projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: add proc entry to show valid block bitmap
[deliverable/linux.git] / fs / f2fs / node.c
1 /*
2 * fs/f2fs/node.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
18
19 #include "f2fs.h"
20 #include "node.h"
21 #include "segment.h"
22 #include "trace.h"
23 #include <trace/events/f2fs.h>
24
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
30
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32 {
33 struct f2fs_nm_info *nm_i = NM_I(sbi);
34 struct sysinfo val;
35 unsigned long avail_ram;
36 unsigned long mem_size = 0;
37 bool res = false;
38
39 si_meminfo(&val);
40
41 /* only uses low memory */
42 avail_ram = val.totalram - val.totalhigh;
43
44 /*
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
46 */
47 if (type == FREE_NIDS) {
48 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
49 PAGE_SHIFT;
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 PAGE_SHIFT;
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 } else if (type == DIRTY_DENTS) {
56 if (sbi->sb->s_bdi->wb.dirty_exceeded)
57 return false;
58 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
59 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
60 } else if (type == INO_ENTRIES) {
61 int i;
62
63 for (i = 0; i <= UPDATE_INO; i++)
64 mem_size += (sbi->im[i].ino_num *
65 sizeof(struct ino_entry)) >> PAGE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
67 } else if (type == EXTENT_CACHE) {
68 mem_size = (atomic_read(&sbi->total_ext_tree) *
69 sizeof(struct extent_tree) +
70 atomic_read(&sbi->total_ext_node) *
71 sizeof(struct extent_node)) >> PAGE_SHIFT;
72 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
73 } else {
74 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
75 return true;
76 }
77 return res;
78 }
79
80 static void clear_node_page_dirty(struct page *page)
81 {
82 struct address_space *mapping = page->mapping;
83 unsigned int long flags;
84
85 if (PageDirty(page)) {
86 spin_lock_irqsave(&mapping->tree_lock, flags);
87 radix_tree_tag_clear(&mapping->page_tree,
88 page_index(page),
89 PAGECACHE_TAG_DIRTY);
90 spin_unlock_irqrestore(&mapping->tree_lock, flags);
91
92 clear_page_dirty_for_io(page);
93 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
94 }
95 ClearPageUptodate(page);
96 }
97
98 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
99 {
100 pgoff_t index = current_nat_addr(sbi, nid);
101 return get_meta_page(sbi, index);
102 }
103
104 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
105 {
106 struct page *src_page;
107 struct page *dst_page;
108 pgoff_t src_off;
109 pgoff_t dst_off;
110 void *src_addr;
111 void *dst_addr;
112 struct f2fs_nm_info *nm_i = NM_I(sbi);
113
114 src_off = current_nat_addr(sbi, nid);
115 dst_off = next_nat_addr(sbi, src_off);
116
117 /* get current nat block page with lock */
118 src_page = get_meta_page(sbi, src_off);
119 dst_page = grab_meta_page(sbi, dst_off);
120 f2fs_bug_on(sbi, PageDirty(src_page));
121
122 src_addr = page_address(src_page);
123 dst_addr = page_address(dst_page);
124 memcpy(dst_addr, src_addr, PAGE_SIZE);
125 set_page_dirty(dst_page);
126 f2fs_put_page(src_page, 1);
127
128 set_to_next_nat(nm_i, nid);
129
130 return dst_page;
131 }
132
133 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
134 {
135 return radix_tree_lookup(&nm_i->nat_root, n);
136 }
137
138 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
139 nid_t start, unsigned int nr, struct nat_entry **ep)
140 {
141 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
142 }
143
144 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
145 {
146 list_del(&e->list);
147 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
148 nm_i->nat_cnt--;
149 kmem_cache_free(nat_entry_slab, e);
150 }
151
152 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
153 struct nat_entry *ne)
154 {
155 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
156 struct nat_entry_set *head;
157
158 if (get_nat_flag(ne, IS_DIRTY))
159 return;
160
161 head = radix_tree_lookup(&nm_i->nat_set_root, set);
162 if (!head) {
163 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
164
165 INIT_LIST_HEAD(&head->entry_list);
166 INIT_LIST_HEAD(&head->set_list);
167 head->set = set;
168 head->entry_cnt = 0;
169 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
170 }
171 list_move_tail(&ne->list, &head->entry_list);
172 nm_i->dirty_nat_cnt++;
173 head->entry_cnt++;
174 set_nat_flag(ne, IS_DIRTY, true);
175 }
176
177 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
178 struct nat_entry *ne)
179 {
180 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
181 struct nat_entry_set *head;
182
183 head = radix_tree_lookup(&nm_i->nat_set_root, set);
184 if (head) {
185 list_move_tail(&ne->list, &nm_i->nat_entries);
186 set_nat_flag(ne, IS_DIRTY, false);
187 head->entry_cnt--;
188 nm_i->dirty_nat_cnt--;
189 }
190 }
191
192 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
193 nid_t start, unsigned int nr, struct nat_entry_set **ep)
194 {
195 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
196 start, nr);
197 }
198
199 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
200 {
201 struct f2fs_nm_info *nm_i = NM_I(sbi);
202 struct nat_entry *e;
203 bool need = false;
204
205 down_read(&nm_i->nat_tree_lock);
206 e = __lookup_nat_cache(nm_i, nid);
207 if (e) {
208 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
209 !get_nat_flag(e, HAS_FSYNCED_INODE))
210 need = true;
211 }
212 up_read(&nm_i->nat_tree_lock);
213 return need;
214 }
215
216 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
217 {
218 struct f2fs_nm_info *nm_i = NM_I(sbi);
219 struct nat_entry *e;
220 bool is_cp = true;
221
222 down_read(&nm_i->nat_tree_lock);
223 e = __lookup_nat_cache(nm_i, nid);
224 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
225 is_cp = false;
226 up_read(&nm_i->nat_tree_lock);
227 return is_cp;
228 }
229
230 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
231 {
232 struct f2fs_nm_info *nm_i = NM_I(sbi);
233 struct nat_entry *e;
234 bool need_update = true;
235
236 down_read(&nm_i->nat_tree_lock);
237 e = __lookup_nat_cache(nm_i, ino);
238 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
239 (get_nat_flag(e, IS_CHECKPOINTED) ||
240 get_nat_flag(e, HAS_FSYNCED_INODE)))
241 need_update = false;
242 up_read(&nm_i->nat_tree_lock);
243 return need_update;
244 }
245
246 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
247 {
248 struct nat_entry *new;
249
250 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
251 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
252 memset(new, 0, sizeof(struct nat_entry));
253 nat_set_nid(new, nid);
254 nat_reset_flag(new);
255 list_add_tail(&new->list, &nm_i->nat_entries);
256 nm_i->nat_cnt++;
257 return new;
258 }
259
260 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
261 struct f2fs_nat_entry *ne)
262 {
263 struct f2fs_nm_info *nm_i = NM_I(sbi);
264 struct nat_entry *e;
265
266 e = __lookup_nat_cache(nm_i, nid);
267 if (!e) {
268 e = grab_nat_entry(nm_i, nid);
269 node_info_from_raw_nat(&e->ni, ne);
270 } else {
271 f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
272 nat_get_blkaddr(e) != ne->block_addr ||
273 nat_get_version(e) != ne->version);
274 }
275 }
276
277 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
278 block_t new_blkaddr, bool fsync_done)
279 {
280 struct f2fs_nm_info *nm_i = NM_I(sbi);
281 struct nat_entry *e;
282
283 down_write(&nm_i->nat_tree_lock);
284 e = __lookup_nat_cache(nm_i, ni->nid);
285 if (!e) {
286 e = grab_nat_entry(nm_i, ni->nid);
287 copy_node_info(&e->ni, ni);
288 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
289 } else if (new_blkaddr == NEW_ADDR) {
290 /*
291 * when nid is reallocated,
292 * previous nat entry can be remained in nat cache.
293 * So, reinitialize it with new information.
