1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 4 * All Rights Reserved. 5 */ 6 #include "xfs_platform.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 = S_IFLNK | S_IRWXUGO; 439 440 error = xfs_dentry_mode_to_name(&name, dentry, mode); 441 if (unlikely(error)) 442 goto out; 443 444 error = xfs_symlink(idmap, XFS_I(dir), &name, symname, mode, &cip); 445 if (unlikely(error)) 446 goto out; 447 448 inode = VFS_I(cip); 449 450 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 451 if (unlikely(error)) 452 goto out_cleanup_inode; 453 454 xfs_setup_iops(cip); 455 456 d_instantiate(dentry, inode); 457 xfs_finish_inode_setup(cip); 458 return 0; 459 460 out_cleanup_inode: 461 xfs_finish_inode_setup(cip); 462 xfs_cleanup_inode(dir, inode, dentry); 463 xfs_irele(cip); 464 out: 465 return error; 466 } 467 468 STATIC int 469 xfs_vn_rename( 470 struct mnt_idmap *idmap, 471 struct inode *odir, 472 struct dentry *odentry, 473 struct inode *ndir, 474 struct dentry *ndentry, 475 unsigned int flags) 476 { 477 struct inode *new_inode = d_inode(ndentry); 478 int omode = 0; 479 int error; 480 struct xfs_name oname; 481 struct xfs_name nname; 482 483 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 484 return -EINVAL; 485 486 /* if we are exchanging files, we need to set i_mode of both files */ 487 if (flags & RENAME_EXCHANGE) 488 omode = d_inode(ndentry)->i_mode; 489 490 error = xfs_dentry_mode_to_name(&oname, odentry, omode); 491 if (omode && unlikely(error)) 492 return error; 493 494 error = xfs_dentry_mode_to_name(&nname, ndentry, 495 d_inode(odentry)->i_mode); 496 if (unlikely(error)) 497 return error; 498 499 return xfs_rename(idmap, XFS_I(odir), &oname, 500 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname, 501 new_inode ? XFS_I(new_inode) : NULL, flags); 502 } 503 504 /* 505 * careful here - this function can get called recursively, so 506 * we need to be very careful about how much stack we use. 507 * uio is kmalloced for this reason... 508 */ 509 STATIC const char * 510 xfs_vn_get_link( 511 struct dentry *dentry, 512 struct inode *inode, 513 struct delayed_call *done) 514 { 515 char *link; 516 int error = -ENOMEM; 517 518 if (!dentry) 519 return ERR_PTR(-ECHILD); 520 521 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); 522 if (!link) 523 goto out_err; 524 525 error = xfs_readlink(XFS_I(d_inode(dentry)), link); 526 if (unlikely(error)) 527 goto out_kfree; 528 529 set_delayed_call(done, kfree_link, link); 530 return link; 531 532 out_kfree: 533 kfree(link); 534 out_err: 535 return ERR_PTR(error); 536 } 537 538 static uint32_t 539 xfs_stat_blksize( 540 struct xfs_inode *ip) 541 { 542 struct xfs_mount *mp = ip->i_mount; 543 544 /* 545 * If the file blocks are being allocated from a realtime volume, then 546 * always return the realtime extent size. 547 */ 548 if (XFS_IS_REALTIME_INODE(ip)) 549 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip) ? : 1); 550 551 /* 552 * Allow large block sizes to be reported to userspace programs if the 553 * "largeio" mount option is used. 554 * 555 * If compatibility mode is specified, simply return the basic unit of 556 * caching so that we don't get inefficient read/modify/write I/O from 557 * user apps. Otherwise.... 558 * 559 * If the underlying volume is a stripe, then return the stripe width in 560 * bytes as the recommended I/O size. It is not a stripe and we've set a 561 * default buffered I/O size, return that, otherwise return the compat 562 * default. 563 */ 564 if (xfs_has_large_iosize(mp)) { 565 if (mp->m_swidth) 566 return XFS_FSB_TO_B(mp, mp->m_swidth); 567 if (xfs_has_allocsize(mp)) 568 return 1U << mp->m_allocsize_log; 569 } 570 571 return max_t(uint32_t, PAGE_SIZE, mp->m_sb.sb_blocksize); 572 } 573 574 static void 575 xfs_report_dioalign( 576 struct xfs_inode *ip, 577 struct kstat *stat) 578 { 579 struct xfs_buftarg *target = xfs_inode_buftarg(ip); 580 struct block_device *bdev = target->bt_bdev; 581 582 stat->result_mask |= STATX_DIOALIGN | STATX_DIO_READ_ALIGN; 583 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; 584 585 /* 586 * For COW inodes, we can only perform out of place writes of entire 587 * allocation units (blocks or RT extents). 