zfs_ctldir.c revision 249643
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 636static int 637zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 638{ 639 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 640 641 if (snapshot_namecheck(name, NULL, NULL) != 0) 642 return (SET_ERROR(EILSEQ)); 643 dmu_objset_name(os, zname); 644 if (strlen(zname) + 1 + strlen(name) >= len) 645 return (SET_ERROR(ENAMETOOLONG)); 646 (void) strcat(zname, "@"); 647 (void) strcat(zname, name); 648 return (0); 649} 650 651static int 652zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 653{ 654 vnode_t *svp = sep->se_root; 655 int error; 656 657 ASSERT(vn_ismntpt(svp)); 658 659 /* this will be dropped by dounmount() */ 660 if ((error = vn_vfswlock(svp)) != 0) 661 return (error); 662 663#ifdef sun 664 VN_HOLD(svp); 665 error = dounmount(vn_mountedvfs(svp), fflags, cr); 666 if (error) { 667 VN_RELE(svp); 668 return (error); 669 } 670 671 /* 672 * We can't use VN_RELE(), as that will try to invoke 673 * zfsctl_snapdir_inactive(), which would cause us to destroy 674 * the sd_lock mutex held by our caller. 675 */ 676 ASSERT(svp->v_count == 1); 677 gfs_vop_inactive(svp, cr, NULL); 678 679 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 680 kmem_free(sep, sizeof (zfs_snapentry_t)); 681 682 return (0); 683#else /* !sun */ 684 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 685#endif /* !sun */ 686} 687 688#ifdef sun 689static void 690zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 691{ 692 avl_index_t where; 693 vfs_t *vfsp; 694 refstr_t *pathref; 695 char newpath[MAXNAMELEN]; 696 char *tail; 697 698 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 699 ASSERT(sep != NULL); 700 701 vfsp = vn_mountedvfs(sep->se_root); 702 ASSERT(vfsp != NULL); 703 704 vfs_lock_wait(vfsp); 705 706 /* 707 * Change the name in the AVL tree. 708 */ 709 avl_remove(&sdp->sd_snaps, sep); 710 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 711 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 712 (void) strcpy(sep->se_name, nm); 713 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 714 avl_insert(&sdp->sd_snaps, sep, where); 715 716 /* 717 * Change the current mountpoint info: 718 * - update the tail of the mntpoint path 719 * - update the tail of the resource path 720 */ 721 pathref = vfs_getmntpoint(vfsp); 722 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 723 VERIFY((tail = strrchr(newpath, '/')) != NULL); 724 *(tail+1) = '\0'; 725 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 726 (void) strcat(newpath, nm); 727 refstr_rele(pathref); 728 vfs_setmntpoint(vfsp, newpath, 0); 729 730 pathref = vfs_getresource(vfsp); 731 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 732 VERIFY((tail = strrchr(newpath, '@')) != NULL); 733 *(tail+1) = '\0'; 734 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 735 (void) strcat(newpath, nm); 736 refstr_rele(pathref); 737 vfs_setresource(vfsp, newpath, 0); 738 739 vfs_unlock(vfsp); 740} 741#endif /* sun */ 742 743#ifdef sun 744/*ARGSUSED*/ 745static int 746zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 747 cred_t *cr, caller_context_t *ct, int flags) 748{ 749 zfsctl_snapdir_t *sdp = sdvp->v_data; 750 zfs_snapentry_t search, *sep; 751 zfsvfs_t *zfsvfs; 752 avl_index_t where; 753 char from[MAXNAMELEN], to[MAXNAMELEN]; 754 char real[MAXNAMELEN], fsname[MAXNAMELEN]; 755 int err; 756 757 zfsvfs = sdvp->v_vfsp->vfs_data; 758 ZFS_ENTER(zfsvfs); 759 760 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 761 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 762 MAXNAMELEN, NULL); 763 if (err == 0) { 764 snm = real; 765 } else if (err != ENOTSUP) { 766 ZFS_EXIT(zfsvfs); 767 return (err); 768 } 769 } 770 771 ZFS_EXIT(zfsvfs); 772 773 dmu_objset_name(zfsvfs->z_os, fsname); 774 775 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 776 if (err == 0) 777 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 778 if (err == 0) 779 err = zfs_secpolicy_rename_perms(from, to, cr); 780 if (err != 0) 781 return (err); 782 783 /* 784 * Cannot move snapshots out of the snapdir. 