zfs_vfsops.c revision 304135
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 return (0); 961} 962 963int 964zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 965{ 966 objset_t *os; 967 zfsvfs_t *zfsvfs; 968 int error; 969 970 /* 971 * XXX: Fix struct statfs so this isn't necessary! 972 * 973 * The 'osname' is used as the filesystem's special node, which means 974 * it must fit in statfs.f_mntfromname, or else it can't be 975 * enumerated, so libzfs_mnttab_find() returns NULL, which causes 976 * 'zfs unmount' to think it's not mounted when it is. 977 */ 978 if (strlen(osname) >= MNAMELEN) 979 return (SET_ERROR(ENAMETOOLONG)); 980 981 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 982 983 /* 984 * We claim to always be readonly so we can open snapshots; 985 * other ZPL code will prevent us from writing to snapshots. 986 */ 987 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 988 if (error) { 989 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 990 return (error); 991 } 992 993 zfsvfs->z_vfs = NULL; 994 zfsvfs->z_parent = zfsvfs; 995 996 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 997 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 998 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 999 offsetof(znode_t, z_link_node)); 1000 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE); 1001 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 1002 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 1003 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1004 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1005 1006 error = zfsvfs_init(zfsvfs, os); 1007 if (error != 0) { 1008 dmu_objset_disown(os, zfsvfs); 1009 *zfvp = NULL; 1010 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1011 return (error); 1012 } 1013 1014 *zfvp = zfsvfs; 1015 return (0); 1016} 1017 1018static int 1019zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 1020{ 1021 int error; 1022 1023 error = zfs_register_callbacks(zfsvfs->z_vfs); 1024 if (error) 1025 return (error); 1026 1027 /* 1028 * Set the objset user_ptr to track its zfsvfs. 1029 */ 1030 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1031 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1032 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1033 1034 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 1035 1036 /* 1037 * If we are not mounting (ie: online recv), then we don't 1038 * have to worry about replaying the log as we blocked all 1039 * operations out since we closed the ZIL. 1040 */ 1041 if (mounting) { 1042 boolean_t readonly; 1043 1044 /* 1045 * During replay we remove the read only flag to 1046 * allow replays to succeed. 1047 */ 1048 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1049 if (readonly != 0) 1050 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1051 else 1052 zfs_unlinked_drain(zfsvfs); 1053 1054 /* 1055 * Parse and replay the intent log. 1056 * 1057 * Because of ziltest, this must be done after 1058 * zfs_unlinked_drain(). (Further note: ziltest 1059 * doesn't use readonly mounts, where 1060 * zfs_unlinked_drain() isn't called.) This is because 1061 * ziltest causes spa_sync() to think it's committed, 1062 * but actually it is not, so the intent log contains 1063 * many txg's worth of changes. 1064 * 1065 * In particular, if object N is in the unlinked set in 1066 * the last txg to actually sync, then it could be 1067 * actually freed in a later txg and then reallocated 1068 * in a yet later txg. This would write a "create 1069 * object N" record to the intent log. Normally, this 1070 * would be fine because the spa_sync() would have 1071 * written out the fact that object N is free, before 1072 * we could write the "create object N" intent log 1073 * record. 1074 * 1075 * But when we are in ziltest mode, we advance the "open 1076 * txg" without actually spa_sync()-ing the changes to 1077 * disk. So we would see that object N is still 1078 * allocated and in the unlinked set, and there is an 1079 * intent log record saying to allocate it. 1080 */ 1081 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1082 if (zil_replay_disable) { 1083 zil_destroy(zfsvfs->z_log, B_FALSE); 1084 } else { 1085 zfsvfs->z_replay = B_TRUE; 1086 zil_replay(zfsvfs->z_os, zfsvfs, 1087 zfs_replay_vector); 1088 zfsvfs->z_replay = B_FALSE; 1089 } 1090 } 1091 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1092 } 1093 1094 return (0); 1095} 1096 1097extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1098 1099void 1100zfsvfs_free(zfsvfs_t *zfsvfs) 1101{ 1102 int i; 1103 1104 /* 1105 * This is a barrier to prevent the filesystem from going away in 1106 * zfs_znode_move() until we can safely ensure that the filesystem is 1107 * not unmounted. We consider the filesystem valid before the barrier 1108 * and invalid after the barrier. 