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_da_format.h" 17 #include "xfs_da_btree.h" 18 #include "xfs_attr.h" 19 #include "xfs_trans.h" 20 #include "xfs_trans_space.h" 21 #include "xfs_bmap_btree.h" 22 #include "xfs_trace.h" 23 #include "xfs_icache.h" 24 #include "xfs_symlink.h" 25 #include "xfs_dir2.h" 26 #include "xfs_iomap.h" 27 #include "xfs_error.h" 28 #include "xfs_ioctl.h" 29 #include "xfs_xattr.h" 30 #include "xfs_file.h" 31 #include "xfs_bmap.h" 32 33 #include <linux/posix_acl.h> 34 #include <linux/security.h> 35 #include <linux/iversion.h> 36 #include <linux/fiemap.h> 37 38 /* 39 * Directories have different lock order w.r.t. mmap_lock compared to regular 40 * files. This is due to readdir potentially triggering page faults on a user 41 * buffer inside filldir(), and this happens with the ilock on the directory 42 * held. For regular files, the lock order is the other way around - the 43 * mmap_lock is taken during the page fault, and then we lock the ilock to do 44 * block mapping. Hence we need a different class for the directory ilock so 45 * that lockdep can tell them apart. 46 */ 47 static struct lock_class_key xfs_nondir_ilock_class; 48 static struct lock_class_key xfs_dir_ilock_class; 49 50 static int 51 xfs_initxattrs( 52 struct inode *inode, 53 const struct xattr *xattr_array, 54 void *fs_info) 55 { 56 const struct xattr *xattr; 57 struct xfs_inode *ip = XFS_I(inode); 58 int error = 0; 59 60 for (xattr = xattr_array; xattr->name != NULL; xattr++) { 61 struct xfs_da_args args = { 62 .dp = ip, 63 .attr_filter = XFS_ATTR_SECURE, 64 .name = xattr->name, 65 .namelen = strlen(xattr->name), 66 .value = xattr->value, 67 .valuelen = xattr->value_len, 68 }; 69 error = xfs_attr_change(&args, XFS_ATTRUPDATE_UPSERT); 70 if (error < 0) 71 break; 72 } 73 return error; 74 } 75 76 /* 77 * Hook in SELinux. This is not quite correct yet, what we really need 78 * here (as we do for default ACLs) is a mechanism by which creation of 79 * these attrs can be journalled at inode creation time (along with the 80 * inode, of course, such that log replay can't cause these to be lost). 81 */ 82 int 83 xfs_inode_init_security( 84 struct inode *inode, 85 struct inode *dir, 86 const struct qstr *qstr) 87 { 88 return security_inode_init_security(inode, dir, qstr, 89 &xfs_initxattrs, NULL); 90 } 91 92 static void 93 xfs_dentry_to_name( 94 struct xfs_name *namep, 95 struct dentry *dentry) 96 { 97 namep->name = dentry->d_name.name; 98 namep->len = dentry->d_name.len; 99 namep->type = XFS_DIR3_FT_UNKNOWN; 100 } 101 102 static int 103 xfs_dentry_mode_to_name( 104 struct xfs_name *namep, 105 struct dentry *dentry, 106 int mode) 107 { 108 namep->name = dentry->d_name.name; 109 namep->len = dentry->d_name.len; 110 namep->type = xfs_mode_to_ftype(mode); 111 112 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) 113 return -EFSCORRUPTED; 114 115 return 0; 116 } 117 118 STATIC void 119 xfs_cleanup_inode( 120 struct inode *dir, 121 struct inode *inode, 122 struct dentry *dentry) 123 { 124 struct xfs_name teardown; 125 126 /* Oh, the horror. 127 * If we can't add the ACL or we fail in 128 * xfs_inode_init_security we must back out. 129 * ENOSPC can hit here, among other things. 130 */ 131 xfs_dentry_to_name(&teardown, dentry); 132 133 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); 134 } 135 136 /* 137 * Check to see if we are likely to need an extended attribute to be added to 138 * the inode we are about to allocate. This allows the attribute fork to be 139 * created during the inode allocation, reducing the number of transactions we 140 * need to do in this fast path. 141 * 142 * The security checks are optimistic, but not guaranteed. The two LSMs that 143 * require xattrs to be added here (selinux and smack) are also the only two 144 * LSMs that add a sb->s_security structure to the superblock. Hence if security 145 * is enabled and sb->s_security is set, we have a pretty good idea that we are 146 * going to be asked to add a security xattr immediately after allocating the 147 * xfs inode and instantiating the VFS inode. 148 */ 149 static inline bool 150 xfs_create_need_xattr( 151 struct inode *dir, 152 struct posix_acl *default_acl, 153 struct posix_acl *acl) 154 { 155 if (acl) 156 return true; 157 if (default_acl) 158 return true; 159 #if IS_ENABLED(CONFIG_SECURITY) 160 if (dir->i_sb->s_security) 161 return true; 162 #endif 163 return false; 164 } 165 166 167 STATIC int 168 xfs_generic_create( 169 struct mnt_idmap *idmap, 170 struct inode *dir, 171 struct dentry *dentry, 172 umode_t mode, 173 dev_t rdev, 174 struct file *tmpfile) /* unnamed file */ 175 { 176 struct xfs_icreate_args args = { 177 .idmap = idmap, 178 .pip = XFS_I(dir), 179 .rdev = rdev, 180 .mode = mode, 181 }; 182 struct inode *inode; 183 struct xfs_inode *ip = NULL; 184 struct posix_acl *default_acl, *acl; 185 struct xfs_name name; 186 int error; 187 188 /* 189 * Irix uses Missed'em'V split, but doesn't want to see 190 * the upper 5 bits of (14bit) major. 