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 * Authors: Artem Bityutskiy (Битюцкий Артём)
24 * This file implements UBIFS journal.
26 * The journal consists of 2 parts - the log and bud LEBs. The log has fixed
27 * length and position, while a bud logical eraseblock is any LEB in the main
28 * area. Buds contain file system data - data nodes, inode nodes, etc. The log
29 * contains only references to buds and some other stuff like commit
30 * start node. The idea is that when we commit the journal, we do
31 * not copy the data, the buds just become indexed. Since after the commit the
32 * nodes in bud eraseblocks become leaf nodes of the file system index tree, we
33 * use term "bud". Analogy is obvious, bud eraseblocks contain nodes which will
34 * become leafs in the future.
36 * The journal is multi-headed because we want to write data to the journal as
37 * optimally as possible. It is nice to have nodes belonging to the same inode
38 * in one LEB, so we may write data owned by different inodes to different
39 * journal heads, although at present only one data head is used.
41 * For recovery reasons, the base head contains all inode nodes, all directory
42 * entry nodes and all truncate nodes. This means that the other heads contain
45 * Bud LEBs may be half-indexed. For example, if the bud was not full at the
46 * time of commit, the bud is retained to continue to be used in the journal,
47 * even though the "front" of the LEB is now indexed. In that case, the log
48 * reference contains the offset where the bud starts for the purposes of the
51 * The journal size has to be limited, because the larger is the journal, the
52 * longer it takes to mount UBIFS (scanning the journal) and the more memory it
53 * takes (indexing in the TNC).
55 * All the journal write operations like 'ubifs_jnl_update()' here, which write
56 * multiple UBIFS nodes to the journal at one go, are atomic with respect to
57 * unclean reboots. Should the unclean reboot happen, the recovery code drops
64 * zero_ino_node_unused - zero out unused fields of an on-flash inode node.
65 * @ino: the inode to zero out
67 static inline void zero_ino_node_unused(struct ubifs_ino_node *ino)
69 memset(ino->padding1, 0, 4);
70 memset(ino->padding2, 0, 26);
74 * zero_dent_node_unused - zero out unused fields of an on-flash directory
76 * @dent: the directory entry to zero out
78 static inline void zero_dent_node_unused(struct ubifs_dent_node *dent)
81 memset(dent->padding2, 0, 4);
85 * zero_data_node_unused - zero out unused fields of an on-flash data node.
86 * @data: the data node to zero out
88 static inline void zero_data_node_unused(struct ubifs_data_node *data)
90 memset(data->padding, 0, 2);
94 * zero_trun_node_unused - zero out unused fields of an on-flash truncation
96 * @trun: the truncation node to zero out
98 static inline void zero_trun_node_unused(struct ubifs_trun_node *trun)
100 memset(trun->padding, 0, 12);
104 * reserve_space - reserve space in the journal.
105 * @c: UBIFS file-system description object
106 * @jhead: journal head number
109 * This function reserves space in journal head @head. If the reservation
110 * succeeded, the journal head stays locked and later has to be unlocked using
111 * 'release_head()'. 'write_node()' and 'write_head()' functions also unlock
112 * it. Returns zero in case of success, %-EAGAIN if commit has to be done, and
113 * other negative error codes in case of other failures.
115 static int reserve_space(struct ubifs_info *c, int jhead, int len)
117 int err = 0, err1, retries = 0, avail, lnum, offs, squeeze;
118 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
121 * Typically, the base head has smaller nodes written to it, so it is
122 * better to try to allocate space at the ends of eraseblocks. This is
123 * what the squeeze parameter does.
125 ubifs_assert(!c->ro_media && !c->ro_mount);
126 squeeze = (jhead == BASEHD);
128 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
135 avail = c->leb_size - wbuf->offs - wbuf->used;
136 if (wbuf->lnum != -1 && avail >= len)
140 * Write buffer wasn't seek'ed or there is no enough space - look for an
141 * LEB with some empty space.
143 lnum = ubifs_find_free_space(c, len, &offs, squeeze);
152 * No free space, we have to run garbage collector to make
153 * some. But the write-buffer mutex has to be unlocked because
156 dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead));
157 mutex_unlock(&wbuf->io_mutex);
159 lnum = ubifs_garbage_collect(c, 0);
166 * GC could not make a free LEB. But someone else may
167 * have allocated new bud for this journal head,
168 * because we dropped @wbuf->io_mutex, so try once
171 dbg_jnl("GC couldn't make a free LEB for jhead %s",
174 dbg_jnl("retry (%d)", retries);
178 dbg_jnl("return -ENOSPC");
182 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
183 dbg_jnl("got LEB %d for jhead %s", lnum, dbg_jhead(jhead));
184 avail = c->leb_size - wbuf->offs - wbuf->used;
186 if (wbuf->lnum != -1 && avail >= len) {
188 * Someone else has switched the journal head and we have
189 * enough space now. This happens when more than one process is
190 * trying to write to the same journal head at the same time.
