projects / linux-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
f2fs: make set_de_type() static
[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 "xattr.h"
23#include "trace.h"
24#include <trace/events/f2fs.h>
25
26#define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
27
28static struct kmem_cache *nat_entry_slab;
29static struct kmem_cache *free_nid_slab;
30static struct kmem_cache *nat_entry_set_slab;
31
32/*
33 * Check whether the given nid is within node id range.
34 */
35int check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
36{
37 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
38 set_sbi_flag(sbi, SBI_NEED_FSCK);
39 f2fs_msg(sbi->sb, KERN_WARNING,
40 "%s: out-of-range nid=%x, run fsck to fix.",
41 __func__, nid);
42 return -EINVAL;
43 }
44 return 0;
45}
46
47bool available_free_memory(struct f2fs_sb_info *sbi, int type)
48{
49 struct f2fs_nm_info *nm_i = NM_I(sbi);
50 struct sysinfo val;
51 unsigned long avail_ram;
52 unsigned long mem_size = 0;
53 bool res = false;
54
55 si_meminfo(&val);
56
57 /* only uses low memory */
58 avail_ram = val.totalram - val.totalhigh;
59
60 /*
61 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
62 */
63 if (type == FREE_NIDS) {
64 mem_size = (nm_i->nid_cnt[FREE_NID] *
65 sizeof(struct free_nid)) >> PAGE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
67 } else if (type == NAT_ENTRIES) {
68 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
69 PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
71 if (excess_cached_nats(sbi))
72 res = false;
73 } else if (type == DIRTY_DENTS) {
74 if (sbi->sb->s_bdi->wb.dirty_exceeded)
75 return false;
76 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
77 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
78 } else if (type == INO_ENTRIES) {
79 int i;
80
81 for (i = 0; i < MAX_INO_ENTRY; i++)
82 mem_size += sbi->im[i].ino_num *
83 sizeof(struct ino_entry);
84 mem_size >>= PAGE_SHIFT;
85 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
86 } else if (type == EXTENT_CACHE) {
87 mem_size = (atomic_read(&sbi->total_ext_tree) *
88 sizeof(struct extent_tree) +
89 atomic_read(&sbi->total_ext_node) *
90 sizeof(struct extent_node)) >> PAGE_SHIFT;
91 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
92 } else if (type == INMEM_PAGES) {
93 /* it allows 20% / total_ram for inmemory pages */
94 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
95 res = mem_size < (val.totalram / 5);
96 } else {
97 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
98 return true;
99 }
100 return res;
101}
102
103static void clear_node_page_dirty(struct page *page)
104{
105 if (PageDirty(page)) {
106 clear_radix_tree_dirty_tag(page);
107 clear_page_dirty_for_io(page);
108 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
109 }
110 ClearPageUptodate(page);
111}
112
113static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
114{
115 pgoff_t index = current_nat_addr(sbi, nid);
116 return get_meta_page(sbi, index);
117}
118
119static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
120{
121 struct page *src_page;
122 struct page *dst_page;
123 pgoff_t src_off;
124 pgoff_t dst_off;
125 void *src_addr;
126 void *dst_addr;
127 struct f2fs_nm_info *nm_i = NM_I(sbi);
128
129 src_off = current_nat_addr(sbi, nid);
130 dst_off = next_nat_addr(sbi, src_off);
131
132 /* get current nat block page with lock */
133 src_page = get_meta_page(sbi, src_off);
134 dst_page = grab_meta_page(sbi, dst_off);
135 f2fs_bug_on(sbi, PageDirty(src_page));
136
137 src_addr = page_address(src_page);
138 dst_addr = page_address(dst_page);
139 memcpy(dst_addr, src_addr, PAGE_SIZE);
140 set_page_dirty(dst_page);
141 f2fs_put_page(src_page, 1);
142
143 set_to_next_nat(nm_i, nid);
144
145 return dst_page;
146}
147
148static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
149{
150 struct nat_entry *new;
151
152 if (no_fail)
153 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
154 else
155 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
156 if (new) {
157 nat_set_nid(new, nid);
158 nat_reset_flag(new);
159 }
160 return new;
161}
162
163static void __free_nat_entry(struct nat_entry *e)
164{
165 kmem_cache_free(nat_entry_slab, e);
166}
167
168/* must be locked by nat_tree_lock */
169static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
170 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
171{
172 if (no_fail)
173 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
174 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
175 return NULL;
176
177 if (raw_ne)
178 node_info_from_raw_nat(&ne->ni, raw_ne);
179 list_add_tail(&ne->list, &nm_i->nat_entries);
180 nm_i->nat_cnt++;
181 return ne;
182}
183
184static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
185{
186 return radix_tree_lookup(&nm_i->nat_root, n);
187}
188
189static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
190 nid_t start, unsigned int nr, struct nat_entry **ep)
191{
192 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
193}
194
195static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
196{
197 list_del(&e->list);
198 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
199 nm_i->nat_cnt--;
200 __free_nat_entry(e);
201}
202
203static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
204 struct nat_entry *ne)
205{
206 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
207 struct nat_entry_set *head;
208
209 head = radix_tree_lookup(&nm_i->nat_set_root, set);
210 if (!head) {
211 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
212
213 INIT_LIST_HEAD(&head->entry_list);
214 INIT_LIST_HEAD(&head->set_list);
215 head->set = set;
216 head->entry_cnt = 0;
217 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
218 }
219 return head;
220}
221
222static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
223 struct nat_entry *ne)
224{
225 struct nat_entry_set *head;
226 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
227
228 if (!new_ne)
229 head = __grab_nat_entry_set(nm_i, ne);
230
231 /*
232 * update entry_cnt in below condition:
233 * 1. update NEW_ADDR to valid block address;
234 * 2. update old block address to new one;
235 */
236 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
237 !get_nat_flag(ne, IS_DIRTY)))
238 head->entry_cnt++;
239
240 set_nat_flag(ne, IS_PREALLOC, new_ne);
241
242 if (get_nat_flag(ne, IS_DIRTY))
243 goto refresh_list;
244
245 nm_i->dirty_nat_cnt++;
246 set_nat_flag(ne, IS_DIRTY, true);
247refresh_list:
248 if (new_ne)
249 list_del_init(&ne->list);
250 else
251 list_move_tail(&ne->list, &head->entry_list);
252}
253
254static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
255 struct nat_entry_set *set, struct nat_entry *ne)
256{
257 list_move_tail(&ne->list, &nm_i->nat_entries);
258 set_nat_flag(ne, IS_DIRTY, false);
259 set->entry_cnt--;
260 nm_i->dirty_nat_cnt--;
261}
262
263static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
264 nid_t start, unsigned int nr, struct nat_entry_set **ep)
265{
266 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
267 start, nr);
268}
269
270int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
271{
272 struct f2fs_nm_info *nm_i = NM_I(sbi);
273 struct nat_entry *e;
274 bool need = false;
275
276 down_read(&nm_i->nat_tree_lock);
277 e = __lookup_nat_cache(nm_i, nid);
278 if (e) {
279 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
280 !get_nat_flag(e, HAS_FSYNCED_INODE))
281 need = true;
282 }
283 up_read(&nm_i->nat_tree_lock);
284 return need;
285}
286
287bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
288{
289 struct f2fs_nm_info *nm_i = NM_I(sbi);
290 struct nat_entry *e;
291 bool is_cp = true;
292
293 down_read(&nm_i->nat_tree_lock);
294 e = __lookup_nat_cache(nm_i, nid);
295 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
296 is_cp = false;
297 up_read(&nm_i->nat_tree_lock);
298 return is_cp;
299}
300
301bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
302{
303 struct f2fs_nm_info *nm_i = NM_I(sbi);
304 struct nat_entry *e;
305 bool need_update = true;
306
307 down_read(&nm_i->nat_tree_lock);
308 e = __lookup_nat_cache(nm_i, ino);
309 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
310 (get_nat_flag(e, IS_CHECKPOINTED) ||
311 get_nat_flag(e, HAS_FSYNCED_INODE)))
312 need_update = false;
313 up_read(&nm_i->nat_tree_lock);
314 return need_update;
315}
316
317/* must be locked by nat_tree_lock */
318static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
319 struct f2fs_nat_entry *ne)
320{
321 struct f2fs_nm_info *nm_i = NM_I(sbi);
322 struct nat_entry *new, *e;
323
324 new = __alloc_nat_entry(nid, false);
325 if (!new)
326 return;
327
328 down_write(&nm_i->nat_tree_lock);
329 e = __lookup_nat_cache(nm_i, nid);
330 if (!e)
331 e = __init_nat_entry(nm_i, new, ne, false);
332 else
333 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
334 nat_get_blkaddr(e) !=
335 le32_to_cpu(ne->block_addr) ||
336 nat_get_version(e) != ne->version);
337 up_write(&nm_i->nat_tree_lock);
338 if (e != new)
339 __free_nat_entry(new);
340}
341
342static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
343 block_t new_blkaddr, bool fsync_done)
344{
345 struct f2fs_nm_info *nm_i = NM_I(sbi);
346 struct nat_entry *e;
347 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
348
349 down_write(&nm_i->nat_tree_lock);
350 e = __lookup_nat_cache(nm_i, ni->nid);
351 if (!e) {
352 e = __init_nat_entry(nm_i, new, NULL, true);
353 copy_node_info(&e->ni, ni);
354 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
355 } else if (new_blkaddr == NEW_ADDR) {
356 /*
357 * when nid is reallocated,
358 * previous nat entry can be remained in nat cache.
359 * So, reinitialize it with new information.
