zfs_vfsops.c revision 196978
1235368Sgnn/* 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 * This function can be called for a snapshot when we update snapshot's 341 * mount point, which isn't really supported. 342 */ 343 if (dmu_objset_is_snapshot(os)) 344 return (EOPNOTSUPP); 345 346 /* 347 * The act of registering our callbacks will destroy any mount 348 * options we may have. In order to enable temporary overrides 349 * of mount options, we stash away the current values and 350 * restore them after we register the callbacks. 351 */ 352 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) { 353 readonly = B_TRUE; 354 do_readonly = B_TRUE; 355 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 356 readonly = B_FALSE; 357 do_readonly = B_TRUE; 358 } 359 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 360 setuid = B_FALSE; 361 do_setuid = B_TRUE; 362 } else { 363 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 364 setuid = B_FALSE; 365 do_setuid = B_TRUE; 366 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 367 setuid = B_TRUE; 368 do_setuid = B_TRUE; 369 } 370 } 371 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 372 exec = B_FALSE; 373 do_exec = B_TRUE; 374 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 375 exec = B_TRUE; 376 do_exec = B_TRUE; 377 } 378 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 379 xattr = B_FALSE; 380 do_xattr = B_TRUE; 381 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 382 xattr = B_TRUE; 383 do_xattr = B_TRUE; 384 } 385 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 386 atime = B_FALSE; 387 do_atime = B_TRUE; 388 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 389 atime = B_TRUE; 390 do_atime = B_TRUE; 391 } 392 393 /* 394 * nbmand is a special property. It can only be changed at 395 * mount time. 396 * 397 * This is weird, but it is documented to only be changeable 398 * at mount time. 399 */ 400 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 401 nbmand = B_FALSE; 402 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 403 nbmand = B_TRUE; 404 } else { 405 char osname[MAXNAMELEN]; 406 407 dmu_objset_name(os, osname); 408 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand, 409 NULL)) { 410 return (error); 411 } 412 } 413 414 /* 415 * Register property callbacks. 416 * 417 * It would probably be fine to just check for i/o error from 418 * the first prop_register(), but I guess I like to go 419 * overboard... 420 */ 421 ds = dmu_objset_ds(os); 422 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs); 423 error = error ? error : dsl_prop_register(ds, 424 "xattr", xattr_changed_cb, zfsvfs); 425 error = error ? error : dsl_prop_register(ds, 426 "recordsize", blksz_changed_cb, zfsvfs); 427 error = error ? error : dsl_prop_register(ds, 428 "readonly", readonly_changed_cb, zfsvfs); 429 error = error ? error : dsl_prop_register(ds, 430 "setuid", setuid_changed_cb, zfsvfs); 431 error = error ? error : dsl_prop_register(ds, 432 "exec", exec_changed_cb, zfsvfs); 433 error = error ? error : dsl_prop_register(ds, 434 "snapdir", snapdir_changed_cb, zfsvfs); 435 error = error ? error : dsl_prop_register(ds, 436 "aclmode", acl_mode_changed_cb, zfsvfs); 437 error = error ? error : dsl_prop_register(ds, 438 "aclinherit", acl_inherit_changed_cb, zfsvfs); 439 error = error ? error : dsl_prop_register(ds, 440 "vscan", vscan_changed_cb, zfsvfs); 441 if (error) 442 goto unregister; 443 444 /* 445 * Invoke our callbacks to restore temporary mount options. 446 */ 447 if (do_readonly) 448 readonly_changed_cb(zfsvfs, readonly); 449 if (do_setuid) 450 setuid_changed_cb(zfsvfs, setuid); 451 if (do_exec) 452 exec_changed_cb(zfsvfs, exec); 453 if (do_xattr) 454 xattr_changed_cb(zfsvfs, xattr); 455 if (do_atime) 456 atime_changed_cb(zfsvfs, atime); 457 458 nbmand_changed_cb(zfsvfs, nbmand); 459 460 return (0); 461 462unregister: 463 /* 464 * We may attempt to unregister some callbacks that are not 465 * registered, but this is OK; it will simply return ENOMSG, 466 * which we will ignore. 