4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
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.
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>
23 #include <trace/events/f2fs.h>
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
33 struct f2fs_nm_info *nm_i = NM_I(sbi);
35 unsigned long avail_ram;
36 unsigned long mem_size = 0;
41 /* only uses low memory */
42 avail_ram = val.totalram - val.totalhigh;
45 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
47 if (type == FREE_NIDS) {
48 mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
50 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 } else if (type == NAT_ENTRIES) {
52 mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
54 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 } else if (type == DIRTY_DENTS) {
56 if (sbi->sb->s_bdi->wb.dirty_exceeded)
58 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
59 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
60 } else if (type == INO_ENTRIES) {
63 for (i = 0; i <= UPDATE_INO; i++)
64 mem_size += (sbi->im[i].ino_num *
65 sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
67 } else if (type == EXTENT_CACHE) {
68 mem_size = (sbi->total_ext_tree * sizeof(struct extent_tree) +
69 atomic_read(&sbi->total_ext_node) *
70 sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
71 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
73 if (sbi->sb->s_bdi->wb.dirty_exceeded)
79 static void clear_node_page_dirty(struct page *page)
81 struct address_space *mapping = page->mapping;
82 unsigned int long flags;
84 if (PageDirty(page)) {
85 spin_lock_irqsave(&mapping->tree_lock, flags);
86 radix_tree_tag_clear(&mapping->page_tree,
89 spin_unlock_irqrestore(&mapping->tree_lock, flags);
91 clear_page_dirty_for_io(page);
92 dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
94 ClearPageUptodate(page);
97 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
99 pgoff_t index = current_nat_addr(sbi, nid);
100 return get_meta_page(sbi, index);
103 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
105 struct page *src_page;
106 struct page *dst_page;
111 struct f2fs_nm_info *nm_i = NM_I(sbi);
113 src_off = current_nat_addr(sbi, nid);
114 dst_off = next_nat_addr(sbi, src_off);
116 /* get current nat block page with lock */
117 src_page = get_meta_page(sbi, src_off);
118 dst_page = grab_meta_page(sbi, dst_off);
119 f2fs_bug_on(sbi, PageDirty(src_page));
121 src_addr = page_address(src_page);
122 dst_addr = page_address(dst_page);
123 memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
124 set_page_dirty(dst_page);
125 f2fs_put_page(src_page, 1);
127 set_to_next_nat(nm_i, nid);
132 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
134 return radix_tree_lookup(&nm_i->nat_root, n);
137 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
138 nid_t start, unsigned int nr, struct nat_entry **ep)
140 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
143 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
146 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
148 kmem_cache_free(nat_entry_slab, e);
151 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
152 struct nat_entry *ne)
154 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
155 struct nat_entry_set *head;
157 if (get_nat_flag(ne, IS_DIRTY))
160 head = radix_tree_lookup(&nm_i->nat_set_root, set);
162 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
164 INIT_LIST_HEAD(&head->entry_list);
165 INIT_LIST_HEAD(&head->set_list);
168 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
170 list_move_tail(&ne->list, &head->entry_list);
171 nm_i->dirty_nat_cnt++;
173 set_nat_flag(ne, IS_DIRTY, true);
176 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
177 struct nat_entry *ne)
179 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
180 struct nat_entry_set *head;
182 head = radix_tree_lookup(&nm_i->nat_set_root, set);
184 list_move_tail(&ne->list, &nm_i->nat_entries);
185 set_nat_flag(ne, IS_DIRTY, false);
187 nm_i->dirty_nat_cnt--;
191 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
192 nid_t start, unsigned int nr, struct nat_entry_set **ep)
194 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
198 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
200 struct f2fs_nm_info *nm_i = NM_I(sbi);
204 down_read(&nm_i->nat_tree_lock);
205 e = __lookup_nat_cache(nm_i, nid);
207 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
208 !get_nat_flag(e, HAS_FSYNCED_INODE))
211 up_read(&nm_i->nat_tree_lock);
215 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
217 struct f2fs_nm_info *nm_i = NM_I(sbi);
221 down_read(&nm_i->nat_tree_lock);
222 e = __lookup_nat_cache(nm_i, nid);
223 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
225 up_read(&nm_i->nat_tree_lock);
229 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
231 struct f2fs_nm_info *nm_i = NM_I(sbi);
233 bool need_update = true;
235 down_read(&nm_i->nat_tree_lock);
236 e = __lookup_nat_cache(nm_i, ino);
237 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
238 (get_nat_flag(e, IS_CHECKPOINTED) ||
239 get_nat_flag(e, HAS_FSYNCED_INODE)))
241 up_read(&nm_i->nat_tree_lock);
245 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
247 struct nat_entry *new;
249 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
250 f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
251 memset(new, 0, sizeof(struct nat_entry));
252 nat_set_nid(new, nid);
254 list_add_tail(&new->list, &nm_i->nat_entries);
259 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
260 struct f2fs_nat_entry *ne)
264 down_write(&nm_i->nat_tree_lock);
265 e = __lookup_nat_cache(nm_i, nid);
267 e = grab_nat_entry(nm_i, nid);
268 node_info_from_raw_nat(&e->ni, ne);
270 up_write(&nm_i->nat_tree_lock);
273 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
274 block_t new_blkaddr, bool fsync_done)
276 struct f2fs_nm_info *nm_i = NM_I(sbi);
279 down_write(&nm_i->nat_tree_lock);
280 e = __lookup_nat_cache(nm_i, ni->nid);
282 e = grab_nat_entry(nm_i, ni->nid);
283 copy_node_info(&e->ni, ni);
284 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
285 } else if (new_blkaddr == NEW_ADDR) {
287 * when nid is reallocated,
288 * previous nat entry can be remained in nat cache.
289 * So, reinitialize it with new information.