785 */ 786 if (sdvp != tdvp) 787 return (SET_ERROR(EINVAL)); 788 789 if (strcmp(snm, tnm) == 0) 790 return (0); 791 792 mutex_enter(&sdp->sd_lock); 793 794 search.se_name = (char *)snm; 795 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 796 mutex_exit(&sdp->sd_lock); 797 return (SET_ERROR(ENOENT)); 798 } 799 800 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0); 801 if (err == 0) 802 zfsctl_rename_snap(sdp, sep, tnm); 803 804 mutex_exit(&sdp->sd_lock); 805 806 return (err); 807} 808#endif /* sun */ 809 810#ifdef sun 811/* ARGSUSED */ 812static int 813zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 814 caller_context_t *ct, int flags) 815{ 816 zfsctl_snapdir_t *sdp = dvp->v_data; 817 zfs_snapentry_t *sep; 818 zfs_snapentry_t search; 819 zfsvfs_t *zfsvfs; 820 char snapname[MAXNAMELEN]; 821 char real[MAXNAMELEN]; 822 int err; 823 824 zfsvfs = dvp->v_vfsp->vfs_data; 825 ZFS_ENTER(zfsvfs); 826 827 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 828 829 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 830 MAXNAMELEN, NULL); 831 if (err == 0) { 832 name = real; 833 } else if (err != ENOTSUP) { 834 ZFS_EXIT(zfsvfs); 835 return (err); 836 } 837 } 838 839 ZFS_EXIT(zfsvfs); 840 841 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 842 if (err == 0) 843 err = zfs_secpolicy_destroy_perms(snapname, cr); 844 if (err != 0) 845 return (err); 846 847 mutex_enter(&sdp->sd_lock); 848 849 search.se_name = name; 850 sep = avl_find(&sdp->sd_snaps, &search, NULL); 851 if (sep) { 852 avl_remove(&sdp->sd_snaps, sep); 853 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 854 if (err != 0) 855 avl_add(&sdp->sd_snaps, sep); 856 else 857 err = dsl_destroy_snapshot(snapname, B_FALSE); 858 } else { 859 err = SET_ERROR(ENOENT); 860 } 861 862 mutex_exit(&sdp->sd_lock); 863 864 return (err); 865} 866#endif /* sun */ 867 868/* 869 * This creates a snapshot under '.zfs/snapshot'. 870 */ 871/* ARGSUSED */ 872static int 873zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 874 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 875{ 876 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 877 char name[MAXNAMELEN]; 878 int err; 879 static enum symfollow follow = NO_FOLLOW; 880 static enum uio_seg seg = UIO_SYSSPACE; 881 882 if (snapshot_namecheck(dirname, NULL, NULL) != 0) 883 return (SET_ERROR(EILSEQ)); 884 885 dmu_objset_name(zfsvfs->z_os, name); 886 887 *vpp = NULL; 888 889 err = zfs_secpolicy_snapshot_perms(name, cr); 890 if (err != 0) 891 return (err); 892 893 if (err == 0) { 894 err = dmu_objset_snapshot_one(name, dirname); 895 if (err != 0) 896 return (err); 897 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 898 } 899 900 return (err); 901} 902 903static int 904zfsctl_freebsd_snapdir_mkdir(ap) 905 struct vop_mkdir_args /* { 906 struct vnode *a_dvp; 907 struct vnode **a_vpp; 908 struct componentname *a_cnp; 909 struct vattr *a_vap; 910 } */ *ap; 911{ 912 913 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 914 915 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 916 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 917} 918 919/* 920 * Lookup entry point for the 'snapshot' directory. Try to open the 921 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 922 * Perform a mount of the associated dataset on top of the vnode. 