zfs_vfsops.c revision 204101
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->mnt_stat.f_iosize = 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 = SPA_MINBLOCKSIZE; 581 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 582 583 vfsp->vfs_data = zfsvfs; 584 vfsp->mnt_flag |= MNT_LOCAL; 585 vfsp->mnt_kern_flag |= MNTK_MPSAFE; 586 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 587 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 588 589 if (error = dsl_prop_get_integer(osname, "readonly", &readonly, NULL)) 590 goto out; 591 592 mode = DS_MODE_OWNER; 593 if (readonly) 594 mode |= DS_MODE_READONLY; 595 596 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os); 597 if (error == EROFS) { 598 mode = DS_MODE_OWNER | DS_MODE_READONLY; 599 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, 600 &zfsvfs->z_os); 601 } 602 603 if (error) 604 goto out; 605 606 if (error = zfs_init_fs(zfsvfs, &zp)) 607 goto out; 608 609 /* 610 * Set features for file system. 611 */ 612 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 613 if (zfsvfs->z_use_fuids) { 614 vfs_set_feature(vfsp, VFSFT_XVATTR); 615 vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS); 616 vfs_set_feature(vfsp, VFSFT_ACEMASKONACCESS); 617 vfs_set_feature(vfsp, VFSFT_ACLONCREATE); 618 } 619 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 620 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 621 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 622 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 623 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 624 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 625 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 626 } 627 628 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 629 uint64_t pval; 630 631 ASSERT(mode & DS_MODE_READONLY); 632 atime_changed_cb(zfsvfs, B_FALSE); 633 readonly_changed_cb(zfsvfs, B_TRUE); 634 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 635 goto out; 636 xattr_changed_cb(zfsvfs, pval); 637 zfsvfs->z_issnap = B_TRUE; 638 } else { 639 error = zfsvfs_setup(zfsvfs, B_TRUE); 640 } 641 642 vfs_mountedfrom(vfsp, osname); 643 644 if (!zfsvfs->z_issnap) 645 zfsctl_create(zfsvfs); 646out: 647 if (error) { 648 if (zfsvfs->z_os) 649 dmu_objset_close(zfsvfs->z_os); 650 zfs_freezfsvfs(zfsvfs); 651 } else { 652 atomic_add_32(&zfs_active_fs_count, 1); 653 } 654 655 return (error); 656} 657 658void 659zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 660{ 661 objset_t *os = zfsvfs->z_os; 662 struct dsl_dataset *ds; 663 664 /* 665 * Unregister properties. 666 */ 667 if (!dmu_objset_is_snapshot(os)) { 668 ds = dmu_objset_ds(os); 669 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, 670 zfsvfs) == 0); 671 672 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, 673 zfsvfs) == 0); 674 675 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, 676 zfsvfs) == 0); 677 678 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, 679 zfsvfs) == 0); 680 681 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, 682 zfsvfs) == 0); 683 684 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, 685 zfsvfs) == 0); 686 687 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, 688 zfsvfs) == 0); 689 690 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, 691 zfsvfs) == 0); 692 693 VERIFY(dsl_prop_unregister(ds, "aclinherit", 694 acl_inherit_changed_cb, zfsvfs) == 0); 695 696 VERIFY(dsl_prop_unregister(ds, "vscan", 697 vscan_changed_cb, zfsvfs) == 0); 698 } 699} 700 701/*ARGSUSED*/ 702static int 703zfs_mount(vfs_t *vfsp) 704{ 705 kthread_t *td = curthread; 706 vnode_t *mvp = vfsp->mnt_vnodecovered; 707 cred_t *cr = td->td_ucred; 708 char *osname; 709 int error = 0; 710 int canwrite; 711 712 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 713 return (EINVAL); 714 715 /* 716 * If full-owner-access is enabled and delegated administration is 717 * turned on, we must set nosuid. 