projects / GitHub / exynos8895 / android_kernel_samsung_universal8895.git / blob
? search:


0af0a715e36790a2e2cc8f58a24f40bee6f751c2
[GitHub/exynos8895/android_kernel_samsung_universal8895.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 static struct kmem_cache *nat_entry_slab;
25 static struct kmem_cache *free_nid_slab;
26
27 static void clear_node_page_dirty(struct page *page)
28 {
29 struct address_space *mapping = page->mapping;
30 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
31 unsigned int long flags;
32
33 if (PageDirty(page)) {
34 spin_lock_irqsave(&mapping->tree_lock, flags);
35 radix_tree_tag_clear(&mapping->page_tree,
36 page_index(page),
37 PAGECACHE_TAG_DIRTY);
38 spin_unlock_irqrestore(&mapping->tree_lock, flags);
39
40 clear_page_dirty_for_io(page);
41 dec_page_count(sbi, F2FS_DIRTY_NODES);
42 }
43 ClearPageUptodate(page);
44 }
45
46 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
47 {
48 pgoff_t index = current_nat_addr(sbi, nid);
49 return get_meta_page(sbi, index);
50 }
51
52 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
53 {
54 struct page *src_page;
55 struct page *dst_page;
56 pgoff_t src_off;
57 pgoff_t dst_off;
58 void *src_addr;
59 void *dst_addr;
60 struct f2fs_nm_info *nm_i = NM_I(sbi);
61
62 src_off = current_nat_addr(sbi, nid);
63 dst_off = next_nat_addr(sbi, src_off);
64
65 /* get current nat block page with lock */
66 src_page = get_meta_page(sbi, src_off);
67
68 /* Dirty src_page means that it is already the new target NAT page. */
69 if (PageDirty(src_page))
70 return src_page;
71
72 dst_page = grab_meta_page(sbi, dst_off);
73
74 src_addr = page_address(src_page);
75 dst_addr = page_address(dst_page);
76 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
77 set_page_dirty(dst_page);
78 f2fs_put_page(src_page, 1);
79
80 set_to_next_nat(nm_i, nid);
81
82 return dst_page;
83 }
84
85 /*
86 * Readahead NAT pages
87 */
88 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
89 {
90 struct address_space *mapping = sbi->meta_inode->i_mapping;
91 struct f2fs_nm_info *nm_i = NM_I(sbi);
92 struct page *page;
93 pgoff_t index;
94 int i;
95 struct f2fs_io_info fio = {
96 .type = META,
97 .rw = READ_SYNC | REQ_META | REQ_PRIO
98 };
99
100
101 for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
102 if (unlikely(nid >= nm_i->max_nid))
103 nid = 0;
104 index = current_nat_addr(sbi, nid);
105
106 page = grab_cache_page(mapping, index);
107 if (!page)
108 continue;
109 if (PageUptodate(page)) {
110 mark_page_accessed(page);
111 f2fs_put_page(page, 1);
112 continue;
113 }
114 f2fs_submit_page_mbio(sbi, page, index, &fio);
115 mark_page_accessed(page);
116 f2fs_put_page(page, 0);
117 }
118 f2fs_submit_merged_bio(sbi, META, READ);
119 }
120
121 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
122 {
123 return radix_tree_lookup(&nm_i->nat_root, n);
124 }
125
126 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
127 nid_t start, unsigned int nr, struct nat_entry **ep)
128 {
129 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
130 }
131
132 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
133 {
134 list_del(&e->list);
135 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
136 nm_i->nat_cnt--;
137 kmem_cache_free(nat_entry_slab, e);
138 }
139
140 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
141 {
142 struct f2fs_nm_info *nm_i = NM_I(sbi);
143 struct nat_entry *e;
144 int is_cp = 1;
145
146 read_lock(&nm_i->nat_tree_lock);
147 e = __lookup_nat_cache(nm_i, nid);
148 if (e && !e->checkpointed)
149 is_cp = 0;
150 read_unlock(&nm_i->nat_tree_lock);
151 return is_cp;
152 }
153
154 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
155 {
156 struct nat_entry *new;
157
158 new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
159 if (!new)
160 return NULL;
161 if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
162 kmem_cache_free(nat_entry_slab, new);
163 return NULL;
164 }
165 memset(new, 0, sizeof(struct nat_entry));
166 nat_set_nid(new, nid);
167 list_add_tail(&new->list, &nm_i->nat_entries);
168 nm_i->nat_cnt++;
169 return new;
170 }
171
172 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
173 struct f2fs_nat_entry *ne)
174 {
175 struct nat_entry *e;
176 retry:
177 write_lock(&nm_i->nat_tree_lock);
178 e = __lookup_nat_cache(nm_i, nid);
179 if (!e) {
180 e = grab_nat_entry(nm_i, nid);
181 if (!e) {
182 write_unlock(&nm_i->nat_tree_lock);
183 goto retry;
184 }
185 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
186 nat_set_ino(e, le32_to_cpu(ne->ino));
187 nat_set_version(e, ne->version);
188 e->checkpointed = true;
189 }
190 write_unlock(&nm_i->nat_tree_lock);
191 }
192
193 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
194 block_t new_blkaddr)
195 {
196 struct f2fs_nm_info *nm_i = NM_I(sbi);
197 struct nat_entry *e;
198 retry:
199 write_lock(&nm_i->nat_tree_lock);
200 e = __lookup_nat_cache(nm_i, ni->nid);
201 if (!e) {
202 e = grab_nat_entry(nm_i, ni->nid);
203 if (!e) {
204 write_unlock(&nm_i->nat_tree_lock);
205 goto retry;
206 }
207 e->ni = *ni;
208 e->checkpointed = true;
209 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
210 } else if (new_blkaddr == NEW_ADDR) {
211 /*
212 * when nid is reallocated,
213 * previous nat entry can be remained in nat cache.
214 * So, reinitialize it with new information.
