zfs_vfsops.c revision 284203
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) 2013 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 * nbmand is a special property. It can only be changed at 470 * mount time. 471 * 472 * This is weird, but it is documented to only be changeable 473 * at mount time. 474 */ 475 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 476 nbmand = B_FALSE; 477 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 478 nbmand = B_TRUE; 479 } else { 480 char osname[MAXNAMELEN]; 481 482 dmu_objset_name(os, osname); 483 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand, 484 NULL)) { 485 return (error); 486 } 487 } 488 489 /* 490 * Register property callbacks. 491 * 492 * It would probably be fine to just check for i/o error from 493 * the first prop_register(), but I guess I like to go 494 * overboard... 495 */ 496 ds = dmu_objset_ds(os); 497 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 498 error = dsl_prop_register(ds, 499 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 500 error = error ? error : dsl_prop_register(ds, 501 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 502 error = error ? error : dsl_prop_register(ds, 503 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 504 error = error ? error : dsl_prop_register(ds, 505 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 506#ifdef illumos 507 error = error ? error : dsl_prop_register(ds, 508 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 509#endif 510 error = error ? error : dsl_prop_register(ds, 511 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 512 error = error ? error : dsl_prop_register(ds, 513 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 514 error = error ? error : dsl_prop_register(ds, 515 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 516 error = error ? error : dsl_prop_register(ds, 517 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 518 error = error ? error : dsl_prop_register(ds, 519 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 520 zfsvfs); 521 error = error ? error : dsl_prop_register(ds, 522 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); 523 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 524 if (error) 525 goto unregister; 526 527 /* 528 * Invoke our callbacks to restore temporary mount options. 529 */ 530 if (do_readonly) 531 readonly_changed_cb(zfsvfs, readonly); 532 if (do_setuid) 533 setuid_changed_cb(zfsvfs, setuid); 534 if (do_exec) 535 exec_changed_cb(zfsvfs, exec); 536 if (do_xattr) 537 xattr_changed_cb(zfsvfs, xattr); 538 if (do_atime) 539 atime_changed_cb(zfsvfs, atime); 540 541 nbmand_changed_cb(zfsvfs, nbmand); 542 543 return (0); 544 545unregister: 546 /* 547 * We may attempt to unregister some callbacks that are not 548 * registered, but this is OK; it will simply return ENOMSG, 549 * which we will ignore. 550 */ 551 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME), 552 atime_changed_cb, zfsvfs); 553 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR), 554 xattr_changed_cb, zfsvfs); 555 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 556 blksz_changed_cb, zfsvfs); 557 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY), 558 readonly_changed_cb, zfsvfs); 559#ifdef illumos 560 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES), 561 devices_changed_cb, zfsvfs); 562#endif 563 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID), 564 setuid_changed_cb, zfsvfs); 565 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC), 566 exec_changed_cb, zfsvfs); 567 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR), 568 snapdir_changed_cb, zfsvfs); 569 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE), 570 acl_mode_changed_cb, zfsvfs); 571 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT), 572 acl_inherit_changed_cb, zfsvfs); 573 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN), 574 vscan_changed_cb, zfsvfs); 575 return (error); 576} 577 578static int 579zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, 580 uint64_t *userp, uint64_t *groupp) 581{ 582 /* 583 * Is it a valid type of object to track? 584 */ 585 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 586 return (SET_ERROR(ENOENT)); 587 588 /* 589 * If we have a NULL data pointer 590 * then assume the id's aren't changing and 591 * return EEXIST to the dmu to let it know to 592 * use the same ids 593 */ 594 if (data == NULL) 595 return (SET_ERROR(EEXIST)); 596 597 if (bonustype == DMU_OT_ZNODE) { 598 znode_phys_t *znp = data; 599 *userp = znp->zp_uid; 600 *groupp = znp->zp_gid; 601 } else { 602 int hdrsize; 603 sa_hdr_phys_t *sap = data; 604 sa_hdr_phys_t sa = *sap; 605 boolean_t swap = B_FALSE; 606 607 ASSERT(bonustype == DMU_OT_SA); 608 609 if (sa.sa_magic == 0) { 610 /* 611 * This should only happen for newly created 612 * files that haven't had the znode data filled 613 * in yet. 614 */ 615 *userp = 0; 616 *groupp = 0; 617 return (0); 618 } 619 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { 620 sa.sa_magic = SA_MAGIC; 621 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); 622 swap = B_TRUE; 623 } else { 624 VERIFY3U(sa.sa_magic, ==, SA_MAGIC); 625 } 626 627 hdrsize = sa_hdrsize(&sa); 628 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); 629 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + 630 SA_UID_OFFSET)); 631 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + 632 SA_GID_OFFSET)); 633 if (swap) { 634 *userp = BSWAP_64(*userp); 635 *groupp = BSWAP_64(*groupp); 636 } 637 } 638 return (0); 639} 640 641static void 642fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, 643 char *domainbuf, int buflen, uid_t *ridp) 644{ 645 uint64_t fuid; 646 const char *domain; 647 648 fuid = strtonum(fuidstr, NULL); 649 650 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); 651 if (domain) 652 (void) strlcpy(domainbuf, domain, buflen); 653 else 654 domainbuf[0] = '\0'; 655 *ridp = FUID_RID(fuid); 656} 657 658static uint64_t 659zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) 660{ 661 switch (type) { 662 case ZFS_PROP_USERUSED: 663 return (DMU_USERUSED_OBJECT); 664 case ZFS_PROP_GROUPUSED: 665 return (DMU_GROUPUSED_OBJECT); 666 case ZFS_PROP_USERQUOTA: 667 return (zfsvfs->z_userquota_obj); 668 case ZFS_PROP_GROUPQUOTA: 669 return (zfsvfs->z_groupquota_obj); 670 } 671 return (0); 672} 673 674int 675zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 676 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) 677{ 678 int error; 679 zap_cursor_t zc; 680 zap_attribute_t za; 681 zfs_useracct_t *buf = vbuf; 682 uint64_t obj; 683 684 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 685 return (SET_ERROR(ENOTSUP)); 686 687 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 688 if (obj == 0) { 689 *bufsizep = 0; 690 return (0); 691 } 692 693 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); 694 (error = zap_cursor_retrieve(&zc, &za)) == 0; 695 zap_cursor_advance(&zc)) { 696 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > 697 *bufsizep) 698 break; 699 700 fuidstr_to_sid(zfsvfs, za.za_name, 701 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); 702 703 buf->zu_space = za.za_first_integer; 704 buf++; 705 } 706 if (error == ENOENT) 707 error = 0; 708 709 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); 710 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; 711 *cookiep = zap_cursor_serialize(&zc); 712 zap_cursor_fini(&zc); 713 return (error); 714} 715 