zfs_ctldir.c revision 182371
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26#pragma ident "%Z%%M% %I% %E% SMI" 27 28/* 29 * ZFS control directory (a.k.a. ".zfs") 30 * 31 * This directory provides a common location for all ZFS meta-objects. 32 * Currently, this is only the 'snapshot' directory, but this may expand in the 33 * future. The elements are built using the GFS primitives, as the hierarchy 34 * does not actually exist on disk. 35 * 36 * For 'snapshot', we don't want to have all snapshots always mounted, because 37 * this would take up a huge amount of space in /etc/mnttab. We have three 38 * types of objects: 39 * 40 * ctldir ------> snapshotdir -------> snapshot 41 * | 42 * | 43 * V 44 * mounted fs 45 * 46 * The 'snapshot' node contains just enough information to lookup '..' and act 47 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we 48 * perform an automount of the underlying filesystem and return the 49 * corresponding vnode. 50 * 51 * All mounts are handled automatically by the kernel, but unmounts are 52 * (currently) handled from user land. The main reason is that there is no 53 * reliable way to auto-unmount the filesystem when it's "no longer in use". 54 * When the user unmounts a filesystem, we call zfsctl_unmount(), which 55 * unmounts any snapshots within the snapshot directory. 56 */ 57 58#include <sys/zfs_context.h> 59#include <sys/zfs_ctldir.h> 60#include <sys/zfs_ioctl.h> 61#include <sys/zfs_vfsops.h> 62#include <sys/namei.h> 63#include <sys/gfs.h> 64#include <sys/stat.h> 65#include <sys/dmu.h> 66#include <sys/mount.h> 67 68typedef struct { 69 char *se_name; 70 vnode_t *se_root; 71 avl_node_t se_node; 72} zfs_snapentry_t; 73 74static int 75snapentry_compare(const void *a, const void *b) 76{ 77 const zfs_snapentry_t *sa = a; 78 const zfs_snapentry_t *sb = b; 79 int ret = strcmp(sa->se_name, sb->se_name); 80 81 if (ret < 0) 82 return (-1); 83 else if (ret > 0) 84 return (1); 85 else 86 return (0); 87} 88 89static struct vop_vector zfsctl_ops_root; 90static struct vop_vector zfsctl_ops_snapdir; 91static struct vop_vector zfsctl_ops_snapshot; 92 93static vnode_t *zfsctl_mknode_snapdir(vnode_t *); 94static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset); 95 96typedef struct zfsctl_node { 97 gfs_dir_t zc_gfs_private; 98 uint64_t zc_id; 99 timestruc_t zc_cmtime; /* ctime and mtime, always the same */ 100} zfsctl_node_t; 101 102typedef struct zfsctl_snapdir { 103 zfsctl_node_t sd_node; 104 kmutex_t sd_lock; 105 avl_tree_t sd_snaps; 106} zfsctl_snapdir_t; 107 108/* 109 * Root directory elements. We have only a single static entry, 'snapshot'. 110 */ 111static gfs_dirent_t zfsctl_root_entries[] = { 112 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE }, 113 { NULL } 114}; 115 116/* include . and .. in the calculation */ 117#define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \ 118 sizeof (gfs_dirent_t)) + 1) 119 120 121/* 122 * Initialize the various GFS pieces we'll need to create and manipulate .zfs 123 * directories. This is called from the ZFS init routine, and initializes the 124 * vnode ops vectors that we'll be using. 125 */ 126void 127zfsctl_init(void) 128{ 129} 130 131void 132zfsctl_fini(void) 133{ 134} 135 136/* 137 * Return the inode number associated with the 'snapshot' directory. 