zfs_ctldir.c revision 248571
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) 2012 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 (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 (EACCES); 340 } else { 341#endif 342 if (accmode & VWRITE) 343 return (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 fidp->fid_len = SHORT_FID_LEN; 401 402 zfid = (zfid_short_t *)fidp; 403 404 zfid->zf_len = SHORT_FID_LEN; 405 406 for (i = 0; i < sizeof (zfid->zf_object); i++) 407 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 408 409 /* .zfs znodes always have a generation number of 0 */ 410 for (i = 0; i < sizeof (zfid->zf_gen); i++) 411 zfid->zf_gen[i] = 0; 412 413 ZFS_EXIT(zfsvfs); 414 return (0); 415} 416 417 418/*ARGSUSED*/ 419static int 420zfsctl_shares_fid(ap) 421 struct vop_fid_args /* { 422 struct vnode *a_vp; 423 struct fid *a_fid; 424 } */ *ap; 425{ 426 vnode_t *vp = ap->a_vp; 427 fid_t *fidp = (void *)ap->a_fid; 428 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 429 znode_t *dzp; 430 int error; 431 432 ZFS_ENTER(zfsvfs); 433 434 if (zfsvfs->z_shares_dir == 0) { 435 ZFS_EXIT(zfsvfs); 436 return (ENOTSUP); 437 } 438 439 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 440 error = VOP_FID(ZTOV(dzp), fidp); 441 VN_RELE(ZTOV(dzp)); 442 } 443 444 ZFS_EXIT(zfsvfs); 445 return (error); 446} 447 448static int 449zfsctl_common_reclaim(ap) 450 struct vop_reclaim_args /* { 451 struct vnode *a_vp; 452 struct thread *a_td; 453 } */ *ap; 454{ 455 vnode_t *vp = ap->a_vp; 456 457 /* 458 * Destroy the vm object and flush associated pages. 459 */ 460 vnode_destroy_vobject(vp); 461 VI_LOCK(vp); 462 vp->v_data = NULL; 463 VI_UNLOCK(vp); 464 return (0); 465} 466 467/* 468 * .zfs inode namespace 469 * 470 * We need to generate unique inode numbers for all files and directories 471 * within the .zfs pseudo-filesystem. We use the following scheme: 472 * 473 * ENTRY ZFSCTL_INODE 474 * .zfs 1 475 * .zfs/snapshot 2 476 * .zfs/snapshot/<snap> objectid(snap) 477 */ 478 479#define ZFSCTL_INO_SNAP(id) (id) 480 481/* 482 * Get root directory attributes. 483 */ 484/* ARGSUSED */ 485static int 486zfsctl_root_getattr(ap) 487 struct vop_getattr_args /* { 488 struct vnode *a_vp; 489 struct vattr *a_vap; 490 struct ucred *a_cred; 491 } */ *ap; 492{ 493 struct vnode *vp = ap->a_vp; 494 struct vattr *vap = ap->a_vap; 495 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 496 zfsctl_node_t *zcp = vp->v_data; 497 498 ZFS_ENTER(zfsvfs); 499 vap->va_nodeid = ZFSCTL_INO_ROOT; 500 vap->va_nlink = vap->va_size = NROOT_ENTRIES; 501 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime; 502 vap->va_birthtime = vap->va_ctime; 503 504 zfsctl_common_getattr(vp, vap); 505 ZFS_EXIT(zfsvfs); 506 507 return (0); 508} 509 510/* 511 * Special case the handling of "..". 512 */ 513/* ARGSUSED */ 514int 515zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, 516 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 517 int *direntflags, pathname_t *realpnp) 518{ 519 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 520 int err; 521 522 /* 523 * No extended attributes allowed under .zfs 524 */ 525 if (flags & LOOKUP_XATTR) 526 return (EINVAL); 527 528 ZFS_ENTER(zfsvfs); 529 530 if (strcmp(nm, "..") == 0) { 531 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp); 532 if (err == 0) 533 VOP_UNLOCK(*vpp, 0); 534 } else { 535 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir, 536 cr, ct, direntflags, realpnp); 537 } 538 539 ZFS_EXIT(zfsvfs); 540 541 return (err); 542} 543 544#ifdef sun 545static int 546zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 547 caller_context_t *ct) 548{ 549 /* 550 * We only care about ACL_ENABLED so that libsec can 551 * display ACL correctly and not default to POSIX draft. 552 */ 553 if (cmd == _PC_ACL_ENABLED) { 554 *valp = _ACL_ACE_ENABLED; 555 return (0); 556 } 557 558 return (fs_pathconf(vp, cmd, valp, cr, ct)); 559} 560#endif /* sun */ 561 562#ifdef sun 563static const fs_operation_def_t zfsctl_tops_root[] = { 564 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 565 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 566 { VOPNAME_IOCTL, { .error = fs_inval } }, 567 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } }, 568 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 569 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 570 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } }, 571 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 572 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 573 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } }, 574 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 575 { NULL } 576}; 577#endif /* sun */ 578 579/* 580 * Special case the handling of "..". 