291 copy_node_info(&e->ni, ni);
292 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
296 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
297 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
298 new_blkaddr == NULL_ADDR);
299 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
300 new_blkaddr == NEW_ADDR);
301 f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
302 nat_get_blkaddr(e) != NULL_ADDR &&
303 new_blkaddr == NEW_ADDR);
305 /* increment version no as node is removed */
306 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
307 unsigned char version = nat_get_version(e);
308 nat_set_version(e, inc_node_version(version));
310 /* in order to reuse the nid */
311 if (nm_i->next_scan_nid > ni->nid)
312 nm_i->next_scan_nid = ni->nid;
316 nat_set_blkaddr(e, new_blkaddr);
317 if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
318 set_nat_flag(e, IS_CHECKPOINTED, false);
319 __set_nat_cache_dirty(nm_i, e);
321 /* update fsync_mark if its inode nat entry is still alive */
322 if (ni->nid != ni->ino)
323 e = __lookup_nat_cache(nm_i, ni->ino);
325 if (fsync_done && ni->nid == ni->ino)
326 set_nat_flag(e, HAS_FSYNCED_INODE, true);
327 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
329 up_write(&nm_i->nat_tree_lock);
332 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
334 struct f2fs_nm_info *nm_i = NM_I(sbi);
337 if (!down_write_trylock(&nm_i->nat_tree_lock))
340 while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
341 struct nat_entry *ne;
342 ne = list_first_entry(&nm_i->nat_entries,
343 struct nat_entry, list);
344 __del_from_nat_cache(nm_i, ne);
347 up_write(&nm_i->nat_tree_lock);
348 return nr - nr_shrink;
352 * This function always returns success
354 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
356 struct f2fs_nm_info *nm_i = NM_I(sbi);
357 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
358 struct f2fs_summary_block *sum = curseg->sum_blk;
359 nid_t start_nid = START_NID(nid);
360 struct f2fs_nat_block *nat_blk;
361 struct page *page = NULL;
362 struct f2fs_nat_entry ne;
368 /* Check nat cache */
369 down_read(&nm_i->nat_tree_lock);
370 e = __lookup_nat_cache(nm_i, nid);
372 ni->ino = nat_get_ino(e);
373 ni->blk_addr = nat_get_blkaddr(e);
374 ni->version = nat_get_version(e);
376 up_read(&nm_i->nat_tree_lock);
380 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
382 /* Check current segment summary */
383 mutex_lock(&curseg->curseg_mutex);
384 i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
386 ne = nat_in_journal(sum, i);
387 node_info_from_raw_nat(ni, &ne);
389 mutex_unlock(&curseg->curseg_mutex);
393 /* Fill node_info from nat page */
394 page = get_current_nat_page(sbi, start_nid);
395 nat_blk = (struct f2fs_nat_block *)page_address(page);
396 ne = nat_blk->entries[nid - start_nid];
397 node_info_from_raw_nat(ni, &ne);
398 f2fs_put_page(page, 1);
400 /* cache nat entry */
401 cache_nat_entry(NM_I(sbi), nid, &ne);
405 * The maximum depth is four.
406 * Offset[0] will have raw inode offset.
408 static int get_node_path(struct f2fs_inode_info *fi, long block,
409 int offset[4], unsigned int noffset[4])
411 const long direct_index = ADDRS_PER_INODE(fi);
412 const long direct_blks = ADDRS_PER_BLOCK;
413 const long dptrs_per_blk = NIDS_PER_BLOCK;
414 const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
415 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
421 if (block < direct_index) {
425 block -= direct_index;
426 if (block < direct_blks) {
427 offset[n++] = NODE_DIR1_BLOCK;
433 block -= direct_blks;
434 if (block < direct_blks) {
435 offset[n++] = NODE_DIR2_BLOCK;
441 block -= direct_blks;
442 if (block < indirect_blks) {
443 offset[n++] = NODE_IND1_BLOCK;
445 offset[n++] = block / direct_blks;
446 noffset[n] = 4 + offset[n - 1];
447 offset[n] = block % direct_blks;
451 block -= indirect_blks;
452 if (block < indirect_blks) {
453 offset[n++] = NODE_IND2_BLOCK;
454 noffset[n] = 4 + dptrs_per_blk;
455 offset[n++] = block / direct_blks;
456 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
457 offset[n] = block % direct_blks;
461 block -= indirect_blks;
462 if (block < dindirect_blks) {
463 offset[n++] = NODE_DIND_BLOCK;
464 noffset[n] = 5 + (dptrs_per_blk * 2);
465 offset[n++] = block / indirect_blks;
466 noffset[n] = 6 + (dptrs_per_blk * 2) +
467 offset[n - 1] * (dptrs_per_blk + 1);
468 offset[n++] = (block / direct_blks) % dptrs_per_blk;
469 noffset[n] = 7 + (dptrs_per_blk * 2) +
470 offset[n - 2] * (dptrs_per_blk + 1) +
472 offset[n] = block % direct_blks;
483 * Caller should call f2fs_put_dnode(dn).
484 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
485 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
486 * In the case of RDONLY_NODE, we don't need to care about mutex.
