zfs_vfsops.c revision 197351
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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26#include <sys/types.h> 27#include <sys/param.h> 28#include <sys/systm.h> 29#include <sys/kernel.h> 30#include <sys/sysmacros.h> 31#include <sys/kmem.h> 32#include <sys/acl.h> 33#include <sys/vnode.h> 34#include <sys/vfs.h> 35#include <sys/mntent.h> 36#include <sys/mount.h> 37#include <sys/cmn_err.h> 38#include <sys/zfs_znode.h> 39#include <sys/zfs_dir.h> 40#include <sys/zil.h> 41#include <sys/fs/zfs.h> 42#include <sys/dmu.h> 43#include <sys/dsl_prop.h> 44#include <sys/dsl_dataset.h> 45#include <sys/dsl_deleg.h> 46#include <sys/spa.h> 47#include <sys/zap.h> 48#include <sys/varargs.h> 49#include <sys/policy.h> 50#include <sys/atomic.h> 51#include <sys/zfs_ioctl.h> 52#include <sys/zfs_ctldir.h> 53#include <sys/zfs_fuid.h> 54#include <sys/sunddi.h> 55#include <sys/dnlc.h> 56#include <sys/dmu_objset.h> 57#include <sys/spa_boot.h> 58#include <sys/vdev_impl.h> /* VDEV_BOOT_VERSION */ 59 60struct mtx zfs_debug_mtx; 61MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF); 62 63SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system"); 64 65int zfs_super_owner = 0; 66SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0, 67 "File system owner can perform privileged operation on his file systems"); 68 69int zfs_debug_level = 0; 70TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level); 71SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0, 72 "Debug level"); 73 74SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions"); 75static int zfs_version_acl = ZFS_ACL_VERSION; 76SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0, 77 "ZFS_ACL_VERSION"); 78static int zfs_version_dmu_backup_header = DMU_BACKUP_HEADER_VERSION; 79SYSCTL_INT(_vfs_zfs_version, OID_AUTO, dmu_backup_header, CTLFLAG_RD, 80 &zfs_version_dmu_backup_header, 0, "DMU_BACKUP_HEADER_VERSION"); 81static int zfs_version_dmu_backup_stream = DMU_BACKUP_STREAM_VERSION; 82SYSCTL_INT(_vfs_zfs_version, OID_AUTO, dmu_backup_stream, CTLFLAG_RD, 83 &zfs_version_dmu_backup_stream, 0, "DMU_BACKUP_STREAM_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_vdev_boot = VDEV_BOOT_VERSION; 88SYSCTL_INT(_vfs_zfs_version, OID_AUTO, vdev_boot, CTLFLAG_RD, 89 &zfs_version_vdev_boot, 0, "VDEV_BOOT_VERSION"); 90static int zfs_version_zpl = ZPL_VERSION; 91SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0, 92 "ZPL_VERSION"); 93 94static int zfs_mount(vfs_t *vfsp); 95static int zfs_umount(vfs_t *vfsp, int fflag); 96static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp); 97static int zfs_statfs(vfs_t *vfsp, struct statfs *statp); 98static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp); 99static int zfs_sync(vfs_t *vfsp, int waitfor); 100static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 101 struct ucred **credanonp, int *numsecflavors, int **secflavors); 102static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, vnode_t **vpp); 103static void zfs_objset_close(zfsvfs_t *zfsvfs); 104static void zfs_freevfs(vfs_t *vfsp); 105 106static struct vfsops zfs_vfsops = { 107 .vfs_mount = zfs_mount, 108 .vfs_unmount = zfs_umount, 109 .vfs_root = zfs_root, 110 .vfs_statfs = zfs_statfs, 111 .vfs_vget = zfs_vget, 112 .vfs_sync = zfs_sync, 113 .vfs_checkexp = zfs_checkexp, 114 .vfs_fhtovp = zfs_fhtovp, 115}; 116 117VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN); 118 119/* 120 * We need to keep a count of active fs's. 121 * This is necessary to prevent our module 122 * from being unloaded after a umount -f 123 */ 124static uint32_t zfs_active_fs_count = 0; 125 126/*ARGSUSED*/ 127static int 128zfs_sync(vfs_t *vfsp, int waitfor) 129{ 130 131 /* 132 * Data integrity is job one. We don't want a compromised kernel 133 * writing to the storage pool, so we never sync during panic. 134 */ 135 if (panicstr) 136 return (0); 137 138 if (vfsp != NULL) { 139 /* 140 * Sync a specific filesystem. 141 */ 142 zfsvfs_t *zfsvfs = vfsp->vfs_data; 143 int error; 144 145 error = vfs_stdsync(vfsp, waitfor); 146 if (error != 0) 147 return (error); 148 149 ZFS_ENTER(zfsvfs); 150 if (zfsvfs->z_log != NULL) 151 zil_commit(zfsvfs->z_log, UINT64_MAX, 0); 152 else 153 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 154 ZFS_EXIT(zfsvfs); 155 } else { 156 /* 157 * Sync all ZFS filesystems. This is what happens when you 158 * run sync(1M). Unlike other filesystems, ZFS honors the 159 * request by waiting for all pools to commit all dirty data. 160 */ 161 spa_sync_allpools(); 162 } 163 164 return (0); 165} 166 167static void 168atime_changed_cb(void *arg, uint64_t newval) 169{ 170 zfsvfs_t *zfsvfs = arg; 171 172 if (newval == TRUE) { 173 zfsvfs->z_atime = TRUE; 174 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; 175 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); 176 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); 177 } else { 178 zfsvfs->z_atime = FALSE; 179 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; 180 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); 181 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); 182 } 183} 184 185static void 186xattr_changed_cb(void *arg, uint64_t newval) 187{ 188 zfsvfs_t *zfsvfs = arg; 189 190 if (newval == TRUE) { 191 /* XXX locking on vfs_flag? */ 192#ifdef TODO 193 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; 194#endif 195 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); 196 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); 197 } else { 198 /* XXX locking on vfs_flag? */ 199#ifdef TODO 200 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR; 201#endif 202 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR); 203 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); 204 } 205} 206 207static void 208blksz_changed_cb(void *arg, uint64_t newval) 209{ 210 zfsvfs_t *zfsvfs = arg; 211 212 if (newval < SPA_MINBLOCKSIZE || 213 newval > SPA_MAXBLOCKSIZE || !ISP2(newval)) 214 newval = SPA_MAXBLOCKSIZE; 215 216 zfsvfs->z_max_blksz = newval; 217 zfsvfs->z_vfs->vfs_bsize = newval; 218} 219 220static void 221readonly_changed_cb(void *arg, uint64_t newval) 222{ 223 zfsvfs_t *zfsvfs = arg; 224 225 if (newval) { 226 /* XXX locking on vfs_flag? */ 227 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; 228 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); 229 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); 230 } else { 231 /* XXX locking on vfs_flag? */ 232 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 233 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); 234 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); 235 } 236} 237 238static void 239setuid_changed_cb(void *arg, uint64_t newval) 240{ 241 zfsvfs_t *zfsvfs = arg; 242 243 if (newval == FALSE) { 244 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; 245 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); 246 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); 247 } else { 248 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; 249 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); 250 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); 251 } 252} 253 254static void 255exec_changed_cb(void *arg, uint64_t newval) 256{ 257 zfsvfs_t *zfsvfs = arg; 258 259 if (newval == FALSE) { 260 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; 261 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); 262 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); 263 } else { 264 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; 265 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); 266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); 267 } 268} 269 270/* 271 * The nbmand mount option can be changed at mount time. 272 * We can't allow it to be toggled on live file systems or incorrect 273 * behavior may be seen from cifs clients 274 * 275 * This property isn't registered via dsl_prop_register(), but this callback 276 * will be called when a file system is first mounted 277 */ 278static void 279nbmand_changed_cb(void *arg, uint64_t newval) 280{ 281 zfsvfs_t *zfsvfs = arg; 282 if (newval == FALSE) { 283 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); 284 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); 285 } else { 286 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); 287 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); 288 } 289} 290 291static void 292snapdir_changed_cb(void *arg, uint64_t newval) 293{ 294 zfsvfs_t *zfsvfs = arg; 295 296 zfsvfs->z_show_ctldir = newval; 297} 298 299static void 300vscan_changed_cb(void *arg, uint64_t newval) 301{ 302 zfsvfs_t *zfsvfs = arg; 303 304 zfsvfs->z_vscan = newval; 305} 306 307static void 308acl_mode_changed_cb(void *arg, uint64_t newval) 309{ 310 zfsvfs_t *zfsvfs = arg; 311 312 zfsvfs->z_acl_mode = newval; 313} 314 315static void 316acl_inherit_changed_cb(void *arg, uint64_t newval) 317{ 318 zfsvfs_t *zfsvfs = arg; 319 320 zfsvfs->z_acl_inherit = newval; 321} 322 323static int 324zfs_register_callbacks(vfs_t *vfsp) 325{ 326 struct dsl_dataset *ds = NULL; 327 objset_t *os = NULL; 328 zfsvfs_t *zfsvfs = NULL; 329 uint64_t nbmand; 330 int readonly, do_readonly = FALSE; 331 int setuid, do_setuid = FALSE; 332 int exec, do_exec = FALSE; 333 int xattr, do_xattr = FALSE; 334 int atime, do_atime = FALSE; 335 int error = 0; 336 337 ASSERT(vfsp); 338 zfsvfs = vfsp->vfs_data; 339 ASSERT(zfsvfs); 340 os = zfsvfs->z_os; 341 342 /* 343 * This function can be called for a snapshot when we update snapshot's 344 * mount point, which isn't really supported. 345 */ 346 if (dmu_objset_is_snapshot(os)) 347 return (EOPNOTSUPP); 348 349 /* 350 * The act of registering our callbacks will destroy any mount 351 * options we may have. In order to enable temporary overrides 352 * of mount options, we stash away the current values and 353 * restore them after we register the callbacks. 