zfs_vfsops.c revision 191990
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_fhtovp(vfs_t *vfsp, fid_t *fidp, vnode_t **vpp); 101static void zfs_objset_close(zfsvfs_t *zfsvfs); 102static void zfs_freevfs(vfs_t *vfsp); 103 104static struct vfsops zfs_vfsops = { 105 .vfs_mount = zfs_mount, 106 .vfs_unmount = zfs_umount, 107 .vfs_root = zfs_root, 108 .vfs_statfs = zfs_statfs, 109 .vfs_vget = zfs_vget, 110 .vfs_sync = zfs_sync, 111 .vfs_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 int error; 141 142 error = vfs_stdsync(vfsp, waitfor); 143 if (error != 0) 144 return (error); 145 146 ZFS_ENTER(zfsvfs); 147 if (zfsvfs->z_log != NULL) 148 zil_commit(zfsvfs->z_log, UINT64_MAX, 0); 149 else 150 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 151 ZFS_EXIT(zfsvfs); 152 } else { 153 /* 154 * Sync all ZFS filesystems. This is what happens when you 155 * run sync(1M). Unlike other filesystems, ZFS honors the 156 * request by waiting for all pools to commit all dirty data. 157 */ 158 spa_sync_allpools(); 159 } 160 161 return (0); 162} 163 164static void 165atime_changed_cb(void *arg, uint64_t newval) 166{ 167 zfsvfs_t *zfsvfs = arg; 168 169 if (newval == TRUE) { 170 zfsvfs->z_atime = TRUE; 171 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; 172 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); 173 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); 174 } else { 175 zfsvfs->z_atime = FALSE; 176 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; 177 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); 178 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); 179 } 180} 181 182static void 183xattr_changed_cb(void *arg, uint64_t newval) 184{ 185 zfsvfs_t *zfsvfs = arg; 186 187 if (newval == TRUE) { 188 /* XXX locking on vfs_flag? */ 189#ifdef TODO 190 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; 191#endif 192 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); 193 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); 194 } else { 195 /* XXX locking on vfs_flag? */ 196#ifdef TODO 197 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR; 198#endif 199 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR); 200 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); 201 } 202} 203 204static void 205blksz_changed_cb(void *arg, uint64_t newval) 206{ 207 zfsvfs_t *zfsvfs = arg; 208 209 if (newval < SPA_MINBLOCKSIZE || 210 newval > SPA_MAXBLOCKSIZE || !ISP2(newval)) 211 newval = SPA_MAXBLOCKSIZE; 212 213 zfsvfs->z_max_blksz = newval; 214 zfsvfs->z_vfs->vfs_bsize = newval; 215} 216 217static void 218readonly_changed_cb(void *arg, uint64_t newval) 219{ 220 zfsvfs_t *zfsvfs = arg; 221 222 if (newval) { 223 /* XXX locking on vfs_flag? */ 224 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; 225 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); 226 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); 227 } else { 228 /* XXX locking on vfs_flag? */ 229 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 230 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); 231 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); 232 } 233} 234 235static void 236setuid_changed_cb(void *arg, uint64_t newval) 237{ 238 zfsvfs_t *zfsvfs = arg; 239 240 if (newval == FALSE) { 241 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; 242 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); 243 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); 244 } else { 245 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; 246 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); 247 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); 248 } 249} 250 251static void 252exec_changed_cb(void *arg, uint64_t newval) 253{ 254 zfsvfs_t *zfsvfs = arg; 255 256 if (newval == FALSE) { 257 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; 258 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); 259 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); 260 } else { 261 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; 262 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); 263 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); 264 } 265} 266 267/* 268 * The nbmand mount option can be changed at mount time. 269 * We can't allow it to be toggled on live file systems or incorrect 270 * behavior may be seen from cifs clients 271 * 272 * This property isn't registered via dsl_prop_register(), but this callback 273 * will be called when a file system is first mounted 274 */ 275static