projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: use percpu_rw_semaphore
[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 if (excess_cached_nats(sbi))
56 res = false;
57 if (nm_i->nat_cnt > DEF_NAT_CACHE_THRESHOLD)
58 res = false;
59 } else if (type == DIRTY_DENTS) {
60 if (sbi->sb->s_bdi->wb.dirty_exceeded)
61 return false;
62 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
63 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
64 } else if (type == INO_ENTRIES) {
65 int i;
66
67 for (i = 0; i <= UPDATE_INO; i++)
68 mem_size += (sbi->im[i].ino_num *
69 sizeof(struct ino_entry)) >> PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
71 } else if (type == EXTENT_CACHE) {
72 mem_size = (atomic_read(&sbi->total_ext_tree) *
73 sizeof(struct extent_tree) +
74 atomic_read(&sbi->total_ext_node) *
75 sizeof(struct extent_node)) >> PAGE_SHIFT;
76 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
77 } else {
78 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
79 return true;
80 }
81 return res;
82 }
83
84 static void clear_node_page_dirty(struct page *page)
85 {
86 struct address_space *mapping = page->mapping;
87 unsigned int long flags;
88
89 if (PageDirty(page)) {
90 spin_lock_irqsave(&mapping->tree_lock, flags);
91 radix_tree_tag_clear(&mapping->page_tree,
92 page_index(page),
93 PAGECACHE_TAG_DIRTY);
94 spin_unlock_irqrestore(&mapping->tree_lock, flags);
95
96 clear_page_dirty_for_io(page);
97 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
98 }
99 ClearPageUptodate(page);
100 }
101
102 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
103 {
104 pgoff_t index = current_nat_addr(sbi, nid);
105 return get_meta_page(sbi, index);
106 }
107
108 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
109 {
110 struct page *src_page;
111 struct page *dst_page;
112 pgoff_t src_off;
113 pgoff_t dst_off;
114 void *src_addr;
115 void *dst_addr;
116 struct f2fs_nm_info *nm_i = NM_I(sbi);
117
118 src_off = current_nat_addr(sbi, nid);
119 dst_off = next_nat_addr(sbi, src_off);
120
121 /* get current nat block page with lock */
122 src_page = get_meta_page(sbi, src_off);
123 dst_page = grab_meta_page(sbi, dst_off);
124 f2fs_bug_on(sbi, PageDirty(src_page));
125
126 src_addr = page_address(src_page);
127 dst_addr = page_address(dst_page);
128 memcpy(dst_addr, src_addr, PAGE_SIZE);
129 set_page_dirty(dst_page);
130 f2fs_put_page(src_page, 1);
131
132 set_to_next_nat(nm_i, nid);
133
134 return dst_page;
135 }
136
137 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
138 {
139 return radix_tree_lookup(&nm_i->nat_root, n);
140 }
141
142 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
143 nid_t start, unsigned int nr, struct nat_entry **ep)
144 {
145 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
146 }
147
148 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
149 {
150 list_del(&e->list);
151 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
152 nm_i->nat_cnt--;
153 kmem_cache_free(nat_entry_slab, e);
154 }
155
156 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
157 struct nat_entry *ne)
158 {
159 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
160 struct nat_entry_set *head;
161
162 if (get_nat_flag(ne, IS_DIRTY))
163 return;
164
165 head = radix_tree_lookup(&nm_i->nat_set_root, set);
166 if (!head) {
167 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
168
169 INIT_LIST_HEAD(&head->entry_list);
170 INIT_LIST_HEAD(&head->set_list);
171 head->set = set;
172 head->entry_cnt = 0;
173 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
174 }
175 list_move_tail(&ne->list, &head->entry_list);
176 nm_i->dirty_nat_cnt++;
177 head->entry_cnt++;
178 set_nat_flag(ne, IS_DIRTY, true);
179 }
180
181 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
182 struct nat_entry *ne)
183 {
184 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
185 struct nat_entry_set *head;
186
187 head = radix_tree_lookup(&nm_i->nat_set_root, set);
188 if (head) {
189 list_move_tail(&ne->list, &nm_i->nat_entries);
190 set_nat_flag(ne, IS_DIRTY, false);
191 head->entry_cnt--;
192 nm_i->dirty_nat_cnt--;
193 }
194 }
195
196 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
197 nid_t start, unsigned int nr, struct nat_entry_set **ep)
198 {
199 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
200 start, nr);
201 }
202
203 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
204 {
205 struct f2fs_nm_info *nm_i = NM_I(sbi);
206 struct nat_entry *e;
207 bool need = false;
208
209 percpu_down_read(&nm_i->nat_tree_lock);
210 e = __lookup_nat_cache(nm_i, nid);
211 if (e) {
212 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
213 !get_nat_flag(e, HAS_FSYNCED_INODE))
214 need = true;
215 }
216 percpu_up_read(&nm_i->nat_tree_lock);
217 return need;
218 }
219
220 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
221 {
222 struct f2fs_nm_info *nm_i = NM_I(sbi);
223 struct nat_entry *e;
224 bool is_cp = true;
225
226 percpu_down_read(&nm_i->nat_tree_lock);
227 e = __lookup_nat_cache(nm_i, nid);
228 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
229 is_cp = false;
230 percpu_up_read(&nm_i->nat_tree_lock);
231 return is_cp;
232 }
233
234 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
235 {
236 struct f2fs_nm_info *nm_i = NM_I(sbi);
237 struct nat_entry *e;
238 bool need_update = true;
239
240 percpu_down_read(&nm_i->nat_tree_lock);
241 e = __lookup_nat_cache(nm_i, ino);
242 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
243 (get_nat_flag(e, IS_CHECKPOINTED) ||
244 get_nat_flag(e, HAS_FSYNCED_INODE)))
245 need_update = false;
246 percpu_up_read(&nm_i->nat_tree_lock);
247 return need_update;
248 }
249
250 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
251 {
252 struct nat_entry *new;
253
254 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
255 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
256 memset(new, 0, sizeof(struct nat_entry));
257 nat_set_nid(new, nid);
258 nat_reset_flag(new);
259 list_add_tail(&new->list, &nm_i->nat_entries);
260 nm_i->nat_cnt++;
261 return new;
262 }
263
264 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
265 struct f2fs_nat_entry *ne)
266 {
267 struct f2fs_nm_info *nm_i = NM_I(sbi);
268 struct nat_entry *e;
269
270 e = __lookup_nat_cache(nm_i, nid);
271 if (!e) {
272 e = grab_nat_entry(nm_i, nid);
273 node_info_from_raw_nat(&e->ni, ne);
274 } else {
275 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
276 nat_get_blkaddr(e) != ne->block_addr ||
277 nat_get_version(e) != ne->version);
278 }
279 }
280
281 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
282 block_t new_blkaddr, bool fsync_done)
283 {
284 struct f2fs_nm_info *nm_i = NM_I(sbi);
285 struct nat_entry *e;
286
287 percpu_down_write(&nm_i->nat_tree_lock);
288 e = __lookup_nat_cache(nm_i, ni->nid);
289 if (!e) {
290 e = grab_nat_entry(nm_i, ni->nid);
291 copy_node_info(&e->ni, ni);
292 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
293 } else if (new_blkaddr == NEW_ADDR) {
294 /*
295 * when nid is reallocated,
296 * previous nat entry can be remained in nat cache.
297 * So, reinitialize it with new information.
