zfs_ctldir.c revision 274326
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 int ltype; 607 608 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE)) 609 return (EOPNOTSUPP); 610 611 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 612 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 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 ltype = VOP_ISLOCKED(dvp); 616 if (flags & ISDOTDOT) { 617 VN_HOLD(*vpp); 618 VOP_UNLOCK(dvp, 0); 619 } 620 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 621 if (flags & ISDOTDOT) { 622 VN_RELE(*vpp); 623 vn_lock(dvp, ltype| LK_RETRY); 624 } 625 } 626 627 return (err); 628} 629 630static struct vop_vector zfsctl_ops_root = { 631 .vop_default = &default_vnodeops, 632 .vop_open = zfsctl_common_open, 633 .vop_close = zfsctl_common_close, 634 .vop_ioctl = VOP_EINVAL, 635 .vop_getattr = zfsctl_root_getattr, 636 .vop_access = zfsctl_common_access, 637 .vop_readdir = gfs_vop_readdir, 638 .vop_lookup = zfsctl_freebsd_root_lookup, 639 .vop_inactive = VOP_NULL, 640 .vop_reclaim = gfs_vop_reclaim, 641#ifdef TODO 642 .vop_pathconf = zfsctl_pathconf, 643#endif 644 .vop_fid = zfsctl_common_fid, 645}; 646 647/* 648 * Gets the full dataset name that corresponds to the given snapshot name 649 * Example: 650 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1" 651 */ 652static int 653zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 654{ 655 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 656 657 if (zfs_component_namecheck(name, NULL, NULL) != 0) 658 return (SET_ERROR(EILSEQ)); 659 dmu_objset_name(os, zname); 660 if (strlen(zname) + 1 + strlen(name) >= len) 661 return (SET_ERROR(ENAMETOOLONG)); 662 (void) strcat(zname, "@"); 663 (void) strcat(zname, name); 664 return (0); 665} 666 667static int 668zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 669{ 670 vnode_t *svp = sep->se_root; 671 int error; 672 673 ASSERT(vn_ismntpt(svp)); 674 675 /* this will be dropped by dounmount() */ 676 if ((error = vn_vfswlock(svp)) != 0) 677 return (error); 678 679#ifdef sun 680 VN_HOLD(svp); 681 error = dounmount(vn_mountedvfs(svp), fflags, cr); 682 if (error) { 683 VN_RELE(svp); 684 return (error); 685 } 686 687 /* 688 * We can't use VN_RELE(), as that will try to invoke 689 * zfsctl_snapdir_inactive(), which would cause us to destroy 690 * the sd_lock mutex held by our caller. 691 */ 692 ASSERT(svp->v_count == 1); 693 gfs_vop_reclaim(svp, cr, NULL); 694 695 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 696 kmem_free(sep, sizeof (zfs_snapentry_t)); 697 698 return (0); 699#else /* !sun */ 700 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 701#endif /* !sun */ 702} 703 704#ifdef sun 705static void 706zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 707{ 708 avl_index_t where; 709 vfs_t *vfsp; 710 refstr_t *pathref; 711 char newpath[MAXNAMELEN]; 712 char *tail; 713 714 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 715 ASSERT(sep != NULL); 716 717 vfsp = vn_mountedvfs(sep->se_root); 718 ASSERT(vfsp != NULL); 719 720 vfs_lock_wait(vfsp); 721 722 /* 723 * Change the name in the AVL tree. 724 */ 725 avl_remove(&sdp->sd_snaps, sep); 726 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 727 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 728 (void) strcpy(sep->se_name, nm); 729 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 730 avl_insert(&sdp->sd_snaps, sep, where); 731 732 /* 733 * Change the current mountpoint info: 734 * - update the tail of the mntpoint path 735 * - update the tail of the resource path 736 */ 737 pathref = vfs_getmntpoint(vfsp); 738 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 739 VERIFY((tail = strrchr(newpath, '/')) != NULL); 740 *(tail+1) = '\0'; 741 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 742 (void) strcat(newpath, nm); 743 refstr_rele(pathref); 744 vfs_setmntpoint(vfsp, newpath, 0); 745 746 pathref = vfs_getresource(vfsp); 747 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 748 VERIFY((tail = strrchr(newpath, '@')) != NULL); 749 *(tail+1) = '\0'; 750 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 751 (void) strcat(newpath, nm); 752 refstr_rele(pathref); 753 vfs_setresource(vfsp, newpath, 0); 754 755 vfs_unlock(vfsp); 756} 757#endif /* sun */ 758 759#ifdef sun 760/*ARGSUSED*/ 761static int 762zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 763 cred_t *cr, caller_context_t *ct, int flags) 764{ 765 zfsctl_snapdir_t *sdp = sdvp->v_data; 766 zfs_snapentry_t search, *sep; 767 zfsvfs_t *zfsvfs; 768 avl_index_t where; 769 char from[MAXNAMELEN], to[MAXNAMELEN]; 770 char real[MAXNAMELEN], fsname[MAXNAMELEN]; 771 int err; 772 773 zfsvfs = sdvp->v_vfsp->vfs_data; 774 ZFS_ENTER(zfsvfs); 775 776 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 777 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 778 MAXNAMELEN, NULL); 779 if (err == 0) { 780 snm = real; 781 } else if (err != ENOTSUP) { 782 ZFS_EXIT(zfsvfs); 783 return (err); 784 } 785 } 786 787 ZFS_EXIT(zfsvfs); 788 789 dmu_objset_name(zfsvfs->z_os, fsname); 790 791 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 792 if (err == 0) 793 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 794 if (err == 0) 795 err = zfs_secpolicy_rename_perms(from, to, cr); 796 if (err != 0) 797 return (err); 798 799 /* 800 * Cannot move snapshots out of the snapdir. 801 */ 802 if (sdvp != tdvp) 803 return (SET_ERROR(EINVAL)); 804 805 if (strcmp(snm, tnm) == 0) 806 return (0); 807 808 mutex_enter(&sdp->sd_lock); 809 810 search.se_name = (char *)snm; 811 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 812 mutex_exit(&sdp->sd_lock); 813 return (SET_ERROR(ENOENT)); 814 } 815 816 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0); 817 if (err == 0) 818 zfsctl_rename_snap(sdp, sep, tnm); 819 820 mutex_exit(&sdp->sd_lock); 821 822 return (err); 823} 824#endif /* sun */ 825 826#ifdef sun 827/* ARGSUSED */ 828static int 829zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 830 caller_context_t *ct, int flags) 831{ 832 zfsctl_snapdir_t *sdp = dvp->v_data; 833 zfs_snapentry_t *sep; 834 zfs_snapentry_t search; 835 zfsvfs_t *zfsvfs; 836 char snapname[MAXNAMELEN]; 837 char real[MAXNAMELEN]; 838 int err; 839 840 zfsvfs = dvp->v_vfsp->vfs_data; 841 ZFS_ENTER(zfsvfs); 842 843 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 844 845 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 846 MAXNAMELEN, NULL); 847 if (err == 0) { 848 name = real; 849 } else if (err != ENOTSUP) { 850 ZFS_EXIT(zfsvfs); 851 return (err); 852 } 853 } 854 855 ZFS_EXIT(zfsvfs); 856 857 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 858 if (err == 0) 859 err = zfs_secpolicy_destroy_perms(snapname, cr); 860 if (err != 0) 861 return (err); 862 863 mutex_enter(&sdp->sd_lock); 864 865 search.se_name = name; 866 sep = avl_find(&sdp->sd_snaps, &search, NULL); 867 if (sep) { 868 avl_remove(&sdp->sd_snaps, sep); 869 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 870 if (err != 0) 871 avl_add(&sdp->sd_snaps, sep); 872 else 873 err = dsl_destroy_snapshot(snapname, B_FALSE); 874 } else { 875 err = SET_ERROR(ENOENT); 876 } 877 878 mutex_exit(&sdp->sd_lock); 879 880 return (err); 881} 882#endif /* sun */ 883 884/* 885 * This creates a snapshot under '.zfs/snapshot'. 886 */ 887/* ARGSUSED */ 888static int 889zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 890 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 891{ 892 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 893 char name[MAXNAMELEN]; 894 int err; 895 static enum symfollow follow = NO_FOLLOW; 896 static enum uio_seg seg = UIO_SYSSPACE; 897 898 if (zfs_component_namecheck(dirname, NULL, NULL) != 0) 899 return (SET_ERROR(EILSEQ)); 900 901 dmu_objset_name(zfsvfs->z_os, name); 902 903 *vpp = NULL; 904 905 err = zfs_secpolicy_snapshot_perms(name, cr); 906 if (err != 0) 907 return (err); 908 909 if (err == 0) { 910 err = dmu_objset_snapshot_one(name, dirname); 911 if (err != 0) 912 return (err); 913 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 