blob: 3bc3fc9470994efbcc7e813716b315053c80a8f2 [file] [log] [blame]
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001/*
2 * This file is part of UBIFS.
3 *
4 * Copyright (C) 2006-2008 Nokia Corporation.
5 *
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.
9 *
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
13 * more details.
14 *
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
18 *
19 * Authors: Artem Bityutskiy (Битюцкий Артём)
20 * Adrian Hunter
21 */
22
23/*
24 * This file implements UBIFS journal.
25 *
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.
35 *
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.
40 *
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
43 * only data nodes.
44 *
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
49 * journal.
50 *
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).
54 *
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
58 * all the nodes.
59 */
60
61#include "ubifs.h"
62
63/**
64 * zero_ino_node_unused - zero out unused fields of an on-flash inode node.
65 * @ino: the inode to zero out
66 */
67static inline void zero_ino_node_unused(struct ubifs_ino_node *ino)
68{
69 memset(ino->padding1, 0, 4);
70 memset(ino->padding2, 0, 26);
71}
72
73/**
74 * zero_dent_node_unused - zero out unused fields of an on-flash directory
75 * entry node.
76 * @dent: the directory entry to zero out
77 */
78static inline void zero_dent_node_unused(struct ubifs_dent_node *dent)
79{
80 dent->padding1 = 0;
81 memset(dent->padding2, 0, 4);
82}
83
84/**
85 * zero_data_node_unused - zero out unused fields of an on-flash data node.
86 * @data: the data node to zero out
87 */
88static inline void zero_data_node_unused(struct ubifs_data_node *data)
89{
90 memset(data->padding, 0, 2);
91}
92
93/**
94 * zero_trun_node_unused - zero out unused fields of an on-flash truncation
95 * node.
96 * @trun: the truncation node to zero out
97 */
98static inline void zero_trun_node_unused(struct ubifs_trun_node *trun)
99{
100 memset(trun->padding, 0, 12);
101}
102
103/**
104 * reserve_space - reserve space in the journal.
105 * @c: UBIFS file-system description object
106 * @jhead: journal head number
107 * @len: node length
108 *
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.
114 */
115static int reserve_space(struct ubifs_info *c, int jhead, int len)
116{
117 int err = 0, err1, retries = 0, avail, lnum, offs, free, squeeze;
118 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
119
120 /*
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.
124 */
125 squeeze = (jhead == BASEHD);
126again:
127 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
128
129 if (c->ro_media) {
130 err = -EROFS;
131 goto out_unlock;
132 }
133
134 avail = c->leb_size - wbuf->offs - wbuf->used;
135 if (wbuf->lnum != -1 && avail >= len)
136 return 0;
137
138 /*
139 * Write buffer wasn't seek'ed or there is no enough space - look for an
140 * LEB with some empty space.
141 */
142 lnum = ubifs_find_free_space(c, len, &free, squeeze);
143 if (lnum >= 0) {
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);
147 if (err)
148 goto out_return;
149 /* A new bud was successfully allocated and added to the log */
150 goto out;
151 }
152
153 err = lnum;
154 if (err != -ENOSPC)
155 goto out_unlock;
156
157 /*
158 * No free space, we have to run garbage collector to make
159 * some. But the write-buffer mutex has to be unlocked because
160 * GC also takes it.
161 */
162 dbg_jnl("no free space jhead %d, run GC", jhead);
163 mutex_unlock(&wbuf->io_mutex);
164
165 lnum = ubifs_garbage_collect(c, 0);
166 if (lnum < 0) {
167 err = lnum;
168 if (err != -ENOSPC)
169 return err;
170
171 /*
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
175 * again.
176 */
177 dbg_jnl("GC couldn't make a free LEB for jhead %d", jhead);
178 if (retries++ < 2) {
179 dbg_jnl("retry (%d)", retries);
180 goto again;
181 }
182
183 dbg_jnl("return -ENOSPC");
184 return err;
185 }
186
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;
190
191 if (wbuf->lnum != -1 && avail >= len) {
192 /*
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.
196 */
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);
200 if (err)
201 goto out_unlock;
202 return 0;
203 }
204
205 err = ubifs_add_bud_to_log(c, jhead, lnum, 0);
206 if (err)
207 goto out_return;
208 offs = 0;
209
210out:
211 err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs, UBI_SHORTTERM);
212 if (err)
213 goto out_unlock;
214
215 return 0;
216
217out_unlock:
218 mutex_unlock(&wbuf->io_mutex);
219 return err;
220
221out_return:
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)
226 /*
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()'.
230 */
231 err = err1;
232 mutex_unlock(&wbuf->io_mutex);
233 return err;
234}
235
236/**
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
241 * @len: node length
242 * @lnum: LEB number written is returned here
243 * @offs: offset written is returned here
244 *
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
247 * failure.
248 */
249static int write_node(struct ubifs_info *c, int jhead, void *node, int len,
250 int *lnum, int *offs)
251{
252 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
253
254 ubifs_assert(jhead != GCHD);
255
256 *lnum = c->jheads[jhead].wbuf.lnum;
257 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
258
259 dbg_jnl("jhead %d, LEB %d:%d, len %d", jhead, *lnum, *offs, len);
260 ubifs_prepare_node(c, node, len, 0);
261
262 return ubifs_wbuf_write_nolock(wbuf, node, len);
263}
264
265/**
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
274 *
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).