294 */
295 copy_node_info(&e->ni, ni);
296 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
297 }
298
299 /* sanity check */
300 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
301 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
302 new_blkaddr == NULL_ADDR);
303 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
304 new_blkaddr == NEW_ADDR);
305 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
306 nat_get_blkaddr(e) != NULL_ADDR &&
307 new_blkaddr == NEW_ADDR);
308
309 /* increment version no as node is removed */
310 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
311 unsigned char version = nat_get_version(e);
312 nat_set_version(e, inc_node_version(version));
313
314 /* in order to reuse the nid */
315 if (nm_i->next_scan_nid > ni->nid)
316 nm_i->next_scan_nid = ni->nid;
317 }
318
319 /* change address */
320 nat_set_blkaddr(e, new_blkaddr);
321 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
322 set_nat_flag(e, IS_CHECKPOINTED, false);
323 __set_nat_cache_dirty(nm_i, e);
324
325 /* update fsync_mark if its inode nat entry is still alive */
326 if (ni->nid != ni->ino)
327 e = __lookup_nat_cache(nm_i, ni->ino);
328 if (e) {
329 if (fsync_done && ni->nid == ni->ino)
330 set_nat_flag(e, HAS_FSYNCED_INODE, true);
331 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
332 }
333 up_write(&nm_i->nat_tree_lock);
334 }
335
336 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
337 {
338 struct f2fs_nm_info *nm_i = NM_I(sbi);
339 int nr = nr_shrink;
340
341 if (!down_write_trylock(&nm_i->nat_tree_lock))
342 return 0;
343
344 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
345 struct nat_entry *ne;
346 ne = list_first_entry(&nm_i->nat_entries,
347 struct nat_entry, list);
348 __del_from_nat_cache(nm_i, ne);
349 nr_shrink--;
350 }
351 up_write(&nm_i->nat_tree_lock);
352 return nr - nr_shrink;
353 }
354
355 /*
356 * This function always returns success
357 */
358 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
359 {
360 struct f2fs_nm_info *nm_i = NM_I(sbi);
361 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
362 struct f2fs_journal *journal = curseg->journal;
363 nid_t start_nid = START_NID(nid);
364 struct f2fs_nat_block *nat_blk;
365 struct page *page = NULL;
366 struct f2fs_nat_entry ne;
367 struct nat_entry *e;
368 int i;
369
370 ni->nid = nid;
371
372 /* Check nat cache */
373 down_read(&nm_i->nat_tree_lock);
374 e = __lookup_nat_cache(nm_i, nid);
375 if (e) {
376 ni->ino = nat_get_ino(e);
377 ni->blk_addr = nat_get_blkaddr(e);
378 ni->version = nat_get_version(e);
379 up_read(&nm_i->nat_tree_lock);
380 return;
381 }
382
383 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
384
385 /* Check current segment summary */
386 down_read(&curseg->journal_rwsem);
387 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
388 if (i >= 0) {
389 ne = nat_in_journal(journal, i);
390 node_info_from_raw_nat(ni, &ne);
391 }
392 up_read(&curseg->journal_rwsem);
393 if (i >= 0)
394 goto cache;
395
396 /* Fill node_info from nat page */
397 page = get_current_nat_page(sbi, start_nid);
398 nat_blk = (struct f2fs_nat_block *)page_address(page);
399 ne = nat_blk->entries[nid - start_nid];
400 node_info_from_raw_nat(ni, &ne);
401 f2fs_put_page(page, 1);
402 cache:
403 up_read(&nm_i->nat_tree_lock);
404 /* cache nat entry */
405 down_write(&nm_i->nat_tree_lock);
406 cache_nat_entry(sbi, nid, &ne);
407 up_write(&nm_i->nat_tree_lock);
408 }
409
410 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
411 {
412 const long direct_index = ADDRS_PER_INODE(dn->inode);
413 const long direct_blks = ADDRS_PER_BLOCK;
414 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
415 unsigned int skipped_unit = ADDRS_PER_BLOCK;
416 int cur_level = dn->cur_level;
417 int max_level = dn->max_level;
418 pgoff_t base = 0;
419
420 if (!dn->max_level)
421 return pgofs + 1;
422
423 while (max_level-- > cur_level)
424 skipped_unit *= NIDS_PER_BLOCK;
425
426 switch (dn->max_level) {
427 case 3:
428 base += 2 * indirect_blks;
429 case 2:
430 base += 2 * direct_blks;
431 case 1:
432 base += direct_index;
433 break;
434 default:
435 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
436 }
437
438 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
439 }
440
441 /*
442 * The maximum depth is four.
443 * Offset[0] will have raw inode offset.
444 */
445 static int get_node_path(struct inode *inode, long block,
446 int offset[4], unsigned int noffset[4])
447 {
448 const long direct_index = ADDRS_PER_INODE(inode);
449 const long direct_blks = ADDRS_PER_BLOCK;
450 const long dptrs_per_blk = NIDS_PER_BLOCK;
451 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
452 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
453 int n = 0;
454 int level = 0;
455
456 noffset[0] = 0;
457
458 if (block < direct_index) {
459 offset[n] = block;
460 goto got;
461 }
462 block -= direct_index;
463 if (block < direct_blks) {
464 offset[n++] = NODE_DIR1_BLOCK;
465 noffset[n] = 1;
466 offset[n] = block;
467 level = 1;
468 goto got;
469 }
470 block -= direct_blks;
471 if (block < direct_blks) {
472 offset[n++] = NODE_DIR2_BLOCK;
473 noffset[n] = 2;
474 offset[n] = block;
475 level = 1;
476 goto got;
477 }
478 block -= direct_blks;
479 if (block < indirect_blks) {
480 offset[n++] = NODE_IND1_BLOCK;
481 noffset[n] = 3;
482 offset[n++] = block / direct_blks;
483 noffset[n] = 4 + offset[n - 1];
484 offset[n] = block % direct_blks;
485 level = 2;
486 goto got;
487 }
488 block -= indirect_blks;
489 if (block < indirect_blks) {
490 offset[n++] = NODE_IND2_BLOCK;
491 noffset[n] = 4 + dptrs_per_blk;
492 offset[n++] = block / direct_blks;
493 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
494 offset[n] = block % direct_blks;
495 level = 2;
496 goto got;
497 }
498 block -= indirect_blks;
499 if (block < dindirect_blks) {
500 offset[n++] = NODE_DIND_BLOCK;
501 noffset[n] = 5 + (dptrs_per_blk * 2);
502 offset[n++] = block / indirect_blks;
503 noffset[n] = 6 + (dptrs_per_blk * 2) +
504 offset[n - 1] * (dptrs_per_blk + 1);
505 offset[n++] = (block / direct_blks) % dptrs_per_blk;
506 noffset[n] = 7 + (dptrs_per_blk * 2) +
507 offset[n - 2] * (dptrs_per_blk + 1) +
508 offset[n - 1];
509 offset[n] = block % direct_blks;
510 level = 3;
511 goto got;
512 } else {
513 BUG();
514 }
515 got:
516 return level;
517 }
518
519 /*
520 * Caller should call f2fs_put_dnode(dn).
521 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
522 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
523 * In the case of RDONLY_NODE, we don't need to care about mutex.
