zfs_vfsops.c revision 328048
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, 2015 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} 849 850/* 851 * Associate this zfsvfs with the given objset, which must be owned. 852 * This will cache a bunch of on-disk state from the objset in the 853 * zfsvfs. 854 */ 855static int 856zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os) 857{ 858 int error; 859 uint64_t val; 860 861 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 862 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 863 zfsvfs->z_os = os; 864 865 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 866 if (error != 0) 867 return (error); 868 if (zfsvfs->z_version > 869 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 870 (void) printf("Can't mount a version %lld file system " 871 "on a version %lld pool\n. Pool must be upgraded to mount " 872 "this file system.", (u_longlong_t)zfsvfs->z_version, 873 (u_longlong_t)spa_version(dmu_objset_spa(os))); 874 return (SET_ERROR(ENOTSUP)); 875 } 876 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val); 877 if (error != 0) 878 return (error); 879 zfsvfs->z_norm = (int)val; 880 881 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val); 882 if (error != 0) 883 return (error); 884 zfsvfs->z_utf8 = (val != 0); 885 886 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val); 887 if (error != 0) 888 return (error); 889 zfsvfs->z_case = (uint_t)val; 890 891 /* 892 * Fold case on file systems that are always or sometimes case 893 * insensitive. 894 */ 895 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 896 zfsvfs->z_case == ZFS_CASE_MIXED) 897 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 898 899 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 900 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 901 902 uint64_t sa_obj = 0; 903 if (zfsvfs->z_use_sa) { 904 /* should either have both of these objects or none */ 905 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 906 &sa_obj); 907 if (error != 0) 908 return (error); 909 } 910 911 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 912 &zfsvfs->z_attr_table); 913 if (error != 0) 914 return (error); 915 916 if (zfsvfs->z_version >= ZPL_VERSION_SA) 917 sa_register_update_callback(os, zfs_sa_upgrade); 918 919 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 920 &zfsvfs->z_root); 921 if (error != 0) 922 return (error); 923 ASSERT(zfsvfs->z_root != 0); 924 925 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 926 &zfsvfs->z_unlinkedobj); 927 if (error != 0) 928 return (error); 929 930 error = zap_lookup(os, MASTER_NODE_OBJ, 931 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 932 8, 1, &zfsvfs->z_userquota_obj); 933 if (error == ENOENT) 934 zfsvfs->z_userquota_obj = 0; 935 else if (error != 0) 936 return (error); 937 938 error = zap_lookup(os, MASTER_NODE_OBJ, 939 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 940 8, 1, &zfsvfs->z_groupquota_obj); 941 if (error == ENOENT) 942 zfsvfs->z_groupquota_obj = 0; 943 else if (error != 0) 944 return (error); 945 946 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 947 &zfsvfs->z_fuid_obj); 948 if (error == ENOENT) 949 zfsvfs->z_fuid_obj = 0; 950 else if (error != 0) 951 return (error); 952 953 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 954 &zfsvfs->z_shares_dir); 955 if (error == ENOENT) 956 zfsvfs->z_shares_dir = 0; 957 else if (error != 0) 958 return (error); 959 960 /* 961 * Only use the name cache if we are looking for a 962 * name on a file system that does not require normalization 963 * or case folding. We can also look there if we happen to be 964 * on a non-normalizing, mixed sensitivity file system IF we 965 * are looking for the exact name (which is always the case on 966 * FreeBSD). 967 */ 968 zfsvfs->z_use_namecache = !zfsvfs->z_norm || 969 ((zfsvfs->z_case == ZFS_CASE_MIXED) && 970 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER)); 971 972 return (0); 973} 974 975int 976zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 977{ 978 objset_t *os; 979 zfsvfs_t *zfsvfs; 980 int error; 981 982 /* 983 * XXX: Fix struct statfs so this isn't necessary! 984 * 985 * The 'osname' is used as the filesystem's special node, which means 986 * it must fit in statfs.f_mntfromname, or else it can't be 987 * enumerated, so libzfs_mnttab_find() returns NULL, which causes 988 * 'zfs unmount' to think it's not mounted when it is. 989 */ 990 if (strlen(osname) >= MNAMELEN) 991 return (SET_ERROR(ENAMETOOLONG)); 992 993 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 994 995 /* 996 * We claim to always be readonly so we can open snapshots; 997 * other ZPL code will prevent us from writing to snapshots. 998 */ 999 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 1000 if (error) { 1001 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1002 return (error); 1003 } 1004 1005 zfsvfs->z_vfs = NULL; 1006 zfsvfs->z_parent = zfsvfs; 1007 1008 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1009 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 1010 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 1011 offsetof(znode_t, z_link_node)); 1012#ifdef DIAGNOSTIC 1013 rrm_init(&zfsvfs->z_teardown_lock, B_TRUE); 1014#else 1015 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE); 1016#endif 1017 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 1018 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 1019 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1020 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1021 1022 error = zfsvfs_init(zfsvfs, os); 1023 if (error != 0) { 1024 dmu_objset_disown(os, zfsvfs); 1025 *zfvp = NULL; 1026 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1027 return (error); 1028 } 1029 1030 *zfvp = zfsvfs; 1031 return (0); 1032} 1033 1034static int 1035zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 1036{ 1037 int error; 1038 1039 error = zfs_register_callbacks(zfsvfs->z_vfs); 1040 if (error) 1041 return (error); 1042 1043 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 1044 1045 /* 1046 * If we are not mounting (ie: online recv), then we don't 1047 * have to worry about replaying the log as we blocked all 1048 * operations out since we closed the ZIL. 1049 */ 1050 if (mounting) { 1051 boolean_t readonly; 1052 1053 /* 1054 * During replay we remove the read only flag to 1055 * allow replays to succeed. 