Loading...
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 "xattr.h"
23#include "trace.h"
24#include <trace/events/f2fs.h>
25
26#define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
27
28static struct kmem_cache *nat_entry_slab;
29static struct kmem_cache *free_nid_slab;
30static struct kmem_cache *nat_entry_set_slab;
31
32bool available_free_memory(struct f2fs_sb_info *sbi, int type)
33{
34 struct f2fs_nm_info *nm_i = NM_I(sbi);
35 struct sysinfo val;
36 unsigned long avail_ram;
37 unsigned long mem_size = 0;
38 bool res = false;
39
40 si_meminfo(&val);
41
42 /* only uses low memory */
43 avail_ram = val.totalram - val.totalhigh;
44
45 /*
46 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
47 */
48 if (type == FREE_NIDS) {
49 mem_size = (nm_i->nid_cnt[FREE_NID] *
50 sizeof(struct free_nid)) >> PAGE_SHIFT;
51 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
52 } else if (type == NAT_ENTRIES) {
53 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
54 PAGE_SHIFT;
55 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
56 if (excess_cached_nats(sbi))
57 res = false;
58 } else if (type == DIRTY_DENTS) {
59 if (sbi->sb->s_bdi->wb.dirty_exceeded)
60 return false;
61 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
62 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
63 } else if (type == INO_ENTRIES) {
64 int i;
65
66 for (i = 0; i < MAX_INO_ENTRY; i++)
67 mem_size += sbi->im[i].ino_num *
68 sizeof(struct ino_entry);
69 mem_size >>= PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
71 } else if (type == EXTENT_CACHE) {
72 mem_size = (atomic_read(&sbi->total_ext_tree) *
73 sizeof(struct extent_tree) +
74 atomic_read(&sbi->total_ext_node) *
75 sizeof(struct extent_node)) >> PAGE_SHIFT;
76 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
77 } else if (type == INMEM_PAGES) {
78 /* it allows 20% / total_ram for inmemory pages */
79 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
80 res = mem_size < (val.totalram / 5);
81 } else {
82 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
83 return true;
84 }
85 return res;
86}
87
88static void clear_node_page_dirty(struct page *page)
89{
90 struct address_space *mapping = page->mapping;
91 unsigned int long flags;
92
93 if (PageDirty(page)) {
94 xa_lock_irqsave(&mapping->i_pages, flags);
95 radix_tree_tag_clear(&mapping->i_pages,
96 page_index(page),
97 PAGECACHE_TAG_DIRTY);
98 xa_unlock_irqrestore(&mapping->i_pages, flags);
99
100 clear_page_dirty_for_io(page);
101 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
102 }
103 ClearPageUptodate(page);
104}
105
106static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
107{
108 pgoff_t index = current_nat_addr(sbi, nid);
109 return get_meta_page(sbi, index);
110}
111
112static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
113{
114 struct page *src_page;
115 struct page *dst_page;
116 pgoff_t src_off;
117 pgoff_t dst_off;
118 void *src_addr;
119 void *dst_addr;
120 struct f2fs_nm_info *nm_i = NM_I(sbi);
121
122 src_off = current_nat_addr(sbi, nid);
123 dst_off = next_nat_addr(sbi, src_off);
124
125 /* get current nat block page with lock */
126 src_page = get_meta_page(sbi, src_off);
127 dst_page = grab_meta_page(sbi, dst_off);
128 f2fs_bug_on(sbi, PageDirty(src_page));
129
130 src_addr = page_address(src_page);
131 dst_addr = page_address(dst_page);
132 memcpy(dst_addr, src_addr, PAGE_SIZE);
133 set_page_dirty(dst_page);
134 f2fs_put_page(src_page, 1);
135
136 set_to_next_nat(nm_i, nid);
137
138 return dst_page;
139}
140
141static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
142{
143 struct nat_entry *new;
144
145 if (no_fail)
146 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
147 else
148 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
149 if (new) {
150 nat_set_nid(new, nid);
151 nat_reset_flag(new);
152 }
153 return new;
154}
155
156static void __free_nat_entry(struct nat_entry *e)
157{
158 kmem_cache_free(nat_entry_slab, e);
159}
160
161/* must be locked by nat_tree_lock */
162static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
163 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
164{
165 if (no_fail)
166 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
167 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
168 return NULL;
169
170 if (raw_ne)
171 node_info_from_raw_nat(&ne->ni, raw_ne);
172 list_add_tail(&ne->list, &nm_i->nat_entries);
173 nm_i->nat_cnt++;
174 return ne;
175}
176
177static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
178{
179 return radix_tree_lookup(&nm_i->nat_root, n);
180}
181
182static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
183 nid_t start, unsigned int nr, struct nat_entry **ep)
184{
185 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
186}
187
188static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
189{
190 list_del(&e->list);
191 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
192 nm_i->nat_cnt--;
193 __free_nat_entry(e);
194}
195
196static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
197 struct nat_entry *ne)
198{
199 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
200 struct nat_entry_set *head;
201
202 head = radix_tree_lookup(&nm_i->nat_set_root, set);
203 if (!head) {
204 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
205
206 INIT_LIST_HEAD(&head->entry_list);
207 INIT_LIST_HEAD(&head->set_list);
208 head->set = set;
209 head->entry_cnt = 0;
210 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
211 }
212 return head;
213}
214
215static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
216 struct nat_entry *ne)
217{
218 struct nat_entry_set *head;
219 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
220
221 if (!new_ne)
222 head = __grab_nat_entry_set(nm_i, ne);
223
224 /*
225 * update entry_cnt in below condition:
226 * 1. update NEW_ADDR to valid block address;
227 * 2. update old block address to new one;
228 */
229 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
230 !get_nat_flag(ne, IS_DIRTY)))
231 head->entry_cnt++;
232
233 set_nat_flag(ne, IS_PREALLOC, new_ne);
234
235 if (get_nat_flag(ne, IS_DIRTY))
236 goto refresh_list;
237
238 nm_i->dirty_nat_cnt++;
239 set_nat_flag(ne, IS_DIRTY, true);
240refresh_list:
241 if (new_ne)
242 list_del_init(&ne->list);
243 else
244 list_move_tail(&ne->list, &head->entry_list);
245}
246
247static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
248 struct nat_entry_set *set, struct nat_entry *ne)
249{
250 list_move_tail(&ne->list, &nm_i->nat_entries);
251 set_nat_flag(ne, IS_DIRTY, false);
252 set->entry_cnt--;
253 nm_i->dirty_nat_cnt--;
254}
255
256static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
257 nid_t start, unsigned int nr, struct nat_entry_set **ep)
258{
259 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
260 start, nr);
261}
262
263int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
264{
265 struct f2fs_nm_info *nm_i = NM_I(sbi);
266 struct nat_entry *e;
267 bool need = false;
268
269 down_read(&nm_i->nat_tree_lock);
270 e = __lookup_nat_cache(nm_i, nid);
271 if (e) {
272 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
273 !get_nat_flag(e, HAS_FSYNCED_INODE))
274 need = true;
275 }
276 up_read(&nm_i->nat_tree_lock);
277 return need;
278}
279
280bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
281{
282 struct f2fs_nm_info *nm_i = NM_I(sbi);
283 struct nat_entry *e;
284 bool is_cp = true;
285
286 down_read(&nm_i->nat_tree_lock);
287 e = __lookup_nat_cache(nm_i, nid);
288 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
289 is_cp = false;
290 up_read(&nm_i->nat_tree_lock);
291 return is_cp;
292}
293
294bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
295{
296 struct f2fs_nm_info *nm_i = NM_I(sbi);
297 struct nat_entry *e;
298 bool need_update = true;
299
300 down_read(&nm_i->nat_tree_lock);
301 e = __lookup_nat_cache(nm_i, ino);
302 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
303 (get_nat_flag(e, IS_CHECKPOINTED) ||
304 get_nat_flag(e, HAS_FSYNCED_INODE)))
305 need_update = false;
306 up_read(&nm_i->nat_tree_lock);
307 return need_update;
308}
309
310/* must be locked by nat_tree_lock */
311static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
312 struct f2fs_nat_entry *ne)
313{
314 struct f2fs_nm_info *nm_i = NM_I(sbi);
315 struct nat_entry *new, *e;
316
317 new = __alloc_nat_entry(nid, false);
318 if (!new)
319 return;
320
321 down_write(&nm_i->nat_tree_lock);
322 e = __lookup_nat_cache(nm_i, nid);
323 if (!e)
324 e = __init_nat_entry(nm_i, new, ne, false);
325 else
326 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
327 nat_get_blkaddr(e) !=
328 le32_to_cpu(ne->block_addr) ||
329 nat_get_version(e) != ne->version);
330 up_write(&nm_i->nat_tree_lock);
331 if (e != new)
332 __free_nat_entry(new);
333}
334
335static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
336 block_t new_blkaddr, bool fsync_done)
337{
338 struct f2fs_nm_info *nm_i = NM_I(sbi);
339 struct nat_entry *e;
340 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
341
342 down_write(&nm_i->nat_tree_lock);
343 e = __lookup_nat_cache(nm_i, ni->nid);
344 if (!e) {
345 e = __init_nat_entry(nm_i, new, NULL, true);
346 copy_node_info(&e->ni, ni);
347 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
348 } else if (new_blkaddr == NEW_ADDR) {
349 /*
350 * when nid is reallocated,
351 * previous nat entry can be remained in nat cache.
352 * So, reinitialize it with new information.
353 */
354 copy_node_info(&e->ni, ni);
355 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
356 }
357 /* let's free early to reduce memory consumption */
358 if (e != new)
359 __free_nat_entry(new);
360
361 /* sanity check */
362 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
363 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
364 new_blkaddr == NULL_ADDR);
365 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
366 new_blkaddr == NEW_ADDR);
367 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
368 nat_get_blkaddr(e) != NULL_ADDR &&
369 new_blkaddr == NEW_ADDR);
370
371 /* increment version no as node is removed */
372 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
373 unsigned char version = nat_get_version(e);
374 nat_set_version(e, inc_node_version(version));
375 }
376
377 /* change address */
378 nat_set_blkaddr(e, new_blkaddr);
379 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
380 set_nat_flag(e, IS_CHECKPOINTED, false);
381 __set_nat_cache_dirty(nm_i, e);
382
383 /* update fsync_mark if its inode nat entry is still alive */
384 if (ni->nid != ni->ino)
385 e = __lookup_nat_cache(nm_i, ni->ino);
386 if (e) {
387 if (fsync_done && ni->nid == ni->ino)
388 set_nat_flag(e, HAS_FSYNCED_INODE, true);
389 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
390 }
391 up_write(&nm_i->nat_tree_lock);
392}
393
394int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
395{
396 struct f2fs_nm_info *nm_i = NM_I(sbi);
397 int nr = nr_shrink;
398
399 if (!down_write_trylock(&nm_i->nat_tree_lock))
400 return 0;
401
402 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
403 struct nat_entry *ne;
404 ne = list_first_entry(&nm_i->nat_entries,
405 struct nat_entry, list);
406 __del_from_nat_cache(nm_i, ne);
407 nr_shrink--;
408 }
409 up_write(&nm_i->nat_tree_lock);
410 return nr - nr_shrink;
411}
412
413/*
414 * This function always returns success
415 */
416void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
417{
418 struct f2fs_nm_info *nm_i = NM_I(sbi);
419 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
420 struct f2fs_journal *journal = curseg->journal;
421 nid_t start_nid = START_NID(nid);
422 struct f2fs_nat_block *nat_blk;
423 struct page *page = NULL;
424 struct f2fs_nat_entry ne;
425 struct nat_entry *e;
426 pgoff_t index;
427 int i;
428
429 ni->nid = nid;
430
431 /* Check nat cache */
432 down_read(&nm_i->nat_tree_lock);
433 e = __lookup_nat_cache(nm_i, nid);
434 if (e) {
435 ni->ino = nat_get_ino(e);
436 ni->blk_addr = nat_get_blkaddr(e);
437 ni->version = nat_get_version(e);
438 up_read(&nm_i->nat_tree_lock);
439 return;
440 }
441
442 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
443
444 /* Check current segment summary */
445 down_read(&curseg->journal_rwsem);
446 i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
447 if (i >= 0) {
448 ne = nat_in_journal(journal, i);
449 node_info_from_raw_nat(ni, &ne);
450 }
451 up_read(&curseg->journal_rwsem);
452 if (i >= 0) {
453 up_read(&nm_i->nat_tree_lock);
454 goto cache;
455 }
456
457 /* Fill node_info from nat page */
458 index = current_nat_addr(sbi, nid);
459 up_read(&nm_i->nat_tree_lock);
460
461 page = get_meta_page(sbi, index);
462 nat_blk = (struct f2fs_nat_block *)page_address(page);
463 ne = nat_blk->entries[nid - start_nid];
464 node_info_from_raw_nat(ni, &ne);
465 f2fs_put_page(page, 1);
466cache:
467 /* cache nat entry */
468 cache_nat_entry(sbi, nid, &ne);
469}
470
471/*
472 * readahead MAX_RA_NODE number of node pages.
473 */
474static void ra_node_pages(struct page *parent, int start, int n)
475{
476 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
477 struct blk_plug plug;
478 int i, end;
479 nid_t nid;
480
481 blk_start_plug(&plug);
482
483 /* Then, try readahead for siblings of the desired node */
484 end = start + n;
485 end = min(end, NIDS_PER_BLOCK);
486 for (i = start; i < end; i++) {
487 nid = get_nid(parent, i, false);
488 ra_node_page(sbi, nid);
489 }
490
491 blk_finish_plug(&plug);
492}
493
494pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
495{
496 const long direct_index = ADDRS_PER_INODE(dn->inode);
497 const long direct_blks = ADDRS_PER_BLOCK;
498 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
499 unsigned int skipped_unit = ADDRS_PER_BLOCK;
500 int cur_level = dn->cur_level;
501 int max_level = dn->max_level;
502 pgoff_t base = 0;
503
504 if (!dn->max_level)
505 return pgofs + 1;
506
507 while (max_level-- > cur_level)
508 skipped_unit *= NIDS_PER_BLOCK;
509
510 switch (dn->max_level) {
511 case 3:
512 base += 2 * indirect_blks;
513 case 2:
514 base += 2 * direct_blks;
515 case 1:
516 base += direct_index;
517 break;
518 default:
519 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
520 }
521
522 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
523}
524
525/*
526 * The maximum depth is four.
527 * Offset[0] will have raw inode offset.
