zfs_ctldir.c revision 229703
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) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>. 24 * 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_deleg.h> 77#include <sys/mount.h> 78#include <sys/sunddi.h> 79 80#include "zfs_namecheck.h" 81 82typedef struct zfsctl_node { 83 gfs_dir_t zc_gfs_private; 84 uint64_t zc_id; 85 timestruc_t zc_cmtime; /* ctime and mtime, always the same */ 86} zfsctl_node_t; 87 88typedef struct zfsctl_snapdir { 89 zfsctl_node_t sd_node; 90 kmutex_t sd_lock; 91 avl_tree_t sd_snaps; 92} zfsctl_snapdir_t; 93 94typedef struct { 95 char *se_name; 96 vnode_t *se_root; 97 avl_node_t se_node; 98} zfs_snapentry_t; 99 100static int 101snapentry_compare(const void *a, const void *b) 102{ 103 const zfs_snapentry_t *sa = a; 104 const zfs_snapentry_t *sb = b; 105 int ret = strcmp(sa->se_name, sb->se_name); 106 107 if (ret < 0) 108 return (-1); 109 else if (ret > 0) 110 return (1); 111 else 112 return (0); 113} 114 115#ifdef sun 116vnodeops_t *zfsctl_ops_root; 117vnodeops_t *zfsctl_ops_snapdir; 118vnodeops_t *zfsctl_ops_snapshot; 119vnodeops_t *zfsctl_ops_shares; 120vnodeops_t *zfsctl_ops_shares_dir; 121 122static const fs_operation_def_t zfsctl_tops_root[]; 123static const fs_operation_def_t zfsctl_tops_snapdir[]; 124static const fs_operation_def_t zfsctl_tops_snapshot[]; 125static const fs_operation_def_t zfsctl_tops_shares[]; 126#else /* !sun */ 127static struct vop_vector zfsctl_ops_root; 128static struct vop_vector zfsctl_ops_snapdir; 129static struct vop_vector zfsctl_ops_snapshot; 130static struct vop_vector zfsctl_ops_shares; 131static struct vop_vector zfsctl_ops_shares_dir; 132#endif /* !sun */ 133 134static vnode_t *zfsctl_mknode_snapdir(vnode_t *); 135static vnode_t *zfsctl_mknode_shares(vnode_t *); 136static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset); 137static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *); 138 139#ifdef sun 140static gfs_opsvec_t zfsctl_opsvec[] = { 141 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root }, 142 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir }, 143 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot }, 144 { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir }, 145 { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares }, 146 { NULL } 147}; 148#endif /* sun */ 149 150/* 151 * Root directory elements. We only have two entries 152 * snapshot and shares. 153 */ 154static gfs_dirent_t zfsctl_root_entries[] = { 155 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE }, 156 { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE }, 157 { NULL } 158}; 159 160/* include . and .. in the calculation */ 161#define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \ 162 sizeof (gfs_dirent_t)) + 1) 163 164 165/* 166 * Initialize the various GFS pieces we'll need to create and manipulate .zfs 167 * directories. This is called from the ZFS init routine, and initializes the 168 * vnode ops vectors that we'll be using. 169 */ 170void 171zfsctl_init(void) 172{ 173#ifdef sun 174 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0); 175#endif 176} 177 178void 179zfsctl_fini(void) 180{ 181#ifdef sun 182 /* 183 * Remove vfsctl vnode ops 184 */ 185 if (zfsctl_ops_root) 186 vn_freevnodeops(zfsctl_ops_root); 187 if (zfsctl_ops_snapdir) 188 vn_freevnodeops(zfsctl_ops_snapdir); 189 if (zfsctl_ops_snapshot) 190 vn_freevnodeops(zfsctl_ops_snapshot); 191 if (zfsctl_ops_shares) 192 vn_freevnodeops(zfsctl_ops_shares); 193 if (zfsctl_ops_shares_dir) 194 vn_freevnodeops(zfsctl_ops_shares_dir); 195 196 zfsctl_ops_root = NULL; 197 zfsctl_ops_snapdir = NULL; 198 zfsctl_ops_snapshot = NULL; 199 zfsctl_ops_shares = NULL; 200 zfsctl_ops_shares_dir = NULL; 201#endif /* sun */ 202} 203 204boolean_t 205zfsctl_is_node(vnode_t *vp) 206{ 207 return (vn_matchops(vp, zfsctl_ops_root) || 208 vn_matchops(vp, zfsctl_ops_snapdir) || 209 vn_matchops(vp, zfsctl_ops_snapshot) || 210 vn_matchops(vp, zfsctl_ops_shares) || 211 vn_matchops(vp, zfsctl_ops_shares_dir)); 212 213} 214 215/* 216 * Return the inode number associated with the 'snapshot' or 217 * 'shares' directory. 218 */ 219/* ARGSUSED */ 220static ino64_t 221zfsctl_root_inode_cb(vnode_t *vp, int index) 222{ 223 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 224 225 ASSERT(index <= 2); 226 227 if (index == 0) 228 return (ZFSCTL_INO_SNAPDIR); 229 230 return (zfsvfs->z_shares_dir); 231} 232 233/* 234 * Create the '.zfs' directory. This directory is cached as part of the VFS 235 * structure. This results in a hold on the vfs_t. The code in zfs_umount() 236 * therefore checks against a vfs_count of 2 instead of 1. This reference 237 * is removed when the ctldir is destroyed in the unmount. 238 */ 239void 240zfsctl_create(zfsvfs_t *zfsvfs) 241{ 242 vnode_t *vp, *rvp; 243 zfsctl_node_t *zcp; 244 uint64_t crtime[2]; 245 246 ASSERT(zfsvfs->z_ctldir == NULL); 247 248 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs, 249 &zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries, 250 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL); 251 zcp = vp->v_data; 252 zcp->zc_id = ZFSCTL_INO_ROOT; 253 254 VERIFY(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp) == 0); 255 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 256 &crtime, sizeof (crtime))); 257 ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime); 258 VN_URELE(rvp); 259 260 /* 261 * We're only faking the fact that we have a root of a filesystem for 262 * the sake of the GFS interfaces. Undo the flag manipulation it did 263 * for us. 264 */ 265 vp->v_vflag &= ~VV_ROOT; 266 267 zfsvfs->z_ctldir = vp; 268 269 VOP_UNLOCK(vp, 0); 270} 271 272/* 273 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted. 274 * There might still be more references if we were force unmounted, but only 275 * new zfs_inactive() calls can occur and they don't reference .zfs 276 */ 277void 278zfsctl_destroy(zfsvfs_t *zfsvfs) 279{ 280 VN_RELE(zfsvfs->z_ctldir); 281 zfsvfs->z_ctldir = NULL; 282} 283 284/* 285 * Given a root znode, retrieve the associated .zfs directory. 