zfs_vfsops.c revision 297082
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 */ 27 28/* Portions Copyright 2010 Robert Milkowski */ 29 30#include <sys/types.h> 31#include <sys/param.h> 32#include <sys/systm.h> 33#include <sys/kernel.h> 34#include <sys/sysmacros.h> 35#include <sys/kmem.h> 36#include <sys/acl.h> 37#include <sys/vnode.h> 38#include <sys/vfs.h> 39#include <sys/mntent.h> 40#include <sys/mount.h> 41#include <sys/cmn_err.h> 42#include <sys/zfs_znode.h> 43#include <sys/zfs_dir.h> 44#include <sys/zil.h> 45#include <sys/fs/zfs.h> 46#include <sys/dmu.h> 47#include <sys/dsl_prop.h> 48#include <sys/dsl_dataset.h> 49#include <sys/dsl_deleg.h> 50#include <sys/spa.h> 51#include <sys/zap.h> 52#include <sys/sa.h> 53#include <sys/sa_impl.h> 54#include <sys/varargs.h> 55#include <sys/policy.h> 56#include <sys/atomic.h> 57#include <sys/zfs_ioctl.h> 58#include <sys/zfs_ctldir.h> 59#include <sys/zfs_fuid.h> 60#include <sys/sunddi.h> 61#include <sys/dnlc.h> 62#include <sys/dmu_objset.h> 63#include <sys/spa_boot.h> 64#include <sys/jail.h> 65#include "zfs_comutil.h" 66 67struct mtx zfs_debug_mtx; 68MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF); 69 70SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system"); 71 72int zfs_super_owner; 73SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0, 74 "File system owner can perform privileged operation on his file systems"); 75 76int zfs_debug_level; 77TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level); 78SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0, 79 "Debug level"); 80 81SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions"); 82static int zfs_version_acl = ZFS_ACL_VERSION; 83SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0, 84 "ZFS_ACL_VERSION"); 85static int zfs_version_spa = SPA_VERSION; 86SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0, 87 "SPA_VERSION"); 88static int zfs_version_zpl = ZPL_VERSION; 89SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0, 90 "ZPL_VERSION"); 91 92static int zfs_mount(vfs_t *vfsp); 93static int zfs_umount(vfs_t *vfsp, int fflag); 94static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp); 95static int zfs_statfs(vfs_t *vfsp, struct statfs *statp); 96static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp); 97static int zfs_sync(vfs_t *vfsp, int waitfor); 98static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 99 struct ucred **credanonp, int *numsecflavors, int **secflavors); 100static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp); 101static void zfs_objset_close(zfsvfs_t *zfsvfs); 102static void zfs_freevfs(vfs_t *vfsp); 103 104static struct vfsops zfs_vfsops = { 105 .vfs_mount = zfs_mount, 106 .vfs_unmount = zfs_umount, 107 .vfs_root = zfs_root, 108 .vfs_statfs = zfs_statfs, 109 .vfs_vget = zfs_vget, 110 .vfs_sync = zfs_sync, 111 .vfs_checkexp = zfs_checkexp, 112 .vfs_fhtovp = zfs_fhtovp, 113}; 114 115VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN); 116 117/* 118 * We need to keep a count of active fs's. 119 * This is necessary to prevent our module 120 * from being unloaded after a umount -f 121 */ 122static uint32_t zfs_active_fs_count = 0; 123 124/*ARGSUSED*/ 125static int 126zfs_sync(vfs_t *vfsp, int waitfor) 127{ 128 129 /* 130 * Data integrity is job one. We don't want a compromised kernel 131 * writing to the storage pool, so we never sync during panic. 132 */ 133 if (panicstr) 134 return (0); 135 136 /* 137 * Ignore the system syncher. ZFS already commits async data 138 * at zfs_txg_timeout intervals. 139 */ 140 if (waitfor == MNT_LAZY) 141 return (0); 142 143 if (vfsp != NULL) { 144 /* 145 * Sync a specific filesystem. 146 */ 147 zfsvfs_t *zfsvfs = vfsp->vfs_data; 148 dsl_pool_t *dp; 149 int error; 150 151 error = vfs_stdsync(vfsp, waitfor); 152 if (error != 0) 153 return (error); 154 155 ZFS_ENTER(zfsvfs); 156 dp = dmu_objset_pool(zfsvfs->z_os); 157 158 /* 159 * If the system is shutting down, then skip any 160 * filesystems which may exist on a suspended pool. 161 */ 162 if (sys_shutdown && spa_suspended(dp->dp_spa)) { 163 ZFS_EXIT(zfsvfs); 164 return (0); 165 } 166 167 if (zfsvfs->z_log != NULL) 168 zil_commit(zfsvfs->z_log, 0); 169 170 ZFS_EXIT(zfsvfs); 171 } else { 172 /* 173 * Sync all ZFS filesystems. This is what happens when you 174 * run sync(1M). Unlike other filesystems, ZFS honors the 175 * request by waiting for all pools to commit all dirty data. 176 */ 177 spa_sync_allpools(); 178 } 179 180 return (0); 181} 182 183#ifndef __FreeBSD_kernel__ 184static int 185zfs_create_unique_device(dev_t *dev) 186{ 187 major_t new_major; 188 189 do { 190 ASSERT3U(zfs_minor, <=, MAXMIN32); 191 minor_t start = zfs_minor; 192 do { 193 mutex_enter(&zfs_dev_mtx); 194 if (zfs_minor >= MAXMIN32) { 195 /* 196 * If we're still using the real major 197 * keep out of /dev/zfs and /dev/zvol minor 198 * number space. If we're using a getudev()'ed 199 * major number, we can use all of its minors. 200 */ 201 if (zfs_major == ddi_name_to_major(ZFS_DRIVER)) 202 zfs_minor = ZFS_MIN_MINOR; 203 else 204 zfs_minor = 0; 205 } else { 206 zfs_minor++; 207 } 208 *dev = makedevice(zfs_major, zfs_minor); 209 mutex_exit(&zfs_dev_mtx); 210 } while (vfs_devismounted(*dev) && zfs_minor != start); 211 if (zfs_minor == start) { 212 /* 213 * We are using all ~262,000 minor numbers for the 214 * current major number. Create a new major number. 215 */ 216 if ((new_major = getudev()) == (major_t)-1) { 217 cmn_err(CE_WARN, 218 "zfs_mount: Can't get unique major " 219 "device number."); 220 return (-1); 221 } 222 mutex_enter(&zfs_dev_mtx); 223 zfs_major = new_major; 224 zfs_minor = 0; 225 226 mutex_exit(&zfs_dev_mtx); 227 } else { 228 break; 229 } 230 /* CONSTANTCONDITION */ 231 } while (1); 232 233 return (0); 234} 235#endif /* !__FreeBSD_kernel__ */ 236 237static void 238atime_changed_cb(void *arg, uint64_t newval) 239{ 240 zfsvfs_t *zfsvfs = arg; 241 242 if (newval == TRUE) { 243 zfsvfs->z_atime = TRUE; 244 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; 245 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); 246 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); 247 } else { 248 zfsvfs->z_atime = FALSE; 249 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; 250 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); 251 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); 252 } 253} 254 255static void 256xattr_changed_cb(void *arg, uint64_t newval) 257{ 258 zfsvfs_t *zfsvfs = arg; 259 260 if (newval == TRUE) { 261 /* XXX locking on vfs_flag? */ 262#ifdef TODO 263 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; 264#endif 265 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); 266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); 267 } else { 268 /* XXX locking on vfs_flag? */ 269#ifdef TODO 270 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR; 271#endif 272 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR); 273 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); 274 } 275} 276 277static void 278blksz_changed_cb(void *arg, uint64_t newval) 279{ 280 zfsvfs_t *zfsvfs = arg; 281 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os))); 282 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); 283 ASSERT(ISP2(newval)); 284 285 zfsvfs->z_max_blksz = newval; 286 zfsvfs->z_vfs->mnt_stat.f_iosize = newval; 287} 288 289static void 290readonly_changed_cb(void *arg, uint64_t newval) 291{ 292 zfsvfs_t *zfsvfs = arg; 293 294 if (newval) { 295 /* XXX locking on vfs_flag? */ 296 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; 297 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); 298 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); 299 } else { 300 /* XXX locking on vfs_flag? */ 301 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 302 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); 303 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); 304 } 305} 306 307static void 308setuid_changed_cb(void *arg, uint64_t newval) 309{ 310 zfsvfs_t *zfsvfs = arg; 311 312 if (newval == FALSE) { 313 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; 314 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); 315 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); 316 } else { 317 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; 318 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); 319 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); 320 } 321} 322 323static void 324exec_changed_cb(void *arg, uint64_t newval) 325{ 326 zfsvfs_t *zfsvfs = arg; 327 328 if (newval == FALSE) { 329 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; 330 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); 331 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); 332 } else { 333 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; 334 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); 335 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); 336 } 337} 338 339/* 340 * The nbmand mount option can be changed at mount time. 341 * We can't allow it to be toggled on live file systems or incorrect 342 * behavior may be seen from cifs clients 343 * 344 * This property isn't registered via dsl_prop_register(), but this callback 345 * will be called when a file system is first mounted 346 */ 347static void 348nbmand_changed_cb(void *arg, uint64_t newval) 349{ 350 zfsvfs_t *zfsvfs = arg; 351 if (newval == FALSE) { 352 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); 353 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); 354 } else { 355 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); 356 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); 357 } 358} 359 360static void 361snapdir_changed_cb(void *arg, uint64_t newval) 362{ 363 zfsvfs_t *zfsvfs = arg; 364 365 zfsvfs->z_show_ctldir = newval; 366} 367 368static void 369vscan_changed_cb(void *arg, uint64_t newval) 370{ 371 zfsvfs_t *zfsvfs = arg; 372 373 zfsvfs->z_vscan = newval; 374} 375 376static void 377acl_mode_changed_cb(void *arg, uint64_t newval) 378{ 379 zfsvfs_t *zfsvfs = arg; 380 381 zfsvfs->z_acl_mode = newval; 382} 383 384static void 385acl_inherit_changed_cb(void *arg, uint64_t newval) 386{ 387 zfsvfs_t *zfsvfs = arg; 388 389 zfsvfs->z_acl_inherit = newval; 390} 391 392static int 393zfs_register_callbacks(vfs_t *vfsp) 394{ 395 struct dsl_dataset *ds = NULL; 396 objset_t *os = NULL; 397 zfsvfs_t *zfsvfs = NULL; 398 uint64_t nbmand; 399 boolean_t readonly = B_FALSE; 400 boolean_t do_readonly = B_FALSE; 401 boolean_t setuid = B_FALSE; 402 boolean_t do_setuid = B_FALSE; 403 boolean_t exec = B_FALSE; 404 boolean_t do_exec = B_FALSE; 405#ifdef illumos 406 boolean_t devices = B_FALSE; 407 boolean_t do_devices = B_FALSE; 408#endif 409 boolean_t xattr = B_FALSE; 410 boolean_t do_xattr = B_FALSE; 411 boolean_t atime = B_FALSE; 412 boolean_t do_atime = B_FALSE; 413 int error = 0; 414 415 ASSERT(vfsp); 416 zfsvfs = vfsp->vfs_data; 417 ASSERT(zfsvfs); 418 os = zfsvfs->z_os; 419 420 /* 421 * This function can be called for a snapshot when we update snapshot's 422 * mount point, which isn't really supported. 423 */ 424 if (dmu_objset_is_snapshot(os)) 425 return (EOPNOTSUPP); 426 427 /* 428 * The act of registering our callbacks will destroy any mount 429 * options we may have. In order to enable temporary overrides 430 * of mount options, we stash away the current values and 431 * restore them after we register the callbacks. 432 */ 433 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) || 434 !spa_writeable(dmu_objset_spa(os))) { 435 readonly = B_TRUE; 436 do_readonly = B_TRUE; 437 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 438 readonly = B_FALSE; 439 do_readonly = B_TRUE; 440 } 441 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 442 setuid = B_FALSE; 443 do_setuid = B_TRUE; 444 } else { 445 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 446 setuid = B_FALSE; 447 do_setuid = B_TRUE; 448 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 449 setuid = B_TRUE; 450 do_setuid = B_TRUE; 451 } 452 } 453 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 454 exec = B_FALSE; 455 do_exec = B_TRUE; 456 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 457 exec = B_TRUE; 458 do_exec = B_TRUE; 459 } 460 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 461 xattr = B_FALSE; 462 do_xattr = B_TRUE; 463 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 464 xattr = B_TRUE; 465 do_xattr = B_TRUE; 466 } 467 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 468 atime = B_FALSE; 469 do_atime = B_TRUE; 470 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 471 atime = B_TRUE; 472 do_atime = B_TRUE; 473 } 474 475 /* 476 * We need to enter pool configuration here, so that we can use 477 * dsl_prop_get_int_ds() to handle the special nbmand property below. 478 * dsl_prop_get_integer() can not be used, because it has to acquire 479 * spa_namespace_lock and we can not do that because we already hold 480 * z_teardown_lock. The problem is that spa_config_sync() is called 481 * with spa_namespace_lock held and the function calls ZFS vnode 482 * operations to write the cache file and thus z_teardown_lock is 483 * acquired after spa_namespace_lock. 484 */ 485 ds = dmu_objset_ds(os); 486 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 487 488 /* 489 * nbmand is a special property. It can only be changed at 490 * mount time. 491 * 492 * This is weird, but it is documented to only be changeable 493 * at mount time. 494 */ 495 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 496 nbmand = B_FALSE; 497 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 498 nbmand = B_TRUE; 499 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) { 500 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 501 return (error); 502 } 503 504 /* 505 * Register property callbacks. 506 * 507 * It would probably be fine to just check for i/o error from 508 * the first prop_register(), but I guess I like to go 509 * overboard... 510 */ 511 error = dsl_prop_register(ds, 512 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 513 error = error ? error : dsl_prop_register(ds, 514 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 515 error = error ? error : dsl_prop_register(ds, 516 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 517 error = error ? error : dsl_prop_register(ds, 518 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 519#ifdef illumos 520 error = error ? error : dsl_prop_register(ds, 521 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 522#endif 523 error = error ? error : dsl_prop_register(ds, 524 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 525 error = error ? error : dsl_prop_register(ds, 526 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 527 error = error ? error : dsl_prop_register(ds, 528 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 529 error = error ? error : dsl_prop_register(ds, 530 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 531 error = error ? error : dsl_prop_register(ds, 532 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 533 zfsvfs); 534 error = error ? error : dsl_prop_register(ds, 535 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); 536 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 537 if (error) 538 goto unregister; 539 540 /* 541 * Invoke our callbacks to restore temporary mount options. 542 */ 543 if (do_readonly) 544 readonly_changed_cb(zfsvfs, readonly); 545 if (do_setuid) 546 setuid_changed_cb(zfsvfs, setuid); 547 if (do_exec) 548 exec_changed_cb(zfsvfs, exec); 549 if (do_xattr) 550 xattr_changed_cb(zfsvfs, xattr); 551 if (do_atime) 552 atime_changed_cb(zfsvfs, atime); 553 554 nbmand_changed_cb(zfsvfs, nbmand); 555 556 return (0); 557 558unregister: 559 dsl_prop_unregister_all(ds, zfsvfs); 560 return (error); 561} 562 563static int 564zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, 565 uint64_t *userp, uint64_t *groupp) 566{ 567 /* 568 * Is it a valid type of object to track? 569 */ 570 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 571 return (SET_ERROR(ENOENT)); 572 573 /* 574 * If we have a NULL data pointer 575 * then assume the id's aren't changing and 576 * return EEXIST to the dmu to let it know to 577 * use the same ids 578 */ 579 if (data == NULL) 580 return (SET_ERROR(EEXIST)); 581 582 if (bonustype == DMU_OT_ZNODE) { 583 znode_phys_t *znp = data; 584 *userp = znp->zp_uid; 585 *groupp = znp->zp_gid; 586 } else { 587 int hdrsize; 588 sa_hdr_phys_t *sap = data; 589 sa_hdr_phys_t sa = *sap; 590 boolean_t swap = B_FALSE; 591 592 ASSERT(bonustype == DMU_OT_SA); 593 594 if (sa.sa_magic == 0) { 595 /* 596 * This should only happen for newly created 597 * files that haven't had the znode data filled 598 * in yet. 599 */ 600 *userp = 0; 601 *groupp = 0; 602 return (0); 603 } 604 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { 605 sa.sa_magic = SA_MAGIC; 606 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); 607 swap = B_TRUE; 608 } else { 609 VERIFY3U(sa.sa_magic, ==, SA_MAGIC); 610 } 611 612 hdrsize = sa_hdrsize(&sa); 613 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); 614 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + 615 SA_UID_OFFSET)); 616 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + 617 SA_GID_OFFSET)); 618 if (swap) { 619 *userp = BSWAP_64(*userp); 620 *groupp = BSWAP_64(*groupp); 621 } 622 } 623 return (0); 624} 625 626static void 627fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, 628 char *domainbuf, int buflen, uid_t *ridp) 629{ 630 uint64_t fuid; 631 const char *domain; 632 633 fuid = strtonum(fuidstr, NULL); 634 635 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); 636 if (domain) 637 (void) strlcpy(domainbuf, domain, buflen); 638 else 639 domainbuf[0] = '\0'; 640 *ridp = FUID_RID(fuid); 641} 642 643static uint64_t 644zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) 