488 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
490 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
491 struct page *npage[4];
492 struct page *parent = NULL;
494 unsigned int noffset[4];
499 level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
501 nids[0] = dn->inode->i_ino;
502 npage[0] = dn->inode_page;
505 npage[0] = get_node_page(sbi, nids[0]);
506 if (IS_ERR(npage[0]))
507 return PTR_ERR(npage[0]);
510 /* if inline_data is set, should not report any block indices */
511 if (f2fs_has_inline_data(dn->inode) && index) {
513 f2fs_put_page(npage[0], 1);
519 nids[1] = get_nid(parent, offset[0], true);
520 dn->inode_page = npage[0];
521 dn->inode_page_locked = true;
523 /* get indirect or direct nodes */
524 for (i = 1; i <= level; i++) {
527 if (!nids[i] && mode == ALLOC_NODE) {
529 if (!alloc_nid(sbi, &(nids[i]))) {
535 npage[i] = new_node_page(dn, noffset[i], NULL);
536 if (IS_ERR(npage[i])) {
537 alloc_nid_failed(sbi, nids[i]);
538 err = PTR_ERR(npage[i]);
542 set_nid(parent, offset[i - 1], nids[i], i == 1);
543 alloc_nid_done(sbi, nids[i]);
545 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
546 npage[i] = get_node_page_ra(parent, offset[i - 1]);
547 if (IS_ERR(npage[i])) {
548 err = PTR_ERR(npage[i]);
554 dn->inode_page_locked = false;
557 f2fs_put_page(parent, 1);
561 npage[i] = get_node_page(sbi, nids[i]);
562 if (IS_ERR(npage[i])) {
563 err = PTR_ERR(npage[i]);
564 f2fs_put_page(npage[0], 0);
570 nids[i + 1] = get_nid(parent, offset[i], false);
573 dn->nid = nids[level];
574 dn->ofs_in_node = offset[level];
575 dn->node_page = npage[level];
576 dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
580 f2fs_put_page(parent, 1);
582 f2fs_put_page(npage[0], 0);
584 dn->inode_page = NULL;
585 dn->node_page = NULL;
589 static void truncate_node(struct dnode_of_data *dn)
591 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
594 get_node_info(sbi, dn->nid, &ni);
595 if (dn->inode->i_blocks == 0) {
596 f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
599 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
601 /* Deallocate node address */
602 invalidate_blocks(sbi, ni.blk_addr);
603 dec_valid_node_count(sbi, dn->inode);
604 set_node_addr(sbi, &ni, NULL_ADDR, false);
606 if (dn->nid == dn->inode->i_ino) {
607 remove_orphan_inode(sbi, dn->nid);
608 dec_valid_inode_count(sbi);
613 clear_node_page_dirty(dn->node_page);
614 set_sbi_flag(sbi, SBI_IS_DIRTY);
616 f2fs_put_page(dn->node_page, 1);
618 invalidate_mapping_pages(NODE_MAPPING(sbi),
619 dn->node_page->index, dn->node_page->index);
621 dn->node_page = NULL;
622 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
625 static int truncate_dnode(struct dnode_of_data *dn)
632 /* get direct node */
633 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
634 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
636 else if (IS_ERR(page))
637 return PTR_ERR(page);
639 /* Make dnode_of_data for parameter */
640 dn->node_page = page;
642 truncate_data_blocks(dn);
647 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
650 struct dnode_of_data rdn = *dn;
652 struct f2fs_node *rn;
654 unsigned int child_nofs;
659 return NIDS_PER_BLOCK + 1;
661 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
663 page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
665 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
666 return PTR_ERR(page);
669 rn = F2FS_NODE(page);
671 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
672 child_nid = le32_to_cpu(rn->in.nid[i]);
676 ret = truncate_dnode(&rdn);
679 set_nid(page, i, 0, false);
682 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
683 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
684 child_nid = le32_to_cpu(rn->in.nid[i]);
685 if (child_nid == 0) {
686 child_nofs += NIDS_PER_BLOCK + 1;
690 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
691 if (ret == (NIDS_PER_BLOCK + 1)) {
692 set_nid(page, i, 0, false);
694 } else if (ret < 0 && ret != -ENOENT) {
702 /* remove current indirect node */
703 dn->node_page = page;
707 f2fs_put_page(page, 1);
709 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
713 f2fs_put_page(page, 1);
714 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
718 static int truncate_partial_nodes(struct dnode_of_data *dn,
719 struct f2fs_inode *ri, int *offset, int depth)
721 struct page *pages[2];
728 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
732 /* get indirect nodes in the path */
733 for (i = 0; i < idx + 1; i++) {
734 /* reference count'll be increased */
735 pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
736 if (IS_ERR(pages[i])) {
737 err = PTR_ERR(pages[i]);
741 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
744 /* free direct nodes linked to a partial indirect node */
745 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
746 child_nid = get_nid(pages[idx], i, false);
750 err = truncate_dnode(dn);
753 set_nid(pages[idx], i, 0, false);
756 if (offset[idx + 1] == 0) {
757 dn->node_page = pages[idx];
761 f2fs_put_page(pages[idx], 1);
767 for (i = idx; i >= 0; i--)
768 f2fs_put_page(pages[i], 1);
770 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
776 * All the block addresses of data and nodes should be nullified.
