projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: refactor __exchange_data_block for speed up
[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 dn->ofs_in_node = offset[level];
653 }
654 return err;
655 }
656
657 static void truncate_node(struct dnode_of_data *dn)
658 {
659 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
660 struct node_info ni;
661
662 get_node_info(sbi, dn->nid, &ni);
663 if (dn->inode->i_blocks == 0) {
664 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
665 goto invalidate;
666 }
667 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
668
669 /* Deallocate node address */
670 invalidate_blocks(sbi, ni.blk_addr);
671 dec_valid_node_count(sbi, dn->inode);
672 set_node_addr(sbi, &ni, NULL_ADDR, false);
673
674 if (dn->nid == dn->inode->i_ino) {
675 remove_orphan_inode(sbi, dn->nid);
676 dec_valid_inode_count(sbi);
677 f2fs_inode_synced(dn->inode);
678 }
679 invalidate:
680 clear_node_page_dirty(dn->node_page);
681 set_sbi_flag(sbi, SBI_IS_DIRTY);
682
683 f2fs_put_page(dn->node_page, 1);
684
685 invalidate_mapping_pages(NODE_MAPPING(sbi),
686 dn->node_page->index, dn->node_page->index);
687
688 dn->node_page = NULL;
689 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
690 }
691
692 static int truncate_dnode(struct dnode_of_data *dn)
693 {
694 struct page *page;
695
696 if (dn->nid == 0)
697 return 1;
698
699 /* get direct node */
700 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
701 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
702 return 1;
703 else if (IS_ERR(page))
704 return PTR_ERR(page);
705
706 /* Make dnode_of_data for parameter */
707 dn->node_page = page;
708 dn->ofs_in_node = 0;
709 truncate_data_blocks(dn);
710 truncate_node(dn);
711 return 1;
712 }
713
714 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
715 int ofs, int depth)
716 {
717 struct dnode_of_data rdn = *dn;
718 struct page *page;
719 struct f2fs_node *rn;
720 nid_t child_nid;
721 unsigned int child_nofs;
722 int freed = 0;
723 int i, ret;
724
725 if (dn->nid == 0)
726 return NIDS_PER_BLOCK + 1;
727
728 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
729
730 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
731 if (IS_ERR(page)) {
732 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
733 return PTR_ERR(page);
734 }
735
736 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
737
738 rn = F2FS_NODE(page);
739 if (depth < 3) {
740 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
741 child_nid = le32_to_cpu(rn->in.nid[i]);
742 if (child_nid == 0)
743 continue;
744 rdn.nid = child_nid;
745 ret = truncate_dnode(&rdn);
746 if (ret < 0)
747 goto out_err;
748 if (set_nid(page, i, 0, false))
749 dn->node_changed = true;
750 }
751 } else {
752 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
753 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
754 child_nid = le32_to_cpu(rn->in.nid[i]);
755 if (child_nid == 0) {
756 child_nofs += NIDS_PER_BLOCK + 1;
757 continue;
758 }
759 rdn.nid = child_nid;
760 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
761 if (ret == (NIDS_PER_BLOCK + 1)) {
762 if (set_nid(page, i, 0, false))
763 dn->node_changed = true;
764 child_nofs += ret;
765 } else if (ret < 0 && ret != -ENOENT) {
766 goto out_err;
767 }
768 }
769 freed = child_nofs;
770 }
771
772 if (!ofs) {
773 /* remove current indirect node */
774 dn->node_page = page;
775 truncate_node(dn);
776 freed++;
777 } else {
778 f2fs_put_page(page, 1);
779 }
780 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
781 return freed;
782
783 out_err:
784 f2fs_put_page(page, 1);
785 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
786 return ret;
787 }
788
789 static int truncate_partial_nodes(struct dnode_of_data *dn,
790 struct f2fs_inode *ri, int *offset, int depth)
791 {
792 struct page *pages[2];
793 nid_t nid[3];
794 nid_t child_nid;
795 int err = 0;
796 int i;
797 int idx = depth - 2;
798
799 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
800 if (!nid[0])
801 return 0;
802
803 /* get indirect nodes in the path */
804 for (i = 0; i < idx + 1; i++) {
805 /* reference count'll be increased */
806 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
807 if (IS_ERR(pages[i])) {
808 err = PTR_ERR(pages[i]);
809 idx = i - 1;
810 goto fail;
811 }
812 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
813 }
814
815 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
816
817 /* free direct nodes linked to a partial indirect node */
818 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
819 child_nid = get_nid(pages[idx], i, false);
820 if (!child_nid)
821 continue;
822 dn->nid = child_nid;
823 err = truncate_dnode(dn);
824 if (err < 0)
825 goto fail;
826 if (set_nid(pages[idx], i, 0, false))
827 dn->node_changed = true;
828 }
829
830 if (offset[idx + 1] == 0) {
831 dn->node_page = pages[idx];
832 dn->nid = nid[idx];
833 truncate_node(dn);
834 } else {
835 f2fs_put_page(pages[idx], 1);
836 }
837 offset[idx]++;
838 offset[idx + 1] = 0;
839 idx--;
840 fail:
841 for (i = idx; i >= 0; i--)
842 f2fs_put_page(pages[i], 1);
843
844 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
845
846 return err;
847 }
848
849 /*
850 * All the block addresses of data and nodes should be nullified.
