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 https://opensource.org/licenses/CDDL-1.0. 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 65 enum { 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 88 static 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 }; 110 111 static void 112 zfsvfs_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 } 121 122 static int 123 zfsvfs_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 */ 214 static int 215 zfsvfs_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 } 252 253 boolean_t 254 zfs_is_readonly(zfsvfs_t *zfsvfs) 255 { 256 return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY)); 257 } 258 259 int 260 zfs_sync(struct super_block *sb, int wait, cred_t *cr) 261 { 262 (void) cr; 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 } 305 306 static void 307 atime_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 } 325 326 static void 327 relatime_changed_cb(void *arg, uint64_t newval) 328 { 329 ((zfsvfs_t *)arg)->z_relatime = newval; 330 } 331 332 static void 333 xattr_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 } 348 349 static void 350 acltype_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 } 373 374 static void 375 blksz_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 } 384 385 static void 386 readonly_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 } 399 400 static void 401 devices_changed_cb(void *arg, uint64_t newval) 402 { 403 } 404 405 static void 406 setuid_changed_cb(void *arg, uint64_t newval) 407 { 408 } 409 410 static void 411 exec_changed_cb(void *arg, uint64_t newval) 412 { 413 } 414 415 static void 416 nbmand_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 } 429 430 static void 431 snapdir_changed_cb(void *arg, uint64_t newval) 432 { 433 ((zfsvfs_t *)arg)->z_show_ctldir = newval; 434 } 435 436 static void 437 acl_mode_changed_cb(void *arg, uint64_t newval) 438 { 439 zfsvfs_t *zfsvfs = arg; 440 441 zfsvfs->z_acl_mode = newval; 442 } 443 444 static void 445 acl_inherit_changed_cb(void *arg, uint64_t newval) 446 { 447 ((zfsvfs_t *)arg)->z_acl_inherit = newval; 448 } 449 450 static int 451 zfs_register_callbacks(vfs_t *vfsp) 452 { 453 struct dsl_dataset *ds = NULL; 454 objset_t *os = NULL; 455 zfsvfs_t *zfsvfs = NULL; 456 int error = 0; 457 458 ASSERT(vfsp); 459 zfsvfs = vfsp->vfs_data; 460 ASSERT(zfsvfs); 461 os = zfsvfs->z_os; 462 463 /* 464 * The act of registering our callbacks will destroy any mount 465 * options we may have. In order to enable temporary overrides 466 * of mount options, we stash away the current values and 467 * restore them after we register the callbacks. 468 */ 469 if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) { 470 vfsp->vfs_do_readonly = B_TRUE; 471 vfsp->vfs_readonly = B_TRUE; 472 } 473 474 /* 475 * Register property callbacks. 476 * 477 * It would probably be fine to just check for i/o error from 478 * the first prop_register(), but I guess I like to go 479 * overboard... 480 */ 481 ds = dmu_objset_ds(os); 482 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 483 error = dsl_prop_register(ds, 484 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 485 error = error ? error : dsl_prop_register(ds, 486 zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs); 487 error = error ? error : dsl_prop_register(ds, 488 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 489 error = error ? error : dsl_prop_register(ds, 490 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 491 error = error ? error : dsl_prop_register(ds, 492 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 493 error = error ? error : dsl_prop_register(ds, 494 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 495 error = error ? error : dsl_prop_register(ds, 496 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 497 error = error ? error : dsl_prop_register(ds, 498 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 499 error = error ? error : dsl_prop_register(ds, 500 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 501 error = error ? error : dsl_prop_register(ds, 502 zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs); 503 error = error ? error : dsl_prop_register(ds, 504 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 505 error = error ? error : dsl_prop_register(ds, 506 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 507 zfsvfs); 508 error = error ? error : dsl_prop_register(ds, 509 zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs); 510 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 511 if (error) 512 goto unregister; 513 514 /* 515 * Invoke our callbacks to restore temporary mount options. 516 */ 517 if (vfsp->vfs_do_readonly) 518 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly); 519 if (vfsp->vfs_do_setuid) 520 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid); 521 if (vfsp->vfs_do_exec) 522 exec_changed_cb(zfsvfs, vfsp->vfs_exec); 523 if (vfsp->vfs_do_devices) 524 devices_changed_cb(zfsvfs, vfsp->vfs_devices); 525 if (vfsp->vfs_do_xattr) 526 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr); 527 if (vfsp->vfs_do_atime) 528 atime_changed_cb(zfsvfs, vfsp->vfs_atime); 529 if (vfsp->vfs_do_relatime) 530 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime); 531 if (vfsp->vfs_do_nbmand) 532 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand); 533 534 return (0); 535 536 unregister: 537 dsl_prop_unregister_all(ds, zfsvfs); 538 return (error); 539 } 540 541 /* 542 * Takes a dataset, a property, a value and that value's setpoint as 543 * found in the ZAP. Checks if the property has been changed in the vfs. 544 * If so, val and setpoint will be overwritten with updated content. 545 * Otherwise, they are left unchanged. 546 */ 547 int 548 zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val, 549 char *setpoint) 550 { 551 int error; 552 zfsvfs_t *zfvp; 553 vfs_t *vfsp; 554 objset_t *os; 555 uint64_t tmp = *val; 556 557 error = dmu_objset_from_ds(ds, &os); 558 if (error != 0) 559 return (error); 560 561 if (dmu_objset_type(os) != DMU_OST_ZFS) 562 return (EINVAL); 563 564 mutex_enter(&os->os_user_ptr_lock); 565 zfvp = dmu_objset_get_user(os); 566 mutex_exit(&os->os_user_ptr_lock); 567 if (zfvp == NULL) 568 return (ESRCH); 569 570 vfsp = zfvp->z_vfs; 571 572 switch (zfs_prop) { 573 case ZFS_PROP_ATIME: 574 if (vfsp->vfs_do_atime) 575 tmp = vfsp->vfs_atime; 576 break; 577 case ZFS_PROP_RELATIME: 578 if (vfsp->vfs_do_relatime) 579 tmp = vfsp->vfs_relatime; 580 break; 581 case ZFS_PROP_DEVICES: 582 if (vfsp->vfs_do_devices) 583 tmp = vfsp->vfs_devices; 584 break; 585 case ZFS_PROP_EXEC: 586 if (vfsp->vfs_do_exec) 587 tmp = vfsp->vfs_exec; 588 break; 589 case ZFS_PROP_SETUID: 590 if (vfsp->vfs_do_setuid) 591 tmp = vfsp->vfs_setuid; 592 break; 593 case ZFS_PROP_READONLY: 594 if (vfsp->vfs_do_readonly) 595 tmp = vfsp->vfs_readonly; 596 break; 597 case ZFS_PROP_XATTR: 598 if (vfsp->vfs_do_xattr) 599 tmp = vfsp->vfs_xattr; 600 break; 601 case ZFS_PROP_NBMAND: 602 if (vfsp->vfs_do_nbmand) 603 tmp = vfsp->vfs_nbmand; 604 break; 605 default: 606 return (ENOENT); 607 } 608 609 if (tmp != *val) { 610 (void) strcpy(setpoint, "temporary"); 611 *val = tmp; 612 } 613 return (0); 614 } 615 616 /* 617 * Associate this zfsvfs with the given objset, which must be owned. 