354 */ 355 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) { 356 readonly = B_TRUE; 357 do_readonly = B_TRUE; 358 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 359 readonly = B_FALSE; 360 do_readonly = B_TRUE; 361 } 362 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 363 setuid = B_FALSE; 364 do_setuid = B_TRUE; 365 } else { 366 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 367 setuid = B_FALSE; 368 do_setuid = B_TRUE; 369 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 370 setuid = B_TRUE; 371 do_setuid = B_TRUE; 372 } 373 } 374 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 375 exec = B_FALSE; 376 do_exec = B_TRUE; 377 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 378 exec = B_TRUE; 379 do_exec = B_TRUE; 380 } 381 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 382 xattr = B_FALSE; 383 do_xattr = B_TRUE; 384 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 385 xattr = B_TRUE; 386 do_xattr = B_TRUE; 387 } 388 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 389 atime = B_FALSE; 390 do_atime = B_TRUE; 391 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 392 atime = B_TRUE; 393 do_atime = B_TRUE; 394 } 395 396 /* 397 * nbmand is a special property. It can only be changed at 398 * mount time. 399 * 400 * This is weird, but it is documented to only be changeable 401 * at mount time. 402 */ 403 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 404 nbmand = B_FALSE; 405 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 406 nbmand = B_TRUE; 407 } else { 408 char osname[MAXNAMELEN]; 409 410 dmu_objset_name(os, osname); 411 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand, 412 NULL)) { 413 return (error); 414 } 415 } 416 417 /* 418 * Register property callbacks. 419 * 420 * It would probably be fine to just check for i/o error from 421 * the first prop_register(), but I guess I like to go 422 * overboard... 423 */ 424 ds = dmu_objset_ds(os); 425 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs); 426 error = error ? error : dsl_prop_register(ds, 427 "xattr", xattr_changed_cb, zfsvfs); 428 error = error ? error : dsl_prop_register(ds, 429 "recordsize", blksz_changed_cb, zfsvfs); 430 error = error ? error : dsl_prop_register(ds, 431 "readonly", readonly_changed_cb, zfsvfs); 432 error = error ? error : dsl_prop_register(ds, 433 "setuid", setuid_changed_cb, zfsvfs); 434 error = error ? error : dsl_prop_register(ds, 435 "exec", exec_changed_cb, zfsvfs); 436 error = error ? error : dsl_prop_register(ds, 437 "snapdir", snapdir_changed_cb, zfsvfs); 438 error = error ? error : dsl_prop_register(ds, 439 "aclmode", acl_mode_changed_cb, zfsvfs); 440 error = error ? error : dsl_prop_register(ds, 441 "aclinherit", acl_inherit_changed_cb, zfsvfs); 442 error = error ? error : dsl_prop_register(ds, 443 "vscan", vscan_changed_cb, zfsvfs); 444 if (error) 445 goto unregister; 446 447 /* 448 * Invoke our callbacks to restore temporary mount options. 449 */ 450 if (do_readonly) 451 readonly_changed_cb(zfsvfs, readonly); 452 if (do_setuid) 453 setuid_changed_cb(zfsvfs, setuid); 454 if (do_exec) 455 exec_changed_cb(zfsvfs, exec); 456 if (do_xattr) 457 xattr_changed_cb(zfsvfs, xattr); 458 if (do_atime) 459 atime_changed_cb(zfsvfs, atime); 460 461 nbmand_changed_cb(zfsvfs, nbmand); 462 463 return (0); 464 465unregister: 466 /* 467 * We may attempt to unregister some callbacks that are not 468 * registered, but this is OK; it will simply return ENOMSG, 469 * which we will ignore. 470 */ 471 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs); 472 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs); 473 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs); 474 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs); 475 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs); 476 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs); 477 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs); 478 (void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs); 479 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb, 480 zfsvfs); 481 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs); 482 return (error); 483 484} 485 486static int 487zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 488{ 489 int error; 490 491 error = zfs_register_callbacks(zfsvfs->z_vfs); 492 if (error) 493 return (error); 494 495 /* 496 * Set the objset user_ptr to track its zfsvfs. 497 */ 498 mutex_enter(&zfsvfs->z_os->os->os_user_ptr_lock); 499 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 500 mutex_exit(&zfsvfs->z_os->os->os_user_ptr_lock); 501 502 /* 503 * If we are not mounting (ie: online recv), then we don't 504 * have to worry about replaying the log as we blocked all 505 * operations out since we closed the ZIL. 506 */ 507 if (mounting) { 508 boolean_t readonly; 509 510 /* 511 * During replay we remove the read only flag to 512 * allow replays to succeed. 513 */ 514 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 515 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 516 517 /* 518 * Parse and replay the intent log. 519 */ 520 zil_replay(zfsvfs->z_os, zfsvfs, &zfsvfs->z_assign, 521 zfs_replay_vector, zfs_unlinked_drain); 522 523 zfs_unlinked_drain(zfsvfs); 524 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 525 } 526 527 if (!zil_disable) 528 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 529 530 return (0); 531} 532 533static void 534zfs_freezfsvfs(zfsvfs_t *zfsvfs) 535{ 536 mutex_destroy(&zfsvfs->z_znodes_lock); 537 mutex_destroy(&zfsvfs->z_online_recv_lock); 538 list_destroy(&zfsvfs->z_all_znodes); 539 rrw_destroy(&zfsvfs->z_teardown_lock); 540 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 541 rw_destroy(&zfsvfs->z_fuid_lock); 542 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 543} 544 545static int 546zfs_domount(vfs_t *vfsp, char *osname) 547{ 548 uint64_t recordsize, readonly; 549 int error = 0; 550 int mode; 551 zfsvfs_t *zfsvfs; 552 znode_t *zp = NULL; 553 554 ASSERT(vfsp); 555 ASSERT(osname); 556 557 /* 558 * Initialize the zfs-specific filesystem structure. 