1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * CDDL HEADER START 4 * 5 * The contents of this file are subject to the terms of the 6 * Common Development and Distribution License (the "License"). 7 * You may not use this file except in compliance with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or https://opensource.org/licenses/CDDL-1.0. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2012, 2018 by Delphix. All rights reserved. 25 */ 26 27 /* Portions Copyright 2010 Robert Milkowski */ 28 29 #include <sys/types.h> 30 #include <sys/param.h> 31 #include <sys/sysmacros.h> 32 #include <sys/kmem.h> 33 #include <sys/pathname.h> 34 #include <sys/vnode.h> 35 #include <sys/vfs.h> 36 #include <sys/mntent.h> 37 #include <sys/cmn_err.h> 38 #include <sys/zfs_znode.h> 39 #include <sys/zfs_vnops.h> 40 #include <sys/zfs_dir.h> 41 #include <sys/zil.h> 42 #include <sys/fs/zfs.h> 43 #include <sys/dmu.h> 44 #include <sys/dsl_prop.h> 45 #include <sys/dsl_dataset.h> 46 #include <sys/dsl_deleg.h> 47 #include <sys/spa.h> 48 #include <sys/zap.h> 49 #include <sys/sa.h> 50 #include <sys/sa_impl.h> 51 #include <sys/policy.h> 52 #include <sys/atomic.h> 53 #include <sys/zfs_ioctl.h> 54 #include <sys/zfs_ctldir.h> 55 #include <sys/zfs_fuid.h> 56 #include <sys/zfs_quota.h> 57 #include <sys/sunddi.h> 58 #include <sys/dmu_objset.h> 59 #include <sys/dsl_dir.h> 60 #include <sys/objlist.h> 61 #include <sys/zfeature.h> 62 #include <sys/zpl.h> 63 #include <linux/vfs_compat.h> 64 #include <linux/fs.h> 65 #include "zfs_comutil.h" 66 67 enum { 68 TOKEN_RO, 69 TOKEN_RW, 70 TOKEN_SETUID, 71 TOKEN_NOSETUID, 72 TOKEN_EXEC, 73 TOKEN_NOEXEC, 74 TOKEN_DEVICES, 75 TOKEN_NODEVICES, 76 TOKEN_DIRXATTR, 77 TOKEN_SAXATTR, 78 TOKEN_XATTR, 79 TOKEN_NOXATTR, 80 TOKEN_ATIME, 81 TOKEN_NOATIME, 82 TOKEN_RELATIME, 83 TOKEN_NORELATIME, 84 TOKEN_NBMAND, 85 TOKEN_NONBMAND, 86 TOKEN_MNTPOINT, 87 TOKEN_LAST, 88 }; 89 90 static const match_table_t zpl_tokens = { 91 { TOKEN_RO, MNTOPT_RO }, 92 { TOKEN_RW, MNTOPT_RW }, 93 { TOKEN_SETUID, MNTOPT_SETUID }, 94 { TOKEN_NOSETUID, MNTOPT_NOSETUID }, 95 { TOKEN_EXEC, MNTOPT_EXEC }, 96 { TOKEN_NOEXEC, MNTOPT_NOEXEC }, 97 { TOKEN_DEVICES, MNTOPT_DEVICES }, 98 { TOKEN_NODEVICES, MNTOPT_NODEVICES }, 99 { TOKEN_DIRXATTR, MNTOPT_DIRXATTR }, 100 { TOKEN_SAXATTR, MNTOPT_SAXATTR }, 101 { TOKEN_XATTR, MNTOPT_XATTR }, 102 { TOKEN_NOXATTR, MNTOPT_NOXATTR }, 103 { TOKEN_ATIME, MNTOPT_ATIME }, 104 { TOKEN_NOATIME, MNTOPT_NOATIME }, 105 { TOKEN_RELATIME, MNTOPT_RELATIME }, 106 { TOKEN_NORELATIME, MNTOPT_NORELATIME }, 107 { TOKEN_NBMAND, MNTOPT_NBMAND }, 108 { TOKEN_NONBMAND, MNTOPT_NONBMAND }, 109 { TOKEN_MNTPOINT, MNTOPT_MNTPOINT "=%s" }, 110 { TOKEN_LAST, NULL }, 111 }; 112 113 static void 114 zfsvfs_vfs_free(vfs_t *vfsp) 115 { 116 if (vfsp != NULL) { 117 if (vfsp->vfs_mntpoint != NULL) 118 kmem_strfree(vfsp->vfs_mntpoint); 119 mutex_destroy(&vfsp->vfs_mntpt_lock); 120 kmem_free(vfsp, sizeof (vfs_t)); 121 } 122 } 123 124 static int 125 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp) 126 { 127 switch (token) { 128 case TOKEN_RO: 129 vfsp->vfs_readonly = B_TRUE; 130 vfsp->vfs_do_readonly = B_TRUE; 131 break; 132 case TOKEN_RW: 133 vfsp->vfs_readonly = B_FALSE; 134 vfsp->vfs_do_readonly = B_TRUE; 135 break; 136 case TOKEN_SETUID: 137 vfsp->vfs_setuid = B_TRUE; 138 vfsp->vfs_do_setuid = B_TRUE; 139 break; 140 case TOKEN_NOSETUID: 141 vfsp->vfs_setuid = B_FALSE; 142 vfsp->vfs_do_setuid = B_TRUE; 143 break; 144 case TOKEN_EXEC: 145 vfsp->vfs_exec = B_TRUE; 146 vfsp->vfs_do_exec = B_TRUE; 147 break; 148 case TOKEN_NOEXEC: 149 vfsp->vfs_exec = B_FALSE; 150 vfsp->vfs_do_exec = B_TRUE; 151 break; 152 case TOKEN_DEVICES: 153 vfsp->vfs_devices = B_TRUE; 154 vfsp->vfs_do_devices = B_TRUE; 155 break; 156 case TOKEN_NODEVICES: 157 vfsp->vfs_devices = B_FALSE; 158 vfsp->vfs_do_devices = B_TRUE; 159 break; 160 case TOKEN_DIRXATTR: 161 vfsp->vfs_xattr = ZFS_XATTR_DIR; 162 vfsp->vfs_do_xattr = B_TRUE; 163 break; 164 case TOKEN_SAXATTR: 165 vfsp->vfs_xattr = ZFS_XATTR_SA; 166 vfsp->vfs_do_xattr = B_TRUE; 167 break; 168 case TOKEN_XATTR: 169 vfsp->vfs_xattr = ZFS_XATTR_SA; 170 vfsp->vfs_do_xattr = B_TRUE; 171 break; 172 case TOKEN_NOXATTR: 173 vfsp->vfs_xattr = ZFS_XATTR_OFF; 174 vfsp->vfs_do_xattr = B_TRUE; 175 break; 176 case TOKEN_ATIME: 177 vfsp->vfs_atime = B_TRUE; 178 vfsp->vfs_do_atime = B_TRUE; 179 break; 180 case TOKEN_NOATIME: 181 vfsp->vfs_atime = B_FALSE; 182 vfsp->vfs_do_atime = B_TRUE; 183 break; 184 case TOKEN_RELATIME: 185 vfsp->vfs_relatime = B_TRUE; 186 vfsp->vfs_do_relatime = B_TRUE; 187 break; 188 case TOKEN_NORELATIME: 189 vfsp->vfs_relatime = B_FALSE; 190 vfsp->vfs_do_relatime = B_TRUE; 191 break; 192 case TOKEN_NBMAND: 193 vfsp->vfs_nbmand = B_TRUE; 194 vfsp->vfs_do_nbmand = B_TRUE; 195 break; 196 case TOKEN_NONBMAND: 197 vfsp->vfs_nbmand = B_FALSE; 198 vfsp->vfs_do_nbmand = B_TRUE; 199 break; 200 case TOKEN_MNTPOINT: 201 if (vfsp->vfs_mntpoint != NULL) 202 kmem_strfree(vfsp->vfs_mntpoint); 203 vfsp->vfs_mntpoint = match_strdup(&args[0]); 204 if (vfsp->vfs_mntpoint == NULL) 205 return (SET_ERROR(ENOMEM)); 206 break; 207 default: 208 break; 209 } 210 211 return (0); 212 } 213 214 /* 215 * Parse the raw mntopts and return a vfs_t describing the options. 216 */ 217 static int 218 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp) 219 { 220 vfs_t *tmp_vfsp; 221 int error; 222 223 tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP); 224 mutex_init(&tmp_vfsp->vfs_mntpt_lock, NULL, MUTEX_DEFAULT, NULL); 225 226 if (mntopts != NULL) { 227 substring_t args[MAX_OPT_ARGS]; 228 char *tmp_mntopts, *p, *t; 229 int token; 230 231 tmp_mntopts = t = kmem_strdup(mntopts); 232 if (tmp_mntopts == NULL) 233 return (SET_ERROR(ENOMEM)); 234 235 while ((p = strsep(&t, ",")) != NULL) { 236 if (!*p) 237 continue; 238 239 args[0].to = args[0].from = NULL; 240 token = match_token(p, zpl_tokens, args); 241 error = zfsvfs_parse_option(p, token, args, tmp_vfsp); 242 if (error) { 243 kmem_strfree(tmp_mntopts); 244 zfsvfs_vfs_free(tmp_vfsp); 245 return (error); 246 } 247 } 248 249 kmem_strfree(tmp_mntopts); 250 } 251 252 *vfsp = tmp_vfsp; 253 254 return (0); 255 } 256 257 boolean_t 258 zfs_is_readonly(zfsvfs_t *zfsvfs) 259 { 260 return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY)); 261 } 262 263 int 264 zfs_sync(struct super_block *sb, int wait, cred_t *cr) 265 { 266 (void) cr; 267 zfsvfs_t *zfsvfs = sb->s_fs_info; 268 ASSERT3P(zfsvfs, !=, NULL); 269 270 /* 271 * Semantically, the only requirement is that the sync be initiated. 272 * The DMU syncs out txgs frequently, so there's nothing to do. 273 */ 274 if (!wait) 275 return (0); 276 277 int err = zfs_enter(zfsvfs, FTAG); 278 if (err != 0) 279 return (err); 280 281 /* 282 * Sync any pending writes, but do not block if the pool is suspended. 