914 } 915 916 return (err); 917} 918 919static int 920zfsctl_freebsd_snapdir_mkdir(ap) 921 struct vop_mkdir_args /* { 922 struct vnode *a_dvp; 923 struct vnode **a_vpp; 924 struct componentname *a_cnp; 925 struct vattr *a_vap; 926 } */ *ap; 927{ 928 929 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 930 931 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 932 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 933} 934 935/* 936 * Lookup entry point for the 'snapshot' directory. Try to open the 937 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 938 * Perform a mount of the associated dataset on top of the vnode. 939 */ 940/* ARGSUSED */ 941int 942zfsctl_snapdir_lookup(ap) 943 struct vop_lookup_args /* { 944 struct vnode *a_dvp; 945 struct vnode **a_vpp; 946 struct componentname *a_cnp; 947 } */ *ap; 948{ 949 vnode_t *dvp = ap->a_dvp; 950 vnode_t **vpp = ap->a_vpp; 951 struct componentname *cnp = ap->a_cnp; 952 char nm[NAME_MAX + 1]; 953 zfsctl_snapdir_t *sdp = dvp->v_data; 954 objset_t *snap; 955 char snapname[MAXNAMELEN]; 956 char real[MAXNAMELEN]; 957 char *mountpoint; 958 zfs_snapentry_t *sep, search; 959 size_t mountpoint_len; 960 avl_index_t where; 961 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 962 int err; 963 int ltype, flags = 0; 964 965 /* 966 * No extended attributes allowed under .zfs 967 */ 968 if (flags & LOOKUP_XATTR) 969 return (SET_ERROR(EINVAL)); 970 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 971 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 972 973 ASSERT(dvp->v_type == VDIR); 974 975 *vpp = NULL; 976 977 /* 978 * If we get a recursive call, that means we got called 979 * from the domount() code while it was trying to look up the 980 * spec (which looks like a local path for zfs). We need to 981 * add some flag to domount() to tell it not to do this lookup. 982 */ 983 if (MUTEX_HELD(&sdp->sd_lock)) 984 return (SET_ERROR(ENOENT)); 985 986 ZFS_ENTER(zfsvfs); 987 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 988 ZFS_EXIT(zfsvfs); 989 return (0); 990 } 991 992 if (flags & FIGNORECASE) { 993 boolean_t conflict = B_FALSE; 994 995 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 996 MAXNAMELEN, &conflict); 997 if (err == 0) { 998 strlcpy(nm, real, sizeof(nm)); 999 } else if (err != ENOTSUP) { 1000 ZFS_EXIT(zfsvfs); 1001 return (err); 1002 } 1003#if 0 1004 if (realpnp) 1005 (void) strlcpy(realpnp->pn_buf, nm, 1006 realpnp->pn_bufsize); 1007 if (conflict && direntflags) 1008 *direntflags = ED_CASE_CONFLICT; 1009#endif 1010 } 1011 1012 mutex_enter(&sdp->sd_lock); 1013 search.se_name = (char *)nm; 1014 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 1015 *vpp = sep->se_root; 1016 VN_HOLD(*vpp); 1017 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 1018 if (err != 0) { 1019 VN_RELE(*vpp); 1020 *vpp = NULL; 1021 } else if (*vpp == sep->se_root) { 1022 /* 1023 * The snapshot was unmounted behind our backs, 1024 * try to remount it. 1025 */ 1026 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1027 goto domount; 1028 } else { 1029 /* 1030 * VROOT was set during the traverse call. We need 1031 * to clear it since we're pretending to be part 1032 * of our parent's vfs. 1033 */ 1034 (*vpp)->v_flag &= ~VROOT; 1035 } 1036 mutex_exit(&sdp->sd_lock); 1037 ZFS_EXIT(zfsvfs); 1038 return (err); 1039 } 1040 1041 /* 1042 * The requested snapshot is not currently mounted, look it up. 1043 */ 1044 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1045 if (err != 0) { 1046 mutex_exit(&sdp->sd_lock); 1047 ZFS_EXIT(zfsvfs); 1048 /* 1049 * handle "ls *" or "?" in a graceful manner, 1050 * forcing EILSEQ to ENOENT. 