192 dbg_jnl("return LEB %d back, already have LEB %d:%d",
193 lnum, wbuf->lnum, wbuf->offs + wbuf->used);
194 err = ubifs_return_leb(c, lnum);
204 * Make sure we synchronize the write-buffer before we add the new bud
205 * to the log. Otherwise we may have a power cut after the log
206 * reference node for the last bud (@lnum) is written but before the
207 * write-buffer data are written to the next-to-last bud
208 * (@wbuf->lnum). And the effect would be that the recovery would see
209 * that there is corruption in the next-to-last bud.
211 err = ubifs_wbuf_sync_nolock(wbuf);
214 err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
217 err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs);
224 mutex_unlock(&wbuf->io_mutex);
228 /* An error occurred and the LEB has to be returned to lprops */
229 ubifs_assert(err < 0);
230 err1 = ubifs_return_leb(c, lnum);
231 if (err1 && err == -EAGAIN)
233 * Return original error code only if it is not %-EAGAIN,
234 * which is not really an error. Otherwise, return the error
235 * code of 'ubifs_return_leb()'.
238 mutex_unlock(&wbuf->io_mutex);
243 * write_node - write node to a journal head.
244 * @c: UBIFS file-system description object
245 * @jhead: journal head
246 * @node: node to write
248 * @lnum: LEB number written is returned here
249 * @offs: offset written is returned here
251 * This function writes a node to reserved space of journal head @jhead.
252 * Returns zero in case of success and a negative error code in case of
255 static int write_node(struct ubifs_info *c, int jhead, void *node, int len,
256 int *lnum, int *offs)
258 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
260 ubifs_assert(jhead != GCHD);
262 *lnum = c->jheads[jhead].wbuf.lnum;
263 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
265 dbg_jnl("jhead %s, LEB %d:%d, len %d",
266 dbg_jhead(jhead), *lnum, *offs, len);
267 ubifs_prepare_node(c, node, len, 0);
269 return ubifs_wbuf_write_nolock(wbuf, node, len);
273 * write_head - write data to a journal head.
274 * @c: UBIFS file-system description object
275 * @jhead: journal head
276 * @buf: buffer to write
277 * @len: length to write
278 * @lnum: LEB number written is returned here
279 * @offs: offset written is returned here
280 * @sync: non-zero if the write-buffer has to by synchronized
282 * This function is the same as 'write_node()' but it does not assume the
283 * buffer it is writing is a node, so it does not prepare it (which means
284 * initializing common header and calculating CRC).
286 static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
287 int *lnum, int *offs, int sync)
290 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
292 ubifs_assert(jhead != GCHD);
294 *lnum = c->jheads[jhead].wbuf.lnum;
295 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
296 dbg_jnl("jhead %s, LEB %d:%d, len %d",
297 dbg_jhead(jhead), *lnum, *offs, len);
299 err = ubifs_wbuf_write_nolock(wbuf, buf, len);
303 err = ubifs_wbuf_sync_nolock(wbuf);
308 * make_reservation - reserve journal space.
309 * @c: UBIFS file-system description object
310 * @jhead: journal head
311 * @len: how many bytes to reserve
313 * This function makes space reservation in journal head @jhead. The function
314 * takes the commit lock and locks the journal head, and the caller has to
315 * unlock the head and finish the reservation with 'finish_reservation()'.
316 * Returns zero in case of success and a negative error code in case of
319 * Note, the journal head may be unlocked as soon as the data is written, while
320 * the commit lock has to be released after the data has been added to the
323 static int make_reservation(struct ubifs_info *c, int jhead, int len)
325 int err, cmt_retries = 0, nospc_retries = 0;
328 down_read(&c->commit_sem);
329 err = reserve_space(c, jhead, len);
332 up_read(&c->commit_sem);
334 if (err == -ENOSPC) {
336 * GC could not make any progress. We should try to commit
337 * once because it could make some dirty space and GC would
338 * make progress, so make the error -EAGAIN so that the below
339 * will commit and re-try.
341 if (nospc_retries++ < 2) {
342 dbg_jnl("no space, retry");
347 * This means that the budgeting is incorrect. We always have
348 * to be able to write to the media, because all operations are
349 * budgeted. Deletions are not budgeted, though, but we reserve
350 * an extra LEB for them.
358 * -EAGAIN means that the journal is full or too large, or the above
359 * code wants to do one commit. Do this and re-try.