283 * This is to help with shutting down with pools suspended, as we don't 284 * want to block in that case. 285 */ 286 err = zil_commit_flags(zfsvfs->z_log, 0, ZIL_COMMIT_NOW); 287 zfs_exit(zfsvfs, FTAG); 288 289 return (err); 290 } 291 292 static void 293 atime_changed_cb(void *arg, uint64_t newval) 294 { 295 zfsvfs_t *zfsvfs = arg; 296 struct super_block *sb = zfsvfs->z_sb; 297 298 if (sb == NULL) 299 return; 300 /* 301 * Update SB_NOATIME bit in VFS super block. Since atime update is 302 * determined by atime_needs_update(), atime_needs_update() needs to 303 * return false if atime is turned off, and not unconditionally return 304 * false if atime is turned on. 305 */ 306 if (newval) 307 sb->s_flags &= ~SB_NOATIME; 308 else 309 sb->s_flags |= SB_NOATIME; 310 } 311 312 static void 313 relatime_changed_cb(void *arg, uint64_t newval) 314 { 315 ((zfsvfs_t *)arg)->z_relatime = newval; 316 } 317 318 static void 319 xattr_changed_cb(void *arg, uint64_t newval) 320 { 321 zfsvfs_t *zfsvfs = arg; 322 323 if (newval == ZFS_XATTR_OFF) { 324 zfsvfs->z_flags &= ~ZSB_XATTR; 325 } else { 326 zfsvfs->z_flags |= ZSB_XATTR; 327 328 if (newval == ZFS_XATTR_SA) 329 zfsvfs->z_xattr_sa = B_TRUE; 330 else 331 zfsvfs->z_xattr_sa = B_FALSE; 332 } 333 } 334 335 static void 336 acltype_changed_cb(void *arg, uint64_t newval) 337 { 338 zfsvfs_t *zfsvfs = arg; 339 340 switch (newval) { 341 case ZFS_ACLTYPE_NFSV4: 342 case ZFS_ACLTYPE_OFF: 343 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF; 344 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL; 345 break; 346 case ZFS_ACLTYPE_POSIX: 347 #ifdef CONFIG_FS_POSIX_ACL 348 zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIX; 349 zfsvfs->z_sb->s_flags |= SB_POSIXACL; 350 #else 351 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF; 352 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL; 353 #endif /* CONFIG_FS_POSIX_ACL */ 354 break; 355 default: 356 break; 357 } 358 } 359 360 static void 361 blksz_changed_cb(void *arg, uint64_t newval) 362 { 363 zfsvfs_t *zfsvfs = arg; 364 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os))); 365 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); 366 ASSERT(ISP2(newval)); 367 368 zfsvfs->z_max_blksz = newval; 369 } 370 371 static void 372 readonly_changed_cb(void *arg, uint64_t newval) 373 { 374 zfsvfs_t *zfsvfs = arg; 375 struct super_block *sb = zfsvfs->z_sb; 376 377 if (sb == NULL) 378 return; 379 380 if (newval) 381 sb->s_flags |= SB_RDONLY; 382 else 383 sb->s_flags &= ~SB_RDONLY; 384 } 385 386 static void 387 devices_changed_cb(void *arg, uint64_t newval) 388 { 389 } 390 391 static void 392 setuid_changed_cb(void *arg, uint64_t newval) 393 { 394 } 395 396 static void 397 exec_changed_cb(void *arg, uint64_t newval) 398 { 399 } 400 401 static void 402 nbmand_changed_cb(void *arg, uint64_t newval) 403 { 404 zfsvfs_t *zfsvfs = arg; 405 struct super_block *sb = zfsvfs->z_sb; 406 407 if (sb == NULL) 408 return; 409 410 if (newval == TRUE) 411 sb->s_flags |= SB_MANDLOCK; 412 else 413 sb->s_flags &= ~SB_MANDLOCK; 414 } 415 416 static void 417 snapdir_changed_cb(void *arg, uint64_t newval) 418 { 419 ((zfsvfs_t *)arg)->z_show_ctldir = newval; 420 } 421 422 static void 423 acl_mode_changed_cb(void *arg, uint64_t newval) 424 { 425 zfsvfs_t *zfsvfs = arg; 426 427 zfsvfs->z_acl_mode = newval; 428 } 429 430 static void 431 acl_inherit_changed_cb(void *arg, uint64_t newval) 432 { 433 ((zfsvfs_t *)arg)->z_acl_inherit = newval; 434 } 435 436 static void 437 longname_changed_cb(void *arg, uint64_t newval) 438 { 439 ((zfsvfs_t *)arg)->z_longname = newval; 440 } 441 442 static int 443 zfs_register_callbacks(vfs_t *vfsp) 444 { 445 struct dsl_dataset *ds = NULL; 446 objset_t *os = NULL; 447 zfsvfs_t *zfsvfs = NULL; 448 int error = 0; 449 450 ASSERT(vfsp); 451 zfsvfs = vfsp->vfs_data; 452 ASSERT(zfsvfs); 453 os = zfsvfs->z_os; 454 455 /* 456 * The act of registering our callbacks will destroy any mount 457 * options we may have. In order to enable temporary overrides 458 * of mount options, we stash away the current values and 459 * restore them after we register the callbacks. 460 */ 461 if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) { 462 vfsp->vfs_do_readonly = B_TRUE; 463 vfsp->vfs_readonly = B_TRUE; 464 } 465 466 /* 467 * Register property callbacks. 468 * 469 * It would probably be fine to just check for i/o error from 470 * the first prop_register(), but I guess I like to go 471 * overboard... 472 */ 473 ds = dmu_objset_ds(os); 474 dsl_pool_config_enter(dmu_objset_pool(os), FTAG); 475 error = dsl_prop_register(ds, 476 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); 477 error = error ? error : dsl_prop_register(ds, 478 zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs); 479 error = error ? error : dsl_prop_register(ds, 480 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); 481 error = error ? error : dsl_prop_register(ds, 482 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); 483 error = error ? error : dsl_prop_register(ds, 484 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); 485 error = error ? error : dsl_prop_register(ds, 486 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs); 487 error = error ? error : dsl_prop_register(ds, 488 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); 489 error = error ? error : dsl_prop_register(ds, 490 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); 491 error = error ? error : dsl_prop_register(ds, 492 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); 493 error = error ? error : dsl_prop_register(ds, 494 zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs); 495 error = error ? error : dsl_prop_register(ds, 496 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); 497 error = error ? error : dsl_prop_register(ds, 498 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, 499 zfsvfs); 500 error = error ? error : dsl_prop_register(ds, 501 zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs); 502 error = error ? error : dsl_prop_register(ds, 503 zfs_prop_to_name(ZFS_PROP_LONGNAME), longname_changed_cb, zfsvfs); 504 dsl_pool_config_exit(dmu_objset_pool(os), FTAG); 505 if (error) 506 goto unregister; 507 508 /* 509 * Invoke our callbacks to restore temporary mount options. 510 */ 511 if (vfsp->vfs_do_readonly) 512 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly); 513 if (vfsp->vfs_do_setuid) 514 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid); 515 if (vfsp->vfs_do_exec) 516 exec_changed_cb(zfsvfs, vfsp->vfs_exec); 517 if (vfsp->vfs_do_devices) 518 devices_changed_cb(zfsvfs, vfsp->vfs_devices); 519 if (vfsp->vfs_do_xattr) 520 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr); 521 if (vfsp->vfs_do_atime) 522 atime_changed_cb(zfsvfs, vfsp->vfs_atime); 523 if (vfsp->vfs_do_relatime) 524 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime); 525 if (vfsp->vfs_do_nbmand) 526 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand); 527 528 return (0); 529 530 unregister: 531 dsl_prop_unregister_all(ds, zfsvfs); 532 return (error); 533 } 534 535 /* 536 * Takes a dataset, a property, a value and that value's setpoint as 537 * found in the ZAP. Checks if the property has been changed in the vfs. 538 * If so, val and setpoint will be overwritten with updated content. 539 * Otherwise, they are left unchanged. 540 */ 541 int 542 zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val, 543 char *setpoint) 544 { 545 int error; 546 zfsvfs_t *zfvp; 547 vfs_t *vfsp; 548 objset_t *os; 549 uint64_t tmp = *val; 550 551 error = dmu_objset_from_ds(ds, &os); 552 if (error != 0) 553 return (error); 554 555 if (dmu_objset_type(os) != DMU_OST_ZFS) 556 return (EINVAL); 557 558 mutex_enter(&os->os_user_ptr_lock); 559 zfvp = dmu_objset_get_user(os); 560 mutex_exit(&os->os_user_ptr_lock); 561 if (zfvp == NULL) 562 return (ESRCH); 563 564 vfsp = zfvp->z_vfs; 565 566 switch (zfs_prop) { 567 case ZFS_PROP_ATIME: 568 if (vfsp->vfs_do_atime) 569 tmp = vfsp->vfs_atime; 570 break; 571 case ZFS_PROP_RELATIME: 572 if (vfsp->vfs_do_relatime) 573 tmp = vfsp->vfs_relatime; 574 break; 575 case ZFS_PROP_DEVICES: 576 if (vfsp->vfs_do_devices) 577 tmp = vfsp->vfs_devices; 578 break; 579 case ZFS_PROP_EXEC: 580 if (vfsp->vfs_do_exec) 581 tmp = vfsp->vfs_exec; 582 break; 583 case ZFS_PROP_SETUID: 584 if (vfsp->vfs_do_setuid) 585 tmp = vfsp->vfs_setuid; 586 break; 587 case ZFS_PROP_READONLY: 588 if (vfsp->vfs_do_readonly) 589 tmp = vfsp->vfs_readonly; 590 break; 591 case ZFS_PROP_XATTR: 592 if (vfsp->vfs_do_xattr) 593 tmp = vfsp->vfs_xattr; 594 break; 595 case ZFS_PROP_NBMAND: 596 if (vfsp->vfs_do_nbmand) 597 tmp = vfsp->vfs_nbmand; 598 break; 599 default: 600 return (ENOENT); 601 } 602 603 if (tmp != *val) { 604 if (setpoint) 605 (void) strcpy(setpoint, "temporary"); 606 *val = tmp; 607 } 608 return (0); 609 } 610 611 /* 612 * Associate this zfsvfs with the given objset, which must be owned. 613 * This will cache a bunch of on-disk state from the objset in the 614 * zfsvfs. 