zfs_vfsops.c revision 243674
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/sa_impl.h> 53#include <sys/varargs.h> 54#include <sys/policy.h> 55#include <sys/atomic.h> 56#include <sys/zfs_ioctl.h> 57#include <sys/zfs_ctldir.h> 58#include <sys/zfs_fuid.h> 59#include <sys/sunddi.h> 60#include <sys/dnlc.h> 61#include <sys/dmu_objset.h> 62#include <sys/spa_boot.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 int error = 0; 554 555 /* 556 * Is it a valid type of object to track? 557 */ 558 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 559 return (ENOENT); 560 561 /* 562 * If we have a NULL data pointer 563 * then assume the id's aren't changing and 564 * return EEXIST to the dmu to let it know to 565 * use the same ids 566 */ 567 if (data == NULL) 568 return (EEXIST); 569 570 if (bonustype == DMU_OT_ZNODE) { 571 znode_phys_t *znp = data; 572 *userp = znp->zp_uid; 573 *groupp = znp->zp_gid; 574 } else { 575 int hdrsize; 576 sa_hdr_phys_t *sap = data; 577 sa_hdr_phys_t sa = *sap; 578 boolean_t swap = B_FALSE; 579 580 ASSERT(bonustype == DMU_OT_SA); 581 582 if (sa.sa_magic == 0) { 583 /* 584 * This should only happen for newly created 585 * files that haven't had the znode data filled 586 * in yet. 587 */ 588 *userp = 0; 589 *groupp = 0; 590 return (0); 591 } 592 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { 593 sa.sa_magic = SA_MAGIC; 594 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); 595 swap = B_TRUE; 596 } else { 597 VERIFY3U(sa.sa_magic, ==, SA_MAGIC); 598 } 599 600 hdrsize = sa_hdrsize(&sa); 601 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); 602 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + 603 SA_UID_OFFSET)); 604 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + 605 SA_GID_OFFSET)); 606 if (swap) { 607 *userp = BSWAP_64(*userp); 608 *groupp = BSWAP_64(*groupp); 609 } 610 } 611 return (error); 612} 613 614static void 615fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, 616 char *domainbuf, int buflen, uid_t *ridp) 617{ 618 uint64_t fuid; 619 const char *domain; 620 621 fuid = strtonum(fuidstr, NULL); 622 623 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); 624 if (domain) 625 (void) strlcpy(domainbuf, domain, buflen); 626 else 627 domainbuf[0] = '\0'; 628 *ridp = FUID_RID(fuid); 629} 630 631static uint64_t 632zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) 633{ 634 switch (type) { 635 case ZFS_PROP_USERUSED: 636 return (DMU_USERUSED_OBJECT); 637 case ZFS_PROP_GROUPUSED: 638 return (DMU_GROUPUSED_OBJECT); 639 case ZFS_PROP_USERQUOTA: 640 return (zfsvfs->z_userquota_obj); 641 case ZFS_PROP_GROUPQUOTA: 642 return (zfsvfs->z_groupquota_obj); 643 } 644 return (0); 645} 646 647int 648zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 649 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) 650{ 651 int error; 652 zap_cursor_t zc; 653 zap_attribute_t za; 654 zfs_useracct_t *buf = vbuf; 655 uint64_t obj; 656 657 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 658 return (ENOTSUP); 659 660 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 661 if (obj == 0) { 662 *bufsizep = 0; 663 return (0); 664 } 665 666 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); 667 (error = zap_cursor_retrieve(&zc, &za)) == 0; 668 zap_cursor_advance(&zc)) { 669 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > 670 *bufsizep) 671 break; 672 673 fuidstr_to_sid(zfsvfs, za.za_name, 674 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); 675 676 buf->zu_space = za.za_first_integer; 677 buf++; 678 } 679 if (error == ENOENT) 680 error = 0; 681 682 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); 683 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; 684 *cookiep = zap_cursor_serialize(&zc); 685 zap_cursor_fini(&zc); 686 return (error); 687} 688 689/* 690 * buf must be big enough (eg, 32 bytes) 691 */ 692static int 693id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, 694 char *buf, boolean_t addok) 695{ 696 uint64_t fuid; 697 int domainid = 0; 698 699 if (domain && domain[0]) { 700 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); 701 if (domainid == -1) 702 return (ENOENT); 703 } 704 fuid = FUID_ENCODE(domainid, rid); 705 (void) sprintf(buf, "%llx", (longlong_t)fuid); 706 return (0); 707} 708 709int 710zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 711 const char *domain, uint64_t rid, uint64_t *valp) 712{ 713 char buf[32]; 714 int err; 715 uint64_t obj; 716 717 *valp = 0; 718 719 if (!dmu_objset_userspace_present(zfsvfs->z_os)) 720 return (ENOTSUP); 721 722 obj = zfs_userquota_prop_to_obj(zfsvfs, type); 723 if (obj == 0) 724 return (0); 725 726 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); 727 if (err) 728 return (err); 729 730 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); 731 if (err == ENOENT) 732 err = 0; 733 return (err); 734} 735 736int 737zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 738 const char *domain, uint64_t rid, uint64_t quota) 739{ 740 char buf[32]; 741 int err; 742 dmu_tx_t *tx; 743 uint64_t *objp; 744 boolean_t fuid_dirtied; 745 746 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) 747 return (EINVAL); 748 749 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) 750 return (ENOTSUP); 751 752 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : 753 &zfsvfs->z_groupquota_obj; 754 755 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); 756 if (err) 757 return (err); 758 fuid_dirtied = zfsvfs->z_fuid_dirty; 759 760 tx = dmu_tx_create(zfsvfs->z_os); 761 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); 762 if (*objp == 0) { 763 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 764 zfs_userquota_prop_prefixes[type]); 765 } 766 if (fuid_dirtied) 767 zfs_fuid_txhold(zfsvfs, tx); 768 err = dmu_tx_assign(tx, TXG_WAIT); 769 if (err) { 770 dmu_tx_abort(tx); 771 return (err); 772 } 773 774 mutex_enter(&zfsvfs->z_lock); 775 if (*objp == 0) { 776 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, 777 DMU_OT_NONE, 0, tx); 778 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 779 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); 780 } 781 mutex_exit(&zfsvfs->z_lock); 782 783 if (quota == 0) { 784 err = zap_remove(zfsvfs->z_os, *objp, buf, tx); 785 if (err == ENOENT) 786 err = 0; 787 } else { 788 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); 789 } 790 ASSERT(err == 0); 791 if (fuid_dirtied) 792 zfs_fuid_sync(zfsvfs, tx); 793 dmu_tx_commit(tx); 794 return (err); 795} 796 797boolean_t 798zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) 799{ 800 char buf[32]; 801 uint64_t used, quota, usedobj, quotaobj; 802 int err; 803 804 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; 805 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 806 807 if (quotaobj == 0 || zfsvfs->z_replay) 808 return (B_FALSE); 809 810 (void) sprintf(buf, "%llx", (longlong_t)fuid); 811 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); 812 if (err != 0) 813 return (B_FALSE); 814 815 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); 816 if (err != 0) 817 return (B_FALSE); 818 return (used >= quota); 819} 820 821boolean_t 822zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) 823{ 824 uint64_t fuid; 825 uint64_t quotaobj; 826 827 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 828 829 fuid = isgroup ? zp->z_gid : zp->z_uid; 830 831 if (quotaobj == 0 || zfsvfs->z_replay) 832 return (B_FALSE); 833 834 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); 835} 836 837int 838zfsvfs_create(const char *osname, zfsvfs_t **zfvp) 839{ 840 objset_t *os; 841 zfsvfs_t *zfsvfs; 842 uint64_t zval; 843 int i, error; 844 uint64_t sa_obj; 845 846 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 847 848 /* 849 * We claim to always be readonly so we can open snapshots; 850 * other ZPL code will prevent us from writing to snapshots. 