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 struct file *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(tmpfile, 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, NULL); 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, NULL); 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 NULL); 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 vfsuid_t vfsuid = i_uid_into_vfsuid(mnt_userns, inode); 562 vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode); 563 564 trace_xfs_getattr(ip); 565 566 if (xfs_is_shutdown(mp)) 567 return -EIO; 568 569 stat->size = XFS_ISIZE(ip); 570 stat->dev = inode->i_sb->s_dev; 571 stat->mode = inode->i_mode; 572 stat->nlink = inode->i_nlink; 573 stat->uid = vfsuid_into_kuid(vfsuid); 574 stat->gid = vfsgid_into_kgid(vfsgid); 575 stat->ino = ip->i_ino; 576 stat->atime = inode->i_atime; 577 stat->mtime = inode->i_mtime; 578 stat->ctime = inode->i_ctime; 579 stat->blocks = XFS_FSB_TO_BB(mp, ip->i_nblocks + ip->i_delayed_blks); 580 581 if (xfs_has_v3inodes(mp)) { 582 if (request_mask & STATX_BTIME) { 583 stat->result_mask |= STATX_BTIME; 584 stat->btime = ip->i_crtime; 585 } 586 } 587 588 /* 589 * Note: If you add another clause to set an attribute flag, please 590 * update attributes_mask below. 591 */ 592 if (ip->i_diflags & XFS_DIFLAG_IMMUTABLE) 593 stat->attributes |= STATX_ATTR_IMMUTABLE; 594 if (ip->i_diflags & XFS_DIFLAG_APPEND) 595 stat->attributes |= STATX_ATTR_APPEND; 596 if (ip->i_diflags & XFS_DIFLAG_NODUMP) 597 stat->attributes |= STATX_ATTR_NODUMP; 598 599 stat->attributes_mask |= (STATX_ATTR_IMMUTABLE | 600 STATX_ATTR_APPEND | 601 STATX_ATTR_NODUMP); 602 603 switch (inode->i_mode & S_IFMT) { 604 case S_IFBLK: 605 case S_IFCHR: 606 stat->blksize = BLKDEV_IOSIZE; 607 stat->rdev = inode->i_rdev; 608 break; 609 case S_IFREG: 610 if (request_mask & STATX_DIOALIGN) { 611 struct xfs_buftarg *target = xfs_inode_buftarg(ip); 612 struct block_device *bdev = target->bt_bdev; 613 614 stat->result_mask |= STATX_DIOALIGN; 615 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; 616 stat->dio_offset_align = bdev_logical_block_size(bdev); 617 } 618 fallthrough; 619 default: 620 stat->blksize = xfs_stat_blksize(ip); 621 stat->rdev = 0; 622 break; 623 } 624 625 return 0; 626 } 627 628 static int 629 xfs_vn_change_ok( 630 struct user_namespace *mnt_userns, 631 struct dentry *dentry, 632 struct iattr *iattr) 633 { 634 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 635 636 if (xfs_is_readonly(mp)) 637 return -EROFS; 638 639 if (xfs_is_shutdown(mp)) 640 return -EIO; 641 642 return setattr_prepare(mnt_userns, dentry, iattr); 643 } 644 645 /* 646 * Set non-size attributes of an inode. 647 * 648 * Caution: The caller of this function is responsible for calling 649 * setattr_prepare() or otherwise verifying the change is fine. 650 */ 651 static int 652 xfs_setattr_nonsize( 653 struct user_namespace *mnt_userns, 654 struct dentry *dentry, 655 struct xfs_inode *ip, 656 struct iattr *iattr) 657 { 658 xfs_mount_t *mp = ip->i_mount; 659 struct inode *inode = VFS_I(ip); 660 int mask = iattr->ia_valid; 661 xfs_trans_t *tp; 662 int error; 663 kuid_t uid = GLOBAL_ROOT_UID; 664 kgid_t gid = GLOBAL_ROOT_GID; 665 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 666 struct xfs_dquot *old_udqp = NULL, *old_gdqp = NULL; 667 668 ASSERT((mask & ATTR_SIZE) == 0); 669 670 /* 671 * If disk quotas is on, we make sure that the dquots do exist on disk, 672 * before we start any other transactions. Trying to do this later 673 * is messy. We don't care to take a readlock to look at the ids 674 * in inode here, because we can't hold it across the trans_reserve. 