716/* 717 * buf must be big enough (eg, 32 bytes) 718 */ 719static int 720id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, 721 char *buf, boolean_t addok) 722{ 723 uint64_t fuid; 724 int domainid = 0; 725 726 if (domain && domain[0]) { 727 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); 728 if (domainid == -1) 729 return (SET_ERROR(ENOENT)); 730 } 731 fuid = FUID_ENCODE(domainid, rid); 732 (void) sprintf(buf, "%llx", (longlong_t)fuid); 733 return (0); 734} 735 736int 737zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 738 const char *domain, uint64_t rid, uint64_t *valp) 739{ 740 char buf[32]; 741 int err; 742 uint64_t obj; 743 744 *valp = 0; 745 746 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 747 return (SET_ERROR(ENOTSUP)); 748 749 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 750 if (obj == 0) 751 return (0); 752 753 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); 754 if (err) 755 return (err); 756 757 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); 758 if (err == ENOENT) 759 err = 0; 760 return (err); 761} 762 763int 764zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 765 const char *domain, uint64_t rid, uint64_t quota) 766{ 767 char buf[32]; 768 int err; 769 dmu_tx_t *tx; 770 uint64_t *objp; 771 boolean_t fuid_dirtied; 772 773 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) 774 return (SET_ERROR(EINVAL)); 775 776 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) 777 return (SET_ERROR(ENOTSUP)); 778 779 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : 780 &zfsvfs->z_groupquota_obj; 781 782 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); 783 if (err) 784 return (err); 785 fuid_dirtied = zfsvfs->z_fuid_dirty; 786 787 tx = dmu_tx_create(zfsvfs->z_os); 788 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); 789 if (*objp == 0) { 790 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 791 zfs_userquota_prop_prefixes[type]); 792 } 793 if (fuid_dirtied) 794 zfs_fuid_txhold(zfsvfs, tx); 795 err = dmu_tx_assign(tx, TXG_WAIT); 796 if (err) { 797 dmu_tx_abort(tx); 798 return (err); 799 } 800 801 mutex_enter(&zfsvfs->z_lock); 802 if (*objp == 0) { 803 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, 804 DMU_OT_NONE, 0, tx); 805 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 806 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); 807 } 808 mutex_exit(&zfsvfs->z_lock); 809 810 if (quota == 0) { 811 err = zap_remove(zfsvfs->z_os, *objp, buf, tx); 812 if (err == ENOENT) 813 err = 0; 814 } else { 815 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); 816 } 817 ASSERT(err == 0); 818 if (fuid_dirtied) 819 zfs_fuid_sync(zfsvfs, tx); 820 dmu_tx_commit(tx); 821 return (err); 822} 823 824boolean_t 825zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) 826{ 827 char buf[32]; 828 uint64_t used, quota, usedobj, quotaobj; 829 int err; 830 831 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; 832 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 833 834 if (quotaobj == 0 || zfsvfs->z_replay) 835 return (B_FALSE); 836 837 (void) sprintf(buf, "%llx", (longlong_t)fuid); 838 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); 839 if (err != 0) 840 return (B_FALSE); 841 842 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); 843 if (err != 0) 844 return (B_FALSE); 845 return (used >= quota); 846} 847 848boolean_t 849zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) 850{ 851 uint64_t fuid; 852 uint64_t quotaobj; 853 854 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 855 856 fuid = isgroup ? zp->z_gid : zp->z_uid; 857 858 if (quotaobj == 0 || zfsvfs->z_replay) 859 return (B_FALSE); 860 861 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); 862} 863 864int 865zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 866{ 867 objset_t *os; 868 zfsvfs_t *zfsvfs; 869 uint64_t zval; 870 int i, error; 871 uint64_t sa_obj; 872 873 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 874 875 /* 876 * We claim to always be readonly so we can open snapshots; 877 * other ZPL code will prevent us from writing to snapshots. 878 */ 879 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 880 if (error) { 881 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 882 return (error); 883 } 884 885 /* 886 * Initialize the zfs-specific filesystem structure. 887 * Should probably make this a kmem cache, shuffle fields, 888 * and just bzero up to z_hold_mtx[]. 889 */ 890 zfsvfs->z_vfs = NULL; 891 zfsvfs->z_parent = zfsvfs; 892 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 893 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 894 zfsvfs->z_os = os; 895 896 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 897 if (error) { 898 goto out; 899 } else if (zfsvfs->z_version > 900 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 901 (void) printf("Can't mount a version %lld file system " 902 "on a version %lld pool\n. Pool must be upgraded to mount " 903 "this file system.", (u_longlong_t)zfsvfs->z_version, 904 (u_longlong_t)spa_version(dmu_objset_spa(os))); 905 error = SET_ERROR(ENOTSUP); 906 goto out; 907 } 908 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) 909 goto out; 910 zfsvfs->z_norm = (int)zval; 911 912 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) 913 goto out; 914 zfsvfs->z_utf8 = (zval != 0); 915 916 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) 917 goto out; 918 zfsvfs->z_case = (uint_t)zval; 919 920 /* 921 * Fold case on file systems that are always or sometimes case 922 * insensitive. 923 */ 924 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 925 zfsvfs->z_case == ZFS_CASE_MIXED) 926 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 927 928 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 929 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 930 931 if (zfsvfs->z_use_sa) { 932 /* should either have both of these objects or none */ 933 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 934 &sa_obj); 935 if (error) 936 return (error); 937 } else { 938 /* 939 * Pre SA versions file systems should never touch 940 * either the attribute registration or layout objects. 941 */ 942 sa_obj = 0; 943 } 944 945 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 946 &zfsvfs->z_attr_table); 947 if (error) 948 goto out; 949 950 if (zfsvfs->z_version >= ZPL_VERSION_SA) 951 sa_register_update_callback(os, zfs_sa_upgrade); 952 953 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 954 &zfsvfs->z_root); 955 if (error) 956 goto out; 957 ASSERT(zfsvfs->z_root != 0); 958 959 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 960 &zfsvfs->z_unlinkedobj); 961 if (error) 962 goto out; 963 964 error = zap_lookup(os, MASTER_NODE_OBJ, 965 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 966 8, 1, &zfsvfs->z_userquota_obj); 967 if (error && error != ENOENT) 968 goto out; 969 970 error = zap_lookup(os, MASTER_NODE_OBJ, 971 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 972 8, 1, &zfsvfs->z_groupquota_obj); 973 if (error && error != ENOENT) 974 goto out; 975 976 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 977 &zfsvfs->z_fuid_obj); 978 if (error && error != ENOENT) 979 goto out; 980 981 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 982 &zfsvfs->z_shares_dir); 983 if (error && error != ENOENT) 984 goto out; 985 986 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 987 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 988 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 989 offsetof(znode_t, z_link_node)); 990 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE); 991 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 992 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 993 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 994 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 995 996 *zfvp = zfsvfs; 997 return (0); 998 999out: 1000 dmu_objset_disown(os, zfsvfs); 1001 *zfvp = NULL; 1002 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1003 return (error); 1004} 1005 1006static int 1007zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 1008{ 1009 int error; 1010 1011 error = zfs_register_callbacks(zfsvfs->z_vfs); 1012 if (error) 1013 return (error); 1014 1015 /* 1016 * Set the objset user_ptr to track its zfsvfs. 