]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - fs/f2fs/node.c
Merge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-3.0-fixes
[~shefty/rdma-dev.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         BUG_ON(ni.blk_addr == NULL_ADDR);
488
489         if (ni.blk_addr != NULL_ADDR)
490                 invalidate_blocks(sbi, ni.blk_addr);
491
492         /* Deallocate node address */
493         dec_valid_node_count(sbi, dn->inode, 1);
494         set_node_addr(sbi, &ni, NULL_ADDR);
495
496         if (dn->nid == dn->inode->i_ino) {
497                 remove_orphan_inode(sbi, dn->nid);
498                 dec_valid_inode_count(sbi);
499         } else {
500                 sync_inode_page(dn);
501         }
502
503         clear_node_page_dirty(dn->node_page);
504         F2FS_SET_SB_DIRT(sbi);
505
506         f2fs_put_page(dn->node_page, 1);
507         dn->node_page = NULL;
508 }
509
510 static int truncate_dnode(struct dnode_of_data *dn)
511 {
512         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
513         struct page *page;
514
515         if (dn->nid == 0)
516                 return 1;
517
518         /* get direct node */
519         page = get_node_page(sbi, dn->nid);
520         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
521                 return 1;
522         else if (IS_ERR(page))
523                 return PTR_ERR(page);
524
525         /* Make dnode_of_data for parameter */
526         dn->node_page = page;
527         dn->ofs_in_node = 0;
528         truncate_data_blocks(dn);
529         truncate_node(dn);
530         return 1;
531 }
532
533 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
534                                                 int ofs, int depth)
535 {
536         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
537         struct dnode_of_data rdn = *dn;
538         struct page *page;
539         struct f2fs_node *rn;
540         nid_t child_nid;
541         unsigned int child_nofs;
542         int freed = 0;
543         int i, ret;
544
545         if (dn->nid == 0)
546                 return NIDS_PER_BLOCK + 1;
547
548         page = get_node_page(sbi, dn->nid);
549         if (IS_ERR(page))
550                 return PTR_ERR(page);
551
552         rn = (struct f2fs_node *)page_address(page);
553         if (depth < 3) {
554                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
555                         child_nid = le32_to_cpu(rn->in.nid[i]);
556                         if (child_nid == 0)
557                                 continue;
558                         rdn.nid = child_nid;
559                         ret = truncate_dnode(&rdn);
560                         if (ret < 0)
561                                 goto out_err;
562                         set_nid(page, i, 0, false);
563                 }
564         } else {
565                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
566                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
567                         child_nid = le32_to_cpu(rn->in.nid[i]);
568                         if (child_nid == 0) {
569                                 child_nofs += NIDS_PER_BLOCK + 1;
570                                 continue;
571                         }
572                         rdn.nid = child_nid;
573                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
574                         if (ret == (NIDS_PER_BLOCK + 1)) {
575                                 set_nid(page, i, 0, false);
576                                 child_nofs += ret;
577                         } else if (ret < 0 && ret != -ENOENT) {
578                                 goto out_err;
579                         }
580                 }
581                 freed = child_nofs;
582         }
583
584         if (!ofs) {
585                 /* remove current indirect node */
586                 dn->node_page = page;
587                 truncate_node(dn);
588                 freed++;
589         } else {
590                 f2fs_put_page(page, 1);
591         }
592         return freed;
593
594 out_err:
595         f2fs_put_page(page, 1);
596         return ret;
597 }
598
599 static int truncate_partial_nodes(struct dnode_of_data *dn,
600                         struct f2fs_inode *ri, int *offset, int depth)
601 {
602         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
603         struct page *pages[2];
604         nid_t nid[3];
605         nid_t child_nid;
606         int err = 0;
607         int i;
608         int idx = depth - 2;
609
610         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
611         if (!nid[0])
612                 return 0;
613
614         /* get indirect nodes in the path */
615         for (i = 0; i < depth - 1; i++) {
616                 /* refernece count'll be increased */
617                 pages[i] = get_node_page(sbi, nid[i]);
618                 if (IS_ERR(pages[i])) {
619                         depth = i + 1;
620                         err = PTR_ERR(pages[i]);
621                         goto fail;
622                 }
623                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
624         }
625
626         /* free direct nodes linked to a partial indirect node */
627         for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
628                 child_nid = get_nid(pages[idx], i, false);
629                 if (!child_nid)
630                         continue;
631                 dn->nid = child_nid;
632                 err = truncate_dnode(dn);
633                 if (err < 0)
634                         goto fail;
635                 set_nid(pages[idx], i, 0, false);
636         }
637
638         if (offset[depth - 1] == 0) {
639                 dn->node_page = pages[idx];
640                 dn->nid = nid[idx];
641                 truncate_node(dn);
642         } else {
643                 f2fs_put_page(pages[idx], 1);
644         }
645         offset[idx]++;
646         offset[depth - 1] = 0;
647 fail:
648         for (i = depth - 3; i >= 0; i--)
649                 f2fs_put_page(pages[i], 1);
650         return err;
651 }
652
653 /*
654  * All the block addresses of data and nodes should be nullified.
