]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/f2fs/node.c
projects / mirror_ubuntu-bionic-kernel.git / blob
? search:


3cc32b8f82044b0a4f85fb68800ff783a2290a7b
[mirror_ubuntu-bionic-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 = (sbi->total_ext_tree * sizeof(struct extent_tree) +
69 atomic_read(&sbi->total_ext_node) *
70 sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
71 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
72 } else {
73 if (sbi->sb->s_bdi->wb.dirty_exceeded)
74 return false;
75 }
76 return res;
77 }
78
79 static void clear_node_page_dirty(struct page *page)
80 {
81 struct address_space *mapping = page->mapping;
82 unsigned int long flags;
83
84 if (PageDirty(page)) {
85 spin_lock_irqsave(&mapping->tree_lock, flags);
86 radix_tree_tag_clear(&mapping->page_tree,
87 page_index(page),
88 PAGECACHE_TAG_DIRTY);
89 spin_unlock_irqrestore(&mapping->tree_lock, flags);
90
91 clear_page_dirty_for_io(page);
92 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
93 }
94 ClearPageUptodate(page);
95 }
96
97 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
98 {
99 pgoff_t index = current_nat_addr(sbi, nid);
100 return get_meta_page(sbi, index);
101 }
102
103 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
104 {
105 struct page *src_page;
106 struct page *dst_page;
107 pgoff_t src_off;
108 pgoff_t dst_off;
109 void *src_addr;
110 void *dst_addr;
111 struct f2fs_nm_info *nm_i = NM_I(sbi);
112
113 src_off = current_nat_addr(sbi, nid);
114 dst_off = next_nat_addr(sbi, src_off);
115
116 /* get current nat block page with lock */
117 src_page = get_meta_page(sbi, src_off);
118 dst_page = grab_meta_page(sbi, dst_off);
119 f2fs_bug_on(sbi, PageDirty(src_page));
120
121 src_addr = page_address(src_page);
122 dst_addr = page_address(dst_page);
123 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
124 set_page_dirty(dst_page);
125 f2fs_put_page(src_page, 1);
126
127 set_to_next_nat(nm_i, nid);
128
129 return dst_page;
130 }
131
132 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
133 {
134 return radix_tree_lookup(&nm_i->nat_root, n);
135 }
136
137 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
138 nid_t start, unsigned int nr, struct nat_entry **ep)
139 {
140 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
141 }
142
143 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
144 {
145 list_del(&e->list);
146 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
147 nm_i->nat_cnt--;
148 kmem_cache_free(nat_entry_slab, e);
149 }
150
151 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
152 struct nat_entry *ne)
153 {
154 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
155 struct nat_entry_set *head;
156
157 if (get_nat_flag(ne, IS_DIRTY))
158 return;
159
160 head = radix_tree_lookup(&nm_i->nat_set_root, set);
161 if (!head) {
162 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
163
164 INIT_LIST_HEAD(&head->entry_list);
165 INIT_LIST_HEAD(&head->set_list);
166 head->set = set;
167 head->entry_cnt = 0;
168 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
169 }
170 list_move_tail(&ne->list, &head->entry_list);
171 nm_i->dirty_nat_cnt++;
172 head->entry_cnt++;
173 set_nat_flag(ne, IS_DIRTY, true);
174 }
175
176 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
177 struct nat_entry *ne)
178 {
179 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
180 struct nat_entry_set *head;
181
182 head = radix_tree_lookup(&nm_i->nat_set_root, set);
183 if (head) {
184 list_move_tail(&ne->list, &nm_i->nat_entries);
185 set_nat_flag(ne, IS_DIRTY, false);
186 head->entry_cnt--;
187 nm_i->dirty_nat_cnt--;
188 }
189 }
190
191 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
192 nid_t start, unsigned int nr, struct nat_entry_set **ep)
193 {
194 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
195 start, nr);
196 }
197
198 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
199 {
200 struct f2fs_nm_info *nm_i = NM_I(sbi);
201 struct nat_entry *e;
202 bool need = false;
203
204 down_read(&nm_i->nat_tree_lock);
205 e = __lookup_nat_cache(nm_i, nid);
206 if (e) {
207 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
208 !get_nat_flag(e, HAS_FSYNCED_INODE))
209 need = true;
210 }
211 up_read(&nm_i->nat_tree_lock);
212 return need;
213 }
214
215 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
216 {
217 struct f2fs_nm_info *nm_i = NM_I(sbi);
218 struct nat_entry *e;
219 bool is_cp = true;
220
221 down_read(&nm_i->nat_tree_lock);
222 e = __lookup_nat_cache(nm_i, nid);
223 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
224 is_cp = false;
225 up_read(&nm_i->nat_tree_lock);
226 return is_cp;
227 }
228
229 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
230 {
231 struct f2fs_nm_info *nm_i = NM_I(sbi);
232 struct nat_entry *e;
233 bool need_update = true;
234
235 down_read(&nm_i->nat_tree_lock);
236 e = __lookup_nat_cache(nm_i, ino);
237 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
238 (get_nat_flag(e, IS_CHECKPOINTED) ||
239 get_nat_flag(e, HAS_FSYNCED_INODE)))
240 need_update = false;
241 up_read(&nm_i->nat_tree_lock);
242 return need_update;
243 }
244
245 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
246 {
247 struct nat_entry *new;
248
249 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
250 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
251 memset(new, 0, sizeof(struct nat_entry));
252 nat_set_nid(new, nid);
253 nat_reset_flag(new);
254 list_add_tail(&new->list, &nm_i->nat_entries);
255 nm_i->nat_cnt++;
256 return new;
257 }
258
259 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
260 struct f2fs_nat_entry *ne)
261 {
262 struct nat_entry *e;
263
264 down_write(&nm_i->nat_tree_lock);
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 up_write(&nm_i->nat_tree_lock);
271 }
272
273 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
274 block_t new_blkaddr, bool fsync_done)
275 {
276 struct f2fs_nm_info *nm_i = NM_I(sbi);
277 struct nat_entry *e;
278
279 down_write(&nm_i->nat_tree_lock);
280 e = __lookup_nat_cache(nm_i, ni->nid);
281 if (!e) {
282 e = grab_nat_entry(nm_i, ni->nid);
283 copy_node_info(&e->ni, ni);
284 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
285 } else if (new_blkaddr == NEW_ADDR) {
286 /*
287 * when nid is reallocated,
288 * previous nat entry can be remained in nat cache.
289 * So, reinitialize it with new information.
