1 /*
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 */
zero_ino_node_unused(struct ubifs_ino_node * ino)67 static 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 */
zero_dent_node_unused(struct ubifs_dent_node * dent)78 static 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 */
zero_data_node_unused(struct ubifs_data_node * data)88 static 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 */
zero_trun_node_unused(struct ubifs_trun_node * trun)98 static 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 */
reserve_space(struct ubifs_info * c,int jhead,int len)115 static int reserve_space(struct ubifs_info *c, int jhead, int len)
116 {
117 int err = 0, err1, retries = 0, avail, lnum, offs, 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 ubifs_assert(!c->ro_media && !c->ro_mount);
126 squeeze = (jhead == BASEHD);
127 again:
128 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
129
130 if (c->ro_error) {
131 err = -EROFS;
132 goto out_unlock;
133 }
134
135 avail = c->leb_size - wbuf->offs - wbuf->used;
136 if (wbuf->lnum != -1 && avail >= len)
137 return 0;
138
139 /*
140 * Write buffer wasn't seek'ed or there is no enough space - look for an
141 * LEB with some empty space.
142 */
143 lnum = ubifs_find_free_space(c, len, &offs, squeeze);
144 if (lnum >= 0)
145 goto out;
146
147 err = lnum;
148 if (err != -ENOSPC)
149 goto out_unlock;
150
151 /*
152 * No free space, we have to run garbage collector to make
153 * some. But the write-buffer mutex has to be unlocked because
154 * GC also takes it.
155 */
156 dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead));
157 mutex_unlock(&wbuf->io_mutex);
158
159 lnum = ubifs_garbage_collect(c, 0);
160 if (lnum < 0) {
161 err = lnum;
162 if (err != -ENOSPC)
163 return err;
164
165 /*
166 * GC could not make a free LEB. But someone else may
167 * have allocated new bud for this journal head,
168 * because we dropped @wbuf->io_mutex, so try once
169 * again.
170 */
171 dbg_jnl("GC couldn't make a free LEB for jhead %s",
172 dbg_jhead(jhead));
173 if (retries++ < 2) {
174 dbg_jnl("retry (%d)", retries);
175 goto again;
176 }
177
178 dbg_jnl("return -ENOSPC");
179 return err;
180 }
181
182 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
183 dbg_jnl("got LEB %d for jhead %s", lnum, dbg_jhead(jhead));
184 avail = c->leb_size - wbuf->offs - wbuf->used;
185
186 if (wbuf->lnum != -1 && avail >= len) {
187 /*
188 * Someone else has switched the journal head and we have
189 * enough space now. This happens when more than one process is
190 * trying to write to the same journal head at the same time.
191 */
192 dbg_jnl("return LEB %d back, already have LEB %d:%d",
193 lnum, wbuf->lnum, wbuf->offs + wbuf->used);
194 err = ubifs_return_leb(c, lnum);
195 if (err)
196 goto out_unlock;
197 return 0;
198 }
199
200 offs = 0;
201
202 out:
203 /*
204 * Make sure we synchronize the write-buffer before we add the new bud
205 * to the log. Otherwise we may have a power cut after the log
206 * reference node for the last bud (@lnum) is written but before the
207 * write-buffer data are written to the next-to-last bud
208 * (@wbuf->lnum). And the effect would be that the recovery would see
209 * that there is corruption in the next-to-last bud.
210 */
211 err = ubifs_wbuf_sync_nolock(wbuf);
212 if (err)
213 goto out_return;
214 err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
215 if (err)
216 goto out_return;
217 err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs);
218 if (err)
219 goto out_unlock;
220
221 return 0;
222
223 out_unlock:
224 mutex_unlock(&wbuf->io_mutex);
225 return err;
226
227 out_return:
228 /* An error occurred and the LEB has to be returned to lprops */
229 ubifs_assert(err < 0);
230 err1 = ubifs_return_leb(c, lnum);
231 if (err1 && err == -EAGAIN)
232 /*
233 * Return original error code only if it is not %-EAGAIN,
234 * which is not really an error. Otherwise, return the error
235 * code of 'ubifs_return_leb()'.
236 */
237 err = err1;
238 mutex_unlock(&wbuf->io_mutex);
239 return err;
240 }
241
242 /**
243 * write_node - write node to a journal head.
244 * @c: UBIFS file-system description object
245 * @jhead: journal head
246 * @node: node to write
247 * @len: node length
248 * @lnum: LEB number written is returned here
249 * @offs: offset written is returned here
250 *
251 * This function writes a node to reserved space of journal head @jhead.
252 * Returns zero in case of success and a negative error code in case of
253 * failure.
254 */
write_node(struct ubifs_info * c,int jhead,void * node,int len,int * lnum,int * offs)255 static int write_node(struct ubifs_info *c, int jhead, void *node, int len,
256 int *lnum, int *offs)
257 {
258 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
259
260 ubifs_assert(jhead != GCHD);
261
262 *lnum = c->jheads[jhead].wbuf.lnum;
263 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
264
265 dbg_jnl("jhead %s, LEB %d:%d, len %d",
266 dbg_jhead(jhead), *lnum, *offs, len);
267 ubifs_prepare_node(c, node, len, 0);
268
269 return ubifs_wbuf_write_nolock(wbuf, node, len);
270 }
271
272 /**
273 * write_head - write data to a journal head.
274 * @c: UBIFS file-system description object
275 * @jhead: journal head
276 * @buf: buffer to write
277 * @len: length to write
278 * @lnum: LEB number written is returned here
279 * @offs: offset written is returned here
280 * @sync: non-zero if the write-buffer has to by synchronized
281 *
282 * This function is the same as 'write_node()' but it does not assume the
283 * buffer it is writing is a node, so it does not prepare it (which means
284 * initializing common header and calculating CRC).
285 */
write_head(struct ubifs_info * c,int jhead,void * buf,int len,int * lnum,int * offs,int sync)286 static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
287 int *lnum, int *offs, int sync)
288 {
289 int err;
290 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
291
292 ubifs_assert(jhead != GCHD);
293
294 *lnum = c->jheads[jhead].wbuf.lnum;
295 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
296 dbg_jnl("jhead %s, LEB %d:%d, len %d",
297 dbg_jhead(jhead), *lnum, *offs, len);
298
299 err = ubifs_wbuf_write_nolock(wbuf, buf, len);
300 if (err)
301 return err;
302 if (sync)
303 err = ubifs_wbuf_sync_nolock(wbuf);
304 return err;
305 }
306
307 /**
308 * make_reservation - reserve journal space.
309 * @c: UBIFS file-system description object
310 * @jhead: journal head
311 * @len: how many bytes to reserve
312 *
313 * This function makes space reservation in journal head @jhead. The function
314 * takes the commit lock and locks the journal head, and the caller has to
315 * unlock the head and finish the reservation with 'finish_reservation()'.