615 */ 616 static int 617 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os) 618 { 619 int error; 620 uint64_t val; 621 622 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; 623 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; 624 zfsvfs->z_os = os; 625 626 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); 627 if (error != 0) 628 return (error); 629 if (zfsvfs->z_version > 630 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { 631 (void) printk("Can't mount a version %lld file system " 632 "on a version %lld pool\n. Pool must be upgraded to mount " 633 "this file system.\n", (u_longlong_t)zfsvfs->z_version, 634 (u_longlong_t)spa_version(dmu_objset_spa(os))); 635 return (SET_ERROR(ENOTSUP)); 636 } 637 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val); 638 if (error != 0) 639 return (error); 640 zfsvfs->z_norm = (int)val; 641 642 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val); 643 if (error != 0) 644 return (error); 645 zfsvfs->z_utf8 = (val != 0); 646 647 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val); 648 if (error != 0) 649 return (error); 650 zfsvfs->z_case = (uint_t)val; 651 652 if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0) 653 return (error); 654 zfsvfs->z_acl_type = (uint_t)val; 655 656 /* 657 * Fold case on file systems that are always or sometimes case 658 * insensitive. 659 */ 660 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 661 zfsvfs->z_case == ZFS_CASE_MIXED) 662 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 663 664 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 665 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 666 667 uint64_t sa_obj = 0; 668 if (zfsvfs->z_use_sa) { 669 /* should either have both of these objects or none */ 670 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, 671 &sa_obj); 672 if (error != 0) 673 return (error); 674 675 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val); 676 if ((error == 0) && (val == ZFS_XATTR_SA)) 677 zfsvfs->z_xattr_sa = B_TRUE; 678 } 679 680 error = zfs_get_zplprop(os, ZFS_PROP_DEFAULTUSERQUOTA, 681 &zfsvfs->z_defaultuserquota); 682 if (error != 0) 683 return (error); 684 685 error = zfs_get_zplprop(os, ZFS_PROP_DEFAULTGROUPQUOTA, 686 &zfsvfs->z_defaultgroupquota); 687 if (error != 0) 688 return (error); 689 690 error = zfs_get_zplprop(os, ZFS_PROP_DEFAULTPROJECTQUOTA, 691 &zfsvfs->z_defaultprojectquota); 692 if (error != 0) 693 return (error); 694 695 error = zfs_get_zplprop(os, ZFS_PROP_DEFAULTUSEROBJQUOTA, 696 &zfsvfs->z_defaultuserobjquota); 697 if (error != 0) 698 return (error); 699 700 error = zfs_get_zplprop(os, ZFS_PROP_DEFAULTGROUPOBJQUOTA, 701 &zfsvfs->z_defaultgroupobjquota); 702 if (error != 0) 703 return (error); 704 705 error = zfs_get_zplprop(os, ZFS_PROP_DEFAULTPROJECTOBJQUOTA, 706 &zfsvfs->z_defaultprojectobjquota); 707 if (error != 0) 708 return (error); 709 710 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 711 &zfsvfs->z_root); 712 if (error != 0) 713 return (error); 714 ASSERT(zfsvfs->z_root != 0); 715 716 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 717 &zfsvfs->z_unlinkedobj); 718 if (error != 0) 719 return (error); 720 721 error = zap_lookup(os, MASTER_NODE_OBJ, 722 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], 723 8, 1, &zfsvfs->z_userquota_obj); 724 if (error == ENOENT) 725 zfsvfs->z_userquota_obj = 0; 726 else if (error != 0) 727 return (error); 728 729 error = zap_lookup(os, MASTER_NODE_OBJ, 730 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], 731 8, 1, &zfsvfs->z_groupquota_obj); 732 if (error == ENOENT) 733 zfsvfs->z_groupquota_obj = 0; 734 else if (error != 0) 735 return (error); 736 737 error = zap_lookup(os, MASTER_NODE_OBJ, 738 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA], 739 8, 1, &zfsvfs->z_projectquota_obj); 740 if (error == ENOENT) 741 zfsvfs->z_projectquota_obj = 0; 742 else if (error != 0) 743 return (error); 744 745 error = zap_lookup(os, MASTER_NODE_OBJ, 746 zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA], 747 8, 1, &zfsvfs->z_userobjquota_obj); 748 if (error == ENOENT) 749 zfsvfs->z_userobjquota_obj = 0; 750 else if (error != 0) 751 return (error); 752 753 error = zap_lookup(os, MASTER_NODE_OBJ, 754 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA], 755 8, 1, &zfsvfs->z_groupobjquota_obj); 756 if (error == ENOENT) 757 zfsvfs->z_groupobjquota_obj = 0; 758 else if (error != 0) 759 return (error); 760 761 error = zap_lookup(os, MASTER_NODE_OBJ, 762 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA], 763 8, 1, &zfsvfs->z_projectobjquota_obj); 764 if (error == ENOENT) 765 zfsvfs->z_projectobjquota_obj = 0; 766 else if (error != 0) 767 return (error); 768 769 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, 770 &zfsvfs->z_fuid_obj); 771 if (error == ENOENT) 772 zfsvfs->z_fuid_obj = 0; 773 else if (error != 0) 774 return (error); 775 776 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, 777 &zfsvfs->z_shares_dir); 778 if (error == ENOENT) 779 zfsvfs->z_shares_dir = 0; 780 else if (error != 0) 781 return (error); 782 783 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 784 &zfsvfs->z_attr_table); 785 if (error != 0) 786 return (error); 787 788 if (zfsvfs->z_version >= ZPL_VERSION_SA) 789 sa_register_update_callback(os, zfs_sa_upgrade); 790 791 return (0); 792 } 793 794 int 795 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp) 796 { 797 objset_t *os; 798 zfsvfs_t *zfsvfs; 799 int error; 800 boolean_t ro = (readonly || (strchr(osname, '@') != NULL)); 801 802 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 803 804 error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os); 805 if (error != 0) { 806 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 807 return (error); 808 } 809 810 error = zfsvfs_create_impl(zfvp, zfsvfs, os); 811 812 return (error); 813 } 814 815 816 /* 817 * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function 818 * on a failure. Do not pass in a statically allocated zfsvfs. 819 */ 820 int 821 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os) 822 { 823 int error; 824 825 zfsvfs->z_vfs = NULL; 826 zfsvfs->z_sb = NULL; 827 zfsvfs->z_parent = zfsvfs; 828 829 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 830 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); 831 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 832 offsetof(znode_t, z_link_node)); 833 ZFS_TEARDOWN_INIT(zfsvfs); 834 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); 835 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); 836 837 int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1), 838 ZFS_OBJ_MTX_MAX); 839 zfsvfs->z_hold_size = size; 840 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size, 841 KM_SLEEP); 842 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP); 843 for (int i = 0; i != size; i++) { 844 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare, 845 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node)); 846 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL); 847 } 848 849 error = zfsvfs_init(zfsvfs, os); 850 if (error != 0) { 851 dmu_objset_disown(os, B_TRUE, zfsvfs); 852 *zfvp = NULL; 853 zfsvfs_free(zfsvfs); 854 return (error); 855 } 856 857 zfsvfs->z_drain_task = TASKQID_INVALID; 858 zfsvfs->z_draining = B_FALSE; 859 zfsvfs->z_drain_cancel = B_TRUE; 860 861 *zfvp = zfsvfs; 862 return (0); 863 } 864 865 static int 866 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) 867 { 868 int error; 869 boolean_t readonly = zfs_is_readonly(zfsvfs); 870 871 error = zfs_register_callbacks(zfsvfs->z_vfs); 872 if (error) 873 return (error); 874 875 /* 876 * If we are not mounting (ie: online recv), then we don't 877 * have to worry about replaying the log as we blocked all 878 * operations out since we closed the ZIL. 879 */ 880 if (mounting) { 881 ASSERT0P(zfsvfs->z_kstat.dk_kstats); 882 error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os); 883 if (error) 884 return (error); 885 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data, 886 &zfsvfs->z_kstat.dk_zil_sums); 887 888 /* 889 * During replay we remove the read only flag to 890 * allow replays to succeed. 