1 // SPDX-License-Identifier: GPL-2.0-only
3 * This file is part of UBIFS.
5 * Copyright (C) 2006-2008 Nokia Corporation.
7 * Authors: Artem Bityutskiy (Битюцкий Артём)
12 * This file implements UBIFS journal.
14 * The journal consists of 2 parts - the log and bud LEBs. The log has fixed
15 * length and position, while a bud logical eraseblock is any LEB in the main
16 * area. Buds contain file system data - data nodes, inode nodes, etc. The log
17 * contains only references to buds and some other stuff like commit
18 * start node. The idea is that when we commit the journal, we do
19 * not copy the data, the buds just become indexed. Since after the commit the
20 * nodes in bud eraseblocks become leaf nodes of the file system index tree, we
21 * use term "bud". Analogy is obvious, bud eraseblocks contain nodes which will
22 * become leafs in the future.
24 * The journal is multi-headed because we want to write data to the journal as
25 * optimally as possible. It is nice to have nodes belonging to the same inode
26 * in one LEB, so we may write data owned by different inodes to different
27 * journal heads, although at present only one data head is used.
29 * For recovery reasons, the base head contains all inode nodes, all directory
30 * entry nodes and all truncate nodes. This means that the other heads contain
33 * Bud LEBs may be half-indexed. For example, if the bud was not full at the
34 * time of commit, the bud is retained to continue to be used in the journal,
35 * even though the "front" of the LEB is now indexed. In that case, the log
36 * reference contains the offset where the bud starts for the purposes of the
39 * The journal size has to be limited, because the larger is the journal, the
40 * longer it takes to mount UBIFS (scanning the journal) and the more memory it
41 * takes (indexing in the TNC).
43 * All the journal write operations like 'ubifs_jnl_update()' here, which write
44 * multiple UBIFS nodes to the journal at one go, are atomic with respect to
45 * unclean reboots. Should the unclean reboot happen, the recovery code drops
52 * zero_ino_node_unused - zero out unused fields of an on-flash inode node.
53 * @ino: the inode to zero out
55 static inline void zero_ino_node_unused(struct ubifs_ino_node
*ino
)
57 memset(ino
->padding1
, 0, 4);
58 memset(ino
->padding2
, 0, 26);
62 * zero_dent_node_unused - zero out unused fields of an on-flash directory
64 * @dent: the directory entry to zero out
66 static inline void zero_dent_node_unused(struct ubifs_dent_node
*dent
)
72 * zero_trun_node_unused - zero out unused fields of an on-flash truncation
74 * @trun: the truncation node to zero out
76 static inline void zero_trun_node_unused(struct ubifs_trun_node
*trun
)
78 memset(trun
->padding
, 0, 12);
81 static void ubifs_add_auth_dirt(struct ubifs_info
*c
, int lnum
)
83 if (ubifs_authenticated(c
))
84 ubifs_add_dirt(c
, lnum
, ubifs_auth_node_sz(c
));
88 * reserve_space - reserve space in the journal.
89 * @c: UBIFS file-system description object
90 * @jhead: journal head number
93 * This function reserves space in journal head @head. If the reservation
94 * succeeded, the journal head stays locked and later has to be unlocked using
95 * 'release_head()'. Returns zero in case of success, %-EAGAIN if commit has to
96 * be done, and other negative error codes in case of other failures.
98 static int reserve_space(struct ubifs_info
*c
, int jhead
, int len
)
100 int err
= 0, err1
, retries
= 0, avail
, lnum
, offs
, squeeze
;
101 struct ubifs_wbuf
*wbuf
= &c
->jheads
[jhead
].wbuf
;
104 * Typically, the base head has smaller nodes written to it, so it is
105 * better to try to allocate space at the ends of eraseblocks. This is
106 * what the squeeze parameter does.
108 ubifs_assert(c
, !c
->ro_media
&& !c
->ro_mount
);
109 squeeze
= (jhead
== BASEHD
);
111 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
118 avail
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
119 if (wbuf
->lnum
!= -1 && avail
>= len
)
123 * Write buffer wasn't seek'ed or there is no enough space - look for an
124 * LEB with some empty space.
126 lnum
= ubifs_find_free_space(c
, len
, &offs
, squeeze
);
135 * No free space, we have to run garbage collector to make
136 * some. But the write-buffer mutex has to be unlocked because
139 dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead
));
140 mutex_unlock(&wbuf
->io_mutex
);
142 lnum
= ubifs_garbage_collect(c
, 0);
149 * GC could not make a free LEB. But someone else may
150 * have allocated new bud for this journal head,
151 * because we dropped @wbuf->io_mutex, so try once
154 dbg_jnl("GC couldn't make a free LEB for jhead %s",
157 dbg_jnl("retry (%d)", retries
);
161 dbg_jnl("return -ENOSPC");
165 mutex_lock_nested(&wbuf
->io_mutex
, wbuf
->jhead
);
166 dbg_jnl("got LEB %d for jhead %s", lnum
, dbg_jhead(jhead
));
167 avail
= c
->leb_size
- wbuf
->offs
- wbuf
->used
;
169 if (wbuf
->lnum
!= -1 && avail
>= len
) {
171 * Someone else has switched the journal head and we have
172 * enough space now. This happens when more than one process is
173 * trying to write to the same journal head at the same time.
175 dbg_jnl("return LEB %d back, already have LEB %d:%d",
176 lnum
, wbuf
->lnum
, wbuf
->offs
+ wbuf
->used
);
177 err
= ubifs_return_leb(c
, lnum
);
187 * Make sure we synchronize the write-buffer before we add the new bud
188 * to the log. Otherwise we may have a power cut after the log
189 * reference node for the last bud (@lnum) is written but before the
190 * write-buffer data are written to the next-to-last bud
191 * (@wbuf->lnum). And the effect would be that the recovery would see
192 * that there is corruption in the next-to-last bud.
194 err
= ubifs_wbuf_sync_nolock(wbuf
);
197 err
= ubifs_add_bud_to_log(c
, jhead
, lnum
, offs
);
200 err
= ubifs_wbuf_seek_nolock(wbuf
, lnum
, offs
);
207 mutex_unlock(&wbuf
->io_mutex
);
211 /* An error occurred and the LEB has to be returned to lprops */
212 ubifs_assert(c
, err
< 0);
213 err1
= ubifs_return_leb(c
, lnum
);
214 if (err1
&& err
== -EAGAIN
)
216 * Return original error code only if it is not %-EAGAIN,
217 * which is not really an error. Otherwise, return the error
218 * code of 'ubifs_return_leb()'.
221 mutex_unlock(&wbuf
->io_mutex
);
225 static int ubifs_hash_nodes(struct ubifs_info
*c
, void *node
,
226 int len
, struct shash_desc
*hash
)
228 int auth_node_size
= ubifs_auth_node_sz(c
);
232 const struct ubifs_ch
*ch
= node
;
233 int nodelen
= le32_to_cpu(ch
->len
);
235 ubifs_assert(c
, len
>= auth_node_size
);
237 if (len
== auth_node_size
)
240 ubifs_assert(c
, len
> nodelen
);
241 ubifs_assert(c
, ch
->magic
== cpu_to_le32(UBIFS_NODE_MAGIC
));
243 err
= ubifs_shash_update(c
, hash
, (void *)node
, nodelen
);
247 node
+= ALIGN(nodelen
, 8);
248 len
-= ALIGN(nodelen
, 8);
251 return ubifs_prepare_auth_node(c
, node
, hash
);
255 * write_head - write data to a journal head.
256 * @c: UBIFS file-system description object
257 * @jhead: journal head
258 * @buf: buffer to write
259 * @len: length to write
260 * @lnum: LEB number written is returned here
261 * @offs: offset written is returned here
262 * @sync: non-zero if the write-buffer has to by synchronized
264 * This function writes data to the reserved space of journal head @jhead.
