1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2018 by Delphix. All rights reserved. 24 */ 25 26/* Portions Copyright 2010 Robert Milkowski */ 27 28#include <sys/types.h> 29#include <sys/param.h> 30#include <sys/sysmacros.h> 31#include <sys/kmem.h> 32#include <sys/pathname.h> 33#include <sys/vnode.h> 34#include <sys/vfs.h> 35#include <sys/mntent.h> 36#include <sys/cmn_err.h> 37#include <sys/zfs_znode.h> 38#include <sys/zfs_vnops.h> 39#include <sys/zfs_dir.h> 40#include <sys/zil.h> 41#include <sys/fs/zfs.h> 42#include <sys/dmu.h> 43#include <sys/dsl_prop.h> 44#include <sys/dsl_dataset.h> 45#include <sys/dsl_deleg.h> 46#include <sys/spa.h> 47#include <sys/zap.h> 48#include <sys/sa.h> 49#include <sys/sa_impl.h> 50#include <sys/policy.h> 51#include <sys/atomic.h> 52#include <sys/zfs_ioctl.h> 53#include <sys/zfs_ctldir.h> 54#include <sys/zfs_fuid.h> 55#include <sys/zfs_quota.h> 56#include <sys/sunddi.h> 57#include <sys/dmu_objset.h> 58#include <sys/dsl_dir.h> 59#include <sys/spa_boot.h> 60#include <sys/objlist.h> 61#include <sys/zpl.h> 62#include <linux/vfs_compat.h> 63#include "zfs_comutil.h" 64 65enum { 66 TOKEN_RO, 67 TOKEN_RW, 68 TOKEN_SETUID, 69 TOKEN_NOSETUID, 70 TOKEN_EXEC, 71 TOKEN_NOEXEC, 72 TOKEN_DEVICES, 73 TOKEN_NODEVICES, 74 TOKEN_DIRXATTR, 75 TOKEN_SAXATTR, 76 TOKEN_XATTR, 77 TOKEN_NOXATTR, 78 TOKEN_ATIME, 79 TOKEN_NOATIME, 80 TOKEN_RELATIME, 81 TOKEN_NORELATIME, 82 TOKEN_NBMAND, 83 TOKEN_NONBMAND, 84 TOKEN_MNTPOINT, 85 TOKEN_LAST, 86}; 87 88static const match_table_t zpl_tokens = { 89 { TOKEN_RO, MNTOPT_RO }, 90 { TOKEN_RW, MNTOPT_RW }, 91 { TOKEN_SETUID, MNTOPT_SETUID }, 92 { TOKEN_NOSETUID, MNTOPT_NOSETUID }, 93 { TOKEN_EXEC, MNTOPT_EXEC }, 94 { TOKEN_NOEXEC, MNTOPT_NOEXEC }, 95 { TOKEN_DEVICES, MNTOPT_DEVICES }, 96 { TOKEN_NODEVICES, MNTOPT_NODEVICES }, 97 { TOKEN_DIRXATTR, MNTOPT_DIRXATTR }, 98 { TOKEN_SAXATTR, MNTOPT_SAXATTR }, 99 { TOKEN_XATTR, MNTOPT_XATTR }, 100 { TOKEN_NOXATTR, MNTOPT_NOXATTR }, 101 { TOKEN_ATIME, MNTOPT_ATIME }, 102 { TOKEN_NOATIME, MNTOPT_NOATIME }, 103 { TOKEN_RELATIME, MNTOPT_RELATIME }, 104 { TOKEN_NORELATIME, MNTOPT_NORELATIME }, 105 { TOKEN_NBMAND, MNTOPT_NBMAND }, 106 { TOKEN_NONBMAND, MNTOPT_NONBMAND }, 107 { TOKEN_MNTPOINT, MNTOPT_MNTPOINT "=%s" }, 108 { TOKEN_LAST, NULL }, 109}; 110 111static void 112zfsvfs_vfs_free(vfs_t *vfsp) 113{ 114 if (vfsp != NULL) { 115 if (vfsp->vfs_mntpoint != NULL) 116 kmem_strfree(vfsp->vfs_mntpoint); 117 118 kmem_free(vfsp, sizeof (vfs_t)); 119 } 120} 121 122static int 123zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp) 124{ 125 switch (token) { 126 case TOKEN_RO: 127 vfsp->vfs_readonly = B_TRUE; 128 vfsp->vfs_do_readonly = B_TRUE; 129 break; 130 case TOKEN_RW: 131 vfsp->vfs_readonly = B_FALSE; 132 vfsp->vfs_do_readonly = B_TRUE; 133 break; 134 case TOKEN_SETUID: 135 vfsp->vfs_setuid = B_TRUE; 136 vfsp->vfs_do_setuid = B_TRUE; 137 break; 138 case TOKEN_NOSETUID: 139 vfsp->vfs_setuid = B_FALSE; 140 vfsp->vfs_do_setuid = B_TRUE; 141 break; 142 case TOKEN_EXEC: 143 vfsp->vfs_exec = B_TRUE; 144 vfsp->vfs_do_exec = B_TRUE; 145 break; 146 case TOKEN_NOEXEC: 147 vfsp->vfs_exec = B_FALSE; 148 vfsp->vfs_do_exec = B_TRUE; 149 break; 150 case TOKEN_DEVICES: 151 vfsp->vfs_devices = B_TRUE; 152 vfsp->vfs_do_devices = B_TRUE; 153 break; 154 case TOKEN_NODEVICES: 155 vfsp->vfs_devices = B_FALSE; 156 vfsp->vfs_do_devices = B_TRUE; 157 break; 158 case TOKEN_DIRXATTR: 159 vfsp->vfs_xattr = ZFS_XATTR_DIR; 160 vfsp->vfs_do_xattr = B_TRUE; 161 break; 162 case TOKEN_SAXATTR: 163 vfsp->vfs_xattr = ZFS_XATTR_SA; 164 vfsp->vfs_do_xattr = B_TRUE; 165 break; 166 case TOKEN_XATTR: 167 vfsp->vfs_xattr = ZFS_XATTR_DIR; 168 vfsp->vfs_do_xattr = B_TRUE; 169 break; 170 case TOKEN_NOXATTR: 171 vfsp->vfs_xattr = ZFS_XATTR_OFF; 172 vfsp->vfs_do_xattr = B_TRUE; 173 break; 174 case TOKEN_ATIME: 175 vfsp->vfs_atime = B_TRUE; 176 vfsp->vfs_do_atime = B_TRUE; 177 break; 178 case TOKEN_NOATIME: 179 vfsp->vfs_atime = B_FALSE; 180 vfsp->vfs_do_atime = B_TRUE; 181 break; 182 case TOKEN_RELATIME: 183 vfsp->vfs_relatime = B_TRUE; 184 vfsp->vfs_do_relatime = B_TRUE; 185 break; 186 case TOKEN_NORELATIME: 187 vfsp->vfs_relatime = B_FALSE; 188 vfsp->vfs_do_relatime = B_TRUE; 189 break; 190 case TOKEN_NBMAND: 191 vfsp->vfs_nbmand = B_TRUE; 192 vfsp->vfs_do_nbmand = B_TRUE; 193 break; 194 case TOKEN_NONBMAND: 195 vfsp->vfs_nbmand = B_FALSE; 196 vfsp->vfs_do_nbmand = B_TRUE; 197 break; 198 case TOKEN_MNTPOINT: 199 vfsp->vfs_mntpoint = match_strdup(&args[0]); 200 if (vfsp->vfs_mntpoint == NULL) 201 return (SET_ERROR(ENOMEM)); 202 203 break; 204 default: 205 break; 206 } 207 208 return (0); 209} 210 211/* 212 * Parse the raw mntopts and return a vfs_t describing the options. 213 */ 214static int 215zfsvfs_parse_options(char *mntopts, vfs_t **vfsp) 216{ 217 vfs_t *tmp_vfsp; 218 int error; 219 220 tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP); 221 222 if (mntopts != NULL) { 223 substring_t args[MAX_OPT_ARGS]; 224 char *tmp_mntopts, *p, *t; 225 int token; 226 227 tmp_mntopts = t = kmem_strdup(mntopts); 228 if (tmp_mntopts == NULL) 229 return (SET_ERROR(ENOMEM)); 230 231 while ((p = strsep(&t, ",")) != NULL) { 232 if (!*p) 233 continue; 234 235 args[0].to = args[0].from = NULL; 236 token = match_token(p, zpl_tokens, args); 237 error = zfsvfs_parse_option(p, token, args, tmp_vfsp); 238 if (error) { 239 kmem_strfree(tmp_mntopts); 240 zfsvfs_vfs_free(tmp_vfsp); 241 return (error); 242 } 243 } 244 245 kmem_strfree(tmp_mntopts); 246 } 247 248 *vfsp = tmp_vfsp; 249 250 return (0); 251} 252 253boolean_t 254zfs_is_readonly(zfsvfs_t *zfsvfs) 255{ 256 return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY)); 257} 258 259/*ARGSUSED*/ 260int 261zfs_sync(struct super_block *sb, int wait, cred_t *cr) 262{ 263 zfsvfs_t *zfsvfs = sb->s_fs_info; 264 265 /* 266 * Semantically, the only requirement is that the sync be initiated. 267 * The DMU syncs out txgs frequently, so there's nothing to do. 268 */ 269 if (!wait) 270 return (0); 271 272 if (zfsvfs != NULL) { 273 /* 274 * Sync a specific filesystem. 275 */ 276 dsl_pool_t *dp; 277 278 ZFS_ENTER(zfsvfs); 279 dp = dmu_objset_pool(zfsvfs->z_os); 280 281 /* 282 * If the system is shutting down, then skip any 283 * filesystems which may exist on a suspended pool. 284 */ 285 if (spa_suspended(dp->dp_spa)) { 286 ZFS_EXIT(zfsvfs); 287 return (0); 288 } 289 290 if (zfsvfs->z_log != NULL) 291 zil_commit(zfsvfs->z_log, 0); 292 293 ZFS_EXIT(zfsvfs); 294 } else { 295 /* 296 * Sync all ZFS filesystems. This is what happens when you 297 * run sync(1). Unlike other filesystems, ZFS honors the 298 * request by waiting for all pools to commit all dirty data. 299 */ 300 spa_sync_allpools(); 301 } 302 303 return (0); 304} 305 306static void 307atime_changed_cb(void *arg, uint64_t newval) 308{ 309 zfsvfs_t *zfsvfs = arg; 310 struct super_block *sb = zfsvfs->z_sb; 311 312 if (sb == NULL) 313 return; 314 /* 315 * Update SB_NOATIME bit in VFS super block. Since atime update is 316 * determined by atime_needs_update(), atime_needs_update() needs to 317 * return false if atime is turned off, and not unconditionally return 318 * false if atime is turned on. 319 */ 320 if (newval) 321 sb->s_flags &= ~SB_NOATIME; 322 else 323 sb->s_flags |= SB_NOATIME; 324} 325 326static void 327relatime_changed_cb(void *arg, uint64_t newval) 328{ 329 ((zfsvfs_t *)arg)->z_relatime = newval; 330} 331 332static void 333xattr_changed_cb(void *arg, uint64_t newval) 334{ 335 zfsvfs_t *zfsvfs = arg; 336 337 if (newval == ZFS_XATTR_OFF) { 338 zfsvfs->z_flags &= ~ZSB_XATTR; 339 } else { 340 zfsvfs->z_flags |= ZSB_XATTR; 341 342 if (newval == ZFS_XATTR_SA) 343 zfsvfs->z_xattr_sa = B_TRUE; 344 else 345 zfsvfs->z_xattr_sa = B_FALSE; 