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
, free
, 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 squeeze
= (jhead
== BASEHD
);
127 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
134 avail
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
135 if (wbuf
->lnum
!= -1 && avail
>= len
)
139 * Write buffer wasn't seek'ed or there is no enough space - look for an
140 * LEB with some empty space.
142 lnum
= ubifs_find_free_space(c
, len
, &free
, squeeze
);
144 /* Found an LEB, add it to the journal head */
145 offs
= c
->leb_size
- free
;
146 err
= ubifs_add_bud_to_log(c
, jhead
, lnum
, offs
);
149 /* A new bud was successfully allocated and added to the log */
158 * No free space, we have to run garbage collector to make
159 * some. But the write-buffer mutex has to be unlocked because
162 dbg_jnl("no free space jhead %d, run GC", jhead
);
163 mutex_unlock(&wbuf
->io_mutex
);
165 lnum
= ubifs_garbage_collect(c
, 0);
172 * GC could not make a free LEB. But someone else may
173 * have allocated new bud for this journal head,
174 * because we dropped @wbuf->io_mutex, so try once
177 dbg_jnl("GC couldn't make a free LEB for jhead %d", jhead
);
179 dbg_jnl("retry (%d)", retries
);
183 dbg_jnl("return -ENOSPC");
187 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
188 dbg_jnl("got LEB %d for jhead %d", lnum
, jhead
);
189 avail
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
191 if (wbuf
->lnum
!= -1 && avail
>= len
) {
193 * Someone else has switched the journal head and we have
194 * enough space now. This happens when more then one process is
195 * trying to write to the same journal head at the same time.
197 dbg_jnl("return LEB %d back, already have LEB %d:%d",
198 lnum
, wbuf
->lnum
, wbuf
->offs
+ wbuf
->used
);
199 err
= ubifs_return_leb(c
, lnum
);
205 err
= ubifs_add_bud_to_log(c
, jhead
, lnum
, 0);
211 err
= ubifs_wbuf_seek_nolock(wbuf
, lnum
, offs
, UBI_SHORTTERM
);
218 mutex_unlock(&wbuf
->io_mutex
);
222 /* An error occurred and the LEB has to be returned to lprops */
223 ubifs_assert(err
< 0);
224 err1
= ubifs_return_leb(c
, lnum
);
225 if (err1
&& err
== -EAGAIN
)
227 * Return original error code only if it is not %-EAGAIN,
228 * which is not really an error. Otherwise, return the error
229 * code of 'ubifs_return_leb()'.
232 mutex_unlock(&wbuf
->io_mutex
);
237 * write_node - write node to a journal head.
238 * @c: UBIFS file-system description object
239 * @jhead: journal head
240 * @node: node to write
242 * @lnum: LEB number written is returned here
243 * @offs: offset written is returned here
245 * This function writes a node to reserved space of journal head @jhead.
246 * Returns zero in case of success and a negative error code in case of
249 static int write_node(struct ubifs_info
*c
, int jhead
, void *node
, int len
,
250 int *lnum
, int *offs
)
252 struct ubifs_wbuf
*wbuf
= &c
->jheads
[jhead
].wbuf
;
254 ubifs_assert(jhead
!= GCHD
);
256 *lnum
= c
->jheads
[jhead
].wbuf
.lnum
;
257 *offs
= c
->jheads
[jhead
].wbuf
.offs
+ c
->jheads
[jhead
].wbuf
.used
;
259 dbg_jnl("jhead %d, LEB %d:%d, len %d", jhead
, *lnum
, *offs
, len
);
260 ubifs_prepare_node(c
, node
, len
, 0);
262 return ubifs_wbuf_write_nolock(wbuf
, node
, len
);
266 * write_head - write data to a journal head.
267 * @c: UBIFS file-system description object
268 * @jhead: journal head
269 * @buf: buffer to write
270 * @len: length to write
271 * @lnum: LEB number written is returned here
272 * @offs: offset written is returned here
273 * @sync: non-zero if the write-buffer has to by synchronized
275 * This function is the same as 'write_node()' but it does not assume the
276 * buffer it is writing is a node, so it does not prepare it (which means
277 * initializing common header and calculating CRC).
279 static int write_head(struct ubifs_info
*c
, int jhead
, void *buf
, int len
,
280 int *lnum
, int *offs
, int sync
)
283 struct ubifs_wbuf
*wbuf
= &c
->jheads
[jhead
].wbuf
;
285 ubifs_assert(jhead
!= GCHD
);
287 *lnum
= c
->jheads
[jhead
].wbuf
.lnum
;
288 *offs
= c
->jheads
[jhead
].wbuf
.offs
+ c
->jheads
[jhead
].wbuf
.used
;
289 dbg_jnl("jhead %d, LEB %d:%d, len %d", jhead
, *lnum
, *offs
, len
);
291 err
= ubifs_wbuf_write_nolock(wbuf
, buf
, len
);
295 err
= ubifs_wbuf_sync_nolock(wbuf
);
300 * make_reservation - reserve journal space.
301 * @c: UBIFS file-system description object
302 * @jhead: journal head
303 * @len: how many bytes to reserve
305 * This function makes space reservation in journal head @jhead. The function
306 * takes the commit lock and locks the journal head, and the caller has to
307 * unlock the head and finish the reservation with 'finish_reservation()'.
