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, 2014 by Delphix. All rights reserved. 24 * Copyright (c) 2014 Integros [integros.com] 25 */ 26 27 /* Portions Copyright 2007 Jeremy Teo */ 28 /* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 29 30 #ifdef _KERNEL 31 #include <sys/types.h> 32 #include <sys/param.h> 33 #include <sys/time.h> 34 #include <sys/systm.h> 35 #include <sys/sysmacros.h> 36 #include <sys/resource.h> 37 #include <sys/mntent.h> 38 #include <sys/u8_textprep.h> 39 #include <sys/dsl_dataset.h> 40 #include <sys/vfs.h> 41 #include <sys/vnode.h> 42 #include <sys/file.h> 43 #include <sys/kmem.h> 44 #include <sys/errno.h> 45 #include <sys/unistd.h> 46 #include <sys/atomic.h> 47 #include <sys/zfs_dir.h> 48 #include <sys/zfs_acl.h> 49 #include <sys/zfs_ioctl.h> 50 #include <sys/zfs_rlock.h> 51 #include <sys/zfs_fuid.h> 52 #include <sys/dnode.h> 53 #include <sys/fs/zfs.h> 54 #endif /* _KERNEL */ 55 56 #include <sys/dmu.h> 57 #include <sys/dmu_objset.h> 58 #include <sys/dmu_tx.h> 59 #include <sys/zfs_refcount.h> 60 #include <sys/stat.h> 61 #include <sys/zap.h> 62 #include <sys/zfs_znode.h> 63 #include <sys/sa.h> 64 #include <sys/zfs_sa.h> 65 #include <sys/zfs_stat.h> 66 67 #include "zfs_prop.h" 68 #include "zfs_comutil.h" 69 70 /* Used by fstat(1). */ 71 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 72 SYSCTL_NULL_INT_PTR, sizeof (znode_t), "sizeof(znode_t)"); 73 74 /* 75 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 76 * turned on when DEBUG is also defined. 77 */ 78 #ifdef ZFS_DEBUG 79 #define ZNODE_STATS 80 #endif /* DEBUG */ 81 82 #ifdef ZNODE_STATS 83 #define ZNODE_STAT_ADD(stat) ((stat)++) 84 #else 85 #define ZNODE_STAT_ADD(stat) /* nothing */ 86 #endif /* ZNODE_STATS */ 87 88 /* 89 * Functions needed for userland (ie: libzpool) are not put under 90 * #ifdef_KERNEL; the rest of the functions have dependencies 91 * (such as VFS logic) that will not compile easily in userland. 92 */ 93 #ifdef _KERNEL 94 #if !defined(KMEM_DEBUG) && __FreeBSD_version >= 1300102 95 #define _ZFS_USE_SMR 96 static uma_zone_t znode_uma_zone; 97 #else 98 static kmem_cache_t *znode_cache = NULL; 99 #endif 100 101 extern struct vop_vector zfs_vnodeops; 102 extern struct vop_vector zfs_fifoops; 103 extern struct vop_vector zfs_shareops; 104 105 106 /* 107 * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on 108 * z_rangelock. It will modify the offset and length of the lock to reflect 109 * znode-specific information, and convert RL_APPEND to RL_WRITER. This is 110 * called with the rangelock_t's rl_lock held, which avoids races. 111 */ 112 static void 113 zfs_rangelock_cb(zfs_locked_range_t *new, void *arg) 114 { 115 znode_t *zp = arg; 116 117 /* 118 * If in append mode, convert to writer and lock starting at the 119 * current end of file. 120 */ 121 if (new->lr_type == RL_APPEND) { 122 new->lr_offset = zp->z_size; 123 new->lr_type = RL_WRITER; 124 } 125 126 /* 127 * If we need to grow the block size then lock the whole file range. 128 */ 129 uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length); 130 if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) || 131 zp->z_blksz < ZTOZSB(zp)->z_max_blksz)) { 132 new->lr_offset = 0; 133 new->lr_length = UINT64_MAX; 134 } 135 } 136 137 static int 138 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 139 { 140 znode_t *zp = buf; 141 142 POINTER_INVALIDATE(&zp->z_zfsvfs); 143 144 list_link_init(&zp->z_link_node); 145 146 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 147 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 148 rw_init(&zp->z_xattr_lock, NULL, RW_DEFAULT, NULL); 149 150 zfs_rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp); 151 152 zp->z_acl_cached = NULL; 153 zp->z_xattr_cached = NULL; 154 zp->z_xattr_parent = 0; 155 zp->z_vnode = NULL; 156 zp->z_sync_writes_cnt = 0; 157 zp->z_async_writes_cnt = 0; 158 159 return (0); 160 } 161 162 static void 163 