projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
f2fs: remove syncing inode page in all the cases
[linux-block.git] / fs / f2fs / node.c
... / ...
CommitLineData
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->fcnt * sizeof(struct free_nid)) >>
49 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 } else if (type == DIRTY_DENTS) {
56 if (sbi->sb->s_bdi->wb.dirty_exceeded)
57 return false;
58 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
59 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
60 } else if (type == INO_ENTRIES) {
61 int i;
62
63 for (i = 0; i <= UPDATE_INO; i++)
64 mem_size += (sbi->im[i].ino_num *
65 sizeof(struct ino_entry)) >> PAGE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
67 } else if (type == EXTENT_CACHE) {
68 mem_size = (atomic_read(&sbi->total_ext_tree) *
69 sizeof(struct extent_tree) +
70 atomic_read(&sbi->total_ext_node) *
71 sizeof(struct extent_node)) >> PAGE_SHIFT;
72 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
73 } else {
74 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
75 return true;
76 }
77 return res;
78}
79
80static void clear_node_page_dirty(struct page *page)
81{
82 struct address_space *mapping = page->mapping;
83 unsigned int long flags;
84
85 if (PageDirty(page)) {
86 spin_lock_irqsave(&mapping->tree_lock, flags);
87 radix_tree_tag_clear(&mapping->page_tree,
88 page_index(page),
89 PAGECACHE_TAG_DIRTY);
90 spin_unlock_irqrestore(&mapping->tree_lock, flags);
91
92 clear_page_dirty_for_io(page);
93 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
94 }
95 ClearPageUptodate(page);
96}
97
98static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
99{
100 pgoff_t index = current_nat_addr(sbi, nid);
101 return get_meta_page(sbi, index);
102}
103
104static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
105{
106 struct page *src_page;
107 struct page *dst_page;
108 pgoff_t src_off;
109 pgoff_t dst_off;
110 void *src_addr;
111 void *dst_addr;
112 struct f2fs_nm_info *nm_i = NM_I(sbi);
113
114 src_off = current_nat_addr(sbi, nid);
115 dst_off = next_nat_addr(sbi, src_off);
116
117 /* get current nat block page with lock */
118 src_page = get_meta_page(sbi, src_off);
119 dst_page = grab_meta_page(sbi, dst_off);
120 f2fs_bug_on(sbi, PageDirty(src_page));
121
122 src_addr = page_address(src_page);
123 dst_addr = page_address(dst_page);
124 memcpy(dst_addr, src_addr, PAGE_SIZE);
125 set_page_dirty(dst_page);
126 f2fs_put_page(src_page, 1);
127
128 set_to_next_nat(nm_i, nid);
129
130 return dst_page;
131}
132
133static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
134{
135 return radix_tree_lookup(&nm_i->nat_root, n);
136}
137
138static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
139 nid_t start, unsigned int nr, struct nat_entry **ep)
140{
141 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
142}
143
144static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
145{
146 list_del(&e->list);
147 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
148 nm_i->nat_cnt--;
149 kmem_cache_free(nat_entry_slab, e);
150}
151
152static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
153 struct nat_entry *ne)
154{
155 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
156 struct nat_entry_set *head;
157
158 if (get_nat_flag(ne, IS_DIRTY))
159 return;
160
161 head = radix_tree_lookup(&nm_i->nat_set_root, set);
162 if (!head) {
163 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
164
165 INIT_LIST_HEAD(&head->entry_list);
166 INIT_LIST_HEAD(&head->set_list);
167 head->set = set;
168 head->entry_cnt = 0;
169 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
170 }
171 list_move_tail(&ne->list, &head->entry_list);
172 nm_i->dirty_nat_cnt++;
173 head->entry_cnt++;
174 set_nat_flag(ne, IS_DIRTY, true);
175}
176
177static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
178 struct nat_entry *ne)
179{
180 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
181 struct nat_entry_set *head;
182
183 head = radix_tree_lookup(&nm_i->nat_set_root, set);
184 if (head) {
185 list_move_tail(&ne->list, &nm_i->nat_entries);
186 set_nat_flag(ne, IS_DIRTY, false);
187 head->entry_cnt--;
188 nm_i->dirty_nat_cnt--;
189 }
190}
191
192static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
193 nid_t start, unsigned int nr, struct nat_entry_set **ep)
194{
195 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
196 start, nr);
197}
198
199int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
200{
201 struct f2fs_nm_info *nm_i = NM_I(sbi);
202 struct nat_entry *e;
203 bool need = false;
204
205 down_read(&nm_i->nat_tree_lock);
206 e = __lookup_nat_cache(nm_i, nid);
207 if (e) {
208 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
209 !get_nat_flag(e, HAS_FSYNCED_INODE))
210 need = true;
211 }
212 up_read(&nm_i->nat_tree_lock);
213 return need;
214}
215
216bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
217{
218 struct f2fs_nm_info *nm_i = NM_I(sbi);
219 struct nat_entry *e;
220 bool is_cp = true;
221
222 down_read(&nm_i->nat_tree_lock);
223 e = __lookup_nat_cache(nm_i, nid);
224 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
225 is_cp = false;
226 up_read(&nm_i->nat_tree_lock);
227 return is_cp;
228}
229
230bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
231{
232 struct f2fs_nm_info *nm_i = NM_I(sbi);
233 struct nat_entry *e;
234 bool need_update = true;
235
236 down_read(&nm_i->nat_tree_lock);
237 e = __lookup_nat_cache(nm_i, ino);
238 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
239 (get_nat_flag(e, IS_CHECKPOINTED) ||
240 get_nat_flag(e, HAS_FSYNCED_INODE)))
241 need_update = false;
242 up_read(&nm_i->nat_tree_lock);
243 return need_update;
244}
245
246static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
247{
248 struct nat_entry *new;
249
250 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
251 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
252 memset(new, 0, sizeof(struct nat_entry));
253 nat_set_nid(new, nid);
254 nat_reset_flag(new);
255 list_add_tail(&new->list, &nm_i->nat_entries);
256 nm_i->nat_cnt++;
257 return new;
258}
259
260static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
261 struct f2fs_nat_entry *ne)
262{
263 struct f2fs_nm_info *nm_i = NM_I(sbi);
264 struct nat_entry *e;
265
266 e = __lookup_nat_cache(nm_i, nid);
267 if (!e) {
268 e = grab_nat_entry(nm_i, nid);
269 node_info_from_raw_nat(&e->ni, ne);
270 } else {
271 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
272 nat_get_blkaddr(e) != ne->block_addr ||
273 nat_get_version(e) != ne->version);
274 }
275}
276
277static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
278 block_t new_blkaddr, bool fsync_done)
279{
280 struct f2fs_nm_info *nm_i = NM_I(sbi);
281 struct nat_entry *e;
282
283 down_write(&nm_i->nat_tree_lock);
284 e = __lookup_nat_cache(nm_i, ni->nid);
285 if (!e) {
286 e = grab_nat_entry(nm_i, ni->nid);
287 copy_node_info(&e->ni, ni);
288 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
289 } else if (new_blkaddr == NEW_ADDR) {
290 /*
291 * when nid is reallocated,
292 * previous nat entry can be remained in nat cache.
293 * So, reinitialize it with new information.
