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 */ 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 */ 78 static inline void zero_dent_node_unused(struct ubifs_dent_node *dent) 79 { 80 dent->padding1 = 0; 81 } 82 83 /** 84 * zero_trun_node_unused - zero out unused fields of an on-flash truncation 85 * node. 86 * @trun: the truncation node to zero out 87 */ 88 static inline void zero_trun_node_unused(struct ubifs_trun_node *trun) 89 { 90 memset(trun->padding, 0, 12); 91 } 92 93 /** 94 * reserve_space - reserve space in the journal. 95 * @c: UBIFS file-system description object 96 * @jhead: journal head number 97 * @len: node length 98 * 99 * This function reserves space in journal head @head. If the reservation 100 * succeeded, the journal head stays locked and later has to be unlocked using 101 * 'release_head()'. 'write_node()' and 'write_head()' functions also unlock 102 * it. Returns zero in case of success, %-EAGAIN if commit has to be done, and 103 * other negative error codes in case of other failures. 104 */ 105 static int reserve_space(struct ubifs_info *c, int jhead, int len) 106 { 107 int err = 0, err1, retries = 0, avail, lnum, offs, squeeze; 108 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf; 109 110 /* 111 * Typically, the base head has smaller nodes written to it, so it is 112 * better to try to allocate space at the ends of eraseblocks. This is 113 * what the squeeze parameter does. 114 */ 115 ubifs_assert(!c->ro_media && !c->ro_mount); 116 squeeze = (jhead == BASEHD); 117 again: 118 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); 119 120 if (c->ro_error) { 121 err = -EROFS; 122 goto out_unlock; 123 } 124 125 avail = c->leb_size - wbuf->offs - wbuf->used; 126 if (wbuf->lnum != -1 && avail >= len) 127 return 0; 128 129 /* 130 * Write buffer wasn't seek'ed or there is no enough space - look for an 131 * LEB with some empty space. 132 */ 133 lnum = ubifs_find_free_space(c, len, &offs, squeeze); 134 if (lnum >= 0) 135 goto out; 136 137 err = lnum; 138 if (err != -ENOSPC) 139 goto out_unlock; 140 141 /* 142 * No free space, we have to run garbage collector to make 143 * some. But the write-buffer mutex has to be unlocked because 144 * GC also takes it. 145 */ 146 dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead)); 147 mutex_unlock(&wbuf->io_mutex); 148 149 lnum = ubifs_garbage_collect(c, 0); 150 if (lnum < 0) { 151 err = lnum; 152 if (err != -ENOSPC) 153 return err; 154 155 /* 156 * GC could not make a free LEB. But someone else may 157 * have allocated new bud for this journal head, 158 * because we dropped @wbuf->io_mutex, so try once 159 * again. 160 */ 161 dbg_jnl("GC couldn't make a free LEB for jhead %s", 162 dbg_jhead(jhead)); 163 if (retries++ < 2) { 164 dbg_jnl("retry (%d)", retries); 165 goto again; 166 } 167 168 dbg_jnl("return -ENOSPC"); 169 return err; 170 } 171 172 mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead); 173 dbg_jnl("got LEB %d for jhead %s", lnum, dbg_jhead(jhead)); 174 avail = c->leb_size - wbuf->offs - wbuf->used; 175 176 if (wbuf->lnum != -1 && avail >= len) { 177 /* 178 * Someone else has switched the journal head and we have 179 * enough space now. This happens when more than one process is 180 * trying to write to the same journal head at the same time. 181 */ 182 dbg_jnl("return LEB %d back, already have LEB %d:%d", 183 lnum, wbuf->lnum, wbuf->offs + wbuf->used); 184 err = ubifs_return_leb(c, lnum); 185 if (err) 186 goto out_unlock; 187 return 0; 188 } 189 190 offs = 0; 191 192 out: 193 /* 194 * Make sure we synchronize the write-buffer before we add the new bud 195 * to the log. Otherwise we may have a power cut after the log 196 * reference node for the last bud (@lnum) is written but before the 197 * write-buffer data are written to the next-to-last bud 198 * (@wbuf->lnum). And the effect would be that the recovery would see 199 * that there is corruption in the next-to-last bud. 200 */ 201 err = ubifs_wbuf_sync_nolock(wbuf); 202 if (err) 203 goto out_return; 204 err = ubifs_add_bud_to_log(c, jhead, lnum, offs); 205 if (err) 206 goto out_return; 207 err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs); 208 if (err) 209 goto out_unlock; 210 211 return 0; 212 213 out_unlock: 214 mutex_unlock(&wbuf->io_mutex); 215 return err; 216 217 out_return: 218 /* An error occurred and the LEB has to be returned to lprops */ 219 ubifs_assert(err < 0); 220 err1 = ubifs_return_leb(c, lnum); 221 if (err1 && err == -EAGAIN) 222 /* 223 * Return original error code only if it is not %-EAGAIN, 224 * which is not really an error. Otherwise, return the error 225 * code of 'ubifs_return_leb()'. 226 */ 227 err = err1; 228 mutex_unlock(&wbuf->io_mutex); 229 return err; 230 } 231 232 /** 233 * write_node - write node to a journal head. 234 * @c: UBIFS file-system description object 235 * @jhead: journal head 236 * @node: node to write 237 * @len: node length 238 * @lnum: LEB number written is returned here 239 * @offs: offset written is returned here 240 * 241 * This function writes a node to reserved space of journal head @jhead. 242 * Returns zero in case of success and a negative error code in case of 243 * failure. 244 */ 245 static int write_node(struct ubifs_info *c, int jhead, void *node, int len, 246 int *lnum, int *offs) 247 { 248 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf; 249 250 ubifs_assert(jhead != GCHD); 251 252 *lnum = c->jheads[jhead].wbuf.lnum; 253 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used; 254 255 dbg_jnl("jhead %s, LEB %d:%d, len %d", 256 dbg_jhead(jhead), *lnum, *offs, len); 257 ubifs_prepare_node(c, node, len, 0); 258 259 return ubifs_wbuf_write_nolock(wbuf, node, len); 260 } 261 262 /** 263 * write_head - write data to a journal head. 264 * @c: UBIFS file-system description object 265 * @jhead: journal head 266 * @buf: buffer to write 267 * @len: length to write 268 * @lnum: LEB number written is returned here 269 * @offs: offset written is returned here 270 * @sync: non-zero if the write-buffer has to by synchronized 271 * 272 * This function is the same as 'write_node()' but it does not assume the 273 * buffer it is writing is a node, so it does not prepare it (which means 274 * initializing common header and calculating CRC). 