346 } 347} 348 349static void 350acltype_changed_cb(void *arg, uint64_t newval) 351{ 352 zfsvfs_t *zfsvfs = arg; 353 354 switch (newval) { 355 case ZFS_ACLTYPE_NFSV4: 356 case ZFS_ACLTYPE_OFF: 357 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF; 358 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL; 359 break; 360 case ZFS_ACLTYPE_POSIX: 361#ifdef CONFIG_FS_POSIX_ACL 362 zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIX; 363 zfsvfs->z_sb->s_flags |= SB_POSIXACL; 364#else 365 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF; 366 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL; 367#endif /* CONFIG_FS_POSIX_ACL */ 368 break; 369 default: 370 break; 371 } 372} 373 374static void 375blksz_changed_cb(void *arg, uint64_t newval) 376{ 377 zfsvfs_t *zfsvfs = arg; 378 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os))); 379 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); 380 ASSERT(ISP2(newval)); 381 382 zfsvfs->z_max_blksz = newval; 383} 384 385static void 386readonly_changed_cb(void *arg, uint64_t newval) 387{ 388 zfsvfs_t *zfsvfs = arg; 389 struct super_block *sb = zfsvfs->z_sb; 390 391 if (sb == NULL) 392 return; 393 394 if (newval) 395 sb->s_flags |= SB_RDONLY; 396 else 397 sb->s_flags &= ~SB_RDONLY; 398} 399 400static void 401devices_changed_cb(void *arg, uint64_t newval) 402{ 403} 404 405static void 406setuid_changed_cb(void *arg, uint64_t newval) 407{ 408} 409 410static void 411exec_changed_cb(void *arg, uint64_t newval) 412{ 413} 414 415static void 416nbmand_changed_cb(void *arg, uint64_t newval) 417{ 418 zfsvfs_t *zfsvfs = arg; 419 struct super_block *sb = zfsvfs->z_sb; 420 421 if (sb == NULL) 422 return; 423 424 if (newval == TRUE) 425 sb->s_flags |= SB_MANDLOCK; 426 else 427 sb->s_flags &= ~SB_MANDLOCK; 428} 429 430static void 431snapdir_changed_cb(void *arg, uint64_t newval) 432{ 433 ((zfsvfs_t *)arg)->z_show_ctldir = newval; 434} 435 436static void 437vscan_changed_cb(void *arg, uint64_t newval) 438{ 439 ((zfsvfs_t *)arg)->z_vscan = newval; 440} 441 442static void 443acl_mode_changed_cb(void *arg, uint64_t newval) 444{ 445 zfsvfs_t *zfsvfs = arg; 446 447 zfsvfs->z_acl_mode = newval; 448} 449 450static void 451acl_inherit_changed_cb(void *arg, uint64_t newval) 452{ 453 ((zfsvfs_t *)arg)->z_acl_inherit = newval; 454} 455 456static int 457zfs_register_callbacks(vfs_t *vfsp) 458{ 459 struct dsl_dataset *ds = NULL; 460 objset_t *os = NULL; 461 zfsvfs_t *zfsvfs = NULL; 462 int error = 0; 463 464 ASSERT(vfsp); 465 zfsvfs = vfsp->vfs_data; 466 ASSERT(zfsvfs); 467 os = zfsvfs->z_os; 468 469 /* 470 * The act of registering our callbacks will destroy any mount 471 * options we may have. In order to enable temporary overrides 472 * of mount options, we stash away the current values and 473 * restore them after we register the callbacks. 474 */ 475 if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) { 476 vfsp->vfs_do_readonly = B_TRUE; 477 vfsp->vfs_readonly = B_TRUE; 478 } 479 480 /* 481 * Register property callbacks. 482 * 483 * It would probably be fine to just check for i/o error from 484 * the first prop_register(), but I guess I like to go 485 * overboard... 486 */ 487 ds = dmu_objset_ds(os); 488 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 489 error = dsl_prop_register(ds, 490 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 491 error = error ? error : dsl_prop_register(ds, 492 zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs); 493 error = error ? error : dsl_prop_register(ds, 494 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 495 error = error ? error : dsl_prop_register(ds, 496 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 497 error = error ? error : dsl_prop_register(ds, 498 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 499 error = error ? error : dsl_prop_register(ds, 500 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 501 error = error ? error : dsl_prop_register(ds, 502 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 503 error = error ? error : dsl_prop_register(ds, 504 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 505 error = error ? error : dsl_prop_register(ds, 506 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 507 error = error ? error : dsl_prop_register(ds, 508 zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs); 509 error = error ? error : dsl_prop_register(ds, 510 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 511 error = error ? error : dsl_prop_register(ds, 512 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 513 zfsvfs); 514 error = error ? error : dsl_prop_register(ds, 515 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs); 516 error = error ? error : dsl_prop_register(ds, 517 zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs); 518 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 519 if (error) 520 goto unregister; 521 522 /* 523 * Invoke our callbacks to restore temporary mount options. 524 */ 525 if (vfsp->vfs_do_readonly) 526 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly); 527 if (vfsp->vfs_do_setuid) 528 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid); 529 if (vfsp->vfs_do_exec) 530 exec_changed_cb(zfsvfs, vfsp->vfs_exec); 531 if (vfsp->vfs_do_devices) 532 devices_changed_cb(zfsvfs, vfsp->vfs_devices); 533 if (vfsp->vfs_do_xattr) 534 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr); 535 if (vfsp->vfs_do_atime) 536 atime_changed_cb(zfsvfs, vfsp->vfs_atime); 537 if (vfsp->vfs_do_relatime) 538 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime); 539 if (vfsp->vfs_do_nbmand) 540 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand); 541 542 return (0); 543 544unregister: 545 dsl_prop_unregister_all(ds, zfsvfs); 546 return (error); 547} 548 549/* 550 * Takes a dataset, a property, a value and that value's setpoint as 551 * found in the ZAP. Checks if the property has been changed in the vfs. 552 * If so, val and setpoint will be overwritten with updated content. 553 * Otherwise, they are left unchanged. 554 */ 555int 556zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val, 557 char *setpoint) 558{ 559 int error; 560 zfsvfs_t *zfvp; 561 vfs_t *vfsp; 562 objset_t *os; 563 uint64_t tmp = *val; 564 565 error = dmu_objset_from_ds(ds, &os); 566 if (error != 0) 567 return (error); 568 569 if (dmu_objset_type(os) != DMU_OST_ZFS) 570 return (EINVAL); 571 572 mutex_enter(&os->os_user_ptr_lock); 573 zfvp = dmu_objset_get_user(os); 574 mutex_exit(&os->os_user_ptr_lock); 575 if (zfvp == NULL) 576 return (ESRCH); 577 578 vfsp = zfvp->z_vfs; 579 580 switch (zfs_prop) { 581 case ZFS_PROP_ATIME: 582 if (vfsp->vfs_do_atime) 583 tmp = vfsp->vfs_atime; 584 break; 585 case ZFS_PROP_RELATIME: 586 if (vfsp->vfs_do_relatime) 587 tmp = vfsp->vfs_relatime; 588 break; 589 case ZFS_PROP_DEVICES: 590 if (vfsp->vfs_do_devices) 591 tmp = vfsp->vfs_devices; 592 break; 593 case ZFS_PROP_EXEC: 594 if (vfsp->vfs_do_exec) 595 tmp = vfsp->vfs_exec; 596 break; 597 case ZFS_PROP_SETUID: 598 if (vfsp->vfs_do_setuid) 599 tmp = vfsp->vfs_setuid; 600 break; 601 case ZFS_PROP_READONLY: 602 if (vfsp->vfs_do_readonly) 603 tmp = vfsp->vfs_readonly; 604 break; 605 case ZFS_PROP_XATTR: 606 if (vfsp->vfs_do_xattr) 607 tmp = vfsp->vfs_xattr; 608 break; 609 case ZFS_PROP_NBMAND: 610 if (vfsp->vfs_do_nbmand) 611 tmp = vfsp->vfs_nbmand; 612 break; 613 default: 614 return (ENOENT); 615 } 616 617 if (tmp != *val) { 618 (void) strcpy(setpoint, "temporary"); 619 *val = tmp; 620 } 621 return (0); 622} 623 624/* 625 * Associate this zfsvfs with the given objset, which must be owned. 626 * This will cache a bunch of on-disk state from the objset in the 627 * zfsvfs. 