zfs_ctldir.c revision 252749
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 */ 25 26/* 27 * ZFS control directory (a.k.a. ".zfs") 28 * 29 * This directory provides a common location for all ZFS meta-objects. 30 * Currently, this is only the 'snapshot' directory, but this may expand in the 31 * future. The elements are built using the GFS primitives, as the hierarchy 32 * does not actually exist on disk. 33 * 34 * For 'snapshot', we don't want to have all snapshots always mounted, because 35 * this would take up a huge amount of space in /etc/mnttab. We have three 36 * types of objects: 37 * 38 * ctldir ------> snapshotdir -------> snapshot 39 * | 40 * | 41 * V 42 * mounted fs 43 * 44 * The 'snapshot' node contains just enough information to lookup '..' and act 45 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we 46 * perform an automount of the underlying filesystem and return the 47 * corresponding vnode. 48 * 49 * All mounts are handled automatically by the kernel, but unmounts are 50 * (currently) handled from user land. The main reason is that there is no 51 * reliable way to auto-unmount the filesystem when it's "no longer in use". 52 * When the user unmounts a filesystem, we call zfsctl_unmount(), which 53 * unmounts any snapshots within the snapshot directory. 54 * 55 * The '.zfs', '.zfs/snapshot', and all directories created under 56 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and 57 * share the same vfs_t as the head filesystem (what '.zfs' lives under). 58 * 59 * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>' 60 * (ie: snapshots) are ZFS nodes and have their own unique vfs_t. 61 * However, vnodes within these mounted on file systems have their v_vfsp 62 * fields set to the head filesystem to make NFS happy (see 63 * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t 64 * so that it cannot be freed until all snapshots have been unmounted. 65 */ 66 67#include <sys/zfs_context.h> 68#include <sys/zfs_ctldir.h> 69#include <sys/zfs_ioctl.h> 70#include <sys/zfs_vfsops.h> 71#include <sys/namei.h> 72#include <sys/gfs.h> 73#include <sys/stat.h> 74#include <sys/dmu.h> 75#include <sys/dsl_destroy.h> 76#include <sys/dsl_deleg.h> 77#include <sys/mount.h> 78#include <sys/sunddi.h> 79 80#include "zfs_namecheck.h" 81 82typedef struct zfsctl_node { 83 gfs_dir_t zc_gfs_private; 84 uint64_t zc_id; 85 timestruc_t zc_cmtime; /* ctime and mtime, always the same */ 86} zfsctl_node_t; 87 88typedef struct zfsctl_snapdir { 89 zfsctl_node_t sd_node; 90 kmutex_t sd_lock; 91 avl_tree_t sd_snaps; 92} zfsctl_snapdir_t; 93 94typedef struct { 95 char *se_name; 96 vnode_t *se_root; 97 avl_node_t se_node; 98} zfs_snapentry_t; 99 100static int 101snapentry_compare(const void *a, const void *b) 102{ 103 const zfs_snapentry_t *sa = a; 104 const zfs_snapentry_t *sb = b; 105 int ret = strcmp(sa->se_name, sb->se_name); 106 107 if (ret < 0) 108 return (-1); 109 else if (ret > 0) 110 return (1); 111 else 112 return (0); 113} 114 115#ifdef sun 116vnodeops_t *zfsctl_ops_root; 117vnodeops_t *zfsctl_ops_snapdir; 118vnodeops_t *zfsctl_ops_snapshot; 119vnodeops_t *zfsctl_ops_shares; 120vnodeops_t *zfsctl_ops_shares_dir; 121 122static const fs_operation_def_t zfsctl_tops_root[]; 123static const fs_operation_def_t zfsctl_tops_snapdir[]; 124static const fs_operation_def_t zfsctl_tops_snapshot[]; 125static const fs_operation_def_t zfsctl_tops_shares[]; 126#else /* !sun */ 127static struct vop_vector zfsctl_ops_root; 128static struct vop_vector zfsctl_ops_snapdir; 129static struct vop_vector zfsctl_ops_snapshot; 130static struct vop_vector zfsctl_ops_shares; 131static struct vop_vector zfsctl_ops_shares_dir; 132#endif /* !sun */ 133 134static vnode_t *zfsctl_mknode_snapdir(vnode_t *); 135static vnode_t *zfsctl_mknode_shares(vnode_t *); 136static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset); 137static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *); 138 139#ifdef sun 140static gfs_opsvec_t zfsctl_opsvec[] = { 141 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root }, 142 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir }, 143 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot }, 144 { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir }, 145 { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares }, 146 { NULL } 147}; 148#endif /* sun */ 149 150/* 151 * Root directory elements. We only have two entries 152 * snapshot and shares. 153 */ 154static gfs_dirent_t zfsctl_root_entries[] = { 155 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE }, 156 { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE }, 157 { NULL } 158}; 159 160/* include . and .. in the calculation */ 161#define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \ 162 sizeof (gfs_dirent_t)) + 1) 163 164 165/* 166 * Initialize the various GFS pieces we'll need to create and manipulate .zfs 167 * directories. This is called from the ZFS init routine, and initializes the 168 * vnode ops vectors that we'll be using. 169 */ 170void 171zfsctl_init(void) 172{ 173#ifdef sun 174 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0); 175#endif 176} 177 178void 179zfsctl_fini(void) 180{ 181#ifdef sun 182 /* 183 * Remove vfsctl vnode ops 184 */ 185 if (zfsctl_ops_root) 186 vn_freevnodeops(zfsctl_ops_root); 187 if (zfsctl_ops_snapdir) 188 vn_freevnodeops(zfsctl_ops_snapdir); 189 if (zfsctl_ops_snapshot) 190 vn_freevnodeops(zfsctl_ops_snapshot); 191 if (zfsctl_ops_shares) 192 vn_freevnodeops(zfsctl_ops_shares); 193 if (zfsctl_ops_shares_dir) 194 vn_freevnodeops(zfsctl_ops_shares_dir); 195 196 zfsctl_ops_root = NULL; 197 zfsctl_ops_snapdir = NULL; 198 zfsctl_ops_snapshot = NULL; 199 zfsctl_ops_shares = NULL; 200 zfsctl_ops_shares_dir = NULL; 201#endif /* sun */ 202} 203 204boolean_t 205zfsctl_is_node(vnode_t *vp) 206{ 207 return (vn_matchops(vp, zfsctl_ops_root) || 208 vn_matchops(vp, zfsctl_ops_snapdir) || 209 vn_matchops(vp, zfsctl_ops_snapshot) || 210 vn_matchops(vp, zfsctl_ops_shares) || 211 vn_matchops(vp, zfsctl_ops_shares_dir)); 212 213} 214 215/* 216 * Return the inode number associated with the 'snapshot' or 217 * 'shares' directory. 