275 */ 276 static int write_head(struct ubifs_info *c, int jhead, void *buf, int len, 277 int *lnum, int *offs, int sync) 278 { 279 int err; 280 struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf; 281 282 ubifs_assert(jhead != GCHD); 283 284 *lnum = c->jheads[jhead].wbuf.lnum; 285 *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used; 286 dbg_jnl("jhead %s, LEB %d:%d, len %d", 287 dbg_jhead(jhead), *lnum, *offs, len); 288 289 err = ubifs_wbuf_write_nolock(wbuf, buf, len); 290 if (err) 291 return err; 292 if (sync) 293 err = ubifs_wbuf_sync_nolock(wbuf); 294 return err; 295 } 296 297 /** 298 * make_reservation - reserve journal space. 299 * @c: UBIFS file-system description object 300 * @jhead: journal head 301 * @len: how many bytes to reserve 302 * 303 * This function makes space reservation in journal head @jhead. The function 304 * takes the commit lock and locks the journal head, and the caller has to 305 * unlock the head and finish the reservation with 'finish_reservation()'. 306 * Returns zero in case of success and a negative error code in case of 307 * failure. 308 * 309 * Note, the journal head may be unlocked as soon as the data is written, while 310 * the commit lock has to be released after the data has been added to the 311 * TNC. 312 */ 313 static int make_reservation(struct ubifs_info *c, int jhead, int len) 314 { 315 int err, cmt_retries = 0, nospc_retries = 0; 316 317 again: 318 down_read(&c->commit_sem); 319 err = reserve_space(c, jhead, len); 320 if (!err) 321 return 0; 322 up_read(&c->commit_sem); 323 324 if (err == -ENOSPC) { 325 /* 326 * GC could not make any progress. We should try to commit 327 * once because it could make some dirty space and GC would 328 * make progress, so make the error -EAGAIN so that the below 329 * will commit and re-try. 330 */ 331 if (nospc_retries++ < 2) { 332 dbg_jnl("no space, retry"); 333 err = -EAGAIN; 334 } 335 336 /* 337 * This means that the budgeting is incorrect. We always have 338 * to be able to write to the media, because all operations are 339 * budgeted. Deletions are not budgeted, though, but we reserve 340 * an extra LEB for them. 341 */ 342 } 343 344 if (err != -EAGAIN) 345 goto out; 346 347 /* 348 * -EAGAIN means that the journal is full or too large, or the above 349 * code wants to do one commit. Do this and re-try. 350 */ 351 if (cmt_retries > 128) { 352 /* 353 * This should not happen unless the journal size limitations 354 * are too tough. 355 */ 356 ubifs_err(c, "stuck in space allocation"); 357 err = -ENOSPC; 358 goto out; 359 } else if (cmt_retries > 32) 360 ubifs_warn(c, "too many space allocation re-tries (%d)", 361 cmt_retries); 362 363 dbg_jnl("-EAGAIN, commit and retry (retried %d times)", 364 cmt_retries); 365 cmt_retries += 1; 366 367 err = ubifs_run_commit(c); 368 if (err) 369 return err; 370 goto again; 371 372 out: 373 ubifs_err(c, "cannot reserve %d bytes in jhead %d, error %d", 374 len, jhead, err); 375 if (err == -ENOSPC) { 376 /* This are some budgeting problems, print useful information */ 377 down_write(&c->commit_sem); 378 dump_stack(); 379 ubifs_dump_budg(c, &c->bi); 380 ubifs_dump_lprops(c); 381 cmt_retries = dbg_check_lprops(c); 382 up_write(&c->commit_sem); 383 } 384 return err; 385 } 386 387 /** 388 * release_head - release a journal head. 389 * @c: UBIFS file-system description object 390 * @jhead: journal head 391 * 392 * This function releases journal head @jhead which was locked by 393 * the 'make_reservation()' function. It has to be called after each successful 394 * 'make_reservation()' invocation. 395 */ 396 static inline void release_head(struct ubifs_info *c, int jhead) 397 { 398 mutex_unlock(&c->jheads[jhead].wbuf.io_mutex); 399 } 400 401 /** 402 * finish_reservation - finish a reservation. 403 * @c: UBIFS file-system description object 404 * 405 * This function finishes journal space reservation. It must be called after 406 * 'make_reservation()'. 407 */ 408 static void finish_reservation(struct ubifs_info *c) 409 { 410 up_read(&c->commit_sem); 411 } 412 413 /** 414 * get_dent_type - translate VFS inode mode to UBIFS directory entry type. 415 * @mode: inode mode 416 */ 417 static int get_dent_type(int mode) 418 { 419 switch (mode & S_IFMT) { 420 case S_IFREG: 421 return UBIFS_ITYPE_REG; 422 case S_IFDIR: 423 return UBIFS_ITYPE_DIR; 424 case S_IFLNK: 425 return UBIFS_ITYPE_LNK; 426 case S_IFBLK: 427 return UBIFS_ITYPE_BLK; 428 case S_IFCHR: 429 return UBIFS_ITYPE_CHR; 430 case S_IFIFO: 431 return UBIFS_ITYPE_FIFO; 432 case S_IFSOCK: 433 return UBIFS_ITYPE_SOCK; 434 default: 435 BUG(); 436 } 437 return 0; 438 } 439 440 /** 441 * pack_inode - pack an inode node. 442 * @c: UBIFS file-system description object 443 * @ino: buffer in which to pack inode node 444 * @inode: inode to pack 445 * @last: indicates the last node of the group 446 */ 447 static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino, 448 const struct inode *inode, int last) 449 { 450 int data_len = 0, last_reference = !inode->i_nlink; 451 struct ubifs_inode *ui = ubifs_inode(inode); 452 453 ino->ch.node_type = UBIFS_INO_NODE; 454 ino_key_init_flash(c, &ino->key, inode->i_ino); 455 ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum); 456 ino->atime_sec = cpu_to_le64(inode->i_atime.tv_sec); 457 ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); 458 ino->ctime_sec = cpu_to_le64(inode->i_ctime.tv_sec); 459 ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); 460 ino->mtime_sec = cpu_to_le64(inode->i_mtime.tv_sec); 461 ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); 462 ino->uid = cpu_to_le32(i_uid_read(inode)); 463 ino->gid = cpu_to_le32(i_gid_read(inode)); 464 ino->mode = cpu_to_le32(inode->i_mode); 465 ino->flags = cpu_to_le32(ui->flags); 466 ino->size = cpu_to_le64(ui->ui_size); 467 ino->nlink = cpu_to_le32(inode->i_nlink); 468 ino->compr_type = cpu_to_le16(ui->compr_type); 469 ino->data_len = cpu_to_le32(ui->data_len); 470 ino->xattr_cnt = cpu_to_le32(ui->xattr_cnt); 471 ino->xattr_size = cpu_to_le32(ui->xattr_size); 472 ino->xattr_names = cpu_to_le32(ui->xattr_names); 473 zero_ino_node_unused(ino); 474 475 /* 476 * Drop the attached data if this is a deletion inode, the data is not 477 * needed anymore. 