294 */
295 copy_node_info(&e->ni, ni);
296 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
297 }
298
299 /* sanity check */
300 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
301 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
302 new_blkaddr == NULL_ADDR);
303 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
304 new_blkaddr == NEW_ADDR);
305 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
306 nat_get_blkaddr(e) != NULL_ADDR &&
307 new_blkaddr == NEW_ADDR);
308
309 /* increment version no as node is removed */
310 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
311 unsigned char version = nat_get_version(e);
312 nat_set_version(e, inc_node_version(version));
313
314 /* in order to reuse the nid */
315 if (nm_i->next_scan_nid > ni->nid)
316 nm_i->next_scan_nid = ni->nid;
317 }
318
319 /* change address */
320 nat_set_blkaddr(e, new_blkaddr);
321 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
322 set_nat_flag(e, IS_CHECKPOINTED, false);
323 __set_nat_cache_dirty(nm_i, e);
324
325 /* update fsync_mark if its inode nat entry is still alive */
326 if (ni->nid != ni->ino)
327 e = __lookup_nat_cache(nm_i, ni->ino);
328 if (e) {
329 if (fsync_done && ni->nid == ni->ino)
330 set_nat_flag(e, HAS_FSYNCED_INODE, true);
331 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
332 }
333 up_write(&nm_i->nat_tree_lock);
334}
335
336int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
337{
338 struct f2fs_nm_info *nm_i = NM_I(sbi);
339 int nr = nr_shrink;
340
341 if (!down_write_trylock(&nm_i->nat_tree_lock))
342 return 0;
343
344 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
345 struct nat_entry *ne;
346 ne = list_first_entry(&nm_i->nat_entries,
347 struct nat_entry, list);
348 __del_from_nat_cache(nm_i, ne);
349 nr_shrink--;
350 }
351 up_write(&nm_i->nat_tree_lock);
352 return nr - nr_shrink;
353}
354
355/*
356 * This function always returns success
357 */
358void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
359{
360 struct f2fs_nm_info *nm_i = NM_I(sbi);
361 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
362 struct f2fs_journal *journal = curseg->journal;
363 nid_t start_nid = START_NID(nid);
364 struct f2fs_nat_block *nat_blk;
365 struct page *page = NULL;
366 struct f2fs_nat_entry ne;
367 struct nat_entry *e;
368 int i;
369
370 ni->nid = nid;
371
372 /* Check nat cache */
373 down_read(&nm_i->nat_tree_lock);
374 e = __lookup_nat_cache(nm_i, nid);
375 if (e) {
376 ni->ino = nat_get_ino(e);
377 ni->blk_addr = nat_get_blkaddr(e);
378 ni->version = nat_get_version(e);
379 up_read(&nm_i->nat_tree_lock);
380 return;
381 }
382
383 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
384
385 /* Check current segment summary */
386 down_read(&curseg->journal_rwsem);
387 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
388 if (i >= 0) {
389 ne = nat_in_journal(journal, i);
390 node_info_from_raw_nat(ni, &ne);
391 }
392 up_read(&curseg->journal_rwsem);
393 if (i >= 0)
394 goto cache;
395
396 /* Fill node_info from nat page */
397 page = get_current_nat_page(sbi, start_nid);
398 nat_blk = (struct f2fs_nat_block *)page_address(page);
399 ne = nat_blk->entries[nid - start_nid];
400 node_info_from_raw_nat(ni, &ne);
401 f2fs_put_page(page, 1);
402cache:
403 up_read(&nm_i->nat_tree_lock);
404 /* cache nat entry */
405 down_write(&nm_i->nat_tree_lock);
406 cache_nat_entry(sbi, nid, &ne);
407 up_write(&nm_i->nat_tree_lock);
408}
409
410/*
411 * readahead MAX_RA_NODE number of node pages.
412 */
413static void ra_node_pages(struct page *parent, int start, int n)
414{
415 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
416 struct blk_plug plug;
417 int i, end;
418 nid_t nid;
419
420 blk_start_plug(&plug);
421
422 /* Then, try readahead for siblings of the desired node */
423 end = start + n;
424 end = min(end, NIDS_PER_BLOCK);
425 for (i = start; i < end; i++) {
426 nid = get_nid(parent, i, false);
427 ra_node_page(sbi, nid);
428 }
429
430 blk_finish_plug(&plug);
431}
432
433pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
434{
435 const long direct_index = ADDRS_PER_INODE(dn->inode);
436 const long direct_blks = ADDRS_PER_BLOCK;
437 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
438 unsigned int skipped_unit = ADDRS_PER_BLOCK;
439 int cur_level = dn->cur_level;
440 int max_level = dn->max_level;
441 pgoff_t base = 0;
442
443 if (!dn->max_level)
444 return pgofs + 1;
445
446 while (max_level-- > cur_level)
447 skipped_unit *= NIDS_PER_BLOCK;
448
449 switch (dn->max_level) {
450 case 3:
451 base += 2 * indirect_blks;
452 case 2:
453 base += 2 * direct_blks;
454 case 1:
455 base += direct_index;
456 break;
457 default:
458 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
459 }
460
461 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
462}
463
464/*
465 * The maximum depth is four.
466 * Offset[0] will have raw inode offset.
467 */
468static int get_node_path(struct inode *inode, long block,
469 int offset[4], unsigned int noffset[4])
470{
471 const long direct_index = ADDRS_PER_INODE(inode);
472 const long direct_blks = ADDRS_PER_BLOCK;
473 const long dptrs_per_blk = NIDS_PER_BLOCK;
474 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
475 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
476 int n = 0;
477 int level = 0;
478
479 noffset[0] = 0;
480
481 if (block < direct_index) {
482 offset[n] = block;
483 goto got;
484 }
485 block -= direct_index;
486 if (block < direct_blks) {
487 offset[n++] = NODE_DIR1_BLOCK;
488 noffset[n] = 1;
489 offset[n] = block;
490 level = 1;
491 goto got;
492 }
493 block -= direct_blks;
494 if (block < direct_blks) {
495 offset[n++] = NODE_DIR2_BLOCK;
496 noffset[n] = 2;
497 offset[n] = block;
498 level = 1;
499 goto got;
500 }
501 block -= direct_blks;
502 if (block < indirect_blks) {
503 offset[n++] = NODE_IND1_BLOCK;
504 noffset[n] = 3;
505 offset[n++] = block / direct_blks;
506 noffset[n] = 4 + offset[n - 1];
507 offset[n] = block % direct_blks;
508 level = 2;
509 goto got;
510 }
511 block -= indirect_blks;
512 if (block < indirect_blks) {
513 offset[n++] = NODE_IND2_BLOCK;
514 noffset[n] = 4 + dptrs_per_blk;
515 offset[n++] = block / direct_blks;
516 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
517 offset[n] = block % direct_blks;
518 level = 2;
519 goto got;
520 }
521 block -= indirect_blks;
522 if (block < dindirect_blks) {
523 offset[n++] = NODE_DIND_BLOCK;
524 noffset[n] = 5 + (dptrs_per_blk * 2);
525 offset[n++] = block / indirect_blks;
526 noffset[n] = 6 + (dptrs_per_blk * 2) +
527 offset[n - 1] * (dptrs_per_blk + 1);
528 offset[n++] = (block / direct_blks) % dptrs_per_blk;
529 noffset[n] = 7 + (dptrs_per_blk * 2) +
530 offset[n - 2] * (dptrs_per_blk + 1) +
531 offset[n - 1];
532 offset[n] = block % direct_blks;
533 level = 3;
534 goto got;
535 } else {
536 BUG();
537 }
538got:
539 return level;
540}
541
542/*
543 * Caller should call f2fs_put_dnode(dn).
544 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
545 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
546 * In the case of RDONLY_NODE, we don't need to care about mutex.