778 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
780 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
781 int err = 0, cont = 1;
782 int level, offset[4], noffset[4];
783 unsigned int nofs = 0;
784 struct f2fs_inode *ri;
785 struct dnode_of_data dn;
788 trace_f2fs_truncate_inode_blocks_enter(inode, from);
790 level = get_node_path(F2FS_I(inode), from, offset, noffset);
792 page = get_node_page(sbi, inode->i_ino);
794 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
795 return PTR_ERR(page);
798 set_new_dnode(&dn, inode, page, NULL, 0);
801 ri = F2FS_INODE(page);
809 if (!offset[level - 1])
811 err = truncate_partial_nodes(&dn, ri, offset, level);
812 if (err < 0 && err != -ENOENT)
814 nofs += 1 + NIDS_PER_BLOCK;
817 nofs = 5 + 2 * NIDS_PER_BLOCK;
818 if (!offset[level - 1])
820 err = truncate_partial_nodes(&dn, ri, offset, level);
821 if (err < 0 && err != -ENOENT)
830 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
832 case NODE_DIR1_BLOCK:
833 case NODE_DIR2_BLOCK:
834 err = truncate_dnode(&dn);
837 case NODE_IND1_BLOCK:
838 case NODE_IND2_BLOCK:
839 err = truncate_nodes(&dn, nofs, offset[1], 2);
842 case NODE_DIND_BLOCK:
843 err = truncate_nodes(&dn, nofs, offset[1], 3);
850 if (err < 0 && err != -ENOENT)
852 if (offset[1] == 0 &&
853 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
855 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
856 f2fs_put_page(page, 1);
859 f2fs_wait_on_page_writeback(page, NODE);
860 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
861 set_page_dirty(page);
869 f2fs_put_page(page, 0);
870 trace_f2fs_truncate_inode_blocks_exit(inode, err);
871 return err > 0 ? 0 : err;
874 int truncate_xattr_node(struct inode *inode, struct page *page)
876 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
877 nid_t nid = F2FS_I(inode)->i_xattr_nid;
878 struct dnode_of_data dn;
884 npage = get_node_page(sbi, nid);
886 return PTR_ERR(npage);
888 F2FS_I(inode)->i_xattr_nid = 0;
890 /* need to do checkpoint during fsync */
891 F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
893 set_new_dnode(&dn, inode, page, npage, nid);
896 dn.inode_page_locked = true;
902 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
905 int remove_inode_page(struct inode *inode)
907 struct dnode_of_data dn;
910 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
911 err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
915 err = truncate_xattr_node(inode, dn.inode_page);
921 /* remove potential inline_data blocks */
922 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
923 S_ISLNK(inode->i_mode))
924 truncate_data_blocks_range(&dn, 1);
926 /* 0 is possible, after f2fs_new_inode() has failed */
927 f2fs_bug_on(F2FS_I_SB(inode),
928 inode->i_blocks != 0 && inode->i_blocks != 1);
930 /* will put inode & node pages */
935 struct page *new_inode_page(struct inode *inode)
937 struct dnode_of_data dn;
939 /* allocate inode page for new inode */
940 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
942 /* caller should f2fs_put_page(page, 1); */
943 return new_node_page(&dn, 0, NULL);
946 struct page *new_node_page(struct dnode_of_data *dn,
947 unsigned int ofs, struct page *ipage)
949 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
950 struct node_info old_ni, new_ni;
954 if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
955 return ERR_PTR(-EPERM);
957 page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
959 return ERR_PTR(-ENOMEM);
961 if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
966 get_node_info(sbi, dn->nid, &old_ni);
968 /* Reinitialize old_ni with new node page */
969 f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
971 new_ni.ino = dn->inode->i_ino;
972 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
974 f2fs_wait_on_page_writeback(page, NODE);
975 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
976 set_cold_node(dn->inode, page);
977 SetPageUptodate(page);
978 set_page_dirty(page);
980 if (f2fs_has_xattr_block(ofs))
981 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
983 dn->node_page = page;
985 update_inode(dn->inode, ipage);
989 inc_valid_inode_count(sbi);
994 clear_node_page_dirty(page);
995 f2fs_put_page(page, 1);
1000 * Caller should do after getting the following values.
1001 * 0: f2fs_put_page(page, 0)
1002 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1004 static int read_node_page(struct page *page, int rw)
1006 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1007 struct node_info ni;
1008 struct f2fs_io_info fio = {
1013 .encrypted_page = NULL,
1016 get_node_info(sbi, page->index, &ni);
1018 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1019 ClearPageUptodate(page);
1023 if (PageUptodate(page))
1026 fio.blk_addr = ni.blk_addr;
1027 return f2fs_submit_page_bio(&fio);
1031 * Readahead a node page
1033 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1038 apage = find_get_page(NODE_MAPPING(sbi), nid);
1039 if (apage && PageUptodate(apage)) {
1040 f2fs_put_page(apage, 0);
1043 f2fs_put_page(apage, 0);
1045 apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1049 err = read_node_page(apage, READA);
1050 f2fs_put_page(apage, err ? 1 : 0);
1053 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1058 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1060 return ERR_PTR(-ENOMEM);
1062 err = read_node_page(page, READ_SYNC);
1064 f2fs_put_page(page, 1);
1065 return ERR_PTR(err);
1066 } else if (err != LOCKED_PAGE) {
1070 if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1071 ClearPageUptodate(page);
1072 f2fs_put_page(page, 1);
1073 return ERR_PTR(-EIO);
1075 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1076 f2fs_put_page(page, 1);
1083 * Return a locked page for the desired node page.
1084 * And, readahead MAX_RA_NODE number of node pages.
1086 struct page *get_node_page_ra(struct page *parent, int start)
1088 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1089 struct blk_plug plug;
1094 /* First, try getting the desired direct node. */
1095 nid = get_nid(parent, start, false);
1097 return ERR_PTR(-ENOENT);
1099 page = grab_cache_page(NODE_MAPPING(sbi), nid);
1101 return ERR_PTR(-ENOMEM);
1103 err = read_node_page(page, READ_SYNC);
1105 f2fs_put_page(page, 1);
1106 return ERR_PTR(err);
1107 } else if (err == LOCKED_PAGE) {
1111 blk_start_plug(&plug);
1113 /* Then, try readahead for siblings of the desired node */
1114 end = start + MAX_RA_NODE;
1115 end = min(end, NIDS_PER_BLOCK);
1116 for (i = start + 1; i < end; i++) {
1117 nid = get_nid(parent, i, false);
1120 ra_node_page(sbi, nid);
1123 blk_finish_plug(&plug);
1126 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1127 f2fs_put_page(page, 1);
1131 if (unlikely(!PageUptodate(page))) {
1132 f2fs_put_page(page, 1);
1133 return ERR_PTR(-EIO);
1138 void sync_inode_page(struct dnode_of_data *dn)
1140 if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1141 update_inode(dn->inode, dn->node_page);
1142 } else if (dn->inode_page) {
1143 if (!dn->inode_page_locked)
1144 lock_page(dn->inode_page);
1145 update_inode(dn->inode, dn->inode_page);
1146 if (!dn->inode_page_locked)
1147 unlock_page(dn->inode_page);
1149 update_inode_page(dn->inode);
1153 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1154 struct writeback_control *wbc)
1157 struct pagevec pvec;
1158 int step = ino ? 2 : 0;
1159 int nwritten = 0, wrote = 0;
1161 pagevec_init(&pvec, 0);
1167 while (index <= end) {
1169 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1170 PAGECACHE_TAG_DIRTY,
1171 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1175 for (i = 0; i < nr_pages; i++) {
1176 struct page *page = pvec.pages[i];
1179 * flushing sequence with step:
1184 if (step == 0 && IS_DNODE(page))
1186 if (step == 1 && (!IS_DNODE(page) ||
1187 is_cold_node(page)))
1189 if (step == 2 && (!IS_DNODE(page) ||
1190 !is_cold_node(page)))
1195 * we should not skip writing node pages.