1051 * Since shell ultimately passes "*" or "?" as name to lookup 1052 */ 1053 return (err == EILSEQ ? ENOENT : err); 1054 } 1055 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1056 mutex_exit(&sdp->sd_lock); 1057#ifdef illumos 1058 ZFS_EXIT(zfsvfs); 1059 return (SET_ERROR(ENOENT)); 1060#else /* !illumos */ 1061 /* Translate errors and add SAVENAME when needed. */ 1062 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1063 err = EJUSTRETURN; 1064 cnp->cn_flags |= SAVENAME; 1065 } else { 1066 err = SET_ERROR(ENOENT); 1067 } 1068 ZFS_EXIT(zfsvfs); 1069 return (err); 1070#endif /* !illumos */ 1071 } 1072 1073 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1074 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1075 (void) strcpy(sep->se_name, nm); 1076 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1077 VN_HOLD(*vpp); 1078 avl_insert(&sdp->sd_snaps, sep, where); 1079 1080 dmu_objset_rele(snap, FTAG); 1081domount: 1082 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1083 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1084 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1085 (void) snprintf(mountpoint, mountpoint_len, 1086 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1087 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1088 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1089 kmem_free(mountpoint, mountpoint_len); 1090 if (err == 0) { 1091 /* 1092 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1093 * 1094 * This is where we lie about our v_vfsp in order to 1095 * make .zfs/snapshot/<snapname> accessible over NFS 1096 * without requiring manual mounts of <snapname>. 1097 */ 1098 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1099 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1100 } 1101 mutex_exit(&sdp->sd_lock); 1102 ZFS_EXIT(zfsvfs); 1103 1104#ifdef illumos 1105 /* 1106 * If we had an error, drop our hold on the vnode and 1107 * zfsctl_snapshot_inactive() will clean up. 1108 */ 1109 if (err != 0) { 1110 VN_RELE(*vpp); 1111 *vpp = NULL; 1112 } 1113#else 1114 if (err != 0) 1115 *vpp = NULL; 1116#endif 1117 return (err); 1118} 1119 1120/* ARGSUSED */ 1121int 1122zfsctl_shares_lookup(ap) 1123 struct vop_lookup_args /* { 1124 struct vnode *a_dvp; 1125 struct vnode **a_vpp; 1126 struct componentname *a_cnp; 1127 } */ *ap; 1128{ 1129 vnode_t *dvp = ap->a_dvp; 1130 vnode_t **vpp = ap->a_vpp; 1131 struct componentname *cnp = ap->a_cnp; 1132 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1133 char nm[NAME_MAX + 1]; 1134 znode_t *dzp; 1135 int error; 1136 1137 ZFS_ENTER(zfsvfs); 1138 1139 ASSERT(cnp->cn_namelen < sizeof(nm)); 1140 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1141 1142 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1143 ZFS_EXIT(zfsvfs); 1144 return (0); 1145 } 1146 1147 if (zfsvfs->z_shares_dir == 0) { 1148 ZFS_EXIT(zfsvfs); 1149 return (SET_ERROR(ENOTSUP)); 1150 } 1151 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) 1152 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1153 1154 VN_RELE(ZTOV(dzp)); 1155 ZFS_EXIT(zfsvfs); 1156 1157 return (error); 1158} 1159 1160/* ARGSUSED */ 1161static int 1162zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1163 offset_t *offp, offset_t *nextp, void *data, int flags) 1164{ 1165 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1166 char snapname[MAXNAMELEN]; 1167 uint64_t id, cookie; 1168 boolean_t case_conflict; 1169 int error; 1170 1171 ZFS_ENTER(zfsvfs); 1172 1173 cookie = *offp; 1174 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1175 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1176 &cookie, &case_conflict); 1177 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1178 if (error) { 1179 ZFS_EXIT(zfsvfs); 1180 if (error == ENOENT) { 1181 *eofp = 1; 1182 return (0); 1183 } 1184 return (error); 1185 } 1186 1187 if (flags & V_RDDIR_ENTFLAGS) { 1188 edirent_t *eodp = dp; 1189 1190 (void) strcpy(eodp->ed_name, snapname); 1191 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1192 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1193 } else { 1194 struct dirent64 *odp = dp; 1195 1196 (void) strcpy(odp->d_name, snapname); 1197 odp->d_ino = ZFSCTL_INO_SNAP(id); 1198 } 1199 *nextp = cookie; 1200 1201 ZFS_EXIT(zfsvfs); 1202 1203 return (0); 1204} 1205 1206/* ARGSUSED */ 1207static int 