547 */
548int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
549{
550 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
551 struct page *npage[4];
552 struct page *parent = NULL;
553 int offset[4];
554 unsigned int noffset[4];
555 nid_t nids[4];
556 int level, i = 0;
557 int err = 0;
558
559 level = get_node_path(dn->inode, index, offset, noffset);
560
561 nids[0] = dn->inode->i_ino;
562 npage[0] = dn->inode_page;
563
564 if (!npage[0]) {
565 npage[0] = get_node_page(sbi, nids[0]);
566 if (IS_ERR(npage[0]))
567 return PTR_ERR(npage[0]);
568 }
569
570 /* if inline_data is set, should not report any block indices */
571 if (f2fs_has_inline_data(dn->inode) && index) {
572 err = -ENOENT;
573 f2fs_put_page(npage[0], 1);
574 goto release_out;
575 }
576
577 parent = npage[0];
578 if (level != 0)
579 nids[1] = get_nid(parent, offset[0], true);
580 dn->inode_page = npage[0];
581 dn->inode_page_locked = true;
582
583 /* get indirect or direct nodes */
584 for (i = 1; i <= level; i++) {
585 bool done = false;
586
587 if (!nids[i] && mode == ALLOC_NODE) {
588 /* alloc new node */
589 if (!alloc_nid(sbi, &(nids[i]))) {
590 err = -ENOSPC;
591 goto release_pages;
592 }
593
594 dn->nid = nids[i];
595 npage[i] = new_node_page(dn, noffset[i], NULL);
596 if (IS_ERR(npage[i])) {
597 alloc_nid_failed(sbi, nids[i]);
598 err = PTR_ERR(npage[i]);
599 goto release_pages;
600 }
601
602 set_nid(parent, offset[i - 1], nids[i], i == 1);
603 alloc_nid_done(sbi, nids[i]);
604 done = true;
605 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
606 npage[i] = get_node_page_ra(parent, offset[i - 1]);
607 if (IS_ERR(npage[i])) {
608 err = PTR_ERR(npage[i]);
609 goto release_pages;
610 }
611 done = true;
612 }
613 if (i == 1) {
614 dn->inode_page_locked = false;
615 unlock_page(parent);
616 } else {
617 f2fs_put_page(parent, 1);
618 }
619
620 if (!done) {
621 npage[i] = get_node_page(sbi, nids[i]);
622 if (IS_ERR(npage[i])) {
623 err = PTR_ERR(npage[i]);
624 f2fs_put_page(npage[0], 0);
625 goto release_out;
626 }
627 }
628 if (i < level) {
629 parent = npage[i];
630 nids[i + 1] = get_nid(parent, offset[i], false);
631 }
632 }
633 dn->nid = nids[level];
634 dn->ofs_in_node = offset[level];
635 dn->node_page = npage[level];
636 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
637 return 0;
638
639release_pages:
640 f2fs_put_page(parent, 1);
641 if (i > 1)
642 f2fs_put_page(npage[0], 0);
643release_out:
644 dn->inode_page = NULL;
645 dn->node_page = NULL;
646 if (err == -ENOENT) {
647 dn->cur_level = i;
648 dn->max_level = level;
649 }
650 return err;
651}
652
653static void truncate_node(struct dnode_of_data *dn)
654{
655 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
656 struct node_info ni;
657
658 get_node_info(sbi, dn->nid, &ni);
659 if (dn->inode->i_blocks == 0) {
660 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
661 goto invalidate;
662 }
663 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
664
665 /* Deallocate node address */
666 invalidate_blocks(sbi, ni.blk_addr);
667 dec_valid_node_count(sbi, dn->inode);
668 set_node_addr(sbi, &ni, NULL_ADDR, false);
669
670 if (dn->nid == dn->inode->i_ino) {
671 remove_orphan_inode(sbi, dn->nid);
672 dec_valid_inode_count(sbi);
673 f2fs_inode_synced(dn->inode);
674 }
675invalidate:
676 clear_node_page_dirty(dn->node_page);
677 set_sbi_flag(sbi, SBI_IS_DIRTY);
678
679 f2fs_put_page(dn->node_page, 1);
680
681 invalidate_mapping_pages(NODE_MAPPING(sbi),
682 dn->node_page->index, dn->node_page->index);
683
684 dn->node_page = NULL;
685 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
686}
687
688static int truncate_dnode(struct dnode_of_data *dn)
689{
690 struct page *page;
691
692 if (dn->nid == 0)
693 return 1;
694
695 /* get direct node */
696 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
697 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
698 return 1;
699 else if (IS_ERR(page))
700 return PTR_ERR(page);
701
702 /* Make dnode_of_data for parameter */
703 dn->node_page = page;
704 dn->ofs_in_node = 0;
705 truncate_data_blocks(dn);
706 truncate_node(dn);
707 return 1;
708}
709
710static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
711 int ofs, int depth)
712{
713 struct dnode_of_data rdn = *dn;
714 struct page *page;
715 struct f2fs_node *rn;
716 nid_t child_nid;
717 unsigned int child_nofs;
718 int freed = 0;
719 int i, ret;
720
721 if (dn->nid == 0)
722 return NIDS_PER_BLOCK + 1;
723
724 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
725
726 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
727 if (IS_ERR(page)) {
728 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
729 return PTR_ERR(page);
730 }
731
732 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
733
734 rn = F2FS_NODE(page);
735 if (depth < 3) {
736 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
737 child_nid = le32_to_cpu(rn->in.nid[i]);
738 if (child_nid == 0)
739 continue;
740 rdn.nid = child_nid;
741 ret = truncate_dnode(&rdn);
742 if (ret < 0)
743 goto out_err;
744 if (set_nid(page, i, 0, false))
745 dn->node_changed = true;
746 }
747 } else {
748 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
749 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
750 child_nid = le32_to_cpu(rn->in.nid[i]);
751 if (child_nid == 0) {
752 child_nofs += NIDS_PER_BLOCK + 1;
753 continue;
754 }
755 rdn.nid = child_nid;
756 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
757 if (ret == (NIDS_PER_BLOCK + 1)) {
758 if (set_nid(page, i, 0, false))
759 dn->node_changed = true;
760 child_nofs += ret;
761 } else if (ret < 0 && ret != -ENOENT) {
762 goto out_err;
763 }
764 }
765 freed = child_nofs;
766 }
767
768 if (!ofs) {
769 /* remove current indirect node */
770 dn->node_page = page;
771 truncate_node(dn);
772 freed++;
773 } else {
774 f2fs_put_page(page, 1);
775 }
776 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
777 return freed;
778
779out_err:
780 f2fs_put_page(page, 1);
781 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
782 return ret;
783}
784
785static int truncate_partial_nodes(struct dnode_of_data *dn,
786 struct f2fs_inode *ri, int *offset, int depth)
787{
788 struct page *pages[2];
789 nid_t nid[3];
790 nid_t child_nid;
791 int err = 0;
792 int i;
793 int idx = depth - 2;
794
795 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
796 if (!nid[0])
797 return 0;
798
799 /* get indirect nodes in the path */
800 for (i = 0; i < idx + 1; i++) {
801 /* reference count'll be increased */
802 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
803 if (IS_ERR(pages[i])) {
804 err = PTR_ERR(pages[i]);
805 idx = i - 1;
806 goto fail;
807 }
808 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
809 }
810
811 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
812
813 /* free direct nodes linked to a partial indirect node */
814 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
815 child_nid = get_nid(pages[idx], i, false);
816 if (!child_nid)
817 continue;
818 dn->nid = child_nid;
819 err = truncate_dnode(dn);
820 if (err < 0)
821 goto fail;
822 if (set_nid(pages[idx], i, 0, false))
823 dn->node_changed = true;
824 }
825
826 if (offset[idx + 1] == 0) {
827 dn->node_page = pages[idx];
828 dn->nid = nid[idx];
829 truncate_node(dn);
830 } else {
831 f2fs_put_page(pages[idx], 1);
832 }
833 offset[idx]++;
834 offset[idx + 1] = 0;
835 idx--;
836fail:
837 for (i = idx; i >= 0; i--)
838 f2fs_put_page(pages[i], 1);
839
840 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
841
842 return err;
843}
844
845/*
846 * All the block addresses of data and nodes should be nullified.
847 */
848int truncate_inode_blocks(struct inode *inode, pgoff_t from)
849{
850 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
851 int err = 0, cont = 1;
852 int level, offset[4], noffset[4];
853 unsigned int nofs = 0;
854 struct f2fs_inode *ri;
855 struct dnode_of_data dn;
856 struct page *page;
857
858 trace_f2fs_truncate_inode_blocks_enter(inode, from);
859
860 level = get_node_path(inode, from, offset, noffset);
861
862 page = get_node_page(sbi, inode->i_ino);
863 if (IS_ERR(page)) {
864 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
865 return PTR_ERR(page);
866 }
867
868 set_new_dnode(&dn, inode, page, NULL, 0);
869 unlock_page(page);
870
871 ri = F2FS_INODE(page);
872 switch (level) {
873 case 0:
874 case 1:
875 nofs = noffset[1];
876 break;
877 case 2:
878 nofs = noffset[1];
879 if (!offset[level - 1])
880 goto skip_partial;
881 err = truncate_partial_nodes(&dn, ri, offset, level);
882 if (err < 0 && err != -ENOENT)
883 goto fail;
884 nofs += 1 + NIDS_PER_BLOCK;
885 break;
886 case 3:
887 nofs = 5 + 2 * NIDS_PER_BLOCK;
888 if (!offset[level - 1])
889 goto skip_partial;
890 err = truncate_partial_nodes(&dn, ri, offset, level);
891 if (err < 0 && err != -ENOENT)
892 goto fail;
893 break;
894 default:
895 BUG();
896 }
897
898skip_partial:
899 while (cont) {
900 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
901 switch (offset[0]) {
902 case NODE_DIR1_BLOCK:
903 case NODE_DIR2_BLOCK:
904 err = truncate_dnode(&dn);
905 break;
906
907 case NODE_IND1_BLOCK:
908 case NODE_IND2_BLOCK:
909 err = truncate_nodes(&dn, nofs, offset[1], 2);
910 break;
911
912 case NODE_DIND_BLOCK:
913 err = truncate_nodes(&dn, nofs, offset[1], 3);
914 cont = 0;
915 break;
916
917 default:
918 BUG();
919 }
920 if (err < 0 && err != -ENOENT)
921 goto fail;
922 if (offset[1] == 0 &&
923 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
924 lock_page(page);
925 BUG_ON(page->mapping != NODE_MAPPING(sbi));
926 f2fs_wait_on_page_writeback(page, NODE, true);
927 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
928 set_page_dirty(page);
929 unlock_page(page);
930 }
931 offset[1] = 0;
932 offset[0]++;
933 nofs += err;
934 }
935fail:
936 f2fs_put_page(page, 0);
937 trace_f2fs_truncate_inode_blocks_exit(inode, err);
938 return err > 0 ? 0 : err;
939}
940
941int truncate_xattr_node(struct inode *inode, struct page *page)
942{
943 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
944 nid_t nid = F2FS_I(inode)->i_xattr_nid;
945 struct dnode_of_data dn;
946 struct page *npage;
947
948 if (!nid)
949 return 0;
950
951 npage = get_node_page(sbi, nid);
952 if (IS_ERR(npage))
953 return PTR_ERR(npage);
954
955 f2fs_i_xnid_write(inode, 0);
956
957 /* need to do checkpoint during fsync */
958 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
959
960 set_new_dnode(&dn, inode, page, npage, nid);
961
962 if (page)
963 dn.inode_page_locked = true;
964 truncate_node(&dn);
965 return 0;
966}
967
968/*
969 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
970 * f2fs_unlock_op().