923 */ 924/* ARGSUSED */ 925int 926zfsctl_snapdir_lookup(ap) 927 struct vop_lookup_args /* { 928 struct vnode *a_dvp; 929 struct vnode **a_vpp; 930 struct componentname *a_cnp; 931 } */ *ap; 932{ 933 vnode_t *dvp = ap->a_dvp; 934 vnode_t **vpp = ap->a_vpp; 935 struct componentname *cnp = ap->a_cnp; 936 char nm[NAME_MAX + 1]; 937 zfsctl_snapdir_t *sdp = dvp->v_data; 938 objset_t *snap; 939 char snapname[MAXNAMELEN]; 940 char real[MAXNAMELEN]; 941 char *mountpoint; 942 zfs_snapentry_t *sep, search; 943 size_t mountpoint_len; 944 avl_index_t where; 945 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 946 int err; 947 int flags = 0; 948 949 /* 950 * No extended attributes allowed under .zfs 951 */ 952 if (flags & LOOKUP_XATTR) 953 return (SET_ERROR(EINVAL)); 954 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 955 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 956 957 ASSERT(dvp->v_type == VDIR); 958 959 *vpp = NULL; 960 961 /* 962 * If we get a recursive call, that means we got called 963 * from the domount() code while it was trying to look up the 964 * spec (which looks like a local path for zfs). We need to 965 * add some flag to domount() to tell it not to do this lookup. 966 */ 967 if (MUTEX_HELD(&sdp->sd_lock)) 968 return (SET_ERROR(ENOENT)); 969 970 ZFS_ENTER(zfsvfs); 971 972 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 973 ZFS_EXIT(zfsvfs); 974 return (0); 975 } 976 977 if (flags & FIGNORECASE) { 978 boolean_t conflict = B_FALSE; 979 980 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 981 MAXNAMELEN, &conflict); 982 if (err == 0) { 983 strlcpy(nm, real, sizeof(nm)); 984 } else if (err != ENOTSUP) { 985 ZFS_EXIT(zfsvfs); 986 return (err); 987 } 988#if 0 989 if (realpnp) 990 (void) strlcpy(realpnp->pn_buf, nm, 991 realpnp->pn_bufsize); 992 if (conflict && direntflags) 993 *direntflags = ED_CASE_CONFLICT; 994#endif 995 } 996 997 mutex_enter(&sdp->sd_lock); 998 search.se_name = (char *)nm; 999 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 1000 *vpp = sep->se_root; 1001 VN_HOLD(*vpp); 1002 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 1003 if (err != 0) { 1004 VN_RELE(*vpp); 1005 *vpp = NULL; 1006 } else if (*vpp == sep->se_root) { 1007 /* 1008 * The snapshot was unmounted behind our backs, 1009 * try to remount it. 1010 */ 1011 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1012 goto domount; 1013 } else { 1014 /* 1015 * VROOT was set during the traverse call. We need 1016 * to clear it since we're pretending to be part 1017 * of our parent's vfs. 1018 */ 1019 (*vpp)->v_flag &= ~VROOT; 1020 } 1021 mutex_exit(&sdp->sd_lock); 1022 ZFS_EXIT(zfsvfs); 1023 return (err); 1024 } 1025 1026 /* 1027 * The requested snapshot is not currently mounted, look it up. 1028 */ 1029 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1030 if (err != 0) { 1031 mutex_exit(&sdp->sd_lock); 1032 ZFS_EXIT(zfsvfs); 1033 /* 1034 * handle "ls *" or "?" in a graceful manner, 1035 * forcing EILSEQ to ENOENT. 1036 * Since shell ultimately passes "*" or "?" as name to lookup 1037 */ 1038 return (err == EILSEQ ? ENOENT : err); 1039 } 1040 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1041 mutex_exit(&sdp->sd_lock); 1042#ifdef illumos 1043 ZFS_EXIT(zfsvfs); 1044 return (SET_ERROR(ENOENT)); 1045#else /* !illumos */ 1046 /* Translate errors and add SAVENAME when needed. */ 1047 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1048 err = EJUSTRETURN; 1049 cnp->cn_flags |= SAVENAME; 1050 } else { 1051 err = SET_ERROR(ENOENT); 1052 } 1053 ZFS_EXIT(zfsvfs); 1054 return (err); 1055#endif /* !illumos */ 1056 } 1057 1058 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1059 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1060 (void) strcpy(sep->se_name, nm); 1061 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1062 VN_HOLD(*vpp); 1063 avl_insert(&sdp->sd_snaps, sep, where); 1064 1065 dmu_objset_rele(snap, FTAG); 1066domount: 1067 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1068 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1069 