projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: avoid unnecessary updating inode during fsync
[deliverable/linux.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_SHIFT;
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 PAGE_SHIFT;
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 } else if (type == DIRTY_DENTS) {
56 if (sbi->sb->s_bdi->wb.dirty_exceeded)
57 return false;
58 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
59 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
60 } else if (type == INO_ENTRIES) {
61 int i;
62
63 for (i = 0; i <= UPDATE_INO; i++)
64 mem_size += (sbi->im[i].ino_num *
65 sizeof(struct ino_entry)) >> PAGE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
67 } else if (type == EXTENT_CACHE) {
68 mem_size = (atomic_read(&sbi->total_ext_tree) *
69 sizeof(struct extent_tree) +
70 atomic_read(&sbi->total_ext_node) *
71 sizeof(struct extent_node)) >> PAGE_SHIFT;
72 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
73 } else {
74 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
75 return true;
76 }
77 return res;
78 }
79
80 static void clear_node_page_dirty(struct page *page)
81 {
82 struct address_space *mapping = page->mapping;
83 unsigned int long flags;
84
85 if (PageDirty(page)) {
86 spin_lock_irqsave(&mapping->tree_lock, flags);
87 radix_tree_tag_clear(&mapping->page_tree,
88 page_index(page),
89 PAGECACHE_TAG_DIRTY);
90 spin_unlock_irqrestore(&mapping->tree_lock, flags);
91
92 clear_page_dirty_for_io(page);
93 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
94 }
95 ClearPageUptodate(page);
96 }
97
98 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
99 {
100 pgoff_t index = current_nat_addr(sbi, nid);
101 return get_meta_page(sbi, index);
102 }
103
104 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
105 {
106 struct page *src_page;
107 struct page *dst_page;
108 pgoff_t src_off;
109 pgoff_t dst_off;
110 void *src_addr;
111 void *dst_addr;
112 struct f2fs_nm_info *nm_i = NM_I(sbi);
113
114 src_off = current_nat_addr(sbi, nid);
115 dst_off = next_nat_addr(sbi, src_off);
116
117 /* get current nat block page with lock */
118 src_page = get_meta_page(sbi, src_off);
119 dst_page = grab_meta_page(sbi, dst_off);
120 f2fs_bug_on(sbi, PageDirty(src_page));
121
122 src_addr = page_address(src_page);
123 dst_addr = page_address(dst_page);
124 memcpy(dst_addr, src_addr, PAGE_SIZE);
125 set_page_dirty(dst_page);
126 f2fs_put_page(src_page, 1);
127
128 set_to_next_nat(nm_i, nid);
129
130 return dst_page;
131 }
132
133 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
134 {
135 return radix_tree_lookup(&nm_i->nat_root, n);
136 }
137
138 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
139 nid_t start, unsigned int nr, struct nat_entry **ep)
140 {
141 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
142 }
143
144 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
145 {
146 list_del(&e->list);
147 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
148 nm_i->nat_cnt--;
149 kmem_cache_free(nat_entry_slab, e);
150 }
151
152 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
153 struct nat_entry *ne)
154 {
155 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
156 struct nat_entry_set *head;
157
158 if (get_nat_flag(ne, IS_DIRTY))
159 return;
160
161 head = radix_tree_lookup(&nm_i->nat_set_root, set);
162 if (!head) {
163 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
164
165 INIT_LIST_HEAD(&head->entry_list);
166 INIT_LIST_HEAD(&head->set_list);
167 head->set = set;
168 head->entry_cnt = 0;
169 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
170 }
171 list_move_tail(&ne->list, &head->entry_list);
172 nm_i->dirty_nat_cnt++;
173 head->entry_cnt++;
174 set_nat_flag(ne, IS_DIRTY, true);
175 }
176
177 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
178 struct nat_entry *ne)
179 {
180 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
181 struct nat_entry_set *head;
182
183 head = radix_tree_lookup(&nm_i->nat_set_root, set);
184 if (head) {
185 list_move_tail(&ne->list, &nm_i->nat_entries);
186 set_nat_flag(ne, IS_DIRTY, false);
187 head->entry_cnt--;
188 nm_i->dirty_nat_cnt--;
189 }
190 }
191
192 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
193 nid_t start, unsigned int nr, struct nat_entry_set **ep)
194 {
195 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
196 start, nr);
197 }
198
199 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
200 {
201 struct f2fs_nm_info *nm_i = NM_I(sbi);
202 struct nat_entry *e;
203 bool need = false;
204
205 down_read(&nm_i->nat_tree_lock);
206 e = __lookup_nat_cache(nm_i, nid);
207 if (e) {
208 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
209 !get_nat_flag(e, HAS_FSYNCED_INODE))
210 need = true;
211 }
212 up_read(&nm_i->nat_tree_lock);
213 return need;
214 }
215
216 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
217 {
218 struct f2fs_nm_info *nm_i = NM_I(sbi);
219 struct nat_entry *e;
220 bool is_cp = true;
221
222 down_read(&nm_i->nat_tree_lock);
223 e = __lookup_nat_cache(nm_i, nid);
224 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
225 is_cp = false;
226 up_read(&nm_i->nat_tree_lock);
227 return is_cp;
228 }
229
230 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
231 {
232 struct f2fs_nm_info *nm_i = NM_I(sbi);
233 struct nat_entry *e;
234 bool need_update = true;
235
236 down_read(&nm_i->nat_tree_lock);
237 e = __lookup_nat_cache(nm_i, ino);
238 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
239 (get_nat_flag(e, IS_CHECKPOINTED) ||
240 get_nat_flag(e, HAS_FSYNCED_INODE)))
241 need_update = false;
242 up_read(&nm_i->nat_tree_lock);
243 return need_update;
244 }
245
246 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
247 {
248 struct nat_entry *new;
249
250 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
251 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
252 memset(new, 0, sizeof(struct nat_entry));
253 nat_set_nid(new, nid);
254 nat_reset_flag(new);
255 list_add_tail(&new->list, &nm_i->nat_entries);
256 nm_i->nat_cnt++;
257 return new;
258 }
259
260 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
261 struct f2fs_nat_entry *ne)
262 {
263 struct f2fs_nm_info *nm_i = NM_I(sbi);
264 struct nat_entry *e;
265
266 e = __lookup_nat_cache(nm_i, nid);
267 if (!e) {
268 e = grab_nat_entry(nm_i, nid);
269 node_info_from_raw_nat(&e->ni, ne);
270 } else {
271 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
272 nat_get_blkaddr(e) != ne->block_addr ||
273 nat_get_version(e) != ne->version);
274 }
275 }
276
277 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
278 block_t new_blkaddr, bool fsync_done)
279 {
280 struct f2fs_nm_info *nm_i = NM_I(sbi);
281 struct nat_entry *e;
282
283 down_write(&nm_i->nat_tree_lock);
284 e = __lookup_nat_cache(nm_i, ni->nid);
285 if (!e) {
286 e = grab_nat_entry(nm_i, ni->nid);
287 copy_node_info(&e->ni, ni);
288 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
289 } else if (new_blkaddr == NEW_ADDR) {
290 /*
291 * when nid is reallocated,
292 * previous nat entry can be remained in nat cache.
293 * So, reinitialize it with new information.