215 */
216 e->ni = *ni;
217 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
218 }
219
220 if (new_blkaddr == NEW_ADDR)
221 e->checkpointed = false;
222
223 /* sanity check */
224 f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
225 f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
226 new_blkaddr == NULL_ADDR);
227 f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
228 new_blkaddr == NEW_ADDR);
229 f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
230 nat_get_blkaddr(e) != NULL_ADDR &&
231 new_blkaddr == NEW_ADDR);
232
233 /* increament version no as node is removed */
234 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
235 unsigned char version = nat_get_version(e);
236 nat_set_version(e, inc_node_version(version));
237 }
238
239 /* change address */
240 nat_set_blkaddr(e, new_blkaddr);
241 __set_nat_cache_dirty(nm_i, e);
242 write_unlock(&nm_i->nat_tree_lock);
243 }
244
245 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
246 {
247 struct f2fs_nm_info *nm_i = NM_I(sbi);
248
249 if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
250 return 0;
251
252 write_lock(&nm_i->nat_tree_lock);
253 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
254 struct nat_entry *ne;
255 ne = list_first_entry(&nm_i->nat_entries,
256 struct nat_entry, list);
257 __del_from_nat_cache(nm_i, ne);
258 nr_shrink--;
259 }
260 write_unlock(&nm_i->nat_tree_lock);
261 return nr_shrink;
262 }
263
264 /*
265 * This function returns always success
266 */
267 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
268 {
269 struct f2fs_nm_info *nm_i = NM_I(sbi);
270 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
271 struct f2fs_summary_block *sum = curseg->sum_blk;
272 nid_t start_nid = START_NID(nid);
273 struct f2fs_nat_block *nat_blk;
274 struct page *page = NULL;
275 struct f2fs_nat_entry ne;
276 struct nat_entry *e;
277 int i;
278
279 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
280 ni->nid = nid;
281
282 /* Check nat cache */
283 read_lock(&nm_i->nat_tree_lock);
284 e = __lookup_nat_cache(nm_i, nid);
285 if (e) {
286 ni->ino = nat_get_ino(e);
287 ni->blk_addr = nat_get_blkaddr(e);
288 ni->version = nat_get_version(e);
289 }
290 read_unlock(&nm_i->nat_tree_lock);
291 if (e)
292 return;
293
294 /* Check current segment summary */
295 mutex_lock(&curseg->curseg_mutex);
296 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
297 if (i >= 0) {
298 ne = nat_in_journal(sum, i);
299 node_info_from_raw_nat(ni, &ne);
300 }
301 mutex_unlock(&curseg->curseg_mutex);
302 if (i >= 0)
303 goto cache;
304
305 /* Fill node_info from nat page */
306 page = get_current_nat_page(sbi, start_nid);
307 nat_blk = (struct f2fs_nat_block *)page_address(page);
308 ne = nat_blk->entries[nid - start_nid];
309 node_info_from_raw_nat(ni, &ne);
310 f2fs_put_page(page, 1);
311 cache:
312 /* cache nat entry */
313 cache_nat_entry(NM_I(sbi), nid, &ne);
314 }
315
316 /*
317 * The maximum depth is four.
318 * Offset[0] will have raw inode offset.
319 */
320 static int get_node_path(struct f2fs_inode_info *fi, long block,
321 int offset[4], unsigned int noffset[4])
322 {
323 const long direct_index = ADDRS_PER_INODE(fi);
324 const long direct_blks = ADDRS_PER_BLOCK;
325 const long dptrs_per_blk = NIDS_PER_BLOCK;
326 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
327 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
328 int n = 0;
329 int level = 0;
330
331 noffset[0] = 0;
332
333 if (block < direct_index) {
334 offset[n] = block;
335 goto got;
336 }
337 block -= direct_index;
338 if (block < direct_blks) {
339 offset[n++] = NODE_DIR1_BLOCK;
340 noffset[n] = 1;
341 offset[n] = block;
342 level = 1;
343 goto got;
344 }
345 block -= direct_blks;
346 if (block < direct_blks) {
347 offset[n++] = NODE_DIR2_BLOCK;
348 noffset[n] = 2;
349 offset[n] = block;
350 level = 1;
351 goto got;
352 }
353 block -= direct_blks;
354 if (block < indirect_blks) {
355 offset[n++] = NODE_IND1_BLOCK;
356 noffset[n] = 3;
357 offset[n++] = block / direct_blks;
358 noffset[n] = 4 + offset[n - 1];
359 offset[n] = block % direct_blks;
360 level = 2;
361 goto got;
362 }
363 block -= indirect_blks;
364 if (block < indirect_blks) {
365 offset[n++] = NODE_IND2_BLOCK;
366 noffset[n] = 4 + dptrs_per_blk;
367 offset[n++] = block / direct_blks;
368 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
369 offset[n] = block % direct_blks;
370 level = 2;
371 goto got;
372 }
373 block -= indirect_blks;
374 if (block < dindirect_blks) {
375 offset[n++] = NODE_DIND_BLOCK;
376 noffset[n] = 5 + (dptrs_per_blk * 2);
377 offset[n++] = block / indirect_blks;
378 noffset[n] = 6 + (dptrs_per_blk * 2) +
379 offset[n - 1] * (dptrs_per_blk + 1);
380 offset[n++] = (block / direct_blks) % dptrs_per_blk;
381 noffset[n] = 7 + (dptrs_per_blk * 2) +
382 offset[n - 2] * (dptrs_per_blk + 1) +
383 offset[n - 1];
384 offset[n] = block % direct_blks;
385 level = 3;
386 goto got;
387 } else {
388 BUG();
389 }
390 got:
391 return level;
392 }
393
394 /*
395 * Caller should call f2fs_put_dnode(dn).
396 * Also, it should grab and release a mutex by calling mutex_lock_op() and
397 * mutex_unlock_op() only if ro is not set RDONLY_NODE.
398 * In the case of RDONLY_NODE, we don't need to care about mutex.