308 * Returns zero in case of success and a negative error code in case of
311 * Note, the journal head may be unlocked as soon as the data is written, while
312 * the commit lock has to be released after the data has been added to the
315 static int make_reservation(struct ubifs_info
*c
, int jhead
, int len
)
317 int err
, cmt_retries
= 0, nospc_retries
= 0;
320 down_read(&c
->commit_sem
);
321 err
= reserve_space(c
, jhead
, len
);
324 up_read(&c
->commit_sem
);
326 if (err
== -ENOSPC
) {
328 * GC could not make any progress. We should try to commit
329 * once because it could make some dirty space and GC would
330 * make progress, so make the error -EAGAIN so that the below
331 * will commit and re-try.
333 if (nospc_retries
++ < 2) {
334 dbg_jnl("no space, retry");
339 * This means that the budgeting is incorrect. We always have
340 * to be able to write to the media, because all operations are
341 * budgeted. Deletions are not budgeted, though, but we reserve
342 * an extra LEB for them.
350 * -EAGAIN means that the journal is full or too large, or the above
351 * code wants to do one commit. Do this and re-try.
353 if (cmt_retries
> 128) {
355 * This should not happen unless the journal size limitations
358 ubifs_err("stuck in space allocation");
361 } else if (cmt_retries
> 32)
362 ubifs_warn("too many space allocation re-tries (%d)",
365 dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
369 err
= ubifs_run_commit(c
);
375 ubifs_err("cannot reserve %d bytes in jhead %d, error %d",
377 if (err
== -ENOSPC
) {
378 /* This are some budgeting problems, print useful information */
379 down_write(&c
->commit_sem
);
380 spin_lock(&c
->space_lock
);
383 spin_unlock(&c
->space_lock
);
385 cmt_retries
= dbg_check_lprops(c
);
386 up_write(&c
->commit_sem
);
392 * release_head - release a journal head.
393 * @c: UBIFS file-system description object
394 * @jhead: journal head
396 * This function releases journal head @jhead which was locked by
397 * the 'make_reservation()' function. It has to be called after each successful
398 * 'make_reservation()' invocation.
400 static inline void release_head(struct ubifs_info
*c
, int jhead
)
402 mutex_unlock(&c
->jheads
[jhead
].wbuf
.io_mutex
);
406 * finish_reservation - finish a reservation.
407 * @c: UBIFS file-system description object
409 * This function finishes journal space reservation. It must be called after
410 * 'make_reservation()'.
412 static void finish_reservation(struct ubifs_info
*c
)
414 up_read(&c
->commit_sem
);
418 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
421 static int get_dent_type(int mode
)
423 switch (mode
& S_IFMT
) {
425 return UBIFS_ITYPE_REG
;
427 return UBIFS_ITYPE_DIR
;
429 return UBIFS_ITYPE_LNK
;
431 return UBIFS_ITYPE_BLK
;
433 return UBIFS_ITYPE_CHR
;
435 return UBIFS_ITYPE_FIFO
;
437 return UBIFS_ITYPE_SOCK
;
445 * pack_inode - pack an inode node.
446 * @c: UBIFS file-system description object
447 * @ino: buffer in which to pack inode node
448 * @inode: inode to pack
449 * @last: indicates the last node of the group
450 * @last_reference: non-zero if this is a deletion inode
452 static void pack_inode(struct ubifs_info
*c
, struct ubifs_ino_node
*ino
,
453 const struct inode
*inode
, int last
,
457 struct ubifs_inode
*ui
= ubifs_inode(inode
);
459 ino
->ch
.node_type
= UBIFS_INO_NODE
;
460 ino_key_init_flash(c
, &ino
->key
, inode
->i_ino
);
461 ino
->creat_sqnum
= cpu_to_le64(ui
->creat_sqnum
);
462 ino
->atime_sec
= cpu_to_le64(inode
->i_atime
.tv_sec
);
463 ino
->atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
464 ino
->ctime_sec
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
465 ino
->ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
466 ino
->mtime_sec
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
467 ino
->mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
468 ino
->uid
= cpu_to_le32(inode
->i_uid
);
469 ino
->gid
= cpu_to_le32(inode
->i_gid
);
470 ino
->mode
= cpu_to_le32(inode
->i_mode
);
471 ino
->flags
= cpu_to_le32(ui
->flags
);
472 ino
->size
= cpu_to_le64(ui
->ui_size
);
473 ino
->nlink
= cpu_to_le32(inode
->i_nlink
);
474 ino
->compr_type
= cpu_to_le16(ui
->compr_type
);
475 ino
->data_len
= cpu_to_le32(ui
->data_len
);
476 ino
->xattr_cnt
= cpu_to_le32(ui
->xattr_cnt
);
477 ino
->xattr_size
= cpu_to_le32(ui
->xattr_size
);
478 ino
->xattr_names
= cpu_to_le32(ui
->xattr_names
);
479 zero_ino_node_unused(ino
);
482 * Drop the attached data if this is a deletion inode, the data is not
485 if (!last_reference
) {
486 memcpy(ino
->data
, ui
->data
, ui
->data_len
);
487 data_len
= ui
->data_len
;
490 ubifs_prep_grp_node(c
, ino
, UBIFS_INO_NODE_SZ
+ data_len
, last
);
494 * mark_inode_clean - mark UBIFS inode as clean.
