2263 * This function is called during the checkpointing process.
2264 */
2265void flush_nat_entries(struct f2fs_sb_info *sbi)
2266{
2267 struct f2fs_nm_info *nm_i = NM_I(sbi);
2268 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2269 struct f2fs_journal *journal = curseg->journal;
2270 struct nat_entry_set *setvec[SETVEC_SIZE];
2271 struct nat_entry_set *set, *tmp;
2272 unsigned int found;
2273 nid_t set_idx = 0;
2274 LIST_HEAD(sets);
2275
2276 if (!nm_i->dirty_nat_cnt)
2277 return;
2278
2279 down_write(&nm_i->nat_tree_lock);
2280
2281 /*
2282 * if there are no enough space in journal to store dirty nat
2283 * entries, remove all entries from journal and merge them
2284 * into nat entry set.
2285 */
2286 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2287 remove_nats_in_journal(sbi);
2288
2289 while ((found = __gang_lookup_nat_set(nm_i,
2290 set_idx, SETVEC_SIZE, setvec))) {
2291 unsigned idx;
2292 set_idx = setvec[found - 1]->set + 1;
2293 for (idx = 0; idx < found; idx++)
2294 __adjust_nat_entry_set(setvec[idx], &sets,
2295 MAX_NAT_JENTRIES(journal));
2296 }
2297
2298 /* flush dirty nats in nat entry set */
2299 list_for_each_entry_safe(set, tmp, &sets, set_list)
2300 __flush_nat_entry_set(sbi, set);
2301
2302 up_write(&nm_i->nat_tree_lock);
2303
2304 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2305}
2306
2307static int init_node_manager(struct f2fs_sb_info *sbi)
2308{
2309 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2310 struct f2fs_nm_info *nm_i = NM_I(sbi);
2311 unsigned char *version_bitmap;
2312 unsigned int nat_segs, nat_blocks;
2313
2314 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2315
2316 /* segment_count_nat includes pair segment so divide to 2. */
2317 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2318 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2319
2320 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2321
2322 /* not used nids: 0, node, meta, (and root counted as valid node) */
2323 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2324 F2FS_RESERVED_NODE_NUM;
2325 nm_i->nid_cnt[FREE_NID_LIST] = 0;
2326 nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2327 nm_i->nat_cnt = 0;
2328 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2329 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2330 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2331
2332 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2333 INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2334 INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2335 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2336 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2337 INIT_LIST_HEAD(&nm_i->nat_entries);
2338
2339 mutex_init(&nm_i->build_lock);
2340 spin_lock_init(&nm_i->nid_list_lock);
2341 init_rwsem(&nm_i->nat_tree_lock);
2342
2343 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2344 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2345 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2346 if (!version_bitmap)
2347 return -EFAULT;
2348
2349 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2350 GFP_KERNEL);
2351 if (!nm_i->nat_bitmap)
2352 return -ENOMEM;
2353 return 0;
2354}
2355
2356int build_node_manager(struct f2fs_sb_info *sbi)
2357{
2358 int err;
2359
2360 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2361 if (!sbi->nm_info)
2362 return -ENOMEM;
2363
2364 err = init_node_manager(sbi);
2365 if (err)
2366 return err;
2367
2368 build_free_nids(sbi, true);
2369 return 0;
2370}
2371
2372void destroy_node_manager(struct f2fs_sb_info *sbi)
2373{
2374 struct f2fs_nm_info *nm_i = NM_I(sbi);
2375 struct free_nid *i, *next_i;
2376 struct nat_entry *natvec[NATVEC_SIZE];
2377 struct nat_entry_set *setvec[SETVEC_SIZE];
2378 nid_t nid = 0;
2379 unsigned int found;
2380
2381 if (!nm_i)
2382 return;
2383
2384 /* destroy free nid list */
2385 spin_lock(&nm_i->nid_list_lock);
2386 list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2387 list) {
2388 __remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2389 spin_unlock(&nm_i->nid_list_lock);
2390 kmem_cache_free(free_nid_slab, i);
2391 spin_lock(&nm_i->nid_list_lock);
2392 }
2393 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2394 f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2395 f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2396 spin_unlock(&nm_i->nid_list_lock);
2397
2398 /* destroy nat cache */
2399 down_write(&nm_i->nat_tree_lock);
2400 while ((found = __gang_lookup_nat_cache(nm_i,
2401 nid, NATVEC_SIZE, natvec))) {
2402 unsigned idx;
2403
2404 nid = nat_get_nid(natvec[found - 1]) + 1;
2405 for (idx = 0; idx < found; idx++)
2406 __del_from_nat_cache(nm_i, natvec[idx]);
2407 }
2408 f2fs_bug_on(sbi, nm_i->nat_cnt);
2409
2410 /* destroy nat set cache */
2411 nid = 0;
2412 while ((found = __gang_lookup_nat_set(nm_i,
2413 nid, SETVEC_SIZE, setvec))) {
2414 unsigned idx;
2415
2416 nid = setvec[found - 1]->set + 1;
2417 for (idx = 0; idx < found; idx++) {
2418 /* entry_cnt is not zero, when cp_error was occurred */
2419 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2420 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2421 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2422 }
2423 }
2424 up_write(&nm_i->nat_tree_lock);
2425
2426 kfree(nm_i->nat_bitmap);
2427 sbi->nm_info = NULL;
2428 kfree(nm_i);
2429}
2430
2431int __init create_node_manager_caches(void)
2432{
2433 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2434 sizeof(struct nat_entry));
2435 if (!nat_entry_slab)
2436 goto fail;
2437
2438 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2439 sizeof(struct free_nid));
2440 if (!free_nid_slab)
2441 goto destroy_nat_entry;
2442
2443 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2444 sizeof(struct nat_entry_set));
2445 if (!nat_entry_set_slab)
2446 goto destroy_free_nid;
2447 return 0;
2448
2449destroy_free_nid:
2450 kmem_cache_destroy(free_nid_slab);
2451destroy_nat_entry:
2452 kmem_cache_destroy(nat_entry_slab);
2453fail:
2454 return -ENOMEM;
2455}
2456
2457void destroy_node_manager_caches(void)
2458{
2459 kmem_cache_destroy(nat_entry_set_slab);
2460 kmem_cache_destroy(free_nid_slab);
2461 kmem_cache_destroy(nat_entry_slab);
2462}