void 276nbmand_changed_cb(void *arg, uint64_t newval) 277{ 278 zfsvfs_t *zfsvfs = arg; 279 if (newval == FALSE) { 280 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); 281 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); 282 } else { 283 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); 284 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); 285 } 286} 287 288static void 289snapdir_changed_cb(void *arg, uint64_t newval) 290{ 291 zfsvfs_t *zfsvfs = arg; 292 293 zfsvfs->z_show_ctldir = newval; 294} 295 296static void 297vscan_changed_cb(void *arg, uint64_t newval) 298{ 299 zfsvfs_t *zfsvfs = arg; 300 301 zfsvfs->z_vscan = newval; 302} 303 304static void 305acl_mode_changed_cb(void *arg, uint64_t newval) 306{ 307 zfsvfs_t *zfsvfs = arg; 308 309 zfsvfs->z_acl_mode = newval; 310} 311 312static void 313acl_inherit_changed_cb(void *arg, uint64_t newval) 314{ 315 zfsvfs_t *zfsvfs = arg; 316 317 zfsvfs->z_acl_inherit = newval; 318} 319 320static int 321zfs_register_callbacks(vfs_t *vfsp) 322{ 323 struct dsl_dataset *ds = NULL; 324 objset_t *os = NULL; 325 zfsvfs_t *zfsvfs = NULL; 326 uint64_t nbmand; 327 int readonly, do_readonly = FALSE; 328 int setuid, do_setuid = FALSE; 329 int exec, do_exec = FALSE; 330 int xattr, do_xattr = FALSE; 331 int atime, do_atime = FALSE; 332 int error = 0; 333 334 ASSERT(vfsp); 335 zfsvfs = vfsp->vfs_data; 336 ASSERT(zfsvfs); 337 os = zfsvfs->z_os; 338 339 /* 340 * The act of registering our callbacks will destroy any mount 341 * options we may have. In order to enable temporary overrides 342 * of mount options, we stash away the current values and 343 * restore them after we register the callbacks. 344 */ 345 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) { 346 readonly = B_TRUE; 347 do_readonly = B_TRUE; 348 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 349 readonly = B_FALSE; 350 do_readonly = B_TRUE; 351 } 352 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 353 setuid = B_FALSE; 354 do_setuid = B_TRUE; 355 } else { 356 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 357 setuid = B_FALSE; 358 do_setuid = B_TRUE; 359 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 360 setuid = B_TRUE; 361 do_setuid = B_TRUE; 362 } 363 } 364 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 365 exec = B_FALSE; 366 do_exec = B_TRUE; 367 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 368 exec = B_TRUE; 369 do_exec = B_TRUE; 370 } 371 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 372 xattr = B_FALSE; 373 do_xattr = B_TRUE; 374 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 375 xattr = B_TRUE; 376 do_xattr = B_TRUE; 377 } 378 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 379 atime = B_FALSE; 380 do_atime = B_TRUE; 381 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 382 atime = B_TRUE; 383 do_atime = B_TRUE; 384 } 385 386 /* 387 * nbmand is a special property. It can only be changed at 388 * mount time. 389 * 390 * This is weird, but it is documented to only be changeable 391 * at mount time. 392 */ 393 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 394 nbmand = B_FALSE; 395 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 396 nbmand = B_TRUE; 397 } else { 398 char osname[MAXNAMELEN]; 399 400 dmu_objset_name(os, osname); 401 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand, 402 NULL)) { 403 return (error); 404 } 405 } 406 407 /* 408 * Register property callbacks. 409 * 410 * It would probably be fine to just check for i/o error from 411 * the first prop_register(), but I guess I like to go 412 * overboard... 413 */ 414 ds = dmu_objset_ds(os); 415 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs); 416 error = error ? error : dsl_prop_register(ds, 417 "xattr", xattr_changed_cb, zfsvfs); 418 error = error ? error : dsl_prop_register(ds, 419 "recordsize", blksz_changed_cb, zfsvfs); 420 error = error ? error : dsl_prop_register(ds, 421 "readonly", readonly_changed_cb, zfsvfs); 422 error = error ? error : dsl_prop_register(ds, 423 "setuid", setuid_changed_cb, zfsvfs); 424 error = error ? error : dsl_prop_register(ds, 425 "exec", exec_changed_cb, zfsvfs); 426 error = error ? error : dsl_prop_register(ds, 427 "snapdir", snapdir_changed_cb, zfsvfs); 428 error = error ? error : dsl_prop_register(ds, 429 "aclmode", acl_mode_changed_cb, zfsvfs); 430 error = error ? error : dsl_prop_register(ds, 431 "aclinherit", acl_inherit_changed_cb, zfsvfs); 432 error = error ? error : dsl_prop_register(ds, 433 "vscan", vscan_changed_cb, zfsvfs); 434 if (error) 435 goto unregister; 436 437 /* 438 * Invoke our callbacks to restore temporary mount options. 