zfs_ctldir.c revision 323762
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved. 24 * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved. 25 */ 26 27/* 28 * ZFS control directory (a.k.a. ".zfs") 29 * 30 * This directory provides a common location for all ZFS meta-objects. 31 * Currently, this is only the 'snapshot' directory, but this may expand in the 32 * future. The elements are built using the GFS primitives, as the hierarchy 33 * does not actually exist on disk. 34 * 35 * For 'snapshot', we don't want to have all snapshots always mounted, because 36 * this would take up a huge amount of space in /etc/mnttab. We have three 37 * types of objects: 38 * 39 * ctldir ------> snapshotdir -------> snapshot 40 * | 41 * | 42 * V 43 * mounted fs 44 * 45 * The 'snapshot' node contains just enough information to lookup '..' and act 46 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we 47 * perform an automount of the underlying filesystem and return the 48 * corresponding vnode. 49 * 50 * All mounts are handled automatically by the kernel, but unmounts are 51 * (currently) handled from user land. The main reason is that there is no 52 * reliable way to auto-unmount the filesystem when it's "no longer in use". 53 * When the user unmounts a filesystem, we call zfsctl_unmount(), which 54 * unmounts any snapshots within the snapshot directory. 55 * 56 * The '.zfs', '.zfs/snapshot', and all directories created under 57 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and 58 * share the same vfs_t as the head filesystem (what '.zfs' lives under). 59 * 60 * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>' 61 * (ie: snapshots) are ZFS nodes and have their own unique vfs_t. 62 * However, vnodes within these mounted on file systems have their v_vfsp 63 * fields set to the head filesystem to make NFS happy (see 64 * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t 65 * so that it cannot be freed until all snapshots have been unmounted. 66 */ 67 68#include <sys/zfs_context.h> 69#include <sys/zfs_ctldir.h> 70#include <sys/zfs_ioctl.h> 71#include <sys/zfs_vfsops.h> 72#include <sys/namei.h> 73#include <sys/stat.h> 74#include <sys/dmu.h> 75#include <sys/dsl_dataset.h> 76#include <sys/dsl_destroy.h> 77#include <sys/dsl_deleg.h> 78#include <sys/mount.h> 79#include <sys/zap.h> 80 81#include "zfs_namecheck.h" 82 83/* 84 * "Synthetic" filesystem implementation. 85 */ 86 87/* 88 * Assert that A implies B. 89 */ 90#define KASSERT_IMPLY(A, B, msg) KASSERT(!(A) || (B), (msg)); 91 92static MALLOC_DEFINE(M_SFSNODES, "sfs_nodes", "synthetic-fs nodes"); 93 94typedef struct sfs_node { 95 char sn_name[ZFS_MAX_DATASET_NAME_LEN]; 96 uint64_t sn_parent_id; 97 uint64_t sn_id; 98} sfs_node_t; 99 100/* 101 * Check the parent's ID as well as the node's to account for a chance 102 * that IDs originating from different domains (snapshot IDs, artifical 103 * IDs, znode IDs) may clash. 104 */ 105static int 106sfs_compare_ids(struct vnode *vp, void *arg) 107{ 108 sfs_node_t *n1 = vp->v_data; 109 sfs_node_t *n2 = arg; 110 bool equal; 111 112 equal = n1->sn_id == n2->sn_id && 113 n1->sn_parent_id == n2->sn_parent_id; 114 115 /* Zero means equality. */ 116 return (!equal); 117} 118 119static int 120sfs_vnode_get(const struct mount *mp, int flags, uint64_t parent_id, 121 uint64_t id, struct vnode **vpp) 122{ 123 sfs_node_t search; 124 int err; 125 126 search.sn_id = id; 127 search.sn_parent_id = parent_id; 128 err = vfs_hash_get(mp, (u_int)id, flags, curthread, vpp, 129 sfs_compare_ids, &search); 130 return (err); 131} 132 133static int 134sfs_vnode_insert(struct vnode *vp, int flags, uint64_t parent_id, 135 uint64_t id, struct vnode **vpp) 136{ 137 int err; 138 139 KASSERT(vp->v_data != NULL, ("sfs_vnode_insert with NULL v_data")); 140 err = vfs_hash_insert(vp, (u_int)id, flags, curthread, vpp, 141 sfs_compare_ids, vp->v_data); 142 return (err); 143} 144 145static void 146sfs_vnode_remove(struct vnode *vp) 147{ 148 vfs_hash_remove(vp); 149} 150 151typedef void sfs_vnode_setup_fn(vnode_t *vp, void *arg); 152 153static int 154sfs_vgetx(struct