298 */
299 copy_node_info(&e->ni, ni);
300 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
301 }
302
303 /* sanity check */
304 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
305 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
306 new_blkaddr == NULL_ADDR);
307 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
308 new_blkaddr == NEW_ADDR);
309 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
310 nat_get_blkaddr(e) != NULL_ADDR &&
311 new_blkaddr == NEW_ADDR);
312
313 /* increment version no as node is removed */
314 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
315 unsigned char version = nat_get_version(e);
316 nat_set_version(e, inc_node_version(version));
317
318 /* in order to reuse the nid */
319 if (nm_i->next_scan_nid > ni->nid)
320 nm_i->next_scan_nid = ni->nid;
321 }
322
323 /* change address */
324 nat_set_blkaddr(e, new_blkaddr);
325 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
326 set_nat_flag(e, IS_CHECKPOINTED, false);
327 __set_nat_cache_dirty(nm_i, e);
328
329 /* update fsync_mark if its inode nat entry is still alive */
330 if (ni->nid != ni->ino)
331 e = __lookup_nat_cache(nm_i, ni->ino);
332 if (e) {
333 if (fsync_done && ni->nid == ni->ino)
334 set_nat_flag(e, HAS_FSYNCED_INODE, true);
335 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
336 }
337 percpu_up_write(&nm_i->nat_tree_lock);
338 }
339
340 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
341 {
342 struct f2fs_nm_info *nm_i = NM_I(sbi);
343 int nr = nr_shrink;
344
345 percpu_down_write(&nm_i->nat_tree_lock);
346
347 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
348 struct nat_entry *ne;
349 ne = list_first_entry(&nm_i->nat_entries,
350 struct nat_entry, list);
351 __del_from_nat_cache(nm_i, ne);
352 nr_shrink--;
353 }
354 percpu_up_write(&nm_i->nat_tree_lock);
355 return nr - nr_shrink;
356 }
357
358 /*
359 * This function always returns success
360 */
361 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
362 {
363 struct f2fs_nm_info *nm_i = NM_I(sbi);
364 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
365 struct f2fs_journal *journal = curseg->journal;
366 nid_t start_nid = START_NID(nid);
367 struct f2fs_nat_block *nat_blk;
368 struct page *page = NULL;
369 struct f2fs_nat_entry ne;
370 struct nat_entry *e;
371 int i;
372
373 ni->nid = nid;
374
375 /* Check nat cache */
376 percpu_down_read(&nm_i->nat_tree_lock);
377 e = __lookup_nat_cache(nm_i, nid);
378 if (e) {
379 ni->ino = nat_get_ino(e);
380 ni->blk_addr = nat_get_blkaddr(e);
381 ni->version = nat_get_version(e);
382 percpu_up_read(&nm_i->nat_tree_lock);
383 return;
384 }
385
386 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
387
388 /* Check current segment summary */
389 down_read(&curseg->journal_rwsem);
390 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
391 if (i >= 0) {
392 ne = nat_in_journal(journal, i);
393 node_info_from_raw_nat(ni, &ne);
394 }
395 up_read(&curseg->journal_rwsem);
396 if (i >= 0)
397 goto cache;
398
399 /* Fill node_info from nat page */
400 page = get_current_nat_page(sbi, start_nid);
401 nat_blk = (struct f2fs_nat_block *)page_address(page);
402 ne = nat_blk->entries[nid - start_nid];
403 node_info_from_raw_nat(ni, &ne);
404 f2fs_put_page(page, 1);
405 cache:
406 percpu_up_read(&nm_i->nat_tree_lock);
407 /* cache nat entry */
408 percpu_down_write(&nm_i->nat_tree_lock);
409 cache_nat_entry(sbi, nid, &ne);
410 percpu_up_write(&nm_i->nat_tree_lock);
411 }
412
413 /*
414 * readahead MAX_RA_NODE number of node pages.
415 */
416 static void ra_node_pages(struct page *parent, int start, int n)
417 {
418 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
419 struct blk_plug plug;
420 int i, end;
421 nid_t nid;
422
423 blk_start_plug(&plug);
424
425 /* Then, try readahead for siblings of the desired node */
426 end = start + n;
427 end = min(end, NIDS_PER_BLOCK);
428 for (i = start; i < end; i++) {
429 nid = get_nid(parent, i, false);
430 ra_node_page(sbi, nid);
431 }
432
433 blk_finish_plug(&plug);
434 }
435
436 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
437 {
438 const long direct_index = ADDRS_PER_INODE(dn->inode);
439 const long direct_blks = ADDRS_PER_BLOCK;
440 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
441 unsigned int skipped_unit = ADDRS_PER_BLOCK;
442 int cur_level = dn->cur_level;
443 int max_level = dn->max_level;
444 pgoff_t base = 0;
445
446 if (!dn->max_level)
447 return pgofs + 1;
448
449 while (max_level-- > cur_level)
450 skipped_unit *= NIDS_PER_BLOCK;
451
452 switch (dn->max_level) {
453 case 3:
454 base += 2 * indirect_blks;
455 case 2:
456 base += 2 * direct_blks;
457 case 1:
458 base += direct_index;
459 break;
460 default:
461 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
462 }
463
464 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
465 }
466
467 /*
468 * The maximum depth is four.
469 * Offset[0] will have raw inode offset.
470 */
471 static int get_node_path(struct inode *inode, long block,
472 int offset[4], unsigned int noffset[4])
473 {
474 const long direct_index = ADDRS_PER_INODE(inode);
475 const long direct_blks = ADDRS_PER_BLOCK;
476 const long dptrs_per_blk = NIDS_PER_BLOCK;
477 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
478 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
479 int n = 0;
480 int level = 0;
481
482 noffset[0] = 0;
483
484 if (block < direct_index) {
485 offset[n] = block;
486 goto got;
487 }
488 block -= direct_index;
489 if (block < direct_blks) {
490 offset[n++] = NODE_DIR1_BLOCK;
491 noffset[n] = 1;
492 offset[n] = block;
493 level = 1;
494 goto got;
495 }
496 block -= direct_blks;
497 if (block < direct_blks) {
498 offset[n++] = NODE_DIR2_BLOCK;
499 noffset[n] = 2;
500 offset[n] = block;
501 level = 1;
502 goto got;
503 }
504 block -= direct_blks;
505 if (block < indirect_blks) {
506 offset[n++] = NODE_IND1_BLOCK;
507 noffset[n] = 3;
508 offset[n++] = block / direct_blks;
509 noffset[n] = 4 + offset[n - 1];
510 offset[n] = block % direct_blks;
511 level = 2;
512 goto got;
513 }
514 block -= indirect_blks;
515 if (block < indirect_blks) {
516 offset[n++] = NODE_IND2_BLOCK;
517 noffset[n] = 4 + dptrs_per_blk;
518 offset[n++] = block / direct_blks;
519 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
520 offset[n] = block % direct_blks;
521 level = 2;
522 goto got;
523 }
524 block -= indirect_blks;
525 if (block < dindirect_blks) {
526 offset[n++] = NODE_DIND_BLOCK;
527 noffset[n] = 5 + (dptrs_per_blk * 2);
528 offset[n++] = block / indirect_blks;
529 noffset[n] = 6 + (dptrs_per_blk * 2) +
530 offset[n - 1] * (dptrs_per_blk + 1);
531 offset[n++] = (block / direct_blks) % dptrs_per_blk;
532 noffset[n] = 7 + (dptrs_per_blk * 2) +
533 offset[n - 2] * (dptrs_per_blk + 1) +
534 offset[n - 1];
535 offset[n] = block % direct_blks;
536 level = 3;
537 goto got;
538 } else {
539 BUG();
540 }
541 got:
542 return level;
543 }
544
545 /*
546 * Caller should call f2fs_put_dnode(dn).
547 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
548 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
549 * In the case of RDONLY_NODE, we don't need to care about mutex.