559 * Should probably make this a kmem cache, shuffle fields, 560 * and just bzero up to z_hold_mtx[]. 561 */ 562 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 563 zfsvfs->z_vfs = vfsp; 564 zfsvfs->z_parent = zfsvfs; 565 zfsvfs->z_assign = TXG_NOWAIT; 566 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE; 567 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 568 569 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 570 mutex_init(&zfsvfs->z_online_recv_lock, NULL, MUTEX_DEFAULT, NULL); 571 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 572 offsetof(znode_t, z_link_node)); 573 rrw_init(&zfsvfs->z_teardown_lock); 574 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 575 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 576 577 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 578 NULL)) 579 goto out; 580 zfsvfs->z_vfs->vfs_bsize = recordsize; 581 582 vfsp->vfs_data = zfsvfs; 583 vfsp->mnt_flag |= MNT_LOCAL; 584 vfsp->mnt_kern_flag |= MNTK_MPSAFE; 585 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 586 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 587 588 if (error = dsl_prop_get_integer(osname, "readonly", &readonly, NULL)) 589 goto out; 590 591 mode = DS_MODE_OWNER; 592 if (readonly) 593 mode |= DS_MODE_READONLY; 594 595 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os); 596 if (error == EROFS) { 597 mode = DS_MODE_OWNER | DS_MODE_READONLY; 598 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, 599 &zfsvfs->z_os); 600 } 601 602 if (error) 603 goto out; 604 605 if (error = zfs_init_fs(zfsvfs, &zp)) 606 goto out; 607 608 /* 609 * Set features for file system. 610 */ 611 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 612 if (zfsvfs->z_use_fuids) { 613 vfs_set_feature(vfsp, VFSFT_XVATTR); 614 vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS); 615 vfs_set_feature(vfsp, VFSFT_ACEMASKONACCESS); 616 vfs_set_feature(vfsp, VFSFT_ACLONCREATE); 617 } 618 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 619 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 620 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 621 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 622 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 623 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 624 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 625 } 626 627 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 628 uint64_t pval; 629 630 ASSERT(mode & DS_MODE_READONLY); 631 atime_changed_cb(zfsvfs, B_FALSE); 632 readonly_changed_cb(zfsvfs, B_TRUE); 633 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 634 goto out; 635 xattr_changed_cb(zfsvfs, pval); 636 zfsvfs->z_issnap = B_TRUE; 637 } else { 638 error = zfsvfs_setup(zfsvfs, B_TRUE); 639 } 640 641 vfs_mountedfrom(vfsp, osname); 642 643 if (!zfsvfs->z_issnap) 644 zfsctl_create(zfsvfs); 645out: 646 if (error) { 647 if (zfsvfs->z_os) 648 dmu_objset_close(zfsvfs->z_os); 649 zfs_freezfsvfs(zfsvfs); 650 } else { 651 atomic_add_32(&zfs_active_fs_count, 1); 652 } 653 654 return (error); 655} 656 657void 658zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 659{ 660 objset_t *os = zfsvfs->z_os; 661 struct dsl_dataset *ds; 662 663 /* 664 * Unregister properties. 665 */ 666 if (!dmu_objset_is_snapshot(os)) { 667 ds = dmu_objset_ds(os); 668 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, 669 zfsvfs) == 0); 670 671 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, 672 zfsvfs) == 0); 673 674 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, 675 zfsvfs) == 0); 676 677 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, 678 zfsvfs) == 0); 679 680 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, 681 zfsvfs) == 0); 682 683 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, 684 zfsvfs) == 0); 685 686 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, 687 zfsvfs) == 0); 688 689 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, 690 zfsvfs) == 0); 691 692 VERIFY(dsl_prop_unregister(ds, "aclinherit", 693 acl_inherit_changed_cb, zfsvfs) == 0); 694 695 VERIFY(dsl_prop_unregister(ds, "vscan", 696 vscan_changed_cb, zfsvfs) == 0); 697 } 698} 699 700/*ARGSUSED*/ 701static int 702zfs_mount(vfs_t *vfsp) 703{ 704 kthread_t *td = curthread; 705 vnode_t *mvp = vfsp->mnt_vnodecovered; 706 cred_t *cr = td->td_ucred; 707 char *osname; 708 int error = 0; 709 int canwrite; 710 711 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 712 return (EINVAL); 713 714 /* 715 * If full-owner-access is enabled and delegated administration is 716 * turned on, we must set nosuid. 717 */ 718 if (zfs_super_owner && 719 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 720 secpolicy_fs_mount_clearopts(cr, vfsp); 721 } 722 723 /* 724 * Check for mount privilege? 