mount *mp, int flags, uint64_t parent_id, uint64_t id, 155 const char *tag, struct vop_vector *vops, 156 sfs_vnode_setup_fn setup, void *arg, 157 struct vnode **vpp) 158{ 159 struct vnode *vp; 160 int error; 161 162 error = sfs_vnode_get(mp, flags, parent_id, id, vpp); 163 if (error != 0 || *vpp != NULL) { 164 KASSERT_IMPLY(error == 0, (*vpp)->v_data != NULL, 165 "sfs vnode with no data"); 166 return (error); 167 } 168 169 /* Allocate a new vnode/inode. */ 170 error = getnewvnode(tag, mp, vops, &vp); 171 if (error != 0) { 172 *vpp = NULL; 173 return (error); 174 } 175 176 /* 177 * Exclusively lock the vnode vnode while it's being constructed. 178 */ 179 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); 180 error = insmntque(vp, mp); 181 if (error != 0) { 182 *vpp = NULL; 183 return (error); 184 } 185 186 setup(vp, arg); 187 188 error = sfs_vnode_insert(vp, flags, parent_id, id, vpp); 189 if (error != 0 || *vpp != NULL) { 190 KASSERT_IMPLY(error == 0, (*vpp)->v_data != NULL, 191 "sfs vnode with no data"); 192 return (error); 193 } 194 195 *vpp = vp; 196 return (0); 197} 198 199static void 200sfs_print_node(sfs_node_t *node) 201{ 202 printf("\tname = %s\n", node->sn_name); 203 printf("\tparent_id = %ju\n", (uintmax_t)node->sn_parent_id); 204 printf("\tid = %ju\n", (uintmax_t)node->sn_id); 205} 206 207static sfs_node_t * 208sfs_alloc_node(size_t size, const char *name, uint64_t parent_id, uint64_t id) 209{ 210 struct sfs_node *node; 211 212 KASSERT(strlen(name) < sizeof(node->sn_name), 213 ("sfs node name is too long")); 214 KASSERT(size >= sizeof(*node), ("sfs node size is too small")); 215 node = malloc(size, M_SFSNODES, M_WAITOK | M_ZERO); 216 strlcpy(node->sn_name, name, sizeof(node->sn_name)); 217 node->sn_parent_id = parent_id; 218 node->sn_id = id; 219 220 return (node); 221} 222 223static void 224sfs_destroy_node(sfs_node_t *node) 225{ 226 free(node, M_SFSNODES); 227} 228 229static void * 230sfs_reclaim_vnode(vnode_t *vp) 231{ 232 sfs_node_t *node; 233 void *data; 234 235 sfs_vnode_remove(vp); 236 data = vp->v_data; 237 vp->v_data = NULL; 238 return (data); 239} 240 241static int 242sfs_readdir_common(uint64_t parent_id, uint64_t id, struct vop_readdir_args *ap, 243 uio_t *uio, off_t *offp) 244{ 245 struct dirent entry; 246 int error; 247 248 /* Reset ncookies for subsequent use of vfs_read_dirent. */ 249 if (ap->a_ncookies != NULL) 250 *ap->a_ncookies = 0; 251 252 if (uio->uio_resid < sizeof(entry)) 253 return (SET_ERROR(EINVAL)); 254 255 if (uio->uio_offset < 0) 256 return (SET_ERROR(EINVAL)); 257 if (uio->uio_offset == 0) { 258 entry.d_fileno = id; 259 entry.d_type = DT_DIR; 260 entry.d_name[0] = '.'; 261 entry.d_name[1] = '\0'; 262 entry.d_namlen = 1; 263 entry.d_reclen = sizeof(entry); 264 error = vfs_read_dirent(ap, &entry, uio->uio_offset); 265 if (error != 0) 266 return (SET_ERROR(error)); 267 } 268 269 if (uio->uio_offset < sizeof(entry)) 270 return (SET_ERROR(EINVAL)); 271 if (uio->uio_offset == sizeof(entry)) { 272 entry.d_fileno = parent_id; 273 entry.d_type = DT_DIR; 274 entry.d_name[0] = '.'; 275 entry.d_name[1] = '.'; 276 entry.d_name[2] = '\0'; 277 entry.d_namlen = 2; 278 entry.d_reclen = sizeof(entry); 279 error = vfs_read_dirent(ap, &entry, uio->uio_offset); 280 if (error != 0) 281 return (SET_ERROR(error)); 282 } 283 284 if (offp != NULL) 285 *offp = 2 * sizeof(entry); 286 return (0); 287} 288 289 290/* 291 * .zfs inode namespace 292 * 293 * We need to generate unique inode numbers for all files and directories 294 * within the .zfs pseudo-filesystem. We use the following scheme: 295 * 296 * ENTRY ZFSCTL_INODE 297 * .zfs 1 298 * .zfs/snapshot 2 299 * .zfs/snapshot/<snap> objectid(snap) 300 */ 301#define ZFSCTL_INO_SNAP(id) (id) 302 303static struct vop_vector zfsctl_ops_root; 304static struct vop_vector zfsctl_ops_snapdir; 305static struct vop_vector zfsctl_ops_snapshot; 306static struct vop_vector zfsctl_ops_shares_dir; 307 308void 309zfsctl_init(void) 310{ 311} 312 313void 314zfsctl_fini(void) 315{ 316} 317 318boolean_t 319zfsctl_is_node(vnode_t *vp) 320{ 321 return (vn_matchops(vp, zfsctl_ops_root) || 322 vn_matchops(vp, zfsctl_ops_snapdir) || 323 vn_matchops(vp, zfsctl_ops_snapshot) || 324 vn_matchops(vp, zfsctl_ops_shares_dir)); 325 326} 327 328typedef struct zfsctl_root { 329 sfs_node_t node; 330 sfs_node_t *snapdir; 331 timestruc_t cmtime; 332} zfsctl_root_t; 333 334 335/* 336 * Create the '.zfs' directory. 