618 * This will cache a bunch of on-disk state from the objset in the 619 * zfsvfs. 620 */ 621 static int 622 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os) 623 { 624 int error; 625 uint64_t val; 626 627 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 628 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 629 zfsvfs->z_os = os; 630 631 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 632 if (error != 0) 633 return (error); 634 if (zfsvfs->z_version > 635 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 636 (void) printk("Can't mount a version %lld file system " 637 "on a version %lld pool\n. Pool must be upgraded to mount " 638 "this file system.\n", (u_longlong_t)zfsvfs->z_version, 639 (u_longlong_t)spa_version(dmu_objset_spa(os))); 640 return (SET_ERROR(ENOTSUP)); 641 } 642 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val); 643 if (error != 0) 644 return (error); 645 zfsvfs->z_norm = (int)val; 646 647 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val); 648 if (error != 0) 649 return (error); 650 zfsvfs->z_utf8 = (val != 0); 651 652 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val); 653 if (error != 0) 654 return (error); 655 zfsvfs->z_case = (uint_t)val; 656 657 if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0) 658 return (error); 659 zfsvfs->z_acl_type = (uint_t)val; 660 661 /* 662 * Fold case on file systems that are always or sometimes case 663 * insensitive. 664 */ 665 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 666 zfsvfs->z_case == ZFS_CASE_MIXED) 667 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 668 669 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 670 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 671 672 uint64_t sa_obj = 0; 673 if (zfsvfs->z_use_sa) { 674 /* should either have both of these objects or none */ 675 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 676 &sa_obj); 677 if (error != 0) 678 return (error); 679 680 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val); 681 if ((error == 0) && (val == ZFS_XATTR_SA)) 682 zfsvfs->z_xattr_sa = B_TRUE; 683 } 684 685 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 686 &zfsvfs->z_root); 687 if (error != 0) 688 return (error); 689 ASSERT(zfsvfs->z_root != 0); 690 691 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 692 &zfsvfs->z_unlinkedobj); 693 if (error != 0) 694 return (error); 695 696 error = zap_lookup(os, MASTER_NODE_OBJ, 697 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 698 8, 1, &zfsvfs->z_userquota_obj); 699 if (error == ENOENT) 700 zfsvfs->z_userquota_obj = 0; 701 else if (error != 0) 702 return (error); 703 704 error = zap_lookup(os, MASTER_NODE_OBJ, 705 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 706 8, 1, &zfsvfs->z_groupquota_obj); 707 if (error == ENOENT) 708 zfsvfs->z_groupquota_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_PROJECTQUOTA], 714 8, 1, &zfsvfs->z_projectquota_obj); 715 if (error == ENOENT) 716 zfsvfs->z_projectquota_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_USEROBJQUOTA], 722 8, 1, &zfsvfs->z_userobjquota_obj); 723 if (error == ENOENT) 724 zfsvfs->z_userobjquota_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_GROUPOBJQUOTA], 730 8, 1, &zfsvfs->z_groupobjquota_obj); 731 if (error == ENOENT) 732 zfsvfs->z_groupobjquota_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_PROJECTOBJQUOTA], 738 8, 1, &zfsvfs->z_projectobjquota_obj); 739 if (error == ENOENT) 740 zfsvfs->z_projectobjquota_obj = 0; 741 else if (error != 0) 742 return (error); 743 744 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 745 &zfsvfs->z_fuid_obj); 746 if (error == ENOENT) 747 zfsvfs->z_fuid_obj = 0; 748 else if (error != 0) 749 return (error); 750 751 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 752 &zfsvfs->z_shares_dir); 753 if (error == ENOENT) 754 zfsvfs->z_shares_dir = 0; 755 else if (error != 0) 756 return (error); 757 758 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 759 &zfsvfs->z_attr_table); 760 if (error != 0) 761 return (error); 762 763 if (zfsvfs->z_version >= ZPL_VERSION_SA) 764 sa_register_update_callback(os, zfs_sa_upgrade); 765 766 return (0); 767 } 768 769 int 770 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp) 771 { 772 objset_t *os; 773 zfsvfs_t *zfsvfs; 774 int error; 775 boolean_t ro = (readonly || (strchr(osname, '@') != NULL)); 776 777 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 778 779 error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os); 780 if (error != 0) { 781 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 782 return (error); 783 } 784 785 error = zfsvfs_create_impl(zfvp, zfsvfs, os); 786 if (error != 0) { 787 dmu_objset_disown(os, B_TRUE, zfsvfs); 788 } 789 return (error); 790 } 791 792 793 /* 794 * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function 795 * on a failure. Do not pass in a statically allocated zfsvfs. 796 */ 797 int 798 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os) 799 { 800 int error; 801 802 zfsvfs->z_vfs = NULL; 803 zfsvfs->z_sb = NULL; 804 zfsvfs->z_parent = zfsvfs; 805 806 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 807 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 808 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 809 offsetof(znode_t, z_link_node)); 810 ZFS_TEARDOWN_INIT(zfsvfs); 811 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 812 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 813 814 int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1), 815 ZFS_OBJ_MTX_MAX); 816 zfsvfs->z_hold_size = size; 817 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size, 818 KM_SLEEP); 819 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP); 820 for (int i = 0; i != size; i++) { 821 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare, 822 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node)); 823 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL); 824 } 825 826 error = zfsvfs_init(zfsvfs, os); 827 if (error != 0) { 828 *zfvp = NULL; 829 zfsvfs_free(zfsvfs); 830 return (error); 831 } 832 833 zfsvfs->z_drain_task = TASKQID_INVALID; 834 zfsvfs->z_draining = B_FALSE; 835 zfsvfs->z_drain_cancel = B_TRUE; 836 837 *zfvp = zfsvfs; 838 return (0); 839 } 840 841 static int 842 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 843 { 844 int error; 845 boolean_t readonly = zfs_is_readonly(zfsvfs); 846 847 error = zfs_register_callbacks(zfsvfs->z_vfs); 848 if (error) 849 return (error); 850 851 /* 852 * If we are not mounting (ie: online recv), then we don't 853 * have to worry about replaying the log as we blocked all 854 * operations out since we closed the ZIL. 855 */ 856 if (mounting) { 857 ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL); 858 error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os); 859 if (error) 860 return (error); 861 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data, 862 &zfsvfs->z_kstat.dk_zil_sums); 863 864 /* 865 * During replay we remove the read only flag to 866 * allow replays to succeed. 