zfs_vfsops.c revision 243486
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 */ 26 27/* Portions Copyright 2010 Robert Milkowski */ 28 29#include <sys/types.h> 30#include <sys/param.h> 31#include <sys/systm.h> 32#include <sys/kernel.h> 33#include <sys/sysmacros.h> 34#include <sys/kmem.h> 35#include <sys/acl.h> 36#include <sys/vnode.h> 37#include <sys/vfs.h> 38#include <sys/mntent.h> 39#include <sys/mount.h> 40#include <sys/cmn_err.h> 41#include <sys/zfs_znode.h> 42#include <sys/zfs_dir.h> 43#include <sys/zil.h> 44#include <sys/fs/zfs.h> 45#include <sys/dmu.h> 46#include <sys/dsl_prop.h> 47#include <sys/dsl_dataset.h> 48#include <sys/dsl_deleg.h> 49#include <sys/spa.h> 50#include <sys/zap.h> 51#include <sys/sa.h> 52#include <sys/varargs.h> 53#include <sys/policy.h> 54#include <sys/atomic.h> 55#include <sys/zfs_ioctl.h> 56#include <sys/zfs_ctldir.h> 57#include <sys/zfs_fuid.h> 58#include <sys/sunddi.h> 59#include <sys/dnlc.h> 60#include <sys/dmu_objset.h> 61#include <sys/spa_boot.h> 62#include <sys/sa.h> 63#include <sys/jail.h> 64#include "zfs_comutil.h" 65 66struct mtx zfs_debug_mtx; 67MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF); 68 69SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system"); 70 71int zfs_super_owner; 72SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0, 73 "File system owner can perform privileged operation on his file systems"); 74 75int zfs_debug_level; 76TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level); 77SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0, 78 "Debug level"); 79 80SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions"); 81static int zfs_version_acl = ZFS_ACL_VERSION; 82SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0, 83 "ZFS_ACL_VERSION"); 84static int zfs_version_spa = SPA_VERSION; 85SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0, 86 "SPA_VERSION"); 87static int zfs_version_zpl = ZPL_VERSION; 88SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0, 89 "ZPL_VERSION"); 90 91static int zfs_mount(vfs_t *vfsp); 92static int zfs_umount(vfs_t *vfsp, int fflag); 93static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp); 94static int zfs_statfs(vfs_t *vfsp, struct statfs *statp); 95static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp); 96static int zfs_sync(vfs_t *vfsp, int waitfor); 97static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 98 struct ucred **credanonp, int *numsecflavors, int **secflavors); 99static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp); 100static void zfs_objset_close(zfsvfs_t *zfsvfs); 101static void zfs_freevfs(vfs_t *vfsp); 102 103static struct vfsops zfs_vfsops = { 104 .vfs_mount = zfs_mount, 105 .vfs_unmount = zfs_umount, 106 .vfs_root = zfs_root, 107 .vfs_statfs = zfs_statfs, 108 .vfs_vget = zfs_vget, 109 .vfs_sync = zfs_sync, 110 .vfs_checkexp = zfs_checkexp, 111 .vfs_fhtovp = zfs_fhtovp, 112}; 113 114VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN); 115 116/* 117 * We need to keep a count of active fs's. 118 * This is necessary to prevent our module 119 * from being unloaded after a umount -f 120 */ 121static uint32_t zfs_active_fs_count = 0; 122 123/*ARGSUSED*/ 124static int 125zfs_sync(vfs_t *vfsp, int waitfor) 126{ 127 128 /* 129 * Data integrity is job one. We don't want a compromised kernel 130 * writing to the storage pool, so we never sync during panic. 131 */ 132 if (panicstr) 133 return (0); 134 135 if (vfsp != NULL) { 136 /* 137 * Sync a specific filesystem. 138 */ 139 zfsvfs_t *zfsvfs = vfsp->vfs_data; 140 dsl_pool_t *dp; 141 int error; 142 143 error = vfs_stdsync(vfsp, waitfor); 144 if (error != 0) 145 return (error); 146 147 ZFS_ENTER(zfsvfs); 148 dp = dmu_objset_pool(zfsvfs->z_os); 149 150 /* 151 * If the system is shutting down, then skip any 152 * filesystems which may exist on a suspended pool. 153 */ 154 if (sys_shutdown && spa_suspended(dp->dp_spa)) { 155 ZFS_EXIT(zfsvfs); 156 return (0); 157 } 158 159 if (zfsvfs->z_log != NULL) 160 zil_commit(zfsvfs->z_log, 0); 161 162 ZFS_EXIT(zfsvfs); 163 } else { 164 /* 165 * Sync all ZFS filesystems. This is what happens when you 166 * run sync(1M). Unlike other filesystems, ZFS honors the 167 * request by waiting for all pools to commit all dirty data. 168 */ 169 spa_sync_allpools(); 170 } 171 172 return (0); 173} 174 175#ifndef __FreeBSD__ 176static int 177zfs_create_unique_device(dev_t *dev) 178{ 179 major_t new_major; 180 181 do { 182 ASSERT3U(zfs_minor, <=, MAXMIN32); 183 minor_t start = zfs_minor; 184 do { 185 mutex_enter(&zfs_dev_mtx); 186 if (zfs_minor >= MAXMIN32) { 187 /* 188 * If we're still using the real major 189 * keep out of /dev/zfs and /dev/zvol minor 190 * number space. If we're using a getudev()'ed 191 * major number, we can use all of its minors. 192 */ 193 if (zfs_major == ddi_name_to_major(ZFS_DRIVER)) 194 zfs_minor = ZFS_MIN_MINOR; 195 else 196 zfs_minor = 0; 197 } else { 198 zfs_minor++; 199 } 200 *dev = makedevice(zfs_major, zfs_minor); 201 mutex_exit(&zfs_dev_mtx); 202 } while (vfs_devismounted(*dev) && zfs_minor != start); 203 if (zfs_minor == start) { 204 /* 205 * We are using all ~262,000 minor numbers for the 206 * current major number. Create a new major number. 207 */ 208 if ((new_major = getudev()) == (major_t)-1) { 209 cmn_err(CE_WARN, 210 "zfs_mount: Can't get unique major " 211 "device number."); 212 return (-1); 213 } 214 mutex_enter(&zfs_dev_mtx); 215 zfs_major = new_major; 216 zfs_minor = 0; 217 218 mutex_exit(&zfs_dev_mtx); 219 } else { 220 break; 221 } 222 /* CONSTANTCONDITION */ 223 } while (1); 224 225 return (0); 226} 227#endif /* !__FreeBSD__ */ 228 229static void 230atime_changed_cb(void *arg, uint64_t newval) 231{ 232 zfsvfs_t *zfsvfs = arg; 233 234 if (newval == TRUE) { 235 zfsvfs->z_atime = TRUE; 236 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; 237 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); 238 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); 239 } else { 240 zfsvfs->z_atime = FALSE; 241 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; 242 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); 243 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); 244 } 245} 246 247static void 248xattr_changed_cb(void *arg, uint64_t newval) 249{ 250 zfsvfs_t *zfsvfs = arg; 251 252 if (newval == TRUE) { 253 /* XXX locking on vfs_flag? */ 254#ifdef TODO 255 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; 256#endif 257 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); 258 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); 259 } else { 260 /* XXX locking on vfs_flag? */ 261#ifdef TODO 262 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR; 263#endif 264 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR); 265 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); 266 } 267} 268 269static void 270blksz_changed_cb(void *arg, uint64_t newval) 271{ 272 zfsvfs_t *zfsvfs = arg; 273 274 if (newval < SPA_MINBLOCKSIZE || 275 newval > SPA_MAXBLOCKSIZE || !ISP2(newval)) 276 newval = SPA_MAXBLOCKSIZE; 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 int readonly, do_readonly = B_FALSE; 393 int setuid, do_setuid = B_FALSE; 394 int exec, do_exec = B_FALSE; 395 int xattr, do_xattr = B_FALSE; 396 int atime, do_atime = B_FALSE; 397 int error = 0; 398 399 ASSERT(vfsp); 400 zfsvfs = vfsp->vfs_data; 401 ASSERT(zfsvfs); 402 os = zfsvfs->z_os; 403 404 /* 405 * This function can be called for a snapshot when we update snapshot's 406 * mount point, which isn't really supported. 