zfs_vfsops.c revision 297081
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>. 24 * All rights reserved. 25 * Copyright (c) 2012, 2014 by Delphix. All rights reserved. 26 */ 27 28/* Portions Copyright 2010 Robert Milkowski */ 29 30#include <sys/types.h> 31#include <sys/param.h> 32#include <sys/systm.h> 33#include <sys/kernel.h> 34#include <sys/sysmacros.h> 35#include <sys/kmem.h> 36#include <sys/acl.h> 37#include <sys/vnode.h> 38#include <sys/vfs.h> 39#include <sys/mntent.h> 40#include <sys/mount.h> 41#include <sys/cmn_err.h> 42#include <sys/zfs_znode.h> 43#include <sys/zfs_dir.h> 44#include <sys/zil.h> 45#include <sys/fs/zfs.h> 46#include <sys/dmu.h> 47#include <sys/dsl_prop.h> 48#include <sys/dsl_dataset.h> 49#include <sys/dsl_deleg.h> 50#include <sys/spa.h> 51#include <sys/zap.h> 52#include <sys/sa.h> 53#include <sys/sa_impl.h> 54#include <sys/varargs.h> 55#include <sys/policy.h> 56#include <sys/atomic.h> 57#include <sys/zfs_ioctl.h> 58#include <sys/zfs_ctldir.h> 59#include <sys/zfs_fuid.h> 60#include <sys/sunddi.h> 61#include <sys/dnlc.h> 62#include <sys/dmu_objset.h> 63#include <sys/spa_boot.h> 64#include <sys/jail.h> 65#include "zfs_comutil.h" 66 67struct mtx zfs_debug_mtx; 68MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF); 69 70SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system"); 71 72int zfs_super_owner; 73SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0, 74 "File system owner can perform privileged operation on his file systems"); 75 76int zfs_debug_level; 77TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level); 78SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0, 79 "Debug level"); 80 81SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions"); 82static int zfs_version_acl = ZFS_ACL_VERSION; 83SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0, 84 "ZFS_ACL_VERSION"); 85static int zfs_version_spa = SPA_VERSION; 86SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0, 87 "SPA_VERSION"); 88static int zfs_version_zpl = ZPL_VERSION; 89SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0, 90 "ZPL_VERSION"); 91 92static int zfs_mount(vfs_t *vfsp); 93static int zfs_umount(vfs_t *vfsp, int fflag); 94static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp); 95static int zfs_statfs(vfs_t *vfsp, struct statfs *statp); 96static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp); 97static int zfs_sync(vfs_t *vfsp, int waitfor); 98static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 99 struct ucred **credanonp, int *numsecflavors, int **secflavors); 100static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp); 101static void zfs_objset_close(zfsvfs_t *zfsvfs); 102static void zfs_freevfs(vfs_t *vfsp); 103 104static struct vfsops zfs_vfsops = { 105 .vfs_mount = zfs_mount, 106 .vfs_unmount = zfs_umount, 107 .vfs_root = zfs_root, 108 .vfs_statfs = zfs_statfs, 109 .vfs_vget = zfs_vget, 110 .vfs_sync = zfs_sync, 111 .vfs_checkexp = zfs_checkexp, 112 .vfs_fhtovp = zfs_fhtovp, 113}; 114 115VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN); 116 117/* 118 * We need to keep a count of active fs's. 119 * This is necessary to prevent our module 120 * from being unloaded after a umount -f 121 */ 122static uint32_t zfs_active_fs_count = 0; 123 124/*ARGSUSED*/ 125static int 126zfs_sync(vfs_t *vfsp, int waitfor) 127{ 128 129 /* 130 * Data integrity is job one. We don't want a compromised kernel 131 * writing to the storage pool, so we never sync during panic. 132 */ 133 if (panicstr) 134 return (0); 135 136 /* 137 * Ignore the system syncher. ZFS already commits async data 138 * at zfs_txg_timeout intervals. 139 */ 140 if (waitfor == MNT_LAZY) 141 return (0); 142 143 if (vfsp != NULL) { 144 /* 145 * Sync a specific filesystem. 146 */ 147 zfsvfs_t *zfsvfs = vfsp->vfs_data; 148 dsl_pool_t *dp; 149 int error; 150 151 error = vfs_stdsync(vfsp, waitfor); 152 if (error != 0) 153 return (error); 154 155 ZFS_ENTER(zfsvfs); 156 dp = dmu_objset_pool(zfsvfs->z_os); 157 158 /* 159 * If the system is shutting down, then skip any 160 * filesystems which may exist on a suspended pool. 161 */ 162 if (sys_shutdown && spa_suspended(dp->dp_spa)) { 163 ZFS_EXIT(zfsvfs); 164 return (0); 165 } 166 167 if (zfsvfs->z_log != NULL) 168 zil_commit(zfsvfs->z_log, 0); 169 170 ZFS_EXIT(zfsvfs); 171 } else { 172 /* 173 * Sync all ZFS filesystems. This is what happens when you 174 * run sync(1M). Unlike other filesystems, ZFS honors the 175 * request by waiting for all pools to commit all dirty data. 176 */ 177 spa_sync_allpools(); 178 } 179 180 return (0); 181} 182 183#ifndef __FreeBSD_kernel__ 184static int 185zfs_create_unique_device(dev_t *dev) 186{ 187 major_t new_major; 188 189 do { 190 ASSERT3U(zfs_minor, <=, MAXMIN32); 191 minor_t start = zfs_minor; 192 do { 193 mutex_enter(&zfs_dev_mtx); 194 if (zfs_minor >= MAXMIN32) { 195 /* 196 * If we're still using the real major 197 * keep out of /dev/zfs and /dev/zvol minor 198 * number space. If we're using a getudev()'ed 199 * major number, we can use all of its minors. 200 */ 201 if (zfs_major == ddi_name_to_major(ZFS_DRIVER)) 202 zfs_minor = ZFS_MIN_MINOR; 203 else 204 zfs_minor = 0; 205 } else { 206 zfs_minor++; 207 } 208 *dev = makedevice(zfs_major, zfs_minor); 209 mutex_exit(&zfs_dev_mtx); 210 } while (vfs_devismounted(*dev) && zfs_minor != start); 211 if (zfs_minor == start) { 212 /* 213 * We are using all ~262,000 minor numbers for the 214 * current major number. Create a new major number. 215 */ 216 if ((new_major = getudev()) == (major_t)-1) { 217 cmn_err(CE_WARN, 218 "zfs_mount: Can't get unique major " 219 "device number."); 220 return (-1); 221 } 222 mutex_enter(&zfs_dev_mtx); 223 zfs_major = new_major; 224 zfs_minor = 0; 225 226 mutex_exit(&zfs_dev_mtx); 227 } else { 228 break; 229 } 230 /* CONSTANTCONDITION */ 231 } while (1); 232 233 return (0); 234} 235#endif /* !__FreeBSD_kernel__ */ 236 237static void 238atime_changed_cb(void *arg, uint64_t newval) 239{ 240 zfsvfs_t *zfsvfs = arg; 241 242 if (newval == TRUE) { 243 zfsvfs->z_atime = TRUE; 244 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; 245 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); 246 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); 247 } else { 248 zfsvfs->z_atime = FALSE; 249 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; 250 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); 251 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); 252 } 253} 254 255static void 256xattr_changed_cb(void *arg, uint64_t newval) 257{ 258 zfsvfs_t *zfsvfs = arg; 259 260 if (newval == TRUE) { 261 /* XXX locking on vfs_flag? */ 262#ifdef TODO 263 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; 264#endif 265 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); 266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); 267 } else { 268 /* XXX locking on vfs_flag? */ 269#ifdef TODO 270 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR; 271#endif 272 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR); 273 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); 274 } 275} 276 277static void 278blksz_changed_cb(void *arg, uint64_t newval) 279{ 280 zfsvfs_t *zfsvfs = arg; 281 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os))); 282 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); 283 ASSERT(ISP2(newval)); 284 285 zfsvfs->z_max_blksz = newval; 286 zfsvfs->z_vfs->mnt_stat.f_iosize = newval; 287} 288 289static void 290readonly_changed_cb(void *arg, uint64_t newval) 291{ 292 zfsvfs_t *zfsvfs = arg; 293 294 if (newval) { 295 /* XXX locking on vfs_flag? */ 296 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; 297 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); 298 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); 299 } else { 300 /* XXX locking on vfs_flag? */ 301 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 302 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); 303 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); 304 } 305} 306 307static void 308setuid_changed_cb(void *arg, uint64_t newval) 309{ 310 zfsvfs_t *zfsvfs = arg; 311 312 if (newval == FALSE) { 313 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; 314 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); 315 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); 316 } else { 317 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; 318 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); 319 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); 320 } 321} 322 323static void 324exec_changed_cb(void *arg, uint64_t newval) 325{ 326 zfsvfs_t *zfsvfs = arg; 327 328 if (newval == FALSE) { 329 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; 330 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); 331 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); 332 } else { 333 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; 