265 * Returns zero in case of success and a negative error code in case of
268 static int write_head(struct ubifs_info
*c
, int jhead
, void *buf
, int len
,
269 int *lnum
, int *offs
, int sync
)
272 struct ubifs_wbuf
*wbuf
= &c
->jheads
[jhead
].wbuf
;
274 ubifs_assert(c
, jhead
!= GCHD
);
276 *lnum
= c
->jheads
[jhead
].wbuf
.lnum
;
277 *offs
= c
->jheads
[jhead
].wbuf
.offs
+ c
->jheads
[jhead
].wbuf
.used
;
278 dbg_jnl("jhead %s, LEB %d:%d, len %d",
279 dbg_jhead(jhead
), *lnum
, *offs
, len
);
281 if (ubifs_authenticated(c
)) {
282 err
= ubifs_hash_nodes(c
, buf
, len
, c
->jheads
[jhead
].log_hash
);
287 err
= ubifs_wbuf_write_nolock(wbuf
, buf
, len
);
291 err
= ubifs_wbuf_sync_nolock(wbuf
);
296 * make_reservation - reserve journal space.
297 * @c: UBIFS file-system description object
298 * @jhead: journal head
299 * @len: how many bytes to reserve
301 * This function makes space reservation in journal head @jhead. The function
302 * takes the commit lock and locks the journal head, and the caller has to
303 * unlock the head and finish the reservation with 'finish_reservation()'.
304 * Returns zero in case of success and a negative error code in case of
307 * Note, the journal head may be unlocked as soon as the data is written, while
308 * the commit lock has to be released after the data has been added to the
311 static int make_reservation(struct ubifs_info
*c
, int jhead
, int len
)
313 int err
, cmt_retries
= 0, nospc_retries
= 0;
316 down_read(&c
->commit_sem
);
317 err
= reserve_space(c
, jhead
, len
);
319 /* c->commit_sem will get released via finish_reservation(). */
321 up_read(&c
->commit_sem
);
323 if (err
== -ENOSPC
) {
325 * GC could not make any progress. We should try to commit
326 * once because it could make some dirty space and GC would
327 * make progress, so make the error -EAGAIN so that the below
328 * will commit and re-try.
330 if (nospc_retries
++ < 2) {
331 dbg_jnl("no space, retry");
336 * This means that the budgeting is incorrect. We always have
337 * to be able to write to the media, because all operations are
338 * budgeted. Deletions are not budgeted, though, but we reserve
339 * an extra LEB for them.
347 * -EAGAIN means that the journal is full or too large, or the above
348 * code wants to do one commit. Do this and re-try.
350 if (cmt_retries
> 128) {
352 * This should not happen unless the journal size limitations
355 ubifs_err(c
, "stuck in space allocation");
358 } else if (cmt_retries
> 32)
359 ubifs_warn(c
, "too many space allocation re-tries (%d)",
362 dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
366 err
= ubifs_run_commit(c
);
372 ubifs_err(c
, "cannot reserve %d bytes in jhead %d, error %d",
374 if (err
== -ENOSPC
) {
375 /* This are some budgeting problems, print useful information */
376 down_write(&c
->commit_sem
);
378 ubifs_dump_budg(c
, &c
->bi
);
379 ubifs_dump_lprops(c
);
380 cmt_retries
= dbg_check_lprops(c
);
381 up_write(&c
->commit_sem
);
387 * release_head - release a journal head.
388 * @c: UBIFS file-system description object
389 * @jhead: journal head
391 * This function releases journal head @jhead which was locked by
392 * the 'make_reservation()' function. It has to be called after each successful
393 * 'make_reservation()' invocation.
395 static inline void release_head(struct ubifs_info
*c
, int jhead
)
397 mutex_unlock(&c
->jheads
[jhead
].wbuf
.io_mutex
);
401 * finish_reservation - finish a reservation.
402 * @c: UBIFS file-system description object
404 * This function finishes journal space reservation. It must be called after
405 * 'make_reservation()'.
407 static void finish_reservation(struct ubifs_info
*c
)
409 up_read(&c
->commit_sem
);
413 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
416 static int get_dent_type(int mode
)
418 switch (mode
& S_IFMT
) {
420 return UBIFS_ITYPE_REG
;
422 return UBIFS_ITYPE_DIR
;
424 return UBIFS_ITYPE_LNK
;
426 return UBIFS_ITYPE_BLK
;
428 return UBIFS_ITYPE_CHR
;
430 return UBIFS_ITYPE_FIFO
;
432 return UBIFS_ITYPE_SOCK
;
440 * pack_inode - pack an inode node.
441 * @c: UBIFS file-system description object
442 * @ino: buffer in which to pack inode node
443 * @inode: inode to pack
444 * @last: indicates the last node of the group
446 static void pack_inode(struct ubifs_info
*c
, struct ubifs_ino_node
*ino
,
447 const struct inode
*inode
, int last
)
449 int data_len
= 0, last_reference
= !inode
->i_nlink
;
450 struct ubifs_inode
*ui
= ubifs_inode(inode
);
452 ino
->ch
.node_type
= UBIFS_INO_NODE
;
453 ino_key_init_flash(c
, &ino
->key
, inode
->i_ino
);
454 ino
->creat_sqnum
= cpu_to_le64(ui
->creat_sqnum
);
455 ino
->atime_sec
= cpu_to_le64(inode
->i_atime
.tv_sec
);
456 ino
->atime_nsec
= cpu_to_le32(inode
->i_atime
.tv_nsec
);
457 ino
->ctime_sec
= cpu_to_le64(inode
->i_ctime
.tv_sec
);
458 ino
->ctime_nsec
= cpu_to_le32(inode
->i_ctime
.tv_nsec
);
459 ino
->mtime_sec
= cpu_to_le64(inode
->i_mtime
.tv_sec
);
460 ino
->mtime_nsec
= cpu_to_le32(inode
->i_mtime
.tv_nsec
);
461 ino
->uid
= cpu_to_le32(i_uid_read(inode
));
462 ino
->gid
= cpu_to_le32(i_gid_read(inode
));
463 ino
->mode
= cpu_to_le32(inode
->i_mode
);
464 ino
->flags
= cpu_to_le32(ui
->flags
);
465 ino
->size
= cpu_to_le64(ui
->ui_size
);
466 ino
->nlink
= cpu_to_le32(inode
->i_nlink
);
467 ino
->compr_type
= cpu_to_le16(ui
->compr_type
);
468 ino
->data_len
= cpu_to_le32(ui
->data_len
);
469 ino
->xattr_cnt
= cpu_to_le32(ui
->xattr_cnt
);
470 ino
->xattr_size
= cpu_to_le32(ui
->xattr_size
);
471 ino
->xattr_names
= cpu_to_le32(ui
->xattr_names
);
472 zero_ino_node_unused(ino
);
475 * Drop the attached data if this is a deletion inode, the data is not
478 if (!last_reference
) {
479 memcpy(ino
->data
, ui
->data
, ui
->data_len
);
480 data_len
= ui
->data_len
;
483 ubifs_prep_grp_node(c
, ino
, UBIFS_INO_NODE_SZ
+ data_len
, last
);
487 * mark_inode_clean - mark UBIFS inode as clean.
488 * @c: UBIFS file-system description object
489 * @ui: UBIFS inode to mark as clean
491 * This helper function marks UBIFS inode @ui as clean by cleaning the
492 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
493 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
496 static void mark_inode_clean(struct ubifs_info
*c
, struct ubifs_inode
*ui
)
499 ubifs_release_dirty_inode_budget(c
, ui
);
503 static void set_dent_cookie(struct ubifs_info
*c
, struct ubifs_dent_node
*dent
)
506 dent
->cookie
= (__force __le32
) prandom_u32();
512 * ubifs_jnl_update - update inode.
513 * @c: UBIFS file-system description object
514 * @dir: parent inode or host inode in case of extended attributes
515 * @nm: directory entry name
516 * @inode: inode to update
517 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
518 * @xent: non-zero if the directory entry is an extended attribute entry
520 * This function updates an inode by writing a directory entry (or extended
521 * attribute entry), the inode itself, and the parent directory inode (or the
522 * host inode) to the journal.