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1070 (void) snprintf(mountpoint, mountpoint_len, 1071 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1072 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1073 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1074 kmem_free(mountpoint, mountpoint_len); 1075 if (err == 0) { 1076 /* 1077 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1078 * 1079 * This is where we lie about our v_vfsp in order to 1080 * make .zfs/snapshot/<snapname> accessible over NFS 1081 * without requiring manual mounts of <snapname>. 1082 */ 1083 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1084 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1085 } 1086 mutex_exit(&sdp->sd_lock); 1087 ZFS_EXIT(zfsvfs); 1088 1089#ifdef illumos 1090 /* 1091 * If we had an error, drop our hold on the vnode and 1092 * zfsctl_snapshot_inactive() will clean up. 1093 */ 1094 if (err != 0) { 1095 VN_RELE(*vpp); 1096 *vpp = NULL; 1097 } 1098#else 1099 if (err != 0) 1100 *vpp = NULL; 1101#endif 1102 return (err); 1103} 1104 1105/* ARGSUSED */ 1106int 1107zfsctl_shares_lookup(ap) 1108 struct vop_lookup_args /* { 1109 struct vnode *a_dvp; 1110 struct vnode **a_vpp; 1111 struct componentname *a_cnp; 1112 } */ *ap; 1113{ 1114 vnode_t *dvp = ap->a_dvp; 1115 vnode_t **vpp = ap->a_vpp; 1116 struct componentname *cnp = ap->a_cnp; 1117 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1118 char nm[NAME_MAX + 1]; 1119 znode_t *dzp; 1120 int error; 1121 1122 ZFS_ENTER(zfsvfs); 1123 1124 ASSERT(cnp->cn_namelen < sizeof(nm)); 1125 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1126 1127 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1128 ZFS_EXIT(zfsvfs); 1129 return (0); 1130 } 1131 1132 if (zfsvfs->z_shares_dir == 0) { 1133 ZFS_EXIT(zfsvfs); 1134 return (SET_ERROR(ENOTSUP)); 1135 } 1136 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) 1137 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1138 1139 VN_RELE(ZTOV(dzp)); 1140 ZFS_EXIT(zfsvfs); 1141 1142 return (error); 1143} 1144 1145/* ARGSUSED */ 1146static int 1147zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1148 offset_t *offp, offset_t *nextp, void *data, int flags) 1149{ 1150 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1151 char snapname[MAXNAMELEN]; 1152 uint64_t id, cookie; 1153 boolean_t case_conflict; 1154 int error; 1155 1156 ZFS_ENTER(zfsvfs); 1157 1158 cookie = *offp; 1159 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1160 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1161 &cookie, &case_conflict); 1162 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1163 if (error) { 1164 ZFS_EXIT(zfsvfs); 1165 if (error == ENOENT) { 1166 *eofp = 1; 1167 return (0); 1168 } 1169 return (error); 1170 } 1171 1172 if (flags & V_RDDIR_ENTFLAGS) { 1173 edirent_t *eodp = dp; 1174 1175 (void) strcpy(eodp->ed_name, snapname); 1176 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1177 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1178 } else { 1179 struct dirent64 *odp = dp; 1180 1181 (void) strcpy(odp->d_name, snapname); 1182 odp->d_ino = ZFSCTL_INO_SNAP(id); 1183 } 1184 *nextp = cookie; 1185 1186 ZFS_EXIT(zfsvfs); 1187 1188 return (0); 1189} 1190 1191/* ARGSUSED */ 1192static int 1193zfsctl_shares_readdir(ap) 1194 struct vop_readdir_args /* { 1195 struct vnode *a_vp; 1196 struct uio *a_uio; 1197 struct ucred *a_cred; 1198 int *a_eofflag; 1199 int *a_ncookies; 1200 u_long **a_cookies; 1201 } */ *ap; 1202{ 1203 vnode_t *vp = ap->a_vp; 1204 uio_t *uiop = ap->a_uio; 1205 cred_t *cr = ap->a_cred; 1206 int *eofp = ap->a_eofflag; 1207 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1208 znode_t *dzp; 1209 int error; 1210 1211 ZFS_ENTER(zfsvfs); 1212 1213 if (zfsvfs->z_shares_dir == 0) { 1214 ZFS_EXIT(zfsvfs); 1215 return (SET_ERROR(ENOTSUP)); 1216 } 1217 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1218 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1219 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1220 VN_URELE(ZTOV(dzp)); 1221 } else { 1222 *eofp = 1; 1223 error = SET_ERROR(ENOENT); 1224 } 1225 1226 ZFS_EXIT(zfsvfs); 1227 return (error); 1228} 1229 1230/* 1231 * pvp is the '.zfs' directory (zfsctl_node_t). 