191 */ 192 if (S_ISCHR(args.mode) || S_ISBLK(args.mode)) { 193 if (unlikely(!sysv_valid_dev(args.rdev) || 194 MAJOR(args.rdev) & ~0x1ff)) 195 return -EINVAL; 196 } else { 197 args.rdev = 0; 198 } 199 200 error = posix_acl_create(dir, &args.mode, &default_acl, &acl); 201 if (error) 202 return error; 203 204 /* Verify mode is valid also for tmpfile case */ 205 error = xfs_dentry_mode_to_name(&name, dentry, args.mode); 206 if (unlikely(error)) 207 goto out_free_acl; 208 209 if (!tmpfile) { 210 if (xfs_create_need_xattr(dir, default_acl, acl)) 211 args.flags |= XFS_ICREATE_INIT_XATTRS; 212 213 error = xfs_create(&args, &name, &ip); 214 } else { 215 args.flags |= XFS_ICREATE_TMPFILE; 216 217 /* 218 * If this temporary file will not be linkable, don't bother 219 * creating an attr fork to receive a parent pointer. 220 */ 221 if (tmpfile->f_flags & O_EXCL) 222 args.flags |= XFS_ICREATE_UNLINKABLE; 223 224 error = xfs_create_tmpfile(&args, &ip); 225 } 226 if (unlikely(error)) 227 goto out_free_acl; 228 229 inode = VFS_I(ip); 230 231 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 232 if (unlikely(error)) 233 goto out_cleanup_inode; 234 235 if (default_acl) { 236 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); 237 if (error) 238 goto out_cleanup_inode; 239 } 240 if (acl) { 241 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); 242 if (error) 243 goto out_cleanup_inode; 244 } 245 246 xfs_setup_iops(ip); 247 248 if (tmpfile) { 249 /* 250 * The VFS requires that any inode fed to d_tmpfile must have 251 * nlink == 1 so that it can decrement the nlink in d_tmpfile. 252 * However, we created the temp file with nlink == 0 because 253 * we're not allowed to put an inode with nlink > 0 on the 254 * unlinked list. Therefore we have to set nlink to 1 so that 255 * d_tmpfile can immediately set it back to zero. 256 */ 257 set_nlink(inode, 1); 258 d_tmpfile(tmpfile, inode); 259 } else 260 d_instantiate(dentry, inode); 261 262 xfs_finish_inode_setup(ip); 263 264 out_free_acl: 265 posix_acl_release(default_acl); 266 posix_acl_release(acl); 267 return error; 268 269 out_cleanup_inode: 270 xfs_finish_inode_setup(ip); 271 if (!tmpfile) 272 xfs_cleanup_inode(dir, inode, dentry); 273 xfs_irele(ip); 274 goto out_free_acl; 275 } 276 277 STATIC int 278 xfs_vn_mknod( 279 struct mnt_idmap *idmap, 280 struct inode *dir, 281 struct dentry *dentry, 282 umode_t mode, 283 dev_t rdev) 284 { 285 return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL); 286 } 287 288 STATIC int 289 xfs_vn_create( 290 struct mnt_idmap *idmap, 291 struct inode *dir, 292 struct dentry *dentry, 293 umode_t mode, 294 bool flags) 295 { 296 return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL); 297 } 298 299 STATIC int 300 xfs_vn_mkdir( 301 struct mnt_idmap *idmap, 302 struct inode *dir, 303 struct dentry *dentry, 304 umode_t mode) 305 { 306 return xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL); 307 } 308 309 STATIC struct dentry * 310 xfs_vn_lookup( 311 struct inode *dir, 312 struct dentry *dentry, 313 unsigned int flags) 314 { 315 struct inode *inode; 316 struct xfs_inode *cip; 317 struct xfs_name name; 318 int error; 319 320 if (dentry->d_name.len >= MAXNAMELEN) 321 return ERR_PTR(-ENAMETOOLONG); 322 323 xfs_dentry_to_name(&name, dentry); 324 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); 325 if (likely(!error)) 326 inode = VFS_I(cip); 327 else if (likely(error == -ENOENT)) 328 inode = NULL; 329 else 330 inode = ERR_PTR(error); 331 return d_splice_alias(inode, dentry); 332 } 333 334 STATIC struct dentry * 335 xfs_vn_ci_lookup( 336 struct inode *dir, 337 struct dentry *dentry, 338 unsigned int flags) 339 { 340 struct xfs_inode *ip; 341 struct xfs_name xname; 342 struct xfs_name ci_name; 343 struct qstr dname; 344 int error; 345 346 if (dentry->d_name.len >= MAXNAMELEN) 347 return ERR_PTR(-ENAMETOOLONG); 348 349 xfs_dentry_to_name(&xname, dentry); 350 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); 351 if (unlikely(error)) { 352 if (unlikely(error != -ENOENT)) 353 return ERR_PTR(error); 354 /* 355 * call d_add(dentry, NULL) here when d_drop_negative_children 356 * is called in xfs_vn_mknod (ie. allow negative dentries 357 * with CI filesystems). 