1 /* 2 * Copyright (c) 2000-2005 Silicon Graphics, Inc. 3 * All Rights Reserved. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License as 7 * published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it would be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write the Free Software Foundation, 16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 17 */ 18 #include "xfs.h" 19 #include "xfs_fs.h" 20 #include "xfs_shared.h" 21 #include "xfs_format.h" 22 #include "xfs_log_format.h" 23 #include "xfs_trans_resv.h" 24 #include "xfs_sb.h" 25 #include "xfs_ag.h" 26 #include "xfs_mount.h" 27 #include "xfs_da_format.h" 28 #include "xfs_inode.h" 29 #include "xfs_bmap.h" 30 #include "xfs_bmap_util.h" 31 #include "xfs_acl.h" 32 #include "xfs_quota.h" 33 #include "xfs_error.h" 34 #include "xfs_attr.h" 35 #include "xfs_trans.h" 36 #include "xfs_trace.h" 37 #include "xfs_icache.h" 38 #include "xfs_symlink.h" 39 #include "xfs_da_btree.h" 40 #include "xfs_dir2_priv.h" 41 #include "xfs_dinode.h" 42 #include "xfs_trans_space.h" 43 44 #include <linux/capability.h> 45 #include <linux/xattr.h> 46 #include <linux/namei.h> 47 #include <linux/posix_acl.h> 48 #include <linux/security.h> 49 #include <linux/fiemap.h> 50 #include <linux/slab.h> 51 52 /* 53 * Directories have different lock order w.r.t. mmap_sem compared to regular 54 * files. This is due to readdir potentially triggering page faults on a user 55 * buffer inside filldir(), and this happens with the ilock on the directory 56 * held. For regular files, the lock order is the other way around - the 57 * mmap_sem is taken during the page fault, and then we lock the ilock to do 58 * block mapping. Hence we need a different class for the directory ilock so 59 * that lockdep can tell them apart. 60 */ 61 static struct lock_class_key xfs_nondir_ilock_class; 62 static struct lock_class_key xfs_dir_ilock_class; 63 64 static int 65 xfs_initxattrs( 66 struct inode *inode, 67 const struct xattr *xattr_array, 68 void *fs_info) 69 { 70 const struct xattr *xattr; 71 struct xfs_inode *ip = XFS_I(inode); 72 int error = 0; 73 74 for (xattr = xattr_array; xattr->name != NULL; xattr++) { 75 error = -xfs_attr_set(ip, xattr->name, xattr->value, 76 xattr->value_len, ATTR_SECURE); 77 if (error < 0) 78 break; 79 } 80 return error; 81 } 82 83 /* 84 * Hook in SELinux. This is not quite correct yet, what we really need 85 * here (as we do for default ACLs) is a mechanism by which creation of 86 * these attrs can be journalled at inode creation time (along with the 87 * inode, of course, such that log replay can't cause these to be lost). 88 */ 89 90 STATIC int 91 xfs_init_security( 92 struct inode *inode, 93 struct inode *dir, 94 const struct qstr *qstr) 95 { 96 return -security_inode_init_security(inode, dir, qstr, 97 &xfs_initxattrs, NULL); 98 } 99 100 static void 101 xfs_dentry_to_name( 102 struct xfs_name *namep, 103 struct dentry *dentry, 104 int mode) 105 { 106 namep->name = dentry->d_name.name; 107 namep->len = dentry->d_name.len; 108 namep->type = xfs_mode_to_ftype[(mode & S_IFMT) >> S_SHIFT]; 109 } 110 111 STATIC void 112 xfs_cleanup_inode( 113 struct inode *dir, 114 struct inode *inode, 115 struct dentry *dentry) 116 { 117 struct xfs_name teardown; 118 119 /* Oh, the horror. 120 * If we can't add the ACL or we fail in 121 * xfs_init_security we must back out. 122 * ENOSPC can hit here, among other things. 123 */ 124 xfs_dentry_to_name(&teardown, dentry, 0); 125 126 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); 127 } 128 129 STATIC int 130 xfs_generic_create( 131 struct inode *dir, 132 struct dentry *dentry, 133 umode_t mode, 134 dev_t rdev, 135 bool tmpfile) /* unnamed file */ 136 { 137 struct inode *inode; 138 struct xfs_inode *ip = NULL; 139 struct posix_acl *default_acl, *acl; 140 struct xfs_name name; 141 int error; 142 143 /* 144 * Irix uses Missed'em'V split, but doesn't want to see 145 * the upper 5 bits of (14bit) major. 