495 * @c: UBIFS file-system description object
496 * @ui: UBIFS inode to mark as clean
498 * This helper function marks UBIFS inode @ui as clean by cleaning the
499 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
500 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
503 static void mark_inode_clean(struct ubifs_info
*c
, struct ubifs_inode
*ui
)
506 ubifs_release_dirty_inode_budget(c
, ui
);
511 * ubifs_jnl_update - update inode.
512 * @c: UBIFS file-system description object
513 * @dir: parent inode or host inode in case of extended attributes
514 * @nm: directory entry name
515 * @inode: inode to update
516 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
517 * @xent: non-zero if the directory entry is an extended attribute entry
519 * This function updates an inode by writing a directory entry (or extended
520 * attribute entry), the inode itself, and the parent directory inode (or the
521 * host inode) to the journal.
523 * The function writes the host inode @dir last, which is important in case of
524 * extended attributes. Indeed, then we guarantee that if the host inode gets
525 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
526 * the extended attribute inode gets flushed too. And this is exactly what the
527 * user expects - synchronizing the host inode synchronizes its extended
528 * attributes. Similarly, this guarantees that if @dir is synchronized, its
529 * directory entry corresponding to @nm gets synchronized too.
531 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
532 * function synchronizes the write-buffer.
534 * This function marks the @dir and @inode inodes as clean and returns zero on
535 * success. In case of failure, a negative error code is returned.
537 int ubifs_jnl_update(struct ubifs_info
*c
, const struct inode
*dir
,
538 const struct qstr
*nm
, const struct inode
*inode
,
539 int deletion
, int xent
)
541 int err
, dlen
, ilen
, len
, lnum
, ino_offs
, dent_offs
;
542 int aligned_dlen
, aligned_ilen
, sync
= IS_DIRSYNC(dir
);
543 int last_reference
= !!(deletion
&& inode
->i_nlink
== 0);
544 struct ubifs_inode
*ui
= ubifs_inode(inode
);
545 struct ubifs_inode
*dir_ui
= ubifs_inode(dir
);
546 struct ubifs_dent_node
*dent
;
547 struct ubifs_ino_node
*ino
;
548 union ubifs_key dent_key
, ino_key
;
550 dbg_jnl("ino %lu, dent '%.*s', data len %d in dir ino %lu",
551 inode
->i_ino
, nm
->len
, nm
->name
, ui
->data_len
, dir
->i_ino
);
552 ubifs_assert(dir_ui
->data_len
== 0);
553 ubifs_assert(mutex_is_locked(&dir_ui
->ui_mutex
));
555 dlen
= UBIFS_DENT_NODE_SZ
+ nm
->len
+ 1;
556 ilen
= UBIFS_INO_NODE_SZ
;
559 * If the last reference to the inode is being deleted, then there is
560 * no need to attach and write inode data, it is being deleted anyway.
561 * And if the inode is being deleted, no need to synchronize
562 * write-buffer even if the inode is synchronous.
564 if (!last_reference
) {
565 ilen
+= ui
->data_len
;
566 sync
|= IS_SYNC(inode
);
569 aligned_dlen
= ALIGN(dlen
, 8);
570 aligned_ilen
= ALIGN(ilen
, 8);
571 len
= aligned_dlen
+ aligned_ilen
+ UBIFS_INO_NODE_SZ
;
572 dent
= kmalloc(len
, GFP_NOFS
);
576 /* Make reservation before allocating sequence numbers */
577 err
= make_reservation(c
, BASEHD
, len
);
582 dent
->ch
.node_type
= UBIFS_DENT_NODE
;
583 dent_key_init(c
, &dent_key
, dir
->i_ino
, nm
);
585 dent
->ch
.node_type
= UBIFS_XENT_NODE
;
586 xent_key_init(c
, &dent_key
, dir
->i_ino
, nm
);
589 key_write(c
, &dent_key
, dent
->key
);
590 dent
->inum
= deletion
? 0 : cpu_to_le64(inode
->i_ino
);
591 dent
->type
= get_dent_type(inode
->i_mode
);
592 dent
->nlen
= cpu_to_le16(nm
->len
);
593 memcpy(dent
->name
, nm
->name
, nm
->len
);
594 dent
->name
[nm
->len
] = '\0';
595 zero_dent_node_unused(dent
);
596 ubifs_prep_grp_node(c
, dent
, dlen
, 0);
598 ino
= (void *)dent
+ aligned_dlen
;
599 pack_inode(c
, ino
, inode
, 0, last_reference
);
600 ino
= (void *)ino
+ aligned_ilen
;
601 pack_inode(c
, ino
, dir
, 1, 0);
603 if (last_reference
) {
604 err
= ubifs_add_orphan(c
, inode
->i_ino
);
606 release_head(c
, BASEHD
);
611 err
= write_head(c
, BASEHD
, dent
, len
, &lnum
, &dent_offs
, sync
);
615 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
617 ubifs_wbuf_add_ino_nolock(wbuf
, inode
->i_ino
);
618 ubifs_wbuf_add_ino_nolock(wbuf
, dir
->i_ino
);
620 release_head(c
, BASEHD
);
624 err
= ubifs_tnc_remove_nm(c
, &dent_key
, nm
);
627 err
= ubifs_add_dirt(c
, lnum
, dlen
);
629 err
= ubifs_tnc_add_nm(c
, &dent_key
, lnum
, dent_offs
, dlen
, nm
);
634 * Note, we do not remove the inode from TNC even if the last reference
635 * to it has just been deleted, because the inode may still be opened.
636 * Instead, the inode has been added to orphan lists and the orphan
637 * subsystem will take further care about it.