581 */ 582/* ARGSUSED */ 583int 584zfsctl_freebsd_root_lookup(ap) 585 struct vop_lookup_args /* { 586 struct vnode *a_dvp; 587 struct vnode **a_vpp; 588 struct componentname *a_cnp; 589 } */ *ap; 590{ 591 vnode_t *dvp = ap->a_dvp; 592 vnode_t **vpp = ap->a_vpp; 593 cred_t *cr = ap->a_cnp->cn_cred; 594 int flags = ap->a_cnp->cn_flags; 595 int nameiop = ap->a_cnp->cn_nameiop; 596 char nm[NAME_MAX + 1]; 597 int err; 598 599 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE)) 600 return (EOPNOTSUPP); 601 602 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 603 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 604 605 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL); 606 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) 607 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 608 return (err); 609} 610 611static struct vop_vector zfsctl_ops_root = { 612 .vop_default = &default_vnodeops, 613 .vop_open = zfsctl_common_open, 614 .vop_close = zfsctl_common_close, 615 .vop_ioctl = VOP_EINVAL, 616 .vop_getattr = zfsctl_root_getattr, 617 .vop_access = zfsctl_common_access, 618 .vop_readdir = gfs_vop_readdir, 619 .vop_lookup = zfsctl_freebsd_root_lookup, 620 .vop_inactive = gfs_vop_inactive, 621 .vop_reclaim = zfsctl_common_reclaim, 622#ifdef TODO 623 .vop_pathconf = zfsctl_pathconf, 624#endif 625 .vop_fid = zfsctl_common_fid, 626}; 627 628static int 629zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 630{ 631 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 632 633 if (snapshot_namecheck(name, NULL, NULL) != 0) 634 return (EILSEQ); 635 dmu_objset_name(os, zname); 636 if (strlen(zname) + 1 + strlen(name) >= len) 637 return (ENAMETOOLONG); 638 (void) strcat(zname, "@"); 639 (void) strcat(zname, name); 640 return (0); 641} 642 643static int 644zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 645{ 646 vnode_t *svp = sep->se_root; 647 int error; 648 649 ASSERT(vn_ismntpt(svp)); 650 651 /* this will be dropped by dounmount() */ 652 if ((error = vn_vfswlock(svp)) != 0) 653 return (error); 654 655#ifdef sun 656 VN_HOLD(svp); 657 error = dounmount(vn_mountedvfs(svp), fflags, cr); 658 if (error) { 659 VN_RELE(svp); 660 return (error); 661 } 662 663 /* 664 * We can't use VN_RELE(), as that will try to invoke 665 * zfsctl_snapdir_inactive(), which would cause us to destroy 666 * the sd_lock mutex held by our caller. 667 */ 668 ASSERT(svp->v_count == 1); 669 gfs_vop_inactive(svp, cr, NULL); 670 671 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 672 kmem_free(sep, sizeof (zfs_snapentry_t)); 673 674 return (0); 675#else /* !sun */ 676 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 677#endif /* !sun */ 678} 679 680#ifdef sun 681static void 682zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 683{ 684 avl_index_t where; 685 vfs_t *vfsp; 686 refstr_t *pathref; 687 char newpath[MAXNAMELEN]; 688 char *tail; 689 690 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 691 ASSERT(sep != NULL); 692 693 vfsp = vn_mountedvfs(sep->se_root); 694 ASSERT(vfsp != NULL); 695 696 vfs_lock_wait(vfsp); 697 698 /* 699 * Change the name in the AVL tree. 700 */ 701 avl_remove(&sdp->sd_snaps, sep); 702 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 703 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 704 (void) strcpy(sep->se_name, nm); 705 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 706 avl_insert(&sdp->sd_snaps, sep, where); 707 708 /* 709 * Change the current mountpoint info: 710 * - update the tail of the mntpoint path 711 * - update the tail of the resource path 712 */ 713 pathref = vfs_getmntpoint(vfsp); 714 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 715 VERIFY((tail = strrchr(newpath, '/')) != NULL); 716 *(tail+1) = '\0'; 717 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 718 (void) strcat(newpath, nm); 719 refstr_rele(pathref); 720 vfs_setmntpoint(vfsp, newpath, 0); 721 722 pathref = vfs_getresource(vfsp); 723 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 724 VERIFY((tail = strrchr(newpath, '@')) != NULL); 725 *(tail+1) = '\0'; 726 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 727 (void) strcat(newpath, nm); 728 refstr_rele(pathref); 729 vfs_setresource(vfsp, newpath, 0); 730 731 vfs_unlock(vfsp); 732} 733#endif /* sun */ 734 735#ifdef sun 736/*ARGSUSED*/ 737static int 738zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 739 cred_t *cr, caller_context_t *ct, int flags) 740{ 741 zfsctl_snapdir_t *sdp = sdvp->v_data; 742 zfs_snapentry_t search, *sep; 743 zfsvfs_t *zfsvfs; 744 avl_index_t where; 745 char from[MAXNAMELEN], to[MAXNAMELEN]; 746 char real[MAXNAMELEN], fsname[MAXNAMELEN]; 747 int err; 748 749 zfsvfs = sdvp->v_vfsp->vfs_data; 750 ZFS_ENTER(zfsvfs); 751 752 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 753 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 754 MAXNAMELEN, NULL); 755 if (err == 0) { 756 snm = real; 757 } else if (err != ENOTSUP) { 758 ZFS_EXIT(zfsvfs); 759 return (err); 760 } 761 } 762 763 ZFS_EXIT(zfsvfs); 764 765 dmu_objset_name(zfsvfs->z_os, fsname); 766 767 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 768 if (err == 0) 769 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 770 if (err == 0) 771 err = zfs_secpolicy_rename_perms(from, to, cr); 772 if (err != 0) 773 return (err); 774 775 /* 776 * Cannot move snapshots out of the snapdir. 777 */ 778 if (sdvp != tdvp) 779 return (EINVAL); 780 781 if (strcmp(snm, tnm) == 0) 782 return (0); 783 784 mutex_enter(&sdp->sd_lock); 785 786 search.se_name = (char *)snm; 787 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 788 mutex_exit(&sdp->sd_lock); 789 return (ENOENT); 790 } 791 792 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0); 793 if (err == 0) 794 zfsctl_rename_snap(sdp, sep, tnm); 795 796 mutex_exit(&sdp->sd_lock); 797 798 return (err); 799} 800#endif /* sun */ 801 802#ifdef sun 803/* ARGSUSED */ 804static int 805zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 806 caller_context_t *ct, int flags) 807{ 808 zfsctl_snapdir_t *sdp = dvp->v_data; 809 zfs_snapentry_t *sep; 810 zfs_snapentry_t search; 811 zfsvfs_t *zfsvfs; 812 char snapname[MAXNAMELEN]; 813 char real[MAXNAMELEN]; 814 int err; 815 816 zfsvfs = dvp->v_vfsp->vfs_data; 817 ZFS_ENTER(zfsvfs); 818 819 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 820 821 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 822 MAXNAMELEN, NULL); 823 if (err == 0) { 824 name = real; 825 } else if (err != ENOTSUP) { 826 ZFS_EXIT(zfsvfs); 827 return (err); 828 } 829 } 830 831 ZFS_EXIT(zfsvfs); 832 833 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 834 if (err == 0) 835 err = zfs_secpolicy_destroy_perms(snapname, cr); 836 if (err != 0) 837 return (err); 838 839 mutex_enter(&sdp->sd_lock); 840 841 search.se_name = name; 842 sep = avl_find(&sdp->sd_snaps, &search, NULL); 843 if (sep) { 844 avl_remove(&sdp->sd_snaps, sep); 845 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 846 if (err != 0) 847 avl_add(&sdp->sd_snaps, sep); 848 else 849 err = dsl_destroy_snapshot(snapname, B_FALSE); 850 } else { 851 err = ENOENT; 852 } 853 854 mutex_exit(&sdp->sd_lock); 855 856 return (err); 857} 858#endif /* sun */ 859 860/* 861 * This creates a snapshot under '.zfs/snapshot'. 862 */ 863/* ARGSUSED */ 864static int 865zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 866 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 867{ 868 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 869 char name[MAXNAMELEN]; 870 int err; 871 static enum symfollow follow = NO_FOLLOW; 872 static enum uio_seg seg = UIO_SYSSPACE; 873 874 if (snapshot_namecheck(dirname, NULL, NULL) != 0) 875 return (EILSEQ); 876 877 dmu_objset_name(zfsvfs->z_os, name); 878 879 *vpp = NULL; 880 881 err = zfs_secpolicy_snapshot_perms(name, cr); 882 if (err != 0) 883 return (err); 884 885 if (err == 0) { 886 err = dmu_objset_snapshot_one(name, dirname); 887 if (err != 0) 888 return (err); 889 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 890 } 891 892 return (err); 893} 894 895static int 896zfsctl_freebsd_snapdir_mkdir(ap) 897 struct vop_mkdir_args /* { 898 struct vnode *a_dvp; 899 struct vnode **a_vpp; 900 struct componentname *a_cnp; 901 struct vattr *a_vap; 902 } */ *ap; 903{ 904 905 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 906 907 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 908 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 909} 910 911/* 912 * Lookup entry point for the 'snapshot' directory. Try to open the 913 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 914 * Perform a mount of the associated dataset on top of the vnode. 