146 */ 147 if (S_ISCHR(mode) || S_ISBLK(mode)) { 148 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) 149 return -EINVAL; 150 rdev = sysv_encode_dev(rdev); 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 if (!tmpfile) { 160 xfs_dentry_to_name(&name, dentry, mode); 161 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip); 162 } else { 163 error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip); 164 } 165 if (unlikely(error)) 166 goto out_free_acl; 167 168 inode = VFS_I(ip); 169 170 error = xfs_init_security(inode, dir, &dentry->d_name); 171 if (unlikely(error)) 172 goto out_cleanup_inode; 173 174 #ifdef CONFIG_XFS_POSIX_ACL 175 if (default_acl) { 176 error = -xfs_set_acl(inode, default_acl, ACL_TYPE_DEFAULT); 177 if (error) 178 goto out_cleanup_inode; 179 } 180 if (acl) { 181 error = -xfs_set_acl(inode, acl, ACL_TYPE_ACCESS); 182 if (error) 183 goto out_cleanup_inode; 184 } 185 #endif 186 187 if (tmpfile) 188 d_tmpfile(dentry, inode); 189 else 190 d_instantiate(dentry, inode); 191 192 out_free_acl: 193 if (default_acl) 194 posix_acl_release(default_acl); 195 if (acl) 196 posix_acl_release(acl); 197 return -error; 198 199 out_cleanup_inode: 200 if (!tmpfile) 201 xfs_cleanup_inode(dir, inode, dentry); 202 iput(inode); 203 goto out_free_acl; 204 } 205 206 STATIC int 207 xfs_vn_mknod( 208 struct inode *dir, 209 struct dentry *dentry, 210 umode_t mode, 211 dev_t rdev) 212 { 213 return xfs_generic_create(dir, dentry, mode, rdev, false); 214 } 215 216 STATIC int 217 xfs_vn_create( 218 struct inode *dir, 219 struct dentry *dentry, 220 umode_t mode, 221 bool flags) 222 { 223 return xfs_vn_mknod(dir, dentry, mode, 0); 224 } 225 226 STATIC int 227 xfs_vn_mkdir( 228 struct inode *dir, 229 struct dentry *dentry, 230 umode_t mode) 231 { 232 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0); 233 } 234 235 STATIC struct dentry * 236 xfs_vn_lookup( 237 struct inode *dir, 238 struct dentry *dentry, 239 unsigned int flags) 240 { 241 struct xfs_inode *cip; 242 struct xfs_name name; 243 int error; 244 245 if (dentry->d_name.len >= MAXNAMELEN) 246 return ERR_PTR(-ENAMETOOLONG); 247 248 xfs_dentry_to_name(&name, dentry, 0); 249 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); 250 if (unlikely(error)) { 251 if (unlikely(error != ENOENT)) 252 return ERR_PTR(-error); 253 d_add(dentry, NULL); 254 return NULL; 255 } 256 257 return d_splice_alias(VFS_I(cip), dentry); 258 } 259 260 STATIC struct dentry * 261 xfs_vn_ci_lookup( 262 struct inode *dir, 263 struct dentry *dentry, 264 unsigned int flags) 265 { 266 struct xfs_inode *ip; 267 struct xfs_name xname; 268 struct xfs_name ci_name; 269 struct qstr dname; 270 int error; 271 272 if (dentry->d_name.len >= MAXNAMELEN) 273 return ERR_PTR(-ENAMETOOLONG); 274 275 xfs_dentry_to_name(&xname, dentry, 0); 276 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); 277 if (unlikely(error)) { 278 if (unlikely(error != ENOENT)) 279 return ERR_PTR(-error); 280 /* 281 * call d_add(dentry, NULL) here when d_drop_negative_children 282 * is called in xfs_vn_mknod (ie. allow negative dentries 283 * with CI filesystems). 284 */ 285 return NULL; 286 } 287 288 /* if exact match, just splice and exit */ 289 if (!ci_name.name) 290 return d_splice_alias(VFS_I(ip), dentry); 291 292 /* else case-insensitive match... */ 293 dname.name = ci_name.name; 294 dname.len = ci_name.len; 295 dentry = d_add_ci(dentry, VFS_I(ip), &dname); 296 kmem_free(ci_name.name); 297 return dentry; 298 } 299 300 STATIC int 301 xfs_vn_link( 302 struct dentry *old_dentry, 303 struct inode *dir, 304 struct dentry *dentry) 305 { 306 struct inode *inode = old_dentry->d_inode; 307 struct xfs_name name; 308 int error; 309 310 xfs_dentry_to_name(&name, dentry, inode->i_mode); 311 312 error = xfs_link(XFS_I(dir), XFS_I(inode), &name); 313 if (unlikely(error)) 314 return -error; 315 316 ihold(inode); 317 d_instantiate(dentry, inode); 318 return 0; 319 } 320 321 STATIC int 322 xfs_vn_unlink( 323 struct inode *dir, 324 struct dentry *dentry) 325 { 326 struct xfs_name name; 327 int error; 328 329 xfs_dentry_to_name(&name, dentry, 0); 330 331 error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode)); 332 if (error) 333 return error; 334 335 /* 336 * With unlink, the VFS makes the dentry "negative": no inode, 337 * but still hashed. This is incompatible with case-insensitive 338 * mode, so invalidate (unhash) the dentry in CI-mode. 339 */ 340 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb)) 341 d_invalidate(dentry); 342 return 0; 343 } 344 345 STATIC int 346 xfs_vn_symlink( 347 struct inode *dir, 348 struct dentry *dentry, 349 const char *symname) 350 { 351 struct inode *inode; 352 struct xfs_inode *cip = NULL; 353 struct xfs_name name; 354 int error; 355 umode_t mode; 356 357 mode = S_IFLNK | 358 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); 359 xfs_dentry_to_name(&name, dentry, mode); 360 361 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip); 362 if (unlikely(error)) 363 goto out; 364 365 inode = VFS_I(cip); 366 367 error = xfs_init_security(inode, dir, &dentry->d_name); 368 if (unlikely(error)) 369 goto out_cleanup_inode; 370 371 d_instantiate(dentry, inode); 372 return 0; 373 374 out_cleanup_inode: 375 xfs_cleanup_inode(dir, inode, dentry); 376 iput(inode); 377 out: 378 return -error; 379 } 380 381 STATIC int 382 xfs_vn_rename( 383 struct inode *odir, 384 struct dentry *odentry, 385 struct inode *ndir, 386 struct dentry *ndentry) 387 { 388 struct inode *new_inode = ndentry->d_inode; 389 struct xfs_name oname; 390 struct xfs_name nname; 391 392 xfs_dentry_to_name(&oname, odentry, 0); 393 xfs_dentry_to_name(&nname, ndentry, odentry->d_inode->i_mode); 394 395 return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode), 396 XFS_I(ndir), &nname, new_inode ? 