1168404Spjd/* 2168404Spjd * CDDL HEADER START 3168404Spjd * 4168404Spjd * The contents of this file are subject to the terms of the 5168404Spjd * Common Development and Distribution License (the "License"). 6168404Spjd * You may not use this file except in compliance with the License. 7168404Spjd * 8168404Spjd * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9168404Spjd * or http://www.opensolaris.org/os/licensing. 10168404Spjd * See the License for the specific language governing permissions 11168404Spjd * and limitations under the License. 12168404Spjd * 13168404Spjd * When distributing Covered Code, include this CDDL HEADER in each 14168404Spjd * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15168404Spjd * If applicable, add the following below this CDDL HEADER, with the 16168404Spjd * fields enclosed by brackets "[]" replaced with your own identifying 17168404Spjd * information: Portions Copyright [yyyy] [name of copyright owner] 18168404Spjd * 19168404Spjd * CDDL HEADER END 20168404Spjd */ 21168404Spjd/* 22219089Spjd * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23226707Spjd * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>. 24226707Spjd * All rights reserved. 25249195Smm * Copyright (c) 2013 by Delphix. All rights reserved. 26168404Spjd */ 27168404Spjd 28219089Spjd/* Portions Copyright 2010 Robert Milkowski */ 29219089Spjd 30168404Spjd#include <sys/types.h> 31168404Spjd#include <sys/param.h> 32168404Spjd#include <sys/systm.h> 33168404Spjd#include <sys/kernel.h> 34168404Spjd#include <sys/sysmacros.h> 35168404Spjd#include <sys/kmem.h> 36168404Spjd#include <sys/acl.h> 37168404Spjd#include <sys/vnode.h> 38168404Spjd#include <sys/vfs.h> 39168404Spjd#include <sys/mntent.h> 40168404Spjd#include <sys/mount.h> 41168404Spjd#include <sys/cmn_err.h> 42168404Spjd#include <sys/zfs_znode.h> 43168404Spjd#include <sys/zfs_dir.h> 44168404Spjd#include <sys/zil.h> 45168404Spjd#include <sys/fs/zfs.h> 46168404Spjd#include <sys/dmu.h> 47168404Spjd#include <sys/dsl_prop.h> 48168404Spjd#include <sys/dsl_dataset.h> 49185029Spjd#include <sys/dsl_deleg.h> 50168404Spjd#include <sys/spa.h> 51168404Spjd#include <sys/zap.h> 52219089Spjd#include <sys/sa.h> 53240955Smm#include <sys/sa_impl.h> 54168404Spjd#include <sys/varargs.h> 55168962Spjd#include <sys/policy.h> 56168404Spjd#include <sys/atomic.h> 57168404Spjd#include <sys/zfs_ioctl.h> 58168404Spjd#include <sys/zfs_ctldir.h> 59185029Spjd#include <sys/zfs_fuid.h> 60168962Spjd#include <sys/sunddi.h> 61168404Spjd#include <sys/dnlc.h> 62185029Spjd#include <sys/dmu_objset.h> 63185029Spjd#include <sys/spa_boot.h> 64232186Smm#include <sys/jail.h> 65219089Spjd#include "zfs_comutil.h" 66168404Spjd 67168404Spjdstruct mtx zfs_debug_mtx; 68168404SpjdMTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF); 69185029Spjd 70168404SpjdSYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system"); 71185029Spjd 72219089Spjdint zfs_super_owner; 73185029SpjdSYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0, 74185029Spjd "File system owner can perform privileged operation on his file systems"); 75185029Spjd 76219089Spjdint zfs_debug_level; 77168713SpjdTUNABLE_INT("vfs.zfs.debug", &zfs_debug_level); 78168404SpjdSYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0, 79168404Spjd "Debug level"); 80168404Spjd 81185029SpjdSYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions"); 82185029Spjdstatic int zfs_version_acl = ZFS_ACL_VERSION; 83185029SpjdSYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0, 84185029Spjd "ZFS_ACL_VERSION"); 85185029Spjdstatic int zfs_version_spa = SPA_VERSION; 86185029SpjdSYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0, 87185029Spjd "SPA_VERSION"); 88185029Spjdstatic int zfs_version_zpl = ZPL_VERSION; 89185029SpjdSYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0, 90185029Spjd "ZPL_VERSION"); 91185029Spjd 92191990Sattiliostatic int zfs_mount(vfs_t *vfsp); 93191990Sattiliostatic int zfs_umount(vfs_t *vfsp, int fflag); 94191990Sattiliostatic int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp); 95191990Sattiliostatic int zfs_statfs(vfs_t *vfsp, struct statfs *statp); 96168404Spjdstatic int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp); 97191990Sattiliostatic int zfs_sync(vfs_t *vfsp, int waitfor); 98196982Spjdstatic int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 99196982Spjd struct ucred **credanonp, int *numsecflavors, int **secflavors); 100222167Srmacklemstatic int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp); 101168404Spjdstatic void zfs_objset_close(zfsvfs_t *zfsvfs); 102168404Spjdstatic void zfs_freevfs(vfs_t *vfsp); 103168404Spjd 104168404Spjdstatic struct vfsops zfs_vfsops = { 105168404Spjd .vfs_mount = zfs_mount, 106168404Spjd .vfs_unmount = zfs_umount, 107168404Spjd .vfs_root = zfs_root, 108168404Spjd .vfs_statfs = zfs_statfs, 109168404Spjd .vfs_vget = zfs_vget, 110168404Spjd .vfs_sync = zfs_sync, 111196982Spjd .vfs_checkexp = zfs_checkexp, 112168404Spjd .vfs_fhtovp = zfs_fhtovp, 113168404Spjd}; 114168404Spjd 115185029SpjdVFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN); 116168404Spjd 117168404Spjd/* 118168404Spjd * We need to keep a count of active fs's. 119168404Spjd * This is necessary to prevent our module 120168404Spjd * from being unloaded after a umount -f 121168404Spjd */ 122168404Spjdstatic uint32_t zfs_active_fs_count = 0; 123168404Spjd 124168404Spjd/*ARGSUSED*/ 125168404Spjdstatic int 126191990Sattiliozfs_sync(vfs_t *vfsp, int waitfor) 127168404Spjd{ 128168404Spjd 129168404Spjd /* 130168404Spjd * Data integrity is job one. We don't want a compromised kernel 131168404Spjd * writing to the storage pool, so we never sync during panic. 132168404Spjd */ 133168404Spjd if (panicstr) 134168404Spjd return (0); 135168404Spjd 136168404Spjd if (vfsp != NULL) { 137168404Spjd /* 138168404Spjd * Sync a specific filesystem. 139168404Spjd */ 140168404Spjd zfsvfs_t *zfsvfs = vfsp->vfs_data; 141209962Smm dsl_pool_t *dp; 142168404Spjd int error; 143168404Spjd 144191990Sattilio error = vfs_stdsync(vfsp, waitfor); 145168404Spjd if (error != 0) 146168404Spjd return (error); 147168404Spjd 148168404Spjd ZFS_ENTER(zfsvfs); 149209962Smm dp = dmu_objset_pool(zfsvfs->z_os); 150209962Smm 151209962Smm /* 152209962Smm * If the system is shutting down, then skip any 153209962Smm * filesystems which may exist on a suspended pool. 154209962Smm */ 155209962Smm if (sys_shutdown && spa_suspended(dp->dp_spa)) { 156209962Smm ZFS_EXIT(zfsvfs); 157209962Smm return (0); 158209962Smm } 159209962Smm 160168404Spjd if (zfsvfs->z_log != NULL) 161219089Spjd zil_commit(zfsvfs->z_log, 0); 162219089Spjd 163168404Spjd ZFS_EXIT(zfsvfs); 164168404Spjd } else { 165168404Spjd /* 166168404Spjd * Sync all ZFS filesystems. This is what happens when you 167168404Spjd * run sync(1M). Unlike other filesystems, ZFS honors the 168168404Spjd * request by waiting for all pools to commit all dirty data. 169168404Spjd */ 170168404Spjd spa_sync_allpools(); 171168404Spjd } 172168404Spjd 173168404Spjd return (0); 174168404Spjd} 175168404Spjd 176252431Srmh#ifndef __FreeBSD_kernel__ 177219089Spjdstatic int 178219089Spjdzfs_create_unique_device(dev_t *dev) 179219089Spjd{ 180219089Spjd major_t new_major; 181219089Spjd 182219089Spjd do { 183219089Spjd ASSERT3U(zfs_minor, <=, MAXMIN32); 184219089Spjd minor_t start = zfs_minor; 185219089Spjd do { 186219089Spjd mutex_enter(&zfs_dev_mtx); 187219089Spjd if (zfs_minor >= MAXMIN32) { 188219089Spjd /* 189219089Spjd * If we're still using the real major 190219089Spjd * keep out of /dev/zfs and /dev/zvol minor 191219089Spjd * number space. If we're using a getudev()'ed 192219089Spjd * major number, we can use all of its minors. 193219089Spjd */ 194219089Spjd if (zfs_major == ddi_name_to_major(ZFS_DRIVER)) 195219089Spjd zfs_minor = ZFS_MIN_MINOR; 196219089Spjd else 197219089Spjd zfs_minor = 0; 198219089Spjd } else { 199219089Spjd zfs_minor++; 200219089Spjd } 201219089Spjd *dev = makedevice(zfs_major, zfs_minor); 202219089Spjd mutex_exit(&zfs_dev_mtx); 203219089Spjd } while (vfs_devismounted(*dev) && zfs_minor != start); 204219089Spjd if (zfs_minor == start) { 205219089Spjd /* 206219089Spjd * We are using all ~262,000 minor numbers for the 207219089Spjd * current major number. Create a new major number. 208219089Spjd */ 209219089Spjd if ((new_major = getudev()) == (major_t)-1) { 210219089Spjd cmn_err(CE_WARN, 211219089Spjd "zfs_mount: Can't get unique major " 212219089Spjd "device number."); 213219089Spjd return (-1); 214219089Spjd } 215219089Spjd mutex_enter(&zfs_dev_mtx); 216219089Spjd zfs_major = new_major; 217219089Spjd zfs_minor = 0; 218219089Spjd 219219089Spjd mutex_exit(&zfs_dev_mtx); 220219089Spjd } else { 221219089Spjd break; 222219089Spjd } 223219089Spjd /* CONSTANTCONDITION */ 224219089Spjd } while (1); 225219089Spjd 226219089Spjd return (0); 227219089Spjd} 228252431Srmh#endif /* !__FreeBSD_kernel__ */ 229219089Spjd 230168404Spjdstatic void 231168404Spjdatime_changed_cb(void *arg, uint64_t newval) 232168404Spjd{ 233168404Spjd zfsvfs_t *zfsvfs = arg; 234168404Spjd 235168404Spjd if (newval == TRUE) { 236168404Spjd zfsvfs->z_atime = TRUE; 237168404Spjd zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; 238168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); 239168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); 240168404Spjd } else { 241168404Spjd zfsvfs->z_atime = FALSE; 242168404Spjd zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; 243168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); 244168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); 245168404Spjd } 246168404Spjd} 247168404Spjd 248168404Spjdstatic void 249168404Spjdxattr_changed_cb(void *arg, uint64_t newval) 250168404Spjd{ 251168404Spjd zfsvfs_t *zfsvfs = arg; 252168404Spjd 253168404Spjd if (newval == TRUE) { 254168404Spjd /* XXX locking on vfs_flag? */ 255168404Spjd#ifdef TODO 256168404Spjd zfsvfs->z_vfs->vfs_flag |= VFS_XATTR; 257168404Spjd#endif 258168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR); 259168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0); 260168404Spjd } else { 261168404Spjd /* XXX locking on vfs_flag? */ 262168404Spjd#ifdef TODO 263168404Spjd zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR; 264168404Spjd#endif 265168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR); 266168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0); 267168404Spjd } 268168404Spjd} 269168404Spjd 270168404Spjdstatic void 271168404Spjdblksz_changed_cb(void *arg, uint64_t newval) 272168404Spjd{ 273168404Spjd zfsvfs_t *zfsvfs = arg; 274168404Spjd 275168404Spjd if (newval < SPA_MINBLOCKSIZE || 276168404Spjd newval > SPA_MAXBLOCKSIZE || !ISP2(newval)) 277168404Spjd newval = SPA_MAXBLOCKSIZE; 278168404Spjd 279168404Spjd zfsvfs->z_max_blksz = newval; 280204101Spjd zfsvfs->z_vfs->mnt_stat.f_iosize = newval; 281168404Spjd} 282168404Spjd 283168404Spjdstatic void 284168404Spjdreadonly_changed_cb(void *arg, uint64_t newval) 285168404Spjd{ 286168404Spjd zfsvfs_t *zfsvfs = arg; 287168404Spjd 288168404Spjd if (newval) { 289168404Spjd /* XXX locking on vfs_flag? */ 290168404Spjd zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; 291168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); 292168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); 293168404Spjd } else { 294168404Spjd /* XXX locking on vfs_flag? */ 295168404Spjd zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 296168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); 297168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); 298168404Spjd } 299168404Spjd} 300168404Spjd 301168404Spjdstatic void 302168404Spjdsetuid_changed_cb(void *arg, uint64_t newval) 303168404Spjd{ 304168404Spjd zfsvfs_t *zfsvfs = arg; 305168404Spjd 306168404Spjd if (newval == FALSE) { 307168404Spjd zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; 308168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); 309168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); 310168404Spjd } else { 311168404Spjd zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; 312168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); 313168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); 314168404Spjd } 315168404Spjd} 316168404Spjd 317168404Spjdstatic void 318168404Spjdexec_changed_cb(void *arg, uint64_t newval) 319168404Spjd{ 320168404Spjd zfsvfs_t *zfsvfs = arg; 321168404Spjd 322168404Spjd if (newval == FALSE) { 323168404Spjd zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; 324168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); 325168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); 326168404Spjd } else { 327168404Spjd zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; 328168404Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); 329168404Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); 330168404Spjd } 331168404Spjd} 332168404Spjd 333185029Spjd/* 334185029Spjd * The nbmand mount option can be changed at mount time. 335185029Spjd * We can't allow it to be toggled on live file systems or incorrect 336185029Spjd * behavior may be seen from cifs clients 337185029Spjd * 338185029Spjd * This property isn't registered via dsl_prop_register(), but this callback 339185029Spjd * will be called when a file system is first mounted 340185029Spjd */ 341168404Spjdstatic void 342185029Spjdnbmand_changed_cb(void *arg, uint64_t newval) 343185029Spjd{ 344185029Spjd zfsvfs_t *zfsvfs = arg; 345185029Spjd if (newval == FALSE) { 346185029Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); 347185029Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); 348185029Spjd } else { 349185029Spjd vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); 350185029Spjd vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); 351185029Spjd } 352185029Spjd} 353185029Spjd 354185029Spjdstatic void 355168404Spjdsnapdir_changed_cb(void *arg, uint64_t newval) 356168404Spjd{ 357168404Spjd zfsvfs_t *zfsvfs = arg; 358168404Spjd 359168404Spjd zfsvfs->z_show_ctldir = newval; 360168404Spjd} 361168404Spjd 362168404Spjdstatic void 363185029Spjdvscan_changed_cb(void *arg, uint64_t newval) 364185029Spjd{ 365185029Spjd zfsvfs_t *zfsvfs = arg; 366185029Spjd 367185029Spjd zfsvfs->z_vscan = newval; 368185029Spjd} 369185029Spjd 370185029Spjdstatic void 371224174Smmacl_mode_changed_cb(void *arg, uint64_t newval) 372224174Smm{ 373224174Smm zfsvfs_t *zfsvfs = arg; 374224174Smm 375224174Smm zfsvfs->z_acl_mode = newval; 376224174Smm} 377224174Smm 378224174Smmstatic void 379168404Spjdacl_inherit_changed_cb(void *arg, uint64_t newval) 380168404Spjd{ 381168404Spjd zfsvfs_t *zfsvfs = arg; 382168404Spjd 383168404Spjd zfsvfs->z_acl_inherit = newval; 384168404Spjd} 385168404Spjd 386168404Spjdstatic int 387168404Spjdzfs_register_callbacks(vfs_t *vfsp) 388168404Spjd{ 389168404Spjd struct dsl_dataset *ds = NULL; 390168404Spjd objset_t *os = NULL; 391168404Spjd zfsvfs_t *zfsvfs = NULL; 392185029Spjd uint64_t nbmand; 393247187Smm boolean_t readonly = B_FALSE; 394247187Smm boolean_t do_readonly = B_FALSE; 395247187Smm boolean_t setuid = B_FALSE; 396247187Smm boolean_t do_setuid = B_FALSE; 397247187Smm boolean_t exec = B_FALSE; 398247187Smm boolean_t do_exec = B_FALSE; 399248571Smm#ifdef illumos 400247187Smm boolean_t devices = B_FALSE; 401247187Smm boolean_t do_devices = B_FALSE; 402248571Smm#endif 403247187Smm boolean_t xattr = B_FALSE; 404247187Smm boolean_t do_xattr = B_FALSE; 405247187Smm boolean_t atime = B_FALSE; 406247187Smm boolean_t do_atime = B_FALSE; 407168404Spjd int error = 0; 408168404Spjd 409168404Spjd ASSERT(vfsp); 410168404Spjd zfsvfs = vfsp->vfs_data; 411168404Spjd ASSERT(zfsvfs); 412168404Spjd os = zfsvfs->z_os; 413168404Spjd 414168404Spjd /* 415196965Spjd * This function can be called for a snapshot when we update snapshot's 416196965Spjd * mount point, which isn't really supported. 417196965Spjd */ 418196965Spjd if (dmu_objset_is_snapshot(os)) 419196965Spjd return (EOPNOTSUPP); 420196965Spjd 421196965Spjd /* 422168404Spjd * The act of registering our callbacks will destroy any mount 423168404Spjd * options we may have. In order to enable temporary overrides 424168404Spjd * of mount options, we stash away the current values and 425168404Spjd * restore them after we register the callbacks. 426168404Spjd */ 427219089Spjd if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) || 428219089Spjd !spa_writeable(dmu_objset_spa(os))) { 429168404Spjd readonly = B_TRUE; 430168404Spjd do_readonly = B_TRUE; 431168404Spjd } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { 432168404Spjd readonly = B_FALSE; 433168404Spjd do_readonly = B_TRUE; 434168404Spjd } 435168404Spjd if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { 436168404Spjd setuid = B_FALSE; 437168404Spjd do_setuid = B_TRUE; 438168404Spjd } else { 439168404Spjd if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { 440168404Spjd setuid = B_FALSE; 441168404Spjd do_setuid = B_TRUE; 442168404Spjd } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { 443168404Spjd setuid = B_TRUE; 444168404Spjd do_setuid = B_TRUE; 445168404Spjd } 446168404Spjd } 447168404Spjd if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { 448168404Spjd exec = B_FALSE; 449168404Spjd do_exec = B_TRUE; 450168404Spjd } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { 451168404Spjd exec = B_TRUE; 452168404Spjd do_exec = B_TRUE; 453168404Spjd } 454168404Spjd if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { 455168404Spjd xattr = B_FALSE; 456168404Spjd do_xattr = B_TRUE; 457168404Spjd } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { 458168404Spjd xattr = B_TRUE; 459168404Spjd do_xattr = B_TRUE; 460168404Spjd } 461185029Spjd if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { 462185029Spjd atime = B_FALSE; 463185029Spjd do_atime = B_TRUE; 464185029Spjd } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { 465185029Spjd atime = B_TRUE; 466185029Spjd do_atime = B_TRUE; 467185029Spjd } 468168404Spjd 469168404Spjd /* 470185029Spjd * nbmand is a special property. It can only be changed at 471185029Spjd * mount time. 472185029Spjd * 473185029Spjd * This is weird, but it is documented to only be changeable 474185029Spjd * at mount time. 475185029Spjd */ 476185029Spjd if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { 477185029Spjd nbmand = B_FALSE; 478185029Spjd } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { 479185029Spjd nbmand = B_TRUE; 480185029Spjd } else { 481185029Spjd char osname[MAXNAMELEN]; 482185029Spjd 483185029Spjd dmu_objset_name(os, osname); 484185029Spjd if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand, 485185029Spjd NULL)) { 486185029Spjd return (error); 487185029Spjd } 488185029Spjd } 489185029Spjd 490185029Spjd /* 491168404Spjd * Register property callbacks. 492168404Spjd * 493168404Spjd * It would probably be fine to just check for i/o error from 494168404Spjd * the first prop_register(), but I guess I like to go 495168404Spjd * overboard... 496168404Spjd */ 497168404Spjd ds = dmu_objset_ds(os); 498248571Smm dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 499248571Smm error = dsl_prop_register(ds, 500248571Smm zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 501168404Spjd error = error ? error : dsl_prop_register(ds, 502248571Smm zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 503168404Spjd error = error ? error : dsl_prop_register(ds, 504248571Smm zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 505168404Spjd error = error ? error : dsl_prop_register(ds, 506248571Smm zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 507248571Smm#ifdef illumos 508168404Spjd error = error ? error : dsl_prop_register(ds, 509248571Smm zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 510248571Smm#endif 511168404Spjd error = error ? error : dsl_prop_register(ds, 512248571Smm zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 513168404Spjd error = error ? error : dsl_prop_register(ds, 514248571Smm zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 515168404Spjd error = error ? error : dsl_prop_register(ds, 516248571Smm zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 517224174Smm error = error ? error : dsl_prop_register(ds, 518248571Smm zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 519185029Spjd error = error ? error : dsl_prop_register(ds, 520248571Smm zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 521248571Smm zfsvfs); 522248571Smm error = error ? error : dsl_prop_register(ds, 523248571Smm zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); 524248571Smm dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 525168404Spjd if (error) 526168404Spjd goto unregister; 527168404Spjd 528168404Spjd /* 529168404Spjd * Invoke our callbacks to restore temporary mount options. 530168404Spjd */ 531168404Spjd if (do_readonly) 532168404Spjd readonly_changed_cb(zfsvfs, readonly); 533168404Spjd if (do_setuid) 534168404Spjd setuid_changed_cb(zfsvfs, setuid); 535168404Spjd if (do_exec) 536168404Spjd exec_changed_cb(zfsvfs, exec); 537168404Spjd if (do_xattr) 538168404Spjd xattr_changed_cb(zfsvfs, xattr); 539185029Spjd if (do_atime) 540185029Spjd atime_changed_cb(zfsvfs, atime); 541168404Spjd 542185029Spjd nbmand_changed_cb(zfsvfs, nbmand); 543185029Spjd 544168404Spjd return (0); 545168404Spjd 546168404Spjdunregister: 547168404Spjd /* 548168404Spjd * We may attempt to unregister some callbacks that are not 549168404Spjd * registered, but this is OK; it will simply return ENOMSG, 550168404Spjd * which we will ignore. 551168404Spjd */ 552248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME), 553248571Smm atime_changed_cb, zfsvfs); 554248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR), 555248571Smm xattr_changed_cb, zfsvfs); 556248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), 557248571Smm blksz_changed_cb, zfsvfs); 558248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY), 559248571Smm readonly_changed_cb, zfsvfs); 560248571Smm#ifdef illumos 561248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES), 562248571Smm devices_changed_cb, zfsvfs); 563248571Smm#endif 564248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID), 565248571Smm setuid_changed_cb, zfsvfs); 566248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC), 567248571Smm exec_changed_cb, zfsvfs); 568248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR), 569248571Smm snapdir_changed_cb, zfsvfs); 570248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE), 571248571Smm acl_mode_changed_cb, zfsvfs); 572248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT), 573248571Smm acl_inherit_changed_cb, zfsvfs); 574248571Smm (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN), 575248571Smm vscan_changed_cb, zfsvfs); 576168404Spjd return (error); 577168404Spjd} 578168404Spjd 579219089Spjdstatic int 580219089Spjdzfs_space_delta_cb(dmu_object_type_t bonustype, void *data, 581219089Spjd uint64_t *userp, uint64_t *groupp) 582209962Smm{ 583219089Spjd /* 584219089Spjd * Is it a valid type of object to track? 