524 */
525 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
526 {
527 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
528 struct page *npage[4];
529 struct page *parent = NULL;
530 int offset[4];
531 unsigned int noffset[4];
532 nid_t nids[4];
533 int level, i = 0;
534 int err = 0;
535
536 level = get_node_path(dn->inode, index, offset, noffset);
537
538 nids[0] = dn->inode->i_ino;
539 npage[0] = dn->inode_page;
540
541 if (!npage[0]) {
542 npage[0] = get_node_page(sbi, nids[0]);
543 if (IS_ERR(npage[0]))
544 return PTR_ERR(npage[0]);
545 }
546
547 /* if inline_data is set, should not report any block indices */
548 if (f2fs_has_inline_data(dn->inode) && index) {
549 err = -ENOENT;
550 f2fs_put_page(npage[0], 1);
551 goto release_out;
552 }
553
554 parent = npage[0];
555 if (level != 0)
556 nids[1] = get_nid(parent, offset[0], true);
557 dn->inode_page = npage[0];
558 dn->inode_page_locked = true;
559
560 /* get indirect or direct nodes */
561 for (i = 1; i <= level; i++) {
562 bool done = false;
563
564 if (!nids[i] && mode == ALLOC_NODE) {
565 /* alloc new node */
566 if (!alloc_nid(sbi, &(nids[i]))) {
567 err = -ENOSPC;
568 goto release_pages;
569 }
570
571 dn->nid = nids[i];
572 npage[i] = new_node_page(dn, noffset[i], NULL);
573 if (IS_ERR(npage[i])) {
574 alloc_nid_failed(sbi, nids[i]);
575 err = PTR_ERR(npage[i]);
576 goto release_pages;
577 }
578
579 set_nid(parent, offset[i - 1], nids[i], i == 1);
580 alloc_nid_done(sbi, nids[i]);
581 done = true;
582 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
583 npage[i] = get_node_page_ra(parent, offset[i - 1]);
584 if (IS_ERR(npage[i])) {
585 err = PTR_ERR(npage[i]);
586 goto release_pages;
587 }
588 done = true;
589 }
590 if (i == 1) {
591 dn->inode_page_locked = false;
592 unlock_page(parent);
593 } else {
594 f2fs_put_page(parent, 1);
595 }
596
597 if (!done) {
598 npage[i] = get_node_page(sbi, nids[i]);
599 if (IS_ERR(npage[i])) {
600 err = PTR_ERR(npage[i]);
601 f2fs_put_page(npage[0], 0);
602 goto release_out;
603 }
604 }
605 if (i < level) {
606 parent = npage[i];
607 nids[i + 1] = get_nid(parent, offset[i], false);
608 }
609 }
610 dn->nid = nids[level];
611 dn->ofs_in_node = offset[level];
612 dn->node_page = npage[level];
613 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
614 return 0;
615
616 release_pages:
617 f2fs_put_page(parent, 1);
618 if (i > 1)
619 f2fs_put_page(npage[0], 0);
620 release_out:
621 dn->inode_page = NULL;
622 dn->node_page = NULL;
623 if (err == -ENOENT) {
624 dn->cur_level = i;
625 dn->max_level = level;
626 }
627 return err;
628 }
629
630 static void truncate_node(struct dnode_of_data *dn)
631 {
632 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
633 struct node_info ni;
634
635 get_node_info(sbi, dn->nid, &ni);
636 if (dn->inode->i_blocks == 0) {
637 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
638 goto invalidate;
639 }
640 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
641
642 /* Deallocate node address */
643 invalidate_blocks(sbi, ni.blk_addr);
644 dec_valid_node_count(sbi, dn->inode);
645 set_node_addr(sbi, &ni, NULL_ADDR, false);
646
647 if (dn->nid == dn->inode->i_ino) {
648 remove_orphan_inode(sbi, dn->nid);
649 dec_valid_inode_count(sbi);
650 } else {
651 sync_inode_page(dn);
652 }
653 invalidate:
654 clear_node_page_dirty(dn->node_page);
655 set_sbi_flag(sbi, SBI_IS_DIRTY);
656
657 f2fs_put_page(dn->node_page, 1);
658
659 invalidate_mapping_pages(NODE_MAPPING(sbi),
660 dn->node_page->index, dn->node_page->index);
661
662 dn->node_page = NULL;
663 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
664 }
665
666 static int truncate_dnode(struct dnode_of_data *dn)
667 {
668 struct page *page;
669
670 if (dn->nid == 0)
671 return 1;
672
673 /* get direct node */
674 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
675 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
676 return 1;
677 else if (IS_ERR(page))
678 return PTR_ERR(page);
679
680 /* Make dnode_of_data for parameter */
681 dn->node_page = page;
682 dn->ofs_in_node = 0;
683 truncate_data_blocks(dn);
684 truncate_node(dn);
685 return 1;
686 }
687
688 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
689 int ofs, int depth)
690 {
691 struct dnode_of_data rdn = *dn;
692 struct page *page;
693 struct f2fs_node *rn;
694 nid_t child_nid;
695 unsigned int child_nofs;
696 int freed = 0;
697 int i, ret;
698
699 if (dn->nid == 0)
700 return NIDS_PER_BLOCK + 1;
701
702 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
703
704 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
705 if (IS_ERR(page)) {
706 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
707 return PTR_ERR(page);
708 }
709
710 rn = F2FS_NODE(page);
711 if (depth < 3) {
712 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
713 child_nid = le32_to_cpu(rn->in.nid[i]);
714 if (child_nid == 0)
715 continue;
716 rdn.nid = child_nid;
717 ret = truncate_dnode(&rdn);
718 if (ret < 0)
719 goto out_err;
720 if (set_nid(page, i, 0, false))
721 dn->node_changed = true;
722 }
723 } else {
724 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
725 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
726 child_nid = le32_to_cpu(rn->in.nid[i]);
727 if (child_nid == 0) {
728 child_nofs += NIDS_PER_BLOCK + 1;
729 continue;
730 }
731 rdn.nid = child_nid;
732 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
733 if (ret == (NIDS_PER_BLOCK + 1)) {
734 if (set_nid(page, i, 0, false))
735 dn->node_changed = true;
736 child_nofs += ret;
737 } else if (ret < 0 && ret != -ENOENT) {
738 goto out_err;
739 }
740 }
741 freed = child_nofs;
742 }
743
744 if (!ofs) {
745 /* remove current indirect node */
746 dn->node_page = page;
747 truncate_node(dn);
748 freed++;
749 } else {
750 f2fs_put_page(page, 1);
751 }
752 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
753 return freed;
754
755 out_err:
756 f2fs_put_page(page, 1);
757 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
758 return ret;
759 }
760
761 static int truncate_partial_nodes(struct dnode_of_data *dn,
762 struct f2fs_inode *ri, int *offset, int depth)
763 {
764 struct page *pages[2];
765 nid_t nid[3];
766 nid_t child_nid;
767 int err = 0;
768 int i;
769 int idx = depth - 2;
770
771 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
772 if (!nid[0])
773 return 0;
774
775 /* get indirect nodes in the path */
776 for (i = 0; i < idx + 1; i++) {
777 /* reference count'll be increased */
778 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
779 if (IS_ERR(pages[i])) {
780 err = PTR_ERR(pages[i]);
781 idx = i - 1;
782 goto fail;
783 }
784 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
785 }
786
787 /* free direct nodes linked to a partial indirect node */
788 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
789 child_nid = get_nid(pages[idx], i, false);
790 if (!child_nid)
791 continue;
792 dn->nid = child_nid;
793 err = truncate_dnode(dn);
794 if (err < 0)
795 goto fail;
796 if (set_nid(pages[idx], i, 0, false))
797 dn->node_changed = true;
798 }
799
800 if (offset[idx + 1] == 0) {
801 dn->node_page = pages[idx];
802 dn->nid = nid[idx];
803 truncate_node(dn);
804 } else {
805 f2fs_put_page(pages[idx], 1);
806 }
807 offset[idx]++;
808 offset[idx + 1] = 0;
809 idx--;
810 fail:
811 for (i = idx; i >= 0; i--)
812 f2fs_put_page(pages[i], 1);
813
814 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
815
816 return err;
817 }
818
819 /*
820 * All the block addresses of data and nodes should be nullified.