725 * 726 * If we don't have privilege then see if 727 * we have local permission to allow it 728 */ 729 error = secpolicy_fs_mount(cr, mvp, vfsp); 730 if (error) { 731 error = dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr); 732 if (error != 0) 733 goto out; 734 735 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 736 vattr_t vattr; 737 738 /* 739 * Make sure user is the owner of the mount point 740 * or has sufficient privileges. 741 */ 742 743 vattr.va_mask = AT_UID; 744 745 vn_lock(mvp, LK_SHARED | LK_RETRY); 746 if (error = VOP_GETATTR(mvp, &vattr, cr)) { 747 VOP_UNLOCK(mvp, 0); 748 goto out; 749 } 750 751#if 0 /* CHECK THIS! Is probably needed for zfs_suser. */ 752 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 753 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 754 error = EPERM; 755 goto out; 756 } 757#else 758 if (error = secpolicy_vnode_owner(mvp, cr, vattr.va_uid)) { 759 VOP_UNLOCK(mvp, 0); 760 goto out; 761 } 762 763 if (error = VOP_ACCESS(mvp, VWRITE, cr, td)) { 764 VOP_UNLOCK(mvp, 0); 765 goto out; 766 } 767 VOP_UNLOCK(mvp, 0); 768#endif 769 } 770 771 secpolicy_fs_mount_clearopts(cr, vfsp); 772 } 773 774 /* 775 * Refuse to mount a filesystem if we are in a local zone and the 776 * dataset is not visible. 777 */ 778 if (!INGLOBALZONE(curthread) && 779 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 780 error = EPERM; 781 goto out; 782 } 783 784 /* 785 * When doing a remount, we simply refresh our temporary properties 786 * according to those options set in the current VFS options. 787 */ 788 if (vfsp->vfs_flag & MS_REMOUNT) { 789 /* refresh mount options */ 790 zfs_unregister_callbacks(vfsp->vfs_data); 791 error = zfs_register_callbacks(vfsp); 792 goto out; 793 } 794 795 DROP_GIANT(); 796 error = zfs_domount(vfsp, osname); 797 PICKUP_GIANT(); 798out: 799 return (error); 800} 801 802static int 803zfs_statfs(vfs_t *vfsp, struct statfs *statp) 804{ 805 zfsvfs_t *zfsvfs = vfsp->vfs_data; 806 uint64_t refdbytes, availbytes, usedobjs, availobjs; 807 808 statp->f_version = STATFS_VERSION; 809 810 ZFS_ENTER(zfsvfs); 811 812 dmu_objset_space(zfsvfs->z_os, 813 &refdbytes, &availbytes, &usedobjs, &availobjs); 814 815 /* 816 * The underlying storage pool actually uses multiple block sizes. 817 * We report the fragsize as the smallest block size we support, 818 * and we report our blocksize as the filesystem's maximum blocksize. 819 */ 820 statp->f_bsize = zfsvfs->z_vfs->vfs_bsize; 821 statp->f_iosize = zfsvfs->z_vfs->vfs_bsize; 822 823 /* 824 * The following report "total" blocks of various kinds in the 825 * file system, but reported in terms of f_frsize - the 826 * "fragment" size. 827 */ 828 829 statp->f_blocks = (refdbytes + availbytes) / statp->f_bsize; 830 statp->f_bfree = availbytes / statp->f_bsize; 831 statp->f_bavail = statp->f_bfree; /* no root reservation */ 832 833 /* 834 * statvfs() should really be called statufs(), because it assumes 835 * static metadata. ZFS doesn't preallocate files, so the best 836 * we can do is report the max that could possibly fit in f_files, 837 * and that minus the number actually used in f_ffree. 838 * For f_ffree, report the smaller of the number of object available 839 * and the number of blocks (each object will take at least a block). 840 */ 841 statp->f_ffree = MIN(availobjs, statp->f_bfree); 842 statp->f_files = statp->f_ffree + usedobjs; 843 844 /* 845 * We're a zfs filesystem. 846 */ 847 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 848 849 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 850 sizeof(statp->f_mntfromname)); 851 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 852 sizeof(statp->f_mntonname)); 853 854 statp->f_namemax = ZFS_MAXNAMELEN; 855 856 ZFS_EXIT(zfsvfs); 857 return (0); 858} 859 860static int 861zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 862{ 863 zfsvfs_t *zfsvfs = vfsp->vfs_data; 864 znode_t *rootzp; 865 int error; 866 867 ZFS_ENTER(zfsvfs); 868 869 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 870 if (error == 0) { 871 *vpp = ZTOV(rootzp); 872 error = vn_lock(*vpp, flags); 873 (*vpp)->v_vflag |= VV_ROOT; 874 } 875 876 ZFS_EXIT(zfsvfs); 877 return (error); 878} 879 880/* 881 * Teardown the zfsvfs::z_os. 882 * 883 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 884 * and 'z_teardown_inactive_lock' held. 885 */ 886static int 887zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 888{ 889 znode_t *zp; 890 891 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 892 893 if (!unmounting) { 894 /* 895 * We purge the parent filesystem's vfsp as the parent 896 * filesystem and all of its snapshots have their vnode's 897 * v_vfsp set to the parent's filesystem's vfsp. Note, 898 * 'z_parent' is self referential for non-snapshots. 899 */ 900 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 901#ifdef FREEBSD_NAMECACHE 902 cache_purgevfs(zfsvfs->z_parent->z_vfs); 903#endif 904 } 905 906 /* 907 * Close the zil. NB: Can't close the zil while zfs_inactive 908 * threads are blocked as zil_close can call zfs_inactive. 909 */ 910 if (zfsvfs->z_log) { 911 zil_close(zfsvfs->z_log); 912 zfsvfs->z_log = NULL; 913 } 914 915 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 916 917 /* 918 * If we are not unmounting (ie: online recv) and someone already 919 * unmounted this file system while we were doing the switcheroo, 920 * or a reopen of z_os failed then just bail out now. 921 */ 922 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 923 rw_exit(&zfsvfs->z_teardown_inactive_lock); 924 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 925 return (EIO); 926 } 927 928 /* 929 * At this point there are no vops active, and any new vops will 930 * fail with EIO since we have z_teardown_lock for writer (only 931 * relavent for forced unmount). 