360 */
361 copy_node_info(&e->ni, ni);
362 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
363 }
364 /* let's free early to reduce memory consumption */
365 if (e != new)
366 __free_nat_entry(new);
367
368 /* sanity check */
369 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
370 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
371 new_blkaddr == NULL_ADDR);
372 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
373 new_blkaddr == NEW_ADDR);
374 f2fs_bug_on(sbi, is_valid_blkaddr(nat_get_blkaddr(e)) &&
375 new_blkaddr == NEW_ADDR);
376
377 /* increment version no as node is removed */
378 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
379 unsigned char version = nat_get_version(e);
380 nat_set_version(e, inc_node_version(version));
381 }
382
383 /* change address */
384 nat_set_blkaddr(e, new_blkaddr);
385 if (!is_valid_blkaddr(new_blkaddr))
386 set_nat_flag(e, IS_CHECKPOINTED, false);
387 __set_nat_cache_dirty(nm_i, e);
388
389 /* update fsync_mark if its inode nat entry is still alive */
390 if (ni->nid != ni->ino)
391 e = __lookup_nat_cache(nm_i, ni->ino);
392 if (e) {
393 if (fsync_done && ni->nid == ni->ino)
394 set_nat_flag(e, HAS_FSYNCED_INODE, true);
395 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
396 }
397 up_write(&nm_i->nat_tree_lock);
398}
399
400int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
401{
402 struct f2fs_nm_info *nm_i = NM_I(sbi);
403 int nr = nr_shrink;
404
405 if (!down_write_trylock(&nm_i->nat_tree_lock))
406 return 0;
407
408 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
409 struct nat_entry *ne;
410 ne = list_first_entry(&nm_i->nat_entries,
411 struct nat_entry, list);
412 __del_from_nat_cache(nm_i, ne);
413 nr_shrink--;
414 }
415 up_write(&nm_i->nat_tree_lock);
416 return nr - nr_shrink;
417}
418
419/*
420 * This function always returns success
421 */
422void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
423{
424 struct f2fs_nm_info *nm_i = NM_I(sbi);
425 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
426 struct f2fs_journal *journal = curseg->journal;
427 nid_t start_nid = START_NID(nid);
428 struct f2fs_nat_block *nat_blk;
429 struct page *page = NULL;
430 struct f2fs_nat_entry ne;
431 struct nat_entry *e;
432 pgoff_t index;
433 int i;
434
435 ni->nid = nid;
436
437 /* Check nat cache */
438 down_read(&nm_i->nat_tree_lock);
439 e = __lookup_nat_cache(nm_i, nid);
440 if (e) {
441 ni->ino = nat_get_ino(e);
442 ni->blk_addr = nat_get_blkaddr(e);
443 ni->version = nat_get_version(e);
444 up_read(&nm_i->nat_tree_lock);
445 return;
446 }
447
448 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
449
450 /* Check current segment summary */
451 down_read(&curseg->journal_rwsem);
452 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
453 if (i >= 0) {
454 ne = nat_in_journal(journal, i);
455 node_info_from_raw_nat(ni, &ne);
456 }
457 up_read(&curseg->journal_rwsem);
458 if (i >= 0) {
459 up_read(&nm_i->nat_tree_lock);
460 goto cache;
461 }
462
463 /* Fill node_info from nat page */
464 index = current_nat_addr(sbi, nid);
465 up_read(&nm_i->nat_tree_lock);
466
467 page = get_meta_page(sbi, index);
468 nat_blk = (struct f2fs_nat_block *)page_address(page);
469 ne = nat_blk->entries[nid - start_nid];
470 node_info_from_raw_nat(ni, &ne);
471 f2fs_put_page(page, 1);
472cache:
473 /* cache nat entry */
474 cache_nat_entry(sbi, nid, &ne);
475}
476
477/*
478 * readahead MAX_RA_NODE number of node pages.
479 */
480static void ra_node_pages(struct page *parent, int start, int n)
481{
482 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
483 struct blk_plug plug;
484 int i, end;
485 nid_t nid;
486
487 blk_start_plug(&plug);
488
489 /* Then, try readahead for siblings of the desired node */
490 end = start + n;
491 end = min(end, NIDS_PER_BLOCK);
492 for (i = start; i < end; i++) {
493 nid = get_nid(parent, i, false);
494 ra_node_page(sbi, nid);
495 }
496
497 blk_finish_plug(&plug);
498}
499
500pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
501{
502 const long direct_index = ADDRS_PER_INODE(dn->inode);
503 const long direct_blks = ADDRS_PER_BLOCK;
504 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
505 unsigned int skipped_unit = ADDRS_PER_BLOCK;
506 int cur_level = dn->cur_level;
507 int max_level = dn->max_level;
508 pgoff_t base = 0;
509
510 if (!dn->max_level)
511 return pgofs + 1;
512
513 while (max_level-- > cur_level)
514 skipped_unit *= NIDS_PER_BLOCK;
515
516 switch (dn->max_level) {
517 case 3:
518 base += 2 * indirect_blks;
519 case 2:
520 base += 2 * direct_blks;
521 case 1:
522 base += direct_index;
523 break;
524 default:
525 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
526 }
527
528 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
529}
530
531/*
532 * The maximum depth is four.
533 * Offset[0] will have raw inode offset.
534 */
535static int get_node_path(struct inode *inode, long block,
536 int offset[4], unsigned int noffset[4])
537{
538 const long direct_index = ADDRS_PER_INODE(inode);
539 const long direct_blks = ADDRS_PER_BLOCK;
540 const long dptrs_per_blk = NIDS_PER_BLOCK;
541 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
542 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
543 int n = 0;
544 int level = 0;
545
546 noffset[0] = 0;
547
548 if (block < direct_index) {
549 offset[n] = block;
550 goto got;
551 }
552 block -= direct_index;
553 if (block < direct_blks) {
554 offset[n++] = NODE_DIR1_BLOCK;
555 noffset[n] = 1;
556 offset[n] = block;
557 level = 1;
558 goto got;
559 }
560 block -= direct_blks;
561 if (block < direct_blks) {
562 offset[n++] = NODE_DIR2_BLOCK;
563 noffset[n] = 2;
564 offset[n] = block;
565 level = 1;
566 goto got;
567 }
568 block -= direct_blks;
569 if (block < indirect_blks) {
570 offset[n++] = NODE_IND1_BLOCK;
571 noffset[n] = 3;
572 offset[n++] = block / direct_blks;
573 noffset[n] = 4 + offset[n - 1];
574 offset[n] = block % direct_blks;
575 level = 2;
576 goto got;
577 }
578 block -= indirect_blks;
579 if (block < indirect_blks) {
580 offset[n++] = NODE_IND2_BLOCK;
581 noffset[n] = 4 + dptrs_per_blk;
582 offset[n++] = block / direct_blks;
583 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
584 offset[n] = block % direct_blks;
585 level = 2;
586 goto got;
587 }
588 block -= indirect_blks;
589 if (block < dindirect_blks) {
590 offset[n++] = NODE_DIND_BLOCK;
591 noffset[n] = 5 + (dptrs_per_blk * 2);
592 offset[n++] = block / indirect_blks;
593 noffset[n] = 6 + (dptrs_per_blk * 2) +
594 offset[n - 1] * (dptrs_per_blk + 1);
595 offset[n++] = (block / direct_blks) % dptrs_per_blk;
596 noffset[n] = 7 + (dptrs_per_blk * 2) +
597 offset[n - 2] * (dptrs_per_blk + 1) +
598 offset[n - 1];
599 offset[n] = block % direct_blks;
600 level = 3;
601 goto got;
602 } else {
603 return -E2BIG;
604 }
605got:
606 return level;
607}
608
609/*
610 * Caller should call f2fs_put_dnode(dn).
611 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
612 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
613 * In the case of RDONLY_NODE, we don't need to care about mutex.
614 */
615int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
616{
617 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
618 struct page *npage[4];
619 struct page *parent = NULL;
620 int offset[4];
621 unsigned int noffset[4];
622 nid_t nids[4];
623 int level, i = 0;
624 int err = 0;
625
626 level = get_node_path(dn->inode, index, offset, noffset);
627 if (level < 0)
628 return level;
629
630 nids[0] = dn->inode->i_ino;
631 npage[0] = dn->inode_page;
632
633 if (!npage[0]) {
634 npage[0] = get_node_page(sbi, nids[0]);
635 if (IS_ERR(npage[0]))
636 return PTR_ERR(npage[0]);
637 }
638
639 /* if inline_data is set, should not report any block indices */
640 if (f2fs_has_inline_data(dn->inode) && index) {
641 err = -ENOENT;
642 f2fs_put_page(npage[0], 1);
643 goto release_out;
644 }
645
646 parent = npage[0];
647 if (level != 0)
648 nids[1] = get_nid(parent, offset[0], true);
649 dn->inode_page = npage[0];
650 dn->inode_page_locked = true;
651
652 /* get indirect or direct nodes */
653 for (i = 1; i <= level; i++) {
654 bool done = false;
655
656 if (!nids[i] && mode == ALLOC_NODE) {
657 /* alloc new node */
658 if (!alloc_nid(sbi, &(nids[i]))) {
659 err = -ENOSPC;
660 goto release_pages;
661 }
662
663 dn->nid = nids[i];
664 npage[i] = new_node_page(dn, noffset[i]);
665 if (IS_ERR(npage[i])) {
666 alloc_nid_failed(sbi, nids[i]);
667 err = PTR_ERR(npage[i]);
668 goto release_pages;
669 }
670
671 set_nid(parent, offset[i - 1], nids[i], i == 1);
672 alloc_nid_done(sbi, nids[i]);
673 done = true;
674 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
675 npage[i] = get_node_page_ra(parent, offset[i - 1]);
676 if (IS_ERR(npage[i])) {
677 err = PTR_ERR(npage[i]);
678 goto release_pages;
679 }
680 done = true;
681 }
682 if (i == 1) {
683 dn->inode_page_locked = false;
684 unlock_page(parent);
685 } else {
686 f2fs_put_page(parent, 1);
687 }
688
689 if (!done) {
690 npage[i] = get_node_page(sbi, nids[i]);
691 if (IS_ERR(npage[i])) {
692 err = PTR_ERR(npage[i]);
693 f2fs_put_page(npage[0], 0);
694 goto release_out;
695 }
696 }
697 if (i < level) {
698 parent = npage[i];
699 nids[i + 1] = get_nid(parent, offset[i], false);
700 }
701 }
702 dn->nid = nids[level];
703 dn->ofs_in_node = offset[level];
704 dn->node_page = npage[level];
705 dn->data_blkaddr = datablock_addr(dn->inode,
706 dn->node_page, dn->ofs_in_node);
707 return 0;
708
709release_pages:
710 f2fs_put_page(parent, 1);
711 if (i > 1)
712 f2fs_put_page(npage[0], 0);
713release_out:
714 dn->inode_page = NULL;
715 dn->node_page = NULL;
716 if (err == -ENOENT) {
717 dn->cur_level = i;
718 dn->max_level = level;
719 dn->ofs_in_node = offset[level];
720 }
721 return err;
722}
723
724static void truncate_node(struct dnode_of_data *dn)
725{
726 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
727 struct node_info ni;
728
729 get_node_info(sbi, dn->nid, &ni);
730
731 /* Deallocate node address */
732 invalidate_blocks(sbi, ni.blk_addr);
733 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