628 */ 629static int 630zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os) 631{ 632 int error; 633 uint64_t val; 634 635 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 636 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 637 zfsvfs->z_os = os; 638 639 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 640 if (error != 0) 641 return (error); 642 if (zfsvfs->z_version > 643 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 644 (void) printk("Can't mount a version %lld file system " 645 "on a version %lld pool\n. Pool must be upgraded to mount " 646 "this file system.\n", (u_longlong_t)zfsvfs->z_version, 647 (u_longlong_t)spa_version(dmu_objset_spa(os))); 648 return (SET_ERROR(ENOTSUP)); 649 } 650 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val); 651 if (error != 0) 652 return (error); 653 zfsvfs->z_norm = (int)val; 654 655 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val); 656 if (error != 0) 657 return (error); 658 zfsvfs->z_utf8 = (val != 0); 659 660 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val); 661 if (error != 0) 662 return (error); 663 zfsvfs->z_case = (uint_t)val; 664 665 if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0) 666 return (error); 667 zfsvfs->z_acl_type = (uint_t)val; 668 669 /* 670 * Fold case on file systems that are always or sometimes case 671 * insensitive. 672 */ 673 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 674 zfsvfs->z_case == ZFS_CASE_MIXED) 675 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 676 677 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 678 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 679 680 uint64_t sa_obj = 0; 681 if (zfsvfs->z_use_sa) { 682 /* should either have both of these objects or none */ 683 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 684 &sa_obj); 685 if (error != 0) 686 return (error); 687 688 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val); 689 if ((error == 0) && (val == ZFS_XATTR_SA)) 690 zfsvfs->z_xattr_sa = B_TRUE; 691 } 692 693 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 694 &zfsvfs->z_root); 695 if (error != 0) 696 return (error); 697 ASSERT(zfsvfs->z_root != 0); 698 699 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 700 &zfsvfs->z_unlinkedobj); 701 if (error != 0) 702 return (error); 703 704 error = zap_lookup(os, MASTER_NODE_OBJ, 705 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 706 8, 1, &zfsvfs->z_userquota_obj); 707 if (error == ENOENT) 708 zfsvfs->z_userquota_obj = 0; 709 else if (error != 0) 710 return (error); 711 712 error = zap_lookup(os, MASTER_NODE_OBJ, 713 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 714 8, 1, &zfsvfs->z_groupquota_obj); 715 if (error == ENOENT) 716 zfsvfs->z_groupquota_obj = 0; 717 else if (error != 0) 718 return (error); 719 720 error = zap_lookup(os, MASTER_NODE_OBJ, 721 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA], 722 8, 1, &zfsvfs->z_projectquota_obj); 723 if (error == ENOENT) 724 zfsvfs->z_projectquota_obj = 0; 725 else if (error != 0) 726 return (error); 727 728 error = zap_lookup(os, MASTER_NODE_OBJ, 729 zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA], 730 8, 1, &zfsvfs->z_userobjquota_obj); 731 if (error == ENOENT) 732 zfsvfs->z_userobjquota_obj = 0; 733 else if (error != 0) 734 return (error); 735 736 error = zap_lookup(os, MASTER_NODE_OBJ, 737 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA], 738 8, 1, &zfsvfs->z_groupobjquota_obj); 739 if (error == ENOENT) 740 zfsvfs->z_groupobjquota_obj = 0; 741 else if (error != 0) 742 return (error); 743 744 error = zap_lookup(os, MASTER_NODE_OBJ, 745 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA], 746 8, 1, &zfsvfs->z_projectobjquota_obj); 747 if (error == ENOENT) 748 zfsvfs->z_projectobjquota_obj = 0; 749 else if (error != 0) 750 return (error); 751 752 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 753 &zfsvfs->z_fuid_obj); 754 if (error == ENOENT) 755 zfsvfs->z_fuid_obj = 0; 756 else if (error != 0) 757 return (error); 758 759 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 760 &zfsvfs->z_shares_dir); 761 if (error == ENOENT) 762 zfsvfs->z_shares_dir = 0; 763 else if (error != 0) 764 return (error); 765 766 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 767 &zfsvfs->z_attr_table); 768 if (error != 0) 769 return (error); 770 771 if (zfsvfs->z_version >= ZPL_VERSION_SA) 772 sa_register_update_callback(os, zfs_sa_upgrade); 773 774 return (0); 775} 776 777int 778zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp) 779{ 780 objset_t *os; 781 zfsvfs_t *zfsvfs; 782 int error; 783 boolean_t ro = (readonly || (strchr(osname, '@') != NULL)); 784 785 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 786 787 error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os); 788 if (error != 0) { 789 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 790 return (error); 791 } 792 793 error = zfsvfs_create_impl(zfvp, zfsvfs, os); 794 if (error != 0) { 795 dmu_objset_disown(os, B_TRUE, zfsvfs); 796 } 797 return (error); 798} 799 800 801/* 802 * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function 803 * on a failure. Do not pass in a statically allocated zfsvfs. 804 */ 805int 806zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os) 807{ 808 int error; 809 810 zfsvfs->z_vfs = NULL; 811 zfsvfs->z_sb = NULL; 812 zfsvfs->z_parent = zfsvfs; 813 814 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 815 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 816 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 817 offsetof(znode_t, z_link_node)); 818 ZFS_TEARDOWN_INIT(zfsvfs); 819 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 820 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 821 822 int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1), 823 ZFS_OBJ_MTX_MAX); 824 zfsvfs->z_hold_size = size; 825 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size, 826 KM_SLEEP); 827 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP); 828 for (int i = 0; i != size; i++) { 829 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare, 830 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node)); 831 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL); 832 } 833 834 error = zfsvfs_init(zfsvfs, os); 835 if (error != 0) { 836 *zfvp = NULL; 837 zfsvfs_free(zfsvfs); 838 return (error); 839 } 840 841 zfsvfs->z_drain_task = TASKQID_INVALID; 842 zfsvfs->z_draining = B_FALSE; 843 zfsvfs->z_drain_cancel = B_TRUE; 844 845 *zfvp = zfsvfs; 846 return (0); 847} 848 849static int 850zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 851{ 852 int error; 853 boolean_t readonly = zfs_is_readonly(zfsvfs); 854 855 error = zfs_register_callbacks(zfsvfs->z_vfs); 856 if (error) 857 return (error); 858 859 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data); 860 861 /* 862 * If we are not mounting (ie: online recv), then we don't 863 * have to worry about replaying the log as we blocked all 864 * operations out since we closed the ZIL. 865 */ 866 if (mounting) { 867 ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL); 868 dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os); 869 870 /* 871 * During replay we remove the read only flag to 872 * allow replays to succeed. 873 */ 874 if (readonly != 0) { 875 readonly_changed_cb(zfsvfs, B_FALSE); 876 } else { 877 zap_stats_t zs; 878 if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj, 879 &zs) == 0) { 880 dataset_kstats_update_nunlinks_kstat( 881 &zfsvfs->z_kstat, zs.zs_num_entries); 882 dprintf_ds(zfsvfs->z_os->os_dsl_dataset, 883 "num_entries in unlinked set: %llu", 884 zs.zs_num_entries); 885 } 886 zfs_unlinked_drain(zfsvfs); 887 dsl_dir_t *dd = zfsvfs->z_os->os_dsl_dataset->ds_dir; 888 dd->dd_activity_cancelled = B_FALSE; 889 } 890 891 /* 892 * Parse and replay the intent log. 893 * 894 * Because of ziltest, this must be done after 895 * zfs_unlinked_drain(). (Further note: ziltest 896 * doesn't use readonly mounts, where 897 * zfs_unlinked_drain() isn't called.) This is because 898 * ziltest causes spa_sync() to think it's committed, 899 * but actually it is not, so the intent log contains 900 * many txg's worth of changes. 