932 * 933 * Release all holds on dbufs. 934 */ 935 mutex_enter(&zfsvfs->z_znodes_lock); 936 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 937 zp = list_next(&zfsvfs->z_all_znodes, zp)) 938 if (zp->z_dbuf) { 939 ASSERT(ZTOV(zp)->v_count >= 0); 940 zfs_znode_dmu_fini(zp); 941 } 942 mutex_exit(&zfsvfs->z_znodes_lock); 943 944 /* 945 * If we are unmounting, set the unmounted flag and let new vops 946 * unblock. zfs_inactive will have the unmounted behavior, and all 947 * other vops will fail with EIO. 948 */ 949 if (unmounting) { 950 zfsvfs->z_unmounted = B_TRUE; 951 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 952 rw_exit(&zfsvfs->z_teardown_inactive_lock); 953 954#ifdef __FreeBSD__ 955 /* 956 * Some znodes might not be fully reclaimed, wait for them. 957 */ 958 mutex_enter(&zfsvfs->z_znodes_lock); 959 while (list_head(&zfsvfs->z_all_znodes) != NULL) { 960 msleep(zfsvfs, &zfsvfs->z_znodes_lock, 0, 961 "zteardown", 0); 962 } 963 mutex_exit(&zfsvfs->z_znodes_lock); 964#endif 965 } 966 967 /* 968 * z_os will be NULL if there was an error in attempting to reopen 969 * zfsvfs, so just return as the properties had already been 970 * unregistered and cached data had been evicted before. 971 */ 972 if (zfsvfs->z_os == NULL) 973 return (0); 974 975 /* 976 * Unregister properties. 977 */ 978 zfs_unregister_callbacks(zfsvfs); 979 980 /* 981 * Evict cached data 982 */ 983 if (dmu_objset_evict_dbufs(zfsvfs->z_os)) { 984 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 985 (void) dmu_objset_evict_dbufs(zfsvfs->z_os); 986 } 987 988 return (0); 989} 990 991/*ARGSUSED*/ 992static int 993zfs_umount(vfs_t *vfsp, int fflag) 994{ 995 zfsvfs_t *zfsvfs = vfsp->vfs_data; 996 objset_t *os; 997 cred_t *cr = curthread->td_ucred; 998 int ret; 999 1000 ret = secpolicy_fs_unmount(cr, vfsp); 1001 if (ret) { 1002 ret = dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1003 ZFS_DELEG_PERM_MOUNT, cr); 1004 if (ret) 1005 return (ret); 1006 } 1007 /* 1008 * We purge the parent filesystem's vfsp as the parent filesystem 1009 * and all of its snapshots have their vnode's v_vfsp set to the 1010 * parent's filesystem's vfsp. Note, 'z_parent' is self 1011 * referential for non-snapshots. 1012 */ 1013 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1014 1015 /* 1016 * Unmount any snapshots mounted under .zfs before unmounting the 1017 * dataset itself. 1018 */ 1019 if (zfsvfs->z_ctldir != NULL) { 1020 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1021 return (ret); 1022 ret = vflush(vfsp, 0, 0, curthread); 1023 ASSERT(ret == EBUSY); 1024 if (!(fflag & MS_FORCE)) { 1025 if (zfsvfs->z_ctldir->v_count > 1) 1026 return (EBUSY); 1027 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1028 } 1029 zfsctl_destroy(zfsvfs); 1030 ASSERT(zfsvfs->z_ctldir == NULL); 1031 } 1032 1033 /* 1034 * Flush all the files. 1035 */ 1036 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, curthread); 1037 if (ret != 0) { 1038 if (!zfsvfs->z_issnap) { 1039 zfsctl_create(zfsvfs); 1040 ASSERT(zfsvfs->z_ctldir != NULL); 1041 } 1042 return (ret); 1043 } 1044 1045 if (!(fflag & MS_FORCE)) { 1046 /* 1047 * Check the number of active vnodes in the file system. 1048 * Our count is maintained in the vfs structure, but the 1049 * number is off by 1 to indicate a hold on the vfs 1050 * structure itself. 1051 * 1052 * The '.zfs' directory maintains a reference of its 1053 * own, and any active references underneath are 1054 * reflected in the vnode count. 1055 */ 1056 if (zfsvfs->z_ctldir == NULL) { 1057 if (vfsp->vfs_count > 1) 1058 return (EBUSY); 1059 } else { 1060 if (vfsp->vfs_count > 2 || 1061 zfsvfs->z_ctldir->v_count > 1) 1062 return (EBUSY); 1063 } 1064 } else { 1065 MNT_ILOCK(vfsp); 1066 vfsp->mnt_kern_flag |= MNTK_UNMOUNTF; 1067 MNT_IUNLOCK(vfsp); 1068 } 1069 1070 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1071 os = zfsvfs->z_os; 1072 1073 /* 1074 * z_os will be NULL if there was an error in 1075 * attempting to reopen zfsvfs. 1076 */ 1077 if (os != NULL) { 1078 /* 1079 * Unset the objset user_ptr. 1080 */ 1081 mutex_enter(&os->os->os_user_ptr_lock); 1082 dmu_objset_set_user(os, NULL); 1083 mutex_exit(&os->os->os_user_ptr_lock); 1084 1085 /* 1086 * Finally release the objset 1087 */ 1088 dmu_objset_close(os); 1089 } 1090 1091 /* 1092 * We can now safely destroy the '.zfs' directory node. 1093 */ 1094 if (zfsvfs->z_ctldir != NULL) 1095 zfsctl_destroy(zfsvfs); 1096 if (zfsvfs->z_issnap) { 1097 vnode_t *svp = vfsp->mnt_vnodecovered; 1098 1099 ASSERT(svp->v_count == 2 || svp->v_count == 1); 1100 if (svp->v_count == 2) 1101 VN_RELE(svp); 1102 } 1103 zfs_freevfs(vfsp); 1104 1105 return (0); 1106} 1107 1108static int 1109zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 1110{ 1111 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1112 znode_t *zp; 1113 int err; 1114 1115 /* 1116 * XXXPJD: zfs_zget() can't operate on virtual entires like .zfs/ or 1117 * .zfs/snapshot/ directories, so for now just return EOPNOTSUPP. 