zfs_ctldir.c revision 307122
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, 2015 by Delphix. All rights reserved. 24 * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved. 25 */ 26 27/* 28 * ZFS control directory (a.k.a. ".zfs") 29 * 30 * This directory provides a common location for all ZFS meta-objects. 31 * Currently, this is only the 'snapshot' directory, but this may expand in the 32 * future. The elements are built using the GFS primitives, as the hierarchy 33 * does not actually exist on disk. 34 * 35 * For 'snapshot', we don't want to have all snapshots always mounted, because 36 * this would take up a huge amount of space in /etc/mnttab. We have three 37 * types of objects: 38 * 39 * ctldir ------> snapshotdir -------> snapshot 40 * | 41 * | 42 * V 43 * mounted fs 44 * 45 * The 'snapshot' node contains just enough information to lookup '..' and act 46 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we 47 * perform an automount of the underlying filesystem and return the 48 * corresponding vnode. 49 * 50 * All mounts are handled automatically by the kernel, but unmounts are 51 * (currently) handled from user land. The main reason is that there is no 52 * reliable way to auto-unmount the filesystem when it's "no longer in use". 53 * When the user unmounts a filesystem, we call zfsctl_unmount(), which 54 * unmounts any snapshots within the snapshot directory. 55 * 56 * The '.zfs', '.zfs/snapshot', and all directories created under 57 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and 58 * share the same vfs_t as the head filesystem (what '.zfs' lives under). 59 * 60 * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>' 61 * (ie: snapshots) are ZFS nodes and have their own unique vfs_t. 62 * However, vnodes within these mounted on file systems have their v_vfsp 63 * fields set to the head filesystem to make NFS happy (see 64 * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t 65 * so that it cannot be freed until all snapshots have been unmounted. 66 */ 67 68#include <sys/zfs_context.h> 69#include <sys/zfs_ctldir.h> 70#include <sys/zfs_ioctl.h> 71#include <sys/zfs_vfsops.h> 72#include <sys/namei.h> 73#include <sys/gfs.h> 74#include <sys/stat.h> 75#include <sys/dmu.h> 76#include <sys/dsl_destroy.h> 77#include <sys/dsl_deleg.h> 78#include <sys/mount.h> 79#include <sys/sunddi.h> 80 81#include "zfs_namecheck.h" 82 83typedef struct zfsctl_node { 84 gfs_dir_t zc_gfs_private; 85 uint64_t zc_id; 86 timestruc_t zc_cmtime; /* ctime and mtime, always the same */ 87} zfsctl_node_t; 88 89typedef struct zfsctl_snapdir { 90 zfsctl_node_t sd_node; 91 kmutex_t sd_lock; 92 avl_tree_t sd_snaps; 93} zfsctl_snapdir_t; 94 95typedef struct { 96 char *se_name; 97 vnode_t *se_root; 98 avl_node_t se_node; 99} zfs_snapentry_t; 100 101static int 102snapentry_compare(const void *a, const void *b) 103{ 104 const zfs_snapentry_t *sa = a; 105 const zfs_snapentry_t *sb = b; 106 int ret = strcmp(sa->se_name, sb->se_name); 107 108 if (ret < 0) 109 return (-1); 110 else if (ret > 0) 111 return (1); 112 else 113 return (0); 114} 115 116#ifdef illumos 117vnodeops_t *zfsctl_ops_root; 118vnodeops_t *zfsctl_ops_snapdir; 119vnodeops_t *zfsctl_ops_snapshot; 120vnodeops_t *zfsctl_ops_shares; 121vnodeops_t *zfsctl_ops_shares_dir; 122 123static const fs_operation_def_t zfsctl_tops_root[]; 124static const fs_operation_def_t zfsctl_tops_snapdir[]; 125static const fs_operation_def_t zfsctl_tops_snapshot[]; 126static const fs_operation_def_t zfsctl_tops_shares[]; 127#else 128static struct vop_vector zfsctl_ops_root; 129static struct vop_vector zfsctl_ops_snapdir; 130static struct vop_vector zfsctl_ops_snapshot; 131static struct vop_vector zfsctl_ops_shares; 132static struct vop_vector zfsctl_ops_shares_dir; 133#endif 134 135static vnode_t *zfsctl_mknode_snapdir(vnode_t *); 136static vnode_t *zfsctl_mknode_shares(vnode_t *); 137static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset); 138static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *); 139 140#ifdef illumos 141static gfs_opsvec_t zfsctl_opsvec[] = { 142 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root }, 143 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir }, 144 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot }, 145 { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir }, 146 { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares }, 147 { NULL } 148}; 149#endif 150 151/* 152 * Root directory elements. We only have two entries 153 * snapshot and shares. 154 */ 155static gfs_dirent_t zfsctl_root_entries[] = { 156 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE }, 157 { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE }, 158 { NULL } 159}; 160 161/* include . and .. in the calculation */ 162#define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \ 163 sizeof (gfs_dirent_t)) + 1) 164 165 166/* 167 * Initialize the various GFS pieces we'll need to create and manipulate .zfs 168 * directories. This is called from the ZFS init routine, and initializes the 169 * vnode ops vectors that we'll be using. 170 */ 171void 172zfsctl_init(void) 173{ 174#ifdef illumos 175 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0); 176#endif 177} 178 179void 180zfsctl_fini(void) 181{ 182#ifdef illumos 183 /* 184 * Remove vfsctl vnode ops 185 */ 186 if (zfsctl_ops_root) 187 vn_freevnodeops(zfsctl_ops_root); 188 if (zfsctl_ops_snapdir) 189 vn_freevnodeops(zfsctl_ops_snapdir); 190 if (zfsctl_ops_snapshot) 191 vn_freevnodeops(zfsctl_ops_snapshot); 192 if (zfsctl_ops_shares) 193 vn_freevnodeops(zfsctl_ops_shares); 194 if (zfsctl_ops_shares_dir) 195 vn_freevnodeops(zfsctl_ops_shares_dir); 196 197 zfsctl_ops_root = NULL; 198 zfsctl_ops_snapdir = NULL; 199 zfsctl_ops_snapshot = NULL; 200 zfsctl_ops_shares = NULL; 201 zfsctl_ops_shares_dir = NULL; 202#endif /* illumos */ 203} 204 205boolean_t 206zfsctl_is_node(vnode_t *vp) 207{ 208 return (vn_matchops(vp, zfsctl_ops_root) || 209 vn_matchops(vp, zfsctl_ops_snapdir) || 210 vn_matchops(vp, zfsctl_ops_snapshot) || 211 vn_matchops(vp, zfsctl_ops_shares) || 212 vn_matchops(vp, zfsctl_ops_shares_dir)); 213 214} 215 216/* 217 * Return the inode number associated with the 'snapshot' or 218 * 'shares' directory. 