218 */ 219/* ARGSUSED */ 220static ino64_t 221zfsctl_root_inode_cb(vnode_t *vp, int index) 222{ 223 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 224 225 ASSERT(index <= 2); 226 227 if (index == 0) 228 return (ZFSCTL_INO_SNAPDIR); 229 230 return (zfsvfs->z_shares_dir); 231} 232 233/* 234 * Create the '.zfs' directory. This directory is cached as part of the VFS 235 * structure. This results in a hold on the vfs_t. The code in zfs_umount() 236 * therefore checks against a vfs_count of 2 instead of 1. This reference 237 * is removed when the ctldir is destroyed in the unmount. 238 */ 239void 240zfsctl_create(zfsvfs_t *zfsvfs) 241{ 242 vnode_t *vp, *rvp; 243 zfsctl_node_t *zcp; 244 uint64_t crtime[2]; 245 246 ASSERT(zfsvfs->z_ctldir == NULL); 247 248 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs, 249 &zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries, 250 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL); 251 zcp = vp->v_data; 252 zcp->zc_id = ZFSCTL_INO_ROOT; 253 254 VERIFY(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp) == 0); 255 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 256 &crtime, sizeof (crtime))); 257 ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime); 258 VN_URELE(rvp); 259 260 /* 261 * We're only faking the fact that we have a root of a filesystem for 262 * the sake of the GFS interfaces. Undo the flag manipulation it did 263 * for us. 264 */ 265 vp->v_vflag &= ~VV_ROOT; 266 267 zfsvfs->z_ctldir = vp; 268 269 VOP_UNLOCK(vp, 0); 270} 271 272/* 273 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted. 274 * There might still be more references if we were force unmounted, but only 275 * new zfs_inactive() calls can occur and they don't reference .zfs 276 */ 277void 278zfsctl_destroy(zfsvfs_t *zfsvfs) 279{ 280 VN_RELE(zfsvfs->z_ctldir); 281 zfsvfs->z_ctldir = NULL; 282} 283 284/* 285 * Given a root znode, retrieve the associated .zfs directory. 286 * Add a hold to the vnode and return it. 287 */ 288vnode_t * 289zfsctl_root(znode_t *zp) 290{ 291 ASSERT(zfs_has_ctldir(zp)); 292 VN_HOLD(zp->z_zfsvfs->z_ctldir); 293 return (zp->z_zfsvfs->z_ctldir); 294} 295 296/* 297 * Common open routine. Disallow any write access. 298 */ 299/* ARGSUSED */ 300static int 301zfsctl_common_open(struct vop_open_args *ap) 302{ 303 int flags = ap->a_mode; 304 305 if (flags & FWRITE) 306 return (SET_ERROR(EACCES)); 307 308 return (0); 309} 310 311/* 312 * Common close routine. Nothing to do here. 313 */ 314/* ARGSUSED */ 315static int 316zfsctl_common_close(struct vop_close_args *ap) 317{ 318 return (0); 319} 320 321/* 322 * Common access routine. Disallow writes. 323 */ 324/* ARGSUSED */ 325static int 326zfsctl_common_access(ap) 327 struct vop_access_args /* { 328 struct vnode *a_vp; 329 accmode_t a_accmode; 330 struct ucred *a_cred; 331 struct thread *a_td; 332 } */ *ap; 333{ 334 accmode_t accmode = ap->a_accmode; 335 336#ifdef TODO 337 if (flags & V_ACE_MASK) { 338 if (accmode & ACE_ALL_WRITE_PERMS) 339 return (SET_ERROR(EACCES)); 340 } else { 341#endif 342 if (accmode & VWRITE) 343 return (SET_ERROR(EACCES)); 344#ifdef TODO 345 } 346#endif 347 348 return (0); 349} 350 351/* 352 * Common getattr function. Fill in basic information. 353 */ 354static void 355zfsctl_common_getattr(vnode_t *vp, vattr_t *vap) 356{ 357 timestruc_t now; 358 359 vap->va_uid = 0; 360 vap->va_gid = 0; 361 vap->va_rdev = 0; 362 /* 363 * We are a purely virtual object, so we have no 364 * blocksize or allocated blocks. 365 */ 366 vap->va_blksize = 0; 367 vap->va_nblocks = 0; 368 vap->va_seq = 0; 369 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 370 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP | 371 S_IROTH | S_IXOTH; 372 vap->va_type = VDIR; 373 /* 374 * We live in the now (for atime). 375 */ 376 gethrestime(&now); 377 vap->va_atime = now; 378 /* FreeBSD: Reset chflags(2) flags. */ 379 vap->va_flags = 0; 380} 381 382/*ARGSUSED*/ 383static int 384zfsctl_common_fid(ap) 385 struct vop_fid_args /* { 386 struct vnode *a_vp; 387 struct fid *a_fid; 388 } */ *ap; 389{ 390 vnode_t *vp = ap->a_vp; 391 fid_t *fidp = (void *)ap->a_fid; 392 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 393 zfsctl_node_t *zcp = vp->v_data; 394 uint64_t object = zcp->zc_id; 395 zfid_short_t *zfid; 396 int i; 397 398 ZFS_ENTER(zfsvfs); 399 400#ifdef illumos 401 if (fidp->fid_len < SHORT_FID_LEN) { 402 fidp->fid_len = SHORT_FID_LEN; 403 ZFS_EXIT(zfsvfs); 404 return (SET_ERROR(ENOSPC)); 405 } 406#else 407 fidp->fid_len = SHORT_FID_LEN; 408#endif 409 410 zfid = (zfid_short_t *)fidp; 411 412 zfid->zf_len = SHORT_FID_LEN; 413 414 for (i = 0; i < sizeof (zfid->zf_object); i++) 415 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 416 417 /* .zfs znodes always have a generation number of 0 */ 418 for (i = 0; i < sizeof (zfid->zf_gen); i++) 419 zfid->zf_gen[i] = 0; 420 421 ZFS_EXIT(zfsvfs); 422 return (0); 423} 424 425 426/*ARGSUSED*/ 427static int 428zfsctl_shares_fid(ap) 429 struct vop_fid_args /* { 430 struct vnode *a_vp; 431 struct fid *a_fid; 432 } */ *ap; 433{ 434 vnode_t *vp = ap->a_vp; 435 fid_t *fidp = (void *)ap->a_fid; 436 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 437 znode_t *dzp; 438 int error; 439 440 ZFS_ENTER(zfsvfs); 441 442 if (zfsvfs->z_shares_dir == 0) { 443 ZFS_EXIT(zfsvfs); 444 return (SET_ERROR(ENOTSUP)); 445 } 446 447 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 448 error = VOP_FID(ZTOV(dzp), fidp); 449 VN_RELE(ZTOV(dzp)); 450 } 451 452 ZFS_EXIT(zfsvfs); 453 return (error); 454} 455 456static int 457zfsctl_common_reclaim(ap) 458 struct vop_reclaim_args /* { 459 struct vnode *a_vp; 460 struct thread *a_td; 461 } */ *ap; 462{ 463 vnode_t *vp = ap->a_vp; 464 465 /* 466 * Destroy the vm object and flush associated pages. 