891 */ 892 if (readonly != 0) { 893 readonly_changed_cb(zfsvfs, B_FALSE); 894 } else { 895 zap_stats_t zs; 896 if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj, 897 &zs) == 0) { 898 dataset_kstats_update_nunlinks_kstat( 899 &zfsvfs->z_kstat, zs.zs_num_entries); 900 dprintf_ds(zfsvfs->z_os->os_dsl_dataset, 901 "num_entries in unlinked set: %llu", 902 zs.zs_num_entries); 903 } 904 zfs_unlinked_drain(zfsvfs); 905 dsl_dir_t *dd = zfsvfs->z_os->os_dsl_dataset->ds_dir; 906 dd->dd_activity_cancelled = B_FALSE; 907 } 908 909 /* 910 * Parse and replay the intent log. 911 * 912 * Because of ziltest, this must be done after 913 * zfs_unlinked_drain(). (Further note: ziltest 914 * doesn't use readonly mounts, where 915 * zfs_unlinked_drain() isn't called.) This is because 916 * ziltest causes spa_sync() to think it's committed, 917 * but actually it is not, so the intent log contains 918 * many txg's worth of changes. 919 * 920 * In particular, if object N is in the unlinked set in 921 * the last txg to actually sync, then it could be 922 * actually freed in a later txg and then reallocated 923 * in a yet later txg. This would write a "create 924 * object N" record to the intent log. Normally, this 925 * would be fine because the spa_sync() would have 926 * written out the fact that object N is free, before 927 * we could write the "create object N" intent log 928 * record. 929 * 930 * But when we are in ziltest mode, we advance the "open 931 * txg" without actually spa_sync()-ing the changes to 932 * disk. So we would see that object N is still 933 * allocated and in the unlinked set, and there is an 934 * intent log record saying to allocate it. 935 */ 936 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { 937 if (zil_replay_disable) { 938 zil_destroy(zfsvfs->z_log, B_FALSE); 939 } else { 940 zfsvfs->z_replay = B_TRUE; 941 zil_replay(zfsvfs->z_os, zfsvfs, 942 zfs_replay_vector); 943 zfsvfs->z_replay = B_FALSE; 944 } 945 } 946 947 /* restore readonly bit */ 948 if (readonly != 0) 949 readonly_changed_cb(zfsvfs, B_TRUE); 950 } else { 951 ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL); 952 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data, 953 &zfsvfs->z_kstat.dk_zil_sums); 954 } 955 956 /* 957 * Set the objset user_ptr to track its zfsvfs. 958 */ 959 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 960 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 961 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 962 963 return (0); 964 } 965 966 void 967 zfsvfs_free(zfsvfs_t *zfsvfs) 968 { 969 int i, size = zfsvfs->z_hold_size; 970 971 zfs_fuid_destroy(zfsvfs); 972 973 mutex_destroy(&zfsvfs->z_znodes_lock); 974 mutex_destroy(&zfsvfs->z_lock); 975 list_destroy(&zfsvfs->z_all_znodes); 976 ZFS_TEARDOWN_DESTROY(zfsvfs); 977 rw_destroy(&zfsvfs->z_teardown_inactive_lock); 978 rw_destroy(&zfsvfs->z_fuid_lock); 979 for (i = 0; i != size; i++) { 980 avl_destroy(&zfsvfs->z_hold_trees[i]); 981 mutex_destroy(&zfsvfs->z_hold_locks[i]); 982 } 983 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size); 984 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size); 985 zfsvfs_vfs_free(zfsvfs->z_vfs); 986 dataset_kstats_destroy(&zfsvfs->z_kstat); 987 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 988 } 989 990 static void 991 zfs_set_fuid_feature(zfsvfs_t *zfsvfs) 992 { 993 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); 994 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); 995 } 996 997 static void 998 zfs_unregister_callbacks(zfsvfs_t *zfsvfs) 999 { 1000 objset_t *os = zfsvfs->z_os; 1001 1002 if (!dmu_objset_is_snapshot(os)) 1003 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); 1004 } 1005 1006 #ifdef HAVE_MLSLABEL 1007 /* 1008 * Check that the hex label string is appropriate for the dataset being 1009 * mounted into the global_zone proper. 1010 * 1011 * Return an error if the hex label string is not default or 1012 * admin_low/admin_high. For admin_low labels, the corresponding 1013 * dataset must be readonly. 1014 */ 1015 int 1016 zfs_check_global_label(const char *dsname, const char *hexsl) 1017 { 1018 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) 1019 return (0); 1020 if (strcasecmp(hexsl, ADMIN_HIGH) == 0) 1021 return (0); 1022 if (strcasecmp(hexsl, ADMIN_LOW) == 0) { 1023 /* must be readonly */ 1024 uint64_t rdonly; 1025 1026 if (dsl_prop_get_integer(dsname, 1027 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) 1028 return (SET_ERROR(EACCES)); 1029 return (rdonly ? 0 : SET_ERROR(EACCES)); 1030 } 1031 return (SET_ERROR(EACCES)); 1032 } 1033 #endif /* HAVE_MLSLABEL */ 1034 1035 static int 1036 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp, 1037 uint32_t bshift) 1038 { 1039 char buf[20 + DMU_OBJACCT_PREFIX_LEN]; 1040 uint64_t offset = DMU_OBJACCT_PREFIX_LEN; 1041 uint64_t quota; 1042 uint64_t used; 1043 int err; 1044 1045 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1); 1046 err = zfs_id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset, 1047 sizeof (buf) - offset, B_FALSE); 1048 if (err) 1049 return (err); 1050 1051 if (zfsvfs->z_projectquota_obj == 0) { 1052 if (zfsvfs->z_defaultprojectquota == 0) 1053 goto objs; 1054 quota = zfsvfs->z_defaultprojectquota; 1055 } else { 1056 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj, 1057 buf + offset, 8, 1, "a); 1058 if (err && (quota = zfsvfs->z_defaultprojectquota) == 0) { 1059 if (err == ENOENT) 1060 goto objs; 1061 return (err); 1062 } 1063 } 1064 1065 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT, 1066 buf + offset, 8, 1, &used); 1067 if (unlikely(err == ENOENT)) { 1068 uint32_t blksize; 1069 u_longlong_t nblocks; 1070 1071 /* 1072 * Quota accounting is async, so it is possible race case. 1073 * There is at least one object with the given project ID. 1074 */ 1075 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 1076 if (unlikely(zp->z_blksz == 0)) 1077 blksize = zfsvfs->z_max_blksz; 1078 1079 used = blksize * nblocks; 1080 } else if (err) { 1081 return (err); 1082 } 1083 1084 statp->f_blocks = quota >> bshift; 1085 statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0; 1086 statp->f_bavail = statp->f_bfree; 1087 1088 objs: 1089 1090 if (zfsvfs->z_projectobjquota_obj == 0) { 1091 if (zfsvfs->z_defaultprojectobjquota == 0) 1092 return (0); 1093 quota = zfsvfs->z_defaultprojectobjquota; 1094 } else { 1095 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj, 1096 buf + offset, 8, 1, "a); 1097 if (err && (quota = zfsvfs->z_defaultprojectobjquota) == 0) { 1098 if (err == ENOENT) 1099 return (0); 1100 return (err); 1101 } 1102 } 1103 1104 1105 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT, 1106 buf, 8, 1, &used); 1107 if (unlikely(err == ENOENT)) { 1108 /* 1109 * Quota accounting is async, so it is possible race case. 1110 * There is at least one object with the given project ID. 1111 */ 1112 used = 1; 1113 } else if (err) { 1114 return (err); 1115 } 1116 1117 statp->f_files = quota; 1118 statp->f_ffree = (quota > used) ? (quota - used) : 0; 1119 1120 return (0); 1121 } 1122 1123 int 1124 zfs_statvfs(struct inode *ip, struct kstatfs *statp) 1125 { 1126 zfsvfs_t *zfsvfs = ITOZSB(ip); 1127 uint64_t refdbytes, availbytes, usedobjs, availobjs; 1128 int err = 0; 1129 1130 if ((err = zfs_enter(zfsvfs, FTAG)) != 0) 1131 return (err); 1132 1133 dmu_objset_space(zfsvfs->z_os, 1134 &refdbytes, &availbytes, &usedobjs, &availobjs); 1135 1136 uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os); 1137 /* 1138 * The underlying storage pool actually uses multiple block 1139 * size. Under Solaris frsize (fragment size) is reported as 1140 * the smallest block size we support, and bsize (block size) 1141 * as the filesystem's maximum block size. Unfortunately, 1142 * under Linux the fragment size and block size are often used 1143 * interchangeably. Thus we are forced to report both of them 1144 * as the filesystem's maximum block size. 1145 */ 1146 statp->f_frsize = zfsvfs->z_max_blksz; 1147 statp->f_bsize = zfsvfs->z_max_blksz; 1148 uint32_t bshift = fls(statp->f_bsize) - 1; 1149 1150 /* 1151 * The following report "total" blocks of various kinds in 1152 * the file system, but reported in terms of f_bsize - the 1153 * "preferred" size. 1154 */ 1155 1156 /* Round up so we never have a filesystem using 0 blocks. */ 1157 refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize); 1158 statp->f_blocks = (refdbytes + availbytes) >> bshift; 1159 statp->f_bfree = availbytes >> bshift; 1160 statp->f_bavail = statp->f_bfree; /* no root reservation */ 1161 1162 /* 1163 * statvfs() should really be called statufs(), because it assumes 1164 * static metadata. ZFS doesn't preallocate files, so the best 1165 * we can do is report the max that could possibly fit in f_files, 1166 * and that minus the number actually used in f_ffree. 