zfs_vfsops.c revision 297077
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 if (vfsp != NULL) { 137 /* 138 * Sync a specific filesystem. 139 */ 140 zfsvfs_t *zfsvfs = vfsp->vfs_data; 141 dsl_pool_t *dp; 142 int error; 143 144 error = vfs_stdsync(vfsp, waitfor); 145 if (error != 0) 146 return (error); 147 148 ZFS_ENTER(zfsvfs); 149 dp = dmu_objset_pool(zfsvfs->z_os); 150 151 /* 152 * If the system is shutting down, then skip any 153 * filesystems which may exist on a suspended pool. 154 */ 155 if (sys_shutdown && spa_suspended(dp->dp_spa)) { 156 ZFS_EXIT(zfsvfs); 157 return (0); 158 } 159 160 if (zfsvfs->z_log != NULL) 161 zil_commit(zfsvfs->z_log, 0); 162 163 ZFS_EXIT(zfsvfs); 164 } else { 165 /* 166 * Sync all ZFS filesystems. This is what happens when you 167 * run sync(1M). Unlike other filesystems, ZFS honors the 168 * request by waiting for all pools to commit all dirty data. 169 */ 170 spa_sync_allpools(); 171 } 172 173 return (0); 174} 175 176#ifndef __FreeBSD_kernel__ 177static int 178zfs_create_unique_device(dev_t *dev) 179{ 180 major_t new_major; 181 182 do { 183 ASSERT3U(zfs_minor, <=, MAXMIN32); 184 minor_t start = zfs_minor; 185 do { 186 mutex_enter(&zfs_dev_mtx); 187 if (zfs_minor >= MAXMIN32) { 188 /* 189 * If we're still using the real major 190 * keep out of /dev/zfs and /dev/zvol minor 191 * number space. If we're using a getudev()'ed 192 * major number, we can use all of its minors. 193 */ 194 if (zfs_major == ddi_name_to_major(ZFS_DRIVER)) 195 zfs_minor = ZFS_MIN_MINOR; 196 else 197 zfs_minor = 0; 198 } else { 199 zfs_minor++; 200 } 201 *dev = makedevice(zfs_major, zfs_minor); 202 mutex_exit(&zfs_dev_mtx); 203 } while (vfs_devismounted(*dev) && zfs_minor != start); 204 if (zfs_minor == start) { 205 /* 206 * We are using all ~262,000 minor numbers for the 207 * current major number. Create a new major number. 208 */ 209 if ((new_major = getudev()) == (major_t)-1) { 210 cmn_err(CE_WARN, 211 "zfs_mount: Can't get unique major " 212 "device number."); 213 return (-1); 214 } 215 mutex_enter(&zfs_dev_mtx); 216 zfs_major = new_major; 217 zfs_minor = 0; 218 219 mutex_exit(&zfs_dev_mtx); 220 } else { 221 break; 222 } 223 /* CONSTANTCONDITION */ 224 } while (1); 225 226 return (0); 227} 228#endif /* !__FreeBSD_kernel__ */ 229 230static void 231atime_changed_cb(void *arg, uint64_t newval) 232{ 233 zfsvfs_t *zfsvfs = arg; 234 235 if (newval == TRUE) { 236 zfsvfs->z_atime = TRUE; 237 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; 238 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); 239 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); 240 } else { 241 zfsvfs->z_atime = FALSE; 242 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; 243 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); 244 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); 245 } 246} 247 248static void 249xattr_changed_cb(void *arg, uint64_t newval) 250{ 251 zfsvfs_t *zfsvfs = arg; 252 253 if (newval == TRUE) { 254 /* XXX locking on vfs_flag? */ 255#ifdef TODO 256 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; 257#endif 258 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); 259 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); 260 } else { 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_XATTR); 266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); 267 } 268} 269 270static void 271blksz_changed_cb(void *arg, uint64_t newval) 272{ 273 zfsvfs_t *zfsvfs = arg; 274 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os))); 275 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); 276 ASSERT(ISP2(newval)); 277 278 zfsvfs->z_max_blksz = newval; 279 zfsvfs->z_vfs->mnt_stat.f_iosize = newval; 280} 281 282static void 283readonly_changed_cb(void *arg, uint64_t newval) 284{ 285 zfsvfs_t *zfsvfs = arg; 286 287 if (newval) { 288 /* XXX locking on vfs_flag? */ 289 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; 290 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); 291 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); 292 } else { 293 /* XXX locking on vfs_flag? */ 294 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 295 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); 296 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); 297 } 298} 299 300static void 301setuid_changed_cb(void *arg, uint64_t newval) 302{ 303 zfsvfs_t *zfsvfs = arg; 304 305 if (newval == FALSE) { 306 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; 307 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); 308 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); 309 } else { 310 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; 311 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); 312 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); 313 } 314} 315 316static void 317exec_changed_cb(void *arg, uint64_t newval) 318{ 319 zfsvfs_t *zfsvfs = arg; 320 321 if (newval == FALSE) { 322 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; 323 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); 324 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); 325 } else { 326 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; 327 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); 328 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); 329 } 330} 331 332/* 333 * The nbmand mount option can be changed at mount time. 334 * We can't allow it to be toggled on live file systems or incorrect 335 * behavior may be seen from cifs clients 336 * 337 * This property isn't registered via dsl_prop_register(), but this callback 338 * will be called when a file system is first mounted 339 */ 340static void 341nbmand_changed_cb(void *arg, uint64_t newval) 342{ 343 zfsvfs_t *zfsvfs = arg; 344 if (newval == FALSE) { 345 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); 346 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); 347 } else { 348 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); 349 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); 350 } 351} 352 353static void 354snapdir_changed_cb(void *arg, uint64_t newval) 355{ 356 zfsvfs_t *zfsvfs = arg; 357 358 zfsvfs->z_show_ctldir = newval; 359} 360 361static void 362vscan_changed_cb(void *arg, uint64_t newval) 363{ 364 zfsvfs_t *zfsvfs = arg; 365 366 zfsvfs->z_vscan = newval; 367} 368 369static void 370acl_mode_changed_cb(void *arg, uint64_t newval) 371{ 372 zfsvfs_t *zfsvfs = arg; 373 374 zfsvfs->z_acl_mode = newval; 375} 376 377static void 378acl_inherit_changed_cb(void *arg, uint64_t newval) 379{ 380 zfsvfs_t *zfsvfs = arg; 381 382 zfsvfs->z_acl_inherit = newval; 383} 384 385static int 386zfs_register_callbacks(vfs_t *vfsp) 387{ 388 struct dsl_dataset *ds = NULL; 389 objset_t *os = NULL; 390 zfsvfs_t *zfsvfs = NULL; 391 uint64_t nbmand; 392 boolean_t readonly = B_FALSE; 393 boolean_t do_readonly = B_FALSE; 394 boolean_t setuid = B_FALSE; 395 boolean_t do_setuid = B_FALSE; 396 boolean_t exec = B_FALSE; 397 boolean_t do_exec = B_FALSE; 398#ifdef illumos 399 boolean_t devices = B_FALSE; 400 boolean_t do_devices = B_FALSE; 401#endif 402 boolean_t xattr = B_FALSE; 403 boolean_t do_xattr = B_FALSE; 404 boolean_t atime = B_FALSE; 405 boolean_t do_atime = B_FALSE; 406 int error = 0; 407 408 ASSERT(vfsp); 409 zfsvfs = vfsp->vfs_data; 410 ASSERT(zfsvfs); 411 os = zfsvfs->z_os; 412 413 /* 414 * This function can be called for a snapshot when we update snapshot's 415 * mount point, which isn't really supported. 416 */ 417 if (dmu_objset_is_snapshot(os)) 418 return (EOPNOTSUPP); 419 420 /* 421 * The act of registering our callbacks will destroy any mount 422 * options we may have. In order to enable temporary overrides 423 * of mount options, we stash away the current values and 424 * restore them after we register the callbacks. 