467 */ 468 vnode_destroy_vobject(vp); 469 VI_LOCK(vp); 470 vp->v_data = NULL; 471 VI_UNLOCK(vp); 472 return (0); 473} 474 475/* 476 * .zfs inode namespace 477 * 478 * We need to generate unique inode numbers for all files and directories 479 * within the .zfs pseudo-filesystem. We use the following scheme: 480 * 481 * ENTRY ZFSCTL_INODE 482 * .zfs 1 483 * .zfs/snapshot 2 484 * .zfs/snapshot/<snap> objectid(snap) 485 */ 486 487#define ZFSCTL_INO_SNAP(id) (id) 488 489/* 490 * Get root directory attributes. 491 */ 492/* ARGSUSED */ 493static int 494zfsctl_root_getattr(ap) 495 struct vop_getattr_args /* { 496 struct vnode *a_vp; 497 struct vattr *a_vap; 498 struct ucred *a_cred; 499 } */ *ap; 500{ 501 struct vnode *vp = ap->a_vp; 502 struct vattr *vap = ap->a_vap; 503 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 504 zfsctl_node_t *zcp = vp->v_data; 505 506 ZFS_ENTER(zfsvfs); 507 vap->va_nodeid = ZFSCTL_INO_ROOT; 508 vap->va_nlink = vap->va_size = NROOT_ENTRIES; 509 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime; 510 vap->va_birthtime = vap->va_ctime; 511 512 zfsctl_common_getattr(vp, vap); 513 ZFS_EXIT(zfsvfs); 514 515 return (0); 516} 517 518/* 519 * Special case the handling of "..". 520 */ 521/* ARGSUSED */ 522int 523zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, 524 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 525 int *direntflags, pathname_t *realpnp) 526{ 527 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 528 int err; 529 530 /* 531 * No extended attributes allowed under .zfs 532 */ 533 if (flags & LOOKUP_XATTR) 534 return (SET_ERROR(EINVAL)); 535 536 ZFS_ENTER(zfsvfs); 537 538 if (strcmp(nm, "..") == 0) { 539 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp); 540 if (err == 0) 541 VOP_UNLOCK(*vpp, 0); 542 } else { 543 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir, 544 cr, ct, direntflags, realpnp); 545 } 546 547 ZFS_EXIT(zfsvfs); 548 549 return (err); 550} 551 552#ifdef sun 553static int 554zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 555 caller_context_t *ct) 556{ 557 /* 558 * We only care about ACL_ENABLED so that libsec can 559 * display ACL correctly and not default to POSIX draft. 560 */ 561 if (cmd == _PC_ACL_ENABLED) { 562 *valp = _ACL_ACE_ENABLED; 563 return (0); 564 } 565 566 return (fs_pathconf(vp, cmd, valp, cr, ct)); 567} 568#endif /* sun */ 569 570#ifdef sun 571static const fs_operation_def_t zfsctl_tops_root[] = { 572 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 573 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 574 { VOPNAME_IOCTL, { .error = fs_inval } }, 575 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } }, 576 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 577 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 578 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } }, 579 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 580 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 581 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } }, 582 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 583 { NULL } 584}; 585#endif /* sun */ 586 587/* 588 * Special case the handling of "..". 589 */ 590/* ARGSUSED */ 591int 592zfsctl_freebsd_root_lookup(ap) 593 struct vop_lookup_args /* { 594 struct vnode *a_dvp; 595 struct vnode **a_vpp; 596 struct componentname *a_cnp; 597 } */ *ap; 598{ 599 vnode_t *dvp = ap->a_dvp; 600 vnode_t **vpp = ap->a_vpp; 601 cred_t *cr = ap->a_cnp->cn_cred; 602 int flags = ap->a_cnp->cn_flags; 603 int nameiop = ap->a_cnp->cn_nameiop; 604 char nm[NAME_MAX + 1]; 605 int err; 606 607 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE)) 608 return (EOPNOTSUPP); 609 610 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 611 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 612 613 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL); 614 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) 615 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 616 return (err); 617} 618 619static struct vop_vector zfsctl_ops_root = { 620 .vop_default = &default_vnodeops, 621 .vop_open = zfsctl_common_open, 622 .vop_close = zfsctl_common_close, 623 .vop_ioctl = VOP_EINVAL, 624 .vop_getattr = zfsctl_root_getattr, 625 .vop_access = zfsctl_common_access, 626 .vop_readdir = gfs_vop_readdir, 627 .vop_lookup = zfsctl_freebsd_root_lookup, 628 .vop_inactive = gfs_vop_inactive, 629 .vop_reclaim = zfsctl_common_reclaim, 630#ifdef TODO 631 .vop_pathconf = zfsctl_pathconf, 632#endif 633 .vop_fid = zfsctl_common_fid, 634}; 635 636/* 637 * Gets the full dataset name that corresponds to the given snapshot name 638 * Example: 639 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1" 640 */ 641static int 642zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 643{ 644 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 645 646 if (snapshot_namecheck(name, NULL, NULL) != 0) 647 return (SET_ERROR(EILSEQ)); 648 dmu_objset_name(os, zname); 649 if (strlen(zname) + 1 + strlen(name) >= len) 650 return (SET_ERROR(ENAMETOOLONG)); 651 (void) strcat(zname, "@"); 652 (void) strcat(zname, name); 653 return (0); 654} 655 656static int 657zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 658{ 659 vnode_t *svp = sep->se_root; 660 int error; 661 662 ASSERT(vn_ismntpt(svp)); 663 664 /* this will be dropped by dounmount() */ 665 if ((error = vn_vfswlock(svp)) != 0) 666 return (error); 667 668#ifdef sun 669 VN_HOLD(svp); 670 error = dounmount(vn_mountedvfs(svp), fflags, cr); 671 if (error) { 672 VN_RELE(svp); 673 return (error); 674 } 675 676 /* 677 * We can't use VN_RELE(), as that will try to invoke 678 * zfsctl_snapdir_inactive(), which would cause us to destroy 679 * the sd_lock mutex held by our caller. 