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_da_format.h" 14 #include "xfs_inode.h" 15 #include "xfs_bmap.h" 16 #include "xfs_bmap_util.h" 17 #include "xfs_acl.h" 18 #include "xfs_quota.h" 19 #include "xfs_error.h" 20 #include "xfs_attr.h" 21 #include "xfs_trans.h" 22 #include "xfs_trace.h" 23 #include "xfs_icache.h" 24 #include "xfs_symlink.h" 25 #include "xfs_da_btree.h" 26 #include "xfs_dir2.h" 27 #include "xfs_trans_space.h" 28 #include "xfs_iomap.h" 29 30 #include <linux/capability.h> 31 #include <linux/xattr.h> 32 #include <linux/posix_acl.h> 33 #include <linux/security.h> 34 #include <linux/iomap.h> 35 #include <linux/slab.h> 36 #include <linux/iversion.h> 37 38 /* 39 * Directories have different lock order w.r.t. mmap_sem compared to regular 40 * files. This is due to readdir potentially triggering page faults on a user 41 * buffer inside filldir(), and this happens with the ilock on the directory 42 * held. For regular files, the lock order is the other way around - the 43 * mmap_sem is taken during the page fault, and then we lock the ilock to do 44 * block mapping. Hence we need a different class for the directory ilock so 45 * that lockdep can tell them apart. 46 */ 47 static struct lock_class_key xfs_nondir_ilock_class; 48 static struct lock_class_key xfs_dir_ilock_class; 49 50 static int 51 xfs_initxattrs( 52 struct inode *inode, 53 const struct xattr *xattr_array, 54 void *fs_info) 55 { 56 const struct xattr *xattr; 57 struct xfs_inode *ip = XFS_I(inode); 58 int error = 0; 59 60 for (xattr = xattr_array; xattr->name != NULL; xattr++) { 61 error = xfs_attr_set(ip, xattr->name, xattr->value, 62 xattr->value_len, ATTR_SECURE); 63 if (error < 0) 64 break; 65 } 66 return error; 67 } 68 69 /* 70 * Hook in SELinux. This is not quite correct yet, what we really need 71 * here (as we do for default ACLs) is a mechanism by which creation of 72 * these attrs can be journalled at inode creation time (along with the 73 * inode, of course, such that log replay can't cause these to be lost). 74 */ 75 76 STATIC int 77 xfs_init_security( 78 struct inode *inode, 79 struct inode *dir, 80 const struct qstr *qstr) 81 { 82 return security_inode_init_security(inode, dir, qstr, 83 &xfs_initxattrs, NULL); 84 } 85 86 static void 87 xfs_dentry_to_name( 88 struct xfs_name *namep, 89 struct dentry *dentry) 90 { 91 namep->name = dentry->d_name.name; 92 namep->len = dentry->d_name.len; 93 namep->type = XFS_DIR3_FT_UNKNOWN; 94 } 95 96 static int 97 xfs_dentry_mode_to_name( 98 struct xfs_name *namep, 99 struct dentry *dentry, 100 int mode) 101 { 102 namep->name = dentry->d_name.name; 103 namep->len = dentry->d_name.len; 104 namep->type = xfs_mode_to_ftype(mode); 105 106 if (unlikely(namep->type == XFS_DIR3_FT_UNKNOWN)) 107 return -EFSCORRUPTED; 108 109 return 0; 110 } 111 112 STATIC void 113 xfs_cleanup_inode( 114 struct inode *dir, 115 struct inode *inode, 116 struct dentry *dentry) 117 { 118 struct xfs_name teardown; 119 120 /* Oh, the horror. 121 * If we can't add the ACL or we fail in 122 * xfs_init_security we must back out. 123 * ENOSPC can hit here, among other things. 124 */ 125 xfs_dentry_to_name(&teardown, dentry); 126 127 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); 128 } 129 130 STATIC int 131 xfs_generic_create( 132 struct inode *dir, 133 struct dentry *dentry, 134 umode_t mode, 135 dev_t rdev, 136 bool tmpfile) /* unnamed file */ 137 { 138 struct inode *inode; 139 struct xfs_inode *ip = NULL; 140 struct posix_acl *default_acl, *acl; 141 struct xfs_name name; 142 int error; 143 144 /* 145 * Irix uses Missed'em'V split, but doesn't want to see 146 * the upper 5 bits of (14bit) major. 147 */ 148 if (S_ISCHR(mode) || S_ISBLK(mode)) { 149 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) 150 return -EINVAL; 151 } else { 152 rdev = 0; 153 } 154 155 error = posix_acl_create(dir, &mode, &default_acl, &acl); 156 if (error) 157 return error; 158 159 /* Verify mode is valid also for tmpfile case */ 160 error = xfs_dentry_mode_to_name(&name, dentry, mode); 161 if (unlikely(error)) 162 goto out_free_acl; 163 164 if (!tmpfile) { 165 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip); 166 } else { 167 error = xfs_create_tmpfile(XFS_I(dir), mode, &ip); 168 } 169 if (unlikely(error)) 170 goto out_free_acl; 171 172 inode = VFS_I(ip); 173 174 error = xfs_init_security(inode, dir, &dentry->d_name); 175 if (unlikely(error)) 176 goto out_cleanup_inode; 177 178 #ifdef CONFIG_XFS_POSIX_ACL 179 if (default_acl) { 180 error = __xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); 181 if (error) 182 goto out_cleanup_inode; 183 } 184 if (acl) { 185 error = __xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); 186 if (error) 187 goto out_cleanup_inode; 188 } 189 #endif 190 191 xfs_setup_iops(ip); 192 193 if (tmpfile) 194 d_tmpfile(dentry, inode); 195 else 196 d_instantiate(dentry, inode); 197 198 xfs_finish_inode_setup(ip); 199 200 out_free_acl: 201 if (default_acl) 202 posix_acl_release(default_acl); 203 if (acl) 204 posix_acl_release(acl); 205 return error; 206 207 out_cleanup_inode: 208 xfs_finish_inode_setup(ip); 209 if (!tmpfile) 210 xfs_cleanup_inode(dir, inode, dentry); 211 iput(inode); 212 goto out_free_acl; 213 } 214 215 STATIC int 216 xfs_vn_mknod( 217 struct inode *dir, 218 struct dentry *dentry, 219 umode_t mode, 220 dev_t rdev) 221 { 222 return xfs_generic_create(dir, dentry, mode, rdev, false); 223 } 224 225 STATIC int 226 xfs_vn_create( 227 struct inode *dir, 228 struct dentry *dentry, 229 umode_t mode, 230 bool flags) 231 { 232 return xfs_vn_mknod(dir, dentry, mode, 0); 233 } 234 235 STATIC int 236 xfs_vn_mkdir( 237 struct inode *dir, 238 struct dentry *dentry, 239 umode_t mode) 240 { 241 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0); 242 } 243 244 STATIC struct dentry * 245 xfs_vn_lookup( 246 struct inode *dir, 247 struct dentry *dentry, 248 unsigned int flags) 249 { 250 struct inode *inode; 251 struct xfs_inode *cip; 252 struct xfs_name name; 253 int error; 254 255 if (dentry->d_name.len >= MAXNAMELEN) 256 return ERR_PTR(-ENAMETOOLONG); 257 258 xfs_dentry_to_name(&name, dentry); 259 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); 260 if (likely(!error)) 261 inode = VFS_I(cip); 262 else if (likely(error == -ENOENT)) 263 inode = NULL; 264 else 265 inode = ERR_PTR(error); 266 return d_splice_alias(inode, dentry); 267 } 268 269 STATIC struct dentry * 270 xfs_vn_ci_lookup( 271 struct inode *dir, 272 struct dentry *dentry, 273 unsigned int flags) 274 { 275 struct xfs_inode *ip; 276 struct xfs_name xname; 277 struct xfs_name ci_name; 278 struct qstr dname; 279 int error; 280 281 if (dentry->d_name.len >= MAXNAMELEN) 282 return ERR_PTR(-ENAMETOOLONG); 283 284 xfs_dentry_to_name(&xname, dentry); 285 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); 286 if (unlikely(error)) { 287 if (unlikely(error != -ENOENT)) 288 return ERR_PTR(error); 289 /* 290 * call d_add(dentry, NULL) here when d_drop_negative_children 291 * is called in xfs_vn_mknod (ie. allow negative dentries 292 * with CI filesystems). 293 */ 294 return NULL; 295 } 296 297 /* if exact match, just splice and exit */ 298 if (!ci_name.name) 299 return d_splice_alias(VFS_I(ip), dentry); 300 301 /* else case-insensitive match... */ 302 dname.name = ci_name.name; 303 dname.len = ci_name.len; 304 dentry = d_add_ci(dentry, VFS_I(ip), &dname); 305 kmem_free(ci_name.name); 306 return dentry; 307 } 308 309 STATIC int 310 xfs_vn_link( 311 struct dentry *old_dentry, 312 struct inode *dir, 313 struct dentry *dentry) 314 { 315 struct inode *inode = d_inode(old_dentry); 316 struct xfs_name name; 317 int error; 318 319 error = xfs_dentry_mode_to_name(&name, dentry, inode->i_mode); 320 if (unlikely(error)) 321 return error; 322 323 error = xfs_link(XFS_I(dir), XFS_I(inode), &name); 324 if (unlikely(error)) 325 return error; 326 327 ihold(inode); 328 d_instantiate(dentry, inode); 329 return 0; 330 } 331 332 STATIC int 333 xfs_vn_unlink( 334 struct inode *dir, 335 struct dentry *dentry) 336 { 337 struct xfs_name name; 338 int error; 339 340 xfs_dentry_to_name(&name, dentry); 341 342 error = xfs_remove(XFS_I(dir), &name, XFS_I(d_inode(dentry))); 343 if (error) 344 return error; 345 346 /* 347 * With unlink, the VFS makes the dentry "negative": no inode, 348 * but still hashed. This is incompatible with case-insensitive 349 * mode, so invalidate (unhash) the dentry in CI-mode. 350 */ 351 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb)) 352 d_invalidate(dentry); 353 return 0; 354 } 355 356 STATIC int 357 xfs_vn_symlink( 358 struct inode *dir, 359 struct dentry *dentry, 360 const char *symname) 361 { 362 struct inode *inode; 363 struct xfs_inode *cip = NULL; 364 struct xfs_name name; 365 int error; 366 umode_t mode; 367 368 mode = S_IFLNK | 369 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); 370 error = xfs_dentry_mode_to_name(&name, dentry, mode); 371 if (unlikely(error)) 372 goto out; 373 374 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip); 375 if (unlikely(error)) 376 goto out; 377 378 inode = VFS_I(cip); 379 380 error = xfs_init_security(inode, dir, &dentry->d_name); 381 if (unlikely(error)) 382 goto out_cleanup_inode; 383 384 xfs_setup_iops(cip); 385 386 d_instantiate(dentry, inode); 