425 */ 426 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) || 427 !spa_writeable(dmu_objset_spa(os))) { 428 readonly = B_TRUE; 429 do_readonly = B_TRUE; 430 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 431 readonly = B_FALSE; 432 do_readonly = B_TRUE; 433 } 434 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 435 setuid = B_FALSE; 436 do_setuid = B_TRUE; 437 } else { 438 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 439 setuid = B_FALSE; 440 do_setuid = B_TRUE; 441 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 442 setuid = B_TRUE; 443 do_setuid = B_TRUE; 444 } 445 } 446 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 447 exec = B_FALSE; 448 do_exec = B_TRUE; 449 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 450 exec = B_TRUE; 451 do_exec = B_TRUE; 452 } 453 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 454 xattr = B_FALSE; 455 do_xattr = B_TRUE; 456 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 457 xattr = B_TRUE; 458 do_xattr = B_TRUE; 459 } 460 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 461 atime = B_FALSE; 462 do_atime = B_TRUE; 463 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 464 atime = B_TRUE; 465 do_atime = B_TRUE; 466 } 467 468 /* 469 * We need to enter pool configuration here, so that we can use 470 * dsl_prop_get_int_ds() to handle the special nbmand property below. 471 * dsl_prop_get_integer() can not be used, because it has to acquire 472 * spa_namespace_lock and we can not do that because we already hold 473 * z_teardown_lock. The problem is that spa_config_sync() is called 474 * with spa_namespace_lock held and the function calls ZFS vnode 475 * operations to write the cache file and thus z_teardown_lock is 476 * acquired after spa_namespace_lock. 477 */ 478 ds = dmu_objset_ds(os); 479 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 480 481 /* 482 * nbmand is a special property. It can only be changed at 483 * mount time. 484 * 485 * This is weird, but it is documented to only be changeable 486 * at mount time. 487 */ 488 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 489 nbmand = B_FALSE; 490 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 491 nbmand = B_TRUE; 492 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) { 493 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 494 return (error); 495 } 496 497 /* 498 * Register property callbacks. 499 * 500 * It would probably be fine to just check for i/o error from 501 * the first prop_register(), but I guess I like to go 502 * overboard... 503 */ 504 error = dsl_prop_register(ds, 505 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 506 error = error ? error : dsl_prop_register(ds, 507 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 508 error = error ? error : dsl_prop_register(ds, 509 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 510 error = error ? error : dsl_prop_register(ds, 511 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 512#ifdef illumos 513 error = error ? error : dsl_prop_register(ds, 514 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 515#endif 516 error = error ? error : dsl_prop_register(ds, 517 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 518 error = error ? error : dsl_prop_register(ds, 519 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 520 error = error ? error : dsl_prop_register(ds, 521 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 522 error = error ? error : dsl_prop_register(ds, 523 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 524 error = error ? error : dsl_prop_register(ds, 525 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 526 zfsvfs); 527 error = error ? error : dsl_prop_register(ds, 528 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); 529 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 530 if (error) 531 goto unregister; 532 533 /* 534 * Invoke our callbacks to restore temporary mount options. 535 */ 536 if (do_readonly) 537 readonly_changed_cb(zfsvfs, readonly); 538 if (do_setuid) 539 setuid_changed_cb(zfsvfs, setuid); 540 if (do_exec) 541 exec_changed_cb(zfsvfs, exec); 542 if (do_xattr) 543 xattr_changed_cb(zfsvfs, xattr); 544 if (do_atime) 545 atime_changed_cb(zfsvfs, atime); 546 547 nbmand_changed_cb(zfsvfs, nbmand); 548 549 return (0); 550 551unregister: 552 dsl_prop_unregister_all(ds, zfsvfs); 553 return (error); 554} 555 556static int 557zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, 558 uint64_t *userp, uint64_t *groupp) 559{ 560 /* 561 * Is it a valid type of object to track? 562 */ 563 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 564 return (SET_ERROR(ENOENT)); 565 566 /* 567 * If we have a NULL data pointer 568 * then assume the id's aren't changing and 569 * return EEXIST to the dmu to let it know to 570 * use the same ids 571 */ 572 if (data == NULL) 573 return (SET_ERROR(EEXIST)); 574 575 if (bonustype == DMU_OT_ZNODE) { 576 znode_phys_t *znp = data; 577 *userp = znp->zp_uid; 578 *groupp = znp->zp_gid; 579 } else { 580 int hdrsize; 581 sa_hdr_phys_t *sap = data; 582 sa_hdr_phys_t sa = *sap; 583 boolean_t swap = B_FALSE; 584 585 ASSERT(bonustype == DMU_OT_SA); 586 587 if (sa.sa_magic == 0) { 588 /* 589 * This should only happen for newly created 590 * files that haven't had the znode data filled 591 * in yet. 592 */ 593 *userp = 0; 594 *groupp = 0; 595 return (0); 596 } 597 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { 598 sa.sa_magic = SA_MAGIC; 599 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); 600 swap = B_TRUE; 601 } else { 602 VERIFY3U(sa.sa_magic, ==, SA_MAGIC); 603 } 604 605 hdrsize = sa_hdrsize(&sa); 606 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); 607 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + 608 SA_UID_OFFSET)); 609 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + 610 SA_GID_OFFSET)); 611 if (swap) { 612 *userp = BSWAP_64(*userp); 613 *groupp = BSWAP_64(*groupp); 614 } 615 } 616 return (0); 617} 618 619static void 620fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, 621 char *domainbuf, int buflen, uid_t *ridp) 622{ 623 uint64_t fuid; 624 const char *domain; 625 626 fuid = strtonum(fuidstr, NULL); 627 628 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); 629 if (domain) 630 (void) strlcpy(domainbuf, domain, buflen); 631 else 632 domainbuf[0] = '\0'; 633 *ridp = FUID_RID(fuid); 634} 635 636static uint64_t 637zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) 638{ 639 switch (type) { 640 case ZFS_PROP_USERUSED: 641 return (DMU_USERUSED_OBJECT); 642 case ZFS_PROP_GROUPUSED: 643 return (DMU_GROUPUSED_OBJECT); 644 case ZFS_PROP_USERQUOTA: 645 return (zfsvfs->z_userquota_obj); 646 case ZFS_PROP_GROUPQUOTA: 647 return (zfsvfs->z_groupquota_obj); 648 } 649 return (0); 650} 651 652int 653zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 654 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) 655{ 656 int error; 657 zap_cursor_t zc; 658 zap_attribute_t za; 659 zfs_useracct_t *buf = vbuf; 660 uint64_t obj; 661 662 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 663 return (SET_ERROR(ENOTSUP)); 664 665 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 666 if (obj == 0) { 667 *bufsizep = 0; 668 return (0); 669 } 670 671 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); 672 (error = zap_cursor_retrieve(&zc, &za)) == 0; 673 zap_cursor_advance(&zc)) { 674 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > 675 *bufsizep) 676 break; 677 678 fuidstr_to_sid(zfsvfs, za.za_name, 679 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); 680 681 buf->zu_space = za.za_first_integer; 682 buf++; 683 } 684 if (error == ENOENT) 685 error = 0; 686 687 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); 688 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; 689 *cookiep = zap_cursor_serialize(&zc); 690 zap_cursor_fini(&zc); 691 return (error); 692} 693 694/* 695 * buf must be big enough (eg, 32 bytes) 696 */ 697static int 698id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, 699 char *buf, boolean_t addok) 700{ 701 uint64_t fuid; 702 int domainid = 0; 703 704 if (domain && domain[0]) { 705 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); 706 if (domainid == -1) 707 return (SET_ERROR(ENOENT)); 708 } 709 fuid = FUID_ENCODE(domainid, rid); 710 (void) sprintf(buf, "%llx", (longlong_t)fuid); 711 return (0); 712} 713 714int 715zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 716 const char *domain, uint64_t rid, uint64_t *valp) 717{ 718 char buf[32]; 719 int err; 720 uint64_t obj; 721 722 *valp = 0; 723 724 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 725 return (SET_ERROR(ENOTSUP)); 726 727 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 728 if (obj == 0) 729 return (0); 730 731 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); 732 if (err) 733 return (err); 734 735 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); 736 if (err == ENOENT) 737 err = 0; 738 return (err); 739} 740 741int 742zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 743 const char *domain, uint64_t