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 orphan 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 static int dbg_check_orphans(struct ubifs_info *c);
58 * ubifs_add_orphan - add an orphan.
59 * @c: UBIFS file-system description object
60 * @inum: orphan inode number
62 * Add an orphan. This function is called when an inodes link count drops to
65 int ubifs_add_orphan(struct ubifs_info *c, ino_t inum)
67 struct ubifs_orphan *orphan, *o;
68 struct rb_node **p, *parent = NULL;
70 orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_NOFS);
76 spin_lock(&c->orphan_lock);
77 if (c->tot_orphans >= c->max_orphans) {
78 spin_unlock(&c->orphan_lock);
82 p = &c->orph_tree.rb_node;
85 o = rb_entry(parent, struct ubifs_orphan, rb);
88 else if (inum > o->inum)
91 ubifs_err("orphaned twice");
92 spin_unlock(&c->orphan_lock);
99 rb_link_node(&orphan->rb, parent, p);
100 rb_insert_color(&orphan->rb, &c->orph_tree);
101 list_add_tail(&orphan->list, &c->orph_list);
102 list_add_tail(&orphan->new_list, &c->orph_new);
103 spin_unlock(&c->orphan_lock);
104 dbg_gen("ino %lu", (unsigned long)inum);
109 * ubifs_delete_orphan - delete an orphan.
110 * @c: UBIFS file-system description object
111 * @inum: orphan inode number
113 * Delete an orphan. This function is called when an inode is deleted.
115 void ubifs_delete_orphan(struct ubifs_info *c, ino_t inum)
117 struct ubifs_orphan *o;
120 spin_lock(&c->orphan_lock);
121 p = c->orph_tree.rb_node;
123 o = rb_entry(p, struct ubifs_orphan, rb);
126 else if (inum > o->inum)
130 spin_unlock(&c->orphan_lock);
131 dbg_gen("deleted twice ino %lu",
132 (unsigned long)inum);
136 o->dnext = c->orph_dnext;
138 spin_unlock(&c->orphan_lock);
139 dbg_gen("delete later ino %lu",
140 (unsigned long)inum);
143 rb_erase(p, &c->orph_tree);
147 list_del(&o->new_list);
150 spin_unlock(&c->orphan_lock);
152 dbg_gen("inum %lu", (unsigned long)inum);
156 spin_unlock(&c->orphan_lock);
157 ubifs_err("missing orphan ino %lu", (unsigned long)inum);
162 * ubifs_orphan_start_commit - start commit of orphans.
163 * @c: UBIFS file-system description object
165 * Start commit of orphans.
167 int ubifs_orphan_start_commit(struct ubifs_info *c)
169 struct ubifs_orphan *orphan, **last;
171 spin_lock(&c->orphan_lock);
172 last = &c->orph_cnext;
173 list_for_each_entry(orphan, &c->orph_new, new_list) {
174 ubifs_assert(orphan->new);
175 ubifs_assert(!orphan->cmt);
179 last = &orphan->cnext;
182 c->cmt_orphans = c->new_orphans;
184 dbg_cmt("%d orphans to commit", c->cmt_orphans);
185 INIT_LIST_HEAD(&c->orph_new);
186 if (c->tot_orphans == 0)
190 spin_unlock(&c->orphan_lock);
195 * avail_orphs - calculate available space.
196 * @c: UBIFS file-system description object
198 * This function returns the number of orphans that can be written in the
201 static int avail_orphs(struct ubifs_info *c)
203 int avail_lebs, avail, gap;
205 avail_lebs = c->orph_lebs - (c->ohead_lnum - c->orph_first) - 1;
207 ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
208 gap = c->leb_size - c->ohead_offs;
209 if (gap >= UBIFS_ORPH_NODE_SZ + sizeof(__le64))
210 avail += (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
215 * tot_avail_orphs - calculate total space.
216 * @c: UBIFS file-system description object
218 * This function returns the number of orphans that can be written in half
219 * the total space. That leaves half the space for adding new orphans.
221 static int tot_avail_orphs(struct ubifs_info *c)
223 int avail_lebs, avail;
225 avail_lebs = c->orph_lebs;
227 ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
232 * do_write_orph_node - write a node to the orphan head.
233 * @c: UBIFS file-system description object
234 * @len: length of node
235 * @atomic: write atomically
237 * This function writes a node to the orphan head from the orphan buffer. If
238 * %atomic is not zero, then the write is done atomically. On success, %0 is
239 * returned, otherwise a negative error code is returned.