358 */ 359 return NULL; 360 } 361 362 /* if exact match, just splice and exit */ 363 if (!ci_name.name) 364 return d_splice_alias(VFS_I(ip), dentry); 365 366 /* else case-insensitive match... */ 367 dname.name = ci_name.name; 368 dname.len = ci_name.len; 369 dentry = d_add_ci(dentry, VFS_I(ip), &dname); 370 kfree(ci_name.name); 371 return dentry; 372 } 373 374 STATIC int 375 xfs_vn_link( 376 struct dentry *old_dentry, 377 struct inode *dir, 378 struct dentry *dentry) 379 { 380 struct inode *inode = d_inode(old_dentry); 381 struct xfs_name name; 382 int error; 383 384 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); 385 if (unlikely(error)) 386 return error; 387 388 if (IS_PRIVATE(inode)) 389 return -EPERM; 390 391 error = xfs_link(XFS_I(dir), XFS_I(inode), &name); 392 if (unlikely(error)) 393 return error; 394 395 ihold(inode); 396 d_instantiate(dentry, inode); 397 return 0; 398 } 399 400 STATIC int 401 xfs_vn_unlink( 402 struct inode *dir, 403 struct dentry *dentry) 404 { 405 struct xfs_name name; 406 int error; 407 408 xfs_dentry_to_name(&name, dentry); 409 410 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); 411 if (error) 412 return error; 413 414 /* 415 * With unlink, the VFS makes the dentry "negative": no inode, 416 * but still hashed. This is incompatible with case-insensitive 417 * mode, so invalidate (unhash) the dentry in CI-mode. 418 */ 419 if (xfs_has_asciici(XFS_M(dir->i_sb))) 420 d_invalidate(dentry); 421 return 0; 422 } 423 424 STATIC int 425 xfs_vn_symlink( 426 struct mnt_idmap *idmap, 427 struct inode *dir, 428 struct dentry *dentry, 429 const char *symname) 430 { 431 struct inode *inode; 432 struct xfs_inode *cip = NULL; 433 struct xfs_name name; 434 int error; 435 umode_t mode; 436 437 mode = S_IFLNK | 438 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); 439 error = xfs_dentry_mode_to_name(&name, dentry, mode); 440 if (unlikely(error)) 441 goto out; 442 443 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip); 444 if (unlikely(error)) 445 goto out; 446 447 inode = VFS_I(cip); 448 449 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 450 if (unlikely(error)) 451 goto out_cleanup_inode; 452 453 xfs_setup_iops(cip); 454 455 d_instantiate(dentry, inode); 456 xfs_finish_inode_setup(cip); 457 return 0; 458 459 out_cleanup_inode: 460 xfs_finish_inode_setup(cip); 461 xfs_cleanup_inode(dir, inode, dentry); 462 xfs_irele(cip); 463 out: 464 return error; 465 } 466 467 STATIC int 468 xfs_vn_rename( 469 struct mnt_idmap *idmap, 470 struct inode *odir, 471 struct dentry *odentry, 472 struct inode *ndir, 473 struct dentry *ndentry, 474 unsigned int flags) 475 { 476 struct inode *new_inode = d_inode(ndentry); 477 int omode = 0; 478 int error; 479 struct xfs_name oname; 480 struct xfs_name nname; 481 482 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 483 return -EINVAL; 484 485 /* if we are exchanging files, we need to set i_mode of both files */ 486 if (flags & RENAME_EXCHANGE) 487 omode = d_inode(ndentry)->i_mode; 488 489 error = xfs_dentry_mode_to_name(&oname, odentry, omode); 490 if (omode && unlikely(error)) 491 return error; 492 493 error = xfs_dentry_mode_to_name(&nname, ndentry, 494 d_inode(odentry)->i_mode); 495 if (unlikely(error)) 496 return error; 497 498 return xfs_rename(idmap, XFS_I(odir), &oname, 499 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname, 500 new_inode ? XFS_I(new_inode) : NULL, flags); 501 } 502 503 /* 504 * careful here - this function can get called recursively, so 505 * we need to be very careful about how much stack we use. 506 * uio is kmalloced for this reason... 507 */ 508 STATIC const char * 509 xfs_vn_get_link( 510 struct dentry *dentry, 511 struct inode *inode, 512 struct delayed_call *done) 513 { 514 char *link; 515 int error = -ENOMEM; 516 517 if (!dentry) 518 return ERR_PTR(-ECHILD); 519 520 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); 521 if (!link) 522 goto out_err; 523 524 error = xfs_readlink(XFS_I(d_inode(dentry)), link); 525 if (unlikely(error)) 526 goto out_kfree; 527 528 set_delayed_call(done, kfree_link, link); 529 return link; 530 531 out_kfree: 532 kfree(link); 533 out_err: 534 return ERR_PTR(error); 535 } 536 537 static uint32_t 538 xfs_stat_blksize( 539 struct xfs_inode *ip) 540 { 541 struct xfs_mount *mp = ip->i_mount; 542 543 /* 544 * If the file blocks are being allocated from a realtime volume, then 545 * always return the realtime extent size. 546 */ 547 if (XFS_IS_REALTIME_INODE(ip)) 548 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1); 549 550 /* 551 * Allow large block sizes to be reported to userspace programs if the 552 * "largeio" mount option is used. 553 * 554 * If compatibility mode is specified, simply return the basic unit of 555 * caching so that we don't get inefficient read/modify/write I/O from 556 * user apps. Otherwise.... 557 * 558 * If the underlying volume is a stripe, then return the stripe width in 559 * bytes as the recommended I/O size. It is not a stripe and we've set a 560 * default buffered I/O size, return that, otherwise return the compat 561 * default. 