]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/f2fs/node.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: limit # of inmemory pages
[mirror_ubuntu-bionic-kernel.git] / fs / f2fs / node.c
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
28 static struct kmem_cache *nat_entry_slab;
29 static struct kmem_cache *free_nid_slab;
30 static struct kmem_cache *nat_entry_set_slab;
31
32 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
33 {
34 struct f2fs_nm_info *nm_i = NM_I(sbi);
35 struct sysinfo val;
36 unsigned long avail_ram;
37 unsigned long mem_size = 0;
38 bool res = false;
39
40 si_meminfo(&val);
41
42 /* only uses low memory */
43 avail_ram = val.totalram - val.totalhigh;
44
45 /*
46 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
47 */
48 if (type == FREE_NIDS) {
49 mem_size = (nm_i->nid_cnt[FREE_NID] *
50 sizeof(struct free_nid)) >> PAGE_SHIFT;
51 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
52 } else if (type == NAT_ENTRIES) {
53 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
54 PAGE_SHIFT;
55 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
56 if (excess_cached_nats(sbi))
57 res = false;
58 } else if (type == DIRTY_DENTS) {
59 if (sbi->sb->s_bdi->wb.dirty_exceeded)
60 return false;
61 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
62 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
63 } else if (type == INO_ENTRIES) {
64 int i;
65
66 for (i = 0; i < MAX_INO_ENTRY; i++)
67 mem_size += sbi->im[i].ino_num *
68 sizeof(struct ino_entry);
69 mem_size >>= PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
71 } else if (type == EXTENT_CACHE) {
72 mem_size = (atomic_read(&sbi->total_ext_tree) *
73 sizeof(struct extent_tree) +
74 atomic_read(&sbi->total_ext_node) *
75 sizeof(struct extent_node)) >> PAGE_SHIFT;
76 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
77 } else if (type == INMEM_PAGES) {
78 /* it allows 20% / total_ram for inmemory pages */
79 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
80 res = mem_size < (val.totalram / 5);
81 } else {
82 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
83 return true;
84 }
85 return res;
86 }
87
88 static void clear_node_page_dirty(struct page *page)
89 {
90 struct address_space *mapping = page->mapping;
91 unsigned int long flags;
92
93 if (PageDirty(page)) {
94 spin_lock_irqsave(&mapping->tree_lock, flags);
95 radix_tree_tag_clear(&mapping->page_tree,
96 page_index(page),
97 PAGECACHE_TAG_DIRTY);
98 spin_unlock_irqrestore(&mapping->tree_lock, flags);
99
100 clear_page_dirty_for_io(page);
101 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
102 }
103 ClearPageUptodate(page);
104 }
105
106 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
107 {
108 pgoff_t index = current_nat_addr(sbi, nid);
109 return get_meta_page(sbi, index);
110 }
111
112 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
113 {
114 struct page *src_page;
115 struct page *dst_page;
116 pgoff_t src_off;
117 pgoff_t dst_off;
118 void *src_addr;
119 void *dst_addr;
120 struct f2fs_nm_info *nm_i = NM_I(sbi);
121
122 src_off = current_nat_addr(sbi, nid);
123 dst_off = next_nat_addr(sbi, src_off);
124
125 /* get current nat block page with lock */
126 src_page = get_meta_page(sbi, src_off);
127 dst_page = grab_meta_page(sbi, dst_off);
128 f2fs_bug_on(sbi, PageDirty(src_page));
129
130 src_addr = page_address(src_page);
131 dst_addr = page_address(dst_page);
132 memcpy(dst_addr, src_addr, PAGE_SIZE);
133 set_page_dirty(dst_page);
134 f2fs_put_page(src_page, 1);
135
136 set_to_next_nat(nm_i, nid);
137
138 return dst_page;
139 }
140
141 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
142 {
143 return radix_tree_lookup(&nm_i->nat_root, n);
144 }
145
146 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
147 nid_t start, unsigned int nr, struct nat_entry **ep)
148 {
149 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
150 }
151
152 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
153 {
154 list_del(&e->list);
155 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
156 nm_i->nat_cnt--;
157 kmem_cache_free(nat_entry_slab, e);
158 }
159
160 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
161 struct nat_entry *ne)
162 {
163 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
164 struct nat_entry_set *head;
165
166 head = radix_tree_lookup(&nm_i->nat_set_root, set);
167 if (!head) {
168 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
169
170 INIT_LIST_HEAD(&head->entry_list);
171 INIT_LIST_HEAD(&head->set_list);
172 head->set = set;
173 head->entry_cnt = 0;
174 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
175 }
176
177 if (get_nat_flag(ne, IS_DIRTY))
178 goto refresh_list;
179
180 nm_i->dirty_nat_cnt++;
181 head->entry_cnt++;
182 set_nat_flag(ne, IS_DIRTY, true);
183 refresh_list:
184 if (nat_get_blkaddr(ne) == NEW_ADDR)
185 list_del_init(&ne->list);
186 else
187 list_move_tail(&ne->list, &head->entry_list);
188 }
189
190 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
191 struct nat_entry_set *set, struct nat_entry *ne)
192 {
193 list_move_tail(&ne->list, &nm_i->nat_entries);
194 set_nat_flag(ne, IS_DIRTY, false);
195 set->entry_cnt--;
196 nm_i->dirty_nat_cnt--;
197 }
198
199 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
200 nid_t start, unsigned int nr, struct nat_entry_set **ep)
201 {
202 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
203 start, nr);
204 }
205
206 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
207 {
208 struct f2fs_nm_info *nm_i = NM_I(sbi);
209 struct nat_entry *e;
210 bool need = false;
211
212 down_read(&nm_i->nat_tree_lock);
213 e = __lookup_nat_cache(nm_i, nid);
214 if (e) {
215 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
216 !get_nat_flag(e, HAS_FSYNCED_INODE))
217 need = true;
218 }
219 up_read(&nm_i->nat_tree_lock);
220 return need;
221 }
222
223 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
224 {
225 struct f2fs_nm_info *nm_i = NM_I(sbi);
226 struct nat_entry *e;
227 bool is_cp = true;
228
229 down_read(&nm_i->nat_tree_lock);
230 e = __lookup_nat_cache(nm_i, nid);
231 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
232 is_cp = false;
233 up_read(&nm_i->nat_tree_lock);
234 return is_cp;
235 }
236
237 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
238 {
239 struct f2fs_nm_info *nm_i = NM_I(sbi);
240 struct nat_entry *e;
241 bool need_update = true;
242
243 down_read(&nm_i->nat_tree_lock);
244 e = __lookup_nat_cache(nm_i, ino);
245 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
246 (get_nat_flag(e, IS_CHECKPOINTED) ||
247 get_nat_flag(e, HAS_FSYNCED_INODE)))
248 need_update = false;
249 up_read(&nm_i->nat_tree_lock);
250 return need_update;
251 }
252
253 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
254 bool no_fail)
255 {
256 struct nat_entry *new;
257
258 if (no_fail) {
259 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
260 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
261 } else {
262 new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
263 if (!new)
264 return NULL;
265 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
266 kmem_cache_free(nat_entry_slab, new);
267 return NULL;
268 }
269 }
270
271 memset(new, 0, sizeof(struct nat_entry));
272 nat_set_nid(new, nid);
273 nat_reset_flag(new);
274 list_add_tail(&new->list, &nm_i->nat_entries);
275 nm_i->nat_cnt++;
276 return new;
277 }
278
279 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
280 struct f2fs_nat_entry *ne)
281 {
282 struct f2fs_nm_info *nm_i = NM_I(sbi);
283 struct nat_entry *e;
284
285 e = __lookup_nat_cache(nm_i, nid);
286 if (!e) {
287 e = grab_nat_entry(nm_i, nid, false);
288 if (e)
289 node_info_from_raw_nat(&e->ni, ne);
290 } else {
291 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
292 nat_get_blkaddr(e) !=
293 le32_to_cpu(ne->block_addr) ||
294 nat_get_version(e) != ne->version);
295 }
296 }
297
298 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
299 block_t new_blkaddr, bool fsync_done)
300 {
301 struct f2fs_nm_info *nm_i = NM_I(sbi);
302 struct nat_entry *e;
303
304 down_write(&nm_i->nat_tree_lock);
305 e = __lookup_nat_cache(nm_i, ni->nid);
306 if (!e) {
307 e = grab_nat_entry(nm_i, ni->nid, true);
308 copy_node_info(&e->ni, ni);
309 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
310 } else if (new_blkaddr == NEW_ADDR) {
311 /*
312 * when nid is reallocated,
313 * previous nat entry can be remained in nat cache.
314 * So, reinitialize it with new information.