1056 */ 1057 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1058 if (readonly != 0) 1059 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1060 else 1061 zfs_unlinked_drain(zfsvfs); 1062 1063 /* 1064 * Parse and replay the intent log. 1065 * 1066 * Because of ziltest, this must be done after 1067 * zfs_unlinked_drain(). (Further note: ziltest 1068 * doesn't use readonly mounts, where 1069 * zfs_unlinked_drain() isn't called.) This is because 1070 * ziltest causes spa_sync() to think it's committed, 1071 * but actually it is not, so the intent log contains 1072 * many txg's worth of changes. 1073 * 1074 * In particular, if object N is in the unlinked set in 1075 * the last txg to actually sync, then it could be 1076 * actually freed in a later txg and then reallocated 1077 * in a yet later txg. This would write a "create 1078 * object N" record to the intent log. Normally, this 1079 * would be fine because the spa_sync() would have 1080 * written out the fact that object N is free, before 1081 * we could write the "create object N" intent log 1082 * record. 1083 * 1084 * But when we are in ziltest mode, we advance the "open 1085 * txg" without actually spa_sync()-ing the changes to 1086 * disk. So we would see that object N is still 1087 * allocated and in the unlinked set, and there is an 1088 * intent log record saying to allocate it. 1089 */ 1090 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1091 if (zil_replay_disable) { 1092 zil_destroy(zfsvfs->z_log, B_FALSE); 1093 } else { 1094 zfsvfs->z_replay = B_TRUE; 1095 zil_replay(zfsvfs->z_os, zfsvfs, 1096 zfs_replay_vector); 1097 zfsvfs->z_replay = B_FALSE; 1098 } 1099 } 1100 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1101 } 1102 1103 /* 1104 * Set the objset user_ptr to track its zfsvfs. 1105 */ 1106 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1107 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1108 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1109 1110 return (0); 1111} 1112 1113extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1114 1115void 1116zfsvfs_free(zfsvfs_t *zfsvfs) 1117{ 1118 int i; 1119 1120 /* 1121 * This is a barrier to prevent the filesystem from going away in 1122 * zfs_znode_move() until we can safely ensure that the filesystem is 1123 * not unmounted. We consider the filesystem valid before the barrier 1124 * and invalid after the barrier. 1125 */ 1126 rw_enter(&zfsvfs_lock, RW_READER); 1127 rw_exit(&zfsvfs_lock); 1128 1129 zfs_fuid_destroy(zfsvfs); 1130 1131 mutex_destroy(&zfsvfs->z_znodes_lock); 1132 mutex_destroy(&zfsvfs->z_lock); 1133 list_destroy(&zfsvfs->z_all_znodes); 1134 rrm_destroy(&zfsvfs->z_teardown_lock); 1135 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1136 rw_destroy(&zfsvfs->z_fuid_lock); 1137 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1138 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1139 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1140} 1141 1142static void 1143zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1144{ 1145 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1146 if (zfsvfs->z_vfs) { 1147 if (zfsvfs->z_use_fuids) { 1148 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1149 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1150 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1151 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1152 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1153 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1154 } else { 1155 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1156 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1157 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1158 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1159 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1160 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1161 } 1162 } 1163 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1164} 1165 1166static int 1167zfs_domount(vfs_t *vfsp, char *osname) 1168{ 1169 uint64_t recordsize, fsid_guid; 1170 int error = 0; 1171 zfsvfs_t *zfsvfs; 1172 vnode_t *vp; 1173 1174 ASSERT(vfsp); 1175 ASSERT(osname); 1176 1177 error = zfsvfs_create(osname, &zfsvfs); 1178 if (error) 1179 return (error); 1180 zfsvfs->z_vfs = vfsp; 1181 1182#ifdef illumos 1183 /* Initialize the generic filesystem structure. */ 1184 vfsp->vfs_bcount = 0; 1185 vfsp->vfs_data = NULL; 1186 1187 if (zfs_create_unique_device(&mount_dev) == -1) { 1188 error = SET_ERROR(ENODEV); 1189 goto out; 1190 } 1191 ASSERT(vfs_devismounted(mount_dev) == 0); 1192#endif 1193 1194 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1195 NULL)) 1196 goto out; 1197 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1198 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1199 1200 vfsp->vfs_data = zfsvfs; 1201 vfsp->mnt_flag |= MNT_LOCAL; 1202 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1203 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1204 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1205 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */ 1206 1207 /* 1208 * The fsid is 64 bits, composed of an 8-bit fs type, which 1209 * separates our fsid from any other filesystem types, and a 1210 * 56-bit objset unique ID. The objset unique ID is unique to 1211 * all objsets open on this system, provided by unique_create(). 1212 * The 8-bit fs type must be put in the low bits of fsid[1] 1213 * because that's where other Solaris filesystems put it. 1214 */ 1215 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1216 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1217 vfsp->vfs_fsid.val[0] = fsid_guid; 1218 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1219 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1220 1221 /* 1222 * Set features for file system. 