467 */ 468 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs); 469 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs); 470 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs); 471 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs); 472 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs); 473 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs); 474 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs); 475 (void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs); 476 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb, 477 zfsvfs); 478 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs); 479 return (error); 480 481} 482 483static int 484zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 485{ 486 int error; 487 488 error = zfs_register_callbacks(zfsvfs->z_vfs); 489 if (error) 490 return (error); 491 492 /* 493 * Set the objset user_ptr to track its zfsvfs. 494 */ 495 mutex_enter(&zfsvfs->z_os->os->os_user_ptr_lock); 496 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 497 mutex_exit(&zfsvfs->z_os->os->os_user_ptr_lock); 498 499 /* 500 * If we are not mounting (ie: online recv), then we don't 501 * have to worry about replaying the log as we blocked all 502 * operations out since we closed the ZIL. 503 */ 504 if (mounting) { 505 boolean_t readonly; 506 507 /* 508 * During replay we remove the read only flag to 509 * allow replays to succeed. 510 */ 511 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 512 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 513 514 /* 515 * Parse and replay the intent log. 516 */ 517 zil_replay(zfsvfs->z_os, zfsvfs, &zfsvfs->z_assign, 518 zfs_replay_vector, zfs_unlinked_drain); 519 520 zfs_unlinked_drain(zfsvfs); 521 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 522 } 523 524 if (!zil_disable) 525 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 526 527 return (0); 528} 529 530static void 531zfs_freezfsvfs(zfsvfs_t *zfsvfs) 532{ 533 mutex_destroy(&zfsvfs->z_znodes_lock); 534 mutex_destroy(&zfsvfs->z_online_recv_lock); 535 list_destroy(&zfsvfs->z_all_znodes); 536 rrw_destroy(&zfsvfs->z_teardown_lock); 537 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 538 rw_destroy(&zfsvfs->z_fuid_lock); 539 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 540} 541 542static int 543zfs_domount(vfs_t *vfsp, char *osname) 544{ 545 uint64_t recordsize, readonly; 546 int error = 0; 547 int mode; 548 zfsvfs_t *zfsvfs; 549 znode_t *zp = NULL; 550 551 ASSERT(vfsp); 552 ASSERT(osname); 553 554 /* 555 * Initialize the zfs-specific filesystem structure. 556 * Should probably make this a kmem cache, shuffle fields, 557 * and just bzero up to z_hold_mtx[]. 558 */ 559 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 560 zfsvfs->z_vfs = vfsp; 561 zfsvfs->z_parent = zfsvfs; 562 zfsvfs->z_assign = TXG_NOWAIT; 563 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE; 564 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 565 566 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 567 mutex_init(&zfsvfs->z_online_recv_lock, NULL, MUTEX_DEFAULT, NULL); 568 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 569 offsetof(znode_t, z_link_node)); 570 rrw_init(&zfsvfs->z_teardown_lock); 571 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 572 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 573 574 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 575 NULL)) 576 goto out; 577 zfsvfs->z_vfs->vfs_bsize = recordsize; 578 579 vfsp->vfs_data = zfsvfs; 580 vfsp->mnt_flag |= MNT_LOCAL; 581 vfsp->mnt_kern_flag |= MNTK_MPSAFE; 582 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 