zfs_vfsops.c revision 302738
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 * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 26 * Copyright (c) 2014 Integros [integros.com] 27 */ 28 29/* Portions Copyright 2010 Robert Milkowski */ 30 31#include <sys/types.h> 32#include <sys/param.h> 33#include <sys/systm.h> 34#include <sys/kernel.h> 35#include <sys/sysmacros.h> 36#include <sys/kmem.h> 37#include <sys/acl.h> 38#include <sys/vnode.h> 39#include <sys/vfs.h> 40#include <sys/mntent.h> 41#include <sys/mount.h> 42#include <sys/cmn_err.h> 43#include <sys/zfs_znode.h> 44#include <sys/zfs_dir.h> 45#include <sys/zil.h> 46#include <sys/fs/zfs.h> 47#include <sys/dmu.h> 48#include <sys/dsl_prop.h> 49#include <sys/dsl_dataset.h> 50#include <sys/dsl_deleg.h> 51#include <sys/spa.h> 52#include <sys/zap.h> 53#include <sys/sa.h> 54#include <sys/sa_impl.h> 55#include <sys/varargs.h> 56#include <sys/policy.h> 57#include <sys/atomic.h> 58#include <sys/zfs_ioctl.h> 59#include <sys/zfs_ctldir.h> 60#include <sys/zfs_fuid.h> 61#include <sys/sunddi.h> 62#include <sys/dnlc.h> 63#include <sys/dmu_objset.h> 64#include <sys/spa_boot.h> 65#include <sys/jail.h> 66#include "zfs_comutil.h" 67 68struct mtx zfs_debug_mtx; 69MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF); 70 71SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system"); 72 73int zfs_super_owner; 74SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0, 75 "File system owner can perform privileged operation on his file systems"); 76 77int zfs_debug_level; 78TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level); 79SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0, 80 "Debug level"); 81 82SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions"); 83static int zfs_version_acl = ZFS_ACL_VERSION; 84SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0, 85 "ZFS_ACL_VERSION"); 86static int zfs_version_spa = SPA_VERSION; 87SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0, 88 "SPA_VERSION"); 89static int zfs_version_zpl = ZPL_VERSION; 90SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0, 91 "ZPL_VERSION"); 92 93static int zfs_mount(vfs_t *vfsp); 94static int zfs_umount(vfs_t *vfsp, int fflag); 95static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp); 96static int zfs_statfs(vfs_t *vfsp, struct statfs *statp); 97static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp); 98static int zfs_sync(vfs_t *vfsp, int waitfor); 99static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 100 struct ucred **credanonp, int *numsecflavors, int **secflavors); 101static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp); 102static void zfs_objset_close(zfsvfs_t *zfsvfs); 103static void zfs_freevfs(vfs_t *vfsp); 104 105static struct vfsops zfs_vfsops = { 106 .vfs_mount = zfs_mount, 107 .vfs_unmount = zfs_umount, 108 .vfs_root = zfs_root, 109 .vfs_statfs = zfs_statfs, 110 .vfs_vget = zfs_vget, 111 .vfs_sync = zfs_sync, 112 .vfs_checkexp = zfs_checkexp, 113 .vfs_fhtovp = zfs_fhtovp, 114}; 115 116VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN); 117 118/* 119 * We need to keep a count of active fs's. 120 * This is necessary to prevent our module 121 * from being unloaded after a umount -f 122 */ 123static uint32_t zfs_active_fs_count = 0; 124 125/*ARGSUSED*/ 126static int 127zfs_sync(vfs_t *vfsp, int waitfor) 128{ 129 130 /* 131 * Data integrity is job one. We don't want a compromised kernel 132 * writing to the storage pool, so we never sync during panic. 133 */ 134 if (panicstr) 135 return (0); 136 137 /* 138 * Ignore the system syncher. ZFS already commits async data 139 * at zfs_txg_timeout intervals. 140 */ 141 if (waitfor == MNT_LAZY) 142 return (0); 143 144 if (vfsp != NULL) { 145 /* 146 * Sync a specific filesystem. 147 */ 148 zfsvfs_t *zfsvfs = vfsp->vfs_data; 149 dsl_pool_t *dp; 150 int error; 151 152 error = vfs_stdsync(vfsp, waitfor); 153 if (error != 0) 154 return (error); 155 156 ZFS_ENTER(zfsvfs); 157 dp = dmu_objset_pool(zfsvfs->z_os); 158 159 /* 160 * If the system is shutting down, then skip any 161 * filesystems which may exist on a suspended pool. 162 */ 163 if (sys_shutdown && spa_suspended(dp->dp_spa)) { 164 ZFS_EXIT(zfsvfs); 165 return (0); 166 } 167 168 if (zfsvfs->z_log != NULL) 169 zil_commit(zfsvfs->z_log, 0); 170 171 ZFS_EXIT(zfsvfs); 172 } else { 173 /* 174 * Sync all ZFS filesystems. This is what happens when you 175 * run sync(1M). Unlike other filesystems, ZFS honors the 176 * request by waiting for all pools to commit all dirty data. 177 */ 178 spa_sync_allpools(); 179 } 180 181 return (0); 182} 183 184#ifndef __FreeBSD_kernel__ 185static int 186zfs_create_unique_device(dev_t *dev) 187{ 188 major_t new_major; 189 190 do { 191 ASSERT3U(zfs_minor, <=, MAXMIN32); 192 minor_t start = zfs_minor; 193 do { 194 mutex_enter(&zfs_dev_mtx); 195 if (zfs_minor >= MAXMIN32) { 196 /* 197 * If we're still using the real major 198 * keep out of /dev/zfs and /dev/zvol minor 199 * number space. If we're using a getudev()'ed 200 * major number, we can use all of its minors. 201 */ 202 if (zfs_major == ddi_name_to_major(ZFS_DRIVER)) 203 zfs_minor = ZFS_MIN_MINOR; 204 else 205 zfs_minor = 0; 206 } else { 207 zfs_minor++; 208 } 209 *dev = makedevice(zfs_major, zfs_minor); 210 mutex_exit(&zfs_dev_mtx); 211 } while (vfs_devismounted(*dev) && zfs_minor != start); 212 if (zfs_minor == start) { 213 /* 214 * We are using all ~262,000 minor numbers for the 215 * current major number. Create a new major number. 216 */ 217 if ((new_major = getudev()) == (major_t)-1) { 218 cmn_err(CE_WARN, 219 "zfs_mount: Can't get unique major " 220 "device number."); 221 return (-1); 222 } 223 mutex_enter(&zfs_dev_mtx); 224 zfs_major = new_major; 225 zfs_minor = 0; 226 227 mutex_exit(&zfs_dev_mtx); 228 } else { 229 break; 230 } 231 /* CONSTANTCONDITION */ 232 } while (1); 233 234 return (0); 235} 236#endif /* !__FreeBSD_kernel__ */ 237 238static void 239atime_changed_cb(void *arg, uint64_t newval) 240{ 241 zfsvfs_t *zfsvfs = arg; 242 243 if (newval == TRUE) { 244 zfsvfs->z_atime = TRUE; 245 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; 246 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); 247 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); 248 } else { 249 zfsvfs->z_atime = FALSE; 250 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; 251 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); 252 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); 253 } 254} 255 256static void 257xattr_changed_cb(void *arg, uint64_t newval) 258{ 259 zfsvfs_t *zfsvfs = arg; 260 261 if (newval == TRUE) { 262 /* XXX locking on vfs_flag? */ 263#ifdef TODO 264 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; 265#endif 266 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); 267 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); 268 } else { 269 /* XXX locking on vfs_flag? */ 270#ifdef TODO 271 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR; 272#endif 273 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR); 274 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); 275 } 276} 277 278static void 279blksz_changed_cb(void *arg, uint64_t newval) 280{ 281 zfsvfs_t *zfsvfs = arg; 282 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os))); 283 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); 284 ASSERT(ISP2(newval)); 285 286 zfsvfs->z_max_blksz = newval; 287 zfsvfs->z_vfs->mnt_stat.f_iosize = newval; 288} 289 290static void 291readonly_changed_cb(void *arg, uint64_t newval) 292{ 293 zfsvfs_t *zfsvfs = arg; 294 295 if (newval) { 296 /* XXX locking on vfs_flag? */ 297 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; 298 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); 299 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); 300 } else { 301 /* XXX locking on vfs_flag? */ 302 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 303 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); 304 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); 305 } 306} 307 308static void 309setuid_changed_cb(void *arg, uint64_t newval) 310{ 311 zfsvfs_t *zfsvfs = arg; 312 313 if (newval == FALSE) { 314 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; 315 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); 316 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); 317 } else { 318 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; 319 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); 320 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); 321 } 322} 323 324static void 325exec_changed_cb(void *arg, uint64_t newval) 326{ 327 zfsvfs_t *zfsvfs = arg; 328 329 if (newval == FALSE) { 330 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; 331 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); 332 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); 333 } else { 334 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; 335 