286 * Add a hold to the vnode and return it. 287 */ 288vnode_t * 289zfsctl_root(znode_t *zp) 290{ 291 ASSERT(zfs_has_ctldir(zp)); 292 VN_HOLD(zp->z_zfsvfs->z_ctldir); 293 return (zp->z_zfsvfs->z_ctldir); 294} 295 296/* 297 * Common open routine. Disallow any write access. 298 */ 299/* ARGSUSED */ 300static int 301zfsctl_common_open(struct vop_open_args *ap) 302{ 303 int flags = ap->a_mode; 304 305 if (flags & FWRITE) 306 return (EACCES); 307 308 return (0); 309} 310 311/* 312 * Common close routine. Nothing to do here. 313 */ 314/* ARGSUSED */ 315static int 316zfsctl_common_close(struct vop_close_args *ap) 317{ 318 return (0); 319} 320 321/* 322 * Common access routine. Disallow writes. 323 */ 324/* ARGSUSED */ 325static int 326zfsctl_common_access(ap) 327 struct vop_access_args /* { 328 struct vnode *a_vp; 329 accmode_t a_accmode; 330 struct ucred *a_cred; 331 struct thread *a_td; 332 } */ *ap; 333{ 334 accmode_t accmode = ap->a_accmode; 335 336#ifdef TODO 337 if (flags & V_ACE_MASK) { 338 if (accmode & ACE_ALL_WRITE_PERMS) 339 return (EACCES); 340 } else { 341#endif 342 if (accmode & VWRITE) 343 return (EACCES); 344#ifdef TODO 345 } 346#endif 347 348 return (0); 349} 350 351/* 352 * Common getattr function. Fill in basic information. 353 */ 354static void 355zfsctl_common_getattr(vnode_t *vp, vattr_t *vap) 356{ 357 timestruc_t now; 358 359 vap->va_uid = 0; 360 vap->va_gid = 0; 361 vap->va_rdev = 0; 362 /* 363 * We are a purely virtual object, so we have no 364 * blocksize or allocated blocks. 365 */ 366 vap->va_blksize = 0; 367 vap->va_nblocks = 0; 368 vap->va_seq = 0; 369 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 370 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP | 371 S_IROTH | S_IXOTH; 372 vap->va_type = VDIR; 373 /* 374 * We live in the now (for atime). 375 */ 376 gethrestime(&now); 377 vap->va_atime = now; 378 /* FreeBSD: Reset chflags(2) flags. */ 379 vap->va_flags = 0; 380} 381 382/*ARGSUSED*/ 383static int 384zfsctl_common_fid(ap) 385 struct vop_fid_args /* { 386 struct vnode *a_vp; 387 struct fid *a_fid; 388 } */ *ap; 389{ 390 vnode_t *vp = ap->a_vp; 391 fid_t *fidp = (void *)ap->a_fid; 392 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 393 zfsctl_node_t *zcp = vp->v_data; 394 uint64_t object = zcp->zc_id; 395 zfid_short_t *zfid; 396 int i; 397 398 ZFS_ENTER(zfsvfs); 399 400 fidp->fid_len = SHORT_FID_LEN; 401 402 zfid = (zfid_short_t *)fidp; 403 404 zfid->zf_len = SHORT_FID_LEN; 405 406 for (i = 0; i < sizeof (zfid->zf_object); i++) 407 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 408 409 /* .zfs znodes always have a generation number of 0 */ 410 for (i = 0; i < sizeof (zfid->zf_gen); i++) 411 zfid->zf_gen[i] = 0; 412 413 ZFS_EXIT(zfsvfs); 414 return (0); 415} 416 417 418/*ARGSUSED*/ 419static int 420zfsctl_shares_fid(ap) 421 struct vop_fid_args /* { 422 struct vnode *a_vp; 423 struct fid *a_fid; 424 } */ *ap; 425{ 426 vnode_t *vp = ap->a_vp; 427 fid_t *fidp = (void *)ap->a_fid; 428 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 429 znode_t *dzp; 430 int error; 431 432 ZFS_ENTER(zfsvfs); 433 434 if (zfsvfs->z_shares_dir == 0) { 435 ZFS_EXIT(zfsvfs); 436 return (ENOTSUP); 437 } 438 439 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 440 error = VOP_FID(ZTOV(dzp), fidp); 441 VN_RELE(ZTOV(dzp)); 442 } 443 444 ZFS_EXIT(zfsvfs); 445 return (error); 446} 447 448static int 449zfsctl_common_reclaim(ap) 450 struct vop_reclaim_args /* { 451 struct vnode *a_vp; 452 struct thread *a_td; 453 } */ *ap; 454{ 455 vnode_t *vp = ap->a_vp; 456 457 /* 458 * Destroy the vm object and flush associated pages. 459 */ 460 vnode_destroy_vobject(vp); 461 VI_LOCK(vp); 462 vp->v_data = NULL; 463 VI_UNLOCK(vp); 464 return (0); 465} 466 467/* 468 * .zfs inode namespace 469 * 470 * We need to generate unique inode numbers for all files and directories 471 * within the .zfs pseudo-filesystem. We use the following scheme: 472 * 473 * ENTRY ZFSCTL_INODE 474 * .zfs 1 475 * .zfs/snapshot 2 476 * .zfs/snapshot/<snap> objectid(snap) 477 */ 478 479#define ZFSCTL_INO_SNAP(id) (id) 480 481/* 482 * Get root directory attributes. 483 */ 484/* ARGSUSED */ 485static int 486zfsctl_root_getattr(ap) 487 struct vop_getattr_args /* { 488 struct vnode *a_vp; 489 struct vattr *a_vap; 490 struct ucred *a_cred; 491 } */ *ap; 492{ 493 struct vnode *vp = ap->a_vp; 494 struct vattr *vap = ap->a_vap; 495 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 496 zfsctl_node_t *zcp = vp->v_data; 497 498 ZFS_ENTER(zfsvfs); 499 vap->va_nodeid = ZFSCTL_INO_ROOT; 500 vap->va_nlink = vap->va_size = NROOT_ENTRIES; 501 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime; 502 vap->va_birthtime = vap->va_ctime; 503 504 zfsctl_common_getattr(vp, vap); 505 ZFS_EXIT(zfsvfs); 506 507 return (0); 508} 509 510/* 511 * Special case the handling of "..". 