rid, uint64_t quota) 744{ 745 char buf[32]; 746 int err; 747 dmu_tx_t *tx; 748 uint64_t *objp; 749 boolean_t fuid_dirtied; 750 751 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) 752 return (SET_ERROR(EINVAL)); 753 754 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) 755 return (SET_ERROR(ENOTSUP)); 756 757 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : 758 &zfsvfs->z_groupquota_obj; 759 760 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); 761 if (err) 762 return (err); 763 fuid_dirtied = zfsvfs->z_fuid_dirty; 764 765 tx = dmu_tx_create(zfsvfs->z_os); 766 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); 767 if (*objp == 0) { 768 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 769 zfs_userquota_prop_prefixes[type]); 770 } 771 if (fuid_dirtied) 772 zfs_fuid_txhold(zfsvfs, tx); 773 err = dmu_tx_assign(tx, TXG_WAIT); 774 if (err) { 775 dmu_tx_abort(tx); 776 return (err); 777 } 778 779 mutex_enter(&zfsvfs->z_lock); 780 if (*objp == 0) { 781 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, 782 DMU_OT_NONE, 0, tx); 783 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 784 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); 785 } 786 mutex_exit(&zfsvfs->z_lock); 787 788 if (quota == 0) { 789 err = zap_remove(zfsvfs->z_os, *objp, buf, tx); 790 if (err == ENOENT) 791 err = 0; 792 } else { 793 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); 794 } 795 ASSERT(err == 0); 796 if (fuid_dirtied) 797 zfs_fuid_sync(zfsvfs, tx); 798 dmu_tx_commit(tx); 799 return (err); 800} 801 802boolean_t 803zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) 804{ 805 char buf[32]; 806 uint64_t used, quota, usedobj, quotaobj; 807 int err; 808 809 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; 810 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 811 812 if (quotaobj == 0 || zfsvfs->z_replay) 813 return (B_FALSE); 814 815 (void) sprintf(buf, "%llx", (longlong_t)fuid); 816 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); 817 if (err != 0) 818 return (B_FALSE); 819 820 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); 821 if (err != 0) 822 return (B_FALSE); 823 return (used >= quota); 824} 825 826boolean_t 827zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) 828{ 829 uint64_t fuid; 830 uint64_t quotaobj; 831 832 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 833 834 fuid = isgroup ? zp->z_gid : zp->z_uid; 835 836 if (quotaobj == 0 || zfsvfs->z_replay) 837 return (B_FALSE); 838 839 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); 840} 841 842int 843zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 844{ 845 objset_t *os; 846 zfsvfs_t *zfsvfs; 847 uint64_t zval; 848 int i, error; 849 uint64_t sa_obj; 850 851 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 852 853 /* 854 * We claim to always be readonly so we can open snapshots; 855 * other ZPL code will prevent us from writing to snapshots. 856 */ 857 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 858 if (error) { 859 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 860 return (error); 861 } 862 863 /* 864 * Initialize the zfs-specific filesystem structure. 865 * Should probably make this a kmem cache, shuffle fields, 866 * and just bzero up to z_hold_mtx[]. 867 */ 868 zfsvfs->z_vfs = NULL; 869 zfsvfs->z_parent = zfsvfs; 870 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 871 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 872 zfsvfs->z_os = os; 873 874 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 875 if (error) { 876 goto out; 877 } else if (zfsvfs->z_version > 878 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 879 (void) printf("Can't mount a version %lld file system " 880 "on a version %lld pool\n. Pool must be upgraded to mount " 881 "this file system.", (u_longlong_t)zfsvfs->z_version, 882 (u_longlong_t)spa_version(dmu_objset_spa(os))); 883 error = SET_ERROR(ENOTSUP); 884 goto out; 885 } 886 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) 887 goto out; 888 zfsvfs->z_norm = (int)zval; 889 890 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) 891 goto out; 892 zfsvfs->z_utf8 = (zval != 0); 893 894 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) 895 goto out; 896 zfsvfs->z_case = (uint_t)zval; 897 898 /* 899 * Fold case on file systems that are always or sometimes case 900 * insensitive. 901 */ 902 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 903 zfsvfs->z_case == ZFS_CASE_MIXED) 904 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 905 906 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 907 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 908 909 if (zfsvfs->z_use_sa) { 910 /* should either have both of these objects or none */ 911 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 912 &sa_obj); 913 if (error) 914 goto out; 915 } else { 916 /* 917 * Pre SA versions file systems should never touch 918 * either the attribute registration or layout objects. 919 */ 920 sa_obj = 0; 921 } 922 923 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 924 &zfsvfs->z_attr_table); 925 if (error) 926 goto out; 927 928 if (zfsvfs->z_version >= ZPL_VERSION_SA) 929 sa_register_update_callback(os, zfs_sa_upgrade); 930 931 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 932 &zfsvfs->z_root); 933 if (error) 934 goto out; 935 ASSERT(zfsvfs->z_root != 0); 936 937 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 938 &zfsvfs->z_unlinkedobj); 939 if (error) 940 goto out; 941 942 error = zap_lookup(os, MASTER_NODE_OBJ, 943 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 944 8, 1, &zfsvfs->z_userquota_obj); 945 if (error && error != ENOENT) 946 goto out; 947 948 error = zap_lookup(os, MASTER_NODE_OBJ, 949 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 950 8, 1, &zfsvfs->z_groupquota_obj); 951 if (error && error != ENOENT) 952 goto out; 953 954 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 955 &zfsvfs->z_fuid_obj); 956 if (error && error != ENOENT) 957 goto out; 958 959 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 960 &zfsvfs->z_shares_dir); 961 if (error && error != ENOENT) 962 goto out; 963 964 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 965 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 966 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 967 offsetof(znode_t, z_link_node)); 968 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE); 969 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 970 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 971 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 972 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 973 974 *zfvp = zfsvfs; 975 return (0); 976 977out: 978 dmu_objset_disown(os, zfsvfs); 979 *zfvp = NULL; 980 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 981 return (error); 982} 983 984static int 985zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 986{ 987 int error; 988 989 error = zfs_register_callbacks(zfsvfs->z_vfs); 990 if (error) 991 return (error); 992 993 /* 994 * Set the objset user_ptr to track its zfsvfs. 995 */ 996 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 997 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 998 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 999 1000 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 1001 1002 /* 1003 * If we are not mounting (ie: online recv), then we don't 1004 * have to worry about replaying the log as we blocked all 1005 * operations out since we closed the ZIL. 1006 */ 1007 if (mounting) { 1008 boolean_t readonly; 1009 1010 /* 1011 * During replay we remove the read only flag to 1012 * allow replays to succeed. 1013 */ 1014 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1015 if (readonly != 0) 1016 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1017 else 1018 zfs_unlinked_drain(zfsvfs); 1019 1020 /* 1021 * Parse and replay the intent log. 1022 * 1023 * Because of ziltest, this must be done after 1024 * zfs_unlinked_drain(). (Further note: ziltest 1025 * doesn't use readonly mounts, where 1026 * zfs_unlinked_drain() isn't called.) This is because 1027 * ziltest causes spa_sync() to think it's committed, 1028 * but actually it is not, so the intent log contains 1029 * many txg's worth of changes. 