524 * The function writes the host inode @dir last, which is important in case of
525 * extended attributes. Indeed, then we guarantee that if the host inode gets
526 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
527 * the extended attribute inode gets flushed too. And this is exactly what the
528 * user expects - synchronizing the host inode synchronizes its extended
529 * attributes. Similarly, this guarantees that if @dir is synchronized, its
530 * directory entry corresponding to @nm gets synchronized too.
532 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
533 * function synchronizes the write-buffer.
535 * This function marks the @dir and @inode inodes as clean and returns zero on
536 * success. In case of failure, a negative error code is returned.
538 int ubifs_jnl_update(struct ubifs_info
*c
, const struct inode
*dir
,
539 const struct fscrypt_name
*nm
, const struct inode
*inode
,
540 int deletion
, int xent
)
542 int err
, dlen
, ilen
, len
, lnum
, ino_offs
, dent_offs
;
543 int aligned_dlen
, aligned_ilen
, sync
= IS_DIRSYNC(dir
);
544 int last_reference
= !!(deletion
&& inode
->i_nlink
== 0);
545 struct ubifs_inode
*ui
= ubifs_inode(inode
);
546 struct ubifs_inode
*host_ui
= ubifs_inode(dir
);
547 struct ubifs_dent_node
*dent
;
548 struct ubifs_ino_node
*ino
;
549 union ubifs_key dent_key
, ino_key
;
550 u8 hash_dent
[UBIFS_HASH_ARR_SZ
];
551 u8 hash_ino
[UBIFS_HASH_ARR_SZ
];
552 u8 hash_ino_host
[UBIFS_HASH_ARR_SZ
];
554 ubifs_assert(c
, mutex_is_locked(&host_ui
->ui_mutex
));
556 dlen
= UBIFS_DENT_NODE_SZ
+ fname_len(nm
) + 1;
557 ilen
= UBIFS_INO_NODE_SZ
;
560 * If the last reference to the inode is being deleted, then there is
561 * no need to attach and write inode data, it is being deleted anyway.
562 * And if the inode is being deleted, no need to synchronize
563 * write-buffer even if the inode is synchronous.
565 if (!last_reference
) {
566 ilen
+= ui
->data_len
;
567 sync
|= IS_SYNC(inode
);
570 aligned_dlen
= ALIGN(dlen
, 8);
571 aligned_ilen
= ALIGN(ilen
, 8);
573 len
= aligned_dlen
+ aligned_ilen
+ UBIFS_INO_NODE_SZ
;
574 /* Make sure to also account for extended attributes */
575 if (ubifs_authenticated(c
))
576 len
+= ALIGN(host_ui
->data_len
, 8) + ubifs_auth_node_sz(c
);
578 len
+= host_ui
->data_len
;
580 dent
= kzalloc(len
, GFP_NOFS
);
584 /* Make reservation before allocating sequence numbers */
585 err
= make_reservation(c
, BASEHD
, len
);
590 dent
->ch
.node_type
= UBIFS_DENT_NODE
;
591 if (fname_name(nm
) == NULL
)
592 dent_key_init_hash(c
, &dent_key
, dir
->i_ino
, nm
->hash
);
594 dent_key_init(c
, &dent_key
, dir
->i_ino
, nm
);
596 dent
->ch
.node_type
= UBIFS_XENT_NODE
;
597 xent_key_init(c
, &dent_key
, dir
->i_ino
, nm
);
600 key_write(c
, &dent_key
, dent
->key
);
601 dent
->inum
= deletion
? 0 : cpu_to_le64(inode
->i_ino
);
602 dent
->type
= get_dent_type(inode
->i_mode
);
603 dent
->nlen
= cpu_to_le16(fname_len(nm
));
604 memcpy(dent
->name
, fname_name(nm
), fname_len(nm
));
605 dent
->name
[fname_len(nm
)] = '\0';
606 set_dent_cookie(c
, dent
);
608 zero_dent_node_unused(dent
);
609 ubifs_prep_grp_node(c
, dent
, dlen
, 0);
610 err
= ubifs_node_calc_hash(c
, dent
, hash_dent
);
614 ino
= (void *)dent
+ aligned_dlen
;
615 pack_inode(c
, ino
, inode
, 0);
616 err
= ubifs_node_calc_hash(c
, ino
, hash_ino
);
620 ino
= (void *)ino
+ aligned_ilen
;
621 pack_inode(c
, ino
, dir
, 1);
622 err
= ubifs_node_calc_hash(c
, ino
, hash_ino_host
);
626 if (last_reference
) {
627 err
= ubifs_add_orphan(c
, inode
->i_ino
);
629 release_head(c
, BASEHD
);
632 ui
->del_cmtno
= c
->cmt_no
;
635 err
= write_head(c
, BASEHD
, dent
, len
, &lnum
, &dent_offs
, sync
);
639 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
641 ubifs_wbuf_add_ino_nolock(wbuf
, inode
->i_ino
);
642 ubifs_wbuf_add_ino_nolock(wbuf
, dir
->i_ino
);
644 release_head(c
, BASEHD
);
646 ubifs_add_auth_dirt(c
, lnum
);
649 if (fname_name(nm
) == NULL
)
650 err
= ubifs_tnc_remove_dh(c
, &dent_key
, nm
->minor_hash
);
652 err
= ubifs_tnc_remove_nm(c
, &dent_key
, nm
);
655 err
= ubifs_add_dirt(c
, lnum
, dlen
);
657 err
= ubifs_tnc_add_nm(c
, &dent_key
, lnum
, dent_offs
, dlen
,
663 * Note, we do not remove the inode from TNC even if the last reference
664 * to it has just been deleted, because the inode may still be opened.
665 * Instead, the inode has been added to orphan lists and the orphan
666 * subsystem will take further care about it.
668 ino_key_init(c
, &ino_key
, inode
->i_ino
);
669 ino_offs
= dent_offs
+ aligned_dlen
;
670 err
= ubifs_tnc_add(c
, &ino_key
, lnum
, ino_offs
, ilen
, hash_ino
);
674 ino_key_init(c
, &ino_key
, dir
->i_ino
);
675 ino_offs
+= aligned_ilen
;
676 err
= ubifs_tnc_add(c
, &ino_key
, lnum
, ino_offs
,
677 UBIFS_INO_NODE_SZ
+ host_ui
->data_len
, hash_ino_host
);
681 finish_reservation(c
);
682 spin_lock(&ui
->ui_lock
);
683 ui
->synced_i_size
= ui
->ui_size
;
684 spin_unlock(&ui
->ui_lock
);
686 spin_lock(&host_ui
->ui_lock
);
687 host_ui
->synced_i_size
= host_ui
->ui_size
;
688 spin_unlock(&host_ui
->ui_lock
);
690 mark_inode_clean(c
, ui
);
691 mark_inode_clean(c
, host_ui
);
695 finish_reservation(c
);
701 release_head(c
, BASEHD
);
704 ubifs_ro_mode(c
, err
);
706 ubifs_delete_orphan(c
, inode
->i_ino
);
707 finish_reservation(c
);
712 * ubifs_jnl_write_data - write a data node to the journal.
713 * @c: UBIFS file-system description object
714 * @inode: inode the data node belongs to
716 * @buf: buffer to write
717 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
719 * This function writes a data node to the journal. Returns %0 if the data node
720 * was successfully written, and a negative error code in case of failure.
722 int ubifs_jnl_write_data(struct ubifs_info
*c
, const struct inode
*inode
,
723 const union ubifs_key
*key
, const void *buf
, int len
)
725 struct ubifs_data_node
*data
;
726 int err
, lnum
, offs
, compr_type
, out_len
, compr_len
, auth_len
;
727 int dlen
= COMPRESSED_DATA_NODE_BUF_SZ
, allocated
= 1;
729 struct ubifs_inode
*ui
= ubifs_inode(inode
);
730 bool encrypted
= IS_ENCRYPTED(inode
);
731 u8 hash
[UBIFS_HASH_ARR_SZ
];
733 dbg_jnlk(key
, "ino %lu, blk %u, len %d, key ",
734 (unsigned long)key_inum(c
, key
), key_block(c
, key
), len
);
735 ubifs_assert(c
, len
<= UBIFS_BLOCK_SIZE
);
738 dlen
+= UBIFS_CIPHER_BLOCK_SIZE
;
740 auth_len
= ubifs_auth_node_sz(c
);
742 data
= kmalloc(dlen
+ auth_len
, GFP_NOFS
| __GFP_NOWARN
);
745 * Fall-back to the write reserve buffer. Note, we might be
746 * currently on the memory reclaim path, when the kernel is
747 * trying to free some memory by writing out dirty pages. The
748 * write reserve buffer helps us to guarantee that we are
749 * always able to write the data.