278 */
279static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
280 int *lnum, int *offs, int sync)
281{
282 int err;
283 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
284
285 ubifs_assert(jhead != GCHD);
286
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);
290
291 err = ubifs_wbuf_write_nolock(wbuf, buf, len);
292 if (err)
293 return err;
294 if (sync)
295 err = ubifs_wbuf_sync_nolock(wbuf);
296 return err;
297}
298
299/**
300 * make_reservation - reserve journal space.
301 * @c: UBIFS file-system description object
302 * @jhead: journal head
303 * @len: how many bytes to reserve
304 *
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
309 * failure.
310 *
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
313 * TNC.
314 */
315static int make_reservation(struct ubifs_info *c, int jhead, int len)
316{
317 int err, cmt_retries = 0, nospc_retries = 0;
318
319again:
320 down_read(&c->commit_sem);
321 err = reserve_space(c, jhead, len);
322 if (!err)
323 return 0;
324 up_read(&c->commit_sem);
325
326 if (err == -ENOSPC) {
327 /*
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.
332 */
333 if (nospc_retries++ < 2) {
334 dbg_jnl("no space, retry");
335 err = -EAGAIN;
336 }
337
338 /*
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.
343 */
344 }
345
346 if (err != -EAGAIN)
347 goto out;
348
349 /*
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.
352 */
353 if (cmt_retries > 128) {
354 /*
355 * This should not happen unless the journal size limitations
356 * are too tough.
357 */
358 ubifs_err("stuck in space allocation");
359 err = -ENOSPC;
360 goto out;
361 } else if (cmt_retries > 32)
362 ubifs_warn("too many space allocation re-tries (%d)",
363 cmt_retries);
364
365 dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
366 cmt_retries);
367 cmt_retries += 1;
368
369 err = ubifs_run_commit(c);
370 if (err)
371 return err;
372 goto again;
373
374out:
375 ubifs_err("cannot reserve %d bytes in jhead %d, error %d",
376 len, jhead, err);
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);
381 dbg_dump_stack();
382 dbg_dump_budg(c);
383 spin_unlock(&c->space_lock);
384 dbg_dump_lprops(c);
385 cmt_retries = dbg_check_lprops(c);
386 up_write(&c->commit_sem);
387 }
388 return err;
389}
390
391/**
392 * release_head - release a journal head.
393 * @c: UBIFS file-system description object
394 * @jhead: journal head
395 *
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.
399 */
400static inline void release_head(struct ubifs_info *c, int jhead)
401{
402 mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
403}
404
405/**
406 * finish_reservation - finish a reservation.
407 * @c: UBIFS file-system description object
408 *
409 * This function finishes journal space reservation. It must be called after
410 * 'make_reservation()'.
411 */
412static void finish_reservation(struct ubifs_info *c)
413{
414 up_read(&c->commit_sem);
415}
416
417/**
418 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
419 * @mode: inode mode
420 */
421static int get_dent_type(int mode)
422{
423 switch (mode & S_IFMT) {
424 case S_IFREG:
425 return UBIFS_ITYPE_REG;
426 case S_IFDIR:
427 return UBIFS_ITYPE_DIR;
428 case S_IFLNK:
429 return UBIFS_ITYPE_LNK;
430 case S_IFBLK:
431 return UBIFS_ITYPE_BLK;
432 case S_IFCHR:
433 return UBIFS_ITYPE_CHR;
434 case S_IFIFO:
435 return UBIFS_ITYPE_FIFO;
436 case S_IFSOCK:
437 return UBIFS_ITYPE_SOCK;
438 default:
439 BUG();
440 }
441 return 0;
442}
443
444/**
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
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300450 */
451static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300452 const struct inode *inode, int last)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300453{
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300454 int data_len = 0, last_reference = !inode->i_nlink;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300455 struct ubifs_inode *ui = ubifs_inode(inode);
456
457 ino->ch.node_type = UBIFS_INO_NODE;
458 ino_key_init_flash(c, &ino->key, inode->i_ino);
459 ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
460 ino->atime_sec = cpu_to_le64(inode->i_atime.tv_sec);
461 ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
462 ino->ctime_sec = cpu_to_le64(inode->i_ctime.tv_sec);
463 ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
464 ino->mtime_sec = cpu_to_le64(inode->i_mtime.tv_sec);
465 ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
466 ino->uid = cpu_to_le32(inode->i_uid);
467 ino->gid = cpu_to_le32(inode->i_gid);
468 ino->mode = cpu_to_le32(inode->i_mode);
469 ino->flags = cpu_to_le32(ui->flags);
470 ino->size = cpu_to_le64(ui->ui_size);
471 ino->nlink = cpu_to_le32(inode->i_nlink);
472 ino->compr_type = cpu_to_le16(ui->compr_type);
473 ino->data_len = cpu_to_le32(ui->data_len);
474 ino->xattr_cnt = cpu_to_le32(ui->xattr_cnt);
475 ino->xattr_size = cpu_to_le32(ui->xattr_size);
476 ino->xattr_names = cpu_to_le32(ui->xattr_names);
477 zero_ino_node_unused(ino);
478
479 /*
480 * Drop the attached data if this is a deletion inode, the data is not
481 * needed anymore.