867 */ 868 if (readonly != 0) { 869 readonly_changed_cb(zfsvfs, B_FALSE); 870 } else { 871 zap_stats_t zs; 872 if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj, 873 &zs) == 0) { 874 dataset_kstats_update_nunlinks_kstat( 875 &zfsvfs->z_kstat, zs.zs_num_entries); 876 dprintf_ds(zfsvfs->z_os->os_dsl_dataset, 877 "num_entries in unlinked set: %llu", 878 zs.zs_num_entries); 879 } 880 zfs_unlinked_drain(zfsvfs); 881 dsl_dir_t *dd = zfsvfs->z_os->os_dsl_dataset->ds_dir; 882 dd->dd_activity_cancelled = B_FALSE; 883 } 884 885 /* 886 * Parse and replay the intent log. 887 * 888 * Because of ziltest, this must be done after 889 * zfs_unlinked_drain(). (Further note: ziltest 890 * doesn't use readonly mounts, where 891 * zfs_unlinked_drain() isn't called.) This is because 892 * ziltest causes spa_sync() to think it's committed, 893 * but actually it is not, so the intent log contains 894 * many txg's worth of changes. 895 * 896 * In particular, if object N is in the unlinked set in 897 * the last txg to actually sync, then it could be 898 * actually freed in a later txg and then reallocated 899 * in a yet later txg. This would write a "create 900 * object N" record to the intent log. Normally, this 901 * would be fine because the spa_sync() would have 902 * written out the fact that object N is free, before 903 * we could write the "create object N" intent log 904 * record. 905 * 906 * But when we are in ziltest mode, we advance the "open 907 * txg" without actually spa_sync()-ing the changes to 908 * disk. So we would see that object N is still 909 * allocated and in the unlinked set, and there is an 910 * intent log record saying to allocate it. 911 */ 912 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 913 if (zil_replay_disable) { 914 zil_destroy(zfsvfs->z_log, B_FALSE); 915 } else { 916 zfsvfs->z_replay = B_TRUE; 917 zil_replay(zfsvfs->z_os, zfsvfs, 918 zfs_replay_vector); 919 zfsvfs->z_replay = B_FALSE; 920 } 921 } 922 923 /* restore readonly bit */ 924 if (readonly != 0) 925 readonly_changed_cb(zfsvfs, B_TRUE); 926 } else { 927 ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL); 928 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data, 929 &zfsvfs->z_kstat.dk_zil_sums); 930 } 931 932 /* 933 * Set the objset user_ptr to track its zfsvfs. 934 */ 935 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 936 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 937 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 938 939 return (0); 940 } 941 942 void 943 zfsvfs_free(zfsvfs_t *zfsvfs) 944 { 945 int i, size = zfsvfs->z_hold_size; 946 947 zfs_fuid_destroy(zfsvfs); 948 949 mutex_destroy(&zfsvfs->z_znodes_lock); 950 mutex_destroy(&zfsvfs->z_lock); 951 list_destroy(&zfsvfs->z_all_znodes); 952 ZFS_TEARDOWN_DESTROY(zfsvfs); 953 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 954 rw_destroy(&zfsvfs->z_fuid_lock); 955 for (i = 0; i != size; i++) { 956 avl_destroy(&zfsvfs->z_hold_trees[i]); 957 mutex_destroy(&zfsvfs->z_hold_locks[i]); 958 } 959 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size); 960 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size); 961 zfsvfs_vfs_free(zfsvfs->z_vfs); 962 dataset_kstats_destroy(&zfsvfs->z_kstat); 963 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 964 } 965 966 static void 967 zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 968 { 969 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 970 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 971 } 972 973 static void 974 zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 975 { 976 objset_t *os = zfsvfs->z_os; 977 978 if (!dmu_objset_is_snapshot(os)) 979 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 980 } 981 982 #ifdef HAVE_MLSLABEL 983 /* 984 * Check that the hex label string is appropriate for the dataset being 985 * mounted into the global_zone proper. 986 * 987 * Return an error if the hex label string is not default or 988 * admin_low/admin_high. For admin_low labels, the corresponding 989 * dataset must be readonly. 990 */ 991 int 992 zfs_check_global_label(const char *dsname, const char *hexsl) 993 { 994 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 995 return (0); 996 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 997 return (0); 998 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 999 /* must be readonly */ 1000 uint64_t rdonly; 1001 1002 if (dsl_prop_get_integer(dsname, 1003 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1004 return (SET_ERROR(EACCES)); 1005 return (rdonly ? 0 : SET_ERROR(EACCES)); 1006 } 1007 return (SET_ERROR(EACCES)); 1008 } 1009 #endif /* HAVE_MLSLABEL */ 1010 1011 static int 1012 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp, 1013 uint32_t bshift) 1014 { 1015 char buf[20 + DMU_OBJACCT_PREFIX_LEN]; 1016 uint64_t offset = DMU_OBJACCT_PREFIX_LEN; 1017 uint64_t quota; 1018 uint64_t used; 1019 int err; 1020 1021 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1); 1022 err = zfs_id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset, 1023 sizeof (buf) - offset, B_FALSE); 1024 if (err) 1025 return (err); 1026 1027 if (zfsvfs->z_projectquota_obj == 0) 1028 goto objs; 1029 1030 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj, 1031 buf + offset, 8, 1, "a); 1032 if (err == ENOENT) 1033 goto objs; 1034 else if (err) 1035 return (err); 1036 1037 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT, 1038 buf + offset, 8, 1, &used); 1039 if (unlikely(err == ENOENT)) { 1040 uint32_t blksize; 1041 u_longlong_t nblocks; 1042 1043 /* 1044 * Quota accounting is async, so it is possible race case. 1045 * There is at least one object with the given project ID. 1046 */ 1047 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 1048 if (unlikely(zp->z_blksz == 0)) 1049 blksize = zfsvfs->z_max_blksz; 1050 1051 used = blksize * nblocks; 1052 } else if (err) { 1053 return (err); 1054 } 1055 1056 statp->f_blocks = quota >> bshift; 1057 statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0; 1058 statp->f_bavail = statp->f_bfree; 