734 set_node_addr(sbi, &ni, NULL_ADDR, false);
735
736 if (dn->nid == dn->inode->i_ino) {
737 remove_orphan_inode(sbi, dn->nid);
738 dec_valid_inode_count(sbi);
739 f2fs_inode_synced(dn->inode);
740 }
741
742 clear_node_page_dirty(dn->node_page);
743 set_sbi_flag(sbi, SBI_IS_DIRTY);
744
745 f2fs_put_page(dn->node_page, 1);
746
747 invalidate_mapping_pages(NODE_MAPPING(sbi),
748 dn->node_page->index, dn->node_page->index);
749
750 dn->node_page = NULL;
751 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
752}
753
754static int truncate_dnode(struct dnode_of_data *dn)
755{
756 struct page *page;
757
758 if (dn->nid == 0)
759 return 1;
760
761 /* get direct node */
762 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
763 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
764 return 1;
765 else if (IS_ERR(page))
766 return PTR_ERR(page);
767
768 /* Make dnode_of_data for parameter */
769 dn->node_page = page;
770 dn->ofs_in_node = 0;
771 truncate_data_blocks(dn);
772 truncate_node(dn);
773 return 1;
774}
775
776static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
777 int ofs, int depth)
778{
779 struct dnode_of_data rdn = *dn;
780 struct page *page;
781 struct f2fs_node *rn;
782 nid_t child_nid;
783 unsigned int child_nofs;
784 int freed = 0;
785 int i, ret;
786
787 if (dn->nid == 0)
788 return NIDS_PER_BLOCK + 1;
789
790 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
791
792 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
793 if (IS_ERR(page)) {
794 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
795 return PTR_ERR(page);
796 }
797
798 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
799
800 rn = F2FS_NODE(page);
801 if (depth < 3) {
802 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
803 child_nid = le32_to_cpu(rn->in.nid[i]);
804 if (child_nid == 0)
805 continue;
806 rdn.nid = child_nid;
807 ret = truncate_dnode(&rdn);
808 if (ret < 0)
809 goto out_err;
810 if (set_nid(page, i, 0, false))
811 dn->node_changed = true;
812 }
813 } else {
814 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
815 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
816 child_nid = le32_to_cpu(rn->in.nid[i]);
817 if (child_nid == 0) {
818 child_nofs += NIDS_PER_BLOCK + 1;
819 continue;
820 }
821 rdn.nid = child_nid;
822 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
823 if (ret == (NIDS_PER_BLOCK + 1)) {
824 if (set_nid(page, i, 0, false))
825 dn->node_changed = true;
826 child_nofs += ret;
827 } else if (ret < 0 && ret != -ENOENT) {
828 goto out_err;
829 }
830 }
831 freed = child_nofs;
832 }
833
834 if (!ofs) {
835 /* remove current indirect node */
836 dn->node_page = page;
837 truncate_node(dn);
838 freed++;
839 } else {
840 f2fs_put_page(page, 1);
841 }
842 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
843 return freed;
844
845out_err:
846 f2fs_put_page(page, 1);
847 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
848 return ret;
849}
850
851static int truncate_partial_nodes(struct dnode_of_data *dn,
852 struct f2fs_inode *ri, int *offset, int depth)
853{
854 struct page *pages[2];
855 nid_t nid[3];
856 nid_t child_nid;
857 int err = 0;
858 int i;
859 int idx = depth - 2;
860
861 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
862 if (!nid[0])
863 return 0;
864
865 /* get indirect nodes in the path */
866 for (i = 0; i < idx + 1; i++) {
867 /* reference count'll be increased */
868 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
869 if (IS_ERR(pages[i])) {
870 err = PTR_ERR(pages[i]);
871 idx = i - 1;
872 goto fail;
873 }
874 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
875 }
876
877 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
878
879 /* free direct nodes linked to a partial indirect node */
880 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
881 child_nid = get_nid(pages[idx], i, false);
882 if (!child_nid)
883 continue;
884 dn->nid = child_nid;
885 err = truncate_dnode(dn);
886 if (err < 0)
887 goto fail;
888 if (set_nid(pages[idx], i, 0, false))
889 dn->node_changed = true;
890 }
891
892 if (offset[idx + 1] == 0) {
893 dn->node_page = pages[idx];
894 dn->nid = nid[idx];
895 truncate_node(dn);
896 } else {
897 f2fs_put_page(pages[idx], 1);
898 }
899 offset[idx]++;
900 offset[idx + 1] = 0;
901 idx--;
902fail:
903 for (i = idx; i >= 0; i--)
904 f2fs_put_page(pages[i], 1);
905
906 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
907
908 return err;
909}
910
911/*
912 * All the block addresses of data and nodes should be nullified.
913 */
914int truncate_inode_blocks(struct inode *inode, pgoff_t from)
915{
916 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
917 int err = 0, cont = 1;
918 int level, offset[4], noffset[4];
919 unsigned int nofs = 0;
920 struct f2fs_inode *ri;
921 struct dnode_of_data dn;
922 struct page *page;
923
924 trace_f2fs_truncate_inode_blocks_enter(inode, from);
925
926 level = get_node_path(inode, from, offset, noffset);
927 if (level < 0)
928 return level;
929
930 page = get_node_page(sbi, inode->i_ino);
931 if (IS_ERR(page)) {
932 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
933 return PTR_ERR(page);
934 }
935
936 set_new_dnode(&dn, inode, page, NULL, 0);
937 unlock_page(page);
938
939 ri = F2FS_INODE(page);
940 switch (level) {
941 case 0:
942 case 1:
943 nofs = noffset[1];
944 break;
945 case 2:
946 nofs = noffset[1];
947 if (!offset[level - 1])
948 goto skip_partial;
949 err = truncate_partial_nodes(&dn, ri, offset, level);
950 if (err < 0 && err != -ENOENT)
951 goto fail;
952 nofs += 1 + NIDS_PER_BLOCK;
953 break;
954 case 3:
955 nofs = 5 + 2 * NIDS_PER_BLOCK;
956 if (!offset[level - 1])
957 goto skip_partial;
958 err = truncate_partial_nodes(&dn, ri, offset, level);
959 if (err < 0 && err != -ENOENT)
960 goto fail;
961 break;
962 default:
963 BUG();
964 }
965
966skip_partial:
967 while (cont) {
968 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
969 switch (offset[0]) {
970 case NODE_DIR1_BLOCK:
971 case NODE_DIR2_BLOCK:
972 err = truncate_dnode(&dn);
973 break;
974
975 case NODE_IND1_BLOCK:
976 case NODE_IND2_BLOCK:
977 err = truncate_nodes(&dn, nofs, offset[1], 2);
978 break;
979
980 case NODE_DIND_BLOCK:
981 err = truncate_nodes(&dn, nofs, offset[1], 3);
982 cont = 0;
983 break;
984
985 default:
986 BUG();
987 }
988 if (err < 0 && err != -ENOENT)
989 goto fail;
990 if (offset[1] == 0 &&
991 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
992 lock_page(page);
993 BUG_ON(page->mapping != NODE_MAPPING(sbi));
994 f2fs_wait_on_page_writeback(page, NODE, true);
995 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
996 set_page_dirty(page);
997 unlock_page(page);
998 }
999 offset[1] = 0;
1000 offset[0]++;
1001 nofs += err;
1002 }
1003fail:
1004 f2fs_put_page(page, 0);
1005 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1006 return err > 0 ? 0 : err;
1007}
1008
1009/* caller must lock inode page */
1010int truncate_xattr_node(struct inode *inode)
1011{
1012 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1013 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1014 struct dnode_of_data dn;
1015 struct page *npage;
1016
1017 if (!nid)
1018 return 0;
1019
1020 npage = get_node_page(sbi, nid);
1021 if (IS_ERR(npage))
1022 return PTR_ERR(npage);
1023
1024 f2fs_i_xnid_write(inode, 0);
1025
1026 set_new_dnode(&dn, inode, NULL, npage, nid);
1027 truncate_node(&dn);
1028 return 0;
1029}
1030
1031/*
1032 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1033 * f2fs_unlock_op().
1034 */
1035int remove_inode_page(struct inode *inode)
1036{
1037 struct dnode_of_data dn;
1038 int err;
1039
1040 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1041 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1042 if (err)
1043 return err;
1044
1045 err = truncate_xattr_node(inode);
1046 if (err) {
1047 f2fs_put_dnode(&dn);
1048 return err;
1049 }
1050
1051 /* remove potential inline_data blocks */
1052 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1053 S_ISLNK(inode->i_mode))
1054 truncate_data_blocks_range(&dn, 1);
1055
1056 /* 0 is possible, after f2fs_new_inode() has failed */
1057 f2fs_bug_on(F2FS_I_SB(inode),
1058 inode->i_blocks != 0 && inode->i_blocks != 8);
1059
1060 /* will put inode & node pages */
1061 truncate_node(&dn);
1062 return 0;
1063}
1064
1065struct page *new_inode_page(struct inode *inode)
1066{
1067 struct dnode_of_data dn;
1068
1069 /* allocate inode page for new inode */
1070 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1071
1072 /* caller should f2fs_put_page(page, 1); */
1073 return new_node_page(&dn, 0);
1074}
1075
1076struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1077{
1078 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1079 struct node_info new_ni;
1080 struct page *page;
1081 int err;
1082
1083 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1084 return ERR_PTR(-EPERM);
1085
1086 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1087 if (!page)
1088 return ERR_PTR(-ENOMEM);
1089
1090 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1091 goto fail;
1092
1093#ifdef CONFIG_F2FS_CHECK_FS
1094 get_node_info(sbi, dn->nid, &new_ni);
1095 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1096#endif
1097 new_ni.nid = dn->nid;
1098 new_ni.ino = dn->inode->i_ino;
1099 new_ni.blk_addr = NULL_ADDR;
1100 new_ni.flag = 0;
1101 new_ni.version = 0;
1102 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1103
1104 f2fs_wait_on_page_writeback(page, NODE, true);
1105 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1106 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1107 if (!PageUptodate(page))
1108 SetPageUptodate(page);
1109 if (set_page_dirty(page))
1110 dn->node_changed = true;
1111
1112 if (f2fs_has_xattr_block(ofs))
1113 f2fs_i_xnid_write(dn->inode, dn->nid);
1114
1115 if (ofs == 0)
1116 inc_valid_inode_count(sbi);
1117 return page;
1118
1119fail:
1120 clear_node_page_dirty(page);
1121 f2fs_put_page(page, 1);
1122 return ERR_PTR(err);
1123}
1124
1125/*
1126 * Caller should do after getting the following values.
