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 unsigned int 605 xfs_get_atomic_write_min( 606 struct xfs_inode *ip) 607 { 608 struct xfs_mount *mp = ip->i_mount; 609 610 /* 611 * If we can complete an atomic write via atomic out of place writes, 612 * then advertise a minimum size of one fsblock. Without this 613 * mechanism, we can only guarantee atomic writes up to a single LBA. 614 * 615 * If out of place writes are not available, we can guarantee an atomic 616 * write of exactly one single fsblock if the bdev will make that 617 * guarantee for us. 618 */ 619 if (xfs_inode_can_hw_atomic_write(ip) || 620 xfs_inode_can_sw_atomic_write(ip)) 621 return mp->m_sb.sb_blocksize; 622 623 return 0; 624 } 625 626 unsigned int 627 xfs_get_atomic_write_max( 628 struct xfs_inode *ip) 629 { 630 struct xfs_mount *mp = ip->i_mount; 631 632 /* 633 * If out of place writes are not available, we can guarantee an atomic 634 * write of exactly one single fsblock if the bdev will make that 635 * guarantee for us. 636 */ 637 if (!xfs_inode_can_sw_atomic_write(ip)) { 638 if (xfs_inode_can_hw_atomic_write(ip)) 639 return mp->m_sb.sb_blocksize; 640 return 0; 641 } 642 643 /* 644 * If we can complete an atomic write via atomic out of place writes, 645 * then advertise a maximum size of whatever we can complete through 646 * that means. Hardware support is reported via max_opt, not here. 647 */ 648 if (XFS_IS_REALTIME_INODE(ip)) 649 return XFS_FSB_TO_B(mp, mp->m_groups[XG_TYPE_RTG].awu_max); 650 return XFS_FSB_TO_B(mp, mp->m_groups[XG_TYPE_AG].awu_max); 651 } 652 653 unsigned int 654 xfs_get_atomic_write_max_opt( 655 struct xfs_inode *ip) 656 { 657 unsigned int awu_max = xfs_get_atomic_write_max(ip); 658 659 /* if the max is 1x block, then just keep behaviour that opt is 0 */ 660 if (awu_max <= ip->i_mount->m_sb.sb_blocksize) 661 return 0; 662 663 /* 664 * Advertise the maximum size of an atomic write that we can tell the 665 * block device to perform for us. In general the bdev limit will be 666 * less than our out of place write limit, but we don't want to exceed 667 * the awu_max. 668 */ 669 return min(awu_max, xfs_inode_buftarg(ip)->bt_awu_max); 670 } 671 672 static void 673 xfs_report_atomic_write( 674 struct xfs_inode *ip, 675 struct kstat *stat) 676 { 677 generic_fill_statx_atomic_writes(stat, 678 xfs_get_atomic_write_min(ip), 679 xfs_get_atomic_write_max(ip), 680 xfs_get_atomic_write_max_opt(ip)); 681 } 682 683 STATIC int 684 xfs_vn_getattr( 685 struct mnt_idmap *idmap, 686 const struct path *path, 687 struct kstat *stat, 688 u32 request_mask, 689 unsigned int query_flags) 690 { 691 struct inode *inode = d_inode(path->dentry); 692 struct xfs_inode *ip = XFS_I(inode); 693 struct xfs_mount *mp = ip->i_mount; 694 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode); 695 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); 696 697 trace_xfs_getattr(ip); 698 699 if (xfs_is_shutdown(mp)) 700 return -EIO; 701 702 stat->size = XFS_ISIZE(ip); 703 stat->dev = inode->i_sb->s_dev; 704 stat->mode = inode->i_mode; 705 stat->nlink = inode->i_nlink; 706 stat->uid = vfsuid_into_kuid(vfsuid); 707 stat->gid = vfsgid_into_kgid(vfsgid); 708 stat->ino = ip->i_ino; 709 stat->atime = inode_get_atime(inode); 710 711 fill_mg_cmtime(stat, request_mask, inode); 712 713 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); 714 715 if (xfs_has_v3inodes(mp)) { 716 if (request_mask & STATX_BTIME) { 717 stat->result_mask |= STATX_BTIME; 718 stat->btime = ip->i_crtime; 719 } 720 } 721 722 /* 723 * Note: If you add another clause to set an attribute flag, please 724 * update attributes_mask below. 