361 if (cmt_retries > 128) {
363 * This should not happen unless the journal size limitations
366 ubifs_err("stuck in space allocation");
369 } else if (cmt_retries > 32)
370 ubifs_warn("too many space allocation re-tries (%d)",
373 dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
377 err = ubifs_run_commit(c);
383 ubifs_err("cannot reserve %d bytes in jhead %d, error %d",
385 if (err == -ENOSPC) {
386 /* This are some budgeting problems, print useful information */
387 down_write(&c->commit_sem);
389 ubifs_dump_budg(c, &c->bi);
390 ubifs_dump_lprops(c);
391 cmt_retries = dbg_check_lprops(c);
392 up_write(&c->commit_sem);
398 * release_head - release a journal head.
399 * @c: UBIFS file-system description object
400 * @jhead: journal head
402 * This function releases journal head @jhead which was locked by
403 * the 'make_reservation()' function. It has to be called after each successful
404 * 'make_reservation()' invocation.
406 static inline void release_head(struct ubifs_info *c, int jhead)
408 mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
412 * finish_reservation - finish a reservation.
413 * @c: UBIFS file-system description object
415 * This function finishes journal space reservation. It must be called after
416 * 'make_reservation()'.
418 static void finish_reservation(struct ubifs_info *c)
420 up_read(&c->commit_sem);
424 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
427 static int get_dent_type(int mode)
429 switch (mode & S_IFMT) {
431 return UBIFS_ITYPE_REG;
433 return UBIFS_ITYPE_DIR;
435 return UBIFS_ITYPE_LNK;
437 return UBIFS_ITYPE_BLK;
439 return UBIFS_ITYPE_CHR;
441 return UBIFS_ITYPE_FIFO;
443 return UBIFS_ITYPE_SOCK;
451 * pack_inode - pack an inode node.
452 * @c: UBIFS file-system description object
453 * @ino: buffer in which to pack inode node
454 * @inode: inode to pack
455 * @last: indicates the last node of the group
457 static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
458 const struct inode *inode, int last)
460 int data_len = 0, last_reference = !inode->i_nlink;
461 struct ubifs_inode *ui = ubifs_inode(inode);
463 ino->ch.node_type = UBIFS_INO_NODE;
464 ino_key_init_flash(c, &ino->key, inode->i_ino);
465 ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
466 ino->atime_sec = cpu_to_le64(inode->i_atime.tv_sec);
467 ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
468 ino->ctime_sec = cpu_to_le64(inode->i_ctime.tv_sec);
469 ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
470 ino->mtime_sec = cpu_to_le64(inode->i_mtime.tv_sec);
471 ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
472 ino->uid = cpu_to_le32(i_uid_read(inode));
473 ino->gid = cpu_to_le32(i_gid_read(inode));
474 ino->mode = cpu_to_le32(inode->i_mode);
475 ino->flags = cpu_to_le32(ui->flags);
476 ino->size = cpu_to_le64(ui->ui_size);
477 ino->nlink = cpu_to_le32(inode->i_nlink);
478 ino->compr_type = cpu_to_le16(ui->compr_type);
479 ino->data_len = cpu_to_le32(ui->data_len);
480 ino->xattr_cnt = cpu_to_le32(ui->xattr_cnt);
481 ino->xattr_size = cpu_to_le32(ui->xattr_size);
482 ino->xattr_names = cpu_to_le32(ui->xattr_names);
483 zero_ino_node_unused(ino);
486 * Drop the attached data if this is a deletion inode, the data is not
489 if (!last_reference) {
490 memcpy(ino->data, ui->data, ui->data_len);
491 data_len = ui->data_len;
494 ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
498 * mark_inode_clean - mark UBIFS inode as clean.
499 * @c: UBIFS file-system description object
500 * @ui: UBIFS inode to mark as clean
502 * This helper function marks UBIFS inode @ui as clean by cleaning the
503 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
504 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
507 static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
510 ubifs_release_dirty_inode_budget(c, ui);
515 * ubifs_jnl_update - update inode.
516 * @c: UBIFS file-system description object
517 * @dir: parent inode or host inode in case of extended attributes
518 * @nm: directory entry name
519 * @inode: inode to update
520 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
521 * @xent: non-zero if the directory entry is an extended attribute entry
523 * This function updates an inode by writing a directory entry (or extended
524 * attribute entry), the inode itself, and the parent directory inode (or the
525 * host inode) to the journal.
527 * The function writes the host inode @dir last, which is important in case of
528 * extended attributes. Indeed, then we guarantee that if the host inode gets
529 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
530 * the extended attribute inode gets flushed too. And this is exactly what the
531 * user expects - synchronizing the host inode synchronizes its extended
532 * attributes. Similarly, this guarantees that if @dir is synchronized, its
533 * directory entry corresponding to @nm gets synchronized too.
535 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
536 * function synchronizes the write-buffer.
538 * This function marks the @dir and @inode inodes as clean and returns zero on
539 * success. In case of failure, a negative error code is returned.