337 */ 338void 339zfsctl_create(zfsvfs_t *zfsvfs) 340{ 341 zfsctl_root_t *dot_zfs; 342 sfs_node_t *snapdir; 343 vnode_t *rvp; 344 uint64_t crtime[2]; 345 346 ASSERT(zfsvfs->z_ctldir == NULL); 347 348 snapdir = sfs_alloc_node(sizeof(*snapdir), "snapshot", ZFSCTL_INO_ROOT, 349 ZFSCTL_INO_SNAPDIR); 350 dot_zfs = (zfsctl_root_t *)sfs_alloc_node(sizeof(*dot_zfs), ".zfs", 0, 351 ZFSCTL_INO_ROOT); 352 dot_zfs->snapdir = snapdir; 353 354 VERIFY(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp) == 0); 355 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 356 &crtime, sizeof(crtime))); 357 ZFS_TIME_DECODE(&dot_zfs->cmtime, crtime); 358 vput(rvp); 359 360 zfsvfs->z_ctldir = dot_zfs; 361} 362 363/* 364 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted. 365 * The nodes must not have any associated vnodes by now as they should be 366 * vflush-ed. 367 */ 368void 369zfsctl_destroy(zfsvfs_t *zfsvfs) 370{ 371 sfs_destroy_node(zfsvfs->z_ctldir->snapdir); 372 sfs_destroy_node((sfs_node_t *)zfsvfs->z_ctldir); 373 zfsvfs->z_ctldir = NULL; 374} 375 376static int 377zfsctl_fs_root_vnode(struct mount *mp, void *arg __unused, int flags, 378 struct vnode **vpp) 379{ 380 return (VFS_ROOT(mp, flags, vpp)); 381} 382 383static void 384zfsctl_common_vnode_setup(vnode_t *vp, void *arg) 385{ 386 ASSERT_VOP_ELOCKED(vp, __func__); 387 388 /* We support shared locking. */ 389 VN_LOCK_ASHARE(vp); 390 vp->v_type = VDIR; 391 vp->v_data = arg; 392} 393 394static int 395zfsctl_root_vnode(struct mount *mp, void *arg __unused, int flags, 396 struct vnode **vpp) 397{ 398 void *node; 399 int err; 400 401 node = ((zfsvfs_t*)mp->mnt_data)->z_ctldir; 402 err = sfs_vgetx(mp, flags, 0, ZFSCTL_INO_ROOT, "zfs", &zfsctl_ops_root, 403 zfsctl_common_vnode_setup, node, vpp); 404 return (err); 405} 406 407static int 408zfsctl_snapdir_vnode(struct mount *mp, void *arg __unused, int flags, 409 struct vnode **vpp) 410{ 411 void *node; 412 int err; 413 414 node = ((zfsvfs_t*)mp->mnt_data)->z_ctldir->snapdir; 415 err = sfs_vgetx(mp, flags, ZFSCTL_INO_ROOT, ZFSCTL_INO_SNAPDIR, "zfs", 416 &zfsctl_ops_snapdir, zfsctl_common_vnode_setup, node, vpp); 417 return (err); 418} 419 420/* 421 * Given a root znode, retrieve the associated .zfs directory. 422 * Add a hold to the vnode and return it. 423 */ 424int 425zfsctl_root(zfsvfs_t *zfsvfs, int flags, vnode_t **vpp) 426{ 427 vnode_t *vp; 428 int error; 429 430 error = zfsctl_root_vnode(zfsvfs->z_vfs, NULL, flags, vpp); 431 return (error); 432} 433 434/* 435 * Common open routine. Disallow any write access. 436 */ 437static int 438zfsctl_common_open(struct vop_open_args *ap) 439{ 440 int flags = ap->a_mode; 441 442 if (flags & FWRITE) 443 return (SET_ERROR(EACCES)); 444 445 return (0); 446} 447 448/* 449 * Common close routine. Nothing to do here. 450 */ 451/* ARGSUSED */ 452static int 453zfsctl_common_close(struct vop_close_args *ap) 454{ 455 return (0); 456} 457 458/* 459 * Common access routine. Disallow writes. 460 */ 461static int 462zfsctl_common_access(ap) 463 struct vop_access_args /* { 464 struct vnode *a_vp; 465 accmode_t a_accmode; 466 struct ucred *a_cred; 467 struct thread *a_td; 468 } */ *ap; 469{ 470 accmode_t accmode = ap->a_accmode; 471 472 if (accmode & VWRITE) 473 return (SET_ERROR(EACCES)); 474 return (0); 475} 476 477/* 478 * Common getattr function. Fill in basic information. 479 */ 480static void 481zfsctl_common_getattr(vnode_t *vp, vattr_t *vap) 482{ 483 timestruc_t now; 484 sfs_node_t *node; 485 486 node = vp->v_data; 487 488 vap->va_uid = 0; 489 vap->va_gid = 0; 490 vap->va_rdev = 0; 491 /* 492 * We are a purely virtual object, so we have no 493 * blocksize or allocated blocks. 