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(c, "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);
137 o->dnext = c->orph_dnext;
139 spin_unlock(&c->orphan_lock);
140 dbg_gen("delete later ino %lu",
141 (unsigned long)inum);
144 rb_erase(p, &c->orph_tree);
148 list_del(&o->new_list);
151 spin_unlock(&c->orphan_lock);
153 dbg_gen("inum %lu", (unsigned long)inum);
157 spin_unlock(&c->orphan_lock);
158 ubifs_err(c, "missing orphan ino %lu", (unsigned long)inum);
163 * ubifs_orphan_start_commit - start commit of orphans.
164 * @c: UBIFS file-system description object
166 * Start commit of orphans.
168 int ubifs_orphan_start_commit(struct ubifs_info *c)
170 struct ubifs_orphan *orphan, **last;
172 spin_lock(&c->orphan_lock);
173 last = &c->orph_cnext;
174 list_for_each_entry(orphan, &c->orph_new, new_list) {
175 ubifs_assert(orphan->new);
176 ubifs_assert(!orphan->cmt);
180 last = &orphan->cnext;
183 c->cmt_orphans = c->new_orphans;
185 dbg_cmt("%d orphans to commit", c->cmt_orphans);
186 INIT_LIST_HEAD(&c->orph_new);
187 if (c->tot_orphans == 0)
191 spin_unlock(&c->orphan_lock);
196 * avail_orphs - calculate available space.
197 * @c: UBIFS file-system description object
199 * This function returns the number of orphans that can be written in the
202 static int avail_orphs(struct ubifs_info *c)
204 int avail_lebs, avail, gap;
206 avail_lebs = c->orph_lebs - (c->ohead_lnum - c->orph_first) - 1;
208 ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
209 gap = c->leb_size - c->ohead_offs;
210 if (gap >= UBIFS_ORPH_NODE_SZ + sizeof(__le64))
211 avail += (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
216 * tot_avail_orphs - calculate total space.
217 * @c: UBIFS file-system description object
219 * This function returns the number of orphans that can be written in half
220 * the total space. That leaves half the space for adding new orphans.
222 static int tot_avail_orphs(struct ubifs_info *c)
224 int avail_lebs, avail;
226 avail_lebs = c->orph_lebs;
228 ((c->leb_size - UBIFS_ORPH_NODE_SZ) / sizeof(__le64));
233 * do_write_orph_node - write a node to the orphan head.
234 * @c: UBIFS file-system description object
235 * @len: length of node
236 * @atomic: write atomically
238 * This function writes a node to the orphan head from the orphan buffer. If
239 * %atomic is not zero, then the write is done atomically. On success, %0 is
240 * returned, otherwise a negative error code is returned.
242 static int do_write_orph_node(struct ubifs_info *c, int len, int atomic)
247 ubifs_assert(c->ohead_offs == 0);
248 ubifs_prepare_node(c, c->orph_buf, len, 1);
249 len = ALIGN(len, c->min_io_size);
250 err = ubifs_leb_change(c, c->ohead_lnum, c->orph_buf, len);
252 if (c->ohead_offs == 0) {
253 /* Ensure LEB has been unmapped */
254 err = ubifs_leb_unmap(c, c->ohead_lnum);
258 err = ubifs_write_node(c, c->orph_buf, len, c->ohead_lnum,
265 * write_orph_node - write an orphan node.
266 * @c: UBIFS file-system description object
267 * @atomic: write atomically
269 * This function builds an orphan node from the cnext list and writes it to the
270 * orphan head. On success, %0 is returned, otherwise a negative error code
273 static int write_orph_node(struct ubifs_info *c, int atomic)
275 struct ubifs_orphan *orphan, *cnext;
276 struct ubifs_orph_node *orph;
277 int gap, err, len, cnt, i;
279 ubifs_assert(c->cmt_orphans > 0);
280 gap = c->leb_size - c->ohead_offs;
281 if (gap < UBIFS_ORPH_NODE_SZ + sizeof(__le64)) {
285 if (c->ohead_lnum > c->orph_last) {
287 * We limit the number of orphans so that this should
290 ubifs_err(c, "out of space in orphan area");
294 cnt = (gap - UBIFS_ORPH_NODE_SZ) / sizeof(__le64);
295 if (cnt > c->cmt_orphans)
296 cnt = c->cmt_orphans;
297 len = UBIFS_ORPH_NODE_SZ + cnt * sizeof(__le64);
298 ubifs_assert(c->orph_buf);
300 orph->ch.node_type = UBIFS_ORPH_NODE;
301 spin_lock(&c->orphan_lock);
302 cnext = c->orph_cnext;
303 for (i = 0; i < cnt; i++) {
305 ubifs_assert(orphan->cmt);
306 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 orphan nodes until there are no more to commit.