851 */
852 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
853 {
854 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
855 int err = 0, cont = 1;
856 int level, offset[4], noffset[4];
857 unsigned int nofs = 0;
858 struct f2fs_inode *ri;
859 struct dnode_of_data dn;
860 struct page *page;
861
862 trace_f2fs_truncate_inode_blocks_enter(inode, from);
863
864 level = get_node_path(inode, from, offset, noffset);
865
866 page = get_node_page(sbi, inode->i_ino);
867 if (IS_ERR(page)) {
868 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
869 return PTR_ERR(page);
870 }
871
872 set_new_dnode(&dn, inode, page, NULL, 0);
873 unlock_page(page);
874
875 ri = F2FS_INODE(page);
876 switch (level) {
877 case 0:
878 case 1:
879 nofs = noffset[1];
880 break;
881 case 2:
882 nofs = noffset[1];
883 if (!offset[level - 1])
884 goto skip_partial;
885 err = truncate_partial_nodes(&dn, ri, offset, level);
886 if (err < 0 && err != -ENOENT)
887 goto fail;
888 nofs += 1 + NIDS_PER_BLOCK;
889 break;
890 case 3:
891 nofs = 5 + 2 * NIDS_PER_BLOCK;
892 if (!offset[level - 1])
893 goto skip_partial;
894 err = truncate_partial_nodes(&dn, ri, offset, level);
895 if (err < 0 && err != -ENOENT)
896 goto fail;
897 break;
898 default:
899 BUG();
900 }
901
902 skip_partial:
903 while (cont) {
904 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
905 switch (offset[0]) {
906 case NODE_DIR1_BLOCK:
907 case NODE_DIR2_BLOCK:
908 err = truncate_dnode(&dn);
909 break;
910
911 case NODE_IND1_BLOCK:
912 case NODE_IND2_BLOCK:
913 err = truncate_nodes(&dn, nofs, offset[1], 2);
914 break;
915
916 case NODE_DIND_BLOCK:
917 err = truncate_nodes(&dn, nofs, offset[1], 3);
918 cont = 0;
919 break;
920
921 default:
922 BUG();
923 }
924 if (err < 0 && err != -ENOENT)
925 goto fail;
926 if (offset[1] == 0 &&
927 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
928 lock_page(page);
929 BUG_ON(page->mapping != NODE_MAPPING(sbi));
930 f2fs_wait_on_page_writeback(page, NODE, true);
931 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
932 set_page_dirty(page);
933 unlock_page(page);
934 }
935 offset[1] = 0;
936 offset[0]++;
937 nofs += err;
938 }
939 fail:
940 f2fs_put_page(page, 0);
941 trace_f2fs_truncate_inode_blocks_exit(inode, err);
942 return err > 0 ? 0 : err;
943 }
944
945 int truncate_xattr_node(struct inode *inode, struct page *page)
946 {
947 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
948 nid_t nid = F2FS_I(inode)->i_xattr_nid;
949 struct dnode_of_data dn;
950 struct page *npage;
951
952 if (!nid)
953 return 0;
954
955 npage = get_node_page(sbi, nid);
956 if (IS_ERR(npage))
957 return PTR_ERR(npage);
958
959 f2fs_i_xnid_write(inode, 0);
960
961 /* need to do checkpoint during fsync */
962 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
963
964 set_new_dnode(&dn, inode, page, npage, nid);
965
966 if (page)
967 dn.inode_page_locked = true;
968 truncate_node(&dn);
969 return 0;
970 }
971
972 /*
973 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
974 * f2fs_unlock_op().
975 */
976 int remove_inode_page(struct inode *inode)
977 {
978 struct dnode_of_data dn;
979 int err;
980
981 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
982 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
983 if (err)
984 return err;
985
986 err = truncate_xattr_node(inode, dn.inode_page);
987 if (err) {
988 f2fs_put_dnode(&dn);
989 return err;
990 }
991
992 /* remove potential inline_data blocks */
993 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
994 S_ISLNK(inode->i_mode))
995 truncate_data_blocks_range(&dn, 1);
996
997 /* 0 is possible, after f2fs_new_inode() has failed */
998 f2fs_bug_on(F2FS_I_SB(inode),
999 inode->i_blocks != 0 && inode->i_blocks != 1);
1000
1001 /* will put inode & node pages */
1002 truncate_node(&dn);
1003 return 0;
1004 }
1005
1006 struct page *new_inode_page(struct inode *inode)
1007 {
1008 struct dnode_of_data dn;
1009
1010 /* allocate inode page for new inode */
1011 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1012
1013 /* caller should f2fs_put_page(page, 1); */
1014 return new_node_page(&dn, 0, NULL);
1015 }
1016
1017 struct page *new_node_page(struct dnode_of_data *dn,
1018 unsigned int ofs, struct page *ipage)
1019 {
1020 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1021 struct node_info old_ni, new_ni;
1022 struct page *page;
1023 int err;
1024
1025 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1026 return ERR_PTR(-EPERM);
1027
1028 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1029 if (!page)
1030 return ERR_PTR(-ENOMEM);
1031
1032 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1033 err = -ENOSPC;
1034 goto fail;
1035 }
1036
1037 get_node_info(sbi, dn->nid, &old_ni);
1038
1039 /* Reinitialize old_ni with new node page */
1040 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1041 new_ni = old_ni;
1042 new_ni.ino = dn->inode->i_ino;
1043 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1044
1045 f2fs_wait_on_page_writeback(page, NODE, true);
1046 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1047 set_cold_node(dn->inode, page);
1048 if (!PageUptodate(page))
1049 SetPageUptodate(page);
1050 if (set_page_dirty(page))
1051 dn->node_changed = true;
1052
1053 if (f2fs_has_xattr_block(ofs))
1054 f2fs_i_xnid_write(dn->inode, dn->nid);
1055
1056 if (ofs == 0)
1057 inc_valid_inode_count(sbi);
1058 return page;
1059
1060 fail:
1061 clear_node_page_dirty(page);
1062 f2fs_put_page(page, 1);
1063 return ERR_PTR(err);
1064 }
1065
1066 /*
1067 * Caller should do after getting the following values.
