zfs_ctldir.c revision 302750
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 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') { 1017 VOP_UNLOCK(dvp, 0); 1018 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE)); 1019 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE)); 1020 } 1021 ZFS_EXIT(zfsvfs); 1022 return (0); 1023 } 1024 1025 if (flags & FIGNORECASE) { 1026 boolean_t conflict = B_FALSE; 1027 1028 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 1029 MAXNAMELEN, &conflict); 1030 if (err == 0) { 1031 strlcpy(nm, real, sizeof(nm)); 1032 } else if (err != ENOTSUP) { 1033 ZFS_EXIT(zfsvfs); 1034 return (err); 1035 } 1036#if 0 1037 if (realpnp) 1038 (void) strlcpy(realpnp->pn_buf, nm, 1039 realpnp->pn_bufsize); 1040 if (conflict && direntflags) 1041 *direntflags = ED_CASE_CONFLICT; 1042#endif 1043 } 1044 1045relookup: 1046 mutex_enter(&sdp->sd_lock); 1047 search.se_name = (char *)nm; 1048 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 1049 *vpp = sep->se_root; 1050 VN_HOLD(*vpp); 1051 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 1052 if (err != 0) { 1053 *vpp = NULL; 1054 } else if (*vpp == sep->se_root) { 1055 /* 1056 * The snapshot was unmounted behind our backs, 1057 * try to remount it. 1058 */ 1059 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1060 goto domount; 1061 } 1062 mutex_exit(&sdp->sd_lock); 1063 ZFS_EXIT(zfsvfs); 1064 return (err); 1065 } 1066 1067 /* 1068 * The requested snapshot is not currently mounted, look it up. 1069 */ 1070 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1071 if (err != 0) { 1072 mutex_exit(&sdp->sd_lock); 1073 ZFS_EXIT(zfsvfs); 1074 /* 1075 * handle "ls *" or "?" in a graceful manner, 1076 * forcing EILSEQ to ENOENT. 1077 * Since shell ultimately passes "*" or "?" as name to lookup 1078 */ 1079 return (err == EILSEQ ? ENOENT : err); 1080 } 1081 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1082 mutex_exit(&sdp->sd_lock); 1083#ifdef illumos 1084 ZFS_EXIT(zfsvfs); 1085 return (SET_ERROR(ENOENT)); 1086#else /* !illumos */ 1087 /* Translate errors and add SAVENAME when needed. */ 1088 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1089 err = EJUSTRETURN; 1090 cnp->cn_flags |= SAVENAME; 1091 } else { 1092 err = SET_ERROR(ENOENT); 1093 } 1094 ZFS_EXIT(zfsvfs); 1095 return (err); 1096#endif /* illumos */ 1097 } 1098 1099 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1100 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1101 (void) strcpy(sep->se_name, nm); 1102 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1103 avl_insert(&sdp->sd_snaps, sep, where); 1104 1105 dmu_objset_rele(snap, FTAG); 1106domount: 1107 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1108 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1109 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1110 (void) snprintf(mountpoint, mountpoint_len, 1111 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1112 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1113 mutex_exit(&sdp->sd_lock); 1114 1115 /* 1116 * The vnode may get reclaimed between dropping sd_lock and 1117 * getting the vnode lock. 1118 * */ 1119 err = vn_lock(*vpp, LK_EXCLUSIVE); 1120 if (err == ENOENT) 1121 goto relookup; 1122 VERIFY0(err); 1123 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1124 kmem_free(mountpoint, mountpoint_len); 1125 if (err == 0) { 1126 /* 1127 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1128 * 1129 * This is where we lie about our v_vfsp in order to 1130 * make .zfs/snapshot/<snapname> accessible over NFS 1131 * without requiring manual mounts of <snapname>. 