projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: add core functions for rb-tree extent cache
[GitHub/exynos8895/android_kernel_samsung_universal8895.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 "trace.h"
23 #include <trace/events/f2fs.h>
24
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
30
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32 {
33 struct f2fs_nm_info *nm_i = NM_I(sbi);
34 struct sysinfo val;
35 unsigned long avail_ram;
36 unsigned long mem_size = 0;
37 bool res = false;
38
39 si_meminfo(&val);
40
41 /* only uses low memory */
42 avail_ram = val.totalram - val.totalhigh;
43
44 /*
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
46 */
47 if (type == FREE_NIDS) {
48 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
49 PAGE_CACHE_SHIFT;
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 PAGE_CACHE_SHIFT;
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 } else if (type == DIRTY_DENTS) {
56 if (sbi->sb->s_bdi->dirty_exceeded)
57 return false;
58 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
59 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
60 } else if (type == INO_ENTRIES) {
61 int i;
62
63 for (i = 0; i <= UPDATE_INO; i++)
64 mem_size += (sbi->im[i].ino_num *
65 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
67 } else if (type == EXTENT_CACHE) {
68 mem_size = (sbi->total_ext_tree * sizeof(struct extent_tree) +
69 atomic_read(&sbi->total_ext_node) *
70 sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
71 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
72 } else {
73 if (sbi->sb->s_bdi->dirty_exceeded)
74 return false;
75 }
76 return res;
77 }
78
79 static void clear_node_page_dirty(struct page *page)
80 {
81 struct address_space *mapping = page->mapping;
82 unsigned int long flags;
83
84 if (PageDirty(page)) {
85 spin_lock_irqsave(&mapping->tree_lock, flags);
86 radix_tree_tag_clear(&mapping->page_tree,
87 page_index(page),
88 PAGECACHE_TAG_DIRTY);
89 spin_unlock_irqrestore(&mapping->tree_lock, flags);
90
91 clear_page_dirty_for_io(page);
92 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
93 }
94 ClearPageUptodate(page);
95 }
96
97 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
98 {
99 pgoff_t index = current_nat_addr(sbi, nid);
100 return get_meta_page(sbi, index);
101 }
102
103 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
104 {
105 struct page *src_page;
106 struct page *dst_page;
107 pgoff_t src_off;
108 pgoff_t dst_off;
109 void *src_addr;
110 void *dst_addr;
111 struct f2fs_nm_info *nm_i = NM_I(sbi);
112
113 src_off = current_nat_addr(sbi, nid);
114 dst_off = next_nat_addr(sbi, src_off);
115
116 /* get current nat block page with lock */
117 src_page = get_meta_page(sbi, src_off);
118 dst_page = grab_meta_page(sbi, dst_off);
119 f2fs_bug_on(sbi, PageDirty(src_page));
120
121 src_addr = page_address(src_page);
122 dst_addr = page_address(dst_page);
123 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
124 set_page_dirty(dst_page);
125 f2fs_put_page(src_page, 1);
126
127 set_to_next_nat(nm_i, nid);
128
129 return dst_page;
130 }
131
132 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
133 {
134 return radix_tree_lookup(&nm_i->nat_root, n);
135 }
136
137 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
138 nid_t start, unsigned int nr, struct nat_entry **ep)
139 {
140 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
141 }
142
143 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
144 {
145 list_del(&e->list);
146 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
147 nm_i->nat_cnt--;
148 kmem_cache_free(nat_entry_slab, e);
149 }
150
151 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
152 struct nat_entry *ne)
153 {
154 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
155 struct nat_entry_set *head;
156
157 if (get_nat_flag(ne, IS_DIRTY))
158 return;
159
160 head = radix_tree_lookup(&nm_i->nat_set_root, set);
161 if (!head) {
162 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
163
164 INIT_LIST_HEAD(&head->entry_list);
165 INIT_LIST_HEAD(&head->set_list);
166 head->set = set;
167 head->entry_cnt = 0;
168 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
169 }
170 list_move_tail(&ne->list, &head->entry_list);
171 nm_i->dirty_nat_cnt++;
172 head->entry_cnt++;
173 set_nat_flag(ne, IS_DIRTY, true);
174 }
175
176 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
177 struct nat_entry *ne)
178 {
179 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
180 struct nat_entry_set *head;
181
182 head = radix_tree_lookup(&nm_i->nat_set_root, set);
183 if (head) {
184 list_move_tail(&ne->list, &nm_i->nat_entries);
185 set_nat_flag(ne, IS_DIRTY, false);
186 head->entry_cnt--;
187 nm_i->dirty_nat_cnt--;
188 }
189 }
190
191 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
192 nid_t start, unsigned int nr, struct nat_entry_set **ep)
193 {
194 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
195 start, nr);
196 }
197
198 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
199 {
200 struct f2fs_nm_info *nm_i = NM_I(sbi);
201 struct nat_entry *e;
202 bool is_cp = true;
203
204 down_read(&nm_i->nat_tree_lock);
205 e = __lookup_nat_cache(nm_i, nid);
206 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
207 is_cp = false;
208 up_read(&nm_i->nat_tree_lock);
209 return is_cp;
210 }
211
212 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
213 {
214 struct f2fs_nm_info *nm_i = NM_I(sbi);
215 struct nat_entry *e;
216 bool fsynced = false;
217
218 down_read(&nm_i->nat_tree_lock);
219 e = __lookup_nat_cache(nm_i, ino);
220 if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
221 fsynced = true;
222 up_read(&nm_i->nat_tree_lock);
223 return fsynced;
224 }
225
226 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
227 {
228 struct f2fs_nm_info *nm_i = NM_I(sbi);
229 struct nat_entry *e;
230 bool need_update = true;
231
232 down_read(&nm_i->nat_tree_lock);
233 e = __lookup_nat_cache(nm_i, ino);
234 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
235 (get_nat_flag(e, IS_CHECKPOINTED) ||
236 get_nat_flag(e, HAS_FSYNCED_INODE)))
237 need_update = false;
238 up_read(&nm_i->nat_tree_lock);
239 return need_update;
240 }
241
242 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
243 {
244 struct nat_entry *new;
245
246 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
247 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
248 memset(new, 0, sizeof(struct nat_entry));
249 nat_set_nid(new, nid);
250 nat_reset_flag(new);
251 list_add_tail(&new->list, &nm_i->nat_entries);
252 nm_i->nat_cnt++;
253 return new;
254 }
255
256 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
257 struct f2fs_nat_entry *ne)
258 {
259 struct nat_entry *e;
260
261 down_write(&nm_i->nat_tree_lock);
262 e = __lookup_nat_cache(nm_i, nid);
263 if (!e) {
264 e = grab_nat_entry(nm_i, nid);
265 node_info_from_raw_nat(&e->ni, ne);
266 }
267 up_write(&nm_i->nat_tree_lock);
268 }
269
270 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
271 block_t new_blkaddr, bool fsync_done)
272 {
273 struct f2fs_nm_info *nm_i = NM_I(sbi);
274 struct nat_entry *e;
275
276 down_write(&nm_i->nat_tree_lock);
277 e = __lookup_nat_cache(nm_i, ni->nid);
278 if (!e) {
279 e = grab_nat_entry(nm_i, ni->nid);
280 copy_node_info(&e->ni, ni);
281 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
282 } else if (new_blkaddr == NEW_ADDR) {
283 /*
284 * when nid is reallocated,
285 * previous nat entry can be remained in nat cache.
286 * So, reinitialize it with new information.