541 int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
542 const struct qstr *nm, const struct inode *inode,
543 int deletion, int xent)
545 int err, dlen, ilen, len, lnum, ino_offs, dent_offs;
546 int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
547 int last_reference = !!(deletion && inode->i_nlink == 0);
548 struct ubifs_inode *ui = ubifs_inode(inode);
549 struct ubifs_inode *host_ui = ubifs_inode(dir);
550 struct ubifs_dent_node *dent;
551 struct ubifs_ino_node *ino;
552 union ubifs_key dent_key, ino_key;
554 dbg_jnl("ino %lu, dent '%.*s', data len %d in dir ino %lu",
555 inode->i_ino, nm->len, nm->name, ui->data_len, dir->i_ino);
556 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
558 dlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
559 ilen = UBIFS_INO_NODE_SZ;
562 * If the last reference to the inode is being deleted, then there is
563 * no need to attach and write inode data, it is being deleted anyway.
564 * And if the inode is being deleted, no need to synchronize
565 * write-buffer even if the inode is synchronous.
567 if (!last_reference) {
568 ilen += ui->data_len;
569 sync |= IS_SYNC(inode);
572 aligned_dlen = ALIGN(dlen, 8);
573 aligned_ilen = ALIGN(ilen, 8);
574 len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
575 dent = kmalloc(len, GFP_NOFS);
579 /* Make reservation before allocating sequence numbers */
580 err = make_reservation(c, BASEHD, len);
585 dent->ch.node_type = UBIFS_DENT_NODE;
586 dent_key_init(c, &dent_key, dir->i_ino, nm);
588 dent->ch.node_type = UBIFS_XENT_NODE;
589 xent_key_init(c, &dent_key, dir->i_ino, nm);
592 key_write(c, &dent_key, dent->key);
593 dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
594 dent->type = get_dent_type(inode->i_mode);
595 dent->nlen = cpu_to_le16(nm->len);
596 memcpy(dent->name, nm->name, nm->len);
597 dent->name[nm->len] = '\0';
598 zero_dent_node_unused(dent);
599 ubifs_prep_grp_node(c, dent, dlen, 0);
601 ino = (void *)dent + aligned_dlen;
602 pack_inode(c, ino, inode, 0);
603 ino = (void *)ino + aligned_ilen;
604 pack_inode(c, ino, dir, 1);
606 if (last_reference) {
607 err = ubifs_add_orphan(c, inode->i_ino);
609 release_head(c, BASEHD);
612 ui->del_cmtno = c->cmt_no;
615 err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
619 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
621 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
622 ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
624 release_head(c, BASEHD);
628 err = ubifs_tnc_remove_nm(c, &dent_key, nm);
631 err = ubifs_add_dirt(c, lnum, dlen);
633 err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
638 * Note, we do not remove the inode from TNC even if the last reference
639 * to it has just been deleted, because the inode may still be opened.
640 * Instead, the inode has been added to orphan lists and the orphan
641 * subsystem will take further care about it.
643 ino_key_init(c, &ino_key, inode->i_ino);
644 ino_offs = dent_offs + aligned_dlen;
645 err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
649 ino_key_init(c, &ino_key, dir->i_ino);
650 ino_offs += aligned_ilen;
651 err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, UBIFS_INO_NODE_SZ);
655 finish_reservation(c);
656 spin_lock(&ui->ui_lock);
657 ui->synced_i_size = ui->ui_size;
658 spin_unlock(&ui->ui_lock);
659 mark_inode_clean(c, ui);
660 mark_inode_clean(c, host_ui);
664 finish_reservation(c);
670 release_head(c, BASEHD);
673 ubifs_ro_mode(c, err);
675 ubifs_delete_orphan(c, inode->i_ino);
676 finish_reservation(c);
681 * ubifs_jnl_write_data - write a data node to the journal.
682 * @c: UBIFS file-system description object
683 * @inode: inode the data node belongs to
685 * @buf: buffer to write
686 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
688 * This function writes a data node to the journal. Returns %0 if the data node
689 * was successfully written, and a negative error code in case of failure.
691 int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
692 const union ubifs_key *key, const void *buf, int len)
694 struct ubifs_data_node *data;
695 int err, lnum, offs, compr_type, out_len;
696 int dlen = COMPRESSED_DATA_NODE_BUF_SZ, allocated = 1;
697 struct ubifs_inode *ui = ubifs_inode(inode);
699 dbg_jnlk(key, "ino %lu, blk %u, len %d, key ",
700 (unsigned long)key_inum(c, key), key_block(c, key), len);
701 ubifs_assert(len <= UBIFS_BLOCK_SIZE);
703 data = kmalloc(dlen, GFP_NOFS | __GFP_NOWARN);
706 * Fall-back to the write reserve buffer. Note, we might be
707 * currently on the memory reclaim path, when the kernel is
708 * trying to free some memory by writing out dirty pages. The
709 * write reserve buffer helps us to guarantee that we are
710 * always able to write the data.
