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 #include "xfs_zone_alloc.h" 33 34 #include <linux/posix_acl.h> 35 #include <linux/security.h> 36 #include <linux/iversion.h> 37 #include <linux/fiemap.h> 38 39 /* 40 * Directories have different lock order w.r.t. mmap_lock compared to regular 41 * files. This is due to readdir potentially triggering page faults on a user 42 * buffer inside filldir(), and this happens with the ilock on the directory 43 * held. For regular files, the lock order is the other way around - the 44 * mmap_lock is taken during the page fault, and then we lock the ilock to do 45 * block mapping. Hence we need a different class for the directory ilock so 46 * that lockdep can tell them apart. Directories in the metadata directory 47 * tree get a separate class so that lockdep reports will warn us if someone 48 * ever tries to lock regular directories after locking metadata directories. 49 */ 50 static struct lock_class_key xfs_nondir_ilock_class; 51 static struct lock_class_key xfs_dir_ilock_class; 52 53 static int 54 xfs_initxattrs( 55 struct inode *inode, 56 const struct xattr *xattr_array, 57 void *fs_info) 58 { 59 const struct xattr *xattr; 60 struct xfs_inode *ip = XFS_I(inode); 61 int error = 0; 62 63 for (xattr = xattr_array; xattr->name != NULL; xattr++) { 64 struct xfs_da_args args = { 65 .dp = ip, 66 .attr_filter = XFS_ATTR_SECURE, 67 .name = xattr->name, 68 .namelen = strlen(xattr->name), 69 .value = xattr->value, 70 .valuelen = xattr->value_len, 71 }; 72 error = xfs_attr_change(&args, XFS_ATTRUPDATE_UPSERT); 73 if (error < 0) 74 break; 75 } 76 return error; 77 } 78 79 /* 80 * Hook in SELinux. This is not quite correct yet, what we really need 81 * here (as we do for default ACLs) is a mechanism by which creation of 82 * these attrs can be journalled at inode creation time (along with the 83 * inode, of course, such that log replay can't cause these to be lost). 84 */ 85 int 86 xfs_inode_init_security( 87 struct inode *inode, 88 struct inode *dir, 89 const struct qstr *qstr) 90 { 91 return security_inode_init_security(inode, dir, qstr, 92 &xfs_initxattrs, NULL); 93 } 94 95 static void 96 xfs_dentry_to_name( 97 struct xfs_name *namep, 98 struct dentry *dentry) 99 { 100 namep->name = dentry->d_name.name; 101 namep->len = dentry->d_name.len; 102 namep->type = XFS_DIR3_FT_UNKNOWN; 103 } 104 105 static int 106 xfs_dentry_mode_to_name( 107 struct xfs_name *namep, 108 struct dentry *dentry, 109 int mode) 110 { 111 namep->name = dentry->d_name.name; 112 namep->len = dentry->d_name.len; 113 namep->type = xfs_mode_to_ftype(mode); 114 115 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) 116 return -EFSCORRUPTED; 117 118 return 0; 119 } 120 121 STATIC void 122 xfs_cleanup_inode( 123 struct inode *dir, 124 struct inode *inode, 125 struct dentry *dentry) 126 { 127 struct xfs_name teardown; 128 129 /* Oh, the horror. 130 * If we can't add the ACL or we fail in 131 * xfs_inode_init_security we must back out. 132 * ENOSPC can hit here, among other things. 133 */ 134 xfs_dentry_to_name(&teardown, dentry); 135 136 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); 137 } 138 139 /* 140 * Check to see if we are likely to need an extended attribute to be added to 141 * the inode we are about to allocate. This allows the attribute fork to be 142 * created during the inode allocation, reducing the number of transactions we 143 * need to do in this fast path. 144 * 145 * The security checks are optimistic, but not guaranteed. The two LSMs that 146 * require xattrs to be added here (selinux and smack) are also the only two 147 * LSMs that add a sb->s_security structure to the superblock. Hence if security 148 * is enabled and sb->s_security is set, we have a pretty good idea that we are 149 * going to be asked to add a security xattr immediately after allocating the 150 * xfs inode and instantiating the VFS inode. 151 */ 152 static inline bool 153 xfs_create_need_xattr( 154 struct inode *dir, 155 struct posix_acl *default_acl, 156 struct posix_acl *acl) 157 { 158 if (acl) 159 return true; 160 if (default_acl) 161 return true; 162 #if IS_ENABLED(CONFIG_SECURITY) 163 if (dir->i_sb->s_security) 164 return true; 165 #endif 166 return false; 167 } 168 169 170 STATIC int 171 xfs_generic_create( 172 struct mnt_idmap *idmap, 173 struct inode *dir, 174 struct dentry *dentry, 175 umode_t mode, 176 dev_t rdev, 177 struct file *tmpfile) /* unnamed file */ 178 { 179 struct xfs_icreate_args args = { 180 .idmap = idmap, 181 .pip = XFS_I(dir), 182 .rdev = rdev, 183 .mode = mode, 184 }; 185 struct inode *inode; 186 struct xfs_inode *ip = NULL; 187 struct posix_acl *default_acl, *acl; 188 struct xfs_name name; 189 int error; 190 191 /* 192 * Irix uses Missed'em'V split, but doesn't want to see 193 * the upper 5 bits of (14bit) major. 