zfs_ctldir.c revision 302760
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); 631 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL); 632 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) { 633 if (flags & ISDOTDOT) 634 VOP_UNLOCK(dvp, 0); 635 err = vn_lock(*vpp, lkflags); 636 if (err != 0) { 637 vrele(*vpp); 638 *vpp = NULL; 639 } 640 if (flags & ISDOTDOT) 641 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); 642 } 643 644 return (err); 645} 646 647static struct vop_vector zfsctl_ops_root = { 648 .vop_default = &default_vnodeops, 649 .vop_open = zfsctl_common_open, 650 .vop_close = zfsctl_common_close, 651 .vop_ioctl = VOP_EINVAL, 652 .vop_getattr = zfsctl_root_getattr, 653 .vop_access = zfsctl_common_access, 654 .vop_readdir = gfs_vop_readdir, 655 .vop_lookup = zfsctl_freebsd_root_lookup, 656 .vop_inactive = VOP_NULL, 657 .vop_reclaim = gfs_vop_reclaim, 658#ifdef TODO 659 .vop_pathconf = zfsctl_pathconf, 660#endif 661 .vop_fid = zfsctl_common_fid, 662 .vop_print = zfsctl_root_print, 663}; 664 665/* 666 * Gets the full dataset name that corresponds to the given snapshot name 667 * Example: 668 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1" 669 */ 670static int 671zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 672{ 673 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 674 675 if (zfs_component_namecheck(name, NULL, NULL) != 0) 676 return (SET_ERROR(EILSEQ)); 677 dmu_objset_name(os, zname); 678 if (strlen(zname) + 1 + strlen(name) >= len) 679 return (SET_ERROR(ENAMETOOLONG)); 680 (void) strcat(zname, "@"); 681 (void) strcat(zname, name); 682 return (0); 683} 684 685static int 686zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 687{ 688 vnode_t *svp = sep->se_root; 689 int error; 690 691 ASSERT(vn_ismntpt(svp)); 692 693 /* this will be dropped by dounmount() */ 694 if ((error = vn_vfswlock(svp)) != 0) 695 return (error); 696 697#ifdef illumos 698 VN_HOLD(svp); 699 error = dounmount(vn_mountedvfs(svp), fflags, cr); 700 if (error) { 701 VN_RELE(svp); 702 return (error); 703 } 704 705 /* 706 * We can't use VN_RELE(), as that will try to invoke 707 * zfsctl_snapdir_inactive(), which would cause us to destroy 708 * the sd_lock mutex held by our caller. 709 */ 710 ASSERT(svp->v_count == 1); 711 gfs_vop_reclaim(svp, cr, NULL); 712 713 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 714 kmem_free(sep, sizeof (zfs_snapentry_t)); 715 716 return (0); 717#else 718 vfs_ref(vn_mountedvfs(svp)); 719 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 720#endif 721} 722 723#ifdef illumos 724static void 725zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 726{ 727 avl_index_t where; 728 vfs_t *vfsp; 729 refstr_t *pathref; 730 char newpath[MAXNAMELEN]; 731 char *tail; 732 733 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 734 ASSERT(sep != NULL); 735 736 vfsp = vn_mountedvfs(sep->se_root); 737 ASSERT(vfsp != NULL); 738 739 vfs_lock_wait(vfsp); 740 741 /* 742 * Change the name in the AVL tree. 743 */ 744 avl_remove(&sdp->sd_snaps, sep); 745 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 746 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 747 (void) strcpy(sep->se_name, nm); 748 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 749 avl_insert(&sdp->sd_snaps, sep, where); 750 751 /* 752 * Change the current mountpoint info: 753 * - update the tail of the mntpoint path 754 * - update the tail of the resource path 755 */ 756 pathref = vfs_getmntpoint(vfsp); 757 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 758 VERIFY((tail = strrchr(newpath, '/')) != NULL); 759 *(tail+1) = '\0'; 760 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 761 (void) strcat(newpath, nm); 762 refstr_rele(pathref); 763 vfs_setmntpoint(vfsp, newpath, 0); 764 765 pathref = vfs_getresource(vfsp); 766 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 