334 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); 335 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); 336 } 337} 338 339/* 340 * The nbmand mount option can be changed at mount time. 341 * We can't allow it to be toggled on live file systems or incorrect 342 * behavior may be seen from cifs clients 343 * 344 * This property isn't registered via dsl_prop_register(), but this callback 345 * will be called when a file system is first mounted 346 */ 347static void 348nbmand_changed_cb(void *arg, uint64_t newval) 349{ 350 zfsvfs_t *zfsvfs = arg; 351 if (newval == FALSE) { 352 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); 353 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); 354 } else { 355 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); 356 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); 357 } 358} 359 360static void 361snapdir_changed_cb(void *arg, uint64_t newval) 362{ 363 zfsvfs_t *zfsvfs = arg; 364 365 zfsvfs->z_show_ctldir = newval; 366} 367 368static void 369vscan_changed_cb(void *arg, uint64_t newval) 370{ 371 zfsvfs_t *zfsvfs = arg; 372 373 zfsvfs->z_vscan = newval; 374} 375 376static void 377acl_mode_changed_cb(void *arg, uint64_t newval) 378{ 379 zfsvfs_t *zfsvfs = arg; 380 381 zfsvfs->z_acl_mode = newval; 382} 383 384static void 385acl_inherit_changed_cb(void *arg, uint64_t newval) 386{ 387 zfsvfs_t *zfsvfs = arg; 388 389 zfsvfs->z_acl_inherit = newval; 390} 391 392static int 393zfs_register_callbacks(vfs_t *vfsp) 394{ 395 struct dsl_dataset *ds = NULL; 396 objset_t *os = NULL; 397 zfsvfs_t *zfsvfs = NULL; 398 uint64_t nbmand; 399 boolean_t readonly = B_FALSE; 400 boolean_t do_readonly = B_FALSE; 401 boolean_t setuid = B_FALSE; 402 boolean_t do_setuid = B_FALSE; 403 boolean_t exec = B_FALSE; 404 boolean_t do_exec = B_FALSE; 405#ifdef illumos 406 boolean_t devices = B_FALSE; 407 boolean_t do_devices = B_FALSE; 408#endif 409 boolean_t xattr = B_FALSE; 410 boolean_t do_xattr = B_FALSE; 411 boolean_t atime = B_FALSE; 412 boolean_t do_atime = B_FALSE; 413 int error = 0; 414 415 ASSERT(vfsp); 416 zfsvfs = vfsp->vfs_data; 417 ASSERT(zfsvfs); 418 os = zfsvfs->z_os; 419 420 /* 421 * This function can be called for a snapshot when we update snapshot's 422 * mount point, which isn't really supported. 423 */ 424 if (dmu_objset_is_snapshot(os)) 425 return (EOPNOTSUPP); 426 427 /* 428 * The act of registering our callbacks will destroy any mount 429 * options we may have. In order to enable temporary overrides 430 * of mount options, we stash away the current values and 431 * restore them after we register the callbacks. 432 */ 433 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) || 434 !spa_writeable(dmu_objset_spa(os))) { 435 readonly = B_TRUE; 436 do_readonly = B_TRUE; 437 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 438 readonly = B_FALSE; 439 do_readonly = B_TRUE; 440 } 441 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 442 setuid = B_FALSE; 443 do_setuid = B_TRUE; 444 } else { 445 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 446 setuid = B_FALSE; 447 do_setuid = B_TRUE; 448 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 449 setuid = B_TRUE; 450 do_setuid = B_TRUE; 451 } 452 } 453 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 454 exec = B_FALSE; 455 do_exec = B_TRUE; 456 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 457 exec = B_TRUE; 458 do_exec = B_TRUE; 459 } 460 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 461 xattr = B_FALSE; 462 do_xattr = B_TRUE; 463 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 464 xattr = B_TRUE; 465 do_xattr = B_TRUE; 466 } 467 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 468 atime = B_FALSE; 469 do_atime = B_TRUE; 470 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 471 atime = B_TRUE; 472 do_atime = B_TRUE; 473 } 474 475 /* 476 * We need to enter pool configuration here, so that we can use 477 * dsl_prop_get_int_ds() to handle the special nbmand property below. 478 * dsl_prop_get_integer() can not be used, because it has to acquire 479 * spa_namespace_lock and we can not do that because we already hold 480 * z_teardown_lock. The problem is that spa_config_sync() is called 481 * with spa_namespace_lock held and the function calls ZFS vnode 482 * operations to write the cache file and thus z_teardown_lock is 483 * acquired after spa_namespace_lock. 484 */ 485 ds = dmu_objset_ds(os); 486 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 487 488 /* 489 * nbmand is a special property. It can only be changed at 490 * mount time. 491 * 492 * This is weird, but it is documented to only be changeable 493 * at mount time. 494 */ 495 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 496 nbmand = B_FALSE; 497 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 498 nbmand = B_TRUE; 499 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) { 500 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 501 return (error); 502 } 503 504 /* 505 * Register property callbacks. 506 * 507 * It would probably be fine to just check for i/o error from 508 * the first prop_register(), but I guess I like to go 509 * overboard... 510 */ 511 error = dsl_prop_register(ds, 512 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 513 error = error ? error : dsl_prop_register(ds, 514 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 515 error = error ? error : dsl_prop_register(ds, 516 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 517 error = error ? error : dsl_prop_register(ds, 518 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 519#ifdef illumos 520 error = error ? error : dsl_prop_register(ds, 521 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 522#endif 523 error = error ? error : dsl_prop_register(ds, 524 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 525 error = error ? error : dsl_prop_register(ds, 526 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 527 error = error ? error : dsl_prop_register(ds, 528 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 529 error = error ? error : dsl_prop_register(ds, 530 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 531 error = error ? error : dsl_prop_register(ds, 532 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 533 zfsvfs); 534 error = error ? error : dsl_prop_register(ds, 535 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); 536 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 537 if (error) 538 goto unregister; 539 540 /* 541 * Invoke our callbacks to restore temporary mount options. 542 */ 543 if (do_readonly) 544 readonly_changed_cb(zfsvfs, readonly); 545 if (do_setuid) 546 setuid_changed_cb(zfsvfs, setuid); 547 if (do_exec) 548 exec_changed_cb(zfsvfs, exec); 549 if (do_xattr) 550 xattr_changed_cb(zfsvfs, xattr); 551 if (do_atime) 552 atime_changed_cb(zfsvfs, atime); 553 554 nbmand_changed_cb(zfsvfs, nbmand); 555 556 return (0); 557 558unregister: 559 dsl_prop_unregister_all(ds, zfsvfs); 560 return (error); 561} 562 563static int 564zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, 565 uint64_t *userp, uint64_t *groupp) 566{ 567 /* 568 * Is it a valid type of object to track? 569 */ 570 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 571 return (SET_ERROR(ENOENT)); 572 573 /* 574 * If we have a NULL data pointer 575 * then assume the id's aren't changing and 576 * return EEXIST to the dmu to let it know to 577 * use the same ids 578 */ 579 if (data == NULL) 580 return (SET_ERROR(EEXIST)); 581 582 if (bonustype == DMU_OT_ZNODE) { 583 znode_phys_t *znp = data; 584 *userp = znp->zp_uid; 585 *groupp = znp->zp_gid; 586 } else { 587 int hdrsize; 588 sa_hdr_phys_t *sap = data; 589 sa_hdr_phys_t sa = *sap; 590 boolean_t swap = B_FALSE; 591 592 ASSERT(bonustype == DMU_OT_SA); 593 594 if (sa.sa_magic == 0) { 595 /* 596 * This should only happen for newly created 597 * files that haven't had the znode data filled 598 * in yet. 599 */ 600 *userp = 0; 601 *groupp = 0; 602 return (0); 603 } 604 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { 605 sa.sa_magic = SA_MAGIC; 606 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); 607 swap = B_TRUE; 608 } else { 609 VERIFY3U(sa.sa_magic, ==, SA_MAGIC); 610 } 611 612 hdrsize = sa_hdrsize(&sa); 613 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); 614 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + 615 SA_UID_OFFSET)); 616 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + 617 SA_GID_OFFSET)); 618 if (swap) { 619 *userp = BSWAP_64(*userp); 620 *groupp = BSWAP_64(*groupp); 621 } 622 } 623 return (0); 624} 625 626static void 627fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, 628 char *domainbuf, int buflen, uid_t *ridp) 629{ 630 uint64_t fuid; 631 const char *domain; 632 633 fuid = strtonum(fuidstr, NULL); 634 635 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); 636 if (domain) 637 (void) strlcpy(domainbuf, domain, buflen); 638 else 639 domainbuf[0] = '\0'; 640 *ridp = FUID_RID(fuid); 641} 642 643static uint64_t 644zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) 645{ 646 switch (type) { 647 case ZFS_PROP_USERUSED: 648 return (DMU_USERUSED_OBJECT); 649 case ZFS_PROP_GROUPUSED: 650 return (DMU_GROUPUSED_OBJECT); 651 case ZFS_PROP_USERQUOTA: 652 return (zfsvfs->z_userquota_obj); 653 case ZFS_PROP_GROUPQUOTA: 654 return (zfsvfs->z_groupquota_obj); 655 } 656 return (0); 657} 658 659int 660zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 661 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) 662{ 663 int error; 664 zap_cursor_t zc; 665 zap_attribute_t za; 666 zfs_useracct_t *buf = vbuf; 667 uint64_t obj; 668 669 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 670 return (SET_ERROR(ENOTSUP)); 671 672 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 673 if (obj == 0) { 674 *bufsizep = 0; 675 return (0); 676 } 677 678 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); 679 (error = zap_cursor_retrieve(&zc, &za)) == 0; 680 zap_cursor_advance(&zc)) { 681 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > 682 *bufsizep) 683 break; 684 685 fuidstr_to_sid(zfsvfs, za.za_name, 686 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); 687 688 buf->zu_space = za.za_first_integer; 689 buf++; 690 } 691 if (error == ENOENT) 692 error = 0; 693 694 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); 695 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; 696 *cookiep = zap_cursor_serialize(&zc); 697 zap_cursor_fini(&zc); 698 return (error); 699} 700 701/* 702 * buf must be big enough (eg, 32 bytes) 703 */ 704static int 705id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, 706 char *buf, boolean_t addok) 707{ 708 uint64_t fuid; 709 int domainid = 0; 710 711 if (domain && domain[0]) { 712 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); 713 if (domainid == -1) 714 return (SET_ERROR(ENOENT)); 715 } 716 fuid = FUID_ENCODE(domainid, rid); 717 (void) sprintf(buf, "%llx", (longlong_t)fuid); 718 return (0); 719} 720 721int 722zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 723 const char *domain, uint64_t rid, uint64_t *valp) 724{ 725 char buf[32]; 726 int err; 727 uint64_t obj; 728 729 *valp = 0; 730 731 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 732 return (SET_ERROR(ENOTSUP)); 733 734 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 735 if (obj == 0) 736 return (0); 737 738 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); 739 if (err) 740 return (err); 741 742 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); 743 if (err == ENOENT) 744 err = 0; 745 return (err); 746} 747 748int 749zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 750 const char *domain, uint64_t rid, uint64_t quota) 751{ 752 char buf[32]; 753 int err; 754 dmu_tx_t *tx; 755 uint64_t *objp; 756 boolean_t fuid_dirtied; 757 758 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) 759 return (SET_ERROR(EINVAL)); 760 761 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) 762 return (SET_ERROR(ENOTSUP)); 763 764 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : 765 &zfsvfs->z_groupquota_obj; 766 767 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); 768 if (err) 769 return (err); 770 fuid_dirtied = zfsvfs->z_fuid_dirty; 771 772 tx = dmu_tx_create(zfsvfs->z_os); 773 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); 774 if (*objp == 0) { 775 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 776 zfs_userquota_prop_prefixes[type]); 777 } 778 if (fuid_dirtied) 779 zfs_fuid_txhold(zfsvfs, tx); 780 err = dmu_tx_assign(tx, TXG_WAIT); 781 if (err) { 782 dmu_tx_abort(tx); 783 return (err); 784 } 785 786 mutex_enter(&zfsvfs->z_lock); 787 if (*objp == 0) { 788 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, 789 DMU_OT_NONE, 0, tx); 790 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 791 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); 792 } 793 mutex_exit(&zfsvfs->z_lock); 794 795 if (quota == 0) { 796 err = zap_remove(zfsvfs->z_os, *objp, buf, tx); 797 if (err == ENOENT) 798 err = 0; 799 } else { 800 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); 801 } 802 ASSERT(err == 0); 803 if (fuid_dirtied) 804 zfs_fuid_sync(zfsvfs, tx); 805 dmu_tx_commit(tx); 806 return (err); 807} 808 809boolean_t 810zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) 811{ 812 char buf[32]; 813 uint64_t used, quota, usedobj, quotaobj; 814 int err; 815 816 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; 817 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 818 819 if (quotaobj == 0 || zfsvfs->z_replay) 820 return (B_FALSE); 821 822 (void) sprintf(buf, "%llx", (longlong_t)fuid); 823 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); 824 if (err != 0) 825 return (B_FALSE); 826 827 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); 828 if (err != 0) 829 return (B_FALSE); 830 return (used >= quota); 831} 832 833boolean_t 834zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) 835{ 836 uint64_t fuid; 837 uint64_t quotaobj; 838 839 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 840 841 fuid = isgroup ? zp->z_gid : zp->z_uid; 842 843 if (quotaobj == 0 || zfsvfs->z_replay) 844 return (B_FALSE); 845 846 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); 847} 848 849int 850zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 851{ 852 objset_t *os; 853 zfsvfs_t *zfsvfs; 854 uint64_t zval; 855 int i, error; 856 uint64_t sa_obj; 857 858 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 859 860 /* 861 * We claim to always be readonly so we can open snapshots; 862 * other ZPL code will prevent us from writing to snapshots. 863 */ 864 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 865 if (error) { 866 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 867 return (error); 868 } 869 870 /* 871 * Initialize the zfs-specific filesystem structure. 872 * Should probably make this a kmem cache, shuffle fields, 873 * and just bzero up to z_hold_mtx[]. 874 */ 875 zfsvfs->z_vfs = NULL; 876 zfsvfs->z_parent = zfsvfs; 877 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 878 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 879 zfsvfs->z_os = os; 880 881 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 882 if (error) { 883 goto out; 884 } else if (zfsvfs->z_version > 885 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 886 (void) printf("Can't mount a version %lld file system " 887 "on a version %lld pool\n. Pool must be upgraded to mount " 888 "this file system.", (u_longlong_t)zfsvfs->z_version, 889 (u_longlong_t)spa_version(dmu_objset_spa(os))); 890 error = SET_ERROR(ENOTSUP); 891 goto out; 892 } 893 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) 894 goto out; 895 zfsvfs->z_norm = (int)zval; 896 897 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) 898 goto out; 899 zfsvfs->z_utf8 = (zval != 0); 900 901 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) 902 goto out; 903 zfsvfs->z_case = (uint_t)zval; 904 905 /* 906 * Fold case on file systems that are always or sometimes case 907 * insensitive. 908 */ 909 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 910 zfsvfs->z_case == ZFS_CASE_MIXED) 911 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 912 913 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 914 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 915 916 if (zfsvfs->z_use_sa) { 917 /* should either have both of these objects or none */ 918 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 919 &sa_obj); 920 if (error) 921 goto out; 922 } else { 923 /* 924 * Pre SA versions file systems should never touch 925 * either the attribute registration or layout objects. 926 */ 927 sa_obj = 0; 928 } 929 930 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 931 &zfsvfs->z_attr_table); 932 if (error) 933 goto out; 934 935 if (zfsvfs->z_version >= ZPL_VERSION_SA) 936 sa_register_update_callback(os, zfs_sa_upgrade); 937 938 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 939 &zfsvfs->z_root); 940 if (error) 941 goto out; 942 ASSERT(zfsvfs->z_root != 0); 943 944 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 945 &zfsvfs->z_unlinkedobj); 946 if (error) 947 goto out; 948 949 error = zap_lookup(os, MASTER_NODE_OBJ, 950 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 951 8, 1, &zfsvfs->z_userquota_obj); 952 if (error && error != ENOENT) 953 goto out; 954 955 error = zap_lookup(os, MASTER_NODE_OBJ, 956 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 957 8, 1, &zfsvfs->z_groupquota_obj); 958 if (error && error != ENOENT) 959 goto out; 960 961 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 962 &zfsvfs->z_fuid_obj); 963 if (error && error != ENOENT) 964 goto out; 965 966 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 967 &zfsvfs->z_shares_dir); 968 if (error && error != ENOENT) 969 goto out; 970 971 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 972 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 973 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 974 offsetof(znode_t, z_link_node)); 975 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE); 976 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 977 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 978 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 979 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 980 981 *zfvp = zfsvfs; 982 return (0); 983 984out: 985 dmu_objset_disown(os, zfsvfs); 986 *zfvp = NULL; 987 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 988 return (error); 989} 990 991static int 992zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 993{ 994 int error; 995 996 error = zfs_register_callbacks(zfsvfs->z_vfs); 997 if (error) 998 return (error); 999 1000 /* 1001 * Set the objset user_ptr to track its zfsvfs. 