639 ino_key_init(c
, &ino_key
, inode
->i_ino
);
640 ino_offs
= dent_offs
+ aligned_dlen
;
641 err
= ubifs_tnc_add(c
, &ino_key
, lnum
, ino_offs
, ilen
);
645 ino_key_init(c
, &ino_key
, dir
->i_ino
);
646 ino_offs
+= aligned_ilen
;
647 err
= ubifs_tnc_add(c
, &ino_key
, lnum
, ino_offs
, UBIFS_INO_NODE_SZ
);
651 finish_reservation(c
);
652 spin_lock(&ui
->ui_lock
);
653 ui
->synced_i_size
= ui
->ui_size
;
654 spin_unlock(&ui
->ui_lock
);
655 mark_inode_clean(c
, ui
);
656 mark_inode_clean(c
, dir_ui
);
660 finish_reservation(c
);
666 release_head(c
, BASEHD
);
668 ubifs_ro_mode(c
, err
);
670 ubifs_delete_orphan(c
, inode
->i_ino
);
671 finish_reservation(c
);
676 * ubifs_jnl_write_data - write a data node to the journal.
677 * @c: UBIFS file-system description object
678 * @inode: inode the data node belongs to
680 * @buf: buffer to write
681 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
683 * This function writes a data node to the journal. Returns %0 if the data node
684 * was successfully written, and a negative error code in case of failure.
686 int ubifs_jnl_write_data(struct ubifs_info
*c
, const struct inode
*inode
,
687 const union ubifs_key
*key
, const void *buf
, int len
)
689 struct ubifs_data_node
*data
;
690 int err
, lnum
, offs
, compr_type
, out_len
;
691 int dlen
= UBIFS_DATA_NODE_SZ
+ UBIFS_BLOCK_SIZE
* WORST_COMPR_FACTOR
;
692 struct ubifs_inode
*ui
= ubifs_inode(inode
);
694 dbg_jnl("ino %lu, blk %u, len %d, key %s", key_inum(c
, key
),
695 key_block(c
, key
), len
, DBGKEY(key
));
696 ubifs_assert(len
<= UBIFS_BLOCK_SIZE
);
698 data
= kmalloc(dlen
, GFP_NOFS
);
702 data
->ch
.node_type
= UBIFS_DATA_NODE
;
703 key_write(c
, key
, &data
->key
);
704 data
->size
= cpu_to_le32(len
);
705 zero_data_node_unused(data
);
707 if (!(ui
->flags
&& UBIFS_COMPR_FL
))
708 /* Compression is disabled for this inode */
709 compr_type
= UBIFS_COMPR_NONE
;
711 compr_type
= ui
->compr_type
;
713 out_len
= dlen
- UBIFS_DATA_NODE_SZ
;
714 ubifs_compress(buf
, len
, &data
->data
, &out_len
, &compr_type
);
715 ubifs_assert(out_len
<= UBIFS_BLOCK_SIZE
);
717 dlen
= UBIFS_DATA_NODE_SZ
+ out_len
;
718 data
->compr_type
= cpu_to_le16(compr_type
);
720 /* Make reservation before allocating sequence numbers */
721 err
= make_reservation(c
, DATAHD
, dlen
);
725 err
= write_node(c
, DATAHD
, data
, dlen
, &lnum
, &offs
);
728 ubifs_wbuf_add_ino_nolock(&c
->jheads
[DATAHD
].wbuf
, key_inum(c
, key
));
729 release_head(c
, DATAHD
);
731 err
= ubifs_tnc_add(c
, key
, lnum
, offs
, dlen
);
735 finish_reservation(c
);
740 release_head(c
, DATAHD
);
742 ubifs_ro_mode(c
, err
);
743 finish_reservation(c
);
750 * ubifs_jnl_write_inode - flush inode to the journal.
751 * @c: UBIFS file-system description object
752 * @inode: inode to flush
754 * This function writes inode @inode to the journal. If the inode is
755 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
756 * success and a negative error code in case of failure.
758 int ubifs_jnl_write_inode(struct ubifs_info
*c
, const struct inode
*inode
)
761 struct ubifs_ino_node
*ino
;
762 struct ubifs_inode
*ui
= ubifs_inode(inode
);
763 int sync
= 0, len
= UBIFS_INO_NODE_SZ
, last_reference
= !inode
->i_nlink
;
765 dbg_jnl("ino %lu, nlink %u", inode
->i_ino
, inode
->i_nlink
);
768 * If the inode is being deleted, do not write the attached data. No
769 * need to synchronize the write-buffer either.
771 if (!last_reference
) {
773 sync
= IS_SYNC(inode
);
775 ino
= kmalloc(len
, GFP_NOFS
);
779 /* Make reservation before allocating sequence numbers */
780 err
= make_reservation(c
, BASEHD
, len
);
784 pack_inode(c
, ino
, inode
, 1, last_reference
);
785 err
= write_head(c
, BASEHD
, ino
, len
, &lnum
, &offs
, sync
);
789 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
,
791 release_head(c
, BASEHD
);
793 if (last_reference
) {
794 err
= ubifs_tnc_remove_ino(c
, inode
->i_ino
);
797 ubifs_delete_orphan(c
, inode
->i_ino
);
798 err
= ubifs_add_dirt(c
, lnum
, len
);
802 ino_key_init(c
, &key
, inode
->i_ino
);
803 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, len
);
808 finish_reservation(c
);
809 spin_lock(&ui
->ui_lock
);
810 ui
->synced_i_size
= ui
->ui_size
;
811 spin_unlock(&ui
->ui_lock
);
816 release_head(c
, BASEHD
);
818 ubifs_ro_mode(c
, err
);
819 finish_reservation(c
);
826 * ubifs_jnl_rename - rename a directory entry.
