zfs_ctldir.c revision 302748
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) 2013 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/gfs.h> 74#include <sys/stat.h> 75#include <sys/dmu.h> 76#include <sys/dsl_destroy.h> 77#include <sys/dsl_deleg.h> 78#include <sys/mount.h> 79#include <sys/sunddi.h> 80 81#include "zfs_namecheck.h" 82 83typedef struct zfsctl_node { 84 gfs_dir_t zc_gfs_private; 85 uint64_t zc_id; 86 timestruc_t zc_cmtime; /* ctime and mtime, always the same */ 87} zfsctl_node_t; 88 89typedef struct zfsctl_snapdir { 90 zfsctl_node_t sd_node; 91 kmutex_t sd_lock; 92 avl_tree_t sd_snaps; 93} zfsctl_snapdir_t; 94 95typedef struct { 96 char *se_name; 97 vnode_t *se_root; 98 avl_node_t se_node; 99} zfs_snapentry_t; 100 101static int 102snapentry_compare(const void *a, const void *b) 103{ 104 const zfs_snapentry_t *sa = a; 105 const zfs_snapentry_t *sb = b; 106 int ret = strcmp(sa->se_name, sb->se_name); 107 108 if (ret < 0) 109 return (-1); 110 else if (ret > 0) 111 return (1); 112 else 113 return (0); 114} 115 116#ifdef illumos 117vnodeops_t *zfsctl_ops_root; 118vnodeops_t *zfsctl_ops_snapdir; 119vnodeops_t *zfsctl_ops_snapshot; 120vnodeops_t *zfsctl_ops_shares; 121vnodeops_t *zfsctl_ops_shares_dir; 122 123static const fs_operation_def_t zfsctl_tops_root[]; 124static const fs_operation_def_t zfsctl_tops_snapdir[]; 125static const fs_operation_def_t zfsctl_tops_snapshot[]; 126static const fs_operation_def_t zfsctl_tops_shares[]; 127#else 128static struct vop_vector zfsctl_ops_root; 129static struct vop_vector zfsctl_ops_snapdir; 130static struct vop_vector zfsctl_ops_snapshot; 131static struct vop_vector zfsctl_ops_shares; 132static struct vop_vector zfsctl_ops_shares_dir; 133#endif 134 135static vnode_t *zfsctl_mknode_snapdir(vnode_t *); 136static vnode_t *zfsctl_mknode_shares(vnode_t *); 137static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset); 138static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *); 139 140#ifdef illumos 141static gfs_opsvec_t zfsctl_opsvec[] = { 142 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root }, 143 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir }, 144 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot }, 145 { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir }, 146 { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares }, 147 { NULL } 148}; 149#endif 150 151/* 152 * Root directory elements. We only have two entries 153 * snapshot and shares. 154 */ 155static gfs_dirent_t zfsctl_root_entries[] = { 156 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE }, 157 { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE }, 158 { NULL } 159}; 160 161/* include . and .. in the calculation */ 162#define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \ 163 sizeof (gfs_dirent_t)) + 1) 164 165 166/* 167 * Initialize the various GFS pieces we'll need to create and manipulate .zfs 168 * directories. This is called from the ZFS init routine, and initializes the 169 * vnode ops vectors that we'll be using. 170 */ 171void 172zfsctl_init(void) 173{ 174#ifdef illumos 175 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0); 176#endif 177} 178 179void 180zfsctl_fini(void) 181{ 182#ifdef illumos 183 /* 184 * Remove vfsctl vnode ops 185 */ 186 if (zfsctl_ops_root) 187 vn_freevnodeops(zfsctl_ops_root); 188 if (zfsctl_ops_snapdir) 189 vn_freevnodeops(zfsctl_ops_snapdir); 190 if (zfsctl_ops_snapshot) 191 vn_freevnodeops(zfsctl_ops_snapshot); 192 if (zfsctl_ops_shares) 193 vn_freevnodeops(zfsctl_ops_shares); 194 if (zfsctl_ops_shares_dir) 195 vn_freevnodeops(zfsctl_ops_shares_dir); 196 197 zfsctl_ops_root = NULL; 198 zfsctl_ops_snapdir = NULL; 199 zfsctl_ops_snapshot = NULL; 200 zfsctl_ops_shares = NULL; 201 zfsctl_ops_shares_dir = NULL; 202#endif /* illumos */ 203} 204 205boolean_t 206zfsctl_is_node(vnode_t *vp) 207{ 208 return (vn_matchops(vp, zfsctl_ops_root) || 209 vn_matchops(vp, zfsctl_ops_snapdir) || 210 vn_matchops(vp, zfsctl_ops_snapshot) || 211 vn_matchops(vp, zfsctl_ops_shares) || 212 vn_matchops(vp, zfsctl_ops_shares_dir)); 213 214} 215 216/* 217 * Return the inode number associated with the 'snapshot' or 218 * 'shares' directory. 219 */ 220/* ARGSUSED */ 221static ino64_t 222zfsctl_root_inode_cb(vnode_t *vp, int index) 223{ 224 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 225 226 ASSERT(index < 2); 227 228 if (index == 0) 229 return (ZFSCTL_INO_SNAPDIR); 230 231 return (zfsvfs->z_shares_dir); 232} 233 234/* 235 * Create the '.zfs' directory. This directory is cached as part of the VFS 236 * structure. This results in a hold on the vfs_t. The code in zfs_umount() 237 * therefore checks against a vfs_count of 2 instead of 1. This reference 238 * is removed when the ctldir is destroyed in the unmount. 239 */ 240void 241zfsctl_create(zfsvfs_t *zfsvfs) 242{ 243 vnode_t *vp, *rvp; 244 zfsctl_node_t *zcp; 245 uint64_t crtime[2]; 246 247 ASSERT(zfsvfs->z_ctldir == NULL); 248 249 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs, 250 &zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries, 251 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL); 252 zcp = vp->v_data; 253 zcp->zc_id = ZFSCTL_INO_ROOT; 254 255 VERIFY(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp) == 0); 256 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 257 &crtime, sizeof (crtime))); 258 ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime); 259 VN_URELE(rvp); 260 261 /* 262 * We're only faking the fact that we have a root of a filesystem for 263 * the sake of the GFS interfaces. Undo the flag manipulation it did 264 * for us. 265 */ 266 vp->v_vflag &= ~VV_ROOT; 267 268 zfsvfs->z_ctldir = vp; 269 270 VOP_UNLOCK(vp, 0); 271} 272 273/* 274 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted. 275 * There might still be more references if we were force unmounted, but only 276 * new zfs_inactive() calls can occur and they don't reference .zfs 277 */ 278void 279zfsctl_destroy(zfsvfs_t *zfsvfs) 280{ 281 VN_RELE(zfsvfs->z_ctldir); 282 zfsvfs->z_ctldir = NULL; 283} 284 285/* 286 * Given a root znode, retrieve the associated .zfs directory. 