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_acl.h" 21 #include "xfs_bit.h" 22 #include "xfs_log.h" 23 #include "xfs_inum.h" 24 #include "xfs_trans.h" 25 #include "xfs_sb.h" 26 #include "xfs_ag.h" 27 #include "xfs_alloc.h" 28 #include "xfs_quota.h" 29 #include "xfs_mount.h" 30 #include "xfs_bmap_btree.h" 31 #include "xfs_dinode.h" 32 #include "xfs_inode.h" 33 #include "xfs_bmap.h" 34 #include "xfs_rtalloc.h" 35 #include "xfs_error.h" 36 #include "xfs_itable.h" 37 #include "xfs_rw.h" 38 #include "xfs_attr.h" 39 #include "xfs_buf_item.h" 40 #include "xfs_utils.h" 41 #include "xfs_vnodeops.h" 42 #include "xfs_inode_item.h" 43 #include "xfs_trace.h" 44 45 #include <linux/capability.h> 46 #include <linux/xattr.h> 47 #include <linux/namei.h> 48 #include <linux/posix_acl.h> 49 #include <linux/security.h> 50 #include <linux/fiemap.h> 51 #include <linux/slab.h> 52 53 /* 54 * Bring the timestamps in the XFS inode uptodate. 55 * 56 * Used before writing the inode to disk. 57 */ 58 void 59 xfs_synchronize_times( 60 xfs_inode_t *ip) 61 { 62 struct inode *inode = VFS_I(ip); 63 64 ip->i_d.di_atime.t_sec = (__int32_t)inode->i_atime.tv_sec; 65 ip->i_d.di_atime.t_nsec = (__int32_t)inode->i_atime.tv_nsec; 66 ip->i_d.di_ctime.t_sec = (__int32_t)inode->i_ctime.tv_sec; 67 ip->i_d.di_ctime.t_nsec = (__int32_t)inode->i_ctime.tv_nsec; 68 ip->i_d.di_mtime.t_sec = (__int32_t)inode->i_mtime.tv_sec; 69 ip->i_d.di_mtime.t_nsec = (__int32_t)inode->i_mtime.tv_nsec; 70 } 71 72 /* 73 * If the linux inode is valid, mark it dirty, else mark the dirty state 74 * in the XFS inode to make sure we pick it up when reclaiming the inode. 75 */ 76 void 77 xfs_mark_inode_dirty_sync( 78 xfs_inode_t *ip) 79 { 80 struct inode *inode = VFS_I(ip); 81 82 if (!(inode->i_state & (I_WILL_FREE|I_FREEING))) 83 mark_inode_dirty_sync(inode); 84 else { 85 barrier(); 86 ip->i_update_core = 1; 87 } 88 } 89 90 void 91 xfs_mark_inode_dirty( 92 xfs_inode_t *ip) 93 { 94 struct inode *inode = VFS_I(ip); 95 96 if (!(inode->i_state & (I_WILL_FREE|I_FREEING))) 97 mark_inode_dirty(inode); 98 else { 99 barrier(); 100 ip->i_update_core = 1; 101 } 102 103 } 104 105 106 int xfs_initxattrs(struct inode *inode, const struct xattr *xattr_array, 107 void *fs_info) 108 { 109 const struct xattr *xattr; 110 struct xfs_inode *ip = XFS_I(inode); 111 int error = 0; 112 113 for (xattr = xattr_array; xattr->name != NULL; xattr++) { 114 error = xfs_attr_set(ip, xattr->name, xattr->value, 115 xattr->value_len, ATTR_SECURE); 116 if (error < 0) 117 break; 118 } 119 return error; 120 } 121 122 /* 123 * Hook in SELinux. This is not quite correct yet, what we really need 124 * here (as we do for default ACLs) is a mechanism by which creation of 125 * these attrs can be journalled at inode creation time (along with the 126 * inode, of course, such that log replay can't cause these to be lost). 127 */ 128 129 STATIC int 130 xfs_init_security( 131 struct inode *inode, 132 struct inode *dir, 133 const struct qstr *qstr) 134 { 135 return security_inode_init_security(inode, dir, qstr, 136 &xfs_initxattrs, NULL); 137 } 138 139 static void 140 xfs_dentry_to_name( 141 struct xfs_name *namep, 142 struct dentry *dentry) 143 { 144 namep->name = dentry->d_name.name; 145 namep->len = dentry->d_name.len; 146 } 147 148 STATIC void 149 xfs_cleanup_inode( 150 struct inode *dir, 151 struct inode *inode, 152 struct dentry *dentry) 153 { 154 struct xfs_name teardown; 155 156 /* Oh, the horror. 157 * If we can't add the ACL or we fail in 158 * xfs_init_security we must back out. 159 * ENOSPC can hit here, among other things. 160 */ 161 xfs_dentry_to_name(&teardown, dentry); 162 163 xfs_remove(XFS_I(dir), &teardown, XFS_I(inode)); 164 iput(inode); 165 } 166 167 STATIC int 168 xfs_vn_mknod( 169 struct inode *dir, 170 struct dentry *dentry, 171 umode_t mode, 172 dev_t rdev) 173 { 174 struct inode *inode; 175 struct xfs_inode *ip = NULL; 176 struct posix_acl *default_acl = NULL; 177 struct xfs_name name; 178 int error; 179 180 /* 181 * Irix uses Missed'em'V split, but doesn't want to see 182 * the upper 5 bits of (14bit) major. 