1132 */ 1133 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1134 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1135 } 1136 ZFS_EXIT(zfsvfs); 1137 1138#ifdef illumos 1139 /* 1140 * If we had an error, drop our hold on the vnode and 1141 * zfsctl_snapshot_inactive() will clean up. 1142 */ 1143 if (err != 0) { 1144 VN_RELE(*vpp); 1145 *vpp = NULL; 1146 } 1147#else 1148 if (err != 0) 1149 *vpp = NULL; 1150#endif 1151 return (err); 1152} 1153 1154/* ARGSUSED */ 1155int 1156zfsctl_shares_lookup(ap) 1157 struct vop_lookup_args /* { 1158 struct vnode *a_dvp; 1159 struct vnode **a_vpp; 1160 struct componentname *a_cnp; 1161 } */ *ap; 1162{ 1163 vnode_t *dvp = ap->a_dvp; 1164 vnode_t **vpp = ap->a_vpp; 1165 struct componentname *cnp = ap->a_cnp; 1166 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1167 char nm[NAME_MAX + 1]; 1168 znode_t *dzp; 1169 int error; 1170 1171 ZFS_ENTER(zfsvfs); 1172 1173 ASSERT(cnp->cn_namelen < sizeof(nm)); 1174 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1175 1176 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1177 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') { 1178 VOP_UNLOCK(dvp, 0); 1179 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE)); 1180 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE)); 1181 } 1182 ZFS_EXIT(zfsvfs); 1183 return (0); 1184 } 1185 1186 if (zfsvfs->z_shares_dir == 0) { 1187 ZFS_EXIT(zfsvfs); 1188 return (SET_ERROR(ENOTSUP)); 1189 } 1190 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1191 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1192 VN_RELE(ZTOV(dzp)); 1193 } 1194 1195 ZFS_EXIT(zfsvfs); 1196 1197 return (error); 1198} 1199 1200/* ARGSUSED */ 1201static int 1202zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1203 offset_t *offp, offset_t *nextp, void *data, int flags) 1204{ 1205 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1206 char snapname[MAXNAMELEN]; 1207 uint64_t id, cookie; 1208 boolean_t case_conflict; 1209 int error; 1210 1211 ZFS_ENTER(zfsvfs); 1212 1213 cookie = *offp; 1214 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1215 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1216 &cookie, &case_conflict); 1217 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1218 if (error) { 1219 ZFS_EXIT(zfsvfs); 1220 if (error == ENOENT) { 1221 *eofp = 1; 1222 return (0); 1223 } 1224 return (error); 1225 } 1226 1227 if (flags & V_RDDIR_ENTFLAGS) { 1228 edirent_t *eodp = dp; 1229 1230 (void) strcpy(eodp->ed_name, snapname); 1231 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1232 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1233 } else { 1234 struct dirent64 *odp = dp; 1235 1236 (void) strcpy(odp->d_name, snapname); 1237 odp->d_ino = ZFSCTL_INO_SNAP(id); 1238 } 1239 *nextp = cookie; 1240 1241 ZFS_EXIT(zfsvfs); 1242 1243 return (0); 1244} 1245 1246/* ARGSUSED */ 1247static int 1248zfsctl_shares_readdir(ap) 1249 struct vop_readdir_args /* { 1250 struct vnode *a_vp; 1251 struct uio *a_uio; 1252 struct ucred *a_cred; 1253 int *a_eofflag; 1254 int *a_ncookies; 1255 u_long **a_cookies; 1256 } */ *ap; 1257{ 1258 vnode_t *vp = ap->a_vp; 1259 uio_t *uiop = ap->a_uio; 1260 cred_t *cr = ap->a_cred; 1261 int *eofp = ap->a_eofflag; 1262 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1263 znode_t *dzp; 1264 int error; 1265 1266 ZFS_ENTER(zfsvfs); 1267 1268 if (zfsvfs->z_shares_dir == 0) { 1269 ZFS_EXIT(zfsvfs); 1270 return (SET_ERROR(ENOTSUP)); 1271 } 1272 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1273 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1274 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1275 VN_URELE(ZTOV(dzp)); 1276 } else { 1277 *eofp = 1; 1278 error = SET_ERROR(ENOENT); 1279 } 1280 1281 ZFS_EXIT(zfsvfs); 1282 return (error); 1283} 1284 1285/* 1286 * pvp is the '.zfs' directory (zfsctl_node_t). 