2 * This file is part of UBIFS.
4 * Copyright (C) 2006-2008 Nokia Corporation.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 * Author: Adrian Hunter
25 * An orphan is an inode number whose inode node has been committed to the index
26 * with a link count of zero. That happens when an open file is deleted
27 * (unlinked) and then a commit is run. In the normal course of events the inode
28 * would be deleted when the file is closed. However in the case of an unclean
29 * unmount, orphans need to be accounted for. After an unclean unmount, the
30 * orphans' inodes must be deleted which means either scanning the entire index
31 * looking for them, or keeping a list on flash somewhere. This unit implements
32 * the latter approach.
34 * The orphan area is a fixed number of LEBs situated between the LPT area and
35 * the main area. The number of orphan area LEBs is specified when the file
36 * system is created. The minimum number is 1. The size of the orphan area
37 * should be so that it can hold the maximum number of orphans that are expected
38 * to ever exist at one time.
40 * The number of orphans that can fit in a LEB is:
42 * (c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64)
44 * For example: a 15872 byte LEB can fit 1980 orphans so 1 LEB may be enough.
46 * Orphans are accumulated in a rb-tree. When an inode's link count drops to
47 * zero, the inode number is added to the rb-tree. It is removed from the tree
48 * when the inode is deleted. Any new orphans that are in the orphan tree when
49 * the commit is run, are written to the orphan area in 1 or more orph nodes.
50 * If the orphan area is full, it is consolidated to make space. There is
51 * always enough space because validation prevents the user from creating more
52 * than the maximum number of orphans allowed.
55 #ifdef CONFIG_UBIFS_FS_DEBUG
56 static int dbg_check_orphans(struct ubifs_info *c);
58 #define dbg_check_orphans(c) 0
62 * ubifs_add_orphan - add an orphan.
63 * @c: UBIFS file-system description object
64 * @inum: orphan inode number
66 * Add an orphan. This function is called when an inodes link count drops to
69 int ubifs_add_orphan(struct ubifs_info *c, ino_t inum)
71 struct ubifs_orphan *orphan, *o;
72 struct rb_node **p, *parent = NULL;
74 orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_NOFS);
80 spin_lock(&c->orphan_lock);
81 if (c->tot_orphans >= c->max_orphans) {
82 spin_unlock(&c->orphan_lock);
86 p = &c->orph_tree.rb_node;
89 o = rb_entry(parent, struct ubifs_orphan, rb);
92 else if (inum > o->inum)
95 dbg_err("orphaned twice");
96 spin_unlock(&c->orphan_lock);
103 rb_link_node(&orphan->rb, parent, p);
104 rb_insert_color(&orphan->rb, &c->orph_tree);
105 list_add_tail(&orphan->list, &c->orph_list);
106 list_add_tail(&orphan->new_list, &c->orph_new);
107 spin_unlock(&c->orphan_lock);
108 dbg_gen("ino %lu", (unsigned long)inum);
113 * ubifs_delete_orphan - delete an orphan.
114 * @c: UBIFS file-system description object
115 * @inum: orphan inode number
117 * Delete an orphan. This function is called when an inode is deleted.
119 void ubifs_delete_orphan(struct ubifs_info *c, ino_t inum)
121 struct ubifs_orphan *o;
124 spin_lock(&c->orphan_lock);
125 p = c->orph_tree.rb_node;
127 o = rb_entry(p, struct ubifs_orphan, rb);
130 else if (inum > o->inum)
134 spin_unlock(&c->orphan_lock);
135 dbg_gen("deleted twice ino %lu",
136 (unsigned long)inum);
140 o->dnext = c->orph_dnext;
142 spin_unlock(&c->orphan_lock);
143 dbg_gen("delete later ino %lu",
144 (unsigned long)inum);
147 rb_erase(p, &c->orph_tree);
151 list_del(&o->new_list);
154 spin_unlock(&c->orphan_lock);
156 dbg_gen("inum %lu", (unsigned long)inum);
160 spin_unlock(&c->orphan_lock);
161 dbg_err("missing orphan ino %lu", (unsigned long)inum);
166 * ubifs_orphan_start_commit - start commit of orphans.