287 * Add a hold to the vnode and return it. 288 */ 289vnode_t * 290zfsctl_root(znode_t *zp) 291{ 292 ASSERT(zfs_has_ctldir(zp)); 293 VN_HOLD(zp->z_zfsvfs->z_ctldir); 294 return (zp->z_zfsvfs->z_ctldir); 295} 296 297static int 298zfsctl_common_print(ap) 299 struct vop_print_args /* { 300 struct vnode *a_vp; 301 } */ *ap; 302{ 303 vnode_t *vp = ap->a_vp; 304 gfs_file_t *fp = vp->v_data; 305 306 printf(" parent = %p\n", fp->gfs_parent); 307 printf(" type = %d\n", fp->gfs_type); 308 printf(" index = %d\n", fp->gfs_index); 309 printf(" ino = %ju\n", (uintmax_t)fp->gfs_ino); 310 return (0); 311} 312 313/* 314 * Common open routine. Disallow any write access. 315 */ 316/* ARGSUSED */ 317static int 318zfsctl_common_open(struct vop_open_args *ap) 319{ 320 int flags = ap->a_mode; 321 322 if (flags & FWRITE) 323 return (SET_ERROR(EACCES)); 324 325 return (0); 326} 327 328/* 329 * Common close routine. Nothing to do here. 330 */ 331/* ARGSUSED */ 332static int 333zfsctl_common_close(struct vop_close_args *ap) 334{ 335 return (0); 336} 337 338/* 339 * Common access routine. Disallow writes. 340 */ 341/* ARGSUSED */ 342static int 343zfsctl_common_access(ap) 344 struct vop_access_args /* { 345 struct vnode *a_vp; 346 accmode_t a_accmode; 347 struct ucred *a_cred; 348 struct thread *a_td; 349 } */ *ap; 350{ 351 accmode_t accmode = ap->a_accmode; 352 353#ifdef TODO 354 if (flags & V_ACE_MASK) { 355 if (accmode & ACE_ALL_WRITE_PERMS) 356 return (SET_ERROR(EACCES)); 357 } else { 358#endif 359 if (accmode & VWRITE) 360 return (SET_ERROR(EACCES)); 361#ifdef TODO 362 } 363#endif 364 365 return (0); 366} 367 368/* 369 * Common getattr function. Fill in basic information. 370 */ 371static void 372zfsctl_common_getattr(vnode_t *vp, vattr_t *vap) 373{ 374 timestruc_t now; 375 376 vap->va_uid = 0; 377 vap->va_gid = 0; 378 vap->va_rdev = 0; 379 /* 380 * We are a purely virtual object, so we have no 381 * blocksize or allocated blocks. 382 */ 383 vap->va_blksize = 0; 384 vap->va_nblocks = 0; 385 vap->va_seq = 0; 386 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 387 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP | 388 S_IROTH | S_IXOTH; 389 vap->va_type = VDIR; 390 /* 391 * We live in the now (for atime). 392 */ 393 gethrestime(&now); 394 vap->va_atime = now; 395 /* FreeBSD: Reset chflags(2) flags. */ 396 vap->va_flags = 0; 397} 398 399/*ARGSUSED*/ 400static int 401zfsctl_common_fid(ap) 402 struct vop_fid_args /* { 403 struct vnode *a_vp; 404 struct fid *a_fid; 405 } */ *ap; 406{ 407 vnode_t *vp = ap->a_vp; 408 fid_t *fidp = (void *)ap->a_fid; 409 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 410 zfsctl_node_t *zcp = vp->v_data; 411 uint64_t object = zcp->zc_id; 412 zfid_short_t *zfid; 413 int i; 414 415 ZFS_ENTER(zfsvfs); 416 417#ifdef illumos 418 if (fidp->fid_len < SHORT_FID_LEN) { 419 fidp->fid_len = SHORT_FID_LEN; 420 ZFS_EXIT(zfsvfs); 421 return (SET_ERROR(ENOSPC)); 422 } 423#endif 424 425 zfid = (zfid_short_t *)fidp; 426 427 zfid->zf_len = SHORT_FID_LEN; 428 429 for (i = 0; i < sizeof (zfid->zf_object); i++) 430 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 431 432 /* .zfs znodes always have a generation number of 0 */ 433 for (i = 0; i < sizeof (zfid->zf_gen); i++) 434 zfid->zf_gen[i] = 0; 435 436 ZFS_EXIT(zfsvfs); 437 return (0); 438} 439 440 441/*ARGSUSED*/ 442static int 443zfsctl_shares_fid(ap) 444 struct vop_fid_args /* { 445 struct vnode *a_vp; 446 struct fid *a_fid; 447 } */ *ap; 448{ 449 vnode_t *vp = ap->a_vp; 450 fid_t *fidp = (void *)ap->a_fid; 451 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 452 znode_t *dzp; 453 int error; 454 455 ZFS_ENTER(zfsvfs); 456 457 if (zfsvfs->z_shares_dir == 0) { 458 ZFS_EXIT(zfsvfs); 459 return (SET_ERROR(ENOTSUP)); 460 } 461 462 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 463 error = VOP_FID(ZTOV(dzp), fidp); 464 VN_RELE(ZTOV(dzp)); 465 } 466 467 ZFS_EXIT(zfsvfs); 468 return (error); 469} 470 471static int 472zfsctl_common_reclaim(ap) 473 struct vop_reclaim_args /* { 474 struct vnode *a_vp; 475 struct thread *a_td; 476 } */ *ap; 477{ 478 vnode_t *vp = ap->a_vp; 479 480 /* 481 * Destroy the vm object and flush associated pages. 482 */ 483 vnode_destroy_vobject(vp); 484 VI_LOCK(vp); 485 vp->v_data = NULL; 486 VI_UNLOCK(vp); 487 return (0); 488} 489 490/* 491 * .zfs inode namespace 492 * 493 * We need to generate unique inode numbers for all files and directories 494 * within the .zfs pseudo-filesystem. We use the following scheme: 495 * 496 * ENTRY ZFSCTL_INODE 497 * .zfs 1 498 * .zfs/snapshot 2 499 * .zfs/snapshot/<snap> objectid(snap) 500 */ 501 502#define ZFSCTL_INO_SNAP(id) (id) 503 504/* 505 * Get root directory attributes. 506 */ 507/* ARGSUSED */ 508static int 509zfsctl_root_getattr(ap) 510 struct vop_getattr_args /* { 511 struct vnode *a_vp; 512 struct vattr *a_vap; 513 struct ucred *a_cred; 514 } */ *ap; 515{ 516 struct vnode *vp = ap->a_vp; 517 struct vattr *vap = ap->a_vap; 518 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 519 zfsctl_node_t *zcp = vp->v_data; 520 521 ZFS_ENTER(zfsvfs); 522 vap->va_nodeid = ZFSCTL_INO_ROOT; 523 vap->va_nlink = vap->va_size = NROOT_ENTRIES; 524 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime; 525 vap->va_birthtime = vap->va_ctime; 526 527 zfsctl_common_getattr(vp, vap); 528 ZFS_EXIT(zfsvfs); 529 530 return (0); 531} 532 533/* 534 * Special case the handling of "..". 