1059 1060 objs: 1061 if (zfsvfs->z_projectobjquota_obj == 0) 1062 return (0); 1063 1064 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj, 1065 buf + offset, 8, 1, "a); 1066 if (err == ENOENT) 1067 return (0); 1068 else if (err) 1069 return (err); 1070 1071 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT, 1072 buf, 8, 1, &used); 1073 if (unlikely(err == ENOENT)) { 1074 /* 1075 * Quota accounting is async, so it is possible race case. 1076 * There is at least one object with the given project ID. 1077 */ 1078 used = 1; 1079 } else if (err) { 1080 return (err); 1081 } 1082 1083 statp->f_files = quota; 1084 statp->f_ffree = (quota > used) ? (quota - used) : 0; 1085 1086 return (0); 1087 } 1088 1089 int 1090 zfs_statvfs(struct inode *ip, struct kstatfs *statp) 1091 { 1092 zfsvfs_t *zfsvfs = ITOZSB(ip); 1093 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1094 int err = 0; 1095 1096 ZFS_ENTER(zfsvfs); 1097 1098 dmu_objset_space(zfsvfs->z_os, 1099 &refdbytes, &availbytes, &usedobjs, &availobjs); 1100 1101 uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os); 1102 /* 1103 * The underlying storage pool actually uses multiple block 1104 * size. Under Solaris frsize (fragment size) is reported as 1105 * the smallest block size we support, and bsize (block size) 1106 * as the filesystem's maximum block size. Unfortunately, 1107 * under Linux the fragment size and block size are often used 1108 * interchangeably. Thus we are forced to report both of them 1109 * as the filesystem's maximum block size. 1110 */ 1111 statp->f_frsize = zfsvfs->z_max_blksz; 1112 statp->f_bsize = zfsvfs->z_max_blksz; 1113 uint32_t bshift = fls(statp->f_bsize) - 1; 1114 1115 /* 1116 * The following report "total" blocks of various kinds in 1117 * the file system, but reported in terms of f_bsize - the 1118 * "preferred" size. 1119 */ 1120 1121 /* Round up so we never have a filesystem using 0 blocks. */ 1122 refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize); 1123 statp->f_blocks = (refdbytes + availbytes) >> bshift; 1124 statp->f_bfree = availbytes >> bshift; 1125 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1126 1127 /* 1128 * statvfs() should really be called statufs(), because it assumes 1129 * static metadata. ZFS doesn't preallocate files, so the best 1130 * we can do is report the max that could possibly fit in f_files, 1131 * and that minus the number actually used in f_ffree. 1132 * For f_ffree, report the smaller of the number of objects available 1133 * and the number of blocks (each object will take at least a block). 1134 */ 1135 statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT); 1136 statp->f_files = statp->f_ffree + usedobjs; 1137 statp->f_fsid.val[0] = (uint32_t)fsid; 1138 statp->f_fsid.val[1] = (uint32_t)(fsid >> 32); 1139 statp->f_type = ZFS_SUPER_MAGIC; 1140 statp->f_namelen = MAXNAMELEN - 1; 1141 1142 /* 1143 * We have all of 40 characters to stuff a string here. 1144 * Is there anything useful we could/should provide? 1145 */ 1146 memset(statp->f_spare, 0, sizeof (statp->f_spare)); 1147 1148 if (dmu_objset_projectquota_enabled(zfsvfs->z_os) && 1149 dmu_objset_projectquota_present(zfsvfs->z_os)) { 1150 znode_t *zp = ITOZ(ip); 1151 1152 if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid && 1153 zpl_is_valid_projid(zp->z_projid)) 1154 err = zfs_statfs_project(zfsvfs, zp, statp, bshift); 1155 } 1156 1157 ZFS_EXIT(zfsvfs); 1158 return (err); 1159 } 1160 1161 static int 1162 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp) 1163 { 1164 znode_t *rootzp; 1165 int error; 1166 1167 ZFS_ENTER(zfsvfs); 1168 1169 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1170 if (error == 0) 1171 *ipp = ZTOI(rootzp); 1172 1173 ZFS_EXIT(zfsvfs); 1174 return (error); 1175 } 1176 1177 /* 1178 * Linux kernels older than 3.1 do not support a per-filesystem shrinker. 1179 * To accommodate this we must improvise and manually walk the list of znodes 1180 * attempting to prune dentries in order to be able to drop the inodes. 1181 * 1182 * To avoid scanning the same znodes multiple times they are always rotated 1183 * to the end of the z_all_znodes list. New znodes are inserted at the 1184 * end of the list so we're always scanning the oldest znodes first. 1185 */ 1186 static int 1187 zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan) 1188 { 1189 znode_t **zp_array, *zp; 1190 int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *)); 1191 int objects = 0; 1192 int i = 0, j = 0; 1193 1194 zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP); 1195 1196 mutex_enter(&zfsvfs->z_znodes_lock); 1197 while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) { 1198 1199 if ((i++ > nr_to_scan) || (j >= max_array)) 1200 break; 1201 1202 ASSERT(list_link_active(&zp->z_link_node)); 1203 list_remove(&zfsvfs->z_all_znodes, zp); 1204 list_insert_tail(&zfsvfs->z_all_znodes, zp); 1205 1206 /* Skip active znodes and .zfs entries */ 1207 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir) 1208 continue; 1209 1210 if (igrab(ZTOI(zp)) == NULL) 1211 continue; 1212 1213 zp_array[j] = zp; 1214 j++; 1215 } 1216 mutex_exit(&zfsvfs->z_znodes_lock); 1217 1218 for (i = 0; i < j; i++) { 1219 zp = zp_array[i]; 1220 1221 ASSERT3P(zp, !=, NULL); 1222 d_prune_aliases(ZTOI(zp)); 1223 1224 if (atomic_read(&ZTOI(zp)->i_count) == 1) 1225 objects++; 1226 1227 zrele(zp); 1228 } 1229 1230 kmem_free(zp_array, max_array * sizeof (znode_t *)); 1231 1232 return (objects); 1233 } 1234 1235 /* 1236 * The ARC has requested that the filesystem drop entries from the dentry 1237 * and inode caches. This can occur when the ARC needs to free meta data 1238 * blocks but can't because they are all pinned by entries in these caches. 1239 */ 1240 int 1241 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects) 1242 { 1243 zfsvfs_t *zfsvfs = sb->s_fs_info; 1244 int error = 0; 1245 struct shrinker *shrinker = &sb->s_shrink; 1246 struct shrink_control sc = { 1247 .nr_to_scan = nr_to_scan, 1248 .gfp_mask = GFP_KERNEL, 1249 }; 1250 1251 ZFS_ENTER(zfsvfs); 1252 1253 #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \ 1254 defined(SHRINK_CONTROL_HAS_NID) && \ 1255 defined(SHRINKER_NUMA_AWARE) 1256 if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) { 1257 *objects = 0; 1258 for_each_online_node(sc.nid) { 1259 *objects += (*shrinker->scan_objects)(shrinker, &sc); 1260 /* 1261 * reset sc.nr_to_scan, modified by 1262 * scan_objects == super_cache_scan 1263 */ 1264 sc.nr_to_scan = nr_to_scan; 1265 } 1266 } else { 1267 *objects = (*shrinker->scan_objects)(shrinker, &sc); 1268 } 1269 1270 #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK) 1271 *objects = (*shrinker->scan_objects)(shrinker, &sc); 1272 #elif defined(HAVE_SINGLE_SHRINKER_CALLBACK) 1273 *objects = (*shrinker->shrink)(shrinker, &sc); 1274 #elif defined(HAVE_D_PRUNE_ALIASES) 1275 #define D_PRUNE_ALIASES_IS_DEFAULT 1276 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan); 1277 #else 1278 #error "No available dentry and inode cache pruning mechanism." 