287 */
288 copy_node_info(&e->ni, ni);
289 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
290 }
291
292 /* sanity check */
293 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
294 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
295 new_blkaddr == NULL_ADDR);
296 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
297 new_blkaddr == NEW_ADDR);
298 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
299 nat_get_blkaddr(e) != NULL_ADDR &&
300 new_blkaddr == NEW_ADDR);
301
302 /* increment version no as node is removed */
303 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
304 unsigned char version = nat_get_version(e);
305 nat_set_version(e, inc_node_version(version));
306 }
307
308 /* change address */
309 nat_set_blkaddr(e, new_blkaddr);
310 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
311 set_nat_flag(e, IS_CHECKPOINTED, false);
312 __set_nat_cache_dirty(nm_i, e);
313
314 /* update fsync_mark if its inode nat entry is still alive */
315 e = __lookup_nat_cache(nm_i, ni->ino);
316 if (e) {
317 if (fsync_done && ni->nid == ni->ino)
318 set_nat_flag(e, HAS_FSYNCED_INODE, true);
319 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
320 }
321 up_write(&nm_i->nat_tree_lock);
322 }
323
324 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
325 {
326 struct f2fs_nm_info *nm_i = NM_I(sbi);
327
328 if (available_free_memory(sbi, NAT_ENTRIES))
329 return 0;
330
331 down_write(&nm_i->nat_tree_lock);
332 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
333 struct nat_entry *ne;
334 ne = list_first_entry(&nm_i->nat_entries,
335 struct nat_entry, list);
336 __del_from_nat_cache(nm_i, ne);
337 nr_shrink--;
338 }
339 up_write(&nm_i->nat_tree_lock);
340 return nr_shrink;
341 }
342
343 /*
344 * This function always returns success
345 */
346 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
347 {
348 struct f2fs_nm_info *nm_i = NM_I(sbi);
349 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
350 struct f2fs_summary_block *sum = curseg->sum_blk;
351 nid_t start_nid = START_NID(nid);
352 struct f2fs_nat_block *nat_blk;
353 struct page *page = NULL;
354 struct f2fs_nat_entry ne;
355 struct nat_entry *e;
356 int i;
357
358 ni->nid = nid;
359
360 /* Check nat cache */
361 down_read(&nm_i->nat_tree_lock);
362 e = __lookup_nat_cache(nm_i, nid);
363 if (e) {
364 ni->ino = nat_get_ino(e);
365 ni->blk_addr = nat_get_blkaddr(e);
366 ni->version = nat_get_version(e);
367 }
368 up_read(&nm_i->nat_tree_lock);
369 if (e)
370 return;
371
372 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
373
374 /* Check current segment summary */
375 mutex_lock(&curseg->curseg_mutex);
376 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
377 if (i >= 0) {
378 ne = nat_in_journal(sum, i);
379 node_info_from_raw_nat(ni, &ne);
380 }
381 mutex_unlock(&curseg->curseg_mutex);
382 if (i >= 0)
383 goto cache;
384
385 /* Fill node_info from nat page */
386 page = get_current_nat_page(sbi, start_nid);
387 nat_blk = (struct f2fs_nat_block *)page_address(page);
388 ne = nat_blk->entries[nid - start_nid];
389 node_info_from_raw_nat(ni, &ne);
390 f2fs_put_page(page, 1);
391 cache:
392 /* cache nat entry */
393 cache_nat_entry(NM_I(sbi), nid, &ne);
394 }
395
396 /*
397 * The maximum depth is four.
398 * Offset[0] will have raw inode offset.
399 */
400 static int get_node_path(struct f2fs_inode_info *fi, long block,
401 int offset[4], unsigned int noffset[4])
402 {
403 const long direct_index = ADDRS_PER_INODE(fi);
404 const long direct_blks = ADDRS_PER_BLOCK;
405 const long dptrs_per_blk = NIDS_PER_BLOCK;
406 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
407 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
408 int n = 0;
409 int level = 0;
410
411 noffset[0] = 0;
412
413 if (block < direct_index) {
414 offset[n] = block;
415 goto got;
416 }
417 block -= direct_index;
418 if (block < direct_blks) {
419 offset[n++] = NODE_DIR1_BLOCK;
420 noffset[n] = 1;
421 offset[n] = block;
422 level = 1;
423 goto got;
424 }
425 block -= direct_blks;
426 if (block < direct_blks) {
427 offset[n++] = NODE_DIR2_BLOCK;
428 noffset[n] = 2;
429 offset[n] = block;
430 level = 1;
431 goto got;
432 }
433 block -= direct_blks;
434 if (block < indirect_blks) {
435 offset[n++] = NODE_IND1_BLOCK;
436 noffset[n] = 3;
437 offset[n++] = block / direct_blks;
438 noffset[n] = 4 + offset[n - 1];
439 offset[n] = block % direct_blks;
440 level = 2;
441 goto got;
442 }
443 block -= indirect_blks;
444 if (block < indirect_blks) {
445 offset[n++] = NODE_IND2_BLOCK;
446 noffset[n] = 4 + dptrs_per_blk;
447 offset[n++] = block / direct_blks;
448 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
449 offset[n] = block % direct_blks;
450 level = 2;
451 goto got;
452 }
453 block -= indirect_blks;
454 if (block < dindirect_blks) {
455 offset[n++] = NODE_DIND_BLOCK;
456 noffset[n] = 5 + (dptrs_per_blk * 2);
457 offset[n++] = block / indirect_blks;
458 noffset[n] = 6 + (dptrs_per_blk * 2) +
459 offset[n - 1] * (dptrs_per_blk + 1);
460 offset[n++] = (block / direct_blks) % dptrs_per_blk;
461 noffset[n] = 7 + (dptrs_per_blk * 2) +
462 offset[n - 2] * (dptrs_per_blk + 1) +
463 offset[n - 1];
464 offset[n] = block % direct_blks;
465 level = 3;
466 goto got;
467 } else {
468 BUG();
469 }
470 got:
471 return level;
472 }
473
474 /*
475 * Caller should call f2fs_put_dnode(dn).
476 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
477 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
478 * In the case of RDONLY_NODE, we don't need to care about mutex.