1118 * This will make NFS to fall back to using READDIR instead of 1119 * READDIRPLUS. 1120 * Also snapshots are stored in AVL tree, but based on their names, 1121 * not inode numbers, so it will be very inefficient to iterate 1122 * over all snapshots to find the right one. 1123 * Note that OpenSolaris READDIRPLUS implementation does LOOKUP on 1124 * d_name, and not VGET on d_fileno as we do. 1125 */ 1126 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR) 1127 return (EOPNOTSUPP); 1128 1129 ZFS_ENTER(zfsvfs); 1130 err = zfs_zget(zfsvfs, ino, &zp); 1131 if (err == 0 && zp->z_unlinked) { 1132 VN_RELE(ZTOV(zp)); 1133 err = EINVAL; 1134 } 1135 if (err != 0) 1136 *vpp = NULL; 1137 else { 1138 *vpp = ZTOV(zp); 1139 vn_lock(*vpp, flags); 1140 } 1141 ZFS_EXIT(zfsvfs); 1142 return (err); 1143} 1144 1145static int 1146zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 1147 struct ucred **credanonp, int *numsecflavors, int **secflavors) 1148{ 1149 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1150 1151 /* 1152 * If this is regular file system vfsp is the same as 1153 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 1154 * zfsvfs->z_parent->z_vfs represents parent file system 1155 * which we have to use here, because only this file system 1156 * has mnt_export configured. 1157 */ 1158 vfsp = zfsvfs->z_parent->z_vfs; 1159 1160 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 1161 credanonp, numsecflavors, secflavors)); 1162} 1163 1164CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 1165CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 1166 1167static int 1168zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, vnode_t **vpp) 1169{ 1170 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1171 znode_t *zp; 1172 uint64_t object = 0; 1173 uint64_t fid_gen = 0; 1174 uint64_t gen_mask; 1175 uint64_t zp_gen; 1176 int i, err; 1177 1178 *vpp = NULL; 1179 1180 ZFS_ENTER(zfsvfs); 1181 1182 /* 1183 * On FreeBSD we can get snapshot's mount point or its parent file 1184 * system mount point depending if snapshot is already mounted or not. 1185 */ 1186 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 1187 zfid_long_t *zlfid = (zfid_long_t *)fidp; 1188 uint64_t objsetid = 0; 1189 uint64_t setgen = 0; 1190 1191 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 1192 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 1193 1194 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 1195 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 1196 1197 ZFS_EXIT(zfsvfs); 1198 1199 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 1200 if (err) 1201 return (EINVAL); 1202 ZFS_ENTER(zfsvfs); 1203 } 1204 1205 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 1206 zfid_short_t *zfid = (zfid_short_t *)fidp; 1207 1208 for (i = 0; i < sizeof (zfid->zf_object); i++) 1209 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 1210 1211 for (i = 0; i < sizeof (zfid->zf_gen); i++) 1212 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 1213 } else { 1214 ZFS_EXIT(zfsvfs); 1215 return (EINVAL); 1216 } 1217 1218 /* A zero fid_gen means we are in the .zfs control directories */ 1219 if (fid_gen == 0 && 1220 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 1221 *vpp = zfsvfs->z_ctldir; 1222 ASSERT(*vpp != NULL); 1223 if (object == ZFSCTL_INO_SNAPDIR) { 1224 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 1225 0, NULL, NULL, NULL, NULL, NULL) == 0); 1226 } else { 1227 VN_HOLD(*vpp); 1228 } 1229 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1230 ZFS_EXIT(zfsvfs); 1231 return (0); 1232 } 1233 1234 gen_mask = -1ULL >> (64 - 8 * i); 1235 1236 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 1237 if (err = zfs_zget(zfsvfs, object, &zp)) { 1238 ZFS_EXIT(zfsvfs); 1239 return (err); 1240 } 1241 zp_gen = zp->z_phys->zp_gen & gen_mask; 1242 if (zp_gen == 0) 1243 zp_gen = 1; 1244 if (zp->z_unlinked || zp_gen != fid_gen) { 1245 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 1246 VN_RELE(ZTOV(zp)); 1247 ZFS_EXIT(zfsvfs); 1248 return (EINVAL); 1249 } 1250 1251 *vpp = ZTOV(zp); 1252 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1253 vnode_create_vobject(*vpp, zp->z_phys->zp_size, curthread); 1254 ZFS_EXIT(zfsvfs); 1255 return (0); 1256} 1257 1258/* 1259 * Block out VOPs and close zfsvfs_t::z_os 1260 * 1261 * Note, if successful, then we return with the 'z_teardown_lock' and 1262 * 'z_teardown_inactive_lock' write held. 1263 */ 1264int 1265zfs_suspend_fs(zfsvfs_t *zfsvfs, char *name, int *mode) 1266{ 1267 int error; 1268 1269 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 1270 return (error); 1271 1272 *mode = zfsvfs->z_os->os_mode; 1273 dmu_objset_name(zfsvfs->z_os, name); 1274 dmu_objset_close(zfsvfs->z_os); 1275 1276 return (0); 1277} 1278 1279/* 1280 * Reopen zfsvfs_t::z_os and release VOPs. 1281 */ 1282int 1283zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname, int mode) 1284{ 1285 int err; 1286 1287 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock)); 1288 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1289 1290 err = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os); 1291 if (err) { 1292 zfsvfs->z_os = NULL; 1293 } else { 1294 znode_t *zp; 1295 1296 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 1297 1298 /* 1299 * Attempt to re-establish all the active znodes with 1300 * their dbufs. If a zfs_rezget() fails, then we'll let 1301 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 1302 * when they try to use their znode. 