752 mutex_lock(&c
->write_reserve_mutex
);
753 data
= c
->write_reserve_buf
;
756 data
->ch
.node_type
= UBIFS_DATA_NODE
;
757 key_write(c
, key
, &data
->key
);
758 data
->size
= cpu_to_le32(len
);
760 if (!(ui
->flags
& UBIFS_COMPR_FL
))
761 /* Compression is disabled for this inode */
762 compr_type
= UBIFS_COMPR_NONE
;
764 compr_type
= ui
->compr_type
;
766 out_len
= compr_len
= dlen
- UBIFS_DATA_NODE_SZ
;
767 ubifs_compress(c
, buf
, len
, &data
->data
, &compr_len
, &compr_type
);
768 ubifs_assert(c
, compr_len
<= UBIFS_BLOCK_SIZE
);
771 err
= ubifs_encrypt(inode
, data
, compr_len
, &out_len
, key_block(c
, key
));
776 data
->compr_size
= 0;
780 dlen
= UBIFS_DATA_NODE_SZ
+ out_len
;
781 if (ubifs_authenticated(c
))
782 write_len
= ALIGN(dlen
, 8) + auth_len
;
786 data
->compr_type
= cpu_to_le16(compr_type
);
788 /* Make reservation before allocating sequence numbers */
789 err
= make_reservation(c
, DATAHD
, write_len
);
793 ubifs_prepare_node(c
, data
, dlen
, 0);
794 err
= write_head(c
, DATAHD
, data
, write_len
, &lnum
, &offs
, 0);
798 err
= ubifs_node_calc_hash(c
, data
, hash
);
802 ubifs_wbuf_add_ino_nolock(&c
->jheads
[DATAHD
].wbuf
, key_inum(c
, key
));
803 release_head(c
, DATAHD
);
805 ubifs_add_auth_dirt(c
, lnum
);
807 err
= ubifs_tnc_add(c
, key
, lnum
, offs
, dlen
, hash
);
811 finish_reservation(c
);
813 mutex_unlock(&c
->write_reserve_mutex
);
819 release_head(c
, DATAHD
);
821 ubifs_ro_mode(c
, err
);
822 finish_reservation(c
);
825 mutex_unlock(&c
->write_reserve_mutex
);
832 * ubifs_jnl_write_inode - flush inode to the journal.
833 * @c: UBIFS file-system description object
834 * @inode: inode to flush
836 * This function writes inode @inode to the journal. If the inode is
837 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
838 * success and a negative error code in case of failure.
840 int ubifs_jnl_write_inode(struct ubifs_info
*c
, const struct inode
*inode
)
843 struct ubifs_ino_node
*ino
, *ino_start
;
844 struct ubifs_inode
*ui
= ubifs_inode(inode
);
845 int sync
= 0, write_len
= 0, ilen
= UBIFS_INO_NODE_SZ
;
846 int last_reference
= !inode
->i_nlink
;
847 int kill_xattrs
= ui
->xattr_cnt
&& last_reference
;
848 u8 hash
[UBIFS_HASH_ARR_SZ
];
850 dbg_jnl("ino %lu, nlink %u", inode
->i_ino
, inode
->i_nlink
);
853 * If the inode is being deleted, do not write the attached data. No
854 * need to synchronize the write-buffer either.
856 if (!last_reference
) {
857 ilen
+= ui
->data_len
;
858 sync
= IS_SYNC(inode
);
859 } else if (kill_xattrs
) {
860 write_len
+= UBIFS_INO_NODE_SZ
* ui
->xattr_cnt
;
863 if (ubifs_authenticated(c
))
864 write_len
+= ALIGN(ilen
, 8) + ubifs_auth_node_sz(c
);
868 ino_start
= ino
= kmalloc(write_len
, GFP_NOFS
);
872 /* Make reservation before allocating sequence numbers */
873 err
= make_reservation(c
, BASEHD
, write_len
);
879 struct fscrypt_name nm
= {0};
881 struct ubifs_dent_node
*xent
, *pxent
= NULL
;
883 if (ui
->xattr_cnt
>= ubifs_xattr_max_cnt(c
)) {
884 ubifs_err(c
, "Cannot delete inode, it has too much xattrs!");
888 lowest_xent_key(c
, &key
, inode
->i_ino
);
890 xent
= ubifs_tnc_next_ent(c
, &key
, &nm
);
899 fname_name(&nm
) = xent
->name
;
900 fname_len(&nm
) = le16_to_cpu(xent
->nlen
);
902 xino
= ubifs_iget(c
->vfs_sb
, le64_to_cpu(xent
->inum
));
905 ubifs_err(c
, "dead directory entry '%s', error %d",
907 ubifs_ro_mode(c
, err
);
911 ubifs_assert(c
, ubifs_inode(xino
)->xattr
);
914 pack_inode(c
, ino
, xino
, 0);
915 ino
= (void *)ino
+ UBIFS_INO_NODE_SZ
;
920 key_read(c
, &xent
->key
, &key
);
925 pack_inode(c
, ino
, inode
, 1);
926 err
= ubifs_node_calc_hash(c
, ino
, hash
);
930 err
= write_head(c
, BASEHD
, ino_start
, write_len
, &lnum
, &offs
, sync
);
934 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
,
936 release_head(c
, BASEHD
);
938 ubifs_add_auth_dirt(c
, lnum
);
940 if (last_reference
) {
941 err
= ubifs_tnc_remove_ino(c
, inode
->i_ino
);
944 ubifs_delete_orphan(c
, inode
->i_ino
);
945 err
= ubifs_add_dirt(c
, lnum
, write_len
);
949 ino_key_init(c
, &key
, inode
->i_ino
);
950 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, ilen
, hash
);
955 finish_reservation(c
);
956 spin_lock(&ui
->ui_lock
);
957 ui
->synced_i_size
= ui
->ui_size
;
958 spin_unlock(&ui
->ui_lock
);
963 release_head(c
, BASEHD
);
965 ubifs_ro_mode(c
, err
);
966 finish_reservation(c
);
973 * ubifs_jnl_delete_inode - delete an inode.
974 * @c: UBIFS file-system description object
975 * @inode: inode to delete
977 * This function deletes inode @inode which includes removing it from orphans,
978 * deleting it from TNC and, in some cases, writing a deletion inode to the
981 * When regular file inodes are unlinked or a directory inode is removed, the
982 * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
983 * direntry to the media, and adds the inode to orphans. After this, when the
984 * last reference to this inode has been dropped, this function is called. In
985 * general, it has to write one more deletion inode to the media, because if
986 * a commit happened between 'ubifs_jnl_update()' and
987 * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
988 * anymore, and in fact it might not be on the flash anymore, because it might
989 * have been garbage-collected already. And for optimization reasons UBIFS does
990 * not read the orphan area if it has been unmounted cleanly, so it would have
991 * no indication in the journal that there is a deleted inode which has to be
994 * However, if there was no commit between 'ubifs_jnl_update()' and
995 * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
996 * inode to the media for the second time. And this is quite a typical case.
998 * This function returns zero in case of success and a negative error code in
1001 int ubifs_jnl_delete_inode(struct ubifs_info
*c
, const struct inode
*inode
)
1004 struct ubifs_inode
*ui
= ubifs_inode(inode
);
1006 ubifs_assert(c
, inode
->i_nlink
== 0);
1008 if (ui
->xattr_cnt
|| ui
->del_cmtno
!= c
->cmt_no
)
1009 /* A commit happened for sure or inode hosts xattrs */
1010 return ubifs_jnl_write_inode(c
, inode
);
1012 down_read(&c
->commit_sem
);
1014 * Check commit number again, because the first test has been done
1015 * without @c->commit_sem, so a commit might have happened.
