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 error = xfs_iext_count_may_overflow(dp, XFS_DATA_FORK, 230 XFS_IEXT_DIR_MANIP_CNT(mp)); 231 if (error) 232 goto out_trans_cancel; 233 234 /* 235 * Allocate an inode for the symlink. 236 */ 237 error = xfs_dialloc(&tp, dp->i_ino, S_IFLNK, &ino); 238 if (!error) 239 error = xfs_init_new_inode(mnt_userns, tp, dp, ino, 240 S_IFLNK | (mode & ~S_IFMT), 1, 0, prid, 241 false, &ip); 242 if (error) 243 goto out_trans_cancel; 244 245 /* 246 * Now we join the directory inode to the transaction. We do not do it 247 * earlier because xfs_dir_ialloc might commit the previous transaction 248 * (and release all the locks). An error from here on will result in 249 * the transaction cancel unlocking dp so don't do it explicitly in the 250 * error path. 251 */ 252 xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL); 253 unlock_dp_on_error = false; 254 255 /* 256 * Also attach the dquot(s) to it, if applicable. 257 */ 258 xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp); 259 260 resblks -= XFS_IALLOC_SPACE_RES(mp); 261 /* 262 * If the symlink will fit into the inode, write it inline. 263 */ 264 if (pathlen <= XFS_IFORK_DSIZE(ip)) { 265 xfs_init_local_fork(ip, XFS_DATA_FORK, target_path, pathlen); 266 267 ip->i_disk_size = pathlen; 268 ip->i_df.if_format = XFS_DINODE_FMT_LOCAL; 269 xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE); 270 } else { 271 int offset; 272 273 first_fsb = 0; 274 nmaps = XFS_SYMLINK_MAPS; 275 276 error = xfs_bmapi_write(tp, ip, first_fsb, fs_blocks, 277 XFS_BMAPI_METADATA, resblks, mval, &nmaps); 278 if (error) 279 goto out_trans_cancel; 280 281 resblks -= fs_blocks; 282 ip->i_disk_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 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d, 293 BTOBB(byte_cnt), 0, &bp); 294 if (error) 295 goto out_trans_cancel; 296 bp->b_ops = &xfs_symlink_buf_ops; 297 298 byte_cnt = XFS_SYMLINK_BUF_SPACE(mp, byte_cnt); 299 byte_cnt = min(byte_cnt, pathlen); 300 301 buf = bp->b_addr; 302 buf += xfs_symlink_hdr_set(mp, ip->i_ino, offset, 303 byte_cnt, bp); 304 305 memcpy(buf, cur_chunk, byte_cnt); 306 307 cur_chunk += byte_cnt; 308 pathlen -= byte_cnt; 309 offset += byte_cnt; 310 311 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_SYMLINK_BUF); 312 xfs_trans_log_buf(tp, bp, 0, (buf + byte_cnt - 1) - 313 (char *)bp->b_addr); 314 } 315 ASSERT(pathlen == 0); 316 } 317 i_size_write(VFS_I(ip), ip->i_disk_size); 318 319 /* 320 * Create the directory entry for the symlink. 321 */ 322 error = xfs_dir_createname(tp, dp, link_name, ip->i_ino, resblks); 323 if (error) 324 goto out_trans_cancel; 325 xfs_trans_ichgtime(tp, dp, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG); 326 xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE); 327 328 /* 329 * If this is a synchronous mount, make sure that the 330 * symlink transaction goes to disk before returning to 331 * the user. 332 */ 333 if (xfs_has_wsync(mp) || xfs_has_dirsync(mp)) 334 xfs_trans_set_sync(tp); 335 336 error = xfs_trans_commit(tp); 337 if (error) 338 goto out_release_inode; 339 340 xfs_qm_dqrele(udqp); 341 xfs_qm_dqrele(gdqp); 342 xfs_qm_dqrele(pdqp); 343 344 *ipp = ip; 345 return 0; 346 347 out_trans_cancel: 348 xfs_trans_cancel(tp); 349 out_release_inode: 350 /* 351 * Wait until after the current transaction is aborted to finish the 352 * setup of the inode and release the inode. This prevents recursive 353 * transactions and deadlocks from xfs_inactive. 354 */ 355 if (ip) { 356 xfs_finish_inode_setup(ip); 357 xfs_irele(ip); 358 } 359 out_release_dquots: 360 xfs_qm_dqrele(udqp); 361 xfs_qm_dqrele(gdqp); 362 xfs_qm_dqrele(pdqp); 363 364 if (unlock_dp_on_error) 365 xfs_iunlock(dp, XFS_ILOCK_EXCL); 366 return error; 367 } 368 369 /* 370 * Free a symlink that has blocks associated with it. 371 * 372 * Note: zero length symlinks are not allowed to exist. When we set the size to 373 * zero, also change it to a regular file so that it does not get written to 374 * disk as a zero length symlink. The inode is on the unlinked list already, so 375 * userspace cannot find this inode anymore, so this change is not user visible 376 * but allows us to catch corrupt zero-length symlinks in the verifiers. 