219 */ 220/* ARGSUSED */ 221static ino64_t 222zfsctl_root_inode_cb(vnode_t *vp, int index) 223{ 224 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 225 226 ASSERT(index < 2); 227 228 if (index == 0) 229 return (ZFSCTL_INO_SNAPDIR); 230 231 return (zfsvfs->z_shares_dir); 232} 233 234/* 235 * Create the '.zfs' directory. This directory is cached as part of the VFS 236 * structure. This results in a hold on the vfs_t. The code in zfs_umount() 237 * therefore checks against a vfs_count of 2 instead of 1. This reference 238 * is removed when the ctldir is destroyed in the unmount. 239 */ 240void 241zfsctl_create(zfsvfs_t *zfsvfs) 242{ 243 vnode_t *vp, *rvp; 244 zfsctl_node_t *zcp; 245 uint64_t crtime[2]; 246 247 ASSERT(zfsvfs->z_ctldir == NULL); 248 249 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs, 250 &zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries, 251 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL); 252 zcp = vp->v_data; 253 zcp->zc_id = ZFSCTL_INO_ROOT; 254 255 VERIFY(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp) == 0); 256 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 257 &crtime, sizeof (crtime))); 258 ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime); 259 VN_URELE(rvp); 260 261 /* 262 * We're only faking the fact that we have a root of a filesystem for 263 * the sake of the GFS interfaces. Undo the flag manipulation it did 264 * for us. 265 */ 266 vp->v_vflag &= ~VV_ROOT; 267 268 zfsvfs->z_ctldir = vp; 269 270 VOP_UNLOCK(vp, 0); 271} 272 273/* 274 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted. 275 * There might still be more references if we were force unmounted, but only 276 * new zfs_inactive() calls can occur and they don't reference .zfs 277 */ 278void 279zfsctl_destroy(zfsvfs_t *zfsvfs) 280{ 281 VN_RELE(zfsvfs->z_ctldir); 282 zfsvfs->z_ctldir = NULL; 283} 284 285/* 286 * Given a root znode, retrieve the associated .zfs directory. 287 * Add a hold to the vnode and return it. 288 */ 289vnode_t * 290zfsctl_root(znode_t *zp) 291{ 292 ASSERT(zfs_has_ctldir(zp)); 293 VN_HOLD(zp->z_zfsvfs->z_ctldir); 294 return (zp->z_zfsvfs->z_ctldir); 295} 296 297static int 298zfsctl_common_print(ap) 299 struct vop_print_args /* { 300 struct vnode *a_vp; 301 } */ *ap; 302{ 303 vnode_t *vp = ap->a_vp; 304 gfs_file_t *fp = vp->v_data; 305 306 printf(" parent = %p\n", fp->gfs_parent); 307 printf(" type = %d\n", fp->gfs_type); 308 printf(" index = %d\n", fp->gfs_index); 309 printf(" ino = %ju\n", (uintmax_t)fp->gfs_ino); 310 return (0); 311} 312 313/* 314 * Common open routine. Disallow any write access. 315 */ 316/* ARGSUSED */ 317static int 318zfsctl_common_open(struct vop_open_args *ap) 319{ 320 int flags = ap->a_mode; 321 322 if (flags & FWRITE) 323 return (SET_ERROR(EACCES)); 324 325 return (0); 326} 327 328/* 329 * Common close routine. Nothing to do here. 330 */ 331/* ARGSUSED */ 332static int 333zfsctl_common_close(struct vop_close_args *ap) 334{ 335 return (0); 336} 337 338/* 339 * Common access routine. Disallow writes. 340 */ 341/* ARGSUSED */ 342static int 343zfsctl_common_access(ap) 344 struct vop_access_args /* { 345 struct vnode *a_vp; 346 accmode_t a_accmode; 347 struct ucred *a_cred; 348 struct thread *a_td; 349 } */ *ap; 350{ 351 accmode_t accmode = ap->a_accmode; 352 353#ifdef TODO 354 if (flags & V_ACE_MASK) { 355 if (accmode & ACE_ALL_WRITE_PERMS) 356 return (SET_ERROR(EACCES)); 357 } else { 358#endif 359 if (accmode & VWRITE) 360 return (SET_ERROR(EACCES)); 361#ifdef TODO 362 } 363#endif 364 365 return (0); 366} 367 368/* 369 * Common getattr function. Fill in basic information. 370 */ 371static void 372zfsctl_common_getattr(vnode_t *vp, vattr_t *vap) 373{ 374 timestruc_t now; 375 376 vap->va_uid = 0; 377 vap->va_gid = 0; 378 vap->va_rdev = 0; 379 /* 380 * We are a purely virtual object, so we have no 381 * blocksize or allocated blocks. 382 */ 383 vap->va_blksize = 0; 384 vap->va_nblocks = 0; 385 vap->va_seq = 0; 386 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 387 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP | 388 S_IROTH | S_IXOTH; 389 vap->va_type = VDIR; 390 /* 391 * We live in the now (for atime). 392 */ 393 gethrestime(&now); 394 vap->va_atime = now; 395 /* FreeBSD: Reset chflags(2) flags. */ 396 vap->va_flags = 0; 397} 398 399/*ARGSUSED*/ 400static int 401zfsctl_common_fid(ap) 402 struct vop_fid_args /* { 403 struct vnode *a_vp; 404 struct fid *a_fid; 405 } */ *ap; 406{ 407 vnode_t *vp = ap->a_vp; 408 fid_t *fidp = (void *)ap->a_fid; 409 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 410 zfsctl_node_t *zcp = vp->v_data; 411 uint64_t object = zcp->zc_id; 412 zfid_short_t *zfid; 413 int i; 414 415 ZFS_ENTER(zfsvfs); 416 417#ifdef illumos 418 if (fidp->fid_len < SHORT_FID_LEN) { 419 fidp->fid_len = SHORT_FID_LEN; 420 ZFS_EXIT(zfsvfs); 421 return (SET_ERROR(ENOSPC)); 422 } 423#endif 424 425 zfid = (zfid_short_t *)fidp; 426 427 zfid->zf_len = SHORT_FID_LEN; 428 429 for (i = 0; i < sizeof (zfid->zf_object); i++) 430 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 431 432 /* .zfs znodes always have a generation number of 0 */ 433 for (i = 0; i < sizeof (zfid->zf_gen); i++) 434 zfid->zf_gen[i] = 0; 435 436 ZFS_EXIT(zfsvfs); 437 return (0); 438} 439 440 441/*ARGSUSED*/ 442static int 443zfsctl_shares_fid(ap) 444 struct vop_fid_args /* { 445 struct vnode *a_vp; 446 struct fid *a_fid; 447 } */ *ap; 448{ 449 vnode_t *vp = ap->a_vp; 450 fid_t *fidp = (void *)ap->a_fid; 451 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 452 znode_t *dzp; 453 int error; 454 455 ZFS_ENTER(zfsvfs); 456 457 if (zfsvfs->z_shares_dir == 0) { 458 ZFS_EXIT(zfsvfs); 459 return (SET_ERROR(ENOTSUP)); 460 } 461 462 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 463 error = VOP_FID(ZTOV(dzp), fidp); 464 VN_RELE(ZTOV(dzp)); 465 } 466 467 ZFS_EXIT(zfsvfs); 468 return (error); 469} 470 471/* 472 * .zfs inode namespace 473 * 474 * We need to generate unique inode numbers for all files and directories 475 * within the .zfs pseudo-filesystem. We use the following scheme: 476 * 477 * ENTRY ZFSCTL_INODE 478 * .zfs 1 479 * .zfs/snapshot 2 480 * .zfs/snapshot/<snap> objectid(snap) 481 */ 482 483#define ZFSCTL_INO_SNAP(id) (id) 484 485/* 486 * Get root directory attributes. 