290 */
291 copy_node_info(&e->ni, ni);
292 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
293 }
294
295 /* sanity check */
296 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
297 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
298 new_blkaddr == NULL_ADDR);
299 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
300 new_blkaddr == NEW_ADDR);
301 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
302 nat_get_blkaddr(e) != NULL_ADDR &&
303 new_blkaddr == NEW_ADDR);
304
305 /* increment version no as node is removed */
306 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
307 unsigned char version = nat_get_version(e);
308 nat_set_version(e, inc_node_version(version));
309
310 /* in order to reuse the nid */
311 if (nm_i->next_scan_nid > ni->nid)
312 nm_i->next_scan_nid = ni->nid;
313 }
314
315 /* change address */
316 nat_set_blkaddr(e, new_blkaddr);
317 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
318 set_nat_flag(e, IS_CHECKPOINTED, false);
319 __set_nat_cache_dirty(nm_i, e);
320
321 /* update fsync_mark if its inode nat entry is still alive */
322 if (ni->nid != ni->ino)
323 e = __lookup_nat_cache(nm_i, ni->ino);
324 if (e) {
325 if (fsync_done && ni->nid == ni->ino)
326 set_nat_flag(e, HAS_FSYNCED_INODE, true);
327 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
328 }
329 up_write(&nm_i->nat_tree_lock);
330 }
331
332 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
333 {
334 struct f2fs_nm_info *nm_i = NM_I(sbi);
335 int nr = nr_shrink;
336
337 if (!down_write_trylock(&nm_i->nat_tree_lock))
338 return 0;
339
340 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
341 struct nat_entry *ne;
342 ne = list_first_entry(&nm_i->nat_entries,
343 struct nat_entry, list);
344 __del_from_nat_cache(nm_i, ne);
345 nr_shrink--;
346 }
347 up_write(&nm_i->nat_tree_lock);
348 return nr - nr_shrink;
349 }
350
351 /*
352 * This function always returns success
353 */
354 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
355 {
356 struct f2fs_nm_info *nm_i = NM_I(sbi);
357 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
358 struct f2fs_summary_block *sum = curseg->sum_blk;
359 nid_t start_nid = START_NID(nid);
360 struct f2fs_nat_block *nat_blk;
361 struct page *page = NULL;
362 struct f2fs_nat_entry ne;
363 struct nat_entry *e;
364 int i;
365
366 ni->nid = nid;
367
368 /* Check nat cache */
369 down_read(&nm_i->nat_tree_lock);
370 e = __lookup_nat_cache(nm_i, nid);
371 if (e) {
372 ni->ino = nat_get_ino(e);
373 ni->blk_addr = nat_get_blkaddr(e);
374 ni->version = nat_get_version(e);
375 }
376 up_read(&nm_i->nat_tree_lock);
377 if (e)
378 return;
379
380 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
381
382 /* Check current segment summary */
383 mutex_lock(&curseg->curseg_mutex);
384 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
385 if (i >= 0) {
386 ne = nat_in_journal(sum, i);
387 node_info_from_raw_nat(ni, &ne);
388 }
389 mutex_unlock(&curseg->curseg_mutex);
390 if (i >= 0)
391 goto cache;
392
393 /* Fill node_info from nat page */
394 page = get_current_nat_page(sbi, start_nid);
395 nat_blk = (struct f2fs_nat_block *)page_address(page);
396 ne = nat_blk->entries[nid - start_nid];
397 node_info_from_raw_nat(ni, &ne);
398 f2fs_put_page(page, 1);
399 cache:
400 /* cache nat entry */
401 cache_nat_entry(NM_I(sbi), nid, &ne);
402 }
403
404 /*
405 * The maximum depth is four.
406 * Offset[0] will have raw inode offset.
407 */
408 static int get_node_path(struct f2fs_inode_info *fi, long block,
409 int offset[4], unsigned int noffset[4])
410 {
411 const long direct_index = ADDRS_PER_INODE(fi);
412 const long direct_blks = ADDRS_PER_BLOCK;
413 const long dptrs_per_blk = NIDS_PER_BLOCK;
414 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
415 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
416 int n = 0;
417 int level = 0;
418
419 noffset[0] = 0;
420
421 if (block < direct_index) {
422 offset[n] = block;
423 goto got;
424 }
425 block -= direct_index;
426 if (block < direct_blks) {
427 offset[n++] = NODE_DIR1_BLOCK;
428 noffset[n] = 1;
429 offset[n] = block;
430 level = 1;
431 goto got;
432 }
433 block -= direct_blks;
434 if (block < direct_blks) {
435 offset[n++] = NODE_DIR2_BLOCK;
436 noffset[n] = 2;
437 offset[n] = block;
438 level = 1;
439 goto got;
440 }
441 block -= direct_blks;
442 if (block < indirect_blks) {
443 offset[n++] = NODE_IND1_BLOCK;
444 noffset[n] = 3;
445 offset[n++] = block / direct_blks;
446 noffset[n] = 4 + offset[n - 1];
447 offset[n] = block % direct_blks;
448 level = 2;
449 goto got;
450 }
451 block -= indirect_blks;
452 if (block < indirect_blks) {
453 offset[n++] = NODE_IND2_BLOCK;
454 noffset[n] = 4 + dptrs_per_blk;
455 offset[n++] = block / direct_blks;
456 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
457 offset[n] = block % direct_blks;
458 level = 2;
459 goto got;
460 }
461 block -= indirect_blks;
462 if (block < dindirect_blks) {
463 offset[n++] = NODE_DIND_BLOCK;
464 noffset[n] = 5 + (dptrs_per_blk * 2);
465 offset[n++] = block / indirect_blks;
466 noffset[n] = 6 + (dptrs_per_blk * 2) +
467 offset[n - 1] * (dptrs_per_blk + 1);
468 offset[n++] = (block / direct_blks) % dptrs_per_blk;
469 noffset[n] = 7 + (dptrs_per_blk * 2) +
470 offset[n - 2] * (dptrs_per_blk + 1) +
471 offset[n - 1];
472 offset[n] = block % direct_blks;
473 level = 3;
474 goto got;
475 } else {
476 BUG();
477 }
478 got:
479 return level;
480 }
481
482 /*
483 * Caller should call f2fs_put_dnode(dn).
484 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
485 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
486 * In the case of RDONLY_NODE, we don't need to care about mutex.
487 */
488 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
489 {
490 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
491 struct page *npage[4];
492 struct page *parent = NULL;
493 int offset[4];
494 unsigned int noffset[4];
495 nid_t nids[4];
496 int level, i;
497 int err = 0;
498
499 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
500
501 nids[0] = dn->inode->i_ino;
502 npage[0] = dn->inode_page;
503
504 if (!npage[0]) {
505 npage[0] = get_node_page(sbi, nids[0]);
506 if (IS_ERR(npage[0]))
507 return PTR_ERR(npage[0]);
508 }
509
510 /* if inline_data is set, should not report any block indices */
511 if (f2fs_has_inline_data(dn->inode) && index) {
512 err = -ENOENT;
513 f2fs_put_page(npage[0], 1);
514 goto release_out;
515 }
516
517 parent = npage[0];
518 if (level != 0)
519 nids[1] = get_nid(parent, offset[0], true);
520 dn->inode_page = npage[0];
521 dn->inode_page_locked = true;
522
523 /* get indirect or direct nodes */
524 for (i = 1; i <= level; i++) {
525 bool done = false;
526
527 if (!nids[i] && mode == ALLOC_NODE) {
528 /* alloc new node */
529 if (!alloc_nid(sbi, &(nids[i]))) {
530 err = -ENOSPC;
531 goto release_pages;
532 }
533
534 dn->nid = nids[i];
535 npage[i] = new_node_page(dn, noffset[i], NULL);
536 if (IS_ERR(npage[i])) {
537 alloc_nid_failed(sbi, nids[i]);
538 err = PTR_ERR(npage[i]);
539 goto release_pages;
540 }
541
542 set_nid(parent, offset[i - 1], nids[i], i == 1);
543 alloc_nid_done(sbi, nids[i]);
544 done = true;
545 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
546 npage[i] = get_node_page_ra(parent, offset[i - 1]);
547 if (IS_ERR(npage[i])) {
548 err = PTR_ERR(npage[i]);
549 goto release_pages;
550 }
551 done = true;
552 }
553 if (i == 1) {
554 dn->inode_page_locked = false;
555 unlock_page(parent);
556 } else {
557 f2fs_put_page(parent, 1);
558 }
559
560 if (!done) {
561 npage[i] = get_node_page(sbi, nids[i]);
562 if (IS_ERR(npage[i])) {
563 err = PTR_ERR(npage[i]);
564 f2fs_put_page(npage[0], 0);
565 goto release_out;
566 }
567 }
568 if (i < level) {
569 parent = npage[i];
570 nids[i + 1] = get_nid(parent, offset[i], false);
571 }
572 }
573 dn->nid = nids[level];
574 dn->ofs_in_node = offset[level];
575 dn->node_page = npage[level];
576 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
577 return 0;
578
579 release_pages:
580 f2fs_put_page(parent, 1);
581 if (i > 1)
