projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mm: non-atomically mark page accessed during page cache allocation where possible
[deliverable/linux.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/events/f2fs.h>
23
24 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
25
26 static struct kmem_cache *nat_entry_slab;
27 static struct kmem_cache *free_nid_slab;
28
29 static inline bool available_free_memory(struct f2fs_nm_info *nm_i, int type)
30 {
31 struct sysinfo val;
32 unsigned long mem_size = 0;
33
34 si_meminfo(&val);
35 if (type == FREE_NIDS)
36 mem_size = nm_i->fcnt * sizeof(struct free_nid);
37 else if (type == NAT_ENTRIES)
38 mem_size += nm_i->nat_cnt * sizeof(struct nat_entry);
39 mem_size >>= 12;
40
41 /* give 50:50 memory for free nids and nat caches respectively */
42 return (mem_size < ((val.totalram * nm_i->ram_thresh) >> 11));
43 }
44
45 static void clear_node_page_dirty(struct page *page)
46 {
47 struct address_space *mapping = page->mapping;
48 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
49 unsigned int long flags;
50
51 if (PageDirty(page)) {
52 spin_lock_irqsave(&mapping->tree_lock, flags);
53 radix_tree_tag_clear(&mapping->page_tree,
54 page_index(page),
55 PAGECACHE_TAG_DIRTY);
56 spin_unlock_irqrestore(&mapping->tree_lock, flags);
57
58 clear_page_dirty_for_io(page);
59 dec_page_count(sbi, F2FS_DIRTY_NODES);
60 }
61 ClearPageUptodate(page);
62 }
63
64 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
65 {
66 pgoff_t index = current_nat_addr(sbi, nid);
67 return get_meta_page(sbi, index);
68 }
69
70 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
71 {
72 struct page *src_page;
73 struct page *dst_page;
74 pgoff_t src_off;
75 pgoff_t dst_off;
76 void *src_addr;
77 void *dst_addr;
78 struct f2fs_nm_info *nm_i = NM_I(sbi);
79
80 src_off = current_nat_addr(sbi, nid);
81 dst_off = next_nat_addr(sbi, src_off);
82
83 /* get current nat block page with lock */
84 src_page = get_meta_page(sbi, src_off);
85
86 /* Dirty src_page means that it is already the new target NAT page. */
87 if (PageDirty(src_page))
88 return src_page;
89
90 dst_page = grab_meta_page(sbi, dst_off);
91
92 src_addr = page_address(src_page);
93 dst_addr = page_address(dst_page);
94 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
95 set_page_dirty(dst_page);
96 f2fs_put_page(src_page, 1);
97
98 set_to_next_nat(nm_i, nid);
99
100 return dst_page;
101 }
102
103 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
104 {
105 return radix_tree_lookup(&nm_i->nat_root, n);
106 }
107
108 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
109 nid_t start, unsigned int nr, struct nat_entry **ep)
110 {
111 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
112 }
113
114 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
115 {
116 list_del(&e->list);
117 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
118 nm_i->nat_cnt--;
119 kmem_cache_free(nat_entry_slab, e);
120 }
121
122 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
123 {
124 struct f2fs_nm_info *nm_i = NM_I(sbi);
125 struct nat_entry *e;
126 int is_cp = 1;
127
128 read_lock(&nm_i->nat_tree_lock);
129 e = __lookup_nat_cache(nm_i, nid);
130 if (e && !e->checkpointed)
131 is_cp = 0;
132 read_unlock(&nm_i->nat_tree_lock);
133 return is_cp;
134 }
135
136 bool fsync_mark_done(struct f2fs_sb_info *sbi, nid_t nid)
137 {
138 struct f2fs_nm_info *nm_i = NM_I(sbi);
139 struct nat_entry *e;
140 bool fsync_done = false;
141
142 read_lock(&nm_i->nat_tree_lock);
143 e = __lookup_nat_cache(nm_i, nid);
144 if (e)
145 fsync_done = e->fsync_done;
146 read_unlock(&nm_i->nat_tree_lock);
147 return fsync_done;
148 }
149
150 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
151 {
152 struct nat_entry *new;
153
154 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
155 if (!new)
156 return NULL;
157 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
158 kmem_cache_free(nat_entry_slab, new);
159 return NULL;
160 }
161 memset(new, 0, sizeof(struct nat_entry));
162 nat_set_nid(new, nid);
163 new->checkpointed = true;
164 list_add_tail(&new->list, &nm_i->nat_entries);
165 nm_i->nat_cnt++;
166 return new;
167 }
168
169 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
170 struct f2fs_nat_entry *ne)
171 {
172 struct nat_entry *e;
173 retry:
174 write_lock(&nm_i->nat_tree_lock);
175 e = __lookup_nat_cache(nm_i, nid);
176 if (!e) {
177 e = grab_nat_entry(nm_i, nid);
178 if (!e) {
179 write_unlock(&nm_i->nat_tree_lock);
180 goto retry;
181 }
182 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
183 nat_set_ino(e, le32_to_cpu(ne->ino));
184 nat_set_version(e, ne->version);
185 }
186 write_unlock(&nm_i->nat_tree_lock);
187 }
188
189 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
190 block_t new_blkaddr, bool fsync_done)
191 {
192 struct f2fs_nm_info *nm_i = NM_I(sbi);
193 struct nat_entry *e;
194 retry:
195 write_lock(&nm_i->nat_tree_lock);
196 e = __lookup_nat_cache(nm_i, ni->nid);
197 if (!e) {
198 e = grab_nat_entry(nm_i, ni->nid);
199 if (!e) {
200 write_unlock(&nm_i->nat_tree_lock);
201 goto retry;
202 }
203 e->ni = *ni;
204 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
205 } else if (new_blkaddr == NEW_ADDR) {
206 /*
207 * when nid is reallocated,
208 * previous nat entry can be remained in nat cache.
209 * So, reinitialize it with new information.