675 * If the IDs do change before we take the ilock, we're covered 676 * because the i_*dquot fields will get updated anyway. 677 */ 678 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 679 uint qflags = 0; 680 681 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 682 uid = from_vfsuid(mnt_userns, i_user_ns(inode), 683 iattr->ia_vfsuid); 684 qflags |= XFS_QMOPT_UQUOTA; 685 } else { 686 uid = inode->i_uid; 687 } 688 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 689 gid = from_vfsgid(mnt_userns, i_user_ns(inode), 690 iattr->ia_vfsgid); 691 qflags |= XFS_QMOPT_GQUOTA; 692 } else { 693 gid = inode->i_gid; 694 } 695 696 /* 697 * We take a reference when we initialize udqp and gdqp, 698 * so it is important that we never blindly double trip on 699 * the same variable. See xfs_create() for an example. 700 */ 701 ASSERT(udqp == NULL); 702 ASSERT(gdqp == NULL); 703 error = xfs_qm_vop_dqalloc(ip, uid, gid, ip->i_projid, 704 qflags, &udqp, &gdqp, NULL); 705 if (error) 706 return error; 707 } 708 709 error = xfs_trans_alloc_ichange(ip, udqp, gdqp, NULL, 710 has_capability_noaudit(current, CAP_FOWNER), &tp); 711 if (error) 712 goto out_dqrele; 713 714 /* 715 * Register quota modifications in the transaction. Must be the owner 716 * or privileged. These IDs could have changed since we last looked at 717 * them. But, we're assured that if the ownership did change while we 718 * didn't have the inode locked, inode's dquot(s) would have changed 719 * also. 720 */ 721 if (XFS_IS_UQUOTA_ON(mp) && 722 i_uid_needs_update(mnt_userns, iattr, inode)) { 723 ASSERT(udqp); 724 old_udqp = xfs_qm_vop_chown(tp, ip, &ip->i_udquot, udqp); 725 } 726 if (XFS_IS_GQUOTA_ON(mp) && 727 i_gid_needs_update(mnt_userns, iattr, inode)) { 728 ASSERT(xfs_has_pquotino(mp) || !XFS_IS_PQUOTA_ON(mp)); 729 ASSERT(gdqp); 730 old_gdqp = xfs_qm_vop_chown(tp, ip, &ip->i_gdquot, gdqp); 731 } 732 733 setattr_copy(mnt_userns, inode, iattr); 734 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 735 736 XFS_STATS_INC(mp, xs_ig_attrchg); 737 738 if (xfs_has_wsync(mp)) 739 xfs_trans_set_sync(tp); 740 error = xfs_trans_commit(tp); 741 742 /* 743 * Release any dquot(s) the inode had kept before chown. 744 */ 745 xfs_qm_dqrele(old_udqp); 746 xfs_qm_dqrele(old_gdqp); 747 xfs_qm_dqrele(udqp); 748 xfs_qm_dqrele(gdqp); 749 750 if (error) 751 return error; 752 753 /* 754 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 755 * update. We could avoid this with linked transactions 756 * and passing down the transaction pointer all the way 757 * to attr_set. No previous user of the generic 758 * Posix ACL code seems to care about this issue either. 759 */ 760 if (mask & ATTR_MODE) { 761 error = posix_acl_chmod(mnt_userns, dentry, inode->i_mode); 762 if (error) 763 return error; 764 } 765 766 return 0; 767 768 out_dqrele: 769 xfs_qm_dqrele(udqp); 770 xfs_qm_dqrele(gdqp); 771 return error; 772 } 773 774 /* 775 * Truncate file. Must have write permission and not be a directory. 