588 * For writes smaller than the allocation unit, we must fall back to 589 * buffered I/O to perform read-modify-write cycles. At best this is 590 * highly inefficient; at worst it leads to page cache invalidation 591 * races. Tell applications to avoid this by reporting the larger write 592 * alignment in dio_offset_align, and the smaller read alignment in 593 * dio_read_offset_align. 594 */ 595 stat->dio_read_offset_align = bdev_logical_block_size(bdev); 596 if (xfs_is_cow_inode(ip)) 597 stat->dio_offset_align = xfs_inode_alloc_unitsize(ip); 598 else 599 stat->dio_offset_align = stat->dio_read_offset_align; 600 } 601 602 unsigned int 603 xfs_get_atomic_write_min( 604 struct xfs_inode *ip) 605 { 606 struct xfs_mount *mp = ip->i_mount; 607 608 /* 609 * If we can complete an atomic write via atomic out of place writes, 610 * then advertise a minimum size of one fsblock. Without this 611 * mechanism, we can only guarantee atomic writes up to a single LBA. 612 * 613 * If out of place writes are not available, we can guarantee an atomic 614 * write of exactly one single fsblock if the bdev will make that 615 * guarantee for us. 616 */ 617 if (xfs_inode_can_hw_atomic_write(ip) || 618 xfs_inode_can_sw_atomic_write(ip)) 619 return mp->m_sb.sb_blocksize; 620 621 return 0; 622 } 623 624 unsigned int 625 xfs_get_atomic_write_max( 626 struct xfs_inode *ip) 627 { 628 struct xfs_mount *mp = ip->i_mount; 629 630 /* 631 * If out of place writes are not available, we can guarantee an atomic 632 * write of exactly one single fsblock if the bdev will make that 633 * guarantee for us. 634 */ 635 if (!xfs_inode_can_sw_atomic_write(ip)) { 636 if (xfs_inode_can_hw_atomic_write(ip)) 637 return mp->m_sb.sb_blocksize; 638 return 0; 639 } 640 641 /* 642 * If we can complete an atomic write via atomic out of place writes, 643 * then advertise a maximum size of whatever we can complete through 644 * that means. Hardware support is reported via max_opt, not here. 645 */ 646 if (XFS_IS_REALTIME_INODE(ip)) 647 return XFS_FSB_TO_B(mp, mp->m_groups[XG_TYPE_RTG].awu_max); 648 return XFS_FSB_TO_B(mp, mp->m_groups[XG_TYPE_AG].awu_max); 649 } 650 651 unsigned int 652 xfs_get_atomic_write_max_opt( 653 struct xfs_inode *ip) 654 { 655 unsigned int awu_max = xfs_get_atomic_write_max(ip); 656 657 /* if the max is 1x block, then just keep behaviour that opt is 0 */ 658 if (awu_max <= ip->i_mount->m_sb.sb_blocksize) 659 return 0; 660 661 /* 662 * Advertise the maximum size of an atomic write that we can tell the 663 * block device to perform for us. In general the bdev limit will be 664 * less than our out of place write limit, but we don't want to exceed 665 * the awu_max. 666 */ 667 return min(awu_max, xfs_inode_buftarg(ip)->bt_awu_max); 668 } 669 670 static void 671 xfs_report_atomic_write( 672 struct xfs_inode *ip, 673 struct kstat *stat) 674 { 675 generic_fill_statx_atomic_writes(stat, 676 xfs_get_atomic_write_min(ip), 677 xfs_get_atomic_write_max(ip), 678 xfs_get_atomic_write_max_opt(ip)); 679 } 680 681 STATIC int 682 xfs_vn_getattr( 683 struct mnt_idmap *idmap, 684 const struct path *path, 685 struct kstat *stat, 686 u32 request_mask, 687 unsigned int query_flags) 688 { 689 struct inode *inode = d_inode(path->dentry); 690 struct xfs_inode *ip = XFS_I(inode); 691 struct xfs_mount *mp = ip->i_mount; 692 vfsuid_t vfsuid = i_uid_into_vfsuid(idmap, inode); 693 vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); 694 695 trace_xfs_getattr(ip); 696 697 if (xfs_is_shutdown(mp)) 698 return -EIO; 699 700 stat->size = XFS_ISIZE(ip); 701 stat->dev = inode->i_sb->s_dev; 702 stat->mode = inode->i_mode; 703 stat->nlink = inode->i_nlink; 704 stat->uid = vfsuid_into_kuid(vfsuid); 705 stat->gid = vfsgid_into_kgid(vfsgid); 706 stat->ino = ip->i_ino; 707 stat->atime = inode_get_atime(inode); 708 709 fill_mg_cmtime(stat, request_mask, inode); 710 711 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); 712 713 if (xfs_has_v3inodes(mp)) { 714 if (request_mask & STATX_BTIME) { 715 stat->result_mask |= STATX_BTIME; 716 stat->btime = ip->i_crtime; 717 } 718 } 719 720 /* 721 * Note: If you add another clause to set an attribute flag, please 722 * update attributes_mask below. 