407 */ 408 if (dmu_objset_is_snapshot(os)) 409 return (EOPNOTSUPP); 410 411 /* 412 * The act of registering our callbacks will destroy any mount 413 * options we may have. In order to enable temporary overrides 414 * of mount options, we stash away the current values and 415 * restore them after we register the callbacks. 416 */ 417 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) || 418 !spa_writeable(dmu_objset_spa(os))) { 419 readonly = B_TRUE; 420 do_readonly = B_TRUE; 421 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 422 readonly = B_FALSE; 423 do_readonly = B_TRUE; 424 } 425 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 426 setuid = B_FALSE; 427 do_setuid = B_TRUE; 428 } else { 429 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 430 setuid = B_FALSE; 431 do_setuid = B_TRUE; 432 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 433 setuid = B_TRUE; 434 do_setuid = B_TRUE; 435 } 436 } 437 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 438 exec = B_FALSE; 439 do_exec = B_TRUE; 440 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 441 exec = B_TRUE; 442 do_exec = B_TRUE; 443 } 444 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 445 xattr = B_FALSE; 446 do_xattr = B_TRUE; 447 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 448 xattr = B_TRUE; 449 do_xattr = B_TRUE; 450 } 451 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 452 atime = B_FALSE; 453 do_atime = B_TRUE; 454 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 455 atime = B_TRUE; 456 do_atime = B_TRUE; 457 } 458 459 /* 460 * nbmand is a special property. It can only be changed at 461 * mount time. 462 * 463 * This is weird, but it is documented to only be changeable 464 * at mount time. 465 */ 466 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 467 nbmand = B_FALSE; 468 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 469 nbmand = B_TRUE; 470 } else { 471 char osname[MAXNAMELEN]; 472 473 dmu_objset_name(os, osname); 474 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand, 475 NULL)) { 476 return (error); 477 } 478 } 479 480 /* 481 * Register property callbacks. 482 * 483 * It would probably be fine to just check for i/o error from 484 * the first prop_register(), but I guess I like to go 485 * overboard... 486 */ 487 ds = dmu_objset_ds(os); 488 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs); 489 error = error ? error : dsl_prop_register(ds, 490 "xattr", xattr_changed_cb, zfsvfs); 491 error = error ? error : dsl_prop_register(ds, 492 "recordsize", blksz_changed_cb, zfsvfs); 493 error = error ? error : dsl_prop_register(ds, 494 "readonly", readonly_changed_cb, zfsvfs); 495 error = error ? error : dsl_prop_register(ds, 496 "setuid", setuid_changed_cb, zfsvfs); 497 error = error ? error : dsl_prop_register(ds, 498 "exec", exec_changed_cb, zfsvfs); 499 error = error ? error : dsl_prop_register(ds, 500 "snapdir", snapdir_changed_cb, zfsvfs); 501 error = error ? error : dsl_prop_register(ds, 502 "aclmode", acl_mode_changed_cb, zfsvfs); 503 error = error ? error : dsl_prop_register(ds, 504 "aclinherit", acl_inherit_changed_cb, zfsvfs); 505 error = error ? error : dsl_prop_register(ds, 506 "vscan", vscan_changed_cb, zfsvfs); 507 if (error) 508 goto unregister; 509 510 /* 511 * Invoke our callbacks to restore temporary mount options. 512 */ 513 if (do_readonly) 514 readonly_changed_cb(zfsvfs, readonly); 515 if (do_setuid) 516 setuid_changed_cb(zfsvfs, setuid); 517 if (do_exec) 518 exec_changed_cb(zfsvfs, exec); 519 if (do_xattr) 520 xattr_changed_cb(zfsvfs, xattr); 521 if (do_atime) 522 atime_changed_cb(zfsvfs, atime); 523 524 nbmand_changed_cb(zfsvfs, nbmand); 525 526 return (0); 527 528unregister: 529 /* 530 * We may attempt to unregister some callbacks that are not 531 * registered, but this is OK; it will simply return ENOMSG, 532 * which we will ignore. 533 */ 534 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs); 535 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs); 536 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs); 537 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs); 538 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs); 539 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs); 540 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs); 541 (void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs); 542 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb, 543 zfsvfs); 544 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs); 545 return (error); 546 547} 548 549static int 550zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, 551 uint64_t *userp, uint64_t *groupp) 552{ 553 znode_phys_t *znp = data; 554 int error = 0; 555 556 /* 557 * Is it a valid type of object to track? 558 */ 559 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 560 return (ENOENT); 561 562 /* 563 * If we have a NULL data pointer 564 * then assume the id's aren't changing and 565 * return EEXIST to the dmu to let it know to 566 * use the same ids 567 */ 568 if (data == NULL) 569 return (EEXIST); 570 571 if (bonustype == DMU_OT_ZNODE) { 572 *userp = znp->zp_uid; 573 *groupp = znp->zp_gid; 574 } else { 575 int hdrsize; 576 577 ASSERT(bonustype == DMU_OT_SA); 578 hdrsize = sa_hdrsize(data); 579 580 if (hdrsize != 0) { 581 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + 582 SA_UID_OFFSET)); 583 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + 584 SA_GID_OFFSET)); 585 } else { 586 /* 587 * This should only happen for newly created 588 * files that haven't had the znode data filled 589 * in yet. 590 */ 591 *userp = 0; 592 *groupp = 0; 593 } 594 } 595 return (error); 596} 597 598static void 599fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, 600 char *domainbuf, int buflen, uid_t *ridp) 601{ 602 uint64_t fuid; 603 const char *domain; 604 605 fuid = strtonum(fuidstr, NULL); 606 607 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); 608 if (domain) 609 (void) strlcpy(domainbuf, domain, buflen); 610 else 611 domainbuf[0] = '\0'; 612 *ridp = FUID_RID(fuid); 613} 614 615static uint64_t 616zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) 617{ 618 switch (type) { 619 case ZFS_PROP_USERUSED: 620 return (DMU_USERUSED_OBJECT); 621 case ZFS_PROP_GROUPUSED: 622 return (DMU_GROUPUSED_OBJECT); 623 case ZFS_PROP_USERQUOTA: 624 return (zfsvfs->z_userquota_obj); 625 case ZFS_PROP_GROUPQUOTA: 626 return (zfsvfs->z_groupquota_obj); 627 } 628 return (0); 629} 630 631int 632zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 633 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) 634{ 635 int error; 636 zap_cursor_t zc; 637 zap_attribute_t za; 638 zfs_useracct_t *buf = vbuf; 639 uint64_t obj; 640 641 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 642 return (ENOTSUP); 643 644 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 645 if (obj == 0) { 646 *bufsizep = 0; 647 return (0); 648 } 649 650 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); 651 (error = zap_cursor_retrieve(&zc, &za)) == 0; 652 zap_cursor_advance(&zc)) { 653 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > 654 *bufsizep) 655 break; 656 657 fuidstr_to_sid(zfsvfs, za.za_name, 658 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); 659 660 buf->zu_space = za.za_first_integer; 661 buf++; 662 } 663 if (error == ENOENT) 664 error = 0; 665 666 