1030 * 1031 * In particular, if object N is in the unlinked set in 1032 * the last txg to actually sync, then it could be 1033 * actually freed in a later txg and then reallocated 1034 * in a yet later txg. This would write a "create 1035 * object N" record to the intent log. Normally, this 1036 * would be fine because the spa_sync() would have 1037 * written out the fact that object N is free, before 1038 * we could write the "create object N" intent log 1039 * record. 1040 * 1041 * But when we are in ziltest mode, we advance the "open 1042 * txg" without actually spa_sync()-ing the changes to 1043 * disk. So we would see that object N is still 1044 * allocated and in the unlinked set, and there is an 1045 * intent log record saying to allocate it. 1046 */ 1047 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1048 if (zil_replay_disable) { 1049 zil_destroy(zfsvfs->z_log, B_FALSE); 1050 } else { 1051 zfsvfs->z_replay = B_TRUE; 1052 zil_replay(zfsvfs->z_os, zfsvfs, 1053 zfs_replay_vector); 1054 zfsvfs->z_replay = B_FALSE; 1055 } 1056 } 1057 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1058 } 1059 1060 return (0); 1061} 1062 1063extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1064 1065void 1066zfsvfs_free(zfsvfs_t *zfsvfs) 1067{ 1068 int i; 1069 1070 /* 1071 * This is a barrier to prevent the filesystem from going away in 1072 * zfs_znode_move() until we can safely ensure that the filesystem is 1073 * not unmounted. We consider the filesystem valid before the barrier 1074 * and invalid after the barrier. 1075 */ 1076 rw_enter(&zfsvfs_lock, RW_READER); 1077 rw_exit(&zfsvfs_lock); 1078 1079 zfs_fuid_destroy(zfsvfs); 1080 1081 mutex_destroy(&zfsvfs->z_znodes_lock); 1082 mutex_destroy(&zfsvfs->z_lock); 1083 list_destroy(&zfsvfs->z_all_znodes); 1084 rrm_destroy(&zfsvfs->z_teardown_lock); 1085 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1086 rw_destroy(&zfsvfs->z_fuid_lock); 1087 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1088 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1089 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1090} 1091 1092static void 1093zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1094{ 1095 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1096 if (zfsvfs->z_vfs) { 1097 if (zfsvfs->z_use_fuids) { 1098 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1099 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1100 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1101 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1102 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1103 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1104 } else { 1105 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1106 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1107 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1108 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1109 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1110 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1111 } 1112 } 1113 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1114} 1115 1116static int 1117zfs_domount(vfs_t *vfsp, char *osname) 1118{ 1119 uint64_t recordsize, fsid_guid; 1120 int error = 0; 1121 zfsvfs_t *zfsvfs; 1122 vnode_t *vp; 1123 1124 ASSERT(vfsp); 1125 ASSERT(osname); 1126 1127 error = zfsvfs_create(osname, &zfsvfs); 1128 if (error) 1129 return (error); 1130 zfsvfs->z_vfs = vfsp; 1131 1132#ifdef illumos 1133 /* Initialize the generic filesystem structure. */ 1134 vfsp->vfs_bcount = 0; 1135 vfsp->vfs_data = NULL; 1136 1137 if (zfs_create_unique_device(&mount_dev) == -1) { 1138 error = SET_ERROR(ENODEV); 1139 goto out; 1140 } 1141 ASSERT(vfs_devismounted(mount_dev) == 0); 1142#endif 1143 1144 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1145 NULL)) 1146 goto out; 1147 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1148 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1149 1150 vfsp->vfs_data = zfsvfs; 1151 vfsp->mnt_flag |= MNT_LOCAL; 1152 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1153 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1154 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1155 1156 /* 1157 * The fsid is 64 bits, composed of an 8-bit fs type, which 1158 * separates our fsid from any other filesystem types, and a 1159 * 56-bit objset unique ID. The objset unique ID is unique to 1160 * all objsets open on this system, provided by unique_create(). 1161 * The 8-bit fs type must be put in the low bits of fsid[1] 1162 * because that's where other Solaris filesystems put it. 1163 */ 1164 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1165 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1166 vfsp->vfs_fsid.val[0] = fsid_guid; 1167 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1168 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1169 1170 /* 1171 * Set features for file system. 1172 */ 1173 zfs_set_fuid_feature(zfsvfs); 1174 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1175 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1176 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1177 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1178 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1179 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1180 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1181 } 1182 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1183 1184 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1185 uint64_t pval; 1186 1187 atime_changed_cb(zfsvfs, B_FALSE); 1188 readonly_changed_cb(zfsvfs, B_TRUE); 1189 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1190 goto out; 1191 xattr_changed_cb(zfsvfs, pval); 1192 zfsvfs->z_issnap = B_TRUE; 1193 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1194 1195 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1196 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1197 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1198 } else { 1199 error = zfsvfs_setup(zfsvfs, B_TRUE); 1200 } 1201 1202 vfs_mountedfrom(vfsp, osname); 1203 1204 if (!zfsvfs->z_issnap) 1205 zfsctl_create(zfsvfs); 1206out: 1207 if (error) { 1208 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1209 zfsvfs_free(zfsvfs); 1210 } else { 1211 atomic_inc_32(&zfs_active_fs_count); 1212 } 1213 1214 return (error); 1215} 1216 1217void 1218zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1219{ 1220 objset_t *os = zfsvfs->z_os; 1221 1222 if (!dmu_objset_is_snapshot(os)) 1223 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 1224} 1225 1226#ifdef SECLABEL 1227/* 1228 * Convert a decimal digit string to a uint64_t integer. 1229 */ 1230static int 1231str_to_uint64(char *str, uint64_t *objnum) 1232{ 1233 uint64_t num = 0; 1234 1235 while (*str) { 1236 if (*str < '0' || *str > '9') 1237 return (SET_ERROR(EINVAL)); 1238 1239 num = num*10 + *str++ - '0'; 1240 } 1241 1242 *objnum = num; 1243 return (0); 1244} 1245 1246/* 1247 * The boot path passed from the boot loader is in the form of 1248 * "rootpool-name/root-filesystem-object-number'. Convert this 1249 * string to a dataset name: "rootpool-name/root-filesystem-name". 1250 */ 1251static int 1252zfs_parse_bootfs(char *bpath, char *outpath) 1253{ 1254 char *slashp; 1255 uint64_t objnum; 1256 int error; 1257 1258 if (*bpath == 0 || *bpath == '/') 1259 return (SET_ERROR(EINVAL)); 1260 1261 (void) strcpy(outpath, bpath); 1262 1263 slashp = strchr(bpath, '/'); 1264 1265 /* if no '/', just return the pool name */ 1266 if (slashp == NULL) { 1267 return (0); 1268 } 1269 1270 /* if not a number, just return the root dataset name */ 1271 if (str_to_uint64(slashp+1, &objnum)) { 1272 return (0); 1273 } 1274 1275 *slashp = '\0'; 1276 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1277 *slashp = '/'; 1278 1279 return (error); 1280} 1281 1282/* 1283 * Check that the hex label string is appropriate for the dataset being 1284 * mounted into the global_zone proper. 1285 * 1286 * Return an error if the hex label string is not default or 1287 * admin_low/admin_high. For admin_low labels, the corresponding 1288 * dataset must be readonly. 1289 */ 1290int 1291zfs_check_global_label(const char *dsname, const char *hexsl) 1292{ 1293 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1294 return (0); 1295 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1296 return (0); 1297 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1298 /* must be readonly */ 1299 uint64_t rdonly; 1300 1301 if (dsl_prop_get_integer(dsname, 1302 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1303 return (SET_ERROR(EACCES)); 1304 return (rdonly ? 