1068 * 0: f2fs_put_page(page, 0)
1069 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1070 */
1071 static int read_node_page(struct page *page, int rw)
1072 {
1073 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1074 struct node_info ni;
1075 struct f2fs_io_info fio = {
1076 .sbi = sbi,
1077 .type = NODE,
1078 .rw = rw,
1079 .page = page,
1080 .encrypted_page = NULL,
1081 };
1082
1083 if (PageUptodate(page))
1084 return LOCKED_PAGE;
1085
1086 get_node_info(sbi, page->index, &ni);
1087
1088 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1089 ClearPageUptodate(page);
1090 return -ENOENT;
1091 }
1092
1093 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1094 return f2fs_submit_page_bio(&fio);
1095 }
1096
1097 /*
1098 * Readahead a node page
1099 */
1100 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1101 {
1102 struct page *apage;
1103 int err;
1104
1105 if (!nid)
1106 return;
1107 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1108
1109 rcu_read_lock();
1110 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1111 rcu_read_unlock();
1112 if (apage)
1113 return;
1114
1115 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1116 if (!apage)
1117 return;
1118
1119 err = read_node_page(apage, READA);
1120 f2fs_put_page(apage, err ? 1 : 0);
1121 }
1122
1123 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1124 struct page *parent, int start)
1125 {
1126 struct page *page;
1127 int err;
1128
1129 if (!nid)
1130 return ERR_PTR(-ENOENT);
1131 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1132 repeat:
1133 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1134 if (!page)
1135 return ERR_PTR(-ENOMEM);
1136
1137 err = read_node_page(page, READ_SYNC);
1138 if (err < 0) {
1139 f2fs_put_page(page, 1);
1140 return ERR_PTR(err);
1141 } else if (err == LOCKED_PAGE) {
1142 goto page_hit;
1143 }
1144
1145 if (parent)
1146 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1147
1148 lock_page(page);
1149
1150 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1151 f2fs_put_page(page, 1);
1152 goto repeat;
1153 }
1154
1155 if (unlikely(!PageUptodate(page)))
1156 goto out_err;
1157 page_hit:
1158 if(unlikely(nid != nid_of_node(page))) {
1159 f2fs_bug_on(sbi, 1);
1160 ClearPageUptodate(page);
1161 out_err:
1162 f2fs_put_page(page, 1);
1163 return ERR_PTR(-EIO);
1164 }
1165 return page;
1166 }
1167
1168 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1169 {
1170 return __get_node_page(sbi, nid, NULL, 0);
1171 }
1172
1173 struct page *get_node_page_ra(struct page *parent, int start)
1174 {
1175 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1176 nid_t nid = get_nid(parent, start, false);
1177
1178 return __get_node_page(sbi, nid, parent, start);
1179 }
1180
1181 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1182 {
1183 struct inode *inode;
1184 struct page *page;
1185 int ret;
1186
1187 /* should flush inline_data before evict_inode */
1188 inode = ilookup(sbi->sb, ino);
1189 if (!inode)
1190 return;
1191
1192 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1193 if (!page)
1194 goto iput_out;
1195
1196 if (!PageUptodate(page))
1197 goto page_out;
1198
1199 if (!PageDirty(page))
1200 goto page_out;
1201
1202 if (!clear_page_dirty_for_io(page))
1203 goto page_out;
1204
1205 ret = f2fs_write_inline_data(inode, page);
1206 inode_dec_dirty_pages(inode);
1207 if (ret)
1208 set_page_dirty(page);
1209 page_out:
1210 f2fs_put_page(page, 1);
1211 iput_out:
1212 iput(inode);
1213 }
1214
1215 void move_node_page(struct page *node_page, int gc_type)
1216 {
1217 if (gc_type == FG_GC) {
1218 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1219 struct writeback_control wbc = {
1220 .sync_mode = WB_SYNC_ALL,
1221 .nr_to_write = 1,
1222 .for_reclaim = 0,
1223 };
1224
1225 set_page_dirty(node_page);
1226 f2fs_wait_on_page_writeback(node_page, NODE, true);
1227
1228 f2fs_bug_on(sbi, PageWriteback(node_page));
1229 if (!clear_page_dirty_for_io(node_page))
1230 goto out_page;
1231
1232 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1233 unlock_page(node_page);
1234 goto release_page;
1235 } else {
1236 /* set page dirty and write it */
1237 if (!PageWriteback(node_page))
1238 set_page_dirty(node_page);
1239 }
1240 out_page:
1241 unlock_page(node_page);
1242 release_page:
1243 f2fs_put_page(node_page, 0);
1244 }
1245
1246 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1247 {
1248 pgoff_t index, end;
1249 struct pagevec pvec;
1250 struct page *last_page = NULL;
1251
1252 pagevec_init(&pvec, 0);
1253 index = 0;
1254 end = ULONG_MAX;
1255
1256 while (index <= end) {
1257 int i, nr_pages;
1258 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1259 PAGECACHE_TAG_DIRTY,
1260 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1261 if (nr_pages == 0)
1262 break;
1263
1264 for (i = 0; i < nr_pages; i++) {
1265 struct page *page = pvec.pages[i];
1266
1267 if (unlikely(f2fs_cp_error(sbi))) {
1268 f2fs_put_page(last_page, 0);
1269 pagevec_release(&pvec);
1270 return ERR_PTR(-EIO);
1271 }
1272
1273 if (!IS_DNODE(page) || !is_cold_node(page))
1274 continue;
1275 if (ino_of_node(page) != ino)
1276 continue;
1277
1278 lock_page(page);
1279
1280 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1281 continue_unlock:
1282 unlock_page(page);
1283 continue;
1284 }
1285 if (ino_of_node(page) != ino)
1286 goto continue_unlock;
1287
1288 if (!PageDirty(page)) {
1289 /* someone wrote it for us */
1290 goto continue_unlock;
1291 }
1292
1293 if (last_page)
1294 f2fs_put_page(last_page, 0);
1295
1296 get_page(page);
1297 last_page = page;
1298 unlock_page(page);
1299 }
1300 pagevec_release(&pvec);
1301 cond_resched();
1302 }
1303 return last_page;
1304 }
1305
1306 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1307 struct writeback_control *wbc, bool atomic)
1308 {
1309 pgoff_t index, end;
1310 struct pagevec pvec;
1311 int ret = 0;
1312 struct page *last_page = NULL;
1313 bool marked = false;
1314 nid_t ino = inode->i_ino;
1315
1316 if (atomic) {
1317 last_page = last_fsync_dnode(sbi, ino);
1318 if (IS_ERR_OR_NULL(last_page))
1319 return PTR_ERR_OR_ZERO(last_page);
1320 }
1321 retry:
1322 pagevec_init(&pvec, 0);
1323 index = 0;
1324 end = ULONG_MAX;
1325
1326 while (index <= end) {
1327 int i, nr_pages;
1328 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1329 PAGECACHE_TAG_DIRTY,
1330 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1331 if (nr_pages == 0)