167 * @c: UBIFS file-system description object
169 * Start commit of orphans.
171 int ubifs_orphan_start_commit(struct ubifs_info *c)
173 struct ubifs_orphan *orphan, **last;
175 spin_lock(&c->orphan_lock);
176 last = &c->orph_cnext;
177 list_for_each_entry(orphan, &c->orph_new, new_list) {
178 ubifs_assert(orphan->new);
181 last = &orphan->cnext;
183 *last = orphan->cnext;
184 c->cmt_orphans = c->new_orphans;
186 dbg_cmt("%d orphans to commit", c->cmt_orphans);
187 INIT_LIST_HEAD(&c->orph_new);
188 if (c->tot_orphans == 0)
192 spin_unlock(&c->orphan_lock);
197 * avail_orphs - calculate available space.
198 * @c: UBIFS file-system description object
200 * This function returns the number of orphans that can be written in the
203 static int avail_orphs(struct ubifs_info *c)
205 int avail_lebs, avail, gap;
207 avail_lebs = c->orph_lebs - (c->ohead_lnum - c->orph_first) - 1;
209 ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
210 gap = c->leb_size - c->ohead_offs;
211 if (gap >= UBIFS_ORPH_NODE_SZ + sizeof(__le64))
212 avail += (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
217 * tot_avail_orphs - calculate total space.
218 * @c: UBIFS file-system description object
220 * This function returns the number of orphans that can be written in half
221 * the total space. That leaves half the space for adding new orphans.
223 static int tot_avail_orphs(struct ubifs_info *c)
225 int avail_lebs, avail;
227 avail_lebs = c->orph_lebs;
229 ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
234 * do_write_orph_node - write a node
235 * @c: UBIFS file-system description object
236 * @len: length of node
237 * @atomic: write atomically
239 * This function writes a node to the orphan head from the orphan buffer. If
240 * %atomic is not zero, then the write is done atomically. On success, %0 is
241 * returned, otherwise a negative error code is returned.
243 static int do_write_orph_node(struct ubifs_info *c, int len, int atomic)
248 ubifs_assert(c->ohead_offs == 0);
249 ubifs_prepare_node(c, c->orph_buf, len, 1);
250 len = ALIGN(len, c->min_io_size);
251 err = ubifs_leb_change(c, c->ohead_lnum, c->orph_buf, len,
254 if (c->ohead_offs == 0) {
255 /* Ensure LEB has been unmapped */
256 err = ubifs_leb_unmap(c, c->ohead_lnum);
260 err = ubifs_write_node(c, c->orph_buf, len, c->ohead_lnum,
261 c->ohead_offs, UBI_SHORTTERM);
267 * write_orph_node - write an orph node
268 * @c: UBIFS file-system description object
269 * @atomic: write atomically
271 * This function builds an orph node from the cnext list and writes it to the
272 * orphan head. On success, %0 is returned, otherwise a negative error code
275 static int write_orph_node(struct ubifs_info *c, int atomic)
277 struct ubifs_orphan *orphan, *cnext;
278 struct ubifs_orph_node *orph;
279 int gap, err, len, cnt, i;
281 ubifs_assert(c->cmt_orphans > 0);
282 gap = c->leb_size - c->ohead_offs;
283 if (gap < UBIFS_ORPH_NODE_SZ + sizeof(__le64)) {
287 if (c->ohead_lnum > c->orph_last) {
289 * We limit the number of orphans so that this should
292 ubifs_err("out of space in orphan area");
296 cnt = (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
297 if (cnt > c->cmt_orphans)
298 cnt = c->cmt_orphans;
299 len = UBIFS_ORPH_NODE_SZ + cnt * sizeof(__le64);