723 */ 724 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) 725 stat->attributes |= STATX_ATTR_IMMUTABLE; 726 if (ip->i_diflags & XFS_DIFLAG_APPEND) 727 stat->attributes |= STATX_ATTR_APPEND; 728 if (ip->i_diflags & XFS_DIFLAG_NODUMP) 729 stat->attributes |= STATX_ATTR_NODUMP; 730 731 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | 732 STATX_ATTR_APPEND | 733 STATX_ATTR_NODUMP); 734 735 switch (inode->i_mode & S_IFMT) { 736 case S_IFBLK: 737 case S_IFCHR: 738 stat->blksize = BLKDEV_IOSIZE; 739 stat->rdev = inode->i_rdev; 740 break; 741 case S_IFREG: 742 if (request_mask & (STATX_DIOALIGN | STATX_DIO_READ_ALIGN)) 743 xfs_report_dioalign(ip, stat); 744 if (request_mask & STATX_WRITE_ATOMIC) 745 xfs_report_atomic_write(ip, stat); 746 fallthrough; 747 default: 748 stat->blksize = xfs_stat_blksize(ip); 749 stat->rdev = 0; 750 break; 751 } 752 753 return 0; 754 } 755 756 static int 757 xfs_vn_change_ok( 758 struct mnt_idmap *idmap, 759 struct dentry *dentry, 760 struct iattr *iattr) 761 { 762 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 763 764 if (xfs_is_readonly(mp)) 765 return -EROFS; 766 767 if (xfs_is_shutdown(mp)) 768 return -EIO; 769 770 return setattr_prepare(idmap, dentry, iattr); 771 } 772 773 /* 774 * Set non-size attributes of an inode. 775 * 776 * Caution: The caller of this function is responsible for calling 777 * setattr_prepare() or otherwise verifying the change is fine. 778 */ 779 static int 780 xfs_setattr_nonsize( 781 struct mnt_idmap *idmap, 782 struct dentry *dentry, 783 struct xfs_inode *ip, 784 struct iattr *iattr) 785 { 786 xfs_mount_t *mp = ip->i_mount; 787 struct inode *inode = VFS_I(ip); 788 int mask = iattr->ia_valid; 789 xfs_trans_t *tp; 790 int error; 791 kuid_t uid = GLOBAL_ROOT_UID; 792 kgid_t gid = GLOBAL_ROOT_GID; 793 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 794 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; 795 796 ASSERT((mask & ATTR_SIZE) == 0); 797 798 /* 799 * If disk quotas is on, we make sure that the dquots do exist on disk, 800 * before we start any other transactions. Trying to do this later 801 * is messy. We don't care to take a readlock to look at the ids 802 * in inode here, because we can't hold it across the trans_reserve. 803 * If the IDs do change before we take the ilock, we're covered 804 * because the i_*dquot fields will get updated anyway. 805 */ 806 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 807 uint qflags = 0; 808 809 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 810 uid = from_vfsuid(idmap, i_user_ns(inode), 811 iattr->ia_vfsuid); 812 qflags |= XFS_QMOPT_UQUOTA; 813 } else { 814 uid = inode->i_uid; 815 } 816 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 817 gid = from_vfsgid(idmap, i_user_ns(inode), 818 iattr->ia_vfsgid); 819 qflags |= XFS_QMOPT_GQUOTA; 820 } else { 821 gid = inode->i_gid; 822 } 823 824 /* 825 * We take a reference when we initialize udqp and gdqp, 826 * so it is important that we never blindly double trip on 827 * the same variable. See xfs_create() for an example. 828 */ 829 ASSERT(udqp == NULL); 830 ASSERT(gdqp == NULL); 831 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, 832 qflags, &udqp, &gdqp, NULL); 833 if (error) 834 return error; 835 } 836 837 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, 838 has_capability_noaudit(current, CAP_FOWNER), &tp); 839 if (error) 840 goto out_dqrele; 841 842 /* 843 * Register quota modifications in the transaction. Must be the owner 844 * or privileged. These IDs could have changed since we last looked at 845 * them. But, we're assured that if the ownership did change while we 846 * didn't have the inode