767 VERIFY((tail = strrchr(newpath, '@')) != NULL); 768 *(tail+1) = '\0'; 769 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 770 (void) strcat(newpath, nm); 771 refstr_rele(pathref); 772 vfs_setresource(vfsp, newpath, 0); 773 774 vfs_unlock(vfsp); 775} 776#endif /* illumos */ 777 778#ifdef illumos 779/*ARGSUSED*/ 780static int 781zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 782 cred_t *cr, caller_context_t *ct, int flags) 783{ 784 zfsctl_snapdir_t *sdp = sdvp->v_data; 785 zfs_snapentry_t search, *sep; 786 zfsvfs_t *zfsvfs; 787 avl_index_t where; 788 char from[MAXNAMELEN], to[MAXNAMELEN]; 789 char real[MAXNAMELEN], fsname[MAXNAMELEN]; 790 int err; 791 792 zfsvfs = sdvp->v_vfsp->vfs_data; 793 ZFS_ENTER(zfsvfs); 794 795 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 796 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 797 MAXNAMELEN, NULL); 798 if (err == 0) { 799 snm = real; 800 } else if (err != ENOTSUP) { 801 ZFS_EXIT(zfsvfs); 802 return (err); 803 } 804 } 805 806 ZFS_EXIT(zfsvfs); 807 808 dmu_objset_name(zfsvfs->z_os, fsname); 809 810 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 811 if (err == 0) 812 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 813 if (err == 0) 814 err = zfs_secpolicy_rename_perms(from, to, cr); 815 if (err != 0) 816 return (err); 817 818 /* 819 * Cannot move snapshots out of the snapdir. 820 */ 821 if (sdvp != tdvp) 822 return (SET_ERROR(EINVAL)); 823 824 if (strcmp(snm, tnm) == 0) 825 return (0); 826 827 mutex_enter(&sdp->sd_lock); 828 829 search.se_name = (char *)snm; 830 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 831 mutex_exit(&sdp->sd_lock); 832 return (SET_ERROR(ENOENT)); 833 } 834 835 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0); 836 if (err == 0) 837 zfsctl_rename_snap(sdp, sep, tnm); 838 839 mutex_exit(&sdp->sd_lock); 840 841 return (err); 842} 843#endif /* illumos */ 844 845#ifdef illumos 846/* ARGSUSED */ 847static int 848zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 849 caller_context_t *ct, int flags) 850{ 851 zfsctl_snapdir_t *sdp = dvp->v_data; 852 zfs_snapentry_t *sep; 853 zfs_snapentry_t search; 854 zfsvfs_t *zfsvfs; 855 char snapname[MAXNAMELEN]; 856 char real[MAXNAMELEN]; 857 int err; 858 859 zfsvfs = dvp->v_vfsp->vfs_data; 860 ZFS_ENTER(zfsvfs); 861 862 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 863 864 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 865 MAXNAMELEN, NULL); 866 if (err == 0) { 867 name = real; 868 } else if (err != ENOTSUP) { 869 ZFS_EXIT(zfsvfs); 870 return (err); 871 } 872 } 873 874 ZFS_EXIT(zfsvfs); 875 876 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 877 if (err == 0) 878 err = zfs_secpolicy_destroy_perms(snapname, cr); 879 if (err != 0) 880 return (err); 881 882 mutex_enter(&sdp->sd_lock); 883 884 search.se_name = name; 885 sep = avl_find(&sdp->sd_snaps, &search, NULL); 886 if (sep) { 887 avl_remove(&sdp->sd_snaps, sep); 888 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 889 if (err != 0) 890 avl_add(&sdp->sd_snaps, sep); 891 else 892 err = dsl_destroy_snapshot(snapname, B_FALSE); 893 } else { 894 err = SET_ERROR(ENOENT); 895 } 896 897 mutex_exit(&sdp->sd_lock); 898 899 return (err); 900} 901#endif /* illumos */ 902 903/* 904 * This creates a snapshot under '.zfs/snapshot'. 