1208zfsctl_shares_readdir(ap) 1209 struct vop_readdir_args /* { 1210 struct vnode *a_vp; 1211 struct uio *a_uio; 1212 struct ucred *a_cred; 1213 int *a_eofflag; 1214 int *a_ncookies; 1215 u_long **a_cookies; 1216 } */ *ap; 1217{ 1218 vnode_t *vp = ap->a_vp; 1219 uio_t *uiop = ap->a_uio; 1220 cred_t *cr = ap->a_cred; 1221 int *eofp = ap->a_eofflag; 1222 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1223 znode_t *dzp; 1224 int error; 1225 1226 ZFS_ENTER(zfsvfs); 1227 1228 if (zfsvfs->z_shares_dir == 0) { 1229 ZFS_EXIT(zfsvfs); 1230 return (SET_ERROR(ENOTSUP)); 1231 } 1232 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1233 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1234 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1235 VN_URELE(ZTOV(dzp)); 1236 } else { 1237 *eofp = 1; 1238 error = SET_ERROR(ENOENT); 1239 } 1240 1241 ZFS_EXIT(zfsvfs); 1242 return (error); 1243} 1244 1245/* 1246 * pvp is the '.zfs' directory (zfsctl_node_t). 1247 * 1248 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1249 * 1250 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1251 * when a lookup is performed on .zfs for "snapshot". 1252 */ 1253vnode_t * 1254zfsctl_mknode_snapdir(vnode_t *pvp) 1255{ 1256 vnode_t *vp; 1257 zfsctl_snapdir_t *sdp; 1258 1259 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1260 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1261 zfsctl_snapdir_readdir_cb, NULL); 1262 sdp = vp->v_data; 1263 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1264 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1265 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1266 avl_create(&sdp->sd_snaps, snapentry_compare, 1267 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1268 VOP_UNLOCK(vp, 0); 1269 return (vp); 1270} 1271 1272vnode_t * 1273zfsctl_mknode_shares(vnode_t *pvp) 1274{ 1275 vnode_t *vp; 1276 zfsctl_node_t *sdp; 1277 1278 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1279 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1280 NULL, NULL); 1281 sdp = vp->v_data; 1282 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1283 VOP_UNLOCK(vp, 0); 1284 return (vp); 1285 1286} 1287 1288/* ARGSUSED */ 1289static int 1290zfsctl_shares_getattr(ap) 1291 struct vop_getattr_args /* { 1292 struct vnode *a_vp; 1293 struct vattr *a_vap; 1294 struct ucred *a_cred; 1295 struct thread *a_td; 1296 } */ *ap; 1297{ 1298 vnode_t *vp = ap->a_vp; 1299 vattr_t *vap = ap->a_vap; 1300 cred_t *cr = ap->a_cred; 1301 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1302 znode_t *dzp; 1303 int error; 1304 1305 ZFS_ENTER(zfsvfs); 1306 if (zfsvfs->z_shares_dir == 0) { 1307 ZFS_EXIT(zfsvfs); 1308 return (SET_ERROR(ENOTSUP)); 1309 } 1310 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1311 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1312 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1313 VN_URELE(ZTOV(dzp)); 1314 } 1315 ZFS_EXIT(zfsvfs); 1316 return (error); 1317 1318 1319} 1320 1321/* ARGSUSED */ 1322static int 1323zfsctl_snapdir_getattr(ap) 1324 struct vop_getattr_args /* { 1325 struct vnode *a_vp; 1326 struct vattr *a_vap; 1327 struct ucred *a_cred; 1328 } */ *ap; 1329{ 1330 vnode_t *vp = ap->a_vp; 1331 vattr_t *vap = ap->a_vap; 1332 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1333 zfsctl_snapdir_t *sdp = vp->v_data; 1334 1335 ZFS_ENTER(zfsvfs); 1336 zfsctl_common_getattr(vp, vap); 1337 vap->va_nodeid = gfs_file_inode(vp); 1338 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1339 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1340 vap->va_birthtime = vap->va_ctime; 1341 ZFS_EXIT(zfsvfs); 1342 1343 return (0); 1344} 1345 1346/* ARGSUSED */ 1347static int 1348zfsctl_snapdir_inactive(ap) 1349 struct vop_inactive_args /* { 1350 struct vnode *a_vp; 1351 struct thread *a_td; 1352 } */ *ap; 1353{ 1354 vnode_t *vp = ap->a_vp; 1355 zfsctl_snapdir_t *sdp = vp->v_data; 1356 zfs_snapentry_t *sep; 1357 1358 /* 1359 * On forced unmount we have to free snapshots from here. 1360 */ 1361 