399 */
400 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
401 {
402 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
403 struct page *npage[4];
404 struct page *parent;
405 int offset[4];
406 unsigned int noffset[4];
407 nid_t nids[4];
408 int level, i;
409 int err = 0;
410
411 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
412
413 nids[0] = dn->inode->i_ino;
414 npage[0] = dn->inode_page;
415
416 if (!npage[0]) {
417 npage[0] = get_node_page(sbi, nids[0]);
418 if (IS_ERR(npage[0]))
419 return PTR_ERR(npage[0]);
420 }
421 parent = npage[0];
422 if (level != 0)
423 nids[1] = get_nid(parent, offset[0], true);
424 dn->inode_page = npage[0];
425 dn->inode_page_locked = true;
426
427 /* get indirect or direct nodes */
428 for (i = 1; i <= level; i++) {
429 bool done = false;
430
431 if (!nids[i] && mode == ALLOC_NODE) {
432 /* alloc new node */
433 if (!alloc_nid(sbi, &(nids[i]))) {
434 err = -ENOSPC;
435 goto release_pages;
436 }
437
438 dn->nid = nids[i];
439 npage[i] = new_node_page(dn, noffset[i], NULL);
440 if (IS_ERR(npage[i])) {
441 alloc_nid_failed(sbi, nids[i]);
442 err = PTR_ERR(npage[i]);
443 goto release_pages;
444 }
445
446 set_nid(parent, offset[i - 1], nids[i], i == 1);
447 alloc_nid_done(sbi, nids[i]);
448 done = true;
449 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
450 npage[i] = get_node_page_ra(parent, offset[i - 1]);
451 if (IS_ERR(npage[i])) {
452 err = PTR_ERR(npage[i]);
453 goto release_pages;
454 }
455 done = true;
456 }
457 if (i == 1) {
458 dn->inode_page_locked = false;
459 unlock_page(parent);
460 } else {
461 f2fs_put_page(parent, 1);
462 }
463
464 if (!done) {
465 npage[i] = get_node_page(sbi, nids[i]);
466 if (IS_ERR(npage[i])) {
467 err = PTR_ERR(npage[i]);
468 f2fs_put_page(npage[0], 0);
469 goto release_out;
470 }
471 }
472 if (i < level) {
473 parent = npage[i];
474 nids[i + 1] = get_nid(parent, offset[i], false);
475 }
476 }
477 dn->nid = nids[level];
478 dn->ofs_in_node = offset[level];
479 dn->node_page = npage[level];
480 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
481 return 0;
482
483 release_pages:
484 f2fs_put_page(parent, 1);
485 if (i > 1)
486 f2fs_put_page(npage[0], 0);
487 release_out:
488 dn->inode_page = NULL;
489 dn->node_page = NULL;
490 return err;
491 }
492
493 static void truncate_node(struct dnode_of_data *dn)
494 {
495 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
496 struct node_info ni;
497
498 get_node_info(sbi, dn->nid, &ni);
499 if (dn->inode->i_blocks == 0) {
500 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
501 goto invalidate;
502 }
503 f2fs_bug_on(ni.blk_addr == NULL_ADDR);
504
505 /* Deallocate node address */
506 invalidate_blocks(sbi, ni.blk_addr);
507 dec_valid_node_count(sbi, dn->inode);
508 set_node_addr(sbi, &ni, NULL_ADDR);
509
510 if (dn->nid == dn->inode->i_ino) {
511 remove_orphan_inode(sbi, dn->nid);
512 dec_valid_inode_count(sbi);
513 } else {
514 sync_inode_page(dn);
515 }
516 invalidate:
517 clear_node_page_dirty(dn->node_page);
518 F2FS_SET_SB_DIRT(sbi);
519
520 f2fs_put_page(dn->node_page, 1);
521 dn->node_page = NULL;
522 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
523 }
524
525 static int truncate_dnode(struct dnode_of_data *dn)
526 {
527 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
528 struct page *page;
529
530 if (dn->nid == 0)
531 return 1;
532
533 /* get direct node */
534 page = get_node_page(sbi, dn->nid);
535 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
536 return 1;
537 else if (IS_ERR(page))
538 return PTR_ERR(page);
539
540 /* Make dnode_of_data for parameter */
541 dn->node_page = page;
542 dn->ofs_in_node = 0;
543 truncate_data_blocks(dn);
544 truncate_node(dn);
545 return 1;
546 }
547
548 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
549 int ofs, int depth)
550 {
551 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
552 struct dnode_of_data rdn = *dn;
553 struct page *page;
554 struct f2fs_node *rn;
555 nid_t child_nid;
556 unsigned int child_nofs;
557 int freed = 0;
558 int i, ret;
559
560 if (dn->nid == 0)
561 return NIDS_PER_BLOCK + 1;
562
563 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
564
565 page = get_node_page(sbi, dn->nid);
566 if (IS_ERR(page)) {
567 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
568 return PTR_ERR(page);
569 }
570
571 rn = F2FS_NODE(page);
572 if (depth < 3) {
573 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
574 child_nid = le32_to_cpu(rn->in.nid[i]);
575 if (child_nid == 0)
576 continue;
577 rdn.nid = child_nid;
578 ret = truncate_dnode(&rdn);
579 if (ret < 0)
580 goto out_err;
581 set_nid(page, i, 0, false);
582 }
583 } else {
584 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
585 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
586 child_nid = le32_to_cpu(rn->in.nid[i]);
587 if (child_nid == 0) {
588 child_nofs += NIDS_PER_BLOCK + 1;
589 continue;
590 }
591 rdn.nid = child_nid;
592 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
593 if (ret == (NIDS_PER_BLOCK + 1)) {
594 set_nid(page, i, 0, false);
595 child_nofs += ret;
596 } else if (ret < 0 && ret != -ENOENT) {
597 goto out_err;
598 }
599 }
600 freed = child_nofs;
601 }
602
603 if (!ofs) {
604 /* remove current indirect node */
605 dn->node_page = page;
606 truncate_node(dn);
607 freed++;
608 } else {
609 f2fs_put_page(page, 1);
610 }
611 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
612 return freed;
613
614 out_err:
615 f2fs_put_page(page, 1);
616 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
617 return ret;
618 }
619
620 static int truncate_partial_nodes(struct dnode_of_data *dn,
621 struct f2fs_inode *ri, int *offset, int depth)
622 {
623 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
624 struct page *pages[2];
625 nid_t nid[3];
626 nid_t child_nid;
627 int err = 0;
628 int i;
629 int idx = depth - 2;
630
631 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
632 if (!nid[0])
633 return 0;
634
635 /* get indirect nodes in the path */
636 for (i = 0; i < depth - 1; i++) {
637 /* refernece count'll be increased */
638 pages[i] = get_node_page(sbi, nid[i]);
639 if (IS_ERR(pages[i])) {
640 depth = i + 1;
641 err = PTR_ERR(pages[i]);
642 goto fail;
643 }
644 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
645 }
646
647 /* free direct nodes linked to a partial indirect node */
648 for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
649 child_nid = get_nid(pages[idx], i, false);
650 if (!child_nid)
651 continue;
652 dn->nid = child_nid;
653 err = truncate_dnode(dn);
654 if (err < 0)
655 goto fail;
656 set_nid(pages[idx], i, 0, false);
657 }
658
659 if (offset[depth - 1] == 0) {
660 dn->node_page = pages[idx];
661 dn->nid = nid[idx];
662 truncate_node(dn);
663 } else {
664 f2fs_put_page(pages[idx], 1);
665 }
666 offset[idx]++;
667 offset[depth - 1] = 0;
668 fail:
669 for (i = depth - 3; i >= 0; i--)
670 f2fs_put_page(pages[i], 1);
671
672 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
673
674 return err;
675 }
676
677 /*
678 * All the block addresses of data and nodes should be nullified.