300 ubifs_assert(c->orph_buf);
302 orph->ch.node_type = UBIFS_ORPH_NODE;
303 spin_lock(&c->orphan_lock);
304 cnext = c->orph_cnext;
305 for (i = 0; i < cnt; i++) {
307 orph->inos[i] = cpu_to_le64(orphan->inum);
308 cnext = orphan->cnext;
309 orphan->cnext = NULL;
311 c->orph_cnext = cnext;
312 c->cmt_orphans -= cnt;
313 spin_unlock(&c->orphan_lock);
315 orph->cmt_no = cpu_to_le64(c->cmt_no);
317 /* Mark the last node of the commit */
318 orph->cmt_no = cpu_to_le64((c->cmt_no) | (1ULL << 63));
319 ubifs_assert(c->ohead_offs + len <= c->leb_size);
320 ubifs_assert(c->ohead_lnum >= c->orph_first);
321 ubifs_assert(c->ohead_lnum <= c->orph_last);
322 err = do_write_orph_node(c, len, atomic);
323 c->ohead_offs += ALIGN(len, c->min_io_size);
324 c->ohead_offs = ALIGN(c->ohead_offs, 8);
329 * write_orph_nodes - write orph nodes until there are no more to commit
330 * @c: UBIFS file-system description object
331 * @atomic: write atomically
333 * This function writes orph nodes for all the orphans to commit. On success,
334 * %0 is returned, otherwise a negative error code is returned.
336 static int write_orph_nodes(struct ubifs_info *c, int atomic)
340 while (c->cmt_orphans > 0) {
341 err = write_orph_node(c, atomic);
348 /* Unmap any unused LEBs after consolidation */
349 lnum = c->ohead_lnum + 1;
350 for (lnum = c->ohead_lnum + 1; lnum <= c->orph_last; lnum++) {
351 err = ubifs_leb_unmap(c, lnum);
360 * consolidate - consolidate the orphan area.
361 * @c: UBIFS file-system description object
363 * This function enables consolidation by putting all the orphans into the list
364 * to commit. The list is in the order that the orphans were added, and the
365 * LEBs are written atomically in order, so at no time can orphans be lost by
366 * an unclean unmount.
368 * This function returns %0 on success and a negative error code on failure.
370 static int consolidate(struct ubifs_info *c)
372 int tot_avail = tot_avail_orphs(c), err = 0;
374 spin_lock(&c->orphan_lock);
375 dbg_cmt("there is space for %d orphans and there are %d",
376 tot_avail, c->tot_orphans);
377 if (c->tot_orphans - c->new_orphans <= tot_avail) {
378 struct ubifs_orphan *orphan, **last;
381 /* Change the cnext list to include all non-new orphans */
382 last = &c->orph_cnext;
383 list_for_each_entry(orphan, &c->orph_list, list) {
387 last = &orphan->cnext;
390 *last = orphan->cnext;
391 ubifs_assert(cnt == c->tot_orphans - c->new_orphans);
392 c->cmt_orphans = cnt;
393 c->ohead_lnum = c->orph_first;
397 * We limit the number of orphans so that this should
400 ubifs_err("out of space in orphan area");
403 spin_unlock(&c->orphan_lock);
408 * commit_orphans - commit orphans.
409 * @c: UBIFS file-system description object
411 * This function commits orphans to flash. On success, %0 is returned,
412 * otherwise a negative error code is returned.
414 static int commit_orphans(struct ubifs_info *c)
416 int avail, atomic = 0, err;
418 ubifs_assert(c->cmt_orphans > 0);
419 avail = avail_orphs(c);
420 if (avail < c->cmt_orphans) {
421 /* Not enough space to write new orphans, so consolidate */
422 err = consolidate(c);
427 err = write_orph_nodes(c, atomic);
432 * erase_deleted - erase the orphans marked for deletion.