1287 * 1288 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1289 * 1290 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1291 * when a lookup is performed on .zfs for "snapshot". 1292 */ 1293vnode_t * 1294zfsctl_mknode_snapdir(vnode_t *pvp) 1295{ 1296 vnode_t *vp; 1297 zfsctl_snapdir_t *sdp; 1298 1299 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1300 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1301 zfsctl_snapdir_readdir_cb, NULL); 1302 sdp = vp->v_data; 1303 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1304 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1305 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1306 avl_create(&sdp->sd_snaps, snapentry_compare, 1307 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1308 VOP_UNLOCK(vp, 0); 1309 return (vp); 1310} 1311 1312vnode_t * 1313zfsctl_mknode_shares(vnode_t *pvp) 1314{ 1315 vnode_t *vp; 1316 zfsctl_node_t *sdp; 1317 1318 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1319 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1320 NULL, NULL); 1321 sdp = vp->v_data; 1322 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1323 VOP_UNLOCK(vp, 0); 1324 return (vp); 1325 1326} 1327 1328/* ARGSUSED */ 1329static int 1330zfsctl_shares_getattr(ap) 1331 struct vop_getattr_args /* { 1332 struct vnode *a_vp; 1333 struct vattr *a_vap; 1334 struct ucred *a_cred; 1335 struct thread *a_td; 1336 } */ *ap; 1337{ 1338 vnode_t *vp = ap->a_vp; 1339 vattr_t *vap = ap->a_vap; 1340 cred_t *cr = ap->a_cred; 1341 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1342 znode_t *dzp; 1343 int error; 1344 1345 ZFS_ENTER(zfsvfs); 1346 if (zfsvfs->z_shares_dir == 0) { 1347 ZFS_EXIT(zfsvfs); 1348 return (SET_ERROR(ENOTSUP)); 1349 } 1350 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1351 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1352 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1353 VN_URELE(ZTOV(dzp)); 1354 } 1355 ZFS_EXIT(zfsvfs); 1356 return (error); 1357 1358 1359} 1360 1361/* ARGSUSED */ 1362static int 1363zfsctl_snapdir_getattr(ap) 1364 struct vop_getattr_args /* { 1365 struct vnode *a_vp; 1366 struct vattr *a_vap; 1367 struct ucred *a_cred; 1368 } */ *ap; 1369{ 1370 vnode_t *vp = ap->a_vp; 1371 vattr_t *vap = ap->a_vap; 1372 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1373 zfsctl_snapdir_t *sdp = vp->v_data; 1374 1375 ZFS_ENTER(zfsvfs); 1376 zfsctl_common_getattr(vp, vap); 1377 vap->va_nodeid = gfs_file_inode(vp); 1378 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1379 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1380 vap->va_birthtime = vap->va_ctime; 1381 ZFS_EXIT(zfsvfs); 1382 1383 return (0); 1384} 1385 1386/* ARGSUSED */ 1387static int 1388zfsctl_snapdir_inactive(ap) 1389 struct vop_inactive_args /* { 1390 struct vnode *a_vp; 1391 struct thread *a_td; 1392 } */ *ap; 1393{ 1394 vnode_t *vp = ap->a_vp; 1395 zfsctl_snapdir_t *sdp = vp->v_data; 1396 zfs_snapentry_t *sep; 1397 1398 /* 1399 * On forced unmount we have to free snapshots from here. 1400 */ 1401 mutex_enter(&sdp->sd_lock); 1402 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) { 1403 avl_remove(&sdp->sd_snaps, sep); 