550 */
551 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
552 {
553 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
554 struct page *npage[4];
555 struct page *parent = NULL;
556 int offset[4];
557 unsigned int noffset[4];
558 nid_t nids[4];
559 int level, i = 0;
560 int err = 0;
561
562 level = get_node_path(dn->inode, index, offset, noffset);
563
564 nids[0] = dn->inode->i_ino;
565 npage[0] = dn->inode_page;
566
567 if (!npage[0]) {
568 npage[0] = get_node_page(sbi, nids[0]);
569 if (IS_ERR(npage[0]))
570 return PTR_ERR(npage[0]);
571 }
572
573 /* if inline_data is set, should not report any block indices */
574 if (f2fs_has_inline_data(dn->inode) && index) {
575 err = -ENOENT;
576 f2fs_put_page(npage[0], 1);
577 goto release_out;
578 }
579
580 parent = npage[0];
581 if (level != 0)
582 nids[1] = get_nid(parent, offset[0], true);
583 dn->inode_page = npage[0];
584 dn->inode_page_locked = true;
585
586 /* get indirect or direct nodes */
587 for (i = 1; i <= level; i++) {
588 bool done = false;
589
590 if (!nids[i] && mode == ALLOC_NODE) {
591 /* alloc new node */
592 if (!alloc_nid(sbi, &(nids[i]))) {
593 err = -ENOSPC;
594 goto release_pages;
595 }
596
597 dn->nid = nids[i];
598 npage[i] = new_node_page(dn, noffset[i], NULL);
599 if (IS_ERR(npage[i])) {
600 alloc_nid_failed(sbi, nids[i]);
601 err = PTR_ERR(npage[i]);
602 goto release_pages;
603 }
604
605 set_nid(parent, offset[i - 1], nids[i], i == 1);
606 alloc_nid_done(sbi, nids[i]);
607 done = true;
608 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
609 npage[i] = get_node_page_ra(parent, offset[i - 1]);
610 if (IS_ERR(npage[i])) {
611 err = PTR_ERR(npage[i]);
612 goto release_pages;
613 }
614 done = true;
615 }
616 if (i == 1) {
617 dn->inode_page_locked = false;
618 unlock_page(parent);
619 } else {
620 f2fs_put_page(parent, 1);
621 }
622
623 if (!done) {
624 npage[i] = get_node_page(sbi, nids[i]);
625 if (IS_ERR(npage[i])) {
626 err = PTR_ERR(npage[i]);
627 f2fs_put_page(npage[0], 0);
628 goto release_out;
629 }
630 }
631 if (i < level) {
632 parent = npage[i];
633 nids[i + 1] = get_nid(parent, offset[i], false);
634 }
635 }
636 dn->nid = nids[level];
637 dn->ofs_in_node = offset[level];
638 dn->node_page = npage[level];
639 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
640 return 0;
641
642 release_pages:
643 f2fs_put_page(parent, 1);
644 if (i > 1)
645 f2fs_put_page(npage[0], 0);
646 release_out:
647 dn->inode_page = NULL;
648 dn->node_page = NULL;
649 if (err == -ENOENT) {
650 dn->cur_level = i;
651 dn->max_level = level;
652 }
653 return err;
654 }
655
656 static void truncate_node(struct dnode_of_data *dn)
657 {
658 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
659 struct node_info ni;
660
661 get_node_info(sbi, dn->nid, &ni);
662 if (dn->inode->i_blocks == 0) {
663 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
664 goto invalidate;
665 }
666 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
667
668 /* Deallocate node address */
669 invalidate_blocks(sbi, ni.blk_addr);
670 dec_valid_node_count(sbi, dn->inode);
671 set_node_addr(sbi, &ni, NULL_ADDR, false);
672
673 if (dn->nid == dn->inode->i_ino) {
674 remove_orphan_inode(sbi, dn->nid);
675 dec_valid_inode_count(sbi);
676 f2fs_inode_synced(dn->inode);
677 }
678 invalidate:
679 clear_node_page_dirty(dn->node_page);
680 set_sbi_flag(sbi, SBI_IS_DIRTY);
681
682 f2fs_put_page(dn->node_page, 1);
683
684 invalidate_mapping_pages(NODE_MAPPING(sbi),
685 dn->node_page->index, dn->node_page->index);
686
687 dn->node_page = NULL;
688 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
689 }
690
691 static int truncate_dnode(struct dnode_of_data *dn)
692 {
693 struct page *page;
694
695 if (dn->nid == 0)
696 return 1;
697
698 /* get direct node */
699 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
700 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
701 return 1;
702 else if (IS_ERR(page))
703 return PTR_ERR(page);
704
705 /* Make dnode_of_data for parameter */
706 dn->node_page = page;
707 dn->ofs_in_node = 0;
708 truncate_data_blocks(dn);
709 truncate_node(dn);
710 return 1;
711 }
712
713 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
714 int ofs, int depth)
715 {
716 struct dnode_of_data rdn = *dn;
717 struct page *page;
718 struct f2fs_node *rn;
719 nid_t child_nid;
720 unsigned int child_nofs;
721 int freed = 0;
722 int i, ret;
723
724 if (dn->nid == 0)
725 return NIDS_PER_BLOCK + 1;
726
727 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
728
729 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
730 if (IS_ERR(page)) {
731 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
732 return PTR_ERR(page);
733 }
734
735 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
736
737 rn = F2FS_NODE(page);
738 if (depth < 3) {
739 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
740 child_nid = le32_to_cpu(rn->in.nid[i]);
741 if (child_nid == 0)
742 continue;
743 rdn.nid = child_nid;
744 ret = truncate_dnode(&rdn);
745 if (ret < 0)
746 goto out_err;
747 if (set_nid(page, i, 0, false))
748 dn->node_changed = true;
749 }
750 } else {
751 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
752 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
753 child_nid = le32_to_cpu(rn->in.nid[i]);
754 if (child_nid == 0) {
755 child_nofs += NIDS_PER_BLOCK + 1;
756 continue;
757 }
758 rdn.nid = child_nid;
759 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
760 if (ret == (NIDS_PER_BLOCK + 1)) {
761 if (set_nid(page, i, 0, false))
762 dn->node_changed = true;
763 child_nofs += ret;
764 } else if (ret < 0 && ret != -ENOENT) {
765 goto out_err;
766 }
767 }
768 freed = child_nofs;
769 }
770
771 if (!ofs) {
772 /* remove current indirect node */
773 dn->node_page = page;
774 truncate_node(dn);
775 freed++;
776 } else {
777 f2fs_put_page(page, 1);
778 }
779 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
780 return freed;
781
782 out_err:
783 f2fs_put_page(page, 1);
784 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
785 return ret;
786 }
787
788 static int truncate_partial_nodes(struct dnode_of_data *dn,
789 struct f2fs_inode *ri, int *offset, int depth)
790 {
791 struct page *pages[2];
792 nid_t nid[3];
793 nid_t child_nid;
794 int err = 0;
795 int i;
796 int idx = depth - 2;
797
798 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
799 if (!nid[0])
800 return 0;
801
802 /* get indirect nodes in the path */
803 for (i = 0; i < idx + 1; i++) {
804 /* reference count'll be increased */
805 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
806 if (IS_ERR(pages[i])) {
807 err = PTR_ERR(pages[i]);
808 idx = i - 1;
809 goto fail;
810 }
811 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
812 }
813
814 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
815
816 /* free direct nodes linked to a partial indirect node */
817 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
818 child_nid = get_nid(pages[idx], i, false);
819 if (!child_nid)
820 continue;
821 dn->nid = child_nid;
822 err = truncate_dnode(dn);
823 if (err < 0)
824 goto fail;
825 if (set_nid(pages[idx], i, 0, false))
826 dn->node_changed = true;
827 }
828
829 if (offset[idx + 1] == 0) {
830 dn->node_page = pages[idx];
831 dn->nid = nid[idx];
832 truncate_node(dn);
833 } else {
834 f2fs_put_page(pages[idx], 1);
835 }
836 offset[idx]++;
837 offset[idx + 1] = 0;
838 idx--;
839 fail:
840 for (i = idx; i >= 0; i--)
841 f2fs_put_page(pages[i], 1);
842
843 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
844
845 return err;
846 }
847
848 /*
849 * All the block addresses of data and nodes should be nullified.
850 */
851 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
852 {
853 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
854 int err = 0, cont = 1;
855 int level, offset[4], noffset[4];
856 unsigned int nofs = 0;
857 struct f2fs_inode *ri;
858 struct dnode_of_data dn;
859 struct page *page;
860
861 trace_f2fs_truncate_inode_blocks_enter(inode, from);
862
863 level = get_node_path(inode, from, offset, noffset);
864
865 page = get_node_page(sbi, inode->i_ino);
866 if (IS_ERR(page)) {
867 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
868 return PTR_ERR(page);
869 }
870
871 set_new_dnode(&dn, inode, page, NULL, 0);
872 unlock_page(page);
873
874 ri = F2FS_INODE(page);
875 switch (level) {
876 case 0:
877 case 1:
878 nofs = noffset[1];
879 break;
880 case 2:
881 nofs = noffset[1];
882 if (!offset[level - 1])
883 goto skip_partial;
884 err = truncate_partial_nodes(&dn, ri, offset, level);
885 if (err < 0 && err != -ENOENT)
886 goto fail;
887 nofs += 1 + NIDS_PER_BLOCK;
888 break;
889 case 3:
890 nofs = 5 + 2 * NIDS_PER_BLOCK;
891 if (!offset[level - 1])
892 goto skip_partial;
893 err = truncate_partial_nodes(&dn, ri, offset, level);
894 if (err < 0 && err != -ENOENT)
895 goto fail;
896 break;
897 default:
898 BUG();
899 }
900
901 skip_partial:
902 while (cont) {
903 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
904 switch (offset[0]) {
905 case NODE_DIR1_BLOCK:
906 case NODE_DIR2_BLOCK:
907 err = truncate_dnode(&dn);
908 break;
909
910 case NODE_IND1_BLOCK:
911 case NODE_IND2_BLOCK:
912 err = truncate_nodes(&dn, nofs, offset[1], 2);
913 break;
914
915 case NODE_DIND_BLOCK:
916 err = truncate_nodes(&dn, nofs, offset[1], 3);
917 cont = 0;
918 break;
919
920 default:
921 BUG();
922 }
923 if (err < 0 && err != -ENOENT)
924 goto fail;
925 if (offset[1] == 0 &&
926 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
927 lock_page(page);
928 BUG_ON(page->mapping != NODE_MAPPING(sbi));
929 f2fs_wait_on_page_writeback(page, NODE, true);
930 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
931 set_page_dirty(page);
932 unlock_page(page);
933 }
934 offset[1] = 0;
935 offset[0]++;
936 nofs += err;
937 }
938 fail:
939 f2fs_put_page(page, 0);
940 trace_f2fs_truncate_inode_blocks_exit(inode, err);
941 return err > 0 ? 0 : err;
942 }
943
944 int truncate_xattr_node(struct inode *inode, struct page *page)
945 {
946 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
947 nid_t nid = F2FS_I(inode)->i_xattr_nid;
948 struct dnode_of_data dn;
949 struct page *npage;
950
951 if (!nid)
952 return 0;
953
954 npage = get_node_page(sbi, nid);
955 if (IS_ERR(npage))
956 return PTR_ERR(npage);
957
958 f2fs_i_xnid_write(inode, 0);
959
960 /* need to do checkpoint during fsync */
961 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
962
963 set_new_dnode(&dn, inode, page, npage, nid);
964
965 if (page)
966 dn.inode_page_locked = true;
967 truncate_node(&dn);
968 return 0;
969 }
970
971 /*
972 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
973 * f2fs_unlock_op().