679 */
680 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
681 {
682 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
683 struct address_space *node_mapping = sbi->node_inode->i_mapping;
684 int err = 0, cont = 1;
685 int level, offset[4], noffset[4];
686 unsigned int nofs = 0;
687 struct f2fs_node *rn;
688 struct dnode_of_data dn;
689 struct page *page;
690
691 trace_f2fs_truncate_inode_blocks_enter(inode, from);
692
693 level = get_node_path(F2FS_I(inode), from, offset, noffset);
694 restart:
695 page = get_node_page(sbi, inode->i_ino);
696 if (IS_ERR(page)) {
697 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
698 return PTR_ERR(page);
699 }
700
701 set_new_dnode(&dn, inode, page, NULL, 0);
702 unlock_page(page);
703
704 rn = F2FS_NODE(page);
705 switch (level) {
706 case 0:
707 case 1:
708 nofs = noffset[1];
709 break;
710 case 2:
711 nofs = noffset[1];
712 if (!offset[level - 1])
713 goto skip_partial;
714 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
715 if (err < 0 && err != -ENOENT)
716 goto fail;
717 nofs += 1 + NIDS_PER_BLOCK;
718 break;
719 case 3:
720 nofs = 5 + 2 * NIDS_PER_BLOCK;
721 if (!offset[level - 1])
722 goto skip_partial;
723 err = truncate_partial_nodes(&dn, &rn->i, offset, level);
724 if (err < 0 && err != -ENOENT)
725 goto fail;
726 break;
727 default:
728 BUG();
729 }
730
731 skip_partial:
732 while (cont) {
733 dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
734 switch (offset[0]) {
735 case NODE_DIR1_BLOCK:
736 case NODE_DIR2_BLOCK:
737 err = truncate_dnode(&dn);
738 break;
739
740 case NODE_IND1_BLOCK:
741 case NODE_IND2_BLOCK:
742 err = truncate_nodes(&dn, nofs, offset[1], 2);
743 break;
744
745 case NODE_DIND_BLOCK:
746 err = truncate_nodes(&dn, nofs, offset[1], 3);
747 cont = 0;
748 break;
749
750 default:
751 BUG();
752 }
753 if (err < 0 && err != -ENOENT)
754 goto fail;
755 if (offset[1] == 0 &&
756 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
757 lock_page(page);
758 if (unlikely(page->mapping != node_mapping)) {
759 f2fs_put_page(page, 1);
760 goto restart;
761 }
762 wait_on_page_writeback(page);
763 rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
764 set_page_dirty(page);
765 unlock_page(page);
766 }
767 offset[1] = 0;
768 offset[0]++;
769 nofs += err;
770 }
771 fail:
772 f2fs_put_page(page, 0);
773 trace_f2fs_truncate_inode_blocks_exit(inode, err);
774 return err > 0 ? 0 : err;
775 }
776
777 int truncate_xattr_node(struct inode *inode, struct page *page)
778 {
779 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
780 nid_t nid = F2FS_I(inode)->i_xattr_nid;
781 struct dnode_of_data dn;
782 struct page *npage;
783
784 if (!nid)
785 return 0;
786
787 npage = get_node_page(sbi, nid);
788 if (IS_ERR(npage))
789 return PTR_ERR(npage);
790
791 F2FS_I(inode)->i_xattr_nid = 0;
792
793 /* need to do checkpoint during fsync */
794 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
795
796 set_new_dnode(&dn, inode, page, npage, nid);
797
798 if (page)
799 dn.inode_page_locked = true;
800 truncate_node(&dn);
801 return 0;
802 }
803
804 /*
805 * Caller should grab and release a mutex by calling mutex_lock_op() and
806 * mutex_unlock_op().
807 */
808 void remove_inode_page(struct inode *inode)
809 {
810 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
811 struct page *page;
812 nid_t ino = inode->i_ino;
813 struct dnode_of_data dn;
814
815 page = get_node_page(sbi, ino);
816 if (IS_ERR(page))
817 return;
818
819 if (truncate_xattr_node(inode, page)) {
820 f2fs_put_page(page, 1);
821 return;
822 }
823 /* 0 is possible, after f2fs_new_inode() is failed */
824 f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
825 set_new_dnode(&dn, inode, page, page, ino);
826 truncate_node(&dn);
827 }
828
829 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
830 {
831 struct dnode_of_data dn;
832
833 /* allocate inode page for new inode */
834 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
835
836 /* caller should f2fs_put_page(page, 1); */
837 return new_node_page(&dn, 0, NULL);
838 }
839
840 struct page *new_node_page(struct dnode_of_data *dn,
841 unsigned int ofs, struct page *ipage)
842 {
843 struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
844 struct address_space *mapping = sbi->node_inode->i_mapping;
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(mapping, dn->nid);
853 if (!page)
854 return ERR_PTR(-ENOMEM);
855
856 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
857 err = -ENOSPC;
858 goto fail;
859 }
860
861 get_node_info(sbi, dn->nid, &old_ni);
862
863 /* Reinitialize old_ni with new node page */
864 f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
865 new_ni = old_ni;
866 new_ni.ino = dn->inode->i_ino;
867 set_node_addr(sbi, &new_ni, NEW_ADDR);
868
869 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
870 set_cold_node(dn->inode, page);
871 SetPageUptodate(page);
872 set_page_dirty(page);
873
874 if (ofs == XATTR_NODE_OFFSET)
875 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
876
877 dn->node_page = page;
878 if (ipage)
879 update_inode(dn->inode, ipage);
880 else
881 sync_inode_page(dn);
882 if (ofs == 0)
883 inc_valid_inode_count(sbi);
884
885 return page;
886
887 fail:
888 clear_node_page_dirty(page);
889 f2fs_put_page(page, 1);
890 return ERR_PTR(err);
891 }
892
893 /*
894 * Caller should do after getting the following values.
895 * 0: f2fs_put_page(page, 0)
896 * LOCKED_PAGE: f2fs_put_page(page, 1)
897 * error: nothing
898 */
899 static int read_node_page(struct page *page, int rw)
900 {
901 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
902 struct node_info ni;
903
904 get_node_info(sbi, page->index, &ni);
905
906 if (unlikely(ni.blk_addr == NULL_ADDR)) {
907 f2fs_put_page(page, 1);
908 return -ENOENT;
909 }
910
911 if (PageUptodate(page))
912 return LOCKED_PAGE;
913
914 return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
915 }
916
917 /*
918 * Readahead a node page
919 */
920 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
921 {
922 struct address_space *mapping = sbi->node_inode->i_mapping;
923 struct page *apage;
924 int err;
925
926 apage = find_get_page(mapping, 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(mapping, 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 address_space *mapping = sbi->node_inode->i_mapping;
947 struct page *page;
948 int err;
949 repeat:
950 page = grab_cache_page(mapping, nid);
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))) {
962 f2fs_put_page(page, 1);
963 return ERR_PTR(-EIO);
964 }
965 if (unlikely(page->mapping != mapping)) {
966 f2fs_put_page(page, 1);
967 goto repeat;
968 }
969 got_it:
970 f2fs_bug_on(nid != nid_of_node(page));
971 mark_page_accessed(page);
972 return page;
973 }
974
975 /*
976 * Return a locked page for the desired node page.