439 */ 440 if (do_readonly) 441 readonly_changed_cb(zfsvfs, readonly); 442 if (do_setuid) 443 setuid_changed_cb(zfsvfs, setuid); 444 if (do_exec) 445 exec_changed_cb(zfsvfs, exec); 446 if (do_xattr) 447 xattr_changed_cb(zfsvfs, xattr); 448 if (do_atime) 449 atime_changed_cb(zfsvfs, atime); 450 451 nbmand_changed_cb(zfsvfs, nbmand); 452 453 return (0); 454 455unregister: 456 /* 457 * We may attempt to unregister some callbacks that are not 458 * registered, but this is OK; it will simply return ENOMSG, 459 * which we will ignore. 460 */ 461 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs); 462 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs); 463 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs); 464 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs); 465 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs); 466 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs); 467 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs); 468 (void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs); 469 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb, 470 zfsvfs); 471 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs); 472 return (error); 473 474} 475 476static int 477zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 478{ 479 int error; 480 481 error = zfs_register_callbacks(zfsvfs->z_vfs); 482 if (error) 483 return (error); 484 485 /* 486 * Set the objset user_ptr to track its zfsvfs. 487 */ 488 mutex_enter(&zfsvfs->z_os->os->os_user_ptr_lock); 489 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 490 mutex_exit(&zfsvfs->z_os->os->os_user_ptr_lock); 491 492 /* 493 * If we are not mounting (ie: online recv), then we don't 494 * have to worry about replaying the log as we blocked all 495 * operations out since we closed the ZIL. 496 */ 497 if (mounting) { 498 boolean_t readonly; 499 500 /* 501 * During replay we remove the read only flag to 502 * allow replays to succeed. 503 */ 504 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 505 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 506 507 /* 508 * Parse and replay the intent log. 509 */ 510 zil_replay(zfsvfs->z_os, zfsvfs, &zfsvfs->z_assign, 511 zfs_replay_vector, zfs_unlinked_drain); 512 513 zfs_unlinked_drain(zfsvfs); 514 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 515 } 516 517 if (!zil_disable) 518 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 519 520 return (0); 521} 522 523static void 524zfs_freezfsvfs(zfsvfs_t *zfsvfs) 525{ 526 mutex_destroy(&zfsvfs->z_znodes_lock); 527 mutex_destroy(&zfsvfs->z_online_recv_lock); 528 list_destroy(&zfsvfs->z_all_znodes); 529 rrw_destroy(&zfsvfs->z_teardown_lock); 530 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 531 rw_destroy(&zfsvfs->z_fuid_lock); 532 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 533} 534 535static int 536zfs_domount(vfs_t *vfsp, char *osname) 537{ 538 uint64_t recordsize, readonly; 539 int error = 0; 540 int mode; 541 zfsvfs_t *zfsvfs; 542 znode_t *zp = NULL; 543 544 ASSERT(vfsp); 545 ASSERT(osname); 546 547 /* 548 * Initialize the zfs-specific filesystem structure. 549 * Should probably make this a kmem cache, shuffle fields, 550 * and just bzero up to z_hold_mtx[]. 551 */ 552 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 553 zfsvfs->z_vfs = vfsp; 554 zfsvfs->z_parent = zfsvfs; 555 zfsvfs->z_assign = TXG_NOWAIT; 556 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE; 557 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 558 559 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 560 mutex_init(&zfsvfs->z_online_recv_lock, NULL, MUTEX_DEFAULT, NULL); 561 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 562 offsetof(znode_t, z_link_node)); 563 rrw_init(&zfsvfs->z_teardown_lock); 564 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 565 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 566 567 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 568 NULL)) 569 goto out; 570 zfsvfs->z_vfs->vfs_bsize = recordsize; 571 572 vfsp->vfs_data = zfsvfs; 573 vfsp->mnt_flag |= MNT_LOCAL; 574 vfsp->mnt_kern_flag |= MNTK_MPSAFE; 575 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 576 577 if (error = dsl_prop_get_integer(osname, "readonly", &readonly, NULL)) 578 goto out; 579 580 mode = DS_MODE_OWNER; 581 if (readonly) 582 mode |= DS_MODE_READONLY; 583 584 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os); 585 if (error == EROFS) { 586 mode = DS_MODE_OWNER | DS_MODE_READONLY; 587 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, 588 &zfsvfs->z_os); 589 } 590 591 if (error) 592 goto out; 593 594 if (error = zfs_init_fs(zfsvfs, &zp)) 595 goto out; 596 597 /* 598 * Set features for file system. 