974 */
975 int remove_inode_page(struct inode *inode)
976 {
977 struct dnode_of_data dn;
978 int err;
979
980 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
981 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
982 if (err)
983 return err;
984
985 err = truncate_xattr_node(inode, dn.inode_page);
986 if (err) {
987 f2fs_put_dnode(&dn);
988 return err;
989 }
990
991 /* remove potential inline_data blocks */
992 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
993 S_ISLNK(inode->i_mode))
994 truncate_data_blocks_range(&dn, 1);
995
996 /* 0 is possible, after f2fs_new_inode() has failed */
997 f2fs_bug_on(F2FS_I_SB(inode),
998 inode->i_blocks != 0 && inode->i_blocks != 1);
999
1000 /* will put inode & node pages */
1001 truncate_node(&dn);
1002 return 0;
1003 }
1004
1005 struct page *new_inode_page(struct inode *inode)
1006 {
1007 struct dnode_of_data dn;
1008
1009 /* allocate inode page for new inode */
1010 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1011
1012 /* caller should f2fs_put_page(page, 1); */
1013 return new_node_page(&dn, 0, NULL);
1014 }
1015
1016 struct page *new_node_page(struct dnode_of_data *dn,
1017 unsigned int ofs, struct page *ipage)
1018 {
1019 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1020 struct node_info old_ni, new_ni;
1021 struct page *page;
1022 int err;
1023
1024 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1025 return ERR_PTR(-EPERM);
1026
1027 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1028 if (!page)
1029 return ERR_PTR(-ENOMEM);
1030
1031 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1032 err = -ENOSPC;
1033 goto fail;
1034 }
1035
1036 get_node_info(sbi, dn->nid, &old_ni);
1037
1038 /* Reinitialize old_ni with new node page */
1039 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1040 new_ni = old_ni;
1041 new_ni.ino = dn->inode->i_ino;
1042 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1043
1044 f2fs_wait_on_page_writeback(page, NODE, true);
1045 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1046 set_cold_node(dn->inode, page);
1047 if (!PageUptodate(page))
1048 SetPageUptodate(page);
1049 if (set_page_dirty(page))
1050 dn->node_changed = true;
1051
1052 if (f2fs_has_xattr_block(ofs))
1053 f2fs_i_xnid_write(dn->inode, dn->nid);
1054
1055 if (ofs == 0)
1056 inc_valid_inode_count(sbi);
1057 return page;
1058
1059 fail:
1060 clear_node_page_dirty(page);
1061 f2fs_put_page(page, 1);
1062 return ERR_PTR(err);
1063 }
1064
1065 /*
1066 * Caller should do after getting the following values.
1067 * 0: f2fs_put_page(page, 0)
1068 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1069 */
1070 static int read_node_page(struct page *page, int rw)
1071 {
1072 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1073 struct node_info ni;
1074 struct f2fs_io_info fio = {
1075 .sbi = sbi,
1076 .type = NODE,
1077 .rw = rw,
1078 .page = page,
1079 .encrypted_page = NULL,
1080 };
1081
1082 if (PageUptodate(page))
1083 return LOCKED_PAGE;
1084
1085 get_node_info(sbi, page->index, &ni);
1086
1087 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1088 ClearPageUptodate(page);
1089 return -ENOENT;
1090 }
1091
1092 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1093 return f2fs_submit_page_bio(&fio);
1094 }
1095
1096 /*
1097 * Readahead a node page
1098 */
1099 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1100 {
1101 struct page *apage;
1102 int err;
1103
1104 if (!nid)
1105 return;
1106 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1107
1108 rcu_read_lock();
1109 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1110 rcu_read_unlock();
1111 if (apage)
1112 return;
1113
1114 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1115 if (!apage)
1116 return;
1117
1118 err = read_node_page(apage, READA);
1119 f2fs_put_page(apage, err ? 1 : 0);
1120 }
1121
1122 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1123 struct page *parent, int start)
1124 {
1125 struct page *page;
1126 int err;
1127
1128 if (!nid)
1129 return ERR_PTR(-ENOENT);
1130 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1131 repeat:
1132 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1133 if (!page)
1134 return ERR_PTR(-ENOMEM);
1135
1136 err = read_node_page(page, READ_SYNC);
1137 if (err < 0) {
1138 f2fs_put_page(page, 1);
1139 return ERR_PTR(err);
1140 } else if (err == LOCKED_PAGE) {
1141 goto page_hit;
1142 }
1143
1144 if (parent)
1145 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1146
1147 lock_page(page);
1148
1149 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1150 f2fs_put_page(page, 1);
1151 goto repeat;
1152 }
1153
1154 if (unlikely(!PageUptodate(page)))
1155 goto out_err;
1156 page_hit:
1157 if(unlikely(nid != nid_of_node(page))) {
1158 f2fs_bug_on(sbi, 1);
1159 ClearPageUptodate(page);
1160 out_err:
1161 f2fs_put_page(page, 1);
1162 return ERR_PTR(-EIO);
1163 }
1164 return page;
1165 }
1166
1167 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1168 {
1169 return __get_node_page(sbi, nid, NULL, 0);
1170 }
1171
1172 struct page *get_node_page_ra(struct page *parent, int start)
1173 {
1174 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1175 nid_t nid = get_nid(parent, start, false);
1176
1177 return __get_node_page(sbi, nid, parent, start);
1178 }
1179
1180 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1181 {
1182 struct inode *inode;
1183 struct page *page;
1184 int ret;
1185
1186 /* should flush inline_data before evict_inode */
1187 inode = ilookup(sbi->sb, ino);
1188 if (!inode)
1189 return;
1190
1191 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1192 if (!page)
1193 goto iput_out;
1194
1195 if (!PageUptodate(page))
1196 goto page_out;
1197
1198 if (!PageDirty(page))
1199 goto page_out;
1200
1201 if (!clear_page_dirty_for_io(page))
1202 goto page_out;
1203
1204 ret = f2fs_write_inline_data(inode, page);
1205 inode_dec_dirty_pages(inode);
1206 if (ret)
1207 set_page_dirty(page);
1208 page_out:
1209 f2fs_put_page(page, 1);
1210 iput_out:
1211 iput(inode);
1212 }
1213
1214 void move_node_page(struct page *node_page, int gc_type)
1215 {
1216 if (gc_type == FG_GC) {
1217 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1218 struct writeback_control wbc = {
1219 .sync_mode = WB_SYNC_ALL,
1220 .nr_to_write = 1,
1221 .for_reclaim = 0,
1222 };
1223
1224 set_page_dirty(node_page);
1225 f2fs_wait_on_page_writeback(node_page, NODE, true);
1226
1227 f2fs_bug_on(sbi, PageWriteback(node_page));
1228 if (!clear_page_dirty_for_io(node_page))
1229 goto out_page;
1230
1231 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1232 unlock_page(node_page);
1233 goto release_page;
1234 } else {
1235 /* set page dirty and write it */
1236 if (!PageWriteback(node_page))
1237 set_page_dirty(node_page);
1238 }
1239 out_page:
1240 unlock_page(node_page);
1241 release_page:
1242 f2fs_put_page(node_page, 0);
1243 }
1244
1245 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1246 {
1247 pgoff_t index, end;
1248 struct pagevec pvec;
1249 struct page *last_page = NULL;
1250
1251 pagevec_init(&pvec, 0);
1252 index = 0;
1253 end = ULONG_MAX;
1254
1255 while (index <= end) {
1256 int i, nr_pages;
1257 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1258 PAGECACHE_TAG_DIRTY,
1259 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1260 if (nr_pages == 0)
1261 break;
1262
1263 for (i = 0; i < nr_pages; i++) {
1264 struct page *page = pvec.pages[i];
1265
1266 if (unlikely(f2fs_cp_error(sbi))) {
1267 f2fs_put_page(last_page, 0);
1268 pagevec_release(&pvec);
1269 return ERR_PTR(-EIO);
1270 }
1271
1272 if (!IS_DNODE(page) || !is_cold_node(page))
1273 continue;
1274 if (ino_of_node(page) != ino)