905 */ 906/* ARGSUSED */ 907static int 908zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 909 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 910{ 911 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 912 char name[MAXNAMELEN]; 913 int err; 914 static enum symfollow follow = NO_FOLLOW; 915 static enum uio_seg seg = UIO_SYSSPACE; 916 917 if (zfs_component_namecheck(dirname, NULL, NULL) != 0) 918 return (SET_ERROR(EILSEQ)); 919 920 dmu_objset_name(zfsvfs->z_os, name); 921 922 *vpp = NULL; 923 924 err = zfs_secpolicy_snapshot_perms(name, cr); 925 if (err != 0) 926 return (err); 927 928 if (err == 0) { 929 err = dmu_objset_snapshot_one(name, dirname); 930 if (err != 0) 931 return (err); 932 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 933 } 934 935 return (err); 936} 937 938static int 939zfsctl_freebsd_snapdir_mkdir(ap) 940 struct vop_mkdir_args /* { 941 struct vnode *a_dvp; 942 struct vnode **a_vpp; 943 struct componentname *a_cnp; 944 struct vattr *a_vap; 945 } */ *ap; 946{ 947 948 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 949 950 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 951 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 952} 953 954/* 955 * Lookup entry point for the 'snapshot' directory. Try to open the 956 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 957 * Perform a mount of the associated dataset on top of the vnode. 958 */ 959/* ARGSUSED */ 960int 961zfsctl_snapdir_lookup(ap) 962 struct vop_lookup_args /* { 963 struct vnode *a_dvp; 964 struct vnode **a_vpp; 965 struct componentname *a_cnp; 966 } */ *ap; 967{ 968 vnode_t *dvp = ap->a_dvp; 969 vnode_t **vpp = ap->a_vpp; 970 struct componentname *cnp = ap->a_cnp; 971 char nm[NAME_MAX + 1]; 972 zfsctl_snapdir_t *sdp = dvp->v_data; 973 objset_t *snap; 974 char snapname[MAXNAMELEN]; 975 char real[MAXNAMELEN]; 976 char *mountpoint; 977 zfs_snapentry_t *sep, search; 978 size_t mountpoint_len; 979 avl_index_t where; 980 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 981 int err; 982 int ltype, flags = 0; 983 984 /* 985 * No extended attributes allowed under .zfs 986 */ 987 if (flags & LOOKUP_XATTR) 988 return (SET_ERROR(EINVAL)); 989 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 990 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 991 992 ASSERT(dvp->v_type == VDIR); 993 994 *vpp = NULL; 995 996 /* 997 * If we get a recursive call, that means we got called 998 * from the domount() code while it was trying to look up the 999 * spec (which looks like a local path for zfs). We need to 1000 * add some flag to domount() to tell it not to do this lookup. 1001 */ 1002 if (MUTEX_HELD(&sdp->sd_lock)) 1003 return (SET_ERROR(ENOENT)); 1004 1005 ZFS_ENTER(zfsvfs); 1006 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1007 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') { 1008 VOP_UNLOCK(dvp, 0); 1009 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE)); 1010 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE)); 1011 } 1012 ZFS_EXIT(zfsvfs); 1013 return (0); 1014 } 1015 1016 if (flags & FIGNORECASE) { 1017 boolean_t conflict = B_FALSE; 1018 1019 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 1020 MAXNAMELEN, &conflict); 1021 if (err == 0) { 1022 strlcpy(nm, real, sizeof(nm)); 1023 } else if (err != ENOTSUP) { 1024 ZFS_EXIT(zfsvfs); 1025 return (err); 1026 } 1027#if 0 1028 if (realpnp) 1029 (void) strlcpy(realpnp->pn_buf, nm, 1030 realpnp->pn_bufsize); 1031 if (conflict && direntflags) 1032 *direntflags = ED_CASE_CONFLICT; 1033#endif 1034 } 1035 1036relookup: 1037 mutex_enter(&sdp->sd_lock); 1038 search.se_name = (char *)nm; 1039 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 1040 *vpp = sep->se_root; 1041 VN_HOLD(*vpp); 1042 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 1043 if (err != 0) { 1044 *vpp = NULL; 1045 } else if (*vpp == sep->se_root) { 1046 /* 1047 * The snapshot was unmounted behind our backs, 1048 * try to remount it. 1049 */ 1050 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0); 1051 goto domount; 1052 } 1053 mutex_exit(&sdp->sd_lock); 1054 ZFS_EXIT(zfsvfs); 1055 return (err); 1056 } 1057 1058 /* 1059 * The requested snapshot is not currently mounted, look it up. 1060 */ 1061 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1062 if (err != 0) { 1063 mutex_exit(&sdp->sd_lock); 1064 ZFS_EXIT(zfsvfs); 1065 /* 1066 * handle "ls *" or "?" in a graceful manner, 1067 * forcing EILSEQ to ENOENT. 