585219089Spjd */ 586219089Spjd if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) 587249195Smm return (SET_ERROR(ENOENT)); 588209962Smm 589219089Spjd /* 590219089Spjd * If we have a NULL data pointer 591219089Spjd * then assume the id's aren't changing and 592219089Spjd * return EEXIST to the dmu to let it know to 593219089Spjd * use the same ids 594219089Spjd */ 595219089Spjd if (data == NULL) 596249195Smm return (SET_ERROR(EEXIST)); 597209962Smm 598219089Spjd if (bonustype == DMU_OT_ZNODE) { 599240955Smm znode_phys_t *znp = data; 600219089Spjd *userp = znp->zp_uid; 601219089Spjd *groupp = znp->zp_gid; 602219089Spjd } else { 603219089Spjd int hdrsize; 604240955Smm sa_hdr_phys_t *sap = data; 605240955Smm sa_hdr_phys_t sa = *sap; 606240955Smm boolean_t swap = B_FALSE; 607209962Smm 608219089Spjd ASSERT(bonustype == DMU_OT_SA); 609209962Smm 610240955Smm if (sa.sa_magic == 0) { 611219089Spjd /* 612219089Spjd * This should only happen for newly created 613219089Spjd * files that haven't had the znode data filled 614219089Spjd * in yet. 615219089Spjd */ 616219089Spjd *userp = 0; 617219089Spjd *groupp = 0; 618240955Smm return (0); 619219089Spjd } 620240955Smm if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { 621240955Smm sa.sa_magic = SA_MAGIC; 622240955Smm sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); 623240955Smm swap = B_TRUE; 624240955Smm } else { 625240955Smm VERIFY3U(sa.sa_magic, ==, SA_MAGIC); 626240955Smm } 627240955Smm 628240955Smm hdrsize = sa_hdrsize(&sa); 629240955Smm VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); 630240955Smm *userp = *((uint64_t *)((uintptr_t)data + hdrsize + 631240955Smm SA_UID_OFFSET)); 632240955Smm *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + 633240955Smm SA_GID_OFFSET)); 634240955Smm if (swap) { 635240955Smm *userp = BSWAP_64(*userp); 636240955Smm *groupp = BSWAP_64(*groupp); 637240955Smm } 638209962Smm } 639248571Smm return (0); 640209962Smm} 641209962Smm 642209962Smmstatic void 643209962Smmfuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr, 644209962Smm char *domainbuf, int buflen, uid_t *ridp) 645209962Smm{ 646209962Smm uint64_t fuid; 647209962Smm const char *domain; 648209962Smm 649209962Smm fuid = strtonum(fuidstr, NULL); 650209962Smm 651209962Smm domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid)); 652209962Smm if (domain) 653209962Smm (void) strlcpy(domainbuf, domain, buflen); 654209962Smm else 655209962Smm domainbuf[0] = '\0'; 656209962Smm *ridp = FUID_RID(fuid); 657209962Smm} 658209962Smm 659209962Smmstatic uint64_t 660209962Smmzfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type) 661209962Smm{ 662209962Smm switch (type) { 663209962Smm case ZFS_PROP_USERUSED: 664209962Smm return (DMU_USERUSED_OBJECT); 665209962Smm case ZFS_PROP_GROUPUSED: 666209962Smm return (DMU_GROUPUSED_OBJECT); 667209962Smm case ZFS_PROP_USERQUOTA: 668209962Smm return (zfsvfs->z_userquota_obj); 669209962Smm case ZFS_PROP_GROUPQUOTA: 670209962Smm return (zfsvfs->z_groupquota_obj); 671209962Smm } 672209962Smm return (0); 673209962Smm} 674209962Smm 675209962Smmint 676209962Smmzfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 677209962Smm uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) 678209962Smm{ 679209962Smm int error; 680209962Smm zap_cursor_t zc; 681209962Smm zap_attribute_t za; 682209962Smm zfs_useracct_t *buf = vbuf; 683209962Smm uint64_t obj; 684209962Smm 685209962Smm if (!dmu_objset_userspace_present(zfsvfs->z_os)) 686249195Smm return (SET_ERROR(ENOTSUP)); 687209962Smm 688209962Smm obj = zfs_userquota_prop_to_obj(zfsvfs, type); 689209962Smm if (obj == 0) { 690209962Smm *bufsizep = 0; 691209962Smm return (0); 692209962Smm } 693209962Smm 694209962Smm for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep); 695209962Smm (error = zap_cursor_retrieve(&zc, &za)) == 0; 696209962Smm zap_cursor_advance(&zc)) { 697209962Smm if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > 698209962Smm *bufsizep) 699209962Smm break; 700209962Smm 701209962Smm fuidstr_to_sid(zfsvfs, za.za_name, 702209962Smm buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); 703209962Smm 704209962Smm buf->zu_space = za.za_first_integer; 705209962Smm buf++; 706209962Smm } 707209962Smm if (error == ENOENT) 708209962Smm error = 0; 709209962Smm 710209962Smm ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); 711209962Smm *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; 712209962Smm *cookiep = zap_cursor_serialize(&zc); 713209962Smm zap_cursor_fini(&zc); 714209962Smm return (error); 715209962Smm} 716209962Smm 717209962Smm/* 718209962Smm * buf must be big enough (eg, 32 bytes) 719209962Smm */ 720168404Spjdstatic int 721209962Smmid_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid, 722209962Smm char *buf, boolean_t addok) 723209962Smm{ 724209962Smm uint64_t fuid; 725209962Smm int domainid = 0; 726209962Smm 727209962Smm if (domain && domain[0]) { 728209962Smm domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok); 729209962Smm if (domainid == -1) 730249195Smm return (SET_ERROR(ENOENT)); 731209962Smm } 732209962Smm fuid = FUID_ENCODE(domainid, rid); 733209962Smm (void) sprintf(buf, "%llx", (longlong_t)fuid); 734209962Smm return (0); 735209962Smm} 736209962Smm 737209962Smmint 738209962Smmzfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 739209962Smm const char *domain, uint64_t rid, uint64_t *valp) 740209962Smm{ 741209962Smm char buf[32]; 742209962Smm int err; 743209962Smm uint64_t obj; 744209962Smm 745209962Smm *valp = 0; 746209962Smm 747209962Smm if (!dmu_objset_userspace_present(zfsvfs->z_os)) 748249195Smm return (SET_ERROR(ENOTSUP)); 749209962Smm 750209962Smm obj = zfs_userquota_prop_to_obj(zfsvfs, type); 751209962Smm if (obj == 0) 752209962Smm return (0); 753209962Smm 754209962Smm err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE); 755209962Smm if (err) 756209962Smm return (err); 757209962Smm 758209962Smm err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp); 759209962Smm if (err == ENOENT) 760209962Smm err = 0; 761209962Smm return (err); 762209962Smm} 763209962Smm 764209962Smmint 765209962Smmzfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type, 766209962Smm const char *domain, uint64_t rid, uint64_t quota) 767209962Smm{ 768209962Smm char buf[32]; 769209962Smm int err; 770209962Smm dmu_tx_t *tx; 771209962Smm uint64_t *objp; 772209962Smm boolean_t fuid_dirtied; 773209962Smm 774209962Smm if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) 775249195Smm return (SET_ERROR(EINVAL)); 776209962Smm 777209962Smm if (zfsvfs->z_version < ZPL_VERSION_USERSPACE) 778249195Smm return (SET_ERROR(ENOTSUP)); 779209962Smm 780209962Smm objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj : 781209962Smm &zfsvfs->z_groupquota_obj; 782209962Smm 783209962Smm err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE); 784209962Smm if (err) 785209962Smm return (err); 786209962Smm fuid_dirtied = zfsvfs->z_fuid_dirty; 787209962Smm 788209962Smm tx = dmu_tx_create(zfsvfs->z_os); 789209962Smm dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); 790209962Smm if (*objp == 0) { 791209962Smm dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 792209962Smm zfs_userquota_prop_prefixes[type]); 793209962Smm } 794209962Smm if (fuid_dirtied) 795209962Smm zfs_fuid_txhold(zfsvfs, tx); 796209962Smm err = dmu_tx_assign(tx, TXG_WAIT); 797209962Smm if (err) { 798209962Smm dmu_tx_abort(tx); 799209962Smm return (err); 800209962Smm } 801209962Smm 802209962Smm mutex_enter(&zfsvfs->z_lock); 803209962Smm if (*objp == 0) { 804209962Smm *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA, 805209962Smm DMU_OT_NONE, 0, tx); 806209962Smm VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 807209962Smm zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); 808209962Smm } 809209962Smm mutex_exit(&zfsvfs->z_lock); 810209962Smm 811209962Smm if (quota == 0) { 812209962Smm err = zap_remove(zfsvfs->z_os, *objp, buf, tx); 813209962Smm if (err == ENOENT) 814209962Smm err = 0; 815209962Smm } else { 816209962Smm err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx); 817209962Smm } 818209962Smm ASSERT(err == 0); 819209962Smm if (fuid_dirtied) 820209962Smm zfs_fuid_sync(zfsvfs, tx); 821209962Smm dmu_tx_commit(tx); 822209962Smm return (err); 823209962Smm} 824209962Smm 825209962Smmboolean_t 826219089Spjdzfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid) 827209962Smm{ 828209962Smm char buf[32]; 829209962Smm uint64_t used, quota, usedobj, quotaobj; 830209962Smm int err; 831209962Smm 832209962Smm usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; 833209962Smm quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 834209962Smm 835209962Smm if (quotaobj == 0 || zfsvfs->z_replay) 836209962Smm return (B_FALSE); 837209962Smm 838209962Smm (void) sprintf(buf, "%llx", (longlong_t)fuid); 839209962Smm err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a); 840209962Smm if (err != 0) 841209962Smm return (B_FALSE); 842209962Smm 843209962Smm err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used); 844209962Smm if (err != 0) 845209962Smm return (B_FALSE); 846209962Smm return (used >= quota); 847209962Smm} 848209962Smm 849219089Spjdboolean_t 850219089Spjdzfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup) 851219089Spjd{ 852219089Spjd uint64_t fuid; 853219089Spjd uint64_t quotaobj; 854219089Spjd 855219089Spjd quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; 856219089Spjd 857219089Spjd fuid = isgroup ? zp->z_gid : zp->z_uid; 858219089Spjd 859219089Spjd if (quotaobj == 0 || zfsvfs->z_replay) 860219089Spjd return (B_FALSE); 861219089Spjd 862219089Spjd return (zfs_fuid_overquota(zfsvfs, isgroup, fuid)); 863219089Spjd} 864219089Spjd 865209962Smmint 866219089Spjdzfsvfs_create(const char *osname, zfsvfs_t **zfvp) 867209962Smm{ 868209962Smm objset_t *os; 869209962Smm zfsvfs_t *zfsvfs; 870209962Smm uint64_t zval; 871209962Smm int i, error; 872219089Spjd uint64_t sa_obj; 873209962Smm 874219089Spjd zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 875219089Spjd 876219089Spjd /* 877219089Spjd * We claim to always be readonly so we can open snapshots; 878219089Spjd * other ZPL code will prevent us from writing to snapshots. 879219089Spjd */ 880219089Spjd error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os); 881219089Spjd if (error) { 882219089Spjd kmem_free(zfsvfs, sizeof (zfsvfs_t)); 883209962Smm return (error); 884209962Smm } 885209962Smm 886209962Smm /* 887209962Smm * Initialize the zfs-specific filesystem structure. 888209962Smm * Should probably make this a kmem cache, shuffle fields, 889209962Smm * and just bzero up to z_hold_mtx[]. 890209962Smm */ 891209962Smm zfsvfs->z_vfs = NULL; 892209962Smm zfsvfs->z_parent = zfsvfs; 893209962Smm zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE; 894209962Smm zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 895209962Smm zfsvfs->z_os = os; 896209962Smm 897209962Smm error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 898209962Smm if (error) { 899209962Smm goto out; 900219089Spjd } else if (zfsvfs->z_version > 901219089Spjd zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 902219089Spjd (void) printf("Can't mount a version %lld file system " 903219089Spjd "on a version %lld pool\n. Pool must be upgraded to mount " 904219089Spjd "this file system.", (u_longlong_t)zfsvfs->z_version, 905219089Spjd (u_longlong_t)spa_version(dmu_objset_spa(os))); 906249195Smm error = SET_ERROR(ENOTSUP); 907209962Smm goto out; 908209962Smm } 909209962Smm if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) 910209962Smm goto out; 911209962Smm zfsvfs->z_norm = (int)zval; 912209962Smm 913209962Smm if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) 914209962Smm goto out; 915209962Smm zfsvfs->z_utf8 = (zval != 0); 916209962Smm 917209962Smm if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) 918209962Smm goto out; 919209962Smm zfsvfs->z_case = (uint_t)zval; 920209962Smm 921209962Smm /* 922209962Smm * Fold case on file systems that are always or sometimes case 923209962Smm * insensitive. 924209962Smm */ 925209962Smm if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 926209962Smm zfsvfs->z_case == ZFS_CASE_MIXED) 927209962Smm zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 928209962Smm 929209962Smm zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 930219089Spjd zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 931209962Smm 932219089Spjd if (zfsvfs->z_use_sa) { 933219089Spjd /* should either have both of these objects or none */ 934219089Spjd error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 935219089Spjd &sa_obj); 936219089Spjd if (error) 937219089Spjd return (error); 938219089Spjd } else { 939219089Spjd /* 940219089Spjd * Pre SA versions file systems should never touch 941219089Spjd * either the attribute registration or layout objects. 