645{ 646 switch (type) { 647 case ZFS_PROP_USERUSED: 648 return (DMU_USERUSED_OBJECT); 649 case ZFS_PROP_GROUPUSED: 650 return (DMU_GROUPUSED_OBJECT); 651 case ZFS_PROP_USERQUOTA: 652 return (zfsvfs->z_userquota_obj); 653 case ZFS_PROP_GROUPQUOTA: 654 return (zfsvfs->z_groupquota_obj); 655 } 656 return (0); 657} 658 659int 660zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 661 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) 662{ 663 int error; 664 zap_cursor_t zc; 665 zap_attribute_t za; 666 zfs_useracct_t *buf = vbuf; 667 uint64_t obj; 668 669 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 670 return (SET_ERROR(ENOTSUP)); 671 672 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 673 if (obj == 0) { 674 *bufsizep = 0; 675 return (0); 676 } 677 678 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); 679 (error = zap_cursor_retrieve(&zc, &za)) == 0; 680 zap_cursor_advance(&zc)) { 681 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > 682 *bufsizep) 683 break; 684 685 fuidstr_to_sid(zfsvfs, za.za_name, 686 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); 687 688 buf->zu_space = za.za_first_integer; 689 buf++; 690 } 691 if (error == ENOENT) 692 error = 0; 693 694 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); 695 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; 696 *cookiep = zap_cursor_serialize(&zc); 697 zap_cursor_fini(&zc); 698 return (error); 699} 700 701/* 702 * buf must be big enough (eg, 32 bytes) 703 */ 704static int 705id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, 706 char *buf, boolean_t addok) 707{ 708 uint64_t fuid; 709 int domainid = 0; 710 711 if (domain && domain[0]) { 712 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); 713 if (domainid == -1) 714 return (SET_ERROR(ENOENT)); 715 } 716 fuid = FUID_ENCODE(domainid, rid); 717 (void) sprintf(buf, "%llx", (longlong_t)fuid); 718 return (0); 719} 720 721int 722zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 723 const char *domain, uint64_t rid, uint64_t *valp) 724{ 725 char buf[32]; 726 int err; 727 uint64_t obj; 728 729 *valp = 0; 730 731 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 732 return (SET_ERROR(ENOTSUP)); 733 734 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 735 if (obj == 0) 736 return (0); 737 738 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); 739 if (err) 740 return (err); 741 742 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); 743 if (err == ENOENT) 744 err = 0; 745 return (err); 746} 747 748int 749zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 750 const char *domain, uint64_t rid, uint64_t quota) 751{ 752 char buf[32]; 753 int err; 754 dmu_tx_t *tx; 755 uint64_t *objp; 756 boolean_t fuid_dirtied; 757 758 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) 759 return (SET_ERROR(EINVAL)); 760 761 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) 762 return (SET_ERROR(ENOTSUP)); 763 764 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : 765 &zfsvfs->z_groupquota_obj; 766 767 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); 768 if (err) 769 return (err); 770 fuid_dirtied = zfsvfs->z_fuid_dirty; 771 772 tx = dmu_tx_create(zfsvfs->z_os); 773 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); 774 if (*objp == 0) { 775 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 776 zfs_userquota_prop_prefixes[type]); 777 } 778 if (fuid_dirtied) 779 zfs_fuid_txhold(zfsvfs, tx); 780 err = dmu_tx_assign(tx, TXG_WAIT); 781 if (err) { 782 dmu_tx_abort(tx); 783 return (err); 784 } 785 786 mutex_enter(&zfsvfs->z_lock); 787 if (*objp == 0) { 788 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, 789 DMU_OT_NONE, 0, tx); 790 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 791 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); 792 } 793 mutex_exit(&zfsvfs->z_lock); 794 795 if (quota == 0) { 796 err = zap_remove(zfsvfs->z_os, *objp, buf, tx); 797 if (err == ENOENT) 798 err = 0; 799 } else { 800 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); 801 } 802 ASSERT(err == 0); 803 if (fuid_dirtied) 804 zfs_fuid_sync(zfsvfs, tx); 805 dmu_tx_commit(tx); 806 return (err); 807} 808 809boolean_t 810zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) 811{ 812 char buf[32]; 813 uint64_t used, quota, usedobj, quotaobj; 814 int err; 815 816 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; 817 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 818 819 if (quotaobj == 0 || zfsvfs->z_replay) 820 return (B_FALSE); 821 822 (void) sprintf(buf, "%llx", (longlong_t)fuid); 823 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); 824 if (err != 0) 825 return (B_FALSE); 826 827 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); 828 if (err != 0) 829 return (B_FALSE); 830 return (used >= quota); 831} 832 833boolean_t 834zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) 835{ 836 uint64_t fuid; 837 uint64_t quotaobj; 838 839 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 840 841 fuid = isgroup ? zp->z_gid : zp->z_uid; 842 843 if (quotaobj == 0 || zfsvfs->z_replay) 844 return (B_FALSE); 845 846 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); 847} 848 849int 850zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 851{ 852 objset_t *os; 853 zfsvfs_t *zfsvfs; 854 uint64_t zval; 855 int i, error; 856 uint64_t sa_obj; 857 858 /* 859 * XXX: Fix struct statfs so this isn't necessary! 860 * 861 * The 'osname' is used as the filesystem's special node, which means 862 * it must fit in statfs.f_mntfromname, or else it can't be 863 * enumerated, so libzfs_mnttab_find() returns NULL, which causes 864 * 'zfs unmount' to think it's not mounted when it is. 865 */ 866 if (strlen(osname) >= MNAMELEN) 867 return (SET_ERROR(ENAMETOOLONG)); 868 869 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 870 871 /* 872 * We claim to always be readonly so we can open snapshots; 873 * other ZPL code will prevent us from writing to snapshots. 874 */ 875 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 876 if (error) { 877 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 878 return (error); 879 } 880 881 /* 882 * Initialize the zfs-specific filesystem structure. 883 * Should probably make this a kmem cache, shuffle fields, 884 * and just bzero up to z_hold_mtx[]. 885 */ 886 zfsvfs->z_vfs = NULL; 887 zfsvfs->z_parent = zfsvfs; 888 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 889 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 890 zfsvfs->z_os = os; 891 892 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 893 if (error) { 894 goto out; 895 } else if (zfsvfs->z_version > 896 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 897 (void) printf("Can't mount a version %lld file system " 898 "on a version %lld pool\n. Pool must be upgraded to mount " 899 "this file system.", (u_longlong_t)zfsvfs->z_version, 900 (u_longlong_t)spa_version(dmu_objset_spa(os))); 901 error = SET_ERROR(ENOTSUP); 902 goto out; 903 } 904 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) 905 goto out; 906 zfsvfs->z_norm = (int)zval; 907 908 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) 909 goto out; 910 zfsvfs->z_utf8 = (zval != 0); 911 912 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) 913 goto out; 914 zfsvfs->z_case = (uint_t)zval; 915 916 /* 917 * Fold case on file systems that are always or sometimes case 918 * insensitive. 919 */ 920 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 921 zfsvfs->z_case == ZFS_CASE_MIXED) 922 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 923 924 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 925 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 926 927 if (zfsvfs->z_use_sa) { 928 /* should either have both of these objects or none */ 929 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 930 &sa_obj); 931 if (error) 932 goto out; 933 } else { 934 /* 935 * Pre SA versions file systems should never touch 936 * either the attribute registration or layout objects. 