901 * 902 * In particular, if object N is in the unlinked set in 903 * the last txg to actually sync, then it could be 904 * actually freed in a later txg and then reallocated 905 * in a yet later txg. This would write a "create 906 * object N" record to the intent log. Normally, this 907 * would be fine because the spa_sync() would have 908 * written out the fact that object N is free, before 909 * we could write the "create object N" intent log 910 * record. 911 * 912 * But when we are in ziltest mode, we advance the "open 913 * txg" without actually spa_sync()-ing the changes to 914 * disk. So we would see that object N is still 915 * allocated and in the unlinked set, and there is an 916 * intent log record saying to allocate it. 917 */ 918 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 919 if (zil_replay_disable) { 920 zil_destroy(zfsvfs->z_log, B_FALSE); 921 } else { 922 zfsvfs->z_replay = B_TRUE; 923 zil_replay(zfsvfs->z_os, zfsvfs, 924 zfs_replay_vector); 925 zfsvfs->z_replay = B_FALSE; 926 } 927 } 928 929 /* restore readonly bit */ 930 if (readonly != 0) 931 readonly_changed_cb(zfsvfs, B_TRUE); 932 } 933 934 /* 935 * Set the objset user_ptr to track its zfsvfs. 936 */ 937 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 938 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 939 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 940 941 return (0); 942} 943 944void 945zfsvfs_free(zfsvfs_t *zfsvfs) 946{ 947 int i, size = zfsvfs->z_hold_size; 948 949 zfs_fuid_destroy(zfsvfs); 950 951 mutex_destroy(&zfsvfs->z_znodes_lock); 952 mutex_destroy(&zfsvfs->z_lock); 953 list_destroy(&zfsvfs->z_all_znodes); 954 ZFS_TEARDOWN_DESTROY(zfsvfs); 955 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 956 rw_destroy(&zfsvfs->z_fuid_lock); 957 for (i = 0; i != size; i++) { 958 avl_destroy(&zfsvfs->z_hold_trees[i]); 959 mutex_destroy(&zfsvfs->z_hold_locks[i]); 960 } 961 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size); 962 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size); 963 zfsvfs_vfs_free(zfsvfs->z_vfs); 964 dataset_kstats_destroy(&zfsvfs->z_kstat); 965 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 966} 967 968static void 969zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 970{ 971 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 972 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 973} 974 975static void 976zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 977{ 978 objset_t *os = zfsvfs->z_os; 979 980 if (!dmu_objset_is_snapshot(os)) 981 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 982} 983 984#ifdef HAVE_MLSLABEL 985/* 986 * Check that the hex label string is appropriate for the dataset being 987 * mounted into the global_zone proper. 988 * 989 * Return an error if the hex label string is not default or 990 * admin_low/admin_high. For admin_low labels, the corresponding 991 * dataset must be readonly. 992 */ 993int 994zfs_check_global_label(const char *dsname, const char *hexsl) 995{ 996 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 997 return (0); 998 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 999 return (0); 1000 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1001 /* must be readonly */ 1002 uint64_t rdonly; 1003 1004 if (dsl_prop_get_integer(dsname, 1005 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1006 return (SET_ERROR(EACCES)); 1007 return (rdonly ? 0 : SET_ERROR(EACCES)); 1008 } 1009 return (SET_ERROR(EACCES)); 1010} 1011#endif /* HAVE_MLSLABEL */ 1012 1013static int 1014zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp, 1015 uint32_t bshift) 1016{ 1017 char buf[20 + DMU_OBJACCT_PREFIX_LEN]; 1018 uint64_t offset = DMU_OBJACCT_PREFIX_LEN; 1019 uint64_t quota; 1020 uint64_t used; 1021 int err; 1022 1023 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1); 1024 err = zfs_id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset, 1025 sizeof (buf) - offset, B_FALSE); 1026 if (err) 1027 return (err); 1028 1029 if (zfsvfs->z_projectquota_obj == 0) 1030 goto objs; 1031 1032 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj, 1033 buf + offset, 8, 1, "a); 1034 if (err == ENOENT) 1035 goto objs; 1036 else if (err) 1037 return (err); 1038 1039 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT, 1040 buf + offset, 8, 1, &used); 1041 if (unlikely(err == ENOENT)) { 1042 uint32_t blksize; 1043 u_longlong_t nblocks; 1044 1045 /* 1046 * Quota accounting is async, so it is possible race case. 1047 * There is at least one object with the given project ID. 1048 */ 1049 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 1050 if (unlikely(zp->z_blksz == 0)) 1051 blksize = zfsvfs->z_max_blksz; 1052 1053 used = blksize * nblocks; 1054 } else if (err) { 1055 return (err); 1056 } 1057 1058 statp->f_blocks = quota >> bshift; 1059 statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0; 1060 statp->f_bavail = statp->f_bfree; 1061 1062objs: 1063 if (zfsvfs->z_projectobjquota_obj == 0) 1064 return (0); 1065 1066 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj, 1067 buf + offset, 8, 1, "a); 1068 if (err == ENOENT) 1069 return (0); 1070 else if (err) 1071 return (err); 1072 1073 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT, 1074 buf, 8, 1, &used); 1075 if (unlikely(err == ENOENT)) { 1076 /* 1077 * Quota accounting is async, so it is possible race case. 1078 * There is at least one object with the given project ID. 1079 */ 1080 used = 1; 1081 } else if (err) { 1082 return (err); 1083 } 1084 1085 statp->f_files = quota; 1086 statp->f_ffree = (quota > used) ? (quota - used) : 0; 1087 1088 return (0); 1089} 1090 1091int 1092zfs_statvfs(struct inode *ip, struct kstatfs *statp) 1093{ 1094 zfsvfs_t *zfsvfs = ITOZSB(ip); 1095 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1096 int err = 0; 1097 1098 ZFS_ENTER(zfsvfs); 1099 1100 dmu_objset_space(zfsvfs->z_os, 1101 &refdbytes, &availbytes, &usedobjs, &availobjs); 1102 1103 uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os); 1104 /* 1105 * The underlying storage pool actually uses multiple block 1106 * size. Under Solaris frsize (fragment size) is reported as 1107 * the smallest block size we support, and bsize (block size) 1108 * as the filesystem's maximum block size. Unfortunately, 1109 * under Linux the fragment size and block size are often used 1110 * interchangeably. Thus we are forced to report both of them 1111 * as the filesystem's maximum block size. 1112 */ 1113 statp->f_frsize = zfsvfs->z_max_blksz; 1114 statp->f_bsize = zfsvfs->z_max_blksz; 1115 uint32_t bshift = fls(statp->f_bsize) - 1; 1116 1117 /* 1118 * The following report "total" blocks of various kinds in 1119 * the file system, but reported in terms of f_bsize - the 1120 * "preferred" size. 1121 */ 1122 1123 /* Round up so we never have a filesystem using 0 blocks. */ 1124 refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize); 1125 statp->f_blocks = (refdbytes + availbytes) >> bshift; 1126 statp->f_bfree = availbytes >> bshift; 1127 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1128 1129 /* 1130 * statvfs() should really be called statufs(), because it assumes 1131 * static metadata. ZFS doesn't preallocate files, so the best 1132 * we can do is report the max that could possibly fit in f_files, 1133 * and that minus the number actually used in f_ffree. 