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); 336 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); 337 } 338} 339 340/* 341 * The nbmand mount option can be changed at mount time. 342 * We can't allow it to be toggled on live file systems or incorrect 343 * behavior may be seen from cifs clients 344 * 345 * This property isn't registered via dsl_prop_register(), but this callback 346 * will be called when a file system is first mounted 347 */ 348static void 349nbmand_changed_cb(void *arg, uint64_t newval) 350{ 351 zfsvfs_t *zfsvfs = arg; 352 if (newval == FALSE) { 353 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); 354 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); 355 } else { 356 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); 357 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); 358 } 359} 360 361static void 362snapdir_changed_cb(void *arg, uint64_t newval) 363{ 364 zfsvfs_t *zfsvfs = arg; 365 366 zfsvfs->z_show_ctldir = newval; 367} 368 369static void 370vscan_changed_cb(void *arg, uint64_t newval) 371{ 372 zfsvfs_t *zfsvfs = arg; 373 374 zfsvfs->z_vscan = newval; 375} 376 377static void 378acl_mode_changed_cb(void *arg, uint64_t newval) 379{ 380 zfsvfs_t *zfsvfs = arg; 381 382 zfsvfs->z_acl_mode = newval; 383} 384 385static void 386acl_inherit_changed_cb(void *arg, uint64_t newval) 387{ 388 zfsvfs_t *zfsvfs = arg; 389 390 zfsvfs->z_acl_inherit = newval; 391} 392 393static int 394zfs_register_callbacks(vfs_t *vfsp) 395{ 396 struct dsl_dataset *ds = NULL; 397 objset_t *os = NULL; 398 zfsvfs_t *zfsvfs = NULL; 399 uint64_t nbmand; 400 boolean_t readonly = B_FALSE; 401 boolean_t do_readonly = B_FALSE; 402 boolean_t setuid = B_FALSE; 403 boolean_t do_setuid = B_FALSE; 404 boolean_t exec = B_FALSE; 405 boolean_t do_exec = B_FALSE; 406#ifdef illumos 407 boolean_t devices = B_FALSE; 408 boolean_t do_devices = B_FALSE; 409#endif 410 boolean_t xattr = B_FALSE; 411 boolean_t do_xattr = B_FALSE; 412 boolean_t atime = B_FALSE; 413 boolean_t do_atime = B_FALSE; 414 int error = 0; 415 416 ASSERT(vfsp); 417 zfsvfs = vfsp->vfs_data; 418 ASSERT(zfsvfs); 419 os = zfsvfs->z_os; 420 421 /* 422 * This function can be called for a snapshot when we update snapshot's 423 * mount point, which isn't really supported. 424 */ 425 if (dmu_objset_is_snapshot(os)) 426 return (EOPNOTSUPP); 427 428 /* 429 * The act of registering our callbacks will destroy any mount 430 * options we may have. In order to enable temporary overrides 431 * of mount options, we stash away the current values and 432 * restore them after we register the callbacks. 433 */ 434 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) || 435 !spa_writeable(dmu_objset_spa(os))) { 436 readonly = B_TRUE; 437 do_readonly = B_TRUE; 438 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 439 readonly = B_FALSE; 440 do_readonly = B_TRUE; 441 } 442 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 443 setuid = B_FALSE; 444 do_setuid = B_TRUE; 445 } else { 446 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 447 setuid = B_FALSE; 448 do_setuid = B_TRUE; 449 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 450 setuid = B_TRUE; 451 do_setuid = B_TRUE; 452 } 453 } 454 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 455 exec = B_FALSE; 456 do_exec = B_TRUE; 457 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 458 exec = B_TRUE; 459 do_exec = B_TRUE; 460 } 461 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 462 xattr = B_FALSE; 463 do_xattr = B_TRUE; 464 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 465 xattr = B_TRUE; 466 do_xattr = B_TRUE; 467 } 468 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 469 atime = B_FALSE; 470 do_atime = B_TRUE; 471 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 472 atime = B_TRUE; 473 do_atime = B_TRUE; 474 } 475 476 /* 477 * We need to enter pool configuration here, so that we can use 478 * dsl_prop_get_int_ds() to handle the special nbmand property below. 479 * dsl_prop_get_integer() can not be used, because it has to acquire 480 * spa_namespace_lock and we can not do that because we already hold 481 * z_teardown_lock. The problem is that spa_config_sync() is called 482 * with spa_namespace_lock held and the function calls ZFS vnode 483 * operations to write the cache file and thus z_teardown_lock is 484 * acquired after spa_namespace_lock. 485 */ 486 ds = dmu_objset_ds(os); 487 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 488 489 /* 490 * nbmand is a special property. It can only be changed at 491 * mount time. 492 * 493 * This is weird, but it is documented to only be changeable 494 * at mount time. 495 */ 496 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 497 nbmand = B_FALSE; 498 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 499 nbmand = B_TRUE; 500 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) { 501 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 502 return (error); 503 } 504 505 /* 506 * Register property callbacks. 507 * 508 * It would probably be fine to just check for i/o error from 509 * the first prop_register(), but I guess I like to go 510 * overboard... 511 */ 512 error = dsl_prop_register(ds, 513 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 514 error = error ? error : dsl_prop_register(ds, 515 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 516 error = error ? error : dsl_prop_register(ds, 517 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 518 error = error ? error : dsl_prop_register(ds, 519 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 520#ifdef illumos 521 error = error ? error : dsl_prop_register(ds, 522 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 523#endif 524 error = error ? error : dsl_prop_register(ds, 525 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 526 error = error ? error : dsl_prop_register(ds, 527 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 528 error = error ? error : dsl_prop_register(ds, 529 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 530 error = error ? error : dsl_prop_register(ds, 531 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 532 error = error ? error : dsl_prop_register(ds, 533 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 534 zfsvfs); 535 error = error ? error : dsl_prop_register(ds, 536 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); 537 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 538 if (error) 539 goto unregister; 540 541 /* 542 * Invoke our callbacks to restore temporary mount options. 543 */ 544 if (do_readonly) 545 readonly_changed_cb(zfsvfs, readonly); 546 if (do_setuid) 547 setuid_changed_cb(zfsvfs, setuid); 548 if (do_exec) 549 exec_changed_cb(zfsvfs, exec); 550 if (do_xattr) 551 xattr_changed_cb(zfsvfs, xattr); 552 if (do_atime) 553 atime_changed_cb(zfsvfs, atime); 554 555 nbmand_changed_cb(zfsvfs, nbmand); 556 557 return (0); 558 559unregister: 560 dsl_prop_unregister_all(ds, zfsvfs); 561 return (error); 562} 563 564static int 565zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, 566 uint64_t *userp, uint64_t *groupp) 567{ 568 /* 569 * Is it a valid type of object to track? 570 */ 571 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 572 return (SET_ERROR(ENOENT)); 573 574 /* 575 * If we have a NULL data pointer 576 * then assume the id's aren't changing and 577 * return EEXIST to the dmu to let it know to 578 * use the same ids 579 */ 580 if (data == NULL) 581 return (SET_ERROR(EEXIST)); 582 583 if (bonustype == DMU_OT_ZNODE) { 584 znode_phys_t *znp = data; 585 *userp = znp->zp_uid; 586 *groupp = znp->zp_gid; 587 } else { 588 int hdrsize; 589 sa_hdr_phys_t *sap = data; 590 sa_hdr_phys_t sa = *sap; 591 boolean_t swap = B_FALSE; 592 593 ASSERT(bonustype == DMU_OT_SA); 594 595 if (sa.sa_magic == 0) { 596 /* 597 * This should only happen for newly created 598 * files that haven't had the znode data filled 599 * in yet. 600 */ 601 *userp = 0; 602 *groupp = 0; 603 return (0); 604 } 605 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { 606 sa.sa_magic = SA_MAGIC; 607 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); 608 swap = B_TRUE; 609 } else { 610 VERIFY3U(sa.sa_magic, ==, SA_MAGIC); 611 } 612 613 hdrsize = sa_hdrsize(&sa); 614 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); 615 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + 616 SA_UID_OFFSET)); 617 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + 618 SA_GID_OFFSET)); 619 if (swap) { 620 *userp = BSWAP_64(*userp); 621 *groupp = BSWAP_64(*groupp); 622 } 623 } 624 return (0); 625} 626 627static void 628fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, 629 char *domainbuf, int buflen, uid_t *ridp) 630{ 631 uint64_t fuid; 632 const char *domain; 633 634 fuid = strtonum(fuidstr, NULL); 635 636 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); 637 if (domain) 638 (void) strlcpy(domainbuf, domain, buflen); 639 else 640 