1279 #endif 1280 1281 #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT) 1282 #undef D_PRUNE_ALIASES_IS_DEFAULT 1283 /* 1284 * Fall back to zfs_prune_aliases if the kernel's per-superblock 1285 * shrinker couldn't free anything, possibly due to the inodes being 1286 * allocated in a different memcg. 1287 */ 1288 if (*objects == 0) 1289 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan); 1290 #endif 1291 1292 ZFS_EXIT(zfsvfs); 1293 1294 dprintf_ds(zfsvfs->z_os->os_dsl_dataset, 1295 "pruning, nr_to_scan=%lu objects=%d error=%d\n", 1296 nr_to_scan, *objects, error); 1297 1298 return (error); 1299 } 1300 1301 /* 1302 * Teardown the zfsvfs_t. 1303 * 1304 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock' 1305 * and 'z_teardown_inactive_lock' held. 1306 */ 1307 static int 1308 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1309 { 1310 znode_t *zp; 1311 1312 zfs_unlinked_drain_stop_wait(zfsvfs); 1313 1314 /* 1315 * If someone has not already unmounted this file system, 1316 * drain the zrele_taskq to ensure all active references to the 1317 * zfsvfs_t have been handled only then can it be safely destroyed. 1318 */ 1319 if (zfsvfs->z_os) { 1320 /* 1321 * If we're unmounting we have to wait for the list to 1322 * drain completely. 1323 * 1324 * If we're not unmounting there's no guarantee the list 1325 * will drain completely, but iputs run from the taskq 1326 * may add the parents of dir-based xattrs to the taskq 1327 * so we want to wait for these. 1328 * 1329 * We can safely read z_nr_znodes without locking because the 1330 * VFS has already blocked operations which add to the 1331 * z_all_znodes list and thus increment z_nr_znodes. 1332 */ 1333 int round = 0; 1334 while (zfsvfs->z_nr_znodes > 0) { 1335 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1336 dmu_objset_pool(zfsvfs->z_os)), 0); 1337 if (++round > 1 && !unmounting) 1338 break; 1339 } 1340 } 1341 1342 ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG); 1343 1344 if (!unmounting) { 1345 /* 1346 * We purge the parent filesystem's super block as the 1347 * parent filesystem and all of its snapshots have their 1348 * inode's super block set to the parent's filesystem's 1349 * super block. Note, 'z_parent' is self referential 1350 * for non-snapshots. 1351 */ 1352 shrink_dcache_sb(zfsvfs->z_parent->z_sb); 1353 } 1354 1355 /* 1356 * Close the zil. NB: Can't close the zil while zfs_inactive 1357 * threads are blocked as zil_close can call zfs_inactive. 1358 */ 1359 if (zfsvfs->z_log) { 1360 zil_close(zfsvfs->z_log); 1361 zfsvfs->z_log = NULL; 1362 } 1363 1364 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1365 1366 /* 1367 * If we are not unmounting (ie: online recv) and someone already 1368 * unmounted this file system while we were doing the switcheroo, 1369 * or a reopen of z_os failed then just bail out now. 1370 */ 1371 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1372 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1373 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1374 return (SET_ERROR(EIO)); 1375 } 1376 1377 /* 1378 * At this point there are no VFS ops active, and any new VFS ops 1379 * will fail with EIO since we have z_teardown_lock for writer (only 1380 * relevant for forced unmount). 1381 * 1382 * Release all holds on dbufs. We also grab an extra reference to all 1383 * the remaining inodes so that the kernel does not attempt to free 1384 * any inodes of a suspended fs. This can cause deadlocks since the 1385 * zfs_resume_fs() process may involve starting threads, which might 1386 * attempt to free unreferenced inodes to free up memory for the new 1387 * thread. 1388 */ 1389 if (!unmounting) { 1390 mutex_enter(&zfsvfs->z_znodes_lock); 1391 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1392 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1393 if (zp->z_sa_hdl) 1394 zfs_znode_dmu_fini(zp); 1395 if (igrab(ZTOI(zp)) != NULL) 1396 zp->z_suspended = B_TRUE; 1397 1398 } 1399 mutex_exit(&zfsvfs->z_znodes_lock); 1400 } 1401 1402 /* 1403 * If we are unmounting, set the unmounted flag and let new VFS ops 1404 * unblock. zfs_inactive will have the unmounted behavior, and all 1405 * other VFS ops will fail with EIO. 1406 */ 1407 if (unmounting) { 1408 zfsvfs->z_unmounted = B_TRUE; 1409 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1410 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1411 } 1412 1413 /* 1414 * z_os will be NULL if there was an error in attempting to reopen 1415 * zfsvfs, so just return as the properties had already been 1416 * 1417 * unregistered and cached data had been evicted before. 1418 */ 1419 if (zfsvfs->z_os == NULL) 1420 return (0); 1421 1422 /* 1423 * Unregister properties. 1424 */ 1425 zfs_unregister_callbacks(zfsvfs); 1426 1427 /* 1428 * Evict cached data. We must write out any dirty data before 1429 * disowning the dataset. 1430 */ 1431 objset_t *os = zfsvfs->z_os; 1432 boolean_t os_dirty = B_FALSE; 1433 for (int t = 0; t < TXG_SIZE; t++) { 1434 if (dmu_objset_is_dirty(os, t)) { 1435 os_dirty = B_TRUE; 1436 break; 1437 } 1438 } 1439 if (!zfs_is_readonly(zfsvfs) && os_dirty) { 1440 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1441 } 1442 dmu_objset_evict_dbufs(zfsvfs->z_os); 1443 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 1444 dsl_dir_cancel_waiters(dd); 1445 1446 return (0); 1447 } 1448 1449 #if defined(HAVE_SUPER_SETUP_BDI_NAME) 1450 atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0); 1451 #endif 1452 1453 int 1454 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent) 1455 { 1456 const char *osname = zm->mnt_osname; 1457 struct inode *root_inode = NULL; 1458 uint64_t recordsize; 1459 int error = 0; 1460 zfsvfs_t *zfsvfs = NULL; 1461 vfs_t *vfs = NULL; 1462 int canwrite; 1463 int dataset_visible_zone; 1464 1465 ASSERT(zm); 1466 ASSERT(osname); 1467 1468 dataset_visible_zone = zone_dataset_visible(osname, &canwrite); 1469 1470 /* 1471 * Refuse to mount a filesystem if we are in a namespace and the 1472 * dataset is not visible or writable in that namespace. 1473 */ 1474 if (!INGLOBALZONE(curproc) && 1475 (!dataset_visible_zone || !canwrite)) { 1476 return (SET_ERROR(EPERM)); 1477 } 1478 1479 error = zfsvfs_parse_options(zm->mnt_data, &vfs); 1480 if (error) 1481 return (error); 1482 1483 /* 1484 * If a non-writable filesystem is being mounted without the 1485 * read-only flag, pretend it was set, as done for snapshots. 