494 */ 495 vap->va_blksize = 0; 496 vap->va_nblocks = 0; 497 vap->va_seq = 0; 498 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 499 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP | 500 S_IROTH | S_IXOTH; 501 vap->va_type = VDIR; 502 /* 503 * We live in the now (for atime). 504 */ 505 gethrestime(&now); 506 vap->va_atime = now; 507 /* FreeBSD: Reset chflags(2) flags. */ 508 vap->va_flags = 0; 509 510 vap->va_nodeid = node->sn_id; 511 512 /* At least '.' and '..'. */ 513 vap->va_nlink = 2; 514} 515 516static int 517zfsctl_common_fid(ap) 518 struct vop_fid_args /* { 519 struct vnode *a_vp; 520 struct fid *a_fid; 521 } */ *ap; 522{ 523 vnode_t *vp = ap->a_vp; 524 fid_t *fidp = (void *)ap->a_fid; 525 sfs_node_t *node = vp->v_data; 526 uint64_t object = node->sn_id; 527 zfid_short_t *zfid; 528 int i; 529 530 zfid = (zfid_short_t *)fidp; 531 zfid->zf_len = SHORT_FID_LEN; 532 533 for (i = 0; i < sizeof(zfid->zf_object); i++) 534 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 535 536 /* .zfs nodes always have a generation number of 0 */ 537 for (i = 0; i < sizeof(zfid->zf_gen); i++) 538 zfid->zf_gen[i] = 0; 539 540 return (0); 541} 542 543static int 544zfsctl_common_reclaim(ap) 545 struct vop_reclaim_args /* { 546 struct vnode *a_vp; 547 struct thread *a_td; 548 } */ *ap; 549{ 550 vnode_t *vp = ap->a_vp; 551 552 (void) sfs_reclaim_vnode(vp); 553 return (0); 554} 555 556static int 557zfsctl_common_print(ap) 558 struct vop_print_args /* { 559 struct vnode *a_vp; 560 } */ *ap; 561{ 562 sfs_print_node(ap->a_vp->v_data); 563 return (0); 564} 565 566/* 567 * Get root directory attributes. 568 */ 569static int 570zfsctl_root_getattr(ap) 571 struct vop_getattr_args /* { 572 struct vnode *a_vp; 573 struct vattr *a_vap; 574 struct ucred *a_cred; 575 } */ *ap; 576{ 577 struct vnode *vp = ap->a_vp; 578 struct vattr *vap = ap->a_vap; 579 zfsctl_root_t *node = vp->v_data; 580 581 zfsctl_common_getattr(vp, vap); 582 vap->va_ctime = node->cmtime; 583 vap->va_mtime = vap->va_ctime; 584 vap->va_birthtime = vap->va_ctime; 585 vap->va_nlink += 1; /* snapdir */ 586 vap->va_size = vap->va_nlink; 587 return (0); 588} 589 590/* 591 * When we lookup "." we still can be asked to lock it 592 * differently, can't we? 593 */ 594int 595zfsctl_relock_dot(vnode_t *dvp, int ltype) 596{ 597 vref(dvp); 598 if (ltype != VOP_ISLOCKED(dvp)) { 599 if (ltype == LK_EXCLUSIVE) 600 vn_lock(dvp, LK_UPGRADE | LK_RETRY); 601 else /* if (ltype == LK_SHARED) */ 602 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY); 603 604 /* Relock for the "." case may left us with reclaimed vnode. */ 605 if ((dvp->v_iflag & VI_DOOMED) != 0) { 606 vrele(dvp); 607 return (SET_ERROR(ENOENT)); 608 } 609 } 610 return (0); 611} 612 613/* 614 * Special case the handling of "..". 615 */ 616int 617zfsctl_root_lookup(ap) 618 struct vop_lookup_args /* { 619 struct vnode *a_dvp; 620 struct vnode **a_vpp; 621 struct componentname *a_cnp; 622 } */ *ap; 623{ 624 struct componentname *cnp = ap->a_cnp; 625 vnode_t *dvp = ap->a_dvp; 626 vnode_t **vpp = ap->a_vpp; 627 cred_t *cr = ap->a_cnp->cn_cred; 628 int flags = ap->a_cnp->cn_flags; 629 int lkflags = ap->a_cnp->cn_lkflags; 630 int nameiop = ap->a_cnp->cn_nameiop; 631 int err; 632 int ltype; 633 634 ASSERT(dvp->v_type == VDIR); 635 636 if ((flags & ISLASTCN) != 0 && nameiop != LOOKUP) 637 return (SET_ERROR(ENOTSUP)); 638 639 if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') { 640 err = zfsctl_relock_dot(dvp, lkflags & LK_TYPE_MASK); 641 if (err == 0) 642 *vpp = dvp; 643 } else if ((flags & ISDOTDOT) != 0) { 644 err = vn_vget_ino_gen(dvp, zfsctl_fs_root_vnode, NULL, 645 lkflags, vpp); 646 } else if (strncmp(cnp->cn_nameptr, "snapshot", cnp->cn_namelen) == 0) { 647 err = zfsctl_snapdir_vnode(dvp->v_mount, NULL, lkflags, vpp); 648 } else { 649 err = SET_ERROR(ENOENT); 650 } 651 if (err != 0) 652 *vpp = NULL; 653 return (err); 654} 655 656static int 657zfsctl_root_readdir(ap) 658 struct vop_readdir_args /* { 659 struct vnode *a_vp; 660 struct uio *a_uio; 661 struct ucred *a_cred; 662 int *a_eofflag; 663 int *ncookies; 664 u_long **a_cookies; 665 } */ *ap; 666{ 667 struct dirent