599 */ 600 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 601 if (zfsvfs->z_use_fuids) { 602 vfs_set_feature(vfsp, VFSFT_XVATTR); 603 vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS); 604 vfs_set_feature(vfsp, VFSFT_ACEMASKONACCESS); 605 vfs_set_feature(vfsp, VFSFT_ACLONCREATE); 606 } 607 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 608 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 609 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 610 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 611 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 612 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 613 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 614 } 615 616 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 617 uint64_t pval; 618 619 ASSERT(mode & DS_MODE_READONLY); 620 atime_changed_cb(zfsvfs, B_FALSE); 621 readonly_changed_cb(zfsvfs, B_TRUE); 622 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 623 goto out; 624 xattr_changed_cb(zfsvfs, pval); 625 zfsvfs->z_issnap = B_TRUE; 626 } else { 627 error = zfsvfs_setup(zfsvfs, B_TRUE); 628 } 629 630 vfs_mountedfrom(vfsp, osname); 631 632 if (!zfsvfs->z_issnap) 633 zfsctl_create(zfsvfs); 634out: 635 if (error) { 636 if (zfsvfs->z_os) 637 dmu_objset_close(zfsvfs->z_os); 638 zfs_freezfsvfs(zfsvfs); 639 } else { 640 atomic_add_32(&zfs_active_fs_count, 1); 641 } 642 643 return (error); 644} 645 646void 647zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 648{ 649 objset_t *os = zfsvfs->z_os; 650 struct dsl_dataset *ds; 651 652 /* 653 * Unregister properties. 654 */ 655 if (!dmu_objset_is_snapshot(os)) { 656 ds = dmu_objset_ds(os); 657 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, 658 zfsvfs) == 0); 659 660 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, 661 zfsvfs) == 0); 662 663 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, 664 zfsvfs) == 0); 665 666 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, 667 zfsvfs) == 0); 668 669 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, 670 zfsvfs) == 0); 671 672 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, 673 zfsvfs) == 0); 674 675 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, 676 zfsvfs) == 0); 677 678 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, 679 zfsvfs) == 0); 680 681 VERIFY(dsl_prop_unregister(ds, "aclinherit", 682 acl_inherit_changed_cb, zfsvfs) == 0); 683 684 VERIFY(dsl_prop_unregister(ds, "vscan", 685 vscan_changed_cb, zfsvfs) == 0); 686 } 687} 688 689/*ARGSUSED*/ 690static int 691zfs_mount(vfs_t *vfsp) 692{ 693 kthread_t *td = curthread; 694 vnode_t *mvp = vfsp->mnt_vnodecovered; 695 cred_t *cr = td->td_ucred; 696 char *osname; 697 int error = 0; 698 int canwrite; 699 700 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 701 return (EINVAL); 702 703 /* 704 * If full-owner-access is enabled and delegated administration is 705 * turned on, we must set nosuid. 706 */ 707 if (zfs_super_owner && 708 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 709 secpolicy_fs_mount_clearopts(cr, vfsp); 710 } 711 712 /* 713 * Check for mount privilege? 714 * 715 * If we don't have privilege then see if 716 * we have local permission to allow it 717 */ 718 error = secpolicy_fs_mount(cr, mvp, vfsp); 719 if (error) { 720 error = dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr); 721 if (error == 0) { 722 vattr_t vattr; 723 724 /* 725 * Make sure user is the owner of the mount point 726 * or has sufficient privileges. 727 */ 728 729 vattr.va_mask = AT_UID; 730 731 if (error = VOP_GETATTR(mvp, &vattr, cr)) { 732 goto out; 733 } 734 735#if 0 /* CHECK THIS! Is probably needed for zfs_suser. */ 736 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 737 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 738 error = EPERM; 739 goto out; 740 } 741#else 742 if (error = secpolicy_vnode_owner(mvp, cr, vattr.va_uid)) { 743 goto out; 744 } 745 746 if (error = VOP_ACCESS(mvp, VWRITE, cr, td)) { 747 goto out; 748 } 749#endif 750 751 secpolicy_fs_mount_clearopts(cr, vfsp); 752 } else { 753 goto out; 754 } 755 } 756 757 /* 758 * Refuse to mount a filesystem if we are in a local zone and the 759 * dataset is not visible. 