821 */
822 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
823 {
824 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
825 int err = 0, cont = 1;
826 int level, offset[4], noffset[4];
827 unsigned int nofs = 0;
828 struct f2fs_inode *ri;
829 struct dnode_of_data dn;
830 struct page *page;
831
832 trace_f2fs_truncate_inode_blocks_enter(inode, from);
833
834 level = get_node_path(inode, from, offset, noffset);
835
836 page = get_node_page(sbi, inode->i_ino);
837 if (IS_ERR(page)) {
838 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
839 return PTR_ERR(page);
840 }
841
842 set_new_dnode(&dn, inode, page, NULL, 0);
843 unlock_page(page);
844
845 ri = F2FS_INODE(page);
846 switch (level) {
847 case 0:
848 case 1:
849 nofs = noffset[1];
850 break;
851 case 2:
852 nofs = noffset[1];
853 if (!offset[level - 1])
854 goto skip_partial;
855 err = truncate_partial_nodes(&dn, ri, offset, level);
856 if (err < 0 && err != -ENOENT)
857 goto fail;
858 nofs += 1 + NIDS_PER_BLOCK;
859 break;
860 case 3:
861 nofs = 5 + 2 * NIDS_PER_BLOCK;
862 if (!offset[level - 1])
863 goto skip_partial;
864 err = truncate_partial_nodes(&dn, ri, offset, level);
865 if (err < 0 && err != -ENOENT)
866 goto fail;
867 break;
868 default:
869 BUG();
870 }
871
872 skip_partial:
873 while (cont) {
874 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
875 switch (offset[0]) {
876 case NODE_DIR1_BLOCK:
877 case NODE_DIR2_BLOCK:
878 err = truncate_dnode(&dn);
879 break;
880
881 case NODE_IND1_BLOCK:
882 case NODE_IND2_BLOCK:
883 err = truncate_nodes(&dn, nofs, offset[1], 2);
884 break;
885
886 case NODE_DIND_BLOCK:
887 err = truncate_nodes(&dn, nofs, offset[1], 3);
888 cont = 0;
889 break;
890
891 default:
892 BUG();
893 }
894 if (err < 0 && err != -ENOENT)
895 goto fail;
896 if (offset[1] == 0 &&
897 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
898 lock_page(page);
899 BUG_ON(page->mapping != NODE_MAPPING(sbi));
900 f2fs_wait_on_page_writeback(page, NODE, true);
901 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
902 set_page_dirty(page);
903 unlock_page(page);
904 }
905 offset[1] = 0;
906 offset[0]++;
907 nofs += err;
908 }
909 fail:
910 f2fs_put_page(page, 0);
911 trace_f2fs_truncate_inode_blocks_exit(inode, err);
912 return err > 0 ? 0 : err;
913 }
914
915 int truncate_xattr_node(struct inode *inode, struct page *page)
916 {
917 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
918 nid_t nid = F2FS_I(inode)->i_xattr_nid;
919 struct dnode_of_data dn;
920 struct page *npage;
921
922 if (!nid)
923 return 0;
924
925 npage = get_node_page(sbi, nid);
926 if (IS_ERR(npage))
927 return PTR_ERR(npage);
928
929 F2FS_I(inode)->i_xattr_nid = 0;
930
931 /* need to do checkpoint during fsync */
932 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
933
934 set_new_dnode(&dn, inode, page, npage, nid);
935
936 if (page)
937 dn.inode_page_locked = true;
938 truncate_node(&dn);
939 return 0;
940 }
941
942 /*
943 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
944 * f2fs_unlock_op().
945 */
946 int remove_inode_page(struct inode *inode)
947 {
948 struct dnode_of_data dn;
949 int err;
950
951 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
952 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
953 if (err)
954 return err;
955
956 err = truncate_xattr_node(inode, dn.inode_page);
957 if (err) {
958 f2fs_put_dnode(&dn);
959 return err;
960 }
961
962 /* remove potential inline_data blocks */
963 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
964 S_ISLNK(inode->i_mode))
965 truncate_data_blocks_range(&dn, 1);
966
967 /* 0 is possible, after f2fs_new_inode() has failed */
968 f2fs_bug_on(F2FS_I_SB(inode),
969 inode->i_blocks != 0 && inode->i_blocks != 1);
970
971 /* will put inode & node pages */
972 truncate_node(&dn);
973 return 0;
974 }
975
976 struct page *new_inode_page(struct inode *inode)
977 {
978 struct dnode_of_data dn;
979
980 /* allocate inode page for new inode */
981 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
982
983 /* caller should f2fs_put_page(page, 1); */
984 return new_node_page(&dn, 0, NULL);
985 }
986
987 struct page *new_node_page(struct dnode_of_data *dn,
988 unsigned int ofs, struct page *ipage)
989 {
990 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
991 struct node_info old_ni, new_ni;
992 struct page *page;
993 int err;
994
995 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
996 return ERR_PTR(-EPERM);
997
998 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
999 if (!page)
1000 return ERR_PTR(-ENOMEM);
1001
1002 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1003 err = -ENOSPC;
1004 goto fail;
1005 }
1006
1007 get_node_info(sbi, dn->nid, &old_ni);
1008
1009 /* Reinitialize old_ni with new node page */
1010 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1011 new_ni = old_ni;
1012 new_ni.ino = dn->inode->i_ino;
1013 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1014
1015 f2fs_wait_on_page_writeback(page, NODE, true);
1016 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1017 set_cold_node(dn->inode, page);
1018 SetPageUptodate(page);
1019 if (set_page_dirty(page))
1020 dn->node_changed = true;
1021
1022 if (f2fs_has_xattr_block(ofs))
1023 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
1024
1025 dn->node_page = page;
1026 if (ipage)
1027 update_inode(dn->inode, ipage);
1028 else
1029 sync_inode_page(dn);
1030 if (ofs == 0)
1031 inc_valid_inode_count(sbi);
1032
1033 return page;
1034
1035 fail:
1036 clear_node_page_dirty(page);
1037 f2fs_put_page(page, 1);
1038 return ERR_PTR(err);
1039 }
1040
1041 /*
1042 * Caller should do after getting the following values.
1043 * 0: f2fs_put_page(page, 0)
1044 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1045 */
1046 static int read_node_page(struct page *page, int rw)
1047 {
1048 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1049 struct node_info ni;
1050 struct f2fs_io_info fio = {
1051 .sbi = sbi,
1052 .type = NODE,
1053 .rw = rw,
1054 .page = page,
1055 .encrypted_page = NULL,
1056 };
1057
1058 get_node_info(sbi, page->index, &ni);
1059
1060 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1061 ClearPageUptodate(page);
1062 return -ENOENT;
1063 }
1064
1065 if (PageUptodate(page))
1066 return LOCKED_PAGE;
1067
1068 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1069 return f2fs_submit_page_bio(&fio);
1070 }
1071
1072 /*
1073 * Readahead a node page
1074 */
1075 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1076 {
1077 struct page *apage;
1078 int err;
1079
1080 if (!nid)
1081 return;
1082 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1083
1084 rcu_read_lock();
1085 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1086 rcu_read_unlock();
1087 if (apage)
1088 return;
1089
1090 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1091 if (!apage)
1092 return;
1093
1094 err = read_node_page(apage, READA);
1095 f2fs_put_page(apage, err ? 1 : 0);
1096 }
1097
1098 /*
1099 * readahead MAX_RA_NODE number of node pages.