942219089Spjd */ 943219089Spjd sa_obj = 0; 944219089Spjd } 945219089Spjd 946219089Spjd error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 947219089Spjd &zfsvfs->z_attr_table); 948219089Spjd if (error) 949219089Spjd goto out; 950219089Spjd 951219089Spjd if (zfsvfs->z_version >= ZPL_VERSION_SA) 952219089Spjd sa_register_update_callback(os, zfs_sa_upgrade); 953219089Spjd 954209962Smm error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 955209962Smm &zfsvfs->z_root); 956209962Smm if (error) 957209962Smm goto out; 958209962Smm ASSERT(zfsvfs->z_root != 0); 959209962Smm 960209962Smm error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 961209962Smm &zfsvfs->z_unlinkedobj); 962209962Smm if (error) 963209962Smm goto out; 964209962Smm 965209962Smm error = zap_lookup(os, MASTER_NODE_OBJ, 966209962Smm zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 967209962Smm 8, 1, &zfsvfs->z_userquota_obj); 968209962Smm if (error && error != ENOENT) 969209962Smm goto out; 970209962Smm 971209962Smm error = zap_lookup(os, MASTER_NODE_OBJ, 972209962Smm zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 973209962Smm 8, 1, &zfsvfs->z_groupquota_obj); 974209962Smm if (error && error != ENOENT) 975209962Smm goto out; 976209962Smm 977209962Smm error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 978209962Smm &zfsvfs->z_fuid_obj); 979209962Smm if (error && error != ENOENT) 980209962Smm goto out; 981209962Smm 982209962Smm error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 983209962Smm &zfsvfs->z_shares_dir); 984209962Smm if (error && error != ENOENT) 985209962Smm goto out; 986209962Smm 987209962Smm mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 988209962Smm mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 989209962Smm list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 990209962Smm offsetof(znode_t, z_link_node)); 991248571Smm rrw_init(&zfsvfs->z_teardown_lock, B_FALSE); 992209962Smm rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 993209962Smm rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 994209962Smm for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 995209962Smm mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 996209962Smm 997219089Spjd *zfvp = zfsvfs; 998209962Smm return (0); 999209962Smm 1000209962Smmout: 1001219089Spjd dmu_objset_disown(os, zfsvfs); 1002219089Spjd *zfvp = NULL; 1003209962Smm kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1004209962Smm return (error); 1005209962Smm} 1006209962Smm 1007209962Smmstatic int 1008185029Spjdzfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 1009168404Spjd{ 1010185029Spjd int error; 1011185029Spjd 1012185029Spjd error = zfs_register_callbacks(zfsvfs->z_vfs); 1013185029Spjd if (error) 1014185029Spjd return (error); 1015185029Spjd 1016185029Spjd /* 1017185029Spjd * Set the objset user_ptr to track its zfsvfs. 1018185029Spjd */ 1019219089Spjd mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1020185029Spjd dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1021219089Spjd mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1022185029Spjd 1023208689Smm zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 1024208689Smm 1025185029Spjd /* 1026185029Spjd * If we are not mounting (ie: online recv), then we don't 1027185029Spjd * have to worry about replaying the log as we blocked all 1028185029Spjd * operations out since we closed the ZIL. 1029185029Spjd */ 1030185029Spjd if (mounting) { 1031185029Spjd boolean_t readonly; 1032185029Spjd 1033185029Spjd /* 1034185029Spjd * During replay we remove the read only flag to 1035185029Spjd * allow replays to succeed. 1036185029Spjd */ 1037185029Spjd readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; 1038208689Smm if (readonly != 0) 1039208689Smm zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; 1040208689Smm else 1041208689Smm zfs_unlinked_drain(zfsvfs); 1042185029Spjd 1043219089Spjd /* 1044219089Spjd * Parse and replay the intent log. 1045219089Spjd * 1046219089Spjd * Because of ziltest, this must be done after 1047219089Spjd * zfs_unlinked_drain(). (Further note: ziltest 1048219089Spjd * doesn't use readonly mounts, where 1049219089Spjd * zfs_unlinked_drain() isn't called.) This is because 1050219089Spjd * ziltest causes spa_sync() to think it's committed, 1051219089Spjd * but actually it is not, so the intent log contains 1052219089Spjd * many txg's worth of changes. 1053219089Spjd * 1054219089Spjd * In particular, if object N is in the unlinked set in 1055219089Spjd * the last txg to actually sync, then it could be 1056219089Spjd * actually freed in a later txg and then reallocated 1057219089Spjd * in a yet later txg. This would write a "create 1058219089Spjd * object N" record to the intent log. Normally, this 1059219089Spjd * would be fine because the spa_sync() would have 1060219089Spjd * written out the fact that object N is free, before 1061219089Spjd * we could write the "create object N" intent log 1062219089Spjd * record. 1063219089Spjd * 1064219089Spjd * But when we are in ziltest mode, we advance the "open 1065219089Spjd * txg" without actually spa_sync()-ing the changes to 1066219089Spjd * disk. So we would see that object N is still 1067219089Spjd * allocated and in the unlinked set, and there is an 1068219089Spjd * intent log record saying to allocate it. 1069219089Spjd */ 1070219089Spjd if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 1071219089Spjd if (zil_replay_disable) { 1072219089Spjd zil_destroy(zfsvfs->z_log, B_FALSE); 1073219089Spjd } else { 1074219089Spjd zfsvfs->z_replay = B_TRUE; 1075219089Spjd zil_replay(zfsvfs->z_os, zfsvfs, 1076219089Spjd zfs_replay_vector); 1077219089Spjd zfsvfs->z_replay = B_FALSE; 1078219089Spjd } 1079208689Smm } 1080185029Spjd zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */ 1081185029Spjd } 1082185029Spjd 1083185029Spjd return (0); 1084185029Spjd} 1085185029Spjd 1086210470Smmextern krwlock_t zfsvfs_lock; /* in zfs_znode.c */ 1087210470Smm 1088209962Smmvoid 1089209962Smmzfsvfs_free(zfsvfs_t *zfsvfs) 1090185029Spjd{ 1091209962Smm int i; 1092209962Smm 1093210470Smm /* 1094210470Smm * This is a barrier to prevent the filesystem from going away in 1095210470Smm * zfs_znode_move() until we can safely ensure that the filesystem is 1096210470Smm * not unmounted. We consider the filesystem valid before the barrier 1097210470Smm * and invalid after the barrier. 1098210470Smm */ 1099210470Smm rw_enter(&zfsvfs_lock, RW_READER); 1100210470Smm rw_exit(&zfsvfs_lock); 1101210470Smm 1102209962Smm zfs_fuid_destroy(zfsvfs); 1103209962Smm 1104185029Spjd mutex_destroy(&zfsvfs->z_znodes_lock); 1105209962Smm mutex_destroy(&zfsvfs->z_lock); 1106185029Spjd list_destroy(&zfsvfs->z_all_znodes); 1107185029Spjd rrw_destroy(&zfsvfs->z_teardown_lock); 1108185029Spjd rw_destroy(&zfsvfs->z_teardown_inactive_lock); 1109185029Spjd rw_destroy(&zfsvfs->z_fuid_lock); 1110209962Smm for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1111209962Smm mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1112185029Spjd kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1113185029Spjd} 1114185029Spjd 1115209962Smmstatic void 1116209962Smmzfs_set_fuid_feature(zfsvfs_t *zfsvfs) 1117209962Smm{ 1118209962Smm zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 1119219089Spjd if (zfsvfs->z_vfs) { 1120219089Spjd if (zfsvfs->z_use_fuids) { 1121219089Spjd vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1122219089Spjd vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1123219089Spjd vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1124219089Spjd vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1125219089Spjd vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1126219089Spjd vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1127219089Spjd } else { 1128219089Spjd vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR); 1129219089Spjd vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS); 1130219089Spjd vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS); 1131219089Spjd vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE); 1132219089Spjd vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER); 1133219089Spjd vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE); 1134219089Spjd } 1135209962Smm } 1136219089Spjd zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 1137209962Smm} 1138209962Smm 1139185029Spjdstatic int 1140185029Spjdzfs_domount(vfs_t *vfsp, char *osname) 1141185029Spjd{ 1142209962Smm uint64_t recordsize, fsid_guid; 1143168404Spjd int error = 0; 1144168404Spjd zfsvfs_t *zfsvfs; 1145209962Smm vnode_t *vp; 1146168404Spjd 1147168404Spjd ASSERT(vfsp); 1148168404Spjd ASSERT(osname); 1149168404Spjd 1150219089Spjd error = zfsvfs_create(osname, &zfsvfs); 1151209962Smm if (error) 1152209962Smm return (error); 1153168404Spjd zfsvfs->z_vfs = vfsp; 1154168404Spjd 1155249195Smm#ifdef illumos 1156249195Smm /* Initialize the generic filesystem structure. */ 1157249195Smm vfsp->vfs_bcount = 0; 1158249195Smm vfsp->vfs_data = NULL; 1159249195Smm 1160249195Smm if (zfs_create_unique_device(&mount_dev) == -1) { 1161249195Smm error = SET_ERROR(ENODEV); 1162249195Smm goto out; 1163249195Smm } 1164249195Smm ASSERT(vfs_devismounted(mount_dev) == 0); 1165249195Smm#endif 1166249195Smm 1167168404Spjd if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize, 1168168404Spjd NULL)) 1169168404Spjd goto out; 1170204101Spjd zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; 1171204101Spjd zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; 1172168404Spjd 1173168404Spjd vfsp->vfs_data = zfsvfs; 1174218386Strasz vfsp->mnt_flag |= MNT_LOCAL; 1175168404Spjd vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; 1176193440Sps vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; 1177242573Savg vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; 1178168404Spjd 1179209962Smm /* 1180209962Smm * The fsid is 64 bits, composed of an 8-bit fs type, which 1181209962Smm * separates our fsid from any other filesystem types, and a 1182209962Smm * 56-bit objset unique ID. The objset unique ID is unique to 1183209962Smm * all objsets open on this system, provided by unique_create(). 1184209962Smm * The 8-bit fs type must be put in the low bits of fsid[1] 1185209962Smm * because that's where other Solaris filesystems put it. 1186209962Smm */ 1187209962Smm fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); 1188209962Smm ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 1189209962Smm vfsp->vfs_fsid.val[0] = fsid_guid; 1190209962Smm vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 1191209962Smm vfsp->mnt_vfc->vfc_typenum & 0xFF; 1192168404Spjd 1193185029Spjd /* 1194185029Spjd * Set features for file system. 1195185029Spjd */ 1196209962Smm zfs_set_fuid_feature(zfsvfs); 1197185029Spjd if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) { 1198185029Spjd vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1199185029Spjd vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1200185029Spjd vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE); 1201185029Spjd } else if (zfsvfs->z_case == ZFS_CASE_MIXED) { 1202185029Spjd vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS); 1203185029Spjd vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE); 1204185029Spjd } 1205219089Spjd vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED); 1206185029Spjd 1207168404Spjd if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1208185029Spjd uint64_t pval; 1209168404Spjd 1210168404Spjd atime_changed_cb(zfsvfs, B_FALSE); 1211168404Spjd readonly_changed_cb(zfsvfs, B_TRUE); 1212185029Spjd if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL)) 1213168404Spjd goto out; 1214185029Spjd xattr_changed_cb(zfsvfs, pval); 1215168404Spjd zfsvfs->z_issnap = B_TRUE; 1216219089Spjd zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1217209962Smm 1218219089Spjd mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1219209962Smm dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1220219089Spjd mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1221168404Spjd } else { 1222185029Spjd error = zfsvfs_setup(zfsvfs, B_TRUE); 1223168404Spjd } 1224168404Spjd 1225168404Spjd vfs_mountedfrom(vfsp, osname); 1226209962Smm /* Grab extra reference. */ 1227209962Smm VERIFY(VFS_ROOT(vfsp, LK_EXCLUSIVE, &vp) == 0); 1228209962Smm VOP_UNLOCK(vp, 0); 1229168404Spjd 1230168404Spjd if (!zfsvfs->z_issnap) 1231168404Spjd zfsctl_create(zfsvfs); 1232168404Spjdout: 1233168404Spjd if (error) { 1234219089Spjd dmu_objset_disown(zfsvfs->z_os, zfsvfs); 1235209962Smm zfsvfs_free(zfsvfs); 1236168404Spjd } else { 1237168404Spjd atomic_add_32(&zfs_active_fs_count, 1); 1238168404Spjd } 1239168404Spjd 1240168404Spjd return (error); 1241168404Spjd} 1242168404Spjd 1243168404Spjdvoid 1244168404Spjdzfs_unregister_callbacks(zfsvfs_t *zfsvfs) 1245168404Spjd{ 1246168404Spjd objset_t *os = zfsvfs->z_os; 1247168404Spjd struct dsl_dataset *ds; 1248168404Spjd 1249168404Spjd /* 1250168404Spjd * Unregister properties. 