zfs_ctldir.c revision 302762
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2013 by Delphix. All rights reserved. 24 * 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_root_print(ap) 561 struct vop_print_args /* { 562 struct vnode *a_vp; 563 } */ *ap; 564{ 565 printf(" .zfs node\n"); 566 zfsctl_common_print(ap); 567 return (0); 568} 569 570#ifdef illumos 571static int 572zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 573 caller_context_t *ct) 574{ 575 /* 576 * We only care about ACL_ENABLED so that libsec can 577 * display ACL correctly and not default to POSIX draft. 578 */ 579 if (cmd == _PC_ACL_ENABLED) { 580 *valp = _ACL_ACE_ENABLED; 581 return (0); 582 } 583 584 return (fs_pathconf(vp, cmd, valp, cr, ct)); 585} 586#endif /* illumos */ 587 588#ifdef illumos 589static const fs_operation_def_t zfsctl_tops_root[] = { 590 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 591 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 592 { VOPNAME_IOCTL, { .error = fs_inval } }, 593 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } }, 594 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 595 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 596 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } }, 597 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 598 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 599 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } }, 600 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 601 { NULL } 602}; 603#endif /* illumos */ 604 605/* 606 * Special case the handling of "..". 607 */ 608/* ARGSUSED */ 609int 610zfsctl_freebsd_root_lookup(ap) 611 struct vop_lookup_args /* { 612 struct vnode *a_dvp; 613 struct vnode **a_vpp; 614 struct componentname *a_cnp; 615 } */ *ap; 616{ 617 vnode_t *dvp = ap->a_dvp; 618 vnode_t **vpp = ap->a_vpp; 619 cred_t *cr = ap->a_cnp->cn_cred; 620 int flags = ap->a_cnp->cn_flags; 621 int lkflags = ap->a_cnp->cn_lkflags; 622 int nameiop = ap->a_cnp->cn_nameiop; 623 char nm[NAME_MAX + 1]; 624 int err; 625 626 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE)) 627 return (EOPNOTSUPP); 628 629 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 630 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 631relookup: 632 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL); 633 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) { 634 if (flags & ISDOTDOT) { 635 VOP_UNLOCK(dvp, 0); 636 err = vn_lock(*vpp, lkflags); 637 if (err != 0) { 638 vrele(*vpp); 639 *vpp = NULL; 640 } 641 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 642 } else { 643 err = vn_lock(*vpp, LK_EXCLUSIVE); 644 if (err != 0) { 645 VERIFY3S(err, ==, ENOENT); 646 goto relookup; 647 } 648 } 649 } 650 return (err); 651} 652 653static struct vop_vector zfsctl_ops_root = { 654 .vop_default = &default_vnodeops, 655 .vop_open = zfsctl_common_open, 656 .vop_close = zfsctl_common_close, 657 .vop_ioctl = VOP_EINVAL, 658 .vop_getattr = zfsctl_root_getattr, 659 .vop_access = zfsctl_common_access, 660 .vop_readdir = gfs_vop_readdir, 661 .vop_lookup = zfsctl_freebsd_root_lookup, 662 .vop_inactive = VOP_NULL, 663 .vop_reclaim = gfs_vop_reclaim, 664#ifdef TODO 665 .vop_pathconf = zfsctl_pathconf, 666#endif 667 .vop_fid = zfsctl_common_fid, 668 .vop_print = zfsctl_root_print, 669}; 670 671/* 672 * Gets the full dataset name that corresponds to the given snapshot name 673 * Example: 674 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1" 675 */ 676static int 677zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 678{ 679 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 680 681 if (zfs_component_namecheck(name, NULL, NULL) != 0) 682 return (SET_ERROR(EILSEQ)); 683 dmu_objset_name(os, zname); 684 if (strlen(zname) + 1 + strlen(name) >= len) 685 return (SET_ERROR(ENAMETOOLONG)); 686 (void) strcat(zname, "@"); 687 (void) strcat(zname, name); 688 return (0); 689} 690 691static int 692zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 693{ 694 vnode_t *svp = sep->se_root; 695 int error; 696 697 ASSERT(vn_ismntpt(svp)); 698 699 /* this will be dropped by dounmount() */ 700 if ((error = vn_vfswlock(svp)) != 0) 701 return (error); 702 703#ifdef illumos 704 VN_HOLD(svp); 705 error = dounmount(vn_mountedvfs(svp), fflags, cr); 706 if (error) { 707 VN_RELE(svp); 708 return (error); 709 } 710 711 /* 712 * We can't use VN_RELE(), as that will try to invoke 713 * zfsctl_snapdir_inactive(), which would cause us to destroy 714 * the sd_lock mutex held by our caller. 