1223 */ 1224 zfs_set_fuid_feature(zfsvfs); 1225 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1226 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1227 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1228 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1229 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1230 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1231 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1232 } 1233 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1234 1235 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1236 uint64_t pval; 1237 1238 atime_changed_cb(zfsvfs, B_FALSE); 1239 readonly_changed_cb(zfsvfs, B_TRUE); 1240 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1241 goto out; 1242 xattr_changed_cb(zfsvfs, pval); 1243 zfsvfs->z_issnap = B_TRUE; 1244 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1245 1246 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1247 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1248 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1249 } else { 1250 error = zfsvfs_setup(zfsvfs, B_TRUE); 1251 } 1252 1253 vfs_mountedfrom(vfsp, osname); 1254 1255 if (!zfsvfs->z_issnap) 1256 zfsctl_create(zfsvfs); 1257out: 1258 if (error) { 1259 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1260 zfsvfs_free(zfsvfs); 1261 } else { 1262 atomic_inc_32(&zfs_active_fs_count); 1263 } 1264 1265 return (error); 1266} 1267 1268void 1269zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1270{ 1271 objset_t *os = zfsvfs->z_os; 1272 1273 if (!dmu_objset_is_snapshot(os)) 1274 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 1275} 1276 1277#ifdef SECLABEL 1278/* 1279 * Convert a decimal digit string to a uint64_t integer. 1280 */ 1281static int 1282str_to_uint64(char *str, uint64_t *objnum) 1283{ 1284 uint64_t num = 0; 1285 1286 while (*str) { 1287 if (*str < '0' || *str > '9') 1288 return (SET_ERROR(EINVAL)); 1289 1290 num = num*10 + *str++ - '0'; 1291 } 1292 1293 *objnum = num; 1294 return (0); 1295} 1296 1297/* 1298 * The boot path passed from the boot loader is in the form of 1299 * "rootpool-name/root-filesystem-object-number'. Convert this 1300 * string to a dataset name: "rootpool-name/root-filesystem-name". 1301 */ 1302static int 1303zfs_parse_bootfs(char *bpath, char *outpath) 1304{ 1305 char *slashp; 1306 uint64_t objnum; 1307 int error; 1308 1309 if (*bpath == 0 || *bpath == '/') 1310 return (SET_ERROR(EINVAL)); 1311 1312 (void) strcpy(outpath, bpath); 1313 1314 slashp = strchr(bpath, '/'); 1315 1316 /* if no '/', just return the pool name */ 1317 if (slashp == NULL) { 1318 return (0); 1319 } 1320 1321 /* if not a number, just return the root dataset name */ 1322 if (str_to_uint64(slashp+1, &objnum)) { 1323 return (0); 1324 } 1325 1326 *slashp = '\0'; 1327 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1328 *slashp = '/'; 1329 1330 return (error); 1331} 1332 1333/* 1334 * Check that the hex label string is appropriate for the dataset being 1335 * mounted into the global_zone proper. 1336 * 1337 * Return an error if the hex label string is not default or 1338 * admin_low/admin_high. For admin_low labels, the corresponding 1339 * dataset must be readonly. 1340 */ 1341int 1342zfs_check_global_label(const char *dsname, const char *hexsl) 1343{ 1344 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1345 return (0); 1346 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1347 return (0); 1348 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1349 /* must be readonly */ 1350 uint64_t rdonly; 1351 1352 if (dsl_prop_get_integer(dsname, 1353 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1354 return (SET_ERROR(EACCES)); 1355 return (rdonly ? 0 : EACCES); 1356 } 1357 return (SET_ERROR(EACCES)); 1358} 1359 1360/* 1361 * Determine whether the mount is allowed according to MAC check. 1362 * by comparing (where appropriate) label of the dataset against 1363 * the label of the zone being mounted into. If the dataset has 1364 * no label, create one. 1365 * 1366 * Returns 0 if access allowed, error otherwise (e.g. EACCES) 1367 */ 1368static int 1369zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1370{ 1371 int error, retv; 1372 zone_t *mntzone = NULL; 1373 ts_label_t *mnt_tsl; 1374 bslabel_t *mnt_sl; 1375 bslabel_t ds_sl; 1376 char ds_hexsl[MAXNAMELEN]; 1377 1378 retv = EACCES; /* assume the worst */ 1379 1380 /* 1381 * Start by getting the dataset label if it exists. 1382 */ 1383 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1384 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1385 if (error) 1386 return (SET_ERROR(EACCES)); 1387 1388 /* 1389 * If labeling is NOT enabled, then disallow the mount of datasets 1390 * which have a non-default label already. No other label checks 1391 * are needed. 1392 */ 1393 if (!is_system_labeled()) { 1394 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1395 return (0); 1396 return (SET_ERROR(EACCES)); 1397 } 1398 1399 /* 1400 * Get the label of the mountpoint. If mounting into the global 1401 * zone (i.e. mountpoint is not within an active zone and the 1402 * zoned property is off), the label must be default or 1403 * admin_low/admin_high only; no other checks are needed. 1404 */ 1405 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1406 if (mntzone->zone_id == GLOBAL_ZONEID) { 1407 uint64_t zoned; 1408 1409 zone_rele(mntzone); 1410 1411 if (dsl_prop_get_integer(osname, 1412 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1413 return (SET_ERROR(EACCES)); 1414 if (!zoned) 1415 return (zfs_check_global_label(osname, ds_hexsl)); 1416 else 1417 /* 1418 * This is the case of a zone dataset being mounted 1419 * initially, before the zone has been fully created; 1420 * allow this mount into global zone. 1421 */ 1422 return (0); 1423 } 1424 1425 mnt_tsl = mntzone->zone_slabel; 1426 ASSERT(mnt_tsl != NULL); 1427 label_hold(mnt_tsl); 1428 mnt_sl = label2bslabel(mnt_tsl); 1429 1430 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1431 /* 1432 * The dataset doesn't have a real label, so fabricate one. 1433 */ 1434 char *str = NULL; 1435 1436 if (l_to_str_internal(mnt_sl, &str) == 0 && 1437 dsl_prop_set_string(osname, 1438 zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1439 ZPROP_SRC_LOCAL, str) == 0) 1440 retv = 0; 1441 if (str != NULL) 1442 kmem_free(str, strlen(str) + 1); 1443 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1444 /* 1445 * Now compare labels to complete the MAC check. If the 1446 * labels are equal then allow access. If the mountpoint 1447 * label dominates the dataset label, allow readonly access. 