1109 */ 1110 rw_enter(&zfsvfs_lock, RW_READER); 1111 rw_exit(&zfsvfs_lock); 1112 1113 zfs_fuid_destroy(zfsvfs); 1114 1115 mutex_destroy(&zfsvfs->z_znodes_lock); 1116 mutex_destroy(&zfsvfs->z_lock); 1117 list_destroy(&zfsvfs->z_all_znodes); 1118 rrm_destroy(&zfsvfs->z_teardown_lock); 1119 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1120 rw_destroy(&zfsvfs->z_fuid_lock); 1121 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1122 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1123 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1124} 1125 1126static void 1127zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1128{ 1129 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1130 if (zfsvfs->z_vfs) { 1131 if (zfsvfs->z_use_fuids) { 1132 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1133 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1134 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1135 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1136 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1137 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1138 } else { 1139 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1140 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1141 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1142 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1143 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1144 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1145 } 1146 } 1147 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1148} 1149 1150static int 1151zfs_domount(vfs_t *vfsp, char *osname) 1152{ 1153 uint64_t recordsize, fsid_guid; 1154 int error = 0; 1155 zfsvfs_t *zfsvfs; 1156 vnode_t *vp; 1157 1158 ASSERT(vfsp); 1159 ASSERT(osname); 1160 1161 error = zfsvfs_create(osname, &zfsvfs); 1162 if (error) 1163 return (error); 1164 zfsvfs->z_vfs = vfsp; 1165 1166#ifdef illumos 1167 /* Initialize the generic filesystem structure. */ 1168 vfsp->vfs_bcount = 0; 1169 vfsp->vfs_data = NULL; 1170 1171 if (zfs_create_unique_device(&mount_dev) == -1) { 1172 error = SET_ERROR(ENODEV); 1173 goto out; 1174 } 1175 ASSERT(vfs_devismounted(mount_dev) == 0); 1176#endif 1177 1178 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1179 NULL)) 1180 goto out; 1181 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1182 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1183 1184 vfsp->vfs_data = zfsvfs; 1185 vfsp->mnt_flag |= MNT_LOCAL; 1186 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1187 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1188 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1189 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */ 1190 1191 /* 1192 * The fsid is 64 bits, composed of an 8-bit fs type, which 1193 * separates our fsid from any other filesystem types, and a 1194 * 56-bit objset unique ID. The objset unique ID is unique to 1195 * all objsets open on this system, provided by unique_create(). 1196 * The 8-bit fs type must be put in the low bits of fsid[1] 1197 * because that's where other Solaris filesystems put it. 1198 */ 1199 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1200 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1201 vfsp->vfs_fsid.val[0] = fsid_guid; 1202 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1203 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1204 1205 /* 1206 * Set features for file system. 1207 */ 1208 zfs_set_fuid_feature(zfsvfs); 1209 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1210 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1211 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1212 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1213 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1214 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1215 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1216 } 1217 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1218 1219 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1220 uint64_t pval; 1221 1222 atime_changed_cb(zfsvfs, B_FALSE); 1223 readonly_changed_cb(zfsvfs, B_TRUE); 1224 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1225 goto out; 1226 xattr_changed_cb(zfsvfs, pval); 1227 zfsvfs->z_issnap = B_TRUE; 1228 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1229 1230 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1231 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1232 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1233 } else { 1234 error = zfsvfs_setup(zfsvfs, B_TRUE); 1235 } 1236 1237 vfs_mountedfrom(vfsp, osname); 1238 1239 if (!zfsvfs->z_issnap) 1240 zfsctl_create(zfsvfs); 1241out: 1242 if (error) { 1243 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1244 zfsvfs_free(zfsvfs); 1245 } else { 1246 atomic_inc_32(&zfs_active_fs_count); 1247 } 1248 1249 return (error); 1250} 1251 1252void 1253zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1254{ 1255 objset_t *os = zfsvfs->z_os; 1256 1257 if (!dmu_objset_is_snapshot(os)) 1258 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 1259} 1260 1261#ifdef SECLABEL 1262/* 1263 * Convert a decimal digit string to a uint64_t integer. 1264 */ 1265static int 1266str_to_uint64(char *str, uint64_t *objnum) 1267{ 1268 uint64_t num = 0; 1269 1270 while (*str) { 1271 if (*str < '0' || *str > '9') 1272 return (SET_ERROR(EINVAL)); 1273 1274 num = num*10 + *str++ - '0'; 1275 } 1276 1277 *objnum = num; 1278 return (0); 1279} 1280 1281/* 1282 * The boot path passed from the boot loader is in the form of 1283 * "rootpool-name/root-filesystem-object-number'. Convert this 1284 * string to a dataset name: "rootpool-name/root-filesystem-name". 1285 */ 1286static int 1287zfs_parse_bootfs(char *bpath, char *outpath) 1288{ 1289 char *slashp; 1290 uint64_t objnum; 1291 int error; 1292 1293 if (*bpath == 0 || *bpath == '/') 1294 return (SET_ERROR(EINVAL)); 1295 1296 (void) strcpy(outpath, bpath); 1297 1298 slashp = strchr(bpath, '/'); 1299 1300 /* if no '/', just return the pool name */ 1301 if (slashp == NULL) { 1302 return (0); 1303 } 1304 1305 /* if not a number, just return the root dataset name */ 1306 if (str_to_uint64(slashp+1, &objnum)) { 1307 return (0); 1308 } 1309 1310 *slashp = '\0'; 1311 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1312 *slashp = '/'; 1313 1314 return (error); 1315} 1316 1317/* 1318 * Check that the hex label string is appropriate for the dataset being 1319 * mounted into the global_zone proper. 