1275 continue;
1276
1277 lock_page(page);
1278
1279 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1280 continue_unlock:
1281 unlock_page(page);
1282 continue;
1283 }
1284 if (ino_of_node(page) != ino)
1285 goto continue_unlock;
1286
1287 if (!PageDirty(page)) {
1288 /* someone wrote it for us */
1289 goto continue_unlock;
1290 }
1291
1292 if (last_page)
1293 f2fs_put_page(last_page, 0);
1294
1295 get_page(page);
1296 last_page = page;
1297 unlock_page(page);
1298 }
1299 pagevec_release(&pvec);
1300 cond_resched();
1301 }
1302 return last_page;
1303 }
1304
1305 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1306 struct writeback_control *wbc, bool atomic)
1307 {
1308 pgoff_t index, end;
1309 struct pagevec pvec;
1310 int ret = 0;
1311 struct page *last_page = NULL;
1312 bool marked = false;
1313 nid_t ino = inode->i_ino;
1314
1315 if (atomic) {
1316 last_page = last_fsync_dnode(sbi, ino);
1317 if (IS_ERR_OR_NULL(last_page))
1318 return PTR_ERR_OR_ZERO(last_page);
1319 }
1320 retry:
1321 pagevec_init(&pvec, 0);
1322 index = 0;
1323 end = ULONG_MAX;
1324
1325 while (index <= end) {
1326 int i, nr_pages;
1327 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1328 PAGECACHE_TAG_DIRTY,
1329 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1330 if (nr_pages == 0)
1331 break;
1332
1333 for (i = 0; i < nr_pages; i++) {
1334 struct page *page = pvec.pages[i];
1335
1336 if (unlikely(f2fs_cp_error(sbi))) {
1337 f2fs_put_page(last_page, 0);
1338 pagevec_release(&pvec);
1339 return -EIO;
1340 }
1341
1342 if (!IS_DNODE(page) || !is_cold_node(page))
1343 continue;
1344 if (ino_of_node(page) != ino)
1345 continue;
1346
1347 lock_page(page);
1348
1349 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1350 continue_unlock:
1351 unlock_page(page);
1352 continue;
1353 }
1354 if (ino_of_node(page) != ino)
1355 goto continue_unlock;
1356
1357 if (!PageDirty(page) && page != last_page) {
1358 /* someone wrote it for us */
1359 goto continue_unlock;
1360 }
1361
1362 f2fs_wait_on_page_writeback(page, NODE, true);
1363 BUG_ON(PageWriteback(page));
1364
1365 if (!atomic || page == last_page) {
1366 set_fsync_mark(page, 1);
1367 if (IS_INODE(page)) {
1368 if (is_inode_flag_set(inode,
1369 FI_DIRTY_INODE))
1370 update_inode(inode, page);
1371 set_dentry_mark(page,
1372 need_dentry_mark(sbi, ino));
1373 }
1374 /* may be written by other thread */
1375 if (!PageDirty(page))
1376 set_page_dirty(page);
1377 }
1378
1379 if (!clear_page_dirty_for_io(page))
1380 goto continue_unlock;
1381
1382 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1383 if (ret) {
1384 unlock_page(page);
1385 f2fs_put_page(last_page, 0);
1386 break;
1387 }
1388 if (page == last_page) {
1389 f2fs_put_page(page, 0);
1390 marked = true;
1391 break;
1392 }
1393 }
1394 pagevec_release(&pvec);
1395 cond_resched();
1396
1397 if (ret || marked)
1398 break;
1399 }
1400 if (!ret && atomic && !marked) {
1401 f2fs_msg(sbi->sb, KERN_DEBUG,
1402 "Retry to write fsync mark: ino=%u, idx=%lx",
1403 ino, last_page->index);
1404 lock_page(last_page);
1405 set_page_dirty(last_page);
1406 unlock_page(last_page);
1407 goto retry;
1408 }
1409 return ret ? -EIO: 0;
1410 }
1411
1412 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1413 {
1414 pgoff_t index, end;
1415 struct pagevec pvec;
1416 int step = 0;
1417 int nwritten = 0;
1418
1419 pagevec_init(&pvec, 0);
1420
1421 next_step:
1422 index = 0;
1423 end = ULONG_MAX;
1424
1425 while (index <= end) {
1426 int i, nr_pages;
1427 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1428 PAGECACHE_TAG_DIRTY,
1429 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1430 if (nr_pages == 0)
1431 break;
1432
1433 for (i = 0; i < nr_pages; i++) {
1434 struct page *page = pvec.pages[i];
1435
1436 if (unlikely(f2fs_cp_error(sbi))) {
1437 pagevec_release(&pvec);
1438 return -EIO;
1439 }
1440
1441 /*
1442 * flushing sequence with step:
1443 * 0. indirect nodes
1444 * 1. dentry dnodes
1445 * 2. file dnodes
1446 */
1447 if (step == 0 && IS_DNODE(page))
1448 continue;
1449 if (step == 1 && (!IS_DNODE(page) ||
1450 is_cold_node(page)))
1451 continue;
1452 if (step == 2 && (!IS_DNODE(page) ||
1453 !is_cold_node(page)))
1454 continue;
1455 lock_node:
1456 if (!trylock_page(page))
1457 continue;
1458
1459 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1460 continue_unlock:
1461 unlock_page(page);
1462 continue;
1463 }
1464
1465 if (!PageDirty(page)) {
1466 /* someone wrote it for us */
1467 goto continue_unlock;
1468 }
1469
1470 /* flush inline_data */
1471 if (is_inline_node(page)) {
1472 clear_inline_node(page);
1473 unlock_page(page);
1474 flush_inline_data(sbi, ino_of_node(page));
1475 goto lock_node;
1476 }
1477
1478 f2fs_wait_on_page_writeback(page, NODE, true);
1479
1480 BUG_ON(PageWriteback(page));
1481 if (!clear_page_dirty_for_io(page))
1482 goto continue_unlock;
1483
1484 set_fsync_mark(page, 0);
1485 set_dentry_mark(page, 0);
1486
1487 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1488 unlock_page(page);
1489
1490 if (--wbc->nr_to_write == 0)
1491 break;
1492 }
1493 pagevec_release(&pvec);
1494 cond_resched();
1495
1496 if (wbc->nr_to_write == 0) {
1497 step = 2;
1498 break;
1499 }
1500 }
1501
1502 if (step < 2) {
1503 step++;
1504 goto next_step;
1505 }
1506 return nwritten;
1507 }
1508
1509 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1510 {
1511 pgoff_t index = 0, end = ULONG_MAX;
1512 struct pagevec pvec;
1513 int ret2 = 0, ret = 0;
1514
1515 pagevec_init(&pvec, 0);
1516
1517 while (index <= end) {
1518 int i, nr_pages;
1519 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1520 PAGECACHE_TAG_WRITEBACK,
1521 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1522 if (nr_pages == 0)
1523 break;
1524
1525 for (i = 0; i < nr_pages; i++) {
1526 struct page *page = pvec.pages[i];
1527
1528 /* until radix tree lookup accepts end_index */
1529 if (unlikely(page->index > end))
1530 continue;
1531
1532 if (ino && ino_of_node(page) == ino) {
1533 f2fs_wait_on_page_writeback(page, NODE, true);
1534 if (TestClearPageError(page))
1535 ret = -EIO;
1536 }
1537 }
1538 pagevec_release(&pvec);
1539 cond_resched();
1540 }
1541
1542 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1543 ret2 = -ENOSPC;
1544 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1545 ret2 = -EIO;
1546 if (!ret)
1547 ret = ret2;
1548 return ret;
1549 }
1550
1551 static int f2fs_write_node_page(struct page *page,
1552 struct writeback_control *wbc)
1553 {
1554 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1555 nid_t nid;
1556 struct node_info ni;
1557 struct f2fs_io_info fio = {
1558 .sbi = sbi,
1559 .type = NODE,
1560 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1561 .page = page,
1562 .encrypted_page = NULL,
1563 };
1564
1565 trace_f2fs_writepage(page, NODE);
1566
1567 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1568 goto redirty_out;
1569 if (unlikely(f2fs_cp_error(sbi)))
1570 goto redirty_out;
1571
1572 /* get old block addr of this node page */
1573 nid = nid_of_node(page);
1574 f2fs_bug_on(sbi, page->index != nid);
1575
1576 if (wbc->for_reclaim) {
1577 if (!down_read_trylock(&sbi->node_write))
1578 goto redirty_out;
1579 } else {
1580 down_read(&sbi->node_write);
1581 }
1582
1583 get_node_info(sbi, nid, &ni);
1584
1585 /* This page is already truncated */
1586 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1587 ClearPageUptodate(page);
1588 dec_page_count(sbi, F2FS_DIRTY_NODES);
1589 up_read(&sbi->node_write);
1590 unlock_page(page);
1591 return 0;
1592 }
1593
1594 set_page_writeback(page);
1595 fio.old_blkaddr = ni.blk_addr;
1596 write_node_page(nid, &fio);