915 */ 916/* ARGSUSED */ 917int 918zfsctl_snapdir_lookup(ap) 919 struct vop_lookup_args /* { 920 struct vnode *a_dvp; 921 struct vnode **a_vpp; 922 struct componentname *a_cnp; 923 } */ *ap; 924{ 925 vnode_t *dvp = ap->a_dvp; 926 vnode_t **vpp = ap->a_vpp; 927 struct componentname *cnp = ap->a_cnp; 928 char nm[NAME_MAX + 1]; 929 zfsctl_snapdir_t *sdp = dvp->v_data; 930 objset_t *snap; 931 char snapname[MAXNAMELEN]; 932 char real[MAXNAMELEN]; 933 char *mountpoint; 934 zfs_snapentry_t *sep, search; 935 size_t mountpoint_len; 936 avl_index_t where; 937 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 938 int err; 939 int flags = 0; 940 941 /* 942 * No extended attributes allowed under .zfs 943 */ 944 if (flags & LOOKUP_XATTR) 945 return (EINVAL); 946 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 947 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 948 949 ASSERT(dvp->v_type == VDIR); 950 951 *vpp = NULL; 952 953 /* 954 * If we get a recursive call, that means we got called 955 * from the domount() code while it was trying to look up the 956 * spec (which looks like a local path for zfs). We need to 957 * add some flag to domount() to tell it not to do this lookup. 958 */ 959 if (MUTEX_HELD(&sdp->sd_lock)) 960 return (ENOENT); 961 962 ZFS_ENTER(zfsvfs); 963 964 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 965 ZFS_EXIT(zfsvfs); 966 return (0); 967 } 968 969 if (flags & FIGNORECASE) { 970 boolean_t conflict = B_FALSE; 971 972 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 973 MAXNAMELEN, &conflict); 974 if (err == 0) { 975 strlcpy(nm, real, sizeof(nm)); 976 } else if (err != ENOTSUP) { 977 ZFS_EXIT(zfsvfs); 978 return (err); 979 } 980#if 0 981 if (realpnp) 982 (void) strlcpy(realpnp->pn_buf, nm, 983 realpnp->pn_bufsize); 984 if (conflict && direntflags) 985 *direntflags = ED_CASE_CONFLICT; 986#endif 987 } 988 989 mutex_enter(&sdp->sd_lock); 990 search.se_name = (char *)nm; 991 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 992 *vpp = sep->se_root; 993 VN_HOLD(*vpp); 994 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 995 if (err != 0) { 996 VN_RELE(*vpp); 997 *vpp = NULL; 998 } else if (*vpp == sep->se_root) { 999 /* 1000 * The snapshot was unmounted behind our backs, 1001 * try to remount it. 1002 */ 1003 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1004 goto domount; 1005 } else { 1006 /* 1007 * VROOT was set during the traverse call. We need 1008 * to clear it since we're pretending to be part 1009 * of our parent's vfs. 1010 */ 1011 (*vpp)->v_flag &= ~VROOT; 1012 } 1013 mutex_exit(&sdp->sd_lock); 1014 ZFS_EXIT(zfsvfs); 1015 return (err); 1016 } 1017 1018 /* 1019 * The requested snapshot is not currently mounted, look it up. 1020 */ 1021 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1022 if (err != 0) { 1023 mutex_exit(&sdp->sd_lock); 1024 ZFS_EXIT(zfsvfs); 1025 /* 1026 * handle "ls *" or "?" in a graceful manner, 1027 * forcing EILSEQ to ENOENT. 1028 * Since shell ultimately passes "*" or "?" as name to lookup 1029 */ 1030 return (err == EILSEQ ? ENOENT : err); 1031 } 1032 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1033 mutex_exit(&sdp->sd_lock); 1034 /* Translate errors and add SAVENAME when needed. */ 1035 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1036 err = EJUSTRETURN; 1037 cnp->cn_flags |= SAVENAME; 1038 } else { 1039 err = ENOENT; 1040 } 1041 ZFS_EXIT(zfsvfs); 1042 return (err); 1043 } 1044 1045 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1046 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1047 (void) strcpy(sep->se_name, nm); 1048 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1049 VN_HOLD(*vpp); 1050 avl_insert(&sdp->sd_snaps, sep, where); 1051 1052 dmu_objset_rele(snap, FTAG); 1053domount: 1054 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1055 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1056 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1057 (void) snprintf(mountpoint, mountpoint_len, 1058 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1059 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1060 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1061 kmem_free(mountpoint, mountpoint_len); 1062 if (err == 0) { 1063 /* 1064 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1065 * 1066 * This is where we lie about our v_vfsp in order to 1067 * make .zfs/snapshot/<snapname> accessible over NFS 1068 * without requiring manual mounts of <snapname>. 