f2fs: avoid issuing small bios due to several dirty node pages
[linux-3.10.git] / fs / f2fs / node.c
1 /*
2  * fs/f2fs/node.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
18
19 #include "f2fs.h"
20 #include "node.h"
21 #include "segment.h"
22
23 static struct kmem_cache *nat_entry_slab;
24 static struct kmem_cache *free_nid_slab;
25
26 static void clear_node_page_dirty(struct page *page)
27 {
28         struct address_space *mapping = page->mapping;
29         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
30         unsigned int long flags;
31
32         if (PageDirty(page)) {
33                 spin_lock_irqsave(&mapping->tree_lock, flags);
34                 radix_tree_tag_clear(&mapping->page_tree,
35                                 page_index(page),
36                                 PAGECACHE_TAG_DIRTY);
37                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
38
39                 clear_page_dirty_for_io(page);
40                 dec_page_count(sbi, F2FS_DIRTY_NODES);
41         }
42         ClearPageUptodate(page);
43 }
44
45 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
46 {
47         pgoff_t index = current_nat_addr(sbi, nid);
48         return get_meta_page(sbi, index);
49 }
50
51 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
52 {
53         struct page *src_page;
54         struct page *dst_page;
55         pgoff_t src_off;
56         pgoff_t dst_off;
57         void *src_addr;
58         void *dst_addr;
59         struct f2fs_nm_info *nm_i = NM_I(sbi);
60
61         src_off = current_nat_addr(sbi, nid);
62         dst_off = next_nat_addr(sbi, src_off);
63
64         /* get current nat block page with lock */
65         src_page = get_meta_page(sbi, src_off);
66
67         /* Dirty src_page means that it is already the new target NAT page. */
68         if (PageDirty(src_page))
69                 return src_page;
70
71         dst_page = grab_meta_page(sbi, dst_off);
72
73         src_addr = page_address(src_page);
74         dst_addr = page_address(dst_page);
75         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
76         set_page_dirty(dst_page);
77         f2fs_put_page(src_page, 1);
78
79         set_to_next_nat(nm_i, nid);
80
81         return dst_page;
82 }
83
84 /*
85  * Readahead NAT pages
86  */
87 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
88 {
89         struct address_space *mapping = sbi->meta_inode->i_mapping;
90         struct f2fs_nm_info *nm_i = NM_I(sbi);
91         struct page *page;
92         pgoff_t index;
93         int i;
94
95         for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
96                 if (nid >= nm_i->max_nid)
97                         nid = 0;
98                 index = current_nat_addr(sbi, nid);
99
100                 page = grab_cache_page(mapping, index);
101                 if (!page)
102                         continue;
103                 if (f2fs_readpage(sbi, page, index, READ)) {
104                         f2fs_put_page(page, 1);
105                         continue;
106                 }
107                 page_cache_release(page);
108         }
109 }
110
111 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
112 {
113         return radix_tree_lookup(&nm_i->nat_root, n);
114 }
115
116 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
117                 nid_t start, unsigned int nr, struct nat_entry **ep)
118 {
119         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
120 }
121
122 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
123 {
124         list_del(&e->list);
125         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
126         nm_i->nat_cnt--;
127         kmem_cache_free(nat_entry_slab, e);
128 }
129
130 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
131 {
132         struct f2fs_nm_info *nm_i = NM_I(sbi);
133         struct nat_entry *e;
134         int is_cp = 1;
135
136         read_lock(&nm_i->nat_tree_lock);
137         e = __lookup_nat_cache(nm_i, nid);
138         if (e && !e->checkpointed)
139                 is_cp = 0;
140         read_unlock(&nm_i->nat_tree_lock);
141         return is_cp;
142 }
143
144 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
145 {
146         struct nat_entry *new;
147
148         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
149         if (!new)
150                 return NULL;
151         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
152                 kmem_cache_free(nat_entry_slab, new);
153                 return NULL;
154         }
155         memset(new, 0, sizeof(struct nat_entry));
156         nat_set_nid(new, nid);
157         list_add_tail(&new->list, &nm_i->nat_entries);
158         nm_i->nat_cnt++;
159         return new;
160 }
161
162 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
163                                                 struct f2fs_nat_entry *ne)
164 {
165         struct nat_entry *e;
166 retry:
167         write_lock(&nm_i->nat_tree_lock);
168         e = __lookup_nat_cache(nm_i, nid);
169         if (!e) {
170                 e = grab_nat_entry(nm_i, nid);
171                 if (!e) {
172                         write_unlock(&nm_i->nat_tree_lock);
173                         goto retry;
174                 }
175                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
176                 nat_set_ino(e, le32_to_cpu(ne->ino));
177                 nat_set_version(e, ne->version);
178                 e->checkpointed = true;
179         }
180         write_unlock(&nm_i->nat_tree_lock);
181 }
182
183 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
184                         block_t new_blkaddr)
185 {
186         struct f2fs_nm_info *nm_i = NM_I(sbi);
187         struct nat_entry *e;
188 retry:
189         write_lock(&nm_i->nat_tree_lock);
190         e = __lookup_nat_cache(nm_i, ni->nid);
191         if (!e) {
192                 e = grab_nat_entry(nm_i, ni->nid);
193                 if (!e) {
194                         write_unlock(&nm_i->nat_tree_lock);
195                         goto retry;
196                 }
197                 e->ni = *ni;
198                 e->checkpointed = true;
199                 BUG_ON(ni->blk_addr == NEW_ADDR);
200         } else if (new_blkaddr == NEW_ADDR) {
201                 /*
202                  * when nid is reallocated,
203                  * previous nat entry can be remained in nat cache.
204                  * So, reinitialize it with new information.