977 * And, readahead MAX_RA_NODE number of node pages.
978 */
979 struct page *get_node_page_ra(struct page *parent, int start)
980 {
981 struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
982 struct address_space *mapping = sbi->node_inode->i_mapping;
983 struct blk_plug plug;
984 struct page *page;
985 int err, i, end;
986 nid_t nid;
987
988 /* First, try getting the desired direct node. */
989 nid = get_nid(parent, start, false);
990 if (!nid)
991 return ERR_PTR(-ENOENT);
992 repeat:
993 page = grab_cache_page(mapping, nid);
994 if (!page)
995 return ERR_PTR(-ENOMEM);
996
997 err = read_node_page(page, READ_SYNC);
998 if (err < 0)
999 return ERR_PTR(err);
1000 else if (err == LOCKED_PAGE)
1001 goto page_hit;
1002
1003 blk_start_plug(&plug);
1004
1005 /* Then, try readahead for siblings of the desired node */
1006 end = start + MAX_RA_NODE;
1007 end = min(end, NIDS_PER_BLOCK);
1008 for (i = start + 1; i < end; i++) {
1009 nid = get_nid(parent, i, false);
1010 if (!nid)
1011 continue;
1012 ra_node_page(sbi, nid);
1013 }
1014
1015 blk_finish_plug(&plug);
1016
1017 lock_page(page);
1018 if (unlikely(page->mapping != mapping)) {
1019 f2fs_put_page(page, 1);
1020 goto repeat;
1021 }
1022 page_hit:
1023 if (unlikely(!PageUptodate(page))) {
1024 f2fs_put_page(page, 1);
1025 return ERR_PTR(-EIO);
1026 }
1027 mark_page_accessed(page);
1028 return page;
1029 }
1030
1031 void sync_inode_page(struct dnode_of_data *dn)
1032 {
1033 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1034 update_inode(dn->inode, dn->node_page);
1035 } else if (dn->inode_page) {
1036 if (!dn->inode_page_locked)
1037 lock_page(dn->inode_page);
1038 update_inode(dn->inode, dn->inode_page);
1039 if (!dn->inode_page_locked)
1040 unlock_page(dn->inode_page);
1041 } else {
1042 update_inode_page(dn->inode);
1043 }
1044 }
1045
1046 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1047 struct writeback_control *wbc)
1048 {
1049 struct address_space *mapping = sbi->node_inode->i_mapping;
1050 pgoff_t index, end;
1051 struct pagevec pvec;
1052 int step = ino ? 2 : 0;
1053 int nwritten = 0, wrote = 0;
1054
1055 pagevec_init(&pvec, 0);
1056
1057 next_step:
1058 index = 0;
1059 end = LONG_MAX;
1060
1061 while (index <= end) {
1062 int i, nr_pages;
1063 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1064 PAGECACHE_TAG_DIRTY,
1065 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1066 if (nr_pages == 0)
1067 break;
1068
1069 for (i = 0; i < nr_pages; i++) {
1070 struct page *page = pvec.pages[i];
1071
1072 /*
1073 * flushing sequence with step:
1074 * 0. indirect nodes
1075 * 1. dentry dnodes
1076 * 2. file dnodes
1077 */
1078 if (step == 0 && IS_DNODE(page))
1079 continue;
1080 if (step == 1 && (!IS_DNODE(page) ||
1081 is_cold_node(page)))
1082 continue;
1083 if (step == 2 && (!IS_DNODE(page) ||
1084 !is_cold_node(page)))
1085 continue;
1086
1087 /*
1088 * If an fsync mode,
1089 * we should not skip writing node pages.
1090 */
1091 if (ino && ino_of_node(page) == ino)
1092 lock_page(page);
1093 else if (!trylock_page(page))
1094 continue;
1095
1096 if (unlikely(page->mapping != mapping)) {
1097 continue_unlock:
1098 unlock_page(page);
1099 continue;
1100 }
1101 if (ino && ino_of_node(page) != ino)
1102 goto continue_unlock;
1103
1104 if (!PageDirty(page)) {
1105 /* someone wrote it for us */
1106 goto continue_unlock;
1107 }
1108
1109 if (!clear_page_dirty_for_io(page))
1110 goto continue_unlock;
1111
1112 /* called by fsync() */
1113 if (ino && IS_DNODE(page)) {
1114 int mark = !is_checkpointed_node(sbi, ino);
1115 set_fsync_mark(page, 1);
1116 if (IS_INODE(page))
1117 set_dentry_mark(page, mark);
1118 nwritten++;
1119 } else {
1120 set_fsync_mark(page, 0);
1121 set_dentry_mark(page, 0);
1122 }
1123 mapping->a_ops->writepage(page, wbc);
1124 wrote++;
1125
1126 if (--wbc->nr_to_write == 0)
1127 break;
1128 }
1129 pagevec_release(&pvec);
1130 cond_resched();
1131
1132 if (wbc->nr_to_write == 0) {
1133 step = 2;
1134 break;
1135 }
1136 }
1137
1138 if (step < 2) {
1139 step++;
1140 goto next_step;
1141 }
1142
1143 if (wrote)
1144 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1145 return nwritten;
1146 }
1147
1148 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1149 {
1150 struct address_space *mapping = sbi->node_inode->i_mapping;
1151 pgoff_t index = 0, end = LONG_MAX;
1152 struct pagevec pvec;
1153 int nr_pages;
1154 int ret2 = 0, ret = 0;
1155
1156 pagevec_init(&pvec, 0);
1157 while ((index <= end) &&
1158 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1159 PAGECACHE_TAG_WRITEBACK,
1160 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
1161 unsigned i;
1162
1163 for (i = 0; i < nr_pages; i++) {
1164 struct page *page = pvec.pages[i];
1165
1166 /* until radix tree lookup accepts end_index */
1167 if (unlikely(page->index > end))
1168 continue;
1169
1170 if (ino && ino_of_node(page) == ino) {
1171 wait_on_page_writeback(page);
1172 if (TestClearPageError(page))
1173 ret = -EIO;
1174 }
1175 }
1176 pagevec_release(&pvec);
1177 cond_resched();
1178 }
1179
1180 if (unlikely(test_and_clear_bit(AS_ENOSPC, &mapping->flags)))
1181 ret2 = -ENOSPC;
1182 if (unlikely(test_and_clear_bit(AS_EIO, &mapping->flags)))
1183 ret2 = -EIO;
1184 if (!ret)
1185 ret = ret2;
1186 return ret;
1187 }
1188
1189 static int f2fs_write_node_page(struct page *page,
1190 struct writeback_control *wbc)
1191 {
1192 struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1193 nid_t nid;
1194 block_t new_addr;
1195 struct node_info ni;
1196
1197 if (unlikely(sbi->por_doing))
1198 goto redirty_out;
1199
1200 wait_on_page_writeback(page);
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, nid, ni.blk_addr, &new_addr);
1221 set_node_addr(sbi, &ni, new_addr);
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 set_page_dirty(page);
1231 return AOP_WRITEPAGE_ACTIVATE;
1232 }
1233
1234 /*
1235 * It is very important to gather dirty pages and write at once, so that we can
1236 * submit a big bio without interfering other data writes.