1448 * Otherwise, access is denied. 1449 */ 1450 if (blequal(mnt_sl, &ds_sl)) 1451 retv = 0; 1452 else if (bldominates(mnt_sl, &ds_sl)) { 1453 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1454 retv = 0; 1455 } 1456 } 1457 1458 label_rele(mnt_tsl); 1459 zone_rele(mntzone); 1460 return (retv); 1461} 1462#endif /* SECLABEL */ 1463 1464#ifdef OPENSOLARIS_MOUNTROOT 1465static int 1466zfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1467{ 1468 int error = 0; 1469 static int zfsrootdone = 0; 1470 zfsvfs_t *zfsvfs = NULL; 1471 znode_t *zp = NULL; 1472 vnode_t *vp = NULL; 1473 char *zfs_bootfs; 1474 char *zfs_devid; 1475 1476 ASSERT(vfsp); 1477 1478 /* 1479 * The filesystem that we mount as root is defined in the 1480 * boot property "zfs-bootfs" with a format of 1481 * "poolname/root-dataset-objnum". 1482 */ 1483 if (why == ROOT_INIT) { 1484 if (zfsrootdone++) 1485 return (SET_ERROR(EBUSY)); 1486 /* 1487 * the process of doing a spa_load will require the 1488 * clock to be set before we could (for example) do 1489 * something better by looking at the timestamp on 1490 * an uberblock, so just set it to -1. 1491 */ 1492 clkset(-1); 1493 1494 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1495 cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1496 "bootfs name"); 1497 return (SET_ERROR(EINVAL)); 1498 } 1499 zfs_devid = spa_get_bootprop("diskdevid"); 1500 error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1501 if (zfs_devid) 1502 spa_free_bootprop(zfs_devid); 1503 if (error) { 1504 spa_free_bootprop(zfs_bootfs); 1505 cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1506 error); 1507 return (error); 1508 } 1509 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1510 spa_free_bootprop(zfs_bootfs); 1511 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1512 error); 1513 return (error); 1514 } 1515 1516 spa_free_bootprop(zfs_bootfs); 1517 1518 if (error = vfs_lock(vfsp)) 1519 return (error); 1520 1521 if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1522 cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1523 goto out; 1524 } 1525 1526 zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1527 ASSERT(zfsvfs); 1528 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1529 cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1530 goto out; 1531 } 1532 1533 vp = ZTOV(zp); 1534 mutex_enter(&vp->v_lock); 1535 vp->v_flag |= VROOT; 1536 mutex_exit(&vp->v_lock); 1537 rootvp = vp; 1538 1539 /* 1540 * Leave rootvp held. The root file system is never unmounted. 1541 */ 1542 1543 vfs_add((struct vnode *)0, vfsp, 1544 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1545out: 1546 vfs_unlock(vfsp); 1547 return (error); 1548 } else if (why == ROOT_REMOUNT) { 1549 readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1550 vfsp->vfs_flag |= VFS_REMOUNT; 1551 1552 /* refresh mount options */ 1553 zfs_unregister_callbacks(vfsp->vfs_data); 1554 return (zfs_register_callbacks(vfsp)); 1555 1556 } else if (why == ROOT_UNMOUNT) { 1557 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1558 (void) zfs_sync(vfsp, 0, 0); 1559 return (0); 1560 } 1561 1562 /* 1563 * if "why" is equal to anything else other than ROOT_INIT, 1564 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1565 */ 1566 return (SET_ERROR(ENOTSUP)); 1567} 1568#endif /* OPENSOLARIS_MOUNTROOT */ 1569 1570static int 1571getpoolname(const char *osname, char *poolname) 1572{ 1573 char *p; 1574 1575 p = strchr(osname, '/'); 1576 if (p == NULL) { 1577 if (strlen(osname) >= MAXNAMELEN) 1578 return (ENAMETOOLONG); 1579 (void) strcpy(poolname, osname); 1580 } else { 1581 if (p - osname >= MAXNAMELEN) 1582 return (ENAMETOOLONG); 1583 (void) strncpy(poolname, osname, p - osname); 1584 poolname[p - osname] = '\0'; 1585 } 1586 return (0); 1587} 1588 1589/*ARGSUSED*/ 1590static int 1591zfs_mount(vfs_t *vfsp) 1592{ 1593 kthread_t *td = curthread; 1594 vnode_t *mvp = vfsp->mnt_vnodecovered; 1595 cred_t *cr = td->td_ucred; 1596 char *osname; 1597 int error = 0; 1598 int canwrite; 1599 1600#ifdef illumos 1601 if (mvp->v_type != VDIR) 1602 return (SET_ERROR(ENOTDIR)); 1603 1604 mutex_enter(&mvp->v_lock); 1605 if ((uap->flags & MS_REMOUNT) == 0 && 1606 (uap->flags & MS_OVERLAY) == 0 && 1607 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 1608 mutex_exit(&mvp->v_lock); 1609 return (SET_ERROR(EBUSY)); 1610 } 1611 mutex_exit(&mvp->v_lock); 1612 1613 /* 1614 * ZFS does not support passing unparsed data in via MS_DATA. 1615 * Users should use the MS_OPTIONSTR interface; this means 1616 * that all option parsing is already done and the options struct 1617 * can be interrogated. 1618 */ 1619 if ((uap->flags & MS_DATA) && uap->datalen > 0) 1620 return (SET_ERROR(EINVAL)); 1621 1622 /* 1623 * Get the objset name (the "special" mount argument). 1624 */ 1625 if (error = pn_get(uap->spec, fromspace, &spn)) 1626 return (error); 1627 1628 osname = spn.pn_path; 1629#else /* !illumos */ 1630 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1631 return (SET_ERROR(EPERM)); 1632 1633 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1634 return (SET_ERROR(EINVAL)); 1635 1636 /* 1637 * If full-owner-access is enabled and delegated administration is 1638 * turned on, we must set nosuid. 1639 */ 1640 if (zfs_super_owner && 1641 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1642 secpolicy_fs_mount_clearopts(cr, vfsp); 1643 } 1644#endif /* illumos */ 1645 1646 /* 1647 * Check for mount privilege? 