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_symlink.h" 21 #include "xfs_trans_space.h" 22 #include "xfs_trace.h" 23 #include "xfs_trans.h" 24 #include "xfs_ialloc.h" 25 #include "xfs_error.h" 26 27 /* ----- Kernel only functions below ----- */ 28 int 29 xfs_readlink_bmap_ilocked( 30 struct xfs_inode *ip, 31 char *link) 32 { 33 struct xfs_mount *mp = ip->i_mount; 34 struct xfs_bmbt_irec mval[XFS_SYMLINK_MAPS]; 35 struct xfs_buf *bp; 36 xfs_daddr_t d; 37 char *cur_chunk; 38 int pathlen = ip->i_disk_size; 39 int nmaps = XFS_SYMLINK_MAPS; 40 int byte_cnt; 41 int n; 42 int error = 0; 43 int fsblocks = 0; 44 int offset; 45 46 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); 47 48 fsblocks = xfs_symlink_blocks(mp, pathlen); 49 error = xfs_bmapi_read(ip, 0, fsblocks, mval, &nmaps, 0); 50 if (error) 51 goto out; 52 53 offset = 0; 54 for (n = 0; n < nmaps; n++) { 55 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); 56 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); 57 58 error = xfs_buf_read(mp->m_ddev_targp, d, BTOBB(byte_cnt), 0, 59 &bp, &xfs_symlink_buf_ops); 60 if (error) 61 return error; 62 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt); 63 if (pathlen < byte_cnt) 64 byte_cnt = pathlen; 65 66 cur_chunk = bp->b_addr; 67 if (xfs_has_crc(mp)) { 68 if (!xfs_symlink_hdr_ok(ip->i_ino, offset, 69 byte_cnt, bp)) { 70 error = -EFSCORRUPTED; 71 xfs_alert(mp, 72 "symlink header does not match required off/len/owner (0x%x/Ox%x,0x%llx)", 73 offset, byte_cnt, ip->i_ino); 74 xfs_buf_relse(bp); 75 goto out; 76 77 } 78 79 cur_chunk += sizeof(struct xfs_dsymlink_hdr); 80 } 81 82 memcpy(link + offset, cur_chunk, byte_cnt); 83 84 pathlen -= byte_cnt; 85 offset += byte_cnt; 86 87 xfs_buf_relse(bp); 88 } 89 ASSERT(pathlen == 0); 90 91 link[ip->i_disk_size] = '\0'; 92 error = 0; 93 94 out: 95 return error; 96 } 97 98 int 99 xfs_readlink( 100 struct xfs_inode *ip, 101 char *link) 102 { 103 struct xfs_mount *mp = ip->i_mount; 104 xfs_fsize_t pathlen; 105 int error = -EFSCORRUPTED; 106 107 trace_xfs_readlink(ip); 108 109 if (xfs_is_shutdown(mp)) 110 return -EIO; 111 112 xfs_ilock(ip, XFS_ILOCK_SHARED); 113 114 pathlen = ip->i_disk_size; 115 if (!pathlen) 116 goto out; 117 118 if (pathlen < 0 || pathlen > XFS_SYMLINK_MAXLEN) { 119 xfs_alert(mp, "%s: inode (%llu) bad symlink length (%lld)", 120 __func__, (unsigned long long) ip->i_ino, 121 (long long) pathlen); 122 ASSERT(0); 123 goto out; 124 } 125 126 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) { 127 /* 128 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED 129 * if if_data is junk. 130 */ 131 if (XFS_IS_CORRUPT(ip->i_mount, !ip->i_df.if_u1.if_data)) 132 goto out; 133 134 memcpy(link, ip->i_df.if_u1.if_data, pathlen + 1); 135 error = 0; 136 } else { 137 error = xfs_readlink_bmap_ilocked(ip, link); 138 } 139 140 out: 141 xfs_iunlock(ip, XFS_ILOCK_SHARED); 142 return error; 143 } 144 145 int 146 xfs_symlink( 147 struct user_namespace *mnt_userns, 148 struct xfs_inode *dp, 149 struct xfs_name *link_name, 150 const char *target_path, 151 umode_t mode, 152 struct xfs_inode **ipp) 153 { 154 struct xfs_mount *mp = dp->i_mount; 155 struct xfs_trans *tp = NULL; 156 struct xfs_inode *ip = NULL; 157 int error = 0; 158 int pathlen; 159 bool unlock_dp_on_error = false; 