zfs_znode_cache_destructor(void *buf, void *arg) 164 { 165 (void) arg; 166 znode_t *zp = buf; 167 168 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 169 ASSERT3P(zp->z_vnode, ==, NULL); 170 ASSERT(!list_link_active(&zp->z_link_node)); 171 mutex_destroy(&zp->z_lock); 172 mutex_destroy(&zp->z_acl_lock); 173 rw_destroy(&zp->z_xattr_lock); 174 zfs_rangelock_fini(&zp->z_rangelock); 175 176 ASSERT3P(zp->z_acl_cached, ==, NULL); 177 ASSERT3P(zp->z_xattr_cached, ==, NULL); 178 179 ASSERT0(atomic_load_32(&zp->z_sync_writes_cnt)); 180 ASSERT0(atomic_load_32(&zp->z_async_writes_cnt)); 181 } 182 183 184 #ifdef _ZFS_USE_SMR 185 VFS_SMR_DECLARE; 186 187 static int 188 zfs_znode_cache_constructor_smr(void *mem, int size __unused, void *private, 189 int flags) 190 { 191 return (zfs_znode_cache_constructor(mem, private, flags)); 192 } 193 194 static void 195 zfs_znode_cache_destructor_smr(void *mem, int size __unused, void *private) 196 { 197 zfs_znode_cache_destructor(mem, private); 198 } 199 200 void 201 zfs_znode_init(void) 202 { 203 /* 204 * Initialize zcache 205 */ 206 ASSERT3P(znode_uma_zone, ==, NULL); 207 znode_uma_zone = uma_zcreate("zfs_znode_cache", 208 sizeof (znode_t), zfs_znode_cache_constructor_smr, 209 zfs_znode_cache_destructor_smr, NULL, NULL, 0, 0); 210 VFS_SMR_ZONE_SET(znode_uma_zone); 211 } 212 213 static znode_t * 214 zfs_znode_alloc_kmem(int flags) 215 { 216 return (uma_zalloc_smr(znode_uma_zone, flags)); 217 } 218 219 static void 220 zfs_znode_free_kmem(znode_t *zp) 221 { 222 if (zp->z_xattr_cached) { 223 nvlist_free(zp->z_xattr_cached); 224 zp->z_xattr_cached = NULL; 225 } 226 uma_zfree_smr(znode_uma_zone, zp); 227 } 228 #else 229 void 230 zfs_znode_init(void) 231 { 232 /* 233 * Initialize zcache 234 */ 235 ASSERT3P(znode_cache, ==, NULL); 236 znode_cache = kmem_cache_create("zfs_znode_cache", 237 sizeof (znode_t), 0, zfs_znode_cache_constructor, 238 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 239 } 240 241 static znode_t * 242 zfs_znode_alloc_kmem(int flags) 243 { 244 return (kmem_cache_alloc(znode_cache, flags)); 245 } 246 247 static void 248 zfs_znode_free_kmem(znode_t *zp) 249 { 250 if (zp->z_xattr_cached) { 251 nvlist_free(zp->z_xattr_cached); 252 zp->z_xattr_cached = NULL; 253 } 254 kmem_cache_free(znode_cache, zp); 255 } 256 #endif 257 258 void 259 zfs_znode_fini(void) 260 { 261 /* 262 * Cleanup zcache 263 */ 264 #ifdef _ZFS_USE_SMR 265 if (znode_uma_zone) { 266 uma_zdestroy(znode_uma_zone); 267 znode_uma_zone = NULL; 268 } 269 #else 270 if (znode_cache) { 271 kmem_cache_destroy(znode_cache); 272 znode_cache = NULL; 273 } 274 #endif 275 } 276 277 278 static int 279 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 280 { 281 zfs_acl_ids_t acl_ids; 282 vattr_t vattr; 283 znode_t *sharezp; 284 znode_t *zp; 285 int error; 286 287 vattr.va_mask = AT_MODE|AT_UID|AT_GID; 288 vattr.va_type = VDIR; 289 vattr.va_mode = S_IFDIR|0555; 290 vattr.va_uid = crgetuid(kcred); 291 vattr.va_gid = crgetgid(kcred); 292 293 sharezp = zfs_znode_alloc_kmem(KM_SLEEP); 294 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 295 sharezp->z_unlinked = 0; 296 sharezp->z_atime_dirty = 0; 297 sharezp->z_zfsvfs = zfsvfs; 298 sharezp->z_is_sa = zfsvfs->z_use_sa; 299 300 VERIFY0(zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 301 kcred, NULL, &acl_ids, NULL)); 302 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 303 ASSERT3P(zp, ==, sharezp); 304 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 305 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 306 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 307 zfsvfs->z_shares_dir = sharezp->z_id; 308 309 zfs_acl_ids_free(&acl_ids); 310 sa_handle_destroy(sharezp->z_sa_hdl); 311 zfs_znode_free_kmem(sharezp); 312 313 return (error); 314 } 315 316 /* 317 * define a couple of values we need available 318 * for both 64 and 32 bit environments. 