1251168404Spjd */ 1252168404Spjd if (!dmu_objset_is_snapshot(os)) { 1253168404Spjd ds = dmu_objset_ds(os); 1254168404Spjd VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, 1255168404Spjd zfsvfs) == 0); 1256168404Spjd 1257168404Spjd VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, 1258168404Spjd zfsvfs) == 0); 1259168404Spjd 1260168404Spjd VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, 1261168404Spjd zfsvfs) == 0); 1262168404Spjd 1263168404Spjd VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, 1264168404Spjd zfsvfs) == 0); 1265168404Spjd 1266168404Spjd VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, 1267168404Spjd zfsvfs) == 0); 1268168404Spjd 1269168404Spjd VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, 1270168404Spjd zfsvfs) == 0); 1271168404Spjd 1272168404Spjd VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, 1273168404Spjd zfsvfs) == 0); 1274168404Spjd 1275224174Smm VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, 1276224174Smm zfsvfs) == 0); 1277224174Smm 1278168404Spjd VERIFY(dsl_prop_unregister(ds, "aclinherit", 1279168404Spjd acl_inherit_changed_cb, zfsvfs) == 0); 1280185029Spjd 1281185029Spjd VERIFY(dsl_prop_unregister(ds, "vscan", 1282185029Spjd vscan_changed_cb, zfsvfs) == 0); 1283168404Spjd } 1284168404Spjd} 1285168404Spjd 1286219089Spjd#ifdef SECLABEL 1287219089Spjd/* 1288219089Spjd * Convert a decimal digit string to a uint64_t integer. 1289219089Spjd */ 1290219089Spjdstatic int 1291219089Spjdstr_to_uint64(char *str, uint64_t *objnum) 1292219089Spjd{ 1293219089Spjd uint64_t num = 0; 1294219089Spjd 1295219089Spjd while (*str) { 1296219089Spjd if (*str < '0' || *str > '9') 1297249195Smm return (SET_ERROR(EINVAL)); 1298219089Spjd 1299219089Spjd num = num*10 + *str++ - '0'; 1300219089Spjd } 1301219089Spjd 1302219089Spjd *objnum = num; 1303219089Spjd return (0); 1304219089Spjd} 1305219089Spjd 1306219089Spjd/* 1307219089Spjd * The boot path passed from the boot loader is in the form of 1308219089Spjd * "rootpool-name/root-filesystem-object-number'. Convert this 1309219089Spjd * string to a dataset name: "rootpool-name/root-filesystem-name". 1310219089Spjd */ 1311219089Spjdstatic int 1312219089Spjdzfs_parse_bootfs(char *bpath, char *outpath) 1313219089Spjd{ 1314219089Spjd char *slashp; 1315219089Spjd uint64_t objnum; 1316219089Spjd int error; 1317219089Spjd 1318219089Spjd if (*bpath == 0 || *bpath == '/') 1319249195Smm return (SET_ERROR(EINVAL)); 1320219089Spjd 1321219089Spjd (void) strcpy(outpath, bpath); 1322219089Spjd 1323219089Spjd slashp = strchr(bpath, '/'); 1324219089Spjd 1325219089Spjd /* if no '/', just return the pool name */ 1326219089Spjd if (slashp == NULL) { 1327219089Spjd return (0); 1328219089Spjd } 1329219089Spjd 1330219089Spjd /* if not a number, just return the root dataset name */ 1331219089Spjd if (str_to_uint64(slashp+1, &objnum)) { 1332219089Spjd return (0); 1333219089Spjd } 1334219089Spjd 1335219089Spjd *slashp = '\0'; 1336219089Spjd error = dsl_dsobj_to_dsname(bpath, objnum, outpath); 1337219089Spjd *slashp = '/'; 1338219089Spjd 1339219089Spjd return (error); 1340219089Spjd} 1341219089Spjd 1342219089Spjd/* 1343251631Sdelphij * Check that the hex label string is appropriate for the dataset being 1344251631Sdelphij * mounted into the global_zone proper. 1345219089Spjd * 1346251631Sdelphij * Return an error if the hex label string is not default or 1347251631Sdelphij * admin_low/admin_high. For admin_low labels, the corresponding 1348251631Sdelphij * dataset must be readonly. 1349219089Spjd */ 1350219089Spjdint 1351219089Spjdzfs_check_global_label(const char *dsname, const char *hexsl) 1352219089Spjd{ 1353219089Spjd if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1354219089Spjd return (0); 1355219089Spjd if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1356219089Spjd return (0); 1357219089Spjd if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1358219089Spjd /* must be readonly */ 1359219089Spjd uint64_t rdonly; 1360219089Spjd 1361219089Spjd if (dsl_prop_get_integer(dsname, 1362219089Spjd zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1363249195Smm return (SET_ERROR(EACCES)); 1364219089Spjd return (rdonly ? 0 : EACCES); 1365219089Spjd } 1366249195Smm return (SET_ERROR(EACCES)); 1367219089Spjd} 1368219089Spjd 1369219089Spjd/* 1370251631Sdelphij * Determine whether the mount is allowed according to MAC check. 1371251631Sdelphij * by comparing (where appropriate) label of the dataset against 1372251631Sdelphij * the label of the zone being mounted into. If the dataset has 1373251631Sdelphij * no label, create one. 1374219089Spjd * 1375251631Sdelphij * Returns 0 if access allowed, error otherwise (e.g. EACCES) 1376219089Spjd */ 1377219089Spjdstatic int 1378219089Spjdzfs_mount_label_policy(vfs_t *vfsp, char *osname) 1379219089Spjd{ 1380219089Spjd int error, retv; 1381219089Spjd zone_t *mntzone = NULL; 1382219089Spjd ts_label_t *mnt_tsl; 1383219089Spjd bslabel_t *mnt_sl; 1384219089Spjd bslabel_t ds_sl; 1385219089Spjd char ds_hexsl[MAXNAMELEN]; 1386219089Spjd 1387219089Spjd retv = EACCES; /* assume the worst */ 1388219089Spjd 1389219089Spjd /* 1390219089Spjd * Start by getting the dataset label if it exists. 1391219089Spjd */ 1392219089Spjd error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1393219089Spjd 1, sizeof (ds_hexsl), &ds_hexsl, NULL); 1394219089Spjd if (error) 1395249195Smm return (SET_ERROR(EACCES)); 1396219089Spjd 1397219089Spjd /* 1398219089Spjd * If labeling is NOT enabled, then disallow the mount of datasets 1399219089Spjd * which have a non-default label already. No other label checks 1400219089Spjd * are needed. 1401219089Spjd */ 1402219089Spjd if (!is_system_labeled()) { 1403219089Spjd if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1404219089Spjd return (0); 1405249195Smm return (SET_ERROR(EACCES)); 1406219089Spjd } 1407219089Spjd 1408219089Spjd /* 1409219089Spjd * Get the label of the mountpoint. If mounting into the global 1410219089Spjd * zone (i.e. mountpoint is not within an active zone and the 1411219089Spjd * zoned property is off), the label must be default or 1412219089Spjd * admin_low/admin_high only; no other checks are needed. 1413219089Spjd */ 1414219089Spjd mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE); 1415219089Spjd if (mntzone->zone_id == GLOBAL_ZONEID) { 1416219089Spjd uint64_t zoned; 1417219089Spjd 1418219089Spjd zone_rele(mntzone); 1419219089Spjd 1420219089Spjd if (dsl_prop_get_integer(osname, 1421219089Spjd zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL)) 1422249195Smm return (SET_ERROR(EACCES)); 1423219089Spjd if (!zoned) 1424219089Spjd return (zfs_check_global_label(osname, ds_hexsl)); 1425219089Spjd else 1426219089Spjd /* 1427219089Spjd * This is the case of a zone dataset being mounted 1428219089Spjd * initially, before the zone has been fully created; 1429219089Spjd * allow this mount into global zone. 1430219089Spjd */ 1431219089Spjd return (0); 1432219089Spjd } 1433219089Spjd 1434219089Spjd mnt_tsl = mntzone->zone_slabel; 1435219089Spjd ASSERT(mnt_tsl != NULL); 1436219089Spjd label_hold(mnt_tsl); 1437219089Spjd mnt_sl = label2bslabel(mnt_tsl); 1438219089Spjd 1439219089Spjd if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) { 1440219089Spjd /* 1441219089Spjd * The dataset doesn't have a real label, so fabricate one. 1442219089Spjd */ 1443219089Spjd char *str = NULL; 1444219089Spjd 1445219089Spjd if (l_to_str_internal(mnt_sl, &str) == 0 && 1446248571Smm dsl_prop_set_string(osname, 1447248571Smm zfs_prop_to_name(ZFS_PROP_MLSLABEL), 1448248571Smm ZPROP_SRC_LOCAL, str) == 0) 1449219089Spjd retv = 0; 1450219089Spjd if (str != NULL) 1451219089Spjd kmem_free(str, strlen(str) + 1); 1452219089Spjd } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) { 1453219089Spjd /* 1454219089Spjd * Now compare labels to complete the MAC check. If the 1455219089Spjd * labels are equal then allow access. If the mountpoint 1456219089Spjd * label dominates the dataset label, allow readonly access. 1457219089Spjd * Otherwise, access is denied. 1458219089Spjd */ 1459219089Spjd if (blequal(mnt_sl, &ds_sl)) 1460219089Spjd retv = 0; 1461219089Spjd else if (bldominates(mnt_sl, &ds_sl)) { 1462219089Spjd vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); 1463219089Spjd retv = 0; 1464219089Spjd } 1465219089Spjd } 1466219089Spjd 1467219089Spjd label_rele(mnt_tsl); 1468219089Spjd zone_rele(mntzone); 1469219089Spjd return (retv); 1470219089Spjd} 1471219089Spjd#endif /* SECLABEL */ 1472219089Spjd 1473219089Spjd#ifdef OPENSOLARIS_MOUNTROOT 1474219089Spjdstatic int 1475219089Spjdzfs_mountroot(vfs_t *vfsp, enum whymountroot why) 1476219089Spjd{ 1477219089Spjd int error = 0; 1478219089Spjd static int zfsrootdone = 0; 1479219089Spjd zfsvfs_t *zfsvfs = NULL; 1480219089Spjd znode_t *zp = NULL; 1481219089Spjd vnode_t *vp = NULL; 1482219089Spjd char *zfs_bootfs; 1483219089Spjd char *zfs_devid; 1484219089Spjd 1485219089Spjd ASSERT(vfsp); 1486219089Spjd 1487219089Spjd /* 1488219089Spjd * The filesystem that we mount as root is defined in the 1489219089Spjd * boot property "zfs-bootfs" with a format of 1490219089Spjd * "poolname/root-dataset-objnum". 1491219089Spjd */ 1492219089Spjd if (why == ROOT_INIT) { 1493219089Spjd if (zfsrootdone++) 1494249195Smm return (SET_ERROR(EBUSY)); 1495219089Spjd /* 1496219089Spjd * the process of doing a spa_load will require the 1497219089Spjd * clock to be set before we could (for example) do 1498219089Spjd * something better by looking at the timestamp on 1499219089Spjd * an uberblock, so just set it to -1. 1500219089Spjd */ 1501219089Spjd clkset(-1); 1502219089Spjd 1503219089Spjd if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) { 1504219089Spjd cmn_err(CE_NOTE, "spa_get_bootfs: can not get " 1505219089Spjd "bootfs name"); 1506249195Smm return (SET_ERROR(EINVAL)); 1507219089Spjd } 1508219089Spjd zfs_devid = spa_get_bootprop("diskdevid"); 1509219089Spjd error = spa_import_rootpool(rootfs.bo_name, zfs_devid); 1510219089Spjd if (zfs_devid) 1511219089Spjd spa_free_bootprop(zfs_devid); 1512219089Spjd if (error) { 1513219089Spjd spa_free_bootprop(zfs_bootfs); 1514219089Spjd cmn_err(CE_NOTE, "spa_import_rootpool: error %d", 1515219089Spjd error); 1516219089Spjd return (error); 1517219089Spjd } 1518219089Spjd if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) { 1519219089Spjd spa_free_bootprop(zfs_bootfs); 1520219089Spjd cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d", 1521219089Spjd error); 1522219089Spjd return (error); 1523219089Spjd } 1524219089Spjd 1525219089Spjd spa_free_bootprop(zfs_bootfs); 1526219089Spjd 1527219089Spjd if (error = vfs_lock(vfsp)) 1528219089Spjd return (error); 1529219089Spjd 1530219089Spjd if (error = zfs_domount(vfsp, rootfs.bo_name)) { 1531219089Spjd cmn_err(CE_NOTE, "zfs_domount: error %d", error); 1532219089Spjd goto out; 1533219089Spjd } 1534219089Spjd 1535219089Spjd zfsvfs = (zfsvfs_t *)vfsp->vfs_data; 1536219089Spjd ASSERT(zfsvfs); 1537219089Spjd if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) { 1538219089Spjd cmn_err(CE_NOTE, "zfs_zget: error %d", error); 1539219089Spjd goto out; 1540219089Spjd } 1541219089Spjd 1542219089Spjd vp = ZTOV(zp); 1543219089Spjd mutex_enter(&vp->v_lock); 1544219089Spjd vp->v_flag |= VROOT; 1545219089Spjd mutex_exit(&vp->v_lock); 1546219089Spjd rootvp = vp; 1547219089Spjd 1548219089Spjd /* 1549219089Spjd * Leave rootvp held. The root file system is never unmounted. 