entry; 668 vnode_t *vp = ap->a_vp; 669 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 670 zfsctl_root_t *node = vp->v_data; 671 uio_t *uio = ap->a_uio; 672 int *eofp = ap->a_eofflag; 673 off_t dots_offset; 674 int error; 675 676 ASSERT(vp->v_type == VDIR); 677 678 error = sfs_readdir_common(zfsvfs->z_root, ZFSCTL_INO_ROOT, ap, uio, 679 &dots_offset); 680 if (error != 0) { 681 if (error == ENAMETOOLONG) /* ran out of destination space */ 682 error = 0; 683 return (error); 684 } 685 if (uio->uio_offset != dots_offset) 686 return (SET_ERROR(EINVAL)); 687 688 CTASSERT(sizeof(node->snapdir->sn_name) <= sizeof(entry.d_name)); 689 entry.d_fileno = node->snapdir->sn_id; 690 entry.d_type = DT_DIR; 691 strcpy(entry.d_name, node->snapdir->sn_name); 692 entry.d_namlen = strlen(entry.d_name); 693 entry.d_reclen = sizeof(entry); 694 error = vfs_read_dirent(ap, &entry, uio->uio_offset); 695 if (error != 0) { 696 if (error == ENAMETOOLONG) 697 error = 0; 698 return (SET_ERROR(error)); 699 } 700 if (eofp != NULL) 701 *eofp = 1; 702 return (0); 703} 704 705static int 706zfsctl_root_vptocnp(struct vop_vptocnp_args *ap) 707{ 708 static const char dotzfs_name[4] = ".zfs"; 709 vnode_t *dvp; 710 int error; 711 712 if (*ap->a_buflen < sizeof (dotzfs_name)) 713 return (SET_ERROR(ENOMEM)); 714 715 error = vn_vget_ino_gen(ap->a_vp, zfsctl_fs_root_vnode, NULL, 716 LK_SHARED, &dvp); 717 if (error != 0) 718 return (SET_ERROR(error)); 719 720 VOP_UNLOCK(dvp, 0); 721 *ap->a_vpp = dvp; 722 *ap->a_buflen -= sizeof (dotzfs_name); 723 bcopy(dotzfs_name, ap->a_buf + *ap->a_buflen, sizeof (dotzfs_name)); 724 return (0); 725} 726 727static struct vop_vector zfsctl_ops_root = { 728 .vop_default = &default_vnodeops, 729 .vop_open = zfsctl_common_open, 730 .vop_close = zfsctl_common_close, 731 .vop_ioctl = VOP_EINVAL, 732 .vop_getattr = zfsctl_root_getattr, 733 .vop_access = zfsctl_common_access, 734 .vop_readdir = zfsctl_root_readdir, 735 .vop_lookup = zfsctl_root_lookup, 736 .vop_inactive = VOP_NULL, 737 .vop_reclaim = zfsctl_common_reclaim, 738 .vop_fid = zfsctl_common_fid, 739 .vop_print = zfsctl_common_print, 740 .vop_vptocnp = zfsctl_root_vptocnp, 741}; 742 743static int 744zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 745{ 746 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 747 748 dmu_objset_name(os, zname); 749 if (strlen(zname) + 1 + strlen(name) >= len) 750 return (SET_ERROR(ENAMETOOLONG)); 751 (void) strcat(zname, "@"); 752 (void) strcat(zname, name); 753 return (0); 754} 755 756static int 757zfsctl_snapshot_lookup(vnode_t *vp, const char *name, uint64_t *id) 758{ 759 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 760 int err; 761 762 err = dsl_dataset_snap_lookup(dmu_objset_ds(os), name, id); 763 return (err); 764} 765 766/* 767 * Given a vnode get a root vnode of a filesystem mounted on top of 768 * the vnode, if any. The root vnode is referenced and locked. 769 * If no filesystem is mounted then the orinal vnode remains referenced 770 * and locked. If any error happens the orinal vnode is unlocked and 771 * released. 772 */ 773static int 774zfsctl_mounted_here(vnode_t **vpp, int flags) 775{ 776 struct mount *mp; 777 int err; 778 779 ASSERT_VOP_LOCKED(*vpp, __func__); 780 ASSERT3S((*vpp)->v_type, ==, VDIR); 781 782 if ((mp = (*vpp)->v_mountedhere) != NULL) { 783 err = vfs_busy(mp, 0); 784 KASSERT(err == 0, ("vfs_busy(mp, 0) failed with %d", err)); 785 KASSERT(vrefcnt(*vpp) > 1, ("unreferenced mountpoint")); 786 vput(*vpp); 787 err = VFS_ROOT(mp, flags, vpp); 788 vfs_unbusy(mp); 789 return (err); 790 } 791 return (EJUSTRETURN); 792} 793 794typedef struct { 795 const char *snap_name; 796 uint64_t snap_id; 797} snapshot_setup_arg_t; 798 799static void 800zfsctl_snapshot_vnode_setup(vnode_t *vp, void *arg) 801{ 802 snapshot_setup_arg_t *ssa = arg; 803 sfs_node_t *node; 804 805 ASSERT_VOP_ELOCKED(vp, __func__); 806 807 node = sfs_alloc_node(sizeof(sfs_node_t), 808 ssa->snap_name, ZFSCTL_INO_SNAPDIR, ssa->snap_id); 809 zfsctl_common_vnode_setup(vp, node); 810 811 /* We have to support recursive locking. */ 812 VN_LOCK_AREC(vp); 813} 814 815/* 816 * Lookup entry point for the 'snapshot' directory. Try to open the 817 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 818 * Perform a mount of the associated dataset on top of the vnode. 