583 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 584 585 if (error = dsl_prop_get_integer(osname, "readonly", &readonly, NULL)) 586 goto out; 587 588 mode = DS_MODE_OWNER; 589 if (readonly) 590 mode |= DS_MODE_READONLY; 591 592 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os); 593 if (error == EROFS) { 594 mode = DS_MODE_OWNER | DS_MODE_READONLY; 595 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, 596 &zfsvfs->z_os); 597 } 598 599 if (error) 600 goto out; 601 602 if (error = zfs_init_fs(zfsvfs, &zp)) 603 goto out; 604 605 /* 606 * Set features for file system. 607 */ 608 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 609 if (zfsvfs->z_use_fuids) { 610 vfs_set_feature(vfsp, VFSFT_XVATTR); 611 vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS); 612 vfs_set_feature(vfsp, VFSFT_ACEMASKONACCESS); 613 vfs_set_feature(vfsp, VFSFT_ACLONCREATE); 614 } 615 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 616 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 617 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 618 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 619 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 620 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 621 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 622 } 623 624 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 625 uint64_t pval; 626 627 ASSERT(mode & DS_MODE_READONLY); 628 atime_changed_cb(zfsvfs, B_FALSE); 629 readonly_changed_cb(zfsvfs, B_TRUE); 630 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 631 goto out; 632 xattr_changed_cb(zfsvfs, pval); 633 zfsvfs->z_issnap = B_TRUE; 634 } else { 635 error = zfsvfs_setup(zfsvfs, B_TRUE); 636 } 637 638 vfs_mountedfrom(vfsp, osname); 639 640 if (!zfsvfs->z_issnap) 641 zfsctl_create(zfsvfs); 642out: 643 if (error) { 644 if (zfsvfs->z_os) 645 dmu_objset_close(zfsvfs->z_os); 646 zfs_freezfsvfs(zfsvfs); 647 } else { 648 atomic_add_32(&zfs_active_fs_count, 1); 649 } 650 651 return (error); 652} 653 654void 655zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 656{ 657 objset_t *os = zfsvfs->z_os; 658 struct dsl_dataset *ds; 659 660 /* 661 * Unregister properties. 662 */ 663 if (!dmu_objset_is_snapshot(os)) { 664 ds = dmu_objset_ds(os); 665 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, 666 zfsvfs) == 0); 667 668 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, 669 zfsvfs) == 0); 670 671 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, 672 zfsvfs) == 0); 673 674 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, 675 zfsvfs) == 0); 676 677 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, 678 zfsvfs) == 0); 679 680 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, 681 zfsvfs) == 0); 682 683 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, 684 zfsvfs) == 0); 685 686 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, 687 zfsvfs) == 0); 688 689 VERIFY(dsl_prop_unregister(ds, "aclinherit", 690 acl_inherit_changed_cb, zfsvfs) == 0); 691 692 VERIFY(dsl_prop_unregister(ds, "vscan", 693 vscan_changed_cb, zfsvfs) == 0); 694 } 695} 696 697/*ARGSUSED*/ 698static int 699zfs_mount(vfs_t *vfsp) 700{ 701 kthread_t *td = curthread; 702 vnode_t *mvp = vfsp->mnt_vnodecovered; 703 cred_t *cr = td->td_ucred; 704 char *osname; 705 int error = 0; 706 int canwrite; 707 708 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 709 return (EINVAL); 710 711 /* 712 * If full-owner-access is enabled and delegated administration is 713 * turned on, we must set nosuid. 714 */ 715 if (zfs_super_owner && 716 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 717 secpolicy_fs_mount_clearopts(cr, vfsp); 718 } 719 720 /* 721 * Check for mount privilege? 