1597 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1598 dec_page_count(sbi, F2FS_DIRTY_NODES);
1599 up_read(&sbi->node_write);
1600
1601 if (wbc->for_reclaim)
1602 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1603
1604 unlock_page(page);
1605
1606 if (unlikely(f2fs_cp_error(sbi)))
1607 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1608
1609 return 0;
1610
1611 redirty_out:
1612 redirty_page_for_writepage(wbc, page);
1613 return AOP_WRITEPAGE_ACTIVATE;
1614 }
1615
1616 static int f2fs_write_node_pages(struct address_space *mapping,
1617 struct writeback_control *wbc)
1618 {
1619 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1620 long diff;
1621
1622 /* balancing f2fs's metadata in background */
1623 f2fs_balance_fs_bg(sbi);
1624
1625 /* collect a number of dirty node pages and write together */
1626 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1627 goto skip_write;
1628
1629 trace_f2fs_writepages(mapping->host, wbc, NODE);
1630
1631 diff = nr_pages_to_write(sbi, NODE, wbc);
1632 wbc->sync_mode = WB_SYNC_NONE;
1633 sync_node_pages(sbi, wbc);
1634 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1635 return 0;
1636
1637 skip_write:
1638 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1639 trace_f2fs_writepages(mapping->host, wbc, NODE);
1640 return 0;
1641 }
1642
1643 static int f2fs_set_node_page_dirty(struct page *page)
1644 {
1645 trace_f2fs_set_page_dirty(page, NODE);
1646
1647 if (!PageUptodate(page))
1648 SetPageUptodate(page);
1649 if (!PageDirty(page)) {
1650 f2fs_set_page_dirty_nobuffers(page);
1651 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1652 SetPagePrivate(page);
1653 f2fs_trace_pid(page);
1654 return 1;
1655 }
1656 return 0;
1657 }
1658
1659 /*
1660 * Structure of the f2fs node operations
1661 */
1662 const struct address_space_operations f2fs_node_aops = {
1663 .writepage = f2fs_write_node_page,
1664 .writepages = f2fs_write_node_pages,
1665 .set_page_dirty = f2fs_set_node_page_dirty,
1666 .invalidatepage = f2fs_invalidate_page,
1667 .releasepage = f2fs_release_page,
1668 };
1669
1670 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1671 nid_t n)
1672 {
1673 return radix_tree_lookup(&nm_i->free_nid_root, n);
1674 }
1675
1676 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1677 struct free_nid *i)
1678 {
1679 list_del(&i->list);
1680 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1681 }
1682
1683 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1684 {
1685 struct f2fs_nm_info *nm_i = NM_I(sbi);
1686 struct free_nid *i;
1687 struct nat_entry *ne;
1688
1689 if (!available_free_memory(sbi, FREE_NIDS))
1690 return -1;
1691
1692 /* 0 nid should not be used */
1693 if (unlikely(nid == 0))
1694 return 0;
1695
1696 if (build) {
1697 /* do not add allocated nids */
1698 ne = __lookup_nat_cache(nm_i, nid);
1699 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1700 nat_get_blkaddr(ne) != NULL_ADDR))
1701 return 0;
1702 }
1703
1704 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1705 i->nid = nid;
1706 i->state = NID_NEW;
1707
1708 if (radix_tree_preload(GFP_NOFS)) {
1709 kmem_cache_free(free_nid_slab, i);
1710 return 0;
1711 }
1712
1713 spin_lock(&nm_i->free_nid_list_lock);
1714 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1715 spin_unlock(&nm_i->free_nid_list_lock);
1716 radix_tree_preload_end();
1717 kmem_cache_free(free_nid_slab, i);
1718 return 0;
1719 }
1720 list_add_tail(&i->list, &nm_i->free_nid_list);
1721 nm_i->fcnt++;
1722 spin_unlock(&nm_i->free_nid_list_lock);
1723 radix_tree_preload_end();
1724 return 1;
1725 }
1726
1727 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1728 {
1729 struct free_nid *i;
1730 bool need_free = false;
1731
1732 spin_lock(&nm_i->free_nid_list_lock);
1733 i = __lookup_free_nid_list(nm_i, nid);
1734 if (i && i->state == NID_NEW) {
1735 __del_from_free_nid_list(nm_i, i);
1736 nm_i->fcnt--;
1737 need_free = true;
1738 }
1739 spin_unlock(&nm_i->free_nid_list_lock);
1740
1741 if (need_free)
1742 kmem_cache_free(free_nid_slab, i);
1743 }
1744
1745 static void scan_nat_page(struct f2fs_sb_info *sbi,
1746 struct page *nat_page, nid_t start_nid)
1747 {
1748 struct f2fs_nm_info *nm_i = NM_I(sbi);
1749 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1750 block_t blk_addr;
1751 int i;
1752
1753 i = start_nid % NAT_ENTRY_PER_BLOCK;
1754
1755 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1756
1757 if (unlikely(start_nid >= nm_i->max_nid))
1758 break;
1759
1760 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1761 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1762 if (blk_addr == NULL_ADDR) {
1763 if (add_free_nid(sbi, start_nid, true) < 0)
1764 break;
1765 }
1766 }
1767 }
1768
1769 void build_free_nids(struct f2fs_sb_info *sbi)
1770 {
1771 struct f2fs_nm_info *nm_i = NM_I(sbi);
1772 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1773 struct f2fs_journal *journal = curseg->journal;
1774 int i = 0;
1775 nid_t nid = nm_i->next_scan_nid;
1776
1777 /* Enough entries */
1778 if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1779 return;
1780
1781 /* readahead nat pages to be scanned */
1782 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1783 META_NAT, true);
1784
1785 percpu_down_read(&nm_i->nat_tree_lock);
1786
1787 while (1) {
1788 struct page *page = get_current_nat_page(sbi, nid);
1789
1790 scan_nat_page(sbi, page, nid);
1791 f2fs_put_page(page, 1);
1792
1793 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1794 if (unlikely(nid >= nm_i->max_nid))
1795 nid = 0;
1796
1797 if (++i >= FREE_NID_PAGES)
1798 break;
1799 }
1800
1801 /* go to the next free nat pages to find free nids abundantly */
1802 nm_i->next_scan_nid = nid;
1803
1804 /* find free nids from current sum_pages */
1805 down_read(&curseg->journal_rwsem);
1806 for (i = 0; i < nats_in_cursum(journal); i++) {
1807 block_t addr;
1808
1809 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1810 nid = le32_to_cpu(nid_in_journal(journal, i));
1811 if (addr == NULL_ADDR)
1812 add_free_nid(sbi, nid, true);
1813 else
1814 remove_free_nid(nm_i, nid);
1815 }
1816 up_read(&curseg->journal_rwsem);
1817 percpu_up_read(&nm_i->nat_tree_lock);
1818
1819 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1820 nm_i->ra_nid_pages, META_NAT, false);
1821 }
1822
1823 /*
1824 * If this function returns success, caller can obtain a new nid
1825 * from second parameter of this function.
1826 * The returned nid could be used ino as well as nid when inode is created.
1827 */
1828 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1829 {
1830 struct f2fs_nm_info *nm_i = NM_I(sbi);
1831 struct free_nid *i = NULL;
1832 retry:
1833 #ifdef CONFIG_F2FS_FAULT_INJECTION
1834 if (time_to_inject(FAULT_ALLOC_NID))
1835 return false;
1836 #endif
1837 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1838 return false;
1839
1840 spin_lock(&nm_i->free_nid_list_lock);
1841
1842 /* We should not use stale free nids created by build_free_nids */
1843 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1844 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1845 list_for_each_entry(i, &nm_i->free_nid_list, list)
1846 if (i->state == NID_NEW)
1847 break;
1848
1849 f2fs_bug_on(sbi, i->state != NID_NEW);
1850 *nid = i->nid;
1851 i->state = NID_ALLOC;
1852 nm_i->fcnt--;
1853 spin_unlock(&nm_i->free_nid_list_lock);
1854 return true;
1855 }
1856 spin_unlock(&nm_i->free_nid_list_lock);
1857
1858 /* Let's scan nat pages and its caches to get free nids */
1859 mutex_lock(&nm_i->build_lock);
1860 build_free_nids(sbi);
1861 mutex_unlock(&nm_i->build_lock);
1862 goto retry;
1863 }
1864
1865 /*
1866 * alloc_nid() should be called prior to this function.