827 * @c: UBIFS file-system description object
828 * @old_dir: parent inode of directory entry to rename
829 * @old_dentry: directory entry to rename
830 * @new_dir: parent inode of directory entry to rename
831 * @new_dentry: new directory entry (or directory entry to replace)
832 * @sync: non-zero if the write-buffer has to be synchronized
834 * This function implements the re-name operation which may involve writing up
835 * to 3 inodes and 2 directory entries. It marks the written inodes as clean
836 * and returns zero on success. In case of failure, a negative error code is
839 int ubifs_jnl_rename(struct ubifs_info
*c
, const struct inode
*old_dir
,
840 const struct dentry
*old_dentry
,
841 const struct inode
*new_dir
,
842 const struct dentry
*new_dentry
, int sync
)
846 struct ubifs_dent_node
*dent
, *dent2
;
847 int err
, dlen1
, dlen2
, ilen
, lnum
, offs
, len
;
848 const struct inode
*old_inode
= old_dentry
->d_inode
;
849 const struct inode
*new_inode
= new_dentry
->d_inode
;
850 int aligned_dlen1
, aligned_dlen2
, plen
= UBIFS_INO_NODE_SZ
;
851 int last_reference
= !!(new_inode
&& new_inode
->i_nlink
== 0);
852 int move
= (old_dir
!= new_dir
);
853 struct ubifs_inode
*uninitialized_var(new_ui
);
855 dbg_jnl("dent '%.*s' in dir ino %lu to dent '%.*s' in dir ino %lu",
856 old_dentry
->d_name
.len
, old_dentry
->d_name
.name
,
857 old_dir
->i_ino
, new_dentry
->d_name
.len
,
858 new_dentry
->d_name
.name
, new_dir
->i_ino
);
859 ubifs_assert(ubifs_inode(old_dir
)->data_len
== 0);
860 ubifs_assert(ubifs_inode(new_dir
)->data_len
== 0);
861 ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir
)->ui_mutex
));
862 ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir
)->ui_mutex
));
864 dlen1
= UBIFS_DENT_NODE_SZ
+ new_dentry
->d_name
.len
+ 1;
865 dlen2
= UBIFS_DENT_NODE_SZ
+ old_dentry
->d_name
.len
+ 1;
867 new_ui
= ubifs_inode(new_inode
);
868 ubifs_assert(mutex_is_locked(&new_ui
->ui_mutex
));
869 ilen
= UBIFS_INO_NODE_SZ
;
871 ilen
+= new_ui
->data_len
;
875 aligned_dlen1
= ALIGN(dlen1
, 8);
876 aligned_dlen2
= ALIGN(dlen2
, 8);
877 len
= aligned_dlen1
+ aligned_dlen2
+ ALIGN(ilen
, 8) + ALIGN(plen
, 8);
878 if (old_dir
!= new_dir
)
880 dent
= kmalloc(len
, GFP_NOFS
);
884 /* Make reservation before allocating sequence numbers */
885 err
= make_reservation(c
, BASEHD
, len
);
890 dent
->ch
.node_type
= UBIFS_DENT_NODE
;
891 dent_key_init_flash(c
, &dent
->key
, new_dir
->i_ino
, &new_dentry
->d_name
);
892 dent
->inum
= cpu_to_le64(old_inode
->i_ino
);
893 dent
->type
= get_dent_type(old_inode
->i_mode
);
894 dent
->nlen
= cpu_to_le16(new_dentry
->d_name
.len
);
895 memcpy(dent
->name
, new_dentry
->d_name
.name
, new_dentry
->d_name
.len
);
896 dent
->name
[new_dentry
->d_name
.len
] = '\0';
897 zero_dent_node_unused(dent
);
898 ubifs_prep_grp_node(c
, dent
, dlen1
, 0);
900 /* Make deletion dent */
901 dent2
= (void *)dent
+ aligned_dlen1
;
902 dent2
->ch
.node_type
= UBIFS_DENT_NODE
;
903 dent_key_init_flash(c
, &dent2
->key
, old_dir
->i_ino
,
904 &old_dentry
->d_name
);
906 dent2
->type
= DT_UNKNOWN
;
907 dent2
->nlen
= cpu_to_le16(old_dentry
->d_name
.len
);
908 memcpy(dent2
->name
, old_dentry
->d_name
.name
, old_dentry
->d_name
.len
);
909 dent2
->name
[old_dentry
->d_name
.len
] = '\0';
910 zero_dent_node_unused(dent2
);
911 ubifs_prep_grp_node(c
, dent2
, dlen2
, 0);
913 p
= (void *)dent2
+ aligned_dlen2
;
915 pack_inode(c
, p
, new_inode
, 0, last_reference
);
920 pack_inode(c
, p
, old_dir
, 1, 0);
922 pack_inode(c
, p
, old_dir
, 0, 0);
924 pack_inode(c
, p
, new_dir
, 1, 0);
927 if (last_reference
) {
928 err
= ubifs_add_orphan(c
, new_inode
->i_ino
);
930 release_head(c
, BASEHD
);
935 err
= write_head(c
, BASEHD
, dent
, len
, &lnum
, &offs
, sync
);
939 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
941 ubifs_wbuf_add_ino_nolock(wbuf
, new_dir
->i_ino
);
942 ubifs_wbuf_add_ino_nolock(wbuf
, old_dir
->i_ino
);
944 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
,
947 release_head(c
, BASEHD
);
949 dent_key_init(c
, &key
, new_dir
->i_ino
, &new_dentry
->d_name
);
950 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen1
, &new_dentry
->d_name
);
954 err