1017 */ 1018 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1019 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1020 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1021 1022 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 1023 1024 /* 1025 * If we are not mounting (ie: online recv), then we don't 1026 * have to worry about replaying the log as we blocked all 1027 * operations out since we closed the ZIL. 1028 */ 1029 if (mounting) { 1030 boolean_t readonly; 1031 1032 /* 1033 * During replay we remove the read only flag to 1034 * allow replays to succeed. 1035 */ 1036 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1037 if (readonly != 0) 1038 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1039 else 1040 zfs_unlinked_drain(zfsvfs); 1041 1042 /* 1043 * Parse and replay the intent log. 1044 * 1045 * Because of ziltest, this must be done after 1046 * zfs_unlinked_drain(). (Further note: ziltest 1047 * doesn't use readonly mounts, where 1048 * zfs_unlinked_drain() isn't called.) This is because 1049 * ziltest causes spa_sync() to think it's committed, 1050 * but actually it is not, so the intent log contains 1051 * many txg's worth of changes. 1052 * 1053 * In particular, if object N is in the unlinked set in 1054 * the last txg to actually sync, then it could be 1055 * actually freed in a later txg and then reallocated 1056 * in a yet later txg. This would write a "create 1057 * object N" record to the intent log. Normally, this 1058 * would be fine because the spa_sync() would have 1059 * written out the fact that object N is free, before 1060 * we could write the "create object N" intent log 1061 * record. 1062 * 1063 * But when we are in ziltest mode, we advance the "open 1064 * txg" without actually spa_sync()-ing the changes to 1065 * disk. So we would see that object N is still 1066 * allocated and in the unlinked set, and there is an 1067 * intent log record saying to allocate it. 1068 */ 1069 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1070 if (zil_replay_disable) { 1071 zil_destroy(zfsvfs->z_log, B_FALSE); 1072 } else { 1073 zfsvfs->z_replay = B_TRUE; 1074 zil_replay(zfsvfs->z_os, zfsvfs, 1075 zfs_replay_vector); 1076 zfsvfs->z_replay = B_FALSE; 1077 } 1078 } 1079 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1080 } 1081 1082 return (0); 1083} 1084 1085extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1086 1087void 1088zfsvfs_free(zfsvfs_t *zfsvfs) 1089{ 1090 int i; 1091 1092 /* 1093 * This is a barrier to prevent the filesystem from going away in 1094 * zfs_znode_move() until we can safely ensure that the filesystem is 1095 * not unmounted. We consider the filesystem valid before the barrier 1096 * and invalid after the barrier. 1097 */ 1098 rw_enter(&zfsvfs_lock, RW_READER); 1099 rw_exit(&zfsvfs_lock); 1100 1101 zfs_fuid_destroy(zfsvfs); 1102 1103 mutex_destroy(&zfsvfs->z_znodes_lock); 1104 mutex_destroy(&zfsvfs->z_lock); 1105 list_destroy(&zfsvfs->z_all_znodes); 1106 rrm_destroy(&zfsvfs->z_teardown_lock); 1107 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1108 rw_destroy(&zfsvfs->z_fuid_lock); 1109 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1110 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1111 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1112} 1113 1114static void 1115zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1116{ 1117 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1118 if (zfsvfs->z_vfs) { 1119 if (zfsvfs->z_use_fuids) { 1120 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1121 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1122 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1123 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1124 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1125 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1126 } else { 1127 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1128 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1129 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1130 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1131 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1132 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1133 } 1134 } 1135 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1136} 1137 1138static int 1139zfs_domount(vfs_t *vfsp, char *osname) 1140{ 1141 uint64_t recordsize, fsid_guid; 1142 int error = 0; 1143 zfsvfs_t *zfsvfs; 1144 vnode_t *vp; 1145 1146 ASSERT(vfsp); 1147 ASSERT(osname); 1148 1149 error = zfsvfs_create(osname, &zfsvfs); 1150 if (error) 1151 return (error); 1152 zfsvfs->z_vfs = vfsp; 1153 1154#ifdef illumos 1155 /* Initialize the generic filesystem structure. */ 1156 vfsp->vfs_bcount = 0; 1157 vfsp->vfs_data = NULL; 1158 1159 if (zfs_create_unique_device(&mount_dev) == -1) { 1160 error = SET_ERROR(ENODEV); 1161 goto out; 1162 } 1163 ASSERT(vfs_devismounted(mount_dev) == 0); 1164#endif 1165 1166 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1167 NULL)) 1168 goto out; 1169 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1170 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1171 1172 vfsp->vfs_data = zfsvfs; 1173 vfsp->mnt_flag |= MNT_LOCAL; 1174 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1175 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1176 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1177 1178 /* 1179 * The fsid is 64 bits, composed of an 8-bit fs type, which 1180 * separates our fsid from any other filesystem types, and a 1181 * 56-bit objset unique ID. The objset unique ID is unique to 1182 * all objsets open on this system, provided by unique_create(). 1183 * The 8-bit fs type must be put in the low bits of fsid[1] 1184 * because that's where other Solaris filesystems put it. 1185 */ 1186 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1187 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1188 vfsp->vfs_fsid.val[0] = fsid_guid; 1189 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1190 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1191 1192 /* 1193 * Set features for file system. 