433 * @c: UBIFS file-system description object
435 * During commit, the orphans being committed cannot be deleted, so they are
436 * marked for deletion and deleted by this function. Also, the recovery
437 * adds killed orphans to the deletion list, and therefore they are deleted
440 static void erase_deleted(struct ubifs_info *c)
442 struct ubifs_orphan *orphan, *dnext;
444 spin_lock(&c->orphan_lock);
445 dnext = c->orph_dnext;
448 dnext = orphan->dnext;
449 ubifs_assert(!orphan->new);
450 rb_erase(&orphan->rb, &c->orph_tree);
451 list_del(&orphan->list);
453 dbg_gen("deleting orphan ino %lu", (unsigned long)orphan->inum);
456 c->orph_dnext = NULL;
457 spin_unlock(&c->orphan_lock);
461 * ubifs_orphan_end_commit - end commit of orphans.
462 * @c: UBIFS file-system description object
464 * End commit of orphans.
466 int ubifs_orphan_end_commit(struct ubifs_info *c)
470 if (c->cmt_orphans != 0) {
471 err = commit_orphans(c);
476 err = dbg_check_orphans(c);
481 * clear_orphans - erase all LEBs used for orphans.
482 * @c: UBIFS file-system description object
484 * If recovery is not required, then the orphans from the previous session
485 * are not needed. This function locates the LEBs used to record
486 * orphans, and un-maps them.
488 static int clear_orphans(struct ubifs_info *c)
492 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
493 err = ubifs_leb_unmap(c, lnum);
497 c->ohead_lnum = c->orph_first;
503 * insert_dead_orphan - insert an orphan.
504 * @c: UBIFS file-system description object
505 * @inum: orphan inode number
507 * This function is a helper to the 'do_kill_orphans()' function. The orphan
508 * must be kept until the next commit, so it is added to the rb-tree and the
511 static int insert_dead_orphan(struct ubifs_info *c, ino_t inum)
513 struct ubifs_orphan *orphan, *o;
514 struct rb_node **p, *parent = NULL;
516 orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_KERNEL);
521 p = &c->orph_tree.rb_node;
524 o = rb_entry(parent, struct ubifs_orphan, rb);
527 else if (inum > o->inum)
530 /* Already added - no problem */
536 rb_link_node(&orphan->rb, parent, p);
537 rb_insert_color(&orphan->rb, &c->orph_tree);
538 list_add_tail(&orphan->list, &c->orph_list);
539 orphan->dnext = c->orph_dnext;
540 c->orph_dnext = orphan;
541 dbg_mnt("ino %lu, new %d, tot %d", (unsigned long)inum,
542 c->new_orphans, c->tot_orphans);
547 * do_kill_orphans - remove orphan inodes from the index.
548 * @c: UBIFS file-system description object
550 * @last_cmt_no: cmt_no of last orph node read is passed and returned here
551 * @outofdate: whether the LEB is out of date is returned here
552 * @last_flagged: whether the end orph node is encountered
554 * This function is a helper to the 'kill_orphans()' function. It goes through
555 * every orphan node in a LEB and for every inode number recorded, removes
556 * all keys for that inode from the TNC.
558 static int do_kill_orphans(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
559 unsigned long long *last_cmt_no, int *outofdate,
562 struct ubifs_scan_node *snod;
563 struct ubifs_orph_node *orph;
564 unsigned long long cmt_no;
566 int i, n, err, first = 1;
568 list_for_each_entry(snod, &sleb->nodes, list) {
569 if (snod->type != UBIFS_ORPH_NODE) {
570 ubifs_err("invalid node type %d in orphan area at "
571 "%d:%d", snod->type, sleb->lnum, snod->offs);
572 dbg_dump_node(c, snod->node);
578 /* Check commit number */
579 cmt_no = le64_to_cpu(orph->cmt_no) & LLONG_MAX;
581 * The commit number on the master node may be less, because
582 * of a failed commit. If there are several failed commits in a
583 * row, the commit number written on orph nodes will continue to
584 * increase (because the commit number is adjusted here) even
585 * though the commit number on the master node stays the same
586 * because the master node has not been re-written.
588 if (cmt_no > c->cmt_no)
590 if (cmt_no < *last_cmt_no && *last_flagged) {
592 * The last orph node had a higher commit number and was
593 * flagged as the last written for that commit number.
594 * That makes this orph node, out of date.