535 */ 536/* ARGSUSED */ 537int 538zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, 539 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 540 int *direntflags, pathname_t *realpnp) 541{ 542 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 543 int err; 544 545 /* 546 * No extended attributes allowed under .zfs 547 */ 548 if (flags & LOOKUP_XATTR) 549 return (SET_ERROR(EINVAL)); 550 551 ZFS_ENTER(zfsvfs); 552 553 if (strcmp(nm, "..") == 0) { 554 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp); 555 if (err == 0) 556 VOP_UNLOCK(*vpp, 0); 557 } else { 558 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir, 559 cr, ct, direntflags, realpnp); 560 } 561 562 ZFS_EXIT(zfsvfs); 563 564 return (err); 565} 566 567static int 568zfsctl_root_print(ap) 569 struct vop_print_args /* { 570 struct vnode *a_vp; 571 } */ *ap; 572{ 573 printf(" .zfs node\n"); 574 zfsctl_common_print(ap); 575 return (0); 576} 577 578#ifdef illumos 579static int 580zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 581 caller_context_t *ct) 582{ 583 /* 584 * We only care about ACL_ENABLED so that libsec can 585 * display ACL correctly and not default to POSIX draft. 586 */ 587 if (cmd == _PC_ACL_ENABLED) { 588 *valp = _ACL_ACE_ENABLED; 589 return (0); 590 } 591 592 return (fs_pathconf(vp, cmd, valp, cr, ct)); 593} 594#endif /* illumos */ 595 596#ifdef illumos 597static const fs_operation_def_t zfsctl_tops_root[] = { 598 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 599 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 600 { VOPNAME_IOCTL, { .error = fs_inval } }, 601 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } }, 602 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 603 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 604 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } }, 605 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 606 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 607 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } }, 608 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 609 { NULL } 610}; 611#endif /* illumos */ 612 613/* 614 * Special case the handling of "..". 615 */ 616/* ARGSUSED */ 617int 618zfsctl_freebsd_root_lookup(ap) 619 struct vop_lookup_args /* { 620 struct vnode *a_dvp; 621 struct vnode **a_vpp; 622 struct componentname *a_cnp; 623 } */ *ap; 624{ 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 nameiop = ap->a_cnp->cn_nameiop; 630 char nm[NAME_MAX + 1]; 631 int err; 632 int ltype; 633 634 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE)) 635 return (EOPNOTSUPP); 636 637 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 638 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 639 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL); 640 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) { 641 ltype = VOP_ISLOCKED(dvp); 642 if (flags & ISDOTDOT) { 643 VN_HOLD(*vpp); 644 VOP_UNLOCK(dvp, 0); 645 } 646 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 647 if (flags & ISDOTDOT) { 648 VN_RELE(*vpp); 649 vn_lock(dvp, ltype| LK_RETRY); 650 } 651 } 652 653 return (err); 654} 655 656static struct vop_vector zfsctl_ops_root = { 657 .vop_default = &default_vnodeops, 658 .vop_open = zfsctl_common_open, 659 .vop_close = zfsctl_common_close, 660 .vop_ioctl = VOP_EINVAL, 661 .vop_getattr = zfsctl_root_getattr, 662 .vop_access = zfsctl_common_access, 663 .vop_readdir = gfs_vop_readdir, 664 .vop_lookup = zfsctl_freebsd_root_lookup, 665 .vop_inactive = VOP_NULL, 666 .vop_reclaim = gfs_vop_reclaim, 667#ifdef TODO 668 .vop_pathconf = zfsctl_pathconf, 669#endif 670 .vop_fid = zfsctl_common_fid, 671 .vop_print = zfsctl_root_print, 672}; 673 674/* 675 * Gets the full dataset name that corresponds to the given snapshot name 676 * Example: 677 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1" 678 */ 679static int 680zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 681{ 682 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 683 684 if (zfs_component_namecheck(name, NULL, NULL) != 0) 685 return (SET_ERROR(EILSEQ)); 686 dmu_objset_name(os, zname); 687 if (strlen(zname) + 1 + strlen(name) >= len) 688 return (SET_ERROR(ENAMETOOLONG)); 689 (void) strcat(zname, "@"); 690 (void) strcat(zname, name); 691 return (0); 692} 693 694static int 695zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 696{ 697 vnode_t *svp = sep->se_root; 698 int error; 699 700 ASSERT(vn_ismntpt(svp)); 701 702 /* this will be dropped by dounmount() */ 703 if ((error = vn_vfswlock(svp)) != 0) 704 return (error); 705 706#ifdef illumos 707 VN_HOLD(svp); 708 error = dounmount(vn_mountedvfs(svp), fflags, cr); 709 if (error) { 710 VN_RELE(svp); 711 return (error); 712 } 713 714 /* 715 * We can't use VN_RELE(), as that will try to invoke 716 * zfsctl_snapdir_inactive(), which would cause us to destroy 717 * the sd_lock mutex held by our caller. 