680 */ 681 ASSERT(svp->v_count == 1); 682 gfs_vop_inactive(svp, cr, NULL); 683 684 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 685 kmem_free(sep, sizeof (zfs_snapentry_t)); 686 687 return (0); 688#else /* !sun */ 689 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 690#endif /* !sun */ 691} 692 693#ifdef sun 694static void 695zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 696{ 697 avl_index_t where; 698 vfs_t *vfsp; 699 refstr_t *pathref; 700 char newpath[MAXNAMELEN]; 701 char *tail; 702 703 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 704 ASSERT(sep != NULL); 705 706 vfsp = vn_mountedvfs(sep->se_root); 707 ASSERT(vfsp != NULL); 708 709 vfs_lock_wait(vfsp); 710 711 /* 712 * Change the name in the AVL tree. 713 */ 714 avl_remove(&sdp->sd_snaps, sep); 715 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 716 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 717 (void) strcpy(sep->se_name, nm); 718 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 719 avl_insert(&sdp->sd_snaps, sep, where); 720 721 /* 722 * Change the current mountpoint info: 723 * - update the tail of the mntpoint path 724 * - update the tail of the resource path 725 */ 726 pathref = vfs_getmntpoint(vfsp); 727 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 728 VERIFY((tail = strrchr(newpath, '/')) != NULL); 729 *(tail+1) = '\0'; 730 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 731 (void) strcat(newpath, nm); 732 refstr_rele(pathref); 733 vfs_setmntpoint(vfsp, newpath, 0); 734 735 pathref = vfs_getresource(vfsp); 736 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 737 VERIFY((tail = strrchr(newpath, '@')) != NULL); 738 *(tail+1) = '\0'; 739 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 740 (void) strcat(newpath, nm); 741 refstr_rele(pathref); 742 vfs_setresource(vfsp, newpath, 0); 743 744 vfs_unlock(vfsp); 745} 746#endif /* sun */ 747 748#ifdef sun 749/*ARGSUSED*/ 750static int 751zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 752 cred_t *cr, caller_context_t *ct, int flags) 753{ 754 zfsctl_snapdir_t *sdp = sdvp->v_data; 755 zfs_snapentry_t search, *sep; 756 zfsvfs_t *zfsvfs; 757 avl_index_t where; 758 char from[MAXNAMELEN], to[MAXNAMELEN]; 759 char real[MAXNAMELEN], fsname[MAXNAMELEN]; 760 int err; 761 762 zfsvfs = sdvp->v_vfsp->vfs_data; 763 ZFS_ENTER(zfsvfs); 764 765 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 766 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 767 MAXNAMELEN, NULL); 768 if (err == 0) { 769 snm = real; 770 } else if (err != ENOTSUP) { 771 ZFS_EXIT(zfsvfs); 772 return (err); 773 } 774 } 775 776 ZFS_EXIT(zfsvfs); 777 778 dmu_objset_name(zfsvfs->z_os, fsname); 779 780 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 781 if (err == 0) 782 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 783 if (err == 0) 784 err = zfs_secpolicy_rename_perms(from, to, cr); 785 if (err != 0) 786 return (err); 787 788 /* 789 * Cannot move snapshots out of the snapdir. 790 */ 791 if (sdvp != tdvp) 792 return (SET_ERROR(EINVAL)); 793 794 if (strcmp(snm, tnm) == 0) 795 return (0); 796 797 mutex_enter(&sdp->sd_lock); 798 799 search.se_name = (char *)snm; 800 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 801 mutex_exit(&sdp->sd_lock); 802 return (SET_ERROR(ENOENT)); 803 } 804 805 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0); 806 if (err == 0) 807 zfsctl_rename_snap(sdp, sep, tnm); 808 809 mutex_exit(&sdp->sd_lock); 810 811 return (err); 812} 813#endif /* sun */ 814 815#ifdef sun 816/* ARGSUSED */ 817static int 818zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 819 caller_context_t *ct, int flags) 820{ 821 zfsctl_snapdir_t *sdp = dvp->v_data; 822 zfs_snapentry_t *sep; 823 zfs_snapentry_t search; 824 zfsvfs_t *zfsvfs; 825 char snapname[MAXNAMELEN]; 826 char real[MAXNAMELEN]; 827 int err; 828 829 zfsvfs = dvp->v_vfsp->vfs_data; 830 ZFS_ENTER(zfsvfs); 831 832 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 833 834 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 835 MAXNAMELEN, NULL); 836 if (err == 0) { 837 name = real; 838 } else if (err != ENOTSUP) { 839 ZFS_EXIT(zfsvfs); 840 return (err); 841 } 842 } 843 844 ZFS_EXIT(zfsvfs); 845 846 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 847 if (err == 0) 848 err = zfs_secpolicy_destroy_perms(snapname, cr); 849 if (err != 0) 850 return (err); 851 852 mutex_enter(&sdp->sd_lock); 853 854 search.se_name = name; 855 sep = avl_find(&sdp->sd_snaps, &search, NULL); 856 if (sep) { 857 avl_remove(&sdp->sd_snaps, sep); 858 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 859 if (err != 0) 860 avl_add(&sdp->sd_snaps, sep); 861 else 862 err = dsl_destroy_snapshot(snapname, B_FALSE); 863 } else { 864 err = SET_ERROR(ENOENT); 865 } 866 867 mutex_exit(&sdp->sd_lock); 868 869 return (err); 870} 871#endif /* sun */ 872 873/* 874 * This creates a snapshot under '.zfs/snapshot'. 875 */ 876/* ARGSUSED */ 877static int 878zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 879 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 880{ 881 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 882 char name[MAXNAMELEN]; 883 int err; 884 static enum symfollow follow = NO_FOLLOW; 885 static enum uio_seg seg = UIO_SYSSPACE; 886 887 if (snapshot_namecheck(dirname, NULL, NULL) != 0) 888 return (SET_ERROR(EILSEQ)); 889 890 dmu_objset_name(zfsvfs->z_os, name); 891 892 *vpp = NULL; 893 894 err = zfs_secpolicy_snapshot_perms(name, cr); 895 if (err != 0) 896 return (err); 897 898 if (err == 0) { 899 err = dmu_objset_snapshot_one(name, dirname); 900 if (err != 0) 901 return (err); 902 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 903 } 904 905 return (err); 906} 907 908static int 909zfsctl_freebsd_snapdir_mkdir(ap) 910 struct vop_mkdir_args /* { 911 struct vnode *a_dvp; 912 struct vnode **a_vpp; 913 struct componentname *a_cnp; 914 struct vattr *a_vap; 915 } */ *ap; 916{ 917 918 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 919 920 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 921 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 922} 923 924/* 925 * Lookup entry point for the 'snapshot' directory. Try to open the 926 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 927 * Perform a mount of the associated dataset on top of the vnode. 