597 ubifs_err("out of order commit number %llu in "
598 "orphan node at %d:%d",
599 cmt_no, sleb->lnum, snod->offs);
600 dbg_dump_node(c, snod->node);
603 dbg_rcvry("out of date LEB %d", sleb->lnum);
611 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
612 for (i = 0; i < n; i++) {
613 inum = le64_to_cpu(orph->inos[i]);
614 dbg_rcvry("deleting orphaned inode %lu",
615 (unsigned long)inum);
616 err = ubifs_tnc_remove_ino(c, inum);
619 err = insert_dead_orphan(c, inum);
624 *last_cmt_no = cmt_no;
625 if (le64_to_cpu(orph->cmt_no) & (1ULL << 63)) {
626 dbg_rcvry("last orph node for commit %llu at %d:%d",
627 cmt_no, sleb->lnum, snod->offs);
637 * kill_orphans - remove all orphan inodes from the index.
638 * @c: UBIFS file-system description object
640 * If recovery is required, then orphan inodes recorded during the previous
641 * session (which ended with an unclean unmount) must be deleted from the index.
642 * This is done by updating the TNC, but since the index is not updated until
643 * the next commit, the LEBs where the orphan information is recorded are not
644 * erased until the next commit.
646 static int kill_orphans(struct ubifs_info *c)
648 unsigned long long last_cmt_no = 0;
649 int lnum, err = 0, outofdate = 0, last_flagged = 0;
651 c->ohead_lnum = c->orph_first;
653 /* Check no-orphans flag and skip this if no orphans */
655 dbg_rcvry("no orphans");
659 * Orph nodes always start at c->orph_first and are written to each
660 * successive LEB in turn. Generally unused LEBs will have been unmapped
661 * but may contain out of date orph nodes if the unmap didn't go
662 * through. In addition, the last orph node written for each commit is
663 * marked (top bit of orph->cmt_no is set to 1). It is possible that
664 * there are orph nodes from the next commit (i.e. the commit did not
665 * complete successfully). In that case, no orphans will have been lost
666 * due to the way that orphans are written, and any orphans added will
667 * be valid orphans anyway and so can be deleted.
669 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
670 struct ubifs_scan_leb *sleb;
672 dbg_rcvry("LEB %d", lnum);
673 sleb = ubifs_scan(c, lnum, 0, c->sbuf);
675 sleb = ubifs_recover_leb(c, lnum, 0, c->sbuf, 0);
681 err = do_kill_orphans(c, sleb, &last_cmt_no, &outofdate,
683 if (err || outofdate) {
684 ubifs_scan_destroy(sleb);
688 c->ohead_lnum = lnum;
689 c->ohead_offs = sleb->endpt;
691 ubifs_scan_destroy(sleb);
697 * ubifs_mount_orphans - delete orphan inodes and erase LEBs that recorded them.
698 * @c: UBIFS file-system description object
699 * @unclean: indicates recovery from unclean unmount
700 * @read_only: indicates read only mount
702 * This function is called when mounting to erase orphans from the previous
703 * session. If UBIFS was not unmounted cleanly, then the inodes recorded as
704 * orphans are deleted.