718 */ 719 ASSERT(svp->v_count == 1); 720 gfs_vop_reclaim(svp, cr, NULL); 721 722 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 723 kmem_free(sep, sizeof (zfs_snapentry_t)); 724 725 return (0); 726#else 727 vfs_ref(vn_mountedvfs(svp)); 728 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 729#endif 730} 731 732#ifdef illumos 733static void 734zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 735{ 736 avl_index_t where; 737 vfs_t *vfsp; 738 refstr_t *pathref; 739 char newpath[MAXNAMELEN]; 740 char *tail; 741 742 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 743 ASSERT(sep != NULL); 744 745 vfsp = vn_mountedvfs(sep->se_root); 746 ASSERT(vfsp != NULL); 747 748 vfs_lock_wait(vfsp); 749 750 /* 751 * Change the name in the AVL tree. 752 */ 753 avl_remove(&sdp->sd_snaps, sep); 754 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 755 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 756 (void) strcpy(sep->se_name, nm); 757 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 758 avl_insert(&sdp->sd_snaps, sep, where); 759 760 /* 761 * Change the current mountpoint info: 762 * - update the tail of the mntpoint path 763 * - update the tail of the resource path 764 */ 765 pathref = vfs_getmntpoint(vfsp); 766 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 767 VERIFY((tail = strrchr(newpath, '/')) != NULL); 768 *(tail+1) = '\0'; 769 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 770 (void) strcat(newpath, nm); 771 refstr_rele(pathref); 772 vfs_setmntpoint(vfsp, newpath, 0); 773 774 pathref = vfs_getresource(vfsp); 775 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 776 VERIFY((tail = strrchr(newpath, '@')) != NULL); 777 *(tail+1) = '\0'; 778 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 779 (void) strcat(newpath, nm); 780 refstr_rele(pathref); 781 vfs_setresource(vfsp, newpath, 0); 782 783 vfs_unlock(vfsp); 784} 785#endif /* illumos */ 786 787#ifdef illumos 788/*ARGSUSED*/ 789static int 790zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 791 cred_t *cr, caller_context_t *ct, int flags) 792{ 793 zfsctl_snapdir_t *sdp = sdvp->v_data; 794 zfs_snapentry_t search, *sep; 795 zfsvfs_t *zfsvfs; 796 avl_index_t where; 797 char from[MAXNAMELEN], to[MAXNAMELEN]; 798 char real[MAXNAMELEN], fsname[MAXNAMELEN]; 799 int err; 800 801 zfsvfs = sdvp->v_vfsp->vfs_data; 802 ZFS_ENTER(zfsvfs); 803 804 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 805 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 806 MAXNAMELEN, NULL); 807 if (err == 0) { 808 snm = real; 809 } else if (err != ENOTSUP) { 810 ZFS_EXIT(zfsvfs); 811 return (err); 812 } 813 } 814 815 ZFS_EXIT(zfsvfs); 816 817 dmu_objset_name(zfsvfs->z_os, fsname); 818 819 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 820 if (err == 0) 821 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 822 if (err == 0) 823 err = zfs_secpolicy_rename_perms(from, to, cr); 824 if (err != 0) 825 return (err); 826 827 /* 828 * Cannot move snapshots out of the snapdir. 829 */ 830 if (sdvp != tdvp) 831 return (SET_ERROR(EINVAL)); 832 833 if (strcmp(snm, tnm) == 0) 834 return (0); 835 836 mutex_enter(&sdp->sd_lock); 837 838 search.se_name = (char *)snm; 839 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 840 mutex_exit(&sdp->sd_lock); 841 return (SET_ERROR(ENOENT)); 842 } 843 844 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0); 845 if (err == 0) 846 zfsctl_rename_snap(sdp, sep, tnm); 847 848 mutex_exit(&sdp->sd_lock); 849 850 return (err); 851} 852#endif /* illumos */ 853 854#ifdef illumos 855/* ARGSUSED */ 856static int 857zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 858 caller_context_t *ct, int flags) 859{ 860 zfsctl_snapdir_t *sdp = dvp->v_data; 861 zfs_snapentry_t *sep; 862 zfs_snapentry_t search; 863 zfsvfs_t *zfsvfs; 864 char snapname[MAXNAMELEN]; 865 char real[MAXNAMELEN]; 866 int err; 867 868 zfsvfs = dvp->v_vfsp->vfs_data; 869 ZFS_ENTER(zfsvfs); 870 871 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 872 873 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 874 MAXNAMELEN, NULL); 875 if (err == 0) { 876 name = real; 877 } else if (err != ENOTSUP) { 878 ZFS_EXIT(zfsvfs); 879 return (err); 880 } 881 } 882 883 ZFS_EXIT(zfsvfs); 884 885 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 886 if (err == 0) 887 err = zfs_secpolicy_destroy_perms(snapname, cr); 888 if (err != 0) 889 return (err); 890 891 mutex_enter(&sdp->sd_lock); 892 893 search.se_name = name; 894 sep = avl_find(&sdp->sd_snaps, &search, NULL); 895 if (sep) { 896 avl_remove(&sdp->sd_snaps, sep); 897 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 898 if (err != 0) 899 avl_add(&sdp->sd_snaps, sep); 900 else 901 err = dsl_destroy_snapshot(snapname, B_FALSE); 902 } else { 903 err = SET_ERROR(ENOENT); 904 } 905 906 mutex_exit(&sdp->sd_lock); 907 908 return (err); 909} 910#endif /* illumos */ 911 912/* 913 * This creates a snapshot under '.zfs/snapshot'. 914 */ 915/* ARGSUSED */ 916static int 917zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 918 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 919{ 920 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 921 char name[MAXNAMELEN]; 922 int err; 923 static enum symfollow follow = NO_FOLLOW; 924 static enum uio_seg seg = UIO_SYSSPACE; 925 926 if (zfs_component_namecheck(dirname, NULL, NULL) != 0) 927 return (SET_ERROR(EILSEQ)); 928 929 dmu_objset_name(zfsvfs->z_os, name); 930 931 *vpp = NULL; 932 933 err = zfs_secpolicy_snapshot_perms(name, cr); 934 if (err != 0) 935 return (err); 936 937 if (err == 0) { 938 err = dmu_objset_snapshot_one(name, dirname); 939 if (err != 0) 940 return (err); 941 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 942 } 943 944 return (err); 945} 946 947static int 948zfsctl_freebsd_snapdir_mkdir(ap) 949 struct vop_mkdir_args /* { 950 struct vnode *a_dvp; 951 struct vnode **a_vpp; 952 struct componentname *a_cnp; 953 struct vattr *a_vap; 954 } */ *ap; 955{ 956 957 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 958 959 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 960 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 961} 962 963/* 964 * Lookup entry point for the 'snapshot' directory. Try to open the 965 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 966 * Perform a mount of the associated dataset on top of the vnode. 