1002 */ 1003 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1004 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1005 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1006 1007 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 1008 1009 /* 1010 * If we are not mounting (ie: online recv), then we don't 1011 * have to worry about replaying the log as we blocked all 1012 * operations out since we closed the ZIL. 1013 */ 1014 if (mounting) { 1015 boolean_t readonly; 1016 1017 /* 1018 * During replay we remove the read only flag to 1019 * allow replays to succeed. 1020 */ 1021 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1022 if (readonly != 0) 1023 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1024 else 1025 zfs_unlinked_drain(zfsvfs); 1026 1027 /* 1028 * Parse and replay the intent log. 1029 * 1030 * Because of ziltest, this must be done after 1031 * zfs_unlinked_drain(). (Further note: ziltest 1032 * doesn't use readonly mounts, where 1033 * zfs_unlinked_drain() isn't called.) This is because 1034 * ziltest causes spa_sync() to think it's committed, 1035 * but actually it is not, so the intent log contains 1036 * many txg's worth of changes. 1037 * 1038 * In particular, if object N is in the unlinked set in 1039 * the last txg to actually sync, then it could be 1040 * actually freed in a later txg and then reallocated 1041 * in a yet later txg. This would write a "create 1042 * object N" record to the intent log. Normally, this 1043 * would be fine because the spa_sync() would have 1044 * written out the fact that object N is free, before 1045 * we could write the "create object N" intent log 1046 * record. 1047 * 1048 * But when we are in ziltest mode, we advance the "open 1049 * txg" without actually spa_sync()-ing the changes to 1050 * disk. So we would see that object N is still 1051 * allocated and in the unlinked set, and there is an 1052 * intent log record saying to allocate it. 1053 */ 1054 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1055 if (zil_replay_disable) { 1056 zil_destroy(zfsvfs->z_log, B_FALSE); 1057 } else { 1058 zfsvfs->z_replay = B_TRUE; 1059 zil_replay(zfsvfs->z_os, zfsvfs, 1060 zfs_replay_vector); 1061 zfsvfs->z_replay = B_FALSE; 1062 } 1063 } 1064 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1065 } 1066 1067 return (0); 1068} 1069 1070extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1071 1072void 1073zfsvfs_free(zfsvfs_t *zfsvfs) 1074{ 1075 int i; 1076 1077 /* 1078 * This is a barrier to prevent the filesystem from going away in 1079 * zfs_znode_move() until we can safely ensure that the filesystem is 1080 * not unmounted. We consider the filesystem valid before the barrier 1081 * and invalid after the barrier. 1082 */ 1083 rw_enter(&zfsvfs_lock, RW_READER); 1084 rw_exit(&zfsvfs_lock); 1085 1086 zfs_fuid_destroy(zfsvfs); 1087 1088 mutex_destroy(&zfsvfs->z_znodes_lock); 1089 mutex_destroy(&zfsvfs->z_lock); 1090 list_destroy(&zfsvfs->z_all_znodes); 1091 rrm_destroy(&zfsvfs->z_teardown_lock); 1092 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1093 rw_destroy(&zfsvfs->z_fuid_lock); 1094 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1095 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1096 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1097} 1098 1099static void 1100zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1101{ 1102 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1103 if (zfsvfs->z_vfs) { 1104 if (zfsvfs->z_use_fuids) { 1105 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1106 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1107 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1108 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1109 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1110 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1111 } else { 1112 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1113 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1114 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1115 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1116 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1117 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1118 } 1119 } 1120 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1121} 1122 1123static int 1124zfs_domount(vfs_t *vfsp, char *osname) 1125{ 1126 uint64_t recordsize, fsid_guid; 1127 int error = 0; 1128 zfsvfs_t *zfsvfs; 1129 vnode_t *vp; 1130 1131 ASSERT(vfsp); 1132 ASSERT(osname); 1133 1134 error = zfsvfs_create(osname, &zfsvfs); 1135 if (error) 1136 return (error); 1137 zfsvfs->z_vfs = vfsp; 1138 1139#ifdef illumos 1140 /* Initialize the generic filesystem structure. */ 1141 vfsp->vfs_bcount = 0; 1142 vfsp->vfs_data = NULL; 1143 1144 if (zfs_create_unique_device(&mount_dev) == -1) { 1145 error = SET_ERROR(ENODEV); 1146 goto out; 1147 } 1148 ASSERT(vfs_devismounted(mount_dev) == 0); 1149#endif 1150 1151 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1152 NULL)) 1153 goto out; 1154 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1155 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1156 1157 vfsp->vfs_data = zfsvfs; 1158 vfsp->mnt_flag |= MNT_LOCAL; 1159 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1160 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1161 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1162 1163 /* 1164 * The fsid is 64 bits, composed of an 8-bit fs type, which 1165 * separates our fsid from any other filesystem types, and a 1166 * 56-bit objset unique ID. The objset unique ID is unique to 1167 * all objsets open on this system, provided by unique_create(). 1168 * The 8-bit fs type must be put in the low bits of fsid[1] 1169 * because that's where other Solaris filesystems put it. 1170 */ 1171 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1172 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1173 vfsp->vfs_fsid.val[0] = fsid_guid; 1174 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1175 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1176 1177 /* 1178 * Set features for file system. 1179 */ 1180 zfs_set_fuid_feature(zfsvfs); 1181 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1182 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1183 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1184 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1185 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1186 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1187 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1188 } 1189 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1190 1191 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1192 uint64_t pval; 1193 1194 atime_changed_cb(zfsvfs, B_FALSE); 1195 readonly_changed_cb(zfsvfs, B_TRUE); 1196 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1197 goto out; 1198 xattr_changed_cb(zfsvfs, pval); 1199 zfsvfs->z_issnap = B_TRUE; 1200 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1201 1202 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1203 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1204 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1205 } else { 1206 error = zfsvfs_setup(zfsvfs, B_TRUE); 1207 } 1208 1209 vfs_mountedfrom(vfsp, osname); 1210 1211 if (!zfsvfs->z_issnap) 1212 zfsctl_create(zfsvfs); 1213out: 1214 if (error) { 1215 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1216 zfsvfs_free(zfsvfs); 1217 } else { 1218 atomic_inc_32(&zfs_active_fs_count); 1219 } 1220 1221 return (error); 1222} 1223 1224void 1225zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1226{ 1227 objset_t *os = zfsvfs->z_os; 1228 1229 if (!dmu_objset_is_snapshot(os)) 1230 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 1231} 1232 1233#ifdef SECLABEL 1234/* 1235 * Convert a decimal digit string to a uint64_t integer. 