205                  */
206                 e->ni = *ni;
207                 BUG_ON(ni->blk_addr != NULL_ADDR);
208         }
209
210         if (new_blkaddr == NEW_ADDR)
211                 e->checkpointed = false;
212
213         /* sanity check */
214         BUG_ON(nat_get_blkaddr(e) != ni->blk_addr);
215         BUG_ON(nat_get_blkaddr(e) == NULL_ADDR &&
216                         new_blkaddr == NULL_ADDR);
217         BUG_ON(nat_get_blkaddr(e) == NEW_ADDR &&
218                         new_blkaddr == NEW_ADDR);
219         BUG_ON(nat_get_blkaddr(e) != NEW_ADDR &&
220                         nat_get_blkaddr(e) != NULL_ADDR &&
221                         new_blkaddr == NEW_ADDR);
222
223         /* increament version no as node is removed */
224         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
225                 unsigned char version = nat_get_version(e);
226                 nat_set_version(e, inc_node_version(version));
227         }
228
229         /* change address */
230         nat_set_blkaddr(e, new_blkaddr);
231         __set_nat_cache_dirty(nm_i, e);
232         write_unlock(&nm_i->nat_tree_lock);
233 }
234
235 static int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
236 {
237         struct f2fs_nm_info *nm_i = NM_I(sbi);
238
239         if (nm_i->nat_cnt < 2 * NM_WOUT_THRESHOLD)
240                 return 0;
241
242         write_lock(&nm_i->nat_tree_lock);
243         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
244                 struct nat_entry *ne;
245                 ne = list_first_entry(&nm_i->nat_entries,
246                                         struct nat_entry, list);
247                 __del_from_nat_cache(nm_i, ne);
248                 nr_shrink--;
249         }
250         write_unlock(&nm_i->nat_tree_lock);
251         return nr_shrink;
252 }
253
254 /*
255  * This function returns always success
256  */
257 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
258 {
259         struct f2fs_nm_info *nm_i = NM_I(sbi);
260         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
261         struct f2fs_summary_block *sum = curseg->sum_blk;
262         nid_t start_nid = START_NID(nid);
263         struct f2fs_nat_block *nat_blk;
264         struct page *page = NULL;
265         struct f2fs_nat_entry ne;
266         struct nat_entry *e;
267         int i;
268
269         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
270         ni->nid = nid;
271
272         /* Check nat cache */
273         read_lock(&nm_i->nat_tree_lock);
274         e = __lookup_nat_cache(nm_i, nid);
275         if (e) {
276                 ni->ino = nat_get_ino(e);
277                 ni->blk_addr = nat_get_blkaddr(e);
278                 ni->version = nat_get_version(e);
279         }
280         read_unlock(&nm_i->nat_tree_lock);
281         if (e)
282                 return;
283
284         /* Check current segment summary */
285         mutex_lock(&curseg->curseg_mutex);
286         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
287         if (i >= 0) {
288                 ne = nat_in_journal(sum, i);
289                 node_info_from_raw_nat(ni, &ne);
290         }
291         mutex_unlock(&curseg->curseg_mutex);
292         if (i >= 0)
293                 goto cache;
294
295         /* Fill node_info from nat page */
296         page = get_current_nat_page(sbi, start_nid);
297         nat_blk = (struct f2fs_nat_block *)page_address(page);
298         ne = nat_blk->entries[nid - start_nid];
299         node_info_from_raw_nat(ni, &ne);
300         f2fs_put_page(page, 1);
301 cache:
302         /* cache nat entry */
303         cache_nat_entry(NM_I(sbi), nid, &ne);
304 }
305
306 /*
307  * The maximum depth is four.
308  * Offset[0] will have raw inode offset.
309  */
310 static int get_node_path(long block, int offset[4], unsigned int noffset[4])
311 {
312         const long direct_index = ADDRS_PER_INODE;
313         const long direct_blks = ADDRS_PER_BLOCK;
314         const long dptrs_per_blk = NIDS_PER_BLOCK;
315         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
316         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
317         int n = 0;
318         int level = 0;
319
320         noffset[0] = 0;
321
322         if (block < direct_index) {
323                 offset[n++] = block;
324                 level = 0;
325                 goto got;
326         }
327         block -= direct_index;
328         if (block < direct_blks) {
329                 offset[n++] = NODE_DIR1_BLOCK;
330                 noffset[n] = 1;
331                 offset[n++] = block;
332                 level = 1;
333                 goto got;
334         }
335         block -= direct_blks;
336         if (block < direct_blks) {
337                 offset[n++] = NODE_DIR2_BLOCK;
338                 noffset[n] = 2;
339                 offset[n++] = block;
340                 level = 1;
341                 goto got;
342         }
343         block -= direct_blks;
344         if (block < indirect_blks) {
345                 offset[n++] = NODE_IND1_BLOCK;
346                 noffset[n] = 3;
347                 offset[n++] = block / direct_blks;
348                 noffset[n] = 4 + offset[n - 1];
349                 offset[n++] = block % direct_blks;
350                 level = 2;
351                 goto got;
352         }
353         block -= indirect_blks;
354         if (block < indirect_blks) {
355                 offset[n++] = NODE_IND2_BLOCK;
356                 noffset[n] = 4 + dptrs_per_blk;
357                 offset[n++] = block / direct_blks;
358                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
359                 offset[n++] = block % direct_blks;
360                 level = 2;
361                 goto got;
362         }
363         block -= indirect_blks;
364         if (block < dindirect_blks) {
365                 offset[n++] = NODE_DIND_BLOCK;
366                 noffset[n] = 5 + (dptrs_per_blk * 2);
367                 offset[n++] = block / indirect_blks;
368                 noffset[n] = 6 + (dptrs_per_blk * 2) +
369                               offset[n - 1] * (dptrs_per_blk + 1);
370                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
371                 noffset[n] = 7 + (dptrs_per_blk * 2) +
372                               offset[n - 2] * (dptrs_per_blk + 1) +
373                               offset[n - 1];
374                 offset[n++] = block % direct_blks;
375                 level = 3;
376                 goto got;
377         } else {
378                 BUG();
379         }
380 got:
381         return level;
382 }
383
384 /*
385  * Caller should call f2fs_put_dnode(dn).