210 */
211 e->ni = *ni;
212 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
213 }
214
215 /* sanity check */
216 f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
217 f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
218 new_blkaddr == NULL_ADDR);
219 f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
220 new_blkaddr == NEW_ADDR);
221 f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
222 nat_get_blkaddr(e) != NULL_ADDR &&
223 new_blkaddr == NEW_ADDR);
224
225 /* increament version no as node is removed */
226 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
227 unsigned char version = nat_get_version(e);
228 nat_set_version(e, inc_node_version(version));
229 }
230
231 /* change address */
232 nat_set_blkaddr(e, new_blkaddr);
233 __set_nat_cache_dirty(nm_i, e);
234
235 /* update fsync_mark if its inode nat entry is still alive */
236 e = __lookup_nat_cache(nm_i, ni->ino);
237 if (e)
238 e->fsync_done = fsync_done;
239 write_unlock(&nm_i->nat_tree_lock);
240 }
241
242 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
243 {
244 struct f2fs_nm_info *nm_i = NM_I(sbi);
245
246 if (available_free_memory(nm_i, NAT_ENTRIES))
247 return 0;
248
249 write_lock(&nm_i->nat_tree_lock);
250 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
251 struct nat_entry *ne;
252 ne = list_first_entry(&nm_i->nat_entries,
253 struct nat_entry, list);
254 __del_from_nat_cache(nm_i, ne);
255 nr_shrink--;
256 }
257 write_unlock(&nm_i->nat_tree_lock);
258 return nr_shrink;
259 }
260
261 /*
262 * This function returns always success
263 */
264 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
265 {
266 struct f2fs_nm_info *nm_i = NM_I(sbi);
267 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
268 struct f2fs_summary_block *sum = curseg->sum_blk;
269 nid_t start_nid = START_NID(nid);
270 struct f2fs_nat_block *nat_blk;
271 struct page *page = NULL;
272 struct f2fs_nat_entry ne;
273 struct nat_entry *e;
274 int i;
275
276 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
277 ni->nid = nid;
278
279 /* Check nat cache */
280 read_lock(&nm_i->nat_tree_lock);
281 e = __lookup_nat_cache(nm_i, nid);
282 if (e) {
283 ni->ino = nat_get_ino(e);
284 ni->blk_addr = nat_get_blkaddr(e);
285 ni->version = nat_get_version(e);
286 }
287 read_unlock(&nm_i->nat_tree_lock);
288 if (e)
289 return;
290
291 /* Check current segment summary */
292 mutex_lock(&curseg->curseg_mutex);
293 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
294 if (i >= 0) {
295 ne = nat_in_journal(sum, i);
296 node_info_from_raw_nat(ni, &ne);
297 }
298 mutex_unlock(&curseg->curseg_mutex);
299 if (i >= 0)
300 goto cache;
301
302 /* Fill node_info from nat page */
303 page = get_current_nat_page(sbi, start_nid);
304 nat_blk = (struct f2fs_nat_block *)page_address(page);
305 ne = nat_blk->entries[nid - start_nid];
306 node_info_from_raw_nat(ni, &ne);
307 f2fs_put_page(page, 1);
308 cache:
309 /* cache nat entry */
310 cache_nat_entry(NM_I(sbi), nid, &ne);
311 }
312
313 /*
314 * The maximum depth is four.
315 * Offset[0] will have raw inode offset.
316 */
317 static int get_node_path(struct f2fs_inode_info *fi, long block,
318 int offset[4], unsigned int noffset[4])
319 {
320 const long direct_index = ADDRS_PER_INODE(fi);
321 const long direct_blks = ADDRS_PER_BLOCK;
322 const long dptrs_per_blk = NIDS_PER_BLOCK;
323 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
324 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
325 int n = 0;
326 int level = 0;
327
328 noffset[0] = 0;
329
330 if (block < direct_index) {
331 offset[n] = block;
332 goto got;
333 }
334 block -= direct_index;
335 if (block < direct_blks) {
336 offset[n++] = NODE_DIR1_BLOCK;
337 noffset[n] = 1;
338 offset[n] = block;
339 level = 1;
340 goto got;
341 }
342 block -= direct_blks;
343 if (block < direct_blks) {
344 offset[n++] = NODE_DIR2_BLOCK;
345 noffset[n] = 2;
346 offset[n] = block;
347 level = 1;
348 goto got;
349 }
350 block -= direct_blks;
351 if (block < indirect_blks) {
352 offset[n++] = NODE_IND1_BLOCK;
353 noffset[n] = 3;
354 offset[n++] = block / direct_blks;
355 noffset[n] = 4 + offset[n - 1];
356 offset[n] = block % direct_blks;
357 level = 2;
358 goto got;
359 }
360 block -= indirect_blks;
361 if (block < indirect_blks) {
362 offset[n++] = NODE_IND2_BLOCK;
363 noffset[n] = 4 + dptrs_per_blk;
364 offset[n++] = block / direct_blks;
365 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
366 offset[n] = block % direct_blks;
367 level = 2;
368 goto got;
369 }
370 block -= indirect_blks;
371 if (block < dindirect_blks) {
372 offset[n++] = NODE_DIND_BLOCK;
373 noffset[n] = 5 + (dptrs_per_blk * 2);
374 offset[n++] = block / indirect_blks;
375 noffset[n] = 6 + (dptrs_per_blk * 2) +
376 offset[n - 1] * (dptrs_per_blk + 1);
377 offset[n++] = (block / direct_blks) % dptrs_per_blk;
378 noffset[n] = 7 + (dptrs_per_blk * 2) +
379 offset[n - 2] * (dptrs_per_blk + 1) +
380 offset[n - 1];
381 offset[n] = block % direct_blks;
382 level = 3;
383 goto got;
384 } else {
385 BUG();
386 }
387 got:
388 return level;
389 }
390
391 /*
392 * Caller should call f2fs_put_dnode(dn).
393 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
394 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
395 * In the case of RDONLY_NODE, we don't need to care about mutex.
396 */
397 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
398 {
399 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
400 struct page *npage[4];
401 struct page *parent;
402 int offset[4];
403 unsigned int noffset[4];
404 nid_t nids[4];
405 int level, i;
406 int err = 0;
407
408 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
409
410 nids[0] = dn->inode->i_ino;
411 npage[0] = dn->inode_page;
412
413 if (!npage[0]) {
414 npage[0] = get_node_page(sbi, nids[0]);
415 if (IS_ERR(npage[0]))
416 return PTR_ERR(npage[0]);
417 }
418 parent = npage[0];
419 if (level != 0)
420 nids[1] = get_nid(parent, offset[0], true);
421 dn->inode_page = npage[0];
422 dn->inode_page_locked = true;
423
424 /* get indirect or direct nodes */
425 for (i = 1; i <= level; i++) {
426 bool done = false;
427
428 if (!nids[i] && mode == ALLOC_NODE) {
429 /* alloc new node */
430 if (!alloc_nid(sbi, &(nids[i]))) {
431 err = -ENOSPC;
432 goto release_pages;
433 }
434
435 dn->nid = nids[i];
436 npage[i] = new_node_page(dn, noffset[i], NULL);
437 if (IS_ERR(npage[i])) {
438 alloc_nid_failed(sbi, nids[i]);
439 err = PTR_ERR(npage[i]);
440 goto release_pages;
441 }
442
443 set_nid(parent, offset[i - 1], nids[i], i == 1);
444 alloc_nid_done(sbi, nids[i]);
445 done = true;
446 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
447 npage[i] = get_node_page_ra(parent, offset[i - 1]);
448 if (IS_ERR(npage[i])) {
449 err = PTR_ERR(npage[i]);
450 goto release_pages;
451 }
452 done = true;
453 }
454 if (i == 1) {
455 dn->inode_page_locked = false;
456 unlock_page(parent);
457 } else {
458 f2fs_put_page(parent, 1);
459 }
460
461 if (!done) {
462 npage[i] = get_node_page(sbi, nids[i]);
463 if (IS_ERR(npage[i])) {
464 err = PTR_ERR(npage[i]);
465 f2fs_put_page(npage[0], 0);
466 goto release_out;
467 }
468 }
469 if (i < level) {
470 parent = npage[i];
471 nids[i + 1] = get_nid(parent, offset[i], false);
472 }
473 }
474 dn->nid = nids[level];
475 dn->ofs_in_node = offset[level];
476 dn->node_page = npage[level];
477 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
478 return 0;
479
480 release_pages:
481 f2fs_put_page(parent, 1);
482 if (i > 1)
483 f2fs_put_page(npage[0], 0);
484 release_out:
485 dn->inode_page = NULL;
486 dn->node_page = NULL;
487 return err;
488 }
489
490 static void truncate_node(struct dnode_of_data *dn)
491 {
492 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
493 struct node_info ni;
494
495 get_node_info(sbi, dn->nid, &ni);
496 if (dn->inode->i_blocks == 0) {