1069 */ 1070 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1071 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1072 } 1073 mutex_exit(&sdp->sd_lock); 1074 ZFS_EXIT(zfsvfs); 1075 1076#ifdef illumos 1077 /* 1078 * If we had an error, drop our hold on the vnode and 1079 * zfsctl_snapshot_inactive() will clean up. 1080 */ 1081 if (err != 0) { 1082 VN_RELE(*vpp); 1083 *vpp = NULL; 1084 } 1085#else 1086 if (err != 0) 1087 *vpp = NULL; 1088#endif 1089 return (err); 1090} 1091 1092/* ARGSUSED */ 1093int 1094zfsctl_shares_lookup(ap) 1095 struct vop_lookup_args /* { 1096 struct vnode *a_dvp; 1097 struct vnode **a_vpp; 1098 struct componentname *a_cnp; 1099 } */ *ap; 1100{ 1101 vnode_t *dvp = ap->a_dvp; 1102 vnode_t **vpp = ap->a_vpp; 1103 struct componentname *cnp = ap->a_cnp; 1104 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1105 char nm[NAME_MAX + 1]; 1106 znode_t *dzp; 1107 int error; 1108 1109 ZFS_ENTER(zfsvfs); 1110 1111 ASSERT(cnp->cn_namelen < sizeof(nm)); 1112 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1113 1114 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1115 ZFS_EXIT(zfsvfs); 1116 return (0); 1117 } 1118 1119 if (zfsvfs->z_shares_dir == 0) { 1120 ZFS_EXIT(zfsvfs); 1121 return (ENOTSUP); 1122 } 1123 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) 1124 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1125 1126 VN_RELE(ZTOV(dzp)); 1127 ZFS_EXIT(zfsvfs); 1128 1129 return (error); 1130} 1131 1132/* ARGSUSED */ 1133static int 1134zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1135 offset_t *offp, offset_t *nextp, void *data, int flags) 1136{ 1137 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1138 char snapname[MAXNAMELEN]; 1139 uint64_t id, cookie; 1140 boolean_t case_conflict; 1141 int error; 1142 1143 ZFS_ENTER(zfsvfs); 1144 1145 cookie = *offp; 1146 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1147 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1148 &cookie, &case_conflict); 1149 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1150 if (error) { 1151 ZFS_EXIT(zfsvfs); 1152 if (error == ENOENT) { 1153 *eofp = 1; 1154 return (0); 1155 } 1156 return (error); 1157 } 1158 1159 if (flags & V_RDDIR_ENTFLAGS) { 1160 edirent_t *eodp = dp; 1161 1162 (void) strcpy(eodp->ed_name, snapname); 1163 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1164 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1165 } else { 1166 struct dirent64 *odp = dp; 1167 1168 (void) strcpy(odp->d_name, snapname); 1169 odp->d_ino = ZFSCTL_INO_SNAP(id); 1170 } 1171 *nextp = cookie; 1172 1173 ZFS_EXIT(zfsvfs); 1174 1175 return (0); 1176} 1177 1178/* ARGSUSED */ 1179static int 1180zfsctl_shares_readdir(ap) 1181 struct vop_readdir_args /* { 1182 struct vnode *a_vp; 1183 struct uio *a_uio; 1184 struct ucred *a_cred; 1185 int *a_eofflag; 1186 int *a_ncookies; 1187 u_long **a_cookies; 1188 } */ *ap; 1189{ 1190 vnode_t *vp = ap->a_vp; 1191 uio_t *uiop = ap->a_uio; 1192 cred_t *cr = ap->a_cred; 1193 int *eofp = ap->a_eofflag; 1194 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1195 znode_t *dzp; 1196 int error; 1197 1198 ZFS_ENTER(zfsvfs); 1199 1200 if (zfsvfs->z_shares_dir == 0) { 1201 ZFS_EXIT(zfsvfs); 1202 return (ENOTSUP); 1203 } 1204 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1205 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1206 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1207 VN_URELE(ZTOV(dzp)); 1208 } else { 1209 *eofp = 1; 1210 error = ENOENT; 1211 } 1212 1213 ZFS_EXIT(zfsvfs); 1214 return (error); 1215} 1216 1217/* 1218 * pvp is the '.zfs' directory (zfsctl_node_t). 1219 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1220 * 1221 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1222 * when a lookup is performed on .zfs for "snapshot". 1223 */ 1224vnode_t * 1225zfsctl_mknode_snapdir(vnode_t *pvp) 1226{ 1227 vnode_t *vp; 1228 zfsctl_snapdir_t *sdp; 1229 1230 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1231 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1232 zfsctl_snapdir_readdir_cb, NULL); 1233 sdp = vp->v_data; 1234 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1235 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1236 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1237 avl_create(&sdp->sd_snaps, snapentry_compare, 1238 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1239 VOP_UNLOCK(vp, 0); 1240 return (vp); 1241} 1242 1243vnode_t * 1244zfsctl_mknode_shares(vnode_t *pvp) 1245{ 1246 vnode_t *vp; 1247 zfsctl_node_t *sdp; 1248 1249 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1250 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1251 NULL, NULL); 1252 sdp = vp->v_data; 1253 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1254 VOP_UNLOCK(vp, 0); 1255 return (vp); 1256 1257} 1258 1259/* ARGSUSED */ 1260static