locked, inode's dquot(s) would have changed 847 * also. 848 */ 849 if (XFS_IS_UQUOTA_ON(mp) && 850 i_uid_needs_update(idmap, iattr, inode)) { 851 ASSERT(udqp); 852 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); 853 } 854 if (XFS_IS_GQUOTA_ON(mp) && 855 i_gid_needs_update(idmap, iattr, inode)) { 856 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp)); 857 ASSERT(gdqp); 858 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp); 859 } 860 861 setattr_copy(idmap, inode, iattr); 862 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 863 864 XFS_STATS_INC(mp, xs_ig_attrchg); 865 866 if (xfs_has_wsync(mp)) 867 xfs_trans_set_sync(tp); 868 error = xfs_trans_commit(tp); 869 870 /* 871 * Release any dquot(s) the inode had kept before chown. 872 */ 873 xfs_qm_dqrele(old_udqp); 874 xfs_qm_dqrele(old_gdqp); 875 xfs_qm_dqrele(udqp); 876 xfs_qm_dqrele(gdqp); 877 878 if (error) 879 return error; 880 881 /* 882 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 883 * update. We could avoid this with linked transactions 884 * and passing down the transaction pointer all the way 885 * to attr_set. No previous user of the generic 886 * Posix ACL code seems to care about this issue either. 887 */ 888 if (mask & ATTR_MODE) { 889 error = posix_acl_chmod(idmap, dentry, inode->i_mode); 890 if (error) 891 return error; 892 } 893 894 return 0; 895 896 out_dqrele: 897 xfs_qm_dqrele(udqp); 898 xfs_qm_dqrele(gdqp); 899 return error; 900 } 901 902 /* 903 * Truncate file. Must have write permission and not be a directory. 904 */ 905 int 906 xfs_vn_setattr_size( 907 struct mnt_idmap *idmap, 908 struct dentry *dentry, 909 struct iattr *iattr) 910 { 911 struct inode *inode = d_inode(dentry); 912 struct xfs_inode *ip = XFS_I(inode); 913 struct xfs_mount *mp = ip->i_mount; 914 xfs_off_t oldsize = inode->i_size; 915 xfs_off_t newsize = iattr->ia_size; 916 struct xfs_trans *tp; 917 int error; 918 uint lock_flags = 0; 919 uint resblks = 0; 920 bool did_zeroing = false; 921 struct xfs_zone_alloc_ctx ac = { }; 922 923 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL); 924 ASSERT(S_ISREG(inode->i_mode)); 925 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 926 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); 927 928 trace_xfs_setattr(ip); 929 930 error = xfs_vn_change_ok(idmap, dentry, iattr); 931 if (error) 932 return error; 933 934 /* 935 * Short circuit the truncate case for zero length files. 936 */ 937 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { 938 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 939 return 0; 940 941 /* 942 * Use the regular setattr path to update the timestamps. 943 */ 944 iattr->ia_valid &= ~ATTR_SIZE; 945 return xfs_setattr_nonsize(idmap, dentry, ip, iattr); 946 } 947 948 /* 949 * Make sure that the dquots are attached to the inode. 950 */ 951 error = xfs_qm_dqattach(ip); 952 if (error) 953 return error; 954 955 /* 956 * Wait for all direct I/O to complete. 957 */ 958 inode_dio_wait(inode); 959 960 /* 961 * Normally xfs_zoned_space_reserve is supposed to be called outside the 962 * IOLOCK. For truncate we can't do that since ->setattr is called with 963 * it already held by the VFS. So for now chicken out and try to 964 * allocate space under it. 965 * 966 * To avoid deadlocks this means we can't block waiting for space, which 967 * can lead to spurious -ENOSPC if there are no directly available 968 * blocks. We mitigate this a bit by allowing zeroing to dip into the 969 * reserved pool, but eventually the VFS calling convention needs to 970 * change. 971 */ 972 if (xfs_is_zoned_inode(ip)) { 973 error = xfs_zoned_space_reserve(mp, 1, 974 XFS_ZR_NOWAIT | XFS_ZR_RESERVED, &ac); 975 if (error) { 976 if (error == -EAGAIN) 977 return -ENOSPC; 978 return error; 979 } 980 } 981 982 /* 983 * File data changes must be complete before we start the transaction to 984 * modify the inode. This needs to be done before joining the inode to 985 * the transaction because the inode cannot be unlocked once it is a 986 * part of the transaction. 