]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/f2fs/node.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: cover more area with nat_tree_lock
[mirror_ubuntu-zesty-kernel.git] / fs / f2fs / node.c
1 /*
2 * fs/f2fs/node.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
18
19 #include "f2fs.h"
20 #include "node.h"
21 #include "segment.h"
22 #include "trace.h"
23 #include <trace/events/f2fs.h>
24
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
30
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32 {
33 struct f2fs_nm_info *nm_i = NM_I(sbi);
34 struct sysinfo val;
35 unsigned long avail_ram;
36 unsigned long mem_size = 0;
37 bool res = false;
38
39 si_meminfo(&val);
40
41 /* only uses low memory */
42 avail_ram = val.totalram - val.totalhigh;
43
44 /*
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
46 */
47 if (type == FREE_NIDS) {
48 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
49 PAGE_CACHE_SHIFT;
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 PAGE_CACHE_SHIFT;
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 } else if (type == DIRTY_DENTS) {
56 if (sbi->sb->s_bdi->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_CACHE_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_CACHE_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 false;
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_CACHE_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_nm_info *nm_i, nid_t nid,
261 struct f2fs_nat_entry *ne)
262 {
263 struct nat_entry *e;
264
265 e = __lookup_nat_cache(nm_i, nid);
266 if (!e) {
267 e = grab_nat_entry(nm_i, nid);
268 node_info_from_raw_nat(&e->ni, ne);
269 }
270 }
271
272 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
273 block_t new_blkaddr, bool fsync_done)
274 {
275 struct f2fs_nm_info *nm_i = NM_I(sbi);
276 struct nat_entry *e;
277
278 down_write(&nm_i->nat_tree_lock);
279 e = __lookup_nat_cache(nm_i, ni->nid);
280 if (!e) {
281 e = grab_nat_entry(nm_i, ni->nid);
282 copy_node_info(&e->ni, ni);
283 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
284 } else if (new_blkaddr == NEW_ADDR) {
285 /*
286 * when nid is reallocated,
287 * previous nat entry can be remained in nat cache.
288 * So, reinitialize it with new information.
289 */
290 copy_node_info(&e->ni, ni);
291 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
292 }
293
294 /* sanity check */
295 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
296 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
297 new_blkaddr == NULL_ADDR);
298 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
299 new_blkaddr == NEW_ADDR);
300 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
301 nat_get_blkaddr(e) != NULL_ADDR &&
302 new_blkaddr == NEW_ADDR);
303
304 /* increment version no as node is removed */
305 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
306 unsigned char version = nat_get_version(e);
307 nat_set_version(e, inc_node_version(version));
308
309 /* in order to reuse the nid */
310 if (nm_i->next_scan_nid > ni->nid)
311 nm_i->next_scan_nid = ni->nid;
312 }
313
314 /* change address */
315 nat_set_blkaddr(e, new_blkaddr);
316 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
317 set_nat_flag(e, IS_CHECKPOINTED, false);
318 __set_nat_cache_dirty(nm_i, e);
319
320 /* update fsync_mark if its inode nat entry is still alive */
321 if (ni->nid != ni->ino)
322 e = __lookup_nat_cache(nm_i, ni->ino);
323 if (e) {
324 if (fsync_done && ni->nid == ni->ino)
325 set_nat_flag(e, HAS_FSYNCED_INODE, true);
326 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
327 }
328 up_write(&nm_i->nat_tree_lock);
329 }
330
331 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
332 {
333 struct f2fs_nm_info *nm_i = NM_I(sbi);
334 int nr = nr_shrink;
335
336 if (!down_write_trylock(&nm_i->nat_tree_lock))
337 return 0;
338
339 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
340 struct nat_entry *ne;
341 ne = list_first_entry(&nm_i->nat_entries,
342 struct nat_entry, list);
343 __del_from_nat_cache(nm_i, ne);
344 nr_shrink--;
345 }
346 up_write(&nm_i->nat_tree_lock);
347 return nr - nr_shrink;
348 }
349
350 /*
351 * This function always returns success
352 */
353 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
354 {
355 struct f2fs_nm_info *nm_i = NM_I(sbi);
356 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
357 struct f2fs_summary_block *sum = curseg->sum_blk;
358 nid_t start_nid = START_NID(nid);
359 struct f2fs_nat_block *nat_blk;
360 struct page *page = NULL;
361 struct f2fs_nat_entry ne;
362 struct nat_entry *e;
363 int i;
364
365 ni->nid = nid;
366
367 /* Check nat cache */
368 down_read(&nm_i->nat_tree_lock);
369 e = __lookup_nat_cache(nm_i, nid);
370 if (e) {
371 ni->ino = nat_get_ino(e);
372 ni->blk_addr = nat_get_blkaddr(e);
373 ni->version = nat_get_version(e);
374 }
375 up_read(&nm_i->nat_tree_lock);
376 if (e)
377 return;
378
379 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
380
381 down_write(&nm_i->nat_tree_lock);
382
383 /* Check current segment summary */
384 mutex_lock(&curseg->curseg_mutex);
385 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
386 if (i >= 0) {
387 ne = nat_in_journal(sum, i);
388 node_info_from_raw_nat(ni, &ne);
389 }
390 mutex_unlock(&curseg->curseg_mutex);
391 if (i >= 0)
392 goto cache;
393
394 /* Fill node_info from nat page */
395 page = get_current_nat_page(sbi, start_nid);
396 nat_blk = (struct f2fs_nat_block *)page_address(page);
397 ne = nat_blk->entries[nid - start_nid];
398 node_info_from_raw_nat(ni, &ne);
399 f2fs_put_page(page, 1);
400 cache:
401 /* cache nat entry */
402 cache_nat_entry(NM_I(sbi), nid, &ne);
403 up_write(&nm_i->nat_tree_lock);
404 }
405
406 /*
407 * The maximum depth is four.
408 * Offset[0] will have raw inode offset.
409 */
410 static int get_node_path(struct f2fs_inode_info *fi, long block,
411 int offset[4], unsigned int noffset[4])
412 {
413 const long direct_index = ADDRS_PER_INODE(fi);
414 const long direct_blks = ADDRS_PER_BLOCK;
415 const long dptrs_per_blk = NIDS_PER_BLOCK;
416 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
417 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
418 int n = 0;
419 int level = 0;
420
421 noffset[0] = 0;
422
423 if (block < direct_index) {
424 offset[n] = block;
425 goto got;
426 }
427 block -= direct_index;
428 if (block < direct_blks) {
429 offset[n++] = NODE_DIR1_BLOCK;
430 noffset[n] = 1;
431 offset[n] = block;
432 level = 1;
433 goto got;
434 }
435 block -= direct_blks;
436 if (block < direct_blks) {
437 offset[n++] = NODE_DIR2_BLOCK;
438 noffset[n] = 2;
439 offset[n] = block;
440 level = 1;
441 goto got;
442 }
443 block -= direct_blks;
444 if (block < indirect_blks) {
445 offset[n++] = NODE_IND1_BLOCK;
446 noffset[n] = 3;
447 offset[n++] = block / direct_blks;
448 noffset[n] = 4 + offset[n - 1];
449 offset[n] = block % direct_blks;
450 level = 2;
451 goto got;
452 }
453 block -= indirect_blks;
454 if (block < indirect_blks) {
455 offset[n++] = NODE_IND2_BLOCK;
456 noffset[n] = 4 + dptrs_per_blk;
457 offset[n++] = block / direct_blks;
458 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
459 offset[n] = block % direct_blks;
460 level = 2;
461 goto got;
462 }
463 block -= indirect_blks;
464 if (block < dindirect_blks) {
465 offset[n++] = NODE_DIND_BLOCK;
466 noffset[n] = 5 + (dptrs_per_blk * 2);
467 offset[n++] = block / indirect_blks;
468 noffset[n] = 6 + (dptrs_per_blk * 2) +
469 offset[n - 1] * (dptrs_per_blk + 1);
470 offset[n++] = (block / direct_blks) % dptrs_per_blk;
471 noffset[n] = 7 + (dptrs_per_blk * 2) +
472 offset[n - 2] * (dptrs_per_blk + 1) +
473 offset[n - 1];
474 offset[n] = block % direct_blks;
475 level = 3;
476 goto got;
477 } else {
478 BUG();
479 }
480 got:
481 return level;
482 }
483
484 /*
485 * Caller should call f2fs_put_dnode(dn).
