1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 4 * All Rights Reserved. 5 */ 6 #include "xfs.h" 7 #include "xfs_fs.h" 8 #include "xfs_shared.h" 9 #include "xfs_format.h" 10 #include "xfs_log_format.h" 11 #include "xfs_trans_resv.h" 12 #include "xfs_mount.h" 13 #include "xfs_inode.h" 14 #include "xfs_acl.h" 15 #include "xfs_quota.h" 16 #include "xfs_attr.h" 17 #include "xfs_trans.h" 18 #include "xfs_trace.h" 19 #include "xfs_icache.h" 20 #include "xfs_symlink.h" 21 #include "xfs_dir2.h" 22 #include "xfs_iomap.h" 23 24 #include <linux/xattr.h> 25 #include <linux/posix_acl.h> 26 #include <linux/security.h> 27 #include <linux/iversion.h> 28 29 /* 30 * Directories have different lock order w.r.t. mmap_sem compared to regular 31 * files. This is due to readdir potentially triggering page faults on a user 32 * buffer inside filldir(), and this happens with the ilock on the directory 33 * held. For regular files, the lock order is the other way around - the 34 * mmap_sem is taken during the page fault, and then we lock the ilock to do 35 * block mapping. Hence we need a different class for the directory ilock so 36 * that lockdep can tell them apart. 37 */ 38 static struct lock_class_key xfs_nondir_ilock_class; 39 static struct lock_class_key xfs_dir_ilock_class; 40 41 static int 42 xfs_initxattrs( 43 struct inode *inode, 44 const struct xattr *xattr_array, 45 void *fs_info) 46 { 47 const struct xattr *xattr; 48 struct xfs_inode *ip = XFS_I(inode); 49 int error = 0; 50 51 for (xattr = xattr_array; xattr->name != NULL; xattr++) { 52 error = xfs_attr_set(ip, xattr->name, xattr->value, 53 xattr->value_len, ATTR_SECURE); 54 if (error < 0) 55 break; 56 } 57 return error; 58 } 59 60 /* 61 * Hook in SELinux. This is not quite correct yet, what we really need 62 * here (as we do for default ACLs) is a mechanism by which creation of 63 * these attrs can be journalled at inode creation time (along with the 64 * inode, of course, such that log replay can't cause these to be lost). 65 */ 66 67 STATIC int 68 xfs_init_security( 69 struct inode *inode, 70 struct inode *dir, 71 const struct qstr *qstr) 72 { 73 return security_inode_init_security(inode, dir, qstr, 74 &xfs_initxattrs, NULL); 75 } 76 77 static void 78 xfs_dentry_to_name( 79 struct xfs_name *namep, 80 struct dentry *dentry) 81 { 82 namep->name = dentry->d_name.name; 83 namep->len = dentry->d_name.len; 84 namep->type = XFS_DIR3_FT_UNKNOWN; 85 } 86 87 static int 88 xfs_dentry_mode_to_name( 89 struct xfs_name *namep, 90 struct dentry *dentry, 91 int mode) 92 { 93 namep->name = dentry->d_name.name; 94 namep->len = dentry->d_name.len; 95 namep->type = xfs_mode_to_ftype(mode); 96 97 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) 98 return -EFSCORRUPTED; 99 100 return 0; 101 } 102 103 STATIC void 104 xfs_cleanup_inode( 105 struct inode *dir, 106 struct inode *inode, 107 struct dentry *dentry) 108 { 109 struct xfs_name teardown; 110 111 /* Oh, the horror. 112 * If we can't add the ACL or we fail in 113 * xfs_init_security we must back out. 114 * ENOSPC can hit here, among other things. 115 */ 116 xfs_dentry_to_name(&teardown, dentry); 117 118 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); 119 } 120 121 STATIC int 122 xfs_generic_create( 123 struct inode *dir, 124 struct dentry *dentry, 125 umode_t mode, 126 dev_t rdev, 127 bool tmpfile) /* unnamed file */ 128 { 129 struct inode *inode; 130 struct xfs_inode *ip = NULL; 131 struct posix_acl *default_acl, *acl; 132 struct xfs_name name; 133 int error; 134 135 /* 136 * Irix uses Missed'em'V split, but doesn't want to see 137 * the upper 5 bits of (14bit) major. 138 */ 139 if (S_ISCHR(mode) || S_ISBLK(mode)) { 140 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) 141 return -EINVAL; 142 } else { 143 rdev = 0; 144 } 145 146 error = posix_acl_create(dir, &mode, &default_acl, &acl); 147 if (error) 148 return error; 149 150 /* Verify mode is valid also for tmpfile case */ 151 error = xfs_dentry_mode_to_name(&name, dentry, mode); 152 if (unlikely(error)) 153 goto out_free_acl; 154 155 if (!tmpfile) { 156 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip); 157 } else { 158 error = xfs_create_tmpfile(XFS_I(dir), mode, &ip); 159 } 160 if (unlikely(error)) 161 goto out_free_acl; 162 163 inode = VFS_I(ip); 164 165 error = xfs_init_security(inode, dir, &dentry->d_name); 166 if (unlikely(error)) 167 goto out_cleanup_inode; 168 169 #ifdef CONFIG_XFS_POSIX_ACL 170 if (default_acl) { 171 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); 172 if (error) 173 goto out_cleanup_inode; 174 } 175 if (acl) { 176 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); 177 if (error) 178 goto out_cleanup_inode; 179 } 180 #endif 181 182 xfs_setup_iops(ip); 183 184 if (tmpfile) { 185 /* 186 * The VFS requires that any inode fed to d_tmpfile must have 187 * nlink == 1 so that it can decrement the nlink in d_tmpfile. 