319 */ 320 #ifndef NBITSMINOR64 321 #define NBITSMINOR64 32 322 #endif 323 #ifndef MAXMAJ64 324 #define MAXMAJ64 0xffffffffUL 325 #endif 326 #ifndef MAXMIN64 327 #define MAXMIN64 0xffffffffUL 328 #endif 329 330 /* 331 * Create special expldev for ZFS private use. 332 * Can't use standard expldev since it doesn't do 333 * what we want. The standard expldev() takes a 334 * dev32_t in LP64 and expands it to a long dev_t. 335 * We need an interface that takes a dev32_t in ILP32 336 * and expands it to a long dev_t. 337 */ 338 static uint64_t 339 zfs_expldev(dev_t dev) 340 { 341 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 342 } 343 /* 344 * Special cmpldev for ZFS private use. 345 * Can't use standard cmpldev since it takes 346 * a long dev_t and compresses it to dev32_t in 347 * LP64. We need to do a compaction of a long dev_t 348 * to a dev32_t in ILP32. 349 */ 350 dev_t 351 zfs_cmpldev(uint64_t dev) 352 { 353 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 354 } 355 356 static void 357 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 358 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 359 { 360 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 361 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 362 363 ASSERT3P(zp->z_sa_hdl, ==, NULL); 364 ASSERT3P(zp->z_acl_cached, ==, NULL); 365 if (sa_hdl == NULL) { 366 VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, zp, 367 SA_HDL_SHARED, &zp->z_sa_hdl)); 368 } else { 369 zp->z_sa_hdl = sa_hdl; 370 sa_set_userp(sa_hdl, zp); 371 } 372 373 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 374 375 /* 376 * Slap on VROOT if we are the root znode unless we are the root 377 * node of a snapshot mounted under .zfs. 378 */ 379 if (zp->z_id == zfsvfs->z_root && zfsvfs->z_parent == zfsvfs) 380 ZTOV(zp)->v_flag |= VROOT; 381 382 vn_exists(ZTOV(zp)); 383 } 384 385 void 386 zfs_znode_dmu_fini(znode_t *zp) 387 { 388 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 389 ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zp->z_zfsvfs)); 390 391 sa_handle_destroy(zp->z_sa_hdl); 392 zp->z_sa_hdl = NULL; 393 } 394 395 static void 396 zfs_vnode_forget(vnode_t *vp) 397 { 398 399 /* copied from insmntque_stddtr */ 400 vp->v_data = NULL; 401 vp->v_op = &dead_vnodeops; 402 vgone(vp); 403 vput(vp); 404 } 405 406 /* 407 * Construct a new znode/vnode and initialize. 408 * 409 * This does not do a call to dmu_set_user() that is 410 * up to the caller to do, in case you don't want to 411 * return the znode 412 */ 413 static znode_t * 414 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 415 dmu_object_type_t obj_type, sa_handle_t *hdl) 416 { 417 znode_t *zp; 418 vnode_t *vp; 419 uint64_t mode; 420 uint64_t parent; 421 #ifdef notyet 422 uint64_t mtime[2], ctime[2]; 423 #endif 424 uint64_t projid = ZFS_DEFAULT_PROJID; 425 sa_bulk_attr_t bulk[9]; 426 int count = 0; 427 int error; 428 429 zp = zfs_znode_alloc_kmem(KM_SLEEP); 430 431 #ifndef _ZFS_USE_SMR 432 KASSERT((zfsvfs->z_parent->z_vfs->mnt_kern_flag & MNTK_FPLOOKUP) == 0, 433 ("%s: fast path lookup enabled without smr", __func__)); 434 #endif 435 436 #if __FreeBSD_version >= 1300076 437 KASSERT(curthread->td_vp_reserved != NULL, 438 ("zfs_znode_alloc: getnewvnode without any vnodes reserved")); 439 #else 440 KASSERT(curthread->td_vp_reserv > 0, 441 ("zfs_znode_alloc: getnewvnode without any vnodes reserved")); 442 #endif 443 error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp); 444 if (error != 0) { 445 zfs_znode_free_kmem(zp); 446 return (NULL); 447 } 448 zp->z_vnode = vp; 449 vp->v_data = zp; 450 451 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 452 453 zp->z_sa_hdl = NULL; 454 zp->z_unlinked = 0; 455 zp->z_atime_dirty = 0; 456 zp->z_mapcnt = 0; 457 zp->z_id = db->db_object; 458 zp->z_blksz = blksz; 459 zp->z_seq = 0x7A4653; 460 zp->z_sync_cnt = 0; 461 zp->z_sync_writes_cnt = 0; 462 zp->z_async_writes_cnt = 0; 463 #if __FreeBSD_version >= 1300139 464 