497 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
498 goto invalidate;
499 }
500 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
501
502 /* Deallocate node address */
503 invalidate_blocks(sbi, ni.blk_addr);
504 dec_valid_node_count(sbi, dn->inode);
505 set_node_addr(sbi, &ni, NULL_ADDR, false);
506
507 if (dn->nid == dn->inode->i_ino) {
508 remove_orphan_inode(sbi, dn->nid);
509 dec_valid_inode_count(sbi);
510 } else {
511 sync_inode_page(dn);
512 }
513 invalidate:
514 clear_node_page_dirty(dn->node_page);
515 F2FS_SET_SB_DIRT(sbi);
516
517 f2fs_put_page(dn->node_page, 1);
518
519 invalidate_mapping_pages(NODE_MAPPING(sbi),
520 dn->node_page->index, dn->node_page->index);
521
522 dn->node_page = NULL;
523 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
524 }
525
526 static int truncate_dnode(struct dnode_of_data *dn)
527 {
528 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
529 struct page *page;
530
531 if (dn->nid == 0)
532 return 1;
533
534 /* get direct node */
535 page = get_node_page(sbi, dn->nid);
536 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
537 return 1;
538 else if (IS_ERR(page))
539 return PTR_ERR(page);
540
541 /* Make dnode_of_data for parameter */
542 dn->node_page = page;
543 dn->ofs_in_node = 0;
544 truncate_data_blocks(dn);
545 truncate_node(dn);
546 return 1;
547 }
548
549 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
550 int ofs, int depth)
551 {
552 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
553 struct dnode_of_data rdn = *dn;
554 struct page *page;
555 struct f2fs_node *rn;
556 nid_t child_nid;
557 unsigned int child_nofs;
558 int freed = 0;
559 int i, ret;
560
561 if (dn->nid == 0)
562 return NIDS_PER_BLOCK + 1;
563
564 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
565
566 page = get_node_page(sbi, dn->nid);
567 if (IS_ERR(page)) {
568 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
569 return PTR_ERR(page);
570 }
571
572 rn = F2FS_NODE(page);
573 if (depth < 3) {
574 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
575 child_nid = le32_to_cpu(rn->in.nid[i]);
576 if (child_nid == 0)
577 continue;
578 rdn.nid = child_nid;
579 ret = truncate_dnode(&rdn);
580 if (ret < 0)
581 goto out_err;
582 set_nid(page, i, 0, false);
583 }
584 } else {
585 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
586 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
587 child_nid = le32_to_cpu(rn->in.nid[i]);
588 if (child_nid == 0) {
589 child_nofs += NIDS_PER_BLOCK + 1;
590 continue;
591 }
592 rdn.nid = child_nid;
593 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
594 if (ret == (NIDS_PER_BLOCK + 1)) {
595 set_nid(page, i, 0, false);
596 child_nofs += ret;
597 } else if (ret < 0 && ret != -ENOENT) {
598 goto out_err;
599 }
600 }
601 freed = child_nofs;
602 }
603
604 if (!ofs) {
605 /* remove current indirect node */
606 dn->node_page = page;
607 truncate_node(dn);
608 freed++;
609 } else {
610 f2fs_put_page(page, 1);
611 }
612 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
613 return freed;
614
615 out_err:
616 f2fs_put_page(page, 1);
617 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
618 return ret;
619 }
620
621 static int truncate_partial_nodes(struct dnode_of_data *dn,
622 struct f2fs_inode *ri, int *offset, int depth)
623 {
624 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
625 struct page *pages[2];
626 nid_t nid[3];
627 nid_t child_nid;
628 int err = 0;
629 int i;
630 int idx = depth - 2;
631
632 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
633 if (!nid[0])
634 return 0;
635
636 /* get indirect nodes in the path */
637 for (i = 0; i < idx + 1; i++) {
638 /* refernece count'll be increased */
639 pages[i] = get_node_page(sbi, nid[i]);
640 if (IS_ERR(pages[i])) {
641 err = PTR_ERR(pages[i]);
642 idx = i - 1;
643 goto fail;
644 }
645 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
646 }
647
648 /* free direct nodes linked to a partial indirect node */
649 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
650 child_nid = get_nid(pages[idx], i, false);
651 if (!child_nid)
652 continue;
653 dn->nid = child_nid;
654 err = truncate_dnode(dn);
655 if (err < 0)
656 goto fail;
657 set_nid(pages[idx], i, 0, false);
658 }
659
660 if (offset[idx + 1] == 0) {
661 dn->node_page = pages[idx];
662 dn->nid = nid[idx];
663 truncate_node(dn);
664 } else {
665 f2fs_put_page(pages[idx], 1);
666 }
667 offset[idx]++;
668 offset[idx + 1] = 0;
669 idx--;
670 fail:
671 for (i = idx; i >= 0; i--)
672 f2fs_put_page(pages[i], 1);
673
674 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
675
676 return err;
677 }
678
679 /*
680 * All the block addresses of data and nodes should be nullified.
681 */
682 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
683 {
684 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
685 int err = 0, cont = 1;
686 int level, offset[4], noffset[4];
687 unsigned int nofs = 0;
688 struct f2fs_inode *ri;
689 struct dnode_of_data dn;
690 struct page *page;
691
692 trace_f2fs_truncate_inode_blocks_enter(inode, from);
693
694 level = get_node_path(F2FS_I(inode), from, offset, noffset);
695 restart:
696 page = get_node_page(sbi, inode->i_ino);
697 if (IS_ERR(page)) {
698 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
699 return PTR_ERR(page);
700 }
701
702 set_new_dnode(&dn, inode, page, NULL, 0);
703 unlock_page(page);
704
705 ri = F2FS_INODE(page);
706 switch (level) {
707 case 0:
708 case 1:
709 nofs = noffset[1];
710 break;
711 case 2:
712 nofs = noffset[1];
713 if (!offset[level - 1])
714 goto skip_partial;
715 err = truncate_partial_nodes(&dn, ri, offset, level);
716 if (err < 0 && err != -ENOENT)
717 goto fail;
718 nofs += 1 + NIDS_PER_BLOCK;
719 break;
720 case 3:
721 nofs = 5 + 2 * NIDS_PER_BLOCK;
722 if (!offset[level - 1])
723 goto skip_partial;
724 err = truncate_partial_nodes(&dn, ri, offset, level);
725 if (err < 0 && err != -ENOENT)
726 goto fail;
727 break;
728 default:
729 BUG();
730 }
731
732 skip_partial:
733 while (cont) {
734 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
735 switch (offset[0]) {
736 case NODE_DIR1_BLOCK:
737 case NODE_DIR2_BLOCK:
738 err = truncate_dnode(&dn);
739 break;
740
741 case NODE_IND1_BLOCK:
742 case NODE_IND2_BLOCK:
743 err = truncate_nodes(&dn, nofs, offset[1], 2);
744 break;
745
746 case NODE_DIND_BLOCK:
747 err = truncate_nodes(&dn, nofs, offset[1], 3);
748 cont = 0;
749 break;
750
751 default:
752 BUG();
753 }
754 if (err < 0 && err != -ENOENT)
755 goto fail;
756 if (offset[1] == 0 &&
757 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
758 lock_page(page);
759 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
760 f2fs_put_page(page, 1);
761 goto restart;
762 }
763 f2fs_wait_on_page_writeback(page, NODE);
764 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
765 set_page_dirty(page);
766 unlock_page(page);
767 }
768 offset[1] = 0;
769 offset[0]++;
770 nofs += err;
771 }
772 fail:
773 f2fs_put_page(page, 0);
774 trace_f2fs_truncate_inode_blocks_exit(inode, err);
775 return err > 0 ? 0 : err;
776 }
777
778 int truncate_xattr_node(struct inode *inode, struct page *page)
779 {
780 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
781 nid_t nid = F2FS_I(inode)->i_xattr_nid;
782 struct dnode_of_data dn;
783 struct page *npage;
784
785 if (!nid)
786 return 0;
787
788 npage = get_node_page(sbi, nid);
789 if (IS_ERR(npage))
790 return PTR_ERR(npage);
791
792 F2FS_I(inode)->i_xattr_nid = 0;
793
794 /* need to do checkpoint during fsync */
795 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
796
797 set_new_dnode(&dn, inode, page, npage, nid);
798
799 if (page)
800 dn.inode_page_locked = true;
801 truncate_node(&dn);
802 return 0;
803 }
804
805 /*
806 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
807 * f2fs_unlock_op().