1197 if (ino && ino_of_node(page) == ino)
1199 else if (!trylock_page(page))
1202 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1207 if (ino && ino_of_node(page) != ino)
1208 goto continue_unlock;
1210 if (!PageDirty(page)) {
1211 /* someone wrote it for us */
1212 goto continue_unlock;
1215 if (!clear_page_dirty_for_io(page))
1216 goto continue_unlock;
1218 /* called by fsync() */
1219 if (ino && IS_DNODE(page)) {
1220 set_fsync_mark(page, 1);
1222 set_dentry_mark(page,
1223 need_dentry_mark(sbi, ino));
1226 set_fsync_mark(page, 0);
1227 set_dentry_mark(page, 0);
1230 if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1235 if (--wbc->nr_to_write == 0)
1238 pagevec_release(&pvec);
1241 if (wbc->nr_to_write == 0) {
1253 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1257 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1259 pgoff_t index = 0, end = LONG_MAX;
1260 struct pagevec pvec;
1261 int ret2 = 0, ret = 0;
1263 pagevec_init(&pvec, 0);
1265 while (index <= end) {
1267 nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1268 PAGECACHE_TAG_WRITEBACK,
1269 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1273 for (i = 0; i < nr_pages; i++) {
1274 struct page *page = pvec.pages[i];
1276 /* until radix tree lookup accepts end_index */
1277 if (unlikely(page->index > end))
1280 if (ino && ino_of_node(page) == ino) {
1281 f2fs_wait_on_page_writeback(page, NODE);
1282 if (TestClearPageError(page))
1286 pagevec_release(&pvec);
1290 if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1292 if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1299 static int f2fs_write_node_page(struct page *page,
1300 struct writeback_control *wbc)
1302 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1304 struct node_info ni;
1305 struct f2fs_io_info fio = {
1308 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1310 .encrypted_page = NULL,
1313 trace_f2fs_writepage(page, NODE);
1315 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1317 if (unlikely(f2fs_cp_error(sbi)))
1320 f2fs_wait_on_page_writeback(page, NODE);
1322 /* get old block addr of this node page */
1323 nid = nid_of_node(page);
1324 f2fs_bug_on(sbi, page->index != nid);
1326 if (wbc->for_reclaim) {
1327 if (!down_read_trylock(&sbi->node_write))
1330 down_read(&sbi->node_write);
1333 get_node_info(sbi, nid, &ni);
1335 /* This page is already truncated */
1336 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1337 ClearPageUptodate(page);
1338 dec_page_count(sbi, F2FS_DIRTY_NODES);
1339 up_read(&sbi->node_write);
1344 set_page_writeback(page);
1345 fio.blk_addr = ni.blk_addr;
1346 write_node_page(nid, &fio);
1347 set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1348 dec_page_count(sbi, F2FS_DIRTY_NODES);
1349 up_read(&sbi->node_write);
1352 if (wbc->for_reclaim)
1353 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1358 redirty_page_for_writepage(wbc, page);
1359 return AOP_WRITEPAGE_ACTIVATE;
1362 static int f2fs_write_node_pages(struct address_space *mapping,
1363 struct writeback_control *wbc)
1365 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1368 trace_f2fs_writepages(mapping->host, wbc, NODE);
1370 /* balancing f2fs's metadata in background */
1371 f2fs_balance_fs_bg(sbi);
1373 /* collect a number of dirty node pages and write together */
1374 if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1377 diff = nr_pages_to_write(sbi, NODE, wbc);
1378 wbc->sync_mode = WB_SYNC_NONE;
1379 sync_node_pages(sbi, 0, wbc);
1380 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1384 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1388 static int f2fs_set_node_page_dirty(struct page *page)
1390 trace_f2fs_set_page_dirty(page, NODE);
1392 SetPageUptodate(page);
1393 if (!PageDirty(page)) {
1394 __set_page_dirty_nobuffers(page);
1395 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1396 SetPagePrivate(page);
1397 f2fs_trace_pid(page);
1404 * Structure of the f2fs node operations
1406 const struct address_space_operations f2fs_node_aops = {
1407 .writepage = f2fs_write_node_page,
1408 .writepages = f2fs_write_node_pages,
1409 .set_page_dirty = f2fs_set_node_page_dirty,
1410 .invalidatepage = f2fs_invalidate_page,
1411 .releasepage = f2fs_release_page,
1414 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1417 return radix_tree_lookup(&nm_i->free_nid_root, n);
1420 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1424 radix_tree_delete(&nm_i->free_nid_root, i->nid);
1427 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1429 struct f2fs_nm_info *nm_i = NM_I(sbi);
1431 struct nat_entry *ne;
1432 bool allocated = false;
1434 if (!available_free_memory(sbi, FREE_NIDS))
1437 /* 0 nid should not be used */
1438 if (unlikely(nid == 0))
1442 /* do not add allocated nids */
1443 down_read(&nm_i->nat_tree_lock);
1444 ne = __lookup_nat_cache(nm_i, nid);
1446 (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1447 nat_get_blkaddr(ne) != NULL_ADDR))
1449 up_read(&nm_i->nat_tree_lock);
1454 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1458 if (radix_tree_preload(GFP_NOFS)) {
1459 kmem_cache_free(free_nid_slab, i);
1463 spin_lock(&nm_i->free_nid_list_lock);
1464 if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1465 spin_unlock(&nm_i->free_nid_list_lock);
1466 radix_tree_preload_end();
1467 kmem_cache_free(free_nid_slab, i);
1470 list_add_tail(&i->list, &nm_i->free_nid_list);
1472 spin_unlock(&nm_i->free_nid_list_lock);
1473 radix_tree_preload_end();
1477 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1480 bool need_free = false;
1482 spin_lock(&nm_i->free_nid_list_lock);
1483 i = __lookup_free_nid_list(nm_i, nid);