138 */ 139/* ARGSUSED */ 140static ino64_t 141zfsctl_root_inode_cb(vnode_t *vp, int index) 142{ 143 ASSERT(index == 0); 144 return (ZFSCTL_INO_SNAPDIR); 145} 146 147/* 148 * Create the '.zfs' directory. This directory is cached as part of the VFS 149 * structure. This results in a hold on the vfs_t. The code in zfs_umount() 150 * therefore checks against a vfs_count of 2 instead of 1. This reference 151 * is removed when the ctldir is destroyed in the unmount. 152 */ 153void 154zfsctl_create(zfsvfs_t *zfsvfs) 155{ 156 vnode_t *vp, *rvp; 157 zfsctl_node_t *zcp; 158 159 ASSERT(zfsvfs->z_ctldir == NULL); 160 161 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs, 162 &zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries, 163 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL); 164 zcp = vp->v_data; 165 zcp->zc_id = ZFSCTL_INO_ROOT; 166 167 VERIFY(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp, curthread) == 0); 168 ZFS_TIME_DECODE(&zcp->zc_cmtime, VTOZ(rvp)->z_phys->zp_crtime); 169 VN_URELE(rvp); 170 171 /* 172 * We're only faking the fact that we have a root of a filesystem for 173 * the sake of the GFS interfaces. Undo the flag manipulation it did 174 * for us. 175 */ 176 vp->v_vflag &= ~VV_ROOT; 177 178 zfsvfs->z_ctldir = vp; 179} 180 181/* 182 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted. 183 * There might still be more references if we were force unmounted, but only 184 * new zfs_inactive() calls can occur and they don't reference .zfs 185 */ 186void 187zfsctl_destroy(zfsvfs_t *zfsvfs) 188{ 189 VN_RELE(zfsvfs->z_ctldir); 190 zfsvfs->z_ctldir = NULL; 191} 192 193/* 194 * Given a root znode, retrieve the associated .zfs directory. 195 * Add a hold to the vnode and return it. 196 */ 197vnode_t * 198zfsctl_root(znode_t *zp) 199{ 200 ASSERT(zfs_has_ctldir(zp)); 201 VN_HOLD(zp->z_zfsvfs->z_ctldir); 202 return (zp->z_zfsvfs->z_ctldir); 203} 204 205/* 206 * Common open routine. Disallow any write access. 207 */ 208/* ARGSUSED */ 209static int 210zfsctl_common_open(struct vop_open_args *ap) 211{ 212 int flags = ap->a_mode; 213 214 if (flags & FWRITE) 215 return (EACCES); 216 217 return (0); 218} 219 220/* 221 * Common close routine. Nothing to do here. 222 */ 223/* ARGSUSED */ 224static int 225zfsctl_common_close(struct vop_close_args *ap) 226{ 227 return (0); 228} 229 230/* 231 * Common access routine. Disallow writes. 232 */ 233/* ARGSUSED */ 234static int 235zfsctl_common_access(ap) 236 struct vop_access_args /* { 237 struct vnode *a_vp; 238 int a_mode; 239 struct ucred *a_cred; 240 struct thread *a_td; 241 } */ *ap; 242{ 243 int mode = ap->a_mode; 244 245 if (mode & VWRITE) 246 return (EACCES); 247 248 return (0); 249} 250 251/* 252 * Common getattr function. Fill in basic information. 253 */ 254static void 255zfsctl_common_getattr(vnode_t *vp, vattr_t *vap) 256{ 257 zfsctl_node_t *zcp = vp->v_data; 258 timestruc_t now; 259 260 vap->va_uid = 0; 261 vap->va_gid = 0; 262 vap->va_rdev = 0; 263 /* 264 * We are a purly virtual object, so we have no 265 * blocksize or allocated blocks. 266 */ 267 vap->va_blksize = 0; 268 vap->va_nblocks = 0; 269 vap->va_seq = 0; 270 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 271 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP | 272 S_IROTH | S_IXOTH; 273 vap->va_type = VDIR; 274 /* 275 * We live in the now (for atime). 