937 */ 938 sa_obj = 0; 939 } 940 941 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 942 &zfsvfs->z_attr_table); 943 if (error) 944 goto out; 945 946 if (zfsvfs->z_version >= ZPL_VERSION_SA) 947 sa_register_update_callback(os, zfs_sa_upgrade); 948 949 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 950 &zfsvfs->z_root); 951 if (error) 952 goto out; 953 ASSERT(zfsvfs->z_root != 0); 954 955 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 956 &zfsvfs->z_unlinkedobj); 957 if (error) 958 goto out; 959 960 error = zap_lookup(os, MASTER_NODE_OBJ, 961 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 962 8, 1, &zfsvfs->z_userquota_obj); 963 if (error && error != ENOENT) 964 goto out; 965 966 error = zap_lookup(os, MASTER_NODE_OBJ, 967 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 968 8, 1, &zfsvfs->z_groupquota_obj); 969 if (error && error != ENOENT) 970 goto out; 971 972 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 973 &zfsvfs->z_fuid_obj); 974 if (error && error != ENOENT) 975 goto out; 976 977 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 978 &zfsvfs->z_shares_dir); 979 if (error && error != ENOENT) 980 goto out; 981 982 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 983 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 984 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 985 offsetof(znode_t, z_link_node)); 986 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE); 987 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 988 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 989 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 990 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 991 992 *zfvp = zfsvfs; 993 return (0); 994 995out: 996 dmu_objset_disown(os, zfsvfs); 997 *zfvp = NULL; 998 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 999 return (error); 1000} 1001 1002static int 1003zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 1004{ 1005 int error; 1006 1007 error = zfs_register_callbacks(zfsvfs->z_vfs); 1008 if (error) 1009 return (error); 1010 1011 /* 1012 * Set the objset user_ptr to track its zfsvfs. 1013 */ 1014 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1015 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1016 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1017 1018 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 1019 1020 /* 1021 * If we are not mounting (ie: online recv), then we don't 1022 * have to worry about replaying the log as we blocked all 1023 * operations out since we closed the ZIL. 1024 */ 1025 if (mounting) { 1026 boolean_t readonly; 1027 1028 /* 1029 * During replay we remove the read only flag to 1030 * allow replays to succeed. 1031 */ 1032 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1033 if (readonly != 0) 1034 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1035 else 1036 zfs_unlinked_drain(zfsvfs); 1037 1038 /* 1039 * Parse and replay the intent log. 1040 * 1041 * Because of ziltest, this must be done after 1042 * zfs_unlinked_drain(). (Further note: ziltest 1043 * doesn't use readonly mounts, where 1044 * zfs_unlinked_drain() isn't called.) This is because 1045 * ziltest causes spa_sync() to think it's committed, 1046 * but actually it is not, so the intent log contains 1047 * many txg's worth of changes. 1048 * 1049 * In particular, if object N is in the unlinked set in 1050 * the last txg to actually sync, then it could be 1051 * actually freed in a later txg and then reallocated 1052 * in a yet later txg. This would write a "create 1053 * object N" record to the intent log. Normally, this 1054 * would be fine because the spa_sync() would have 1055 * written out the fact that object N is free, before 1056 * we could write the "create object N" intent log 1057 * record. 1058 * 1059 * But when we are in ziltest mode, we advance the "open 1060 * txg" without actually spa_sync()-ing the changes to 1061 * disk. So we would see that object N is still 1062 * allocated and in the unlinked set, and there is an 1063 * intent log record saying to allocate it. 1064 */ 1065 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1066 if (zil_replay_disable) { 1067 zil_destroy(zfsvfs->z_log, B_FALSE); 1068 } else { 1069 zfsvfs->z_replay = B_TRUE; 1070 zil_replay(zfsvfs->z_os, zfsvfs, 1071 zfs_replay_vector); 1072 zfsvfs->z_replay = B_FALSE; 1073 } 1074 } 1075 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1076 } 1077 1078 return (0); 1079} 1080 1081extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1082 1083void 1084zfsvfs_free(zfsvfs_t *zfsvfs) 1085{ 1086 int i; 1087 1088 /* 1089 * This is a barrier to prevent the filesystem from going away in 1090 * zfs_znode_move() until we can safely ensure that the filesystem is 1091 * not unmounted. We consider the filesystem valid before the barrier 1092 * and invalid after the barrier. 1093 */ 1094 rw_enter(&zfsvfs_lock, RW_READER); 1095 rw_exit(&zfsvfs_lock); 1096 1097 zfs_fuid_destroy(zfsvfs); 1098 1099 mutex_destroy(&zfsvfs->z_znodes_lock); 1100 mutex_destroy(&zfsvfs->z_lock); 1101 list_destroy(&zfsvfs->z_all_znodes); 1102 rrm_destroy(&zfsvfs->z_teardown_lock); 1103 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1104 rw_destroy(&zfsvfs->z_fuid_lock); 1105 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1106 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1107 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1108} 1109 1110static void 1111zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1112{ 1113 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1114 if (zfsvfs->z_vfs) { 1115 if (zfsvfs->z_use_fuids) { 1116 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1117 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1118 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1119 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1120 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1121 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1122 } else { 1123 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1124 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1125 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1126 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1127 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1128 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1129 } 1130 } 1131 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1132} 1133 1134static int 1135zfs_domount(vfs_t *vfsp, char *osname) 1136{ 1137 uint64_t recordsize, fsid_guid; 1138 int error = 0; 1139 zfsvfs_t *zfsvfs; 1140 vnode_t *vp; 1141 1142 ASSERT(vfsp); 1143 ASSERT(osname); 1144 1145 error = zfsvfs_create(osname, &zfsvfs); 1146 if (error) 1147 return (error); 1148 zfsvfs->z_vfs = vfsp; 1149 1150#ifdef illumos 1151 /* Initialize the generic filesystem structure. */ 1152 vfsp->vfs_bcount = 0; 1153 vfsp->vfs_data = NULL; 1154 1155 if (zfs_create_unique_device(&mount_dev) == -1) { 1156 error = SET_ERROR(ENODEV); 1157 goto out; 1158 } 1159 ASSERT(vfs_devismounted(mount_dev) == 0); 1160#endif 1161 1162 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1163 NULL)) 1164 goto out; 1165 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1166 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1167 1168 vfsp->vfs_data = zfsvfs; 1169 vfsp->mnt_flag |= MNT_LOCAL; 1170 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1171 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1172 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1173 1174 /* 1175 * The fsid is 64 bits, composed of an 8-bit fs type, which 1176 * separates our fsid from any other filesystem types, and a 1177 * 56-bit objset unique ID. The objset unique ID is unique to 1178 * all objsets open on this system, provided by unique_create(). 1179 * The 8-bit fs type must be put in the low bits of fsid[1] 1180 * because that's where other Solaris filesystems put it. 1181 */ 1182 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1183 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1184 vfsp->vfs_fsid.val[0] = fsid_guid; 1185 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1186 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1187 1188 /* 1189 * Set features for file system. 1190 */ 1191 zfs_set_fuid_feature(zfsvfs); 1192 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1193 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1194 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1195 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1196 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1197 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1198 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1199 } 1200 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1201 1202 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1203 uint64_t pval; 1204 1205 atime_changed_cb(zfsvfs, B_FALSE); 1206 readonly_changed_cb(zfsvfs, B_TRUE); 1207 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1208 goto out; 1209 xattr_changed_cb(zfsvfs, pval); 1210 zfsvfs->z_issnap = B_TRUE; 1211 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1212 1213 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1214 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1215 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1216 } else { 1217 error = zfsvfs_setup(zfsvfs, B_TRUE); 1218 } 1219 1220 vfs_mountedfrom(vfsp, osname); 1221 1222 if (!zfsvfs->z_issnap) 1223 zfsctl_create(zfsvfs); 1224out: 1225 if (error) { 1226 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1227 zfsvfs_free(zfsvfs); 1228 } else { 1229 atomic_inc_32(&zfs_active_fs_count); 1230 } 1231 1232 return (error); 1233} 1234 1235void 1236zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1237{ 1238 objset_t *os = zfsvfs->z_os; 1239 1240 if (!dmu_objset_is_snapshot(os)) 1241 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 1242} 1243 1244#ifdef SECLABEL 1245/* 1246 * Convert a decimal digit string to a uint64_t integer. 