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