1068 * Since shell ultimately passes "*" or "?" as name to lookup 1069 */ 1070 return (err == EILSEQ ? ENOENT : err); 1071 } 1072 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1073 mutex_exit(&sdp->sd_lock); 1074#ifdef illumos 1075 ZFS_EXIT(zfsvfs); 1076 return (SET_ERROR(ENOENT)); 1077#else /* !illumos */ 1078 /* Translate errors and add SAVENAME when needed. */ 1079 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1080 err = EJUSTRETURN; 1081 cnp->cn_flags |= SAVENAME; 1082 } else { 1083 err = SET_ERROR(ENOENT); 1084 } 1085 ZFS_EXIT(zfsvfs); 1086 return (err); 1087#endif /* illumos */ 1088 } 1089 1090 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1091 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1092 (void) strcpy(sep->se_name, nm); 1093 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1094 avl_insert(&sdp->sd_snaps, sep, where); 1095 1096 dmu_objset_rele(snap, FTAG); 1097domount: 1098 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1099 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1100 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1101 (void) snprintf(mountpoint, mountpoint_len, 1102 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1103 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1104 mutex_exit(&sdp->sd_lock); 1105 1106 /* 1107 * The vnode may get reclaimed between dropping sd_lock and 1108 * getting the vnode lock. 1109 * */ 1110 err = vn_lock(*vpp, LK_EXCLUSIVE); 1111 if (err == ENOENT) 1112 goto relookup; 1113 VERIFY0(err); 1114 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1115 kmem_free(mountpoint, mountpoint_len); 1116 if (err == 0) { 1117 /* 1118 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1119 * 1120 * This is where we lie about our v_vfsp in order to 1121 * make .zfs/snapshot/<snapname> accessible over NFS 1122 * without requiring manual mounts of <snapname>. 1123 */ 1124 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1125 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1126 } 1127 ZFS_EXIT(zfsvfs); 1128 1129#ifdef illumos 1130 /* 1131 * If we had an error, drop our hold on the vnode and 1132 * zfsctl_snapshot_inactive() will clean up. 1133 */ 1134 if (err != 0) { 1135 VN_RELE(*vpp); 1136 *vpp = NULL; 1137 } 1138#else 1139 if (err != 0) 1140 *vpp = NULL; 1141#endif 1142 return (err); 1143} 1144 1145/* ARGSUSED */ 1146int 1147zfsctl_shares_lookup(ap) 1148 struct vop_lookup_args /* { 1149 struct vnode *a_dvp; 1150 struct vnode **a_vpp; 1151 struct componentname *a_cnp; 1152 } */ *ap; 1153{ 1154 vnode_t *dvp = ap->a_dvp; 1155 vnode_t **vpp = ap->a_vpp; 1156 struct componentname *cnp = ap->a_cnp; 1157 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1158 char nm[NAME_MAX + 1]; 1159 znode_t *dzp; 1160 int error; 1161 1162 ZFS_ENTER(zfsvfs); 1163 1164 ASSERT(cnp->cn_namelen < sizeof(nm)); 1165 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1166 1167 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1168 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') { 1169 VOP_UNLOCK(dvp, 0); 1170 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE)); 1171 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE)); 1172 } 1173 ZFS_EXIT(zfsvfs); 1174 return (0); 1175 } 1176 1177 if (zfsvfs->z_shares_dir == 0) { 1178 ZFS_EXIT(zfsvfs); 1179 return (SET_ERROR(ENOTSUP)); 1180 } 1181 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1182 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1183 VN_RELE(ZTOV(dzp)); 1184 } 1185 1186 ZFS_EXIT(zfsvfs); 1187 1188 return (error); 1189} 1190 1191/* ARGSUSED */ 1192static int 1193zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1194 offset_t *offp, offset_t *nextp, void *data, int flags) 1195{ 1196 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1197 char snapname[MAXNAMELEN]; 1198 uint64_t id, cookie; 1199 boolean_t