713 mutex_lock(&c->write_reserve_mutex);
714 data = c->write_reserve_buf;
717 data->ch.node_type = UBIFS_DATA_NODE;
718 key_write(c, key, &data->key);
719 data->size = cpu_to_le32(len);
720 zero_data_node_unused(data);
722 if (!(ui->flags & UBIFS_COMPR_FL))
723 /* Compression is disabled for this inode */
724 compr_type = UBIFS_COMPR_NONE;
726 compr_type = ui->compr_type;
728 out_len = dlen - UBIFS_DATA_NODE_SZ;
729 ubifs_compress(buf, len, &data->data, &out_len, &compr_type);
730 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
732 dlen = UBIFS_DATA_NODE_SZ + out_len;
733 data->compr_type = cpu_to_le16(compr_type);
735 /* Make reservation before allocating sequence numbers */
736 err = make_reservation(c, DATAHD, dlen);
740 err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
743 ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
744 release_head(c, DATAHD);
746 err = ubifs_tnc_add(c, key, lnum, offs, dlen);
750 finish_reservation(c);
752 mutex_unlock(&c->write_reserve_mutex);
758 release_head(c, DATAHD);
760 ubifs_ro_mode(c, err);
761 finish_reservation(c);
764 mutex_unlock(&c->write_reserve_mutex);
771 * ubifs_jnl_write_inode - flush inode to the journal.
772 * @c: UBIFS file-system description object
773 * @inode: inode to flush
775 * This function writes inode @inode to the journal. If the inode is
776 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
777 * success and a negative error code in case of failure.
779 int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
782 struct ubifs_ino_node *ino;
783 struct ubifs_inode *ui = ubifs_inode(inode);
784 int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
786 dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
789 * If the inode is being deleted, do not write the attached data. No
790 * need to synchronize the write-buffer either.
792 if (!last_reference) {
794 sync = IS_SYNC(inode);
796 ino = kmalloc(len, GFP_NOFS);
800 /* Make reservation before allocating sequence numbers */
801 err = make_reservation(c, BASEHD, len);
805 pack_inode(c, ino, inode, 1);
806 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
810 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
812 release_head(c, BASEHD);
814 if (last_reference) {
815 err = ubifs_tnc_remove_ino(c, inode->i_ino);
818 ubifs_delete_orphan(c, inode->i_ino);
819 err = ubifs_add_dirt(c, lnum, len);
823 ino_key_init(c, &key, inode->i_ino);
824 err = ubifs_tnc_add(c, &key, lnum, offs, len);
829 finish_reservation(c);
830 spin_lock(&ui->ui_lock);
831 ui->synced_i_size = ui->ui_size;
832 spin_unlock(&ui->ui_lock);
837 release_head(c, BASEHD);
839 ubifs_ro_mode(c, err);
840 finish_reservation(c);
847 * ubifs_jnl_delete_inode - delete an inode.
848 * @c: UBIFS file-system description object
849 * @inode: inode to delete
851 * This function deletes inode @inode which includes removing it from orphans,
852 * deleting it from TNC and, in some cases, writing a deletion inode to the
855 * When regular file inodes are unlinked or a directory inode is removed, the
856 * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
857 * direntry to the media, and adds the inode to orphans. After this, when the
858 * last reference to this inode has been dropped, this function is called. In
859 * general, it has to write one more deletion inode to the media, because if
860 * a commit happened between 'ubifs_jnl_update()' and
861 * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
862 * anymore, and in fact it might not be on the flash anymore, because it might
863 * have been garbage-collected already. And for optimization reasons UBIFS does
864 * not read the orphan area if it has been unmounted cleanly, so it would have
865 * no indication in the journal that there is a deleted inode which has to be
868 * However, if there was no commit between 'ubifs_jnl_update()' and
869 * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
870 * inode to the media for the second time. And this is quite a typical case.
872 * This function returns zero in case of success and a negative error code in
875 int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
878 struct ubifs_inode *ui = ubifs_inode(inode);
880 ubifs_assert(inode->i_nlink == 0);
882 if (ui->del_cmtno != c->cmt_no)
883 /* A commit happened for sure */
884 return ubifs_jnl_write_inode(c, inode);
886 down_read(&c->commit_sem);
888 * Check commit number again, because the first test has been done
889 * without @c->commit_sem, so a commit might have happened.
891 if (ui->del_cmtno != c->cmt_no) {
892 up_read(&c->commit_sem);
893 return ubifs_jnl_write_inode(c, inode);
896 err = ubifs_tnc_remove_ino(c, inode->i_ino);
898 ubifs_ro_mode(c, err);
900 ubifs_delete_orphan(c, inode->i_ino);
901 up_read(&c->commit_sem);
906 * ubifs_jnl_rename - rename a directory entry.