1320 * 1321 * Return an error if the hex label string is not default or 1322 * admin_low/admin_high. For admin_low labels, the corresponding 1323 * dataset must be readonly. 1324 */ 1325int 1326zfs_check_global_label(const char *dsname, const char *hexsl) 1327{ 1328 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1329 return (0); 1330 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1331 return (0); 1332 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1333 /* must be readonly */ 1334 uint64_t rdonly; 1335 1336 if (dsl_prop_get_integer(dsname, 1337 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1338 return (SET_ERROR(EACCES)); 1339 return (rdonly ? 0 : EACCES); 1340 } 1341 return (SET_ERROR(EACCES)); 1342} 1343 1344/* 1345 * Determine whether the mount is allowed according to MAC check. 1346 * by comparing (where appropriate) label of the dataset against 1347 * the label of the zone being mounted into. If the dataset has 1348 * no label, create one. 1349 * 1350 * Returns 0 if access allowed, error otherwise (e.g. EACCES) 1351 */ 1352static int 1353zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1354{ 1355 int error, retv; 1356 zone_t *mntzone = NULL; 1357 ts_label_t *mnt_tsl; 1358 bslabel_t *mnt_sl; 1359 bslabel_t ds_sl; 1360 char ds_hexsl[MAXNAMELEN]; 1361 1362 retv = EACCES; /* assume the worst */ 1363 1364 /* 1365 * Start by getting the dataset label if it exists. 1366 */ 1367 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1368 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1369 if (error) 1370 return (SET_ERROR(EACCES)); 1371 1372 /* 1373 * If labeling is NOT enabled, then disallow the mount of datasets 1374 * which have a non-default label already. No other label checks 1375 * are needed. 1376 */ 1377 if (!is_system_labeled()) { 1378 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1379 return (0); 1380 return (SET_ERROR(EACCES)); 1381 } 1382 1383 /* 1384 * Get the label of the mountpoint. If mounting into the global 1385 * zone (i.e. mountpoint is not within an active zone and the 1386 * zoned property is off), the label must be default or 1387 * admin_low/admin_high only; no other checks are needed. 1388 */ 1389 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1390 if (mntzone->zone_id == GLOBAL_ZONEID) { 1391 uint64_t zoned; 1392 1393 zone_rele(mntzone); 1394 1395 if (dsl_prop_get_integer(osname, 1396 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1397 return (SET_ERROR(EACCES)); 1398 if (!zoned) 1399 return (zfs_check_global_label(osname, ds_hexsl)); 1400 else 1401 /* 1402 * This is the case of a zone dataset being mounted 1403 * initially, before the zone has been fully created; 1404 * allow this mount into global zone. 1405 */ 1406 return (0); 1407 } 1408 1409 mnt_tsl = mntzone->zone_slabel; 1410 ASSERT(mnt_tsl != NULL); 1411 label_hold(mnt_tsl); 1412 mnt_sl = label2bslabel(mnt_tsl); 1413 1414 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1415 /* 1416 * The dataset doesn't have a real label, so fabricate one. 1417 */ 1418 char *str = NULL; 1419 1420 if (l_to_str_internal(mnt_sl, &str) == 0 && 1421 dsl_prop_set_string(osname, 1422 zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1423 ZPROP_SRC_LOCAL, str) == 0) 1424 retv = 0; 1425 if (str != NULL) 1426 kmem_free(str, strlen(str) + 1); 1427 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1428 /* 1429 * Now compare labels to complete the MAC check. If the 1430 * labels are equal then allow access. If the mountpoint 1431 * label dominates the dataset label, allow readonly access. 1432 * Otherwise, access is denied. 1433 */ 1434 if (blequal(mnt_sl, &ds_sl)) 1435 retv = 0; 1436 else if (bldominates(mnt_sl, &ds_sl)) { 1437 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1438 retv = 0; 1439 } 1440 } 1441 1442 label_rele(mnt_tsl); 1443 zone_rele(mntzone); 1444 return (retv); 1445} 1446#endif /* SECLABEL */ 1447 1448#ifdef OPENSOLARIS_MOUNTROOT 1449static int 1450zfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1451{ 1452 int error = 0; 1453 static int zfsrootdone = 0; 1454 zfsvfs_t *zfsvfs = NULL; 1455 znode_t *zp = NULL; 1456 vnode_t *vp = NULL; 1457 char *zfs_bootfs; 1458 char *zfs_devid; 1459 1460 ASSERT(vfsp); 1461 1462 /* 1463 * The filesystem that we mount as root is defined in the 1464 * boot property "zfs-bootfs" with a format of 1465 * "poolname/root-dataset-objnum". 1466 */ 1467 if (why == ROOT_INIT) { 1468 if (zfsrootdone++) 1469 return (SET_ERROR(EBUSY)); 1470 /* 1471 * the process of doing a spa_load will require the 1472 * clock to be set before we could (for example) do 1473 * something better by looking at the timestamp on 1474 * an uberblock, so just set it to -1. 1475 */ 1476 clkset(-1); 1477 1478 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1479 cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1480 "bootfs name"); 1481 return (SET_ERROR(EINVAL)); 1482 } 1483 zfs_devid = spa_get_bootprop("diskdevid"); 1484 error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1485 if (zfs_devid) 1486 spa_free_bootprop(zfs_devid); 1487 if (error) { 1488 spa_free_bootprop(zfs_bootfs); 1489 cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1490 error); 1491 return (error); 1492 } 1493 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1494 spa_free_bootprop(zfs_bootfs); 1495 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1496 error); 1497 return (error); 1498 } 1499 1500 spa_free_bootprop(zfs_bootfs); 1501 1502 if (error = vfs_lock(vfsp)) 1503 return (error); 1504 1505 if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1506 cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1507 goto out; 1508 } 1509 1510 zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1511 ASSERT(zfsvfs); 1512 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1513 cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1514 goto out; 1515 } 1516 1517 vp = ZTOV(zp); 1518 mutex_enter(&vp->v_lock); 1519 vp->v_flag |= VROOT; 1520 mutex_exit(&vp->v_lock); 1521 rootvp = vp; 1522 1523 /* 1524 * Leave rootvp held. The root file system is never unmounted. 