808 */
809 void remove_inode_page(struct inode *inode)
810 {
811 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
812 struct page *page;
813 nid_t ino = inode->i_ino;
814 struct dnode_of_data dn;
815
816 page = get_node_page(sbi, ino);
817 if (IS_ERR(page))
818 return;
819
820 if (truncate_xattr_node(inode, page)) {
821 f2fs_put_page(page, 1);
822 return;
823 }
824 /* 0 is possible, after f2fs_new_inode() is failed */
825 f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
826 set_new_dnode(&dn, inode, page, page, ino);
827 truncate_node(&dn);
828 }
829
830 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
831 {
832 struct dnode_of_data dn;
833
834 /* allocate inode page for new inode */
835 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
836
837 /* caller should f2fs_put_page(page, 1); */
838 return new_node_page(&dn, 0, NULL);
839 }
840
841 struct page *new_node_page(struct dnode_of_data *dn,
842 unsigned int ofs, struct page *ipage)
843 {
844 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
845 struct node_info old_ni, new_ni;
846 struct page *page;
847 int err;
848
849 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
850 return ERR_PTR(-EPERM);
851
852 page = grab_cache_page_write_begin(NODE_MAPPING(sbi),
853 dn->nid, AOP_FLAG_NOFS);
854 if (!page)
855 return ERR_PTR(-ENOMEM);
856
857 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
858 err = -ENOSPC;
859 goto fail;
860 }
861
862 get_node_info(sbi, dn->nid, &old_ni);
863
864 /* Reinitialize old_ni with new node page */
865 f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
866 new_ni = old_ni;
867 new_ni.ino = dn->inode->i_ino;
868 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
869
870 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
871 set_cold_node(dn->inode, page);
872 SetPageUptodate(page);
873 set_page_dirty(page);
874
875 if (f2fs_has_xattr_block(ofs))
876 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
877
878 dn->node_page = page;
879 if (ipage)
880 update_inode(dn->inode, ipage);
881 else
882 sync_inode_page(dn);
883 if (ofs == 0)
884 inc_valid_inode_count(sbi);
885
886 return page;
887
888 fail:
889 clear_node_page_dirty(page);
890 f2fs_put_page(page, 1);
891 return ERR_PTR(err);
892 }
893
894 /*
895 * Caller should do after getting the following values.
896 * 0: f2fs_put_page(page, 0)
897 * LOCKED_PAGE: f2fs_put_page(page, 1)
898 * error: nothing
899 */
900 static int read_node_page(struct page *page, int rw)
901 {
902 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
903 struct node_info ni;
904
905 get_node_info(sbi, page->index, &ni);
906
907 if (unlikely(ni.blk_addr == NULL_ADDR)) {
908 f2fs_put_page(page, 1);
909 return -ENOENT;
910 }
911
912 if (PageUptodate(page))
913 return LOCKED_PAGE;
914
915 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
916 }
917
918 /*
919 * Readahead a node page
920 */
921 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
922 {
923 struct page *apage;
924 int err;
925
926 apage = find_get_page(NODE_MAPPING(sbi), nid);
927 if (apage && PageUptodate(apage)) {
928 f2fs_put_page(apage, 0);
929 return;
930 }
931 f2fs_put_page(apage, 0);
932
933 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
934 if (!apage)
935 return;
936
937 err = read_node_page(apage, READA);
938 if (err == 0)
939 f2fs_put_page(apage, 0);
940 else if (err == LOCKED_PAGE)
941 f2fs_put_page(apage, 1);
942 }
943
944 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
945 {
946 struct page *page;
947 int err;
948 repeat:
949 page = grab_cache_page_write_begin(NODE_MAPPING(sbi),
950 nid, AOP_FLAG_NOFS);
951 if (!page)
952 return ERR_PTR(-ENOMEM);
953
954 err = read_node_page(page, READ_SYNC);
955 if (err < 0)
956 return ERR_PTR(err);
957 else if (err == LOCKED_PAGE)
958 goto got_it;
959
960 lock_page(page);
961 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
962 f2fs_put_page(page, 1);
963 return ERR_PTR(-EIO);
964 }
965 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
966 f2fs_put_page(page, 1);
967 goto repeat;
968 }
969 got_it:
970 return page;
971 }
972
973 /*
974 * Return a locked page for the desired node page.
975 * And, readahead MAX_RA_NODE number of node pages.
976 */
977 struct page *get_node_page_ra(struct page *parent, int start)
978 {
979 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
980 struct blk_plug plug;
981 struct page *page;
982 int err, i, end;
983 nid_t nid;
984
985 /* First, try getting the desired direct node. */
986 nid = get_nid(parent, start, false);
987 if (!nid)
988 return ERR_PTR(-ENOENT);
989 repeat:
990 page = grab_cache_page(NODE_MAPPING(sbi), nid);
991 if (!page)
992 return ERR_PTR(-ENOMEM);
993
994 err = read_node_page(page, READ_SYNC);
995 if (err < 0)
996 return ERR_PTR(err);
997 else if (err == LOCKED_PAGE)
998 goto page_hit;
999
1000 blk_start_plug(&plug);
1001
1002 /* Then, try readahead for siblings of the desired node */
1003 end = start + MAX_RA_NODE;
1004 end = min(end, NIDS_PER_BLOCK);
1005 for (i = start + 1; i < end; i++) {
1006 nid = get_nid(parent, i, false);
1007 if (!nid)
1008 continue;
1009 ra_node_page(sbi, nid);
1010 }
1011
1012 blk_finish_plug(&plug);
1013
1014 lock_page(page);
1015 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1016 f2fs_put_page(page, 1);
1017 goto repeat;
1018 }
1019 page_hit:
1020 if (unlikely(!PageUptodate(page))) {
1021 f2fs_put_page(page, 1);
1022 return ERR_PTR(-EIO);
1023 }
1024 return page;
1025 }
1026
1027 void sync_inode_page(struct dnode_of_data *dn)
1028 {
1029 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1030 update_inode(dn->inode, dn->node_page);
1031 } else if (dn->inode_page) {
1032 if (!dn->inode_page_locked)
1033 lock_page(dn->inode_page);
1034 update_inode(dn->inode, dn->inode_page);
1035 if (!dn->inode_page_locked)
1036 unlock_page(dn->inode_page);
1037 } else {
1038 update_inode_page(dn->inode);
1039 }
1040 }
1041
1042 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1043 struct writeback_control *wbc)
1044 {
1045 pgoff_t index, end;
1046 struct pagevec pvec;
1047 int step = ino ? 2 : 0;
1048 int nwritten = 0, wrote = 0;
1049
1050 pagevec_init(&pvec, 0);
1051
1052 next_step:
1053 index = 0;
1054 end = LONG_MAX;
1055
1056 while (index <= end) {
1057 int i, nr_pages;
1058 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1059 PAGECACHE_TAG_DIRTY,
1060 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1061 if (nr_pages == 0)
1062 break;
1063
1064 for (i = 0; i < nr_pages; i++) {
1065 struct page *page = pvec.pages[i];
1066
1067 /*
1068 * flushing sequence with step:
1069 * 0. indirect nodes
1070 * 1. dentry dnodes
1071 * 2. file dnodes
1072 */
1073 if (step == 0 && IS_DNODE(page))
1074 continue;
1075 if (step == 1 && (!IS_DNODE(page) ||
1076 is_cold_node(page)))
1077 continue;
1078 if (step == 2 && (!IS_DNODE(page) ||
1079 !is_cold_node(page)))
1080 continue;
1081
1082 /*
1083 * If an fsync mode,
1084 * we should not skip writing node pages.