760 */ 761 if (!INGLOBALZONE(curthread) && 762 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 763 error = EPERM; 764 goto out; 765 } 766 767 /* 768 * When doing a remount, we simply refresh our temporary properties 769 * according to those options set in the current VFS options. 770 */ 771 if (vfsp->vfs_flag & MS_REMOUNT) { 772 /* refresh mount options */ 773 zfs_unregister_callbacks(vfsp->vfs_data); 774 error = zfs_register_callbacks(vfsp); 775 goto out; 776 } 777 778 DROP_GIANT(); 779 error = zfs_domount(vfsp, osname); 780 PICKUP_GIANT(); 781out: 782 return (error); 783} 784 785static int 786zfs_statfs(vfs_t *vfsp, struct statfs *statp) 787{ 788 zfsvfs_t *zfsvfs = vfsp->vfs_data; 789 uint64_t refdbytes, availbytes, usedobjs, availobjs; 790 791 statp->f_version = STATFS_VERSION; 792 793 ZFS_ENTER(zfsvfs); 794 795 dmu_objset_space(zfsvfs->z_os, 796 &refdbytes, &availbytes, &usedobjs, &availobjs); 797 798 /* 799 * The underlying storage pool actually uses multiple block sizes. 800 * We report the fragsize as the smallest block size we support, 801 * and we report our blocksize as the filesystem's maximum blocksize. 802 */ 803 statp->f_bsize = zfsvfs->z_vfs->vfs_bsize; 804 statp->f_iosize = zfsvfs->z_vfs->vfs_bsize; 805 806 /* 807 * The following report "total" blocks of various kinds in the 808 * file system, but reported in terms of f_frsize - the 809 * "fragment" size. 810 */ 811 812 statp->f_blocks = (refdbytes + availbytes) / statp->f_bsize; 813 statp->f_bfree = availbytes / statp->f_bsize; 814 statp->f_bavail = statp->f_bfree; /* no root reservation */ 815 816 /* 817 * statvfs() should really be called statufs(), because it assumes 818 * static metadata. ZFS doesn't preallocate files, so the best 819 * we can do is report the max that could possibly fit in f_files, 820 * and that minus the number actually used in f_ffree. 821 * For f_ffree, report the smaller of the number of object available 822 * and the number of blocks (each object will take at least a block). 823 */ 824 statp->f_ffree = MIN(availobjs, statp->f_bfree); 825 statp->f_files = statp->f_ffree + usedobjs; 826 827 /* 828 * We're a zfs filesystem. 829 */ 830 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 831 832 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 833 sizeof(statp->f_mntfromname)); 834 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 835 sizeof(statp->f_mntonname)); 836 837 statp->f_namemax = ZFS_MAXNAMELEN; 838 839 ZFS_EXIT(zfsvfs); 840 return (0); 841} 842 843static int 844zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 845{ 846 zfsvfs_t *zfsvfs = vfsp->vfs_data; 847 znode_t *rootzp; 848 int error; 849 850 ZFS_ENTER(zfsvfs); 851 852 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 853 if (error == 0) { 854 *vpp = ZTOV(rootzp); 855 error = vn_lock(*vpp, flags); 856 (*vpp)->v_vflag |= VV_ROOT; 857 } 858 859 ZFS_EXIT(zfsvfs); 860 return (error); 861} 862 863/* 864 * Teardown the zfsvfs::z_os. 865 * 866 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 867 * and 'z_teardown_inactive_lock' held. 868 */ 869static int 870zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 871{ 872 znode_t *zp; 873 874 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 875 876 if (!unmounting) { 877 /* 878 * We purge the parent filesystem's vfsp as the parent 879 * filesystem and all of its snapshots have their vnode's 880 * v_vfsp set to the parent's filesystem's vfsp. Note, 881 * 'z_parent' is self referential for non-snapshots. 882 */ 883 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 884 } 885 886 /* 887 * Close the zil. NB: Can't close the zil while zfs_inactive 888 * threads are blocked as zil_close can call zfs_inactive. 889 */ 890 if (zfsvfs->z_log) { 891 zil_close(zfsvfs->z_log); 892 zfsvfs->z_log = NULL; 893 } 894 895 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 896 897 /* 898 * If we are not unmounting (ie: online recv) and someone already 899 * unmounted this file system while we were doing the switcheroo, 900 * or a reopen of z_os failed then just bail out now. 901 */ 902 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 903 rw_exit(&zfsvfs->z_teardown_inactive_lock); 904 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 905 return (EIO); 906 } 907 908 /* 909 * At this point there are no vops active, and any new vops will 910 * fail with EIO since we have z_teardown_lock for writer (only 911 * relavent for forced unmount). 912 * 913 * Release all holds on dbufs. 