582 f2fs_put_page(npage[0], 0);
583 release_out:
584 dn->inode_page = NULL;
585 dn->node_page = NULL;
586 return err;
587 }
588
589 static void truncate_node(struct dnode_of_data *dn)
590 {
591 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
592 struct node_info ni;
593
594 get_node_info(sbi, dn->nid, &ni);
595 if (dn->inode->i_blocks == 0) {
596 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
597 goto invalidate;
598 }
599 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
600
601 /* Deallocate node address */
602 invalidate_blocks(sbi, ni.blk_addr);
603 dec_valid_node_count(sbi, dn->inode);
604 set_node_addr(sbi, &ni, NULL_ADDR, false);
605
606 if (dn->nid == dn->inode->i_ino) {
607 remove_orphan_inode(sbi, dn->nid);
608 dec_valid_inode_count(sbi);
609 } else {
610 sync_inode_page(dn);
611 }
612 invalidate:
613 clear_node_page_dirty(dn->node_page);
614 set_sbi_flag(sbi, SBI_IS_DIRTY);
615
616 f2fs_put_page(dn->node_page, 1);
617
618 invalidate_mapping_pages(NODE_MAPPING(sbi),
619 dn->node_page->index, dn->node_page->index);
620
621 dn->node_page = NULL;
622 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
623 }
624
625 static int truncate_dnode(struct dnode_of_data *dn)
626 {
627 struct page *page;
628
629 if (dn->nid == 0)
630 return 1;
631
632 /* get direct node */
633 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
634 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
635 return 1;
636 else if (IS_ERR(page))
637 return PTR_ERR(page);
638
639 /* Make dnode_of_data for parameter */
640 dn->node_page = page;
641 dn->ofs_in_node = 0;
642 truncate_data_blocks(dn);
643 truncate_node(dn);
644 return 1;
645 }
646
647 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
648 int ofs, int depth)
649 {
650 struct dnode_of_data rdn = *dn;
651 struct page *page;
652 struct f2fs_node *rn;
653 nid_t child_nid;
654 unsigned int child_nofs;
655 int freed = 0;
656 int i, ret;
657
658 if (dn->nid == 0)
659 return NIDS_PER_BLOCK + 1;
660
661 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
662
663 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
664 if (IS_ERR(page)) {
665 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
666 return PTR_ERR(page);
667 }
668
669 rn = F2FS_NODE(page);
670 if (depth < 3) {
671 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
672 child_nid = le32_to_cpu(rn->in.nid[i]);
673 if (child_nid == 0)
674 continue;
675 rdn.nid = child_nid;
676 ret = truncate_dnode(&rdn);
677 if (ret < 0)
678 goto out_err;
679 set_nid(page, i, 0, false);
680 }
681 } else {
682 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
683 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
684 child_nid = le32_to_cpu(rn->in.nid[i]);
685 if (child_nid == 0) {
686 child_nofs += NIDS_PER_BLOCK + 1;
687 continue;
688 }
689 rdn.nid = child_nid;
690 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
691 if (ret == (NIDS_PER_BLOCK + 1)) {
692 set_nid(page, i, 0, false);
693 child_nofs += ret;
694 } else if (ret < 0 && ret != -ENOENT) {
695 goto out_err;
696 }
697 }
698 freed = child_nofs;
699 }
700
701 if (!ofs) {
702 /* remove current indirect node */
703 dn->node_page = page;
704 truncate_node(dn);
705 freed++;
706 } else {
707 f2fs_put_page(page, 1);
708 }
709 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
710 return freed;
711
712 out_err:
713 f2fs_put_page(page, 1);
714 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
715 return ret;
716 }
717
718 static int truncate_partial_nodes(struct dnode_of_data *dn,
719 struct f2fs_inode *ri, int *offset, int depth)
720 {
721 struct page *pages[2];
722 nid_t nid[3];
723 nid_t child_nid;
724 int err = 0;
725 int i;
726 int idx = depth - 2;
727
728 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
729 if (!nid[0])
730 return 0;
731
732 /* get indirect nodes in the path */
733 for (i = 0; i < idx + 1; i++) {
734 /* reference count'll be increased */
735 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
736 if (IS_ERR(pages[i])) {
737 err = PTR_ERR(pages[i]);
738 idx = i - 1;
739 goto fail;
740 }
741 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
742 }
743
744 /* free direct nodes linked to a partial indirect node */
745 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
746 child_nid = get_nid(pages[idx], i, false);
747 if (!child_nid)
748 continue;
749 dn->nid = child_nid;
750 err = truncate_dnode(dn);
751 if (err < 0)
752 goto fail;
753 set_nid(pages[idx], i, 0, false);
754 }
755
756 if (offset[idx + 1] == 0) {
757 dn->node_page = pages[idx];
758 dn->nid = nid[idx];
759 truncate_node(dn);
760 } else {
761 f2fs_put_page(pages[idx], 1);
762 }
763 offset[idx]++;
764 offset[idx + 1] = 0;
765 idx--;
766 fail:
767 for (i = idx; i >= 0; i--)
768 f2fs_put_page(pages[i], 1);
769
770 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
771
772 return err;
773 }
774
775 /*
776 * All the block addresses of data and nodes should be nullified.
777 */
778 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
779 {
780 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
781 int err = 0, cont = 1;
782 int level, offset[4], noffset[4];
783 unsigned int nofs = 0;
784 struct f2fs_inode *ri;
785 struct dnode_of_data dn;
786 struct page *page;
787
788 trace_f2fs_truncate_inode_blocks_enter(inode, from);
789
790 level = get_node_path(F2FS_I(inode), from, offset, noffset);
791 restart:
792 page = get_node_page(sbi, inode->i_ino);
793 if (IS_ERR(page)) {
794 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
795 return PTR_ERR(page);
796 }
797
798 set_new_dnode(&dn, inode, page, NULL, 0);
799 unlock_page(page);
800
801 ri = F2FS_INODE(page);
802 switch (level) {
803 case 0:
804 case 1:
805 nofs = noffset[1];
806 break;
807 case 2:
808 nofs = noffset[1];
809 if (!offset[level - 1])
810 goto skip_partial;
811 err = truncate_partial_nodes(&dn, ri, offset, level);
812 if (err < 0 && err != -ENOENT)
813 goto fail;
814 nofs += 1 + NIDS_PER_BLOCK;
815 break;
816 case 3:
817 nofs = 5 + 2 * NIDS_PER_BLOCK;
818 if (!offset[level - 1])
819 goto skip_partial;
820 err = truncate_partial_nodes(&dn, ri, offset, level);
821 if (err < 0 && err != -ENOENT)
822 goto fail;
823 break;
824 default:
825 BUG();
826 }
827
828 skip_partial:
829 while (cont) {
830 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
831 switch (offset[0]) {
832 case NODE_DIR1_BLOCK:
833 case NODE_DIR2_BLOCK:
834 err = truncate_dnode(&dn);
835 break;
836
837 case NODE_IND1_BLOCK:
838 case NODE_IND2_BLOCK:
839 err = truncate_nodes(&dn, nofs, offset[1], 2);
840 break;
841
842 case NODE_DIND_BLOCK:
843 err = truncate_nodes(&dn, nofs, offset[1], 3);
844 cont = 0;
845 break;
846
847 default:
848 BUG();
849 }
850 if (err < 0 && err != -ENOENT)
851 goto fail;
852 if (offset[1] == 0 &&
853 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
854 lock_page(page);
855 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
856 f2fs_put_page(page, 1);
857 goto restart;
858 }
859 f2fs_wait_on_page_writeback(page, NODE);
860 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
861 set_page_dirty(page);
862 unlock_page(page);
863 }
864 offset[1] = 0;
865 offset[0]++;
866 nofs += err;
867 }
868 fail:
869 f2fs_put_page(page, 0);
870 trace_f2fs_truncate_inode_blocks_exit(inode, err);
871 return err > 0 ? 0 : err;
872 }
873
874 int truncate_xattr_node(struct inode *inode, struct page *page)
875 {
876 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
877 nid_t nid = F2FS_I(inode)->i_xattr_nid;
878 struct dnode_of_data dn;
879 struct page *npage;
880
881 if (!nid)
882 return 0;
883
884 npage = get_node_page(sbi, nid);
885 if (IS_ERR(npage))
886 return PTR_ERR(npage);
887
888 F2FS_I(inode)->i_xattr_nid = 0;
889
890 /* need to do checkpoint during fsync */
891 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
892
893 set_new_dnode(&dn, inode, page, npage, nid);
894
895 if (page)
896 dn.inode_page_locked = true;
897 truncate_node(&dn);
898 return 0;
899 }
900
901 /*
902 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
903 * f2fs_unlock_op().