528 */
529static int get_node_path(struct inode *inode, long block,
530 int offset[4], unsigned int noffset[4])
531{
532 const long direct_index = ADDRS_PER_INODE(inode);
533 const long direct_blks = ADDRS_PER_BLOCK;
534 const long dptrs_per_blk = NIDS_PER_BLOCK;
535 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
536 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
537 int n = 0;
538 int level = 0;
539
540 noffset[0] = 0;
541
542 if (block < direct_index) {
543 offset[n] = block;
544 goto got;
545 }
546 block -= direct_index;
547 if (block < direct_blks) {
548 offset[n++] = NODE_DIR1_BLOCK;
549 noffset[n] = 1;
550 offset[n] = block;
551 level = 1;
552 goto got;
553 }
554 block -= direct_blks;
555 if (block < direct_blks) {
556 offset[n++] = NODE_DIR2_BLOCK;
557 noffset[n] = 2;
558 offset[n] = block;
559 level = 1;
560 goto got;
561 }
562 block -= direct_blks;
563 if (block < indirect_blks) {
564 offset[n++] = NODE_IND1_BLOCK;
565 noffset[n] = 3;
566 offset[n++] = block / direct_blks;
567 noffset[n] = 4 + offset[n - 1];
568 offset[n] = block % direct_blks;
569 level = 2;
570 goto got;
571 }
572 block -= indirect_blks;
573 if (block < indirect_blks) {
574 offset[n++] = NODE_IND2_BLOCK;
575 noffset[n] = 4 + dptrs_per_blk;
576 offset[n++] = block / direct_blks;
577 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
578 offset[n] = block % direct_blks;
579 level = 2;
580 goto got;
581 }
582 block -= indirect_blks;
583 if (block < dindirect_blks) {
584 offset[n++] = NODE_DIND_BLOCK;
585 noffset[n] = 5 + (dptrs_per_blk * 2);
586 offset[n++] = block / indirect_blks;
587 noffset[n] = 6 + (dptrs_per_blk * 2) +
588 offset[n - 1] * (dptrs_per_blk + 1);
589 offset[n++] = (block / direct_blks) % dptrs_per_blk;
590 noffset[n] = 7 + (dptrs_per_blk * 2) +
591 offset[n - 2] * (dptrs_per_blk + 1) +
592 offset[n - 1];
593 offset[n] = block % direct_blks;
594 level = 3;
595 goto got;
596 } else {
597 return -E2BIG;
598 }
599got:
600 return level;
601}
602
603/*
604 * Caller should call f2fs_put_dnode(dn).
605 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
606 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
607 * In the case of RDONLY_NODE, we don't need to care about mutex.
608 */
609int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
610{
611 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
612 struct page *npage[4];
613 struct page *parent = NULL;
614 int offset[4];
615 unsigned int noffset[4];
616 nid_t nids[4];
617 int level, i = 0;
618 int err = 0;
619
620 level = get_node_path(dn->inode, index, offset, noffset);
621 if (level < 0)
622 return level;
623
624 nids[0] = dn->inode->i_ino;
625 npage[0] = dn->inode_page;
626
627 if (!npage[0]) {
628 npage[0] = get_node_page(sbi, nids[0]);
629 if (IS_ERR(npage[0]))
630 return PTR_ERR(npage[0]);
631 }
632
633 /* if inline_data is set, should not report any block indices */
634 if (f2fs_has_inline_data(dn->inode) && index) {
635 err = -ENOENT;
636 f2fs_put_page(npage[0], 1);
637 goto release_out;
638 }
639
640 parent = npage[0];
641 if (level != 0)
642 nids[1] = get_nid(parent, offset[0], true);
643 dn->inode_page = npage[0];
644 dn->inode_page_locked = true;
645
646 /* get indirect or direct nodes */
647 for (i = 1; i <= level; i++) {
648 bool done = false;
649
650 if (!nids[i] && mode == ALLOC_NODE) {
651 /* alloc new node */
652 if (!alloc_nid(sbi, &(nids[i]))) {
653 err = -ENOSPC;
654 goto release_pages;
655 }
656
657 dn->nid = nids[i];
658 npage[i] = new_node_page(dn, noffset[i]);
659 if (IS_ERR(npage[i])) {
660 alloc_nid_failed(sbi, nids[i]);
661 err = PTR_ERR(npage[i]);
662 goto release_pages;
663 }
664
665 set_nid(parent, offset[i - 1], nids[i], i == 1);
666 alloc_nid_done(sbi, nids[i]);
667 done = true;
668 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
669 npage[i] = get_node_page_ra(parent, offset[i - 1]);
670 if (IS_ERR(npage[i])) {
671 err = PTR_ERR(npage[i]);
672 goto release_pages;
673 }
674 done = true;
675 }
676 if (i == 1) {
677 dn->inode_page_locked = false;
678 unlock_page(parent);
679 } else {
680 f2fs_put_page(parent, 1);
681 }
682
683 if (!done) {
684 npage[i] = get_node_page(sbi, nids[i]);
685 if (IS_ERR(npage[i])) {
686 err = PTR_ERR(npage[i]);
687 f2fs_put_page(npage[0], 0);
688 goto release_out;
689 }
690 }
691 if (i < level) {
692 parent = npage[i];
693 nids[i + 1] = get_nid(parent, offset[i], false);
694 }
695 }
696 dn->nid = nids[level];
697 dn->ofs_in_node = offset[level];
698 dn->node_page = npage[level];
699 dn->data_blkaddr = datablock_addr(dn->inode,
700 dn->node_page, dn->ofs_in_node);
701 return 0;
702
703release_pages:
704 f2fs_put_page(parent, 1);
705 if (i > 1)
706 f2fs_put_page(npage[0], 0);
707release_out:
708 dn->inode_page = NULL;
709 dn->node_page = NULL;
710 if (err == -ENOENT) {
711 dn->cur_level = i;
712 dn->max_level = level;
713 dn->ofs_in_node = offset[level];
714 }
715 return err;
716}
717
718static void truncate_node(struct dnode_of_data *dn)
719{
720 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
721 struct node_info ni;
722
723 get_node_info(sbi, dn->nid, &ni);
724
725 /* Deallocate node address */
726 invalidate_blocks(sbi, ni.blk_addr);
727 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
728 set_node_addr(sbi, &ni, NULL_ADDR, false);
729
730 if (dn->nid == dn->inode->i_ino) {
731 remove_orphan_inode(sbi, dn->nid);
732 dec_valid_inode_count(sbi);
733 f2fs_inode_synced(dn->inode);
734 }
735
736 clear_node_page_dirty(dn->node_page);
737 set_sbi_flag(sbi, SBI_IS_DIRTY);
738
739 f2fs_put_page(dn->node_page, 1);
740
741 invalidate_mapping_pages(NODE_MAPPING(sbi),
742 dn->node_page->index, dn->node_page->index);
743
744 dn->node_page = NULL;
745 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
746}
747
748static int truncate_dnode(struct dnode_of_data *dn)
749{
750 struct page *page;
751
752 if (dn->nid == 0)
753 return 1;
754
755 /* get direct node */
756 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
757 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
758 return 1;
759 else if (IS_ERR(page))
760 return PTR_ERR(page);
761
762 /* Make dnode_of_data for parameter */
763 dn->node_page = page;
764 dn->ofs_in_node = 0;
765 truncate_data_blocks(dn);
766 truncate_node(dn);
767 return 1;
768}
769
770static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
771 int ofs, int depth)
772{
773 struct dnode_of_data rdn = *dn;
774 struct page *page;
775 struct f2fs_node *rn;
776 nid_t child_nid;
777 unsigned int child_nofs;
778 int freed = 0;
779 int i, ret;
780
781 if (dn->nid == 0)
782 return NIDS_PER_BLOCK + 1;
783
784 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
785
786 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
787 if (IS_ERR(page)) {
788 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
789 return PTR_ERR(page);
790 }
791
792 ra_node_pages(page, ofs, NIDS_PER_BLOCK);
793
794 rn = F2FS_NODE(page);
795 if (depth < 3) {
796 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
797 child_nid = le32_to_cpu(rn->in.nid[i]);
798 if (child_nid == 0)
799 continue;
800 rdn.nid = child_nid;
801 ret = truncate_dnode(&rdn);
802 if (ret < 0)
803 goto out_err;
804 if (set_nid(page, i, 0, false))
805 dn->node_changed = true;
806 }
807 } else {
808 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
809 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
810 child_nid = le32_to_cpu(rn->in.nid[i]);
811 if (child_nid == 0) {
812 child_nofs += NIDS_PER_BLOCK + 1;
813 continue;
814 }
815 rdn.nid = child_nid;
816 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
817 if (ret == (NIDS_PER_BLOCK + 1)) {
818 if (set_nid(page, i, 0, false))
819 dn->node_changed = true;
820 child_nofs += ret;
821 } else if (ret < 0 && ret != -ENOENT) {
822 goto out_err;
823 }
824 }
825 freed = child_nofs;
826 }
827
828 if (!ofs) {
829 /* remove current indirect node */
830 dn->node_page = page;
831 truncate_node(dn);
832 freed++;
833 } else {
834 f2fs_put_page(page, 1);
835 }
836 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
837 return freed;
838
839out_err:
840 f2fs_put_page(page, 1);
841 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
842 return ret;
843}
844
845static int truncate_partial_nodes(struct dnode_of_data *dn,
846 struct f2fs_inode *ri, int *offset, int depth)
847{
848 struct page *pages[2];
849 nid_t nid[3];
850 nid_t child_nid;
851 int err = 0;
852 int i;
853 int idx = depth - 2;
854
855 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
856 if (!nid[0])
857 return 0;
858
859 /* get indirect nodes in the path */
860 for (i = 0; i < idx + 1; i++) {
861 /* reference count'll be increased */
862 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
863 if (IS_ERR(pages[i])) {
864 err = PTR_ERR(pages[i]);
865 idx = i - 1;
866 goto fail;
867 }
868 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
869 }
870
871 ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
872
873 /* free direct nodes linked to a partial indirect node */
874 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
875 child_nid = get_nid(pages[idx], i, false);
876 if (!child_nid)
877 continue;
878 dn->nid = child_nid;
879 err = truncate_dnode(dn);
880 if (err < 0)
881 goto fail;
882 if (set_nid(pages[idx], i, 0, false))
883 dn->node_changed = true;
884 }
885
886 if (offset[idx + 1] == 0) {
887 dn->node_page = pages[idx];
888 dn->nid = nid[idx];
889 truncate_node(dn);
890 } else {
891 f2fs_put_page(pages[idx], 1);
892 }
893 offset[idx]++;
894 offset[idx + 1] = 0;
895 idx--;
896fail:
897 for (i = idx; i >= 0; i--)
898 f2fs_put_page(pages[i], 1);
899
900 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
901
902 return err;
903}
904
905/*
906 * All the block addresses of data and nodes should be nullified.
907 */
908int truncate_inode_blocks(struct inode *inode, pgoff_t from)
909{
910 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
911 int err = 0, cont = 1;
912 int level, offset[4], noffset[4];
913 unsigned int nofs = 0;
914 struct f2fs_inode *ri;
915 struct dnode_of_data dn;
916 struct page *page;
917
918 trace_f2fs_truncate_inode_blocks_enter(inode, from);
919
920 level = get_node_path(inode, from, offset, noffset);
921 if (level < 0)
922 return level;
923
924 page = get_node_page(sbi, inode->i_ino);
925 if (IS_ERR(page)) {
926 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
927 return PTR_ERR(page);
928 }
929
930 set_new_dnode(&dn, inode, page, NULL, 0);
931 unlock_page(page);
932
933 ri = F2FS_INODE(page);
934 switch (level) {
935 case 0:
936 case 1:
937 nofs = noffset[1];
938 break;
939 case 2:
940 nofs = noffset[1];
941 if (!offset[level - 1])
942 goto skip_partial;
943 err = truncate_partial_nodes(&dn, ri, offset, level);
944 if (err < 0 && err != -ENOENT)
945 goto fail;
946 nofs += 1 + NIDS_PER_BLOCK;
947 break;
948 case 3:
949 nofs = 5 + 2 * NIDS_PER_BLOCK;
950 if (!offset[level - 1])
951 goto skip_partial;
952 err = truncate_partial_nodes(&dn, ri, offset, level);
953 if (err < 0 && err != -ENOENT)
954 goto fail;
955 break;
956 default:
957 BUG();
958 }
959
960skip_partial:
961 while (cont) {
962 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
963 switch (offset[0]) {
964 case NODE_DIR1_BLOCK:
965 case NODE_DIR2_BLOCK:
966 err = truncate_dnode(&dn);
967 break;
968
969 case NODE_IND1_BLOCK:
970 case NODE_IND2_BLOCK:
971 err = truncate_nodes(&dn, nofs, offset[1], 2);
972 break;
973
974 case NODE_DIND_BLOCK:
975 err = truncate_nodes(&dn, nofs, offset[1], 3);
976 cont = 0;
977 break;
978
979 default:
980 BUG();
981 }
982 if (err < 0 && err != -ENOENT)
983 goto fail;
984 if (offset[1] == 0 &&
985 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
986 lock_page(page);
987 BUG_ON(page->mapping != NODE_MAPPING(sbi));
988 f2fs_wait_on_page_writeback(page, NODE, true);
989 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
990 set_page_dirty(page);
991 unlock_page(page);
992 }
993 offset[1] = 0;
994 offset[0]++;
995 nofs += err;
996 }
997fail:
998 f2fs_put_page(page, 0);
999 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1000 return err > 0 ? 0 : err;
1001}
1002
1003/* caller must lock inode page */
1004int truncate_xattr_node(struct inode *inode)
1005{
1006 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1007 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1008 struct dnode_of_data dn;
1009 struct page *npage;
1010
1011 if (!nid)
1012 return 0;
1013
1014 npage = get_node_page(sbi, nid);
1015 if (IS_ERR(npage))
1016 return PTR_ERR(npage);
1017
1018 f2fs_i_xnid_write(inode, 0);
1019
1020 set_new_dnode(&dn, inode, NULL, npage, nid);
1021 truncate_node(&dn);
1022 return 0;
1023}
1024
1025/*
1026 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1027 * f2fs_unlock_op().
1028 */
1029int remove_inode_page(struct inode *inode)
1030{
1031 struct dnode_of_data dn;
1032 int err;
1033
1034 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1035 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1036 if (err)
1037 return err;
1038
1039 err = truncate_xattr_node(inode);
1040 if (err) {
1041 f2fs_put_dnode(&dn);
1042 return err;
1043 }
1044
1045 /* remove potential inline_data blocks */
1046 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1047 S_ISLNK(inode->i_mode))
1048 truncate_data_blocks_range(&dn, 1);
1049
1050 /* 0 is possible, after f2fs_new_inode() has failed */
1051 f2fs_bug_on(F2FS_I_SB(inode),
1052 inode->i_blocks != 0 && inode->i_blocks != 8);
1053
1054 /* will put inode & node pages */
1055 truncate_node(&dn);
1056 return 0;
1057}
1058
1059struct page *new_inode_page(struct inode *inode)
1060{
1061 struct dnode_of_data dn;
1062
1063 /* allocate inode page for new inode */
1064 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1065
1066 /* caller should f2fs_put_page(page, 1); */
1067 return new_node_page(&dn, 0);
1068}
1069
1070struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1071{
1072 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1073 struct node_info new_ni;
1074 struct page *page;
1075 int err;
1076
1077 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1078 return ERR_PTR(-EPERM);
1079
1080 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1081 if (!page)
1082 return ERR_PTR(-ENOMEM);
1083
1084 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1085 goto fail;
1086
1087#ifdef CONFIG_F2FS_CHECK_FS
1088 get_node_info(sbi, dn->nid, &new_ni);
1089 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1090#endif
1091 new_ni.nid = dn->nid;
1092 new_ni.ino = dn->inode->i_ino;
1093 new_ni.blk_addr = NULL_ADDR;
1094 new_ni.flag = 0;
1095 new_ni.version = 0;
1096 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1097
1098 f2fs_wait_on_page_writeback(page, NODE, true);
1099 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1100 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1101 if (!PageUptodate(page))
1102 SetPageUptodate(page);
1103 if (set_page_dirty(page))
1104 dn->node_changed = true;
1105
1106 if (f2fs_has_xattr_block(ofs))
1107 f2fs_i_xnid_write(dn->inode, dn->nid);
1108
1109 if (ofs == 0)
1110 inc_valid_inode_count(sbi);
1111 return page;
1112
1113fail:
1114 clear_node_page_dirty(page);
1115 f2fs_put_page(page, 1);
1116 return ERR_PTR(err);
1117}
1118
1119/*
1120 * Caller should do after getting the following values.