domainbuf[0] = '\0'; 641 *ridp = FUID_RID(fuid); 642} 643 644static uint64_t 645zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) 646{ 647 switch (type) { 648 case ZFS_PROP_USERUSED: 649 return (DMU_USERUSED_OBJECT); 650 case ZFS_PROP_GROUPUSED: 651 return (DMU_GROUPUSED_OBJECT); 652 case ZFS_PROP_USERQUOTA: 653 return (zfsvfs->z_userquota_obj); 654 case ZFS_PROP_GROUPQUOTA: 655 return (zfsvfs->z_groupquota_obj); 656 } 657 return (0); 658} 659 660int 661zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 662 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) 663{ 664 int error; 665 zap_cursor_t zc; 666 zap_attribute_t za; 667 zfs_useracct_t *buf = vbuf; 668 uint64_t obj; 669 670 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 671 return (SET_ERROR(ENOTSUP)); 672 673 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 674 if (obj == 0) { 675 *bufsizep = 0; 676 return (0); 677 } 678 679 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); 680 (error = zap_cursor_retrieve(&zc, &za)) == 0; 681 zap_cursor_advance(&zc)) { 682 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > 683 *bufsizep) 684 break; 685 686 fuidstr_to_sid(zfsvfs, za.za_name, 687 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); 688 689 buf->zu_space = za.za_first_integer; 690 buf++; 691 } 692 if (error == ENOENT) 693 error = 0; 694 695 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); 696 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; 697 *cookiep = zap_cursor_serialize(&zc); 698 zap_cursor_fini(&zc); 699 return (error); 700} 701 702/* 703 * buf must be big enough (eg, 32 bytes) 704 */ 705static int 706id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, 707 char *buf, boolean_t addok) 708{ 709 uint64_t fuid; 710 int domainid = 0; 711 712 if (domain && domain[0]) { 713 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); 714 if (domainid == -1) 715 return (SET_ERROR(ENOENT)); 716 } 717 fuid = FUID_ENCODE(domainid, rid); 718 (void) sprintf(buf, "%llx", (longlong_t)fuid); 719 return (0); 720} 721 722int 723zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 724 const char *domain, uint64_t rid, uint64_t *valp) 725{ 726 char buf[32]; 727 int err; 728 uint64_t obj; 729 730 *valp = 0; 731 732 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 733 return (SET_ERROR(ENOTSUP)); 734 735 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 736 if (obj == 0) 737 return (0); 738 739 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); 740 if (err) 741 return (err); 742 743 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); 744 if (err == ENOENT) 745 err = 0; 746 return (err); 747} 748 749int 750zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 751 const char *domain, uint64_t rid, uint64_t quota) 752{ 753 char buf[32]; 754 int err; 755 dmu_tx_t *tx; 756 uint64_t *objp; 757 boolean_t fuid_dirtied; 758 759 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) 760 return (SET_ERROR(EINVAL)); 761 762 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) 763 return (SET_ERROR(ENOTSUP)); 764 765 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : 766 &zfsvfs->z_groupquota_obj; 767 768 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); 769 if (err) 770 return (err); 771 fuid_dirtied = zfsvfs->z_fuid_dirty; 772 773 tx = dmu_tx_create(zfsvfs->z_os); 774 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); 775 if (*objp == 0) { 776 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 777 zfs_userquota_prop_prefixes[type]); 778 } 779 if (fuid_dirtied) 780 zfs_fuid_txhold(zfsvfs, tx); 781 err = dmu_tx_assign(tx, TXG_WAIT); 782 if (err) { 783 dmu_tx_abort(tx); 784 return (err); 785 } 786 787 mutex_enter(&zfsvfs->z_lock); 788 if (*objp == 0) { 789 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, 790 DMU_OT_NONE, 0, tx); 791 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 792 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); 793 } 794 mutex_exit(&zfsvfs->z_lock); 795 796 if (quota == 0) { 797 err = zap_remove(zfsvfs->z_os, *objp, buf, tx); 798 if (err == ENOENT) 799 err = 0; 800 } else { 801 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); 802 } 803 ASSERT(err == 0); 804 if (fuid_dirtied) 805 zfs_fuid_sync(zfsvfs, tx); 806 dmu_tx_commit(tx); 807 return (err); 808} 809 810boolean_t 811zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) 812{ 813 char buf[32]; 814 uint64_t used, quota, usedobj, quotaobj; 815 int err; 816 817 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; 818 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 819 820 if (quotaobj == 0 || zfsvfs->z_replay) 821 return (B_FALSE); 822 823 (void) sprintf(buf, "%llx", (longlong_t)fuid); 824 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); 825 if (err != 0) 826 return (B_FALSE); 827 828 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); 829 if (err != 0) 830 return (B_FALSE); 831 return (used >= quota); 832} 833 834boolean_t 835zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) 836{ 837 uint64_t fuid; 838 uint64_t quotaobj; 839 840 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 841 842 fuid = isgroup ? zp->z_gid : zp->z_uid; 843 844 if (quotaobj == 0 || zfsvfs->z_replay) 845 return (B_FALSE); 846 847 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); 848} 849 850int 851zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 852{ 853 objset_t *os; 854 zfsvfs_t *zfsvfs; 855 uint64_t zval; 856 int i, error; 857 uint64_t sa_obj; 858 859 /* 860 * XXX: Fix struct statfs so this isn't necessary! 861 * 862 * The 'osname' is used as the filesystem's special node, which means 863 * it must fit in statfs.f_mntfromname, or else it can't be 864 * enumerated, so libzfs_mnttab_find() returns NULL, which causes 865 * 'zfs unmount' to think it's not mounted when it is. 866 */ 867 if (strlen(osname) >= MNAMELEN) 868 return (SET_ERROR(ENAMETOOLONG)); 869 870 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 871 872 /* 873 * We claim to always be readonly so we can open snapshots; 874 * other ZPL code will prevent us from writing to snapshots. 875 */ 876 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 877 if (error) { 878 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 879 return (error); 880 } 881 882 /* 883 * Initialize the zfs-specific filesystem structure. 884 * Should probably make this a kmem cache, shuffle fields, 885 * and just bzero up to z_hold_mtx[]. 886 */ 887 zfsvfs->z_vfs = NULL; 888 zfsvfs->z_parent = zfsvfs; 889 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 890 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 891 zfsvfs->z_os = os; 892 893 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 894 if (error) { 895 goto out; 896 } else if (zfsvfs->z_version > 897 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 898 (void) printf("Can't mount a version %lld file system " 899 "on a version %lld pool\n. Pool must be upgraded to mount " 900 "this file system.", (u_longlong_t)zfsvfs->z_version, 901 (u_longlong_t)spa_version(dmu_objset_spa(os))); 902 error = SET_ERROR(ENOTSUP); 903 goto out; 904 } 905 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) 906 goto out; 907 zfsvfs->z_norm = (int)zval; 908 909 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) 910 goto out; 911 zfsvfs->z_utf8 = (zval != 0); 912 913 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) 914 goto out; 915 zfsvfs->z_case = (uint_t)zval; 916 917 /* 918 * Fold case on file systems that are always or sometimes case 919 * insensitive. 920 */ 921 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 922 zfsvfs->z_case == ZFS_CASE_MIXED) 923 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 924 925 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 926 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 927 928 if (zfsvfs->z_use_sa) { 929 /* should either have both of these objects or none */ 930 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 931 &sa_obj); 932 if (error) 933 goto out; 934 } else { 935 /* 936 * Pre SA versions file systems should never touch 937 * either the attribute registration or layout objects. 