387 xfs_finish_inode_setup(cip); 388 return 0; 389 390 out_cleanup_inode: 391 xfs_finish_inode_setup(cip); 392 xfs_cleanup_inode(dir, inode, dentry); 393 iput(inode); 394 out: 395 return error; 396 } 397 398 STATIC int 399 xfs_vn_rename( 400 struct inode *odir, 401 struct dentry *odentry, 402 struct inode *ndir, 403 struct dentry *ndentry, 404 unsigned int flags) 405 { 406 struct inode *new_inode = d_inode(ndentry); 407 int omode = 0; 408 int error; 409 struct xfs_name oname; 410 struct xfs_name nname; 411 412 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 413 return -EINVAL; 414 415 /* if we are exchanging files, we need to set i_mode of both files */ 416 if (flags & RENAME_EXCHANGE) 417 omode = d_inode(ndentry)->i_mode; 418 419 error = xfs_dentry_mode_to_name(&oname, odentry, omode); 420 if (omode && unlikely(error)) 421 return error; 422 423 error = xfs_dentry_mode_to_name(&nname, ndentry, 424 d_inode(odentry)->i_mode); 425 if (unlikely(error)) 426 return error; 427 428 return xfs_rename(XFS_I(odir), &oname, XFS_I(d_inode(odentry)), 429 XFS_I(ndir), &nname, 430 new_inode ? XFS_I(new_inode) : NULL, flags); 431 } 432 433 /* 434 * careful here - this function can get called recursively, so 435 * we need to be very careful about how much stack we use. 436 * uio is kmalloced for this reason... 437 */ 438 STATIC const char * 439 xfs_vn_get_link( 440 struct dentry *dentry, 441 struct inode *inode, 442 struct delayed_call *done) 443 { 444 char *link; 445 int error = -ENOMEM; 446 447 if (!dentry) 448 return ERR_PTR(-ECHILD); 449 450 link = kmalloc(XFS_SYMLINK_MAXLEN+1, GFP_KERNEL); 451 if (!link) 452 goto out_err; 453 454 error = xfs_readlink(XFS_I(d_inode(dentry)), link); 455 if (unlikely(error)) 456 goto out_kfree; 457 458 set_delayed_call(done, kfree_link, link); 459 return link; 460 461 out_kfree: 462 kfree(link); 463 out_err: 464 return ERR_PTR(error); 465 } 466 467 STATIC const char * 468 xfs_vn_get_link_inline( 469 struct dentry *dentry, 470 struct inode *inode, 471 struct delayed_call *done) 472 { 473 ASSERT(XFS_I(inode)->i_df.if_flags & XFS_IFINLINE); 474 return XFS_I(inode)->i_df.if_u1.if_data; 475 } 476 477 STATIC int 478 xfs_vn_getattr( 479 const struct path *path, 480 struct kstat *stat, 481 u32 request_mask, 482 unsigned int query_flags) 483 { 484 struct inode *inode = d_inode(path->dentry); 485 struct xfs_inode *ip = XFS_I(inode); 486 struct xfs_mount *mp = ip->i_mount; 487 488 trace_xfs_getattr(ip); 489 490 if (XFS_FORCED_SHUTDOWN(mp)) 491 return -EIO; 492 493 stat->size = XFS_ISIZE(ip); 494 stat->dev = inode->i_sb->s_dev; 495 stat->mode = inode->i_mode; 496 stat->nlink = inode->i_nlink; 497 stat->uid = inode->i_uid; 498 stat->gid = inode->i_gid; 499 stat->ino = ip->i_ino; 500 stat->atime = inode->i_atime; 501 stat->mtime = inode->i_mtime; 502 stat->ctime = inode->i_ctime; 503 stat->blocks = 504 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); 505 506 if (ip->i_d.di_version == 3) { 507 if (request_mask & STATX_BTIME) { 508 stat->result_mask |= STATX_BTIME; 509 stat->btime.tv_sec = ip->i_d.di_crtime.t_sec; 510 stat->btime.tv_nsec = ip->i_d.di_crtime.t_nsec; 511 } 512 } 513 514 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE) 515 stat->attributes |= STATX_ATTR_IMMUTABLE; 516 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND) 517 stat->attributes |= STATX_ATTR_APPEND; 518 if (ip->i_d.di_flags & XFS_DIFLAG_NODUMP) 519 stat->attributes |= STATX_ATTR_NODUMP; 520 521 switch (inode->i_mode & S_IFMT) { 522 case S_IFBLK: 523 case S_IFCHR: 524 stat->blksize = BLKDEV_IOSIZE; 525 stat->rdev = inode->i_rdev; 526 break; 527 default: 528 if (XFS_IS_REALTIME_INODE(ip)) { 529 /* 530 * If the file blocks are being allocated from a 531 * realtime volume, then return the inode's realtime 532 * extent size or the realtime volume's extent size. 533 */ 534 stat->blksize = 535 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog; 536 } else 537 stat->blksize = xfs_preferred_iosize(mp); 538 stat->rdev = 0; 539 break; 540 } 541 542 return 0; 543 } 544 545 static void 546 xfs_setattr_mode( 547 struct xfs_inode *ip, 548 struct iattr *iattr) 549 { 550 struct inode *inode = VFS_I(ip); 551 umode_t mode = iattr->ia_mode; 552 553 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 554 555 inode->i_mode &= S_IFMT; 556 inode->i_mode |= mode & ~S_IFMT; 557 } 558 559 void 560 xfs_setattr_time( 561 struct xfs_inode *ip, 562 struct iattr *iattr) 563 { 564 struct inode *inode = VFS_I(ip); 565 566 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 567 568 if (iattr->ia_valid & ATTR_ATIME) 569 inode->i_atime = iattr->ia_atime; 570 if (iattr->ia_valid & ATTR_CTIME) 571 inode->i_ctime = iattr->ia_ctime; 572 if (iattr->ia_valid & ATTR_MTIME) 573 inode->i_mtime = iattr->ia_mtime; 574 } 575 576 static int 577 xfs_vn_change_ok( 578 struct dentry *dentry, 579 struct iattr *iattr) 580 { 581 struct xfs_mount *mp = XFS_I(d_inode(dentry))->i_mount; 582 583 if (mp->m_flags & XFS_MOUNT_RDONLY) 584 return -EROFS; 585 586 if (XFS_FORCED_SHUTDOWN(mp)) 587 return -EIO; 588 589 return setattr_prepare(dentry, iattr); 590 } 591 592 /* 593 * Set non-size attributes of an inode. 