330 * @c: UBIFS file-system description object
331 * @atomic: write atomically
333 * This function writes orphan 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 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(c, "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 ubifs_assert(orphan->del);
451 rb_erase(&orphan->rb, &c->orph_tree);
452 list_del(&orphan->list);
454 dbg_gen("deleting orphan ino %lu", (unsigned long)orphan->inum);
457 c->orph_dnext = NULL;
458 spin_unlock(&c->orphan_lock);
462 * ubifs_orphan_end_commit - end commit of orphans.
463 * @c: UBIFS file-system description object
465 * End commit of orphans.
467 int ubifs_orphan_end_commit(struct ubifs_info *c)
471 if (c->cmt_orphans != 0) {
472 err = commit_orphans(c);
477 err = dbg_check_orphans(c);
482 * ubifs_clear_orphans - erase all LEBs used for orphans.
483 * @c: UBIFS file-system description object
485 * If recovery is not required, then the orphans from the previous session
486 * are not needed. This function locates the LEBs used to record
487 * orphans, and un-maps them.
489 int ubifs_clear_orphans(struct ubifs_info *c)
493 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
494 err = ubifs_leb_unmap(c, lnum);
498 c->ohead_lnum = c->orph_first;
504 * insert_dead_orphan - insert an orphan.
505 * @c: UBIFS file-system description object
506 * @inum: orphan inode number
508 * This function is a helper to the 'do_kill_orphans()' function. The orphan
509 * must be kept until the next commit, so it is added to the rb-tree and the
512 static int insert_dead_orphan(struct ubifs_info *c, ino_t inum)
514 struct ubifs_orphan *orphan, *o;
515 struct rb_node **p, *parent = NULL;
517 orphan = kzalloc(sizeof(struct ubifs_orphan), GFP_KERNEL);
522 p = &c->orph_tree.rb_node;
525 o = rb_entry(parent, struct ubifs_orphan, rb);
528 else if (inum > o->inum)
531 /* Already added - no problem */
537 rb_link_node(&orphan->rb, parent, p);
538 rb_insert_color(&orphan->rb, &c->orph_tree);
539 list_add_tail(&orphan->list, &c->orph_list);
541 orphan->dnext = c->orph_dnext;
542 c->orph_dnext = orphan;
543 dbg_mnt("ino %lu, new %d, tot %d", (unsigned long)inum,
544 c->new_orphans, c->tot_orphans);
549 * do_kill_orphans - remove orphan inodes from the index.
550 * @c: UBIFS file-system description object
552 * @last_cmt_no: cmt_no of last orphan node read is passed and returned here
553 * @outofdate: whether the LEB is out of date is returned here
554 * @last_flagged: whether the end orphan node is encountered
556 * This function is a helper to the 'kill_orphans()' function. It goes through
557 * every orphan node in a LEB and for every inode number recorded, removes
558 * all keys for that inode from the TNC.
560 static int do_kill_orphans(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
561 unsigned long long *last_cmt_no, int *outofdate,
564 struct ubifs_scan_node *snod;
565 struct ubifs_orph_node *orph;
566 unsigned long long cmt_no;
568 int i, n, err, first = 1;
570 list_for_each_entry(snod, &sleb->nodes, list) {
571 if (snod->type != UBIFS_ORPH_NODE) {
572 ubifs_err(c, "invalid node type %d in orphan area at %d:%d",
573 snod->type, sleb->lnum, snod->offs);
574 ubifs_dump_node(c, snod->node);
580 /* Check commit number */
581 cmt_no = le64_to_cpu(orph->cmt_no) & LLONG_MAX;
583 * The commit number on the master node may be less, because
584 * of a failed commit. If there are several failed commits in a
585 * row, the commit number written on orphan nodes will continue
586 * to increase (because the commit number is adjusted here) even
587 * though the commit number on the master node stays the same
588 * because the master node has not been re-written.