715 */ 716 ASSERT(svp->v_count == 1); 717 gfs_vop_reclaim(svp, cr, NULL); 718 719 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 720 kmem_free(sep, sizeof (zfs_snapentry_t)); 721 722 return (0); 723#else 724 vfs_ref(vn_mountedvfs(svp)); 725 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 726#endif 727} 728 729#ifdef illumos 730static void 731zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 732{ 733 avl_index_t where; 734 vfs_t *vfsp; 735 refstr_t *pathref; 736 char newpath[MAXNAMELEN]; 737 char *tail; 738 739 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 740 ASSERT(sep != NULL); 741 742 vfsp = vn_mountedvfs(sep->se_root); 743 ASSERT(vfsp != NULL); 744 745 vfs_lock_wait(vfsp); 746 747 /* 748 * Change the name in the AVL tree. 749 */ 750 avl_remove(&sdp->sd_snaps, sep); 751 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 752 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 753 (void) strcpy(sep->se_name, nm); 754 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 755 avl_insert(&sdp->sd_snaps, sep, where); 756 757 /* 758 * Change the current mountpoint info: 759 * - update the tail of the mntpoint path 760 * - update the tail of the resource path 761 */ 762 pathref = vfs_getmntpoint(vfsp); 763 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 764 VERIFY((tail = strrchr(newpath, '/')) != NULL); 765 *(tail+1) = '\0'; 766 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 767 (void) strcat(newpath, nm); 768 refstr_rele(pathref); 769 vfs_setmntpoint(vfsp, newpath, 0); 770 771 pathref = vfs_getresource(vfsp); 772 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 773 VERIFY((tail = strrchr(newpath, '@')) != NULL); 774 *(tail+1) = '\0'; 775 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 776 (void) strcat(newpath, nm); 777 refstr_rele(pathref); 778 vfs_setresource(vfsp, newpath, 0); 779 780 vfs_unlock(vfsp); 781} 782#endif /* illumos */ 783 784#ifdef illumos 785/*ARGSUSED*/ 786static int 787zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 788 cred_t *cr, caller_context_t *ct, int flags) 789{ 790 zfsctl_snapdir_t *sdp = sdvp->v_data; 791 zfs_snapentry_t search, *sep; 792 zfsvfs_t *zfsvfs; 793 avl_index_t where; 794 char from[MAXNAMELEN], to[MAXNAMELEN]; 795 char real[MAXNAMELEN], fsname[MAXNAMELEN]; 796 int err; 797 798 zfsvfs = sdvp->v_vfsp->vfs_data; 799 ZFS_ENTER(zfsvfs); 800 801 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 802 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 803 MAXNAMELEN, NULL); 804 if (err == 0) { 805 snm = real; 806 } else if (err != ENOTSUP) { 807 ZFS_EXIT(zfsvfs); 808 return (err); 809 } 810 } 811 812 ZFS_EXIT(zfsvfs); 813 814 dmu_objset_name(zfsvfs->z_os, fsname); 815 816 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 817 if (err == 0) 818 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 819 if (err == 0) 820 err = zfs_secpolicy_rename_perms(from, to, cr); 821 if (err != 0) 822 return (err); 823 824 /* 825 * Cannot move snapshots out of the snapdir. 826 */ 827 if (sdvp != tdvp) 828 return (SET_ERROR(EINVAL)); 829 830 if (strcmp(snm, tnm) == 0) 831 return (0); 832 833 mutex_enter(&sdp->sd_lock); 834 835 search.se_name = (char *)snm; 836 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 837 mutex_exit(&sdp->sd_lock); 838 return (SET_ERROR(ENOENT)); 839 } 840 841 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0); 842 if (err == 0) 843 zfsctl_rename_snap(sdp, sep, tnm); 844 845 mutex_exit(&sdp->sd_lock); 846 847 return (err); 848} 849#endif /* illumos */ 850 851#ifdef illumos 852/* ARGSUSED */ 853static int 854zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 855 caller_context_t *ct, int flags) 856{ 857 zfsctl_snapdir_t *sdp = dvp->v_data; 858 zfs_snapentry_t *sep; 859 zfs_snapentry_t search; 860 