1134 * For f_ffree, report the smaller of the number of objects available 1135 * and the number of blocks (each object will take at least a block). 1136 */ 1137 statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT); 1138 statp->f_files = statp->f_ffree + usedobjs; 1139 statp->f_fsid.val[0] = (uint32_t)fsid; 1140 statp->f_fsid.val[1] = (uint32_t)(fsid >> 32); 1141 statp->f_type = ZFS_SUPER_MAGIC; 1142 statp->f_namelen = MAXNAMELEN - 1; 1143 1144 /* 1145 * We have all of 40 characters to stuff a string here. 1146 * Is there anything useful we could/should provide? 1147 */ 1148 bzero(statp->f_spare, sizeof (statp->f_spare)); 1149 1150 if (dmu_objset_projectquota_enabled(zfsvfs->z_os) && 1151 dmu_objset_projectquota_present(zfsvfs->z_os)) { 1152 znode_t *zp = ITOZ(ip); 1153 1154 if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid && 1155 zpl_is_valid_projid(zp->z_projid)) 1156 err = zfs_statfs_project(zfsvfs, zp, statp, bshift); 1157 } 1158 1159 ZFS_EXIT(zfsvfs); 1160 return (err); 1161} 1162 1163static int 1164zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp) 1165{ 1166 znode_t *rootzp; 1167 int error; 1168 1169 ZFS_ENTER(zfsvfs); 1170 1171 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1172 if (error == 0) 1173 *ipp = ZTOI(rootzp); 1174 1175 ZFS_EXIT(zfsvfs); 1176 return (error); 1177} 1178 1179/* 1180 * Linux kernels older than 3.1 do not support a per-filesystem shrinker. 1181 * To accommodate this we must improvise and manually walk the list of znodes 1182 * attempting to prune dentries in order to be able to drop the inodes. 1183 * 1184 * To avoid scanning the same znodes multiple times they are always rotated 1185 * to the end of the z_all_znodes list. New znodes are inserted at the 1186 * end of the list so we're always scanning the oldest znodes first. 1187 */ 1188static int 1189zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan) 1190{ 1191 znode_t **zp_array, *zp; 1192 int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *)); 1193 int objects = 0; 1194 int i = 0, j = 0; 1195 1196 zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP); 1197 1198 mutex_enter(&zfsvfs->z_znodes_lock); 1199 while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) { 1200 1201 if ((i++ > nr_to_scan) || (j >= max_array)) 1202 break; 1203 1204 ASSERT(list_link_active(&zp->z_link_node)); 1205 list_remove(&zfsvfs->z_all_znodes, zp); 1206 list_insert_tail(&zfsvfs->z_all_znodes, zp); 1207 1208 /* Skip active znodes and .zfs entries */ 1209 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir) 1210 continue; 1211 1212 if (igrab(ZTOI(zp)) == NULL) 1213 continue; 1214 1215 zp_array[j] = zp; 1216 j++; 1217 } 1218 mutex_exit(&zfsvfs->z_znodes_lock); 1219 1220 for (i = 0; i < j; i++) { 1221 zp = zp_array[i]; 1222 1223 ASSERT3P(zp, !=, NULL); 1224 d_prune_aliases(ZTOI(zp)); 1225 1226 if (atomic_read(&ZTOI(zp)->i_count) == 1) 1227 objects++; 1228 1229 zrele(zp); 1230 } 1231 1232 kmem_free(zp_array, max_array * sizeof (znode_t *)); 1233 1234 return (objects); 1235} 1236 1237/* 1238 * The ARC has requested that the filesystem drop entries from the dentry 1239 * and inode caches. This can occur when the ARC needs to free meta data 1240 * blocks but can't because they are all pinned by entries in these caches. 1241 */ 1242int 1243zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects) 1244{ 1245 zfsvfs_t *zfsvfs = sb->s_fs_info; 1246 int error = 0; 1247 struct shrinker *shrinker = &sb->s_shrink; 1248 struct shrink_control sc = { 1249 .nr_to_scan = nr_to_scan, 1250 .gfp_mask = GFP_KERNEL, 1251 }; 1252 1253 ZFS_ENTER(zfsvfs); 1254 1255#if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \ 1256 defined(SHRINK_CONTROL_HAS_NID) && \ 1257 defined(SHRINKER_NUMA_AWARE) 1258 if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) { 1259 *objects = 0; 1260 for_each_online_node(sc.nid) { 1261 *objects += (*shrinker->scan_objects)(shrinker, &sc); 1262 } 1263 } else { 1264 *objects = (*shrinker->scan_objects)(shrinker, &sc); 1265 } 1266 1267#elif defined(HAVE_SPLIT_SHRINKER_CALLBACK) 1268 *objects = (*shrinker->scan_objects)(shrinker, &sc); 1269#elif defined(HAVE_SINGLE_SHRINKER_CALLBACK) 1270 *objects = (*shrinker->shrink)(shrinker, &sc); 1271#elif defined(HAVE_D_PRUNE_ALIASES) 1272#define D_PRUNE_ALIASES_IS_DEFAULT 1273 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan); 1274#else 1275#error "No available dentry and inode cache pruning mechanism." 1276#endif 1277 1278#if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT) 1279#undef D_PRUNE_ALIASES_IS_DEFAULT 1280 /* 1281 * Fall back to zfs_prune_aliases if the kernel's per-superblock 1282 * shrinker couldn't free anything, possibly due to the inodes being 1283 * allocated in a different memcg. 1284 */ 1285 if (*objects == 0) 1286 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan); 1287#endif 1288 1289 ZFS_EXIT(zfsvfs); 1290 1291 dprintf_ds(zfsvfs->z_os->os_dsl_dataset, 1292 "pruning, nr_to_scan=%lu objects=%d error=%d\n", 1293 nr_to_scan, *objects, error); 1294 1295 return (error); 1296} 1297 1298/* 1299 * Teardown the zfsvfs_t. 1300 * 1301 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock' 1302 * and 'z_teardown_inactive_lock' held. 1303 */ 1304static int 1305zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1306{ 1307 znode_t *zp; 1308 1309 zfs_unlinked_drain_stop_wait(zfsvfs); 1310 1311 /* 1312 * If someone has not already unmounted this file system, 1313 * drain the zrele_taskq to ensure all active references to the 1314 * zfsvfs_t have been handled only then can it be safely destroyed. 1315 */ 1316 if (zfsvfs->z_os) { 1317 /* 1318 * If we're unmounting we have to wait for the list to 1319 * drain completely. 1320 * 1321 * If we're not unmounting there's no guarantee the list 1322 * will drain completely, but iputs run from the taskq 1323 * may add the parents of dir-based xattrs to the taskq 1324 * so we want to wait for these. 1325 * 1326 * We can safely read z_nr_znodes without locking because the 1327 * VFS has already blocked operations which add to the 1328 * z_all_znodes list and thus increment z_nr_znodes. 1329 */ 1330 int round = 0; 1331 while (zfsvfs->z_nr_znodes > 0) { 1332 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1333 dmu_objset_pool(zfsvfs->z_os)), 0); 1334 if (++round > 1 && !unmounting) 1335 break; 1336 } 1337 } 1338 1339 ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG); 1340 1341 if (!unmounting) { 1342 /* 1343 * We purge the parent filesystem's super block as the 1344 * parent filesystem and all of its snapshots have their 1345 * inode's super block set to the parent's filesystem's 1346 * super block. Note, 'z_parent' is self referential 1347 * for non-snapshots. 1348 */ 1349 shrink_dcache_sb(zfsvfs->z_parent->z_sb); 1350 } 1351 1352 /* 1353 * Close the zil. NB: Can't close the zil while zfs_inactive 1354 * threads are blocked as zil_close can call zfs_inactive. 1355 */ 1356 if (zfsvfs->z_log) { 1357 zil_close(zfsvfs->z_log); 1358 zfsvfs->z_log = NULL; 1359 } 1360 1361 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1362 1363 /* 1364 * If we are not unmounting (ie: online recv) and someone already 1365 * unmounted this file system while we were doing the switcheroo, 1366 * or a reopen of z_os failed then just bail out now. 1367 */ 1368 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1369 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1370 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1371 return (SET_ERROR(EIO)); 1372 } 1373 1374 /* 1375 * At this point there are no VFS ops active, and any new VFS ops 1376 * will fail with EIO since we have z_teardown_lock for writer (only 1377 * relevant for forced unmount). 