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); 667 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; 668 *cookiep = zap_cursor_serialize(&zc); 669 zap_cursor_fini(&zc); 670 return (error); 671} 672 673/* 674 * buf must be big enough (eg, 32 bytes) 675 */ 676static int 677id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, 678 char *buf, boolean_t addok) 679{ 680 uint64_t fuid; 681 int domainid = 0; 682 683 if (domain && domain[0]) { 684 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); 685 if (domainid == -1) 686 return (ENOENT); 687 } 688 fuid = FUID_ENCODE(domainid, rid); 689 (void) sprintf(buf, "%llx", (longlong_t)fuid); 690 return (0); 691} 692 693int 694zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 695 const char *domain, uint64_t rid, uint64_t *valp) 696{ 697 char buf[32]; 698 int err; 699 uint64_t obj; 700 701 *valp = 0; 702 703 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 704 return (ENOTSUP); 705 706 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 707 if (obj == 0) 708 return (0); 709 710 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); 711 if (err) 712 return (err); 713 714 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); 715 if (err == ENOENT) 716 err = 0; 717 return (err); 718} 719 720int 721zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 722 const char *domain, uint64_t rid, uint64_t quota) 723{ 724 char buf[32]; 725 int err; 726 dmu_tx_t *tx; 727 uint64_t *objp; 728 boolean_t fuid_dirtied; 729 730 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) 731 return (EINVAL); 732 733 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) 734 return (ENOTSUP); 735 736 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : 737 &zfsvfs->z_groupquota_obj; 738 739 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); 740 if (err) 741 return (err); 742 fuid_dirtied = zfsvfs->z_fuid_dirty; 743 744 tx = dmu_tx_create(zfsvfs->z_os); 745 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); 746 if (*objp == 0) { 747 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 748 zfs_userquota_prop_prefixes[type]); 749 } 750 if (fuid_dirtied) 751 zfs_fuid_txhold(zfsvfs, tx); 752 err = dmu_tx_assign(tx, TXG_WAIT); 753 if (err) { 754 dmu_tx_abort(tx); 755 return (err); 756 } 757 758 mutex_enter(&zfsvfs->z_lock); 759 if (*objp == 0) { 760 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, 761 DMU_OT_NONE, 0, tx); 762 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 763 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); 764 } 765 mutex_exit(&zfsvfs->z_lock); 766 767 if (quota == 0) { 768 err = zap_remove(zfsvfs->z_os, *objp, buf, tx); 769 if (err == ENOENT) 770 err = 0; 771 } else { 772 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); 773 } 774 ASSERT(err == 0); 775 if (fuid_dirtied) 776 zfs_fuid_sync(zfsvfs, tx); 777 dmu_tx_commit(tx); 778 return (err); 779} 780 781boolean_t 782zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) 783{ 784 char buf[32]; 785 uint64_t used, quota, usedobj, quotaobj; 786 int err; 787 788 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; 789 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 790 791 if (quotaobj == 0 || zfsvfs->z_replay) 792 return (B_FALSE); 793 794 (void) sprintf(buf, "%llx", (longlong_t)fuid); 795 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); 796 if (err != 0) 797 return (B_FALSE); 798 799 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); 800 if (err != 0) 801 return (B_FALSE); 802 return (used >= quota); 803} 804 805boolean_t 806zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) 807{ 808 uint64_t fuid; 809 uint64_t quotaobj; 810 811 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 812 813 fuid = isgroup ? zp->z_gid : zp->z_uid; 814 815 if (quotaobj == 0 || zfsvfs->z_replay) 816 return (B_FALSE); 817 818 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); 819} 820 821int 822zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 823{ 824 objset_t *os; 825 zfsvfs_t *zfsvfs; 826 uint64_t zval; 827 int i, error; 828 uint64_t sa_obj; 829 830 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 831 832 /* 833 * We claim to always be readonly so we can open snapshots; 834 * other ZPL code will prevent us from writing to snapshots. 835 */ 836 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 837 if (error) { 838 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 839 return (error); 840 } 841 842 /* 843 * Initialize the zfs-specific filesystem structure. 844 * Should probably make this a kmem cache, shuffle fields, 845 * and just bzero up to z_hold_mtx[]. 846 */ 847 zfsvfs->z_vfs = NULL; 848 zfsvfs->z_parent = zfsvfs; 849 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE; 850 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 851 zfsvfs->z_os = os; 852 853 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 854 if (error) { 855 goto out; 856 } else if (zfsvfs->z_version > 857 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 858 (void) printf("Can't mount a version %lld file system " 859 "on a version %lld pool\n. Pool must be upgraded to mount " 860 "this file system.", (u_longlong_t)zfsvfs->z_version, 861 (u_longlong_t)spa_version(dmu_objset_spa(os))); 862 error = ENOTSUP; 863 goto out; 864 } 865 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) 866 goto out; 867 zfsvfs->z_norm = (int)zval; 868 869 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) 870 goto out; 871 zfsvfs->z_utf8 = (zval != 0); 872 873 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) 874 goto out; 875 zfsvfs->z_case = (uint_t)zval; 876 877 /* 878 * Fold case on file systems that are always or sometimes case 879 * insensitive. 880 */ 881 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 882 zfsvfs->z_case == ZFS_CASE_MIXED) 883 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 884 885 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 886 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 887 888 if (zfsvfs->z_use_sa) { 889 /* should either have both of these objects or none */ 890 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 891 &sa_obj); 892 if (error) 893 return (error); 894 } else { 895 /* 896 * Pre SA versions file systems should never touch 897 * either the attribute registration or layout objects. 898 */ 899 sa_obj = 0; 900 } 901 902 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 903 &zfsvfs->z_attr_table); 904 if (error) 905 goto out; 906 907 if (zfsvfs->z_version >= ZPL_VERSION_SA) 908 sa_register_update_callback(os, zfs_sa_upgrade); 909 910 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 911 &zfsvfs->z_root); 912 if (error) 913 goto out; 914 ASSERT(zfsvfs->z_root != 0); 915 916 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 917 &zfsvfs->z_unlinkedobj); 918 if (error) 919 goto out; 920 921 error = zap_lookup(os, MASTER_NODE_OBJ, 922 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 923 8, 1, &zfsvfs->z_userquota_obj); 924 if (error && error != ENOENT) 925 goto out; 926 927 error = zap_lookup(os, MASTER_NODE_OBJ, 928 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 929 8, 1, &zfsvfs->z_groupquota_obj); 930 if (error && error != ENOENT) 931 goto out; 932 933 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 934 &zfsvfs->z_fuid_obj); 935 if (error && error != ENOENT) 936 goto out; 937 938 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 939 &zfsvfs->z_shares_dir); 940 if (error && error != ENOENT) 941 goto out; 942 943 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 944 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 945 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 946 offsetof(znode_t, z_link_node)); 947 rrw_init(&zfsvfs->z_teardown_lock); 948 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 949 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 950 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 951 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 952 953 *zfvp = zfsvfs; 954 return (0); 955 956out: 957 dmu_objset_disown(os, zfsvfs); 958 *zfvp = NULL; 959 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 960 return (error); 961} 962 963static int 964zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 965{ 966 int error; 967 968 error = zfs_register_callbacks(zfsvfs->z_vfs); 969 if (error) 970 return (error); 971 972 /* 973 * Set the objset user_ptr to track its zfsvfs. 