315 */
316 copy_node_info(&e->ni, ni);
317 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
318 }
319
320 /* sanity check */
321 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
322 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
323 new_blkaddr == NULL_ADDR);
324 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
325 new_blkaddr == NEW_ADDR);
326 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
327 nat_get_blkaddr(e) != NULL_ADDR &&
328 new_blkaddr == NEW_ADDR);
329
330 /* increment version no as node is removed */
331 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
332 unsigned char version = nat_get_version(e);
333 nat_set_version(e, inc_node_version(version));
334 }
335
336 /* change address */
337 nat_set_blkaddr(e, new_blkaddr);
338 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
339 set_nat_flag(e, IS_CHECKPOINTED, false);
340 __set_nat_cache_dirty(nm_i, e);
341
342 /* update fsync_mark if its inode nat entry is still alive */
343 if (ni->nid != ni->ino)
344 e = __lookup_nat_cache(nm_i, ni->ino);
345 if (e) {
346 if (fsync_done && ni->nid == ni->ino)
347 set_nat_flag(e, HAS_FSYNCED_INODE, true);
348 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
349 }
350 up_write(&nm_i->nat_tree_lock);
351 }
352
353 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
354 {
355 struct f2fs_nm_info *nm_i = NM_I(sbi);
356 int nr = nr_shrink;
357
358 if (!down_write_trylock(&nm_i->nat_tree_lock))
359 return 0;
360
361 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
362 struct nat_entry *ne;
363 ne = list_first_entry(&nm_i->nat_entries,
364 struct nat_entry, list);
365 __del_from_nat_cache(nm_i, ne);
366 nr_shrink--;
367 }
368 up_write(&nm_i->nat_tree_lock);
369 return nr - nr_shrink;
370 }
371
372 /*
373 * This function always returns success
374 */
375 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
376 {
377 struct f2fs_nm_info *nm_i = NM_I(sbi);
378 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
379 struct f2fs_journal *journal = curseg->journal;
380 nid_t start_nid = START_NID(nid);
381 struct f2fs_nat_block *nat_blk;
382 struct page *page = NULL;
383 struct f2fs_nat_entry ne;
384 struct nat_entry *e;
385 pgoff_t index;
386 int i;
387
388 ni->nid = nid;
389
390 /* Check nat cache */
391 down_read(&nm_i->nat_tree_lock);
392 e = __lookup_nat_cache(nm_i, nid);
393 if (e) {
394 ni->ino = nat_get_ino(e);
395 ni->blk_addr = nat_get_blkaddr(e);
396 ni->version = nat_get_version(e);
397 up_read(&nm_i->nat_tree_lock);
398 return;
399 }
400
401 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
402
403 /* Check current segment summary */
404 down_read(&curseg->journal_rwsem);
405 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
406 if (i >= 0) {
407 ne = nat_in_journal(journal, i);
408 node_info_from_raw_nat(ni, &ne);
409 }
410 up_read(&curseg->journal_rwsem);
411 if (i >= 0) {
412 up_read(&nm_i->nat_tree_lock);
413 goto cache;
414 }
415
416 /* Fill node_info from nat page */
417 index = current_nat_addr(sbi, nid);
418 up_read(&nm_i->nat_tree_lock);
419
420 page = get_meta_page(sbi, index);
421 nat_blk = (struct f2fs_nat_block *)page_address(page);
422 ne = nat_blk->entries[nid - start_nid];
423 node_info_from_raw_nat(ni, &ne);
424 f2fs_put_page(page, 1);
425 cache:
426 /* cache nat entry */
427 down_write(&nm_i->nat_tree_lock);
428 cache_nat_entry(sbi, nid, &ne);
429 up_write(&nm_i->nat_tree_lock);
430 }
431
432 /*
433 * readahead MAX_RA_NODE number of node pages.
434 */
435 static void ra_node_pages(struct page *parent, int start, int n)
436 {
437 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
438 struct blk_plug plug;
439 int i, end;
440 nid_t nid;
441
442 blk_start_plug(&plug);
443
444 /* Then, try readahead for siblings of the desired node */
445 end = start + n;
446 end = min(end, NIDS_PER_BLOCK);
447 for (i = start; i < end; i++) {
448 nid = get_nid(parent, i, false);
449 ra_node_page(sbi, nid);
450 }
451
452 blk_finish_plug(&plug);
453 }
454
455 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
456 {
457 const long direct_index = ADDRS_PER_INODE(dn->inode);
458 const long direct_blks = ADDRS_PER_BLOCK;
459 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
460 unsigned int skipped_unit = ADDRS_PER_BLOCK;
461 int cur_level = dn->cur_level;
462 int max_level = dn->max_level;
463 pgoff_t base = 0;
464
465 if (!dn->max_level)
466 return pgofs + 1;
467
468 while (max_level-- > cur_level)
469 skipped_unit *= NIDS_PER_BLOCK;
470
471 switch (dn->max_level) {
472 case 3:
473 base += 2 * indirect_blks;
474 case 2:
475 base += 2 * direct_blks;
476 case 1:
477 base += direct_index;
478 break;
479 default:
480 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
481 }
482
483 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
484 }
485
486 /*
487 * The maximum depth is four.
488 * Offset[0] will have raw inode offset.
489 */
490 static int get_node_path(struct inode *inode, long block,
491 int offset[4], unsigned int noffset[4])
492 {
493 const long direct_index = ADDRS_PER_INODE(inode);
494 const long direct_blks = ADDRS_PER_BLOCK;
495 const long dptrs_per_blk = NIDS_PER_BLOCK;
496 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
497 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
498 int n = 0;
499 int level = 0;
500
501 noffset[0] = 0;
502
503 if (block < direct_index) {
504 offset[n] = block;
505 goto got;
506 }
507 block -= direct_index;
508 if (block < direct_blks) {
509 offset[n++] = NODE_DIR1_BLOCK;
510 noffset[n] = 1;
511 offset[n] = block;
512 level = 1;
513 goto got;
514 }
515 block -= direct_blks;
516 if (block < direct_blks) {
517 offset[n++] = NODE_DIR2_BLOCK;
518 noffset[n] = 2;
519 offset[n] = block;
520 level = 1;
521 goto got;
522 }
523 block -= direct_blks;
524 if (block < indirect_blks) {
525 offset[n++] = NODE_IND1_BLOCK;
526 noffset[n] = 3;
527 offset[n++] = block / direct_blks;
528 noffset[n] = 4 + offset[n - 1];
529 offset[n] = block % direct_blks;
530 level = 2;
531 goto got;
532 }
533 block -= indirect_blks;
534 if (block < indirect_blks) {
535 offset[n++] = NODE_IND2_BLOCK;
536 noffset[n] = 4 + dptrs_per_blk;
537 offset[n++] = block / direct_blks;
538 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
539 offset[n] = block % direct_blks;
540 level = 2;
541 goto got;
542 }
543 block -= indirect_blks;
544 if (block < dindirect_blks) {
545 offset[n++] = NODE_DIND_BLOCK;
546 noffset[n] = 5 + (dptrs_per_blk * 2);
547 offset[n++] = block / indirect_blks;
548 noffset[n] = 6 + (dptrs_per_blk * 2) +
549 offset[n - 1] * (dptrs_per_blk + 1);
550 offset[n++] = (block / direct_blks) % dptrs_per_blk;
551 noffset[n] = 7 + (dptrs_per_blk * 2) +
552 offset[n - 2] * (dptrs_per_blk + 1) +
553 offset[n - 1];
554 offset[n] = block % direct_blks;
555 level = 3;
556 goto got;
557 } else {
558 return -E2BIG;
559 }
560 got:
561 return level;
562 }
563
564 /*
565 * Caller should call f2fs_put_dnode(dn).
566 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
567 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
568 * In the case of RDONLY_NODE, we don't need to care about mutex.
569 */
570 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
571 {
572 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
573 struct page *npage[4];
574 struct page *parent = NULL;
575 int offset[4];
576 unsigned int noffset[4];
577 nid_t nids[4];
578 int level, i = 0;
579 int err = 0;
580
581 level = get_node_path(dn->inode, index, offset, noffset);
582 if (level < 0)
583 return level;
584
585 nids[0] = dn->inode->i_ino;
586 npage[0] = dn->inode_page;
587
588 if (!npage[0]) {
589 npage[0] = get_node_page(sbi, nids[0]);
590 if (IS_ERR(npage[0]))
591 return PTR_ERR(npage[0]);
592 }
593
594 /* if inline_data is set, should not report any block indices */
595 if (f2fs_has_inline_data(dn->inode) && index) {
596 err = -ENOENT;
597 f2fs_put_page(npage[0], 1);
598 goto release_out;
599 }
600
601 parent = npage[0];
602 if (level != 0)
603 nids[1] = get_nid(parent, offset[0], true);
604 dn->inode_page = npage[0];
605 dn->inode_page_locked = true;
606
607 /* get indirect or direct nodes */
608 for (i = 1; i <= level; i++) {
609 bool done = false;
610
611 if (!nids[i] && mode == ALLOC_NODE) {
612 /* alloc new node */
613 if (!alloc_nid(sbi, &(nids[i]))) {
614 err = -ENOSPC;
615 goto release_pages;
616 }
617
618 dn->nid = nids[i];
619 npage[i] = new_node_page(dn, noffset[i]);
620 if (IS_ERR(npage[i])) {
621 alloc_nid_failed(sbi, nids[i]);
622 err = PTR_ERR(npage[i]);
623 goto release_pages;
624 }
625
626 set_nid(parent, offset[i - 1], nids[i], i == 1);
627 alloc_nid_done(sbi, nids[i]);
628 done = true;
629 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
630 npage[i] = get_node_page_ra(parent, offset[i - 1]);
631 if (IS_ERR(npage[i])) {
632 err = PTR_ERR(npage[i]);
633 goto release_pages;
634 }
635 done = true;
636 }
637 if (i == 1) {
638 dn->inode_page_locked = false;
639 unlock_page(parent);
640 } else {
641 f2fs_put_page(parent, 1);
642 }
643
644 if (!done) {
645 npage[i] = get_node_page(sbi, nids[i]);
646 if (IS_ERR(npage[i])) {
647 err = PTR_ERR(npage[i]);
648 f2fs_put_page(npage[0], 0);
649 goto release_out;
650 }
651 }
652 if (i < level) {
653 parent = npage[i];
654 nids[i + 1] = get_nid(parent, offset[i], false);
655 }
656 }
657 dn->nid = nids[level];
658 dn->ofs_in_node = offset[level];
659 dn->node_page = npage[level];
660 dn->data_blkaddr = datablock_addr(dn->inode,
661 dn->node_page, dn->ofs_in_node);
662 return 0;
663
664 release_pages:
665 f2fs_put_page(parent, 1);
666 if (i > 1)
667 f2fs_put_page(npage[0], 0);
668 release_out:
669 dn->inode_page = NULL;
670 dn->node_page = NULL;
671 if (err == -ENOENT) {
672 dn->cur_level = i;
673 dn->max_level = level;
674 dn->ofs_in_node = offset[level];
675 }
676 return err;
677 }
678
679 static void truncate_node(struct dnode_of_data *dn)
680 {
681 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
682 struct node_info ni;
683
684 get_node_info(sbi, dn->nid, &ni);
685 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
686
687 /* Deallocate node address */
688 invalidate_blocks(sbi, ni.blk_addr);