487 */ 488/* ARGSUSED */ 489static int 490zfsctl_root_getattr(ap) 491 struct vop_getattr_args /* { 492 struct vnode *a_vp; 493 struct vattr *a_vap; 494 struct ucred *a_cred; 495 } */ *ap; 496{ 497 struct vnode *vp = ap->a_vp; 498 struct vattr *vap = ap->a_vap; 499 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 500 zfsctl_node_t *zcp = vp->v_data; 501 502 ZFS_ENTER(zfsvfs); 503 vap->va_nodeid = ZFSCTL_INO_ROOT; 504 vap->va_nlink = vap->va_size = NROOT_ENTRIES; 505 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime; 506 vap->va_birthtime = vap->va_ctime; 507 508 zfsctl_common_getattr(vp, vap); 509 ZFS_EXIT(zfsvfs); 510 511 return (0); 512} 513 514/* 515 * Special case the handling of "..". 516 */ 517/* ARGSUSED */ 518int 519zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, 520 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 521 int *direntflags, pathname_t *realpnp) 522{ 523 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 524 int err; 525 526 /* 527 * No extended attributes allowed under .zfs 528 */ 529 if (flags & LOOKUP_XATTR) 530 return (SET_ERROR(EINVAL)); 531 532 ZFS_ENTER(zfsvfs); 533 534 if (strcmp(nm, "..") == 0) { 535#ifdef illumos 536 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp); 537#else 538 /* 539 * NB: can not use VFS_ROOT here as it would acquire 540 * the vnode lock of the parent (root) vnode while 541 * holding the child's (.zfs) lock. 542 */ 543 znode_t *rootzp; 544 545 err = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 546 if (err == 0) 547 *vpp = ZTOV(rootzp); 548#endif 549 } else { 550 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir, 551 cr, ct, direntflags, realpnp); 552 } 553 554 ZFS_EXIT(zfsvfs); 555 556 return (err); 557} 558 559static int 560zfsctl_freebsd_root_lookup(ap) 561 struct vop_lookup_args /* { 562 struct vnode *a_dvp; 563 struct vnode **a_vpp; 564 struct componentname *a_cnp; 565 } */ *ap; 566{ 567 vnode_t *dvp = ap->a_dvp; 568 vnode_t **vpp = ap->a_vpp; 569 cred_t *cr = ap->a_cnp->cn_cred; 570 int flags = ap->a_cnp->cn_flags; 571 int lkflags = ap->a_cnp->cn_lkflags; 572 int nameiop = ap->a_cnp->cn_nameiop; 573 char nm[NAME_MAX + 1]; 574 int err; 575 576 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE)) 577 return (EOPNOTSUPP); 578 579 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 580 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 581relookup: 582 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL); 583 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) { 584 if (flags & ISDOTDOT) { 585 VOP_UNLOCK(dvp, 0); 586 err = vn_lock(*vpp, lkflags); 587 if (err != 0) { 588 vrele(*vpp); 589 *vpp = NULL; 590 } 591 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 592 } else { 593 err = vn_lock(*vpp, LK_EXCLUSIVE); 594 if (err != 0) { 595 VERIFY3S(err, ==, ENOENT); 596 goto relookup; 597 } 598 } 599 } 600 return (err); 601} 602 603static int 604zfsctl_root_print(ap) 605 struct vop_print_args /* { 606 struct vnode *a_vp; 607 } */ *ap; 608{ 609 printf(" .zfs node\n"); 610 zfsctl_common_print(ap); 611 return (0); 612} 613 614#ifdef illumos 615static int 616zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 617 caller_context_t *ct) 618{ 619 /* 620 * We only care about ACL_ENABLED so that libsec can 621 * display ACL correctly and not default to POSIX draft. 622 */ 623 if (cmd == _PC_ACL_ENABLED) { 624 *valp = _ACL_ACE_ENABLED; 625 return (0); 626 } 627 628 return (fs_pathconf(vp, cmd, valp, cr, ct)); 629} 630#endif /* illumos */ 631 632#ifdef illumos 633static const fs_operation_def_t zfsctl_tops_root[] = { 634 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 635 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 636 { VOPNAME_IOCTL, { .error = fs_inval } }, 637 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } }, 638 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 639 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 640 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } }, 641 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 642 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 643 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } }, 644 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 645 { NULL } 646}; 647#endif /* illumos */ 648 649static struct vop_vector zfsctl_ops_root = { 650 .vop_default = &default_vnodeops, 651 .vop_open = zfsctl_common_open, 652 .vop_close = zfsctl_common_close, 653 .vop_ioctl = VOP_EINVAL, 654 .vop_getattr = zfsctl_root_getattr, 655 .vop_access = zfsctl_common_access, 656 .vop_readdir = gfs_vop_readdir, 657 .vop_lookup = zfsctl_freebsd_root_lookup, 658 .vop_inactive = VOP_NULL, 659 .vop_reclaim = gfs_vop_reclaim, 660#ifdef TODO 661 .vop_pathconf = zfsctl_pathconf, 662#endif 663 .vop_fid = zfsctl_common_fid, 664 .vop_print = zfsctl_root_print, 665}; 666 667/* 668 * Gets the full dataset name that corresponds to the given snapshot name 669 * Example: 670 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1" 671 */ 672static int 673zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 674{ 675 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 676 677 if (zfs_component_namecheck(name, NULL, NULL) != 0) 678 return (SET_ERROR(EILSEQ)); 679 dmu_objset_name(os, zname); 680 if (strlen(zname) + 1 + strlen(name) >= len) 681 return (SET_ERROR(ENAMETOOLONG)); 682 (void) strcat(zname, "@"); 683 (void) strcat(zname, name); 684 return (0); 685} 686 687static int 688zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 689{ 690 vnode_t *svp = sep->se_root; 691 int error; 692 693 ASSERT(vn_ismntpt(svp)); 694 695 /* this will be dropped by dounmount() */ 696 if ((error = vn_vfswlock(svp)) != 0) 697 return (error); 698 699#ifdef illumos 700 VN_HOLD(svp); 701 error = dounmount(vn_mountedvfs(svp), fflags, cr); 702 if (error) { 703 VN_RELE(svp); 704 return (error); 705 } 706 707 /* 708 * We can't use VN_RELE(), as that will try to invoke 709 * zfsctl_snapdir_inactive(), which would cause us to destroy 710 * the sd_lock mutex held by our caller. 