1525 */ 1526 1527 vfs_add((struct vnode *)0, vfsp, 1528 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1529out: 1530 vfs_unlock(vfsp); 1531 return (error); 1532 } else if (why == ROOT_REMOUNT) { 1533 readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1534 vfsp->vfs_flag |= VFS_REMOUNT; 1535 1536 /* refresh mount options */ 1537 zfs_unregister_callbacks(vfsp->vfs_data); 1538 return (zfs_register_callbacks(vfsp)); 1539 1540 } else if (why == ROOT_UNMOUNT) { 1541 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1542 (void) zfs_sync(vfsp, 0, 0); 1543 return (0); 1544 } 1545 1546 /* 1547 * if "why" is equal to anything else other than ROOT_INIT, 1548 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1549 */ 1550 return (SET_ERROR(ENOTSUP)); 1551} 1552#endif /* OPENSOLARIS_MOUNTROOT */ 1553 1554static int 1555getpoolname(const char *osname, char *poolname) 1556{ 1557 char *p; 1558 1559 p = strchr(osname, '/'); 1560 if (p == NULL) { 1561 if (strlen(osname) >= MAXNAMELEN) 1562 return (ENAMETOOLONG); 1563 (void) strcpy(poolname, osname); 1564 } else { 1565 if (p - osname >= MAXNAMELEN) 1566 return (ENAMETOOLONG); 1567 (void) strncpy(poolname, osname, p - osname); 1568 poolname[p - osname] = '\0'; 1569 } 1570 return (0); 1571} 1572 1573/*ARGSUSED*/ 1574static int 1575zfs_mount(vfs_t *vfsp) 1576{ 1577 kthread_t *td = curthread; 1578 vnode_t *mvp = vfsp->mnt_vnodecovered; 1579 cred_t *cr = td->td_ucred; 1580 char *osname; 1581 int error = 0; 1582 int canwrite; 1583 1584#ifdef illumos 1585 if (mvp->v_type != VDIR) 1586 return (SET_ERROR(ENOTDIR)); 1587 1588 mutex_enter(&mvp->v_lock); 1589 if ((uap->flags & MS_REMOUNT) == 0 && 1590 (uap->flags & MS_OVERLAY) == 0 && 1591 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 1592 mutex_exit(&mvp->v_lock); 1593 return (SET_ERROR(EBUSY)); 1594 } 1595 mutex_exit(&mvp->v_lock); 1596 1597 /* 1598 * ZFS does not support passing unparsed data in via MS_DATA. 1599 * Users should use the MS_OPTIONSTR interface; this means 1600 * that all option parsing is already done and the options struct 1601 * can be interrogated. 1602 */ 1603 if ((uap->flags & MS_DATA) && uap->datalen > 0) 1604#else /* !illumos */ 1605 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1606 return (SET_ERROR(EPERM)); 1607 1608 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1609 return (SET_ERROR(EINVAL)); 1610#endif /* illumos */ 1611 1612 /* 1613 * If full-owner-access is enabled and delegated administration is 1614 * turned on, we must set nosuid. 1615 */ 1616 if (zfs_super_owner && 1617 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1618 secpolicy_fs_mount_clearopts(cr, vfsp); 1619 } 1620 1621 /* 1622 * Check for mount privilege? 1623 * 1624 * If we don't have privilege then see if 1625 * we have local permission to allow it 1626 */ 1627 error = secpolicy_fs_mount(cr, mvp, vfsp); 1628 if (error) { 1629 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1630 goto out; 1631 1632 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1633 vattr_t vattr; 1634 1635 /* 1636 * Make sure user is the owner of the mount point 1637 * or has sufficient privileges. 1638 */ 1639 1640 vattr.va_mask = AT_UID; 1641 1642 vn_lock(mvp, LK_SHARED | LK_RETRY); 1643 if (VOP_GETATTR(mvp, &vattr, cr)) { 1644 VOP_UNLOCK(mvp, 0); 1645 goto out; 1646 } 1647 1648 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1649 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1650 VOP_UNLOCK(mvp, 0); 1651 goto out; 1652 } 1653 VOP_UNLOCK(mvp, 0); 1654 } 1655 1656 secpolicy_fs_mount_clearopts(cr, vfsp); 1657 } 1658 1659 /* 1660 * Refuse to mount a filesystem if we are in a local zone and the 1661 * dataset is not visible. 1662 */ 1663 if (!INGLOBALZONE(curthread) && 1664 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1665 error = SET_ERROR(EPERM); 1666 goto out; 1667 } 1668 1669#ifdef SECLABEL 1670 error = zfs_mount_label_policy(vfsp, osname); 1671 if (error) 1672 goto out; 1673#endif 1674 1675 vfsp->vfs_flag |= MNT_NFS4ACLS; 1676 1677 /* 1678 * When doing a remount, we simply refresh our temporary properties 1679 * according to those options set in the current VFS options. 1680 */ 1681 if (vfsp->vfs_flag & MS_REMOUNT) { 1682 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1683 1684 /* 1685 * Refresh mount options with z_teardown_lock blocking I/O while 1686 * the filesystem is in an inconsistent state. 1687 * The lock also serializes this code with filesystem 1688 * manipulations between entry to zfs_suspend_fs() and return 1689 * from zfs_resume_fs(). 1690 */ 1691 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1692 zfs_unregister_callbacks(zfsvfs); 1693 error = zfs_register_callbacks(vfsp); 1694 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1695 goto out; 1696 } 1697 1698 /* Initial root mount: try hard to import the requested root pool. */ 1699 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1700 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1701 char pname[MAXNAMELEN]; 1702 1703 error = getpoolname(osname, pname); 1704 if (error == 0) 1705 error = spa_import_rootpool(pname); 1706 if (error) 1707 goto out; 1708 } 1709 DROP_GIANT(); 1710 error = zfs_domount(vfsp, osname); 1711 PICKUP_GIANT(); 1712 1713#ifdef illumos 1714 /* 1715 * Add an extra VFS_HOLD on our parent vfs so that it can't 1716 * disappear due to a forced unmount. 