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