1017 if (ui
->del_cmtno
!= c
->cmt_no
) {
1018 up_read(&c
->commit_sem
);
1019 return ubifs_jnl_write_inode(c
, inode
);
1022 err
= ubifs_tnc_remove_ino(c
, inode
->i_ino
);
1024 ubifs_ro_mode(c
, err
);
1026 ubifs_delete_orphan(c
, inode
->i_ino
);
1027 up_read(&c
->commit_sem
);
1032 * ubifs_jnl_xrename - cross rename two directory entries.
1033 * @c: UBIFS file-system description object
1034 * @fst_dir: parent inode of 1st directory entry to exchange
1035 * @fst_inode: 1st inode to exchange
1036 * @fst_nm: name of 1st inode to exchange
1037 * @snd_dir: parent inode of 2nd directory entry to exchange
1038 * @snd_inode: 2nd inode to exchange
1039 * @snd_nm: name of 2nd inode to exchange
1040 * @sync: non-zero if the write-buffer has to be synchronized
1042 * This function implements the cross rename operation which may involve
1043 * writing 2 inodes and 2 directory entries. It marks the written inodes as clean
1044 * and returns zero on success. In case of failure, a negative error code is
1047 int ubifs_jnl_xrename(struct ubifs_info
*c
, const struct inode
*fst_dir
,
1048 const struct inode
*fst_inode
,
1049 const struct fscrypt_name
*fst_nm
,
1050 const struct inode
*snd_dir
,
1051 const struct inode
*snd_inode
,
1052 const struct fscrypt_name
*snd_nm
, int sync
)
1054 union ubifs_key key
;
1055 struct ubifs_dent_node
*dent1
, *dent2
;
1056 int err
, dlen1
, dlen2
, lnum
, offs
, len
, plen
= UBIFS_INO_NODE_SZ
;
1057 int aligned_dlen1
, aligned_dlen2
;
1058 int twoparents
= (fst_dir
!= snd_dir
);
1060 u8 hash_dent1
[UBIFS_HASH_ARR_SZ
];
1061 u8 hash_dent2
[UBIFS_HASH_ARR_SZ
];
1062 u8 hash_p1
[UBIFS_HASH_ARR_SZ
];
1063 u8 hash_p2
[UBIFS_HASH_ARR_SZ
];
1065 ubifs_assert(c
, ubifs_inode(fst_dir
)->data_len
== 0);
1066 ubifs_assert(c
, ubifs_inode(snd_dir
)->data_len
== 0);
1067 ubifs_assert(c
, mutex_is_locked(&ubifs_inode(fst_dir
)->ui_mutex
));
1068 ubifs_assert(c
, mutex_is_locked(&ubifs_inode(snd_dir
)->ui_mutex
));
1070 dlen1
= UBIFS_DENT_NODE_SZ
+ fname_len(snd_nm
) + 1;
1071 dlen2
= UBIFS_DENT_NODE_SZ
+ fname_len(fst_nm
) + 1;
1072 aligned_dlen1
= ALIGN(dlen1
, 8);
1073 aligned_dlen2
= ALIGN(dlen2
, 8);
1075 len
= aligned_dlen1
+ aligned_dlen2
+ ALIGN(plen
, 8);
1079 len
+= ubifs_auth_node_sz(c
);
1081 dent1
= kzalloc(len
, GFP_NOFS
);
1085 /* Make reservation before allocating sequence numbers */
1086 err
= make_reservation(c
, BASEHD
, len
);
1090 /* Make new dent for 1st entry */
1091 dent1
->ch
.node_type
= UBIFS_DENT_NODE
;
1092 dent_key_init_flash(c
, &dent1
->key
, snd_dir
->i_ino
, snd_nm
);
1093 dent1
->inum
= cpu_to_le64(fst_inode
->i_ino
);
1094 dent1
->type
= get_dent_type(fst_inode
->i_mode
);
1095 dent1
->nlen
= cpu_to_le16(fname_len(snd_nm
));
1096 memcpy(dent1
->name
, fname_name(snd_nm
), fname_len(snd_nm
));
1097 dent1
->name
[fname_len(snd_nm
)] = '\0';
1098 set_dent_cookie(c
, dent1
);
1099 zero_dent_node_unused(dent1
);
1100 ubifs_prep_grp_node(c
, dent1
, dlen1
, 0);
1101 err
= ubifs_node_calc_hash(c
, dent1
, hash_dent1
);
1105 /* Make new dent for 2nd entry */
1106 dent2
= (void *)dent1
+ aligned_dlen1
;
1107 dent2
->ch
.node_type
= UBIFS_DENT_NODE
;
1108 dent_key_init_flash(c
, &dent2
->key
, fst_dir
->i_ino
, fst_nm
);
1109 dent2
->inum
= cpu_to_le64(snd_inode
->i_ino
);
1110 dent2
->type
= get_dent_type(snd_inode
->i_mode
);
1111 dent2
->nlen
= cpu_to_le16(fname_len(fst_nm
));
1112 memcpy(dent2
->name
, fname_name(fst_nm
), fname_len(fst_nm
));
1113 dent2
->name
[fname_len(fst_nm
)] = '\0';
1114 set_dent_cookie(c
, dent2
);
1115 zero_dent_node_unused(dent2
);
1116 ubifs_prep_grp_node(c
, dent2
, dlen2
, 0);
1117 err
= ubifs_node_calc_hash(c
, dent2
, hash_dent2
);
1121 p
= (void *)dent2
+ aligned_dlen2
;
1123 pack_inode(c
, p
, fst_dir
, 1);
1124 err
= ubifs_node_calc_hash(c
, p
, hash_p1
);
1128 pack_inode(c
, p
, fst_dir
, 0);
1129 err
= ubifs_node_calc_hash(c
, p
, hash_p1
);
1132 p
+= ALIGN(plen
, 8);
1133 pack_inode(c
, p
, snd_dir
, 1);
1134 err
= ubifs_node_calc_hash(c
, p
, hash_p2
);
1139 err
= write_head(c
, BASEHD
, dent1
, len
, &lnum
, &offs
, sync
);
1143 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
1145 ubifs_wbuf_add_ino_nolock(wbuf
, fst_dir
->i_ino
);
1146 ubifs_wbuf_add_ino_nolock(wbuf
, snd_dir
->i_ino
);
1148 release_head(c
, BASEHD
);
1150 ubifs_add_auth_dirt(c
, lnum
);
1152 dent_key_init(c
, &key
, snd_dir
->i_ino
, snd_nm
);
1153 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen1
, hash_dent1
, snd_nm
);
1157 offs
+= aligned_dlen1
;
1158 dent_key_init(c
, &key
, fst_dir
->i_ino
, fst_nm
);
1159 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen2
, hash_dent2
, fst_nm
);
1163 offs
+= aligned_dlen2
;
1165 ino_key_init(c
, &key
, fst_dir
->i_ino
);
1166 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
, hash_p1
);
1171 offs
+= ALIGN(plen
, 8);
1172 ino_key_init(c
, &key
, snd_dir
->i_ino
);
1173 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
, hash_p2
);
1178 finish_reservation(c
);
1180 mark_inode_clean(c
, ubifs_inode(fst_dir
));
1182 mark_inode_clean(c
, ubifs_inode(snd_dir
));
1187 release_head(c
, BASEHD
);
1189 ubifs_ro_mode(c
, err
);
1190 finish_reservation(c
);
1197 * ubifs_jnl_rename - rename a directory entry.