594 * 595 * Caution: The caller of this function is responsible for calling 596 * setattr_prepare() or otherwise verifying the change is fine. 597 */ 598 int 599 xfs_setattr_nonsize( 600 struct xfs_inode *ip, 601 struct iattr *iattr, 602 int flags) 603 { 604 xfs_mount_t *mp = ip->i_mount; 605 struct inode *inode = VFS_I(ip); 606 int mask = iattr->ia_valid; 607 xfs_trans_t *tp; 608 int error; 609 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID; 610 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID; 611 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 612 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL; 613 614 ASSERT((mask & ATTR_SIZE) == 0); 615 616 /* 617 * If disk quotas is on, we make sure that the dquots do exist on disk, 618 * before we start any other transactions. Trying to do this later 619 * is messy. We don't care to take a readlock to look at the ids 620 * in inode here, because we can't hold it across the trans_reserve. 621 * If the IDs do change before we take the ilock, we're covered 622 * because the i_*dquot fields will get updated anyway. 623 */ 624 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 625 uint qflags = 0; 626 627 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 628 uid = iattr->ia_uid; 629 qflags |= XFS_QMOPT_UQUOTA; 630 } else { 631 uid = inode->i_uid; 632 } 633 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 634 gid = iattr->ia_gid; 635 qflags |= XFS_QMOPT_GQUOTA; 636 } else { 637 gid = inode->i_gid; 638 } 639 640 /* 641 * We take a reference when we initialize udqp and gdqp, 642 * so it is important that we never blindly double trip on 643 * the same variable. See xfs_create() for an example. 644 */ 645 ASSERT(udqp == NULL); 646 ASSERT(gdqp == NULL); 647 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid), 648 xfs_kgid_to_gid(gid), 649 xfs_get_projid(ip), 650 qflags, &udqp, &gdqp, NULL); 651 if (error) 652 return error; 653 } 654 655 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_ichange, 0, 0, 0, &tp); 656 if (error) 657 goto out_dqrele; 658 659 xfs_ilock(ip, XFS_ILOCK_EXCL); 660 xfs_trans_ijoin(tp, ip, 0); 661 662 /* 663 * Change file ownership. Must be the owner or privileged. 664 */ 665 if (mask & (ATTR_UID|ATTR_GID)) { 666 /* 667 * These IDs could have changed since we last looked at them. 668 * But, we're assured that if the ownership did change 669 * while we didn't have the inode locked, inode's dquot(s) 670 * would have changed also. 671 */ 672 iuid = inode->i_uid; 673 igid = inode->i_gid; 674 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; 675 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; 676 677 /* 678 * Do a quota reservation only if uid/gid is actually 679 * going to change. 680 */ 681 if (XFS_IS_QUOTA_RUNNING(mp) && 682 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) || 683 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) { 684 ASSERT(tp); 685 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp, 686 NULL, capable(CAP_FOWNER) ? 687 XFS_QMOPT_FORCE_RES : 0); 688 if (error) /* out of quota */ 689 goto out_cancel; 690 } 691 } 692 693 /* 694 * Change file ownership. Must be the owner or privileged. 695 */ 696 if (mask & (ATTR_UID|ATTR_GID)) { 697 /* 698 * CAP_FSETID overrides the following restrictions: 699 * 700 * The set-user-ID and set-group-ID bits of a file will be 701 * cleared upon successful return from chown() 702 */ 703 if ((inode->i_mode & (S_ISUID|S_ISGID)) && 704 !capable(CAP_FSETID)) 705 inode->i_mode &= ~(S_ISUID|S_ISGID); 706 707 /* 708 * Change the ownerships and register quota modifications 709 * in the transaction. 