1303 */ 1304 mutex_enter(&zfsvfs->z_znodes_lock); 1305 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 1306 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1307 (void) zfs_rezget(zp); 1308 } 1309 mutex_exit(&zfsvfs->z_znodes_lock); 1310 1311 } 1312 1313 /* release the VOPs */ 1314 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1315 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1316 1317 if (err) { 1318 /* 1319 * Since we couldn't reopen zfsvfs::z_os, force 1320 * unmount this file system. 1321 */ 1322 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) 1323 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 1324 } 1325 return (err); 1326} 1327 1328static void 1329zfs_freevfs(vfs_t *vfsp) 1330{ 1331 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1332 int i; 1333 1334 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1335 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1336 1337 zfs_fuid_destroy(zfsvfs); 1338 zfs_freezfsvfs(zfsvfs); 1339 1340 atomic_add_32(&zfs_active_fs_count, -1); 1341} 1342 1343#ifdef __i386__ 1344static int desiredvnodes_backup; 1345#endif 1346 1347static void 1348zfs_vnodes_adjust(void) 1349{ 1350#ifdef __i386__ 1351 int newdesiredvnodes; 1352 1353 desiredvnodes_backup = desiredvnodes; 1354 1355 /* 1356 * We calculate newdesiredvnodes the same way it is done in 1357 * vntblinit(). If it is equal to desiredvnodes, it means that 1358 * it wasn't tuned by the administrator and we can tune it down. 1359 */ 1360 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 1361 vm_kmem_size / (5 * (sizeof(struct vm_object) + 1362 sizeof(struct vnode)))); 1363 if (newdesiredvnodes == desiredvnodes) 1364 desiredvnodes = (3 * newdesiredvnodes) / 4; 1365#endif 1366} 1367 1368static void 1369zfs_vnodes_adjust_back(void) 1370{ 1371 1372#ifdef __i386__ 1373 desiredvnodes = desiredvnodes_backup; 1374#endif 1375} 1376 1377void 1378zfs_init(void) 1379{ 1380 1381 printf("ZFS filesystem version " SPA_VERSION_STRING "\n"); 1382 1383 /* 1384 * Initialize znode cache, vnode ops, etc... 1385 */ 1386 zfs_znode_init(); 1387 1388 /* 1389 * Initialize .zfs directory structures 1390 */ 1391 zfsctl_init(); 1392 1393 /* 1394 * Reduce number of vnode. Originally number of vnodes is calculated 1395 * with UFS inode in mind. We reduce it here, because it's too big for 1396 * ZFS/i386. 1397 */ 1398 zfs_vnodes_adjust(); 1399} 1400 1401void 1402zfs_fini(void) 1403{ 1404 zfsctl_fini(); 1405 zfs_znode_fini(); 1406 zfs_vnodes_adjust_back(); 1407} 1408 1409int 1410zfs_busy(void) 1411{ 1412 return (zfs_active_fs_count != 0); 1413} 1414 1415int 1416zfs_set_version(const char *name, uint64_t newvers) 1417{ 1418 int error; 1419 objset_t *os; 1420 dmu_tx_t *tx; 1421 uint64_t curvers; 1422 1423 /* 1424 * XXX for now, require that the filesystem be unmounted. Would 1425 * be nice to find the zfsvfs_t and just update that if 1426 * possible. 1427 */ 1428 1429 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 1430 return (EINVAL); 1431 1432 error = dmu_objset_open(name, DMU_OST_ZFS, DS_MODE_OWNER, &os); 1433 if (error) 1434 return (error); 1435 1436 error = zap_lookup(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 1437 8, 1, &curvers); 1438 if (error) 1439 goto out; 1440 if (newvers < curvers) { 1441 error = EINVAL; 1442 goto out; 1443 } 1444 1445 tx = dmu_tx_create(os); 1446 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, 0, ZPL_VERSION_STR); 1447 error = dmu_tx_assign(tx, TXG_WAIT); 1448 if (error) { 1449 dmu_tx_abort(tx); 1450 goto out; 1451 } 1452 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 8, 1, 1453 &newvers, tx); 1454 1455 spa_history_internal_log(LOG_DS_UPGRADE, 1456 dmu_objset_spa(os), tx, CRED(), 1457 "oldver=%llu newver=%llu dataset = %llu", curvers, newvers, 1458 dmu_objset_id(os)); 1459 dmu_tx_commit(tx); 1460 1461out: 1462 dmu_objset_close(os); 1463 return (error); 1464} 1465/* 1466 * Read a property stored within the master node. 1467 */ 1468int 1469zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 1470{ 1471 const char *pname; 1472 int error = ENOENT; 1473 1474 /* 1475 * Look up the file system's value for the property. For the 1476 * version property, we look up a slightly different string. 1477 */ 1478 if (prop == ZFS_PROP_VERSION) 1479 pname = ZPL_VERSION_STR; 1480 else 1481 pname = zfs_prop_to_name(prop); 1482 1483 if (os != NULL) 1484 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 1485 1486 if (error == ENOENT) { 1487 /* No value set, use the default value */ 1488 switch (prop) { 1489 case ZFS_PROP_VERSION: 1490 *value = ZPL_VERSION; 1491 break; 1492 case ZFS_PROP_NORMALIZE: 1493 case ZFS_PROP_UTF8ONLY: 1494 *value = 0; 1495 break; 1496 case ZFS_PROP_CASE: 1497 *value = ZFS_CASE_SENSITIVE; 1498 break; 1499 default: 1500 return (error); 1501 } 1502 error = 0; 1503 } 1504 return (error); 1505} 1506