1648 * 1649 * If we don't have privilege then see if 1650 * we have local permission to allow it 1651 */ 1652 error = secpolicy_fs_mount(cr, mvp, vfsp); 1653 if (error) { 1654 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1655 goto out; 1656 1657 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1658 vattr_t vattr; 1659 1660 /* 1661 * Make sure user is the owner of the mount point 1662 * or has sufficient privileges. 1663 */ 1664 1665 vattr.va_mask = AT_UID; 1666 1667 vn_lock(mvp, LK_SHARED | LK_RETRY); 1668 if (VOP_GETATTR(mvp, &vattr, cr)) { 1669 VOP_UNLOCK(mvp, 0); 1670 goto out; 1671 } 1672 1673 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1674 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1675 VOP_UNLOCK(mvp, 0); 1676 goto out; 1677 } 1678 VOP_UNLOCK(mvp, 0); 1679 } 1680 1681 secpolicy_fs_mount_clearopts(cr, vfsp); 1682 } 1683 1684 /* 1685 * Refuse to mount a filesystem if we are in a local zone and the 1686 * dataset is not visible. 1687 */ 1688 if (!INGLOBALZONE(curthread) && 1689 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1690 error = SET_ERROR(EPERM); 1691 goto out; 1692 } 1693 1694#ifdef SECLABEL 1695 error = zfs_mount_label_policy(vfsp, osname); 1696 if (error) 1697 goto out; 1698#endif 1699 1700 vfsp->vfs_flag |= MNT_NFS4ACLS; 1701 1702 /* 1703 * When doing a remount, we simply refresh our temporary properties 1704 * according to those options set in the current VFS options. 1705 */ 1706 if (vfsp->vfs_flag & MS_REMOUNT) { 1707 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1708 1709 /* 1710 * Refresh mount options with z_teardown_lock blocking I/O while 1711 * the filesystem is in an inconsistent state. 1712 * The lock also serializes this code with filesystem 1713 * manipulations between entry to zfs_suspend_fs() and return 1714 * from zfs_resume_fs(). 1715 */ 1716 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1717 zfs_unregister_callbacks(zfsvfs); 1718 error = zfs_register_callbacks(vfsp); 1719 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1720 goto out; 1721 } 1722 1723 /* Initial root mount: try hard to import the requested root pool. */ 1724 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1725 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1726 char pname[MAXNAMELEN]; 1727 1728 error = getpoolname(osname, pname); 1729 if (error == 0) 1730 error = spa_import_rootpool(pname); 1731 if (error) 1732 goto out; 1733 } 1734 DROP_GIANT(); 1735 error = zfs_domount(vfsp, osname); 1736 PICKUP_GIANT(); 1737 1738#ifdef illumos 1739 /* 1740 * Add an extra VFS_HOLD on our parent vfs so that it can't 1741 * disappear due to a forced unmount. 1742 */ 1743 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1744 VFS_HOLD(mvp->v_vfsp); 1745#endif 1746 1747out: 1748 return (error); 1749} 1750 1751static int 1752zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1753{ 1754 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1755 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1756 1757 statp->f_version = STATFS_VERSION; 1758 1759 ZFS_ENTER(zfsvfs); 1760 1761 dmu_objset_space(zfsvfs->z_os, 1762 &refdbytes, &availbytes, &usedobjs, &availobjs); 1763 1764 /* 1765 * The underlying storage pool actually uses multiple block sizes. 1766 * We report the fragsize as the smallest block size we support, 1767 * and we report our blocksize as the filesystem's maximum blocksize. 1768 */ 1769 statp->f_bsize = SPA_MINBLOCKSIZE; 1770 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1771 1772 /* 1773 * The following report "total" blocks of various kinds in the 1774 * file system, but reported in terms of f_frsize - the 1775 * "fragment" size. 1776 */ 1777 1778 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1779 statp->f_bfree = availbytes / statp->f_bsize; 1780 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1781 1782 /* 1783 * statvfs() should really be called statufs(), because it assumes 1784 * static metadata. ZFS doesn't preallocate files, so the best 1785 * we can do is report the max that could possibly fit in f_files, 1786 * and that minus the number actually used in f_ffree. 1787 * For f_ffree, report the smaller of the number of object available 1788 * and the number of blocks (each object will take at least a block). 1789 */ 1790 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1791 statp->f_files = statp->f_ffree + usedobjs; 1792 1793 /* 1794 * We're a zfs filesystem. 1795 */ 1796 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1797 1798 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1799 sizeof(statp->f_mntfromname)); 1800 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1801 sizeof(statp->f_mntonname)); 1802 1803 statp->f_namemax = MAXNAMELEN - 1; 1804 1805 ZFS_EXIT(zfsvfs); 1806 return (0); 1807} 1808 1809static int 1810zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1811{ 1812 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1813 znode_t *rootzp; 1814 int error; 1815 1816 ZFS_ENTER(zfsvfs); 1817 1818 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1819 if (error == 0) 1820 *vpp = ZTOV(rootzp); 1821 1822 ZFS_EXIT(zfsvfs); 1823 1824 if (error == 0) { 1825 error = vn_lock(*vpp, flags); 1826 if (error != 0) { 1827 VN_RELE(*vpp); 1828 *vpp = NULL; 1829 } 1830 } 1831 return (error); 1832} 1833 1834/* 1835 * Teardown the zfsvfs::z_os. 1836 * 1837 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1838 * and 'z_teardown_inactive_lock' held. 1839 */ 1840static int 1841zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1842{ 1843 znode_t *zp; 1844 1845 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1846 1847 if (!unmounting) { 1848 /* 1849 * We purge the parent filesystem's vfsp as the parent 1850 * filesystem and all of its snapshots have their vnode's 1851 * v_vfsp set to the parent's filesystem's vfsp. Note, 1852 * 'z_parent' is self referential for non-snapshots. 1853 */ 1854 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1855#ifdef FREEBSD_NAMECACHE 1856 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1857#endif 1858 } 1859 1860 /* 1861 * Close the zil. NB: Can't close the zil while zfs_inactive 1862 * threads are blocked as zil_close can call zfs_inactive. 