1167 * For f_ffree, report the smaller of the number of objects available 1168 * and the number of blocks (each object will take at least a block). 1169 */ 1170 statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT); 1171 statp->f_files = statp->f_ffree + usedobjs; 1172 statp->f_fsid.val[0] = (uint32_t)fsid; 1173 statp->f_fsid.val[1] = (uint32_t)(fsid >> 32); 1174 statp->f_type = ZFS_SUPER_MAGIC; 1175 statp->f_namelen = 1176 zfsvfs->z_longname ? (ZAP_MAXNAMELEN_NEW - 1) : (MAXNAMELEN - 1); 1177 1178 /* 1179 * We have all of 40 characters to stuff a string here. 1180 * Is there anything useful we could/should provide? 1181 */ 1182 memset(statp->f_spare, 0, sizeof (statp->f_spare)); 1183 1184 if (dmu_objset_projectquota_enabled(zfsvfs->z_os) && 1185 dmu_objset_projectquota_present(zfsvfs->z_os)) { 1186 znode_t *zp = ITOZ(ip); 1187 1188 if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid && 1189 zpl_is_valid_projid(zp->z_projid)) 1190 err = zfs_statfs_project(zfsvfs, zp, statp, bshift); 1191 } 1192 1193 zfs_exit(zfsvfs, FTAG); 1194 return (err); 1195 } 1196 1197 static int 1198 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp) 1199 { 1200 znode_t *rootzp; 1201 int error; 1202 1203 if ((error = zfs_enter(zfsvfs, FTAG)) != 0) 1204 return (error); 1205 1206 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); 1207 if (error == 0) 1208 *ipp = ZTOI(rootzp); 1209 1210 zfs_exit(zfsvfs, FTAG); 1211 return (error); 1212 } 1213 1214 /* 1215 * Dentry and inode caches referenced by a task in non-root memcg are 1216 * not going to be scanned by the kernel-provided shrinker. So, if 1217 * kernel prunes nothing, fall back to this manual walk to free dnodes. 1218 * To avoid scanning the same znodes multiple times they are always rotated 1219 * to the end of the z_all_znodes list. New znodes are inserted at the 1220 * end of the list so we're always scanning the oldest znodes first. 1221 */ 1222 static int 1223 zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan) 1224 { 1225 znode_t **zp_array, *zp; 1226 int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *)); 1227 int objects = 0; 1228 int i = 0, j = 0; 1229 1230 zp_array = vmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP); 1231 1232 mutex_enter(&zfsvfs->z_znodes_lock); 1233 while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) { 1234 1235 if ((i++ > nr_to_scan) || (j >= max_array)) 1236 break; 1237 1238 ASSERT(list_link_active(&zp->z_link_node)); 1239 list_remove(&zfsvfs->z_all_znodes, zp); 1240 list_insert_tail(&zfsvfs->z_all_znodes, zp); 1241 1242 /* Skip active znodes and .zfs entries */ 1243 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir) 1244 continue; 1245 1246 if (igrab(ZTOI(zp)) == NULL) 1247 continue; 1248 1249 zp_array[j] = zp; 1250 j++; 1251 } 1252 mutex_exit(&zfsvfs->z_znodes_lock); 1253 1254 for (i = 0; i < j; i++) { 1255 zp = zp_array[i]; 1256 1257 ASSERT3P(zp, !=, NULL); 1258 d_prune_aliases(ZTOI(zp)); 1259 1260 if (atomic_read(&ZTOI(zp)->i_count) == 1) 1261 objects++; 1262 1263 zrele(zp); 1264 } 1265 1266 vmem_free(zp_array, max_array * sizeof (znode_t *)); 1267 1268 return (objects); 1269 } 1270 1271 /* 1272 * The ARC has requested that the filesystem drop entries from the dentry 1273 * and inode caches. This can occur when the ARC needs to free meta data 1274 * blocks but can't because they are all pinned by entries in these caches. 1275 */ 1276 #if defined(HAVE_SUPER_BLOCK_S_SHRINK) 1277 #define S_SHRINK(sb) (&(sb)->s_shrink) 1278 #elif defined(HAVE_SUPER_BLOCK_S_SHRINK_PTR) 1279 #define S_SHRINK(sb) ((sb)->s_shrink) 1280 #endif 1281 1282 int 1283 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects) 1284 { 1285 zfsvfs_t *zfsvfs = sb->s_fs_info; 1286 int error = 0; 1287 struct shrinker *shrinker = S_SHRINK(sb); 1288 struct shrink_control sc = { 1289 .nr_to_scan = nr_to_scan, 1290 .gfp_mask = GFP_KERNEL, 1291 }; 1292 1293 if ((error = zfs_enter(zfsvfs, FTAG)) != 0) 1294 return (error); 1295 1296 #ifdef SHRINKER_NUMA_AWARE 1297 if (shrinker->flags & SHRINKER_NUMA_AWARE) { 1298 long tc = 1; 1299 for_each_online_node(sc.nid) { 1300 long c = shrinker->count_objects(shrinker, &sc); 1301 if (c == 0 || c == SHRINK_EMPTY) 1302 continue; 1303 tc += c; 1304 } 1305 *objects = 0; 1306 for_each_online_node(sc.nid) { 1307 long c = shrinker->count_objects(shrinker, &sc); 1308 if (c == 0 || c == SHRINK_EMPTY) 1309 continue; 1310 if (c > tc) 1311 tc = c; 1312 sc.nr_to_scan = mult_frac(nr_to_scan, c, tc) + 1; 1313 *objects += (*shrinker->scan_objects)(shrinker, &sc); 1314 } 1315 } else { 1316 *objects = (*shrinker->scan_objects)(shrinker, &sc); 1317 } 1318 #else 1319 *objects = (*shrinker->scan_objects)(shrinker, &sc); 1320 #endif 1321 1322 /* 1323 * Fall back to zfs_prune_aliases if kernel's shrinker did nothing 1324 * due to dentry and inode caches being referenced by a task running 1325 * in non-root memcg. 1326 */ 1327 if (*objects == 0) 1328 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan); 1329 1330 zfs_exit(zfsvfs, FTAG); 1331 1332 dprintf_ds(zfsvfs->z_os->os_dsl_dataset, 1333 "pruning, nr_to_scan=%lu objects=%d error=%d\n", 1334 nr_to_scan, *objects, error); 1335 1336 return (error); 1337 } 1338 1339 /* 1340 * Teardown the zfsvfs_t. 1341 * 1342 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock' 1343 * and 'z_teardown_inactive_lock' held. 1344 */ 1345 static int 1346 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) 1347 { 1348 znode_t *zp; 1349 1350 zfs_unlinked_drain_stop_wait(zfsvfs); 1351 1352 /* 1353 * If someone has not already unmounted this file system, 1354 * drain the zrele_taskq to ensure all active references to the 1355 * zfsvfs_t have been handled only then can it be safely destroyed. 1356 */ 1357 if (zfsvfs->z_os) { 1358 /* 1359 * If we're unmounting we have to wait for the list to 1360 * drain completely. 1361 * 1362 * If we're not unmounting there's no guarantee the list 1363 * will drain completely, but iputs run from the taskq 1364 * may add the parents of dir-based xattrs to the taskq 1365 * so we want to wait for these. 1366 * 1367 * We can safely check z_all_znodes for being empty because the 1368 * VFS has already blocked operations which add to it. 1369 */ 1370 int round = 0; 1371 while (!list_is_empty(&zfsvfs->z_all_znodes)) { 1372 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1373 dmu_objset_pool(zfsvfs->z_os)), 0); 1374 if (++round > 1 && !unmounting) 1375 break; 1376 } 1377 } 1378 1379 ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG); 1380 1381 if (!unmounting) { 1382 /* 1383 * We purge the parent filesystem's super block as the 1384 * parent filesystem and all of its snapshots have their 1385 * inode's super block set to the parent's filesystem's 1386 * super block. Note, 'z_parent' is self referential 1387 * for non-snapshots. 1388 */ 1389 shrink_dcache_sb(zfsvfs->z_parent->z_sb); 1390 } 1391 1392 /* 1393 * Close the zil. NB: Can't close the zil while zfs_inactive 1394 * threads are blocked as zil_close can call zfs_inactive. 