512 */ 513/* ARGSUSED */ 514int 515zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp, 516 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 517 int *direntflags, pathname_t *realpnp) 518{ 519 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 520 int err; 521 522 /* 523 * No extended attributes allowed under .zfs 524 */ 525 if (flags & LOOKUP_XATTR) 526 return (EINVAL); 527 528 ZFS_ENTER(zfsvfs); 529 530 if (strcmp(nm, "..") == 0) { 531 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp); 532 if (err == 0) 533 VOP_UNLOCK(*vpp, 0); 534 } else { 535 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir, 536 cr, ct, direntflags, realpnp); 537 } 538 539 ZFS_EXIT(zfsvfs); 540 541 return (err); 542} 543 544#ifdef sun 545static int 546zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 547 caller_context_t *ct) 548{ 549 /* 550 * We only care about ACL_ENABLED so that libsec can 551 * display ACL correctly and not default to POSIX draft. 552 */ 553 if (cmd == _PC_ACL_ENABLED) { 554 *valp = _ACL_ACE_ENABLED; 555 return (0); 556 } 557 558 return (fs_pathconf(vp, cmd, valp, cr, ct)); 559} 560#endif /* sun */ 561 562#ifdef sun 563static const fs_operation_def_t zfsctl_tops_root[] = { 564 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 565 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 566 { VOPNAME_IOCTL, { .error = fs_inval } }, 567 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } }, 568 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 569 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 570 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } }, 571 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 572 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 573 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } }, 574 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 575 { NULL } 576}; 577#endif /* sun */ 578 579/* 580 * Special case the handling of "..". 581 */ 582/* ARGSUSED */ 583int 584zfsctl_freebsd_root_lookup(ap) 585 struct vop_lookup_args /* { 586 struct vnode *a_dvp; 587 struct vnode **a_vpp; 588 struct componentname *a_cnp; 589 } */ *ap; 590{ 591 vnode_t *dvp = ap->a_dvp; 592 vnode_t **vpp = ap->a_vpp; 593 cred_t *cr = ap->a_cnp->cn_cred; 594 int flags = ap->a_cnp->cn_flags; 595 int nameiop = ap->a_cnp->cn_nameiop; 596 char nm[NAME_MAX + 1]; 597 int err; 598 599 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE)) 600 return (EOPNOTSUPP); 601 602 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 603 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 604 605 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL); 606 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) 607 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 608 return (err); 609} 610 611static struct vop_vector zfsctl_ops_root = { 612 .vop_default = &default_vnodeops, 613 .vop_open = zfsctl_common_open, 614 .vop_close = zfsctl_common_close, 615 .vop_ioctl = VOP_EINVAL, 616 .vop_getattr = zfsctl_root_getattr, 617 .vop_access = zfsctl_common_access, 618 .vop_readdir = gfs_vop_readdir, 619 .vop_lookup = zfsctl_freebsd_root_lookup, 620 .vop_inactive = gfs_vop_inactive, 621 .vop_reclaim = zfsctl_common_reclaim, 622#ifdef TODO 623 .vop_pathconf = zfsctl_pathconf, 624#endif 625 .vop_fid = zfsctl_common_fid, 626}; 627 628static int 629zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname) 630{ 631 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os; 632 633 if (snapshot_namecheck(name, NULL, NULL) != 0) 634 return (EILSEQ); 635 dmu_objset_name(os, zname); 636 if (strlen(zname) + 1 + strlen(name) >= len) 637 return (ENAMETOOLONG); 638 (void) strcat(zname, "@"); 639 (void) strcat(zname, name); 640 return (0); 641} 642 643static int 644zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr) 645{ 646 vnode_t *svp = sep->se_root; 647 int error; 648 649 ASSERT(vn_ismntpt(svp)); 650 651 /* this will be dropped by dounmount() */ 652 if ((error = vn_vfswlock(svp)) != 0) 653 return (error); 654 655#ifdef sun 656 VN_HOLD(svp); 657 error = dounmount(vn_mountedvfs(svp), fflags, cr); 658 if (error) { 659 VN_RELE(svp); 660 return (error); 661 } 662 663 /* 664 * We can't use VN_RELE(), as that will try to invoke 665 * zfsctl_snapdir_inactive(), which would cause us to destroy 666 * the sd_lock mutex held by our caller. 667 */ 668 ASSERT(svp->v_count == 1); 669 gfs_vop_inactive(svp, cr, NULL); 670 671 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 672 kmem_free(sep, sizeof (zfs_snapentry_t)); 673 674 return (0); 675#else /* !sun */ 676 return (dounmount(vn_mountedvfs(svp), fflags, curthread)); 677#endif /* !sun */ 678} 679 680#ifdef sun 681static void 682zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm) 683{ 684 avl_index_t where; 685 vfs_t *vfsp; 686 refstr_t *pathref; 687 char newpath[MAXNAMELEN]; 688 char *tail; 689 690 ASSERT(MUTEX_HELD(&sdp->sd_lock)); 691 ASSERT(sep != NULL); 692 693 vfsp = vn_mountedvfs(sep->se_root); 694 ASSERT(vfsp != NULL); 695 696 vfs_lock_wait(vfsp); 697 698 /* 699 * Change the name in the AVL tree. 700 */ 701 avl_remove(&sdp->sd_snaps, sep); 702 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 703 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 704 (void) strcpy(sep->se_name, nm); 705 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL); 706 avl_insert(&sdp->sd_snaps, sep, where); 707 708 /* 709 * Change the current mountpoint info: 710 * - update the tail of the mntpoint path 711 * - update the tail of the resource path 712 */ 713 pathref = vfs_getmntpoint(vfsp); 714 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 715 VERIFY((tail = strrchr(newpath, '/')) != NULL); 716 *(tail+1) = '\0'; 