294 */
295 copy_node_info(&e->ni, ni);
296 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
297 }
298
299 /* sanity check */
300 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
301 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
302 new_blkaddr == NULL_ADDR);
303 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
304 new_blkaddr == NEW_ADDR);
305 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
306 nat_get_blkaddr(e) != NULL_ADDR &&
307 new_blkaddr == NEW_ADDR);
308
309 /* increment version no as node is removed */
310 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
311 unsigned char version = nat_get_version(e);
312 nat_set_version(e, inc_node_version(version));
313
314 /* in order to reuse the nid */
315 if (nm_i->next_scan_nid > ni->nid)
316 nm_i->next_scan_nid = ni->nid;
317 }
318
319 /* change address */
320 nat_set_blkaddr(e, new_blkaddr);
321 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
322 set_nat_flag(e, IS_CHECKPOINTED, false);
323 __set_nat_cache_dirty(nm_i, e);
324
325 /* update fsync_mark if its inode nat entry is still alive */
326 if (ni->nid != ni->ino)
327 e = __lookup_nat_cache(nm_i, ni->ino);
328 if (e) {
329 if (fsync_done && ni->nid == ni->ino)
330 set_nat_flag(e, HAS_FSYNCED_INODE, true);
331 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
332 }
333 up_write(&nm_i->nat_tree_lock);
334 }
335
336 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
337 {
338 struct f2fs_nm_info *nm_i = NM_I(sbi);
339 int nr = nr_shrink;
340
341 if (!down_write_trylock(&nm_i->nat_tree_lock))
342 return 0;
343
344 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
345 struct nat_entry *ne;
346 ne = list_first_entry(&nm_i->nat_entries,
347 struct nat_entry, list);
348 __del_from_nat_cache(nm_i, ne);
349 nr_shrink--;
350 }
351 up_write(&nm_i->nat_tree_lock);
352 return nr - nr_shrink;
353 }
354
355 /*
356 * This function always returns success
357 */
358 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
359 {
360 struct f2fs_nm_info *nm_i = NM_I(sbi);
361 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
362 struct f2fs_journal *journal = curseg->journal;
363 nid_t start_nid = START_NID(nid);
364 struct f2fs_nat_block *nat_blk;
365 struct page *page = NULL;
366 struct f2fs_nat_entry ne;
367 struct nat_entry *e;
368 int i;
369
370 ni->nid = nid;
371
372 /* Check nat cache */
373 down_read(&nm_i->nat_tree_lock);
374 e = __lookup_nat_cache(nm_i, nid);
375 if (e) {
376 ni->ino = nat_get_ino(e);
377 ni->blk_addr = nat_get_blkaddr(e);
378 ni->version = nat_get_version(e);
379 up_read(&nm_i->nat_tree_lock);
380 return;
381 }
382
383 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
384
385 /* Check current segment summary */
386 down_read(&curseg->journal_rwsem);
387 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
388 if (i >= 0) {
389 ne = nat_in_journal(journal, i);
390 node_info_from_raw_nat(ni, &ne);
391 }
392 up_read(&curseg->journal_rwsem);
393 if (i >= 0)
394 goto cache;
395
396 /* Fill node_info from nat page */
397 page = get_current_nat_page(sbi, start_nid);
398 nat_blk = (struct f2fs_nat_block *)page_address(page);
399 ne = nat_blk->entries[nid - start_nid];
400 node_info_from_raw_nat(ni, &ne);
401 f2fs_put_page(page, 1);
402 cache:
403 up_read(&nm_i->nat_tree_lock);
404 /* cache nat entry */
405 down_write(&nm_i->nat_tree_lock);
406 cache_nat_entry(sbi, nid, &ne);
407 up_write(&nm_i->nat_tree_lock);
408 }
409
410 /*
411 * readahead MAX_RA_NODE number of node pages.
412 */
413 static void ra_node_pages(struct page *parent, int start, int n)
414 {
415 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
416 struct blk_plug plug;
417 int i, end;
418 nid_t nid;
419
420 blk_start_plug(&plug);
421
422 /* Then, try readahead for siblings of the desired node */
423 end = start + n;
424 end = min(end, NIDS_PER_BLOCK);
425 for (i = start; i < end; i++) {
426 nid = get_nid(parent, i, false);
427 ra_node_page(sbi, nid);
428 }
429
430 blk_finish_plug(&plug);
431 }
432
433 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
434 {
435 const long direct_index = ADDRS_PER_INODE(dn->inode);
436 const long direct_blks = ADDRS_PER_BLOCK;
437 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
438 unsigned int skipped_unit = ADDRS_PER_BLOCK;
439 int cur_level = dn->cur_level;
440 int max_level = dn->max_level;
441 pgoff_t base = 0;
442
443 if (!dn->max_level)
444 return pgofs + 1;
445
446 while (max_level-- > cur_level)
447 skipped_unit *= NIDS_PER_BLOCK;
448
449 switch (dn->max_level) {
450 case 3:
451 base += 2 * indirect_blks;
452 case 2:
453 base += 2 * direct_blks;
454 case 1:
455 base += direct_index;
456 break;
457 default:
458 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
459 }
460
461 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
462 }
463
464 /*
465 * The maximum depth is four.
466 * Offset[0] will have raw inode offset.
467 */
468 static int get_node_path(struct inode *inode, long block,
469 int offset[4], unsigned int noffset[4])
470 {
471 const long direct_index = ADDRS_PER_INODE(inode);
472 const long direct_blks = ADDRS_PER_BLOCK;
473 const long dptrs_per_blk = NIDS_PER_BLOCK;
474 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
475 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
476 int n = 0;
477 int level = 0;
478
479 noffset[0] = 0;
480
481 if (block < direct_index) {
482 offset[n] = block;
483 goto got;
484 }
485 block -= direct_index;
486 if (block < direct_blks) {
487 offset[n++] = NODE_DIR1_BLOCK;
488 noffset[n] = 1;
489 offset[n] = block;
490 level = 1;
491 goto got;
492 }
493 block -= direct_blks;
494 if (block < direct_blks) {
495 offset[n++] = NODE_DIR2_BLOCK;
496 noffset[n] = 2;
497 offset[n] = block;
498 level = 1;
499 goto got;
500 }
501 block -= direct_blks;
502 if (block < indirect_blks) {
503 offset[n++] = NODE_IND1_BLOCK;
504 noffset[n] = 3;
505 offset[n++] = block / direct_blks;
506 noffset[n] = 4 + offset[n - 1];
507 offset[n] = block % direct_blks;
508 level = 2;
509 goto got;
510 }
511 block -= indirect_blks;
512 if (block < indirect_blks) {
513 offset[n++] = NODE_IND2_BLOCK;
514 noffset[n] = 4 + dptrs_per_blk;
515 offset[n++] = block / direct_blks;
516 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
517 offset[n] = block % direct_blks;
518 level = 2;
519 goto got;
520 }
521 block -= indirect_blks;
522 if (block < dindirect_blks) {
523 offset[n++] = NODE_DIND_BLOCK;
524 noffset[n] = 5 + (dptrs_per_blk * 2);
525 offset[n++] = block / indirect_blks;
526 noffset[n] = 6 + (dptrs_per_blk * 2) +
527 offset[n - 1] * (dptrs_per_blk + 1);
528 offset[n++] = (block / direct_blks) % dptrs_per_blk;
529 noffset[n] = 7 + (dptrs_per_blk * 2) +
530 offset[n - 2] * (dptrs_per_blk + 1) +
531 offset[n - 1];
532 offset[n] = block % direct_blks;
533 level = 3;
534 goto got;
535 } else {
536 BUG();
537 }
538 got:
539 return level;
540 }
541
542 /*
543 * Caller should call f2fs_put_dnode(dn).
544 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
545 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
546 * In the case of RDONLY_NODE, we don't need to care about mutex.