1378 * 1379 * Release all holds on dbufs. We also grab an extra reference to all 1380 * the remaining inodes so that the kernel does not attempt to free 1381 * any inodes of a suspended fs. This can cause deadlocks since the 1382 * zfs_resume_fs() process may involve starting threads, which might 1383 * attempt to free unreferenced inodes to free up memory for the new 1384 * thread. 1385 */ 1386 if (!unmounting) { 1387 mutex_enter(&zfsvfs->z_znodes_lock); 1388 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1389 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1390 if (zp->z_sa_hdl) 1391 zfs_znode_dmu_fini(zp); 1392 if (igrab(ZTOI(zp)) != NULL) 1393 zp->z_suspended = B_TRUE; 1394 1395 } 1396 mutex_exit(&zfsvfs->z_znodes_lock); 1397 } 1398 1399 /* 1400 * If we are unmounting, set the unmounted flag and let new VFS ops 1401 * unblock. zfs_inactive will have the unmounted behavior, and all 1402 * other VFS ops will fail with EIO. 1403 */ 1404 if (unmounting) { 1405 zfsvfs->z_unmounted = B_TRUE; 1406 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1407 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1408 } 1409 1410 /* 1411 * z_os will be NULL if there was an error in attempting to reopen 1412 * zfsvfs, so just return as the properties had already been 1413 * 1414 * unregistered and cached data had been evicted before. 1415 */ 1416 if (zfsvfs->z_os == NULL) 1417 return (0); 1418 1419 /* 1420 * Unregister properties. 1421 */ 1422 zfs_unregister_callbacks(zfsvfs); 1423 1424 /* 1425 * Evict cached data. We must write out any dirty data before 1426 * disowning the dataset. 1427 */ 1428 objset_t *os = zfsvfs->z_os; 1429 boolean_t os_dirty = B_FALSE; 1430 for (int t = 0; t < TXG_SIZE; t++) { 1431 if (dmu_objset_is_dirty(os, t)) { 1432 os_dirty = B_TRUE; 1433 break; 1434 } 1435 } 1436 if (!zfs_is_readonly(zfsvfs) && os_dirty) { 1437 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1438 } 1439 dmu_objset_evict_dbufs(zfsvfs->z_os); 1440 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 1441 dsl_dir_cancel_waiters(dd); 1442 1443 return (0); 1444} 1445 1446#if defined(HAVE_SUPER_SETUP_BDI_NAME) 1447atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0); 1448#endif 1449 1450int 1451zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent) 1452{ 1453 const char *osname = zm->mnt_osname; 1454 struct inode *root_inode = NULL; 1455 uint64_t recordsize; 1456 int error = 0; 1457 zfsvfs_t *zfsvfs = NULL; 1458 vfs_t *vfs = NULL; 1459 1460 ASSERT(zm); 1461 ASSERT(osname); 1462 1463 error = zfsvfs_parse_options(zm->mnt_data, &vfs); 1464 if (error) 1465 return (error); 1466 1467 error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs); 1468 if (error) { 1469 zfsvfs_vfs_free(vfs); 1470 goto out; 1471 } 1472 1473 if ((error = dsl_prop_get_integer(osname, "recordsize", 1474 &recordsize, NULL))) { 1475 zfsvfs_vfs_free(vfs); 1476 goto out; 1477 } 1478 1479 vfs->vfs_data = zfsvfs; 1480 zfsvfs->z_vfs = vfs; 1481 zfsvfs->z_sb = sb; 1482 sb->s_fs_info = zfsvfs; 1483 sb->s_magic = ZFS_SUPER_MAGIC; 1484 sb->s_maxbytes = MAX_LFS_FILESIZE; 1485 sb->s_time_gran = 1; 1486 sb->s_blocksize = recordsize; 1487 sb->s_blocksize_bits = ilog2(recordsize); 1488 1489 error = -zpl_bdi_setup(sb, "zfs"); 1490 if (error) 1491 goto out; 1492 1493 sb->s_bdi->ra_pages = 0; 1494 1495 /* Set callback operations for the file system. */ 1496 sb->s_op = &zpl_super_operations; 1497 sb->s_xattr = zpl_xattr_handlers; 1498 sb->s_export_op = &zpl_export_operations; 1499 sb->s_d_op = &zpl_dentry_operations; 1500 1501 /* Set features for file system. */ 1502 zfs_set_fuid_feature(zfsvfs); 1503 1504 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1505 uint64_t pval; 1506 1507 atime_changed_cb(zfsvfs, B_FALSE); 1508 readonly_changed_cb(zfsvfs, B_TRUE); 1509 if ((error = dsl_prop_get_integer(osname, 1510 "xattr", &pval, NULL))) 1511 goto out; 1512 xattr_changed_cb(zfsvfs, pval); 1513 if ((error = dsl_prop_get_integer(osname, 1514 "acltype", &pval, NULL))) 1515 goto out; 1516 acltype_changed_cb(zfsvfs, pval); 1517 zfsvfs->z_issnap = B_TRUE; 1518 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1519 zfsvfs->z_snap_defer_time = jiffies; 1520 1521 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1522 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1523 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1524 } else { 1525 if ((error = zfsvfs_setup(zfsvfs, B_TRUE))) 1526 goto out; 1527 } 1528 1529 /* Allocate a root inode for the filesystem. */ 1530 error = zfs_root(zfsvfs, &root_inode); 1531 if (error) { 1532 (void) zfs_umount(sb); 1533 goto out; 1534 } 1535 1536 /* Allocate a root dentry for the filesystem */ 1537 sb->s_root = d_make_root(root_inode); 1538 if (sb->s_root == NULL) { 1539 (void) zfs_umount(sb); 1540 error = SET_ERROR(ENOMEM); 1541 goto out; 1542 } 1543 1544 if (!zfsvfs->z_issnap) 1545 zfsctl_create(zfsvfs); 1546 1547 zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb); 1548out: 1549 if (error) { 1550 if (zfsvfs != NULL) { 1551 dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs); 1552 zfsvfs_free(zfsvfs); 1553 } 1554 /* 1555 * make sure we don't have dangling sb->s_fs_info which 1556 * zfs_preumount will use. 1557 */ 1558 sb->s_fs_info = NULL; 1559 } 1560 1561 return (error); 1562} 1563 1564/* 1565 * Called when an unmount is requested and certain sanity checks have 1566 * already passed. At this point no dentries or inodes have been reclaimed 1567 * from their respective caches. We drop the extra reference on the .zfs 1568 * control directory to allow everything to be reclaimed. All snapshots 1569 * must already have been unmounted to reach this point. 1570 */ 1571void 1572zfs_preumount(struct super_block *sb) 1573{ 1574 zfsvfs_t *zfsvfs = sb->s_fs_info; 1575 1576 /* zfsvfs is NULL when zfs_domount fails during mount */ 1577 if (zfsvfs) { 1578 zfs_unlinked_drain_stop_wait(zfsvfs); 1579 zfsctl_destroy(sb->s_fs_info); 1580 /* 1581 * Wait for zrele_async before entering evict_inodes in 1582 * generic_shutdown_super. The reason we must finish before 1583 * evict_inodes is when lazytime is on, or when zfs_purgedir 1584 * calls zfs_zget, zrele would bump i_count from 0 to 1. This 1585 * would race with the i_count check in evict_inodes. This means 1586 * it could destroy the inode while we are still using it. 1587 * 1588 * We wait for two passes. xattr directories in the first pass 1589 * may add xattr entries in zfs_purgedir, so in the second pass 1590 * we wait for them. We don't use taskq_wait here because it is 1591 * a pool wide taskq. Other mounted filesystems can constantly 1592 * do zrele_async and there's no guarantee when taskq will be 1593 * empty. 1594 */ 1595 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1596 dmu_objset_pool(zfsvfs->z_os)), 0); 1597 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1598 dmu_objset_pool(zfsvfs->z_os)), 0); 1599 } 1600} 1601 1602/* 1603 * Called once all other unmount released tear down has occurred. 1604 * It is our responsibility to release any remaining infrastructure. 1605 */ 1606/*ARGSUSED*/ 1607int 1608zfs_umount(struct super_block *sb) 1609{ 1610 zfsvfs_t *zfsvfs = sb->s_fs_info; 1611 objset_t *os; 1612 1613 if (zfsvfs->z_arc_prune != NULL) 1614 arc_remove_prune_callback(zfsvfs->z_arc_prune); 1615 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1616 os = zfsvfs->z_os; 1617 zpl_bdi_destroy(sb); 1618 1619 /* 1620 * z_os will be NULL if there was an error in 1621 * attempting to reopen zfsvfs. 