478 */ 479 if (!last_reference) { 480 memcpy(ino->data, ui->data, ui->data_len); 481 data_len = ui->data_len; 482 } 483 484 ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last); 485 } 486 487 /** 488 * mark_inode_clean - mark UBIFS inode as clean. 489 * @c: UBIFS file-system description object 490 * @ui: UBIFS inode to mark as clean 491 * 492 * This helper function marks UBIFS inode @ui as clean by cleaning the 493 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the 494 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would 495 * just do nothing. 496 */ 497 static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui) 498 { 499 if (ui->dirty) 500 ubifs_release_dirty_inode_budget(c, ui); 501 ui->dirty = 0; 502 } 503 504 static void set_dent_cookie(struct ubifs_info *c, struct ubifs_dent_node *dent) 505 { 506 if (c->double_hash) 507 dent->cookie = prandom_u32(); 508 else 509 dent->cookie = 0; 510 } 511 512 /** 513 * ubifs_jnl_update - update inode. 514 * @c: UBIFS file-system description object 515 * @dir: parent inode or host inode in case of extended attributes 516 * @nm: directory entry name 517 * @inode: inode to update 518 * @deletion: indicates a directory entry deletion i.e unlink or rmdir 519 * @xent: non-zero if the directory entry is an extended attribute entry 520 * 521 * This function updates an inode by writing a directory entry (or extended 522 * attribute entry), the inode itself, and the parent directory inode (or the 523 * host inode) to the journal. 524 * 525 * The function writes the host inode @dir last, which is important in case of 526 * extended attributes. Indeed, then we guarantee that if the host inode gets 527 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed, 528 * the extended attribute inode gets flushed too. And this is exactly what the 529 * user expects - synchronizing the host inode synchronizes its extended 530 * attributes. Similarly, this guarantees that if @dir is synchronized, its 531 * directory entry corresponding to @nm gets synchronized too. 532 * 533 * If the inode (@inode) or the parent directory (@dir) are synchronous, this 534 * function synchronizes the write-buffer. 535 * 536 * This function marks the @dir and @inode inodes as clean and returns zero on 537 * success. In case of failure, a negative error code is returned. 538 */ 539 int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir, 540 const struct fscrypt_name *nm, const struct inode *inode, 541 int deletion, int xent) 542 { 543 int err, dlen, ilen, len, lnum, ino_offs, dent_offs; 544 int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir); 545 int last_reference = !!(deletion && inode->i_nlink == 0); 546 struct ubifs_inode *ui = ubifs_inode(inode); 547 struct ubifs_inode *host_ui = ubifs_inode(dir); 548 struct ubifs_dent_node *dent; 549 struct ubifs_ino_node *ino; 550 union ubifs_key dent_key, ino_key; 551 552 //dbg_jnl("ino %lu, dent '%.*s', data len %d in dir ino %lu", 553 // inode->i_ino, nm->len, nm->name, ui->data_len, dir->i_ino); 554 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex)); 555 556 dlen = UBIFS_DENT_NODE_SZ + fname_len(nm) + 1; 557 ilen = UBIFS_INO_NODE_SZ; 558 559 /* 560 * If the last reference to the inode is being deleted, then there is 561 * no need to attach and write inode data, it is being deleted anyway. 562 * And if the inode is being deleted, no need to synchronize 563 * write-buffer even if the inode is synchronous. 564 */ 565 if (!last_reference) { 566 ilen += ui->data_len; 567 sync |= IS_SYNC(inode); 568 } 569 570 aligned_dlen = ALIGN(dlen, 8); 571 aligned_ilen = ALIGN(ilen, 8); 572 573 len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ; 574 /* Make sure to also account for extended attributes */ 575 len += host_ui->data_len; 576 577 dent = kmalloc(len, GFP_NOFS); 578 if (!dent) 579 return -ENOMEM; 580 581 /* Make reservation before allocating sequence numbers */ 582 err = make_reservation(c, BASEHD, len); 583 if (err) 584 goto out_free; 585 586 if (!xent) { 587 dent->ch.node_type = UBIFS_DENT_NODE; 588 dent_key_init(c, &dent_key, dir->i_ino, nm); 589 } else { 590 dent->ch.node_type = UBIFS_XENT_NODE; 591 xent_key_init(c, &dent_key, dir->i_ino, nm); 592 } 593 594 key_write(c, &dent_key, dent->key); 595 dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino); 596 dent->type = get_dent_type(inode->i_mode); 597 dent->nlen = cpu_to_le16(fname_len(nm)); 598 memcpy(dent->name, fname_name(nm), fname_len(nm)); 599 dent->name[fname_len(nm)] = '\0'; 600 set_dent_cookie(c, dent); 601 602 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 */ 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, compr_len; 701 int dlen = COMPRESSED_DATA_NODE_BUF_SZ, allocated = 1; 702 struct ubifs_inode *ui = ubifs_inode(inode); 703 bool encrypted = ubifs_crypt_is_encrypted(inode); 704 705 dbg_jnlk(key, "ino %lu, blk %u, len %d, key ", 706 (unsigned long)key_inum(c, key), key_block(c, key), len); 707 ubifs_assert(len <= UBIFS_BLOCK_SIZE); 708 709 if (encrypted) 710 dlen += UBIFS_CIPHER_BLOCK_SIZE; 711 712 data = kmalloc(dlen, GFP_NOFS | __GFP_NOWARN); 713 if (!data) { 714 /* 715 * Fall-back to the write reserve buffer. Note, we might be 716 * currently on the memory reclaim path, when the kernel is 717 * trying to free some memory by writing out dirty pages. The 718 * write reserve buffer helps us to guarantee that we are 719 * always able to write the data. 720 */ 721 allocated = 0; 722 mutex_lock(&c->write_reserve_mutex); 723 data = c->write_reserve_buf; 724 } 725 726 data->ch.node_type = UBIFS_DATA_NODE; 727 key_write(c, key, &data->key); 728 data->size = cpu_to_le32(len); 729 730 if (!