479 */
480 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
481 {
482 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
483 struct page *npage[4];
484 struct page *parent = NULL;
485 int offset[4];
486 unsigned int noffset[4];
487 nid_t nids[4];
488 int level, i;
489 int err = 0;
490
491 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
492
493 nids[0] = dn->inode->i_ino;
494 npage[0] = dn->inode_page;
495
496 if (!npage[0]) {
497 npage[0] = get_node_page(sbi, nids[0]);
498 if (IS_ERR(npage[0]))
499 return PTR_ERR(npage[0]);
500 }
501
502 /* if inline_data is set, should not report any block indices */
503 if (f2fs_has_inline_data(dn->inode) && index) {
504 err = -EINVAL;
505 f2fs_put_page(npage[0], 1);
506 goto release_out;
507 }
508
509 parent = npage[0];
510 if (level != 0)
511 nids[1] = get_nid(parent, offset[0], true);
512 dn->inode_page = npage[0];
513 dn->inode_page_locked = true;
514
515 /* get indirect or direct nodes */
516 for (i = 1; i <= level; i++) {
517 bool done = false;
518
519 if (!nids[i] && mode == ALLOC_NODE) {
520 /* alloc new node */
521 if (!alloc_nid(sbi, &(nids[i]))) {
522 err = -ENOSPC;
523 goto release_pages;
524 }
525
526 dn->nid = nids[i];
527 npage[i] = new_node_page(dn, noffset[i], NULL);
528 if (IS_ERR(npage[i])) {
529 alloc_nid_failed(sbi, nids[i]);
530 err = PTR_ERR(npage[i]);
531 goto release_pages;
532 }
533
534 set_nid(parent, offset[i - 1], nids[i], i == 1);
535 alloc_nid_done(sbi, nids[i]);
536 done = true;
537 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
538 npage[i] = get_node_page_ra(parent, offset[i - 1]);
539 if (IS_ERR(npage[i])) {
540 err = PTR_ERR(npage[i]);
541 goto release_pages;
542 }
543 done = true;
544 }
545 if (i == 1) {
546 dn->inode_page_locked = false;
547 unlock_page(parent);
548 } else {
549 f2fs_put_page(parent, 1);
550 }
551
552 if (!done) {
553 npage[i] = get_node_page(sbi, nids[i]);
554 if (IS_ERR(npage[i])) {
555 err = PTR_ERR(npage[i]);
556 f2fs_put_page(npage[0], 0);
557 goto release_out;
558 }
559 }
560 if (i < level) {
561 parent = npage[i];
562 nids[i + 1] = get_nid(parent, offset[i], false);
563 }
564 }
565 dn->nid = nids[level];
566 dn->ofs_in_node = offset[level];
567 dn->node_page = npage[level];
568 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
569 return 0;
570
571 release_pages:
572 f2fs_put_page(parent, 1);
573 if (i > 1)
574 f2fs_put_page(npage[0], 0);
575 release_out:
576 dn->inode_page = NULL;
577 dn->node_page = NULL;
578 return err;
579 }
580
581 static void truncate_node(struct dnode_of_data *dn)
582 {
583 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
584 struct node_info ni;
585
586 get_node_info(sbi, dn->nid, &ni);
587 if (dn->inode->i_blocks == 0) {
588 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
589 goto invalidate;
590 }
591 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
592
593 /* Deallocate node address */
594 invalidate_blocks(sbi, ni.blk_addr);
595 dec_valid_node_count(sbi, dn->inode);
596 set_node_addr(sbi, &ni, NULL_ADDR, false);
597
598 if (dn->nid == dn->inode->i_ino) {
599 remove_orphan_inode(sbi, dn->nid);
600 dec_valid_inode_count(sbi);
601 } else {
602 sync_inode_page(dn);
603 }
604 invalidate:
605 clear_node_page_dirty(dn->node_page);
606 set_sbi_flag(sbi, SBI_IS_DIRTY);
607
608 f2fs_put_page(dn->node_page, 1);
609
610 invalidate_mapping_pages(NODE_MAPPING(sbi),
611 dn->node_page->index, dn->node_page->index);
612
613 dn->node_page = NULL;
614 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
615 }
616
617 static int truncate_dnode(struct dnode_of_data *dn)
618 {
619 struct page *page;
620
621 if (dn->nid == 0)
622 return 1;
623
624 /* get direct node */
625 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
626 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
627 return 1;
628 else if (IS_ERR(page))
629 return PTR_ERR(page);
630
631 /* Make dnode_of_data for parameter */
632 dn->node_page = page;
633 dn->ofs_in_node = 0;
634 truncate_data_blocks(dn);
635 truncate_node(dn);
636 return 1;
637 }
638
639 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
640 int ofs, int depth)
641 {
642 struct dnode_of_data rdn = *dn;
643 struct page *page;
644 struct f2fs_node *rn;
645 nid_t child_nid;
646 unsigned int child_nofs;
647 int freed = 0;
648 int i, ret;
649
650 if (dn->nid == 0)
651 return NIDS_PER_BLOCK + 1;
652
653 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
654
655 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
656 if (IS_ERR(page)) {
657 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
658 return PTR_ERR(page);
659 }
660
661 rn = F2FS_NODE(page);
662 if (depth < 3) {
663 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
664 child_nid = le32_to_cpu(rn->in.nid[i]);
665 if (child_nid == 0)
666 continue;
667 rdn.nid = child_nid;
668 ret = truncate_dnode(&rdn);
669 if (ret < 0)
670 goto out_err;
671 set_nid(page, i, 0, false);
672 }
673 } else {
674 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
675 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
676 child_nid = le32_to_cpu(rn->in.nid[i]);
677 if (child_nid == 0) {
678 child_nofs += NIDS_PER_BLOCK + 1;
679 continue;
680 }
681 rdn.nid = child_nid;
682 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
683 if (ret == (NIDS_PER_BLOCK + 1)) {
684 set_nid(page, i, 0, false);
685 child_nofs += ret;
686 } else if (ret < 0 && ret != -ENOENT) {
687 goto out_err;
688 }
689 }
690 freed = child_nofs;
691 }
692
693 if (!ofs) {
694 /* remove current indirect node */
695 dn->node_page = page;
696 truncate_node(dn);
697 freed++;
698 } else {
699 f2fs_put_page(page, 1);
700 }
701 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
702 return freed;
703
704 out_err:
705 f2fs_put_page(page, 1);
706 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
707 return ret;
708 }
709
710 static int truncate_partial_nodes(struct dnode_of_data *dn,
711 struct f2fs_inode *ri, int *offset, int depth)
712 {
713 struct page *pages[2];
714 nid_t nid[3];
715 nid_t child_nid;
716 int err = 0;
717 int i;
718 int idx = depth - 2;
719
720 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
721 if (!nid[0])
722 return 0;
723
724 /* get indirect nodes in the path */
725 for (i = 0; i < idx + 1; i++) {
726 /* reference count'll be increased */
727 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
728 if (IS_ERR(pages[i])) {
729 err = PTR_ERR(pages[i]);
730 idx = i - 1;
731 goto fail;
732 }
733 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
734 }
735
736 /* free direct nodes linked to a partial indirect node */
737 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
738 child_nid = get_nid(pages[idx], i, false);
739 if (!child_nid)
740 continue;
741 dn->nid = child_nid;
742 err = truncate_dnode(dn);
743 if (err < 0)
744 goto fail;
745 set_nid(pages[idx], i, 0, false);
746 }
747
748 if (offset[idx + 1] == 0) {
749 dn->node_page = pages[idx];
750 dn->nid = nid[idx];
751 truncate_node(dn);
752 } else {
753 f2fs_put_page(pages[idx], 1);
754 }
755 offset[idx]++;
756 offset[idx + 1] = 0;
757 idx--;
758 fail:
759 for (i = idx; i >= 0; i--)
760 f2fs_put_page(pages[i], 1);
761
762 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
763
764 return err;
765 }
766
767 /*
768 * All the block addresses of data and nodes should be nullified.