562 */ 563 if (xfs_has_large_iosize(mp)) { 564 if (mp->m_swidth) 565 return XFS_FSB_TO_B(mp, mp->m_swidth); 566 if (xfs_has_allocsize(mp)) 567 return 1U << mp->m_allocsize_log; 568 } 569 570 return max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize); 571 } 572 573 STATIC int 574 xfs_vn_getattr( 575 struct mnt_idmap *idmap, 576 const struct path *path, 577 struct kstat *stat, 578 u32 request_mask, 579 unsigned int query_flags) 580 { 581 struct inode *inode = d_inode(path->dentry); 582 struct xfs_inode *ip = XFS_I(inode); 583 struct xfs_mount *mp = ip->i_mount; 584 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode); 585 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); 586 587 trace_xfs_getattr(ip); 588 589 if (xfs_is_shutdown(mp)) 590 return -EIO; 591 592 stat->size = XFS_ISIZE(ip); 593 stat->dev = inode->i_sb->s_dev; 594 stat->mode = inode->i_mode; 595 stat->nlink = inode->i_nlink; 596 stat->uid = vfsuid_into_kuid(vfsuid); 597 stat->gid = vfsgid_into_kgid(vfsgid); 598 stat->ino = ip->i_ino; 599 stat->atime = inode_get_atime(inode); 600 601 fill_mg_cmtime(stat, request_mask, inode); 602 603 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); 604 605 if (xfs_has_v3inodes(mp)) { 606 if (request_mask & STATX_BTIME) { 607 stat->result_mask |= STATX_BTIME; 608 stat->btime = ip->i_crtime; 609 } 610 } 611 612 /* 613 * Note: If you add another clause to set an attribute flag, please 614 * update attributes_mask below. 615 */ 616 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) 617 stat->attributes |= STATX_ATTR_IMMUTABLE; 618 if (ip->i_diflags & XFS_DIFLAG_APPEND) 619 stat->attributes |= STATX_ATTR_APPEND; 620 if (ip->i_diflags & XFS_DIFLAG_NODUMP) 621 stat->attributes |= STATX_ATTR_NODUMP; 622 623 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | 624 STATX_ATTR_APPEND | 625 STATX_ATTR_NODUMP); 626 627 switch (inode->i_mode & S_IFMT) { 628 case S_IFBLK: 629 case S_IFCHR: 630 stat->blksize = BLKDEV_IOSIZE; 631 stat->rdev = inode->i_rdev; 632 break; 633 case S_IFREG: 634 if (request_mask & STATX_DIOALIGN) { 635 struct xfs_buftarg *target = xfs_inode_buftarg(ip); 636 struct block_device *bdev = target->bt_bdev; 637 638 stat->result_mask |= STATX_DIOALIGN; 639 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; 640 stat->dio_offset_align = bdev_logical_block_size(bdev); 641 } 642 fallthrough; 643 default: 644 stat->blksize = xfs_stat_blksize(ip); 645 stat->rdev = 0; 646 break; 647 } 648 649 return 0; 650 } 651 652 static int 653 xfs_vn_change_ok( 654 struct mnt_idmap *idmap, 655 struct dentry *dentry, 656 struct iattr *iattr) 657 { 658 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 659 660 if (xfs_is_readonly(mp)) 661 return -EROFS; 662 663 if (xfs_is_shutdown(mp)) 664 return -EIO; 665 666 return setattr_prepare(idmap, dentry, iattr); 667 } 668 669 /* 670 * Set non-size attributes of an inode. 671 * 672 * Caution: The caller of this function is responsible for calling 673 * setattr_prepare() or otherwise verifying the change is fine. 674 */ 675 static int 676 xfs_setattr_nonsize( 677 struct mnt_idmap *idmap, 678 struct dentry *dentry, 679 struct xfs_inode *ip, 680 struct iattr *iattr) 681 { 682 xfs_mount_t *mp = ip->i_mount; 683 struct inode *inode = VFS_I(ip); 684 int mask = iattr->ia_valid; 685 xfs_trans_t *tp; 686 int error; 687 kuid_t uid = GLOBAL_ROOT_UID; 688 kgid_t gid = GLOBAL_ROOT_GID; 689 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 690 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; 691 692 ASSERT((mask & ATTR_SIZE) == 0); 693 694 /* 695 * If disk quotas is on, we make sure that the dquots do exist on disk, 696 * before we start any other transactions. Trying to do this later 697 * is messy. We don't care to take a readlock to look at the ids 698 * in inode here, because we can't hold it across the trans_reserve. 699 * If the IDs do change before we take the ilock, we're covered 700 * because the i_*dquot fields will get updated anyway. 701 */ 702 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 703 uint qflags = 0; 704 705 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 706 uid = from_vfsuid(idmap, i_user_ns(inode), 707 iattr->ia_vfsuid); 708 qflags |= XFS_QMOPT_UQUOTA; 709 } else { 710 uid = inode->i_uid; 711 } 712 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 713 gid = from_vfsgid(idmap, i_user_ns(inode), 714 iattr->ia_vfsgid); 715 qflags |= XFS_QMOPT_GQUOTA; 716 } else { 717 gid = inode->i_gid; 718 } 719 720 /* 721 * We take a reference when we initialize udqp and gdqp, 722 * so it is important that we never blindly double trip on 723 * the same variable. See xfs_create() for an example. 