590 if (cmt_no > c->cmt_no)
592 if (cmt_no < *last_cmt_no && *last_flagged) {
594 * The last orphan node had a higher commit number and
595 * was flagged as the last written for that commit
596 * number. That makes this orphan node, out of date.
599 ubifs_err(c, "out of order commit number %llu in orphan node at %d:%d",
600 cmt_no, sleb->lnum, snod->offs);
601 ubifs_dump_node(c, snod->node);
604 dbg_rcvry("out of date LEB %d", sleb->lnum);
612 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
613 for (i = 0; i < n; i++) {
614 inum = le64_to_cpu(orph->inos[i]);
615 dbg_rcvry("deleting orphaned inode %lu",
616 (unsigned long)inum);
617 err = ubifs_tnc_remove_ino(c, inum);
620 err = insert_dead_orphan(c, inum);
625 *last_cmt_no = cmt_no;
626 if (le64_to_cpu(orph->cmt_no) & (1ULL << 63)) {
627 dbg_rcvry("last orph node for commit %llu at %d:%d",
628 cmt_no, sleb->lnum, snod->offs);
638 * kill_orphans - remove all orphan inodes from the index.
639 * @c: UBIFS file-system description object
641 * If recovery is required, then orphan inodes recorded during the previous
642 * session (which ended with an unclean unmount) must be deleted from the index.
643 * This is done by updating the TNC, but since the index is not updated until
644 * the next commit, the LEBs where the orphan information is recorded are not
645 * erased until the next commit.
647 static int kill_orphans(struct ubifs_info *c)
649 unsigned long long last_cmt_no = 0;
650 int lnum, err = 0, outofdate = 0, last_flagged = 0;
652 c->ohead_lnum = c->orph_first;
654 /* Check no-orphans flag and skip this if no orphans */
656 dbg_rcvry("no orphans");
660 * Orph nodes always start at c->orph_first and are written to each
661 * successive LEB in turn. Generally unused LEBs will have been unmapped
662 * but may contain out of date orphan nodes if the unmap didn't go
663 * through. In addition, the last orphan node written for each commit is
664 * marked (top bit of orph->cmt_no is set to 1). It is possible that
665 * there are orphan nodes from the next commit (i.e. the commit did not
666 * complete successfully). In that case, no orphans will have been lost
667 * due to the way that orphans are written, and any orphans added will
668 * be valid orphans anyway and so can be deleted.
670 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
671 struct ubifs_scan_leb *sleb;
673 dbg_rcvry("LEB %d", lnum);
674 sleb = ubifs_scan(c, lnum, 0, c->sbuf, 1);
676 if (PTR_ERR(sleb) == -EUCLEAN)
677 sleb = ubifs_recover_leb(c, lnum, 0,
684 err = do_kill_orphans(c, sleb, &last_cmt_no, &outofdate,
686 if (err || outofdate) {
687 ubifs_scan_destroy(sleb);
691 c->ohead_lnum = lnum;
692 c->ohead_offs = sleb->endpt;
694 ubifs_scan_destroy(sleb);
700 * ubifs_mount_orphans - delete orphan inodes and erase LEBs that recorded them.
701 * @c: UBIFS file-system description object
702 * @unclean: indicates recovery from unclean unmount
703 * @read_only: indicates read only mount
705 * This function is called when mounting to erase orphans from the previous
706 * session. If UBIFS was not unmounted cleanly, then the inodes recorded as
707 * orphans are deleted.
709 int ubifs_mount_orphans(struct ubifs_info *c, int unclean, int read_only)
713 c->max_orphans = tot_avail_orphs(c);
716 c->orph_buf = vmalloc(c->leb_size);
722 err = kill_orphans(c);
724 err = ubifs_clear_orphans(c);
730 * Everything below is related to debugging.