655  */
656 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
657 {
658         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
659         int err = 0, cont = 1;
660         int level, offset[4], noffset[4];
661         unsigned int nofs;
662         struct f2fs_node *rn;
663         struct dnode_of_data dn;
664         struct page *page;
665
666         level = get_node_path(from, offset, noffset);
667
668         page = get_node_page(sbi, inode->i_ino);
669         if (IS_ERR(page))
670                 return PTR_ERR(page);
671
672         set_new_dnode(&dn, inode, page, NULL, 0);
673         unlock_page(page);
674
675         rn = page_address(page);
676         switch (level) {
677         case 0:
678         case 1:
679                 nofs = noffset[1];
680                 break;
681         case 2:
682                 nofs = noffset[1];
683                 if (!offset[level - 1])
684                         goto skip_partial;
685                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
686                 if (err < 0 && err != -ENOENT)
687                         goto fail;
688                 nofs += 1 + NIDS_PER_BLOCK;
689                 break;
690         case 3:
691                 nofs = 5 + 2 * NIDS_PER_BLOCK;
692                 if (!offset[level - 1])
693                         goto skip_partial;
694                 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
695                 if (err < 0 && err != -ENOENT)
696                         goto fail;
697                 break;
698         default:
699                 BUG();
700         }
701
702 skip_partial:
703         while (cont) {
704                 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
705                 switch (offset[0]) {
706                 case NODE_DIR1_BLOCK:
707                 case NODE_DIR2_BLOCK:
708                         err = truncate_dnode(&dn);
709                         break;
710
711                 case NODE_IND1_BLOCK:
712                 case NODE_IND2_BLOCK:
713                         err = truncate_nodes(&dn, nofs, offset[1], 2);
714                         break;
715
716                 case NODE_DIND_BLOCK:
717                         err = truncate_nodes(&dn, nofs, offset[1], 3);
718                         cont = 0;
719                         break;
720
721                 default:
722                         BUG();
723                 }
724                 if (err < 0 && err != -ENOENT)
725                         goto fail;
726                 if (offset[1] == 0 &&
727                                 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
728                         lock_page(page);
729                         wait_on_page_writeback(page);
730                         rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
731                         set_page_dirty(page);
732                         unlock_page(page);
733                 }
734                 offset[1] = 0;
735                 offset[0]++;
736                 nofs += err;
737         }
738 fail:
739         f2fs_put_page(page, 0);
740         return err > 0 ? 0 : err;
741 }
742
743 int remove_inode_page(struct inode *inode)
744 {
745         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
746         struct page *page;
747         nid_t ino = inode->i_ino;
748         struct dnode_of_data dn;
749
750         mutex_lock_op(sbi, NODE_TRUNC);
751         page = get_node_page(sbi, ino);
752         if (IS_ERR(page)) {
753                 mutex_unlock_op(sbi, NODE_TRUNC);
754                 return PTR_ERR(page);
755         }
756
757         if (F2FS_I(inode)->i_xattr_nid) {
758                 nid_t nid = F2FS_I(inode)->i_xattr_nid;
759                 struct page *npage = get_node_page(sbi, nid);
760
761                 if (IS_ERR(npage)) {
762                         mutex_unlock_op(sbi, NODE_TRUNC);
763                         return PTR_ERR(npage);
764                 }
765
766                 F2FS_I(inode)->i_xattr_nid = 0;
767                 set_new_dnode(&dn, inode, page, npage, nid);
768                 dn.inode_page_locked = 1;
769                 truncate_node(&dn);
770         }
771         if (inode->i_blocks == 1) {
772                 /* inernally call f2fs_put_page() */
773                 set_new_dnode(&dn, inode, page, page, ino);
774                 truncate_node(&dn);
775         } else if (inode->i_blocks == 0) {
776                 struct node_info ni;
777                 get_node_info(sbi, inode->i_ino, &ni);
778
779                 /* called after f2fs_new_inode() is failed */
780                 BUG_ON(ni.blk_addr != NULL_ADDR);
781                 f2fs_put_page(page, 1);
782         } else {
783                 BUG();
784         }
785         mutex_unlock_op(sbi, NODE_TRUNC);
786         return 0;
787 }
788
789 int new_inode_page(struct inode *inode, struct dentry *dentry)
790 {
791         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
792         struct page *page;
793         struct dnode_of_data dn;
794
795         /* allocate inode page for new inode */
796         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
797         mutex_lock_op(sbi, NODE_NEW);
798         page = new_node_page(&dn, 0);
799         init_dent_inode(dentry, page);
800         mutex_unlock_op(sbi, NODE_NEW);
801         if (IS_ERR(page))
802                 return PTR_ERR(page);
803         f2fs_put_page(page, 1);
804         return 0;
805 }
806
807 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
808 {
809         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
810         struct address_space *mapping = sbi->node_inode->i_mapping;
811         struct node_info old_ni, new_ni;
812         struct page *page;
813         int err;
814
815         if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
816                 return ERR_PTR(-EPERM);
817
818         page = grab_cache_page(mapping, dn->nid);
819         if (!page)
820                 return ERR_PTR(-ENOMEM);
821
822         get_node_info(sbi, dn->nid, &old_ni);
823
824         SetPageUptodate(page);
825         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
826
827         /* Reinitialize old_ni with new node page */
828         BUG_ON(old_ni.blk_addr != NULL_ADDR);
829         new_ni = old_ni;
830         new_ni.ino = dn->inode->i_ino;
831
832         if (!inc_valid_node_count(sbi, dn->inode, 1)) {
833                 err = -ENOSPC;
834                 goto fail;