1486 */ 1487 if (!canwrite) 1488 vfs->vfs_readonly = true; 1489 1490 error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs); 1491 if (error) { 1492 zfsvfs_vfs_free(vfs); 1493 goto out; 1494 } 1495 1496 if ((error = dsl_prop_get_integer(osname, "recordsize", 1497 &recordsize, NULL))) { 1498 zfsvfs_vfs_free(vfs); 1499 goto out; 1500 } 1501 1502 vfs->vfs_data = zfsvfs; 1503 zfsvfs->z_vfs = vfs; 1504 zfsvfs->z_sb = sb; 1505 sb->s_fs_info = zfsvfs; 1506 sb->s_magic = ZFS_SUPER_MAGIC; 1507 sb->s_maxbytes = MAX_LFS_FILESIZE; 1508 sb->s_time_gran = 1; 1509 sb->s_blocksize = recordsize; 1510 sb->s_blocksize_bits = ilog2(recordsize); 1511 1512 error = -zpl_bdi_setup(sb, "zfs"); 1513 if (error) 1514 goto out; 1515 1516 sb->s_bdi->ra_pages = 0; 1517 1518 /* Set callback operations for the file system. */ 1519 sb->s_op = &zpl_super_operations; 1520 sb->s_xattr = zpl_xattr_handlers; 1521 sb->s_export_op = &zpl_export_operations; 1522 sb->s_d_op = &zpl_dentry_operations; 1523 1524 /* Set features for file system. */ 1525 zfs_set_fuid_feature(zfsvfs); 1526 1527 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1528 uint64_t pval; 1529 1530 atime_changed_cb(zfsvfs, B_FALSE); 1531 readonly_changed_cb(zfsvfs, B_TRUE); 1532 if ((error = dsl_prop_get_integer(osname, 1533 "xattr", &pval, NULL))) 1534 goto out; 1535 xattr_changed_cb(zfsvfs, pval); 1536 if ((error = dsl_prop_get_integer(osname, 1537 "acltype", &pval, NULL))) 1538 goto out; 1539 acltype_changed_cb(zfsvfs, pval); 1540 zfsvfs->z_issnap = B_TRUE; 1541 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1542 zfsvfs->z_snap_defer_time = jiffies; 1543 1544 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1545 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1546 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1547 } else { 1548 if ((error = zfsvfs_setup(zfsvfs, B_TRUE))) 1549 goto out; 1550 } 1551 1552 /* Allocate a root inode for the filesystem. */ 1553 error = zfs_root(zfsvfs, &root_inode); 1554 if (error) { 1555 (void) zfs_umount(sb); 1556 goto out; 1557 } 1558 1559 /* Allocate a root dentry for the filesystem */ 1560 sb->s_root = d_make_root(root_inode); 1561 if (sb->s_root == NULL) { 1562 (void) zfs_umount(sb); 1563 error = SET_ERROR(ENOMEM); 1564 goto out; 1565 } 1566 1567 if (!zfsvfs->z_issnap) 1568 zfsctl_create(zfsvfs); 1569 1570 zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb); 1571 out: 1572 if (error) { 1573 if (zfsvfs != NULL) { 1574 dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs); 1575 zfsvfs_free(zfsvfs); 1576 } 1577 /* 1578 * make sure we don't have dangling sb->s_fs_info which 1579 * zfs_preumount will use. 1580 */ 1581 sb->s_fs_info = NULL; 1582 } 1583 1584 return (error); 1585 } 1586 1587 /* 1588 * Called when an unmount is requested and certain sanity checks have 1589 * already passed. At this point no dentries or inodes have been reclaimed 1590 * from their respective caches. We drop the extra reference on the .zfs 1591 * control directory to allow everything to be reclaimed. All snapshots 1592 * must already have been unmounted to reach this point. 1593 */ 1594 void 1595 zfs_preumount(struct super_block *sb) 1596 { 1597 zfsvfs_t *zfsvfs = sb->s_fs_info; 1598 1599 /* zfsvfs is NULL when zfs_domount fails during mount */ 1600 if (zfsvfs) { 1601 zfs_unlinked_drain_stop_wait(zfsvfs); 1602 zfsctl_destroy(sb->s_fs_info); 1603 /* 1604 * Wait for zrele_async before entering evict_inodes in 1605 * generic_shutdown_super. The reason we must finish before 1606 * evict_inodes is when lazytime is on, or when zfs_purgedir 1607 * calls zfs_zget, zrele would bump i_count from 0 to 1. This 1608 * would race with the i_count check in evict_inodes. This means 1609 * it could destroy the inode while we are still using it. 1610 * 1611 * We wait for two passes. xattr directories in the first pass 1612 * may add xattr entries in zfs_purgedir, so in the second pass 1613 * we wait for them. We don't use taskq_wait here because it is 1614 * a pool wide taskq. Other mounted filesystems can constantly 1615 * do zrele_async and there's no guarantee when taskq will be 1616 * empty. 1617 */ 1618 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1619 dmu_objset_pool(zfsvfs->z_os)), 0); 1620 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1621 dmu_objset_pool(zfsvfs->z_os)), 0); 1622 } 1623 } 1624 1625 /* 1626 * Called once all other unmount released tear down has occurred. 1627 * It is our responsibility to release any remaining infrastructure. 1628 */ 1629 int 1630 zfs_umount(struct super_block *sb) 1631 { 1632 zfsvfs_t *zfsvfs = sb->s_fs_info; 1633 objset_t *os; 1634 1635 if (zfsvfs->z_arc_prune != NULL) 1636 arc_remove_prune_callback(zfsvfs->z_arc_prune); 1637 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0); 1638 os = zfsvfs->z_os; 1639 zpl_bdi_destroy(sb); 1640 1641 /* 1642 * z_os will be NULL if there was an error in 1643 * attempting to reopen zfsvfs. 1644 */ 1645 if (os != NULL) { 1646 /* 1647 * Unset the objset user_ptr. 1648 */ 1649 mutex_enter(&os->os_user_ptr_lock); 1650 dmu_objset_set_user(os, NULL); 1651 mutex_exit(&os->os_user_ptr_lock); 1652 1653 /* 1654 * Finally release the objset 1655 */ 1656 dmu_objset_disown(os, B_TRUE, zfsvfs); 1657 } 1658 1659 zfsvfs_free(zfsvfs); 1660 return (0); 1661 } 1662 1663 int 1664 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm) 1665 { 1666 zfsvfs_t *zfsvfs = sb->s_fs_info; 1667 vfs_t *vfsp; 1668 boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os); 1669 int error; 1670 1671 if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) && 1672 !(*flags & SB_RDONLY)) { 1673 *flags |= SB_RDONLY; 1674 return (EROFS); 1675 } 1676 1677 error = zfsvfs_parse_options(zm->mnt_data, &vfsp); 1678 if (error) 1679 return (error); 1680 1681 if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY)) 1682 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1683 1684 zfs_unregister_callbacks(zfsvfs); 1685 zfsvfs_vfs_free(zfsvfs->z_vfs); 1686 1687 vfsp->vfs_data = zfsvfs; 1688 zfsvfs->z_vfs = vfsp; 1689 if (!issnap) 1690 (void) zfs_register_callbacks(vfsp); 1691 1692 return (error); 1693 } 1694 1695 int 1696 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp) 1697 { 1698 zfsvfs_t *zfsvfs = sb->s_fs_info; 1699 znode_t *zp; 1700 uint64_t object = 0; 1701 uint64_t fid_gen = 0; 1702 uint64_t gen_mask; 1703 uint64_t zp_gen; 1704 int i, err; 1705 1706 *ipp = NULL; 1707 1708 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 1709 zfid_short_t *zfid = (zfid_short_t *)fidp; 1710 1711 for (i = 0; i < sizeof (zfid->zf_object); i++) 1712 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 1713 1714 for (i = 0; i < sizeof (zfid->zf_gen); i++) 1715 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 1716 } else { 1717 return (SET_ERROR(EINVAL)); 1718 } 1719 1720 /* LONG_FID_LEN means snapdirs */ 1721 if (fidp->fid_len == LONG_FID_LEN) { 1722 zfid_long_t *zlfid = (zfid_long_t *)fidp; 