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