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