718 */ 719 if (zfs_super_owner && 720 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 721 secpolicy_fs_mount_clearopts(cr, vfsp); 722 } 723 724 /* 725 * Check for mount privilege? 726 * 727 * If we don't have privilege then see if 728 * we have local permission to allow it 729 */ 730 error = secpolicy_fs_mount(cr, mvp, vfsp); 731 if (error) { 732 error = dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr); 733 if (error != 0) 734 goto out; 735 736 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 737 vattr_t vattr; 738 739 /* 740 * Make sure user is the owner of the mount point 741 * or has sufficient privileges. 742 */ 743 744 vattr.va_mask = AT_UID; 745 746 vn_lock(mvp, LK_SHARED | LK_RETRY); 747 if (error = VOP_GETATTR(mvp, &vattr, cr)) { 748 VOP_UNLOCK(mvp, 0); 749 goto out; 750 } 751 752#if 0 /* CHECK THIS! Is probably needed for zfs_suser. */ 753 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 754 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 755 error = EPERM; 756 goto out; 757 } 758#else 759 if (error = secpolicy_vnode_owner(mvp, cr, vattr.va_uid)) { 760 VOP_UNLOCK(mvp, 0); 761 goto out; 762 } 763 764 if (error = VOP_ACCESS(mvp, VWRITE, cr, td)) { 765 VOP_UNLOCK(mvp, 0); 766 goto out; 767 } 768 VOP_UNLOCK(mvp, 0); 769#endif 770 } 771 772 secpolicy_fs_mount_clearopts(cr, vfsp); 773 } 774 775 /* 776 * Refuse to mount a filesystem if we are in a local zone and the 777 * dataset is not visible. 778 */ 779 if (!INGLOBALZONE(curthread) && 780 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 781 error = EPERM; 782 goto out; 783 } 784 785 /* 786 * When doing a remount, we simply refresh our temporary properties 787 * according to those options set in the current VFS options. 788 */ 789 if (vfsp->vfs_flag & MS_REMOUNT) { 790 /* refresh mount options */ 791 zfs_unregister_callbacks(vfsp->vfs_data); 792 error = zfs_register_callbacks(vfsp); 793 goto out; 794 } 795 796 DROP_GIANT(); 797 error = zfs_domount(vfsp, osname); 798 PICKUP_GIANT(); 799out: 800 return (error); 801} 802 803static int 804zfs_statfs(vfs_t *vfsp, struct statfs *statp) 805{ 806 zfsvfs_t *zfsvfs = vfsp->vfs_data; 807 uint64_t refdbytes, availbytes, usedobjs, availobjs; 808 809 statp->f_version = STATFS_VERSION; 810 811 ZFS_ENTER(zfsvfs); 812 813 dmu_objset_space(zfsvfs->z_os, 814 &refdbytes, &availbytes, &usedobjs, &availobjs); 815 816 /* 817 * The underlying storage pool actually uses multiple block sizes. 818 * We report the fragsize as the smallest block size we support, 819 * and we report our blocksize as the filesystem's maximum blocksize. 820 */ 821 statp->f_bsize = SPA_MINBLOCKSIZE; 822 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 823 824 /* 825 * The following report "total" blocks of various kinds in the 826 * file system, but reported in terms of f_frsize - the 827 * "fragment" size. 828 */ 829 830 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 831 statp->f_bfree = availbytes / statp->f_bsize; 832 statp->f_bavail = statp->f_bfree; /* no root reservation */ 833 834 /* 835 * statvfs() should really be called statufs(), because it assumes 836 * static metadata. ZFS doesn't preallocate files, so the best 837 * we can do is report the max that could possibly fit in f_files, 838 * and that minus the number actually used in f_ffree. 