183 */ 184 if (S_ISCHR(mode) || S_ISBLK(mode)) { 185 if (unlikely(!sysv_valid_dev(rdev) || MAJOR(rdev) & ~0x1ff)) 186 return -EINVAL; 187 rdev = sysv_encode_dev(rdev); 188 } else { 189 rdev = 0; 190 } 191 192 if (IS_POSIXACL(dir)) { 193 default_acl = xfs_get_acl(dir, ACL_TYPE_DEFAULT); 194 if (IS_ERR(default_acl)) 195 return PTR_ERR(default_acl); 196 197 if (!default_acl) 198 mode &= ~current_umask(); 199 } 200 201 xfs_dentry_to_name(&name, dentry); 202 error = xfs_create(XFS_I(dir), &name, mode, rdev, &ip); 203 if (unlikely(error)) 204 goto out_free_acl; 205 206 inode = VFS_I(ip); 207 208 error = xfs_init_security(inode, dir, &dentry->d_name); 209 if (unlikely(error)) 210 goto out_cleanup_inode; 211 212 if (default_acl) { 213 error = -xfs_inherit_acl(inode, default_acl); 214 default_acl = NULL; 215 if (unlikely(error)) 216 goto out_cleanup_inode; 217 } 218 219 220 d_instantiate(dentry, inode); 221 return -error; 222 223 out_cleanup_inode: 224 xfs_cleanup_inode(dir, inode, dentry); 225 out_free_acl: 226 posix_acl_release(default_acl); 227 return -error; 228 } 229 230 STATIC int 231 xfs_vn_create( 232 struct inode *dir, 233 struct dentry *dentry, 234 umode_t mode, 235 struct nameidata *nd) 236 { 237 return xfs_vn_mknod(dir, dentry, mode, 0); 238 } 239 240 STATIC int 241 xfs_vn_mkdir( 242 struct inode *dir, 243 struct dentry *dentry, 244 umode_t mode) 245 { 246 return xfs_vn_mknod(dir, dentry, mode|S_IFDIR, 0); 247 } 248 249 STATIC struct dentry * 250 xfs_vn_lookup( 251 struct inode *dir, 252 struct dentry *dentry, 253 struct nameidata *nd) 254 { 255 struct xfs_inode *cip; 256 struct xfs_name name; 257 int error; 258 259 if (dentry->d_name.len >= MAXNAMELEN) 260 return ERR_PTR(-ENAMETOOLONG); 261 262 xfs_dentry_to_name(&name, dentry); 263 error = xfs_lookup(XFS_I(dir), &name, &cip, NULL); 264 if (unlikely(error)) { 265 if (unlikely(error != ENOENT)) 266 return ERR_PTR(-error); 267 d_add(dentry, NULL); 268 return NULL; 269 } 270 271 return d_splice_alias(VFS_I(cip), dentry); 272 } 273 274 STATIC struct dentry * 275 xfs_vn_ci_lookup( 276 struct inode *dir, 277 struct dentry *dentry, 278 struct nameidata *nd) 279 { 280 struct xfs_inode *ip; 281 struct xfs_name xname; 282 struct xfs_name ci_name; 283 struct qstr dname; 284 int error; 285 286 if (dentry->d_name.len >= MAXNAMELEN) 287 return ERR_PTR(-ENAMETOOLONG); 288 289 xfs_dentry_to_name(&xname, dentry); 290 error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name); 291 if (unlikely(error)) { 292 if (unlikely(error != ENOENT)) 293 return ERR_PTR(-error); 294 /* 295 * call d_add(dentry, NULL) here when d_drop_negative_children 296 * is called in xfs_vn_mknod (ie. allow negative dentries 297 * with CI filesystems). 298 */ 299 return NULL; 300 } 301 302 /* if exact match, just splice and exit */ 303 if (!ci_name.name) 304 return d_splice_alias(VFS_I(ip), dentry); 305 306 /* else case-insensitive match... */ 307 dname.name = ci_name.name; 308 dname.len = ci_name.len; 309 dentry = d_add_ci(dentry, VFS_I(ip), &dname); 310 kmem_free(ci_name.name); 311 return dentry; 312 } 313 314 STATIC int 315 xfs_vn_link( 316 struct dentry *old_dentry, 317 struct inode *dir, 318 struct dentry *dentry) 319 { 320 struct inode *inode = old_dentry->d_inode; 321 struct xfs_name name; 322 int error; 323 324 xfs_dentry_to_name(&name, dentry); 325 326 error = xfs_link(XFS_I(dir), XFS_I(inode), &name); 327 if (unlikely(error)) 328 return -error; 329 330 ihold(inode); 331 d_instantiate(dentry, inode); 332 return 0; 333 } 334 335 STATIC int 336 xfs_vn_unlink( 337 struct inode *dir, 338 struct dentry *dentry) 339 { 340 struct xfs_name name; 341 int error; 342 343 xfs_dentry_to_name(&name, dentry); 344 345 error = -xfs_remove(XFS_I(dir), &name, XFS_I(dentry->d_inode)); 346 if (error) 347 return error; 348 349 /* 350 * With unlink, the VFS makes the dentry "negative": no inode, 351 * but still hashed. This is incompatible with case-insensitive 352 * mode, so invalidate (unhash) the dentry in CI-mode. 353 */ 354 if (xfs_sb_version_hasasciici(&XFS_M(dir->i_sb)->m_sb)) 355 d_invalidate(dentry); 356 return 0; 357 } 358 359 STATIC int 360 xfs_vn_symlink( 361 struct inode *dir, 362 struct dentry *dentry, 363 const char *symname) 364 { 365 struct inode *inode; 366 struct xfs_inode *cip = NULL; 367 struct xfs_name name; 368 int error; 369 umode_t mode; 370 371 mode = S_IFLNK | 372 (irix_symlink_mode ? 