160 xfs_fileoff_t first_fsb; 161 xfs_filblks_t fs_blocks; 162 int nmaps; 163 struct xfs_bmbt_irec mval[XFS_SYMLINK_MAPS]; 164 xfs_daddr_t d; 165 const char *cur_chunk; 166 int byte_cnt; 167 int n; 168 struct xfs_buf *bp; 169 prid_t prid; 170 struct xfs_dquot *udqp = NULL; 171 struct xfs_dquot *gdqp = NULL; 172 struct xfs_dquot *pdqp = NULL; 173 uint resblks; 174 xfs_ino_t ino; 175 176 *ipp = NULL; 177 178 trace_xfs_symlink(dp, link_name); 179 180 if (xfs_is_shutdown(mp)) 181 return -EIO; 182 183 /* 184 * Check component lengths of the target path name. 185 */ 186 pathlen = strlen(target_path); 187 if (pathlen >= XFS_SYMLINK_MAXLEN) /* total string too long */ 188 return -ENAMETOOLONG; 189 ASSERT(pathlen > 0); 190 191 prid = xfs_get_initial_prid(dp); 192 193 /* 194 * Make sure that we have allocated dquot(s) on disk. 195 */ 196 error = xfs_qm_vop_dqalloc(dp, mapped_fsuid(mnt_userns, &init_user_ns), 197 mapped_fsgid(mnt_userns, &init_user_ns), prid, 198 XFS_QMOPT_QUOTALL | XFS_QMOPT_INHERIT, 199 &udqp, &gdqp, &pdqp); 200 if (error) 201 return error; 202 203 /* 204 * The symlink will fit into the inode data fork? 205 * There can't be any attributes so we get the whole variable part. 206 */ 207 if (pathlen <= XFS_LITINO(mp)) 208 fs_blocks = 0; 209 else 210 fs_blocks = xfs_symlink_blocks(mp, pathlen); 211 resblks = XFS_SYMLINK_SPACE_RES(mp, link_name->len, fs_blocks); 212 213 error = xfs_trans_alloc_icreate(mp, &M_RES(mp)->tr_symlink, udqp, gdqp, 214 pdqp, resblks, &tp); 215 if (error) 216 goto out_release_dquots; 217 218 xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT); 219 unlock_dp_on_error = true; 220 221 /* 222 * Check whether the directory allows new symlinks or not. 223 */ 224 if (dp->i_diflags & XFS_DIFLAG_NOSYMLINKS) { 225 error = -EPERM; 226 goto out_trans_cancel; 227 } 228 229 /* 230 * Allocate an inode for the symlink. 231 */ 232 error = xfs_dialloc(&tp, dp->i_ino, S_IFLNK, &ino); 233 if (!error) 234 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, 235 S_IFLNK | (mode & ~S_IFMT), 1, 0, prid, 236 false, &ip); 237 if (error) 238 goto out_trans_cancel; 239 240 /* 241 * Now we join the directory inode to the transaction. We do not do it 242 * earlier because xfs_dir_ialloc might commit the previous transaction 243 * (and release all the locks). An error from here on will result in 244 * the transaction cancel unlocking dp so don't do it explicitly in the 245 * error path. 246 */ 247 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 248 unlock_dp_on_error = false; 249 250 /* 251 * Also attach the dquot(s) to it, if applicable. 252 */ 253 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 254 255 resblks -= XFS_IALLOC_SPACE_RES(mp); 256 /* 257 * If the symlink will fit into the inode, write it inline. 258 */ 259 if (pathlen <= xfs_inode_data_fork_size(ip)) { 260 xfs_init_local_fork(ip, XFS_DATA_FORK, target_path, pathlen); 261 262 ip->i_disk_size = pathlen; 263 ip->i_df.if_format = XFS_DINODE_FMT_LOCAL; 264 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); 265 } else { 266 int offset; 267 268 first_fsb = 0; 269 nmaps = XFS_SYMLINK_MAPS; 270 271 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks, 272 XFS_BMAPI_METADATA, resblks, mval, &nmaps); 273 if (error) 274 goto out_trans_cancel; 275 276 resblks -= fs_blocks; 277 ip->i_disk_size = pathlen; 278 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 279 280 cur_chunk = target_path; 281 offset = 0; 282 for (n = 0; n < nmaps; n++) { 283 char *buf; 284 285 d = XFS_FSB_TO_DADDR(mp, mval[n].br_startblock); 286 byte_cnt = XFS_FSB_TO_B(mp, mval[n].br_blockcount); 287 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, 288 BTOBB(byte_cnt), 0, &bp); 289 if (error) 290 goto out_trans_cancel; 291 bp->b_ops = &xfs_symlink_buf_ops; 292 293 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt); 294 byte_cnt = min(byte_cnt, pathlen); 295 296 buf = bp->b_addr; 297 buf += xfs_symlink_hdr_set(mp, ip->i_ino, offset, 298 byte_cnt, bp); 299 300 memcpy(buf, cur_chunk, byte_cnt); 301 302 cur_chunk += byte_cnt; 303 pathlen -= byte_cnt; 304 offset += byte_cnt; 305 306 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SYMLINK_BUF); 307 xfs_trans_log_buf(tp, bp, 0, (buf + byte_cnt - 1) - 308 (char *)bp->b_addr); 309 } 310 ASSERT(pathlen == 0); 311 } 312 i_size_write(VFS_I(ip), ip->i_disk_size); 313 314 /* 315 * Create the directory entry for the symlink. 316 */ 317 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, resblks); 318 if (error) 319 goto out_trans_cancel; 320 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 321 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 322 323 /* 324 * If this is a synchronous mount, make sure that the 325 * symlink transaction goes to disk before returning to 326 * the user. 327 */ 328 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp)) 329 xfs_trans_set_sync(tp); 330 331 error = xfs_trans_commit(tp); 332 if (error) 333 goto out_release_inode; 334 335 xfs_qm_dqrele(udqp); 336 xfs_qm_dqrele(gdqp); 337 xfs_qm_dqrele(pdqp); 338 339 *ipp = ip; 340 return 0; 341 342 out_trans_cancel: 343 xfs_trans_cancel(tp); 344 out_release_inode: 345 /* 346 * Wait until after the current transaction is aborted to finish the 347 * setup of the inode and release the inode. This prevents recursive 348 * transactions and deadlocks from xfs_inactive. 349 */ 350 if (ip) { 351 xfs_finish_inode_setup(ip); 352 xfs_irele(ip); 353 } 354 out_release_dquots: 355 xfs_qm_dqrele(udqp); 356 xfs_qm_dqrele(gdqp); 357 xfs_qm_dqrele(pdqp); 358 359 if (unlock_dp_on_error) 360 xfs_iunlock(dp, XFS_ILOCK_EXCL); 361 return error; 362 } 363 364 /* 365 * Free a symlink that has blocks associated with it. 366 * 367 * Note: zero length symlinks are not allowed to exist. When we set the size to 368 * zero, also change it to a regular file so that it does not get written to 369 * disk as a zero length symlink. The inode is on the unlinked list already, so 370 * userspace cannot find this inode anymore, so this change is not user visible 371 * but allows us to catch corrupt zero-length symlinks in the verifiers. 372 */ 373 STATIC int 374 xfs_inactive_symlink_rmt( 375 struct xfs_inode *ip) 376 { 377 struct xfs_buf *bp; 378 int done; 379 int error; 380 int i; 381 xfs_mount_t *mp; 382 xfs_bmbt_irec_t mval[XFS_SYMLINK_MAPS]; 383 int nmaps; 384 int size; 385 xfs_trans_t *tp; 386 387 mp = ip->i_mount; 388 ASSERT(!xfs_need_iread_extents(&ip->i_df)); 389 /* 390 * We're freeing a symlink that has some 391 * blocks allocated to it. Free the 392 * blocks here. We know that we've got 393 * either 1 or 2 extents