904 */
905 void remove_inode_page(struct inode *inode)
906 {
907 struct dnode_of_data dn;
908
909 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
910 if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
911 return;
912
913 if (truncate_xattr_node(inode, dn.inode_page)) {
914 f2fs_put_dnode(&dn);
915 return;
916 }
917
918 /* remove potential inline_data blocks */
919 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
920 S_ISLNK(inode->i_mode))
921 truncate_data_blocks_range(&dn, 1);
922
923 /* 0 is possible, after f2fs_new_inode() has failed */
924 f2fs_bug_on(F2FS_I_SB(inode),
925 inode->i_blocks != 0 && inode->i_blocks != 1);
926
927 /* will put inode & node pages */
928 truncate_node(&dn);
929 }
930
931 struct page *new_inode_page(struct inode *inode)
932 {
933 struct dnode_of_data dn;
934
935 /* allocate inode page for new inode */
936 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
937
938 /* caller should f2fs_put_page(page, 1); */
939 return new_node_page(&dn, 0, NULL);
940 }
941
942 struct page *new_node_page(struct dnode_of_data *dn,
943 unsigned int ofs, struct page *ipage)
944 {
945 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
946 struct node_info old_ni, new_ni;
947 struct page *page;
948 int err;
949
950 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
951 return ERR_PTR(-EPERM);
952
953 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
954 if (!page)
955 return ERR_PTR(-ENOMEM);
956
957 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
958 err = -ENOSPC;
959 goto fail;
960 }
961
962 get_node_info(sbi, dn->nid, &old_ni);
963
964 /* Reinitialize old_ni with new node page */
965 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
966 new_ni = old_ni;
967 new_ni.ino = dn->inode->i_ino;
968 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
969
970 f2fs_wait_on_page_writeback(page, NODE);
971 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
972 set_cold_node(dn->inode, page);
973 SetPageUptodate(page);
974 set_page_dirty(page);
975
976 if (f2fs_has_xattr_block(ofs))
977 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
978
979 dn->node_page = page;
980 if (ipage)
981 update_inode(dn->inode, ipage);
982 else
983 sync_inode_page(dn);
984 if (ofs == 0)
985 inc_valid_inode_count(sbi);
986
987 return page;
988
989 fail:
990 clear_node_page_dirty(page);
991 f2fs_put_page(page, 1);
992 return ERR_PTR(err);
993 }
994
995 /*
996 * Caller should do after getting the following values.
997 * 0: f2fs_put_page(page, 0)
998 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
999 */
1000 static int read_node_page(struct page *page, int rw)
1001 {
1002 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1003 struct node_info ni;
1004 struct f2fs_io_info fio = {
1005 .sbi = sbi,
1006 .type = NODE,
1007 .rw = rw,
1008 .page = page,
1009 .encrypted_page = NULL,
1010 };
1011
1012 get_node_info(sbi, page->index, &ni);
1013
1014 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1015 ClearPageUptodate(page);
1016 return -ENOENT;
1017 }
1018
1019 if (PageUptodate(page))
1020 return LOCKED_PAGE;
1021
1022 fio.blk_addr = ni.blk_addr;
1023 return f2fs_submit_page_bio(&fio);
1024 }
1025
1026 /*
1027 * Readahead a node page
1028 */
1029 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1030 {
1031 struct page *apage;
1032 int err;
1033
1034 apage = find_get_page(NODE_MAPPING(sbi), nid);
1035 if (apage && PageUptodate(apage)) {
1036 f2fs_put_page(apage, 0);
1037 return;
1038 }
1039 f2fs_put_page(apage, 0);
1040
1041 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1042 if (!apage)
1043 return;
1044
1045 err = read_node_page(apage, READA);
1046 f2fs_put_page(apage, err ? 1 : 0);
1047 }
1048
1049 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1050 {
1051 struct page *page;
1052 int err;
1053 repeat:
1054 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1055 if (!page)
1056 return ERR_PTR(-ENOMEM);
1057
1058 err = read_node_page(page, READ_SYNC);
1059 if (err < 0) {
1060 f2fs_put_page(page, 1);
1061 return ERR_PTR(err);
1062 } else if (err != LOCKED_PAGE) {
1063 lock_page(page);
1064 }
1065
1066 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1067 ClearPageUptodate(page);
1068 f2fs_put_page(page, 1);
1069 return ERR_PTR(-EIO);
1070 }
1071 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1072 f2fs_put_page(page, 1);
1073 goto repeat;
1074 }
1075 return page;
1076 }
1077
1078 /*
1079 * Return a locked page for the desired node page.
1080 * And, readahead MAX_RA_NODE number of node pages.
1081 */
1082 struct page *get_node_page_ra(struct page *parent, int start)
1083 {
1084 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1085 struct blk_plug plug;
1086 struct page *page;
1087 int err, i, end;
1088 nid_t nid;
1089
1090 /* First, try getting the desired direct node. */
1091 nid = get_nid(parent, start, false);
1092 if (!nid)
1093 return ERR_PTR(-ENOENT);
1094 repeat:
1095 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1096 if (!page)
1097 return ERR_PTR(-ENOMEM);
1098
1099 err = read_node_page(page, READ_SYNC);
1100 if (err < 0) {
1101 f2fs_put_page(page, 1);
1102 return ERR_PTR(err);
1103 } else if (err == LOCKED_PAGE) {
1104 goto page_hit;
1105 }
1106
1107 blk_start_plug(&plug);
1108
1109 /* Then, try readahead for siblings of the desired node */
1110 end = start + MAX_RA_NODE;
1111 end = min(end, NIDS_PER_BLOCK);
1112 for (i = start + 1; i < end; i++) {
1113 nid = get_nid(parent, i, false);
1114 if (!nid)
1115 continue;
1116 ra_node_page(sbi, nid);
1117 }
1118
1119 blk_finish_plug(&plug);
1120
1121 lock_page(page);
1122 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1123 f2fs_put_page(page, 1);
1124 goto repeat;
1125 }
1126 page_hit:
1127 if (unlikely(!PageUptodate(page))) {
1128 f2fs_put_page(page, 1);
1129 return ERR_PTR(-EIO);
1130 }
1131 return page;
1132 }
1133
1134 void sync_inode_page(struct dnode_of_data *dn)
1135 {
1136 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1137 update_inode(dn->inode, dn->node_page);
1138 } else if (dn->inode_page) {
1139 if (!dn->inode_page_locked)
1140 lock_page(dn->inode_page);
1141 update_inode(dn->inode, dn->inode_page);
1142 if (!dn->inode_page_locked)
1143 unlock_page(dn->inode_page);
1144 } else {
1145 update_inode_page(dn->inode);
1146 }
1147 }
1148
1149 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1150 struct writeback_control *wbc)
1151 {
1152 pgoff_t index, end;
1153 struct pagevec pvec;
1154 int step = ino ? 2 : 0;
1155 int nwritten = 0, wrote = 0;
1156
1157 pagevec_init(&pvec, 0);
1158
1159 next_step:
1160 index = 0;
1161 end = LONG_MAX;
1162
1163 while (index <= end) {
1164 int i, nr_pages;
1165 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1166 PAGECACHE_TAG_DIRTY,
1167 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1168 if (nr_pages == 0)
1169 break;
1170
1171 for (i = 0; i < nr_pages; i++) {
1172 struct page *page = pvec.pages[i];
1173
1174 /*
1175 * flushing sequence with step:
1176 * 0. indirect nodes
1177 * 1. dentry dnodes
1178 * 2. file dnodes
1179 */
1180 if (step == 0 && IS_DNODE(page))
1181 continue;
1182 if (step == 1 && (!IS_DNODE(page) ||
1183 is_cold_node(page)))
1184 continue;
1185 if (step == 2 && (!IS_DNODE(page) ||
1186 !is_cold_node(page)))
1187 continue;
1188
1189 /*
1190 * If an fsync mode,
1191 * we should not skip writing node pages.