689 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
690 set_node_addr(sbi, &ni, NULL_ADDR, false);
691
692 if (dn->nid == dn->inode->i_ino) {
693 remove_orphan_inode(sbi, dn->nid);
694 dec_valid_inode_count(sbi);
695 f2fs_inode_synced(dn->inode);
696 }
697
698 clear_node_page_dirty(dn->node_page);
699 set_sbi_flag(sbi, SBI_IS_DIRTY);
700
701 f2fs_put_page(dn->node_page, 1);
702
703 invalidate_mapping_pages(NODE_MAPPING(sbi),
704 dn->node_page->index, dn->node_page->index);
705
706 dn->node_page = NULL;
707 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
708 }
709
710 static int truncate_dnode(struct dnode_of_data *dn)
711 {
712 struct page *page;
713
714 if (dn->nid == 0)
715 return 1;
716
717 /* get direct node */
718 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
719 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
720 return 1;
721 else if (IS_ERR(page))
722 return PTR_ERR(page);
723
724 /* Make dnode_of_data for parameter */
725 dn->node_page = page;
726 dn->ofs_in_node = 0;
727 truncate_data_blocks(dn);
728 truncate_node(dn);
729 return 1;
730 }
731
732 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
733 int ofs, int depth)
734 {
735 struct dnode_of_data rdn = *dn;
736 struct page *page;
737 struct f2fs_node *rn;
738 nid_t child_nid;
739 unsigned int child_nofs;
740 int freed = 0;
741 int i, ret;
742
743 if (dn->nid == 0)
744 return NIDS_PER_BLOCK + 1;
745
746 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
747
748 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
749 if (IS_ERR(page)) {
750 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
751 return PTR_ERR(page);
752 }
753
754 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
755
756 rn = F2FS_NODE(page);
757 if (depth < 3) {
758 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
759 child_nid = le32_to_cpu(rn->in.nid[i]);
760 if (child_nid == 0)
761 continue;
762 rdn.nid = child_nid;
763 ret = truncate_dnode(&rdn);
764 if (ret < 0)
765 goto out_err;
766 if (set_nid(page, i, 0, false))
767 dn->node_changed = true;
768 }
769 } else {
770 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
771 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
772 child_nid = le32_to_cpu(rn->in.nid[i]);
773 if (child_nid == 0) {
774 child_nofs += NIDS_PER_BLOCK + 1;
775 continue;
776 }
777 rdn.nid = child_nid;
778 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
779 if (ret == (NIDS_PER_BLOCK + 1)) {
780 if (set_nid(page, i, 0, false))
781 dn->node_changed = true;
782 child_nofs += ret;
783 } else if (ret < 0 && ret != -ENOENT) {
784 goto out_err;
785 }
786 }
787 freed = child_nofs;
788 }
789
790 if (!ofs) {
791 /* remove current indirect node */
792 dn->node_page = page;
793 truncate_node(dn);
794 freed++;
795 } else {
796 f2fs_put_page(page, 1);
797 }
798 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
799 return freed;
800
801 out_err:
802 f2fs_put_page(page, 1);
803 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
804 return ret;
805 }
806
807 static int truncate_partial_nodes(struct dnode_of_data *dn,
808 struct f2fs_inode *ri, int *offset, int depth)
809 {
810 struct page *pages[2];
811 nid_t nid[3];
812 nid_t child_nid;
813 int err = 0;
814 int i;
815 int idx = depth - 2;
816
817 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
818 if (!nid[0])
819 return 0;
820
821 /* get indirect nodes in the path */
822 for (i = 0; i < idx + 1; i++) {
823 /* reference count'll be increased */
824 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
825 if (IS_ERR(pages[i])) {
826 err = PTR_ERR(pages[i]);
827 idx = i - 1;
828 goto fail;
829 }
830 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
831 }
832
833 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
834
835 /* free direct nodes linked to a partial indirect node */
836 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
837 child_nid = get_nid(pages[idx], i, false);
838 if (!child_nid)
839 continue;
840 dn->nid = child_nid;
841 err = truncate_dnode(dn);
842 if (err < 0)
843 goto fail;
844 if (set_nid(pages[idx], i, 0, false))
845 dn->node_changed = true;
846 }
847
848 if (offset[idx + 1] == 0) {
849 dn->node_page = pages[idx];
850 dn->nid = nid[idx];
851 truncate_node(dn);
852 } else {
853 f2fs_put_page(pages[idx], 1);
854 }
855 offset[idx]++;
856 offset[idx + 1] = 0;
857 idx--;
858 fail:
859 for (i = idx; i >= 0; i--)
860 f2fs_put_page(pages[i], 1);
861
862 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
863
864 return err;
865 }
866
867 /*
868 * All the block addresses of data and nodes should be nullified.
869 */
870 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
871 {
872 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
873 int err = 0, cont = 1;
874 int level, offset[4], noffset[4];
875 unsigned int nofs = 0;
876 struct f2fs_inode *ri;
877 struct dnode_of_data dn;
878 struct page *page;
879
880 trace_f2fs_truncate_inode_blocks_enter(inode, from);
881
882 level = get_node_path(inode, from, offset, noffset);
883 if (level < 0)
884 return level;
885
886 page = get_node_page(sbi, inode->i_ino);
887 if (IS_ERR(page)) {
888 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
889 return PTR_ERR(page);
890 }
891
892 set_new_dnode(&dn, inode, page, NULL, 0);
893 unlock_page(page);
894
895 ri = F2FS_INODE(page);
896 switch (level) {
897 case 0:
898 case 1:
899 nofs = noffset[1];
900 break;
901 case 2:
902 nofs = noffset[1];
903 if (!offset[level - 1])
904 goto skip_partial;
905 err = truncate_partial_nodes(&dn, ri, offset, level);
906 if (err < 0 && err != -ENOENT)
907 goto fail;
908 nofs += 1 + NIDS_PER_BLOCK;
909 break;
910 case 3:
911 nofs = 5 + 2 * NIDS_PER_BLOCK;
912 if (!offset[level - 1])
913 goto skip_partial;
914 err = truncate_partial_nodes(&dn, ri, offset, level);
915 if (err < 0 && err != -ENOENT)
916 goto fail;
917 break;
918 default:
919 BUG();
920 }
921
922 skip_partial:
923 while (cont) {
924 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
925 switch (offset[0]) {
926 case NODE_DIR1_BLOCK:
927 case NODE_DIR2_BLOCK:
928 err = truncate_dnode(&dn);
929 break;
930
931 case NODE_IND1_BLOCK:
932 case NODE_IND2_BLOCK:
933 err = truncate_nodes(&dn, nofs, offset[1], 2);
934 break;
935
936 case NODE_DIND_BLOCK:
937 err = truncate_nodes(&dn, nofs, offset[1], 3);
938 cont = 0;
939 break;
940
941 default:
942 BUG();
943 }
944 if (err < 0 && err != -ENOENT)
945 goto fail;
946 if (offset[1] == 0 &&
947 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
948 lock_page(page);
949 BUG_ON(page->mapping != NODE_MAPPING(sbi));
950 f2fs_wait_on_page_writeback(page, NODE, true);
951 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
952 set_page_dirty(page);
953 unlock_page(page);
954 }
955 offset[1] = 0;
956 offset[0]++;
957 nofs += err;
958 }
959 fail:
960 f2fs_put_page(page, 0);
961 trace_f2fs_truncate_inode_blocks_exit(inode, err);
962 return err > 0 ? 0 : err;
963 }
964
965 int truncate_xattr_node(struct inode *inode, struct page *page)
966 {
967 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
968 nid_t nid = F2FS_I(inode)->i_xattr_nid;
969 struct dnode_of_data dn;
970 struct page *npage;
971
972 if (!nid)
973 return 0;
974
975 npage = get_node_page(sbi, nid);
976 if (IS_ERR(npage))
977 return PTR_ERR(npage);
978
979 f2fs_i_xnid_write(inode, 0);
980
981 set_new_dnode(&dn, inode, page, npage, nid);
982
983 if (page)
984 dn.inode_page_locked = true;
985 truncate_node(&dn);
986 return 0;
987 }
988
989 /*
990 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
991 * f2fs_unlock_op().
992 */
993 int remove_inode_page(struct inode *inode)
994 {
995 struct dnode_of_data dn;
996 int err;
997
998 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
999 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1000 if (err)
1001 return err;
1002
1003 err = truncate_xattr_node(inode, dn.inode_page);
1004 if (err) {
1005 f2fs_put_dnode(&dn);
1006 return err;
1007 }
1008
1009 /* remove potential inline_data blocks */
1010 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1011 S_ISLNK(inode->i_mode))
1012 truncate_data_blocks_range(&dn, 1);
1013
1014 /* 0 is possible, after f2fs_new_inode() has failed */
1015 f2fs_bug_on(F2FS_I_SB(inode),
1016 inode->i_blocks != 0 && inode->i_blocks != 8);
1017
1018 /* will put inode & node pages */
1019 truncate_node(&dn);
1020 return 0;
1021 }
1022
1023 struct page *new_inode_page(struct inode *inode)
1024 {
1025 struct dnode_of_data dn;
1026
1027 /* allocate inode page for new inode */
1028 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1029
1030 /* caller should f2fs_put_page(page, 1); */
1031 return new_node_page(&dn, 0);
1032 }
1033
1034 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1035 {
1036 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1037 struct node_info new_ni;
1038 struct page *page;
1039 int err;
1040
1041 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1042 return ERR_PTR(-EPERM);
1043
1044 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1045 if (!page)
1046 return ERR_PTR(-ENOMEM);
1047
1048 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1049 goto fail;
1050
1051 #ifdef CONFIG_F2FS_CHECK_FS
1052 get_node_info(sbi, dn->nid, &new_ni);
1053 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1054 #endif
1055 new_ni.nid = dn->nid;
1056 new_ni.ino = dn->inode->i_ino;
1057 new_ni.blk_addr = NULL_ADDR;
1058 new_ni.flag = 0;
1059 new_ni.version = 0;
1060 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1061
1062 f2fs_wait_on_page_writeback(page, NODE, true);
1063 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1064 set_cold_node(dn->inode, page);
1065 if (!PageUptodate(page))
1066 SetPageUptodate(page);
1067 if (set_page_dirty(page))
1068 dn->node_changed = true;
1069
1070 if (f2fs_has_xattr_block(ofs))
1071 f2fs_i_xnid_write(dn->inode, dn->nid);
1072
1073 if (ofs == 0)
1074 inc_valid_inode_count(sbi);
1075 return page;
1076
1077 fail:
1078 clear_node_page_dirty(page);
1079 f2fs_put_page(page, 1);
1080 return ERR_PTR(err);
1081 }
1082
1083 /*
1084 * Caller should do after getting the following values.