276 */ 277 gethrestime(&now); 278 vap->va_atime = now; 279 vap->va_mtime = vap->va_ctime = vap->va_birthtime = zcp->zc_cmtime; 280 /* FreeBSD: Reset chflags(2) flags. */ 281 vap->va_flags = 0; 282} 283 284static int 285zfsctl_common_fid(ap) 286 struct vop_fid_args /* { 287 struct vnode *a_vp; 288 struct fid *a_fid; 289 } */ *ap; 290{ 291 vnode_t *vp = ap->a_vp; 292 fid_t *fidp = (void *)ap->a_fid; 293 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 294 zfsctl_node_t *zcp = vp->v_data; 295 uint64_t object = zcp->zc_id; 296 zfid_short_t *zfid; 297 int i; 298 299 ZFS_ENTER(zfsvfs); 300 301 fidp->fid_len = SHORT_FID_LEN; 302 303 zfid = (zfid_short_t *)fidp; 304 305 zfid->zf_len = SHORT_FID_LEN; 306 307 for (i = 0; i < sizeof (zfid->zf_object); i++) 308 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 309 310 /* .zfs znodes always have a generation number of 0 */ 311 for (i = 0; i < sizeof (zfid->zf_gen); i++) 312 zfid->zf_gen[i] = 0; 313 314 ZFS_EXIT(zfsvfs); 315 return (0); 316} 317 318static int 319zfsctl_common_reclaim(ap) 320 struct vop_reclaim_args /* { 321 struct vnode *a_vp; 322 struct thread *a_td; 323 } */ *ap; 324{ 325 vnode_t *vp = ap->a_vp; 326 327 /* 328 * Destroy the vm object and flush associated pages. 329 */ 330 vnode_destroy_vobject(vp); 331 VI_LOCK(vp); 332 vp->v_data = NULL; 333 VI_UNLOCK(vp); 334 return (0); 335} 336 337/* 338 * .zfs inode namespace 339 * 340 * We need to generate unique inode numbers for all files and directories 341 * within the .zfs pseudo-filesystem. We use the following scheme: 342 * 343 * ENTRY ZFSCTL_INODE 344 * .zfs 1 345 * .zfs/snapshot 2 346 * .zfs/snapshot/<snap> objectid(snap) 347 */ 348 349#define ZFSCTL_INO_SNAP(id) (id) 350 351/* 352 * Get root directory attributes. 353 */ 354/* ARGSUSED */ 355static int 356zfsctl_root_getattr(ap) 357 struct vop_getattr_args /* { 358 struct vnode *a_vp; 359 struct vattr *a_vap; 360 struct ucred *a_cred; 361 } */ *ap; 362{ 363 struct vnode *vp = ap->a_vp; 364 struct vattr *vap = ap->a_vap; 365 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 366 367 ZFS_ENTER(zfsvfs); 368 vap->va_nodeid = ZFSCTL_INO_ROOT; 369 vap->va_nlink = vap->va_size = NROOT_ENTRIES; 370 371 zfsctl_common_getattr(vp, vap); 372 ZFS_EXIT(zfsvfs); 373 374 return (0); 375} 376 377/* 378 * Special case the handling of "..". 379 */ 380/* ARGSUSED */ 381int 382zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, 383 int flags, vnode_t *rdir, cred_t *cr) 384{ 385 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 386 int err; 387 388 ZFS_ENTER(zfsvfs); 389 390 if (strcmp(nm, "..") == 0) { 391 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp, curthread); 392 if (err == 0) 393 VOP_UNLOCK(*vpp, 0); 394 } else { 395 err = gfs_dir_lookup(dvp, nm, vpp); 396 } 397 398 ZFS_EXIT(zfsvfs); 399 400 return (err); 401} 402 403/* 404 * Special case the handling of "..". 405 */ 406/* ARGSUSED */ 407int 408zfsctl_root_lookup_vop(ap) 409 struct vop_lookup_args /* { 410 struct vnode *a_dvp; 411 struct vnode **a_vpp; 412 struct componentname *a_cnp; 413 } */ *ap; 414{ 415 vnode_t *dvp = ap->a_dvp; 416 vnode_t **vpp = ap->a_vpp; 417 cred_t *cr = ap->a_cnp->cn_cred; 418 int flags = ap->a_cnp->cn_flags; 419 int nameiop = ap->a_cnp->cn_nameiop; 420 char nm[NAME_MAX + 1]; 421 int err; 422 423 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE)) 424 return (EOPNOTSUPP); 425 426 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 427 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 428 429 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr); 430 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) 431 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 432 433 return (err); 434} 435 436static struct vop_vector zfsctl_ops_root = { 437 .vop_default = &default_vnodeops, 438 .vop_open = zfsctl_common_open, 439 .vop_close = zfsctl_common_close, 440 .vop_ioctl = VOP_EINVAL, 441 .vop_getattr = zfsctl_root_getattr, 442 .vop_access = zfsctl_common_access, 443 .vop_readdir = gfs_vop_readdir, 444 .vop_lookup = zfsctl_root_lookup_vop, 445 .vop_inactive = gfs_vop_inactive, 446 .vop_reclaim = zfsctl_common_reclaim, 447 .vop_fid = zfsctl_common_fid, 448}; 449 450static int 451zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 452{ 453 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 454 455 dmu_objset_name(os, zname); 456 if (strlen(zname) + 1 + strlen(name) >= len) 457 return (ENAMETOOLONG); 458 (void) strcat(zname, "@"); 459 (void) strcat(zname, name); 460 return (0); 461} 462 463static int 464zfsctl_unmount_snap(vnode_t *dvp, const char *name, int force, cred_t *cr) 465{ 466 zfsctl_snapdir_t *sdp = dvp->v_data; 467 zfs_snapentry_t search, *sep; 468 struct vop_inactive_args ap; 469 avl_index_t where; 470 int err; 471 472 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 473 474 search.se_name = (char *)name; 475 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) 476 return (ENOENT); 477 478 ASSERT(vn_ismntpt(sep->se_root)); 479 480 /* this will be dropped by dounmount() */ 481 if ((err = vn_vfswlock(sep->se_root)) != 0) 482 return (err); 483 484 err = dounmount(vn_mountedvfs(sep->se_root), force, curthread); 485 if (err) 486 return (err); 487 ASSERT(sep->se_root->v_count == 1); 488 ap.a_vp = sep->se_root; 489 gfs_vop_inactive(&ap); 490 491 avl_remove(&sdp->sd_snaps, sep); 492 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 493 kmem_free(sep, sizeof (zfs_snapentry_t)); 494 495 return (0); 496} 497 498#if 0 499static void 500zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 501{ 502 avl_index_t where; 503 vfs_t *vfsp; 504 refstr_t *pathref; 505 char newpath[MAXNAMELEN]; 506 char *tail; 507 508 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 509 ASSERT(sep != NULL); 510 511 vfsp = vn_mountedvfs(sep->se_root); 512 ASSERT(vfsp != NULL); 513 514 vfs_lock_wait(vfsp); 515 516 /* 517 * Change the name in the AVL tree. 518 */ 519 avl_remove(&sdp->sd_snaps, sep); 520 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 521 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 522 (void) strcpy(sep->se_name, nm); 523 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 524 avl_insert(&sdp->sd_snaps, sep, where); 525 526 /* 527 * Change the current mountpoint info: 528 * - update the tail of the mntpoint path 529 * - update the tail of the resource path 530 */ 531 pathref = vfs_getmntpoint(vfsp); 532 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 533 VERIFY((tail = strrchr(newpath, '/')) != NULL); 534 *(tail+1) = '\0'; 535 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 536 (void) strcat(newpath, nm); 537 refstr_rele(pathref); 538 vfs_setmntpoint(vfsp, newpath); 539 540 pathref = vfs_getresource(vfsp); 541 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 542 VERIFY((tail = strrchr(newpath, '@')) != NULL); 543 *(tail+1) = '\0'; 544 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 545 (void) strcat(newpath, nm); 546 refstr_rele(pathref); 547 vfs_setresource(vfsp, newpath); 548 549 vfs_unlock(vfsp); 550} 551#endif 552 553#if 0 554static int 555zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 556 cred_t *cr) 557{ 558 zfsctl_snapdir_t *sdp = sdvp->v_data; 559 zfs_snapentry_t search, *sep; 560 avl_index_t where; 561 char from[MAXNAMELEN], to[MAXNAMELEN]; 562 int err; 563 564 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 565 if (err) 566 return (err); 567 err = zfs_secpolicy_write(from, cr); 568 if (err) 569 return (err); 570 571 /* 572 * Cannot move snapshots out of the snapdir. 