1404 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1405 kmem_free(sep, sizeof (zfs_snapentry_t)); 1406 } 1407 mutex_exit(&sdp->sd_lock); 1408 gfs_dir_inactive(vp); 1409 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1410 mutex_destroy(&sdp->sd_lock); 1411 avl_destroy(&sdp->sd_snaps); 1412 kmem_free(sdp, sizeof (zfsctl_snapdir_t)); 1413 1414 return (0); 1415} 1416 1417static int 1418zfsctl_shares_print(ap) 1419 struct vop_print_args /* { 1420 struct vnode *a_vp; 1421 } */ *ap; 1422{ 1423 printf(" .zfs/shares node\n"); 1424 zfsctl_common_print(ap); 1425 return (0); 1426} 1427 1428static int 1429zfsctl_snapdir_print(ap) 1430 struct vop_print_args /* { 1431 struct vnode *a_vp; 1432 } */ *ap; 1433{ 1434 vnode_t *vp = ap->a_vp; 1435 zfsctl_snapdir_t *sdp = vp->v_data; 1436 1437 printf(" .zfs/snapshot node\n"); 1438 printf(" number of children = %lu\n", avl_numnodes(&sdp->sd_snaps)); 1439 zfsctl_common_print(ap); 1440 return (0); 1441} 1442 1443#ifdef illumos 1444static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1445 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1446 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1447 { VOPNAME_IOCTL, { .error = fs_inval } }, 1448 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1449 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1450 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1451 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1452 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1453 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1454 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1455 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1456 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1457 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1458 { NULL } 1459}; 1460 1461static const fs_operation_def_t zfsctl_tops_shares[] = { 1462 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1463 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1464 { VOPNAME_IOCTL, { .error = fs_inval } }, 1465 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1466 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1467 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1468 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1469 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1470 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1471 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1472 { NULL } 1473}; 1474#else /* !illumos */ 1475static struct vop_vector zfsctl_ops_snapdir = { 1476 .vop_default = &default_vnodeops, 1477 .vop_open = zfsctl_common_open, 1478 .vop_close = zfsctl_common_close, 1479 .vop_ioctl = VOP_EINVAL, 1480 .vop_getattr = zfsctl_snapdir_getattr, 1481 .vop_access = zfsctl_common_access, 1482 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1483 .vop_readdir = gfs_vop_readdir, 1484 .vop_lookup = zfsctl_snapdir_lookup, 1485 .vop_inactive = zfsctl_snapdir_inactive, 1486 .vop_reclaim = zfsctl_common_reclaim, 1487 .vop_fid = zfsctl_common_fid, 1488 .vop_print = zfsctl_snapdir_print, 1489}; 1490 1491static struct vop_vector zfsctl_ops_shares = { 1492 .vop_default = &default_vnodeops, 1493 .vop_open = zfsctl_common_open, 1494 .vop_close = zfsctl_common_close, 1495 .vop_ioctl = VOP_EINVAL, 1496 .vop_getattr = zfsctl_shares_getattr, 1497 .vop_access = zfsctl_common_access, 1498 .vop_readdir = zfsctl_shares_readdir, 1499 .vop_lookup = zfsctl_shares_lookup, 1500 .vop_inactive = VOP_NULL, 1501 .vop_reclaim = gfs_vop_reclaim, 1502 .vop_fid = zfsctl_shares_fid, 1503 .vop_print = zfsctl_shares_print, 1504}; 1505#endif /* illumos */ 1506 1507/* 1508 * pvp is the GFS vnode '.zfs/snapshot'. 