0777 & ~current_umask() : S_IRWXUGO); 373 xfs_dentry_to_name(&name, dentry); 374 375 error = xfs_symlink(XFS_I(dir), &name, symname, mode, &cip); 376 if (unlikely(error)) 377 goto out; 378 379 inode = VFS_I(cip); 380 381 error = xfs_init_security(inode, dir, &dentry->d_name); 382 if (unlikely(error)) 383 goto out_cleanup_inode; 384 385 d_instantiate(dentry, inode); 386 return 0; 387 388 out_cleanup_inode: 389 xfs_cleanup_inode(dir, inode, dentry); 390 out: 391 return -error; 392 } 393 394 STATIC int 395 xfs_vn_rename( 396 struct inode *odir, 397 struct dentry *odentry, 398 struct inode *ndir, 399 struct dentry *ndentry) 400 { 401 struct inode *new_inode = ndentry->d_inode; 402 struct xfs_name oname; 403 struct xfs_name nname; 404 405 xfs_dentry_to_name(&oname, odentry); 406 xfs_dentry_to_name(&nname, ndentry); 407 408 return -xfs_rename(XFS_I(odir), &oname, XFS_I(odentry->d_inode), 409 XFS_I(ndir), &nname, new_inode ? 410 XFS_I(new_inode) : NULL); 411 } 412 413 /* 414 * careful here - this function can get called recursively, so 415 * we need to be very careful about how much stack we use. 416 * uio is kmalloced for this reason... 417 */ 418 STATIC void * 419 xfs_vn_follow_link( 420 struct dentry *dentry, 421 struct nameidata *nd) 422 { 423 char *link; 424 int error = -ENOMEM; 425 426 link = kmalloc(MAXPATHLEN+1, GFP_KERNEL); 427 if (!link) 428 goto out_err; 429 430 error = -xfs_readlink(XFS_I(dentry->d_inode), link); 431 if (unlikely(error)) 432 goto out_kfree; 433 434 nd_set_link(nd, link); 435 return NULL; 436 437 out_kfree: 438 kfree(link); 439 out_err: 440 nd_set_link(nd, ERR_PTR(error)); 441 return NULL; 442 } 443 444 STATIC void 445 xfs_vn_put_link( 446 struct dentry *dentry, 447 struct nameidata *nd, 448 void *p) 449 { 450 char *s = nd_get_link(nd); 451 452 if (!IS_ERR(s)) 453 kfree(s); 454 } 455 456 STATIC int 457 xfs_vn_getattr( 458 struct vfsmount *mnt, 459 struct dentry *dentry, 460 struct kstat *stat) 461 { 462 struct inode *inode = dentry->d_inode; 463 struct xfs_inode *ip = XFS_I(inode); 464 struct xfs_mount *mp = ip->i_mount; 465 466 trace_xfs_getattr(ip); 467 468 if (XFS_FORCED_SHUTDOWN(mp)) 469 return -XFS_ERROR(EIO); 470 471 stat->size = XFS_ISIZE(ip); 472 stat->dev = inode->i_sb->s_dev; 473 stat->mode = ip->i_d.di_mode; 474 stat->nlink = ip->i_d.di_nlink; 475 stat->uid = ip->i_d.di_uid; 476 stat->gid = ip->i_d.di_gid; 477 stat->ino = ip->i_ino; 478 stat->atime = inode->i_atime; 479 stat->mtime = inode->i_mtime; 480 stat->ctime = inode->i_ctime; 481 stat->blocks = 482 XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); 483 484 485 switch (inode->i_mode & S_IFMT) { 486 case S_IFBLK: 487 case S_IFCHR: 488 stat->blksize = BLKDEV_IOSIZE; 489 stat->rdev = MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff, 490 sysv_minor(ip->i_df.if_u2.if_rdev)); 491 break; 492 default: 493 if (XFS_IS_REALTIME_INODE(ip)) { 494 /* 495 * If the file blocks are being allocated from a 496 * realtime volume, then return the inode's realtime 497 * extent size or the realtime volume's extent size. 498 */ 499 stat->blksize = 500 xfs_get_extsz_hint(ip) << mp->m_sb.sb_blocklog; 501 } else 502 stat->blksize = xfs_preferred_iosize(mp); 503 stat->rdev = 0; 504 break; 505 } 506 507 return 0; 508 } 509 510 int 511 xfs_setattr_nonsize( 512 struct xfs_inode *ip, 513 struct iattr *iattr, 514 int flags) 515 { 516 xfs_mount_t *mp = ip->i_mount; 517 struct inode *inode = VFS_I(ip); 518 int mask = iattr->ia_valid; 519 xfs_trans_t *tp; 520 int error; 521 uid_t uid = 0, iuid = 0; 522 gid_t gid = 0, igid = 0; 523 struct xfs_dquot *udqp = NULL, *gdqp = NULL; 524 struct xfs_dquot *olddquot1 = NULL, *olddquot2 = NULL; 525 526 trace_xfs_setattr(ip); 527 528 if (mp->m_flags & XFS_MOUNT_RDONLY) 529 return XFS_ERROR(EROFS); 530 531 if (XFS_FORCED_SHUTDOWN(mp)) 532 return XFS_ERROR(EIO); 533 534 error = -inode_change_ok(inode, iattr); 535 if (error) 536 return XFS_ERROR(error); 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 = ip->i_d.di_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 = ip->i_d.di_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, uid, gid, xfs_get_projid(ip), 572 qflags, &udqp, &gdqp); 573 if (error) 574 return error; 575 } 576 577 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_NOT_SIZE); 578 error = xfs_trans_reserve(tp, 0, XFS_ICHANGE_LOG_RES(mp), 0, 0, 0); 579 if (error) 580 goto out_dqrele; 581 582 xfs_ilock(ip, XFS_ILOCK_EXCL); 583 584 /* 585 * Change file ownership. Must be the owner or privileged. 586 */ 587 if (mask & (ATTR_UID|ATTR_GID)) { 588 /* 589 * These IDs could have changed since we last looked at them. 590 * But, we're assured that if the ownership did change 591 * while we didn't have the inode locked, inode's dquot(s) 592 * would have changed also. 593 */ 594 iuid = ip->i_d.di_uid; 595 igid = ip->i_d.di_gid; 596 gid = (mask & ATTR_GID) ? iattr->ia_gid : igid; 597 uid = (mask & ATTR_UID) ? iattr->ia_uid : iuid; 598 599 /* 600 * Do a quota reservation only if uid/gid is actually 601 * going to change. 602 */ 603 if (XFS_IS_QUOTA_RUNNING(mp) && 604 ((XFS_IS_UQUOTA_ON(mp) && iuid != uid) || 605 (XFS_IS_GQUOTA_ON(mp) && igid != gid))) { 606 ASSERT(tp); 607 error = xfs_qm_vop_chown_reserve(tp, ip, udqp, gdqp, 608 capable(CAP_FOWNER) ? 609 XFS_QMOPT_FORCE_RES : 0); 610 if (error) /* out of quota */ 611 goto out_trans_cancel; 612 } 613 } 614 615 xfs_trans_ijoin(tp, ip, 0); 616 617 /* 618 * Change file ownership. Must be the owner or privileged. 619 */ 620 if (mask & (ATTR_UID|ATTR_GID)) { 621 /* 622 * CAP_FSETID overrides the following restrictions: 623 * 624 * The set-user-ID and set-group-ID bits of a file will be 625 * cleared upon successful return from chown() 626 */ 627 if ((ip->i_d.di_mode & (S_ISUID|S_ISGID)) && 628 !capable(CAP_FSETID)) 629 ip->i_d.di_mode &= ~(S_ISUID|S_ISGID); 630 631 /* 632 * Change the ownerships and register quota modifications 633 * in the transaction. 634 */ 635 if (iuid != uid) { 636 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_UQUOTA_ON(mp)) { 637 ASSERT(mask & ATTR_UID); 638 ASSERT(udqp); 639 olddquot1 = xfs_qm_vop_chown(tp, ip, 640 &ip->i_udquot, udqp); 641 } 642 ip->i_d.di_uid = uid; 643 inode->i_uid = uid; 644 } 645 if (igid != gid) { 646 if (XFS_IS_QUOTA_RUNNING(mp) && XFS_IS_GQUOTA_ON(mp)) { 647 ASSERT(!XFS_IS_PQUOTA_ON(mp)); 648 ASSERT(mask & ATTR_GID); 649 ASSERT(gdqp); 650 olddquot2 = xfs_qm_vop_chown(tp, ip, 651 &ip->i_gdquot, gdqp); 652 } 653 ip->i_d.di_gid = gid; 654 inode->i_gid = gid; 655 } 656 } 657 658 /* 659 * Change file access modes. 660 */ 661 if (mask & ATTR_MODE) { 662 umode_t mode = iattr->ia_mode; 663 664 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID)) 665 mode &= ~S_ISGID; 666 667 ip->i_d.di_mode &= S_IFMT; 668 ip->i_d.di_mode |= mode & ~S_IFMT; 669 670 inode->i_mode &= S_IFMT; 671 inode->i_mode |= mode & ~S_IFMT; 672 } 673 674 /* 675 * Change file access or modified times. 676 */ 677 if (mask & ATTR_ATIME) { 678 inode->i_atime = iattr->ia_atime; 679 ip->i_d.di_atime.t_sec = iattr->ia_atime.tv_sec; 680 ip->i_d.di_atime.t_nsec = iattr->ia_atime.tv_nsec; 681 ip->i_update_core = 1; 682 } 683 if (mask & ATTR_CTIME) { 684 inode->i_ctime = iattr->ia_ctime; 685 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec; 686 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec; 687 ip->i_update_core = 1; 688 } 689 if (mask & ATTR_MTIME) { 690 inode->i_mtime = iattr->ia_mtime; 691 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec; 692 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec; 693 ip->i_update_core = 1; 694 } 695 696 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 697 698 XFS_STATS_INC(xs_ig_attrchg); 699 700 if (mp->m_flags & XFS_MOUNT_WSYNC) 701 xfs_trans_set_sync(tp); 702 error = xfs_trans_commit(tp, 0); 703 704 xfs_iunlock(ip, XFS_ILOCK_EXCL); 705 706 /* 707 * Release any dquot(s) the inode had kept before chown. 