769 */
770 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
771 {
772 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
773 int err = 0, cont = 1;
774 int level, offset[4], noffset[4];
775 unsigned int nofs = 0;
776 struct f2fs_inode *ri;
777 struct dnode_of_data dn;
778 struct page *page;
779
780 trace_f2fs_truncate_inode_blocks_enter(inode, from);
781
782 level = get_node_path(F2FS_I(inode), from, offset, noffset);
783 restart:
784 page = get_node_page(sbi, inode->i_ino);
785 if (IS_ERR(page)) {
786 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
787 return PTR_ERR(page);
788 }
789
790 set_new_dnode(&dn, inode, page, NULL, 0);
791 unlock_page(page);
792
793 ri = F2FS_INODE(page);
794 switch (level) {
795 case 0:
796 case 1:
797 nofs = noffset[1];
798 break;
799 case 2:
800 nofs = noffset[1];
801 if (!offset[level - 1])
802 goto skip_partial;
803 err = truncate_partial_nodes(&dn, ri, offset, level);
804 if (err < 0 && err != -ENOENT)
805 goto fail;
806 nofs += 1 + NIDS_PER_BLOCK;
807 break;
808 case 3:
809 nofs = 5 + 2 * NIDS_PER_BLOCK;
810 if (!offset[level - 1])
811 goto skip_partial;
812 err = truncate_partial_nodes(&dn, ri, offset, level);
813 if (err < 0 && err != -ENOENT)
814 goto fail;
815 break;
816 default:
817 BUG();
818 }
819
820 skip_partial:
821 while (cont) {
822 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
823 switch (offset[0]) {
824 case NODE_DIR1_BLOCK:
825 case NODE_DIR2_BLOCK:
826 err = truncate_dnode(&dn);
827 break;
828
829 case NODE_IND1_BLOCK:
830 case NODE_IND2_BLOCK:
831 err = truncate_nodes(&dn, nofs, offset[1], 2);
832 break;
833
834 case NODE_DIND_BLOCK:
835 err = truncate_nodes(&dn, nofs, offset[1], 3);
836 cont = 0;
837 break;
838
839 default:
840 BUG();
841 }
842 if (err < 0 && err != -ENOENT)
843 goto fail;
844 if (offset[1] == 0 &&
845 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
846 lock_page(page);
847 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
848 f2fs_put_page(page, 1);
849 goto restart;
850 }
851 f2fs_wait_on_page_writeback(page, NODE);
852 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
853 set_page_dirty(page);
854 unlock_page(page);
855 }
856 offset[1] = 0;
857 offset[0]++;
858 nofs += err;
859 }
860 fail:
861 f2fs_put_page(page, 0);
862 trace_f2fs_truncate_inode_blocks_exit(inode, err);
863 return err > 0 ? 0 : err;
864 }
865
866 int truncate_xattr_node(struct inode *inode, struct page *page)
867 {
868 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
869 nid_t nid = F2FS_I(inode)->i_xattr_nid;
870 struct dnode_of_data dn;
871 struct page *npage;
872
873 if (!nid)
874 return 0;
875
876 npage = get_node_page(sbi, nid);
877 if (IS_ERR(npage))
878 return PTR_ERR(npage);
879
880 F2FS_I(inode)->i_xattr_nid = 0;
881
882 /* need to do checkpoint during fsync */
883 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
884
885 set_new_dnode(&dn, inode, page, npage, nid);
886
887 if (page)
888 dn.inode_page_locked = true;
889 truncate_node(&dn);
890 return 0;
891 }
892
893 /*
894 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
895 * f2fs_unlock_op().
896 */
897 void remove_inode_page(struct inode *inode)
898 {
899 struct dnode_of_data dn;
900
901 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
902 if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
903 return;
904
905 if (truncate_xattr_node(inode, dn.inode_page)) {
906 f2fs_put_dnode(&dn);
907 return;
908 }
909
910 /* remove potential inline_data blocks */
911 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
912 S_ISLNK(inode->i_mode))
913 truncate_data_blocks_range(&dn, 1);
914
915 /* 0 is possible, after f2fs_new_inode() has failed */
916 f2fs_bug_on(F2FS_I_SB(inode),
917 inode->i_blocks != 0 && inode->i_blocks != 1);
918
919 /* will put inode & node pages */
920 truncate_node(&dn);
921 }
922
923 struct page *new_inode_page(struct inode *inode)
924 {
925 struct dnode_of_data dn;
926
927 /* allocate inode page for new inode */
928 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
929
930 /* caller should f2fs_put_page(page, 1); */
931 return new_node_page(&dn, 0, NULL);
932 }
933
934 struct page *new_node_page(struct dnode_of_data *dn,
935 unsigned int ofs, struct page *ipage)
936 {
937 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
938 struct node_info old_ni, new_ni;
939 struct page *page;
940 int err;
941
942 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
943 return ERR_PTR(-EPERM);
944
945 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
946 if (!page)
947 return ERR_PTR(-ENOMEM);
948
949 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
950 err = -ENOSPC;
951 goto fail;
952 }
953
954 get_node_info(sbi, dn->nid, &old_ni);
955
956 /* Reinitialize old_ni with new node page */
957 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
958 new_ni = old_ni;
959 new_ni.ino = dn->inode->i_ino;
960 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
961
962 f2fs_wait_on_page_writeback(page, NODE);
963 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
964 set_cold_node(dn->inode, page);
965 SetPageUptodate(page);
966 set_page_dirty(page);
967
968 if (f2fs_has_xattr_block(ofs))
969 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
970
971 dn->node_page = page;
972 if (ipage)
973 update_inode(dn->inode, ipage);
974 else
975 sync_inode_page(dn);
976 if (ofs == 0)
977 inc_valid_inode_count(sbi);
978
979 return page;
980
981 fail:
982 clear_node_page_dirty(page);
983 f2fs_put_page(page, 1);
984 return ERR_PTR(err);
985 }
986
987 /*
988 * Caller should do after getting the following values.
989 * 0: f2fs_put_page(page, 0)
990 * LOCKED_PAGE: f2fs_put_page(page, 1)
991 * error: nothing
992 */
993 static int read_node_page(struct page *page, int rw)
994 {
995 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
996 struct node_info ni;
997 struct f2fs_io_info fio = {
998 .type = NODE,
999 .rw = rw,
1000 };
1001
1002 get_node_info(sbi, page->index, &ni);
1003
1004 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1005 f2fs_put_page(page, 1);
1006 return -ENOENT;
1007 }
1008
1009 if (PageUptodate(page))
1010 return LOCKED_PAGE;
1011
1012 fio.blk_addr = ni.blk_addr;
1013 return f2fs_submit_page_bio(sbi, page, &fio);
1014 }
1015
1016 /*
1017 * Readahead a node page
1018 */
1019 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1020 {
1021 struct page *apage;
1022 int err;
1023
1024 apage = find_get_page(NODE_MAPPING(sbi), nid);
1025 if (apage && PageUptodate(apage)) {
1026 f2fs_put_page(apage, 0);
1027 return;
1028 }
1029 f2fs_put_page(apage, 0);
1030
1031 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1032 if (!apage)
1033 return;
1034
1035 err = read_node_page(apage, READA);
1036 if (err == 0)
1037 f2fs_put_page(apage, 0);
1038 else if (err == LOCKED_PAGE)
1039 f2fs_put_page(apage, 1);
1040 }
1041
1042 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1043 {
1044 struct page *page;
1045 int err;
1046 repeat:
1047 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1048 if (!page)
1049 return ERR_PTR(-ENOMEM);
1050
1051 err = read_node_page(page, READ_SYNC);
1052 if (err < 0)
1053 return ERR_PTR(err);
1054 else if (err != LOCKED_PAGE)
1055 lock_page(page);
1056
1057 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1058 ClearPageUptodate(page);
1059 f2fs_put_page(page, 1);
1060 return ERR_PTR(-EIO);
1061 }
1062 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1063 f2fs_put_page(page, 1);
1064 goto repeat;
1065 }
1066 return page;
1067 }
1068
1069 /*
1070 * Return a locked page for the desired node page.
1071 * And, readahead MAX_RA_NODE number of node pages.