1237 * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1238 */
1239 #define COLLECT_DIRTY_NODES 1536
1240 static int f2fs_write_node_pages(struct address_space *mapping,
1241 struct writeback_control *wbc)
1242 {
1243 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1244 long nr_to_write = wbc->nr_to_write;
1245
1246 /* balancing f2fs's metadata in background */
1247 f2fs_balance_fs_bg(sbi);
1248
1249 /* collect a number of dirty node pages and write together */
1250 if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1251 return 0;
1252
1253 /* if mounting is failed, skip writing node pages */
1254 wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1255 sync_node_pages(sbi, 0, wbc);
1256 wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1257 wbc->nr_to_write);
1258 return 0;
1259 }
1260
1261 static int f2fs_set_node_page_dirty(struct page *page)
1262 {
1263 struct address_space *mapping = page->mapping;
1264 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1265
1266 trace_f2fs_set_page_dirty(page, NODE);
1267
1268 SetPageUptodate(page);
1269 if (!PageDirty(page)) {
1270 __set_page_dirty_nobuffers(page);
1271 inc_page_count(sbi, F2FS_DIRTY_NODES);
1272 SetPagePrivate(page);
1273 return 1;
1274 }
1275 return 0;
1276 }
1277
1278 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1279 unsigned int length)
1280 {
1281 struct inode *inode = page->mapping->host;
1282 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1283 if (PageDirty(page))
1284 dec_page_count(sbi, F2FS_DIRTY_NODES);
1285 ClearPagePrivate(page);
1286 }
1287
1288 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1289 {
1290 ClearPagePrivate(page);
1291 return 1;
1292 }
1293
1294 /*
1295 * Structure of the f2fs node operations
1296 */
1297 const struct address_space_operations f2fs_node_aops = {
1298 .writepage = f2fs_write_node_page,
1299 .writepages = f2fs_write_node_pages,
1300 .set_page_dirty = f2fs_set_node_page_dirty,
1301 .invalidatepage = f2fs_invalidate_node_page,
1302 .releasepage = f2fs_release_node_page,
1303 };
1304
1305 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1306 {
1307 struct list_head *this;
1308 struct free_nid *i;
1309 list_for_each(this, head) {
1310 i = list_entry(this, struct free_nid, list);
1311 if (i->nid == n)
1312 return i;
1313 }
1314 return NULL;
1315 }
1316
1317 static void __del_from_free_nid_list(struct free_nid *i)
1318 {
1319 list_del(&i->list);
1320 kmem_cache_free(free_nid_slab, i);
1321 }
1322
1323 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1324 {
1325 struct free_nid *i;
1326 struct nat_entry *ne;
1327 bool allocated = false;
1328
1329 if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1330 return -1;
1331
1332 /* 0 nid should not be used */
1333 if (unlikely(nid == 0))
1334 return 0;
1335
1336 if (build) {
1337 /* do not add allocated nids */
1338 read_lock(&nm_i->nat_tree_lock);
1339 ne = __lookup_nat_cache(nm_i, nid);
1340 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1341 allocated = true;
1342 read_unlock(&nm_i->nat_tree_lock);
1343 if (allocated)
1344 return 0;
1345 }
1346
1347 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1348 i->nid = nid;
1349 i->state = NID_NEW;
1350
1351 spin_lock(&nm_i->free_nid_list_lock);
1352 if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1353 spin_unlock(&nm_i->free_nid_list_lock);
1354 kmem_cache_free(free_nid_slab, i);
1355 return 0;
1356 }
1357 list_add_tail(&i->list, &nm_i->free_nid_list);
1358 nm_i->fcnt++;
1359 spin_unlock(&nm_i->free_nid_list_lock);
1360 return 1;
1361 }
1362
1363 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1364 {
1365 struct free_nid *i;
1366 spin_lock(&nm_i->free_nid_list_lock);
1367 i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1368 if (i && i->state == NID_NEW) {
1369 __del_from_free_nid_list(i);
1370 nm_i->fcnt--;
1371 }
1372 spin_unlock(&nm_i->free_nid_list_lock);
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_nat_pages(sbi, nid);
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 struct list_head *this;
1453 retry:
1454 if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1455 return false;
1456
1457 spin_lock(&nm_i->free_nid_list_lock);
1458
1459 /* We should not use stale free nids created by build_free_nids */
1460 if (nm_i->fcnt && !sbi->on_build_free_nids) {
1461 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1462 list_for_each(this, &nm_i->free_nid_list) {
1463 i = list_entry(this, struct free_nid, list);
1464 if (i->state == NID_NEW)
1465 break;
1466 }
1467
1468 f2fs_bug_on(i->state != NID_NEW);
1469 *nid = i->nid;
1470 i->state = NID_ALLOC;
1471 nm_i->fcnt--;
1472 spin_unlock(&nm_i->free_nid_list_lock);
1473 return true;
1474 }
1475 spin_unlock(&nm_i->free_nid_list_lock);
1476
1477 /* Let's scan nat pages and its caches to get free nids */
1478 mutex_lock(&nm_i->build_lock);
1479 sbi->on_build_free_nids = true;
1480 build_free_nids(sbi);
1481 sbi->on_build_free_nids = false;
1482 mutex_unlock(&nm_i->build_lock);
1483 goto retry;
1484 }
1485
1486 /*
1487 * alloc_nid() should be called prior to this function.
1488 */
1489 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1490 {
1491 struct f2fs_nm_info *nm_i = NM_I(sbi);
1492 struct free_nid *i;
1493
1494 spin_lock(&nm_i->free_nid_list_lock);
1495 i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1496 f2fs_bug_on(!i || i->state != NID_ALLOC);
1497 __del_from_free_nid_list(i);
1498 spin_unlock(&nm_i->free_nid_list_lock);
1499 }
1500
1501 /*
1502 * alloc_nid() should be called prior to this function.