987 * 988 * Start with zeroing any data beyond EOF that we may expose on file 989 * extension, or zeroing out the rest of the block on a downward 990 * truncate. 991 */ 992 if (newsize > oldsize) { 993 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 994 error = xfs_zero_range(ip, oldsize, newsize - oldsize, 995 &ac, &did_zeroing); 996 } else { 997 error = xfs_truncate_page(ip, newsize, &ac, &did_zeroing); 998 } 999 1000 if (xfs_is_zoned_inode(ip)) 1001 xfs_zoned_space_unreserve(mp, &ac); 1002 1003 if (error) 1004 return error; 1005 1006 /* 1007 * We've already locked out new page faults, so now we can safely remove 1008 * pages from the page cache knowing they won't get refaulted until we 1009 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 1010 * complete. The truncate_setsize() call also cleans partial EOF page 1011 * PTEs on extending truncates and hence ensures sub-page block size 1012 * filesystems are correctly handled, too. 1013 * 1014 * We have to do all the page cache truncate work outside the 1015 * transaction context as the "lock" order is page lock->log space 1016 * reservation as defined by extent allocation in the writeback path. 1017 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 1018 * having already truncated the in-memory version of the file (i.e. made 1019 * user visible changes). There's not much we can do about this, except 1020 * to hope that the caller sees ENOMEM and retries the truncate 1021 * operation. 1022 * 1023 * And we update in-core i_size and truncate page cache beyond newsize 1024 * before writeback the [i_disk_size, newsize] range, so we're 1025 * guaranteed not to write stale data past the new EOF on truncate down. 1026 */ 1027 truncate_setsize(inode, newsize); 1028 1029 /* 1030 * We are going to log the inode size change in this transaction so 1031 * any previous writes that are beyond the on disk EOF and the new 1032 * EOF that have not been written out need to be written here. If we 1033 * do not write the data out, we expose ourselves to the null files 1034 * problem. Note that this includes any block zeroing we did above; 1035 * otherwise those blocks may not be zeroed after a crash. 1036 */ 1037 if (did_zeroing || 1038 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { 1039 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 1040 ip->i_disk_size, newsize - 1); 1041 if (error) 1042 return error; 1043 } 1044 1045 /* 1046 * For realtime inode with more than one block rtextsize, we need the 1047 * block reservation for bmap btree block allocations/splits that can 1048 * happen since it could split the tail written extent and convert the 1049 * right beyond EOF one to unwritten. 1050 */ 1051 if (xfs_inode_has_bigrtalloc(ip)) 1052 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0); 1053 1054 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, resblks, 1055 0, 0, &tp); 1056 if (error) 1057 return error; 1058 1059 lock_flags |= XFS_ILOCK_EXCL; 1060 xfs_ilock(ip, XFS_ILOCK_EXCL); 1061 xfs_trans_ijoin(tp, ip, 0); 1062 1063 /* 1064 * Only change the c/mtime if we are changing the size or we are 1065 * explicitly asked to change it. This handles the semantic difference 1066 * between truncate() and ftruncate() as implemented in the VFS. 1067 * 1068 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 1069 * special case where we need to update the times despite not having 1070 * these flags set. For all other operations the VFS set these flags 1071 * explicitly if it wants a timestamp update. 1072 */ 1073 if (newsize != oldsize && 1074 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 1075 iattr->ia_ctime = iattr->ia_mtime = 1076 current_time(inode); 1077 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 1078 } 1079 1080 /* 1081 * The first thing we do is set the size to new_size permanently on 1082 * disk. This way we don't have to worry about anyone ever being able 1083 * to look at the data being freed even in the face of a crash. 