835         }
836         set_node_addr(sbi, &new_ni, NEW_ADDR);
837
838         dn->node_page = page;
839         sync_inode_page(dn);
840         set_page_dirty(page);
841         set_cold_node(dn->inode, page);
842         if (ofs == 0)
843                 inc_valid_inode_count(sbi);
844
845         return page;
846
847 fail:
848         f2fs_put_page(page, 1);
849         return ERR_PTR(err);
850 }
851
852 static int read_node_page(struct page *page, int type)
853 {
854         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
855         struct node_info ni;
856
857         get_node_info(sbi, page->index, &ni);
858
859         if (ni.blk_addr == NULL_ADDR)
860                 return -ENOENT;
861         return f2fs_readpage(sbi, page, ni.blk_addr, type);
862 }
863
864 /*
865  * Readahead a node page
866  */
867 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
868 {
869         struct address_space *mapping = sbi->node_inode->i_mapping;
870         struct page *apage;
871
872         apage = find_get_page(mapping, nid);
873         if (apage && PageUptodate(apage))
874                 goto release_out;
875         f2fs_put_page(apage, 0);
876
877         apage = grab_cache_page(mapping, nid);
878         if (!apage)
879                 return;
880
881         if (read_node_page(apage, READA))
882                 goto unlock_out;
883
884         page_cache_release(apage);
885         return;
886
887 unlock_out:
888         unlock_page(apage);
889 release_out:
890         page_cache_release(apage);
891 }
892
893 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
894 {
895         int err;
896         struct page *page;
897         struct address_space *mapping = sbi->node_inode->i_mapping;
898
899         page = grab_cache_page(mapping, nid);
900         if (!page)
901                 return ERR_PTR(-ENOMEM);
902
903         err = read_node_page(page, READ_SYNC);
904         if (err) {
905                 f2fs_put_page(page, 1);
906                 return ERR_PTR(err);
907         }
908
909         BUG_ON(nid != nid_of_node(page));
910         mark_page_accessed(page);
911         return page;
912 }
913
914 /*
915  * Return a locked page for the desired node page.
916  * And, readahead MAX_RA_NODE number of node pages.
917  */
918 struct page *get_node_page_ra(struct page *parent, int start)
919 {
920         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
921         struct address_space *mapping = sbi->node_inode->i_mapping;
922         int i, end;
923         int err = 0;
924         nid_t nid;
925         struct page *page;
926
927         /* First, try getting the desired direct node. */
928         nid = get_nid(parent, start, false);
929         if (!nid)
930                 return ERR_PTR(-ENOENT);
931
932         page = find_get_page(mapping, nid);
933         if (page && PageUptodate(page))
934                 goto page_hit;
935         f2fs_put_page(page, 0);
936
937 repeat:
938         page = grab_cache_page(mapping, nid);
939         if (!page)
940                 return ERR_PTR(-ENOMEM);
941
942         err = read_node_page(page, READA);
943         if (err) {
944                 f2fs_put_page(page, 1);
945                 return ERR_PTR(err);
946         }
947
948         /* Then, try readahead for siblings of the desired node */
949         end = start + MAX_RA_NODE;
950         end = min(end, NIDS_PER_BLOCK);
951         for (i = start + 1; i < end; i++) {
952                 nid = get_nid(parent, i, false);
953                 if (!nid)
954                         continue;
955                 ra_node_page(sbi, nid);
956         }
957
958 page_hit:
959         lock_page(page);
960         if (PageError(page)) {
961                 f2fs_put_page(page, 1);
962                 return ERR_PTR(-EIO);
963         }
964
965         /* Has the page been truncated? */
966         if (page->mapping != mapping) {
967                 f2fs_put_page(page, 1);
968                 goto repeat;
969         }
970         return page;
971 }
972
973 void sync_inode_page(struct dnode_of_data *dn)
974 {
975         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
976                 update_inode(dn->inode, dn->node_page);
977         } else if (dn->inode_page) {
978                 if (!dn->inode_page_locked)
979                         lock_page(dn->inode_page);
980                 update_inode(dn->inode, dn->inode_page);
981                 if (!dn->inode_page_locked)
982                         unlock_page(dn->inode_page);
983         } else {
984                 f2fs_write_inode(dn->inode, NULL);
985         }
986 }
987
988 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
989                                         struct writeback_control *wbc)
990 {
991         struct address_space *mapping = sbi->node_inode->i_mapping;
992         pgoff_t index, end;
993         struct pagevec pvec;
994         int step = ino ? 2 : 0;
995         int nwritten = 0, wrote = 0;
996
997         pagevec_init(&pvec, 0);
998
999 next_step:
1000         index = 0;
1001         end = LONG_MAX;
1002
1003         while (index <= end) {
1004                 int i, nr_pages;
1005                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1006                                 PAGECACHE_TAG_DIRTY,
1007                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1008                 if (nr_pages == 0)
1009                         break;
1010
1011                 for (i = 0; i < nr_pages; i++) {
1012                         struct page *page = pvec.pages[i];
1013
1014                         /*
1015                          * flushing sequence with step:
1016                          * 0. indirect nodes
1017                          * 1. dentry dnodes
1018                          * 2. file dnodes
1019                          */
1020                         if (step == 0 && IS_DNODE(page))
1021                                 continue;
1022                         if (step == 1 && (!IS_DNODE(page) ||
1023                                                 is_cold_node(page)))
1024                                 continue;
1025                         if (step == 2 && (!IS_DNODE(page) ||
1026                                                 !is_cold_node(page)))
1027                                 continue;
1028
1029                         /*
1030                          * If an fsync mode,
1031                          * we should not skip writing node pages.