776 * 777 * Caution: The caller of this function is responsible for calling 778 * setattr_prepare() or otherwise verifying the change is fine. 779 */ 780 STATIC int 781 xfs_setattr_size( 782 struct user_namespace *mnt_userns, 783 struct dentry *dentry, 784 struct xfs_inode *ip, 785 struct iattr *iattr) 786 { 787 struct xfs_mount *mp = ip->i_mount; 788 struct inode *inode = VFS_I(ip); 789 xfs_off_t oldsize, newsize; 790 struct xfs_trans *tp; 791 int error; 792 uint lock_flags = 0; 793 bool did_zeroing = false; 794 795 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 796 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 797 ASSERT(S_ISREG(inode->i_mode)); 798 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 799 ATTR_MTIME_SET|ATTR_TIMES_SET)) == 0); 800 801 oldsize = inode->i_size; 802 newsize = iattr->ia_size; 803 804 /* 805 * Short circuit the truncate case for zero length files. 806 */ 807 if (newsize == 0 && oldsize == 0 && ip->i_df.if_nextents == 0) { 808 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 809 return 0; 810 811 /* 812 * Use the regular setattr path to update the timestamps. 813 */ 814 iattr->ia_valid &= ~ATTR_SIZE; 815 return xfs_setattr_nonsize(mnt_userns, dentry, ip, iattr); 816 } 817 818 /* 819 * Make sure that the dquots are attached to the inode. 820 */ 821 error = xfs_qm_dqattach(ip); 822 if (error) 823 return error; 824 825 /* 826 * Wait for all direct I/O to complete. 827 */ 828 inode_dio_wait(inode); 829 830 /* 831 * File data changes must be complete before we start the transaction to 832 * modify the inode. This needs to be done before joining the inode to 833 * the transaction because the inode cannot be unlocked once it is a 834 * part of the transaction. 835 * 836 * Start with zeroing any data beyond EOF that we may expose on file 837 * extension, or zeroing out the rest of the block on a downward 838 * truncate. 839 */ 840 if (newsize > oldsize) { 841 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 842 error = xfs_zero_range(ip, oldsize, newsize - oldsize, 843 &did_zeroing); 844 } else { 845 /* 846 * iomap won't detect a dirty page over an unwritten block (or a 847 * cow block over a hole) and subsequently skips zeroing the 848 * newly post-EOF portion of the page. Flush the new EOF to 849 * convert the block before the pagecache truncate. 850 */ 851 error = filemap_write_and_wait_range(inode->i_mapping, newsize, 852 newsize); 853 if (error) 854 return error; 855 error = xfs_truncate_page(ip, newsize, &did_zeroing); 856 } 857 858 if (error) 859 return error; 860 861 /* 862 * We've already locked out new page faults, so now we can safely remove 863 * pages from the page cache knowing they won't get refaulted until we 864 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 865 * complete. The truncate_setsize() call also cleans partial EOF page 866 * PTEs on extending truncates and hence ensures sub-page block size 867 * filesystems are correctly handled, too. 868 * 869 * We have to do all the page cache truncate work outside the 870 * transaction context as the "lock" order is page lock->log space 871 * reservation as defined by extent allocation in the writeback path. 872 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 873 * having already truncated the in-memory version of the file (i.e. made 874 * user visible changes). There's not much we can do about this, except 875 * to hope that the caller sees ENOMEM and retries the truncate 876 * operation. 