851 */ 852 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 853 if (error) { 854 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 855 return (error); 856 } 857 858 /* 859 * Initialize the zfs-specific filesystem structure. 860 * Should probably make this a kmem cache, shuffle fields, 861 * and just bzero up to z_hold_mtx[]. 862 */ 863 zfsvfs->z_vfs = NULL; 864 zfsvfs->z_parent = zfsvfs; 865 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE; 866 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 867 zfsvfs->z_os = os; 868 869 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 870 if (error) { 871 goto out; 872 } else if (zfsvfs->z_version > 873 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 874 (void) printf("Can't mount a version %lld file system " 875 "on a version %lld pool\n. Pool must be upgraded to mount " 876 "this file system.", (u_longlong_t)zfsvfs->z_version, 877 (u_longlong_t)spa_version(dmu_objset_spa(os))); 878 error = ENOTSUP; 879 goto out; 880 } 881 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) 882 goto out; 883 zfsvfs->z_norm = (int)zval; 884 885 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) 886 goto out; 887 zfsvfs->z_utf8 = (zval != 0); 888 889 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) 890 goto out; 891 zfsvfs->z_case = (uint_t)zval; 892 893 /* 894 * Fold case on file systems that are always or sometimes case 895 * insensitive. 896 */ 897 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 898 zfsvfs->z_case == ZFS_CASE_MIXED) 899 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 900 901 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 902 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 903 904 if (zfsvfs->z_use_sa) { 905 /* should either have both of these objects or none */ 906 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 907 &sa_obj); 908 if (error) 909 return (error); 910 } else { 911 /* 912 * Pre SA versions file systems should never touch 913 * either the attribute registration or layout objects. 914 */ 915 sa_obj = 0; 916 } 917 918 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 919 &zfsvfs->z_attr_table); 920 if (error) 921 goto out; 922 923 if (zfsvfs->z_version >= ZPL_VERSION_SA) 924 sa_register_update_callback(os, zfs_sa_upgrade); 925 926 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 927 &zfsvfs->z_root); 928 if (error) 929 goto out; 930 ASSERT(zfsvfs->z_root != 0); 931 932 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 933 &zfsvfs->z_unlinkedobj); 934 if (error) 935 goto out; 936 937 error = zap_lookup(os, MASTER_NODE_OBJ, 938 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 939 8, 1, &zfsvfs->z_userquota_obj); 940 if (error && error != ENOENT) 941 goto out; 942 943 error = zap_lookup(os, MASTER_NODE_OBJ, 944 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 945 8, 1, &zfsvfs->z_groupquota_obj); 946 if (error && error != ENOENT) 947 goto out; 948 949 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 950 &zfsvfs->z_fuid_obj); 951 if (error && error != ENOENT) 952 goto out; 953 954 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 955 &zfsvfs->z_shares_dir); 956 if (error && error != ENOENT) 957 goto out; 958 959 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 960 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 961 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 962 offsetof(znode_t, z_link_node)); 963 rrw_init(&zfsvfs->z_teardown_lock); 964 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 965 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 966 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 967 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 968 969 *zfvp = zfsvfs; 970 return (0); 971 972out: 973 dmu_objset_disown(os, zfsvfs); 974 *zfvp = NULL; 975 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 976 return (error); 977} 978 979static int 980zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 981{ 982 int error; 983 984 error = zfs_register_callbacks(zfsvfs->z_vfs); 985 if (error) 986 return (error); 987 988 /* 989 * Set the objset user_ptr to track its zfsvfs. 990 */ 991 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 992 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 993 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 994 995 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 996 997 /* 998 * If we are not mounting (ie: online recv), then we don't 999 * have to worry about replaying the log as we blocked all 1000 * operations out since we closed the ZIL. 1001 */ 1002 if (mounting) { 1003 boolean_t readonly; 1004 1005 /* 1006 * During replay we remove the read only flag to 1007 * allow replays to succeed. 1008 */ 1009 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1010 if (readonly != 0) 1011 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1012 else 1013 zfs_unlinked_drain(zfsvfs); 1014 1015 /* 1016 * Parse and replay the intent log. 1017 * 1018 * Because of ziltest, this must be done after 1019 * zfs_unlinked_drain(). (Further note: ziltest 1020 * doesn't use readonly mounts, where 1021 * zfs_unlinked_drain() isn't called.) This is because 1022 * ziltest causes spa_sync() to think it's committed, 1023 * but actually it is not, so the intent log contains 1024 * many txg's worth of changes. 1025 * 1026 * In particular, if object N is in the unlinked set in 1027 * the last txg to actually sync, then it could be 1028 * actually freed in a later txg and then reallocated 1029 * in a yet later txg. This would write a "create 1030 * object N" record to the intent log. Normally, this 1031 * would be fine because the spa_sync() would have 1032 * written out the fact that object N is free, before 1033 * we could write the "create object N" intent log 1034 * record. 