1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2006 Silicon Graphics, Inc. 4 * Copyright (c) 2012-2013 Red Hat, Inc. 5 * All rights reserved. 6 */ 7 #include "xfs.h" 8 #include "xfs_shared.h" 9 #include "xfs_fs.h" 10 #include "xfs_format.h" 11 #include "xfs_log_format.h" 12 #include "xfs_trans_resv.h" 13 #include "xfs_bit.h" 14 #include "xfs_mount.h" 15 #include "xfs_dir2.h" 16 #include "xfs_inode.h" 17 #include "xfs_bmap.h" 18 #include "xfs_bmap_btree.h" 19 #include "xfs_quota.h" 20 #include "xfs_trans_space.h" 21 #include "xfs_trace.h" 22 #include "xfs_trans.h" 23 24 /* ----- Kernel only functions below ----- */ 25 int 26 xfs_readlink_bmap_ilocked( 27 struct xfs_inode *ip, 28 char *link) 29 { 30 struct xfs_mount *mp = ip->i_mount; 31 struct xfs_bmbt_irec mval[XFS_SYMLINK_MAPS]; 32 struct xfs_buf *bp; 33 xfs_daddr_t d; 34 char *cur_chunk; 35 int pathlen = ip->i_d.di_size; 36 int nmaps = XFS_SYMLINK_MAPS; 37 int byte_cnt; 38 int n; 39 int error = 0; 40 int fsblocks = 0; 41 int offset; 42 43 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 44 45 fsblocks = xfs_symlink_blocks(mp, pathlen); 46 error = xfs_bmapi_read(ip, 0, fsblocks, mval, &nmaps, 0); 47 if (error) 48 goto out; 49 50 offset = 0; 51 for (n = 0; n < nmaps; n++) { 52 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); 53 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); 54 55 bp = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0, 56 &xfs_symlink_buf_ops); 57 if (!bp) 58 return -ENOMEM; 59 error = bp->b_error; 60 if (error) { 61 xfs_buf_ioerror_alert(bp, __func__); 62 xfs_buf_relse(bp); 63 64 /* bad CRC means corrupted metadata */ 65 if (error == -EFSBADCRC) 66 error = -EFSCORRUPTED; 67 goto out; 68 } 69 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt); 70 if (pathlen < byte_cnt) 71 byte_cnt = pathlen; 72 73 cur_chunk = bp->b_addr; 74 if (xfs_sb_version_hascrc(&mp->m_sb)) { 75 if (!xfs_symlink_hdr_ok(ip->i_ino, offset, 76 byte_cnt, bp)) { 77 error = -EFSCORRUPTED; 78 xfs_alert(mp, 79 "symlink header does not match required off/len/owner (0x%x/Ox%x,0x%llx)", 80 offset, byte_cnt, ip->i_ino); 81 xfs_buf_relse(bp); 82 goto out; 83 84 } 85 86 cur_chunk += sizeof(struct xfs_dsymlink_hdr); 87 } 88 89 memcpy(link + offset, cur_chunk, byte_cnt); 90 91 pathlen -= byte_cnt; 92 offset += byte_cnt; 93 94 xfs_buf_relse(bp); 95 } 96 ASSERT(pathlen == 0); 97 98 link[ip->i_d.di_size] = '\0'; 99 error = 0; 100 101 out: 102 return error; 103 } 104 105 int 106 xfs_readlink( 107 struct xfs_inode *ip, 108 char *link) 109 { 110 struct xfs_mount *mp = ip->i_mount; 111 xfs_fsize_t pathlen; 112 int error = 0; 113 114 trace_xfs_readlink(ip); 115 116 ASSERT(!(ip->i_df.if_flags & XFS_IFINLINE)); 117 118 if (XFS_FORCED_SHUTDOWN(mp)) 119 return -EIO; 120 121 xfs_ilock(ip, XFS_ILOCK_SHARED); 122 123 pathlen = ip->i_d.di_size; 124 if (!pathlen) 125 goto out; 126 127 if (pathlen < 0 || pathlen > XFS_SYMLINK_MAXLEN) { 128 xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)", 129 __func__, (unsigned long long) ip->i_ino, 130 (long long) pathlen); 131 ASSERT(0); 132 error = -EFSCORRUPTED; 133 goto out; 134 } 135 136 137 error = xfs_readlink_bmap_ilocked(ip, link); 138 139 out: 140 xfs_iunlock(ip, XFS_ILOCK_SHARED); 141 return error; 142 } 143 144 int 145 xfs_symlink( 146 struct xfs_inode *dp, 147 struct xfs_name *link_name, 148 const char *target_path, 149 umode_t mode, 