1032                          */
1033                         if (ino && ino_of_node(page) == ino)
1034                                 lock_page(page);
1035                         else if (!trylock_page(page))
1036                                 continue;
1037
1038                         if (unlikely(page->mapping != mapping)) {
1039 continue_unlock:
1040                                 unlock_page(page);
1041                                 continue;
1042                         }
1043                         if (ino && ino_of_node(page) != ino)
1044                                 goto continue_unlock;
1045
1046                         if (!PageDirty(page)) {
1047                                 /* someone wrote it for us */
1048                                 goto continue_unlock;
1049                         }
1050
1051                         if (!clear_page_dirty_for_io(page))
1052                                 goto continue_unlock;
1053
1054                         /* called by fsync() */
1055                         if (ino && IS_DNODE(page)) {
1056                                 int mark = !is_checkpointed_node(sbi, ino);
1057                                 set_fsync_mark(page, 1);
1058                                 if (IS_INODE(page))
1059                                         set_dentry_mark(page, mark);
1060                                 nwritten++;
1061                         } else {
1062                                 set_fsync_mark(page, 0);
1063                                 set_dentry_mark(page, 0);
1064                         }
1065                         mapping->a_ops->writepage(page, wbc);
1066                         wrote++;
1067
1068                         if (--wbc->nr_to_write == 0)
1069                                 break;
1070                 }
1071                 pagevec_release(&pvec);
1072                 cond_resched();
1073
1074                 if (wbc->nr_to_write == 0) {
1075                         step = 2;
1076                         break;
1077                 }
1078         }
1079
1080         if (step < 2) {
1081                 step++;
1082                 goto next_step;
1083         }
1084
1085         if (wrote)
1086                 f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1087
1088         return nwritten;
1089 }
1090
1091 static int f2fs_write_node_page(struct page *page,
1092                                 struct writeback_control *wbc)
1093 {
1094         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1095         nid_t nid;
1096         unsigned int nofs;
1097         block_t new_addr;
1098         struct node_info ni;
1099
1100         if (wbc->for_reclaim) {
1101                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1102                 wbc->pages_skipped++;
1103                 set_page_dirty(page);
1104                 return AOP_WRITEPAGE_ACTIVATE;
1105         }
1106
1107         wait_on_page_writeback(page);
1108
1109         mutex_lock_op(sbi, NODE_WRITE);
1110
1111         /* get old block addr of this node page */
1112         nid = nid_of_node(page);
1113         nofs = ofs_of_node(page);
1114         BUG_ON(page->index != nid);
1115
1116         get_node_info(sbi, nid, &ni);
1117
1118         /* This page is already truncated */
1119         if (ni.blk_addr == NULL_ADDR)
1120                 return 0;
1121
1122         set_page_writeback(page);
1123
1124         /* insert node offset */
1125         write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1126         set_node_addr(sbi, &ni, new_addr);
1127         dec_page_count(sbi, F2FS_DIRTY_NODES);
1128
1129         mutex_unlock_op(sbi, NODE_WRITE);
1130         unlock_page(page);
1131         return 0;
1132 }
1133
1134 static int f2fs_write_node_pages(struct address_space *mapping,
1135                             struct writeback_control *wbc)
1136 {
1137         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1138         struct block_device *bdev = sbi->sb->s_bdev;
1139         long nr_to_write = wbc->nr_to_write;
1140
1141         if (wbc->for_kupdate)
1142                 return 0;
1143
1144         if (get_pages(sbi, F2FS_DIRTY_NODES) == 0)
1145                 return 0;
1146
1147         if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
1148                 write_checkpoint(sbi, false, false);
1149                 return 0;
1150         }
1151
1152         /* if mounting is failed, skip writing node pages */
1153         wbc->nr_to_write = bio_get_nr_vecs(bdev);
1154         sync_node_pages(sbi, 0, wbc);
1155         wbc->nr_to_write = nr_to_write -
1156                 (bio_get_nr_vecs(bdev) - wbc->nr_to_write);
1157         return 0;
1158 }
1159
1160 static int f2fs_set_node_page_dirty(struct page *page)
1161 {
1162         struct address_space *mapping = page->mapping;
1163         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1164
1165         SetPageUptodate(page);
1166         if (!PageDirty(page)) {
1167                 __set_page_dirty_nobuffers(page);
1168                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1169                 SetPagePrivate(page);
1170                 return 1;
1171         }
1172         return 0;
1173 }
1174
1175 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
1176 {
1177         struct inode *inode = page->mapping->host;
1178         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1179         if (PageDirty(page))
1180                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1181         ClearPagePrivate(page);
1182 }
1183
1184 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1185 {