907 * @c: UBIFS file-system description object
908 * @old_dir: parent inode of directory entry to rename
909 * @old_dentry: directory entry to rename
910 * @new_dir: parent inode of directory entry to rename
911 * @new_dentry: new directory entry (or directory entry to replace)
912 * @sync: non-zero if the write-buffer has to be synchronized
914 * This function implements the re-name operation which may involve writing up
915 * to 3 inodes and 2 directory entries. It marks the written inodes as clean
916 * and returns zero on success. In case of failure, a negative error code is
919 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
920 const struct dentry *old_dentry,
921 const struct inode *new_dir,
922 const struct dentry *new_dentry, int sync)
926 struct ubifs_dent_node *dent, *dent2;
927 int err, dlen1, dlen2, ilen, lnum, offs, len;
928 const struct inode *old_inode = old_dentry->d_inode;
929 const struct inode *new_inode = new_dentry->d_inode;
930 int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
931 int last_reference = !!(new_inode && new_inode->i_nlink == 0);
932 int move = (old_dir != new_dir);
933 struct ubifs_inode *uninitialized_var(new_ui);
935 dbg_jnl("dent '%pd' in dir ino %lu to dent '%pd' in dir ino %lu",
936 old_dentry, old_dir->i_ino, new_dentry, new_dir->i_ino);
937 ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
938 ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
939 ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
940 ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
942 dlen1 = UBIFS_DENT_NODE_SZ + new_dentry->d_name.len + 1;
943 dlen2 = UBIFS_DENT_NODE_SZ + old_dentry->d_name.len + 1;
945 new_ui = ubifs_inode(new_inode);
946 ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
947 ilen = UBIFS_INO_NODE_SZ;
949 ilen += new_ui->data_len;
953 aligned_dlen1 = ALIGN(dlen1, 8);
954 aligned_dlen2 = ALIGN(dlen2, 8);
955 len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
956 if (old_dir != new_dir)
958 dent = kmalloc(len, GFP_NOFS);
962 /* Make reservation before allocating sequence numbers */
963 err = make_reservation(c, BASEHD, len);
968 dent->ch.node_type = UBIFS_DENT_NODE;
969 dent_key_init_flash(c, &dent->key, new_dir->i_ino, &new_dentry->d_name);
970 dent->inum = cpu_to_le64(old_inode->i_ino);
971 dent->type = get_dent_type(old_inode->i_mode);
972 dent->nlen = cpu_to_le16(new_dentry->d_name.len);
973 memcpy(dent->name, new_dentry->d_name.name, new_dentry->d_name.len);
974 dent->name[new_dentry->d_name.len] = '\0';
975 zero_dent_node_unused(dent);
976 ubifs_prep_grp_node(c, dent, dlen1, 0);
978 /* Make deletion dent */
979 dent2 = (void *)dent + aligned_dlen1;
980 dent2->ch.node_type = UBIFS_DENT_NODE;
981 dent_key_init_flash(c, &dent2->key, old_dir->i_ino,
982 &old_dentry->d_name);
984 dent2->type = DT_UNKNOWN;
985 dent2->nlen = cpu_to_le16(old_dentry->d_name.len);
986 memcpy(dent2->name, old_dentry->d_name.name, old_dentry->d_name.len);
987 dent2->name[old_dentry->d_name.len] = '\0';
988 zero_dent_node_unused(dent2);
989 ubifs_prep_grp_node(c, dent2, dlen2, 0);
991 p = (void *)dent2 + aligned_dlen2;
993 pack_inode(c, p, new_inode, 0);
998 pack_inode(c, p, old_dir, 1);
1000 pack_inode(c, p, old_dir, 0);
1001 p += ALIGN(plen, 8);
1002 pack_inode(c, p, new_dir, 1);
1005 if (last_reference) {
1006 err = ubifs_add_orphan(c, new_inode->i_ino);
1008 release_head(c, BASEHD);
1011 new_ui->del_cmtno = c->cmt_no;
1014 err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
1018 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1020 ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1021 ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1023 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1026 release_head(c, BASEHD);
1028 dent_key_init(c, &key, new_dir->i_ino, &new_dentry->d_name);
1029 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &new_dentry->d_name);
1033 err = ubifs_add_dirt(c, lnum, dlen2);
1037 dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
1038 err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
1042 offs += aligned_dlen1 + aligned_dlen2;
1044 ino_key_init(c, &key, new_inode->i_ino);
1045 err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
1048 offs += ALIGN(ilen, 8);
1051 ino_key_init(c, &key, old_dir->i_ino);
1052 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1056 if (old_dir != new_dir) {
1057 offs += ALIGN(plen, 8);
1058 ino_key_init(c, &key, new_dir->i_ino);
1059 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1064 finish_reservation(c);
1066 mark_inode_clean(c, new_ui);
1067 spin_lock(&new_ui->ui_lock);
1068 new_ui->synced_i_size = new_ui->ui_size;
1069 spin_unlock(&new_ui->ui_lock);
1071 mark_inode_clean(c, ubifs_inode(old_dir));
1073 mark_inode_clean(c, ubifs_inode(new_dir));
1078 release_head(c, BASEHD);
1080 ubifs_ro_mode(c, err);
1082 ubifs_delete_orphan(c, new_inode->i_ino);
1084 finish_reservation(c);
1091 * recomp_data_node - re-compress a truncated data node.