316 * Returns zero in case of success and a negative error code in case of
317 * failure.
318 *
319 * Note, the journal head may be unlocked as soon as the data is written, while
320 * the commit lock has to be released after the data has been added to the
321 * TNC.
322 */
make_reservation(struct ubifs_info * c,int jhead,int len)323 static int make_reservation(struct ubifs_info *c, int jhead, int len)
324 {
325 int err, cmt_retries = 0, nospc_retries = 0;
326
327 again:
328 down_read(&c->commit_sem);
329 err = reserve_space(c, jhead, len);
330 if (!err)
331 return 0;
332 up_read(&c->commit_sem);
333
334 if (err == -ENOSPC) {
335 /*
336 * GC could not make any progress. We should try to commit
337 * once because it could make some dirty space and GC would
338 * make progress, so make the error -EAGAIN so that the below
339 * will commit and re-try.
340 */
341 if (nospc_retries++ < 2) {
342 dbg_jnl("no space, retry");
343 err = -EAGAIN;
344 }
345
346 /*
347 * This means that the budgeting is incorrect. We always have
348 * to be able to write to the media, because all operations are
349 * budgeted. Deletions are not budgeted, though, but we reserve
350 * an extra LEB for them.
351 */
352 }
353
354 if (err != -EAGAIN)
355 goto out;
356
357 /*
358 * -EAGAIN means that the journal is full or too large, or the above
359 * code wants to do one commit. Do this and re-try.
360 */
361 if (cmt_retries > 128) {
362 /*
363 * This should not happen unless the journal size limitations
364 * are too tough.
365 */
366 ubifs_err(c, "stuck in space allocation");
367 err = -ENOSPC;
368 goto out;
369 } else if (cmt_retries > 32)
370 ubifs_warn(c, "too many space allocation re-tries (%d)",
371 cmt_retries);
372
373 dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
374 cmt_retries);
375 cmt_retries += 1;
376
377 err = ubifs_run_commit(c);
378 if (err)
379 return err;
380 goto again;
381
382 out:
383 ubifs_err(c, "cannot reserve %d bytes in jhead %d, error %d",
384 len, jhead, err);
385 if (err == -ENOSPC) {
386 /* This are some budgeting problems, print useful information */
387 down_write(&c->commit_sem);
388 dump_stack();
389 ubifs_dump_budg(c, &c->bi);
390 ubifs_dump_lprops(c);
391 cmt_retries = dbg_check_lprops(c);
392 up_write(&c->commit_sem);
393 }
394 return err;
395 }
396
397 /**
398 * release_head - release a journal head.
399 * @c: UBIFS file-system description object
400 * @jhead: journal head
401 *
402 * This function releases journal head @jhead which was locked by
403 * the 'make_reservation()' function. It has to be called after each successful
404 * 'make_reservation()' invocation.
405 */
release_head(struct ubifs_info * c,int jhead)406 static inline void release_head(struct ubifs_info *c, int jhead)
407 {
408 mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
409 }
410
411 /**
412 * finish_reservation - finish a reservation.
413 * @c: UBIFS file-system description object
414 *
415 * This function finishes journal space reservation. It must be called after
416 * 'make_reservation()'.
417 */
finish_reservation(struct ubifs_info * c)418 static void finish_reservation(struct ubifs_info *c)
419 {
420 up_read(&c->commit_sem);
421 }
422
423 /**
424 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
425 * @mode: inode mode
426 */
get_dent_type(int mode)427 static int get_dent_type(int mode)
428 {
429 switch (mode & S_IFMT) {
430 case S_IFREG:
431 return UBIFS_ITYPE_REG;
432 case S_IFDIR:
433 return UBIFS_ITYPE_DIR;
434 case S_IFLNK:
435 return UBIFS_ITYPE_LNK;
436 case S_IFBLK:
437 return UBIFS_ITYPE_BLK;
438 case S_IFCHR:
439 return UBIFS_ITYPE_CHR;
440 case S_IFIFO:
441 return UBIFS_ITYPE_FIFO;
442 case S_IFSOCK:
443 return UBIFS_ITYPE_SOCK;
444 default:
445 BUG();
446 }
447 return 0;
448 }
449
450 /**
451 * pack_inode - pack an inode node.
452 * @c: UBIFS file-system description object
453 * @ino: buffer in which to pack inode node
454 * @inode: inode to pack
455 * @last: indicates the last node of the group
456 */
pack_inode(struct ubifs_info * c,struct ubifs_ino_node * ino,const struct inode * inode,int last)457 static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
458 const struct inode *inode, int last)
459 {
460 int data_len = 0, last_reference = !inode->i_nlink;
461 struct ubifs_inode *ui = ubifs_inode(inode);
462
463 ino->ch.node_type = UBIFS_INO_NODE;
464 ino_key_init_flash(c, &ino->key, inode->i_ino);
465 ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
466 ino->atime_sec = cpu_to_le64(inode->i_atime.tv_sec);
467 ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
468 ino->ctime_sec = cpu_to_le64(inode->i_ctime.tv_sec);
469 ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
470 ino->mtime_sec = cpu_to_le64(inode->i_mtime.tv_sec);
471 ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
472 ino->uid = cpu_to_le32(i_uid_read(inode));
473 ino->gid = cpu_to_le32(i_gid_read(inode));
474 ino->mode = cpu_to_le32(inode->i_mode);
475 ino->flags = cpu_to_le32(ui->flags);
476 ino->size = cpu_to_le64(ui->ui_size);
477 ino->nlink = cpu_to_le32(inode->i_nlink);
478 ino->compr_type = cpu_to_le16(ui->compr_type);
479 ino->data_len = cpu_to_le32(ui->data_len);
480 ino->xattr_cnt = cpu_to_le32(ui->xattr_cnt);
481 ino->xattr_size = cpu_to_le32(ui->xattr_size);
482 ino->xattr_names = cpu_to_le32(ui->xattr_names);
483 zero_ino_node_unused(ino);
484
485 /*
486 * Drop the attached data if this is a deletion inode, the data is not
487 * needed anymore.
488 */
489 if (!last_reference) {
490 memcpy(ino->data, ui->data, ui->data_len);
491 data_len = ui->data_len;
492 }
493
494 ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
495 }
496
497 /**
498 * mark_inode_clean - mark UBIFS inode as clean.
499 * @c: UBIFS file-system description object
500 * @ui: UBIFS inode to mark as clean
501 *
502 * This helper function marks UBIFS inode @ui as clean by cleaning the
503 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
504 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
505 * just do nothing.
506 */
mark_inode_clean(struct ubifs_info * c,struct ubifs_inode * ui)507 static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
508 {
509 if (ui->dirty)
510 ubifs_release_dirty_inode_budget(c, ui);
511 ui->dirty = 0;
512 }
513
514 /**
515 * ubifs_jnl_update - update inode.