194 */ 195 if (S_ISCHR(args.mode) || S_ISBLK(args.mode)) { 196 if (unlikely(!sysv_valid_dev(args.rdev) || 197 MAJOR(args.rdev) & ~0x1ff)) 198 return -EINVAL; 199 } else { 200 args.rdev = 0; 201 } 202 203 error = posix_acl_create(dir, &args.mode, &default_acl, &acl); 204 if (error) 205 return error; 206 207 /* Verify mode is valid also for tmpfile case */ 208 error = xfs_dentry_mode_to_name(&name, dentry, args.mode); 209 if (unlikely(error)) 210 goto out_free_acl; 211 212 if (!tmpfile) { 213 if (xfs_create_need_xattr(dir, default_acl, acl)) 214 args.flags |= XFS_ICREATE_INIT_XATTRS; 215 216 error = xfs_create(&args, &name, &ip); 217 } else { 218 args.flags |= XFS_ICREATE_TMPFILE; 219 220 /* 221 * If this temporary file will not be linkable, don't bother 222 * creating an attr fork to receive a parent pointer. 223 */ 224 if (tmpfile->f_flags & O_EXCL) 225 args.flags |= XFS_ICREATE_UNLINKABLE; 226 227 error = xfs_create_tmpfile(&args, &ip); 228 } 229 if (unlikely(error)) 230 goto out_free_acl; 231 232 inode = VFS_I(ip); 233 234 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 235 if (unlikely(error)) 236 goto out_cleanup_inode; 237 238 if (default_acl) { 239 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); 240 if (error) 241 goto out_cleanup_inode; 242 } 243 if (acl) { 244 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); 245 if (error) 246 goto out_cleanup_inode; 247 } 248 249 xfs_setup_iops(ip); 250 251 if (tmpfile) { 252 /* 253 * The VFS requires that any inode fed to d_tmpfile must have 254 * nlink == 1 so that it can decrement the nlink in d_tmpfile. 255 * However, we created the temp file with nlink == 0 because 256 * we're not allowed to put an inode with nlink > 0 on the 257 * unlinked list. Therefore we have to set nlink to 1 so that 258 * d_tmpfile can immediately set it back to zero. 259 */ 260 set_nlink(inode, 1); 261 d_tmpfile(tmpfile, inode); 262 } else 263 d_instantiate(dentry, inode); 264 265 xfs_finish_inode_setup(ip); 266 267 out_free_acl: 268 posix_acl_release(default_acl); 269 posix_acl_release(acl); 270 return error; 271 272 out_cleanup_inode: 273 xfs_finish_inode_setup(ip); 274 if (!tmpfile) 275 xfs_cleanup_inode(dir, inode, dentry); 276 xfs_irele(ip); 277 goto out_free_acl; 278 } 279 280 STATIC int 281 xfs_vn_mknod( 282 struct mnt_idmap *idmap, 283 struct inode *dir, 284 struct dentry *dentry, 285 umode_t mode, 286 dev_t rdev) 287 { 288 return xfs_generic_create(idmap, dir, dentry, mode, rdev, NULL); 289 } 290 291 STATIC int 292 xfs_vn_create( 293 struct mnt_idmap *idmap, 294 struct inode *dir, 295 struct dentry *dentry, 296 umode_t mode, 297 bool flags) 298 { 299 return xfs_generic_create(idmap, dir, dentry, mode, 0, NULL); 300 } 301 302 STATIC struct dentry * 303 xfs_vn_mkdir( 304 struct mnt_idmap *idmap, 305 struct inode *dir, 306 struct dentry *dentry, 307 umode_t mode) 308 { 309 return ERR_PTR(xfs_generic_create(idmap, dir, dentry, mode | S_IFDIR, 0, NULL)); 310 } 311 312 STATIC struct dentry * 313 xfs_vn_lookup( 314 struct inode *dir, 315 struct dentry *dentry, 316 unsigned int flags) 317 { 318 struct inode *inode; 319 struct xfs_inode *cip; 320 struct xfs_name name; 321 int error; 322 323 if (dentry->d_name.len >= MAXNAMELEN) 324 return ERR_PTR(-ENAMETOOLONG); 325 326 xfs_dentry_to_name(&name, dentry); 327 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); 328 if (likely(!error)) 329 inode = VFS_I(cip); 330 else if (likely(error == -ENOENT)) 331 inode = NULL; 332 else 333 inode = ERR_PTR(error); 334 return d_splice_alias(inode, dentry); 335 } 336 337 STATIC struct dentry * 338 xfs_vn_ci_lookup( 339 struct inode *dir, 340 struct dentry *dentry, 341 unsigned int flags) 342 { 343 struct xfs_inode *ip; 344 struct xfs_name xname; 345 struct xfs_name ci_name; 346 struct qstr dname; 347 int error; 348 349 if (dentry->d_name.len >= MAXNAMELEN) 350 return ERR_PTR(-ENAMETOOLONG); 351 352 xfs_dentry_to_name(&xname, dentry); 353 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); 354 if (unlikely(error)) { 355 if (unlikely(error != -ENOENT)) 356 return ERR_PTR(error); 357 /* 358 * call d_add(dentry, NULL) here when d_drop_negative_children 359 * is called in xfs_vn_mknod (ie. allow negative dentries 360 * with CI filesystems). 361 */ 362 return NULL; 363 } 364 365 /* if exact match, just splice and exit */ 366 if (!ci_name.name) 367 return d_splice_alias(VFS_I(ip), dentry); 368 369 /* else case-insensitive match... */ 370 dname.name = ci_name.name; 371 dname.len = ci_name.len; 372 dentry = d_add_ci(dentry, VFS_I(ip), &dname); 373 kfree(ci_name.name); 374 return dentry; 375 } 376 377 STATIC int 378 xfs_vn_link( 379 struct dentry *old_dentry, 380 struct inode *dir, 381 struct dentry *dentry) 382 { 383 struct inode *inode = d_inode(old_dentry); 384 struct xfs_name name; 385 int error; 386 387 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); 388 if (unlikely(error)) 389 return error; 390 391 if (IS_PRIVATE(inode)) 392 return -EPERM; 393 394 error = xfs_link(XFS_I(dir), XFS_I(inode), &name); 395 if (unlikely(error)) 396 return error; 397 398 ihold(inode); 399 d_instantiate(dentry, inode); 400 return 0; 401 } 402 403 STATIC int 404 xfs_vn_unlink( 405 struct inode *dir, 406 struct dentry *dentry) 407 { 408 struct xfs_name name; 409 int error; 410 411 xfs_dentry_to_name(&name, dentry); 412 413 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); 414 if (error) 415 return error; 416 417 /* 418 * With unlink, the VFS makes the dentry "negative": no