atomic_store_ptr(&zp->z_cached_symlink, NULL); 465 #endif 466 467 vp = ZTOV(zp); 468 469 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 470 471 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 472 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 473 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 474 &zp->z_size, 8); 475 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 476 &zp->z_links, 8); 477 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 478 &zp->z_pflags, 8); 479 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 480 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 481 &zp->z_atime, 16); 482 #ifdef notyet 483 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 484 &mtime, 16); 485 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 486 &ctime, 16); 487 #endif 488 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 489 &zp->z_uid, 8); 490 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 491 &zp->z_gid, 8); 492 493 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0 || 494 (dmu_objset_projectquota_enabled(zfsvfs->z_os) && 495 (zp->z_pflags & ZFS_PROJID) && 496 sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) { 497 if (hdl == NULL) 498 sa_handle_destroy(zp->z_sa_hdl); 499 zfs_vnode_forget(vp); 500 zp->z_vnode = NULL; 501 zfs_znode_free_kmem(zp); 502 return (NULL); 503 } 504 505 zp->z_projid = projid; 506 zp->z_mode = mode; 507 508 /* Cache the xattr parent id */ 509 if (zp->z_pflags & ZFS_XATTR) 510 zp->z_xattr_parent = parent; 511 512 vp->v_type = IFTOVT((mode_t)mode); 513 514 switch (vp->v_type) { 515 case VDIR: 516 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 517 break; 518 case VFIFO: 519 vp->v_op = &zfs_fifoops; 520 break; 521 case VREG: 522 if (parent == zfsvfs->z_shares_dir) { 523 ASSERT0(zp->z_uid); 524 ASSERT0(zp->z_gid); 525 vp->v_op = &zfs_shareops; 526 } 527 break; 528 default: 529 break; 530 } 531 532 mutex_enter(&zfsvfs->z_znodes_lock); 533 list_insert_tail(&zfsvfs->z_all_znodes, zp); 534 zfsvfs->z_nr_znodes++; 535 zp->z_zfsvfs = zfsvfs; 536 mutex_exit(&zfsvfs->z_znodes_lock); 537 538 /* 539 * Acquire vnode lock before making it available to the world. 540 */ 541 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 542 #if __FreeBSD_version >= 1400077 543 vn_set_state(vp, VSTATE_CONSTRUCTED); 544 #endif 545 VN_LOCK_AREC(vp); 546 if (vp->v_type != VFIFO) 547 VN_LOCK_ASHARE(vp); 548 549 return (zp); 550 } 551 552 static uint64_t empty_xattr; 553 static uint64_t pad[4]; 554 static zfs_acl_phys_t acl_phys; 555 /* 556 * Create a new DMU object to hold a zfs znode. 557 * 558 * IN: dzp - parent directory for new znode 559 * vap - file attributes for new znode 560 * tx - dmu transaction id for zap operations 561 * cr - credentials of caller 562 * flag - flags: 563 * IS_ROOT_NODE - new object will be root 564 * IS_XATTR - new object is an attribute 565 * bonuslen - length of bonus buffer 566 * setaclp - File/Dir initial ACL 567 * fuidp - Tracks fuid allocation. 568 * 569 * OUT: zpp - allocated znode 570 * 571 */ 572 void 573 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 574 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 575 { 576 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 577 uint64_t mode, size, links, parent, pflags; 578 uint64_t dzp_pflags = 0; 579 uint64_t rdev = 0; 580 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 581 dmu_buf_t *db; 582 timestruc_t now; 583 uint64_t gen, obj; 584 int bonuslen; 585 int dnodesize; 586 sa_handle_t *sa_hdl; 587 dmu_object_type_t obj_type; 588 sa_bulk_attr_t *sa_attrs; 589 int cnt = 0; 590 zfs_acl_locator_cb_t locate = { 0 }; 591 592 ASSERT3P(vap, !