711 */ 712 ASSERT(svp->v_count == 1); 713 gfs_vop_reclaim(svp, cr, NULL); 714 715 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 716 kmem_free(sep, sizeof (zfs_snapentry_t)); 717 718 return (0); 719#else 720 vfs_ref(vn_mountedvfs(svp)); 721 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 722#endif 723} 724 725#ifdef illumos 726static void 727zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 728{ 729 avl_index_t where; 730 vfs_t *vfsp; 731 refstr_t *pathref; 732 char newpath[MAXNAMELEN]; 733 char *tail; 734 735 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 736 ASSERT(sep != NULL); 737 738 vfsp = vn_mountedvfs(sep->se_root); 739 ASSERT(vfsp != NULL); 740 741 vfs_lock_wait(vfsp); 742 743 /* 744 * Change the name in the AVL tree. 745 */ 746 avl_remove(&sdp->sd_snaps, sep); 747 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 748 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 749 (void) strcpy(sep->se_name, nm); 750 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 751 avl_insert(&sdp->sd_snaps, sep, where); 752 753 /* 754 * Change the current mountpoint info: 755 * - update the tail of the mntpoint path 756 * - update the tail of the resource path 757 */ 758 pathref = vfs_getmntpoint(vfsp); 759 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 760 VERIFY((tail = strrchr(newpath, '/')) != NULL); 761 *(tail+1) = '\0'; 762 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 763 (void) strcat(newpath, nm); 764 refstr_rele(pathref); 765 vfs_setmntpoint(vfsp, newpath, 0); 766 767 pathref = vfs_getresource(vfsp); 768 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 769 VERIFY((tail = strrchr(newpath, '@')) != NULL); 770 *(tail+1) = '\0'; 771 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 772 (void) strcat(newpath, nm); 773 refstr_rele(pathref); 774 vfs_setresource(vfsp, newpath, 0); 775 776 vfs_unlock(vfsp); 777} 778#endif /* illumos */ 779 780#ifdef illumos 781/*ARGSUSED*/ 782static int 783zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 784 cred_t *cr, caller_context_t *ct, int flags) 785{ 786 zfsctl_snapdir_t *sdp = sdvp->v_data; 787 zfs_snapentry_t search, *sep; 788 zfsvfs_t *zfsvfs; 789 avl_index_t where; 790 char from[ZFS_MAX_DATASET_NAME_LEN], to[ZFS_MAX_DATASET_NAME_LEN]; 791 char real[ZFS_MAX_DATASET_NAME_LEN], fsname[ZFS_MAX_DATASET_NAME_LEN]; 792 int err; 793 794 zfsvfs = sdvp->v_vfsp->vfs_data; 795 ZFS_ENTER(zfsvfs); 796 797 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 798 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 799 sizeof (real), NULL); 800 if (err == 0) { 801 snm = real; 802 } else if (err != ENOTSUP) { 803 ZFS_EXIT(zfsvfs); 804 return (err); 805 } 806 } 807 808 ZFS_EXIT(zfsvfs); 809 810 dmu_objset_name(zfsvfs->z_os, fsname); 811 812 err = zfsctl_snapshot_zname(sdvp, snm, sizeof (from), from); 813 if (err == 0) 814 err = zfsctl_snapshot_zname(tdvp, tnm, sizeof (to), to); 815 if (err == 0) 816 err = zfs_secpolicy_rename_perms(from, to, cr); 817 if (err != 0) 818 return (err); 819 820 /* 821 * Cannot move snapshots out of the snapdir. 822 */ 823 if (sdvp != tdvp) 824 return (SET_ERROR(EINVAL)); 825 826 if (strcmp(snm, tnm) == 0) 827 return (0); 828 829 mutex_enter(&sdp->sd_lock); 830 831 search.se_name = (char *)snm; 832 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 833 mutex_exit(&sdp->sd_lock); 834 return (SET_ERROR(ENOENT)); 835 } 836 837 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0); 838 if (err == 0) 839 zfsctl_rename_snap(sdp, sep, tnm); 840 841 mutex_exit(&sdp->sd_lock); 842 843 return (err); 844} 845#endif /* illumos */ 846 847#ifdef illumos 848/* ARGSUSED */ 849static int 850zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 851 caller_context_t *ct, int flags) 852{ 853 zfsctl_snapdir_t *sdp = dvp->v_data; 854 zfs_snapentry_t *sep; 855 zfs_snapentry_t search; 856 zfsvfs_t *zfsvfs; 857 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 858 char real[ZFS_MAX_DATASET_NAME_LEN]; 859 int err; 860 861 zfsvfs = dvp->v_vfsp->vfs_data; 862 ZFS_ENTER(zfsvfs); 863 864 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 865 866 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 867 sizeof (real), NULL); 868 if (err == 0) { 869 name = real; 870 } else if (err != ENOTSUP) { 871 ZFS_EXIT(zfsvfs); 872 return (err); 873 } 874 } 875 876 ZFS_EXIT(zfsvfs); 877 878 err = zfsctl_snapshot_zname(dvp, name, sizeof (snapname), snapname); 879 if (err == 0) 880 err = zfs_secpolicy_destroy_perms(snapname, cr); 881 if (err != 0) 882 return (err); 883 884 mutex_enter(&sdp->sd_lock); 885 886 search.se_name = name; 887 sep = avl_find(&sdp->sd_snaps, &search, NULL); 888 if (sep) { 889 avl_remove(&sdp->sd_snaps, sep); 890 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 891 if (err != 0) 892 avl_add(&sdp->sd_snaps, sep); 893 else 894 err = dsl_destroy_snapshot(snapname, B_FALSE); 895 } else { 896 err = SET_ERROR(ENOENT); 897 } 898 899 mutex_exit(&sdp->sd_lock); 900 901 return (err); 902} 903#endif /* illumos */ 904 905/* 906 * This creates a snapshot under '.zfs/snapshot'. 