1236 */ 1237static int 1238str_to_uint64(char *str, uint64_t *objnum) 1239{ 1240 uint64_t num = 0; 1241 1242 while (*str) { 1243 if (*str < '0' || *str > '9') 1244 return (SET_ERROR(EINVAL)); 1245 1246 num = num*10 + *str++ - '0'; 1247 } 1248 1249 *objnum = num; 1250 return (0); 1251} 1252 1253/* 1254 * The boot path passed from the boot loader is in the form of 1255 * "rootpool-name/root-filesystem-object-number'. Convert this 1256 * string to a dataset name: "rootpool-name/root-filesystem-name". 1257 */ 1258static int 1259zfs_parse_bootfs(char *bpath, char *outpath) 1260{ 1261 char *slashp; 1262 uint64_t objnum; 1263 int error; 1264 1265 if (*bpath == 0 || *bpath == '/') 1266 return (SET_ERROR(EINVAL)); 1267 1268 (void) strcpy(outpath, bpath); 1269 1270 slashp = strchr(bpath, '/'); 1271 1272 /* if no '/', just return the pool name */ 1273 if (slashp == NULL) { 1274 return (0); 1275 } 1276 1277 /* if not a number, just return the root dataset name */ 1278 if (str_to_uint64(slashp+1, &objnum)) { 1279 return (0); 1280 } 1281 1282 *slashp = '\0'; 1283 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1284 *slashp = '/'; 1285 1286 return (error); 1287} 1288 1289/* 1290 * Check that the hex label string is appropriate for the dataset being 1291 * mounted into the global_zone proper. 1292 * 1293 * Return an error if the hex label string is not default or 1294 * admin_low/admin_high. For admin_low labels, the corresponding 1295 * dataset must be readonly. 1296 */ 1297int 1298zfs_check_global_label(const char *dsname, const char *hexsl) 1299{ 1300 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1301 return (0); 1302 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1303 return (0); 1304 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1305 /* must be readonly */ 1306 uint64_t rdonly; 1307 1308 if (dsl_prop_get_integer(dsname, 1309 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1310 return (SET_ERROR(EACCES)); 1311 return (rdonly ? 0 : EACCES); 1312 } 1313 return (SET_ERROR(EACCES)); 1314} 1315 1316/* 1317 * Determine whether the mount is allowed according to MAC check. 1318 * by comparing (where appropriate) label of the dataset against 1319 * the label of the zone being mounted into. If the dataset has 1320 * no label, create one. 1321 * 1322 * Returns 0 if access allowed, error otherwise (e.g. EACCES) 1323 */ 1324static int 1325zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1326{ 1327 int error, retv; 1328 zone_t *mntzone = NULL; 1329 ts_label_t *mnt_tsl; 1330 bslabel_t *mnt_sl; 1331 bslabel_t ds_sl; 1332 char ds_hexsl[MAXNAMELEN]; 1333 1334 retv = EACCES; /* assume the worst */ 1335 1336 /* 1337 * Start by getting the dataset label if it exists. 1338 */ 1339 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1340 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1341 if (error) 1342 return (SET_ERROR(EACCES)); 1343 1344 /* 1345 * If labeling is NOT enabled, then disallow the mount of datasets 1346 * which have a non-default label already. No other label checks 1347 * are needed. 1348 */ 1349 if (!is_system_labeled()) { 1350 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1351 return (0); 1352 return (SET_ERROR(EACCES)); 1353 } 1354 1355 /* 1356 * Get the label of the mountpoint. If mounting into the global 1357 * zone (i.e. mountpoint is not within an active zone and the 1358 * zoned property is off), the label must be default or 1359 * admin_low/admin_high only; no other checks are needed. 1360 */ 1361 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1362 if (mntzone->zone_id == GLOBAL_ZONEID) { 1363 uint64_t zoned; 1364 1365 zone_rele(mntzone); 1366 1367 if (dsl_prop_get_integer(osname, 1368 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1369 return (SET_ERROR(EACCES)); 1370 if (!zoned) 1371 return (zfs_check_global_label(osname, ds_hexsl)); 1372 else 1373 /* 1374 * This is the case of a zone dataset being mounted 1375 * initially, before the zone has been fully created; 1376 * allow this mount into global zone. 1377 */ 1378 return (0); 1379 } 1380 1381 mnt_tsl = mntzone->zone_slabel; 1382 ASSERT(mnt_tsl != NULL); 1383 label_hold(mnt_tsl); 1384 mnt_sl = label2bslabel(mnt_tsl); 1385 1386 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1387 /* 1388 * The dataset doesn't have a real label, so fabricate one. 1389 */ 1390 char *str = NULL; 1391 1392 if (l_to_str_internal(mnt_sl, &str) == 0 && 1393 dsl_prop_set_string(osname, 1394 zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1395 ZPROP_SRC_LOCAL, str) == 0) 1396 retv = 0; 1397 if (str != NULL) 1398 kmem_free(str, strlen(str) + 1); 1399 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1400 /* 1401 * Now compare labels to complete the MAC check. If the 1402 * labels are equal then allow access. If the mountpoint 1403 * label dominates the dataset label, allow readonly access. 1404 * Otherwise, access is denied. 1405 */ 1406 if (blequal(mnt_sl, &ds_sl)) 1407 retv = 0; 1408 else if (bldominates(mnt_sl, &ds_sl)) { 1409 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1410 retv = 0; 1411 } 1412 } 1413 1414 label_rele(mnt_tsl); 1415 zone_rele(mntzone); 1416 return (retv); 1417} 1418#endif /* SECLABEL */ 1419 1420#ifdef OPENSOLARIS_MOUNTROOT 1421static int 1422zfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1423{ 1424 int error = 0; 1425 static int zfsrootdone = 0; 1426 zfsvfs_t *zfsvfs = NULL; 1427 znode_t *zp = NULL; 1428 vnode_t *vp = NULL; 1429 char *zfs_bootfs; 1430 char *zfs_devid; 1431 1432 ASSERT(vfsp); 1433 1434 /* 1435 * The filesystem that we mount as root is defined in the 1436 * boot property "zfs-bootfs" with a format of 1437 * "poolname/root-dataset-objnum". 1438 */ 1439 if (why == ROOT_INIT) { 1440 if (zfsrootdone++) 1441 return (SET_ERROR(EBUSY)); 1442 /* 1443 * the process of doing a spa_load will require the 1444 * clock to be set before we could (for example) do 1445 * something better by looking at the timestamp on 1446 * an uberblock, so just set it to -1. 1447 */ 1448 clkset(-1); 1449 1450 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1451 cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1452 "bootfs name"); 1453 return (SET_ERROR(EINVAL)); 1454 } 1455 zfs_devid = spa_get_bootprop("diskdevid"); 1456 error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1457 if (zfs_devid) 1458 spa_free_bootprop(zfs_devid); 1459 if (error) { 1460 spa_free_bootprop(zfs_bootfs); 1461 cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1462 error); 1463 return (error); 1464 } 1465 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1466 spa_free_bootprop(zfs_bootfs); 1467 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1468 error); 1469 return (error); 1470 } 1471 1472 spa_free_bootprop(zfs_bootfs); 1473 1474 if (error = vfs_lock(vfsp)) 1475 return (error); 1476 1477 if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1478 cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1479 goto out; 1480 } 1481 1482 zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1483 ASSERT(zfsvfs); 1484 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1485 cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1486 goto out; 1487 } 1488 1489 vp = ZTOV(zp); 1490 mutex_enter(&vp->v_lock); 1491 vp->v_flag |= VROOT; 1492 mutex_exit(&vp->v_lock); 1493 rootvp = vp; 1494 1495 /* 1496 * Leave rootvp held. The root file system is never unmounted. 1497 */ 1498 1499 vfs_add((struct vnode *)0, vfsp, 1500 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1501out: 1502 vfs_unlock(vfsp); 1503 return (error); 1504 } else if (why == ROOT_REMOUNT) { 1505 readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1506 vfsp->vfs_flag |= VFS_REMOUNT; 1507 1508 /* refresh mount options */ 1509 zfs_unregister_callbacks(vfsp->vfs_data); 1510 return (zfs_register_callbacks(vfsp)); 1511 1512 } else if (why == ROOT_UNMOUNT) { 1513 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1514 (void) zfs_sync(vfsp, 0, 0); 1515 return (0); 1516 } 1517 1518 /* 1519 * if "why" is equal to anything else other than ROOT_INIT, 1520 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1521 */ 1522 return (SET_ERROR(ENOTSUP)); 1523} 1524#endif /* OPENSOLARIS_MOUNTROOT */ 1525 1526static int 1527getpoolname(const char *osname, char *poolname) 1528{ 1529 char *p; 1530 1531 p = strchr(osname, '/'); 1532 if (p == NULL) { 1533 if (strlen(osname) >= MAXNAMELEN) 1534 return (ENAMETOOLONG); 1535 (void) strcpy(poolname, osname); 1536 } else { 1537 if (p - osname >= MAXNAMELEN) 1538 return (ENAMETOOLONG); 1539 (void) strncpy(poolname, osname, p - osname); 1540 poolname[p - osname] = '\0'; 1541 } 1542 return (0); 1543} 1544 1545/*ARGSUSED*/ 1546static int 1547zfs_mount(vfs_t *vfsp) 1548{ 1549 kthread_t *td = curthread; 1550 vnode_t *mvp = vfsp->mnt_vnodecovered; 1551 cred_t *cr = td->td_ucred; 1552 char *osname; 1553 int error = 0; 1554 int canwrite; 1555 1556#ifdef illumos 1557 if (mvp->v_type != VDIR) 1558 return (SET_ERROR(ENOTDIR)); 1559 1560 mutex_enter(&mvp->v_lock); 1561 if ((uap->flags & MS_REMOUNT) == 0 && 1562 (uap->flags & MS_OVERLAY) == 0 && 1563 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 1564 mutex_exit(&mvp->v_lock); 1565 return (SET_ERROR(EBUSY)); 1566 } 1567 mutex_exit(&mvp->v_lock); 1568 1569 /* 1570 * ZFS does not support passing unparsed data in via MS_DATA. 1571 * Users should use the MS_OPTIONSTR interface; this means 1572 * that all option parsing is already done and the options struct 1573 * can be interrogated. 