482 */
483 if (!last_reference) {
484 memcpy(ino->data, ui->data, ui->data_len);
485 data_len = ui->data_len;
486 }
487
488 ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
489}
490
491/**
492 * mark_inode_clean - mark UBIFS inode as clean.
493 * @c: UBIFS file-system description object
494 * @ui: UBIFS inode to mark as clean
495 *
496 * This helper function marks UBIFS inode @ui as clean by cleaning the
497 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
498 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
499 * just do nothing.
500 */
501static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
502{
503 if (ui->dirty)
504 ubifs_release_dirty_inode_budget(c, ui);
505 ui->dirty = 0;
506}
507
508/**
509 * ubifs_jnl_update - update inode.
510 * @c: UBIFS file-system description object
511 * @dir: parent inode or host inode in case of extended attributes
512 * @nm: directory entry name
513 * @inode: inode to update
514 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
515 * @xent: non-zero if the directory entry is an extended attribute entry
516 *
517 * This function updates an inode by writing a directory entry (or extended
518 * attribute entry), the inode itself, and the parent directory inode (or the
519 * host inode) to the journal.
520 *
521 * The function writes the host inode @dir last, which is important in case of
522 * extended attributes. Indeed, then we guarantee that if the host inode gets
523 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
524 * the extended attribute inode gets flushed too. And this is exactly what the
525 * user expects - synchronizing the host inode synchronizes its extended
526 * attributes. Similarly, this guarantees that if @dir is synchronized, its
527 * directory entry corresponding to @nm gets synchronized too.
528 *
529 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
530 * function synchronizes the write-buffer.
531 *
532 * This function marks the @dir and @inode inodes as clean and returns zero on
533 * success. In case of failure, a negative error code is returned.
534 */
535int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
536 const struct qstr *nm, const struct inode *inode,
537 int deletion, int xent)
538{
539 int err, dlen, ilen, len, lnum, ino_offs, dent_offs;
540 int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
541 int last_reference = !!(deletion && inode->i_nlink == 0);
542 struct ubifs_inode *ui = ubifs_inode(inode);
543 struct ubifs_inode *dir_ui = ubifs_inode(dir);
544 struct ubifs_dent_node *dent;
545 struct ubifs_ino_node *ino;
546 union ubifs_key dent_key, ino_key;
547
548 dbg_jnl("ino %lu, dent '%.*s', data len %d in dir ino %lu",
549 inode->i_ino, nm->len, nm->name, ui->data_len, dir->i_ino);
550 ubifs_assert(dir_ui->data_len == 0);
551 ubifs_assert(mutex_is_locked(&dir_ui->ui_mutex));
552
553 dlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
554 ilen = UBIFS_INO_NODE_SZ;
555
556 /*
557 * If the last reference to the inode is being deleted, then there is
558 * no need to attach and write inode data, it is being deleted anyway.
559 * And if the inode is being deleted, no need to synchronize
560 * write-buffer even if the inode is synchronous.
561 */
562 if (!last_reference) {
563 ilen += ui->data_len;
564 sync |= IS_SYNC(inode);
565 }
566
567 aligned_dlen = ALIGN(dlen, 8);
568 aligned_ilen = ALIGN(ilen, 8);
569 len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
570 dent = kmalloc(len, GFP_NOFS);
571 if (!dent)
572 return -ENOMEM;
573
574 /* Make reservation before allocating sequence numbers */
575 err = make_reservation(c, BASEHD, len);
576 if (err)
577 goto out_free;
578
579 if (!xent) {
580 dent->ch.node_type = UBIFS_DENT_NODE;
581 dent_key_init(c, &dent_key, dir->i_ino, nm);
582 } else {
583 dent->ch.node_type = UBIFS_XENT_NODE;
584 xent_key_init(c, &dent_key, dir->i_ino, nm);
585 }
586
587 key_write(c, &dent_key, dent->key);
588 dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
589 dent->type = get_dent_type(inode->i_mode);
590 dent->nlen = cpu_to_le16(nm->len);
591 memcpy(dent->name, nm->name, nm->len);
592 dent->name[nm->len] = '\0';
593 zero_dent_node_unused(dent);
594 ubifs_prep_grp_node(c, dent, dlen, 0);
595
596 ino = (void *)dent + aligned_dlen;
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300597 pack_inode(c, ino, inode, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300598 ino = (void *)ino + aligned_ilen;
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300599 pack_inode(c, ino, dir, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300600
601 if (last_reference) {
602 err = ubifs_add_orphan(c, inode->i_ino);
603 if (err) {
604 release_head(c, BASEHD);
605 goto out_finish;
606 }
607 }
608
609 err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
610 if (err)
611 goto out_release;
612 if (!sync) {
613 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
614
615 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
616 ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
617 }
618 release_head(c, BASEHD);
619 kfree(dent);
620
621 if (deletion) {
622 err = ubifs_tnc_remove_nm(c, &dent_key, nm);
623 if (err)
624 goto out_ro;
625 err = ubifs_add_dirt(c, lnum, dlen);
626 } else
627 err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
628 if (err)
629 goto out_ro;
630
631 /*
632 * Note, we do not remove the inode from TNC even if the last reference
633 * to it has just been deleted, because the inode may still be opened.