mutex_enter(&sdp->sd_lock); 1362 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) { 1363 avl_remove(&sdp->sd_snaps, sep); 1364 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1365 kmem_free(sep, sizeof (zfs_snapentry_t)); 1366 } 1367 mutex_exit(&sdp->sd_lock); 1368 gfs_dir_inactive(vp); 1369 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1370 mutex_destroy(&sdp->sd_lock); 1371 avl_destroy(&sdp->sd_snaps); 1372 kmem_free(sdp, sizeof (zfsctl_snapdir_t)); 1373 1374 return (0); 1375} 1376 1377#ifdef sun 1378static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1379 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1380 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1381 { VOPNAME_IOCTL, { .error = fs_inval } }, 1382 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1383 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1384 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1385 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1386 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1387 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1388 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1389 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1390 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1391 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1392 { NULL } 1393}; 1394 1395static const fs_operation_def_t zfsctl_tops_shares[] = { 1396 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1397 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1398 { VOPNAME_IOCTL, { .error = fs_inval } }, 1399 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1400 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1401 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1402 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1403 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1404 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1405 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1406 { NULL } 1407}; 1408#else /* !sun */ 1409static struct vop_vector zfsctl_ops_snapdir = { 1410 .vop_default = &default_vnodeops, 1411 .vop_open = zfsctl_common_open, 1412 .vop_close = zfsctl_common_close, 1413 .vop_ioctl = VOP_EINVAL, 1414 .vop_getattr = zfsctl_snapdir_getattr, 1415 .vop_access = zfsctl_common_access, 1416 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1417 .vop_readdir = gfs_vop_readdir, 1418 .vop_lookup = zfsctl_snapdir_lookup, 1419 .vop_inactive = zfsctl_snapdir_inactive, 1420 .vop_reclaim = zfsctl_common_reclaim, 1421 .vop_fid = zfsctl_common_fid, 1422}; 1423 1424static struct vop_vector zfsctl_ops_shares = { 1425 .vop_default = &default_vnodeops, 1426 .vop_open = zfsctl_common_open, 1427 .vop_close = zfsctl_common_close, 1428 .vop_ioctl = VOP_EINVAL, 1429 .vop_getattr = zfsctl_shares_getattr, 1430 .vop_access = zfsctl_common_access, 1431 .vop_readdir = zfsctl_shares_readdir, 1432 .vop_lookup = zfsctl_shares_lookup, 1433 .vop_inactive = VOP_NULL, 1434 .vop_reclaim = gfs_vop_reclaim, 1435 .vop_fid = zfsctl_shares_fid, 1436}; 1437#endif /* !sun */ 1438 1439/* 1440 * pvp is the GFS vnode '.zfs/snapshot'. 1441 * 1442 * This creates a GFS node under '.zfs/snapshot' representing each 1443 * snapshot. This newly created GFS node is what we mount snapshot 1444 * vfs_t's ontop of. 1445 */ 1446static vnode_t * 1447zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1448{ 1449 vnode_t *vp; 1450 zfsctl_node_t *zcp; 1451 1452 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1453 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1454 VN_HOLD(vp); 1455 zcp = vp->v_data; 1456 zcp->zc_id = objset; 1457 VOP_UNLOCK(vp, 0); 1458 1459 return (vp); 1460} 1461 1462 1463static int 1464zfsctl_snapshot_reclaim(ap) 1465 struct vop_inactive_args /* { 1466 struct vnode *a_vp; 1467 struct thread *a_td; 1468 } */ *ap; 1469{ 1470 vnode_t *vp = ap->a_vp; 1471 cred_t *cr = ap->a_td->td_ucred; 1472 struct vop_reclaim_args iap; 1473 zfsctl_snapdir_t *sdp; 1474 zfs_snapentry_t *sep, *next; 1475 