]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/f2fs/node.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
f2fs: check node id earily when readaheading node page
[mirror_ubuntu-zesty-kernel.git] / fs / f2fs / node.c
1 /*
2 * fs/f2fs/node.c
3 *
4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
18
19 #include "f2fs.h"
20 #include "node.h"
21 #include "segment.h"
22 #include "trace.h"
23 #include <trace/events/f2fs.h>
24
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
30
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32 {
33 struct f2fs_nm_info *nm_i = NM_I(sbi);
34 struct sysinfo val;
35 unsigned long avail_ram;
36 unsigned long mem_size = 0;
37 bool res = false;
38
39 si_meminfo(&val);
40
41 /* only uses low memory */
42 avail_ram = val.totalram - val.totalhigh;
43
44 /*
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
46 */
47 if (type == FREE_NIDS) {
48 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
49 PAGE_CACHE_SHIFT;
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 PAGE_CACHE_SHIFT;
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 } else if (type == DIRTY_DENTS) {
56 if (sbi->sb->s_bdi->wb.dirty_exceeded)
57 return false;
58 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
59 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
60 } else if (type == INO_ENTRIES) {
61 int i;
62
63 for (i = 0; i <= UPDATE_INO; i++)
64 mem_size += (sbi->im[i].ino_num *
65 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
67 } else if (type == EXTENT_CACHE) {
68 mem_size = (atomic_read(&sbi->total_ext_tree) *
69 sizeof(struct extent_tree) +
70 atomic_read(&sbi->total_ext_node) *
71 sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
72 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
73 } else {
74 if (sbi->sb->s_bdi->wb.dirty_exceeded)
75 return false;
76 }
77 return res;
78 }
79
80 static void clear_node_page_dirty(struct page *page)
81 {
82 struct address_space *mapping = page->mapping;
83 unsigned int long flags;
84
85 if (PageDirty(page)) {
86 spin_lock_irqsave(&mapping->tree_lock, flags);
87 radix_tree_tag_clear(&mapping->page_tree,
88 page_index(page),
89 PAGECACHE_TAG_DIRTY);
90 spin_unlock_irqrestore(&mapping->tree_lock, flags);
91
92 clear_page_dirty_for_io(page);
93 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
94 }
95 ClearPageUptodate(page);
96 }
97
98 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
99 {
100 pgoff_t index = current_nat_addr(sbi, nid);
101 return get_meta_page(sbi, index);
102 }
103
104 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
105 {
106 struct page *src_page;
107 struct page *dst_page;
108 pgoff_t src_off;
109 pgoff_t dst_off;
110 void *src_addr;
111 void *dst_addr;
112 struct f2fs_nm_info *nm_i = NM_I(sbi);
113
114 src_off = current_nat_addr(sbi, nid);
115 dst_off = next_nat_addr(sbi, src_off);
116
117 /* get current nat block page with lock */
118 src_page = get_meta_page(sbi, src_off);
119 dst_page = grab_meta_page(sbi, dst_off);
120 f2fs_bug_on(sbi, PageDirty(src_page));
121
122 src_addr = page_address(src_page);
123 dst_addr = page_address(dst_page);
124 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
125 set_page_dirty(dst_page);
126 f2fs_put_page(src_page, 1);
127
128 set_to_next_nat(nm_i, nid);
129
130 return dst_page;
131 }
132
133 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
134 {
135 return radix_tree_lookup(&nm_i->nat_root, n);
136 }
137
138 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
139 nid_t start, unsigned int nr, struct nat_entry **ep)
140 {
141 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
142 }
143
144 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
145 {
146 list_del(&e->list);
147 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
148 nm_i->nat_cnt--;
149 kmem_cache_free(nat_entry_slab, e);
150 }
151
152 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
153 struct nat_entry *ne)
154 {
155 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
156 struct nat_entry_set *head;
157
158 if (get_nat_flag(ne, IS_DIRTY))
159 return;
160
161 head = radix_tree_lookup(&nm_i->nat_set_root, set);
162 if (!head) {
163 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
164
165 INIT_LIST_HEAD(&head->entry_list);
166 INIT_LIST_HEAD(&head->set_list);
167 head->set = set;
168 head->entry_cnt = 0;
169 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
170 }
171 list_move_tail(&ne->list, &head->entry_list);
172 nm_i->dirty_nat_cnt++;
173 head->entry_cnt++;
174 set_nat_flag(ne, IS_DIRTY, true);
175 }
176
177 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
178 struct nat_entry *ne)
179 {
180 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
181 struct nat_entry_set *head;
182
183 head = radix_tree_lookup(&nm_i->nat_set_root, set);
184 if (head) {
185 list_move_tail(&ne->list, &nm_i->nat_entries);
186 set_nat_flag(ne, IS_DIRTY, false);
187 head->entry_cnt--;
188 nm_i->dirty_nat_cnt--;
189 }
190 }
191
192 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
193 nid_t start, unsigned int nr, struct nat_entry_set **ep)
194 {
195 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
196 start, nr);
197 }
198
199 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
200 {
201 struct f2fs_nm_info *nm_i = NM_I(sbi);
202 struct nat_entry *e;
203 bool need = false;
204
205 down_read(&nm_i->nat_tree_lock);
206 e = __lookup_nat_cache(nm_i, nid);
207 if (e) {
208 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
209 !get_nat_flag(e, HAS_FSYNCED_INODE))
210 need = true;
211 }
212 up_read(&nm_i->nat_tree_lock);
213 return need;
214 }
215
216 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
217 {
218 struct f2fs_nm_info *nm_i = NM_I(sbi);
219 struct nat_entry *e;
220 bool is_cp = true;
221
222 down_read(&nm_i->nat_tree_lock);
223 e = __lookup_nat_cache(nm_i, nid);
224 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
225 is_cp = false;
226 up_read(&nm_i->nat_tree_lock);
227 return is_cp;
228 }
229
230 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
231 {
232 struct f2fs_nm_info *nm_i = NM_I(sbi);
233 struct nat_entry *e;
234 bool need_update = true;
235
236 down_read(&nm_i->nat_tree_lock);
237 e = __lookup_nat_cache(nm_i, ino);
238 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
239 (get_nat_flag(e, IS_CHECKPOINTED) ||
240 get_nat_flag(e, HAS_FSYNCED_INODE)))
241 need_update = false;
242 up_read(&nm_i->nat_tree_lock);
243 return need_update;
244 }
245
246 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
247 {
248 struct nat_entry *new;
249
250 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
251 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
252 memset(new, 0, sizeof(struct nat_entry));
253 nat_set_nid(new, nid);
254 nat_reset_flag(new);
255 list_add_tail(&new->list, &nm_i->nat_entries);
256 nm_i->nat_cnt++;
257 return new;
258 }
259
260 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
261 struct f2fs_nat_entry *ne)
262 {
263 struct nat_entry *e;
264
265 e = __lookup_nat_cache(nm_i, nid);
266 if (!e) {
267 e = grab_nat_entry(nm_i, nid);
268 node_info_from_raw_nat(&e->ni, ne);
269 }
270 }
271
272 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
273 block_t new_blkaddr, bool fsync_done)
274 {
275 struct f2fs_nm_info *nm_i = NM_I(sbi);
276 struct nat_entry *e;
277
278 down_write(&nm_i->nat_tree_lock);
279 e = __lookup_nat_cache(nm_i, ni->nid);
280 if (!e) {
281 e = grab_nat_entry(nm_i, ni->nid);
282 copy_node_info(&e->ni, ni);
283 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
284 } else if (new_blkaddr == NEW_ADDR) {
285 /*
286 * when nid is reallocated,
287 * previous nat entry can be remained in nat cache.
288 * So, reinitialize it with new information.