573 */ 574 if (sdvp != tdvp) 575 return (EINVAL); 576 577 if (strcmp(snm, tnm) == 0) 578 return (0); 579 580 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 581 if (err) 582 return (err); 583 584 mutex_enter(&sdp->sd_lock); 585 586 search.se_name = (char *)snm; 587 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 588 mutex_exit(&sdp->sd_lock); 589 return (ENOENT); 590 } 591 592 err = dmu_objset_rename(from, to, B_FALSE); 593 if (err == 0) 594 zfsctl_rename_snap(sdp, sep, tnm); 595 596 mutex_exit(&sdp->sd_lock); 597 598 return (err); 599} 600#endif 601 602#if 0 603/* ARGSUSED */ 604static int 605zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 606{ 607 zfsctl_snapdir_t *sdp = dvp->v_data; 608 char snapname[MAXNAMELEN]; 609 int err; 610 611 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 612 if (err) 613 return (err); 614 err = zfs_secpolicy_write(snapname, cr); 615 if (err) 616 return (err); 617 618 mutex_enter(&sdp->sd_lock); 619 620 err = zfsctl_unmount_snap(dvp, name, 0, cr); 621 if (err) { 622 mutex_exit(&sdp->sd_lock); 623 return (err); 624 } 625 626 err = dmu_objset_destroy(snapname); 627 628 mutex_exit(&sdp->sd_lock); 629 630 return (err); 631} 632#endif 633 634/* 635 * Lookup entry point for the 'snapshot' directory. Try to open the 636 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 637 * Perform a mount of the associated dataset on top of the vnode. 638 */ 639/* ARGSUSED */ 640int 641zfsctl_snapdir_lookup(ap) 642 struct vop_lookup_args /* { 643 struct vnode *a_dvp; 644 struct vnode **a_vpp; 645 struct componentname *a_cnp; 646 } */ *ap; 647{ 648 vnode_t *dvp = ap->a_dvp; 649 vnode_t **vpp = ap->a_vpp; 650 char nm[NAME_MAX + 1]; 651 zfsctl_snapdir_t *sdp = dvp->v_data; 652 objset_t *snap; 653 char snapname[MAXNAMELEN]; 654 char *mountpoint; 655 zfs_snapentry_t *sep, search; 656 size_t mountpoint_len; 657 avl_index_t where; 658 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 659 int err; 660 661 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 662 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 663 664 ASSERT(dvp->v_type == VDIR); 665 666 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) 667 return (0); 668 669 *vpp = NULL; 670 671 /* 672 * If we get a recursive call, that means we got called 673 * from the domount() code while it was trying to look up the 674 * spec (which looks like a local path for zfs). We need to 675 * add some flag to domount() to tell it not to do this lookup. 676 */ 677 if (MUTEX_HELD(&sdp->sd_lock)) 678 return (ENOENT); 679 680 ZFS_ENTER(zfsvfs); 681 682 mutex_enter(&sdp->sd_lock); 683 search.se_name = (char *)nm; 684 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 685 *vpp = sep->se_root; 686 VN_HOLD(*vpp); 687 if ((*vpp)->v_mountedhere == NULL) { 688 /* 689 * The snapshot was unmounted behind our backs, 690 * try to remount it. 691 */ 692 goto domount; 693 } 694 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 695 mutex_exit(&sdp->sd_lock); 696 ZFS_EXIT(zfsvfs); 697 return (0); 698 } 699 700 /* 701 * The requested snapshot is not currently mounted, look it up. 702 */ 703 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 704 if (err) { 705 mutex_exit(&sdp->sd_lock); 706 ZFS_EXIT(zfsvfs); 707 return (err); 708 } 709 if (dmu_objset_open(snapname, DMU_OST_ZFS, 710 DS_MODE_STANDARD | DS_MODE_READONLY, &snap) != 0) { 711 mutex_exit(&sdp->sd_lock); 712 ZFS_EXIT(zfsvfs); 713 return (ENOENT); 714 } 715 716 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 717 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 718 (void) strcpy(sep->se_name, nm); 719 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 720 VN_HOLD(*vpp); 721 avl_insert(&sdp->sd_snaps, sep, where); 722 723 dmu_objset_close(snap); 724domount: 725 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 726 strlen("/.zfs/snapshot/") + strlen(nm) + 1; 727 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 728 (void) snprintf(mountpoint, mountpoint_len, "%s/.zfs/snapshot/%s", 729 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 730 err = domount(curthread, *vpp, "zfs", mountpoint, snapname, 0); 731 kmem_free(mountpoint, mountpoint_len); 732 /* FreeBSD: This line was moved from below to avoid a lock recursion. */ 733 if (err == 0) 734 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 735 mutex_exit(&sdp->sd_lock); 736 737 /* 738 * If we had an error, drop our hold on the vnode and 739 * zfsctl_snapshot_inactive() will clean up. 740 */ 741 if (err) { 742 VN_RELE(*vpp); 743 *vpp = NULL; 744 } 745 return (err); 746} 747 748/* ARGSUSED */ 749static int 750zfsctl_snapdir_readdir_cb(vnode_t *vp, struct dirent64 *dp, int *eofp, 751 offset_t *offp, offset_t *nextp, void *data) 752{ 753 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 754 char snapname[MAXNAMELEN]; 755 uint64_t id, cookie; 756 757 ZFS_ENTER(zfsvfs); 758 759 cookie = *offp; 760 if (dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 761 &cookie) == ENOENT) { 762 *eofp = 1; 763 ZFS_EXIT(zfsvfs); 764 return (0); 765 } 766 767 (void) strcpy(dp->d_name, snapname); 768 dp->d_ino = ZFSCTL_INO_SNAP(id); 769 *nextp = cookie; 770 771 ZFS_EXIT(zfsvfs); 772 773 return (0); 774} 775 776vnode_t * 777zfsctl_mknode_snapdir(vnode_t *pvp) 778{ 779 vnode_t *vp; 780 zfsctl_snapdir_t *sdp; 781 782 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 783 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 784 zfsctl_snapdir_readdir_cb, NULL); 785 sdp = vp->v_data; 786 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 787 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 788 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 789 avl_create(&sdp->sd_snaps, snapentry_compare, 790 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 791 return (vp); 792} 793 794/* ARGSUSED */ 795static int 796zfsctl_snapdir_getattr(ap) 797 struct vop_getattr_args /* { 798 struct vnode *a_vp; 799 struct vattr *a_vap; 800 struct ucred *a_cred; 801 } */ *ap; 802{ 803 struct vnode *vp = ap->a_vp; 804 struct vattr *vap = ap->a_vap; 805 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 806 zfsctl_snapdir_t *sdp = vp->v_data; 807 808 ZFS_ENTER(zfsvfs); 809 zfsctl_common_getattr(vp, vap); 810 vap->va_nodeid = gfs_file_inode(vp); 811 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 812 ZFS_EXIT(zfsvfs); 813 814 return (0); 815} 816 817/* ARGSUSED */ 818static int 819zfsctl_snapdir_inactive(ap) 820 struct vop_inactive_args /* { 821 struct vnode *a_vp; 822 struct thread *a_td; 823 } */ *ap; 824{ 825 vnode_t *vp = ap->a_vp; 826 zfsctl_snapdir_t *sdp = vp->v_data; 827 void *private; 828 829 private = gfs_dir_inactive(vp); 830 if (private != NULL) { 831 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 832 mutex_destroy(&sdp->sd_lock); 