zfsvfs_t *zfsvfs; 861 char snapname[MAXNAMELEN]; 862 char real[MAXNAMELEN]; 863 int err; 864 865 zfsvfs = dvp->v_vfsp->vfs_data; 866 ZFS_ENTER(zfsvfs); 867 868 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 869 870 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 871 MAXNAMELEN, NULL); 872 if (err == 0) { 873 name = real; 874 } else if (err != ENOTSUP) { 875 ZFS_EXIT(zfsvfs); 876 return (err); 877 } 878 } 879 880 ZFS_EXIT(zfsvfs); 881 882 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 883 if (err == 0) 884 err = zfs_secpolicy_destroy_perms(snapname, cr); 885 if (err != 0) 886 return (err); 887 888 mutex_enter(&sdp->sd_lock); 889 890 search.se_name = name; 891 sep = avl_find(&sdp->sd_snaps, &search, NULL); 892 if (sep) { 893 avl_remove(&sdp->sd_snaps, sep); 894 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 895 if (err != 0) 896 avl_add(&sdp->sd_snaps, sep); 897 else 898 err = dsl_destroy_snapshot(snapname, B_FALSE); 899 } else { 900 err = SET_ERROR(ENOENT); 901 } 902 903 mutex_exit(&sdp->sd_lock); 904 905 return (err); 906} 907#endif /* illumos */ 908 909/* 910 * This creates a snapshot under '.zfs/snapshot'. 911 */ 912/* ARGSUSED */ 913static int 914zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 915 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 916{ 917 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 918 char name[MAXNAMELEN]; 919 int err; 920 static enum symfollow follow = NO_FOLLOW; 921 static enum uio_seg seg = UIO_SYSSPACE; 922 923 if (zfs_component_namecheck(dirname, NULL, NULL) != 0) 924 return (SET_ERROR(EILSEQ)); 925 926 dmu_objset_name(zfsvfs->z_os, name); 927 928 *vpp = NULL; 929 930 err = zfs_secpolicy_snapshot_perms(name, cr); 931 if (err != 0) 932 return (err); 933 934 if (err == 0) { 935 err = dmu_objset_snapshot_one(name, dirname); 936 if (err != 0) 937 return (err); 938 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 939 } 940 941 return (err); 942} 943 944static int 945zfsctl_freebsd_snapdir_mkdir(ap) 946 struct vop_mkdir_args /* { 947 struct vnode *a_dvp; 948 struct vnode **a_vpp; 949 struct componentname *a_cnp; 950 struct vattr *a_vap; 951 } */ *ap; 952{ 953 954 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 955 956 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 957 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 958} 959 960/* 961 * Lookup entry point for the 'snapshot' directory. Try to open the 962 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 963 * Perform a mount of the associated dataset on top of the vnode. 964 */ 965/* ARGSUSED */ 966int 967zfsctl_snapdir_lookup(ap) 968 struct vop_lookup_args /* { 969 struct vnode *a_dvp; 970 struct vnode **a_vpp; 971 struct componentname *a_cnp; 972 } */ *ap; 973{ 974 vnode_t *dvp = ap->a_dvp; 975 vnode_t **vpp = ap->a_vpp; 976 struct componentname *cnp = ap->a_cnp; 977 char nm[NAME_MAX + 1]; 978 zfsctl_snapdir_t *sdp = dvp->v_data; 979 objset_t *snap; 980 char snapname[MAXNAMELEN]; 981 char real[MAXNAMELEN]; 982 char *mountpoint; 983 zfs_snapentry_t *sep, search; 984 size_t mountpoint_len; 985 avl_index_t where; 986 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 987 int err; 988 int ltype, flags = 0; 989 990 /* 991 * No extended attributes allowed under .zfs 992 */ 993 if (flags & LOOKUP_XATTR) 994 return (SET_ERROR(EINVAL)); 995 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 996 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 997 998 ASSERT(dvp->v_type == VDIR); 999 1000 *vpp = NULL; 1001 1002 /* 1003 * If we get a recursive call, that means we got called 1004 * from the domount() code while it was trying to look up the 1005 * spec (which looks like a local path for zfs). We need to 1006 * add some flag to domount() to tell it not to do this lookup. 