1085 * 0: f2fs_put_page(page, 0)
1086 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1087 */
1088 static int read_node_page(struct page *page, int op_flags)
1089 {
1090 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1091 struct node_info ni;
1092 struct f2fs_io_info fio = {
1093 .sbi = sbi,
1094 .type = NODE,
1095 .op = REQ_OP_READ,
1096 .op_flags = op_flags,
1097 .page = page,
1098 .encrypted_page = NULL,
1099 };
1100
1101 if (PageUptodate(page))
1102 return LOCKED_PAGE;
1103
1104 get_node_info(sbi, page->index, &ni);
1105
1106 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1107 ClearPageUptodate(page);
1108 return -ENOENT;
1109 }
1110
1111 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1112 return f2fs_submit_page_bio(&fio);
1113 }
1114
1115 /*
1116 * Readahead a node page
1117 */
1118 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1119 {
1120 struct page *apage;
1121 int err;
1122
1123 if (!nid)
1124 return;
1125 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1126
1127 rcu_read_lock();
1128 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1129 rcu_read_unlock();
1130 if (apage)
1131 return;
1132
1133 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1134 if (!apage)
1135 return;
1136
1137 err = read_node_page(apage, REQ_RAHEAD);
1138 f2fs_put_page(apage, err ? 1 : 0);
1139 }
1140
1141 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1142 struct page *parent, int start)
1143 {
1144 struct page *page;
1145 int err;
1146
1147 if (!nid)
1148 return ERR_PTR(-ENOENT);
1149 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1150 repeat:
1151 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1152 if (!page)
1153 return ERR_PTR(-ENOMEM);
1154
1155 err = read_node_page(page, 0);
1156 if (err < 0) {
1157 f2fs_put_page(page, 1);
1158 return ERR_PTR(err);
1159 } else if (err == LOCKED_PAGE) {
1160 err = 0;
1161 goto page_hit;
1162 }
1163
1164 if (parent)
1165 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1166
1167 lock_page(page);
1168
1169 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1170 f2fs_put_page(page, 1);
1171 goto repeat;
1172 }
1173
1174 if (unlikely(!PageUptodate(page))) {
1175 err = -EIO;
1176 goto out_err;
1177 }
1178
1179 if (!f2fs_inode_chksum_verify(sbi, page)) {
1180 err = -EBADMSG;
1181 goto out_err;
1182 }
1183 page_hit:
1184 if(unlikely(nid != nid_of_node(page))) {
1185 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1186 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1187 nid, nid_of_node(page), ino_of_node(page),
1188 ofs_of_node(page), cpver_of_node(page),
1189 next_blkaddr_of_node(page));
1190 err = -EINVAL;
1191 out_err:
1192 ClearPageUptodate(page);
1193 f2fs_put_page(page, 1);
1194 return ERR_PTR(err);
1195 }
1196 return page;
1197 }
1198
1199 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1200 {
1201 return __get_node_page(sbi, nid, NULL, 0);
1202 }
1203
1204 struct page *get_node_page_ra(struct page *parent, int start)
1205 {
1206 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1207 nid_t nid = get_nid(parent, start, false);
1208
1209 return __get_node_page(sbi, nid, parent, start);
1210 }
1211
1212 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1213 {
1214 struct inode *inode;
1215 struct page *page;
1216 int ret;
1217
1218 /* should flush inline_data before evict_inode */
1219 inode = ilookup(sbi->sb, ino);
1220 if (!inode)
1221 return;
1222
1223 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1224 if (!page)
1225 goto iput_out;
1226
1227 if (!PageUptodate(page))
1228 goto page_out;
1229
1230 if (!PageDirty(page))
1231 goto page_out;
1232
1233 if (!clear_page_dirty_for_io(page))
1234 goto page_out;
1235
1236 ret = f2fs_write_inline_data(inode, page);
1237 inode_dec_dirty_pages(inode);
1238 remove_dirty_inode(inode);
1239 if (ret)
1240 set_page_dirty(page);
1241 page_out:
1242 f2fs_put_page(page, 1);
1243 iput_out:
1244 iput(inode);
1245 }
1246
1247 void move_node_page(struct page *node_page, int gc_type)
1248 {
1249 if (gc_type == FG_GC) {
1250 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1251 struct writeback_control wbc = {
1252 .sync_mode = WB_SYNC_ALL,
1253 .nr_to_write = 1,
1254 .for_reclaim = 0,
1255 };
1256
1257 set_page_dirty(node_page);
1258 f2fs_wait_on_page_writeback(node_page, NODE, true);
1259
1260 f2fs_bug_on(sbi, PageWriteback(node_page));
1261 if (!clear_page_dirty_for_io(node_page))
1262 goto out_page;
1263
1264 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1265 unlock_page(node_page);
1266 goto release_page;
1267 } else {
1268 /* set page dirty and write it */
1269 if (!PageWriteback(node_page))
1270 set_page_dirty(node_page);
1271 }
1272 out_page:
1273 unlock_page(node_page);
1274 release_page:
1275 f2fs_put_page(node_page, 0);
1276 }
1277
1278 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1279 {
1280 pgoff_t index, end;
1281 struct pagevec pvec;
1282 struct page *last_page = NULL;
1283
1284 pagevec_init(&pvec, 0);
1285 index = 0;
1286 end = ULONG_MAX;
1287
1288 while (index <= end) {
1289 int i, nr_pages;
1290 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1291 PAGECACHE_TAG_DIRTY,
1292 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1293 if (nr_pages == 0)
1294 break;
1295
1296 for (i = 0; i < nr_pages; i++) {
1297 struct page *page = pvec.pages[i];
1298
1299 if (unlikely(f2fs_cp_error(sbi))) {
1300 f2fs_put_page(last_page, 0);
1301 pagevec_release(&pvec);
1302 return ERR_PTR(-EIO);
1303 }
1304
1305 if (!IS_DNODE(page) || !is_cold_node(page))
1306 continue;
1307 if (ino_of_node(page) != ino)
1308 continue;
1309
1310 lock_page(page);
1311
1312 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1313 continue_unlock:
1314 unlock_page(page);
1315 continue;
1316 }
1317 if (ino_of_node(page) != ino)
1318 goto continue_unlock;
1319
1320 if (!PageDirty(page)) {
1321 /* someone wrote it for us */
1322 goto continue_unlock;
1323 }
1324
1325 if (last_page)
1326 f2fs_put_page(last_page, 0);
1327
1328 get_page(page);
1329 last_page = page;
1330 unlock_page(page);
1331 }
1332 pagevec_release(&pvec);
1333 cond_resched();
1334 }
1335 return last_page;
1336 }
1337
1338 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1339 struct writeback_control *wbc, bool do_balance,
1340 enum iostat_type io_type)
1341 {
1342 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1343 nid_t nid;
1344 struct node_info ni;
1345 struct f2fs_io_info fio = {
1346 .sbi = sbi,
1347 .ino = ino_of_node(page),
1348 .type = NODE,
1349 .op = REQ_OP_WRITE,
1350 .op_flags = wbc_to_write_flags(wbc),
1351 .page = page,
1352 .encrypted_page = NULL,
1353 .submitted = false,
1354 .io_type = io_type,
1355 };
1356
1357 trace_f2fs_writepage(page, NODE);
1358
1359 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1360 goto redirty_out;
1361 if (unlikely(f2fs_cp_error(sbi)))
1362 goto redirty_out;
1363
1364 /* get old block addr of this node page */
1365 nid = nid_of_node(page);
1366 f2fs_bug_on(sbi, page->index != nid);
1367
1368 if (wbc->for_reclaim) {
1369 if (!down_read_trylock(&sbi->node_write))
1370 goto redirty_out;
1371 } else {
1372 down_read(&sbi->node_write);
1373 }
1374
1375 get_node_info(sbi, nid, &ni);
1376
1377 /* This page is already truncated */
1378 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1379 ClearPageUptodate(page);
1380 dec_page_count(sbi, F2FS_DIRTY_NODES);
1381 up_read(&sbi->node_write);
1382 unlock_page(page);
1383 return 0;
1384 }
1385
1386 if (atomic && !test_opt(sbi, NOBARRIER))
1387 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1388
1389 set_page_writeback(page);
1390 fio.old_blkaddr = ni.blk_addr;
1391 write_node_page(nid, &fio);
1392 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1393 dec_page_count(sbi, F2FS_DIRTY_NODES);
1394 up_read(&sbi->node_write);
1395
1396 if (wbc->for_reclaim) {
1397 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1398 page->index, NODE);
1399 submitted = NULL;
1400 }
1401
1402 unlock_page(page);
1403
1404 if (unlikely(f2fs_cp_error(sbi))) {
1405 f2fs_submit_merged_write(sbi, NODE);
1406 submitted = NULL;
1407 }
1408 if (submitted)
1409 *submitted = fio.submitted;
1410
1411 if (do_balance)
1412 f2fs_balance_fs(sbi, false);
1413 return 0;
1414
1415 redirty_out:
1416 redirty_page_for_writepage(wbc, page);
1417 return AOP_WRITEPAGE_ACTIVATE;
1418 }
1419
1420 static int f2fs_write_node_page(struct page *page,
1421 struct writeback_control *wbc)
1422 {
1423 return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1424 }
1425
1426 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1427 struct writeback_control *wbc, bool atomic)
1428 {
1429 pgoff_t index, end;
1430 pgoff_t last_idx = ULONG_MAX;
1431 struct pagevec pvec;
1432 int ret = 0;
1433 struct page *last_page = NULL;
1434 bool marked = false;
1435 nid_t ino = inode->i_ino;
1436
1437 if (atomic) {
1438 last_page = last_fsync_dnode(sbi, ino);
1439 if (IS_ERR_OR_NULL(last_page))
1440 return PTR_ERR_OR_ZERO(last_page);