1127 * 0: f2fs_put_page(page, 0)
1128 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1129 */
1130static int read_node_page(struct page *page, int op_flags)
1131{
1132 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1133 struct node_info ni;
1134 struct f2fs_io_info fio = {
1135 .sbi = sbi,
1136 .type = NODE,
1137 .op = REQ_OP_READ,
1138 .op_flags = op_flags,
1139 .page = page,
1140 .encrypted_page = NULL,
1141 };
1142
1143 if (PageUptodate(page))
1144 return LOCKED_PAGE;
1145
1146 get_node_info(sbi, page->index, &ni);
1147
1148 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1149 ClearPageUptodate(page);
1150 return -ENOENT;
1151 }
1152
1153 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1154 return f2fs_submit_page_bio(&fio);
1155}
1156
1157/*
1158 * Readahead a node page
1159 */
1160void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1161{
1162 struct page *apage;
1163 int err;
1164
1165 if (!nid)
1166 return;
1167 if (check_nid_range(sbi, nid))
1168 return;
1169
1170 rcu_read_lock();
1171 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->i_pages, nid);
1172 rcu_read_unlock();
1173 if (apage)
1174 return;
1175
1176 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1177 if (!apage)
1178 return;
1179
1180 err = read_node_page(apage, REQ_RAHEAD);
1181 f2fs_put_page(apage, err ? 1 : 0);
1182}
1183
1184static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1185 struct page *parent, int start)
1186{
1187 struct page *page;
1188 int err;
1189
1190 if (!nid)
1191 return ERR_PTR(-ENOENT);
1192 if (check_nid_range(sbi, nid))
1193 return ERR_PTR(-EINVAL);
1194repeat:
1195 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1196 if (!page)
1197 return ERR_PTR(-ENOMEM);
1198
1199 err = read_node_page(page, 0);
1200 if (err < 0) {
1201 f2fs_put_page(page, 1);
1202 return ERR_PTR(err);
1203 } else if (err == LOCKED_PAGE) {
1204 err = 0;
1205 goto page_hit;
1206 }
1207
1208 if (parent)
1209 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1210
1211 lock_page(page);
1212
1213 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1214 f2fs_put_page(page, 1);
1215 goto repeat;
1216 }
1217
1218 if (unlikely(!PageUptodate(page))) {
1219 err = -EIO;
1220 goto out_err;
1221 }
1222
1223 if (!f2fs_inode_chksum_verify(sbi, page)) {
1224 err = -EBADMSG;
1225 goto out_err;
1226 }
1227page_hit:
1228 if(unlikely(nid != nid_of_node(page))) {
1229 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1230 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1231 nid, nid_of_node(page), ino_of_node(page),
1232 ofs_of_node(page), cpver_of_node(page),
1233 next_blkaddr_of_node(page));
1234 err = -EINVAL;
1235out_err:
1236 ClearPageUptodate(page);
1237 f2fs_put_page(page, 1);
1238 return ERR_PTR(err);
1239 }
1240 return page;
1241}
1242
1243struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1244{
1245 return __get_node_page(sbi, nid, NULL, 0);
1246}
1247
1248struct page *get_node_page_ra(struct page *parent, int start)
1249{
1250 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1251 nid_t nid = get_nid(parent, start, false);
1252
1253 return __get_node_page(sbi, nid, parent, start);
1254}
1255
1256static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1257{
1258 struct inode *inode;
1259 struct page *page;
1260 int ret;
1261
1262 /* should flush inline_data before evict_inode */
1263 inode = ilookup(sbi->sb, ino);
1264 if (!inode)
1265 return;
1266
1267 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1268 FGP_LOCK|FGP_NOWAIT, 0);
1269 if (!page)
1270 goto iput_out;
1271
1272 if (!PageUptodate(page))
1273 goto page_out;
1274
1275 if (!PageDirty(page))
1276 goto page_out;
1277
1278 if (!clear_page_dirty_for_io(page))
1279 goto page_out;
1280
1281 ret = f2fs_write_inline_data(inode, page);
1282 inode_dec_dirty_pages(inode);
1283 remove_dirty_inode(inode);
1284 if (ret)
1285 set_page_dirty(page);
1286page_out:
1287 f2fs_put_page(page, 1);
1288iput_out:
1289 iput(inode);
1290}
1291
1292static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1293{
1294 pgoff_t index;
1295 struct pagevec pvec;
1296 struct page *last_page = NULL;
1297 int nr_pages;
1298
1299 pagevec_init(&pvec);
1300 index = 0;
1301
1302 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1303 PAGECACHE_TAG_DIRTY))) {
1304 int i;
1305
1306 for (i = 0; i < nr_pages; i++) {
1307 struct page *page = pvec.pages[i];
1308
1309 if (unlikely(f2fs_cp_error(sbi))) {
1310 f2fs_put_page(last_page, 0);
1311 pagevec_release(&pvec);
1312 return ERR_PTR(-EIO);
1313 }
1314
1315 if (!IS_DNODE(page) || !is_cold_node(page))
1316 continue;
1317 if (ino_of_node(page) != ino)
1318 continue;
1319
1320 lock_page(page);
1321
1322 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1323continue_unlock:
1324 unlock_page(page);
1325 continue;
1326 }
1327 if (ino_of_node(page) != ino)
1328 goto continue_unlock;
1329
1330 if (!PageDirty(page)) {
1331 /* someone wrote it for us */
1332 goto continue_unlock;
1333 }
1334
1335 if (last_page)
1336 f2fs_put_page(last_page, 0);
1337
1338 get_page(page);
1339 last_page = page;
1340 unlock_page(page);
1341 }
1342 pagevec_release(&pvec);
1343 cond_resched();
1344 }
1345 return last_page;
1346}
1347
1348static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1349 struct writeback_control *wbc, bool do_balance,
1350 enum iostat_type io_type)
1351{
1352 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1353 nid_t nid;
1354 struct node_info ni;
1355 struct f2fs_io_info fio = {
1356 .sbi = sbi,
1357 .ino = ino_of_node(page),
1358 .type = NODE,
1359 .op = REQ_OP_WRITE,
1360 .op_flags = wbc_to_write_flags(wbc),
1361 .page = page,
1362 .encrypted_page = NULL,
1363 .submitted = false,
1364 .io_type = io_type,
1365 .io_wbc = wbc,
1366 };
1367
1368 trace_f2fs_writepage(page, NODE);
1369
1370 if (unlikely(f2fs_cp_error(sbi)))
1371 goto redirty_out;
1372
1373 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1374 goto redirty_out;
1375
1376 /* get old block addr of this node page */
1377 nid = nid_of_node(page);
1378 f2fs_bug_on(sbi, page->index != nid);
1379
1380 if (wbc->for_reclaim) {
1381 if (!down_read_trylock(&sbi->node_write))
1382 goto redirty_out;
1383 } else {
1384 down_read(&sbi->node_write);
1385 }
1386
1387 get_node_info(sbi, nid, &ni);
1388
1389 /* This page is already truncated */
1390 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1391 ClearPageUptodate(page);
1392 dec_page_count(sbi, F2FS_DIRTY_NODES);
1393 up_read(&sbi->node_write);
1394 unlock_page(page);
1395 return 0;
1396 }
1397
1398 if (atomic && !test_opt(sbi, NOBARRIER))
1399 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1400
1401 set_page_writeback(page);
1402 ClearPageError(page);
1403 fio.old_blkaddr = ni.blk_addr;
1404 write_node_page(nid, &fio);
1405 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1406 dec_page_count(sbi, F2FS_DIRTY_NODES);
1407 up_read(&sbi->node_write);
1408
1409 if (wbc->for_reclaim) {
1410 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1411 page->index, NODE);
1412 submitted = NULL;
1413 }
1414
1415 unlock_page(page);
1416
1417 if (unlikely(f2fs_cp_error(sbi))) {
1418 f2fs_submit_merged_write(sbi, NODE);
1419 submitted = NULL;
1420 }
1421 if (submitted)
1422 *submitted = fio.submitted;
1423
1424 if (do_balance)
1425 f2fs_balance_fs(sbi, false);
1426 return 0;
1427
1428redirty_out:
1429 redirty_page_for_writepage(wbc, page);
1430 return AOP_WRITEPAGE_ACTIVATE;
1431}
1432
1433void move_node_page(struct page *node_page, int gc_type)
1434{
1435 if (gc_type == FG_GC) {
1436 struct writeback_control wbc = {
1437 .sync_mode = WB_SYNC_ALL,
1438 .nr_to_write = 1,
1439 .for_reclaim = 0,
1440 };
1441
1442 set_page_dirty(node_page);
1443 f2fs_wait_on_page_writeback(node_page, NODE, true);
1444
1445 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1446 if (!clear_page_dirty_for_io(node_page))
1447 goto out_page;
1448
1449 if (__write_node_page(node_page, false, NULL,
1450 &wbc, false, FS_GC_NODE_IO))
1451 unlock_page(node_page);
1452 goto release_page;
1453 } else {
1454 /* set page dirty and write it */
1455 if (!PageWriteback(node_page))
1456 set_page_dirty(node_page);
1457 }
1458out_page:
1459 unlock_page(node_page);
1460release_page:
1461 f2fs_put_page(node_page, 0);
1462}
1463
1464static int f2fs_write_node_page(struct page *page,
1465 struct writeback_control *wbc)
1466{
1467 return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1468}
1469
1470int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1471 struct writeback_control *wbc, bool atomic)
1472{
1473 pgoff_t index;
1474 pgoff_t last_idx = ULONG_MAX;
1475 struct pagevec pvec;
1476 int ret = 0;
1477 struct page *last_page = NULL;
1478 bool marked = false;
1479 nid_t ino = inode->i_ino;
1480 int nr_pages;
1481
1482 if (atomic) {
1483 last_page = last_fsync_dnode(sbi, ino);
1484 if (IS_ERR_OR_NULL(last_page))
1485 return PTR_ERR_OR_ZERO(last_page);