877 * 878 * And we update in-core i_size and truncate page cache beyond newsize 879 * before writeback the [i_disk_size, newsize] range, so we're 880 * guaranteed not to write stale data past the new EOF on truncate down. 881 */ 882 truncate_setsize(inode, newsize); 883 884 /* 885 * We are going to log the inode size change in this transaction so 886 * any previous writes that are beyond the on disk EOF and the new 887 * EOF that have not been written out need to be written here. If we 888 * do not write the data out, we expose ourselves to the null files 889 * problem. Note that this includes any block zeroing we did above; 890 * otherwise those blocks may not be zeroed after a crash. 891 */ 892 if (did_zeroing || 893 (newsize > ip->i_disk_size && oldsize != ip->i_disk_size)) { 894 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 895 ip->i_disk_size, newsize - 1); 896 if (error) 897 return error; 898 } 899 900 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 901 if (error) 902 return error; 903 904 lock_flags |= XFS_ILOCK_EXCL; 905 xfs_ilock(ip, XFS_ILOCK_EXCL); 906 xfs_trans_ijoin(tp, ip, 0); 907 908 /* 909 * Only change the c/mtime if we are changing the size or we are 910 * explicitly asked to change it. This handles the semantic difference 911 * between truncate() and ftruncate() as implemented in the VFS. 912 * 913 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 914 * special case where we need to update the times despite not having 915 * these flags set. For all other operations the VFS set these flags 916 * explicitly if it wants a timestamp update. 917 */ 918 if (newsize != oldsize && 919 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 920 iattr->ia_ctime = iattr->ia_mtime = 921 current_time(inode); 922 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 923 } 924 925 /* 926 * The first thing we do is set the size to new_size permanently on 927 * disk. This way we don't have to worry about anyone ever being able 928 * to look at the data being freed even in the face of a crash. 929 * What we're getting around here is the case where we free a block, it 930 * is allocated to another file, it is written to, and then we crash. 931 * If the new data gets written to the file but the log buffers 932 * containing the free and reallocation don't, then we'd end up with 933 * garbage in the blocks being freed. As long as we make the new size 934 * permanent before actually freeing any blocks it doesn't matter if 935 * they get written to. 936 */ 937 ip->i_disk_size = newsize; 938 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 939 940 if (newsize <= oldsize) { 941 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 942 if (error) 943 goto out_trans_cancel; 944 945 /* 946 * Truncated "down", so we're removing references to old data 947 * here - if we delay flushing for a long time, we expose 948 * ourselves unduly to the notorious NULL files problem. So, 949 * we mark this inode and flush it when the file is closed, 950 * and do not wait the usual (long) time for writeout. 951 */ 952 xfs_iflags_set(ip, XFS_ITRUNCATED); 953 954 /* A truncate down always removes post-EOF blocks. */ 955 xfs_inode_clear_eofblocks_tag(ip); 956 } 957 958 ASSERT(!