486 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
487 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
488 * In the case of RDONLY_NODE, we don't need to care about mutex.
489 */
490 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
491 {
492 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
493 struct page *npage[4];
494 struct page *parent = NULL;
495 int offset[4];
496 unsigned int noffset[4];
497 nid_t nids[4];
498 int level, i;
499 int err = 0;
500
501 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
502
503 nids[0] = dn->inode->i_ino;
504 npage[0] = dn->inode_page;
505
506 if (!npage[0]) {
507 npage[0] = get_node_page(sbi, nids[0]);
508 if (IS_ERR(npage[0]))
509 return PTR_ERR(npage[0]);
510 }
511
512 /* if inline_data is set, should not report any block indices */
513 if (f2fs_has_inline_data(dn->inode) && index) {
514 err = -ENOENT;
515 f2fs_put_page(npage[0], 1);
516 goto release_out;
517 }
518
519 parent = npage[0];
520 if (level != 0)
521 nids[1] = get_nid(parent, offset[0], true);
522 dn->inode_page = npage[0];
523 dn->inode_page_locked = true;
524
525 /* get indirect or direct nodes */
526 for (i = 1; i <= level; i++) {
527 bool done = false;
528
529 if (!nids[i] && mode == ALLOC_NODE) {
530 /* alloc new node */
531 if (!alloc_nid(sbi, &(nids[i]))) {
532 err = -ENOSPC;
533 goto release_pages;
534 }
535
536 dn->nid = nids[i];
537 npage[i] = new_node_page(dn, noffset[i], NULL);
538 if (IS_ERR(npage[i])) {
539 alloc_nid_failed(sbi, nids[i]);
540 err = PTR_ERR(npage[i]);
541 goto release_pages;
542 }
543
544 set_nid(parent, offset[i - 1], nids[i], i == 1);
545 alloc_nid_done(sbi, nids[i]);
546 dn->node_changed = true;
547 done = true;
548 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
549 npage[i] = get_node_page_ra(parent, offset[i - 1]);
550 if (IS_ERR(npage[i])) {
551 err = PTR_ERR(npage[i]);
552 goto release_pages;
553 }
554 done = true;
555 }
556 if (i == 1) {
557 dn->inode_page_locked = false;
558 unlock_page(parent);
559 } else {
560 f2fs_put_page(parent, 1);
561 }
562
563 if (!done) {
564 npage[i] = get_node_page(sbi, nids[i]);
565 if (IS_ERR(npage[i])) {
566 err = PTR_ERR(npage[i]);
567 f2fs_put_page(npage[0], 0);
568 goto release_out;
569 }
570 }
571 if (i < level) {
572 parent = npage[i];
573 nids[i + 1] = get_nid(parent, offset[i], false);
574 }
575 }
576 dn->nid = nids[level];
577 dn->ofs_in_node = offset[level];
578 dn->node_page = npage[level];
579 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
580 return 0;
581
582 release_pages:
583 f2fs_put_page(parent, 1);
584 if (i > 1)
585 f2fs_put_page(npage[0], 0);
586 release_out:
587 dn->inode_page = NULL;
588 dn->node_page = NULL;
589 return err;
590 }
591
592 static void truncate_node(struct dnode_of_data *dn)
593 {
594 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
595 struct node_info ni;
596
597 get_node_info(sbi, dn->nid, &ni);
598 if (dn->inode->i_blocks == 0) {
599 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
600 goto invalidate;
601 }
602 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
603
604 /* Deallocate node address */
605 invalidate_blocks(sbi, ni.blk_addr);
606 dec_valid_node_count(sbi, dn->inode);
607 set_node_addr(sbi, &ni, NULL_ADDR, false);
608
609 if (dn->nid == dn->inode->i_ino) {
610 remove_orphan_inode(sbi, dn->nid);
611 dec_valid_inode_count(sbi);
612 } else {
613 sync_inode_page(dn);
614 }
615 invalidate:
616 clear_node_page_dirty(dn->node_page);
617 set_sbi_flag(sbi, SBI_IS_DIRTY);
618
619 f2fs_put_page(dn->node_page, 1);
620
621 invalidate_mapping_pages(NODE_MAPPING(sbi),
622 dn->node_page->index, dn->node_page->index);
623
624 dn->node_page = NULL;
625 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
626 }
627
628 static int truncate_dnode(struct dnode_of_data *dn)
629 {
630 struct page *page;
631
632 if (dn->nid == 0)
633 return 1;
634
635 /* get direct node */
636 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
637 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
638 return 1;
639 else if (IS_ERR(page))
640 return PTR_ERR(page);
641
642 /* Make dnode_of_data for parameter */
643 dn->node_page = page;
644 dn->ofs_in_node = 0;
645 truncate_data_blocks(dn);
646 truncate_node(dn);
647 return 1;
648 }
649
650 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
651 int ofs, int depth)
652 {
653 struct dnode_of_data rdn = *dn;
654 struct page *page;
655 struct f2fs_node *rn;
656 nid_t child_nid;
657 unsigned int child_nofs;
658 int freed = 0;
659 int i, ret;
660
661 if (dn->nid == 0)
662 return NIDS_PER_BLOCK + 1;
663
664 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
665
666 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
667 if (IS_ERR(page)) {
668 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
669 return PTR_ERR(page);
670 }
671
672 rn = F2FS_NODE(page);
673 if (depth < 3) {
674 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
675 child_nid = le32_to_cpu(rn->in.nid[i]);
676 if (child_nid == 0)
677 continue;
678 rdn.nid = child_nid;
679 ret = truncate_dnode(&rdn);
680 if (ret < 0)
681 goto out_err;
682 set_nid(page, i, 0, false);
683 dn->node_changed = true;
684 }
685 } else {
686 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
687 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
688 child_nid = le32_to_cpu(rn->in.nid[i]);
689 if (child_nid == 0) {
690 child_nofs += NIDS_PER_BLOCK + 1;
691 continue;
692 }
693 rdn.nid = child_nid;
694 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
695 if (ret == (NIDS_PER_BLOCK + 1)) {
696 set_nid(page, i, 0, false);
697 dn->node_changed = true;
698 child_nofs += ret;
699 } else if (ret < 0 && ret != -ENOENT) {
700 goto out_err;
701 }
702 }
703 freed = child_nofs;
704 }
705
706 if (!ofs) {
707 /* remove current indirect node */
708 dn->node_page = page;
709 truncate_node(dn);
710 freed++;
711 } else {
712 f2fs_put_page(page, 1);
713 }
714 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
715 return freed;
716
717 out_err:
718 f2fs_put_page(page, 1);
719 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
720 return ret;
721 }
722
723 static int truncate_partial_nodes(struct dnode_of_data *dn,
724 struct f2fs_inode *ri, int *offset, int depth)
725 {
726 struct page *pages[2];
727 nid_t nid[3];
728 nid_t child_nid;
729 int err = 0;
730 int i;
731 int idx = depth - 2;
732
733 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
734 if (!nid[0])
735 return 0;
736
737 /* get indirect nodes in the path */
738 for (i = 0; i < idx + 1; i++) {
739 /* reference count'll be increased */
740 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
741 if (IS_ERR(pages[i])) {
742 err = PTR_ERR(pages[i]);
743 idx = i - 1;
744 goto fail;
745 }
746 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
747 }
748
749 /* free direct nodes linked to a partial indirect node */
750 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
751 child_nid = get_nid(pages[idx], i, false);
752 if (!child_nid)
753 continue;
754 dn->nid = child_nid;
755 err = truncate_dnode(dn);
756 if (err < 0)
757 goto fail;
758 set_nid(pages[idx], i, 0, false);
759 dn->node_changed = true;
760 }
761
762 if (offset[idx + 1] == 0) {
763 dn->node_page = pages[idx];
764 dn->nid = nid[idx];
765 truncate_node(dn);
766 } else {
767 f2fs_put_page(pages[idx], 1);
768 }
769 offset[idx]++;
770 offset[idx + 1] = 0;
771 idx--;
772 fail:
773 for (i = idx; i >= 0; i--)
774 f2fs_put_page(pages[i], 1);
775
776 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
777
778 return err;
779 }
780
781 /*
782 * All the block addresses of data and nodes should be nullified.