1232 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1233 * 1234 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1235 * when a lookup is performed on .zfs for "snapshot". 1236 */ 1237vnode_t * 1238zfsctl_mknode_snapdir(vnode_t *pvp) 1239{ 1240 vnode_t *vp; 1241 zfsctl_snapdir_t *sdp; 1242 1243 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1244 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1245 zfsctl_snapdir_readdir_cb, NULL); 1246 sdp = vp->v_data; 1247 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1248 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1249 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1250 avl_create(&sdp->sd_snaps, snapentry_compare, 1251 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1252 VOP_UNLOCK(vp, 0); 1253 return (vp); 1254} 1255 1256vnode_t * 1257zfsctl_mknode_shares(vnode_t *pvp) 1258{ 1259 vnode_t *vp; 1260 zfsctl_node_t *sdp; 1261 1262 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1263 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1264 NULL, NULL); 1265 sdp = vp->v_data; 1266 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1267 VOP_UNLOCK(vp, 0); 1268 return (vp); 1269 1270} 1271 1272/* ARGSUSED */ 1273static int 1274zfsctl_shares_getattr(ap) 1275 struct vop_getattr_args /* { 1276 struct vnode *a_vp; 1277 struct vattr *a_vap; 1278 struct ucred *a_cred; 1279 struct thread *a_td; 1280 } */ *ap; 1281{ 1282 vnode_t *vp = ap->a_vp; 1283 vattr_t *vap = ap->a_vap; 1284 cred_t *cr = ap->a_cred; 1285 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1286 znode_t *dzp; 1287 int error; 1288 1289 ZFS_ENTER(zfsvfs); 1290 if (zfsvfs->z_shares_dir == 0) { 1291 ZFS_EXIT(zfsvfs); 1292 return (SET_ERROR(ENOTSUP)); 1293 } 1294 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1295 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1296 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1297 VN_URELE(ZTOV(dzp)); 1298 } 1299 ZFS_EXIT(zfsvfs); 1300 return (error); 1301 1302 1303} 1304 1305/* ARGSUSED */ 1306static int 1307zfsctl_snapdir_getattr(ap) 1308 struct vop_getattr_args /* { 1309 struct vnode *a_vp; 1310 struct vattr *a_vap; 1311 struct ucred *a_cred; 1312 } */ *ap; 1313{ 1314 vnode_t *vp = ap->a_vp; 1315 vattr_t *vap = ap->a_vap; 1316 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1317 zfsctl_snapdir_t *sdp = vp->v_data; 1318 1319 ZFS_ENTER(zfsvfs); 1320 zfsctl_common_getattr(vp, vap); 1321 vap->va_nodeid = gfs_file_inode(vp); 1322 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1323 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1324 vap->va_birthtime = vap->va_ctime; 1325 ZFS_EXIT(zfsvfs); 1326 1327 return (0); 1328} 1329 1330/* ARGSUSED */ 1331static int 1332zfsctl_snapdir_inactive(ap) 1333 struct vop_inactive_args /* { 1334 struct vnode *a_vp; 1335 struct thread *a_td; 1336 } */ *ap; 1337{ 1338 vnode_t *vp = ap->a_vp; 1339 zfsctl_snapdir_t *sdp = vp->v_data; 1340 zfs_snapentry_t *sep; 1341 1342 /* 1343 * On forced unmount we have to free snapshots from here. 1344 */ 1345 mutex_enter(&sdp->sd_lock); 1346 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) { 1347 avl_remove(&sdp->sd_snaps, sep); 1348 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1349 kmem_free(sep, sizeof (zfs_snapentry_t)); 1350 } 1351 mutex_exit(&sdp->sd_lock); 1352 gfs_dir_inactive(vp); 1353 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1354 