967 */ 968/* ARGSUSED */ 969int 970zfsctl_snapdir_lookup(ap) 971 struct vop_lookup_args /* { 972 struct vnode *a_dvp; 973 struct vnode **a_vpp; 974 struct componentname *a_cnp; 975 } */ *ap; 976{ 977 vnode_t *dvp = ap->a_dvp; 978 vnode_t **vpp = ap->a_vpp; 979 struct componentname *cnp = ap->a_cnp; 980 char nm[NAME_MAX + 1]; 981 zfsctl_snapdir_t *sdp = dvp->v_data; 982 objset_t *snap; 983 char snapname[MAXNAMELEN]; 984 char real[MAXNAMELEN]; 985 char *mountpoint; 986 zfs_snapentry_t *sep, search; 987 size_t mountpoint_len; 988 avl_index_t where; 989 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 990 int err; 991 int ltype, flags = 0; 992 993 /* 994 * No extended attributes allowed under .zfs 995 */ 996 if (flags & LOOKUP_XATTR) 997 return (SET_ERROR(EINVAL)); 998 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 999 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 1000 1001 ASSERT(dvp->v_type == VDIR); 1002 1003 *vpp = NULL; 1004 1005 /* 1006 * If we get a recursive call, that means we got called 1007 * from the domount() code while it was trying to look up the 1008 * spec (which looks like a local path for zfs). We need to 1009 * add some flag to domount() to tell it not to do this lookup. 1010 */ 1011 if (MUTEX_HELD(&sdp->sd_lock)) 1012 return (SET_ERROR(ENOENT)); 1013 1014 ZFS_ENTER(zfsvfs); 1015 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1016 ZFS_EXIT(zfsvfs); 1017 return (0); 1018 } 1019 1020 if (flags & FIGNORECASE) { 1021 boolean_t conflict = B_FALSE; 1022 1023 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 1024 MAXNAMELEN, &conflict); 1025 if (err == 0) { 1026 strlcpy(nm, real, sizeof(nm)); 1027 } else if (err != ENOTSUP) { 1028 ZFS_EXIT(zfsvfs); 1029 return (err); 1030 } 1031#if 0 1032 if (realpnp) 1033 (void) strlcpy(realpnp->pn_buf, nm, 1034 realpnp->pn_bufsize); 1035 if (conflict && direntflags) 1036 *direntflags = ED_CASE_CONFLICT; 1037#endif 1038 } 1039 1040relookup: 1041 mutex_enter(&sdp->sd_lock); 1042 search.se_name = (char *)nm; 1043 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 1044 *vpp = sep->se_root; 1045 VN_HOLD(*vpp); 1046 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 1047 if (err != 0) { 1048 *vpp = NULL; 1049 } else if (*vpp == sep->se_root) { 1050 /* 1051 * The snapshot was unmounted behind our backs, 1052 * try to remount it. 1053 */ 1054 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1055 goto domount; 1056 } 1057 mutex_exit(&sdp->sd_lock); 1058 ZFS_EXIT(zfsvfs); 1059 return (err); 1060 } 1061 1062 /* 1063 * The requested snapshot is not currently mounted, look it up. 1064 */ 1065 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1066 if (err != 0) { 1067 mutex_exit(&sdp->sd_lock); 1068 ZFS_EXIT(zfsvfs); 1069 /* 1070 * handle "ls *" or "?" in a graceful manner, 1071 * forcing EILSEQ to ENOENT. 1072 * Since shell ultimately passes "*" or "?" as name to lookup 1073 */ 1074 return (err == EILSEQ ? ENOENT : err); 1075 } 1076 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1077 mutex_exit(&sdp->sd_lock); 1078#ifdef illumos 1079 ZFS_EXIT(zfsvfs); 1080 return (SET_ERROR(ENOENT)); 1081#else /* !illumos */ 1082 /* Translate errors and add SAVENAME when needed. */ 1083 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1084 err = EJUSTRETURN; 1085 cnp->cn_flags |= SAVENAME; 1086 } else { 1087 err = SET_ERROR(ENOENT); 1088 } 1089 ZFS_EXIT(zfsvfs); 1090 return (err); 1091#endif /* illumos */ 1092 } 1093 1094 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1095 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1096 (void) strcpy(sep->se_name, nm); 1097 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1098 avl_insert(&sdp->sd_snaps, sep, where); 1099 1100 dmu_objset_rele(snap, FTAG); 1101domount: 1102 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1103 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1104 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1105 (void) snprintf(mountpoint, mountpoint_len, 1106 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1107 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1108 mutex_exit(&sdp->sd_lock); 1109 1110 /* 1111 * The vnode may get reclaimed between dropping sd_lock and 1112 * getting the vnode lock. 1113 * */ 1114 err = vn_lock(*vpp, LK_EXCLUSIVE); 1115 if (err == ENOENT) 1116 goto relookup; 1117 VERIFY0(err); 1118 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1119 kmem_free(mountpoint, mountpoint_len); 1120 if (err == 0) { 1121 /* 1122 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1123 * 1124 * This is where we lie about our v_vfsp in order to 1125 * make .zfs/snapshot/<snapname> accessible over NFS 1126 * without requiring manual mounts of <snapname>. 