case_conflict; 1200 int error; 1201 1202 ZFS_ENTER(zfsvfs); 1203 1204 cookie = *offp; 1205 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG); 1206 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1207 &cookie, &case_conflict); 1208 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG); 1209 if (error) { 1210 ZFS_EXIT(zfsvfs); 1211 if (error == ENOENT) { 1212 *eofp = 1; 1213 return (0); 1214 } 1215 return (error); 1216 } 1217 1218 if (flags & V_RDDIR_ENTFLAGS) { 1219 edirent_t *eodp = dp; 1220 1221 (void) strcpy(eodp->ed_name, snapname); 1222 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1223 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1224 } else { 1225 struct dirent64 *odp = dp; 1226 1227 (void) strcpy(odp->d_name, snapname); 1228 odp->d_ino = ZFSCTL_INO_SNAP(id); 1229 } 1230 *nextp = cookie; 1231 1232 ZFS_EXIT(zfsvfs); 1233 1234 return (0); 1235} 1236 1237/* ARGSUSED */ 1238static int 1239zfsctl_shares_readdir(ap) 1240 struct vop_readdir_args /* { 1241 struct vnode *a_vp; 1242 struct uio *a_uio; 1243 struct ucred *a_cred; 1244 int *a_eofflag; 1245 int *a_ncookies; 1246 u_long **a_cookies; 1247 } */ *ap; 1248{ 1249 vnode_t *vp = ap->a_vp; 1250 uio_t *uiop = ap->a_uio; 1251 cred_t *cr = ap->a_cred; 1252 int *eofp = ap->a_eofflag; 1253 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1254 znode_t *dzp; 1255 int error; 1256 1257 ZFS_ENTER(zfsvfs); 1258 1259 if (zfsvfs->z_shares_dir == 0) { 1260 ZFS_EXIT(zfsvfs); 1261 return (SET_ERROR(ENOTSUP)); 1262 } 1263 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1264 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1265 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1266 VN_URELE(ZTOV(dzp)); 1267 } else { 1268 *eofp = 1; 1269 error = SET_ERROR(ENOENT); 1270 } 1271 1272 ZFS_EXIT(zfsvfs); 1273 return (error); 1274} 1275 1276/* 1277 * pvp is the '.zfs' directory (zfsctl_node_t). 1278 * 1279 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1280 * 1281 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1282 * when a lookup is performed on .zfs for "snapshot". 1283 */ 1284vnode_t * 1285zfsctl_mknode_snapdir(vnode_t *pvp) 1286{ 1287 vnode_t *vp; 1288 zfsctl_snapdir_t *sdp; 1289 1290 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1291 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1292 zfsctl_snapdir_readdir_cb, NULL); 1293 sdp = vp->v_data; 1294 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1295 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1296 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1297 avl_create(&sdp->sd_snaps, snapentry_compare, 1298 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1299 VOP_UNLOCK(vp, 0); 1300 return (vp); 1301} 1302 1303vnode_t * 1304zfsctl_mknode_shares(vnode_t *pvp) 1305{ 1306 vnode_t *vp; 1307 zfsctl_node_t *sdp; 1308 1309 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1310 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1311 NULL, NULL); 1312 sdp = vp->v_data; 1313 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1314 VOP_UNLOCK(vp, 0); 1315 return (vp); 1316 1317} 1318 1319/* ARGSUSED */ 1320static int 1321zfsctl_shares_getattr(ap) 1322 struct vop_getattr_args /* { 1323 struct vnode *a_vp; 1324 struct vattr *a_vap; 1325 struct ucred *a_cred; 1326 struct thread *a_td; 1327 } */ *ap; 1328{ 1329 vnode_t *vp = ap->a_vp; 1330 vattr_t *vap = ap->a_vap; 1331 cred_t *cr = ap->a_cred; 1332 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1333 znode_t *dzp; 1334 int error; 1335 1336 ZFS_ENTER(zfsvfs); 1337 if (zfsvfs->z_shares_dir == 0) { 1338 ZFS_EXIT(zfsvfs); 1339 return (SET_ERROR(ENOTSUP)); 1340 } 1341 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1342 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1343 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1344 VN_URELE(ZTOV(dzp)); 1345 } 1346 ZFS_EXIT(zfsvfs); 