928 */ 929/* ARGSUSED */ 930int 931zfsctl_snapdir_lookup(ap) 932 struct vop_lookup_args /* { 933 struct vnode *a_dvp; 934 struct vnode **a_vpp; 935 struct componentname *a_cnp; 936 } */ *ap; 937{ 938 vnode_t *dvp = ap->a_dvp; 939 vnode_t **vpp = ap->a_vpp; 940 struct componentname *cnp = ap->a_cnp; 941 char nm[NAME_MAX + 1]; 942 zfsctl_snapdir_t *sdp = dvp->v_data; 943 objset_t *snap; 944 char snapname[MAXNAMELEN]; 945 char real[MAXNAMELEN]; 946 char *mountpoint; 947 zfs_snapentry_t *sep, search; 948 size_t mountpoint_len; 949 avl_index_t where; 950 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 951 int err; 952 int flags = 0; 953 954 /* 955 * No extended attributes allowed under .zfs 956 */ 957 if (flags & LOOKUP_XATTR) 958 return (SET_ERROR(EINVAL)); 959 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 960 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 961 962 ASSERT(dvp->v_type == VDIR); 963 964 *vpp = NULL; 965 966 /* 967 * If we get a recursive call, that means we got called 968 * from the domount() code while it was trying to look up the 969 * spec (which looks like a local path for zfs). We need to 970 * add some flag to domount() to tell it not to do this lookup. 971 */ 972 if (MUTEX_HELD(&sdp->sd_lock)) 973 return (SET_ERROR(ENOENT)); 974 975 ZFS_ENTER(zfsvfs); 976 977 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 978 ZFS_EXIT(zfsvfs); 979 return (0); 980 } 981 982 if (flags & FIGNORECASE) { 983 boolean_t conflict = B_FALSE; 984 985 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 986 MAXNAMELEN, &conflict); 987 if (err == 0) { 988 strlcpy(nm, real, sizeof(nm)); 989 } else if (err != ENOTSUP) { 990 ZFS_EXIT(zfsvfs); 991 return (err); 992 } 993#if 0 994 if (realpnp) 995 (void) strlcpy(realpnp->pn_buf, nm, 996 realpnp->pn_bufsize); 997 if (conflict && direntflags) 998 *direntflags = ED_CASE_CONFLICT; 999#endif 1000 } 1001 1002 mutex_enter(&sdp->sd_lock); 1003 search.se_name = (char *)nm; 1004 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 1005 *vpp = sep->se_root; 1006 VN_HOLD(*vpp); 1007 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 1008 if (err != 0) { 1009 VN_RELE(*vpp); 1010 *vpp = NULL; 1011 } else if (*vpp == sep->se_root) { 1012 /* 1013 * The snapshot was unmounted behind our backs, 1014 * try to remount it. 1015 */ 1016 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1017 goto domount; 1018 } else { 1019 /* 1020 * VROOT was set during the traverse call. We need 1021 * to clear it since we're pretending to be part 1022 * of our parent's vfs. 1023 */ 1024 (*vpp)->v_flag &= ~VROOT; 1025 } 1026 mutex_exit(&sdp->sd_lock); 1027 ZFS_EXIT(zfsvfs); 1028 return (err); 1029 } 1030 1031 /* 1032 * The requested snapshot is not currently mounted, look it up. 1033 */ 1034 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1035 if (err != 0) { 1036 mutex_exit(&sdp->sd_lock); 1037 ZFS_EXIT(zfsvfs); 1038 /* 1039 * handle "ls *" or "?" in a graceful manner, 1040 * forcing EILSEQ to ENOENT. 1041 * Since shell ultimately passes "*" or "?" as name to lookup 1042 */ 1043 return (err == EILSEQ ? ENOENT : err); 1044 } 1045 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1046 mutex_exit(&sdp->sd_lock); 1047#ifdef illumos 1048 ZFS_EXIT(zfsvfs); 1049 return (SET_ERROR(ENOENT)); 1050#else /* !illumos */ 1051 /* Translate errors and add SAVENAME when needed. */ 1052 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1053 err = EJUSTRETURN; 1054 cnp->cn_flags |= SAVENAME; 1055 } else { 1056 err = SET_ERROR(ENOENT); 1057 } 1058 ZFS_EXIT(zfsvfs); 1059 return (err); 1060#endif /* !illumos */ 1061 } 1062 1063 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1064 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1065 (void) strcpy(sep->se_name, nm); 1066 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1067 VN_HOLD(*vpp); 1068 avl_insert(&sdp->sd_snaps, sep, where); 1069 1070 dmu_objset_rele(snap, FTAG); 1071domount: 1072 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1073 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1074 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1075 (void) snprintf(mountpoint, mountpoint_len, 1076 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1077 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1078 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1079 kmem_free(mountpoint, mountpoint_len); 1080 if (err == 0) { 1081 /* 1082 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1083 * 1084 * This is where we lie about our v_vfsp in order to 1085 * make .zfs/snapshot/<snapname> accessible over NFS 1086 * without requiring manual mounts of <snapname>. 