241 static int do_write_orph_node(struct ubifs_info *c, int len, int atomic)
246 ubifs_assert(c->ohead_offs == 0);
247 ubifs_prepare_node(c, c->orph_buf, len, 1);
248 len = ALIGN(len, c->min_io_size);
249 err = ubifs_leb_change(c, c->ohead_lnum, c->orph_buf, len);
251 if (c->ohead_offs == 0) {
252 /* Ensure LEB has been unmapped */
253 err = ubifs_leb_unmap(c, c->ohead_lnum);
257 err = ubifs_write_node(c, c->orph_buf, len, c->ohead_lnum,
264 * write_orph_node - write an orphan node.
265 * @c: UBIFS file-system description object
266 * @atomic: write atomically
268 * This function builds an orphan node from the cnext list and writes it to the
269 * orphan head. On success, %0 is returned, otherwise a negative error code
272 static int write_orph_node(struct ubifs_info *c, int atomic)
274 struct ubifs_orphan *orphan, *cnext;
275 struct ubifs_orph_node *orph;
276 int gap, err, len, cnt, i;
278 ubifs_assert(c->cmt_orphans > 0);
279 gap = c->leb_size - c->ohead_offs;
280 if (gap < UBIFS_ORPH_NODE_SZ + sizeof(__le64)) {
284 if (c->ohead_lnum > c->orph_last) {
286 * We limit the number of orphans so that this should
289 ubifs_err("out of space in orphan area");
293 cnt = (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
294 if (cnt > c->cmt_orphans)
295 cnt = c->cmt_orphans;
296 len = UBIFS_ORPH_NODE_SZ + cnt * sizeof(__le64);
297 ubifs_assert(c->orph_buf);
299 orph->ch.node_type = UBIFS_ORPH_NODE;
300 spin_lock(&c->orphan_lock);
301 cnext = c->orph_cnext;
302 for (i = 0; i < cnt; i++) {
304 ubifs_assert(orphan->cmt);
305 orph->inos[i] = cpu_to_le64(orphan->inum);
307 cnext = orphan->cnext;
308 orphan->cnext = NULL;
310 c->orph_cnext = cnext;
311 c->cmt_orphans -= cnt;
312 spin_unlock(&c->orphan_lock);
314 orph->cmt_no = cpu_to_le64(c->cmt_no);
316 /* Mark the last node of the commit */
317 orph->cmt_no = cpu_to_le64((c->cmt_no) | (1ULL << 63));
318 ubifs_assert(c->ohead_offs + len <= c->leb_size);
319 ubifs_assert(c->ohead_lnum >= c->orph_first);
320 ubifs_assert(c->ohead_lnum <= c->orph_last);
321 err = do_write_orph_node(c, len, atomic);
322 c->ohead_offs += ALIGN(len, c->min_io_size);
323 c->ohead_offs = ALIGN(c->ohead_offs, 8);
328 * write_orph_nodes - write orphan nodes until there are no more to commit.
329 * @c: UBIFS file-system description object
330 * @atomic: write atomically
332 * This function writes orphan nodes for all the orphans to commit. On success,
333 * %0 is returned, otherwise a negative error code is returned.
335 static int write_orph_nodes(struct ubifs_info *c, int atomic)
339 while (c->cmt_orphans > 0) {
340 err = write_orph_node(c, atomic);
347 /* Unmap any unused LEBs after consolidation */
348 lnum = c->ohead_lnum + 1;
349 for (lnum = c->ohead_lnum + 1; lnum <= c->orph_last; lnum++) {
350 err = ubifs_leb_unmap(c, lnum);
359 * consolidate - consolidate the orphan area.
360 * @c: UBIFS file-system description object
362 * This function enables consolidation by putting all the orphans into the list
363 * to commit. The list is in the order that the orphans were added, and the
364 * LEBs are written atomically in order, so at no time can orphans be lost by
365 * an unclean unmount.
367 * This function returns %0 on success and a negative error code on failure.