1550219089Spjd */ 1551219089Spjd 1552219089Spjd vfs_add((struct vnode *)0, vfsp, 1553219089Spjd (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0); 1554219089Spjdout: 1555219089Spjd vfs_unlock(vfsp); 1556219089Spjd return (error); 1557219089Spjd } else if (why == ROOT_REMOUNT) { 1558219089Spjd readonly_changed_cb(vfsp->vfs_data, B_FALSE); 1559219089Spjd vfsp->vfs_flag |= VFS_REMOUNT; 1560219089Spjd 1561219089Spjd /* refresh mount options */ 1562219089Spjd zfs_unregister_callbacks(vfsp->vfs_data); 1563219089Spjd return (zfs_register_callbacks(vfsp)); 1564219089Spjd 1565219089Spjd } else if (why == ROOT_UNMOUNT) { 1566219089Spjd zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data); 1567219089Spjd (void) zfs_sync(vfsp, 0, 0); 1568219089Spjd return (0); 1569219089Spjd } 1570219089Spjd 1571219089Spjd /* 1572219089Spjd * if "why" is equal to anything else other than ROOT_INIT, 1573219089Spjd * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it. 1574219089Spjd */ 1575249195Smm return (SET_ERROR(ENOTSUP)); 1576219089Spjd} 1577219089Spjd#endif /* OPENSOLARIS_MOUNTROOT */ 1578219089Spjd 1579241286Savgstatic int 1580241286Savggetpoolname(const char *osname, char *poolname) 1581241286Savg{ 1582241286Savg char *p; 1583241286Savg 1584241286Savg p = strchr(osname, '/'); 1585241286Savg if (p == NULL) { 1586241286Savg if (strlen(osname) >= MAXNAMELEN) 1587241286Savg return (ENAMETOOLONG); 1588241286Savg (void) strcpy(poolname, osname); 1589241286Savg } else { 1590241286Savg if (p - osname >= MAXNAMELEN) 1591241286Savg return (ENAMETOOLONG); 1592241286Savg (void) strncpy(poolname, osname, p - osname); 1593241286Savg poolname[p - osname] = '\0'; 1594241286Savg } 1595241286Savg return (0); 1596241286Savg} 1597241286Savg 1598168404Spjd/*ARGSUSED*/ 1599168404Spjdstatic int 1600191990Sattiliozfs_mount(vfs_t *vfsp) 1601168404Spjd{ 1602191990Sattilio kthread_t *td = curthread; 1603185029Spjd vnode_t *mvp = vfsp->mnt_vnodecovered; 1604185029Spjd cred_t *cr = td->td_ucred; 1605185029Spjd char *osname; 1606185029Spjd int error = 0; 1607185029Spjd int canwrite; 1608168404Spjd 1609249195Smm#ifdef illumos 1610249195Smm if (mvp->v_type != VDIR) 1611249195Smm return (SET_ERROR(ENOTDIR)); 1612249195Smm 1613249195Smm mutex_enter(&mvp->v_lock); 1614249195Smm if ((uap->flags & MS_REMOUNT) == 0 && 1615249195Smm (uap->flags & MS_OVERLAY) == 0 && 1616249195Smm (mvp->v_count != 1 || (mvp->v_flag & VROOT))) { 1617249195Smm mutex_exit(&mvp->v_lock); 1618249195Smm return (SET_ERROR(EBUSY)); 1619249195Smm } 1620249195Smm mutex_exit(&mvp->v_lock); 1621249195Smm 1622249195Smm /* 1623249195Smm * ZFS does not support passing unparsed data in via MS_DATA. 1624249195Smm * Users should use the MS_OPTIONSTR interface; this means 1625249195Smm * that all option parsing is already done and the options struct 1626249195Smm * can be interrogated. 1627249195Smm */ 1628249195Smm if ((uap->flags & MS_DATA) && uap->datalen > 0) 1629249195Smm#else 1630232186Smm if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS)) 1631249195Smm return (SET_ERROR(EPERM)); 1632232186Smm 1633185029Spjd if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) 1634249195Smm return (SET_ERROR(EINVAL)); 1635249195Smm#endif /* ! illumos */ 1636185029Spjd 1637168404Spjd /* 1638185029Spjd * If full-owner-access is enabled and delegated administration is 1639185029Spjd * turned on, we must set nosuid. 1640185029Spjd */ 1641185029Spjd if (zfs_super_owner && 1642185029Spjd dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { 1643185029Spjd secpolicy_fs_mount_clearopts(cr, vfsp); 1644185029Spjd } 1645185029Spjd 1646185029Spjd /* 1647185029Spjd * Check for mount privilege? 1648185029Spjd * 1649185029Spjd * If we don't have privilege then see if 1650185029Spjd * we have local permission to allow it 1651185029Spjd */ 1652185029Spjd error = secpolicy_fs_mount(cr, mvp, vfsp); 1653185029Spjd if (error) { 1654212694Smm if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) 1655196944Spjd goto out; 1656196944Spjd 1657196944Spjd if (!(vfsp->vfs_flag & MS_REMOUNT)) { 1658185029Spjd vattr_t vattr; 1659185029Spjd 1660185029Spjd /* 1661185029Spjd * Make sure user is the owner of the mount point 1662185029Spjd * or has sufficient privileges. 1663185029Spjd */ 1664185029Spjd 1665185029Spjd vattr.va_mask = AT_UID; 1666185029Spjd 1667196662Spjd vn_lock(mvp, LK_SHARED | LK_RETRY); 1668212694Smm if (VOP_GETATTR(mvp, &vattr, cr)) { 1669196662Spjd VOP_UNLOCK(mvp, 0); 1670185029Spjd goto out; 1671185029Spjd } 1672185029Spjd 1673185029Spjd if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && 1674185029Spjd VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { 1675196662Spjd VOP_UNLOCK(mvp, 0); 1676185029Spjd goto out; 1677185029Spjd } 1678196662Spjd VOP_UNLOCK(mvp, 0); 1679196944Spjd } 1680185029Spjd 1681196944Spjd secpolicy_fs_mount_clearopts(cr, vfsp); 1682185029Spjd } 1683185029Spjd 1684185029Spjd /* 1685185029Spjd * Refuse to mount a filesystem if we are in a local zone and the 1686185029Spjd * dataset is not visible. 1687185029Spjd */ 1688185029Spjd if (!INGLOBALZONE(curthread) && 1689185029Spjd (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { 1690249195Smm error = SET_ERROR(EPERM); 1691185029Spjd goto out; 1692185029Spjd } 1693185029Spjd 1694219089Spjd#ifdef SECLABEL 1695219089Spjd error = zfs_mount_label_policy(vfsp, osname); 1696219089Spjd if (error) 1697219089Spjd goto out; 1698219089Spjd#endif 1699219089Spjd 1700218386Strasz vfsp->vfs_flag |= MNT_NFS4ACLS; 1701218386Strasz 1702185029Spjd /* 1703168404Spjd * When doing a remount, we simply refresh our temporary properties 1704168404Spjd * according to those options set in the current VFS options. 1705168404Spjd */ 1706185029Spjd if (vfsp->vfs_flag & MS_REMOUNT) { 1707185029Spjd /* refresh mount options */ 1708185029Spjd zfs_unregister_callbacks(vfsp->vfs_data); 1709185029Spjd error = zfs_register_callbacks(vfsp); 1710185029Spjd goto out; 1711185029Spjd } 1712168404Spjd 1713241286Savg /* Initial root mount: try hard to import the requested root pool. */ 1714241286Savg if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && 1715241286Savg (vfsp->vfs_flag & MNT_UPDATE) == 0) { 1716241286Savg char pname[MAXNAMELEN]; 1717241286Savg 1718241286Savg error = getpoolname(osname, pname); 1719242567Savg if (error == 0) 1720242567Savg error = spa_import_rootpool(pname); 1721241286Savg if (error) 1722241286Savg goto out; 1723241286Savg } 1724168510Spjd DROP_GIANT(); 1725185029Spjd error = zfs_domount(vfsp, osname); 1726168510Spjd PICKUP_GIANT(); 1727209962Smm 1728215260Smm#ifdef sun 1729209962Smm /* 1730209962Smm * Add an extra VFS_HOLD on our parent vfs so that it can't 1731209962Smm * disappear due to a forced unmount. 1732209962Smm */ 1733209962Smm if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap) 1734209962Smm VFS_HOLD(mvp->v_vfsp); 1735215260Smm#endif /* sun */ 1736209962Smm 1737185029Spjdout: 1738168510Spjd return (error); 1739168404Spjd} 1740168404Spjd 1741168404Spjdstatic int 1742191990Sattiliozfs_statfs(vfs_t *vfsp, struct statfs *statp) 1743169170Spjd{ 1744168404Spjd zfsvfs_t *zfsvfs = vfsp->vfs_data; 1745168404Spjd uint64_t refdbytes, availbytes, usedobjs, availobjs; 1746168404Spjd 1747168404Spjd statp->f_version = STATFS_VERSION; 1748168404Spjd 1749168404Spjd ZFS_ENTER(zfsvfs); 1750168404Spjd 1751168404Spjd dmu_objset_space(zfsvfs->z_os, 1752168404Spjd &refdbytes, &availbytes, &usedobjs, &availobjs); 1753168404Spjd 1754168404Spjd /* 1755168404Spjd * The underlying storage pool actually uses multiple block sizes. 1756168404Spjd * We report the fragsize as the smallest block size we support, 1757168404Spjd * and we report our blocksize as the filesystem's maximum blocksize. 1758168404Spjd */ 1759204101Spjd statp->f_bsize = SPA_MINBLOCKSIZE; 1760204101Spjd statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; 1761168404Spjd 1762168404Spjd /* 1763168404Spjd * The following report "total" blocks of various kinds in the 1764168404Spjd * file system, but reported in terms of f_frsize - the 1765168404Spjd * "fragment" size. 1766168404Spjd */ 1767168404Spjd 1768204101Spjd statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; 1769168404Spjd statp->f_bfree = availbytes / statp->f_bsize; 1770168404Spjd statp->f_bavail = statp->f_bfree; /* no root reservation */ 1771168404Spjd 1772168404Spjd /* 1773168404Spjd * statvfs() should really be called statufs(), because it assumes 1774168404Spjd * static metadata. ZFS doesn't preallocate files, so the best 1775168404Spjd * we can do is report the max that could possibly fit in f_files, 1776168404Spjd * and that minus the number actually used in f_ffree. 1777168404Spjd * For f_ffree, report the smaller of the number of object available 1778168404Spjd * and the number of blocks (each object will take at least a block). 1779168404Spjd */ 1780168404Spjd statp->f_ffree = MIN(availobjs, statp->f_bfree); 1781168404Spjd statp->f_files = statp->f_ffree + usedobjs; 1782168404Spjd 1783168404Spjd /* 1784168404Spjd * We're a zfs filesystem. 1785168404Spjd */ 1786168404Spjd (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename)); 1787168404Spjd 1788168404Spjd strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, 1789168404Spjd sizeof(statp->f_mntfromname)); 1790168404Spjd strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, 1791168404Spjd sizeof(statp->f_mntonname)); 1792168404Spjd 1793168404Spjd statp->f_namemax = ZFS_MAXNAMELEN; 1794168404Spjd 1795168404Spjd ZFS_EXIT(zfsvfs); 1796168404Spjd return (0); 1797168404Spjd} 1798168404Spjd 1799168404Spjdstatic int 1800191990Sattiliozfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) 1801168404Spjd{ 1802168404Spjd zfsvfs_t *zfsvfs = vfsp->vfs_data; 1803168404Spjd znode_t *rootzp; 1804168404Spjd int error; 1805168404Spjd 1806197459Spjd ZFS_ENTER_NOERROR(zfsvfs); 1807168404Spjd 1808168404Spjd error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1809219089Spjd if (error == 0) 1810219089Spjd *vpp = ZTOV(rootzp); 1811206667Spjd 1812206667Spjd ZFS_EXIT(zfsvfs); 1813206667Spjd 1814168404Spjd if (error == 0) { 1815254711Savg error = vn_lock(*vpp, flags); 1816219089Spjd if (error == 0) 1817219089Spjd (*vpp)->v_vflag |= VV_ROOT; 1818168404Spjd } 1819219089Spjd if (error != 0) 1820219089Spjd *vpp = NULL; 1821168404Spjd 1822168404Spjd return (error); 1823168404Spjd} 1824168404Spjd 1825185029Spjd/* 1826185029Spjd * Teardown the zfsvfs::z_os. 1827185029Spjd * 1828185029Spjd * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' 1829185029Spjd * and 'z_teardown_inactive_lock' held. 1830185029Spjd */ 1831185029Spjdstatic int 1832185029Spjdzfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1833185029Spjd{ 1834185029Spjd znode_t *zp; 1835185029Spjd 1836185029Spjd rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1837185029Spjd 1838185029Spjd if (!unmounting) { 1839185029Spjd /* 1840185029Spjd * We purge the parent filesystem's vfsp as the parent 1841185029Spjd * filesystem and all of its snapshots have their vnode's 1842185029Spjd * v_vfsp set to the parent's filesystem's vfsp. Note, 1843185029Spjd * 'z_parent' is self referential for non-snapshots. 1844185029Spjd */ 1845185029Spjd (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1846197351Spjd#ifdef FREEBSD_NAMECACHE 1847197351Spjd cache_purgevfs(zfsvfs->z_parent->z_vfs); 1848197351Spjd#endif 1849185029Spjd } 1850185029Spjd 1851185029Spjd /* 1852185029Spjd * Close the zil. NB: Can't close the zil while zfs_inactive 1853185029Spjd * threads are blocked as zil_close can call zfs_inactive. 1854185029Spjd */ 1855185029Spjd if (zfsvfs->z_log) { 1856185029Spjd zil_close(zfsvfs->z_log); 1857185029Spjd zfsvfs->z_log = NULL; 1858185029Spjd } 1859185029Spjd 1860185029Spjd rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1861185029Spjd 1862185029Spjd /* 1863185029Spjd * If we are not unmounting (ie: online recv) and someone already 1864185029Spjd * unmounted this file system while we were doing the switcheroo, 1865185029Spjd * or a reopen of z_os failed then just bail out now. 1866185029Spjd */ 1867185029Spjd if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1868185029Spjd rw_exit(&zfsvfs->z_teardown_inactive_lock); 1869185029Spjd rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1870249195Smm return (SET_ERROR(EIO)); 1871185029Spjd } 1872185029Spjd 1873185029Spjd /* 1874185029Spjd * At this point there are no vops active, and any new vops will 1875185029Spjd * fail with EIO since we have z_teardown_lock for writer (only 1876185029Spjd * relavent for forced unmount). 1877185029Spjd * 1878185029Spjd * Release all holds on dbufs. 1879185029Spjd */ 1880185029Spjd mutex_enter(&zfsvfs->z_znodes_lock); 1881185029Spjd for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1882185029Spjd zp = list_next(&zfsvfs->z_all_znodes, zp)) 1883219089Spjd if (zp->z_sa_hdl) { 1884196297Spjd ASSERT(ZTOV(zp)->v_count >= 0); 1885185029Spjd zfs_znode_dmu_fini(zp); 1886185029Spjd } 1887185029Spjd mutex_exit(&zfsvfs->z_znodes_lock); 1888185029Spjd 1889185029Spjd /* 1890185029Spjd * If we are unmounting, set the unmounted flag and let new vops 1891185029Spjd * unblock. zfs_inactive will have the unmounted behavior, and all 1892185029Spjd * other vops will fail with EIO. 1893185029Spjd */ 1894185029Spjd if (unmounting) { 1895185029Spjd zfsvfs->z_unmounted = B_TRUE; 1896185029Spjd rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1897185029Spjd rw_exit(&zfsvfs->z_teardown_inactive_lock); 1898185029Spjd } 1899185029Spjd 1900185029Spjd /* 1901185029Spjd * z_os will be NULL if there was an error in attempting to reopen 1902185029Spjd * zfsvfs, so just return as the properties had already been 1903185029Spjd * unregistered and cached data had been evicted before. 