1484 if (i && i->state == NID_NEW) {
1485 __del_from_free_nid_list(nm_i, i);
1489 spin_unlock(&nm_i->free_nid_list_lock);
1492 kmem_cache_free(free_nid_slab, i);
1495 static void scan_nat_page(struct f2fs_sb_info *sbi,
1496 struct page *nat_page, nid_t start_nid)
1498 struct f2fs_nm_info *nm_i = NM_I(sbi);
1499 struct f2fs_nat_block *nat_blk = page_address(nat_page);
1503 i = start_nid % NAT_ENTRY_PER_BLOCK;
1505 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1507 if (unlikely(start_nid >= nm_i->max_nid))
1510 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1511 f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1512 if (blk_addr == NULL_ADDR) {
1513 if (add_free_nid(sbi, start_nid, true) < 0)
1519 static void build_free_nids(struct f2fs_sb_info *sbi)
1521 struct f2fs_nm_info *nm_i = NM_I(sbi);
1522 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1523 struct f2fs_summary_block *sum = curseg->sum_blk;
1525 nid_t nid = nm_i->next_scan_nid;
1527 /* Enough entries */
1528 if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1531 /* readahead nat pages to be scanned */
1532 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1536 struct page *page = get_current_nat_page(sbi, nid);
1538 scan_nat_page(sbi, page, nid);
1539 f2fs_put_page(page, 1);
1541 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1542 if (unlikely(nid >= nm_i->max_nid))
1545 if (++i >= FREE_NID_PAGES)
1549 /* go to the next free nat pages to find free nids abundantly */
1550 nm_i->next_scan_nid = nid;
1552 /* find free nids from current sum_pages */
1553 mutex_lock(&curseg->curseg_mutex);
1554 for (i = 0; i < nats_in_cursum(sum); i++) {
1555 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1556 nid = le32_to_cpu(nid_in_journal(sum, i));
1557 if (addr == NULL_ADDR)
1558 add_free_nid(sbi, nid, true);
1560 remove_free_nid(nm_i, nid);
1562 mutex_unlock(&curseg->curseg_mutex);
1564 ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1565 nm_i->ra_nid_pages, META_NAT, false);
1569 * If this function returns success, caller can obtain a new nid
1570 * from second parameter of this function.
1571 * The returned nid could be used ino as well as nid when inode is created.
1573 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1575 struct f2fs_nm_info *nm_i = NM_I(sbi);
1576 struct free_nid *i = NULL;
1578 if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1581 spin_lock(&nm_i->free_nid_list_lock);
1583 /* We should not use stale free nids created by build_free_nids */
1584 if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1585 struct node_info ni;
1587 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1588 list_for_each_entry(i, &nm_i->free_nid_list, list)
1589 if (i->state == NID_NEW)
1592 f2fs_bug_on(sbi, i->state != NID_NEW);
1594 i->state = NID_ALLOC;
1596 spin_unlock(&nm_i->free_nid_list_lock);
1598 /* check nid is allocated already */
1599 get_node_info(sbi, *nid, &ni);
1600 if (ni.blk_addr != NULL_ADDR) {
1601 alloc_nid_done(sbi, *nid);
1606 spin_unlock(&nm_i->free_nid_list_lock);
1608 /* Let's scan nat pages and its caches to get free nids */
1609 mutex_lock(&nm_i->build_lock);
1610 build_free_nids(sbi);
1611 mutex_unlock(&nm_i->build_lock);
1616 * alloc_nid() should be called prior to this function.
1618 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1620 struct f2fs_nm_info *nm_i = NM_I(sbi);
1623 spin_lock(&nm_i->free_nid_list_lock);
1624 i = __lookup_free_nid_list(nm_i, nid);
1625 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1626 __del_from_free_nid_list(nm_i, i);
1627 spin_unlock(&nm_i->free_nid_list_lock);
1629 kmem_cache_free(free_nid_slab, i);
1633 * alloc_nid() should be called prior to this function.
1635 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1637 struct f2fs_nm_info *nm_i = NM_I(sbi);
1639 bool need_free = false;
1644 spin_lock(&nm_i->free_nid_list_lock);
1645 i = __lookup_free_nid_list(nm_i, nid);
1646 f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1647 if (!available_free_memory(sbi, FREE_NIDS)) {
1648 __del_from_free_nid_list(nm_i, i);
1654 spin_unlock(&nm_i->free_nid_list_lock);
1657 kmem_cache_free(free_nid_slab, i);
1660 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1662 struct f2fs_nm_info *nm_i = NM_I(sbi);
1663 struct free_nid *i, *next;
1666 if (!mutex_trylock(&nm_i->build_lock))
1669 spin_lock(&nm_i->free_nid_list_lock);
1670 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1671 if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1673 if (i->state == NID_ALLOC)
1675 __del_from_free_nid_list(nm_i, i);
1676 kmem_cache_free(free_nid_slab, i);
1680 spin_unlock(&nm_i->free_nid_list_lock);
1681 mutex_unlock(&nm_i->build_lock);
1683 return nr - nr_shrink;
1686 void recover_inline_xattr(struct inode *inode, struct page *page)
1688 void *src_addr, *dst_addr;
1691 struct f2fs_inode *ri;
1693 ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1694 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1696 ri = F2FS_INODE(page);
1697 if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1698 clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1702 dst_addr = inline_xattr_addr(ipage);
1703 src_addr = inline_xattr_addr(page);
1704 inline_size = inline_xattr_size(inode);
1706 f2fs_wait_on_page_writeback(ipage, NODE);
1707 memcpy(dst_addr, src_addr, inline_size);
1709 update_inode(inode, ipage);
1710 f2fs_put_page(ipage, 1);
1713 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1715 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1716 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1717 nid_t new_xnid = nid_of_node(page);
1718 struct node_info ni;
1720 /* 1: invalidate the previous xattr nid */