1072 */
1073 struct page *get_node_page_ra(struct page *parent, int start)
1074 {
1075 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1076 struct blk_plug plug;
1077 struct page *page;
1078 int err, i, end;
1079 nid_t nid;
1080
1081 /* First, try getting the desired direct node. */
1082 nid = get_nid(parent, start, false);
1083 if (!nid)
1084 return ERR_PTR(-ENOENT);
1085 repeat:
1086 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1087 if (!page)
1088 return ERR_PTR(-ENOMEM);
1089
1090 err = read_node_page(page, READ_SYNC);
1091 if (err < 0)
1092 return ERR_PTR(err);
1093 else if (err == LOCKED_PAGE)
1094 goto page_hit;
1095
1096 blk_start_plug(&plug);
1097
1098 /* Then, try readahead for siblings of the desired node */
1099 end = start + MAX_RA_NODE;
1100 end = min(end, NIDS_PER_BLOCK);
1101 for (i = start + 1; i < end; i++) {
1102 nid = get_nid(parent, i, false);
1103 if (!nid)
1104 continue;
1105 ra_node_page(sbi, nid);
1106 }
1107
1108 blk_finish_plug(&plug);
1109
1110 lock_page(page);
1111 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1112 f2fs_put_page(page, 1);
1113 goto repeat;
1114 }
1115 page_hit:
1116 if (unlikely(!PageUptodate(page))) {
1117 f2fs_put_page(page, 1);
1118 return ERR_PTR(-EIO);
1119 }
1120 return page;
1121 }
1122
1123 void sync_inode_page(struct dnode_of_data *dn)
1124 {
1125 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1126 update_inode(dn->inode, dn->node_page);
1127 } else if (dn->inode_page) {
1128 if (!dn->inode_page_locked)
1129 lock_page(dn->inode_page);
1130 update_inode(dn->inode, dn->inode_page);
1131 if (!dn->inode_page_locked)
1132 unlock_page(dn->inode_page);
1133 } else {
1134 update_inode_page(dn->inode);
1135 }
1136 }
1137
1138 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1139 struct writeback_control *wbc)
1140 {
1141 pgoff_t index, end;
1142 struct pagevec pvec;
1143 int step = ino ? 2 : 0;
1144 int nwritten = 0, wrote = 0;
1145
1146 pagevec_init(&pvec, 0);
1147
1148 next_step:
1149 index = 0;
1150 end = LONG_MAX;
1151
1152 while (index <= end) {
1153 int i, nr_pages;
1154 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1155 PAGECACHE_TAG_DIRTY,
1156 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1157 if (nr_pages == 0)
1158 break;
1159
1160 for (i = 0; i < nr_pages; i++) {
1161 struct page *page = pvec.pages[i];
1162
1163 /*
1164 * flushing sequence with step:
1165 * 0. indirect nodes
1166 * 1. dentry dnodes
1167 * 2. file dnodes
1168 */
1169 if (step == 0 && IS_DNODE(page))
1170 continue;
1171 if (step == 1 && (!IS_DNODE(page) ||
1172 is_cold_node(page)))
1173 continue;
1174 if (step == 2 && (!IS_DNODE(page) ||
1175 !is_cold_node(page)))
1176 continue;
1177
1178 /*
1179 * If an fsync mode,
1180 * we should not skip writing node pages.
1181 */
1182 if (ino && ino_of_node(page) == ino)
1183 lock_page(page);
1184 else if (!trylock_page(page))
1185 continue;
1186
1187 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1188 continue_unlock:
1189 unlock_page(page);
1190 continue;
1191 }
1192 if (ino && ino_of_node(page) != ino)
1193 goto continue_unlock;
1194
1195 if (!PageDirty(page)) {
1196 /* someone wrote it for us */
1197 goto continue_unlock;
1198 }
1199
1200 if (!clear_page_dirty_for_io(page))
1201 goto continue_unlock;
1202
1203 /* called by fsync() */
1204 if (ino && IS_DNODE(page)) {
1205 set_fsync_mark(page, 1);
1206 if (IS_INODE(page)) {
1207 if (!is_checkpointed_node(sbi, ino) &&
1208 !has_fsynced_inode(sbi, ino))
1209 set_dentry_mark(page, 1);
1210 else
1211 set_dentry_mark(page, 0);
1212 }
1213 nwritten++;
1214 } else {
1215 set_fsync_mark(page, 0);
1216 set_dentry_mark(page, 0);
1217 }
1218
1219 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1220 unlock_page(page);
1221 else
1222 wrote++;
1223
1224 if (--wbc->nr_to_write == 0)
1225 break;
1226 }
1227 pagevec_release(&pvec);
1228 cond_resched();
1229
1230 if (wbc->nr_to_write == 0) {
1231 step = 2;
1232 break;
1233 }
1234 }
1235
1236 if (step < 2) {
1237 step++;
1238 goto next_step;
1239 }
1240
1241 if (wrote)
1242 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1243 return nwritten;
1244 }
1245
1246 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1247 {
1248 pgoff_t index = 0, end = LONG_MAX;
1249 struct pagevec pvec;
1250 int ret2 = 0, ret = 0;
1251
1252 pagevec_init(&pvec, 0);
1253
1254 while (index <= end) {
1255 int i, nr_pages;
1256 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1257 PAGECACHE_TAG_WRITEBACK,
1258 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1259 if (nr_pages == 0)
1260 break;
1261
1262 for (i = 0; i < nr_pages; i++) {
1263 struct page *page = pvec.pages[i];
1264
1265 /* until radix tree lookup accepts end_index */
1266 if (unlikely(page->index > end))
1267 continue;
1268
1269 if (ino && ino_of_node(page) == ino) {
1270 f2fs_wait_on_page_writeback(page, NODE);
1271 if (TestClearPageError(page))
1272 ret = -EIO;
1273 }
1274 }
1275 pagevec_release(&pvec);
1276 cond_resched();
1277 }
1278
1279 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1280 ret2 = -ENOSPC;
1281 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1282 ret2 = -EIO;
1283 if (!ret)
1284 ret = ret2;
1285 return ret;
1286 }
1287
1288 static int f2fs_write_node_page(struct page *page,
1289 struct writeback_control *wbc)
1290 {
1291 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1292 nid_t nid;
1293 struct node_info ni;
1294 struct f2fs_io_info fio = {
1295 .type = NODE,
1296 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1297 };
1298
1299 trace_f2fs_writepage(page, NODE);
1300
1301 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1302 goto redirty_out;
1303 if (unlikely(f2fs_cp_error(sbi)))
1304 goto redirty_out;
1305
1306 f2fs_wait_on_page_writeback(page, NODE);
1307
1308 /* get old block addr of this node page */
1309 nid = nid_of_node(page);
1310 f2fs_bug_on(sbi, page->index != nid);
1311
1312 get_node_info(sbi, nid, &ni);
1313
1314 /* This page is already truncated */
1315 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1316 dec_page_count(sbi, F2FS_DIRTY_NODES);
1317 unlock_page(page);
1318 return 0;
1319 }
1320
1321 if (wbc->for_reclaim) {
1322 if (!down_read_trylock(&sbi->node_write))
1323 goto redirty_out;
1324 } else {
1325 down_read(&sbi->node_write);
1326 }
1327
1328 set_page_writeback(page);
1329 fio.blk_addr = ni.blk_addr;
1330 write_node_page(sbi, page, nid, &fio);
1331 set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1332 dec_page_count(sbi, F2FS_DIRTY_NODES);
1333 up_read(&sbi->node_write);
1334 unlock_page(page);
1335
1336 if (wbc->for_reclaim)
1337 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1338
1339 return 0;
1340
1341 redirty_out:
1342 redirty_page_for_writepage(wbc, page);
1343 return AOP_WRITEPAGE_ACTIVATE;
1344 }
1345
1346 static int f2fs_write_node_pages(struct address_space *mapping,
1347 struct writeback_control *wbc)
1348 {
1349 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1350 long diff;
1351
1352 trace_f2fs_writepages(mapping->host, wbc, NODE);
1353
1354 /* balancing f2fs's metadata in background */