1194 */ 1195 zfs_set_fuid_feature(zfsvfs); 1196 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1197 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1198 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1199 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1200 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1201 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1202 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1203 } 1204 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1205 1206 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1207 uint64_t pval; 1208 1209 atime_changed_cb(zfsvfs, B_FALSE); 1210 readonly_changed_cb(zfsvfs, B_TRUE); 1211 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1212 goto out; 1213 xattr_changed_cb(zfsvfs, pval); 1214 zfsvfs->z_issnap = B_TRUE; 1215 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1216 1217 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1218 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1219 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1220 } else { 1221 error = zfsvfs_setup(zfsvfs, B_TRUE); 1222 } 1223 1224 vfs_mountedfrom(vfsp, osname); 1225 1226 if (!zfsvfs->z_issnap) 1227 zfsctl_create(zfsvfs); 1228out: 1229 if (error) { 1230 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1231 zfsvfs_free(zfsvfs); 1232 } else { 1233 atomic_inc_32(&zfs_active_fs_count); 1234 } 1235 1236 return (error); 1237} 1238 1239void 1240zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1241{ 1242 objset_t *os = zfsvfs->z_os; 1243 struct dsl_dataset *ds; 1244 1245 /* 1246 * Unregister properties. 1247 */ 1248 if (!dmu_objset_is_snapshot(os)) { 1249 ds = dmu_objset_ds(os); 1250 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, 1251 zfsvfs) == 0); 1252 1253 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, 1254 zfsvfs) == 0); 1255 1256 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, 1257 zfsvfs) == 0); 1258 1259 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, 1260 zfsvfs) == 0); 1261 1262 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, 1263 zfsvfs) == 0); 1264 1265 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, 1266 zfsvfs) == 0); 1267 1268 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, 1269 zfsvfs) == 0); 1270 1271 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, 1272 zfsvfs) == 0); 1273 1274 VERIFY(dsl_prop_unregister(ds, "aclinherit", 1275 acl_inherit_changed_cb, zfsvfs) == 0); 1276 1277 VERIFY(dsl_prop_unregister(ds, "vscan", 1278 vscan_changed_cb, zfsvfs) == 0); 1279 } 1280} 1281 1282#ifdef SECLABEL 1283/* 1284 * Convert a decimal digit string to a uint64_t integer. 1285 */ 1286static int 1287str_to_uint64(char *str, uint64_t *objnum) 1288{ 1289 uint64_t num = 0; 1290 1291 while (*str) { 1292 if (*str < '0' || *str > '9') 1293 return (SET_ERROR(EINVAL)); 1294 1295 num = num*10 + *str++ - '0'; 1296 } 1297 1298 *objnum = num; 1299 return (0); 1300} 1301 1302/* 1303 * The boot path passed from the boot loader is in the form of 1304 * "rootpool-name/root-filesystem-object-number'. Convert this 1305 * string to a dataset name: "rootpool-name/root-filesystem-name". 1306 */ 1307static int 1308zfs_parse_bootfs(char *bpath, char *outpath) 1309{ 1310 char *slashp; 1311 uint64_t objnum; 1312 int error; 1313 1314 if (*bpath == 0 || *bpath == '/') 1315 return (SET_ERROR(EINVAL)); 1316 1317 (void) strcpy(outpath, bpath); 1318 1319 slashp = strchr(bpath, '/'); 1320 1321 /* if no '/', just return the pool name */ 1322 if (slashp == NULL) { 1323 return (0); 1324 } 1325 1326 /* if not a number, just return the root dataset name */ 1327 if (str_to_uint64(slashp+1, &objnum)) { 1328 return (0); 1329 } 1330 1331 *slashp = '\0'; 1332 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1333 *slashp = '/'; 1334 1335 return (error); 1336} 1337 1338/* 1339 * Check that the hex label string is appropriate for the dataset being 1340 * mounted into the global_zone proper. 1341 * 1342 * Return an error if the hex label string is not default or 1343 * admin_low/admin_high. For admin_low labels, the corresponding 1344 * dataset must be readonly. 1345 */ 1346int 1347zfs_check_global_label(const char *dsname, const char *hexsl) 1348{ 1349 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1350 return (0); 1351 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1352 return (0); 1353 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1354 /* must be readonly */ 1355 uint64_t rdonly; 1356 1357 if (dsl_prop_get_integer(dsname, 1358 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1359 return (SET_ERROR(EACCES)); 1360 return (rdonly ? 0 : EACCES); 1361 } 1362 return (SET_ERROR(EACCES)); 1363} 1364 1365/* 1366 * Determine whether the mount is allowed according to MAC check. 1367 * by comparing (where appropriate) label of the dataset against 1368 * the label of the zone being mounted into. If the dataset has 1369 * no label, create one. 1370 * 1371 * Returns 0 if access allowed, error otherwise (e.g. EACCES) 1372 */ 1373static int 1374zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1375{ 1376 int error, retv; 1377 zone_t *mntzone = NULL; 1378 ts_label_t *mnt_tsl; 1379 bslabel_t *mnt_sl; 1380 bslabel_t ds_sl; 1381 char ds_hexsl[MAXNAMELEN]; 1382 1383 retv = EACCES; /* assume the worst */ 1384 1385 /* 1386 * Start by getting the dataset label if it exists. 1387 */ 1388 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1389 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1390 if (error) 1391 return (SET_ERROR(EACCES)); 1392 1393 /* 1394 * If labeling is NOT enabled, then disallow the mount of datasets 1395 * which have a non-default label already. No other label checks 1396 * are needed. 1397 */ 1398 if (!is_system_labeled()) { 1399 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1400 return (0); 1401 return (SET_ERROR(EACCES)); 1402 } 1403 1404 /* 1405 * Get the label of the mountpoint. If mounting into the global 1406 * zone (i.e. mountpoint is not within an active zone and the 1407 * zoned property is off), the label must be default or 1408 * admin_low/admin_high only; no other checks are needed. 1409 */ 1410 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1411 if (mntzone->zone_id == GLOBAL_ZONEID) { 1412 uint64_t zoned; 1413 1414 zone_rele(mntzone); 1415 1416 if (dsl_prop_get_integer(osname, 1417 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1418 return (SET_ERROR(EACCES)); 1419 if (!zoned) 1420 return (zfs_check_global_label(osname, ds_hexsl)); 1421 else 1422 /* 1423 * This is the case of a zone dataset being mounted 1424 * initially, before the zone has been fully created; 1425 * allow this mount into global zone. 1426 */ 1427 return (0); 1428 } 1429 1430 mnt_tsl = mntzone->zone_slabel; 1431 ASSERT(mnt_tsl != NULL); 1432 label_hold(mnt_tsl); 1433 mnt_sl = label2bslabel(mnt_tsl); 1434 1435 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1436 /* 1437 * The dataset doesn't have a real label, so fabricate one. 1438 */ 1439 char *str = NULL; 1440 1441 if (l_to_str_internal(mnt_sl, &str) == 0 && 1442 dsl_prop_set_string(osname, 1443 zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1444 ZPROP_SRC_LOCAL, str) == 0) 1445 retv = 0; 1446 if (str != NULL) 1447 kmem_free(str, strlen(str) + 1); 1448 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1449 /* 1450 * Now compare labels to complete the MAC check. If the 1451 * labels are equal then allow access. If the mountpoint 1452 * label dominates the dataset label, allow readonly access. 1453 * Otherwise, access is denied. 1454 */ 1455 if (blequal(mnt_sl, &ds_sl)) 1456 retv = 0; 1457 else if (bldominates(mnt_sl, &ds_sl)) { 1458 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1459 retv = 0; 1460 } 1461 } 1462 1463 label_rele(mnt_tsl); 1464 zone_rele(mntzone); 1465 return (retv); 1466} 1467#endif /* SECLABEL */ 1468 1469#ifdef OPENSOLARIS_MOUNTROOT 1470static int 1471zfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1472{ 1473 int error = 0; 1474 static int zfsrootdone = 0; 1475 zfsvfs_t *zfsvfs = NULL; 1476 znode_t *zp = NULL; 1477 vnode_t *vp = NULL; 1478 char *zfs_bootfs; 1479 char *zfs_devid; 1480 1481 ASSERT(vfsp); 1482 1483 /* 1484 * The filesystem that we mount as root is defined in the 1485 * boot property "zfs-bootfs" with a format of 1486 * "poolname/root-dataset-objnum". 