1724 /* Deallocate node address */
1725 get_node_info(sbi, prev_xnid, &ni);
1726 f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1727 invalidate_blocks(sbi, ni.blk_addr);
1728 dec_valid_node_count(sbi, inode);
1729 set_node_addr(sbi, &ni, NULL_ADDR, false);
1732 /* 2: allocate new xattr nid */
1733 if (unlikely(!inc_valid_node_count(sbi, inode)))
1734 f2fs_bug_on(sbi, 1);
1736 remove_free_nid(NM_I(sbi), new_xnid);
1737 get_node_info(sbi, new_xnid, &ni);
1738 ni.ino = inode->i_ino;
1739 set_node_addr(sbi, &ni, NEW_ADDR, false);
1740 F2FS_I(inode)->i_xattr_nid = new_xnid;
1742 /* 3: update xattr blkaddr */
1743 refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1744 set_node_addr(sbi, &ni, blkaddr, false);
1746 update_inode_page(inode);
1749 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1751 struct f2fs_inode *src, *dst;
1752 nid_t ino = ino_of_node(page);
1753 struct node_info old_ni, new_ni;
1756 get_node_info(sbi, ino, &old_ni);
1758 if (unlikely(old_ni.blk_addr != NULL_ADDR))
1761 ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1765 /* Should not use this inode from free nid list */
1766 remove_free_nid(NM_I(sbi), ino);
1768 SetPageUptodate(ipage);
1769 fill_node_footer(ipage, ino, ino, 0, true);
1771 src = F2FS_INODE(page);
1772 dst = F2FS_INODE(ipage);
1774 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1776 dst->i_blocks = cpu_to_le64(1);
1777 dst->i_links = cpu_to_le32(1);
1778 dst->i_xattr_nid = 0;
1779 dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1784 if (unlikely(!inc_valid_node_count(sbi, NULL)))
1786 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1787 inc_valid_inode_count(sbi);
1788 set_page_dirty(ipage);
1789 f2fs_put_page(ipage, 1);
1793 int restore_node_summary(struct f2fs_sb_info *sbi,
1794 unsigned int segno, struct f2fs_summary_block *sum)
1796 struct f2fs_node *rn;
1797 struct f2fs_summary *sum_entry;
1799 int bio_blocks = MAX_BIO_BLOCKS(sbi);
1800 int i, idx, last_offset, nrpages;
1802 /* scan the node segment */
1803 last_offset = sbi->blocks_per_seg;
1804 addr = START_BLOCK(sbi, segno);
1805 sum_entry = &sum->entries[0];
1807 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1808 nrpages = min(last_offset - i, bio_blocks);
1810 /* readahead node pages */
1811 ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1813 for (idx = addr; idx < addr + nrpages; idx++) {
1814 struct page *page = get_tmp_page(sbi, idx);
1816 rn = F2FS_NODE(page);
1817 sum_entry->nid = rn->footer.nid;
1818 sum_entry->version = 0;
1819 sum_entry->ofs_in_node = 0;
1821 f2fs_put_page(page, 1);
1824 invalidate_mapping_pages(META_MAPPING(sbi), addr,
1830 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1832 struct f2fs_nm_info *nm_i = NM_I(sbi);
1833 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1834 struct f2fs_summary_block *sum = curseg->sum_blk;
1837 mutex_lock(&curseg->curseg_mutex);
1838 for (i = 0; i < nats_in_cursum(sum); i++) {
1839 struct nat_entry *ne;
1840 struct f2fs_nat_entry raw_ne;
1841 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1843 raw_ne = nat_in_journal(sum, i);
1845 down_write(&nm_i->nat_tree_lock);
1846 ne = __lookup_nat_cache(nm_i, nid);
1848 ne = grab_nat_entry(nm_i, nid);
1849 node_info_from_raw_nat(&ne->ni, &raw_ne);
1851 __set_nat_cache_dirty(nm_i, ne);
1852 up_write(&nm_i->nat_tree_lock);
1854 update_nats_in_cursum(sum, -i);
1855 mutex_unlock(&curseg->curseg_mutex);
1858 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1859 struct list_head *head, int max)
1861 struct nat_entry_set *cur;
1863 if (nes->entry_cnt >= max)
1866 list_for_each_entry(cur, head, set_list) {
1867 if (cur->entry_cnt >= nes->entry_cnt) {
1868 list_add(&nes->set_list, cur->set_list.prev);
1873 list_add_tail(&nes->set_list, head);
1876 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1877 struct nat_entry_set *set)
1879 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1880 struct f2fs_summary_block *sum = curseg->sum_blk;
1881 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1882 bool to_journal = true;
1883 struct f2fs_nat_block *nat_blk;
1884 struct nat_entry *ne, *cur;
1885 struct page *page = NULL;
1886 struct f2fs_nm_info *nm_i = NM_I(sbi);
1889 * there are two steps to flush nat entries:
1890 * #1, flush nat entries to journal in current hot data summary block.
1891 * #2, flush nat entries to nat page.
1893 if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1897 mutex_lock(&curseg->curseg_mutex);
1899 page = get_next_nat_page(sbi, start_nid);
1900 nat_blk = page_address(page);
1901 f2fs_bug_on(sbi, !nat_blk);
1904 /* flush dirty nats in nat entry set */
1905 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1906 struct f2fs_nat_entry *raw_ne;
1907 nid_t nid = nat_get_nid(ne);
1910 if (nat_get_blkaddr(ne) == NEW_ADDR)
1914 offset = lookup_journal_in_cursum(sum,
1915 NAT_JOURNAL, nid, 1);
1916 f2fs_bug_on(sbi, offset < 0);
1917 raw_ne = &nat_in_journal(sum, offset);
1918 nid_in_journal(sum, offset) = cpu_to_le32(nid);
1920 raw_ne = &nat_blk->entries[nid - start_nid];
1922 raw_nat_from_node_info(raw_ne, &ne->ni);
1924 down_write(&NM_I(sbi)->nat_tree_lock);
1926 __clear_nat_cache_dirty(NM_I(sbi), ne);
1927 up_write(&NM_I(sbi)->nat_tree_lock);
1929 if (nat_get_blkaddr(ne) == NULL_ADDR)
1930 add_free_nid(sbi, nid, false);
1934 mutex_unlock(&curseg->curseg_mutex);
1936 f2fs_put_page(page, 1);
1938 f2fs_bug_on(sbi, set->entry_cnt);
1940 down_write(&nm_i->nat_tree_lock);
1941 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1942 up_write(&nm_i->nat_tree_lock);
1943 kmem_cache_free(nat_entry_set_slab, set);
1947 * This function is called during the checkpointing process.