1717 */ 1718 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1719 VFS_HOLD(mvp->v_vfsp); 1720#endif 1721 1722out: 1723 return (error); 1724} 1725 1726static int 1727zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1728{ 1729 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1730 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1731 1732 statp->f_version = STATFS_VERSION; 1733 1734 ZFS_ENTER(zfsvfs); 1735 1736 dmu_objset_space(zfsvfs->z_os, 1737 &refdbytes, &availbytes, &usedobjs, &availobjs); 1738 1739 /* 1740 * The underlying storage pool actually uses multiple block sizes. 1741 * We report the fragsize as the smallest block size we support, 1742 * and we report our blocksize as the filesystem's maximum blocksize. 1743 */ 1744 statp->f_bsize = SPA_MINBLOCKSIZE; 1745 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1746 1747 /* 1748 * The following report "total" blocks of various kinds in the 1749 * file system, but reported in terms of f_frsize - the 1750 * "fragment" size. 1751 */ 1752 1753 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1754 statp->f_bfree = availbytes / statp->f_bsize; 1755 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1756 1757 /* 1758 * statvfs() should really be called statufs(), because it assumes 1759 * static metadata. ZFS doesn't preallocate files, so the best 1760 * we can do is report the max that could possibly fit in f_files, 1761 * and that minus the number actually used in f_ffree. 1762 * For f_ffree, report the smaller of the number of object available 1763 * and the number of blocks (each object will take at least a block). 1764 */ 1765 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1766 statp->f_files = statp->f_ffree + usedobjs; 1767 1768 /* 1769 * We're a zfs filesystem. 1770 */ 1771 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1772 1773 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1774 sizeof(statp->f_mntfromname)); 1775 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1776 sizeof(statp->f_mntonname)); 1777 1778 statp->f_namemax = ZFS_MAXNAMELEN; 1779 1780 ZFS_EXIT(zfsvfs); 1781 return (0); 1782} 1783 1784static int 1785zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1786{ 1787 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1788 znode_t *rootzp; 1789 int error; 1790 1791 ZFS_ENTER(zfsvfs); 1792 1793 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1794 if (error == 0) 1795 *vpp = ZTOV(rootzp); 1796 1797 ZFS_EXIT(zfsvfs); 1798 1799 if (error == 0) { 1800 error = vn_lock(*vpp, flags); 1801 if (error != 0) { 1802 VN_RELE(*vpp); 1803 *vpp = NULL; 1804 } 1805 } 1806 return (error); 1807} 1808 1809/* 1810 * Teardown the zfsvfs::z_os. 1811 * 1812 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1813 * and 'z_teardown_inactive_lock' held. 1814 */ 1815static int 1816zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1817{ 1818 znode_t *zp; 1819 1820 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1821 1822 if (!unmounting) { 1823 /* 1824 * We purge the parent filesystem's vfsp as the parent 1825 * filesystem and all of its snapshots have their vnode's 1826 * v_vfsp set to the parent's filesystem's vfsp. Note, 1827 * 'z_parent' is self referential for non-snapshots. 1828 */ 1829 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1830#ifdef FREEBSD_NAMECACHE 1831 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1832#endif 1833 } 1834 1835 /* 1836 * Close the zil. NB: Can't close the zil while zfs_inactive 1837 * threads are blocked as zil_close can call zfs_inactive. 1838 */ 1839 if (zfsvfs->z_log) { 1840 zil_close(zfsvfs->z_log); 1841 zfsvfs->z_log = NULL; 1842 } 1843 1844 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1845 1846 /* 1847 * If we are not unmounting (ie: online recv) and someone already 1848 * unmounted this file system while we were doing the switcheroo, 1849 * or a reopen of z_os failed then just bail out now. 1850 */ 1851 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1852 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1853 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1854 return (SET_ERROR(EIO)); 1855 } 1856 1857 /* 1858 * At this point there are no vops active, and any new vops will 1859 * fail with EIO since we have z_teardown_lock for writer (only 1860 * relavent for forced unmount). 1861 * 1862 * Release all holds on dbufs. 1863 */ 1864 mutex_enter(&zfsvfs->z_znodes_lock); 1865 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1866 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1867 if (zp->z_sa_hdl) { 1868 ASSERT(ZTOV(zp)->v_count >= 0); 1869 zfs_znode_dmu_fini(zp); 1870 } 1871 mutex_exit(&zfsvfs->z_znodes_lock); 1872 1873 /* 1874 * If we are unmounting, set the unmounted flag and let new vops 1875 * unblock. zfs_inactive will have the unmounted behavior, and all 1876 * other vops will fail with EIO. 1877 */ 1878 if (unmounting) { 1879 zfsvfs->z_unmounted = B_TRUE; 1880 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1881 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1882 } 1883 1884 /* 1885 * z_os will be NULL if there was an error in attempting to reopen 1886 * zfsvfs, so just return as the properties had already been 1887 * unregistered and cached data had been evicted before. 