724 */ 725 ASSERT(udqp == NULL); 726 ASSERT(gdqp == NULL); 727 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, 728 qflags, &udqp, &gdqp, NULL); 729 if (error) 730 return error; 731 } 732 733 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, 734 has_capability_noaudit(current, CAP_FOWNER), &tp); 735 if (error) 736 goto out_dqrele; 737 738 /* 739 * Register quota modifications in the transaction. Must be the owner 740 * or privileged. These IDs could have changed since we last looked at 741 * them. But, we're assured that if the ownership did change while we 742 * didn't have the inode locked, inode's dquot(s) would have changed 743 * also. 744 */ 745 if (XFS_IS_UQUOTA_ON(mp) && 746 i_uid_needs_update(idmap, iattr, inode)) { 747 ASSERT(udqp); 748 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); 749 } 750 if (XFS_IS_GQUOTA_ON(mp) && 751 i_gid_needs_update(idmap, iattr, inode)) { 752 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp)); 753 ASSERT(gdqp); 754 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp); 755 } 756 757 setattr_copy(idmap, inode, iattr); 758 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 759 760 XFS_STATS_INC(mp, xs_ig_attrchg); 761 762 if (xfs_has_wsync(mp)) 763 xfs_trans_set_sync(tp); 764 error = xfs_trans_commit(tp); 765 766 /* 767 * Release any dquot(s) the inode had kept before chown. 768 */ 769 xfs_qm_dqrele(old_udqp); 770 xfs_qm_dqrele(old_gdqp); 771 xfs_qm_dqrele(udqp); 772 xfs_qm_dqrele(gdqp); 773 774 if (error) 775 return error; 776 777 /* 778 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 779 * update. We could avoid this with linked transactions 780 * and passing down the transaction pointer all the way 781 * to attr_set. No previous user of the generic 782 * Posix ACL code seems to care about this issue either. 783 */ 784 if (mask & ATTR_MODE) { 785 error = posix_acl_chmod(idmap, dentry, inode->i_mode); 786 if (error) 787 return error; 788 } 789 790 return 0; 791 792 out_dqrele: 793 xfs_qm_dqrele(udqp); 794 xfs_qm_dqrele(gdqp); 795 return error; 796 } 797 798 /* 799 * Truncate file. Must have write permission and not be a directory. 800 * 801 * Caution: The caller of this function is responsible for calling 802 * setattr_prepare() or otherwise verifying the change is fine. 803 */ 804 STATIC int 805 xfs_setattr_size( 806 struct mnt_idmap *idmap, 807 struct dentry *dentry, 808 struct xfs_inode *ip, 809 struct iattr *iattr) 810 { 811 struct xfs_mount *mp = ip->i_mount; 812 struct inode *inode = VFS_I(ip); 813 xfs_off_t oldsize, newsize; 814 struct xfs_trans *tp; 815 int error; 816 uint lock_flags = 0; 817 uint resblks = 0; 818 bool did_zeroing = false; 819 820 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL); 821 ASSERT(S_ISREG(inode->i_mode)); 822 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 823 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); 824 825 oldsize = inode->i_size; 826 newsize = iattr->ia_size; 827 828 /* 829 * Short circuit the truncate case for zero length files. 830 */ 831 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { 832 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 833 return 0; 834 835 /* 836 * Use the regular setattr path to update the timestamps. 837 */ 838 iattr->ia_valid &= ~ATTR_SIZE; 839 return xfs_setattr_nonsize(idmap, dentry, ip, iattr); 840 } 841 842 /* 843 * Make sure that the dquots are attached to the inode. 844 */ 845 error = xfs_qm_dqattach(ip); 846 if (error) 847 return error; 848 849 /* 850 * Wait for all direct I/O to complete. 851 */ 852 inode_dio_wait(inode); 853 854 /* 855 * File data changes must be complete before we start the transaction to 856 * modify the inode. This needs to be done before joining the inode to 857 * the transaction because the inode cannot be unlocked once it is a 858 * part of the transaction. 859 * 860 * Start with zeroing any data beyond EOF that we may expose on file 861 * extension, or zeroing out the rest of the block on a downward 862 * truncate. 863 */ 864 if (newsize > oldsize) { 865 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 866 error = xfs_zero_range(ip, oldsize, newsize - oldsize, 867 &did_zeroing); 868 } else { 869 error = xfs_truncate_page(ip, newsize, &did_zeroing); 870 } 871 872 if (error) 873 return error; 874 875 /* 876 * We've already locked out new page faults, so now we can safely remove 877 * pages from the page cache knowing they won't get refaulted until we 878 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 879 * complete. The truncate_setsize() call also cleans partial EOF page 880 * PTEs on extending truncates and hence ensures sub-page block size 881 * filesystems are correctly handled, too. 882 * 883 * We have to do all the page cache truncate work outside the 884 * transaction context as the "lock" order is page lock->log space 885 * reservation as defined by extent allocation in the writeback path. 886 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 887 * having already truncated the in-memory version of the file (i.e. made 888 * user visible changes). There's not much we can do about this, except 889 * to hope that the caller sees ENOMEM and retries the truncate 890 * operation. 