839 * For f_ffree, report the smaller of the number of object available 840 * and the number of blocks (each object will take at least a block). 841 */ 842 statp->f_ffree = MIN(availobjs, statp->f_bfree); 843 statp->f_files = statp->f_ffree + usedobjs; 844 845 /* 846 * We're a zfs filesystem. 847 */ 848 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 849 850 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 851 sizeof(statp->f_mntfromname)); 852 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 853 sizeof(statp->f_mntonname)); 854 855 statp->f_namemax = ZFS_MAXNAMELEN; 856 857 ZFS_EXIT(zfsvfs); 858 return (0); 859} 860 861static int 862zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 863{ 864 zfsvfs_t *zfsvfs = vfsp->vfs_data; 865 znode_t *rootzp; 866 int error; 867 868 ZFS_ENTER_NOERROR(zfsvfs); 869 870 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 871 if (error == 0) { 872 *vpp = ZTOV(rootzp); 873 error = vn_lock(*vpp, flags); 874 (*vpp)->v_vflag |= VV_ROOT; 875 } 876 877 ZFS_EXIT(zfsvfs); 878 return (error); 879} 880 881/* 882 * Teardown the zfsvfs::z_os. 883 * 884 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 885 * and 'z_teardown_inactive_lock' held. 886 */ 887static int 888zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 889{ 890 znode_t *zp; 891 892 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 893 894 if (!unmounting) { 895 /* 896 * We purge the parent filesystem's vfsp as the parent 897 * filesystem and all of its snapshots have their vnode's 898 * v_vfsp set to the parent's filesystem's vfsp. Note, 899 * 'z_parent' is self referential for non-snapshots. 900 */ 901 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 902#ifdef FREEBSD_NAMECACHE 903 cache_purgevfs(zfsvfs->z_parent->z_vfs); 904#endif 905 } 906 907 /* 908 * Close the zil. NB: Can't close the zil while zfs_inactive 909 * threads are blocked as zil_close can call zfs_inactive. 910 */ 911 if (zfsvfs->z_log) { 912 zil_close(zfsvfs->z_log); 913 zfsvfs->z_log = NULL; 914 } 915 916 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 917 918 /* 919 * If we are not unmounting (ie: online recv) and someone already 920 * unmounted this file system while we were doing the switcheroo, 921 * or a reopen of z_os failed then just bail out now. 922 */ 923 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 924 rw_exit(&zfsvfs->z_teardown_inactive_lock); 925 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 926 return (EIO); 927 } 928 929 /* 930 * At this point there are no vops active, and any new vops will 931 * fail with EIO since we have z_teardown_lock for writer (only 932 * relavent for forced unmount). 933 * 934 * Release all holds on dbufs. 935 */ 936 mutex_enter(&zfsvfs->z_znodes_lock); 937 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 938 zp = list_next(&zfsvfs->z_all_znodes, zp)) 939 if (zp->z_dbuf) { 940 ASSERT(ZTOV(zp)->v_count >= 0); 941 zfs_znode_dmu_fini(zp); 942 } 943 mutex_exit(&zfsvfs->z_znodes_lock); 944 945 /* 946 * If we are unmounting, set the unmounted flag and let new vops 947 * unblock. zfs_inactive will have the unmounted behavior, and all 948 * other vops will fail with EIO. 949 */ 950 if (unmounting) { 951 zfsvfs->z_unmounted = B_TRUE; 952 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 953 rw_exit(&zfsvfs->z_teardown_inactive_lock); 954 955#ifdef __FreeBSD__ 956 /* 957 * Some znodes might not be fully reclaimed, wait for them. 958 */ 959 mutex_enter(&zfsvfs->z_znodes_lock); 960 while (list_head(&zfsvfs->z_all_znodes) != NULL) { 961 msleep(zfsvfs, &zfsvfs->z_znodes_lock, 0, 962 "zteardown", 0); 963 } 964 mutex_exit(&zfsvfs->z_znodes_lock); 965#endif 966 } 967 968 /* 969 * z_os will be NULL if there was an error in attempting to reopen 970 * zfsvfs, so just return as the properties had already been 971 * unregistered and cached data had been evicted before. 