1395 */ 1396 if (zfsvfs->z_log) { 1397 zil_close(zfsvfs->z_log); 1398 zfsvfs->z_log = NULL; 1399 } 1400 1401 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER); 1402 1403 /* 1404 * If we are not unmounting (ie: online recv) and someone already 1405 * unmounted this file system while we were doing the switcheroo, 1406 * or a reopen of z_os failed then just bail out now. 1407 */ 1408 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { 1409 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1410 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1411 return (SET_ERROR(EIO)); 1412 } 1413 1414 /* 1415 * At this point there are no VFS ops active, and any new VFS ops 1416 * will fail with EIO since we have z_teardown_lock for writer (only 1417 * relevant for forced unmount). 1418 * 1419 * Release all holds on dbufs. We also grab an extra reference to all 1420 * the remaining inodes so that the kernel does not attempt to free 1421 * any inodes of a suspended fs. This can cause deadlocks since the 1422 * zfs_resume_fs() process may involve starting threads, which might 1423 * attempt to free unreferenced inodes to free up memory for the new 1424 * thread. 1425 */ 1426 if (!unmounting) { 1427 mutex_enter(&zfsvfs->z_znodes_lock); 1428 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; 1429 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1430 if (zp->z_sa_hdl) 1431 zfs_znode_dmu_fini(zp); 1432 if (igrab(ZTOI(zp)) != NULL) 1433 zp->z_suspended = B_TRUE; 1434 1435 } 1436 mutex_exit(&zfsvfs->z_znodes_lock); 1437 } 1438 1439 /* 1440 * If we are unmounting, set the unmounted flag and let new VFS ops 1441 * unblock. zfs_inactive will have the unmounted behavior, and all 1442 * other VFS ops will fail with EIO. 1443 */ 1444 if (unmounting) { 1445 zfsvfs->z_unmounted = B_TRUE; 1446 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1447 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1448 } 1449 1450 /* 1451 * z_os will be NULL if there was an error in attempting to reopen 1452 * zfsvfs, so just return as the properties had already been 1453 * 1454 * unregistered and cached data had been evicted before. 1455 */ 1456 if (zfsvfs->z_os == NULL) 1457 return (0); 1458 1459 /* 1460 * Unregister properties. 1461 */ 1462 zfs_unregister_callbacks(zfsvfs); 1463 1464 /* 1465 * Evict cached data. We must write out any dirty data before 1466 * disowning the dataset. 1467 */ 1468 objset_t *os = zfsvfs->z_os; 1469 boolean_t os_dirty = B_FALSE; 1470 for (int t = 0; t < TXG_SIZE; t++) { 1471 if (dmu_objset_is_dirty(os, t)) { 1472 os_dirty = B_TRUE; 1473 break; 1474 } 1475 } 1476 if (!zfs_is_readonly(zfsvfs) && os_dirty) { 1477 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1478 } 1479 dmu_objset_evict_dbufs(zfsvfs->z_os); 1480 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir; 1481 dsl_dir_cancel_waiters(dd); 1482 1483 return (0); 1484 } 1485 1486 static atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0); 1487 1488 int 1489 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent) 1490 { 1491 const char *osname = zm->mnt_osname; 1492 struct inode *root_inode = NULL; 1493 uint64_t recordsize; 1494 int error = 0; 1495 zfsvfs_t *zfsvfs = NULL; 1496 vfs_t *vfs = NULL; 1497 int canwrite; 1498 int dataset_visible_zone; 1499 1500 ASSERT(zm); 1501 ASSERT(osname); 1502 1503 dataset_visible_zone = zone_dataset_visible(osname, &canwrite); 1504 1505 /* 1506 * Refuse to mount a filesystem if we are in a namespace and the 1507 * dataset is not visible or writable in that namespace. 1508 */ 1509 if (!INGLOBALZONE(curproc) && 1510 (!dataset_visible_zone || !canwrite)) { 1511 return (SET_ERROR(EPERM)); 1512 } 1513 1514 error = zfsvfs_parse_options(zm->mnt_data, &vfs); 1515 if (error) 1516 return (error); 1517 1518 /* 1519 * If a non-writable filesystem is being mounted without the 1520 * read-only flag, pretend it was set, as done for snapshots. 1521 */ 1522 if (!canwrite) 1523 vfs->vfs_readonly = B_TRUE; 1524 1525 error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs); 1526 if (error) { 1527 zfsvfs_vfs_free(vfs); 1528 goto out; 1529 } 1530 1531 if ((error = dsl_prop_get_integer(osname, "recordsize", 1532 &recordsize, NULL))) { 1533 zfsvfs_vfs_free(vfs); 1534 goto out; 1535 } 1536 1537 vfs->vfs_data = zfsvfs; 1538 zfsvfs->z_vfs = vfs; 1539 zfsvfs->z_sb = sb; 1540 sb->s_fs_info = zfsvfs; 1541 sb->s_magic = ZFS_SUPER_MAGIC; 1542 sb->s_maxbytes = MAX_LFS_FILESIZE; 1543 sb->s_time_gran = 1; 1544 sb->s_blocksize = recordsize; 1545 sb->s_blocksize_bits = ilog2(recordsize); 1546 1547 error = -super_setup_bdi_name(sb, "%.28s-%ld", "zfs", 1548 atomic_long_inc_return(&zfs_bdi_seq)); 1549 if (error) 1550 goto out; 1551 1552 sb->s_bdi->ra_pages = 0; 1553 1554 /* Set callback operations for the file system. */ 1555 sb->s_op = &zpl_super_operations; 1556 sb->s_xattr = zpl_xattr_handlers; 1557 sb->s_export_op = &zpl_export_operations; 1558 1559 /* Set features for file system. */ 1560 zfs_set_fuid_feature(zfsvfs); 1561 1562 if (dmu_objset_is_snapshot(zfsvfs->z_os)) { 1563 uint64_t pval; 1564 1565 atime_changed_cb(zfsvfs, B_FALSE); 1566 readonly_changed_cb(zfsvfs, B_TRUE); 1567 if ((error = dsl_prop_get_integer(osname, 1568 "xattr", &pval, NULL))) 1569 goto out; 1570 xattr_changed_cb(zfsvfs, pval); 1571 if ((error = dsl_prop_get_integer(osname, 1572 "acltype", &pval, NULL))) 1573 goto out; 1574 acltype_changed_cb(zfsvfs, pval); 1575 zfsvfs->z_issnap = B_TRUE; 1576 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; 1577 zfsvfs->z_snap_defer_time = jiffies; 1578 1579 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); 1580 dmu_objset_set_user(zfsvfs->z_os, zfsvfs); 1581 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); 1582 } else { 1583 if ((error = zfsvfs_setup(zfsvfs, B_TRUE))) 1584 goto out; 1585 } 1586 1587 /* Allocate a root inode for the filesystem. */ 1588 error = zfs_root(zfsvfs, &root_inode); 1589 if (error) { 1590 (void) zfs_umount(sb); 1591 zfsvfs = NULL; /* avoid double-free; first in zfs_umount */ 1592 goto out; 1593 } 1594 1595 /* Allocate a root dentry for the filesystem */ 1596 sb->s_root = d_make_root(root_inode); 1597 if (sb->s_root == NULL) { 1598 (void) zfs_umount(sb); 1599 zfsvfs = NULL; /* avoid double-free; first in zfs_umount */ 1600 error = SET_ERROR(ENOMEM); 1601 goto out; 1602 } 1603 1604 if (!zfsvfs->z_issnap) 1605 zfsctl_create(zfsvfs); 1606 1607 zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb); 1608 out: 1609 if (error) { 1610 if (zfsvfs != NULL) { 1611 dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs); 1612 zfsvfs_free(zfsvfs); 1613 } 1614 /* 1615 * make sure we don't have dangling sb->s_fs_info which 1616 * zfs_preumount will use. 1617 */ 1618 sb->s_fs_info = NULL; 1619 } 1620 1621 return (error); 1622 } 1623 1624 /* 1625 * Called when an unmount is requested and certain sanity checks have 1626 * already passed. At this point no dentries or inodes have been reclaimed 1627 * from their respective caches. We drop the extra reference on the .zfs 1628 * control directory to allow everything to be reclaimed. All snapshots 1629 * must already have been unmounted to reach this point. 1630 */ 1631 void 1632 zfs_preumount(struct super_block *sb) 1633 { 1634 zfsvfs_t *zfsvfs = sb->s_fs_info; 1635 1636 /* zfsvfs is NULL when zfs_domount fails during mount */ 1637 if (zfsvfs) { 1638 zfs_unlinked_drain_stop_wait(zfsvfs); 1639 zfsctl_destroy(sb->s_fs_info); 1640 /* 1641 * Wait for zrele_async before entering evict_inodes in 1642 * generic_shutdown_super. The reason we must finish before 1643 * evict_inodes is when lazytime is on, or when zfs_purgedir 1644 * calls zfs_zget, zrele would bump i_count from 0 to 1. This 1645 * would race with the i_count check in evict_inodes. This means 1646 * it could destroy the inode while we are still using it. 1647 * 1648 * We wait for two passes. xattr directories in the first pass 1649 * may add xattr entries in zfs_purgedir, so in the second pass 1650 * we wait for them. We don't use taskq_wait here because it is 1651 * a pool wide taskq. Other mounted filesystems can constantly 1652 * do zrele_async and there's no guarantee when taskq will be 1653 * empty. 