974 */ 975 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 976 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 977 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 978 979 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 980 981 /* 982 * If we are not mounting (ie: online recv), then we don't 983 * have to worry about replaying the log as we blocked all 984 * operations out since we closed the ZIL. 985 */ 986 if (mounting) { 987 boolean_t readonly; 988 989 /* 990 * During replay we remove the read only flag to 991 * allow replays to succeed. 992 */ 993 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 994 if (readonly != 0) 995 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 996 else 997 zfs_unlinked_drain(zfsvfs); 998 999 /* 1000 * Parse and replay the intent log. 1001 * 1002 * Because of ziltest, this must be done after 1003 * zfs_unlinked_drain(). (Further note: ziltest 1004 * doesn't use readonly mounts, where 1005 * zfs_unlinked_drain() isn't called.) This is because 1006 * ziltest causes spa_sync() to think it's committed, 1007 * but actually it is not, so the intent log contains 1008 * many txg's worth of changes. 1009 * 1010 * In particular, if object N is in the unlinked set in 1011 * the last txg to actually sync, then it could be 1012 * actually freed in a later txg and then reallocated 1013 * in a yet later txg. This would write a "create 1014 * object N" record to the intent log. Normally, this 1015 * would be fine because the spa_sync() would have 1016 * written out the fact that object N is free, before 1017 * we could write the "create object N" intent log 1018 * record. 1019 * 1020 * But when we are in ziltest mode, we advance the "open 1021 * txg" without actually spa_sync()-ing the changes to 1022 * disk. So we would see that object N is still 1023 * allocated and in the unlinked set, and there is an 1024 * intent log record saying to allocate it. 1025 */ 1026 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1027 if (zil_replay_disable) { 1028 zil_destroy(zfsvfs->z_log, B_FALSE); 1029 } else { 1030 zfsvfs->z_replay = B_TRUE; 1031 zil_replay(zfsvfs->z_os, zfsvfs, 1032 zfs_replay_vector); 1033 zfsvfs->z_replay = B_FALSE; 1034 } 1035 } 1036 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1037 } 1038 1039 return (0); 1040} 1041 1042extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1043 1044void 1045zfsvfs_free(zfsvfs_t *zfsvfs) 1046{ 1047 int i; 1048 1049 /* 1050 * This is a barrier to prevent the filesystem from going away in 1051 * zfs_znode_move() until we can safely ensure that the filesystem is 1052 * not unmounted. We consider the filesystem valid before the barrier 1053 * and invalid after the barrier. 1054 */ 1055 rw_enter(&zfsvfs_lock, RW_READER); 1056 rw_exit(&zfsvfs_lock); 1057 1058 zfs_fuid_destroy(zfsvfs); 1059 1060 mutex_destroy(&zfsvfs->z_znodes_lock); 1061 mutex_destroy(&zfsvfs->z_lock); 1062 list_destroy(&zfsvfs->z_all_znodes); 1063 rrw_destroy(&zfsvfs->z_teardown_lock); 1064 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1065 rw_destroy(&zfsvfs->z_fuid_lock); 1066 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1067 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1068 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1069} 1070 1071static void 1072zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1073{ 1074 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1075 if (zfsvfs->z_vfs) { 1076 if (zfsvfs->z_use_fuids) { 1077 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1078 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1079 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1080 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1081 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1082 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1083 } else { 1084 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1085 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1086 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1087 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1088 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1089 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1090 } 1091 } 1092 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1093} 1094 1095static int 1096zfs_domount(vfs_t *vfsp, char *osname) 1097{ 1098 uint64_t recordsize, fsid_guid; 1099 int error = 0; 1100 zfsvfs_t *zfsvfs; 1101 vnode_t *vp; 1102 1103 ASSERT(vfsp); 1104 ASSERT(osname); 1105 1106 error = zfsvfs_create(osname, &zfsvfs); 1107 if (error) 1108 return (error); 1109 zfsvfs->z_vfs = vfsp; 1110 1111 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1112 NULL)) 1113 goto out; 1114 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1115 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1116 1117 vfsp->vfs_data = zfsvfs; 1118 vfsp->mnt_flag |= MNT_LOCAL; 1119 vfsp->mnt_kern_flag |= MNTK_MPSAFE; 1120 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1121 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1122 1123 /* 1124 * The fsid is 64 bits, composed of an 8-bit fs type, which 1125 * separates our fsid from any other filesystem types, and a 1126 * 56-bit objset unique ID. The objset unique ID is unique to 1127 * all objsets open on this system, provided by unique_create(). 1128 * The 8-bit fs type must be put in the low bits of fsid[1] 1129 * because that's where other Solaris filesystems put it. 1130 */ 1131 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1132 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1133 vfsp->vfs_fsid.val[0] = fsid_guid; 1134 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1135 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1136 1137 /* 1138 * Set features for file system. 