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_trace.h" 21 #include "xfs_icache.h" 22 #include "xfs_symlink.h" 23 #include "xfs_dir2.h" 24 #include "xfs_iomap.h" 25 #include "xfs_error.h" 26 #include "xfs_ioctl.h" 27 #include "xfs_xattr.h" 28 29 #include <linux/posix_acl.h> 30 #include <linux/security.h> 31 #include <linux/iversion.h> 32 #include <linux/fiemap.h> 33 34 /* 35 * Directories have different lock order w.r.t. mmap_lock compared to regular 36 * files. This is due to readdir potentially triggering page faults on a user 37 * buffer inside filldir(), and this happens with the ilock on the directory 38 * held. For regular files, the lock order is the other way around - the 39 * mmap_lock is taken during the page fault, and then we lock the ilock to do 40 * block mapping. Hence we need a different class for the directory ilock so 41 * that lockdep can tell them apart. 42 */ 43 static struct lock_class_key xfs_nondir_ilock_class; 44 static struct lock_class_key xfs_dir_ilock_class; 45 46 static int 47 xfs_initxattrs( 48 struct inode *inode, 49 const struct xattr *xattr_array, 50 void *fs_info) 51 { 52 const struct xattr *xattr; 53 struct xfs_inode *ip = XFS_I(inode); 54 int error = 0; 55 56 for (xattr = xattr_array; xattr->name != NULL; xattr++) { 57 struct xfs_da_args args = { 58 .dp = ip, 59 .attr_filter = XFS_ATTR_SECURE, 60 .name = xattr->name, 61 .namelen = strlen(xattr->name), 62 .value = xattr->value, 63 .valuelen = xattr->value_len, 64 }; 65 error = xfs_attr_change(&args); 66 if (error < 0) 67 break; 68 } 69 return error; 70 } 71 72 /* 73 * Hook in SELinux. This is not quite correct yet, what we really need 74 * here (as we do for default ACLs) is a mechanism by which creation of 75 * these attrs can be journalled at inode creation time (along with the 76 * inode, of course, such that log replay can't cause these to be lost). 77 */ 78 int 79 xfs_inode_init_security( 80 struct inode *inode, 81 struct inode *dir, 82 const struct qstr *qstr) 83 { 84 return security_inode_init_security(inode, dir, qstr, 85 &xfs_initxattrs, NULL); 86 } 87 88 static void 89 xfs_dentry_to_name( 90 struct xfs_name *namep, 91 struct dentry *dentry) 92 { 93 namep->name = dentry->d_name.name; 94 namep->len = dentry->d_name.len; 95 namep->type = XFS_DIR3_FT_UNKNOWN; 96 } 97 98 static int 99 xfs_dentry_mode_to_name( 100 struct xfs_name *namep, 101 struct dentry *dentry, 102 int mode) 103 { 104 namep->name = dentry->d_name.name; 105 namep->len = dentry->d_name.len; 106 namep->type = xfs_mode_to_ftype(mode); 107 108 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) 109 return -EFSCORRUPTED; 110 111 return 0; 112 } 113 114 STATIC void 115 xfs_cleanup_inode( 116 struct inode *dir, 117 struct inode *inode, 118 struct dentry *dentry) 119 { 120 struct xfs_name teardown; 121 122 /* Oh, the horror. 123 * If we can't add the ACL or we fail in 124 * xfs_inode_init_security we must back out. 125 * ENOSPC can hit here, among other things. 126 */ 127 xfs_dentry_to_name(&teardown, dentry); 128 129 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); 130 } 131 132 /* 133 * Check to see if we are likely to need an extended attribute to be added to 134 * the inode we are about to allocate. This allows the attribute fork to be 135 * created during the inode allocation, reducing the number of transactions we 136 * need to do in this fast path. 137 * 138 * The security checks are optimistic, but not guaranteed. The two LSMs that 139 * require xattrs to be added here (selinux and smack) are also the only two 140 * LSMs that add a sb->s_security structure to the superblock. Hence if security 141 * is enabled and sb->s_security is set, we have a pretty good idea that we are 142 * going to be asked to add a security xattr immediately after allocating the 143 * xfs inode and instantiating the VFS inode. 144 */ 145 static inline bool 146 xfs_create_need_xattr( 147 struct inode *dir, 148 struct posix_acl *default_acl, 149 struct posix_acl *acl) 150 { 151 if (acl) 152 return true; 153 if (default_acl) 154 return true; 155 #if IS_ENABLED(CONFIG_SECURITY) 156 if (dir->i_sb->s_security) 157 return true; 158 #endif 159 return false; 160 } 161 162 163 STATIC int 164 xfs_generic_create( 165 struct user_namespace *mnt_userns, 166 struct inode *dir, 167 struct dentry *dentry, 168 umode_t mode, 169 dev_t rdev, 170 bool tmpfile) /* unnamed file */ 171 { 172 struct inode *inode; 173 struct xfs_inode *ip = NULL; 174 struct posix_acl *default_acl, *acl; 175 struct xfs_name name; 176 int error; 177 178 /* 179 * Irix uses Missed'em'V split, but doesn't want to see 180 * the upper 5 bits of (14bit) major. 