1867 */
1868 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1869 {
1870 struct f2fs_nm_info *nm_i = NM_I(sbi);
1871 struct free_nid *i;
1872
1873 spin_lock(&nm_i->free_nid_list_lock);
1874 i = __lookup_free_nid_list(nm_i, nid);
1875 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1876 __del_from_free_nid_list(nm_i, i);
1877 spin_unlock(&nm_i->free_nid_list_lock);
1878
1879 kmem_cache_free(free_nid_slab, i);
1880 }
1881
1882 /*
1883 * alloc_nid() should be called prior to this function.
1884 */
1885 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1886 {
1887 struct f2fs_nm_info *nm_i = NM_I(sbi);
1888 struct free_nid *i;
1889 bool need_free = false;
1890
1891 if (!nid)
1892 return;
1893
1894 spin_lock(&nm_i->free_nid_list_lock);
1895 i = __lookup_free_nid_list(nm_i, nid);
1896 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1897 if (!available_free_memory(sbi, FREE_NIDS)) {
1898 __del_from_free_nid_list(nm_i, i);
1899 need_free = true;
1900 } else {
1901 i->state = NID_NEW;
1902 nm_i->fcnt++;
1903 }
1904 spin_unlock(&nm_i->free_nid_list_lock);
1905
1906 if (need_free)
1907 kmem_cache_free(free_nid_slab, i);
1908 }
1909
1910 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1911 {
1912 struct f2fs_nm_info *nm_i = NM_I(sbi);
1913 struct free_nid *i, *next;
1914 int nr = nr_shrink;
1915
1916 if (nm_i->fcnt <= MAX_FREE_NIDS)
1917 return 0;
1918
1919 if (!mutex_trylock(&nm_i->build_lock))
1920 return 0;
1921
1922 spin_lock(&nm_i->free_nid_list_lock);
1923 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1924 if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1925 break;
1926 if (i->state == NID_ALLOC)
1927 continue;
1928 __del_from_free_nid_list(nm_i, i);
1929 kmem_cache_free(free_nid_slab, i);
1930 nm_i->fcnt--;
1931 nr_shrink--;
1932 }
1933 spin_unlock(&nm_i->free_nid_list_lock);
1934 mutex_unlock(&nm_i->build_lock);
1935
1936 return nr - nr_shrink;
1937 }
1938
1939 void recover_inline_xattr(struct inode *inode, struct page *page)
1940 {
1941 void *src_addr, *dst_addr;
1942 size_t inline_size;
1943 struct page *ipage;
1944 struct f2fs_inode *ri;
1945
1946 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1947 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1948
1949 ri = F2FS_INODE(page);
1950 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1951 clear_inode_flag(inode, FI_INLINE_XATTR);
1952 goto update_inode;
1953 }
1954
1955 dst_addr = inline_xattr_addr(ipage);
1956 src_addr = inline_xattr_addr(page);
1957 inline_size = inline_xattr_size(inode);
1958
1959 f2fs_wait_on_page_writeback(ipage, NODE, true);
1960 memcpy(dst_addr, src_addr, inline_size);
1961 update_inode:
1962 update_inode(inode, ipage);
1963 f2fs_put_page(ipage, 1);
1964 }
1965
1966 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1967 {
1968 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1969 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1970 nid_t new_xnid = nid_of_node(page);
1971 struct node_info ni;
1972
1973 /* 1: invalidate the previous xattr nid */
1974 if (!prev_xnid)
1975 goto recover_xnid;
1976
1977 /* Deallocate node address */
1978 get_node_info(sbi, prev_xnid, &ni);
1979 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1980 invalidate_blocks(sbi, ni.blk_addr);
1981 dec_valid_node_count(sbi, inode);
1982 set_node_addr(sbi, &ni, NULL_ADDR, false);
1983
1984 recover_xnid:
1985 /* 2: allocate new xattr nid */
1986 if (unlikely(!inc_valid_node_count(sbi, inode)))
1987 f2fs_bug_on(sbi, 1);
1988
1989 remove_free_nid(NM_I(sbi), new_xnid);
1990 get_node_info(sbi, new_xnid, &ni);
1991 ni.ino = inode->i_ino;
1992 set_node_addr(sbi, &ni, NEW_ADDR, false);
1993 f2fs_i_xnid_write(inode, new_xnid);
1994
1995 /* 3: update xattr blkaddr */
1996 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1997 set_node_addr(sbi, &ni, blkaddr, false);
1998 }
1999
2000 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2001 {
2002 struct f2fs_inode *src, *dst;
2003 nid_t ino = ino_of_node(page);
2004 struct node_info old_ni, new_ni;
2005 struct page *ipage;
2006
2007 get_node_info(sbi, ino, &old_ni);
2008
2009 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2010 return -EINVAL;
2011
2012 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2013 if (!ipage)
2014 return -ENOMEM;
2015
2016 /* Should not use this inode from free nid list */
2017 remove_free_nid(NM_I(sbi), ino);
2018
2019 if (!PageUptodate(ipage))
2020 SetPageUptodate(ipage);
2021 fill_node_footer(ipage, ino, ino, 0, true);
2022
2023 src = F2FS_INODE(page);
2024 dst = F2FS_INODE(ipage);
2025
2026 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2027 dst->i_size = 0;
2028 dst->i_blocks = cpu_to_le64(1);
2029 dst->i_links = cpu_to_le32(1);
2030 dst->i_xattr_nid = 0;
2031 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2032
2033 new_ni = old_ni;
2034 new_ni.ino = ino;
2035
2036 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2037 WARN_ON(1);
2038 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2039 inc_valid_inode_count(sbi);
2040 set_page_dirty(ipage);
2041 f2fs_put_page(ipage, 1);
2042 return 0;
2043 }
2044
2045 int restore_node_summary(struct f2fs_sb_info *sbi,
2046 unsigned int segno, struct f2fs_summary_block *sum)
2047 {
2048 struct f2fs_node *rn;
2049 struct f2fs_summary *sum_entry;
2050 block_t addr;
2051 int bio_blocks = MAX_BIO_BLOCKS(sbi);
2052 int i, idx, last_offset, nrpages;
2053
2054 /* scan the node segment */
2055 last_offset = sbi->blocks_per_seg;
2056 addr = START_BLOCK(sbi, segno);
2057 sum_entry = &sum->entries[0];
2058
2059 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2060 nrpages = min(last_offset - i, bio_blocks);
2061
2062 /* readahead node pages */
2063 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2064
2065 for (idx = addr; idx < addr + nrpages; idx++) {
2066 struct page *page = get_tmp_page(sbi, idx);
2067
2068 rn = F2FS_NODE(page);
2069 sum_entry->nid = rn->footer.nid;
2070 sum_entry->version = 0;
2071 sum_entry->ofs_in_node = 0;
2072 sum_entry++;
2073 f2fs_put_page(page, 1);
2074 }
2075
2076 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2077 addr + nrpages);
2078 }
2079 return 0;
2080 }
2081
2082 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2083 {
2084 struct f2fs_nm_info *nm_i = NM_I(sbi);
2085 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2086 struct f2fs_journal *journal = curseg->journal;
2087 int i;
2088
2089 down_write(&curseg->journal_rwsem);
2090 for (i = 0; i < nats_in_cursum(journal); i++) {
2091 struct nat_entry *ne;
2092 struct f2fs_nat_entry raw_ne;
2093 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2094
2095 raw_ne = nat_in_journal(journal, i);
2096
2097 ne = __lookup_nat_cache(nm_i, nid);
2098 if (!ne) {
2099 ne = grab_nat_entry(nm_i, nid);
2100 node_info_from_raw_nat(&ne->ni, &raw_ne);
2101 }
2102 __set_nat_cache_dirty(nm_i, ne);
2103 }
2104 update_nats_in_cursum(journal, -i);
2105 up_write(&curseg->journal_rwsem);
2106 }
2107
2108 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2109 struct list_head *head, int max)
2110 {
2111 struct nat_entry_set *cur;
2112
2113 if (nes->entry_cnt >= max)
2114 goto add_out;
2115
2116 list_for_each_entry(cur, head, set_list) {
2117 if (cur->entry_cnt >= nes->entry_cnt) {
2118 list_add(&nes->set_list, cur->set_list.prev);
2119 return;
2120 }
2121 }
2122 add_out:
2123 list_add_tail(&nes->set_list, head);
2124 }
2125
2126 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2127 struct nat_entry_set *set)
2128 {
2129 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2130 struct f2fs_journal *journal = curseg->journal;
2131 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2132 bool to_journal = true;
2133 struct f2fs_nat_block *nat_blk;
2134 struct nat_entry *ne, *cur;
2135 struct page *page = NULL;
2136
2137 /*
2138 * there are two steps to flush nat entries:
2139 * #1, flush nat entries to journal in current hot data summary block.