1084 * What we're getting around here is the case where we free a block, it 1085 * is allocated to another file, it is written to, and then we crash. 1086 * If the new data gets written to the file but the log buffers 1087 * containing the free and reallocation don't, then we'd end up with 1088 * garbage in the blocks being freed. As long as we make the new size 1089 * permanent before actually freeing any blocks it doesn't matter if 1090 * they get written to. 1091 */ 1092 ip->i_disk_size = newsize; 1093 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1094 1095 if (newsize <= oldsize) { 1096 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 1097 if (error) 1098 goto out_trans_cancel; 1099 1100 /* 1101 * Truncated "down", so we're removing references to old data 1102 * here - if we delay flushing for a long time, we expose 1103 * ourselves unduly to the notorious NULL files problem. So, 1104 * we mark this inode and flush it when the file is closed, 1105 * and do not wait the usual (long) time for writeout. 1106 */ 1107 xfs_iflags_set(ip, XFS_ITRUNCATED); 1108 1109 /* A truncate down always removes post-EOF blocks. */ 1110 xfs_inode_clear_eofblocks_tag(ip); 1111 } 1112 1113 setattr_copy(idmap, inode, iattr); 1114 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 1115 1116 XFS_STATS_INC(mp, xs_ig_attrchg); 1117 1118 if (xfs_has_wsync(mp)) 1119 xfs_trans_set_sync(tp); 1120 1121 error = xfs_trans_commit(tp); 1122 out_unlock: 1123 if (lock_flags) 1124 xfs_iunlock(ip, lock_flags); 1125 return error; 1126 1127 out_trans_cancel: 1128 xfs_trans_cancel(tp); 1129 goto out_unlock; 1130 } 1131 1132 STATIC int 1133 xfs_vn_setattr( 1134 struct mnt_idmap *idmap, 1135 struct dentry *dentry, 1136 struct iattr *iattr) 1137 { 1138 struct inode *inode = d_inode(dentry); 1139 struct xfs_inode *ip = XFS_I(inode); 1140 int error; 1141 1142 if (iattr->ia_valid & ATTR_SIZE) { 1143 uint iolock; 1144 1145 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1146 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1147 1148 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1149 if (error) { 1150 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1151 return error; 1152 } 1153 1154 error = xfs_vn_setattr_size(idmap, dentry, iattr); 1155 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1156 } else { 1157 trace_xfs_setattr(ip); 1158 1159 error = xfs_vn_change_ok(idmap, dentry, iattr); 1160 if (!error) 1161 error = xfs_setattr_nonsize(idmap, dentry, ip, iattr); 1162 } 1163 1164 return error; 1165 } 1166 1167 STATIC int 1168 xfs_vn_update_time( 1169 struct inode *inode, 1170 enum fs_update_time type, 1171 unsigned int flags) 1172 { 1173 struct xfs_inode *ip = XFS_I(inode); 1174 struct xfs_mount *mp = ip->i_mount; 1175 int log_flags = XFS_ILOG_TIMESTAMP; 1176 struct xfs_trans *tp; 1177 int error; 1178 1179 trace_xfs_update_time(ip); 1180 1181 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1182 int dirty; 1183 1184 dirty = inode_update_time(inode, type, flags); 1185 if (dirty <= 0) 1186 return dirty; 1187 if (dirty == I_DIRTY_TIME) { 1188 __mark_inode_dirty(inode, I_DIRTY_TIME); 1189 return 0; 1190 } 1191 1192 /* Capture the iversion update that just occurred */ 1193 log_flags |= XFS_ILOG_CORE; 1194 } else { 1195 if (flags & IOCB_NOWAIT) 1196 return -EAGAIN; 1197 } 1198 1199 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1200 if (error) 1201 return error; 1202 1203 xfs_ilock(ip, XFS_ILOCK_EXCL); 1204 if (type == FS_UPD_ATIME) 1205 inode_set_atime_to_ts(inode, current_time(inode)); 1206 else 1207 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode)); 1208 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1209 xfs_trans_log_inode(tp, ip, log_flags); 1210 return xfs_trans_commit(tp); 1211 } 1212 1213 static void 1214 xfs_vn_sync_lazytime( 1215 struct inode *inode) 1216 { 1217 struct xfs_inode *ip = XFS_I(inode); 1218 struct xfs_mount *mp = ip->i_mount; 1219 struct xfs_trans *tp; 1220 1221 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp)) 1222 return; 1223 