516 * @c: UBIFS file-system description object
517 * @dir: parent inode or host inode in case of extended attributes
518 * @nm: directory entry name
519 * @inode: inode to update
520 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
521 * @xent: non-zero if the directory entry is an extended attribute entry
522 *
523 * This function updates an inode by writing a directory entry (or extended
524 * attribute entry), the inode itself, and the parent directory inode (or the
525 * host inode) to the journal.
526 *
527 * The function writes the host inode @dir last, which is important in case of
528 * extended attributes. Indeed, then we guarantee that if the host inode gets
529 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
530 * the extended attribute inode gets flushed too. And this is exactly what the
531 * user expects - synchronizing the host inode synchronizes its extended
532 * attributes. Similarly, this guarantees that if @dir is synchronized, its
533 * directory entry corresponding to @nm gets synchronized too.
534 *
535 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
536 * function synchronizes the write-buffer.
537 *
538 * This function marks the @dir and @inode inodes as clean and returns zero on
539 * success. In case of failure, a negative error code is returned.
540 */
ubifs_jnl_update(struct ubifs_info * c,const struct inode * dir,const struct qstr * nm,const struct inode * inode,int deletion,int xent)541 int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
542 const struct qstr *nm, const struct inode *inode,
543 int deletion, int xent)
544 {
545 int err, dlen, ilen, len, lnum, ino_offs, dent_offs;
546 int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
547 int last_reference = !!(deletion && inode->i_nlink == 0);
548 struct ubifs_inode *ui = ubifs_inode(inode);
549 struct ubifs_inode *host_ui = ubifs_inode(dir);
550 struct ubifs_dent_node *dent;
551 struct ubifs_ino_node *ino;
552 union ubifs_key dent_key, ino_key;
553
554 dbg_jnl("ino %lu, dent '%.*s', data len %d in dir ino %lu",
555 inode->i_ino, nm->len, nm->name, ui->data_len, dir->i_ino);
556 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
557
558 dlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
559 ilen = UBIFS_INO_NODE_SZ;
560
561 /*
562 * If the last reference to the inode is being deleted, then there is
563 * no need to attach and write inode data, it is being deleted anyway.
564 * And if the inode is being deleted, no need to synchronize
565 * write-buffer even if the inode is synchronous.
566 */
567 if (!last_reference) {
568 ilen += ui->data_len;
569 sync |= IS_SYNC(inode);
570 }
571
572 aligned_dlen = ALIGN(dlen, 8);
573 aligned_ilen = ALIGN(ilen, 8);
574
575 len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
576 /* Make sure to also account for extended attributes */
577 len += host_ui->data_len;
578
579 dent = kzalloc(len, GFP_NOFS);
580 if (!dent)
581 return -ENOMEM;
582
583 /* Make reservation before allocating sequence numbers */
584 err = make_reservation(c, BASEHD, len);
585 if (err)
586 goto out_free;
587
588 if (!xent) {
589 dent->ch.node_type = UBIFS_DENT_NODE;
590 dent_key_init(c, &dent_key, dir->i_ino, nm);
591 } else {
592 dent->ch.node_type = UBIFS_XENT_NODE;
593 xent_key_init(c, &dent_key, dir->i_ino, nm);
594 }
595
596 key_write(c, &dent_key, dent->key);
597 dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
598 dent->type = get_dent_type(inode->i_mode);
599 dent->nlen = cpu_to_le16(nm->len);
600 memcpy(dent->name, nm->name, nm->len);
601 dent->name[nm->len] = '\0';
602 zero_dent_node_unused(dent);
603 ubifs_prep_grp_node(c, dent, dlen, 0);
604
605 ino = (void *)dent + aligned_dlen;
606 pack_inode(c, ino, inode, 0);
607 ino = (void *)ino + aligned_ilen;
608 pack_inode(c, ino, dir, 1);
609
610 if (last_reference) {
611 err = ubifs_add_orphan(c, inode->i_ino);
612 if (err) {
613 release_head(c, BASEHD);
614 goto out_finish;
615 }
616 ui->del_cmtno = c->cmt_no;
617 }
618
619 err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
620 if (err)
621 goto out_release;
622 if (!sync) {
623 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
624
625 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
626 ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
627 }
628 release_head(c, BASEHD);
629 kfree(dent);
630
631 if (deletion) {
632 err = ubifs_tnc_remove_nm(c, &dent_key, nm);
633 if (err)
634 goto out_ro;
635 err = ubifs_add_dirt(c, lnum, dlen);
636 } else
637 err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
638 if (err)
639 goto out_ro;
640
641 /*
642 * Note, we do not remove the inode from TNC even if the last reference
643 * to it has just been deleted, because the inode may still be opened.
644 * Instead, the inode has been added to orphan lists and the orphan
645 * subsystem will take further care about it.
646 */
647 ino_key_init(c, &ino_key, inode->i_ino);
648 ino_offs = dent_offs + aligned_dlen;
649 err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
650 if (err)
651 goto out_ro;
652
653 ino_key_init(c, &ino_key, dir->i_ino);
654 ino_offs += aligned_ilen;
655 err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs,
656 UBIFS_INO_NODE_SZ + host_ui->data_len);
657 if (err)
658 goto out_ro;
659
660 finish_reservation(c);
661 spin_lock(&ui->ui_lock);
662 ui->synced_i_size = ui->ui_size;
663 spin_unlock(&ui->ui_lock);
664 mark_inode_clean(c, ui);
665 mark_inode_clean(c, host_ui);
666 return 0;
667
668 out_finish:
669 finish_reservation(c);
670 out_free:
671 kfree(dent);
672 return err;
673
674 out_release:
675 release_head(c, BASEHD);
676 kfree(dent);
677 out_ro:
678 ubifs_ro_mode(c, err);
679 if (last_reference)
680 ubifs_delete_orphan(c, inode->i_ino);
681 finish_reservation(c);
682 return err;
683 }
684
685 /**
686 * ubifs_jnl_write_data - write a data node to the journal.
687 * @c: UBIFS file-system description object
688 * @inode: inode the data node belongs to
689 * @key: node key
690 * @buf: buffer to write
691 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
692 *
693 * This function writes a data node to the journal. Returns %0 if the data node
694 * was successfully written, and a negative error code in case of failure.