1654 */ 1655 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1656 dmu_objset_pool(zfsvfs->z_os)), 0); 1657 taskq_wait_outstanding(dsl_pool_zrele_taskq( 1658 dmu_objset_pool(zfsvfs->z_os)), 0); 1659 } 1660 } 1661 1662 /* 1663 * Called once all other unmount released tear down has occurred. 1664 * It is our responsibility to release any remaining infrastructure. 1665 */ 1666 int 1667 zfs_umount(struct super_block *sb) 1668 { 1669 zfsvfs_t *zfsvfs = sb->s_fs_info; 1670 objset_t *os; 1671 1672 if (zfsvfs->z_arc_prune != NULL) 1673 arc_remove_prune_callback(zfsvfs->z_arc_prune); 1674 VERIFY0(zfsvfs_teardown(zfsvfs, B_TRUE)); 1675 os = zfsvfs->z_os; 1676 1677 /* 1678 * z_os will be NULL if there was an error in 1679 * attempting to reopen zfsvfs. 1680 */ 1681 if (os != NULL) { 1682 /* 1683 * Unset the objset user_ptr. 1684 */ 1685 mutex_enter(&os->os_user_ptr_lock); 1686 dmu_objset_set_user(os, NULL); 1687 mutex_exit(&os->os_user_ptr_lock); 1688 1689 /* 1690 * Finally release the objset 1691 */ 1692 dmu_objset_disown(os, B_TRUE, zfsvfs); 1693 } 1694 1695 zfsvfs_free(zfsvfs); 1696 sb->s_fs_info = NULL; 1697 return (0); 1698 } 1699 1700 int 1701 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm) 1702 { 1703 zfsvfs_t *zfsvfs = sb->s_fs_info; 1704 vfs_t *vfsp; 1705 boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os); 1706 int error; 1707 1708 if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) && 1709 !(*flags & SB_RDONLY)) { 1710 *flags |= SB_RDONLY; 1711 return (EROFS); 1712 } 1713 1714 error = zfsvfs_parse_options(zm->mnt_data, &vfsp); 1715 if (error) 1716 return (error); 1717 1718 if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY)) 1719 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); 1720 1721 zfs_unregister_callbacks(zfsvfs); 1722 zfsvfs_vfs_free(zfsvfs->z_vfs); 1723 1724 vfsp->vfs_data = zfsvfs; 1725 zfsvfs->z_vfs = vfsp; 1726 if (!issnap) 1727 (void) zfs_register_callbacks(vfsp); 1728 1729 return (error); 1730 } 1731 1732 int 1733 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp) 1734 { 1735 zfsvfs_t *zfsvfs = sb->s_fs_info; 1736 znode_t *zp; 1737 uint64_t object = 0; 1738 uint64_t fid_gen = 0; 1739 uint64_t gen_mask; 1740 uint64_t zp_gen; 1741 int i, err; 1742 1743 *ipp = NULL; 1744 1745 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { 1746 zfid_short_t *zfid = (zfid_short_t *)fidp; 1747 1748 for (i = 0; i < sizeof (zfid->zf_object); i++) 1749 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); 1750 1751 for (i = 0; i < sizeof (zfid->zf_gen); i++) 1752 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); 1753 } else { 1754 return (SET_ERROR(EINVAL)); 1755 } 1756 1757 /* LONG_FID_LEN means snapdirs */ 1758 if (fidp->fid_len == LONG_FID_LEN) { 1759 zfid_long_t *zlfid = (zfid_long_t *)fidp; 1760 uint64_t objsetid = 0; 1761 uint64_t setgen = 0; 1762 1763 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 1764 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); 1765 1766 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 1767 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); 1768 1769 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) { 1770 dprintf("snapdir fid: objsetid (%llu) != " 1771 "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n", 1772 objsetid, ZFSCTL_INO_SNAPDIRS, object); 1773 1774 return (SET_ERROR(EINVAL)); 1775 } 1776 1777 if (fid_gen > 1 || setgen != 0) { 1778 dprintf("snapdir fid: fid_gen (%llu) and setgen " 1779 "(%llu)\n", fid_gen, setgen); 1780 return (SET_ERROR(EINVAL)); 1781 } 1782 1783 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp)); 1784 } 1785 1786 if ((err = zfs_enter(zfsvfs, FTAG)) != 0) 1787 return (err); 1788 /* A zero fid_gen means we are in the .zfs control directories */ 1789 if (fid_gen == 0 && 1790 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { 1791 if (zfsvfs->z_show_ctldir == ZFS_SNAPDIR_DISABLED) { 1792 zfs_exit(zfsvfs, FTAG); 1793 return (SET_ERROR(ENOENT)); 1794 } 1795 1796 *ipp = zfsvfs->z_ctldir; 1797 ASSERT(*ipp != NULL); 1798 1799 if (object == ZFSCTL_INO_SNAPDIR) { 1800 VERIFY0(zfsctl_root_lookup(*ipp, "snapshot", ipp, 1801 0, kcred, NULL, NULL)); 1802 } else { 1803 /* 1804 * Must have an existing ref, so igrab() 1805 * cannot return NULL 1806 */ 1807 VERIFY3P(igrab(*ipp), !=, NULL); 1808 } 1809 zfs_exit(zfsvfs, FTAG); 1810 return (0); 1811 } 1812 1813 gen_mask = -1ULL >> (64 - 8 * i); 1814 1815 dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask); 1816 if ((err = zfs_zget(zfsvfs, object, &zp))) { 1817 zfs_exit(zfsvfs, FTAG); 1818 return (err); 1819 } 1820 1821 /* Don't export xattr stuff */ 1822 if (zp->z_pflags & ZFS_XATTR) { 1823 zrele(zp); 1824 zfs_exit(zfsvfs, FTAG); 1825 return (SET_ERROR(ENOENT)); 1826 } 1827 1828 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, 1829 sizeof (uint64_t)); 1830 zp_gen = zp_gen & gen_mask; 1831 if (zp_gen == 0) 1832 zp_gen = 1; 1833 if ((fid_gen == 0) && (zfsvfs->z_root == object)) 1834 fid_gen = zp_gen; 1835 if (zp->z_unlinked || zp_gen != fid_gen) { 1836 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen, 1837 fid_gen); 1838 zrele(zp); 1839 zfs_exit(zfsvfs, FTAG); 1840 return (SET_ERROR(ENOENT)); 1841 } 1842 1843 *ipp = ZTOI(zp); 1844 if (*ipp) 1845 zfs_znode_update_vfs(ITOZ(*ipp)); 1846 1847 zfs_exit(zfsvfs, FTAG); 1848 return (0); 1849 } 1850 1851 /* 1852 * Block out VFS ops and close zfsvfs_t 1853 * 1854 * Note, if successful, then we return with the 'z_teardown_lock' and 1855 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying 1856 * dataset and objset intact so that they can be atomically handed off during 1857 * a subsequent rollback or recv operation and the resume thereafter. 1858 */ 1859 int 1860 zfs_suspend_fs(zfsvfs_t *zfsvfs) 1861 { 1862 int error; 1863 1864 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) 1865 return (error); 1866 1867 return (0); 1868 } 1869 1870 /* 1871 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset 1872 * is an invariant across any of the operations that can be performed while the 1873 * filesystem was suspended. Whether it succeeded or failed, the preconditions 1874 * are the same: the relevant objset and associated dataset are owned by 1875 * zfsvfs, held, and long held on entry. 1876 */ 1877 int 1878 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) 1879 { 1880 int err, err2; 1881 znode_t *zp; 1882 1883 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs)); 1884 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1885 1886 /* 1887 * We already own this, so just update the objset_t, as the one we 1888 * had before may have been evicted. 1889 */ 1890 objset_t *os; 1891 VERIFY3P(ds->ds_owner, ==, zfsvfs); 1892 VERIFY(dsl_dataset_long_held(ds)); 1893 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds)); 1894 dsl_pool_config_enter(dp, FTAG); 1895 VERIFY0(dmu_objset_from_ds(ds, &os)); 1896 dsl_pool_config_exit(dp, FTAG); 1897 1898 err = zfsvfs_init(zfsvfs, os); 1899 if (err != 0) 1900 goto bail; 1901 1902 ds->ds_dir->dd_activity_cancelled = B_FALSE; 1903 VERIFY0(zfsvfs_setup(zfsvfs, B_FALSE)); 1904 1905 zfs_set_fuid_feature(zfsvfs); 1906 zfsvfs->z_rollback_time = jiffies; 1907 1908 /* 1909 * Attempt to re-establish all the active inodes with their 1910 * dbufs. If a zfs_rezget() fails, then we unhash the inode 1911 * and mark it stale. This prevents a collision if a new 1912 * inode/object is created which must use the same inode 1913 * number. The stale inode will be be released when the 1914 * VFS prunes the dentry holding the remaining references 1915 * on the stale inode. 1916 */ 1917 mutex_enter(&zfsvfs->z_znodes_lock); 1918 for (zp = list_head(&zfsvfs->z_all_znodes); zp; 1919 zp = list_next(&zfsvfs->z_all_znodes, zp)) { 1920 err2 = zfs_rezget(zp); 1921 if (err2) { 1922 zpl_d_drop_aliases(ZTOI(zp)); 1923 remove_inode_hash(ZTOI(zp)); 1924 } 1925 1926 /* see comment in zfs_suspend_fs() */ 1927 if (zp->z_suspended) { 1928 zfs_zrele_async(zp); 1929 zp->z_suspended = B_FALSE; 1930 } 1931 } 1932 mutex_exit(&zfsvfs->z_znodes_lock); 1933 1934 if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) { 1935 /* 1936 * zfs_suspend_fs() could have interrupted freeing 1937 * of dnodes. We need to restart this freeing so 1938 * that we don't "leak" the space. 