907 */ 908/* ARGSUSED */ 909static int 910zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 911 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 912{ 913 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 914 char name[ZFS_MAX_DATASET_NAME_LEN]; 915 int err; 916 static enum symfollow follow = NO_FOLLOW; 917 static enum uio_seg seg = UIO_SYSSPACE; 918 919 if (zfs_component_namecheck(dirname, NULL, NULL) != 0) 920 return (SET_ERROR(EILSEQ)); 921 922 dmu_objset_name(zfsvfs->z_os, name); 923 924 *vpp = NULL; 925 926 err = zfs_secpolicy_snapshot_perms(name, cr); 927 if (err != 0) 928 return (err); 929 930 if (err == 0) { 931 err = dmu_objset_snapshot_one(name, dirname); 932 if (err != 0) 933 return (err); 934 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 935 } 936 937 return (err); 938} 939 940static int 941zfsctl_freebsd_snapdir_mkdir(ap) 942 struct vop_mkdir_args /* { 943 struct vnode *a_dvp; 944 struct vnode **a_vpp; 945 struct componentname *a_cnp; 946 struct vattr *a_vap; 947 } */ *ap; 948{ 949 950 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 951 952 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 953 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 954} 955 956/* 957 * Lookup entry point for the 'snapshot' directory. Try to open the 958 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 959 * Perform a mount of the associated dataset on top of the vnode. 960 */ 961/* ARGSUSED */ 962int 963zfsctl_snapdir_lookup(ap) 964 struct vop_lookup_args /* { 965 struct vnode *a_dvp; 966 struct vnode **a_vpp; 967 struct componentname *a_cnp; 968 } */ *ap; 969{ 970 vnode_t *dvp = ap->a_dvp; 971 vnode_t **vpp = ap->a_vpp; 972 struct componentname *cnp = ap->a_cnp; 973 char nm[NAME_MAX + 1]; 974 zfsctl_snapdir_t *sdp = dvp->v_data; 975 objset_t *snap; 976 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 977 char real[ZFS_MAX_DATASET_NAME_LEN]; 978 char *mountpoint; 979 zfs_snapentry_t *sep, search; 980 size_t mountpoint_len; 981 avl_index_t where; 982 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 983 int err; 984 int ltype, flags = 0; 985 986 /* 987 * No extended attributes allowed under .zfs 988 */ 989 if (flags & LOOKUP_XATTR) 990 return (SET_ERROR(EINVAL)); 991 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 992 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 993 994 ASSERT(dvp->v_type == VDIR); 995 996 *vpp = NULL; 997 998 /* 999 * If we get a recursive call, that means we got called 1000 * from the domount() code while it was trying to look up the 1001 * spec (which looks like a local path for zfs). We need to 1002 * add some flag to domount() to tell it not to do this lookup. 1003 */ 1004 if (MUTEX_HELD(&sdp->sd_lock)) 1005 return (SET_ERROR(ENOENT)); 1006 1007 ZFS_ENTER(zfsvfs); 1008 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1009 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') { 1010 VOP_UNLOCK(dvp, 0); 1011 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE)); 1012 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE)); 1013 } 1014 ZFS_EXIT(zfsvfs); 1015 return (0); 1016 } 1017 1018 if (flags & FIGNORECASE) { 1019 boolean_t conflict = B_FALSE; 1020 1021 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 1022 sizeof (real), &conflict); 1023 if (err == 0) { 1024 strlcpy(nm, real, sizeof(nm)); 1025 } else if (err != ENOTSUP) { 1026 ZFS_EXIT(zfsvfs); 1027 return (err); 1028 } 1029#if 0 1030 if (realpnp) 1031 (void) strlcpy(realpnp->pn_buf, nm, 1032 realpnp->pn_bufsize); 1033 if (conflict && direntflags) 1034 *direntflags = ED_CASE_CONFLICT; 1035#endif 1036 } 1037 1038relookup: 1039 mutex_enter(&sdp->sd_lock); 1040 search.se_name = (char *)nm; 1041 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 1042 *vpp = sep->se_root; 1043 VN_HOLD(*vpp); 1044 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 1045 if (err != 0) { 1046 *vpp = NULL; 1047 } else if (*vpp == sep->se_root) { 1048 /* 1049 * The snapshot was unmounted behind our backs, 1050 * try to remount it. 1051 */ 1052 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1053 goto domount; 1054 } 1055 mutex_exit(&sdp->sd_lock); 1056 ZFS_EXIT(zfsvfs); 1057 return (err); 1058 } 1059 1060 /* 1061 * The requested snapshot is not currently mounted, look it up. 1062 */ 1063 err = zfsctl_snapshot_zname(dvp, nm, sizeof (snapname), snapname); 1064 if (err != 0) { 1065 mutex_exit(&sdp->sd_lock); 1066 ZFS_EXIT(zfsvfs); 1067 /* 1068 * handle "ls *" or "?" in a graceful manner, 1069 * forcing EILSEQ to ENOENT. 1070 * Since shell ultimately passes "*" or "?" as name to lookup 1071 */ 1072 return (err == EILSEQ ? ENOENT : err); 1073 } 1074 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1075 mutex_exit(&sdp->sd_lock); 1076#ifdef illumos 1077 ZFS_EXIT(zfsvfs); 1078 return (SET_ERROR(ENOENT)); 1079#else /* !illumos */ 1080 /* Translate errors and add SAVENAME when needed. */ 1081 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1082 err = EJUSTRETURN; 1083 cnp->cn_flags |= SAVENAME; 1084 } else { 1085 err = SET_ERROR(ENOENT); 1086 } 1087 ZFS_EXIT(zfsvfs); 1088 return (err); 1089#endif /* illumos */ 1090 } 1091 1092 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1093 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1094 (void) strcpy(sep->se_name, nm); 1095 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1096 avl_insert(&sdp->sd_snaps, sep, where); 1097 1098 dmu_objset_rele(snap, FTAG); 1099domount: 1100 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1101 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1102 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1103 (void) snprintf(mountpoint, mountpoint_len, 1104 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1105 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1106 mutex_exit(&sdp->sd_lock); 1107 1108 /* 1109 * The vnode may get reclaimed between dropping sd_lock and 1110 * getting the vnode lock. 1111 * */ 1112 err = vn_lock(*vpp, LK_EXCLUSIVE); 1113 if (err == ENOENT) 1114 goto relookup; 1115 VERIFY0(err); 1116 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1117 kmem_free(mountpoint, mountpoint_len); 1118 if (err == 0) { 1119 /* 1120 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1121 * 1122 * This is where we lie about our v_vfsp in order to 1123 * make .zfs/snapshot/<snapname> accessible over NFS 1124 * without requiring manual mounts of <snapname>. 