1487 */ 1488 if (why == ROOT_INIT) { 1489 if (zfsrootdone++) 1490 return (SET_ERROR(EBUSY)); 1491 /* 1492 * the process of doing a spa_load will require the 1493 * clock to be set before we could (for example) do 1494 * something better by looking at the timestamp on 1495 * an uberblock, so just set it to -1. 1496 */ 1497 clkset(-1); 1498 1499 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1500 cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1501 "bootfs name"); 1502 return (SET_ERROR(EINVAL)); 1503 } 1504 zfs_devid = spa_get_bootprop("diskdevid"); 1505 error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1506 if (zfs_devid) 1507 spa_free_bootprop(zfs_devid); 1508 if (error) { 1509 spa_free_bootprop(zfs_bootfs); 1510 cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1511 error); 1512 return (error); 1513 } 1514 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1515 spa_free_bootprop(zfs_bootfs); 1516 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1517 error); 1518 return (error); 1519 } 1520 1521 spa_free_bootprop(zfs_bootfs); 1522 1523 if (error = vfs_lock(vfsp)) 1524 return (error); 1525 1526 if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1527 cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1528 goto out; 1529 } 1530 1531 zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1532 ASSERT(zfsvfs); 1533 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1534 cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1535 goto out; 1536 } 1537 1538 vp = ZTOV(zp); 1539 mutex_enter(&vp->v_lock); 1540 vp->v_flag |= VROOT; 1541 mutex_exit(&vp->v_lock); 1542 rootvp = vp; 1543 1544 /* 1545 * Leave rootvp held. The root file system is never unmounted. 1546 */ 1547 1548 vfs_add((struct vnode *)0, vfsp, 1549 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1550out: 1551 vfs_unlock(vfsp); 1552 return (error); 1553 } else if (why == ROOT_REMOUNT) { 1554 readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1555 vfsp->vfs_flag |= VFS_REMOUNT; 1556 1557 /* refresh mount options */ 1558 zfs_unregister_callbacks(vfsp->vfs_data); 1559 return (zfs_register_callbacks(vfsp)); 1560 1561 } else if (why == ROOT_UNMOUNT) { 1562 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1563 (void) zfs_sync(vfsp, 0, 0); 1564 return (0); 1565 } 1566 1567 /* 1568 * if "why" is equal to anything else other than ROOT_INIT, 1569 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1570 */ 1571 return (SET_ERROR(ENOTSUP)); 1572} 1573#endif /* OPENSOLARIS_MOUNTROOT */ 1574 1575static int 1576getpoolname(const char *osname, char *poolname) 1577{ 1578 char *p; 1579 1580 p = strchr(osname, '/'); 1581 if (p == NULL) { 1582 if (strlen(osname) >= MAXNAMELEN) 1583 return (ENAMETOOLONG); 1584 (void) strcpy(poolname, osname); 1585 } else { 1586 if (p - osname >= MAXNAMELEN) 1587 return (ENAMETOOLONG); 1588 (void) strncpy(poolname, osname, p - osname); 1589 poolname[p - osname] = '\0'; 1590 } 1591 return (0); 1592} 1593 1594/*ARGSUSED*/ 1595static int 1596zfs_mount(vfs_t *vfsp) 1597{ 1598 kthread_t *td = curthread; 1599 vnode_t *mvp = vfsp->mnt_vnodecovered; 1600 cred_t *cr = td->td_ucred; 1601 char *osname; 1602 int error = 0; 1603 int canwrite; 1604 1605#ifdef illumos 1606 if (mvp->v_type != VDIR) 1607 return (SET_ERROR(ENOTDIR)); 1608 1609 mutex_enter(&mvp->v_lock); 1610 if ((uap->flags & MS_REMOUNT) == 0 && 1611 (uap->flags & MS_OVERLAY) == 0 && 1612 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 1613 mutex_exit(&mvp->v_lock); 1614 return (SET_ERROR(EBUSY)); 1615 } 1616 mutex_exit(&mvp->v_lock); 1617 1618 /* 1619 * ZFS does not support passing unparsed data in via MS_DATA. 1620 * Users should use the MS_OPTIONSTR interface; this means 1621 * that all option parsing is already done and the options struct 1622 * can be interrogated. 1623 */ 1624 if ((uap->flags & MS_DATA) && uap->datalen > 0) 1625#else 1626 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1627 return (SET_ERROR(EPERM)); 1628 1629 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1630 return (SET_ERROR(EINVAL)); 1631#endif /* ! illumos */ 1632 1633 /* 1634 * If full-owner-access is enabled and delegated administration is 1635 * turned on, we must set nosuid. 1636 */ 1637 if (zfs_super_owner && 1638 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1639 secpolicy_fs_mount_clearopts(cr, vfsp); 1640 } 1641 1642 /* 1643 * Check for mount privilege? 1644 * 1645 * If we don't have privilege then see if 1646 * we have local permission to allow it 1647 */ 1648 error = secpolicy_fs_mount(cr, mvp, vfsp); 1649 if (error) { 1650 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1651 goto out; 1652 1653 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1654 vattr_t vattr; 1655 1656 /* 1657 * Make sure user is the owner of the mount point 1658 * or has sufficient privileges. 1659 */ 1660 1661 vattr.va_mask = AT_UID; 1662 1663 vn_lock(mvp, LK_SHARED | LK_RETRY); 1664 if (VOP_GETATTR(mvp, &vattr, cr)) { 1665 VOP_UNLOCK(mvp, 0); 1666 goto out; 1667 } 1668 1669 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1670 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1671 VOP_UNLOCK(mvp, 0); 1672 goto out; 1673 } 1674 VOP_UNLOCK(mvp, 0); 1675 } 1676 1677 secpolicy_fs_mount_clearopts(cr, vfsp); 1678 } 1679 1680 /* 1681 * Refuse to mount a filesystem if we are in a local zone and the 1682 * dataset is not visible. 1683 */ 1684 if (!INGLOBALZONE(curthread) && 1685 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1686 error = SET_ERROR(EPERM); 1687 goto out; 1688 } 1689 1690#ifdef SECLABEL 1691 error = zfs_mount_label_policy(vfsp, osname); 1692 if (error) 1693 goto out; 1694#endif 1695 1696 vfsp->vfs_flag |= MNT_NFS4ACLS; 1697 1698 /* 1699 * When doing a remount, we simply refresh our temporary properties 1700 * according to those options set in the current VFS options. 1701 */ 1702 if (vfsp->vfs_flag & MS_REMOUNT) { 1703 /* refresh mount options */ 1704 zfs_unregister_callbacks(vfsp->vfs_data); 1705 error = zfs_register_callbacks(vfsp); 1706 goto out; 1707 } 1708 1709 /* Initial root mount: try hard to import the requested root pool. */ 1710 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1711 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1712 char pname[MAXNAMELEN]; 1713 1714 error = getpoolname(osname, pname); 1715 if (error == 0) 1716 error = spa_import_rootpool(pname); 1717 if (error) 1718 goto out; 1719 } 1720 DROP_GIANT(); 1721 error = zfs_domount(vfsp, osname); 1722 PICKUP_GIANT(); 1723 1724#ifdef sun 1725 /* 1726 * Add an extra VFS_HOLD on our parent vfs so that it can't 1727 * disappear due to a forced unmount. 1728 */ 1729 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1730 VFS_HOLD(mvp->v_vfsp); 1731#endif /* sun */ 1732 1733out: 1734 return (error); 1735} 1736 1737static int 1738zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1739{ 1740 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1741 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1742 1743 statp->f_version = STATFS_VERSION; 1744 1745 ZFS_ENTER(zfsvfs); 1746 1747 dmu_objset_space(zfsvfs->z_os, 1748 &refdbytes, &availbytes, &usedobjs, &availobjs); 1749 1750 /* 1751 * The underlying storage pool actually uses multiple block sizes. 