1441 }
1442 retry:
1443 pagevec_init(&pvec, 0);
1444 index = 0;
1445 end = ULONG_MAX;
1446
1447 while (index <= end) {
1448 int i, nr_pages;
1449 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1450 PAGECACHE_TAG_DIRTY,
1451 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1452 if (nr_pages == 0)
1453 break;
1454
1455 for (i = 0; i < nr_pages; i++) {
1456 struct page *page = pvec.pages[i];
1457 bool submitted = false;
1458
1459 if (unlikely(f2fs_cp_error(sbi))) {
1460 f2fs_put_page(last_page, 0);
1461 pagevec_release(&pvec);
1462 ret = -EIO;
1463 goto out;
1464 }
1465
1466 if (!IS_DNODE(page) || !is_cold_node(page))
1467 continue;
1468 if (ino_of_node(page) != ino)
1469 continue;
1470
1471 lock_page(page);
1472
1473 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1474 continue_unlock:
1475 unlock_page(page);
1476 continue;
1477 }
1478 if (ino_of_node(page) != ino)
1479 goto continue_unlock;
1480
1481 if (!PageDirty(page) && page != last_page) {
1482 /* someone wrote it for us */
1483 goto continue_unlock;
1484 }
1485
1486 f2fs_wait_on_page_writeback(page, NODE, true);
1487 BUG_ON(PageWriteback(page));
1488
1489 set_fsync_mark(page, 0);
1490 set_dentry_mark(page, 0);
1491
1492 if (!atomic || page == last_page) {
1493 set_fsync_mark(page, 1);
1494 if (IS_INODE(page)) {
1495 if (is_inode_flag_set(inode,
1496 FI_DIRTY_INODE))
1497 update_inode(inode, page);
1498 set_dentry_mark(page,
1499 need_dentry_mark(sbi, ino));
1500 }
1501 /* may be written by other thread */
1502 if (!PageDirty(page))
1503 set_page_dirty(page);
1504 }
1505
1506 if (!clear_page_dirty_for_io(page))
1507 goto continue_unlock;
1508
1509 ret = __write_node_page(page, atomic &&
1510 page == last_page,
1511 &submitted, wbc, true,
1512 FS_NODE_IO);
1513 if (ret) {
1514 unlock_page(page);
1515 f2fs_put_page(last_page, 0);
1516 break;
1517 } else if (submitted) {
1518 last_idx = page->index;
1519 }
1520
1521 if (page == last_page) {
1522 f2fs_put_page(page, 0);
1523 marked = true;
1524 break;
1525 }
1526 }
1527 pagevec_release(&pvec);
1528 cond_resched();
1529
1530 if (ret || marked)
1531 break;
1532 }
1533 if (!ret && atomic && !marked) {
1534 f2fs_msg(sbi->sb, KERN_DEBUG,
1535 "Retry to write fsync mark: ino=%u, idx=%lx",
1536 ino, last_page->index);
1537 lock_page(last_page);
1538 f2fs_wait_on_page_writeback(last_page, NODE, true);
1539 set_page_dirty(last_page);
1540 unlock_page(last_page);
1541 goto retry;
1542 }
1543 out:
1544 if (last_idx != ULONG_MAX)
1545 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1546 return ret ? -EIO: 0;
1547 }
1548
1549 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1550 bool do_balance, enum iostat_type io_type)
1551 {
1552 pgoff_t index, end;
1553 struct pagevec pvec;
1554 int step = 0;
1555 int nwritten = 0;
1556 int ret = 0;
1557
1558 pagevec_init(&pvec, 0);
1559
1560 next_step:
1561 index = 0;
1562 end = ULONG_MAX;
1563
1564 while (index <= end) {
1565 int i, nr_pages;
1566 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1567 PAGECACHE_TAG_DIRTY,
1568 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1569 if (nr_pages == 0)
1570 break;
1571
1572 for (i = 0; i < nr_pages; i++) {
1573 struct page *page = pvec.pages[i];
1574 bool submitted = false;
1575
1576 if (unlikely(f2fs_cp_error(sbi))) {
1577 pagevec_release(&pvec);
1578 ret = -EIO;
1579 goto out;
1580 }
1581
1582 /*
1583 * flushing sequence with step:
1584 * 0. indirect nodes
1585 * 1. dentry dnodes
1586 * 2. file dnodes
1587 */
1588 if (step == 0 && IS_DNODE(page))
1589 continue;
1590 if (step == 1 && (!IS_DNODE(page) ||
1591 is_cold_node(page)))
1592 continue;
1593 if (step == 2 && (!IS_DNODE(page) ||
1594 !is_cold_node(page)))
1595 continue;
1596 lock_node:
1597 if (!trylock_page(page))
1598 continue;
1599
1600 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1601 continue_unlock:
1602 unlock_page(page);
1603 continue;
1604 }
1605
1606 if (!PageDirty(page)) {
1607 /* someone wrote it for us */
1608 goto continue_unlock;
1609 }
1610
1611 /* flush inline_data */
1612 if (is_inline_node(page)) {
1613 clear_inline_node(page);
1614 unlock_page(page);
1615 flush_inline_data(sbi, ino_of_node(page));
1616 goto lock_node;
1617 }
1618
1619 f2fs_wait_on_page_writeback(page, NODE, true);
1620
1621 BUG_ON(PageWriteback(page));
1622 if (!clear_page_dirty_for_io(page))
1623 goto continue_unlock;
1624
1625 set_fsync_mark(page, 0);
1626 set_dentry_mark(page, 0);
1627
1628 ret = __write_node_page(page, false, &submitted,
1629 wbc, do_balance, io_type);
1630 if (ret)
1631 unlock_page(page);
1632 else if (submitted)
1633 nwritten++;
1634
1635 if (--wbc->nr_to_write == 0)
1636 break;
1637 }
1638 pagevec_release(&pvec);
1639 cond_resched();
1640
1641 if (wbc->nr_to_write == 0) {
1642 step = 2;
1643 break;
1644 }
1645 }
1646
1647 if (step < 2) {
1648 step++;
1649 goto next_step;
1650 }
1651 out:
1652 if (nwritten)
1653 f2fs_submit_merged_write(sbi, NODE);
1654 return ret;
1655 }
1656
1657 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1658 {
1659 pgoff_t index = 0, end = ULONG_MAX;
1660 struct pagevec pvec;
1661 int ret2, ret = 0;
1662
1663 pagevec_init(&pvec, 0);
1664
1665 while (index <= end) {
1666 int i, nr_pages;
1667 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1668 PAGECACHE_TAG_WRITEBACK,
1669 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1670 if (nr_pages == 0)
1671 break;
1672
1673 for (i = 0; i < nr_pages; i++) {
1674 struct page *page = pvec.pages[i];
1675
1676 /* until radix tree lookup accepts end_index */
1677 if (unlikely(page->index > end))
1678 continue;
1679
1680 if (ino && ino_of_node(page) == ino) {
1681 f2fs_wait_on_page_writeback(page, NODE, true);
1682 if (TestClearPageError(page))
1683 ret = -EIO;
1684 }
1685 }
1686 pagevec_release(&pvec);
1687 cond_resched();
1688 }
1689
1690 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1691 if (!ret)
1692 ret = ret2;
1693 return ret;
1694 }
1695
1696 static int f2fs_write_node_pages(struct address_space *mapping,
1697 struct writeback_control *wbc)
1698 {
1699 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1700 struct blk_plug plug;
1701 long diff;
1702
1703 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1704 goto skip_write;
1705
1706 /* balancing f2fs's metadata in background */
1707 f2fs_balance_fs_bg(sbi);
1708
1709 /* collect a number of dirty node pages and write together */
1710 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1711 goto skip_write;
1712
1713 trace_f2fs_writepages(mapping->host, wbc, NODE);
1714
1715 diff = nr_pages_to_write(sbi, NODE, wbc);
1716 wbc->sync_mode = WB_SYNC_NONE;
1717 blk_start_plug(&plug);
1718 sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1719 blk_finish_plug(&plug);
1720 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1721 return 0;
1722
1723 skip_write:
1724 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1725 trace_f2fs_writepages(mapping->host, wbc, NODE);
1726 return 0;
1727 }
1728
1729 static int f2fs_set_node_page_dirty(struct page *page)
1730 {
1731 trace_f2fs_set_page_dirty(page, NODE);
1732
1733 if (!PageUptodate(page))
1734 SetPageUptodate(page);
1735 if (!PageDirty(page)) {
1736 f2fs_set_page_dirty_nobuffers(page);
1737 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1738 SetPagePrivate(page);
1739 f2fs_trace_pid(page);
1740 return 1;
1741 }
1742 return 0;
1743 }
1744
1745 /*
1746 * Structure of the f2fs node operations
1747 */
1748 const struct address_space_operations f2fs_node_aops = {
1749 .writepage = f2fs_write_node_page,
1750 .writepages = f2fs_write_node_pages,
1751 .set_page_dirty = f2fs_set_node_page_dirty,
1752 .invalidatepage = f2fs_invalidate_page,
1753 .releasepage = f2fs_release_page,
1754 #ifdef CONFIG_MIGRATION
1755 .migratepage = f2fs_migrate_page,
1756 #endif
1757 };
1758
1759 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1760 nid_t n)
1761 {
1762 return radix_tree_lookup(&nm_i->free_nid_root, n);
1763 }
1764
1765 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1766 struct free_nid *i, enum nid_state state, bool new)
1767 {
1768 struct f2fs_nm_info *nm_i = NM_I(sbi);
1769
1770 if (new) {
1771 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1772 if (err)
1773 return err;
1774 }
1775
1776 f2fs_bug_on(sbi, state != i->state);
1777 nm_i->nid_cnt[state]++;
1778 if (state == FREE_NID)
1779 list_add_tail(&i->list, &nm_i->free_nid_list);
1780 return 0;
1781 }
1782
1783 static void __remove_free_nid(struct f2fs_sb_info *sbi,
1784 struct free_nid *i, enum nid_state state, bool reuse)
1785 {
1786 struct f2fs_nm_info *nm_i = NM_I(sbi);
1787
1788 f2fs_bug_on(sbi, state != i->state);
1789 nm_i->nid_cnt[state]--;
1790 if (state == FREE_NID)
1791 list_del(&i->list);
1792 if (!reuse)
1793 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1794 }
1795
1796 /* return if the nid is recognized as free */
1797 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1798 {
1799 struct f2fs_nm_info *nm_i = NM_I(sbi);