1035 * 1036 * But when we are in ziltest mode, we advance the "open 1037 * txg" without actually spa_sync()-ing the changes to 1038 * disk. So we would see that object N is still 1039 * allocated and in the unlinked set, and there is an 1040 * intent log record saying to allocate it. 1041 */ 1042 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1043 if (zil_replay_disable) { 1044 zil_destroy(zfsvfs->z_log, B_FALSE); 1045 } else { 1046 zfsvfs->z_replay = B_TRUE; 1047 zil_replay(zfsvfs->z_os, zfsvfs, 1048 zfs_replay_vector); 1049 zfsvfs->z_replay = B_FALSE; 1050 } 1051 } 1052 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1053 } 1054 1055 return (0); 1056} 1057 1058extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1059 1060void 1061zfsvfs_free(zfsvfs_t *zfsvfs) 1062{ 1063 int i; 1064 1065 /* 1066 * This is a barrier to prevent the filesystem from going away in 1067 * zfs_znode_move() until we can safely ensure that the filesystem is 1068 * not unmounted. We consider the filesystem valid before the barrier 1069 * and invalid after the barrier. 1070 */ 1071 rw_enter(&zfsvfs_lock, RW_READER); 1072 rw_exit(&zfsvfs_lock); 1073 1074 zfs_fuid_destroy(zfsvfs); 1075 1076 mutex_destroy(&zfsvfs->z_znodes_lock); 1077 mutex_destroy(&zfsvfs->z_lock); 1078 list_destroy(&zfsvfs->z_all_znodes); 1079 rrw_destroy(&zfsvfs->z_teardown_lock); 1080 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1081 rw_destroy(&zfsvfs->z_fuid_lock); 1082 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1083 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1084 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1085} 1086 1087static void 1088zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1089{ 1090 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1091 if (zfsvfs->z_vfs) { 1092 if (zfsvfs->z_use_fuids) { 1093 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1094 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1095 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1096 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1097 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1098 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1099 } else { 1100 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1101 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1102 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1103 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1104 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1105 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1106 } 1107 } 1108 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1109} 1110 1111static int 1112zfs_domount(vfs_t *vfsp, char *osname) 1113{ 1114 uint64_t recordsize, fsid_guid; 1115 int error = 0; 1116 zfsvfs_t *zfsvfs; 1117 vnode_t *vp; 1118 1119 ASSERT(vfsp); 1120 ASSERT(osname); 1121 1122 error = zfsvfs_create(osname, &zfsvfs); 1123 if (error) 1124 return (error); 1125 zfsvfs->z_vfs = vfsp; 1126 1127 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1128 NULL)) 1129 goto out; 1130 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1131 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1132 1133 vfsp->vfs_data = zfsvfs; 1134 vfsp->mnt_flag |= MNT_LOCAL; 1135 vfsp->mnt_kern_flag |= MNTK_MPSAFE; 1136 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1137 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1138 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1139 1140 /* 1141 * The fsid is 64 bits, composed of an 8-bit fs type, which 1142 * separates our fsid from any other filesystem types, and a 1143 * 56-bit objset unique ID. The objset unique ID is unique to 1144 * all objsets open on this system, provided by unique_create(). 1145 * The 8-bit fs type must be put in the low bits of fsid[1] 1146 * because that's where other Solaris filesystems put it. 1147 */ 1148 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1149 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1150 vfsp->vfs_fsid.val[0] = fsid_guid; 1151 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1152 vfsp->mnt_vfc->vfc_typenum & 0xFF; 1153 1154 /* 1155 * Set features for file system. 1156 */ 1157 zfs_set_fuid_feature(zfsvfs); 1158 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1159 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1160 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1161 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1162 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1163 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1164 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1165 } 1166 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1167 1168 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1169 uint64_t pval; 1170 1171 atime_changed_cb(zfsvfs, B_FALSE); 1172 readonly_changed_cb(zfsvfs, B_TRUE); 1173 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1174 goto out; 1175 xattr_changed_cb(zfsvfs, pval); 1176 zfsvfs->z_issnap = B_TRUE; 1177 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1178 1179 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1180 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1181 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1182 } else { 1183 error = zfsvfs_setup(zfsvfs, B_TRUE); 1184 } 1185 1186 vfs_mountedfrom(vfsp, osname); 1187 /* Grab extra reference. */ 1188 VERIFY(VFS_ROOT(vfsp, LK_EXCLUSIVE, &vp) == 0); 1189 VOP_UNLOCK(vp, 0); 1190 1191 if (!zfsvfs->z_issnap) 1192 zfsctl_create(zfsvfs); 1193out: 1194 if (error) { 1195 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1196 zfsvfs_free(zfsvfs); 1197 } else { 1198 atomic_add_32(&zfs_active_fs_count, 1); 1199 } 1200 1201 return (error); 1202} 1203 1204void 1205zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1206{ 1207 objset_t *os = zfsvfs->z_os; 1208 struct dsl_dataset *ds; 1209 1210 /* 1211 * Unregister properties. 1212 */ 1213 if (!dmu_objset_is_snapshot(os)) { 1214 ds = dmu_objset_ds(os); 1215 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, 1216 zfsvfs) == 0); 1217 1218 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, 1219 zfsvfs) == 0); 1220 1221 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, 1222 zfsvfs) == 0); 1223 1224 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, 1225 zfsvfs) == 0); 1226 1227 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, 1228 zfsvfs) == 0); 1229 1230 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, 1231 zfsvfs) == 0); 1232 1233 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, 1234 zfsvfs) == 0); 1235 1236 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, 1237 zfsvfs) == 0); 1238 1239 VERIFY(dsl_prop_unregister(ds, "aclinherit", 1240 acl_inherit_changed_cb, zfsvfs) == 0); 1241 1242 VERIFY(dsl_prop_unregister(ds, "vscan", 1243 vscan_changed_cb, zfsvfs) == 0); 1244 } 1245} 1246 1247#ifdef SECLABEL 1248/* 1249 * Convert a decimal digit string to a uint64_t integer. 