1127 */ 1128 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1129 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1130 } 1131 ZFS_EXIT(zfsvfs); 1132 1133#ifdef illumos 1134 /* 1135 * If we had an error, drop our hold on the vnode and 1136 * zfsctl_snapshot_inactive() will clean up. 1137 */ 1138 if (err != 0) { 1139 VN_RELE(*vpp); 1140 *vpp = NULL; 1141 } 1142#else 1143 if (err != 0) 1144 *vpp = NULL; 1145#endif 1146 return (err); 1147} 1148 1149/* ARGSUSED */ 1150int 1151zfsctl_shares_lookup(ap) 1152 struct vop_lookup_args /* { 1153 struct vnode *a_dvp; 1154 struct vnode **a_vpp; 1155 struct componentname *a_cnp; 1156 } */ *ap; 1157{ 1158 vnode_t *dvp = ap->a_dvp; 1159 vnode_t **vpp = ap->a_vpp; 1160 struct componentname *cnp = ap->a_cnp; 1161 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1162 char nm[NAME_MAX + 1]; 1163 znode_t *dzp; 1164 int error; 1165 1166 ZFS_ENTER(zfsvfs); 1167 1168 ASSERT(cnp->cn_namelen < sizeof(nm)); 1169 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1170 1171 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1172 ZFS_EXIT(zfsvfs); 1173 return (0); 1174 } 1175 1176 if (zfsvfs->z_shares_dir == 0) { 1177 ZFS_EXIT(zfsvfs); 1178 return (SET_ERROR(ENOTSUP)); 1179 } 1180 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1181 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1182 VN_RELE(ZTOV(dzp)); 1183 } 1184 1185 ZFS_EXIT(zfsvfs); 1186 1187 return (error); 1188} 1189 1190/* ARGSUSED */ 1191static int 1192zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1193 offset_t *offp, offset_t *nextp, void *data, int flags) 1194{ 1195 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1196 char snapname[MAXNAMELEN]; 1197 uint64_t id, cookie; 1198 boolean_t case_conflict; 1199 int error; 1200 1201 ZFS_ENTER(zfsvfs); 1202 1203 cookie = *offp; 1204 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1205 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1206 &cookie, &case_conflict); 1207 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1208 if (error) { 1209 ZFS_EXIT(zfsvfs); 1210 if (error == ENOENT) { 1211 *eofp = 1; 1212 return (0); 1213 } 1214 return (error); 1215 } 1216 1217 if (flags & V_RDDIR_ENTFLAGS) { 1218 edirent_t *eodp = dp; 1219 1220 (void) strcpy(eodp->ed_name, snapname); 1221 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1222 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1223 } else { 1224 struct dirent64 *odp = dp; 1225 1226 (void) strcpy(odp->d_name, snapname); 1227 odp->d_ino = ZFSCTL_INO_SNAP(id); 1228 } 1229 *nextp = cookie; 1230 1231 ZFS_EXIT(zfsvfs); 1232 1233 return (0); 1234} 1235 1236/* ARGSUSED */ 1237static int 1238zfsctl_shares_readdir(ap) 1239 struct vop_readdir_args /* { 1240 struct vnode *a_vp; 1241 struct uio *a_uio; 1242 struct ucred *a_cred; 1243 int *a_eofflag; 1244 int *a_ncookies; 1245 u_long **a_cookies; 1246 } */ *ap; 1247{ 1248 vnode_t *vp = ap->a_vp; 1249 uio_t *uiop = ap->a_uio; 1250 cred_t *cr = ap->a_cred; 1251 int *eofp = ap->a_eofflag; 1252 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1253 znode_t *dzp; 1254 int error; 1255 1256 ZFS_ENTER(zfsvfs); 1257 1258 if (zfsvfs->z_shares_dir == 0) { 1259 ZFS_EXIT(zfsvfs); 1260 return (SET_ERROR(ENOTSUP)); 1261 } 1262 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1263 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1264 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1265 VN_URELE(ZTOV(dzp)); 1266 } else { 1267 *eofp = 1; 1268 error = SET_ERROR(ENOENT); 1269 } 1270 1271 ZFS_EXIT(zfsvfs); 1272 return (error); 1273} 1274 1275/* 1276 * pvp is the '.zfs' directory (zfsctl_node_t). 1277 * 1278 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1279 * 1280 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1281 * when a lookup is performed on .zfs for "snapshot". 1282 */ 1283vnode_t * 1284zfsctl_mknode_snapdir(vnode_t *pvp) 1285{ 1286 vnode_t *vp; 1287 zfsctl_snapdir_t *sdp; 1288 1289 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1290 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1291 zfsctl_snapdir_readdir_cb, NULL); 1292 sdp = vp->v_data; 1293 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1294 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1295 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1296 avl_create(&sdp->sd_snaps, snapentry_compare, 1297 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1298 VOP_UNLOCK(vp, 0); 1299 return (vp); 1300} 1301 1302vnode_t * 1303zfsctl_mknode_shares(vnode_t *pvp) 1304{ 1305 vnode_t *vp; 1306 zfsctl_node_t *sdp; 1307 1308 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1309 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1310 NULL, NULL); 1311 sdp = vp->v_data; 1312 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1313 VOP_UNLOCK(vp, 0); 1314 return (vp); 1315 1316} 1317 1318/* ARGSUSED */ 1319static int 1320zfsctl_shares_getattr(ap) 1321 struct vop_getattr_args /* { 1322 struct vnode *a_vp; 1323 struct vattr *a_vap; 1324 struct ucred *a_cred; 1325 struct thread *a_td; 1326 } */ *ap; 1327{ 1328 vnode_t *vp = ap->a_vp; 1329 vattr_t *vap = ap->a_vap; 1330 cred_t *cr = ap->a_cred; 1331 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1332 znode_t *dzp; 1333 int error; 1334 1335 ZFS_ENTER(zfsvfs); 1336 if (zfsvfs->z_shares_dir == 0) { 1337 ZFS_EXIT(zfsvfs); 1338 return (SET_ERROR(ENOTSUP)); 1339 } 1340 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1341 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1342 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1343 VN_URELE(ZTOV(dzp)); 1344 } 1345 ZFS_EXIT(zfsvfs); 1346 return (error); 1347 1348 1349} 1350 1351/* ARGSUSED */ 1352static int 1353zfsctl_snapdir_getattr(ap) 1354 struct vop_getattr_args /* { 1355 struct vnode *a_vp; 1356 struct vattr *a_vap; 1357 struct ucred *a_cred; 1358 } */ *ap; 1359{ 1360 vnode_t *vp = ap->a_vp; 1361 vattr_t *vap = ap->a_vap; 1362 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1363 zfsctl_snapdir_t *sdp = vp->v_data; 1364 1365 ZFS_ENTER(zfsvfs); 1366 zfsctl_common_getattr(vp, vap); 1367 vap->va_nodeid = gfs_file_inode(vp); 1368 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1369 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1370 