0 : EACCES); 1305 } 1306 return (SET_ERROR(EACCES)); 1307} 1308 1309/* 1310 * Determine whether the mount is allowed according to MAC check. 1311 * by comparing (where appropriate) label of the dataset against 1312 * the label of the zone being mounted into. If the dataset has 1313 * no label, create one. 1314 * 1315 * Returns 0 if access allowed, error otherwise (e.g. EACCES) 1316 */ 1317static int 1318zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1319{ 1320 int error, retv; 1321 zone_t *mntzone = NULL; 1322 ts_label_t *mnt_tsl; 1323 bslabel_t *mnt_sl; 1324 bslabel_t ds_sl; 1325 char ds_hexsl[MAXNAMELEN]; 1326 1327 retv = EACCES; /* assume the worst */ 1328 1329 /* 1330 * Start by getting the dataset label if it exists. 1331 */ 1332 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1333 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1334 if (error) 1335 return (SET_ERROR(EACCES)); 1336 1337 /* 1338 * If labeling is NOT enabled, then disallow the mount of datasets 1339 * which have a non-default label already. No other label checks 1340 * are needed. 1341 */ 1342 if (!is_system_labeled()) { 1343 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1344 return (0); 1345 return (SET_ERROR(EACCES)); 1346 } 1347 1348 /* 1349 * Get the label of the mountpoint. If mounting into the global 1350 * zone (i.e. mountpoint is not within an active zone and the 1351 * zoned property is off), the label must be default or 1352 * admin_low/admin_high only; no other checks are needed. 1353 */ 1354 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1355 if (mntzone->zone_id == GLOBAL_ZONEID) { 1356 uint64_t zoned; 1357 1358 zone_rele(mntzone); 1359 1360 if (dsl_prop_get_integer(osname, 1361 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1362 return (SET_ERROR(EACCES)); 1363 if (!zoned) 1364 return (zfs_check_global_label(osname, ds_hexsl)); 1365 else 1366 /* 1367 * This is the case of a zone dataset being mounted 1368 * initially, before the zone has been fully created; 1369 * allow this mount into global zone. 1370 */ 1371 return (0); 1372 } 1373 1374 mnt_tsl = mntzone->zone_slabel; 1375 ASSERT(mnt_tsl != NULL); 1376 label_hold(mnt_tsl); 1377 mnt_sl = label2bslabel(mnt_tsl); 1378 1379 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1380 /* 1381 * The dataset doesn't have a real label, so fabricate one. 1382 */ 1383 char *str = NULL; 1384 1385 if (l_to_str_internal(mnt_sl, &str) == 0 && 1386 dsl_prop_set_string(osname, 1387 zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1388 ZPROP_SRC_LOCAL, str) == 0) 1389 retv = 0; 1390 if (str != NULL) 1391 kmem_free(str, strlen(str) + 1); 1392 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1393 /* 1394 * Now compare labels to complete the MAC check. If the 1395 * labels are equal then allow access. If the mountpoint 1396 * label dominates the dataset label, allow readonly access. 1397 * Otherwise, access is denied. 1398 */ 1399 if (blequal(mnt_sl, &ds_sl)) 1400 retv = 0; 1401 else if (bldominates(mnt_sl, &ds_sl)) { 1402 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1403 retv = 0; 1404 } 1405 } 1406 1407 label_rele(mnt_tsl); 1408 zone_rele(mntzone); 1409 return (retv); 1410} 1411#endif /* SECLABEL */ 1412 1413#ifdef OPENSOLARIS_MOUNTROOT 1414static int 1415zfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1416{ 1417 int error = 0; 1418 static int zfsrootdone = 0; 1419 zfsvfs_t *zfsvfs = NULL; 1420 znode_t *zp = NULL; 1421 vnode_t *vp = NULL; 1422 char *zfs_bootfs; 1423 char *zfs_devid; 1424 1425 ASSERT(vfsp); 1426 1427 /* 1428 * The filesystem that we mount as root is defined in the 1429 * boot property "zfs-bootfs" with a format of 1430 * "poolname/root-dataset-objnum". 1431 */ 1432 if (why == ROOT_INIT) { 1433 if (zfsrootdone++) 1434 return (SET_ERROR(EBUSY)); 1435 /* 1436 * the process of doing a spa_load will require the 1437 * clock to be set before we could (for example) do 1438 * something better by looking at the timestamp on 1439 * an uberblock, so just set it to -1. 1440 */ 1441 clkset(-1); 1442 1443 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1444 cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1445 "bootfs name"); 1446 return (SET_ERROR(EINVAL)); 1447 } 1448 zfs_devid = spa_get_bootprop("diskdevid"); 1449 error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1450 if (zfs_devid) 1451 spa_free_bootprop(zfs_devid); 1452 if (error) { 1453 spa_free_bootprop(zfs_bootfs); 1454 cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1455 error); 1456 return (error); 1457 } 1458 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1459 spa_free_bootprop(zfs_bootfs); 1460 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1461 error); 1462 return (error); 1463 } 1464 1465 spa_free_bootprop(zfs_bootfs); 1466 1467 if (error = vfs_lock(vfsp)) 1468 return (error); 1469 1470 if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1471 cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1472 goto out; 1473 } 1474 1475 zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1476 ASSERT(zfsvfs); 1477 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1478 cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1479 goto out; 1480 } 1481 1482 vp = ZTOV(zp); 1483 mutex_enter(&vp->v_lock); 1484 vp->v_flag |= VROOT; 1485 mutex_exit(&vp->v_lock); 1486 rootvp = vp; 1487 1488 /* 1489 * Leave rootvp held. The root file system is never unmounted. 1490 */ 1491 1492 vfs_add((struct vnode *)0, vfsp, 1493 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1494out: 1495 vfs_unlock(vfsp); 1496 return (error); 1497 } else if (why == ROOT_REMOUNT) { 1498 readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1499 vfsp->vfs_flag |= VFS_REMOUNT; 1500 1501 /* refresh mount options */ 1502 zfs_unregister_callbacks(vfsp->vfs_data); 1503 return (zfs_register_callbacks(vfsp)); 1504 1505 } else if (why == ROOT_UNMOUNT) { 1506 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1507 (void) zfs_sync(vfsp, 0, 0); 1508 return (0); 1509 } 1510 1511 /* 1512 * if "why" is equal to anything else other than ROOT_INIT, 1513 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1514 */ 1515 return (SET_ERROR(ENOTSUP)); 1516} 1517#endif /* OPENSOLARIS_MOUNTROOT */ 1518 1519static int 1520getpoolname(const char *osname, char *poolname) 1521{ 1522 char *p; 1523 1524 p = strchr(osname, '/'); 1525 if (p == NULL) { 1526 if (strlen(osname) >= MAXNAMELEN) 1527 return (ENAMETOOLONG); 1528 (void) strcpy(poolname, osname); 1529 } else { 1530 if (p - osname >= MAXNAMELEN) 1531 return (ENAMETOOLONG); 1532 (void) strncpy(poolname, osname, p - osname); 1533 poolname[p - osname] = '\0'; 1534 } 1535 return (0); 1536} 1537 1538/*ARGSUSED*/ 1539static int 1540zfs_mount(vfs_t *vfsp) 1541{ 1542 kthread_t *td = curthread; 1543 vnode_t *mvp = vfsp->mnt_vnodecovered; 1544 cred_t *cr = td->td_ucred; 1545 char *osname; 1546 int error = 0; 1547 int canwrite; 1548 1549#ifdef illumos 1550 if (mvp->v_type != VDIR) 1551 return (SET_ERROR(ENOTDIR)); 1552 1553 mutex_enter(&mvp->v_lock); 1554 if ((uap->flags & MS_REMOUNT) == 0 && 1555 (uap->flags & MS_OVERLAY) == 0 && 1556 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 1557 mutex_exit(&mvp->v_lock); 1558 return (SET_ERROR(EBUSY)); 1559 } 1560 mutex_exit(&mvp->v_lock); 1561 1562 /* 1563 * ZFS does not support passing unparsed data in via MS_DATA. 1564 * Users should use the MS_OPTIONSTR interface; this means 1565 * that all option parsing is already done and the options struct 1566 * can be interrogated. 1567 */ 1568 if ((uap->flags & MS_DATA) && uap->datalen > 0) 1569#else /* !illumos */ 1570 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1571 return (SET_ERROR(EPERM)); 1572 1573 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1574 return (SET_ERROR(EINVAL)); 1575#endif /* illumos */ 1576 1577 /* 1578 * If full-owner-access is enabled and delegated administration is 1579 * turned on, we must set nosuid. 1580 */ 1581 if (zfs_super_owner && 1582 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1583 secpolicy_fs_mount_clearopts(cr, vfsp); 1584 } 1585 1586 /* 1587 * Check for mount privilege? 1588 * 1589 * If we don't have privilege then see if 1590 * we have local permission to allow it 1591 */ 1592 error = secpolicy_fs_mount(cr, mvp, vfsp); 1593 if (error) { 1594 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1595 goto out; 1596 1597 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1598 vattr_t vattr; 1599 1600 /* 1601 * Make sure user is the owner of the mount point 1602 * or has sufficient privileges. 