914 */ 915 mutex_enter(&zfsvfs->z_znodes_lock); 916 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 917 zp = list_next(&zfsvfs->z_all_znodes, zp)) 918 if (zp->z_dbuf) { 919 ASSERT(ZTOV(zp)->v_count > 0); 920 zfs_znode_dmu_fini(zp); 921 } 922 mutex_exit(&zfsvfs->z_znodes_lock); 923 924 /* 925 * If we are unmounting, set the unmounted flag and let new vops 926 * unblock. zfs_inactive will have the unmounted behavior, and all 927 * other vops will fail with EIO. 928 */ 929 if (unmounting) { 930 zfsvfs->z_unmounted = B_TRUE; 931 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 932 rw_exit(&zfsvfs->z_teardown_inactive_lock); 933 } 934 935 /* 936 * z_os will be NULL if there was an error in attempting to reopen 937 * zfsvfs, so just return as the properties had already been 938 * unregistered and cached data had been evicted before. 939 */ 940 if (zfsvfs->z_os == NULL) 941 return (0); 942 943 /* 944 * Unregister properties. 945 */ 946 zfs_unregister_callbacks(zfsvfs); 947 948 /* 949 * Evict cached data 950 */ 951 if (dmu_objset_evict_dbufs(zfsvfs->z_os)) { 952 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 953 (void) dmu_objset_evict_dbufs(zfsvfs->z_os); 954 } 955 956 return (0); 957} 958 959/*ARGSUSED*/ 960static int 961zfs_umount(vfs_t *vfsp, int fflag) 962{ 963 zfsvfs_t *zfsvfs = vfsp->vfs_data; 964 objset_t *os; 965 cred_t *cr = curthread->td_ucred; 966 int ret; 967 968 if (fflag & MS_FORCE) { 969 /* TODO: Force unmount is not well implemented yet, so deny it. */ 970 ZFS_LOG(0, "Force unmount is not supported, removing FORCE flag."); 971 fflag &= ~MS_FORCE; 972 } 973 974 ret = secpolicy_fs_unmount(cr, vfsp); 975 if (ret) { 976 ret = dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 977 ZFS_DELEG_PERM_MOUNT, cr); 978 if (ret) 979 return (ret); 980 } 981 /* 982 * We purge the parent filesystem's vfsp as the parent filesystem 983 * and all of its snapshots have their vnode's v_vfsp set to the 984 * parent's filesystem's vfsp. Note, 'z_parent' is self 985 * referential for non-snapshots. 986 */ 987 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 988 989 /* 990 * Unmount any snapshots mounted under .zfs before unmounting the 991 * dataset itself. 992 */ 993 if (zfsvfs->z_ctldir != NULL) { 994 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 995 return (ret); 996 ret = vflush(vfsp, 0, 0, curthread); 997 ASSERT(ret == EBUSY); 998 if (!(fflag & MS_FORCE)) { 999 if (zfsvfs->z_ctldir->v_count > 1) 1000 return (EBUSY); 1001 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1002 } 1003 zfsctl_destroy(zfsvfs); 1004 ASSERT(zfsvfs->z_ctldir == NULL); 1005 } 1006 1007 /* 1008 * Flush all the files. 1009 */ 1010 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, curthread); 1011 if (ret != 0) { 1012 if (!zfsvfs->z_issnap) { 1013 zfsctl_create(zfsvfs); 1014 ASSERT(zfsvfs->z_ctldir != NULL); 1015 } 1016 return (ret); 1017 } 1018 1019 if (!(fflag & MS_FORCE)) { 1020 /* 1021 * Check the number of active vnodes in the file system. 1022 * Our count is maintained in the vfs structure, but the 1023 * number is off by 1 to indicate a hold on the vfs 1024 * structure itself. 1025 * 1026 * The '.zfs' directory maintains a reference of its 1027 * own, and any active references underneath are 1028 * reflected in the vnode count. 1029 */ 1030 if (zfsvfs->z_ctldir == NULL) { 1031 if (vfsp->vfs_count > 1) 1032 return (EBUSY); 1033 } else { 1034 if (vfsp->vfs_count > 2 || 1035 zfsvfs->z_ctldir->v_count > 1) 1036 return (EBUSY); 1037 } 1038 } else { 1039 MNT_ILOCK(vfsp); 1040 vfsp->mnt_kern_flag |= MNTK_UNMOUNTF; 1041 MNT_IUNLOCK(vfsp); 1042 } 1043 1044 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1045 os = zfsvfs->z_os; 1046 1047 /* 1048 * z_os will be NULL if there was an error in 1049 * attempting to reopen zfsvfs. 1050 */ 1051 if (os != NULL) { 1052 /* 1053 * Unset the objset user_ptr. 1054 */ 1055 mutex_enter(&os->os->os_user_ptr_lock); 1056 dmu_objset_set_user(os, NULL); 1057 mutex_exit(&os->os->os_user_ptr_lock); 1058 1059 /* 1060 * Finally release the objset 1061 */ 1062 dmu_objset_close(os); 1063 } 1064 1065 /* 1066 * We can now safely destroy the '.zfs' directory node. 1067 */ 1068 if (zfsvfs->z_ctldir != NULL) 1069 zfsctl_destroy(zfsvfs); 1070 if (zfsvfs->z_issnap) { 1071 vnode_t *svp = vfsp->mnt_vnodecovered; 1072 1073 ASSERT(svp->v_count == 2); 1074 VN_RELE(svp); 1075 } 1076 zfs_freevfs(vfsp); 1077 1078 return (0); 1079} 1080 1081static int 1082zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 1083{ 1084 