1125 */ 1126 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1127 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1128 (*vpp)->v_flag &= ~VROOT; 1129 } 1130 ZFS_EXIT(zfsvfs); 1131 1132#ifdef illumos 1133 /* 1134 * If we had an error, drop our hold on the vnode and 1135 * zfsctl_snapshot_inactive() will clean up. 1136 */ 1137 if (err != 0) { 1138 VN_RELE(*vpp); 1139 *vpp = NULL; 1140 } 1141#else 1142 if (err != 0) 1143 *vpp = NULL; 1144#endif 1145 return (err); 1146} 1147 1148/* ARGSUSED */ 1149int 1150zfsctl_shares_lookup(ap) 1151 struct vop_lookup_args /* { 1152 struct vnode *a_dvp; 1153 struct vnode **a_vpp; 1154 struct componentname *a_cnp; 1155 } */ *ap; 1156{ 1157 vnode_t *dvp = ap->a_dvp; 1158 vnode_t **vpp = ap->a_vpp; 1159 struct componentname *cnp = ap->a_cnp; 1160 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1161 char nm[NAME_MAX + 1]; 1162 znode_t *dzp; 1163 int error; 1164 1165 ZFS_ENTER(zfsvfs); 1166 1167 ASSERT(cnp->cn_namelen < sizeof(nm)); 1168 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1169 1170 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1171 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') { 1172 VOP_UNLOCK(dvp, 0); 1173 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE)); 1174 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE)); 1175 } 1176 ZFS_EXIT(zfsvfs); 1177 return (0); 1178 } 1179 1180 if (zfsvfs->z_shares_dir == 0) { 1181 ZFS_EXIT(zfsvfs); 1182 return (SET_ERROR(ENOTSUP)); 1183 } 1184 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1185 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1186 VN_RELE(ZTOV(dzp)); 1187 } 1188 1189 ZFS_EXIT(zfsvfs); 1190 1191 return (error); 1192} 1193 1194/* ARGSUSED */ 1195static int 1196zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1197 offset_t *offp, offset_t *nextp, void *data, int flags) 1198{ 1199 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1200 char snapname[ZFS_MAX_DATASET_NAME_LEN]; 1201 uint64_t id, cookie; 1202 boolean_t case_conflict; 1203 int error; 1204 1205 ZFS_ENTER(zfsvfs); 1206 1207 cookie = *offp; 1208 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1209 error = dmu_snapshot_list_next(zfsvfs->z_os, 1210 sizeof (snapname), snapname, &id, &cookie, &case_conflict); 1211 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1212 if (error) { 1213 ZFS_EXIT(zfsvfs); 1214 if (error == ENOENT) { 1215 *eofp = 1; 1216 return (0); 1217 } 1218 return (error); 1219 } 1220 1221 if (flags & V_RDDIR_ENTFLAGS) { 1222 edirent_t *eodp = dp; 1223 1224 (void) strcpy(eodp->ed_name, snapname); 1225 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1226 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1227 } else { 1228 struct dirent64 *odp = dp; 1229 1230 (void) strcpy(odp->d_name, snapname); 1231 odp->d_ino = ZFSCTL_INO_SNAP(id); 1232 } 1233 *nextp = cookie; 1234 1235 ZFS_EXIT(zfsvfs); 1236 1237 return (0); 1238} 1239 1240/* ARGSUSED */ 1241static int 1242zfsctl_shares_readdir(ap) 1243 struct vop_readdir_args /* { 1244 struct vnode *a_vp; 1245 struct uio *a_uio; 1246 struct ucred *a_cred; 1247 int *a_eofflag; 1248 int *a_ncookies; 1249 u_long **a_cookies; 1250 } */ *ap; 1251{ 1252 vnode_t *vp = ap->a_vp; 1253 uio_t *uiop = ap->a_uio; 1254 cred_t *cr = ap->a_cred; 1255 int *eofp = ap->a_eofflag; 1256 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1257 znode_t *dzp; 1258 int error; 1259 1260 ZFS_ENTER(zfsvfs); 1261 1262 if (zfsvfs->z_shares_dir == 0) { 1263 ZFS_EXIT(zfsvfs); 1264 return (SET_ERROR(ENOTSUP)); 1265 } 1266 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1267 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1268 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1269 VN_URELE(ZTOV(dzp)); 1270 } else { 1271 *eofp = 1; 1272 error = SET_ERROR(ENOENT); 1273 } 1274 1275 ZFS_EXIT(zfsvfs); 1276 return (error); 1277} 1278 1279/* 1280 * pvp is the '.zfs' directory (zfsctl_node_t). 1281 * 1282 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1283 * 1284 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1285 * when a lookup is performed on .zfs for "snapshot". 1286 */ 1287vnode_t * 1288zfsctl_mknode_snapdir(vnode_t *pvp) 1289{ 1290 vnode_t *vp; 1291 zfsctl_snapdir_t *sdp; 1292 1293 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1294 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1295 zfsctl_snapdir_readdir_cb, NULL); 1296 sdp = vp->v_data; 1297 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1298 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1299 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1300 avl_create(&sdp->sd_snaps, snapentry_compare, 1301 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1302 VOP_UNLOCK(vp, 0); 1303 return (vp); 1304} 1305 1306vnode_t * 1307zfsctl_mknode_shares(vnode_t *pvp) 1308{ 1309 vnode_t *vp; 1310 zfsctl_node_t *sdp; 1311 1312 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1313 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1314 NULL, NULL); 1315 sdp = vp->v_data; 1316 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1317 VOP_UNLOCK(vp, 0); 1318 return (vp); 1319 1320} 1321 1322/* ARGSUSED */ 1323static int 1324zfsctl_shares_getattr(ap) 1325 struct vop_getattr_args /* { 1326 struct vnode *a_vp; 1327 struct vattr *a_vap; 1328 struct ucred *a_cred; 1329 struct thread *a_td; 1330 } */ *ap; 1331{ 1332 vnode_t *vp = ap->a_vp; 1333 vattr_t *vap = ap->a_vap; 1334 cred_t *cr = ap->a_cred; 1335 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1336 znode_t *dzp; 1337 int error; 1338 1339 ZFS_ENTER(zfsvfs); 1340 if (zfsvfs->z_shares_dir == 0) { 1341 ZFS_EXIT(zfsvfs); 