1007 */ 1008 if (MUTEX_HELD(&sdp->sd_lock)) 1009 return (SET_ERROR(ENOENT)); 1010 1011 ZFS_ENTER(zfsvfs); 1012 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1013 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') { 1014 VOP_UNLOCK(dvp, 0); 1015 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE)); 1016 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE)); 1017 } 1018 ZFS_EXIT(zfsvfs); 1019 return (0); 1020 } 1021 1022 if (flags & FIGNORECASE) { 1023 boolean_t conflict = B_FALSE; 1024 1025 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 1026 MAXNAMELEN, &conflict); 1027 if (err == 0) { 1028 strlcpy(nm, real, sizeof(nm)); 1029 } else if (err != ENOTSUP) { 1030 ZFS_EXIT(zfsvfs); 1031 return (err); 1032 } 1033#if 0 1034 if (realpnp) 1035 (void) strlcpy(realpnp->pn_buf, nm, 1036 realpnp->pn_bufsize); 1037 if (conflict && direntflags) 1038 *direntflags = ED_CASE_CONFLICT; 1039#endif 1040 } 1041 1042relookup: 1043 mutex_enter(&sdp->sd_lock); 1044 search.se_name = (char *)nm; 1045 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 1046 *vpp = sep->se_root; 1047 VN_HOLD(*vpp); 1048 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 1049 if (err != 0) { 1050 *vpp = NULL; 1051 } else if (*vpp == sep->se_root) { 1052 /* 1053 * The snapshot was unmounted behind our backs, 1054 * try to remount it. 1055 */ 1056 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1057 goto domount; 1058 } 1059 mutex_exit(&sdp->sd_lock); 1060 ZFS_EXIT(zfsvfs); 1061 return (err); 1062 } 1063 1064 /* 1065 * The requested snapshot is not currently mounted, look it up. 1066 */ 1067 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1068 if (err != 0) { 1069 mutex_exit(&sdp->sd_lock); 1070 ZFS_EXIT(zfsvfs); 1071 /* 1072 * handle "ls *" or "?" in a graceful manner, 1073 * forcing EILSEQ to ENOENT. 1074 * Since shell ultimately passes "*" or "?" as name to lookup 1075 */ 1076 return (err == EILSEQ ? ENOENT : err); 1077 } 1078 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1079 mutex_exit(&sdp->sd_lock); 1080#ifdef illumos 1081 ZFS_EXIT(zfsvfs); 1082 return (SET_ERROR(ENOENT)); 1083#else /* !illumos */ 1084 /* Translate errors and add SAVENAME when needed. */ 1085 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1086 err = EJUSTRETURN; 1087 cnp->cn_flags |= SAVENAME; 1088 } else { 1089 err = SET_ERROR(ENOENT); 1090 } 1091 ZFS_EXIT(zfsvfs); 1092 return (err); 1093#endif /* illumos */ 1094 } 1095 1096 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1097 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1098 (void) strcpy(sep->se_name, nm); 1099 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1100 avl_insert(&sdp->sd_snaps, sep, where); 1101 1102 dmu_objset_rele(snap, FTAG); 1103domount: 1104 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1105 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1106 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1107 (void) snprintf(mountpoint, mountpoint_len, 1108 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1109 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1110 mutex_exit(&sdp->sd_lock); 1111 1112 /* 1113 * The vnode may get reclaimed between dropping sd_lock and 1114 * getting the vnode lock. 1115 * */ 1116 err = vn_lock(*vpp, LK_EXCLUSIVE); 1117 if (err == ENOENT) 1118 goto relookup; 1119 VERIFY0(err); 1120 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1121 kmem_free(mountpoint, mountpoint_len); 1122 if (err == 0) { 1123 /* 1124 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1125 * 1126 * This is where we lie about our v_vfsp in order to 1127 * make .zfs/snapshot/<snapname> accessible over NFS 1128 * without requiring manual mounts of <snapname>. 1129 */ 1130 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1131 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1132 } 1133 ZFS_EXIT(zfsvfs); 1134 1135#ifdef illumos 1136 /* 1137 * If we had an error, drop our hold on the vnode and 1138 * zfsctl_snapshot_inactive() will clean up. 1139 */ 1140 if (err != 0) { 1141 VN_RELE(*vpp); 1142 *vpp = NULL; 1143 } 1144#else 1145 if (err != 0) 1146 *vpp = NULL; 1147#endif 1148 return (err); 1149} 1150 1151/* ARGSUSED */ 1152int 1153zfsctl_shares_lookup(ap) 1154 struct vop_lookup_args /* { 1155 struct vnode *a_dvp; 1156 struct vnode **a_vpp; 1157 struct componentname *a_cnp; 1158 } */ *ap; 1159{ 1160 vnode_t *dvp = ap->a_dvp; 1161 vnode_t **vpp = ap->a_vpp; 1162 struct componentname *cnp = ap->a_cnp; 1163 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1164 