1250 */ 1251static int 1252str_to_uint64(char *str, uint64_t *objnum) 1253{ 1254 uint64_t num = 0; 1255 1256 while (*str) { 1257 if (*str < '0' || *str > '9') 1258 return (EINVAL); 1259 1260 num = num*10 + *str++ - '0'; 1261 } 1262 1263 *objnum = num; 1264 return (0); 1265} 1266 1267/* 1268 * The boot path passed from the boot loader is in the form of 1269 * "rootpool-name/root-filesystem-object-number'. Convert this 1270 * string to a dataset name: "rootpool-name/root-filesystem-name". 1271 */ 1272static int 1273zfs_parse_bootfs(char *bpath, char *outpath) 1274{ 1275 char *slashp; 1276 uint64_t objnum; 1277 int error; 1278 1279 if (*bpath == 0 || *bpath == '/') 1280 return (EINVAL); 1281 1282 (void) strcpy(outpath, bpath); 1283 1284 slashp = strchr(bpath, '/'); 1285 1286 /* if no '/', just return the pool name */ 1287 if (slashp == NULL) { 1288 return (0); 1289 } 1290 1291 /* if not a number, just return the root dataset name */ 1292 if (str_to_uint64(slashp+1, &objnum)) { 1293 return (0); 1294 } 1295 1296 *slashp = '\0'; 1297 error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1298 *slashp = '/'; 1299 1300 return (error); 1301} 1302 1303/* 1304 * zfs_check_global_label: 1305 * Check that the hex label string is appropriate for the dataset 1306 * being mounted into the global_zone proper. 1307 * 1308 * Return an error if the hex label string is not default or 1309 * admin_low/admin_high. For admin_low labels, the corresponding 1310 * dataset must be readonly. 1311 */ 1312int 1313zfs_check_global_label(const char *dsname, const char *hexsl) 1314{ 1315 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1316 return (0); 1317 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1318 return (0); 1319 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1320 /* must be readonly */ 1321 uint64_t rdonly; 1322 1323 if (dsl_prop_get_integer(dsname, 1324 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1325 return (EACCES); 1326 return (rdonly ? 0 : EACCES); 1327 } 1328 return (EACCES); 1329} 1330 1331/* 1332 * zfs_mount_label_policy: 1333 * Determine whether the mount is allowed according to MAC check. 1334 * by comparing (where appropriate) label of the dataset against 1335 * the label of the zone being mounted into. If the dataset has 1336 * no label, create one. 1337 * 1338 * Returns: 1339 * 0 : access allowed 1340 * >0 : error code, such as EACCES 1341 */ 1342static int 1343zfs_mount_label_policy(vfs_t *vfsp, char *osname) 1344{ 1345 int error, retv; 1346 zone_t *mntzone = NULL; 1347 ts_label_t *mnt_tsl; 1348 bslabel_t *mnt_sl; 1349 bslabel_t ds_sl; 1350 char ds_hexsl[MAXNAMELEN]; 1351 1352 retv = EACCES; /* assume the worst */ 1353 1354 /* 1355 * Start by getting the dataset label if it exists. 1356 */ 1357 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1358 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1359 if (error) 1360 return (EACCES); 1361 1362 /* 1363 * If labeling is NOT enabled, then disallow the mount of datasets 1364 * which have a non-default label already. No other label checks 1365 * are needed. 1366 */ 1367 if (!is_system_labeled()) { 1368 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1369 return (0); 1370 return (EACCES); 1371 } 1372 1373 /* 1374 * Get the label of the mountpoint. If mounting into the global 1375 * zone (i.e. mountpoint is not within an active zone and the 1376 * zoned property is off), the label must be default or 1377 * admin_low/admin_high only; no other checks are needed. 1378 */ 1379 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1380 if (mntzone->zone_id == GLOBAL_ZONEID) { 1381 uint64_t zoned; 1382 1383 zone_rele(mntzone); 1384 1385 if (dsl_prop_get_integer(osname, 1386 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1387 return (EACCES); 1388 if (!zoned) 1389 return (zfs_check_global_label(osname, ds_hexsl)); 1390 else 1391 /* 1392 * This is the case of a zone dataset being mounted 1393 * initially, before the zone has been fully created; 1394 * allow this mount into global zone. 1395 */ 1396 return (0); 1397 } 1398 1399 mnt_tsl = mntzone->zone_slabel; 1400 ASSERT(mnt_tsl != NULL); 1401 label_hold(mnt_tsl); 1402 mnt_sl = label2bslabel(mnt_tsl); 1403 1404 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1405 /* 1406 * The dataset doesn't have a real label, so fabricate one. 1407 */ 1408 char *str = NULL; 1409 1410 if (l_to_str_internal(mnt_sl, &str) == 0 && 1411 dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1412 ZPROP_SRC_LOCAL, 1, strlen(str) + 1, str) == 0) 1413 retv = 0; 1414 if (str != NULL) 1415 kmem_free(str, strlen(str) + 1); 1416 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1417 /* 1418 * Now compare labels to complete the MAC check. If the 1419 * labels are equal then allow access. If the mountpoint 1420 * label dominates the dataset label, allow readonly access. 1421 * Otherwise, access is denied. 1422 */ 1423 if (blequal(mnt_sl, &ds_sl)) 1424 retv = 0; 1425 else if (bldominates(mnt_sl, &ds_sl)) { 1426 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1427 retv = 0; 1428 } 1429 } 1430 1431 label_rele(mnt_tsl); 1432 zone_rele(mntzone); 1433 return (retv); 1434} 1435#endif /* SECLABEL */ 1436 1437#ifdef OPENSOLARIS_MOUNTROOT 1438static int 1439zfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1440{ 1441 int error = 0; 1442 static int zfsrootdone = 0; 1443 zfsvfs_t *zfsvfs = NULL; 1444 znode_t *zp = NULL; 1445 vnode_t *vp = NULL; 1446 char *zfs_bootfs; 1447 char *zfs_devid; 1448 1449 ASSERT(vfsp); 1450 1451 /* 1452 * The filesystem that we mount as root is defined in the 1453 * boot property "zfs-bootfs" with a format of 1454 * "poolname/root-dataset-objnum". 1455 */ 1456 if (why == ROOT_INIT) { 1457 if (zfsrootdone++) 1458 return (EBUSY); 1459 /* 1460 * the process of doing a spa_load will require the 1461 * clock to be set before we could (for example) do 1462 * something better by looking at the timestamp on 1463 * an uberblock, so just set it to -1. 1464 */ 1465 clkset(-1); 1466 1467 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1468 cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1469 "bootfs name"); 1470 return (EINVAL); 1471 } 1472 zfs_devid = spa_get_bootprop("diskdevid"); 1473 error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1474 if (zfs_devid) 1475 spa_free_bootprop(zfs_devid); 1476 if (error) { 1477 spa_free_bootprop(zfs_bootfs); 1478 cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1479 error); 1480 return (error); 1481 } 1482 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1483 spa_free_bootprop(zfs_bootfs); 1484 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1485 error); 1486 return (error); 1487 } 1488 1489 spa_free_bootprop(zfs_bootfs); 1490 1491 if (error = vfs_lock(vfsp)) 1492 return (error); 1493 1494 if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1495 cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1496 goto out; 1497 } 1498 1499 zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1500 ASSERT(zfsvfs); 1501 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1502 cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1503 goto out; 1504 } 1505 1506 vp = ZTOV(zp); 1507 mutex_enter(&vp->v_lock); 1508 vp->v_flag |= VROOT; 1509 mutex_exit(&vp->v_lock); 1510 rootvp = vp; 1511 1512 /* 1513 * Leave rootvp held. The root file system is never unmounted. 