722 * 723 * If we don't have privilege then see if 724 * we have local permission to allow it 725 */ 726 error = secpolicy_fs_mount(cr, mvp, vfsp); 727 if (error) { 728 error = dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr); 729 if (error != 0) 730 goto out; 731 732 if (!(vfsp->vfs_flag & MS_REMOUNT)) { 733 vattr_t vattr; 734 735 /* 736 * Make sure user is the owner of the mount point 737 * or has sufficient privileges. 738 */ 739 740 vattr.va_mask = AT_UID; 741 742 vn_lock(mvp, LK_SHARED | LK_RETRY); 743 if (error = VOP_GETATTR(mvp, &vattr, cr)) { 744 VOP_UNLOCK(mvp, 0); 745 goto out; 746 } 747 748#if 0 /* CHECK THIS! Is probably needed for zfs_suser. */ 749 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 750 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 751 error = EPERM; 752 goto out; 753 } 754#else 755 if (error = secpolicy_vnode_owner(mvp, cr, vattr.va_uid)) { 756 VOP_UNLOCK(mvp, 0); 757 goto out; 758 } 759 760 if (error = VOP_ACCESS(mvp, VWRITE, cr, td)) { 761 VOP_UNLOCK(mvp, 0); 762 goto out; 763 } 764 VOP_UNLOCK(mvp, 0); 765#endif 766 } 767 768 secpolicy_fs_mount_clearopts(cr, vfsp); 769 } 770 771 /* 772 * Refuse to mount a filesystem if we are in a local zone and the 773 * dataset is not visible. 774 */ 775 if (!INGLOBALZONE(curthread) && 776 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 777 error = EPERM; 778 goto out; 779 } 780 781 /* 782 * When doing a remount, we simply refresh our temporary properties 783 * according to those options set in the current VFS options. 784 */ 785 if (vfsp->vfs_flag & MS_REMOUNT) { 786 /* refresh mount options */ 787 zfs_unregister_callbacks(vfsp->vfs_data); 788 error = zfs_register_callbacks(vfsp); 789 goto out; 790 } 791 792 DROP_GIANT(); 793 error = zfs_domount(vfsp, osname); 794 PICKUP_GIANT(); 795out: 796 return (error); 797} 798 799static int 800zfs_statfs(vfs_t *vfsp, struct statfs *statp) 801{ 802 zfsvfs_t *zfsvfs = vfsp->vfs_data; 803 uint64_t refdbytes, availbytes, usedobjs, availobjs; 804 805 statp->f_version = STATFS_VERSION; 806 807 ZFS_ENTER(zfsvfs); 808 809 dmu_objset_space(zfsvfs->z_os, 810 &refdbytes, &availbytes, &usedobjs, &availobjs); 811 812 /* 813 * The underlying storage pool actually uses multiple block sizes. 814 * We report the fragsize as the smallest block size we support, 815 * and we report our blocksize as the filesystem's maximum blocksize. 816 */ 817 statp->f_bsize = zfsvfs->z_vfs->vfs_bsize; 818 statp->f_iosize = zfsvfs->z_vfs->vfs_bsize; 819 820 /* 821 * The following report "total" blocks of various kinds in the 822 * file system, but reported in terms of f_frsize - the 823 * "fragment" size. 824 */ 825 826 statp->f_blocks = (refdbytes + availbytes) / statp->f_bsize; 827 statp->f_bfree = availbytes / statp->f_bsize; 828 statp->f_bavail = statp->f_bfree; /* no root reservation */ 829 830 /* 831 * statvfs() should really be called statufs(), because it assumes 832 * static metadata. ZFS doesn't preallocate files, so the best 833 * we can do is report the max that could possibly fit in f_files, 834 * and that minus the number actually used in f_ffree. 835 * For f_ffree, report the smaller of the number of object available 836 * and the number of blocks (each object will take at least a block). 837 */ 838 statp->f_ffree = MIN(availobjs, statp->f_bfree); 839 statp->f_files = statp->f_ffree + usedobjs; 840 841 /* 842 * We're a zfs filesystem. 