289 */
290 copy_node_info(&e->ni, ni);
291 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
292 }
293
294 /* sanity check */
295 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
296 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
297 new_blkaddr == NULL_ADDR);
298 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
299 new_blkaddr == NEW_ADDR);
300 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
301 nat_get_blkaddr(e) != NULL_ADDR &&
302 new_blkaddr == NEW_ADDR);
303
304 /* increment version no as node is removed */
305 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
306 unsigned char version = nat_get_version(e);
307 nat_set_version(e, inc_node_version(version));
308
309 /* in order to reuse the nid */
310 if (nm_i->next_scan_nid > ni->nid)
311 nm_i->next_scan_nid = ni->nid;
312 }
313
314 /* change address */
315 nat_set_blkaddr(e, new_blkaddr);
316 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
317 set_nat_flag(e, IS_CHECKPOINTED, false);
318 __set_nat_cache_dirty(nm_i, e);
319
320 /* update fsync_mark if its inode nat entry is still alive */
321 if (ni->nid != ni->ino)
322 e = __lookup_nat_cache(nm_i, ni->ino);
323 if (e) {
324 if (fsync_done && ni->nid == ni->ino)
325 set_nat_flag(e, HAS_FSYNCED_INODE, true);
326 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
327 }
328 up_write(&nm_i->nat_tree_lock);
329 }
330
331 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
332 {
333 struct f2fs_nm_info *nm_i = NM_I(sbi);
334 int nr = nr_shrink;
335
336 if (!down_write_trylock(&nm_i->nat_tree_lock))
337 return 0;
338
339 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
340 struct nat_entry *ne;
341 ne = list_first_entry(&nm_i->nat_entries,
342 struct nat_entry, list);
343 __del_from_nat_cache(nm_i, ne);
344 nr_shrink--;
345 }
346 up_write(&nm_i->nat_tree_lock);
347 return nr - nr_shrink;
348 }
349
350 /*
351 * This function always returns success
352 */
353 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
354 {
355 struct f2fs_nm_info *nm_i = NM_I(sbi);
356 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
357 struct f2fs_summary_block *sum = curseg->sum_blk;
358 nid_t start_nid = START_NID(nid);
359 struct f2fs_nat_block *nat_blk;
360 struct page *page = NULL;
361 struct f2fs_nat_entry ne;
362 struct nat_entry *e;
363 int i;
364
365 ni->nid = nid;
366
367 /* Check nat cache */
368 down_read(&nm_i->nat_tree_lock);
369 e = __lookup_nat_cache(nm_i, nid);
370 if (e) {
371 ni->ino = nat_get_ino(e);
372 ni->blk_addr = nat_get_blkaddr(e);
373 ni->version = nat_get_version(e);
374 }
375 up_read(&nm_i->nat_tree_lock);
376 if (e)
377 return;
378
379 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
380
381 down_write(&nm_i->nat_tree_lock);
382
383 /* Check current segment summary */
384 mutex_lock(&curseg->curseg_mutex);
385 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
386 if (i >= 0) {
387 ne = nat_in_journal(sum, i);
388 node_info_from_raw_nat(ni, &ne);
389 }
390 mutex_unlock(&curseg->curseg_mutex);
391 if (i >= 0)
392 goto cache;
393
394 /* Fill node_info from nat page */
395 page = get_current_nat_page(sbi, start_nid);
396 nat_blk = (struct f2fs_nat_block *)page_address(page);
397 ne = nat_blk->entries[nid - start_nid];
398 node_info_from_raw_nat(ni, &ne);
399 f2fs_put_page(page, 1);
400 cache:
401 /* cache nat entry */
402 cache_nat_entry(NM_I(sbi), nid, &ne);
403 up_write(&nm_i->nat_tree_lock);
404 }
405
406 /*
407 * The maximum depth is four.
408 * Offset[0] will have raw inode offset.
409 */
410 static int get_node_path(struct f2fs_inode_info *fi, long block,
411 int offset[4], unsigned int noffset[4])
412 {
413 const long direct_index = ADDRS_PER_INODE(fi);
414 const long direct_blks = ADDRS_PER_BLOCK;
415 const long dptrs_per_blk = NIDS_PER_BLOCK;
416 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
417 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
418 int n = 0;
419 int level = 0;
420
421 noffset[0] = 0;
422
423 if (block < direct_index) {
424 offset[n] = block;
425 goto got;
426 }
427 block -= direct_index;
428 if (block < direct_blks) {
429 offset[n++] = NODE_DIR1_BLOCK;
430 noffset[n] = 1;
431 offset[n] = block;
432 level = 1;
433 goto got;
434 }
435 block -= direct_blks;
436 if (block < direct_blks) {
437 offset[n++] = NODE_DIR2_BLOCK;
438 noffset[n] = 2;
439 offset[n] = block;
440 level = 1;
441 goto got;
442 }
443 block -= direct_blks;
444 if (block < indirect_blks) {
445 offset[n++] = NODE_IND1_BLOCK;
446 noffset[n] = 3;
447 offset[n++] = block / direct_blks;
448 noffset[n] = 4 + offset[n - 1];
449 offset[n] = block % direct_blks;
450 level = 2;
451 goto got;
452 }
453 block -= indirect_blks;
454 if (block < indirect_blks) {
455 offset[n++] = NODE_IND2_BLOCK;
456 noffset[n] = 4 + dptrs_per_blk;
457 offset[n++] = block / direct_blks;
458 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
459 offset[n] = block % direct_blks;
460 level = 2;
461 goto got;
462 }
463 block -= indirect_blks;
464 if (block < dindirect_blks) {
465 offset[n++] = NODE_DIND_BLOCK;
466 noffset[n] = 5 + (dptrs_per_blk * 2);
467 offset[n++] = block / indirect_blks;
468 noffset[n] = 6 + (dptrs_per_blk * 2) +
469 offset[n - 1] * (dptrs_per_blk + 1);
470 offset[n++] = (block / direct_blks) % dptrs_per_blk;
471 noffset[n] = 7 + (dptrs_per_blk * 2) +
472 offset[n - 2] * (dptrs_per_blk + 1) +
473 offset[n - 1];
474 offset[n] = block % direct_blks;
475 level = 3;
476 goto got;
477 } else {
478 BUG();
479 }
480 got:
481 return level;
482 }
483
484 /*
485 * Caller should call f2fs_put_dnode(dn).
486 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
487 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
488 * In the case of RDONLY_NODE, we don't need to care about mutex.