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1085 znode_t *zp; 1086 int err; 1087 1088 ZFS_ENTER(zfsvfs); 1089 err = zfs_zget(zfsvfs, ino, &zp); 1090 if (err == 0 && zp->z_unlinked) { 1091 VN_RELE(ZTOV(zp)); 1092 err = EINVAL; 1093 } 1094 if (err != 0) 1095 *vpp = NULL; 1096 else { 1097 *vpp = ZTOV(zp); 1098 vn_lock(*vpp, flags); 1099 } 1100 ZFS_EXIT(zfsvfs); 1101 return (err); 1102} 1103 1104static int 1105zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, vnode_t **vpp) 1106{ 1107 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1108 znode_t *zp; 1109 uint64_t object = 0; 1110 uint64_t fid_gen = 0; 1111 uint64_t gen_mask; 1112 uint64_t zp_gen; 1113 int i, err; 1114 1115 *vpp = NULL; 1116 1117 ZFS_ENTER(zfsvfs); 1118 1119 if (fidp->fid_len == LONG_FID_LEN) { 1120 zfid_long_t *zlfid = (zfid_long_t *)fidp; 1121 uint64_t objsetid = 0; 1122 uint64_t setgen = 0; 1123 1124 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 1125 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 1126 1127 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 1128 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 1129 1130 ZFS_EXIT(zfsvfs); 1131 1132 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 1133 if (err) 1134 return (EINVAL); 1135 ZFS_ENTER(zfsvfs); 1136 } 1137 1138 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 1139 zfid_short_t *zfid = (zfid_short_t *)fidp; 1140 1141 for (i = 0; i < sizeof (zfid->zf_object); i++) 1142 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 1143 1144 for (i = 0; i < sizeof (zfid->zf_gen); i++) 1145 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 1146 } else { 1147 ZFS_EXIT(zfsvfs); 1148 return (EINVAL); 1149 } 1150 1151 /* A zero fid_gen means we are in the .zfs control directories */ 1152 if (fid_gen == 0 && 1153 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 1154 *vpp = zfsvfs->z_ctldir; 1155 ASSERT(*vpp != NULL); 1156 if (object == ZFSCTL_INO_SNAPDIR) { 1157 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 1158 0, NULL, NULL, NULL, NULL, NULL) == 0); 1159 } else { 1160 VN_HOLD(*vpp); 1161 } 1162 ZFS_EXIT(zfsvfs); 1163 /* XXX: LK_RETRY? */ 1164 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1165 return (0); 1166 } 1167 1168 gen_mask = -1ULL >> (64 - 8 * i); 1169 1170 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 1171 if (err = zfs_zget(zfsvfs, object, &zp)) { 1172 ZFS_EXIT(zfsvfs); 1173 return (err); 1174 } 1175 zp_gen = zp->z_phys->zp_gen & gen_mask; 1176 if (zp_gen == 0) 1177 zp_gen = 1; 1178 if (zp->z_unlinked || zp_gen != fid_gen) { 1179 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 1180 VN_RELE(ZTOV(zp)); 1181 ZFS_EXIT(zfsvfs); 1182 return (EINVAL); 1183 } 1184 1185 *vpp = ZTOV(zp); 1186 /* XXX: LK_RETRY? */ 1187 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1188 vnode_create_vobject(*vpp, zp->z_phys->zp_size, curthread); 1189 ZFS_EXIT(zfsvfs); 1190 return (0); 1191} 1192 1193/* 1194 * Block out VOPs and close zfsvfs_t::z_os 1195 * 1196 * Note, if successful, then we return with the 'z_teardown_lock' and 1197 * 'z_teardown_inactive_lock' write held. 1198 */ 1199int 1200zfs_suspend_fs(zfsvfs_t *zfsvfs, char *name, int *mode) 1201{ 1202 int error; 1203 1204 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 1205 return (error); 1206 1207 *mode = zfsvfs->z_os->os_mode; 1208 dmu_objset_name(zfsvfs->z_os, name); 1209 dmu_objset_close(zfsvfs->z_os); 1210 1211 return (0); 1212} 1213 1214/* 1215 * Reopen zfsvfs_t::z_os and release VOPs. 1216 */ 1217int 1218zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname, int mode) 1219{ 1220 int err; 1221 1222 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock)); 1223 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1224 1225 err = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os); 1226 if (err) { 1227 zfsvfs->z_os = NULL; 1228 } else { 1229 znode_t *zp; 1230 1231 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 1232 1233 /* 1234 * Attempt to re-establish all the active znodes with 1235 * their dbufs. If a zfs_rezget() fails, then we'll let 1236 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 1237 * when they try to use their znode. 1238 */ 1239 mutex_enter(&zfsvfs->z_znodes_lock); 1240 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 1241 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1242 (void) zfs_rezget(zp); 1243 } 1244 mutex_exit(&zfsvfs->z_znodes_lock); 1245 1246 } 1247 1248 /* release the VOPs */ 1249 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1250 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1251 1252 if (err) { 1253 /* 1254 * Since we couldn't reopen zfsvfs::z_os, force 1255 * unmount this file system. 