1100 */
1101 static void ra_node_pages(struct page *parent, int start)
1102 {
1103 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1104 struct blk_plug plug;
1105 int i, end;
1106 nid_t nid;
1107
1108 blk_start_plug(&plug);
1109
1110 /* Then, try readahead for siblings of the desired node */
1111 end = start + MAX_RA_NODE;
1112 end = min(end, NIDS_PER_BLOCK);
1113 for (i = start; i < end; i++) {
1114 nid = get_nid(parent, i, false);
1115 ra_node_page(sbi, nid);
1116 }
1117
1118 blk_finish_plug(&plug);
1119 }
1120
1121 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1122 struct page *parent, int start)
1123 {
1124 struct page *page;
1125 int err;
1126
1127 if (!nid)
1128 return ERR_PTR(-ENOENT);
1129 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1130 repeat:
1131 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1132 if (!page)
1133 return ERR_PTR(-ENOMEM);
1134
1135 err = read_node_page(page, READ_SYNC);
1136 if (err < 0) {
1137 f2fs_put_page(page, 1);
1138 return ERR_PTR(err);
1139 } else if (err == LOCKED_PAGE) {
1140 goto page_hit;
1141 }
1142
1143 if (parent)
1144 ra_node_pages(parent, start + 1);
1145
1146 lock_page(page);
1147
1148 if (unlikely(!PageUptodate(page))) {
1149 f2fs_put_page(page, 1);
1150 return ERR_PTR(-EIO);
1151 }
1152 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1153 f2fs_put_page(page, 1);
1154 goto repeat;
1155 }
1156 page_hit:
1157 f2fs_bug_on(sbi, nid != nid_of_node(page));
1158 return page;
1159 }
1160
1161 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1162 {
1163 return __get_node_page(sbi, nid, NULL, 0);
1164 }
1165
1166 struct page *get_node_page_ra(struct page *parent, int start)
1167 {
1168 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1169 nid_t nid = get_nid(parent, start, false);
1170
1171 return __get_node_page(sbi, nid, parent, start);
1172 }
1173
1174 void sync_inode_page(struct dnode_of_data *dn)
1175 {
1176 int ret = 0;
1177
1178 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1179 ret = update_inode(dn->inode, dn->node_page);
1180 } else if (dn->inode_page) {
1181 if (!dn->inode_page_locked)
1182 lock_page(dn->inode_page);
1183 ret = update_inode(dn->inode, dn->inode_page);
1184 if (!dn->inode_page_locked)
1185 unlock_page(dn->inode_page);
1186 } else {
1187 ret = update_inode_page(dn->inode);
1188 }
1189 dn->node_changed = ret ? true: false;
1190 }
1191
1192 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1193 {
1194 struct inode *inode;
1195 struct page *page;
1196
1197 /* should flush inline_data before evict_inode */
1198 inode = ilookup(sbi->sb, ino);
1199 if (!inode)
1200 return;
1201
1202 page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1203 if (!page)
1204 goto iput_out;
1205
1206 if (!PageUptodate(page))
1207 goto page_out;
1208
1209 if (!PageDirty(page))
1210 goto page_out;
1211
1212 if (!clear_page_dirty_for_io(page))
1213 goto page_out;
1214
1215 if (!f2fs_write_inline_data(inode, page))
1216 inode_dec_dirty_pages(inode);
1217 else
1218 set_page_dirty(page);
1219 page_out:
1220 f2fs_put_page(page, 1);
1221 iput_out:
1222 iput(inode);
1223 }
1224
1225 void move_node_page(struct page *node_page, int gc_type)
1226 {
1227 if (gc_type == FG_GC) {
1228 struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1229 struct writeback_control wbc = {
1230 .sync_mode = WB_SYNC_ALL,
1231 .nr_to_write = 1,
1232 .for_reclaim = 0,
1233 };
1234
1235 set_page_dirty(node_page);
1236 f2fs_wait_on_page_writeback(node_page, NODE, true);
1237
1238 f2fs_bug_on(sbi, PageWriteback(node_page));
1239 if (!clear_page_dirty_for_io(node_page))
1240 goto out_page;
1241
1242 if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1243 unlock_page(node_page);
1244 goto release_page;
1245 } else {
1246 /* set page dirty and write it */
1247 if (!PageWriteback(node_page))
1248 set_page_dirty(node_page);
1249 }
1250 out_page:
1251 unlock_page(node_page);
1252 release_page:
1253 f2fs_put_page(node_page, 0);
1254 }
1255
1256 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1257 {
1258 pgoff_t index, end;
1259 struct pagevec pvec;
1260 struct page *last_page = NULL;
1261
1262 pagevec_init(&pvec, 0);
1263 index = 0;
1264 end = ULONG_MAX;
1265
1266 while (index <= end) {
1267 int i, nr_pages;
1268 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1269 PAGECACHE_TAG_DIRTY,
1270 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1271 if (nr_pages == 0)
1272 break;
1273
1274 for (i = 0; i < nr_pages; i++) {
1275 struct page *page = pvec.pages[i];
1276
1277 if (unlikely(f2fs_cp_error(sbi))) {
1278 f2fs_put_page(last_page, 0);
1279 pagevec_release(&pvec);
1280 return ERR_PTR(-EIO);
1281 }
1282
1283 if (!IS_DNODE(page) || !is_cold_node(page))
1284 continue;
1285 if (ino_of_node(page) != ino)
1286 continue;
1287
1288 lock_page(page);
1289
1290 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1291 continue_unlock:
1292 unlock_page(page);
1293 continue;
1294 }
1295 if (ino_of_node(page) != ino)
1296 goto continue_unlock;
1297
1298 if (!PageDirty(page)) {
1299 /* someone wrote it for us */
1300 goto continue_unlock;
1301 }
1302
1303 if (last_page)
1304 f2fs_put_page(last_page, 0);
1305
1306 get_page(page);
1307 last_page = page;
1308 unlock_page(page);
1309 }
1310 pagevec_release(&pvec);
1311 cond_resched();
1312 }
1313 return last_page;
1314 }
1315
1316 int fsync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1317 struct writeback_control *wbc, bool atomic)
1318 {
1319 pgoff_t index, end;
1320 struct pagevec pvec;
1321 int ret = 0;
1322 struct page *last_page = NULL;
1323 bool marked = false;
1324
1325 if (atomic) {
1326 last_page = last_fsync_dnode(sbi, ino);
1327 if (IS_ERR_OR_NULL(last_page))
1328 return PTR_ERR_OR_ZERO(last_page);
1329 }
1330 retry:
1331 pagevec_init(&pvec, 0);
1332 index = 0;
1333 end = ULONG_MAX;
1334
1335 while (index <= end) {
1336 int i, nr_pages;
1337 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1338 PAGECACHE_TAG_DIRTY,
1339 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1340 if (nr_pages == 0)
1341 break;
1342
1343 for (i = 0; i < nr_pages; i++) {
1344 struct page *page = pvec.pages[i];
1345
1346 if (unlikely(f2fs_cp_error(sbi))) {
1347 f2fs_put_page(last_page, 0);
1348 pagevec_release(&pvec);
1349 return -EIO;
1350 }
1351
1352 if (!IS_DNODE(page) || !is_cold_node(page))
1353 continue;
1354 if (ino_of_node(page) != ino)
1355 continue;
1356
1357 lock_page(page);
1358
1359 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1360 continue_unlock:
1361 unlock_page(page);
1362 continue;
1363 }
1364 if (ino_of_node(page) != ino)
1365 goto continue_unlock;
1366
1367 if (!PageDirty(page) && page != last_page) {
1368 /* someone wrote it for us */
1369 goto continue_unlock;
1370 }
1371
1372 f2fs_wait_on_page_writeback(page, NODE, true);