1198 * @c: UBIFS file-system description object
1199 * @old_dir: parent inode of directory entry to rename
1200 * @old_dentry: directory entry to rename
1201 * @new_dir: parent inode of directory entry to rename
1202 * @new_dentry: new directory entry (or directory entry to replace)
1203 * @sync: non-zero if the write-buffer has to be synchronized
1205 * This function implements the re-name operation which may involve writing up
1206 * to 4 inodes and 2 directory entries. It marks the written inodes as clean
1207 * and returns zero on success. In case of failure, a negative error code is
1210 int ubifs_jnl_rename(struct ubifs_info
*c
, const struct inode
*old_dir
,
1211 const struct inode
*old_inode
,
1212 const struct fscrypt_name
*old_nm
,
1213 const struct inode
*new_dir
,
1214 const struct inode
*new_inode
,
1215 const struct fscrypt_name
*new_nm
,
1216 const struct inode
*whiteout
, int sync
)
1219 union ubifs_key key
;
1220 struct ubifs_dent_node
*dent
, *dent2
;
1221 int err
, dlen1
, dlen2
, ilen
, lnum
, offs
, len
;
1222 int aligned_dlen1
, aligned_dlen2
, plen
= UBIFS_INO_NODE_SZ
;
1223 int last_reference
= !!(new_inode
&& new_inode
->i_nlink
== 0);
1224 int move
= (old_dir
!= new_dir
);
1225 struct ubifs_inode
*uninitialized_var(new_ui
);
1226 u8 hash_old_dir
[UBIFS_HASH_ARR_SZ
];
1227 u8 hash_new_dir
[UBIFS_HASH_ARR_SZ
];
1228 u8 hash_new_inode
[UBIFS_HASH_ARR_SZ
];
1229 u8 hash_dent1
[UBIFS_HASH_ARR_SZ
];
1230 u8 hash_dent2
[UBIFS_HASH_ARR_SZ
];
1232 ubifs_assert(c
, ubifs_inode(old_dir
)->data_len
== 0);
1233 ubifs_assert(c
, ubifs_inode(new_dir
)->data_len
== 0);
1234 ubifs_assert(c
, mutex_is_locked(&ubifs_inode(old_dir
)->ui_mutex
));
1235 ubifs_assert(c
, mutex_is_locked(&ubifs_inode(new_dir
)->ui_mutex
));
1237 dlen1
= UBIFS_DENT_NODE_SZ
+ fname_len(new_nm
) + 1;
1238 dlen2
= UBIFS_DENT_NODE_SZ
+ fname_len(old_nm
) + 1;
1240 new_ui
= ubifs_inode(new_inode
);
1241 ubifs_assert(c
, mutex_is_locked(&new_ui
->ui_mutex
));
1242 ilen
= UBIFS_INO_NODE_SZ
;
1243 if (!last_reference
)
1244 ilen
+= new_ui
->data_len
;
1248 aligned_dlen1
= ALIGN(dlen1
, 8);
1249 aligned_dlen2
= ALIGN(dlen2
, 8);
1250 len
= aligned_dlen1
+ aligned_dlen2
+ ALIGN(ilen
, 8) + ALIGN(plen
, 8);
1254 len
+= ubifs_auth_node_sz(c
);
1256 dent
= kzalloc(len
, GFP_NOFS
);
1260 /* Make reservation before allocating sequence numbers */
1261 err
= make_reservation(c
, BASEHD
, len
);
1266 dent
->ch
.node_type
= UBIFS_DENT_NODE
;
1267 dent_key_init_flash(c
, &dent
->key
, new_dir
->i_ino
, new_nm
);
1268 dent
->inum
= cpu_to_le64(old_inode
->i_ino
);
1269 dent
->type
= get_dent_type(old_inode
->i_mode
);
1270 dent
->nlen
= cpu_to_le16(fname_len(new_nm
));
1271 memcpy(dent
->name
, fname_name(new_nm
), fname_len(new_nm
));
1272 dent
->name
[fname_len(new_nm
)] = '\0';
1273 set_dent_cookie(c
, dent
);
1274 zero_dent_node_unused(dent
);
1275 ubifs_prep_grp_node(c
, dent
, dlen1
, 0);
1276 err
= ubifs_node_calc_hash(c
, dent
, hash_dent1
);
1280 dent2
= (void *)dent
+ aligned_dlen1
;
1281 dent2
->ch
.node_type
= UBIFS_DENT_NODE
;
1282 dent_key_init_flash(c
, &dent2
->key
, old_dir
->i_ino
, old_nm
);
1285 dent2
->inum
= cpu_to_le64(whiteout
->i_ino
);
1286 dent2
->type
= get_dent_type(whiteout
->i_mode
);
1288 /* Make deletion dent */
1290 dent2
->type
= DT_UNKNOWN
;
1292 dent2
->nlen
= cpu_to_le16(fname_len(old_nm
));
1293 memcpy(dent2
->name
, fname_name(old_nm
), fname_len(old_nm
));
1294 dent2
->name
[fname_len(old_nm
)] = '\0';
1295 set_dent_cookie(c
, dent2
);
1296 zero_dent_node_unused(dent2
);
1297 ubifs_prep_grp_node(c
, dent2
, dlen2
, 0);
1298 err
= ubifs_node_calc_hash(c
, dent2
, hash_dent2
);
1302 p
= (void *)dent2
+ aligned_dlen2
;
1304 pack_inode(c
, p
, new_inode
, 0);
1305 err
= ubifs_node_calc_hash(c
, p
, hash_new_inode
);
1309 p
+= ALIGN(ilen
, 8);
1313 pack_inode(c
, p
, old_dir
, 1);
1314 err
= ubifs_node_calc_hash(c
, p
, hash_old_dir
);
1318 pack_inode(c
, p
, old_dir
, 0);
1319 err
= ubifs_node_calc_hash(c
, p
, hash_old_dir
);
1323 p
+= ALIGN(plen
, 8);
1324 pack_inode(c
, p
, new_dir
, 1);
1325 err
= ubifs_node_calc_hash(c
, p
, hash_new_dir
);
1330 if (last_reference
) {
1331 err
= ubifs_add_orphan(c
, new_inode
->i_ino
);
1333 release_head(c
, BASEHD
);
1336 new_ui
->del_cmtno
= c
->cmt_no
;
1339 err
= write_head(c
, BASEHD
, dent
, len
, &lnum
, &offs
, sync
);
1343 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
1345 ubifs_wbuf_add_ino_nolock(wbuf
, new_dir
->i_ino
);
1346 ubifs_wbuf_add_ino_nolock(wbuf
, old_dir
->i_ino
);
1348 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
,
1351 release_head(c
, BASEHD
);
1353 ubifs_add_auth_dirt(c
, lnum
);
1355 dent_key_init(c
, &key
, new_dir
->i_ino
, new_nm
);
1356 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen1
, hash_dent1
, new_nm
);
1360 offs
+= aligned_dlen1
;
1362 dent_key_init(c
, &key
, old_dir
->i_ino
, old_nm
);
1363 err
= ubifs_tnc_add_nm(c
, &key
, lnum
, offs
, dlen2
, hash_dent2
, old_nm
);
1367 ubifs_delete_orphan(c
, whiteout
->i_ino
);
1369 err
= ubifs_add_dirt(c
, lnum
, dlen2
);
1373 dent_key_init(c
, &key
, old_dir
->i_ino
, old_nm
);
1374 err
= ubifs_tnc_remove_nm(c
, &key
, old_nm
);
1379 offs
+= aligned_dlen2
;
1381 ino_key_init(c
, &key
, new_inode
->i_ino
);
1382 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, ilen
, hash_new_inode
);
1385 offs
+= ALIGN(ilen
, 8);
1388 ino_key_init(c
, &key
, old_dir
->i_ino
);
1389 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
, hash_old_dir
);
1394 offs
+= ALIGN(plen
, 8);
1395 ino_key_init(c
, &key
, new_dir
->i_ino
);
1396 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, plen
, hash_new_dir
);
1401 finish_reservation(c
);
1403 mark_inode_clean(c
, new_ui
);
1404 spin_lock(&new_ui
->ui_lock
);
1405 new_ui
->synced_i_size
= new_ui
->ui_size
;
1406 spin_unlock(&new_ui
->ui_lock
);
1408 mark_inode_clean(c
, ubifs_inode(old_dir
));
1410 mark_inode_clean(c
, ubifs_inode(new_dir
));
1415 release_head(c
, BASEHD
);
1417 ubifs_ro_mode(c
, err
);
1419 ubifs_delete_orphan(c
, new_inode
->i_ino
);
1421 finish_reservation(c
);
1428 * truncate_data_node - re-compress/encrypt a truncated data node.