634 * Instead, the inode has been added to orphan lists and the orphan
635 * subsystem will take further care about it.
636 */
637 ino_key_init(c, &ino_key, inode->i_ino);
638 ino_offs = dent_offs + aligned_dlen;
639 err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
640 if (err)
641 goto out_ro;
642
643 ino_key_init(c, &ino_key, dir->i_ino);
644 ino_offs += aligned_ilen;
645 err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, UBIFS_INO_NODE_SZ);
646 if (err)
647 goto out_ro;
648
649 finish_reservation(c);
650 spin_lock(&ui->ui_lock);
651 ui->synced_i_size = ui->ui_size;
652 spin_unlock(&ui->ui_lock);
653 mark_inode_clean(c, ui);
654 mark_inode_clean(c, dir_ui);
655 return 0;
656
657out_finish:
658 finish_reservation(c);
659out_free:
660 kfree(dent);
661 return err;
662
663out_release:
664 release_head(c, BASEHD);
665out_ro:
666 ubifs_ro_mode(c, err);
667 if (last_reference)
668 ubifs_delete_orphan(c, inode->i_ino);
669 finish_reservation(c);
670 return err;
671}
672
673/**
674 * ubifs_jnl_write_data - write a data node to the journal.
675 * @c: UBIFS file-system description object
676 * @inode: inode the data node belongs to
677 * @key: node key
678 * @buf: buffer to write
679 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
680 *
681 * This function writes a data node to the journal. Returns %0 if the data node
682 * was successfully written, and a negative error code in case of failure.
683 */
684int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
685 const union ubifs_key *key, const void *buf, int len)
686{
687 struct ubifs_data_node *data;
688 int err, lnum, offs, compr_type, out_len;
689 int dlen = UBIFS_DATA_NODE_SZ + UBIFS_BLOCK_SIZE * WORST_COMPR_FACTOR;
690 struct ubifs_inode *ui = ubifs_inode(inode);
691
692 dbg_jnl("ino %lu, blk %u, len %d, key %s", key_inum(c, key),
693 key_block(c, key), len, DBGKEY(key));
694 ubifs_assert(len <= UBIFS_BLOCK_SIZE);
695
696 data = kmalloc(dlen, GFP_NOFS);
697 if (!data)
698 return -ENOMEM;
699
700 data->ch.node_type = UBIFS_DATA_NODE;
701 key_write(c, key, &data->key);
702 data->size = cpu_to_le32(len);
703 zero_data_node_unused(data);
704
705 if (!(ui->flags && UBIFS_COMPR_FL))
706 /* Compression is disabled for this inode */
707 compr_type = UBIFS_COMPR_NONE;
708 else
709 compr_type = ui->compr_type;
710
711 out_len = dlen - UBIFS_DATA_NODE_SZ;
712 ubifs_compress(buf, len, &data->data, &out_len, &compr_type);
713 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
714
715 dlen = UBIFS_DATA_NODE_SZ + out_len;
716 data->compr_type = cpu_to_le16(compr_type);
717
718 /* Make reservation before allocating sequence numbers */
719 err = make_reservation(c, DATAHD, dlen);
720 if (err)
721 goto out_free;
722
723 err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
724 if (err)
725 goto out_release;
726 ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
727 release_head(c, DATAHD);
728
729 err = ubifs_tnc_add(c, key, lnum, offs, dlen);
730 if (err)
731 goto out_ro;
732
733 finish_reservation(c);
734 kfree(data);
735 return 0;
736
737out_release:
738 release_head(c, DATAHD);
739out_ro:
740 ubifs_ro_mode(c, err);
741 finish_reservation(c);
742out_free:
743 kfree(data);
744 return err;
745}
746
747/**
748 * ubifs_jnl_write_inode - flush inode to the journal.
749 * @c: UBIFS file-system description object
750 * @inode: inode to flush
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300751 *
752 * This function writes inode @inode to the journal. If the inode is
753 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
754 * success and a negative error code in case of failure.
755 */
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300756int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300757{
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300758 int err, lnum, offs;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300759 struct ubifs_ino_node *ino;
760 struct ubifs_inode *ui = ubifs_inode(inode);
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300761 int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300762
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300763 dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300764
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300765 /*
766 * If the inode is being deleted, do not write the attached data. No
767 * need to synchronize the write-buffer either.