708 */ 709 xfs_qm_dqrele(olddquot1); 710 xfs_qm_dqrele(olddquot2); 711 xfs_qm_dqrele(udqp); 712 xfs_qm_dqrele(gdqp); 713 714 if (error) 715 return XFS_ERROR(error); 716 717 /* 718 * XXX(hch): Updating the ACL entries is not atomic vs the i_mode 719 * update. We could avoid this with linked transactions 720 * and passing down the transaction pointer all the way 721 * to attr_set. No previous user of the generic 722 * Posix ACL code seems to care about this issue either. 723 */ 724 if ((mask & ATTR_MODE) && !(flags & XFS_ATTR_NOACL)) { 725 error = -xfs_acl_chmod(inode); 726 if (error) 727 return XFS_ERROR(error); 728 } 729 730 return 0; 731 732 out_trans_cancel: 733 xfs_trans_cancel(tp, 0); 734 xfs_iunlock(ip, XFS_ILOCK_EXCL); 735 out_dqrele: 736 xfs_qm_dqrele(udqp); 737 xfs_qm_dqrele(gdqp); 738 return error; 739 } 740 741 /* 742 * Truncate file. Must have write permission and not be a directory. 743 */ 744 int 745 xfs_setattr_size( 746 struct xfs_inode *ip, 747 struct iattr *iattr, 748 int flags) 749 { 750 struct xfs_mount *mp = ip->i_mount; 751 struct inode *inode = VFS_I(ip); 752 int mask = iattr->ia_valid; 753 xfs_off_t oldsize, newsize; 754 struct xfs_trans *tp; 755 int error; 756 uint lock_flags; 757 uint commit_flags = 0; 758 759 trace_xfs_setattr(ip); 760 761 if (mp->m_flags & XFS_MOUNT_RDONLY) 762 return XFS_ERROR(EROFS); 763 764 if (XFS_FORCED_SHUTDOWN(mp)) 765 return XFS_ERROR(EIO); 766 767 error = -inode_change_ok(inode, iattr); 768 if (error) 769 return XFS_ERROR(error); 770 771 ASSERT(S_ISREG(ip->i_d.di_mode)); 772 ASSERT((mask & (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_ATIME|ATTR_ATIME_SET| 773 ATTR_MTIME_SET|ATTR_KILL_SUID|ATTR_KILL_SGID| 774 ATTR_KILL_PRIV|ATTR_TIMES_SET)) == 0); 775 776 lock_flags = XFS_ILOCK_EXCL; 777 if (!(flags & XFS_ATTR_NOLOCK)) 778 lock_flags |= XFS_IOLOCK_EXCL; 779 xfs_ilock(ip, lock_flags); 780 781 oldsize = inode->i_size; 782 newsize = iattr->ia_size; 783 784 /* 785 * Short circuit the truncate case for zero length files. 786 */ 787 if (newsize == 0 && oldsize == 0 && ip->i_d.di_nextents == 0) { 788 if (!(mask & (ATTR_CTIME|ATTR_MTIME))) 789 goto out_unlock; 790 791 /* 792 * Use the regular setattr path to update the timestamps. 793 */ 794 xfs_iunlock(ip, lock_flags); 795 iattr->ia_valid &= ~ATTR_SIZE; 796 return xfs_setattr_nonsize(ip, iattr, 0); 797 } 798 799 /* 800 * Make sure that the dquots are attached to the inode. 801 */ 802 error = xfs_qm_dqattach_locked(ip, 0); 803 if (error) 804 goto out_unlock; 805 806 /* 807 * Now we can make the changes. Before we join the inode to the 808 * transaction, take care of the part of the truncation that must be 809 * done without the inode lock. This needs to be done before joining 810 * the inode to the transaction, because the inode cannot be unlocked 811 * once it is a part of the transaction. 812 */ 813 if (newsize > oldsize) { 814 /* 815 * Do the first part of growing a file: zero any data in the 816 * last block that is beyond the old EOF. We need to do this 817 * before the inode is joined to the transaction to modify 818 * i_size. 819 */ 820 error = xfs_zero_eof(ip, newsize, oldsize); 821 if (error) 822 goto out_unlock; 823 } 824 xfs_iunlock(ip, XFS_ILOCK_EXCL); 825 lock_flags &= ~XFS_ILOCK_EXCL; 826 827 /* 828 * We are going to log the inode size change in this transaction so 829 * any previous writes that are beyond the on disk EOF and the new 830 * EOF that have not been written out need to be written here. If we 831 * do not write the data out, we expose ourselves to the null files 832 * problem. 833 * 834 * Only flush from the on disk size to the smaller of the in memory 835 * file size or the new size as that's the range we really care about 836 * here and prevents waiting for other data not within the range we 837 * care about here. 838 */ 839 if (oldsize != ip->i_d.di_size && newsize > ip->i_d.di_size) { 840 error = xfs_flush_pages(ip, ip->i_d.di_size, newsize, 0, 841 FI_NONE); 842 if (error) 843 goto out_unlock; 844 } 845 846 /* 847 * Wait for all direct I/O to complete. 