1355 f2fs_balance_fs_bg(sbi);
1356
1357 /* collect a number of dirty node pages and write together */
1358 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1359 goto skip_write;
1360
1361 diff = nr_pages_to_write(sbi, NODE, wbc);
1362 wbc->sync_mode = WB_SYNC_NONE;
1363 sync_node_pages(sbi, 0, wbc);
1364 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1365 return 0;
1366
1367 skip_write:
1368 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1369 return 0;
1370 }
1371
1372 static int f2fs_set_node_page_dirty(struct page *page)
1373 {
1374 trace_f2fs_set_page_dirty(page, NODE);
1375
1376 SetPageUptodate(page);
1377 if (!PageDirty(page)) {
1378 __set_page_dirty_nobuffers(page);
1379 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1380 SetPagePrivate(page);
1381 f2fs_trace_pid(page);
1382 return 1;
1383 }
1384 return 0;
1385 }
1386
1387 /*
1388 * Structure of the f2fs node operations
1389 */
1390 const struct address_space_operations f2fs_node_aops = {
1391 .writepage = f2fs_write_node_page,
1392 .writepages = f2fs_write_node_pages,
1393 .set_page_dirty = f2fs_set_node_page_dirty,
1394 .invalidatepage = f2fs_invalidate_page,
1395 .releasepage = f2fs_release_page,
1396 };
1397
1398 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1399 nid_t n)
1400 {
1401 return radix_tree_lookup(&nm_i->free_nid_root, n);
1402 }
1403
1404 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1405 struct free_nid *i)
1406 {
1407 list_del(&i->list);
1408 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1409 }
1410
1411 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1412 {
1413 struct f2fs_nm_info *nm_i = NM_I(sbi);
1414 struct free_nid *i;
1415 struct nat_entry *ne;
1416 bool allocated = false;
1417
1418 if (!available_free_memory(sbi, FREE_NIDS))
1419 return -1;
1420
1421 /* 0 nid should not be used */
1422 if (unlikely(nid == 0))
1423 return 0;
1424
1425 if (build) {
1426 /* do not add allocated nids */
1427 down_read(&nm_i->nat_tree_lock);
1428 ne = __lookup_nat_cache(nm_i, nid);
1429 if (ne &&
1430 (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1431 nat_get_blkaddr(ne) != NULL_ADDR))
1432 allocated = true;
1433 up_read(&nm_i->nat_tree_lock);
1434 if (allocated)
1435 return 0;
1436 }
1437
1438 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1439 i->nid = nid;
1440 i->state = NID_NEW;
1441
1442 if (radix_tree_preload(GFP_NOFS)) {
1443 kmem_cache_free(free_nid_slab, i);
1444 return 0;
1445 }
1446
1447 spin_lock(&nm_i->free_nid_list_lock);
1448 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1449 spin_unlock(&nm_i->free_nid_list_lock);
1450 radix_tree_preload_end();
1451 kmem_cache_free(free_nid_slab, i);
1452 return 0;
1453 }
1454 list_add_tail(&i->list, &nm_i->free_nid_list);
1455 nm_i->fcnt++;
1456 spin_unlock(&nm_i->free_nid_list_lock);
1457 radix_tree_preload_end();
1458 return 1;
1459 }
1460
1461 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1462 {
1463 struct free_nid *i;
1464 bool need_free = false;
1465
1466 spin_lock(&nm_i->free_nid_list_lock);
1467 i = __lookup_free_nid_list(nm_i, nid);
1468 if (i && i->state == NID_NEW) {
1469 __del_from_free_nid_list(nm_i, i);
1470 nm_i->fcnt--;
1471 need_free = true;
1472 }
1473 spin_unlock(&nm_i->free_nid_list_lock);
1474
1475 if (need_free)
1476 kmem_cache_free(free_nid_slab, i);
1477 }
1478
1479 static void scan_nat_page(struct f2fs_sb_info *sbi,
1480 struct page *nat_page, nid_t start_nid)
1481 {
1482 struct f2fs_nm_info *nm_i = NM_I(sbi);
1483 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1484 block_t blk_addr;
1485 int i;
1486
1487 i = start_nid % NAT_ENTRY_PER_BLOCK;
1488
1489 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1490
1491 if (unlikely(start_nid >= nm_i->max_nid))
1492 break;
1493
1494 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1495 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1496 if (blk_addr == NULL_ADDR) {
1497 if (add_free_nid(sbi, start_nid, true) < 0)
1498 break;
1499 }
1500 }
1501 }
1502
1503 static void build_free_nids(struct f2fs_sb_info *sbi)
1504 {
1505 struct f2fs_nm_info *nm_i = NM_I(sbi);
1506 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1507 struct f2fs_summary_block *sum = curseg->sum_blk;
1508 int i = 0;
1509 nid_t nid = nm_i->next_scan_nid;
1510
1511 /* Enough entries */
1512 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1513 return;
1514
1515 /* readahead nat pages to be scanned */
1516 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1517
1518 while (1) {
1519 struct page *page = get_current_nat_page(sbi, nid);
1520
1521 scan_nat_page(sbi, page, nid);
1522 f2fs_put_page(page, 1);
1523
1524 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1525 if (unlikely(nid >= nm_i->max_nid))
1526 nid = 0;
1527
1528 if (i++ == FREE_NID_PAGES)
1529 break;
1530 }
1531
1532 /* go to the next free nat pages to find free nids abundantly */
1533 nm_i->next_scan_nid = nid;
1534
1535 /* find free nids from current sum_pages */
1536 mutex_lock(&curseg->curseg_mutex);
1537 for (i = 0; i < nats_in_cursum(sum); i++) {
1538 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1539 nid = le32_to_cpu(nid_in_journal(sum, i));
1540 if (addr == NULL_ADDR)
1541 add_free_nid(sbi, nid, true);
1542 else
1543 remove_free_nid(nm_i, nid);
1544 }
1545 mutex_unlock(&curseg->curseg_mutex);
1546 }
1547
1548 /*
1549 * If this function returns success, caller can obtain a new nid
1550 * from second parameter of this function.
1551 * The returned nid could be used ino as well as nid when inode is created.
1552 */
1553 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1554 {
1555 struct f2fs_nm_info *nm_i = NM_I(sbi);
1556 struct free_nid *i = NULL;
1557 retry:
1558 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1559 return false;
1560
1561 spin_lock(&nm_i->free_nid_list_lock);
1562
1563 /* We should not use stale free nids created by build_free_nids */
1564 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1565 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1566 list_for_each_entry(i, &nm_i->free_nid_list, list)
1567 if (i->state == NID_NEW)
1568 break;
1569
1570 f2fs_bug_on(sbi, i->state != NID_NEW);
1571 *nid = i->nid;
1572 i->state = NID_ALLOC;
1573 nm_i->fcnt--;
1574 spin_unlock(&nm_i->free_nid_list_lock);
1575 return true;
1576 }
1577 spin_unlock(&nm_i->free_nid_list_lock);
1578
1579 /* Let's scan nat pages and its caches to get free nids */
1580 mutex_lock(&nm_i->build_lock);
1581 build_free_nids(sbi);
1582 mutex_unlock(&nm_i->build_lock);
1583 goto retry;
1584 }
1585
1586 /*
1587 * alloc_nid() should be called prior to this function.
1588 */
1589 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1590 {
1591 struct f2fs_nm_info *nm_i = NM_I(sbi);
1592 struct free_nid *i;
1593
1594 spin_lock(&nm_i->free_nid_list_lock);
1595 i = __lookup_free_nid_list(nm_i, nid);
1596 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1597 __del_from_free_nid_list(nm_i, i);
1598 spin_unlock(&nm_i->free_nid_list_lock);
1599
1600 kmem_cache_free(free_nid_slab, i);
1601 }
1602
1603 /*
1604 * alloc_nid() should be called prior to this function.