(ui->flags & UBIFS_COMPR_FL)) 731 /* Compression is disabled for this inode */ 732 compr_type = UBIFS_COMPR_NONE; 733 else 734 compr_type = ui->compr_type; 735 736 out_len = compr_len = dlen - UBIFS_DATA_NODE_SZ; 737 ubifs_compress(c, buf, len, &data->data, &compr_len, &compr_type); 738 ubifs_assert(compr_len <= UBIFS_BLOCK_SIZE); 739 740 if (encrypted) { 741 err = ubifs_encrypt(inode, data, compr_len, &out_len, key_block(c, key)); 742 if (err) 743 goto out_free; 744 745 } else { 746 data->compr_size = 0; 747 } 748 749 dlen = UBIFS_DATA_NODE_SZ + out_len; 750 data->compr_type = cpu_to_le16(compr_type); 751 752 /* Make reservation before allocating sequence numbers */ 753 err = make_reservation(c, DATAHD, dlen); 754 if (err) 755 goto out_free; 756 757 err = write_node(c, DATAHD, data, dlen, &lnum, &offs); 758 if (err) 759 goto out_release; 760 ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key)); 761 release_head(c, DATAHD); 762 763 err = ubifs_tnc_add(c, key, lnum, offs, dlen); 764 if (err) 765 goto out_ro; 766 767 finish_reservation(c); 768 if (!allocated) 769 mutex_unlock(&c->write_reserve_mutex); 770 else 771 kfree(data); 772 return 0; 773 774 out_release: 775 release_head(c, DATAHD); 776 out_ro: 777 ubifs_ro_mode(c, err); 778 finish_reservation(c); 779 out_free: 780 if (!allocated) 781 mutex_unlock(&c->write_reserve_mutex); 782 else 783 kfree(data); 784 return err; 785 } 786 787 /** 788 * ubifs_jnl_write_inode - flush inode to the journal. 789 * @c: UBIFS file-system description object 790 * @inode: inode to flush 791 * 792 * This function writes inode @inode to the journal. If the inode is 793 * synchronous, it also synchronizes the write-buffer. Returns zero in case of 794 * success and a negative error code in case of failure. 795 */ 796 int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode) 797 { 798 int err, lnum, offs; 799 struct ubifs_ino_node *ino; 800 struct ubifs_inode *ui = ubifs_inode(inode); 801 int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink; 802 803 dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink); 804 805 /* 806 * If the inode is being deleted, do not write the attached data. No 807 * need to synchronize the write-buffer either. 808 */ 809 if (!last_reference) { 810 len += ui->data_len; 811 sync = IS_SYNC(inode); 812 } 813 ino = kmalloc(len, GFP_NOFS); 814 if (!ino) 815 return -ENOMEM; 816 817 /* Make reservation before allocating sequence numbers */ 818 err = make_reservation(c, BASEHD, len); 819 if (err) 820 goto out_free; 821 822 pack_inode(c, ino, inode, 1); 823 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync); 824 if (err) 825 goto out_release; 826 if (!sync) 827 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, 828 inode->i_ino); 829 release_head(c, BASEHD); 830 831 if (last_reference) { 832 err = ubifs_tnc_remove_ino(c, inode->i_ino); 833 if (err) 834 goto out_ro; 835 ubifs_delete_orphan(c, inode->i_ino); 836 err = ubifs_add_dirt(c, lnum, len); 837 } else { 838 union ubifs_key key; 839 840 ino_key_init(c, &key, inode->i_ino); 841 err = ubifs_tnc_add(c, &key, lnum, offs, len); 842 } 843 if (err) 844 goto out_ro; 845 846 finish_reservation(c); 847 spin_lock(&ui->ui_lock); 848 ui->synced_i_size = ui->ui_size; 849 spin_unlock(&ui->ui_lock); 850 kfree(ino); 851 return 0; 852 853 out_release: 854 release_head(c, BASEHD); 855 out_ro: 856 ubifs_ro_mode(c, err); 857 finish_reservation(c); 858 out_free: 859 kfree(ino); 860 return err; 861 } 862 863 /** 864 * ubifs_jnl_delete_inode - delete an inode. 865 * @c: UBIFS file-system description object 866 * @inode: inode to delete 867 * 868 * This function deletes inode @inode which includes removing it from orphans, 869 * deleting it from TNC and, in some cases, writing a deletion inode to the 870 * journal. 871 * 872 * When regular file inodes are unlinked or a directory inode is removed, the 873 * 'ubifs_jnl_update()' function writes a corresponding deletion inode and 874 * direntry to the media, and adds the inode to orphans. After this, when the 875 * last reference to this inode has been dropped, this function is called. In 876 * general, it has to write one more deletion inode to the media, because if 877 * a commit happened between 'ubifs_jnl_update()' and 878 * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal 879 * anymore, and in fact it might not be on the flash anymore, because it might 880 * have been garbage-collected already. And for optimization reasons UBIFS does 881 * not read the orphan area if it has been unmounted cleanly, so it would have 882 * no indication in the journal that there is a deleted inode which has to be 883 * removed from TNC. 884 * 885 * However, if there was no commit between 'ubifs_jnl_update()' and 886 * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion 887 * inode to the media for the second time. And this is quite a typical case. 888 * 889 * This function returns zero in case of success and a negative error code in 890 * case of failure. 891 */ 892 int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode) 893 { 894 int err; 895 struct ubifs_inode *ui = ubifs_inode(inode); 896 897 ubifs_assert(inode->i_nlink == 0); 898 899 if (ui->del_cmtno != c->cmt_no) 900 /* A commit happened for sure */ 901 return ubifs_jnl_write_inode(c, inode); 902 903 down_read(&c->commit_sem); 904 /* 905 * Check commit number again, because the first test has been done 906 * without @c->commit_sem, so a commit might have happened. 907 */ 908 if (ui->del_cmtno != c->cmt_no) { 909 up_read(&c->commit_sem); 910 return ubifs_jnl_write_inode(c, inode); 911 } 912 913 err = ubifs_tnc_remove_ino(c, inode->i_ino); 914 if (err) 915 ubifs_ro_mode(c, err); 916 else 917 ubifs_delete_orphan(c, inode->i_ino); 918 up_read(&c->commit_sem); 919 return err; 920 } 921 922 /** 923 * ubifs_jnl_xrename - cross rename two directory entries. 