1342 return (SET_ERROR(ENOTSUP)); 1343 } 1344 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1345 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1346 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1347 VN_URELE(ZTOV(dzp)); 1348 } 1349 ZFS_EXIT(zfsvfs); 1350 return (error); 1351 1352 1353} 1354 1355/* ARGSUSED */ 1356static int 1357zfsctl_snapdir_getattr(ap) 1358 struct vop_getattr_args /* { 1359 struct vnode *a_vp; 1360 struct vattr *a_vap; 1361 struct ucred *a_cred; 1362 } */ *ap; 1363{ 1364 vnode_t *vp = ap->a_vp; 1365 vattr_t *vap = ap->a_vap; 1366 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1367 zfsctl_snapdir_t *sdp = vp->v_data; 1368 1369 ZFS_ENTER(zfsvfs); 1370 zfsctl_common_getattr(vp, vap); 1371 vap->va_nodeid = gfs_file_inode(vp); 1372 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1373 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1374 vap->va_birthtime = vap->va_ctime; 1375 ZFS_EXIT(zfsvfs); 1376 1377 return (0); 1378} 1379 1380/* ARGSUSED */ 1381static int 1382zfsctl_snapdir_reclaim(ap) 1383 struct vop_reclaim_args /* { 1384 struct vnode *a_vp; 1385 struct thread *a_td; 1386 } */ *ap; 1387{ 1388 vnode_t *vp = ap->a_vp; 1389 zfsctl_snapdir_t *sdp = vp->v_data; 1390 zfs_snapentry_t *sep; 1391 1392 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1393 mutex_destroy(&sdp->sd_lock); 1394 avl_destroy(&sdp->sd_snaps); 1395 gfs_vop_reclaim(ap); 1396 1397 return (0); 1398} 1399 1400static int 1401zfsctl_shares_print(ap) 1402 struct vop_print_args /* { 1403 struct vnode *a_vp; 1404 } */ *ap; 1405{ 1406 printf(" .zfs/shares node\n"); 1407 zfsctl_common_print(ap); 1408 return (0); 1409} 1410 1411static int 1412zfsctl_snapdir_print(ap) 1413 struct vop_print_args /* { 1414 struct vnode *a_vp; 1415 } */ *ap; 1416{ 1417 vnode_t *vp = ap->a_vp; 1418 zfsctl_snapdir_t *sdp = vp->v_data; 1419 1420 printf(" .zfs/snapshot node\n"); 1421 printf(" number of children = %lu\n", avl_numnodes(&sdp->sd_snaps)); 1422 zfsctl_common_print(ap); 1423 return (0); 1424} 1425 1426#ifdef illumos 1427static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1428 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1429 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1430 { VOPNAME_IOCTL, { .error = fs_inval } }, 1431 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1432 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1433 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1434 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1435 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1436 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1437 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1438 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1439 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1440 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1441 { NULL } 1442}; 1443 1444static const fs_operation_def_t zfsctl_tops_shares[] = { 1445 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1446 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1447 { VOPNAME_IOCTL, { .error = fs_inval } }, 1448 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1449 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1450 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1451 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1452 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1453 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1454 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1455 { NULL } 1456}; 1457#else /* !illumos */ 1458static struct vop_vector zfsctl_ops_snapdir = { 1459 .vop_default = &default_vnodeops, 1460 .vop_open = zfsctl_common_open, 1461 .vop_close = zfsctl_common_close, 1462 .vop_ioctl = VOP_EINVAL, 1463 .vop_getattr = zfsctl_snapdir_getattr, 1464 .vop_access = zfsctl_common_access, 1465 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1466 .vop_readdir = gfs_vop_readdir, 1467 .vop_lookup = zfsctl_snapdir_lookup, 1468 .vop_inactive = VOP_NULL, 1469 .vop_reclaim = zfsctl_snapdir_reclaim, 1470 .vop_fid = zfsctl_common_fid, 1471 .vop_print = zfsctl_snapdir_print, 1472}; 1473 1474static struct vop_vector zfsctl_ops_shares = { 1475 .vop_default = &default_vnodeops, 1476 .vop_open = zfsctl_common_open, 1477 .vop_close = zfsctl_common_close, 1478 .vop_ioctl = VOP_EINVAL, 1479 .vop_getattr = zfsctl_shares_getattr, 1480 .vop_access = zfsctl_common_access, 1481 .vop_readdir = zfsctl_shares_readdir, 1482 .vop_lookup = zfsctl_shares_lookup, 1483 .vop_inactive = VOP_NULL, 1484 .vop_reclaim = gfs_vop_reclaim, 1485 .vop_fid = zfsctl_shares_fid, 1486 .vop_print = zfsctl_shares_print, 1487}; 1488#endif /* illumos */ 1489 1490/* 1491 * pvp is the GFS vnode '.zfs/snapshot'. 1492 * 1493 * This creates a GFS node under '.zfs/snapshot' representing each 1494 * snapshot. This newly created GFS node is what we mount snapshot 1495 * vfs_t's ontop of. 1496 */ 1497static vnode_t * 1498zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1499{ 1500 vnode_t *vp; 1501 zfsctl_node_t *zcp; 1502 1503 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1504 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1505 zcp = vp->v_data; 1506 zcp->zc_id = objset; 1507 VOP_UNLOCK(vp, 0); 1508 1509 return (vp); 1510} 1511 1512static int 1513zfsctl_snapshot_inactive(ap) 1514 struct vop_inactive_args /* { 1515 struct vnode *a_vp; 1516 struct thread *a_td; 1517 } */ *ap; 1518{ 1519 vnode_t *vp = ap->a_vp; 1520 1521 vrecycle(vp); 1522 return (0); 1523} 1524 1525static int 1526zfsctl_snapshot_reclaim(ap) 1527 struct vop_reclaim_args /* { 1528 struct vnode *a_vp; 1529 struct thread *a_td; 1530 } */ *ap; 1531{ 1532 vnode_t *vp = ap->a_vp; 1533 cred_t *cr = ap->a_td->td_ucred; 1534 zfsctl_snapdir_t *sdp; 1535 zfs_snapentry_t *sep, *next; 1536 int locked; 1537 vnode_t *dvp; 1538 1539 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1540 sdp = dvp->v_data; 1541 /* this may already have been unmounted */ 1542 if (sdp == NULL) { 1543 VN_RELE(dvp); 1544 return (0); 1545 } 1546 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1547 mutex_enter(&sdp->sd_lock); 1548 1549 ASSERT(!vn_ismntpt(vp)); 1550 1551 sep = avl_first(&sdp->sd_snaps); 1552 while (sep != NULL) { 1553 next = AVL_NEXT(&sdp->sd_snaps, sep); 1554 1555 if (sep->se_root == vp) { 1556 avl_remove(&sdp->sd_snaps, sep); 1557 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1558 kmem_free(sep, sizeof (zfs_snapentry_t)); 1559 break; 1560 } 1561 sep = next; 1562 } 1563 ASSERT(sep != NULL); 1564 1565 if (!locked) 1566 mutex_exit(&sdp->sd_lock); 1567 VN_RELE(dvp); 1568 1569 /* 1570 * Dispose of the vnode for the snapshot mount point. 