717 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 718 (void) strcat(newpath, nm); 719 refstr_rele(pathref); 720 vfs_setmntpoint(vfsp, newpath, 0); 721 722 pathref = vfs_getresource(vfsp); 723 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath)); 724 VERIFY((tail = strrchr(newpath, '@')) != NULL); 725 *(tail+1) = '\0'; 726 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath)); 727 (void) strcat(newpath, nm); 728 refstr_rele(pathref); 729 vfs_setresource(vfsp, newpath, 0); 730 731 vfs_unlock(vfsp); 732} 733#endif /* sun */ 734 735#ifdef sun 736/*ARGSUSED*/ 737static int 738zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, 739 cred_t *cr, caller_context_t *ct, int flags) 740{ 741 zfsctl_snapdir_t *sdp = sdvp->v_data; 742 zfs_snapentry_t search, *sep; 743 zfsvfs_t *zfsvfs; 744 avl_index_t where; 745 char from[MAXNAMELEN], to[MAXNAMELEN]; 746 char real[MAXNAMELEN]; 747 int err; 748 749 zfsvfs = sdvp->v_vfsp->vfs_data; 750 ZFS_ENTER(zfsvfs); 751 752 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 753 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real, 754 MAXNAMELEN, NULL); 755 if (err == 0) { 756 snm = real; 757 } else if (err != ENOTSUP) { 758 ZFS_EXIT(zfsvfs); 759 return (err); 760 } 761 } 762 763 ZFS_EXIT(zfsvfs); 764 765 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from); 766 if (!err) 767 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to); 768 if (!err) 769 err = zfs_secpolicy_rename_perms(from, to, cr); 770 if (err) 771 return (err); 772 773 /* 774 * Cannot move snapshots out of the snapdir. 775 */ 776 if (sdvp != tdvp) 777 return (EINVAL); 778 779 if (strcmp(snm, tnm) == 0) 780 return (0); 781 782 mutex_enter(&sdp->sd_lock); 783 784 search.se_name = (char *)snm; 785 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) { 786 mutex_exit(&sdp->sd_lock); 787 return (ENOENT); 788 } 789 790 err = dmu_objset_rename(from, to, 0); 791 if (err == 0) 792 zfsctl_rename_snap(sdp, sep, tnm); 793 794 mutex_exit(&sdp->sd_lock); 795 796 return (err); 797} 798#endif /* sun */ 799 800#ifdef sun 801/* ARGSUSED */ 802static int 803zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 804 caller_context_t *ct, int flags) 805{ 806 zfsctl_snapdir_t *sdp = dvp->v_data; 807 zfs_snapentry_t *sep; 808 zfs_snapentry_t search; 809 zfsvfs_t *zfsvfs; 810 char snapname[MAXNAMELEN]; 811 char real[MAXNAMELEN]; 812 int err; 813 814 zfsvfs = dvp->v_vfsp->vfs_data; 815 ZFS_ENTER(zfsvfs); 816 817 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 818 819 err = dmu_snapshot_realname(zfsvfs->z_os, name, real, 820 MAXNAMELEN, NULL); 821 if (err == 0) { 822 name = real; 823 } else if (err != ENOTSUP) { 824 ZFS_EXIT(zfsvfs); 825 return (err); 826 } 827 } 828 829 ZFS_EXIT(zfsvfs); 830 831 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname); 832 if (!err) 833 err = zfs_secpolicy_destroy_perms(snapname, cr); 834 if (err) 835 return (err); 836 837 mutex_enter(&sdp->sd_lock); 838 839 search.se_name = name; 840 sep = avl_find(&sdp->sd_snaps, &search, NULL); 841 if (sep) { 842 avl_remove(&sdp->sd_snaps, sep); 843 err = zfsctl_unmount_snap(sep, MS_FORCE, cr); 844 if (err) { 845 avl_index_t where; 846 847 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 848 avl_insert(&sdp->sd_snaps, sep, where); 849 } else 850 err = dmu_objset_destroy(snapname, B_FALSE); 851 } else { 852 err = ENOENT; 853 } 854 855 mutex_exit(&sdp->sd_lock); 856 857 return (err); 858} 859#endif /* sun */ 860 861/* 862 * This creates a snapshot under '.zfs/snapshot'. 863 */ 864/* ARGSUSED */ 865static int 866zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, 867 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp) 868{ 869 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 870 char name[MAXNAMELEN]; 871 int err; 872 static enum symfollow follow = NO_FOLLOW; 873 static enum uio_seg seg = UIO_SYSSPACE; 874 875 if (snapshot_namecheck(dirname, NULL, NULL) != 0) 876 return (EILSEQ); 877 878 dmu_objset_name(zfsvfs->z_os, name); 879 880 *vpp = NULL; 881 882 err = zfs_secpolicy_snapshot_perms(name, cr); 883 if (err) 884 return (err); 885 886 if (err == 0) { 887 err = dmu_objset_snapshot(name, dirname, NULL, NULL, 888 B_FALSE, B_FALSE, -1); 889 if (err) 890 return (err); 891 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp); 892 } 893 894 return (err); 895} 896 897static int 898zfsctl_freebsd_snapdir_mkdir(ap) 899 struct vop_mkdir_args /* { 900 struct vnode *a_dvp; 901 struct vnode **a_vpp; 902 struct componentname *a_cnp; 903 struct vattr *a_vap; 904 } */ *ap; 905{ 906 907 ASSERT(ap->a_cnp->cn_flags & SAVENAME); 908 909 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL, 910 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL)); 911} 912 913/* 914 * Lookup entry point for the 'snapshot' directory. Try to open the 915 * snapshot if it exist, creating the pseudo filesystem vnode as necessary. 916 * Perform a mount of the associated dataset on top of the vnode. 