843 */ 844 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 845 846 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 847 sizeof(statp->f_mntfromname)); 848 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 849 sizeof(statp->f_mntonname)); 850 851 statp->f_namemax = ZFS_MAXNAMELEN; 852 853 ZFS_EXIT(zfsvfs); 854 return (0); 855} 856 857static int 858zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 859{ 860 zfsvfs_t *zfsvfs = vfsp->vfs_data; 861 znode_t *rootzp; 862 int error; 863 864 ZFS_ENTER(zfsvfs); 865 866 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 867 if (error == 0) { 868 *vpp = ZTOV(rootzp); 869 error = vn_lock(*vpp, flags); 870 (*vpp)->v_vflag |= VV_ROOT; 871 } 872 873 ZFS_EXIT(zfsvfs); 874 return (error); 875} 876 877/* 878 * Teardown the zfsvfs::z_os. 879 * 880 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 881 * and 'z_teardown_inactive_lock' held. 882 */ 883static int 884zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 885{ 886 znode_t *zp; 887 888 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 889 890 if (!unmounting) { 891 /* 892 * We purge the parent filesystem's vfsp as the parent 893 * filesystem and all of its snapshots have their vnode's 894 * v_vfsp set to the parent's filesystem's vfsp. Note, 895 * 'z_parent' is self referential for non-snapshots. 896 */ 897 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 898 } 899 900 /* 901 * Close the zil. NB: Can't close the zil while zfs_inactive 902 * threads are blocked as zil_close can call zfs_inactive. 903 */ 904 if (zfsvfs->z_log) { 905 zil_close(zfsvfs->z_log); 906 zfsvfs->z_log = NULL; 907 } 908 909 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 910 911 /* 912 * If we are not unmounting (ie: online recv) and someone already 913 * unmounted this file system while we were doing the switcheroo, 914 * or a reopen of z_os failed then just bail out now. 915 */ 916 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 917 rw_exit(&zfsvfs->z_teardown_inactive_lock); 918 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 919 return (EIO); 920 } 921 922 /* 923 * At this point there are no vops active, and any new vops will 924 * fail with EIO since we have z_teardown_lock for writer (only 925 * relavent for forced unmount). 926 * 927 * Release all holds on dbufs. 928 */ 929 mutex_enter(&zfsvfs->z_znodes_lock); 930 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 931 zp = list_next(&zfsvfs->z_all_znodes, zp)) 932 if (zp->z_dbuf) { 933 ASSERT(ZTOV(zp)->v_count >= 0); 934 zfs_znode_dmu_fini(zp); 935 } 936 mutex_exit(&zfsvfs->z_znodes_lock); 937 938 /* 939 * If we are unmounting, set the unmounted flag and let new vops 940 * unblock. zfs_inactive will have the unmounted behavior, and all 941 * other vops will fail with EIO. 942 */ 943 if (unmounting) { 944 zfsvfs->z_unmounted = B_TRUE; 945 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 946 rw_exit(&zfsvfs->z_teardown_inactive_lock); 947 } 948 949 /* 950 * z_os will be NULL if there was an error in attempting to reopen 951 * zfsvfs, so just return as the properties had already been 952 * unregistered and cached data had been evicted before. 953 */ 954 if (zfsvfs->z_os == NULL) 955 return (0); 956 957 /* 958 * Unregister properties. 959 */ 960 zfs_unregister_callbacks(zfsvfs); 961 962 /* 963 * Evict cached data 964 */ 965 if (dmu_objset_evict_dbufs(zfsvfs->z_os)) { 966 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 967 (void) dmu_objset_evict_dbufs(zfsvfs->z_os); 968 } 969 970 return (0); 971} 972 973/*ARGSUSED*/ 974static int 975zfs_umount(vfs_t *vfsp, int fflag) 976{ 977 zfsvfs_t *zfsvfs = vfsp->vfs_data; 978 objset_t *os; 979 cred_t *cr = curthread->td_ucred; 980 int ret; 981 982 if (fflag & MS_FORCE) { 983 /* TODO: Force unmount is not well implemented yet, so deny it. */ 984 ZFS_LOG(0, "Force unmount is experimental - report any problems."); 985 } 986 987 ret = secpolicy_fs_unmount(cr, vfsp); 988 if (ret) { 989 ret = dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 990 ZFS_DELEG_PERM_MOUNT, cr); 991 if (ret) 992 return (ret); 993 } 994 /* 995 * We purge the parent filesystem's vfsp as the parent filesystem 996 * and all of its snapshots have their vnode's v_vfsp set to the 997 * parent's filesystem's vfsp. Note, 'z_parent' is self 998 * referential for non-snapshots. 