1373 BUG_ON(PageWriteback(page));
1374
1375 if (!atomic || page == last_page) {
1376 set_fsync_mark(page, 1);
1377 if (IS_INODE(page))
1378 set_dentry_mark(page,
1379 need_dentry_mark(sbi, ino));
1380 /* may be written by other thread */
1381 if (!PageDirty(page))
1382 set_page_dirty(page);
1383 }
1384
1385 if (!clear_page_dirty_for_io(page))
1386 goto continue_unlock;
1387
1388 ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1389 if (ret) {
1390 unlock_page(page);
1391 f2fs_put_page(last_page, 0);
1392 break;
1393 }
1394 if (page == last_page) {
1395 f2fs_put_page(page, 0);
1396 marked = true;
1397 break;
1398 }
1399 }
1400 pagevec_release(&pvec);
1401 cond_resched();
1402
1403 if (ret || marked)
1404 break;
1405 }
1406 if (!ret && atomic && !marked) {
1407 f2fs_msg(sbi->sb, KERN_DEBUG,
1408 "Retry to write fsync mark: ino=%u, idx=%lx",
1409 ino, last_page->index);
1410 lock_page(last_page);
1411 set_page_dirty(last_page);
1412 unlock_page(last_page);
1413 goto retry;
1414 }
1415 return ret ? -EIO: 0;
1416 }
1417
1418 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1419 {
1420 pgoff_t index, end;
1421 struct pagevec pvec;
1422 int step = 0;
1423 int nwritten = 0;
1424
1425 pagevec_init(&pvec, 0);
1426
1427 next_step:
1428 index = 0;
1429 end = ULONG_MAX;
1430
1431 while (index <= end) {
1432 int i, nr_pages;
1433 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1434 PAGECACHE_TAG_DIRTY,
1435 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1436 if (nr_pages == 0)
1437 break;
1438
1439 for (i = 0; i < nr_pages; i++) {
1440 struct page *page = pvec.pages[i];
1441
1442 if (unlikely(f2fs_cp_error(sbi))) {
1443 pagevec_release(&pvec);
1444 return -EIO;
1445 }
1446
1447 /*
1448 * flushing sequence with step:
1449 * 0. indirect nodes
1450 * 1. dentry dnodes
1451 * 2. file dnodes
1452 */
1453 if (step == 0 && IS_DNODE(page))
1454 continue;
1455 if (step == 1 && (!IS_DNODE(page) ||
1456 is_cold_node(page)))
1457 continue;
1458 if (step == 2 && (!IS_DNODE(page) ||
1459 !is_cold_node(page)))
1460 continue;
1461 lock_node:
1462 if (!trylock_page(page))
1463 continue;
1464
1465 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1466 continue_unlock:
1467 unlock_page(page);
1468 continue;
1469 }
1470
1471 if (!PageDirty(page)) {
1472 /* someone wrote it for us */
1473 goto continue_unlock;
1474 }
1475
1476 /* flush inline_data */
1477 if (is_inline_node(page)) {
1478 clear_inline_node(page);
1479 unlock_page(page);
1480 flush_inline_data(sbi, ino_of_node(page));
1481 goto lock_node;
1482 }
1483
1484 f2fs_wait_on_page_writeback(page, NODE, true);
1485
1486 BUG_ON(PageWriteback(page));
1487 if (!clear_page_dirty_for_io(page))
1488 goto continue_unlock;
1489
1490 set_fsync_mark(page, 0);
1491 set_dentry_mark(page, 0);
1492
1493 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1494 unlock_page(page);
1495
1496 if (--wbc->nr_to_write == 0)
1497 break;
1498 }
1499 pagevec_release(&pvec);
1500 cond_resched();
1501
1502 if (wbc->nr_to_write == 0) {
1503 step = 2;
1504 break;
1505 }
1506 }
1507
1508 if (step < 2) {
1509 step++;
1510 goto next_step;
1511 }
1512 return nwritten;
1513 }
1514
1515 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1516 {
1517 pgoff_t index = 0, end = ULONG_MAX;
1518 struct pagevec pvec;
1519 int ret2 = 0, ret = 0;
1520
1521 pagevec_init(&pvec, 0);
1522
1523 while (index <= end) {
1524 int i, nr_pages;
1525 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1526 PAGECACHE_TAG_WRITEBACK,
1527 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1528 if (nr_pages == 0)
1529 break;
1530
1531 for (i = 0; i < nr_pages; i++) {
1532 struct page *page = pvec.pages[i];
1533
1534 /* until radix tree lookup accepts end_index */
1535 if (unlikely(page->index > end))
1536 continue;
1537
1538 if (ino && ino_of_node(page) == ino) {
1539 f2fs_wait_on_page_writeback(page, NODE, true);
1540 if (TestClearPageError(page))
1541 ret = -EIO;
1542 }
1543 }
1544 pagevec_release(&pvec);
1545 cond_resched();
1546 }
1547
1548 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1549 ret2 = -ENOSPC;
1550 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1551 ret2 = -EIO;
1552 if (!ret)
1553 ret = ret2;
1554 return ret;
1555 }
1556
1557 static int f2fs_write_node_page(struct page *page,
1558 struct writeback_control *wbc)
1559 {
1560 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1561 nid_t nid;
1562 struct node_info ni;
1563 struct f2fs_io_info fio = {
1564 .sbi = sbi,
1565 .type = NODE,
1566 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1567 .page = page,
1568 .encrypted_page = NULL,
1569 };
1570
1571 trace_f2fs_writepage(page, NODE);
1572
1573 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1574 goto redirty_out;
1575 if (unlikely(f2fs_cp_error(sbi)))
1576 goto redirty_out;
1577
1578 /* get old block addr of this node page */
1579 nid = nid_of_node(page);
1580 f2fs_bug_on(sbi, page->index != nid);
1581
1582 if (wbc->for_reclaim) {
1583 if (!down_read_trylock(&sbi->node_write))
1584 goto redirty_out;
1585 } else {
1586 down_read(&sbi->node_write);
1587 }
1588
1589 get_node_info(sbi, nid, &ni);
1590
1591 /* This page is already truncated */
1592 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1593 ClearPageUptodate(page);
1594 dec_page_count(sbi, F2FS_DIRTY_NODES);
1595 up_read(&sbi->node_write);
1596 unlock_page(page);
1597 return 0;
1598 }
1599
1600 set_page_writeback(page);
1601 fio.old_blkaddr = ni.blk_addr;
1602 write_node_page(nid, &fio);
1603 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1604 dec_page_count(sbi, F2FS_DIRTY_NODES);
1605 up_read(&sbi->node_write);
1606
1607 if (wbc->for_reclaim)
1608 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1609
1610 unlock_page(page);
1611
1612 if (unlikely(f2fs_cp_error(sbi)))
1613 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1614
1615 return 0;
1616
1617 redirty_out:
1618 redirty_page_for_writepage(wbc, page);
1619 return AOP_WRITEPAGE_ACTIVATE;
1620 }
1621
1622 static int f2fs_write_node_pages(struct address_space *mapping,
1623 struct writeback_control *wbc)
1624 {
1625 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1626 long diff;
1627
1628 /* balancing f2fs's metadata in background */
1629 f2fs_balance_fs_bg(sbi);
1630
1631 /* collect a number of dirty node pages and write together */
1632 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1633 goto skip_write;
1634
1635 trace_f2fs_writepages(mapping->host, wbc, NODE);
1636
1637 diff = nr_pages_to_write(sbi, NODE, wbc);
1638 wbc->sync_mode = WB_SYNC_NONE;
1639 sync_node_pages(sbi, wbc);
1640 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1641 return 0;
1642
1643 skip_write:
1644 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1645 trace_f2fs_writepages(mapping->host, wbc, NODE);
1646 return 0;