783 */
784 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
785 {
786 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
787 int err = 0, cont = 1;
788 int level, offset[4], noffset[4];
789 unsigned int nofs = 0;
790 struct f2fs_inode *ri;
791 struct dnode_of_data dn;
792 struct page *page;
793
794 trace_f2fs_truncate_inode_blocks_enter(inode, from);
795
796 level = get_node_path(F2FS_I(inode), from, offset, noffset);
797 restart:
798 page = get_node_page(sbi, inode->i_ino);
799 if (IS_ERR(page)) {
800 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
801 return PTR_ERR(page);
802 }
803
804 set_new_dnode(&dn, inode, page, NULL, 0);
805 unlock_page(page);
806
807 ri = F2FS_INODE(page);
808 switch (level) {
809 case 0:
810 case 1:
811 nofs = noffset[1];
812 break;
813 case 2:
814 nofs = noffset[1];
815 if (!offset[level - 1])
816 goto skip_partial;
817 err = truncate_partial_nodes(&dn, ri, offset, level);
818 if (err < 0 && err != -ENOENT)
819 goto fail;
820 nofs += 1 + NIDS_PER_BLOCK;
821 break;
822 case 3:
823 nofs = 5 + 2 * NIDS_PER_BLOCK;
824 if (!offset[level - 1])
825 goto skip_partial;
826 err = truncate_partial_nodes(&dn, ri, offset, level);
827 if (err < 0 && err != -ENOENT)
828 goto fail;
829 break;
830 default:
831 BUG();
832 }
833
834 skip_partial:
835 while (cont) {
836 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
837 switch (offset[0]) {
838 case NODE_DIR1_BLOCK:
839 case NODE_DIR2_BLOCK:
840 err = truncate_dnode(&dn);
841 break;
842
843 case NODE_IND1_BLOCK:
844 case NODE_IND2_BLOCK:
845 err = truncate_nodes(&dn, nofs, offset[1], 2);
846 break;
847
848 case NODE_DIND_BLOCK:
849 err = truncate_nodes(&dn, nofs, offset[1], 3);
850 cont = 0;
851 break;
852
853 default:
854 BUG();
855 }
856 if (err < 0 && err != -ENOENT)
857 goto fail;
858 if (offset[1] == 0 &&
859 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
860 lock_page(page);
861 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
862 f2fs_put_page(page, 1);
863 goto restart;
864 }
865 f2fs_wait_on_page_writeback(page, NODE);
866 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
867 set_page_dirty(page);
868 unlock_page(page);
869 }
870 offset[1] = 0;
871 offset[0]++;
872 nofs += err;
873 }
874 fail:
875 f2fs_put_page(page, 0);
876 trace_f2fs_truncate_inode_blocks_exit(inode, err);
877 return err > 0 ? 0 : err;
878 }
879
880 int truncate_xattr_node(struct inode *inode, struct page *page)
881 {
882 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
883 nid_t nid = F2FS_I(inode)->i_xattr_nid;
884 struct dnode_of_data dn;
885 struct page *npage;
886
887 if (!nid)
888 return 0;
889
890 npage = get_node_page(sbi, nid);
891 if (IS_ERR(npage))
892 return PTR_ERR(npage);
893
894 F2FS_I(inode)->i_xattr_nid = 0;
895
896 /* need to do checkpoint during fsync */
897 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
898
899 set_new_dnode(&dn, inode, page, npage, nid);
900
901 if (page)
902 dn.inode_page_locked = true;
903 truncate_node(&dn);
904 return 0;
905 }
906
907 /*
908 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
909 * f2fs_unlock_op().
910 */
911 int remove_inode_page(struct inode *inode)
912 {
913 struct dnode_of_data dn;
914 int err;
915
916 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
917 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
918 if (err)
919 return err;
920
921 err = truncate_xattr_node(inode, dn.inode_page);
922 if (err) {
923 f2fs_put_dnode(&dn);
924 return err;
925 }
926
927 /* remove potential inline_data blocks */
928 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
929 S_ISLNK(inode->i_mode))
930 truncate_data_blocks_range(&dn, 1);
931
932 /* 0 is possible, after f2fs_new_inode() has failed */
933 f2fs_bug_on(F2FS_I_SB(inode),
934 inode->i_blocks != 0 && inode->i_blocks != 1);
935
936 /* will put inode & node pages */
937 truncate_node(&dn);
938 return 0;
939 }
940
941 struct page *new_inode_page(struct inode *inode)
942 {
943 struct dnode_of_data dn;
944
945 /* allocate inode page for new inode */
946 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
947
948 /* caller should f2fs_put_page(page, 1); */
949 return new_node_page(&dn, 0, NULL);
950 }
951
952 struct page *new_node_page(struct dnode_of_data *dn,
953 unsigned int ofs, struct page *ipage)
954 {
955 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
956 struct node_info old_ni, new_ni;
957 struct page *page;
958 int err;
959
960 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
961 return ERR_PTR(-EPERM);
962
963 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
964 if (!page)
965 return ERR_PTR(-ENOMEM);
966
967 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
968 err = -ENOSPC;
969 goto fail;
970 }
971
972 get_node_info(sbi, dn->nid, &old_ni);
973
974 /* Reinitialize old_ni with new node page */
975 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
976 new_ni = old_ni;
977 new_ni.ino = dn->inode->i_ino;
978 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
979
980 f2fs_wait_on_page_writeback(page, NODE);
981 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
982 set_cold_node(dn->inode, page);
983 SetPageUptodate(page);
984 set_page_dirty(page);
985
986 if (f2fs_has_xattr_block(ofs))
987 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
988
989 dn->node_page = page;
990 if (ipage)
991 update_inode(dn->inode, ipage);
992 else
993 sync_inode_page(dn);
994 if (ofs == 0)
995 inc_valid_inode_count(sbi);
996
997 return page;
998
999 fail:
1000 clear_node_page_dirty(page);
1001 f2fs_put_page(page, 1);
1002 return ERR_PTR(err);
1003 }
1004
1005 /*
1006 * Caller should do after getting the following values.