188 * However, we created the temp file with nlink == 0 because 189 * we're not allowed to put an inode with nlink > 0 on the 190 * unlinked list. Therefore we have to set nlink to 1 so that 191 * d_tmpfile can immediately set it back to zero. 192 */ 193 set_nlink(inode, 1); 194 d_tmpfile(dentry, inode); 195 } else 196 d_instantiate(dentry, inode); 197 198 xfs_finish_inode_setup(ip); 199 200 out_free_acl: 201 if (default_acl) 202 posix_acl_release(default_acl); 203 if (acl) 204 posix_acl_release(acl); 205 return error; 206 207 out_cleanup_inode: 208 xfs_finish_inode_setup(ip); 209 if (!tmpfile) 210 xfs_cleanup_inode(dir, inode, dentry); 211 xfs_irele(ip); 212 goto out_free_acl; 213 } 214 215 STATIC int 216 xfs_vn_mknod( 217 struct inode *dir, 218 struct dentry *dentry, 219 umode_t mode, 220 dev_t rdev) 221 { 222 return xfs_generic_create(dir, dentry, mode, rdev, false); 223 } 224 225 STATIC int 226 xfs_vn_create( 227 struct inode *dir, 228 struct dentry *dentry, 229 umode_t mode, 230 bool flags) 231 { 232 return xfs_vn_mknod(dir, dentry, mode, 0); 233 } 234 235 STATIC int 236 xfs_vn_mkdir( 237 struct inode *dir, 238 struct dentry *dentry, 239 umode_t mode) 240 { 241 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0); 242 } 243 244 STATIC struct dentry * 245 xfs_vn_lookup( 246 struct inode *dir, 247 struct dentry *dentry, 248 unsigned int flags) 249 { 250 struct inode *inode; 251 struct xfs_inode *cip; 252 struct xfs_name name; 253 int error; 254 255 if (dentry->d_name.len >= MAXNAMELEN) 256 return ERR_PTR(-ENAMETOOLONG); 257 258 xfs_dentry_to_name(&name, dentry); 259 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); 260 if (likely(!error)) 261 inode = VFS_I(cip); 262 else if (likely(error == -ENOENT)) 263 inode = NULL; 264 else 265 inode = ERR_PTR(error); 266 return d_splice_alias(inode, dentry); 267 } 268 269 STATIC struct dentry * 270 xfs_vn_ci_lookup( 271 struct inode *dir, 272 struct dentry *dentry, 273 unsigned int flags) 274 { 275 struct xfs_inode *ip; 276 struct xfs_name xname; 277 struct xfs_name ci_name; 278 struct qstr dname; 279 int error; 280 281 if (dentry->d_name.len >= MAXNAMELEN) 282 return ERR_PTR(-ENAMETOOLONG); 283 284 xfs_dentry_to_name(&xname, dentry); 285 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); 286 if (unlikely(error)) { 287 if (unlikely(error != -ENOENT)) 288 return ERR_PTR(error); 289 /* 290 * call d_add(dentry, NULL) here when d_drop_negative_children 291 * is called in xfs_vn_mknod (ie. allow negative dentries 292 * with CI filesystems). 293 */ 294 return NULL; 295 } 296 297 /* if exact match, just splice and exit */ 298 if (!ci_name.name) 299 return d_splice_alias(VFS_I(ip), dentry); 300 301 /* else case-insensitive match... */ 302 dname.name = ci_name.name; 303 dname.len = ci_name.len; 304 dentry = d_add_ci(dentry, VFS_I(ip), &dname); 305 kmem_free(ci_name.name); 306 return dentry; 307 } 308 309 STATIC int 310 xfs_vn_link( 311 struct dentry *old_dentry, 312 struct inode *dir, 313 struct dentry *dentry) 314 { 315 struct inode *inode = d_inode(old_dentry); 316 struct xfs_name name; 317 int error; 318 319 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); 320 if (unlikely(error)) 321 return error; 322 323 error = xfs_link(XFS_I(dir), XFS_I(inode), &name); 324 if (unlikely(error)) 325 return error; 326 327 ihold(inode); 328 d_instantiate(dentry, inode); 329 return 0; 330 } 331 332 STATIC int 333 xfs_vn_unlink( 334 struct inode *dir, 335 struct dentry *dentry) 336 { 337 struct xfs_name name; 338 int error; 339 340 xfs_dentry_to_name(&name, dentry); 341 342 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); 343 if (error) 344 return error; 345 346 /* 347 * With unlink, the VFS makes the dentry "negative": no inode, 348 * but still hashed. This is incompatible with case-insensitive 349 * mode, so invalidate (unhash) the dentry in CI-mode. 350 */ 351 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb)) 352 d_invalidate(dentry); 353 return 0; 354 } 355 356 STATIC int 357 xfs_vn_symlink( 358 struct inode *dir, 359 struct dentry *dentry, 360 const char *symname) 361 { 362 struct inode *inode; 363 struct xfs_inode *cip = NULL; 364 struct xfs_name name; 365 int error; 366 umode_t mode; 367 368 mode = S_IFLNK | 369 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); 370 error = xfs_dentry_mode_to_name(&name, dentry, mode); 371 if (unlikely(error)) 372 goto out; 373 374 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip); 375 if (unlikely(error)) 376 goto out; 377 378 inode = VFS_I(cip); 379 380 error = xfs_init_security(inode, dir, &dentry->d_name); 381 if (unlikely(error)) 382 goto out_cleanup_inode; 383 384 xfs_setup_iops(cip); 385 386 d_instantiate(dentry, inode); 387 xfs_finish_inode_setup(cip); 388 return 0; 389 390 out_cleanup_inode: 391 xfs_finish_inode_setup(cip); 392 xfs_cleanup_inode(dir, inode, dentry); 393 xfs_irele(cip); 394 out: 395 return error; 396 } 397 398 STATIC int 399 xfs_vn_rename( 400 struct inode *odir, 401 struct dentry *odentry, 402 struct inode *ndir, 403 struct dentry *ndentry, 404 unsigned int flags) 405 { 406 struct inode *new_inode = d_inode(ndentry); 407 int omode = 0; 408 int error; 409 struct xfs_name oname; 410 struct xfs_name nname; 411 412 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 413 return -EINVAL; 414 415 /* if we are exchanging files, we need to set i_mode of both files */ 416 if (flags & RENAME_EXCHANGE) 417 omode = d_inode(ndentry)->i_mode; 418 419 error = xfs_dentry_mode_to_name(&oname, odentry, omode); 420 if (omode && unlikely(error)) 421 return error; 422 423 error = xfs_dentry_mode_to_name(&nname, ndentry, 424 d_inode(odentry)->i_mode); 425 if (unlikely(error)) 426 return error; 427 428 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)), 429 XFS_I(ndir), &nname, 430 new_inode ? XFS_I(new_inode) : NULL, flags); 431 } 432 433 /* 434 * careful here - this function can get called recursively, so 435 * we need to be very careful about how much stack we use. 436 * uio is kmalloced for this reason... 437 */ 438 STATIC const char * 439 xfs_vn_get_link( 440 struct dentry *dentry, 441 struct inode *inode, 442 struct delayed_call *done) 443 { 444 char *link; 445 int error = -ENOMEM; 446 447 if (!dentry) 448 return ERR_PTR(-ECHILD); 449 450 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); 451 if (!link) 452 goto out_err; 453 454 error = xfs_readlink(XFS_I(d_inode(dentry)), link); 455 if (unlikely(error)) 456 goto out_kfree; 457 458 set_delayed_call(done, kfree_link, link); 459 return link; 460 461 out_kfree: 462 kfree(link); 463 out_err: 464 return ERR_PTR(error); 465 } 466 467 STATIC const char * 468 xfs_vn_get_link_inline( 469 struct dentry *dentry, 470 struct inode *inode, 471 struct delayed_call *done) 472 { 473 char *link; 474 475 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE); 476 477 /* 478 * The VFS crashes on a NULL pointer, so return -EFSCORRUPTED if 479 * if_data is junk. 480 */ 481 link = XFS_I(inode)->i_df.if_u1.if_data; 482 if (!link) 483 return ERR_PTR(-EFSCORRUPTED); 484 return link; 485 } 486 487 STATIC int 488 xfs_vn_getattr( 489 const struct path *path, 490 struct kstat *stat, 491 u32 request_mask, 492 unsigned int query_flags) 493 { 494 struct inode *inode = d_inode(path->dentry); 495 struct xfs_inode *ip = XFS_I(inode); 496 struct xfs_mount *mp = ip->i_mount; 497 498 trace_xfs_getattr(ip); 499 500 if (XFS_FORCED_SHUTDOWN(mp)) 501 return -EIO; 502 503 stat->size = XFS_ISIZE(ip); 504 stat->dev = inode->i_sb->s_dev; 505 stat->mode = inode->i_mode; 506 stat->nlink = inode->i_nlink; 507 stat->uid = inode->i_uid; 508 stat->gid = inode->i_gid; 509 stat->ino = ip->i_ino; 510 stat->atime = inode->i_atime; 511 stat->mtime = inode->i_mtime; 512 stat->ctime = inode->i_ctime; 513 stat->blocks = 514 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); 515 516 if (ip->i_d.di_version == 3) { 517 if (request_mask & STATX_BTIME) { 518 stat->result_mask |= STATX_BTIME; 519 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec; 520 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec; 521 } 522 } 523 524 /* 525 * Note: If you add another clause to set an attribute flag, please 526 * update attributes_mask below. 527 */ 528 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE) 529 stat->attributes |= STATX_ATTR_IMMUTABLE; 530 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND) 531 stat->attributes |= STATX_ATTR_APPEND; 532 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP) 533 stat->attributes |= STATX_ATTR_NODUMP; 534 535 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | 536 STATX_ATTR_APPEND | 537 STATX_ATTR_NODUMP); 538 539 switch (inode->i_mode & S_IFMT) { 540 case S_IFBLK: 541 case S_IFCHR: 542 stat->blksize = BLKDEV_IOSIZE; 543 stat->rdev = inode->i_rdev; 544 break; 545 default: 546 if (XFS_IS_REALTIME_INODE(ip)) { 547 /* 548 * If the file blocks are being allocated from a 549 * realtime volume, then return the inode's realtime 550 * extent size or the realtime volume's extent size. 551 */ 552 stat->blksize = 553 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog; 554 } else 555 stat->blksize = xfs_preferred_iosize(mp); 556 stat->rdev = 0; 557 break; 558 } 559 560 return 0; 561 } 562 563 static void 564 xfs_setattr_mode( 565 struct xfs_inode *ip, 566 