1603 */ 1604 1605 vattr.va_mask = AT_UID; 1606 1607 vn_lock(mvp, LK_SHARED | LK_RETRY); 1608 if (VOP_GETATTR(mvp, &vattr, cr)) { 1609 VOP_UNLOCK(mvp, 0); 1610 goto out; 1611 } 1612 1613 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1614 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1615 VOP_UNLOCK(mvp, 0); 1616 goto out; 1617 } 1618 VOP_UNLOCK(mvp, 0); 1619 } 1620 1621 secpolicy_fs_mount_clearopts(cr, vfsp); 1622 } 1623 1624 /* 1625 * Refuse to mount a filesystem if we are in a local zone and the 1626 * dataset is not visible. 1627 */ 1628 if (!INGLOBALZONE(curthread) && 1629 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1630 error = SET_ERROR(EPERM); 1631 goto out; 1632 } 1633 1634#ifdef SECLABEL 1635 error = zfs_mount_label_policy(vfsp, osname); 1636 if (error) 1637 goto out; 1638#endif 1639 1640 vfsp->vfs_flag |= MNT_NFS4ACLS; 1641 1642 /* 1643 * When doing a remount, we simply refresh our temporary properties 1644 * according to those options set in the current VFS options. 1645 */ 1646 if (vfsp->vfs_flag & MS_REMOUNT) { 1647 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1648 1649 /* 1650 * Refresh mount options with z_teardown_lock blocking I/O while 1651 * the filesystem is in an inconsistent state. 1652 * The lock also serializes this code with filesystem 1653 * manipulations between entry to zfs_suspend_fs() and return 1654 * from zfs_resume_fs(). 1655 */ 1656 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1657 zfs_unregister_callbacks(zfsvfs); 1658 error = zfs_register_callbacks(vfsp); 1659 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1660 goto out; 1661 } 1662 1663 /* Initial root mount: try hard to import the requested root pool. */ 1664 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1665 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1666 char pname[MAXNAMELEN]; 1667 1668 error = getpoolname(osname, pname); 1669 if (error == 0) 1670 error = spa_import_rootpool(pname); 1671 if (error) 1672 goto out; 1673 } 1674 DROP_GIANT(); 1675 error = zfs_domount(vfsp, osname); 1676 PICKUP_GIANT(); 1677 1678#ifdef illumos 1679 /* 1680 * Add an extra VFS_HOLD on our parent vfs so that it can't 1681 * disappear due to a forced unmount. 1682 */ 1683 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1684 VFS_HOLD(mvp->v_vfsp); 1685#endif 1686 1687out: 1688 return (error); 1689} 1690 1691static int 1692zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1693{ 1694 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1695 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1696 1697 statp->f_version = STATFS_VERSION; 1698 1699 ZFS_ENTER(zfsvfs); 1700 1701 dmu_objset_space(zfsvfs->z_os, 1702 &refdbytes, &availbytes, &usedobjs, &availobjs); 1703 1704 /* 1705 * The underlying storage pool actually uses multiple block sizes. 1706 * We report the fragsize as the smallest block size we support, 1707 * and we report our blocksize as the filesystem's maximum blocksize. 1708 */ 1709 statp->f_bsize = SPA_MINBLOCKSIZE; 1710 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1711 1712 /* 1713 * The following report "total" blocks of various kinds in the 1714 * file system, but reported in terms of f_frsize - the 1715 * "fragment" size. 1716 */ 1717 1718 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1719 statp->f_bfree = availbytes / statp->f_bsize; 1720 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1721 1722 /* 1723 * statvfs() should really be called statufs(), because it assumes 1724 * static metadata. ZFS doesn't preallocate files, so the best 1725 * we can do is report the max that could possibly fit in f_files, 1726 * and that minus the number actually used in f_ffree. 1727 * For f_ffree, report the smaller of the number of object available 1728 * and the number of blocks (each object will take at least a block). 1729 */ 1730 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1731 statp->f_files = statp->f_ffree + usedobjs; 1732 1733 /* 1734 * We're a zfs filesystem. 1735 */ 1736 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1737 1738 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1739 sizeof(statp->f_mntfromname)); 1740 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1741 sizeof(statp->f_mntonname)); 1742 1743 statp->f_namemax = ZFS_MAXNAMELEN; 1744 1745 ZFS_EXIT(zfsvfs); 1746 return (0); 1747} 1748 1749static int 1750zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1751{ 1752 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1753 znode_t *rootzp; 1754 int error; 1755 1756 ZFS_ENTER(zfsvfs); 1757 1758 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1759 if (error == 0) 1760 *vpp = ZTOV(rootzp); 1761 1762 ZFS_EXIT(zfsvfs); 1763 1764 if (error == 0) { 1765 error = vn_lock(*vpp, flags); 1766 if (error == 0) 1767 (*vpp)->v_vflag |= VV_ROOT; 1768 } 1769 if (error != 0) 1770 *vpp = NULL; 1771 1772 return (error); 1773} 1774 1775/* 1776 * Teardown the zfsvfs::z_os. 1777 * 1778 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1779 * and 'z_teardown_inactive_lock' held. 1780 */ 1781static int 1782zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1783{ 1784 znode_t *zp; 1785 1786 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1787 1788 if (!unmounting) { 1789 /* 1790 * We purge the parent filesystem's vfsp as the parent 1791 * filesystem and all of its snapshots have their vnode's 1792 * v_vfsp set to the parent's filesystem's vfsp. Note, 1793 * 'z_parent' is self referential for non-snapshots. 1794 */ 1795 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1796#ifdef FREEBSD_NAMECACHE 1797 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1798#endif 1799 } 1800 1801 /* 1802 * Close the zil. NB: Can't close the zil while zfs_inactive 1803 * threads are blocked as zil_close can call zfs_inactive. 1804 */ 1805 if (zfsvfs->z_log) { 1806 zil_close(zfsvfs->z_log); 1807 zfsvfs->z_log = NULL; 1808 } 1809 1810 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1811 1812 /* 1813 * If we are not unmounting (ie: online recv) and someone already 1814 * unmounted this file system while we were doing the switcheroo, 1815 * or a reopen of z_os failed then just bail out now. 1816 */ 1817 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1818 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1819 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1820 return (SET_ERROR(EIO)); 1821 } 1822 1823 /* 1824 * At this point there are no vops active, and any new vops will 1825 * fail with EIO since we have z_teardown_lock for writer (only 1826 * relavent for forced unmount). 1827 * 1828 * Release all holds on dbufs. 1829 */ 1830 mutex_enter(&zfsvfs->z_znodes_lock); 1831 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1832 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1833 if (zp->z_sa_hdl) { 1834 ASSERT(ZTOV(zp)->v_count >= 0); 1835 zfs_znode_dmu_fini(zp); 1836 } 1837 mutex_exit(&zfsvfs->z_znodes_lock); 1838 1839 /* 1840 * If we are unmounting, set the unmounted flag and let new vops 1841 * unblock. zfs_inactive will have the unmounted behavior, and all 1842 * other vops will fail with EIO. 1843 */ 1844 if (unmounting) { 1845 zfsvfs->z_unmounted = B_TRUE; 1846 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1847 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1848 } 1849 1850 /* 1851 * z_os will be NULL if there was an error in attempting to reopen 1852 * zfsvfs, so just return as the properties had already been 1853 * unregistered and cached data had been evicted before. 1854 */ 1855 if (zfsvfs->z_os == NULL) 1856 return (0); 1857 1858 /* 1859 * Unregister properties. 1860 */ 1861 zfs_unregister_callbacks(zfsvfs); 1862 1863 /* 1864 * Evict cached data 1865 */ 1866 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1867 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1868 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1869 dmu_objset_evict_dbufs(zfsvfs->z_os); 1870 1871 return (0); 1872} 1873 1874/*ARGSUSED*/ 1875static int 1876zfs_umount(vfs_t *vfsp, int fflag) 1877{ 1878 kthread_t *td = curthread; 1879 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1880 objset_t *os; 1881 cred_t *cr = td->td_ucred; 1882 int ret; 1883 1884 ret = secpolicy_fs_unmount(cr, vfsp); 1885 if (ret) { 1886 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1887 ZFS_DELEG_PERM_MOUNT, cr)) 1888 return (ret); 1889 } 1890 1891 /* 1892 * We purge the parent filesystem's vfsp as the parent filesystem 1893 * and all of its snapshots have their vnode's v_vfsp set to the 1894 * parent's filesystem's vfsp. Note, 'z_parent' is self 1895 * referential for non-snapshots. 