1256 */ 1257 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) 1258 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 1259 } 1260 return (err); 1261} 1262 1263static void 1264zfs_freevfs(vfs_t *vfsp) 1265{ 1266 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1267 int i; 1268 1269 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1270 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1271 1272 zfs_fuid_destroy(zfsvfs); 1273 zfs_freezfsvfs(zfsvfs); 1274 1275 atomic_add_32(&zfs_active_fs_count, -1); 1276} 1277 1278#ifdef __i386__ 1279static int desiredvnodes_backup; 1280#endif 1281 1282static void 1283zfs_vnodes_adjust(void) 1284{ 1285#ifdef __i386__ 1286 int newdesiredvnodes; 1287 1288 desiredvnodes_backup = desiredvnodes; 1289 1290 /* 1291 * We calculate newdesiredvnodes the same way it is done in 1292 * vntblinit(). If it is equal to desiredvnodes, it means that 1293 * it wasn't tuned by the administrator and we can tune it down. 1294 */ 1295 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 1296 vm_kmem_size / (5 * (sizeof(struct vm_object) + 1297 sizeof(struct vnode)))); 1298 if (newdesiredvnodes == desiredvnodes) 1299 desiredvnodes = (3 * newdesiredvnodes) / 4; 1300#endif 1301} 1302 1303static void 1304zfs_vnodes_adjust_back(void) 1305{ 1306 1307#ifdef __i386__ 1308 desiredvnodes = desiredvnodes_backup; 1309#endif 1310} 1311 1312void 1313zfs_init(void) 1314{ 1315 1316 printf("ZFS filesystem version " SPA_VERSION_STRING "\n"); 1317 1318 /* 1319 * Initialize znode cache, vnode ops, etc... 1320 */ 1321 zfs_znode_init(); 1322 1323 /* 1324 * Initialize .zfs directory structures 1325 */ 1326 zfsctl_init(); 1327 1328 /* 1329 * Reduce number of vnode. Originally number of vnodes is calculated 1330 * with UFS inode in mind. We reduce it here, because it's too big for 1331 * ZFS/i386. 1332 */ 1333 zfs_vnodes_adjust(); 1334} 1335 1336void 1337zfs_fini(void) 1338{ 1339 zfsctl_fini(); 1340 zfs_znode_fini(); 1341 zfs_vnodes_adjust_back(); 1342} 1343 1344int 1345zfs_busy(void) 1346{ 1347 return (zfs_active_fs_count != 0); 1348} 1349 1350int 1351zfs_set_version(const char *name, uint64_t newvers) 1352{ 1353 int error; 1354 objset_t *os; 1355 dmu_tx_t *tx; 1356 uint64_t curvers; 1357 1358 /* 1359 * XXX for now, require that the filesystem be unmounted. Would 1360 * be nice to find the zfsvfs_t and just update that if 1361 * possible. 1362 */ 1363 1364 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 1365 return (EINVAL); 1366 1367 error = dmu_objset_open(name, DMU_OST_ZFS, DS_MODE_OWNER, &os); 1368 if (error) 1369 return (error); 1370 1371 error = zap_lookup(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 1372 8, 1, &curvers); 1373 if (error) 1374 goto out; 1375 if (newvers < curvers) { 1376 error = EINVAL; 1377 goto out; 1378 } 1379 1380 tx = dmu_tx_create(os); 1381 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, 0, ZPL_VERSION_STR); 1382 error = dmu_tx_assign(tx, TXG_WAIT); 1383 if (error) { 1384 dmu_tx_abort(tx); 1385 goto out; 1386 } 1387 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 8, 1, 1388 &newvers, tx); 1389 1390 spa_history_internal_log(LOG_DS_UPGRADE, 1391 dmu_objset_spa(os), tx, CRED(), 1392 "oldver=%llu newver=%llu dataset = %llu", curvers, newvers, 1393 dmu_objset_id(os)); 1394 dmu_tx_commit(tx); 1395 1396out: 1397 dmu_objset_close(os); 1398 return (error); 1399} 1400/* 1401 * Read a property stored within the master node. 1402 */ 1403int 1404zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 1405{ 1406 const char *pname; 1407 int error = ENOENT; 1408 1409 /* 1410 * Look up the file system's value for the property. For the 1411 * version property, we look up a slightly different string. 1412 */ 1413 if (prop == ZFS_PROP_VERSION) 1414 pname = ZPL_VERSION_STR; 1415 else 1416 pname = zfs_prop_to_name(prop); 1417 1418 if (os != NULL) 1419 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 1420 1421 if (error == ENOENT) { 1422 /* No value set, use the default value */ 1423 switch (prop) { 1424 case ZFS_PROP_VERSION: 1425 *value = ZPL_VERSION; 1426 break; 1427 case ZFS_PROP_NORMALIZE: 1428 case ZFS_PROP_UTF8ONLY: 1429 *value = 0; 1430 break; 1431 case ZFS_PROP_CASE: 1432 *value = ZFS_CASE_SENSITIVE; 1433 break; 1434 default: 1435 return (error); 1436 } 1437 error = 0; 1438 } 1439 return (error); 1440} 1441