int 1261zfsctl_shares_getattr(ap) 1262 struct vop_getattr_args /* { 1263 struct vnode *a_vp; 1264 struct vattr *a_vap; 1265 struct ucred *a_cred; 1266 struct thread *a_td; 1267 } */ *ap; 1268{ 1269 vnode_t *vp = ap->a_vp; 1270 vattr_t *vap = ap->a_vap; 1271 cred_t *cr = ap->a_cred; 1272 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1273 znode_t *dzp; 1274 int error; 1275 1276 ZFS_ENTER(zfsvfs); 1277 if (zfsvfs->z_shares_dir == 0) { 1278 ZFS_EXIT(zfsvfs); 1279 return (ENOTSUP); 1280 } 1281 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1282 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1283 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1284 VN_URELE(ZTOV(dzp)); 1285 } 1286 ZFS_EXIT(zfsvfs); 1287 return (error); 1288 1289 1290} 1291 1292/* ARGSUSED */ 1293static int 1294zfsctl_snapdir_getattr(ap) 1295 struct vop_getattr_args /* { 1296 struct vnode *a_vp; 1297 struct vattr *a_vap; 1298 struct ucred *a_cred; 1299 } */ *ap; 1300{ 1301 vnode_t *vp = ap->a_vp; 1302 vattr_t *vap = ap->a_vap; 1303 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1304 zfsctl_snapdir_t *sdp = vp->v_data; 1305 1306 ZFS_ENTER(zfsvfs); 1307 zfsctl_common_getattr(vp, vap); 1308 vap->va_nodeid = gfs_file_inode(vp); 1309 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1310 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1311 vap->va_birthtime = vap->va_ctime; 1312 ZFS_EXIT(zfsvfs); 1313 1314 return (0); 1315} 1316 1317/* ARGSUSED */ 1318static int 1319zfsctl_snapdir_inactive(ap) 1320 struct vop_inactive_args /* { 1321 struct vnode *a_vp; 1322 struct thread *a_td; 1323 } */ *ap; 1324{ 1325 vnode_t *vp = ap->a_vp; 1326 zfsctl_snapdir_t *sdp = vp->v_data; 1327 zfs_snapentry_t *sep; 1328 1329 /* 1330 * On forced unmount we have to free snapshots from here. 1331 */ 1332 mutex_enter(&sdp->sd_lock); 1333 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) { 1334 avl_remove(&sdp->sd_snaps, sep); 1335 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1336 kmem_free(sep, sizeof (zfs_snapentry_t)); 1337 } 1338 mutex_exit(&sdp->sd_lock); 1339 gfs_dir_inactive(vp); 1340 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1341 mutex_destroy(&sdp->sd_lock); 1342 avl_destroy(&sdp->sd_snaps); 1343 kmem_free(sdp, sizeof (zfsctl_snapdir_t)); 1344 1345 return (0); 1346} 1347 1348#ifdef sun 1349static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1350 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1351 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1352 { VOPNAME_IOCTL, { .error = fs_inval } }, 1353 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1354 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1355 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1356 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1357 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1358 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1359 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1360 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1361 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1362 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1363 { NULL } 1364}; 1365 1366static const fs_operation_def_t zfsctl_tops_shares[] = { 1367 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1368 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1369 { VOPNAME_IOCTL, { .error = fs_inval } }, 1370 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1371 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1372 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1373 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1374 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1375 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1376 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1377 { NULL } 1378}; 1379#else /* !sun */ 1380static struct vop_vector zfsctl_ops_snapdir = { 1381 .vop_default = &default_vnodeops, 1382 .vop_open = zfsctl_common_open, 1383 .vop_close = zfsctl_common_close, 1384 .vop_ioctl = VOP_EINVAL, 1385 .vop_getattr = zfsctl_snapdir_getattr, 1386 .vop_access = zfsctl_common_access, 1387 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1388 .vop_readdir = gfs_vop_readdir, 1389 .vop_lookup = zfsctl_snapdir_lookup, 1390 .vop_inactive = zfsctl_snapdir_inactive, 1391 .vop_reclaim = zfsctl_common_reclaim, 1392 .vop_fid = zfsctl_common_fid, 1393}; 1394 1395static struct vop_vector zfsctl_ops_shares = { 1396 .vop_default = &default_vnodeops, 1397 .vop_open = zfsctl_common_open, 1398 .vop_close = zfsctl_common_close, 1399 .vop_ioctl = VOP_EINVAL, 1400 .vop_getattr = zfsctl_shares_getattr, 1401 .vop_access = zfsctl_common_access, 1402 .vop_readdir = zfsctl_shares_readdir, 1403 .vop_lookup = zfsctl_shares_lookup, 1404 .vop_inactive = gfs_vop_inactive, 1405 .vop_reclaim = zfsctl_common_reclaim, 1406 .vop_fid = zfsctl_shares_fid, 1407}; 1408#endif /* !sun */ 1409 1410/* 1411 * pvp is the GFS vnode '.zfs/snapshot'. 