1087 */ 1088 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1089 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1090 } 1091 mutex_exit(&sdp->sd_lock); 1092 ZFS_EXIT(zfsvfs); 1093 1094#ifdef illumos 1095 /* 1096 * If we had an error, drop our hold on the vnode and 1097 * zfsctl_snapshot_inactive() will clean up. 1098 */ 1099 if (err != 0) { 1100 VN_RELE(*vpp); 1101 *vpp = NULL; 1102 } 1103#else 1104 if (err != 0) 1105 *vpp = NULL; 1106#endif 1107 return (err); 1108} 1109 1110/* ARGSUSED */ 1111int 1112zfsctl_shares_lookup(ap) 1113 struct vop_lookup_args /* { 1114 struct vnode *a_dvp; 1115 struct vnode **a_vpp; 1116 struct componentname *a_cnp; 1117 } */ *ap; 1118{ 1119 vnode_t *dvp = ap->a_dvp; 1120 vnode_t **vpp = ap->a_vpp; 1121 struct componentname *cnp = ap->a_cnp; 1122 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1123 char nm[NAME_MAX + 1]; 1124 znode_t *dzp; 1125 int error; 1126 1127 ZFS_ENTER(zfsvfs); 1128 1129 ASSERT(cnp->cn_namelen < sizeof(nm)); 1130 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1131 1132 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1133 ZFS_EXIT(zfsvfs); 1134 return (0); 1135 } 1136 1137 if (zfsvfs->z_shares_dir == 0) { 1138 ZFS_EXIT(zfsvfs); 1139 return (SET_ERROR(ENOTSUP)); 1140 } 1141 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) 1142 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1143 1144 VN_RELE(ZTOV(dzp)); 1145 ZFS_EXIT(zfsvfs); 1146 1147 return (error); 1148} 1149 1150/* ARGSUSED */ 1151static int 1152zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1153 offset_t *offp, offset_t *nextp, void *data, int flags) 1154{ 1155 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1156 char snapname[MAXNAMELEN]; 1157 uint64_t id, cookie; 1158 boolean_t case_conflict; 1159 int error; 1160 1161 ZFS_ENTER(zfsvfs); 1162 1163 cookie = *offp; 1164 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1165 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1166 &cookie, &case_conflict); 1167 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1168 if (error) { 1169 ZFS_EXIT(zfsvfs); 1170 if (error == ENOENT) { 1171 *eofp = 1; 1172 return (0); 1173 } 1174 return (error); 1175 } 1176 1177 if (flags & V_RDDIR_ENTFLAGS) { 1178 edirent_t *eodp = dp; 1179 1180 (void) strcpy(eodp->ed_name, snapname); 1181 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1182 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1183 } else { 1184 struct dirent64 *odp = dp; 1185 1186 (void) strcpy(odp->d_name, snapname); 1187 odp->d_ino = ZFSCTL_INO_SNAP(id); 1188 } 1189 *nextp = cookie; 1190 1191 ZFS_EXIT(zfsvfs); 1192 1193 return (0); 1194} 1195 1196/* ARGSUSED */ 1197static int 1198zfsctl_shares_readdir(ap) 1199 struct vop_readdir_args /* { 1200 struct vnode *a_vp; 1201 struct uio *a_uio; 1202 struct ucred *a_cred; 1203 int *a_eofflag; 1204 int *a_ncookies; 1205 u_long **a_cookies; 1206 } */ *ap; 1207{ 1208 vnode_t *vp = ap->a_vp; 1209 uio_t *uiop = ap->a_uio; 1210 cred_t *cr = ap->a_cred; 1211 int *eofp = ap->a_eofflag; 1212 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1213 znode_t *dzp; 1214 int error; 1215 1216 ZFS_ENTER(zfsvfs); 1217 1218 if (zfsvfs->z_shares_dir == 0) { 1219 ZFS_EXIT(zfsvfs); 1220 return (SET_ERROR(ENOTSUP)); 1221 } 1222 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1223 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1224 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1225 VN_URELE(ZTOV(dzp)); 1226 } else { 1227 *eofp = 1; 1228 error = SET_ERROR(ENOENT); 1229 } 1230 1231 ZFS_EXIT(zfsvfs); 1232 return (error); 1233} 1234 1235/* 1236 * pvp is the '.zfs' directory (zfsctl_node_t). 1237 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1238 * 1239 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1240 * when a lookup is performed on .zfs for "snapshot". 1241 */ 1242vnode_t * 1243zfsctl_mknode_snapdir(vnode_t *pvp) 1244{ 1245 vnode_t *vp; 1246 zfsctl_snapdir_t *sdp; 1247 1248 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1249 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1250 zfsctl_snapdir_readdir_cb, NULL); 1251 sdp = vp->v_data; 1252 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1253 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1254 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1255 avl_create(&sdp->sd_snaps, snapentry_compare, 1256 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1257 VOP_UNLOCK(vp, 0); 1258 return (vp); 1259} 1260 1261vnode_t * 1262zfsctl_mknode_shares(vnode_t *pvp) 1263{ 1264 vnode_t *vp; 1265 zfsctl_node_t *sdp; 1266 1267 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1268 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1269 NULL, NULL); 1270 sdp = vp->v_data; 1271 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1272 VOP_UNLOCK(vp, 0); 1273 return (vp); 1274 1275} 1276 1277/* ARGSUSED */ 1278static int 1279zfsctl_shares_getattr(ap) 1280 struct vop_getattr_args /* { 1281 struct vnode *a_vp; 1282 struct vattr *a_vap; 1283 struct ucred *a_cred; 1284 struct thread *a_td; 1285 } */ *ap; 1286{ 1287 vnode_t *vp = ap->a_vp; 1288 vattr_t *vap = ap->a_vap; 1289 cred_t *cr = ap->a_cred; 1290 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1291 znode_t *dzp; 1292 int error; 1293 1294 ZFS_ENTER(zfsvfs); 1295 if (zfsvfs->z_shares_dir == 0) { 1296 ZFS_EXIT(zfsvfs); 1297 return (SET_ERROR(ENOTSUP)); 1298 } 1299 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1300 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1301 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1302 VN_URELE(ZTOV(dzp)); 1303 } 1304 ZFS_EXIT(zfsvfs); 1305 return (error); 1306 1307 1308} 1309 1310/* ARGSUSED */ 1311static int 1312zfsctl_snapdir_getattr(ap) 1313 struct vop_getattr_args /* { 1314 struct vnode *a_vp; 1315 struct vattr *a_vap; 1316 struct ucred *a_cred; 1317 } */ *ap; 1318{ 1319 vnode_t *vp = ap->a_vp; 1320 vattr_t *vap = ap->a_vap; 1321 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1322 zfsctl_snapdir_t *sdp = vp->v_data; 1323 1324 ZFS_ENTER(zfsvfs); 1325 zfsctl_common_getattr(vp, vap); 1326 vap->va_nodeid = gfs_file_inode(vp); 1327 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1328 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1329 vap->va_birthtime = vap->va_ctime; 1330 ZFS_EXIT(zfsvfs); 1331 1332 return (0); 1333} 1334 1335/* ARGSUSED */ 1336static int 1337zfsctl_snapdir_inactive(ap) 1338 struct vop_inactive_args /* { 1339 struct vnode *a_vp; 1340 struct thread *a_td; 1341 } */ *ap; 1342{ 1343 vnode_t *vp = ap->a_vp; 1344 zfsctl_snapdir_t *sdp = vp->v_data; 1345 zfs_snapentry_t *sep; 1346 1347 /* 