397 XFS_I(new_inode) : NULL); 398 } 399 400 /* 401 * careful here - this function can get called recursively, so 402 * we need to be very careful about how much stack we use. 403 * uio is kmalloced for this reason... 404 */ 405 STATIC void * 406 xfs_vn_follow_link( 407 struct dentry *dentry, 408 struct nameidata *nd) 409 { 410 char *link; 411 int error = -ENOMEM; 412 413 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL); 414 if (!link) 415 goto out_err; 416 417 error = -xfs_readlink(XFS_I(dentry->d_inode), link); 418 if (unlikely(error)) 419 goto out_kfree; 420 421 nd_set_link(nd, link); 422 return NULL; 423 424 out_kfree: 425 kfree(link); 426 out_err: 427 nd_set_link(nd, ERR_PTR(error)); 428 return NULL; 429 } 430 431 STATIC int 432 xfs_vn_getattr( 433 struct vfsmount *mnt, 434 struct dentry *dentry, 435 struct kstat *stat) 436 { 437 struct inode *inode = dentry->d_inode; 438 struct xfs_inode *ip = XFS_I(inode); 439 struct xfs_mount *mp = ip->i_mount; 440 441 trace_xfs_getattr(ip); 442 443 if (XFS_FORCED_SHUTDOWN(mp)) 444 return -XFS_ERROR(EIO); 445 446 stat->size = XFS_ISIZE(ip); 447 stat->dev = inode->i_sb->s_dev; 448 stat->mode = ip->i_d.di_mode; 449 stat->nlink = ip->i_d.di_nlink; 450 stat->uid = inode->i_uid; 451 stat->gid = inode->i_gid; 452 stat->ino = ip->i_ino; 453 stat->atime = inode->i_atime; 454 stat->mtime = inode->i_mtime; 455 stat->ctime = inode->i_ctime; 456 stat->blocks = 457 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); 458 459 460 switch (inode->i_mode & S_IFMT) { 461 case S_IFBLK: 462 case S_IFCHR: 463 stat->blksize = BLKDEV_IOSIZE; 464 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff, 465 sysv_minor(ip->i_df.if_u2.if_rdev)); 466 break; 467 default: 468 if (XFS_IS_REALTIME_INODE(ip)) { 469 /* 470 * If the file blocks are being allocated from a 471 * realtime volume, then return the inode's realtime 472 * extent size or the realtime volume's extent size. 473 */ 474 stat->blksize = 475 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog; 476 } else 477 stat->blksize = xfs_preferred_iosize(mp); 478 stat->rdev = 0; 479 break; 480 } 481 482 return 0; 483 } 484 485 static void 486 xfs_setattr_mode( 487 struct xfs_inode *ip, 488 struct iattr *iattr) 489 { 490 struct inode *inode = VFS_I(ip); 491 umode_t mode = iattr->ia_mode; 492 493 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 494 495 ip->i_d.di_mode &= S_IFMT; 496 ip->i_d.di_mode |= mode & ~S_IFMT; 497 498 inode->i_mode &= S_IFMT; 499 inode->i_mode |= mode & ~S_IFMT; 500 } 501 502 static void 503 xfs_setattr_time( 504 struct xfs_inode *ip, 505 struct iattr *iattr) 506 { 507 struct inode *inode = VFS_I(ip); 508 509 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); 510 511 if (iattr->ia_valid & ATTR_ATIME) { 512 inode->i_atime = iattr->ia_atime; 513 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec; 514 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec; 515 } 516 if (iattr->ia_valid & ATTR_CTIME) { 517 inode->i_ctime = iattr->ia_ctime; 518 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec; 519 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec; 520 } 521 if (iattr->ia_valid & ATTR_MTIME) { 522 inode->i_mtime = iattr->ia_mtime; 523 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec; 524 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec; 525 } 526 } 527 528 int 529 xfs_setattr_nonsize( 530 struct xfs_inode *ip, 531 struct iattr *iattr, 532 int flags) 533 { 534 xfs_mount_t *mp = ip->i_mount; 535 struct inode *inode = VFS_I(ip); 536 int mask = iattr->ia_valid; 537 xfs_trans_t *tp; 538 int error; 539 kuid_t uid = GLOBAL_ROOT_UID, iuid = GLOBAL_ROOT_UID; 540 kgid_t gid = GLOBAL_ROOT_GID, igid = GLOBAL_ROOT_GID; 541 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 542 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL; 543 544 trace_xfs_setattr(ip); 545 546 /* If acls are being inherited, we already have this checked */ 547 if (!