mutex_destroy(&sdp->sd_lock); 1355 avl_destroy(&sdp->sd_snaps); 1356 kmem_free(sdp, sizeof (zfsctl_snapdir_t)); 1357 1358 return (0); 1359} 1360 1361#ifdef sun 1362static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1363 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1364 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1365 { VOPNAME_IOCTL, { .error = fs_inval } }, 1366 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1367 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1368 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1369 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1370 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1371 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1372 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1373 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1374 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1375 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1376 { NULL } 1377}; 1378 1379static const fs_operation_def_t zfsctl_tops_shares[] = { 1380 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1381 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1382 { VOPNAME_IOCTL, { .error = fs_inval } }, 1383 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1384 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1385 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1386 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1387 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1388 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1389 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1390 { NULL } 1391}; 1392#else /* !sun */ 1393static struct vop_vector zfsctl_ops_snapdir = { 1394 .vop_default = &default_vnodeops, 1395 .vop_open = zfsctl_common_open, 1396 .vop_close = zfsctl_common_close, 1397 .vop_ioctl = VOP_EINVAL, 1398 .vop_getattr = zfsctl_snapdir_getattr, 1399 .vop_access = zfsctl_common_access, 1400 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1401 .vop_readdir = gfs_vop_readdir, 1402 .vop_lookup = zfsctl_snapdir_lookup, 1403 .vop_inactive = zfsctl_snapdir_inactive, 1404 .vop_reclaim = zfsctl_common_reclaim, 1405 .vop_fid = zfsctl_common_fid, 1406}; 1407 1408static struct vop_vector zfsctl_ops_shares = { 1409 .vop_default = &default_vnodeops, 1410 .vop_open = zfsctl_common_open, 1411 .vop_close = zfsctl_common_close, 1412 .vop_ioctl = VOP_EINVAL, 1413 .vop_getattr = zfsctl_shares_getattr, 1414 .vop_access = zfsctl_common_access, 1415 .vop_readdir = zfsctl_shares_readdir, 1416 .vop_lookup = zfsctl_shares_lookup, 1417 .vop_inactive = gfs_vop_inactive, 1418 .vop_reclaim = zfsctl_common_reclaim, 1419 .vop_fid = zfsctl_shares_fid, 1420}; 1421#endif /* !sun */ 1422 1423/* 1424 * pvp is the GFS vnode '.zfs/snapshot'. 1425 * 1426 * This creates a GFS node under '.zfs/snapshot' representing each 1427 * snapshot. This newly created GFS node is what we mount snapshot 1428 * vfs_t's ontop of. 1429 */ 1430static vnode_t * 1431zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1432{ 1433 vnode_t *vp; 1434 zfsctl_node_t *zcp; 1435 1436 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1437 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1438 VN_HOLD(vp); 1439 zcp = vp->v_data; 1440 zcp->zc_id = objset; 1441 VOP_UNLOCK(vp, 0); 1442 1443 return (vp); 1444} 1445 1446static int 1447zfsctl_snapshot_inactive(ap) 1448 struct vop_inactive_args /* { 1449 struct vnode *a_vp; 1450 struct thread *a_td; 1451 } */ *ap; 1452{ 1453 vnode_t *vp = ap->a_vp; 1454 cred_t *cr = ap->a_td->td_ucred; 1455 struct vop_inactive_args iap; 1456 zfsctl_snapdir_t *sdp; 1457 zfs_snapentry_t *sep, *next; 1458 int locked; 1459 vnode_t *dvp; 1460 1461 if (vp->v_count > 0) 1462 goto end; 1463 1464 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1465 sdp = dvp->v_data; 