struct iattr *iattr) 567 { 568 struct inode *inode = VFS_I(ip); 569 umode_t mode = iattr->ia_mode; 570 571 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 572 573 inode->i_mode &= S_IFMT; 574 inode->i_mode |= mode & ~S_IFMT; 575 } 576 577 void 578 xfs_setattr_time( 579 struct xfs_inode *ip, 580 struct iattr *iattr) 581 { 582 struct inode *inode = VFS_I(ip); 583 584 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 585 586 if (iattr->ia_valid & ATTR_ATIME) 587 inode->i_atime = iattr->ia_atime; 588 if (iattr->ia_valid & ATTR_CTIME) 589 inode->i_ctime = iattr->ia_ctime; 590 if (iattr->ia_valid & ATTR_MTIME) 591 inode->i_mtime = iattr->ia_mtime; 592 } 593 594 static int 595 xfs_vn_change_ok( 596 struct dentry *dentry, 597 struct iattr *iattr) 598 { 599 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 600 601 if (mp->m_flags & XFS_MOUNT_RDONLY) 602 return -EROFS; 603 604 if (XFS_FORCED_SHUTDOWN(mp)) 605 return -EIO; 606 607 return setattr_prepare(dentry, iattr); 608 } 609 610 /* 611 * Set non-size attributes of an inode. 612 * 613 * Caution: The caller of this function is responsible for calling 614 * setattr_prepare() or otherwise verifying the change is fine. 615 */ 616 int 617 xfs_setattr_nonsize( 618 struct xfs_inode *ip, 619 struct iattr *iattr, 620 int flags) 621 { 622 xfs_mount_t *mp = ip->i_mount; 623 struct inode *inode = VFS_I(ip); 624 int mask = iattr->ia_valid; 625 xfs_trans_t *tp; 626 int error; 627 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID; 628 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID; 629 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 630 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL; 631 632 ASSERT((mask & ATTR_SIZE) == 0); 633 634 /* 635 * If disk quotas is on, we make sure that the dquots do exist on disk, 636 * before we start any other transactions. Trying to do this later 637 * is messy. We don't care to take a readlock to look at the ids 638 * in inode here, because we can't hold it across the trans_reserve. 639 * If the IDs do change before we take the ilock, we're covered 640 * because the i_*dquot fields will get updated anyway. 641 */ 642 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 643 uint qflags = 0; 644 645 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 646 uid = iattr->ia_uid; 647 qflags |= XFS_QMOPT_UQUOTA; 648 } else { 649 uid = inode->i_uid; 650 } 651 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 652 gid = iattr->ia_gid; 653 qflags |= XFS_QMOPT_GQUOTA; 654 } else { 655 gid = inode->i_gid; 656 } 657 658 /* 659 * We take a reference when we initialize udqp and gdqp, 660 * so it is important that we never blindly double trip on 661 * the same variable. See xfs_create() for an example. 662 */ 663 ASSERT(udqp == NULL); 664 ASSERT(gdqp == NULL); 665 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid), 666 xfs_kgid_to_gid(gid), 667 xfs_get_projid(ip), 668 qflags, &udqp, &gdqp, NULL); 669 if (error) 670 return error; 671 } 672 673 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); 674 if (error) 675 goto out_dqrele; 676 677 xfs_ilock(ip, XFS_ILOCK_EXCL); 678 xfs_trans_ijoin(tp, ip, 0); 679 680 /* 681 * Change file ownership. Must be the owner or privileged. 682 */ 683 if (mask & (ATTR_UID|ATTR_GID)) { 684 /* 685 * These IDs could have changed since we last looked at them. 686 * But, we're assured that if the ownership did change 687 * while we didn't have the inode locked, inode's dquot(s) 688 * would have changed also. 689 */ 690 iuid = inode->i_uid; 691 igid = inode->i_gid; 692 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; 693 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; 694 695 /* 696 * Do a quota reservation only if uid/gid is actually 697 * going to change. 698 */ 699 if (XFS_IS_QUOTA_RUNNING(mp) && 700 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) || 701 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) { 702 ASSERT(tp); 703 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp, 704 NULL, capable(CAP_FOWNER) ? 705 XFS_QMOPT_FORCE_RES : 0); 706 if (error) /* out of quota */ 707 goto out_cancel; 708 } 709 } 710 711 /* 712 * Change file ownership. Must be the owner or privileged. 713 */ 714 if (mask & (ATTR_UID|ATTR_GID)) { 715 /* 716 * CAP_FSETID overrides the following restrictions: 717 * 718 * The set-user-ID and set-group-ID bits of a file will be 719 * cleared upon successful return from chown() 720 */ 721 if ((inode->i_mode & (S_ISUID|S_ISGID)) && 722 !capable(CAP_FSETID)) 723 inode->i_mode &= ~(S_ISUID|S_ISGID); 724 725 /* 726 * Change the ownerships and register quota modifications 727 * in the transaction. 