938 */ 939 sa_obj = 0; 940 } 941 942 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 943 &zfsvfs->z_attr_table); 944 if (error) 945 goto out; 946 947 if (zfsvfs->z_version >= ZPL_VERSION_SA) 948 sa_register_update_callback(os, zfs_sa_upgrade); 949 950 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 951 &zfsvfs->z_root); 952 if (error) 953 goto out; 954 ASSERT(zfsvfs->z_root != 0); 955 956 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 957 &zfsvfs->z_unlinkedobj); 958 if (error) 959 goto out; 960 961 error = zap_lookup(os, MASTER_NODE_OBJ, 962 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 963 8, 1, &zfsvfs->z_userquota_obj); 964 if (error && error != ENOENT) 965 goto out; 966 967 error = zap_lookup(os, MASTER_NODE_OBJ, 968 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 969 8, 1, &zfsvfs->z_groupquota_obj); 970 if (error && error != ENOENT) 971 goto out; 972 973 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 974 &zfsvfs->z_fuid_obj); 975 if (error && error != ENOENT) 976 goto out; 977 978 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 979 &zfsvfs->z_shares_dir); 980 if (error && error != ENOENT) 981 goto out; 982 983 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 984 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 985 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 986 offsetof(znode_t, z_link_node)); 987 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE); 988 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 989 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 990 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 991 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 992 993 *zfvp = zfsvfs; 994 return (0); 995 996out: 997 dmu_objset_disown(os, zfsvfs); 998 *zfvp = NULL; 999 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1000 return (error); 1001} 1002 1003static int 1004zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 1005{ 1006 int error; 1007 1008 error = zfs_register_callbacks(zfsvfs->z_vfs); 1009 if (error) 1010 return (error); 1011 1012 /* 1013 * Set the objset user_ptr to track its zfsvfs. 1014 */ 1015 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1016 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1017 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1018 1019 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 1020 1021 /* 1022 * If we are not mounting (ie: online recv), then we don't 1023 * have to worry about replaying the log as we blocked all 1024 * operations out since we closed the ZIL. 1025 */ 1026 if (mounting) { 1027 boolean_t readonly; 1028 1029 /* 1030 * During replay we remove the read only flag to 1031 * allow replays to succeed. 1032 */ 1033 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1034 if (readonly != 0) 1035 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1036 else 1037 zfs_unlinked_drain(zfsvfs); 1038 1039 /* 1040 * Parse and replay the intent log. 1041 * 1042 * Because of ziltest, this must be done after 1043 * zfs_unlinked_drain(). (Further note: ziltest 1044 * doesn't use readonly mounts, where 1045 * zfs_unlinked_drain() isn't called.) This is because 1046 * ziltest causes spa_sync() to think it's committed, 1047 * but actually it is not, so the intent log contains 1048 * many txg's worth of changes. 1049 * 1050 * In particular, if object N is in the unlinked set in 1051 * the last txg to actually sync, then it could be 1052 * actually freed in a later txg and then reallocated 1053 * in a yet later txg. This would write a "create 1054 * object N" record to the intent log. Normally, this 1055 * would be fine because the spa_sync() would have 1056 * written out the fact that object N is free, before 1057 * we could write the "create object N" intent log 1058 * record. 1059 * 1060 * But when we are in ziltest mode, we advance the "open 1061 * txg" without actually spa_sync()-ing the changes to 1062 * disk. So we would see that object N is still 1063 * allocated and in the unlinked set, and there is an 1064 * intent log record saying to allocate it. 1065 */ 1066 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1067 if (zil_replay_disable) { 1068 zil_destroy(zfsvfs->z_log, B_FALSE); 1069 } else { 1070 zfsvfs->z_replay = B_TRUE; 1071 zil_replay(zfsvfs->z_os, zfsvfs, 1072 zfs_replay_vector); 1073 zfsvfs->z_replay = B_FALSE; 1074 } 1075 } 1076 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1077 } 1078 1079 return (0); 1080} 1081 1082extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1083 1084void 1085zfsvfs_free(zfsvfs_t *zfsvfs) 1086{ 1087 int i; 1088 1089 /* 1090 * This is a barrier to prevent the filesystem from going away in 1091 * zfs_znode_move() until we can safely ensure that the filesystem is 1092 * not unmounted. We consider the filesystem valid before the barrier 1093 * and invalid after the barrier. 1094 */ 1095 rw_enter(&zfsvfs_lock, RW_READER); 1096 rw_exit(&zfsvfs_lock); 1097 1098 zfs_fuid_destroy(zfsvfs); 1099 1100 mutex_destroy(&zfsvfs->z_znodes_lock); 1101 mutex_destroy(&zfsvfs->z_lock); 1102 list_destroy(&zfsvfs->z_all_znodes); 1103 rrm_destroy(&zfsvfs->z_teardown_lock); 1104 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1105 rw_destroy(&zfsvfs->z_fuid_lock); 1106 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1107 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1108 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1109} 1110 1111static void 1112zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1113{ 1114 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1115 if (zfsvfs->z_vfs) { 1116 if (zfsvfs->z_use_fuids) { 1117 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1118 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1119 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1120 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1121 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1122 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1123 } else { 1124 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1125 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1126 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1127 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1128 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1129 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1130 } 1131 } 1132 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1133} 1134 1135static int 1136zfs_domount(vfs_t *vfsp, char *osname) 1137{ 1138 uint64_t recordsize, fsid_guid; 1139 int error = 0; 1140 zfsvfs_t *zfsvfs; 1141 vnode_t *vp; 1142 1143 ASSERT(vfsp); 1144 ASSERT(osname); 1145 1146 error = zfsvfs_create(osname, &zfsvfs); 1147 if (error) 1148 return (error); 1149 zfsvfs->z_vfs = vfsp; 1150 1151#ifdef illumos 1152 /* Initialize the generic filesystem structure. */ 1153 vfsp->vfs_bcount = 0; 1154 vfsp->vfs_data = NULL; 1155 1156 if (zfs_create_unique_device(&mount_dev) == -1) { 1157 error = SET_ERROR(ENODEV); 1158 goto out; 1159 } 1160 ASSERT(vfs_devismounted(mount_dev) == 0); 1161#endif 1162 1163 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1164 NULL)) 1165 goto out; 1166 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1167 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1168 1169 vfsp->vfs_data = zfsvfs; 1170 vfsp->mnt_flag |= MNT_LOCAL; 1171 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1172 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1173 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1174 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */ 1175 1176 /* 1177 * The fsid is 64 bits, composed of an 8-bit fs type, which 1178 * separates our fsid from any other filesystem types, and a 1179 * 56-bit objset unique ID. The objset unique ID is unique to 1180 * all objsets open on this system, provided by unique_create(). 1181 * The 8-bit fs type must be put in the low bits of fsid[1] 1182 * because that's where other Solaris filesystems put it. 1183 */ 1184 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1185 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1186 vfsp->vfs_fsid.val[0] = fsid_guid; 1187 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1188 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1189 1190 /* 1191 * Set features for file system. 1192 */ 1193 zfs_set_fuid_feature(zfsvfs); 1194 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1195 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1196 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1197 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1198 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1199 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1200 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1201 } 1202 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1203 1204 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1205 uint64_t pval; 1206 1207 atime_changed_cb(zfsvfs, B_FALSE); 1208 readonly_changed_cb(zfsvfs, B_TRUE); 1209 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1210 goto out; 1211 xattr_changed_cb(zfsvfs, pval); 1212 zfsvfs->z_issnap = B_TRUE; 1213 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1214 1215 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1216 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1217 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1218 } else { 1219 error = zfsvfs_setup(zfsvfs, B_TRUE); 1220 } 1221 1222 vfs_mountedfrom(vfsp, osname); 1223 1224 if (!zfsvfs->z_issnap) 1225 zfsctl_create(zfsvfs); 1226out: 1227 if (error) { 1228 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1229 zfsvfs_free(zfsvfs); 1230 } else { 1231 atomic_inc_32(&zfs_active_fs_count); 1232 } 1233 1234 return (error); 1235} 1236 1237void 1238zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1239{ 1240 objset_t *os = zfsvfs->z_os; 1241 1242 if (!dmu_objset_is_snapshot(os)) 1243 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 1244} 1245 1246#ifdef SECLABEL 1247/* 1248 * Convert a decimal digit string to a uint64_t integer. 