1486 }
1487retry:
1488 pagevec_init(&pvec);
1489 index = 0;
1490
1491 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1492 PAGECACHE_TAG_DIRTY))) {
1493 int i;
1494
1495 for (i = 0; i < nr_pages; i++) {
1496 struct page *page = pvec.pages[i];
1497 bool submitted = false;
1498
1499 if (unlikely(f2fs_cp_error(sbi))) {
1500 f2fs_put_page(last_page, 0);
1501 pagevec_release(&pvec);
1502 ret = -EIO;
1503 goto out;
1504 }
1505
1506 if (!IS_DNODE(page) || !is_cold_node(page))
1507 continue;
1508 if (ino_of_node(page) != ino)
1509 continue;
1510
1511 lock_page(page);
1512
1513 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1514continue_unlock:
1515 unlock_page(page);
1516 continue;
1517 }
1518 if (ino_of_node(page) != ino)
1519 goto continue_unlock;
1520
1521 if (!PageDirty(page) && page != last_page) {
1522 /* someone wrote it for us */
1523 goto continue_unlock;
1524 }
1525
1526 f2fs_wait_on_page_writeback(page, NODE, true);
1527 BUG_ON(PageWriteback(page));
1528
1529 set_fsync_mark(page, 0);
1530 set_dentry_mark(page, 0);
1531
1532 if (!atomic || page == last_page) {
1533 set_fsync_mark(page, 1);
1534 if (IS_INODE(page)) {
1535 if (is_inode_flag_set(inode,
1536 FI_DIRTY_INODE))
1537 update_inode(inode, page);
1538 set_dentry_mark(page,
1539 need_dentry_mark(sbi, ino));
1540 }
1541 /* may be written by other thread */
1542 if (!PageDirty(page))
1543 set_page_dirty(page);
1544 }
1545
1546 if (!clear_page_dirty_for_io(page))
1547 goto continue_unlock;
1548
1549 ret = __write_node_page(page, atomic &&
1550 page == last_page,
1551 &submitted, wbc, true,
1552 FS_NODE_IO);
1553 if (ret) {
1554 unlock_page(page);
1555 f2fs_put_page(last_page, 0);
1556 break;
1557 } else if (submitted) {
1558 last_idx = page->index;
1559 }
1560
1561 if (page == last_page) {
1562 f2fs_put_page(page, 0);
1563 marked = true;
1564 break;
1565 }
1566 }
1567 pagevec_release(&pvec);
1568 cond_resched();
1569
1570 if (ret || marked)
1571 break;
1572 }
1573 if (!ret && atomic && !marked) {
1574 f2fs_msg(sbi->sb, KERN_DEBUG,
1575 "Retry to write fsync mark: ino=%u, idx=%lx",
1576 ino, last_page->index);
1577 lock_page(last_page);
1578 f2fs_wait_on_page_writeback(last_page, NODE, true);
1579 set_page_dirty(last_page);
1580 unlock_page(last_page);
1581 goto retry;
1582 }
1583out:
1584 if (last_idx != ULONG_MAX)
1585 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1586 return ret ? -EIO: 0;
1587}
1588
1589int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1590 bool do_balance, enum iostat_type io_type)
1591{
1592 pgoff_t index;
1593 struct pagevec pvec;
1594 int step = 0;
1595 int nwritten = 0;
1596 int ret = 0;
1597 int nr_pages;
1598
1599 pagevec_init(&pvec);
1600
1601next_step:
1602 index = 0;
1603
1604 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1605 PAGECACHE_TAG_DIRTY))) {
1606 int i;
1607
1608 for (i = 0; i < nr_pages; i++) {
1609 struct page *page = pvec.pages[i];
1610 bool submitted = false;
1611
1612 /*
1613 * flushing sequence with step:
1614 * 0. indirect nodes
1615 * 1. dentry dnodes
1616 * 2. file dnodes
1617 */
1618 if (step == 0 && IS_DNODE(page))
1619 continue;
1620 if (step == 1 && (!IS_DNODE(page) ||
1621 is_cold_node(page)))
1622 continue;
1623 if (step == 2 && (!IS_DNODE(page) ||
1624 !is_cold_node(page)))
1625 continue;
1626lock_node:
1627 if (!trylock_page(page))
1628 continue;
1629
1630 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1631continue_unlock:
1632 unlock_page(page);
1633 continue;
1634 }
1635
1636 if (!PageDirty(page)) {
1637 /* someone wrote it for us */
1638 goto continue_unlock;
1639 }
1640
1641 /* flush inline_data */
1642 if (is_inline_node(page)) {
1643 clear_inline_node(page);
1644 unlock_page(page);
1645 flush_inline_data(sbi, ino_of_node(page));
1646 goto lock_node;
1647 }
1648
1649 f2fs_wait_on_page_writeback(page, NODE, true);
1650
1651 BUG_ON(PageWriteback(page));
1652 if (!clear_page_dirty_for_io(page))
1653 goto continue_unlock;
1654
1655 set_fsync_mark(page, 0);
1656 set_dentry_mark(page, 0);
1657
1658 ret = __write_node_page(page, false, &submitted,
1659 wbc, do_balance, io_type);
1660 if (ret)
1661 unlock_page(page);
1662 else if (submitted)
1663 nwritten++;
1664
1665 if (--wbc->nr_to_write == 0)
1666 break;
1667 }
1668 pagevec_release(&pvec);
1669 cond_resched();
1670
1671 if (wbc->nr_to_write == 0) {
1672 step = 2;
1673 break;
1674 }
1675 }
1676
1677 if (step < 2) {
1678 step++;
1679 goto next_step;
1680 }
1681
1682 if (nwritten)
1683 f2fs_submit_merged_write(sbi, NODE);
1684
1685 if (unlikely(f2fs_cp_error(sbi)))
1686 return -EIO;
1687 return ret;
1688}
1689
1690int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1691{
1692 pgoff_t index = 0;
1693 struct pagevec pvec;
1694 int ret2, ret = 0;
1695 int nr_pages;
1696
1697 pagevec_init(&pvec);
1698
1699 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1700 PAGECACHE_TAG_WRITEBACK))) {
1701 int i;
1702
1703 for (i = 0; i < nr_pages; i++) {
1704 struct page *page = pvec.pages[i];
1705
1706 if (ino && ino_of_node(page) == ino) {
1707 f2fs_wait_on_page_writeback(page, NODE, true);
1708 if (TestClearPageError(page))
1709 ret = -EIO;
1710 }
1711 }
1712 pagevec_release(&pvec);
1713 cond_resched();
1714 }
1715
1716 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1717 if (!ret)
1718 ret = ret2;
1719 return ret;
1720}
1721
1722static int f2fs_write_node_pages(struct address_space *mapping,
1723 struct writeback_control *wbc)
1724{
1725 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1726 struct blk_plug plug;
1727 long diff;
1728
1729 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1730 goto skip_write;
1731
1732 /* balancing f2fs's metadata in background */
1733 f2fs_balance_fs_bg(sbi);
1734
1735 /* collect a number of dirty node pages and write together */
1736 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1737 goto skip_write;
1738
1739 trace_f2fs_writepages(mapping->host, wbc, NODE);
1740
1741 diff = nr_pages_to_write(sbi, NODE, wbc);
1742 wbc->sync_mode = WB_SYNC_NONE;
1743 blk_start_plug(&plug);
1744 sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1745 blk_finish_plug(&plug);
1746 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1747 return 0;
1748
1749skip_write:
1750 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1751 trace_f2fs_writepages(mapping->host, wbc, NODE);
1752 return 0;
1753}
1754
1755static int f2fs_set_node_page_dirty(struct page *page)
1756{
1757 trace_f2fs_set_page_dirty(page, NODE);
1758
1759 if (!PageUptodate(page))
1760 SetPageUptodate(page);
1761 if (!PageDirty(page)) {
1762 __set_page_dirty_nobuffers(page);
1763 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1764 SetPagePrivate(page);
1765 f2fs_trace_pid(page);
1766 return 1;
1767 }
1768 return 0;
1769}
1770
1771/*
1772 * Structure of the f2fs node operations
1773 */
1774const struct address_space_operations f2fs_node_aops = {
1775 .writepage = f2fs_write_node_page,
1776 .writepages = f2fs_write_node_pages,
1777 .set_page_dirty = f2fs_set_node_page_dirty,
1778 .invalidatepage = f2fs_invalidate_page,
1779 .releasepage = f2fs_release_page,
1780#ifdef CONFIG_MIGRATION
1781 .migratepage = f2fs_migrate_page,
1782#endif
1783};
1784
1785static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1786 nid_t n)
1787{
1788 return radix_tree_lookup(&nm_i->free_nid_root, n);
1789}
1790
1791static int __insert_free_nid(struct f2fs_sb_info *sbi,
1792 struct free_nid *i, enum nid_state state)
1793{
1794 struct f2fs_nm_info *nm_i = NM_I(sbi);
1795
1796 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1797 if (err)
1798 return err;
1799
1800 f2fs_bug_on(sbi, state != i->state);
1801 nm_i->nid_cnt[state]++;
1802 if (state == FREE_NID)
1803 list_add_tail(&i->list, &nm_i->free_nid_list);
1804 return 0;
1805}
1806
1807static void __remove_free_nid(struct f2fs_sb_info *sbi,
1808 struct free_nid *i, enum nid_state state)
1809{
1810 struct f2fs_nm_info *nm_i = NM_I(sbi);
1811
1812 f2fs_bug_on(sbi, state != i->state);
1813 nm_i->nid_cnt[state]--;
1814 if (state == FREE_NID)
1815 list_del(&i->list);
1816 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1817}
1818
1819static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
1820 enum nid_state org_state, enum nid_state dst_state)
1821{
1822 struct f2fs_nm_info *nm_i = NM_I(sbi);
1823
1824 f2fs_bug_on(sbi, org_state != i->state);
1825 i->state = dst_state;
1826 nm_i->nid_cnt[org_state]--;
1827 nm_i->nid_cnt[dst_state]++;
1828
1829 switch (dst_state) {
1830 case PREALLOC_NID:
1831 list_del(&i->list);
1832 break;
1833 case FREE_NID:
1834 list_add_tail(&i->list, &nm_i->free_nid_list);
1835 break;
1836 default:
1837 BUG_ON(1);
1838 }
1839}
1840
1841static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1842 bool set, bool build)
1843{
1844 struct f2fs_nm_info *nm_i = NM_I(sbi);
1845 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1846 unsigned int nid_ofs = nid - START_NID(nid);