819 */ 820int 821zfsctl_snapdir_lookup(ap) 822 struct vop_lookup_args /* { 823 struct vnode *a_dvp; 824 struct vnode **a_vpp; 825 struct componentname *a_cnp; 826 } */ *ap; 827{ 828 vnode_t *dvp = ap->a_dvp; 829 vnode_t **vpp = ap->a_vpp; 830 struct componentname *cnp = ap->a_cnp; 831 char name[NAME_MAX + 1]; 832 char fullname[ZFS_MAX_DATASET_NAME_LEN]; 833 char *mountpoint; 834 size_t mountpoint_len; 835 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 836 uint64_t snap_id; 837 int nameiop = cnp->cn_nameiop; 838 int lkflags = cnp->cn_lkflags; 839 int flags = cnp->cn_flags; 840 int err; 841 842 ASSERT(dvp->v_type == VDIR); 843 844 if ((flags & ISLASTCN) != 0 && nameiop != LOOKUP) 845 return (SET_ERROR(ENOTSUP)); 846 847 if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') { 848 err = zfsctl_relock_dot(dvp, lkflags & LK_TYPE_MASK); 849 if (err == 0) 850 *vpp = dvp; 851 return (err); 852 } 853 if (flags & ISDOTDOT) { 854 err = vn_vget_ino_gen(dvp, zfsctl_root_vnode, NULL, lkflags, 855 vpp); 856 return (err); 857 } 858 859 if (cnp->cn_namelen >= sizeof(name)) 860 return (SET_ERROR(ENAMETOOLONG)); 861 862 strlcpy(name, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 863 err = zfsctl_snapshot_lookup(dvp, name, &snap_id); 864 if (err != 0) 865 return (SET_ERROR(ENOENT)); 866 867 for (;;) { 868 snapshot_setup_arg_t ssa; 869 870 ssa.snap_name = name; 871 ssa.snap_id = snap_id; 872 err = sfs_vgetx(dvp->v_mount, LK_SHARED, ZFSCTL_INO_SNAPDIR, 873 snap_id, "zfs", &zfsctl_ops_snapshot, 874 zfsctl_snapshot_vnode_setup, &ssa, vpp); 875 if (err != 0) 876 return (err); 877 878 /* Check if a new vnode has just been created. */ 879 if (VOP_ISLOCKED(*vpp) == LK_EXCLUSIVE) 880 break; 881 882 /* 883 * The vnode must be referenced at least by this thread and 884 * the mounted snapshot or the thread doing the mounting. 885 * There can be more references from concurrent lookups. 886 */ 887 KASSERT(vrefcnt(*vpp) > 1, ("found unreferenced mountpoint")); 888 889 /* 890 * Check if a snapshot is already mounted on top of the vnode. 891 */ 892 err = zfsctl_mounted_here(vpp, lkflags); 893 if (err != EJUSTRETURN) 894 return (err); 895 896#ifdef INVARIANTS 897 /* 898 * If the vnode not covered yet, then the mount operation 899 * must be in progress. 900 */ 901 VI_LOCK(*vpp); 902 KASSERT(((*vpp)->v_iflag & VI_MOUNT) != 0, 903 ("snapshot vnode not covered")); 904 VI_UNLOCK(*vpp); 905#endif 906 vput(*vpp); 907 908 /* 909 * In this situation we can loop on uncontested locks and starve 910 * the thread doing the lengthy, non-trivial mount operation. 911 */ 912 kern_yield(PRI_USER); 913 } 914 915 VERIFY0(zfsctl_snapshot_zname(dvp, name, sizeof(fullname), fullname)); 916 917 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 918 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(name) + 1; 919 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 920 (void) snprintf(mountpoint, mountpoint_len, 921 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 922 dvp->v_vfsp->mnt_stat.f_mntonname, name); 923 924 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, fullname, 0); 925 kmem_free(mountpoint, mountpoint_len); 926 if (err == 0) { 927 /* 928 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 929 * 930 * This is where we lie about our v_vfsp in order to 931 * make .zfs/snapshot/<snapname> accessible over NFS 932 * without requiring manual mounts of <snapname>. 