1896 */ 1897 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1898 1899 /* 1900 * Unmount any snapshots mounted under .zfs before unmounting the 1901 * dataset itself. 1902 */ 1903 if (zfsvfs->z_ctldir != NULL) { 1904 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1905 return (ret); 1906 ret = vflush(vfsp, 0, 0, td); 1907 ASSERT(ret == EBUSY); 1908 if (!(fflag & MS_FORCE)) { 1909 if (zfsvfs->z_ctldir->v_count > 1) 1910 return (EBUSY); 1911 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1912 } 1913 zfsctl_destroy(zfsvfs); 1914 ASSERT(zfsvfs->z_ctldir == NULL); 1915 } 1916 1917 if (fflag & MS_FORCE) { 1918 /* 1919 * Mark file system as unmounted before calling 1920 * vflush(FORCECLOSE). This way we ensure no future vnops 1921 * will be called and risk operating on DOOMED vnodes. 1922 */ 1923 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1924 zfsvfs->z_unmounted = B_TRUE; 1925 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1926 } 1927 1928 /* 1929 * Flush all the files. 1930 */ 1931 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1932 if (ret != 0) { 1933 if (!zfsvfs->z_issnap) { 1934 zfsctl_create(zfsvfs); 1935 ASSERT(zfsvfs->z_ctldir != NULL); 1936 } 1937 return (ret); 1938 } 1939 1940#ifdef illumos 1941 if (!(fflag & MS_FORCE)) { 1942 /* 1943 * Check the number of active vnodes in the file system. 1944 * Our count is maintained in the vfs structure, but the 1945 * number is off by 1 to indicate a hold on the vfs 1946 * structure itself. 1947 * 1948 * The '.zfs' directory maintains a reference of its 1949 * own, and any active references underneath are 1950 * reflected in the vnode count. 1951 */ 1952 if (zfsvfs->z_ctldir == NULL) { 1953 if (vfsp->vfs_count > 1) 1954 return (SET_ERROR(EBUSY)); 1955 } else { 1956 if (vfsp->vfs_count > 2 || 1957 zfsvfs->z_ctldir->v_count > 1) 1958 return (SET_ERROR(EBUSY)); 1959 } 1960 } 1961#endif 1962 1963 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1964 os = zfsvfs->z_os; 1965 1966 /* 1967 * z_os will be NULL if there was an error in 1968 * attempting to reopen zfsvfs. 1969 */ 1970 if (os != NULL) { 1971 /* 1972 * Unset the objset user_ptr. 1973 */ 1974 mutex_enter(&os->os_user_ptr_lock); 1975 dmu_objset_set_user(os, NULL); 1976 mutex_exit(&os->os_user_ptr_lock); 1977 1978 /* 1979 * Finally release the objset 1980 */ 1981 dmu_objset_disown(os, zfsvfs); 1982 } 1983 1984 /* 1985 * We can now safely destroy the '.zfs' directory node. 1986 */ 1987 if (zfsvfs->z_ctldir != NULL) 1988 zfsctl_destroy(zfsvfs); 1989 if (zfsvfs->z_issnap) { 1990 vnode_t *svp = vfsp->mnt_vnodecovered; 1991 1992 if (svp->v_count >= 2) 1993 VN_RELE(svp); 1994 } 1995 zfs_freevfs(vfsp); 1996 1997 return (0); 1998} 1999 2000static int 2001zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2002{ 2003 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2004 znode_t *zp; 2005 int err; 2006 2007 /* 2008 * zfs_zget() can't operate on virtual entries like .zfs/ or 2009 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2010 * This will make NFS to switch to LOOKUP instead of using VGET. 2011 */ 2012 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || 2013 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) 2014 return (EOPNOTSUPP); 2015 2016 ZFS_ENTER(zfsvfs); 2017 err = zfs_zget(zfsvfs, ino, &zp); 2018 if (err == 0 && zp->z_unlinked) { 2019 VN_RELE(ZTOV(zp)); 2020 err = EINVAL; 2021 } 2022 if (err == 0) 2023 *vpp = ZTOV(zp); 2024 ZFS_EXIT(zfsvfs); 2025 if (err == 0) 2026 err = vn_lock(*vpp, flags); 2027 if (err != 0) 2028 *vpp = NULL; 2029 return (err); 2030} 2031 2032static int 2033zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2034 struct ucred **credanonp, int *numsecflavors, int **secflavors) 2035{ 2036 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2037 2038 /* 2039 * If this is regular file system vfsp is the same as 2040 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2041 * zfsvfs->z_parent->z_vfs represents parent file system 2042 * which we have to use here, because only this file system 2043 * has mnt_export configured. 2044 */ 2045 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2046 credanonp, numsecflavors, secflavors)); 2047} 2048 2049CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2050CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2051 2052static int 2053zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2054{ 2055 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2056 znode_t *zp; 2057 uint64_t object = 0; 2058 uint64_t fid_gen = 0; 2059 uint64_t gen_mask; 2060 uint64_t zp_gen; 2061 int i, err; 2062 2063 *vpp = NULL; 2064 2065 ZFS_ENTER(zfsvfs); 2066 2067 /* 2068 * On FreeBSD we can get snapshot's mount point or its parent file 2069 * system mount point depending if snapshot is already mounted or not. 2070 */ 2071 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2072 zfid_long_t *zlfid = (zfid_long_t *)fidp; 2073 uint64_t objsetid = 0; 2074 uint64_t setgen = 0; 2075 2076 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2077 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2078 2079 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2080 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2081 2082 ZFS_EXIT(zfsvfs); 2083 2084 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2085 if (err) 2086 return (SET_ERROR(EINVAL)); 2087 ZFS_ENTER(zfsvfs); 2088 } 2089 2090 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2091 zfid_short_t *zfid = (zfid_short_t *)fidp; 2092 2093 for (i = 0; i < sizeof (zfid->zf_object); i++) 2094 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2095 2096 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2097 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2098 } else { 2099 ZFS_EXIT(zfsvfs); 2100 return (SET_ERROR(EINVAL)); 2101 } 2102 2103 /* 2104 * A zero fid_gen means we are in .zfs or the .zfs/snapshot 2105 * directory tree. If the object == zfsvfs->z_shares_dir, then 2106 * we are in the .zfs/shares directory tree. 2107 */ 2108 if ((fid_gen == 0 && 2109 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || 2110 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { 2111 *vpp = zfsvfs->z_ctldir; 2112 ASSERT(*vpp != NULL); 2113 if (object == ZFSCTL_INO_SNAPDIR) { 2114 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2115 0, NULL, NULL, NULL, NULL, NULL) == 0); 2116 } else if (object == zfsvfs->z_shares_dir) { 2117 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL, 2118 0, NULL, NULL, NULL, NULL, NULL) == 0); 2119 } else { 2120 VN_HOLD(*vpp); 2121 } 2122 ZFS_EXIT(zfsvfs); 2123 err = vn_lock(*vpp, flags); 2124 if (err != 0) 2125 *vpp = NULL; 2126 return (err); 2127 } 2128 2129 gen_mask = -1ULL >> (64 - 8 * i); 2130 2131 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2132 if (err = zfs_zget(zfsvfs, object, &zp)) { 2133 ZFS_EXIT(zfsvfs); 2134 return (err); 2135 } 2136 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2137 sizeof (uint64_t)); 2138 zp_gen = zp_gen & gen_mask; 2139 if (zp_gen == 0) 2140 zp_gen = 1; 2141 if (zp->z_unlinked || zp_gen != fid_gen) { 2142 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2143 VN_RELE(ZTOV(zp)); 2144 ZFS_EXIT(zfsvfs); 2145 return (SET_ERROR(EINVAL)); 2146 } 2147 2148 *vpp = ZTOV(zp); 2149 ZFS_EXIT(zfsvfs); 2150 err = vn_lock(*vpp, flags | LK_RETRY); 2151 if (err == 0) 2152 vnode_create_vobject(*vpp, zp->z_size, curthread); 2153 else 2154 *vpp = NULL; 2155 return (err); 2156} 2157 2158/* 2159 * Block out VOPs and close zfsvfs_t::z_os 2160 * 2161 * Note, if successful, then we return with the 'z_teardown_lock' and 2162 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 2163 * dataset and objset intact so that they can be atomically handed off during 2164 * a subsequent rollback or recv operation and the resume thereafter. 