(iattr->ia_valid & (ATTR_UID | ATTR_GID))); 959 setattr_copy(mnt_userns, inode, iattr); 960 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 961 962 XFS_STATS_INC(mp, xs_ig_attrchg); 963 964 if (xfs_has_wsync(mp)) 965 xfs_trans_set_sync(tp); 966 967 error = xfs_trans_commit(tp); 968 out_unlock: 969 if (lock_flags) 970 xfs_iunlock(ip, lock_flags); 971 return error; 972 973 out_trans_cancel: 974 xfs_trans_cancel(tp); 975 goto out_unlock; 976 } 977 978 int 979 xfs_vn_setattr_size( 980 struct user_namespace *mnt_userns, 981 struct dentry *dentry, 982 struct iattr *iattr) 983 { 984 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 985 int error; 986 987 trace_xfs_setattr(ip); 988 989 error = xfs_vn_change_ok(mnt_userns, dentry, iattr); 990 if (error) 991 return error; 992 return xfs_setattr_size(mnt_userns, dentry, ip, iattr); 993 } 994 995 STATIC int 996 xfs_vn_setattr( 997 struct user_namespace *mnt_userns, 998 struct dentry *dentry, 999 struct iattr *iattr) 1000 { 1001 struct inode *inode = d_inode(dentry); 1002 struct xfs_inode *ip = XFS_I(inode); 1003 int error; 1004 1005 if (iattr->ia_valid & ATTR_SIZE) { 1006 uint iolock; 1007 1008 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1009 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1010 1011 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1012 if (error) { 1013 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1014 return error; 1015 } 1016 1017 error = xfs_vn_setattr_size(mnt_userns, dentry, iattr); 1018 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1019 } else { 1020 trace_xfs_setattr(ip); 1021 1022 error = xfs_vn_change_ok(mnt_userns, dentry, iattr); 1023 if (!error) 1024 error = xfs_setattr_nonsize(mnt_userns, dentry, ip, iattr); 1025 } 1026 1027 return error; 1028 } 1029 1030 STATIC int 1031 xfs_vn_update_time( 1032 struct inode *inode, 1033 struct timespec64 *now, 1034 int flags) 1035 { 1036 struct xfs_inode *ip = XFS_I(inode); 1037 struct xfs_mount *mp = ip->i_mount; 1038 int log_flags = XFS_ILOG_TIMESTAMP; 1039 struct xfs_trans *tp; 1040 int error; 1041 1042 trace_xfs_update_time(ip); 1043 1044 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1045 if (!((flags & S_VERSION) && 1046 inode_maybe_inc_iversion(inode, false))) 1047 return generic_update_time(inode, now, flags); 1048 1049 /* Capture the iversion update that just occurred */ 1050 log_flags |= XFS_ILOG_CORE; 1051 } 1052 1053 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1054 if (error) 1055 return error; 1056 1057 xfs_ilock(ip, XFS_ILOCK_EXCL); 1058 if (flags & S_CTIME) 1059 inode->i_ctime = *now; 1060 if (flags & S_MTIME) 1061 inode->i_mtime = *now; 1062 if (flags & S_ATIME) 1063 inode->i_atime = *now; 1064 1065 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1066 xfs_trans_log_inode(tp, ip, log_flags); 1067 return xfs_trans_commit(tp); 1068 } 1069 1070 STATIC int 1071 xfs_vn_fiemap( 1072 struct inode *inode, 1073 struct fiemap_extent_info *fieinfo, 1074 u64 start, 1075 u64 length) 1076 { 1077 int error; 1078 1079 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1080 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1081 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1082 error = iomap_fiemap(inode, fieinfo, start, length, 1083 &xfs_xattr_iomap_ops); 1084 } else { 1085 error = iomap_fiemap(inode, fieinfo, start, length, 1086 &xfs_read_iomap_ops); 1087 } 1088 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1089 1090 return error; 1091 } 1092 1093 STATIC int 1094 xfs_vn_tmpfile( 1095 struct user_namespace *mnt_userns, 1096 struct inode *dir, 1097 struct file *file, 1098 umode_t mode) 1099 { 1100 int err = xfs_generic_create(mnt_userns, dir, file->f_path.dentry, mode, 0, file); 1101 1102 return finish_open_simple(file, err); 1103 } 1104 1105 static const struct inode_operations