150 struct xfs_inode **ipp) 151 { 152 struct xfs_mount *mp = dp->i_mount; 153 struct xfs_trans *tp = NULL; 154 struct xfs_inode *ip = NULL; 155 int error = 0; 156 int pathlen; 157 bool unlock_dp_on_error = false; 158 xfs_fileoff_t first_fsb; 159 xfs_filblks_t fs_blocks; 160 int nmaps; 161 struct xfs_bmbt_irec mval[XFS_SYMLINK_MAPS]; 162 xfs_daddr_t d; 163 const char *cur_chunk; 164 int byte_cnt; 165 int n; 166 xfs_buf_t *bp; 167 prid_t prid; 168 struct xfs_dquot *udqp = NULL; 169 struct xfs_dquot *gdqp = NULL; 170 struct xfs_dquot *pdqp = NULL; 171 uint resblks; 172 173 *ipp = NULL; 174 175 trace_xfs_symlink(dp, link_name); 176 177 if (XFS_FORCED_SHUTDOWN(mp)) 178 return -EIO; 179 180 /* 181 * Check component lengths of the target path name. 182 */ 183 pathlen = strlen(target_path); 184 if (pathlen >= XFS_SYMLINK_MAXLEN) /* total string too long */ 185 return -ENAMETOOLONG; 186 ASSERT(pathlen > 0); 187 188 udqp = gdqp = NULL; 189 prid = xfs_get_initial_prid(dp); 190 191 /* 192 * Make sure that we have allocated dquot(s) on disk. 193 */ 194 error = xfs_qm_vop_dqalloc(dp, 195 xfs_kuid_to_uid(current_fsuid()), 196 xfs_kgid_to_gid(current_fsgid()), prid, 197 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 198 &udqp, &gdqp, &pdqp); 199 if (error) 200 return error; 201 202 /* 203 * The symlink will fit into the inode data fork? 204 * There can't be any attributes so we get the whole variable part. 205 */ 206 if (pathlen <= XFS_LITINO(mp, dp->i_d.di_version)) 207 fs_blocks = 0; 208 else 209 fs_blocks = xfs_symlink_blocks(mp, pathlen); 210 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks); 211 212 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_symlink, resblks, 0, 0, &tp); 213 if (error) 214 goto out_release_inode; 215 216 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); 217 unlock_dp_on_error = true; 218 219 /* 220 * Check whether the directory allows new symlinks or not. 221 */ 222 if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) { 223 error = -EPERM; 224 goto out_trans_cancel; 225 } 226 227 /* 228 * Reserve disk quota : blocks and inode. 229 */ 230 error = xfs_trans_reserve_quota(tp, mp, udqp, gdqp, 231 pdqp, resblks, 1, 0); 232 if (error) 233 goto out_trans_cancel; 234 235 /* 236 * Allocate an inode for the symlink. 237 */ 238 error = xfs_dir_ialloc(&tp, dp, S_IFLNK | (mode & ~S_IFMT), 1, 0, 239 prid, &ip); 240 if (error) 241 goto out_trans_cancel; 242 243 /* 244 * Now we join the directory inode to the transaction. We do not do it 245 * earlier because xfs_dir_ialloc might commit the previous transaction 246 * (and release all the locks). An error from here on will result in 247 * the transaction cancel unlocking dp so don't do it explicitly in the 248 * error path. 249 */ 250 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 251 unlock_dp_on_error = false; 252 253 /* 254 * Also attach the dquot(s) to it, if applicable. 255 */ 256 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 257 258 if (resblks) 259 resblks -= XFS_IALLOC_SPACE_RES(mp); 260 /* 261 * If the symlink will fit into the inode, write it inline. 