833 avl_destroy(&sdp->sd_snaps); 834 kmem_free(private, sizeof (zfsctl_snapdir_t)); 835 } 836 return (0); 837} 838 839static struct vop_vector zfsctl_ops_snapdir = { 840 .vop_default = &default_vnodeops, 841 .vop_open = zfsctl_common_open, 842 .vop_close = zfsctl_common_close, 843 .vop_ioctl = VOP_EINVAL, 844 .vop_getattr = zfsctl_snapdir_getattr, 845 .vop_access = zfsctl_common_access, 846 .vop_readdir = gfs_vop_readdir, 847 .vop_lookup = zfsctl_snapdir_lookup, 848 .vop_inactive = zfsctl_snapdir_inactive, 849 .vop_reclaim = zfsctl_common_reclaim, 850 .vop_fid = zfsctl_common_fid, 851}; 852 853static vnode_t * 854zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 855{ 856 vnode_t *vp; 857 zfsctl_node_t *zcp; 858 859 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 860 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 861 zcp = vp->v_data; 862 zcp->zc_id = objset; 863 864 return (vp); 865} 866 867static int 868zfsctl_snapshot_inactive(ap) 869 struct vop_inactive_args /* { 870 struct vnode *a_vp; 871 struct thread *a_td; 872 } */ *ap; 873{ 874 vnode_t *vp = ap->a_vp; 875 struct vop_inactive_args iap; 876 zfsctl_snapdir_t *sdp; 877 zfs_snapentry_t *sep, *next; 878 int locked; 879 vnode_t *dvp; 880 881 VERIFY(gfs_dir_lookup(vp, "..", &dvp) == 0); 882 sdp = dvp->v_data; 883 VOP_UNLOCK(dvp, 0); 884 885 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 886 mutex_enter(&sdp->sd_lock); 887 888 if (vp->v_count > 1) { 889 if (!locked) 890 mutex_exit(&sdp->sd_lock); 891 return (0); 892 } 893 ASSERT(!vn_ismntpt(vp)); 894 895 sep = avl_first(&sdp->sd_snaps); 896 while (sep != NULL) { 897 next = AVL_NEXT(&sdp->sd_snaps, sep); 898 899 if (sep->se_root == vp) { 900 avl_remove(&sdp->sd_snaps, sep); 901 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 902 kmem_free(sep, sizeof (zfs_snapentry_t)); 903 break; 904 } 905 sep = next; 906 } 907 ASSERT(sep != NULL); 908 909 if (!locked) 910 mutex_exit(&sdp->sd_lock); 911 VN_RELE(dvp); 912 913 /* 914 * Dispose of the vnode for the snapshot mount point. 915 * This is safe to do because once this entry has been removed 916 * from the AVL tree, it can't be found again, so cannot become 917 * "active". If we lookup the same name again we will end up 918 * creating a new vnode. 919 */ 920 iap.a_vp = vp; 921 return (gfs_vop_inactive(&iap)); 922} 923 924static int 925zfsctl_traverse_begin(vnode_t **vpp, int lktype) 926{ 927 928 VN_HOLD(*vpp); 929 930 /* Snapshot should be already mounted, but just in case. */ 931 if (vn_mountedvfs(*vpp) == NULL) 932 return (ENOENT); 933 return (traverse(vpp, lktype)); 934} 935 936static void 937zfsctl_traverse_end(vnode_t *vp, int err) 938{ 939 940 if (err == 0) 941 vput(vp); 942 else 943 VN_RELE(vp); 944} 945 946static int 947zfsctl_snapshot_getattr(ap) 948 struct vop_getattr_args /* { 949 struct vnode *a_vp; 950 struct vattr *a_vap; 951 struct ucred *a_cred; 952 } */ *ap; 953{ 954 vnode_t *vp = ap->a_vp; 955 int err; 956 957 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 958 if (err == 0) 959 err = VOP_GETATTR(vp, ap->a_vap, ap->a_cred); 960 zfsctl_traverse_end(vp, err); 961 return (err); 962} 963 964static int 965zfsctl_snapshot_fid(ap) 966 struct vop_fid_args /* { 967 struct vnode *a_vp; 968 struct fid *a_fid; 969 } */ *ap; 970{ 971 vnode_t *vp = ap->a_vp; 972 int err; 973 974 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 975 if (err == 0) 976 err = VOP_VPTOFH(vp, (void *)ap->a_fid); 977 zfsctl_traverse_end(vp, err); 978 return (err); 979} 980 981/* 982 * These VP's should never see the light of day. They should always 983 * be covered. 