917 */ 918/* ARGSUSED */ 919int 920zfsctl_snapdir_lookup(ap) 921 struct vop_lookup_args /* { 922 struct vnode *a_dvp; 923 struct vnode **a_vpp; 924 struct componentname *a_cnp; 925 } */ *ap; 926{ 927 vnode_t *dvp = ap->a_dvp; 928 vnode_t **vpp = ap->a_vpp; 929 struct componentname *cnp = ap->a_cnp; 930 char nm[NAME_MAX + 1]; 931 zfsctl_snapdir_t *sdp = dvp->v_data; 932 objset_t *snap; 933 char snapname[MAXNAMELEN]; 934 char real[MAXNAMELEN]; 935 char *mountpoint; 936 zfs_snapentry_t *sep, search; 937 size_t mountpoint_len; 938 avl_index_t where; 939 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 940 int err; 941 int flags = 0; 942 943 /* 944 * No extended attributes allowed under .zfs 945 */ 946 if (flags & LOOKUP_XATTR) 947 return (EINVAL); 948 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm)); 949 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1); 950 951 ASSERT(dvp->v_type == VDIR); 952 953 *vpp = NULL; 954 955 /* 956 * If we get a recursive call, that means we got called 957 * from the domount() code while it was trying to look up the 958 * spec (which looks like a local path for zfs). We need to 959 * add some flag to domount() to tell it not to do this lookup. 960 */ 961 if (MUTEX_HELD(&sdp->sd_lock)) 962 return (ENOENT); 963 964 ZFS_ENTER(zfsvfs); 965 966 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 967 ZFS_EXIT(zfsvfs); 968 return (0); 969 } 970 971 if (flags & FIGNORECASE) { 972 boolean_t conflict = B_FALSE; 973 974 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real, 975 MAXNAMELEN, &conflict); 976 if (err == 0) { 977 strlcpy(nm, real, sizeof(nm)); 978 } else if (err != ENOTSUP) { 979 ZFS_EXIT(zfsvfs); 980 return (err); 981 } 982#if 0 983 if (realpnp) 984 (void) strlcpy(realpnp->pn_buf, nm, 985 realpnp->pn_bufsize); 986 if (conflict && direntflags) 987 *direntflags = ED_CASE_CONFLICT; 988#endif 989 } 990 991 mutex_enter(&sdp->sd_lock); 992 search.se_name = (char *)nm; 993 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) { 994 *vpp = sep->se_root; 995 VN_HOLD(*vpp); 996 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY); 997 if (err) { 998 VN_RELE(*vpp); 999 *vpp = NULL; 1000 } else if (*vpp == sep->se_root) { 1001 /* 1002 * The snapshot was unmounted behind our backs, 1003 * try to remount it. 1004 */ 1005 goto domount; 1006 } else { 1007 /* 1008 * VROOT was set during the traverse call. We need 1009 * to clear it since we're pretending to be part 1010 * of our parent's vfs. 1011 */ 1012 (*vpp)->v_flag &= ~VROOT; 1013 } 1014 mutex_exit(&sdp->sd_lock); 1015 ZFS_EXIT(zfsvfs); 1016 return (err); 1017 } 1018 1019 /* 1020 * The requested snapshot is not currently mounted, look it up. 1021 */ 1022 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname); 1023 if (err) { 1024 mutex_exit(&sdp->sd_lock); 1025 ZFS_EXIT(zfsvfs); 1026 /* 1027 * handle "ls *" or "?" in a graceful manner, 1028 * forcing EILSEQ to ENOENT. 1029 * Since shell ultimately passes "*" or "?" as name to lookup 1030 */ 1031 return (err == EILSEQ ? ENOENT : err); 1032 } 1033 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) { 1034 mutex_exit(&sdp->sd_lock); 1035 /* Translate errors and add SAVENAME when needed. */ 1036 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 1037 err = EJUSTRETURN; 1038 cnp->cn_flags |= SAVENAME; 1039 } else { 1040 err = ENOENT; 1041 } 1042 ZFS_EXIT(zfsvfs); 1043 return (err); 1044 } 1045 1046 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP); 1047 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP); 1048 (void) strcpy(sep->se_name, nm); 1049 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap)); 1050 VN_HOLD(*vpp); 1051 avl_insert(&sdp->sd_snaps, sep, where); 1052 1053 dmu_objset_rele(snap, FTAG); 1054domount: 1055 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) + 1056 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1; 1057 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP); 1058 (void) snprintf(mountpoint, mountpoint_len, 1059 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s", 1060 dvp->v_vfsp->mnt_stat.f_mntonname, nm); 1061 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0); 1062 kmem_free(mountpoint, mountpoint_len); 1063 if (err == 0) { 1064 /* 1065 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>. 1066 * 1067 * This is where we lie about our v_vfsp in order to 1068 * make .zfs/snapshot/<snapname> accessible over NFS 1069 * without requiring manual mounts of <snapname>. 1070 */ 1071 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs); 1072 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs; 1073 } 1074 mutex_exit(&sdp->sd_lock); 1075 ZFS_EXIT(zfsvfs); 1076 if (err != 0) 1077 *vpp = NULL; 1078 return (err); 1079} 1080 1081/* ARGSUSED */ 1082int 1083zfsctl_shares_lookup(ap) 1084 struct vop_lookup_args /* { 1085 struct vnode *a_dvp; 1086 struct vnode **a_vpp; 1087 struct componentname *a_cnp; 1088 } */ *ap; 1089{ 1090 vnode_t *dvp = ap->a_dvp; 1091 vnode_t **vpp = ap->a_vpp; 1092 struct componentname *cnp = ap->a_cnp; 1093 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1094 char nm[NAME_MAX + 1]; 1095 znode_t *dzp; 1096 int error; 1097 1098 ZFS_ENTER(zfsvfs); 1099 1100 ASSERT(cnp->cn_namelen < sizeof(nm)); 1101 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1); 1102 1103 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) { 1104 ZFS_EXIT(zfsvfs); 1105 return (0); 1106 } 1107 1108 if (zfsvfs->z_shares_dir == 0) { 1109 ZFS_EXIT(zfsvfs); 1110 return (ENOTSUP); 1111 } 1112 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) 1113 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp); 1114 1115 VN_RELE(ZTOV(dzp)); 1116 ZFS_EXIT(zfsvfs); 1117 1118 return (error); 1119} 1120 1121/* ARGSUSED */ 1122static int 1123zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp, 1124 offset_t *offp, offset_t *nextp, void *data, int flags) 1125{ 1126 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1127 char snapname[MAXNAMELEN]; 1128 uint64_t id, cookie; 1129 boolean_t case_conflict; 1130 int error; 1131 1132 ZFS_ENTER(zfsvfs); 1133 1134 cookie = *offp; 1135 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id, 