999 */ 1000 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1001 1002 /* 1003 * Unmount any snapshots mounted under .zfs before unmounting the 1004 * dataset itself. 1005 */ 1006 if (zfsvfs->z_ctldir != NULL) { 1007 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1008 return (ret); 1009 ret = vflush(vfsp, 0, 0, curthread); 1010 ASSERT(ret == EBUSY); 1011 if (!(fflag & MS_FORCE)) { 1012 if (zfsvfs->z_ctldir->v_count > 1) 1013 return (EBUSY); 1014 ASSERT(zfsvfs->z_ctldir->v_count == 1); 1015 } 1016 zfsctl_destroy(zfsvfs); 1017 ASSERT(zfsvfs->z_ctldir == NULL); 1018 } 1019 1020 /* 1021 * Flush all the files. 1022 */ 1023 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, curthread); 1024 if (ret != 0) { 1025 if (!zfsvfs->z_issnap) { 1026 zfsctl_create(zfsvfs); 1027 ASSERT(zfsvfs->z_ctldir != NULL); 1028 } 1029 return (ret); 1030 } 1031 1032 if (!(fflag & MS_FORCE)) { 1033 /* 1034 * Check the number of active vnodes in the file system. 1035 * Our count is maintained in the vfs structure, but the 1036 * number is off by 1 to indicate a hold on the vfs 1037 * structure itself. 1038 * 1039 * The '.zfs' directory maintains a reference of its 1040 * own, and any active references underneath are 1041 * reflected in the vnode count. 1042 */ 1043 if (zfsvfs->z_ctldir == NULL) { 1044 if (vfsp->vfs_count > 1) 1045 return (EBUSY); 1046 } else { 1047 if (vfsp->vfs_count > 2 || 1048 zfsvfs->z_ctldir->v_count > 1) 1049 return (EBUSY); 1050 } 1051 } else { 1052 MNT_ILOCK(vfsp); 1053 vfsp->mnt_kern_flag |= MNTK_UNMOUNTF; 1054 MNT_IUNLOCK(vfsp); 1055 } 1056 1057 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1058 os = zfsvfs->z_os; 1059 1060 /* 1061 * z_os will be NULL if there was an error in 1062 * attempting to reopen zfsvfs. 1063 */ 1064 if (os != NULL) { 1065 /* 1066 * Unset the objset user_ptr. 1067 */ 1068 mutex_enter(&os->os->os_user_ptr_lock); 1069 dmu_objset_set_user(os, NULL); 1070 mutex_exit(&os->os->os_user_ptr_lock); 1071 1072 /* 1073 * Finally release the objset 1074 */ 1075 dmu_objset_close(os); 1076 } 1077 1078 /* 1079 * We can now safely destroy the '.zfs' directory node. 1080 */ 1081 if (zfsvfs->z_ctldir != NULL) 1082 zfsctl_destroy(zfsvfs); 1083 if (zfsvfs->z_issnap) { 1084 vnode_t *svp = vfsp->mnt_vnodecovered; 1085 1086 ASSERT(svp->v_count == 2 || svp->v_count == 1); 1087 if (svp->v_count == 2) 1088 VN_RELE(svp); 1089 } 1090 zfs_freevfs(vfsp); 1091 1092 return (0); 1093} 1094 1095static int 1096zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 1097{ 1098 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1099 znode_t *zp; 1100 int err; 1101 1102 ZFS_ENTER(zfsvfs); 1103 err = zfs_zget(zfsvfs, ino, &zp); 1104 if (err == 0 && zp->z_unlinked) { 1105 VN_RELE(ZTOV(zp)); 1106 err = EINVAL; 1107 } 1108 if (err != 0) 1109 *vpp = NULL; 1110 else { 1111 *vpp = ZTOV(zp); 1112 vn_lock(*vpp, flags); 1113 } 1114 ZFS_EXIT(zfsvfs); 1115 return (err); 1116} 1117 1118static int 1119zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, vnode_t **vpp) 1120{ 1121 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1122 znode_t *zp; 1123 uint64_t object = 0; 1124 uint64_t fid_gen = 0; 1125 uint64_t gen_mask; 1126 uint64_t zp_gen; 1127 int i, err; 1128 1129 *vpp = NULL; 1130 1131 ZFS_ENTER(zfsvfs); 1132 1133 if (fidp->fid_len == LONG_FID_LEN) { 1134 zfid_long_t *zlfid = (zfid_long_t *)fidp; 1135 uint64_t objsetid = 0; 1136 uint64_t setgen = 0; 1137 1138 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 1139 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 1140 1141 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 1142 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 1143 1144 ZFS_EXIT(zfsvfs); 1145 1146 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 1147 if (err) 1148 return (EINVAL); 1149 ZFS_ENTER(zfsvfs); 1150 } 1151 1152 