489 */
490 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
491 {
492 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
493 struct page *npage[4];
494 struct page *parent = NULL;
495 int offset[4];
496 unsigned int noffset[4];
497 nid_t nids[4];
498 int level, i;
499 int err = 0;
500
501 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
502
503 nids[0] = dn->inode->i_ino;
504 npage[0] = dn->inode_page;
505
506 if (!npage[0]) {
507 npage[0] = get_node_page(sbi, nids[0]);
508 if (IS_ERR(npage[0]))
509 return PTR_ERR(npage[0]);
510 }
511
512 /* if inline_data is set, should not report any block indices */
513 if (f2fs_has_inline_data(dn->inode) && index) {
514 err = -ENOENT;
515 f2fs_put_page(npage[0], 1);
516 goto release_out;
517 }
518
519 parent = npage[0];
520 if (level != 0)
521 nids[1] = get_nid(parent, offset[0], true);
522 dn->inode_page = npage[0];
523 dn->inode_page_locked = true;
524
525 /* get indirect or direct nodes */
526 for (i = 1; i <= level; i++) {
527 bool done = false;
528
529 if (!nids[i] && mode == ALLOC_NODE) {
530 /* alloc new node */
531 if (!alloc_nid(sbi, &(nids[i]))) {
532 err = -ENOSPC;
533 goto release_pages;
534 }
535
536 dn->nid = nids[i];
537 npage[i] = new_node_page(dn, noffset[i], NULL);
538 if (IS_ERR(npage[i])) {
539 alloc_nid_failed(sbi, nids[i]);
540 err = PTR_ERR(npage[i]);
541 goto release_pages;
542 }
543
544 set_nid(parent, offset[i - 1], nids[i], i == 1);
545 alloc_nid_done(sbi, nids[i]);
546 dn->node_changed = true;
547 done = true;
548 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
549 npage[i] = get_node_page_ra(parent, offset[i - 1]);
550 if (IS_ERR(npage[i])) {
551 err = PTR_ERR(npage[i]);
552 goto release_pages;
553 }
554 done = true;
555 }
556 if (i == 1) {
557 dn->inode_page_locked = false;
558 unlock_page(parent);
559 } else {
560 f2fs_put_page(parent, 1);
561 }
562
563 if (!done) {
564 npage[i] = get_node_page(sbi, nids[i]);
565 if (IS_ERR(npage[i])) {
566 err = PTR_ERR(npage[i]);
567 f2fs_put_page(npage[0], 0);
568 goto release_out;
569 }
570 }
571 if (i < level) {
572 parent = npage[i];
573 nids[i + 1] = get_nid(parent, offset[i], false);
574 }
575 }
576 dn->nid = nids[level];
577 dn->ofs_in_node = offset[level];
578 dn->node_page = npage[level];
579 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
580 return 0;
581
582 release_pages:
583 f2fs_put_page(parent, 1);
584 if (i > 1)
585 f2fs_put_page(npage[0], 0);
586 release_out:
587 dn->inode_page = NULL;
588 dn->node_page = NULL;
589 return err;
590 }
591
592 static void truncate_node(struct dnode_of_data *dn)
593 {
594 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
595 struct node_info ni;
596
597 get_node_info(sbi, dn->nid, &ni);
598 if (dn->inode->i_blocks == 0) {
599 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
600 goto invalidate;
601 }
602 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
603
604 /* Deallocate node address */
605 invalidate_blocks(sbi, ni.blk_addr);
606 dec_valid_node_count(sbi, dn->inode);
607 set_node_addr(sbi, &ni, NULL_ADDR, false);
608
609 if (dn->nid == dn->inode->i_ino) {
610 remove_orphan_inode(sbi, dn->nid);
611 dec_valid_inode_count(sbi);
612 } else {
613 sync_inode_page(dn);
614 }
615 invalidate:
616 clear_node_page_dirty(dn->node_page);
617 set_sbi_flag(sbi, SBI_IS_DIRTY);
618
619 f2fs_put_page(dn->node_page, 1);
620
621 invalidate_mapping_pages(NODE_MAPPING(sbi),
622 dn->node_page->index, dn->node_page->index);
623
624 dn->node_page = NULL;
625 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
626 }
627
628 static int truncate_dnode(struct dnode_of_data *dn)
629 {
630 struct page *page;
631
632 if (dn->nid == 0)
633 return 1;
634
635 /* get direct node */
636 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
637 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
638 return 1;
639 else if (IS_ERR(page))
640 return PTR_ERR(page);
641
642 /* Make dnode_of_data for parameter */
643 dn->node_page = page;
644 dn->ofs_in_node = 0;
645 truncate_data_blocks(dn);
646 truncate_node(dn);
647 return 1;
648 }
649
650 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
651 int ofs, int depth)
652 {
653 struct dnode_of_data rdn = *dn;
654 struct page *page;
655 struct f2fs_node *rn;
656 nid_t child_nid;
657 unsigned int child_nofs;
658 int freed = 0;
659 int i, ret;
660
661 if (dn->nid == 0)
662 return NIDS_PER_BLOCK + 1;
663
664 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
665
666 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
667 if (IS_ERR(page)) {
668 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
669 return PTR_ERR(page);
670 }
671
672 rn = F2FS_NODE(page);
673 if (depth < 3) {
674 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
675 child_nid = le32_to_cpu(rn->in.nid[i]);
676 if (child_nid == 0)
677 continue;
678 rdn.nid = child_nid;
679 ret = truncate_dnode(&rdn);
680 if (ret < 0)
681 goto out_err;
682 set_nid(page, i, 0, false);
683 dn->node_changed = true;
684 }
685 } else {
686 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
687 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
688 child_nid = le32_to_cpu(rn->in.nid[i]);
689 if (child_nid == 0) {
690 child_nofs += NIDS_PER_BLOCK + 1;
691 continue;
692 }
693 rdn.nid = child_nid;
694 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
695 if (ret == (NIDS_PER_BLOCK + 1)) {
696 set_nid(page, i, 0, false);
697 dn->node_changed = true;
698 child_nofs += ret;
699 } else if (ret < 0 && ret != -ENOENT) {
700 goto out_err;
701 }
702 }
703 freed = child_nofs;
704 }
705
706 if (!ofs) {
707 /* remove current indirect node */
708 dn->node_page = page;
709 truncate_node(dn);
710 freed++;
711 } else {
712 f2fs_put_page(page, 1);
713 }
714 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
715 return freed;
716
717 out_err:
718 f2fs_put_page(page, 1);
719 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
720 return ret;
721 }
722
723 static int truncate_partial_nodes(struct dnode_of_data *dn,
724 struct f2fs_inode *ri, int *offset, int depth)
725 {
726 struct page *pages[2];
727 nid_t nid[3];
728 nid_t child_nid;
729 int err = 0;
730 int i;
731 int idx = depth - 2;
732
733 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
734 if (!nid[0])
735 return 0;
736
737 /* get indirect nodes in the path */
738 for (i = 0; i < idx + 1; i++) {
739 /* reference count'll be increased */
740 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
741 if (IS_ERR(pages[i])) {
742 err = PTR_ERR(pages[i]);
743 idx = i - 1;
744 goto fail;
745 }
746 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
747 }
748
749 /* free direct nodes linked to a partial indirect node */
750 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
751 child_nid = get_nid(pages[idx], i, false);
752 if (!child_nid)
753 continue;
754 dn->nid = child_nid;
755 err = truncate_dnode(dn);
756 if (err < 0)
757 goto fail;
758 set_nid(pages[idx], i, 0, false);
759 dn->node_changed = true;
760 }
761
762 if (offset[idx + 1] == 0) {
763 dn->node_page = pages[idx];
764 dn->nid = nid[idx];
765 truncate_node(dn);
766 } else {
767 f2fs_put_page(pages[idx], 1);
768 }
769 offset[idx]++;
770 offset[idx + 1] = 0;
771 idx--;
772 fail:
773 for (i = idx; i >= 0; i--)
774 f2fs_put_page(pages[i], 1);
775
776 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
777
778 return err;
779 }
780
781 /*
782 * All the block addresses of data and nodes should be nullified.