1085 */
1086 if (ino && ino_of_node(page) == ino)
1087 lock_page(page);
1088 else if (!trylock_page(page))
1089 continue;
1090
1091 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1092 continue_unlock:
1093 unlock_page(page);
1094 continue;
1095 }
1096 if (ino && ino_of_node(page) != ino)
1097 goto continue_unlock;
1098
1099 if (!PageDirty(page)) {
1100 /* someone wrote it for us */
1101 goto continue_unlock;
1102 }
1103
1104 if (!clear_page_dirty_for_io(page))
1105 goto continue_unlock;
1106
1107 /* called by fsync() */
1108 if (ino && IS_DNODE(page)) {
1109 int mark = !is_checkpointed_node(sbi, ino);
1110 set_fsync_mark(page, 1);
1111 if (IS_INODE(page))
1112 set_dentry_mark(page, mark);
1113 nwritten++;
1114 } else {
1115 set_fsync_mark(page, 0);
1116 set_dentry_mark(page, 0);
1117 }
1118 NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1119 wrote++;
1120
1121 if (--wbc->nr_to_write == 0)
1122 break;
1123 }
1124 pagevec_release(&pvec);
1125 cond_resched();
1126
1127 if (wbc->nr_to_write == 0) {
1128 step = 2;
1129 break;
1130 }
1131 }
1132
1133 if (step < 2) {
1134 step++;
1135 goto next_step;
1136 }
1137
1138 if (wrote)
1139 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1140 return nwritten;
1141 }
1142
1143 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1144 {
1145 pgoff_t index = 0, end = LONG_MAX;
1146 struct pagevec pvec;
1147 int ret2 = 0, ret = 0;
1148
1149 pagevec_init(&pvec, 0);
1150
1151 while (index <= end) {
1152 int i, nr_pages;
1153 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1154 PAGECACHE_TAG_WRITEBACK,
1155 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1156 if (nr_pages == 0)
1157 break;
1158
1159 for (i = 0; i < nr_pages; i++) {
1160 struct page *page = pvec.pages[i];
1161
1162 /* until radix tree lookup accepts end_index */
1163 if (unlikely(page->index > end))
1164 continue;
1165
1166 if (ino && ino_of_node(page) == ino) {
1167 f2fs_wait_on_page_writeback(page, NODE);
1168 if (TestClearPageError(page))
1169 ret = -EIO;
1170 }
1171 }
1172 pagevec_release(&pvec);
1173 cond_resched();
1174 }
1175
1176 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1177 ret2 = -ENOSPC;
1178 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1179 ret2 = -EIO;
1180 if (!ret)
1181 ret = ret2;
1182 return ret;
1183 }
1184
1185 static int f2fs_write_node_page(struct page *page,
1186 struct writeback_control *wbc)
1187 {
1188 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1189 nid_t nid;
1190 block_t new_addr;
1191 struct node_info ni;
1192 struct f2fs_io_info fio = {
1193 .type = NODE,
1194 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1195 };
1196
1197 if (unlikely(sbi->por_doing))
1198 goto redirty_out;
1199
1200 f2fs_wait_on_page_writeback(page, NODE);
1201
1202 /* get old block addr of this node page */
1203 nid = nid_of_node(page);
1204 f2fs_bug_on(page->index != nid);
1205
1206 get_node_info(sbi, nid, &ni);
1207
1208 /* This page is already truncated */
1209 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1210 dec_page_count(sbi, F2FS_DIRTY_NODES);
1211 unlock_page(page);
1212 return 0;
1213 }
1214
1215 if (wbc->for_reclaim)
1216 goto redirty_out;
1217
1218 mutex_lock(&sbi->node_write);
1219 set_page_writeback(page);
1220 write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1221 set_node_addr(sbi, &ni, new_addr, is_fsync_dnode(page));
1222 dec_page_count(sbi, F2FS_DIRTY_NODES);
1223 mutex_unlock(&sbi->node_write);
1224 unlock_page(page);
1225 return 0;
1226
1227 redirty_out:
1228 dec_page_count(sbi, F2FS_DIRTY_NODES);
1229 wbc->pages_skipped++;
1230 account_page_redirty(page);
1231 set_page_dirty(page);
1232 return AOP_WRITEPAGE_ACTIVATE;
1233 }
1234
1235 static int f2fs_write_node_pages(struct address_space *mapping,
1236 struct writeback_control *wbc)
1237 {
1238 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1239 long diff;
1240
1241 /* balancing f2fs's metadata in background */
1242 f2fs_balance_fs_bg(sbi);
1243
1244 /* collect a number of dirty node pages and write together */
1245 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1246 goto skip_write;
1247
1248 diff = nr_pages_to_write(sbi, NODE, wbc);
1249 wbc->sync_mode = WB_SYNC_NONE;
1250 sync_node_pages(sbi, 0, wbc);
1251 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1252 return 0;
1253
1254 skip_write:
1255 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1256 return 0;
1257 }
1258
1259 static int f2fs_set_node_page_dirty(struct page *page)
1260 {
1261 struct address_space *mapping = page->mapping;
1262 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1263
1264 trace_f2fs_set_page_dirty(page, NODE);
1265
1266 SetPageUptodate(page);
1267 if (!PageDirty(page)) {
1268 __set_page_dirty_nobuffers(page);
1269 inc_page_count(sbi, F2FS_DIRTY_NODES);
1270 SetPagePrivate(page);
1271 return 1;
1272 }
1273 return 0;
1274 }
1275
1276 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1277 unsigned int length)
1278 {
1279 struct inode *inode = page->mapping->host;
1280 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1281 if (PageDirty(page))
1282 dec_page_count(sbi, F2FS_DIRTY_NODES);
1283 ClearPagePrivate(page);
1284 }
1285
1286 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1287 {
1288 ClearPagePrivate(page);
1289 return 1;
1290 }
1291
1292 /*
1293 * Structure of the f2fs node operations
1294 */
1295 const struct address_space_operations f2fs_node_aops = {
1296 .writepage = f2fs_write_node_page,
1297 .writepages = f2fs_write_node_pages,
1298 .set_page_dirty = f2fs_set_node_page_dirty,
1299 .invalidatepage = f2fs_invalidate_node_page,