386  */
387 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int ro)
388 {
389         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
390         struct page *npage[4];
391         struct page *parent;
392         int offset[4];
393         unsigned int noffset[4];
394         nid_t nids[4];
395         int level, i;
396         int err = 0;
397
398         level = get_node_path(index, offset, noffset);
399
400         nids[0] = dn->inode->i_ino;
401         npage[0] = get_node_page(sbi, nids[0]);
402         if (IS_ERR(npage[0]))
403                 return PTR_ERR(npage[0]);
404
405         parent = npage[0];
406         nids[1] = get_nid(parent, offset[0], true);
407         dn->inode_page = npage[0];
408         dn->inode_page_locked = true;
409
410         /* get indirect or direct nodes */
411         for (i = 1; i <= level; i++) {
412                 bool done = false;
413
414                 if (!nids[i] && !ro) {
415                         mutex_lock_op(sbi, NODE_NEW);
416
417                         /* alloc new node */
418                         if (!alloc_nid(sbi, &(nids[i]))) {
419                                 mutex_unlock_op(sbi, NODE_NEW);
420                                 err = -ENOSPC;
421                                 goto release_pages;
422                         }
423
424                         dn->nid = nids[i];
425                         npage[i] = new_node_page(dn, noffset[i]);
426                         if (IS_ERR(npage[i])) {
427                                 alloc_nid_failed(sbi, nids[i]);
428                                 mutex_unlock_op(sbi, NODE_NEW);
429                                 err = PTR_ERR(npage[i]);
430                                 goto release_pages;
431                         }
432
433                         set_nid(parent, offset[i - 1], nids[i], i == 1);
434                         alloc_nid_done(sbi, nids[i]);
435                         mutex_unlock_op(sbi, NODE_NEW);
436                         done = true;
437                 } else if (ro && i == level && level > 1) {
438                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
439                         if (IS_ERR(npage[i])) {
440                                 err = PTR_ERR(npage[i]);
441                                 goto release_pages;
442                         }
443                         done = true;
444                 }
445                 if (i == 1) {
446                         dn->inode_page_locked = false;
447                         unlock_page(parent);
448                 } else {
449                         f2fs_put_page(parent, 1);
450                 }
451
452                 if (!done) {
453                         npage[i] = get_node_page(sbi, nids[i]);
454                         if (IS_ERR(npage[i])) {
455                                 err = PTR_ERR(npage[i]);
456                                 f2fs_put_page(npage[0], 0);
457                                 goto release_out;
458                         }
459                 }
460                 if (i < level) {
461                         parent = npage[i];
462                         nids[i + 1] = get_nid(parent, offset[i], false);
463                 }
464         }
465         dn->nid = nids[level];
466         dn->ofs_in_node = offset[level];
467         dn->node_page = npage[level];
468         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
469         return 0;
470
471 release_pages:
472         f2fs_put_page(parent, 1);
473         if (i > 1)
474                 f2fs_put_page(npage[0], 0);
475 release_out:
476         dn->inode_page = NULL;
477         dn->node_page = NULL;
478         return err;
479 }
480
481 static void truncate_node(struct dnode_of_data *dn)
482 {
483         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
484         struct node_info ni;
485
486         get_node_info(sbi, dn->nid, &ni);
487         if (dn->inode->i_blocks == 0) {
488                 BUG_ON(ni.blk_addr != NULL_ADDR);
489                 goto invalidate;
490         }
491         BUG_ON(ni.blk_addr == NULL_ADDR);
492
493         /* Deallocate node address */
494         invalidate_blocks(sbi, ni.blk_addr);
495         dec_valid_node_count(sbi, dn->inode, 1);
496         set_node_addr(sbi, &ni, NULL_ADDR);
497
498         if (dn->nid == dn->inode->i_ino) {
499                 remove_orphan_inode(sbi, dn->nid);
500                 dec_valid_inode_count(sbi);
501         } else {
502                 sync_inode_page(dn);
503         }
504 invalidate:
505         clear_node_page_dirty(dn->node_page);
506         F2FS_SET_SB_DIRT(sbi);
507
508         f2fs_put_page(dn->node_page, 1);
509         dn->node_page = NULL;
510 }
511
512 static int truncate_dnode(struct dnode_of_data *dn)
513 {
514         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
515         struct page *page;
516
517         if (dn->nid == 0)
518                 return 1;
519
520         /* get direct node */
521         page = get_node_page(sbi, dn->nid);
522         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
523                 return 1;
524         else if (IS_ERR(page))
525                 return PTR_ERR(page);
526
527         /* Make dnode_of_data for parameter */
528         dn->node_page = page;
529         dn->ofs_in_node = 0;
530         truncate_data_blocks(dn);
531         truncate_node(dn);
532         return 1;
533 }
534
535 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
536                                                 int ofs, int depth)
537 {
538         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
539         struct dnode_of_data rdn = *dn;
540         struct page *page;
541         struct f2fs_node *rn;
542         nid_t child_nid;
543         unsigned int child_nofs;
544         int freed = 0;
545         int i, ret;
546
547         if (dn->nid == 0)
548                 return NIDS_PER_BLOCK + 1;
549
550         page = get_node_page(sbi, dn->nid);
551         if (IS_ERR(page))
552                 return PTR_ERR(page);
553
554         rn = (struct f2fs_node *)page_address(page);
555         if (depth < 3) {
556                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
557                         child_nid = le32_to_cpu(rn->in.nid[i]);
558                         if (child_nid == 0)
559                                 continue;
560                         rdn.nid = child_nid;
561                         ret = truncate_dnode(&rdn);
562                         if (ret < 0)
563                                 goto out_err;
564                         set_nid(page, i, 0, false);
565                 }
566         } else {
567                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
568                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
569                         child_nid = le32_to_cpu(rn->in.nid[i]);
570                         if (child_nid == 0) {
571                                 child_nofs += NIDS_PER_BLOCK + 1;
572                                 continue;
573                         }
574                         rdn.nid = child_nid;
575                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
576                         if (ret == (NIDS_PER_BLOCK + 1)) {
577                                 set_nid(page, i, 0, false);
578                                 child_nofs += ret;
579                         } else if (ret < 0 && ret != -ENOENT) {
580                                 goto out_err;
581                         }
582                 }
583                 freed = child_nofs;
584         }
585
586         if (!ofs) {
587                 /* remove current indirect node */
588                 dn->node_page = page;
589                 truncate_node(dn);
590                 freed++;
591         } else {
592                 f2fs_put_page(page, 1);
593         }
594         return freed;
595
596 out_err:
597         f2fs_put_page(page, 1);
598         return ret;
599 }
600
601 static int truncate_partial_nodes(struct dnode_of_data *dn,
602                         struct f2fs_inode *ri, int *offset, int depth)
603 {
604         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
605         struct page *pages[2];
606         nid_t nid[3];
607         nid_t child_nid;
608         int err = 0;
609         int i;
610         int idx = depth - 2;
611
612         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
613         if (!nid[0])
614                 return 0;
615
616         /* get indirect nodes in the path */
617         for (i = 0; i < depth - 1; i++) {
618                 /* refernece count'll be increased */
619                 pages[i] = get_node_page(sbi, nid[i]);
620                 if (IS_ERR(pages[i])) {
621                         depth = i + 1;
622                         err = PTR_ERR(pages[i]);
623                         goto fail;
624                 }
625                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
626         }
627
628         /* free direct nodes linked to a partial indirect node */
629         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
630                 child_nid = get_nid(pages[idx], i, false);
631                 if (!child_nid)
632                         continue;
633                 dn->nid = child_nid;
634                 err = truncate_dnode(dn);
635                 if (err < 0)
636                         goto fail;
637                 set_nid(pages[idx], i, 0, false);
638         }
639
640         if (offset[depth - 1] == 0) {
641                 dn->node_page = pages[idx];
642                 dn->nid = nid[idx];
643                 truncate_node(dn);
644         } else {
645                 f2fs_put_page(pages[idx], 1);
646         }
647         offset[idx]++;
648         offset[depth - 1] = 0;
649 fail:
650         for (i = depth - 3; i >= 0; i--)
651                 f2fs_put_page(pages[i], 1);
652         return err;
653 }
654
655 /*
656  * All the block addresses of data and nodes should be nullified.