971 */
972int remove_inode_page(struct inode *inode)
973{
974 struct dnode_of_data dn;
975 int err;
976
977 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
978 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
979 if (err)
980 return err;
981
982 err = truncate_xattr_node(inode, dn.inode_page);
983 if (err) {
984 f2fs_put_dnode(&dn);
985 return err;
986 }
987
988 /* remove potential inline_data blocks */
989 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
990 S_ISLNK(inode->i_mode))
991 truncate_data_blocks_range(&dn, 1);
992
993 /* 0 is possible, after f2fs_new_inode() has failed */
994 f2fs_bug_on(F2FS_I_SB(inode),
995 inode->i_blocks != 0 && inode->i_blocks != 1);
996
997 /* will put inode & node pages */
998 truncate_node(&dn);
999 return 0;
1000}
1001
1002struct page *new_inode_page(struct inode *inode)
1003{
1004 struct dnode_of_data dn;
1005
1006 /* allocate inode page for new inode */
1007 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1008
1009 /* caller should f2fs_put_page(page, 1); */
1010 return new_node_page(&dn, 0, NULL);
1011}
1012
1013struct page *new_node_page(struct dnode_of_data *dn,
1014 unsigned int ofs, struct page *ipage)
1015{
1016 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1017 struct node_info old_ni, new_ni;
1018 struct page *page;
1019 int err;
1020
1021 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1022 return ERR_PTR(-EPERM);
1023
1024 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1025 if (!page)
1026 return ERR_PTR(-ENOMEM);
1027
1028 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1029 err = -ENOSPC;
1030 goto fail;
1031 }
1032
1033 get_node_info(sbi, dn->nid, &old_ni);
1034
1035 /* Reinitialize old_ni with new node page */
1036 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1037 new_ni = old_ni;
1038 new_ni.ino = dn->inode->i_ino;
1039 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1040
1041 f2fs_wait_on_page_writeback(page, NODE, true);
1042 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1043 set_cold_node(dn->inode, page);
1044 SetPageUptodate(page);
1045 if (set_page_dirty(page))
1046 dn->node_changed = true;
1047
1048 if (f2fs_has_xattr_block(ofs))
1049 f2fs_i_xnid_write(dn->inode, dn->nid);
1050
1051 if (ofs == 0)
1052 inc_valid_inode_count(sbi);
1053 return page;
1054
1055fail:
1056 clear_node_page_dirty(page);
1057 f2fs_put_page(page, 1);
1058 return ERR_PTR(err);
1059}
1060
1061/*
1062 * Caller should do after getting the following values.
1063 * 0: f2fs_put_page(page, 0)
1064 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1065 */
1066static int read_node_page(struct page *page, int rw)
1067{
1068 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1069 struct node_info ni;
1070 struct f2fs_io_info fio = {
1071 .sbi = sbi,
1072 .type = NODE,
1073 .rw = rw,
1074 .page = page,
1075 .encrypted_page = NULL,
1076 };
1077
1078 get_node_info(sbi, page->index, &ni);
1079
1080 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1081 ClearPageUptodate(page);
1082 return -ENOENT;
1083 }
1084
1085 if (PageUptodate(page))
1086 return LOCKED_PAGE;
1087
1088 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1089 return f2fs_submit_page_bio(&fio);
1090}
1091
1092/*
1093 * Readahead a node page
1094 */
1095void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1096{
1097 struct page *apage;
1098 int err;
1099
1100 if (!nid)
1101 return;
1102 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1103
1104 rcu_read_lock();
1105 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1106 rcu_read_unlock();
1107 if (apage)
1108 return;
1109
1110 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1111 if (!apage)
1112 return;
1113
1114 err = read_node_page(apage, READA);
1115 f2fs_put_page(apage, err ? 1 : 0);
1116}
1117
1118static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1119 struct page *parent, int start)
1120{
1121 struct page *page;
1122 int err;
1123
1124 if (!nid)
1125 return ERR_PTR(-ENOENT);
1126 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1127repeat:
1128 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1129 if (!page)
1130 return ERR_PTR(-ENOMEM);
1131
1132 err = read_node_page(page, READ_SYNC);
1133 if (err < 0) {
1134 f2fs_put_page(page, 1);
1135 return ERR_PTR(err);
1136 } else if (err == LOCKED_PAGE) {
1137 goto page_hit;
1138 }
1139
1140 if (parent)
1141 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1142
1143 lock_page(page);
1144
1145 if (unlikely(!PageUptodate(page))) {
1146 f2fs_put_page(page, 1);
1147 return ERR_PTR(-EIO);
1148 }
1149 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1150 f2fs_put_page(page, 1);
1151 goto repeat;
1152 }
1153page_hit:
1154 f2fs_bug_on(sbi, nid != nid_of_node(page));
1155 return page;
1156}
1157
1158struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1159{
1160 return __get_node_page(sbi, nid, NULL, 0);
1161}
1162
1163struct page *get_node_page_ra(struct page *parent, int start)
1164{
1165 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1166 nid_t nid = get_nid(parent, start, false);
1167
1168 return __get_node_page(sbi, nid, parent, start);
1169}
1170
1171static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1172{
1173 struct inode *inode;
1174 struct page *page;
1175 int ret;
1176
1177 /* should flush inline_data before evict_inode */
1178 inode = ilookup(sbi->sb, ino);
1179 if (!inode)
1180 return;
1181
1182 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1183 if (!page)
1184 goto iput_out;
1185
1186 if (!PageUptodate(page))
1187 goto page_out;
1188
1189 if (!PageDirty(page))
1190 goto page_out;
1191
1192 if (!clear_page_dirty_for_io(page))
1193 goto page_out;
1194
1195 ret = f2fs_write_inline_data(inode, page);
1196 inode_dec_dirty_pages(inode);
1197 if (ret)
1198 set_page_dirty(page);
1199page_out:
1200 f2fs_put_page(page, 1);
1201iput_out:
1202 iput(inode);
1203}
1204
1205void move_node_page(struct page *node_page, int gc_type)
1206{
1207 if (gc_type == FG_GC) {
1208 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1209 struct writeback_control wbc = {
1210 .sync_mode = WB_SYNC_ALL,
1211 .nr_to_write = 1,
1212 .for_reclaim = 0,
1213 };
1214
1215 set_page_dirty(node_page);
1216 f2fs_wait_on_page_writeback(node_page, NODE, true);
1217
1218 f2fs_bug_on(sbi, PageWriteback(node_page));
1219 if (!clear_page_dirty_for_io(node_page))
1220 goto out_page;
1221
1222 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1223 unlock_page(node_page);
1224 goto release_page;
1225 } else {
1226 /* set page dirty and write it */
1227 if (!PageWriteback(node_page))
1228 set_page_dirty(node_page);
1229 }
1230out_page:
1231 unlock_page(node_page);
1232release_page:
1233 f2fs_put_page(node_page, 0);
1234}
1235
1236static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1237{
1238 pgoff_t index, end;
1239 struct pagevec pvec;
1240 struct page *last_page = NULL;
1241
1242 pagevec_init(&pvec, 0);
1243 index = 0;
1244 end = ULONG_MAX;
1245
1246 while (index <= end) {
1247 int i, nr_pages;
1248 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1249 PAGECACHE_TAG_DIRTY,
1250 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1251 if (nr_pages == 0)
1252 break;
1253
1254 for (i = 0; i < nr_pages; i++) {
1255 struct page *page = pvec.pages[i];
1256
1257 if (unlikely(f2fs_cp_error(sbi))) {
1258 f2fs_put_page(last_page, 0);
1259 pagevec_release(&pvec);
1260 return ERR_PTR(-EIO);
1261 }
1262
1263 if (!IS_DNODE(page) || !is_cold_node(page))
1264 continue;
1265 if (ino_of_node(page) != ino)
1266 continue;
1267