725 */ 726 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) 727 stat->attributes |= STATX_ATTR_IMMUTABLE; 728 if (ip->i_diflags & XFS_DIFLAG_APPEND) 729 stat->attributes |= STATX_ATTR_APPEND; 730 if (ip->i_diflags & XFS_DIFLAG_NODUMP) 731 stat->attributes |= STATX_ATTR_NODUMP; 732 733 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | 734 STATX_ATTR_APPEND | 735 STATX_ATTR_NODUMP); 736 737 switch (inode->i_mode & S_IFMT) { 738 case S_IFBLK: 739 case S_IFCHR: 740 stat->blksize = BLKDEV_IOSIZE; 741 stat->rdev = inode->i_rdev; 742 break; 743 case S_IFREG: 744 if (request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN)) 745 xfs_report_dioalign(ip, stat); 746 if (request_mask & STATX_WRITE_ATOMIC) 747 xfs_report_atomic_write(ip, stat); 748 fallthrough; 749 default: 750 stat->blksize = xfs_stat_blksize(ip); 751 stat->rdev = 0; 752 break; 753 } 754 755 return 0; 756 } 757 758 static int 759 xfs_vn_change_ok( 760 struct mnt_idmap *idmap, 761 struct dentry *dentry, 762 struct iattr *iattr) 763 { 764 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 765 766 if (xfs_is_readonly(mp)) 767 return -EROFS; 768 769 if (xfs_is_shutdown(mp)) 770 return -EIO; 771 772 return setattr_prepare(idmap, dentry, iattr); 773 } 774 775 /* 776 * Set non-size attributes of an inode. 777 * 778 * Caution: The caller of this function is responsible for calling 779 * setattr_prepare() or otherwise verifying the change is fine. 780 */ 781 static int 782 xfs_setattr_nonsize( 783 struct mnt_idmap *idmap, 784 struct dentry *dentry, 785 struct xfs_inode *ip, 786 struct iattr *iattr) 787 { 788 xfs_mount_t *mp = ip->i_mount; 789 struct inode *inode = VFS_I(ip); 790 int mask = iattr->ia_valid; 791 xfs_trans_t *tp; 792 int error; 793 kuid_t uid = GLOBAL_ROOT_UID; 794 kgid_t gid = GLOBAL_ROOT_GID; 795 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 796 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; 797 798 ASSERT((mask & ATTR_SIZE) == 0); 799 800 /* 801 * If disk quotas is on, we make sure that the dquots do exist on disk, 802 * before we start any other transactions. Trying to do this later 803 * is messy. We don't care to take a readlock to look at the ids 804 * in inode here, because we can't hold it across the trans_reserve. 805 * If the IDs do change before we take the ilock, we're covered 806 * because the i_*dquot fields will get updated anyway. 807 */ 808 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 809 uint qflags = 0; 810 811 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 812 uid = from_vfsuid(idmap, i_user_ns(inode), 813 iattr->ia_vfsuid); 814 qflags |= XFS_QMOPT_UQUOTA; 815 } else { 816 uid = inode->i_uid; 817 } 818 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 819 gid = from_vfsgid(idmap, i_user_ns(inode), 820 iattr->ia_vfsgid); 821 qflags |= XFS_QMOPT_GQUOTA; 822 } else { 823 gid = inode->i_gid; 824 } 825 826 /* 827 * We take a reference when we initialize udqp and gdqp, 828 * so it is important that we never blindly double trip on 829 * the same variable. See xfs_create() for an example. 