1509 * 1510 * This creates a GFS node under '.zfs/snapshot' representing each 1511 * snapshot. This newly created GFS node is what we mount snapshot 1512 * vfs_t's ontop of. 1513 */ 1514static vnode_t * 1515zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1516{ 1517 vnode_t *vp; 1518 zfsctl_node_t *zcp; 1519 1520 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1521 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1522 zcp = vp->v_data; 1523 zcp->zc_id = objset; 1524 VOP_UNLOCK(vp, 0); 1525 1526 return (vp); 1527} 1528 1529 1530static int 1531zfsctl_snapshot_reclaim(ap) 1532 struct vop_inactive_args /* { 1533 struct vnode *a_vp; 1534 struct thread *a_td; 1535 } */ *ap; 1536{ 1537 vnode_t *vp = ap->a_vp; 1538 cred_t *cr = ap->a_td->td_ucred; 1539 struct vop_reclaim_args iap; 1540 zfsctl_snapdir_t *sdp; 1541 zfs_snapentry_t *sep, *next; 1542 int locked; 1543 vnode_t *dvp; 1544 1545 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1546 sdp = dvp->v_data; 1547 /* this may already have been unmounted */ 1548 if (sdp == NULL) { 1549 VN_RELE(dvp); 1550 return (0); 1551 } 1552 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1553 mutex_enter(&sdp->sd_lock); 1554 1555 ASSERT(!vn_ismntpt(vp)); 1556 1557 sep = avl_first(&sdp->sd_snaps); 1558 while (sep != NULL) { 1559 next = AVL_NEXT(&sdp->sd_snaps, sep); 1560 1561 if (sep->se_root == vp) { 1562 avl_remove(&sdp->sd_snaps, sep); 1563 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1564 kmem_free(sep, sizeof (zfs_snapentry_t)); 1565 break; 1566 } 1567 sep = next; 1568 } 1569 ASSERT(sep != NULL); 1570 1571 if (!locked) 1572 mutex_exit(&sdp->sd_lock); 1573 VN_RELE(dvp); 1574 1575 /* 1576 * Dispose of the vnode for the snapshot mount point. 1577 * This is safe to do because once this entry has been removed 1578 * from the AVL tree, it can't be found again, so cannot become 1579 * "active". If we lookup the same name again we will end up 1580 * creating a new vnode. 1581 */ 1582 iap.a_vp = vp; 1583 gfs_vop_reclaim(&iap); 1584 return (0); 1585 1586} 1587 1588static int 1589zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1590{ 1591 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1592 vnode_t *dvp, *vp; 1593 zfsctl_snapdir_t *sdp; 1594 zfs_snapentry_t *sep; 1595 int error; 1596 1597 ASSERT(zfsvfs->z_ctldir != NULL); 1598 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1599 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1600 if (error != 0) 1601 return (error); 1602 sdp = dvp->v_data; 1603 1604 mutex_enter(&sdp->sd_lock); 1605 sep = avl_first(&sdp->sd_snaps); 1606 while (sep != NULL) { 1607 vp = sep->se_root; 1608 if (vp == ap->a_vp) 1609 break; 1610 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1611 } 1612 if (sep == NULL) { 1613 mutex_exit(&sdp->sd_lock); 1614 error = ENOENT; 1615 } else { 1616 size_t len; 1617 1618 len = strlen(sep->se_name); 1619 *ap->a_buflen -= len; 1620 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1621 mutex_exit(&sdp->sd_lock); 1622 vref(dvp); 1623 *ap->a_vpp = dvp; 1624 } 1625 VN_RELE(dvp); 1626 1627 return (error); 1628} 1629 1630static int 1631zfsctl_snaphot_print(ap) 1632 struct vop_print_args /* { 1633 struct vnode *a_vp; 1634 } */ *ap; 1635{ 1636 vnode_t *vp = ap->a_vp; 1637 zfsctl_node_t *zcp = vp->v_data; 1638 1639 printf(" .zfs/snapshot/<snap> node\n"); 1640 printf(" id = %ju\n", (uintmax_t)zcp->zc_id); 1641 zfsctl_common_print(ap); 1642 return (0); 1643} 1644 1645/* 1646 * These VP's should never see the light of day. They should always 1647 * be covered. 