and that we can 394 * free them all in one bunmapi call. 395 */ 396 ASSERT(ip->i_df.if_nextents > 0 && ip->i_df.if_nextents <= 2); 397 398 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 399 if (error) 400 return error; 401 402 xfs_ilock(ip, XFS_ILOCK_EXCL); 403 xfs_trans_ijoin(tp, ip, 0); 404 405 /* 406 * Lock the inode, fix the size, turn it into a regular file and join it 407 * to the transaction. Hold it so in the normal path, we still have it 408 * locked for the second transaction. In the error paths we need it 409 * held so the cancel won't rele it, see below. 410 */ 411 size = (int)ip->i_disk_size; 412 ip->i_disk_size = 0; 413 VFS_I(ip)->i_mode = (VFS_I(ip)->i_mode & ~S_IFMT) | S_IFREG; 414 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 415 /* 416 * Find the block(s) so we can inval and unmap them. 417 */ 418 done = 0; 419 nmaps = ARRAY_SIZE(mval); 420 error = xfs_bmapi_read(ip, 0, xfs_symlink_blocks(mp, size), 421 mval, &nmaps, 0); 422 if (error) 423 goto error_trans_cancel; 424 /* 425 * Invalidate the block(s). No validation is done. 426 */ 427 for (i = 0; i < nmaps; i++) { 428 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, 429 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), 430 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0, 431 &bp); 432 if (error) 433 goto error_trans_cancel; 434 xfs_trans_binval(tp, bp); 435 } 436 /* 437 * Unmap the dead block(s) to the dfops. 438 */ 439 error = xfs_bunmapi(tp, ip, 0, size, 0, nmaps, &done); 440 if (error) 441 goto error_trans_cancel; 442 ASSERT(done); 443 444 /* 445 * Commit the transaction. This first logs the EFI and the inode, then 446 * rolls and commits the transaction that frees the extents. 447 */ 448 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 449 error = xfs_trans_commit(tp); 450 if (error) { 451 ASSERT(xfs_is_shutdown(mp)); 452 goto error_unlock; 453 } 454 455 /* 456 * Remove the memory for extent descriptions (just bookkeeping). 457 */ 458 if (ip->i_df.if_bytes) 459 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); 460 ASSERT(ip->i_df.if_bytes == 0); 461 462 xfs_iunlock(ip, XFS_ILOCK_EXCL); 463 return 0; 464 465 error_trans_cancel: 466 xfs_trans_cancel(tp); 467 error_unlock: 468 xfs_iunlock(ip, XFS_ILOCK_EXCL); 469 return error; 470 } 471 472 /* 473 * xfs_inactive_symlink - free a symlink 474 */ 475 int 476 xfs_inactive_symlink( 477 struct xfs_inode *ip) 478 { 479 struct xfs_mount *mp = ip->i_mount; 480 int pathlen; 481 482 trace_xfs_inactive_symlink(ip); 483 484 if (xfs_is_shutdown(mp)) 485 return -EIO; 486 487 xfs_ilock(ip, XFS_ILOCK_EXCL); 488 pathlen = (int)ip->i_disk_size; 489 ASSERT(pathlen); 490 491 if (pathlen <= 0 || pathlen > XFS_SYMLINK_MAXLEN) { 492 xfs_alert(mp, "%s: inode (0x%llx) bad symlink length (%d)", 493 __func__, (unsigned long long)ip->i_ino, pathlen); 494 xfs_iunlock(ip, XFS_ILOCK_EXCL); 495 ASSERT(0); 496 return -EFSCORRUPTED; 497 } 498 499 /* 500 * Inline fork state gets removed by xfs_difree() so we have nothing to 501 * do here in that case. 502 */ 503 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) { 504 xfs_iunlock(ip, XFS_ILOCK_EXCL); 505 return 0; 506 } 507 508 xfs_iunlock(ip, XFS_ILOCK_EXCL); 509 510 /* remove the remote symlink */ 511 return xfs_inactive_symlink_rmt(ip); 512 } 513