695 */
ubifs_jnl_write_data(struct ubifs_info * c,const struct inode * inode,const union ubifs_key * key,const void * buf,int len)696 int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
697 const union ubifs_key *key, const void *buf, int len)
698 {
699 struct ubifs_data_node *data;
700 int err, lnum, offs, compr_type, out_len;
701 int dlen = COMPRESSED_DATA_NODE_BUF_SZ, allocated = 1;
702 struct ubifs_inode *ui = ubifs_inode(inode);
703
704 dbg_jnlk(key, "ino %lu, blk %u, len %d, key ",
705 (unsigned long)key_inum(c, key), key_block(c, key), len);
706 ubifs_assert(len <= UBIFS_BLOCK_SIZE);
707
708 data = kmalloc(dlen, GFP_NOFS | __GFP_NOWARN);
709 if (!data) {
710 /*
711 * Fall-back to the write reserve buffer. Note, we might be
712 * currently on the memory reclaim path, when the kernel is
713 * trying to free some memory by writing out dirty pages. The
714 * write reserve buffer helps us to guarantee that we are
715 * always able to write the data.
716 */
717 allocated = 0;
718 mutex_lock(&c->write_reserve_mutex);
719 data = c->write_reserve_buf;
720 }
721
722 data->ch.node_type = UBIFS_DATA_NODE;
723 key_write(c, key, &data->key);
724 data->size = cpu_to_le32(len);
725 zero_data_node_unused(data);
726
727 if (!(ui->flags & UBIFS_COMPR_FL))
728 /* Compression is disabled for this inode */
729 compr_type = UBIFS_COMPR_NONE;
730 else
731 compr_type = ui->compr_type;
732
733 out_len = dlen - UBIFS_DATA_NODE_SZ;
734 ubifs_compress(c, buf, len, &data->data, &out_len, &compr_type);
735 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
736
737 dlen = UBIFS_DATA_NODE_SZ + out_len;
738 data->compr_type = cpu_to_le16(compr_type);
739
740 /* Make reservation before allocating sequence numbers */
741 err = make_reservation(c, DATAHD, dlen);
742 if (err)
743 goto out_free;
744
745 err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
746 if (err)
747 goto out_release;
748 ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
749 release_head(c, DATAHD);
750
751 err = ubifs_tnc_add(c, key, lnum, offs, dlen);
752 if (err)
753 goto out_ro;
754
755 finish_reservation(c);
756 if (!allocated)
757 mutex_unlock(&c->write_reserve_mutex);
758 else
759 kfree(data);
760 return 0;
761
762 out_release:
763 release_head(c, DATAHD);
764 out_ro:
765 ubifs_ro_mode(c, err);
766 finish_reservation(c);
767 out_free:
768 if (!allocated)
769 mutex_unlock(&c->write_reserve_mutex);
770 else
771 kfree(data);
772 return err;
773 }
774
775 /**
776 * ubifs_jnl_write_inode - flush inode to the journal.
777 * @c: UBIFS file-system description object
778 * @inode: inode to flush
779 *
780 * This function writes inode @inode to the journal. If the inode is
781 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
782 * success and a negative error code in case of failure.
783 */
ubifs_jnl_write_inode(struct ubifs_info * c,const struct inode * inode)784 int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
785 {
786 int err, lnum, offs;
787 struct ubifs_ino_node *ino;
788 struct ubifs_inode *ui = ubifs_inode(inode);
789 int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
790
791 dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
792
793 /*
794 * If the inode is being deleted, do not write the attached data. No
795 * need to synchronize the write-buffer either.
796 */
797 if (!last_reference) {
798 len += ui->data_len;
799 sync = IS_SYNC(inode);
800 }
801 ino = kmalloc(len, GFP_NOFS);
802 if (!ino)
803 return -ENOMEM;
804
805 /* Make reservation before allocating sequence numbers */
806 err = make_reservation(c, BASEHD, len);
807 if (err)
808 goto out_free;
809
810 pack_inode(c, ino, inode, 1);
811 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
812 if (err)
813 goto out_release;
814 if (!sync)
815 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
816 inode->i_ino);
817 release_head(c, BASEHD);
818
819 if (last_reference) {
820 err = ubifs_tnc_remove_ino(c, inode->i_ino);
821 if (err)
822 goto out_ro;
823 ubifs_delete_orphan(c, inode->i_ino);
824 err = ubifs_add_dirt(c, lnum, len);
825 } else {
826 union ubifs_key key;
827
828 ino_key_init(c, &key, inode->i_ino);
829 err = ubifs_tnc_add(c, &key, lnum, offs, len);
830 }
831 if (err)
832 goto out_ro;
833
834 finish_reservation(c);
835 spin_lock(&ui->ui_lock);
836 ui->synced_i_size = ui->ui_size;
837 spin_unlock(&ui->ui_lock);
838 kfree(ino);
839 return 0;
840
841 out_release:
842 release_head(c, BASEHD);
843 out_ro:
844 ubifs_ro_mode(c, err);
845 finish_reservation(c);
846 out_free:
847 kfree(ino);
848 return err;
849 }
850
851 /**
852 * ubifs_jnl_delete_inode - delete an inode.
853 * @c: UBIFS file-system description object
854 * @inode: inode to delete
855 *
856 * This function deletes inode @inode which includes removing it from orphans,
857 * deleting it from TNC and, in some cases, writing a deletion inode to the
858 * journal.
859 *
860 * When regular file inodes are unlinked or a directory inode is removed, the
861 * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
862 * direntry to the media, and adds the inode to orphans. After this, when the
863 * last reference to this inode has been dropped, this function is called. In
864 * general, it has to write one more deletion inode to the media, because if
865 * a commit happened between 'ubifs_jnl_update()' and
866 * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
867 * anymore, and in fact it might not be on the flash anymore, because it might
868 * have been garbage-collected already. And for optimization reasons UBIFS does
869 * not read the orphan area if it has been unmounted cleanly, so it would have
870 * no indication in the journal that there is a deleted inode which has to be
871 * removed from TNC.
872 *
873 * However, if there was no commit between 'ubifs_jnl_update()' and
874 * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
875 * inode to the media for the second time. And this is quite a typical case.
876 *
877 * This function returns zero in case of success and a negative error code in
878 * case of failure.
879 */
ubifs_jnl_delete_inode(struct ubifs_info * c,const struct inode * inode)880 int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
881 {
882 int err;
883 struct ubifs_inode *ui = ubifs_inode(inode);
884
885 ubifs_assert(inode->i_nlink == 0);
886
887 if (ui->del_cmtno != c->cmt_no)
888 /* A commit happened for sure */
889 return ubifs_jnl_write_inode(c, inode);
890
891 down_read(&c->commit_sem);
892 /*
893 * Check commit number again, because the first test has been done
894 * without @c->commit_sem, so a commit might have happened.
895 */
896 if (ui->del_cmtno != c->cmt_no) {
897 up_read(&c->commit_sem);
898 return ubifs_jnl_write_inode(c, inode);
899 }
900
901 err = ubifs_tnc_remove_ino(c, inode->i_ino);
902 if (err)
903 ubifs_ro_mode(c, err);
904 else
905 ubifs_delete_orphan(c, inode->i_ino);
906 up_read(&c->commit_sem);
907 return err;
908 }
909
910 /**
911 * ubifs_jnl_xrename - cross rename two directory entries.