972 */ 973 if (zfsvfs->z_os == NULL) 974 return (0); 975 976 /* 977 * Unregister properties. 978 */ 979 zfs_unregister_callbacks(zfsvfs); 980 981 /* 982 * Evict cached data 983 */ 984 if (dmu_objset_evict_dbufs(zfsvfs->z_os)) { 985 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 986 (void) dmu_objset_evict_dbufs(zfsvfs->z_os); 987 } 988 989 return (0); 990} 991 992/*ARGSUSED*/ 993static int 994zfs_umount(vfs_t *vfsp, int fflag) 995{ 996 zfsvfs_t *zfsvfs = vfsp->vfs_data; 997 objset_t *os; 998 cred_t *cr = curthread->td_ucred; 999 int ret; 1000 1001 ret = secpolicy_fs_unmount(cr, vfsp); 1002 if (ret) { 1003 ret = dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1004 ZFS_DELEG_PERM_MOUNT, cr); 1005 if (ret) 1006 return (ret); 1007 } 1008 /* 1009 * We purge the parent filesystem's vfsp as the parent filesystem 1010 * and all of its snapshots have their vnode's v_vfsp set to the 1011 * parent's filesystem's vfsp. Note, 'z_parent' is self 1012 * referential for non-snapshots. 1013 */ 1014 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1015 1016 /* 1017 * Unmount any snapshots mounted under .zfs before unmounting the 1018 * dataset itself. 1019 */ 1020 if (zfsvfs->z_ctldir != NULL) { 1021 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1022 return (ret); 1023 ret = vflush(vfsp, 0, 0, curthread); 1024 ASSERT(ret == EBUSY); 1025 if (!(fflag & MS_FORCE)) { 1026 if (zfsvfs->z_ctldir->v_count > 1) 1027 return (EBUSY); 1028 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1029 } 1030 zfsctl_destroy(zfsvfs); 1031 ASSERT(zfsvfs->z_ctldir == NULL); 1032 } 1033 1034 if (fflag & MS_FORCE) { 1035 /* 1036 * Mark file system as unmounted before calling 1037 * vflush(FORCECLOSE). This way we ensure no future vnops 1038 * will be called and risk operating on DOOMED vnodes. 1039 */ 1040 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1041 zfsvfs->z_unmounted = B_TRUE; 1042 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1043 } 1044 1045 /* 1046 * Flush all the files. 1047 */ 1048 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, curthread); 1049 if (ret != 0) { 1050 if (!zfsvfs->z_issnap) { 1051 zfsctl_create(zfsvfs); 1052 ASSERT(zfsvfs->z_ctldir != NULL); 1053 } 1054 return (ret); 1055 } 1056 1057 if (!(fflag & MS_FORCE)) { 1058 /* 1059 * Check the number of active vnodes in the file system. 1060 * Our count is maintained in the vfs structure, but the 1061 * number is off by 1 to indicate a hold on the vfs 1062 * structure itself. 1063 * 1064 * The '.zfs' directory maintains a reference of its 1065 * own, and any active references underneath are 1066 * reflected in the vnode count. 1067 */ 1068 if (zfsvfs->z_ctldir == NULL) { 1069 if (vfsp->vfs_count > 1) 1070 return (EBUSY); 1071 } else { 1072 if (vfsp->vfs_count > 2 || 1073 zfsvfs->z_ctldir->v_count > 1) 1074 return (EBUSY); 1075 } 1076 } else { 1077 MNT_ILOCK(vfsp); 1078 vfsp->mnt_kern_flag |= MNTK_UNMOUNTF; 1079 MNT_IUNLOCK(vfsp); 1080 } 1081 1082 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1083 os = zfsvfs->z_os; 1084 1085 /* 1086 * z_os will be NULL if there was an error in 1087 * attempting to reopen zfsvfs. 