1136 &cookie, &case_conflict); 1137 if (error) { 1138 ZFS_EXIT(zfsvfs); 1139 if (error == ENOENT) { 1140 *eofp = 1; 1141 return (0); 1142 } 1143 return (error); 1144 } 1145 1146 if (flags & V_RDDIR_ENTFLAGS) { 1147 edirent_t *eodp = dp; 1148 1149 (void) strcpy(eodp->ed_name, snapname); 1150 eodp->ed_ino = ZFSCTL_INO_SNAP(id); 1151 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0; 1152 } else { 1153 struct dirent64 *odp = dp; 1154 1155 (void) strcpy(odp->d_name, snapname); 1156 odp->d_ino = ZFSCTL_INO_SNAP(id); 1157 } 1158 *nextp = cookie; 1159 1160 ZFS_EXIT(zfsvfs); 1161 1162 return (0); 1163} 1164 1165/* ARGSUSED */ 1166static int 1167zfsctl_shares_readdir(ap) 1168 struct vop_readdir_args /* { 1169 struct vnode *a_vp; 1170 struct uio *a_uio; 1171 struct ucred *a_cred; 1172 int *a_eofflag; 1173 int *a_ncookies; 1174 u_long **a_cookies; 1175 } */ *ap; 1176{ 1177 vnode_t *vp = ap->a_vp; 1178 uio_t *uiop = ap->a_uio; 1179 cred_t *cr = ap->a_cred; 1180 int *eofp = ap->a_eofflag; 1181 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1182 znode_t *dzp; 1183 int error; 1184 1185 ZFS_ENTER(zfsvfs); 1186 1187 if (zfsvfs->z_shares_dir == 0) { 1188 ZFS_EXIT(zfsvfs); 1189 return (ENOTSUP); 1190 } 1191 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1192 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1193 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies); 1194 VN_URELE(ZTOV(dzp)); 1195 } else { 1196 *eofp = 1; 1197 error = ENOENT; 1198 } 1199 1200 ZFS_EXIT(zfsvfs); 1201 return (error); 1202} 1203 1204/* 1205 * pvp is the '.zfs' directory (zfsctl_node_t). 1206 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t). 1207 * 1208 * This function is the callback to create a GFS vnode for '.zfs/snapshot' 1209 * when a lookup is performed on .zfs for "snapshot". 1210 */ 1211vnode_t * 1212zfsctl_mknode_snapdir(vnode_t *pvp) 1213{ 1214 vnode_t *vp; 1215 zfsctl_snapdir_t *sdp; 1216 1217 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp, 1218 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN, 1219 zfsctl_snapdir_readdir_cb, NULL); 1220 sdp = vp->v_data; 1221 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR; 1222 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1223 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL); 1224 avl_create(&sdp->sd_snaps, snapentry_compare, 1225 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node)); 1226 VOP_UNLOCK(vp, 0); 1227 return (vp); 1228} 1229 1230vnode_t * 1231zfsctl_mknode_shares(vnode_t *pvp) 1232{ 1233 vnode_t *vp; 1234 zfsctl_node_t *sdp; 1235 1236 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1237 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN, 1238 NULL, NULL); 1239 sdp = vp->v_data; 1240 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime; 1241 VOP_UNLOCK(vp, 0); 1242 return (vp); 1243 1244} 1245 1246/* ARGSUSED */ 1247static int 1248zfsctl_shares_getattr(ap) 1249 struct vop_getattr_args /* { 1250 struct vnode *a_vp; 1251 struct vattr *a_vap; 1252 struct ucred *a_cred; 1253 struct thread *a_td; 1254 } */ *ap; 1255{ 1256 vnode_t *vp = ap->a_vp; 1257 vattr_t *vap = ap->a_vap; 1258 cred_t *cr = ap->a_cred; 1259 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1260 znode_t *dzp; 1261 int error; 1262 1263 ZFS_ENTER(zfsvfs); 1264 if (zfsvfs->z_shares_dir == 0) { 1265 ZFS_EXIT(zfsvfs); 1266 return (ENOTSUP); 1267 } 1268 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) { 1269 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY); 1270 error = VOP_GETATTR(ZTOV(dzp), vap, cr); 1271 VN_URELE(ZTOV(dzp)); 1272 } 1273 ZFS_EXIT(zfsvfs); 1274 return (error); 1275 1276 1277} 1278 1279/* ARGSUSED */ 1280static int 1281zfsctl_snapdir_getattr(ap) 1282 struct vop_getattr_args /* { 1283 struct vnode *a_vp; 1284 struct vattr *a_vap; 1285 struct ucred *a_cred; 1286 } */ *ap; 1287{ 1288 vnode_t *vp = ap->a_vp; 1289 vattr_t *vap = ap->a_vap; 1290 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data; 1291 zfsctl_snapdir_t *sdp = vp->v_data; 1292 1293 ZFS_ENTER(zfsvfs); 1294 zfsctl_common_getattr(vp, vap); 1295 vap->va_nodeid = gfs_file_inode(vp); 1296 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2; 1297 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os); 1298 vap->va_birthtime = vap->va_ctime; 1299 ZFS_EXIT(zfsvfs); 1300 1301 return (0); 1302} 1303 1304/* ARGSUSED */ 1305static int 1306zfsctl_snapdir_inactive(ap) 1307 struct vop_inactive_args /* { 1308 struct vnode *a_vp; 1309 struct thread *a_td; 1310 } */ *ap; 1311{ 1312 vnode_t *vp = ap->a_vp; 1313 zfsctl_snapdir_t *sdp = vp->v_data; 1314 zfs_snapentry_t *sep; 1315 1316 /* 1317 * On forced unmount we have to free snapshots from here. 1318 */ 1319 mutex_enter(&sdp->sd_lock); 1320 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) { 1321 avl_remove(&sdp->sd_snaps, sep); 1322 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1323 kmem_free(sep, sizeof (zfs_snapentry_t)); 1324 } 1325 mutex_exit(&sdp->sd_lock); 1326 gfs_dir_inactive(vp); 1327 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0); 1328 mutex_destroy(&sdp->sd_lock); 1329 avl_destroy(&sdp->sd_snaps); 1330 kmem_free(sdp, sizeof (zfsctl_snapdir_t)); 1331 1332 return (0); 1333} 1334 1335#ifdef sun 1336static const fs_operation_def_t zfsctl_tops_snapdir[] = { 1337 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1338 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1339 { VOPNAME_IOCTL, { .error = fs_inval } }, 1340 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } }, 1341 