181 */ 182 if (S_ISCHR(mode) || S_ISBLK(mode)) { 183 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) 184 return -EINVAL; 185 } else { 186 rdev = 0; 187 } 188 189 error = posix_acl_create(dir, &mode, &default_acl, &acl); 190 if (error) 191 return error; 192 193 /* Verify mode is valid also for tmpfile case */ 194 error = xfs_dentry_mode_to_name(&name, dentry, mode); 195 if (unlikely(error)) 196 goto out_free_acl; 197 198 if (!tmpfile) { 199 error = xfs_create(mnt_userns, XFS_I(dir), &name, mode, rdev, 200 xfs_create_need_xattr(dir, default_acl, acl), 201 &ip); 202 } else { 203 error = xfs_create_tmpfile(mnt_userns, XFS_I(dir), mode, &ip); 204 } 205 if (unlikely(error)) 206 goto out_free_acl; 207 208 inode = VFS_I(ip); 209 210 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 211 if (unlikely(error)) 212 goto out_cleanup_inode; 213 214 if (default_acl) { 215 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); 216 if (error) 217 goto out_cleanup_inode; 218 } 219 if (acl) { 220 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); 221 if (error) 222 goto out_cleanup_inode; 223 } 224 225 xfs_setup_iops(ip); 226 227 if (tmpfile) { 228 /* 229 * The VFS requires that any inode fed to d_tmpfile must have 230 * nlink == 1 so that it can decrement the nlink in d_tmpfile. 231 * However, we created the temp file with nlink == 0 because 232 * we're not allowed to put an inode with nlink > 0 on the 233 * unlinked list. Therefore we have to set nlink to 1 so that 234 * d_tmpfile can immediately set it back to zero. 235 */ 236 set_nlink(inode, 1); 237 d_tmpfile(dentry, inode); 238 } else 239 d_instantiate(dentry, inode); 240 241 xfs_finish_inode_setup(ip); 242 243 out_free_acl: 244 posix_acl_release(default_acl); 245 posix_acl_release(acl); 246 return error; 247 248 out_cleanup_inode: 249 xfs_finish_inode_setup(ip); 250 if (!tmpfile) 251 xfs_cleanup_inode(dir, inode, dentry); 252 xfs_irele(ip); 253 goto out_free_acl; 254 } 255 256 STATIC int 257 xfs_vn_mknod( 258 struct user_namespace *mnt_userns, 259 struct inode *dir, 260 struct dentry *dentry, 261 umode_t mode, 262 dev_t rdev) 263 { 264 return xfs_generic_create(mnt_userns, dir, dentry, mode, rdev, false); 265 } 266 267 STATIC int 268 xfs_vn_create( 269 struct user_namespace *mnt_userns, 270 struct inode *dir, 271 struct dentry *dentry, 272 umode_t mode, 273 bool flags) 274 { 275 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, false); 276 } 277 278 STATIC int 279 xfs_vn_mkdir( 280 struct user_namespace *mnt_userns, 281 struct inode *dir, 282 struct dentry *dentry, 283 umode_t mode) 284 { 285 return xfs_generic_create(mnt_userns, dir, dentry, mode | S_IFDIR, 0, 286 false); 287 } 288 289 STATIC struct dentry * 290 xfs_vn_lookup( 291 struct inode *dir, 292 struct dentry *dentry, 293 unsigned int flags) 294 { 295 struct inode *inode; 296 struct xfs_inode *cip; 297 struct xfs_name name; 298 int error; 299 300 if (dentry->d_name.len >= MAXNAMELEN) 301 return ERR_PTR(-ENAMETOOLONG); 302 303 xfs_dentry_to_name(&name, dentry); 304 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); 305 if (likely(!error)) 306 inode = VFS_I(cip); 307 else if (likely(error == -ENOENT)) 308 inode = NULL; 309 else 310 inode = ERR_PTR(error); 311 return d_splice_alias(inode, dentry); 312 } 313 314 STATIC struct dentry * 315 xfs_vn_ci_lookup( 316 struct inode *dir, 317 struct dentry *dentry, 318 unsigned int flags) 319 { 320 struct xfs_inode *ip; 321 struct xfs_name xname; 322 struct xfs_name ci_name; 323 struct qstr dname; 324 int error; 325 326 if (dentry->d_name.len >= MAXNAMELEN) 327 return ERR_PTR(-ENAMETOOLONG); 328 329 xfs_dentry_to_name(&xname, dentry); 330 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); 331 if (unlikely(error)) { 332 if (unlikely(error != -ENOENT)) 333 return ERR_PTR(error); 334 /* 335 * call d_add(dentry, NULL) here when d_drop_negative_children 336 * is called in xfs_vn_mknod (ie. allow negative dentries 337 * with CI filesystems). 338 */ 339 return NULL; 340 } 341 342 /* if exact match, just splice and exit */ 343 if (!ci_name.name) 344 return d_splice_alias(VFS_I(ip), dentry); 345 346 /* else case-insensitive match... */ 347 dname.name = ci_name.name; 348 dname.len = ci_name.len; 349 dentry = d_add_ci(dentry, VFS_I(ip), &dname); 350 kmem_free(ci_name.name); 351 return dentry; 352 } 353 354 STATIC int 355 xfs_vn_link( 356 struct dentry *old_dentry, 357 struct inode *dir, 358 struct dentry *dentry) 359 { 360 struct inode *inode = d_inode(old_dentry); 361 struct xfs_name name; 362 int error; 363 364 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); 365 if (unlikely(error)) 366 return error; 367 368 error = xfs_link(XFS_I(dir), XFS_I(inode), &name); 369 if (unlikely(error)) 370 return error; 371 372 ihold(inode); 373 d_instantiate(dentry, inode); 