657  */
658 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
659 {
660         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
661         int err = 0, cont = 1;
662         int level, offset[4], noffset[4];
663         unsigned int nofs;
664         struct f2fs_node *rn;
665         struct dnode_of_data dn;
666         struct page *page;
667
668         level = get_node_path(from, offset, noffset);
669
670         page = get_node_page(sbi, inode->i_ino);
671         if (IS_ERR(page))
672                 return PTR_ERR(page);
673
674         set_new_dnode(&dn, inode, page, NULL, 0);
675         unlock_page(page);
676
677         rn = page_address(page);
678         switch (level) {
679         case 0:
680         case 1:
681                 nofs = noffset[1];
682                 break;
683         case 2:
684                 nofs = noffset[1];
685                 if (!offset[level - 1])
686                         goto skip_partial;
687                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
688                 if (err < 0 && err != -ENOENT)
689                         goto fail;
690                 nofs += 1 + NIDS_PER_BLOCK;
691                 break;
692         case 3:
693                 nofs = 5 + 2 * NIDS_PER_BLOCK;
694                 if (!offset[level - 1])
695                         goto skip_partial;
696                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
697                 if (err < 0 && err != -ENOENT)
698                         goto fail;
699                 break;
700         default:
701                 BUG();
702         }
703
704 skip_partial:
705         while (cont) {
706                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
707                 switch (offset[0]) {
708                 case NODE_DIR1_BLOCK:
709                 case NODE_DIR2_BLOCK:
710                         err = truncate_dnode(&dn);
711                         break;
712
713                 case NODE_IND1_BLOCK:
714                 case NODE_IND2_BLOCK:
715                         err = truncate_nodes(&dn, nofs, offset[1], 2);
716                         break;
717
718                 case NODE_DIND_BLOCK:
719                         err = truncate_nodes(&dn, nofs, offset[1], 3);
720                         cont = 0;
721                         break;
722
723                 default:
724                         BUG();
725                 }
726                 if (err < 0 && err != -ENOENT)
727                         goto fail;
728                 if (offset[1] == 0 &&
729                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
730                         lock_page(page);
731                         wait_on_page_writeback(page);
732                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
733                         set_page_dirty(page);
734                         unlock_page(page);
735                 }
736                 offset[1] = 0;
737                 offset[0]++;
738                 nofs += err;
739         }
740 fail:
741         f2fs_put_page(page, 0);
742         return err > 0 ? 0 : err;
743 }
744
745 int remove_inode_page(struct inode *inode)
746 {
747         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
748         struct page *page;
749         nid_t ino = inode->i_ino;
750         struct dnode_of_data dn;
751
752         mutex_lock_op(sbi, NODE_TRUNC);
753         page = get_node_page(sbi, ino);
754         if (IS_ERR(page)) {
755                 mutex_unlock_op(sbi, NODE_TRUNC);
756                 return PTR_ERR(page);
757         }
758
759         if (F2FS_I(inode)->i_xattr_nid) {
760                 nid_t nid = F2FS_I(inode)->i_xattr_nid;
761                 struct page *npage = get_node_page(sbi, nid);
762
763                 if (IS_ERR(npage)) {
764                         mutex_unlock_op(sbi, NODE_TRUNC);
765                         return PTR_ERR(npage);
766                 }
767
768                 F2FS_I(inode)->i_xattr_nid = 0;
769                 set_new_dnode(&dn, inode, page, npage, nid);
770                 dn.inode_page_locked = 1;
771                 truncate_node(&dn);
772         }
773
774         /* 0 is possible, after f2fs_new_inode() is failed */
775         BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
776         set_new_dnode(&dn, inode, page, page, ino);
777         truncate_node(&dn);
778
779         mutex_unlock_op(sbi, NODE_TRUNC);
780         return 0;
781 }
782
783 int new_inode_page(struct inode *inode, struct dentry *dentry)
784 {
785         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
786         struct page *page;
787         struct dnode_of_data dn;
788
789         /* allocate inode page for new inode */
790         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
791         mutex_lock_op(sbi, NODE_NEW);
792         page = new_node_page(&dn, 0);
793         init_dent_inode(dentry, page);
794         mutex_unlock_op(sbi, NODE_NEW);
795         if (IS_ERR(page))
796                 return PTR_ERR(page);
797         f2fs_put_page(page, 1);
798         return 0;
799 }
800
801 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
802 {
803         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
804         struct address_space *mapping = sbi->node_inode->i_mapping;
805         struct node_info old_ni, new_ni;
806         struct page *page;
807         int err;
808
809         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
810                 return ERR_PTR(-EPERM);
811
812         page = grab_cache_page(mapping, dn->nid);
813         if (!page)
814                 return ERR_PTR(-ENOMEM);
815
816         get_node_info(sbi, dn->nid, &old_ni);
817
818         SetPageUptodate(page);
819         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
820
821         /* Reinitialize old_ni with new node page */
822         BUG_ON(old_ni.blk_addr != NULL_ADDR);
823         new_ni = old_ni;
824         new_ni.ino = dn->inode->i_ino;
825
826         if (!inc_valid_node_count(sbi, dn->inode, 1)) {
827                 err = -ENOSPC;
828                 goto fail;
829         }
830         set_node_addr(sbi, &new_ni, NEW_ADDR);
831         set_cold_node(dn->inode, page);
832
833         dn->node_page = page;
834         sync_inode_page(dn);
835         set_page_dirty(page);
836         if (ofs == 0)
837                 inc_valid_inode_count(sbi);
838
839         return page;
840
841 fail:
842         clear_node_page_dirty(page);
843         f2fs_put_page(page, 1);
844         return ERR_PTR(err);
845 }
846
847 static int read_node_page(struct page *page, int type)
848 {
849         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
850         struct node_info ni;
851
852         get_node_info(sbi, page->index, &ni);
853
854         if (ni.blk_addr == NULL_ADDR)
855                 return -ENOENT;
856         return f2fs_readpage(sbi, page, ni.blk_addr, type);
857 }
858
859 /*
860  * Readahead a node page
861  */
862 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
863 {
864         struct address_space *mapping = sbi->node_inode->i_mapping;
865         struct page *apage;
866
867         apage = find_get_page(mapping, nid);
868         if (apage && PageUptodate(apage))
869                 goto release_out;
870         f2fs_put_page(apage, 0);
871
872         apage = grab_cache_page(mapping, nid);
873         if (!apage)
874                 return;
875
876         if (read_node_page(apage, READA))
877                 goto unlock_out;
878
879         page_cache_release(apage);
880         return;
881
882 unlock_out:
883         unlock_page(apage);
884 release_out:
885         page_cache_release(apage);
886 }
887
888 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
889 {
890         int err;
891         struct page *page;
892         struct address_space *mapping = sbi->node_inode->i_mapping;
893
894         page = grab_cache_page(mapping, nid);
895         if (!page)
896                 return ERR_PTR(-ENOMEM);
897
898         err = read_node_page(page, READ_SYNC);
899         if (err) {
900                 f2fs_put_page(page, 1);
901                 return ERR_PTR(err);
902         }
903
904         BUG_ON(nid != nid_of_node(page));
905         mark_page_accessed(page);
906         return page;
907 }
908
909 /*
910  * Return a locked page for the desired node page.
911  * And, readahead MAX_RA_NODE number of node pages.