int locked; 1476 vnode_t *dvp; 1477 1478 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1479 sdp = dvp->v_data; 1480 VOP_UNLOCK(dvp, 0); 1481 /* this may already have been unmounted */ 1482 if (sdp == NULL) { 1483 VN_RELE(dvp); 1484 return (0); 1485 } 1486 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1487 mutex_enter(&sdp->sd_lock); 1488 1489 ASSERT(!vn_ismntpt(vp)); 1490 1491 sep = avl_first(&sdp->sd_snaps); 1492 while (sep != NULL) { 1493 next = AVL_NEXT(&sdp->sd_snaps, sep); 1494 1495 if (sep->se_root == vp) { 1496 avl_remove(&sdp->sd_snaps, sep); 1497 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1498 kmem_free(sep, sizeof (zfs_snapentry_t)); 1499 break; 1500 } 1501 sep = next; 1502 } 1503 ASSERT(sep != NULL); 1504 1505 if (!locked) 1506 mutex_exit(&sdp->sd_lock); 1507 VN_RELE(dvp); 1508 1509 /* 1510 * Dispose of the vnode for the snapshot mount point. 1511 * This is safe to do because once this entry has been removed 1512 * from the AVL tree, it can't be found again, so cannot become 1513 * "active". If we lookup the same name again we will end up 1514 * creating a new vnode. 1515 */ 1516 iap.a_vp = vp; 1517 gfs_vop_reclaim(&iap); 1518 return (0); 1519 1520} 1521 1522static int 1523zfsctl_traverse_begin(vnode_t **vpp, int lktype) 1524{ 1525 1526 VN_HOLD(*vpp); 1527 /* Snapshot should be already mounted, but just in case. */ 1528 if (vn_mountedvfs(*vpp) == NULL) 1529 return (ENOENT); 1530 return (traverse(vpp, lktype)); 1531} 1532 1533static void 1534zfsctl_traverse_end(vnode_t *vp, int err) 1535{ 1536 1537 if (err == 0) 1538 vput(vp); 1539 else 1540 VN_RELE(vp); 1541} 1542 1543static int 1544zfsctl_snapshot_getattr(ap) 1545 struct vop_getattr_args /* { 1546 struct vnode *a_vp; 1547 struct vattr *a_vap; 1548 struct ucred *a_cred; 1549 } */ *ap; 1550{ 1551 vnode_t *vp = ap->a_vp; 1552 int err; 1553 1554 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1555 if (err == 0) 1556 err = VOP_GETATTR(vp, ap->a_vap, ap->a_cred); 1557 zfsctl_traverse_end(vp, err); 1558 return (err); 1559} 1560 1561static int 1562zfsctl_snapshot_fid(ap) 1563 struct vop_fid_args /* { 1564 struct vnode *a_vp; 1565 struct fid *a_fid; 1566 } */ *ap; 1567{ 1568 vnode_t *vp = ap->a_vp; 1569 int err; 1570 1571 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1572 if (err == 0) 1573 err = VOP_VPTOFH(vp, (void *)ap->a_fid); 1574 zfsctl_traverse_end(vp, err); 1575 return (err); 1576} 1577 1578static int 1579zfsctl_snapshot_lookup(ap) 1580 struct vop_lookup_args /* { 1581 struct vnode *a_dvp; 1582 struct vnode **a_vpp; 1583 struct componentname *a_cnp; 1584 } */ *ap; 1585{ 1586 vnode_t *dvp = ap->a_dvp; 1587 vnode_t **vpp = ap->a_vpp; 1588 struct componentname *cnp = ap->a_cnp; 1589 cred_t *cr = ap->a_cnp->cn_cred; 1590 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1591 int error; 1592 1593 if (cnp->cn_namelen != 2 || cnp->cn_nameptr[0] != '.' || 1594 cnp->cn_nameptr[1] != '.') { 1595 return (ENOENT); 1596 } 1597 1598 ASSERT(dvp->v_type == VDIR); 1599 ASSERT(zfsvfs->z_ctldir != NULL); 1600 1601 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", vpp, 1602 NULL, 0, NULL, cr, NULL, NULL, NULL); 1603 if (error == 0) { 1604 int ltype = VOP_ISLOCKED(dvp); 1605 VN_HOLD(*vpp); 1606 VOP_UNLOCK(dvp, 0); 1607 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1608 VN_RELE(*vpp); 1609 vn_lock(dvp, ltype | LK_RETRY); 1610 } 1611 1612 return (error); 1613} 1614 1615static int 1616zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1617{ 1618 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1619 vnode_t *dvp, *vp; 1620 zfsctl_snapdir_t *sdp; 1621 zfs_snapentry_t *sep; 1622 int error; 1623 1624 ASSERT(zfsvfs->z_ctldir != NULL); 1625 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1626 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1627 if (error != 0) 1628 return (error); 1629 sdp = dvp->v_data; 1630 1631 mutex_enter(&sdp->sd_lock); 1632 sep = avl_first(&sdp->sd_snaps); 1633 while (sep != NULL) { 1634 vp = sep->se_root; 1635 if (vp == ap->a_vp) 1636 break; 1637 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1638 } 1639 if (sep == NULL) { 1640 mutex_exit(&sdp->sd_lock); 1641 error = ENOENT; 1642 } else { 1643 size_t len; 1644 1645 len = strlen(sep->se_name); 1646 *ap->a_buflen -= len; 1647 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1648 mutex_exit(&sdp->sd_lock); 1649 vref(dvp); 1650 *ap->a_vpp = dvp; 1651 } 1652 VN_RELE(dvp); 1653 1654 return (error); 1655} 1656 1657/* 1658 * These VP's should never see the light of day. They should always 1659 * be covered. 