= ubifs_add_dirt(c
, lnum
, dlen2
);
958 dent_key_init(c
, &key
, old_dir
->i_ino
, &old_dentry
->d_name
);
959 err
= ubifs_tnc_remove_nm(c
, &key
, &old_dentry
->d_name
);
963 offs
+= aligned_dlen1
+ aligned_dlen2
;
965 ino_key_init(c
, &key
, new_inode
->i_ino
);
966 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, ilen
);
969 offs
+= ALIGN(ilen
, 8);
972 ino_key_init(c
, &key
, old_dir
->i_ino
);
973 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
);
977 if (old_dir
!= new_dir
) {
978 offs
+= ALIGN(plen
, 8);
979 ino_key_init(c
, &key
, new_dir
->i_ino
);
980 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
);
985 finish_reservation(c
);
987 mark_inode_clean(c
, new_ui
);
988 spin_lock(&new_ui
->ui_lock
);
989 new_ui
->synced_i_size
= new_ui
->ui_size
;
990 spin_unlock(&new_ui
->ui_lock
);
992 mark_inode_clean(c
, ubifs_inode(old_dir
));
994 mark_inode_clean(c
, ubifs_inode(new_dir
));
999 release_head(c
, BASEHD
);
1001 ubifs_ro_mode(c
, err
);
1003 ubifs_delete_orphan(c
, new_inode
->i_ino
);
1005 finish_reservation(c
);
1012 * recomp_data_node - re-compress a truncated data node.
1013 * @dn: data node to re-compress
1014 * @new_len: new length
1016 * This function is used when an inode is truncated and the last data node of
1017 * the inode has to be re-compressed and re-written.
1019 static int recomp_data_node(struct ubifs_data_node
*dn
, int *new_len
)
1022 int err
, len
, compr_type
, out_len
;
1024 out_len
= le32_to_cpu(dn
->size
);
1025 buf
= kmalloc(out_len
* WORST_COMPR_FACTOR
, GFP_NOFS
);
1029 len
= le32_to_cpu(dn
->ch
.len
) - UBIFS_DATA_NODE_SZ
;
1030 compr_type
= le16_to_cpu(dn
->compr_type
);
1031 err
= ubifs_decompress(&dn
->data
, len
, buf
, &out_len
, compr_type
);
1035 ubifs_compress(buf
, *new_len
, &dn
->data
, &out_len
, &compr_type
);
1036 ubifs_assert(out_len
<= UBIFS_BLOCK_SIZE
);
1037 dn
->compr_type
= cpu_to_le16(compr_type
);
1038 dn
->size
= cpu_to_le32(*new_len
);
1039 *new_len
= UBIFS_DATA_NODE_SZ
+ out_len
;
1046 * ubifs_jnl_truncate - update the journal for a truncation.
1047 * @c: UBIFS file-system description object
1048 * @inode: inode to truncate
1049 * @old_size: old size
1050 * @new_size: new size
1052 * When the size of a file decreases due to truncation, a truncation node is
1053 * written, the journal tree is updated, and the last data block is re-written
1054 * if it has been affected. The inode is also updated in order to synchronize
1055 * the new inode size.
1057 * This function marks the inode as clean and returns zero on success. In case
1058 * of failure, a negative error code is returned.
1060 int ubifs_jnl_truncate(struct ubifs_info
*c
, const struct inode
*inode
,
1061 loff_t old_size
, loff_t new_size
)
1063 union ubifs_key key
, to_key
;
1064 struct ubifs_ino_node
*ino
;
1065 struct ubifs_trun_node
*trun
;
1066 struct ubifs_data_node
*uninitialized_var(dn
);
1067 int err
, dlen
, len
, lnum
, offs
, bit
, sz
, sync
= IS_SYNC(inode
);
1068 struct ubifs_inode
*ui
= ubifs_inode(inode
);
1069 ino_t inum
= inode
->i_ino
;
1072 dbg_jnl("ino %lu, size %lld -> %lld", inum
, old_size
, new_size
);
1073 ubifs_assert(!ui
->data_len
);
1074 ubifs_assert(S_ISREG(inode
->i_mode
));
1075 ubifs_assert(mutex_is_locked(&ui
->ui_mutex
));
1077 sz
= UBIFS_TRUN_NODE_SZ
+ UBIFS_INO_NODE_SZ
+
1078 UBIFS_MAX_DATA_NODE_SZ
* WORST_COMPR_FACTOR
;
1079 ino
= kmalloc(sz
, GFP_NOFS
);
1083 trun
= (void *)ino
+ UBIFS_INO_NODE_SZ
;
1084 trun
->ch
.node_type
= UBIFS_TRUN_NODE
;
1085 trun
->inum
= cpu_to_le32(inum
);
1086 trun
->old_size
= cpu_to_le64(old_size
);
1087 trun
->new_size
= cpu_to_le64(new_size
);
1088 zero_trun_node_unused(trun
);
1090 dlen
= new_size
& (UBIFS_BLOCK_SIZE
- 1);
1092 /* Get last data block so it can be truncated */
1093 dn
= (void *)trun
+ UBIFS_TRUN_NODE_SZ
;
1094 blk
= new_size
>> UBIFS_BLOCK_SHIFT
;
1095 data_key_init(c
, &key
, inum
, blk
);
1096 dbg_jnl("last block key %s", DBGKEY(&key
));
1097 err
= ubifs_tnc_lookup(c
, &key
, dn
);
1099 dlen
= 0; /* Not found (so it is a hole) */
1103 if (le32_to_cpu(dn
->size
) <= dlen
)
1104 dlen
= 0; /* Nothing to do */
1106 int compr_type
= le16_to_cpu(dn
->compr_type
);
1108 if (compr_type