830 */ 831 ASSERT(udqp == NULL); 832 ASSERT(gdqp == NULL); 833 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, 834 qflags, &udqp, &gdqp, NULL); 835 if (error) 836 return error; 837 } 838 839 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, 840 has_capability_noaudit(current, CAP_FOWNER), &tp); 841 if (error) 842 goto out_dqrele; 843 844 /* 845 * Register quota modifications in the transaction. Must be the owner 846 * or privileged. These IDs could have changed since we last looked at 847 * them. But, we're assured that if the ownership did change while we 848 * didn't have the inode locked, inode's dquot(s) would have changed 849 * also. 850 */ 851 if (XFS_IS_UQUOTA_ON(mp) && 852 i_uid_needs_update(idmap, iattr, inode)) { 853 ASSERT(udqp); 854 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); 855 } 856 if (XFS_IS_GQUOTA_ON(mp) && 857 i_gid_needs_update(idmap, iattr, inode)) { 858 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp)); 859 ASSERT(gdqp); 860 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp); 861 } 862 863 setattr_copy(idmap, inode, iattr); 864 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 865 866 XFS_STATS_INC(mp, xs_ig_attrchg); 867 868 if (xfs_has_wsync(mp)) 869 xfs_trans_set_sync(tp); 870 error = xfs_trans_commit(tp); 871 872 /* 873 * Release any dquot(s) the inode had kept before chown. 874 */ 875 xfs_qm_dqrele(old_udqp); 876 xfs_qm_dqrele(old_gdqp); 877 xfs_qm_dqrele(udqp); 878 xfs_qm_dqrele(gdqp); 879 880 if (error) 881 return error; 882 883 /* 884 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 885 * update. We could avoid this with linked transactions 886 * and passing down the transaction pointer all the way 887 * to attr_set. No previous user of the generic 888 * Posix ACL code seems to care about this issue either. 889 */ 890 if (mask & ATTR_MODE) { 891 error = posix_acl_chmod(idmap, dentry, inode->i_mode); 892 if (error) 893 return error; 894 } 895 896 return 0; 897 898 out_dqrele: 899 xfs_qm_dqrele(udqp); 900 xfs_qm_dqrele(gdqp); 901 return error; 902 } 903 904 /* 905 * Truncate file. Must have write permission and not be a directory. 906 * 907 * Caution: The caller of this function is responsible for calling 908 * setattr_prepare() or otherwise verifying the change is fine. 909 */ 910 STATIC int 911 xfs_setattr_size( 912 struct mnt_idmap *idmap, 913 struct dentry *dentry, 914 struct xfs_inode *ip, 915 struct iattr *iattr) 916 { 917 struct xfs_mount *mp = ip->i_mount; 918 struct inode *inode = VFS_I(ip); 919 xfs_off_t oldsize, newsize; 920 struct xfs_trans *tp; 921 int error; 922 uint lock_flags = 0; 923 uint resblks = 0; 924 bool did_zeroing = false; 925 struct xfs_zone_alloc_ctx ac = { }; 926 927 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL); 928 ASSERT(S_ISREG(inode->i_mode)); 929 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 930 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); 931 932 oldsize = inode->i_size; 933 newsize = iattr->ia_size; 934 935 /* 936 * Short circuit the truncate case for zero length files. 937 */ 938 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { 939 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 940 return 0; 941 942 /* 943 * Use the regular setattr path to update the timestamps. 944 */ 945 iattr->ia_valid &= ~ATTR_SIZE; 946 return xfs_setattr_nonsize(idmap, dentry, ip, iattr); 947 } 948 949 /* 950 * Make sure that the dquots are attached to the inode. 951 */ 952 error = xfs_qm_dqattach(ip); 953 if (error) 954 return error; 955 956 /* 957 * Wait for all direct I/O to complete. 