1007 * 0: f2fs_put_page(page, 0)
1008 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1009 */
1010 static int read_node_page(struct page *page, int rw)
1011 {
1012 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1013 struct node_info ni;
1014 struct f2fs_io_info fio = {
1015 .sbi = sbi,
1016 .type = NODE,
1017 .rw = rw,
1018 .page = page,
1019 .encrypted_page = NULL,
1020 };
1021
1022 get_node_info(sbi, page->index, &ni);
1023
1024 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1025 ClearPageUptodate(page);
1026 return -ENOENT;
1027 }
1028
1029 if (PageUptodate(page))
1030 return LOCKED_PAGE;
1031
1032 fio.blk_addr = ni.blk_addr;
1033 return f2fs_submit_page_bio(&fio);
1034 }
1035
1036 /*
1037 * Readahead a node page
1038 */
1039 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1040 {
1041 struct page *apage;
1042 int err;
1043
1044 apage = find_get_page(NODE_MAPPING(sbi), nid);
1045 if (apage && PageUptodate(apage)) {
1046 f2fs_put_page(apage, 0);
1047 return;
1048 }
1049 f2fs_put_page(apage, 0);
1050
1051 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1052 if (!apage)
1053 return;
1054
1055 err = read_node_page(apage, READA);
1056 f2fs_put_page(apage, err ? 1 : 0);
1057 }
1058
1059 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1060 {
1061 struct page *page;
1062 int err;
1063
1064 if (!nid)
1065 return ERR_PTR(-ENOENT);
1066 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1067 repeat:
1068 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1069 if (!page)
1070 return ERR_PTR(-ENOMEM);
1071
1072 err = read_node_page(page, READ_SYNC);
1073 if (err < 0) {
1074 f2fs_put_page(page, 1);
1075 return ERR_PTR(err);
1076 } else if (err == LOCKED_PAGE) {
1077 goto page_hit;
1078 }
1079
1080 lock_page(page);
1081
1082 if (unlikely(!PageUptodate(page))) {
1083 f2fs_put_page(page, 1);
1084 return ERR_PTR(-EIO);
1085 }
1086 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1087 f2fs_put_page(page, 1);
1088 goto repeat;
1089 }
1090 page_hit:
1091 f2fs_bug_on(sbi, nid != nid_of_node(page));
1092 return page;
1093 }
1094
1095 /*
1096 * Return a locked page for the desired node page.
1097 * And, readahead MAX_RA_NODE number of node pages.
1098 */
1099 struct page *get_node_page_ra(struct page *parent, int start)
1100 {
1101 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1102 struct blk_plug plug;
1103 struct page *page;
1104 int err, i, end;
1105 nid_t nid;
1106
1107 /* First, try getting the desired direct node. */
1108 nid = get_nid(parent, start, false);
1109 if (!nid)
1110 return ERR_PTR(-ENOENT);
1111 repeat:
1112 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1113 if (!page)
1114 return ERR_PTR(-ENOMEM);
1115
1116 err = read_node_page(page, READ_SYNC);
1117 if (err < 0) {
1118 f2fs_put_page(page, 1);
1119 return ERR_PTR(err);
1120 } else if (err == LOCKED_PAGE) {
1121 goto page_hit;
1122 }
1123
1124 blk_start_plug(&plug);
1125
1126 /* Then, try readahead for siblings of the desired node */
1127 end = start + MAX_RA_NODE;
1128 end = min(end, NIDS_PER_BLOCK);
1129 for (i = start + 1; i < end; i++) {
1130 nid_t tnid = get_nid(parent, i, false);
1131 if (!tnid)
1132 continue;
1133 ra_node_page(sbi, tnid);
1134 }
1135
1136 blk_finish_plug(&plug);
1137
1138 lock_page(page);
1139 if (unlikely(!PageUptodate(page))) {
1140 f2fs_put_page(page, 1);
1141 return ERR_PTR(-EIO);
1142 }
1143 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1144 f2fs_put_page(page, 1);
1145 goto repeat;
1146 }
1147 page_hit:
1148 f2fs_bug_on(sbi, nid != nid_of_node(page));
1149 return page;
1150 }
1151
1152 void sync_inode_page(struct dnode_of_data *dn)
1153 {
1154 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1155 update_inode(dn->inode, dn->node_page);
1156 } else if (dn->inode_page) {
1157 if (!dn->inode_page_locked)
1158 lock_page(dn->inode_page);
1159 update_inode(dn->inode, dn->inode_page);
1160 if (!dn->inode_page_locked)
1161 unlock_page(dn->inode_page);
1162 } else {
1163 update_inode_page(dn->inode);
1164 }
1165 dn->node_changed = true;
1166 }
1167
1168 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1169 struct writeback_control *wbc)
1170 {
1171 pgoff_t index, end;
1172 struct pagevec pvec;
1173 int step = ino ? 2 : 0;
1174 int nwritten = 0, wrote = 0;
1175
1176 pagevec_init(&pvec, 0);
1177
1178 next_step:
1179 index = 0;
1180 end = LONG_MAX;
1181
1182 while (index <= end) {
1183 int i, nr_pages;
1184 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1185 PAGECACHE_TAG_DIRTY,
1186 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1187 if (nr_pages == 0)
1188 break;
1189
1190 for (i = 0; i < nr_pages; i++) {
1191 struct page *page = pvec.pages[i];
1192
1193 if (unlikely(f2fs_cp_error(sbi))) {
1194 pagevec_release(&pvec);
1195 return -EIO;
1196 }
1197
1198 /*
1199 * flushing sequence with step:
1200 * 0. indirect nodes
1201 * 1. dentry dnodes
1202 * 2. file dnodes
1203 */
1204 if (step == 0 && IS_DNODE(page))
1205 continue;
1206 if (step == 1 && (!IS_DNODE(page) ||
1207 is_cold_node(page)))
1208 continue;
1209 if (step == 2 && (!IS_DNODE(page) ||
1210 !is_cold_node(page)))
1211 continue;
1212
1213 /*
1214 * If an fsync mode,
1215 * we should not skip writing node pages.