(flags & XFS_ATTR_NOACL)) { 548 if (mp->m_flags & XFS_MOUNT_RDONLY) 549 return XFS_ERROR(EROFS); 550 551 if (XFS_FORCED_SHUTDOWN(mp)) 552 return XFS_ERROR(EIO); 553 554 error = -inode_change_ok(inode, iattr); 555 if (error) 556 return XFS_ERROR(error); 557 } 558 559 ASSERT((mask & ATTR_SIZE) == 0); 560 561 /* 562 * If disk quotas is on, we make sure that the dquots do exist on disk, 563 * before we start any other transactions. Trying to do this later 564 * is messy. We don't care to take a readlock to look at the ids 565 * in inode here, because we can't hold it across the trans_reserve. 566 * If the IDs do change before we take the ilock, we're covered 567 * because the i_*dquot fields will get updated anyway. 568 */ 569 if (XFS_IS_QUOTA_ON(mp) && (mask & (ATTR_UID|ATTR_GID))) { 570 uint qflags = 0; 571 572 if ((mask & ATTR_UID) && XFS_IS_UQUOTA_ON(mp)) { 573 uid = iattr->ia_uid; 574 qflags |= XFS_QMOPT_UQUOTA; 575 } else { 576 uid = inode->i_uid; 577 } 578 if ((mask & ATTR_GID) && XFS_IS_GQUOTA_ON(mp)) { 579 gid = iattr->ia_gid; 580 qflags |= XFS_QMOPT_GQUOTA; 581 } else { 582 gid = inode->i_gid; 583 } 584 585 /* 586 * We take a reference when we initialize udqp and gdqp, 587 * so it is important that we never blindly double trip on 588 * the same variable. See xfs_create() for an example. 589 */ 590 ASSERT(udqp == NULL); 591 ASSERT(gdqp == NULL); 592 error = xfs_qm_vop_dqalloc(ip, xfs_kuid_to_uid(uid), 593 xfs_kgid_to_gid(gid), 594 xfs_get_projid(ip), 595 qflags, &udqp, &gdqp, NULL); 596 if (error) 597 return error; 598 } 599 600 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); 601 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0); 602 if (error) 603 goto out_dqrele; 604 605 xfs_ilock(ip, XFS_ILOCK_EXCL); 606 607 /* 608 * Change file ownership. Must be the owner or privileged. 609 */ 610 if (mask & (ATTR_UID|ATTR_GID)) { 611 /* 612 * These IDs could have changed since we last looked at them. 613 * But, we're assured that if the ownership did change 614 * while we didn't have the inode locked, inode's dquot(s) 615 * would have changed also. 616 */ 617 iuid = inode->i_uid; 618 igid = inode->i_gid; 619 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; 620 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; 621 622 /* 623 * Do a quota reservation only if uid/gid is actually 624 * going to change. 625 */ 626 if (XFS_IS_QUOTA_RUNNING(mp) && 627 ((XFS_IS_UQUOTA_ON(mp) && !uid_eq(iuid, uid)) || 628 (XFS_IS_GQUOTA_ON(mp) && !gid_eq(igid, gid)))) { 629 ASSERT(tp); 630 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp, 631 NULL, capable(CAP_FOWNER) ? 632 XFS_QMOPT_FORCE_RES : 0); 633 if (error) /* out of quota */ 634 goto out_trans_cancel; 635 } 636 } 637 638 xfs_trans_ijoin(tp, ip, 0); 639 640 /* 641 * Change file ownership. Must be the owner or privileged. 642 */ 643 if (mask & (ATTR_UID|ATTR_GID)) { 644 /* 645 * CAP_FSETID overrides the following restrictions: 646 * 647 * The set-user-ID and set-group-ID bits of a file will be 648 * cleared upon successful return from chown() 649 */ 650 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && 651 !capable(CAP_FSETID)) 652 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); 653 654 /* 655 * Change the ownerships and register quota modifications 656 * in the transaction. 657 */ 658 if (!uid_eq(iuid, uid)) { 659 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { 660 ASSERT(mask & ATTR_UID); 661 ASSERT(udqp); 662 olddquot1 = xfs_qm_vop_chown(tp, ip, 663 &ip->i_udquot, udqp); 664 } 665 ip->i_d.di_uid = xfs_kuid_to_uid(uid); 666 inode->i_uid = uid; 667 } 668 if (!gid_eq(igid, gid)) { 669 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { 670 ASSERT(xfs_sb_version_has_pquotino(&mp->m_sb) || 671 !XFS_IS_PQUOTA_ON(mp)); 672 ASSERT(mask & ATTR_GID); 673 ASSERT(gdqp); 674 olddquot2 = xfs_qm_vop_chown(tp, ip, 675 &ip->i_gdquot, gdqp); 676 } 677 ip->i_d.di_gid = xfs_kgid_to_gid(gid); 678 inode->i_gid = gid; 679 } 680 } 681 682 if (mask & ATTR_MODE) 683 xfs_setattr_mode(ip, iattr); 684 if (mask & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) 685 xfs_setattr_time(ip, iattr); 686 687 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 688 689 XFS_STATS_INC(xs_ig_attrchg); 690 691 if (mp->m_flags & XFS_MOUNT_WSYNC) 692 xfs_trans_set_sync(tp); 693 error = xfs_trans_commit(tp, 0); 694 695 xfs_iunlock(ip, XFS_ILOCK_EXCL); 696 697 /* 698 * Release any dquot(s) the inode had kept before chown. 699 */ 700 xfs_qm_dqrele(olddquot1); 701 xfs_qm_dqrele(olddquot2); 702 xfs_qm_dqrele(udqp); 703 xfs_qm_dqrele(gdqp); 704 705 if (error) 706 return XFS_ERROR(error); 707 708 /* 709 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 710 * update. We could avoid this with linked transactions 711 * and passing down the transaction pointer all the way 712 * to attr_set. No previous user of the generic 713 * Posix ACL code seems to care about this issue either. 