1660 */ 1661static struct vop_vector zfsctl_ops_snapshot = { 1662 .vop_default = &default_vnodeops, 1663 .vop_inactive = VOP_NULL, 1664 .vop_lookup = zfsctl_snapshot_lookup, 1665 .vop_reclaim = zfsctl_snapshot_reclaim, 1666 .vop_getattr = zfsctl_snapshot_getattr, 1667 .vop_fid = zfsctl_snapshot_fid, 1668 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1669}; 1670 1671int 1672zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1673{ 1674 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1675 vnode_t *dvp, *vp; 1676 zfsctl_snapdir_t *sdp; 1677 zfsctl_node_t *zcp; 1678 zfs_snapentry_t *sep; 1679 int error; 1680 1681 ASSERT(zfsvfs->z_ctldir != NULL); 1682 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1683 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1684 if (error != 0) 1685 return (error); 1686 sdp = dvp->v_data; 1687 1688 mutex_enter(&sdp->sd_lock); 1689 sep = avl_first(&sdp->sd_snaps); 1690 while (sep != NULL) { 1691 vp = sep->se_root; 1692 zcp = vp->v_data; 1693 if (zcp->zc_id == objsetid) 1694 break; 1695 1696 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1697 } 1698 1699 if (sep != NULL) { 1700 VN_HOLD(vp); 1701 /* 1702 * Return the mounted root rather than the covered mount point. 1703 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1704 * and returns the ZFS vnode mounted on top of the GFS node. 1705 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1706 */ 1707 error = traverse(&vp, LK_SHARED | LK_RETRY); 1708 if (error == 0) { 1709 if (vp == sep->se_root) 1710 error = SET_ERROR(EINVAL); 1711 else 1712 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1713 } 1714 mutex_exit(&sdp->sd_lock); 1715 if (error == 0) 1716 VN_URELE(vp); 1717 else 1718 VN_RELE(vp); 1719 } else { 1720 error = SET_ERROR(EINVAL); 1721 mutex_exit(&sdp->sd_lock); 1722 } 1723 1724 VN_RELE(dvp); 1725 1726 return (error); 1727} 1728 1729/* 1730 * Unmount any snapshots for the given filesystem. This is called from 1731 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1732 * snapshots. 1733 */ 1734int 1735zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1736{ 1737 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1738 vnode_t *dvp; 1739 zfsctl_snapdir_t *sdp; 1740 zfs_snapentry_t *sep, *next; 1741 int error; 1742 1743 ASSERT(zfsvfs->z_ctldir != NULL); 1744 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1745 NULL, 0, NULL, cr, NULL, NULL, NULL); 1746 if (error != 0) 1747 return (error); 1748 sdp = dvp->v_data; 1749 1750 mutex_enter(&sdp->sd_lock); 1751 1752 sep = avl_first(&sdp->sd_snaps); 1753 while (sep != NULL) { 1754 next = AVL_NEXT(&sdp->sd_snaps, sep); 1755 1756 /* 1757 * If this snapshot is not mounted, then it must 1758 * have just been unmounted by somebody else, and 1759 * will be cleaned up by zfsctl_snapdir_inactive(). 1760 */ 1761 if (vn_ismntpt(sep->se_root)) { 1762 error = zfsctl_unmount_snap(sep, fflags, cr); 1763 if (error) { 1764 avl_index_t where; 1765 1766 /* 1767 * Before reinserting snapshot to the tree, 1768 * check if it was actually removed. For example 1769 * when snapshot mount point is busy, we will 1770 * have an error here, but there will be no need 1771 * to reinsert snapshot. 1772 */ 1773 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1774 avl_insert(&sdp->sd_snaps, sep, where); 1775 break; 1776 } 1777 } 1778 sep = next; 1779 } 1780 1781 mutex_exit(&sdp->sd_lock); 1782 VN_RELE(dvp); 1783 1784 return (error); 1785} 1786