1949 void flush_nat_entries(struct f2fs_sb_info *sbi)
1951 struct f2fs_nm_info *nm_i = NM_I(sbi);
1952 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1953 struct f2fs_summary_block *sum = curseg->sum_blk;
1954 struct nat_entry_set *setvec[SETVEC_SIZE];
1955 struct nat_entry_set *set, *tmp;
1960 if (!nm_i->dirty_nat_cnt)
1963 * if there are no enough space in journal to store dirty nat
1964 * entries, remove all entries from journal and merge them
1965 * into nat entry set.
1967 if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1968 remove_nats_in_journal(sbi);
1970 down_write(&nm_i->nat_tree_lock);
1971 while ((found = __gang_lookup_nat_set(nm_i,
1972 set_idx, SETVEC_SIZE, setvec))) {
1974 set_idx = setvec[found - 1]->set + 1;
1975 for (idx = 0; idx < found; idx++)
1976 __adjust_nat_entry_set(setvec[idx], &sets,
1977 MAX_NAT_JENTRIES(sum));
1979 up_write(&nm_i->nat_tree_lock);
1981 /* flush dirty nats in nat entry set */
1982 list_for_each_entry_safe(set, tmp, &sets, set_list)
1983 __flush_nat_entry_set(sbi, set);
1985 f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1988 static int init_node_manager(struct f2fs_sb_info *sbi)
1990 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1991 struct f2fs_nm_info *nm_i = NM_I(sbi);
1992 unsigned char *version_bitmap;
1993 unsigned int nat_segs, nat_blocks;
1995 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1997 /* segment_count_nat includes pair segment so divide to 2. */
1998 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1999 nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2001 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2003 /* not used nids: 0, node, meta, (and root counted as valid node) */
2004 nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2007 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2008 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2010 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2011 INIT_LIST_HEAD(&nm_i->free_nid_list);
2012 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2013 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2014 INIT_LIST_HEAD(&nm_i->nat_entries);
2016 mutex_init(&nm_i->build_lock);
2017 spin_lock_init(&nm_i->free_nid_list_lock);
2018 init_rwsem(&nm_i->nat_tree_lock);
2020 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2021 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2022 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2023 if (!version_bitmap)
2026 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2028 if (!nm_i->nat_bitmap)
2033 int build_node_manager(struct f2fs_sb_info *sbi)
2037 sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2041 err = init_node_manager(sbi);
2045 build_free_nids(sbi);
2049 void destroy_node_manager(struct f2fs_sb_info *sbi)
2051 struct f2fs_nm_info *nm_i = NM_I(sbi);
2052 struct free_nid *i, *next_i;
2053 struct nat_entry *natvec[NATVEC_SIZE];
2054 struct nat_entry_set *setvec[SETVEC_SIZE];
2061 /* destroy free nid list */
2062 spin_lock(&nm_i->free_nid_list_lock);
2063 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2064 f2fs_bug_on(sbi, i->state == NID_ALLOC);
2065 __del_from_free_nid_list(nm_i, i);
2067 spin_unlock(&nm_i->free_nid_list_lock);
2068 kmem_cache_free(free_nid_slab, i);
2069 spin_lock(&nm_i->free_nid_list_lock);
2071 f2fs_bug_on(sbi, nm_i->fcnt);
2072 spin_unlock(&nm_i->free_nid_list_lock);
2074 /* destroy nat cache */
2075 down_write(&nm_i->nat_tree_lock);
2076 while ((found = __gang_lookup_nat_cache(nm_i,
2077 nid, NATVEC_SIZE, natvec))) {
2080 nid = nat_get_nid(natvec[found - 1]) + 1;
2081 for (idx = 0; idx < found; idx++)
2082 __del_from_nat_cache(nm_i, natvec[idx]);
2084 f2fs_bug_on(sbi, nm_i->nat_cnt);
2086 /* destroy nat set cache */
2088 while ((found = __gang_lookup_nat_set(nm_i,
2089 nid, SETVEC_SIZE, setvec))) {
2092 nid = setvec[found - 1]->set + 1;
2093 for (idx = 0; idx < found; idx++) {
2094 /* entry_cnt is not zero, when cp_error was occurred */
2095 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2096 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2097 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2100 up_write(&nm_i->nat_tree_lock);
2102 kfree(nm_i->nat_bitmap);
2103 sbi->nm_info = NULL;
2107 int __init create_node_manager_caches(void)
2109 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2110 sizeof(struct nat_entry));
2111 if (!nat_entry_slab)
2114 free_nid_slab = f2fs_kmem_cache_create("free_nid",
2115 sizeof(struct free_nid));
2117 goto destroy_nat_entry;
2119 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2120 sizeof(struct nat_entry_set));
2121 if (!nat_entry_set_slab)
2122 goto destroy_free_nid;
2126 kmem_cache_destroy(free_nid_slab);
2128 kmem_cache_destroy(nat_entry_slab);
2133 void destroy_node_manager_caches(void)
2135 kmem_cache_destroy(nat_entry_set_slab);
2136 kmem_cache_destroy(free_nid_slab);
2137 kmem_cache_destroy(nat_entry_slab);