1247 */ 1248static int 1249str_to_uint64(char *str, uint64_t *objnum) 1250{ 1251 uint64_t num = 0; 1252 1253 while (*str) { 1254 if (*str < '0' || *str > '9') 1255 return (SET_ERROR(EINVAL)); 1256 1257 num = num*10 + *str++ - '0'; 1258 } 1259 1260 *objnum = num; 1261 return (0); 1262} 1263 1264/* 1265 * The boot path passed from the boot loader is in the form of 1266 * "rootpool-name/root-filesystem-object-number'. Convert this 1267 * string to a dataset name: "rootpool-name/root-filesystem-name". 1268 */ 1269static int 1270zfs_parse_bootfs(char *bpath, char *outpath) 1271{ 1272 char *slashp; 1273 uint64_t objnum; 1274 int error; 1275 1276 if (*bpath == 0 || *bpath == '/') 1277 return (SET_ERROR(EINVAL)); 1278 1279 (void) strcpy(outpath, bpath); 1280 1281 slashp = strchr(bpath, '/'); 1282 1283 /* if no '/', just return the pool name */ 1284 if (slashp == NULL) { 1285 return (0); 1286 } 1287 1288 /* if not a number, just return the root dataset name */ 1289 if (str_to_uint64(slashp+1, &objnum)) { 1290 return (0); 1291 } 1292 1293 *slashp = '\0'; 1294 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1295 *slashp = '/'; 1296 1297 return (error); 1298} 1299 1300/* 1301 * Check that the hex label string is appropriate for the dataset being 1302 * mounted into the global_zone proper. 1303 * 1304 * Return an error if the hex label string is not default or 1305 * admin_low/admin_high. For admin_low labels, the corresponding 1306 * dataset must be readonly. 1307 */ 1308int 1309zfs_check_global_label(const char *dsname, const char *hexsl) 1310{ 1311 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1312 return (0); 1313 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1314 return (0); 1315 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1316 /* must be readonly */ 1317 uint64_t rdonly; 1318 1319 if (dsl_prop_get_integer(dsname, 1320 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1321 return (SET_ERROR(EACCES)); 1322 return (rdonly ? 0 : EACCES); 1323 } 1324 return (SET_ERROR(EACCES)); 1325} 1326 1327/* 1328 * Determine whether the mount is allowed according to MAC check. 1329 * by comparing (where appropriate) label of the dataset against 1330 * the label of the zone being mounted into. If the dataset has 1331 * no label, create one. 1332 * 1333 * Returns 0 if access allowed, error otherwise (e.g. EACCES) 1334 */ 1335static int 1336zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1337{ 1338 int error, retv; 1339 zone_t *mntzone = NULL; 1340 ts_label_t *mnt_tsl; 1341 bslabel_t *mnt_sl; 1342 bslabel_t ds_sl; 1343 char ds_hexsl[MAXNAMELEN]; 1344 1345 retv = EACCES; /* assume the worst */ 1346 1347 /* 1348 * Start by getting the dataset label if it exists. 1349 */ 1350 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1351 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1352 if (error) 1353 return (SET_ERROR(EACCES)); 1354 1355 /* 1356 * If labeling is NOT enabled, then disallow the mount of datasets 1357 * which have a non-default label already. No other label checks 1358 * are needed. 1359 */ 1360 if (!is_system_labeled()) { 1361 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1362 return (0); 1363 return (SET_ERROR(EACCES)); 1364 } 1365 1366 /* 1367 * Get the label of the mountpoint. If mounting into the global 1368 * zone (i.e. mountpoint is not within an active zone and the 1369 * zoned property is off), the label must be default or 1370 * admin_low/admin_high only; no other checks are needed. 1371 */ 1372 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1373 if (mntzone->zone_id == GLOBAL_ZONEID) { 1374 uint64_t zoned; 1375 1376 zone_rele(mntzone); 1377 1378 if (dsl_prop_get_integer(osname, 1379 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1380 return (SET_ERROR(EACCES)); 1381 if (!zoned) 1382 return (zfs_check_global_label(osname, ds_hexsl)); 1383 else 1384 /* 1385 * This is the case of a zone dataset being mounted 1386 * initially, before the zone has been fully created; 1387 * allow this mount into global zone. 1388 */ 1389 return (0); 1390 } 1391 1392 mnt_tsl = mntzone->zone_slabel; 1393 ASSERT(mnt_tsl != NULL); 1394 label_hold(mnt_tsl); 1395 mnt_sl = label2bslabel(mnt_tsl); 1396 1397 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1398 /* 1399 * The dataset doesn't have a real label, so fabricate one. 1400 */ 1401 char *str = NULL; 1402 1403 if (l_to_str_internal(mnt_sl, &str) == 0 && 1404 dsl_prop_set_string(osname, 1405 zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1406 ZPROP_SRC_LOCAL, str) == 0) 1407 retv = 0; 1408 if (str != NULL) 1409 kmem_free(str, strlen(str) + 1); 1410 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1411 /* 1412 * Now compare labels to complete the MAC check. If the 1413 * labels are equal then allow access. If the mountpoint 1414 * label dominates the dataset label, allow readonly access. 1415 * Otherwise, access is denied. 1416 */ 1417 if (blequal(mnt_sl, &ds_sl)) 1418 retv = 0; 1419 else if (bldominates(mnt_sl, &ds_sl)) { 1420 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1421 retv = 0; 1422 } 1423 } 1424 1425 label_rele(mnt_tsl); 1426 zone_rele(mntzone); 1427 return (retv); 1428} 1429#endif /* SECLABEL */ 1430 1431#ifdef OPENSOLARIS_MOUNTROOT 1432static int 1433zfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1434{ 1435 int error = 0; 1436 static int zfsrootdone = 0; 1437 zfsvfs_t *zfsvfs = NULL; 1438 znode_t *zp = NULL; 1439 vnode_t *vp = NULL; 1440 char *zfs_bootfs; 1441 char *zfs_devid; 1442 1443 ASSERT(vfsp); 1444 1445 /* 1446 * The filesystem that we mount as root is defined in the 1447 * boot property "zfs-bootfs" with a format of 1448 * "poolname/root-dataset-objnum". 1449 */ 1450 if (why == ROOT_INIT) { 1451 if (zfsrootdone++) 1452 return (SET_ERROR(EBUSY)); 1453 /* 1454 * the process of doing a spa_load will require the 1455 * clock to be set before we could (for example) do 1456 * something better by looking at the timestamp on 1457 * an uberblock, so just set it to -1. 1458 */ 1459 clkset(-1); 1460 1461 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1462 cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1463 "bootfs name"); 1464 return (SET_ERROR(EINVAL)); 1465 } 1466 zfs_devid = spa_get_bootprop("diskdevid"); 1467 error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1468 if (zfs_devid) 1469 spa_free_bootprop(zfs_devid); 1470 if (error) { 1471 spa_free_bootprop(zfs_bootfs); 1472 cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1473 error); 1474 return (error); 1475 } 1476 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1477 spa_free_bootprop(zfs_bootfs); 1478 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1479 error); 1480 return (error); 1481 } 1482 1483 spa_free_bootprop(zfs_bootfs); 1484 1485 if (error = vfs_lock(vfsp)) 1486 return (error); 1487 1488 if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1489 cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1490 goto out; 1491 } 1492 1493 zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1494 ASSERT(zfsvfs); 1495 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1496 cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1497 goto out; 1498 } 1499 1500 vp = ZTOV(zp); 1501 mutex_enter(&vp->v_lock); 1502 vp->v_flag |= VROOT; 1503 mutex_exit(&vp->v_lock); 1504 rootvp = vp; 1505 1506 /* 1507 * Leave rootvp held. The root file system is never unmounted. 1508 */ 1509 1510 vfs_add((struct vnode *)0, vfsp, 1511 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1512out: 1513 vfs_unlock(vfsp); 1514 return (error); 1515 } else if (why == ROOT_REMOUNT) { 1516 readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1517 vfsp->vfs_flag |= VFS_REMOUNT; 1518 1519 /* refresh mount options */ 1520 zfs_unregister_callbacks(vfsp->vfs_data); 1521 return (zfs_register_callbacks(vfsp)); 1522 1523 } else if (why == ROOT_UNMOUNT) { 1524 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1525 (void) zfs_sync(vfsp, 0, 0); 1526 return (0); 1527 } 1528 1529 /* 1530 * if "why" is equal to anything else other than ROOT_INIT, 1531 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1532 */ 1533 return (SET_ERROR(ENOTSUP)); 1534} 1535#endif /* OPENSOLARIS_MOUNTROOT */ 1536 1537static int 1538getpoolname(const char *osname, char *poolname) 1539{ 1540 char *p; 1541 1542 p = strchr(osname, '/'); 1543 if (p == NULL) { 1544 if (strlen(osname) >= MAXNAMELEN) 1545 return (ENAMETOOLONG); 1546 (void) strcpy(poolname, osname); 1547 } else { 1548 if (p - osname >= MAXNAMELEN) 1549 return (ENAMETOOLONG); 1550 (void) strncpy(poolname, osname, p - osname); 1551 poolname[p - osname] = '\0'; 1552 } 1553 return (0); 1554} 1555 1556/*ARGSUSED*/ 1557static int 1558zfs_mount(vfs_t *vfsp) 1559{ 1560 kthread_t *td = curthread; 1561 vnode_t *mvp = vfsp->mnt_vnodecovered; 1562 cred_t *cr = td->td_ucred; 1563 char *osname; 1564 int error = 0; 1565 int canwrite; 1566 1567#ifdef illumos 1568 if (mvp->v_type != VDIR) 1569 return (SET_ERROR(ENOTDIR)); 1570 1571 mutex_enter(&mvp->v_lock); 1572 if ((uap->flags & MS_REMOUNT) == 0 && 1573 (uap->flags & MS_OVERLAY) == 0 && 1574 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 1575 mutex_exit(&mvp->v_lock); 1576 return (SET_ERROR(EBUSY)); 1577 } 1578 mutex_exit(&mvp->v_lock); 1579 1580 /* 1581 * ZFS does not support passing unparsed data in via MS_DATA. 1582 * Users should use the MS_OPTIONSTR interface; this means 1583 * that all option parsing is already done and the options struct 1584 * can be interrogated. 