848 */ 849 inode_dio_wait(inode); 850 851 error = -block_truncate_page(inode->i_mapping, newsize, xfs_get_blocks); 852 if (error) 853 goto out_unlock; 854 855 tp = xfs_trans_alloc(mp, XFS_TRANS_SETATTR_SIZE); 856 error = xfs_trans_reserve(tp, 0, XFS_ITRUNCATE_LOG_RES(mp), 0, 857 XFS_TRANS_PERM_LOG_RES, 858 XFS_ITRUNCATE_LOG_COUNT); 859 if (error) 860 goto out_trans_cancel; 861 862 truncate_setsize(inode, newsize); 863 864 commit_flags = XFS_TRANS_RELEASE_LOG_RES; 865 lock_flags |= XFS_ILOCK_EXCL; 866 867 xfs_ilock(ip, XFS_ILOCK_EXCL); 868 869 xfs_trans_ijoin(tp, ip, 0); 870 871 /* 872 * Only change the c/mtime if we are changing the size or we are 873 * explicitly asked to change it. This handles the semantic difference 874 * between truncate() and ftruncate() as implemented in the VFS. 875 * 876 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a 877 * special case where we need to update the times despite not having 878 * these flags set. For all other operations the VFS set these flags 879 * explicitly if it wants a timestamp update. 880 */ 881 if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME)))) { 882 iattr->ia_ctime = iattr->ia_mtime = 883 current_fs_time(inode->i_sb); 884 mask |= ATTR_CTIME | ATTR_MTIME; 885 } 886 887 /* 888 * The first thing we do is set the size to new_size permanently on 889 * disk. This way we don't have to worry about anyone ever being able 890 * to look at the data being freed even in the face of a crash. 891 * What we're getting around here is the case where we free a block, it 892 * is allocated to another file, it is written to, and then we crash. 893 * If the new data gets written to the file but the log buffers 894 * containing the free and reallocation don't, then we'd end up with 895 * garbage in the blocks being freed. As long as we make the new size 896 * permanent before actually freeing any blocks it doesn't matter if 897 * they get written to. 898 */ 899 ip->i_d.di_size = newsize; 900 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 901 902 if (newsize <= oldsize) { 903 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK, newsize); 904 if (error) 905 goto out_trans_abort; 906 907 /* 908 * Truncated "down", so we're removing references to old data 909 * here - if we delay flushing for a long time, we expose 910 * ourselves unduly to the notorious NULL files problem. So, 911 * we mark this inode and flush it when the file is closed, 912 * and do not wait the usual (long) time for writeout. 913 */ 914 xfs_iflags_set(ip, XFS_ITRUNCATED); 915 } 916 917 if (mask & ATTR_CTIME) { 918 inode->i_ctime = iattr->ia_ctime; 919 ip->i_d.di_ctime.t_sec = iattr->ia_ctime.tv_sec; 920 ip->i_d.di_ctime.t_nsec = iattr->ia_ctime.tv_nsec; 921 ip->i_update_core = 1; 922 } 923 if (mask & ATTR_MTIME) { 924 inode->i_mtime = iattr->ia_mtime; 925 ip->i_d.di_mtime.t_sec = iattr->ia_mtime.tv_sec; 926 ip->i_d.di_mtime.t_nsec = iattr->ia_mtime.tv_nsec; 927 ip->i_update_core = 1; 928 } 929 930 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE); 931 932 XFS_STATS_INC(xs_ig_attrchg); 933 934 if (mp->m_flags & XFS_MOUNT_WSYNC) 935 xfs_trans_set_sync(tp); 936 937 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES); 938 out_unlock: 939 if (lock_flags) 940 xfs_iunlock(ip, lock_flags); 941 return error; 942 943 out_trans_abort: 944 commit_flags |= XFS_TRANS_ABORT; 945 out_trans_cancel: 946 xfs_trans_cancel(tp, commit_flags); 947 goto out_unlock; 948 } 949 950 STATIC int 951 xfs_vn_setattr( 952 struct dentry *dentry, 953 struct iattr *iattr) 954 { 955 if (iattr->ia_valid & ATTR_SIZE) 956 return -xfs_setattr_size(XFS_I(dentry->d_inode), iattr, 0); 957 return -xfs_setattr_nonsize(XFS_I(dentry->d_inode), iattr, 0); 958 } 959 960 #define XFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC|FIEMAP_FLAG_XATTR) 961 962 /* 963 * Call fiemap helper to fill in user data. 964 * Returns positive errors to xfs_getbmap. 