728 */ 729 if (!uid_eq(iuid, uid)) { 730 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { 731 ASSERT(mask & ATTR_UID); 732 ASSERT(udqp); 733 olddquot1 = xfs_qm_vop_chown(tp, ip, 734 &ip->i_udquot, udqp); 735 } 736 ip->i_d.di_uid = xfs_kuid_to_uid(uid); 737 inode->i_uid = uid; 738 } 739 if (!gid_eq(igid, gid)) { 740 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { 741 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) || 742 !XFS_IS_PQUOTA_ON(mp)); 743 ASSERT(mask & ATTR_GID); 744 ASSERT(gdqp); 745 olddquot2 = xfs_qm_vop_chown(tp, ip, 746 &ip->i_gdquot, gdqp); 747 } 748 ip->i_d.di_gid = xfs_kgid_to_gid(gid); 749 inode->i_gid = gid; 750 } 751 } 752 753 if (mask & ATTR_MODE) 754 xfs_setattr_mode(ip, iattr); 755 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) 756 xfs_setattr_time(ip, iattr); 757 758 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 759 760 XFS_STATS_INC(mp, xs_ig_attrchg); 761 762 if (mp->m_flags & XFS_MOUNT_WSYNC) 763 xfs_trans_set_sync(tp); 764 error = xfs_trans_commit(tp); 765 766 xfs_iunlock(ip, XFS_ILOCK_EXCL); 767 768 /* 769 * Release any dquot(s) the inode had kept before chown. 770 */ 771 xfs_qm_dqrele(olddquot1); 772 xfs_qm_dqrele(olddquot2); 773 xfs_qm_dqrele(udqp); 774 xfs_qm_dqrele(gdqp); 775 776 if (error) 777 return error; 778 779 /* 780 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 781 * update. We could avoid this with linked transactions 782 * and passing down the transaction pointer all the way 783 * to attr_set. No previous user of the generic 784 * Posix ACL code seems to care about this issue either. 785 */ 786 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) { 787 error = posix_acl_chmod(inode, inode->i_mode); 788 if (error) 789 return error; 790 } 791 792 return 0; 793 794 out_cancel: 795 xfs_trans_cancel(tp); 796 out_dqrele: 797 xfs_qm_dqrele(udqp); 798 xfs_qm_dqrele(gdqp); 799 return error; 800 } 801 802 int 803 xfs_vn_setattr_nonsize( 804 struct dentry *dentry, 805 struct iattr *iattr) 806 { 807 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 808 int error; 809 810 trace_xfs_setattr(ip); 811 812 error = xfs_vn_change_ok(dentry, iattr); 813 if (error) 814 return error; 815 return xfs_setattr_nonsize(ip, iattr, 0); 816 } 817 818 /* 819 * Truncate file. Must have write permission and not be a directory. 820 * 821 * Caution: The caller of this function is responsible for calling 822 * setattr_prepare() or otherwise verifying the change is fine. 823 */ 824 STATIC int 825 xfs_setattr_size( 826 struct xfs_inode *ip, 827 struct iattr *iattr) 828 { 829 struct xfs_mount *mp = ip->i_mount; 830 struct inode *inode = VFS_I(ip); 831 xfs_off_t oldsize, newsize; 832 struct xfs_trans *tp; 833 int error; 834 uint lock_flags = 0; 835 bool did_zeroing = false; 836 837 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 838 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 839 ASSERT(S_ISREG(inode->i_mode)); 840 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 841 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0); 842 843 oldsize = inode->i_size; 844 newsize = iattr->ia_size; 845 846 /* 847 * Short circuit the truncate case for zero length files. 848 */ 849 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) { 850 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 851 return 0; 852 853 /* 854 * Use the regular setattr path to update the timestamps. 855 */ 856 iattr->ia_valid &= ~ATTR_SIZE; 857 return xfs_setattr_nonsize(ip, iattr, 0); 858 } 859 860 /* 861 * Make sure that the dquots are attached to the inode. 862 */ 863 error = xfs_qm_dqattach(ip); 864 if (error) 865 return error; 866 867 /* 868 * Wait for all direct I/O to complete. 869 */ 870 inode_dio_wait(inode); 871 872 /* 873 * File data changes must be complete before we start the transaction to 874 * modify the inode. This needs to be done before joining the inode to 875 * the transaction because the inode cannot be unlocked once it is a 876 * part of the transaction. 877 * 878 * Start with zeroing any data beyond EOF that we may expose on file 879 * extension, or zeroing out the rest of the block on a downward 880 * truncate. 881 */ 882 if (newsize > oldsize) { 883 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 884 error = iomap_zero_range(inode, oldsize, newsize - oldsize, 885 &did_zeroing, &xfs_iomap_ops); 886 } else { 887 error = iomap_truncate_page(inode, newsize, &did_zeroing, 888 &xfs_iomap_ops); 889 } 890 891 if (error) 892 return error; 893 894 /* 895 * We've already locked out new page faults, so now we can safely remove 896 * pages from the page cache knowing they won't get refaulted until we 897 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 898 * complete. The truncate_setsize() call also cleans partial EOF page 899 * PTEs on extending truncates and hence ensures sub-page block size 900 * filesystems are correctly handled, too. 901 * 902 * We have to do all the page cache truncate work outside the 903 * transaction context as the "lock" order is page lock->log space 904 * reservation as defined by extent allocation in the writeback path. 905 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 906 * having already truncated the in-memory version of the file (i.e. made 907 * user visible changes). There's not much we can do about this, except 908 * to hope that the caller sees ENOMEM and retries the truncate 909 * operation. 