1800 struct free_nid *i, *e;
1801 struct nat_entry *ne;
1802 int err = -EINVAL;
1803 bool ret = false;
1804
1805 /* 0 nid should not be used */
1806 if (unlikely(nid == 0))
1807 return false;
1808
1809 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1810 i->nid = nid;
1811 i->state = FREE_NID;
1812
1813 if (radix_tree_preload(GFP_NOFS))
1814 goto err;
1815
1816 spin_lock(&nm_i->nid_list_lock);
1817
1818 if (build) {
1819 /*
1820 * Thread A Thread B
1821 * - f2fs_create
1822 * - f2fs_new_inode
1823 * - alloc_nid
1824 * - __insert_nid_to_list(PREALLOC_NID)
1825 * - f2fs_balance_fs_bg
1826 * - build_free_nids
1827 * - __build_free_nids
1828 * - scan_nat_page
1829 * - add_free_nid
1830 * - __lookup_nat_cache
1831 * - f2fs_add_link
1832 * - init_inode_metadata
1833 * - new_inode_page
1834 * - new_node_page
1835 * - set_node_addr
1836 * - alloc_nid_done
1837 * - __remove_nid_from_list(PREALLOC_NID)
1838 * - __insert_nid_to_list(FREE_NID)
1839 */
1840 ne = __lookup_nat_cache(nm_i, nid);
1841 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1842 nat_get_blkaddr(ne) != NULL_ADDR))
1843 goto err_out;
1844
1845 e = __lookup_free_nid_list(nm_i, nid);
1846 if (e) {
1847 if (e->state == FREE_NID)
1848 ret = true;
1849 goto err_out;
1850 }
1851 }
1852 ret = true;
1853 err = __insert_free_nid(sbi, i, FREE_NID, true);
1854 err_out:
1855 spin_unlock(&nm_i->nid_list_lock);
1856 radix_tree_preload_end();
1857 err:
1858 if (err)
1859 kmem_cache_free(free_nid_slab, i);
1860 return ret;
1861 }
1862
1863 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1864 {
1865 struct f2fs_nm_info *nm_i = NM_I(sbi);
1866 struct free_nid *i;
1867 bool need_free = false;
1868
1869 spin_lock(&nm_i->nid_list_lock);
1870 i = __lookup_free_nid_list(nm_i, nid);
1871 if (i && i->state == FREE_NID) {
1872 __remove_free_nid(sbi, i, FREE_NID, false);
1873 need_free = true;
1874 }
1875 spin_unlock(&nm_i->nid_list_lock);
1876
1877 if (need_free)
1878 kmem_cache_free(free_nid_slab, i);
1879 }
1880
1881 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1882 bool set, bool build)
1883 {
1884 struct f2fs_nm_info *nm_i = NM_I(sbi);
1885 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1886 unsigned int nid_ofs = nid - START_NID(nid);
1887
1888 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1889 return;
1890
1891 if (set)
1892 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1893 else
1894 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1895
1896 if (set)
1897 nm_i->free_nid_count[nat_ofs]++;
1898 else if (!build)
1899 nm_i->free_nid_count[nat_ofs]--;
1900 }
1901
1902 static void scan_nat_page(struct f2fs_sb_info *sbi,
1903 struct page *nat_page, nid_t start_nid)
1904 {
1905 struct f2fs_nm_info *nm_i = NM_I(sbi);
1906 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1907 block_t blk_addr;
1908 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1909 int i;
1910
1911 if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1912 return;
1913
1914 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1915
1916 i = start_nid % NAT_ENTRY_PER_BLOCK;
1917
1918 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1919 bool freed = false;
1920
1921 if (unlikely(start_nid >= nm_i->max_nid))
1922 break;
1923
1924 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1925 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1926 if (blk_addr == NULL_ADDR)
1927 freed = add_free_nid(sbi, start_nid, true);
1928 spin_lock(&NM_I(sbi)->nid_list_lock);
1929 update_free_nid_bitmap(sbi, start_nid, freed, true);
1930 spin_unlock(&NM_I(sbi)->nid_list_lock);
1931 }
1932 }
1933
1934 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1935 {
1936 struct f2fs_nm_info *nm_i = NM_I(sbi);
1937 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1938 struct f2fs_journal *journal = curseg->journal;
1939 unsigned int i, idx;
1940
1941 down_read(&nm_i->nat_tree_lock);
1942
1943 for (i = 0; i < nm_i->nat_blocks; i++) {
1944 if (!test_bit_le(i, nm_i->nat_block_bitmap))
1945 continue;
1946 if (!nm_i->free_nid_count[i])
1947 continue;
1948 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1949 nid_t nid;
1950
1951 if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1952 continue;
1953
1954 nid = i * NAT_ENTRY_PER_BLOCK + idx;
1955 add_free_nid(sbi, nid, true);
1956
1957 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
1958 goto out;
1959 }
1960 }
1961 out:
1962 down_read(&curseg->journal_rwsem);
1963 for (i = 0; i < nats_in_cursum(journal); i++) {
1964 block_t addr;
1965 nid_t nid;
1966
1967 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1968 nid = le32_to_cpu(nid_in_journal(journal, i));
1969 if (addr == NULL_ADDR)
1970 add_free_nid(sbi, nid, true);
1971 else
1972 remove_free_nid(sbi, nid);
1973 }
1974 up_read(&curseg->journal_rwsem);
1975 up_read(&nm_i->nat_tree_lock);
1976 }
1977
1978 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1979 {
1980 struct f2fs_nm_info *nm_i = NM_I(sbi);
1981 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1982 struct f2fs_journal *journal = curseg->journal;
1983 int i = 0;
1984 nid_t nid = nm_i->next_scan_nid;
1985
1986 if (unlikely(nid >= nm_i->max_nid))
1987 nid = 0;
1988
1989 /* Enough entries */
1990 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
1991 return;
1992
1993 if (!sync && !available_free_memory(sbi, FREE_NIDS))
1994 return;
1995
1996 if (!mount) {
1997 /* try to find free nids in free_nid_bitmap */
1998 scan_free_nid_bits(sbi);
1999
2000 if (nm_i->nid_cnt[FREE_NID])
2001 return;
2002 }
2003
2004 /* readahead nat pages to be scanned */
2005 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2006 META_NAT, true);
2007
2008 down_read(&nm_i->nat_tree_lock);
2009
2010 while (1) {
2011 struct page *page = get_current_nat_page(sbi, nid);
2012
2013 scan_nat_page(sbi, page, nid);
2014 f2fs_put_page(page, 1);
2015
2016 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2017 if (unlikely(nid >= nm_i->max_nid))
2018 nid = 0;
2019
2020 if (++i >= FREE_NID_PAGES)
2021 break;
2022 }
2023
2024 /* go to the next free nat pages to find free nids abundantly */
2025 nm_i->next_scan_nid = nid;
2026
2027 /* find free nids from current sum_pages */
2028 down_read(&curseg->journal_rwsem);
2029 for (i = 0; i < nats_in_cursum(journal); i++) {
2030 block_t addr;
2031
2032 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2033 nid = le32_to_cpu(nid_in_journal(journal, i));
2034 if (addr == NULL_ADDR)
2035 add_free_nid(sbi, nid, true);
2036 else
2037 remove_free_nid(sbi, nid);
2038 }
2039 up_read(&curseg->journal_rwsem);
2040 up_read(&nm_i->nat_tree_lock);
2041
2042 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2043 nm_i->ra_nid_pages, META_NAT, false);
2044 }
2045
2046 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2047 {
2048 mutex_lock(&NM_I(sbi)->build_lock);
2049 __build_free_nids(sbi, sync, mount);
2050 mutex_unlock(&NM_I(sbi)->build_lock);
2051 }
2052
2053 /*
2054 * If this function returns success, caller can obtain a new nid
2055 * from second parameter of this function.
2056 * The returned nid could be used ino as well as nid when inode is created.
2057 */
2058 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2059 {
2060 struct f2fs_nm_info *nm_i = NM_I(sbi);
2061 struct free_nid *i = NULL;
2062 retry:
2063 #ifdef CONFIG_F2FS_FAULT_INJECTION
2064 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2065 f2fs_show_injection_info(FAULT_ALLOC_NID);
2066 return false;
2067 }
2068 #endif
2069 spin_lock(&nm_i->nid_list_lock);
2070
2071 if (unlikely(nm_i->available_nids == 0)) {
2072 spin_unlock(&nm_i->nid_list_lock);
2073 return false;
2074 }
2075
2076 /* We should not use stale free nids created by build_free_nids */
2077 if (nm_i->nid_cnt[FREE_NID] && !on_build_free_nids(nm_i)) {
2078 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2079 i = list_first_entry(&nm_i->free_nid_list,
2080 struct free_nid, list);
2081 *nid = i->nid;
2082
2083 __remove_free_nid(sbi, i, FREE_NID, true);
2084 i->state = PREALLOC_NID;
2085 __insert_free_nid(sbi, i, PREALLOC_NID, false);
2086 nm_i->available_nids--;
2087
2088 update_free_nid_bitmap(sbi, *nid, false, false);
2089
2090 spin_unlock(&nm_i->nid_list_lock);
2091 return true;
2092 }
2093 spin_unlock(&nm_i->nid_list_lock);
2094
2095 /* Let's scan nat pages and its caches to get free nids */
2096 build_free_nids(sbi, true, false);
2097 goto retry;
2098 }
2099
2100 /*
2101 * alloc_nid() should be called prior to this function.
2102 */
2103 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2104 {
2105 struct f2fs_nm_info *nm_i = NM_I(sbi);
2106 struct free_nid *i;
2107
2108 spin_lock(&nm_i->nid_list_lock);
2109 i = __lookup_free_nid_list(nm_i, nid);
2110 f2fs_bug_on(sbi, !i);
2111 __remove_free_nid(sbi, i, PREALLOC_NID, false);
2112 spin_unlock(&nm_i->nid_list_lock);
2113
2114 kmem_cache_free(free_nid_slab, i);
2115 }
2116
2117 /*
2118 * alloc_nid() should be called prior to this function.