374 return 0; 375 } 376 377 STATIC int 378 xfs_vn_unlink( 379 struct inode *dir, 380 struct dentry *dentry) 381 { 382 struct xfs_name name; 383 int error; 384 385 xfs_dentry_to_name(&name, dentry); 386 387 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); 388 if (error) 389 return error; 390 391 /* 392 * With unlink, the VFS makes the dentry "negative": no inode, 393 * but still hashed. This is incompatible with case-insensitive 394 * mode, so invalidate (unhash) the dentry in CI-mode. 395 */ 396 if (xfs_has_asciici(XFS_M(dir->i_sb))) 397 d_invalidate(dentry); 398 return 0; 399 } 400 401 STATIC int 402 xfs_vn_symlink( 403 struct user_namespace *mnt_userns, 404 struct inode *dir, 405 struct dentry *dentry, 406 const char *symname) 407 { 408 struct inode *inode; 409 struct xfs_inode *cip = NULL; 410 struct xfs_name name; 411 int error; 412 umode_t mode; 413 414 mode = S_IFLNK | 415 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); 416 error = xfs_dentry_mode_to_name(&name, dentry, mode); 417 if (unlikely(error)) 418 goto out; 419 420 error = xfs_symlink(mnt_userns, XFS_I(dir), &name, symname, mode, &cip); 421 if (unlikely(error)) 422 goto out; 423 424 inode = VFS_I(cip); 425 426 error = xfs_inode_init_security(inode, dir, &dentry->d_name); 427 if (unlikely(error)) 428 goto out_cleanup_inode; 429 430 xfs_setup_iops(cip); 431 432 d_instantiate(dentry, inode); 433 xfs_finish_inode_setup(cip); 434 return 0; 435 436 out_cleanup_inode: 437 xfs_finish_inode_setup(cip); 438 xfs_cleanup_inode(dir, inode, dentry); 439 xfs_irele(cip); 440 out: 441 return error; 442 } 443 444 STATIC int 445 xfs_vn_rename( 446 struct user_namespace *mnt_userns, 447 struct inode *odir, 448 struct dentry *odentry, 449 struct inode *ndir, 450 struct dentry *ndentry, 451 unsigned int flags) 452 { 453 struct inode *new_inode = d_inode(ndentry); 454 int omode = 0; 455 int error; 456 struct xfs_name oname; 457 struct xfs_name nname; 458 459 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 460 return -EINVAL; 461 462 /* if we are exchanging files, we need to set i_mode of both files */ 463 if (flags & RENAME_EXCHANGE) 464 omode = d_inode(ndentry)->i_mode; 465 466 error = xfs_dentry_mode_to_name(&oname, odentry, omode); 467 if (omode && unlikely(error)) 468 return error; 469 470 error = xfs_dentry_mode_to_name(&nname, ndentry, 471 d_inode(odentry)->i_mode); 472 if (unlikely(error)) 473 return error; 474 475 return xfs_rename(mnt_userns, XFS_I(odir), &oname, 476 XFS_I(d_inode(odentry)), XFS_I(ndir), &nname, 477 new_inode ? XFS_I(new_inode) : NULL, flags); 478 } 479 480 /* 481 * careful here - this function can get called recursively, so 482 * we need to be very careful about how much stack we use. 483 * uio is kmalloced for this reason... 484 */ 485 STATIC const char * 486 xfs_vn_get_link( 487 struct dentry *dentry, 488 struct inode *inode, 489 struct delayed_call *done) 490 { 491 char *link; 492 int error = -ENOMEM; 493 494 if (!dentry) 495 return ERR_PTR(-ECHILD); 496 497 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); 498 if (!link) 499 goto out_err; 500 501 error = xfs_readlink(XFS_I(d_inode(dentry)), link); 502 if (unlikely(error)) 503 goto out_kfree; 504 505 set_delayed_call(done, kfree_link, link); 506 return link; 507 508 out_kfree: 509 kfree(link); 510 out_err: 511 return ERR_PTR(error); 512 } 513 514 static uint32_t 515 xfs_stat_blksize( 516 struct xfs_inode *ip) 517 { 518 struct xfs_mount *mp = ip->i_mount; 519 520 /* 521 * If the file blocks are being allocated from a realtime volume, then 522 * always return the realtime extent size. 523 */ 524 if (XFS_IS_REALTIME_INODE(ip)) 525 return XFS_FSB_TO_B(mp, xfs_get_extsz_hint(ip)); 526 527 /* 528 * Allow large block sizes to be reported to userspace programs if the 529 * "largeio" mount option is used. 530 * 531 * If compatibility mode is specified, simply return the basic unit of 532 * caching so that we don't get inefficient read/modify/write I/O from 533 * user apps. Otherwise.... 534 * 535 * If the underlying volume is a stripe, then return the stripe width in 536 * bytes as the recommended I/O size. It is not a stripe and we've set a 537 * default buffered I/O size, return that, otherwise return the compat 538 * default. 