1332 break;
1333
1334 for (i = 0; i < nr_pages; i++) {
1335 struct page *page = pvec.pages[i];
1336
1337 if (unlikely(f2fs_cp_error(sbi))) {
1338 f2fs_put_page(last_page, 0);
1339 pagevec_release(&pvec);
1340 return -EIO;
1341 }
1342
1343 if (!IS_DNODE(page) || !is_cold_node(page))
1344 continue;
1345 if (ino_of_node(page) != ino)
1346 continue;
1347
1348 lock_page(page);
1349
1350 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1351 continue_unlock:
1352 unlock_page(page);
1353 continue;
1354 }
1355 if (ino_of_node(page) != ino)
1356 goto continue_unlock;
1357
1358 if (!PageDirty(page) && page != last_page) {
1359 /* someone wrote it for us */
1360 goto continue_unlock;
1361 }
1362
1363 f2fs_wait_on_page_writeback(page, NODE, true);
1364 BUG_ON(PageWriteback(page));
1365
1366 if (!atomic || page == last_page) {
1367 set_fsync_mark(page, 1);
1368 if (IS_INODE(page)) {
1369 if (is_inode_flag_set(inode,
1370 FI_DIRTY_INODE))
1371 update_inode(inode, page);
1372 set_dentry_mark(page,
1373 need_dentry_mark(sbi, ino));
1374 }
1375 /* may be written by other thread */
1376 if (!PageDirty(page))
1377 set_page_dirty(page);
1378 }
1379
1380 if (!clear_page_dirty_for_io(page))
1381 goto continue_unlock;
1382
1383 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1384 if (ret) {
1385 unlock_page(page);
1386 f2fs_put_page(last_page, 0);
1387 break;
1388 }
1389 if (page == last_page) {
1390 f2fs_put_page(page, 0);
1391 marked = true;
1392 break;
1393 }
1394 }
1395 pagevec_release(&pvec);
1396 cond_resched();
1397
1398 if (ret || marked)
1399 break;
1400 }
1401 if (!ret && atomic && !marked) {
1402 f2fs_msg(sbi->sb, KERN_DEBUG,
1403 "Retry to write fsync mark: ino=%u, idx=%lx",
1404 ino, last_page->index);
1405 lock_page(last_page);
1406 set_page_dirty(last_page);
1407 unlock_page(last_page);
1408 goto retry;
1409 }
1410 return ret ? -EIO: 0;
1411 }
1412
1413 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1414 {
1415 pgoff_t index, end;
1416 struct pagevec pvec;
1417 int step = 0;
1418 int nwritten = 0;
1419
1420 pagevec_init(&pvec, 0);
1421
1422 next_step:
1423 index = 0;
1424 end = ULONG_MAX;
1425
1426 while (index <= end) {
1427 int i, nr_pages;
1428 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1429 PAGECACHE_TAG_DIRTY,
1430 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1431 if (nr_pages == 0)
1432 break;
1433
1434 for (i = 0; i < nr_pages; i++) {
1435 struct page *page = pvec.pages[i];
1436
1437 if (unlikely(f2fs_cp_error(sbi))) {
1438 pagevec_release(&pvec);
1439 return -EIO;
1440 }
1441
1442 /*
1443 * flushing sequence with step:
1444 * 0. indirect nodes
1445 * 1. dentry dnodes
1446 * 2. file dnodes
1447 */
1448 if (step == 0 && IS_DNODE(page))
1449 continue;
1450 if (step == 1 && (!IS_DNODE(page) ||
1451 is_cold_node(page)))
1452 continue;
1453 if (step == 2 && (!IS_DNODE(page) ||
1454 !is_cold_node(page)))
1455 continue;
1456 lock_node:
1457 if (!trylock_page(page))
1458 continue;
1459
1460 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1461 continue_unlock:
1462 unlock_page(page);
1463 continue;
1464 }
1465
1466 if (!PageDirty(page)) {
1467 /* someone wrote it for us */
1468 goto continue_unlock;
1469 }
1470
1471 /* flush inline_data */
1472 if (is_inline_node(page)) {
1473 clear_inline_node(page);
1474 unlock_page(page);
1475 flush_inline_data(sbi, ino_of_node(page));
1476 goto lock_node;
1477 }
1478
1479 f2fs_wait_on_page_writeback(page, NODE, true);
1480
1481 BUG_ON(PageWriteback(page));
1482 if (!clear_page_dirty_for_io(page))
1483 goto continue_unlock;
1484
1485 set_fsync_mark(page, 0);
1486 set_dentry_mark(page, 0);
1487
1488 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1489 unlock_page(page);
1490
1491 if (--wbc->nr_to_write == 0)
1492 break;
1493 }
1494 pagevec_release(&pvec);
1495 cond_resched();
1496
1497 if (wbc->nr_to_write == 0) {
1498 step = 2;
1499 break;
1500 }
1501 }
1502
1503 if (step < 2) {
1504 step++;
1505 goto next_step;
1506 }
1507 return nwritten;
1508 }
1509
1510 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1511 {
1512 pgoff_t index = 0, end = ULONG_MAX;
1513 struct pagevec pvec;
1514 int ret2 = 0, ret = 0;
1515
1516 pagevec_init(&pvec, 0);
1517
1518 while (index <= end) {
1519 int i, nr_pages;
1520 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1521 PAGECACHE_TAG_WRITEBACK,
1522 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1523 if (nr_pages == 0)
1524 break;
1525
1526 for (i = 0; i < nr_pages; i++) {
1527 struct page *page = pvec.pages[i];
1528
1529 /* until radix tree lookup accepts end_index */
1530 if (unlikely(page->index > end))
1531 continue;
1532
1533 if (ino && ino_of_node(page) == ino) {
1534 f2fs_wait_on_page_writeback(page, NODE, true);
1535 if (TestClearPageError(page))
1536 ret = -EIO;
1537 }
1538 }
1539 pagevec_release(&pvec);
1540 cond_resched();
1541 }
1542
1543 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1544 ret2 = -ENOSPC;
1545 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1546 ret2 = -EIO;
1547 if (!ret)
1548 ret = ret2;
1549 return ret;
1550 }
1551
1552 static int f2fs_write_node_page(struct page *page,
1553 struct writeback_control *wbc)
1554 {
1555 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1556 nid_t nid;
1557 struct node_info ni;
1558 struct f2fs_io_info fio = {
1559 .sbi = sbi,
1560 .type = NODE,
1561 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1562 .page = page,
1563 .encrypted_page = NULL,
1564 };
1565
1566 trace_f2fs_writepage(page, NODE);
1567
1568 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1569 goto redirty_out;
1570 if (unlikely(f2fs_cp_error(sbi)))
1571 goto redirty_out;
1572
1573 /* get old block addr of this node page */
1574 nid = nid_of_node(page);
1575 f2fs_bug_on(sbi, page->index != nid);
1576
1577 if (wbc->for_reclaim) {
1578 if (!down_read_trylock(&sbi->node_write))
1579 goto redirty_out;
1580 } else {
1581 down_read(&sbi->node_write);
1582 }
1583
1584 get_node_info(sbi, nid, &ni);
1585
1586 /* This page is already truncated */
1587 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1588 ClearPageUptodate(page);
1589 dec_page_count(sbi, F2FS_DIRTY_NODES);
1590 up_read(&sbi->node_write);
1591 unlock_page(page);
1592 return 0;
1593 }
1594
1595 set_page_writeback(page);
1596 fio.old_blkaddr = ni.blk_addr;
1597 write_node_page(nid, &fio);