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1342 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } }, 1343 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } }, 1344 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } }, 1345 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } }, 1346 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } }, 1347 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1348 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } }, 1349 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } }, 1350 { NULL } 1351}; 1352 1353static const fs_operation_def_t zfsctl_tops_shares[] = { 1354 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } }, 1355 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } }, 1356 { VOPNAME_IOCTL, { .error = fs_inval } }, 1357 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } }, 1358 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } }, 1359 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } }, 1360 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } }, 1361 { VOPNAME_SEEK, { .vop_seek = fs_seek } }, 1362 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } }, 1363 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } }, 1364 { NULL } 1365}; 1366#else /* !sun */ 1367static struct vop_vector zfsctl_ops_snapdir = { 1368 .vop_default = &default_vnodeops, 1369 .vop_open = zfsctl_common_open, 1370 .vop_close = zfsctl_common_close, 1371 .vop_ioctl = VOP_EINVAL, 1372 .vop_getattr = zfsctl_snapdir_getattr, 1373 .vop_access = zfsctl_common_access, 1374 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir, 1375 .vop_readdir = gfs_vop_readdir, 1376 .vop_lookup = zfsctl_snapdir_lookup, 1377 .vop_inactive = zfsctl_snapdir_inactive, 1378 .vop_reclaim = zfsctl_common_reclaim, 1379 .vop_fid = zfsctl_common_fid, 1380}; 1381 1382static struct vop_vector zfsctl_ops_shares = { 1383 .vop_default = &default_vnodeops, 1384 .vop_open = zfsctl_common_open, 1385 .vop_close = zfsctl_common_close, 1386 .vop_ioctl = VOP_EINVAL, 1387 .vop_getattr = zfsctl_shares_getattr, 1388 .vop_access = zfsctl_common_access, 1389 .vop_readdir = zfsctl_shares_readdir, 1390 .vop_lookup = zfsctl_shares_lookup, 1391 .vop_inactive = gfs_vop_inactive, 1392 .vop_reclaim = zfsctl_common_reclaim, 1393 .vop_fid = zfsctl_shares_fid, 1394}; 1395#endif /* !sun */ 1396 1397/* 1398 * pvp is the GFS vnode '.zfs/snapshot'. 1399 * 1400 * This creates a GFS node under '.zfs/snapshot' representing each 1401 * snapshot. This newly created GFS node is what we mount snapshot 1402 * vfs_t's ontop of. 1403 */ 1404static vnode_t * 1405zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset) 1406{ 1407 vnode_t *vp; 1408 zfsctl_node_t *zcp; 1409 1410 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp, 1411 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL); 1412 VN_HOLD(vp); 1413 zcp = vp->v_data; 1414 zcp->zc_id = objset; 1415 VOP_UNLOCK(vp, 0); 1416 1417 return (vp); 1418} 1419 1420static int 1421zfsctl_snapshot_inactive(ap) 1422 struct vop_inactive_args /* { 1423 struct vnode *a_vp; 1424 struct thread *a_td; 1425 } */ *ap; 1426{ 1427 vnode_t *vp = ap->a_vp; 1428 cred_t *cr = ap->a_td->td_ucred; 1429 struct vop_inactive_args iap; 1430 zfsctl_snapdir_t *sdp; 1431 zfs_snapentry_t *sep, *next; 1432 int locked; 1433 vnode_t *dvp; 1434 1435 if (vp->v_count > 0) 1436 goto end; 1437 1438 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0); 1439 sdp = dvp->v_data; 1440 VOP_UNLOCK(dvp, 0); 1441 1442 if (!(locked = MUTEX_HELD(&sdp->sd_lock))) 1443 mutex_enter(&sdp->sd_lock); 1444 1445 ASSERT(!vn_ismntpt(vp)); 1446 1447 sep = avl_first(&sdp->sd_snaps); 1448 while (sep != NULL) { 1449 next = AVL_NEXT(&sdp->sd_snaps, sep); 1450 1451 if (sep->se_root == vp) { 1452 avl_remove(&sdp->sd_snaps, sep); 1453 kmem_free(sep->se_name, strlen(sep->se_name) + 1); 1454 kmem_free(sep, sizeof (zfs_snapentry_t)); 1455 break; 1456 } 1457 sep = next; 1458 } 1459 ASSERT(sep != NULL); 1460 1461 if (!locked) 1462 mutex_exit(&sdp->sd_lock); 1463 VN_RELE(dvp); 1464 1465end: 1466 /* 1467 * Dispose of the vnode for the snapshot mount point. 1468 * This is safe to do because once this entry has been removed 1469 * from the AVL tree, it can't be found again, so cannot become 1470 * "active". If we lookup the same name again we will end up 1471 * creating a new vnode. 1472 */ 1473 iap.a_vp = vp; 1474 return (gfs_vop_inactive(&iap)); 1475} 1476 1477static int 1478zfsctl_traverse_begin(vnode_t **vpp, int lktype) 1479{ 1480 1481 VN_HOLD(*vpp); 1482 /* Snapshot should be already mounted, but just in case. */ 1483 if (vn_mountedvfs(*vpp) == NULL) 1484 return (ENOENT); 1485 return (traverse(vpp, lktype)); 1486} 1487 1488static void 1489zfsctl_traverse_end(vnode_t *vp, int err) 1490{ 1491 1492 if (err == 0) 1493 vput(vp); 1494 else 1495 VN_RELE(vp); 1496} 1497 1498static int 1499zfsctl_snapshot_getattr(ap) 1500 struct vop_getattr_args /* { 1501 struct vnode *a_vp; 1502 struct vattr *a_vap; 1503 struct ucred *a_cred; 1504 } */ *ap; 1505{ 1506 vnode_t *vp = ap->a_vp; 1507 int err; 1508 1509 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1510 if (err == 0) 1511 err = VOP_GETATTR(vp, ap->a_vap, ap->a_cred); 1512 zfsctl_traverse_end(vp, err); 1513 return (err); 1514} 1515 1516static int 1517zfsctl_snapshot_fid(ap) 1518 struct vop_fid_args /* { 1519 struct vnode *a_vp; 1520 struct fid *a_fid; 1521 } */ *ap; 1522{ 1523 vnode_t *vp = ap->a_vp; 1524 int err; 1525 1526 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY); 1527 if (err == 0) 1528 