1939 */ 1940 zfs_unlinked_drain(zfsvfs); 1941 } 1942 1943 /* 1944 * Most of the time zfs_suspend_fs is used for changing the contents 1945 * of the underlying dataset. ZFS rollback and receive operations 1946 * might create files for which negative dentries are present in 1947 * the cache. Since walking the dcache would require a lot of GPL-only 1948 * code duplication, it's much easier on these rather rare occasions 1949 * just to flush the whole dcache for the given dataset/filesystem. 1950 */ 1951 shrink_dcache_sb(zfsvfs->z_sb); 1952 1953 bail: 1954 if (err != 0) 1955 zfsvfs->z_unmounted = B_TRUE; 1956 1957 /* release the VFS ops */ 1958 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1959 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1960 1961 if (err != 0) { 1962 /* 1963 * Since we couldn't setup the sa framework, try to force 1964 * unmount this file system. 1965 */ 1966 if (zfsvfs->z_os) 1967 (void) zfs_umount(zfsvfs->z_sb); 1968 } 1969 return (err); 1970 } 1971 1972 /* 1973 * Release VOPs and unmount a suspended filesystem. 1974 */ 1975 int 1976 zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) 1977 { 1978 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs)); 1979 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)); 1980 1981 /* 1982 * We already own this, so just hold and rele it to update the 1983 * objset_t, as the one we had before may have been evicted. 1984 */ 1985 objset_t *os; 1986 VERIFY3P(ds->ds_owner, ==, zfsvfs); 1987 VERIFY(dsl_dataset_long_held(ds)); 1988 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds)); 1989 dsl_pool_config_enter(dp, FTAG); 1990 VERIFY0(dmu_objset_from_ds(ds, &os)); 1991 dsl_pool_config_exit(dp, FTAG); 1992 zfsvfs->z_os = os; 1993 1994 /* release the VOPs */ 1995 rw_exit(&zfsvfs->z_teardown_inactive_lock); 1996 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); 1997 1998 /* 1999 * Try to force unmount this file system. 2000 */ 2001 (void) zfs_umount(zfsvfs->z_sb); 2002 zfsvfs->z_unmounted = B_TRUE; 2003 return (0); 2004 } 2005 2006 /* 2007 * Automounted snapshots rely on periodic revalidation 2008 * to defer snapshots from being automatically unmounted. 2009 */ 2010 2011 inline void 2012 zfs_exit_fs(zfsvfs_t *zfsvfs) 2013 { 2014 if (!zfsvfs->z_issnap) 2015 return; 2016 2017 if (time_after(jiffies, zfsvfs->z_snap_defer_time + 2018 MAX(zfs_expire_snapshot * HZ / 2, HZ))) { 2019 zfsvfs->z_snap_defer_time = jiffies; 2020 zfsctl_snapshot_unmount_delay(zfsvfs->z_os->os_spa, 2021 dmu_objset_id(zfsvfs->z_os), 2022 zfs_expire_snapshot); 2023 } 2024 } 2025 2026 int 2027 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) 2028 { 2029 int error; 2030 objset_t *os = zfsvfs->z_os; 2031 dmu_tx_t *tx; 2032 2033 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) 2034 return (SET_ERROR(EINVAL)); 2035 2036 if (newvers < zfsvfs->z_version) 2037 return (SET_ERROR(EINVAL)); 2038 2039 if (zfs_spa_version_map(newvers) > 2040 spa_version(dmu_objset_spa(zfsvfs->z_os))) 2041 return (SET_ERROR(ENOTSUP)); 2042 2043 tx = dmu_tx_create(os); 2044 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); 2045 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2046 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, 2047 ZFS_SA_ATTRS); 2048 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 2049 } 2050 error = dmu_tx_assign(tx, DMU_TX_WAIT); 2051 if (error) { 2052 dmu_tx_abort(tx); 2053 return (error); 2054 } 2055 2056 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, 2057 8, 1, &newvers, tx); 2058 2059 if (error) { 2060 dmu_tx_commit(tx); 2061 return (error); 2062 } 2063 2064 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { 2065 uint64_t sa_obj; 2066 2067 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, 2068 SPA_VERSION_SA); 2069 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 2070 DMU_OT_NONE, 0, tx); 2071 2072 error = zap_add(os, MASTER_NODE_OBJ, 2073 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 2074 ASSERT0(error); 2075 2076 VERIFY0(sa_set_sa_object(os, sa_obj)); 2077 sa_register_update_callback(os, zfs_sa_upgrade); 2078 } 2079 2080 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, 2081 "from %llu to %llu", zfsvfs->z_version, newvers); 2082 2083 dmu_tx_commit(tx); 2084 2085 zfsvfs->z_version = newvers; 2086 os->os_version = newvers; 2087 2088 zfs_set_fuid_feature(zfsvfs); 2089 2090 return (0); 2091 } 2092 2093 int 2094 zfs_set_default_quota(zfsvfs_t *zfsvfs, zfs_prop_t prop, uint64_t quota) 2095 { 2096 int error; 2097 objset_t *os = zfsvfs->z_os; 2098 const char *propstr = zfs_prop_to_name(prop); 2099 dmu_tx_t *tx; 2100 2101 tx = dmu_tx_create(os); 2102 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, propstr); 2103 error = dmu_tx_assign(tx, DMU_TX_WAIT); 2104 if (error) { 2105 dmu_tx_abort(tx); 2106 return (error); 2107 } 2108 2109 if (quota == 0) { 2110 error = zap_remove(os, MASTER_NODE_OBJ, propstr, tx); 2111 if (error == ENOENT) 2112 error = 0; 2113 } else { 2114 error = zap_update(os, MASTER_NODE_OBJ, propstr, 8, 1, 2115 "a, tx); 2116 } 2117 2118 if (error) 2119 goto out; 2120 2121 switch (prop) { 2122 case ZFS_PROP_DEFAULTUSERQUOTA: 2123 zfsvfs->z_defaultuserquota = quota; 2124 break; 2125 case ZFS_PROP_DEFAULTGROUPQUOTA: 2126 zfsvfs->z_defaultgroupquota = quota; 2127 break; 2128 case ZFS_PROP_DEFAULTPROJECTQUOTA: 2129 zfsvfs->z_defaultprojectquota = quota; 2130 break; 2131 case ZFS_PROP_DEFAULTUSEROBJQUOTA: 2132 zfsvfs->z_defaultuserobjquota = quota; 2133 break; 2134 case ZFS_PROP_DEFAULTGROUPOBJQUOTA: 2135 zfsvfs->z_defaultgroupobjquota = quota; 2136 break; 2137 case ZFS_PROP_DEFAULTPROJECTOBJQUOTA: 2138 zfsvfs->z_defaultprojectobjquota = quota; 2139 break; 2140 default: 2141 break; 2142 } 2143 2144 out: 2145 dmu_tx_commit(tx); 2146 return (error); 2147 } 2148 2149 /* 2150 * Return true if the corresponding vfs's unmounted flag is set. 2151 * Otherwise return false. 2152 * If this function returns true we know VFS unmount has been initiated. 2153 */ 2154 boolean_t 2155 zfs_get_vfs_flag_unmounted(objset_t *os) 2156 { 2157 zfsvfs_t *zfvp; 2158 boolean_t unmounted = B_FALSE; 2159 2160 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS); 2161 2162 mutex_enter(&os->os_user_ptr_lock); 2163 zfvp = dmu_objset_get_user(os); 2164 if (zfvp != NULL && zfvp->z_unmounted) 2165 unmounted = B_TRUE; 2166 mutex_exit(&os->os_user_ptr_lock); 2167 2168 return (unmounted); 2169 } 2170 2171 void 2172 zfsvfs_update_fromname(const char *oldname, const char *newname) 2173 { 2174 /* 2175 * We don't need to do anything here, the devname is always current by 2176 * virtue of zfsvfs->z_sb->s_op->show_devname. 2177 */ 2178 (void) oldname, (void) newname; 2179 } 2180 2181 void 2182 zfs_init(void) 2183 { 2184 zfsctl_init(); 2185 zfs_znode_init(); 2186 dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info); 2187 register_filesystem(&zpl_fs_type); 2188 } 2189 2190 void 2191 zfs_fini(void) 2192 { 2193 /* 2194 * we don't use outstanding because zpl_posix_acl_free might add more. 2195 */ 2196 taskq_wait(system_delay_taskq); 2197 taskq_wait(system_taskq); 2198 unregister_filesystem(&zpl_fs_type); 2199 zfs_znode_fini(); 2200 zfsctl_fini(); 2201 } 2202 2203 #if defined(_KERNEL) 2204 EXPORT_SYMBOL(zfs_suspend_fs); 2205 EXPORT_SYMBOL(zfs_resume_fs); 2206 EXPORT_SYMBOL(zfs_set_version); 2207 EXPORT_SYMBOL(zfsvfs_create); 2208 EXPORT_SYMBOL(zfsvfs_free); 2209 EXPORT_SYMBOL(zfs_is_readonly); 2210 EXPORT_SYMBOL(zfs_domount); 2211 EXPORT_SYMBOL(zfs_preumount); 2212 EXPORT_SYMBOL(zfs_umount); 2213 EXPORT_SYMBOL(zfs_remount); 2214 EXPORT_SYMBOL(zfs_statvfs); 2215 EXPORT_SYMBOL(zfs_vget); 2216 EXPORT_SYMBOL(zfs_prune); 2217 EXPORT_SYMBOL(zfs_set_default_quota); 2218 #endif 2219