inode, 419 * but still hashed. This is incompatible with case-insensitive 420 * mode, so invalidate (unhash) the dentry in CI-mode. 421 */ 422 if (xfs_has_asciici(XFS_M(dir->i_sb))) 423 d_invalidate(dentry); 424 return 0; 425 } 426 427 STATIC int 428 xfs_vn_symlink( 429 struct mnt_idmap *idmap, 430 struct inode *dir, 431 struct dentry *dentry, 432 const char *symname) 433 { 434 struct inode *inode; 435 struct xfs_inode *cip = NULL; 436 struct xfs_name name; 437 int error; 438 umode_t mode; 439 440 mode = S_IFLNK | 441 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); 442 error = xfs_dentry_mode_to_name(&name, dentry, mode); 443 if (unlikely(error)) 444 goto out; 445 446 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip); 447 if (unlikely(error)) 448 goto out; 449 450 inode = VFS_I(cip); 451 452 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 453 if (unlikely(error)) 454 goto out_cleanup_inode; 455 456 xfs_setup_iops(cip); 457 458 d_instantiate(dentry, inode); 459 xfs_finish_inode_setup(cip); 460 return 0; 461 462 out_cleanup_inode: 463 xfs_finish_inode_setup(cip); 464 xfs_cleanup_inode(dir, inode, dentry); 465 xfs_irele(cip); 466 out: 467 return error; 468 } 469 470 STATIC int 471 xfs_vn_rename( 472 struct mnt_idmap *idmap, 473 struct inode *odir, 474 struct dentry *odentry, 475 struct inode *ndir, 476 struct dentry *ndentry, 477 unsigned int flags) 478 { 479 struct inode *new_inode = d_inode(ndentry); 480 int omode = 0; 481 int error; 482 struct xfs_name oname; 483 struct xfs_name nname; 484 485 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 486 return -EINVAL; 487 488 /* if we are exchanging files, we need to set i_mode of both files */ 489 if (flags & RENAME_EXCHANGE) 490 omode = d_inode(ndentry)->i_mode; 491 492 error = xfs_dentry_mode_to_name(&oname, odentry, omode); 493 if (omode && unlikely(error)) 494 return error; 495 496 error = xfs_dentry_mode_to_name(&nname, ndentry, 497 d_inode(odentry)->i_mode); 498 if (unlikely(error)) 499 return error; 500 501 return xfs_rename(idmap, XFS_I(odir), &oname, 502 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname, 503 new_inode ? XFS_I(new_inode) : NULL, flags); 504 } 505 506 /* 507 * careful here - this function can get called recursively, so 508 * we need to be very careful about how much stack we use. 509 * uio is kmalloced for this reason... 510 */ 511 STATIC const char * 512 xfs_vn_get_link( 513 struct dentry *dentry, 514 struct inode *inode, 515 struct delayed_call *done) 516 { 517 char *link; 518 int error = -ENOMEM; 519 520 if (!dentry) 521 return ERR_PTR(-ECHILD); 522 523 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); 524 if (!link) 525 goto out_err; 526 527 error = xfs_readlink(XFS_I(d_inode(dentry)), link); 528 if (unlikely(error)) 529 goto out_kfree; 530 531 set_delayed_call(done, kfree_link, link); 532 return link; 533 534 out_kfree: 535 kfree(link); 536 out_err: 537 return ERR_PTR(error); 538 } 539 540 static uint32_t 541 xfs_stat_blksize( 542 struct xfs_inode *ip) 543 { 544 struct xfs_mount *mp = ip->i_mount; 545 546 /* 547 * If the file blocks are being allocated from a realtime volume, then 548 * always return the realtime extent size. 549 */ 550 if (XFS_IS_REALTIME_INODE(ip)) 551 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1); 552 553 /* 554 * Allow large block sizes to be reported to userspace programs if the 555 * "largeio" mount option is used. 556 * 557 * If compatibility mode is specified, simply return the basic unit of 558 * caching so that we don't get inefficient read/modify/write I/O from 559 * user apps. Otherwise.... 560 * 561 * If the underlying volume is a stripe, then return the stripe width in 562 * bytes as the recommended I/O size. It is not a stripe and we've set a 563 * default buffered I/O size, return that, otherwise return the compat 564 * default. 565 */ 566 if (xfs_has_large_iosize(mp)) { 567 if (mp->m_swidth) 568 return XFS_FSB_TO_B(mp, mp->m_swidth); 569 if (xfs_has_allocsize(mp)) 570 return 1U << mp->m_allocsize_log; 571 } 572 573 return max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize); 574 } 575 576 static void 577 xfs_report_dioalign( 578 struct xfs_inode *ip, 579 struct kstat *stat) 580 { 581 struct xfs_buftarg *target = xfs_inode_buftarg(ip); 582 struct block_device *bdev = target->bt_bdev; 583 584 stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN; 585 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; 586 587 /* 588 * For COW inodes, we can only perform out of place writes of entire 589 * allocation units (blocks or RT extents). 590 * For writes smaller than the allocation unit, we must fall back to 591 * buffered I/O to perform read-modify-write cycles. At best this is 592 * highly inefficient; at worst it leads to page cache invalidation 593 * races. Tell applications to avoid this by reporting the larger write 594 * alignment in dio_offset_align, and the smaller read alignment in 595 * dio_read_offset_align. 