1847
1848 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1849 return;
1850
1851 if (set) {
1852 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1853 return;
1854 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1855 nm_i->free_nid_count[nat_ofs]++;
1856 } else {
1857 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1858 return;
1859 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1860 if (!build)
1861 nm_i->free_nid_count[nat_ofs]--;
1862 }
1863}
1864
1865/* return if the nid is recognized as free */
1866static bool add_free_nid(struct f2fs_sb_info *sbi,
1867 nid_t nid, bool build, bool update)
1868{
1869 struct f2fs_nm_info *nm_i = NM_I(sbi);
1870 struct free_nid *i, *e;
1871 struct nat_entry *ne;
1872 int err = -EINVAL;
1873 bool ret = false;
1874
1875 /* 0 nid should not be used */
1876 if (unlikely(nid == 0))
1877 return false;
1878
1879 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1880 i->nid = nid;
1881 i->state = FREE_NID;
1882
1883 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
1884
1885 spin_lock(&nm_i->nid_list_lock);
1886
1887 if (build) {
1888 /*
1889 * Thread A Thread B
1890 * - f2fs_create
1891 * - f2fs_new_inode
1892 * - alloc_nid
1893 * - __insert_nid_to_list(PREALLOC_NID)
1894 * - f2fs_balance_fs_bg
1895 * - build_free_nids
1896 * - __build_free_nids
1897 * - scan_nat_page
1898 * - add_free_nid
1899 * - __lookup_nat_cache
1900 * - f2fs_add_link
1901 * - init_inode_metadata
1902 * - new_inode_page
1903 * - new_node_page
1904 * - set_node_addr
1905 * - alloc_nid_done
1906 * - __remove_nid_from_list(PREALLOC_NID)
1907 * - __insert_nid_to_list(FREE_NID)
1908 */
1909 ne = __lookup_nat_cache(nm_i, nid);
1910 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1911 nat_get_blkaddr(ne) != NULL_ADDR))
1912 goto err_out;
1913
1914 e = __lookup_free_nid_list(nm_i, nid);
1915 if (e) {
1916 if (e->state == FREE_NID)
1917 ret = true;
1918 goto err_out;
1919 }
1920 }
1921 ret = true;
1922 err = __insert_free_nid(sbi, i, FREE_NID);
1923err_out:
1924 if (update) {
1925 update_free_nid_bitmap(sbi, nid, ret, build);
1926 if (!build)
1927 nm_i->available_nids++;
1928 }
1929 spin_unlock(&nm_i->nid_list_lock);
1930 radix_tree_preload_end();
1931
1932 if (err)
1933 kmem_cache_free(free_nid_slab, i);
1934 return ret;
1935}
1936
1937static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1938{
1939 struct f2fs_nm_info *nm_i = NM_I(sbi);
1940 struct free_nid *i;
1941 bool need_free = false;
1942
1943 spin_lock(&nm_i->nid_list_lock);
1944 i = __lookup_free_nid_list(nm_i, nid);
1945 if (i && i->state == FREE_NID) {
1946 __remove_free_nid(sbi, i, FREE_NID);
1947 need_free = true;
1948 }
1949 spin_unlock(&nm_i->nid_list_lock);
1950
1951 if (need_free)
1952 kmem_cache_free(free_nid_slab, i);
1953}
1954
1955static void scan_nat_page(struct f2fs_sb_info *sbi,
1956 struct page *nat_page, nid_t start_nid)
1957{
1958 struct f2fs_nm_info *nm_i = NM_I(sbi);
1959 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1960 block_t blk_addr;
1961 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1962 int i;
1963
1964 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1965
1966 i = start_nid % NAT_ENTRY_PER_BLOCK;
1967
1968 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1969 if (unlikely(start_nid >= nm_i->max_nid))
1970 break;
1971
1972 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1973 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1974 if (blk_addr == NULL_ADDR) {
1975 add_free_nid(sbi, start_nid, true, true);
1976 } else {
1977 spin_lock(&NM_I(sbi)->nid_list_lock);
1978 update_free_nid_bitmap(sbi, start_nid, false, true);
1979 spin_unlock(&NM_I(sbi)->nid_list_lock);
1980 }
1981 }
1982}
1983
1984static void scan_curseg_cache(struct f2fs_sb_info *sbi)
1985{
1986 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1987 struct f2fs_journal *journal = curseg->journal;
1988 int i;
1989
1990 down_read(&curseg->journal_rwsem);
1991 for (i = 0; i < nats_in_cursum(journal); i++) {
1992 block_t addr;
1993 nid_t nid;
1994
1995 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1996 nid = le32_to_cpu(nid_in_journal(journal, i));
1997 if (addr == NULL_ADDR)
1998 add_free_nid(sbi, nid, true, false);
1999 else
2000 remove_free_nid(sbi, nid);
2001 }
2002 up_read(&curseg->journal_rwsem);
2003}
2004
2005static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2006{
2007 struct f2fs_nm_info *nm_i = NM_I(sbi);
2008 unsigned int i, idx;
2009 nid_t nid;
2010
2011 down_read(&nm_i->nat_tree_lock);
2012
2013 for (i = 0; i < nm_i->nat_blocks; i++) {
2014 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2015 continue;
2016 if (!nm_i->free_nid_count[i])
2017 continue;
2018 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2019 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2020 NAT_ENTRY_PER_BLOCK, idx);
2021 if (idx >= NAT_ENTRY_PER_BLOCK)
2022 break;
2023
2024 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2025 add_free_nid(sbi, nid, true, false);
2026
2027 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2028 goto out;
2029 }
2030 }
2031out:
2032 scan_curseg_cache(sbi);
2033
2034 up_read(&nm_i->nat_tree_lock);
2035}
2036
2037static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2038{
2039 struct f2fs_nm_info *nm_i = NM_I(sbi);
2040 int i = 0;
2041 nid_t nid = nm_i->next_scan_nid;
2042
2043 if (unlikely(nid >= nm_i->max_nid))
2044 nid = 0;
2045
2046 /* Enough entries */
2047 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2048 return;
2049
2050 if (!sync && !available_free_memory(sbi, FREE_NIDS))
2051 return;
2052
2053 if (!mount) {
2054 /* try to find free nids in free_nid_bitmap */
2055 scan_free_nid_bits(sbi);
2056
2057 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2058 return;
2059 }
2060
2061 /* readahead nat pages to be scanned */
2062 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2063 META_NAT, true);
2064
2065 down_read(&nm_i->nat_tree_lock);
2066
2067 while (1) {
2068 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2069 nm_i->nat_block_bitmap)) {
2070 struct page *page = get_current_nat_page(sbi, nid);
2071
2072 scan_nat_page(sbi, page, nid);
2073 f2fs_put_page(page, 1);
2074 }
2075
2076 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2077 if (unlikely(nid >= nm_i->max_nid))
2078 nid = 0;
2079
2080 if (++i >= FREE_NID_PAGES)
2081 break;
2082 }
2083
2084 /* go to the next free nat pages to find free nids abundantly */
2085 nm_i->next_scan_nid = nid;
2086
2087 /* find free nids from current sum_pages */
2088 scan_curseg_cache(sbi);
2089
2090 up_read(&nm_i->nat_tree_lock);
2091
2092 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2093 nm_i->ra_nid_pages, META_NAT, false);
2094}
2095
2096void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2097{
2098 mutex_lock(&NM_I(sbi)->build_lock);
2099 __build_free_nids(sbi, sync, mount);
2100 mutex_unlock(&NM_I(sbi)->build_lock);
2101}
2102
2103/*
2104 * If this function returns success, caller can obtain a new nid
2105 * from second parameter of this function.
2106 * The returned nid could be used ino as well as nid when inode is created.
2107 */
2108bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2109{
2110 struct f2fs_nm_info *nm_i = NM_I(sbi);
2111 struct free_nid *i = NULL;
2112retry:
2113#ifdef CONFIG_F2FS_FAULT_INJECTION
2114 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2115 f2fs_show_injection_info(FAULT_ALLOC_NID);
2116 return false;
2117 }
2118#endif
2119 spin_lock(&nm_i->nid_list_lock);
2120
2121 if (unlikely(nm_i->available_nids == 0)) {
2122 spin_unlock(&nm_i->nid_list_lock);
2123 return false;
2124 }
2125
2126 /* We should not use stale free nids created by build_free_nids */
2127 if (nm_i->nid_cnt[FREE_NID] && !on_build_free_nids(nm_i)) {
2128 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2129 i = list_first_entry(&nm_i->free_nid_list,
2130 struct free_nid, list);
2131 *nid = i->nid;
2132
2133 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2134 nm_i->available_nids--;
2135
2136 update_free_nid_bitmap(sbi, *nid, false, false);
2137
2138 spin_unlock(&nm_i->nid_list_lock);
2139 return true;
2140 }
2141 spin_unlock(&nm_i->nid_list_lock);
2142
2143 /* Let's scan nat pages and its caches to get free nids */
2144 build_free_nids(sbi, true, false);
2145 goto retry;
2146}
2147
2148/*
2149 * alloc_nid() should be called prior to this function.
2150 */
2151void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2152{
2153 struct f2fs_nm_info *nm_i = NM_I(sbi);
2154 struct free_nid *i;
2155
2156 spin_lock(&nm_i->nid_list_lock);
2157 i = __lookup_free_nid_list(nm_i, nid);
2158 f2fs_bug_on(sbi, !i);
2159 __remove_free_nid(sbi, i, PREALLOC_NID);
2160 spin_unlock(&nm_i->nid_list_lock);
2161
2162 kmem_cache_free(free_nid_slab, i);
2163}
2164
2165/*
2166 * alloc_nid() should be called prior to this function.