369 static int consolidate(struct ubifs_info *c)
371 int tot_avail = tot_avail_orphs(c), err = 0;
373 spin_lock(&c->orphan_lock);
374 dbg_cmt("there is space for %d orphans and there are %d",
375 tot_avail, c->tot_orphans);
376 if (c->tot_orphans - c->new_orphans <= tot_avail) {
377 struct ubifs_orphan *orphan, **last;
380 /* Change the cnext list to include all non-new orphans */
381 last = &c->orph_cnext;
382 list_for_each_entry(orphan, &c->orph_list, list) {
387 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 * ubifs_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 int ubifs_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 orphan 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 orphan 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 %d:%d",
571 snod->type, sleb->lnum, snod->offs);
572 ubifs_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 orphan nodes will continue
584 * to 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 orphan node had a higher commit number and
593 * was flagged as the last written for that commit
594 * number. That makes this orphan node, out of date.
597 ubifs_err("out of order commit number %llu in orphan node at %d:%d",
598 cmt_no, sleb->lnum, snod->offs);
599 ubifs_dump_node(c, snod->node);
602 dbg_rcvry("out of date LEB %d", sleb->lnum);
610 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
611 for (i = 0; i < n; i++) {
612 inum = le64_to_cpu(orph->inos[i]);
613 dbg_rcvry("deleting orphaned inode %lu",
614 (unsigned long)inum);
615 err = ubifs_tnc_remove_ino(c, inum);
618 err = insert_dead_orphan(c, inum);
623 *last_cmt_no = cmt_no;
624 if (le64_to_cpu(orph->cmt_no) & (1ULL << 63)) {
625 dbg_rcvry("last orph node for commit %llu at %d:%d",
626 cmt_no, sleb->lnum, snod->offs);
636 * kill_orphans - remove all orphan inodes from the index.
637 * @c: UBIFS file-system description object
639 * If recovery is required, then orphan inodes recorded during the previous
640 * session (which ended with an unclean unmount) must be deleted from the index.
641 * This is done by updating the TNC, but since the index is not updated until
642 * the next commit, the LEBs where the orphan information is recorded are not
643 * erased until the next commit.
645 static int kill_orphans(struct ubifs_info *c)
647 unsigned long long last_cmt_no = 0;
648 int lnum, err = 0, outofdate = 0, last_flagged = 0;
650 c->ohead_lnum = c->orph_first;
652 /* Check no-orphans flag and skip this if no orphans */
654 dbg_rcvry("no orphans");
658 * Orph nodes always start at c->orph_first and are written to each
659 * successive LEB in turn. Generally unused LEBs will have been unmapped
660 * but may contain out of date orphan nodes if the unmap didn't go
661 * through. In addition, the last orphan node written for each commit is
662 * marked (top bit of orph->cmt_no is set to 1). It is possible that
663 * there are orphan nodes from the next commit (i.e. the commit did not
664 * complete successfully). In that case, no orphans will have been lost
665 * due to the way that orphans are written, and any orphans added will
666 * be valid orphans anyway and so can be deleted.
668 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
669 struct ubifs_scan_leb *sleb;
671 dbg_rcvry("LEB %d", lnum);
672 sleb = ubifs_scan(c, lnum, 0, c->sbuf, 1);
674 if (PTR_ERR(sleb) == -EUCLEAN)
675 sleb = ubifs_recover_leb(c, lnum, 0,
682 err = do_kill_orphans(c, sleb, &last_cmt_no, &outofdate,
684 if (err || outofdate) {
685 ubifs_scan_destroy(sleb);
689 c->ohead_lnum = lnum;
690 c->ohead_offs = sleb->endpt;
692 ubifs_scan_destroy(sleb);
698 * ubifs_mount_orphans - delete orphan inodes and erase LEBs that recorded them.
699 * @c: UBIFS file-system description object
700 * @unclean: indicates recovery from unclean unmount
701 * @read_only: indicates read only mount
703 * This function is called when mounting to erase orphans from the previous
704 * session. If UBIFS was not unmounted cleanly, then the inodes recorded as
705 * orphans are deleted.
707 int ubifs_mount_orphans(struct ubifs_info *c, int unclean, int read_only)
711 c->max_orphans = tot_avail_orphs(c);
714 c->orph_buf = vmalloc(c->leb_size);
720 err = kill_orphans(c);
722 err = ubifs_clear_orphans(c);
728 * Everything below is related to debugging.