1585 */ 1586 if ((uap->flags & MS_DATA) && uap->datalen > 0) 1587#else /* !illumos */ 1588 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1589 return (SET_ERROR(EPERM)); 1590 1591 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1592 return (SET_ERROR(EINVAL)); 1593#endif /* illumos */ 1594 1595 /* 1596 * If full-owner-access is enabled and delegated administration is 1597 * turned on, we must set nosuid. 1598 */ 1599 if (zfs_super_owner && 1600 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1601 secpolicy_fs_mount_clearopts(cr, vfsp); 1602 } 1603 1604 /* 1605 * Check for mount privilege? 1606 * 1607 * If we don't have privilege then see if 1608 * we have local permission to allow it 1609 */ 1610 error = secpolicy_fs_mount(cr, mvp, vfsp); 1611 if (error) { 1612 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1613 goto out; 1614 1615 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1616 vattr_t vattr; 1617 1618 /* 1619 * Make sure user is the owner of the mount point 1620 * or has sufficient privileges. 1621 */ 1622 1623 vattr.va_mask = AT_UID; 1624 1625 vn_lock(mvp, LK_SHARED | LK_RETRY); 1626 if (VOP_GETATTR(mvp, &vattr, cr)) { 1627 VOP_UNLOCK(mvp, 0); 1628 goto out; 1629 } 1630 1631 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1632 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1633 VOP_UNLOCK(mvp, 0); 1634 goto out; 1635 } 1636 VOP_UNLOCK(mvp, 0); 1637 } 1638 1639 secpolicy_fs_mount_clearopts(cr, vfsp); 1640 } 1641 1642 /* 1643 * Refuse to mount a filesystem if we are in a local zone and the 1644 * dataset is not visible. 1645 */ 1646 if (!INGLOBALZONE(curthread) && 1647 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1648 error = SET_ERROR(EPERM); 1649 goto out; 1650 } 1651 1652#ifdef SECLABEL 1653 error = zfs_mount_label_policy(vfsp, osname); 1654 if (error) 1655 goto out; 1656#endif 1657 1658 vfsp->vfs_flag |= MNT_NFS4ACLS; 1659 1660 /* 1661 * When doing a remount, we simply refresh our temporary properties 1662 * according to those options set in the current VFS options. 1663 */ 1664 if (vfsp->vfs_flag & MS_REMOUNT) { 1665 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1666 1667 /* 1668 * Refresh mount options with z_teardown_lock blocking I/O while 1669 * the filesystem is in an inconsistent state. 1670 * The lock also serializes this code with filesystem 1671 * manipulations between entry to zfs_suspend_fs() and return 1672 * from zfs_resume_fs(). 1673 */ 1674 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1675 zfs_unregister_callbacks(zfsvfs); 1676 error = zfs_register_callbacks(vfsp); 1677 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1678 goto out; 1679 } 1680 1681 /* Initial root mount: try hard to import the requested root pool. */ 1682 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1683 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1684 char pname[MAXNAMELEN]; 1685 1686 error = getpoolname(osname, pname); 1687 if (error == 0) 1688 error = spa_import_rootpool(pname); 1689 if (error) 1690 goto out; 1691 } 1692 DROP_GIANT(); 1693 error = zfs_domount(vfsp, osname); 1694 PICKUP_GIANT(); 1695 1696#ifdef illumos 1697 /* 1698 * Add an extra VFS_HOLD on our parent vfs so that it can't 1699 * disappear due to a forced unmount. 1700 */ 1701 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1702 VFS_HOLD(mvp->v_vfsp); 1703#endif 1704 1705out: 1706 return (error); 1707} 1708 1709static int 1710zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1711{ 1712 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1713 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1714 1715 statp->f_version = STATFS_VERSION; 1716 1717 ZFS_ENTER(zfsvfs); 1718 1719 dmu_objset_space(zfsvfs->z_os, 1720 &refdbytes, &availbytes, &usedobjs, &availobjs); 1721 1722 /* 1723 * The underlying storage pool actually uses multiple block sizes. 1724 * We report the fragsize as the smallest block size we support, 1725 * and we report our blocksize as the filesystem's maximum blocksize. 1726 */ 1727 statp->f_bsize = SPA_MINBLOCKSIZE; 1728 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1729 1730 /* 1731 * The following report "total" blocks of various kinds in the 1732 * file system, but reported in terms of f_frsize - the 1733 * "fragment" size. 1734 */ 1735 1736 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1737 statp->f_bfree = availbytes / statp->f_bsize; 1738 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1739 1740 /* 1741 * statvfs() should really be called statufs(), because it assumes 1742 * static metadata. ZFS doesn't preallocate files, so the best 1743 * we can do is report the max that could possibly fit in f_files, 1744 * and that minus the number actually used in f_ffree. 1745 * For f_ffree, report the smaller of the number of object available 1746 * and the number of blocks (each object will take at least a block). 1747 */ 1748 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1749 statp->f_files = statp->f_ffree + usedobjs; 1750 1751 /* 1752 * We're a zfs filesystem. 1753 */ 1754 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1755 1756 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1757 sizeof(statp->f_mntfromname)); 1758 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1759 sizeof(statp->f_mntonname)); 1760 1761 statp->f_namemax = ZFS_MAXNAMELEN; 1762 1763 ZFS_EXIT(zfsvfs); 1764 return (0); 1765} 1766 1767static int 1768zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1769{ 1770 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1771 znode_t *rootzp; 1772 int error; 1773 1774 ZFS_ENTER(zfsvfs); 1775 1776 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1777 if (error == 0) 1778 *vpp = ZTOV(rootzp); 1779 1780 ZFS_EXIT(zfsvfs); 1781 1782 if (error == 0) { 1783 error = vn_lock(*vpp, flags); 1784 if (error == 0) 1785 (*vpp)->v_vflag |= VV_ROOT; 1786 } 1787 if (error != 0) 1788 *vpp = NULL; 1789 1790 return (error); 1791} 1792 1793/* 1794 * Teardown the zfsvfs::z_os. 1795 * 1796 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1797 * and 'z_teardown_inactive_lock' held. 1798 */ 1799static int 1800zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1801{ 1802 znode_t *zp; 1803 1804 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1805 1806 if (!unmounting) { 1807 /* 1808 * We purge the parent filesystem's vfsp as the parent 1809 * filesystem and all of its snapshots have their vnode's 1810 * v_vfsp set to the parent's filesystem's vfsp. Note, 1811 * 'z_parent' is self referential for non-snapshots. 1812 */ 1813 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1814#ifdef FREEBSD_NAMECACHE 1815 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1816#endif 1817 } 1818 1819 /* 1820 * Close the zil. NB: Can't close the zil while zfs_inactive 1821 * threads are blocked as zil_close can call zfs_inactive. 1822 */ 1823 if (zfsvfs->z_log) { 1824 zil_close(zfsvfs->z_log); 1825 zfsvfs->z_log = NULL; 1826 } 1827 1828 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1829 1830 /* 1831 * If we are not unmounting (ie: online recv) and someone already 1832 * unmounted this file system while we were doing the switcheroo, 1833 * or a reopen of z_os failed then just bail out now. 1834 */ 1835 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1836 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1837 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1838 return (SET_ERROR(EIO)); 1839 } 1840 1841 /* 1842 * At this point there are no vops active, and any new vops will 1843 * fail with EIO since we have z_teardown_lock for writer (only 1844 * relavent for forced unmount). 1845 * 1846 * Release all holds on dbufs. 1847 */ 1848 mutex_enter(&zfsvfs->z_znodes_lock); 1849 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1850 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1851 if (zp->z_sa_hdl) { 1852 ASSERT(ZTOV(zp)->v_count >= 0); 1853 zfs_znode_dmu_fini(zp); 1854 } 1855 mutex_exit(&zfsvfs->z_znodes_lock); 1856 1857 /* 1858 * If we are unmounting, set the unmounted flag and let new vops 1859 * unblock. zfs_inactive will have the unmounted behavior, and all 1860 * other vops will fail with EIO. 1861 */ 1862 if (unmounting) { 1863 zfsvfs->z_unmounted = B_TRUE; 1864 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1865 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1866 } 1867 1868 /* 1869 * z_os will be NULL if there was an error in attempting to reopen 1870 * zfsvfs, so just return as the properties had already been 1871 * unregistered and cached data had been evicted before. 