1503 */
1504 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1505 {
1506 struct f2fs_nm_info *nm_i = NM_I(sbi);
1507 struct free_nid *i;
1508
1509 if (!nid)
1510 return;
1511
1512 spin_lock(&nm_i->free_nid_list_lock);
1513 i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1514 f2fs_bug_on(!i || i->state != NID_ALLOC);
1515 if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1516 __del_from_free_nid_list(i);
1517 } else {
1518 i->state = NID_NEW;
1519 nm_i->fcnt++;
1520 }
1521 spin_unlock(&nm_i->free_nid_list_lock);
1522 }
1523
1524 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1525 struct f2fs_summary *sum, struct node_info *ni,
1526 block_t new_blkaddr)
1527 {
1528 rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1529 set_node_addr(sbi, ni, new_blkaddr);
1530 clear_node_page_dirty(page);
1531 }
1532
1533 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1534 {
1535 struct address_space *mapping = sbi->node_inode->i_mapping;
1536 struct f2fs_node *src, *dst;
1537 nid_t ino = ino_of_node(page);
1538 struct node_info old_ni, new_ni;
1539 struct page *ipage;
1540
1541 ipage = grab_cache_page(mapping, ino);
1542 if (!ipage)
1543 return -ENOMEM;
1544
1545 /* Should not use this inode from free nid list */
1546 remove_free_nid(NM_I(sbi), ino);
1547
1548 get_node_info(sbi, ino, &old_ni);
1549 SetPageUptodate(ipage);
1550 fill_node_footer(ipage, ino, ino, 0, true);
1551
1552 src = F2FS_NODE(page);
1553 dst = F2FS_NODE(ipage);
1554
1555 memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1556 dst->i.i_size = 0;
1557 dst->i.i_blocks = cpu_to_le64(1);
1558 dst->i.i_links = cpu_to_le32(1);
1559 dst->i.i_xattr_nid = 0;
1560
1561 new_ni = old_ni;
1562 new_ni.ino = ino;
1563
1564 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1565 WARN_ON(1);
1566 set_node_addr(sbi, &new_ni, NEW_ADDR);
1567 inc_valid_inode_count(sbi);
1568 f2fs_put_page(ipage, 1);
1569 return 0;
1570 }
1571
1572 /*
1573 * ra_sum_pages() merge contiguous pages into one bio and submit.
1574 * these pre-readed pages are linked in pages list.
1575 */
1576 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1577 int start, int nrpages)
1578 {
1579 struct page *page;
1580 int page_idx = start;
1581 struct f2fs_io_info fio = {
1582 .type = META,
1583 .rw = READ_SYNC | REQ_META | REQ_PRIO
1584 };
1585
1586 for (; page_idx < start + nrpages; page_idx++) {
1587 /* alloc temporal page for read node summary info*/
1588 page = alloc_page(GFP_F2FS_ZERO);
1589 if (!page) {
1590 struct page *tmp;
1591 list_for_each_entry_safe(page, tmp, pages, lru) {
1592 list_del(&page->lru);
1593 unlock_page(page);
1594 __free_pages(page, 0);
1595 }
1596 return -ENOMEM;
1597 }
1598
1599 lock_page(page);
1600 page->index = page_idx;
1601 list_add_tail(&page->lru, pages);
1602 }
1603
1604 list_for_each_entry(page, pages, lru)
1605 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1606
1607 f2fs_submit_merged_bio(sbi, META, READ);
1608 return 0;
1609 }
1610
1611 int restore_node_summary(struct f2fs_sb_info *sbi,
1612 unsigned int segno, struct f2fs_summary_block *sum)
1613 {
1614 struct f2fs_node *rn;
1615 struct f2fs_summary *sum_entry;
1616 struct page *page, *tmp;
1617 block_t addr;
1618 int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1619 int i, last_offset, nrpages, err = 0;
1620 LIST_HEAD(page_list);
1621
1622 /* scan the node segment */
1623 last_offset = sbi->blocks_per_seg;
1624 addr = START_BLOCK(sbi, segno);
1625 sum_entry = &sum->entries[0];
1626
1627 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1628 nrpages = min(last_offset - i, bio_blocks);
1629
1630 /* read ahead node pages */
1631 err = ra_sum_pages(sbi, &page_list, addr, nrpages);
1632 if (err)
1633 return err;
1634
1635 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1636
1637 lock_page(page);
1638 if (unlikely(!PageUptodate(page))) {
1639 err = -EIO;
1640 } else {
1641 rn = F2FS_NODE(page);
1642 sum_entry->nid = rn->footer.nid;
1643 sum_entry->version = 0;
1644 sum_entry->ofs_in_node = 0;
1645 sum_entry++;
1646 }
1647
1648 list_del(&page->lru);
1649 unlock_page(page);
1650 __free_pages(page, 0);
1651 }
1652 }
1653 return err;
1654 }
1655
1656 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1657 {
1658 struct f2fs_nm_info *nm_i = NM_I(sbi);
1659 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1660 struct f2fs_summary_block *sum = curseg->sum_blk;
1661 int i;
1662
1663 mutex_lock(&curseg->curseg_mutex);
1664
1665 if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1666 mutex_unlock(&curseg->curseg_mutex);
1667 return false;
1668 }
1669
1670 for (i = 0; i < nats_in_cursum(sum); i++) {
1671 struct nat_entry *ne;
1672 struct f2fs_nat_entry raw_ne;
1673 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1674
1675 raw_ne = nat_in_journal(sum, i);
1676 retry:
1677 write_lock(&nm_i->nat_tree_lock);
1678 ne = __lookup_nat_cache(nm_i, nid);
1679 if (ne) {
1680 __set_nat_cache_dirty(nm_i, ne);
1681 write_unlock(&nm_i->nat_tree_lock);
1682 continue;
1683 }
1684 ne = grab_nat_entry(nm_i, nid);
1685 if (!ne) {
1686 write_unlock(&nm_i->nat_tree_lock);
1687 goto retry;
1688 }
1689 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1690 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1691 nat_set_version(ne, raw_ne.version);
1692 __set_nat_cache_dirty(nm_i, ne);
1693 write_unlock(&nm_i->nat_tree_lock);
1694 }
1695 update_nats_in_cursum(sum, -i);
1696 mutex_unlock(&curseg->curseg_mutex);
1697 return true;
1698 }
1699
1700 /*
1701 * This function is called during the checkpointing process.