596 */ 597 stat->dio_read_offset_align = bdev_logical_block_size(bdev); 598 if (xfs_is_cow_inode(ip)) 599 stat->dio_offset_align = xfs_inode_alloc_unitsize(ip); 600 else 601 stat->dio_offset_align = stat->dio_read_offset_align; 602 } 603 604 static void 605 xfs_report_atomic_write( 606 struct xfs_inode *ip, 607 struct kstat *stat) 608 { 609 unsigned int unit_min = 0, unit_max = 0; 610 611 if (xfs_inode_can_atomicwrite(ip)) 612 unit_min = unit_max = ip->i_mount->m_sb.sb_blocksize; 613 generic_fill_statx_atomic_writes(stat, unit_min, unit_max); 614 } 615 616 STATIC int 617 xfs_vn_getattr( 618 struct mnt_idmap *idmap, 619 const struct path *path, 620 struct kstat *stat, 621 u32 request_mask, 622 unsigned int query_flags) 623 { 624 struct inode *inode = d_inode(path->dentry); 625 struct xfs_inode *ip = XFS_I(inode); 626 struct xfs_mount *mp = ip->i_mount; 627 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode); 628 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); 629 630 trace_xfs_getattr(ip); 631 632 if (xfs_is_shutdown(mp)) 633 return -EIO; 634 635 stat->size = XFS_ISIZE(ip); 636 stat->dev = inode->i_sb->s_dev; 637 stat->mode = inode->i_mode; 638 stat->nlink = inode->i_nlink; 639 stat->uid = vfsuid_into_kuid(vfsuid); 640 stat->gid = vfsgid_into_kgid(vfsgid); 641 stat->ino = ip->i_ino; 642 stat->atime = inode_get_atime(inode); 643 644 fill_mg_cmtime(stat, request_mask, inode); 645 646 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); 647 648 if (xfs_has_v3inodes(mp)) { 649 if (request_mask & STATX_BTIME) { 650 stat->result_mask |= STATX_BTIME; 651 stat->btime = ip->i_crtime; 652 } 653 } 654 655 /* 656 * Note: If you add another clause to set an attribute flag, please 657 * update attributes_mask below. 658 */ 659 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) 660 stat->attributes |= STATX_ATTR_IMMUTABLE; 661 if (ip->i_diflags & XFS_DIFLAG_APPEND) 662 stat->attributes |= STATX_ATTR_APPEND; 663 if (ip->i_diflags & XFS_DIFLAG_NODUMP) 664 stat->attributes |= STATX_ATTR_NODUMP; 665 666 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | 667 STATX_ATTR_APPEND | 668 STATX_ATTR_NODUMP); 669 670 switch (inode->i_mode & S_IFMT) { 671 case S_IFBLK: 672 case S_IFCHR: 673 stat->blksize = BLKDEV_IOSIZE; 674 stat->rdev = inode->i_rdev; 675 break; 676 case S_IFREG: 677 if (request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN)) 678 xfs_report_dioalign(ip, stat); 679 if (request_mask & STATX_WRITE_ATOMIC) 680 xfs_report_atomic_write(ip, stat); 681 fallthrough; 682 default: 683 stat->blksize = xfs_stat_blksize(ip); 684 stat->rdev = 0; 685 break; 686 } 687 688 return 0; 689 } 690 691 static int 692 xfs_vn_change_ok( 693 struct mnt_idmap *idmap, 694 struct dentry *dentry, 695 struct iattr *iattr) 696 { 697 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 698 699 if (xfs_is_readonly(mp)) 700 return -EROFS; 701 702 if (xfs_is_shutdown(mp)) 703 return -EIO; 704 705 return setattr_prepare(idmap, dentry, iattr); 706 } 707 708 /* 709 * Set non-size attributes of an inode. 710 * 711 * Caution: The caller of this function is responsible for calling 712 * setattr_prepare() or otherwise verifying the change is fine. 713 */ 714 static int 715 xfs_setattr_nonsize( 716 struct mnt_idmap *idmap, 717 struct dentry *dentry, 718 struct xfs_inode *ip, 719 struct iattr *iattr) 720 { 721 xfs_mount_t *mp = ip->i_mount; 722 struct inode *inode = VFS_I(ip); 723 int mask = iattr->ia_valid; 724 xfs_trans_t *tp; 725 int error; 726 kuid_t uid = GLOBAL_ROOT_UID; 727 kgid_t gid = GLOBAL_ROOT_GID; 728 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 729 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; 730 731 ASSERT((mask & ATTR_SIZE) == 0); 732 733 /* 734 * If disk quotas is on, we make sure that the dquots do exist on disk, 735 * before we start any other transactions. Trying to do this later 736 * is messy. We don't care to take a readlock to look at the ids 737 * in inode here, because we can't hold it across the trans_reserve. 738 * If the IDs do change before we take the ilock, we're covered 739 * because the i_*dquot fields will get updated anyway. 740 */ 741 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 742 uint qflags = 0; 743 744 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 745 uid = from_vfsuid(idmap, i_user_ns(inode), 746 iattr->ia_vfsuid); 747 qflags |= XFS_QMOPT_UQUOTA; 748 } else { 749 uid = inode->i_uid; 750 } 751 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 752 gid = from_vfsgid(idmap, i_user_ns(inode), 753 iattr->ia_vfsgid); 754 qflags |= XFS_QMOPT_GQUOTA; 755 } else { 756 gid = inode->i_gid; 757 } 758 759 /* 760 * We take a reference when we initialize udqp and gdqp, 761 * so it is important that we never blindly double trip on 762 * the same variable. See xfs_create() for an example. 