768 */
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300769 if (!last_reference) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300770 len += ui->data_len;
771 sync = IS_SYNC(inode);
772 }
773 ino = kmalloc(len, GFP_NOFS);
774 if (!ino)
775 return -ENOMEM;
776
777 /* Make reservation before allocating sequence numbers */
778 err = make_reservation(c, BASEHD, len);
779 if (err)
780 goto out_free;
781
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300782 pack_inode(c, ino, inode, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300783 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
784 if (err)
785 goto out_release;
786 if (!sync)
787 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
788 inode->i_ino);
789 release_head(c, BASEHD);
790
Artem Bityutskiy1f286812008-07-22 12:06:13 +0300791 if (last_reference) {
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300792 err = ubifs_tnc_remove_ino(c, inode->i_ino);
793 if (err)
794 goto out_ro;
795 ubifs_delete_orphan(c, inode->i_ino);
796 err = ubifs_add_dirt(c, lnum, len);
797 } else {
798 union ubifs_key key;
799
800 ino_key_init(c, &key, inode->i_ino);
801 err = ubifs_tnc_add(c, &key, lnum, offs, len);
802 }
803 if (err)
804 goto out_ro;
805
806 finish_reservation(c);
807 spin_lock(&ui->ui_lock);
808 ui->synced_i_size = ui->ui_size;
809 spin_unlock(&ui->ui_lock);
810 kfree(ino);
811 return 0;
812
813out_release:
814 release_head(c, BASEHD);
815out_ro:
816 ubifs_ro_mode(c, err);
817 finish_reservation(c);
818out_free:
819 kfree(ino);
820 return err;
821}
822
823/**
824 * ubifs_jnl_rename - rename a directory entry.
825 * @c: UBIFS file-system description object
826 * @old_dir: parent inode of directory entry to rename
827 * @old_dentry: directory entry to rename
828 * @new_dir: parent inode of directory entry to rename
829 * @new_dentry: new directory entry (or directory entry to replace)
830 * @sync: non-zero if the write-buffer has to be synchronized
831 *
832 * This function implements the re-name operation which may involve writing up
833 * to 3 inodes and 2 directory entries. It marks the written inodes as clean
834 * and returns zero on success. In case of failure, a negative error code is
835 * returned.
836 */
837int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
838 const struct dentry *old_dentry,
839 const struct inode *new_dir,
840 const struct dentry *new_dentry, int sync)
841{
842 void *p;
843 union ubifs_key key;
844 struct ubifs_dent_node *dent, *dent2;
845 int err, dlen1, dlen2, ilen, lnum, offs, len;
846 const struct inode *old_inode = old_dentry->d_inode;
847 const struct inode *new_inode = new_dentry->d_inode;
848 int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
849 int last_reference = !!(new_inode && new_inode->i_nlink == 0);
850 int move = (old_dir != new_dir);
851 struct ubifs_inode *uninitialized_var(new_ui);
852
853 dbg_jnl("dent '%.*s' in dir ino %lu to dent '%.*s' in dir ino %lu",
854 old_dentry->d_name.len, old_dentry->d_name.name,
855 old_dir->i_ino, new_dentry->d_name.len,
856 new_dentry->d_name.name, new_dir->i_ino);
857 ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
858 ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
859 ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
860 ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
861
862 dlen1 = UBIFS_DENT_NODE_SZ + new_dentry->d_name.len + 1;
863 dlen2 = UBIFS_DENT_NODE_SZ + old_dentry->d_name.len + 1;
864 if (new_inode) {
865 new_ui = ubifs_inode(new_inode);
866 ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
867 ilen = UBIFS_INO_NODE_SZ;
868 if (!last_reference)
869 ilen += new_ui->data_len;
870 } else
871 ilen = 0;
872
873 aligned_dlen1 = ALIGN(dlen1, 8);
874 aligned_dlen2 = ALIGN(dlen2, 8);
875 len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
876 if (old_dir != new_dir)
877 len += plen;
878 dent = kmalloc(len, GFP_NOFS);
879 if (!dent)
880 return -ENOMEM;
881
882 /* Make reservation before allocating sequence numbers */
883 err = make_reservation(c, BASEHD, len);
884 if (err)
885 goto out_free;
886
887 /* Make new dent */
888 dent->ch.node_type = UBIFS_DENT_NODE;
889 dent_key_init_flash(c, &dent->key, new_dir->i_ino, &new_dentry->d_name);
890 dent->inum = cpu_to_le64(old_inode->i_ino);
891 dent->type = get_dent_type(old_inode->i_mode);
892 dent->nlen = cpu_to_le16(new_dentry->d_name.len);
893 memcpy(dent->name, new_dentry->d_name.name, new_dentry->d_name.len);
894 dent->name[new_dentry->d_name.len] = '\0';
895 zero_dent_node_unused(dent);
896 ubifs_prep_grp_node(c, dent, dlen1, 0);
897
898 /* Make deletion dent */
899 dent2 = (void *)dent + aligned_dlen1;
900 dent2->ch.node_type = UBIFS_DENT_NODE;
901 dent_key_init_flash(c, &dent2->key, old_dir->i_ino,
902 &old_dentry->d_name);
903 dent2->inum = 0;
904 dent2->type = DT_UNKNOWN;
905 dent2->nlen = cpu_to_le16(old_dentry->d_name.len);