1088 */ 1089 if (os != NULL) { 1090 /* 1091 * Unset the objset user_ptr. 1092 */ 1093 mutex_enter(&os->os->os_user_ptr_lock); 1094 dmu_objset_set_user(os, NULL); 1095 mutex_exit(&os->os->os_user_ptr_lock); 1096 1097 /* 1098 * Finally release the objset 1099 */ 1100 dmu_objset_close(os); 1101 } 1102 1103 /* 1104 * We can now safely destroy the '.zfs' directory node. 1105 */ 1106 if (zfsvfs->z_ctldir != NULL) 1107 zfsctl_destroy(zfsvfs); 1108 if (zfsvfs->z_issnap) { 1109 vnode_t *svp = vfsp->mnt_vnodecovered; 1110 1111 if (svp->v_count >= 2) 1112 VN_RELE(svp); 1113 } 1114 zfs_freevfs(vfsp); 1115 1116 return (0); 1117} 1118 1119static int 1120zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 1121{ 1122 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1123 znode_t *zp; 1124 int err; 1125 1126 /* 1127 * XXXPJD: zfs_zget() can't operate on virtual entires like .zfs/ or 1128 * .zfs/snapshot/ directories, so for now just return EOPNOTSUPP. 1129 * This will make NFS to fall back to using READDIR instead of 1130 * READDIRPLUS. 1131 * Also snapshots are stored in AVL tree, but based on their names, 1132 * not inode numbers, so it will be very inefficient to iterate 1133 * over all snapshots to find the right one. 1134 * Note that OpenSolaris READDIRPLUS implementation does LOOKUP on 1135 * d_name, and not VGET on d_fileno as we do. 1136 */ 1137 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR) 1138 return (EOPNOTSUPP); 1139 1140 ZFS_ENTER(zfsvfs); 1141 err = zfs_zget(zfsvfs, ino, &zp); 1142 if (err == 0 && zp->z_unlinked) { 1143 VN_RELE(ZTOV(zp)); 1144 err = EINVAL; 1145 } 1146 if (err != 0) 1147 *vpp = NULL; 1148 else { 1149 *vpp = ZTOV(zp); 1150 vn_lock(*vpp, flags); 1151 } 1152 ZFS_EXIT(zfsvfs); 1153 return (err); 1154} 1155 1156static int 1157zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 1158 struct ucred **credanonp, int *numsecflavors, int **secflavors) 1159{ 1160 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1161 1162 /* 1163 * If this is regular file system vfsp is the same as 1164 * zfsvfs->z_parent->z_vfs, but if it is snapshot, 1165 * zfsvfs->z_parent->z_vfs represents parent file system 1166 * which we have to use here, because only this file system 1167 * has mnt_export configured. 1168 */ 1169 vfsp = zfsvfs->z_parent->z_vfs; 1170 1171 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 1172 credanonp, numsecflavors, secflavors)); 1173} 1174 1175CTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 1176CTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 1177 1178static int 1179zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, vnode_t **vpp) 1180{ 1181 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1182 znode_t *zp; 1183 uint64_t object = 0; 1184 uint64_t fid_gen = 0; 1185 uint64_t gen_mask; 1186 uint64_t zp_gen; 1187 int i, err; 1188 1189 *vpp = NULL; 1190 1191 ZFS_ENTER(zfsvfs); 1192 1193 /* 1194 * On FreeBSD we can get snapshot's mount point or its parent file 1195 * system mount point depending if snapshot is already mounted or not. 1196 */ 1197 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 1198 zfid_long_t *zlfid = (zfid_long_t *)fidp; 1199 uint64_t objsetid = 0; 1200 uint64_t setgen = 0; 1201 1202 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 1203 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 1204 1205 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 1206 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 1207 1208 ZFS_EXIT(zfsvfs); 1209 1210 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 1211 if (err) 1212 return (EINVAL); 1213 ZFS_ENTER(zfsvfs); 1214 } 1215 1216 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 1217 zfid_short_t *zfid = (zfid_short_t *)fidp; 1218 1219 for (i = 0; i < sizeof (zfid->zf_object); i++) 1220 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 1221 1222 for (i = 0; i < sizeof (zfid->zf_gen); i++) 1223 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 1224 } else { 1225 ZFS_EXIT(zfsvfs); 1226 