910 * 911 * And we update in-core i_size and truncate page cache beyond newsize 912 * before writeback the [di_size, newsize] range, so we're guaranteed 913 * not to write stale data past the new EOF on truncate down. 914 */ 915 truncate_setsize(inode, newsize); 916 917 /* 918 * We are going to log the inode size change in this transaction so 919 * any previous writes that are beyond the on disk EOF and the new 920 * EOF that have not been written out need to be written here. If we 921 * do not write the data out, we expose ourselves to the null files 922 * problem. Note that this includes any block zeroing we did above; 923 * otherwise those blocks may not be zeroed after a crash. 924 */ 925 if (did_zeroing || 926 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) { 927 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 928 ip->i_d.di_size, newsize - 1); 929 if (error) 930 return error; 931 } 932 933 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 934 if (error) 935 return error; 936 937 lock_flags |= XFS_ILOCK_EXCL; 938 xfs_ilock(ip, XFS_ILOCK_EXCL); 939 xfs_trans_ijoin(tp, ip, 0); 940 941 /* 942 * Only change the c/mtime if we are changing the size or we are 943 * explicitly asked to change it. This handles the semantic difference 944 * between truncate() and ftruncate() as implemented in the VFS. 945 * 946 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 947 * special case where we need to update the times despite not having 948 * these flags set. For all other operations the VFS set these flags 949 * explicitly if it wants a timestamp update. 950 */ 951 if (newsize != oldsize && 952 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 953 iattr->ia_ctime = iattr->ia_mtime = 954 current_time(inode); 955 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 956 } 957 958 /* 959 * The first thing we do is set the size to new_size permanently on 960 * disk. This way we don't have to worry about anyone ever being able 961 * to look at the data being freed even in the face of a crash. 962 * What we're getting around here is the case where we free a block, it 963 * is allocated to another file, it is written to, and then we crash. 964 * If the new data gets written to the file but the log buffers 965 * containing the free and reallocation don't, then we'd end up with 966 * garbage in the blocks being freed. As long as we make the new size 967 * permanent before actually freeing any blocks it doesn't matter if 968 * they get written to. 969 */ 970 ip->i_d.di_size = newsize; 971 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 972 973 if (newsize <= oldsize) { 974 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 975 if (error) 976 goto out_trans_cancel; 977 978 /* 979 * Truncated "down", so we're removing references to old data 980 * here - if we delay flushing for a long time, we expose 981 * ourselves unduly to the notorious NULL files problem. So, 982 * we mark this inode and flush it when the file is closed, 983 * and do not wait the usual (long) time for writeout. 984 */ 985 xfs_iflags_set(ip, XFS_ITRUNCATED); 986 987 /* A truncate down always removes post-EOF blocks. */ 988 xfs_inode_clear_eofblocks_tag(ip); 989 } 990 991 if (iattr->ia_valid & ATTR_MODE) 992 xfs_setattr_mode(ip, iattr); 993 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) 994 xfs_setattr_time(ip, iattr); 995 996 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 997 998 XFS_STATS_INC(mp, xs_ig_attrchg); 999 1000 if (mp->m_flags & XFS_MOUNT_WSYNC) 1001 xfs_trans_set_sync(tp); 1002 1003 error = xfs_trans_commit(tp); 1004 out_unlock: 1005 if (lock_flags) 1006 xfs_iunlock(ip, lock_flags); 1007 return error; 1008 1009 out_trans_cancel: 1010 xfs_trans_cancel(tp); 1011 goto out_unlock; 1012 } 1013 1014 int 1015 xfs_vn_setattr_size( 1016 struct dentry *dentry, 1017 struct iattr *iattr) 1018 { 1019 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 1020 int error; 1021 1022 trace_xfs_setattr(ip); 1023 1024 error = xfs_vn_change_ok(dentry, iattr); 1025 if (error) 1026 return error; 1027 return xfs_setattr_size(ip, iattr); 1028 } 1029 1030 STATIC int 1031 xfs_vn_setattr( 1032 