1139 */ 1140 zfs_set_fuid_feature(zfsvfs); 1141 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1142 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1143 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1144 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1145 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1146 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1147 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1148 } 1149 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1150 1151 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1152 uint64_t pval; 1153 1154 atime_changed_cb(zfsvfs, B_FALSE); 1155 readonly_changed_cb(zfsvfs, B_TRUE); 1156 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1157 goto out; 1158 xattr_changed_cb(zfsvfs, pval); 1159 zfsvfs->z_issnap = B_TRUE; 1160 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1161 1162 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1163 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1164 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1165 } else { 1166 error = zfsvfs_setup(zfsvfs, B_TRUE); 1167 } 1168 1169 vfs_mountedfrom(vfsp, osname); 1170 /* Grab extra reference. */ 1171 VERIFY(VFS_ROOT(vfsp, LK_EXCLUSIVE, &vp) == 0); 1172 VOP_UNLOCK(vp, 0); 1173 1174 if (!zfsvfs->z_issnap) 1175 zfsctl_create(zfsvfs); 1176out: 1177 if (error) { 1178 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1179 zfsvfs_free(zfsvfs); 1180 } else { 1181 atomic_add_32(&zfs_active_fs_count, 1); 1182 } 1183 1184 return (error); 1185} 1186 1187void 1188zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1189{ 1190 objset_t *os = zfsvfs->z_os; 1191 struct dsl_dataset *ds; 1192 1193 /* 1194 * Unregister properties. 1195 */ 1196 if (!dmu_objset_is_snapshot(os)) { 1197 ds = dmu_objset_ds(os); 1198 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, 1199 zfsvfs) == 0); 1200 1201 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, 1202 zfsvfs) == 0); 1203 1204 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, 1205 zfsvfs) == 0); 1206 1207 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, 1208 zfsvfs) == 0); 1209 1210 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, 1211 zfsvfs) == 0); 1212 1213 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, 1214 zfsvfs) == 0); 1215 1216 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, 1217 zfsvfs) == 0); 1218 1219 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, 1220 zfsvfs) == 0); 1221 1222 VERIFY(dsl_prop_unregister(ds, "aclinherit", 1223 acl_inherit_changed_cb, zfsvfs) == 0); 1224 1225 VERIFY(dsl_prop_unregister(ds, "vscan", 1226 vscan_changed_cb, zfsvfs) == 0); 1227 } 1228} 1229 1230#ifdef SECLABEL 1231/* 1232 * Convert a decimal digit string to a uint64_t integer. 1233 */ 1234static int 1235str_to_uint64(char *str, uint64_t *objnum) 1236{ 1237 uint64_t num = 0; 1238 1239 while (*str) { 1240 if (*str < '0' || *str > '9') 1241 return (EINVAL); 1242 1243 num = num*10 + *str++ - '0'; 1244 } 1245 1246 *objnum = num; 1247 return (0); 1248} 1249 1250/* 1251 * The boot path passed from the boot loader is in the form of 1252 * "rootpool-name/root-filesystem-object-number'. Convert this 1253 * string to a dataset name: "rootpool-name/root-filesystem-name". 1254 */ 1255static int 1256zfs_parse_bootfs(char *bpath, char *outpath) 1257{ 1258 char *slashp; 1259 uint64_t objnum; 1260 int error; 1261 1262 if (*bpath == 0 || *bpath == '/') 1263 return (EINVAL); 1264 1265 (void) strcpy(outpath, bpath); 1266 1267 slashp = strchr(bpath, '/'); 1268 1269 /* if no '/', just return the pool name */ 1270 if (slashp == NULL) { 1271 return (0); 1272 } 1273 1274 /* if not a number, just return the root dataset name */ 1275 if (str_to_uint64(slashp+1, &objnum)) { 1276 return (0); 1277 } 1278 1279 *slashp = '\0'; 1280 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1281 *slashp = '/'; 1282 1283 return (error); 1284} 1285 1286/* 1287 * zfs_check_global_label: 1288 * Check that the hex label string is appropriate for the dataset 1289 * being mounted into the global_zone proper. 1290 * 1291 * Return an error if the hex label string is not default or 1292 * admin_low/admin_high. For admin_low labels, the corresponding 1293 * dataset must be readonly. 1294 */ 1295int 1296zfs_check_global_label(const char *dsname, const char *hexsl) 1297{ 1298 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1299 return (0); 1300 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1301 return (0); 1302 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1303 /* must be readonly */ 1304 uint64_t rdonly; 1305 1306 if (dsl_prop_get_integer(dsname, 1307 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1308 return (EACCES); 1309 return (rdonly ? 0 : EACCES); 1310 } 1311 return (EACCES); 1312} 1313 1314/* 1315 * zfs_mount_label_policy: 1316 * Determine whether the mount is allowed according to MAC check. 1317 * by comparing (where appropriate) label of the dataset against 1318 * the label of the zone being mounted into. If the dataset has 1319 * no label, create one. 1320 * 1321 * Returns: 1322 * 0 : access allowed 1323 * >0 : error code, such as EACCES 1324 */ 1325static int 1326zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1327{ 1328 int error, retv; 1329 zone_t *mntzone = NULL; 1330 ts_label_t *mnt_tsl; 1331 bslabel_t *mnt_sl; 1332 bslabel_t ds_sl; 1333 char ds_hexsl[MAXNAMELEN]; 1334 1335 retv = EACCES; /* assume the worst */ 1336 1337 /* 1338 * Start by getting the dataset label if it exists. 1339 */ 1340 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1341 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1342 if (error) 1343 return (EACCES); 1344 1345 /* 1346 * If labeling is NOT enabled, then disallow the mount of datasets 1347 * which have a non-default label already. No other label checks 1348 * are needed. 1349 */ 1350 if (!is_system_labeled()) { 1351 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1352 return (0); 1353 return (EACCES); 1354 } 1355 1356 /* 1357 * Get the label of the mountpoint. If mounting into the global 1358 * zone (i.e. mountpoint is not within an active zone and the 1359 * zoned property is off), the label must be default or 1360 * admin_low/admin_high only; no other checks are needed. 1361 */ 1362 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1363 if (mntzone->zone_id == GLOBAL_ZONEID) { 1364 uint64_t zoned; 1365 1366 zone_rele(mntzone); 1367 1368 if (dsl_prop_get_integer(osname, 1369 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1370 return (EACCES); 1371 if (!zoned) 1372 return (zfs_check_global_label(osname, ds_hexsl)); 1373 else 1374 /* 1375 * This is the case of a zone dataset being mounted 1376 * initially, before the zone has been fully created; 1377 * allow this mount into global zone. 1378 */ 1379 return (0); 1380 } 1381 1382 mnt_tsl = mntzone->zone_slabel; 1383 ASSERT(mnt_tsl != NULL); 1384 label_hold(mnt_tsl); 1385 mnt_sl = label2bslabel(mnt_tsl); 1386 1387 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1388 /* 1389 * The dataset doesn't have a real label, so fabricate one. 1390 */ 1391 char *str = NULL; 1392 1393 if (l_to_str_internal(mnt_sl, &str) == 0 && 1394 dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1395 ZPROP_SRC_LOCAL, 1, strlen(str) + 1, 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 (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 (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 (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 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1557 return (EPERM); 1558 1559 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1560 return (EINVAL); 1561 1562 /* 1563 * If full-owner-access is enabled and delegated administration is 1564 * turned on, we must set nosuid. 1565 */ 1566 if (zfs_super_owner && 1567 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1568 secpolicy_fs_mount_clearopts(cr, vfsp); 1569 } 1570 1571 /* 1572 * Check for mount privilege? 