783 */
784 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
785 {
786 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
787 int err = 0, cont = 1;
788 int level, offset[4], noffset[4];
789 unsigned int nofs = 0;
790 struct f2fs_inode *ri;
791 struct dnode_of_data dn;
792 struct page *page;
793
794 trace_f2fs_truncate_inode_blocks_enter(inode, from);
795
796 level = get_node_path(F2FS_I(inode), from, offset, noffset);
797 restart:
798 page = get_node_page(sbi, inode->i_ino);
799 if (IS_ERR(page)) {
800 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
801 return PTR_ERR(page);
802 }
803
804 set_new_dnode(&dn, inode, page, NULL, 0);
805 unlock_page(page);
806
807 ri = F2FS_INODE(page);
808 switch (level) {
809 case 0:
810 case 1:
811 nofs = noffset[1];
812 break;
813 case 2:
814 nofs = noffset[1];
815 if (!offset[level - 1])
816 goto skip_partial;
817 err = truncate_partial_nodes(&dn, ri, offset, level);
818 if (err < 0 && err != -ENOENT)
819 goto fail;
820 nofs += 1 + NIDS_PER_BLOCK;
821 break;
822 case 3:
823 nofs = 5 + 2 * NIDS_PER_BLOCK;
824 if (!offset[level - 1])
825 goto skip_partial;
826 err = truncate_partial_nodes(&dn, ri, offset, level);
827 if (err < 0 && err != -ENOENT)
828 goto fail;
829 break;
830 default:
831 BUG();
832 }
833
834 skip_partial:
835 while (cont) {
836 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
837 switch (offset[0]) {
838 case NODE_DIR1_BLOCK:
839 case NODE_DIR2_BLOCK:
840 err = truncate_dnode(&dn);
841 break;
842
843 case NODE_IND1_BLOCK:
844 case NODE_IND2_BLOCK:
845 err = truncate_nodes(&dn, nofs, offset[1], 2);
846 break;
847
848 case NODE_DIND_BLOCK:
849 err = truncate_nodes(&dn, nofs, offset[1], 3);
850 cont = 0;
851 break;
852
853 default:
854 BUG();
855 }
856 if (err < 0 && err != -ENOENT)
857 goto fail;
858 if (offset[1] == 0 &&
859 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
860 lock_page(page);
861 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
862 f2fs_put_page(page, 1);
863 goto restart;
864 }
865 f2fs_wait_on_page_writeback(page, NODE);
866 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
867 set_page_dirty(page);
868 unlock_page(page);
869 }
870 offset[1] = 0;
871 offset[0]++;
872 nofs += err;
873 }
874 fail:
875 f2fs_put_page(page, 0);
876 trace_f2fs_truncate_inode_blocks_exit(inode, err);
877 return err > 0 ? 0 : err;
878 }
879
880 int truncate_xattr_node(struct inode *inode, struct page *page)
881 {
882 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
883 nid_t nid = F2FS_I(inode)->i_xattr_nid;
884 struct dnode_of_data dn;
885 struct page *npage;
886
887 if (!nid)
888 return 0;
889
890 npage = get_node_page(sbi, nid);
891 if (IS_ERR(npage))
892 return PTR_ERR(npage);
893
894 F2FS_I(inode)->i_xattr_nid = 0;
895
896 /* need to do checkpoint during fsync */
897 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
898
899 set_new_dnode(&dn, inode, page, npage, nid);
900
901 if (page)
902 dn.inode_page_locked = true;
903 truncate_node(&dn);
904 return 0;
905 }
906
907 /*
908 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
909 * f2fs_unlock_op().
910 */
911 int remove_inode_page(struct inode *inode)
912 {
913 struct dnode_of_data dn;
914 int err;
915
916 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
917 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
918 if (err)
919 return err;
920
921 err = truncate_xattr_node(inode, dn.inode_page);
922 if (err) {
923 f2fs_put_dnode(&dn);
924 return err;
925 }
926
927 /* remove potential inline_data blocks */
928 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
929 S_ISLNK(inode->i_mode))
930 truncate_data_blocks_range(&dn, 1);
931
932 /* 0 is possible, after f2fs_new_inode() has failed */
933 f2fs_bug_on(F2FS_I_SB(inode),
934 inode->i_blocks != 0 && inode->i_blocks != 1);
935
936 /* will put inode & node pages */
937 truncate_node(&dn);
938 return 0;
939 }
940
941 struct page *new_inode_page(struct inode *inode)
942 {
943 struct dnode_of_data dn;
944
945 /* allocate inode page for new inode */
946 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
947
948 /* caller should f2fs_put_page(page, 1); */
949 return new_node_page(&dn, 0, NULL);
950 }
951
952 struct page *new_node_page(struct dnode_of_data *dn,
953 unsigned int ofs, struct page *ipage)
954 {
955 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
956 struct node_info old_ni, new_ni;
957 struct page *page;
958 int err;
959
960 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
961 return ERR_PTR(-EPERM);
962
963 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
964 if (!page)
965 return ERR_PTR(-ENOMEM);
966
967 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
968 err = -ENOSPC;
969 goto fail;
970 }
971
972 get_node_info(sbi, dn->nid, &old_ni);
973
974 /* Reinitialize old_ni with new node page */
975 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
976 new_ni = old_ni;
977 new_ni.ino = dn->inode->i_ino;
978 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
979
980 f2fs_wait_on_page_writeback(page, NODE);
981 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
982 set_cold_node(dn->inode, page);
983 SetPageUptodate(page);
984 set_page_dirty(page);
985
986 if (f2fs_has_xattr_block(ofs))
987 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
988
989 dn->node_page = page;
990 if (ipage)
991 update_inode(dn->inode, ipage);
992 else
993 sync_inode_page(dn);
994 if (ofs == 0)
995 inc_valid_inode_count(sbi);
996
997 return page;
998
999 fail:
1000 clear_node_page_dirty(page);
1001 f2fs_put_page(page, 1);
1002 return ERR_PTR(err);
1003 }
1004
1005 /*
1006 * Caller should do after getting the following values.
1007 * 0: f2fs_put_page(page, 0)
1008 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1009 */
1010 static int read_node_page(struct page *page, int rw)
1011 {
1012 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1013 struct node_info ni;
1014 struct f2fs_io_info fio = {
1015 .sbi = sbi,
1016 .type = NODE,
1017 .rw = rw,
1018 .page = page,
1019 .encrypted_page = NULL,
1020 };
1021
1022 get_node_info(sbi, page->index, &ni);
1023
1024 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1025 ClearPageUptodate(page);
1026 return -ENOENT;
1027 }
1028
1029 if (PageUptodate(page))
1030 return LOCKED_PAGE;
1031
1032 fio.blk_addr = ni.blk_addr;
1033 return f2fs_submit_page_bio(&fio);
1034 }
1035
1036 /*
1037 * Readahead a node page
1038 */
1039 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1040 {
1041 struct page *apage;
1042 int err;
1043
1044 if (!nid)
1045 return;
1046 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1047
1048 apage = find_get_page(NODE_MAPPING(sbi), nid);
1049 if (apage && PageUptodate(apage)) {
1050 f2fs_put_page(apage, 0);
1051 return;
1052 }
1053 f2fs_put_page(apage, 0);
1054
1055 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1056 if (!apage)
1057 return;
1058
1059 err = read_node_page(apage, READA);
1060 f2fs_put_page(apage, err ? 1 : 0);
1061 }
1062
1063 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1064 {
1065 struct page *page;
1066 int err;
1067
1068 if (!nid)
1069 return ERR_PTR(-ENOENT);
1070 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1071 repeat:
1072 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1073 if (!page)
1074 return ERR_PTR(-ENOMEM);
1075
1076 err = read_node_page(page, READ_SYNC);
1077 if (err < 0) {
1078 f2fs_put_page(page, 1);
1079 return ERR_PTR(err);
1080 } else if (err == LOCKED_PAGE) {
1081 goto page_hit;
1082 }
1083
1084 lock_page(page);
1085
1086 if (unlikely(!PageUptodate(page))) {
1087 f2fs_put_page(page, 1);
1088 return ERR_PTR(-EIO);
1089 }
1090 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1091 f2fs_put_page(page, 1);
1092 goto repeat;
1093 }
1094 page_hit:
1095 f2fs_bug_on(sbi, nid != nid_of_node(page));
1096 return page;
1097 }
1098
1099 /*
1100 * Return a locked page for the desired node page.
1101 * And, readahead MAX_RA_NODE number of node pages.