1872 */ 1873 if (zfsvfs->z_os == NULL) 1874 return (0); 1875 1876 /* 1877 * Unregister properties. 1878 */ 1879 zfs_unregister_callbacks(zfsvfs); 1880 1881 /* 1882 * Evict cached data 1883 */ 1884 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1885 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1886 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1887 dmu_objset_evict_dbufs(zfsvfs->z_os); 1888 1889 return (0); 1890} 1891 1892/*ARGSUSED*/ 1893static int 1894zfs_umount(vfs_t *vfsp, int fflag) 1895{ 1896 kthread_t *td = curthread; 1897 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1898 objset_t *os; 1899 cred_t *cr = td->td_ucred; 1900 int ret; 1901 1902 ret = secpolicy_fs_unmount(cr, vfsp); 1903 if (ret) { 1904 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1905 ZFS_DELEG_PERM_MOUNT, cr)) 1906 return (ret); 1907 } 1908 1909 /* 1910 * We purge the parent filesystem's vfsp as the parent filesystem 1911 * and all of its snapshots have their vnode's v_vfsp set to the 1912 * parent's filesystem's vfsp. Note, 'z_parent' is self 1913 * referential for non-snapshots. 1914 */ 1915 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1916 1917 /* 1918 * Unmount any snapshots mounted under .zfs before unmounting the 1919 * dataset itself. 1920 */ 1921 if (zfsvfs->z_ctldir != NULL) { 1922 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1923 return (ret); 1924 ret = vflush(vfsp, 0, 0, td); 1925 ASSERT(ret == EBUSY); 1926 if (!(fflag & MS_FORCE)) { 1927 if (zfsvfs->z_ctldir->v_count > 1) 1928 return (EBUSY); 1929 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1930 } 1931 zfsctl_destroy(zfsvfs); 1932 ASSERT(zfsvfs->z_ctldir == NULL); 1933 } 1934 1935 if (fflag & MS_FORCE) { 1936 /* 1937 * Mark file system as unmounted before calling 1938 * vflush(FORCECLOSE). This way we ensure no future vnops 1939 * will be called and risk operating on DOOMED vnodes. 1940 */ 1941 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1942 zfsvfs->z_unmounted = B_TRUE; 1943 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1944 } 1945 1946 /* 1947 * Flush all the files. 1948 */ 1949 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1950 if (ret != 0) { 1951 if (!zfsvfs->z_issnap) { 1952 zfsctl_create(zfsvfs); 1953 ASSERT(zfsvfs->z_ctldir != NULL); 1954 } 1955 return (ret); 1956 } 1957 1958#ifdef illumos 1959 if (!(fflag & MS_FORCE)) { 1960 /* 1961 * Check the number of active vnodes in the file system. 1962 * Our count is maintained in the vfs structure, but the 1963 * number is off by 1 to indicate a hold on the vfs 1964 * structure itself. 1965 * 1966 * The '.zfs' directory maintains a reference of its 1967 * own, and any active references underneath are 1968 * reflected in the vnode count. 1969 */ 1970 if (zfsvfs->z_ctldir == NULL) { 1971 if (vfsp->vfs_count > 1) 1972 return (SET_ERROR(EBUSY)); 1973 } else { 1974 if (vfsp->vfs_count > 2 || 1975 zfsvfs->z_ctldir->v_count > 1) 1976 return (SET_ERROR(EBUSY)); 1977 } 1978 } 1979#endif 1980 1981 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1982 os = zfsvfs->z_os; 1983 1984 /* 1985 * z_os will be NULL if there was an error in 1986 * attempting to reopen zfsvfs. 1987 */ 1988 if (os != NULL) { 1989 /* 1990 * Unset the objset user_ptr. 1991 */ 1992 mutex_enter(&os->os_user_ptr_lock); 1993 dmu_objset_set_user(os, NULL); 1994 mutex_exit(&os->os_user_ptr_lock); 1995 1996 /* 1997 * Finally release the objset 1998 */ 1999 dmu_objset_disown(os, zfsvfs); 2000 } 2001 2002 /* 2003 * We can now safely destroy the '.zfs' directory node. 2004 */ 2005 if (zfsvfs->z_ctldir != NULL) 2006 zfsctl_destroy(zfsvfs); 2007 if (zfsvfs->z_issnap) { 2008 vnode_t *svp = vfsp->mnt_vnodecovered; 2009 2010 if (svp->v_count >= 2) 2011 VN_RELE(svp); 2012 } 2013 zfs_freevfs(vfsp); 2014 2015 return (0); 2016} 2017 2018static int 2019zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2020{ 2021 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2022 znode_t *zp; 2023 int err; 2024 2025 /* 2026 * zfs_zget() can't operate on virtual entries like .zfs/ or 2027 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2028 * This will make NFS to switch to LOOKUP instead of using VGET. 2029 */ 2030 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || 2031 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) 2032 return (EOPNOTSUPP); 2033 2034 ZFS_ENTER(zfsvfs); 2035 err = zfs_zget(zfsvfs, ino, &zp); 2036 if (err == 0 && zp->z_unlinked) { 2037 VN_RELE(ZTOV(zp)); 2038 err = EINVAL; 2039 } 2040 if (err == 0) 2041 *vpp = ZTOV(zp); 2042 ZFS_EXIT(zfsvfs); 2043 if (err == 0) 2044 err = vn_lock(*vpp, flags); 2045 if (err != 0) 2046 *vpp = NULL; 2047 return (err); 2048} 2049 2050static int 2051zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2052 struct ucred **credanonp, int *numsecflavors, int **secflavors) 2053{ 2054 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2055 2056 /* 2057 * If this is regular file system vfsp is the same as 2058 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2059 * zfsvfs->z_parent->z_vfs represents parent file system 2060 * which we have to use here, because only this file system 2061 * has mnt_export configured. 2062 */ 2063 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2064 credanonp, numsecflavors, secflavors)); 2065} 2066 2067CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2068CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2069 2070static int 2071zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2072{ 2073 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2074 znode_t *zp; 2075 uint64_t object = 0; 2076 uint64_t fid_gen = 0; 2077 uint64_t gen_mask; 2078 uint64_t zp_gen; 2079 int i, err; 2080 2081 *vpp = NULL; 2082 2083 ZFS_ENTER(zfsvfs); 2084 2085 /* 2086 * On FreeBSD we can get snapshot's mount point or its parent file 2087 * system mount point depending if snapshot is already mounted or not. 2088 */ 2089 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2090 zfid_long_t *zlfid = (zfid_long_t *)fidp; 2091 uint64_t objsetid = 0; 2092 uint64_t setgen = 0; 2093 2094 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2095 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2096 2097 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2098 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2099 2100 ZFS_EXIT(zfsvfs); 2101 2102 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2103 if (err) 2104 return (SET_ERROR(EINVAL)); 2105 ZFS_ENTER(zfsvfs); 2106 } 2107 2108 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2109 zfid_short_t *zfid = (zfid_short_t *)fidp; 2110 2111 for (i = 0; i < sizeof (zfid->zf_object); i++) 2112 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2113 2114 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2115 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2116 } else { 2117 ZFS_EXIT(zfsvfs); 2118 return (SET_ERROR(EINVAL)); 2119 } 2120 2121 /* 2122 * A zero fid_gen means we are in .zfs or the .zfs/snapshot 2123 * directory tree. If the object == zfsvfs->z_shares_dir, then 2124 * we are in the .zfs/shares directory tree. 2125 */ 2126 if ((fid_gen == 0 && 2127 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || 2128 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { 2129 *vpp = zfsvfs->z_ctldir; 2130 ASSERT(*vpp != NULL); 2131 if (object == ZFSCTL_INO_SNAPDIR) { 2132 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2133 0, NULL, NULL, NULL, NULL, NULL) == 0); 2134 } else if (object == zfsvfs->z_shares_dir) { 2135 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL, 2136 0, NULL, NULL, NULL, NULL, NULL) == 0); 2137 } else { 2138 VN_HOLD(*vpp); 2139 } 2140 ZFS_EXIT(zfsvfs); 2141 err = vn_lock(*vpp, flags); 2142 if (err != 0) 2143 *vpp = NULL; 2144 return (err); 2145 } 2146 2147 gen_mask = -1ULL >> (64 - 8 * i); 2148 2149 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2150 if (err = zfs_zget(zfsvfs, object, &zp)) { 2151 ZFS_EXIT(zfsvfs); 2152 return (err); 2153 } 2154 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2155 sizeof (uint64_t)); 2156 zp_gen = zp_gen & gen_mask; 2157 if (zp_gen == 0) 2158 zp_gen = 1; 2159 if (zp->z_unlinked || zp_gen != fid_gen) { 2160 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2161 VN_RELE(ZTOV(zp)); 2162 ZFS_EXIT(zfsvfs); 2163 return (SET_ERROR(EINVAL)); 2164 } 2165 2166 *vpp = ZTOV(zp); 2167 ZFS_EXIT(zfsvfs); 2168 err = vn_lock(*vpp, flags | LK_RETRY); 2169 if (err == 0) 2170 vnode_create_vobject(*vpp, zp->z_size, curthread); 2171 else 2172 *vpp = NULL; 2173 return (err); 2174} 2175 2176/* 2177 * Block out VOPs and close zfsvfs_t::z_os 2178 * 2179 * Note, if successful, then we return with the 'z_teardown_lock' and 2180 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 2181 * dataset and objset intact so that they can be atomically handed off during 2182 * a subsequent rollback or recv operation and the resume thereafter. 