714 */ 715 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) { 716 error = -posix_acl_chmod(inode, inode->i_mode); 717 if (error) 718 return XFS_ERROR(error); 719 } 720 721 return 0; 722 723 out_trans_cancel: 724 xfs_trans_cancel(tp, 0); 725 xfs_iunlock(ip, XFS_ILOCK_EXCL); 726 out_dqrele: 727 xfs_qm_dqrele(udqp); 728 xfs_qm_dqrele(gdqp); 729 return error; 730 } 731 732 /* 733 * Truncate file. Must have write permission and not be a directory. 734 */ 735 int 736 xfs_setattr_size( 737 struct xfs_inode *ip, 738 struct iattr *iattr) 739 { 740 struct xfs_mount *mp = ip->i_mount; 741 struct inode *inode = VFS_I(ip); 742 xfs_off_t oldsize, newsize; 743 struct xfs_trans *tp; 744 int error; 745 uint lock_flags = 0; 746 uint commit_flags = 0; 747 748 trace_xfs_setattr(ip); 749 750 if (mp->m_flags & XFS_MOUNT_RDONLY) 751 return XFS_ERROR(EROFS); 752 753 if (XFS_FORCED_SHUTDOWN(mp)) 754 return XFS_ERROR(EIO); 755 756 error = -inode_change_ok(inode, iattr); 757 if (error) 758 return XFS_ERROR(error); 759 760 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL)); 761 ASSERT(S_ISREG(ip->i_d.di_mode)); 762 ASSERT((iattr->ia_valid & (ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 763 ATTR_MTIME_SET|ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0); 764 765 oldsize = inode->i_size; 766 newsize = iattr->ia_size; 767 768 /* 769 * Short circuit the truncate case for zero length files. 770 */ 771 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) { 772 if (!(iattr->ia_valid & (ATTR_CTIME|ATTR_MTIME))) 773 return 0; 774 775 /* 776 * Use the regular setattr path to update the timestamps. 777 */ 778 iattr->ia_valid &= ~ATTR_SIZE; 779 return xfs_setattr_nonsize(ip, iattr, 0); 780 } 781 782 /* 783 * Make sure that the dquots are attached to the inode. 784 */ 785 error = xfs_qm_dqattach(ip, 0); 786 if (error) 787 return error; 788 789 /* 790 * Now we can make the changes. Before we join the inode to the 791 * transaction, take care of the part of the truncation that must be 792 * done without the inode lock. This needs to be done before joining 793 * the inode to the transaction, because the inode cannot be unlocked 794 * once it is a part of the transaction. 795 */ 796 if (newsize > oldsize) { 797 /* 798 * Do the first part of growing a file: zero any data in the 799 * last block that is beyond the old EOF. We need to do this 800 * before the inode is joined to the transaction to modify 801 * i_size. 802 */ 803 error = xfs_zero_eof(ip, newsize, oldsize); 804 if (error) 805 return error; 806 } 807 808 /* 809 * We are going to log the inode size change in this transaction so 810 * any previous writes that are beyond the on disk EOF and the new 811 * EOF that have not been written out need to be written here. If we 812 * do not write the data out, we expose ourselves to the null files 813 * problem. 814 * 815 * Only flush from the on disk size to the smaller of the in memory 816 * file size or the new size as that's the range we really care about 817 * here and prevents waiting for other data not within the range we 818 * care about here. 819 */ 820 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) { 821 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping, 822 ip->i_d.di_size, newsize); 823 if (error) 824 return error; 825 } 826 827 /* 828 * Wait for all direct I/O to complete. 829 */ 830 inode_dio_wait(inode); 831 832 /* 833 * Do all the page cache truncate work outside the transaction context 834 * as the "lock" order is page lock->log space reservation. i.e. 835 * locking pages inside the transaction can ABBA deadlock with 836 * writeback. We have to do the VFS inode size update before we truncate 837 * the pagecache, however, to avoid racing with page faults beyond the 838 * new EOF they are not serialised against truncate operations except by 839 * page locks and size updates. 840 * 841 * Hence we are in a situation where a truncate can fail with ENOMEM 842 * from xfs_trans_reserve(), but having already truncated the in-memory 843 * version of the file (i.e. made user visible changes). There's not 844 * much we can do about this, except to hope that the caller sees ENOMEM 845 * and retries the truncate operation. 