547 */
548 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
549 {
550 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
551 struct page *npage[4];
552 struct page *parent = NULL;
553 int offset[4];
554 unsigned int noffset[4];
555 nid_t nids[4];
556 int level, i = 0;
557 int err = 0;
558
559 level = get_node_path(dn->inode, index, offset, noffset);
560
561 nids[0] = dn->inode->i_ino;
562 npage[0] = dn->inode_page;
563
564 if (!npage[0]) {
565 npage[0] = get_node_page(sbi, nids[0]);
566 if (IS_ERR(npage[0]))
567 return PTR_ERR(npage[0]);
568 }
569
570 /* if inline_data is set, should not report any block indices */
571 if (f2fs_has_inline_data(dn->inode) && index) {
572 err = -ENOENT;
573 f2fs_put_page(npage[0], 1);
574 goto release_out;
575 }
576
577 parent = npage[0];
578 if (level != 0)
579 nids[1] = get_nid(parent, offset[0], true);
580 dn->inode_page = npage[0];
581 dn->inode_page_locked = true;
582
583 /* get indirect or direct nodes */
584 for (i = 1; i <= level; i++) {
585 bool done = false;
586
587 if (!nids[i] && mode == ALLOC_NODE) {
588 /* alloc new node */
589 if (!alloc_nid(sbi, &(nids[i]))) {
590 err = -ENOSPC;
591 goto release_pages;
592 }
593
594 dn->nid = nids[i];
595 npage[i] = new_node_page(dn, noffset[i], NULL);
596 if (IS_ERR(npage[i])) {
597 alloc_nid_failed(sbi, nids[i]);
598 err = PTR_ERR(npage[i]);
599 goto release_pages;
600 }
601
602 set_nid(parent, offset[i - 1], nids[i], i == 1);
603 alloc_nid_done(sbi, nids[i]);
604 done = true;
605 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
606 npage[i] = get_node_page_ra(parent, offset[i - 1]);
607 if (IS_ERR(npage[i])) {
608 err = PTR_ERR(npage[i]);
609 goto release_pages;
610 }
611 done = true;
612 }
613 if (i == 1) {
614 dn->inode_page_locked = false;
615 unlock_page(parent);
616 } else {
617 f2fs_put_page(parent, 1);
618 }
619
620 if (!done) {
621 npage[i] = get_node_page(sbi, nids[i]);
622 if (IS_ERR(npage[i])) {
623 err = PTR_ERR(npage[i]);
624 f2fs_put_page(npage[0], 0);
625 goto release_out;
626 }
627 }
628 if (i < level) {
629 parent = npage[i];
630 nids[i + 1] = get_nid(parent, offset[i], false);
631 }
632 }
633 dn->nid = nids[level];
634 dn->ofs_in_node = offset[level];
635 dn->node_page = npage[level];
636 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
637 return 0;
638
639 release_pages:
640 f2fs_put_page(parent, 1);
641 if (i > 1)
642 f2fs_put_page(npage[0], 0);
643 release_out:
644 dn->inode_page = NULL;
645 dn->node_page = NULL;
646 if (err == -ENOENT) {
647 dn->cur_level = i;
648 dn->max_level = level;
649 }
650 return err;
651 }
652
653 static void truncate_node(struct dnode_of_data *dn)
654 {
655 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
656 struct node_info ni;
657
658 get_node_info(sbi, dn->nid, &ni);
659 if (dn->inode->i_blocks == 0) {
660 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
661 goto invalidate;
662 }
663 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
664
665 /* Deallocate node address */
666 invalidate_blocks(sbi, ni.blk_addr);
667 dec_valid_node_count(sbi, dn->inode);
668 set_node_addr(sbi, &ni, NULL_ADDR, false);
669
670 if (dn->nid == dn->inode->i_ino) {
671 remove_orphan_inode(sbi, dn->nid);
672 dec_valid_inode_count(sbi);
673 f2fs_inode_synced(dn->inode);
674 }
675 invalidate:
676 clear_node_page_dirty(dn->node_page);
677 set_sbi_flag(sbi, SBI_IS_DIRTY);
678
679 f2fs_put_page(dn->node_page, 1);
680
681 invalidate_mapping_pages(NODE_MAPPING(sbi),
682 dn->node_page->index, dn->node_page->index);
683
684 dn->node_page = NULL;
685 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
686 }
687
688 static int truncate_dnode(struct dnode_of_data *dn)
689 {
690 struct page *page;
691
692 if (dn->nid == 0)
693 return 1;
694
695 /* get direct node */
696 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
697 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
698 return 1;
699 else if (IS_ERR(page))
700 return PTR_ERR(page);
701
702 /* Make dnode_of_data for parameter */
703 dn->node_page = page;
704 dn->ofs_in_node = 0;
705 truncate_data_blocks(dn);
706 truncate_node(dn);
707 return 1;
708 }
709
710 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
711 int ofs, int depth)
712 {
713 struct dnode_of_data rdn = *dn;
714 struct page *page;
715 struct f2fs_node *rn;
716 nid_t child_nid;
717 unsigned int child_nofs;
718 int freed = 0;
719 int i, ret;
720
721 if (dn->nid == 0)
722 return NIDS_PER_BLOCK + 1;
723
724 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
725
726 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
727 if (IS_ERR(page)) {
728 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
729 return PTR_ERR(page);
730 }
731
732 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
733
734 rn = F2FS_NODE(page);
735 if (depth < 3) {
736 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
737 child_nid = le32_to_cpu(rn->in.nid[i]);
738 if (child_nid == 0)
739 continue;
740 rdn.nid = child_nid;
741 ret = truncate_dnode(&rdn);
742 if (ret < 0)
743 goto out_err;
744 if (set_nid(page, i, 0, false))
745 dn->node_changed = true;
746 }
747 } else {
748 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
749 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
750 child_nid = le32_to_cpu(rn->in.nid[i]);
751 if (child_nid == 0) {
752 child_nofs += NIDS_PER_BLOCK + 1;
753 continue;
754 }
755 rdn.nid = child_nid;
756 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
757 if (ret == (NIDS_PER_BLOCK + 1)) {
758 if (set_nid(page, i, 0, false))
759 dn->node_changed = true;
760 child_nofs += ret;
761 } else if (ret < 0 && ret != -ENOENT) {
762 goto out_err;
763 }
764 }
765 freed = child_nofs;
766 }
767
768 if (!ofs) {
769 /* remove current indirect node */
770 dn->node_page = page;
771 truncate_node(dn);
772 freed++;
773 } else {
774 f2fs_put_page(page, 1);
775 }
776 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
777 return freed;
778
779 out_err:
780 f2fs_put_page(page, 1);
781 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
782 return ret;
783 }
784
785 static int truncate_partial_nodes(struct dnode_of_data *dn,
786 struct f2fs_inode *ri, int *offset, int depth)
787 {
788 struct page *pages[2];
789 nid_t nid[3];
790 nid_t child_nid;
791 int err = 0;
792 int i;
793 int idx = depth - 2;
794
795 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
796 if (!nid[0])
797 return 0;
798
799 /* get indirect nodes in the path */
800 for (i = 0; i < idx + 1; i++) {
801 /* reference count'll be increased */
802 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
803 if (IS_ERR(pages[i])) {
804 err = PTR_ERR(pages[i]);
805 idx = i - 1;
806 goto fail;
807 }
808 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
809 }
810
811 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
812
813 /* free direct nodes linked to a partial indirect node */
814 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
815 child_nid = get_nid(pages[idx], i, false);
816 if (!child_nid)
817 continue;
818 dn->nid = child_nid;
819 err = truncate_dnode(dn);
820 if (err < 0)
821 goto fail;
822 if (set_nid(pages[idx], i, 0, false))
823 dn->node_changed = true;
824 }
825
826 if (offset[idx + 1] == 0) {
827 dn->node_page = pages[idx];
828 dn->nid = nid[idx];
829 truncate_node(dn);
830 } else {
831 f2fs_put_page(pages[idx], 1);
832 }
833 offset[idx]++;
834 offset[idx + 1] = 0;
835 idx--;
836 fail:
837 for (i = idx; i >= 0; i--)
838 f2fs_put_page(pages[i], 1);
839
840 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
841
842 return err;
843 }
844
845 /*
846 * All the block addresses of data and nodes should be nullified.
847 */
848 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
849 {
850 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
851 int err = 0, cont = 1;
852 int level, offset[4], noffset[4];
853 unsigned int nofs = 0;
854 struct f2fs_inode *ri;
855 struct dnode_of_data dn;
856 struct page *page;
857
858 trace_f2fs_truncate_inode_blocks_enter(inode, from);
859
860 level = get_node_path(inode, from, offset, noffset);
861
862 page = get_node_page(sbi, inode->i_ino);
863 if (IS_ERR(page)) {
864 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
865 return PTR_ERR(page);
866 }
867
868 set_new_dnode(&dn, inode, page, NULL, 0);
869 unlock_page(page);
870
871 ri = F2FS_INODE(page);
872 switch (level) {
873 case 0:
874 case 1:
875 nofs = noffset[1];
876 break;
877 case 2:
878 nofs = noffset[1];
879 if (!offset[level - 1])
880 goto skip_partial;
881 err = truncate_partial_nodes(&dn, ri, offset, level);
882 if (err < 0 && err != -ENOENT)
883 goto fail;
884 nofs += 1 + NIDS_PER_BLOCK;
885 break;
886 case 3:
887 nofs = 5 + 2 * NIDS_PER_BLOCK;
888 if (!offset[level - 1])
889 goto skip_partial;
890 err = truncate_partial_nodes(&dn, ri, offset, level);
891 if (err < 0 && err != -ENOENT)
892 goto fail;
893 break;
894 default:
895 BUG();
896 }
897
898 skip_partial:
899 while (cont) {
900 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
901 switch (offset[0]) {
902 case NODE_DIR1_BLOCK:
903 case NODE_DIR2_BLOCK:
904 err = truncate_dnode(&dn);
905 break;
906
907 case NODE_IND1_BLOCK:
908 case NODE_IND2_BLOCK:
909 err = truncate_nodes(&dn, nofs, offset[1], 2);
910 break;
911
912 case NODE_DIND_BLOCK:
913 err = truncate_nodes(&dn, nofs, offset[1], 3);
914 cont = 0;
915 break;
916
917 default:
918 BUG();
919 }
920 if (err < 0 && err != -ENOENT)
921 goto fail;
922 if (offset[1] == 0 &&
923 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
924 lock_page(page);
925 BUG_ON(page->mapping != NODE_MAPPING(sbi));
926 f2fs_wait_on_page_writeback(page, NODE, true);
927 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
928 set_page_dirty(page);
929 unlock_page(page);
930 }
931 offset[1] = 0;
932 offset[0]++;
933 nofs += err;
934 }
935 fail:
936 f2fs_put_page(page, 0);
937 trace_f2fs_truncate_inode_blocks_exit(inode, err);
938 return err > 0 ? 0 : err;
939 }
940
941 int truncate_xattr_node(struct inode *inode, struct page *page)
942 {
943 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
944 nid_t nid = F2FS_I(inode)->i_xattr_nid;
945 struct dnode_of_data dn;
946 struct page *npage;
947
948 if (!nid)
949 return 0;
950
951 npage = get_node_page(sbi, nid);
952 if (IS_ERR(npage))
953 return PTR_ERR(npage);
954
955 f2fs_i_xnid_write(inode, 0);
956
957 /* need to do checkpoint during fsync */
958 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
959
960 set_new_dnode(&dn, inode, page, npage, nid);
961
962 if (page)
963 dn.inode_page_locked = true;
964 truncate_node(&dn);
965 return 0;
966 }
967
968 /*
969 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
970 * f2fs_unlock_op().