2165 */ 2166int 2167zfs_suspend_fs(zfsvfs_t *zfsvfs) 2168{ 2169 int error; 2170 2171 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2172 return (error); 2173 2174 return (0); 2175} 2176 2177/* 2178 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 2179 * is an invariant across any of the operations that can be performed while the 2180 * filesystem was suspended. Whether it succeeded or failed, the preconditions 2181 * are the same: the relevant objset and associated dataset are owned by 2182 * zfsvfs, held, and long held on entry. 2183 */ 2184int 2185zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2186{ 2187 int err; 2188 znode_t *zp; 2189 uint64_t sa_obj = 0; 2190 2191 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2192 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2193 2194 /* 2195 * We already own this, so just hold and rele it to update the 2196 * objset_t, as the one we had before may have been evicted. 2197 */ 2198 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os)); 2199 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs); 2200 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset)); 2201 dmu_objset_rele(zfsvfs->z_os, zfsvfs); 2202 2203 /* 2204 * Make sure version hasn't changed 2205 */ 2206 2207 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, 2208 &zfsvfs->z_version); 2209 2210 if (err) 2211 goto bail; 2212 2213 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, 2214 ZFS_SA_ATTRS, 8, 1, &sa_obj); 2215 2216 if (err && zfsvfs->z_version >= ZPL_VERSION_SA) 2217 goto bail; 2218 2219 if ((err = sa_setup(zfsvfs->z_os, sa_obj, 2220 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) 2221 goto bail; 2222 2223 if (zfsvfs->z_version >= ZPL_VERSION_SA) 2224 sa_register_update_callback(zfsvfs->z_os, 2225 zfs_sa_upgrade); 2226 2227 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2228 2229 zfs_set_fuid_feature(zfsvfs); 2230 2231 /* 2232 * Attempt to re-establish all the active znodes with 2233 * their dbufs. If a zfs_rezget() fails, then we'll let 2234 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2235 * when they try to use their znode. 2236 */ 2237 mutex_enter(&zfsvfs->z_znodes_lock); 2238 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2239 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2240 (void) zfs_rezget(zp); 2241 } 2242 mutex_exit(&zfsvfs->z_znodes_lock); 2243 2244bail: 2245 /* release the VOPs */ 2246 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2247 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 2248 2249 if (err) { 2250 /* 2251 * Since we couldn't setup the sa framework, try to force 2252 * unmount this file system. 2253 */ 2254 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) { 2255 vfs_ref(zfsvfs->z_vfs); 2256 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2257 } 2258 } 2259 return (err); 2260} 2261 2262static void 2263zfs_freevfs(vfs_t *vfsp) 2264{ 2265 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2266 2267#ifdef illumos 2268 /* 2269 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2270 * from zfs_mount(). Release it here. If we came through 2271 * zfs_mountroot() instead, we didn't grab an extra hold, so 2272 * skip the VFS_RELE for rootvfs. 2273 */ 2274 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2275 VFS_RELE(zfsvfs->z_parent->z_vfs); 2276#endif 2277 2278 zfsvfs_free(zfsvfs); 2279 2280 atomic_dec_32(&zfs_active_fs_count); 2281} 2282 2283#ifdef __i386__ 2284static int desiredvnodes_backup; 2285#endif 2286 2287static void 2288zfs_vnodes_adjust(void) 2289{ 2290#ifdef __i386__ 2291 int newdesiredvnodes; 2292 2293 desiredvnodes_backup = desiredvnodes; 2294 2295 /* 2296 * We calculate newdesiredvnodes the same way it is done in 2297 * vntblinit(). If it is equal to desiredvnodes, it means that 2298 * it wasn't tuned by the administrator and we can tune it down. 2299 */ 2300 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2301 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2302 sizeof(struct vnode)))); 2303 if (newdesiredvnodes == desiredvnodes) 2304 desiredvnodes = (3 * newdesiredvnodes) / 4; 2305#endif 2306} 2307 2308static void 2309zfs_vnodes_adjust_back(void) 2310{ 2311 2312#ifdef __i386__ 2313 desiredvnodes = desiredvnodes_backup; 2314#endif 2315} 2316 2317void 2318zfs_init(void) 2319{ 2320 2321 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2322 2323 /* 2324 * Initialize .zfs directory structures 2325 */ 2326 zfsctl_init(); 2327 2328 /* 2329 * Initialize znode cache, vnode ops, etc... 2330 */ 2331 zfs_znode_init(); 2332 2333 /* 2334 * Reduce number of vnodes. Originally number of vnodes is calculated 2335 * with UFS inode in mind. We reduce it here, because it's too big for 2336 * ZFS/i386. 2337 */ 2338 zfs_vnodes_adjust(); 2339 2340 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2341} 2342 2343void 2344zfs_fini(void) 2345{ 2346 zfsctl_fini(); 2347 zfs_znode_fini(); 2348 zfs_vnodes_adjust_back(); 2349} 2350 2351int 2352zfs_busy(void) 2353{ 2354 return (zfs_active_fs_count != 0); 2355} 2356 2357int 2358zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2359{ 2360 int error; 2361 objset_t *os = zfsvfs->z_os; 2362 dmu_tx_t *tx; 2363 2364 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2365 return (SET_ERROR(EINVAL)); 2366 2367 if (newvers < zfsvfs->z_version) 2368 return (SET_ERROR(EINVAL)); 2369 2370 if (zfs_spa_version_map(newvers) > 2371 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2372 return (SET_ERROR(ENOTSUP)); 2373 2374 tx = dmu_tx_create(os); 2375 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2376 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2377 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2378 ZFS_SA_ATTRS); 2379 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2380 } 2381 error = dmu_tx_assign(tx, TXG_WAIT); 2382 if (error) { 2383 dmu_tx_abort(tx); 2384 return (error); 2385 } 2386 2387 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2388 8, 1, &newvers, tx); 2389 2390 if (error) { 2391 dmu_tx_commit(tx); 2392 return (error); 2393 } 2394 2395 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2396 uint64_t sa_obj; 2397 2398 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2399 SPA_VERSION_SA); 2400 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2401 DMU_OT_NONE, 0, tx); 2402 2403 error = zap_add(os, MASTER_NODE_OBJ, 2404 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2405 ASSERT0(error); 2406 2407 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2408 sa_register_update_callback(os, zfs_sa_upgrade); 2409 } 2410 2411 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2412 "from %llu to %llu", zfsvfs->z_version, newvers); 2413 2414 dmu_tx_commit(tx); 2415 2416 zfsvfs->z_version = newvers; 2417 2418 zfs_set_fuid_feature(zfsvfs); 2419 2420 return (0); 2421} 2422 2423/* 2424 * Read a property stored within the master node. 2425 */ 2426int 2427zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2428{ 2429 const char *pname; 2430 int error = ENOENT; 2431 2432 /* 2433 * Look up the file system's value for the property. For the 2434 * version property, we look up a slightly different string. 2435 */ 2436 if (prop == ZFS_PROP_VERSION) 2437 pname = ZPL_VERSION_STR; 2438 else 2439 pname = zfs_prop_to_name(prop); 2440 2441 if (os != NULL) 2442 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2443 2444 if (error == ENOENT) { 2445 /* No value set, use the default value */ 2446 switch (prop) { 2447 case ZFS_PROP_VERSION: 2448 *value = ZPL_VERSION; 2449 break; 2450 case ZFS_PROP_NORMALIZE: 2451 case ZFS_PROP_UTF8ONLY: 2452 *value = 0; 2453 break; 2454 case ZFS_PROP_CASE: 2455 *value = ZFS_CASE_SENSITIVE; 2456 break; 2457 default: 2458 return (error); 2459 } 2460 error = 0; 2461 } 2462 return (error); 2463} 2464 2465#ifdef _KERNEL 2466void 2467zfsvfs_update_fromname(const char *oldname, const char *newname) 2468{ 2469 char tmpbuf[MAXPATHLEN]; 2470 struct mount *mp; 2471 char *fromname; 2472 size_t oldlen; 2473 2474 oldlen = strlen(oldname); 2475 2476 mtx_lock(&mountlist_mtx); 2477 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2478 fromname = mp->mnt_stat.f_mntfromname; 2479 if (strcmp(fromname, oldname) == 0) { 2480 (void)strlcpy(fromname, newname, 2481 sizeof(mp->mnt_stat.f_mntfromname)); 2482 continue; 2483 } 2484 if (strncmp(fromname, oldname, oldlen) == 0 && 2485 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2486 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2487 newname, fromname + oldlen); 2488 (void)strlcpy(fromname, tmpbuf, 2489 sizeof(mp->mnt_stat.f_mntfromname)); 2490 continue; 2491 } 2492 } 2493 mtx_unlock(&mountlist_mtx); 2494} 2495#endif 2496