1186         ClearPagePrivate(page);
1187         return 0;
1188 }
1189
1190 /*
1191  * Structure of the f2fs node operations
1192  */
1193 const struct address_space_operations f2fs_node_aops = {
1194         .writepage      = f2fs_write_node_page,
1195         .writepages     = f2fs_write_node_pages,
1196         .set_page_dirty = f2fs_set_node_page_dirty,
1197         .invalidatepage = f2fs_invalidate_node_page,
1198         .releasepage    = f2fs_release_node_page,
1199 };
1200
1201 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1202 {
1203         struct list_head *this;
1204         struct free_nid *i = NULL;
1205         list_for_each(this, head) {
1206                 i = list_entry(this, struct free_nid, list);
1207                 if (i->nid == n)
1208                         break;
1209                 i = NULL;
1210         }
1211         return i;
1212 }
1213
1214 static void __del_from_free_nid_list(struct free_nid *i)
1215 {
1216         list_del(&i->list);
1217         kmem_cache_free(free_nid_slab, i);
1218 }
1219
1220 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1221 {
1222         struct free_nid *i;
1223
1224         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1225                 return 0;
1226 retry:
1227         i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1228         if (!i) {
1229                 cond_resched();
1230                 goto retry;
1231         }
1232         i->nid = nid;
1233         i->state = NID_NEW;
1234
1235         spin_lock(&nm_i->free_nid_list_lock);
1236         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1237                 spin_unlock(&nm_i->free_nid_list_lock);
1238                 kmem_cache_free(free_nid_slab, i);
1239                 return 0;
1240         }
1241         list_add_tail(&i->list, &nm_i->free_nid_list);
1242         nm_i->fcnt++;
1243         spin_unlock(&nm_i->free_nid_list_lock);
1244         return 1;
1245 }
1246
1247 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1248 {
1249         struct free_nid *i;
1250         spin_lock(&nm_i->free_nid_list_lock);
1251         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1252         if (i && i->state == NID_NEW) {
1253                 __del_from_free_nid_list(i);
1254                 nm_i->fcnt--;
1255         }
1256         spin_unlock(&nm_i->free_nid_list_lock);
1257 }
1258
1259 static int scan_nat_page(struct f2fs_nm_info *nm_i,
1260                         struct page *nat_page, nid_t start_nid)
1261 {
1262         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1263         block_t blk_addr;
1264         int fcnt = 0;
1265         int i;
1266
1267         /* 0 nid should not be used */
1268         if (start_nid == 0)
1269                 ++start_nid;
1270
1271         i = start_nid % NAT_ENTRY_PER_BLOCK;
1272
1273         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1274                 blk_addr  = le32_to_cpu(nat_blk->entries[i].block_addr);
1275                 BUG_ON(blk_addr == NEW_ADDR);
1276                 if (blk_addr == NULL_ADDR)
1277                         fcnt += add_free_nid(nm_i, start_nid);
1278         }
1279         return fcnt;
1280 }
1281
1282 static void build_free_nids(struct f2fs_sb_info *sbi)
1283 {
1284         struct free_nid *fnid, *next_fnid;
1285         struct f2fs_nm_info *nm_i = NM_I(sbi);
1286         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1287         struct f2fs_summary_block *sum = curseg->sum_blk;
1288         nid_t nid = 0;
1289         bool is_cycled = false;
1290         int fcnt = 0;
1291         int i;
1292
1293         nid = nm_i->next_scan_nid;
1294         nm_i->init_scan_nid = nid;
1295
1296         ra_nat_pages(sbi, nid);
1297
1298         while (1) {
1299                 struct page *page = get_current_nat_page(sbi, nid);
1300
1301                 fcnt += scan_nat_page(nm_i, page, nid);
1302                 f2fs_put_page(page, 1);
1303
1304                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1305
1306                 if (nid >= nm_i->max_nid) {
1307                         nid = 0;
1308                         is_cycled = true;
1309                 }
1310                 if (fcnt > MAX_FREE_NIDS)
1311                         break;
1312                 if (is_cycled && nm_i->init_scan_nid <= nid)
1313                         break;
1314         }
1315
1316         nm_i->next_scan_nid = nid;
1317
1318         /* find free nids from current sum_pages */
1319         mutex_lock(&curseg->curseg_mutex);
1320         for (i = 0; i < nats_in_cursum(sum); i++) {
1321                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1322                 nid = le32_to_cpu(nid_in_journal(sum, i));
1323                 if (addr == NULL_ADDR)
1324                         add_free_nid(nm_i, nid);
1325                 else
1326                         remove_free_nid(nm_i, nid);
1327         }
1328         mutex_unlock(&curseg->curseg_mutex);
1329
1330         /* remove the free nids from current allocated nids */
1331         list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) {
1332                 struct nat_entry *ne;
1333
1334                 read_lock(&nm_i->nat_tree_lock);
1335                 ne = __lookup_nat_cache(nm_i, fnid->nid);
1336                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1337                         remove_free_nid(nm_i, fnid->nid);
1338                 read_unlock(&nm_i->nat_tree_lock);
1339         }
1340 }
1341
1342 /*
1343  * If this function returns success, caller can obtain a new nid
1344  * from second parameter of this function.