912  */
913 struct page *get_node_page_ra(struct page *parent, int start)
914 {
915         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
916         struct address_space *mapping = sbi->node_inode->i_mapping;
917         int i, end;
918         int err = 0;
919         nid_t nid;
920         struct page *page;
921
922         /* First, try getting the desired direct node. */
923         nid = get_nid(parent, start, false);
924         if (!nid)
925                 return ERR_PTR(-ENOENT);
926
927         page = find_get_page(mapping, nid);
928         if (page && PageUptodate(page))
929                 goto page_hit;
930         f2fs_put_page(page, 0);
931
932 repeat:
933         page = grab_cache_page(mapping, nid);
934         if (!page)
935                 return ERR_PTR(-ENOMEM);
936
937         err = read_node_page(page, READA);
938         if (err) {
939                 f2fs_put_page(page, 1);
940                 return ERR_PTR(err);
941         }
942
943         /* Then, try readahead for siblings of the desired node */
944         end = start + MAX_RA_NODE;
945         end = min(end, NIDS_PER_BLOCK);
946         for (i = start + 1; i < end; i++) {
947                 nid = get_nid(parent, i, false);
948                 if (!nid)
949                         continue;
950                 ra_node_page(sbi, nid);
951         }
952
953 page_hit:
954         lock_page(page);
955         if (PageError(page)) {
956                 f2fs_put_page(page, 1);
957                 return ERR_PTR(-EIO);
958         }
959
960         /* Has the page been truncated? */
961         if (page->mapping != mapping) {
962                 f2fs_put_page(page, 1);
963                 goto repeat;
964         }
965         return page;
966 }
967
968 void sync_inode_page(struct dnode_of_data *dn)
969 {
970         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
971                 update_inode(dn->inode, dn->node_page);
972         } else if (dn->inode_page) {
973                 if (!dn->inode_page_locked)
974                         lock_page(dn->inode_page);
975                 update_inode(dn->inode, dn->inode_page);
976                 if (!dn->inode_page_locked)
977                         unlock_page(dn->inode_page);
978         } else {
979                 f2fs_write_inode(dn->inode, NULL);
980         }
981 }
982
983 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
984                                         struct writeback_control *wbc)
985 {
986         struct address_space *mapping = sbi->node_inode->i_mapping;
987         pgoff_t index, end;
988         struct pagevec pvec;
989         int step = ino ? 2 : 0;
990         int nwritten = 0, wrote = 0;
991
992         pagevec_init(&pvec, 0);
993
994 next_step:
995         index = 0;
996         end = LONG_MAX;
997
998         while (index <= end) {
999                 int i, nr_pages;
1000                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1001                                 PAGECACHE_TAG_DIRTY,
1002                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1003                 if (nr_pages == 0)
1004                         break;
1005
1006                 for (i = 0; i < nr_pages; i++) {
1007                         struct page *page = pvec.pages[i];
1008
1009                         /*
1010                          * flushing sequence with step:
1011                          * 0. indirect nodes
1012                          * 1. dentry dnodes
1013                          * 2. file dnodes
1014                          */
1015                         if (step == 0 && IS_DNODE(page))
1016                                 continue;
1017                         if (step == 1 && (!IS_DNODE(page) ||
1018                                                 is_cold_node(page)))
1019                                 continue;
1020                         if (step == 2 && (!IS_DNODE(page) ||
1021                                                 !is_cold_node(page)))
1022                                 continue;
1023
1024                         /*
1025                          * If an fsync mode,
1026                          * we should not skip writing node pages.
1027                          */
1028                         if (ino && ino_of_node(page) == ino)
1029                                 lock_page(page);
1030                         else if (!trylock_page(page))
1031                                 continue;
1032
1033                         if (unlikely(page->mapping != mapping)) {
1034 continue_unlock:
1035                                 unlock_page(page);
1036                                 continue;
1037                         }
1038                         if (ino && ino_of_node(page) != ino)
1039                                 goto continue_unlock;
1040
1041                         if (!PageDirty(page)) {
1042                                 /* someone wrote it for us */
1043                                 goto continue_unlock;
1044                         }
1045
1046                         if (!clear_page_dirty_for_io(page))
1047                                 goto continue_unlock;
1048
1049                         /* called by fsync() */
1050                         if (ino && IS_DNODE(page)) {
1051                                 int mark = !is_checkpointed_node(sbi, ino);
1052                                 set_fsync_mark(page, 1);
1053                                 if (IS_INODE(page))
1054                                         set_dentry_mark(page, mark);
1055                                 nwritten++;
1056                         } else {
1057                                 set_fsync_mark(page, 0);
1058                                 set_dentry_mark(page, 0);
1059                         }
1060                         mapping->a_ops->writepage(page, wbc);
1061                         wrote++;
1062
1063                         if (--wbc->nr_to_write == 0)
1064                                 break;
1065                 }
1066                 pagevec_release(&pvec);
1067                 cond_resched();
1068
1069                 if (wbc->nr_to_write == 0) {
1070                         step = 2;
1071                         break;
1072                 }
1073         }
1074
1075         if (step < 2) {
1076                 step++;
1077                 goto next_step;
1078         }
1079
1080         if (wrote)
1081                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1082
1083         return nwritten;
1084 }
1085
1086 static int f2fs_write_node_page(struct page *page,
1087                                 struct writeback_control *wbc)
1088 {
1089         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1090         nid_t nid;
1091         block_t new_addr;
1092         struct node_info ni;
1093
1094         if (wbc->for_reclaim) {
1095                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1096                 wbc->pages_skipped++;
1097                 set_page_dirty(page);
1098                 return AOP_WRITEPAGE_ACTIVATE;
1099         }
1100
1101         wait_on_page_writeback(page);
1102
1103         mutex_lock_op(sbi, NODE_WRITE);
1104
1105         /* get old block addr of this node page */
1106         nid = nid_of_node(page);
1107         BUG_ON(page->index != nid);
1108
1109         get_node_info(sbi, nid, &ni);
1110
1111         /* This page is already truncated */
1112         if (ni.blk_addr == NULL_ADDR)
1113                 return 0;
1114
1115         set_page_writeback(page);
1116
1117         /* insert node offset */
1118         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1119         set_node_addr(sbi, &ni, new_addr);
1120         dec_page_count(sbi, F2FS_DIRTY_NODES);
1121
1122         mutex_unlock_op(sbi, NODE_WRITE);
1123         unlock_page(page);
1124         return 0;
1125 }
1126
1127 /*
1128  * It is very important to gather dirty pages and write at once, so that we can
1129  * submit a big bio without interfering other data writes.
1130  * Be default, 512 pages (2MB), a segment size, is quite reasonable.