struct dentry *dentry, 1033 struct iattr *iattr) 1034 { 1035 int error; 1036 1037 if (iattr->ia_valid & ATTR_SIZE) { 1038 struct inode *inode = d_inode(dentry); 1039 struct xfs_inode *ip = XFS_I(inode); 1040 uint iolock; 1041 1042 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1043 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1044 1045 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1046 if (error) { 1047 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1048 return error; 1049 } 1050 1051 error = xfs_vn_setattr_size(dentry, iattr); 1052 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1053 } else { 1054 error = xfs_vn_setattr_nonsize(dentry, iattr); 1055 } 1056 1057 return error; 1058 } 1059 1060 STATIC int 1061 xfs_vn_update_time( 1062 struct inode *inode, 1063 struct timespec64 *now, 1064 int flags) 1065 { 1066 struct xfs_inode *ip = XFS_I(inode); 1067 struct xfs_mount *mp = ip->i_mount; 1068 int log_flags = XFS_ILOG_TIMESTAMP; 1069 struct xfs_trans *tp; 1070 int error; 1071 1072 trace_xfs_update_time(ip); 1073 1074 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1075 if (!((flags & S_VERSION) && 1076 inode_maybe_inc_iversion(inode, false))) 1077 return generic_update_time(inode, now, flags); 1078 1079 /* Capture the iversion update that just occurred */ 1080 log_flags |= XFS_ILOG_CORE; 1081 } 1082 1083 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1084 if (error) 1085 return error; 1086 1087 xfs_ilock(ip, XFS_ILOCK_EXCL); 1088 if (flags & S_CTIME) 1089 inode->i_ctime = *now; 1090 if (flags & S_MTIME) 1091 inode->i_mtime = *now; 1092 if (flags & S_ATIME) 1093 inode->i_atime = *now; 1094 1095 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1096 xfs_trans_log_inode(tp, ip, log_flags); 1097 return xfs_trans_commit(tp); 1098 } 1099 1100 STATIC int 1101 xfs_vn_fiemap( 1102 struct inode *inode, 1103 struct fiemap_extent_info *fieinfo, 1104 u64 start, 1105 u64 length) 1106 { 1107 int error; 1108 1109 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1110 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1111 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1112 error = iomap_fiemap(inode, fieinfo, start, length, 1113 &xfs_xattr_iomap_ops); 1114 } else { 1115 error = iomap_fiemap(inode, fieinfo, start, length, 1116 &xfs_iomap_ops); 1117 } 1118 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1119 1120 return error; 1121 } 1122 1123 STATIC int 1124 xfs_vn_tmpfile( 1125 struct inode *dir, 1126 struct dentry *dentry, 1127 umode_t mode) 1128 { 1129 return xfs_generic_create(dir, dentry, mode, 0, true); 1130 } 1131 1132 static const struct inode_operations xfs_inode_operations = { 1133 .get_acl = xfs_get_acl, 1134 .set_acl = xfs_set_acl, 1135 .getattr = xfs_vn_getattr, 1136 .setattr = xfs_vn_setattr, 1137 .listxattr = xfs_vn_listxattr, 1138 .fiemap = xfs_vn_fiemap, 1139 .update_time = xfs_vn_update_time, 1140 }; 1141 1142 static const struct inode_operations xfs_dir_inode_operations = { 1143 .create = xfs_vn_create, 1144 .lookup = xfs_vn_lookup, 1145 .link = xfs_vn_link, 1146 .unlink = xfs_vn_unlink, 1147 .symlink = xfs_vn_symlink, 1148 .mkdir = xfs_vn_mkdir, 1149 /* 1150 * Yes, XFS uses the same method for rmdir and unlink. 1151 * 1152 * There are some subtile differences deeper in the code, 1153 * but we use S_ISDIR to check for those. 1154 */ 1155 .rmdir = xfs_vn_unlink, 1156 .mknod = xfs_vn_mknod, 1157 .rename = xfs_vn_rename, 1158 .get_acl = xfs_get_acl, 1159 .set_acl = xfs_set_acl, 1160 .getattr = xfs_vn_getattr, 1161 .setattr = xfs_vn_setattr, 1162 .listxattr = xfs_vn_listxattr, 1163 .update_time = xfs_vn_update_time, 1164 .tmpfile = xfs_vn_tmpfile, 1165 }; 1166 1167 static const struct inode_operations xfs_dir_ci_inode_operations = { 1168 .create = xfs_vn_create, 1169 .lookup = xfs_vn_ci_lookup, 1170 .link = xfs_vn_link, 1171 .unlink = xfs_vn_unlink, 1172 .symlink = xfs_vn_symlink, 1173 .mkdir = xfs_vn_mkdir, 1174 /* 1175 * Yes, XFS uses the same method for rmdir and unlink. 1176 * 1177 * There are some subtile differences deeper in the code, 1178 * but we use S_ISDIR to check for those. 