706 int ubifs_mount_orphans(struct ubifs_info *c, int unclean, int read_only)
710 c->max_orphans = tot_avail_orphs(c);
713 c->orph_buf = vmalloc(c->leb_size);
719 err = kill_orphans(c);
721 err = clear_orphans(c);
726 #ifdef CONFIG_UBIFS_FS_DEBUG
728 struct check_orphan {
734 unsigned long last_ino;
735 unsigned long tot_inos;
736 unsigned long missing;
737 unsigned long long leaf_cnt;
738 struct ubifs_ino_node *node;
742 static int dbg_find_orphan(struct ubifs_info *c, ino_t inum)
744 struct ubifs_orphan *o;
747 spin_lock(&c->orphan_lock);
748 p = c->orph_tree.rb_node;
750 o = rb_entry(p, struct ubifs_orphan, rb);
753 else if (inum > o->inum)
756 spin_unlock(&c->orphan_lock);
760 spin_unlock(&c->orphan_lock);
764 static int dbg_ins_check_orphan(struct rb_root *root, ino_t inum)
766 struct check_orphan *orphan, *o;
767 struct rb_node **p, *parent = NULL;
769 orphan = kzalloc(sizeof(struct check_orphan), GFP_NOFS);
777 o = rb_entry(parent, struct check_orphan, rb);
780 else if (inum > o->inum)
787 rb_link_node(&orphan->rb, parent, p);
788 rb_insert_color(&orphan->rb, root);
792 static int dbg_find_check_orphan(struct rb_root *root, ino_t inum)
794 struct check_orphan *o;
799 o = rb_entry(p, struct check_orphan, rb);
802 else if (inum > o->inum)
810 static void dbg_free_check_tree(struct rb_root *root)
812 struct rb_node *this = root->rb_node;
813 struct check_orphan *o;
817 this = this->rb_left;
819 } else if (this->rb_right) {
820 this = this->rb_right;
823 o = rb_entry(this, struct check_orphan, rb);
824 this = rb_parent(this);
826 if (this->rb_left == &o->rb)
827 this->rb_left = NULL;
829 this->rb_right = NULL;
835 static int dbg_orphan_check(struct ubifs_info *c, struct ubifs_zbranch *zbr,
838 struct check_info *ci = priv;
842 inum = key_inum(c, &zbr->key);
843 if (inum != ci->last_ino) {
844 /* Lowest node type is the inode node, so it comes first */
845 if (key_type(c, &zbr->key) != UBIFS_INO_KEY)
846 ubifs_err("found orphan node ino %lu, type %d",
847 (unsigned long)inum, key_type(c, &zbr->key));
850 err = ubifs_tnc_read_node(c, zbr, ci->node);
852 ubifs_err("node read failed, error %d", err);
855 if (ci->node->nlink == 0)
856 /* Must be recorded as an orphan */
857 if (!dbg_find_check_orphan(&ci->root, inum) &&
858 !dbg_find_orphan(c, inum)) {
859 ubifs_err("missing orphan, ino %lu",
860 (unsigned long)inum);
868 static int dbg_read_orphans(struct check_info *ci, struct ubifs_scan_leb *sleb)
870 struct ubifs_scan_node *snod;
871 struct ubifs_orph_node *orph;
875 list_for_each_entry(snod, &sleb->nodes, list) {
877 if (snod->type != UBIFS_ORPH_NODE)
880 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
881 for (i = 0; i < n; i++) {
882 inum = le64_to_cpu(orph->inos[i]);
883 err = dbg_ins_check_orphan(&ci->root, inum);
891 static int dbg_scan_orphans(struct ubifs_info *c, struct check_info *ci)
895 /* Check no-orphans flag and skip this if no orphans */
899 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
900 struct ubifs_scan_leb *sleb;
902 sleb = ubifs_scan(c, lnum, 0, c->dbg->buf);
908 err = dbg_read_orphans(ci, sleb);
909 ubifs_scan_destroy(sleb);
917 static int dbg_check_orphans(struct ubifs_info *c)
919 struct check_info ci;
922 if (!(ubifs_chk_flags & UBIFS_CHK_ORPH))
930 ci.node = kmalloc(UBIFS_MAX_INO_NODE_SZ, GFP_NOFS);
932 ubifs_err("out of memory");
936 err = dbg_scan_orphans(c, &ci);
940 err = dbg_walk_index(c, &dbg_orphan_check, NULL, &ci);
942 ubifs_err("cannot scan TNC, error %d", err);
947 ubifs_err("%lu missing orphan(s)", ci.missing);
952 dbg_cmt("last inode number is %lu", ci.last_ino);
953 dbg_cmt("total number of inodes is %lu", ci.tot_inos);
954 dbg_cmt("total number of leaf nodes is %llu", ci.leaf_cnt);
957 dbg_free_check_tree(&ci.root);
962 #endif /* CONFIG_UBIFS_FS_DEBUG */