1574 */ 1575 if ((uap->flags & MS_DATA) && uap->datalen > 0) 1576#else /* !illumos */ 1577 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1578 return (SET_ERROR(EPERM)); 1579 1580 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1581 return (SET_ERROR(EINVAL)); 1582#endif /* illumos */ 1583 1584 /* 1585 * If full-owner-access is enabled and delegated administration is 1586 * turned on, we must set nosuid. 1587 */ 1588 if (zfs_super_owner && 1589 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1590 secpolicy_fs_mount_clearopts(cr, vfsp); 1591 } 1592 1593 /* 1594 * Check for mount privilege? 1595 * 1596 * If we don't have privilege then see if 1597 * we have local permission to allow it 1598 */ 1599 error = secpolicy_fs_mount(cr, mvp, vfsp); 1600 if (error) { 1601 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1602 goto out; 1603 1604 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1605 vattr_t vattr; 1606 1607 /* 1608 * Make sure user is the owner of the mount point 1609 * or has sufficient privileges. 1610 */ 1611 1612 vattr.va_mask = AT_UID; 1613 1614 vn_lock(mvp, LK_SHARED | LK_RETRY); 1615 if (VOP_GETATTR(mvp, &vattr, cr)) { 1616 VOP_UNLOCK(mvp, 0); 1617 goto out; 1618 } 1619 1620 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1621 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1622 VOP_UNLOCK(mvp, 0); 1623 goto out; 1624 } 1625 VOP_UNLOCK(mvp, 0); 1626 } 1627 1628 secpolicy_fs_mount_clearopts(cr, vfsp); 1629 } 1630 1631 /* 1632 * Refuse to mount a filesystem if we are in a local zone and the 1633 * dataset is not visible. 1634 */ 1635 if (!INGLOBALZONE(curthread) && 1636 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1637 error = SET_ERROR(EPERM); 1638 goto out; 1639 } 1640 1641#ifdef SECLABEL 1642 error = zfs_mount_label_policy(vfsp, osname); 1643 if (error) 1644 goto out; 1645#endif 1646 1647 vfsp->vfs_flag |= MNT_NFS4ACLS; 1648 1649 /* 1650 * When doing a remount, we simply refresh our temporary properties 1651 * according to those options set in the current VFS options. 1652 */ 1653 if (vfsp->vfs_flag & MS_REMOUNT) { 1654 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1655 1656 /* 1657 * Refresh mount options with z_teardown_lock blocking I/O while 1658 * the filesystem is in an inconsistent state. 1659 * The lock also serializes this code with filesystem 1660 * manipulations between entry to zfs_suspend_fs() and return 1661 * from zfs_resume_fs(). 1662 */ 1663 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1664 zfs_unregister_callbacks(zfsvfs); 1665 error = zfs_register_callbacks(vfsp); 1666 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1667 goto out; 1668 } 1669 1670 /* Initial root mount: try hard to import the requested root pool. */ 1671 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1672 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1673 char pname[MAXNAMELEN]; 1674 1675 error = getpoolname(osname, pname); 1676 if (error == 0) 1677 error = spa_import_rootpool(pname); 1678 if (error) 1679 goto out; 1680 } 1681 DROP_GIANT(); 1682 error = zfs_domount(vfsp, osname); 1683 PICKUP_GIANT(); 1684 1685#ifdef illumos 1686 /* 1687 * Add an extra VFS_HOLD on our parent vfs so that it can't 1688 * disappear due to a forced unmount. 1689 */ 1690 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1691 VFS_HOLD(mvp->v_vfsp); 1692#endif 1693 1694out: 1695 return (error); 1696} 1697 1698static int 1699zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1700{ 1701 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1702 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1703 1704 statp->f_version = STATFS_VERSION; 1705 1706 ZFS_ENTER(zfsvfs); 1707 1708 dmu_objset_space(zfsvfs->z_os, 1709 &refdbytes, &availbytes, &usedobjs, &availobjs); 1710 1711 /* 1712 * The underlying storage pool actually uses multiple block sizes. 1713 * We report the fragsize as the smallest block size we support, 1714 * and we report our blocksize as the filesystem's maximum blocksize. 1715 */ 1716 statp->f_bsize = SPA_MINBLOCKSIZE; 1717 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1718 1719 /* 1720 * The following report "total" blocks of various kinds in the 1721 * file system, but reported in terms of f_frsize - the 1722 * "fragment" size. 1723 */ 1724 1725 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1726 statp->f_bfree = availbytes / statp->f_bsize; 1727 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1728 1729 /* 1730 * statvfs() should really be called statufs(), because it assumes 1731 * static metadata. ZFS doesn't preallocate files, so the best 1732 * we can do is report the max that could possibly fit in f_files, 1733 * and that minus the number actually used in f_ffree. 1734 * For f_ffree, report the smaller of the number of object available 1735 * and the number of blocks (each object will take at least a block). 1736 */ 1737 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1738 statp->f_files = statp->f_ffree + usedobjs; 1739 1740 /* 1741 * We're a zfs filesystem. 1742 */ 1743 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1744 1745 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1746 sizeof(statp->f_mntfromname)); 1747 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1748 sizeof(statp->f_mntonname)); 1749 1750 statp->f_namemax = ZFS_MAXNAMELEN; 1751 1752 ZFS_EXIT(zfsvfs); 1753 return (0); 1754} 1755 1756static int 1757zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1758{ 1759 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1760 znode_t *rootzp; 1761 int error; 1762 1763 ZFS_ENTER(zfsvfs); 1764 1765 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1766 if (error == 0) 1767 *vpp = ZTOV(rootzp); 1768 1769 ZFS_EXIT(zfsvfs); 1770 1771 if (error == 0) { 1772 error = vn_lock(*vpp, flags); 1773 if (error == 0) 1774 (*vpp)->v_vflag |= VV_ROOT; 1775 } 1776 if (error != 0) 1777 *vpp = NULL; 1778 1779 return (error); 1780} 1781 1782/* 1783 * Teardown the zfsvfs::z_os. 1784 * 1785 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1786 * and 'z_teardown_inactive_lock' held. 1787 */ 1788static int 1789zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1790{ 1791 znode_t *zp; 1792 1793 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1794 1795 if (!unmounting) { 1796 /* 1797 * We purge the parent filesystem's vfsp as the parent 1798 * filesystem and all of its snapshots have their vnode's 1799 * v_vfsp set to the parent's filesystem's vfsp. Note, 1800 * 'z_parent' is self referential for non-snapshots. 1801 */ 1802 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1803#ifdef FREEBSD_NAMECACHE 1804 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1805#endif 1806 } 1807 1808 /* 1809 * Close the zil. NB: Can't close the zil while zfs_inactive 1810 * threads are blocked as zil_close can call zfs_inactive. 1811 */ 1812 if (zfsvfs->z_log) { 1813 zil_close(zfsvfs->z_log); 1814 zfsvfs->z_log = NULL; 1815 } 1816 1817 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1818 1819 /* 1820 * If we are not unmounting (ie: online recv) and someone already 1821 * unmounted this file system while we were doing the switcheroo, 1822 * or a reopen of z_os failed then just bail out now. 1823 */ 1824 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1825 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1826 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1827 return (SET_ERROR(EIO)); 1828 } 1829 1830 /* 1831 * At this point there are no vops active, and any new vops will 1832 * fail with EIO since we have z_teardown_lock for writer (only 1833 * relavent for forced unmount). 1834 * 1835 * Release all holds on dbufs. 1836 */ 1837 mutex_enter(&zfsvfs->z_znodes_lock); 1838 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1839 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1840 if (zp->z_sa_hdl) { 1841 ASSERT(ZTOV(zp)->v_count >= 0); 1842 zfs_znode_dmu_fini(zp); 1843 } 1844 mutex_exit(&zfsvfs->z_znodes_lock); 1845 1846 /* 1847 * If we are unmounting, set the unmounted flag and let new vops 1848 * unblock. zfs_inactive will have the unmounted behavior, and all 1849 * other vops will fail with EIO. 1850 */ 1851 if (unmounting) { 1852 zfsvfs->z_unmounted = B_TRUE; 1853 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1854 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1855 } 1856 1857 /* 1858 * z_os will be NULL if there was an error in attempting to reopen 1859 * zfsvfs, so just return as the properties had already been 1860 * unregistered and cached data had been evicted before. 