906 memcpy(dent2->name, old_dentry->d_name.name, old_dentry->d_name.len);
907 dent2->name[old_dentry->d_name.len] = '\0';
908 zero_dent_node_unused(dent2);
909 ubifs_prep_grp_node(c, dent2, dlen2, 0);
910
911 p = (void *)dent2 + aligned_dlen2;
912 if (new_inode) {
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300913 pack_inode(c, p, new_inode, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300914 p += ALIGN(ilen, 8);
915 }
916
917 if (!move)
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300918 pack_inode(c, p, old_dir, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300919 else {
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300920 pack_inode(c, p, old_dir, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300921 p += ALIGN(plen, 8);
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +0300922 pack_inode(c, p, new_dir, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +0300923 }
924
925 if (last_reference) {
926 err = ubifs_add_orphan(c, new_inode->i_ino);
927 if (err) {
928 release_head(c, BASEHD);
929 goto out_finish;
930 }
931 }
932
933 err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
934 if (err)
935 goto out_release;
936 if (!sync) {
937 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
938
939 ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
940 ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
941 if (new_inode)
942 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
943 new_inode->i_ino);
944 }
945 release_head(c, BASEHD);
946
947 dent_key_init(c, &key, new_dir->i_ino, &new_dentry->d_name);
948 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &new_dentry->d_name);
949 if (err)
950 goto out_ro;
951
952 err = ubifs_add_dirt(c, lnum, dlen2);
953 if (err)
954 goto out_ro;
955
956 dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
957 err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
958 if (err)
959 goto out_ro;
960
961 offs += aligned_dlen1 + aligned_dlen2;
962 if (new_inode) {
963 ino_key_init(c, &key, new_inode->i_ino);
964 err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
965 if (err)
966 goto out_ro;
967 offs += ALIGN(ilen, 8);
968 }
969
970 ino_key_init(c, &key, old_dir->i_ino);
971 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
972 if (err)
973 goto out_ro;
974
975 if (old_dir != new_dir) {
976 offs += ALIGN(plen, 8);
977 ino_key_init(c, &key, new_dir->i_ino);
978 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
979 if (err)
980 goto out_ro;
981 }
982
983 finish_reservation(c);
984 if (new_inode) {
985 mark_inode_clean(c, new_ui);
986 spin_lock(&new_ui->ui_lock);
987 new_ui->synced_i_size = new_ui->ui_size;
988 spin_unlock(&new_ui->ui_lock);
989 }
990 mark_inode_clean(c, ubifs_inode(old_dir));
991 if (move)
992 mark_inode_clean(c, ubifs_inode(new_dir));
993 kfree(dent);
994 return 0;
995
996out_release:
997 release_head(c, BASEHD);
998out_ro:
999 ubifs_ro_mode(c, err);
1000 if (last_reference)
1001 ubifs_delete_orphan(c, new_inode->i_ino);
1002out_finish:
1003 finish_reservation(c);
1004out_free:
1005 kfree(dent);
1006 return err;
1007}
1008
1009/**
1010 * recomp_data_node - re-compress a truncated data node.
1011 * @dn: data node to re-compress
1012 * @new_len: new length
1013 *
1014 * This function is used when an inode is truncated and the last data node of
1015 * the inode has to be re-compressed and re-written.
1016 */
1017static int recomp_data_node(struct ubifs_data_node *dn, int *new_len)
1018{
1019 void *buf;
1020 int err, len, compr_type, out_len;
1021
1022 out_len = le32_to_cpu(dn->size);
1023 buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS);
1024 if (!buf)
1025 return -ENOMEM;
1026
1027 len = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1028 compr_type = le16_to_cpu(dn->compr_type);
1029 err = ubifs_decompress(&dn->data, len, buf, &out_len, compr_type);
1030 if (err)
1031 goto out;
1032
1033 ubifs_compress(buf, *new_len, &dn->data, &out_len, &compr_type);
1034 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1035 dn->compr_type = cpu_to_le16(compr_type);
1036 dn->size = cpu_to_le32(*new_len);
1037 *new_len = UBIFS_DATA_NODE_SZ + out_len;
1038out:
1039 kfree(buf);
1040 return err;
1041}
1042
1043/**
1044 * ubifs_jnl_truncate - update the journal for a truncation.
1045 * @c: UBIFS file-system description object
1046 * @inode: inode to truncate
1047 * @old_size: old size
1048 * @new_size: new size
1049 *
1050 * When the size of a file decreases due to truncation, a truncation node is
1051 * written, the journal tree is updated, and the last data block is re-written
1052 * if it has been affected. The inode is also updated in order to synchronize
1053 * the new inode size.
1054 *
1055 * This function marks the inode as clean and returns zero on success. In case
1056 * of failure, a negative error code is returned.