1723 uint64_t objsetid = 0; 1724 uint64_t setgen = 0; 1725 1726 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 1727 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 1728 1729 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 1730 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 1731 1732 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) { 1733 dprintf("snapdir fid: objsetid (%llu) != " 1734 "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n", 1735 objsetid, ZFSCTL_INO_SNAPDIRS, object); 1736 1737 return (SET_ERROR(EINVAL)); 1738 } 1739 1740 if (fid_gen > 1 || setgen != 0) { 1741 dprintf("snapdir fid: fid_gen (%llu) and setgen " 1742 "(%llu)\n", fid_gen, setgen); 1743 return (SET_ERROR(EINVAL)); 1744 } 1745 1746 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp)); 1747 } 1748 1749 ZFS_ENTER(zfsvfs); 1750 /* A zero fid_gen means we are in the .zfs control directories */ 1751 if (fid_gen == 0 && 1752 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 1753 *ipp = zfsvfs->z_ctldir; 1754 ASSERT(*ipp != NULL); 1755 if (object == ZFSCTL_INO_SNAPDIR) { 1756 VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp, 1757 0, kcred, NULL, NULL) == 0); 1758 } else { 1759 /* 1760 * Must have an existing ref, so igrab() 1761 * cannot return NULL 1762 */ 1763 VERIFY3P(igrab(*ipp), !=, NULL); 1764 } 1765 ZFS_EXIT(zfsvfs); 1766 return (0); 1767 } 1768 1769 gen_mask = -1ULL >> (64 - 8 * i); 1770 1771 dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask); 1772 if ((err = zfs_zget(zfsvfs, object, &zp))) { 1773 ZFS_EXIT(zfsvfs); 1774 return (err); 1775 } 1776 1777 /* Don't export xattr stuff */ 1778 if (zp->z_pflags & ZFS_XATTR) { 1779 zrele(zp); 1780 ZFS_EXIT(zfsvfs); 1781 return (SET_ERROR(ENOENT)); 1782 } 1783 1784 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 1785 sizeof (uint64_t)); 1786 zp_gen = zp_gen & gen_mask; 1787 if (zp_gen == 0) 1788 zp_gen = 1; 1789 if ((fid_gen == 0) && (zfsvfs->z_root == object)) 1790 fid_gen = zp_gen; 1791 if (zp->z_unlinked || zp_gen != fid_gen) { 1792 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen, 1793 fid_gen); 1794 zrele(zp); 1795 ZFS_EXIT(zfsvfs); 1796 return (SET_ERROR(ENOENT)); 1797 } 1798 1799 *ipp = ZTOI(zp); 1800 if (*ipp) 1801 zfs_znode_update_vfs(ITOZ(*ipp)); 1802 1803 ZFS_EXIT(zfsvfs); 1804 return (0); 1805 } 1806 1807 /* 1808 * Block out VFS ops and close zfsvfs_t 1809 * 1810 * Note, if successful, then we return with the 'z_teardown_lock' and 1811 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 1812 * dataset and objset intact so that they can be atomically handed off during 1813 * a subsequent rollback or recv operation and the resume thereafter. 1814 */ 1815 int 1816 zfs_suspend_fs(zfsvfs_t *zfsvfs) 1817 { 1818 int error; 1819 1820 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 1821 return (error); 1822 1823 return (0); 1824 } 1825 1826 /* 1827 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 1828 * is an invariant across any of the operations that can be performed while the 1829 * filesystem was suspended. Whether it succeeded or failed, the preconditions 1830 * are the same: the relevant objset and associated dataset are owned by 1831 * zfsvfs, held, and long held on entry. 1832 */ 1833 int 1834 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) 1835 { 1836 int err, err2; 1837 znode_t *zp; 1838 1839 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs)); 1840 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1841 1842 /* 1843 * We already own this, so just update the objset_t, as the one we 1844 * had before may have been evicted. 1845 */ 1846 objset_t *os; 1847 VERIFY3P(ds->ds_owner, ==, zfsvfs); 1848 VERIFY(dsl_dataset_long_held(ds)); 1849 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds)); 1850 dsl_pool_config_enter(dp, FTAG); 1851 VERIFY0(dmu_objset_from_ds(ds, &os)); 1852 dsl_pool_config_exit(dp, FTAG); 1853 1854 err = zfsvfs_init(zfsvfs, os); 1855 if (err != 0) 1856 goto bail; 1857 1858 ds->ds_dir->dd_activity_cancelled = B_FALSE; 1859 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0); 1860 1861 zfs_set_fuid_feature(zfsvfs); 1862 zfsvfs->z_rollback_time = jiffies; 1863 1864 /* 1865 * Attempt to re-establish all the active inodes with their 1866 * dbufs. If a zfs_rezget() fails, then we unhash the inode 1867 * and mark it stale. This prevents a collision if a new 1868 * inode/object is created which must use the same inode 1869 * number. The stale inode will be be released when the 1870 * VFS prunes the dentry holding the remaining references 1871 * on the stale inode. 1872 */ 1873 mutex_enter(&zfsvfs->z_znodes_lock); 1874 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 1875 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1876 err2 = zfs_rezget(zp); 1877 if (err2) { 1878 remove_inode_hash(ZTOI(zp)); 1879 zp->z_is_stale = B_TRUE; 1880 } 1881 1882 /* see comment in zfs_suspend_fs() */ 1883 if (zp->z_suspended) { 1884 zfs_zrele_async(zp); 1885 zp->z_suspended = B_FALSE; 1886 } 1887 } 1888 mutex_exit(&zfsvfs->z_znodes_lock); 1889 1890 if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) { 1891 /* 1892 * zfs_suspend_fs() could have interrupted freeing 1893 * of dnodes. We need to restart this freeing so 1894 * that we don't "leak" the space. 1895 */ 1896 zfs_unlinked_drain(zfsvfs); 1897 } 1898 1899 /* 1900 * Most of the time zfs_suspend_fs is used for changing the contents 1901 * of the underlying dataset. ZFS rollback and receive operations 1902 * might create files for which negative dentries are present in 1903 * the cache. Since walking the dcache would require a lot of GPL-only 1904 * code duplication, it's much easier on these rather rare occasions 1905 * just to flush the whole dcache for the given dataset/filesystem. 1906 */ 1907 shrink_dcache_sb(zfsvfs->z_sb); 1908 1909 bail: 1910 if (err != 0) 1911 zfsvfs->z_unmounted = B_TRUE; 1912 1913 /* release the VFS ops */ 1914 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1915 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1916 1917 if (err != 0) { 1918 /* 1919 * Since we couldn't setup the sa framework, try to force 1920 * unmount this file system. 1921 */ 1922 if (zfsvfs->z_os) 1923 (void) zfs_umount(zfsvfs->z_sb); 1924 } 1925 return (err); 1926 } 1927 1928 /* 1929 * Release VOPs and unmount a suspended filesystem. 1930 */ 1931 int 1932 zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) 1933 { 1934 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs)); 1935 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1936 1937 /* 1938 * We already own this, so just hold and rele it to update the 1939 * objset_t, as the one we had before may have been evicted. 