1102 */
1103 struct page *get_node_page_ra(struct page *parent, int start)
1104 {
1105 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1106 struct blk_plug plug;
1107 struct page *page;
1108 int err, i, end;
1109 nid_t nid;
1110
1111 /* First, try getting the desired direct node. */
1112 nid = get_nid(parent, start, false);
1113 if (!nid)
1114 return ERR_PTR(-ENOENT);
1115 f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1116 repeat:
1117 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1118 if (!page)
1119 return ERR_PTR(-ENOMEM);
1120
1121 err = read_node_page(page, READ_SYNC);
1122 if (err < 0) {
1123 f2fs_put_page(page, 1);
1124 return ERR_PTR(err);
1125 } else if (err == LOCKED_PAGE) {
1126 goto page_hit;
1127 }
1128
1129 blk_start_plug(&plug);
1130
1131 /* Then, try readahead for siblings of the desired node */
1132 end = start + MAX_RA_NODE;
1133 end = min(end, NIDS_PER_BLOCK);
1134 for (i = start + 1; i < end; i++) {
1135 nid_t tnid;
1136
1137 tnid = get_nid(parent, i, false);
1138 ra_node_page(sbi, tnid);
1139 }
1140
1141 blk_finish_plug(&plug);
1142
1143 lock_page(page);
1144 if (unlikely(!PageUptodate(page))) {
1145 f2fs_put_page(page, 1);
1146 return ERR_PTR(-EIO);
1147 }
1148 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1149 f2fs_put_page(page, 1);
1150 goto repeat;
1151 }
1152 page_hit:
1153 f2fs_bug_on(sbi, nid != nid_of_node(page));
1154 return page;
1155 }
1156
1157 void sync_inode_page(struct dnode_of_data *dn)
1158 {
1159 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1160 update_inode(dn->inode, dn->node_page);
1161 } else if (dn->inode_page) {
1162 if (!dn->inode_page_locked)
1163 lock_page(dn->inode_page);
1164 update_inode(dn->inode, dn->inode_page);
1165 if (!dn->inode_page_locked)
1166 unlock_page(dn->inode_page);
1167 } else {
1168 update_inode_page(dn->inode);
1169 }
1170 dn->node_changed = true;
1171 }
1172
1173 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1174 struct writeback_control *wbc)
1175 {
1176 pgoff_t index, end;
1177 struct pagevec pvec;
1178 int step = ino ? 2 : 0;
1179 int nwritten = 0, wrote = 0;
1180
1181 pagevec_init(&pvec, 0);
1182
1183 next_step:
1184 index = 0;
1185 end = LONG_MAX;
1186
1187 while (index <= end) {
1188 int i, nr_pages;
1189 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1190 PAGECACHE_TAG_DIRTY,
1191 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1192 if (nr_pages == 0)
1193 break;
1194
1195 for (i = 0; i < nr_pages; i++) {
1196 struct page *page = pvec.pages[i];
1197
1198 if (unlikely(f2fs_cp_error(sbi))) {
1199 pagevec_release(&pvec);
1200 return -EIO;
1201 }
1202
1203 /*
1204 * flushing sequence with step:
1205 * 0. indirect nodes
1206 * 1. dentry dnodes
1207 * 2. file dnodes
1208 */
1209 if (step == 0 && IS_DNODE(page))
1210 continue;
1211 if (step == 1 && (!IS_DNODE(page) ||
1212 is_cold_node(page)))
1213 continue;
1214 if (step == 2 && (!IS_DNODE(page) ||
1215 !is_cold_node(page)))
1216 continue;
1217
1218 /*
1219 * If an fsync mode,
1220 * we should not skip writing node pages.
1221 */
1222 if (ino && ino_of_node(page) == ino)
1223 lock_page(page);
1224 else if (!trylock_page(page))
1225 continue;
1226
1227 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1228 continue_unlock:
1229 unlock_page(page);
1230 continue;
1231 }
1232 if (ino && ino_of_node(page) != ino)
1233 goto continue_unlock;
1234
1235 if (!PageDirty(page)) {
1236 /* someone wrote it for us */
1237 goto continue_unlock;
1238 }
1239
1240 if (!clear_page_dirty_for_io(page))
1241 goto continue_unlock;
1242
1243 /* called by fsync() */
1244 if (ino && IS_DNODE(page)) {
1245 set_fsync_mark(page, 1);
1246 if (IS_INODE(page))
1247 set_dentry_mark(page,
1248 need_dentry_mark(sbi, ino));
1249 nwritten++;
1250 } else {
1251 set_fsync_mark(page, 0);
1252 set_dentry_mark(page, 0);
1253 }
1254
1255 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1256 unlock_page(page);
1257 else
1258 wrote++;
1259
1260 if (--wbc->nr_to_write == 0)
1261 break;
1262 }
1263 pagevec_release(&pvec);
1264 cond_resched();
1265
1266 if (wbc->nr_to_write == 0) {
1267 step = 2;
1268 break;
1269 }
1270 }
1271
1272 if (step < 2) {
1273 step++;
1274 goto next_step;
1275 }
1276
1277 if (wrote)
1278 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1279 return nwritten;
1280 }
1281
1282 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1283 {
1284 pgoff_t index = 0, end = LONG_MAX;
1285 struct pagevec pvec;
1286 int ret2 = 0, ret = 0;
1287
1288 pagevec_init(&pvec, 0);
1289
1290 while (index <= end) {
1291 int i, nr_pages;
1292 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1293 PAGECACHE_TAG_WRITEBACK,
1294 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1295 if (nr_pages == 0)
1296 break;
1297
1298 for (i = 0; i < nr_pages; i++) {
1299 struct page *page = pvec.pages[i];
1300
1301 /* until radix tree lookup accepts end_index */
1302 if (unlikely(page->index > end))
1303 continue;
1304
1305 if (ino && ino_of_node(page) == ino) {
1306 f2fs_wait_on_page_writeback(page, NODE);
1307 if (TestClearPageError(page))
1308 ret = -EIO;
1309 }
1310 }
1311 pagevec_release(&pvec);
1312 cond_resched();
1313 }
1314
1315 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1316 ret2 = -ENOSPC;
1317 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1318 ret2 = -EIO;
1319 if (!ret)
1320 ret = ret2;
1321 return ret;
1322 }
1323
1324 static int f2fs_write_node_page(struct page *page,
1325 struct writeback_control *wbc)
1326 {
1327 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1328 nid_t nid;
1329 struct node_info ni;
1330 struct f2fs_io_info fio = {
1331 .sbi = sbi,
1332 .type = NODE,
1333 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1334 .page = page,
1335 .encrypted_page = NULL,
1336 };
1337
1338 trace_f2fs_writepage(page, NODE);
1339
1340 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1341 goto redirty_out;
1342 if (unlikely(f2fs_cp_error(sbi)))
1343 goto redirty_out;
1344
1345 f2fs_wait_on_page_writeback(page, NODE);
1346
1347 /* get old block addr of this node page */
1348 nid = nid_of_node(page);
1349 f2fs_bug_on(sbi, page->index != nid);
1350
1351 if (wbc->for_reclaim) {
1352 if (!down_read_trylock(&sbi->node_write))
1353 goto redirty_out;
1354 } else {
1355 down_read(&sbi->node_write);
1356 }
1357
1358 get_node_info(sbi, nid, &ni);
1359
1360 /* This page is already truncated */
1361 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1362 ClearPageUptodate(page);
1363 dec_page_count(sbi, F2FS_DIRTY_NODES);
1364 up_read(&sbi->node_write);
1365 unlock_page(page);
1366 return 0;
1367 }
1368
1369 set_page_writeback(page);
1370 fio.blk_addr = ni.blk_addr;
1371 write_node_page(nid, &fio);
1372 set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1373 dec_page_count(sbi, F2FS_DIRTY_NODES);
1374 up_read(&sbi->node_write);
1375 unlock_page(page);
1376
1377 if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi)))
1378 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1379
1380 return 0;
1381
1382 redirty_out:
1383 redirty_page_for_writepage(wbc, page);
1384 return AOP_WRITEPAGE_ACTIVATE;
1385 }
1386
1387 static int f2fs_write_node_pages(struct address_space *mapping,
1388 struct writeback_control *wbc)
1389 {
1390 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1391 long diff;
1392