1249 */ 1250static int 1251str_to_uint64(char *str, uint64_t *objnum) 1252{ 1253 uint64_t num = 0; 1254 1255 while (*str) { 1256 if (*str < '0' || *str > '9') 1257 return (SET_ERROR(EINVAL)); 1258 1259 num = num*10 + *str++ - '0'; 1260 } 1261 1262 *objnum = num; 1263 return (0); 1264} 1265 1266/* 1267 * The boot path passed from the boot loader is in the form of 1268 * "rootpool-name/root-filesystem-object-number'. Convert this 1269 * string to a dataset name: "rootpool-name/root-filesystem-name". 1270 */ 1271static int 1272zfs_parse_bootfs(char *bpath, char *outpath) 1273{ 1274 char *slashp; 1275 uint64_t objnum; 1276 int error; 1277 1278 if (*bpath == 0 || *bpath == '/') 1279 return (SET_ERROR(EINVAL)); 1280 1281 (void) strcpy(outpath, bpath); 1282 1283 slashp = strchr(bpath, '/'); 1284 1285 /* if no '/', just return the pool name */ 1286 if (slashp == NULL) { 1287 return (0); 1288 } 1289 1290 /* if not a number, just return the root dataset name */ 1291 if (str_to_uint64(slashp+1, &objnum)) { 1292 return (0); 1293 } 1294 1295 *slashp = '\0'; 1296 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1297 *slashp = '/'; 1298 1299 return (error); 1300} 1301 1302/* 1303 * Check that the hex label string is appropriate for the dataset being 1304 * mounted into the global_zone proper. 1305 * 1306 * Return an error if the hex label string is not default or 1307 * admin_low/admin_high. For admin_low labels, the corresponding 1308 * dataset must be readonly. 1309 */ 1310int 1311zfs_check_global_label(const char *dsname, const char *hexsl) 1312{ 1313 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1314 return (0); 1315 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1316 return (0); 1317 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1318 /* must be readonly */ 1319 uint64_t rdonly; 1320 1321 if (dsl_prop_get_integer(dsname, 1322 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1323 return (SET_ERROR(EACCES)); 1324 return (rdonly ? 0 : EACCES); 1325 } 1326 return (SET_ERROR(EACCES)); 1327} 1328 1329/* 1330 * Determine whether the mount is allowed according to MAC check. 1331 * by comparing (where appropriate) label of the dataset against 1332 * the label of the zone being mounted into. If the dataset has 1333 * no label, create one. 1334 * 1335 * Returns 0 if access allowed, error otherwise (e.g. EACCES) 1336 */ 1337static int 1338zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1339{ 1340 int error, retv; 1341 zone_t *mntzone = NULL; 1342 ts_label_t *mnt_tsl; 1343 bslabel_t *mnt_sl; 1344 bslabel_t ds_sl; 1345 char ds_hexsl[MAXNAMELEN]; 1346 1347 retv = EACCES; /* assume the worst */ 1348 1349 /* 1350 * Start by getting the dataset label if it exists. 1351 */ 1352 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1353 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1354 if (error) 1355 return (SET_ERROR(EACCES)); 1356 1357 /* 1358 * If labeling is NOT enabled, then disallow the mount of datasets 1359 * which have a non-default label already. No other label checks 1360 * are needed. 1361 */ 1362 if (!is_system_labeled()) { 1363 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1364 return (0); 1365 return (SET_ERROR(EACCES)); 1366 } 1367 1368 /* 1369 * Get the label of the mountpoint. If mounting into the global 1370 * zone (i.e. mountpoint is not within an active zone and the 1371 * zoned property is off), the label must be default or 1372 * admin_low/admin_high only; no other checks are needed. 1373 */ 1374 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1375 if (mntzone->zone_id == GLOBAL_ZONEID) { 1376 uint64_t zoned; 1377 1378 zone_rele(mntzone); 1379 1380 if (dsl_prop_get_integer(osname, 1381 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1382 return (SET_ERROR(EACCES)); 1383 if (!zoned) 1384 return (zfs_check_global_label(osname, ds_hexsl)); 1385 else 1386 /* 1387 * This is the case of a zone dataset being mounted 1388 * initially, before the zone has been fully created; 1389 * allow this mount into global zone. 1390 */ 1391 return (0); 1392 } 1393 1394 mnt_tsl = mntzone->zone_slabel; 1395 ASSERT(mnt_tsl != NULL); 1396 label_hold(mnt_tsl); 1397 mnt_sl = label2bslabel(mnt_tsl); 1398 1399 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1400 /* 1401 * The dataset doesn't have a real label, so fabricate one. 1402 */ 1403 char *str = NULL; 1404 1405 if (l_to_str_internal(mnt_sl, &str) == 0 && 1406 dsl_prop_set_string(osname, 1407 zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1408 ZPROP_SRC_LOCAL, str) == 0) 1409 retv = 0; 1410 if (str != NULL) 1411 kmem_free(str, strlen(str) + 1); 1412 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1413 /* 1414 * Now compare labels to complete the MAC check. If the 1415 * labels are equal then allow access. If the mountpoint 1416 * label dominates the dataset label, allow readonly access. 1417 * Otherwise, access is denied. 1418 */ 1419 if (blequal(mnt_sl, &ds_sl)) 1420 retv = 0; 1421 else if (bldominates(mnt_sl, &ds_sl)) { 1422 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1423 retv = 0; 1424 } 1425 } 1426 1427 label_rele(mnt_tsl); 1428 zone_rele(mntzone); 1429 return (retv); 1430} 1431#endif /* SECLABEL */ 1432 1433#ifdef OPENSOLARIS_MOUNTROOT 1434static int 1435zfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1436{ 1437 int error = 0; 1438 static int zfsrootdone = 0; 1439 zfsvfs_t *zfsvfs = NULL; 1440 znode_t *zp = NULL; 1441 vnode_t *vp = NULL; 1442 char *zfs_bootfs; 1443 char *zfs_devid; 1444 1445 ASSERT(vfsp); 1446 1447 /* 1448 * The filesystem that we mount as root is defined in the 1449 * boot property "zfs-bootfs" with a format of 1450 * "poolname/root-dataset-objnum". 1451 */ 1452 if (why == ROOT_INIT) { 1453 if (zfsrootdone++) 1454 return (SET_ERROR(EBUSY)); 1455 /* 1456 * the process of doing a spa_load will require the 1457 * clock to be set before we could (for example) do 1458 * something better by looking at the timestamp on 1459 * an uberblock, so just set it to -1. 1460 */ 1461 clkset(-1); 1462 1463 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1464 cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1465 "bootfs name"); 1466 return (SET_ERROR(EINVAL)); 1467 } 1468 zfs_devid = spa_get_bootprop("diskdevid"); 1469 error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1470 if (zfs_devid) 1471 spa_free_bootprop(zfs_devid); 1472 if (error) { 1473 spa_free_bootprop(zfs_bootfs); 1474 cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1475 error); 1476 return (error); 1477 } 1478 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1479 spa_free_bootprop(zfs_bootfs); 1480 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1481 error); 1482 return (error); 1483 } 1484 1485 spa_free_bootprop(zfs_bootfs); 1486 1487 if (error = vfs_lock(vfsp)) 1488 return (error); 1489 1490 if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1491 cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1492 goto out; 1493 } 1494 1495 zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1496 ASSERT(zfsvfs); 1497 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1498 cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1499 goto out; 1500 } 1501 1502 vp = ZTOV(zp); 1503 mutex_enter(&vp->v_lock); 1504 vp->v_flag |= VROOT; 1505 mutex_exit(&vp->v_lock); 1506 rootvp = vp; 1507 1508 /* 1509 * Leave rootvp held. The root file system is never unmounted. 1510 */ 1511 1512 vfs_add((struct vnode *)0, vfsp, 1513 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1514out: 1515 vfs_unlock(vfsp); 1516 return (error); 1517 } else if (why == ROOT_REMOUNT) { 1518 readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1519 vfsp->vfs_flag |= VFS_REMOUNT; 1520 1521 /* refresh mount options */ 1522 zfs_unregister_callbacks(vfsp->vfs_data); 1523 return (zfs_register_callbacks(vfsp)); 1524 1525 } else if (why == ROOT_UNMOUNT) { 1526 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1527 (void) zfs_sync(vfsp, 0, 0); 1528 return (0); 1529 } 1530 1531 /* 1532 * if "why" is equal to anything else other than ROOT_INIT, 1533 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1534 */ 1535 return (SET_ERROR(ENOTSUP)); 1536} 1537#endif /* OPENSOLARIS_MOUNTROOT */ 1538 1539static int 1540getpoolname(const char *osname, char *poolname) 1541{ 1542 char *p; 1543 1544 p = strchr(osname, '/'); 1545 if (p == NULL) { 1546 if (strlen(osname) >= MAXNAMELEN) 1547 return (ENAMETOOLONG); 1548 (void) strcpy(poolname, osname); 1549 } else { 1550 if (p - osname >= MAXNAMELEN) 1551 return (ENAMETOOLONG); 1552 (void) strncpy(poolname, osname, p - osname); 1553 poolname[p - osname] = '\0'; 1554 } 1555 return (0); 1556} 1557 1558/*ARGSUSED*/ 1559static int 1560zfs_mount(vfs_t *vfsp) 1561{ 1562 kthread_t *td = curthread; 1563 vnode_t *mvp = vfsp->mnt_vnodecovered; 1564 cred_t *cr = td->td_ucred; 1565 char *osname; 1566 int error = 0; 1567 int canwrite; 1568 1569#ifdef illumos 1570 if (mvp->v_type != VDIR) 1571 return (SET_ERROR(ENOTDIR)); 1572 1573 mutex_enter(&mvp->v_lock); 1574 if ((uap->flags & MS_REMOUNT) == 0 && 1575 (uap->flags & MS_OVERLAY) == 0 && 1576 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 1577 mutex_exit(&mvp->v_lock); 1578 return (SET_ERROR(EBUSY)); 1579 } 1580 mutex_exit(&mvp->v_lock); 1581 1582 /* 1583 * ZFS does not support passing unparsed data in via MS_DATA. 1584 * Users should use the MS_OPTIONSTR interface; this means 1585 * that all option parsing is already done and the options struct 1586 * can be interrogated. 