1598 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1599 dec_page_count(sbi, F2FS_DIRTY_NODES);
1600 up_read(&sbi->node_write);
1601
1602 if (wbc->for_reclaim)
1603 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1604
1605 unlock_page(page);
1606
1607 if (unlikely(f2fs_cp_error(sbi)))
1608 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1609
1610 return 0;
1611
1612 redirty_out:
1613 redirty_page_for_writepage(wbc, page);
1614 return AOP_WRITEPAGE_ACTIVATE;
1615 }
1616
1617 static int f2fs_write_node_pages(struct address_space *mapping,
1618 struct writeback_control *wbc)
1619 {
1620 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1621 long diff;
1622
1623 /* balancing f2fs's metadata in background */
1624 f2fs_balance_fs_bg(sbi);
1625
1626 /* collect a number of dirty node pages and write together */
1627 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1628 goto skip_write;
1629
1630 trace_f2fs_writepages(mapping->host, wbc, NODE);
1631
1632 diff = nr_pages_to_write(sbi, NODE, wbc);
1633 wbc->sync_mode = WB_SYNC_NONE;
1634 sync_node_pages(sbi, wbc);
1635 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1636 return 0;
1637
1638 skip_write:
1639 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1640 trace_f2fs_writepages(mapping->host, wbc, NODE);
1641 return 0;
1642 }
1643
1644 static int f2fs_set_node_page_dirty(struct page *page)
1645 {
1646 trace_f2fs_set_page_dirty(page, NODE);
1647
1648 if (!PageUptodate(page))
1649 SetPageUptodate(page);
1650 if (!PageDirty(page)) {
1651 f2fs_set_page_dirty_nobuffers(page);
1652 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1653 SetPagePrivate(page);
1654 f2fs_trace_pid(page);
1655 return 1;
1656 }
1657 return 0;
1658 }
1659
1660 /*
1661 * Structure of the f2fs node operations
1662 */
1663 const struct address_space_operations f2fs_node_aops = {
1664 .writepage = f2fs_write_node_page,
1665 .writepages = f2fs_write_node_pages,
1666 .set_page_dirty = f2fs_set_node_page_dirty,
1667 .invalidatepage = f2fs_invalidate_page,
1668 .releasepage = f2fs_release_page,
1669 };
1670
1671 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1672 nid_t n)
1673 {
1674 return radix_tree_lookup(&nm_i->free_nid_root, n);
1675 }
1676
1677 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1678 struct free_nid *i)
1679 {
1680 list_del(&i->list);
1681 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1682 }
1683
1684 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1685 {
1686 struct f2fs_nm_info *nm_i = NM_I(sbi);
1687 struct free_nid *i;
1688 struct nat_entry *ne;
1689
1690 if (!available_free_memory(sbi, FREE_NIDS))
1691 return -1;
1692
1693 /* 0 nid should not be used */
1694 if (unlikely(nid == 0))
1695 return 0;
1696
1697 if (build) {
1698 /* do not add allocated nids */
1699 ne = __lookup_nat_cache(nm_i, nid);
1700 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1701 nat_get_blkaddr(ne) != NULL_ADDR))
1702 return 0;
1703 }
1704
1705 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1706 i->nid = nid;
1707 i->state = NID_NEW;
1708
1709 if (radix_tree_preload(GFP_NOFS)) {
1710 kmem_cache_free(free_nid_slab, i);
1711 return 0;
1712 }
1713
1714 spin_lock(&nm_i->free_nid_list_lock);
1715 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1716 spin_unlock(&nm_i->free_nid_list_lock);
1717 radix_tree_preload_end();
1718 kmem_cache_free(free_nid_slab, i);
1719 return 0;
1720 }
1721 list_add_tail(&i->list, &nm_i->free_nid_list);
1722 nm_i->fcnt++;
1723 spin_unlock(&nm_i->free_nid_list_lock);
1724 radix_tree_preload_end();
1725 return 1;
1726 }
1727
1728 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1729 {
1730 struct free_nid *i;
1731 bool need_free = false;
1732
1733 spin_lock(&nm_i->free_nid_list_lock);
1734 i = __lookup_free_nid_list(nm_i, nid);
1735 if (i && i->state == NID_NEW) {
1736 __del_from_free_nid_list(nm_i, i);
1737 nm_i->fcnt--;
1738 need_free = true;
1739 }
1740 spin_unlock(&nm_i->free_nid_list_lock);
1741
1742 if (need_free)
1743 kmem_cache_free(free_nid_slab, i);
1744 }
1745
1746 static void scan_nat_page(struct f2fs_sb_info *sbi,
1747 struct page *nat_page, nid_t start_nid)
1748 {
1749 struct f2fs_nm_info *nm_i = NM_I(sbi);
1750 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1751 block_t blk_addr;
1752 int i;
1753
1754 i = start_nid % NAT_ENTRY_PER_BLOCK;
1755
1756 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1757
1758 if (unlikely(start_nid >= nm_i->max_nid))
1759 break;
1760
1761 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1762 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1763 if (blk_addr == NULL_ADDR) {
1764 if (add_free_nid(sbi, start_nid, true) < 0)
1765 break;
1766 }
1767 }
1768 }
1769
1770 void build_free_nids(struct f2fs_sb_info *sbi)
1771 {
1772 struct f2fs_nm_info *nm_i = NM_I(sbi);
1773 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1774 struct f2fs_journal *journal = curseg->journal;
1775 int i = 0;
1776 nid_t nid = nm_i->next_scan_nid;
1777
1778 /* Enough entries */
1779 if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1780 return;
1781
1782 /* readahead nat pages to be scanned */
1783 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1784 META_NAT, true);
1785
1786 percpu_down_read(&nm_i->nat_tree_lock);
1787
1788 while (1) {
1789 struct page *page = get_current_nat_page(sbi, nid);
1790
1791 scan_nat_page(sbi, page, nid);
1792 f2fs_put_page(page, 1);
1793
1794 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1795 if (unlikely(nid >= nm_i->max_nid))
1796 nid = 0;
1797
1798 if (++i >= FREE_NID_PAGES)
1799 break;
1800 }
1801
1802 /* go to the next free nat pages to find free nids abundantly */
1803 nm_i->next_scan_nid = nid;
1804
1805 /* find free nids from current sum_pages */
1806 down_read(&curseg->journal_rwsem);
1807 for (i = 0; i < nats_in_cursum(journal); i++) {
1808 block_t addr;
1809
1810 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1811 nid = le32_to_cpu(nid_in_journal(journal, i));
1812 if (addr == NULL_ADDR)
1813 add_free_nid(sbi, nid, true);
1814 else
1815 remove_free_nid(nm_i, nid);
1816 }
1817 up_read(&curseg->journal_rwsem);
1818 percpu_up_read(&nm_i->nat_tree_lock);
1819
1820 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1821 nm_i->ra_nid_pages, META_NAT, false);
1822 }
1823
1824 /*
1825 * If this function returns success, caller can obtain a new nid
1826 * from second parameter of this function.