924 * @c: UBIFS file-system description object 925 * @fst_dir: parent inode of 1st directory entry to exchange 926 * @fst_inode: 1st inode to exchange 927 * @fst_nm: name of 1st inode to exchange 928 * @snd_dir: parent inode of 2nd directory entry to exchange 929 * @snd_inode: 2nd inode to exchange 930 * @snd_nm: name of 2nd inode to exchange 931 * @sync: non-zero if the write-buffer has to be synchronized 932 * 933 * This function implements the cross rename operation which may involve 934 * writing 2 inodes and 2 directory entries. It marks the written inodes as clean 935 * and returns zero on success. In case of failure, a negative error code is 936 * returned. 937 */ 938 int ubifs_jnl_xrename(struct ubifs_info *c, const struct inode *fst_dir, 939 const struct inode *fst_inode, 940 const struct fscrypt_name *fst_nm, 941 const struct inode *snd_dir, 942 const struct inode *snd_inode, 943 const struct fscrypt_name *snd_nm, int sync) 944 { 945 union ubifs_key key; 946 struct ubifs_dent_node *dent1, *dent2; 947 int err, dlen1, dlen2, lnum, offs, len, plen = UBIFS_INO_NODE_SZ; 948 int aligned_dlen1, aligned_dlen2; 949 int twoparents = (fst_dir != snd_dir); 950 void *p; 951 952 //dbg_jnl("dent '%pd' in dir ino %lu between dent '%pd' in dir ino %lu", 953 // fst_dentry, fst_dir->i_ino, snd_dentry, snd_dir->i_ino); 954 955 ubifs_assert(ubifs_inode(fst_dir)->data_len == 0); 956 ubifs_assert(ubifs_inode(snd_dir)->data_len == 0); 957 ubifs_assert(mutex_is_locked(&ubifs_inode(fst_dir)->ui_mutex)); 958 ubifs_assert(mutex_is_locked(&ubifs_inode(snd_dir)->ui_mutex)); 959 960 dlen1 = UBIFS_DENT_NODE_SZ + fname_len(snd_nm) + 1; 961 dlen2 = UBIFS_DENT_NODE_SZ + fname_len(fst_nm) + 1; 962 aligned_dlen1 = ALIGN(dlen1, 8); 963 aligned_dlen2 = ALIGN(dlen2, 8); 964 965 len = aligned_dlen1 + aligned_dlen2 + ALIGN(plen, 8); 966 if (twoparents) 967 len += plen; 968 969 dent1 = kmalloc(len, GFP_NOFS); 970 if (!dent1) 971 return -ENOMEM; 972 973 /* Make reservation before allocating sequence numbers */ 974 err = make_reservation(c, BASEHD, len); 975 if (err) 976 goto out_free; 977 978 /* Make new dent for 1st entry */ 979 dent1->ch.node_type = UBIFS_DENT_NODE; 980 dent_key_init_flash(c, &dent1->key, snd_dir->i_ino, snd_nm); 981 dent1->inum = cpu_to_le64(fst_inode->i_ino); 982 dent1->type = get_dent_type(fst_inode->i_mode); 983 dent1->nlen = cpu_to_le16(fname_len(snd_nm)); 984 memcpy(dent1->name, fname_name(snd_nm), fname_len(snd_nm)); 985 dent1->name[fname_len(snd_nm)] = '\0'; 986 zero_dent_node_unused(dent1); 987 ubifs_prep_grp_node(c, dent1, dlen1, 0); 988 989 /* Make new dent for 2nd entry */ 990 dent2 = (void *)dent1 + aligned_dlen1; 991 dent2->ch.node_type = UBIFS_DENT_NODE; 992 dent_key_init_flash(c, &dent2->key, fst_dir->i_ino, fst_nm); 993 dent2->inum = cpu_to_le64(snd_inode->i_ino); 994 dent2->type = get_dent_type(snd_inode->i_mode); 995 dent2->nlen = cpu_to_le16(fname_len(fst_nm)); 996 memcpy(dent2->name, fname_name(fst_nm), fname_len(fst_nm)); 997 dent2->name[fname_len(fst_nm)] = '\0'; 998 zero_dent_node_unused(dent2); 999 ubifs_prep_grp_node(c, dent2, dlen2, 0); 1000 1001 p = (void *)dent2 + aligned_dlen2; 1002 if (!twoparents) 1003 pack_inode(c, p, fst_dir, 1); 1004 else { 1005 pack_inode(c, p, fst_dir, 0); 1006 p += ALIGN(plen, 8); 1007 pack_inode(c, p, snd_dir, 1); 1008 } 1009 1010 err = write_head(c, BASEHD, dent1, len, &lnum, &offs, sync); 1011 if (err) 1012 goto out_release; 1013 if (!sync) { 1014 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf; 1015 1016 ubifs_wbuf_add_ino_nolock(wbuf, fst_dir->i_ino); 1017 ubifs_wbuf_add_ino_nolock(wbuf, snd_dir->i_ino); 1018 } 1019 release_head(c, BASEHD); 1020 1021 dent_key_init(c, &key, snd_dir->i_ino, snd_nm); 1022 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, snd_nm); 1023 if (err) 1024 goto out_ro; 1025 1026 offs += aligned_dlen1; 1027 dent_key_init(c, &key, fst_dir->i_ino, fst_nm); 1028 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, fst_nm); 1029 if (err) 1030 goto out_ro; 1031 1032 offs += aligned_dlen2; 1033 1034 ino_key_init(c, &key, fst_dir->i_ino); 1035 err = ubifs_tnc_add(c, &key, lnum, offs, plen); 1036 if (err) 1037 goto out_ro; 1038 1039 if (twoparents) { 1040 offs += ALIGN(plen, 8); 1041 ino_key_init(c, &key, snd_dir->i_ino); 1042 err = ubifs_tnc_add(c, &key, lnum, offs, plen); 1043 if (err) 1044 goto out_ro; 1045 } 1046 1047 finish_reservation(c); 1048 1049 mark_inode_clean(c, ubifs_inode(fst_dir)); 1050 if (twoparents) 1051 mark_inode_clean(c, ubifs_inode(snd_dir)); 1052 kfree(dent1); 1053 return 0; 1054 1055 out_release: 1056 release_head(c, BASEHD); 1057 out_ro: 1058 ubifs_ro_mode(c, err); 1059 finish_reservation(c); 1060 out_free: 1061 kfree(dent1); 1062 return err; 1063 } 1064 1065 /** 1066 * ubifs_jnl_rename - rename a directory entry. 1067 * @c: UBIFS file-system description object 1068 * @old_dir: parent inode of directory entry to rename 1069 * @old_dentry: directory entry to rename 1070 * @new_dir: parent inode of directory entry to rename 1071 * @new_dentry: new directory entry (or directory entry to replace) 1072 * @sync: non-zero if the write-buffer has to be synchronized 1073 * 1074 * This function implements the re-name operation which may involve writing up 1075 * to 4 inodes and 2 directory entries. It marks the written inodes as clean 1076 * and returns zero on success. In case of failure, a negative error code is 1077 * returned. 1078 */ 1079 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir, 1080 const struct inode *old_inode, 1081 const struct fscrypt_name *old_nm, 1082 const struct inode *new_dir, 1083 const struct inode *new_inode, 1084 const struct fscrypt_name *new_nm, 1085 const struct inode *whiteout, int sync) 1086 { 1087 void *p; 1088 union ubifs_key key; 1089 struct ubifs_dent_node *dent, *dent2; 1090 int err, dlen1, dlen2, ilen, lnum, offs, len; 1091 int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ; 1092 int last_reference = !!