1121 * 0: f2fs_put_page(page, 0)
1122 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1123 */
1124static int read_node_page(struct page *page, int op_flags)
1125{
1126 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1127 struct node_info ni;
1128 struct f2fs_io_info fio = {
1129 .sbi = sbi,
1130 .type = NODE,
1131 .op = REQ_OP_READ,
1132 .op_flags = op_flags,
1133 .page = page,
1134 .encrypted_page = NULL,
1135 };
1136
1137 if (PageUptodate(page))
1138 return LOCKED_PAGE;
1139
1140 get_node_info(sbi, page->index, &ni);
1141
1142 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1143 ClearPageUptodate(page);
1144 return -ENOENT;
1145 }
1146
1147 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1148 return f2fs_submit_page_bio(&fio);
1149}
1150
1151/*
1152 * Readahead a node page
1153 */
1154void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1155{
1156 struct page *apage;
1157 int err;
1158
1159 if (!nid)
1160 return;
1161 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1162
1163 rcu_read_lock();
1164 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->i_pages, nid);
1165 rcu_read_unlock();
1166 if (apage)
1167 return;
1168
1169 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1170 if (!apage)
1171 return;
1172
1173 err = read_node_page(apage, REQ_RAHEAD);
1174 f2fs_put_page(apage, err ? 1 : 0);
1175}
1176
1177static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1178 struct page *parent, int start)
1179{
1180 struct page *page;
1181 int err;
1182
1183 if (!nid)
1184 return ERR_PTR(-ENOENT);
1185 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1186repeat:
1187 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1188 if (!page)
1189 return ERR_PTR(-ENOMEM);
1190
1191 err = read_node_page(page, 0);
1192 if (err < 0) {
1193 f2fs_put_page(page, 1);
1194 return ERR_PTR(err);
1195 } else if (err == LOCKED_PAGE) {
1196 err = 0;
1197 goto page_hit;
1198 }
1199
1200 if (parent)
1201 ra_node_pages(parent, start + 1, MAX_RA_NODE);
1202
1203 lock_page(page);
1204
1205 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1206 f2fs_put_page(page, 1);
1207 goto repeat;
1208 }
1209
1210 if (unlikely(!PageUptodate(page))) {
1211 err = -EIO;
1212 goto out_err;
1213 }
1214
1215 if (!f2fs_inode_chksum_verify(sbi, page)) {
1216 err = -EBADMSG;
1217 goto out_err;
1218 }
1219page_hit:
1220 if(unlikely(nid != nid_of_node(page))) {
1221 f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1222 "nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1223 nid, nid_of_node(page), ino_of_node(page),
1224 ofs_of_node(page), cpver_of_node(page),
1225 next_blkaddr_of_node(page));
1226 err = -EINVAL;
1227out_err:
1228 ClearPageUptodate(page);
1229 f2fs_put_page(page, 1);
1230 return ERR_PTR(err);
1231 }
1232 return page;
1233}
1234
1235struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1236{
1237 return __get_node_page(sbi, nid, NULL, 0);
1238}
1239
1240struct page *get_node_page_ra(struct page *parent, int start)
1241{
1242 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1243 nid_t nid = get_nid(parent, start, false);
1244
1245 return __get_node_page(sbi, nid, parent, start);
1246}
1247
1248static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1249{
1250 struct inode *inode;
1251 struct page *page;
1252 int ret;
1253
1254 /* should flush inline_data before evict_inode */
1255 inode = ilookup(sbi->sb, ino);
1256 if (!inode)
1257 return;
1258
1259 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1260 FGP_LOCK|FGP_NOWAIT, 0);
1261 if (!page)
1262 goto iput_out;
1263
1264 if (!PageUptodate(page))
1265 goto page_out;
1266
1267 if (!PageDirty(page))
1268 goto page_out;
1269
1270 if (!clear_page_dirty_for_io(page))
1271 goto page_out;
1272
1273 ret = f2fs_write_inline_data(inode, page);
1274 inode_dec_dirty_pages(inode);
1275 remove_dirty_inode(inode);
1276 if (ret)
1277 set_page_dirty(page);
1278page_out:
1279 f2fs_put_page(page, 1);
1280iput_out:
1281 iput(inode);
1282}
1283
1284static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1285{
1286 pgoff_t index;
1287 struct pagevec pvec;
1288 struct page *last_page = NULL;
1289 int nr_pages;
1290
1291 pagevec_init(&pvec);
1292 index = 0;
1293
1294 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1295 PAGECACHE_TAG_DIRTY))) {
1296 int i;
1297
1298 for (i = 0; i < nr_pages; i++) {
1299 struct page *page = pvec.pages[i];
1300
1301 if (unlikely(f2fs_cp_error(sbi))) {
1302 f2fs_put_page(last_page, 0);
1303 pagevec_release(&pvec);
1304 return ERR_PTR(-EIO);
1305 }
1306
1307 if (!IS_DNODE(page) || !is_cold_node(page))
1308 continue;
1309 if (ino_of_node(page) != ino)
1310 continue;
1311
1312 lock_page(page);
1313
1314 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1315continue_unlock:
1316 unlock_page(page);
1317 continue;
1318 }
1319 if (ino_of_node(page) != ino)
1320 goto continue_unlock;
1321
1322 if (!PageDirty(page)) {
1323 /* someone wrote it for us */
1324 goto continue_unlock;
1325 }
1326
1327 if (last_page)
1328 f2fs_put_page(last_page, 0);
1329
1330 get_page(page);
1331 last_page = page;
1332 unlock_page(page);
1333 }
1334 pagevec_release(&pvec);
1335 cond_resched();
1336 }
1337 return last_page;
1338}
1339
1340static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1341 struct writeback_control *wbc, bool do_balance,
1342 enum iostat_type io_type)
1343{
1344 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1345 nid_t nid;
1346 struct node_info ni;
1347 struct f2fs_io_info fio = {
1348 .sbi = sbi,
1349 .ino = ino_of_node(page),
1350 .type = NODE,
1351 .op = REQ_OP_WRITE,
1352 .op_flags = wbc_to_write_flags(wbc),
1353 .page = page,
1354 .encrypted_page = NULL,
1355 .submitted = false,
1356 .io_type = io_type,
1357 .io_wbc = wbc,
1358 };
1359
1360 trace_f2fs_writepage(page, NODE);
1361
1362 if (unlikely(f2fs_cp_error(sbi))) {
1363 dec_page_count(sbi, F2FS_DIRTY_NODES);
1364 unlock_page(page);
1365 return 0;
1366 }
1367
1368 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1369 goto redirty_out;
1370
1371 /* get old block addr of this node page */
1372 nid = nid_of_node(page);
1373 f2fs_bug_on(sbi, page->index != nid);
1374
1375 if (wbc->for_reclaim) {
1376 if (!down_read_trylock(&sbi->node_write))
1377 goto redirty_out;
1378 } else {
1379 down_read(&sbi->node_write);
1380 }
1381
1382 get_node_info(sbi, nid, &ni);
1383
1384 /* This page is already truncated */
1385 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1386 ClearPageUptodate(page);
1387 dec_page_count(sbi, F2FS_DIRTY_NODES);
1388 up_read(&sbi->node_write);
1389 unlock_page(page);
1390 return 0;
1391 }
1392
1393 if (atomic && !test_opt(sbi, NOBARRIER))
1394 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1395
1396 set_page_writeback(page);
1397 fio.old_blkaddr = ni.blk_addr;
1398 write_node_page(nid, &fio);
1399 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1400 dec_page_count(sbi, F2FS_DIRTY_NODES);
1401 up_read(&sbi->node_write);
1402
1403 if (wbc->for_reclaim) {
1404 f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1405 page->index, NODE);
1406 submitted = NULL;
1407 }
1408
1409 unlock_page(page);
1410
1411 if (unlikely(f2fs_cp_error(sbi))) {
1412 f2fs_submit_merged_write(sbi, NODE);
1413 submitted = NULL;
1414 }
1415 if (submitted)
1416 *submitted = fio.submitted;
1417
1418 if (do_balance)
1419 f2fs_balance_fs(sbi, false);
1420 return 0;
1421
1422redirty_out:
1423 redirty_page_for_writepage(wbc, page);
1424 return AOP_WRITEPAGE_ACTIVATE;
1425}
1426
1427void move_node_page(struct page *node_page, int gc_type)
1428{
1429 if (gc_type == FG_GC) {
1430 struct writeback_control wbc = {
1431 .sync_mode = WB_SYNC_ALL,
1432 .nr_to_write = 1,
1433 .for_reclaim = 0,
1434 };
1435
1436 set_page_dirty(node_page);
1437 f2fs_wait_on_page_writeback(node_page, NODE, true);
1438
1439 f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1440 if (!clear_page_dirty_for_io(node_page))
1441 goto out_page;
1442
1443 if (__write_node_page(node_page, false, NULL,
1444 &wbc, false, FS_GC_NODE_IO))
1445 unlock_page(node_page);
1446 goto release_page;
1447 } else {
1448 /* set page dirty and write it */
1449 if (!PageWriteback(node_page))
1450 set_page_dirty(node_page);
1451 }
1452out_page:
1453 unlock_page(node_page);
1454release_page:
1455 f2fs_put_page(node_page, 0);
1456}
1457
1458static int f2fs_write_node_page(struct page *page,
1459 struct writeback_control *wbc)
1460{
1461 return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1462}
1463
1464int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1465 struct writeback_control *wbc, bool atomic)
1466{
1467 pgoff_t index;
1468 pgoff_t last_idx = ULONG_MAX;
1469 struct pagevec pvec;
1470 int ret = 0;
1471 struct page *last_page = NULL;
1472 bool marked = false;
1473 nid_t ino = inode->i_ino;
1474 int nr_pages;
1475
1476 if (atomic) {
1477 last_page = last_fsync_dnode(sbi, ino);
1478 if (IS_ERR_OR_NULL(last_page))
1479 return PTR_ERR_OR_ZERO(last_page);
1480 }
1481retry:
1482 pagevec_init(&pvec);
1483 index = 0;
1484
1485 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1486 PAGECACHE_TAG_DIRTY))) {
1487 int i;
1488
1489 for (i = 0; i < nr_pages; i++) {
1490 struct page *page = pvec.pages[i];
1491 bool submitted = false;
1492
1493 if (unlikely(f2fs_cp_error(sbi))) {
1494 f2fs_put_page(last_page, 0);
1495 pagevec_release(&pvec);
1496 ret = -EIO;
1497 goto out;
1498 }
1499
1500 if (!IS_DNODE(page) || !is_cold_node(page))
1501 continue;
1502 if (ino_of_node(page) != ino)
1503 continue;
1504
1505 lock_page(page);
1506
1507 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1508continue_unlock:
1509 unlock_page(page);
1510 continue;
1511 }
1512 if (ino_of_node(page) != ino)
1513 goto continue_unlock;
1514
1515 if (!PageDirty(page) && page != last_page) {
1516 /* someone wrote it for us */
1517 goto continue_unlock;
1518 }
1519
1520 f2fs_wait_on_page_writeback(page, NODE, true);
1521 BUG_ON(PageWriteback(page));
1522
1523 set_fsync_mark(page, 0);
1524 set_dentry_mark(page, 0);
1525
1526 if (!atomic || page == last_page) {
1527 set_fsync_mark(page, 1);
1528 if (IS_INODE(page)) {
1529 if (is_inode_flag_set(inode,
1530 FI_DIRTY_INODE))
1531 update_inode(inode, page);
1532 set_dentry_mark(page,
1533 need_dentry_mark(sbi, ino));
1534 }
1535 /* may be written by other thread */
1536 if (!PageDirty(page))
1537 set_page_dirty(page);
1538 }
1539
1540 if (!clear_page_dirty_for_io(page))
1541 goto continue_unlock;
1542
1543 ret = __write_node_page(page, atomic &&
1544 page == last_page,
1545 &submitted, wbc, true,
1546 FS_NODE_IO);
1547 if (ret) {
1548 unlock_page(page);
1549 f2fs_put_page(last_page, 0);
1550 break;
1551 } else if (submitted) {
1552 last_idx = page->index;
1553 }
1554
1555 if (page == last_page) {
1556 f2fs_put_page(page, 0);
1557 marked = true;
1558 break;
1559 }
1560 }
1561 pagevec_release(&pvec);
1562 cond_resched();
1563
1564 if (ret || marked)
1565 break;
1566 }
1567 if (!ret && atomic && !marked) {
1568 f2fs_msg(sbi->sb, KERN_DEBUG,
1569 "Retry to write fsync mark: ino=%u, idx=%lx",
1570 ino, last_page->index);
1571 lock_page(last_page);
1572 f2fs_wait_on_page_writeback(last_page, NODE, true);
1573 set_page_dirty(last_page);
1574 unlock_page(last_page);
1575 goto retry;
1576 }
1577out:
1578 if (last_idx != ULONG_MAX)
1579 f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1580 return ret ? -EIO: 0;
1581}
1582
1583int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1584 bool do_balance, enum iostat_type io_type)
1585{
1586 pgoff_t index;
1587 struct pagevec pvec;
1588 int step = 0;
1589 int nwritten = 0;
1590 int ret = 0;
1591 int nr_pages;
1592
1593 pagevec_init(&pvec);
1594
1595next_step:
1596 index = 0;
1597
1598 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1599 PAGECACHE_TAG_DIRTY))) {
1600 int i;
1601
1602 for (i = 0; i < nr_pages; i++) {
1603 struct page *page = pvec.pages[i];
1604 bool submitted = false;
1605
1606 /*
1607 * flushing sequence with step:
1608 * 0. indirect nodes
1609 * 1. dentry dnodes
1610 * 2. file dnodes
1611 */
1612 if (step == 0 && IS_DNODE(page))
1613 continue;
1614 if (step == 1 && (!IS_DNODE(page) ||
1615 is_cold_node(page)))
1616 continue;
1617 if (step == 2 && (!IS_DNODE(page) ||
1618 !is_cold_node(page)))
1619 continue;
1620lock_node:
1621 if (!trylock_page(page))
1622 continue;
1623
1624 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1625continue_unlock:
1626 unlock_page(page);
1627 continue;
1628 }
1629
1630 if (!PageDirty(page)) {
1631 /* someone wrote it for us */
1632 goto continue_unlock;
1633 }
1634
1635 /* flush inline_data */
1636 if (is_inline_node(page)) {
1637 clear_inline_node(page);
1638 unlock_page(page);
1639 flush_inline_data(sbi, ino_of_node(page));
1640 goto lock_node;
1641 }
1642
1643 f2fs_wait_on_page_writeback(page, NODE, true);
1644
1645 BUG_ON(PageWriteback(page));
1646 if (!clear_page_dirty_for_io(page))
1647 goto continue_unlock;
1648
1649 set_fsync_mark(page, 0);
1650 set_dentry_mark(page, 0);
1651
1652 ret = __write_node_page(page, false, &submitted,
1653 wbc, do_balance, io_type);
1654 if (ret)
1655 unlock_page(page);
1656 else if (submitted)
1657 nwritten++;
1658
1659 if (--wbc->nr_to_write == 0)
1660 break;
1661 }
1662 pagevec_release(&pvec);
1663 cond_resched();
1664
1665 if (wbc->nr_to_write == 0) {
1666 step = 2;
1667 break;
1668 }
1669 }
1670
1671 if (step < 2) {