1092 * @dn: data node to re-compress
1093 * @new_len: new length
1095 * This function is used when an inode is truncated and the last data node of
1096 * the inode has to be re-compressed and re-written.
1098 static int recomp_data_node(struct ubifs_data_node *dn, int *new_len)
1101 int err, len, compr_type, out_len;
1103 out_len = le32_to_cpu(dn->size);
1104 buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS);
1108 len = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1109 compr_type = le16_to_cpu(dn->compr_type);
1110 err = ubifs_decompress(&dn->data, len, buf, &out_len, compr_type);
1114 ubifs_compress(buf, *new_len, &dn->data, &out_len, &compr_type);
1115 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1116 dn->compr_type = cpu_to_le16(compr_type);
1117 dn->size = cpu_to_le32(*new_len);
1118 *new_len = UBIFS_DATA_NODE_SZ + out_len;
1125 * ubifs_jnl_truncate - update the journal for a truncation.
1126 * @c: UBIFS file-system description object
1127 * @inode: inode to truncate
1128 * @old_size: old size
1129 * @new_size: new size
1131 * When the size of a file decreases due to truncation, a truncation node is
1132 * written, the journal tree is updated, and the last data block is re-written
1133 * if it has been affected. The inode is also updated in order to synchronize
1134 * the new inode size.
1136 * This function marks the inode as clean and returns zero on success. In case
1137 * of failure, a negative error code is returned.
1139 int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1140 loff_t old_size, loff_t new_size)
1142 union ubifs_key key, to_key;
1143 struct ubifs_ino_node *ino;
1144 struct ubifs_trun_node *trun;
1145 struct ubifs_data_node *uninitialized_var(dn);
1146 int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1147 struct ubifs_inode *ui = ubifs_inode(inode);
1148 ino_t inum = inode->i_ino;
1151 dbg_jnl("ino %lu, size %lld -> %lld",
1152 (unsigned long)inum, old_size, new_size);
1153 ubifs_assert(!ui->data_len);
1154 ubifs_assert(S_ISREG(inode->i_mode));
1155 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1157 sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1158 UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1159 ino = kmalloc(sz, GFP_NOFS);
1163 trun = (void *)ino + UBIFS_INO_NODE_SZ;
1164 trun->ch.node_type = UBIFS_TRUN_NODE;
1165 trun->inum = cpu_to_le32(inum);
1166 trun->old_size = cpu_to_le64(old_size);
1167 trun->new_size = cpu_to_le64(new_size);
1168 zero_trun_node_unused(trun);
1170 dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1172 /* Get last data block so it can be truncated */
1173 dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1174 blk = new_size >> UBIFS_BLOCK_SHIFT;
1175 data_key_init(c, &key, inum, blk);
1176 dbg_jnlk(&key, "last block key ");
1177 err = ubifs_tnc_lookup(c, &key, dn);
1179 dlen = 0; /* Not found (so it is a hole) */
1183 if (le32_to_cpu(dn->size) <= dlen)
1184 dlen = 0; /* Nothing to do */
1186 int compr_type = le16_to_cpu(dn->compr_type);
1188 if (compr_type != UBIFS_COMPR_NONE) {
1189 err = recomp_data_node(dn, &dlen);
1193 dn->size = cpu_to_le32(dlen);
1194 dlen += UBIFS_DATA_NODE_SZ;
1196 zero_data_node_unused(dn);
1201 /* Must make reservation before allocating sequence numbers */
1202 len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1205 err = make_reservation(c, BASEHD, len);
1209 pack_inode(c, ino, inode, 0);
1210 ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1212 ubifs_prep_grp_node(c, dn, dlen, 1);
1214 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1218 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1219 release_head(c, BASEHD);
1222 sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1223 err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1228 ino_key_init(c, &key, inum);
1229 err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1233 err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1237 bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1238 blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1239 data_key_init(c, &key, inum, blk);
1241 bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1242 blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
1243 data_key_init(c, &to_key, inum, blk);
1245 err = ubifs_tnc_remove_range(c, &key, &to_key);
1249 finish_reservation(c);
1250 spin_lock(&ui->ui_lock);
1251 ui->synced_i_size = ui->ui_size;
1252 spin_unlock(&ui->ui_lock);
1253 mark_inode_clean(c, ui);
1258 release_head(c, BASEHD);
1260 ubifs_ro_mode(c, err);
1261 finish_reservation(c);
1269 * ubifs_jnl_delete_xattr - delete an extended attribute.