1131  */
1132 #define COLLECT_DIRTY_NODES     512
1133 static int f2fs_write_node_pages(struct address_space *mapping,
1134                             struct writeback_control *wbc)
1135 {
1136         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1137         struct block_device *bdev = sbi->sb->s_bdev;
1138         long nr_to_write = wbc->nr_to_write;
1139
1140         /* First check balancing cached NAT entries */
1141         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
1142                 write_checkpoint(sbi, false, false);
1143                 return 0;
1144         }
1145
1146         /* collect a number of dirty node pages and write together */
1147         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1148                 return 0;
1149
1150         /* if mounting is failed, skip writing node pages */
1151         wbc->nr_to_write = bio_get_nr_vecs(bdev);
1152         sync_node_pages(sbi, 0, wbc);
1153         wbc->nr_to_write = nr_to_write -
1154                 (bio_get_nr_vecs(bdev) - wbc->nr_to_write);
1155         return 0;
1156 }
1157
1158 static int f2fs_set_node_page_dirty(struct page *page)
1159 {
1160         struct address_space *mapping = page->mapping;
1161         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1162
1163         SetPageUptodate(page);
1164         if (!PageDirty(page)) {
1165                 __set_page_dirty_nobuffers(page);
1166                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1167                 SetPagePrivate(page);
1168                 return 1;
1169         }
1170         return 0;
1171 }
1172
1173 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
1174 {
1175         struct inode *inode = page->mapping->host;
1176         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1177         if (PageDirty(page))
1178                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1179         ClearPagePrivate(page);
1180 }
1181
1182 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1183 {
1184         ClearPagePrivate(page);
1185         return 0;
1186 }
1187
1188 /*
1189  * Structure of the f2fs node operations
1190  */
1191 const struct address_space_operations f2fs_node_aops = {
1192         .writepage      = f2fs_write_node_page,
1193         .writepages     = f2fs_write_node_pages,
1194         .set_page_dirty = f2fs_set_node_page_dirty,
1195         .invalidatepage = f2fs_invalidate_node_page,
1196         .releasepage    = f2fs_release_node_page,
1197 };
1198
1199 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1200 {
1201         struct list_head *this;
1202         struct free_nid *i = NULL;
1203         list_for_each(this, head) {
1204                 i = list_entry(this, struct free_nid, list);
1205                 if (i->nid == n)
1206                         break;
1207                 i = NULL;
1208         }
1209         return i;
1210 }
1211
1212 static void __del_from_free_nid_list(struct free_nid *i)
1213 {
1214         list_del(&i->list);
1215         kmem_cache_free(free_nid_slab, i);
1216 }
1217
1218 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1219 {
1220         struct free_nid *i;
1221
1222         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1223                 return 0;
1224 retry:
1225         i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1226         if (!i) {
1227                 cond_resched();
1228                 goto retry;
1229         }
1230         i->nid = nid;
1231         i->state = NID_NEW;
1232
1233         spin_lock(&nm_i->free_nid_list_lock);
1234         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1235                 spin_unlock(&nm_i->free_nid_list_lock);
1236                 kmem_cache_free(free_nid_slab, i);
1237                 return 0;
1238         }
1239         list_add_tail(&i->list, &nm_i->free_nid_list);
1240         nm_i->fcnt++;
1241         spin_unlock(&nm_i->free_nid_list_lock);
1242         return 1;
1243 }
1244
1245 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1246 {
1247         struct free_nid *i;
1248         spin_lock(&nm_i->free_nid_list_lock);
1249         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1250         if (i && i->state == NID_NEW) {
1251                 __del_from_free_nid_list(i);
1252                 nm_i->fcnt--;
1253         }
1254         spin_unlock(&nm_i->free_nid_list_lock);
1255 }
1256
1257 static int scan_nat_page(struct f2fs_nm_info *nm_i,
1258                         struct page *nat_page, nid_t start_nid)
1259 {
1260         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1261         block_t blk_addr;
1262         int fcnt = 0;
1263         int i;
1264
1265         /* 0 nid should not be used */
1266         if (start_nid == 0)
1267                 ++start_nid;
1268
1269         i = start_nid % NAT_ENTRY_PER_BLOCK;
1270
1271         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1272                 blk_addr  = le32_to_cpu(nat_blk->entries[i].block_addr);
1273                 BUG_ON(blk_addr == NEW_ADDR);
1274                 if (blk_addr == NULL_ADDR)
1275                         fcnt += add_free_nid(nm_i, start_nid);
1276         }
1277         return fcnt;
1278 }
1279
1280 static void build_free_nids(struct f2fs_sb_info *sbi)
1281 {
1282         struct free_nid *fnid, *next_fnid;
1283         struct f2fs_nm_info *nm_i = NM_I(sbi);
1284         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1285         struct f2fs_summary_block *sum = curseg->sum_blk;
1286         nid_t nid = 0;
1287         bool is_cycled = false;
1288         int fcnt = 0;
1289         int i;
1290
1291         nid = nm_i->next_scan_nid;
1292         nm_i->init_scan_nid = nid;
1293
1294         ra_nat_pages(sbi, nid);
1295
1296         while (1) {
1297                 struct page *page = get_current_nat_page(sbi, nid);
1298
1299                 fcnt += scan_nat_page(nm_i, page, nid);
1300                 f2fs_put_page(page, 1);
1301
1302                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1303
1304                 if (nid >= nm_i->max_nid) {
1305                         nid = 0;
1306                         is_cycled = true;
1307                 }
1308                 if (fcnt > MAX_FREE_NIDS)
1309                         break;
1310                 if (is_cycled && nm_i->init_scan_nid <= nid)
1311                         break;
1312         }
1313
1314         nm_i->next_scan_nid = nid;
1315
1316         /* find free nids from current sum_pages */
1317         mutex_lock(&curseg->curseg_mutex);
1318         for (i = 0; i < nats_in_cursum(sum); i++) {
1319                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1320                 nid = le32_to_cpu(nid_in_journal(sum, i));
1321                 if (addr == NULL_ADDR)
1322                         add_free_nid(nm_i, nid);
1323                 else
1324                         remove_free_nid(nm_i, nid);
1325         }
1326         mutex_unlock(&curseg->curseg_mutex);
1327
1328         /* remove the free nids from current allocated nids */
1329         list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) {
1330                 struct nat_entry *ne;
1331
1332                 read_lock(&nm_i->nat_tree_lock);
1333                 ne = __lookup_nat_cache(nm_i, fnid->nid);
1334                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1335                         remove_free_nid(nm_i, fnid->nid);
1336                 read_unlock(&nm_i->nat_tree_lock);
1337         }
1338 }
1339
1340 /*
1341  * If this function returns success, caller can obtain a new nid
1342  * from second parameter of this function.
1343  * The returned nid could be used ino as well as nid when inode is created.