1888 */ 1889 if (zfsvfs->z_os == NULL) 1890 return (0); 1891 1892 /* 1893 * Unregister properties. 1894 */ 1895 zfs_unregister_callbacks(zfsvfs); 1896 1897 /* 1898 * Evict cached data 1899 */ 1900 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1901 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1902 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1903 dmu_objset_evict_dbufs(zfsvfs->z_os); 1904 1905 return (0); 1906} 1907 1908/*ARGSUSED*/ 1909static int 1910zfs_umount(vfs_t *vfsp, int fflag) 1911{ 1912 kthread_t *td = curthread; 1913 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1914 objset_t *os; 1915 cred_t *cr = td->td_ucred; 1916 int ret; 1917 1918 ret = secpolicy_fs_unmount(cr, vfsp); 1919 if (ret) { 1920 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1921 ZFS_DELEG_PERM_MOUNT, cr)) 1922 return (ret); 1923 } 1924 1925 /* 1926 * We purge the parent filesystem's vfsp as the parent filesystem 1927 * and all of its snapshots have their vnode's v_vfsp set to the 1928 * parent's filesystem's vfsp. Note, 'z_parent' is self 1929 * referential for non-snapshots. 1930 */ 1931 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1932 1933 /* 1934 * Unmount any snapshots mounted under .zfs before unmounting the 1935 * dataset itself. 1936 */ 1937 if (zfsvfs->z_ctldir != NULL) { 1938 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1939 return (ret); 1940 ret = vflush(vfsp, 0, 0, td); 1941 ASSERT(ret == EBUSY); 1942 if (!(fflag & MS_FORCE)) { 1943 if (zfsvfs->z_ctldir->v_count > 1) 1944 return (EBUSY); 1945 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1946 } 1947 zfsctl_destroy(zfsvfs); 1948 ASSERT(zfsvfs->z_ctldir == NULL); 1949 } 1950 1951 if (fflag & MS_FORCE) { 1952 /* 1953 * Mark file system as unmounted before calling 1954 * vflush(FORCECLOSE). This way we ensure no future vnops 1955 * will be called and risk operating on DOOMED vnodes. 1956 */ 1957 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1958 zfsvfs->z_unmounted = B_TRUE; 1959 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1960 } 1961 1962 /* 1963 * Flush all the files. 1964 */ 1965 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1966 if (ret != 0) { 1967 if (!zfsvfs->z_issnap) { 1968 zfsctl_create(zfsvfs); 1969 ASSERT(zfsvfs->z_ctldir != NULL); 1970 } 1971 return (ret); 1972 } 1973 1974#ifdef illumos 1975 if (!(fflag & MS_FORCE)) { 1976 /* 1977 * Check the number of active vnodes in the file system. 1978 * Our count is maintained in the vfs structure, but the 1979 * number is off by 1 to indicate a hold on the vfs 1980 * structure itself. 1981 * 1982 * The '.zfs' directory maintains a reference of its 1983 * own, and any active references underneath are 1984 * reflected in the vnode count. 1985 */ 1986 if (zfsvfs->z_ctldir == NULL) { 1987 if (vfsp->vfs_count > 1) 1988 return (SET_ERROR(EBUSY)); 1989 } else { 1990 if (vfsp->vfs_count > 2 || 1991 zfsvfs->z_ctldir->v_count > 1) 1992 return (SET_ERROR(EBUSY)); 1993 } 1994 } 1995#endif 1996 1997 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1998 os = zfsvfs->z_os; 1999 2000 /* 2001 * z_os will be NULL if there was an error in 2002 * attempting to reopen zfsvfs. 2003 */ 2004 if (os != NULL) { 2005 /* 2006 * Unset the objset user_ptr. 2007 */ 2008 mutex_enter(&os->os_user_ptr_lock); 2009 dmu_objset_set_user(os, NULL); 2010 mutex_exit(&os->os_user_ptr_lock); 2011 2012 /* 2013 * Finally release the objset 2014 */ 2015 dmu_objset_disown(os, zfsvfs); 2016 } 2017 2018 /* 2019 * We can now safely destroy the '.zfs' directory node. 2020 */ 2021 if (zfsvfs->z_ctldir != NULL) 2022 zfsctl_destroy(zfsvfs); 2023 zfs_freevfs(vfsp); 2024 2025 return (0); 2026} 2027 2028static int 2029zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2030{ 2031 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2032 znode_t *zp; 2033 int err; 2034 2035 /* 2036 * zfs_zget() can't operate on virtual entries like .zfs/ or 2037 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2038 * This will make NFS to switch to LOOKUP instead of using VGET. 2039 */ 2040 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || 2041 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) 2042 return (EOPNOTSUPP); 2043 2044 ZFS_ENTER(zfsvfs); 2045 err = zfs_zget(zfsvfs, ino, &zp); 2046 if (err == 0 && zp->z_unlinked) { 2047 VN_RELE(ZTOV(zp)); 2048 err = EINVAL; 2049 } 2050 if (err == 0) 2051 *vpp = ZTOV(zp); 2052 ZFS_EXIT(zfsvfs); 2053 if (err == 0) 2054 err = vn_lock(*vpp, flags); 2055 if (err != 0) 2056 *vpp = NULL; 2057 return (err); 2058} 2059 2060static int 2061zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2062 struct ucred **credanonp, int *numsecflavors, int **secflavors) 2063{ 2064 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2065 2066 /* 2067 * If this is regular file system vfsp is the same as 2068 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2069 * zfsvfs->z_parent->z_vfs represents parent file system 2070 * which we have to use here, because only this file system 2071 * has mnt_export configured. 