262 */ 263 if (pathlen <= XFS_IFORK_DSIZE(ip)) { 264 xfs_init_local_fork(ip, XFS_DATA_FORK, target_path, pathlen); 265 266 ip->i_d.di_size = pathlen; 267 ip->i_d.di_format = XFS_DINODE_FMT_LOCAL; 268 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); 269 } else { 270 int offset; 271 272 first_fsb = 0; 273 nmaps = XFS_SYMLINK_MAPS; 274 275 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks, 276 XFS_BMAPI_METADATA, resblks, mval, &nmaps); 277 if (error) 278 goto out_trans_cancel; 279 280 if (resblks) 281 resblks -= fs_blocks; 282 ip->i_d.di_size = pathlen; 283 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 284 285 cur_chunk = target_path; 286 offset = 0; 287 for (n = 0; n < nmaps; n++) { 288 char *buf; 289 290 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); 291 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); 292 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, 293 BTOBB(byte_cnt), 0); 294 if (!bp) { 295 error = -ENOMEM; 296 goto out_trans_cancel; 297 } 298 bp->b_ops = &xfs_symlink_buf_ops; 299 300 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt); 301 byte_cnt = min(byte_cnt, pathlen); 302 303 buf = bp->b_addr; 304 buf += xfs_symlink_hdr_set(mp, ip->i_ino, offset, 305 byte_cnt, bp); 306 307 memcpy(buf, cur_chunk, byte_cnt); 308 309 cur_chunk += byte_cnt; 310 pathlen -= byte_cnt; 311 offset += byte_cnt; 312 313 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SYMLINK_BUF); 314 xfs_trans_log_buf(tp, bp, 0, (buf + byte_cnt - 1) - 315 (char *)bp->b_addr); 316 } 317 ASSERT(pathlen == 0); 318 } 319 320 /* 321 * Create the directory entry for the symlink. 322 */ 323 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, resblks); 324 if (error) 325 goto out_trans_cancel; 326 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 327 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 328 329 /* 330 * If this is a synchronous mount, make sure that the 331 * symlink transaction goes to disk before returning to 332 * the user. 333 */ 334 if (mp->m_flags & (XFS_MOUNT_WSYNC|XFS_MOUNT_DIRSYNC)) { 335 xfs_trans_set_sync(tp); 336 } 337 338 error = xfs_trans_commit(tp); 339 if (error) 340 goto out_release_inode; 341 342 xfs_qm_dqrele(udqp); 343 xfs_qm_dqrele(gdqp); 344 xfs_qm_dqrele(pdqp); 345 346 *ipp = ip; 347 return 0; 348 349 out_trans_cancel: 350 xfs_trans_cancel(tp); 351 out_release_inode: 352 /* 353 * Wait until after the current transaction is aborted to finish the 354 * setup of the inode and release the inode. This prevents recursive 355 * transactions and deadlocks from xfs_inactive. 356 */ 357 if (ip) { 358 xfs_finish_inode_setup(ip); 359 xfs_irele(ip); 360 } 361 362 xfs_qm_dqrele(udqp); 363 xfs_qm_dqrele(gdqp); 364 xfs_qm_dqrele(pdqp); 365 366 if (unlock_dp_on_error) 367 xfs_iunlock(dp, XFS_ILOCK_EXCL); 368 return error; 369 } 370 371 /* 372 * Free a symlink that has blocks associated with it. 373 * 374 * Note: zero length symlinks are not allowed to exist. When we set the size to 375 * zero, also change it to a regular file so that it does not get written to 376 * disk as a zero length symlink. The inode is on the unlinked list already, so 377 * userspace cannot find this inode anymore, so this change is not user visible 378 * but allows us to catch corrupt zero-length symlinks in the verifiers. 