1466 VOP_UNLOCK(dvp, 0); 1467 1468 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1469 mutex_enter(&sdp->sd_lock); 1470 1471 ASSERT(!vn_ismntpt(vp)); 1472 1473 sep = avl_first(&sdp->sd_snaps); 1474 while (sep != NULL) { 1475 next = AVL_NEXT(&sdp->sd_snaps, sep); 1476 1477 if (sep->se_root == vp) { 1478 avl_remove(&sdp->sd_snaps, sep); 1479 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1480 kmem_free(sep, sizeof (zfs_snapentry_t)); 1481 break; 1482 } 1483 sep = next; 1484 } 1485 ASSERT(sep != NULL); 1486 1487 if (!locked) 1488 mutex_exit(&sdp->sd_lock); 1489 VN_RELE(dvp); 1490 1491end: 1492 /* 1493 * Dispose of the vnode for the snapshot mount point. 1494 * This is safe to do because once this entry has been removed 1495 * from the AVL tree, it can't be found again, so cannot become 1496 * "active". If we lookup the same name again we will end up 1497 * creating a new vnode. 1498 */ 1499 iap.a_vp = vp; 1500 return (gfs_vop_inactive(&iap)); 1501} 1502 1503static int 1504zfsctl_traverse_begin(vnode_t **vpp, int lktype) 1505{ 1506 1507 VN_HOLD(*vpp); 1508 /* Snapshot should be already mounted, but just in case. */ 1509 if (vn_mountedvfs(*vpp) == NULL) 1510 return (ENOENT); 1511 return (traverse(vpp, lktype)); 1512} 1513 1514static void 1515zfsctl_traverse_end(vnode_t *vp, int err) 1516{ 1517 1518 if (err == 0) 1519 vput(vp); 1520 else 1521 VN_RELE(vp); 1522} 1523 1524static int 1525zfsctl_snapshot_getattr(ap) 1526 struct vop_getattr_args /* { 1527 struct vnode *a_vp; 1528 struct vattr *a_vap; 1529 struct ucred *a_cred; 1530 } */ *ap; 1531{ 1532 vnode_t *vp = ap->a_vp; 1533 int err; 1534 1535 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1536 if (err == 0) 1537 err = VOP_GETATTR(vp, ap->a_vap, ap->a_cred); 1538 zfsctl_traverse_end(vp, err); 1539 return (err); 1540} 1541 1542static int 1543zfsctl_snapshot_fid(ap) 1544 struct vop_fid_args /* { 1545 struct vnode *a_vp; 1546 struct fid *a_fid; 1547 } */ *ap; 1548{ 1549 vnode_t *vp = ap->a_vp; 1550 int err; 1551 1552 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1553 if (err == 0) 1554 err = VOP_VPTOFH(vp, (void *)ap->a_fid); 1555 zfsctl_traverse_end(vp, err); 1556 return (err); 1557} 1558 1559static int 1560zfsctl_snapshot_lookup(ap) 1561 struct vop_lookup_args /* { 1562 struct vnode *a_dvp; 1563 struct vnode **a_vpp; 1564 struct componentname *a_cnp; 1565 } */ *ap; 1566{ 1567 vnode_t *dvp = ap->a_dvp; 1568 vnode_t **vpp = ap->a_vpp; 1569 struct componentname *cnp = ap->a_cnp; 1570 cred_t *cr = ap->a_cnp->cn_cred; 1571 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1572 int error; 1573 1574 if (cnp->cn_namelen != 2 || cnp->cn_nameptr[0] != '.' || 1575 cnp->cn_nameptr[1] != '.') { 1576 return (ENOENT); 1577 } 1578 1579 ASSERT(dvp->v_type == VDIR); 1580 ASSERT(zfsvfs->z_ctldir != NULL); 1581 1582 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", vpp, 1583 NULL, 0, NULL, cr, NULL, NULL, NULL); 1584 if (error == 0) 1585 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1586 return (error); 1587} 1588 1589static int 1590zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1591{ 1592 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1593 vnode_t *dvp, *vp; 1594 zfsctl_snapdir_t *sdp; 1595 zfs_snapentry_t *sep; 1596 int error; 1597 1598 ASSERT(zfsvfs->z_ctldir != NULL); 1599 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1600 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1601 if (error != 0) 1602 return (error); 1603 sdp = dvp->v_data; 1604 1605 mutex_enter(&sdp->sd_lock); 1606 sep = avl_first(&sdp->sd_snaps); 1607 while (sep != NULL) { 1608 vp = sep->se_root; 1609 if (vp == ap->a_vp) 1610 break; 1611 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1612 } 1613 if (sep == NULL) { 1614 mutex_exit(&sdp->sd_lock); 1615 error = ENOENT; 1616 } else { 1617 size_t len; 1618 1619 len = strlen(sep->se_name); 1620 *ap->a_buflen -= len; 1621 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1622 mutex_exit(&sdp->sd_lock); 1623 vref(dvp); 1624 *ap->a_vpp = dvp; 1625 } 1626 VN_RELE(dvp); 1627 1628 return (error); 1629} 1630 1631/* 1632 * These VP's should never see the light of day. They should always 1633 * be covered. 