846 */ 847 error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks); 848 if (error) 849 return error; 850 truncate_setsize(inode, newsize); 851 852 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); 853 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0); 854 if (error) 855 goto out_trans_cancel; 856 857 commit_flags = XFS_TRANS_RELEASE_LOG_RES; 858 lock_flags |= XFS_ILOCK_EXCL; 859 xfs_ilock(ip, XFS_ILOCK_EXCL); 860 xfs_trans_ijoin(tp, ip, 0); 861 862 /* 863 * Only change the c/mtime if we are changing the size or we are 864 * explicitly asked to change it. This handles the semantic difference 865 * between truncate() and ftruncate() as implemented in the VFS. 866 * 867 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 868 * special case where we need to update the times despite not having 869 * these flags set. For all other operations the VFS set these flags 870 * explicitly if it wants a timestamp update. 871 */ 872 if (newsize != oldsize && 873 !(iattr->ia_valid & (ATTR_CTIME | ATTR_MTIME))) { 874 iattr->ia_ctime = iattr->ia_mtime = 875 current_fs_time(inode->i_sb); 876 iattr->ia_valid |= ATTR_CTIME | ATTR_MTIME; 877 } 878 879 /* 880 * The first thing we do is set the size to new_size permanently on 881 * disk. This way we don't have to worry about anyone ever being able 882 * to look at the data being freed even in the face of a crash. 883 * What we're getting around here is the case where we free a block, it 884 * is allocated to another file, it is written to, and then we crash. 885 * If the new data gets written to the file but the log buffers 886 * containing the free and reallocation don't, then we'd end up with 887 * garbage in the blocks being freed. As long as we make the new size 888 * permanent before actually freeing any blocks it doesn't matter if 889 * they get written to. 890 */ 891 ip->i_d.di_size = newsize; 892 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 893 894 if (newsize <= oldsize) { 895 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 896 if (error) 897 goto out_trans_abort; 898 899 /* 900 * Truncated "down", so we're removing references to old data 901 * here - if we delay flushing for a long time, we expose 902 * ourselves unduly to the notorious NULL files problem. So, 903 * we mark this inode and flush it when the file is closed, 904 * and do not wait the usual (long) time for writeout. 905 */ 906 xfs_iflags_set(ip, XFS_ITRUNCATED); 907 908 /* A truncate down always removes post-EOF blocks. */ 909 xfs_inode_clear_eofblocks_tag(ip); 910 } 911 912 if (iattr->ia_valid & ATTR_MODE) 913 xfs_setattr_mode(ip, iattr); 914 if (iattr->ia_valid & (ATTR_ATIME|ATTR_CTIME|ATTR_MTIME)) 915 xfs_setattr_time(ip, iattr); 916 917 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 918 919 XFS_STATS_INC(xs_ig_attrchg); 920 921 if (mp->m_flags & XFS_MOUNT_WSYNC) 922 xfs_trans_set_sync(tp); 923 924 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 925 out_unlock: 926 if (lock_flags) 927 xfs_iunlock(ip, lock_flags); 928 return error; 929 930 out_trans_abort: 931 commit_flags |= XFS_TRANS_ABORT; 932 out_trans_cancel: 933 xfs_trans_cancel(tp, commit_flags); 934 goto out_unlock; 935 } 936 937 STATIC int 938 xfs_vn_setattr( 939 struct dentry *dentry, 940 struct iattr *iattr) 941 { 942 struct xfs_inode *ip = XFS_I(dentry->d_inode); 943 int error; 944 945 if (iattr->ia_valid & ATTR_SIZE) { 946 xfs_ilock(ip, XFS_IOLOCK_EXCL); 947 error = xfs_setattr_size(ip, iattr); 948 xfs_iunlock(ip, XFS_IOLOCK_EXCL); 949 } else { 950 error = xfs_setattr_nonsize(ip, iattr, 0); 951 } 952 953 return -error; 954 } 955 956 STATIC int 957 xfs_vn_update_time( 958 struct inode *inode, 959 struct timespec *now, 960 int flags) 961 { 962 struct xfs_inode *ip = XFS_I(inode); 963 struct xfs_mount *mp = ip->i_mount; 964 struct xfs_trans *tp; 965 int error; 966 967 trace_xfs_update_time(ip); 968 969 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS); 970 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0); 971 if (error) { 972 xfs_trans_cancel(tp, 0); 973 return -error; 974 } 975 976 xfs_ilock(ip, XFS_ILOCK_EXCL); 977 if (flags & S_CTIME) { 978 inode->i_ctime = *now; 979 ip->i_d.di_ctime.t_sec = (__int32_t)now->tv_sec; 980 ip->i_d.di_ctime.t_nsec = (__int32_t)now->tv_nsec; 981 } 982 if (flags & S_MTIME) { 983 inode->i_mtime = *now; 984 ip->i_d.di_mtime.t_sec = (__int32_t)now->tv_sec; 985 ip->i_d.di_mtime.t_nsec = (__int32_t)now->tv_nsec; 986 } 987 if (flags & S_ATIME) { 988 inode->i_atime = *now; 989 ip->i_d.di_atime.t_sec = (__int32_t)now->tv_sec; 990 ip->i_d.di_atime.t_nsec = (__int32_t)now->tv_nsec; 991 } 992 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL); 993 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP); 994 return -xfs_trans_commit(tp, 0); 995 } 996 997 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) 998 999 /* 1000 * Call fiemap helper to fill in user data. 1001 * Returns positive errors to xfs_getbmap. 