763 */ 764 ASSERT(udqp == NULL); 765 ASSERT(gdqp == NULL); 766 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, 767 qflags, &udqp, &gdqp, NULL); 768 if (error) 769 return error; 770 } 771 772 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, 773 has_capability_noaudit(current, CAP_FOWNER), &tp); 774 if (error) 775 goto out_dqrele; 776 777 /* 778 * Register quota modifications in the transaction. Must be the owner 779 * or privileged. These IDs could have changed since we last looked at 780 * them. But, we're assured that if the ownership did change while we 781 * didn't have the inode locked, inode's dquot(s) would have changed 782 * also. 783 */ 784 if (XFS_IS_UQUOTA_ON(mp) && 785 i_uid_needs_update(idmap, iattr, inode)) { 786 ASSERT(udqp); 787 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); 788 } 789 if (XFS_IS_GQUOTA_ON(mp) && 790 i_gid_needs_update(idmap, iattr, inode)) { 791 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp)); 792 ASSERT(gdqp); 793 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp); 794 } 795 796 setattr_copy(idmap, inode, iattr); 797 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 798 799 XFS_STATS_INC(mp, xs_ig_attrchg); 800 801 if (xfs_has_wsync(mp)) 802 xfs_trans_set_sync(tp); 803 error = xfs_trans_commit(tp); 804 805 /* 806 * Release any dquot(s) the inode had kept before chown. 807 */ 808 xfs_qm_dqrele(old_udqp); 809 xfs_qm_dqrele(old_gdqp); 810 xfs_qm_dqrele(udqp); 811 xfs_qm_dqrele(gdqp); 812 813 if (error) 814 return error; 815 816 /* 817 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 818 * update. We could avoid this with linked transactions 819 * and passing down the transaction pointer all the way 820 * to attr_set. No previous user of the generic 821 * Posix ACL code seems to care about this issue either. 822 */ 823 if (mask & ATTR_MODE) { 824 error = posix_acl_chmod(idmap, dentry, inode->i_mode); 825 if (error) 826 return error; 827 } 828 829 return 0; 830 831 out_dqrele: 832 xfs_qm_dqrele(udqp); 833 xfs_qm_dqrele(gdqp); 834 return error; 835 } 836 837 /* 838 * Truncate file. Must have write permission and not be a directory. 839 * 840 * Caution: The caller of this function is responsible for calling 841 * setattr_prepare() or otherwise verifying the change is fine. 842 */ 843 STATIC int 844 xfs_setattr_size( 845 struct mnt_idmap *idmap, 846 struct dentry *dentry, 847 struct xfs_inode *ip, 848 struct iattr *iattr) 849 { 850 struct xfs_mount *mp = ip->i_mount; 851 struct inode *inode = VFS_I(ip); 852 xfs_off_t oldsize, newsize; 853 struct xfs_trans *tp; 854 int error; 855 uint lock_flags = 0; 856 uint resblks = 0; 857 bool did_zeroing = false; 858 struct xfs_zone_alloc_ctx ac = { }; 859 860 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL); 861 ASSERT(S_ISREG(inode->i_mode)); 862 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 863 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); 864 865 oldsize = inode->i_size; 866 newsize = iattr->ia_size; 867 868 /* 869 * Short circuit the truncate case for zero length files. 870 */ 871 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { 872 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 873 return 0; 874 875 /* 876 * Use the regular setattr path to update the timestamps. 877 */ 878 iattr->ia_valid &= ~ATTR_SIZE; 879 return xfs_setattr_nonsize(idmap, dentry, ip, iattr); 880 } 881 882 /* 883 * Make sure that the dquots are attached to the inode. 884 */ 885 error = xfs_qm_dqattach(ip); 886 if (error) 887 return error; 888 889 /* 890 * Wait for all direct I/O to complete. 891 */ 892 inode_dio_wait(inode); 893 894 /* 895 * Normally xfs_zoned_space_reserve is supposed to be called outside the 896 * IOLOCK. For truncate we can't do that since ->setattr is called with 897 * it already held by the VFS. So for now chicken out and try to 898 * allocate space under it. 899 * 900 * To avoid deadlocks this means we can't block waiting for space, which 901 * can lead to spurious -ENOSPC if there are no directly available 902 * blocks. We mitigate this a bit by allowing zeroing to dip into the 903 * reserved pool, but eventually the VFS calling convention needs to 904 * change. 905 */ 906 if (xfs_is_zoned_inode(ip)) { 907 error = xfs_zoned_space_reserve(ip, 1, 908 XFS_ZR_NOWAIT | XFS_ZR_RESERVED, &ac); 909 if (error) { 910 if (error == -EAGAIN) 911 return -ENOSPC; 912 return error; 913 } 914 } 915 916 /* 917 * File data changes must be complete before we start the transaction to 918 * modify the inode. This needs to be done before joining the inode to 919 * the transaction because the inode cannot be unlocked once it is a 920 * part of the transaction. 921 * 922 * Start with zeroing any data beyond EOF that we may expose on file 923 * extension, or zeroing out the rest of the block on a downward 924 * truncate. 925 */ 926 if (newsize > oldsize) { 927 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 928 error = xfs_zero_range(ip, oldsize, newsize - oldsize, 929 &ac, &did_zeroing); 930 } else { 931 error = xfs_truncate_page(ip, newsize, &ac, &did_zeroing); 932 } 933 934 if (xfs_is_zoned_inode(ip)) 935 xfs_zoned_space_unreserve(ip, &ac); 936 937 if (error) 938 return error; 939 940 /* 941 * We've already locked out new page faults, so now we can safely remove 942 * pages from the page cache knowing they won't get refaulted until we 943 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 944 * complete. The truncate_setsize() call also cleans partial EOF page 945 * PTEs on extending truncates and hence ensures sub-page block size 946 * filesystems are correctly handled, too. 947 * 948 * We have to do all the page cache truncate work outside the 949 * transaction context as the "lock" order is page lock->log space 950 * reservation as defined by extent allocation in the writeback path. 951 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 952 * having already truncated the in-memory version of the file (i.e. made 953 * user visible changes). There's not much we can do about this, except 954 * to hope that the caller sees ENOMEM and retries the truncate 955 * operation. 