1672 step++;
1673 goto next_step;
1674 }
1675
1676 if (nwritten)
1677 f2fs_submit_merged_write(sbi, NODE);
1678
1679 if (unlikely(f2fs_cp_error(sbi)))
1680 return -EIO;
1681 return ret;
1682}
1683
1684int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1685{
1686 pgoff_t index = 0;
1687 struct pagevec pvec;
1688 int ret2, ret = 0;
1689 int nr_pages;
1690
1691 pagevec_init(&pvec);
1692
1693 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1694 PAGECACHE_TAG_WRITEBACK))) {
1695 int i;
1696
1697 for (i = 0; i < nr_pages; i++) {
1698 struct page *page = pvec.pages[i];
1699
1700 if (ino && ino_of_node(page) == ino) {
1701 f2fs_wait_on_page_writeback(page, NODE, true);
1702 if (TestClearPageError(page))
1703 ret = -EIO;
1704 }
1705 }
1706 pagevec_release(&pvec);
1707 cond_resched();
1708 }
1709
1710 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1711 if (!ret)
1712 ret = ret2;
1713 return ret;
1714}
1715
1716static int f2fs_write_node_pages(struct address_space *mapping,
1717 struct writeback_control *wbc)
1718{
1719 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1720 struct blk_plug plug;
1721 long diff;
1722
1723 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1724 goto skip_write;
1725
1726 /* balancing f2fs's metadata in background */
1727 f2fs_balance_fs_bg(sbi);
1728
1729 /* collect a number of dirty node pages and write together */
1730 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1731 goto skip_write;
1732
1733 trace_f2fs_writepages(mapping->host, wbc, NODE);
1734
1735 diff = nr_pages_to_write(sbi, NODE, wbc);
1736 wbc->sync_mode = WB_SYNC_NONE;
1737 blk_start_plug(&plug);
1738 sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1739 blk_finish_plug(&plug);
1740 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1741 return 0;
1742
1743skip_write:
1744 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1745 trace_f2fs_writepages(mapping->host, wbc, NODE);
1746 return 0;
1747}
1748
1749static int f2fs_set_node_page_dirty(struct page *page)
1750{
1751 trace_f2fs_set_page_dirty(page, NODE);
1752
1753 if (!PageUptodate(page))
1754 SetPageUptodate(page);
1755 if (!PageDirty(page)) {
1756 f2fs_set_page_dirty_nobuffers(page);
1757 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1758 SetPagePrivate(page);
1759 f2fs_trace_pid(page);
1760 return 1;
1761 }
1762 return 0;
1763}
1764
1765/*
1766 * Structure of the f2fs node operations
1767 */
1768const struct address_space_operations f2fs_node_aops = {
1769 .writepage = f2fs_write_node_page,
1770 .writepages = f2fs_write_node_pages,
1771 .set_page_dirty = f2fs_set_node_page_dirty,
1772 .invalidatepage = f2fs_invalidate_page,
1773 .releasepage = f2fs_release_page,
1774#ifdef CONFIG_MIGRATION
1775 .migratepage = f2fs_migrate_page,
1776#endif
1777};
1778
1779static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1780 nid_t n)
1781{
1782 return radix_tree_lookup(&nm_i->free_nid_root, n);
1783}
1784
1785static int __insert_free_nid(struct f2fs_sb_info *sbi,
1786 struct free_nid *i, enum nid_state state)
1787{
1788 struct f2fs_nm_info *nm_i = NM_I(sbi);
1789
1790 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1791 if (err)
1792 return err;
1793
1794 f2fs_bug_on(sbi, state != i->state);
1795 nm_i->nid_cnt[state]++;
1796 if (state == FREE_NID)
1797 list_add_tail(&i->list, &nm_i->free_nid_list);
1798 return 0;
1799}
1800
1801static void __remove_free_nid(struct f2fs_sb_info *sbi,
1802 struct free_nid *i, enum nid_state state)
1803{
1804 struct f2fs_nm_info *nm_i = NM_I(sbi);
1805
1806 f2fs_bug_on(sbi, state != i->state);
1807 nm_i->nid_cnt[state]--;
1808 if (state == FREE_NID)
1809 list_del(&i->list);
1810 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1811}
1812
1813static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
1814 enum nid_state org_state, enum nid_state dst_state)
1815{
1816 struct f2fs_nm_info *nm_i = NM_I(sbi);
1817
1818 f2fs_bug_on(sbi, org_state != i->state);
1819 i->state = dst_state;
1820 nm_i->nid_cnt[org_state]--;
1821 nm_i->nid_cnt[dst_state]++;
1822
1823 switch (dst_state) {
1824 case PREALLOC_NID:
1825 list_del(&i->list);
1826 break;
1827 case FREE_NID:
1828 list_add_tail(&i->list, &nm_i->free_nid_list);
1829 break;
1830 default:
1831 BUG_ON(1);
1832 }
1833}
1834
1835static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1836 bool set, bool build)
1837{
1838 struct f2fs_nm_info *nm_i = NM_I(sbi);
1839 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1840 unsigned int nid_ofs = nid - START_NID(nid);
1841
1842 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1843 return;
1844
1845 if (set) {
1846 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1847 return;
1848 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1849 nm_i->free_nid_count[nat_ofs]++;
1850 } else {
1851 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1852 return;
1853 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1854 if (!build)
1855 nm_i->free_nid_count[nat_ofs]--;
1856 }
1857}
1858
1859/* return if the nid is recognized as free */
1860static bool add_free_nid(struct f2fs_sb_info *sbi,
1861 nid_t nid, bool build, bool update)
1862{
1863 struct f2fs_nm_info *nm_i = NM_I(sbi);
1864 struct free_nid *i, *e;
1865 struct nat_entry *ne;
1866 int err = -EINVAL;
1867 bool ret = false;
1868
1869 /* 0 nid should not be used */
1870 if (unlikely(nid == 0))
1871 return false;
1872
1873 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1874 i->nid = nid;
1875 i->state = FREE_NID;
1876
1877 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
1878
1879 spin_lock(&nm_i->nid_list_lock);
1880
1881 if (build) {
1882 /*
1883 * Thread A Thread B
1884 * - f2fs_create
1885 * - f2fs_new_inode
1886 * - alloc_nid
1887 * - __insert_nid_to_list(PREALLOC_NID)
1888 * - f2fs_balance_fs_bg
1889 * - build_free_nids
1890 * - __build_free_nids
1891 * - scan_nat_page
1892 * - add_free_nid
1893 * - __lookup_nat_cache
1894 * - f2fs_add_link
1895 * - init_inode_metadata
1896 * - new_inode_page
1897 * - new_node_page
1898 * - set_node_addr
1899 * - alloc_nid_done
1900 * - __remove_nid_from_list(PREALLOC_NID)
1901 * - __insert_nid_to_list(FREE_NID)
1902 */
1903 ne = __lookup_nat_cache(nm_i, nid);
1904 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1905 nat_get_blkaddr(ne) != NULL_ADDR))
1906 goto err_out;
1907
1908 e = __lookup_free_nid_list(nm_i, nid);
1909 if (e) {
1910 if (e->state == FREE_NID)
1911 ret = true;
1912 goto err_out;
1913 }
1914 }
1915 ret = true;
1916 err = __insert_free_nid(sbi, i, FREE_NID);
1917err_out:
1918 if (update) {
1919 update_free_nid_bitmap(sbi, nid, ret, build);
1920 if (!build)
1921 nm_i->available_nids++;
1922 }
1923 spin_unlock(&nm_i->nid_list_lock);
1924 radix_tree_preload_end();
1925
1926 if (err)
1927 kmem_cache_free(free_nid_slab, i);
1928 return ret;
1929}
1930
1931static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1932{
1933 struct f2fs_nm_info *nm_i = NM_I(sbi);
1934 struct free_nid *i;
1935 bool need_free = false;
1936
1937 spin_lock(&nm_i->nid_list_lock);
1938 i = __lookup_free_nid_list(nm_i, nid);
1939 if (i && i->state == FREE_NID) {
1940 __remove_free_nid(sbi, i, FREE_NID);
1941 need_free = true;
1942 }
1943 spin_unlock(&nm_i->nid_list_lock);
1944
1945 if (need_free)
1946 kmem_cache_free(free_nid_slab, i);
1947}
1948
1949static void scan_nat_page(struct f2fs_sb_info *sbi,
1950 struct page *nat_page, nid_t start_nid)
1951{
1952 struct f2fs_nm_info *nm_i = NM_I(sbi);
1953 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1954 block_t blk_addr;
1955 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1956 int i;
1957
1958 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1959
1960 i = start_nid % NAT_ENTRY_PER_BLOCK;
1961
1962 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1963 if (unlikely(start_nid >= nm_i->max_nid))
1964 break;
1965
1966 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1967 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1968 if (blk_addr == NULL_ADDR) {
1969 add_free_nid(sbi, start_nid, true, true);
1970 } else {
1971 spin_lock(&NM_I(sbi)->nid_list_lock);
1972 update_free_nid_bitmap(sbi, start_nid, false, true);
1973 spin_unlock(&NM_I(sbi)->nid_list_lock);
1974 }
1975 }
1976}
1977
1978static void scan_curseg_cache(struct f2fs_sb_info *sbi)
1979{
1980 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1981 struct f2fs_journal *journal = curseg->journal;
1982 int i;
1983
1984 down_read(&curseg->journal_rwsem);
1985 for (i = 0; i < nats_in_cursum(journal); i++) {
1986 block_t addr;
1987 nid_t nid;
1988
1989 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1990 nid = le32_to_cpu(nid_in_journal(journal, i));
1991 if (addr == NULL_ADDR)
1992 add_free_nid(sbi, nid, true, false);
1993 else
1994 remove_free_nid(sbi, nid);
1995 }
1996 up_read(&curseg->journal_rwsem);
1997}
1998
1999static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2000{
2001 struct f2fs_nm_info *nm_i = NM_I(sbi);
2002 unsigned int i, idx;
2003 nid_t nid;
2004
2005 down_read(&nm_i->nat_tree_lock);
2006
2007 for (i = 0; i < nm_i->nat_blocks; i++) {
2008 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2009 continue;
2010 if (!nm_i->free_nid_count[i])
2011 continue;
2012 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2013 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2014 NAT_ENTRY_PER_BLOCK, idx);
2015 if (idx >= NAT_ENTRY_PER_BLOCK)
2016 break;
2017
2018 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2019 add_free_nid(sbi, nid, true, false);
2020
2021 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2022 goto out;
2023 }
2024 }
2025out:
2026 scan_curseg_cache(sbi);
2027
2028 up_read(&nm_i->nat_tree_lock);
2029}
2030
2031static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2032{
2033 struct f2fs_nm_info *nm_i = NM_I(sbi);
2034 int i = 0;
2035 nid_t nid = nm_i->next_scan_nid;
2036
2037 if (unlikely(nid >= nm_i->max_nid))
2038 nid = 0;
2039
2040 /* Enough entries */
2041 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2042 return;
2043
2044 if (!sync && !available_free_memory(sbi, FREE_NIDS))
2045 return;
2046
2047 if (!mount) {
2048 /* try to find free nids in free_nid_bitmap */
2049 scan_free_nid_bits(sbi);
2050
2051 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2052 return;
2053 }
2054
2055 /* readahead nat pages to be scanned */
2056 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2057 META_NAT, true);
2058
2059 down_read(&nm_i->nat_tree_lock);
2060
2061 while (1) {
2062 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2063 nm_i->nat_block_bitmap)) {
2064 struct page *page = get_current_nat_page(sbi, nid);
2065
2066 scan_nat_page(sbi, page, nid);
2067 f2fs_put_page(page, 1);
2068 }
2069
2070 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2071 if (unlikely(nid >= nm_i->max_nid))
2072 nid = 0;
2073
2074 if (++i >= FREE_NID_PAGES)
2075 break;
2076 }
2077
2078 /* go to the next free nat pages to find free nids abundantly */
2079 nm_i->next_scan_nid = nid;
2080
2081 /* find free nids from current sum_pages */
2082 scan_curseg_cache(sbi);
2083
2084 up_read(&nm_i->nat_tree_lock);
2085
2086 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2087 nm_i->ra_nid_pages, META_NAT, false);
2088}
2089
2090void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2091{
2092 mutex_lock(&NM_I(sbi)->build_lock);
2093 __build_free_nids(sbi, sync, mount);
2094 mutex_unlock(&NM_I(sbi)->build_lock);
2095}
2096
2097/*
2098 * If this function returns success, caller can obtain a new nid
2099 * from second parameter of this function.
2100 * The returned nid could be used ino as well as nid when inode is created.
2101 */
2102bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2103{
2104 struct f2fs_nm_info *nm_i = NM_I(sbi);
2105 struct free_nid *i = NULL;
2106retry:
2107#ifdef CONFIG_F2FS_FAULT_INJECTION
2108 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2109 f2fs_show_injection_info(FAULT_ALLOC_NID);
2110 return false;
2111 }
2112#endif
2113 spin_lock(&nm_i->nid_list_lock);
2114
2115 if (unlikely(nm_i->available_nids == 0)) {
2116 spin_unlock(&nm_i->nid_list_lock);
2117 return false;
2118 }
2119
2120 /* We should not use stale free nids created by build_free_nids */
2121 if (nm_i->nid_cnt[FREE_NID] && !on_build_free_nids(nm_i)) {
2122 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2123 i = list_first_entry(&nm_i->free_nid_list,
2124 struct free_nid, list);
2125 *nid = i->nid;
2126
2127 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2128 nm_i->available_nids--;
2129
2130 update_free_nid_bitmap(sbi, *nid, false, false);
2131
2132 spin_unlock(&nm_i->nid_list_lock);
2133 return true;
2134 }
2135 spin_unlock(&nm_i->nid_list_lock);
2136
2137 /* Let's scan nat pages and its caches to get free nids */
2138 build_free_nids(sbi, true, false);
2139 goto retry;
2140}
2141
2142/*
2143 * alloc_nid() should be called prior to this function.
2144 */
2145void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2146{
2147 struct f2fs_nm_info *nm_i = NM_I(sbi);
2148 struct free_nid *i;
2149
2150 spin_lock(&nm_i->nid_list_lock);
2151 i = __lookup_free_nid_list(nm_i, nid);
2152 f2fs_bug_on(sbi, !i);
2153 __remove_free_nid(sbi, i, PREALLOC_NID);
2154 spin_unlock(&nm_i->nid_list_lock);
2155
2156 kmem_cache_free(free_nid_slab, i);
2157}
2158
2159/*
2160 * alloc_nid() should be called prior to this function.