1002 */ 1003 STATIC int 1004 xfs_fiemap_format( 1005 void **arg, 1006 struct getbmapx *bmv, 1007 int *full) 1008 { 1009 int error; 1010 struct fiemap_extent_info *fieinfo = *arg; 1011 u32 fiemap_flags = 0; 1012 u64 logical, physical, length; 1013 1014 /* Do nothing for a hole */ 1015 if (bmv->bmv_block == -1LL) 1016 return 0; 1017 1018 logical = BBTOB(bmv->bmv_offset); 1019 physical = BBTOB(bmv->bmv_block); 1020 length = BBTOB(bmv->bmv_length); 1021 1022 if (bmv->bmv_oflags & BMV_OF_PREALLOC) 1023 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN; 1024 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) { 1025 fiemap_flags |= (FIEMAP_EXTENT_DELALLOC | 1026 FIEMAP_EXTENT_UNKNOWN); 1027 physical = 0; /* no block yet */ 1028 } 1029 if (bmv->bmv_oflags & BMV_OF_LAST) 1030 fiemap_flags |= FIEMAP_EXTENT_LAST; 1031 1032 error = fiemap_fill_next_extent(fieinfo, logical, physical, 1033 length, fiemap_flags); 1034 if (error > 0) { 1035 error = 0; 1036 *full = 1; /* user array now full */ 1037 } 1038 1039 return -error; 1040 } 1041 1042 STATIC int 1043 xfs_vn_fiemap( 1044 struct inode *inode, 1045 struct fiemap_extent_info *fieinfo, 1046 u64 start, 1047 u64 length) 1048 { 1049 xfs_inode_t *ip = XFS_I(inode); 1050 struct getbmapx bm; 1051 int error; 1052 1053 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS); 1054 if (error) 1055 return error; 1056 1057 /* Set up bmap header for xfs internal routine */ 1058 bm.bmv_offset = BTOBB(start); 1059 /* Special case for whole file */ 1060 if (length == FIEMAP_MAX_OFFSET) 1061 bm.bmv_length = -1LL; 1062 else 1063 bm.bmv_length = BTOBB(length); 1064 1065 /* We add one because in getbmap world count includes the header */ 1066 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM : 1067 fieinfo->fi_extents_max + 1; 1068 bm.bmv_count = min_t(__s32, bm.bmv_count, 1069 (PAGE_SIZE * 16 / sizeof(struct getbmapx))); 1070 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES; 1071 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) 1072 bm.bmv_iflags |= BMV_IF_ATTRFORK; 1073 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC)) 1074 bm.bmv_iflags |= BMV_IF_DELALLOC; 1075 1076 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo); 1077 if (error) 1078 return -error; 1079 1080 return 0; 1081 } 1082 1083 STATIC int 1084 xfs_vn_tmpfile( 1085 struct inode *dir, 1086 struct dentry *dentry, 1087 umode_t mode) 1088 { 1089 return xfs_generic_create(dir, dentry, mode, 0, true); 1090 } 1091 1092 static const struct inode_operations xfs_inode_operations = { 1093 .get_acl = xfs_get_acl, 1094 .set_acl = xfs_set_acl, 1095 .getattr = xfs_vn_getattr, 1096 .setattr = xfs_vn_setattr, 1097 .setxattr = generic_setxattr, 1098 .getxattr = generic_getxattr, 1099 .removexattr = generic_removexattr, 1100 .listxattr = xfs_vn_listxattr, 1101 .fiemap = xfs_vn_fiemap, 1102 .update_time = xfs_vn_update_time, 1103 }; 1104 1105 static const struct inode_operations xfs_dir_inode_operations = { 1106 .create = xfs_vn_create, 1107 .lookup = xfs_vn_lookup, 1108 .link = xfs_vn_link, 1109 .unlink = xfs_vn_unlink, 1110 .symlink = xfs_vn_symlink, 1111 .mkdir = xfs_vn_mkdir, 1112 /* 1113 * Yes, XFS uses the same method for rmdir and unlink. 1114 * 1115 * There are some subtile differences deeper in the code, 1116 * but we use S_ISDIR to check for those. 1117 */ 1118 .rmdir = xfs_vn_unlink, 1119 .mknod = xfs_vn_mknod, 1120 .rename = xfs_vn_rename, 1121 .get_acl = xfs_get_acl, 1122 .set_acl = xfs_set_acl, 1123 .getattr = xfs_vn_getattr, 1124 .setattr = xfs_vn_setattr, 1125 .setxattr = generic_setxattr, 1126 .getxattr = generic_getxattr, 1127 .removexattr = generic_removexattr, 1128 .listxattr = xfs_vn_listxattr, 1129 .update_time = xfs_vn_update_time, 1130 .tmpfile = xfs_vn_tmpfile, 1131 }; 1132 1133 static const struct inode_operations xfs_dir_ci_inode_operations = { 1134 .create = xfs_vn_create, 1135 .lookup = xfs_vn_ci_lookup, 1136 .link = xfs_vn_link, 1137 .unlink = xfs_vn_unlink, 1138 .symlink = xfs_vn_symlink, 1139 .mkdir = xfs_vn_mkdir, 1140 /* 1141 * Yes, XFS uses the same method for rmdir and unlink. 1142 * 1143 * There are some subtile differences deeper in the code, 1144 * but we use S_ISDIR to check for those. 