1571 * This is safe to do because once this entry has been removed 1572 * from the AVL tree, it can't be found again, so cannot become 1573 * "active". If we lookup the same name again we will end up 1574 * creating a new vnode. 1575 */ 1576 gfs_vop_reclaim(ap); 1577 return (0); 1578 1579} 1580 1581static int 1582zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1583{ 1584 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1585 vnode_t *dvp, *vp; 1586 zfsctl_snapdir_t *sdp; 1587 zfs_snapentry_t *sep; 1588 int error; 1589 1590 ASSERT(zfsvfs->z_ctldir != NULL); 1591 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1592 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1593 if (error != 0) 1594 return (error); 1595 sdp = dvp->v_data; 1596 1597 mutex_enter(&sdp->sd_lock); 1598 sep = avl_first(&sdp->sd_snaps); 1599 while (sep != NULL) { 1600 vp = sep->se_root; 1601 if (vp == ap->a_vp) 1602 break; 1603 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1604 } 1605 if (sep == NULL) { 1606 mutex_exit(&sdp->sd_lock); 1607 error = ENOENT; 1608 } else { 1609 size_t len; 1610 1611 len = strlen(sep->se_name); 1612 *ap->a_buflen -= len; 1613 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1614 mutex_exit(&sdp->sd_lock); 1615 vref(dvp); 1616 *ap->a_vpp = dvp; 1617 } 1618 VN_RELE(dvp); 1619 1620 return (error); 1621} 1622 1623static int 1624zfsctl_snaphot_print(ap) 1625 struct vop_print_args /* { 1626 struct vnode *a_vp; 1627 } */ *ap; 1628{ 1629 vnode_t *vp = ap->a_vp; 1630 zfsctl_node_t *zcp = vp->v_data; 1631 1632 printf(" .zfs/snapshot/<snap> node\n"); 1633 printf(" id = %ju\n", (uintmax_t)zcp->zc_id); 1634 zfsctl_common_print(ap); 1635 return (0); 1636} 1637 1638/* 1639 * These VP's should never see the light of day. They should always 1640 * be covered. 1641 */ 1642static struct vop_vector zfsctl_ops_snapshot = { 1643 .vop_default = &default_vnodeops, 1644 .vop_inactive = zfsctl_snapshot_inactive, 1645 .vop_reclaim = zfsctl_snapshot_reclaim, 1646 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1647 .vop_print = zfsctl_snaphot_print, 1648}; 1649 1650int 1651zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1652{ 1653 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1654 vnode_t *dvp, *vp; 1655 zfsctl_snapdir_t *sdp; 1656 zfsctl_node_t *zcp; 1657 zfs_snapentry_t *sep; 1658 int error; 1659 1660 ASSERT(zfsvfs->z_ctldir != NULL); 1661 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1662 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1663 if (error != 0) 1664 return (error); 1665 sdp = dvp->v_data; 1666 1667 mutex_enter(&sdp->sd_lock); 1668 sep = avl_first(&sdp->sd_snaps); 1669 while (sep != NULL) { 1670 vp = sep->se_root; 1671 zcp = vp->v_data; 1672 if (zcp->zc_id == objsetid) 1673 break; 1674 1675 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1676 } 1677 1678 if (sep != NULL) { 1679 VN_HOLD(vp); 1680 /* 1681 * Return the mounted root rather than the covered mount point. 1682 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1683 * and returns the ZFS vnode mounted on top of the GFS node. 1684 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1685 */ 1686 error = traverse(&vp, LK_SHARED | LK_RETRY); 1687 if (error == 0) { 1688 if (vp == sep->se_root) { 1689 VN_RELE(vp); /* release covered vp */ 1690 error = SET_ERROR(EINVAL); 1691 } else { 1692 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1693 VN_URELE(vp); /* put snapshot's root vp */ 1694 } 1695 } 1696 mutex_exit(&sdp->sd_lock); 1697 } else { 1698 error = SET_ERROR(EINVAL); 1699 mutex_exit(&sdp->sd_lock); 1700 } 1701 1702 VN_RELE(dvp); 1703 1704 return (error); 1705} 1706 1707/* 1708 * Unmount any snapshots for the given filesystem. This is called from 1709 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1710 * snapshots. 1711 */ 1712int 1713zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1714{ 1715 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1716 vnode_t *dvp; 1717 zfsctl_snapdir_t *sdp; 1718 zfs_snapentry_t *sep, *next; 1719 int error; 1720 1721 ASSERT(zfsvfs->z_ctldir != NULL); 1722 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1723 NULL, 0, NULL, cr, NULL, NULL, NULL); 1724 if (error != 0) 1725 return (error); 1726 sdp = dvp->v_data; 1727 1728 mutex_enter(&sdp->sd_lock); 1729 1730 sep = avl_first(&sdp->sd_snaps); 1731 while (sep != NULL) { 1732 next = AVL_NEXT(&sdp->sd_snaps, sep); 1733 1734 /* 1735 * If this snapshot is not mounted, then it must 1736 * have just been unmounted by somebody else, and 1737 * will be cleaned up by zfsctl_snapdir_inactive(). 1738 */ 1739 if (vn_ismntpt(sep->se_root)) { 1740 error = zfsctl_unmount_snap(sep, fflags, cr); 1741 if (error) { 1742 avl_index_t where; 1743 1744 /* 1745 * Before reinserting snapshot to the tree, 1746 * check if it was actually removed. For example 1747 * when snapshot mount point is busy, we will 1748 * have an error here, but there will be no need 1749 * to reinsert snapshot. 1750 */ 1751 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1752 avl_insert(&sdp->sd_snaps, sep, where); 1753 break; 1754 } 1755 } 1756 sep = next; 1757 } 1758 1759 mutex_exit(&sdp->sd_lock); 1760 VN_RELE(dvp); 1761 1762 return (error); 1763} 1764