1393 trace_f2fs_writepages(mapping->host, wbc, NODE);
1394
1395 /* balancing f2fs's metadata in background */
1396 f2fs_balance_fs_bg(sbi);
1397
1398 /* collect a number of dirty node pages and write together */
1399 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1400 goto skip_write;
1401
1402 diff = nr_pages_to_write(sbi, NODE, wbc);
1403 wbc->sync_mode = WB_SYNC_NONE;
1404 sync_node_pages(sbi, 0, wbc);
1405 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1406 return 0;
1407
1408 skip_write:
1409 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1410 return 0;
1411 }
1412
1413 static int f2fs_set_node_page_dirty(struct page *page)
1414 {
1415 trace_f2fs_set_page_dirty(page, NODE);
1416
1417 SetPageUptodate(page);
1418 if (!PageDirty(page)) {
1419 __set_page_dirty_nobuffers(page);
1420 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1421 SetPagePrivate(page);
1422 f2fs_trace_pid(page);
1423 return 1;
1424 }
1425 return 0;
1426 }
1427
1428 /*
1429 * Structure of the f2fs node operations
1430 */
1431 const struct address_space_operations f2fs_node_aops = {
1432 .writepage = f2fs_write_node_page,
1433 .writepages = f2fs_write_node_pages,
1434 .set_page_dirty = f2fs_set_node_page_dirty,
1435 .invalidatepage = f2fs_invalidate_page,
1436 .releasepage = f2fs_release_page,
1437 };
1438
1439 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1440 nid_t n)
1441 {
1442 return radix_tree_lookup(&nm_i->free_nid_root, n);
1443 }
1444
1445 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1446 struct free_nid *i)
1447 {
1448 list_del(&i->list);
1449 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1450 }
1451
1452 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1453 {
1454 struct f2fs_nm_info *nm_i = NM_I(sbi);
1455 struct free_nid *i;
1456 struct nat_entry *ne;
1457 bool allocated = false;
1458
1459 if (!available_free_memory(sbi, FREE_NIDS))
1460 return -1;
1461
1462 /* 0 nid should not be used */
1463 if (unlikely(nid == 0))
1464 return 0;
1465
1466 if (build) {
1467 /* do not add allocated nids */
1468 ne = __lookup_nat_cache(nm_i, nid);
1469 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1470 nat_get_blkaddr(ne) != NULL_ADDR))
1471 allocated = true;
1472 if (allocated)
1473 return 0;
1474 }
1475
1476 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1477 i->nid = nid;
1478 i->state = NID_NEW;
1479
1480 if (radix_tree_preload(GFP_NOFS)) {
1481 kmem_cache_free(free_nid_slab, i);
1482 return 0;
1483 }
1484
1485 spin_lock(&nm_i->free_nid_list_lock);
1486 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1487 spin_unlock(&nm_i->free_nid_list_lock);
1488 radix_tree_preload_end();
1489 kmem_cache_free(free_nid_slab, i);
1490 return 0;
1491 }
1492 list_add_tail(&i->list, &nm_i->free_nid_list);
1493 nm_i->fcnt++;
1494 spin_unlock(&nm_i->free_nid_list_lock);
1495 radix_tree_preload_end();
1496 return 1;
1497 }
1498
1499 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1500 {
1501 struct free_nid *i;
1502 bool need_free = false;
1503
1504 spin_lock(&nm_i->free_nid_list_lock);
1505 i = __lookup_free_nid_list(nm_i, nid);
1506 if (i && i->state == NID_NEW) {
1507 __del_from_free_nid_list(nm_i, i);
1508 nm_i->fcnt--;
1509 need_free = true;
1510 }
1511 spin_unlock(&nm_i->free_nid_list_lock);
1512
1513 if (need_free)
1514 kmem_cache_free(free_nid_slab, i);
1515 }
1516
1517 static void scan_nat_page(struct f2fs_sb_info *sbi,
1518 struct page *nat_page, nid_t start_nid)
1519 {
1520 struct f2fs_nm_info *nm_i = NM_I(sbi);
1521 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1522 block_t blk_addr;
1523 int i;
1524
1525 i = start_nid % NAT_ENTRY_PER_BLOCK;
1526
1527 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1528
1529 if (unlikely(start_nid >= nm_i->max_nid))
1530 break;
1531
1532 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1533 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1534 if (blk_addr == NULL_ADDR) {
1535 if (add_free_nid(sbi, start_nid, true) < 0)
1536 break;
1537 }
1538 }
1539 }
1540
1541 static void build_free_nids(struct f2fs_sb_info *sbi)
1542 {
1543 struct f2fs_nm_info *nm_i = NM_I(sbi);
1544 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1545 struct f2fs_summary_block *sum = curseg->sum_blk;
1546 int i = 0;
1547 nid_t nid = nm_i->next_scan_nid;
1548
1549 /* Enough entries */
1550 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1551 return;
1552
1553 /* readahead nat pages to be scanned */
1554 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1555 META_NAT, true);
1556
1557 down_read(&nm_i->nat_tree_lock);
1558
1559 while (1) {
1560 struct page *page = get_current_nat_page(sbi, nid);
1561
1562 scan_nat_page(sbi, page, nid);
1563 f2fs_put_page(page, 1);
1564
1565 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1566 if (unlikely(nid >= nm_i->max_nid))
1567 nid = 0;
1568
1569 if (++i >= FREE_NID_PAGES)
1570 break;
1571 }
1572
1573 /* go to the next free nat pages to find free nids abundantly */
1574 nm_i->next_scan_nid = nid;
1575
1576 /* find free nids from current sum_pages */
1577 mutex_lock(&curseg->curseg_mutex);
1578 for (i = 0; i < nats_in_cursum(sum); i++) {
1579 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1580 nid = le32_to_cpu(nid_in_journal(sum, i));
1581 if (addr == NULL_ADDR)
1582 add_free_nid(sbi, nid, true);
1583 else
1584 remove_free_nid(nm_i, nid);
1585 }
1586 mutex_unlock(&curseg->curseg_mutex);
1587 up_read(&nm_i->nat_tree_lock);
1588
1589 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1590 nm_i->ra_nid_pages, META_NAT, false);
1591 }
1592
1593 /*
1594 * If this function returns success, caller can obtain a new nid
1595 * from second parameter of this function.
1596 * The returned nid could be used ino as well as nid when inode is created.
1597 */
1598 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1599 {
1600 struct f2fs_nm_info *nm_i = NM_I(sbi);
1601 struct free_nid *i = NULL;
1602 retry:
1603 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1604 return false;
1605
1606 spin_lock(&nm_i->free_nid_list_lock);
1607
1608 /* We should not use stale free nids created by build_free_nids */
1609 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1610 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1611 list_for_each_entry(i, &nm_i->free_nid_list, list)
1612 if (i->state == NID_NEW)
1613 break;
1614
1615 f2fs_bug_on(sbi, i->state != NID_NEW);
1616 *nid = i->nid;
1617 i->state = NID_ALLOC;
1618 nm_i->fcnt--;
1619 spin_unlock(&nm_i->free_nid_list_lock);
1620 return true;
1621 }
1622 spin_unlock(&nm_i->free_nid_list_lock);
1623
1624 /* Let's scan nat pages and its caches to get free nids */
1625 mutex_lock(&nm_i->build_lock);
1626 build_free_nids(sbi);
1627 mutex_unlock(&nm_i->build_lock);
1628 goto retry;
1629 }
1630
1631 /*
1632 * alloc_nid() should be called prior to this function.
1633 */
1634 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1635 {
1636 struct f2fs_nm_info *nm_i = NM_I(sbi);
1637 struct free_nid *i;
1638
1639 spin_lock(&nm_i->free_nid_list_lock);
1640 i = __lookup_free_nid_list(nm_i, nid);
1641 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1642 __del_from_free_nid_list(nm_i, i);
1643 spin_unlock(&nm_i->free_nid_list_lock);
1644
1645 kmem_cache_free(free_nid_slab, i);
1646 }
1647
1648 /*
1649 * alloc_nid() should be called prior to this function.