912 * @c: UBIFS file-system description object
913 * @fst_dir: parent inode of 1st directory entry to exchange
914 * @fst_dentry: 1st directory entry to exchange
915 * @snd_dir: parent inode of 2nd directory entry to exchange
916 * @snd_dentry: 2nd directory entry to exchange
917 * @sync: non-zero if the write-buffer has to be synchronized
918 *
919 * This function implements the cross rename operation which may involve
920 * writing 2 inodes and 2 directory entries. It marks the written inodes as clean
921 * and returns zero on success. In case of failure, a negative error code is
922 * returned.
923 */
ubifs_jnl_xrename(struct ubifs_info * c,const struct inode * fst_dir,const struct dentry * fst_dentry,const struct inode * snd_dir,const struct dentry * snd_dentry,int sync)924 int ubifs_jnl_xrename(struct ubifs_info *c, const struct inode *fst_dir,
925 const struct dentry *fst_dentry,
926 const struct inode *snd_dir,
927 const struct dentry *snd_dentry, int sync)
928 {
929 union ubifs_key key;
930 struct ubifs_dent_node *dent1, *dent2;
931 int err, dlen1, dlen2, lnum, offs, len, plen = UBIFS_INO_NODE_SZ;
932 int aligned_dlen1, aligned_dlen2;
933 int twoparents = (fst_dir != snd_dir);
934 const struct inode *fst_inode = d_inode(fst_dentry);
935 const struct inode *snd_inode = d_inode(snd_dentry);
936 void *p;
937
938 dbg_jnl("dent '%pd' in dir ino %lu between dent '%pd' in dir ino %lu",
939 fst_dentry, fst_dir->i_ino, snd_dentry, snd_dir->i_ino);
940
941 ubifs_assert(ubifs_inode(fst_dir)->data_len == 0);
942 ubifs_assert(ubifs_inode(snd_dir)->data_len == 0);
943 ubifs_assert(mutex_is_locked(&ubifs_inode(fst_dir)->ui_mutex));
944 ubifs_assert(mutex_is_locked(&ubifs_inode(snd_dir)->ui_mutex));
945
946 dlen1 = UBIFS_DENT_NODE_SZ + snd_dentry->d_name.len + 1;
947 dlen2 = UBIFS_DENT_NODE_SZ + fst_dentry->d_name.len + 1;
948 aligned_dlen1 = ALIGN(dlen1, 8);
949 aligned_dlen2 = ALIGN(dlen2, 8);
950
951 len = aligned_dlen1 + aligned_dlen2 + ALIGN(plen, 8);
952 if (twoparents)
953 len += plen;
954
955 dent1 = kzalloc(len, GFP_NOFS);
956 if (!dent1)
957 return -ENOMEM;
958
959 /* Make reservation before allocating sequence numbers */
960 err = make_reservation(c, BASEHD, len);
961 if (err)
962 goto out_free;
963
964 /* Make new dent for 1st entry */
965 dent1->ch.node_type = UBIFS_DENT_NODE;
966 dent_key_init_flash(c, &dent1->key, snd_dir->i_ino, &snd_dentry->d_name);
967 dent1->inum = cpu_to_le64(fst_inode->i_ino);
968 dent1->type = get_dent_type(fst_inode->i_mode);
969 dent1->nlen = cpu_to_le16(snd_dentry->d_name.len);
970 memcpy(dent1->name, snd_dentry->d_name.name, snd_dentry->d_name.len);
971 dent1->name[snd_dentry->d_name.len] = '\0';
972 zero_dent_node_unused(dent1);
973 ubifs_prep_grp_node(c, dent1, dlen1, 0);
974
975 /* Make new dent for 2nd entry */
976 dent2 = (void *)dent1 + aligned_dlen1;
977 dent2->ch.node_type = UBIFS_DENT_NODE;
978 dent_key_init_flash(c, &dent2->key, fst_dir->i_ino, &fst_dentry->d_name);
979 dent2->inum = cpu_to_le64(snd_inode->i_ino);
980 dent2->type = get_dent_type(snd_inode->i_mode);
981 dent2->nlen = cpu_to_le16(fst_dentry->d_name.len);
982 memcpy(dent2->name, fst_dentry->d_name.name, fst_dentry->d_name.len);
983 dent2->name[fst_dentry->d_name.len] = '\0';
984 zero_dent_node_unused(dent2);
985 ubifs_prep_grp_node(c, dent2, dlen2, 0);
986
987 p = (void *)dent2 + aligned_dlen2;
988 if (!twoparents)
989 pack_inode(c, p, fst_dir, 1);
990 else {
991 pack_inode(c, p, fst_dir, 0);
992 p += ALIGN(plen, 8);
993 pack_inode(c, p, snd_dir, 1);
994 }
995
996 err = write_head(c, BASEHD, dent1, len, &lnum, &offs, sync);
997 if (err)
998 goto out_release;
999 if (!sync) {
1000 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1001
1002 ubifs_wbuf_add_ino_nolock(wbuf, fst_dir->i_ino);
1003 ubifs_wbuf_add_ino_nolock(wbuf, snd_dir->i_ino);
1004 }
1005 release_head(c, BASEHD);
1006
1007 dent_key_init(c, &key, snd_dir->i_ino, &snd_dentry->d_name);
1008 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &snd_dentry->d_name);
1009 if (err)
1010 goto out_ro;
1011
1012 offs += aligned_dlen1;
1013 dent_key_init(c, &key, fst_dir->i_ino, &fst_dentry->d_name);
1014 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, &fst_dentry->d_name);
1015 if (err)
1016 goto out_ro;
1017
1018 offs += aligned_dlen2;
1019
1020 ino_key_init(c, &key, fst_dir->i_ino);
1021 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1022 if (err)
1023 goto out_ro;
1024
1025 if (twoparents) {
1026 offs += ALIGN(plen, 8);
1027 ino_key_init(c, &key, snd_dir->i_ino);
1028 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1029 if (err)
1030 goto out_ro;
1031 }
1032
1033 finish_reservation(c);
1034
1035 mark_inode_clean(c, ubifs_inode(fst_dir));
1036 if (twoparents)
1037 mark_inode_clean(c, ubifs_inode(snd_dir));
1038 kfree(dent1);
1039 return 0;
1040
1041 out_release:
1042 release_head(c, BASEHD);
1043 out_ro:
1044 ubifs_ro_mode(c, err);
1045 finish_reservation(c);
1046 out_free:
1047 kfree(dent1);
1048 return err;
1049 }
1050
1051 /**
1052 * ubifs_jnl_rename - rename a directory entry.