1904185029Spjd */ 1905185029Spjd if (zfsvfs->z_os == NULL) 1906185029Spjd return (0); 1907185029Spjd 1908185029Spjd /* 1909185029Spjd * Unregister properties. 1910185029Spjd */ 1911185029Spjd zfs_unregister_callbacks(zfsvfs); 1912185029Spjd 1913185029Spjd /* 1914185029Spjd * Evict cached data 1915185029Spjd */ 1916239774Smm if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) && 1917239774Smm !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)) 1918239774Smm txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1919248571Smm dmu_objset_evict_dbufs(zfsvfs->z_os); 1920185029Spjd 1921185029Spjd return (0); 1922185029Spjd} 1923185029Spjd 1924168404Spjd/*ARGSUSED*/ 1925168404Spjdstatic int 1926191990Sattiliozfs_umount(vfs_t *vfsp, int fflag) 1927168404Spjd{ 1928209962Smm kthread_t *td = curthread; 1929168404Spjd zfsvfs_t *zfsvfs = vfsp->vfs_data; 1930185029Spjd objset_t *os; 1931209962Smm cred_t *cr = td->td_ucred; 1932168404Spjd int ret; 1933168404Spjd 1934185029Spjd ret = secpolicy_fs_unmount(cr, vfsp); 1935185029Spjd if (ret) { 1936212694Smm if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource), 1937212694Smm ZFS_DELEG_PERM_MOUNT, cr)) 1938185029Spjd return (ret); 1939185029Spjd } 1940219089Spjd 1941185029Spjd /* 1942185029Spjd * We purge the parent filesystem's vfsp as the parent filesystem 1943185029Spjd * and all of its snapshots have their vnode's v_vfsp set to the 1944185029Spjd * parent's filesystem's vfsp. Note, 'z_parent' is self 1945185029Spjd * referential for non-snapshots. 1946185029Spjd */ 1947185029Spjd (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0); 1948168404Spjd 1949168404Spjd /* 1950168404Spjd * Unmount any snapshots mounted under .zfs before unmounting the 1951168404Spjd * dataset itself. 1952168404Spjd */ 1953169170Spjd if (zfsvfs->z_ctldir != NULL) { 1954168404Spjd if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) 1955168404Spjd return (ret); 1956209962Smm ret = vflush(vfsp, 0, 0, td); 1957168404Spjd ASSERT(ret == EBUSY); 1958168404Spjd if (!(fflag & MS_FORCE)) { 1959168404Spjd if (zfsvfs->z_ctldir->v_count > 1) 1960168404Spjd return (EBUSY); 1961168404Spjd ASSERT(zfsvfs->z_ctldir->v_count == 1); 1962168404Spjd } 1963168404Spjd zfsctl_destroy(zfsvfs); 1964168404Spjd ASSERT(zfsvfs->z_ctldir == NULL); 1965168404Spjd } 1966168404Spjd 1967197459Spjd if (fflag & MS_FORCE) { 1968197459Spjd /* 1969197459Spjd * Mark file system as unmounted before calling 1970197459Spjd * vflush(FORCECLOSE). This way we ensure no future vnops 1971197459Spjd * will be called and risk operating on DOOMED vnodes. 1972197459Spjd */ 1973197459Spjd rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG); 1974197459Spjd zfsvfs->z_unmounted = B_TRUE; 1975197459Spjd rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 1976197459Spjd } 1977197459Spjd 1978168404Spjd /* 1979168404Spjd * Flush all the files. 1980168404Spjd */ 1981209962Smm ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); 1982168404Spjd if (ret != 0) { 1983168404Spjd if (!zfsvfs->z_issnap) { 1984168404Spjd zfsctl_create(zfsvfs); 1985168404Spjd ASSERT(zfsvfs->z_ctldir != NULL); 1986168404Spjd } 1987168404Spjd return (ret); 1988168404Spjd } 1989168404Spjd 1990248653Swill#ifdef sun 1991185029Spjd if (!(fflag & MS_FORCE)) { 1992185029Spjd /* 1993185029Spjd * Check the number of active vnodes in the file system. 1994185029Spjd * Our count is maintained in the vfs structure, but the 1995185029Spjd * number is off by 1 to indicate a hold on the vfs 1996185029Spjd * structure itself. 1997185029Spjd * 1998185029Spjd * The '.zfs' directory maintains a reference of its 1999185029Spjd * own, and any active references underneath are 2000185029Spjd * reflected in the vnode count. 2001185029Spjd */ 2002185029Spjd if (zfsvfs->z_ctldir == NULL) { 2003185029Spjd if (vfsp->vfs_count > 1) 2004249195Smm return (SET_ERROR(EBUSY)); 2005185029Spjd } else { 2006185029Spjd if (vfsp->vfs_count > 2 || 2007185029Spjd zfsvfs->z_ctldir->v_count > 1) 2008249195Smm return (SET_ERROR(EBUSY)); 2009185029Spjd } 2010185029Spjd } 2011248653Swill#endif 2012168404Spjd 2013185029Spjd VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 2014185029Spjd os = zfsvfs->z_os; 2015185029Spjd 2016185029Spjd /* 2017185029Spjd * z_os will be NULL if there was an error in 2018185029Spjd * attempting to reopen zfsvfs. 2019185029Spjd */ 2020185029Spjd if (os != NULL) { 2021168404Spjd /* 2022185029Spjd * Unset the objset user_ptr. 2023168404Spjd */ 2024219089Spjd mutex_enter(&os->os_user_ptr_lock); 2025185029Spjd dmu_objset_set_user(os, NULL); 2026219089Spjd mutex_exit(&os->os_user_ptr_lock); 2027185029Spjd 2028185029Spjd /* 2029185029Spjd * Finally release the objset 2030185029Spjd */ 2031219089Spjd dmu_objset_disown(os, zfsvfs); 2032168404Spjd } 2033168404Spjd 2034185029Spjd /* 2035185029Spjd * We can now safely destroy the '.zfs' directory node. 2036185029Spjd */ 2037185029Spjd if (zfsvfs->z_ctldir != NULL) 2038185029Spjd zfsctl_destroy(zfsvfs); 2039185029Spjd if (zfsvfs->z_issnap) { 2040185029Spjd vnode_t *svp = vfsp->mnt_vnodecovered; 2041185029Spjd 2042197515Spjd if (svp->v_count >= 2) 2043192211Skmacy VN_RELE(svp); 2044185029Spjd } 2045168404Spjd zfs_freevfs(vfsp); 2046168404Spjd 2047168404Spjd return (0); 2048168404Spjd} 2049168404Spjd 2050168404Spjdstatic int 2051168404Spjdzfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) 2052168404Spjd{ 2053168404Spjd zfsvfs_t *zfsvfs = vfsp->vfs_data; 2054168404Spjd znode_t *zp; 2055168404Spjd int err; 2056168404Spjd 2057197167Spjd /* 2058215397Savg * zfs_zget() can't operate on virtual entries like .zfs/ or 2059211855Spjd * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. 2060211855Spjd * This will make NFS to switch to LOOKUP instead of using VGET. 2061197167Spjd */ 2062246532Savg if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || 2063246532Savg (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) 2064197167Spjd return (EOPNOTSUPP); 2065197167Spjd 2066168404Spjd ZFS_ENTER(zfsvfs); 2067168404Spjd err = zfs_zget(zfsvfs, ino, &zp); 2068168404Spjd if (err == 0 && zp->z_unlinked) { 2069168404Spjd VN_RELE(ZTOV(zp)); 2070168404Spjd err = EINVAL; 2071168404Spjd } 2072219089Spjd if (err == 0) 2073219089Spjd *vpp = ZTOV(zp); 2074206667Spjd ZFS_EXIT(zfsvfs); 2075219089Spjd if (err == 0) 2076254711Savg err = vn_lock(*vpp, flags); 2077168404Spjd if (err != 0) 2078168404Spjd *vpp = NULL; 2079245409Skib else 2080245409Skib (*vpp)->v_hash = ino; 2081171063Sdfr return (err); 2082168404Spjd} 2083168404Spjd 2084168404Spjdstatic int 2085196982Spjdzfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, 2086196982Spjd struct ucred **credanonp, int *numsecflavors, int **secflavors) 2087196982Spjd{ 2088196982Spjd zfsvfs_t *zfsvfs = vfsp->vfs_data; 2089196982Spjd 2090196982Spjd /* 2091196982Spjd * If this is regular file system vfsp is the same as 2092196982Spjd * zfsvfs->z_parent->z_vfs, but if it is snapshot, 2093196982Spjd * zfsvfs->z_parent->z_vfs represents parent file system 2094196982Spjd * which we have to use here, because only this file system 2095196982Spjd * has mnt_export configured. 2096196982Spjd */ 2097196982Spjd return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, 2098196982Spjd credanonp, numsecflavors, secflavors)); 2099196982Spjd} 2100196982Spjd 2101197151SpjdCTASSERT(SHORT_FID_LEN <= sizeof(struct fid)); 2102197151SpjdCTASSERT(LONG_FID_LEN <= sizeof(struct fid)); 2103196982Spjd 2104196982Spjdstatic int 2105222167Srmacklemzfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) 2106168404Spjd{ 2107168404Spjd zfsvfs_t *zfsvfs = vfsp->vfs_data; 2108168404Spjd znode_t *zp; 2109168404Spjd uint64_t object = 0; 2110168404Spjd uint64_t fid_gen = 0; 2111168404Spjd uint64_t gen_mask; 2112168404Spjd uint64_t zp_gen; 2113219089Spjd int i, err; 2114168404Spjd 2115168404Spjd *vpp = NULL; 2116168404Spjd 2117168404Spjd ZFS_ENTER(zfsvfs); 2118168404Spjd 2119196979Spjd /* 2120197177Spjd * On FreeBSD we can get snapshot's mount point or its parent file 2121197177Spjd * system mount point depending if snapshot is already mounted or not. 2122196979Spjd */ 2123197177Spjd if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { 2124168404Spjd zfid_long_t *zlfid = (zfid_long_t *)fidp; 2125168404Spjd uint64_t objsetid = 0; 2126168404Spjd uint64_t setgen = 0; 2127168404Spjd 2128168404Spjd for (i = 0; i < sizeof (zlfid->zf_setid); i++) 2129168404Spjd objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 2130168404Spjd 2131168404Spjd for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 2132168404Spjd setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 2133168404Spjd 2134168404Spjd ZFS_EXIT(zfsvfs); 2135168404Spjd 2136168404Spjd err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); 2137168404Spjd if (err) 2138249195Smm return (SET_ERROR(EINVAL)); 2139168404Spjd ZFS_ENTER(zfsvfs); 2140168404Spjd } 2141168404Spjd 2142168404Spjd if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 2143168404Spjd zfid_short_t *zfid = (zfid_short_t *)fidp; 2144168404Spjd 2145168404Spjd for (i = 0; i < sizeof (zfid->zf_object); i++) 2146168404Spjd object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 2147168404Spjd 2148168404Spjd for (i = 0; i < sizeof (zfid->zf_gen); i++) 2149168404Spjd fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 2150168404Spjd } else { 2151168404Spjd ZFS_EXIT(zfsvfs); 2152249195Smm return (SET_ERROR(EINVAL)); 2153168404Spjd } 2154168404Spjd 2155246532Savg /* 2156246532Savg * A zero fid_gen means we are in .zfs or the .zfs/snapshot 2157246532Savg * directory tree. If the object == zfsvfs->z_shares_dir, then 2158246532Savg * we are in the .zfs/shares directory tree. 2159246532Savg */ 2160246532Savg if ((fid_gen == 0 && 2161246532Savg (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || 2162246532Savg (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { 2163168404Spjd *vpp = zfsvfs->z_ctldir; 2164168404Spjd ASSERT(*vpp != NULL); 2165168404Spjd if (object == ZFSCTL_INO_SNAPDIR) { 2166168404Spjd VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL, 2167185029Spjd 0, NULL, NULL, NULL, NULL, NULL) == 0); 2168246532Savg } else if (object == zfsvfs->z_shares_dir) { 2169246532Savg VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL, 2170246532Savg 0, NULL, NULL, NULL, NULL, NULL) == 0); 2171168404Spjd } else { 2172168404Spjd VN_HOLD(*vpp); 2173168404Spjd } 2174206667Spjd ZFS_EXIT(zfsvfs); 2175254711Savg err = vn_lock(*vpp, flags); 2176219089Spjd if (err != 0) 2177219089Spjd *vpp = NULL; 2178219089Spjd return (err); 2179168404Spjd } 2180168404Spjd 2181168404Spjd gen_mask = -1ULL >> (64 - 8 * i); 2182168404Spjd 2183168404Spjd dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask); 2184168404Spjd if (err = zfs_zget(zfsvfs, object, &zp)) { 2185168404Spjd ZFS_EXIT(zfsvfs); 2186168404Spjd return (err); 2187168404Spjd } 2188219089Spjd (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 2189219089Spjd sizeof (uint64_t)); 2190219089Spjd zp_gen = zp_gen & gen_mask; 2191168404Spjd if (zp_gen == 0) 2192168404Spjd zp_gen = 1; 2193168404Spjd if (zp->z_unlinked || zp_gen != fid_gen) { 2194168404Spjd dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen); 2195168404Spjd VN_RELE(ZTOV(zp)); 2196168404Spjd ZFS_EXIT(zfsvfs); 2197249195Smm return (SET_ERROR(EINVAL)); 2198168404Spjd } 2199168404Spjd 2200219089Spjd *vpp = ZTOV(zp); 2201206667Spjd ZFS_EXIT(zfsvfs); 2202254711Savg err = vn_lock(*vpp, flags | LK_RETRY); 2203219089Spjd if (err == 0) 2204219089Spjd vnode_create_vobject(*vpp, zp->z_size, curthread); 2205219089Spjd else 2206219089Spjd *vpp = NULL; 2207219089Spjd return (err); 2208168404Spjd} 2209168404Spjd 2210185029Spjd/* 2211185029Spjd * Block out VOPs and close zfsvfs_t::z_os 2212185029Spjd * 2213185029Spjd * Note, if successful, then we return with the 'z_teardown_lock' and 2214253816Sdelphij * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 2215253816Sdelphij * dataset and objset intact so that they can be atomically handed off during 2216253816Sdelphij * a subsequent rollback or recv operation and the resume thereafter. 