1179 */ 1180 .rmdir = xfs_vn_unlink, 1181 .mknod = xfs_vn_mknod, 1182 .rename = xfs_vn_rename, 1183 .get_acl = xfs_get_acl, 1184 .set_acl = xfs_set_acl, 1185 .getattr = xfs_vn_getattr, 1186 .setattr = xfs_vn_setattr, 1187 .listxattr = xfs_vn_listxattr, 1188 .update_time = xfs_vn_update_time, 1189 .tmpfile = xfs_vn_tmpfile, 1190 }; 1191 1192 static const struct inode_operations xfs_symlink_inode_operations = { 1193 .get_link = xfs_vn_get_link, 1194 .getattr = xfs_vn_getattr, 1195 .setattr = xfs_vn_setattr, 1196 .listxattr = xfs_vn_listxattr, 1197 .update_time = xfs_vn_update_time, 1198 }; 1199 1200 static const struct inode_operations xfs_inline_symlink_inode_operations = { 1201 .get_link = xfs_vn_get_link_inline, 1202 .getattr = xfs_vn_getattr, 1203 .setattr = xfs_vn_setattr, 1204 .listxattr = xfs_vn_listxattr, 1205 .update_time = xfs_vn_update_time, 1206 }; 1207 1208 /* Figure out if this file actually supports DAX. */ 1209 static bool 1210 xfs_inode_supports_dax( 1211 struct xfs_inode *ip) 1212 { 1213 struct xfs_mount *mp = ip->i_mount; 1214 1215 /* Only supported on non-reflinked files. */ 1216 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip)) 1217 return false; 1218 1219 /* DAX mount option or DAX iflag must be set. */ 1220 if (!(mp->m_flags & XFS_MOUNT_DAX) && 1221 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)) 1222 return false; 1223 1224 /* Block size must match page size */ 1225 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1226 return false; 1227 1228 /* Device has to support DAX too. */ 1229 return xfs_find_daxdev_for_inode(VFS_I(ip)) != NULL; 1230 } 1231 1232 STATIC void 1233 xfs_diflags_to_iflags( 1234 struct inode *inode, 1235 struct xfs_inode *ip) 1236 { 1237 uint16_t flags = ip->i_d.di_flags; 1238 1239 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | 1240 S_NOATIME | S_DAX); 1241 1242 if (flags & XFS_DIFLAG_IMMUTABLE) 1243 inode->i_flags |= S_IMMUTABLE; 1244 if (flags & XFS_DIFLAG_APPEND) 1245 inode->i_flags |= S_APPEND; 1246 if (flags & XFS_DIFLAG_SYNC) 1247 inode->i_flags |= S_SYNC; 1248 if (flags & XFS_DIFLAG_NOATIME) 1249 inode->i_flags |= S_NOATIME; 1250 if (xfs_inode_supports_dax(ip)) 1251 inode->i_flags |= S_DAX; 1252 } 1253 1254 /* 1255 * Initialize the Linux inode. 1256 * 1257 * When reading existing inodes from disk this is called directly from xfs_iget, 1258 * when creating a new inode it is called from xfs_ialloc after setting up the 1259 * inode. These callers have different criteria for clearing XFS_INEW, so leave 1260 * it up to the caller to deal with unlocking the inode appropriately. 1261 */ 1262 void 1263 xfs_setup_inode( 1264 struct xfs_inode *ip) 1265 { 1266 struct inode *inode = &ip->i_vnode; 1267 gfp_t gfp_mask; 1268 1269 inode->i_ino = ip->i_ino; 1270 inode->i_state = I_NEW; 1271 1272 inode_sb_list_add(inode); 1273 /* make the inode look hashed for the writeback code */ 1274 inode_fake_hash(inode); 1275 1276 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid); 1277 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid); 1278 1279 i_size_write(inode, ip->i_d.di_size); 1280 xfs_diflags_to_iflags(inode, ip); 1281 1282 if (S_ISDIR(inode->i_mode)) { 1283 /* 1284 * We set the i_rwsem class here to avoid potential races with 1285 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1286 * after a filehandle lookup has already found the inode in 1287 * cache before it has been unlocked via unlock_new_inode(). 1288 */ 1289 lockdep_set_class(&inode->i_rwsem, 1290 &inode->i_sb->s_type->i_mutex_dir_key); 1291 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); 1292 ip->d_ops = ip->i_mount->m_dir_inode_ops; 1293 } else { 1294 ip->d_ops = ip->i_mount->m_nondir_inode_ops; 1295 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); 1296 } 1297 1298 /* 1299 * Ensure all page cache allocations are done from GFP_NOFS context to 1300 * prevent direct reclaim recursion back into the filesystem and blowing 1301 * stacks or deadlocking. 1302 */ 1303 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1304 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1305 1306 /* 1307 * If there is no attribute fork no ACL can exist on this inode, 1308 * and it can't have any file capabilities attached to it either. 1309 */ 1310 if (!XFS_IFORK_Q(ip)) { 1311 inode_has_no_xattr(inode); 1312 cache_no_acl(inode); 1313 } 1314 } 1315 1316 void 1317 xfs_setup_iops( 1318 struct xfs_inode *ip) 1319 { 1320 struct inode *inode = &ip->i_vnode; 1321 1322 switch (inode->i_mode & S_IFMT) { 1323 case S_IFREG: 1324 inode->i_op = &xfs_inode_operations; 1325 inode->i_fop = &xfs_file_operations; 1326 if (IS_DAX(inode)) 1327 inode->i_mapping->a_ops = &xfs_dax_aops; 1328 else 1329 inode->i_mapping->a_ops = &xfs_address_space_operations; 1330 break; 1331 case S_IFDIR: 1332 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb)) 1333 inode->i_op = &xfs_dir_ci_inode_operations; 1334 else 1335 inode->i_op = &xfs_dir_inode_operations; 1336 inode->i_fop = &xfs_dir_file_operations; 1337 break; 1338 case S_IFLNK: 1339 if (ip->i_df.if_flags & XFS_IFINLINE) 1340 inode->i_op = &xfs_inline_symlink_inode_operations; 1341 else 1342 inode->i_op = &xfs_symlink_inode_operations; 1343 break; 1344 default: 1345 inode->i_op = &xfs_inode_operations; 1346 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1347 break; 1348 } 1349 } 1350