958 */ 959 inode_dio_wait(inode); 960 961 /* 962 * Normally xfs_zoned_space_reserve is supposed to be called outside the 963 * IOLOCK. For truncate we can't do that since ->setattr is called with 964 * it already held by the VFS. So for now chicken out and try to 965 * allocate space under it. 966 * 967 * To avoid deadlocks this means we can't block waiting for space, which 968 * can lead to spurious -ENOSPC if there are no directly available 969 * blocks. We mitigate this a bit by allowing zeroing to dip into the 970 * reserved pool, but eventually the VFS calling convention needs to 971 * change. 972 */ 973 if (xfs_is_zoned_inode(ip)) { 974 error = xfs_zoned_space_reserve(mp, 1, 975 XFS_ZR_NOWAIT | XFS_ZR_RESERVED, &ac); 976 if (error) { 977 if (error == -EAGAIN) 978 return -ENOSPC; 979 return error; 980 } 981 } 982 983 /* 984 * File data changes must be complete before we start the transaction to 985 * modify the inode. This needs to be done before joining the inode to 986 * the transaction because the inode cannot be unlocked once it is a 987 * part of the transaction. 988 * 989 * Start with zeroing any data beyond EOF that we may expose on file 990 * extension, or zeroing out the rest of the block on a downward 991 * truncate. 992 */ 993 if (newsize > oldsize) { 994 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 995 error = xfs_zero_range(ip, oldsize, newsize - oldsize, 996 &ac, &did_zeroing); 997 } else { 998 error = xfs_truncate_page(ip, newsize, &ac, &did_zeroing); 999 } 1000 1001 if (xfs_is_zoned_inode(ip)) 1002 xfs_zoned_space_unreserve(mp, &ac); 1003 1004 if (error) 1005 return error; 1006 1007 /* 1008 * We've already locked out new page faults, so now we can safely remove 1009 * pages from the page cache knowing they won't get refaulted until we 1010 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 1011 * complete. The truncate_setsize() call also cleans partial EOF page 1012 * PTEs on extending truncates and hence ensures sub-page block size 1013 * filesystems are correctly handled, too. 1014 * 1015 * We have to do all the page cache truncate work outside the 1016 * transaction context as the "lock" order is page lock->log space 1017 * reservation as defined by extent allocation in the writeback path. 1018 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 1019 * having already truncated the in-memory version of the file (i.e. made 1020 * user visible changes). There's not much we can do about this, except 1021 * to hope that the caller sees ENOMEM and retries the truncate 1022 * operation. 1023 * 1024 * And we update in-core i_size and truncate page cache beyond newsize 1025 * before writeback the [i_disk_size, newsize] range, so we're 1026 * guaranteed not to write stale data past the new EOF on truncate down. 1027 */ 1028 truncate_setsize(inode, newsize); 1029 1030 /* 1031 * We are going to log the inode size change in this transaction so 1032 * any previous writes that are beyond the on disk EOF and the new 1033 * EOF that have not been written out need to be written here. If we 1034 * do not write the data out, we expose ourselves to the null files 1035 * problem. Note that this includes any block zeroing we did above; 1036 * otherwise those blocks may not be zeroed after a crash. 1037 */ 1038 if (did_zeroing || 1039 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { 1040 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 1041 ip->i_disk_size, newsize - 1); 1042 if (error) 1043 return error; 1044 } 1045 1046 /* 1047 * For realtime inode with more than one block rtextsize, we need the 1048 * block reservation for bmap btree block allocations/splits that can 1049 * happen since it could split the tail written extent and convert the 1050 * right beyond EOF one to unwritten. 