710 */ 711 if (!uid_eq(iuid, uid)) { 712 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { 713 ASSERT(mask & ATTR_UID); 714 ASSERT(udqp); 715 olddquot1 = xfs_qm_vop_chown(tp, ip, 716 &ip->i_udquot, udqp); 717 } 718 ip->i_d.di_uid = xfs_kuid_to_uid(uid); 719 inode->i_uid = uid; 720 } 721 if (!gid_eq(igid, gid)) { 722 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { 723 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) || 724 !XFS_IS_PQUOTA_ON(mp)); 725 ASSERT(mask & ATTR_GID); 726 ASSERT(gdqp); 727 olddquot2 = xfs_qm_vop_chown(tp, ip, 728 &ip->i_gdquot, gdqp); 729 } 730 ip->i_d.di_gid = xfs_kgid_to_gid(gid); 731 inode->i_gid = gid; 732 } 733 } 734 735 if (mask & ATTR_MODE) 736 xfs_setattr_mode(ip, iattr); 737 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) 738 xfs_setattr_time(ip, iattr); 739 740 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 741 742 XFS_STATS_INC(mp, xs_ig_attrchg); 743 744 if (mp->m_flags & XFS_MOUNT_WSYNC) 745 xfs_trans_set_sync(tp); 746 error = xfs_trans_commit(tp); 747 748 xfs_iunlock(ip, XFS_ILOCK_EXCL); 749 750 /* 751 * Release any dquot(s) the inode had kept before chown. 752 */ 753 xfs_qm_dqrele(olddquot1); 754 xfs_qm_dqrele(olddquot2); 755 xfs_qm_dqrele(udqp); 756 xfs_qm_dqrele(gdqp); 757 758 if (error) 759 return error; 760 761 /* 762 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 763 * update. We could avoid this with linked transactions 764 * and passing down the transaction pointer all the way 765 * to attr_set. No previous user of the generic 766 * Posix ACL code seems to care about this issue either. 767 */ 768 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) { 769 error = posix_acl_chmod(inode, inode->i_mode); 770 if (error) 771 return error; 772 } 773 774 return 0; 775 776 out_cancel: 777 xfs_trans_cancel(tp); 778 out_dqrele: 779 xfs_qm_dqrele(udqp); 780 xfs_qm_dqrele(gdqp); 781 return error; 782 } 783 784 int 785 xfs_vn_setattr_nonsize( 786 struct dentry *dentry, 787 struct iattr *iattr) 788 { 789 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 790 int error; 791 792 trace_xfs_setattr(ip); 793 794 error = xfs_vn_change_ok(dentry, iattr); 795 if (error) 796 return error; 797 return xfs_setattr_nonsize(ip, iattr, 0); 798 } 799 800 /* 801 * Truncate file. Must have write permission and not be a directory. 802 * 803 * Caution: The caller of this function is responsible for calling 804 * setattr_prepare() or otherwise verifying the change is fine. 805 */ 806 STATIC int 807 xfs_setattr_size( 808 struct xfs_inode *ip, 809 struct iattr *iattr) 810 { 811 struct xfs_mount *mp = ip->i_mount; 812 struct inode *inode = VFS_I(ip); 813 xfs_off_t oldsize, newsize; 814 struct xfs_trans *tp; 815 int error; 816 uint lock_flags = 0; 817 bool did_zeroing = false; 818 819 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 820 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL)); 821 ASSERT(S_ISREG(inode->i_mode)); 822 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 823 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0); 824 825 oldsize = inode->i_size; 826 newsize = iattr->ia_size; 827 828 /* 829 * Short circuit the truncate case for zero length files. 830 */ 831 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) { 832 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 833 return 0; 834 835 /* 836 * Use the regular setattr path to update the timestamps. 837 */ 838 iattr->ia_valid &= ~ATTR_SIZE; 839 return xfs_setattr_nonsize(ip, iattr, 0); 840 } 841 842 /* 843 * Make sure that the dquots are attached to the inode. 844 */ 845 error = xfs_qm_dqattach(ip); 846 if (error) 847 return error; 848 849 /* 850 * Wait for all direct I/O to complete. 851 */ 852 inode_dio_wait(inode); 853 854 /* 855 * File data changes must be complete before we start the transaction to 856 * modify the inode. This needs to be done before joining the inode to 857 * the transaction because the inode cannot be unlocked once it is a 858 * part of the transaction. 859 * 860 * Start with zeroing any data beyond EOF that we may expose on file 861 * extension, or zeroing out the rest of the block on a downward 862 * truncate. 863 */ 864 if (newsize > oldsize) { 865 trace_xfs_zero_eof(ip, oldsize, newsize - oldsize); 866 error = iomap_zero_range(inode, oldsize, newsize - oldsize, 867 &did_zeroing, &xfs_iomap_ops); 868 } else { 869 error = iomap_truncate_page(inode, newsize, &did_zeroing, 870 &xfs_iomap_ops); 871 } 872 873 if (error) 874 return error; 875 876 /* 877 * We've already locked out new page faults, so now we can safely remove 878 * pages from the page cache knowing they won't get refaulted until we 879 * drop the XFS_MMAP_EXCL lock after the extent manipulations are 880 * complete. The truncate_setsize() call also cleans partial EOF page 881 * PTEs on extending truncates and hence ensures sub-page block size 882 * filesystems are correctly handled, too. 883 * 884 * We have to do all the page cache truncate work outside the 885 * transaction context as the "lock" order is page lock->log space 886 * reservation as defined by extent allocation in the writeback path. 887 * Hence a truncate can fail with ENOMEM from xfs_trans_alloc(), but 888 * having already truncated the in-memory version of the file (i.e. made 889 * user visible changes). There's not much we can do about this, except 890 * to hope that the caller sees ENOMEM and retries the truncate 891 * operation. 