1650 */
1651 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1652 {
1653 struct f2fs_nm_info *nm_i = NM_I(sbi);
1654 struct free_nid *i;
1655 bool need_free = false;
1656
1657 if (!nid)
1658 return;
1659
1660 spin_lock(&nm_i->free_nid_list_lock);
1661 i = __lookup_free_nid_list(nm_i, nid);
1662 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1663 if (!available_free_memory(sbi, FREE_NIDS)) {
1664 __del_from_free_nid_list(nm_i, i);
1665 need_free = true;
1666 } else {
1667 i->state = NID_NEW;
1668 nm_i->fcnt++;
1669 }
1670 spin_unlock(&nm_i->free_nid_list_lock);
1671
1672 if (need_free)
1673 kmem_cache_free(free_nid_slab, i);
1674 }
1675
1676 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1677 {
1678 struct f2fs_nm_info *nm_i = NM_I(sbi);
1679 struct free_nid *i, *next;
1680 int nr = nr_shrink;
1681
1682 if (!mutex_trylock(&nm_i->build_lock))
1683 return 0;
1684
1685 spin_lock(&nm_i->free_nid_list_lock);
1686 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1687 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1688 break;
1689 if (i->state == NID_ALLOC)
1690 continue;
1691 __del_from_free_nid_list(nm_i, i);
1692 kmem_cache_free(free_nid_slab, i);
1693 nm_i->fcnt--;
1694 nr_shrink--;
1695 }
1696 spin_unlock(&nm_i->free_nid_list_lock);
1697 mutex_unlock(&nm_i->build_lock);
1698
1699 return nr - nr_shrink;
1700 }
1701
1702 void recover_inline_xattr(struct inode *inode, struct page *page)
1703 {
1704 void *src_addr, *dst_addr;
1705 size_t inline_size;
1706 struct page *ipage;
1707 struct f2fs_inode *ri;
1708
1709 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1710 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1711
1712 ri = F2FS_INODE(page);
1713 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1714 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1715 goto update_inode;
1716 }
1717
1718 dst_addr = inline_xattr_addr(ipage);
1719 src_addr = inline_xattr_addr(page);
1720 inline_size = inline_xattr_size(inode);
1721
1722 f2fs_wait_on_page_writeback(ipage, NODE);
1723 memcpy(dst_addr, src_addr, inline_size);
1724 update_inode:
1725 update_inode(inode, ipage);
1726 f2fs_put_page(ipage, 1);
1727 }
1728
1729 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1730 {
1731 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1732 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1733 nid_t new_xnid = nid_of_node(page);
1734 struct node_info ni;
1735
1736 /* 1: invalidate the previous xattr nid */
1737 if (!prev_xnid)
1738 goto recover_xnid;
1739
1740 /* Deallocate node address */
1741 get_node_info(sbi, prev_xnid, &ni);
1742 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1743 invalidate_blocks(sbi, ni.blk_addr);
1744 dec_valid_node_count(sbi, inode);
1745 set_node_addr(sbi, &ni, NULL_ADDR, false);
1746
1747 recover_xnid:
1748 /* 2: allocate new xattr nid */
1749 if (unlikely(!inc_valid_node_count(sbi, inode)))
1750 f2fs_bug_on(sbi, 1);
1751
1752 remove_free_nid(NM_I(sbi), new_xnid);
1753 get_node_info(sbi, new_xnid, &ni);
1754 ni.ino = inode->i_ino;
1755 set_node_addr(sbi, &ni, NEW_ADDR, false);
1756 F2FS_I(inode)->i_xattr_nid = new_xnid;
1757
1758 /* 3: update xattr blkaddr */
1759 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1760 set_node_addr(sbi, &ni, blkaddr, false);
1761
1762 update_inode_page(inode);
1763 }
1764
1765 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1766 {
1767 struct f2fs_inode *src, *dst;
1768 nid_t ino = ino_of_node(page);
1769 struct node_info old_ni, new_ni;
1770 struct page *ipage;
1771
1772 get_node_info(sbi, ino, &old_ni);
1773
1774 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1775 return -EINVAL;
1776
1777 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1778 if (!ipage)
1779 return -ENOMEM;
1780
1781 /* Should not use this inode from free nid list */
1782 remove_free_nid(NM_I(sbi), ino);
1783
1784 SetPageUptodate(ipage);
1785 fill_node_footer(ipage, ino, ino, 0, true);
1786
1787 src = F2FS_INODE(page);
1788 dst = F2FS_INODE(ipage);
1789
1790 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1791 dst->i_size = 0;
1792 dst->i_blocks = cpu_to_le64(1);
1793 dst->i_links = cpu_to_le32(1);
1794 dst->i_xattr_nid = 0;
1795 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1796
1797 new_ni = old_ni;
1798 new_ni.ino = ino;
1799
1800 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1801 WARN_ON(1);
1802 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1803 inc_valid_inode_count(sbi);
1804 set_page_dirty(ipage);
1805 f2fs_put_page(ipage, 1);
1806 return 0;
1807 }
1808
1809 int restore_node_summary(struct f2fs_sb_info *sbi,
1810 unsigned int segno, struct f2fs_summary_block *sum)
1811 {
1812 struct f2fs_node *rn;
1813 struct f2fs_summary *sum_entry;
1814 block_t addr;
1815 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1816 int i, idx, last_offset, nrpages;
1817
1818 /* scan the node segment */
1819 last_offset = sbi->blocks_per_seg;
1820 addr = START_BLOCK(sbi, segno);
1821 sum_entry = &sum->entries[0];
1822
1823 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1824 nrpages = min(last_offset - i, bio_blocks);
1825
1826 /* readahead node pages */
1827 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1828
1829 for (idx = addr; idx < addr + nrpages; idx++) {
1830 struct page *page = get_tmp_page(sbi, idx);
1831
1832 rn = F2FS_NODE(page);
1833 sum_entry->nid = rn->footer.nid;
1834 sum_entry->version = 0;
1835 sum_entry->ofs_in_node = 0;
1836 sum_entry++;
1837 f2fs_put_page(page, 1);
1838 }
1839
1840 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1841 addr + nrpages);
1842 }
1843 return 0;
1844 }
1845
1846 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1847 {
1848 struct f2fs_nm_info *nm_i = NM_I(sbi);
1849 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1850 struct f2fs_summary_block *sum = curseg->sum_blk;
1851 int i;
1852
1853 mutex_lock(&curseg->curseg_mutex);
1854 for (i = 0; i < nats_in_cursum(sum); i++) {
1855 struct nat_entry *ne;
1856 struct f2fs_nat_entry raw_ne;
1857 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1858
1859 raw_ne = nat_in_journal(sum, i);
1860
1861 ne = __lookup_nat_cache(nm_i, nid);
1862 if (!ne) {
1863 ne = grab_nat_entry(nm_i, nid);
1864 node_info_from_raw_nat(&ne->ni, &raw_ne);
1865 }
1866 __set_nat_cache_dirty(nm_i, ne);
1867 }
1868 update_nats_in_cursum(sum, -i);
1869 mutex_unlock(&curseg->curseg_mutex);
1870 }
1871
1872 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1873 struct list_head *head, int max)
1874 {
1875 struct nat_entry_set *cur;
1876
1877 if (nes->entry_cnt >= max)
1878 goto add_out;
1879
1880 list_for_each_entry(cur, head, set_list) {
1881 if (cur->entry_cnt >= nes->entry_cnt) {
1882 list_add(&nes->set_list, cur->set_list.prev);
1883 return;
1884 }
1885 }
1886 add_out:
1887 list_add_tail(&nes->set_list, head);
1888 }
1889
1890 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1891 struct nat_entry_set *set)
1892 {
1893 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1894 struct f2fs_summary_block *sum = curseg->sum_blk;
1895 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1896 bool to_journal = true;
1897 struct f2fs_nat_block *nat_blk;
1898 struct nat_entry *ne, *cur;
1899 struct page *page = NULL;
1900
1901 /*
1902 * there are two steps to flush nat entries:
1903 * #1, flush nat entries to journal in current hot data summary block.
1904 * #2, flush nat entries to nat page.