1192 */
1193 if (ino && ino_of_node(page) == ino)
1194 lock_page(page);
1195 else if (!trylock_page(page))
1196 continue;
1197
1198 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1199 continue_unlock:
1200 unlock_page(page);
1201 continue;
1202 }
1203 if (ino && ino_of_node(page) != ino)
1204 goto continue_unlock;
1205
1206 if (!PageDirty(page)) {
1207 /* someone wrote it for us */
1208 goto continue_unlock;
1209 }
1210
1211 if (!clear_page_dirty_for_io(page))
1212 goto continue_unlock;
1213
1214 /* called by fsync() */
1215 if (ino && IS_DNODE(page)) {
1216 set_fsync_mark(page, 1);
1217 if (IS_INODE(page))
1218 set_dentry_mark(page,
1219 need_dentry_mark(sbi, ino));
1220 nwritten++;
1221 } else {
1222 set_fsync_mark(page, 0);
1223 set_dentry_mark(page, 0);
1224 }
1225
1226 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1227 unlock_page(page);
1228 else
1229 wrote++;
1230
1231 if (--wbc->nr_to_write == 0)
1232 break;
1233 }
1234 pagevec_release(&pvec);
1235 cond_resched();
1236
1237 if (wbc->nr_to_write == 0) {
1238 step = 2;
1239 break;
1240 }
1241 }
1242
1243 if (step < 2) {
1244 step++;
1245 goto next_step;
1246 }
1247
1248 if (wrote)
1249 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1250 return nwritten;
1251 }
1252
1253 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1254 {
1255 pgoff_t index = 0, end = LONG_MAX;
1256 struct pagevec pvec;
1257 int ret2 = 0, ret = 0;
1258
1259 pagevec_init(&pvec, 0);
1260
1261 while (index <= end) {
1262 int i, nr_pages;
1263 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1264 PAGECACHE_TAG_WRITEBACK,
1265 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1266 if (nr_pages == 0)
1267 break;
1268
1269 for (i = 0; i < nr_pages; i++) {
1270 struct page *page = pvec.pages[i];
1271
1272 /* until radix tree lookup accepts end_index */
1273 if (unlikely(page->index > end))
1274 continue;
1275
1276 if (ino && ino_of_node(page) == ino) {
1277 f2fs_wait_on_page_writeback(page, NODE);
1278 if (TestClearPageError(page))
1279 ret = -EIO;
1280 }
1281 }
1282 pagevec_release(&pvec);
1283 cond_resched();
1284 }
1285
1286 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1287 ret2 = -ENOSPC;
1288 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1289 ret2 = -EIO;
1290 if (!ret)
1291 ret = ret2;
1292 return ret;
1293 }
1294
1295 static int f2fs_write_node_page(struct page *page,
1296 struct writeback_control *wbc)
1297 {
1298 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1299 nid_t nid;
1300 struct node_info ni;
1301 struct f2fs_io_info fio = {
1302 .sbi = sbi,
1303 .type = NODE,
1304 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1305 .page = page,
1306 .encrypted_page = NULL,
1307 };
1308
1309 trace_f2fs_writepage(page, NODE);
1310
1311 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1312 goto redirty_out;
1313 if (unlikely(f2fs_cp_error(sbi)))
1314 goto redirty_out;
1315
1316 f2fs_wait_on_page_writeback(page, NODE);
1317
1318 /* get old block addr of this node page */
1319 nid = nid_of_node(page);
1320 f2fs_bug_on(sbi, page->index != nid);
1321
1322 get_node_info(sbi, nid, &ni);
1323
1324 /* This page is already truncated */
1325 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1326 ClearPageUptodate(page);
1327 dec_page_count(sbi, F2FS_DIRTY_NODES);
1328 unlock_page(page);
1329 return 0;
1330 }
1331
1332 if (wbc->for_reclaim) {
1333 if (!down_read_trylock(&sbi->node_write))
1334 goto redirty_out;
1335 } else {
1336 down_read(&sbi->node_write);
1337 }
1338
1339 set_page_writeback(page);
1340 fio.blk_addr = ni.blk_addr;
1341 write_node_page(nid, &fio);
1342 set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1343 dec_page_count(sbi, F2FS_DIRTY_NODES);
1344 up_read(&sbi->node_write);
1345 unlock_page(page);
1346
1347 if (wbc->for_reclaim)
1348 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1349
1350 return 0;
1351
1352 redirty_out:
1353 redirty_page_for_writepage(wbc, page);
1354 return AOP_WRITEPAGE_ACTIVATE;
1355 }
1356
1357 static int f2fs_write_node_pages(struct address_space *mapping,
1358 struct writeback_control *wbc)
1359 {
1360 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1361 long diff;
1362
1363 trace_f2fs_writepages(mapping->host, wbc, NODE);
1364
1365 /* balancing f2fs's metadata in background */
1366 f2fs_balance_fs_bg(sbi);
1367
1368 /* collect a number of dirty node pages and write together */
1369 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1370 goto skip_write;
1371
1372 diff = nr_pages_to_write(sbi, NODE, wbc);
1373 wbc->sync_mode = WB_SYNC_NONE;
1374 sync_node_pages(sbi, 0, wbc);
1375 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1376 return 0;
1377
1378 skip_write:
1379 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1380 return 0;
1381 }
1382
1383 static int f2fs_set_node_page_dirty(struct page *page)
1384 {
1385 trace_f2fs_set_page_dirty(page, NODE);
1386
1387 SetPageUptodate(page);
1388 if (!PageDirty(page)) {
1389 __set_page_dirty_nobuffers(page);
1390 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1391 SetPagePrivate(page);
1392 f2fs_trace_pid(page);
1393 return 1;
1394 }
1395 return 0;
1396 }
1397
1398 /*
1399 * Structure of the f2fs node operations
1400 */
1401 const struct address_space_operations f2fs_node_aops = {
1402 .writepage = f2fs_write_node_page,
1403 .writepages = f2fs_write_node_pages,
1404 .set_page_dirty = f2fs_set_node_page_dirty,
1405 .invalidatepage = f2fs_invalidate_page,
1406 .releasepage = f2fs_release_page,
1407 };
1408
1409 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1410 nid_t n)
1411 {
1412 return radix_tree_lookup(&nm_i->free_nid_root, n);
1413 }
1414
1415 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1416 struct free_nid *i)