xfs_inode_operations = { 1106 .get_inode_acl = xfs_get_acl, 1107 .set_acl = xfs_set_acl, 1108 .getattr = xfs_vn_getattr, 1109 .setattr = xfs_vn_setattr, 1110 .listxattr = xfs_vn_listxattr, 1111 .fiemap = xfs_vn_fiemap, 1112 .update_time = xfs_vn_update_time, 1113 .fileattr_get = xfs_fileattr_get, 1114 .fileattr_set = xfs_fileattr_set, 1115 }; 1116 1117 static const struct inode_operations xfs_dir_inode_operations = { 1118 .create = xfs_vn_create, 1119 .lookup = xfs_vn_lookup, 1120 .link = xfs_vn_link, 1121 .unlink = xfs_vn_unlink, 1122 .symlink = xfs_vn_symlink, 1123 .mkdir = xfs_vn_mkdir, 1124 /* 1125 * Yes, XFS uses the same method for rmdir and unlink. 1126 * 1127 * There are some subtile differences deeper in the code, 1128 * but we use S_ISDIR to check for those. 1129 */ 1130 .rmdir = xfs_vn_unlink, 1131 .mknod = xfs_vn_mknod, 1132 .rename = xfs_vn_rename, 1133 .get_inode_acl = xfs_get_acl, 1134 .set_acl = xfs_set_acl, 1135 .getattr = xfs_vn_getattr, 1136 .setattr = xfs_vn_setattr, 1137 .listxattr = xfs_vn_listxattr, 1138 .update_time = xfs_vn_update_time, 1139 .tmpfile = xfs_vn_tmpfile, 1140 .fileattr_get = xfs_fileattr_get, 1141 .fileattr_set = xfs_fileattr_set, 1142 }; 1143 1144 static const struct inode_operations xfs_dir_ci_inode_operations = { 1145 .create = xfs_vn_create, 1146 .lookup = xfs_vn_ci_lookup, 1147 .link = xfs_vn_link, 1148 .unlink = xfs_vn_unlink, 1149 .symlink = xfs_vn_symlink, 1150 .mkdir = xfs_vn_mkdir, 1151 /* 1152 * Yes, XFS uses the same method for rmdir and unlink. 1153 * 1154 * There are some subtile differences deeper in the code, 1155 * but we use S_ISDIR to check for those. 1156 */ 1157 .rmdir = xfs_vn_unlink, 1158 .mknod = xfs_vn_mknod, 1159 .rename = xfs_vn_rename, 1160 .get_inode_acl = xfs_get_acl, 1161 .set_acl = xfs_set_acl, 1162 .getattr = xfs_vn_getattr, 1163 .setattr = xfs_vn_setattr, 1164 .listxattr = xfs_vn_listxattr, 1165 .update_time = xfs_vn_update_time, 1166 .tmpfile = xfs_vn_tmpfile, 1167 .fileattr_get = xfs_fileattr_get, 1168 .fileattr_set = xfs_fileattr_set, 1169 }; 1170 1171 static const struct inode_operations xfs_symlink_inode_operations = { 1172 .get_link = xfs_vn_get_link, 1173 .getattr = xfs_vn_getattr, 1174 .setattr = xfs_vn_setattr, 1175 .listxattr = xfs_vn_listxattr, 1176 .update_time = xfs_vn_update_time, 1177 }; 1178 1179 /* Figure out if this file actually supports DAX. */ 1180 static bool 1181 xfs_inode_supports_dax( 1182 struct xfs_inode *ip) 1183 { 1184 struct xfs_mount *mp = ip->i_mount; 1185 1186 /* Only supported on regular files. */ 1187 if (!S_ISREG(VFS_I(ip)->i_mode)) 1188 return false; 1189 1190 /* Block size must match page size */ 1191 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1192 return false; 1193 1194 /* Device has to support DAX too. */ 1195 return xfs_inode_buftarg(ip)->bt_daxdev != NULL; 1196 } 1197 1198 static bool 1199 xfs_inode_should_enable_dax( 1200 struct xfs_inode *ip) 1201 { 1202 if (!IS_ENABLED(CONFIG_FS_DAX)) 1203 return false; 1204 if (xfs_has_dax_never(ip->i_mount)) 1205 return false; 1206 if (!xfs_inode_supports_dax(ip)) 1207 return false; 1208 if (xfs_has_dax_always(ip->i_mount)) 1209 return true; 1210 if (ip->i_diflags2 & XFS_DIFLAG2_DAX) 1211 return true; 1212 return false; 1213 } 1214 1215 void 1216 xfs_diflags_to_iflags( 1217 struct xfs_inode *ip, 1218 bool init) 1219 { 1220 struct inode *inode = VFS_I(ip); 1221 unsigned int xflags = xfs_ip2xflags(ip); 1222 unsigned int flags = 0; 1223 1224 ASSERT(!