1412 * 1413 * This creates a GFS node under '.zfs/snapshot' representing each 1414 * snapshot. This newly created GFS node is what we mount snapshot 1415 * vfs_t's ontop of. 1416 */ 1417static vnode_t * 1418zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1419{ 1420 vnode_t *vp; 1421 zfsctl_node_t *zcp; 1422 1423 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1424 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1425 VN_HOLD(vp); 1426 zcp = vp->v_data; 1427 zcp->zc_id = objset; 1428 VOP_UNLOCK(vp, 0); 1429 1430 return (vp); 1431} 1432 1433static int 1434zfsctl_snapshot_inactive(ap) 1435 struct vop_inactive_args /* { 1436 struct vnode *a_vp; 1437 struct thread *a_td; 1438 } */ *ap; 1439{ 1440 vnode_t *vp = ap->a_vp; 1441 cred_t *cr = ap->a_td->td_ucred; 1442 struct vop_inactive_args iap; 1443 zfsctl_snapdir_t *sdp; 1444 zfs_snapentry_t *sep, *next; 1445 int locked; 1446 vnode_t *dvp; 1447 1448 if (vp->v_count > 0) 1449 goto end; 1450 1451 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1452 sdp = dvp->v_data; 1453 VOP_UNLOCK(dvp, 0); 1454 1455 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1456 mutex_enter(&sdp->sd_lock); 1457 1458 ASSERT(!vn_ismntpt(vp)); 1459 1460 sep = avl_first(&sdp->sd_snaps); 1461 while (sep != NULL) { 1462 next = AVL_NEXT(&sdp->sd_snaps, sep); 1463 1464 if (sep->se_root == vp) { 1465 avl_remove(&sdp->sd_snaps, sep); 1466 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1467 kmem_free(sep, sizeof (zfs_snapentry_t)); 1468 break; 1469 } 1470 sep = next; 1471 } 1472 ASSERT(sep != NULL); 1473 1474 if (!locked) 1475 mutex_exit(&sdp->sd_lock); 1476 VN_RELE(dvp); 1477 1478end: 1479 /* 1480 * Dispose of the vnode for the snapshot mount point. 1481 * This is safe to do because once this entry has been removed 1482 * from the AVL tree, it can't be found again, so cannot become 1483 * "active". If we lookup the same name again we will end up 1484 * creating a new vnode. 1485 */ 1486 iap.a_vp = vp; 1487 return (gfs_vop_inactive(&iap)); 1488} 1489 1490static int 1491zfsctl_traverse_begin(vnode_t **vpp, int lktype) 1492{ 1493 1494 VN_HOLD(*vpp); 1495 /* Snapshot should be already mounted, but just in case. */ 1496 if (vn_mountedvfs(*vpp) == NULL) 1497 return (ENOENT); 1498 return (traverse(vpp, lktype)); 1499} 1500 1501static void 1502zfsctl_traverse_end(vnode_t *vp, int err) 1503{ 1504 1505 if (err == 0) 1506 vput(vp); 1507 else 1508 VN_RELE(vp); 1509} 1510 1511static int 1512zfsctl_snapshot_getattr(ap) 1513 struct vop_getattr_args /* { 1514 struct vnode *a_vp; 1515 struct vattr *a_vap; 1516 struct ucred *a_cred; 1517 } */ *ap; 1518{ 1519 vnode_t *vp = ap->a_vp; 1520 int err; 1521 1522 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1523 if (err == 0) 1524 err = VOP_GETATTR(vp, ap->a_vap, ap->a_cred); 1525 zfsctl_traverse_end(vp, err); 1526 return (err); 1527} 1528 1529static int 1530zfsctl_snapshot_fid(ap) 1531 struct vop_fid_args /* { 1532 struct vnode *a_vp; 1533 struct fid *a_fid; 1534 } */ *ap; 1535{ 1536 vnode_t *vp = ap->a_vp; 1537 int err; 1538 1539 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1540 if (err == 0) 1541 err = VOP_VPTOFH(vp, (void *)ap->a_fid); 1542 zfsctl_traverse_end(vp, err); 1543 return (err); 1544} 1545 1546static int 1547zfsctl_snapshot_lookup(ap) 1548 struct vop_lookup_args /* { 1549 struct vnode *a_dvp; 1550 struct vnode **a_vpp; 1551 struct componentname *a_cnp; 1552 } */ *ap; 1553{ 1554 vnode_t *dvp = ap->a_dvp; 1555 vnode_t **vpp = ap->a_vpp; 1556 struct componentname *cnp = ap->a_cnp; 1557 cred_t *cr = ap->a_cnp->cn_cred; 1558 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1559 int error; 1560 1561 if (cnp->cn_namelen != 2 || cnp->cn_nameptr[0] != '.' || 1562 cnp->cn_nameptr[1] != '.') { 1563 return (ENOENT); 1564 } 1565 1566 ASSERT(dvp->v_type == VDIR); 1567 ASSERT(zfsvfs->z_ctldir != NULL); 1568 1569 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", vpp, 1570 NULL, 0, NULL, cr, NULL, NULL, NULL); 1571 if (error == 0) 1572 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1573 return (error); 1574} 1575 1576static int 1577zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1578{ 