1827 * The returned nid could be used ino as well as nid when inode is created.
1828 */
1829 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1830 {
1831 struct f2fs_nm_info *nm_i = NM_I(sbi);
1832 struct free_nid *i = NULL;
1833 retry:
1834 #ifdef CONFIG_F2FS_FAULT_INJECTION
1835 if (time_to_inject(FAULT_ALLOC_NID))
1836 return false;
1837 #endif
1838 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1839 return false;
1840
1841 spin_lock(&nm_i->free_nid_list_lock);
1842
1843 /* We should not use stale free nids created by build_free_nids */
1844 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1845 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1846 list_for_each_entry(i, &nm_i->free_nid_list, list)
1847 if (i->state == NID_NEW)
1848 break;
1849
1850 f2fs_bug_on(sbi, i->state != NID_NEW);
1851 *nid = i->nid;
1852 i->state = NID_ALLOC;
1853 nm_i->fcnt--;
1854 spin_unlock(&nm_i->free_nid_list_lock);
1855 return true;
1856 }
1857 spin_unlock(&nm_i->free_nid_list_lock);
1858
1859 /* Let's scan nat pages and its caches to get free nids */
1860 mutex_lock(&nm_i->build_lock);
1861 build_free_nids(sbi);
1862 mutex_unlock(&nm_i->build_lock);
1863 goto retry;
1864 }
1865
1866 /*
1867 * alloc_nid() should be called prior to this function.
1868 */
1869 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1870 {
1871 struct f2fs_nm_info *nm_i = NM_I(sbi);
1872 struct free_nid *i;
1873
1874 spin_lock(&nm_i->free_nid_list_lock);
1875 i = __lookup_free_nid_list(nm_i, nid);
1876 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1877 __del_from_free_nid_list(nm_i, i);
1878 spin_unlock(&nm_i->free_nid_list_lock);
1879
1880 kmem_cache_free(free_nid_slab, i);
1881 }
1882
1883 /*
1884 * alloc_nid() should be called prior to this function.
1885 */
1886 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1887 {
1888 struct f2fs_nm_info *nm_i = NM_I(sbi);
1889 struct free_nid *i;
1890 bool need_free = false;
1891
1892 if (!nid)
1893 return;
1894
1895 spin_lock(&nm_i->free_nid_list_lock);
1896 i = __lookup_free_nid_list(nm_i, nid);
1897 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1898 if (!available_free_memory(sbi, FREE_NIDS)) {
1899 __del_from_free_nid_list(nm_i, i);
1900 need_free = true;
1901 } else {
1902 i->state = NID_NEW;
1903 nm_i->fcnt++;
1904 }
1905 spin_unlock(&nm_i->free_nid_list_lock);
1906
1907 if (need_free)
1908 kmem_cache_free(free_nid_slab, i);
1909 }
1910
1911 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1912 {
1913 struct f2fs_nm_info *nm_i = NM_I(sbi);
1914 struct free_nid *i, *next;
1915 int nr = nr_shrink;
1916
1917 if (nm_i->fcnt <= MAX_FREE_NIDS)
1918 return 0;
1919
1920 if (!mutex_trylock(&nm_i->build_lock))
1921 return 0;
1922
1923 spin_lock(&nm_i->free_nid_list_lock);
1924 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1925 if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1926 break;
1927 if (i->state == NID_ALLOC)
1928 continue;
1929 __del_from_free_nid_list(nm_i, i);
1930 kmem_cache_free(free_nid_slab, i);
1931 nm_i->fcnt--;
1932 nr_shrink--;
1933 }
1934 spin_unlock(&nm_i->free_nid_list_lock);
1935 mutex_unlock(&nm_i->build_lock);
1936
1937 return nr - nr_shrink;
1938 }
1939
1940 void recover_inline_xattr(struct inode *inode, struct page *page)
1941 {
1942 void *src_addr, *dst_addr;
1943 size_t inline_size;
1944 struct page *ipage;
1945 struct f2fs_inode *ri;
1946
1947 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1948 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1949
1950 ri = F2FS_INODE(page);
1951 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1952 clear_inode_flag(inode, FI_INLINE_XATTR);
1953 goto update_inode;
1954 }
1955
1956 dst_addr = inline_xattr_addr(ipage);
1957 src_addr = inline_xattr_addr(page);
1958 inline_size = inline_xattr_size(inode);
1959
1960 f2fs_wait_on_page_writeback(ipage, NODE, true);
1961 memcpy(dst_addr, src_addr, inline_size);
1962 update_inode:
1963 update_inode(inode, ipage);
1964 f2fs_put_page(ipage, 1);
1965 }
1966
1967 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1968 {
1969 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1970 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1971 nid_t new_xnid = nid_of_node(page);
1972 struct node_info ni;
1973
1974 /* 1: invalidate the previous xattr nid */
1975 if (!prev_xnid)
1976 goto recover_xnid;
1977
1978 /* Deallocate node address */
1979 get_node_info(sbi, prev_xnid, &ni);
1980 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1981 invalidate_blocks(sbi, ni.blk_addr);
1982 dec_valid_node_count(sbi, inode);
1983 set_node_addr(sbi, &ni, NULL_ADDR, false);
1984
1985 recover_xnid:
1986 /* 2: allocate new xattr nid */
1987 if (unlikely(!inc_valid_node_count(sbi, inode)))
1988 f2fs_bug_on(sbi, 1);
1989
1990 remove_free_nid(NM_I(sbi), new_xnid);
1991 get_node_info(sbi, new_xnid, &ni);
1992 ni.ino = inode->i_ino;
1993 set_node_addr(sbi, &ni, NEW_ADDR, false);
1994 f2fs_i_xnid_write(inode, new_xnid);
1995
1996 /* 3: update xattr blkaddr */
1997 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1998 set_node_addr(sbi, &ni, blkaddr, false);
1999 }
2000
2001 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2002 {
2003 struct f2fs_inode *src, *dst;
2004 nid_t ino = ino_of_node(page);
2005 struct node_info old_ni, new_ni;
2006 struct page *ipage;
2007