(IS_DAX(inode) && init)); 1225 1226 if (xflags & FS_XFLAG_IMMUTABLE) 1227 flags |= S_IMMUTABLE; 1228 if (xflags & FS_XFLAG_APPEND) 1229 flags |= S_APPEND; 1230 if (xflags & FS_XFLAG_SYNC) 1231 flags |= S_SYNC; 1232 if (xflags & FS_XFLAG_NOATIME) 1233 flags |= S_NOATIME; 1234 if (init && xfs_inode_should_enable_dax(ip)) 1235 flags |= S_DAX; 1236 1237 /* 1238 * S_DAX can only be set during inode initialization and is never set by 1239 * the VFS, so we cannot mask off S_DAX in i_flags. 1240 */ 1241 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | S_NOATIME); 1242 inode->i_flags |= flags; 1243 } 1244 1245 /* 1246 * Initialize the Linux inode. 1247 * 1248 * When reading existing inodes from disk this is called directly from xfs_iget, 1249 * when creating a new inode it is called from xfs_init_new_inode after setting 1250 * up the inode. These callers have different criteria for clearing XFS_INEW, so 1251 * leave it up to the caller to deal with unlocking the inode appropriately. 1252 */ 1253 void 1254 xfs_setup_inode( 1255 struct xfs_inode *ip) 1256 { 1257 struct inode *inode = &ip->i_vnode; 1258 gfp_t gfp_mask; 1259 1260 inode->i_ino = ip->i_ino; 1261 inode->i_state |= I_NEW; 1262 1263 inode_sb_list_add(inode); 1264 /* make the inode look hashed for the writeback code */ 1265 inode_fake_hash(inode); 1266 1267 i_size_write(inode, ip->i_disk_size); 1268 xfs_diflags_to_iflags(ip, true); 1269 1270 if (S_ISDIR(inode->i_mode)) { 1271 /* 1272 * We set the i_rwsem class here to avoid potential races with 1273 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1274 * after a filehandle lookup has already found the inode in 1275 * cache before it has been unlocked via unlock_new_inode(). 1276 */ 1277 lockdep_set_class(&inode->i_rwsem, 1278 &inode->i_sb->s_type->i_mutex_dir_key); 1279 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); 1280 } else { 1281 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); 1282 } 1283 1284 /* 1285 * Ensure all page cache allocations are done from GFP_NOFS context to 1286 * prevent direct reclaim recursion back into the filesystem and blowing 1287 * stacks or deadlocking. 1288 */ 1289 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1290 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1291 1292 /* 1293 * If there is no attribute fork no ACL can exist on this inode, 1294 * and it can't have any file capabilities attached to it either. 1295 */ 1296 if (!xfs_inode_has_attr_fork(ip)) { 1297 inode_has_no_xattr(inode); 1298 cache_no_acl(inode); 1299 } 1300 } 1301 1302 void 1303 xfs_setup_iops( 1304 struct xfs_inode *ip) 1305 { 1306 struct inode *inode = &ip->i_vnode; 1307 1308 switch (inode->i_mode & S_IFMT) { 1309 case S_IFREG: 1310 inode->i_op = &xfs_inode_operations; 1311 inode->i_fop = &xfs_file_operations; 1312 if (IS_DAX(inode)) 1313 inode->i_mapping->a_ops = &xfs_dax_aops; 1314 else 1315 inode->i_mapping->a_ops = &xfs_address_space_operations; 1316 break; 1317 case S_IFDIR: 1318 if (xfs_has_asciici(XFS_M(inode->i_sb))) 1319 inode->i_op = &xfs_dir_ci_inode_operations; 1320 else 1321 inode->i_op = &xfs_dir_inode_operations; 1322 inode->i_fop = &xfs_dir_file_operations; 1323 break; 1324 case S_IFLNK: 1325 inode->i_op = &xfs_symlink_inode_operations; 1326 break; 1327 default: 1328 inode->i_op = &xfs_inode_operations; 1329 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1330 break; 1331 } 1332 } 1333