1647 }
1648
1649 static int f2fs_set_node_page_dirty(struct page *page)
1650 {
1651 trace_f2fs_set_page_dirty(page, NODE);
1652
1653 SetPageUptodate(page);
1654 if (!PageDirty(page)) {
1655 __set_page_dirty_nobuffers(page);
1656 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1657 SetPagePrivate(page);
1658 f2fs_trace_pid(page);
1659 return 1;
1660 }
1661 return 0;
1662 }
1663
1664 /*
1665 * Structure of the f2fs node operations
1666 */
1667 const struct address_space_operations f2fs_node_aops = {
1668 .writepage = f2fs_write_node_page,
1669 .writepages = f2fs_write_node_pages,
1670 .set_page_dirty = f2fs_set_node_page_dirty,
1671 .invalidatepage = f2fs_invalidate_page,
1672 .releasepage = f2fs_release_page,
1673 };
1674
1675 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1676 nid_t n)
1677 {
1678 return radix_tree_lookup(&nm_i->free_nid_root, n);
1679 }
1680
1681 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1682 struct free_nid *i)
1683 {
1684 list_del(&i->list);
1685 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1686 }
1687
1688 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1689 {
1690 struct f2fs_nm_info *nm_i = NM_I(sbi);
1691 struct free_nid *i;
1692 struct nat_entry *ne;
1693 bool allocated = false;
1694
1695 if (!available_free_memory(sbi, FREE_NIDS))
1696 return -1;
1697
1698 /* 0 nid should not be used */
1699 if (unlikely(nid == 0))
1700 return 0;
1701
1702 if (build) {
1703 /* do not add allocated nids */
1704 ne = __lookup_nat_cache(nm_i, nid);
1705 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1706 nat_get_blkaddr(ne) != NULL_ADDR))
1707 allocated = true;
1708 if (allocated)
1709 return 0;
1710 }
1711
1712 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1713 i->nid = nid;
1714 i->state = NID_NEW;
1715
1716 if (radix_tree_preload(GFP_NOFS)) {
1717 kmem_cache_free(free_nid_slab, i);
1718 return 0;
1719 }
1720
1721 spin_lock(&nm_i->free_nid_list_lock);
1722 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1723 spin_unlock(&nm_i->free_nid_list_lock);
1724 radix_tree_preload_end();
1725 kmem_cache_free(free_nid_slab, i);
1726 return 0;
1727 }
1728 list_add_tail(&i->list, &nm_i->free_nid_list);
1729 nm_i->fcnt++;
1730 spin_unlock(&nm_i->free_nid_list_lock);
1731 radix_tree_preload_end();
1732 return 1;
1733 }
1734
1735 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1736 {
1737 struct free_nid *i;
1738 bool need_free = false;
1739
1740 spin_lock(&nm_i->free_nid_list_lock);
1741 i = __lookup_free_nid_list(nm_i, nid);
1742 if (i && i->state == NID_NEW) {
1743 __del_from_free_nid_list(nm_i, i);
1744 nm_i->fcnt--;
1745 need_free = true;
1746 }
1747 spin_unlock(&nm_i->free_nid_list_lock);
1748
1749 if (need_free)
1750 kmem_cache_free(free_nid_slab, i);
1751 }
1752
1753 static void scan_nat_page(struct f2fs_sb_info *sbi,
1754 struct page *nat_page, nid_t start_nid)
1755 {
1756 struct f2fs_nm_info *nm_i = NM_I(sbi);
1757 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1758 block_t blk_addr;
1759 int i;
1760
1761 i = start_nid % NAT_ENTRY_PER_BLOCK;
1762
1763 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1764
1765 if (unlikely(start_nid >= nm_i->max_nid))
1766 break;
1767
1768 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1769 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1770 if (blk_addr == NULL_ADDR) {
1771 if (add_free_nid(sbi, start_nid, true) < 0)
1772 break;
1773 }
1774 }
1775 }
1776
1777 static void build_free_nids(struct f2fs_sb_info *sbi)
1778 {
1779 struct f2fs_nm_info *nm_i = NM_I(sbi);
1780 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1781 struct f2fs_journal *journal = curseg->journal;
1782 int i = 0;
1783 nid_t nid = nm_i->next_scan_nid;
1784
1785 /* Enough entries */
1786 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1787 return;
1788
1789 /* readahead nat pages to be scanned */
1790 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1791 META_NAT, true);
1792
1793 down_read(&nm_i->nat_tree_lock);
1794
1795 while (1) {
1796 struct page *page = get_current_nat_page(sbi, nid);
1797
1798 scan_nat_page(sbi, page, nid);
1799 f2fs_put_page(page, 1);
1800
1801 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1802 if (unlikely(nid >= nm_i->max_nid))
1803 nid = 0;
1804
1805 if (++i >= FREE_NID_PAGES)
1806 break;
1807 }
1808
1809 /* go to the next free nat pages to find free nids abundantly */
1810 nm_i->next_scan_nid = nid;
1811
1812 /* find free nids from current sum_pages */
1813 down_read(&curseg->journal_rwsem);
1814 for (i = 0; i < nats_in_cursum(journal); i++) {
1815 block_t addr;
1816
1817 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1818 nid = le32_to_cpu(nid_in_journal(journal, i));
1819 if (addr == NULL_ADDR)
1820 add_free_nid(sbi, nid, true);
1821 else
1822 remove_free_nid(nm_i, nid);
1823 }
1824 up_read(&curseg->journal_rwsem);
1825 up_read(&nm_i->nat_tree_lock);
1826
1827 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1828 nm_i->ra_nid_pages, META_NAT, false);
1829 }
1830
1831 /*
1832 * If this function returns success, caller can obtain a new nid
1833 * from second parameter of this function.
1834 * The returned nid could be used ino as well as nid when inode is created.
1835 */
1836 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1837 {
1838 struct f2fs_nm_info *nm_i = NM_I(sbi);
1839 struct free_nid *i = NULL;
1840 retry:
1841 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1842 return false;
1843
1844 spin_lock(&nm_i->free_nid_list_lock);
1845
1846 /* We should not use stale free nids created by build_free_nids */
1847 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1848 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1849 list_for_each_entry(i, &nm_i->free_nid_list, list)
1850 if (i->state == NID_NEW)
1851 break;
1852
1853 f2fs_bug_on(sbi, i->state != NID_NEW);
1854 *nid = i->nid;
1855 i->state = NID_ALLOC;
1856 nm_i->fcnt--;
1857 spin_unlock(&nm_i->free_nid_list_lock);
1858 return true;
1859 }
1860 spin_unlock(&nm_i->free_nid_list_lock);
1861
1862 /* Let's scan nat pages and its caches to get free nids */
1863 mutex_lock(&nm_i->build_lock);
1864 build_free_nids(sbi);
1865 mutex_unlock(&nm_i->build_lock);
1866 goto retry;
1867 }
1868
1869 /*
1870 * alloc_nid() should be called prior to this function.
1871 */
1872 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1873 {
1874 struct f2fs_nm_info *nm_i = NM_I(sbi);
1875 struct free_nid *i;
1876
1877 spin_lock(&nm_i->free_nid_list_lock);
1878 i = __lookup_free_nid_list(nm_i, nid);
1879 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1880 __del_from_free_nid_list(nm_i, i);
1881 spin_unlock(&nm_i->free_nid_list_lock);
1882
1883 kmem_cache_free(free_nid_slab, i);
1884 }
1885
1886 /*
1887 * alloc_nid() should be called prior to this function.