1300 .releasepage = f2fs_release_node_page,
1301 };
1302
1303 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1304 nid_t n)
1305 {
1306 return radix_tree_lookup(&nm_i->free_nid_root, n);
1307 }
1308
1309 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1310 struct free_nid *i)
1311 {
1312 list_del(&i->list);
1313 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1314 }
1315
1316 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1317 {
1318 struct free_nid *i;
1319 struct nat_entry *ne;
1320 bool allocated = false;
1321
1322 if (!available_free_memory(nm_i, FREE_NIDS))
1323 return -1;
1324
1325 /* 0 nid should not be used */
1326 if (unlikely(nid == 0))
1327 return 0;
1328
1329 if (build) {
1330 /* do not add allocated nids */
1331 read_lock(&nm_i->nat_tree_lock);
1332 ne = __lookup_nat_cache(nm_i, nid);
1333 if (ne &&
1334 (!ne->checkpointed || nat_get_blkaddr(ne) != NULL_ADDR))
1335 allocated = true;
1336 read_unlock(&nm_i->nat_tree_lock);
1337 if (allocated)
1338 return 0;
1339 }
1340
1341 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1342 i->nid = nid;
1343 i->state = NID_NEW;
1344
1345 spin_lock(&nm_i->free_nid_list_lock);
1346 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1347 spin_unlock(&nm_i->free_nid_list_lock);
1348 kmem_cache_free(free_nid_slab, i);
1349 return 0;
1350 }
1351 list_add_tail(&i->list, &nm_i->free_nid_list);
1352 nm_i->fcnt++;
1353 spin_unlock(&nm_i->free_nid_list_lock);
1354 return 1;
1355 }
1356
1357 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1358 {
1359 struct free_nid *i;
1360 bool need_free = false;
1361
1362 spin_lock(&nm_i->free_nid_list_lock);
1363 i = __lookup_free_nid_list(nm_i, nid);
1364 if (i && i->state == NID_NEW) {
1365 __del_from_free_nid_list(nm_i, i);
1366 nm_i->fcnt--;
1367 need_free = true;
1368 }
1369 spin_unlock(&nm_i->free_nid_list_lock);
1370
1371 if (need_free)
1372 kmem_cache_free(free_nid_slab, i);
1373 }
1374
1375 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1376 struct page *nat_page, nid_t start_nid)
1377 {
1378 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1379 block_t blk_addr;
1380 int i;
1381
1382 i = start_nid % NAT_ENTRY_PER_BLOCK;
1383
1384 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1385
1386 if (unlikely(start_nid >= nm_i->max_nid))
1387 break;
1388
1389 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1390 f2fs_bug_on(blk_addr == NEW_ADDR);
1391 if (blk_addr == NULL_ADDR) {
1392 if (add_free_nid(nm_i, start_nid, true) < 0)
1393 break;
1394 }
1395 }
1396 }
1397
1398 static void build_free_nids(struct f2fs_sb_info *sbi)
1399 {
1400 struct f2fs_nm_info *nm_i = NM_I(sbi);
1401 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1402 struct f2fs_summary_block *sum = curseg->sum_blk;
1403 int i = 0;
1404 nid_t nid = nm_i->next_scan_nid;
1405
1406 /* Enough entries */
1407 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1408 return;
1409
1410 /* readahead nat pages to be scanned */
1411 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1412
1413 while (1) {
1414 struct page *page = get_current_nat_page(sbi, nid);
1415
1416 scan_nat_page(nm_i, page, nid);
1417 f2fs_put_page(page, 1);
1418
1419 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1420 if (unlikely(nid >= nm_i->max_nid))
1421 nid = 0;
1422
1423 if (i++ == FREE_NID_PAGES)
1424 break;
1425 }
1426
1427 /* go to the next free nat pages to find free nids abundantly */
1428 nm_i->next_scan_nid = nid;
1429
1430 /* find free nids from current sum_pages */
1431 mutex_lock(&curseg->curseg_mutex);
1432 for (i = 0; i < nats_in_cursum(sum); i++) {
1433 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1434 nid = le32_to_cpu(nid_in_journal(sum, i));
1435 if (addr == NULL_ADDR)
1436 add_free_nid(nm_i, nid, true);
1437 else
1438 remove_free_nid(nm_i, nid);
1439 }
1440 mutex_unlock(&curseg->curseg_mutex);
1441 }
1442
1443 /*
1444 * If this function returns success, caller can obtain a new nid
1445 * from second parameter of this function.
1446 * The returned nid could be used ino as well as nid when inode is created.
1447 */
1448 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1449 {
1450 struct f2fs_nm_info *nm_i = NM_I(sbi);
1451 struct free_nid *i = NULL;
1452 retry:
1453 if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1454 return false;
1455
1456 spin_lock(&nm_i->free_nid_list_lock);
1457
1458 /* We should not use stale free nids created by build_free_nids */
1459 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1460 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1461 list_for_each_entry(i, &nm_i->free_nid_list, list)
1462 if (i->state == NID_NEW)
1463 break;
1464
1465 f2fs_bug_on(i->state != NID_NEW);
1466 *nid = i->nid;
1467 i->state = NID_ALLOC;
1468 nm_i->fcnt--;
1469 spin_unlock(&nm_i->free_nid_list_lock);
1470 return true;
1471 }
1472 spin_unlock(&nm_i->free_nid_list_lock);
1473
1474 /* Let's scan nat pages and its caches to get free nids */
1475 mutex_lock(&nm_i->build_lock);
1476 build_free_nids(sbi);
1477 mutex_unlock(&nm_i->build_lock);
1478 goto retry;
1479 }
1480
1481 /*
1482 * alloc_nid() should be called prior to this function.
1483 */
1484 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1485 {
1486 struct f2fs_nm_info *nm_i = NM_I(sbi);
1487 struct free_nid *i;
1488
1489 spin_lock(&nm_i->free_nid_list_lock);
1490 i = __lookup_free_nid_list(nm_i, nid);
1491 f2fs_bug_on(!i || i->state != NID_ALLOC);
1492 __del_from_free_nid_list(nm_i, i);
1493 spin_unlock(&nm_i->free_nid_list_lock);
1494
1495 kmem_cache_free(free_nid_slab, i);
1496 }
1497
1498 /*
1499 * alloc_nid() should be called prior to this function.