2161 */
2162void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2163{
2164 struct f2fs_nm_info *nm_i = NM_I(sbi);
2165 struct free_nid *i;
2166 bool need_free = false;
2167
2168 if (!nid)
2169 return;
2170
2171 spin_lock(&nm_i->nid_list_lock);
2172 i = __lookup_free_nid_list(nm_i, nid);
2173 f2fs_bug_on(sbi, !i);
2174
2175 if (!available_free_memory(sbi, FREE_NIDS)) {
2176 __remove_free_nid(sbi, i, PREALLOC_NID);
2177 need_free = true;
2178 } else {
2179 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2180 }
2181
2182 nm_i->available_nids++;
2183
2184 update_free_nid_bitmap(sbi, nid, true, false);
2185
2186 spin_unlock(&nm_i->nid_list_lock);
2187
2188 if (need_free)
2189 kmem_cache_free(free_nid_slab, i);
2190}
2191
2192int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2193{
2194 struct f2fs_nm_info *nm_i = NM_I(sbi);
2195 struct free_nid *i, *next;
2196 int nr = nr_shrink;
2197
2198 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2199 return 0;
2200
2201 if (!mutex_trylock(&nm_i->build_lock))
2202 return 0;
2203
2204 spin_lock(&nm_i->nid_list_lock);
2205 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2206 if (nr_shrink <= 0 ||
2207 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2208 break;
2209
2210 __remove_free_nid(sbi, i, FREE_NID);
2211 kmem_cache_free(free_nid_slab, i);
2212 nr_shrink--;
2213 }
2214 spin_unlock(&nm_i->nid_list_lock);
2215 mutex_unlock(&nm_i->build_lock);
2216
2217 return nr - nr_shrink;
2218}
2219
2220void recover_inline_xattr(struct inode *inode, struct page *page)
2221{
2222 void *src_addr, *dst_addr;
2223 size_t inline_size;
2224 struct page *ipage;
2225 struct f2fs_inode *ri;
2226
2227 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2228 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2229
2230 ri = F2FS_INODE(page);
2231 if (ri->i_inline & F2FS_INLINE_XATTR) {
2232 set_inode_flag(inode, FI_INLINE_XATTR);
2233 } else {
2234 clear_inode_flag(inode, FI_INLINE_XATTR);
2235 goto update_inode;
2236 }
2237
2238 dst_addr = inline_xattr_addr(inode, ipage);
2239 src_addr = inline_xattr_addr(inode, page);
2240 inline_size = inline_xattr_size(inode);
2241
2242 f2fs_wait_on_page_writeback(ipage, NODE, true);
2243 memcpy(dst_addr, src_addr, inline_size);
2244update_inode:
2245 update_inode(inode, ipage);
2246 f2fs_put_page(ipage, 1);
2247}
2248
2249int recover_xattr_data(struct inode *inode, struct page *page)
2250{
2251 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2252 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2253 nid_t new_xnid;
2254 struct dnode_of_data dn;
2255 struct node_info ni;
2256 struct page *xpage;
2257
2258 if (!prev_xnid)
2259 goto recover_xnid;
2260
2261 /* 1: invalidate the previous xattr nid */
2262 get_node_info(sbi, prev_xnid, &ni);
2263 invalidate_blocks(sbi, ni.blk_addr);
2264 dec_valid_node_count(sbi, inode, false);
2265 set_node_addr(sbi, &ni, NULL_ADDR, false);
2266
2267recover_xnid:
2268 /* 2: update xattr nid in inode */
2269 if (!alloc_nid(sbi, &new_xnid))
2270 return -ENOSPC;
2271
2272 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2273 xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2274 if (IS_ERR(xpage)) {
2275 alloc_nid_failed(sbi, new_xnid);
2276 return PTR_ERR(xpage);
2277 }
2278
2279 alloc_nid_done(sbi, new_xnid);
2280 update_inode_page(inode);
2281
2282 /* 3: update and set xattr node page dirty */
2283 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2284
2285 set_page_dirty(xpage);
2286 f2fs_put_page(xpage, 1);
2287
2288 return 0;
2289}
2290
2291int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2292{
2293 struct f2fs_inode *src, *dst;
2294 nid_t ino = ino_of_node(page);
2295 struct node_info old_ni, new_ni;
2296 struct page *ipage;
2297
2298 get_node_info(sbi, ino, &old_ni);
2299
2300 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2301 return -EINVAL;
2302retry:
2303 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2304 if (!ipage) {
2305 congestion_wait(BLK_RW_ASYNC, HZ/50);
2306 goto retry;
2307 }
2308
2309 /* Should not use this inode from free nid list */
2310 remove_free_nid(sbi, ino);
2311
2312 if (!PageUptodate(ipage))
2313 SetPageUptodate(ipage);
2314 fill_node_footer(ipage, ino, ino, 0, true);
2315 set_cold_node(page, false);
2316
2317 src = F2FS_INODE(page);
2318 dst = F2FS_INODE(ipage);
2319
2320 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2321 dst->i_size = 0;
2322 dst->i_blocks = cpu_to_le64(1);
2323 dst->i_links = cpu_to_le32(1);
2324 dst->i_xattr_nid = 0;
2325 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2326 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2327 dst->i_extra_isize = src->i_extra_isize;
2328
2329 if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2330 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2331 i_inline_xattr_size))
2332 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2333
2334 if (f2fs_sb_has_project_quota(sbi->sb) &&
2335 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2336 i_projid))
2337 dst->i_projid = src->i_projid;
2338 }
2339
2340 new_ni = old_ni;
2341 new_ni.ino = ino;
2342
2343 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2344 WARN_ON(1);
2345 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2346 inc_valid_inode_count(sbi);
2347 set_page_dirty(ipage);
2348 f2fs_put_page(ipage, 1);
2349 return 0;
2350}
2351
2352void restore_node_summary(struct f2fs_sb_info *sbi,
2353 unsigned int segno, struct f2fs_summary_block *sum)
2354{
2355 struct f2fs_node *rn;
2356 struct f2fs_summary *sum_entry;
2357 block_t addr;
2358 int i, idx, last_offset, nrpages;
2359
2360 /* scan the node segment */
2361 last_offset = sbi->blocks_per_seg;
2362 addr = START_BLOCK(sbi, segno);
2363 sum_entry = &sum->entries[0];
2364
2365 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2366 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2367
2368 /* readahead node pages */
2369 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2370
2371 for (idx = addr; idx < addr + nrpages; idx++) {
2372 struct page *page = get_tmp_page(sbi, idx);
2373
2374 rn = F2FS_NODE(page);
2375 sum_entry->nid = rn->footer.nid;
2376 sum_entry->version = 0;
2377 sum_entry->ofs_in_node = 0;
2378 sum_entry++;
2379 f2fs_put_page(page, 1);
2380 }
2381
2382 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2383 addr + nrpages);
2384 }
2385}
2386
2387static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2388{
2389 struct f2fs_nm_info *nm_i = NM_I(sbi);
2390 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2391 struct f2fs_journal *journal = curseg->journal;
2392 int i;
2393
2394 down_write(&curseg->journal_rwsem);
2395 for (i = 0; i < nats_in_cursum(journal); i++) {
2396 struct nat_entry *ne;
2397 struct f2fs_nat_entry raw_ne;
2398 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2399
2400 raw_ne = nat_in_journal(journal, i);
2401
2402 ne = __lookup_nat_cache(nm_i, nid);
2403 if (!ne) {
2404 ne = __alloc_nat_entry(nid, true);
2405 __init_nat_entry(nm_i, ne, &raw_ne, true);
2406 }
2407
2408 /*
2409 * if a free nat in journal has not been used after last
2410 * checkpoint, we should remove it from available nids,
2411 * since later we will add it again.
2412 */
2413 if (!get_nat_flag(ne, IS_DIRTY) &&
2414 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2415 spin_lock(&nm_i->nid_list_lock);
2416 nm_i->available_nids--;
2417 spin_unlock(&nm_i->nid_list_lock);
2418 }
2419
2420 __set_nat_cache_dirty(nm_i, ne);
2421 }
2422 update_nats_in_cursum(journal, -i);
2423 up_write(&curseg->journal_rwsem);
2424}
2425
2426static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2427 struct list_head *head, int max)
2428{
2429 struct nat_entry_set *cur;
2430
2431 if (nes->entry_cnt >= max)
2432 goto add_out;
2433
2434 list_for_each_entry(cur, head, set_list) {
2435 if (cur->entry_cnt >= nes->entry_cnt) {
2436 list_add(&nes->set_list, cur->set_list.prev);
2437 return;
2438 }
2439 }
2440add_out:
2441 list_add_tail(&nes->set_list, head);
2442}
2443
2444static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2445 struct page *page)
2446{
2447 struct f2fs_nm_info *nm_i = NM_I(sbi);
2448 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2449 struct f2fs_nat_block *nat_blk = page_address(page);
2450 int valid = 0;
2451 int i = 0;
2452
2453 if (!enabled_nat_bits(sbi, NULL))
2454 return;
2455
2456 if (nat_index == 0) {
2457 valid = 1;
2458 i = 1;
2459 }
2460 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2461 if (nat_blk->entries[i].block_addr != NULL_ADDR)
2462 valid++;
2463 }
2464 if (valid == 0) {
2465 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2466 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2467 return;
2468 }
2469
2470 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2471 if (valid == NAT_ENTRY_PER_BLOCK)
2472 __set_bit_le(nat_index, nm_i->full_nat_bits);
2473 else
2474 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2475}
2476
2477static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2478 struct nat_entry_set *set, struct cp_control *cpc)
2479{
2480 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2481 struct f2fs_journal *journal = curseg->journal;
2482 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2483 bool to_journal = true;
2484 struct f2fs_nat_block *nat_blk;
2485 struct nat_entry *ne, *cur;
2486 struct page *page = NULL;
2487
2488 /*
2489 * there are two steps to flush nat entries:
2490 * #1, flush nat entries to journal in current hot data summary block.
2491 * #2, flush nat entries to nat page.
2492 */
2493 if (enabled_nat_bits(sbi, cpc) ||
2494 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2495 to_journal = false;
2496
2497 if (to_journal) {
2498 down_write(&curseg->journal_rwsem);
2499 } else {
2500 page = get_next_nat_page(sbi, start_nid);
2501 nat_blk = page_address(page);
2502 f2fs_bug_on(sbi, !nat_blk);
2503 }
2504
2505 /* flush dirty nats in nat entry set */
2506 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2507 struct f2fs_nat_entry *raw_ne;
2508 nid_t nid = nat_get_nid(ne);
2509 int offset;
2510
2511 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2512
2513 if (to_journal) {
2514 offset = lookup_journal_in_cursum(journal,
2515 NAT_JOURNAL, nid, 1);
2516 f2fs_bug_on(sbi, offset < 0);
2517 raw_ne = &nat_in_journal(journal, offset);
2518 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2519 } else {
2520 raw_ne = &nat_blk->entries[nid - start_nid];
2521 }
2522 raw_nat_from_node_info(raw_ne, &ne->ni);
2523 nat_reset_flag(ne);
2524 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2525 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2526 add_free_nid(sbi, nid, false, true);
2527 } else {
2528 spin_lock(&NM_I(sbi)->nid_list_lock);
2529 update_free_nid_bitmap(sbi, nid, false, false);
2530 spin_unlock(&NM_I(sbi)->nid_list_lock);
2531 }
2532 }
2533
2534 if (to_journal) {
2535 up_write(&curseg->journal_rwsem);
2536 } else {
2537 __update_nat_bits(sbi, start_nid, page);
2538 f2fs_put_page(page, 1);
2539 }
2540
2541 /* Allow dirty nats by node block allocation in write_begin */
2542 if (!set->entry_cnt) {
2543 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2544 kmem_cache_free(nat_entry_set_slab, set);
2545 }
2546}
2547
2548/*
2549 * This function is called during the checkpointing process.
2550 */
2551void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2552{
2553 struct f2fs_nm_info *nm_i = NM_I(sbi);
2554 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2555 struct f2fs_journal *journal = curseg->journal;
2556 struct nat_entry_set *setvec[SETVEC_SIZE];
2557 struct nat_entry_set *set, *tmp;
2558 unsigned int found;
2559 nid_t set_idx = 0;
2560 LIST_HEAD(sets);
2561
2562 if (!nm_i->dirty_nat_cnt)
2563 return;
2564
2565 down_write(&nm_i->nat_tree_lock);
2566
2567 /*
2568 * if there are no enough space in journal to store dirty nat
2569 * entries, remove all entries from journal and merge them
2570 * into nat entry set.