1268 lock_page(page);
1269
1270 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1271continue_unlock:
1272 unlock_page(page);
1273 continue;
1274 }
1275 if (ino_of_node(page) != ino)
1276 goto continue_unlock;
1277
1278 if (!PageDirty(page)) {
1279 /* someone wrote it for us */
1280 goto continue_unlock;
1281 }
1282
1283 if (last_page)
1284 f2fs_put_page(last_page, 0);
1285
1286 get_page(page);
1287 last_page = page;
1288 unlock_page(page);
1289 }
1290 pagevec_release(&pvec);
1291 cond_resched();
1292 }
1293 return last_page;
1294}
1295
1296int fsync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1297 struct writeback_control *wbc, bool atomic)
1298{
1299 pgoff_t index, end;
1300 struct pagevec pvec;
1301 int ret = 0;
1302 struct page *last_page = NULL;
1303 bool marked = false;
1304
1305 if (atomic) {
1306 last_page = last_fsync_dnode(sbi, ino);
1307 if (IS_ERR_OR_NULL(last_page))
1308 return PTR_ERR_OR_ZERO(last_page);
1309 }
1310retry:
1311 pagevec_init(&pvec, 0);
1312 index = 0;
1313 end = ULONG_MAX;
1314
1315 while (index <= end) {
1316 int i, nr_pages;
1317 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1318 PAGECACHE_TAG_DIRTY,
1319 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1320 if (nr_pages == 0)
1321 break;
1322
1323 for (i = 0; i < nr_pages; i++) {
1324 struct page *page = pvec.pages[i];
1325
1326 if (unlikely(f2fs_cp_error(sbi))) {
1327 f2fs_put_page(last_page, 0);
1328 pagevec_release(&pvec);
1329 return -EIO;
1330 }
1331
1332 if (!IS_DNODE(page) || !is_cold_node(page))
1333 continue;
1334 if (ino_of_node(page) != ino)
1335 continue;
1336
1337 lock_page(page);
1338
1339 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1340continue_unlock:
1341 unlock_page(page);
1342 continue;
1343 }
1344 if (ino_of_node(page) != ino)
1345 goto continue_unlock;
1346
1347 if (!PageDirty(page) && page != last_page) {
1348 /* someone wrote it for us */
1349 goto continue_unlock;
1350 }
1351
1352 f2fs_wait_on_page_writeback(page, NODE, true);
1353 BUG_ON(PageWriteback(page));
1354
1355 if (!atomic || page == last_page) {
1356 set_fsync_mark(page, 1);
1357 if (IS_INODE(page))
1358 set_dentry_mark(page,
1359 need_dentry_mark(sbi, ino));
1360 /* may be written by other thread */
1361 if (!PageDirty(page))
1362 set_page_dirty(page);
1363 }
1364
1365 if (!clear_page_dirty_for_io(page))
1366 goto continue_unlock;
1367
1368 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1369 if (ret) {
1370 unlock_page(page);
1371 f2fs_put_page(last_page, 0);
1372 break;
1373 }
1374 if (page == last_page) {
1375 f2fs_put_page(page, 0);
1376 marked = true;
1377 break;
1378 }
1379 }
1380 pagevec_release(&pvec);
1381 cond_resched();
1382
1383 if (ret || marked)
1384 break;
1385 }
1386 if (!ret && atomic && !marked) {
1387 f2fs_msg(sbi->sb, KERN_DEBUG,
1388 "Retry to write fsync mark: ino=%u, idx=%lx",
1389 ino, last_page->index);
1390 lock_page(last_page);
1391 set_page_dirty(last_page);
1392 unlock_page(last_page);
1393 goto retry;
1394 }
1395 return ret ? -EIO: 0;
1396}
1397
1398int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1399{
1400 pgoff_t index, end;
1401 struct pagevec pvec;
1402 int step = 0;
1403 int nwritten = 0;
1404
1405 pagevec_init(&pvec, 0);
1406
1407next_step:
1408 index = 0;
1409 end = ULONG_MAX;
1410
1411 while (index <= end) {
1412 int i, nr_pages;
1413 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1414 PAGECACHE_TAG_DIRTY,
1415 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1416 if (nr_pages == 0)
1417 break;
1418
1419 for (i = 0; i < nr_pages; i++) {
1420 struct page *page = pvec.pages[i];
1421
1422 if (unlikely(f2fs_cp_error(sbi))) {
1423 pagevec_release(&pvec);
1424 return -EIO;
1425 }
1426
1427 /*
1428 * flushing sequence with step:
1429 * 0. indirect nodes
1430 * 1. dentry dnodes
1431 * 2. file dnodes
1432 */
1433 if (step == 0 && IS_DNODE(page))
1434 continue;
1435 if (step == 1 && (!IS_DNODE(page) ||
1436 is_cold_node(page)))
1437 continue;
1438 if (step == 2 && (!IS_DNODE(page) ||
1439 !is_cold_node(page)))
1440 continue;
1441lock_node:
1442 if (!trylock_page(page))
1443 continue;
1444
1445 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1446continue_unlock:
1447 unlock_page(page);
1448 continue;
1449 }
1450
1451 if (!PageDirty(page)) {
1452 /* someone wrote it for us */
1453 goto continue_unlock;
1454 }
1455
1456 /* flush inline_data */
1457 if (is_inline_node(page)) {
1458 clear_inline_node(page);
1459 unlock_page(page);
1460 flush_inline_data(sbi, ino_of_node(page));
1461 goto lock_node;
1462 }
1463
1464 f2fs_wait_on_page_writeback(page, NODE, true);
1465
1466 BUG_ON(PageWriteback(page));
1467 if (!clear_page_dirty_for_io(page))
1468 goto continue_unlock;
1469
1470 set_fsync_mark(page, 0);
1471 set_dentry_mark(page, 0);
1472
1473 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1474 unlock_page(page);
1475
1476 if (--wbc->nr_to_write == 0)
1477 break;
1478 }
1479 pagevec_release(&pvec);
1480 cond_resched();
1481
1482 if (wbc->nr_to_write == 0) {
1483 step = 2;
1484 break;
1485 }
1486 }
1487
1488 if (step < 2) {
1489 step++;
1490 goto next_step;
1491 }
1492 return nwritten;
1493}
1494
1495int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1496{
1497 pgoff_t index = 0, end = ULONG_MAX;
1498 struct pagevec pvec;
1499 int ret2 = 0, ret = 0;
1500
1501 pagevec_init(&pvec, 0);
1502
1503 while (index <= end) {
1504 int i, nr_pages;
1505 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1506 PAGECACHE_TAG_WRITEBACK,
1507 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1508 if (nr_pages == 0)
1509 break;
1510
1511 for (i = 0; i < nr_pages; i++) {
1512 struct page *page = pvec.pages[i];
1513
1514 /* until radix tree lookup accepts end_index */
1515 if (unlikely(page->index > end))
1516 continue;
1517
1518 if (ino && ino_of_node(page) == ino) {
1519 f2fs_wait_on_page_writeback(page, NODE, true);
1520 if (TestClearPageError(page))
1521 ret = -EIO;
1522 }
1523 }
1524 pagevec_release(&pvec);
1525 cond_resched();
1526 }
1527
1528 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1529 ret2 = -ENOSPC;
1530 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1531 ret2 = -EIO;
1532 if (!ret)
1533 ret = ret2;
1534 return ret;
1535}
1536
1537static int f2fs_write_node_page(struct page *page,
1538 struct writeback_control *wbc)
1539{
1540 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1541 nid_t nid;
1542 struct node_info ni;
1543 struct f2fs_io_info fio = {
1544 .sbi = sbi,
1545 .type = NODE,
1546 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1547 .page = page,
1548 .encrypted_page = NULL,
1549 };
1550
1551 trace_f2fs_writepage(page, NODE);
1552
1553 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1554 goto redirty_out;
1555 if (unlikely(f2fs_cp_error(sbi)))
1556 goto redirty_out;
1557
1558 /* get old block addr of this node page */
1559 nid = nid_of_node(page);
1560 f2fs_bug_on(sbi, page->index != nid);
1561
1562 if (wbc->for_reclaim) {
1563 if (!down_read_trylock(&sbi->node_write))
1564 goto redirty_out;
1565 } else {
1566 down_read(&sbi->node_write);
1567 }
1568
1569 get_node_info(sbi, nid, &ni);
1570
1571 /* This page is already truncated */
1572 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1573 ClearPageUptodate(page);
1574 dec_page_count(sbi, F2FS_DIRTY_NODES);
1575 up_read(&sbi->node_write);
1576 unlock_page(page);
1577 return 0;
1578 }
1579
1580 set_page_writeback(page);
1581 fio.old_blkaddr = ni.blk_addr;
1582 write_node_page(nid, &fio);
1583 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1584 dec_page_count(sbi, F2FS_DIRTY_NODES);