1514 */ 1515 1516 vfs_add((struct vnode *)0, vfsp, 1517 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1518out: 1519 vfs_unlock(vfsp); 1520 return (error); 1521 } else if (why == ROOT_REMOUNT) { 1522 readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1523 vfsp->vfs_flag |= VFS_REMOUNT; 1524 1525 /* refresh mount options */ 1526 zfs_unregister_callbacks(vfsp->vfs_data); 1527 return (zfs_register_callbacks(vfsp)); 1528 1529 } else if (why == ROOT_UNMOUNT) { 1530 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1531 (void) zfs_sync(vfsp, 0, 0); 1532 return (0); 1533 } 1534 1535 /* 1536 * if "why" is equal to anything else other than ROOT_INIT, 1537 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1538 */ 1539 return (ENOTSUP); 1540} 1541#endif /* OPENSOLARIS_MOUNTROOT */ 1542 1543static int 1544getpoolname(const char *osname, char *poolname) 1545{ 1546 char *p; 1547 1548 p = strchr(osname, '/'); 1549 if (p == NULL) { 1550 if (strlen(osname) >= MAXNAMELEN) 1551 return (ENAMETOOLONG); 1552 (void) strcpy(poolname, osname); 1553 } else { 1554 if (p - osname >= MAXNAMELEN) 1555 return (ENAMETOOLONG); 1556 (void) strncpy(poolname, osname, p - osname); 1557 poolname[p - osname] = '\0'; 1558 } 1559 return (0); 1560} 1561 1562/*ARGSUSED*/ 1563static int 1564zfs_mount(vfs_t *vfsp) 1565{ 1566 kthread_t *td = curthread; 1567 vnode_t *mvp = vfsp->mnt_vnodecovered; 1568 cred_t *cr = td->td_ucred; 1569 char *osname; 1570 int error = 0; 1571 int canwrite; 1572 1573 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1574 return (EPERM); 1575 1576 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1577 return (EINVAL); 1578 1579 /* 1580 * If full-owner-access is enabled and delegated administration is 1581 * turned on, we must set nosuid. 1582 */ 1583 if (zfs_super_owner && 1584 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1585 secpolicy_fs_mount_clearopts(cr, vfsp); 1586 } 1587 1588 /* 1589 * Check for mount privilege? 1590 * 1591 * If we don't have privilege then see if 1592 * we have local permission to allow it 1593 */ 1594 error = secpolicy_fs_mount(cr, mvp, vfsp); 1595 if (error) { 1596 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1597 goto out; 1598 1599 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1600 vattr_t vattr; 1601 1602 /* 1603 * Make sure user is the owner of the mount point 1604 * or has sufficient privileges. 1605 */ 1606 1607 vattr.va_mask = AT_UID; 1608 1609 vn_lock(mvp, LK_SHARED | LK_RETRY); 1610 if (VOP_GETATTR(mvp, &vattr, cr)) { 1611 VOP_UNLOCK(mvp, 0); 1612 goto out; 1613 } 1614 1615 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1616 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1617 VOP_UNLOCK(mvp, 0); 1618 goto out; 1619 } 1620 VOP_UNLOCK(mvp, 0); 1621 } 1622 1623 secpolicy_fs_mount_clearopts(cr, vfsp); 1624 } 1625 1626 /* 1627 * Refuse to mount a filesystem if we are in a local zone and the 1628 * dataset is not visible. 1629 */ 1630 if (!INGLOBALZONE(curthread) && 1631 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1632 error = EPERM; 1633 goto out; 1634 } 1635 1636#ifdef SECLABEL 1637 error = zfs_mount_label_policy(vfsp, osname); 1638 if (error) 1639 goto out; 1640#endif 1641 1642 vfsp->vfs_flag |= MNT_NFS4ACLS; 1643 1644 /* 1645 * When doing a remount, we simply refresh our temporary properties 1646 * according to those options set in the current VFS options. 1647 */ 1648 if (vfsp->vfs_flag & MS_REMOUNT) { 1649 /* refresh mount options */ 1650 zfs_unregister_callbacks(vfsp->vfs_data); 1651 error = zfs_register_callbacks(vfsp); 1652 goto out; 1653 } 1654 1655 /* Initial root mount: try hard to import the requested root pool. */ 1656 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1657 (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1658 char pname[MAXNAMELEN]; 1659 spa_t *spa; 1660 int prefer_cache; 1661 1662 error = getpoolname(osname, pname); 1663 if (error) 1664 goto out; 1665 1666 prefer_cache = 1; 1667 TUNABLE_INT_FETCH("vfs.zfs.rootpool.prefer_cached_config", 1668 &prefer_cache); 1669 mutex_enter(&spa_namespace_lock); 1670 spa = spa_lookup(pname); 1671 mutex_exit(&spa_namespace_lock); 1672 if (!prefer_cache || spa == NULL) { 1673 error = spa_import_rootpool(pname); 1674 if (error) 1675 goto out; 1676 } 1677 } 1678 DROP_GIANT(); 1679 error = zfs_domount(vfsp, osname); 1680 PICKUP_GIANT(); 1681 1682#ifdef sun 1683 /* 1684 * Add an extra VFS_HOLD on our parent vfs so that it can't 1685 * disappear due to a forced unmount. 1686 */ 1687 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1688 VFS_HOLD(mvp->v_vfsp); 1689#endif /* sun */ 1690 1691out: 1692 return (error); 1693} 1694 1695static int 1696zfs_statfs(vfs_t *vfsp, struct statfs *statp) 1697{ 1698 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1699 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1700 1701 statp->f_version = STATFS_VERSION; 1702 1703 ZFS_ENTER(zfsvfs); 1704 1705 dmu_objset_space(zfsvfs->z_os, 1706 &refdbytes, &availbytes, &usedobjs, &availobjs); 1707 1708 /* 1709 * The underlying storage pool actually uses multiple block sizes. 1710 * We report the fragsize as the smallest block size we support, 1711 * and we report our blocksize as the filesystem's maximum blocksize. 1712 */ 1713 statp->f_bsize = SPA_MINBLOCKSIZE; 1714 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1715 1716 /* 1717 * The following report "total" blocks of various kinds in the 1718 * file system, but reported in terms of f_frsize - the 1719 * "fragment" size. 1720 */ 1721 1722 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1723 statp->f_bfree = availbytes / statp->f_bsize; 1724 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1725 1726 /* 1727 * statvfs() should really be called statufs(), because it assumes 1728 * static metadata. ZFS doesn't preallocate files, so the best 1729 * we can do is report the max that could possibly fit in f_files, 1730 * and that minus the number actually used in f_ffree. 1731 * For f_ffree, report the smaller of the number of object available 1732 * and the number of blocks (each object will take at least a block). 1733 */ 1734 statp->f_ffree = MIN(availobjs, statp->f_bfree); 1735 statp->f_files = statp->f_ffree + usedobjs; 1736 1737 /* 1738 * We're a zfs filesystem. 