2167 */
2168void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2169{
2170 struct f2fs_nm_info *nm_i = NM_I(sbi);
2171 struct free_nid *i;
2172 bool need_free = false;
2173
2174 if (!nid)
2175 return;
2176
2177 spin_lock(&nm_i->nid_list_lock);
2178 i = __lookup_free_nid_list(nm_i, nid);
2179 f2fs_bug_on(sbi, !i);
2180
2181 if (!available_free_memory(sbi, FREE_NIDS)) {
2182 __remove_free_nid(sbi, i, PREALLOC_NID);
2183 need_free = true;
2184 } else {
2185 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2186 }
2187
2188 nm_i->available_nids++;
2189
2190 update_free_nid_bitmap(sbi, nid, true, false);
2191
2192 spin_unlock(&nm_i->nid_list_lock);
2193
2194 if (need_free)
2195 kmem_cache_free(free_nid_slab, i);
2196}
2197
2198int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2199{
2200 struct f2fs_nm_info *nm_i = NM_I(sbi);
2201 struct free_nid *i, *next;
2202 int nr = nr_shrink;
2203
2204 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2205 return 0;
2206
2207 if (!mutex_trylock(&nm_i->build_lock))
2208 return 0;
2209
2210 spin_lock(&nm_i->nid_list_lock);
2211 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2212 if (nr_shrink <= 0 ||
2213 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2214 break;
2215
2216 __remove_free_nid(sbi, i, FREE_NID);
2217 kmem_cache_free(free_nid_slab, i);
2218 nr_shrink--;
2219 }
2220 spin_unlock(&nm_i->nid_list_lock);
2221 mutex_unlock(&nm_i->build_lock);
2222
2223 return nr - nr_shrink;
2224}
2225
2226void recover_inline_xattr(struct inode *inode, struct page *page)
2227{
2228 void *src_addr, *dst_addr;
2229 size_t inline_size;
2230 struct page *ipage;
2231 struct f2fs_inode *ri;
2232
2233 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2234 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2235
2236 ri = F2FS_INODE(page);
2237 if (ri->i_inline & F2FS_INLINE_XATTR) {
2238 set_inode_flag(inode, FI_INLINE_XATTR);
2239 } else {
2240 clear_inode_flag(inode, FI_INLINE_XATTR);
2241 goto update_inode;
2242 }
2243
2244 dst_addr = inline_xattr_addr(inode, ipage);
2245 src_addr = inline_xattr_addr(inode, page);
2246 inline_size = inline_xattr_size(inode);
2247
2248 f2fs_wait_on_page_writeback(ipage, NODE, true);
2249 memcpy(dst_addr, src_addr, inline_size);
2250update_inode:
2251 update_inode(inode, ipage);
2252 f2fs_put_page(ipage, 1);
2253}
2254
2255int recover_xattr_data(struct inode *inode, struct page *page)
2256{
2257 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2258 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2259 nid_t new_xnid;
2260 struct dnode_of_data dn;
2261 struct node_info ni;
2262 struct page *xpage;
2263
2264 if (!prev_xnid)
2265 goto recover_xnid;
2266
2267 /* 1: invalidate the previous xattr nid */
2268 get_node_info(sbi, prev_xnid, &ni);
2269 invalidate_blocks(sbi, ni.blk_addr);
2270 dec_valid_node_count(sbi, inode, false);
2271 set_node_addr(sbi, &ni, NULL_ADDR, false);
2272
2273recover_xnid:
2274 /* 2: update xattr nid in inode */
2275 if (!alloc_nid(sbi, &new_xnid))
2276 return -ENOSPC;
2277
2278 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2279 xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2280 if (IS_ERR(xpage)) {
2281 alloc_nid_failed(sbi, new_xnid);
2282 return PTR_ERR(xpage);
2283 }
2284
2285 alloc_nid_done(sbi, new_xnid);
2286 update_inode_page(inode);
2287
2288 /* 3: update and set xattr node page dirty */
2289 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2290
2291 set_page_dirty(xpage);
2292 f2fs_put_page(xpage, 1);
2293
2294 return 0;
2295}
2296
2297int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2298{
2299 struct f2fs_inode *src, *dst;
2300 nid_t ino = ino_of_node(page);
2301 struct node_info old_ni, new_ni;
2302 struct page *ipage;
2303
2304 get_node_info(sbi, ino, &old_ni);
2305
2306 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2307 return -EINVAL;
2308retry:
2309 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2310 if (!ipage) {
2311 congestion_wait(BLK_RW_ASYNC, HZ/50);
2312 goto retry;
2313 }
2314
2315 /* Should not use this inode from free nid list */
2316 remove_free_nid(sbi, ino);
2317
2318 if (!PageUptodate(ipage))
2319 SetPageUptodate(ipage);
2320 fill_node_footer(ipage, ino, ino, 0, true);
2321 set_cold_node(page, false);
2322
2323 src = F2FS_INODE(page);
2324 dst = F2FS_INODE(ipage);
2325
2326 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2327 dst->i_size = 0;
2328 dst->i_blocks = cpu_to_le64(1);
2329 dst->i_links = cpu_to_le32(1);
2330 dst->i_xattr_nid = 0;
2331 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2332 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2333 dst->i_extra_isize = src->i_extra_isize;
2334
2335 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2336 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2337 i_inline_xattr_size))
2338 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2339
2340 if (f2fs_sb_has_project_quota(sbi->sb) &&
2341 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2342 i_projid))
2343 dst->i_projid = src->i_projid;
2344 }
2345
2346 new_ni = old_ni;
2347 new_ni.ino = ino;
2348
2349 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2350 WARN_ON(1);
2351 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2352 inc_valid_inode_count(sbi);
2353 set_page_dirty(ipage);
2354 f2fs_put_page(ipage, 1);
2355 return 0;
2356}
2357
2358void restore_node_summary(struct f2fs_sb_info *sbi,
2359 unsigned int segno, struct f2fs_summary_block *sum)
2360{
2361 struct f2fs_node *rn;
2362 struct f2fs_summary *sum_entry;
2363 block_t addr;
2364 int i, idx, last_offset, nrpages;
2365
2366 /* scan the node segment */
2367 last_offset = sbi->blocks_per_seg;
2368 addr = START_BLOCK(sbi, segno);
2369 sum_entry = &sum->entries[0];
2370
2371 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2372 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2373
2374 /* readahead node pages */
2375 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2376
2377 for (idx = addr; idx < addr + nrpages; idx++) {
2378 struct page *page = get_tmp_page(sbi, idx);
2379
2380 rn = F2FS_NODE(page);
2381 sum_entry->nid = rn->footer.nid;
2382 sum_entry->version = 0;
2383 sum_entry->ofs_in_node = 0;
2384 sum_entry++;
2385 f2fs_put_page(page, 1);
2386 }
2387
2388 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2389 addr + nrpages);
2390 }
2391}
2392
2393static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2394{
2395 struct f2fs_nm_info *nm_i = NM_I(sbi);
2396 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2397 struct f2fs_journal *journal = curseg->journal;
2398 int i;
2399
2400 down_write(&curseg->journal_rwsem);
2401 for (i = 0; i < nats_in_cursum(journal); i++) {
2402 struct nat_entry *ne;
2403 struct f2fs_nat_entry raw_ne;
2404 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2405
2406 raw_ne = nat_in_journal(journal, i);
2407
2408 ne = __lookup_nat_cache(nm_i, nid);
2409 if (!ne) {
2410 ne = __alloc_nat_entry(nid, true);
2411 __init_nat_entry(nm_i, ne, &raw_ne, true);
2412 }
2413
2414 /*
2415 * if a free nat in journal has not been used after last
2416 * checkpoint, we should remove it from available nids,
2417 * since later we will add it again.
2418 */
2419 if (!get_nat_flag(ne, IS_DIRTY) &&
2420 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2421 spin_lock(&nm_i->nid_list_lock);
2422 nm_i->available_nids--;
2423 spin_unlock(&nm_i->nid_list_lock);
2424 }
2425
2426 __set_nat_cache_dirty(nm_i, ne);
2427 }
2428 update_nats_in_cursum(journal, -i);
2429 up_write(&curseg->journal_rwsem);
2430}
2431
2432static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2433 struct list_head *head, int max)
2434{
2435 struct nat_entry_set *cur;
2436
2437 if (nes->entry_cnt >= max)
2438 goto add_out;
2439
2440 list_for_each_entry(cur, head, set_list) {
2441 if (cur->entry_cnt >= nes->entry_cnt) {
2442 list_add(&nes->set_list, cur->set_list.prev);
2443 return;
2444 }
2445 }
2446add_out:
2447 list_add_tail(&nes->set_list, head);
2448}
2449
2450static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2451 struct page *page)
2452{
2453 struct f2fs_nm_info *nm_i = NM_I(sbi);
2454 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2455 struct f2fs_nat_block *nat_blk = page_address(page);
2456 int valid = 0;
2457 int i = 0;
2458
2459 if (!enabled_nat_bits(sbi, NULL))
2460 return;
2461
2462 if (nat_index == 0) {
2463 valid = 1;
2464 i = 1;
2465 }
2466 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2467 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2468 valid++;
2469 }
2470 if (valid == 0) {
2471 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2472 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2473 return;
2474 }
2475
2476 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2477 if (valid == NAT_ENTRY_PER_BLOCK)
2478 __set_bit_le(nat_index, nm_i->full_nat_bits);
2479 else
2480 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2481}
2482
2483static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2484 struct nat_entry_set *set, struct cp_control *cpc)
2485{
2486 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2487 struct f2fs_journal *journal = curseg->journal;
2488 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2489 bool to_journal = true;
2490 struct f2fs_nat_block *nat_blk;
2491 struct nat_entry *ne, *cur;
2492 struct page *page = NULL;
2493
2494 /*
2495 * there are two steps to flush nat entries:
2496 * #1, flush nat entries to journal in current hot data summary block.
2497 * #2, flush nat entries to nat page.
2498 */
2499 if (enabled_nat_bits(sbi, cpc) ||
2500 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2501 to_journal = false;
2502
2503 if (to_journal) {
2504 down_write(&curseg->journal_rwsem);
2505 } else {
2506 page = get_next_nat_page(sbi, start_nid);
2507 nat_blk = page_address(page);
2508 f2fs_bug_on(sbi, !nat_blk);
2509 }
2510
2511 /* flush dirty nats in nat entry set */
2512 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2513 struct f2fs_nat_entry *raw_ne;
2514 nid_t nid = nat_get_nid(ne);
2515 int offset;
2516
2517 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2518
2519 if (to_journal) {
2520 offset = lookup_journal_in_cursum(journal,
2521 NAT_JOURNAL, nid, 1);
2522 f2fs_bug_on(sbi, offset < 0);
2523 raw_ne = &nat_in_journal(journal, offset);
2524 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2525 } else {
2526 raw_ne = &nat_blk->entries[nid - start_nid];
2527 }
2528 raw_nat_from_node_info(raw_ne, &ne->ni);
2529 nat_reset_flag(ne);
2530 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2531 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2532 add_free_nid(sbi, nid, false, true);
2533 } else {
2534 spin_lock(&NM_I(sbi)->nid_list_lock);
2535 update_free_nid_bitmap(sbi, nid, false, false);
2536 spin_unlock(&NM_I(sbi)->nid_list_lock);
2537 }
2538 }
2539
2540 if (to_journal) {
2541 up_write(&curseg->journal_rwsem);
2542 } else {
2543 __update_nat_bits(sbi, start_nid, page);
2544 f2fs_put_page(page, 1);
2545 }
2546
2547 /* Allow dirty nats by node block allocation in write_begin */
2548 if (!set->entry_cnt) {
2549 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2550 kmem_cache_free(nat_entry_set_slab, set);
2551 }
2552}
2553
2554/*
2555 * This function is called during the checkpointing process.