933 */ 934 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 935 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 936 937 /* Clear the root flag (set via VFS_ROOT) as well. */ 938 (*vpp)->v_vflag &= ~VV_ROOT; 939 } 940 941 if (err != 0) 942 *vpp = NULL; 943 return (err); 944} 945 946static int 947zfsctl_snapdir_readdir(ap) 948 struct vop_readdir_args /* { 949 struct vnode *a_vp; 950 struct uio *a_uio; 951 struct ucred *a_cred; 952 int *a_eofflag; 953 int *ncookies; 954 u_long **a_cookies; 955 } */ *ap; 956{ 957 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 958 struct dirent entry; 959 vnode_t *vp = ap->a_vp; 960 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 961 uio_t *uio = ap->a_uio; 962 int *eofp = ap->a_eofflag; 963 off_t dots_offset; 964 int error; 965 966 ASSERT(vp->v_type == VDIR); 967 968 error = sfs_readdir_common(ZFSCTL_INO_ROOT, ZFSCTL_INO_SNAPDIR, ap, uio, 969 &dots_offset); 970 if (error != 0) { 971 if (error == ENAMETOOLONG) /* ran out of destination space */ 972 error = 0; 973 return (error); 974 } 975 976 for (;;) { 977 uint64_t cookie; 978 uint64_t id; 979 980 cookie = uio->uio_offset - dots_offset; 981 982 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 983 error = dmu_snapshot_list_next(zfsvfs->z_os, sizeof(snapname), 984 snapname, &id, &cookie, NULL); 985 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 986 if (error != 0) { 987 if (error == ENOENT) { 988 if (eofp != NULL) 989 *eofp = 1; 990 error = 0; 991 } 992 return (error); 993 } 994 995 entry.d_fileno = id; 996 entry.d_type = DT_DIR; 997 strcpy(entry.d_name, snapname); 998 entry.d_namlen = strlen(entry.d_name); 999 entry.d_reclen = sizeof(entry); 1000 error = vfs_read_dirent(ap, &entry, uio->uio_offset); 1001 if (error != 0) { 1002 if (error == ENAMETOOLONG) 1003 error = 0; 1004 return (SET_ERROR(error)); 1005 } 1006 uio->uio_offset = cookie + dots_offset; 1007 } 1008 /* NOTREACHED */ 1009} 1010 1011static int 1012zfsctl_snapdir_getattr(ap) 1013 struct vop_getattr_args /* { 1014 struct vnode *a_vp; 1015 struct vattr *a_vap; 1016 struct ucred *a_cred; 1017 } */ *ap; 1018{ 1019 vnode_t *vp = ap->a_vp; 1020 vattr_t *vap = ap->a_vap; 1021 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1022 dsl_dataset_t *ds = dmu_objset_ds(zfsvfs->z_os); 1023 sfs_node_t *node = vp->v_data; 1024 uint64_t snap_count; 1025 int err; 1026 1027 zfsctl_common_getattr(vp, vap); 1028 vap->va_ctime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1029 vap->va_mtime = vap->va_ctime; 1030 vap->va_birthtime = vap->va_ctime; 1031 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0) { 1032 err = zap_count(dmu_objset_pool(ds->ds_objset)->dp_meta_objset, 1033 dsl_dataset_phys(ds)->ds_snapnames_zapobj, &snap_count); 1034 if (err != 0) 1035 return (err); 1036 vap->va_nlink += snap_count; 1037 } 1038 vap->va_size = vap->va_nlink; 1039 1040 return (0); 1041} 1042 1043static struct vop_vector zfsctl_ops_snapdir = { 1044 .vop_default = &default_vnodeops, 1045 .vop_open = zfsctl_common_open, 1046 .vop_close = zfsctl_common_close, 1047 .vop_getattr = zfsctl_snapdir_getattr, 1048 .vop_access = zfsctl_common_access, 1049 .vop_readdir = zfsctl_snapdir_readdir, 1050 .vop_lookup = zfsctl_snapdir_lookup, 1051 .vop_reclaim = zfsctl_common_reclaim, 1052 .vop_fid = zfsctl_common_fid, 1053 .vop_print = zfsctl_common_print, 1054}; 1055 1056static int 1057zfsctl_snapshot_inactive(ap) 1058 struct vop_inactive_args /* { 1059 struct vnode *a_vp; 1060 struct thread *a_td; 1061 } */ *ap; 1062{ 1063 vnode_t *vp = ap->a_vp; 1064 1065 VERIFY(vrecycle(vp) == 1); 1066 return (0); 1067} 1068 1069static int 1070zfsctl_snapshot_reclaim(ap) 1071 struct vop_reclaim_args /* { 1072 struct vnode *a_vp; 1073 struct thread *a_td; 1074 } */ *ap; 1075{ 1076 vnode_t *vp = ap->a_vp; 1077 void *data = vp->v_data; 1078 1079 sfs_reclaim_vnode(vp); 1080 sfs_destroy_node(data); 1081 return (0); 1082} 1083 1084static int 1085zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1086{ 1087 struct mount *mp; 1088 vnode_t *dvp; 1089 vnode_t *vp; 1090 sfs_node_t *node; 1091 size_t len; 1092 int locked; 1093 int error; 1094 1095 vp = ap->a_vp; 1096 node = vp->v_data; 1097 len = strlen(node->sn_name); 1098 if (*ap->a_buflen < len) 1099 return (SET_ERROR(ENOMEM)); 1100 1101 /* 1102 * Prevent unmounting of the snapshot while the vnode lock 1103 * is not held. That is not strictly required, but allows 1104 * us to assert that an uncovered snapshot vnode is never 1105 * "leaked". 