539 */ 540 if (xfs_has_large_iosize(mp)) { 541 if (mp->m_swidth) 542 return XFS_FSB_TO_B(mp, mp->m_swidth); 543 if (xfs_has_allocsize(mp)) 544 return 1U << mp->m_allocsize_log; 545 } 546 547 return PAGE_SIZE; 548 } 549 550 STATIC int 551 xfs_vn_getattr( 552 struct user_namespace *mnt_userns, 553 const struct path *path, 554 struct kstat *stat, 555 u32 request_mask, 556 unsigned int query_flags) 557 { 558 struct inode *inode = d_inode(path->dentry); 559 struct xfs_inode *ip = XFS_I(inode); 560 struct xfs_mount *mp = ip->i_mount; 561 562 trace_xfs_getattr(ip); 563 564 if (xfs_is_shutdown(mp)) 565 return -EIO; 566 567 stat->size = XFS_ISIZE(ip); 568 stat->dev = inode->i_sb->s_dev; 569 stat->mode = inode->i_mode; 570 stat->nlink = inode->i_nlink; 571 stat->uid = i_uid_into_mnt(mnt_userns, inode); 572 stat->gid = i_gid_into_mnt(mnt_userns, inode); 573 stat->ino = ip->i_ino; 574 stat->atime = inode->i_atime; 575 stat->mtime = inode->i_mtime; 576 stat->ctime = inode->i_ctime; 577 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); 578 579 if (xfs_has_v3inodes(mp)) { 580 if (request_mask & STATX_BTIME) { 581 stat->result_mask |= STATX_BTIME; 582 stat->btime = ip->i_crtime; 583 } 584 } 585 586 /* 587 * Note: If you add another clause to set an attribute flag, please 588 * update attributes_mask below. 589 */ 590 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) 591 stat->attributes |= STATX_ATTR_IMMUTABLE; 592 if (ip->i_diflags & XFS_DIFLAG_APPEND) 593 stat->attributes |= STATX_ATTR_APPEND; 594 if (ip->i_diflags & XFS_DIFLAG_NODUMP) 595 stat->attributes |= STATX_ATTR_NODUMP; 596 597 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | 598 STATX_ATTR_APPEND | 599 STATX_ATTR_NODUMP); 600 601 switch (inode->i_mode & S_IFMT) { 602 case S_IFBLK: 603 case S_IFCHR: 604 stat->blksize = BLKDEV_IOSIZE; 605 stat->rdev = inode->i_rdev; 606 break; 607 case S_IFREG: 608 if (request_mask & STATX_DIOALIGN) { 609 struct xfs_buftarg *target = xfs_inode_buftarg(ip); 610 struct block_device *bdev = target->bt_bdev; 611 612 stat->result_mask |= STATX_DIOALIGN; 613 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; 614 stat->dio_offset_align = bdev_logical_block_size(bdev); 615 } 616 fallthrough; 617 default: 618 stat->blksize = xfs_stat_blksize(ip); 619 stat->rdev = 0; 620 break; 621 } 622 623 return 0; 624 } 625 626 static int 627 xfs_vn_change_ok( 628 struct user_namespace *mnt_userns, 629 struct dentry *dentry, 630 struct iattr *iattr) 631 { 632 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 633 634 if (xfs_is_readonly(mp)) 635 return -EROFS; 636 637 if (xfs_is_shutdown(mp)) 638 return -EIO; 639 640 return setattr_prepare(mnt_userns, dentry, iattr); 641 } 642 643 /* 644 * Set non-size attributes of an inode. 645 * 646 * Caution: The caller of this function is responsible for calling 647 * setattr_prepare() or otherwise verifying the change is fine. 648 */ 649 static int 650 xfs_setattr_nonsize( 651 struct user_namespace *mnt_userns, 652 struct xfs_inode *ip, 653 struct iattr *iattr) 654 { 655 xfs_mount_t *mp = ip->i_mount; 656 struct inode *inode = VFS_I(ip); 657 int mask = iattr->ia_valid; 658 xfs_trans_t *tp; 659 int error; 660 kuid_t uid = GLOBAL_ROOT_UID; 661 kgid_t gid = GLOBAL_ROOT_GID; 662 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 663 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; 664 665 ASSERT((mask & ATTR_SIZE) == 0); 666 667 /* 668 * If disk quotas is on, we make sure that the dquots do exist on disk, 669 * before we start any other transactions. Trying to do this later 670 * is messy. We don't care to take a readlock to look at the ids 671 * in inode here, because we can't hold it across the trans_reserve. 672 * If the IDs do change before we take the ilock, we're covered 673 * because the i_*dquot fields will get updated anyway. 674 */ 675 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 676 uint qflags = 0; 677 678 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 679 uid = from_vfsuid(mnt_userns, i_user_ns(inode), 680 iattr->ia_vfsuid); 681 qflags |= XFS_QMOPT_UQUOTA; 682 } else { 683 uid = inode->i_uid; 684 } 685 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 686 gid = from_vfsgid(mnt_userns, i_user_ns(inode), 687 iattr->ia_vfsgid); 688 qflags |= XFS_QMOPT_GQUOTA; 689 } else { 690 gid = inode->i_gid; 691 } 692 693 /* 694 * We take a reference when we initialize udqp and gdqp, 695 * so it is important that we never blindly double trip on 696 * the same variable. See xfs_create() for an example. 