1051 */ 1052 if (xfs_inode_has_bigrtalloc(ip)) 1053 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 1054 1055 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks, 1056 0, 0, &tp); 1057 if (error) 1058 return error; 1059 1060 lock_flags |= XFS_ILOCK_EXCL; 1061 xfs_ilock(ip, XFS_ILOCK_EXCL); 1062 xfs_trans_ijoin(tp, ip, 0); 1063 1064 /* 1065 * Only change the c/mtime if we are changing the size or we are 1066 * explicitly asked to change it. This handles the semantic difference 1067 * between truncate() and ftruncate() as implemented in the VFS. 1068 * 1069 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 1070 * special case where we need to update the times despite not having 1071 * these flags set. For all other operations the VFS set these flags 1072 * explicitly if it wants a timestamp update. 1073 */ 1074 if (newsize != oldsize && 1075 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 1076 iattr->ia_ctime = iattr->ia_mtime = 1077 current_time(inode); 1078 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 1079 } 1080 1081 /* 1082 * The first thing we do is set the size to new_size permanently on 1083 * disk. This way we don't have to worry about anyone ever being able 1084 * to look at the data being freed even in the face of a crash. 1085 * What we're getting around here is the case where we free a block, it 1086 * is allocated to another file, it is written to, and then we crash. 1087 * If the new data gets written to the file but the log buffers 1088 * containing the free and reallocation don't, then we'd end up with 1089 * garbage in the blocks being freed. As long as we make the new size 1090 * permanent before actually freeing any blocks it doesn't matter if 1091 * they get written to. 1092 */ 1093 ip->i_disk_size = newsize; 1094 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1095 1096 if (newsize <= oldsize) { 1097 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 1098 if (error) 1099 goto out_trans_cancel; 1100 1101 /* 1102 * Truncated "down", so we're removing references to old data 1103 * here - if we delay flushing for a long time, we expose 1104 * ourselves unduly to the notorious NULL files problem. So, 1105 * we mark this inode and flush it when the file is closed, 1106 * and do not wait the usual (long) time for writeout. 1107 */ 1108 xfs_iflags_set(ip, XFS_ITRUNCATED); 1109 1110 /* A truncate down always removes post-EOF blocks. */ 1111 xfs_inode_clear_eofblocks_tag(ip); 1112 } 1113 1114 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID))); 1115 setattr_copy(idmap, inode, iattr); 1116 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1117 1118 XFS_STATS_INC(mp, xs_ig_attrchg); 1119 1120 if (xfs_has_wsync(mp)) 1121 xfs_trans_set_sync(tp); 1122 1123 error = xfs_trans_commit(tp); 1124 out_unlock: 1125 if (lock_flags) 1126 xfs_iunlock(ip, lock_flags); 1127 return error; 1128 1129 out_trans_cancel: 1130 xfs_trans_cancel(tp); 1131 goto out_unlock; 1132 } 1133 1134 int 1135 xfs_vn_setattr_size( 1136 struct mnt_idmap *idmap, 1137 struct dentry *dentry, 1138 struct iattr *iattr) 1139 { 1140 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 1141 int error; 1142 1143 trace_xfs_setattr(ip); 1144 1145 error = xfs_vn_change_ok(idmap, dentry, iattr); 1146 if (error) 1147 return error; 1148 return xfs_setattr_size(idmap, dentry, ip, iattr); 1149 } 1150 1151 STATIC int 1152 xfs_vn_setattr( 1153 