1585 up_read(&sbi->node_write);
1586
1587 if (wbc->for_reclaim)
1588 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1589
1590 unlock_page(page);
1591
1592 if (unlikely(f2fs_cp_error(sbi)))
1593 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1594
1595 return 0;
1596
1597redirty_out:
1598 redirty_page_for_writepage(wbc, page);
1599 return AOP_WRITEPAGE_ACTIVATE;
1600}
1601
1602static int f2fs_write_node_pages(struct address_space *mapping,
1603 struct writeback_control *wbc)
1604{
1605 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1606 long diff;
1607
1608 /* balancing f2fs's metadata in background */
1609 f2fs_balance_fs_bg(sbi);
1610
1611 /* collect a number of dirty node pages and write together */
1612 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1613 goto skip_write;
1614
1615 trace_f2fs_writepages(mapping->host, wbc, NODE);
1616
1617 diff = nr_pages_to_write(sbi, NODE, wbc);
1618 wbc->sync_mode = WB_SYNC_NONE;
1619 sync_node_pages(sbi, wbc);
1620 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1621 return 0;
1622
1623skip_write:
1624 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1625 trace_f2fs_writepages(mapping->host, wbc, NODE);
1626 return 0;
1627}
1628
1629static int f2fs_set_node_page_dirty(struct page *page)
1630{
1631 trace_f2fs_set_page_dirty(page, NODE);
1632
1633 SetPageUptodate(page);
1634 if (!PageDirty(page)) {
1635 __set_page_dirty_nobuffers(page);
1636 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1637 SetPagePrivate(page);
1638 f2fs_trace_pid(page);
1639 return 1;
1640 }
1641 return 0;
1642}
1643
1644/*
1645 * Structure of the f2fs node operations
1646 */
1647const struct address_space_operations f2fs_node_aops = {
1648 .writepage = f2fs_write_node_page,
1649 .writepages = f2fs_write_node_pages,
1650 .set_page_dirty = f2fs_set_node_page_dirty,
1651 .invalidatepage = f2fs_invalidate_page,
1652 .releasepage = f2fs_release_page,
1653};
1654
1655static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1656 nid_t n)
1657{
1658 return radix_tree_lookup(&nm_i->free_nid_root, n);
1659}
1660
1661static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1662 struct free_nid *i)
1663{
1664 list_del(&i->list);
1665 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1666}
1667
1668static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1669{
1670 struct f2fs_nm_info *nm_i = NM_I(sbi);
1671 struct free_nid *i;
1672 struct nat_entry *ne;
1673
1674 if (!available_free_memory(sbi, FREE_NIDS))
1675 return -1;
1676
1677 /* 0 nid should not be used */
1678 if (unlikely(nid == 0))
1679 return 0;
1680
1681 if (build) {
1682 /* do not add allocated nids */
1683 ne = __lookup_nat_cache(nm_i, nid);
1684 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1685 nat_get_blkaddr(ne) != NULL_ADDR))
1686 return 0;
1687 }
1688
1689 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1690 i->nid = nid;
1691 i->state = NID_NEW;
1692
1693 if (radix_tree_preload(GFP_NOFS)) {
1694 kmem_cache_free(free_nid_slab, i);
1695 return 0;
1696 }
1697
1698 spin_lock(&nm_i->free_nid_list_lock);
1699 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1700 spin_unlock(&nm_i->free_nid_list_lock);
1701 radix_tree_preload_end();
1702 kmem_cache_free(free_nid_slab, i);
1703 return 0;
1704 }
1705 list_add_tail(&i->list, &nm_i->free_nid_list);
1706 nm_i->fcnt++;
1707 spin_unlock(&nm_i->free_nid_list_lock);
1708 radix_tree_preload_end();
1709 return 1;
1710}
1711
1712static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1713{
1714 struct free_nid *i;
1715 bool need_free = false;
1716
1717 spin_lock(&nm_i->free_nid_list_lock);
1718 i = __lookup_free_nid_list(nm_i, nid);
1719 if (i && i->state == NID_NEW) {
1720 __del_from_free_nid_list(nm_i, i);
1721 nm_i->fcnt--;
1722 need_free = true;
1723 }
1724 spin_unlock(&nm_i->free_nid_list_lock);
1725
1726 if (need_free)
1727 kmem_cache_free(free_nid_slab, i);
1728}
1729
1730static void scan_nat_page(struct f2fs_sb_info *sbi,
1731 struct page *nat_page, nid_t start_nid)
1732{
1733 struct f2fs_nm_info *nm_i = NM_I(sbi);
1734 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1735 block_t blk_addr;
1736 int i;
1737
1738 i = start_nid % NAT_ENTRY_PER_BLOCK;
1739
1740 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1741
1742 if (unlikely(start_nid >= nm_i->max_nid))
1743 break;
1744
1745 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1746 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1747 if (blk_addr == NULL_ADDR) {
1748 if (add_free_nid(sbi, start_nid, true) < 0)
1749 break;
1750 }
1751 }
1752}
1753
1754static void build_free_nids(struct f2fs_sb_info *sbi)
1755{
1756 struct f2fs_nm_info *nm_i = NM_I(sbi);
1757 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1758 struct f2fs_journal *journal = curseg->journal;
1759 int i = 0;
1760 nid_t nid = nm_i->next_scan_nid;
1761
1762 /* Enough entries */
1763 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1764 return;
1765
1766 /* readahead nat pages to be scanned */
1767 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1768 META_NAT, true);
1769
1770 down_read(&nm_i->nat_tree_lock);
1771
1772 while (1) {
1773 struct page *page = get_current_nat_page(sbi, nid);
1774
1775 scan_nat_page(sbi, page, nid);
1776 f2fs_put_page(page, 1);
1777
1778 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1779 if (unlikely(nid >= nm_i->max_nid))
1780 nid = 0;
1781
1782 if (++i >= FREE_NID_PAGES)
1783 break;
1784 }
1785
1786 /* go to the next free nat pages to find free nids abundantly */
1787 nm_i->next_scan_nid = nid;
1788
1789 /* find free nids from current sum_pages */
1790 down_read(&curseg->journal_rwsem);
1791 for (i = 0; i < nats_in_cursum(journal); i++) {
1792 block_t addr;
1793
1794 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1795 nid = le32_to_cpu(nid_in_journal(journal, i));
1796 if (addr == NULL_ADDR)
1797 add_free_nid(sbi, nid, true);
1798 else
1799 remove_free_nid(nm_i, nid);
1800 }
1801 up_read(&curseg->journal_rwsem);
1802 up_read(&nm_i->nat_tree_lock);
1803
1804 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1805 nm_i->ra_nid_pages, META_NAT, false);
1806}
1807
1808/*
1809 * If this function returns success, caller can obtain a new nid
1810 * from second parameter of this function.
1811 * The returned nid could be used ino as well as nid when inode is created.
1812 */
1813bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1814{
1815 struct f2fs_nm_info *nm_i = NM_I(sbi);
1816 struct free_nid *i = NULL;
1817retry:
1818#ifdef CONFIG_F2FS_FAULT_INJECTION
1819 if (time_to_inject(FAULT_ALLOC_NID))
1820 return false;
1821#endif
1822 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1823 return false;
1824
1825 spin_lock(&nm_i->free_nid_list_lock);
1826
1827 /* We should not use stale free nids created by build_free_nids */
1828 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1829 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1830 list_for_each_entry(i, &nm_i->free_nid_list, list)
1831 if (i->state == NID_NEW)
1832 break;
1833
1834 f2fs_bug_on(sbi, i->state != NID_NEW);
1835 *nid = i->nid;
1836 i->state = NID_ALLOC;
1837 nm_i->fcnt--;
1838 spin_unlock(&nm_i->free_nid_list_lock);
1839 return true;
1840 }
1841 spin_unlock(&nm_i->free_nid_list_lock);
1842
1843 /* Let's scan nat pages and its caches to get free nids */
1844 mutex_lock(&nm_i->build_lock);
1845 build_free_nids(sbi);
1846 mutex_unlock(&nm_i->build_lock);
1847 goto retry;
1848}
1849
1850/*
1851 * alloc_nid() should be called prior to this function.