vap->va_birthtime = vap->va_ctime; 1371 ZFS_EXIT(zfsvfs); 1372 1373 return (0); 1374} 1375 1376/* ARGSUSED */ 1377static int 1378zfsctl_snapdir_inactive(ap) 1379 struct vop_inactive_args /* { 1380 struct vnode *a_vp; 1381 struct thread *a_td; 1382 } */ *ap; 1383{ 1384 vnode_t *vp = ap->a_vp; 1385 zfsctl_snapdir_t *sdp = vp->v_data; 1386 zfs_snapentry_t *sep; 1387 1388 /* 1389 * On forced unmount we have to free snapshots from here. 1390 */ 1391 mutex_enter(&sdp->sd_lock); 1392 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) { 1393 avl_remove(&sdp->sd_snaps, sep); 1394 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1395 kmem_free(sep, sizeof (zfs_snapentry_t)); 1396 } 1397 mutex_exit(&sdp->sd_lock); 1398 gfs_dir_inactive(vp); 1399 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1400 mutex_destroy(&sdp->sd_lock); 1401 avl_destroy(&sdp->sd_snaps); 1402 kmem_free(sdp, sizeof (zfsctl_snapdir_t)); 1403 1404 return (0); 1405} 1406 1407static int 1408zfsctl_shares_print(ap) 1409 struct vop_print_args /* { 1410 struct vnode *a_vp; 1411 } */ *ap; 1412{ 1413 printf(" .zfs/shares node\n"); 1414 zfsctl_common_print(ap); 1415 return (0); 1416} 1417 1418static int 1419zfsctl_snapdir_print(ap) 1420 struct vop_print_args /* { 1421 struct vnode *a_vp; 1422 } */ *ap; 1423{ 1424 vnode_t *vp = ap->a_vp; 1425 zfsctl_snapdir_t *sdp = vp->v_data; 1426 1427 printf(" .zfs/snapshot node\n"); 1428 printf(" number of children = %lu\n", avl_numnodes(&sdp->sd_snaps)); 1429 zfsctl_common_print(ap); 1430 return (0); 1431} 1432 1433#ifdef illumos 1434static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1435 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1436 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1437 { VOPNAME_IOCTL, { .error = fs_inval } }, 1438 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1439 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1440 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1441 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1442 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1443 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1444 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1445 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1446 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1447 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1448 { NULL } 1449}; 1450 1451static const fs_operation_def_t zfsctl_tops_shares[] = { 1452 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1453 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1454 { VOPNAME_IOCTL, { .error = fs_inval } }, 1455 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1456 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1457 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1458 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1459 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1460 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1461 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1462 { NULL } 1463}; 1464#else /* !illumos */ 1465static struct vop_vector zfsctl_ops_snapdir = { 1466 .vop_default = &default_vnodeops, 1467 .vop_open = zfsctl_common_open, 1468 .vop_close = zfsctl_common_close, 1469 .vop_ioctl = VOP_EINVAL, 1470 .vop_getattr = zfsctl_snapdir_getattr, 1471 .vop_access = zfsctl_common_access, 1472 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1473 .vop_readdir = gfs_vop_readdir, 1474 .vop_lookup = zfsctl_snapdir_lookup, 1475 .vop_inactive = zfsctl_snapdir_inactive, 1476 .vop_reclaim = zfsctl_common_reclaim, 1477 .vop_fid = zfsctl_common_fid, 1478 .vop_print = zfsctl_snapdir_print, 1479}; 1480 1481static struct vop_vector zfsctl_ops_shares = { 1482 .vop_default = &default_vnodeops, 1483 .vop_open = zfsctl_common_open, 1484 .vop_close = zfsctl_common_close, 1485 .vop_ioctl = VOP_EINVAL, 1486 .vop_getattr = zfsctl_shares_getattr, 1487 .vop_access = zfsctl_common_access, 1488 .vop_readdir = zfsctl_shares_readdir, 1489 .vop_lookup = zfsctl_shares_lookup, 1490 .vop_inactive = VOP_NULL, 1491 .vop_reclaim = gfs_vop_reclaim, 1492 .vop_fid = zfsctl_shares_fid, 1493 .vop_print = zfsctl_shares_print, 1494}; 1495#endif /* illumos */ 1496 1497/* 1498 * pvp is the GFS vnode '.zfs/snapshot'. 1499 * 1500 * This creates a GFS node under '.zfs/snapshot' representing each 1501 * snapshot. This newly created GFS node is what we mount snapshot 1502 * vfs_t's ontop of. 1503 */ 1504static vnode_t * 1505zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1506{ 1507 vnode_t *vp; 1508 zfsctl_node_t *zcp; 1509 1510 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1511 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1512 zcp = vp->v_data; 1513 zcp->zc_id = objset; 1514 VOP_UNLOCK(vp, 0); 1515 1516 return (vp); 1517} 1518 1519 1520static int 1521zfsctl_snapshot_reclaim(ap) 1522 struct vop_inactive_args /* { 1523 struct vnode *a_vp; 1524 struct thread *a_td; 1525 } */ *ap; 1526{ 1527 vnode_t *vp = ap->a_vp; 1528 cred_t *cr = ap->a_td->td_ucred; 1529 struct vop_reclaim_args iap; 1530 zfsctl_snapdir_t *sdp; 1531 zfs_snapentry_t *sep, *next; 1532 int locked; 1533 vnode_t *dvp; 1534 1535 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1536 sdp = dvp->v_data; 1537 VOP_UNLOCK(dvp, 0); 1538 /* this may already have been unmounted */ 1539 if (sdp == NULL) { 1540 VN_RELE(dvp); 1541 return (0); 1542 } 1543 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1544 mutex_enter(&sdp->sd_lock); 1545 1546 ASSERT(!vn_ismntpt(vp)); 1547 1548 sep = avl_first(&sdp->sd_snaps); 1549 while (sep != NULL) { 1550 next = AVL_NEXT(&sdp->sd_snaps, sep); 1551 1552 if (sep->se_root == vp) { 1553 avl_remove(&sdp->sd_snaps, sep); 1554 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1555 kmem_free(sep, sizeof (zfs_snapentry_t)); 1556 break; 1557 } 1558 sep = next; 1559 } 1560 ASSERT(sep != NULL); 1561 1562 if (!locked) 1563 mutex_exit(&sdp->sd_lock); 1564 VN_RELE(dvp); 1565 1566 /* 1567 * Dispose of the vnode for the snapshot mount point. 