=, NULL); 593 ASSERT3U((vap->va_mask & AT_MODE), ==, AT_MODE); 594 595 if (zfsvfs->z_replay) { 596 obj = vap->va_nodeid; 597 now = vap->va_ctime; /* see zfs_replay_create() */ 598 gen = vap->va_nblocks; /* ditto */ 599 dnodesize = vap->va_fsid; /* ditto */ 600 } else { 601 obj = 0; 602 vfs_timestamp(&now); 603 gen = dmu_tx_get_txg(tx); 604 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os); 605 } 606 607 if (dnodesize == 0) 608 dnodesize = DNODE_MIN_SIZE; 609 610 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 611 bonuslen = (obj_type == DMU_OT_SA) ? 612 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE; 613 614 /* 615 * Create a new DMU object. 616 */ 617 /* 618 * There's currently no mechanism for pre-reading the blocks that will 619 * be needed to allocate a new object, so we accept the small chance 620 * that there will be an i/o error and we will fail one of the 621 * assertions below. 622 */ 623 if (vap->va_type == VDIR) { 624 if (zfsvfs->z_replay) { 625 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj, 626 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 627 obj_type, bonuslen, dnodesize, tx)); 628 } else { 629 obj = zap_create_norm_dnsize(zfsvfs->z_os, 630 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 631 obj_type, bonuslen, dnodesize, tx); 632 } 633 } else { 634 if (zfsvfs->z_replay) { 635 VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj, 636 DMU_OT_PLAIN_FILE_CONTENTS, 0, 637 obj_type, bonuslen, dnodesize, tx)); 638 } else { 639 obj = dmu_object_alloc_dnsize(zfsvfs->z_os, 640 DMU_OT_PLAIN_FILE_CONTENTS, 0, 641 obj_type, bonuslen, dnodesize, tx); 642 } 643 } 644 645 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 646 VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 647 648 /* 649 * If this is the root, fix up the half-initialized parent pointer 650 * to reference the just-allocated physical data area. 651 */ 652 if (flag & IS_ROOT_NODE) { 653 dzp->z_id = obj; 654 } else { 655 dzp_pflags = dzp->z_pflags; 656 } 657 658 /* 659 * If parent is an xattr, so am I. 660 */ 661 if (dzp_pflags & ZFS_XATTR) { 662 flag |= IS_XATTR; 663 } 664 665 if (zfsvfs->z_use_fuids) 666 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 667 else 668 pflags = 0; 669 670 if (vap->va_type == VDIR) { 671 size = 2; /* contents ("." and "..") */ 672 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 673 } else { 674 size = links = 0; 675 } 676 677 if (vap->va_type == VBLK || vap->va_type == VCHR) { 678 rdev = zfs_expldev(vap->va_rdev); 679 } 680 681 parent = dzp->z_id; 682 mode = acl_ids->z_mode; 683 if (flag & IS_XATTR) 684 pflags |= ZFS_XATTR; 685 686 /* 687 * No execs denied will be determined when zfs_mode_compute() is called. 688 */ 689 pflags |= acl_ids->z_aclp->z_hints & 690 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 691 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 692 693 ZFS_TIME_ENCODE(&now, crtime); 694 ZFS_TIME_ENCODE(&now, ctime); 695 696 if (vap->va_mask & AT_ATIME) { 697 ZFS_TIME_ENCODE(&vap->va_atime, atime); 698 } else { 699 ZFS_TIME_ENCODE(&now, atime); 700 } 701 702 if (vap->va_mask & AT_MTIME) { 703 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 704 } else { 705 ZFS_TIME_ENCODE(&now, mtime); 706 } 707 708 /* Now add in all of the "SA" attributes */ 709 VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 710 &sa_hdl)); 711 712 /* 713 * Setup the array of attributes to be replaced/set on the new file 714 * 715 * order for DMU_OT_ZNODE is critical since it needs to be constructed 716 * in the old znode_phys_t format. Don't change this ordering 717 */ 718 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP); 719 720 if (obj_type == DMU_OT_ZNODE) { 721 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 722 NULL, &atime, 16); 723 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 724 NULL, &mtime, 16); 725 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 726 NULL, &ctime, 16); 727 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 728 NULL, &crtime, 16); 729 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 730 NULL, &gen, 8); 731 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 732 NULL, &mode, 8); 733 