1344  */
1345 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1346 {
1347         struct f2fs_nm_info *nm_i = NM_I(sbi);
1348         struct free_nid *i = NULL;
1349         struct list_head *this;
1350 retry:
1351         mutex_lock(&nm_i->build_lock);
1352         if (!nm_i->fcnt) {
1353                 /* scan NAT in order to build free nid list */
1354                 build_free_nids(sbi);
1355                 if (!nm_i->fcnt) {
1356                         mutex_unlock(&nm_i->build_lock);
1357                         return false;
1358                 }
1359         }
1360         mutex_unlock(&nm_i->build_lock);
1361
1362         /*
1363          * We check fcnt again since previous check is racy as
1364          * we didn't hold free_nid_list_lock. So other thread
1365          * could consume all of free nids.
1366          */
1367         spin_lock(&nm_i->free_nid_list_lock);
1368         if (!nm_i->fcnt) {
1369                 spin_unlock(&nm_i->free_nid_list_lock);
1370                 goto retry;
1371         }
1372
1373         BUG_ON(list_empty(&nm_i->free_nid_list));
1374         list_for_each(this, &nm_i->free_nid_list) {
1375                 i = list_entry(this, struct free_nid, list);
1376                 if (i->state == NID_NEW)
1377                         break;
1378         }
1379
1380         BUG_ON(i->state != NID_NEW);
1381         *nid = i->nid;
1382         i->state = NID_ALLOC;
1383         nm_i->fcnt--;
1384         spin_unlock(&nm_i->free_nid_list_lock);
1385         return true;
1386 }
1387
1388 /*
1389  * alloc_nid() should be called prior to this function.
1390  */
1391 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1392 {
1393         struct f2fs_nm_info *nm_i = NM_I(sbi);
1394         struct free_nid *i;
1395
1396         spin_lock(&nm_i->free_nid_list_lock);
1397         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1398         if (i) {
1399                 BUG_ON(i->state != NID_ALLOC);
1400                 __del_from_free_nid_list(i);
1401         }
1402         spin_unlock(&nm_i->free_nid_list_lock);
1403 }
1404
1405 /*
1406  * alloc_nid() should be called prior to this function.
1407  */
1408 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1409 {
1410         alloc_nid_done(sbi, nid);
1411         add_free_nid(NM_I(sbi), nid);
1412 }
1413
1414 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1415                 struct f2fs_summary *sum, struct node_info *ni,
1416                 block_t new_blkaddr)
1417 {
1418         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1419         set_node_addr(sbi, ni, new_blkaddr);
1420         clear_node_page_dirty(page);
1421 }
1422
1423 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1424 {
1425         struct address_space *mapping = sbi->node_inode->i_mapping;
1426         struct f2fs_node *src, *dst;
1427         nid_t ino = ino_of_node(page);
1428         struct node_info old_ni, new_ni;
1429         struct page *ipage;
1430
1431         ipage = grab_cache_page(mapping, ino);
1432         if (!ipage)
1433                 return -ENOMEM;
1434
1435         /* Should not use this inode  from free nid list */
1436         remove_free_nid(NM_I(sbi), ino);
1437
1438         get_node_info(sbi, ino, &old_ni);
1439         SetPageUptodate(ipage);
1440         fill_node_footer(ipage, ino, ino, 0, true);
1441
1442         src = (struct f2fs_node *)page_address(page);
1443         dst = (struct f2fs_node *)page_address(ipage);
1444
1445         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1446         dst->i.i_size = 0;
1447         dst->i.i_blocks = cpu_to_le64(1);
1448         dst->i.i_links = cpu_to_le32(1);
1449         dst->i.i_xattr_nid = 0;
1450
1451         new_ni = old_ni;
1452         new_ni.ino = ino;
1453
1454         set_node_addr(sbi, &new_ni, NEW_ADDR);
1455         inc_valid_inode_count(sbi);
1456
1457         f2fs_put_page(ipage, 1);
1458         return 0;
1459 }
1460
1461 int restore_node_summary(struct f2fs_sb_info *sbi,
1462                         unsigned int segno, struct f2fs_summary_block *sum)
1463 {
1464         struct f2fs_node *rn;
1465         struct f2fs_summary *sum_entry;
1466         struct page *page;
1467         block_t addr;
1468         int i, last_offset;
1469
1470         /* alloc temporal page for read node */
1471         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1472         if (IS_ERR(page))
1473                 return PTR_ERR(page);
1474         lock_page(page);
1475
1476         /* scan the node segment */
1477         last_offset = sbi->blocks_per_seg;
1478         addr = START_BLOCK(sbi, segno);
1479         sum_entry = &sum->entries[0];
1480
1481         for (i = 0; i < last_offset; i++, sum_entry++) {
1482                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1483                         goto out;
1484
1485                 rn = (struct f2fs_node *)page_address(page);
1486                 sum_entry->nid = rn->footer.nid;
1487                 sum_entry->version = 0;
1488                 sum_entry->ofs_in_node = 0;
1489                 addr++;
1490
1491                 /*
1492                  * In order to read next node page,
1493                  * we must clear PageUptodate flag.
1494                  */
1495                 ClearPageUptodate(page);
1496         }
1497 out:
1498         unlock_page(page);
1499         __free_pages(page, 0);
1500         return 0;
1501 }
1502
1503 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1504 {
1505         struct f2fs_nm_info *nm_i = NM_I(sbi);
1506         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1507         struct f2fs_summary_block *sum = curseg->sum_blk;
1508         int i;
1509
1510         mutex_lock(&curseg->curseg_mutex);
1511
1512         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1513                 mutex_unlock(&curseg->curseg_mutex);
1514                 return false;
1515         }
1516
1517         for (i = 0; i < nats_in_cursum(sum); i++) {
1518                 struct nat_entry *ne;
1519                 struct f2fs_nat_entry raw_ne;
1520                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1521
1522                 raw_ne = nat_in_journal(sum, i);
1523 retry:
1524                 write_lock(&nm_i->nat_tree_lock);
1525                 ne = __lookup_nat_cache(nm_i, nid);
1526                 if (ne) {
1527                         __set_nat_cache_dirty(nm_i, ne);
1528                         write_unlock(&nm_i->nat_tree_lock);
1529                         continue;
1530                 }
1531                 ne = grab_nat_entry(nm_i, nid);
1532                 if (!ne) {
1533                         write_unlock(&nm_i->nat_tree_lock);
1534                         goto retry;
1535                 }
1536                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1537                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1538                 nat_set_version(ne, raw_ne.version);
1539                 __set_nat_cache_dirty(nm_i, ne);
1540                 write_unlock(&nm_i->nat_tree_lock);
1541         }
1542         update_nats_in_cursum(sum, -i);
1543         mutex_unlock(&curseg->curseg_mutex);
1544         return true;
1545 }
1546
1547 /*
1548  * This function is called during the checkpointing process.