956 * 957 * And we update in-core i_size and truncate page cache beyond newsize 958 * before writeback the [i_disk_size, newsize] range, so we're 959 * guaranteed not to write stale data past the new EOF on truncate down. 960 */ 961 truncate_setsize(inode, newsize); 962 963 /* 964 * We are going to log the inode size change in this transaction so 965 * any previous writes that are beyond the on disk EOF and the new 966 * EOF that have not been written out need to be written here. If we 967 * do not write the data out, we expose ourselves to the null files 968 * problem. Note that this includes any block zeroing we did above; 969 * otherwise those blocks may not be zeroed after a crash. 970 */ 971 if (did_zeroing || 972 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { 973 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 974 ip->i_disk_size, newsize - 1); 975 if (error) 976 return error; 977 } 978 979 /* 980 * For realtime inode with more than one block rtextsize, we need the 981 * block reservation for bmap btree block allocations/splits that can 982 * happen since it could split the tail written extent and convert the 983 * right beyond EOF one to unwritten. 984 */ 985 if (xfs_inode_has_bigrtalloc(ip)) 986 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 987 988 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks, 989 0, 0, &tp); 990 if (error) 991 return error; 992 993 lock_flags |= XFS_ILOCK_EXCL; 994 xfs_ilock(ip, XFS_ILOCK_EXCL); 995 xfs_trans_ijoin(tp, ip, 0); 996 997 /* 998 * Only change the c/mtime if we are changing the size or we are 999 * explicitly asked to change it. This handles the semantic difference 1000 * between truncate() and ftruncate() as implemented in the VFS. 1001 * 1002 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 1003 * special case where we need to update the times despite not having 1004 * these flags set. For all other operations the VFS set these flags 1005 * explicitly if it wants a timestamp update. 1006 */ 1007 if (newsize != oldsize && 1008 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 1009 iattr->ia_ctime = iattr->ia_mtime = 1010 current_time(inode); 1011 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 1012 } 1013 1014 /* 1015 * The first thing we do is set the size to new_size permanently on 1016 * disk. This way we don't have to worry about anyone ever being able 1017 * to look at the data being freed even in the face of a crash. 1018 * What we're getting around here is the case where we free a block, it 1019 * is allocated to another file, it is written to, and then we crash. 1020 * If the new data gets written to the file but the log buffers 1021 * containing the free and reallocation don't, then we'd end up with 1022 * garbage in the blocks being freed. As long as we make the new size 1023 * permanent before actually freeing any blocks it doesn't matter if 1024 * they get written to. 1025 */ 1026 ip->i_disk_size = newsize; 1027 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1028 1029 if (newsize <= oldsize) { 1030 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 1031 if (error) 1032 goto out_trans_cancel; 1033 1034 /* 1035 * Truncated "down", so we're removing references to old data 1036 * here - if we delay flushing for a long time, we expose 1037 * ourselves unduly to the notorious NULL files problem. So, 1038 * we mark this inode and flush it when the file is closed, 1039 * and do not wait the usual (long) time for writeout. 1040 */ 1041 xfs_iflags_set(ip, XFS_ITRUNCATED); 1042 1043 /* A truncate down always removes post-EOF blocks. */ 1044 xfs_inode_clear_eofblocks_tag(ip); 1045 } 1046 1047 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID))); 1048 setattr_copy(idmap, inode, iattr); 1049 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1050 1051 XFS_STATS_INC(mp, xs_ig_attrchg); 1052 1053 if (xfs_has_wsync(mp)) 1054 xfs_trans_set_sync(tp); 1055 1056 error = xfs_trans_commit(tp); 1057 out_unlock: 1058 if (lock_flags) 1059 xfs_iunlock(ip, lock_flags); 1060 return error; 1061 1062 out_trans_cancel: 1063 xfs_trans_cancel(tp); 1064 goto out_unlock; 1065 } 1066 1067 int 1068 xfs_vn_setattr_size( 1069 struct mnt_idmap *idmap, 1070 struct dentry *dentry, 1071 struct iattr *iattr) 1072 { 1073 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 1074 int error; 1075 1076 trace_xfs_setattr(ip); 1077 1078 error = xfs_vn_change_ok(idmap, dentry, iattr); 1079 if (error) 1080 return error; 1081 return xfs_setattr_size(idmap, dentry, ip, iattr); 1082 } 1083 1084 STATIC int 1085 xfs_vn_setattr( 1086 struct mnt_idmap *idmap, 1087 struct dentry *dentry, 1088 struct iattr *iattr) 1089 { 1090 struct inode *inode = d_inode(dentry); 1091 struct xfs_inode *ip = XFS_I(inode); 1092 int error; 1093 1094 if (iattr->ia_valid & ATTR_SIZE) { 1095 uint iolock; 1096 1097 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1098 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1099 1100 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1101 if (error) { 1102 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1103 return error; 1104 } 1105 1106 error = xfs_vn_setattr_size(idmap, dentry, iattr); 1107 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1108 } else { 1109 trace_xfs_setattr(ip); 1110 1111 error = xfs_vn_change_ok(idmap, dentry, iattr); 1112 if (!error) 1113 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr); 1114 } 1115 1116 return error; 1117 } 1118 1119 STATIC int 1120 xfs_vn_update_time( 1121 struct inode *inode, 1122 int flags) 1123 { 1124 struct xfs_inode *ip = XFS_I(inode); 1125 struct xfs_mount *mp = ip->i_mount; 1126 int log_flags = XFS_ILOG_TIMESTAMP; 1127 struct xfs_trans *tp; 1128 int error; 1129 struct timespec64 now; 1130 1131 trace_xfs_update_time(ip); 1132 1133 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1134 if (!((flags & S_VERSION) && 1135 inode_maybe_inc_iversion(inode, false))) { 1136 generic_update_time(inode, flags); 1137 return 0; 1138 } 1139 1140 /* Capture the iversion update that just occurred */ 1141 log_flags |= XFS_ILOG_CORE; 1142 } 1143 1144 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1145 if (error) 1146 return error; 1147 1148 xfs_ilock(ip, XFS_ILOCK_EXCL); 1149 if (flags & (S_CTIME|S_MTIME)) 1150 now = inode_set_ctime_current(inode); 1151 else 1152 now = current_time(inode); 1153 1154 if (flags & S_MTIME) 1155 inode_set_mtime_to_ts(inode, now); 1156 if (flags & S_ATIME) 1157 inode_set_atime_to_ts(inode, now); 1158 1159 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1160 xfs_trans_log_inode(tp, ip, log_flags); 1161 return xfs_trans_commit(tp); 1162 } 1163 1164 STATIC int 1165 xfs_vn_fiemap( 1166 struct inode *inode, 1167 struct fiemap_extent_info *fieinfo, 1168 u64 start, 1169 u64 length) 1170 { 1171 int error; 1172 1173 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1174 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1175 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1176 error = iomap_fiemap(inode, fieinfo, start, length, 1177 &xfs_xattr_iomap_ops); 1178 } else { 1179 error = iomap_fiemap(inode, fieinfo, start, length, 1180 &xfs_read_iomap_ops); 1181 } 1182 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1183 1184 return error; 1185 } 1186 1187 STATIC int 1188 xfs_vn_tmpfile( 1189 struct mnt_idmap *idmap, 1190 struct inode *dir, 1191 struct file *file, 1192 umode_t mode) 1193 { 1194 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file); 1195 1196 return finish_open_simple(file, err); 1197 } 1198 1199 static const struct inode_operations xfs_inode_operations = { 1200 .get_inode_acl = xfs_get_acl, 1201 .set_acl = xfs_set_acl, 1202 .getattr = xfs_vn_getattr, 1203 .setattr = xfs_vn_setattr, 1204 .listxattr = xfs_vn_listxattr, 1205 .fiemap = xfs_vn_fiemap, 1206 .update_time = xfs_vn_update_time, 1207 .fileattr_get = xfs_fileattr_get, 1208 .fileattr_set = xfs_fileattr_set, 1209 }; 1210 1211 static const struct inode_operations xfs_dir_inode_operations = { 1212 .create = xfs_vn_create, 1213 .lookup = xfs_vn_lookup, 1214 .link = xfs_vn_link, 1215 .unlink = xfs_vn_unlink, 1216 .symlink = xfs_vn_symlink, 1217 .mkdir = xfs_vn_mkdir, 1218 /* 1219 * Yes, XFS uses the same method for rmdir and unlink. 1220 * 1221 * There are some subtile differences deeper in the code, 1222 * but we use S_ISDIR to check for those. 1223 */ 1224 .rmdir = xfs_vn_unlink, 1225 .mknod = xfs_vn_mknod, 1226 .rename = xfs_vn_rename, 1227 .get_inode_acl = xfs_get_acl, 1228 .set_acl = xfs_set_acl, 1229 .getattr = xfs_vn_getattr, 1230 .setattr = xfs_vn_setattr, 1231 .listxattr = xfs_vn_listxattr, 1232 .update_time = xfs_vn_update_time, 1233 .tmpfile = xfs_vn_tmpfile, 1234 .fileattr_get = xfs_fileattr_get, 1235 .fileattr_set = xfs_fileattr_set, 1236 }; 1237 1238 static const struct inode_operations xfs_dir_ci_inode_operations = { 1239 .create = xfs_vn_create, 1240 .lookup = xfs_vn_ci_lookup, 1241 .link = xfs_vn_link, 1242 .unlink = xfs_vn_unlink, 1243 .symlink = xfs_vn_symlink, 1244 .mkdir = xfs_vn_mkdir, 1245 /* 1246 * Yes, XFS uses the same method for rmdir and unlink. 1247 * 1248 * There are some subtile differences deeper in the code, 1249 * but we use S_ISDIR to check for those. 