891 * 892 * And we update in-core i_size and truncate page cache beyond newsize 893 * before writeback the [i_disk_size, newsize] range, so we're 894 * guaranteed not to write stale data past the new EOF on truncate down. 895 */ 896 truncate_setsize(inode, newsize); 897 898 /* 899 * We are going to log the inode size change in this transaction so 900 * any previous writes that are beyond the on disk EOF and the new 901 * EOF that have not been written out need to be written here. If we 902 * do not write the data out, we expose ourselves to the null files 903 * problem. Note that this includes any block zeroing we did above; 904 * otherwise those blocks may not be zeroed after a crash. 905 */ 906 if (did_zeroing || 907 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { 908 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 909 ip->i_disk_size, newsize - 1); 910 if (error) 911 return error; 912 } 913 914 /* 915 * For realtime inode with more than one block rtextsize, we need the 916 * block reservation for bmap btree block allocations/splits that can 917 * happen since it could split the tail written extent and convert the 918 * right beyond EOF one to unwritten. 919 */ 920 if (xfs_inode_has_bigrtalloc(ip)) 921 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 922 923 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks, 924 0, 0, &tp); 925 if (error) 926 return error; 927 928 lock_flags |= XFS_ILOCK_EXCL; 929 xfs_ilock(ip, XFS_ILOCK_EXCL); 930 xfs_trans_ijoin(tp, ip, 0); 931 932 /* 933 * Only change the c/mtime if we are changing the size or we are 934 * explicitly asked to change it. This handles the semantic difference 935 * between truncate() and ftruncate() as implemented in the VFS. 936 * 937 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 938 * special case where we need to update the times despite not having 939 * these flags set. For all other operations the VFS set these flags 940 * explicitly if it wants a timestamp update. 941 */ 942 if (newsize != oldsize && 943 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 944 iattr->ia_ctime = iattr->ia_mtime = 945 current_time(inode); 946 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 947 } 948 949 /* 950 * The first thing we do is set the size to new_size permanently on 951 * disk. This way we don't have to worry about anyone ever being able 952 * to look at the data being freed even in the face of a crash. 953 * What we're getting around here is the case where we free a block, it 954 * is allocated to another file, it is written to, and then we crash. 955 * If the new data gets written to the file but the log buffers 956 * containing the free and reallocation don't, then we'd end up with 957 * garbage in the blocks being freed. As long as we make the new size 958 * permanent before actually freeing any blocks it doesn't matter if 959 * they get written to. 960 */ 961 ip->i_disk_size = newsize; 962 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 963 964 if (newsize <= oldsize) { 965 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 966 if (error) 967 goto out_trans_cancel; 968 969 /* 970 * Truncated "down", so we're removing references to old data 971 * here - if we delay flushing for a long time, we expose 972 * ourselves unduly to the notorious NULL files problem. So, 973 * we mark this inode and flush it when the file is closed, 974 * and do not wait the usual (long) time for writeout. 975 */ 976 xfs_iflags_set(ip, XFS_ITRUNCATED); 977 978 /* A truncate down always removes post-EOF blocks. */ 979 xfs_inode_clear_eofblocks_tag(ip); 980 } 981 982 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID))); 983 setattr_copy(idmap, inode, iattr); 984 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 985 986 XFS_STATS_INC(mp, xs_ig_attrchg); 987 988 if (xfs_has_wsync(mp)) 989 xfs_trans_set_sync(tp); 990 991 error = xfs_trans_commit(tp); 992 out_unlock: 993 if (lock_flags) 994 xfs_iunlock(ip, lock_flags); 995 return error; 996 997 out_trans_cancel: 998 xfs_trans_cancel(tp); 999 goto out_unlock; 1000 } 1001 1002 int 1003 xfs_vn_setattr_size( 1004 struct mnt_idmap *idmap, 1005 struct dentry *dentry, 1006 struct iattr *iattr) 1007 { 1008 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 1009 int error; 1010 1011 trace_xfs_setattr(ip); 1012 1013 error = xfs_vn_change_ok(idmap, dentry, iattr); 1014 if (error) 1015 return error; 1016 return xfs_setattr_size(idmap, dentry, ip, iattr); 1017 } 1018 1019 STATIC int 1020 xfs_vn_setattr( 1021 struct mnt_idmap *idmap, 1022 struct dentry *dentry, 1023 struct iattr *iattr) 1024 { 1025 struct inode *inode = d_inode(dentry); 1026 struct xfs_inode *ip = XFS_I(inode); 