892 * 893 * And we update in-core i_size and truncate page cache beyond newsize 894 * before writeback the [di_size, newsize] range, so we're guaranteed 895 * not to write stale data past the new EOF on truncate down. 896 */ 897 truncate_setsize(inode, newsize); 898 899 /* 900 * We are going to log the inode size change in this transaction so 901 * any previous writes that are beyond the on disk EOF and the new 902 * EOF that have not been written out need to be written here. If we 903 * do not write the data out, we expose ourselves to the null files 904 * problem. Note that this includes any block zeroing we did above; 905 * otherwise those blocks may not be zeroed after a crash. 906 */ 907 if (did_zeroing || 908 (newsize > ip->i_d.di_size && oldsize != ip->i_d.di_size)) { 909 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 910 ip->i_d.di_size, newsize - 1); 911 if (error) 912 return error; 913 } 914 915 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp); 916 if (error) 917 return error; 918 919 lock_flags |= XFS_ILOCK_EXCL; 920 xfs_ilock(ip, XFS_ILOCK_EXCL); 921 xfs_trans_ijoin(tp, ip, 0); 922 923 /* 924 * Only change the c/mtime if we are changing the size or we are 925 * explicitly asked to change it. This handles the semantic difference 926 * between truncate() and ftruncate() as implemented in the VFS. 927 * 928 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 929 * special case where we need to update the times despite not having 930 * these flags set. For all other operations the VFS set these flags 931 * explicitly if it wants a timestamp update. 932 */ 933 if (newsize != oldsize && 934 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 935 iattr->ia_ctime = iattr->ia_mtime = 936 current_time(inode); 937 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 938 } 939 940 /* 941 * The first thing we do is set the size to new_size permanently on 942 * disk. This way we don't have to worry about anyone ever being able 943 * to look at the data being freed even in the face of a crash. 944 * What we're getting around here is the case where we free a block, it 945 * is allocated to another file, it is written to, and then we crash. 946 * If the new data gets written to the file but the log buffers 947 * containing the free and reallocation don't, then we'd end up with 948 * garbage in the blocks being freed. As long as we make the new size 949 * permanent before actually freeing any blocks it doesn't matter if 950 * they get written to. 951 */ 952 ip->i_d.di_size = newsize; 953 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 954 955 if (newsize <= oldsize) { 956 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 957 if (error) 958 goto out_trans_cancel; 959 960 /* 961 * Truncated "down", so we're removing references to old data 962 * here - if we delay flushing for a long time, we expose 963 * ourselves unduly to the notorious NULL files problem. So, 964 * we mark this inode and flush it when the file is closed, 965 * and do not wait the usual (long) time for writeout. 966 */ 967 xfs_iflags_set(ip, XFS_ITRUNCATED); 968 969 /* A truncate down always removes post-EOF blocks. */ 970 xfs_inode_clear_eofblocks_tag(ip); 971 } 972 973 if (iattr->ia_valid & ATTR_MODE) 974 xfs_setattr_mode(ip, iattr); 975 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) 976 xfs_setattr_time(ip, iattr); 977 978 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 979 980 XFS_STATS_INC(mp, xs_ig_attrchg); 981 982 if (mp->m_flags & XFS_MOUNT_WSYNC) 983 xfs_trans_set_sync(tp); 984 985 error = xfs_trans_commit(tp); 986 out_unlock: 987 if (lock_flags) 988 xfs_iunlock(ip, lock_flags); 989 return error; 990 991 out_trans_cancel: 992 xfs_trans_cancel(tp); 993 goto out_unlock; 994 } 995 996 int 997 xfs_vn_setattr_size( 998 struct dentry *dentry, 999 struct iattr *iattr) 1000 { 1001 struct xfs_inode *ip = XFS_I(d_inode(dentry)); 1002 int error; 1003 1004 trace_xfs_setattr(ip); 1005 1006 error = xfs_vn_change_ok(dentry, iattr); 1007 if (error) 1008 return error; 1009 return xfs_setattr_size(ip, iattr); 1010 } 1011 1012 STATIC int 1013 xfs_vn_setattr( 1014 struct dentry *dentry, 1015 struct iattr *iattr) 1016 { 1017 int error; 1018 1019 if (iattr->ia_valid & ATTR_SIZE) { 1020 struct inode *inode = d_inode(dentry); 