1429 * @c: UBIFS file-system description object
1430 * @inode: inode which referes to the data node
1431 * @block: data block number
1432 * @dn: data node to re-compress
1433 * @new_len: new length
1435 * This function is used when an inode is truncated and the last data node of
1436 * the inode has to be re-compressed/encrypted and re-written.
1438 static int truncate_data_node(const struct ubifs_info
*c
, const struct inode
*inode
,
1439 unsigned int block
, struct ubifs_data_node
*dn
,
1443 int err
, dlen
, compr_type
, out_len
, old_dlen
;
1445 out_len
= le32_to_cpu(dn
->size
);
1446 buf
= kmalloc_array(out_len
, WORST_COMPR_FACTOR
, GFP_NOFS
);
1450 dlen
= old_dlen
= le32_to_cpu(dn
->ch
.len
) - UBIFS_DATA_NODE_SZ
;
1451 compr_type
= le16_to_cpu(dn
->compr_type
);
1453 if (IS_ENCRYPTED(inode
)) {
1454 err
= ubifs_decrypt(inode
, dn
, &dlen
, block
);
1459 if (compr_type
== UBIFS_COMPR_NONE
) {
1462 err
= ubifs_decompress(c
, &dn
->data
, dlen
, buf
, &out_len
, compr_type
);
1466 ubifs_compress(c
, buf
, *new_len
, &dn
->data
, &out_len
, &compr_type
);
1469 if (IS_ENCRYPTED(inode
)) {
1470 err
= ubifs_encrypt(inode
, dn
, out_len
, &old_dlen
, block
);
1479 ubifs_assert(c
, out_len
<= UBIFS_BLOCK_SIZE
);
1480 dn
->compr_type
= cpu_to_le16(compr_type
);
1481 dn
->size
= cpu_to_le32(*new_len
);
1482 *new_len
= UBIFS_DATA_NODE_SZ
+ out_len
;
1490 * ubifs_jnl_truncate - update the journal for a truncation.
1491 * @c: UBIFS file-system description object
1492 * @inode: inode to truncate
1493 * @old_size: old size
1494 * @new_size: new size
1496 * When the size of a file decreases due to truncation, a truncation node is
1497 * written, the journal tree is updated, and the last data block is re-written
1498 * if it has been affected. The inode is also updated in order to synchronize
1499 * the new inode size.
1501 * This function marks the inode as clean and returns zero on success. In case
1502 * of failure, a negative error code is returned.
1504 int ubifs_jnl_truncate(struct ubifs_info
*c
, const struct inode
*inode
,
1505 loff_t old_size
, loff_t new_size
)
1507 union ubifs_key key
, to_key
;
1508 struct ubifs_ino_node
*ino
;
1509 struct ubifs_trun_node
*trun
;
1510 struct ubifs_data_node
*uninitialized_var(dn
);
1511 int err
, dlen
, len
, lnum
, offs
, bit
, sz
, sync
= IS_SYNC(inode
);
1512 struct ubifs_inode
*ui
= ubifs_inode(inode
);
1513 ino_t inum
= inode
->i_ino
;
1515 u8 hash_ino
[UBIFS_HASH_ARR_SZ
];
1516 u8 hash_dn
[UBIFS_HASH_ARR_SZ
];
1518 dbg_jnl("ino %lu, size %lld -> %lld",
1519 (unsigned long)inum
, old_size
, new_size
);
1520 ubifs_assert(c
, !ui
->data_len
);
1521 ubifs_assert(c
, S_ISREG(inode
->i_mode
));
1522 ubifs_assert(c
, mutex_is_locked(&ui
->ui_mutex
));
1524 sz
= UBIFS_TRUN_NODE_SZ
+ UBIFS_INO_NODE_SZ
+
1525 UBIFS_MAX_DATA_NODE_SZ
* WORST_COMPR_FACTOR
;
1527 sz
+= ubifs_auth_node_sz(c
);
1529 ino
= kmalloc(sz
, GFP_NOFS
);
1533 trun
= (void *)ino
+ UBIFS_INO_NODE_SZ
;
1534 trun
->ch
.node_type
= UBIFS_TRUN_NODE
;
1535 trun
->inum
= cpu_to_le32(inum
);
1536 trun
->old_size
= cpu_to_le64(old_size
);
1537 trun
->new_size
= cpu_to_le64(new_size
);
1538 zero_trun_node_unused(trun
);
1540 dlen
= new_size
& (UBIFS_BLOCK_SIZE
- 1);
1542 /* Get last data block so it can be truncated */
1543 dn
= (void *)trun
+ UBIFS_TRUN_NODE_SZ
;
1544 blk
= new_size
>> UBIFS_BLOCK_SHIFT
;
1545 data_key_init(c
, &key
, inum
, blk
);
1546 dbg_jnlk(&key
, "last block key ");
1547 err
= ubifs_tnc_lookup(c
, &key
, dn
);
1549 dlen
= 0; /* Not found (so it is a hole) */
1553 int dn_len
= le32_to_cpu(dn
->size
);
1555 if (dn_len
<= 0 || dn_len
> UBIFS_BLOCK_SIZE
) {
1556 ubifs_err(c
, "bad data node (block %u, inode %lu)",
1558 ubifs_dump_node(c
, dn
);
1563 dlen
= 0; /* Nothing to do */
1565 err
= truncate_data_node(c
, inode
, blk
, dn
, &dlen
);
1572 /* Must make reservation before allocating sequence numbers */
1573 len
= UBIFS_TRUN_NODE_SZ
+ UBIFS_INO_NODE_SZ
;
1575 if (ubifs_authenticated(c
))
1576 len
+= ALIGN(dlen
, 8) + ubifs_auth_node_sz(c
);
1580 err
= make_reservation(c
, BASEHD
, len
);
1584 pack_inode(c
, ino
, inode
, 0);
1585 err
= ubifs_node_calc_hash(c
, ino
, hash_ino
);
1589 ubifs_prep_grp_node(c
, trun
, UBIFS_TRUN_NODE_SZ
, dlen
? 0 : 1);
1591 ubifs_prep_grp_node(c
, dn
, dlen
, 1);
1592 err
= ubifs_node_calc_hash(c
, dn
, hash_dn
);
1597 err
= write_head(c
, BASEHD
, ino
, len
, &lnum
, &offs
, sync
);
1601 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
, inum
);
1602 release_head(c
, BASEHD
);
1604 ubifs_add_auth_dirt(c
, lnum
);
1607 sz
= offs
+ UBIFS_INO_NODE_SZ
+ UBIFS_TRUN_NODE_SZ
;
1608 err
= ubifs_tnc_add(c
, &key
, lnum
, sz
, dlen
, hash_dn
);
1613 ino_key_init(c
, &key
, inum
);
1614 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, UBIFS_INO_NODE_SZ
, hash_ino
);
1618 err
= ubifs_add_dirt(c
, lnum
, UBIFS_TRUN_NODE_SZ
);
1622 bit
= new_size
& (UBIFS_BLOCK_SIZE
- 1);
1623 blk
= (new_size
>> UBIFS_BLOCK_SHIFT
) + (bit
? 1 : 0);
1624 data_key_init(c
, &key
, inum
, blk
);
1626 bit
= old_size
& (UBIFS_BLOCK_SIZE
- 1);
1627 blk
= (old_size
>> UBIFS_BLOCK_SHIFT
) - (bit
? 0 : 1);
1628 data_key_init(c
, &to_key
, inum
, blk
);
1630 err
= ubifs_tnc_remove_range(c
, &key
, &to_key
);
1634 finish_reservation(c
);
1635 spin_lock(&ui
->ui_lock
);
1636 ui
->synced_i_size
= ui
->ui_size
;
1637 spin_unlock(&ui
->ui_lock
);
1638 mark_inode_clean(c
, ui
);
1643 release_head(c
, BASEHD
);
1645 ubifs_ro_mode(c
, err
);
1646 finish_reservation(c
);
1654 * ubifs_jnl_delete_xattr - delete an extended attribute.
1655 * @c: UBIFS file-system description object
1657 * @inode: extended attribute inode
1658 * @nm: extended attribute entry name
1660 * This function delete an extended attribute which is very similar to
1661 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1662 * updates the target inode. Returns zero in case of success and a negative
1663 * error code in case of failure.