xfs_ilock(ip, XFS_ILOCK_EXCL); 1224 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1225 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP); 1226 xfs_trans_commit(tp); 1227 } 1228 1229 STATIC int 1230 xfs_vn_fiemap( 1231 struct inode *inode, 1232 struct fiemap_extent_info *fieinfo, 1233 u64 start, 1234 u64 length) 1235 { 1236 int error; 1237 1238 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1239 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1240 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1241 error = iomap_fiemap(inode, fieinfo, start, length, 1242 &xfs_xattr_iomap_ops); 1243 } else { 1244 error = iomap_fiemap(inode, fieinfo, start, length, 1245 &xfs_read_iomap_ops); 1246 } 1247 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1248 1249 return error; 1250 } 1251 1252 STATIC int 1253 xfs_vn_tmpfile( 1254 struct mnt_idmap *idmap, 1255 struct inode *dir, 1256 struct file *file, 1257 umode_t mode) 1258 { 1259 int err = xfs_generic_create(idmap, dir, file->f_path.dentry, mode, 0, file); 1260 1261 return finish_open_simple(file, err); 1262 } 1263 1264 static const struct inode_operations xfs_inode_operations = { 1265 .get_inode_acl = xfs_get_acl, 1266 .set_acl = xfs_set_acl, 1267 .getattr = xfs_vn_getattr, 1268 .setattr = xfs_vn_setattr, 1269 .listxattr = xfs_vn_listxattr, 1270 .fiemap = xfs_vn_fiemap, 1271 .update_time = xfs_vn_update_time, 1272 .sync_lazytime = xfs_vn_sync_lazytime, 1273 .fileattr_get = xfs_fileattr_get, 1274 .fileattr_set = xfs_fileattr_set, 1275 }; 1276 1277 static const struct inode_operations xfs_dir_inode_operations = { 1278 .create = xfs_vn_create, 1279 .lookup = xfs_vn_lookup, 1280 .link = xfs_vn_link, 1281 .unlink = xfs_vn_unlink, 1282 .symlink = xfs_vn_symlink, 1283 .mkdir = xfs_vn_mkdir, 1284 /* 1285 * Yes, XFS uses the same method for rmdir and unlink. 1286 * 1287 * There are some subtile differences deeper in the code, 1288 * but we use S_ISDIR to check for those. 1289 */ 1290 .rmdir = xfs_vn_unlink, 1291 .mknod = xfs_vn_mknod, 1292 .rename = xfs_vn_rename, 1293 .get_inode_acl = xfs_get_acl, 1294 .set_acl = xfs_set_acl, 1295 .getattr = xfs_vn_getattr, 1296 .setattr = xfs_vn_setattr, 1297 .listxattr = xfs_vn_listxattr, 1298 .update_time = xfs_vn_update_time, 1299 .sync_lazytime = xfs_vn_sync_lazytime, 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 .sync_lazytime = xfs_vn_sync_lazytime, 1328 .tmpfile = xfs_vn_tmpfile, 1329 .fileattr_get = xfs_fileattr_get, 1330 .fileattr_set = xfs_fileattr_set, 1331 }; 1332 1333 static const struct inode_operations xfs_symlink_inode_operations = { 1334 .get_link = xfs_vn_get_link, 1335 .getattr = xfs_vn_getattr, 1336 .setattr = xfs_vn_setattr, 1337 .listxattr = xfs_vn_listxattr, 1338 .update_time = xfs_vn_update_time, 1339 .sync_lazytime = xfs_vn_sync_lazytime, 1340 .fileattr_get = xfs_fileattr_get, 1341 .fileattr_set = xfs_fileattr_set, 1342 }; 1343 1344 /* Figure out if this file actually supports DAX. */ 1345 static bool 1346 xfs_inode_supports_dax( 1347 struct xfs_inode *ip) 1348 { 1349 struct xfs_mount *mp = ip->i_mount; 1350 1351 /* Only supported on regular files. */ 1352 if (!S_ISREG(VFS_I(ip)->i_mode)) 1353 return false; 1354 1355 /* Block size must match page size */ 1356 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1357 return false; 1358 1359 /* Device has to support DAX too. */ 1360 return xfs_inode_buftarg(ip)->bt_daxdev != NULL; 1361 } 1362 1363 static bool 1364 xfs_inode_should_enable_dax( 1365 struct xfs_inode *ip) 1366 { 1367 if (!IS_ENABLED(CONFIG_FS_DAX)) 1368 return false; 1369 if (xfs_has_dax_never(ip->i_mount)) 1370 return false; 1371 if (!xfs_inode_supports_dax(ip)) 1372 return false; 1373 if (xfs_has_dax_always(ip->i_mount)) 1374 return true; 1375 if (ip->i_diflags2 & XFS_DIFLAG2_DAX) 1376 return true; 1377 return false; 1378 } 1379 1380 void 1381 xfs_diflags_to_iflags( 1382 struct xfs_inode *ip, 1383 bool init) 1384 { 1385 struct inode *inode = VFS_I(ip); 1386 unsigned int xflags = xfs_ip2xflags(ip); 1387 unsigned int flags = 0; 1388 1389 ASSERT(!