697 */ 698 ASSERT(udqp == NULL); 699 ASSERT(gdqp == NULL); 700 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, 701 qflags, &udqp, &gdqp, NULL); 702 if (error) 703 return error; 704 } 705 706 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, 707 has_capability_noaudit(current, CAP_FOWNER), &tp); 708 if (error) 709 goto out_dqrele; 710 711 /* 712 * Register quota modifications in the transaction. Must be the owner 713 * or privileged. These IDs could have changed since we last looked at 714 * them. But, we're assured that if the ownership did change while we 715 * didn't have the inode locked, inode's dquot(s) would have changed 716 * also. 717 */ 718 if (XFS_IS_UQUOTA_ON(mp) && 719 i_uid_needs_update(mnt_userns, iattr, inode)) { 720 ASSERT(udqp); 721 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); 722 } 723 if (XFS_IS_GQUOTA_ON(mp) && 724 i_gid_needs_update(mnt_userns, iattr, inode)) { 725 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp)); 726 ASSERT(gdqp); 727 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp); 728 } 729 730 setattr_copy(mnt_userns, inode, iattr); 731 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 732 733 XFS_STATS_INC(mp, xs_ig_attrchg); 734 735 if (xfs_has_wsync(mp)) 736 xfs_trans_set_sync(tp); 737 error = xfs_trans_commit(tp); 738 739 /* 740 * Release any dquot(s) the inode had kept before chown. 741 */ 742 xfs_qm_dqrele(old_udqp); 743 xfs_qm_dqrele(old_gdqp); 744 xfs_qm_dqrele(udqp); 745 xfs_qm_dqrele(gdqp); 746 747 if (error) 748 return error; 749 750 /* 751 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 752 * update. We could avoid this with linked transactions 753 * and passing down the transaction pointer all the way 754 * to attr_set. No previous user of the generic 755 * Posix ACL code seems to care about this issue either. 756 */ 757 if (mask & ATTR_MODE) { 758 error = posix_acl_chmod(mnt_userns, inode, inode->i_mode); 759 if (error) 760 return error; 761 } 762 763 return 0; 764 765 out_dqrele: 766 xfs_qm_dqrele(udqp); 767 xfs_qm_dqrele(gdqp); 768 return error; 769 } 770 771 /* 772 * Truncate file. Must have write permission and not be a directory. 773 * 774 * Caution: The caller of this function is responsible for calling 775 * setattr_prepare() or otherwise verifying the change is fine. 776 */ 777 STATIC int 778 xfs_setattr_size( 779 struct user_namespace *mnt_userns, 780 struct xfs_inode *ip, 781 struct iattr *iattr) 782 { 783 struct xfs_mount *mp = ip->i_mount; 784 struct inode *inode = VFS_I(ip); 785 xfs_off_t oldsize, newsize; 786 struct xfs_trans *tp; 787 int error; 788 uint lock_flags = 0; 789 bool did_zeroing = false; 790 791 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 792 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 793 ASSERT(S_ISREG(inode->i_mode)); 794 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 795 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); 796 797 oldsize = inode->i_size; 798 newsize = iattr->ia_size; 799 800 /* 801 * Short circuit the truncate case for zero length files. 802 */ 803 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { 804 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 805 return 0; 806 807 /* 808 * Use the regular setattr path to update the timestamps. 809 */ 810 iattr->ia_valid &= ~ATTR_SIZE; 811 return xfs_setattr_nonsize(mnt_userns, ip, iattr); 812 } 813 814 /* 815 * Make sure that the dquots are attached to the inode. 816 */ 817 error = xfs_qm_dqattach(ip); 818 if (error) 819 return error; 820 821 /* 822 * Wait for all direct I/O to complete. 823 */ 824 inode_dio_wait(inode); 825 826 /* 827 * File data changes must be complete before we start the transaction to 828 * modify the inode. This needs to be done before joining the inode to 829 * the transaction because the inode cannot be unlocked once it is a 830 * part of the transaction. 831 * 832 * Start with zeroing any data beyond EOF that we may expose on file 833 * extension, or zeroing out the rest of the block on a downward 834 * truncate. 835 */ 836 if (newsize > oldsize) { 837 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 838 error = xfs_zero_range(ip, oldsize, newsize - oldsize, 839 &did_zeroing); 840 } else { 841 /* 842 * iomap won't detect a dirty page over an unwritten block (or a 843 * cow block over a hole) and subsequently skips zeroing the 844 * newly post-EOF portion of the page. Flush the new EOF to 845 * convert the block before the pagecache truncate. 846 */ 847 error = filemap_write_and_wait_range(inode->i_mapping, newsize, 848 newsize); 849 if (error) 850 return error; 851 error = xfs_truncate_page(ip, newsize, &did_zeroing); 852 } 853 854 if (error) 855 return error; 856 857 /* 858 * We've already locked out new page faults, so now we can safely remove 859 * pages from the page cache knowing they won't get refaulted until we 860 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 861 * complete. The truncate_setsize() call also cleans partial EOF page 862 * PTEs on extending truncates and hence ensures sub-page block size 863 * filesystems are correctly handled, too. 864 * 865 * We have to do all the page cache truncate work outside the 866 * transaction context as the "lock" order is page lock->log space 867 * reservation as defined by extent allocation in the writeback path. 868 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 869 * having already truncated the in-memory version of the file (i.e. made 870 * user visible changes). There's not much we can do about this, except 871 * to hope that the caller sees ENOMEM and retries the truncate 872 * operation. 