377 */ 378 STATIC int 379 xfs_inactive_symlink_rmt( 380 struct xfs_inode *ip) 381 { 382 struct xfs_buf *bp; 383 int done; 384 int error; 385 int i; 386 xfs_mount_t *mp; 387 xfs_bmbt_irec_t mval[XFS_SYMLINK_MAPS]; 388 int nmaps; 389 int size; 390 xfs_trans_t *tp; 391 392 mp = ip->i_mount; 393 ASSERT(!xfs_need_iread_extents(&ip->i_df)); 394 /* 395 * We're freeing a symlink that has some 396 * blocks allocated to it. Free the 397 * blocks here. We know that we've got 398 * either 1 or 2 extents and that we can 399 * free them all in one bunmapi call. 400 */ 401 ASSERT(ip->i_df.if_nextents > 0 && ip->i_df.if_nextents <= 2); 402 403 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 404 if (error) 405 return error; 406 407 xfs_ilock(ip, XFS_ILOCK_EXCL); 408 xfs_trans_ijoin(tp, ip, 0); 409 410 /* 411 * Lock the inode, fix the size, turn it into a regular file and join it 412 * to the transaction. Hold it so in the normal path, we still have it 413 * locked for the second transaction. In the error paths we need it 414 * held so the cancel won't rele it, see below. 415 */ 416 size = (int)ip->i_disk_size; 417 ip->i_disk_size = 0; 418 VFS_I(ip)->i_mode = (VFS_I(ip)->i_mode & ~S_IFMT) | S_IFREG; 419 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 420 /* 421 * Find the block(s) so we can inval and unmap them. 422 */ 423 done = 0; 424 nmaps = ARRAY_SIZE(mval); 425 error = xfs_bmapi_read(ip, 0, xfs_symlink_blocks(mp, size), 426 mval, &nmaps, 0); 427 if (error) 428 goto error_trans_cancel; 429 /* 430 * Invalidate the block(s). No validation is done. 431 */ 432 for (i = 0; i < nmaps; i++) { 433 error = xfs_trans_get_buf(tp, mp->m_ddev_targp, 434 XFS_FSB_TO_DADDR(mp, mval[i].br_startblock), 435 XFS_FSB_TO_BB(mp, mval[i].br_blockcount), 0, 436 &bp); 437 if (error) 438 goto error_trans_cancel; 439 xfs_trans_binval(tp, bp); 440 } 441 /* 442 * Unmap the dead block(s) to the dfops. 443 */ 444 error = xfs_bunmapi(tp, ip, 0, size, 0, nmaps, &done); 445 if (error) 446 goto error_trans_cancel; 447 ASSERT(done); 448 449 /* 450 * Commit the transaction. This first logs the EFI and the inode, then 451 * rolls and commits the transaction that frees the extents. 452 */ 453 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 454 error = xfs_trans_commit(tp); 455 if (error) { 456 ASSERT(xfs_is_shutdown(mp)); 457 goto error_unlock; 458 } 459 460 /* 461 * Remove the memory for extent descriptions (just bookkeeping). 462 */ 463 if (ip->i_df.if_bytes) 464 xfs_idata_realloc(ip, -ip->i_df.if_bytes, XFS_DATA_FORK); 465 ASSERT(ip->i_df.if_bytes == 0); 466 467 xfs_iunlock(ip, XFS_ILOCK_EXCL); 468 return 0; 469 470 error_trans_cancel: 471 xfs_trans_cancel(tp); 472 error_unlock: 473 xfs_iunlock(ip, XFS_ILOCK_EXCL); 474 return error; 475 } 476 477 /* 478 * xfs_inactive_symlink - free a symlink 479 */ 480 int 481 xfs_inactive_symlink( 482 struct xfs_inode *ip) 483 { 484 struct xfs_mount *mp = ip->i_mount; 485 int pathlen; 486 487 trace_xfs_inactive_symlink(ip); 488 489 if (xfs_is_shutdown(mp)) 490 return -EIO; 491 492 xfs_ilock(ip, XFS_ILOCK_EXCL); 493 pathlen = (int)ip->i_disk_size; 494 ASSERT(pathlen); 495 496 if (pathlen <= 0 || pathlen > XFS_SYMLINK_MAXLEN) { 497 xfs_alert(mp, "%s: inode (0x%llx) bad symlink length (%d)", 498 __func__, (unsigned long long)ip->i_ino, pathlen); 499 xfs_iunlock(ip, XFS_ILOCK_EXCL); 500 ASSERT(0); 501 return -EFSCORRUPTED; 502 } 503 504 /* 505 * Inline fork state gets removed by xfs_difree() so we have nothing to 506 * do here in that case. 507 */ 508 if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) { 509 xfs_iunlock(ip, XFS_ILOCK_EXCL); 510 return 0; 511 } 512 513 xfs_iunlock(ip, XFS_ILOCK_EXCL); 514 515 /* remove the remote symlink */ 516 return xfs_inactive_symlink_rmt(ip); 517 } 518