1648 */ 1649static struct vop_vector zfsctl_ops_snapshot = { 1650 .vop_default = &default_vnodeops, 1651 .vop_inactive = VOP_NULL, 1652 .vop_reclaim = zfsctl_snapshot_reclaim, 1653 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1654 .vop_print = zfsctl_snaphot_print, 1655}; 1656 1657int 1658zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1659{ 1660 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1661 vnode_t *dvp, *vp; 1662 zfsctl_snapdir_t *sdp; 1663 zfsctl_node_t *zcp; 1664 zfs_snapentry_t *sep; 1665 int error; 1666 1667 ASSERT(zfsvfs->z_ctldir != NULL); 1668 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1669 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1670 if (error != 0) 1671 return (error); 1672 sdp = dvp->v_data; 1673 1674 mutex_enter(&sdp->sd_lock); 1675 sep = avl_first(&sdp->sd_snaps); 1676 while (sep != NULL) { 1677 vp = sep->se_root; 1678 zcp = vp->v_data; 1679 if (zcp->zc_id == objsetid) 1680 break; 1681 1682 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1683 } 1684 1685 if (sep != NULL) { 1686 VN_HOLD(vp); 1687 /* 1688 * Return the mounted root rather than the covered mount point. 1689 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1690 * and returns the ZFS vnode mounted on top of the GFS node. 1691 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1692 */ 1693 error = traverse(&vp, LK_SHARED | LK_RETRY); 1694 if (error == 0) { 1695 if (vp == sep->se_root) { 1696 VN_RELE(vp); /* release covered vp */ 1697 error = SET_ERROR(EINVAL); 1698 } else { 1699 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1700 VN_URELE(vp); /* put snapshot's root vp */ 1701 } 1702 } 1703 mutex_exit(&sdp->sd_lock); 1704 } else { 1705 error = SET_ERROR(EINVAL); 1706 mutex_exit(&sdp->sd_lock); 1707 } 1708 1709 VN_RELE(dvp); 1710 1711 return (error); 1712} 1713 1714/* 1715 * Unmount any snapshots for the given filesystem. This is called from 1716 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1717 * snapshots. 1718 */ 1719int 1720zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1721{ 1722 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1723 vnode_t *dvp; 1724 zfsctl_snapdir_t *sdp; 1725 zfs_snapentry_t *sep, *next; 1726 int error; 1727 1728 ASSERT(zfsvfs->z_ctldir != NULL); 1729 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1730 NULL, 0, NULL, cr, NULL, NULL, NULL); 1731 if (error != 0) 1732 return (error); 1733 sdp = dvp->v_data; 1734 1735 mutex_enter(&sdp->sd_lock); 1736 1737 sep = avl_first(&sdp->sd_snaps); 1738 while (sep != NULL) { 1739 next = AVL_NEXT(&sdp->sd_snaps, sep); 1740 1741 /* 1742 * If this snapshot is not mounted, then it must 1743 * have just been unmounted by somebody else, and 1744 * will be cleaned up by zfsctl_snapdir_inactive(). 1745 */ 1746 if (vn_ismntpt(sep->se_root)) { 1747 error = zfsctl_unmount_snap(sep, fflags, cr); 1748 if (error) { 1749 avl_index_t where; 1750 1751 /* 1752 * Before reinserting snapshot to the tree, 1753 * check if it was actually removed. For example 1754 * when snapshot mount point is busy, we will 1755 * have an error here, but there will be no need 1756 * to reinsert snapshot. 1757 */ 1758 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1759 avl_insert(&sdp->sd_snaps, sep, where); 1760 break; 1761 } 1762 } 1763 sep = next; 1764 } 1765 1766 mutex_exit(&sdp->sd_lock); 1767 VN_RELE(dvp); 1768 1769 return (error); 1770} 1771