1905 */
1906 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1907 to_journal = false;
1908
1909 if (to_journal) {
1910 mutex_lock(&curseg->curseg_mutex);
1911 } else {
1912 page = get_next_nat_page(sbi, start_nid);
1913 nat_blk = page_address(page);
1914 f2fs_bug_on(sbi, !nat_blk);
1915 }
1916
1917 /* flush dirty nats in nat entry set */
1918 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1919 struct f2fs_nat_entry *raw_ne;
1920 nid_t nid = nat_get_nid(ne);
1921 int offset;
1922
1923 if (nat_get_blkaddr(ne) == NEW_ADDR)
1924 continue;
1925
1926 if (to_journal) {
1927 offset = lookup_journal_in_cursum(sum,
1928 NAT_JOURNAL, nid, 1);
1929 f2fs_bug_on(sbi, offset < 0);
1930 raw_ne = &nat_in_journal(sum, offset);
1931 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1932 } else {
1933 raw_ne = &nat_blk->entries[nid - start_nid];
1934 }
1935 raw_nat_from_node_info(raw_ne, &ne->ni);
1936 nat_reset_flag(ne);
1937 __clear_nat_cache_dirty(NM_I(sbi), ne);
1938 if (nat_get_blkaddr(ne) == NULL_ADDR)
1939 add_free_nid(sbi, nid, false);
1940 }
1941
1942 if (to_journal)
1943 mutex_unlock(&curseg->curseg_mutex);
1944 else
1945 f2fs_put_page(page, 1);
1946
1947 f2fs_bug_on(sbi, set->entry_cnt);
1948
1949 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1950 kmem_cache_free(nat_entry_set_slab, set);
1951 }
1952
1953 /*
1954 * This function is called during the checkpointing process.
1955 */
1956 void flush_nat_entries(struct f2fs_sb_info *sbi)
1957 {
1958 struct f2fs_nm_info *nm_i = NM_I(sbi);
1959 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1960 struct f2fs_summary_block *sum = curseg->sum_blk;
1961 struct nat_entry_set *setvec[SETVEC_SIZE];
1962 struct nat_entry_set *set, *tmp;
1963 unsigned int found;
1964 nid_t set_idx = 0;
1965 LIST_HEAD(sets);
1966
1967 if (!nm_i->dirty_nat_cnt)
1968 return;
1969
1970 down_write(&nm_i->nat_tree_lock);
1971
1972 /*
1973 * if there are no enough space in journal to store dirty nat
1974 * entries, remove all entries from journal and merge them
1975 * into nat entry set.
1976 */
1977 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1978 remove_nats_in_journal(sbi);
1979
1980 while ((found = __gang_lookup_nat_set(nm_i,
1981 set_idx, SETVEC_SIZE, setvec))) {
1982 unsigned idx;
1983 set_idx = setvec[found - 1]->set + 1;
1984 for (idx = 0; idx < found; idx++)
1985 __adjust_nat_entry_set(setvec[idx], &sets,
1986 MAX_NAT_JENTRIES(sum));
1987 }
1988
1989 /* flush dirty nats in nat entry set */
1990 list_for_each_entry_safe(set, tmp, &sets, set_list)
1991 __flush_nat_entry_set(sbi, set);
1992
1993 up_write(&nm_i->nat_tree_lock);
1994
1995 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1996 }
1997
1998 static int init_node_manager(struct f2fs_sb_info *sbi)
1999 {
2000 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2001 struct f2fs_nm_info *nm_i = NM_I(sbi);
2002 unsigned char *version_bitmap;
2003 unsigned int nat_segs, nat_blocks;
2004
2005 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2006
2007 /* segment_count_nat includes pair segment so divide to 2. */
2008 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2009 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2010
2011 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2012
2013 /* not used nids: 0, node, meta, (and root counted as valid node) */
2014 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2015 nm_i->fcnt = 0;
2016 nm_i->nat_cnt = 0;
2017 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2018 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2019
2020 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2021 INIT_LIST_HEAD(&nm_i->free_nid_list);
2022 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2023 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2024 INIT_LIST_HEAD(&nm_i->nat_entries);
2025
2026 mutex_init(&nm_i->build_lock);
2027 spin_lock_init(&nm_i->free_nid_list_lock);
2028 init_rwsem(&nm_i->nat_tree_lock);
2029
2030 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2031 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2032 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2033 if (!version_bitmap)
2034 return -EFAULT;
2035
2036 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2037 GFP_KERNEL);
2038 if (!nm_i->nat_bitmap)
2039 return -ENOMEM;
2040 return 0;
2041 }
2042
2043 int build_node_manager(struct f2fs_sb_info *sbi)
2044 {
2045 int err;
2046
2047 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2048 if (!sbi->nm_info)
2049 return -ENOMEM;
2050
2051 err = init_node_manager(sbi);
2052 if (err)
2053 return err;
2054
2055 build_free_nids(sbi);
2056 return 0;
2057 }
2058
2059 void destroy_node_manager(struct f2fs_sb_info *sbi)
2060 {
2061 struct f2fs_nm_info *nm_i = NM_I(sbi);
2062 struct free_nid *i, *next_i;
2063 struct nat_entry *natvec[NATVEC_SIZE];
2064 struct nat_entry_set *setvec[SETVEC_SIZE];
2065 nid_t nid = 0;
2066 unsigned int found;
2067
2068 if (!nm_i)
2069 return;
2070
2071 /* destroy free nid list */
2072 spin_lock(&nm_i->free_nid_list_lock);
2073 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2074 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2075 __del_from_free_nid_list(nm_i, i);
2076 nm_i->fcnt--;
2077 spin_unlock(&nm_i->free_nid_list_lock);
2078 kmem_cache_free(free_nid_slab, i);
2079 spin_lock(&nm_i->free_nid_list_lock);
2080 }
2081 f2fs_bug_on(sbi, nm_i->fcnt);
2082 spin_unlock(&nm_i->free_nid_list_lock);
2083
2084 /* destroy nat cache */
2085 down_write(&nm_i->nat_tree_lock);
2086 while ((found = __gang_lookup_nat_cache(nm_i,
2087 nid, NATVEC_SIZE, natvec))) {
2088 unsigned idx;
2089
2090 nid = nat_get_nid(natvec[found - 1]) + 1;
2091 for (idx = 0; idx < found; idx++)
2092 __del_from_nat_cache(nm_i, natvec[idx]);
2093 }
2094 f2fs_bug_on(sbi, nm_i->nat_cnt);
2095
2096 /* destroy nat set cache */
2097 nid = 0;
2098 while ((found = __gang_lookup_nat_set(nm_i,
2099 nid, SETVEC_SIZE, setvec))) {
2100 unsigned idx;
2101
2102 nid = setvec[found - 1]->set + 1;
2103 for (idx = 0; idx < found; idx++) {
2104 /* entry_cnt is not zero, when cp_error was occurred */
2105 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2106 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2107 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2108 }
2109 }
2110 up_write(&nm_i->nat_tree_lock);
2111
2112 kfree(nm_i->nat_bitmap);
2113 sbi->nm_info = NULL;
2114 kfree(nm_i);
2115 }
2116
2117 int __init create_node_manager_caches(void)
2118 {
2119 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2120 sizeof(struct nat_entry));
2121 if (!nat_entry_slab)
2122 goto fail;
2123
2124 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2125 sizeof(struct free_nid));
2126 if (!free_nid_slab)
2127 goto destroy_nat_entry;
2128
2129 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2130 sizeof(struct nat_entry_set));
2131 if (!nat_entry_set_slab)
2132 goto destroy_free_nid;
2133 return 0;
2134
2135 destroy_free_nid:
2136 kmem_cache_destroy(free_nid_slab);
2137 destroy_nat_entry:
2138 kmem_cache_destroy(nat_entry_slab);
2139 fail:
2140 return -ENOMEM;
2141 }
2142
2143 void destroy_node_manager_caches(void)
2144 {
2145 kmem_cache_destroy(nat_entry_set_slab);
2146 kmem_cache_destroy(free_nid_slab);
2147 kmem_cache_destroy(nat_entry_slab);
2148 }
ubuntu-zesty-kernel mirror