2556 */
2557void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2558{
2559 struct f2fs_nm_info *nm_i = NM_I(sbi);
2560 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2561 struct f2fs_journal *journal = curseg->journal;
2562 struct nat_entry_set *setvec[SETVEC_SIZE];
2563 struct nat_entry_set *set, *tmp;
2564 unsigned int found;
2565 nid_t set_idx = 0;
2566 LIST_HEAD(sets);
2567
2568 if (!nm_i->dirty_nat_cnt)
2569 return;
2570
2571 down_write(&nm_i->nat_tree_lock);
2572
2573 /*
2574 * if there are no enough space in journal to store dirty nat
2575 * entries, remove all entries from journal and merge them
2576 * into nat entry set.
2577 */
2578 if (enabled_nat_bits(sbi, cpc) ||
2579 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2580 remove_nats_in_journal(sbi);
2581
2582 while ((found = __gang_lookup_nat_set(nm_i,
2583 set_idx, SETVEC_SIZE, setvec))) {
2584 unsigned idx;
2585 set_idx = setvec[found - 1]->set + 1;
2586 for (idx = 0; idx < found; idx++)
2587 __adjust_nat_entry_set(setvec[idx], &sets,
2588 MAX_NAT_JENTRIES(journal));
2589 }
2590
2591 /* flush dirty nats in nat entry set */
2592 list_for_each_entry_safe(set, tmp, &sets, set_list)
2593 __flush_nat_entry_set(sbi, set, cpc);
2594
2595 up_write(&nm_i->nat_tree_lock);
2596 /* Allow dirty nats by node block allocation in write_begin */
2597}
2598
2599static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2600{
2601 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2602 struct f2fs_nm_info *nm_i = NM_I(sbi);
2603 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2604 unsigned int i;
2605 __u64 cp_ver = cur_cp_version(ckpt);
2606 block_t nat_bits_addr;
2607
2608 if (!enabled_nat_bits(sbi, NULL))
2609 return 0;
2610
2611 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2612 nm_i->nat_bits = f2fs_kzalloc(sbi,
2613 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2614 if (!nm_i->nat_bits)
2615 return -ENOMEM;
2616
2617 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2618 nm_i->nat_bits_blocks;
2619 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2620 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2621
2622 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2623 page_address(page), F2FS_BLKSIZE);
2624 f2fs_put_page(page, 1);
2625 }
2626
2627 cp_ver |= (cur_cp_crc(ckpt) << 32);
2628 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2629 disable_nat_bits(sbi, true);
2630 return 0;
2631 }
2632
2633 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2634 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2635
2636 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2637 return 0;
2638}
2639
2640static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2641{
2642 struct f2fs_nm_info *nm_i = NM_I(sbi);
2643 unsigned int i = 0;
2644 nid_t nid, last_nid;
2645
2646 if (!enabled_nat_bits(sbi, NULL))
2647 return;
2648
2649 for (i = 0; i < nm_i->nat_blocks; i++) {
2650 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2651 if (i >= nm_i->nat_blocks)
2652 break;
2653
2654 __set_bit_le(i, nm_i->nat_block_bitmap);
2655
2656 nid = i * NAT_ENTRY_PER_BLOCK;
2657 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2658
2659 spin_lock(&NM_I(sbi)->nid_list_lock);
2660 for (; nid < last_nid; nid++)
2661 update_free_nid_bitmap(sbi, nid, true, true);
2662 spin_unlock(&NM_I(sbi)->nid_list_lock);
2663 }
2664
2665 for (i = 0; i < nm_i->nat_blocks; i++) {
2666 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2667 if (i >= nm_i->nat_blocks)
2668 break;
2669
2670 __set_bit_le(i, nm_i->nat_block_bitmap);
2671 }
2672}
2673
2674static int init_node_manager(struct f2fs_sb_info *sbi)
2675{
2676 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2677 struct f2fs_nm_info *nm_i = NM_I(sbi);
2678 unsigned char *version_bitmap;
2679 unsigned int nat_segs;
2680 int err;
2681
2682 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2683
2684 /* segment_count_nat includes pair segment so divide to 2. */
2685 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2686 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2687 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2688
2689 /* not used nids: 0, node, meta, (and root counted as valid node) */
2690 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2691 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2692 nm_i->nid_cnt[FREE_NID] = 0;
2693 nm_i->nid_cnt[PREALLOC_NID] = 0;
2694 nm_i->nat_cnt = 0;
2695 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2696 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2697 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2698
2699 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2700 INIT_LIST_HEAD(&nm_i->free_nid_list);
2701 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2702 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2703 INIT_LIST_HEAD(&nm_i->nat_entries);
2704
2705 mutex_init(&nm_i->build_lock);
2706 spin_lock_init(&nm_i->nid_list_lock);
2707 init_rwsem(&nm_i->nat_tree_lock);
2708
2709 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2710 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2711 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2712 if (!version_bitmap)
2713 return -EFAULT;
2714
2715 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2716 GFP_KERNEL);
2717 if (!nm_i->nat_bitmap)
2718 return -ENOMEM;
2719
2720 err = __get_nat_bitmaps(sbi);
2721 if (err)
2722 return err;
2723
2724#ifdef CONFIG_F2FS_CHECK_FS
2725 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2726 GFP_KERNEL);
2727 if (!nm_i->nat_bitmap_mir)
2728 return -ENOMEM;
2729#endif
2730
2731 return 0;
2732}
2733
2734static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2735{
2736 struct f2fs_nm_info *nm_i = NM_I(sbi);
2737 int i;
2738
2739 nm_i->free_nid_bitmap = f2fs_kzalloc(sbi, nm_i->nat_blocks *
2740 sizeof(unsigned char *), GFP_KERNEL);
2741 if (!nm_i->free_nid_bitmap)
2742 return -ENOMEM;
2743
2744 for (i = 0; i < nm_i->nat_blocks; i++) {
2745 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
2746 NAT_ENTRY_BITMAP_SIZE_ALIGNED, GFP_KERNEL);
2747 if (!nm_i->free_nid_bitmap)
2748 return -ENOMEM;
2749 }
2750
2751 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
2752 GFP_KERNEL);
2753 if (!nm_i->nat_block_bitmap)
2754 return -ENOMEM;
2755
2756 nm_i->free_nid_count = f2fs_kvzalloc(sbi, nm_i->nat_blocks *
2757 sizeof(unsigned short), GFP_KERNEL);
2758 if (!nm_i->free_nid_count)
2759 return -ENOMEM;
2760 return 0;
2761}
2762
2763int build_node_manager(struct f2fs_sb_info *sbi)
2764{
2765 int err;
2766
2767 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
2768 GFP_KERNEL);
2769 if (!sbi->nm_info)
2770 return -ENOMEM;
2771
2772 err = init_node_manager(sbi);
2773 if (err)
2774 return err;
2775
2776 err = init_free_nid_cache(sbi);
2777 if (err)
2778 return err;
2779
2780 /* load free nid status from nat_bits table */
2781 load_free_nid_bitmap(sbi);
2782
2783 build_free_nids(sbi, true, true);
2784 return 0;
2785}
2786
2787void destroy_node_manager(struct f2fs_sb_info *sbi)
2788{
2789 struct f2fs_nm_info *nm_i = NM_I(sbi);
2790 struct free_nid *i, *next_i;
2791 struct nat_entry *natvec[NATVEC_SIZE];
2792 struct nat_entry_set *setvec[SETVEC_SIZE];
2793 nid_t nid = 0;
2794 unsigned int found;
2795
2796 if (!nm_i)
2797 return;
2798
2799 /* destroy free nid list */
2800 spin_lock(&nm_i->nid_list_lock);
2801 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2802 __remove_free_nid(sbi, i, FREE_NID);
2803 spin_unlock(&nm_i->nid_list_lock);
2804 kmem_cache_free(free_nid_slab, i);
2805 spin_lock(&nm_i->nid_list_lock);
2806 }
2807 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2808 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2809 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2810 spin_unlock(&nm_i->nid_list_lock);
2811
2812 /* destroy nat cache */
2813 down_write(&nm_i->nat_tree_lock);
2814 while ((found = __gang_lookup_nat_cache(nm_i,
2815 nid, NATVEC_SIZE, natvec))) {
2816 unsigned idx;
2817
2818 nid = nat_get_nid(natvec[found - 1]) + 1;
2819 for (idx = 0; idx < found; idx++)
2820 __del_from_nat_cache(nm_i, natvec[idx]);
2821 }
2822 f2fs_bug_on(sbi, nm_i->nat_cnt);
2823
2824 /* destroy nat set cache */
2825 nid = 0;
2826 while ((found = __gang_lookup_nat_set(nm_i,
2827 nid, SETVEC_SIZE, setvec))) {
2828 unsigned idx;
2829
2830 nid = setvec[found - 1]->set + 1;
2831 for (idx = 0; idx < found; idx++) {
2832 /* entry_cnt is not zero, when cp_error was occurred */
2833 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2834 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2835 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2836 }
2837 }
2838 up_write(&nm_i->nat_tree_lock);
2839
2840 kvfree(nm_i->nat_block_bitmap);
2841 if (nm_i->free_nid_bitmap) {
2842 int i;
2843
2844 for (i = 0; i < nm_i->nat_blocks; i++)
2845 kvfree(nm_i->free_nid_bitmap[i]);
2846 kfree(nm_i->free_nid_bitmap);
2847 }
2848 kvfree(nm_i->free_nid_count);
2849
2850 kfree(nm_i->nat_bitmap);
2851 kfree(nm_i->nat_bits);
2852#ifdef CONFIG_F2FS_CHECK_FS
2853 kfree(nm_i->nat_bitmap_mir);
2854#endif
2855 sbi->nm_info = NULL;
2856 kfree(nm_i);
2857}
2858
2859int __init create_node_manager_caches(void)
2860{
2861 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2862 sizeof(struct nat_entry));
2863 if (!nat_entry_slab)
2864 goto fail;
2865
2866 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2867 sizeof(struct free_nid));
2868 if (!free_nid_slab)
2869 goto destroy_nat_entry;
2870
2871 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2872 sizeof(struct nat_entry_set));
2873 if (!nat_entry_set_slab)
2874 goto destroy_free_nid;
2875 return 0;
2876
2877destroy_free_nid:
2878 kmem_cache_destroy(free_nid_slab);
2879destroy_nat_entry:
2880 kmem_cache_destroy(nat_entry_slab);
2881fail:
2882 return -ENOMEM;
2883}
2884
2885void destroy_node_manager_caches(void)
2886{
2887 kmem_cache_destroy(nat_entry_set_slab);
2888 kmem_cache_destroy(free_nid_slab);
2889 kmem_cache_destroy(nat_entry_slab);
2890}
Linux 6.x block layer and io_uring tree(s)