1417 {
1418 list_del(&i->list);
1419 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1420 }
1421
1422 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1423 {
1424 struct f2fs_nm_info *nm_i = NM_I(sbi);
1425 struct free_nid *i;
1426 struct nat_entry *ne;
1427 bool allocated = false;
1428
1429 if (!available_free_memory(sbi, FREE_NIDS))
1430 return -1;
1431
1432 /* 0 nid should not be used */
1433 if (unlikely(nid == 0))
1434 return 0;
1435
1436 if (build) {
1437 /* do not add allocated nids */
1438 down_read(&nm_i->nat_tree_lock);
1439 ne = __lookup_nat_cache(nm_i, nid);
1440 if (ne &&
1441 (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1442 nat_get_blkaddr(ne) != NULL_ADDR))
1443 allocated = true;
1444 up_read(&nm_i->nat_tree_lock);
1445 if (allocated)
1446 return 0;
1447 }
1448
1449 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1450 i->nid = nid;
1451 i->state = NID_NEW;
1452
1453 if (radix_tree_preload(GFP_NOFS)) {
1454 kmem_cache_free(free_nid_slab, i);
1455 return 0;
1456 }
1457
1458 spin_lock(&nm_i->free_nid_list_lock);
1459 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1460 spin_unlock(&nm_i->free_nid_list_lock);
1461 radix_tree_preload_end();
1462 kmem_cache_free(free_nid_slab, i);
1463 return 0;
1464 }
1465 list_add_tail(&i->list, &nm_i->free_nid_list);
1466 nm_i->fcnt++;
1467 spin_unlock(&nm_i->free_nid_list_lock);
1468 radix_tree_preload_end();
1469 return 1;
1470 }
1471
1472 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1473 {
1474 struct free_nid *i;
1475 bool need_free = false;
1476
1477 spin_lock(&nm_i->free_nid_list_lock);
1478 i = __lookup_free_nid_list(nm_i, nid);
1479 if (i && i->state == NID_NEW) {
1480 __del_from_free_nid_list(nm_i, i);
1481 nm_i->fcnt--;
1482 need_free = true;
1483 }
1484 spin_unlock(&nm_i->free_nid_list_lock);
1485
1486 if (need_free)
1487 kmem_cache_free(free_nid_slab, i);
1488 }
1489
1490 static void scan_nat_page(struct f2fs_sb_info *sbi,
1491 struct page *nat_page, nid_t start_nid)
1492 {
1493 struct f2fs_nm_info *nm_i = NM_I(sbi);
1494 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1495 block_t blk_addr;
1496 int i;
1497
1498 i = start_nid % NAT_ENTRY_PER_BLOCK;
1499
1500 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1501
1502 if (unlikely(start_nid >= nm_i->max_nid))
1503 break;
1504
1505 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1506 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1507 if (blk_addr == NULL_ADDR) {
1508 if (add_free_nid(sbi, start_nid, true) < 0)
1509 break;
1510 }
1511 }
1512 }
1513
1514 static void build_free_nids(struct f2fs_sb_info *sbi)
1515 {
1516 struct f2fs_nm_info *nm_i = NM_I(sbi);
1517 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1518 struct f2fs_summary_block *sum = curseg->sum_blk;
1519 int i = 0;
1520 nid_t nid = nm_i->next_scan_nid;
1521
1522 /* Enough entries */
1523 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1524 return;
1525
1526 /* readahead nat pages to be scanned */
1527 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1528
1529 while (1) {
1530 struct page *page = get_current_nat_page(sbi, nid);
1531
1532 scan_nat_page(sbi, page, nid);
1533 f2fs_put_page(page, 1);
1534
1535 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1536 if (unlikely(nid >= nm_i->max_nid))
1537 nid = 0;
1538
1539 if (++i >= FREE_NID_PAGES)
1540 break;
1541 }
1542
1543 /* go to the next free nat pages to find free nids abundantly */
1544 nm_i->next_scan_nid = nid;
1545
1546 /* find free nids from current sum_pages */
1547 mutex_lock(&curseg->curseg_mutex);
1548 for (i = 0; i < nats_in_cursum(sum); i++) {
1549 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1550 nid = le32_to_cpu(nid_in_journal(sum, i));
1551 if (addr == NULL_ADDR)
1552 add_free_nid(sbi, nid, true);
1553 else
1554 remove_free_nid(nm_i, nid);
1555 }
1556 mutex_unlock(&curseg->curseg_mutex);
1557 }
1558
1559 /*
1560 * If this function returns success, caller can obtain a new nid
1561 * from second parameter of this function.
1562 * The returned nid could be used ino as well as nid when inode is created.
1563 */
1564 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1565 {
1566 struct f2fs_nm_info *nm_i = NM_I(sbi);
1567 struct free_nid *i = NULL;
1568 retry:
1569 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1570 return false;
1571
1572 spin_lock(&nm_i->free_nid_list_lock);
1573
1574 /* We should not use stale free nids created by build_free_nids */
1575 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1576 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1577 list_for_each_entry(i, &nm_i->free_nid_list, list)
1578 if (i->state == NID_NEW)
1579 break;
1580
1581 f2fs_bug_on(sbi, i->state != NID_NEW);
1582 *nid = i->nid;
1583 i->state = NID_ALLOC;
1584 nm_i->fcnt--;
1585 spin_unlock(&nm_i->free_nid_list_lock);
1586 return true;
1587 }
1588 spin_unlock(&nm_i->free_nid_list_lock);
1589
1590 /* Let's scan nat pages and its caches to get free nids */
1591 mutex_lock(&nm_i->build_lock);
1592 build_free_nids(sbi);
1593 mutex_unlock(&nm_i->build_lock);
1594 goto retry;
1595 }
1596
1597 /*
1598 * alloc_nid() should be called prior to this function.
1599 */
1600 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1601 {
1602 struct f2fs_nm_info *nm_i = NM_I(sbi);
1603 struct free_nid *i;
1604
1605 spin_lock(&nm_i->free_nid_list_lock);
1606 i = __lookup_free_nid_list(nm_i, nid);
1607 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1608 __del_from_free_nid_list(nm_i, i);
1609 spin_unlock(&nm_i->free_nid_list_lock);
1610
1611 kmem_cache_free(free_nid_slab, i);
1612 }
1613
1614 /*
1615 * alloc_nid() should be called prior to this function.