(new_inode && new_inode->i_nlink == 0); 1093 int move = (old_dir != new_dir); 1094 struct ubifs_inode *uninitialized_var(new_ui); 1095 1096 //dbg_jnl("dent '%pd' in dir ino %lu to dent '%pd' in dir ino %lu", 1097 // old_dentry, old_dir->i_ino, new_dentry, new_dir->i_ino); 1098 ubifs_assert(ubifs_inode(old_dir)->data_len == 0); 1099 ubifs_assert(ubifs_inode(new_dir)->data_len == 0); 1100 ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex)); 1101 ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex)); 1102 1103 dlen1 = UBIFS_DENT_NODE_SZ + fname_len(new_nm) + 1; 1104 dlen2 = UBIFS_DENT_NODE_SZ + fname_len(old_nm) + 1; 1105 if (new_inode) { 1106 new_ui = ubifs_inode(new_inode); 1107 ubifs_assert(mutex_is_locked(&new_ui->ui_mutex)); 1108 ilen = UBIFS_INO_NODE_SZ; 1109 if (!last_reference) 1110 ilen += new_ui->data_len; 1111 } else 1112 ilen = 0; 1113 1114 aligned_dlen1 = ALIGN(dlen1, 8); 1115 aligned_dlen2 = ALIGN(dlen2, 8); 1116 len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8); 1117 if (move) 1118 len += plen; 1119 dent = kmalloc(len, GFP_NOFS); 1120 if (!dent) 1121 return -ENOMEM; 1122 1123 /* Make reservation before allocating sequence numbers */ 1124 err = make_reservation(c, BASEHD, len); 1125 if (err) 1126 goto out_free; 1127 1128 /* Make new dent */ 1129 dent->ch.node_type = UBIFS_DENT_NODE; 1130 dent_key_init_flash(c, &dent->key, new_dir->i_ino, new_nm); 1131 dent->inum = cpu_to_le64(old_inode->i_ino); 1132 dent->type = get_dent_type(old_inode->i_mode); 1133 dent->nlen = cpu_to_le16(fname_len(new_nm)); 1134 memcpy(dent->name, fname_name(new_nm), fname_len(new_nm)); 1135 dent->name[fname_len(new_nm)] = '\0'; 1136 set_dent_cookie(c, dent); 1137 zero_dent_node_unused(dent); 1138 ubifs_prep_grp_node(c, dent, dlen1, 0); 1139 1140 dent2 = (void *)dent + aligned_dlen1; 1141 dent2->ch.node_type = UBIFS_DENT_NODE; 1142 dent_key_init_flash(c, &dent2->key, old_dir->i_ino, old_nm); 1143 1144 if (whiteout) { 1145 dent2->inum = cpu_to_le64(whiteout->i_ino); 1146 dent2->type = get_dent_type(whiteout->i_mode); 1147 } else { 1148 /* Make deletion dent */ 1149 dent2->inum = 0; 1150 dent2->type = DT_UNKNOWN; 1151 } 1152 dent2->nlen = cpu_to_le16(fname_len(old_nm)); 1153 memcpy(dent2->name, fname_name(old_nm), fname_len(old_nm)); 1154 dent2->name[fname_len(old_nm)] = '\0'; 1155 set_dent_cookie(c, dent2); 1156 zero_dent_node_unused(dent2); 1157 ubifs_prep_grp_node(c, dent2, dlen2, 0); 1158 1159 p = (void *)dent2 + aligned_dlen2; 1160 if (new_inode) { 1161 pack_inode(c, p, new_inode, 0); 1162 p += ALIGN(ilen, 8); 1163 } 1164 1165 if (!move) 1166 pack_inode(c, p, old_dir, 1); 1167 else { 1168 pack_inode(c, p, old_dir, 0); 1169 p += ALIGN(plen, 8); 1170 pack_inode(c, p, new_dir, 1); 1171 } 1172 1173 if (last_reference) { 1174 err = ubifs_add_orphan(c, new_inode->i_ino); 1175 if (err) { 1176 release_head(c, BASEHD); 1177 goto out_finish; 1178 } 1179 new_ui->del_cmtno = c->cmt_no; 1180 } 1181 1182 err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync); 1183 if (err) 1184 goto out_release; 1185 if (!sync) { 1186 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf; 1187 1188 ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino); 1189 ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino); 1190 if (new_inode) 1191 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, 1192 new_inode->i_ino); 1193 } 1194 release_head(c, BASEHD); 1195 1196 dent_key_init(c, &key, new_dir->i_ino, new_nm); 1197 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, new_nm); 1198 if (err) 1199 goto out_ro; 1200 1201 offs += aligned_dlen1; 1202 if (whiteout) { 1203 dent_key_init(c, &key, old_dir->i_ino, old_nm); 1204 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, old_nm); 1205 if (err) 1206 goto out_ro; 1207 1208 ubifs_delete_orphan(c, whiteout->i_ino); 1209 } else { 1210 err = ubifs_add_dirt(c, lnum, dlen2); 1211 if (err) 1212 goto out_ro; 1213 1214 dent_key_init(c, &key, old_dir->i_ino, old_nm); 1215 err = ubifs_tnc_remove_nm(c, &key, old_nm); 1216 if (err) 1217 goto out_ro; 1218 } 1219 1220 offs += aligned_dlen2; 1221 if (new_inode) { 1222 ino_key_init(c, &key, new_inode->i_ino); 1223 err = ubifs_tnc_add(c, &key, lnum, offs, ilen); 1224 if (err) 1225 goto out_ro; 1226 offs += ALIGN(ilen, 8); 1227 } 1228 1229 ino_key_init(c, &key, old_dir->i_ino); 1230 err = ubifs_tnc_add(c, &key, lnum, offs, plen); 1231 if (err) 1232 goto out_ro; 1233 1234 if (move) { 1235 offs += ALIGN(plen, 8); 1236 ino_key_init(c, &key, new_dir->i_ino); 1237 err = ubifs_tnc_add(c, &key, lnum, offs, plen); 1238 if (err) 1239 goto out_ro; 1240 } 1241 1242 finish_reservation(c); 1243 if (new_inode) { 1244 mark_inode_clean(c, new_ui); 1245 spin_lock(&new_ui->ui_lock); 1246 new_ui->synced_i_size = new_ui->ui_size; 1247 spin_unlock(&new_ui->ui_lock); 1248 } 1249 mark_inode_clean(c, ubifs_inode(old_dir)); 1250 if (move) 1251 mark_inode_clean(c, ubifs_inode(new_dir)); 1252 kfree(dent); 1253 return 0; 1254 1255 out_release: 1256 release_head(c, BASEHD); 1257 out_ro: 1258 ubifs_ro_mode(c, err); 1259 if (last_reference) 1260 ubifs_delete_orphan(c, new_inode->i_ino); 1261 out_finish: 1262 finish_reservation(c); 1263 out_free: 1264 kfree(dent); 1265 return err; 1266 } 1267 1268 /** 1269 * truncate_data_node - re-compress/encrypt a truncated data node. 1270 * @c: UBIFS file-system description object 1271 * @inode: inode which referes to the data node 1272 * @block: data block number 1273 * @dn: data node to re-compress 1274 * @new_len: new length 1275 * 1276 * This function is used when an inode is truncated and the last data node of 1277 * the inode has to be re-compressed/encrypted and re-written. 