char nm[NAME_MAX + 1]; 1165 znode_t *dzp; 1166 int error; 1167 1168 ZFS_ENTER(zfsvfs); 1169 1170 ASSERT(cnp->cn_namelen < sizeof(nm)); 1171 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1172 1173 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1174 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') { 1175 VOP_UNLOCK(dvp, 0); 1176 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE)); 1177 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE)); 1178 } 1179 ZFS_EXIT(zfsvfs); 1180 return (0); 1181 } 1182 1183 if (zfsvfs->z_shares_dir == 0) { 1184 ZFS_EXIT(zfsvfs); 1185 return (SET_ERROR(ENOTSUP)); 1186 } 1187 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1188 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1189 VN_RELE(ZTOV(dzp)); 1190 } 1191 1192 ZFS_EXIT(zfsvfs); 1193 1194 return (error); 1195} 1196 1197/* ARGSUSED */ 1198static int 1199zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1200 offset_t *offp, offset_t *nextp, void *data, int flags) 1201{ 1202 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1203 char snapname[MAXNAMELEN]; 1204 uint64_t id, cookie; 1205 boolean_t case_conflict; 1206 int error; 1207 1208 ZFS_ENTER(zfsvfs); 1209 1210 cookie = *offp; 1211 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1212 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1213 &cookie, &case_conflict); 1214 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1215 if (error) { 1216 ZFS_EXIT(zfsvfs); 1217 if (error == ENOENT) { 1218 *eofp = 1; 1219 return (0); 1220 } 1221 return (error); 1222 } 1223 1224 if (flags & V_RDDIR_ENTFLAGS) { 1225 edirent_t *eodp = dp; 1226 1227 (void) strcpy(eodp->ed_name, snapname); 1228 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1229 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1230 } else { 1231 struct dirent64 *odp = dp; 1232 1233 (void) strcpy(odp->d_name, snapname); 1234 odp->d_ino = ZFSCTL_INO_SNAP(id); 1235 } 1236 *nextp = cookie; 1237 1238 ZFS_EXIT(zfsvfs); 1239 1240 return (0); 1241} 1242 1243/* ARGSUSED */ 1244static int 1245zfsctl_shares_readdir(ap) 1246 struct vop_readdir_args /* { 1247 struct vnode *a_vp; 1248 struct uio *a_uio; 1249 struct ucred *a_cred; 1250 int *a_eofflag; 1251 int *a_ncookies; 1252 u_long **a_cookies; 1253 } */ *ap; 1254{ 1255 vnode_t *vp = ap->a_vp; 1256 uio_t *uiop = ap->a_uio; 1257 cred_t *cr = ap->a_cred; 1258 int *eofp = ap->a_eofflag; 1259 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1260 znode_t *dzp; 1261 int error; 1262 1263 ZFS_ENTER(zfsvfs); 1264 1265 if (zfsvfs->z_shares_dir == 0) { 1266 ZFS_EXIT(zfsvfs); 1267 return (SET_ERROR(ENOTSUP)); 1268 } 1269 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1270 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1271 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1272 VN_URELE(ZTOV(dzp)); 1273 } else { 1274 *eofp = 1; 1275 error = SET_ERROR(ENOENT); 1276 } 1277 1278 ZFS_EXIT(zfsvfs); 1279 return (error); 1280} 1281 1282/* 1283 * pvp is the '.zfs' directory (zfsctl_node_t). 1284 * 1285 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1286 * 1287 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1288 * when a lookup is performed on .zfs for "snapshot". 1289 */ 1290vnode_t * 1291zfsctl_mknode_snapdir(vnode_t *pvp) 1292{ 1293 vnode_t *vp; 1294 zfsctl_snapdir_t *sdp; 1295 1296 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1297 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1298 zfsctl_snapdir_readdir_cb, NULL); 1299 sdp = vp->v_data; 1300 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1301 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1302 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1303 avl_create(&sdp->sd_snaps, snapentry_compare, 1304 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1305 VOP_UNLOCK(vp, 0); 1306 return (vp); 1307} 1308 1309vnode_t * 1310zfsctl_mknode_shares(vnode_t *pvp) 1311{ 1312 vnode_t *vp; 1313 zfsctl_node_t *sdp; 1314 1315 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1316 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1317 NULL, NULL); 1318 sdp = vp->v_data; 1319 