1739 */ 1740 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1741 1742 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1743 sizeof(statp->f_mntfromname)); 1744 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1745 sizeof(statp->f_mntonname)); 1746 1747 statp->f_namemax = ZFS_MAXNAMELEN; 1748 1749 ZFS_EXIT(zfsvfs); 1750 return (0); 1751} 1752 1753int 1754zfs_vnode_lock(vnode_t *vp, int flags) 1755{ 1756 int error; 1757 1758 ASSERT(vp != NULL); 1759 1760 error = vn_lock(vp, flags); 1761 return (error); 1762} 1763 1764static int 1765zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1766{ 1767 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1768 znode_t *rootzp; 1769 int error; 1770 1771 ZFS_ENTER_NOERROR(zfsvfs); 1772 1773 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1774 if (error == 0) 1775 *vpp = ZTOV(rootzp); 1776 1777 ZFS_EXIT(zfsvfs); 1778 1779 if (error == 0) { 1780 error = zfs_vnode_lock(*vpp, flags); 1781 if (error == 0) 1782 (*vpp)->v_vflag |= VV_ROOT; 1783 } 1784 if (error != 0) 1785 *vpp = NULL; 1786 1787 return (error); 1788} 1789 1790/* 1791 * Teardown the zfsvfs::z_os. 1792 * 1793 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1794 * and 'z_teardown_inactive_lock' held. 1795 */ 1796static int 1797zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1798{ 1799 znode_t *zp; 1800 1801 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1802 1803 if (!unmounting) { 1804 /* 1805 * We purge the parent filesystem's vfsp as the parent 1806 * filesystem and all of its snapshots have their vnode's 1807 * v_vfsp set to the parent's filesystem's vfsp. Note, 1808 * 'z_parent' is self referential for non-snapshots. 1809 */ 1810 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1811#ifdef FREEBSD_NAMECACHE 1812 cache_purgevfs(zfsvfs->z_parent->z_vfs); 1813#endif 1814 } 1815 1816 /* 1817 * Close the zil. NB: Can't close the zil while zfs_inactive 1818 * threads are blocked as zil_close can call zfs_inactive. 1819 */ 1820 if (zfsvfs->z_log) { 1821 zil_close(zfsvfs->z_log); 1822 zfsvfs->z_log = NULL; 1823 } 1824 1825 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1826 1827 /* 1828 * If we are not unmounting (ie: online recv) and someone already 1829 * unmounted this file system while we were doing the switcheroo, 1830 * or a reopen of z_os failed then just bail out now. 1831 */ 1832 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1833 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1834 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1835 return (EIO); 1836 } 1837 1838 /* 1839 * At this point there are no vops active, and any new vops will 1840 * fail with EIO since we have z_teardown_lock for writer (only 1841 * relavent for forced unmount). 1842 * 1843 * Release all holds on dbufs. 1844 */ 1845 mutex_enter(&zfsvfs->z_znodes_lock); 1846 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1847 zp = list_next(&zfsvfs->z_all_znodes, zp)) 1848 if (zp->z_sa_hdl) { 1849 ASSERT(ZTOV(zp)->v_count >= 0); 1850 zfs_znode_dmu_fini(zp); 1851 } 1852 mutex_exit(&zfsvfs->z_znodes_lock); 1853 1854 /* 1855 * If we are unmounting, set the unmounted flag and let new vops 1856 * unblock. zfs_inactive will have the unmounted behavior, and all 1857 * other vops will fail with EIO. 1858 */ 1859 if (unmounting) { 1860 zfsvfs->z_unmounted = B_TRUE; 1861 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1862 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1863 1864#ifdef __FreeBSD__ 1865 /* 1866 * Some znodes might not be fully reclaimed, wait for them. 1867 */ 1868 mutex_enter(&zfsvfs->z_znodes_lock); 1869 while (list_head(&zfsvfs->z_all_znodes) != NULL) { 1870 msleep(zfsvfs, &zfsvfs->z_znodes_lock, 0, 1871 "zteardown", 0); 1872 } 1873 mutex_exit(&zfsvfs->z_znodes_lock); 1874#endif 1875 } 1876 1877 /* 1878 * z_os will be NULL if there was an error in attempting to reopen 1879 * zfsvfs, so just return as the properties had already been 1880 * unregistered and cached data had been evicted before. 1881 */ 1882 if (zfsvfs->z_os == NULL) 1883 return (0); 1884 1885 /* 1886 * Unregister properties. 1887 */ 1888 zfs_unregister_callbacks(zfsvfs); 1889 1890 /* 1891 * Evict cached data 1892 */ 1893 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1894 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1895 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1896 (void) dmu_objset_evict_dbufs(zfsvfs->z_os); 1897 1898 return (0); 1899} 1900 1901/*ARGSUSED*/ 1902static int 1903zfs_umount(vfs_t *vfsp, int fflag) 1904{ 1905 kthread_t *td = curthread; 1906 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1907 objset_t *os; 1908 cred_t *cr = td->td_ucred; 1909 int ret; 1910 1911 ret = secpolicy_fs_unmount(cr, vfsp); 1912 if (ret) { 1913 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1914 ZFS_DELEG_PERM_MOUNT, cr)) 1915 return (ret); 1916 } 1917 1918 /* 1919 * We purge the parent filesystem's vfsp as the parent filesystem 1920 * and all of its snapshots have their vnode's v_vfsp set to the 1921 * parent's filesystem's vfsp. Note, 'z_parent' is self 1922 * referential for non-snapshots. 1923 */ 1924 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1925 1926 /* 1927 * Unmount any snapshots mounted under .zfs before unmounting the 1928 * dataset itself. 1929 */ 1930 if (zfsvfs->z_ctldir != NULL) { 1931 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1932 return (ret); 1933 ret = vflush(vfsp, 0, 0, td); 1934 ASSERT(ret == EBUSY); 1935 if (!(fflag & MS_FORCE)) { 1936 if (zfsvfs->z_ctldir->v_count > 1) 1937 return (EBUSY); 1938 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1939 } 1940 zfsctl_destroy(zfsvfs); 1941 ASSERT(zfsvfs->z_ctldir == NULL); 1942 } 1943 1944 if (fflag & MS_FORCE) { 1945 /* 1946 * Mark file system as unmounted before calling 1947 * vflush(FORCECLOSE). This way we ensure no future vnops 1948 * will be called and risk operating on DOOMED vnodes. 1949 */ 1950 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1951 zfsvfs->z_unmounted = B_TRUE; 1952 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1953 } 1954 1955 /* 1956 * Flush all the files. 1957 */ 1958 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1959 if (ret != 0) { 1960 if (!zfsvfs->z_issnap) { 1961 zfsctl_create(zfsvfs); 1962 ASSERT(zfsvfs->z_ctldir != NULL); 1963 } 1964 return (ret); 1965 } 1966 1967 if (!(fflag & MS_FORCE)) { 1968 /* 1969 * Check the number of active vnodes in the file system. 1970 * Our count is maintained in the vfs structure, but the 1971 * number is off by 1 to indicate a hold on the vfs 1972 * structure itself. 1973 * 1974 * The '.zfs' directory maintains a reference of its 1975 * own, and any active references underneath are 1976 * reflected in the vnode count. 