1587 */ 1588 if ((uap->flags & MS_DATA) && uap->datalen > 0) 1589#else /* !illumos */ 1590 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1591 return (SET_ERROR(EPERM)); 1592 1593 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1594 return (SET_ERROR(EINVAL)); 1595#endif /* illumos */ 1596 1597 /* 1598 * If full-owner-access is enabled and delegated administration is 1599 * turned on, we must set nosuid. 1600 */ 1601 if (zfs_super_owner && 1602 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1603 secpolicy_fs_mount_clearopts(cr, vfsp); 1604 } 1605 1606 /* 1607 * Check for mount privilege? 1608 * 1609 * If we don't have privilege then see if 1610 * we have local permission to allow it 1611 */ 1612 error = secpolicy_fs_mount(cr, mvp, vfsp); 1613 if (error) { 1614 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1615 goto out; 1616 1617 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1618 vattr_t vattr; 1619 1620 /* 1621 * Make sure user is the owner of the mount point 1622 * or has sufficient privileges. 1623 */ 1624 1625 vattr.va_mask = AT_UID; 1626 1627 vn_lock(mvp, LK_SHARED | LK_RETRY); 1628 if (VOP_GETATTR(mvp, &vattr, cr)) { 1629 VOP_UNLOCK(mvp, 0); 1630 goto out; 1631 } 1632 1633 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1634 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1635 VOP_UNLOCK(mvp, 0); 1636 goto out; 1637 } 1638 VOP_UNLOCK(mvp, 0); 1639 } 1640 1641 secpolicy_fs_mount_clearopts(cr, vfsp); 1642 } 1643 1644 /* 1645 * Refuse to mount a filesystem if we are in a local zone and the 1646 * dataset is not visible. 1647 */ 1648 if (!INGLOBALZONE(curthread) && 1649 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1650 error = SET_ERROR(EPERM); 1651 goto out; 1652 } 1653 1654#ifdef SECLABEL 1655 error = zfs_mount_label_policy(vfsp, osname); 1656 if (error) 1657 goto out; 1658#endif 1659 1660 vfsp->vfs_flag |= MNT_NFS4ACLS; 1661 1662 /* 1663 * When doing a remount, we simply refresh our temporary properties 1664 * according to those options set in the current VFS options. 1665 */ 1666 if (vfsp->vfs_flag & MS_REMOUNT) { 1667 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1668 1669 /* 1670 * Refresh mount options with z_teardown_lock blocking I/O while 1671 * the filesystem is in an inconsistent state. 1672 * The lock also serializes this code with filesystem 1673 * manipulations between entry to zfs_suspend_fs() and return 1674 * from zfs_resume_fs(). 1675 */ 1676 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1677 zfs_unregister_callbacks(zfsvfs); 1678 error = zfs_register_callbacks(vfsp); 1679 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1680 goto out; 1681 } 1682 1683 /* Initial root mount: try hard to import the requested root pool. */ 1684 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1685 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1686 char pname[MAXNAMELEN]; 1687 1688 error = getpoolname(osname, pname); 1689 if (error == 0) 1690 error = spa_import_rootpool(pname); 1691 if (error) 1692 goto out; 1693 } 1694 DROP_GIANT(); 1695 error = zfs_domount(vfsp, osname); 1696 PICKUP_GIANT(); 1697 1698#ifdef illumos 1699 /* 1700 * Add an extra VFS_HOLD on our parent vfs so that it can't 1701 * disappear due to a forced unmount. 1702 */ 1703 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1704 VFS_HOLD(mvp->v_vfsp); 1705#endif 1706 1707out: 1708 return (error); 1709} 1710 1711static int 1712zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1713{ 1714 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1715 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1716 1717 statp->f_version = STATFS_VERSION; 1718 1719 ZFS_ENTER(zfsvfs); 1720 1721 dmu_objset_space(zfsvfs->z_os, 1722 &refdbytes, &availbytes, &usedobjs, &availobjs); 1723 1724 /* 1725 * The underlying storage pool actually uses multiple block sizes. 1726 * We report the fragsize as the smallest block size we support, 1727 * and we report our blocksize as the filesystem's maximum blocksize. 1728 */ 1729 statp->f_bsize = SPA_MINBLOCKSIZE; 1730 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1731 1732 /* 1733 * The following report "total" blocks of various kinds in the 1734 * file system, but reported in terms of f_frsize - the 1735 * "fragment" size. 1736 */ 1737 1738 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1739 statp->f_bfree = availbytes / statp->f_bsize; 1740 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1741 1742 /* 1743 * statvfs() should really be called statufs(), because it assumes 1744 * static metadata. ZFS doesn't preallocate files, so the best 1745 * we can do is report the max that could possibly fit in f_files, 1746 * and that minus the number actually used in f_ffree. 1747 * For f_ffree, report the smaller of the number of object available 1748 * and the number of blocks (each object will take at least a block). 1749 */ 1750 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1751 statp->f_files = statp->f_ffree + usedobjs; 1752 1753 /* 1754 * We're a zfs filesystem. 1755 */ 1756 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1757 1758 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1759 sizeof(statp->f_mntfromname)); 1760 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1761 sizeof(statp->f_mntonname)); 1762 1763 statp->f_namemax = ZFS_MAXNAMELEN; 1764 1765 ZFS_EXIT(zfsvfs); 1766 return (0); 1767} 1768 1769static int 1770zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1771{ 1772 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1773 znode_t *rootzp; 1774 int error; 1775 1776 ZFS_ENTER(zfsvfs); 1777 1778 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1779 if (error == 0) 1780 *vpp = ZTOV(rootzp); 1781 1782 ZFS_EXIT(zfsvfs); 1783 1784 if (error == 0) 1785 error = vn_lock(*vpp, flags); 1786 if (error != 0) 1787 *vpp = NULL; 1788 1789 return (error); 1790} 1791 1792/* 1793 * Teardown the zfsvfs::z_os. 1794 * 1795 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1796 * and 'z_teardown_inactive_lock' held. 1797 */ 1798static int 1799zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1800{ 1801 znode_t *zp; 1802 1803 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1804 1805 if (!unmounting) { 1806 /* 1807 * We purge the parent filesystem's vfsp as the parent 1808 * filesystem and all of its snapshots have their vnode's 1809 * v_vfsp set to the parent's filesystem's vfsp. Note, 1810 * 'z_parent' is self referential for non-snapshots. 1811 */ 1812 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1813#ifdef FREEBSD_NAMECACHE 1814 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1815#endif 1816 } 1817 1818 /* 1819 * Close the zil. NB: Can't close the zil while zfs_inactive 1820 * threads are blocked as zil_close can call zfs_inactive. 1821 */ 1822 if (zfsvfs->z_log) { 1823 zil_close(zfsvfs->z_log); 1824 zfsvfs->z_log = NULL; 1825 } 1826 1827 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1828 1829 /* 1830 * If we are not unmounting (ie: online recv) and someone already 1831 * unmounted this file system while we were doing the switcheroo, 1832 * or a reopen of z_os failed then just bail out now. 1833 */ 1834 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1835 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1836 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1837 return (SET_ERROR(EIO)); 1838 } 1839 1840 /* 1841 * At this point there are no vops active, and any new vops will 1842 * fail with EIO since we have z_teardown_lock for writer (only 1843 * relavent for forced unmount). 1844 * 1845 * Release all holds on dbufs. 1846 */ 1847 mutex_enter(&zfsvfs->z_znodes_lock); 1848 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1849 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1850 if (zp->z_sa_hdl) { 1851 ASSERT(ZTOV(zp)->v_count >= 0); 1852 zfs_znode_dmu_fini(zp); 1853 } 1854 mutex_exit(&zfsvfs->z_znodes_lock); 1855 1856 /* 1857 * If we are unmounting, set the unmounted flag and let new vops 1858 * unblock. zfs_inactive will have the unmounted behavior, and all 1859 * other vops will fail with EIO. 1860 */ 1861 if (unmounting) { 1862 zfsvfs->z_unmounted = B_TRUE; 1863 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1864 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1865 } 1866 1867 /* 1868 * z_os will be NULL if there was an error in attempting to reopen 1869 * zfsvfs, so just return as the properties had already been 1870 * unregistered and cached data had been evicted before. 