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1320 VOP_UNLOCK(vp, 0); 1321 return (vp); 1322 1323} 1324 1325/* ARGSUSED */ 1326static int 1327zfsctl_shares_getattr(ap) 1328 struct vop_getattr_args /* { 1329 struct vnode *a_vp; 1330 struct vattr *a_vap; 1331 struct ucred *a_cred; 1332 struct thread *a_td; 1333 } */ *ap; 1334{ 1335 vnode_t *vp = ap->a_vp; 1336 vattr_t *vap = ap->a_vap; 1337 cred_t *cr = ap->a_cred; 1338 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1339 znode_t *dzp; 1340 int error; 1341 1342 ZFS_ENTER(zfsvfs); 1343 if (zfsvfs->z_shares_dir == 0) { 1344 ZFS_EXIT(zfsvfs); 1345 return (SET_ERROR(ENOTSUP)); 1346 } 1347 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1348 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1349 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1350 VN_URELE(ZTOV(dzp)); 1351 } 1352 ZFS_EXIT(zfsvfs); 1353 return (error); 1354 1355 1356} 1357 1358/* ARGSUSED */ 1359static int 1360zfsctl_snapdir_getattr(ap) 1361 struct vop_getattr_args /* { 1362 struct vnode *a_vp; 1363 struct vattr *a_vap; 1364 struct ucred *a_cred; 1365 } */ *ap; 1366{ 1367 vnode_t *vp = ap->a_vp; 1368 vattr_t *vap = ap->a_vap; 1369 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1370 zfsctl_snapdir_t *sdp = vp->v_data; 1371 1372 ZFS_ENTER(zfsvfs); 1373 zfsctl_common_getattr(vp, vap); 1374 vap->va_nodeid = gfs_file_inode(vp); 1375 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1376 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1377 vap->va_birthtime = vap->va_ctime; 1378 ZFS_EXIT(zfsvfs); 1379 1380 return (0); 1381} 1382 1383/* ARGSUSED */ 1384static int 1385zfsctl_snapdir_reclaim(ap) 1386 struct vop_reclaim_args /* { 1387 struct vnode *a_vp; 1388 struct thread *a_td; 1389 } */ *ap; 1390{ 1391 vnode_t *vp = ap->a_vp; 1392 zfsctl_snapdir_t *sdp = vp->v_data; 1393 zfs_snapentry_t *sep; 1394 1395 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1396 mutex_destroy(&sdp->sd_lock); 1397 avl_destroy(&sdp->sd_snaps); 1398 gfs_vop_reclaim(ap); 1399 1400 return (0); 1401} 1402 1403static int 1404zfsctl_shares_print(ap) 1405 struct vop_print_args /* { 1406 struct vnode *a_vp; 1407 } */ *ap; 1408{ 1409 printf(" .zfs/shares node\n"); 1410 zfsctl_common_print(ap); 1411 return (0); 1412} 1413 1414static int 1415zfsctl_snapdir_print(ap) 1416 struct vop_print_args /* { 1417 struct vnode *a_vp; 1418 } */ *ap; 1419{ 1420 vnode_t *vp = ap->a_vp; 1421 zfsctl_snapdir_t *sdp = vp->v_data; 1422 1423 printf(" .zfs/snapshot node\n"); 1424 printf(" number of children = %lu\n", avl_numnodes(&sdp->sd_snaps)); 1425 zfsctl_common_print(ap); 1426 return (0); 1427} 1428 1429#ifdef illumos 1430static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1431 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1432 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1433 { VOPNAME_IOCTL, { .error = fs_inval } }, 1434 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1435 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1436 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1437 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1438 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1439 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1440 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1441 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1442 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1443 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1444 { NULL } 1445}; 1446 1447static const fs_operation_def_t zfsctl_tops_shares[] = { 1448 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1449 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1450 { VOPNAME_IOCTL, { .error = fs_inval } }, 1451 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1452 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1453 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1454 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1455 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1456 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1457 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1458 { NULL } 1459}; 1460#else /* !illumos */ 1461static