971 */
972 int remove_inode_page(struct inode *inode)
973 {
974 struct dnode_of_data dn;
975 int err;
976
977 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
978 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
979 if (err)
980 return err;
981
982 err = truncate_xattr_node(inode, dn.inode_page);
983 if (err) {
984 f2fs_put_dnode(&dn);
985 return err;
986 }
987
988 /* remove potential inline_data blocks */
989 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
990 S_ISLNK(inode->i_mode))
991 truncate_data_blocks_range(&dn, 1);
992
993 /* 0 is possible, after f2fs_new_inode() has failed */
994 f2fs_bug_on(F2FS_I_SB(inode),
995 inode->i_blocks != 0 && inode->i_blocks != 1);
996
997 /* will put inode & node pages */
998 truncate_node(&dn);
999 return 0;
1000 }
1001
1002 struct page *new_inode_page(struct inode *inode)
1003 {
1004 struct dnode_of_data dn;
1005
1006 /* allocate inode page for new inode */
1007 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1008
1009 /* caller should f2fs_put_page(page, 1); */
1010 return new_node_page(&dn, 0, NULL);
1011 }
1012
1013 struct page *new_node_page(struct dnode_of_data *dn,
1014 unsigned int ofs, struct page *ipage)
1015 {
1016 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1017 struct node_info old_ni, new_ni;
1018 struct page *page;
1019 int err;
1020
1021 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1022 return ERR_PTR(-EPERM);
1023
1024 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1025 if (!page)
1026 return ERR_PTR(-ENOMEM);
1027
1028 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1029 err = -ENOSPC;
1030 goto fail;
1031 }
1032
1033 get_node_info(sbi, dn->nid, &old_ni);
1034
1035 /* Reinitialize old_ni with new node page */
1036 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1037 new_ni = old_ni;
1038 new_ni.ino = dn->inode->i_ino;
1039 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1040
1041 f2fs_wait_on_page_writeback(page, NODE, true);
1042 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1043 set_cold_node(dn->inode, page);
1044 SetPageUptodate(page);
1045 if (set_page_dirty(page))
1046 dn->node_changed = true;
1047
1048 if (f2fs_has_xattr_block(ofs))
1049 f2fs_i_xnid_write(dn->inode, dn->nid);
1050
1051 if (ofs == 0)
1052 inc_valid_inode_count(sbi);
1053 return page;
1054
1055 fail:
1056 clear_node_page_dirty(page);
1057 f2fs_put_page(page, 1);
1058 return ERR_PTR(err);
1059 }
1060
1061 /*
1062 * Caller should do after getting the following values.
1063 * 0: f2fs_put_page(page, 0)
1064 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1065 */
1066 static int read_node_page(struct page *page, int rw)
1067 {
1068 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1069 struct node_info ni;
1070 struct f2fs_io_info fio = {
1071 .sbi = sbi,
1072 .type = NODE,
1073 .rw = rw,
1074 .page = page,
1075 .encrypted_page = NULL,
1076 };
1077
1078 get_node_info(sbi, page->index, &ni);
1079
1080 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1081 ClearPageUptodate(page);
1082 return -ENOENT;
1083 }
1084
1085 if (PageUptodate(page))
1086 return LOCKED_PAGE;
1087
1088 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1089 return f2fs_submit_page_bio(&fio);
1090 }
1091
1092 /*
1093 * Readahead a node page
1094 */
1095 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1096 {
1097 struct page *apage;
1098 int err;
1099
1100 if (!nid)
1101 return;
1102 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1103
1104 rcu_read_lock();
1105 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1106 rcu_read_unlock();
1107 if (apage)
1108 return;
1109
1110 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1111 if (!apage)
1112 return;
1113
1114 err = read_node_page(apage, READA);
1115 f2fs_put_page(apage, err ? 1 : 0);
1116 }
1117
1118 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1119 struct page *parent, int start)
1120 {
1121 struct page *page;
1122 int err;
1123
1124 if (!nid)
1125 return ERR_PTR(-ENOENT);
1126 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1127 repeat:
1128 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1129 if (!page)
1130 return ERR_PTR(-ENOMEM);
1131
1132 err = read_node_page(page, READ_SYNC);
1133 if (err < 0) {
1134 f2fs_put_page(page, 1);
1135 return ERR_PTR(err);
1136 } else if (err == LOCKED_PAGE) {
1137 goto page_hit;
1138 }
1139
1140 if (parent)
1141 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1142
1143 lock_page(page);
1144
1145 if (unlikely(!PageUptodate(page))) {
1146 f2fs_put_page(page, 1);
1147 return ERR_PTR(-EIO);
1148 }
1149 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1150 f2fs_put_page(page, 1);
1151 goto repeat;
1152 }
1153 page_hit:
1154 f2fs_bug_on(sbi, nid != nid_of_node(page));
1155 return page;
1156 }
1157
1158 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1159 {
1160 return __get_node_page(sbi, nid, NULL, 0);
1161 }
1162
1163 struct page *get_node_page_ra(struct page *parent, int start)
1164 {
1165 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1166 nid_t nid = get_nid(parent, start, false);
1167
1168 return __get_node_page(sbi, nid, parent, start);
1169 }
1170
1171 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1172 {
1173 struct inode *inode;
1174 struct page *page;
1175 int ret;
1176
1177 /* should flush inline_data before evict_inode */
1178 inode = ilookup(sbi->sb, ino);
1179 if (!inode)
1180 return;
1181
1182 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1183 if (!page)
1184 goto iput_out;
1185
1186 if (!PageUptodate(page))
1187 goto page_out;
1188
1189 if (!PageDirty(page))
1190 goto page_out;
1191
1192 if (!clear_page_dirty_for_io(page))
1193 goto page_out;
1194
1195 ret = f2fs_write_inline_data(inode, page);
1196 inode_dec_dirty_pages(inode);
1197 if (ret)
1198 set_page_dirty(page);
1199 page_out:
1200 f2fs_put_page(page, 1);
1201 iput_out:
1202 iput(inode);
1203 }
1204
1205 void move_node_page(struct page *node_page, int gc_type)
1206 {
1207 if (gc_type == FG_GC) {
1208 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1209 struct writeback_control wbc = {
1210 .sync_mode = WB_SYNC_ALL,
1211 .nr_to_write = 1,
1212 .for_reclaim = 0,
1213 };
1214
1215 set_page_dirty(node_page);
1216 f2fs_wait_on_page_writeback(node_page, NODE, true);
1217
1218 f2fs_bug_on(sbi, PageWriteback(node_page));
1219 if (!clear_page_dirty_for_io(node_page))
1220 goto out_page;
1221
1222 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1223 unlock_page(node_page);
1224 goto release_page;
1225 } else {
1226 /* set page dirty and write it */
1227 if (!PageWriteback(node_page))
1228 set_page_dirty(node_page);
1229 }
1230 out_page:
1231 unlock_page(node_page);
1232 release_page:
1233 f2fs_put_page(node_page, 0);
1234 }
1235
1236 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1237 {
1238 pgoff_t index, end;
1239 struct pagevec pvec;
1240 struct page *last_page = NULL;
1241
1242 pagevec_init(&pvec, 0);
1243 index = 0;
1244 end = ULONG_MAX;
1245
1246 while (index <= end) {
1247 int i, nr_pages;
1248 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1249 PAGECACHE_TAG_DIRTY,
1250 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1251 if (nr_pages == 0)
1252 break;
1253
1254 for (i = 0; i < nr_pages; i++) {
1255 struct page *page = pvec.pages[i];
1256
1257 if (unlikely(f2fs_cp_error(sbi))) {
1258 f2fs_put_page(last_page, 0);
1259 pagevec_release(&pvec);
1260 return ERR_PTR(-EIO);
1261 }
1262
1263 if (!IS_DNODE(page) || !is_cold_node(page))
1264 continue;
1265 if (ino_of_node(page) != ino)
1266 continue;
1267
1268 lock_page(page);
1269