struct mnt_idmap *idmap, 1154 struct dentry *dentry, 1155 struct iattr *iattr) 1156 { 1157 struct inode *inode = d_inode(dentry); 1158 struct xfs_inode *ip = XFS_I(inode); 1159 int error; 1160 1161 if (iattr->ia_valid & ATTR_SIZE) { 1162 uint iolock; 1163 1164 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1165 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1166 1167 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1168 if (error) { 1169 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1170 return error; 1171 } 1172 1173 error = xfs_vn_setattr_size(idmap, dentry, iattr); 1174 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1175 } else { 1176 trace_xfs_setattr(ip); 1177 1178 error = xfs_vn_change_ok(idmap, dentry, iattr); 1179 if (!error) 1180 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr); 1181 } 1182 1183 return error; 1184 } 1185 1186 STATIC int 1187 xfs_vn_update_time( 1188 struct inode *inode, 1189 int flags) 1190 { 1191 struct xfs_inode *ip = XFS_I(inode); 1192 struct xfs_mount *mp = ip->i_mount; 1193 int log_flags = XFS_ILOG_TIMESTAMP; 1194 struct xfs_trans *tp; 1195 int error; 1196 struct timespec64 now; 1197 1198 trace_xfs_update_time(ip); 1199 1200 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1201 if (!((flags & S_VERSION) && 1202 inode_maybe_inc_iversion(inode, false))) { 1203 generic_update_time(inode, flags); 1204 return 0; 1205 } 1206 1207 /* Capture the iversion update that just occurred */ 1208 log_flags |= XFS_ILOG_CORE; 1209 } 1210 1211 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1212 if (error) 1213 return error; 1214 1215 xfs_ilock(ip, XFS_ILOCK_EXCL); 1216 if (flags & (S_CTIME|S_MTIME)) 1217 now = inode_set_ctime_current(inode); 1218 else 1219 now = current_time(inode); 1220 1221 if (flags & S_MTIME) 1222 inode_set_mtime_to_ts(inode, now); 1223 if (flags & S_ATIME) 1224 inode_set_atime_to_ts(inode, now); 1225 1226 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1227 xfs_trans_log_inode(tp, ip, log_flags); 1228 return xfs_trans_commit(tp); 1229 } 1230 1231 STATIC int 1232 xfs_vn_fiemap( 1233 struct inode *inode, 1234 struct fiemap_extent_info *fieinfo, 1235 u64 start, 1236 u64 length) 1237 { 1238 int error; 1239 1240 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1241 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1242 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1243 error = iomap_fiemap(inode, fieinfo, start, length, 1244 &xfs_xattr_iomap_ops); 1245 } else { 1246 error = iomap_fiemap(inode, fieinfo, start, length, 1247 &xfs_read_iomap_ops); 1248 } 1249 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1250 1251 return error; 1252 } 1253 1254 STATIC int 1255 xfs_vn_tmpfile( 1256 struct mnt_idmap *idmap, 1257 struct inode *dir, 1258 struct file *file, 1259 umode_t mode) 1260 { 1261 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file); 1262 1263 return finish_open_simple(file, err); 1264 } 1265 1266 static const struct inode_operations xfs_inode_operations = { 1267 .get_inode_acl = xfs_get_acl, 1268 .set_acl = xfs_set_acl, 1269 .getattr = xfs_vn_getattr, 1270 .setattr = xfs_vn_setattr, 1271 .listxattr = xfs_vn_listxattr, 1272 .fiemap = xfs_vn_fiemap, 1273 .update_time = xfs_vn_update_time, 1274 .fileattr_get = xfs_fileattr_get, 1275 .fileattr_set = xfs_fileattr_set, 1276 }; 1277 1278 static const struct inode_operations xfs_dir_inode_operations = { 1279 .create = xfs_vn_create, 1280 .lookup = xfs_vn_lookup, 1281 .link = xfs_vn_link, 1282 .unlink = xfs_vn_unlink, 1283 .symlink = xfs_vn_symlink, 1284 .mkdir = xfs_vn_mkdir, 1285 /* 1286 * Yes, XFS uses the same method for rmdir and unlink. 