struct vop_vector zfsctl_ops_snapdir = { 1462 .vop_default = &default_vnodeops, 1463 .vop_open = zfsctl_common_open, 1464 .vop_close = zfsctl_common_close, 1465 .vop_ioctl = VOP_EINVAL, 1466 .vop_getattr = zfsctl_snapdir_getattr, 1467 .vop_access = zfsctl_common_access, 1468 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1469 .vop_readdir = gfs_vop_readdir, 1470 .vop_lookup = zfsctl_snapdir_lookup, 1471 .vop_inactive = VOP_NULL, 1472 .vop_reclaim = zfsctl_snapdir_reclaim, 1473 .vop_fid = zfsctl_common_fid, 1474 .vop_print = zfsctl_snapdir_print, 1475}; 1476 1477static struct vop_vector zfsctl_ops_shares = { 1478 .vop_default = &default_vnodeops, 1479 .vop_open = zfsctl_common_open, 1480 .vop_close = zfsctl_common_close, 1481 .vop_ioctl = VOP_EINVAL, 1482 .vop_getattr = zfsctl_shares_getattr, 1483 .vop_access = zfsctl_common_access, 1484 .vop_readdir = zfsctl_shares_readdir, 1485 .vop_lookup = zfsctl_shares_lookup, 1486 .vop_inactive = VOP_NULL, 1487 .vop_reclaim = gfs_vop_reclaim, 1488 .vop_fid = zfsctl_shares_fid, 1489 .vop_print = zfsctl_shares_print, 1490}; 1491#endif /* illumos */ 1492 1493/* 1494 * pvp is the GFS vnode '.zfs/snapshot'. 1495 * 1496 * This creates a GFS node under '.zfs/snapshot' representing each 1497 * snapshot. This newly created GFS node is what we mount snapshot 1498 * vfs_t's ontop of. 1499 */ 1500static vnode_t * 1501zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1502{ 1503 vnode_t *vp; 1504 zfsctl_node_t *zcp; 1505 1506 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1507 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1508 zcp = vp->v_data; 1509 zcp->zc_id = objset; 1510 VOP_UNLOCK(vp, 0); 1511 1512 return (vp); 1513} 1514 1515static int 1516zfsctl_snapshot_inactive(ap) 1517 struct vop_inactive_args /* { 1518 struct vnode *a_vp; 1519 struct thread *a_td; 1520 } */ *ap; 1521{ 1522 vnode_t *vp = ap->a_vp; 1523 1524 vrecycle(vp); 1525 return (0); 1526} 1527 1528static int 1529zfsctl_snapshot_reclaim(ap) 1530 struct vop_reclaim_args /* { 1531 struct vnode *a_vp; 1532 struct thread *a_td; 1533 } */ *ap; 1534{ 1535 vnode_t *vp = ap->a_vp; 1536 cred_t *cr = ap->a_td->td_ucred; 1537 zfsctl_snapdir_t *sdp; 1538 zfs_snapentry_t *sep, *next; 1539 int locked; 1540 vnode_t *dvp; 1541 1542 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1543 sdp = dvp->v_data; 1544 /* this may already have been unmounted */ 1545 if (sdp == NULL) { 1546 VN_RELE(dvp); 1547 return (0); 1548 } 1549 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1550 mutex_enter(&sdp->sd_lock); 1551 1552 ASSERT(!vn_ismntpt(vp)); 1553 1554 sep = avl_first(&sdp->sd_snaps); 1555 while (sep != NULL) { 1556 next = AVL_NEXT(&sdp->sd_snaps, sep); 1557 1558 if (sep->se_root == vp) { 1559 avl_remove(&sdp->sd_snaps, sep); 1560 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1561 kmem_free(sep, sizeof (zfs_snapentry_t)); 1562 break; 1563 } 1564 sep = next; 1565 } 1566 ASSERT(sep != NULL); 1567 1568 if (!locked) 1569 mutex_exit(&sdp->sd_lock); 1570 VN_RELE(dvp); 1571 1572 /* 1573 * Dispose of the vnode for the snapshot mount point. 1574 * This is safe to do because once this entry has been removed 1575 * from the AVL tree, it can't be found again, so cannot become 1576 * "active". If we lookup the same name again we will end up 1577 * creating a new vnode. 1578 */ 1579 gfs_vop_reclaim(ap); 1580 return (0); 1581 1582} 1583 1584static int 1585zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1586{ 1587 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1588 vnode_t *dvp, *vp; 1589 zfsctl_snapdir_t *sdp; 1590 zfs_snapentry_t *sep; 1591 int error; 1592 1593 ASSERT(zfsvfs->z_ctldir != NULL); 1594 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1595 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1596 if (error != 0) 1597 return (error); 1598 sdp = dvp->v_data; 1599 1600 mutex_enter(&sdp->sd_lock); 1601 sep = avl_first(&sdp->sd_snaps); 1602 while (sep != NULL) { 1603 vp = sep->se_root; 1604 if (vp == ap->a_vp) 1605 break; 1606 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1607 } 1608 if (sep == NULL) { 1609 mutex_exit(&sdp->sd_lock); 1610 error = ENOENT; 