1616 */
1617 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1618 {
1619 struct f2fs_nm_info *nm_i = NM_I(sbi);
1620 struct free_nid *i;
1621 bool need_free = false;
1622
1623 if (!nid)
1624 return;
1625
1626 spin_lock(&nm_i->free_nid_list_lock);
1627 i = __lookup_free_nid_list(nm_i, nid);
1628 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1629 if (!available_free_memory(sbi, FREE_NIDS)) {
1630 __del_from_free_nid_list(nm_i, i);
1631 need_free = true;
1632 } else {
1633 i->state = NID_NEW;
1634 nm_i->fcnt++;
1635 }
1636 spin_unlock(&nm_i->free_nid_list_lock);
1637
1638 if (need_free)
1639 kmem_cache_free(free_nid_slab, i);
1640 }
1641
1642 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1643 {
1644 struct f2fs_nm_info *nm_i = NM_I(sbi);
1645 struct free_nid *i, *next;
1646 int nr = nr_shrink;
1647
1648 if (!mutex_trylock(&nm_i->build_lock))
1649 return 0;
1650
1651 spin_lock(&nm_i->free_nid_list_lock);
1652 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1653 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1654 break;
1655 if (i->state == NID_ALLOC)
1656 continue;
1657 __del_from_free_nid_list(nm_i, i);
1658 nm_i->fcnt--;
1659 spin_unlock(&nm_i->free_nid_list_lock);
1660 kmem_cache_free(free_nid_slab, i);
1661 nr_shrink--;
1662 spin_lock(&nm_i->free_nid_list_lock);
1663 }
1664 spin_unlock(&nm_i->free_nid_list_lock);
1665 mutex_unlock(&nm_i->build_lock);
1666
1667 return nr - nr_shrink;
1668 }
1669
1670 void recover_inline_xattr(struct inode *inode, struct page *page)
1671 {
1672 void *src_addr, *dst_addr;
1673 size_t inline_size;
1674 struct page *ipage;
1675 struct f2fs_inode *ri;
1676
1677 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1678 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1679
1680 ri = F2FS_INODE(page);
1681 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1682 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1683 goto update_inode;
1684 }
1685
1686 dst_addr = inline_xattr_addr(ipage);
1687 src_addr = inline_xattr_addr(page);
1688 inline_size = inline_xattr_size(inode);
1689
1690 f2fs_wait_on_page_writeback(ipage, NODE);
1691 memcpy(dst_addr, src_addr, inline_size);
1692 update_inode:
1693 update_inode(inode, ipage);
1694 f2fs_put_page(ipage, 1);
1695 }
1696
1697 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1698 {
1699 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1700 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1701 nid_t new_xnid = nid_of_node(page);
1702 struct node_info ni;
1703
1704 /* 1: invalidate the previous xattr nid */
1705 if (!prev_xnid)
1706 goto recover_xnid;
1707
1708 /* Deallocate node address */
1709 get_node_info(sbi, prev_xnid, &ni);
1710 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1711 invalidate_blocks(sbi, ni.blk_addr);
1712 dec_valid_node_count(sbi, inode);
1713 set_node_addr(sbi, &ni, NULL_ADDR, false);
1714
1715 recover_xnid:
1716 /* 2: allocate new xattr nid */
1717 if (unlikely(!inc_valid_node_count(sbi, inode)))
1718 f2fs_bug_on(sbi, 1);
1719
1720 remove_free_nid(NM_I(sbi), new_xnid);
1721 get_node_info(sbi, new_xnid, &ni);
1722 ni.ino = inode->i_ino;
1723 set_node_addr(sbi, &ni, NEW_ADDR, false);
1724 F2FS_I(inode)->i_xattr_nid = new_xnid;
1725
1726 /* 3: update xattr blkaddr */
1727 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1728 set_node_addr(sbi, &ni, blkaddr, false);
1729
1730 update_inode_page(inode);
1731 }
1732
1733 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1734 {
1735 struct f2fs_inode *src, *dst;
1736 nid_t ino = ino_of_node(page);
1737 struct node_info old_ni, new_ni;
1738 struct page *ipage;
1739
1740 get_node_info(sbi, ino, &old_ni);
1741
1742 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1743 return -EINVAL;
1744
1745 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1746 if (!ipage)
1747 return -ENOMEM;
1748
1749 /* Should not use this inode from free nid list */
1750 remove_free_nid(NM_I(sbi), ino);
1751
1752 SetPageUptodate(ipage);
1753 fill_node_footer(ipage, ino, ino, 0, true);
1754
1755 src = F2FS_INODE(page);
1756 dst = F2FS_INODE(ipage);
1757
1758 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1759 dst->i_size = 0;
1760 dst->i_blocks = cpu_to_le64(1);
1761 dst->i_links = cpu_to_le32(1);
1762 dst->i_xattr_nid = 0;
1763 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1764
1765 new_ni = old_ni;
1766 new_ni.ino = ino;
1767
1768 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1769 WARN_ON(1);
1770 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1771 inc_valid_inode_count(sbi);
1772 set_page_dirty(ipage);
1773 f2fs_put_page(ipage, 1);
1774 return 0;
1775 }
1776
1777 int restore_node_summary(struct f2fs_sb_info *sbi,
1778 unsigned int segno, struct f2fs_summary_block *sum)
1779 {
1780 struct f2fs_node *rn;
1781 struct f2fs_summary *sum_entry;
1782 block_t addr;
1783 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1784 int i, idx, last_offset, nrpages;
1785
1786 /* scan the node segment */
1787 last_offset = sbi->blocks_per_seg;
1788 addr = START_BLOCK(sbi, segno);
1789 sum_entry = &sum->entries[0];
1790
1791 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1792 nrpages = min(last_offset - i, bio_blocks);
1793
1794 /* readahead node pages */
1795 ra_meta_pages(sbi, addr, nrpages, META_POR);
1796
1797 for (idx = addr; idx < addr + nrpages; idx++) {
1798 struct page *page = get_meta_page(sbi, idx);
1799
1800 rn = F2FS_NODE(page);
1801 sum_entry->nid = rn->footer.nid;
1802 sum_entry->version = 0;
1803 sum_entry->ofs_in_node = 0;
1804 sum_entry++;
1805 f2fs_put_page(page, 1);
1806 }
1807
1808 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1809 addr + nrpages);
1810 }
1811 return 0;
1812 }
1813
1814 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1815 {
1816 struct f2fs_nm_info *nm_i = NM_I(sbi);
1817 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1818 struct f2fs_summary_block *sum = curseg->sum_blk;
1819 int i;
1820
1821 mutex_lock(&curseg->curseg_mutex);
1822 for (i = 0; i < nats_in_cursum(sum); i++) {
1823 struct nat_entry *ne;
1824 struct f2fs_nat_entry raw_ne;
1825 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1826
1827 raw_ne = nat_in_journal(sum, i);
1828
1829 down_write(&nm_i->nat_tree_lock);
1830 ne = __lookup_nat_cache(nm_i, nid);
1831 if (!ne) {
1832 ne = grab_nat_entry(nm_i, nid);
1833 node_info_from_raw_nat(&ne->ni, &raw_ne);
1834 }
1835 __set_nat_cache_dirty(nm_i, ne);
1836 up_write(&nm_i->nat_tree_lock);
1837 }
1838 update_nats_in_cursum(sum, -i);
1839 mutex_unlock(&curseg->curseg_mutex);
1840 }
1841
1842 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1843 struct list_head *head, int max)
1844 {
1845 struct nat_entry_set *cur;
1846
1847 if (nes->entry_cnt >= max)
1848 goto add_out;
1849
1850 list_for_each_entry(cur, head, set_list) {
1851 if (cur->entry_cnt >= nes->entry_cnt) {
1852 list_add(&nes->set_list, cur->set_list.prev);
1853 return;
1854 }
1855 }
1856 add_out:
1857 list_add_tail(&nes->set_list, head);
1858 }
1859
1860 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1861 struct nat_entry_set *set)
1862 {
1863 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1864 struct f2fs_summary_block *sum = curseg->sum_blk;
1865 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1866 bool to_journal = true;
1867 struct f2fs_nat_block *nat_blk;
1868 struct nat_entry *ne, *cur;
1869 struct page *page = NULL;
1870 struct f2fs_nm_info *nm_i = NM_I(sbi);
1871
1872 /*
1873 * there are two steps to flush nat entries:
1874 * #1, flush nat entries to journal in current hot data summary block.
1875 * #2, flush nat entries to nat page.