1270 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1271 continue_unlock:
1272 unlock_page(page);
1273 continue;
1274 }
1275 if (ino_of_node(page) != ino)
1276 goto continue_unlock;
1277
1278 if (!PageDirty(page)) {
1279 /* someone wrote it for us */
1280 goto continue_unlock;
1281 }
1282
1283 if (last_page)
1284 f2fs_put_page(last_page, 0);
1285
1286 get_page(page);
1287 last_page = page;
1288 unlock_page(page);
1289 }
1290 pagevec_release(&pvec);
1291 cond_resched();
1292 }
1293 return last_page;
1294 }
1295
1296 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1297 struct writeback_control *wbc, bool atomic)
1298 {
1299 pgoff_t index, end;
1300 struct pagevec pvec;
1301 int ret = 0;
1302 struct page *last_page = NULL;
1303 bool marked = false;
1304 nid_t ino = inode->i_ino;
1305
1306 if (atomic) {
1307 last_page = last_fsync_dnode(sbi, ino);
1308 if (IS_ERR_OR_NULL(last_page))
1309 return PTR_ERR_OR_ZERO(last_page);
1310 }
1311 retry:
1312 pagevec_init(&pvec, 0);
1313 index = 0;
1314 end = ULONG_MAX;
1315
1316 while (index <= end) {
1317 int i, nr_pages;
1318 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1319 PAGECACHE_TAG_DIRTY,
1320 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1321 if (nr_pages == 0)
1322 break;
1323
1324 for (i = 0; i < nr_pages; i++) {
1325 struct page *page = pvec.pages[i];
1326
1327 if (unlikely(f2fs_cp_error(sbi))) {
1328 f2fs_put_page(last_page, 0);
1329 pagevec_release(&pvec);
1330 return -EIO;
1331 }
1332
1333 if (!IS_DNODE(page) || !is_cold_node(page))
1334 continue;
1335 if (ino_of_node(page) != ino)
1336 continue;
1337
1338 lock_page(page);
1339
1340 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1341 continue_unlock:
1342 unlock_page(page);
1343 continue;
1344 }
1345 if (ino_of_node(page) != ino)
1346 goto continue_unlock;
1347
1348 if (!PageDirty(page) && page != last_page) {
1349 /* someone wrote it for us */
1350 goto continue_unlock;
1351 }
1352
1353 f2fs_wait_on_page_writeback(page, NODE, true);
1354 BUG_ON(PageWriteback(page));
1355
1356 if (!atomic || page == last_page) {
1357 set_fsync_mark(page, 1);
1358 if (IS_INODE(page)) {
1359 if (is_inode_flag_set(inode,
1360 FI_DIRTY_INODE))
1361 update_inode(inode, page);
1362 set_dentry_mark(page,
1363 need_dentry_mark(sbi, ino));
1364 }
1365 /* may be written by other thread */
1366 if (!PageDirty(page))
1367 set_page_dirty(page);
1368 }
1369
1370 if (!clear_page_dirty_for_io(page))
1371 goto continue_unlock;
1372
1373 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1374 if (ret) {
1375 unlock_page(page);
1376 f2fs_put_page(last_page, 0);
1377 break;
1378 }
1379 if (page == last_page) {
1380 f2fs_put_page(page, 0);
1381 marked = true;
1382 break;
1383 }
1384 }
1385 pagevec_release(&pvec);
1386 cond_resched();
1387
1388 if (ret || marked)
1389 break;
1390 }
1391 if (!ret && atomic && !marked) {
1392 f2fs_msg(sbi->sb, KERN_DEBUG,
1393 "Retry to write fsync mark: ino=%u, idx=%lx",
1394 ino, last_page->index);
1395 lock_page(last_page);
1396 set_page_dirty(last_page);
1397 unlock_page(last_page);
1398 goto retry;
1399 }
1400 return ret ? -EIO: 0;
1401 }
1402
1403 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1404 {
1405 pgoff_t index, end;
1406 struct pagevec pvec;
1407 int step = 0;
1408 int nwritten = 0;
1409
1410 pagevec_init(&pvec, 0);
1411
1412 next_step:
1413 index = 0;
1414 end = ULONG_MAX;
1415
1416 while (index <= end) {
1417 int i, nr_pages;
1418 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1419 PAGECACHE_TAG_DIRTY,
1420 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1421 if (nr_pages == 0)
1422 break;
1423
1424 for (i = 0; i < nr_pages; i++) {
1425 struct page *page = pvec.pages[i];
1426
1427 if (unlikely(f2fs_cp_error(sbi))) {
1428 pagevec_release(&pvec);
1429 return -EIO;
1430 }
1431
1432 /*
1433 * flushing sequence with step:
1434 * 0. indirect nodes
1435 * 1. dentry dnodes
1436 * 2. file dnodes
1437 */
1438 if (step == 0 && IS_DNODE(page))
1439 continue;
1440 if (step == 1 && (!IS_DNODE(page) ||
1441 is_cold_node(page)))
1442 continue;
1443 if (step == 2 && (!IS_DNODE(page) ||
1444 !is_cold_node(page)))
1445 continue;
1446 lock_node:
1447 if (!trylock_page(page))
1448 continue;
1449
1450 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1451 continue_unlock:
1452 unlock_page(page);
1453 continue;
1454 }
1455
1456 if (!PageDirty(page)) {
1457 /* someone wrote it for us */
1458 goto continue_unlock;
1459 }
1460
1461 /* flush inline_data */
1462 if (is_inline_node(page)) {
1463 clear_inline_node(page);
1464 unlock_page(page);
1465 flush_inline_data(sbi, ino_of_node(page));
1466 goto lock_node;
1467 }
1468
1469 f2fs_wait_on_page_writeback(page, NODE, true);
1470
1471 BUG_ON(PageWriteback(page));
1472 if (!clear_page_dirty_for_io(page))
1473 goto continue_unlock;
1474
1475 set_fsync_mark(page, 0);
1476 set_dentry_mark(page, 0);
1477
1478 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1479 unlock_page(page);
1480
1481 if (--wbc->nr_to_write == 0)
1482 break;
1483 }
1484 pagevec_release(&pvec);
1485 cond_resched();
1486
1487 if (wbc->nr_to_write == 0) {
1488 step = 2;
1489 break;
1490 }
1491 }
1492
1493 if (step < 2) {
1494 step++;
1495 goto next_step;
1496 }
1497 return nwritten;
1498 }
1499
1500 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1501 {
1502 pgoff_t index = 0, end = ULONG_MAX;
1503 struct pagevec pvec;
1504 int ret2 = 0, ret = 0;
1505
1506 pagevec_init(&pvec, 0);
1507
1508 while (index <= end) {
1509 int i, nr_pages;
1510 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1511 PAGECACHE_TAG_WRITEBACK,
1512 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1513 if (nr_pages == 0)
1514 break;
1515
1516 for (i = 0; i < nr_pages; i++) {
1517 struct page *page = pvec.pages[i];
1518
1519 /* until radix tree lookup accepts end_index */
1520 if (unlikely(page->index > end))
1521 continue;
1522
1523 if (ino && ino_of_node(page) == ino) {
1524 f2fs_wait_on_page_writeback(page, NODE, true);
1525 if (TestClearPageError(page))
1526 ret = -EIO;
1527 }
1528 }
1529 pagevec_release(&pvec);
1530 cond_resched();
1531 }
1532
1533 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1534 ret2 = -ENOSPC;
1535 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1536 ret2 = -EIO;
1537 if (!ret)
1538 ret = ret2;
1539 return ret;
1540 }
1541
1542 static int f2fs_write_node_page(struct page *page,
1543 struct writeback_control *wbc)
1544 {
1545 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1546 nid_t nid;
1547 struct node_info ni;
1548 struct f2fs_io_info fio = {
1549 .sbi = sbi,
1550 .type = NODE,
1551 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1552 .page = page,
1553 .encrypted_page = NULL,
1554 };
1555
1556 trace_f2fs_writepage(page, NODE);
1557
1558 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1559 goto redirty_out;
1560 if (unlikely(f2fs_cp_error(sbi)))
1561 goto redirty_out;
1562
1563 /* get old block addr of this node page */
1564 nid = nid_of_node(page);
1565 f2fs_bug_on(sbi, page->index != nid);
1566
1567 if (wbc->for_reclaim) {
1568 if (!down_read_trylock(&sbi->node_write))
1569 goto redirty_out;
1570 } else {
1571 down_read(&sbi->node_write);
1572 }
1573
1574 get_node_info(sbi, nid, &ni);
1575
1576 /* This page is already truncated */
1577 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1578 ClearPageUptodate(page);
1579 dec_page_count(sbi, F2FS_DIRTY_NODES);
1580 up_read(&sbi->node_write);
1581 unlock_page(page);
1582 return 0;
1583 }
1584
1585 set_page_writeback(page);
1586 fio.old_blkaddr = ni.blk_addr;
1587 write_node_page(nid, &fio);