1611 } else { 1612 size_t len; 1613 1614 len = strlen(sep->se_name); 1615 *ap->a_buflen -= len; 1616 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1617 mutex_exit(&sdp->sd_lock); 1618 vref(dvp); 1619 *ap->a_vpp = dvp; 1620 } 1621 VN_RELE(dvp); 1622 1623 return (error); 1624} 1625 1626static int 1627zfsctl_snaphot_print(ap) 1628 struct vop_print_args /* { 1629 struct vnode *a_vp; 1630 } */ *ap; 1631{ 1632 vnode_t *vp = ap->a_vp; 1633 zfsctl_node_t *zcp = vp->v_data; 1634 1635 printf(" .zfs/snapshot/<snap> node\n"); 1636 printf(" id = %ju\n", (uintmax_t)zcp->zc_id); 1637 zfsctl_common_print(ap); 1638 return (0); 1639} 1640 1641/* 1642 * These VP's should never see the light of day. They should always 1643 * be covered. 1644 */ 1645static struct vop_vector zfsctl_ops_snapshot = { 1646 .vop_default = &default_vnodeops, 1647 .vop_inactive = zfsctl_snapshot_inactive, 1648 .vop_reclaim = zfsctl_snapshot_reclaim, 1649 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1650 .vop_print = zfsctl_snaphot_print, 1651}; 1652 1653int 1654zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1655{ 1656 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1657 vnode_t *dvp, *vp; 1658 zfsctl_snapdir_t *sdp; 1659 zfsctl_node_t *zcp; 1660 zfs_snapentry_t *sep; 1661 int error; 1662 1663 ASSERT(zfsvfs->z_ctldir != NULL); 1664 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1665 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1666 if (error != 0) 1667 return (error); 1668 sdp = dvp->v_data; 1669 1670 mutex_enter(&sdp->sd_lock); 1671 sep = avl_first(&sdp->sd_snaps); 1672 while (sep != NULL) { 1673 vp = sep->se_root; 1674 zcp = vp->v_data; 1675 if (zcp->zc_id == objsetid) 1676 break; 1677 1678 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1679 } 1680 1681 if (sep != NULL) { 1682 VN_HOLD(vp); 1683 /* 1684 * Return the mounted root rather than the covered mount point. 1685 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1686 * and returns the ZFS vnode mounted on top of the GFS node. 1687 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1688 */ 1689 error = traverse(&vp, LK_SHARED | LK_RETRY); 1690 if (error == 0) { 1691 if (vp == sep->se_root) { 1692 VN_RELE(vp); /* release covered vp */ 1693 error = SET_ERROR(EINVAL); 1694 } else { 1695 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1696 VN_URELE(vp); /* put snapshot's root vp */ 1697 } 1698 } 1699 mutex_exit(&sdp->sd_lock); 1700 } else { 1701 error = SET_ERROR(EINVAL); 1702 mutex_exit(&sdp->sd_lock); 1703 } 1704 1705 VN_RELE(dvp); 1706 1707 return (error); 1708} 1709 1710/* 1711 * Unmount any snapshots for the given filesystem. This is called from 1712 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1713 * snapshots. 1714 */ 1715int 1716zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1717{ 1718 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1719 vnode_t *dvp; 1720 zfsctl_snapdir_t *sdp; 1721 zfs_snapentry_t *sep, *next; 1722 int error; 1723 1724 ASSERT(zfsvfs->z_ctldir != NULL); 1725 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1726 NULL, 0, NULL, cr, NULL, NULL, NULL); 1727 if (error != 0) 1728 return (error); 1729 sdp = dvp->v_data; 1730 1731 mutex_enter(&sdp->sd_lock); 1732 1733 sep = avl_first(&sdp->sd_snaps); 1734 while (sep != NULL) { 1735 next = AVL_NEXT(&sdp->sd_snaps, sep); 1736 1737 /* 1738 * If this snapshot is not mounted, then it must 1739 * have just been unmounted by somebody else, and 1740 * will be cleaned up by zfsctl_snapdir_inactive(). 1741 */ 1742 if (vn_ismntpt(sep->se_root)) { 1743 error = zfsctl_unmount_snap(sep, fflags, cr); 1744 if (error) { 1745 avl_index_t where; 1746 1747 /* 1748 * Before reinserting snapshot to the tree, 1749 * check if it was actually removed. For example 1750 * when snapshot mount point is busy, we will 1751 * have an error here, but there will be no need 1752 * to reinsert snapshot. 1753 */ 1754 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1755 avl_insert(&sdp->sd_snaps, sep, where); 1756 break; 1757 } 1758 } 1759 sep = next; 1760 } 1761 1762 mutex_exit(&sdp->sd_lock); 1763 VN_RELE(dvp); 1764 1765 return (error); 1766} 1767