2571 */
2572 if (enabled_nat_bits(sbi, cpc) ||
2573 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2574 remove_nats_in_journal(sbi);
2575
2576 while ((found = __gang_lookup_nat_set(nm_i,
2577 set_idx, SETVEC_SIZE, setvec))) {
2578 unsigned idx;
2579 set_idx = setvec[found - 1]->set + 1;
2580 for (idx = 0; idx < found; idx++)
2581 __adjust_nat_entry_set(setvec[idx], &sets,
2582 MAX_NAT_JENTRIES(journal));
2583 }
2584
2585 /* flush dirty nats in nat entry set */
2586 list_for_each_entry_safe(set, tmp, &sets, set_list)
2587 __flush_nat_entry_set(sbi, set, cpc);
2588
2589 up_write(&nm_i->nat_tree_lock);
2590 /* Allow dirty nats by node block allocation in write_begin */
2591}
2592
2593static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2594{
2595 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2596 struct f2fs_nm_info *nm_i = NM_I(sbi);
2597 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2598 unsigned int i;
2599 __u64 cp_ver = cur_cp_version(ckpt);
2600 block_t nat_bits_addr;
2601
2602 if (!enabled_nat_bits(sbi, NULL))
2603 return 0;
2604
2605 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2606 nm_i->nat_bits = f2fs_kzalloc(sbi,
2607 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2608 if (!nm_i->nat_bits)
2609 return -ENOMEM;
2610
2611 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2612 nm_i->nat_bits_blocks;
2613 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2614 struct page *page = get_meta_page(sbi, nat_bits_addr++);
2615
2616 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2617 page_address(page), F2FS_BLKSIZE);
2618 f2fs_put_page(page, 1);
2619 }
2620
2621 cp_ver |= (cur_cp_crc(ckpt) << 32);
2622 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2623 disable_nat_bits(sbi, true);
2624 return 0;
2625 }
2626
2627 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2628 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2629
2630 f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2631 return 0;
2632}
2633
2634static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2635{
2636 struct f2fs_nm_info *nm_i = NM_I(sbi);
2637 unsigned int i = 0;
2638 nid_t nid, last_nid;
2639
2640 if (!enabled_nat_bits(sbi, NULL))
2641 return;
2642
2643 for (i = 0; i < nm_i->nat_blocks; i++) {
2644 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2645 if (i >= nm_i->nat_blocks)
2646 break;
2647
2648 __set_bit_le(i, nm_i->nat_block_bitmap);
2649
2650 nid = i * NAT_ENTRY_PER_BLOCK;
2651 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2652
2653 spin_lock(&NM_I(sbi)->nid_list_lock);
2654 for (; nid < last_nid; nid++)
2655 update_free_nid_bitmap(sbi, nid, true, true);
2656 spin_unlock(&NM_I(sbi)->nid_list_lock);
2657 }
2658
2659 for (i = 0; i < nm_i->nat_blocks; i++) {
2660 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2661 if (i >= nm_i->nat_blocks)
2662 break;
2663
2664 __set_bit_le(i, nm_i->nat_block_bitmap);
2665 }
2666}
2667
2668static int init_node_manager(struct f2fs_sb_info *sbi)
2669{
2670 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2671 struct f2fs_nm_info *nm_i = NM_I(sbi);
2672 unsigned char *version_bitmap;
2673 unsigned int nat_segs;
2674 int err;
2675
2676 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2677
2678 /* segment_count_nat includes pair segment so divide to 2. */
2679 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2680 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2681 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2682
2683 /* not used nids: 0, node, meta, (and root counted as valid node) */
2684 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2685 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2686 nm_i->nid_cnt[FREE_NID] = 0;
2687 nm_i->nid_cnt[PREALLOC_NID] = 0;
2688 nm_i->nat_cnt = 0;
2689 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2690 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2691 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2692
2693 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2694 INIT_LIST_HEAD(&nm_i->free_nid_list);
2695 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2696 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2697 INIT_LIST_HEAD(&nm_i->nat_entries);
2698
2699 mutex_init(&nm_i->build_lock);
2700 spin_lock_init(&nm_i->nid_list_lock);
2701 init_rwsem(&nm_i->nat_tree_lock);
2702
2703 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2704 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2705 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2706 if (!version_bitmap)
2707 return -EFAULT;
2708
2709 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2710 GFP_KERNEL);
2711 if (!nm_i->nat_bitmap)
2712 return -ENOMEM;
2713
2714 err = __get_nat_bitmaps(sbi);
2715 if (err)
2716 return err;
2717
2718#ifdef CONFIG_F2FS_CHECK_FS
2719 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2720 GFP_KERNEL);
2721 if (!nm_i->nat_bitmap_mir)
2722 return -ENOMEM;
2723#endif
2724
2725 return 0;
2726}
2727
2728static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2729{
2730 struct f2fs_nm_info *nm_i = NM_I(sbi);
2731 int i;
2732
2733 nm_i->free_nid_bitmap = f2fs_kzalloc(sbi, nm_i->nat_blocks *
2734 sizeof(unsigned char *), GFP_KERNEL);
2735 if (!nm_i->free_nid_bitmap)
2736 return -ENOMEM;
2737
2738 for (i = 0; i < nm_i->nat_blocks; i++) {
2739 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
2740 NAT_ENTRY_BITMAP_SIZE_ALIGNED, GFP_KERNEL);
2741 if (!nm_i->free_nid_bitmap)
2742 return -ENOMEM;
2743 }
2744
2745 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
2746 GFP_KERNEL);
2747 if (!nm_i->nat_block_bitmap)
2748 return -ENOMEM;
2749
2750 nm_i->free_nid_count = f2fs_kvzalloc(sbi, nm_i->nat_blocks *
2751 sizeof(unsigned short), GFP_KERNEL);
2752 if (!nm_i->free_nid_count)
2753 return -ENOMEM;
2754 return 0;
2755}
2756
2757int build_node_manager(struct f2fs_sb_info *sbi)
2758{
2759 int err;
2760
2761 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
2762 GFP_KERNEL);
2763 if (!sbi->nm_info)
2764 return -ENOMEM;
2765
2766 err = init_node_manager(sbi);
2767 if (err)
2768 return err;
2769
2770 err = init_free_nid_cache(sbi);
2771 if (err)
2772 return err;
2773
2774 /* load free nid status from nat_bits table */
2775 load_free_nid_bitmap(sbi);
2776
2777 build_free_nids(sbi, true, true);
2778 return 0;
2779}
2780
2781void destroy_node_manager(struct f2fs_sb_info *sbi)
2782{
2783 struct f2fs_nm_info *nm_i = NM_I(sbi);
2784 struct free_nid *i, *next_i;
2785 struct nat_entry *natvec[NATVEC_SIZE];
2786 struct nat_entry_set *setvec[SETVEC_SIZE];
2787 nid_t nid = 0;
2788 unsigned int found;
2789
2790 if (!nm_i)
2791 return;
2792
2793 /* destroy free nid list */
2794 spin_lock(&nm_i->nid_list_lock);
2795 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2796 __remove_free_nid(sbi, i, FREE_NID);
2797 spin_unlock(&nm_i->nid_list_lock);
2798 kmem_cache_free(free_nid_slab, i);
2799 spin_lock(&nm_i->nid_list_lock);
2800 }
2801 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2802 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2803 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2804 spin_unlock(&nm_i->nid_list_lock);
2805
2806 /* destroy nat cache */
2807 down_write(&nm_i->nat_tree_lock);
2808 while ((found = __gang_lookup_nat_cache(nm_i,
2809 nid, NATVEC_SIZE, natvec))) {
2810 unsigned idx;
2811
2812 nid = nat_get_nid(natvec[found - 1]) + 1;
2813 for (idx = 0; idx < found; idx++)
2814 __del_from_nat_cache(nm_i, natvec[idx]);
2815 }
2816 f2fs_bug_on(sbi, nm_i->nat_cnt);
2817
2818 /* destroy nat set cache */
2819 nid = 0;
2820 while ((found = __gang_lookup_nat_set(nm_i,
2821 nid, SETVEC_SIZE, setvec))) {
2822 unsigned idx;
2823
2824 nid = setvec[found - 1]->set + 1;
2825 for (idx = 0; idx < found; idx++) {
2826 /* entry_cnt is not zero, when cp_error was occurred */
2827 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2828 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2829 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2830 }
2831 }
2832 up_write(&nm_i->nat_tree_lock);
2833
2834 kvfree(nm_i->nat_block_bitmap);
2835 if (nm_i->free_nid_bitmap) {
2836 int i;
2837
2838 for (i = 0; i < nm_i->nat_blocks; i++)
2839 kvfree(nm_i->free_nid_bitmap[i]);
2840 kfree(nm_i->free_nid_bitmap);
2841 }
2842 kvfree(nm_i->free_nid_count);
2843
2844 kfree(nm_i->nat_bitmap);
2845 kfree(nm_i->nat_bits);
2846#ifdef CONFIG_F2FS_CHECK_FS
2847 kfree(nm_i->nat_bitmap_mir);
2848#endif
2849 sbi->nm_info = NULL;
2850 kfree(nm_i);
2851}
2852
2853int __init create_node_manager_caches(void)
2854{
2855 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2856 sizeof(struct nat_entry));
2857 if (!nat_entry_slab)
2858 goto fail;
2859
2860 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2861 sizeof(struct free_nid));
2862 if (!free_nid_slab)
2863 goto destroy_nat_entry;
2864
2865 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2866 sizeof(struct nat_entry_set));
2867 if (!nat_entry_set_slab)
2868 goto destroy_free_nid;
2869 return 0;
2870
2871destroy_free_nid:
2872 kmem_cache_destroy(free_nid_slab);
2873destroy_nat_entry:
2874 kmem_cache_destroy(nat_entry_slab);
2875fail:
2876 return -ENOMEM;
2877}
2878
2879void destroy_node_manager_caches(void)
2880{
2881 kmem_cache_destroy(nat_entry_set_slab);
2882 kmem_cache_destroy(free_nid_slab);
2883 kmem_cache_destroy(nat_entry_slab);
2884}
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
27static struct kmem_cache *nat_entry_slab;
28static struct kmem_cache *free_nid_slab;
29static struct kmem_cache *nat_entry_set_slab;
30
31bool 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
80static 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
98static 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
104static 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
133static 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
138static 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
144static 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
152static 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
177static 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
192static 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
199int 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
216bool 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
230bool 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
246static 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
260static 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
277static 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
336int 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 */
358void 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);
402cache:
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
410pgoff_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 */
445static 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 }
515got:
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 */
525int 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
616release_pages:
617 f2fs_put_page(parent, 1);
618 if (i > 1)
619 f2fs_put_page(npage[0], 0);
620release_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
630static 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 }
653invalidate:
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
666static 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
688static 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
755out_err:
756 f2fs_put_page(page, 1);
757 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
758 return ret;
759}
760
761static 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--;
810fail:
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 */
822int 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);
835restart:
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
872skip_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 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
900 f2fs_put_page(page, 1);
901 goto restart;
902 }
903 f2fs_wait_on_page_writeback(page, NODE, true);
904 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
905 set_page_dirty(page);
906 unlock_page(page);
907 }
908 offset[1] = 0;
909 offset[0]++;
910 nofs += err;
911 }
912fail:
913 f2fs_put_page(page, 0);
914 trace_f2fs_truncate_inode_blocks_exit(inode, err);
915 return err > 0 ? 0 : err;
916}
917
918int truncate_xattr_node(struct inode *inode, struct page *page)
919{
920 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
921 nid_t nid = F2FS_I(inode)->i_xattr_nid;
922 struct dnode_of_data dn;
923 struct page *npage;
924
925 if (!nid)
926 return 0;
927
928 npage = get_node_page(sbi, nid);
929 if (IS_ERR(npage))
930 return PTR_ERR(npage);
931
932 F2FS_I(inode)->i_xattr_nid = 0;
933
934 /* need to do checkpoint during fsync */
935 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
936
937 set_new_dnode(&dn, inode, page, npage, nid);
938
939 if (page)
940 dn.inode_page_locked = true;
941 truncate_node(&dn);
942 return 0;
943}
944
945/*
946 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
947 * f2fs_unlock_op().
948 */
949int remove_inode_page(struct inode *inode)
950{
951 struct dnode_of_data dn;
952 int err;
953
954 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
955 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
956 if (err)
957 return err;
958
959 err = truncate_xattr_node(inode, dn.inode_page);
960 if (err) {
961 f2fs_put_dnode(&dn);
962 return err;
963 }
964
965 /* remove potential inline_data blocks */
966 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
967 S_ISLNK(inode->i_mode))
968 truncate_data_blocks_range(&dn, 1);
969
970 /* 0 is possible, after f2fs_new_inode() has failed */
971 f2fs_bug_on(F2FS_I_SB(inode),
972 inode->i_blocks != 0 && inode->i_blocks != 1);
973
974 /* will put inode & node pages */
975 truncate_node(&dn);
976 return 0;
977}
978
979struct page *new_inode_page(struct inode *inode)
980{
981 struct dnode_of_data dn;
982
983 /* allocate inode page for new inode */
984 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
985
986 /* caller should f2fs_put_page(page, 1); */
987 return new_node_page(&dn, 0, NULL);
988}
989
990struct page *new_node_page(struct dnode_of_data *dn,
991 unsigned int ofs, struct page *ipage)
992{
993 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
994 struct node_info old_ni, new_ni;
995 struct page *page;
996 int err;
997
998 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
999 return ERR_PTR(-EPERM);
1000
1001 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
1002 if (!page)
1003 return ERR_PTR(-ENOMEM);
1004
1005 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1006 err = -ENOSPC;
1007 goto fail;
1008 }
1009
1010 get_node_info(sbi, dn->nid, &old_ni);
1011
1012 /* Reinitialize old_ni with new node page */
1013 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1014 new_ni = old_ni;
1015 new_ni.ino = dn->inode->i_ino;
1016 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1017
1018 f2fs_wait_on_page_writeback(page, NODE, true);
1019 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1020 set_cold_node(dn->inode, page);
1021 SetPageUptodate(page);
1022 if (set_page_dirty(page))
1023 dn->node_changed = true;
1024
1025 if (f2fs_has_xattr_block(ofs))
1026 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
1027
1028 dn->node_page = page;
1029 if (ipage)
1030 update_inode(dn->inode, ipage);
1031 else
1032 sync_inode_page(dn);
1033 if (ofs == 0)
1034 inc_valid_inode_count(sbi);
1035
1036 return page;
1037
1038fail:
1039 clear_node_page_dirty(page);
1040 f2fs_put_page(page, 1);
1041 return ERR_PTR(err);
1042}
1043
1044/*
1045 * Caller should do after getting the following values.
1046 * 0: f2fs_put_page(page, 0)
1047 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1048 */
1049static int read_node_page(struct page *page, int rw)
1050{
1051 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1052 struct node_info ni;
1053 struct f2fs_io_info fio = {
1054 .sbi = sbi,
1055 .type = NODE,
1056 .rw = rw,
1057 .page = page,
1058 .encrypted_page = NULL,
1059 };
1060
1061 get_node_info(sbi, page->index, &ni);
1062
1063 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1064 ClearPageUptodate(page);
1065 return -ENOENT;
1066 }
1067
1068 if (PageUptodate(page))
1069 return LOCKED_PAGE;
1070
1071 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1072 return f2fs_submit_page_bio(&fio);
1073}
1074
1075/*
1076 * Readahead a node page
1077 */
1078void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1079{
1080 struct page *apage;
1081 int err;
1082
1083 if (!nid)
1084 return;
1085 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1086
1087 rcu_read_lock();
1088 apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1089 rcu_read_unlock();
1090 if (apage)
1091 return;
1092
1093 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1094 if (!apage)
1095 return;
1096
1097 err = read_node_page(apage, READA);
1098 f2fs_put_page(apage, err ? 1 : 0);
1099}
1100
1101/*
1102 * readahead MAX_RA_NODE number of node pages.
1103 */
1104static void ra_node_pages(struct page *parent, int start)
1105{
1106 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1107 struct blk_plug plug;
1108 int i, end;
1109 nid_t nid;
1110
1111 blk_start_plug(&plug);
1112
1113 /* Then, try readahead for siblings of the desired node */
1114 end = start + MAX_RA_NODE;
1115 end = min(end, NIDS_PER_BLOCK);
1116 for (i = start; i < end; i++) {
1117 nid = get_nid(parent, i, false);
1118 ra_node_page(sbi, nid);
1119 }
1120
1121 blk_finish_plug(&plug);
1122}
1123
1124static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1125 struct page *parent, int start)
1126{
1127 struct page *page;
1128 int err;
1129
1130 if (!nid)
1131 return ERR_PTR(-ENOENT);
1132 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1133repeat:
1134 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1135 if (!page)
1136 return ERR_PTR(-ENOMEM);
1137
1138 err = read_node_page(page, READ_SYNC);
1139 if (err < 0) {
1140 f2fs_put_page(page, 1);
1141 return ERR_PTR(err);
1142 } else if (err == LOCKED_PAGE) {
1143 goto page_hit;
1144 }
1145
1146 if (parent)
1147 ra_node_pages(parent, start + 1);
1148
1149 lock_page(page);
1150
1151 if (unlikely(!PageUptodate(page))) {
1152 f2fs_put_page(page, 1);
1153 return ERR_PTR(-EIO);
1154 }
1155 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1156 f2fs_put_page(page, 1);
1157 goto repeat;
1158 }
1159page_hit:
1160 f2fs_bug_on(sbi, nid != nid_of_node(page));
1161 return page;
1162}
1163
1164struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1165{
1166 return __get_node_page(sbi, nid, NULL, 0);
1167}
1168
1169struct page *get_node_page_ra(struct page *parent, int start)
1170{
1171 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1172 nid_t nid = get_nid(parent, start, false);
1173
1174 return __get_node_page(sbi, nid, parent, start);
1175}
1176
1177void sync_inode_page(struct dnode_of_data *dn)
1178{
1179 int ret = 0;
1180
1181 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1182 ret = update_inode(dn->inode, dn->node_page);
1183 } else if (dn->inode_page) {
1184 if (!dn->inode_page_locked)
1185 lock_page(dn->inode_page);
1186 ret = update_inode(dn->inode, dn->inode_page);
1187 if (!dn->inode_page_locked)
1188 unlock_page(dn->inode_page);
1189 } else {
1190 ret = update_inode_page(dn->inode);
1191 }
1192 dn->node_changed = ret ? true: false;
1193}
1194
1195static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1196{
1197 struct inode *inode;
1198 struct page *page;
1199
1200 /* should flush inline_data before evict_inode */
1201 inode = ilookup(sbi->sb, ino);
1202 if (!inode)
1203 return;
1204
1205 page = pagecache_get_page(inode->i_mapping, 0, FGP_NOWAIT, 0);
1206 if (!page)
1207 goto iput_out;
1208
1209 if (!trylock_page(page))
1210 goto release_out;
1211
1212 if (!PageUptodate(page))
1213 goto page_out;
1214
1215 if (!PageDirty(page))
1216 goto page_out;
1217
1218 if (!clear_page_dirty_for_io(page))
1219 goto page_out;
1220
1221 if (!f2fs_write_inline_data(inode, page))
1222 inode_dec_dirty_pages(inode);
1223 else
1224 set_page_dirty(page);
1225page_out:
1226 unlock_page(page);
1227release_out:
1228 f2fs_put_page(page, 0);
1229iput_out:
1230 iput(inode);
1231}
1232
1233int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1234 struct writeback_control *wbc)
1235{
1236 pgoff_t index, end;
1237 struct pagevec pvec;
1238 int step = ino ? 2 : 0;
1239 int nwritten = 0;
1240
1241 pagevec_init(&pvec, 0);
1242
1243next_step:
1244 index = 0;
1245 end = ULONG_MAX;
1246
1247 while (index <= end) {
1248 int i, nr_pages;
1249 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1250 PAGECACHE_TAG_DIRTY,
1251 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1252 if (nr_pages == 0)
1253 break;
1254
1255 for (i = 0; i < nr_pages; i++) {
1256 struct page *page = pvec.pages[i];
1257
1258 if (unlikely(f2fs_cp_error(sbi))) {
1259 pagevec_release(&pvec);
1260 return -EIO;
1261 }
1262
1263 /*
1264 * flushing sequence with step:
1265 * 0. indirect nodes
1266 * 1. dentry dnodes
1267 * 2. file dnodes
1268 */
1269 if (step == 0 && IS_DNODE(page))
1270 continue;
1271 if (step == 1 && (!IS_DNODE(page) ||
1272 is_cold_node(page)))
1273 continue;
1274 if (step == 2 && (!IS_DNODE(page) ||
1275 !is_cold_node(page)))
1276 continue;
1277
1278 /*
1279 * If an fsync mode,
1280 * we should not skip writing node pages.