2119 */
2120 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2121 {
2122 struct f2fs_nm_info *nm_i = NM_I(sbi);
2123 struct free_nid *i;
2124 bool need_free = false;
2125
2126 if (!nid)
2127 return;
2128
2129 spin_lock(&nm_i->nid_list_lock);
2130 i = __lookup_free_nid_list(nm_i, nid);
2131 f2fs_bug_on(sbi, !i);
2132
2133 if (!available_free_memory(sbi, FREE_NIDS)) {
2134 __remove_free_nid(sbi, i, PREALLOC_NID, false);
2135 need_free = true;
2136 } else {
2137 __remove_free_nid(sbi, i, PREALLOC_NID, true);
2138 i->state = FREE_NID;
2139 __insert_free_nid(sbi, i, FREE_NID, false);
2140 }
2141
2142 nm_i->available_nids++;
2143
2144 update_free_nid_bitmap(sbi, nid, true, false);
2145
2146 spin_unlock(&nm_i->nid_list_lock);
2147
2148 if (need_free)
2149 kmem_cache_free(free_nid_slab, i);
2150 }
2151
2152 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2153 {
2154 struct f2fs_nm_info *nm_i = NM_I(sbi);
2155 struct free_nid *i, *next;
2156 int nr = nr_shrink;
2157
2158 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2159 return 0;
2160
2161 if (!mutex_trylock(&nm_i->build_lock))
2162 return 0;
2163
2164 spin_lock(&nm_i->nid_list_lock);
2165 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2166 if (nr_shrink <= 0 ||
2167 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2168 break;
2169
2170 __remove_free_nid(sbi, i, FREE_NID, false);
2171 kmem_cache_free(free_nid_slab, i);
2172 nr_shrink--;
2173 }
2174 spin_unlock(&nm_i->nid_list_lock);
2175 mutex_unlock(&nm_i->build_lock);
2176
2177 return nr - nr_shrink;
2178 }
2179
2180 void recover_inline_xattr(struct inode *inode, struct page *page)
2181 {
2182 void *src_addr, *dst_addr;
2183 size_t inline_size;
2184 struct page *ipage;
2185 struct f2fs_inode *ri;
2186
2187 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2188 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2189
2190 ri = F2FS_INODE(page);
2191 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2192 clear_inode_flag(inode, FI_INLINE_XATTR);
2193 goto update_inode;
2194 }
2195
2196 dst_addr = inline_xattr_addr(ipage);
2197 src_addr = inline_xattr_addr(page);
2198 inline_size = inline_xattr_size(inode);
2199
2200 f2fs_wait_on_page_writeback(ipage, NODE, true);
2201 memcpy(dst_addr, src_addr, inline_size);
2202 update_inode:
2203 update_inode(inode, ipage);
2204 f2fs_put_page(ipage, 1);
2205 }
2206
2207 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2208 {
2209 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2210 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2211 nid_t new_xnid;
2212 struct dnode_of_data dn;
2213 struct node_info ni;
2214 struct page *xpage;
2215
2216 if (!prev_xnid)
2217 goto recover_xnid;
2218
2219 /* 1: invalidate the previous xattr nid */
2220 get_node_info(sbi, prev_xnid, &ni);
2221 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2222 invalidate_blocks(sbi, ni.blk_addr);
2223 dec_valid_node_count(sbi, inode, false);
2224 set_node_addr(sbi, &ni, NULL_ADDR, false);
2225
2226 recover_xnid:
2227 /* 2: update xattr nid in inode */
2228 if (!alloc_nid(sbi, &new_xnid))
2229 return -ENOSPC;
2230
2231 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2232 xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2233 if (IS_ERR(xpage)) {
2234 alloc_nid_failed(sbi, new_xnid);
2235 return PTR_ERR(xpage);
2236 }
2237
2238 alloc_nid_done(sbi, new_xnid);
2239 update_inode_page(inode);
2240
2241 /* 3: update and set xattr node page dirty */
2242 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2243
2244 set_page_dirty(xpage);
2245 f2fs_put_page(xpage, 1);
2246
2247 return 0;
2248 }
2249
2250 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2251 {
2252 struct f2fs_inode *src, *dst;
2253 nid_t ino = ino_of_node(page);
2254 struct node_info old_ni, new_ni;
2255 struct page *ipage;
2256
2257 get_node_info(sbi, ino, &old_ni);
2258
2259 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2260 return -EINVAL;
2261 retry:
2262 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2263 if (!ipage) {
2264 congestion_wait(BLK_RW_ASYNC, HZ/50);
2265 goto retry;
2266 }
2267
2268 /* Should not use this inode from free nid list */
2269 remove_free_nid(sbi, ino);
2270
2271 if (!PageUptodate(ipage))
2272 SetPageUptodate(ipage);
2273 fill_node_footer(ipage, ino, ino, 0, true);
2274
2275 src = F2FS_INODE(page);
2276 dst = F2FS_INODE(ipage);
2277
2278 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2279 dst->i_size = 0;
2280 dst->i_blocks = cpu_to_le64(1);
2281 dst->i_links = cpu_to_le32(1);
2282 dst->i_xattr_nid = 0;
2283 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2284 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2285 dst->i_extra_isize = src->i_extra_isize;
2286 if (f2fs_sb_has_project_quota(sbi->sb) &&
2287 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2288 i_projid))
2289 dst->i_projid = src->i_projid;
2290 }
2291
2292 new_ni = old_ni;
2293 new_ni.ino = ino;
2294
2295 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2296 WARN_ON(1);
2297 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2298 inc_valid_inode_count(sbi);
2299 set_page_dirty(ipage);
2300 f2fs_put_page(ipage, 1);
2301 return 0;
2302 }
2303
2304 int restore_node_summary(struct f2fs_sb_info *sbi,
2305 unsigned int segno, struct f2fs_summary_block *sum)
2306 {
2307 struct f2fs_node *rn;
2308 struct f2fs_summary *sum_entry;
2309 block_t addr;
2310 int i, idx, last_offset, nrpages;
2311
2312 /* scan the node segment */
2313 last_offset = sbi->blocks_per_seg;
2314 addr = START_BLOCK(sbi, segno);
2315 sum_entry = &sum->entries[0];
2316
2317 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2318 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2319
2320 /* readahead node pages */
2321 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2322
2323 for (idx = addr; idx < addr + nrpages; idx++) {
2324 struct page *page = get_tmp_page(sbi, idx);
2325
2326 rn = F2FS_NODE(page);
2327 sum_entry->nid = rn->footer.nid;
2328 sum_entry->version = 0;
2329 sum_entry->ofs_in_node = 0;
2330 sum_entry++;
2331 f2fs_put_page(page, 1);
2332 }
2333
2334 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2335 addr + nrpages);
2336 }
2337 return 0;
2338 }
2339
2340 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2341 {
2342 struct f2fs_nm_info *nm_i = NM_I(sbi);
2343 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2344 struct f2fs_journal *journal = curseg->journal;
2345 int i;
2346
2347 down_write(&curseg->journal_rwsem);
2348 for (i = 0; i < nats_in_cursum(journal); i++) {
2349 struct nat_entry *ne;
2350 struct f2fs_nat_entry raw_ne;
2351 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2352
2353 raw_ne = nat_in_journal(journal, i);
2354
2355 ne = __lookup_nat_cache(nm_i, nid);
2356 if (!ne) {
2357 ne = grab_nat_entry(nm_i, nid, true);
2358 node_info_from_raw_nat(&ne->ni, &raw_ne);
2359 }
2360
2361 /*
2362 * if a free nat in journal has not been used after last
2363 * checkpoint, we should remove it from available nids,
2364 * since later we will add it again.
2365 */
2366 if (!get_nat_flag(ne, IS_DIRTY) &&
2367 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2368 spin_lock(&nm_i->nid_list_lock);
2369 nm_i->available_nids--;
2370 spin_unlock(&nm_i->nid_list_lock);
2371 }
2372
2373 __set_nat_cache_dirty(nm_i, ne);
2374 }
2375 update_nats_in_cursum(journal, -i);
2376 up_write(&curseg->journal_rwsem);
2377 }
2378
2379 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2380 struct list_head *head, int max)
2381 {
2382 struct nat_entry_set *cur;
2383
2384 if (nes->entry_cnt >= max)
2385 goto add_out;
2386
2387 list_for_each_entry(cur, head, set_list) {
2388 if (cur->entry_cnt >= nes->entry_cnt) {
2389 list_add(&nes->set_list, cur->set_list.prev);
2390 return;
2391 }
2392 }
2393 add_out:
2394 list_add_tail(&nes->set_list, head);
2395 }
2396
2397 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2398 struct page *page)
2399 {
2400 struct f2fs_nm_info *nm_i = NM_I(sbi);
2401 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2402 struct f2fs_nat_block *nat_blk = page_address(page);
2403 int valid = 0;
2404 int i;
2405
2406 if (!enabled_nat_bits(sbi, NULL))
2407 return;
2408
2409 for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2410 if (start_nid == 0 && i == 0)
2411 valid++;
2412 if (nat_blk->entries[i].block_addr)
2413 valid++;
2414 }
2415 if (valid == 0) {
2416 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2417 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2418 return;
2419 }
2420
2421 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2422 if (valid == NAT_ENTRY_PER_BLOCK)
2423 __set_bit_le(nat_index, nm_i->full_nat_bits);
2424 else
2425 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2426 }
2427
2428 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2429 struct nat_entry_set *set, struct cp_control *cpc)
2430 {
2431 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2432 struct f2fs_journal *journal = curseg->journal;
2433 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2434 bool to_journal = true;
2435 struct f2fs_nat_block *nat_blk;
2436 struct nat_entry *ne, *cur;
2437 struct page *page = NULL;
2438
2439 /*
2440 * there are two steps to flush nat entries:
2441 * #1, flush nat entries to journal in current hot data summary block.
2442 * #2, flush nat entries to nat page.
2443 */
2444 if (enabled_nat_bits(sbi, cpc) ||
2445 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2446 to_journal = false;
2447
2448 if (to_journal) {
2449 down_write(&curseg->journal_rwsem);
2450 } else {
2451 page = get_next_nat_page(sbi, start_nid);
2452 nat_blk = page_address(page);
2453 f2fs_bug_on(sbi, !nat_blk);
2454 }
2455
2456 /* flush dirty nats in nat entry set */
2457 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2458 struct f2fs_nat_entry *raw_ne;
2459 nid_t nid = nat_get_nid(ne);
2460 int offset;
2461
2462 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2463
2464 if (to_journal) {
2465 offset = lookup_journal_in_cursum(journal,
2466 NAT_JOURNAL, nid, 1);
2467 f2fs_bug_on(sbi, offset < 0);
2468 raw_ne = &nat_in_journal(journal, offset);
2469 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2470 } else {
2471 raw_ne = &nat_blk->entries[nid - start_nid];
2472 }
2473 raw_nat_from_node_info(raw_ne, &ne->ni);
2474 nat_reset_flag(ne);
2475 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2476 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2477 add_free_nid(sbi, nid, false);
2478 spin_lock(&NM_I(sbi)->nid_list_lock);
2479 NM_I(sbi)->available_nids++;
2480 update_free_nid_bitmap(sbi, nid, true, false);
2481 spin_unlock(&NM_I(sbi)->nid_list_lock);
2482 } else {
2483 spin_lock(&NM_I(sbi)->nid_list_lock);
2484 update_free_nid_bitmap(sbi, nid, false, false);
2485 spin_unlock(&NM_I(sbi)->nid_list_lock);
2486 }
2487 }
2488
2489 if (to_journal) {
2490 up_write(&curseg->journal_rwsem);
2491 } else {
2492 __update_nat_bits(sbi, start_nid, page);
2493 f2fs_put_page(page, 1);
2494 }
2495
2496 /* Allow dirty nats by node block allocation in write_begin */
2497 if (!set->entry_cnt) {
2498 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2499 kmem_cache_free(nat_entry_set_slab, set);
2500 }
2501 }
2502
2503 /*
2504 * This function is called during the checkpointing process.