1287 * 1288 * There are some subtile differences deeper in the code, 1289 * but we use S_ISDIR to check for those. 1290 */ 1291 .rmdir = xfs_vn_unlink, 1292 .mknod = xfs_vn_mknod, 1293 .rename = xfs_vn_rename, 1294 .get_inode_acl = xfs_get_acl, 1295 .set_acl = xfs_set_acl, 1296 .getattr = xfs_vn_getattr, 1297 .setattr = xfs_vn_setattr, 1298 .listxattr = xfs_vn_listxattr, 1299 .update_time = xfs_vn_update_time, 1300 .tmpfile = xfs_vn_tmpfile, 1301 .fileattr_get = xfs_fileattr_get, 1302 .fileattr_set = xfs_fileattr_set, 1303 }; 1304 1305 static const struct inode_operations xfs_dir_ci_inode_operations = { 1306 .create = xfs_vn_create, 1307 .lookup = xfs_vn_ci_lookup, 1308 .link = xfs_vn_link, 1309 .unlink = xfs_vn_unlink, 1310 .symlink = xfs_vn_symlink, 1311 .mkdir = xfs_vn_mkdir, 1312 /* 1313 * Yes, XFS uses the same method for rmdir and unlink. 1314 * 1315 * There are some subtile differences deeper in the code, 1316 * but we use S_ISDIR to check for those. 1317 */ 1318 .rmdir = xfs_vn_unlink, 1319 .mknod = xfs_vn_mknod, 1320 .rename = xfs_vn_rename, 1321 .get_inode_acl = xfs_get_acl, 1322 .set_acl = xfs_set_acl, 1323 .getattr = xfs_vn_getattr, 1324 .setattr = xfs_vn_setattr, 1325 .listxattr = xfs_vn_listxattr, 1326 .update_time = xfs_vn_update_time, 1327 .tmpfile = xfs_vn_tmpfile, 1328 .fileattr_get = xfs_fileattr_get, 1329 .fileattr_set = xfs_fileattr_set, 1330 }; 1331 1332 static const struct inode_operations xfs_symlink_inode_operations = { 1333 .get_link = xfs_vn_get_link, 1334 .getattr = xfs_vn_getattr, 1335 .setattr = xfs_vn_setattr, 1336 .listxattr = xfs_vn_listxattr, 1337 .update_time = xfs_vn_update_time, 1338 }; 1339 1340 /* Figure out if this file actually supports DAX. */ 1341 static bool 1342 xfs_inode_supports_dax( 1343 struct xfs_inode *ip) 1344 { 1345 struct xfs_mount *mp = ip->i_mount; 1346 1347 /* Only supported on regular files. */ 1348 if (!S_ISREG(VFS_I(ip)->i_mode)) 1349 return false; 1350 1351 /* Block size must match page size */ 1352 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1353 return false; 1354 1355 /* Device has to support DAX too. */ 1356 return xfs_inode_buftarg(ip)->bt_daxdev != NULL; 1357 } 1358 1359 static bool 1360 xfs_inode_should_enable_dax( 1361 struct xfs_inode *ip) 1362 { 1363 if (!IS_ENABLED(CONFIG_FS_DAX)) 1364 return false; 1365 if (xfs_has_dax_never(ip->i_mount)) 1366 return false; 1367 if (!xfs_inode_supports_dax(ip)) 1368 return false; 1369 if (xfs_has_dax_always(ip->i_mount)) 1370 return true; 1371 if (ip->i_diflags2 & XFS_DIFLAG2_DAX) 1372 return true; 1373 return false; 1374 } 1375 1376 void 1377 xfs_diflags_to_iflags( 1378 struct xfs_inode *ip, 1379 bool init) 1380 { 1381 struct inode *inode = VFS_I(ip); 1382 unsigned int xflags = xfs_ip2xflags(ip); 1383 unsigned int flags = 0; 1384 1385 ASSERT(!(IS_DAX(inode) && init)); 1386 1387 if (xflags & FS_XFLAG_IMMUTABLE) 1388 flags |= S_IMMUTABLE; 1389 if (xflags & FS_XFLAG_APPEND) 1390 flags |= S_APPEND; 1391 if (xflags & FS_XFLAG_SYNC) 1392 flags |= S_SYNC; 1393 if (xflags & FS_XFLAG_NOATIME) 1394 flags |= S_NOATIME; 1395 if (init && xfs_inode_should_enable_dax(ip)) 1396 flags |= S_DAX; 1397 1398 /* 1399 * S_DAX can only be set during inode initialization and is never set by 1400 * the VFS, so we cannot mask off S_DAX in i_flags. 