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 1153 zfid_short_t *zfid = (zfid_short_t *)fidp; 1154 1155 for (i = 0; i < sizeof (zfid->zf_object); i++) 1156 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 1157 1158 for (i = 0; i < sizeof (zfid->zf_gen); i++) 1159 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 1160 } else { 1161 ZFS_EXIT(zfsvfs); 1162 return (EINVAL); 1163 } 1164 1165 /* A zero fid_gen means we are in the .zfs control directories */ 1166 if (fid_gen == 0 && 1167 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 1168 *vpp = zfsvfs->z_ctldir; 1169 ASSERT(*vpp != NULL); 1170 if (object == ZFSCTL_INO_SNAPDIR) { 1171 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 1172 0, NULL, NULL, NULL, NULL, NULL) == 0); 1173 } else { 1174 VN_HOLD(*vpp); 1175 } 1176 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1177 ZFS_EXIT(zfsvfs); 1178 return (0); 1179 } 1180 1181 gen_mask = -1ULL >> (64 - 8 * i); 1182 1183 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 1184 if (err = zfs_zget(zfsvfs, object, &zp)) { 1185 ZFS_EXIT(zfsvfs); 1186 return (err); 1187 } 1188 zp_gen = zp->z_phys->zp_gen & gen_mask; 1189 if (zp_gen == 0) 1190 zp_gen = 1; 1191 if (zp->z_unlinked || zp_gen != fid_gen) { 1192 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 1193 VN_RELE(ZTOV(zp)); 1194 ZFS_EXIT(zfsvfs); 1195 return (EINVAL); 1196 } 1197 1198 *vpp = ZTOV(zp); 1199 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 1200 vnode_create_vobject(*vpp, zp->z_phys->zp_size, curthread); 1201 ZFS_EXIT(zfsvfs); 1202 return (0); 1203} 1204 1205/* 1206 * Block out VOPs and close zfsvfs_t::z_os 1207 * 1208 * Note, if successful, then we return with the 'z_teardown_lock' and 1209 * 'z_teardown_inactive_lock' write held. 1210 */ 1211int 1212zfs_suspend_fs(zfsvfs_t *zfsvfs, char *name, int *mode) 1213{ 1214 int error; 1215 1216 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 1217 return (error); 1218 1219 *mode = zfsvfs->z_os->os_mode; 1220 dmu_objset_name(zfsvfs->z_os, name); 1221 dmu_objset_close(zfsvfs->z_os); 1222 1223 return (0); 1224} 1225 1226/* 1227 * Reopen zfsvfs_t::z_os and release VOPs. 1228 */ 1229int 1230zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname, int mode) 1231{ 1232 int err; 1233 1234 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock)); 1235 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1236 1237 err = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os); 1238 if (err) { 1239 zfsvfs->z_os = NULL; 1240 } else { 1241 znode_t *zp; 1242 1243 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 1244 1245 /* 1246 * Attempt to re-establish all the active znodes with 1247 * their dbufs. If a zfs_rezget() fails, then we'll let 1248 * any potential callers discover that via ZFS_ENTER_VERIFY_VP 1249 * when they try to use their znode. 1250 */ 1251 mutex_enter(&zfsvfs->z_znodes_lock); 1252 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 1253 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1254 (void) zfs_rezget(zp); 1255 } 1256 mutex_exit(&zfsvfs->z_znodes_lock); 1257 1258 } 1259 1260 /* release the VOPs */ 1261 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1262 rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1263 1264 if (err) { 1265 /* 1266 * Since we couldn't reopen zfsvfs::z_os, force 1267 * unmount this file system. 