1605 */
1606 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1607 {
1608 struct f2fs_nm_info *nm_i = NM_I(sbi);
1609 struct free_nid *i;
1610 bool need_free = false;
1611
1612 if (!nid)
1613 return;
1614
1615 spin_lock(&nm_i->free_nid_list_lock);
1616 i = __lookup_free_nid_list(nm_i, nid);
1617 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1618 if (!available_free_memory(sbi, FREE_NIDS)) {
1619 __del_from_free_nid_list(nm_i, i);
1620 need_free = true;
1621 } else {
1622 i->state = NID_NEW;
1623 nm_i->fcnt++;
1624 }
1625 spin_unlock(&nm_i->free_nid_list_lock);
1626
1627 if (need_free)
1628 kmem_cache_free(free_nid_slab, i);
1629 }
1630
1631 void recover_inline_xattr(struct inode *inode, struct page *page)
1632 {
1633 void *src_addr, *dst_addr;
1634 size_t inline_size;
1635 struct page *ipage;
1636 struct f2fs_inode *ri;
1637
1638 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1639 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1640
1641 ri = F2FS_INODE(page);
1642 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1643 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1644 goto update_inode;
1645 }
1646
1647 dst_addr = inline_xattr_addr(ipage);
1648 src_addr = inline_xattr_addr(page);
1649 inline_size = inline_xattr_size(inode);
1650
1651 f2fs_wait_on_page_writeback(ipage, NODE);
1652 memcpy(dst_addr, src_addr, inline_size);
1653 update_inode:
1654 update_inode(inode, ipage);
1655 f2fs_put_page(ipage, 1);
1656 }
1657
1658 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1659 {
1660 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1661 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1662 nid_t new_xnid = nid_of_node(page);
1663 struct node_info ni;
1664
1665 /* 1: invalidate the previous xattr nid */
1666 if (!prev_xnid)
1667 goto recover_xnid;
1668
1669 /* Deallocate node address */
1670 get_node_info(sbi, prev_xnid, &ni);
1671 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1672 invalidate_blocks(sbi, ni.blk_addr);
1673 dec_valid_node_count(sbi, inode);
1674 set_node_addr(sbi, &ni, NULL_ADDR, false);
1675
1676 recover_xnid:
1677 /* 2: allocate new xattr nid */
1678 if (unlikely(!inc_valid_node_count(sbi, inode)))
1679 f2fs_bug_on(sbi, 1);
1680
1681 remove_free_nid(NM_I(sbi), new_xnid);
1682 get_node_info(sbi, new_xnid, &ni);
1683 ni.ino = inode->i_ino;
1684 set_node_addr(sbi, &ni, NEW_ADDR, false);
1685 F2FS_I(inode)->i_xattr_nid = new_xnid;
1686
1687 /* 3: update xattr blkaddr */
1688 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1689 set_node_addr(sbi, &ni, blkaddr, false);
1690
1691 update_inode_page(inode);
1692 }
1693
1694 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1695 {
1696 struct f2fs_inode *src, *dst;
1697 nid_t ino = ino_of_node(page);
1698 struct node_info old_ni, new_ni;
1699 struct page *ipage;
1700
1701 get_node_info(sbi, ino, &old_ni);
1702
1703 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1704 return -EINVAL;
1705
1706 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1707 if (!ipage)
1708 return -ENOMEM;
1709
1710 /* Should not use this inode from free nid list */
1711 remove_free_nid(NM_I(sbi), ino);
1712
1713 SetPageUptodate(ipage);
1714 fill_node_footer(ipage, ino, ino, 0, true);
1715
1716 src = F2FS_INODE(page);
1717 dst = F2FS_INODE(ipage);
1718
1719 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1720 dst->i_size = 0;
1721 dst->i_blocks = cpu_to_le64(1);
1722 dst->i_links = cpu_to_le32(1);
1723 dst->i_xattr_nid = 0;
1724 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1725
1726 new_ni = old_ni;
1727 new_ni.ino = ino;
1728
1729 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1730 WARN_ON(1);
1731 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1732 inc_valid_inode_count(sbi);
1733 set_page_dirty(ipage);
1734 f2fs_put_page(ipage, 1);
1735 return 0;
1736 }
1737
1738 int restore_node_summary(struct f2fs_sb_info *sbi,
1739 unsigned int segno, struct f2fs_summary_block *sum)
1740 {
1741 struct f2fs_node *rn;
1742 struct f2fs_summary *sum_entry;
1743 block_t addr;
1744 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1745 int i, idx, last_offset, nrpages;
1746
1747 /* scan the node segment */
1748 last_offset = sbi->blocks_per_seg;
1749 addr = START_BLOCK(sbi, segno);
1750 sum_entry = &sum->entries[0];
1751
1752 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1753 nrpages = min(last_offset - i, bio_blocks);
1754
1755 /* readahead node pages */
1756 ra_meta_pages(sbi, addr, nrpages, META_POR);
1757
1758 for (idx = addr; idx < addr + nrpages; idx++) {
1759 struct page *page = get_meta_page(sbi, idx);
1760
1761 rn = F2FS_NODE(page);
1762 sum_entry->nid = rn->footer.nid;
1763 sum_entry->version = 0;
1764 sum_entry->ofs_in_node = 0;
1765 sum_entry++;
1766 f2fs_put_page(page, 1);
1767 }
1768
1769 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1770 addr + nrpages);
1771 }
1772 return 0;
1773 }
1774
1775 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1776 {
1777 struct f2fs_nm_info *nm_i = NM_I(sbi);
1778 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1779 struct f2fs_summary_block *sum = curseg->sum_blk;
1780 int i;
1781
1782 mutex_lock(&curseg->curseg_mutex);
1783 for (i = 0; i < nats_in_cursum(sum); i++) {
1784 struct nat_entry *ne;
1785 struct f2fs_nat_entry raw_ne;
1786 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1787
1788 raw_ne = nat_in_journal(sum, i);
1789
1790 down_write(&nm_i->nat_tree_lock);
1791 ne = __lookup_nat_cache(nm_i, nid);
1792 if (!ne) {
1793 ne = grab_nat_entry(nm_i, nid);
1794 node_info_from_raw_nat(&ne->ni, &raw_ne);
1795 }
1796 __set_nat_cache_dirty(nm_i, ne);
1797 up_write(&nm_i->nat_tree_lock);
1798 }
1799 update_nats_in_cursum(sum, -i);
1800 mutex_unlock(&curseg->curseg_mutex);
1801 }
1802
1803 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1804 struct list_head *head, int max)
1805 {
1806 struct nat_entry_set *cur;
1807
1808 if (nes->entry_cnt >= max)
1809 goto add_out;
1810
1811 list_for_each_entry(cur, head, set_list) {
1812 if (cur->entry_cnt >= nes->entry_cnt) {
1813 list_add(&nes->set_list, cur->set_list.prev);
1814 return;
1815 }
1816 }
1817 add_out:
1818 list_add_tail(&nes->set_list, head);
1819 }
1820
1821 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1822 struct nat_entry_set *set)
1823 {
1824 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1825 struct f2fs_summary_block *sum = curseg->sum_blk;
1826 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1827 bool to_journal = true;
1828 struct f2fs_nat_block *nat_blk;
1829 struct nat_entry *ne, *cur;
1830 struct page *page = NULL;
1831
1832 /*
1833 * there are two steps to flush nat entries:
1834 * #1, flush nat entries to journal in current hot data summary block.
1835 * #2, flush nat entries to nat page.