965 */ 966 STATIC int 967 xfs_fiemap_format( 968 void **arg, 969 struct getbmapx *bmv, 970 int *full) 971 { 972 int error; 973 struct fiemap_extent_info *fieinfo = *arg; 974 u32 fiemap_flags = 0; 975 u64 logical, physical, length; 976 977 /* Do nothing for a hole */ 978 if (bmv->bmv_block == -1LL) 979 return 0; 980 981 logical = BBTOB(bmv->bmv_offset); 982 physical = BBTOB(bmv->bmv_block); 983 length = BBTOB(bmv->bmv_length); 984 985 if (bmv->bmv_oflags & BMV_OF_PREALLOC) 986 fiemap_flags |= FIEMAP_EXTENT_UNWRITTEN; 987 else if (bmv->bmv_oflags & BMV_OF_DELALLOC) { 988 fiemap_flags |= FIEMAP_EXTENT_DELALLOC; 989 physical = 0; /* no block yet */ 990 } 991 if (bmv->bmv_oflags & BMV_OF_LAST) 992 fiemap_flags |= FIEMAP_EXTENT_LAST; 993 994 error = fiemap_fill_next_extent(fieinfo, logical, physical, 995 length, fiemap_flags); 996 if (error > 0) { 997 error = 0; 998 *full = 1; /* user array now full */ 999 } 1000 1001 return -error; 1002 } 1003 1004 STATIC int 1005 xfs_vn_fiemap( 1006 struct inode *inode, 1007 struct fiemap_extent_info *fieinfo, 1008 u64 start, 1009 u64 length) 1010 { 1011 xfs_inode_t *ip = XFS_I(inode); 1012 struct getbmapx bm; 1013 int error; 1014 1015 error = fiemap_check_flags(fieinfo, XFS_FIEMAP_FLAGS); 1016 if (error) 1017 return error; 1018 1019 /* Set up bmap header for xfs internal routine */ 1020 bm.bmv_offset = BTOBB(start); 1021 /* Special case for whole file */ 1022 if (length == FIEMAP_MAX_OFFSET) 1023 bm.bmv_length = -1LL; 1024 else 1025 bm.bmv_length = BTOBB(length); 1026 1027 /* We add one because in getbmap world count includes the header */ 1028 bm.bmv_count = !fieinfo->fi_extents_max ? MAXEXTNUM : 1029 fieinfo->fi_extents_max + 1; 1030 bm.bmv_count = min_t(__s32, bm.bmv_count, 1031 (PAGE_SIZE * 16 / sizeof(struct getbmapx))); 1032 bm.bmv_iflags = BMV_IF_PREALLOC | BMV_IF_NO_HOLES; 1033 if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) 1034 bm.bmv_iflags |= BMV_IF_ATTRFORK; 1035 if (!(fieinfo->fi_flags & FIEMAP_FLAG_SYNC)) 1036 bm.bmv_iflags |= BMV_IF_DELALLOC; 1037 1038 error = xfs_getbmap(ip, &bm, xfs_fiemap_format, fieinfo); 1039 if (error) 1040 return -error; 1041 1042 return 0; 1043 } 1044 1045 static const struct inode_operations xfs_inode_operations = { 1046 .get_acl = xfs_get_acl, 1047 .getattr = xfs_vn_getattr, 1048 .setattr = xfs_vn_setattr, 1049 .setxattr = generic_setxattr, 1050 .getxattr = generic_getxattr, 1051 .removexattr = generic_removexattr, 1052 .listxattr = xfs_vn_listxattr, 1053 .fiemap = xfs_vn_fiemap, 1054 }; 1055 1056 static const struct inode_operations xfs_dir_inode_operations = { 1057 .create = xfs_vn_create, 1058 .lookup = xfs_vn_lookup, 1059 .link = xfs_vn_link, 1060 .unlink = xfs_vn_unlink, 1061 .symlink = xfs_vn_symlink, 1062 .mkdir = xfs_vn_mkdir, 1063 /* 1064 * Yes, XFS uses the same method for rmdir and unlink. 1065 * 1066 * There are some subtile differences deeper in the code, 1067 * but we use S_ISDIR to check for those. 1068 */ 1069 .rmdir = xfs_vn_unlink, 1070 .mknod = xfs_vn_mknod, 1071 .rename = xfs_vn_rename, 1072 .get_acl = xfs_get_acl, 1073 .getattr = xfs_vn_getattr, 1074 .setattr = xfs_vn_setattr, 1075 .setxattr = generic_setxattr, 1076 .getxattr = generic_getxattr, 1077 .removexattr = generic_removexattr, 1078 .listxattr = xfs_vn_listxattr, 1079 }; 1080 1081 static const struct inode_operations xfs_dir_ci_inode_operations = { 1082 .create = xfs_vn_create, 1083 .lookup = xfs_vn_ci_lookup, 1084 .link = xfs_vn_link, 1085 .unlink = xfs_vn_unlink, 1086 .symlink = xfs_vn_symlink, 1087 .mkdir = xfs_vn_mkdir, 1088 /* 1089 * Yes, XFS uses the same method for rmdir and unlink. 1090 * 1091 * There are some subtile differences deeper in the code, 1092 * but we use S_ISDIR to check for those. 