1345  * The returned nid could be used ino as well as nid when inode is created.
1346  */
1347 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1348 {
1349         struct f2fs_nm_info *nm_i = NM_I(sbi);
1350         struct free_nid *i = NULL;
1351         struct list_head *this;
1352 retry:
1353         mutex_lock(&nm_i->build_lock);
1354         if (!nm_i->fcnt) {
1355                 /* scan NAT in order to build free nid list */
1356                 build_free_nids(sbi);
1357                 if (!nm_i->fcnt) {
1358                         mutex_unlock(&nm_i->build_lock);
1359                         return false;
1360                 }
1361         }
1362         mutex_unlock(&nm_i->build_lock);
1363
1364         /*
1365          * We check fcnt again since previous check is racy as
1366          * we didn't hold free_nid_list_lock. So other thread
1367          * could consume all of free nids.
1368          */
1369         spin_lock(&nm_i->free_nid_list_lock);
1370         if (!nm_i->fcnt) {
1371                 spin_unlock(&nm_i->free_nid_list_lock);
1372                 goto retry;
1373         }
1374
1375         BUG_ON(list_empty(&nm_i->free_nid_list));
1376         list_for_each(this, &nm_i->free_nid_list) {
1377                 i = list_entry(this, struct free_nid, list);
1378                 if (i->state == NID_NEW)
1379                         break;
1380         }
1381
1382         BUG_ON(i->state != NID_NEW);
1383         *nid = i->nid;
1384         i->state = NID_ALLOC;
1385         nm_i->fcnt--;
1386         spin_unlock(&nm_i->free_nid_list_lock);
1387         return true;
1388 }
1389
1390 /*
1391  * alloc_nid() should be called prior to this function.
1392  */
1393 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1394 {
1395         struct f2fs_nm_info *nm_i = NM_I(sbi);
1396         struct free_nid *i;
1397
1398         spin_lock(&nm_i->free_nid_list_lock);
1399         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1400         if (i) {
1401                 BUG_ON(i->state != NID_ALLOC);
1402                 __del_from_free_nid_list(i);
1403         }
1404         spin_unlock(&nm_i->free_nid_list_lock);
1405 }
1406
1407 /*
1408  * alloc_nid() should be called prior to this function.
1409  */
1410 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1411 {
1412         alloc_nid_done(sbi, nid);
1413         add_free_nid(NM_I(sbi), nid);
1414 }
1415
1416 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1417                 struct f2fs_summary *sum, struct node_info *ni,
1418                 block_t new_blkaddr)
1419 {
1420         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1421         set_node_addr(sbi, ni, new_blkaddr);
1422         clear_node_page_dirty(page);
1423 }
1424
1425 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1426 {
1427         struct address_space *mapping = sbi->node_inode->i_mapping;
1428         struct f2fs_node *src, *dst;
1429         nid_t ino = ino_of_node(page);
1430         struct node_info old_ni, new_ni;
1431         struct page *ipage;
1432
1433         ipage = grab_cache_page(mapping, ino);
1434         if (!ipage)
1435                 return -ENOMEM;
1436
1437         /* Should not use this inode  from free nid list */
1438         remove_free_nid(NM_I(sbi), ino);
1439
1440         get_node_info(sbi, ino, &old_ni);
1441         SetPageUptodate(ipage);
1442         fill_node_footer(ipage, ino, ino, 0, true);
1443
1444         src = (struct f2fs_node *)page_address(page);
1445         dst = (struct f2fs_node *)page_address(ipage);
1446
1447         memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1448         dst->i.i_size = 0;
1449         dst->i.i_blocks = cpu_to_le64(1);
1450         dst->i.i_links = cpu_to_le32(1);
1451         dst->i.i_xattr_nid = 0;
1452
1453         new_ni = old_ni;
1454         new_ni.ino = ino;
1455
1456         set_node_addr(sbi, &new_ni, NEW_ADDR);
1457         inc_valid_inode_count(sbi);
1458
1459         f2fs_put_page(ipage, 1);
1460         return 0;
1461 }
1462
1463 int restore_node_summary(struct f2fs_sb_info *sbi,
1464                         unsigned int segno, struct f2fs_summary_block *sum)
1465 {
1466         struct f2fs_node *rn;
1467         struct f2fs_summary *sum_entry;
1468         struct page *page;
1469         block_t addr;
1470         int i, last_offset;
1471
1472         /* alloc temporal page for read node */
1473         page = alloc_page(GFP_NOFS | __GFP_ZERO);
1474         if (IS_ERR(page))
1475                 return PTR_ERR(page);
1476         lock_page(page);
1477
1478         /* scan the node segment */
1479         last_offset = sbi->blocks_per_seg;
1480         addr = START_BLOCK(sbi, segno);
1481         sum_entry = &sum->entries[0];
1482
1483         for (i = 0; i < last_offset; i++, sum_entry++) {
1484                 if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1485                         goto out;
1486
1487                 rn = (struct f2fs_node *)page_address(page);
1488                 sum_entry->nid = rn->footer.nid;
1489                 sum_entry->version = 0;
1490                 sum_entry->ofs_in_node = 0;
1491                 addr++;
1492
1493                 /*
1494                  * In order to read next node page,
1495                  * we must clear PageUptodate flag.