1861 */ 1862 if (zfsvfs->z_os == NULL) 1863 return (0); 1864 1865 /* 1866 * Unregister properties. 1867 */ 1868 zfs_unregister_callbacks(zfsvfs); 1869 1870 /* 1871 * Evict cached data 1872 */ 1873 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1874 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1875 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1876 dmu_objset_evict_dbufs(zfsvfs->z_os); 1877 1878 return (0); 1879} 1880 1881/*ARGSUSED*/ 1882static int 1883zfs_umount(vfs_t *vfsp, int fflag) 1884{ 1885 kthread_t *td = curthread; 1886 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1887 objset_t *os; 1888 cred_t *cr = td->td_ucred; 1889 int ret; 1890 1891 ret = secpolicy_fs_unmount(cr, vfsp); 1892 if (ret) { 1893 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1894 ZFS_DELEG_PERM_MOUNT, cr)) 1895 return (ret); 1896 } 1897 1898 /* 1899 * We purge the parent filesystem's vfsp as the parent filesystem 1900 * and all of its snapshots have their vnode's v_vfsp set to the 1901 * parent's filesystem's vfsp. Note, 'z_parent' is self 1902 * referential for non-snapshots. 1903 */ 1904 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1905 1906 /* 1907 * Unmount any snapshots mounted under .zfs before unmounting the 1908 * dataset itself. 1909 */ 1910 if (zfsvfs->z_ctldir != NULL) { 1911 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1912 return (ret); 1913 ret = vflush(vfsp, 0, 0, td); 1914 ASSERT(ret == EBUSY); 1915 if (!(fflag & MS_FORCE)) { 1916 if (zfsvfs->z_ctldir->v_count > 1) 1917 return (EBUSY); 1918 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1919 } 1920 zfsctl_destroy(zfsvfs); 1921 ASSERT(zfsvfs->z_ctldir == NULL); 1922 } 1923 1924 if (fflag & MS_FORCE) { 1925 /* 1926 * Mark file system as unmounted before calling 1927 * vflush(FORCECLOSE). This way we ensure no future vnops 1928 * will be called and risk operating on DOOMED vnodes. 1929 */ 1930 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1931 zfsvfs->z_unmounted = B_TRUE; 1932 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1933 } 1934 1935 /* 1936 * Flush all the files. 1937 */ 1938 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1939 if (ret != 0) { 1940 if (!zfsvfs->z_issnap) { 1941 zfsctl_create(zfsvfs); 1942 ASSERT(zfsvfs->z_ctldir != NULL); 1943 } 1944 return (ret); 1945 } 1946 1947#ifdef illumos 1948 if (!(fflag & MS_FORCE)) { 1949 /* 1950 * Check the number of active vnodes in the file system. 1951 * Our count is maintained in the vfs structure, but the 1952 * number is off by 1 to indicate a hold on the vfs 1953 * structure itself. 1954 * 1955 * The '.zfs' directory maintains a reference of its 1956 * own, and any active references underneath are 1957 * reflected in the vnode count. 1958 */ 1959 if (zfsvfs->z_ctldir == NULL) { 1960 if (vfsp->vfs_count > 1) 1961 return (SET_ERROR(EBUSY)); 1962 } else { 1963 if (vfsp->vfs_count > 2 || 1964 zfsvfs->z_ctldir->v_count > 1) 1965 return (SET_ERROR(EBUSY)); 1966 } 1967 } 1968#endif 1969 1970 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1971 os = zfsvfs->z_os; 1972 1973 /* 1974 * z_os will be NULL if there was an error in 1975 * attempting to reopen zfsvfs. 1976 */ 1977 if (os != NULL) { 1978 /* 1979 * Unset the objset user_ptr. 1980 */ 1981 mutex_enter(&os->os_user_ptr_lock); 1982 dmu_objset_set_user(os, NULL); 1983 mutex_exit(&os->os_user_ptr_lock); 1984 1985 /* 1986 * Finally release the objset 1987 */ 1988 dmu_objset_disown(os, zfsvfs); 1989 } 1990 1991 /* 1992 * We can now safely destroy the '.zfs' directory node. 1993 */ 1994 if (zfsvfs->z_ctldir != NULL) 1995 zfsctl_destroy(zfsvfs); 1996 if (zfsvfs->z_issnap) { 1997 vnode_t *svp = vfsp->mnt_vnodecovered; 1998 1999 if (svp->v_count >= 2) 2000 VN_RELE(svp); 2001 } 2002 zfs_freevfs(vfsp); 2003 2004 return (0); 2005} 2006 2007static int 2008zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2009{ 2010 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2011 znode_t *zp; 2012 int err; 2013 2014 /* 2015 * zfs_zget() can't operate on virtual entries like .zfs/ or 2016 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2017 * This will make NFS to switch to LOOKUP instead of using VGET. 2018 */ 2019 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || 2020 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) 2021 return (EOPNOTSUPP); 2022 2023 ZFS_ENTER(zfsvfs); 2024 err = zfs_zget(zfsvfs, ino, &zp); 2025 if (err == 0 && zp->z_unlinked) { 2026 VN_RELE(ZTOV(zp)); 2027 err = EINVAL; 2028 } 2029 if (err == 0) 2030 *vpp = ZTOV(zp); 2031 ZFS_EXIT(zfsvfs); 2032 if (err == 0) 2033 err = vn_lock(*vpp, flags); 2034 if (err != 0) 2035 *vpp = NULL; 2036 return (err); 2037} 2038 2039static int 2040zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2041 struct ucred **credanonp, int *numsecflavors, int **secflavors) 2042{ 2043 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2044 2045 /* 2046 * If this is regular file system vfsp is the same as 2047 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2048 * zfsvfs->z_parent->z_vfs represents parent file system 2049 * which we have to use here, because only this file system 2050 * has mnt_export configured. 2051 */ 2052 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2053 credanonp, numsecflavors, secflavors)); 2054} 2055 2056CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2057CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2058 2059static int 2060zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2061{ 2062 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2063 znode_t *zp; 2064 uint64_t object = 0; 2065 uint64_t fid_gen = 0; 2066 uint64_t gen_mask; 2067 uint64_t zp_gen; 2068 int i, err; 2069 2070 *vpp = NULL; 2071 2072 ZFS_ENTER(zfsvfs); 2073 2074 /* 2075 * On FreeBSD we can get snapshot's mount point or its parent file 2076 * system mount point depending if snapshot is already mounted or not. 2077 */ 2078 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2079 zfid_long_t *zlfid = (zfid_long_t *)fidp; 2080 uint64_t objsetid = 0; 2081 uint64_t setgen = 0; 2082 2083 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2084 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2085 2086 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2087 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2088 2089 ZFS_EXIT(zfsvfs); 2090 2091 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2092 if (err) 2093 return (SET_ERROR(EINVAL)); 2094 ZFS_ENTER(zfsvfs); 2095 } 2096 2097 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2098 zfid_short_t *zfid = (zfid_short_t *)fidp; 2099 2100 for (i = 0; i < sizeof (zfid->zf_object); i++) 2101 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2102 2103 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2104 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2105 } else { 2106 ZFS_EXIT(zfsvfs); 2107 return (SET_ERROR(EINVAL)); 2108 } 2109 2110 /* 2111 * A zero fid_gen means we are in .zfs or the .zfs/snapshot 2112 * directory tree. If the object == zfsvfs->z_shares_dir, then 2113 * we are in the .zfs/shares directory tree. 2114 */ 2115 if ((fid_gen == 0 && 2116 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || 2117 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { 2118 *vpp = zfsvfs->z_ctldir; 2119 ASSERT(*vpp != NULL); 2120 if (object == ZFSCTL_INO_SNAPDIR) { 2121 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2122 0, NULL, NULL, NULL, NULL, NULL) == 0); 2123 } else if (object == zfsvfs->z_shares_dir) { 2124 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL, 2125 0, NULL, NULL, NULL, NULL, NULL) == 0); 2126 } else { 2127 VN_HOLD(*vpp); 2128 } 2129 ZFS_EXIT(zfsvfs); 2130 err = vn_lock(*vpp, flags); 2131 if (err != 0) 2132 *vpp = NULL; 2133 return (err); 2134 } 2135 2136 gen_mask = -1ULL >> (64 - 8 * i); 2137 2138 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2139 if (err = zfs_zget(zfsvfs, object, &zp)) { 2140 ZFS_EXIT(zfsvfs); 2141 return (err); 2142 } 2143 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2144 sizeof (uint64_t)); 2145 zp_gen = zp_gen & gen_mask; 2146 if (zp_gen == 0) 2147 zp_gen = 1; 2148 if (zp->z_unlinked || zp_gen != fid_gen) { 2149 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2150 VN_RELE(ZTOV(zp)); 2151 ZFS_EXIT(zfsvfs); 2152 return (SET_ERROR(EINVAL)); 2153 } 2154 2155 *vpp = ZTOV(zp); 2156 ZFS_EXIT(zfsvfs); 2157 err = vn_lock(*vpp, flags | LK_RETRY); 2158 if (err == 0) 2159 vnode_create_vobject(*vpp, zp->z_size, curthread); 2160 else 2161 *vpp = NULL; 2162 return (err); 2163} 2164 2165/* 2166 * Block out VOPs and close zfsvfs_t::z_os 2167 * 2168 * Note, if successful, then we return with the 'z_teardown_lock' and 2169 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 2170 * dataset and objset intact so that they can be atomically handed off during 2171 * a subsequent rollback or recv operation and the resume thereafter. 