1093 */ 1094 .rmdir = xfs_vn_unlink, 1095 .mknod = xfs_vn_mknod, 1096 .rename = xfs_vn_rename, 1097 .get_acl = xfs_get_acl, 1098 .getattr = xfs_vn_getattr, 1099 .setattr = xfs_vn_setattr, 1100 .setxattr = generic_setxattr, 1101 .getxattr = generic_getxattr, 1102 .removexattr = generic_removexattr, 1103 .listxattr = xfs_vn_listxattr, 1104 }; 1105 1106 static const struct inode_operations xfs_symlink_inode_operations = { 1107 .readlink = generic_readlink, 1108 .follow_link = xfs_vn_follow_link, 1109 .put_link = xfs_vn_put_link, 1110 .get_acl = xfs_get_acl, 1111 .getattr = xfs_vn_getattr, 1112 .setattr = xfs_vn_setattr, 1113 .setxattr = generic_setxattr, 1114 .getxattr = generic_getxattr, 1115 .removexattr = generic_removexattr, 1116 .listxattr = xfs_vn_listxattr, 1117 }; 1118 1119 STATIC void 1120 xfs_diflags_to_iflags( 1121 struct inode *inode, 1122 struct xfs_inode *ip) 1123 { 1124 if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE) 1125 inode->i_flags |= S_IMMUTABLE; 1126 else 1127 inode->i_flags &= ~S_IMMUTABLE; 1128 if (ip->i_d.di_flags & XFS_DIFLAG_APPEND) 1129 inode->i_flags |= S_APPEND; 1130 else 1131 inode->i_flags &= ~S_APPEND; 1132 if (ip->i_d.di_flags & XFS_DIFLAG_SYNC) 1133 inode->i_flags |= S_SYNC; 1134 else 1135 inode->i_flags &= ~S_SYNC; 1136 if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME) 1137 inode->i_flags |= S_NOATIME; 1138 else 1139 inode->i_flags &= ~S_NOATIME; 1140 } 1141 1142 /* 1143 * Initialize the Linux inode, set up the operation vectors and 1144 * unlock the inode. 1145 * 1146 * When reading existing inodes from disk this is called directly 1147 * from xfs_iget, when creating a new inode it is called from 1148 * xfs_ialloc after setting up the inode. 1149 * 1150 * We are always called with an uninitialised linux inode here. 1151 * We need to initialise the necessary fields and take a reference 1152 * on it. 1153 */ 1154 void 1155 xfs_setup_inode( 1156 struct xfs_inode *ip) 1157 { 1158 struct inode *inode = &ip->i_vnode; 1159 1160 inode->i_ino = ip->i_ino; 1161 inode->i_state = I_NEW; 1162 1163 inode_sb_list_add(inode); 1164 /* make the inode look hashed for the writeback code */ 1165 hlist_add_fake(&inode->i_hash); 1166 1167 inode->i_mode = ip->i_d.di_mode; 1168 set_nlink(inode, ip->i_d.di_nlink); 1169 inode->i_uid = ip->i_d.di_uid; 1170 inode->i_gid = ip->i_d.di_gid; 1171 1172 switch (inode->i_mode & S_IFMT) { 1173 case S_IFBLK: 1174 case S_IFCHR: 1175 inode->i_rdev = 1176 MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff, 1177 sysv_minor(ip->i_df.if_u2.if_rdev)); 1178 break; 1179 default: 1180 inode->i_rdev = 0; 1181 break; 1182 } 1183 1184 inode->i_generation = ip->i_d.di_gen; 1185 i_size_write(inode, ip->i_d.di_size); 1186 inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec; 1187 inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec; 1188 inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec; 1189 inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec; 1190 inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec; 1191 inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec; 1192 xfs_diflags_to_iflags(inode, ip); 1193 1194 switch (inode->i_mode & S_IFMT) { 1195 case S_IFREG: 1196 inode->i_op = &xfs_inode_operations; 1197 inode->i_fop = &xfs_file_operations; 1198 inode->i_mapping->a_ops = &xfs_address_space_operations; 1199 break; 1200 case S_IFDIR: 1201 if (xfs_sb_version_hasasciici(&XFS_M(inode->i_sb)->m_sb)) 1202 inode->i_op = &xfs_dir_ci_inode_operations; 1203 else 1204 inode->i_op = &xfs_dir_inode_operations; 1205 inode->i_fop = &xfs_dir_file_operations; 1206 break; 1207 case S_IFLNK: 1208 inode->i_op = &xfs_symlink_inode_operations; 1209 if (!(ip->i_df.if_flags & XFS_IFINLINE)) 1210 inode->i_mapping->a_ops = &xfs_address_space_operations; 1211 break; 1212 default: 1213 inode->i_op = &xfs_inode_operations; 1214 init_special_inode(inode, inode->i_mode, inode->i_rdev); 1215 break; 1216 } 1217 1218 /* 1219 * If there is no attribute fork no ACL can exist on this inode, 1220 * and it can't have any file capabilities attached to it either. 1221 */ 1222 if (!XFS_IFORK_Q(ip)) { 1223 inode_has_no_xattr(inode); 1224 cache_no_acl(inode); 1225 } 1226 1227 xfs_iflags_clear(ip, XFS_INEW); 1228 barrier(); 1229 1230 unlock_new_inode(inode); 1231 } 1232