1863 */ 1864 if (zfsvfs->z_log) { 1865 zil_close(zfsvfs->z_log); 1866 zfsvfs->z_log = NULL; 1867 } 1868 1869 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1870 1871 /* 1872 * If we are not unmounting (ie: online recv) and someone already 1873 * unmounted this file system while we were doing the switcheroo, 1874 * or a reopen of z_os failed then just bail out now. 1875 */ 1876 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1877 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1878 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1879 return (SET_ERROR(EIO)); 1880 } 1881 1882 /* 1883 * At this point there are no vops active, and any new vops will 1884 * fail with EIO since we have z_teardown_lock for writer (only 1885 * relavent for forced unmount). 1886 * 1887 * Release all holds on dbufs. 1888 */ 1889 mutex_enter(&zfsvfs->z_znodes_lock); 1890 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1891 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1892 if (zp->z_sa_hdl) { 1893 ASSERT(ZTOV(zp)->v_count >= 0); 1894 zfs_znode_dmu_fini(zp); 1895 } 1896 mutex_exit(&zfsvfs->z_znodes_lock); 1897 1898 /* 1899 * If we are unmounting, set the unmounted flag and let new vops 1900 * unblock. zfs_inactive will have the unmounted behavior, and all 1901 * other vops will fail with EIO. 1902 */ 1903 if (unmounting) { 1904 zfsvfs->z_unmounted = B_TRUE; 1905 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1906 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1907 } 1908 1909 /* 1910 * z_os will be NULL if there was an error in attempting to reopen 1911 * zfsvfs, so just return as the properties had already been 1912 * unregistered and cached data had been evicted before. 1913 */ 1914 if (zfsvfs->z_os == NULL) 1915 return (0); 1916 1917 /* 1918 * Unregister properties. 1919 */ 1920 zfs_unregister_callbacks(zfsvfs); 1921 1922 /* 1923 * Evict cached data 1924 */ 1925 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1926 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1927 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1928 dmu_objset_evict_dbufs(zfsvfs->z_os); 1929 1930 return (0); 1931} 1932 1933/*ARGSUSED*/ 1934static int 1935zfs_umount(vfs_t *vfsp, int fflag) 1936{ 1937 kthread_t *td = curthread; 1938 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1939 objset_t *os; 1940 cred_t *cr = td->td_ucred; 1941 int ret; 1942 1943 ret = secpolicy_fs_unmount(cr, vfsp); 1944 if (ret) { 1945 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1946 ZFS_DELEG_PERM_MOUNT, cr)) 1947 return (ret); 1948 } 1949 1950 /* 1951 * We purge the parent filesystem's vfsp as the parent filesystem 1952 * and all of its snapshots have their vnode's v_vfsp set to the 1953 * parent's filesystem's vfsp. Note, 'z_parent' is self 1954 * referential for non-snapshots. 1955 */ 1956 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1957 1958 /* 1959 * Unmount any snapshots mounted under .zfs before unmounting the 1960 * dataset itself. 1961 */ 1962 if (zfsvfs->z_ctldir != NULL) { 1963 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1964 return (ret); 1965 } 1966 1967 if (fflag & MS_FORCE) { 1968 /* 1969 * Mark file system as unmounted before calling 1970 * vflush(FORCECLOSE). This way we ensure no future vnops 1971 * will be called and risk operating on DOOMED vnodes. 1972 */ 1973 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1974 zfsvfs->z_unmounted = B_TRUE; 1975 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1976 } 1977 1978 /* 1979 * Flush all the files. 1980 */ 1981 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1982 if (ret != 0) 1983 return (ret); 1984 1985#ifdef illumos 1986 if (!(fflag & MS_FORCE)) { 1987 /* 1988 * Check the number of active vnodes in the file system. 1989 * Our count is maintained in the vfs structure, but the 1990 * number is off by 1 to indicate a hold on the vfs 1991 * structure itself. 1992 * 1993 * The '.zfs' directory maintains a reference of its 1994 * own, and any active references underneath are 1995 * reflected in the vnode count. 1996 */ 1997 if (zfsvfs->z_ctldir == NULL) { 1998 if (vfsp->vfs_count > 1) 1999 return (SET_ERROR(EBUSY)); 2000 } else { 2001 if (vfsp->vfs_count > 2 || 2002 zfsvfs->z_ctldir->v_count > 1) 2003 return (SET_ERROR(EBUSY)); 2004 } 2005 } 2006#endif 2007 2008 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 2009 os = zfsvfs->z_os; 2010 2011 /* 2012 * z_os will be NULL if there was an error in 2013 * attempting to reopen zfsvfs. 2014 */ 2015 if (os != NULL) { 2016 /* 2017 * Unset the objset user_ptr. 2018 */ 2019 mutex_enter(&os->os_user_ptr_lock); 2020 dmu_objset_set_user(os, NULL); 2021 mutex_exit(&os->os_user_ptr_lock); 2022 2023 /* 2024 * Finally release the objset 2025 */ 2026 dmu_objset_disown(os, zfsvfs); 2027 } 2028 2029 /* 2030 * We can now safely destroy the '.zfs' directory node. 2031 */ 2032 if (zfsvfs->z_ctldir != NULL) 2033 zfsctl_destroy(zfsvfs); 2034 zfs_freevfs(vfsp); 2035 2036 return (0); 2037} 2038 2039static int 2040zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2041{ 2042 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2043 znode_t *zp; 2044 int err; 2045 2046 /* 2047 * zfs_zget() can't operate on virtual entries like .zfs/ or 2048 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2049 * This will make NFS to switch to LOOKUP instead of using VGET. 2050 */ 2051 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || 2052 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) 2053 return (EOPNOTSUPP); 2054 2055 ZFS_ENTER(zfsvfs); 2056 err = zfs_zget(zfsvfs, ino, &zp); 2057 if (err == 0 && zp->z_unlinked) { 2058 vrele(ZTOV(zp)); 2059 err = EINVAL; 2060 } 2061 if (err == 0) 2062 *vpp = ZTOV(zp); 2063 ZFS_EXIT(zfsvfs); 2064 if (err == 0) 2065 err = vn_lock(*vpp, flags); 2066 if (err != 0) 2067 *vpp = NULL; 2068 return (err); 2069} 2070 2071static int 2072zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2073 struct ucred **credanonp, int *numsecflavors, int **secflavors) 2074{ 2075 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2076 2077 /* 2078 * If this is regular file system vfsp is the same as 2079 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2080 * zfsvfs->z_parent->z_vfs represents parent file system 2081 * which we have to use here, because only this file system 2082 * has mnt_export configured. 