1021 struct xfs_inode *ip = XFS_I(inode); 1022 uint iolock; 1023 1024 xfs_ilock(ip, XFS_MMAPLOCK_EXCL); 1025 iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL; 1026 1027 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP); 1028 if (error) { 1029 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1030 return error; 1031 } 1032 1033 error = xfs_vn_setattr_size(dentry, iattr); 1034 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL); 1035 } else { 1036 error = xfs_vn_setattr_nonsize(dentry, iattr); 1037 } 1038 1039 return error; 1040 } 1041 1042 STATIC int 1043 xfs_vn_update_time( 1044 struct inode *inode, 1045 struct timespec64 *now, 1046 int flags) 1047 { 1048 struct xfs_inode *ip = XFS_I(inode); 1049 struct xfs_mount *mp = ip->i_mount; 1050 int log_flags = XFS_ILOG_TIMESTAMP; 1051 struct xfs_trans *tp; 1052 int error; 1053 1054 trace_xfs_update_time(ip); 1055 1056 if (inode->i_sb->s_flags & SB_LAZYTIME) { 1057 if (!((flags & S_VERSION) && 1058 inode_maybe_inc_iversion(inode, false))) 1059 return generic_update_time(inode, now, flags); 1060 1061 /* Capture the iversion update that just occurred */ 1062 log_flags |= XFS_ILOG_CORE; 1063 } 1064 1065 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp); 1066 if (error) 1067 return error; 1068 1069 xfs_ilock(ip, XFS_ILOCK_EXCL); 1070 if (flags & S_CTIME) 1071 inode->i_ctime = *now; 1072 if (flags & S_MTIME) 1073 inode->i_mtime = *now; 1074 if (flags & S_ATIME) 1075 inode->i_atime = *now; 1076 1077 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 1078 xfs_trans_log_inode(tp, ip, log_flags); 1079 return xfs_trans_commit(tp); 1080 } 1081 1082 STATIC int 1083 xfs_vn_fiemap( 1084 struct inode *inode, 1085 struct fiemap_extent_info *fieinfo, 1086 u64 start, 1087 u64 length) 1088 { 1089 int error; 1090 1091 xfs_ilock(XFS_I(inode), XFS_IOLOCK_SHARED); 1092 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) { 1093 fieinfo->fi_flags &= ~FIEMAP_FLAG_XATTR; 1094 error = iomap_fiemap(inode, fieinfo, start, length, 1095 &xfs_xattr_iomap_ops); 1096 } else { 1097 error = iomap_fiemap(inode, fieinfo, start, length, 1098 &xfs_iomap_ops); 1099 } 1100 xfs_iunlock(XFS_I(inode), XFS_IOLOCK_SHARED); 1101 1102 return error; 1103 } 1104 1105 STATIC int 1106 xfs_vn_tmpfile( 1107 struct inode *dir, 1108 struct dentry *dentry, 1109 umode_t mode) 1110 { 1111 return xfs_generic_create(dir, dentry, mode, 0, true); 1112 } 1113 1114 static const struct inode_operations xfs_inode_operations = { 1115 .get_acl = xfs_get_acl, 1116 .set_acl = xfs_set_acl, 1117 .getattr = xfs_vn_getattr, 1118 .setattr = xfs_vn_setattr, 1119 .listxattr = xfs_vn_listxattr, 1120 .fiemap = xfs_vn_fiemap, 1121 .update_time = xfs_vn_update_time, 1122 }; 1123 1124 static const struct inode_operations xfs_dir_inode_operations = { 1125 .create = xfs_vn_create, 1126 .lookup = xfs_vn_lookup, 1127 .link = xfs_vn_link, 1128 .unlink = xfs_vn_unlink, 1129 .symlink = xfs_vn_symlink, 1130 .mkdir = xfs_vn_mkdir, 1131 /* 1132 * Yes, XFS uses the same method for rmdir and unlink. 1133 * 1134 * There are some subtile differences deeper in the code, 1135 * but we use S_ISDIR to check for those. 1136 */ 1137 .rmdir = xfs_vn_unlink, 1138 .mknod = xfs_vn_mknod, 1139 .rename = xfs_vn_rename, 1140 .get_acl = xfs_get_acl, 1141 .set_acl = xfs_set_acl, 1142 .getattr = xfs_vn_getattr, 1143 .setattr = xfs_vn_setattr, 1144 .listxattr = xfs_vn_listxattr, 1145 .update_time = xfs_vn_update_time, 1146 .tmpfile = xfs_vn_tmpfile, 1147 }; 1148 1149 static const struct inode_operations xfs_dir_ci_inode_operations = { 1150 .create = xfs_vn_create, 1151 .lookup = xfs_vn_ci_lookup, 1152 .link = xfs_vn_link, 1153 .unlink = xfs_vn_unlink, 1154 .symlink = xfs_vn_symlink, 1155 .mkdir = xfs_vn_mkdir, 1156 /* 1157 * Yes, XFS uses the same method for rmdir and unlink. 1158 * 1159 * There are some subtile differences deeper in the code, 1160 * but we use S_ISDIR to check for those. 1161 */ 1162 .rmdir = xfs_vn_unlink, 1163 .mknod = xfs_vn_mknod, 1164 .rename = xfs_vn_rename, 1165 .get_acl = xfs_get_acl, 1166 .set_acl = xfs_set_acl, 1167 .getattr = xfs_vn_getattr, 1168 .setattr = xfs_vn_setattr, 1169 .listxattr = xfs_vn_listxattr, 1170 .update_time = xfs_vn_update_time, 1171 .tmpfile = xfs_vn_tmpfile, 1172 }; 1173 1174 static const struct inode_operations xfs_symlink_inode_operations = { 1175 .get_link = xfs_vn_get_link, 1176 .getattr = xfs_vn_getattr, 1177 .setattr = xfs_vn_setattr, 1178 .listxattr = xfs_vn_listxattr, 1179 .update_time = xfs_vn_update_time, 1180 }; 1181 1182 static const struct inode_operations xfs_inline_symlink_inode_operations = { 1183 .get_link = xfs_vn_get_link_inline, 1184 .getattr = xfs_vn_getattr, 1185 .setattr = xfs_vn_setattr, 1186 .listxattr = xfs_vn_listxattr, 1187 .update_time = xfs_vn_update_time, 1188 }; 1189 1190 /* Figure out if this file actually supports DAX. */ 1191 static bool 1192 xfs_inode_supports_dax( 1193 struct xfs_inode *ip) 1194 { 1195 struct xfs_mount *mp = ip->i_mount; 1196 1197 /* Only supported on non-reflinked files. */ 1198 if (!S_ISREG(VFS_I(ip)->i_mode) || xfs_is_reflink_inode(ip)) 1199 return false; 1200 1201 /* DAX mount option or DAX iflag must be set. */ 1202 if (!