1106 */ 1107 mp = vp->v_mountedhere; 1108 if (mp == NULL) 1109 return (SET_ERROR(ENOENT)); 1110 error = vfs_busy(mp, 0); 1111 KASSERT(error == 0, ("vfs_busy(mp, 0) failed with %d", error)); 1112 1113 /* 1114 * We can vput the vnode as we can now depend on the reference owned 1115 * by the busied mp. But we also need to hold the vnode, because 1116 * the reference may go after vfs_unbusy() which has to be called 1117 * before we can lock the vnode again. 1118 */ 1119 locked = VOP_ISLOCKED(vp); 1120 vhold(vp); 1121 vput(vp); 1122 1123 /* Look up .zfs/snapshot, our parent. */ 1124 error = zfsctl_snapdir_vnode(vp->v_mount, NULL, LK_SHARED, &dvp); 1125 if (error == 0) { 1126 VOP_UNLOCK(dvp, 0); 1127 *ap->a_vpp = dvp; 1128 *ap->a_buflen -= len; 1129 bcopy(node->sn_name, ap->a_buf + *ap->a_buflen, len); 1130 } 1131 vfs_unbusy(mp); 1132 vget(vp, locked | LK_RETRY, curthread); 1133 vdrop(vp); 1134 return (error); 1135} 1136 1137/* 1138 * These VP's should never see the light of day. They should always 1139 * be covered. 1140 */ 1141static struct vop_vector zfsctl_ops_snapshot = { 1142 .vop_default = NULL, /* ensure very restricted access */ 1143 .vop_inactive = zfsctl_snapshot_inactive, 1144 .vop_reclaim = zfsctl_snapshot_reclaim, 1145 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1146 .vop_lock1 = vop_stdlock, 1147 .vop_unlock = vop_stdunlock, 1148 .vop_islocked = vop_stdislocked, 1149 .vop_advlockpurge = vop_stdadvlockpurge, /* called by vgone */ 1150 .vop_print = zfsctl_common_print, 1151}; 1152 1153int 1154zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1155{ 1156 struct mount *mp; 1157 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1158 vnode_t *vp; 1159 int error; 1160 1161 ASSERT(zfsvfs->z_ctldir != NULL); 1162 *zfsvfsp = NULL; 1163 error = sfs_vnode_get(vfsp, LK_EXCLUSIVE, 1164 ZFSCTL_INO_SNAPDIR, objsetid, &vp); 1165 if (error == 0 && vp != NULL) { 1166 /* 1167 * XXX Probably need to at least reference, if not busy, the mp. 1168 */ 1169 if (vp->v_mountedhere != NULL) 1170 *zfsvfsp = vp->v_mountedhere->mnt_data; 1171 vput(vp); 1172 } 1173 if (*zfsvfsp == NULL) 1174 return (SET_ERROR(EINVAL)); 1175 return (0); 1176} 1177 1178/* 1179 * Unmount any snapshots for the given filesystem. This is called from 1180 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1181 * snapshots. 1182 */ 1183int 1184zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1185{ 1186 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 1187 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1188 struct mount *mp; 1189 vnode_t *dvp; 1190 vnode_t *vp; 1191 sfs_node_t *node; 1192 sfs_node_t *snap; 1193 uint64_t cookie; 1194 int error; 1195 1196 ASSERT(zfsvfs->z_ctldir != NULL); 1197 1198 cookie = 0; 1199 for (;;) { 1200 uint64_t id; 1201 1202 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1203 error = dmu_snapshot_list_next(zfsvfs->z_os, sizeof(snapname), 1204 snapname, &id, &cookie, NULL); 1205 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1206 if (error != 0) { 1207 if (error == ENOENT) 1208 error = 0; 1209 break; 1210 } 1211 1212 for (;;) { 1213 error = sfs_vnode_get(vfsp, LK_EXCLUSIVE, 1214 ZFSCTL_INO_SNAPDIR, id, &vp); 1215 if (error != 0 || vp == NULL) 1216 break; 1217 1218 mp = vp->v_mountedhere; 1219 1220 /* 1221 * v_mountedhere being NULL means that the 1222 * (uncovered) vnode is in a transient state 1223 * (mounting or unmounting), so loop until it 1224 * settles down. 1225 */ 1226 if (mp != NULL) 1227 break; 1228 vput(vp); 1229 } 1230 if (error != 0) 1231 break; 1232 if (vp == NULL) 1233 continue; /* no mountpoint, nothing to do */ 1234 1235 /* 1236 * The mount-point vnode is kept locked to avoid spurious EBUSY 1237 * from a concurrent umount. 1238 * The vnode lock must have recursive locking enabled. 1239 */ 1240 vfs_ref(mp); 1241 error = dounmount(mp, fflags, curthread); 1242 KASSERT_IMPLY(error == 0, vrefcnt(vp) == 1, 1243 ("extra references after unmount")); 1244 vput(vp); 1245 if (error != 0) 1246 break; 1247 } 1248 KASSERT_IMPLY((fflags & MS_FORCE) != 0, error == 0, 1249 ("force unmounting failed")); 1250 return (error); 1251} 1252 1253