2072 */ 2073 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2074 credanonp, numsecflavors, secflavors)); 2075} 2076 2077CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2078CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2079 2080static int 2081zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2082{ 2083 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2084 znode_t *zp; 2085 uint64_t object = 0; 2086 uint64_t fid_gen = 0; 2087 uint64_t gen_mask; 2088 uint64_t zp_gen; 2089 int i, err; 2090 2091 *vpp = NULL; 2092 2093 ZFS_ENTER(zfsvfs); 2094 2095 /* 2096 * On FreeBSD we can get snapshot's mount point or its parent file 2097 * system mount point depending if snapshot is already mounted or not. 2098 */ 2099 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2100 zfid_long_t *zlfid = (zfid_long_t *)fidp; 2101 uint64_t objsetid = 0; 2102 uint64_t setgen = 0; 2103 2104 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2105 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2106 2107 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2108 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2109 2110 ZFS_EXIT(zfsvfs); 2111 2112 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2113 if (err) 2114 return (SET_ERROR(EINVAL)); 2115 ZFS_ENTER(zfsvfs); 2116 } 2117 2118 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2119 zfid_short_t *zfid = (zfid_short_t *)fidp; 2120 2121 for (i = 0; i < sizeof (zfid->zf_object); i++) 2122 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2123 2124 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2125 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2126 } else { 2127 ZFS_EXIT(zfsvfs); 2128 return (SET_ERROR(EINVAL)); 2129 } 2130 2131 /* 2132 * A zero fid_gen means we are in .zfs or the .zfs/snapshot 2133 * directory tree. If the object == zfsvfs->z_shares_dir, then 2134 * we are in the .zfs/shares directory tree. 2135 */ 2136 if ((fid_gen == 0 && 2137 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || 2138 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { 2139 *vpp = zfsvfs->z_ctldir; 2140 ASSERT(*vpp != NULL); 2141 if (object == ZFSCTL_INO_SNAPDIR) { 2142 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2143 0, NULL, NULL, NULL, NULL, NULL) == 0); 2144 } else if (object == zfsvfs->z_shares_dir) { 2145 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL, 2146 0, NULL, NULL, NULL, NULL, NULL) == 0); 2147 } else { 2148 VN_HOLD(*vpp); 2149 } 2150 ZFS_EXIT(zfsvfs); 2151 err = vn_lock(*vpp, flags); 2152 if (err != 0) 2153 *vpp = NULL; 2154 return (err); 2155 } 2156 2157 gen_mask = -1ULL >> (64 - 8 * i); 2158 2159 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2160 if (err = zfs_zget(zfsvfs, object, &zp)) { 2161 ZFS_EXIT(zfsvfs); 2162 return (err); 2163 } 2164 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2165 sizeof (uint64_t)); 2166 zp_gen = zp_gen & gen_mask; 2167 if (zp_gen == 0) 2168 zp_gen = 1; 2169 if (zp->z_unlinked || zp_gen != fid_gen) { 2170 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2171 VN_RELE(ZTOV(zp)); 2172 ZFS_EXIT(zfsvfs); 2173 return (SET_ERROR(EINVAL)); 2174 } 2175 2176 *vpp = ZTOV(zp); 2177 ZFS_EXIT(zfsvfs); 2178 err = vn_lock(*vpp, flags | LK_RETRY); 2179 if (err == 0) 2180 vnode_create_vobject(*vpp, zp->z_size, curthread); 2181 else 2182 *vpp = NULL; 2183 return (err); 2184} 2185 2186/* 2187 * Block out VOPs and close zfsvfs_t::z_os 2188 * 2189 * Note, if successful, then we return with the 'z_teardown_lock' and 2190 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 2191 * dataset and objset intact so that they can be atomically handed off during 2192 * a subsequent rollback or recv operation and the resume thereafter. 2193 */ 2194int 2195zfs_suspend_fs(zfsvfs_t *zfsvfs) 2196{ 2197 int error; 2198 2199 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2200 return (error); 2201 2202 return (0); 2203} 2204 2205/* 2206 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 2207 * is an invariant across any of the operations that can be performed while the 2208 * filesystem was suspended. Whether it succeeded or failed, the preconditions 2209 * are the same: the relevant objset and associated dataset are owned by 2210 * zfsvfs, held, and long held on entry. 2211 */ 2212int 2213zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2214{ 2215 int err; 2216 znode_t *zp; 2217 2218 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2219 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2220 2221 /* 2222 * We already own this, so just hold and rele it to update the 2223 * objset_t, as the one we had before may have been evicted. 2224 */ 2225 objset_t *os; 2226 VERIFY0(dmu_objset_hold(osname, zfsvfs, &os)); 2227 VERIFY3P(os->os_dsl_dataset->ds_owner, ==, zfsvfs); 2228 VERIFY(dsl_dataset_long_held(os->os_dsl_dataset)); 2229 dmu_objset_rele(os, zfsvfs); 2230 2231 err = zfsvfs_init(zfsvfs, os); 2232 if (err != 0) 2233 goto bail; 2234 2235 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2236 2237 zfs_set_fuid_feature(zfsvfs); 2238 2239 /* 2240 * Attempt to re-establish all the active znodes with 2241 * their dbufs. If a zfs_rezget() fails, then we'll let 2242 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2243 * when they try to use their znode. 2244 */ 2245 mutex_enter(&zfsvfs->z_znodes_lock); 2246 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2247 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2248 (void) zfs_rezget(zp); 2249 } 2250 mutex_exit(&zfsvfs->z_znodes_lock); 2251 2252bail: 2253 /* release the VOPs */ 2254 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2255 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 2256 2257 if (err) { 2258 /* 2259 * Since we couldn't setup the sa framework, try to force 2260 * unmount this file system. 2261 */ 2262 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) { 2263 vfs_ref(zfsvfs->z_vfs); 2264 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2265 } 2266 } 2267 return (err); 2268} 2269 2270static void 2271zfs_freevfs(vfs_t *vfsp) 2272{ 2273 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2274 2275#ifdef illumos 2276 /* 2277 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2278 * from zfs_mount(). Release it here. If we came through 2279 * zfs_mountroot() instead, we didn't grab an extra hold, so 2280 * skip the VFS_RELE for rootvfs. 