1216 */
1217 if (ino && ino_of_node(page) == ino)
1218 lock_page(page);
1219 else if (!trylock_page(page))
1220 continue;
1221
1222 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1223 continue_unlock:
1224 unlock_page(page);
1225 continue;
1226 }
1227 if (ino && ino_of_node(page) != ino)
1228 goto continue_unlock;
1229
1230 if (!PageDirty(page)) {
1231 /* someone wrote it for us */
1232 goto continue_unlock;
1233 }
1234
1235 if (!clear_page_dirty_for_io(page))
1236 goto continue_unlock;
1237
1238 /* called by fsync() */
1239 if (ino && IS_DNODE(page)) {
1240 set_fsync_mark(page, 1);
1241 if (IS_INODE(page))
1242 set_dentry_mark(page,
1243 need_dentry_mark(sbi, ino));
1244 nwritten++;
1245 } else {
1246 set_fsync_mark(page, 0);
1247 set_dentry_mark(page, 0);
1248 }
1249
1250 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1251 unlock_page(page);
1252 else
1253 wrote++;
1254
1255 if (--wbc->nr_to_write == 0)
1256 break;
1257 }
1258 pagevec_release(&pvec);
1259 cond_resched();
1260
1261 if (wbc->nr_to_write == 0) {
1262 step = 2;
1263 break;
1264 }
1265 }
1266
1267 if (step < 2) {
1268 step++;
1269 goto next_step;
1270 }
1271
1272 if (wrote)
1273 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1274 return nwritten;
1275 }
1276
1277 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1278 {
1279 pgoff_t index = 0, end = LONG_MAX;
1280 struct pagevec pvec;
1281 int ret2 = 0, ret = 0;
1282
1283 pagevec_init(&pvec, 0);
1284
1285 while (index <= end) {
1286 int i, nr_pages;
1287 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1288 PAGECACHE_TAG_WRITEBACK,
1289 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1290 if (nr_pages == 0)
1291 break;
1292
1293 for (i = 0; i < nr_pages; i++) {
1294 struct page *page = pvec.pages[i];
1295
1296 /* until radix tree lookup accepts end_index */
1297 if (unlikely(page->index > end))
1298 continue;
1299
1300 if (ino && ino_of_node(page) == ino) {
1301 f2fs_wait_on_page_writeback(page, NODE);
1302 if (TestClearPageError(page))
1303 ret = -EIO;
1304 }
1305 }
1306 pagevec_release(&pvec);
1307 cond_resched();
1308 }
1309
1310 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1311 ret2 = -ENOSPC;
1312 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1313 ret2 = -EIO;
1314 if (!ret)
1315 ret = ret2;
1316 return ret;
1317 }
1318
1319 static int f2fs_write_node_page(struct page *page,
1320 struct writeback_control *wbc)
1321 {
1322 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1323 nid_t nid;
1324 struct node_info ni;
1325 struct f2fs_io_info fio = {
1326 .sbi = sbi,
1327 .type = NODE,
1328 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1329 .page = page,
1330 .encrypted_page = NULL,
1331 };
1332
1333 trace_f2fs_writepage(page, NODE);
1334
1335 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1336 goto redirty_out;
1337 if (unlikely(f2fs_cp_error(sbi)))
1338 goto redirty_out;
1339
1340 f2fs_wait_on_page_writeback(page, NODE);
1341
1342 /* get old block addr of this node page */
1343 nid = nid_of_node(page);
1344 f2fs_bug_on(sbi, page->index != nid);
1345
1346 if (wbc->for_reclaim) {
1347 if (!down_read_trylock(&sbi->node_write))
1348 goto redirty_out;
1349 } else {
1350 down_read(&sbi->node_write);
1351 }
1352
1353 get_node_info(sbi, nid, &ni);
1354
1355 /* This page is already truncated */
1356 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1357 ClearPageUptodate(page);
1358 dec_page_count(sbi, F2FS_DIRTY_NODES);
1359 up_read(&sbi->node_write);
1360 unlock_page(page);
1361 return 0;
1362 }
1363
1364 set_page_writeback(page);
1365 fio.blk_addr = ni.blk_addr;
1366 write_node_page(nid, &fio);
1367 set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1368 dec_page_count(sbi, F2FS_DIRTY_NODES);
1369 up_read(&sbi->node_write);
1370 unlock_page(page);
1371
1372 if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi)))
1373 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1374
1375 return 0;
1376
1377 redirty_out:
1378 redirty_page_for_writepage(wbc, page);
1379 return AOP_WRITEPAGE_ACTIVATE;
1380 }
1381
1382 static int f2fs_write_node_pages(struct address_space *mapping,
1383 struct writeback_control *wbc)
1384 {
1385 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1386 long diff;
1387
1388 trace_f2fs_writepages(mapping->host, wbc, NODE);
1389
1390 /* balancing f2fs's metadata in background */
1391 f2fs_balance_fs_bg(sbi);
1392
1393 /* collect a number of dirty node pages and write together */
1394 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1395 goto skip_write;
1396
1397 diff = nr_pages_to_write(sbi, NODE, wbc);
1398 wbc->sync_mode = WB_SYNC_NONE;
1399 sync_node_pages(sbi, 0, wbc);
1400 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1401 return 0;
1402
1403 skip_write:
1404 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1405 return 0;
1406 }
1407
1408 static int f2fs_set_node_page_dirty(struct page *page)
1409 {
1410 trace_f2fs_set_page_dirty(page, NODE);
1411
1412 SetPageUptodate(page);
1413 if (!PageDirty(page)) {
1414 __set_page_dirty_nobuffers(page);
1415 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1416 SetPagePrivate(page);
1417 f2fs_trace_pid(page);
1418 return 1;
1419 }
1420 return 0;
1421 }
1422
1423 /*
1424 * Structure of the f2fs node operations
1425 */
1426 const struct address_space_operations f2fs_node_aops = {
1427 .writepage = f2fs_write_node_page,
1428 .writepages = f2fs_write_node_pages,
1429 .set_page_dirty = f2fs_set_node_page_dirty,
1430 .invalidatepage = f2fs_invalidate_page,
1431 .releasepage = f2fs_release_page,
1432 };
1433
1434 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1435 nid_t n)
1436 {
1437 return radix_tree_lookup(&nm_i->free_nid_root, n);
1438 }
1439
1440 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1441 struct free_nid *i)
1442 {
1443 list_del(&i->list);
1444 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1445 }