1876 */
1877 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1878 to_journal = false;
1879
1880 if (to_journal) {
1881 mutex_lock(&curseg->curseg_mutex);
1882 } else {
1883 page = get_next_nat_page(sbi, start_nid);
1884 nat_blk = page_address(page);
1885 f2fs_bug_on(sbi, !nat_blk);
1886 }
1887
1888 /* flush dirty nats in nat entry set */
1889 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1890 struct f2fs_nat_entry *raw_ne;
1891 nid_t nid = nat_get_nid(ne);
1892 int offset;
1893
1894 if (nat_get_blkaddr(ne) == NEW_ADDR)
1895 continue;
1896
1897 if (to_journal) {
1898 offset = lookup_journal_in_cursum(sum,
1899 NAT_JOURNAL, nid, 1);
1900 f2fs_bug_on(sbi, offset < 0);
1901 raw_ne = &nat_in_journal(sum, offset);
1902 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1903 } else {
1904 raw_ne = &nat_blk->entries[nid - start_nid];
1905 }
1906 raw_nat_from_node_info(raw_ne, &ne->ni);
1907
1908 down_write(&NM_I(sbi)->nat_tree_lock);
1909 nat_reset_flag(ne);
1910 __clear_nat_cache_dirty(NM_I(sbi), ne);
1911 up_write(&NM_I(sbi)->nat_tree_lock);
1912
1913 if (nat_get_blkaddr(ne) == NULL_ADDR)
1914 add_free_nid(sbi, nid, false);
1915 }
1916
1917 if (to_journal)
1918 mutex_unlock(&curseg->curseg_mutex);
1919 else
1920 f2fs_put_page(page, 1);
1921
1922 f2fs_bug_on(sbi, set->entry_cnt);
1923
1924 down_write(&nm_i->nat_tree_lock);
1925 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1926 up_write(&nm_i->nat_tree_lock);
1927 kmem_cache_free(nat_entry_set_slab, set);
1928 }
1929
1930 /*
1931 * This function is called during the checkpointing process.
1932 */
1933 void flush_nat_entries(struct f2fs_sb_info *sbi)
1934 {
1935 struct f2fs_nm_info *nm_i = NM_I(sbi);
1936 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1937 struct f2fs_summary_block *sum = curseg->sum_blk;
1938 struct nat_entry_set *setvec[SETVEC_SIZE];
1939 struct nat_entry_set *set, *tmp;
1940 unsigned int found;
1941 nid_t set_idx = 0;
1942 LIST_HEAD(sets);
1943
1944 if (!nm_i->dirty_nat_cnt)
1945 return;
1946 /*
1947 * if there are no enough space in journal to store dirty nat
1948 * entries, remove all entries from journal and merge them
1949 * into nat entry set.
1950 */
1951 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1952 remove_nats_in_journal(sbi);
1953
1954 down_write(&nm_i->nat_tree_lock);
1955 while ((found = __gang_lookup_nat_set(nm_i,
1956 set_idx, SETVEC_SIZE, setvec))) {
1957 unsigned idx;
1958 set_idx = setvec[found - 1]->set + 1;
1959 for (idx = 0; idx < found; idx++)
1960 __adjust_nat_entry_set(setvec[idx], &sets,
1961 MAX_NAT_JENTRIES(sum));
1962 }
1963 up_write(&nm_i->nat_tree_lock);
1964
1965 /* flush dirty nats in nat entry set */
1966 list_for_each_entry_safe(set, tmp, &sets, set_list)
1967 __flush_nat_entry_set(sbi, set);
1968
1969 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1970 }
1971
1972 static int init_node_manager(struct f2fs_sb_info *sbi)
1973 {
1974 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1975 struct f2fs_nm_info *nm_i = NM_I(sbi);
1976 unsigned char *version_bitmap;
1977 unsigned int nat_segs, nat_blocks;
1978
1979 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1980
1981 /* segment_count_nat includes pair segment so divide to 2. */
1982 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1983 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1984
1985 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1986
1987 /* not used nids: 0, node, meta, (and root counted as valid node) */
1988 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1989 nm_i->fcnt = 0;
1990 nm_i->nat_cnt = 0;
1991 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1992
1993 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1994 INIT_LIST_HEAD(&nm_i->free_nid_list);
1995 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1996 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1997 INIT_LIST_HEAD(&nm_i->nat_entries);
1998
1999 mutex_init(&nm_i->build_lock);
2000 spin_lock_init(&nm_i->free_nid_list_lock);
2001 init_rwsem(&nm_i->nat_tree_lock);
2002
2003 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2004 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2005 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2006 if (!version_bitmap)
2007 return -EFAULT;
2008
2009 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2010 GFP_KERNEL);
2011 if (!nm_i->nat_bitmap)
2012 return -ENOMEM;
2013 return 0;
2014 }
2015
2016 int build_node_manager(struct f2fs_sb_info *sbi)
2017 {
2018 int err;
2019
2020 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2021 if (!sbi->nm_info)
2022 return -ENOMEM;
2023
2024 err = init_node_manager(sbi);
2025 if (err)
2026 return err;
2027
2028 build_free_nids(sbi);
2029 return 0;
2030 }
2031
2032 void destroy_node_manager(struct f2fs_sb_info *sbi)
2033 {
2034 struct f2fs_nm_info *nm_i = NM_I(sbi);
2035 struct free_nid *i, *next_i;
2036 struct nat_entry *natvec[NATVEC_SIZE];
2037 struct nat_entry_set *setvec[SETVEC_SIZE];
2038 nid_t nid = 0;
2039 unsigned int found;
2040
2041 if (!nm_i)
2042 return;
2043
2044 /* destroy free nid list */
2045 spin_lock(&nm_i->free_nid_list_lock);
2046 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2047 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2048 __del_from_free_nid_list(nm_i, i);
2049 nm_i->fcnt--;
2050 spin_unlock(&nm_i->free_nid_list_lock);
2051 kmem_cache_free(free_nid_slab, i);
2052 spin_lock(&nm_i->free_nid_list_lock);
2053 }
2054 f2fs_bug_on(sbi, nm_i->fcnt);
2055 spin_unlock(&nm_i->free_nid_list_lock);
2056
2057 /* destroy nat cache */
2058 down_write(&nm_i->nat_tree_lock);
2059 while ((found = __gang_lookup_nat_cache(nm_i,
2060 nid, NATVEC_SIZE, natvec))) {
2061 unsigned idx;
2062
2063 nid = nat_get_nid(natvec[found - 1]) + 1;
2064 for (idx = 0; idx < found; idx++)
2065 __del_from_nat_cache(nm_i, natvec[idx]);
2066 }
2067 f2fs_bug_on(sbi, nm_i->nat_cnt);
2068
2069 /* destroy nat set cache */
2070 nid = 0;
2071 while ((found = __gang_lookup_nat_set(nm_i,
2072 nid, SETVEC_SIZE, setvec))) {
2073 unsigned idx;
2074
2075 nid = setvec[found - 1]->set + 1;
2076 for (idx = 0; idx < found; idx++) {
2077 /* entry_cnt is not zero, when cp_error was occurred */
2078 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2079 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2080 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2081 }
2082 }
2083 up_write(&nm_i->nat_tree_lock);
2084
2085 kfree(nm_i->nat_bitmap);
2086 sbi->nm_info = NULL;
2087 kfree(nm_i);
2088 }
2089
2090 int __init create_node_manager_caches(void)
2091 {
2092 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2093 sizeof(struct nat_entry));
2094 if (!nat_entry_slab)
2095 goto fail;
2096
2097 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2098 sizeof(struct free_nid));
2099 if (!free_nid_slab)
2100 goto destroy_nat_entry;
2101
2102 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2103 sizeof(struct nat_entry_set));
2104 if (!nat_entry_set_slab)
2105 goto destroy_free_nid;
2106 return 0;
2107
2108 destroy_free_nid:
2109 kmem_cache_destroy(free_nid_slab);
2110 destroy_nat_entry:
2111 kmem_cache_destroy(nat_entry_slab);
2112 fail:
2113 return -ENOMEM;
2114 }
2115
2116 void destroy_node_manager_caches(void)
2117 {
2118 kmem_cache_destroy(nat_entry_set_slab);
2119 kmem_cache_destroy(free_nid_slab);
2120 kmem_cache_destroy(nat_entry_slab);
2121 }
ubuntu-bionic-kernel mirror