1588 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1589 dec_page_count(sbi, F2FS_DIRTY_NODES);
1590 up_read(&sbi->node_write);
1591
1592 if (wbc->for_reclaim)
1593 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1594
1595 unlock_page(page);
1596
1597 if (unlikely(f2fs_cp_error(sbi)))
1598 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1599
1600 return 0;
1601
1602 redirty_out:
1603 redirty_page_for_writepage(wbc, page);
1604 return AOP_WRITEPAGE_ACTIVATE;
1605 }
1606
1607 static int f2fs_write_node_pages(struct address_space *mapping,
1608 struct writeback_control *wbc)
1609 {
1610 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1611 long diff;
1612
1613 /* balancing f2fs's metadata in background */
1614 f2fs_balance_fs_bg(sbi);
1615
1616 /* collect a number of dirty node pages and write together */
1617 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1618 goto skip_write;
1619
1620 trace_f2fs_writepages(mapping->host, wbc, NODE);
1621
1622 diff = nr_pages_to_write(sbi, NODE, wbc);
1623 wbc->sync_mode = WB_SYNC_NONE;
1624 sync_node_pages(sbi, wbc);
1625 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1626 return 0;
1627
1628 skip_write:
1629 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1630 trace_f2fs_writepages(mapping->host, wbc, NODE);
1631 return 0;
1632 }
1633
1634 static int f2fs_set_node_page_dirty(struct page *page)
1635 {
1636 trace_f2fs_set_page_dirty(page, NODE);
1637
1638 SetPageUptodate(page);
1639 if (!PageDirty(page)) {
1640 __set_page_dirty_nobuffers(page);
1641 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1642 SetPagePrivate(page);
1643 f2fs_trace_pid(page);
1644 return 1;
1645 }
1646 return 0;
1647 }
1648
1649 /*
1650 * Structure of the f2fs node operations
1651 */
1652 const struct address_space_operations f2fs_node_aops = {
1653 .writepage = f2fs_write_node_page,
1654 .writepages = f2fs_write_node_pages,
1655 .set_page_dirty = f2fs_set_node_page_dirty,
1656 .invalidatepage = f2fs_invalidate_page,
1657 .releasepage = f2fs_release_page,
1658 };
1659
1660 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1661 nid_t n)
1662 {
1663 return radix_tree_lookup(&nm_i->free_nid_root, n);
1664 }
1665
1666 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1667 struct free_nid *i)
1668 {
1669 list_del(&i->list);
1670 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1671 }
1672
1673 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1674 {
1675 struct f2fs_nm_info *nm_i = NM_I(sbi);
1676 struct free_nid *i;
1677 struct nat_entry *ne;
1678
1679 if (!available_free_memory(sbi, FREE_NIDS))
1680 return -1;
1681
1682 /* 0 nid should not be used */
1683 if (unlikely(nid == 0))
1684 return 0;
1685
1686 if (build) {
1687 /* do not add allocated nids */
1688 ne = __lookup_nat_cache(nm_i, nid);
1689 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1690 nat_get_blkaddr(ne) != NULL_ADDR))
1691 return 0;
1692 }
1693
1694 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1695 i->nid = nid;
1696 i->state = NID_NEW;
1697
1698 if (radix_tree_preload(GFP_NOFS)) {
1699 kmem_cache_free(free_nid_slab, i);
1700 return 0;
1701 }
1702
1703 spin_lock(&nm_i->free_nid_list_lock);
1704 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1705 spin_unlock(&nm_i->free_nid_list_lock);
1706 radix_tree_preload_end();
1707 kmem_cache_free(free_nid_slab, i);
1708 return 0;
1709 }
1710 list_add_tail(&i->list, &nm_i->free_nid_list);
1711 nm_i->fcnt++;
1712 spin_unlock(&nm_i->free_nid_list_lock);
1713 radix_tree_preload_end();
1714 return 1;
1715 }
1716
1717 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1718 {
1719 struct free_nid *i;
1720 bool need_free = false;
1721
1722 spin_lock(&nm_i->free_nid_list_lock);
1723 i = __lookup_free_nid_list(nm_i, nid);
1724 if (i && i->state == NID_NEW) {
1725 __del_from_free_nid_list(nm_i, i);
1726 nm_i->fcnt--;
1727 need_free = true;
1728 }
1729 spin_unlock(&nm_i->free_nid_list_lock);
1730
1731 if (need_free)
1732 kmem_cache_free(free_nid_slab, i);
1733 }
1734
1735 static void scan_nat_page(struct f2fs_sb_info *sbi,
1736 struct page *nat_page, nid_t start_nid)
1737 {
1738 struct f2fs_nm_info *nm_i = NM_I(sbi);
1739 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1740 block_t blk_addr;
1741 int i;
1742
1743 i = start_nid % NAT_ENTRY_PER_BLOCK;
1744
1745 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1746
1747 if (unlikely(start_nid >= nm_i->max_nid))
1748 break;
1749
1750 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1751 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1752 if (blk_addr == NULL_ADDR) {
1753 if (add_free_nid(sbi, start_nid, true) < 0)
1754 break;
1755 }
1756 }
1757 }
1758
1759 static void build_free_nids(struct f2fs_sb_info *sbi)
1760 {
1761 struct f2fs_nm_info *nm_i = NM_I(sbi);
1762 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1763 struct f2fs_journal *journal = curseg->journal;
1764 int i = 0;
1765 nid_t nid = nm_i->next_scan_nid;
1766
1767 /* Enough entries */
1768 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1769 return;
1770
1771 /* readahead nat pages to be scanned */
1772 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1773 META_NAT, true);
1774
1775 down_read(&nm_i->nat_tree_lock);
1776
1777 while (1) {
1778 struct page *page = get_current_nat_page(sbi, nid);
1779
1780 scan_nat_page(sbi, page, nid);
1781 f2fs_put_page(page, 1);
1782
1783 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1784 if (unlikely(nid >= nm_i->max_nid))
1785 nid = 0;
1786
1787 if (++i >= FREE_NID_PAGES)
1788 break;
1789 }
1790
1791 /* go to the next free nat pages to find free nids abundantly */
1792 nm_i->next_scan_nid = nid;
1793
1794 /* find free nids from current sum_pages */
1795 down_read(&curseg->journal_rwsem);
1796 for (i = 0; i < nats_in_cursum(journal); i++) {
1797 block_t addr;
1798
1799 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1800 nid = le32_to_cpu(nid_in_journal(journal, i));
1801 if (addr == NULL_ADDR)
1802 add_free_nid(sbi, nid, true);
1803 else
1804 remove_free_nid(nm_i, nid);
1805 }
1806 up_read(&curseg->journal_rwsem);
1807 up_read(&nm_i->nat_tree_lock);
1808
1809 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1810 nm_i->ra_nid_pages, META_NAT, false);
1811 }
1812
1813 /*
1814 * If this function returns success, caller can obtain a new nid
1815 * from second parameter of this function.
1816 * The returned nid could be used ino as well as nid when inode is created.
1817 */
1818 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1819 {
1820 struct f2fs_nm_info *nm_i = NM_I(sbi);
1821 struct free_nid *i = NULL;
1822 retry:
1823 #ifdef CONFIG_F2FS_FAULT_INJECTION
1824 if (time_to_inject(FAULT_ALLOC_NID))
1825 return false;
1826 #endif
1827 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1828 return false;
1829
1830 spin_lock(&nm_i->free_nid_list_lock);
1831
1832 /* We should not use stale free nids created by build_free_nids */
1833 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1834 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1835 list_for_each_entry(i, &nm_i->free_nid_list, list)
1836 if (i->state == NID_NEW)
1837 break;
1838
1839 f2fs_bug_on(sbi, i->state != NID_NEW);
1840 *nid = i->nid;
1841 i->state = NID_ALLOC;
1842 nm_i->fcnt--;
1843 spin_unlock(&nm_i->free_nid_list_lock);
1844 return true;
1845 }
1846 spin_unlock(&nm_i->free_nid_list_lock);
1847
1848 /* Let's scan nat pages and its caches to get free nids */
1849 mutex_lock(&nm_i->build_lock);
1850 build_free_nids(sbi);
1851 mutex_unlock(&nm_i->build_lock);
1852 goto retry;
1853 }
1854
1855 /*
1856 * alloc_nid() should be called prior to this function.