1057 */
1058int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1059 loff_t old_size, loff_t new_size)
1060{
1061 union ubifs_key key, to_key;
1062 struct ubifs_ino_node *ino;
1063 struct ubifs_trun_node *trun;
1064 struct ubifs_data_node *uninitialized_var(dn);
1065 int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1066 struct ubifs_inode *ui = ubifs_inode(inode);
1067 ino_t inum = inode->i_ino;
1068 unsigned int blk;
1069
1070 dbg_jnl("ino %lu, size %lld -> %lld", inum, old_size, new_size);
1071 ubifs_assert(!ui->data_len);
1072 ubifs_assert(S_ISREG(inode->i_mode));
1073 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1074
1075 sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1076 UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1077 ino = kmalloc(sz, GFP_NOFS);
1078 if (!ino)
1079 return -ENOMEM;
1080
1081 trun = (void *)ino + UBIFS_INO_NODE_SZ;
1082 trun->ch.node_type = UBIFS_TRUN_NODE;
1083 trun->inum = cpu_to_le32(inum);
1084 trun->old_size = cpu_to_le64(old_size);
1085 trun->new_size = cpu_to_le64(new_size);
1086 zero_trun_node_unused(trun);
1087
1088 dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1089 if (dlen) {
1090 /* Get last data block so it can be truncated */
1091 dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1092 blk = new_size >> UBIFS_BLOCK_SHIFT;
1093 data_key_init(c, &key, inum, blk);
1094 dbg_jnl("last block key %s", DBGKEY(&key));
1095 err = ubifs_tnc_lookup(c, &key, dn);
1096 if (err == -ENOENT)
1097 dlen = 0; /* Not found (so it is a hole) */
1098 else if (err)
1099 goto out_free;
1100 else {
1101 if (le32_to_cpu(dn->size) <= dlen)
1102 dlen = 0; /* Nothing to do */
1103 else {
1104 int compr_type = le16_to_cpu(dn->compr_type);
1105
1106 if (compr_type != UBIFS_COMPR_NONE) {
1107 err = recomp_data_node(dn, &dlen);
1108 if (err)
1109 goto out_free;
1110 } else {
1111 dn->size = cpu_to_le32(dlen);
1112 dlen += UBIFS_DATA_NODE_SZ;
1113 }
1114 zero_data_node_unused(dn);
1115 }
1116 }
1117 }
1118
1119 /* Must make reservation before allocating sequence numbers */
1120 len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1121 if (dlen)
1122 len += dlen;
1123 err = make_reservation(c, BASEHD, len);
1124 if (err)
1125 goto out_free;
1126
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +03001127 pack_inode(c, ino, inode, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001128 ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1129 if (dlen)
1130 ubifs_prep_grp_node(c, dn, dlen, 1);
1131
1132 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1133 if (err)
1134 goto out_release;
1135 if (!sync)
1136 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1137 release_head(c, BASEHD);
1138
1139 if (dlen) {
1140 sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1141 err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1142 if (err)
1143 goto out_ro;
1144 }
1145
1146 ino_key_init(c, &key, inum);
1147 err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1148 if (err)
1149 goto out_ro;
1150
1151 err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1152 if (err)
1153 goto out_ro;
1154
1155 bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1156 blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1157 data_key_init(c, &key, inum, blk);
1158
1159 bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1160 blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0: 1);
1161 data_key_init(c, &to_key, inum, blk);
1162
1163 err = ubifs_tnc_remove_range(c, &key, &to_key);
1164 if (err)
1165 goto out_ro;
1166
1167 finish_reservation(c);
1168 spin_lock(&ui->ui_lock);
1169 ui->synced_i_size = ui->ui_size;
1170 spin_unlock(&ui->ui_lock);
1171 mark_inode_clean(c, ui);
1172 kfree(ino);
1173 return 0;
1174
1175out_release:
1176 release_head(c, BASEHD);
1177out_ro:
1178 ubifs_ro_mode(c, err);
1179 finish_reservation(c);
1180out_free:
1181 kfree(ino);
1182 return err;
1183}
1184
1185#ifdef CONFIG_UBIFS_FS_XATTR
1186
1187/**
1188 * ubifs_jnl_delete_xattr - delete an extended attribute.
1189 * @c: UBIFS file-system description object
1190 * @host: host inode
1191 * @inode: extended attribute inode
1192 * @nm: extended attribute entry name
1193 *
1194 * This function delete an extended attribute which is very similar to
1195 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1196 * updates the target inode. Returns zero in case of success and a negative
1197 * error code in case of failure.