1665 int ubifs_jnl_delete_xattr(struct ubifs_info
*c
, const struct inode
*host
,
1666 const struct inode
*inode
,
1667 const struct fscrypt_name
*nm
)
1669 int err
, xlen
, hlen
, len
, lnum
, xent_offs
, aligned_xlen
, write_len
;
1670 struct ubifs_dent_node
*xent
;
1671 struct ubifs_ino_node
*ino
;
1672 union ubifs_key xent_key
, key1
, key2
;
1673 int sync
= IS_DIRSYNC(host
);
1674 struct ubifs_inode
*host_ui
= ubifs_inode(host
);
1675 u8 hash
[UBIFS_HASH_ARR_SZ
];
1677 ubifs_assert(c
, inode
->i_nlink
== 0);
1678 ubifs_assert(c
, mutex_is_locked(&host_ui
->ui_mutex
));
1681 * Since we are deleting the inode, we do not bother to attach any data
1682 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1684 xlen
= UBIFS_DENT_NODE_SZ
+ fname_len(nm
) + 1;
1685 aligned_xlen
= ALIGN(xlen
, 8);
1686 hlen
= host_ui
->data_len
+ UBIFS_INO_NODE_SZ
;
1687 len
= aligned_xlen
+ UBIFS_INO_NODE_SZ
+ ALIGN(hlen
, 8);
1689 write_len
= len
+ ubifs_auth_node_sz(c
);
1691 xent
= kzalloc(write_len
, GFP_NOFS
);
1695 /* Make reservation before allocating sequence numbers */
1696 err
= make_reservation(c
, BASEHD
, write_len
);
1702 xent
->ch
.node_type
= UBIFS_XENT_NODE
;
1703 xent_key_init(c
, &xent_key
, host
->i_ino
, nm
);
1704 key_write(c
, &xent_key
, xent
->key
);
1706 xent
->type
= get_dent_type(inode
->i_mode
);
1707 xent
->nlen
= cpu_to_le16(fname_len(nm
));
1708 memcpy(xent
->name
, fname_name(nm
), fname_len(nm
));
1709 xent
->name
[fname_len(nm
)] = '\0';
1710 zero_dent_node_unused(xent
);
1711 ubifs_prep_grp_node(c
, xent
, xlen
, 0);
1713 ino
= (void *)xent
+ aligned_xlen
;
1714 pack_inode(c
, ino
, inode
, 0);
1715 ino
= (void *)ino
+ UBIFS_INO_NODE_SZ
;
1716 pack_inode(c
, ino
, host
, 1);
1717 err
= ubifs_node_calc_hash(c
, ino
, hash
);
1721 err
= write_head(c
, BASEHD
, xent
, write_len
, &lnum
, &xent_offs
, sync
);
1723 ubifs_wbuf_add_ino_nolock(&c
->jheads
[BASEHD
].wbuf
, host
->i_ino
);
1724 release_head(c
, BASEHD
);
1726 ubifs_add_auth_dirt(c
, lnum
);
1731 /* Remove the extended attribute entry from TNC */
1732 err
= ubifs_tnc_remove_nm(c
, &xent_key
, nm
);
1735 err
= ubifs_add_dirt(c
, lnum
, xlen
);
1740 * Remove all nodes belonging to the extended attribute inode from TNC.
1741 * Well, there actually must be only one node - the inode itself.
1743 lowest_ino_key(c
, &key1
, inode
->i_ino
);
1744 highest_ino_key(c
, &key2
, inode
->i_ino
);
1745 err
= ubifs_tnc_remove_range(c
, &key1
, &key2
);
1748 err
= ubifs_add_dirt(c
, lnum
, UBIFS_INO_NODE_SZ
);
1752 /* And update TNC with the new host inode position */
1753 ino_key_init(c
, &key1
, host
->i_ino
);
1754 err
= ubifs_tnc_add(c
, &key1
, lnum
, xent_offs
+ len
- hlen
, hlen
, hash
);
1758 finish_reservation(c
);
1759 spin_lock(&host_ui
->ui_lock
);
1760 host_ui
->synced_i_size
= host_ui
->ui_size
;
1761 spin_unlock(&host_ui
->ui_lock
);
1762 mark_inode_clean(c
, host_ui
);
1767 release_head(c
, BASEHD
);
1769 ubifs_ro_mode(c
, err
);
1770 finish_reservation(c
);
1775 * ubifs_jnl_change_xattr - change an extended attribute.
1776 * @c: UBIFS file-system description object
1777 * @inode: extended attribute inode
1780 * This function writes the updated version of an extended attribute inode and
1781 * the host inode to the journal (to the base head). The host inode is written
1782 * after the extended attribute inode in order to guarantee that the extended
1783 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1784 * consequently, the write-buffer is synchronized. This function returns zero
1785 * in case of success and a negative error code in case of failure.
1787 int ubifs_jnl_change_xattr(struct ubifs_info
*c
, const struct inode
*inode
,
1788 const struct inode
*host
)
1790 int err
, len1
, len2
, aligned_len
, aligned_len1
, lnum
, offs
;
1791 struct ubifs_inode
*host_ui
= ubifs_inode(host
);
1792 struct ubifs_ino_node
*ino
;
1793 union ubifs_key key
;
1794 int sync
= IS_DIRSYNC(host
);
1795 u8 hash_host
[UBIFS_HASH_ARR_SZ
];
1796 u8 hash
[UBIFS_HASH_ARR_SZ
];
1798 dbg_jnl("ino %lu, ino %lu", host
->i_ino
, inode
->i_ino
);
1799 ubifs_assert(c
, host
->i_nlink
> 0);
1800 ubifs_assert(c
, inode
->i_nlink
> 0);
1801 ubifs_assert(c
, mutex_is_locked(&host_ui
->ui_mutex
));
1803 len1
= UBIFS_INO_NODE_SZ
+ host_ui
->data_len
;
1804 len2
= UBIFS_INO_NODE_SZ
+ ubifs_inode(inode
)->data_len
;
1805 aligned_len1
= ALIGN(len1
, 8);
1806 aligned_len
= aligned_len1
+ ALIGN(len2
, 8);
1808 aligned_len
+= ubifs_auth_node_sz(c
);
1810 ino
= kzalloc(aligned_len
, GFP_NOFS
);
1814 /* Make reservation before allocating sequence numbers */
1815 err
= make_reservation(c
, BASEHD
, aligned_len
);
1819 pack_inode(c
, ino
, host
, 0);
1820 err
= ubifs_node_calc_hash(c
, ino
, hash_host
);
1823 pack_inode(c
, (void *)ino
+ aligned_len1
, inode
, 1);
1824 err
= ubifs_node_calc_hash(c
, (void *)ino
+ aligned_len1
, hash
);
1828 err
= write_head(c
, BASEHD
, ino
, aligned_len
, &lnum
, &offs
, 0);
1829 if (!sync
&& !err
) {
1830 struct ubifs_wbuf
*wbuf
= &c
->jheads
[BASEHD
].wbuf
;
1832 ubifs_wbuf_add_ino_nolock(wbuf
, host
->i_ino
);
1833 ubifs_wbuf_add_ino_nolock(wbuf
, inode
->i_ino
);
1835 release_head(c
, BASEHD
);
1839 ubifs_add_auth_dirt(c
, lnum
);
1841 ino_key_init(c
, &key
, host
->i_ino
);
1842 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
, len1
, hash_host
);
1846 ino_key_init(c
, &key
, inode
->i_ino
);
1847 err
= ubifs_tnc_add(c
, &key
, lnum
, offs
+ aligned_len1
, len2
, hash
);
1851 finish_reservation(c
);
1852 spin_lock(&host_ui
->ui_lock
);
1853 host_ui
->synced_i_size
= host_ui
->ui_size
;
1854 spin_unlock(&host_ui
->ui_lock
);
1855 mark_inode_clean(c
, host_ui
);
1860 release_head(c
, BASEHD
);
1862 ubifs_ro_mode(c
, err
);
1863 finish_reservation(c
);