1446
1447 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1448 {
1449 struct f2fs_nm_info *nm_i = NM_I(sbi);
1450 struct free_nid *i;
1451 struct nat_entry *ne;
1452 bool allocated = false;
1453
1454 if (!available_free_memory(sbi, FREE_NIDS))
1455 return -1;
1456
1457 /* 0 nid should not be used */
1458 if (unlikely(nid == 0))
1459 return 0;
1460
1461 if (build) {
1462 /* do not add allocated nids */
1463 ne = __lookup_nat_cache(nm_i, nid);
1464 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1465 nat_get_blkaddr(ne) != NULL_ADDR))
1466 allocated = true;
1467 if (allocated)
1468 return 0;
1469 }
1470
1471 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1472 i->nid = nid;
1473 i->state = NID_NEW;
1474
1475 if (radix_tree_preload(GFP_NOFS)) {
1476 kmem_cache_free(free_nid_slab, i);
1477 return 0;
1478 }
1479
1480 spin_lock(&nm_i->free_nid_list_lock);
1481 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1482 spin_unlock(&nm_i->free_nid_list_lock);
1483 radix_tree_preload_end();
1484 kmem_cache_free(free_nid_slab, i);
1485 return 0;
1486 }
1487 list_add_tail(&i->list, &nm_i->free_nid_list);
1488 nm_i->fcnt++;
1489 spin_unlock(&nm_i->free_nid_list_lock);
1490 radix_tree_preload_end();
1491 return 1;
1492 }
1493
1494 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1495 {
1496 struct free_nid *i;
1497 bool need_free = false;
1498
1499 spin_lock(&nm_i->free_nid_list_lock);
1500 i = __lookup_free_nid_list(nm_i, nid);
1501 if (i && i->state == NID_NEW) {
1502 __del_from_free_nid_list(nm_i, i);
1503 nm_i->fcnt--;
1504 need_free = true;
1505 }
1506 spin_unlock(&nm_i->free_nid_list_lock);
1507
1508 if (need_free)
1509 kmem_cache_free(free_nid_slab, i);
1510 }
1511
1512 static void scan_nat_page(struct f2fs_sb_info *sbi,
1513 struct page *nat_page, nid_t start_nid)
1514 {
1515 struct f2fs_nm_info *nm_i = NM_I(sbi);
1516 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1517 block_t blk_addr;
1518 int i;
1519
1520 i = start_nid % NAT_ENTRY_PER_BLOCK;
1521
1522 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1523
1524 if (unlikely(start_nid >= nm_i->max_nid))
1525 break;
1526
1527 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1528 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1529 if (blk_addr == NULL_ADDR) {
1530 if (add_free_nid(sbi, start_nid, true) < 0)
1531 break;
1532 }
1533 }
1534 }
1535
1536 static void build_free_nids(struct f2fs_sb_info *sbi)
1537 {
1538 struct f2fs_nm_info *nm_i = NM_I(sbi);
1539 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1540 struct f2fs_summary_block *sum = curseg->sum_blk;
1541 int i = 0;
1542 nid_t nid = nm_i->next_scan_nid;
1543
1544 /* Enough entries */
1545 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1546 return;
1547
1548 /* readahead nat pages to be scanned */
1549 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1550 META_NAT, true);
1551
1552 down_read(&nm_i->nat_tree_lock);
1553
1554 while (1) {
1555 struct page *page = get_current_nat_page(sbi, nid);
1556
1557 scan_nat_page(sbi, page, nid);
1558 f2fs_put_page(page, 1);
1559
1560 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1561 if (unlikely(nid >= nm_i->max_nid))
1562 nid = 0;
1563
1564 if (++i >= FREE_NID_PAGES)
1565 break;
1566 }
1567
1568 /* go to the next free nat pages to find free nids abundantly */
1569 nm_i->next_scan_nid = nid;
1570
1571 /* find free nids from current sum_pages */
1572 mutex_lock(&curseg->curseg_mutex);
1573 for (i = 0; i < nats_in_cursum(sum); i++) {
1574 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1575 nid = le32_to_cpu(nid_in_journal(sum, i));
1576 if (addr == NULL_ADDR)
1577 add_free_nid(sbi, nid, true);
1578 else
1579 remove_free_nid(nm_i, nid);
1580 }
1581 mutex_unlock(&curseg->curseg_mutex);
1582 up_read(&nm_i->nat_tree_lock);
1583
1584 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1585 nm_i->ra_nid_pages, META_NAT, false);
1586 }
1587
1588 /*
1589 * If this function returns success, caller can obtain a new nid
1590 * from second parameter of this function.
1591 * The returned nid could be used ino as well as nid when inode is created.
1592 */
1593 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1594 {
1595 struct f2fs_nm_info *nm_i = NM_I(sbi);
1596 struct free_nid *i = NULL;
1597 retry:
1598 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1599 return false;
1600
1601 spin_lock(&nm_i->free_nid_list_lock);
1602
1603 /* We should not use stale free nids created by build_free_nids */
1604 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1605 struct node_info ni;
1606
1607 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1608 list_for_each_entry(i, &nm_i->free_nid_list, list)
1609 if (i->state == NID_NEW)
1610 break;
1611
1612 f2fs_bug_on(sbi, i->state != NID_NEW);
1613 *nid = i->nid;
1614 i->state = NID_ALLOC;
1615 nm_i->fcnt--;
1616 spin_unlock(&nm_i->free_nid_list_lock);
1617
1618 /* check nid is allocated already */
1619 get_node_info(sbi, *nid, &ni);
1620 if (ni.blk_addr != NULL_ADDR) {
1621 alloc_nid_done(sbi, *nid);
1622 goto retry;
1623 }
1624 return true;
1625 }
1626 spin_unlock(&nm_i->free_nid_list_lock);
1627
1628 /* Let's scan nat pages and its caches to get free nids */
1629 mutex_lock(&nm_i->build_lock);
1630 build_free_nids(sbi);
1631 mutex_unlock(&nm_i->build_lock);
1632 goto retry;
1633 }
1634
1635 /*
1636 * alloc_nid() should be called prior to this function.
1637 */
1638 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1639 {
1640 struct f2fs_nm_info *nm_i = NM_I(sbi);
1641 struct free_nid *i;
1642
1643 spin_lock(&nm_i->free_nid_list_lock);
1644 i = __lookup_free_nid_list(nm_i, nid);
1645 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1646 __del_from_free_nid_list(nm_i, i);
1647 spin_unlock(&nm_i->free_nid_list_lock);
1648
1649 kmem_cache_free(free_nid_slab, i);
1650 }
1651
1652 /*
1653 * alloc_nid() should be called prior to this function.