1622 */ 1623 if (os != NULL) { 1624 /* 1625 * Unset the objset user_ptr. 1626 */ 1627 mutex_enter(&os->os_user_ptr_lock); 1628 dmu_objset_set_user(os, NULL); 1629 mutex_exit(&os->os_user_ptr_lock); 1630 1631 /* 1632 * Finally release the objset 1633 */ 1634 dmu_objset_disown(os, B_TRUE, zfsvfs); 1635 } 1636 1637 zfsvfs_free(zfsvfs); 1638 return (0); 1639} 1640 1641int 1642zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm) 1643{ 1644 zfsvfs_t *zfsvfs = sb->s_fs_info; 1645 vfs_t *vfsp; 1646 boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os); 1647 int error; 1648 1649 if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) && 1650 !(*flags & SB_RDONLY)) { 1651 *flags |= SB_RDONLY; 1652 return (EROFS); 1653 } 1654 1655 error = zfsvfs_parse_options(zm->mnt_data, &vfsp); 1656 if (error) 1657 return (error); 1658 1659 if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY)) 1660 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1661 1662 zfs_unregister_callbacks(zfsvfs); 1663 zfsvfs_vfs_free(zfsvfs->z_vfs); 1664 1665 vfsp->vfs_data = zfsvfs; 1666 zfsvfs->z_vfs = vfsp; 1667 if (!issnap) 1668 (void) zfs_register_callbacks(vfsp); 1669 1670 return (error); 1671} 1672 1673int 1674zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp) 1675{ 1676 zfsvfs_t *zfsvfs = sb->s_fs_info; 1677 znode_t *zp; 1678 uint64_t object = 0; 1679 uint64_t fid_gen = 0; 1680 uint64_t gen_mask; 1681 uint64_t zp_gen; 1682 int i, err; 1683 1684 *ipp = NULL; 1685 1686 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 1687 zfid_short_t *zfid = (zfid_short_t *)fidp; 1688 1689 for (i = 0; i < sizeof (zfid->zf_object); i++) 1690 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 1691 1692 for (i = 0; i < sizeof (zfid->zf_gen); i++) 1693 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 1694 } else { 1695 return (SET_ERROR(EINVAL)); 1696 } 1697 1698 /* LONG_FID_LEN means snapdirs */ 1699 if (fidp->fid_len == LONG_FID_LEN) { 1700 zfid_long_t *zlfid = (zfid_long_t *)fidp; 1701 uint64_t objsetid = 0; 1702 uint64_t setgen = 0; 1703 1704 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 1705 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 1706 1707 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 1708 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 1709 1710 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) { 1711 dprintf("snapdir fid: objsetid (%llu) != " 1712 "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n", 1713 objsetid, ZFSCTL_INO_SNAPDIRS, object); 1714 1715 return (SET_ERROR(EINVAL)); 1716 } 1717 1718 if (fid_gen > 1 || setgen != 0) { 1719 dprintf("snapdir fid: fid_gen (%llu) and setgen " 1720 "(%llu)\n", fid_gen, setgen); 1721 return (SET_ERROR(EINVAL)); 1722 } 1723 1724 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp)); 1725 } 1726 1727 ZFS_ENTER(zfsvfs); 1728 /* A zero fid_gen means we are in the .zfs control directories */ 1729 if (fid_gen == 0 && 1730 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 1731 *ipp = zfsvfs->z_ctldir; 1732 ASSERT(*ipp != NULL); 1733 if (object == ZFSCTL_INO_SNAPDIR) { 1734 VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp, 1735 0, kcred, NULL, NULL) == 0); 1736 } else { 1737 /* 1738 * Must have an existing ref, so igrab() 1739 * cannot return NULL 1740 */ 1741 VERIFY3P(igrab(*ipp), !=, NULL); 1742 } 1743 ZFS_EXIT(zfsvfs); 1744 return (0); 1745 } 1746 1747 gen_mask = -1ULL >> (64 - 8 * i); 1748 1749 dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask); 1750 if ((err = zfs_zget(zfsvfs, object, &zp))) { 1751 ZFS_EXIT(zfsvfs); 1752 return (err); 1753 } 1754 1755 /* Don't export xattr stuff */ 1756 if (zp->z_pflags & ZFS_XATTR) { 1757 zrele(zp); 1758 ZFS_EXIT(zfsvfs); 1759 return (SET_ERROR(ENOENT)); 1760 } 1761 1762 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 1763 sizeof (uint64_t)); 1764 zp_gen = zp_gen & gen_mask; 1765 if (zp_gen == 0) 1766 zp_gen = 1; 1767 if ((fid_gen == 0) && (zfsvfs->z_root == object)) 1768 fid_gen = zp_gen; 1769 if (zp->z_unlinked || zp_gen != fid_gen) { 1770 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen, 1771 fid_gen); 1772 zrele(zp); 1773 ZFS_EXIT(zfsvfs); 1774 return (SET_ERROR(ENOENT)); 1775 } 1776 1777 *ipp = ZTOI(zp); 1778 if (*ipp) 1779 zfs_znode_update_vfs(ITOZ(*ipp)); 1780 1781 ZFS_EXIT(zfsvfs); 1782 return (0); 1783} 1784 1785/* 1786 * Block out VFS ops and close zfsvfs_t 1787 * 1788 * Note, if successful, then we return with the 'z_teardown_lock' and 1789 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 1790 * dataset and objset intact so that they can be atomically handed off during 1791 * a subsequent rollback or recv operation and the resume thereafter. 1792 */ 1793int 1794zfs_suspend_fs(zfsvfs_t *zfsvfs) 1795{ 1796 int error; 1797 1798 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 1799 return (error); 1800 1801 return (0); 1802} 1803 1804/* 1805 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 1806 * is an invariant across any of the operations that can be performed while the 1807 * filesystem was suspended. Whether it succeeded or failed, the preconditions 1808 * are the same: the relevant objset and associated dataset are owned by 1809 * zfsvfs, held, and long held on entry. 1810 */ 1811int 1812zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) 1813{ 1814 int err, err2; 1815 znode_t *zp; 1816 1817 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs)); 1818 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1819 1820 /* 1821 * We already own this, so just update the objset_t, as the one we 1822 * had before may have been evicted. 1823 */ 1824 objset_t *os; 1825 VERIFY3P(ds->ds_owner, ==, zfsvfs); 1826 VERIFY(dsl_dataset_long_held(ds)); 1827 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds)); 1828 dsl_pool_config_enter(dp, FTAG); 1829 VERIFY0(dmu_objset_from_ds(ds, &os)); 1830 dsl_pool_config_exit(dp, FTAG); 1831 1832 err = zfsvfs_init(zfsvfs, os); 1833 if (err != 0) 1834 goto bail; 1835 1836 ds->ds_dir->dd_activity_cancelled = B_FALSE; 1837 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 1838 1839 zfs_set_fuid_feature(zfsvfs); 1840 zfsvfs->z_rollback_time = jiffies; 1841 1842 /* 1843 * Attempt to re-establish all the active inodes with their 1844 * dbufs. If a zfs_rezget() fails, then we unhash the inode 1845 * and mark it stale. This prevents a collision if a new 1846 * inode/object is created which must use the same inode 1847 * number. The stale inode will be be released when the 1848 * VFS prunes the dentry holding the remaining references 1849 * on the stale inode. 1850 */ 1851 mutex_enter(&zfsvfs->z_znodes_lock); 1852 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 1853 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1854 err2 = zfs_rezget(zp); 1855 if (err2) { 1856 remove_inode_hash(ZTOI(zp)); 1857 zp->z_is_stale = B_TRUE; 1858 } 1859 1860 /* see comment in zfs_suspend_fs() */ 1861 if (zp->z_suspended) { 1862 zfs_zrele_async(zp); 1863 zp->z_suspended = B_FALSE; 1864 } 1865 } 1866 mutex_exit(&zfsvfs->z_znodes_lock); 1867 1868 if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) { 1869 /* 1870 * zfs_suspend_fs() could have interrupted freeing 1871 * of dnodes. We need to restart this freeing so 1872 * that we don't "leak" the space. 1873 */ 1874 zfs_unlinked_drain(zfsvfs); 1875 } 1876 1877 /* 1878 * Most of the time zfs_suspend_fs is used for changing the contents 1879 * of the underlying dataset. ZFS rollback and receive operations 1880 * might create files for which negative dentries are present in 1881 * the cache. Since walking the dcache would require a lot of GPL-only 1882 * code duplication, it's much easier on these rather rare occasions 1883 * just to flush the whole dcache for the given dataset/filesystem. 