873 * 874 * And we update in-core i_size and truncate page cache beyond newsize 875 * before writeback the [i_disk_size, newsize] range, so we're 876 * guaranteed not to write stale data past the new EOF on truncate down. 877 */ 878 truncate_setsize(inode, newsize); 879 880 /* 881 * We are going to log the inode size change in this transaction so 882 * any previous writes that are beyond the on disk EOF and the new 883 * EOF that have not been written out need to be written here. If we 884 * do not write the data out, we expose ourselves to the null files 885 * problem. Note that this includes any block zeroing we did above; 886 * otherwise those blocks may not be zeroed after a crash. 887 */ 888 if (did_zeroing || 889 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { 890 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 891 ip->i_disk_size, newsize - 1); 892 if (error) 893 return error; 894 } 895 896 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 897 if (error) 898 return error; 899 900 lock_flags |= XFS_ILOCK_EXCL; 901 xfs_ilock(ip, XFS_ILOCK_EXCL); 902 xfs_trans_ijoin(tp, ip, 0); 903 904 /* 905 * Only change the c/mtime if we are changing the size or we are 906 * explicitly asked to change it. This handles the semantic difference 907 * between truncate() and ftruncate() as implemented in the VFS. 908 * 909 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 910 * special case where we need to update the times despite not having 911 * these flags set. For all other operations the VFS set these flags 912 * explicitly if it wants a timestamp update. 913 */ 914 if (newsize != oldsize && 915 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 916 iattr->ia_ctime = iattr->ia_mtime = 917 current_time(inode); 918 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 919 } 920 921 /* 922 * The first thing we do is set the size to new_size permanently on 923 * disk. This way we don't have to worry about anyone ever being able 924 * to look at the data being freed even in the face of a crash. 925 * What we're getting around here is the case where we free a block, it 926 * is allocated to another file, it is written to, and then we crash. 927 * If the new data gets written to the file but the log buffers 928 * containing the free and reallocation don't, then we'd end up with 929 * garbage in the blocks being freed. As long as we make the new size 930 * permanent before actually freeing any blocks it doesn't matter if 931 * they get written to. 932 */ 933 ip->i_disk_size = newsize; 934 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 935 936 if (newsize <= oldsize) { 937 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 938 if (error) 939 goto out_trans_cancel; 940 941 /* 942 * Truncated "down", so we're removing references to old data 943 * here - if we delay flushing for a long time, we expose 944 * ourselves unduly to the notorious NULL files problem. So, 945 * we mark this inode and flush it when the file is closed, 946 * and do not wait the usual (long) time for writeout. 947 */ 948 xfs_iflags_set(ip, XFS_ITRUNCATED); 949 950 /* A truncate down always removes post-EOF blocks. */ 951 xfs_inode_clear_eofblocks_tag(ip); 952 } 953 954 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID))); 955 setattr_copy(mnt_userns, inode, iattr); 956 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 957 958 XFS_STATS_INC(mp, xs_ig_attrchg); 959 960 if (xfs_has_wsync(mp)) 961 xfs_trans_set_sync(tp); 962 963 error = xfs_trans_commit(tp); 964 out_unlock: 965 if (lock_flags) 966 xfs_iunlock(ip, lock_flags); 967 return error; 968 969 out_trans_cancel: 970 xfs_trans_cancel(tp); 971 goto out_unlock; 972 } 973 974 int 975 xfs_vn_setattr_size( 976 struct user_namespace *mnt_userns, 977 struct dentry *dentry, 978 struct iattr *iattr) 979 { 980 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 981 int error; 982 983 trace_xfs_setattr(ip); 984 985 error = xfs_vn_change_ok(mnt_userns, dentry, iattr); 986 if (error) 987 return error; 988 return xfs_setattr_size(mnt_userns, ip, iattr); 989 } 990 991 STATIC int 992 xfs_vn_setattr( 993 struct user_namespace *mnt_userns, 994 struct dentry *dentry, 995 struct iattr *iattr) 996 { 997 struct inode *inode = d_inode(dentry); 998 struct xfs_inode *ip = XFS_I(inode); 999 int error; 1000 1001 if (iattr->ia_valid & ATTR_SIZE) { 1002 uint iolock; 1003 1004 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1005 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1006 