1857 */
1858 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1859 {
1860 struct f2fs_nm_info *nm_i = NM_I(sbi);
1861 struct free_nid *i;
1862
1863 spin_lock(&nm_i->free_nid_list_lock);
1864 i = __lookup_free_nid_list(nm_i, nid);
1865 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1866 __del_from_free_nid_list(nm_i, i);
1867 spin_unlock(&nm_i->free_nid_list_lock);
1868
1869 kmem_cache_free(free_nid_slab, i);
1870 }
1871
1872 /*
1873 * alloc_nid() should be called prior to this function.
1874 */
1875 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1876 {
1877 struct f2fs_nm_info *nm_i = NM_I(sbi);
1878 struct free_nid *i;
1879 bool need_free = false;
1880
1881 if (!nid)
1882 return;
1883
1884 spin_lock(&nm_i->free_nid_list_lock);
1885 i = __lookup_free_nid_list(nm_i, nid);
1886 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1887 if (!available_free_memory(sbi, FREE_NIDS)) {
1888 __del_from_free_nid_list(nm_i, i);
1889 need_free = true;
1890 } else {
1891 i->state = NID_NEW;
1892 nm_i->fcnt++;
1893 }
1894 spin_unlock(&nm_i->free_nid_list_lock);
1895
1896 if (need_free)
1897 kmem_cache_free(free_nid_slab, i);
1898 }
1899
1900 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1901 {
1902 struct f2fs_nm_info *nm_i = NM_I(sbi);
1903 struct free_nid *i, *next;
1904 int nr = nr_shrink;
1905
1906 if (!mutex_trylock(&nm_i->build_lock))
1907 return 0;
1908
1909 spin_lock(&nm_i->free_nid_list_lock);
1910 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1911 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1912 break;
1913 if (i->state == NID_ALLOC)
1914 continue;
1915 __del_from_free_nid_list(nm_i, i);
1916 kmem_cache_free(free_nid_slab, i);
1917 nm_i->fcnt--;
1918 nr_shrink--;
1919 }
1920 spin_unlock(&nm_i->free_nid_list_lock);
1921 mutex_unlock(&nm_i->build_lock);
1922
1923 return nr - nr_shrink;
1924 }
1925
1926 void recover_inline_xattr(struct inode *inode, struct page *page)
1927 {
1928 void *src_addr, *dst_addr;
1929 size_t inline_size;
1930 struct page *ipage;
1931 struct f2fs_inode *ri;
1932
1933 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1934 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1935
1936 ri = F2FS_INODE(page);
1937 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1938 clear_inode_flag(inode, FI_INLINE_XATTR);
1939 goto update_inode;
1940 }
1941
1942 dst_addr = inline_xattr_addr(ipage);
1943 src_addr = inline_xattr_addr(page);
1944 inline_size = inline_xattr_size(inode);
1945
1946 f2fs_wait_on_page_writeback(ipage, NODE, true);
1947 memcpy(dst_addr, src_addr, inline_size);
1948 update_inode:
1949 update_inode(inode, ipage);
1950 f2fs_put_page(ipage, 1);
1951 }
1952
1953 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1954 {
1955 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1956 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1957 nid_t new_xnid = nid_of_node(page);
1958 struct node_info ni;
1959
1960 /* 1: invalidate the previous xattr nid */
1961 if (!prev_xnid)
1962 goto recover_xnid;
1963
1964 /* Deallocate node address */
1965 get_node_info(sbi, prev_xnid, &ni);
1966 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1967 invalidate_blocks(sbi, ni.blk_addr);
1968 dec_valid_node_count(sbi, inode);
1969 set_node_addr(sbi, &ni, NULL_ADDR, false);
1970
1971 recover_xnid:
1972 /* 2: allocate new xattr nid */
1973 if (unlikely(!inc_valid_node_count(sbi, inode)))
1974 f2fs_bug_on(sbi, 1);
1975
1976 remove_free_nid(NM_I(sbi), new_xnid);
1977 get_node_info(sbi, new_xnid, &ni);
1978 ni.ino = inode->i_ino;
1979 set_node_addr(sbi, &ni, NEW_ADDR, false);
1980 f2fs_i_xnid_write(inode, new_xnid);
1981
1982 /* 3: update xattr blkaddr */
1983 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1984 set_node_addr(sbi, &ni, blkaddr, false);
1985 }
1986
1987 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1988 {
1989 struct f2fs_inode *src, *dst;
1990 nid_t ino = ino_of_node(page);
1991 struct node_info old_ni, new_ni;
1992 struct page *ipage;
1993
1994 get_node_info(sbi, ino, &old_ni);
1995
1996 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1997 return -EINVAL;
1998
1999 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2000 if (!ipage)
2001 return -ENOMEM;
2002
2003 /* Should not use this inode from free nid list */
2004 remove_free_nid(NM_I(sbi), ino);
2005
2006 SetPageUptodate(ipage);
2007 fill_node_footer(ipage, ino, ino, 0, true);
2008
2009 src = F2FS_INODE(page);
2010 dst = F2FS_INODE(ipage);
2011
2012 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2013 dst->i_size = 0;
2014 dst->i_blocks = cpu_to_le64(1);
2015 dst->i_links = cpu_to_le32(1);
2016 dst->i_xattr_nid = 0;
2017 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2018
2019 new_ni = old_ni;
2020 new_ni.ino = ino;
2021
2022 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2023 WARN_ON(1);
2024 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2025 inc_valid_inode_count(sbi);
2026 set_page_dirty(ipage);
2027 f2fs_put_page(ipage, 1);
2028 return 0;
2029 }
2030
2031 int restore_node_summary(struct f2fs_sb_info *sbi,
2032 unsigned int segno, struct f2fs_summary_block *sum)
2033 {
2034 struct f2fs_node *rn;
2035 struct f2fs_summary *sum_entry;
2036 block_t addr;
2037 int bio_blocks = MAX_BIO_BLOCKS(sbi);
2038 int i, idx, last_offset, nrpages;
2039
2040 /* scan the node segment */
2041 last_offset = sbi->blocks_per_seg;
2042 addr = START_BLOCK(sbi, segno);
2043 sum_entry = &sum->entries[0];
2044
2045 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2046 nrpages = min(last_offset - i, bio_blocks);
2047
2048 /* readahead node pages */
2049 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2050
2051 for (idx = addr; idx < addr + nrpages; idx++) {
2052 struct page *page = get_tmp_page(sbi, idx);
2053
2054 rn = F2FS_NODE(page);
2055 sum_entry->nid = rn->footer.nid;
2056 sum_entry->version = 0;
2057 sum_entry->ofs_in_node = 0;
2058 sum_entry++;
2059 f2fs_put_page(page, 1);
2060 }
2061
2062 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2063 addr + nrpages);
2064 }
2065 return 0;
2066 }
2067
2068 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2069 {
2070 struct f2fs_nm_info *nm_i = NM_I(sbi);
2071 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2072 struct f2fs_journal *journal = curseg->journal;
2073 int i;
2074
2075 down_write(&curseg->journal_rwsem);
2076 for (i = 0; i < nats_in_cursum(journal); i++) {
2077 struct nat_entry *ne;
2078 struct f2fs_nat_entry raw_ne;
2079 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2080
2081 raw_ne = nat_in_journal(journal, i);
2082
2083 ne = __lookup_nat_cache(nm_i, nid);
2084 if (!ne) {
2085 ne = grab_nat_entry(nm_i, nid);
2086 node_info_from_raw_nat(&ne->ni, &raw_ne);
2087 }
2088 __set_nat_cache_dirty(nm_i, ne);
2089 }
2090 update_nats_in_cursum(journal, -i);
2091 up_write(&curseg->journal_rwsem);
2092 }
2093
2094 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2095 struct list_head *head, int max)
2096 {
2097 struct nat_entry_set *cur;
2098
2099 if (nes->entry_cnt >= max)
2100 goto add_out;
2101
2102 list_for_each_entry(cur, head, set_list) {
2103 if (cur->entry_cnt >= nes->entry_cnt) {
2104 list_add(&nes->set_list, cur->set_list.prev);
2105 return;
2106 }
2107 }
2108 add_out:
2109 list_add_tail(&nes->set_list, head);
2110 }
2111
2112 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2113 struct nat_entry_set *set)
2114 {
2115 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2116 struct f2fs_journal *journal = curseg->journal;
2117 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2118 bool to_journal = true;
2119 struct f2fs_nat_block *nat_blk;
2120 struct nat_entry *ne, *cur;
2121 struct page *page = NULL;
2122
2123 /*
2124 * there are two steps to flush nat entries:
2125 * #1, flush nat entries to journal in current hot data summary block.