1573 * 1574 * If we don't have privilege then see if 1575 * we have local permission to allow it 1576 */ 1577 error = secpolicy_fs_mount(cr, mvp, vfsp); 1578 if (error) { 1579 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1580 goto out; 1581 1582 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1583 vattr_t vattr; 1584 1585 /* 1586 * Make sure user is the owner of the mount point 1587 * or has sufficient privileges. 1588 */ 1589 1590 vattr.va_mask = AT_UID; 1591 1592 vn_lock(mvp, LK_SHARED | LK_RETRY); 1593 if (VOP_GETATTR(mvp, &vattr, cr)) { 1594 VOP_UNLOCK(mvp, 0); 1595 goto out; 1596 } 1597 1598 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1599 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1600 VOP_UNLOCK(mvp, 0); 1601 goto out; 1602 } 1603 VOP_UNLOCK(mvp, 0); 1604 } 1605 1606 secpolicy_fs_mount_clearopts(cr, vfsp); 1607 } 1608 1609 /* 1610 * Refuse to mount a filesystem if we are in a local zone and the 1611 * dataset is not visible. 1612 */ 1613 if (!INGLOBALZONE(curthread) && 1614 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1615 error = EPERM; 1616 goto out; 1617 } 1618 1619#ifdef SECLABEL 1620 error = zfs_mount_label_policy(vfsp, osname); 1621 if (error) 1622 goto out; 1623#endif 1624 1625 vfsp->vfs_flag |= MNT_NFS4ACLS; 1626 1627 /* 1628 * When doing a remount, we simply refresh our temporary properties 1629 * according to those options set in the current VFS options. 1630 */ 1631 if (vfsp->vfs_flag & MS_REMOUNT) { 1632 /* refresh mount options */ 1633 zfs_unregister_callbacks(vfsp->vfs_data); 1634 error = zfs_register_callbacks(vfsp); 1635 goto out; 1636 } 1637 1638 /* Initial root mount: try hard to import the requested root pool. */ 1639 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1640 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1641 char pname[MAXNAMELEN]; 1642 spa_t *spa; 1643 int prefer_cache; 1644 1645 error = getpoolname(osname, pname); 1646 if (error) 1647 goto out; 1648 1649 prefer_cache = 1; 1650 TUNABLE_INT_FETCH("vfs.zfs.rootpool.prefer_cached_config", 1651 &prefer_cache); 1652 mutex_enter(&spa_namespace_lock); 1653 spa = spa_lookup(pname); 1654 mutex_exit(&spa_namespace_lock); 1655 if (!prefer_cache || spa == NULL) { 1656 error = spa_import_rootpool(pname); 1657 if (error) 1658 goto out; 1659 } 1660 } 1661 DROP_GIANT(); 1662 error = zfs_domount(vfsp, osname); 1663 PICKUP_GIANT(); 1664 1665#ifdef sun 1666 /* 1667 * Add an extra VFS_HOLD on our parent vfs so that it can't 1668 * disappear due to a forced unmount. 1669 */ 1670 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1671 VFS_HOLD(mvp->v_vfsp); 1672#endif /* sun */ 1673 1674out: 1675 return (error); 1676} 1677 1678static int 1679zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1680{ 1681 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1682 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1683 1684 statp->f_version = STATFS_VERSION; 1685 1686 ZFS_ENTER(zfsvfs); 1687 1688 dmu_objset_space(zfsvfs->z_os, 1689 &refdbytes, &availbytes, &usedobjs, &availobjs); 1690 1691 /* 1692 * The underlying storage pool actually uses multiple block sizes. 1693 * We report the fragsize as the smallest block size we support, 1694 * and we report our blocksize as the filesystem's maximum blocksize. 1695 */ 1696 statp->f_bsize = SPA_MINBLOCKSIZE; 1697 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1698 1699 /* 1700 * The following report "total" blocks of various kinds in the 1701 * file system, but reported in terms of f_frsize - the 1702 * "fragment" size. 1703 */ 1704 1705 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1706 statp->f_bfree = availbytes / statp->f_bsize; 1707 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1708 1709 /* 1710 * statvfs() should really be called statufs(), because it assumes 1711 * static metadata. ZFS doesn't preallocate files, so the best 1712 * we can do is report the max that could possibly fit in f_files, 1713 * and that minus the number actually used in f_ffree. 1714 * For f_ffree, report the smaller of the number of object available 1715 * and the number of blocks (each object will take at least a block). 1716 */ 1717 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1718 statp->f_files = statp->f_ffree + usedobjs; 1719 1720 /* 1721 * We're a zfs filesystem. 1722 */ 1723 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1724 1725 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1726 sizeof(statp->f_mntfromname)); 1727 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1728 sizeof(statp->f_mntonname)); 1729 1730 statp->f_namemax = ZFS_MAXNAMELEN; 1731 1732 ZFS_EXIT(zfsvfs); 1733 return (0); 1734} 1735 1736int 1737zfs_vnode_lock(vnode_t *vp, int flags) 1738{ 1739 int error; 1740 1741 ASSERT(vp != NULL); 1742 1743 error = vn_lock(vp, flags); 1744 return (error); 1745} 1746 1747static int 1748zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1749{ 1750 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1751 znode_t *rootzp; 1752 int error; 1753 1754 ZFS_ENTER_NOERROR(zfsvfs); 1755 1756 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1757 if (error == 0) 1758 *vpp = ZTOV(rootzp); 1759 1760 ZFS_EXIT(zfsvfs); 1761 1762 if (error == 0) { 1763 error = zfs_vnode_lock(*vpp, flags); 1764 if (error == 0) 1765 (*vpp)->v_vflag |= VV_ROOT; 1766 } 1767 if (error != 0) 1768 *vpp = NULL; 1769 1770 return (error); 1771} 1772 1773/* 1774 * Teardown the zfsvfs::z_os. 1775 * 1776 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1777 * and 'z_teardown_inactive_lock' held. 1778 */ 1779static int 1780zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1781{ 1782 znode_t *zp; 1783 1784 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1785 1786 if (!unmounting) { 1787 /* 1788 * We purge the parent filesystem's vfsp as the parent 1789 * filesystem and all of its snapshots have their vnode's 1790 * v_vfsp set to the parent's filesystem's vfsp. Note, 1791 * 'z_parent' is self referential for non-snapshots. 1792 */ 1793 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1794#ifdef FREEBSD_NAMECACHE 1795 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1796#endif 1797 } 1798 1799 /* 1800 * Close the zil. NB: Can't close the zil while zfs_inactive 1801 * threads are blocked as zil_close can call zfs_inactive. 1802 */ 1803 if (zfsvfs->z_log) { 1804 zil_close(zfsvfs->z_log); 1805 zfsvfs->z_log = NULL; 1806 } 1807 1808 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1809 1810 /* 1811 * If we are not unmounting (ie: online recv) and someone already 1812 * unmounted this file system while we were doing the switcheroo, 1813 * or a reopen of z_os failed then just bail out now. 1814 */ 1815 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1816 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1817 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1818 return (EIO); 1819 } 1820 1821 /* 1822 * At this point there are no vops active, and any new vops will 1823 * fail with EIO since we have z_teardown_lock for writer (only 1824 * relavent for forced unmount). 1825 * 1826 * Release all holds on dbufs. 