1270 * @c: UBIFS file-system description object
1272 * @inode: extended attribute inode
1273 * @nm: extended attribute entry name
1275 * This function delete an extended attribute which is very similar to
1276 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1277 * updates the target inode. Returns zero in case of success and a negative
1278 * error code in case of failure.
1280 int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1281 const struct inode *inode, const struct qstr *nm)
1283 int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1284 struct ubifs_dent_node *xent;
1285 struct ubifs_ino_node *ino;
1286 union ubifs_key xent_key, key1, key2;
1287 int sync = IS_DIRSYNC(host);
1288 struct ubifs_inode *host_ui = ubifs_inode(host);
1290 dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d",
1291 host->i_ino, inode->i_ino, nm->name,
1292 ubifs_inode(inode)->data_len);
1293 ubifs_assert(inode->i_nlink == 0);
1294 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1297 * Since we are deleting the inode, we do not bother to attach any data
1298 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1300 xlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
1301 aligned_xlen = ALIGN(xlen, 8);
1302 hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1303 len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1305 xent = kmalloc(len, GFP_NOFS);
1309 /* Make reservation before allocating sequence numbers */
1310 err = make_reservation(c, BASEHD, len);
1316 xent->ch.node_type = UBIFS_XENT_NODE;
1317 xent_key_init(c, &xent_key, host->i_ino, nm);
1318 key_write(c, &xent_key, xent->key);
1320 xent->type = get_dent_type(inode->i_mode);
1321 xent->nlen = cpu_to_le16(nm->len);
1322 memcpy(xent->name, nm->name, nm->len);
1323 xent->name[nm->len] = '\0';
1324 zero_dent_node_unused(xent);
1325 ubifs_prep_grp_node(c, xent, xlen, 0);
1327 ino = (void *)xent + aligned_xlen;
1328 pack_inode(c, ino, inode, 0);
1329 ino = (void *)ino + UBIFS_INO_NODE_SZ;
1330 pack_inode(c, ino, host, 1);
1332 err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1334 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1335 release_head(c, BASEHD);
1340 /* Remove the extended attribute entry from TNC */
1341 err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1344 err = ubifs_add_dirt(c, lnum, xlen);
1349 * Remove all nodes belonging to the extended attribute inode from TNC.
1350 * Well, there actually must be only one node - the inode itself.
1352 lowest_ino_key(c, &key1, inode->i_ino);
1353 highest_ino_key(c, &key2, inode->i_ino);
1354 err = ubifs_tnc_remove_range(c, &key1, &key2);
1357 err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1361 /* And update TNC with the new host inode position */
1362 ino_key_init(c, &key1, host->i_ino);
1363 err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1367 finish_reservation(c);
1368 spin_lock(&host_ui->ui_lock);
1369 host_ui->synced_i_size = host_ui->ui_size;
1370 spin_unlock(&host_ui->ui_lock);
1371 mark_inode_clean(c, host_ui);
1375 ubifs_ro_mode(c, err);
1376 finish_reservation(c);
1381 * ubifs_jnl_change_xattr - change an extended attribute.
1382 * @c: UBIFS file-system description object
1383 * @inode: extended attribute inode
1386 * This function writes the updated version of an extended attribute inode and
1387 * the host inode to the journal (to the base head). The host inode is written
1388 * after the extended attribute inode in order to guarantee that the extended
1389 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1390 * consequently, the write-buffer is synchronized. This function returns zero
1391 * in case of success and a negative error code in case of failure.
1393 int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1394 const struct inode *host)
1396 int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1397 struct ubifs_inode *host_ui = ubifs_inode(host);
1398 struct ubifs_ino_node *ino;
1399 union ubifs_key key;
1400 int sync = IS_DIRSYNC(host);
1402 dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1403 ubifs_assert(host->i_nlink > 0);
1404 ubifs_assert(inode->i_nlink > 0);
1405 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1407 len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1408 len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1409 aligned_len1 = ALIGN(len1, 8);
1410 aligned_len = aligned_len1 + ALIGN(len2, 8);
1412 ino = kmalloc(aligned_len, GFP_NOFS);
1416 /* Make reservation before allocating sequence numbers */
1417 err = make_reservation(c, BASEHD, aligned_len);
1421 pack_inode(c, ino, host, 0);
1422 pack_inode(c, (void *)ino + aligned_len1, inode, 1);
1424 err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1425 if (!sync && !err) {
1426 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1428 ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1429 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1431 release_head(c, BASEHD);
1435 ino_key_init(c, &key, host->i_ino);
1436 err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1440 ino_key_init(c, &key, inode->i_ino);
1441 err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1445 finish_reservation(c);
1446 spin_lock(&host_ui->ui_lock);
1447 host_ui->synced_i_size = host_ui->ui_size;
1448 spin_unlock(&host_ui->ui_lock);
1449 mark_inode_clean(c, host_ui);
1454 ubifs_ro_mode(c, err);
1455 finish_reservation(c);