1940 */ 1941 objset_t *os; 1942 VERIFY3P(ds->ds_owner, ==, zfsvfs); 1943 VERIFY(dsl_dataset_long_held(ds)); 1944 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds)); 1945 dsl_pool_config_enter(dp, FTAG); 1946 VERIFY0(dmu_objset_from_ds(ds, &os)); 1947 dsl_pool_config_exit(dp, FTAG); 1948 zfsvfs->z_os = os; 1949 1950 /* release the VOPs */ 1951 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1952 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1953 1954 /* 1955 * Try to force unmount this file system. 1956 */ 1957 (void) zfs_umount(zfsvfs->z_sb); 1958 zfsvfs->z_unmounted = B_TRUE; 1959 return (0); 1960 } 1961 1962 /* 1963 * Automounted snapshots rely on periodic revalidation 1964 * to defer snapshots from being automatically unmounted. 1965 */ 1966 1967 inline void 1968 zfs_exit_fs(zfsvfs_t *zfsvfs) 1969 { 1970 if (!zfsvfs->z_issnap) 1971 return; 1972 1973 if (time_after(jiffies, zfsvfs->z_snap_defer_time + 1974 MAX(zfs_expire_snapshot * HZ / 2, HZ))) { 1975 zfsvfs->z_snap_defer_time = jiffies; 1976 zfsctl_snapshot_unmount_delay(zfsvfs->z_os->os_spa, 1977 dmu_objset_id(zfsvfs->z_os), 1978 zfs_expire_snapshot); 1979 } 1980 } 1981 1982 int 1983 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 1984 { 1985 int error; 1986 objset_t *os = zfsvfs->z_os; 1987 dmu_tx_t *tx; 1988 1989 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 1990 return (SET_ERROR(EINVAL)); 1991 1992 if (newvers < zfsvfs->z_version) 1993 return (SET_ERROR(EINVAL)); 1994 1995 if (zfs_spa_version_map(newvers) > 1996 spa_version(dmu_objset_spa(zfsvfs->z_os))) 1997 return (SET_ERROR(ENOTSUP)); 1998 1999 tx = dmu_tx_create(os); 2000 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2001 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2002 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2003 ZFS_SA_ATTRS); 2004 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2005 } 2006 error = dmu_tx_assign(tx, TXG_WAIT); 2007 if (error) { 2008 dmu_tx_abort(tx); 2009 return (error); 2010 } 2011 2012 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2013 8, 1, &newvers, tx); 2014 2015 if (error) { 2016 dmu_tx_commit(tx); 2017 return (error); 2018 } 2019 2020 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2021 uint64_t sa_obj; 2022 2023 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2024 SPA_VERSION_SA); 2025 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2026 DMU_OT_NONE, 0, tx); 2027 2028 error = zap_add(os, MASTER_NODE_OBJ, 2029 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2030 ASSERT0(error); 2031 2032 VERIFY(0 == sa_set_sa_object(os, sa_obj)); 2033 sa_register_update_callback(os, zfs_sa_upgrade); 2034 } 2035 2036 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2037 "from %llu to %llu", zfsvfs->z_version, newvers); 2038 2039 dmu_tx_commit(tx); 2040 2041 zfsvfs->z_version = newvers; 2042 os->os_version = newvers; 2043 2044 zfs_set_fuid_feature(zfsvfs); 2045 2046 return (0); 2047 } 2048 2049 /* 2050 * Read a property stored within the master node. 2051 */ 2052 int 2053 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) 2054 { 2055 uint64_t *cached_copy = NULL; 2056 2057 /* 2058 * Figure out where in the objset_t the cached copy would live, if it 2059 * is available for the requested property. 2060 */ 2061 if (os != NULL) { 2062 switch (prop) { 2063 case ZFS_PROP_VERSION: 2064 cached_copy = &os->os_version; 2065 break; 2066 case ZFS_PROP_NORMALIZE: 2067 cached_copy = &os->os_normalization; 2068 break; 2069 case ZFS_PROP_UTF8ONLY: 2070 cached_copy = &os->os_utf8only; 2071 break; 2072 case ZFS_PROP_CASE: 2073 cached_copy = &os->os_casesensitivity; 2074 break; 2075 default: 2076 break; 2077 } 2078 } 2079 if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) { 2080 *value = *cached_copy; 2081 return (0); 2082 } 2083 2084 /* 2085 * If the property wasn't cached, look up the file system's value for 2086 * the property. For the version property, we look up a slightly 2087 * different string. 2088 */ 2089 const char *pname; 2090 int error = ENOENT; 2091 if (prop == ZFS_PROP_VERSION) 2092 pname = ZPL_VERSION_STR; 2093 else 2094 pname = zfs_prop_to_name(prop); 2095 2096 if (os != NULL) { 2097 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS); 2098 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); 2099 } 2100 2101 if (error == ENOENT) { 2102 /* No value set, use the default value */ 2103 switch (prop) { 2104 case ZFS_PROP_VERSION: 2105 *value = ZPL_VERSION; 2106 break; 2107 case ZFS_PROP_NORMALIZE: 2108 case ZFS_PROP_UTF8ONLY: 2109 *value = 0; 2110 break; 2111 case ZFS_PROP_CASE: 2112 *value = ZFS_CASE_SENSITIVE; 2113 break; 2114 case ZFS_PROP_ACLTYPE: 2115 *value = ZFS_ACLTYPE_OFF; 2116 break; 2117 default: 2118 return (error); 2119 } 2120 error = 0; 2121 } 2122 2123 /* 2124 * If one of the methods for getting the property value above worked, 2125 * copy it into the objset_t's cache. 2126 */ 2127 if (error == 0 && cached_copy != NULL) { 2128 *cached_copy = *value; 2129 } 2130 2131 return (error); 2132 } 2133 2134 /* 2135 * Return true if the corresponding vfs's unmounted flag is set. 2136 * Otherwise return false. 2137 * If this function returns true we know VFS unmount has been initiated. 2138 */ 2139 boolean_t 2140 zfs_get_vfs_flag_unmounted(objset_t *os) 2141 { 2142 zfsvfs_t *zfvp; 2143 boolean_t unmounted = B_FALSE; 2144 2145 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS); 2146 2147 mutex_enter(&os->os_user_ptr_lock); 2148 zfvp = dmu_objset_get_user(os); 2149 if (zfvp != NULL && zfvp->z_unmounted) 2150 unmounted = B_TRUE; 2151 mutex_exit(&os->os_user_ptr_lock); 2152 2153 return (unmounted); 2154 } 2155 2156 void 2157 zfsvfs_update_fromname(const char *oldname, const char *newname) 2158 { 2159 /* 2160 * We don't need to do anything here, the devname is always current by 2161 * virtue of zfsvfs->z_sb->s_op->show_devname. 2162 */ 2163 (void) oldname, (void) newname; 2164 } 2165 2166 void 2167 zfs_init(void) 2168 { 2169 zfsctl_init(); 2170 zfs_znode_init(); 2171 dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info); 2172 register_filesystem(&zpl_fs_type); 2173 } 2174 2175 void 2176 zfs_fini(void) 2177 { 2178 /* 2179 * we don't use outstanding because zpl_posix_acl_free might add more. 2180 */ 2181 taskq_wait(system_delay_taskq); 2182 taskq_wait(system_taskq); 2183 unregister_filesystem(&zpl_fs_type); 2184 zfs_znode_fini(); 2185 zfsctl_fini(); 2186 } 2187 2188 #if defined(_KERNEL) 2189 EXPORT_SYMBOL(zfs_suspend_fs); 2190 EXPORT_SYMBOL(zfs_resume_fs); 2191 EXPORT_SYMBOL(zfs_set_version); 2192 EXPORT_SYMBOL(zfsvfs_create); 2193 EXPORT_SYMBOL(zfsvfs_free); 2194 EXPORT_SYMBOL(zfs_is_readonly); 2195 EXPORT_SYMBOL(zfs_domount); 2196 EXPORT_SYMBOL(zfs_preumount); 2197 EXPORT_SYMBOL(zfs_umount); 2198 EXPORT_SYMBOL(zfs_remount); 2199 EXPORT_SYMBOL(zfs_statvfs); 2200 EXPORT_SYMBOL(zfs_vget); 2201 EXPORT_SYMBOL(zfs_prune); 2202 #endif 2203