1634 */ 1635static struct vop_vector zfsctl_ops_snapshot = { 1636 .vop_default = &default_vnodeops, 1637 .vop_inactive = zfsctl_snapshot_inactive, 1638 .vop_lookup = zfsctl_snapshot_lookup, 1639 .vop_reclaim = zfsctl_common_reclaim, 1640 .vop_getattr = zfsctl_snapshot_getattr, 1641 .vop_fid = zfsctl_snapshot_fid, 1642 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1643}; 1644 1645int 1646zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1647{ 1648 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1649 vnode_t *dvp, *vp; 1650 zfsctl_snapdir_t *sdp; 1651 zfsctl_node_t *zcp; 1652 zfs_snapentry_t *sep; 1653 int error; 1654 1655 ASSERT(zfsvfs->z_ctldir != NULL); 1656 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1657 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1658 if (error != 0) 1659 return (error); 1660 sdp = dvp->v_data; 1661 1662 mutex_enter(&sdp->sd_lock); 1663 sep = avl_first(&sdp->sd_snaps); 1664 while (sep != NULL) { 1665 vp = sep->se_root; 1666 zcp = vp->v_data; 1667 if (zcp->zc_id == objsetid) 1668 break; 1669 1670 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1671 } 1672 1673 if (sep != NULL) { 1674 VN_HOLD(vp); 1675 /* 1676 * Return the mounted root rather than the covered mount point. 1677 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1678 * and returns the ZFS vnode mounted on top of the GFS node. 1679 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1680 */ 1681 error = traverse(&vp, LK_SHARED | LK_RETRY); 1682 if (error == 0) { 1683 if (vp == sep->se_root) 1684 error = SET_ERROR(EINVAL); 1685 else 1686 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1687 } 1688 mutex_exit(&sdp->sd_lock); 1689 if (error == 0) 1690 VN_URELE(vp); 1691 else 1692 VN_RELE(vp); 1693 } else { 1694 error = SET_ERROR(EINVAL); 1695 mutex_exit(&sdp->sd_lock); 1696 } 1697 1698 VN_RELE(dvp); 1699 1700 return (error); 1701} 1702 1703/* 1704 * Unmount any snapshots for the given filesystem. This is called from 1705 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1706 * snapshots. 1707 */ 1708int 1709zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1710{ 1711 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1712 vnode_t *dvp; 1713 zfsctl_snapdir_t *sdp; 1714 zfs_snapentry_t *sep, *next; 1715 int error; 1716 1717 ASSERT(zfsvfs->z_ctldir != NULL); 1718 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1719 NULL, 0, NULL, cr, NULL, NULL, NULL); 1720 if (error != 0) 1721 return (error); 1722 sdp = dvp->v_data; 1723 1724 mutex_enter(&sdp->sd_lock); 1725 1726 sep = avl_first(&sdp->sd_snaps); 1727 while (sep != NULL) { 1728 next = AVL_NEXT(&sdp->sd_snaps, sep); 1729 1730 /* 1731 * If this snapshot is not mounted, then it must 1732 * have just been unmounted by somebody else, and 1733 * will be cleaned up by zfsctl_snapdir_inactive(). 1734 */ 1735 if (vn_ismntpt(sep->se_root)) { 1736 error = zfsctl_unmount_snap(sep, fflags, cr); 1737 if (error) { 1738 avl_index_t where; 1739 1740 /* 1741 * Before reinserting snapshot to the tree, 1742 * check if it was actually removed. For example 1743 * when snapshot mount point is busy, we will 1744 * have an error here, but there will be no need 1745 * to reinsert snapshot. 1746 */ 1747 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1748 avl_insert(&sdp->sd_snaps, sep, where); 1749 break; 1750 } 1751 } 1752 sep = next; 1753 } 1754 1755 mutex_exit(&sdp->sd_lock); 1756 VN_RELE(dvp); 1757 1758 return (error); 1759} 1760