1053 * @c: UBIFS file-system description object
1054 * @old_dir: parent inode of directory entry to rename
1055 * @old_dentry: directory entry to rename
1056 * @new_dir: parent inode of directory entry to rename
1057 * @new_dentry: new directory entry (or directory entry to replace)
1058 * @sync: non-zero if the write-buffer has to be synchronized
1059 *
1060 * This function implements the re-name operation which may involve writing up
1061 * to 4 inodes and 2 directory entries. It marks the written inodes as clean
1062 * and returns zero on success. In case of failure, a negative error code is
1063 * returned.
1064 */
ubifs_jnl_rename(struct ubifs_info * c,const struct inode * old_dir,const struct dentry * old_dentry,const struct inode * new_dir,const struct dentry * new_dentry,const struct inode * whiteout,int sync)1065 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
1066 const struct dentry *old_dentry,
1067 const struct inode *new_dir,
1068 const struct dentry *new_dentry,
1069 const struct inode *whiteout, int sync)
1070 {
1071 void *p;
1072 union ubifs_key key;
1073 struct ubifs_dent_node *dent, *dent2;
1074 int err, dlen1, dlen2, ilen, lnum, offs, len;
1075 const struct inode *old_inode = d_inode(old_dentry);
1076 const struct inode *new_inode = d_inode(new_dentry);
1077 int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
1078 int last_reference = !!(new_inode && new_inode->i_nlink == 0);
1079 int move = (old_dir != new_dir);
1080 struct ubifs_inode *uninitialized_var(new_ui);
1081
1082 dbg_jnl("dent '%pd' in dir ino %lu to dent '%pd' in dir ino %lu",
1083 old_dentry, old_dir->i_ino, new_dentry, new_dir->i_ino);
1084 ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
1085 ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
1086 ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
1087 ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
1088
1089 dlen1 = UBIFS_DENT_NODE_SZ + new_dentry->d_name.len + 1;
1090 dlen2 = UBIFS_DENT_NODE_SZ + old_dentry->d_name.len + 1;
1091 if (new_inode) {
1092 new_ui = ubifs_inode(new_inode);
1093 ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
1094 ilen = UBIFS_INO_NODE_SZ;
1095 if (!last_reference)
1096 ilen += new_ui->data_len;
1097 } else
1098 ilen = 0;
1099
1100 aligned_dlen1 = ALIGN(dlen1, 8);
1101 aligned_dlen2 = ALIGN(dlen2, 8);
1102 len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
1103 if (move)
1104 len += plen;
1105 dent = kzalloc(len, GFP_NOFS);
1106 if (!dent)
1107 return -ENOMEM;
1108
1109 /* Make reservation before allocating sequence numbers */
1110 err = make_reservation(c, BASEHD, len);
1111 if (err)
1112 goto out_free;
1113
1114 /* Make new dent */
1115 dent->ch.node_type = UBIFS_DENT_NODE;
1116 dent_key_init_flash(c, &dent->key, new_dir->i_ino, &new_dentry->d_name);
1117 dent->inum = cpu_to_le64(old_inode->i_ino);
1118 dent->type = get_dent_type(old_inode->i_mode);
1119 dent->nlen = cpu_to_le16(new_dentry->d_name.len);
1120 memcpy(dent->name, new_dentry->d_name.name, new_dentry->d_name.len);
1121 dent->name[new_dentry->d_name.len] = '\0';
1122 zero_dent_node_unused(dent);
1123 ubifs_prep_grp_node(c, dent, dlen1, 0);
1124
1125 dent2 = (void *)dent + aligned_dlen1;
1126 dent2->ch.node_type = UBIFS_DENT_NODE;
1127 dent_key_init_flash(c, &dent2->key, old_dir->i_ino,
1128 &old_dentry->d_name);
1129
1130 if (whiteout) {
1131 dent2->inum = cpu_to_le64(whiteout->i_ino);
1132 dent2->type = get_dent_type(whiteout->i_mode);
1133 } else {
1134 /* Make deletion dent */
1135 dent2->inum = 0;
1136 dent2->type = DT_UNKNOWN;
1137 }
1138 dent2->nlen = cpu_to_le16(old_dentry->d_name.len);
1139 memcpy(dent2->name, old_dentry->d_name.name, old_dentry->d_name.len);
1140 dent2->name[old_dentry->d_name.len] = '\0';
1141 zero_dent_node_unused(dent2);
1142 ubifs_prep_grp_node(c, dent2, dlen2, 0);
1143
1144 p = (void *)dent2 + aligned_dlen2;
1145 if (new_inode) {
1146 pack_inode(c, p, new_inode, 0);
1147 p += ALIGN(ilen, 8);
1148 }
1149
1150 if (!move)
1151 pack_inode(c, p, old_dir, 1);
1152 else {
1153 pack_inode(c, p, old_dir, 0);
1154 p += ALIGN(plen, 8);
1155 pack_inode(c, p, new_dir, 1);
1156 }
1157
1158 if (last_reference) {
1159 err = ubifs_add_orphan(c, new_inode->i_ino);
1160 if (err) {
1161 release_head(c, BASEHD);
1162 goto out_finish;
1163 }
1164 new_ui->del_cmtno = c->cmt_no;
1165 }
1166
1167 err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
1168 if (err)
1169 goto out_release;
1170 if (!sync) {
1171 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1172
1173 ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1174 ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1175 if (new_inode)
1176 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1177 new_inode->i_ino);
1178 }
1179 release_head(c, BASEHD);
1180
1181 dent_key_init(c, &key, new_dir->i_ino, &new_dentry->d_name);
1182 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &new_dentry->d_name);
1183 if (err)
1184 goto out_ro;
1185
1186 offs += aligned_dlen1;
1187 if (whiteout) {
1188 dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
1189 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, &old_dentry->d_name);
1190 if (err)
1191 goto out_ro;
1192
1193 ubifs_delete_orphan(c, whiteout->i_ino);
1194 } else {
1195 err = ubifs_add_dirt(c, lnum, dlen2);
1196 if (err)
1197 goto out_ro;
1198
1199 dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
1200 err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
1201 if (err)
1202 goto out_ro;
1203 }
1204
1205 offs += aligned_dlen2;
1206 if (new_inode) {
1207 ino_key_init(c, &key, new_inode->i_ino);
1208 err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
1209 if (err)
1210 goto out_ro;
1211 offs += ALIGN(ilen, 8);
1212 }
1213
1214 ino_key_init(c, &key, old_dir->i_ino);
1215 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1216 if (err)
1217 goto out_ro;
1218
1219 if (move) {
1220 offs += ALIGN(plen, 8);
1221 ino_key_init(c, &key, new_dir->i_ino);
1222 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1223 if (err)
1224 goto out_ro;
1225 }
1226
1227 finish_reservation(c);
1228 if (new_inode) {
1229 mark_inode_clean(c, new_ui);
1230 spin_lock(&new_ui->ui_lock);
1231 new_ui->synced_i_size = new_ui->ui_size;
1232 spin_unlock(&new_ui->ui_lock);
1233 }
1234 mark_inode_clean(c, ubifs_inode(old_dir));
1235 if (move)
1236 mark_inode_clean(c, ubifs_inode(new_dir));
1237 kfree(dent);
1238 return 0;
1239
1240 out_release:
1241 release_head(c, BASEHD);
1242 out_ro:
1243 ubifs_ro_mode(c, err);
1244 if (last_reference)
1245 ubifs_delete_orphan(c, new_inode->i_ino);
1246 out_finish:
1247 finish_reservation(c);
1248 out_free:
1249 kfree(dent);
1250 return err;
1251 }
1252
1253 /**
1254 * recomp_data_node - re-compress a truncated data node.