1871 */ 1872 if (zfsvfs->z_os == NULL) 1873 return (0); 1874 1875 /* 1876 * Unregister properties. 1877 */ 1878 zfs_unregister_callbacks(zfsvfs); 1879 1880 /* 1881 * Evict cached data 1882 */ 1883 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1884 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1885 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1886 dmu_objset_evict_dbufs(zfsvfs->z_os); 1887 1888 return (0); 1889} 1890 1891/*ARGSUSED*/ 1892static int 1893zfs_umount(vfs_t *vfsp, int fflag) 1894{ 1895 kthread_t *td = curthread; 1896 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1897 objset_t *os; 1898 cred_t *cr = td->td_ucred; 1899 int ret; 1900 1901 ret = secpolicy_fs_unmount(cr, vfsp); 1902 if (ret) { 1903 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1904 ZFS_DELEG_PERM_MOUNT, cr)) 1905 return (ret); 1906 } 1907 1908 /* 1909 * We purge the parent filesystem's vfsp as the parent filesystem 1910 * and all of its snapshots have their vnode's v_vfsp set to the 1911 * parent's filesystem's vfsp. Note, 'z_parent' is self 1912 * referential for non-snapshots. 1913 */ 1914 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1915 1916 /* 1917 * Unmount any snapshots mounted under .zfs before unmounting the 1918 * dataset itself. 1919 */ 1920 if (zfsvfs->z_ctldir != NULL) { 1921 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1922 return (ret); 1923 ret = vflush(vfsp, 0, 0, td); 1924 ASSERT(ret == EBUSY); 1925 if (!(fflag & MS_FORCE)) { 1926 if (zfsvfs->z_ctldir->v_count > 1) 1927 return (EBUSY); 1928 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1929 } 1930 zfsctl_destroy(zfsvfs); 1931 ASSERT(zfsvfs->z_ctldir == NULL); 1932 } 1933 1934 if (fflag & MS_FORCE) { 1935 /* 1936 * Mark file system as unmounted before calling 1937 * vflush(FORCECLOSE). This way we ensure no future vnops 1938 * will be called and risk operating on DOOMED vnodes. 1939 */ 1940 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1941 zfsvfs->z_unmounted = B_TRUE; 1942 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1943 } 1944 1945 /* 1946 * Flush all the files. 1947 */ 1948 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1949 if (ret != 0) { 1950 if (!zfsvfs->z_issnap) { 1951 zfsctl_create(zfsvfs); 1952 ASSERT(zfsvfs->z_ctldir != NULL); 1953 } 1954 return (ret); 1955 } 1956 1957#ifdef illumos 1958 if (!(fflag & MS_FORCE)) { 1959 /* 1960 * Check the number of active vnodes in the file system. 1961 * Our count is maintained in the vfs structure, but the 1962 * number is off by 1 to indicate a hold on the vfs 1963 * structure itself. 1964 * 1965 * The '.zfs' directory maintains a reference of its 1966 * own, and any active references underneath are 1967 * reflected in the vnode count. 1968 */ 1969 if (zfsvfs->z_ctldir == NULL) { 1970 if (vfsp->vfs_count > 1) 1971 return (SET_ERROR(EBUSY)); 1972 } else { 1973 if (vfsp->vfs_count > 2 || 1974 zfsvfs->z_ctldir->v_count > 1) 1975 return (SET_ERROR(EBUSY)); 1976 } 1977 } 1978#endif 1979 1980 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1981 os = zfsvfs->z_os; 1982 1983 /* 1984 * z_os will be NULL if there was an error in 1985 * attempting to reopen zfsvfs. 1986 */ 1987 if (os != NULL) { 1988 /* 1989 * Unset the objset user_ptr. 1990 */ 1991 mutex_enter(&os->os_user_ptr_lock); 1992 dmu_objset_set_user(os, NULL); 1993 mutex_exit(&os->os_user_ptr_lock); 1994 1995 /* 1996 * Finally release the objset 1997 */ 1998 dmu_objset_disown(os, zfsvfs); 1999 } 2000 2001 /* 2002 * We can now safely destroy the '.zfs' directory node. 2003 */ 2004 if (zfsvfs->z_ctldir != NULL) 2005 zfsctl_destroy(zfsvfs); 2006 zfs_freevfs(vfsp); 2007 2008 return (0); 2009} 2010 2011static int 2012zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2013{ 2014 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2015 znode_t *zp; 2016 int err; 2017 2018 /* 2019 * zfs_zget() can't operate on virtual entries like .zfs/ or 2020 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2021 * This will make NFS to switch to LOOKUP instead of using VGET. 2022 */ 2023 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || 2024 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) 2025 return (EOPNOTSUPP); 2026 2027 ZFS_ENTER(zfsvfs); 2028 err = zfs_zget(zfsvfs, ino, &zp); 2029 if (err == 0 && zp->z_unlinked) { 2030 VN_RELE(ZTOV(zp)); 2031 err = EINVAL; 2032 } 2033 if (err == 0) 2034 *vpp = ZTOV(zp); 2035 ZFS_EXIT(zfsvfs); 2036 if (err == 0) 2037 err = vn_lock(*vpp, flags); 2038 if (err != 0) 2039 *vpp = NULL; 2040 return (err); 2041} 2042 2043static int 2044zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2045 struct ucred **credanonp, int *numsecflavors, int **secflavors) 2046{ 2047 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2048 2049 /* 2050 * If this is regular file system vfsp is the same as 2051 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2052 * zfsvfs->z_parent->z_vfs represents parent file system 2053 * which we have to use here, because only this file system 2054 * has mnt_export configured. 2055 */ 2056 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2057 credanonp, numsecflavors, secflavors)); 2058} 2059 2060CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2061CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2062 2063static int 2064zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2065{ 2066 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2067 znode_t *zp; 2068 uint64_t object = 0; 2069 uint64_t fid_gen = 0; 2070 uint64_t gen_mask; 2071 uint64_t zp_gen; 2072 int i, err; 2073 2074 *vpp = NULL; 2075 2076 ZFS_ENTER(zfsvfs); 2077 2078 /* 2079 * On FreeBSD we can get snapshot's mount point or its parent file 2080 * system mount point depending if snapshot is already mounted or not. 2081 */ 2082 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2083 zfid_long_t *zlfid = (zfid_long_t *)fidp; 2084 uint64_t objsetid = 0; 2085 uint64_t setgen = 0; 2086 2087 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2088 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2089 2090 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2091 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2092 2093 ZFS_EXIT(zfsvfs); 2094 2095 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2096 if (err) 2097 return (SET_ERROR(EINVAL)); 2098 ZFS_ENTER(zfsvfs); 2099 } 2100 2101 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2102 zfid_short_t *zfid = (zfid_short_t *)fidp; 2103 2104 for (i = 0; i < sizeof (zfid->zf_object); i++) 2105 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2106 2107 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2108 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2109 } else { 2110 ZFS_EXIT(zfsvfs); 2111 return (SET_ERROR(EINVAL)); 2112 } 2113 2114 /* 2115 * A zero fid_gen means we are in .zfs or the .zfs/snapshot 2116 * directory tree. If the object == zfsvfs->z_shares_dir, then 2117 * we are in the .zfs/shares directory tree. 2118 */ 2119 if ((fid_gen == 0 && 2120 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || 2121 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { 2122 *vpp = zfsvfs->z_ctldir; 2123 ASSERT(*vpp != NULL); 2124 if (object == ZFSCTL_INO_SNAPDIR) { 2125 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2126 0, NULL, NULL, NULL, NULL, NULL) == 0); 2127 } else if (object == zfsvfs->z_shares_dir) { 2128 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL, 2129 0, NULL, NULL, NULL, NULL, NULL) == 0); 2130 } else { 2131 VN_HOLD(*vpp); 2132 } 2133 ZFS_EXIT(zfsvfs); 2134 err = vn_lock(*vpp, flags); 2135 if (err != 0) 2136 *vpp = NULL; 2137 return (err); 2138 } 2139 2140 gen_mask = -1ULL >> (64 - 8 * i); 2141 2142 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2143 if (err = zfs_zget(zfsvfs, object, &zp)) { 2144 ZFS_EXIT(zfsvfs); 2145 return (err); 2146 } 2147 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2148 sizeof (uint64_t)); 2149 zp_gen = zp_gen & gen_mask; 2150 if (zp_gen == 0) 2151 zp_gen = 1; 2152 if (zp->z_unlinked || zp_gen != fid_gen) { 2153 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2154 VN_RELE(ZTOV(zp)); 2155 ZFS_EXIT(zfsvfs); 2156 return (SET_ERROR(EINVAL)); 2157 } 2158 2159 *vpp = ZTOV(zp); 2160 ZFS_EXIT(zfsvfs); 2161 err = vn_lock(*vpp, flags | LK_RETRY); 2162 if (err == 0) 2163 vnode_create_vobject(*vpp, zp->z_size, curthread); 2164 else 2165 *vpp = NULL; 2166 return (err); 2167} 2168 2169/* 2170 * Block out VOPs and close zfsvfs_t::z_os 2171 * 2172 * Note, if successful, then we return with the 'z_teardown_lock' and 2173 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 2174 * dataset and objset intact so that they can be atomically handed off during 2175 * a subsequent rollback or recv operation and the resume thereafter. 