1752 * We report the fragsize as the smallest block size we support, 1753 * and we report our blocksize as the filesystem's maximum blocksize. 1754 */ 1755 statp->f_bsize = SPA_MINBLOCKSIZE; 1756 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1757 1758 /* 1759 * The following report "total" blocks of various kinds in the 1760 * file system, but reported in terms of f_frsize - the 1761 * "fragment" size. 1762 */ 1763 1764 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1765 statp->f_bfree = availbytes / statp->f_bsize; 1766 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1767 1768 /* 1769 * statvfs() should really be called statufs(), because it assumes 1770 * static metadata. ZFS doesn't preallocate files, so the best 1771 * we can do is report the max that could possibly fit in f_files, 1772 * and that minus the number actually used in f_ffree. 1773 * For f_ffree, report the smaller of the number of object available 1774 * and the number of blocks (each object will take at least a block). 1775 */ 1776 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1777 statp->f_files = statp->f_ffree + usedobjs; 1778 1779 /* 1780 * We're a zfs filesystem. 1781 */ 1782 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1783 1784 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1785 sizeof(statp->f_mntfromname)); 1786 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1787 sizeof(statp->f_mntonname)); 1788 1789 statp->f_namemax = ZFS_MAXNAMELEN; 1790 1791 ZFS_EXIT(zfsvfs); 1792 return (0); 1793} 1794 1795static int 1796zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1797{ 1798 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1799 znode_t *rootzp; 1800 int error; 1801 1802 ZFS_ENTER(zfsvfs); 1803 1804 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1805 if (error == 0) 1806 *vpp = ZTOV(rootzp); 1807 1808 ZFS_EXIT(zfsvfs); 1809 1810 if (error == 0) { 1811 error = vn_lock(*vpp, flags); 1812 if (error == 0) 1813 (*vpp)->v_vflag |= VV_ROOT; 1814 } 1815 if (error != 0) 1816 *vpp = NULL; 1817 1818 return (error); 1819} 1820 1821/* 1822 * Teardown the zfsvfs::z_os. 1823 * 1824 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1825 * and 'z_teardown_inactive_lock' held. 1826 */ 1827static int 1828zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1829{ 1830 znode_t *zp; 1831 1832 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1833 1834 if (!unmounting) { 1835 /* 1836 * We purge the parent filesystem's vfsp as the parent 1837 * filesystem and all of its snapshots have their vnode's 1838 * v_vfsp set to the parent's filesystem's vfsp. Note, 1839 * 'z_parent' is self referential for non-snapshots. 1840 */ 1841 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1842#ifdef FREEBSD_NAMECACHE 1843 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1844#endif 1845 } 1846 1847 /* 1848 * Close the zil. NB: Can't close the zil while zfs_inactive 1849 * threads are blocked as zil_close can call zfs_inactive. 1850 */ 1851 if (zfsvfs->z_log) { 1852 zil_close(zfsvfs->z_log); 1853 zfsvfs->z_log = NULL; 1854 } 1855 1856 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1857 1858 /* 1859 * If we are not unmounting (ie: online recv) and someone already 1860 * unmounted this file system while we were doing the switcheroo, 1861 * or a reopen of z_os failed then just bail out now. 1862 */ 1863 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1864 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1865 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1866 return (SET_ERROR(EIO)); 1867 } 1868 1869 /* 1870 * At this point there are no vops active, and any new vops will 1871 * fail with EIO since we have z_teardown_lock for writer (only 1872 * relavent for forced unmount). 1873 * 1874 * Release all holds on dbufs. 1875 */ 1876 mutex_enter(&zfsvfs->z_znodes_lock); 1877 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1878 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1879 if (zp->z_sa_hdl) { 1880 ASSERT(ZTOV(zp)->v_count >= 0); 1881 zfs_znode_dmu_fini(zp); 1882 } 1883 mutex_exit(&zfsvfs->z_znodes_lock); 1884 1885 /* 1886 * If we are unmounting, set the unmounted flag and let new vops 1887 * unblock. zfs_inactive will have the unmounted behavior, and all 1888 * other vops will fail with EIO. 1889 */ 1890 if (unmounting) { 1891 zfsvfs->z_unmounted = B_TRUE; 1892 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1893 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1894 } 1895 1896 /* 1897 * z_os will be NULL if there was an error in attempting to reopen 1898 * zfsvfs, so just return as the properties had already been 1899 * unregistered and cached data had been evicted before. 1900 */ 1901 if (zfsvfs->z_os == NULL) 1902 return (0); 1903 1904 /* 1905 * Unregister properties. 1906 */ 1907 zfs_unregister_callbacks(zfsvfs); 1908 1909 /* 1910 * Evict cached data 1911 */ 1912 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1913 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1914 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1915 dmu_objset_evict_dbufs(zfsvfs->z_os); 1916 1917 return (0); 1918} 1919 1920/*ARGSUSED*/ 1921static int 1922zfs_umount(vfs_t *vfsp, int fflag) 1923{ 1924 kthread_t *td = curthread; 1925 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1926 objset_t *os; 1927 cred_t *cr = td->td_ucred; 1928 int ret; 1929 1930 ret = secpolicy_fs_unmount(cr, vfsp); 1931 if (ret) { 1932 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1933 ZFS_DELEG_PERM_MOUNT, cr)) 1934 return (ret); 1935 } 1936 1937 /* 1938 * We purge the parent filesystem's vfsp as the parent filesystem 1939 * and all of its snapshots have their vnode's v_vfsp set to the 1940 * parent's filesystem's vfsp. Note, 'z_parent' is self 1941 * referential for non-snapshots. 1942 */ 1943 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1944 1945 /* 1946 * Unmount any snapshots mounted under .zfs before unmounting the 1947 * dataset itself. 1948 */ 1949 if (zfsvfs->z_ctldir != NULL) { 1950 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1951 return (ret); 1952 ret = vflush(vfsp, 0, 0, td); 1953 ASSERT(ret == EBUSY); 1954 if (!(fflag & MS_FORCE)) { 1955 if (zfsvfs->z_ctldir->v_count > 1) 1956 return (EBUSY); 1957 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1958 } 1959 zfsctl_destroy(zfsvfs); 1960 ASSERT(zfsvfs->z_ctldir == NULL); 1961 } 1962 1963 if (fflag & MS_FORCE) { 1964 /* 1965 * Mark file system as unmounted before calling 1966 * vflush(FORCECLOSE). This way we ensure no future vnops 1967 * will be called and risk operating on DOOMED vnodes. 1968 */ 1969 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1970 zfsvfs->z_unmounted = B_TRUE; 1971 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 1972 } 1973 1974 /* 1975 * Flush all the files. 1976 */ 1977 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1978 if (ret != 0) { 1979 if (!zfsvfs->z_issnap) { 1980 zfsctl_create(zfsvfs); 1981 ASSERT(zfsvfs->z_ctldir != NULL); 1982 } 1983 return (ret); 1984 } 1985 1986#ifdef sun 1987 if (!(fflag & MS_FORCE)) { 1988 /* 1989 * Check the number of active vnodes in the file system. 1990 * Our count is maintained in the vfs structure, but the 1991 * number is off by 1 to indicate a hold on the vfs 1992 * structure itself. 