2008 get_node_info(sbi, ino, &old_ni);
2009
2010 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2011 return -EINVAL;
2012
2013 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2014 if (!ipage)
2015 return -ENOMEM;
2016
2017 /* Should not use this inode from free nid list */
2018 remove_free_nid(NM_I(sbi), ino);
2019
2020 if (!PageUptodate(ipage))
2021 SetPageUptodate(ipage);
2022 fill_node_footer(ipage, ino, ino, 0, true);
2023
2024 src = F2FS_INODE(page);
2025 dst = F2FS_INODE(ipage);
2026
2027 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2028 dst->i_size = 0;
2029 dst->i_blocks = cpu_to_le64(1);
2030 dst->i_links = cpu_to_le32(1);
2031 dst->i_xattr_nid = 0;
2032 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2033
2034 new_ni = old_ni;
2035 new_ni.ino = ino;
2036
2037 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2038 WARN_ON(1);
2039 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2040 inc_valid_inode_count(sbi);
2041 set_page_dirty(ipage);
2042 f2fs_put_page(ipage, 1);
2043 return 0;
2044 }
2045
2046 int restore_node_summary(struct f2fs_sb_info *sbi,
2047 unsigned int segno, struct f2fs_summary_block *sum)
2048 {
2049 struct f2fs_node *rn;
2050 struct f2fs_summary *sum_entry;
2051 block_t addr;
2052 int bio_blocks = MAX_BIO_BLOCKS(sbi);
2053 int i, idx, last_offset, nrpages;
2054
2055 /* scan the node segment */
2056 last_offset = sbi->blocks_per_seg;
2057 addr = START_BLOCK(sbi, segno);
2058 sum_entry = &sum->entries[0];
2059
2060 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2061 nrpages = min(last_offset - i, bio_blocks);
2062
2063 /* readahead node pages */
2064 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2065
2066 for (idx = addr; idx < addr + nrpages; idx++) {
2067 struct page *page = get_tmp_page(sbi, idx);
2068
2069 rn = F2FS_NODE(page);
2070 sum_entry->nid = rn->footer.nid;
2071 sum_entry->version = 0;
2072 sum_entry->ofs_in_node = 0;
2073 sum_entry++;
2074 f2fs_put_page(page, 1);
2075 }
2076
2077 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2078 addr + nrpages);
2079 }
2080 return 0;
2081 }
2082
2083 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2084 {
2085 struct f2fs_nm_info *nm_i = NM_I(sbi);
2086 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2087 struct f2fs_journal *journal = curseg->journal;
2088 int i;
2089
2090 down_write(&curseg->journal_rwsem);
2091 for (i = 0; i < nats_in_cursum(journal); i++) {
2092 struct nat_entry *ne;
2093 struct f2fs_nat_entry raw_ne;
2094 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2095
2096 raw_ne = nat_in_journal(journal, i);
2097
2098 ne = __lookup_nat_cache(nm_i, nid);
2099 if (!ne) {
2100 ne = grab_nat_entry(nm_i, nid);
2101 node_info_from_raw_nat(&ne->ni, &raw_ne);
2102 }
2103 __set_nat_cache_dirty(nm_i, ne);
2104 }
2105 update_nats_in_cursum(journal, -i);
2106 up_write(&curseg->journal_rwsem);
2107 }
2108
2109 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2110 struct list_head *head, int max)
2111 {
2112 struct nat_entry_set *cur;
2113
2114 if (nes->entry_cnt >= max)
2115 goto add_out;
2116
2117 list_for_each_entry(cur, head, set_list) {
2118 if (cur->entry_cnt >= nes->entry_cnt) {
2119 list_add(&nes->set_list, cur->set_list.prev);
2120 return;
2121 }
2122 }
2123 add_out:
2124 list_add_tail(&nes->set_list, head);
2125 }
2126
2127 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2128 struct nat_entry_set *set)
2129 {
2130 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2131 struct f2fs_journal *journal = curseg->journal;
2132 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2133 bool to_journal = true;
2134 struct f2fs_nat_block *nat_blk;
2135 struct nat_entry *ne, *cur;
2136 struct page *page = NULL;
2137
2138 /*
2139 * there are two steps to flush nat entries:
2140 * #1, flush nat entries to journal in current hot data summary block.
2141 * #2, flush nat entries to nat page.
2142 */
2143 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2144 to_journal = false;
2145
2146 if (to_journal) {
2147 down_write(&curseg->journal_rwsem);
2148 } else {
2149 page = get_next_nat_page(sbi, start_nid);
2150 nat_blk = page_address(page);
2151 f2fs_bug_on(sbi, !nat_blk);
2152 }
2153
2154 /* flush dirty nats in nat entry set */
2155 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2156 struct f2fs_nat_entry *raw_ne;
2157 nid_t nid = nat_get_nid(ne);
2158 int offset;
2159
2160 if (nat_get_blkaddr(ne) == NEW_ADDR)
2161 continue;
2162
2163 if (to_journal) {
2164 offset = lookup_journal_in_cursum(journal,
2165 NAT_JOURNAL, nid, 1);
2166 f2fs_bug_on(sbi, offset < 0);
2167 raw_ne = &nat_in_journal(journal, offset);
2168 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2169 } else {
2170 raw_ne = &nat_blk->entries[nid - start_nid];
2171 }
2172 raw_nat_from_node_info(raw_ne, &ne->ni);
2173 nat_reset_flag(ne);
2174 __clear_nat_cache_dirty(NM_I(sbi), ne);
2175 if (nat_get_blkaddr(ne) == NULL_ADDR)
2176 add_free_nid(sbi, nid, false);
2177 }
2178
2179 if (to_journal)
2180 up_write(&curseg->journal_rwsem);
2181 else
2182 f2fs_put_page(page, 1);
2183
2184 f2fs_bug_on(sbi, set->entry_cnt);
2185
2186 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2187 kmem_cache_free(nat_entry_set_slab, set);
2188 }
2189
2190 /*
2191 * This function is called during the checkpointing process.