1977 */ 1978 if (zfsvfs->z_ctldir == NULL) { 1979 if (vfsp->vfs_count > 1) 1980 return (EBUSY); 1981 } else { 1982 if (vfsp->vfs_count > 2 || 1983 zfsvfs->z_ctldir->v_count > 1) 1984 return (EBUSY); 1985 } 1986 } 1987 1988 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1989 os = zfsvfs->z_os; 1990 1991 /* 1992 * z_os will be NULL if there was an error in 1993 * attempting to reopen zfsvfs. 1994 */ 1995 if (os != NULL) { 1996 /* 1997 * Unset the objset user_ptr. 1998 */ 1999 mutex_enter(&os->os_user_ptr_lock); 2000 dmu_objset_set_user(os, NULL); 2001 mutex_exit(&os->os_user_ptr_lock); 2002 2003 /* 2004 * Finally release the objset 2005 */ 2006 dmu_objset_disown(os, zfsvfs); 2007 } 2008 2009 /* 2010 * We can now safely destroy the '.zfs' directory node. 2011 */ 2012 if (zfsvfs->z_ctldir != NULL) 2013 zfsctl_destroy(zfsvfs); 2014 if (zfsvfs->z_issnap) { 2015 vnode_t *svp = vfsp->mnt_vnodecovered; 2016 2017 if (svp->v_count >= 2) 2018 VN_RELE(svp); 2019 } 2020 zfs_freevfs(vfsp); 2021 2022 return (0); 2023} 2024 2025static int 2026zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2027{ 2028 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2029 znode_t *zp; 2030 int err; 2031 2032 /* 2033 * zfs_zget() can't operate on virtual entries like .zfs/ or 2034 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2035 * This will make NFS to switch to LOOKUP instead of using VGET. 2036 */ 2037 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR) 2038 return (EOPNOTSUPP); 2039 2040 ZFS_ENTER(zfsvfs); 2041 err = zfs_zget(zfsvfs, ino, &zp); 2042 if (err == 0 && zp->z_unlinked) { 2043 VN_RELE(ZTOV(zp)); 2044 err = EINVAL; 2045 } 2046 if (err == 0) 2047 *vpp = ZTOV(zp); 2048 ZFS_EXIT(zfsvfs); 2049 if (err == 0) 2050 err = zfs_vnode_lock(*vpp, flags); 2051 if (err != 0) 2052 *vpp = NULL; 2053 return (err); 2054} 2055 2056static int 2057zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2058 struct ucred **credanonp, int *numsecflavors, int **secflavors) 2059{ 2060 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2061 2062 /* 2063 * If this is regular file system vfsp is the same as 2064 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2065 * zfsvfs->z_parent->z_vfs represents parent file system 2066 * which we have to use here, because only this file system 2067 * has mnt_export configured. 2068 */ 2069 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2070 credanonp, numsecflavors, secflavors)); 2071} 2072 2073CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2074CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2075 2076static int 2077zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2078{ 2079 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2080 znode_t *zp; 2081 uint64_t object = 0; 2082 uint64_t fid_gen = 0; 2083 uint64_t gen_mask; 2084 uint64_t zp_gen; 2085 int i, err; 2086 2087 *vpp = NULL; 2088 2089 ZFS_ENTER(zfsvfs); 2090 2091 /* 2092 * On FreeBSD we can get snapshot's mount point or its parent file 2093 * system mount point depending if snapshot is already mounted or not. 2094 */ 2095 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2096 zfid_long_t *zlfid = (zfid_long_t *)fidp; 2097 uint64_t objsetid = 0; 2098 uint64_t setgen = 0; 2099 2100 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2101 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2102 2103 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2104 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2105 2106 ZFS_EXIT(zfsvfs); 2107 2108 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2109 if (err) 2110 return (EINVAL); 2111 ZFS_ENTER(zfsvfs); 2112 } 2113 2114 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2115 zfid_short_t *zfid = (zfid_short_t *)fidp; 2116 2117 for (i = 0; i < sizeof (zfid->zf_object); i++) 2118 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2119 2120 for (i = 0; i < sizeof (zfid->zf_gen); i++) 2121 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2122 } else { 2123 ZFS_EXIT(zfsvfs); 2124 return (EINVAL); 2125 } 2126 2127 /* A zero fid_gen means we are in the .zfs control directories */ 2128 if (fid_gen == 0 && 2129 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 2130 *vpp = zfsvfs->z_ctldir; 2131 ASSERT(*vpp != NULL); 2132 if (object == ZFSCTL_INO_SNAPDIR) { 2133 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2134 0, NULL, NULL, NULL, NULL, NULL) == 0); 2135 } else { 2136 VN_HOLD(*vpp); 2137 } 2138 ZFS_EXIT(zfsvfs); 2139 err = zfs_vnode_lock(*vpp, flags | LK_RETRY); 2140 if (err != 0) 2141 *vpp = NULL; 2142 return (err); 2143 } 2144 2145 gen_mask = -1ULL >> (64 - 8 * i); 2146 2147 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2148 if (err = zfs_zget(zfsvfs, object, &zp)) { 2149 ZFS_EXIT(zfsvfs); 2150 return (err); 2151 } 2152 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2153 sizeof (uint64_t)); 2154 zp_gen = zp_gen & gen_mask; 2155 if (zp_gen == 0) 2156 zp_gen = 1; 2157 if (zp->z_unlinked || zp_gen != fid_gen) { 2158 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2159 VN_RELE(ZTOV(zp)); 2160 ZFS_EXIT(zfsvfs); 2161 return (EINVAL); 2162 } 2163 2164 *vpp = ZTOV(zp); 2165 ZFS_EXIT(zfsvfs); 2166 err = zfs_vnode_lock(*vpp, flags | LK_RETRY); 2167 if (err == 0) 2168 vnode_create_vobject(*vpp, zp->z_size, curthread); 2169 else 2170 *vpp = NULL; 2171 return (err); 2172} 2173 2174/* 2175 * Block out VOPs and close zfsvfs_t::z_os 2176 * 2177 * Note, if successful, then we return with the 'z_teardown_lock' and 2178 * 'z_teardown_inactive_lock' write held. 2179 */ 2180int 2181zfs_suspend_fs(zfsvfs_t *zfsvfs) 2182{ 2183 int error; 2184 2185 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2186 return (error); 2187 dmu_objset_disown(zfsvfs->z_os, zfsvfs); 2188 2189 return (0); 2190} 2191 2192/* 2193 * Reopen zfsvfs_t::z_os and release VOPs. 2194 */ 2195int 2196zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2197{ 2198 int err; 2199 2200 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2201 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2202 2203 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs, 2204 &zfsvfs->z_os); 2205 if (err) { 2206 zfsvfs->z_os = NULL; 2207 } else { 2208 znode_t *zp; 2209 uint64_t sa_obj = 0; 2210 2211 /* 2212 * Make sure version hasn't changed 2213 */ 2214 2215 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, 2216 &zfsvfs->z_version); 2217 2218 if (err) 2219 goto bail; 2220 2221 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, 2222 ZFS_SA_ATTRS, 8, 1, &sa_obj); 2223 2224 if (err && zfsvfs->z_version >= ZPL_VERSION_SA) 2225 goto bail; 2226 2227 if ((err = sa_setup(zfsvfs->z_os, sa_obj, 2228 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) 2229 goto bail; 2230 2231 if (zfsvfs->z_version >= ZPL_VERSION_SA) 2232 sa_register_update_callback(zfsvfs->z_os, 2233 zfs_sa_upgrade); 2234 2235 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2236 2237 zfs_set_fuid_feature(zfsvfs); 2238 2239 /* 2240 * Attempt to re-establish all the active znodes with 2241 * their dbufs. If a zfs_rezget() fails, then we'll let 2242 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2243 * when they try to use their znode. 