2281 */ 2282 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2283 VFS_RELE(zfsvfs->z_parent->z_vfs); 2284#endif 2285 2286 zfsvfs_free(zfsvfs); 2287 2288 atomic_dec_32(&zfs_active_fs_count); 2289} 2290 2291#ifdef __i386__ 2292static int desiredvnodes_backup; 2293#endif 2294 2295static void 2296zfs_vnodes_adjust(void) 2297{ 2298#ifdef __i386__ 2299 int newdesiredvnodes; 2300 2301 desiredvnodes_backup = desiredvnodes; 2302 2303 /* 2304 * We calculate newdesiredvnodes the same way it is done in 2305 * vntblinit(). If it is equal to desiredvnodes, it means that 2306 * it wasn't tuned by the administrator and we can tune it down. 2307 */ 2308 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2309 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2310 sizeof(struct vnode)))); 2311 if (newdesiredvnodes == desiredvnodes) 2312 desiredvnodes = (3 * newdesiredvnodes) / 4; 2313#endif 2314} 2315 2316static void 2317zfs_vnodes_adjust_back(void) 2318{ 2319 2320#ifdef __i386__ 2321 desiredvnodes = desiredvnodes_backup; 2322#endif 2323} 2324 2325void 2326zfs_init(void) 2327{ 2328 2329 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2330 2331 /* 2332 * Initialize .zfs directory structures 2333 */ 2334 zfsctl_init(); 2335 2336 /* 2337 * Initialize znode cache, vnode ops, etc... 2338 */ 2339 zfs_znode_init(); 2340 2341 /* 2342 * Reduce number of vnodes. Originally number of vnodes is calculated 2343 * with UFS inode in mind. We reduce it here, because it's too big for 2344 * ZFS/i386. 2345 */ 2346 zfs_vnodes_adjust(); 2347 2348 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2349} 2350 2351void 2352zfs_fini(void) 2353{ 2354 zfsctl_fini(); 2355 zfs_znode_fini(); 2356 zfs_vnodes_adjust_back(); 2357} 2358 2359int 2360zfs_busy(void) 2361{ 2362 return (zfs_active_fs_count != 0); 2363} 2364 2365int 2366zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2367{ 2368 int error; 2369 objset_t *os = zfsvfs->z_os; 2370 dmu_tx_t *tx; 2371 2372 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2373 return (SET_ERROR(EINVAL)); 2374 2375 if (newvers < zfsvfs->z_version) 2376 return (SET_ERROR(EINVAL)); 2377 2378 if (zfs_spa_version_map(newvers) > 2379 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2380 return (SET_ERROR(ENOTSUP)); 2381 2382 tx = dmu_tx_create(os); 2383 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2384 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2385 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2386 ZFS_SA_ATTRS); 2387 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2388 } 2389 error = dmu_tx_assign(tx, TXG_WAIT); 2390 if (error) { 2391 dmu_tx_abort(tx); 2392 return (error); 2393 } 2394 2395 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2396 8, 1, &newvers, tx); 2397 2398 if (error) { 2399 dmu_tx_commit(tx); 2400 return (error); 2401 } 2402 2403 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2404 uint64_t sa_obj; 2405 2406 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2407 SPA_VERSION_SA); 2408 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2409 DMU_OT_NONE, 0, tx); 2410 2411 error = zap_add(os, MASTER_NODE_OBJ, 2412 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2413 ASSERT0(error); 2414 2415 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2416 sa_register_update_callback(os, zfs_sa_upgrade); 2417 } 2418 2419 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2420 "from %llu to %llu", zfsvfs->z_version, newvers); 2421 2422 dmu_tx_commit(tx); 2423 2424 zfsvfs->z_version = newvers; 2425 2426 zfs_set_fuid_feature(zfsvfs); 2427 2428 return (0); 2429} 2430 2431/* 2432 * Read a property stored within the master node. 2433 */ 2434int 2435zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2436{ 2437 const char *pname; 2438 int error = ENOENT; 2439 2440 /* 2441 * Look up the file system's value for the property. For the 2442 * version property, we look up a slightly different string. 2443 */ 2444 if (prop == ZFS_PROP_VERSION) 2445 pname = ZPL_VERSION_STR; 2446 else 2447 pname = zfs_prop_to_name(prop); 2448 2449 if (os != NULL) 2450 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2451 2452 if (error == ENOENT) { 2453 /* No value set, use the default value */ 2454 switch (prop) { 2455 case ZFS_PROP_VERSION: 2456 *value = ZPL_VERSION; 2457 break; 2458 case ZFS_PROP_NORMALIZE: 2459 case ZFS_PROP_UTF8ONLY: 2460 *value = 0; 2461 break; 2462 case ZFS_PROP_CASE: 2463 *value = ZFS_CASE_SENSITIVE; 2464 break; 2465 default: 2466 return (error); 2467 } 2468 error = 0; 2469 } 2470 return (error); 2471} 2472 2473#ifdef _KERNEL 2474void 2475zfsvfs_update_fromname(const char *oldname, const char *newname) 2476{ 2477 char tmpbuf[MAXPATHLEN]; 2478 struct mount *mp; 2479 char *fromname; 2480 size_t oldlen; 2481 2482 oldlen = strlen(oldname); 2483 2484 mtx_lock(&mountlist_mtx); 2485 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2486 fromname = mp->mnt_stat.f_mntfromname; 2487 if (strcmp(fromname, oldname) == 0) { 2488 (void)strlcpy(fromname, newname, 2489 sizeof(mp->mnt_stat.f_mntfromname)); 2490 continue; 2491 } 2492 if (strncmp(fromname, oldname, oldlen) == 0 && 2493 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2494 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2495 newname, fromname + oldlen); 2496 (void)strlcpy(fromname, tmpbuf, 2497 sizeof(mp->mnt_stat.f_mntfromname)); 2498 continue; 2499 } 2500 } 2501 mtx_unlock(&mountlist_mtx); 2502} 2503#endif 2504