379 */ 380 STATIC int 381 xfs_inactive_symlink_rmt( 382 struct xfs_inode *ip) 383 { 384 xfs_buf_t *bp; 385 int done; 386 int error; 387 int i; 388 xfs_mount_t *mp; 389 xfs_bmbt_irec_t mval[XFS_SYMLINK_MAPS]; 390 int nmaps; 391 int size; 392 xfs_trans_t *tp; 393 394 mp = ip->i_mount; 395 ASSERT(ip->i_df.if_flags & XFS_IFEXTENTS); 396 /* 397 * We're freeing a symlink that has some 398 * blocks allocated to it. Free the 399 * blocks here. We know that we've got 400 * either 1 or 2 extents and that we can 401 * free them all in one bunmapi call. 402 */ 403 ASSERT(ip->i_d.di_nextents > 0 && ip->i_d.di_nextents <= 2); 404 405 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 406 if (error) 407 return error; 408 409 xfs_ilock(ip, XFS_ILOCK_EXCL); 410 xfs_trans_ijoin(tp, ip, 0); 411 412 /* 413 * Lock the inode, fix the size, turn it into a regular file and join it 414 * to the transaction. Hold it so in the normal path, we still have it 415 * locked for the second transaction. In the error paths we need it 416 * held so the cancel won't rele it, see below. 417 */ 418 size = (int)ip->i_d.di_size; 419 ip->i_d.di_size = 0; 420 VFS_I(ip)->i_mode = (VFS_I(ip)->i_mode & ~S_IFMT) | S_IFREG; 421 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 422 /* 423 * Find the block(s) so we can inval and unmap them. 424 */ 425 done = 0; 426 nmaps = ARRAY_SIZE(mval); 427 error = xfs_bmapi_read(ip, 0, xfs_symlink_blocks(mp, size), 428 mval, &nmaps, 0); 429 if (error) 430 goto error_trans_cancel; 431 /* 432 * Invalidate the block(s). No validation is done. 433 */ 434 for (i = 0; i < nmaps; i++) { 435 bp = xfs_trans_get_buf(tp, mp->m_ddev_targp, 436 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), 437 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0); 438 if (!bp) { 439 error = -ENOMEM; 440 goto error_trans_cancel; 441 } 442 xfs_trans_binval(tp, bp); 443 } 444 /* 445 * Unmap the dead block(s) to the dfops. 446 */ 447 error = xfs_bunmapi(tp, ip, 0, size, 0, nmaps, &done); 448 if (error) 449 goto error_trans_cancel; 450 ASSERT(done); 451 452 /* 453 * Commit the transaction. This first logs the EFI and the inode, then 454 * rolls and commits the transaction that frees the extents. 455 */ 456 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 457 error = xfs_trans_commit(tp); 458 if (error) { 459 ASSERT(XFS_FORCED_SHUTDOWN(mp)); 460 goto error_unlock; 461 } 462 463 /* 464 * Remove the memory for extent descriptions (just bookkeeping). 465 */ 466 if (ip->i_df.if_bytes) 467 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); 468 ASSERT(ip->i_df.if_bytes == 0); 469 470 xfs_iunlock(ip, XFS_ILOCK_EXCL); 471 return 0; 472 473 error_trans_cancel: 474 xfs_trans_cancel(tp); 475 error_unlock: 476 xfs_iunlock(ip, XFS_ILOCK_EXCL); 477 return error; 478 } 479 480 /* 481 * xfs_inactive_symlink - free a symlink 482 */ 483 int 484 xfs_inactive_symlink( 485 struct xfs_inode *ip) 486 { 487 struct xfs_mount *mp = ip->i_mount; 488 int pathlen; 489 490 trace_xfs_inactive_symlink(ip); 491 492 if (XFS_FORCED_SHUTDOWN(mp)) 493 return -EIO; 494 495 xfs_ilock(ip, XFS_ILOCK_EXCL); 496 pathlen = (int)ip->i_d.di_size; 497 ASSERT(pathlen); 498 499 if (pathlen <= 0 || pathlen > XFS_SYMLINK_MAXLEN) { 500 xfs_alert(mp, "%s: inode (0x%llx) bad symlink length (%d)", 501 __func__, (unsigned long long)ip->i_ino, pathlen); 502 xfs_iunlock(ip, XFS_ILOCK_EXCL); 503 ASSERT(0); 504 return -EFSCORRUPTED; 505 } 506 507 /* 508 * Inline fork state gets removed by xfs_difree() so we have nothing to 509 * do here in that case. 510 */ 511 if (ip->i_df.if_flags & XFS_IFINLINE) { 512 xfs_iunlock(ip, XFS_ILOCK_EXCL); 513 return 0; 514 } 515 516 xfs_iunlock(ip, XFS_ILOCK_EXCL); 517 518 /* remove the remote symlink */ 519 return xfs_inactive_symlink_rmt(ip); 520 } 521