1348 * On forced unmount we have to free snapshots from here. 1349 */ 1350 mutex_enter(&sdp->sd_lock); 1351 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) { 1352 avl_remove(&sdp->sd_snaps, sep); 1353 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1354 kmem_free(sep, sizeof (zfs_snapentry_t)); 1355 } 1356 mutex_exit(&sdp->sd_lock); 1357 gfs_dir_inactive(vp); 1358 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1359 mutex_destroy(&sdp->sd_lock); 1360 avl_destroy(&sdp->sd_snaps); 1361 kmem_free(sdp, sizeof (zfsctl_snapdir_t)); 1362 1363 return (0); 1364} 1365 1366#ifdef sun 1367static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1368 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1369 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1370 { VOPNAME_IOCTL, { .error = fs_inval } }, 1371 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1372 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1373 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1374 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1375 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1376 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1377 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1378 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1379 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1380 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1381 { NULL } 1382}; 1383 1384static const fs_operation_def_t zfsctl_tops_shares[] = { 1385 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1386 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1387 { VOPNAME_IOCTL, { .error = fs_inval } }, 1388 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1389 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1390 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1391 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1392 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1393 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1394 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1395 { NULL } 1396}; 1397#else /* !sun */ 1398static struct vop_vector zfsctl_ops_snapdir = { 1399 .vop_default = &default_vnodeops, 1400 .vop_open = zfsctl_common_open, 1401 .vop_close = zfsctl_common_close, 1402 .vop_ioctl = VOP_EINVAL, 1403 .vop_getattr = zfsctl_snapdir_getattr, 1404 .vop_access = zfsctl_common_access, 1405 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1406 .vop_readdir = gfs_vop_readdir, 1407 .vop_lookup = zfsctl_snapdir_lookup, 1408 .vop_inactive = zfsctl_snapdir_inactive, 1409 .vop_reclaim = zfsctl_common_reclaim, 1410 .vop_fid = zfsctl_common_fid, 1411}; 1412 1413static struct vop_vector zfsctl_ops_shares = { 1414 .vop_default = &default_vnodeops, 1415 .vop_open = zfsctl_common_open, 1416 .vop_close = zfsctl_common_close, 1417 .vop_ioctl = VOP_EINVAL, 1418 .vop_getattr = zfsctl_shares_getattr, 1419 .vop_access = zfsctl_common_access, 1420 .vop_readdir = zfsctl_shares_readdir, 1421 .vop_lookup = zfsctl_shares_lookup, 1422 .vop_inactive = gfs_vop_inactive, 1423 .vop_reclaim = zfsctl_common_reclaim, 1424 .vop_fid = zfsctl_shares_fid, 1425}; 1426#endif /* !sun */ 1427 1428/* 1429 * pvp is the GFS vnode '.zfs/snapshot'. 1430 * 1431 * This creates a GFS node under '.zfs/snapshot' representing each 1432 * snapshot. This newly created GFS node is what we mount snapshot 1433 * vfs_t's ontop of. 1434 */ 1435static vnode_t * 1436zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1437{ 1438 vnode_t *vp; 1439 zfsctl_node_t *zcp; 1440 1441 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1442 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1443 VN_HOLD(vp); 1444 zcp = vp->v_data; 1445 zcp->zc_id = objset; 1446 VOP_UNLOCK(vp, 0); 1447 1448 return (vp); 1449} 1450 1451static int 1452zfsctl_snapshot_inactive(ap) 1453 struct vop_inactive_args /* { 1454 struct vnode *a_vp; 1455 struct thread *a_td; 1456 } */ *ap; 1457{ 1458 vnode_t *vp = ap->a_vp; 1459 cred_t *cr = ap->a_td->td_ucred; 1460 struct vop_inactive_args iap; 1461 zfsctl_snapdir_t *sdp; 1462 zfs_snapentry_t *sep, *next; 1463 int locked; 1464 vnode_t *dvp; 1465 1466 if (vp->v_count > 0) 1467 goto end; 1468 1469 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1470 sdp = dvp->v_data; 1471 VOP_UNLOCK(dvp, 0); 1472 1473 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1474 mutex_enter(&sdp->sd_lock); 1475 1476 ASSERT(!vn_ismntpt(vp)); 1477 1478 sep = avl_first(&sdp->sd_snaps); 1479 while (sep != NULL) { 1480 next = AVL_NEXT(&sdp->sd_snaps, sep); 1481 1482 if (sep->se_root == vp) { 1483 avl_remove(&sdp->sd_snaps, sep); 1484 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1485 kmem_free(sep, sizeof (zfs_snapentry_t)); 1486 break; 1487 } 1488 sep = next; 1489 } 1490 ASSERT(sep != NULL); 1491 1492 if (!locked) 1493 mutex_exit(&sdp->sd_lock); 1494 VN_RELE(dvp); 1495 1496end: 1497 /* 1498 * Dispose of the vnode for the snapshot mount point. 1499 * This is safe to do because once this entry has been removed 1500 * from the AVL tree, it can't be found again, so cannot become 1501 * "active". If we lookup the same name again we will end up 1502 * creating a new vnode. 