1654 */
1655 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1656 {
1657 struct f2fs_nm_info *nm_i = NM_I(sbi);
1658 struct free_nid *i;
1659 bool need_free = false;
1660
1661 if (!nid)
1662 return;
1663
1664 spin_lock(&nm_i->free_nid_list_lock);
1665 i = __lookup_free_nid_list(nm_i, nid);
1666 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1667 if (!available_free_memory(sbi, FREE_NIDS)) {
1668 __del_from_free_nid_list(nm_i, i);
1669 need_free = true;
1670 } else {
1671 i->state = NID_NEW;
1672 nm_i->fcnt++;
1673 }
1674 spin_unlock(&nm_i->free_nid_list_lock);
1675
1676 if (need_free)
1677 kmem_cache_free(free_nid_slab, i);
1678 }
1679
1680 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1681 {
1682 struct f2fs_nm_info *nm_i = NM_I(sbi);
1683 struct free_nid *i, *next;
1684 int nr = nr_shrink;
1685
1686 if (!mutex_trylock(&nm_i->build_lock))
1687 return 0;
1688
1689 spin_lock(&nm_i->free_nid_list_lock);
1690 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1691 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1692 break;
1693 if (i->state == NID_ALLOC)
1694 continue;
1695 __del_from_free_nid_list(nm_i, i);
1696 kmem_cache_free(free_nid_slab, i);
1697 nm_i->fcnt--;
1698 nr_shrink--;
1699 }
1700 spin_unlock(&nm_i->free_nid_list_lock);
1701 mutex_unlock(&nm_i->build_lock);
1702
1703 return nr - nr_shrink;
1704 }
1705
1706 void recover_inline_xattr(struct inode *inode, struct page *page)
1707 {
1708 void *src_addr, *dst_addr;
1709 size_t inline_size;
1710 struct page *ipage;
1711 struct f2fs_inode *ri;
1712
1713 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1714 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1715
1716 ri = F2FS_INODE(page);
1717 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1718 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1719 goto update_inode;
1720 }
1721
1722 dst_addr = inline_xattr_addr(ipage);
1723 src_addr = inline_xattr_addr(page);
1724 inline_size = inline_xattr_size(inode);
1725
1726 f2fs_wait_on_page_writeback(ipage, NODE);
1727 memcpy(dst_addr, src_addr, inline_size);
1728 update_inode:
1729 update_inode(inode, ipage);
1730 f2fs_put_page(ipage, 1);
1731 }
1732
1733 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1734 {
1735 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1736 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1737 nid_t new_xnid = nid_of_node(page);
1738 struct node_info ni;
1739
1740 /* 1: invalidate the previous xattr nid */
1741 if (!prev_xnid)
1742 goto recover_xnid;
1743
1744 /* Deallocate node address */
1745 get_node_info(sbi, prev_xnid, &ni);
1746 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1747 invalidate_blocks(sbi, ni.blk_addr);
1748 dec_valid_node_count(sbi, inode);
1749 set_node_addr(sbi, &ni, NULL_ADDR, false);
1750
1751 recover_xnid:
1752 /* 2: allocate new xattr nid */
1753 if (unlikely(!inc_valid_node_count(sbi, inode)))
1754 f2fs_bug_on(sbi, 1);
1755
1756 remove_free_nid(NM_I(sbi), new_xnid);
1757 get_node_info(sbi, new_xnid, &ni);
1758 ni.ino = inode->i_ino;
1759 set_node_addr(sbi, &ni, NEW_ADDR, false);
1760 F2FS_I(inode)->i_xattr_nid = new_xnid;
1761
1762 /* 3: update xattr blkaddr */
1763 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1764 set_node_addr(sbi, &ni, blkaddr, false);
1765
1766 update_inode_page(inode);
1767 }
1768
1769 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1770 {
1771 struct f2fs_inode *src, *dst;
1772 nid_t ino = ino_of_node(page);
1773 struct node_info old_ni, new_ni;
1774 struct page *ipage;
1775
1776 get_node_info(sbi, ino, &old_ni);
1777
1778 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1779 return -EINVAL;
1780
1781 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1782 if (!ipage)
1783 return -ENOMEM;
1784
1785 /* Should not use this inode from free nid list */
1786 remove_free_nid(NM_I(sbi), ino);
1787
1788 SetPageUptodate(ipage);
1789 fill_node_footer(ipage, ino, ino, 0, true);
1790
1791 src = F2FS_INODE(page);
1792 dst = F2FS_INODE(ipage);
1793
1794 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1795 dst->i_size = 0;
1796 dst->i_blocks = cpu_to_le64(1);
1797 dst->i_links = cpu_to_le32(1);
1798 dst->i_xattr_nid = 0;
1799 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1800
1801 new_ni = old_ni;
1802 new_ni.ino = ino;
1803
1804 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1805 WARN_ON(1);
1806 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1807 inc_valid_inode_count(sbi);
1808 set_page_dirty(ipage);
1809 f2fs_put_page(ipage, 1);
1810 return 0;
1811 }
1812
1813 int restore_node_summary(struct f2fs_sb_info *sbi,
1814 unsigned int segno, struct f2fs_summary_block *sum)
1815 {
1816 struct f2fs_node *rn;
1817 struct f2fs_summary *sum_entry;
1818 block_t addr;
1819 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1820 int i, idx, last_offset, nrpages;
1821
1822 /* scan the node segment */
1823 last_offset = sbi->blocks_per_seg;
1824 addr = START_BLOCK(sbi, segno);
1825 sum_entry = &sum->entries[0];
1826
1827 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1828 nrpages = min(last_offset - i, bio_blocks);
1829
1830 /* readahead node pages */
1831 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1832
1833 for (idx = addr; idx < addr + nrpages; idx++) {
1834 struct page *page = get_tmp_page(sbi, idx);
1835
1836 rn = F2FS_NODE(page);
1837 sum_entry->nid = rn->footer.nid;
1838 sum_entry->version = 0;
1839 sum_entry->ofs_in_node = 0;
1840 sum_entry++;
1841 f2fs_put_page(page, 1);
1842 }
1843
1844 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1845 addr + nrpages);
1846 }
1847 return 0;
1848 }
1849
1850 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1851 {
1852 struct f2fs_nm_info *nm_i = NM_I(sbi);
1853 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1854 struct f2fs_summary_block *sum = curseg->sum_blk;
1855 int i;
1856
1857 mutex_lock(&curseg->curseg_mutex);
1858 for (i = 0; i < nats_in_cursum(sum); i++) {
1859 struct nat_entry *ne;
1860 struct f2fs_nat_entry raw_ne;
1861 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1862
1863 raw_ne = nat_in_journal(sum, i);
1864
1865 ne = __lookup_nat_cache(nm_i, nid);
1866 if (!ne) {
1867 ne = grab_nat_entry(nm_i, nid);
1868 node_info_from_raw_nat(&ne->ni, &raw_ne);
1869 }
1870 __set_nat_cache_dirty(nm_i, ne);
1871 }
1872 update_nats_in_cursum(sum, -i);
1873 mutex_unlock(&curseg->curseg_mutex);
1874 }
1875
1876 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1877 struct list_head *head, int max)
1878 {
1879 struct nat_entry_set *cur;
1880
1881 if (nes->entry_cnt >= max)
1882 goto add_out;
1883
1884 list_for_each_entry(cur, head, set_list) {
1885 if (cur->entry_cnt >= nes->entry_cnt) {
1886 list_add(&nes->set_list, cur->set_list.prev);
1887 return;
1888 }
1889 }
1890 add_out:
1891 list_add_tail(&nes->set_list, head);
1892 }
1893
1894 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1895 struct nat_entry_set *set)
1896 {
1897 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1898 struct f2fs_summary_block *sum = curseg->sum_blk;
1899 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1900 bool to_journal = true;
1901 struct f2fs_nat_block *nat_blk;
1902 struct nat_entry *ne, *cur;
1903 struct page *page = NULL;
1904
1905 /*
1906 * there are two steps to flush nat entries:
1907 * #1, flush nat entries to journal in current hot data summary block.
1908 * #2, flush nat entries to nat page.
1909 */
1910 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1911 to_journal = false;
1912
1913 if (to_journal) {
1914 mutex_lock(&curseg->curseg_mutex);
1915 } else {
1916 page = get_next_nat_page(sbi, start_nid);
1917 nat_blk = page_address(page);
1918 f2fs_bug_on(sbi, !nat_blk);
1919 }
1920
1921 /* flush dirty nats in nat entry set */
1922 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1923 struct f2fs_nat_entry *raw_ne;
1924 nid_t nid = nat_get_nid(ne);
1925 int offset;
1926
1927 if (nat_get_blkaddr(ne) == NEW_ADDR)
1928 continue;
1929
1930 if (to_journal) {
1931 offset = lookup_journal_in_cursum(sum,
1932 NAT_JOURNAL, nid, 1);
1933 f2fs_bug_on(sbi, offset < 0);
1934 raw_ne = &nat_in_journal(sum, offset);
1935 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1936 } else {
1937 raw_ne = &nat_blk->entries[nid - start_nid];
1938 }
1939 raw_nat_from_node_info(raw_ne, &ne->ni);
1940 nat_reset_flag(ne);
1941 __clear_nat_cache_dirty(NM_I(sbi), ne);
1942 if (nat_get_blkaddr(ne) == NULL_ADDR)
1943 add_free_nid(sbi, nid, false);
1944 }
1945
1946 if (to_journal)
1947 mutex_unlock(&curseg->curseg_mutex);
1948 else
1949 f2fs_put_page(page, 1);
1950
1951 f2fs_bug_on(sbi, set->entry_cnt);
1952
1953 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1954 kmem_cache_free(nat_entry_set_slab, set);
1955 }
1956
1957 /*
1958 * This function is called during the checkpointing process.