1250 */ 1251 .rmdir = xfs_vn_unlink, 1252 .mknod = xfs_vn_mknod, 1253 .rename = xfs_vn_rename, 1254 .get_inode_acl = xfs_get_acl, 1255 .set_acl = xfs_set_acl, 1256 .getattr = xfs_vn_getattr, 1257 .setattr = xfs_vn_setattr, 1258 .listxattr = xfs_vn_listxattr, 1259 .update_time = xfs_vn_update_time, 1260 .tmpfile = xfs_vn_tmpfile, 1261 .fileattr_get = xfs_fileattr_get, 1262 .fileattr_set = xfs_fileattr_set, 1263 }; 1264 1265 static const struct inode_operations xfs_symlink_inode_operations = { 1266 .get_link = xfs_vn_get_link, 1267 .getattr = xfs_vn_getattr, 1268 .setattr = xfs_vn_setattr, 1269 .listxattr = xfs_vn_listxattr, 1270 .update_time = xfs_vn_update_time, 1271 }; 1272 1273 /* Figure out if this file actually supports DAX. */ 1274 static bool 1275 xfs_inode_supports_dax( 1276 struct xfs_inode *ip) 1277 { 1278 struct xfs_mount *mp = ip->i_mount; 1279 1280 /* Only supported on regular files. */ 1281 if (!S_ISREG(VFS_I(ip)->i_mode)) 1282 return false; 1283 1284 /* Block size must match page size */ 1285 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1286 return false; 1287 1288 /* Device has to support DAX too. */ 1289 return xfs_inode_buftarg(ip)->bt_daxdev != NULL; 1290 } 1291 1292 static bool 1293 xfs_inode_should_enable_dax( 1294 struct xfs_inode *ip) 1295 { 1296 if (!IS_ENABLED(CONFIG_FS_DAX)) 1297 return false; 1298 if (xfs_has_dax_never(ip->i_mount)) 1299 return false; 1300 if (!xfs_inode_supports_dax(ip)) 1301 return false; 1302 if (xfs_has_dax_always(ip->i_mount)) 1303 return true; 1304 if (ip->i_diflags2 & XFS_DIFLAG2_DAX) 1305 return true; 1306 return false; 1307 } 1308 1309 void 1310 xfs_diflags_to_iflags( 1311 struct xfs_inode *ip, 1312 bool init) 1313 { 1314 struct inode *inode = VFS_I(ip); 1315 unsigned int xflags = xfs_ip2xflags(ip); 1316 unsigned int flags = 0; 1317 1318 ASSERT(!(IS_DAX(inode) && init)); 1319 1320 if (xflags & FS_XFLAG_IMMUTABLE) 1321 flags |= S_IMMUTABLE; 1322 if (xflags & FS_XFLAG_APPEND) 1323 flags |= S_APPEND; 1324 if (xflags & FS_XFLAG_SYNC) 1325 flags |= S_SYNC; 1326 if (xflags & FS_XFLAG_NOATIME) 1327 flags |= S_NOATIME; 1328 if (init && xfs_inode_should_enable_dax(ip)) 1329 flags |= S_DAX; 1330 1331 /* 1332 * S_DAX can only be set during inode initialization and is never set by 1333 * the VFS, so we cannot mask off S_DAX in i_flags. 1334 */ 1335 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); 1336 inode->i_flags |= flags; 1337 } 1338 1339 /* 1340 * Initialize the Linux inode. 1341 * 1342 * When reading existing inodes from disk this is called directly from xfs_iget, 1343 * when creating a new inode it is called from xfs_init_new_inode after setting 1344 * up the inode. These callers have different criteria for clearing XFS_INEW, so 1345 * leave it up to the caller to deal with unlocking the inode appropriately. 1346 */ 1347 void 1348 xfs_setup_inode( 1349 struct xfs_inode *ip) 1350 { 1351 struct inode *inode = &ip->i_vnode; 1352 gfp_t gfp_mask; 1353 bool is_meta = xfs_is_internal_inode(ip); 1354 1355 inode->i_ino = ip->i_ino; 1356 inode->i_state |= I_NEW; 1357 1358 inode_sb_list_add(inode); 1359 /* make the inode look hashed for the writeback code */ 1360 inode_fake_hash(inode); 1361 1362 i_size_write(inode, ip->i_disk_size); 1363 xfs_diflags_to_iflags(ip, true); 1364 1365 /* 1366 * Mark our metadata files as private so that LSMs and the ACL code 1367 * don't try to add their own metadata or reason about these files, 1368 * and users cannot ever obtain file handles to them. 1369 */ 1370 if (is_meta) { 1371 inode->i_flags |= S_PRIVATE; 1372 inode->i_opflags &= ~IOP_XATTR; 1373 } 1374 1375 if (S_ISDIR(inode->i_mode)) { 1376 /* 1377 * We set the i_rwsem class here to avoid potential races with 1378 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1379 * after a filehandle lookup has already found the inode in 1380 * cache before it has been unlocked via unlock_new_inode(). 1381 */ 1382 lockdep_set_class(&inode->i_rwsem, 1383 &inode->i_sb->s_type->i_mutex_dir_key); 1384 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class); 1385 } else { 1386 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class); 1387 } 1388 1389 /* 1390 * Ensure all page cache allocations are done from GFP_NOFS context to 1391 * prevent direct reclaim recursion back into the filesystem and blowing 1392 * stacks or deadlocking. 1393 */ 1394 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1395 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1396 1397 /* 1398 * For real-time inodes update the stable write flags to that of the RT 1399 * device instead of the data device. 1400 */ 1401 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip)) 1402 xfs_update_stable_writes(ip); 1403 1404 /* 1405 * If there is no attribute fork no ACL can exist on this inode, 1406 * and it can't have any file capabilities attached to it either. 1407 */ 1408 if (!xfs_inode_has_attr_fork(ip)) { 1409 inode_has_no_xattr(inode); 1410 cache_no_acl(inode); 1411 } 1412 } 1413 1414 void 1415 xfs_setup_iops( 1416 struct xfs_inode *ip) 1417 { 1418 struct inode *inode = &ip->i_vnode; 1419 1420 switch (inode->i_mode & S_IFMT) { 1421 case S_IFREG: 1422 inode->i_op = &xfs_inode_operations; 1423 inode->i_fop = &xfs_file_operations; 1424 if (IS_DAX(inode)) 1425 inode->i_mapping->a_ops = &xfs_dax_aops; 1426 else 1427 inode->i_mapping->a_ops = &xfs_address_space_operations; 1428 break; 1429 case S_IFDIR: 1430 if (xfs_has_asciici(XFS_M(inode->i_sb))) 1431 inode->i_op = &xfs_dir_ci_inode_operations; 1432 else 1433 inode->i_op = &xfs_dir_inode_operations; 1434 inode->i_fop = &xfs_dir_file_operations; 1435 break; 1436 case S_IFLNK: 1437 inode->i_op = &xfs_symlink_inode_operations; 1438 break; 1439 default: 1440 inode->i_op = &xfs_inode_operations; 1441 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1442 break; 1443 } 1444 } 1445