1579 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1580 vnode_t *dvp, *vp; 1581 zfsctl_snapdir_t *sdp; 1582 zfs_snapentry_t *sep; 1583 int error; 1584 1585 ASSERT(zfsvfs->z_ctldir != NULL); 1586 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1587 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1588 if (error != 0) 1589 return (error); 1590 sdp = dvp->v_data; 1591 1592 mutex_enter(&sdp->sd_lock); 1593 sep = avl_first(&sdp->sd_snaps); 1594 while (sep != NULL) { 1595 vp = sep->se_root; 1596 if (vp == ap->a_vp) 1597 break; 1598 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1599 } 1600 if (sep == NULL) { 1601 mutex_exit(&sdp->sd_lock); 1602 error = ENOENT; 1603 } else { 1604 size_t len; 1605 1606 len = strlen(sep->se_name); 1607 *ap->a_buflen -= len; 1608 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1609 mutex_exit(&sdp->sd_lock); 1610 vref(dvp); 1611 *ap->a_vpp = dvp; 1612 } 1613 VN_RELE(dvp); 1614 1615 return (error); 1616} 1617 1618/* 1619 * These VP's should never see the light of day. They should always 1620 * be covered. 1621 */ 1622static struct vop_vector zfsctl_ops_snapshot = { 1623 .vop_default = &default_vnodeops, 1624 .vop_inactive = zfsctl_snapshot_inactive, 1625 .vop_lookup = zfsctl_snapshot_lookup, 1626 .vop_reclaim = zfsctl_common_reclaim, 1627 .vop_getattr = zfsctl_snapshot_getattr, 1628 .vop_fid = zfsctl_snapshot_fid, 1629 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1630}; 1631 1632int 1633zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1634{ 1635 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1636 vnode_t *dvp, *vp; 1637 zfsctl_snapdir_t *sdp; 1638 zfsctl_node_t *zcp; 1639 zfs_snapentry_t *sep; 1640 int error; 1641 1642 ASSERT(zfsvfs->z_ctldir != NULL); 1643 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1644 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1645 if (error != 0) 1646 return (error); 1647 sdp = dvp->v_data; 1648 1649 mutex_enter(&sdp->sd_lock); 1650 sep = avl_first(&sdp->sd_snaps); 1651 while (sep != NULL) { 1652 vp = sep->se_root; 1653 zcp = vp->v_data; 1654 if (zcp->zc_id == objsetid) 1655 break; 1656 1657 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1658 } 1659 1660 if (sep != NULL) { 1661 VN_HOLD(vp); 1662 /* 1663 * Return the mounted root rather than the covered mount point. 1664 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1665 * and returns the ZFS vnode mounted on top of the GFS node. 1666 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1667 */ 1668 error = traverse(&vp, LK_SHARED | LK_RETRY); 1669 if (error == 0) { 1670 if (vp == sep->se_root) 1671 error = EINVAL; 1672 else 1673 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1674 } 1675 mutex_exit(&sdp->sd_lock); 1676 if (error == 0) 1677 VN_URELE(vp); 1678 else 1679 VN_RELE(vp); 1680 } else { 1681 error = EINVAL; 1682 mutex_exit(&sdp->sd_lock); 1683 } 1684 1685 VN_RELE(dvp); 1686 1687 return (error); 1688} 1689 1690/* 1691 * Unmount any snapshots for the given filesystem. This is called from 1692 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1693 * snapshots. 1694 */ 1695int 1696zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1697{ 1698 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1699 vnode_t *dvp; 1700 zfsctl_snapdir_t *sdp; 1701 zfs_snapentry_t *sep, *next; 1702 int error; 1703 1704 ASSERT(zfsvfs->z_ctldir != NULL); 1705 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1706 NULL, 0, NULL, cr, NULL, NULL, NULL); 1707 if (error != 0) 1708 return (error); 1709 sdp = dvp->v_data; 1710 1711 mutex_enter(&sdp->sd_lock); 1712 1713 sep = avl_first(&sdp->sd_snaps); 1714 while (sep != NULL) { 1715 next = AVL_NEXT(&sdp->sd_snaps, sep); 1716 1717 /* 1718 * If this snapshot is not mounted, then it must 1719 * have just been unmounted by somebody else, and 1720 * will be cleaned up by zfsctl_snapdir_inactive(). 1721 */ 1722 if (vn_ismntpt(sep->se_root)) { 1723 error = zfsctl_unmount_snap(sep, fflags, cr); 1724 if (error) { 1725 avl_index_t where; 1726 1727 /* 1728 * Before reinserting snapshot to the tree, 1729 * check if it was actually removed. For example 1730 * when snapshot mount point is busy, we will 1731 * have an error here, but there will be no need 1732 * to reinsert snapshot. 1733 */ 1734 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1735 avl_insert(&sdp->sd_snaps, sep, where); 1736 break; 1737 } 1738 } 1739 sep = next; 1740 } 1741 1742 mutex_exit(&sdp->sd_lock); 1743 VN_RELE(dvp); 1744 1745 return (error); 1746} 1747