1568 * This is safe to do because once this entry has been removed 1569 * from the AVL tree, it can't be found again, so cannot become 1570 * "active". If we lookup the same name again we will end up 1571 * creating a new vnode. 1572 */ 1573 iap.a_vp = vp; 1574 gfs_vop_reclaim(&iap); 1575 return (0); 1576 1577} 1578 1579static int 1580zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1581{ 1582 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1583 vnode_t *dvp, *vp; 1584 zfsctl_snapdir_t *sdp; 1585 zfs_snapentry_t *sep; 1586 int error; 1587 1588 ASSERT(zfsvfs->z_ctldir != NULL); 1589 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1590 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1591 if (error != 0) 1592 return (error); 1593 sdp = dvp->v_data; 1594 1595 mutex_enter(&sdp->sd_lock); 1596 sep = avl_first(&sdp->sd_snaps); 1597 while (sep != NULL) { 1598 vp = sep->se_root; 1599 if (vp == ap->a_vp) 1600 break; 1601 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1602 } 1603 if (sep == NULL) { 1604 mutex_exit(&sdp->sd_lock); 1605 error = ENOENT; 1606 } else { 1607 size_t len; 1608 1609 len = strlen(sep->se_name); 1610 *ap->a_buflen -= len; 1611 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1612 mutex_exit(&sdp->sd_lock); 1613 vref(dvp); 1614 *ap->a_vpp = dvp; 1615 } 1616 VN_RELE(dvp); 1617 1618 return (error); 1619} 1620 1621static int 1622zfsctl_snaphot_print(ap) 1623 struct vop_print_args /* { 1624 struct vnode *a_vp; 1625 } */ *ap; 1626{ 1627 vnode_t *vp = ap->a_vp; 1628 zfsctl_node_t *zcp = vp->v_data; 1629 1630 printf(" .zfs/snapshot/<snap> node\n"); 1631 printf(" id = %ju\n", (uintmax_t)zcp->zc_id); 1632 zfsctl_common_print(ap); 1633 return (0); 1634} 1635 1636/* 1637 * These VP's should never see the light of day. They should always 1638 * be covered. 1639 */ 1640static struct vop_vector zfsctl_ops_snapshot = { 1641 .vop_default = &default_vnodeops, 1642 .vop_inactive = VOP_NULL, 1643 .vop_reclaim = zfsctl_snapshot_reclaim, 1644 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1645 .vop_print = zfsctl_snaphot_print, 1646}; 1647 1648int 1649zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1650{ 1651 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1652 vnode_t *dvp, *vp; 1653 zfsctl_snapdir_t *sdp; 1654 zfsctl_node_t *zcp; 1655 zfs_snapentry_t *sep; 1656 int error; 1657 1658 ASSERT(zfsvfs->z_ctldir != NULL); 1659 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1660 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1661 if (error != 0) 1662 return (error); 1663 sdp = dvp->v_data; 1664 1665 mutex_enter(&sdp->sd_lock); 1666 sep = avl_first(&sdp->sd_snaps); 1667 while (sep != NULL) { 1668 vp = sep->se_root; 1669 zcp = vp->v_data; 1670 if (zcp->zc_id == objsetid) 1671 break; 1672 1673 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1674 } 1675 1676 if (sep != NULL) { 1677 VN_HOLD(vp); 1678 /* 1679 * Return the mounted root rather than the covered mount point. 1680 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1681 * and returns the ZFS vnode mounted on top of the GFS node. 1682 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1683 */ 1684 error = traverse(&vp, LK_SHARED | LK_RETRY); 1685 if (error == 0) { 1686 if (vp == sep->se_root) { 1687 VN_RELE(vp); /* release covered vp */ 1688 error = SET_ERROR(EINVAL); 1689 } else { 1690 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1691 VN_URELE(vp); /* put snapshot's root vp */ 1692 } 1693 } 1694 mutex_exit(&sdp->sd_lock); 1695 } else { 1696 error = SET_ERROR(EINVAL); 1697 mutex_exit(&sdp->sd_lock); 1698 } 1699 1700 VN_RELE(dvp); 1701 1702 return (error); 1703} 1704 1705/* 1706 * Unmount any snapshots for the given filesystem. This is called from 1707 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1708 * snapshots. 1709 */ 1710int 1711zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1712{ 1713 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1714 vnode_t *dvp; 1715 zfsctl_snapdir_t *sdp; 1716 zfs_snapentry_t *sep, *next; 1717 int error; 1718 1719 ASSERT(zfsvfs->z_ctldir != NULL); 1720 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1721 NULL, 0, NULL, cr, NULL, NULL, NULL); 1722 if (error != 0) 1723 return (error); 1724 sdp = dvp->v_data; 1725 1726 mutex_enter(&sdp->sd_lock); 1727 1728 sep = avl_first(&sdp->sd_snaps); 1729 while (sep != NULL) { 1730 next = AVL_NEXT(&sdp->sd_snaps, sep); 1731 1732 /* 1733 * If this snapshot is not mounted, then it must 1734 * have just been unmounted by somebody else, and 1735 * will be cleaned up by zfsctl_snapdir_inactive(). 1736 */ 1737 if (vn_ismntpt(sep->se_root)) { 1738 error = zfsctl_unmount_snap(sep, fflags, cr); 1739 if (error) { 1740 avl_index_t where; 1741 1742 /* 1743 * Before reinserting snapshot to the tree, 1744 * check if it was actually removed. For example 1745 * when snapshot mount point is busy, we will 1746 * have an error here, but there will be no need 1747 * to reinsert snapshot. 1748 */ 1749 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1750 avl_insert(&sdp->sd_snaps, sep, where); 1751 break; 1752 } 1753 } 1754 sep = next; 1755 } 1756 1757 mutex_exit(&sdp->sd_lock); 1758 VN_RELE(dvp); 1759 1760 return (error); 1761} 1762