984 */ 985static struct vop_vector zfsctl_ops_snapshot = { 986 .vop_default = &default_vnodeops, 987 .vop_inactive = zfsctl_snapshot_inactive, 988 .vop_reclaim = zfsctl_common_reclaim, 989 .vop_getattr = zfsctl_snapshot_getattr, 990 .vop_fid = zfsctl_snapshot_fid, 991}; 992 993int 994zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 995{ 996 zfsvfs_t *zfsvfs = vfsp->vfs_data; 997 vnode_t *dvp, *vp; 998 zfsctl_snapdir_t *sdp; 999 zfsctl_node_t *zcp; 1000 zfs_snapentry_t *sep; 1001 int error; 1002 1003 ASSERT(zfsvfs->z_ctldir != NULL); 1004 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1005 NULL, 0, NULL, kcred); 1006 if (error != 0) 1007 return (error); 1008 sdp = dvp->v_data; 1009 1010 mutex_enter(&sdp->sd_lock); 1011 sep = avl_first(&sdp->sd_snaps); 1012 while (sep != NULL) { 1013 vp = sep->se_root; 1014 zcp = vp->v_data; 1015 if (zcp->zc_id == objsetid) 1016 break; 1017 1018 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1019 } 1020 1021 if (sep != NULL) { 1022 VN_HOLD(vp); 1023 error = traverse(&vp, LK_SHARED | LK_RETRY); 1024 if (error == 0) { 1025 if (vp == sep->se_root) 1026 error = EINVAL; 1027 else 1028 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1029 } 1030 mutex_exit(&sdp->sd_lock); 1031 if (error == 0) 1032 VN_URELE(vp); 1033 else 1034 VN_RELE(vp); 1035 } else { 1036 error = EINVAL; 1037 mutex_exit(&sdp->sd_lock); 1038 } 1039 1040 VN_RELE(dvp); 1041 1042 return (error); 1043} 1044 1045/* 1046 * Unmount any snapshots for the given filesystem. This is called from 1047 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1048 * snapshots. 1049 */ 1050int 1051zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1052{ 1053 struct vop_inactive_args ap; 1054 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1055 vnode_t *dvp, *svp; 1056 zfsctl_snapdir_t *sdp; 1057 zfs_snapentry_t *sep, *next; 1058 int error; 1059 1060 ASSERT(zfsvfs->z_ctldir != NULL); 1061 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1062 NULL, 0, NULL, cr); 1063 if (error != 0) 1064 return (error); 1065 sdp = dvp->v_data; 1066 1067 mutex_enter(&sdp->sd_lock); 1068 1069 sep = avl_first(&sdp->sd_snaps); 1070 while (sep != NULL) { 1071 svp = sep->se_root; 1072 next = AVL_NEXT(&sdp->sd_snaps, sep); 1073 1074 /* 1075 * If this snapshot is not mounted, then it must 1076 * have just been unmounted by somebody else, and 1077 * will be cleaned up by zfsctl_snapdir_inactive(). 1078 */ 1079 if (vn_ismntpt(svp)) { 1080 if ((error = vn_vfswlock(svp)) != 0) 1081 goto out; 1082 1083 /* 1084 * Increase usecount, so dounmount() won't vrele() it 1085 * to 0 and call zfsctl_snapdir_inactive(). 1086 */ 1087 VN_HOLD(svp); 1088 vfsp = vn_mountedvfs(svp); 1089 mtx_lock(&Giant); 1090 error = dounmount(vfsp, fflags, curthread); 1091 mtx_unlock(&Giant); 1092 if (error != 0) { 1093 VN_RELE(svp); 1094 goto out; 1095 } 1096 1097 avl_remove(&sdp->sd_snaps, sep); 1098 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1099 kmem_free(sep, sizeof (zfs_snapentry_t)); 1100 1101 /* 1102 * We can't use VN_RELE(), as that will try to 1103 * invoke zfsctl_snapdir_inactive(), and that 1104 * would lead to an attempt to re-grab the sd_lock. 1105 */ 1106 ASSERT3U(svp->v_count, ==, 1); 1107 ap.a_vp = svp; 1108 gfs_vop_inactive(&ap); 1109 } 1110 sep = next; 1111 } 1112out: 1113 mutex_exit(&sdp->sd_lock); 1114 VN_RELE(dvp); 1115 1116 return (error); 1117} 1118