1852 */
1853void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1854{
1855 struct f2fs_nm_info *nm_i = NM_I(sbi);
1856 struct free_nid *i;
1857
1858 spin_lock(&nm_i->free_nid_list_lock);
1859 i = __lookup_free_nid_list(nm_i, nid);
1860 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1861 __del_from_free_nid_list(nm_i, i);
1862 spin_unlock(&nm_i->free_nid_list_lock);
1863
1864 kmem_cache_free(free_nid_slab, i);
1865}
1866
1867/*
1868 * alloc_nid() should be called prior to this function.
1869 */
1870void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1871{
1872 struct f2fs_nm_info *nm_i = NM_I(sbi);
1873 struct free_nid *i;
1874 bool need_free = false;
1875
1876 if (!nid)
1877 return;
1878
1879 spin_lock(&nm_i->free_nid_list_lock);
1880 i = __lookup_free_nid_list(nm_i, nid);
1881 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1882 if (!available_free_memory(sbi, FREE_NIDS)) {
1883 __del_from_free_nid_list(nm_i, i);
1884 need_free = true;
1885 } else {
1886 i->state = NID_NEW;
1887 nm_i->fcnt++;
1888 }
1889 spin_unlock(&nm_i->free_nid_list_lock);
1890
1891 if (need_free)
1892 kmem_cache_free(free_nid_slab, i);
1893}
1894
1895int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1896{
1897 struct f2fs_nm_info *nm_i = NM_I(sbi);
1898 struct free_nid *i, *next;
1899 int nr = nr_shrink;
1900
1901 if (!mutex_trylock(&nm_i->build_lock))
1902 return 0;
1903
1904 spin_lock(&nm_i->free_nid_list_lock);
1905 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1906 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1907 break;
1908 if (i->state == NID_ALLOC)
1909 continue;
1910 __del_from_free_nid_list(nm_i, i);
1911 kmem_cache_free(free_nid_slab, i);
1912 nm_i->fcnt--;
1913 nr_shrink--;
1914 }
1915 spin_unlock(&nm_i->free_nid_list_lock);
1916 mutex_unlock(&nm_i->build_lock);
1917
1918 return nr - nr_shrink;
1919}
1920
1921void recover_inline_xattr(struct inode *inode, struct page *page)
1922{
1923 void *src_addr, *dst_addr;
1924 size_t inline_size;
1925 struct page *ipage;
1926 struct f2fs_inode *ri;
1927
1928 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1929 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1930
1931 ri = F2FS_INODE(page);
1932 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1933 clear_inode_flag(inode, FI_INLINE_XATTR);
1934 goto update_inode;
1935 }
1936
1937 dst_addr = inline_xattr_addr(ipage);
1938 src_addr = inline_xattr_addr(page);
1939 inline_size = inline_xattr_size(inode);
1940
1941 f2fs_wait_on_page_writeback(ipage, NODE, true);
1942 memcpy(dst_addr, src_addr, inline_size);
1943update_inode:
1944 update_inode(inode, ipage);
1945 f2fs_put_page(ipage, 1);
1946}
1947
1948void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1949{
1950 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1951 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1952 nid_t new_xnid = nid_of_node(page);
1953 struct node_info ni;
1954
1955 /* 1: invalidate the previous xattr nid */
1956 if (!prev_xnid)
1957 goto recover_xnid;
1958
1959 /* Deallocate node address */
1960 get_node_info(sbi, prev_xnid, &ni);
1961 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1962 invalidate_blocks(sbi, ni.blk_addr);
1963 dec_valid_node_count(sbi, inode);
1964 set_node_addr(sbi, &ni, NULL_ADDR, false);
1965
1966recover_xnid:
1967 /* 2: allocate new xattr nid */
1968 if (unlikely(!inc_valid_node_count(sbi, inode)))
1969 f2fs_bug_on(sbi, 1);
1970
1971 remove_free_nid(NM_I(sbi), new_xnid);
1972 get_node_info(sbi, new_xnid, &ni);
1973 ni.ino = inode->i_ino;
1974 set_node_addr(sbi, &ni, NEW_ADDR, false);
1975 f2fs_i_xnid_write(inode, new_xnid);
1976
1977 /* 3: update xattr blkaddr */
1978 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1979 set_node_addr(sbi, &ni, blkaddr, false);
1980}
1981
1982int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1983{
1984 struct f2fs_inode *src, *dst;
1985 nid_t ino = ino_of_node(page);
1986 struct node_info old_ni, new_ni;
1987 struct page *ipage;
1988
1989 get_node_info(sbi, ino, &old_ni);
1990
1991 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1992 return -EINVAL;
1993
1994 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
1995 if (!ipage)
1996 return -ENOMEM;
1997
1998 /* Should not use this inode from free nid list */
1999 remove_free_nid(NM_I(sbi), ino);
2000
2001 SetPageUptodate(ipage);
2002 fill_node_footer(ipage, ino, ino, 0, true);
2003
2004 src = F2FS_INODE(page);
2005 dst = F2FS_INODE(ipage);
2006
2007 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2008 dst->i_size = 0;
2009 dst->i_blocks = cpu_to_le64(1);
2010 dst->i_links = cpu_to_le32(1);
2011 dst->i_xattr_nid = 0;
2012 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2013
2014 new_ni = old_ni;
2015 new_ni.ino = ino;
2016
2017 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2018 WARN_ON(1);
2019 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2020 inc_valid_inode_count(sbi);
2021 set_page_dirty(ipage);
2022 f2fs_put_page(ipage, 1);
2023 return 0;
2024}
2025
2026int restore_node_summary(struct f2fs_sb_info *sbi,
2027 unsigned int segno, struct f2fs_summary_block *sum)
2028{
2029 struct f2fs_node *rn;
2030 struct f2fs_summary *sum_entry;
2031 block_t addr;
2032 int bio_blocks = MAX_BIO_BLOCKS(sbi);
2033 int i, idx, last_offset, nrpages;
2034
2035 /* scan the node segment */
2036 last_offset = sbi->blocks_per_seg;
2037 addr = START_BLOCK(sbi, segno);
2038 sum_entry = &sum->entries[0];
2039
2040 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2041 nrpages = min(last_offset - i, bio_blocks);
2042
2043 /* readahead node pages */
2044 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2045
2046 for (idx = addr; idx < addr + nrpages; idx++) {
2047 struct page *page = get_tmp_page(sbi, idx);
2048
2049 rn = F2FS_NODE(page);
2050 sum_entry->nid = rn->footer.nid;
2051 sum_entry->version = 0;
2052 sum_entry->ofs_in_node = 0;
2053 sum_entry++;
2054 f2fs_put_page(page, 1);
2055 }
2056
2057 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2058 addr + nrpages);
2059 }
2060 return 0;
2061}
2062
2063static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2064{
2065 struct f2fs_nm_info *nm_i = NM_I(sbi);
2066 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2067 struct f2fs_journal *journal = curseg->journal;
2068 int i;
2069
2070 down_write(&curseg->journal_rwsem);
2071 for (i = 0; i < nats_in_cursum(journal); i++) {
2072 struct nat_entry *ne;
2073 struct f2fs_nat_entry raw_ne;
2074 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2075
2076 raw_ne = nat_in_journal(journal, i);
2077
2078 ne = __lookup_nat_cache(nm_i, nid);
2079 if (!ne) {
2080 ne = grab_nat_entry(nm_i, nid);
2081 node_info_from_raw_nat(&ne->ni, &raw_ne);
2082 }
2083 __set_nat_cache_dirty(nm_i, ne);
2084 }
2085 update_nats_in_cursum(journal, -i);
2086 up_write(&curseg->journal_rwsem);
2087}
2088
2089static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2090 struct list_head *head, int max)
2091{
2092 struct nat_entry_set *cur;
2093
2094 if (nes->entry_cnt >= max)
2095 goto add_out;
2096
2097 list_for_each_entry(cur, head, set_list) {
2098 if (cur->entry_cnt >= nes->entry_cnt) {
2099 list_add(&nes->set_list, cur->set_list.prev);
2100 return;
2101 }
2102 }
2103add_out:
2104 list_add_tail(&nes->set_list, head);
2105}
2106
2107static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2108 struct nat_entry_set *set)
2109{
2110 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2111 struct f2fs_journal *journal = curseg->journal;
2112 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2113 bool to_journal = true;
2114 struct f2fs_nat_block *nat_blk;
2115 struct nat_entry *ne, *cur;
2116 struct page *page = NULL;
2117
2118 /*
2119 * there are two steps to flush nat entries:
2120 * #1, flush nat entries to journal in current hot data summary block.