return (EINVAL); 1227 } 1228 1229 /* A zero fid_gen means we are in the .zfs control directories */ 1230 if (fid_gen == 0 && 1231 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 1232 *vpp = zfsvfs->z_ctldir; 1233 ASSERT(*vpp != NULL); 1234 if (object == ZFSCTL_INO_SNAPDIR) { 1235 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 1236 0, NULL, NULL, NULL, NULL, NULL) == 0); 1237 } else { 1238 VN_HOLD(*vpp); 1239 } 1240 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1241 ZFS_EXIT(zfsvfs); 1242 return (0); 1243 } 1244 1245 gen_mask = -1ULL >> (64 - 8 * i); 1246 1247 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 1248 if (err = zfs_zget(zfsvfs, object, &zp)) { 1249 ZFS_EXIT(zfsvfs); 1250 return (err); 1251 } 1252 zp_gen = zp->z_phys->zp_gen & gen_mask; 1253 if (zp_gen == 0) 1254 zp_gen = 1; 1255 if (zp->z_unlinked || zp_gen != fid_gen) { 1256 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 1257 VN_RELE(ZTOV(zp)); 1258 ZFS_EXIT(zfsvfs); 1259 return (EINVAL); 1260 } 1261 1262 *vpp = ZTOV(zp); 1263 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1264 vnode_create_vobject(*vpp, zp->z_phys->zp_size, curthread); 1265 ZFS_EXIT(zfsvfs); 1266 return (0); 1267} 1268 1269/* 1270 * Block out VOPs and close zfsvfs_t::z_os 1271 * 1272 * Note, if successful, then we return with the 'z_teardown_lock' and 1273 * 'z_teardown_inactive_lock' write held. 1274 */ 1275int 1276zfs_suspend_fs(zfsvfs_t *zfsvfs, char *name, int *mode) 1277{ 1278 int error; 1279 1280 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 1281 return (error); 1282 1283 *mode = zfsvfs->z_os->os_mode; 1284 dmu_objset_name(zfsvfs->z_os, name); 1285 dmu_objset_close(zfsvfs->z_os); 1286 1287 return (0); 1288} 1289 1290/* 1291 * Reopen zfsvfs_t::z_os and release VOPs. 1292 */ 1293int 1294zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname, int mode) 1295{ 1296 int err; 1297 1298 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock)); 1299 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1300 1301 err = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os); 1302 if (err) { 1303 zfsvfs->z_os = NULL; 1304 } else { 1305 znode_t *zp; 1306 1307 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 1308 1309 /* 1310 * Attempt to re-establish all the active znodes with 1311 * their dbufs. If a zfs_rezget() fails, then we'll let 1312 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 1313 * when they try to use their znode. 1314 */ 1315 mutex_enter(&zfsvfs->z_znodes_lock); 1316 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 1317 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1318 (void) zfs_rezget(zp); 1319 } 1320 mutex_exit(&zfsvfs->z_znodes_lock); 1321 1322 } 1323 1324 /* release the VOPs */ 1325 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1326 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1327 1328 if (err) { 1329 /* 1330 * Since we couldn't reopen zfsvfs::z_os, force 1331 * unmount this file system. 1332 */ 1333 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) 1334 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 1335 } 1336 return (err); 1337} 1338 1339static void 1340zfs_freevfs(vfs_t *vfsp) 1341{ 1342 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1343 int i; 1344 1345 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1346 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1347 1348 zfs_fuid_destroy(zfsvfs); 1349 zfs_freezfsvfs(zfsvfs); 1350 1351 atomic_add_32(&zfs_active_fs_count, -1); 1352} 1353 1354#ifdef __i386__ 1355static int desiredvnodes_backup; 1356#endif 1357 1358static void 1359zfs_vnodes_adjust(void) 1360{ 1361#ifdef __i386__ 1362 