1268 */ 1269 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) 1270 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 1271 } 1272 return (err); 1273} 1274 1275static void 1276zfs_freevfs(vfs_t *vfsp) 1277{ 1278 zfsvfs_t *zfsvfs = vfsp->vfs_data; 1279 int i; 1280 1281 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1282 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1283 1284 zfs_fuid_destroy(zfsvfs); 1285 zfs_freezfsvfs(zfsvfs); 1286 1287 atomic_add_32(&zfs_active_fs_count, -1); 1288} 1289 1290#ifdef __i386__ 1291static int desiredvnodes_backup; 1292#endif 1293 1294static void 1295zfs_vnodes_adjust(void) 1296{ 1297#ifdef __i386__ 1298 int newdesiredvnodes; 1299 1300 desiredvnodes_backup = desiredvnodes; 1301 1302 /* 1303 * We calculate newdesiredvnodes the same way it is done in 1304 * vntblinit(). If it is equal to desiredvnodes, it means that 1305 * it wasn't tuned by the administrator and we can tune it down. 1306 */ 1307 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 1308 vm_kmem_size / (5 * (sizeof(struct vm_object) + 1309 sizeof(struct vnode)))); 1310 if (newdesiredvnodes == desiredvnodes) 1311 desiredvnodes = (3 * newdesiredvnodes) / 4; 1312#endif 1313} 1314 1315static void 1316zfs_vnodes_adjust_back(void) 1317{ 1318 1319#ifdef __i386__ 1320 desiredvnodes = desiredvnodes_backup; 1321#endif 1322} 1323 1324void 1325zfs_init(void) 1326{ 1327 1328 printf("ZFS filesystem version " SPA_VERSION_STRING "\n"); 1329 1330 /* 1331 * Initialize znode cache, vnode ops, etc... 1332 */ 1333 zfs_znode_init(); 1334 1335 /* 1336 * Initialize .zfs directory structures 1337 */ 1338 zfsctl_init(); 1339 1340 /* 1341 * Reduce number of vnode. Originally number of vnodes is calculated 1342 * with UFS inode in mind. We reduce it here, because it's too big for 1343 * ZFS/i386. 1344 */ 1345 zfs_vnodes_adjust(); 1346} 1347 1348void 1349zfs_fini(void) 1350{ 1351 zfsctl_fini(); 1352 zfs_znode_fini(); 1353 zfs_vnodes_adjust_back(); 1354} 1355 1356int 1357zfs_busy(void) 1358{ 1359 return (zfs_active_fs_count != 0); 1360} 1361 1362int 1363zfs_set_version(const char *name, uint64_t newvers) 1364{ 1365 int error; 1366 objset_t *os; 1367 dmu_tx_t *tx; 1368 uint64_t curvers; 1369 1370 /* 1371 * XXX for now, require that the filesystem be unmounted. Would 1372 * be nice to find the zfsvfs_t and just update that if 1373 * possible. 1374 */ 1375 1376 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 1377 return (EINVAL); 1378 1379 error = dmu_objset_open(name, DMU_OST_ZFS, DS_MODE_OWNER, &os); 1380 if (error) 1381 return (error); 1382 1383 error = zap_lookup(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 1384 8, 1, &curvers); 1385 if (error) 1386 goto out; 1387 if (newvers < curvers) { 1388 error = EINVAL; 1389 goto out; 1390 } 1391 1392 tx = dmu_tx_create(os); 1393 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, 0, ZPL_VERSION_STR); 1394 error = dmu_tx_assign(tx, TXG_WAIT); 1395 if (error) { 1396 dmu_tx_abort(tx); 1397 goto out; 1398 } 1399 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 8, 1, 1400 &newvers, tx); 1401 1402 spa_history_internal_log(LOG_DS_UPGRADE, 1403 dmu_objset_spa(os), tx, CRED(), 1404 "oldver=%llu newver=%llu dataset = %llu", curvers, newvers, 1405 dmu_objset_id(os)); 1406 dmu_tx_commit(tx); 1407 1408out: 1409 dmu_objset_close(os); 1410 return (error); 1411} 1412/* 1413 * Read a property stored within the master node. 1414 */ 1415int 1416zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 1417{ 1418 const char *pname; 1419 int error = ENOENT; 1420 1421 /* 1422 * Look up the file system's value for the property. For the 1423 * version property, we look up a slightly different string. 1424 */ 1425 if (prop == ZFS_PROP_VERSION) 1426 pname = ZPL_VERSION_STR; 1427 else 1428 pname = zfs_prop_to_name(prop); 1429 1430 if (os != NULL) 1431 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 1432 1433 if (error == ENOENT) { 1434 /* No value set, use the default value */ 1435 switch (prop) { 1436 case ZFS_PROP_VERSION: 1437 *value = ZPL_VERSION; 1438 break; 1439 case ZFS_PROP_NORMALIZE: 1440 case ZFS_PROP_UTF8ONLY: 1441 *value = 0; 1442 break; 1443 case ZFS_PROP_CASE: 1444 *value = ZFS_CASE_SENSITIVE; 1445 break; 1446 default: 1447 return (error); 1448 } 1449 error = 0; 1450 } 1451 return (error); 1452} 1453