2183 */ 2184int 2185zfs_suspend_fs(zfsvfs_t *zfsvfs) 2186{ 2187 int error; 2188 2189 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2190 return (error); 2191 2192 return (0); 2193} 2194 2195/* 2196 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 2197 * is an invariant across any of the operations that can be performed while the 2198 * filesystem was suspended. Whether it succeeded or failed, the preconditions 2199 * are the same: the relevant objset and associated dataset are owned by 2200 * zfsvfs, held, and long held on entry. 2201 */ 2202int 2203zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2204{ 2205 int err; 2206 znode_t *zp; 2207 uint64_t sa_obj = 0; 2208 2209 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2210 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2211 2212 /* 2213 * We already own this, so just hold and rele it to update the 2214 * objset_t, as the one we had before may have been evicted. 2215 */ 2216 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os)); 2217 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs); 2218 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset)); 2219 dmu_objset_rele(zfsvfs->z_os, zfsvfs); 2220 2221 /* 2222 * Make sure version hasn't changed 2223 */ 2224 2225 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, 2226 &zfsvfs->z_version); 2227 2228 if (err) 2229 goto bail; 2230 2231 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, 2232 ZFS_SA_ATTRS, 8, 1, &sa_obj); 2233 2234 if (err && zfsvfs->z_version >= ZPL_VERSION_SA) 2235 goto bail; 2236 2237 if ((err = sa_setup(zfsvfs->z_os, sa_obj, 2238 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) 2239 goto bail; 2240 2241 if (zfsvfs->z_version >= ZPL_VERSION_SA) 2242 sa_register_update_callback(zfsvfs->z_os, 2243 zfs_sa_upgrade); 2244 2245 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2246 2247 zfs_set_fuid_feature(zfsvfs); 2248 2249 /* 2250 * Attempt to re-establish all the active znodes with 2251 * their dbufs. If a zfs_rezget() fails, then we'll let 2252 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2253 * when they try to use their znode. 2254 */ 2255 mutex_enter(&zfsvfs->z_znodes_lock); 2256 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2257 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2258 (void) zfs_rezget(zp); 2259 } 2260 mutex_exit(&zfsvfs->z_znodes_lock); 2261 2262bail: 2263 /* release the VOPs */ 2264 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2265 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 2266 2267 if (err) { 2268 /* 2269 * Since we couldn't setup the sa framework, try to force 2270 * unmount this file system. 2271 */ 2272 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) { 2273 vfs_ref(zfsvfs->z_vfs); 2274 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2275 } 2276 } 2277 return (err); 2278} 2279 2280static void 2281zfs_freevfs(vfs_t *vfsp) 2282{ 2283 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2284 2285#ifdef illumos 2286 /* 2287 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2288 * from zfs_mount(). Release it here. If we came through 2289 * zfs_mountroot() instead, we didn't grab an extra hold, so 2290 * skip the VFS_RELE for rootvfs. 2291 */ 2292 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2293 VFS_RELE(zfsvfs->z_parent->z_vfs); 2294#endif 2295 2296 zfsvfs_free(zfsvfs); 2297 2298 atomic_dec_32(&zfs_active_fs_count); 2299} 2300 2301#ifdef __i386__ 2302static int desiredvnodes_backup; 2303#endif 2304 2305static void 2306zfs_vnodes_adjust(void) 2307{ 2308#ifdef __i386__ 2309 int newdesiredvnodes; 2310 2311 desiredvnodes_backup = desiredvnodes; 2312 2313 /* 2314 * We calculate newdesiredvnodes the same way it is done in 2315 * vntblinit(). If it is equal to desiredvnodes, it means that 2316 * it wasn't tuned by the administrator and we can tune it down. 2317 */ 2318 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2319 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2320 sizeof(struct vnode)))); 2321 if (newdesiredvnodes == desiredvnodes) 2322 desiredvnodes = (3 * newdesiredvnodes) / 4; 2323#endif 2324} 2325 2326static void 2327zfs_vnodes_adjust_back(void) 2328{ 2329 2330#ifdef __i386__ 2331 desiredvnodes = desiredvnodes_backup; 2332#endif 2333} 2334 2335void 2336zfs_init(void) 2337{ 2338 2339 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2340 2341 /* 2342 * Initialize .zfs directory structures 2343 */ 2344 zfsctl_init(); 2345 2346 /* 2347 * Initialize znode cache, vnode ops, etc... 2348 */ 2349 zfs_znode_init(); 2350 2351 /* 2352 * Reduce number of vnodes. Originally number of vnodes is calculated 2353 * with UFS inode in mind. We reduce it here, because it's too big for 2354 * ZFS/i386. 2355 */ 2356 zfs_vnodes_adjust(); 2357 2358 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2359} 2360 2361void 2362zfs_fini(void) 2363{ 2364 zfsctl_fini(); 2365 zfs_znode_fini(); 2366 zfs_vnodes_adjust_back(); 2367} 2368 2369int 2370zfs_busy(void) 2371{ 2372 return (zfs_active_fs_count != 0); 2373} 2374 2375int 2376zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2377{ 2378 int error; 2379 objset_t *os = zfsvfs->z_os; 2380 dmu_tx_t *tx; 2381 2382 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2383 return (SET_ERROR(EINVAL)); 2384 2385 if (newvers < zfsvfs->z_version) 2386 return (SET_ERROR(EINVAL)); 2387 2388 if (zfs_spa_version_map(newvers) > 2389 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2390 return (SET_ERROR(ENOTSUP)); 2391 2392 tx = dmu_tx_create(os); 2393 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2394 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2395 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2396 ZFS_SA_ATTRS); 2397 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2398 } 2399 error = dmu_tx_assign(tx, TXG_WAIT); 2400 if (error) { 2401 dmu_tx_abort(tx); 2402 return (error); 2403 } 2404 2405 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2406 8, 1, &newvers, tx); 2407 2408 if (error) { 2409 dmu_tx_commit(tx); 2410 return (error); 2411 } 2412 2413 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2414 uint64_t sa_obj; 2415 2416 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2417 SPA_VERSION_SA); 2418 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2419 DMU_OT_NONE, 0, tx); 2420 2421 error = zap_add(os, MASTER_NODE_OBJ, 2422 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2423 ASSERT0(error); 2424 2425 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2426 sa_register_update_callback(os, zfs_sa_upgrade); 2427 } 2428 2429 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2430 "from %llu to %llu", zfsvfs->z_version, newvers); 2431 2432 dmu_tx_commit(tx); 2433 2434 zfsvfs->z_version = newvers; 2435 2436 zfs_set_fuid_feature(zfsvfs); 2437 2438 return (0); 2439} 2440 2441/* 2442 * Read a property stored within the master node. 2443 */ 2444int 2445zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2446{ 2447 const char *pname; 2448 int error = ENOENT; 2449 2450 /* 2451 * Look up the file system's value for the property. For the 2452 * version property, we look up a slightly different string. 2453 */ 2454 if (prop == ZFS_PROP_VERSION) 2455 pname = ZPL_VERSION_STR; 2456 else 2457 pname = zfs_prop_to_name(prop); 2458 2459 if (os != NULL) 2460 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2461 2462 if (error == ENOENT) { 2463 /* No value set, use the default value */ 2464 switch (prop) { 2465 case ZFS_PROP_VERSION: 2466 *value = ZPL_VERSION; 2467 break; 2468 case ZFS_PROP_NORMALIZE: 2469 case ZFS_PROP_UTF8ONLY: 2470 *value = 0; 2471 break; 2472 case ZFS_PROP_CASE: 2473 *value = ZFS_CASE_SENSITIVE; 2474 break; 2475 default: 2476 return (error); 2477 } 2478 error = 0; 2479 } 2480 return (error); 2481} 2482 2483#ifdef _KERNEL 2484void 2485zfsvfs_update_fromname(const char *oldname, const char *newname) 2486{ 2487 char tmpbuf[MAXPATHLEN]; 2488 struct mount *mp; 2489 char *fromname; 2490 size_t oldlen; 2491 2492 oldlen = strlen(oldname); 2493 2494 mtx_lock(&mountlist_mtx); 2495 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2496 fromname = mp->mnt_stat.f_mntfromname; 2497 if (strcmp(fromname, oldname) == 0) { 2498 (void)strlcpy(fromname, newname, 2499 sizeof(mp->mnt_stat.f_mntfromname)); 2500 continue; 2501 } 2502 if (strncmp(fromname, oldname, oldlen) == 0 && 2503 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2504 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2505 newname, fromname + oldlen); 2506 (void)strlcpy(fromname, tmpbuf, 2507 sizeof(mp->mnt_stat.f_mntfromname)); 2508 continue; 2509 } 2510 } 2511 mtx_unlock(&mountlist_mtx); 2512} 2513#endif 2514