1347 return (error); 1348 1349 1350} 1351 1352/* ARGSUSED */ 1353static int 1354zfsctl_snapdir_getattr(ap) 1355 struct vop_getattr_args /* { 1356 struct vnode *a_vp; 1357 struct vattr *a_vap; 1358 struct ucred *a_cred; 1359 } */ *ap; 1360{ 1361 vnode_t *vp = ap->a_vp; 1362 vattr_t *vap = ap->a_vap; 1363 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1364 zfsctl_snapdir_t *sdp = vp->v_data; 1365 1366 ZFS_ENTER(zfsvfs); 1367 zfsctl_common_getattr(vp, vap); 1368 vap->va_nodeid = gfs_file_inode(vp); 1369 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1370 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1371 vap->va_birthtime = vap->va_ctime; 1372 ZFS_EXIT(zfsvfs); 1373 1374 return (0); 1375} 1376 1377/* ARGSUSED */ 1378static int 1379zfsctl_snapdir_reclaim(ap) 1380 struct vop_reclaim_args /* { 1381 struct vnode *a_vp; 1382 struct thread *a_td; 1383 } */ *ap; 1384{ 1385 vnode_t *vp = ap->a_vp; 1386 zfsctl_snapdir_t *sdp = vp->v_data; 1387 zfs_snapentry_t *sep; 1388 1389 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1390 mutex_destroy(&sdp->sd_lock); 1391 avl_destroy(&sdp->sd_snaps); 1392 gfs_vop_reclaim(ap); 1393 1394 return (0); 1395} 1396 1397static int 1398zfsctl_shares_print(ap) 1399 struct vop_print_args /* { 1400 struct vnode *a_vp; 1401 } */ *ap; 1402{ 1403 printf(" .zfs/shares node\n"); 1404 zfsctl_common_print(ap); 1405 return (0); 1406} 1407 1408static int 1409zfsctl_snapdir_print(ap) 1410 struct vop_print_args /* { 1411 struct vnode *a_vp; 1412 } */ *ap; 1413{ 1414 vnode_t *vp = ap->a_vp; 1415 zfsctl_snapdir_t *sdp = vp->v_data; 1416 1417 printf(" .zfs/snapshot node\n"); 1418 printf(" number of children = %lu\n", avl_numnodes(&sdp->sd_snaps)); 1419 zfsctl_common_print(ap); 1420 return (0); 1421} 1422 1423#ifdef illumos 1424static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1425 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1426 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1427 { VOPNAME_IOCTL, { .error = fs_inval } }, 1428 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1429 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1430 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1431 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1432 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1433 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1434 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1435 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1436 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1437 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1438 { NULL } 1439}; 1440 1441static const fs_operation_def_t zfsctl_tops_shares[] = { 1442 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1443 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1444 { VOPNAME_IOCTL, { .error = fs_inval } }, 1445 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1446 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1447 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1448 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1449 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1450 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1451 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1452 { NULL } 1453}; 1454#else /* !illumos */ 1455static struct vop_vector zfsctl_ops_snapdir = { 1456 .vop_default = &default_vnodeops, 1457 .vop_open = zfsctl_common_open, 1458 .vop_close = zfsctl_common_close, 1459 .vop_ioctl = VOP_EINVAL, 1460 .vop_getattr = zfsctl_snapdir_getattr, 1461 .vop_access = zfsctl_common_access, 1462 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1463 .vop_readdir = gfs_vop_readdir, 1464 .vop_lookup = zfsctl_snapdir_lookup, 1465 .vop_inactive = VOP_NULL, 1466 .vop_reclaim = zfsctl_snapdir_reclaim, 1467 .vop_fid = zfsctl_common_fid, 1468 .vop_print = zfsctl_snapdir_print, 1469}; 1470 