2505 */
2506 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2507 {
2508 struct f2fs_nm_info *nm_i = NM_I(sbi);
2509 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2510 struct f2fs_journal *journal = curseg->journal;
2511 struct nat_entry_set *setvec[SETVEC_SIZE];
2512 struct nat_entry_set *set, *tmp;
2513 unsigned int found;
2514 nid_t set_idx = 0;
2515 LIST_HEAD(sets);
2516
2517 if (!nm_i->dirty_nat_cnt)
2518 return;
2519
2520 down_write(&nm_i->nat_tree_lock);
2521
2522 /*
2523 * if there are no enough space in journal to store dirty nat
2524 * entries, remove all entries from journal and merge them
2525 * into nat entry set.
2526 */
2527 if (enabled_nat_bits(sbi, cpc) ||
2528 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2529 remove_nats_in_journal(sbi);
2530
2531 while ((found = __gang_lookup_nat_set(nm_i,
2532 set_idx, SETVEC_SIZE, setvec))) {
2533 unsigned idx;
2534 set_idx = setvec[found - 1]->set + 1;
2535 for (idx = 0; idx < found; idx++)
2536 __adjust_nat_entry_set(setvec[idx], &sets,
2537 MAX_NAT_JENTRIES(journal));
2538 }
2539
2540 /* flush dirty nats in nat entry set */
2541 list_for_each_entry_safe(set, tmp, &sets, set_list)
2542 __flush_nat_entry_set(sbi, set, cpc);
2543
2544 up_write(&nm_i->nat_tree_lock);
2545 /* Allow dirty nats by node block allocation in write_begin */
2546 }
2547
2548 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2549 {
2550 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2551 struct f2fs_nm_info *nm_i = NM_I(sbi);
2552 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2553 unsigned int i;
2554 __u64 cp_ver = cur_cp_version(ckpt);
2555 block_t nat_bits_addr;
2556
2557 if (!enabled_nat_bits(sbi, NULL))
2558 return 0;
2559
2560 nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2561 F2FS_BLKSIZE - 1);
2562 nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2563 GFP_KERNEL);
2564 if (!nm_i->nat_bits)
2565 return -ENOMEM;
2566
2567 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2568 nm_i->nat_bits_blocks;
2569 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2570 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2571
2572 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2573 page_address(page), F2FS_BLKSIZE);
2574 f2fs_put_page(page, 1);
2575 }
2576
2577 cp_ver |= (cur_cp_crc(ckpt) << 32);
2578 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2579 disable_nat_bits(sbi, true);
2580 return 0;
2581 }
2582
2583 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2584 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2585
2586 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2587 return 0;
2588 }
2589
2590 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2591 {
2592 struct f2fs_nm_info *nm_i = NM_I(sbi);
2593 unsigned int i = 0;
2594 nid_t nid, last_nid;
2595
2596 if (!enabled_nat_bits(sbi, NULL))
2597 return;
2598
2599 for (i = 0; i < nm_i->nat_blocks; i++) {
2600 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2601 if (i >= nm_i->nat_blocks)
2602 break;
2603
2604 __set_bit_le(i, nm_i->nat_block_bitmap);
2605
2606 nid = i * NAT_ENTRY_PER_BLOCK;
2607 last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2608
2609 spin_lock(&NM_I(sbi)->nid_list_lock);
2610 for (; nid < last_nid; nid++)
2611 update_free_nid_bitmap(sbi, nid, true, true);
2612 spin_unlock(&NM_I(sbi)->nid_list_lock);
2613 }
2614
2615 for (i = 0; i < nm_i->nat_blocks; i++) {
2616 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2617 if (i >= nm_i->nat_blocks)
2618 break;
2619
2620 __set_bit_le(i, nm_i->nat_block_bitmap);
2621 }
2622 }
2623
2624 static int init_node_manager(struct f2fs_sb_info *sbi)
2625 {
2626 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2627 struct f2fs_nm_info *nm_i = NM_I(sbi);
2628 unsigned char *version_bitmap;
2629 unsigned int nat_segs;
2630 int err;
2631
2632 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2633
2634 /* segment_count_nat includes pair segment so divide to 2. */
2635 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2636 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2637 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2638
2639 /* not used nids: 0, node, meta, (and root counted as valid node) */
2640 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2641 F2FS_RESERVED_NODE_NUM;
2642 nm_i->nid_cnt[FREE_NID] = 0;
2643 nm_i->nid_cnt[PREALLOC_NID] = 0;
2644 nm_i->nat_cnt = 0;
2645 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2646 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2647 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2648
2649 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2650 INIT_LIST_HEAD(&nm_i->free_nid_list);
2651 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2652 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2653 INIT_LIST_HEAD(&nm_i->nat_entries);
2654
2655 mutex_init(&nm_i->build_lock);
2656 spin_lock_init(&nm_i->nid_list_lock);
2657 init_rwsem(&nm_i->nat_tree_lock);
2658
2659 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2660 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2661 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2662 if (!version_bitmap)
2663 return -EFAULT;
2664
2665 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2666 GFP_KERNEL);
2667 if (!nm_i->nat_bitmap)
2668 return -ENOMEM;
2669
2670 err = __get_nat_bitmaps(sbi);
2671 if (err)
2672 return err;
2673
2674 #ifdef CONFIG_F2FS_CHECK_FS
2675 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2676 GFP_KERNEL);
2677 if (!nm_i->nat_bitmap_mir)
2678 return -ENOMEM;
2679 #endif
2680
2681 return 0;
2682 }
2683
2684 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2685 {
2686 struct f2fs_nm_info *nm_i = NM_I(sbi);
2687
2688 nm_i->free_nid_bitmap = kvzalloc(nm_i->nat_blocks *
2689 NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2690 if (!nm_i->free_nid_bitmap)
2691 return -ENOMEM;
2692
2693 nm_i->nat_block_bitmap = kvzalloc(nm_i->nat_blocks / 8,
2694 GFP_KERNEL);
2695 if (!nm_i->nat_block_bitmap)
2696 return -ENOMEM;
2697
2698 nm_i->free_nid_count = kvzalloc(nm_i->nat_blocks *
2699 sizeof(unsigned short), GFP_KERNEL);
2700 if (!nm_i->free_nid_count)
2701 return -ENOMEM;
2702 return 0;
2703 }
2704
2705 int build_node_manager(struct f2fs_sb_info *sbi)
2706 {
2707 int err;
2708
2709 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2710 if (!sbi->nm_info)
2711 return -ENOMEM;
2712
2713 err = init_node_manager(sbi);
2714 if (err)
2715 return err;
2716
2717 err = init_free_nid_cache(sbi);
2718 if (err)
2719 return err;
2720
2721 /* load free nid status from nat_bits table */
2722 load_free_nid_bitmap(sbi);
2723
2724 build_free_nids(sbi, true, true);
2725 return 0;
2726 }
2727
2728 void destroy_node_manager(struct f2fs_sb_info *sbi)
2729 {
2730 struct f2fs_nm_info *nm_i = NM_I(sbi);
2731 struct free_nid *i, *next_i;
2732 struct nat_entry *natvec[NATVEC_SIZE];
2733 struct nat_entry_set *setvec[SETVEC_SIZE];
2734 nid_t nid = 0;
2735 unsigned int found;
2736
2737 if (!nm_i)
2738 return;
2739
2740 /* destroy free nid list */
2741 spin_lock(&nm_i->nid_list_lock);
2742 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2743 __remove_free_nid(sbi, i, FREE_NID, false);
2744 spin_unlock(&nm_i->nid_list_lock);
2745 kmem_cache_free(free_nid_slab, i);
2746 spin_lock(&nm_i->nid_list_lock);
2747 }
2748 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2749 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2750 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2751 spin_unlock(&nm_i->nid_list_lock);
2752
2753 /* destroy nat cache */
2754 down_write(&nm_i->nat_tree_lock);
2755 while ((found = __gang_lookup_nat_cache(nm_i,
2756 nid, NATVEC_SIZE, natvec))) {
2757 unsigned idx;
2758
2759 nid = nat_get_nid(natvec[found - 1]) + 1;
2760 for (idx = 0; idx < found; idx++)
2761 __del_from_nat_cache(nm_i, natvec[idx]);
2762 }
2763 f2fs_bug_on(sbi, nm_i->nat_cnt);
2764
2765 /* destroy nat set cache */
2766 nid = 0;
2767 while ((found = __gang_lookup_nat_set(nm_i,
2768 nid, SETVEC_SIZE, setvec))) {
2769 unsigned idx;
2770
2771 nid = setvec[found - 1]->set + 1;
2772 for (idx = 0; idx < found; idx++) {
2773 /* entry_cnt is not zero, when cp_error was occurred */
2774 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2775 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2776 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2777 }
2778 }
2779 up_write(&nm_i->nat_tree_lock);
2780
2781 kvfree(nm_i->nat_block_bitmap);
2782 kvfree(nm_i->free_nid_bitmap);
2783 kvfree(nm_i->free_nid_count);
2784
2785 kfree(nm_i->nat_bitmap);
2786 kfree(nm_i->nat_bits);
2787 #ifdef CONFIG_F2FS_CHECK_FS
2788 kfree(nm_i->nat_bitmap_mir);
2789 #endif
2790 sbi->nm_info = NULL;
2791 kfree(nm_i);
2792 }
2793
2794 int __init create_node_manager_caches(void)
2795 {
2796 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2797 sizeof(struct nat_entry));
2798 if (!nat_entry_slab)
2799 goto fail;
2800
2801 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2802 sizeof(struct free_nid));
2803 if (!free_nid_slab)
2804 goto destroy_nat_entry;
2805
2806 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2807 sizeof(struct nat_entry_set));
2808 if (!nat_entry_set_slab)
2809 goto destroy_free_nid;
2810 return 0;
2811
2812 destroy_free_nid:
2813 kmem_cache_destroy(free_nid_slab);
2814 destroy_nat_entry:
2815 kmem_cache_destroy(nat_entry_slab);
2816 fail:
2817 return -ENOMEM;
2818 }
2819
2820 void destroy_node_manager_caches(void)
2821 {
2822 kmem_cache_destroy(nat_entry_set_slab);
2823 kmem_cache_destroy(free_nid_slab);
2824 kmem_cache_destroy(nat_entry_slab);
2825 }
ubuntu-bionic-kernel mirror