err = VOP_VPTOFH(vp, (void *)ap->a_fid); 1529 zfsctl_traverse_end(vp, err); 1530 return (err); 1531} 1532 1533static int 1534zfsctl_snapshot_lookup(ap) 1535 struct vop_lookup_args /* { 1536 struct vnode *a_dvp; 1537 struct vnode **a_vpp; 1538 struct componentname *a_cnp; 1539 } */ *ap; 1540{ 1541 vnode_t *dvp = ap->a_dvp; 1542 vnode_t **vpp = ap->a_vpp; 1543 struct componentname *cnp = ap->a_cnp; 1544 cred_t *cr = ap->a_cnp->cn_cred; 1545 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data; 1546 int error; 1547 1548 if (cnp->cn_namelen != 2 || cnp->cn_nameptr[0] != '.' || 1549 cnp->cn_nameptr[1] != '.') { 1550 return (ENOENT); 1551 } 1552 1553 ASSERT(dvp->v_type == VDIR); 1554 ASSERT(zfsvfs->z_ctldir != NULL); 1555 1556 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", vpp, 1557 NULL, 0, NULL, cr, NULL, NULL, NULL); 1558 if (error == 0) 1559 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1560 return (error); 1561} 1562 1563static int 1564zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap) 1565{ 1566 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data; 1567 vnode_t *dvp, *vp; 1568 zfsctl_snapdir_t *sdp; 1569 zfs_snapentry_t *sep; 1570 int error; 1571 1572 ASSERT(zfsvfs->z_ctldir != NULL); 1573 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1574 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1575 if (error != 0) 1576 return (error); 1577 sdp = dvp->v_data; 1578 1579 mutex_enter(&sdp->sd_lock); 1580 sep = avl_first(&sdp->sd_snaps); 1581 while (sep != NULL) { 1582 vp = sep->se_root; 1583 if (vp == ap->a_vp) 1584 break; 1585 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1586 } 1587 if (sep == NULL) { 1588 mutex_exit(&sdp->sd_lock); 1589 error = ENOENT; 1590 } else { 1591 size_t len; 1592 1593 len = strlen(sep->se_name); 1594 *ap->a_buflen -= len; 1595 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len); 1596 mutex_exit(&sdp->sd_lock); 1597 vref(dvp); 1598 *ap->a_vpp = dvp; 1599 } 1600 VN_RELE(dvp); 1601 1602 return (error); 1603} 1604 1605/* 1606 * These VP's should never see the light of day. They should always 1607 * be covered. 1608 */ 1609static struct vop_vector zfsctl_ops_snapshot = { 1610 .vop_default = &default_vnodeops, 1611 .vop_inactive = zfsctl_snapshot_inactive, 1612 .vop_lookup = zfsctl_snapshot_lookup, 1613 .vop_reclaim = zfsctl_common_reclaim, 1614 .vop_getattr = zfsctl_snapshot_getattr, 1615 .vop_fid = zfsctl_snapshot_fid, 1616 .vop_vptocnp = zfsctl_snapshot_vptocnp, 1617}; 1618 1619int 1620zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp) 1621{ 1622 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1623 vnode_t *dvp, *vp; 1624 zfsctl_snapdir_t *sdp; 1625 zfsctl_node_t *zcp; 1626 zfs_snapentry_t *sep; 1627 int error; 1628 1629 ASSERT(zfsvfs->z_ctldir != NULL); 1630 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1631 NULL, 0, NULL, kcred, NULL, NULL, NULL); 1632 if (error != 0) 1633 return (error); 1634 sdp = dvp->v_data; 1635 1636 mutex_enter(&sdp->sd_lock); 1637 sep = avl_first(&sdp->sd_snaps); 1638 while (sep != NULL) { 1639 vp = sep->se_root; 1640 zcp = vp->v_data; 1641 if (zcp->zc_id == objsetid) 1642 break; 1643 1644 sep = AVL_NEXT(&sdp->sd_snaps, sep); 1645 } 1646 1647 if (sep != NULL) { 1648 VN_HOLD(vp); 1649 /* 1650 * Return the mounted root rather than the covered mount point. 1651 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid> 1652 * and returns the ZFS vnode mounted on top of the GFS node. 1653 * This ZFS vnode is the root of the vfs for objset 'objsetid'. 1654 */ 1655 error = traverse(&vp, LK_SHARED | LK_RETRY); 1656 if (error == 0) { 1657 if (vp == sep->se_root) 1658 error = EINVAL; 1659 else 1660 *zfsvfsp = VTOZ(vp)->z_zfsvfs; 1661 } 1662 mutex_exit(&sdp->sd_lock); 1663 if (error == 0) 1664 VN_URELE(vp); 1665 else 1666 VN_RELE(vp); 1667 } else { 1668 error = EINVAL; 1669 mutex_exit(&sdp->sd_lock); 1670 } 1671 1672 VN_RELE(dvp); 1673 1674 return (error); 1675} 1676 1677/* 1678 * Unmount any snapshots for the given filesystem. This is called from 1679 * zfs_umount() - if we have a ctldir, then go through and unmount all the 1680 * snapshots. 1681 */ 1682int 1683zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr) 1684{ 1685 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1686 vnode_t *dvp; 1687 zfsctl_snapdir_t *sdp; 1688 zfs_snapentry_t *sep, *next; 1689 int error; 1690 1691 ASSERT(zfsvfs->z_ctldir != NULL); 1692 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp, 1693 NULL, 0, NULL, cr, NULL, NULL, NULL); 1694 if (error != 0) 1695 return (error); 1696 sdp = dvp->v_data; 1697 1698 mutex_enter(&sdp->sd_lock); 1699 1700 sep = avl_first(&sdp->sd_snaps); 1701 while (sep != NULL) { 1702 next = AVL_NEXT(&sdp->sd_snaps, sep); 1703 1704 /* 1705 * If this snapshot is not mounted, then it must 1706 * have just been unmounted by somebody else, and 1707 * will be cleaned up by zfsctl_snapdir_inactive(). 1708 */ 1709 if (vn_ismntpt(sep->se_root)) { 1710 error = zfsctl_unmount_snap(sep, fflags, cr); 1711 if (error) { 1712 avl_index_t where; 1713 1714 /* 1715 * Before reinserting snapshot to the tree, 1716 * check if it was actually removed. For example 1717 * when snapshot mount point is busy, we will 1718 * have an error here, but there will be no need 1719 * to reinsert snapshot. 1720 */ 1721 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL) 1722 avl_insert(&sdp->sd_snaps, sep, where); 1723 break; 1724 } 1725 } 1726 sep = next; 1727 } 1728 1729 mutex_exit(&sdp->sd_lock); 1730 VN_RELE(dvp); 1731 1732 return (error); 1733} 1734