1007 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1008 if (error) { 1009 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1010 return error; 1011 } 1012 1013 error = xfs_vn_setattr_size(mnt_userns, dentry, iattr); 1014 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1015 } else { 1016 trace_xfs_setattr(ip); 1017 1018 error = xfs_vn_change_ok(mnt_userns, dentry, iattr); 1019 if (!error) 1020 error = xfs_setattr_nonsize(mnt_userns, ip, iattr); 1021 } 1022 1023 return error; 1024 } 1025 1026 STATIC int 1027 xfs_vn_update_time( 1028 struct inode *inode, 1029 struct timespec64 *now, 1030 int flags) 1031 { 1032 struct xfs_inode *ip = XFS_I(inode); 1033 struct xfs_mount *mp = ip->i_mount; 1034 int log_flags = XFS_ILOG_TIMESTAMP; 1035 struct xfs_trans *tp; 1036 int error; 1037 1038 trace_xfs_update_time(ip); 1039 1040 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1041 if (!((flags & S_VERSION) && 1042 inode_maybe_inc_iversion(inode, false))) 1043 return generic_update_time(inode, now, flags); 1044 1045 /* Capture the iversion update that just occurred */ 1046 log_flags |= XFS_ILOG_CORE; 1047 } 1048 1049 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1050 if (error) 1051 return error; 1052 1053 xfs_ilock(ip, XFS_ILOCK_EXCL); 1054 if (flags & S_CTIME) 1055 inode->i_ctime = *now; 1056 if (flags & S_MTIME) 1057 inode->i_mtime = *now; 1058 if (flags & S_ATIME) 1059 inode->i_atime = *now; 1060 1061 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1062 xfs_trans_log_inode(tp, ip, log_flags); 1063 return xfs_trans_commit(tp); 1064 } 1065 1066 STATIC int 1067 xfs_vn_fiemap( 1068 struct inode *inode, 1069 struct fiemap_extent_info *fieinfo, 1070 u64 start, 1071 u64 length) 1072 { 1073 int error; 1074 1075 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1076 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1077 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1078 error = iomap_fiemap(inode, fieinfo, start, length, 1079 &xfs_xattr_iomap_ops); 1080 } else { 1081 error = iomap_fiemap(inode, fieinfo, start, length, 1082 &xfs_read_iomap_ops); 1083 } 1084 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1085 1086 return error; 1087 } 1088 1089 STATIC int 1090 xfs_vn_tmpfile( 1091 struct user_namespace *mnt_userns, 1092 struct inode *dir, 1093 struct dentry *dentry, 1094 umode_t mode) 1095 { 1096 return xfs_generic_create(mnt_userns, dir, dentry, mode, 0, true); 1097 } 1098 1099 static const struct inode_operations xfs_inode_operations = { 1100 .get_acl = xfs_get_acl, 1101 .set_acl = xfs_set_acl, 1102 .getattr = xfs_vn_getattr, 1103 .setattr = xfs_vn_setattr, 1104 .listxattr = xfs_vn_listxattr, 1105 .fiemap = xfs_vn_fiemap, 1106 .update_time = xfs_vn_update_time, 1107 .fileattr_get = xfs_fileattr_get, 1108 .fileattr_set = xfs_fileattr_set, 1109 }; 1110 1111 static const struct inode_operations xfs_dir_inode_operations = { 1112 .create = xfs_vn_create, 1113 .lookup = xfs_vn_lookup, 1114 .link = xfs_vn_link, 1115 .unlink = xfs_vn_unlink, 1116 .symlink = xfs_vn_symlink, 1117 .mkdir = xfs_vn_mkdir, 1118 /* 1119 * Yes, XFS uses the same method for rmdir and unlink. 1120 * 1121 * There are some subtile differences deeper in the code, 1122 * but we use S_ISDIR to check for those. 1123 */ 1124 .rmdir = xfs_vn_unlink, 1125 .mknod = xfs_vn_mknod, 1126 .rename = xfs_vn_rename, 1127 .get_acl = xfs_get_acl, 1128 .set_acl = xfs_set_acl, 1129 .getattr = xfs_vn_getattr, 1130 .setattr = xfs_vn_setattr, 1131 .listxattr = xfs_vn_listxattr, 1132 .update_time = xfs_vn_update_time, 1133 .tmpfile = xfs_vn_tmpfile, 1134 .fileattr_get = xfs_fileattr_get, 1135 .fileattr_set = xfs_fileattr_set, 1136 }; 1137 1138 static const struct inode_operations xfs_dir_ci_inode_operations = { 1139 .create = xfs_vn_create, 1140 .lookup = xfs_vn_ci_lookup, 1141 .link = xfs_vn_link, 1142 .unlink = xfs_vn_unlink, 1143 .symlink = xfs_vn_symlink, 1144 .mkdir = xfs_vn_mkdir, 1145 /* 1146 * Yes, XFS uses the same method for rmdir and unlink. 1147 * 1148 * There are some subtile differences deeper in the code, 1149 * but we use S_ISDIR to check for those. 