2083 */ 2084 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2085 credanonp, numsecflavors, secflavors)); 2086} 2087 2088CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2089CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2090 2091static int 2092zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2093{ 2094 struct componentname cn; 2095 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2096 znode_t *zp; 2097 vnode_t *dvp; 2098 uint64_t object = 0; 2099 uint64_t fid_gen = 0; 2100 uint64_t gen_mask; 2101 uint64_t zp_gen; 2102 int i, err; 2103 2104 *vpp = NULL; 2105 2106 ZFS_ENTER(zfsvfs); 2107 2108 /* 2109 * On FreeBSD we can get snapshot's mount point or its parent file 2110 * system mount point depending if snapshot is already mounted or not. 2111 */ 2112 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2113 zfid_long_t *zlfid = (zfid_long_t *)fidp; 2114 uint64_t objsetid = 0; 2115 uint64_t setgen = 0; 2116 2117 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2118 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2119 2120 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2121 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2122 2123 ZFS_EXIT(zfsvfs); 2124 2125 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2126 if (err) 2127 return (SET_ERROR(EINVAL)); 2128 ZFS_ENTER(zfsvfs); 2129 } 2130 2131 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2132 zfid_short_t *zfid = (zfid_short_t *)fidp; 2133 2134 for (i = 0; i < sizeof (zfid->zf_object); i++) 2135 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2136 2137 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2138 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2139 } else { 2140 ZFS_EXIT(zfsvfs); 2141 return (SET_ERROR(EINVAL)); 2142 } 2143 2144 /* 2145 * A zero fid_gen means we are in .zfs or the .zfs/snapshot 2146 * directory tree. If the object == zfsvfs->z_shares_dir, then 2147 * we are in the .zfs/shares directory tree. 2148 */ 2149 if ((fid_gen == 0 && 2150 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || 2151 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { 2152 ZFS_EXIT(zfsvfs); 2153 VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp)); 2154 if (object == ZFSCTL_INO_SNAPDIR) { 2155 cn.cn_nameptr = "snapshot"; 2156 cn.cn_namelen = strlen(cn.cn_nameptr); 2157 cn.cn_nameiop = LOOKUP; 2158 cn.cn_flags = ISLASTCN | LOCKLEAF; 2159 cn.cn_lkflags = flags; 2160 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn)); 2161 vput(dvp); 2162 } else if (object == zfsvfs->z_shares_dir) { 2163 /* 2164 * XXX This branch must not be taken, 2165 * if it is, then the lookup below will 2166 * explode. 2167 */ 2168 cn.cn_nameptr = "shares"; 2169 cn.cn_namelen = strlen(cn.cn_nameptr); 2170 cn.cn_nameiop = LOOKUP; 2171 cn.cn_flags = ISLASTCN; 2172 cn.cn_lkflags = flags; 2173 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn)); 2174 vput(dvp); 2175 } else { 2176 *vpp = dvp; 2177 } 2178 return (err); 2179 } 2180 2181 gen_mask = -1ULL >> (64 - 8 * i); 2182 2183 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2184 if (err = zfs_zget(zfsvfs, object, &zp)) { 2185 ZFS_EXIT(zfsvfs); 2186 return (err); 2187 } 2188 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2189 sizeof (uint64_t)); 2190 zp_gen = zp_gen & gen_mask; 2191 if (zp_gen == 0) 2192 zp_gen = 1; 2193 if (zp->z_unlinked || zp_gen != fid_gen) { 2194 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2195 vrele(ZTOV(zp)); 2196 ZFS_EXIT(zfsvfs); 2197 return (SET_ERROR(EINVAL)); 2198 } 2199 2200 *vpp = ZTOV(zp); 2201 ZFS_EXIT(zfsvfs); 2202 err = vn_lock(*vpp, flags); 2203 if (err == 0) 2204 vnode_create_vobject(*vpp, zp->z_size, curthread); 2205 else 2206 *vpp = NULL; 2207 return (err); 2208} 2209 2210/* 2211 * Block out VOPs and close zfsvfs_t::z_os 2212 * 2213 * Note, if successful, then we return with the 'z_teardown_lock' and 2214 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 2215 * dataset and objset intact so that they can be atomically handed off during 2216 * a subsequent rollback or recv operation and the resume thereafter. 2217 */ 2218int 2219zfs_suspend_fs(zfsvfs_t *zfsvfs) 2220{ 2221 int error; 2222 2223 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2224 return (error); 2225 2226 return (0); 2227} 2228 2229/* 2230 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 2231 * is an invariant across any of the operations that can be performed while the 2232 * filesystem was suspended. Whether it succeeded or failed, the preconditions 2233 * are the same: the relevant objset and associated dataset are owned by 2234 * zfsvfs, held, and long held on entry. 2235 */ 2236int 2237zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) 2238{ 2239 int err; 2240 znode_t *zp; 2241 2242 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2243 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2244 2245 /* 2246 * We already own this, so just update the objset_t, as the one we 2247 * had before may have been evicted. 2248 */ 2249 objset_t *os; 2250 VERIFY3P(ds->ds_owner, ==, zfsvfs); 2251 VERIFY(dsl_dataset_long_held(ds)); 2252 VERIFY0(dmu_objset_from_ds(ds, &os)); 2253 2254 err = zfsvfs_init(zfsvfs, os); 2255 if (err != 0) 2256 goto bail; 2257 2258 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2259 2260 zfs_set_fuid_feature(zfsvfs); 2261 2262 /* 2263 * Attempt to re-establish all the active znodes with 2264 * their dbufs. If a zfs_rezget() fails, then we'll let 2265 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2266 * when they try to use their znode. 2267 */ 2268 mutex_enter(&zfsvfs->z_znodes_lock); 2269 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2270 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2271 (void) zfs_rezget(zp); 2272 } 2273 mutex_exit(&zfsvfs->z_znodes_lock); 2274 2275bail: 2276 /* release the VOPs */ 2277 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2278 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 2279 2280 if (err) { 2281 /* 2282 * Since we couldn't setup the sa framework, try to force 2283 * unmount this file system. 