1827 */ 1828 mutex_enter(&zfsvfs->z_znodes_lock); 1829 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1830 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1831 if (zp->z_sa_hdl) { 1832 ASSERT(ZTOV(zp)->v_count >= 0); 1833 zfs_znode_dmu_fini(zp); 1834 } 1835 mutex_exit(&zfsvfs->z_znodes_lock); 1836 1837 /* 1838 * If we are unmounting, set the unmounted flag and let new vops 1839 * unblock. zfs_inactive will have the unmounted behavior, and all 1840 * other vops will fail with EIO. 1841 */ 1842 if (unmounting) { 1843 zfsvfs->z_unmounted = B_TRUE; 1844 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1845 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1846 1847#ifdef __FreeBSD__ 1848 /* 1849 * Some znodes might not be fully reclaimed, wait for them. 1850 */ 1851 mutex_enter(&zfsvfs->z_znodes_lock); 1852 while (list_head(&zfsvfs->z_all_znodes) != NULL) { 1853 msleep(zfsvfs, &zfsvfs->z_znodes_lock, 0, 1854 "zteardown", 0); 1855 } 1856 mutex_exit(&zfsvfs->z_znodes_lock); 1857#endif 1858 } 1859 1860 /* 1861 * z_os will be NULL if there was an error in attempting to reopen 1862 * zfsvfs, so just return as the properties had already been 1863 * unregistered and cached data had been evicted before. 1864 */ 1865 if (zfsvfs->z_os == NULL) 1866 return (0); 1867 1868 /* 1869 * Unregister properties. 1870 */ 1871 zfs_unregister_callbacks(zfsvfs); 1872 1873 /* 1874 * Evict cached data 1875 */ 1876 if (dmu_objset_is_dirty_anywhere(zfsvfs->z_os)) 1877 if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1878 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1879 (void) dmu_objset_evict_dbufs(zfsvfs->z_os); 1880 1881 return (0); 1882} 1883 1884/*ARGSUSED*/ 1885static int 1886zfs_umount(vfs_t *vfsp, int fflag) 1887{ 1888 kthread_t *td = curthread; 1889 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1890 objset_t *os; 1891 cred_t *cr = td->td_ucred; 1892 int ret; 1893 1894 ret = secpolicy_fs_unmount(cr, vfsp); 1895 if (ret) { 1896 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1897 ZFS_DELEG_PERM_MOUNT, cr)) 1898 return (ret); 1899 } 1900 1901 /* 1902 * We purge the parent filesystem's vfsp as the parent filesystem 1903 * and all of its snapshots have their vnode's v_vfsp set to the 1904 * parent's filesystem's vfsp. Note, 'z_parent' is self 1905 * referential for non-snapshots. 1906 */ 1907 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1908 1909 /* 1910 * Unmount any snapshots mounted under .zfs before unmounting the 1911 * dataset itself. 1912 */ 1913 if (zfsvfs->z_ctldir != NULL) { 1914 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1915 return (ret); 1916 ret = vflush(vfsp, 0, 0, td); 1917 ASSERT(ret == EBUSY); 1918 if (!(fflag & MS_FORCE)) { 1919 if (zfsvfs->z_ctldir->v_count > 1) 1920 return (EBUSY); 1921 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1922 } 1923 zfsctl_destroy(zfsvfs); 1924 ASSERT(zfsvfs->z_ctldir == NULL); 1925 } 1926 1927 if (fflag & MS_FORCE) { 1928 /* 1929 * Mark file system as unmounted before calling 1930 * vflush(FORCECLOSE). This way we ensure no future vnops 1931 * will be called and risk operating on DOOMED vnodes. 1932 */ 1933 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1934 zfsvfs->z_unmounted = B_TRUE; 1935 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1936 } 1937 1938 /* 1939 * Flush all the files. 1940 */ 1941 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1942 if (ret != 0) { 1943 if (!zfsvfs->z_issnap) { 1944 zfsctl_create(zfsvfs); 1945 ASSERT(zfsvfs->z_ctldir != NULL); 1946 } 1947 return (ret); 1948 } 1949 1950 if (!(fflag & MS_FORCE)) { 1951 /* 1952 * Check the number of active vnodes in the file system. 1953 * Our count is maintained in the vfs structure, but the 1954 * number is off by 1 to indicate a hold on the vfs 1955 * structure itself. 1956 * 1957 * The '.zfs' directory maintains a reference of its 1958 * own, and any active references underneath are 1959 * reflected in the vnode count. 1960 */ 1961 if (zfsvfs->z_ctldir == NULL) { 1962 if (vfsp->vfs_count > 1) 1963 return (EBUSY); 1964 } else { 1965 if (vfsp->vfs_count > 2 || 1966 zfsvfs->z_ctldir->v_count > 1) 1967 return (EBUSY); 1968 } 1969 } 1970 1971 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1972 os = zfsvfs->z_os; 1973 1974 /* 1975 * z_os will be NULL if there was an error in 1976 * attempting to reopen zfsvfs. 1977 */ 1978 if (os != NULL) { 1979 /* 1980 * Unset the objset user_ptr. 1981 */ 1982 mutex_enter(&os->os_user_ptr_lock); 1983 dmu_objset_set_user(os, NULL); 1984 mutex_exit(&os->os_user_ptr_lock); 1985 1986 /* 1987 * Finally release the objset 1988 */ 1989 dmu_objset_disown(os, zfsvfs); 1990 } 1991 1992 /* 1993 * We can now safely destroy the '.zfs' directory node. 1994 */ 1995 if (zfsvfs->z_ctldir != NULL) 1996 zfsctl_destroy(zfsvfs); 1997 if (zfsvfs->z_issnap) { 1998 vnode_t *svp = vfsp->mnt_vnodecovered; 1999 2000 if (svp->v_count >= 2) 2001 VN_RELE(svp); 2002 } 2003 zfs_freevfs(vfsp); 2004 2005 return (0); 2006} 2007 2008static int 2009zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2010{ 2011 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2012 znode_t *zp; 2013 int err; 2014 2015 /* 2016 * zfs_zget() can't operate on virtual entries like .zfs/ or 2017 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2018 * This will make NFS to switch to LOOKUP instead of using VGET. 2019 */ 2020 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR) 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 = zfs_vnode_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 (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 (EINVAL); 2108 } 2109 2110 /* A zero fid_gen means we are in the .zfs control directories */ 2111 if (fid_gen == 0 && 2112 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 2113 *vpp = zfsvfs->z_ctldir; 2114 ASSERT(*vpp != NULL); 2115 if (object == ZFSCTL_INO_SNAPDIR) { 2116 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2117 0, NULL, NULL, NULL, NULL, NULL) == 0); 2118 } else { 2119 VN_HOLD(*vpp); 2120 } 2121 ZFS_EXIT(zfsvfs); 2122 err = zfs_vnode_lock(*vpp, flags | LK_RETRY); 2123 if (err != 0) 2124 *vpp = NULL; 2125 return (err); 2126 } 2127 2128 gen_mask = -1ULL >> (64 - 8 * i); 2129 2130 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2131 if (err = zfs_zget(zfsvfs, object, &zp)) { 2132 ZFS_EXIT(zfsvfs); 2133 return (err); 2134 } 2135 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2136 sizeof (uint64_t)); 2137 zp_gen = zp_gen & gen_mask; 2138 if (zp_gen == 0) 2139 zp_gen = 1; 2140 if (zp->z_unlinked || zp_gen != fid_gen) { 2141 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2142 VN_RELE(ZTOV(zp)); 2143 ZFS_EXIT(zfsvfs); 2144 return (EINVAL); 2145 } 2146 2147 *vpp = ZTOV(zp); 2148 ZFS_EXIT(zfsvfs); 2149 err = zfs_vnode_lock(*vpp, flags | LK_RETRY); 2150 if (err == 0) 2151 vnode_create_vobject(*vpp, zp->z_size, curthread); 2152 else 2153 *vpp = NULL; 2154 return (err); 2155} 2156 2157/* 2158 * Block out VOPs and close zfsvfs_t::z_os 2159 * 2160 * Note, if successful, then we return with the 'z_teardown_lock' and 2161 * 'z_teardown_inactive_lock' write held. 2162 */ 2163int 2164zfs_suspend_fs(zfsvfs_t *zfsvfs) 2165{ 2166 int error; 2167 2168 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2169 return (error); 2170 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 2171 2172 return (0); 2173} 2174 2175/* 2176 * Reopen zfsvfs_t::z_os and release VOPs. 2177 */ 2178int 2179zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2180{ 2181 int err; 2182 2183 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2184 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2185 2186 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs, 2187 &zfsvfs->z_os); 2188 if (err) { 2189 zfsvfs->z_os = NULL; 2190 } else { 2191 znode_t *zp; 2192 uint64_t sa_obj = 0; 2193 2194 /* 2195 * Make sure version hasn't changed 2196 */ 2197 2198 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, 2199 &zfsvfs->z_version); 2200 2201 if (err) 2202 goto bail; 2203 2204 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, 2205 ZFS_SA_ATTRS, 8, 1, &sa_obj); 2206 2207 if (err && zfsvfs->z_version >= ZPL_VERSION_SA) 2208 goto bail; 2209 2210 if ((err = sa_setup(zfsvfs->z_os, sa_obj, 2211 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) 2212 goto bail; 2213 2214 if (zfsvfs->z_version >= ZPL_VERSION_SA) 2215 sa_register_update_callback(zfsvfs->z_os, 2216 zfs_sa_upgrade); 2217 2218 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2219 2220 zfs_set_fuid_feature(zfsvfs); 2221 2222 /* 2223 * Attempt to re-establish all the active znodes with 2224 * their dbufs. If a zfs_rezget() fails, then we'll let 2225 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2226 * when they try to use their znode. 