int newdesiredvnodes; 1363 1364 desiredvnodes_backup = desiredvnodes; 1365 1366 /* 1367 * We calculate newdesiredvnodes the same way it is done in 1368 * vntblinit(). If it is equal to desiredvnodes, it means that 1369 * it wasn't tuned by the administrator and we can tune it down. 1370 */ 1371 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 1372 vm_kmem_size / (5 * (sizeof(struct vm_object) + 1373 sizeof(struct vnode)))); 1374 if (newdesiredvnodes == desiredvnodes) 1375 desiredvnodes = (3 * newdesiredvnodes) / 4; 1376#endif 1377} 1378 1379static void 1380zfs_vnodes_adjust_back(void) 1381{ 1382 1383#ifdef __i386__ 1384 desiredvnodes = desiredvnodes_backup; 1385#endif 1386} 1387 1388void 1389zfs_init(void) 1390{ 1391 1392 printf("ZFS filesystem version " ZPL_VERSION_STRING "\n"); 1393 1394 /* 1395 * Initialize znode cache, vnode ops, etc... 1396 */ 1397 zfs_znode_init(); 1398 1399 /* 1400 * Initialize .zfs directory structures 1401 */ 1402 zfsctl_init(); 1403 1404 /* 1405 * Reduce number of vnode. Originally number of vnodes is calculated 1406 * with UFS inode in mind. We reduce it here, because it's too big for 1407 * ZFS/i386. 1408 */ 1409 zfs_vnodes_adjust(); 1410} 1411 1412void 1413zfs_fini(void) 1414{ 1415 zfsctl_fini(); 1416 zfs_znode_fini(); 1417 zfs_vnodes_adjust_back(); 1418} 1419 1420int 1421zfs_busy(void) 1422{ 1423 return (zfs_active_fs_count != 0); 1424} 1425 1426int 1427zfs_set_version(const char *name, uint64_t newvers) 1428{ 1429 int error; 1430 objset_t *os; 1431 dmu_tx_t *tx; 1432 uint64_t curvers; 1433 1434 /* 1435 * XXX for now, require that the filesystem be unmounted. Would 1436 * be nice to find the zfsvfs_t and just update that if 1437 * possible. 1438 */ 1439 1440 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 1441 return (EINVAL); 1442 1443 error = dmu_objset_open(name, DMU_OST_ZFS, DS_MODE_OWNER, &os); 1444 if (error) 1445 return (error); 1446 1447 error = zap_lookup(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 1448 8, 1, &curvers); 1449 if (error) 1450 goto out; 1451 if (newvers < curvers) { 1452 error = EINVAL; 1453 goto out; 1454 } 1455 1456 tx = dmu_tx_create(os); 1457 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, 0, ZPL_VERSION_STR); 1458 error = dmu_tx_assign(tx, TXG_WAIT); 1459 if (error) { 1460 dmu_tx_abort(tx); 1461 goto out; 1462 } 1463 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 8, 1, 1464 &newvers, tx); 1465 1466 spa_history_internal_log(LOG_DS_UPGRADE, 1467 dmu_objset_spa(os), tx, CRED(), 1468 "oldver=%llu newver=%llu dataset = %llu", curvers, newvers, 1469 dmu_objset_id(os)); 1470 dmu_tx_commit(tx); 1471 1472out: 1473 dmu_objset_close(os); 1474 return (error); 1475} 1476/* 1477 * Read a property stored within the master node. 1478 */ 1479int 1480zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 1481{ 1482 const char *pname; 1483 int error = ENOENT; 1484 1485 /* 1486 * Look up the file system's value for the property. For the 1487 * version property, we look up a slightly different string. 1488 */ 1489 if (prop == ZFS_PROP_VERSION) 1490 pname = ZPL_VERSION_STR; 1491 else 1492 pname = zfs_prop_to_name(prop); 1493 1494 if (os != NULL) 1495 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 1496 1497 if (error == ENOENT) { 1498 /* No value set, use the default value */ 1499 switch (prop) { 1500 case ZFS_PROP_VERSION: 1501 *value = ZPL_VERSION; 1502 break; 1503 case ZFS_PROP_NORMALIZE: 1504 case ZFS_PROP_UTF8ONLY: 1505 *value = 0; 1506 break; 1507 case ZFS_PROP_CASE: 1508 *value = ZFS_CASE_SENSITIVE; 1509 break; 1510 default: 1511 return (error); 1512 } 1513 error = 0; 1514 } 1515 return (error); 1516} 1517