2244 */ 2245 mutex_enter(&zfsvfs->z_znodes_lock); 2246 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2247 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2248 (void) zfs_rezget(zp); 2249 } 2250 mutex_exit(&zfsvfs->z_znodes_lock); 2251 } 2252 2253bail: 2254 /* release the VOPs */ 2255 rw_exit(&zfsvfs->z_teardown_inactive_lock); 2256 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 2257 2258 if (err) { 2259 /* 2260 * Since we couldn't reopen zfsvfs::z_os, or 2261 * setup the sa framework force unmount this file system. 2262 */ 2263 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) 2264 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2265 } 2266 return (err); 2267} 2268 2269static void 2270zfs_freevfs(vfs_t *vfsp) 2271{ 2272 zfsvfs_t *zfsvfs = vfsp->vfs_data; 2273 2274#ifdef sun 2275 /* 2276 * If this is a snapshot, we have an extra VFS_HOLD on our parent 2277 * from zfs_mount(). Release it here. If we came through 2278 * zfs_mountroot() instead, we didn't grab an extra hold, so 2279 * skip the VFS_RELE for rootvfs. 2280 */ 2281 if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2282 VFS_RELE(zfsvfs->z_parent->z_vfs); 2283#endif /* sun */ 2284 2285 zfsvfs_free(zfsvfs); 2286 2287 atomic_add_32(&zfs_active_fs_count, -1); 2288} 2289 2290#ifdef __i386__ 2291static int desiredvnodes_backup; 2292#endif 2293 2294static void 2295zfs_vnodes_adjust(void) 2296{ 2297#ifdef __i386__ 2298 int newdesiredvnodes; 2299 2300 desiredvnodes_backup = desiredvnodes; 2301 2302 /* 2303 * We calculate newdesiredvnodes the same way it is done in 2304 * vntblinit(). If it is equal to desiredvnodes, it means that 2305 * it wasn't tuned by the administrator and we can tune it down. 2306 */ 2307 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2308 vm_kmem_size / (5 * (sizeof(struct vm_object) + 2309 sizeof(struct vnode)))); 2310 if (newdesiredvnodes == desiredvnodes) 2311 desiredvnodes = (3 * newdesiredvnodes) / 4; 2312#endif 2313} 2314 2315static void 2316zfs_vnodes_adjust_back(void) 2317{ 2318 2319#ifdef __i386__ 2320 desiredvnodes = desiredvnodes_backup; 2321#endif 2322} 2323 2324void 2325zfs_init(void) 2326{ 2327 2328 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2329 2330 /* 2331 * Initialize .zfs directory structures 2332 */ 2333 zfsctl_init(); 2334 2335 /* 2336 * Initialize znode cache, vnode ops, etc... 2337 */ 2338 zfs_znode_init(); 2339 2340 /* 2341 * Reduce number of vnodes. Originally number of vnodes is calculated 2342 * with UFS inode in mind. We reduce it here, because it's too big for 2343 * ZFS/i386. 2344 */ 2345 zfs_vnodes_adjust(); 2346 2347 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2348} 2349 2350void 2351zfs_fini(void) 2352{ 2353 zfsctl_fini(); 2354 zfs_znode_fini(); 2355 zfs_vnodes_adjust_back(); 2356} 2357 2358int 2359zfs_busy(void) 2360{ 2361 return (zfs_active_fs_count != 0); 2362} 2363 2364int 2365zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2366{ 2367 int error; 2368 objset_t *os = zfsvfs->z_os; 2369 dmu_tx_t *tx; 2370 2371 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2372 return (EINVAL); 2373 2374 if (newvers < zfsvfs->z_version) 2375 return (EINVAL); 2376 2377 if (zfs_spa_version_map(newvers) > 2378 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2379 return (ENOTSUP); 2380 2381 tx = dmu_tx_create(os); 2382 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2383 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2384 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2385 ZFS_SA_ATTRS); 2386 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2387 } 2388 error = dmu_tx_assign(tx, TXG_WAIT); 2389 if (error) { 2390 dmu_tx_abort(tx); 2391 return (error); 2392 } 2393 2394 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2395 8, 1, &newvers, tx); 2396 2397 if (error) { 2398 dmu_tx_commit(tx); 2399 return (error); 2400 } 2401 2402 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2403 uint64_t sa_obj; 2404 2405 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2406 SPA_VERSION_SA); 2407 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2408 DMU_OT_NONE, 0, tx); 2409 2410 error = zap_add(os, MASTER_NODE_OBJ, 2411 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2412 ASSERT0(error); 2413 2414 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2415 sa_register_update_callback(os, zfs_sa_upgrade); 2416 } 2417 2418 spa_history_log_internal(LOG_DS_UPGRADE, 2419 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu", 2420 zfsvfs->z_version, newvers, dmu_objset_id(os)); 2421 2422 dmu_tx_commit(tx); 2423 2424 zfsvfs->z_version = newvers; 2425 2426 zfs_set_fuid_feature(zfsvfs); 2427 2428 return (0); 2429} 2430 2431/* 2432 * Read a property stored within the master node. 2433 */ 2434int 2435zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2436{ 2437 const char *pname; 2438 int error = ENOENT; 2439 2440 /* 2441 * Look up the file system's value for the property. For the 2442 * version property, we look up a slightly different string. 2443 */ 2444 if (prop == ZFS_PROP_VERSION) 2445 pname = ZPL_VERSION_STR; 2446 else 2447 pname = zfs_prop_to_name(prop); 2448 2449 if (os != NULL) 2450 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2451 2452 if (error == ENOENT) { 2453 /* No value set, use the default value */ 2454 switch (prop) { 2455 case ZFS_PROP_VERSION: 2456 *value = ZPL_VERSION; 2457 break; 2458 case ZFS_PROP_NORMALIZE: 2459 case ZFS_PROP_UTF8ONLY: 2460 *value = 0; 2461 break; 2462 case ZFS_PROP_CASE: 2463 *value = ZFS_CASE_SENSITIVE; 2464 break; 2465 default: 2466 return (error); 2467 } 2468 error = 0; 2469 } 2470 return (error); 2471} 2472 2473#ifdef _KERNEL 2474void 2475zfsvfs_update_fromname(const char *oldname, const char *newname) 2476{ 2477 char tmpbuf[MAXPATHLEN]; 2478 struct mount *mp; 2479 char *fromname; 2480 size_t oldlen; 2481 2482 oldlen = strlen(oldname); 2483 2484 mtx_lock(&mountlist_mtx); 2485 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2486 fromname = mp->mnt_stat.f_mntfromname; 2487 if (strcmp(fromname, oldname) == 0) { 2488 (void)strlcpy(fromname, newname, 2489 sizeof(mp->mnt_stat.f_mntfromname)); 2490 continue; 2491 } 2492 if (strncmp(fromname, oldname, oldlen) == 0 && 2493 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2494 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2495 newname, fromname + oldlen); 2496 (void)strlcpy(fromname, tmpbuf, 2497 sizeof(mp->mnt_stat.f_mntfromname)); 2498 continue; 2499 } 2500 } 2501 mtx_unlock(&mountlist_mtx); 2502} 2503#endif 2504