2227 */ 2228 mutex_enter(&zfsvfs->z_znodes_lock); 2229 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2230 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2231 (void) zfs_rezget(zp); 2232 } 2233 mutex_exit(&zfsvfs->z_znodes_lock); 2234 } 2235 2236bail: 2237 /* release the VOPs */ 2238 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2239 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 2240 2241 if (err) { 2242 /* 2243 * Since we couldn't reopen zfsvfs::z_os, or 2244 * setup the sa framework force unmount this file system. 2245 */ 2246 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) 2247 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2248 } 2249 return (err); 2250} 2251 2252static void 2253zfs_freevfs(vfs_t *vfsp) 2254{ 2255 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2256 2257#ifdef sun 2258 /* 2259 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2260 * from zfs_mount(). Release it here. If we came through 2261 * zfs_mountroot() instead, we didn't grab an extra hold, so 2262 * skip the VFS_RELE for rootvfs. 2263 */ 2264 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2265 VFS_RELE(zfsvfs->z_parent->z_vfs); 2266#endif /* sun */ 2267 2268 zfsvfs_free(zfsvfs); 2269 2270 atomic_add_32(&zfs_active_fs_count, -1); 2271} 2272 2273#ifdef __i386__ 2274static int desiredvnodes_backup; 2275#endif 2276 2277static void 2278zfs_vnodes_adjust(void) 2279{ 2280#ifdef __i386__ 2281 int newdesiredvnodes; 2282 2283 desiredvnodes_backup = desiredvnodes; 2284 2285 /* 2286 * We calculate newdesiredvnodes the same way it is done in 2287 * vntblinit(). If it is equal to desiredvnodes, it means that 2288 * it wasn't tuned by the administrator and we can tune it down. 2289 */ 2290 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2291 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2292 sizeof(struct vnode)))); 2293 if (newdesiredvnodes == desiredvnodes) 2294 desiredvnodes = (3 * newdesiredvnodes) / 4; 2295#endif 2296} 2297 2298static void 2299zfs_vnodes_adjust_back(void) 2300{ 2301 2302#ifdef __i386__ 2303 desiredvnodes = desiredvnodes_backup; 2304#endif 2305} 2306 2307void 2308zfs_init(void) 2309{ 2310 2311 printf("ZFS filesystem version " ZPL_VERSION_STRING "\n"); 2312 2313 /* 2314 * Initialize .zfs directory structures 2315 */ 2316 zfsctl_init(); 2317 2318 /* 2319 * Initialize znode cache, vnode ops, etc... 2320 */ 2321 zfs_znode_init(); 2322 2323 /* 2324 * Reduce number of vnodes. Originally number of vnodes is calculated 2325 * with UFS inode in mind. We reduce it here, because it's too big for 2326 * ZFS/i386. 2327 */ 2328 zfs_vnodes_adjust(); 2329 2330 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2331} 2332 2333void 2334zfs_fini(void) 2335{ 2336 zfsctl_fini(); 2337 zfs_znode_fini(); 2338 zfs_vnodes_adjust_back(); 2339} 2340 2341int 2342zfs_busy(void) 2343{ 2344 return (zfs_active_fs_count != 0); 2345} 2346 2347int 2348zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2349{ 2350 int error; 2351 objset_t *os = zfsvfs->z_os; 2352 dmu_tx_t *tx; 2353 2354 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2355 return (EINVAL); 2356 2357 if (newvers < zfsvfs->z_version) 2358 return (EINVAL); 2359 2360 if (zfs_spa_version_map(newvers) > 2361 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2362 return (ENOTSUP); 2363 2364 tx = dmu_tx_create(os); 2365 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2366 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2367 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2368 ZFS_SA_ATTRS); 2369 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2370 } 2371 error = dmu_tx_assign(tx, TXG_WAIT); 2372 if (error) { 2373 dmu_tx_abort(tx); 2374 return (error); 2375 } 2376 2377 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2378 8, 1, &newvers, tx); 2379 2380 if (error) { 2381 dmu_tx_commit(tx); 2382 return (error); 2383 } 2384 2385 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2386 uint64_t sa_obj; 2387 2388 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2389 SPA_VERSION_SA); 2390 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2391 DMU_OT_NONE, 0, tx); 2392 2393 error = zap_add(os, MASTER_NODE_OBJ, 2394 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2395 ASSERT3U(error, ==, 0); 2396 2397 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2398 sa_register_update_callback(os, zfs_sa_upgrade); 2399 } 2400 2401 spa_history_log_internal(LOG_DS_UPGRADE, 2402 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu", 2403 zfsvfs->z_version, newvers, dmu_objset_id(os)); 2404 2405 dmu_tx_commit(tx); 2406 2407 zfsvfs->z_version = newvers; 2408 2409 zfs_set_fuid_feature(zfsvfs); 2410 2411 return (0); 2412} 2413 2414/* 2415 * Read a property stored within the master node. 2416 */ 2417int 2418zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2419{ 2420 const char *pname; 2421 int error = ENOENT; 2422 2423 /* 2424 * Look up the file system's value for the property. For the 2425 * version property, we look up a slightly different string. 2426 */ 2427 if (prop == ZFS_PROP_VERSION) 2428 pname = ZPL_VERSION_STR; 2429 else 2430 pname = zfs_prop_to_name(prop); 2431 2432 if (os != NULL) 2433 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2434 2435 if (error == ENOENT) { 2436 /* No value set, use the default value */ 2437 switch (prop) { 2438 case ZFS_PROP_VERSION: 2439 *value = ZPL_VERSION; 2440 break; 2441 case ZFS_PROP_NORMALIZE: 2442 case ZFS_PROP_UTF8ONLY: 2443 *value = 0; 2444 break; 2445 case ZFS_PROP_CASE: 2446 *value = ZFS_CASE_SENSITIVE; 2447 break; 2448 default: 2449 return (error); 2450 } 2451 error = 0; 2452 } 2453 return (error); 2454} 2455 2456#ifdef _KERNEL 2457void 2458zfsvfs_update_fromname(const char *oldname, const char *newname) 2459{ 2460 char tmpbuf[MAXPATHLEN]; 2461 struct mount *mp; 2462 char *fromname; 2463 size_t oldlen; 2464 2465 oldlen = strlen(oldname); 2466 2467 mtx_lock(&mountlist_mtx); 2468 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2469 fromname = mp->mnt_stat.f_mntfromname; 2470 if (strcmp(fromname, oldname) == 0) { 2471 (void)strlcpy(fromname, newname, 2472 sizeof(mp->mnt_stat.f_mntfromname)); 2473 continue; 2474 } 2475 if (strncmp(fromname, oldname, oldlen) == 0 && 2476 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2477 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2478 newname, fromname + oldlen); 2479 (void)strlcpy(fromname, tmpbuf, 2480 sizeof(mp->mnt_stat.f_mntfromname)); 2481 continue; 2482 } 2483 } 2484 mtx_unlock(&mountlist_mtx); 2485} 2486#endif 2487