1027 int error; 1028 1029 if (iattr->ia_valid & ATTR_SIZE) { 1030 uint iolock; 1031 1032 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1033 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1034 1035 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1036 if (error) { 1037 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1038 return error; 1039 } 1040 1041 error = xfs_vn_setattr_size(idmap, dentry, iattr); 1042 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1043 } else { 1044 trace_xfs_setattr(ip); 1045 1046 error = xfs_vn_change_ok(idmap, dentry, iattr); 1047 if (!error) 1048 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr); 1049 } 1050 1051 return error; 1052 } 1053 1054 STATIC int 1055 xfs_vn_update_time( 1056 struct inode *inode, 1057 int flags) 1058 { 1059 struct xfs_inode *ip = XFS_I(inode); 1060 struct xfs_mount *mp = ip->i_mount; 1061 int log_flags = XFS_ILOG_TIMESTAMP; 1062 struct xfs_trans *tp; 1063 int error; 1064 struct timespec64 now; 1065 1066 trace_xfs_update_time(ip); 1067 1068 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1069 if (!((flags & S_VERSION) && 1070 inode_maybe_inc_iversion(inode, false))) { 1071 generic_update_time(inode, flags); 1072 return 0; 1073 } 1074 1075 /* Capture the iversion update that just occurred */ 1076 log_flags |= XFS_ILOG_CORE; 1077 } 1078 1079 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1080 if (error) 1081 return error; 1082 1083 xfs_ilock(ip, XFS_ILOCK_EXCL); 1084 if (flags & (S_CTIME|S_MTIME)) 1085 now = inode_set_ctime_current(inode); 1086 else 1087 now = current_time(inode); 1088 1089 if (flags & S_MTIME) 1090 inode_set_mtime_to_ts(inode, now); 1091 if (flags & S_ATIME) 1092 inode_set_atime_to_ts(inode, now); 1093 1094 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1095 xfs_trans_log_inode(tp, ip, log_flags); 1096 return xfs_trans_commit(tp); 1097 } 1098 1099 STATIC int 1100 xfs_vn_fiemap( 1101 struct inode *inode, 1102 struct fiemap_extent_info *fieinfo, 1103 u64 start, 1104 u64 length) 1105 { 1106 int error; 1107 1108 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1109 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1110 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1111 error = iomap_fiemap(inode, fieinfo, start, length, 1112 &xfs_xattr_iomap_ops); 1113 } else { 1114 error = iomap_fiemap(inode, fieinfo, start, length, 1115 &xfs_read_iomap_ops); 1116 } 1117 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1118 1119 return error; 1120 } 1121 1122 STATIC int 1123 xfs_vn_tmpfile( 1124 struct mnt_idmap *idmap, 1125 struct inode *dir, 1126 struct file *file, 1127 umode_t mode) 1128 { 1129 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file); 1130 1131 return finish_open_simple(file, err); 1132 } 1133 1134 static const struct inode_operations xfs_inode_operations = { 1135 .get_inode_acl = xfs_get_acl, 1136 .set_acl = xfs_set_acl, 1137 .getattr = xfs_vn_getattr, 1138 .setattr = xfs_vn_setattr, 1139 .listxattr = xfs_vn_listxattr, 1140 .fiemap = xfs_vn_fiemap, 1141 .update_time = xfs_vn_update_time, 1142 .fileattr_get = xfs_fileattr_get, 1143 .fileattr_set = xfs_fileattr_set, 1144 }; 1145 1146 static const struct inode_operations xfs_dir_inode_operations = { 1147 .create = xfs_vn_create, 1148 .lookup = xfs_vn_lookup, 1149 .link = xfs_vn_link, 1150 .unlink = xfs_vn_unlink, 1151 .symlink = xfs_vn_symlink, 1152 .mkdir = xfs_vn_mkdir, 1153 /* 1154 * Yes, XFS uses the same method for rmdir and unlink. 1155 * 1156 * There are some subtile differences deeper in the code, 1157 * but we use S_ISDIR to check for those. 1158 */ 1159 .rmdir = xfs_vn_unlink, 1160 .mknod = xfs_vn_mknod, 1161 .rename = xfs_vn_rename, 1162 .get_inode_acl = xfs_get_acl, 1163 .set_acl = xfs_set_acl, 1164 .getattr = xfs_vn_getattr, 1165 .setattr = xfs_vn_setattr, 1166 .listxattr = xfs_vn_listxattr, 1167 .update_time = xfs_vn_update_time, 1168 .tmpfile = xfs_vn_tmpfile, 1169 .fileattr_get = xfs_fileattr_get, 1170 .fileattr_set = xfs_fileattr_set, 1171 }; 1172 1173 static const struct inode_operations xfs_dir_ci_inode_operations = { 1174 .create = xfs_vn_create, 1175 .lookup = xfs_vn_ci_lookup, 1176 .link = xfs_vn_link, 1177 .unlink = xfs_vn_unlink, 1178 .symlink = xfs_vn_symlink, 1179 .mkdir = xfs_vn_mkdir, 1180 /* 1181 * Yes, XFS uses the same method for rmdir and unlink. 1182 * 1183 * There are some subtile differences deeper in the code, 1184 * but we use S_ISDIR to check for those. 