731 struct check_orphan {
737 unsigned long last_ino;
738 unsigned long tot_inos;
739 unsigned long missing;
740 unsigned long long leaf_cnt;
741 struct ubifs_ino_node *node;
745 static int dbg_find_orphan(struct ubifs_info *c, ino_t inum)
747 struct ubifs_orphan *o;
750 spin_lock(&c->orphan_lock);
751 p = c->orph_tree.rb_node;
753 o = rb_entry(p, struct ubifs_orphan, rb);
756 else if (inum > o->inum)
759 spin_unlock(&c->orphan_lock);
763 spin_unlock(&c->orphan_lock);
767 static int dbg_ins_check_orphan(struct rb_root *root, ino_t inum)
769 struct check_orphan *orphan, *o;
770 struct rb_node **p, *parent = NULL;
772 orphan = kzalloc(sizeof(struct check_orphan), GFP_NOFS);
780 o = rb_entry(parent, struct check_orphan, rb);
783 else if (inum > o->inum)
790 rb_link_node(&orphan->rb, parent, p);
791 rb_insert_color(&orphan->rb, root);
795 static int dbg_find_check_orphan(struct rb_root *root, ino_t inum)
797 struct check_orphan *o;
802 o = rb_entry(p, struct check_orphan, rb);
805 else if (inum > o->inum)
813 static void dbg_free_check_tree(struct rb_root *root)
815 struct rb_node *this = root->rb_node;
816 struct check_orphan *o;
820 this = this->rb_left;
822 } else if (this->rb_right) {
823 this = this->rb_right;
826 o = rb_entry(this, struct check_orphan, rb);
827 this = rb_parent(this);
829 if (this->rb_left == &o->rb)
830 this->rb_left = NULL;
832 this->rb_right = NULL;
838 static int dbg_orphan_check(struct ubifs_info *c, struct ubifs_zbranch *zbr,
841 struct check_info *ci = priv;
845 inum = key_inum(c, &zbr->key);
846 if (inum != ci->last_ino) {
847 /* Lowest node type is the inode node, so it comes first */
848 if (key_type(c, &zbr->key) != UBIFS_INO_KEY)
849 ubifs_err("found orphan node ino %lu, type %d",
850 (unsigned long)inum, key_type(c, &zbr->key));
853 err = ubifs_tnc_read_node(c, zbr, ci->node);
855 ubifs_err("node read failed, error %d", err);
858 if (ci->node->nlink == 0)
859 /* Must be recorded as an orphan */
860 if (!dbg_find_check_orphan(&ci->root, inum) &&
861 !dbg_find_orphan(c, inum)) {
862 ubifs_err("missing orphan, ino %lu",
863 (unsigned long)inum);
871 static int dbg_read_orphans(struct check_info *ci, struct ubifs_scan_leb *sleb)
873 struct ubifs_scan_node *snod;
874 struct ubifs_orph_node *orph;
878 list_for_each_entry(snod, &sleb->nodes, list) {
880 if (snod->type != UBIFS_ORPH_NODE)
883 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
884 for (i = 0; i < n; i++) {
885 inum = le64_to_cpu(orph->inos[i]);
886 err = dbg_ins_check_orphan(&ci->root, inum);
894 static int dbg_scan_orphans(struct ubifs_info *c, struct check_info *ci)
899 /* Check no-orphans flag and skip this if no orphans */
903 buf = __vmalloc(c->leb_size, GFP_NOFS, PAGE_KERNEL);
905 ubifs_err("cannot allocate memory to check orphans");
909 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
910 struct ubifs_scan_leb *sleb;
912 sleb = ubifs_scan(c, lnum, 0, buf, 0);
918 err = dbg_read_orphans(ci, sleb);
919 ubifs_scan_destroy(sleb);
928 static int dbg_check_orphans(struct ubifs_info *c)
930 struct check_info ci;
933 if (!dbg_is_chk_orph(c))
941 ci.node = kmalloc(UBIFS_MAX_INO_NODE_SZ, GFP_NOFS);
943 ubifs_err("out of memory");
947 err = dbg_scan_orphans(c, &ci);
951 err = dbg_walk_index(c, &dbg_orphan_check, NULL, &ci);
953 ubifs_err("cannot scan TNC, error %d", err);
958 ubifs_err("%lu missing orphan(s)", ci.missing);
963 dbg_cmt("last inode number is %lu", ci.last_ino);
964 dbg_cmt("total number of inodes is %lu", ci.tot_inos);
965 dbg_cmt("total number of leaf nodes is %llu", ci.leaf_cnt);
968 dbg_free_check_tree(&ci.root);