1549  */
1550 void flush_nat_entries(struct f2fs_sb_info *sbi)
1551 {
1552         struct f2fs_nm_info *nm_i = NM_I(sbi);
1553         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1554         struct f2fs_summary_block *sum = curseg->sum_blk;
1555         struct list_head *cur, *n;
1556         struct page *page = NULL;
1557         struct f2fs_nat_block *nat_blk = NULL;
1558         nid_t start_nid = 0, end_nid = 0;
1559         bool flushed;
1560
1561         flushed = flush_nats_in_journal(sbi);
1562
1563         if (!flushed)
1564                 mutex_lock(&curseg->curseg_mutex);
1565
1566         /* 1) flush dirty nat caches */
1567         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1568                 struct nat_entry *ne;
1569                 nid_t nid;
1570                 struct f2fs_nat_entry raw_ne;
1571                 int offset = -1;
1572                 block_t new_blkaddr;
1573
1574                 ne = list_entry(cur, struct nat_entry, list);
1575                 nid = nat_get_nid(ne);
1576
1577                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1578                         continue;
1579                 if (flushed)
1580                         goto to_nat_page;
1581
1582                 /* if there is room for nat enries in curseg->sumpage */
1583                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1584                 if (offset >= 0) {
1585                         raw_ne = nat_in_journal(sum, offset);
1586                         goto flush_now;
1587                 }
1588 to_nat_page:
1589                 if (!page || (start_nid > nid || nid > end_nid)) {
1590                         if (page) {
1591                                 f2fs_put_page(page, 1);
1592                                 page = NULL;
1593                         }
1594                         start_nid = START_NID(nid);
1595                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1596
1597                         /*
1598                          * get nat block with dirty flag, increased reference
1599                          * count, mapped and lock
1600                          */
1601                         page = get_next_nat_page(sbi, start_nid);
1602                         nat_blk = page_address(page);
1603                 }
1604
1605                 BUG_ON(!nat_blk);
1606                 raw_ne = nat_blk->entries[nid - start_nid];
1607 flush_now:
1608                 new_blkaddr = nat_get_blkaddr(ne);
1609
1610                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1611                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1612                 raw_ne.version = nat_get_version(ne);
1613
1614                 if (offset < 0) {
1615                         nat_blk->entries[nid - start_nid] = raw_ne;
1616                 } else {
1617                         nat_in_journal(sum, offset) = raw_ne;
1618                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1619                 }
1620
1621                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
1622                         write_lock(&nm_i->nat_tree_lock);
1623                         __del_from_nat_cache(nm_i, ne);
1624                         write_unlock(&nm_i->nat_tree_lock);
1625
1626                         /* We can reuse this freed nid at this point */
1627                         add_free_nid(NM_I(sbi), nid);
1628                 } else {
1629                         write_lock(&nm_i->nat_tree_lock);
1630                         __clear_nat_cache_dirty(nm_i, ne);
1631                         ne->checkpointed = true;
1632                         write_unlock(&nm_i->nat_tree_lock);
1633                 }
1634         }
1635         if (!flushed)
1636                 mutex_unlock(&curseg->curseg_mutex);
1637         f2fs_put_page(page, 1);
1638
1639         /* 2) shrink nat caches if necessary */
1640         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1641 }
1642
1643 static int init_node_manager(struct f2fs_sb_info *sbi)
1644 {
1645         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1646         struct f2fs_nm_info *nm_i = NM_I(sbi);
1647         unsigned char *version_bitmap;
1648         unsigned int nat_segs, nat_blocks;
1649
1650         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1651
1652         /* segment_count_nat includes pair segment so divide to 2. */
1653         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1654         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1655         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1656         nm_i->fcnt = 0;
1657         nm_i->nat_cnt = 0;
1658
1659         INIT_LIST_HEAD(&nm_i->free_nid_list);
1660         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1661         INIT_LIST_HEAD(&nm_i->nat_entries);
1662         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1663
1664         mutex_init(&nm_i->build_lock);
1665         spin_lock_init(&nm_i->free_nid_list_lock);
1666         rwlock_init(&nm_i->nat_tree_lock);
1667
1668         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1669         nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1670         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1671
1672         nm_i->nat_bitmap = kzalloc(nm_i->bitmap_size, GFP_KERNEL);
1673         if (!nm_i->nat_bitmap)
1674                 return -ENOMEM;
1675         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1676         if (!version_bitmap)
1677                 return -EFAULT;
1678
1679         /* copy version bitmap */
1680         memcpy(nm_i->nat_bitmap, version_bitmap, nm_i->bitmap_size);
1681         return 0;
1682 }
1683
1684 int build_node_manager(struct f2fs_sb_info *sbi)
1685 {
1686         int err;
1687
1688         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1689         if (!sbi->nm_info)
1690                 return -ENOMEM;
1691
1692         err = init_node_manager(sbi);
1693         if (err)
1694                 return err;
1695
1696         build_free_nids(sbi);
1697         return 0;
1698 }
1699
1700 void destroy_node_manager(struct f2fs_sb_info *sbi)
1701 {
1702         struct f2fs_nm_info *nm_i = NM_I(sbi);
1703         struct free_nid *i, *next_i;
1704         struct nat_entry *natvec[NATVEC_SIZE];
1705         nid_t nid = 0;
1706         unsigned int found;
1707
1708         if (!nm_i)
1709                 return;
1710
1711         /* destroy free nid list */
1712         spin_lock(&nm_i->free_nid_list_lock);
1713         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1714                 BUG_ON(i->state == NID_ALLOC);
1715                 __del_from_free_nid_list(i);
1716                 nm_i->fcnt--;
1717         }
1718         BUG_ON(nm_i->fcnt);
1719         spin_unlock(&nm_i->free_nid_list_lock);
1720
1721         /* destroy nat cache */
1722         write_lock(&nm_i->nat_tree_lock);
1723         while ((found = __gang_lookup_nat_cache(nm_i,
1724                                         nid, NATVEC_SIZE, natvec))) {
1725                 unsigned idx;
1726                 for (idx = 0; idx < found; idx++) {
1727                         struct nat_entry *e = natvec[idx];
1728                         nid = nat_get_nid(e) + 1;
1729                         __del_from_nat_cache(nm_i, e);
1730                 }
1731         }
1732         BUG_ON(nm_i->nat_cnt);
1733         write_unlock(&nm_i->nat_tree_lock);
1734
1735         kfree(nm_i->nat_bitmap);
1736         sbi->nm_info = NULL;
1737         kfree(nm_i);
1738 }
1739
1740 int __init create_node_manager_caches(void)
1741 {
1742         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1743                         sizeof(struct nat_entry), NULL);
1744         if (!nat_entry_slab)
1745                 return -ENOMEM;
1746
1747         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1748                         sizeof(struct free_nid), NULL);
1749         if (!free_nid_slab) {
1750                 kmem_cache_destroy(nat_entry_slab);
1751                 return -ENOMEM;
1752         }
1753         return 0;
1754 }
1755
1756 void destroy_node_manager_caches(void)
1757 {
1758         kmem_cache_destroy(free_nid_slab);
1759         kmem_cache_destroy(nat_entry_slab);
1760 }