1503 */ 1504 iap.a_vp = vp; 1505 return (gfs_vop_inactive(&iap)); 1506} 1507 1508static int 1509zfsctl_traverse_begin(vnode_t **vpp, int lktype) 1510{ 1511 1512 VN_HOLD(*vpp); 1513 /* Snapshot should be already mounted, but just in case. */ 1514 if (vn_mountedvfs(*vpp) == NULL) 1515 return (ENOENT); 1516 return (traverse(vpp, lktype)); 1517} 1518 1519static void 1520zfsctl_traverse_end(vnode_t *vp, int err) 1521{ 1522 1523 if (err == 0) 1524 vput(vp); 1525 else 1526 VN_RELE(vp); 1527} 1528 1529static int 1530zfsctl_snapshot_getattr(ap) 1531 struct vop_getattr_args /* { 1532 struct vnode *a_vp; 1533 struct vattr *a_vap; 1534 struct ucred *a_cred; 1535 } */ *ap; 1536{ 1537 vnode_t *vp = ap->a_vp; 1538 int err; 1539 1540 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1541 if (err == 0) 1542 err = VOP_GETATTR(vp, ap->a_vap, ap->a_cred); 1543 zfsctl_traverse_end(vp, err); 1544 return (err); 1545} 1546 1547static int 1548zfsctl_snapshot_fid(ap) 1549 struct vop_fid_args /* { 1550 struct vnode *a_vp; 1551 struct fid *a_fid; 1552 } */ *ap; 1553{ 1554 vnode_t *vp = ap->a_vp; 1555 int err; 1556 1557 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1558 if (err == 0) 1559 err = VOP_VPTOFH(vp, (void *)ap->a_fid); 1560 zfsctl_traverse_end(vp, err); 1561 return (err); 1562} 1563 1564static int 1565zfsctl_snapshot_lookup(ap) 1566 struct vop_lookup_args /* { 1567 struct vnode *a_dvp; 1568 struct vnode **a_vpp; 1569 struct componentname *a_cnp; 1570 } */ *ap; 1571{ 1572 vnode_t *dvp = ap->a_dvp; 1573 vnode_t **vpp = ap->a_vpp; 1574 struct componentname *cnp = ap->a_cnp; 1575 cred_t *cr = ap->a_cnp->cn_cred; 1576 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1577 int error; 1578 1579 if (cnp->cn_namelen != 2 || cnp->cn_nameptr[0] != '.' || 1580 cnp->cn_nameptr[1] != '.') { 1581 return (ENOENT); 1582 } 1583 1584 ASSERT(dvp->v_type == VDIR); 1585 ASSERT(zfsvfs->z_ctldir != NULL); 1586 1587 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", vpp, 1588 NULL, 0, NULL, cr, NULL, NULL, NULL); 1589 if (error == 0) 1590 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1591 return (error); 1592} 1593 1594static int 1595zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1596{ 1597 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1598 vnode_t *dvp, *vp; 1599 zfsctl_snapdir_t *sdp; 1600 zfs_snapentry_t *sep; 1601 int error; 1602 1603 ASSERT(zfsvfs->z_ctldir != NULL); 1604 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1605 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1606 if (error != 0) 1607 return (error); 1608 sdp = dvp->v_data; 1609 1610 mutex_enter(&sdp->sd_lock); 1611 sep = avl_first(&sdp->sd_snaps); 1612 while (sep != NULL) { 1613 vp = sep->se_root; 1614 if (vp == ap->a_vp) 1615 break; 1616 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1617 } 1618 if (sep == NULL) { 1619 mutex_exit(&sdp->sd_lock); 1620 error = ENOENT; 1621 } else { 1622 size_t len; 1623 1624 len = strlen(sep->se_name); 1625 *ap->a_buflen -= len; 1626 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1627 mutex_exit(&sdp->sd_lock); 1628 vref(dvp); 1629 *ap->a_vpp = dvp; 1630 } 1631 VN_RELE(dvp); 1632 1633 return (error); 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 = zfsctl_snapshot_inactive, 1643 .vop_lookup = zfsctl_snapshot_lookup, 1644 .vop_reclaim = zfsctl_common_reclaim, 1645 .vop_getattr = zfsctl_snapshot_getattr, 1646 .vop_fid = zfsctl_snapshot_fid, 1647 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1648}; 1649 1650int 1651zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1652{ 1653 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1654 vnode_t *dvp, *vp; 1655 zfsctl_snapdir_t *sdp; 1656 zfsctl_node_t *zcp; 1657 zfs_snapentry_t *sep; 1658 int error; 1659 1660 ASSERT(zfsvfs->z_ctldir != NULL); 1661 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1662 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1663 if (error != 0) 1664 return (error); 1665 sdp = dvp->v_data; 1666 1667 mutex_enter(&sdp->sd_lock); 1668 sep = avl_first(&sdp->sd_snaps); 1669 while (sep != NULL) { 1670 vp = sep->se_root; 1671 zcp = vp->v_data; 1672 if (zcp->zc_id == objsetid) 1673 break; 1674 1675 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1676 } 1677 1678 if (sep != NULL) { 1679 VN_HOLD(vp); 1680 /* 1681 * Return the mounted root rather than the covered mount point. 1682 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1683 * and returns the ZFS vnode mounted on top of the GFS node. 1684 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1685 */ 1686 error = traverse(&vp, LK_SHARED | LK_RETRY); 1687 if (error == 0) { 1688 if (vp == sep->se_root) 1689 error = SET_ERROR(EINVAL); 1690 else 1691 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1692 } 1693 mutex_exit(&sdp->sd_lock); 1694 if (error == 0) 1695 VN_URELE(vp); 1696 else 1697 VN_RELE(vp); 1698 } else { 1699 error = SET_ERROR(EINVAL); 1700 mutex_exit(&sdp->sd_lock); 1701 } 1702 1703 VN_RELE(dvp); 1704 1705 return (error); 1706} 1707 1708/* 1709 * Unmount any snapshots for the given filesystem. This is called from 1710 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1711 * snapshots. 1712 */ 1713int 1714zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1715{ 1716 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1717 vnode_t *dvp; 1718 zfsctl_snapdir_t *sdp; 1719 zfs_snapentry_t *sep, *next; 1720 int error; 1721 1722 ASSERT(zfsvfs->z_ctldir != NULL); 1723 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1724 NULL, 0, NULL, cr, NULL, NULL, NULL); 1725 if (error != 0) 1726 return (error); 1727 sdp = dvp->v_data; 1728 1729 mutex_enter(&sdp->sd_lock); 1730 1731 sep = avl_first(&sdp->sd_snaps); 1732 while (sep != NULL) { 1733 next = AVL_NEXT(&sdp->sd_snaps, sep); 1734 1735 /* 1736 * If this snapshot is not mounted, then it must 1737 * have just been unmounted by somebody else, and 1738 * will be cleaned up by zfsctl_snapdir_inactive(). 1739 */ 1740 if (vn_ismntpt(sep->se_root)) { 1741 error = zfsctl_unmount_snap(sep, fflags, cr); 1742 if (error) { 1743 avl_index_t where; 1744 1745 /* 1746 * Before reinserting snapshot to the tree, 1747 * check if it was actually removed. For example 1748 * when snapshot mount point is busy, we will 1749 * have an error here, but there will be no need 1750 * to reinsert snapshot. 1751 */ 1752 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1753 avl_insert(&sdp->sd_snaps, sep, where); 1754 break; 1755 } 1756 } 1757 sep = next; 1758 } 1759 1760 mutex_exit(&sdp->sd_lock); 1761 VN_RELE(dvp); 1762 1763 return (error); 1764} 1765