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 734 NULL, &size, 8); 735 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 736 NULL, &parent, 8); 737 } else { 738 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 739 NULL, &mode, 8); 740 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 741 NULL, &size, 8); 742 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 743 NULL, &gen, 8); 744 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), 745 NULL, &acl_ids->z_fuid, 8); 746 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), 747 NULL, &acl_ids->z_fgid, 8); 748 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 749 NULL, &parent, 8); 750 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 751 NULL, &pflags, 8); 752 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 753 NULL, &atime, 16); 754 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 755 NULL, &mtime, 16); 756 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 757 NULL, &ctime, 16); 758 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 759 NULL, &crtime, 16); 760 } 761 762 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 763 764 if (obj_type == DMU_OT_ZNODE) { 765 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 766 &empty_xattr, 8); 767 } 768 if (obj_type == DMU_OT_ZNODE || 769 (vap->va_type == VBLK || vap->va_type == VCHR)) { 770 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 771 NULL, &rdev, 8); 772 773 } 774 if (obj_type == DMU_OT_ZNODE) { 775 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 776 NULL, &pflags, 8); 777 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 778 &acl_ids->z_fuid, 8); 779 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 780 &acl_ids->z_fgid, 8); 781 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 782 sizeof (uint64_t) * 4); 783 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 784 &acl_phys, sizeof (zfs_acl_phys_t)); 785 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 786 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 787 &acl_ids->z_aclp->z_acl_count, 8); 788 locate.cb_aclp = acl_ids->z_aclp; 789 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 790 zfs_acl_data_locator, &locate, 791 acl_ids->z_aclp->z_acl_bytes); 792 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 793 acl_ids->z_fuid, acl_ids->z_fgid); 794 } 795 796 VERIFY0(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx)); 797 798 if (!(flag & IS_ROOT_NODE)) { 799 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 800 ASSERT3P(*zpp, !=, NULL); 801 } else { 802 /* 803 * If we are creating the root node, the "parent" we 804 * passed in is the znode for the root. 805 */ 806 *zpp = dzp; 807 808 (*zpp)->z_sa_hdl = sa_hdl; 809 } 810 811 (*zpp)->z_pflags = pflags; 812 (*zpp)->z_mode = mode; 813 (*zpp)->z_dnodesize = dnodesize; 814 815 if (vap->va_mask & AT_XVATTR) 816 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 817 818 if (obj_type == DMU_OT_ZNODE || 819 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 820 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx)); 821 } 822 if (!(flag & IS_ROOT_NODE)) { 823 vnode_t *vp = ZTOV(*zpp); 824 vp->v_vflag |= VV_FORCEINSMQ; 825 int err = insmntque(vp, zfsvfs->z_vfs); 826 vp->v_vflag &= ~VV_FORCEINSMQ; 827 (void) err; 828 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 829 } 830 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END); 831 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 832 } 833 834 /* 835 * Update in-core attributes. It is assumed the caller will be doing an 836 * sa_bulk_update to push the changes out. 837 */ 838 void 839 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 840 { 841 xoptattr_t *xoap; 842 843 xoap = xva_getxoptattr(xvap); 844 ASSERT3P(xoap, !