1836 */
1837 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1838 to_journal = false;
1839
1840 if (to_journal) {
1841 mutex_lock(&curseg->curseg_mutex);
1842 } else {
1843 page = get_next_nat_page(sbi, start_nid);
1844 nat_blk = page_address(page);
1845 f2fs_bug_on(sbi, !nat_blk);
1846 }
1847
1848 /* flush dirty nats in nat entry set */
1849 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1850 struct f2fs_nat_entry *raw_ne;
1851 nid_t nid = nat_get_nid(ne);
1852 int offset;
1853
1854 if (nat_get_blkaddr(ne) == NEW_ADDR)
1855 continue;
1856
1857 if (to_journal) {
1858 offset = lookup_journal_in_cursum(sum,
1859 NAT_JOURNAL, nid, 1);
1860 f2fs_bug_on(sbi, offset < 0);
1861 raw_ne = &nat_in_journal(sum, offset);
1862 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1863 } else {
1864 raw_ne = &nat_blk->entries[nid - start_nid];
1865 }
1866 raw_nat_from_node_info(raw_ne, &ne->ni);
1867
1868 down_write(&NM_I(sbi)->nat_tree_lock);
1869 nat_reset_flag(ne);
1870 __clear_nat_cache_dirty(NM_I(sbi), ne);
1871 up_write(&NM_I(sbi)->nat_tree_lock);
1872
1873 if (nat_get_blkaddr(ne) == NULL_ADDR)
1874 add_free_nid(sbi, nid, false);
1875 }
1876
1877 if (to_journal)
1878 mutex_unlock(&curseg->curseg_mutex);
1879 else
1880 f2fs_put_page(page, 1);
1881
1882 f2fs_bug_on(sbi, set->entry_cnt);
1883
1884 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1885 kmem_cache_free(nat_entry_set_slab, set);
1886 }
1887
1888 /*
1889 * This function is called during the checkpointing process.
1890 */
1891 void flush_nat_entries(struct f2fs_sb_info *sbi)
1892 {
1893 struct f2fs_nm_info *nm_i = NM_I(sbi);
1894 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1895 struct f2fs_summary_block *sum = curseg->sum_blk;
1896 struct nat_entry_set *setvec[SETVEC_SIZE];
1897 struct nat_entry_set *set, *tmp;
1898 unsigned int found;
1899 nid_t set_idx = 0;
1900 LIST_HEAD(sets);
1901
1902 if (!nm_i->dirty_nat_cnt)
1903 return;
1904 /*
1905 * if there are no enough space in journal to store dirty nat
1906 * entries, remove all entries from journal and merge them
1907 * into nat entry set.
1908 */
1909 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1910 remove_nats_in_journal(sbi);
1911
1912 while ((found = __gang_lookup_nat_set(nm_i,
1913 set_idx, SETVEC_SIZE, setvec))) {
1914 unsigned idx;
1915 set_idx = setvec[found - 1]->set + 1;
1916 for (idx = 0; idx < found; idx++)
1917 __adjust_nat_entry_set(setvec[idx], &sets,
1918 MAX_NAT_JENTRIES(sum));
1919 }
1920
1921 /* flush dirty nats in nat entry set */
1922 list_for_each_entry_safe(set, tmp, &sets, set_list)
1923 __flush_nat_entry_set(sbi, set);
1924
1925 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1926 }
1927
1928 static int init_node_manager(struct f2fs_sb_info *sbi)
1929 {
1930 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1931 struct f2fs_nm_info *nm_i = NM_I(sbi);
1932 unsigned char *version_bitmap;
1933 unsigned int nat_segs, nat_blocks;
1934
1935 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1936
1937 /* segment_count_nat includes pair segment so divide to 2. */
1938 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1939 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1940
1941 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1942
1943 /* not used nids: 0, node, meta, (and root counted as valid node) */
1944 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1945 nm_i->fcnt = 0;
1946 nm_i->nat_cnt = 0;
1947 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1948
1949 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1950 INIT_LIST_HEAD(&nm_i->free_nid_list);
1951 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1952 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1953 INIT_LIST_HEAD(&nm_i->nat_entries);
1954
1955 mutex_init(&nm_i->build_lock);
1956 spin_lock_init(&nm_i->free_nid_list_lock);
1957 init_rwsem(&nm_i->nat_tree_lock);
1958
1959 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1960 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1961 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1962 if (!version_bitmap)
1963 return -EFAULT;
1964
1965 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1966 GFP_KERNEL);
1967 if (!nm_i->nat_bitmap)
1968 return -ENOMEM;
1969 return 0;
1970 }
1971
1972 int build_node_manager(struct f2fs_sb_info *sbi)
1973 {
1974 int err;
1975
1976 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1977 if (!sbi->nm_info)
1978 return -ENOMEM;
1979
1980 err = init_node_manager(sbi);
1981 if (err)
1982 return err;
1983
1984 build_free_nids(sbi);
1985 return 0;
1986 }
1987
1988 void destroy_node_manager(struct f2fs_sb_info *sbi)
1989 {
1990 struct f2fs_nm_info *nm_i = NM_I(sbi);
1991 struct free_nid *i, *next_i;
1992 struct nat_entry *natvec[NATVEC_SIZE];
1993 struct nat_entry_set *setvec[SETVEC_SIZE];
1994 nid_t nid = 0;
1995 unsigned int found;
1996
1997 if (!nm_i)
1998 return;
1999
2000 /* destroy free nid list */
2001 spin_lock(&nm_i->free_nid_list_lock);
2002 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2003 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2004 __del_from_free_nid_list(nm_i, i);
2005 nm_i->fcnt--;
2006 spin_unlock(&nm_i->free_nid_list_lock);
2007 kmem_cache_free(free_nid_slab, i);
2008 spin_lock(&nm_i->free_nid_list_lock);
2009 }
2010 f2fs_bug_on(sbi, nm_i->fcnt);
2011 spin_unlock(&nm_i->free_nid_list_lock);
2012
2013 /* destroy nat cache */
2014 down_write(&nm_i->nat_tree_lock);
2015 while ((found = __gang_lookup_nat_cache(nm_i,
2016 nid, NATVEC_SIZE, natvec))) {
2017 unsigned idx;
2018
2019 nid = nat_get_nid(natvec[found - 1]) + 1;
2020 for (idx = 0; idx < found; idx++)
2021 __del_from_nat_cache(nm_i, natvec[idx]);
2022 }
2023 f2fs_bug_on(sbi, nm_i->nat_cnt);
2024
2025 /* destroy nat set cache */
2026 nid = 0;
2027 while ((found = __gang_lookup_nat_set(nm_i,
2028 nid, SETVEC_SIZE, setvec))) {
2029 unsigned idx;
2030
2031 nid = setvec[found - 1]->set + 1;
2032 for (idx = 0; idx < found; idx++) {
2033 /* entry_cnt is not zero, when cp_error was occurred */
2034 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2035 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2036 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2037 }
2038 }
2039 up_write(&nm_i->nat_tree_lock);
2040
2041 kfree(nm_i->nat_bitmap);
2042 sbi->nm_info = NULL;
2043 kfree(nm_i);
2044 }
2045
2046 int __init create_node_manager_caches(void)
2047 {
2048 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2049 sizeof(struct nat_entry));
2050 if (!nat_entry_slab)
2051 goto fail;
2052
2053 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2054 sizeof(struct free_nid));
2055 if (!free_nid_slab)
2056 goto destroy_nat_entry;
2057
2058 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2059 sizeof(struct nat_entry_set));
2060 if (!nat_entry_set_slab)
2061 goto destroy_free_nid;
2062 return 0;
2063
2064 destroy_free_nid:
2065 kmem_cache_destroy(free_nid_slab);
2066 destroy_nat_entry:
2067 kmem_cache_destroy(nat_entry_slab);
2068 fail:
2069 return -ENOMEM;
2070 }
2071
2072 void destroy_node_manager_caches(void)
2073 {
2074 kmem_cache_destroy(nat_entry_set_slab);
2075 kmem_cache_destroy(free_nid_slab);
2076 kmem_cache_destroy(nat_entry_slab);
2077 }