1993 * 1994 * The '.zfs' directory maintains a reference of its 1995 * own, and any active references underneath are 1996 * reflected in the vnode count. 1997 */ 1998 if (zfsvfs->z_ctldir == NULL) { 1999 if (vfsp->vfs_count > 1) 2000 return (SET_ERROR(EBUSY)); 2001 } else { 2002 if (vfsp->vfs_count > 2 || 2003 zfsvfs->z_ctldir->v_count > 1) 2004 return (SET_ERROR(EBUSY)); 2005 } 2006 } 2007#endif 2008 2009 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 2010 os = zfsvfs->z_os; 2011 2012 /* 2013 * z_os will be NULL if there was an error in 2014 * attempting to reopen zfsvfs. 2015 */ 2016 if (os != NULL) { 2017 /* 2018 * Unset the objset user_ptr. 2019 */ 2020 mutex_enter(&os->os_user_ptr_lock); 2021 dmu_objset_set_user(os, NULL); 2022 mutex_exit(&os->os_user_ptr_lock); 2023 2024 /* 2025 * Finally release the objset 2026 */ 2027 dmu_objset_disown(os, zfsvfs); 2028 } 2029 2030 /* 2031 * We can now safely destroy the '.zfs' directory node. 2032 */ 2033 if (zfsvfs->z_ctldir != NULL) 2034 zfsctl_destroy(zfsvfs); 2035 if (zfsvfs->z_issnap) { 2036 vnode_t *svp = vfsp->mnt_vnodecovered; 2037 2038 if (svp->v_count >= 2) 2039 VN_RELE(svp); 2040 } 2041 zfs_freevfs(vfsp); 2042 2043 return (0); 2044} 2045 2046static int 2047zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2048{ 2049 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2050 znode_t *zp; 2051 int err; 2052 2053 /* 2054 * zfs_zget() can't operate on virtual entries like .zfs/ or 2055 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2056 * This will make NFS to switch to LOOKUP instead of using VGET. 2057 */ 2058 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || 2059 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) 2060 return (EOPNOTSUPP); 2061 2062 ZFS_ENTER(zfsvfs); 2063 err = zfs_zget(zfsvfs, ino, &zp); 2064 if (err == 0 && zp->z_unlinked) { 2065 VN_RELE(ZTOV(zp)); 2066 err = EINVAL; 2067 } 2068 if (err == 0) 2069 *vpp = ZTOV(zp); 2070 ZFS_EXIT(zfsvfs); 2071 if (err == 0) 2072 err = vn_lock(*vpp, flags); 2073 if (err != 0) 2074 *vpp = NULL; 2075 return (err); 2076} 2077 2078static int 2079zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2080 struct ucred **credanonp, int *numsecflavors, int **secflavors) 2081{ 2082 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2083 2084 /* 2085 * If this is regular file system vfsp is the same as 2086 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2087 * zfsvfs->z_parent->z_vfs represents parent file system 2088 * which we have to use here, because only this file system 2089 * has mnt_export configured. 2090 */ 2091 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2092 credanonp, numsecflavors, secflavors)); 2093} 2094 2095CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2096CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2097 2098static int 2099zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2100{ 2101 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2102 znode_t *zp; 2103 uint64_t object = 0; 2104 uint64_t fid_gen = 0; 2105 uint64_t gen_mask; 2106 uint64_t zp_gen; 2107 int i, err; 2108 2109 *vpp = NULL; 2110 2111 ZFS_ENTER(zfsvfs); 2112 2113 /* 2114 * On FreeBSD we can get snapshot's mount point or its parent file 2115 * system mount point depending if snapshot is already mounted or not. 2116 */ 2117 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2118 zfid_long_t *zlfid = (zfid_long_t *)fidp; 2119 uint64_t objsetid = 0; 2120 uint64_t setgen = 0; 2121 2122 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2123 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2124 2125 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2126 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2127 2128 ZFS_EXIT(zfsvfs); 2129 2130 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2131 if (err) 2132 return (SET_ERROR(EINVAL)); 2133 ZFS_ENTER(zfsvfs); 2134 } 2135 2136 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2137 zfid_short_t *zfid = (zfid_short_t *)fidp; 2138 2139 for (i = 0; i < sizeof (zfid->zf_object); i++) 2140 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2141 2142 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2143 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2144 } else { 2145 ZFS_EXIT(zfsvfs); 2146 return (SET_ERROR(EINVAL)); 2147 } 2148 2149 /* 2150 * A zero fid_gen means we are in .zfs or the .zfs/snapshot 2151 * directory tree. If the object == zfsvfs->z_shares_dir, then 2152 * we are in the .zfs/shares directory tree. 2153 */ 2154 if ((fid_gen == 0 && 2155 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || 2156 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { 2157 *vpp = zfsvfs->z_ctldir; 2158 ASSERT(*vpp != NULL); 2159 if (object == ZFSCTL_INO_SNAPDIR) { 2160 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2161 0, NULL, NULL, NULL, NULL, NULL) == 0); 2162 } else if (object == zfsvfs->z_shares_dir) { 2163 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL, 2164 0, NULL, NULL, NULL, NULL, NULL) == 0); 2165 } else { 2166 VN_HOLD(*vpp); 2167 } 2168 ZFS_EXIT(zfsvfs); 2169 err = vn_lock(*vpp, flags); 2170 if (err != 0) 2171 *vpp = NULL; 2172 return (err); 2173 } 2174 2175 gen_mask = -1ULL >> (64 - 8 * i); 2176 2177 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2178 if (err = zfs_zget(zfsvfs, object, &zp)) { 2179 ZFS_EXIT(zfsvfs); 2180 return (err); 2181 } 2182 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2183 sizeof (uint64_t)); 2184 zp_gen = zp_gen & gen_mask; 2185 if (zp_gen == 0) 2186 zp_gen = 1; 2187 if (zp->z_unlinked || zp_gen != fid_gen) { 2188 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2189 VN_RELE(ZTOV(zp)); 2190 ZFS_EXIT(zfsvfs); 2191 return (SET_ERROR(EINVAL)); 2192 } 2193 2194 *vpp = ZTOV(zp); 2195 ZFS_EXIT(zfsvfs); 2196 err = vn_lock(*vpp, flags | LK_RETRY); 2197 if (err == 0) 2198 vnode_create_vobject(*vpp, zp->z_size, curthread); 2199 else 2200 *vpp = NULL; 2201 return (err); 2202} 2203 2204/* 2205 * Block out VOPs and close zfsvfs_t::z_os 2206 * 2207 * Note, if successful, then we return with the 'z_teardown_lock' and 2208 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 2209 * dataset and objset intact so that they can be atomically handed off during 2210 * a subsequent rollback or recv operation and the resume thereafter. 2211 */ 2212int 2213zfs_suspend_fs(zfsvfs_t *zfsvfs) 2214{ 2215 int error; 2216 2217 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2218 return (error); 2219 2220 return (0); 2221} 2222 2223/* 2224 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 2225 * is an invariant across any of the operations that can be performed while the 2226 * filesystem was suspended. Whether it succeeded or failed, the preconditions 2227 * are the same: the relevant objset and associated dataset are owned by 2228 * zfsvfs, held, and long held on entry. 2229 */ 2230int 2231zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2232{ 2233 int err; 2234 znode_t *zp; 2235 uint64_t sa_obj = 0; 2236 2237 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2238 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2239 2240 /* 2241 * We already own this, so just hold and rele it to update the 2242 * objset_t, as the one we had before may have been evicted. 