!= UBIFS_COMPR_NONE
) {
1109 err
= recomp_data_node(dn
, &dlen
);
1113 dn
->size
= cpu_to_le32(dlen
);
1114 dlen
+= UBIFS_DATA_NODE_SZ
;
1116 zero_data_node_unused(dn
);
1121 /* Must make reservation before allocating sequence numbers */
1122 len
= UBIFS_TRUN_NODE_SZ
+ UBIFS_INO_NODE_SZ
;
1125 err
= make_reservation(c
, BASEHD
, len
);
1129 pack_inode(c
, ino
, inode
, 0, 0);
1130 ubifs_prep_grp_node(c
, trun
, UBIFS_TRUN_NODE_SZ
, dlen
? 0 : 1);
1132 ubifs_prep_grp_node(c
, dn
, dlen
, 1);
1134 err
= write_head(c
, BASEHD
, ino
, len
, &lnum
, &offs
, sync
);
1138 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
, inum
);
1139 release_head(c
, BASEHD
);
1142 sz
= offs
+ UBIFS_INO_NODE_SZ
+ UBIFS_TRUN_NODE_SZ
;
1143 err
= ubifs_tnc_add(c
, &key
, lnum
, sz
, dlen
);
1148 ino_key_init(c
, &key
, inum
);
1149 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, UBIFS_INO_NODE_SZ
);
1153 err
= ubifs_add_dirt(c
, lnum
, UBIFS_TRUN_NODE_SZ
);
1157 bit
= new_size
& (UBIFS_BLOCK_SIZE
- 1);
1158 blk
= (new_size
>> UBIFS_BLOCK_SHIFT
) + (bit
? 1 : 0);
1159 data_key_init(c
, &key
, inum
, blk
);
1161 bit
= old_size
& (UBIFS_BLOCK_SIZE
- 1);
1162 blk
= (old_size
>> UBIFS_BLOCK_SHIFT
) - (bit
? 0: 1);
1163 data_key_init(c
, &to_key
, inum
, blk
);
1165 err
= ubifs_tnc_remove_range(c
, &key
, &to_key
);
1169 finish_reservation(c
);
1170 spin_lock(&ui
->ui_lock
);
1171 ui
->synced_i_size
= ui
->ui_size
;
1172 spin_unlock(&ui
->ui_lock
);
1173 mark_inode_clean(c
, ui
);
1178 release_head(c
, BASEHD
);
1180 ubifs_ro_mode(c
, err
);
1181 finish_reservation(c
);
1187 #ifdef CONFIG_UBIFS_FS_XATTR
1190 * ubifs_jnl_delete_xattr - delete an extended attribute.
1191 * @c: UBIFS file-system description object
1193 * @inode: extended attribute inode
1194 * @nm: extended attribute entry name
1196 * This function delete an extended attribute which is very similar to
1197 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1198 * updates the target inode. Returns zero in case of success and a negative
1199 * error code in case of failure.
1201 int ubifs_jnl_delete_xattr(struct ubifs_info
*c
, const struct inode
*host
,
1202 const struct inode
*inode
, const struct qstr
*nm
)
1204 int err
, xlen
, hlen
, len
, lnum
, xent_offs
, aligned_xlen
;
1205 struct ubifs_dent_node
*xent
;
1206 struct ubifs_ino_node
*ino
;
1207 union ubifs_key xent_key
, key1
, key2
;
1208 int sync
= IS_DIRSYNC(host
);
1209 struct ubifs_inode
*host_ui
= ubifs_inode(host
);
1211 dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d",
1212 host
->i_ino
, inode
->i_ino
, nm
->name
,
1213 ubifs_inode(inode
)->data_len
);
1214 ubifs_assert(inode
->i_nlink
== 0);
1215 ubifs_assert(mutex_is_locked(&host_ui
->ui_mutex
));
1218 * Since we are deleting the inode, we do not bother to attach any data
1219 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1221 xlen
= UBIFS_DENT_NODE_SZ
+ nm
->len
+ 1;
1222 aligned_xlen
= ALIGN(xlen
, 8);
1223 hlen
= host_ui
->data_len
+ UBIFS_INO_NODE_SZ
;
1224 len
= aligned_xlen
+ UBIFS_INO_NODE_SZ
+ ALIGN(hlen
, 8);
1226 xent
= kmalloc(len
, GFP_NOFS
);
1230 /* Make reservation before allocating sequence numbers */
1231 err
= make_reservation(c
, BASEHD
, len
);
1237 xent
->ch
.node_type
= UBIFS_XENT_NODE
;
1238 xent_key_init(c
, &xent_key
, host
->i_ino
, nm
);
1239 key_write(c
, &xent_key
, xent
->key
);
1241 xent
->type
= get_dent_type(inode
->i_mode
);
1242 xent
->nlen
= cpu_to_le16(nm
->len
);
1243 memcpy(xent
->name
, nm
->name
, nm
->len
);
1244 xent
->name
[nm
->len
] = '\0';
1245 zero_dent_node_unused(xent
);
1246 ubifs_prep_grp_node(c
, xent
, xlen
, 0);
1248 ino
= (void *)xent
+ aligned_xlen
;
1249 pack_inode(c
, ino
, inode
, 0, 1);
1250 ino
= (void *)ino
+ UBIFS_INO_NODE_SZ
;
1251 pack_inode(c
, ino
, host
, 1, 0);
1253 err
= write_head(c
, BASEHD
, xent
, len
, &lnum
, &xent_offs
, sync
);
1255 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
, host
->i_ino
);
1256 release_head(c
, BASEHD
);
1261 /* Remove the extended attribute entry from TNC */
1262 err
= ubifs_tnc_remove_nm(c
, &xent_key
, nm
);
1265 err
= ubifs_add_dirt(c
, lnum
, xlen
);
1270 * Remove all nodes belonging to the extended attribute inode from TNC.