1500 */
1501 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1502 {
1503 struct f2fs_nm_info *nm_i = NM_I(sbi);
1504 struct free_nid *i;
1505 bool need_free = false;
1506
1507 if (!nid)
1508 return;
1509
1510 spin_lock(&nm_i->free_nid_list_lock);
1511 i = __lookup_free_nid_list(nm_i, nid);
1512 f2fs_bug_on(!i || i->state != NID_ALLOC);
1513 if (!available_free_memory(nm_i, FREE_NIDS)) {
1514 __del_from_free_nid_list(nm_i, i);
1515 need_free = true;
1516 } else {
1517 i->state = NID_NEW;
1518 nm_i->fcnt++;
1519 }
1520 spin_unlock(&nm_i->free_nid_list_lock);
1521
1522 if (need_free)
1523 kmem_cache_free(free_nid_slab, i);
1524 }
1525
1526 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1527 struct f2fs_summary *sum, struct node_info *ni,
1528 block_t new_blkaddr)
1529 {
1530 rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1531 set_node_addr(sbi, ni, new_blkaddr, false);
1532 clear_node_page_dirty(page);
1533 }
1534
1535 void recover_inline_xattr(struct inode *inode, struct page *page)
1536 {
1537 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1538 void *src_addr, *dst_addr;
1539 size_t inline_size;
1540 struct page *ipage;
1541 struct f2fs_inode *ri;
1542
1543 if (!f2fs_has_inline_xattr(inode))
1544 return;
1545
1546 if (!IS_INODE(page))
1547 return;
1548
1549 ri = F2FS_INODE(page);
1550 if (!(ri->i_inline & F2FS_INLINE_XATTR))
1551 return;
1552
1553 ipage = get_node_page(sbi, inode->i_ino);
1554 f2fs_bug_on(IS_ERR(ipage));
1555
1556 dst_addr = inline_xattr_addr(ipage);
1557 src_addr = inline_xattr_addr(page);
1558 inline_size = inline_xattr_size(inode);
1559
1560 memcpy(dst_addr, src_addr, inline_size);
1561
1562 update_inode(inode, ipage);
1563 f2fs_put_page(ipage, 1);
1564 }
1565
1566 bool recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1567 {
1568 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1569 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1570 nid_t new_xnid = nid_of_node(page);
1571 struct node_info ni;
1572
1573 recover_inline_xattr(inode, page);
1574
1575 if (!f2fs_has_xattr_block(ofs_of_node(page)))
1576 return false;
1577
1578 /* 1: invalidate the previous xattr nid */
1579 if (!prev_xnid)
1580 goto recover_xnid;
1581
1582 /* Deallocate node address */
1583 get_node_info(sbi, prev_xnid, &ni);
1584 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
1585 invalidate_blocks(sbi, ni.blk_addr);
1586 dec_valid_node_count(sbi, inode);
1587 set_node_addr(sbi, &ni, NULL_ADDR, false);
1588
1589 recover_xnid:
1590 /* 2: allocate new xattr nid */
1591 if (unlikely(!inc_valid_node_count(sbi, inode)))
1592 f2fs_bug_on(1);
1593
1594 remove_free_nid(NM_I(sbi), new_xnid);
1595 get_node_info(sbi, new_xnid, &ni);
1596 ni.ino = inode->i_ino;
1597 set_node_addr(sbi, &ni, NEW_ADDR, false);
1598 F2FS_I(inode)->i_xattr_nid = new_xnid;
1599
1600 /* 3: update xattr blkaddr */
1601 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1602 set_node_addr(sbi, &ni, blkaddr, false);
1603
1604 update_inode_page(inode);
1605 return true;
1606 }
1607
1608 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1609 {
1610 struct f2fs_inode *src, *dst;
1611 nid_t ino = ino_of_node(page);
1612 struct node_info old_ni, new_ni;
1613 struct page *ipage;
1614
1615 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1616 if (!ipage)
1617 return -ENOMEM;
1618
1619 /* Should not use this inode from free nid list */
1620 remove_free_nid(NM_I(sbi), ino);
1621
1622 get_node_info(sbi, ino, &old_ni);
1623 SetPageUptodate(ipage);
1624 fill_node_footer(ipage, ino, ino, 0, true);
1625
1626 src = F2FS_INODE(page);
1627 dst = F2FS_INODE(ipage);
1628
1629 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1630 dst->i_size = 0;
1631 dst->i_blocks = cpu_to_le64(1);
1632 dst->i_links = cpu_to_le32(1);
1633 dst->i_xattr_nid = 0;
1634
1635 new_ni = old_ni;
1636 new_ni.ino = ino;
1637
1638 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1639 WARN_ON(1);
1640 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1641 inc_valid_inode_count(sbi);
1642 f2fs_put_page(ipage, 1);
1643 return 0;
1644 }
1645
1646 /*
1647 * ra_sum_pages() merge contiguous pages into one bio and submit.
1648 * these pre-readed pages are linked in pages list.
1649 */
1650 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1651 int start, int nrpages)
1652 {
1653 struct page *page;
1654 int page_idx = start;
1655 struct f2fs_io_info fio = {
1656 .type = META,
1657 .rw = READ_SYNC | REQ_META | REQ_PRIO
1658 };
1659
1660 for (; page_idx < start + nrpages; page_idx++) {
1661 /* alloc temporal page for read node summary info*/
1662 page = alloc_page(GFP_F2FS_ZERO);
1663 if (!page)
1664 break;
1665
1666 lock_page(page);
1667 page->index = page_idx;
1668 list_add_tail(&page->lru, pages);
1669 }
1670
1671 list_for_each_entry(page, pages, lru)
1672 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1673
1674 f2fs_submit_merged_bio(sbi, META, READ);
1675
1676 return page_idx - start;
1677 }
1678
1679 int restore_node_summary(struct f2fs_sb_info *sbi,
1680 unsigned int segno, struct f2fs_summary_block *sum)
1681 {
1682 struct f2fs_node *rn;
1683 struct f2fs_summary *sum_entry;
1684 struct page *page, *tmp;
1685 block_t addr;
1686 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1687 int i, last_offset, nrpages, err = 0;
1688 LIST_HEAD(page_list);
1689
1690 /* scan the node segment */
1691 last_offset = sbi->blocks_per_seg;
1692 addr = START_BLOCK(sbi, segno);
1693 sum_entry = &sum->entries[0];
1694
1695 for (i = 0; !err && i < last_offset; i += nrpages, addr += nrpages) {
1696 nrpages = min(last_offset - i, bio_blocks);
1697
1698 /* read ahead node pages */
1699 nrpages = ra_sum_pages(sbi, &page_list, addr, nrpages);
1700 if (!nrpages)
1701 return -ENOMEM;
1702
1703 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1704 if (err)
1705 goto skip;
1706
1707 lock_page(page);
1708 if (unlikely(!PageUptodate(page))) {
1709 err = -EIO;
1710 } else {
1711 rn = F2FS_NODE(page);
1712 sum_entry->nid = rn->footer.nid;
1713 sum_entry->version = 0;
1714 sum_entry->ofs_in_node = 0;
1715 sum_entry++;
1716 }
1717 unlock_page(page);
1718 skip:
1719 list_del(&page->lru);
1720 __free_pages(page, 0);
1721 }
1722 }
1723 return err;
1724 }
1725
1726 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1727 {
1728 struct f2fs_nm_info *nm_i = NM_I(sbi);
1729 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1730 struct f2fs_summary_block *sum = curseg->sum_blk;
1731 int i;
1732
1733 mutex_lock(&curseg->curseg_mutex);
1734
1735 if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1736 mutex_unlock(&curseg->curseg_mutex);
1737 return false;
1738 }
1739
1740 for (i = 0; i < nats_in_cursum(sum); i++) {
1741 struct nat_entry *ne;
1742 struct f2fs_nat_entry raw_ne;
1743 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1744
1745 raw_ne = nat_in_journal(sum, i);
1746 retry:
1747 write_lock(&nm_i->nat_tree_lock);
1748 ne = __lookup_nat_cache(nm_i, nid);
1749 if (ne) {
1750 __set_nat_cache_dirty(nm_i, ne);
1751 write_unlock(&nm_i->nat_tree_lock);
1752 continue;
1753 }
1754 ne = grab_nat_entry(nm_i, nid);
1755 if (!ne) {
1756 write_unlock(&nm_i->nat_tree_lock);
1757 goto retry;
1758 }
1759 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1760 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1761 nat_set_version(ne, raw_ne.version);
1762 __set_nat_cache_dirty(nm_i, ne);
1763 write_unlock(&nm_i->nat_tree_lock);
1764 }
1765 update_nats_in_cursum(sum, -i);
1766 mutex_unlock(&curseg->curseg_mutex);
1767 return true;
1768 }
1769
1770 /*
1771 * This function is called during the checkpointing process.