2176 */ 2177int 2178zfs_suspend_fs(zfsvfs_t *zfsvfs) 2179{ 2180 int error; 2181 2182 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2183 return (error); 2184 2185 return (0); 2186} 2187 2188/* 2189 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 2190 * is an invariant across any of the operations that can be performed while the 2191 * filesystem was suspended. Whether it succeeded or failed, the preconditions 2192 * are the same: the relevant objset and associated dataset are owned by 2193 * zfsvfs, held, and long held on entry. 2194 */ 2195int 2196zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2197{ 2198 int err; 2199 znode_t *zp; 2200 uint64_t sa_obj = 0; 2201 2202 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2203 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2204 2205 /* 2206 * We already own this, so just hold and rele it to update the 2207 * objset_t, as the one we had before may have been evicted. 2208 */ 2209 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os)); 2210 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs); 2211 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset)); 2212 dmu_objset_rele(zfsvfs->z_os, zfsvfs); 2213 2214 /* 2215 * Make sure version hasn't changed 2216 */ 2217 2218 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, 2219 &zfsvfs->z_version); 2220 2221 if (err) 2222 goto bail; 2223 2224 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, 2225 ZFS_SA_ATTRS, 8, 1, &sa_obj); 2226 2227 if (err && zfsvfs->z_version >= ZPL_VERSION_SA) 2228 goto bail; 2229 2230 if ((err = sa_setup(zfsvfs->z_os, sa_obj, 2231 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) 2232 goto bail; 2233 2234 if (zfsvfs->z_version >= ZPL_VERSION_SA) 2235 sa_register_update_callback(zfsvfs->z_os, 2236 zfs_sa_upgrade); 2237 2238 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2239 2240 zfs_set_fuid_feature(zfsvfs); 2241 2242 /* 2243 * Attempt to re-establish all the active znodes with 2244 * their dbufs. If a zfs_rezget() fails, then we'll let 2245 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2246 * when they try to use their znode. 2247 */ 2248 mutex_enter(&zfsvfs->z_znodes_lock); 2249 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2250 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2251 (void) zfs_rezget(zp); 2252 } 2253 mutex_exit(&zfsvfs->z_znodes_lock); 2254 2255bail: 2256 /* release the VOPs */ 2257 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2258 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 2259 2260 if (err) { 2261 /* 2262 * Since we couldn't setup the sa framework, try to force 2263 * unmount this file system. 2264 */ 2265 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) { 2266 vfs_ref(zfsvfs->z_vfs); 2267 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2268 } 2269 } 2270 return (err); 2271} 2272 2273static void 2274zfs_freevfs(vfs_t *vfsp) 2275{ 2276 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2277 2278#ifdef illumos 2279 /* 2280 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2281 * from zfs_mount(). Release it here. If we came through 2282 * zfs_mountroot() instead, we didn't grab an extra hold, so 2283 * skip the VFS_RELE for rootvfs. 2284 */ 2285 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2286 VFS_RELE(zfsvfs->z_parent->z_vfs); 2287#endif 2288 2289 zfsvfs_free(zfsvfs); 2290 2291 atomic_dec_32(&zfs_active_fs_count); 2292} 2293 2294#ifdef __i386__ 2295static int desiredvnodes_backup; 2296#endif 2297 2298static void 2299zfs_vnodes_adjust(void) 2300{ 2301#ifdef __i386__ 2302 int newdesiredvnodes; 2303 2304 desiredvnodes_backup = desiredvnodes; 2305 2306 /* 2307 * We calculate newdesiredvnodes the same way it is done in 2308 * vntblinit(). If it is equal to desiredvnodes, it means that 2309 * it wasn't tuned by the administrator and we can tune it down. 2310 */ 2311 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2312 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2313 sizeof(struct vnode)))); 2314 if (newdesiredvnodes == desiredvnodes) 2315 desiredvnodes = (3 * newdesiredvnodes) / 4; 2316#endif 2317} 2318 2319static void 2320zfs_vnodes_adjust_back(void) 2321{ 2322 2323#ifdef __i386__ 2324 desiredvnodes = desiredvnodes_backup; 2325#endif 2326} 2327 2328void 2329zfs_init(void) 2330{ 2331 2332 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2333 2334 /* 2335 * Initialize .zfs directory structures 2336 */ 2337 zfsctl_init(); 2338 2339 /* 2340 * Initialize znode cache, vnode ops, etc... 2341 */ 2342 zfs_znode_init(); 2343 2344 /* 2345 * Reduce number of vnodes. Originally number of vnodes is calculated 2346 * with UFS inode in mind. We reduce it here, because it's too big for 2347 * ZFS/i386. 2348 */ 2349 zfs_vnodes_adjust(); 2350 2351 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2352} 2353 2354void 2355zfs_fini(void) 2356{ 2357 zfsctl_fini(); 2358 zfs_znode_fini(); 2359 zfs_vnodes_adjust_back(); 2360} 2361 2362int 2363zfs_busy(void) 2364{ 2365 return (zfs_active_fs_count != 0); 2366} 2367 2368int 2369zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2370{ 2371 int error; 2372 objset_t *os = zfsvfs->z_os; 2373 dmu_tx_t *tx; 2374 2375 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2376 return (SET_ERROR(EINVAL)); 2377 2378 if (newvers < zfsvfs->z_version) 2379 return (SET_ERROR(EINVAL)); 2380 2381 if (zfs_spa_version_map(newvers) > 2382 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2383 return (SET_ERROR(ENOTSUP)); 2384 2385 tx = dmu_tx_create(os); 2386 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2387 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2388 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2389 ZFS_SA_ATTRS); 2390 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2391 } 2392 error = dmu_tx_assign(tx, TXG_WAIT); 2393 if (error) { 2394 dmu_tx_abort(tx); 2395 return (error); 2396 } 2397 2398 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2399 8, 1, &newvers, tx); 2400 2401 if (error) { 2402 dmu_tx_commit(tx); 2403 return (error); 2404 } 2405 2406 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2407 uint64_t sa_obj; 2408 2409 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2410 SPA_VERSION_SA); 2411 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2412 DMU_OT_NONE, 0, tx); 2413 2414 error = zap_add(os, MASTER_NODE_OBJ, 2415 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2416 ASSERT0(error); 2417 2418 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2419 sa_register_update_callback(os, zfs_sa_upgrade); 2420 } 2421 2422 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2423 "from %llu to %llu", zfsvfs->z_version, newvers); 2424 2425 dmu_tx_commit(tx); 2426 2427 zfsvfs->z_version = newvers; 2428 2429 zfs_set_fuid_feature(zfsvfs); 2430 2431 return (0); 2432} 2433 2434/* 2435 * Read a property stored within the master node. 2436 */ 2437int 2438zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2439{ 2440 const char *pname; 2441 int error = ENOENT; 2442 2443 /* 2444 * Look up the file system's value for the property. For the 2445 * version property, we look up a slightly different string. 2446 */ 2447 if (prop == ZFS_PROP_VERSION) 2448 pname = ZPL_VERSION_STR; 2449 else 2450 pname = zfs_prop_to_name(prop); 2451 2452 if (os != NULL) 2453 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2454 2455 if (error == ENOENT) { 2456 /* No value set, use the default value */ 2457 switch (prop) { 2458 case ZFS_PROP_VERSION: 2459 *value = ZPL_VERSION; 2460 break; 2461 case ZFS_PROP_NORMALIZE: 2462 case ZFS_PROP_UTF8ONLY: 2463 *value = 0; 2464 break; 2465 case ZFS_PROP_CASE: 2466 *value = ZFS_CASE_SENSITIVE; 2467 break; 2468 default: 2469 return (error); 2470 } 2471 error = 0; 2472 } 2473 return (error); 2474} 2475 2476#ifdef _KERNEL 2477void 2478zfsvfs_update_fromname(const char *oldname, const char *newname) 2479{ 2480 char tmpbuf[MAXPATHLEN]; 2481 struct mount *mp; 2482 char *fromname; 2483 size_t oldlen; 2484 2485 oldlen = strlen(oldname); 2486 2487 mtx_lock(&mountlist_mtx); 2488 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2489 fromname = mp->mnt_stat.f_mntfromname; 2490 if (strcmp(fromname, oldname) == 0) { 2491 (void)strlcpy(fromname, newname, 2492 sizeof(mp->mnt_stat.f_mntfromname)); 2493 continue; 2494 } 2495 if (strncmp(fromname, oldname, oldlen) == 0 && 2496 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2497 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2498 newname, fromname + oldlen); 2499 (void)strlcpy(fromname, tmpbuf, 2500 sizeof(mp->mnt_stat.f_mntfromname)); 2501 continue; 2502 } 2503 } 2504 mtx_unlock(&mountlist_mtx); 2505} 2506#endif 2507