1471static struct vop_vector zfsctl_ops_shares = { 1472 .vop_default = &default_vnodeops, 1473 .vop_open = zfsctl_common_open, 1474 .vop_close = zfsctl_common_close, 1475 .vop_ioctl = VOP_EINVAL, 1476 .vop_getattr = zfsctl_shares_getattr, 1477 .vop_access = zfsctl_common_access, 1478 .vop_readdir = zfsctl_shares_readdir, 1479 .vop_lookup = zfsctl_shares_lookup, 1480 .vop_inactive = VOP_NULL, 1481 .vop_reclaim = gfs_vop_reclaim, 1482 .vop_fid = zfsctl_shares_fid, 1483 .vop_print = zfsctl_shares_print, 1484}; 1485#endif /* illumos */ 1486 1487/* 1488 * pvp is the GFS vnode '.zfs/snapshot'. 1489 * 1490 * This creates a GFS node under '.zfs/snapshot' representing each 1491 * snapshot. This newly created GFS node is what we mount snapshot 1492 * vfs_t's ontop of. 1493 */ 1494static vnode_t * 1495zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1496{ 1497 vnode_t *vp; 1498 zfsctl_node_t *zcp; 1499 1500 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1501 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1502 zcp = vp->v_data; 1503 zcp->zc_id = objset; 1504 VOP_UNLOCK(vp, 0); 1505 1506 return (vp); 1507} 1508 1509static int 1510zfsctl_snapshot_inactive(ap) 1511 struct vop_inactive_args /* { 1512 struct vnode *a_vp; 1513 struct thread *a_td; 1514 } */ *ap; 1515{ 1516 vnode_t *vp = ap->a_vp; 1517 1518 vrecycle(vp); 1519 return (0); 1520} 1521 1522static int 1523zfsctl_snapshot_reclaim(ap) 1524 struct vop_reclaim_args /* { 1525 struct vnode *a_vp; 1526 struct thread *a_td; 1527 } */ *ap; 1528{ 1529 vnode_t *vp = ap->a_vp; 1530 cred_t *cr = ap->a_td->td_ucred; 1531 zfsctl_snapdir_t *sdp; 1532 zfs_snapentry_t *sep, *next; 1533 int locked; 1534 vnode_t *dvp; 1535 1536 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1537 sdp = dvp->v_data; 1538 /* this may already have been unmounted */ 1539 if (sdp == NULL) { 1540 VN_RELE(dvp); 1541 return (0); 1542 } 1543 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1544 mutex_enter(&sdp->sd_lock); 1545 1546 ASSERT(!vn_ismntpt(vp)); 1547 1548 sep = avl_first(&sdp->sd_snaps); 1549 while (sep != NULL) { 1550 next = AVL_NEXT(&sdp->sd_snaps, sep); 1551 1552 if (sep->se_root == vp) { 1553 avl_remove(&sdp->sd_snaps, sep); 1554 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1555 kmem_free(sep, sizeof (zfs_snapentry_t)); 1556 break; 1557 } 1558 sep = next; 1559 } 1560 ASSERT(sep != NULL); 1561 1562 if (!locked) 1563 mutex_exit(&sdp->sd_lock); 1564 VN_RELE(dvp); 1565 1566 /* 1567 * Dispose of the vnode for the snapshot mount point. 1568 * This is safe to do because once this entry has been removed 1569 * from the AVL tree, it can't be found again, so cannot become 1570 * "active". If we lookup the same name again we will end up 1571 * creating a new vnode. 1572 */ 1573 gfs_vop_reclaim(ap); 1574 return (0); 1575 1576} 1577 1578static int 1579zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1580{ 1581 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1582 vnode_t *dvp, *vp; 1583 zfsctl_snapdir_t *sdp; 1584 zfs_snapentry_t *sep; 1585 int error; 1586 1587 ASSERT(zfsvfs->z_ctldir != NULL); 1588 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1589 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1590 if (error != 0) 1591 return (error); 1592 sdp = dvp->v_data; 1593 1594 mutex_enter(&sdp->sd_lock); 1595 sep = avl_first(&sdp->sd_snaps); 1596 while (sep != NULL) { 1597 vp = sep->se_root; 1598 if (vp == ap->a_vp) 1599 break; 1600 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1601 } 1602 if (sep == NULL) { 1603 mutex_exit(&sdp->sd_lock); 1604 error = ENOENT; 1605 } else { 1606 size_t len; 1607 1608 len = strlen(sep->se_name); 1609 *ap->a_buflen -= len; 1610 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1611 mutex_exit(&sdp->sd_lock); 1612 vref(dvp); 1613 *ap->a_vpp = dvp; 1614 } 1615 VN_RELE(dvp); 1616 1617 return (error); 1618} 1619 1620static int 1621zfsctl_snaphot_print(ap) 1622 struct vop_print_args /* { 1623 struct vnode *a_vp; 1624 } */ *ap; 1625{ 1626 vnode_t *vp = ap->a_vp; 1627 zfsctl_node_t *zcp = vp->v_data; 1628 1629 printf(" .zfs/snapshot/<snap> node\n"); 1630 printf(" id = %ju\n", (uintmax_t)zcp->zc_id); 1631 zfsctl_common_print(ap); 1632 return (0); 1633} 1634 1635/* 1636 * These VP's should never see the light of day. They should always 1637 * be covered. 