1278 */ 1279 static int truncate_data_node(const struct ubifs_info *c, const struct inode *inode, 1280 unsigned int block, struct ubifs_data_node *dn, 1281 int *new_len) 1282 { 1283 void *buf; 1284 int err, dlen, compr_type, out_len, old_dlen; 1285 1286 out_len = le32_to_cpu(dn->size); 1287 buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS); 1288 if (!buf) 1289 return -ENOMEM; 1290 1291 dlen = old_dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ; 1292 compr_type = le16_to_cpu(dn->compr_type); 1293 1294 if (ubifs_crypt_is_encrypted(inode)) { 1295 err = ubifs_decrypt(inode, dn, &dlen, block); 1296 if (err) 1297 goto out; 1298 } 1299 1300 if (compr_type != UBIFS_COMPR_NONE) { 1301 err = ubifs_decompress(c, &dn->data, dlen, buf, &out_len, compr_type); 1302 if (err) 1303 goto out; 1304 1305 ubifs_compress(c, buf, *new_len, &dn->data, &out_len, &compr_type); 1306 } 1307 1308 if (ubifs_crypt_is_encrypted(inode)) { 1309 err = ubifs_encrypt(inode, dn, out_len, &old_dlen, block); 1310 if (err) 1311 goto out; 1312 1313 out_len = old_dlen; 1314 } else { 1315 dn->compr_size = 0; 1316 } 1317 1318 ubifs_assert(out_len <= UBIFS_BLOCK_SIZE); 1319 dn->compr_type = cpu_to_le16(compr_type); 1320 dn->size = cpu_to_le32(*new_len); 1321 *new_len = UBIFS_DATA_NODE_SZ + out_len; 1322 out: 1323 kfree(buf); 1324 return err; 1325 } 1326 1327 /** 1328 * ubifs_jnl_truncate - update the journal for a truncation. 1329 * @c: UBIFS file-system description object 1330 * @inode: inode to truncate 1331 * @old_size: old size 1332 * @new_size: new size 1333 * 1334 * When the size of a file decreases due to truncation, a truncation node is 1335 * written, the journal tree is updated, and the last data block is re-written 1336 * if it has been affected. The inode is also updated in order to synchronize 1337 * the new inode size. 1338 * 1339 * This function marks the inode as clean and returns zero on success. In case 1340 * of failure, a negative error code is returned. 1341 */ 1342 int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode, 1343 loff_t old_size, loff_t new_size) 1344 { 1345 union ubifs_key key, to_key; 1346 struct ubifs_ino_node *ino; 1347 struct ubifs_trun_node *trun; 1348 struct ubifs_data_node *uninitialized_var(dn); 1349 int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode); 1350 struct ubifs_inode *ui = ubifs_inode(inode); 1351 ino_t inum = inode->i_ino; 1352 unsigned int blk; 1353 1354 dbg_jnl("ino %lu, size %lld -> %lld", 1355 (unsigned long)inum, old_size, new_size); 1356 ubifs_assert(!ui->data_len); 1357 ubifs_assert(S_ISREG(inode->i_mode)); 1358 ubifs_assert(mutex_is_locked(&ui->ui_mutex)); 1359 1360 sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ + 1361 UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR; 1362 ino = kmalloc(sz, GFP_NOFS); 1363 if (!ino) 1364 return -ENOMEM; 1365 1366 trun = (void *)ino + UBIFS_INO_NODE_SZ; 1367 trun->ch.node_type = UBIFS_TRUN_NODE; 1368 trun->inum = cpu_to_le32(inum); 1369 trun->old_size = cpu_to_le64(old_size); 1370 trun->new_size = cpu_to_le64(new_size); 1371 zero_trun_node_unused(trun); 1372 1373 dlen = new_size & (UBIFS_BLOCK_SIZE - 1); 1374 if (dlen) { 1375 /* Get last data block so it can be truncated */ 1376 dn = (void *)trun + UBIFS_TRUN_NODE_SZ; 1377 blk = new_size >> UBIFS_BLOCK_SHIFT; 1378 data_key_init(c, &key, inum, blk); 1379 dbg_jnlk(&key, "last block key "); 1380 err = ubifs_tnc_lookup(c, &key, dn); 1381 if (err == -ENOENT) 1382 dlen = 0; /* Not found (so it is a hole) */ 1383 else if (err) 1384 goto out_free; 1385 else { 1386 if (le32_to_cpu(dn->size) <= dlen) 1387 dlen = 0; /* Nothing to do */ 1388 else { 1389 err = truncate_data_node(c, inode, blk, dn, &dlen); 1390 if (err) 1391 goto out_free; 1392 } 1393 } 1394 } 1395 1396 /* Must make reservation before allocating sequence numbers */ 1397 len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ; 1398 if (dlen) 1399 len += dlen; 1400 err = make_reservation(c, BASEHD, len); 1401 if (err) 1402 goto out_free; 1403 1404 pack_inode(c, ino, inode, 0); 1405 ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1); 1406 if (dlen) 1407 ubifs_prep_grp_node(c, dn, dlen, 1); 1408 1409 err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync); 1410 if (err) 1411 goto out_release; 1412 if (!sync) 1413 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum); 1414 release_head(c, BASEHD); 1415 1416 if (dlen) { 1417 sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ; 1418 err = ubifs_tnc_add(c, &key, lnum, sz, dlen); 1419 if (err) 1420 goto out_ro; 1421 } 1422 1423 ino_key_init(c, &key, inum); 1424 err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ); 1425 if (err) 1426 goto out_ro; 1427 1428 err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ); 1429 if (err) 1430 goto out_ro; 1431 1432 bit = new_size & (UBIFS_BLOCK_SIZE - 1); 1433 blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0); 1434 data_key_init(c, &key, inum, blk); 1435 1436 bit = old_size & (UBIFS_BLOCK_SIZE - 1); 1437 blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1); 1438 data_key_init(c, &to_key, inum, blk); 1439 1440 err = ubifs_tnc_remove_range(c, &key, &to_key); 1441 if (err) 1442 goto out_ro; 1443 1444 finish_reservation(c); 1445 spin_lock(&ui->ui_lock); 1446 ui->synced_i_size = ui->ui_size; 1447 spin_unlock(&ui->ui_lock); 1448 mark_inode_clean(c, ui); 1449 kfree(ino); 1450 return 0; 1451 1452 out_release: 1453 release_head(c, BASEHD); 1454 out_ro: 1455 ubifs_ro_mode(c, err); 1456 finish_reservation(c); 1457 out_free: 1458 kfree(ino); 1459 return err; 1460 } 1461 1462 1463 /** 1464 * ubifs_jnl_delete_xattr - delete an extended attribute. 1465 * @c: UBIFS file-system description object 1466 * @host: host inode 1467 * @inode: extended attribute inode 1468 * @nm: extended attribute entry name 1469 * 1470 * This function delete an extended attribute which is very similar to 1471 * un-linking regular files - it writes a deletion xentry, a deletion inode and 1472 * updates the target inode. Returns zero in case of success and a negative 1473 * error code in case of failure. 