1401 */ 1402 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); 1403 inode->i_flags |= flags; 1404 } 1405 1406 /* 1407 * Initialize the Linux inode. 1408 * 1409 * When reading existing inodes from disk this is called directly from xfs_iget, 1410 * when creating a new inode it is called from xfs_init_new_inode after setting 1411 * up the inode. These callers have different criteria for clearing XFS_INEW, so 1412 * leave it up to the caller to deal with unlocking the inode appropriately. 1413 */ 1414 void 1415 xfs_setup_inode( 1416 struct xfs_inode *ip) 1417 { 1418 struct inode *inode = &ip->i_vnode; 1419 gfp_t gfp_mask; 1420 bool is_meta = xfs_is_internal_inode(ip); 1421 1422 inode->i_ino = ip->i_ino; 1423 inode->i_state |= I_NEW; 1424 1425 inode_sb_list_add(inode); 1426 /* make the inode look hashed for the writeback code */ 1427 inode_fake_hash(inode); 1428 1429 i_size_write(inode, ip->i_disk_size); 1430 xfs_diflags_to_iflags(ip, true); 1431 1432 /* 1433 * Mark our metadata files as private so that LSMs and the ACL code 1434 * don't try to add their own metadata or reason about these files, 1435 * and users cannot ever obtain file handles to them. 1436 */ 1437 if (is_meta) { 1438 inode->i_flags |= S_PRIVATE; 1439 inode->i_opflags &= ~IOP_XATTR; 1440 } 1441 1442 if (S_ISDIR(inode->i_mode)) { 1443 /* 1444 * We set the i_rwsem class here to avoid potential races with 1445 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1446 * after a filehandle lookup has already found the inode in 1447 * cache before it has been unlocked via unlock_new_inode(). 1448 */ 1449 lockdep_set_class(&inode->i_rwsem, 1450 &inode->i_sb->s_type->i_mutex_dir_key); 1451 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class); 1452 } else { 1453 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class); 1454 } 1455 1456 /* 1457 * Ensure all page cache allocations are done from GFP_NOFS context to 1458 * prevent direct reclaim recursion back into the filesystem and blowing 1459 * stacks or deadlocking. 1460 */ 1461 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1462 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1463 1464 /* 1465 * For real-time inodes update the stable write flags to that of the RT 1466 * device instead of the data device. 1467 */ 1468 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip)) 1469 xfs_update_stable_writes(ip); 1470 1471 /* 1472 * If there is no attribute fork no ACL can exist on this inode, 1473 * and it can't have any file capabilities attached to it either. 1474 */ 1475 if (!xfs_inode_has_attr_fork(ip)) { 1476 inode_has_no_xattr(inode); 1477 cache_no_acl(inode); 1478 } 1479 } 1480 1481 void 1482 xfs_setup_iops( 1483 struct xfs_inode *ip) 1484 { 1485 struct inode *inode = &ip->i_vnode; 1486 1487 switch (inode->i_mode & S_IFMT) { 1488 case S_IFREG: 1489 inode->i_op = &xfs_inode_operations; 1490 inode->i_fop = &xfs_file_operations; 1491 if (IS_DAX(inode)) 1492 inode->i_mapping->a_ops = &xfs_dax_aops; 1493 else 1494 inode->i_mapping->a_ops = &xfs_address_space_operations; 1495 break; 1496 case S_IFDIR: 1497 if (xfs_has_asciici(XFS_M(inode->i_sb))) 1498 inode->i_op = &xfs_dir_ci_inode_operations; 1499 else 1500 inode->i_op = &xfs_dir_inode_operations; 1501 inode->i_fop = &xfs_dir_file_operations; 1502 break; 1503 case S_IFLNK: 1504 inode->i_op = &xfs_symlink_inode_operations; 1505 break; 1506 default: 1507 inode->i_op = &xfs_inode_operations; 1508 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1509 break; 1510 } 1511 } 1512