1185 */ 1186 .rmdir = xfs_vn_unlink, 1187 .mknod = xfs_vn_mknod, 1188 .rename = xfs_vn_rename, 1189 .get_inode_acl = xfs_get_acl, 1190 .set_acl = xfs_set_acl, 1191 .getattr = xfs_vn_getattr, 1192 .setattr = xfs_vn_setattr, 1193 .listxattr = xfs_vn_listxattr, 1194 .update_time = xfs_vn_update_time, 1195 .tmpfile = xfs_vn_tmpfile, 1196 .fileattr_get = xfs_fileattr_get, 1197 .fileattr_set = xfs_fileattr_set, 1198 }; 1199 1200 static const struct inode_operations xfs_symlink_inode_operations = { 1201 .get_link = xfs_vn_get_link, 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 regular files. */ 1216 if (!S_ISREG(VFS_I(ip)->i_mode)) 1217 return false; 1218 1219 /* Block size must match page size */ 1220 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1221 return false; 1222 1223 /* Device has to support DAX too. */ 1224 return xfs_inode_buftarg(ip)->bt_daxdev != NULL; 1225 } 1226 1227 static bool 1228 xfs_inode_should_enable_dax( 1229 struct xfs_inode *ip) 1230 { 1231 if (!IS_ENABLED(CONFIG_FS_DAX)) 1232 return false; 1233 if (xfs_has_dax_never(ip->i_mount)) 1234 return false; 1235 if (!xfs_inode_supports_dax(ip)) 1236 return false; 1237 if (xfs_has_dax_always(ip->i_mount)) 1238 return true; 1239 if (ip->i_diflags2 & XFS_DIFLAG2_DAX) 1240 return true; 1241 return false; 1242 } 1243 1244 void 1245 xfs_diflags_to_iflags( 1246 struct xfs_inode *ip, 1247 bool init) 1248 { 1249 struct inode *inode = VFS_I(ip); 1250 unsigned int xflags = xfs_ip2xflags(ip); 1251 unsigned int flags = 0; 1252 1253 ASSERT(!(IS_DAX(inode) && init)); 1254 1255 if (xflags & FS_XFLAG_IMMUTABLE) 1256 flags |= S_IMMUTABLE; 1257 if (xflags & FS_XFLAG_APPEND) 1258 flags |= S_APPEND; 1259 if (xflags & FS_XFLAG_SYNC) 1260 flags |= S_SYNC; 1261 if (xflags & FS_XFLAG_NOATIME) 1262 flags |= S_NOATIME; 1263 if (init && xfs_inode_should_enable_dax(ip)) 1264 flags |= S_DAX; 1265 1266 /* 1267 * S_DAX can only be set during inode initialization and is never set by 1268 * the VFS, so we cannot mask off S_DAX in i_flags. 1269 */ 1270 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); 1271 inode->i_flags |= flags; 1272 } 1273 1274 /* 1275 * Initialize the Linux inode. 1276 * 1277 * When reading existing inodes from disk this is called directly from xfs_iget, 1278 * when creating a new inode it is called from xfs_init_new_inode after setting 1279 * up the inode. These callers have different criteria for clearing XFS_INEW, so 1280 * leave it up to the caller to deal with unlocking the inode appropriately. 1281 */ 1282 void 1283 xfs_setup_inode( 1284 struct xfs_inode *ip) 1285 { 1286 struct inode *inode = &ip->i_vnode; 1287 gfp_t gfp_mask; 1288 1289 inode->i_ino = ip->i_ino; 1290 inode->i_state |= I_NEW; 1291 1292 inode_sb_list_add(inode); 1293 /* make the inode look hashed for the writeback code */ 1294 inode_fake_hash(inode); 1295 1296 i_size_write(inode, ip->i_disk_size); 1297 xfs_diflags_to_iflags(ip, true); 1298 1299 if (S_ISDIR(inode->i_mode)) { 1300 /* 1301 * We set the i_rwsem class here to avoid potential races with 1302 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1303 * after a filehandle lookup has already found the inode in 1304 * cache before it has been unlocked via unlock_new_inode(). 1305 */ 1306 lockdep_set_class(&inode->i_rwsem, 1307 &inode->i_sb->s_type->i_mutex_dir_key); 1308 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class); 1309 } else { 1310 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class); 1311 } 1312 1313 /* 1314 * Ensure all page cache allocations are done from GFP_NOFS context to 1315 * prevent direct reclaim recursion back into the filesystem and blowing 1316 * stacks or deadlocking. 1317 */ 1318 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1319 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1320 1321 /* 1322 * For real-time inodes update the stable write flags to that of the RT 1323 * device instead of the data device. 1324 */ 1325 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip)) 1326 xfs_update_stable_writes(ip); 1327 1328 /* 1329 * If there is no attribute fork no ACL can exist on this inode, 1330 * and it can't have any file capabilities attached to it either. 1331 */ 1332 if (!xfs_inode_has_attr_fork(ip)) { 1333 inode_has_no_xattr(inode); 1334 cache_no_acl(inode); 1335 } 1336 } 1337 1338 void 1339 xfs_setup_iops( 1340 struct xfs_inode *ip) 1341 { 1342 struct inode *inode = &ip->i_vnode; 1343 1344 switch (inode->i_mode & S_IFMT) { 1345 case S_IFREG: 1346 inode->i_op = &xfs_inode_operations; 1347 inode->i_fop = &xfs_file_operations; 1348 if (IS_DAX(inode)) 1349 inode->i_mapping->a_ops = &xfs_dax_aops; 1350 else 1351 inode->i_mapping->a_ops = &xfs_address_space_operations; 1352 break; 1353 case S_IFDIR: 1354 if (xfs_has_asciici(XFS_M(inode->i_sb))) 1355 inode->i_op = &xfs_dir_ci_inode_operations; 1356 else 1357 inode->i_op = &xfs_dir_inode_operations; 1358 inode->i_fop = &xfs_dir_file_operations; 1359 break; 1360 case S_IFLNK: 1361 inode->i_op = &xfs_symlink_inode_operations; 1362 break; 1363 default: 1364 inode->i_op = &xfs_inode_operations; 1365 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1366 break; 1367 } 1368 } 1369