2172 */ 2173int 2174zfs_suspend_fs(zfsvfs_t *zfsvfs) 2175{ 2176 int error; 2177 2178 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2179 return (error); 2180 2181 return (0); 2182} 2183 2184/* 2185 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 2186 * is an invariant across any of the operations that can be performed while the 2187 * filesystem was suspended. Whether it succeeded or failed, the preconditions 2188 * are the same: the relevant objset and associated dataset are owned by 2189 * zfsvfs, held, and long held on entry. 2190 */ 2191int 2192zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2193{ 2194 int err; 2195 znode_t *zp; 2196 uint64_t sa_obj = 0; 2197 2198 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2199 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2200 2201 /* 2202 * We already own this, so just hold and rele it to update the 2203 * objset_t, as the one we had before may have been evicted. 2204 */ 2205 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os)); 2206 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs); 2207 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset)); 2208 dmu_objset_rele(zfsvfs->z_os, zfsvfs); 2209 2210 /* 2211 * Make sure version hasn't changed 2212 */ 2213 2214 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, 2215 &zfsvfs->z_version); 2216 2217 if (err) 2218 goto bail; 2219 2220 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, 2221 ZFS_SA_ATTRS, 8, 1, &sa_obj); 2222 2223 if (err && zfsvfs->z_version >= ZPL_VERSION_SA) 2224 goto bail; 2225 2226 if ((err = sa_setup(zfsvfs->z_os, sa_obj, 2227 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) 2228 goto bail; 2229 2230 if (zfsvfs->z_version >= ZPL_VERSION_SA) 2231 sa_register_update_callback(zfsvfs->z_os, 2232 zfs_sa_upgrade); 2233 2234 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2235 2236 zfs_set_fuid_feature(zfsvfs); 2237 2238 /* 2239 * Attempt to re-establish all the active znodes with 2240 * their dbufs. If a zfs_rezget() fails, then we'll let 2241 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2242 * when they try to use their znode. 2243 */ 2244 mutex_enter(&zfsvfs->z_znodes_lock); 2245 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2246 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2247 (void) zfs_rezget(zp); 2248 } 2249 mutex_exit(&zfsvfs->z_znodes_lock); 2250 2251bail: 2252 /* release the VOPs */ 2253 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2254 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 2255 2256 if (err) { 2257 /* 2258 * Since we couldn't setup the sa framework, try to force 2259 * unmount this file system. 2260 */ 2261 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) { 2262 vfs_ref(zfsvfs->z_vfs); 2263 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2264 } 2265 } 2266 return (err); 2267} 2268 2269static void 2270zfs_freevfs(vfs_t *vfsp) 2271{ 2272 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2273 2274#ifdef illumos 2275 /* 2276 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2277 * from zfs_mount(). Release it here. If we came through 2278 * zfs_mountroot() instead, we didn't grab an extra hold, so 2279 * skip the VFS_RELE for rootvfs. 2280 */ 2281 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2282 VFS_RELE(zfsvfs->z_parent->z_vfs); 2283#endif 2284 2285 zfsvfs_free(zfsvfs); 2286 2287 atomic_dec_32(&zfs_active_fs_count); 2288} 2289 2290#ifdef __i386__ 2291static int desiredvnodes_backup; 2292#endif 2293 2294static void 2295zfs_vnodes_adjust(void) 2296{ 2297#ifdef __i386__ 2298 int newdesiredvnodes; 2299 2300 desiredvnodes_backup = desiredvnodes; 2301 2302 /* 2303 * We calculate newdesiredvnodes the same way it is done in 2304 * vntblinit(). If it is equal to desiredvnodes, it means that 2305 * it wasn't tuned by the administrator and we can tune it down. 2306 */ 2307 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2308 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2309 sizeof(struct vnode)))); 2310 if (newdesiredvnodes == desiredvnodes) 2311 desiredvnodes = (3 * newdesiredvnodes) / 4; 2312#endif 2313} 2314 2315static void 2316zfs_vnodes_adjust_back(void) 2317{ 2318 2319#ifdef __i386__ 2320 desiredvnodes = desiredvnodes_backup; 2321#endif 2322} 2323 2324void 2325zfs_init(void) 2326{ 2327 2328 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2329 2330 /* 2331 * Initialize .zfs directory structures 2332 */ 2333 zfsctl_init(); 2334 2335 /* 2336 * Initialize znode cache, vnode ops, etc... 2337 */ 2338 zfs_znode_init(); 2339 2340 /* 2341 * Reduce number of vnodes. Originally number of vnodes is calculated 2342 * with UFS inode in mind. We reduce it here, because it's too big for 2343 * ZFS/i386. 2344 */ 2345 zfs_vnodes_adjust(); 2346 2347 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2348} 2349 2350void 2351zfs_fini(void) 2352{ 2353 zfsctl_fini(); 2354 zfs_znode_fini(); 2355 zfs_vnodes_adjust_back(); 2356} 2357 2358int 2359zfs_busy(void) 2360{ 2361 return (zfs_active_fs_count != 0); 2362} 2363 2364int 2365zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2366{ 2367 int error; 2368 objset_t *os = zfsvfs->z_os; 2369 dmu_tx_t *tx; 2370 2371 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2372 return (SET_ERROR(EINVAL)); 2373 2374 if (newvers < zfsvfs->z_version) 2375 return (SET_ERROR(EINVAL)); 2376 2377 if (zfs_spa_version_map(newvers) > 2378 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2379 return (SET_ERROR(ENOTSUP)); 2380 2381 tx = dmu_tx_create(os); 2382 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2383 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2384 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2385 ZFS_SA_ATTRS); 2386 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2387 } 2388 error = dmu_tx_assign(tx, TXG_WAIT); 2389 if (error) { 2390 dmu_tx_abort(tx); 2391 return (error); 2392 } 2393 2394 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2395 8, 1, &newvers, tx); 2396 2397 if (error) { 2398 dmu_tx_commit(tx); 2399 return (error); 2400 } 2401 2402 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2403 uint64_t sa_obj; 2404 2405 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2406 SPA_VERSION_SA); 2407 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2408 DMU_OT_NONE, 0, tx); 2409 2410 error = zap_add(os, MASTER_NODE_OBJ, 2411 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2412 ASSERT0(error); 2413 2414 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2415 sa_register_update_callback(os, zfs_sa_upgrade); 2416 } 2417 2418 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2419 "from %llu to %llu", zfsvfs->z_version, newvers); 2420 2421 dmu_tx_commit(tx); 2422 2423 zfsvfs->z_version = newvers; 2424 2425 zfs_set_fuid_feature(zfsvfs); 2426 2427 return (0); 2428} 2429 2430/* 2431 * Read a property stored within the master node. 2432 */ 2433int 2434zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2435{ 2436 const char *pname; 2437 int error = ENOENT; 2438 2439 /* 2440 * Look up the file system's value for the property. For the 2441 * version property, we look up a slightly different string. 2442 */ 2443 if (prop == ZFS_PROP_VERSION) 2444 pname = ZPL_VERSION_STR; 2445 else 2446 pname = zfs_prop_to_name(prop); 2447 2448 if (os != NULL) 2449 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2450 2451 if (error == ENOENT) { 2452 /* No value set, use the default value */ 2453 switch (prop) { 2454 case ZFS_PROP_VERSION: 2455 *value = ZPL_VERSION; 2456 break; 2457 case ZFS_PROP_NORMALIZE: 2458 case ZFS_PROP_UTF8ONLY: 2459 *value = 0; 2460 break; 2461 case ZFS_PROP_CASE: 2462 *value = ZFS_CASE_SENSITIVE; 2463 break; 2464 default: 2465 return (error); 2466 } 2467 error = 0; 2468 } 2469 return (error); 2470} 2471 2472#ifdef _KERNEL 2473void 2474zfsvfs_update_fromname(const char *oldname, const char *newname) 2475{ 2476 char tmpbuf[MAXPATHLEN]; 2477 struct mount *mp; 2478 char *fromname; 2479 size_t oldlen; 2480 2481 oldlen = strlen(oldname); 2482 2483 mtx_lock(&mountlist_mtx); 2484 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2485 fromname = mp->mnt_stat.f_mntfromname; 2486 if (strcmp(fromname, oldname) == 0) { 2487 (void)strlcpy(fromname, newname, 2488 sizeof(mp->mnt_stat.f_mntfromname)); 2489 continue; 2490 } 2491 if (strncmp(fromname, oldname, oldlen) == 0 && 2492 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2493 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2494 newname, fromname + oldlen); 2495 (void)strlcpy(fromname, tmpbuf, 2496 sizeof(mp->mnt_stat.f_mntfromname)); 2497 continue; 2498 } 2499 } 2500 mtx_unlock(&mountlist_mtx); 2501} 2502#endif 2503