1198 */
1199int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1200 const struct inode *inode, const struct qstr *nm)
1201{
1202 int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1203 struct ubifs_dent_node *xent;
1204 struct ubifs_ino_node *ino;
1205 union ubifs_key xent_key, key1, key2;
1206 int sync = IS_DIRSYNC(host);
1207 struct ubifs_inode *host_ui = ubifs_inode(host);
1208
1209 dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d",
1210 host->i_ino, inode->i_ino, nm->name,
1211 ubifs_inode(inode)->data_len);
1212 ubifs_assert(inode->i_nlink == 0);
1213 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1214
1215 /*
1216 * Since we are deleting the inode, we do not bother to attach any data
1217 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1218 */
1219 xlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
1220 aligned_xlen = ALIGN(xlen, 8);
1221 hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1222 len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1223
1224 xent = kmalloc(len, GFP_NOFS);
1225 if (!xent)
1226 return -ENOMEM;
1227
1228 /* Make reservation before allocating sequence numbers */
1229 err = make_reservation(c, BASEHD, len);
1230 if (err) {
1231 kfree(xent);
1232 return err;
1233 }
1234
1235 xent->ch.node_type = UBIFS_XENT_NODE;
1236 xent_key_init(c, &xent_key, host->i_ino, nm);
1237 key_write(c, &xent_key, xent->key);
1238 xent->inum = 0;
1239 xent->type = get_dent_type(inode->i_mode);
1240 xent->nlen = cpu_to_le16(nm->len);
1241 memcpy(xent->name, nm->name, nm->len);
1242 xent->name[nm->len] = '\0';
1243 zero_dent_node_unused(xent);
1244 ubifs_prep_grp_node(c, xent, xlen, 0);
1245
1246 ino = (void *)xent + aligned_xlen;
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +03001247 pack_inode(c, ino, inode, 0);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001248 ino = (void *)ino + UBIFS_INO_NODE_SZ;
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +03001249 pack_inode(c, ino, host, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001250
1251 err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1252 if (!sync && !err)
1253 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1254 release_head(c, BASEHD);
1255 kfree(xent);
1256 if (err)
1257 goto out_ro;
1258
1259 /* Remove the extended attribute entry from TNC */
1260 err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1261 if (err)
1262 goto out_ro;
1263 err = ubifs_add_dirt(c, lnum, xlen);
1264 if (err)
1265 goto out_ro;
1266
1267 /*
1268 * Remove all nodes belonging to the extended attribute inode from TNC.
1269 * Well, there actually must be only one node - the inode itself.
1270 */
1271 lowest_ino_key(c, &key1, inode->i_ino);
1272 highest_ino_key(c, &key2, inode->i_ino);
1273 err = ubifs_tnc_remove_range(c, &key1, &key2);
1274 if (err)
1275 goto out_ro;
1276 err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1277 if (err)
1278 goto out_ro;
1279
1280 /* And update TNC with the new host inode position */
1281 ino_key_init(c, &key1, host->i_ino);
1282 err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1283 if (err)
1284 goto out_ro;
1285
1286 finish_reservation(c);
1287 spin_lock(&host_ui->ui_lock);
1288 host_ui->synced_i_size = host_ui->ui_size;
1289 spin_unlock(&host_ui->ui_lock);
1290 mark_inode_clean(c, host_ui);
1291 return 0;
1292
1293out_ro:
1294 ubifs_ro_mode(c, err);
1295 finish_reservation(c);
1296 return err;
1297}
1298
1299/**
1300 * ubifs_jnl_change_xattr - change an extended attribute.
1301 * @c: UBIFS file-system description object
1302 * @inode: extended attribute inode
1303 * @host: host inode
1304 *
1305 * This function writes the updated version of an extended attribute inode and
1306 * the host inode tho the journal (to the base head). The host inode is written
1307 * after the extended attribute inode in order to guarantee that the extended
1308 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1309 * consequently, the write-buffer is synchronized. This function returns zero
1310 * in case of success and a negative error code in case of failure.
1311 */
1312int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1313 const struct inode *host)
1314{
1315 int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1316 struct ubifs_inode *host_ui = ubifs_inode(inode);
1317 struct ubifs_ino_node *ino;
1318 union ubifs_key key;
1319 int sync = IS_DIRSYNC(host);
1320
1321 dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1322 ubifs_assert(host->i_nlink > 0);
1323 ubifs_assert(inode->i_nlink > 0);
1324 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1325
1326 len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1327 len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1328 aligned_len1 = ALIGN(len1, 8);
1329 aligned_len = aligned_len1 + ALIGN(len2, 8);
1330
1331 ino = kmalloc(aligned_len, GFP_NOFS);
1332 if (!ino)
1333 return -ENOMEM;
1334
1335 /* Make reservation before allocating sequence numbers */
1336 err = make_reservation(c, BASEHD, aligned_len);
1337 if (err)
1338 goto out_free;
1339
Artem Bityutskiyfd6c6b52008-07-22 12:19:09 +03001340 pack_inode(c, ino, host, 0);
1341 pack_inode(c, (void *)ino + aligned_len1, inode, 1);
Artem Bityutskiy1e517642008-07-14 19:08:37 +03001342
1343 err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1344 if (!sync && !err) {
1345 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1346
1347 ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1348 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1349 }
1350 release_head(c, BASEHD);
1351 if (err)
1352 goto out_ro;
1353
1354 ino_key_init(c, &key, host->i_ino);
1355 err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1356 if (err)
1357 goto out_ro;
1358
1359 ino_key_init(c, &key, inode->i_ino);
1360 err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1361 if (err)
1362 goto out_ro;
1363
1364 finish_reservation(c);
1365 spin_lock(&host_ui->ui_lock);
1366 host_ui->synced_i_size = host_ui->ui_size;
1367 spin_unlock(&host_ui->ui_lock);
1368 mark_inode_clean(c, host_ui);
1369 kfree(ino);
1370 return 0;
1371
1372out_ro:
1373 ubifs_ro_mode(c, err);
1374 finish_reservation(c);
1375out_free:
1376 kfree(ino);
1377 return err;
1378}
1379
1380#endif /* CONFIG_UBIFS_FS_XATTR */