2217185029Spjd */ 2218185029Spjdint 2219219089Spjdzfs_suspend_fs(zfsvfs_t *zfsvfs) 2220168404Spjd{ 2221185029Spjd int error; 2222168404Spjd 2223185029Spjd if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 2224185029Spjd return (error); 2225168404Spjd 2226185029Spjd return (0); 2227185029Spjd} 2228168404Spjd 2229185029Spjd/* 2230253816Sdelphij * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 2231253816Sdelphij * is an invariant across any of the operations that can be performed while the 2232253816Sdelphij * filesystem was suspended. Whether it succeeded or failed, the preconditions 2233253816Sdelphij * are the same: the relevant objset and associated dataset are owned by 2234253816Sdelphij * zfsvfs, held, and long held on entry. 2235185029Spjd */ 2236185029Spjdint 2237219089Spjdzfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname) 2238185029Spjd{ 2239185029Spjd int err; 2240253816Sdelphij znode_t *zp; 2241253816Sdelphij uint64_t sa_obj = 0; 2242168404Spjd 2243185029Spjd ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock)); 2244185029Spjd ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 2245185029Spjd 2246253816Sdelphij /* 2247253816Sdelphij * We already own this, so just hold and rele it to update the 2248253816Sdelphij * objset_t, as the one we had before may have been evicted. 2249253816Sdelphij */ 2250253816Sdelphij VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os)); 2251253816Sdelphij VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs); 2252253816Sdelphij VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset)); 2253253816Sdelphij dmu_objset_rele(zfsvfs->z_os, zfsvfs); 2254185029Spjd 2255253816Sdelphij /* 2256253816Sdelphij * Make sure version hasn't changed 2257253816Sdelphij */ 2258219089Spjd 2259253816Sdelphij err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION, 2260253816Sdelphij &zfsvfs->z_version); 2261219089Spjd 2262253816Sdelphij if (err) 2263253816Sdelphij goto bail; 2264219089Spjd 2265253816Sdelphij err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ, 2266253816Sdelphij ZFS_SA_ATTRS, 8, 1, &sa_obj); 2267219089Spjd 2268253816Sdelphij if (err && zfsvfs->z_version >= ZPL_VERSION_SA) 2269253816Sdelphij goto bail; 2270219089Spjd 2271253816Sdelphij if ((err = sa_setup(zfsvfs->z_os, sa_obj, 2272253816Sdelphij zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0) 2273253816Sdelphij goto bail; 2274219089Spjd 2275253816Sdelphij if (zfsvfs->z_version >= ZPL_VERSION_SA) 2276253816Sdelphij sa_register_update_callback(zfsvfs->z_os, 2277253816Sdelphij zfs_sa_upgrade); 2278219089Spjd 2279253816Sdelphij VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 2280185029Spjd 2281253816Sdelphij zfs_set_fuid_feature(zfsvfs); 2282219089Spjd 2283253816Sdelphij /* 2284253816Sdelphij * Attempt to re-establish all the active znodes with 2285253816Sdelphij * their dbufs. If a zfs_rezget() fails, then we'll let 2286253816Sdelphij * any potential callers discover that via ZFS_ENTER_VERIFY_VP 2287253816Sdelphij * when they try to use their znode. 2288253816Sdelphij */ 2289253816Sdelphij mutex_enter(&zfsvfs->z_znodes_lock); 2290253816Sdelphij for (zp = list_head(&zfsvfs->z_all_znodes); zp; 2291253816Sdelphij zp = list_next(&zfsvfs->z_all_znodes, zp)) { 2292253816Sdelphij (void) zfs_rezget(zp); 2293168404Spjd } 2294253816Sdelphij mutex_exit(&zfsvfs->z_znodes_lock); 2295168404Spjd 2296219089Spjdbail: 2297185029Spjd /* release the VOPs */ 2298185029Spjd rw_exit(&zfsvfs->z_teardown_inactive_lock); 2299185029Spjd rrw_exit(&zfsvfs->z_teardown_lock, FTAG); 2300185029Spjd 2301185029Spjd if (err) { 2302185029Spjd /* 2303253816Sdelphij * Since we couldn't setup the sa framework, try to force 2304253816Sdelphij * unmount this file system. 2305185029Spjd */ 2306185029Spjd if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) 2307185029Spjd (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); 2308168404Spjd } 2309185029Spjd return (err); 2310168404Spjd} 2311168404Spjd 2312168404Spjdstatic void 2313168404Spjdzfs_freevfs(vfs_t *vfsp) 2314168404Spjd{ 2315168404Spjd zfsvfs_t *zfsvfs = vfsp->vfs_data; 2316168404Spjd 2317215260Smm#ifdef sun 2318209962Smm /* 2319209962Smm * If this is a snapshot, we have an extra VFS_HOLD on our parent 2320219089Spjd * from zfs_mount(). Release it here. If we came through 2321219089Spjd * zfs_mountroot() instead, we didn't grab an extra hold, so 2322219089Spjd * skip the VFS_RELE for rootvfs. 2323209962Smm */ 2324219089Spjd if (zfsvfs->z_issnap && (vfsp != rootvfs)) 2325209962Smm VFS_RELE(zfsvfs->z_parent->z_vfs); 2326215260Smm#endif /* sun */ 2327168404Spjd 2328209962Smm zfsvfs_free(zfsvfs); 2329185029Spjd 2330168404Spjd atomic_add_32(&zfs_active_fs_count, -1); 2331168404Spjd} 2332168404Spjd 2333172135Spjd#ifdef __i386__ 2334172135Spjdstatic int desiredvnodes_backup; 2335172135Spjd#endif 2336172135Spjd 2337172135Spjdstatic void 2338172135Spjdzfs_vnodes_adjust(void) 2339172135Spjd{ 2340172135Spjd#ifdef __i386__ 2341185029Spjd int newdesiredvnodes; 2342172135Spjd 2343172135Spjd desiredvnodes_backup = desiredvnodes; 2344172135Spjd 2345172135Spjd /* 2346172135Spjd * We calculate newdesiredvnodes the same way it is done in 2347172135Spjd * vntblinit(). If it is equal to desiredvnodes, it means that 2348172135Spjd * it wasn't tuned by the administrator and we can tune it down. 2349172135Spjd */ 2350185029Spjd newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * 2351185029Spjd vm_kmem_size / (5 * (sizeof(struct vm_object) + 2352185029Spjd sizeof(struct vnode)))); 2353185029Spjd if (newdesiredvnodes == desiredvnodes) 2354185029Spjd desiredvnodes = (3 * newdesiredvnodes) / 4; 2355172135Spjd#endif 2356172135Spjd} 2357172135Spjd 2358172135Spjdstatic void 2359172135Spjdzfs_vnodes_adjust_back(void) 2360172135Spjd{ 2361172135Spjd 2362172135Spjd#ifdef __i386__ 2363172135Spjd desiredvnodes = desiredvnodes_backup; 2364172135Spjd#endif 2365172135Spjd} 2366172135Spjd 2367168404Spjdvoid 2368168404Spjdzfs_init(void) 2369168404Spjd{ 2370168404Spjd 2371236884Smm printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); 2372168404Spjd 2373168404Spjd /* 2374219089Spjd * Initialize .zfs directory structures 2375168404Spjd */ 2376219089Spjd zfsctl_init(); 2377168404Spjd 2378168404Spjd /* 2379219089Spjd * Initialize znode cache, vnode ops, etc... 2380168404Spjd */ 2381219089Spjd zfs_znode_init(); 2382172135Spjd 2383172135Spjd /* 2384219089Spjd * Reduce number of vnodes. Originally number of vnodes is calculated 2385172135Spjd * with UFS inode in mind. We reduce it here, because it's too big for 2386172135Spjd * ZFS/i386. 2387172135Spjd */ 2388172135Spjd zfs_vnodes_adjust(); 2389209962Smm 2390209962Smm dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); 2391168404Spjd} 2392168404Spjd 2393168404Spjdvoid 2394168404Spjdzfs_fini(void) 2395168404Spjd{ 2396168404Spjd zfsctl_fini(); 2397168404Spjd zfs_znode_fini(); 2398172135Spjd zfs_vnodes_adjust_back(); 2399168404Spjd} 2400168404Spjd 2401168404Spjdint 2402168404Spjdzfs_busy(void) 2403168404Spjd{ 2404168404Spjd return (zfs_active_fs_count != 0); 2405168404Spjd} 2406185029Spjd 2407185029Spjdint 2408209962Smmzfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2409185029Spjd{ 2410185029Spjd int error; 2411209962Smm objset_t *os = zfsvfs->z_os; 2412185029Spjd dmu_tx_t *tx; 2413185029Spjd 2414185029Spjd if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2415249195Smm return (SET_ERROR(EINVAL)); 2416185029Spjd 2417209962Smm if (newvers < zfsvfs->z_version) 2418249195Smm return (SET_ERROR(EINVAL)); 2419185029Spjd 2420219089Spjd if (zfs_spa_version_map(newvers) > 2421219089Spjd spa_version(dmu_objset_spa(zfsvfs->z_os))) 2422249195Smm return (SET_ERROR(ENOTSUP)); 2423219089Spjd 2424185029Spjd tx = dmu_tx_create(os); 2425209962Smm dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2426219089Spjd if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2427219089Spjd dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2428219089Spjd ZFS_SA_ATTRS); 2429219089Spjd dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2430219089Spjd } 2431185029Spjd error = dmu_tx_assign(tx, TXG_WAIT); 2432185029Spjd if (error) { 2433185029Spjd dmu_tx_abort(tx); 2434209962Smm return (error); 2435185029Spjd } 2436219089Spjd 2437209962Smm error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2438209962Smm 8, 1, &newvers, tx); 2439185029Spjd 2440209962Smm if (error) { 2441209962Smm dmu_tx_commit(tx); 2442209962Smm return (error); 2443209962Smm } 2444209962Smm 2445219089Spjd if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2446219089Spjd uint64_t sa_obj; 2447219089Spjd 2448219089Spjd ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2449219089Spjd SPA_VERSION_SA); 2450219089Spjd sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2451219089Spjd DMU_OT_NONE, 0, tx); 2452219089Spjd 2453219089Spjd error = zap_add(os, MASTER_NODE_OBJ, 2454219089Spjd ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2455240415Smm ASSERT0(error); 2456219089Spjd 2457219089Spjd VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2458219089Spjd sa_register_update_callback(os, zfs_sa_upgrade); 2459219089Spjd } 2460219089Spjd 2461248571Smm spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2462248571Smm "from %llu to %llu", zfsvfs->z_version, newvers); 2463209962Smm 2464185029Spjd dmu_tx_commit(tx); 2465185029Spjd 2466209962Smm zfsvfs->z_version = newvers; 2467209962Smm 2468219089Spjd zfs_set_fuid_feature(zfsvfs); 2469209962Smm 2470209962Smm return (0); 2471185029Spjd} 2472219089Spjd 2473185029Spjd/* 2474185029Spjd * Read a property stored within the master node. 2475185029Spjd */ 2476185029Spjdint 2477185029Spjdzfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2478185029Spjd{ 2479185029Spjd const char *pname; 2480185029Spjd int error = ENOENT; 2481185029Spjd 2482185029Spjd /* 2483185029Spjd * Look up the file system's value for the property. For the 2484185029Spjd * version property, we look up a slightly different string. 2485185029Spjd */ 2486185029Spjd if (prop == ZFS_PROP_VERSION) 2487185029Spjd pname = ZPL_VERSION_STR; 2488185029Spjd else 2489185029Spjd pname = zfs_prop_to_name(prop); 2490185029Spjd 2491185029Spjd if (os != NULL) 2492185029Spjd error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2493185029Spjd 2494185029Spjd if (error == ENOENT) { 2495185029Spjd /* No value set, use the default value */ 2496185029Spjd switch (prop) { 2497185029Spjd case ZFS_PROP_VERSION: 2498185029Spjd *value = ZPL_VERSION; 2499185029Spjd break; 2500185029Spjd case ZFS_PROP_NORMALIZE: 2501185029Spjd case ZFS_PROP_UTF8ONLY: 2502185029Spjd *value = 0; 2503185029Spjd break; 2504185029Spjd case ZFS_PROP_CASE: 2505185029Spjd *value = ZFS_CASE_SENSITIVE; 2506185029Spjd break; 2507185029Spjd default: 2508185029Spjd return (error); 2509185029Spjd } 2510185029Spjd error = 0; 2511185029Spjd } 2512185029Spjd return (error); 2513185029Spjd} 2514226676Spjd 2515226676Spjd#ifdef _KERNEL 2516226676Spjdvoid 2517226676Spjdzfsvfs_update_fromname(const char *oldname, const char *newname) 2518226676Spjd{ 2519226676Spjd char tmpbuf[MAXPATHLEN]; 2520226676Spjd struct mount *mp; 2521226676Spjd char *fromname; 2522226676Spjd size_t oldlen; 2523226676Spjd 2524226676Spjd oldlen = strlen(oldname); 2525226676Spjd 2526226676Spjd mtx_lock(&mountlist_mtx); 2527226676Spjd TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2528226676Spjd fromname = mp->mnt_stat.f_mntfromname; 2529226676Spjd if (strcmp(fromname, oldname) == 0) { 2530226676Spjd (void)strlcpy(fromname, newname, 2531226676Spjd sizeof(mp->mnt_stat.f_mntfromname)); 2532226676Spjd continue; 2533226676Spjd } 2534226676Spjd if (strncmp(fromname, oldname, oldlen) == 0 && 2535226700Spjd (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { 2536226676Spjd (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s", 2537226676Spjd newname, fromname + oldlen); 2538226676Spjd (void)strlcpy(fromname, tmpbuf, 2539226676Spjd sizeof(mp->mnt_stat.f_mntfromname)); 2540226676Spjd continue; 2541226676Spjd } 2542226676Spjd } 2543226676Spjd mtx_unlock(&mountlist_mtx); 2544226676Spjd} 2545226676Spjd#endif 2546