1888 */
1889 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1890 {
1891 struct f2fs_nm_info *nm_i = NM_I(sbi);
1892 struct free_nid *i;
1893 bool need_free = false;
1894
1895 if (!nid)
1896 return;
1897
1898 spin_lock(&nm_i->free_nid_list_lock);
1899 i = __lookup_free_nid_list(nm_i, nid);
1900 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1901 if (!available_free_memory(sbi, FREE_NIDS)) {
1902 __del_from_free_nid_list(nm_i, i);
1903 need_free = true;
1904 } else {
1905 i->state = NID_NEW;
1906 nm_i->fcnt++;
1907 }
1908 spin_unlock(&nm_i->free_nid_list_lock);
1909
1910 if (need_free)
1911 kmem_cache_free(free_nid_slab, i);
1912 }
1913
1914 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1915 {
1916 struct f2fs_nm_info *nm_i = NM_I(sbi);
1917 struct free_nid *i, *next;
1918 int nr = nr_shrink;
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 <= NAT_ENTRY_PER_BLOCK)
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(F2FS_I(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(inode)->i_xattr_nid = 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 update_inode_page(inode);
2001 }
2002
2003 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2004 {
2005 struct f2fs_inode *src, *dst;
2006 nid_t ino = ino_of_node(page);
2007 struct node_info old_ni, new_ni;
2008 struct page *ipage;
2009
2010 get_node_info(sbi, ino, &old_ni);
2011
2012 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2013 return -EINVAL;
2014
2015 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
2016 if (!ipage)
2017 return -ENOMEM;
2018
2019 /* Should not use this inode from free nid list */
2020 remove_free_nid(NM_I(sbi), ino);
2021
2022 SetPageUptodate(ipage);
2023 fill_node_footer(ipage, ino, ino, 0, true);
2024
2025 src = F2FS_INODE(page);
2026 dst = F2FS_INODE(ipage);
2027
2028 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2029 dst->i_size = 0;
2030 dst->i_blocks = cpu_to_le64(1);
2031 dst->i_links = cpu_to_le32(1);
2032 dst->i_xattr_nid = 0;
2033 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2034
2035 new_ni = old_ni;
2036 new_ni.ino = ino;
2037
2038 if (unlikely(!inc_valid_node_count(sbi, NULL)))
2039 WARN_ON(1);
2040 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2041 inc_valid_inode_count(sbi);
2042 set_page_dirty(ipage);
2043 f2fs_put_page(ipage, 1);
2044 return 0;
2045 }
2046
2047 int restore_node_summary(struct f2fs_sb_info *sbi,
2048 unsigned int segno, struct f2fs_summary_block *sum)
2049 {
2050 struct f2fs_node *rn;
2051 struct f2fs_summary *sum_entry;
2052 block_t addr;
2053 int bio_blocks = MAX_BIO_BLOCKS(sbi);
2054 int i, idx, last_offset, nrpages;
2055
2056 /* scan the node segment */
2057 last_offset = sbi->blocks_per_seg;
2058 addr = START_BLOCK(sbi, segno);
2059 sum_entry = &sum->entries[0];
2060
2061 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2062 nrpages = min(last_offset - i, bio_blocks);
2063
2064 /* readahead node pages */
2065 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2066
2067 for (idx = addr; idx < addr + nrpages; idx++) {
2068 struct page *page = get_tmp_page(sbi, idx);
2069
2070 rn = F2FS_NODE(page);
2071 sum_entry->nid = rn->footer.nid;
2072 sum_entry->version = 0;
2073 sum_entry->ofs_in_node = 0;
2074 sum_entry++;
2075 f2fs_put_page(page, 1);
2076 }
2077
2078 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2079 addr + nrpages);
2080 }
2081 return 0;
2082 }
2083
2084 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2085 {
2086 struct f2fs_nm_info *nm_i = NM_I(sbi);
2087 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2088 struct f2fs_journal *journal = curseg->journal;
2089 int i;
2090
2091 down_write(&curseg->journal_rwsem);
2092 for (i = 0; i < nats_in_cursum(journal); i++) {
2093 struct nat_entry *ne;
2094 struct f2fs_nat_entry raw_ne;
2095 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2096
2097 raw_ne = nat_in_journal(journal, i);
2098
2099 ne = __lookup_nat_cache(nm_i, nid);
2100 if (!ne) {
2101 ne = grab_nat_entry(nm_i, nid);
2102 node_info_from_raw_nat(&ne->ni, &raw_ne);
2103 }
2104 __set_nat_cache_dirty(nm_i, ne);
2105 }
2106 update_nats_in_cursum(journal, -i);
2107 up_write(&curseg->journal_rwsem);
2108 }
2109
2110 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2111 struct list_head *head, int max)
2112 {
2113 struct nat_entry_set *cur;
2114
2115 if (nes->entry_cnt >= max)
2116 goto add_out;
2117
2118 list_for_each_entry(cur, head, set_list) {
2119 if (cur->entry_cnt >= nes->entry_cnt) {
2120 list_add(&nes->set_list, cur->set_list.prev);
2121 return;
2122 }
2123 }
2124 add_out:
2125 list_add_tail(&nes->set_list, head);
2126 }
2127
2128 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2129 struct nat_entry_set *set)
2130 {
2131 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2132 struct f2fs_journal *journal = curseg->journal;
2133 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2134 bool to_journal = true;
2135 struct f2fs_nat_block *nat_blk;
2136 struct nat_entry *ne, *cur;
2137 struct page *page = NULL;
2138
2139 /*
2140 * there are two steps to flush nat entries:
2141 * #1, flush nat entries to journal in current hot data summary block.
2142 * #2, flush nat entries to nat page.
2143 */
2144 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2145 to_journal = false;
2146
2147 if (to_journal) {
2148 down_write(&curseg->journal_rwsem);
2149 } else {
2150 page = get_next_nat_page(sbi, start_nid);
2151 nat_blk = page_address(page);
2152 f2fs_bug_on(sbi, !nat_blk);
2153 }
2154
2155 /* flush dirty nats in nat entry set */
2156 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2157 struct f2fs_nat_entry *raw_ne;
2158 nid_t nid = nat_get_nid(ne);
2159 int offset;
2160
2161 if (nat_get_blkaddr(ne) == NEW_ADDR)
2162 continue;
2163
2164 if (to_journal) {
2165 offset = lookup_journal_in_cursum(journal,
2166 NAT_JOURNAL, nid, 1);
2167 f2fs_bug_on(sbi, offset < 0);
2168 raw_ne = &nat_in_journal(journal, offset);
2169 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2170 } else {
2171 raw_ne = &nat_blk->entries[nid - start_nid];
2172 }
2173 raw_nat_from_node_info(raw_ne, &ne->ni);
2174 nat_reset_flag(ne);
2175 __clear_nat_cache_dirty(NM_I(sbi), ne);
2176 if (nat_get_blkaddr(ne) == NULL_ADDR)
2177 add_free_nid(sbi, nid, false);
2178 }
2179
2180 if (to_journal)
2181 up_write(&curseg->journal_rwsem);
2182 else
2183 f2fs_put_page(page, 1);
2184
2185 f2fs_bug_on(sbi, set->entry_cnt);
2186
2187 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2188 kmem_cache_free(nat_entry_set_slab, set);
2189 }
2190
2191 /*
2192 * This function is called during the checkpointing process.
2193 */
2194 void flush_nat_entries(struct f2fs_sb_info *sbi)
2195 {
2196 struct f2fs_nm_info *nm_i = NM_I(sbi);
2197 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2198 struct f2fs_journal *journal = curseg->journal;
2199 struct nat_entry_set *setvec[SETVEC_SIZE];
2200 struct nat_entry_set *set, *tmp;
2201 unsigned int found;
2202 nid_t set_idx = 0;
2203 LIST_HEAD(sets);
2204
2205 if (!nm_i->dirty_nat_cnt)
2206 return;
2207
2208 down_write(&nm_i->nat_tree_lock);
2209
2210 /*
2211 * if there are no enough space in journal to store dirty nat
2212 * entries, remove all entries from journal and merge them
2213 * into nat entry set.
2214 */
2215 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2216 remove_nats_in_journal(sbi);
2217
2218 while ((found = __gang_lookup_nat_set(nm_i,
2219 set_idx, SETVEC_SIZE, setvec))) {
2220 unsigned idx;
2221 set_idx = setvec[found - 1]->set + 1;
2222 for (idx = 0; idx < found; idx++)
2223 __adjust_nat_entry_set(setvec[idx], &sets,
2224 MAX_NAT_JENTRIES(journal));
2225 }
2226
2227 /* flush dirty nats in nat entry set */
2228 list_for_each_entry_safe(set, tmp, &sets, set_list)
2229 __flush_nat_entry_set(sbi, set);
2230
2231 up_write(&nm_i->nat_tree_lock);
2232
2233 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2234 }
2235
2236 static int init_node_manager(struct f2fs_sb_info *sbi)
2237 {
2238 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2239 struct f2fs_nm_info *nm_i = NM_I(sbi);
2240 unsigned char *version_bitmap;
2241 unsigned int nat_segs, nat_blocks;
2242
2243 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2244
2245 /* segment_count_nat includes pair segment so divide to 2. */
2246 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2247 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2248
2249 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2250
2251 /* not used nids: 0, node, meta, (and root counted as valid node) */
2252 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2253 nm_i->fcnt = 0;
2254 nm_i->nat_cnt = 0;
2255 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2256 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2257 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2258
2259 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2260 INIT_LIST_HEAD(&nm_i->free_nid_list);
2261 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2262 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2263 INIT_LIST_HEAD(&nm_i->nat_entries);
2264
2265 mutex_init(&nm_i->build_lock);
2266 spin_lock_init(&nm_i->free_nid_list_lock);
2267 init_rwsem(&nm_i->nat_tree_lock);
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 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 up_write(&nm_i->nat_tree_lock);
2350
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 0.107463 seconds and 5 git commands to generate.