1145 */ 1146 .rmdir = xfs_vn_unlink, 1147 .mknod = xfs_vn_mknod, 1148 .rename = xfs_vn_rename, 1149 .get_acl = xfs_get_acl, 1150 .set_acl = xfs_set_acl, 1151 .getattr = xfs_vn_getattr, 1152 .setattr = xfs_vn_setattr, 1153 .setxattr = generic_setxattr, 1154 .getxattr = generic_getxattr, 1155 .removexattr = generic_removexattr, 1156 .listxattr = xfs_vn_listxattr, 1157 .update_time = xfs_vn_update_time, 1158 .tmpfile = xfs_vn_tmpfile, 1159 }; 1160 1161 static const struct inode_operations xfs_symlink_inode_operations = { 1162 .readlink = generic_readlink, 1163 .follow_link = xfs_vn_follow_link, 1164 .put_link = kfree_put_link, 1165 .getattr = xfs_vn_getattr, 1166 .setattr = xfs_vn_setattr, 1167 .setxattr = generic_setxattr, 1168 .getxattr = generic_getxattr, 1169 .removexattr = generic_removexattr, 1170 .listxattr = xfs_vn_listxattr, 1171 .update_time = xfs_vn_update_time, 1172 }; 1173 1174 STATIC void 1175 xfs_diflags_to_iflags( 1176 struct inode *inode, 1177 struct xfs_inode *ip) 1178 { 1179 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE) 1180 inode->i_flags |= S_IMMUTABLE; 1181 else 1182 inode->i_flags &= ~S_IMMUTABLE; 1183 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND) 1184 inode->i_flags |= S_APPEND; 1185 else 1186 inode->i_flags &= ~S_APPEND; 1187 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC) 1188 inode->i_flags |= S_SYNC; 1189 else 1190 inode->i_flags &= ~S_SYNC; 1191 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME) 1192 inode->i_flags |= S_NOATIME; 1193 else 1194 inode->i_flags &= ~S_NOATIME; 1195 } 1196 1197 /* 1198 * Initialize the Linux inode, set up the operation vectors and 1199 * unlock the inode. 1200 * 1201 * When reading existing inodes from disk this is called directly 1202 * from xfs_iget, when creating a new inode it is called from 1203 * xfs_ialloc after setting up the inode. 1204 * 1205 * We are always called with an uninitialised linux inode here. 1206 * We need to initialise the necessary fields and take a reference 1207 * on it. 1208 */ 1209 void 1210 xfs_setup_inode( 1211 struct xfs_inode *ip) 1212 { 1213 struct inode *inode = &ip->i_vnode; 1214 gfp_t gfp_mask; 1215 1216 inode->i_ino = ip->i_ino; 1217 inode->i_state = I_NEW; 1218 1219 inode_sb_list_add(inode); 1220 /* make the inode look hashed for the writeback code */ 1221 hlist_add_fake(&inode->i_hash); 1222 1223 inode->i_mode = ip->i_d.di_mode; 1224 set_nlink(inode, ip->i_d.di_nlink); 1225 inode->i_uid = xfs_uid_to_kuid(ip->i_d.di_uid); 1226 inode->i_gid = xfs_gid_to_kgid(ip->i_d.di_gid); 1227 1228 switch (inode->i_mode & S_IFMT) { 1229 case S_IFBLK: 1230 case S_IFCHR: 1231 inode->i_rdev = 1232 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff, 1233 sysv_minor(ip->i_df.if_u2.if_rdev)); 1234 break; 1235 default: 1236 inode->i_rdev = 0; 1237 break; 1238 } 1239 1240 inode->i_generation = ip->i_d.di_gen; 1241 i_size_write(inode, ip->i_d.di_size); 1242 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec; 1243 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec; 1244 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec; 1245 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec; 1246 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec; 1247 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec; 1248 xfs_diflags_to_iflags(inode, ip); 1249 1250 ip->d_ops = ip->i_mount->m_nondir_inode_ops; 1251 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_nondir_ilock_class); 1252 switch (inode->i_mode & S_IFMT) { 1253 case S_IFREG: 1254 inode->i_op = &xfs_inode_operations; 1255 inode->i_fop = &xfs_file_operations; 1256 inode->i_mapping->a_ops = &xfs_address_space_operations; 1257 break; 1258 case S_IFDIR: 1259 lockdep_set_class(&ip->i_lock.mr_lock, &xfs_dir_ilock_class); 1260 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb)) 1261 inode->i_op = &xfs_dir_ci_inode_operations; 1262 else 1263 inode->i_op = &xfs_dir_inode_operations; 1264 inode->i_fop = &xfs_dir_file_operations; 1265 ip->d_ops = ip->i_mount->m_dir_inode_ops; 1266 break; 1267 case S_IFLNK: 1268 inode->i_op = &xfs_symlink_inode_operations; 1269 if (!(ip->i_df.if_flags & XFS_IFINLINE)) 1270 inode->i_mapping->a_ops = &xfs_address_space_operations; 1271 break; 1272 default: 1273 inode->i_op = &xfs_inode_operations; 1274 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1275 break; 1276 } 1277 1278 /* 1279 * Ensure all page cache allocations are done from GFP_NOFS context to 1280 * prevent direct reclaim recursion back into the filesystem and blowing 1281 * stacks or deadlocking. 1282 */ 1283 gfp_mask = mapping_gfp_mask(inode->i_mapping); 1284 mapping_set_gfp_mask(inode->i_mapping, (gfp_mask & ~(__GFP_FS))); 1285 1286 /* 1287 * If there is no attribute fork no ACL can exist on this inode, 1288 * and it can't have any file capabilities attached to it either. 1289 */ 1290 if (!XFS_IFORK_Q(ip)) { 1291 inode_has_no_xattr(inode); 1292 cache_no_acl(inode); 1293 } 1294 1295 xfs_iflags_clear(ip, XFS_INEW); 1296 barrier(); 1297 1298 unlock_new_inode(inode); 1299 } 1300