2284 */ 2285 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) { 2286 vfs_ref(zfsvfs->z_vfs); 2287 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2288 } 2289 } 2290 return (err); 2291} 2292 2293static void 2294zfs_freevfs(vfs_t *vfsp) 2295{ 2296 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2297 2298#ifdef illumos 2299 /* 2300 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2301 * from zfs_mount(). Release it here. If we came through 2302 * zfs_mountroot() instead, we didn't grab an extra hold, so 2303 * skip the VFS_RELE for rootvfs. 2304 */ 2305 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2306 VFS_RELE(zfsvfs->z_parent->z_vfs); 2307#endif 2308 2309 zfsvfs_free(zfsvfs); 2310 2311 atomic_dec_32(&zfs_active_fs_count); 2312} 2313 2314#ifdef __i386__ 2315static int desiredvnodes_backup; 2316#endif 2317 2318static void 2319zfs_vnodes_adjust(void) 2320{ 2321#ifdef __i386__ 2322 int newdesiredvnodes; 2323 2324 desiredvnodes_backup = desiredvnodes; 2325 2326 /* 2327 * We calculate newdesiredvnodes the same way it is done in 2328 * vntblinit(). If it is equal to desiredvnodes, it means that 2329 * it wasn't tuned by the administrator and we can tune it down. 2330 */ 2331 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2332 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2333 sizeof(struct vnode)))); 2334 if (newdesiredvnodes == desiredvnodes) 2335 desiredvnodes = (3 * newdesiredvnodes) / 4; 2336#endif 2337} 2338 2339static void 2340zfs_vnodes_adjust_back(void) 2341{ 2342 2343#ifdef __i386__ 2344 desiredvnodes = desiredvnodes_backup; 2345#endif 2346} 2347 2348void 2349zfs_init(void) 2350{ 2351 2352 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2353 2354 /* 2355 * Initialize .zfs directory structures 2356 */ 2357 zfsctl_init(); 2358 2359 /* 2360 * Initialize znode cache, vnode ops, etc... 2361 */ 2362 zfs_znode_init(); 2363 2364 /* 2365 * Reduce number of vnodes. Originally number of vnodes is calculated 2366 * with UFS inode in mind. We reduce it here, because it's too big for 2367 * ZFS/i386. 2368 */ 2369 zfs_vnodes_adjust(); 2370 2371 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2372} 2373 2374void 2375zfs_fini(void) 2376{ 2377 zfsctl_fini(); 2378 zfs_znode_fini(); 2379 zfs_vnodes_adjust_back(); 2380} 2381 2382int 2383zfs_busy(void) 2384{ 2385 return (zfs_active_fs_count != 0); 2386} 2387 2388int 2389zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2390{ 2391 int error; 2392 objset_t *os = zfsvfs->z_os; 2393 dmu_tx_t *tx; 2394 2395 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2396 return (SET_ERROR(EINVAL)); 2397 2398 if (newvers < zfsvfs->z_version) 2399 return (SET_ERROR(EINVAL)); 2400 2401 if (zfs_spa_version_map(newvers) > 2402 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2403 return (SET_ERROR(ENOTSUP)); 2404 2405 tx = dmu_tx_create(os); 2406 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2407 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2408 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2409 ZFS_SA_ATTRS); 2410 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2411 } 2412 error = dmu_tx_assign(tx, TXG_WAIT); 2413 if (error) { 2414 dmu_tx_abort(tx); 2415 return (error); 2416 } 2417 2418 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2419 8, 1, &newvers, tx); 2420 2421 if (error) { 2422 dmu_tx_commit(tx); 2423 return (error); 2424 } 2425 2426 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2427 uint64_t sa_obj; 2428 2429 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2430 SPA_VERSION_SA); 2431 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2432 DMU_OT_NONE, 0, tx); 2433 2434 error = zap_add(os, MASTER_NODE_OBJ, 2435 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2436 ASSERT0(error); 2437 2438 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2439 sa_register_update_callback(os, zfs_sa_upgrade); 2440 } 2441 2442 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2443 "from %llu to %llu", zfsvfs->z_version, newvers); 2444 2445 dmu_tx_commit(tx); 2446 2447 zfsvfs->z_version = newvers; 2448 2449 zfs_set_fuid_feature(zfsvfs); 2450 2451 return (0); 2452} 2453 2454/* 2455 * Read a property stored within the master node. 2456 */ 2457int 2458zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2459{ 2460 const char *pname; 2461 int error = ENOENT; 2462 2463 /* 2464 * Look up the file system's value for the property. For the 2465 * version property, we look up a slightly different string. 2466 */ 2467 if (prop == ZFS_PROP_VERSION) 2468 pname = ZPL_VERSION_STR; 2469 else 2470 pname = zfs_prop_to_name(prop); 2471 2472 if (os != NULL) 2473 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2474 2475 if (error == ENOENT) { 2476 /* No value set, use the default value */ 2477 switch (prop) { 2478 case ZFS_PROP_VERSION: 2479 *value = ZPL_VERSION; 2480 break; 2481 case ZFS_PROP_NORMALIZE: 2482 case ZFS_PROP_UTF8ONLY: 2483 *value = 0; 2484 break; 2485 case ZFS_PROP_CASE: 2486 *value = ZFS_CASE_SENSITIVE; 2487 break; 2488 default: 2489 return (error); 2490 } 2491 error = 0; 2492 } 2493 return (error); 2494} 2495 2496#ifdef _KERNEL 2497void 2498zfsvfs_update_fromname(const char *oldname, const char *newname) 2499{ 2500 char tmpbuf[MAXPATHLEN]; 2501 struct mount *mp; 2502 char *fromname; 2503 size_t oldlen; 2504 2505 oldlen = strlen(oldname); 2506 2507 mtx_lock(&mountlist_mtx); 2508 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2509 fromname = mp->mnt_stat.f_mntfromname; 2510 if (strcmp(fromname, oldname) == 0) { 2511 (void)strlcpy(fromname, newname, 2512 sizeof(mp->mnt_stat.f_mntfromname)); 2513 continue; 2514 } 2515 if (strncmp(fromname, oldname, oldlen) == 0 && 2516 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2517 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2518 newname, fromname + oldlen); 2519 (void)strlcpy(fromname, tmpbuf, 2520 sizeof(mp->mnt_stat.f_mntfromname)); 2521 continue; 2522 } 2523 } 2524 mtx_unlock(&mountlist_mtx); 2525} 2526#endif 2527