1281 */
1282lock_node:
1283 if (ino && ino_of_node(page) == ino)
1284 lock_page(page);
1285 else if (!trylock_page(page))
1286 continue;
1287
1288 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1289continue_unlock:
1290 unlock_page(page);
1291 continue;
1292 }
1293 if (ino && ino_of_node(page) != ino)
1294 goto continue_unlock;
1295
1296 if (!PageDirty(page)) {
1297 /* someone wrote it for us */
1298 goto continue_unlock;
1299 }
1300
1301 /* flush inline_data */
1302 if (!ino && is_inline_node(page)) {
1303 clear_inline_node(page);
1304 unlock_page(page);
1305 flush_inline_data(sbi, ino_of_node(page));
1306 goto lock_node;
1307 }
1308
1309 f2fs_wait_on_page_writeback(page, NODE, true);
1310
1311 BUG_ON(PageWriteback(page));
1312 if (!clear_page_dirty_for_io(page))
1313 goto continue_unlock;
1314
1315 /* called by fsync() */
1316 if (ino && IS_DNODE(page)) {
1317 set_fsync_mark(page, 1);
1318 if (IS_INODE(page))
1319 set_dentry_mark(page,
1320 need_dentry_mark(sbi, ino));
1321 nwritten++;
1322 } else {
1323 set_fsync_mark(page, 0);
1324 set_dentry_mark(page, 0);
1325 }
1326
1327 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1328 unlock_page(page);
1329
1330 if (--wbc->nr_to_write == 0)
1331 break;
1332 }
1333 pagevec_release(&pvec);
1334 cond_resched();
1335
1336 if (wbc->nr_to_write == 0) {
1337 step = 2;
1338 break;
1339 }
1340 }
1341
1342 if (step < 2) {
1343 step++;
1344 goto next_step;
1345 }
1346 return nwritten;
1347}
1348
1349int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1350{
1351 pgoff_t index = 0, end = ULONG_MAX;
1352 struct pagevec pvec;
1353 int ret2 = 0, ret = 0;
1354
1355 pagevec_init(&pvec, 0);
1356
1357 while (index <= end) {
1358 int i, nr_pages;
1359 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1360 PAGECACHE_TAG_WRITEBACK,
1361 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1362 if (nr_pages == 0)
1363 break;
1364
1365 for (i = 0; i < nr_pages; i++) {
1366 struct page *page = pvec.pages[i];
1367
1368 /* until radix tree lookup accepts end_index */
1369 if (unlikely(page->index > end))
1370 continue;
1371
1372 if (ino && ino_of_node(page) == ino) {
1373 f2fs_wait_on_page_writeback(page, NODE, true);
1374 if (TestClearPageError(page))
1375 ret = -EIO;
1376 }
1377 }
1378 pagevec_release(&pvec);
1379 cond_resched();
1380 }
1381
1382 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1383 ret2 = -ENOSPC;
1384 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1385 ret2 = -EIO;
1386 if (!ret)
1387 ret = ret2;
1388 return ret;
1389}
1390
1391static int f2fs_write_node_page(struct page *page,
1392 struct writeback_control *wbc)
1393{
1394 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1395 nid_t nid;
1396 struct node_info ni;
1397 struct f2fs_io_info fio = {
1398 .sbi = sbi,
1399 .type = NODE,
1400 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1401 .page = page,
1402 .encrypted_page = NULL,
1403 };
1404
1405 trace_f2fs_writepage(page, NODE);
1406
1407 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1408 goto redirty_out;
1409 if (unlikely(f2fs_cp_error(sbi)))
1410 goto redirty_out;
1411
1412 /* get old block addr of this node page */
1413 nid = nid_of_node(page);
1414 f2fs_bug_on(sbi, page->index != nid);
1415
1416 if (wbc->for_reclaim) {
1417 if (!down_read_trylock(&sbi->node_write))
1418 goto redirty_out;
1419 } else {
1420 down_read(&sbi->node_write);
1421 }
1422
1423 get_node_info(sbi, nid, &ni);
1424
1425 /* This page is already truncated */
1426 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1427 ClearPageUptodate(page);
1428 dec_page_count(sbi, F2FS_DIRTY_NODES);
1429 up_read(&sbi->node_write);
1430 unlock_page(page);
1431 return 0;
1432 }
1433
1434 set_page_writeback(page);
1435 fio.old_blkaddr = ni.blk_addr;
1436 write_node_page(nid, &fio);
1437 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1438 dec_page_count(sbi, F2FS_DIRTY_NODES);
1439 up_read(&sbi->node_write);
1440
1441 if (wbc->for_reclaim)
1442 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1443
1444 unlock_page(page);
1445
1446 if (unlikely(f2fs_cp_error(sbi)))
1447 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1448
1449 return 0;
1450
1451redirty_out:
1452 redirty_page_for_writepage(wbc, page);
1453 return AOP_WRITEPAGE_ACTIVATE;
1454}
1455
1456static int f2fs_write_node_pages(struct address_space *mapping,
1457 struct writeback_control *wbc)
1458{
1459 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1460 long diff;
1461
1462 /* balancing f2fs's metadata in background */
1463 f2fs_balance_fs_bg(sbi);
1464
1465 /* collect a number of dirty node pages and write together */
1466 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1467 goto skip_write;
1468
1469 trace_f2fs_writepages(mapping->host, wbc, NODE);
1470
1471 diff = nr_pages_to_write(sbi, NODE, wbc);
1472 wbc->sync_mode = WB_SYNC_NONE;
1473 sync_node_pages(sbi, 0, wbc);
1474 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1475 return 0;
1476
1477skip_write:
1478 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1479 trace_f2fs_writepages(mapping->host, wbc, NODE);
1480 return 0;
1481}
1482
1483static int f2fs_set_node_page_dirty(struct page *page)
1484{
1485 trace_f2fs_set_page_dirty(page, NODE);
1486
1487 SetPageUptodate(page);
1488 if (!PageDirty(page)) {
1489 __set_page_dirty_nobuffers(page);
1490 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1491 SetPagePrivate(page);
1492 f2fs_trace_pid(page);
1493 return 1;
1494 }
1495 return 0;
1496}
1497
1498/*
1499 * Structure of the f2fs node operations
1500 */
1501const struct address_space_operations f2fs_node_aops = {
1502 .writepage = f2fs_write_node_page,
1503 .writepages = f2fs_write_node_pages,
1504 .set_page_dirty = f2fs_set_node_page_dirty,
1505 .invalidatepage = f2fs_invalidate_page,
1506 .releasepage = f2fs_release_page,
1507};
1508
1509static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1510 nid_t n)
1511{
1512 return radix_tree_lookup(&nm_i->free_nid_root, n);
1513}
1514
1515static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1516 struct free_nid *i)
1517{
1518 list_del(&i->list);
1519 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1520}
1521
1522static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1523{
1524 struct f2fs_nm_info *nm_i = NM_I(sbi);
1525 struct free_nid *i;
1526 struct nat_entry *ne;
1527 bool allocated = false;
1528
1529 if (!available_free_memory(sbi, FREE_NIDS))
1530 return -1;
1531
1532 /* 0 nid should not be used */
1533 if (unlikely(nid == 0))
1534 return 0;
1535
1536 if (build) {
1537 /* do not add allocated nids */
1538 ne = __lookup_nat_cache(nm_i, nid);
1539 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1540 nat_get_blkaddr(ne) != NULL_ADDR))
1541 allocated = true;
1542 if (allocated)
1543 return 0;
1544 }
1545
1546 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1547 i->nid = nid;
1548 i->state = NID_NEW;
1549
1550 if (radix_tree_preload(GFP_NOFS)) {
1551 kmem_cache_free(free_nid_slab, i);
1552 return 0;
1553 }
1554
1555 spin_lock(&nm_i->free_nid_list_lock);
1556 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1557 spin_unlock(&nm_i->free_nid_list_lock);
1558 radix_tree_preload_end();
1559 kmem_cache_free(free_nid_slab, i);
1560 return 0;
1561 }
1562 list_add_tail(&i->list, &nm_i->free_nid_list);
1563 nm_i->fcnt++;
1564 spin_unlock(&nm_i->free_nid_list_lock);
1565 radix_tree_preload_end();
1566 return 1;
1567}
1568
1569static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1570{
1571 struct free_nid *i;
1572 bool need_free = false;
1573
1574 spin_lock(&nm_i->free_nid_list_lock);
1575 i = __lookup_free_nid_list(nm_i, nid);
1576 if (i && i->state == NID_NEW) {
1577 __del_from_free_nid_list(nm_i, i);
1578 nm_i->fcnt--;
1579 need_free = true;
1580 }
1581 spin_unlock(&nm_i->free_nid_list_lock);
1582
1583 if (need_free)
1584 kmem_cache_free(free_nid_slab, i);
1585}
1586
1587static void scan_nat_page(struct f2fs_sb_info *sbi,
1588 struct page *nat_page, nid_t start_nid)
1589{
1590 struct f2fs_nm_info *nm_i = NM_I(sbi);
1591 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1592 block_t blk_addr;
1593 int i;
1594
1595 i = start_nid % NAT_ENTRY_PER_BLOCK;
1596
1597 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1598
1599 if (unlikely(start_nid >= nm_i->max_nid))
1600 break;
1601
1602 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1603 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1604 if (blk_addr == NULL_ADDR) {
1605 if (add_free_nid(sbi, start_nid, true) < 0)
1606 break;
1607 }
1608 }
1609}
1610
1611static void build_free_nids(struct f2fs_sb_info *sbi)
1612{
1613 struct f2fs_nm_info *nm_i = NM_I(sbi);
1614 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1615 struct f2fs_journal *journal = curseg->journal;
1616 int i = 0;
1617 nid_t nid = nm_i->next_scan_nid;
1618
1619 /* Enough entries */
1620 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1621 return;
1622
1623 /* readahead nat pages to be scanned */
1624 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1625 META_NAT, true);
1626
1627 down_read(&nm_i->nat_tree_lock);
1628
1629 while (1) {
1630 struct page *page = get_current_nat_page(sbi, nid);
1631
1632 scan_nat_page(sbi, page, nid);
1633 f2fs_put_page(page, 1);
1634
1635 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1636 if (unlikely(nid >= nm_i->max_nid))
1637 nid = 0;
1638
1639 if (++i >= FREE_NID_PAGES)
1640 break;
1641 }
1642
1643 /* go to the next free nat pages to find free nids abundantly */
1644 nm_i->next_scan_nid = nid;
1645
1646 /* find free nids from current sum_pages */
1647 down_read(&curseg->journal_rwsem);
1648 for (i = 0; i < nats_in_cursum(journal); i++) {
1649 block_t addr;
1650
1651 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1652 nid = le32_to_cpu(nid_in_journal(journal, i));
1653 if (addr == NULL_ADDR)
1654 add_free_nid(sbi, nid, true);
1655 else
1656 remove_free_nid(nm_i, nid);
1657 }
1658 up_read(&curseg->journal_rwsem);
1659 up_read(&nm_i->nat_tree_lock);
1660
1661 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1662 nm_i->ra_nid_pages, META_NAT, false);
1663}
1664
1665/*
1666 * If this function returns success, caller can obtain a new nid
1667 * from second parameter of this function.
1668 * The returned nid could be used ino as well as nid when inode is created.
1669 */
1670bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1671{
1672 struct f2fs_nm_info *nm_i = NM_I(sbi);
1673 struct free_nid *i = NULL;
1674retry:
1675 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1676 return false;
1677
1678 spin_lock(&nm_i->free_nid_list_lock);
1679
1680 /* We should not use stale free nids created by build_free_nids */
1681 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1682 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1683 list_for_each_entry(i, &nm_i->free_nid_list, list)
1684 if (i->state == NID_NEW)
1685 break;
1686
1687 f2fs_bug_on(sbi, i->state != NID_NEW);
1688 *nid = i->nid;
1689 i->state = NID_ALLOC;
1690 nm_i->fcnt--;
1691 spin_unlock(&nm_i->free_nid_list_lock);
1692 return true;
1693 }
1694 spin_unlock(&nm_i->free_nid_list_lock);
1695
1696 /* Let's scan nat pages and its caches to get free nids */
1697 mutex_lock(&nm_i->build_lock);
1698 build_free_nids(sbi);
1699 mutex_unlock(&nm_i->build_lock);
1700 goto retry;
1701}
1702
1703/*
1704 * alloc_nid() should be called prior to this function.
1705 */
1706void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1707{
1708 struct f2fs_nm_info *nm_i = NM_I(sbi);
1709 struct free_nid *i;
1710
1711 spin_lock(&nm_i->free_nid_list_lock);
1712 i = __lookup_free_nid_list(nm_i, nid);
1713 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1714 __del_from_free_nid_list(nm_i, i);
1715 spin_unlock(&nm_i->free_nid_list_lock);
1716
1717 kmem_cache_free(free_nid_slab, i);
1718}
1719
1720/*
1721 * alloc_nid() should be called prior to this function.