733 struct check_orphan {
739 unsigned long last_ino;
740 unsigned long tot_inos;
741 unsigned long missing;
742 unsigned long long leaf_cnt;
743 struct ubifs_ino_node *node;
747 static int dbg_find_orphan(struct ubifs_info *c, ino_t inum)
749 struct ubifs_orphan *o;
752 spin_lock(&c->orphan_lock);
753 p = c->orph_tree.rb_node;
755 o = rb_entry(p, struct ubifs_orphan, rb);
758 else if (inum > o->inum)
761 spin_unlock(&c->orphan_lock);
765 spin_unlock(&c->orphan_lock);
769 static int dbg_ins_check_orphan(struct rb_root *root, ino_t inum)
771 struct check_orphan *orphan, *o;
772 struct rb_node **p, *parent = NULL;
774 orphan = kzalloc(sizeof(struct check_orphan), GFP_NOFS);
782 o = rb_entry(parent, struct check_orphan, rb);
785 else if (inum > o->inum)
792 rb_link_node(&orphan->rb, parent, p);
793 rb_insert_color(&orphan->rb, root);
797 static int dbg_find_check_orphan(struct rb_root *root, ino_t inum)
799 struct check_orphan *o;
804 o = rb_entry(p, struct check_orphan, rb);
807 else if (inum > o->inum)
815 static void dbg_free_check_tree(struct rb_root *root)
817 struct check_orphan *o, *n;
819 rbtree_postorder_for_each_entry_safe(o, n, root, rb)
823 static int dbg_orphan_check(struct ubifs_info *c, struct ubifs_zbranch *zbr,
826 struct check_info *ci = priv;
830 inum = key_inum(c, &zbr->key);
831 if (inum != ci->last_ino) {
832 /* Lowest node type is the inode node, so it comes first */
833 if (key_type(c, &zbr->key) != UBIFS_INO_KEY)
834 ubifs_err(c, "found orphan node ino %lu, type %d",
835 (unsigned long)inum, key_type(c, &zbr->key));
838 err = ubifs_tnc_read_node(c, zbr, ci->node);
840 ubifs_err(c, "node read failed, error %d", err);
843 if (ci->node->nlink == 0)
844 /* Must be recorded as an orphan */
845 if (!dbg_find_check_orphan(&ci->root, inum) &&
846 !dbg_find_orphan(c, inum)) {
847 ubifs_err(c, "missing orphan, ino %lu",
848 (unsigned long)inum);
856 static int dbg_read_orphans(struct check_info *ci, struct ubifs_scan_leb *sleb)
858 struct ubifs_scan_node *snod;
859 struct ubifs_orph_node *orph;
863 list_for_each_entry(snod, &sleb->nodes, list) {
865 if (snod->type != UBIFS_ORPH_NODE)
868 n = (le32_to_cpu(orph->ch.len) - UBIFS_ORPH_NODE_SZ) >> 3;
869 for (i = 0; i < n; i++) {
870 inum = le64_to_cpu(orph->inos[i]);
871 err = dbg_ins_check_orphan(&ci->root, inum);
879 static int dbg_scan_orphans(struct ubifs_info *c, struct check_info *ci)
884 /* Check no-orphans flag and skip this if no orphans */
888 buf = __vmalloc(c->leb_size, GFP_NOFS, PAGE_KERNEL);
890 ubifs_err(c, "cannot allocate memory to check orphans");
894 for (lnum = c->orph_first; lnum <= c->orph_last; lnum++) {
895 struct ubifs_scan_leb *sleb;
897 sleb = ubifs_scan(c, lnum, 0, buf, 0);
903 err = dbg_read_orphans(ci, sleb);
904 ubifs_scan_destroy(sleb);
913 static int dbg_check_orphans(struct ubifs_info *c)
915 struct check_info ci;
918 if (!dbg_is_chk_orph(c))
926 ci.node = kmalloc(UBIFS_MAX_INO_NODE_SZ, GFP_NOFS);
928 ubifs_err(c, "out of memory");
932 err = dbg_scan_orphans(c, &ci);
936 err = dbg_walk_index(c, &dbg_orphan_check, NULL, &ci);
938 ubifs_err(c, "cannot scan TNC, error %d", err);
943 ubifs_err(c, "%lu missing orphan(s)", ci.missing);
948 dbg_cmt("last inode number is %lu", ci.last_ino);
949 dbg_cmt("total number of inodes is %lu", ci.tot_inos);
950 dbg_cmt("total number of leaf nodes is %llu", ci.leaf_cnt);
953 dbg_free_check_tree(&ci.root);