1496                  */
1497                 ClearPageUptodate(page);
1498         }
1499 out:
1500         unlock_page(page);
1501         __free_pages(page, 0);
1502         return 0;
1503 }
1504
1505 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1506 {
1507         struct f2fs_nm_info *nm_i = NM_I(sbi);
1508         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1509         struct f2fs_summary_block *sum = curseg->sum_blk;
1510         int i;
1511
1512         mutex_lock(&curseg->curseg_mutex);
1513
1514         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1515                 mutex_unlock(&curseg->curseg_mutex);
1516                 return false;
1517         }
1518
1519         for (i = 0; i < nats_in_cursum(sum); i++) {
1520                 struct nat_entry *ne;
1521                 struct f2fs_nat_entry raw_ne;
1522                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1523
1524                 raw_ne = nat_in_journal(sum, i);
1525 retry:
1526                 write_lock(&nm_i->nat_tree_lock);
1527                 ne = __lookup_nat_cache(nm_i, nid);
1528                 if (ne) {
1529                         __set_nat_cache_dirty(nm_i, ne);
1530                         write_unlock(&nm_i->nat_tree_lock);
1531                         continue;
1532                 }
1533                 ne = grab_nat_entry(nm_i, nid);
1534                 if (!ne) {
1535                         write_unlock(&nm_i->nat_tree_lock);
1536                         goto retry;
1537                 }
1538                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1539                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1540                 nat_set_version(ne, raw_ne.version);
1541                 __set_nat_cache_dirty(nm_i, ne);
1542                 write_unlock(&nm_i->nat_tree_lock);
1543         }
1544         update_nats_in_cursum(sum, -i);
1545         mutex_unlock(&curseg->curseg_mutex);
1546         return true;
1547 }
1548
1549 /*
1550  * This function is called during the checkpointing process.
1551  */
1552 void flush_nat_entries(struct f2fs_sb_info *sbi)
1553 {
1554         struct f2fs_nm_info *nm_i = NM_I(sbi);
1555         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1556         struct f2fs_summary_block *sum = curseg->sum_blk;
1557         struct list_head *cur, *n;
1558         struct page *page = NULL;
1559         struct f2fs_nat_block *nat_blk = NULL;
1560         nid_t start_nid = 0, end_nid = 0;
1561         bool flushed;
1562
1563         flushed = flush_nats_in_journal(sbi);
1564
1565         if (!flushed)
1566                 mutex_lock(&curseg->curseg_mutex);
1567
1568         /* 1) flush dirty nat caches */
1569         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1570                 struct nat_entry *ne;
1571                 nid_t nid;
1572                 struct f2fs_nat_entry raw_ne;
1573                 int offset = -1;
1574                 block_t old_blkaddr, new_blkaddr;
1575
1576                 ne = list_entry(cur, struct nat_entry, list);
1577                 nid = nat_get_nid(ne);
1578
1579                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1580                         continue;
1581                 if (flushed)
1582                         goto to_nat_page;
1583
1584                 /* if there is room for nat enries in curseg->sumpage */
1585                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1586                 if (offset >= 0) {
1587                         raw_ne = nat_in_journal(sum, offset);
1588                         old_blkaddr = le32_to_cpu(raw_ne.block_addr);
1589                         goto flush_now;
1590                 }
1591 to_nat_page:
1592                 if (!page || (start_nid > nid || nid > end_nid)) {
1593                         if (page) {
1594                                 f2fs_put_page(page, 1);
1595                                 page = NULL;
1596                         }
1597                         start_nid = START_NID(nid);
1598                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1599
1600                         /*
1601                          * get nat block with dirty flag, increased reference
1602                          * count, mapped and lock
1603                          */
1604                         page = get_next_nat_page(sbi, start_nid);
1605                         nat_blk = page_address(page);
1606                 }
1607
1608                 BUG_ON(!nat_blk);
1609                 raw_ne = nat_blk->entries[nid - start_nid];
1610                 old_blkaddr = le32_to_cpu(raw_ne.block_addr);
1611 flush_now:
1612                 new_blkaddr = nat_get_blkaddr(ne);
1613
1614                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1615                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1616                 raw_ne.version = nat_get_version(ne);
1617
1618                 if (offset < 0) {
1619                         nat_blk->entries[nid - start_nid] = raw_ne;
1620                 } else {
1621                         nat_in_journal(sum, offset) = raw_ne;
1622                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1623                 }