1255 * @dn: data node to re-compress
1256 * @new_len: new length
1257 *
1258 * This function is used when an inode is truncated and the last data node of
1259 * the inode has to be re-compressed and re-written.
1260 */
recomp_data_node(const struct ubifs_info * c,struct ubifs_data_node * dn,int * new_len)1261 static int recomp_data_node(const struct ubifs_info *c,
1262 struct ubifs_data_node *dn, int *new_len)
1263 {
1264 void *buf;
1265 int err, len, compr_type, out_len;
1266
1267 out_len = le32_to_cpu(dn->size);
1268 buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS);
1269 if (!buf)
1270 return -ENOMEM;
1271
1272 len = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1273 compr_type = le16_to_cpu(dn->compr_type);
1274 err = ubifs_decompress(c, &dn->data, len, buf, &out_len, compr_type);
1275 if (err)
1276 goto out;
1277
1278 ubifs_compress(c, buf, *new_len, &dn->data, &out_len, &compr_type);
1279 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1280 dn->compr_type = cpu_to_le16(compr_type);
1281 dn->size = cpu_to_le32(*new_len);
1282 *new_len = UBIFS_DATA_NODE_SZ + out_len;
1283 out:
1284 kfree(buf);
1285 return err;
1286 }
1287
1288 /**
1289 * ubifs_jnl_truncate - update the journal for a truncation.
1290 * @c: UBIFS file-system description object
1291 * @inode: inode to truncate
1292 * @old_size: old size
1293 * @new_size: new size
1294 *
1295 * When the size of a file decreases due to truncation, a truncation node is
1296 * written, the journal tree is updated, and the last data block is re-written
1297 * if it has been affected. The inode is also updated in order to synchronize
1298 * the new inode size.
1299 *
1300 * This function marks the inode as clean and returns zero on success. In case
1301 * of failure, a negative error code is returned.
1302 */
ubifs_jnl_truncate(struct ubifs_info * c,const struct inode * inode,loff_t old_size,loff_t new_size)1303 int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1304 loff_t old_size, loff_t new_size)
1305 {
1306 union ubifs_key key, to_key;
1307 struct ubifs_ino_node *ino;
1308 struct ubifs_trun_node *trun;
1309 struct ubifs_data_node *uninitialized_var(dn);
1310 int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1311 struct ubifs_inode *ui = ubifs_inode(inode);
1312 ino_t inum = inode->i_ino;
1313 unsigned int blk;
1314
1315 dbg_jnl("ino %lu, size %lld -> %lld",
1316 (unsigned long)inum, old_size, new_size);
1317 ubifs_assert(!ui->data_len);
1318 ubifs_assert(S_ISREG(inode->i_mode));
1319 ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1320
1321 sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1322 UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1323 ino = kmalloc(sz, GFP_NOFS);
1324 if (!ino)
1325 return -ENOMEM;
1326
1327 trun = (void *)ino + UBIFS_INO_NODE_SZ;
1328 trun->ch.node_type = UBIFS_TRUN_NODE;
1329 trun->inum = cpu_to_le32(inum);
1330 trun->old_size = cpu_to_le64(old_size);
1331 trun->new_size = cpu_to_le64(new_size);
1332 zero_trun_node_unused(trun);
1333
1334 dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1335 if (dlen) {
1336 /* Get last data block so it can be truncated */
1337 dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1338 blk = new_size >> UBIFS_BLOCK_SHIFT;
1339 data_key_init(c, &key, inum, blk);
1340 dbg_jnlk(&key, "last block key ");
1341 err = ubifs_tnc_lookup(c, &key, dn);
1342 if (err == -ENOENT)
1343 dlen = 0; /* Not found (so it is a hole) */
1344 else if (err)
1345 goto out_free;
1346 else {
1347 if (le32_to_cpu(dn->size) <= dlen)
1348 dlen = 0; /* Nothing to do */
1349 else {
1350 int compr_type = le16_to_cpu(dn->compr_type);
1351
1352 if (compr_type != UBIFS_COMPR_NONE) {
1353 err = recomp_data_node(c, dn, &dlen);
1354 if (err)
1355 goto out_free;
1356 } else {
1357 dn->size = cpu_to_le32(dlen);
1358 dlen += UBIFS_DATA_NODE_SZ;
1359 }
1360 zero_data_node_unused(dn);
1361 }
1362 }
1363 }
1364
1365 /* Must make reservation before allocating sequence numbers */
1366 len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1367 if (dlen)
1368 len += dlen;
1369 err = make_reservation(c, BASEHD, len);
1370 if (err)
1371 goto out_free;
1372
1373 pack_inode(c, ino, inode, 0);
1374 ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1375 if (dlen)
1376 ubifs_prep_grp_node(c, dn, dlen, 1);
1377
1378 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1379 if (err)
1380 goto out_release;
1381 if (!sync)
1382 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1383 release_head(c, BASEHD);
1384
1385 if (dlen) {
1386 sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1387 err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1388 if (err)
1389 goto out_ro;
1390 }
1391
1392 ino_key_init(c, &key, inum);
1393 err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1394 if (err)
1395 goto out_ro;
1396
1397 err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1398 if (err)
1399 goto out_ro;
1400
1401 bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1402 blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1403 data_key_init(c, &key, inum, blk);
1404
1405 bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1406 blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
1407 data_key_init(c, &to_key, inum, blk);
1408
1409 err = ubifs_tnc_remove_range(c, &key, &to_key);
1410 if (err)
1411 goto out_ro;
1412
1413 finish_reservation(c);
1414 spin_lock(&ui->ui_lock);
1415 ui->synced_i_size = ui->ui_size;
1416 spin_unlock(&ui->ui_lock);
1417 mark_inode_clean(c, ui);
1418 kfree(ino);
1419 return 0;
1420
1421 out_release:
1422 release_head(c, BASEHD);
1423 out_ro:
1424 ubifs_ro_mode(c, err);
1425 finish_reservation(c);
1426 out_free:
1427 kfree(ino);
1428 return err;
1429 }
1430
1431
1432 /**
1433 * ubifs_jnl_delete_xattr - delete an extended attribute.