1959 */
1960 void flush_nat_entries(struct f2fs_sb_info *sbi)
1961 {
1962 struct f2fs_nm_info *nm_i = NM_I(sbi);
1963 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1964 struct f2fs_summary_block *sum = curseg->sum_blk;
1965 struct nat_entry_set *setvec[SETVEC_SIZE];
1966 struct nat_entry_set *set, *tmp;
1967 unsigned int found;
1968 nid_t set_idx = 0;
1969 LIST_HEAD(sets);
1970
1971 if (!nm_i->dirty_nat_cnt)
1972 return;
1973
1974 down_write(&nm_i->nat_tree_lock);
1975
1976 /*
1977 * if there are no enough space in journal to store dirty nat
1978 * entries, remove all entries from journal and merge them
1979 * into nat entry set.
1980 */
1981 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1982 remove_nats_in_journal(sbi);
1983
1984 while ((found = __gang_lookup_nat_set(nm_i,
1985 set_idx, SETVEC_SIZE, setvec))) {
1986 unsigned idx;
1987 set_idx = setvec[found - 1]->set + 1;
1988 for (idx = 0; idx < found; idx++)
1989 __adjust_nat_entry_set(setvec[idx], &sets,
1990 MAX_NAT_JENTRIES(sum));
1991 }
1992
1993 /* flush dirty nats in nat entry set */
1994 list_for_each_entry_safe(set, tmp, &sets, set_list)
1995 __flush_nat_entry_set(sbi, set);
1996
1997 up_write(&nm_i->nat_tree_lock);
1998
1999 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2000 }
2001
2002 static int init_node_manager(struct f2fs_sb_info *sbi)
2003 {
2004 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2005 struct f2fs_nm_info *nm_i = NM_I(sbi);
2006 unsigned char *version_bitmap;
2007 unsigned int nat_segs, nat_blocks;
2008
2009 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2010
2011 /* segment_count_nat includes pair segment so divide to 2. */
2012 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2013 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2014
2015 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2016
2017 /* not used nids: 0, node, meta, (and root counted as valid node) */
2018 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2019 nm_i->fcnt = 0;
2020 nm_i->nat_cnt = 0;
2021 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2022 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2023
2024 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2025 INIT_LIST_HEAD(&nm_i->free_nid_list);
2026 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2027 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2028 INIT_LIST_HEAD(&nm_i->nat_entries);
2029
2030 mutex_init(&nm_i->build_lock);
2031 spin_lock_init(&nm_i->free_nid_list_lock);
2032 init_rwsem(&nm_i->nat_tree_lock);
2033
2034 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2035 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2036 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2037 if (!version_bitmap)
2038 return -EFAULT;
2039
2040 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2041 GFP_KERNEL);
2042 if (!nm_i->nat_bitmap)
2043 return -ENOMEM;
2044 return 0;
2045 }
2046
2047 int build_node_manager(struct f2fs_sb_info *sbi)
2048 {
2049 int err;
2050
2051 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2052 if (!sbi->nm_info)
2053 return -ENOMEM;
2054
2055 err = init_node_manager(sbi);
2056 if (err)
2057 return err;
2058
2059 build_free_nids(sbi);
2060 return 0;
2061 }
2062
2063 void destroy_node_manager(struct f2fs_sb_info *sbi)
2064 {
2065 struct f2fs_nm_info *nm_i = NM_I(sbi);
2066 struct free_nid *i, *next_i;
2067 struct nat_entry *natvec[NATVEC_SIZE];
2068 struct nat_entry_set *setvec[SETVEC_SIZE];
2069 nid_t nid = 0;
2070 unsigned int found;
2071
2072 if (!nm_i)
2073 return;
2074
2075 /* destroy free nid list */
2076 spin_lock(&nm_i->free_nid_list_lock);
2077 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2078 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2079 __del_from_free_nid_list(nm_i, i);
2080 nm_i->fcnt--;
2081 spin_unlock(&nm_i->free_nid_list_lock);
2082 kmem_cache_free(free_nid_slab, i);
2083 spin_lock(&nm_i->free_nid_list_lock);
2084 }
2085 f2fs_bug_on(sbi, nm_i->fcnt);
2086 spin_unlock(&nm_i->free_nid_list_lock);
2087
2088 /* destroy nat cache */
2089 down_write(&nm_i->nat_tree_lock);
2090 while ((found = __gang_lookup_nat_cache(nm_i,
2091 nid, NATVEC_SIZE, natvec))) {
2092 unsigned idx;
2093
2094 nid = nat_get_nid(natvec[found - 1]) + 1;
2095 for (idx = 0; idx < found; idx++)
2096 __del_from_nat_cache(nm_i, natvec[idx]);
2097 }
2098 f2fs_bug_on(sbi, nm_i->nat_cnt);
2099
2100 /* destroy nat set cache */
2101 nid = 0;
2102 while ((found = __gang_lookup_nat_set(nm_i,
2103 nid, SETVEC_SIZE, setvec))) {
2104 unsigned idx;
2105
2106 nid = setvec[found - 1]->set + 1;
2107 for (idx = 0; idx < found; idx++) {
2108 /* entry_cnt is not zero, when cp_error was occurred */
2109 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2110 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2111 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2112 }
2113 }
2114 up_write(&nm_i->nat_tree_lock);
2115
2116 kfree(nm_i->nat_bitmap);
2117 sbi->nm_info = NULL;
2118 kfree(nm_i);
2119 }
2120
2121 int __init create_node_manager_caches(void)
2122 {
2123 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2124 sizeof(struct nat_entry));
2125 if (!nat_entry_slab)
2126 goto fail;
2127
2128 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2129 sizeof(struct free_nid));
2130 if (!free_nid_slab)
2131 goto destroy_nat_entry;
2132
2133 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2134 sizeof(struct nat_entry_set));
2135 if (!nat_entry_set_slab)
2136 goto destroy_free_nid;
2137 return 0;
2138
2139 destroy_free_nid:
2140 kmem_cache_destroy(free_nid_slab);
2141 destroy_nat_entry:
2142 kmem_cache_destroy(nat_entry_slab);
2143 fail:
2144 return -ENOMEM;
2145 }
2146
2147 void destroy_node_manager_caches(void)
2148 {
2149 kmem_cache_destroy(nat_entry_set_slab);
2150 kmem_cache_destroy(free_nid_slab);
2151 kmem_cache_destroy(nat_entry_slab);
2152 }
ubuntu-zesty-kernel mirror