1772 */
1773 void flush_nat_entries(struct f2fs_sb_info *sbi)
1774 {
1775 struct f2fs_nm_info *nm_i = NM_I(sbi);
1776 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1777 struct f2fs_summary_block *sum = curseg->sum_blk;
1778 struct nat_entry *ne, *cur;
1779 struct page *page = NULL;
1780 struct f2fs_nat_block *nat_blk = NULL;
1781 nid_t start_nid = 0, end_nid = 0;
1782 bool flushed;
1783
1784 flushed = flush_nats_in_journal(sbi);
1785
1786 if (!flushed)
1787 mutex_lock(&curseg->curseg_mutex);
1788
1789 /* 1) flush dirty nat caches */
1790 list_for_each_entry_safe(ne, cur, &nm_i->dirty_nat_entries, list) {
1791 nid_t nid;
1792 struct f2fs_nat_entry raw_ne;
1793 int offset = -1;
1794 block_t new_blkaddr;
1795
1796 if (nat_get_blkaddr(ne) == NEW_ADDR)
1797 continue;
1798
1799 nid = nat_get_nid(ne);
1800
1801 if (flushed)
1802 goto to_nat_page;
1803
1804 /* if there is room for nat enries in curseg->sumpage */
1805 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1806 if (offset >= 0) {
1807 raw_ne = nat_in_journal(sum, offset);
1808 goto flush_now;
1809 }
1810 to_nat_page:
1811 if (!page || (start_nid > nid || nid > end_nid)) {
1812 if (page) {
1813 f2fs_put_page(page, 1);
1814 page = NULL;
1815 }
1816 start_nid = START_NID(nid);
1817 end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1818
1819 /*
1820 * get nat block with dirty flag, increased reference
1821 * count, mapped and lock
1822 */
1823 page = get_next_nat_page(sbi, start_nid);
1824 nat_blk = page_address(page);
1825 }
1826
1827 f2fs_bug_on(!nat_blk);
1828 raw_ne = nat_blk->entries[nid - start_nid];
1829 flush_now:
1830 new_blkaddr = nat_get_blkaddr(ne);
1831
1832 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1833 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1834 raw_ne.version = nat_get_version(ne);
1835
1836 if (offset < 0) {
1837 nat_blk->entries[nid - start_nid] = raw_ne;
1838 } else {
1839 nat_in_journal(sum, offset) = raw_ne;
1840 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1841 }
1842
1843 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1844 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1845 write_lock(&nm_i->nat_tree_lock);
1846 __del_from_nat_cache(nm_i, ne);
1847 write_unlock(&nm_i->nat_tree_lock);
1848 } else {
1849 write_lock(&nm_i->nat_tree_lock);
1850 __clear_nat_cache_dirty(nm_i, ne);
1851 write_unlock(&nm_i->nat_tree_lock);
1852 }
1853 }
1854 if (!flushed)
1855 mutex_unlock(&curseg->curseg_mutex);
1856 f2fs_put_page(page, 1);
1857 }
1858
1859 static int init_node_manager(struct f2fs_sb_info *sbi)
1860 {
1861 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1862 struct f2fs_nm_info *nm_i = NM_I(sbi);
1863 unsigned char *version_bitmap;
1864 unsigned int nat_segs, nat_blocks;
1865
1866 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1867
1868 /* segment_count_nat includes pair segment so divide to 2. */
1869 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1870 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1871
1872 /* not used nids: 0, node, meta, (and root counted as valid node) */
1873 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks - 3;
1874 nm_i->fcnt = 0;
1875 nm_i->nat_cnt = 0;
1876 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1877
1878 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1879 INIT_LIST_HEAD(&nm_i->free_nid_list);
1880 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1881 INIT_LIST_HEAD(&nm_i->nat_entries);
1882 INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1883
1884 mutex_init(&nm_i->build_lock);
1885 spin_lock_init(&nm_i->free_nid_list_lock);
1886 rwlock_init(&nm_i->nat_tree_lock);
1887
1888 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1889 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1890 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1891 if (!version_bitmap)
1892 return -EFAULT;
1893
1894 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1895 GFP_KERNEL);
1896 if (!nm_i->nat_bitmap)
1897 return -ENOMEM;
1898 return 0;
1899 }
1900
1901 int build_node_manager(struct f2fs_sb_info *sbi)
1902 {
1903 int err;
1904
1905 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1906 if (!sbi->nm_info)
1907 return -ENOMEM;
1908
1909 err = init_node_manager(sbi);
1910 if (err)
1911 return err;
1912
1913 build_free_nids(sbi);
1914 return 0;
1915 }
1916
1917 void destroy_node_manager(struct f2fs_sb_info *sbi)
1918 {
1919 struct f2fs_nm_info *nm_i = NM_I(sbi);
1920 struct free_nid *i, *next_i;
1921 struct nat_entry *natvec[NATVEC_SIZE];
1922 nid_t nid = 0;
1923 unsigned int found;
1924
1925 if (!nm_i)
1926 return;
1927
1928 /* destroy free nid list */
1929 spin_lock(&nm_i->free_nid_list_lock);
1930 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1931 f2fs_bug_on(i->state == NID_ALLOC);
1932 __del_from_free_nid_list(nm_i, i);
1933 nm_i->fcnt--;
1934 spin_unlock(&nm_i->free_nid_list_lock);
1935 kmem_cache_free(free_nid_slab, i);
1936 spin_lock(&nm_i->free_nid_list_lock);
1937 }
1938 f2fs_bug_on(nm_i->fcnt);
1939 spin_unlock(&nm_i->free_nid_list_lock);
1940
1941 /* destroy nat cache */
1942 write_lock(&nm_i->nat_tree_lock);
1943 while ((found = __gang_lookup_nat_cache(nm_i,
1944 nid, NATVEC_SIZE, natvec))) {
1945 unsigned idx;
1946 nid = nat_get_nid(natvec[found - 1]) + 1;
1947 for (idx = 0; idx < found; idx++)
1948 __del_from_nat_cache(nm_i, natvec[idx]);
1949 }
1950 f2fs_bug_on(nm_i->nat_cnt);
1951 write_unlock(&nm_i->nat_tree_lock);
1952
1953 kfree(nm_i->nat_bitmap);
1954 sbi->nm_info = NULL;
1955 kfree(nm_i);
1956 }
1957
1958 int __init create_node_manager_caches(void)
1959 {
1960 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1961 sizeof(struct nat_entry));
1962 if (!nat_entry_slab)
1963 return -ENOMEM;
1964
1965 free_nid_slab = f2fs_kmem_cache_create("free_nid",
1966 sizeof(struct free_nid));
1967 if (!free_nid_slab) {
1968 kmem_cache_destroy(nat_entry_slab);
1969 return -ENOMEM;
1970 }
1971 return 0;
1972 }
1973
1974 void destroy_node_manager_caches(void)
1975 {
1976 kmem_cache_destroy(free_nid_slab);
1977 kmem_cache_destroy(nat_entry_slab);
1978 }
Linux kernel - Deliverables
This page took 0.082117 seconds and 5 git commands to generate.