1434 * @c: UBIFS file-system description object
1435 * @host: host inode
1436 * @inode: extended attribute inode
1437 * @nm: extended attribute entry name
1438 *
1439 * This function delete an extended attribute which is very similar to
1440 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1441 * updates the target inode. Returns zero in case of success and a negative
1442 * error code in case of failure.
1443 */
ubifs_jnl_delete_xattr(struct ubifs_info * c,const struct inode * host,const struct inode * inode,const struct qstr * nm)1444 int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1445 const struct inode *inode, const struct qstr *nm)
1446 {
1447 int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1448 struct ubifs_dent_node *xent;
1449 struct ubifs_ino_node *ino;
1450 union ubifs_key xent_key, key1, key2;
1451 int sync = IS_DIRSYNC(host);
1452 struct ubifs_inode *host_ui = ubifs_inode(host);
1453
1454 dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d",
1455 host->i_ino, inode->i_ino, nm->name,
1456 ubifs_inode(inode)->data_len);
1457 ubifs_assert(inode->i_nlink == 0);
1458 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1459
1460 /*
1461 * Since we are deleting the inode, we do not bother to attach any data
1462 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1463 */
1464 xlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
1465 aligned_xlen = ALIGN(xlen, 8);
1466 hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1467 len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1468
1469 xent = kzalloc(len, GFP_NOFS);
1470 if (!xent)
1471 return -ENOMEM;
1472
1473 /* Make reservation before allocating sequence numbers */
1474 err = make_reservation(c, BASEHD, len);
1475 if (err) {
1476 kfree(xent);
1477 return err;
1478 }
1479
1480 xent->ch.node_type = UBIFS_XENT_NODE;
1481 xent_key_init(c, &xent_key, host->i_ino, nm);
1482 key_write(c, &xent_key, xent->key);
1483 xent->inum = 0;
1484 xent->type = get_dent_type(inode->i_mode);
1485 xent->nlen = cpu_to_le16(nm->len);
1486 memcpy(xent->name, nm->name, nm->len);
1487 xent->name[nm->len] = '\0';
1488 zero_dent_node_unused(xent);
1489 ubifs_prep_grp_node(c, xent, xlen, 0);
1490
1491 ino = (void *)xent + aligned_xlen;
1492 pack_inode(c, ino, inode, 0);
1493 ino = (void *)ino + UBIFS_INO_NODE_SZ;
1494 pack_inode(c, ino, host, 1);
1495
1496 err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1497 if (!sync && !err)
1498 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1499 release_head(c, BASEHD);
1500 kfree(xent);
1501 if (err)
1502 goto out_ro;
1503
1504 /* Remove the extended attribute entry from TNC */
1505 err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1506 if (err)
1507 goto out_ro;
1508 err = ubifs_add_dirt(c, lnum, xlen);
1509 if (err)
1510 goto out_ro;
1511
1512 /*
1513 * Remove all nodes belonging to the extended attribute inode from TNC.
1514 * Well, there actually must be only one node - the inode itself.
1515 */
1516 lowest_ino_key(c, &key1, inode->i_ino);
1517 highest_ino_key(c, &key2, inode->i_ino);
1518 err = ubifs_tnc_remove_range(c, &key1, &key2);
1519 if (err)
1520 goto out_ro;
1521 err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1522 if (err)
1523 goto out_ro;
1524
1525 /* And update TNC with the new host inode position */
1526 ino_key_init(c, &key1, host->i_ino);
1527 err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1528 if (err)
1529 goto out_ro;
1530
1531 finish_reservation(c);
1532 spin_lock(&host_ui->ui_lock);
1533 host_ui->synced_i_size = host_ui->ui_size;
1534 spin_unlock(&host_ui->ui_lock);
1535 mark_inode_clean(c, host_ui);
1536 return 0;
1537
1538 out_ro:
1539 ubifs_ro_mode(c, err);
1540 finish_reservation(c);
1541 return err;
1542 }
1543
1544 /**
1545 * ubifs_jnl_change_xattr - change an extended attribute.
1546 * @c: UBIFS file-system description object
1547 * @inode: extended attribute inode
1548 * @host: host inode
1549 *
1550 * This function writes the updated version of an extended attribute inode and
1551 * the host inode to the journal (to the base head). The host inode is written
1552 * after the extended attribute inode in order to guarantee that the extended
1553 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1554 * consequently, the write-buffer is synchronized. This function returns zero
1555 * in case of success and a negative error code in case of failure.
1556 */
ubifs_jnl_change_xattr(struct ubifs_info * c,const struct inode * inode,const struct inode * host)1557 int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1558 const struct inode *host)
1559 {
1560 int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1561 struct ubifs_inode *host_ui = ubifs_inode(host);
1562 struct ubifs_ino_node *ino;
1563 union ubifs_key key;
1564 int sync = IS_DIRSYNC(host);
1565
1566 dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1567 ubifs_assert(host->i_nlink > 0);
1568 ubifs_assert(inode->i_nlink > 0);
1569 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1570
1571 len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1572 len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1573 aligned_len1 = ALIGN(len1, 8);
1574 aligned_len = aligned_len1 + ALIGN(len2, 8);
1575
1576 ino = kzalloc(aligned_len, GFP_NOFS);
1577 if (!ino)
1578 return -ENOMEM;
1579
1580 /* Make reservation before allocating sequence numbers */
1581 err = make_reservation(c, BASEHD, aligned_len);
1582 if (err)
1583 goto out_free;
1584
1585 pack_inode(c, ino, host, 0);
1586 pack_inode(c, (void *)ino + aligned_len1, inode, 1);
1587
1588 err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1589 if (!sync && !err) {
1590 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1591
1592 ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1593 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1594 }
1595 release_head(c, BASEHD);
1596 if (err)
1597 goto out_ro;
1598
1599 ino_key_init(c, &key, host->i_ino);
1600 err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1601 if (err)
1602 goto out_ro;
1603
1604 ino_key_init(c, &key, inode->i_ino);
1605 err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1606 if (err)
1607 goto out_ro;
1608
1609 finish_reservation(c);
1610 spin_lock(&host_ui->ui_lock);
1611 host_ui->synced_i_size = host_ui->ui_size;
1612 spin_unlock(&host_ui->ui_lock);
1613 mark_inode_clean(c, host_ui);
1614 kfree(ino);
1615 return 0;
1616
1617 out_ro:
1618 ubifs_ro_mode(c, err);
1619 finish_reservation(c);
1620 out_free:
1621 kfree(ino);
1622 return err;
1623 }
1624
1625