2126 * #2, flush nat entries to nat page.
2127 */
2128 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2129 to_journal = false;
2130
2131 if (to_journal) {
2132 down_write(&curseg->journal_rwsem);
2133 } else {
2134 page = get_next_nat_page(sbi, start_nid);
2135 nat_blk = page_address(page);
2136 f2fs_bug_on(sbi, !nat_blk);
2137 }
2138
2139 /* flush dirty nats in nat entry set */
2140 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2141 struct f2fs_nat_entry *raw_ne;
2142 nid_t nid = nat_get_nid(ne);
2143 int offset;
2144
2145 if (nat_get_blkaddr(ne) == NEW_ADDR)
2146 continue;
2147
2148 if (to_journal) {
2149 offset = lookup_journal_in_cursum(journal,
2150 NAT_JOURNAL, nid, 1);
2151 f2fs_bug_on(sbi, offset < 0);
2152 raw_ne = &nat_in_journal(journal, offset);
2153 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2154 } else {
2155 raw_ne = &nat_blk->entries[nid - start_nid];
2156 }
2157 raw_nat_from_node_info(raw_ne, &ne->ni);
2158 nat_reset_flag(ne);
2159 __clear_nat_cache_dirty(NM_I(sbi), ne);
2160 if (nat_get_blkaddr(ne) == NULL_ADDR)
2161 add_free_nid(sbi, nid, false);
2162 }
2163
2164 if (to_journal)
2165 up_write(&curseg->journal_rwsem);
2166 else
2167 f2fs_put_page(page, 1);
2168
2169 f2fs_bug_on(sbi, set->entry_cnt);
2170
2171 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2172 kmem_cache_free(nat_entry_set_slab, set);
2173 }
2174
2175 /*
2176 * This function is called during the checkpointing process.
2177 */
2178 void flush_nat_entries(struct f2fs_sb_info *sbi)
2179 {
2180 struct f2fs_nm_info *nm_i = NM_I(sbi);
2181 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2182 struct f2fs_journal *journal = curseg->journal;
2183 struct nat_entry_set *setvec[SETVEC_SIZE];
2184 struct nat_entry_set *set, *tmp;
2185 unsigned int found;
2186 nid_t set_idx = 0;
2187 LIST_HEAD(sets);
2188
2189 if (!nm_i->dirty_nat_cnt)
2190 return;
2191
2192 down_write(&nm_i->nat_tree_lock);
2193
2194 /*
2195 * if there are no enough space in journal to store dirty nat
2196 * entries, remove all entries from journal and merge them
2197 * into nat entry set.
2198 */
2199 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2200 remove_nats_in_journal(sbi);
2201
2202 while ((found = __gang_lookup_nat_set(nm_i,
2203 set_idx, SETVEC_SIZE, setvec))) {
2204 unsigned idx;
2205 set_idx = setvec[found - 1]->set + 1;
2206 for (idx = 0; idx < found; idx++)
2207 __adjust_nat_entry_set(setvec[idx], &sets,
2208 MAX_NAT_JENTRIES(journal));
2209 }
2210
2211 /* flush dirty nats in nat entry set */
2212 list_for_each_entry_safe(set, tmp, &sets, set_list)
2213 __flush_nat_entry_set(sbi, set);
2214
2215 up_write(&nm_i->nat_tree_lock);
2216
2217 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2218 }
2219
2220 static int init_node_manager(struct f2fs_sb_info *sbi)
2221 {
2222 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2223 struct f2fs_nm_info *nm_i = NM_I(sbi);
2224 unsigned char *version_bitmap;
2225 unsigned int nat_segs, nat_blocks;
2226
2227 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2228
2229 /* segment_count_nat includes pair segment so divide to 2. */
2230 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2231 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2232
2233 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2234
2235 /* not used nids: 0, node, meta, (and root counted as valid node) */
2236 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2237 nm_i->fcnt = 0;
2238 nm_i->nat_cnt = 0;
2239 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2240 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2241 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2242
2243 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2244 INIT_LIST_HEAD(&nm_i->free_nid_list);
2245 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2246 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2247 INIT_LIST_HEAD(&nm_i->nat_entries);
2248
2249 mutex_init(&nm_i->build_lock);
2250 spin_lock_init(&nm_i->free_nid_list_lock);
2251 init_rwsem(&nm_i->nat_tree_lock);
2252
2253 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2254 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2255 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2256 if (!version_bitmap)
2257 return -EFAULT;
2258
2259 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2260 GFP_KERNEL);
2261 if (!nm_i->nat_bitmap)
2262 return -ENOMEM;
2263 return 0;
2264 }
2265
2266 int build_node_manager(struct f2fs_sb_info *sbi)
2267 {
2268 int err;
2269
2270 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2271 if (!sbi->nm_info)
2272 return -ENOMEM;
2273
2274 err = init_node_manager(sbi);
2275 if (err)
2276 return err;
2277
2278 build_free_nids(sbi);
2279 return 0;
2280 }
2281
2282 void destroy_node_manager(struct f2fs_sb_info *sbi)
2283 {
2284 struct f2fs_nm_info *nm_i = NM_I(sbi);
2285 struct free_nid *i, *next_i;
2286 struct nat_entry *natvec[NATVEC_SIZE];
2287 struct nat_entry_set *setvec[SETVEC_SIZE];
2288 nid_t nid = 0;
2289 unsigned int found;
2290
2291 if (!nm_i)
2292 return;
2293
2294 /* destroy free nid list */
2295 spin_lock(&nm_i->free_nid_list_lock);
2296 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2297 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2298 __del_from_free_nid_list(nm_i, i);
2299 nm_i->fcnt--;
2300 spin_unlock(&nm_i->free_nid_list_lock);
2301 kmem_cache_free(free_nid_slab, i);
2302 spin_lock(&nm_i->free_nid_list_lock);
2303 }
2304 f2fs_bug_on(sbi, nm_i->fcnt);
2305 spin_unlock(&nm_i->free_nid_list_lock);
2306
2307 /* destroy nat cache */
2308 down_write(&nm_i->nat_tree_lock);
2309 while ((found = __gang_lookup_nat_cache(nm_i,
2310 nid, NATVEC_SIZE, natvec))) {
2311 unsigned idx;
2312
2313 nid = nat_get_nid(natvec[found - 1]) + 1;
2314 for (idx = 0; idx < found; idx++)
2315 __del_from_nat_cache(nm_i, natvec[idx]);
2316 }
2317 f2fs_bug_on(sbi, nm_i->nat_cnt);
2318
2319 /* destroy nat set cache */
2320 nid = 0;
2321 while ((found = __gang_lookup_nat_set(nm_i,
2322 nid, SETVEC_SIZE, setvec))) {
2323 unsigned idx;
2324
2325 nid = setvec[found - 1]->set + 1;
2326 for (idx = 0; idx < found; idx++) {
2327 /* entry_cnt is not zero, when cp_error was occurred */
2328 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2329 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2330 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2331 }
2332 }
2333 up_write(&nm_i->nat_tree_lock);
2334
2335 kfree(nm_i->nat_bitmap);
2336 sbi->nm_info = NULL;
2337 kfree(nm_i);
2338 }
2339
2340 int __init create_node_manager_caches(void)
2341 {
2342 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2343 sizeof(struct nat_entry));
2344 if (!nat_entry_slab)
2345 goto fail;
2346
2347 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2348 sizeof(struct free_nid));
2349 if (!free_nid_slab)
2350 goto destroy_nat_entry;
2351
2352 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2353 sizeof(struct nat_entry_set));
2354 if (!nat_entry_set_slab)
2355 goto destroy_free_nid;
2356 return 0;
2357
2358 destroy_free_nid:
2359 kmem_cache_destroy(free_nid_slab);
2360 destroy_nat_entry:
2361 kmem_cache_destroy(nat_entry_slab);
2362 fail:
2363 return -ENOMEM;
2364 }
2365
2366 void destroy_node_manager_caches(void)
2367 {
2368 kmem_cache_destroy(nat_entry_set_slab);
2369 kmem_cache_destroy(free_nid_slab);
2370 kmem_cache_destroy(nat_entry_slab);
2371 }
Linux kernel - Deliverables
This page took 0.077983 seconds and 6 git commands to generate.