1474 */ 1475 int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host, 1476 const struct inode *inode, 1477 const struct fscrypt_name *nm) 1478 { 1479 int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen; 1480 struct ubifs_dent_node *xent; 1481 struct ubifs_ino_node *ino; 1482 union ubifs_key xent_key, key1, key2; 1483 int sync = IS_DIRSYNC(host); 1484 struct ubifs_inode *host_ui = ubifs_inode(host); 1485 1486 //dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d", 1487 // host->i_ino, inode->i_ino, nm->name, 1488 // ubifs_inode(inode)->data_len); 1489 ubifs_assert(inode->i_nlink == 0); 1490 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex)); 1491 1492 /* 1493 * Since we are deleting the inode, we do not bother to attach any data 1494 * to it and assume its length is %UBIFS_INO_NODE_SZ. 1495 */ 1496 xlen = UBIFS_DENT_NODE_SZ + fname_len(nm) + 1; 1497 aligned_xlen = ALIGN(xlen, 8); 1498 hlen = host_ui->data_len + UBIFS_INO_NODE_SZ; 1499 len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8); 1500 1501 xent = kmalloc(len, GFP_NOFS); 1502 if (!xent) 1503 return -ENOMEM; 1504 1505 /* Make reservation before allocating sequence numbers */ 1506 err = make_reservation(c, BASEHD, len); 1507 if (err) { 1508 kfree(xent); 1509 return err; 1510 } 1511 1512 xent->ch.node_type = UBIFS_XENT_NODE; 1513 xent_key_init(c, &xent_key, host->i_ino, nm); 1514 key_write(c, &xent_key, xent->key); 1515 xent->inum = 0; 1516 xent->type = get_dent_type(inode->i_mode); 1517 xent->nlen = cpu_to_le16(fname_len(nm)); 1518 memcpy(xent->name, fname_name(nm), fname_len(nm)); 1519 xent->name[fname_len(nm)] = '\0'; 1520 zero_dent_node_unused(xent); 1521 ubifs_prep_grp_node(c, xent, xlen, 0); 1522 1523 ino = (void *)xent + aligned_xlen; 1524 pack_inode(c, ino, inode, 0); 1525 ino = (void *)ino + UBIFS_INO_NODE_SZ; 1526 pack_inode(c, ino, host, 1); 1527 1528 err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync); 1529 if (!sync && !err) 1530 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino); 1531 release_head(c, BASEHD); 1532 kfree(xent); 1533 if (err) 1534 goto out_ro; 1535 1536 /* Remove the extended attribute entry from TNC */ 1537 err = ubifs_tnc_remove_nm(c, &xent_key, nm); 1538 if (err) 1539 goto out_ro; 1540 err = ubifs_add_dirt(c, lnum, xlen); 1541 if (err) 1542 goto out_ro; 1543 1544 /* 1545 * Remove all nodes belonging to the extended attribute inode from TNC. 1546 * Well, there actually must be only one node - the inode itself. 1547 */ 1548 lowest_ino_key(c, &key1, inode->i_ino); 1549 highest_ino_key(c, &key2, inode->i_ino); 1550 err = ubifs_tnc_remove_range(c, &key1, &key2); 1551 if (err) 1552 goto out_ro; 1553 err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ); 1554 if (err) 1555 goto out_ro; 1556 1557 /* And update TNC with the new host inode position */ 1558 ino_key_init(c, &key1, host->i_ino); 1559 err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen); 1560 if (err) 1561 goto out_ro; 1562 1563 finish_reservation(c); 1564 spin_lock(&host_ui->ui_lock); 1565 host_ui->synced_i_size = host_ui->ui_size; 1566 spin_unlock(&host_ui->ui_lock); 1567 mark_inode_clean(c, host_ui); 1568 return 0; 1569 1570 out_ro: 1571 ubifs_ro_mode(c, err); 1572 finish_reservation(c); 1573 return err; 1574 } 1575 1576 /** 1577 * ubifs_jnl_change_xattr - change an extended attribute. 1578 * @c: UBIFS file-system description object 1579 * @inode: extended attribute inode 1580 * @host: host inode 1581 * 1582 * This function writes the updated version of an extended attribute inode and 1583 * the host inode to the journal (to the base head). The host inode is written 1584 * after the extended attribute inode in order to guarantee that the extended 1585 * attribute will be flushed when the inode is synchronized by 'fsync()' and 1586 * consequently, the write-buffer is synchronized. This function returns zero 1587 * in case of success and a negative error code in case of failure. 1588 */ 1589 int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode, 1590 const struct inode *host) 1591 { 1592 int err, len1, len2, aligned_len, aligned_len1, lnum, offs; 1593 struct ubifs_inode *host_ui = ubifs_inode(host); 1594 struct ubifs_ino_node *ino; 1595 union ubifs_key key; 1596 int sync = IS_DIRSYNC(host); 1597 1598 dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino); 1599 ubifs_assert(host->i_nlink > 0); 1600 ubifs_assert(inode->i_nlink > 0); 1601 ubifs_assert(mutex_is_locked(&host_ui->ui_mutex)); 1602 1603 len1 = UBIFS_INO_NODE_SZ + host_ui->data_len; 1604 len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len; 1605 aligned_len1 = ALIGN(len1, 8); 1606 aligned_len = aligned_len1 + ALIGN(len2, 8); 1607 1608 ino = kmalloc(aligned_len, GFP_NOFS); 1609 if (!ino) 1610 return -ENOMEM; 1611 1612 /* Make reservation before allocating sequence numbers */ 1613 err = make_reservation(c, BASEHD, aligned_len); 1614 if (err) 1615 goto out_free; 1616 1617 pack_inode(c, ino, host, 0); 1618 pack_inode(c, (void *)ino + aligned_len1, inode, 1); 1619 1620 err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0); 1621 if (!sync && !err) { 1622 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf; 1623 1624 ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino); 1625 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino); 1626 } 1627 release_head(c, BASEHD); 1628 if (err) 1629 goto out_ro; 1630 1631 ino_key_init(c, &key, host->i_ino); 1632 err = ubifs_tnc_add(c, &key, lnum, offs, len1); 1633 if (err) 1634 goto out_ro; 1635 1636 ino_key_init(c, &key, inode->i_ino); 1637 err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2); 1638 if (err) 1639 goto out_ro; 1640 1641 finish_reservation(c); 1642 spin_lock(&host_ui->ui_lock); 1643 host_ui->synced_i_size = host_ui->ui_size; 1644 spin_unlock(&host_ui->ui_lock); 1645 mark_inode_clean(c, host_ui); 1646 kfree(ino); 1647 return 0; 1648 1649 out_ro: 1650 ubifs_ro_mode(c, err); 1651 finish_reservation(c); 1652 out_free: 1653 kfree(ino); 1654 return err; 1655 } 1656 1657