1722 */
1723void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1724{
1725 struct f2fs_nm_info *nm_i = NM_I(sbi);
1726 struct free_nid *i;
1727 bool need_free = false;
1728
1729 if (!nid)
1730 return;
1731
1732 spin_lock(&nm_i->free_nid_list_lock);
1733 i = __lookup_free_nid_list(nm_i, nid);
1734 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1735 if (!available_free_memory(sbi, FREE_NIDS)) {
1736 __del_from_free_nid_list(nm_i, i);
1737 need_free = true;
1738 } else {
1739 i->state = NID_NEW;
1740 nm_i->fcnt++;
1741 }
1742 spin_unlock(&nm_i->free_nid_list_lock);
1743
1744 if (need_free)
1745 kmem_cache_free(free_nid_slab, i);
1746}
1747
1748int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1749{
1750 struct f2fs_nm_info *nm_i = NM_I(sbi);
1751 struct free_nid *i, *next;
1752 int nr = nr_shrink;
1753
1754 if (!mutex_trylock(&nm_i->build_lock))
1755 return 0;
1756
1757 spin_lock(&nm_i->free_nid_list_lock);
1758 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1759 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1760 break;
1761 if (i->state == NID_ALLOC)
1762 continue;
1763 __del_from_free_nid_list(nm_i, i);
1764 kmem_cache_free(free_nid_slab, i);
1765 nm_i->fcnt--;
1766 nr_shrink--;
1767 }
1768 spin_unlock(&nm_i->free_nid_list_lock);
1769 mutex_unlock(&nm_i->build_lock);
1770
1771 return nr - nr_shrink;
1772}
1773
1774void recover_inline_xattr(struct inode *inode, struct page *page)
1775{
1776 void *src_addr, *dst_addr;
1777 size_t inline_size;
1778 struct page *ipage;
1779 struct f2fs_inode *ri;
1780
1781 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1782 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1783
1784 ri = F2FS_INODE(page);
1785 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1786 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1787 goto update_inode;
1788 }
1789
1790 dst_addr = inline_xattr_addr(ipage);
1791 src_addr = inline_xattr_addr(page);
1792 inline_size = inline_xattr_size(inode);
1793
1794 f2fs_wait_on_page_writeback(ipage, NODE, true);
1795 memcpy(dst_addr, src_addr, inline_size);
1796update_inode:
1797 update_inode(inode, ipage);
1798 f2fs_put_page(ipage, 1);
1799}
1800
1801void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1802{
1803 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1804 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1805 nid_t new_xnid = nid_of_node(page);
1806 struct node_info ni;
1807
1808 /* 1: invalidate the previous xattr nid */
1809 if (!prev_xnid)
1810 goto recover_xnid;
1811
1812 /* Deallocate node address */
1813 get_node_info(sbi, prev_xnid, &ni);
1814 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1815 invalidate_blocks(sbi, ni.blk_addr);
1816 dec_valid_node_count(sbi, inode);
1817 set_node_addr(sbi, &ni, NULL_ADDR, false);
1818
1819recover_xnid:
1820 /* 2: allocate new xattr nid */
1821 if (unlikely(!inc_valid_node_count(sbi, inode)))
1822 f2fs_bug_on(sbi, 1);
1823
1824 remove_free_nid(NM_I(sbi), new_xnid);
1825 get_node_info(sbi, new_xnid, &ni);
1826 ni.ino = inode->i_ino;
1827 set_node_addr(sbi, &ni, NEW_ADDR, false);
1828 F2FS_I(inode)->i_xattr_nid = new_xnid;
1829
1830 /* 3: update xattr blkaddr */
1831 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1832 set_node_addr(sbi, &ni, blkaddr, false);
1833
1834 update_inode_page(inode);
1835}
1836
1837int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1838{
1839 struct f2fs_inode *src, *dst;
1840 nid_t ino = ino_of_node(page);
1841 struct node_info old_ni, new_ni;
1842 struct page *ipage;
1843
1844 get_node_info(sbi, ino, &old_ni);
1845
1846 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1847 return -EINVAL;
1848
1849 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1850 if (!ipage)
1851 return -ENOMEM;
1852
1853 /* Should not use this inode from free nid list */
1854 remove_free_nid(NM_I(sbi), ino);
1855
1856 SetPageUptodate(ipage);
1857 fill_node_footer(ipage, ino, ino, 0, true);
1858
1859 src = F2FS_INODE(page);
1860 dst = F2FS_INODE(ipage);
1861
1862 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1863 dst->i_size = 0;
1864 dst->i_blocks = cpu_to_le64(1);
1865 dst->i_links = cpu_to_le32(1);
1866 dst->i_xattr_nid = 0;
1867 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1868
1869 new_ni = old_ni;
1870 new_ni.ino = ino;
1871
1872 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1873 WARN_ON(1);
1874 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1875 inc_valid_inode_count(sbi);
1876 set_page_dirty(ipage);
1877 f2fs_put_page(ipage, 1);
1878 return 0;
1879}
1880
1881int restore_node_summary(struct f2fs_sb_info *sbi,
1882 unsigned int segno, struct f2fs_summary_block *sum)
1883{
1884 struct f2fs_node *rn;
1885 struct f2fs_summary *sum_entry;
1886 block_t addr;
1887 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1888 int i, idx, last_offset, nrpages;
1889
1890 /* scan the node segment */
1891 last_offset = sbi->blocks_per_seg;
1892 addr = START_BLOCK(sbi, segno);
1893 sum_entry = &sum->entries[0];
1894
1895 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1896 nrpages = min(last_offset - i, bio_blocks);
1897
1898 /* readahead node pages */
1899 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1900
1901 for (idx = addr; idx < addr + nrpages; idx++) {
1902 struct page *page = get_tmp_page(sbi, idx);
1903
1904 rn = F2FS_NODE(page);
1905 sum_entry->nid = rn->footer.nid;
1906 sum_entry->version = 0;
1907 sum_entry->ofs_in_node = 0;
1908 sum_entry++;
1909 f2fs_put_page(page, 1);
1910 }
1911
1912 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1913 addr + nrpages);
1914 }
1915 return 0;
1916}
1917
1918static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1919{
1920 struct f2fs_nm_info *nm_i = NM_I(sbi);
1921 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1922 struct f2fs_journal *journal = curseg->journal;
1923 int i;
1924
1925 down_write(&curseg->journal_rwsem);
1926 for (i = 0; i < nats_in_cursum(journal); i++) {
1927 struct nat_entry *ne;
1928 struct f2fs_nat_entry raw_ne;
1929 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
1930
1931 raw_ne = nat_in_journal(journal, i);
1932
1933 ne = __lookup_nat_cache(nm_i, nid);
1934 if (!ne) {
1935 ne = grab_nat_entry(nm_i, nid);
1936 node_info_from_raw_nat(&ne->ni, &raw_ne);
1937 }
1938 __set_nat_cache_dirty(nm_i, ne);
1939 }
1940 update_nats_in_cursum(journal, -i);
1941 up_write(&curseg->journal_rwsem);
1942}
1943
1944static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1945 struct list_head *head, int max)
1946{
1947 struct nat_entry_set *cur;
1948
1949 if (nes->entry_cnt >= max)
1950 goto add_out;
1951
1952 list_for_each_entry(cur, head, set_list) {
1953 if (cur->entry_cnt >= nes->entry_cnt) {
1954 list_add(&nes->set_list, cur->set_list.prev);
1955 return;
1956 }
1957 }
1958add_out:
1959 list_add_tail(&nes->set_list, head);
1960}
1961
1962static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1963 struct nat_entry_set *set)
1964{
1965 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1966 struct f2fs_journal *journal = curseg->journal;
1967 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1968 bool to_journal = true;
1969 struct f2fs_nat_block *nat_blk;
1970 struct nat_entry *ne, *cur;
1971 struct page *page = NULL;
1972
1973 /*
1974 * there are two steps to flush nat entries:
1975 * #1, flush nat entries to journal in current hot data summary block.
1976 * #2, flush nat entries to nat page.
1977 */
1978 if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
1979 to_journal = false;
1980
1981 if (to_journal) {
1982 down_write(&curseg->journal_rwsem);
1983 } else {
1984 page = get_next_nat_page(sbi, start_nid);
1985 nat_blk = page_address(page);
1986 f2fs_bug_on(sbi, !nat_blk);
1987 }
1988
1989 /* flush dirty nats in nat entry set */
1990 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1991 struct f2fs_nat_entry *raw_ne;
1992 nid_t nid = nat_get_nid(ne);
1993 int offset;
1994
1995 if (nat_get_blkaddr(ne) == NEW_ADDR)
1996 continue;
1997
1998 if (to_journal) {
1999 offset = lookup_journal_in_cursum(journal,
2000 NAT_JOURNAL, nid, 1);
2001 f2fs_bug_on(sbi, offset < 0);
2002 raw_ne = &nat_in_journal(journal, offset);
2003 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2004 } else {
2005 raw_ne = &nat_blk->entries[nid - start_nid];
2006 }
2007 raw_nat_from_node_info(raw_ne, &ne->ni);
2008 nat_reset_flag(ne);
2009 __clear_nat_cache_dirty(NM_I(sbi), ne);
2010 if (nat_get_blkaddr(ne) == NULL_ADDR)
2011 add_free_nid(sbi, nid, false);
2012 }
2013
2014 if (to_journal)
2015 up_write(&curseg->journal_rwsem);
2016 else
2017 f2fs_put_page(page, 1);
2018
2019 f2fs_bug_on(sbi, set->entry_cnt);
2020
2021 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2022 kmem_cache_free(nat_entry_set_slab, set);
2023}
2024
2025/*
2026 * This function is called during the checkpointing process.
2027 */
2028void flush_nat_entries(struct f2fs_sb_info *sbi)
2029{
2030 struct f2fs_nm_info *nm_i = NM_I(sbi);
2031 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2032 struct f2fs_journal *journal = curseg->journal;
2033 struct nat_entry_set *setvec[SETVEC_SIZE];
2034 struct nat_entry_set *set, *tmp;
2035 unsigned int found;
2036 nid_t set_idx = 0;
2037 LIST_HEAD(sets);
2038
2039 if (!nm_i->dirty_nat_cnt)
2040 return;
2041
2042 down_write(&nm_i->nat_tree_lock);
2043
2044 /*
2045 * if there are no enough space in journal to store dirty nat
2046 * entries, remove all entries from journal and merge them
2047 * into nat entry set.
2048 */
2049 if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2050 remove_nats_in_journal(sbi);
2051
2052 while ((found = __gang_lookup_nat_set(nm_i,
2053 set_idx, SETVEC_SIZE, setvec))) {
2054 unsigned idx;
2055 set_idx = setvec[found - 1]->set + 1;
2056 for (idx = 0; idx < found; idx++)
2057 __adjust_nat_entry_set(setvec[idx], &sets,
2058 MAX_NAT_JENTRIES(journal));
2059 }
2060
2061 /* flush dirty nats in nat entry set */
2062 list_for_each_entry_safe(set, tmp, &sets, set_list)
2063 __flush_nat_entry_set(sbi, set);
2064
2065 up_write(&nm_i->nat_tree_lock);
2066
2067 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2068}
2069
2070static int init_node_manager(struct f2fs_sb_info *sbi)
2071{
2072 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2073 struct f2fs_nm_info *nm_i = NM_I(sbi);
2074 unsigned char *version_bitmap;
2075 unsigned int nat_segs, nat_blocks;
2076
2077 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2078
2079 /* segment_count_nat includes pair segment so divide to 2. */
2080 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2081 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2082
2083 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2084
2085 /* not used nids: 0, node, meta, (and root counted as valid node) */
2086 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2087 nm_i->fcnt = 0;
2088 nm_i->nat_cnt = 0;
2089 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2090 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2091 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2092
2093 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2094 INIT_LIST_HEAD(&nm_i->free_nid_list);
2095 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2096 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2097 INIT_LIST_HEAD(&nm_i->nat_entries);
2098
2099 mutex_init(&nm_i->build_lock);
2100 spin_lock_init(&nm_i->free_nid_list_lock);
2101 init_rwsem(&nm_i->nat_tree_lock);
2102
2103 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2104 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2105 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2106 if (!version_bitmap)
2107 return -EFAULT;
2108
2109 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2110 GFP_KERNEL);
2111 if (!nm_i->nat_bitmap)
2112 return -ENOMEM;
2113 return 0;
2114}
2115
2116int build_node_manager(struct f2fs_sb_info *sbi)
2117{
2118 int err;
2119
2120 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2121 if (!sbi->nm_info)
2122 return -ENOMEM;
2123
2124 err = init_node_manager(sbi);
2125 if (err)
2126 return err;
2127
2128 build_free_nids(sbi);
2129 return 0;
2130}
2131
2132void destroy_node_manager(struct f2fs_sb_info *sbi)
2133{
2134 struct f2fs_nm_info *nm_i = NM_I(sbi);
2135 struct free_nid *i, *next_i;
2136 struct nat_entry *natvec[NATVEC_SIZE];
2137 struct nat_entry_set *setvec[SETVEC_SIZE];
2138 nid_t nid = 0;
2139 unsigned int found;
2140
2141 if (!nm_i)
2142 return;
2143
2144 /* destroy free nid list */
2145 spin_lock(&nm_i->free_nid_list_lock);
2146 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2147 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2148 __del_from_free_nid_list(nm_i, i);
2149 nm_i->fcnt--;
2150 spin_unlock(&nm_i->free_nid_list_lock);
2151 kmem_cache_free(free_nid_slab, i);
2152 spin_lock(&nm_i->free_nid_list_lock);
2153 }
2154 f2fs_bug_on(sbi, nm_i->fcnt);
2155 spin_unlock(&nm_i->free_nid_list_lock);
2156
2157 /* destroy nat cache */
2158 down_write(&nm_i->nat_tree_lock);
2159 while ((found = __gang_lookup_nat_cache(nm_i,
2160 nid, NATVEC_SIZE, natvec))) {
2161 unsigned idx;
2162
2163 nid = nat_get_nid(natvec[found - 1]) + 1;
2164 for (idx = 0; idx < found; idx++)
2165 __del_from_nat_cache(nm_i, natvec[idx]);
2166 }
2167 f2fs_bug_on(sbi, nm_i->nat_cnt);
2168
2169 /* destroy nat set cache */
2170 nid = 0;
2171 while ((found = __gang_lookup_nat_set(nm_i,
2172 nid, SETVEC_SIZE, setvec))) {
2173 unsigned idx;
2174
2175 nid = setvec[found - 1]->set + 1;
2176 for (idx = 0; idx < found; idx++) {
2177 /* entry_cnt is not zero, when cp_error was occurred */
2178 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2179 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2180 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2181 }
2182 }
2183 up_write(&nm_i->nat_tree_lock);
2184
2185 kfree(nm_i->nat_bitmap);
2186 sbi->nm_info = NULL;
2187 kfree(nm_i);
2188}
2189
2190int __init create_node_manager_caches(void)
2191{
2192 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2193 sizeof(struct nat_entry));
2194 if (!nat_entry_slab)
2195 goto fail;
2196
2197 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2198 sizeof(struct free_nid));
2199 if (!free_nid_slab)
2200 goto destroy_nat_entry;
2201
2202 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2203 sizeof(struct nat_entry_set));
2204 if (!nat_entry_set_slab)
2205 goto destroy_free_nid;
2206 return 0;
2207
2208destroy_free_nid:
2209 kmem_cache_destroy(free_nid_slab);
2210destroy_nat_entry:
2211 kmem_cache_destroy(nat_entry_slab);
2212fail:
2213 return -ENOMEM;
2214}
2215
2216void destroy_node_manager_caches(void)
2217{
2218 kmem_cache_destroy(nat_entry_set_slab);
2219 kmem_cache_destroy(free_nid_slab);
2220 kmem_cache_destroy(nat_entry_slab);
2221}