2121 * #2, flush nat entries to nat page.
2122 */
2123 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2124 to_journal = false;
2125
2126 if (to_journal) {
2127 down_write(&curseg->journal_rwsem);
2128 } else {
2129 page = get_next_nat_page(sbi, start_nid);
2130 nat_blk = page_address(page);
2131 f2fs_bug_on(sbi, !nat_blk);
2132 }
2133
2134 /* flush dirty nats in nat entry set */
2135 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2136 struct f2fs_nat_entry *raw_ne;
2137 nid_t nid = nat_get_nid(ne);
2138 int offset;
2139
2140 if (nat_get_blkaddr(ne) == NEW_ADDR)
2141 continue;
2142
2143 if (to_journal) {
2144 offset = lookup_journal_in_cursum(journal,
2145 NAT_JOURNAL, nid, 1);
2146 f2fs_bug_on(sbi, offset < 0);
2147 raw_ne = &nat_in_journal(journal, offset);
2148 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2149 } else {
2150 raw_ne = &nat_blk->entries[nid - start_nid];
2151 }
2152 raw_nat_from_node_info(raw_ne, &ne->ni);
2153 nat_reset_flag(ne);
2154 __clear_nat_cache_dirty(NM_I(sbi), ne);
2155 if (nat_get_blkaddr(ne) == NULL_ADDR)
2156 add_free_nid(sbi, nid, false);
2157 }
2158
2159 if (to_journal)
2160 up_write(&curseg->journal_rwsem);
2161 else
2162 f2fs_put_page(page, 1);
2163
2164 f2fs_bug_on(sbi, set->entry_cnt);
2165
2166 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2167 kmem_cache_free(nat_entry_set_slab, set);
2168}
2169
2170/*
2171 * This function is called during the checkpointing process.
2172 */
2173void flush_nat_entries(struct f2fs_sb_info *sbi)
2174{
2175 struct f2fs_nm_info *nm_i = NM_I(sbi);
2176 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2177 struct f2fs_journal *journal = curseg->journal;
2178 struct nat_entry_set *setvec[SETVEC_SIZE];
2179 struct nat_entry_set *set, *tmp;
2180 unsigned int found;
2181 nid_t set_idx = 0;
2182 LIST_HEAD(sets);
2183
2184 if (!nm_i->dirty_nat_cnt)
2185 return;
2186
2187 down_write(&nm_i->nat_tree_lock);
2188
2189 /*
2190 * if there are no enough space in journal to store dirty nat
2191 * entries, remove all entries from journal and merge them
2192 * into nat entry set.
2193 */
2194 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2195 remove_nats_in_journal(sbi);
2196
2197 while ((found = __gang_lookup_nat_set(nm_i,
2198 set_idx, SETVEC_SIZE, setvec))) {
2199 unsigned idx;
2200 set_idx = setvec[found - 1]->set + 1;
2201 for (idx = 0; idx < found; idx++)
2202 __adjust_nat_entry_set(setvec[idx], &sets,
2203 MAX_NAT_JENTRIES(journal));
2204 }
2205
2206 /* flush dirty nats in nat entry set */
2207 list_for_each_entry_safe(set, tmp, &sets, set_list)
2208 __flush_nat_entry_set(sbi, set);
2209
2210 up_write(&nm_i->nat_tree_lock);
2211
2212 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2213}
2214
2215static int init_node_manager(struct f2fs_sb_info *sbi)
2216{
2217 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2218 struct f2fs_nm_info *nm_i = NM_I(sbi);
2219 unsigned char *version_bitmap;
2220 unsigned int nat_segs, nat_blocks;
2221
2222 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2223
2224 /* segment_count_nat includes pair segment so divide to 2. */
2225 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2226 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2227
2228 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2229
2230 /* not used nids: 0, node, meta, (and root counted as valid node) */
2231 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2232 nm_i->fcnt = 0;
2233 nm_i->nat_cnt = 0;
2234 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2235 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2236 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2237
2238 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2239 INIT_LIST_HEAD(&nm_i->free_nid_list);
2240 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2241 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2242 INIT_LIST_HEAD(&nm_i->nat_entries);
2243
2244 mutex_init(&nm_i->build_lock);
2245 spin_lock_init(&nm_i->free_nid_list_lock);
2246 init_rwsem(&nm_i->nat_tree_lock);
2247
2248 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2249 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2250 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2251 if (!version_bitmap)
2252 return -EFAULT;
2253
2254 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2255 GFP_KERNEL);
2256 if (!nm_i->nat_bitmap)
2257 return -ENOMEM;
2258 return 0;
2259}
2260
2261int build_node_manager(struct f2fs_sb_info *sbi)
2262{
2263 int err;
2264
2265 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2266 if (!sbi->nm_info)
2267 return -ENOMEM;
2268
2269 err = init_node_manager(sbi);
2270 if (err)
2271 return err;
2272
2273 build_free_nids(sbi);
2274 return 0;
2275}
2276
2277void destroy_node_manager(struct f2fs_sb_info *sbi)
2278{
2279 struct f2fs_nm_info *nm_i = NM_I(sbi);
2280 struct free_nid *i, *next_i;
2281 struct nat_entry *natvec[NATVEC_SIZE];
2282 struct nat_entry_set *setvec[SETVEC_SIZE];
2283 nid_t nid = 0;
2284 unsigned int found;
2285
2286 if (!nm_i)
2287 return;
2288
2289 /* destroy free nid list */
2290 spin_lock(&nm_i->free_nid_list_lock);
2291 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2292 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2293 __del_from_free_nid_list(nm_i, i);
2294 nm_i->fcnt--;
2295 spin_unlock(&nm_i->free_nid_list_lock);
2296 kmem_cache_free(free_nid_slab, i);
2297 spin_lock(&nm_i->free_nid_list_lock);
2298 }
2299 f2fs_bug_on(sbi, nm_i->fcnt);
2300 spin_unlock(&nm_i->free_nid_list_lock);
2301
2302 /* destroy nat cache */
2303 down_write(&nm_i->nat_tree_lock);
2304 while ((found = __gang_lookup_nat_cache(nm_i,
2305 nid, NATVEC_SIZE, natvec))) {
2306 unsigned idx;
2307
2308 nid = nat_get_nid(natvec[found - 1]) + 1;
2309 for (idx = 0; idx < found; idx++)
2310 __del_from_nat_cache(nm_i, natvec[idx]);
2311 }
2312 f2fs_bug_on(sbi, nm_i->nat_cnt);
2313
2314 /* destroy nat set cache */
2315 nid = 0;
2316 while ((found = __gang_lookup_nat_set(nm_i,
2317 nid, SETVEC_SIZE, setvec))) {
2318 unsigned idx;
2319
2320 nid = setvec[found - 1]->set + 1;
2321 for (idx = 0; idx < found; idx++) {
2322 /* entry_cnt is not zero, when cp_error was occurred */
2323 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2324 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2325 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2326 }
2327 }
2328 up_write(&nm_i->nat_tree_lock);
2329
2330 kfree(nm_i->nat_bitmap);
2331 sbi->nm_info = NULL;
2332 kfree(nm_i);
2333}
2334
2335int __init create_node_manager_caches(void)
2336{
2337 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2338 sizeof(struct nat_entry));
2339 if (!nat_entry_slab)
2340 goto fail;
2341
2342 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2343 sizeof(struct free_nid));
2344 if (!free_nid_slab)
2345 goto destroy_nat_entry;
2346
2347 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2348 sizeof(struct nat_entry_set));
2349 if (!nat_entry_set_slab)
2350 goto destroy_free_nid;
2351 return 0;
2352
2353destroy_free_nid:
2354 kmem_cache_destroy(free_nid_slab);
2355destroy_nat_entry:
2356 kmem_cache_destroy(nat_entry_slab);
2357fail:
2358 return -ENOMEM;
2359}
2360
2361void destroy_node_manager_caches(void)
2362{
2363 kmem_cache_destroy(nat_entry_set_slab);
2364 kmem_cache_destroy(free_nid_slab);
2365 kmem_cache_destroy(nat_entry_slab);
2366}
Linux 6.x block layer and io_uring tree(s)