(mp->m_flags & XFS_MOUNT_DAX) && 1203 !(ip->i_d.di_flags2 & XFS_DIFLAG2_DAX)) 1204 return false; 1205 1206 /* Block size must match page size */ 1207 if (mp->m_sb.sb_blocksize != PAGE_SIZE) 1208 return false; 1209 1210 /* Device has to support DAX too. */ 1211 return xfs_find_daxdev_for_inode(VFS_I(ip)) != NULL; 1212 } 1213 1214 STATIC void 1215 xfs_diflags_to_iflags( 1216 struct inode *inode, 1217 struct xfs_inode *ip) 1218 { 1219 uint16_t flags = ip->i_d.di_flags; 1220 1221 inode->i_flags &= ~(S_IMMUTABLE | S_APPEND | S_SYNC | 1222 S_NOATIME | S_DAX); 1223 1224 if (flags & XFS_DIFLAG_IMMUTABLE) 1225 inode->i_flags |= S_IMMUTABLE; 1226 if (flags & XFS_DIFLAG_APPEND) 1227 inode->i_flags |= S_APPEND; 1228 if (flags & XFS_DIFLAG_SYNC) 1229 inode->i_flags |= S_SYNC; 1230 if (flags & XFS_DIFLAG_NOATIME) 1231 inode->i_flags |= S_NOATIME; 1232 if (xfs_inode_supports_dax(ip)) 1233 inode->i_flags |= S_DAX; 1234 } 1235 1236 /* 1237 * Initialize the Linux inode. 1238 * 1239 * When reading existing inodes from disk this is called directly from xfs_iget, 1240 * when creating a new inode it is called from xfs_ialloc after setting up the 1241 * inode. These callers have different criteria for clearing XFS_INEW, so leave 1242 * it up to the caller to deal with unlocking the inode appropriately. 1243 */ 1244 void 1245 xfs_setup_inode( 1246 struct xfs_inode *ip) 1247 { 1248 struct inode *inode = &ip->i_vnode; 1249 gfp_t gfp_mask; 1250 1251 inode->i_ino = ip->i_ino; 1252 inode->i_state = I_NEW; 1253 1254 inode_sb_list_add(inode); 1255 /* make the inode look hashed for the writeback code */ 1256 hlist_add_fake(&inode->i_hash); 1257 1258 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid); 1259 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid); 1260 1261 i_size_write(inode, ip->i_d.di_size); 1262 xfs_diflags_to_iflags(inode, ip); 1263 1264 if (S_ISDIR(inode->i_mode)) { 1265 /* 1266 * We set the i_rwsem class here to avoid potential races with 1267 * lockdep_annotate_inode_mutex_key() reinitialising the lock 1268 * after a filehandle lookup has already found the inode in 1269 * cache before it has been unlocked via unlock_new_inode(). 1270 */ 1271 lockdep_set_class(&inode->i_rwsem, 1272 &inode->i_sb->s_type->i_mutex_dir_key); 1273 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); 1274 ip->d_ops = ip->i_mount->m_dir_inode_ops; 1275 } else { 1276 ip->d_ops = ip->i_mount->m_nondir_inode_ops; 1277 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); 1278 } 1279 1280 /* 1281 * Ensure all page cache allocations are done from GFP_NOFS context to 1282 * prevent direct reclaim recursion back into the filesystem and blowing 1283 * stacks or deadlocking. 1284 */ 1285 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1286 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1287 1288 /* 1289 * If there is no attribute fork no ACL can exist on this inode, 1290 * and it can't have any file capabilities attached to it either. 1291 */ 1292 if (!XFS_IFORK_Q(ip)) { 1293 inode_has_no_xattr(inode); 1294 cache_no_acl(inode); 1295 } 1296 } 1297 1298 void 1299 xfs_setup_iops( 1300 struct xfs_inode *ip) 1301 { 1302 struct inode *inode = &ip->i_vnode; 1303 1304 switch (inode->i_mode & S_IFMT) { 1305 case S_IFREG: 1306 inode->i_op = &xfs_inode_operations; 1307 inode->i_fop = &xfs_file_operations; 1308 if (IS_DAX(inode)) 1309 inode->i_mapping->a_ops = &xfs_dax_aops; 1310 else 1311 inode->i_mapping->a_ops = &xfs_address_space_operations; 1312 break; 1313 case S_IFDIR: 1314 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb)) 1315 inode->i_op = &xfs_dir_ci_inode_operations; 1316 else 1317 inode->i_op = &xfs_dir_inode_operations; 1318 inode->i_fop = &xfs_dir_file_operations; 1319 break; 1320 case S_IFLNK: 1321 if (ip->i_df.if_flags & XFS_IFINLINE) 1322 inode->i_op = &xfs_inline_symlink_inode_operations; 1323 else 1324 inode->i_op = &xfs_symlink_inode_operations; 1325 break; 1326 default: 1327 inode->i_op = &xfs_inode_operations; 1328 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1329 break; 1330 } 1331 } 1332