1702 */
1703 void flush_nat_entries(struct f2fs_sb_info *sbi)
1704 {
1705 struct f2fs_nm_info *nm_i = NM_I(sbi);
1706 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1707 struct f2fs_summary_block *sum = curseg->sum_blk;
1708 struct list_head *cur, *n;
1709 struct page *page = NULL;
1710 struct f2fs_nat_block *nat_blk = NULL;
1711 nid_t start_nid = 0, end_nid = 0;
1712 bool flushed;
1713
1714 flushed = flush_nats_in_journal(sbi);
1715
1716 if (!flushed)
1717 mutex_lock(&curseg->curseg_mutex);
1718
1719 /* 1) flush dirty nat caches */
1720 list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1721 struct nat_entry *ne;
1722 nid_t nid;
1723 struct f2fs_nat_entry raw_ne;
1724 int offset = -1;
1725 block_t new_blkaddr;
1726
1727 ne = list_entry(cur, struct nat_entry, list);
1728 nid = nat_get_nid(ne);
1729
1730 if (nat_get_blkaddr(ne) == NEW_ADDR)
1731 continue;
1732 if (flushed)
1733 goto to_nat_page;
1734
1735 /* if there is room for nat enries in curseg->sumpage */
1736 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1737 if (offset >= 0) {
1738 raw_ne = nat_in_journal(sum, offset);
1739 goto flush_now;
1740 }
1741 to_nat_page:
1742 if (!page || (start_nid > nid || nid > end_nid)) {
1743 if (page) {
1744 f2fs_put_page(page, 1);
1745 page = NULL;
1746 }
1747 start_nid = START_NID(nid);
1748 end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1749
1750 /*
1751 * get nat block with dirty flag, increased reference
1752 * count, mapped and lock
1753 */
1754 page = get_next_nat_page(sbi, start_nid);
1755 nat_blk = page_address(page);
1756 }
1757
1758 f2fs_bug_on(!nat_blk);
1759 raw_ne = nat_blk->entries[nid - start_nid];
1760 flush_now:
1761 new_blkaddr = nat_get_blkaddr(ne);
1762
1763 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1764 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1765 raw_ne.version = nat_get_version(ne);
1766
1767 if (offset < 0) {
1768 nat_blk->entries[nid - start_nid] = raw_ne;
1769 } else {
1770 nat_in_journal(sum, offset) = raw_ne;
1771 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1772 }
1773
1774 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1775 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1776 write_lock(&nm_i->nat_tree_lock);
1777 __del_from_nat_cache(nm_i, ne);
1778 write_unlock(&nm_i->nat_tree_lock);
1779 } else {
1780 write_lock(&nm_i->nat_tree_lock);
1781 __clear_nat_cache_dirty(nm_i, ne);
1782 ne->checkpointed = true;
1783 write_unlock(&nm_i->nat_tree_lock);
1784 }
1785 }
1786 if (!flushed)
1787 mutex_unlock(&curseg->curseg_mutex);
1788 f2fs_put_page(page, 1);
1789
1790 /* 2) shrink nat caches if necessary */
1791 try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1792 }
1793
1794 static int init_node_manager(struct f2fs_sb_info *sbi)
1795 {
1796 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1797 struct f2fs_nm_info *nm_i = NM_I(sbi);
1798 unsigned char *version_bitmap;
1799 unsigned int nat_segs, nat_blocks;
1800
1801 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1802
1803 /* segment_count_nat includes pair segment so divide to 2. */
1804 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1805 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1806 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1807 nm_i->fcnt = 0;
1808 nm_i->nat_cnt = 0;
1809
1810 INIT_LIST_HEAD(&nm_i->free_nid_list);
1811 INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1812 INIT_LIST_HEAD(&nm_i->nat_entries);
1813 INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1814
1815 mutex_init(&nm_i->build_lock);
1816 spin_lock_init(&nm_i->free_nid_list_lock);
1817 rwlock_init(&nm_i->nat_tree_lock);
1818
1819 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1820 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1821 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1822 if (!version_bitmap)
1823 return -EFAULT;
1824
1825 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1826 GFP_KERNEL);
1827 if (!nm_i->nat_bitmap)
1828 return -ENOMEM;
1829 return 0;
1830 }
1831
1832 int build_node_manager(struct f2fs_sb_info *sbi)
1833 {
1834 int err;
1835
1836 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1837 if (!sbi->nm_info)
1838 return -ENOMEM;
1839
1840 err = init_node_manager(sbi);
1841 if (err)
1842 return err;
1843
1844 build_free_nids(sbi);
1845 return 0;
1846 }
1847
1848 void destroy_node_manager(struct f2fs_sb_info *sbi)
1849 {
1850 struct f2fs_nm_info *nm_i = NM_I(sbi);
1851 struct free_nid *i, *next_i;
1852 struct nat_entry *natvec[NATVEC_SIZE];
1853 nid_t nid = 0;
1854 unsigned int found;
1855
1856 if (!nm_i)
1857 return;
1858
1859 /* destroy free nid list */
1860 spin_lock(&nm_i->free_nid_list_lock);
1861 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1862 f2fs_bug_on(i->state == NID_ALLOC);
1863 __del_from_free_nid_list(i);
1864 nm_i->fcnt--;
1865 }
1866 f2fs_bug_on(nm_i->fcnt);
1867 spin_unlock(&nm_i->free_nid_list_lock);
1868
1869 /* destroy nat cache */
1870 write_lock(&nm_i->nat_tree_lock);
1871 while ((found = __gang_lookup_nat_cache(nm_i,
1872 nid, NATVEC_SIZE, natvec))) {
1873 unsigned idx;
1874 for (idx = 0; idx < found; idx++) {
1875 struct nat_entry *e = natvec[idx];
1876 nid = nat_get_nid(e) + 1;
1877 __del_from_nat_cache(nm_i, e);
1878 }
1879 }
1880 f2fs_bug_on(nm_i->nat_cnt);
1881 write_unlock(&nm_i->nat_tree_lock);
1882
1883 kfree(nm_i->nat_bitmap);
1884 sbi->nm_info = NULL;
1885 kfree(nm_i);
1886 }
1887
1888 int __init create_node_manager_caches(void)
1889 {
1890 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1891 sizeof(struct nat_entry), NULL);
1892 if (!nat_entry_slab)
1893 return -ENOMEM;
1894
1895 free_nid_slab = f2fs_kmem_cache_create("free_nid",
1896 sizeof(struct free_nid), NULL);
1897 if (!free_nid_slab) {
1898 kmem_cache_destroy(nat_entry_slab);
1899 return -ENOMEM;
1900 }
1901 return 0;
1902 }
1903
1904 void destroy_node_manager_caches(void)
1905 {
1906 kmem_cache_destroy(free_nid_slab);
1907 kmem_cache_destroy(nat_entry_slab);
1908 }