1638 */ 1639static struct vop_vector zfsctl_ops_snapshot = { 1640 .vop_default = &default_vnodeops, 1641 .vop_inactive = zfsctl_snapshot_inactive, 1642 .vop_reclaim = zfsctl_snapshot_reclaim, 1643 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1644 .vop_print = zfsctl_snaphot_print, 1645}; 1646 1647int 1648zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1649{ 1650 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1651 vnode_t *dvp, *vp; 1652 zfsctl_snapdir_t *sdp; 1653 zfsctl_node_t *zcp; 1654 zfs_snapentry_t *sep; 1655 int error; 1656 1657 ASSERT(zfsvfs->z_ctldir != NULL); 1658 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1659 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1660 if (error != 0) 1661 return (error); 1662 sdp = dvp->v_data; 1663 1664 mutex_enter(&sdp->sd_lock); 1665 sep = avl_first(&sdp->sd_snaps); 1666 while (sep != NULL) { 1667 vp = sep->se_root; 1668 zcp = vp->v_data; 1669 if (zcp->zc_id == objsetid) 1670 break; 1671 1672 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1673 } 1674 1675 if (sep != NULL) { 1676 VN_HOLD(vp); 1677 /* 1678 * Return the mounted root rather than the covered mount point. 1679 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1680 * and returns the ZFS vnode mounted on top of the GFS node. 1681 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1682 */ 1683 error = traverse(&vp, LK_SHARED | LK_RETRY); 1684 if (error == 0) { 1685 if (vp == sep->se_root) { 1686 VN_RELE(vp); /* release covered vp */ 1687 error = SET_ERROR(EINVAL); 1688 } else { 1689 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1690 VN_URELE(vp); /* put snapshot's root vp */ 1691 } 1692 } 1693 mutex_exit(&sdp->sd_lock); 1694 } else { 1695 error = SET_ERROR(EINVAL); 1696 mutex_exit(&sdp->sd_lock); 1697 } 1698 1699 VN_RELE(dvp); 1700 1701 return (error); 1702} 1703 1704/* 1705 * Unmount any snapshots for the given filesystem. This is called from 1706 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1707 * snapshots. 1708 */ 1709int 1710zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1711{ 1712 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1713 vnode_t *dvp; 1714 zfsctl_snapdir_t *sdp; 1715 zfs_snapentry_t *sep, *next; 1716 int error; 1717 1718 ASSERT(zfsvfs->z_ctldir != NULL); 1719 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1720 NULL, 0, NULL, cr, NULL, NULL, NULL); 1721 if (error != 0) 1722 return (error); 1723 sdp = dvp->v_data; 1724 1725 mutex_enter(&sdp->sd_lock); 1726 1727 sep = avl_first(&sdp->sd_snaps); 1728 while (sep != NULL) { 1729 next = AVL_NEXT(&sdp->sd_snaps, sep); 1730 1731 /* 1732 * If this snapshot is not mounted, then it must 1733 * have just been unmounted by somebody else, and 1734 * will be cleaned up by zfsctl_snapdir_inactive(). 1735 */ 1736 if (vn_ismntpt(sep->se_root)) { 1737 error = zfsctl_unmount_snap(sep, fflags, cr); 1738 if (error) { 1739 avl_index_t where; 1740 1741 /* 1742 * Before reinserting snapshot to the tree, 1743 * check if it was actually removed. For example 1744 * when snapshot mount point is busy, we will 1745 * have an error here, but there will be no need 1746 * to reinsert snapshot. 1747 */ 1748 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1749 avl_insert(&sdp->sd_snaps, sep, where); 1750 break; 1751 } 1752 } 1753 sep = next; 1754 } 1755 1756 mutex_exit(&sdp->sd_lock); 1757 VN_RELE(dvp); 1758 1759 return (error); 1760} 1761