1884 */ 1885 shrink_dcache_sb(zfsvfs->z_sb); 1886 1887bail: 1888 if (err != 0) 1889 zfsvfs->z_unmounted = B_TRUE; 1890 1891 /* release the VFS ops */ 1892 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1893 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1894 1895 if (err != 0) { 1896 /* 1897 * Since we couldn't setup the sa framework, try to force 1898 * unmount this file system. 1899 */ 1900 if (zfsvfs->z_os) 1901 (void) zfs_umount(zfsvfs->z_sb); 1902 } 1903 return (err); 1904} 1905 1906/* 1907 * Release VOPs and unmount a suspended filesystem. 1908 */ 1909int 1910zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) 1911{ 1912 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs)); 1913 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1914 1915 /* 1916 * We already own this, so just hold and rele it to update the 1917 * objset_t, as the one we had before may have been evicted. 1918 */ 1919 objset_t *os; 1920 VERIFY3P(ds->ds_owner, ==, zfsvfs); 1921 VERIFY(dsl_dataset_long_held(ds)); 1922 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds)); 1923 dsl_pool_config_enter(dp, FTAG); 1924 VERIFY0(dmu_objset_from_ds(ds, &os)); 1925 dsl_pool_config_exit(dp, FTAG); 1926 zfsvfs->z_os = os; 1927 1928 /* release the VOPs */ 1929 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1930 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1931 1932 /* 1933 * Try to force unmount this file system. 1934 */ 1935 (void) zfs_umount(zfsvfs->z_sb); 1936 zfsvfs->z_unmounted = B_TRUE; 1937 return (0); 1938} 1939 1940/* 1941 * Automounted snapshots rely on periodic revalidation 1942 * to defer snapshots from being automatically unmounted. 1943 */ 1944 1945inline void 1946zfs_exit_fs(zfsvfs_t *zfsvfs) 1947{ 1948 if (!zfsvfs->z_issnap) 1949 return; 1950 1951 if (time_after(jiffies, zfsvfs->z_snap_defer_time + 1952 MAX(zfs_expire_snapshot * HZ / 2, HZ))) { 1953 zfsvfs->z_snap_defer_time = jiffies; 1954 zfsctl_snapshot_unmount_delay(zfsvfs->z_os->os_spa, 1955 dmu_objset_id(zfsvfs->z_os), 1956 zfs_expire_snapshot); 1957 } 1958} 1959 1960int 1961zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 1962{ 1963 int error; 1964 objset_t *os = zfsvfs->z_os; 1965 dmu_tx_t *tx; 1966 1967 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 1968 return (SET_ERROR(EINVAL)); 1969 1970 if (newvers < zfsvfs->z_version) 1971 return (SET_ERROR(EINVAL)); 1972 1973 if (zfs_spa_version_map(newvers) > 1974 spa_version(dmu_objset_spa(zfsvfs->z_os))) 1975 return (SET_ERROR(ENOTSUP)); 1976 1977 tx = dmu_tx_create(os); 1978 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 1979 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 1980 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 1981 ZFS_SA_ATTRS); 1982 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1983 } 1984 error = dmu_tx_assign(tx, TXG_WAIT); 1985 if (error) { 1986 dmu_tx_abort(tx); 1987 return (error); 1988 } 1989 1990 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 1991 8, 1, &newvers, tx); 1992 1993 if (error) { 1994 dmu_tx_commit(tx); 1995 return (error); 1996 } 1997 1998 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 1999 uint64_t sa_obj; 2000 2001 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2002 SPA_VERSION_SA); 2003 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2004 DMU_OT_NONE, 0, tx); 2005 2006 error = zap_add(os, MASTER_NODE_OBJ, 2007 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2008 ASSERT0(error); 2009 2010 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2011 sa_register_update_callback(os, zfs_sa_upgrade); 2012 } 2013 2014 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2015 "from %llu to %llu", zfsvfs->z_version, newvers); 2016 2017 dmu_tx_commit(tx); 2018 2019 zfsvfs->z_version = newvers; 2020 os->os_version = newvers; 2021 2022 zfs_set_fuid_feature(zfsvfs); 2023 2024 return (0); 2025} 2026 2027/* 2028 * Read a property stored within the master node. 2029 */ 2030int 2031zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2032{ 2033 uint64_t *cached_copy = NULL; 2034 2035 /* 2036 * Figure out where in the objset_t the cached copy would live, if it 2037 * is available for the requested property. 2038 */ 2039 if (os != NULL) { 2040 switch (prop) { 2041 case ZFS_PROP_VERSION: 2042 cached_copy = &os->os_version; 2043 break; 2044 case ZFS_PROP_NORMALIZE: 2045 cached_copy = &os->os_normalization; 2046 break; 2047 case ZFS_PROP_UTF8ONLY: 2048 cached_copy = &os->os_utf8only; 2049 break; 2050 case ZFS_PROP_CASE: 2051 cached_copy = &os->os_casesensitivity; 2052 break; 2053 default: 2054 break; 2055 } 2056 } 2057 if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) { 2058 *value = *cached_copy; 2059 return (0); 2060 } 2061 2062 /* 2063 * If the property wasn't cached, look up the file system's value for 2064 * the property. For the version property, we look up a slightly 2065 * different string. 2066 */ 2067 const char *pname; 2068 int error = ENOENT; 2069 if (prop == ZFS_PROP_VERSION) 2070 pname = ZPL_VERSION_STR; 2071 else 2072 pname = zfs_prop_to_name(prop); 2073 2074 if (os != NULL) { 2075 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS); 2076 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2077 } 2078 2079 if (error == ENOENT) { 2080 /* No value set, use the default value */ 2081 switch (prop) { 2082 case ZFS_PROP_VERSION: 2083 *value = ZPL_VERSION; 2084 break; 2085 case ZFS_PROP_NORMALIZE: 2086 case ZFS_PROP_UTF8ONLY: 2087 *value = 0; 2088 break; 2089 case ZFS_PROP_CASE: 2090 *value = ZFS_CASE_SENSITIVE; 2091 break; 2092 case ZFS_PROP_ACLTYPE: 2093 *value = ZFS_ACLTYPE_OFF; 2094 break; 2095 default: 2096 return (error); 2097 } 2098 error = 0; 2099 } 2100 2101 /* 2102 * If one of the methods for getting the property value above worked, 2103 * copy it into the objset_t's cache. 2104 */ 2105 if (error == 0 && cached_copy != NULL) { 2106 *cached_copy = *value; 2107 } 2108 2109 return (error); 2110} 2111 2112/* 2113 * Return true if the corresponding vfs's unmounted flag is set. 2114 * Otherwise return false. 2115 * If this function returns true we know VFS unmount has been initiated. 2116 */ 2117boolean_t 2118zfs_get_vfs_flag_unmounted(objset_t *os) 2119{ 2120 zfsvfs_t *zfvp; 2121 boolean_t unmounted = B_FALSE; 2122 2123 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS); 2124 2125 mutex_enter(&os->os_user_ptr_lock); 2126 zfvp = dmu_objset_get_user(os); 2127 if (zfvp != NULL && zfvp->z_unmounted) 2128 unmounted = B_TRUE; 2129 mutex_exit(&os->os_user_ptr_lock); 2130 2131 return (unmounted); 2132} 2133 2134/*ARGSUSED*/ 2135void 2136zfsvfs_update_fromname(const char *oldname, const char *newname) 2137{ 2138 /* 2139 * We don't need to do anything here, the devname is always current by 2140 * virtue of zfsvfs->z_sb->s_op->show_devname. 2141 */ 2142} 2143 2144void 2145zfs_init(void) 2146{ 2147 zfsctl_init(); 2148 zfs_znode_init(); 2149 dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info); 2150 register_filesystem(&zpl_fs_type); 2151} 2152 2153void 2154zfs_fini(void) 2155{ 2156 /* 2157 * we don't use outstanding because zpl_posix_acl_free might add more. 2158 */ 2159 taskq_wait(system_delay_taskq); 2160 taskq_wait(system_taskq); 2161 unregister_filesystem(&zpl_fs_type); 2162 zfs_znode_fini(); 2163 zfsctl_fini(); 2164} 2165 2166#if defined(_KERNEL) 2167EXPORT_SYMBOL(zfs_suspend_fs); 2168EXPORT_SYMBOL(zfs_resume_fs); 2169EXPORT_SYMBOL(zfs_set_version); 2170EXPORT_SYMBOL(zfsvfs_create); 2171EXPORT_SYMBOL(zfsvfs_free); 2172EXPORT_SYMBOL(zfs_is_readonly); 2173EXPORT_SYMBOL(zfs_domount); 2174EXPORT_SYMBOL(zfs_preumount); 2175EXPORT_SYMBOL(zfs_umount); 2176EXPORT_SYMBOL(zfs_remount); 2177EXPORT_SYMBOL(zfs_statvfs); 2178EXPORT_SYMBOL(zfs_vget); 2179EXPORT_SYMBOL(zfs_prune); 2180#endif 2181