2243 */ 2244 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os)); 2245 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs); 2246 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset)); 2247 dmu_objset_rele(zfsvfs->z_os, zfsvfs); 2248 2249 /* 2250 * Make sure version hasn't changed 2251 */ 2252 2253 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, 2254 &zfsvfs->z_version); 2255 2256 if (err) 2257 goto bail; 2258 2259 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, 2260 ZFS_SA_ATTRS, 8, 1, &sa_obj); 2261 2262 if (err && zfsvfs->z_version >= ZPL_VERSION_SA) 2263 goto bail; 2264 2265 if ((err = sa_setup(zfsvfs->z_os, sa_obj, 2266 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) 2267 goto bail; 2268 2269 if (zfsvfs->z_version >= ZPL_VERSION_SA) 2270 sa_register_update_callback(zfsvfs->z_os, 2271 zfs_sa_upgrade); 2272 2273 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2274 2275 zfs_set_fuid_feature(zfsvfs); 2276 2277 /* 2278 * Attempt to re-establish all the active znodes with 2279 * their dbufs. If a zfs_rezget() fails, then we'll let 2280 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2281 * when they try to use their znode. 2282 */ 2283 mutex_enter(&zfsvfs->z_znodes_lock); 2284 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2285 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2286 (void) zfs_rezget(zp); 2287 } 2288 mutex_exit(&zfsvfs->z_znodes_lock); 2289 2290bail: 2291 /* release the VOPs */ 2292 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2293 rrm_exit(&zfsvfs->z_teardown_lock, FTAG); 2294 2295 if (err) { 2296 /* 2297 * Since we couldn't setup the sa framework, try to force 2298 * unmount this file system. 2299 */ 2300 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) { 2301 vfs_ref(zfsvfs->z_vfs); 2302 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2303 } 2304 } 2305 return (err); 2306} 2307 2308static void 2309zfs_freevfs(vfs_t *vfsp) 2310{ 2311 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2312 2313#ifdef sun 2314 /* 2315 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2316 * from zfs_mount(). Release it here. If we came through 2317 * zfs_mountroot() instead, we didn't grab an extra hold, so 2318 * skip the VFS_RELE for rootvfs. 2319 */ 2320 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2321 VFS_RELE(zfsvfs->z_parent->z_vfs); 2322#endif /* sun */ 2323 2324 zfsvfs_free(zfsvfs); 2325 2326 atomic_dec_32(&zfs_active_fs_count); 2327} 2328 2329#ifdef __i386__ 2330static int desiredvnodes_backup; 2331#endif 2332 2333static void 2334zfs_vnodes_adjust(void) 2335{ 2336#ifdef __i386__ 2337 int newdesiredvnodes; 2338 2339 desiredvnodes_backup = desiredvnodes; 2340 2341 /* 2342 * We calculate newdesiredvnodes the same way it is done in 2343 * vntblinit(). If it is equal to desiredvnodes, it means that 2344 * it wasn't tuned by the administrator and we can tune it down. 2345 */ 2346 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2347 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2348 sizeof(struct vnode)))); 2349 if (newdesiredvnodes == desiredvnodes) 2350 desiredvnodes = (3 * newdesiredvnodes) / 4; 2351#endif 2352} 2353 2354static void 2355zfs_vnodes_adjust_back(void) 2356{ 2357 2358#ifdef __i386__ 2359 desiredvnodes = desiredvnodes_backup; 2360#endif 2361} 2362 2363void 2364zfs_init(void) 2365{ 2366 2367 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2368 2369 /* 2370 * Initialize .zfs directory structures 2371 */ 2372 zfsctl_init(); 2373 2374 /* 2375 * Initialize znode cache, vnode ops, etc... 2376 */ 2377 zfs_znode_init(); 2378 2379 /* 2380 * Reduce number of vnodes. Originally number of vnodes is calculated 2381 * with UFS inode in mind. We reduce it here, because it's too big for 2382 * ZFS/i386. 2383 */ 2384 zfs_vnodes_adjust(); 2385 2386 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2387} 2388 2389void 2390zfs_fini(void) 2391{ 2392 zfsctl_fini(); 2393 zfs_znode_fini(); 2394 zfs_vnodes_adjust_back(); 2395} 2396 2397int 2398zfs_busy(void) 2399{ 2400 return (zfs_active_fs_count != 0); 2401} 2402 2403int 2404zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2405{ 2406 int error; 2407 objset_t *os = zfsvfs->z_os; 2408 dmu_tx_t *tx; 2409 2410 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2411 return (SET_ERROR(EINVAL)); 2412 2413 if (newvers < zfsvfs->z_version) 2414 return (SET_ERROR(EINVAL)); 2415 2416 if (zfs_spa_version_map(newvers) > 2417 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2418 return (SET_ERROR(ENOTSUP)); 2419 2420 tx = dmu_tx_create(os); 2421 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2422 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2423 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2424 ZFS_SA_ATTRS); 2425 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2426 } 2427 error = dmu_tx_assign(tx, TXG_WAIT); 2428 if (error) { 2429 dmu_tx_abort(tx); 2430 return (error); 2431 } 2432 2433 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2434 8, 1, &newvers, tx); 2435 2436 if (error) { 2437 dmu_tx_commit(tx); 2438 return (error); 2439 } 2440 2441 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2442 uint64_t sa_obj; 2443 2444 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2445 SPA_VERSION_SA); 2446 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2447 DMU_OT_NONE, 0, tx); 2448 2449 error = zap_add(os, MASTER_NODE_OBJ, 2450 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2451 ASSERT0(error); 2452 2453 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2454 sa_register_update_callback(os, zfs_sa_upgrade); 2455 } 2456 2457 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2458 "from %llu to %llu", zfsvfs->z_version, newvers); 2459 2460 dmu_tx_commit(tx); 2461 2462 zfsvfs->z_version = newvers; 2463 2464 zfs_set_fuid_feature(zfsvfs); 2465 2466 return (0); 2467} 2468 2469/* 2470 * Read a property stored within the master node. 2471 */ 2472int 2473zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2474{ 2475 const char *pname; 2476 int error = ENOENT; 2477 2478 /* 2479 * Look up the file system's value for the property. For the 2480 * version property, we look up a slightly different string. 2481 */ 2482 if (prop == ZFS_PROP_VERSION) 2483 pname = ZPL_VERSION_STR; 2484 else 2485 pname = zfs_prop_to_name(prop); 2486 2487 if (os != NULL) 2488 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2489 2490 if (error == ENOENT) { 2491 /* No value set, use the default value */ 2492 switch (prop) { 2493 case ZFS_PROP_VERSION: 2494 *value = ZPL_VERSION; 2495 break; 2496 case ZFS_PROP_NORMALIZE: 2497 case ZFS_PROP_UTF8ONLY: 2498 *value = 0; 2499 break; 2500 case ZFS_PROP_CASE: 2501 *value = ZFS_CASE_SENSITIVE; 2502 break; 2503 default: 2504 return (error); 2505 } 2506 error = 0; 2507 } 2508 return (error); 2509} 2510 2511#ifdef _KERNEL 2512void 2513zfsvfs_update_fromname(const char *oldname, const char *newname) 2514{ 2515 char tmpbuf[MAXPATHLEN]; 2516 struct mount *mp; 2517 char *fromname; 2518 size_t oldlen; 2519 2520 oldlen = strlen(oldname); 2521 2522 mtx_lock(&mountlist_mtx); 2523 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2524 fromname = mp->mnt_stat.f_mntfromname; 2525 if (strcmp(fromname, oldname) == 0) { 2526 (void)strlcpy(fromname, newname, 2527 sizeof(mp->mnt_stat.f_mntfromname)); 2528 continue; 2529 } 2530 if (strncmp(fromname, oldname, oldlen) == 0 && 2531 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2532 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2533 newname, fromname + oldlen); 2534 (void)strlcpy(fromname, tmpbuf, 2535 sizeof(mp->mnt_stat.f_mntfromname)); 2536 continue; 2537 } 2538 } 2539 mtx_unlock(&mountlist_mtx); 2540} 2541#endif 2542