2140 * #2, flush nat entries to nat page.
2141 */
2142 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2143 to_journal = false;
2144
2145 if (to_journal) {
2146 down_write(&curseg->journal_rwsem);
2147 } else {
2148 page = get_next_nat_page(sbi, start_nid);
2149 nat_blk = page_address(page);
2150 f2fs_bug_on(sbi, !nat_blk);
2151 }
2152
2153 /* flush dirty nats in nat entry set */
2154 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2155 struct f2fs_nat_entry *raw_ne;
2156 nid_t nid = nat_get_nid(ne);
2157 int offset;
2158
2159 if (nat_get_blkaddr(ne) == NEW_ADDR)
2160 continue;
2161
2162 if (to_journal) {
2163 offset = lookup_journal_in_cursum(journal,
2164 NAT_JOURNAL, nid, 1);
2165 f2fs_bug_on(sbi, offset < 0);
2166 raw_ne = &nat_in_journal(journal, offset);
2167 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2168 } else {
2169 raw_ne = &nat_blk->entries[nid - start_nid];
2170 }
2171 raw_nat_from_node_info(raw_ne, &ne->ni);
2172 nat_reset_flag(ne);
2173 __clear_nat_cache_dirty(NM_I(sbi), ne);
2174 if (nat_get_blkaddr(ne) == NULL_ADDR)
2175 add_free_nid(sbi, nid, false);
2176 }
2177
2178 if (to_journal)
2179 up_write(&curseg->journal_rwsem);
2180 else
2181 f2fs_put_page(page, 1);
2182
2183 f2fs_bug_on(sbi, set->entry_cnt);
2184
2185 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2186 kmem_cache_free(nat_entry_set_slab, set);
2187 }
2188
2189 /*
2190 * This function is called during the checkpointing process.
2191 */
2192 void flush_nat_entries(struct f2fs_sb_info *sbi)
2193 {
2194 struct f2fs_nm_info *nm_i = NM_I(sbi);
2195 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2196 struct f2fs_journal *journal = curseg->journal;
2197 struct nat_entry_set *setvec[SETVEC_SIZE];
2198 struct nat_entry_set *set, *tmp;
2199 unsigned int found;
2200 nid_t set_idx = 0;
2201 LIST_HEAD(sets);
2202
2203 if (!nm_i->dirty_nat_cnt)
2204 return;
2205
2206 percpu_down_write(&nm_i->nat_tree_lock);
2207
2208 /*
2209 * if there are no enough space in journal to store dirty nat
2210 * entries, remove all entries from journal and merge them
2211 * into nat entry set.
2212 */
2213 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2214 remove_nats_in_journal(sbi);
2215
2216 while ((found = __gang_lookup_nat_set(nm_i,
2217 set_idx, SETVEC_SIZE, setvec))) {
2218 unsigned idx;
2219 set_idx = setvec[found - 1]->set + 1;
2220 for (idx = 0; idx < found; idx++)
2221 __adjust_nat_entry_set(setvec[idx], &sets,
2222 MAX_NAT_JENTRIES(journal));
2223 }
2224
2225 /* flush dirty nats in nat entry set */
2226 list_for_each_entry_safe(set, tmp, &sets, set_list)
2227 __flush_nat_entry_set(sbi, set);
2228
2229 percpu_up_write(&nm_i->nat_tree_lock);
2230
2231 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2232 }
2233
2234 static int init_node_manager(struct f2fs_sb_info *sbi)
2235 {
2236 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2237 struct f2fs_nm_info *nm_i = NM_I(sbi);
2238 unsigned char *version_bitmap;
2239 unsigned int nat_segs, nat_blocks;
2240
2241 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2242
2243 /* segment_count_nat includes pair segment so divide to 2. */
2244 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2245 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2246
2247 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2248
2249 /* not used nids: 0, node, meta, (and root counted as valid node) */
2250 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2251 nm_i->fcnt = 0;
2252 nm_i->nat_cnt = 0;
2253 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2254 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2255 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2256
2257 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2258 INIT_LIST_HEAD(&nm_i->free_nid_list);
2259 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2260 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2261 INIT_LIST_HEAD(&nm_i->nat_entries);
2262
2263 mutex_init(&nm_i->build_lock);
2264 spin_lock_init(&nm_i->free_nid_list_lock);
2265 if (percpu_init_rwsem(&nm_i->nat_tree_lock))
2266 return -ENOMEM;
2267
2268 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2269 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2270 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2271 if (!version_bitmap)
2272 return -EFAULT;
2273
2274 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2275 GFP_KERNEL);
2276 if (!nm_i->nat_bitmap)
2277 return -ENOMEM;
2278 return 0;
2279 }
2280
2281 int build_node_manager(struct f2fs_sb_info *sbi)
2282 {
2283 int err;
2284
2285 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2286 if (!sbi->nm_info)
2287 return -ENOMEM;
2288
2289 err = init_node_manager(sbi);
2290 if (err)
2291 return err;
2292
2293 build_free_nids(sbi);
2294 return 0;
2295 }
2296
2297 void destroy_node_manager(struct f2fs_sb_info *sbi)
2298 {
2299 struct f2fs_nm_info *nm_i = NM_I(sbi);
2300 struct free_nid *i, *next_i;
2301 struct nat_entry *natvec[NATVEC_SIZE];
2302 struct nat_entry_set *setvec[SETVEC_SIZE];
2303 nid_t nid = 0;
2304 unsigned int found;
2305
2306 if (!nm_i)
2307 return;
2308
2309 /* destroy free nid list */
2310 spin_lock(&nm_i->free_nid_list_lock);
2311 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2312 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2313 __del_from_free_nid_list(nm_i, i);
2314 nm_i->fcnt--;
2315 spin_unlock(&nm_i->free_nid_list_lock);
2316 kmem_cache_free(free_nid_slab, i);
2317 spin_lock(&nm_i->free_nid_list_lock);
2318 }
2319 f2fs_bug_on(sbi, nm_i->fcnt);
2320 spin_unlock(&nm_i->free_nid_list_lock);
2321
2322 /* destroy nat cache */
2323 percpu_down_write(&nm_i->nat_tree_lock);
2324 while ((found = __gang_lookup_nat_cache(nm_i,
2325 nid, NATVEC_SIZE, natvec))) {
2326 unsigned idx;
2327
2328 nid = nat_get_nid(natvec[found - 1]) + 1;
2329 for (idx = 0; idx < found; idx++)
2330 __del_from_nat_cache(nm_i, natvec[idx]);
2331 }
2332 f2fs_bug_on(sbi, nm_i->nat_cnt);
2333
2334 /* destroy nat set cache */
2335 nid = 0;
2336 while ((found = __gang_lookup_nat_set(nm_i,
2337 nid, SETVEC_SIZE, setvec))) {
2338 unsigned idx;
2339
2340 nid = setvec[found - 1]->set + 1;
2341 for (idx = 0; idx < found; idx++) {
2342 /* entry_cnt is not zero, when cp_error was occurred */
2343 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2344 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2345 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2346 }
2347 }
2348 percpu_up_write(&nm_i->nat_tree_lock);
2349
2350 percpu_free_rwsem(&nm_i->nat_tree_lock);
2351 kfree(nm_i->nat_bitmap);
2352 sbi->nm_info = NULL;
2353 kfree(nm_i);
2354 }
2355
2356 int __init create_node_manager_caches(void)
2357 {
2358 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2359 sizeof(struct nat_entry));
2360 if (!nat_entry_slab)
2361 goto fail;
2362
2363 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2364 sizeof(struct free_nid));
2365 if (!free_nid_slab)
2366 goto destroy_nat_entry;
2367
2368 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2369 sizeof(struct nat_entry_set));
2370 if (!nat_entry_set_slab)
2371 goto destroy_free_nid;
2372 return 0;
2373
2374 destroy_free_nid:
2375 kmem_cache_destroy(free_nid_slab);
2376 destroy_nat_entry:
2377 kmem_cache_destroy(nat_entry_slab);
2378 fail:
2379 return -ENOMEM;
2380 }
2381
2382 void destroy_node_manager_caches(void)
2383 {
2384 kmem_cache_destroy(nat_entry_set_slab);
2385 kmem_cache_destroy(free_nid_slab);
2386 kmem_cache_destroy(nat_entry_slab);
2387 }
Linux kernel - Deliverables
This page took 0.112972 seconds and 5 git commands to generate.