(IS_DAX(inode) && init)); 1390 1391 if (xflags & FS_XFLAG_IMMUTABLE) 1392 flags |= S_IMMUTABLE; 1393 if (xflags & FS_XFLAG_APPEND) 1394 flags |= S_APPEND; 1395 if (xflags & FS_XFLAG_SYNC) 1396 flags |= S_SYNC; 1397 if (xflags & FS_XFLAG_NOATIME) 1398 flags |= S_NOATIME; 1399 if (init && xfs_inode_should_enable_dax(ip)) 1400 flags |= S_DAX; 1401 1402 /* 1403 * S_DAX can only be set during inode initialization and is never set by 1404 * the VFS, so we cannot mask off S_DAX in i_flags. 1405 */ 1406 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); 1407 inode->i_flags |= flags; 1408 } 1409 1410 /* 1411 * Initialize the Linux inode. 1412 * 1413 * When reading existing inodes from disk this is called directly from xfs_iget, 1414 * when creating a new inode it is called from xfs_init_new_inode after setting 1415 * up the inode. These callers have different criteria for clearing XFS_INEW, so 1416 * leave it up to the caller to deal with unlocking the inode appropriately. 1417 */ 1418 void 1419 xfs_setup_inode( 1420 struct xfs_inode *ip) 1421 { 1422 struct inode *inode = &ip->i_vnode; 1423 gfp_t gfp_mask; 1424 bool is_meta = xfs_is_internal_inode(ip); 1425 1426 inode->i_ino = ip->i_ino; 1427 inode_state_set_raw(inode, I_NEW); 1428 1429 inode_sb_list_add(inode); 1430 /* make the inode look hashed for the writeback code */ 1431 inode_fake_hash(inode); 1432 1433 i_size_write(inode, ip->i_disk_size); 1434 xfs_diflags_to_iflags(ip, true); 1435 1436 /* 1437 * Mark our metadata files as private so that LSMs and the ACL code 1438 * don't try to add their own metadata or reason about these files, 1439 * and users cannot ever obtain file handles to them. 1440 */ 1441 if (is_meta) { 1442 inode->i_flags |= S_PRIVATE; 1443 inode->i_opflags &= ~IOP_XATTR; 1444 } 1445 1446 if (S_ISDIR(inode->i_mode)) { 1447 /* 1448 * We set the i_rwsem class here to avoid potential races with 1449 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1450 * after a filehandle lookup has already found the inode in 1451 * cache before it has been unlocked via unlock_new_inode(). 1452 */ 1453 lockdep_set_class(&inode->i_rwsem, 1454 &inode->i_sb->s_type->i_mutex_dir_key); 1455 lockdep_set_class(&ip->i_lock, &xfs_dir_ilock_class); 1456 } else { 1457 lockdep_set_class(&ip->i_lock, &xfs_nondir_ilock_class); 1458 } 1459 1460 /* 1461 * Ensure all page cache allocations are done from GFP_NOFS context to 1462 * prevent direct reclaim recursion back into the filesystem and blowing 1463 * stacks or deadlocking. 1464 */ 1465 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1466 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1467 1468 /* 1469 * For real-time inodes update the stable write flags to that of the RT 1470 * device instead of the data device. 1471 */ 1472 if (S_ISREG(inode->i_mode) && XFS_IS_REALTIME_INODE(ip)) 1473 xfs_update_stable_writes(ip); 1474 1475 /* 1476 * If there is no attribute fork no ACL can exist on this inode, 1477 * and it can't have any file capabilities attached to it either. 1478 */ 1479 if (!xfs_inode_has_attr_fork(ip)) { 1480 inode_has_no_xattr(inode); 1481 cache_no_acl(inode); 1482 } 1483 } 1484 1485 void 1486 xfs_setup_iops( 1487 struct xfs_inode *ip) 1488 { 1489 struct inode *inode = &ip->i_vnode; 1490 1491 switch (inode->i_mode & S_IFMT) { 1492 case S_IFREG: 1493 inode->i_op = &xfs_inode_operations; 1494 inode->i_fop = &xfs_file_operations; 1495 if (IS_DAX(inode)) 1496 inode->i_mapping->a_ops = &xfs_dax_aops; 1497 else 1498 inode->i_mapping->a_ops = &xfs_address_space_operations; 1499 break; 1500 case S_IFDIR: 1501 if (xfs_has_asciici(XFS_M(inode->i_sb))) 1502 inode->i_op = &xfs_dir_ci_inode_operations; 1503 else 1504 inode->i_op = &xfs_dir_inode_operations; 1505 inode->i_fop = &xfs_dir_file_operations; 1506 break; 1507 case S_IFLNK: 1508 inode->i_op = &xfs_symlink_inode_operations; 1509 break; 1510 default: 1511 inode->i_op = &xfs_inode_operations; 1512 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1513 break; 1514 } 1515 } 1516