1271 * Well, there actually must be only one node - the inode itself.
1273 lowest_ino_key(c
, &key1
, inode
->i_ino
);
1274 highest_ino_key(c
, &key2
, inode
->i_ino
);
1275 err
= ubifs_tnc_remove_range(c
, &key1
, &key2
);
1278 err
= ubifs_add_dirt(c
, lnum
, UBIFS_INO_NODE_SZ
);
1282 /* And update TNC with the new host inode position */
1283 ino_key_init(c
, &key1
, host
->i_ino
);
1284 err
= ubifs_tnc_add(c
, &key1
, lnum
, xent_offs
+ len
- hlen
, hlen
);
1288 finish_reservation(c
);
1289 spin_lock(&host_ui
->ui_lock
);
1290 host_ui
->synced_i_size
= host_ui
->ui_size
;
1291 spin_unlock(&host_ui
->ui_lock
);
1292 mark_inode_clean(c
, host_ui
);
1296 ubifs_ro_mode(c
, err
);
1297 finish_reservation(c
);
1302 * ubifs_jnl_change_xattr - change an extended attribute.
1303 * @c: UBIFS file-system description object
1304 * @inode: extended attribute inode
1307 * This function writes the updated version of an extended attribute inode and
1308 * the host inode tho the journal (to the base head). The host inode is written
1309 * after the extended attribute inode in order to guarantee that the extended
1310 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1311 * consequently, the write-buffer is synchronized. This function returns zero
1312 * in case of success and a negative error code in case of failure.
1314 int ubifs_jnl_change_xattr(struct ubifs_info
*c
, const struct inode
*inode
,
1315 const struct inode
*host
)
1317 int err
, len1
, len2
, aligned_len
, aligned_len1
, lnum
, offs
;
1318 struct ubifs_inode
*host_ui
= ubifs_inode(inode
);
1319 struct ubifs_ino_node
*ino
;
1320 union ubifs_key key
;
1321 int sync
= IS_DIRSYNC(host
);
1323 dbg_jnl("ino %lu, ino %lu", host
->i_ino
, inode
->i_ino
);
1324 ubifs_assert(host
->i_nlink
> 0);
1325 ubifs_assert(inode
->i_nlink
> 0);
1326 ubifs_assert(mutex_is_locked(&host_ui
->ui_mutex
));
1328 len1
= UBIFS_INO_NODE_SZ
+ host_ui
->data_len
;
1329 len2
= UBIFS_INO_NODE_SZ
+ ubifs_inode(inode
)->data_len
;
1330 aligned_len1
= ALIGN(len1
, 8);
1331 aligned_len
= aligned_len1
+ ALIGN(len2
, 8);
1333 ino
= kmalloc(aligned_len
, GFP_NOFS
);
1337 /* Make reservation before allocating sequence numbers */
1338 err
= make_reservation(c
, BASEHD
, aligned_len
);
1342 pack_inode(c
, ino
, host
, 0, 0);
1343 pack_inode(c
, (void *)ino
+ aligned_len1
, inode
, 1, 0);
1345 err
= write_head(c
, BASEHD
, ino
, aligned_len
, &lnum
, &offs
, 0);
1346 if (!sync
&& !err
) {
1347 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
1349 ubifs_wbuf_add_ino_nolock(wbuf
, host
->i_ino
);
1350 ubifs_wbuf_add_ino_nolock(wbuf
, inode
->i_ino
);
1352 release_head(c
, BASEHD
);
1356 ino_key_init(c
, &key
, host
->i_ino
);
1357 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, len1
);
1361 ino_key_init(c
, &key
, inode
->i_ino
);
1362 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
+ aligned_len1
, len2
);
1366 finish_reservation(c
);
1367 spin_lock(&host_ui
->ui_lock
);
1368 host_ui
->synced_i_size
= host_ui
->ui_size
;
1369 spin_unlock(&host_ui
->ui_lock
);
1370 mark_inode_clean(c
, host_ui
);
1375 ubifs_ro_mode(c
, err
);
1376 finish_reservation(c
);
1382 #endif /* CONFIG_UBIFS_FS_XATTR */