1624
1625                 if (nat_get_blkaddr(ne) == NULL_ADDR) {
1626                         write_lock(&nm_i->nat_tree_lock);
1627                         __del_from_nat_cache(nm_i, ne);
1628                         write_unlock(&nm_i->nat_tree_lock);
1629
1630                         /* We can reuse this freed nid at this point */
1631                         add_free_nid(NM_I(sbi), nid);
1632                 } else {
1633                         write_lock(&nm_i->nat_tree_lock);
1634                         __clear_nat_cache_dirty(nm_i, ne);
1635                         ne->checkpointed = true;
1636                         write_unlock(&nm_i->nat_tree_lock);
1637                 }
1638         }
1639         if (!flushed)
1640                 mutex_unlock(&curseg->curseg_mutex);
1641         f2fs_put_page(page, 1);
1642
1643         /* 2) shrink nat caches if necessary */
1644         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1645 }
1646
1647 static int init_node_manager(struct f2fs_sb_info *sbi)
1648 {
1649         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1650         struct f2fs_nm_info *nm_i = NM_I(sbi);
1651         unsigned char *version_bitmap;
1652         unsigned int nat_segs, nat_blocks;
1653
1654         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1655
1656         /* segment_count_nat includes pair segment so divide to 2. */
1657         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1658         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1659         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1660         nm_i->fcnt = 0;
1661         nm_i->nat_cnt = 0;
1662
1663         INIT_LIST_HEAD(&nm_i->free_nid_list);
1664         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1665         INIT_LIST_HEAD(&nm_i->nat_entries);
1666         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1667
1668         mutex_init(&nm_i->build_lock);
1669         spin_lock_init(&nm_i->free_nid_list_lock);
1670         rwlock_init(&nm_i->nat_tree_lock);
1671
1672         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1673         nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1674         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1675
1676         nm_i->nat_bitmap = kzalloc(nm_i->bitmap_size, GFP_KERNEL);
1677         if (!nm_i->nat_bitmap)
1678                 return -ENOMEM;
1679         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1680         if (!version_bitmap)
1681                 return -EFAULT;
1682
1683         /* copy version bitmap */
1684         memcpy(nm_i->nat_bitmap, version_bitmap, nm_i->bitmap_size);
1685         return 0;
1686 }
1687
1688 int build_node_manager(struct f2fs_sb_info *sbi)
1689 {
1690         int err;
1691
1692         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1693         if (!sbi->nm_info)
1694                 return -ENOMEM;
1695
1696         err = init_node_manager(sbi);
1697         if (err)
1698                 return err;
1699
1700         build_free_nids(sbi);
1701         return 0;
1702 }
1703
1704 void destroy_node_manager(struct f2fs_sb_info *sbi)
1705 {
1706         struct f2fs_nm_info *nm_i = NM_I(sbi);
1707         struct free_nid *i, *next_i;
1708         struct nat_entry *natvec[NATVEC_SIZE];
1709         nid_t nid = 0;
1710         unsigned int found;
1711
1712         if (!nm_i)
1713                 return;
1714
1715         /* destroy free nid list */
1716         spin_lock(&nm_i->free_nid_list_lock);
1717         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1718                 BUG_ON(i->state == NID_ALLOC);
1719                 __del_from_free_nid_list(i);
1720                 nm_i->fcnt--;
1721         }
1722         BUG_ON(nm_i->fcnt);
1723         spin_unlock(&nm_i->free_nid_list_lock);
1724
1725         /* destroy nat cache */
1726         write_lock(&nm_i->nat_tree_lock);
1727         while ((found = __gang_lookup_nat_cache(nm_i,
1728                                         nid, NATVEC_SIZE, natvec))) {
1729                 unsigned idx;
1730                 for (idx = 0; idx < found; idx++) {
1731                         struct nat_entry *e = natvec[idx];
1732                         nid = nat_get_nid(e) + 1;
1733                         __del_from_nat_cache(nm_i, e);
1734                 }
1735         }
1736         BUG_ON(nm_i->nat_cnt);
1737         write_unlock(&nm_i->nat_tree_lock);
1738
1739         kfree(nm_i->nat_bitmap);
1740         sbi->nm_info = NULL;
1741         kfree(nm_i);
1742 }
1743
1744 int create_node_manager_caches(void)
1745 {
1746         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1747                         sizeof(struct nat_entry), NULL);
1748         if (!nat_entry_slab)
1749                 return -ENOMEM;
1750
1751         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1752                         sizeof(struct free_nid), NULL);
1753         if (!free_nid_slab) {
1754                 kmem_cache_destroy(nat_entry_slab);
1755                 return -ENOMEM;
1756         }
1757         return 0;
1758 }
1759
1760 void destroy_node_manager_caches(void)
1761 {
1762         kmem_cache_destroy(free_nid_slab);
1763         kmem_cache_destroy(nat_entry_slab);
1764 }