2192 */
2193 void flush_nat_entries(struct f2fs_sb_info *sbi)
2194 {
2195 struct f2fs_nm_info *nm_i = NM_I(sbi);
2196 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2197 struct f2fs_journal *journal = curseg->journal;
2198 struct nat_entry_set *setvec[SETVEC_SIZE];
2199 struct nat_entry_set *set, *tmp;
2200 unsigned int found;
2201 nid_t set_idx = 0;
2202 LIST_HEAD(sets);
2203
2204 if (!nm_i->dirty_nat_cnt)
2205 return;
2206
2207 percpu_down_write(&nm_i->nat_tree_lock);
2208
2209 /*
2210 * if there are no enough space in journal to store dirty nat
2211 * entries, remove all entries from journal and merge them
2212 * into nat entry set.
2213 */
2214 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2215 remove_nats_in_journal(sbi);
2216
2217 while ((found = __gang_lookup_nat_set(nm_i,
2218 set_idx, SETVEC_SIZE, setvec))) {
2219 unsigned idx;
2220 set_idx = setvec[found - 1]->set + 1;
2221 for (idx = 0; idx < found; idx++)
2222 __adjust_nat_entry_set(setvec[idx], &sets,
2223 MAX_NAT_JENTRIES(journal));
2224 }
2225
2226 /* flush dirty nats in nat entry set */
2227 list_for_each_entry_safe(set, tmp, &sets, set_list)
2228 __flush_nat_entry_set(sbi, set);
2229
2230 percpu_up_write(&nm_i->nat_tree_lock);
2231
2232 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2233 }
2234
2235 static int init_node_manager(struct f2fs_sb_info *sbi)
2236 {
2237 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2238 struct f2fs_nm_info *nm_i = NM_I(sbi);
2239 unsigned char *version_bitmap;
2240 unsigned int nat_segs, nat_blocks;
2241
2242 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2243
2244 /* segment_count_nat includes pair segment so divide to 2. */
2245 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2246 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2247
2248 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2249
2250 /* not used nids: 0, node, meta, (and root counted as valid node) */
2251 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2252 nm_i->fcnt = 0;
2253 nm_i->nat_cnt = 0;
2254 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2255 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2256 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2257
2258 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2259 INIT_LIST_HEAD(&nm_i->free_nid_list);
2260 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2261 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2262 INIT_LIST_HEAD(&nm_i->nat_entries);
2263
2264 mutex_init(&nm_i->build_lock);
2265 spin_lock_init(&nm_i->free_nid_list_lock);
2266 if (percpu_init_rwsem(&nm_i->nat_tree_lock))
2267 return -ENOMEM;
2268
2269 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2270 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2271 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2272 if (!version_bitmap)
2273 return -EFAULT;
2274
2275 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2276 GFP_KERNEL);
2277 if (!nm_i->nat_bitmap)
2278 return -ENOMEM;
2279 return 0;
2280 }
2281
2282 int build_node_manager(struct f2fs_sb_info *sbi)
2283 {
2284 int err;
2285
2286 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2287 if (!sbi->nm_info)
2288 return -ENOMEM;
2289
2290 err = init_node_manager(sbi);
2291 if (err)
2292 return err;
2293
2294 build_free_nids(sbi);
2295 return 0;
2296 }
2297
2298 void destroy_node_manager(struct f2fs_sb_info *sbi)
2299 {
2300 struct f2fs_nm_info *nm_i = NM_I(sbi);
2301 struct free_nid *i, *next_i;
2302 struct nat_entry *natvec[NATVEC_SIZE];
2303 struct nat_entry_set *setvec[SETVEC_SIZE];
2304 nid_t nid = 0;
2305 unsigned int found;
2306
2307 if (!nm_i)
2308 return;
2309
2310 /* destroy free nid list */
2311 spin_lock(&nm_i->free_nid_list_lock);
2312 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2313 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2314 __del_from_free_nid_list(nm_i, i);
2315 nm_i->fcnt--;
2316 spin_unlock(&nm_i->free_nid_list_lock);
2317 kmem_cache_free(free_nid_slab, i);
2318 spin_lock(&nm_i->free_nid_list_lock);
2319 }
2320 f2fs_bug_on(sbi, nm_i->fcnt);
2321 spin_unlock(&nm_i->free_nid_list_lock);
2322
2323 /* destroy nat cache */
2324 percpu_down_write(&nm_i->nat_tree_lock);
2325 while ((found = __gang_lookup_nat_cache(nm_i,
2326 nid, NATVEC_SIZE, natvec))) {
2327 unsigned idx;
2328
2329 nid = nat_get_nid(natvec[found - 1]) + 1;
2330 for (idx = 0; idx < found; idx++)
2331 __del_from_nat_cache(nm_i, natvec[idx]);
2332 }
2333 f2fs_bug_on(sbi, nm_i->nat_cnt);
2334
2335 /* destroy nat set cache */
2336 nid = 0;
2337 while ((found = __gang_lookup_nat_set(nm_i,
2338 nid, SETVEC_SIZE, setvec))) {
2339 unsigned idx;
2340
2341 nid = setvec[found - 1]->set + 1;
2342 for (idx = 0; idx < found; idx++) {
2343 /* entry_cnt is not zero, when cp_error was occurred */
2344 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2345 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2346 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2347 }
2348 }
2349 percpu_up_write(&nm_i->nat_tree_lock);
2350
2351 percpu_free_rwsem(&nm_i->nat_tree_lock);
2352 kfree(nm_i->nat_bitmap);
2353 sbi->nm_info = NULL;
2354 kfree(nm_i);
2355 }
2356
2357 int __init create_node_manager_caches(void)
2358 {
2359 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2360 sizeof(struct nat_entry));
2361 if (!nat_entry_slab)
2362 goto fail;
2363
2364 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2365 sizeof(struct free_nid));
2366 if (!free_nid_slab)
2367 goto destroy_nat_entry;
2368
2369 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2370 sizeof(struct nat_entry_set));
2371 if (!nat_entry_set_slab)
2372 goto destroy_free_nid;
2373 return 0;
2374
2375 destroy_free_nid:
2376 kmem_cache_destroy(free_nid_slab);
2377 destroy_nat_entry:
2378 kmem_cache_destroy(nat_entry_slab);
2379 fail:
2380 return -ENOMEM;
2381 }
2382
2383 void destroy_node_manager_caches(void)
2384 {
2385 kmem_cache_destroy(nat_entry_set_slab);
2386 kmem_cache_destroy(free_nid_slab);
2387 kmem_cache_destroy(nat_entry_slab);
2388 }
Linux kernel - Deliverables
This page took 0.081266 seconds and 6 git commands to generate.