1150 */ 1151 .rmdir = xfs_vn_unlink, 1152 .mknod = xfs_vn_mknod, 1153 .rename = xfs_vn_rename, 1154 .get_acl = xfs_get_acl, 1155 .set_acl = xfs_set_acl, 1156 .getattr = xfs_vn_getattr, 1157 .setattr = xfs_vn_setattr, 1158 .listxattr = xfs_vn_listxattr, 1159 .update_time = xfs_vn_update_time, 1160 .tmpfile = xfs_vn_tmpfile, 1161 .fileattr_get = xfs_fileattr_get, 1162 .fileattr_set = xfs_fileattr_set, 1163 }; 1164 1165 static const struct inode_operations xfs_symlink_inode_operations = { 1166 .get_link = xfs_vn_get_link, 1167 .getattr = xfs_vn_getattr, 1168 .setattr = xfs_vn_setattr, 1169 .listxattr = xfs_vn_listxattr, 1170 .update_time = xfs_vn_update_time, 1171 }; 1172 1173 /* Figure out if this file actually supports DAX. */ 1174 static bool 1175 xfs_inode_supports_dax( 1176 struct xfs_inode *ip) 1177 { 1178 struct xfs_mount *mp = ip->i_mount; 1179 1180 /* Only supported on regular files. */ 1181 if (!S_ISREG(VFS_I(ip)->i_mode)) 1182 return false; 1183 1184 /* Only supported on non-reflinked files. */ 1185 if (xfs_is_reflink_inode(ip)) 1186 return false; 1187 1188 /* Block size must match page size */ 1189 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1190 return false; 1191 1192 /* Device has to support DAX too. */ 1193 return xfs_inode_buftarg(ip)->bt_daxdev != NULL; 1194 } 1195 1196 static bool 1197 xfs_inode_should_enable_dax( 1198 struct xfs_inode *ip) 1199 { 1200 if (!IS_ENABLED(CONFIG_FS_DAX)) 1201 return false; 1202 if (xfs_has_dax_never(ip->i_mount)) 1203 return false; 1204 if (!xfs_inode_supports_dax(ip)) 1205 return false; 1206 if (xfs_has_dax_always(ip->i_mount)) 1207 return true; 1208 if (ip->i_diflags2 & XFS_DIFLAG2_DAX) 1209 return true; 1210 return false; 1211 } 1212 1213 void 1214 xfs_diflags_to_iflags( 1215 struct xfs_inode *ip, 1216 bool init) 1217 { 1218 struct inode *inode = VFS_I(ip); 1219 unsigned int xflags = xfs_ip2xflags(ip); 1220 unsigned int flags = 0; 1221 1222 ASSERT(!(IS_DAX(inode) && init)); 1223 1224 if (xflags & FS_XFLAG_IMMUTABLE) 1225 flags |= S_IMMUTABLE; 1226 if (xflags & FS_XFLAG_APPEND) 1227 flags |= S_APPEND; 1228 if (xflags & FS_XFLAG_SYNC) 1229 flags |= S_SYNC; 1230 if (xflags & FS_XFLAG_NOATIME) 1231 flags |= S_NOATIME; 1232 if (init && xfs_inode_should_enable_dax(ip)) 1233 flags |= S_DAX; 1234 1235 /* 1236 * S_DAX can only be set during inode initialization and is never set by 1237 * the VFS, so we cannot mask off S_DAX in i_flags. 1238 */ 1239 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); 1240 inode->i_flags |= flags; 1241 } 1242 1243 /* 1244 * Initialize the Linux inode. 1245 * 1246 * When reading existing inodes from disk this is called directly from xfs_iget, 1247 * when creating a new inode it is called from xfs_init_new_inode after setting 1248 * up the inode. These callers have different criteria for clearing XFS_INEW, so 1249 * leave it up to the caller to deal with unlocking the inode appropriately. 1250 */ 1251 void 1252 xfs_setup_inode( 1253 struct xfs_inode *ip) 1254 { 1255 struct inode *inode = &ip->i_vnode; 1256 gfp_t gfp_mask; 1257 1258 inode->i_ino = ip->i_ino; 1259 inode->i_state |= I_NEW; 1260 1261 inode_sb_list_add(inode); 1262 /* make the inode look hashed for the writeback code */ 1263 inode_fake_hash(inode); 1264 1265 i_size_write(inode, ip->i_disk_size); 1266 xfs_diflags_to_iflags(ip, true); 1267 1268 if (S_ISDIR(inode->i_mode)) { 1269 /* 1270 * We set the i_rwsem class here to avoid potential races with 1271 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1272 * after a filehandle lookup has already found the inode in 1273 * cache before it has been unlocked via unlock_new_inode(). 1274 */ 1275 lockdep_set_class(&inode->i_rwsem, 1276 &inode->i_sb->s_type->i_mutex_dir_key); 1277 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); 1278 } else { 1279 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); 1280 } 1281 1282 /* 1283 * Ensure all page cache allocations are done from GFP_NOFS context to 1284 * prevent direct reclaim recursion back into the filesystem and blowing 1285 * stacks or deadlocking. 1286 */ 1287 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1288 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1289 1290 /* 1291 * If there is no attribute fork no ACL can exist on this inode, 1292 * and it can't have any file capabilities attached to it either. 1293 */ 1294 if (!xfs_inode_has_attr_fork(ip)) { 1295 inode_has_no_xattr(inode); 1296 cache_no_acl(inode); 1297 } 1298 } 1299 1300 void 1301 xfs_setup_iops( 1302 struct xfs_inode *ip) 1303 { 1304 struct inode *inode = &ip->i_vnode; 1305 1306 switch (inode->i_mode & S_IFMT) { 1307 case S_IFREG: 1308 inode->i_op = &xfs_inode_operations; 1309 inode->i_fop = &xfs_file_operations; 1310 if (IS_DAX(inode)) 1311 inode->i_mapping->a_ops = &xfs_dax_aops; 1312 else 1313 inode->i_mapping->a_ops = &xfs_address_space_operations; 1314 break; 1315 case S_IFDIR: 1316 if (xfs_has_asciici(XFS_M(inode->i_sb))) 1317 inode->i_op = &xfs_dir_ci_inode_operations; 1318 else 1319 inode->i_op = &xfs_dir_inode_operations; 1320 inode->i_fop = &xfs_dir_file_operations; 1321 break; 1322 case S_IFLNK: 1323 inode->i_op = &xfs_symlink_inode_operations; 1324 break; 1325 default: 1326 inode->i_op = &xfs_inode_operations; 1327 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1328 break; 1329 } 1330 } 1331