=, NULL); 845 846 if (zp->z_zfsvfs->z_replay == B_FALSE) { 847 ASSERT_VOP_IN_SEQC(ZTOV(zp)); 848 } 849 850 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 851 uint64_t times[2]; 852 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 853 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 854 ×, sizeof (times), tx); 855 XVA_SET_RTN(xvap, XAT_CREATETIME); 856 } 857 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 858 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 859 zp->z_pflags, tx); 860 XVA_SET_RTN(xvap, XAT_READONLY); 861 } 862 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 863 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 864 zp->z_pflags, tx); 865 XVA_SET_RTN(xvap, XAT_HIDDEN); 866 } 867 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 868 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 869 zp->z_pflags, tx); 870 XVA_SET_RTN(xvap, XAT_SYSTEM); 871 } 872 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 873 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 874 zp->z_pflags, tx); 875 XVA_SET_RTN(xvap, XAT_ARCHIVE); 876 } 877 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 878 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 879 zp->z_pflags, tx); 880 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 881 } 882 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 883 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 884 zp->z_pflags, tx); 885 XVA_SET_RTN(xvap, XAT_NOUNLINK); 886 } 887 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 888 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 889 zp->z_pflags, tx); 890 XVA_SET_RTN(xvap, XAT_APPENDONLY); 891 } 892 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 893 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 894 zp->z_pflags, tx); 895 XVA_SET_RTN(xvap, XAT_NODUMP); 896 } 897 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 898 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 899 zp->z_pflags, tx); 900 XVA_SET_RTN(xvap, XAT_OPAQUE); 901 } 902 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 903 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 904 xoap->xoa_av_quarantined, zp->z_pflags, tx); 905 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 906 } 907 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 908 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 909 zp->z_pflags, tx); 910 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 911 } 912 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 913 zfs_sa_set_scanstamp(zp, xvap, tx); 914 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 915 } 916 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 917 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 918 zp->z_pflags, tx); 919 XVA_SET_RTN(xvap, XAT_REPARSE); 920 } 921 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 922 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 923 zp->z_pflags, tx); 924 XVA_SET_RTN(xvap, XAT_OFFLINE); 925 } 926 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 927 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 928 zp->z_pflags, tx); 929 XVA_SET_RTN(xvap, XAT_SPARSE); 930 } 931 } 932 933 int 934 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 935 { 936 dmu_object_info_t doi; 937 dmu_buf_t *db; 938 znode_t *zp; 939 vnode_t *vp; 940 sa_handle_t *hdl; 941 int locked; 942 int err; 943 944 getnewvnode_reserve_(); 945 again: 946 *zpp = NULL; 947 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 948 949 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 950 if (err) { 951 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 952 getnewvnode_drop_reserve(); 953 return (err); 954 } 955 956 dmu_object_info_from_db(db, &doi); 957 if (doi.doi_bonus_type != DMU_OT_SA && 958 (doi.doi_bonus_type != DMU_OT_ZNODE || 959 (doi.doi_bonus_type == DMU_OT_ZNODE && 960 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 961 sa_buf_rele(db, NULL); 962 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 963 getnewvnode_drop_reserve(); 964 return (SET_ERROR(EINVAL)); 965 } 966 967 hdl = dmu_buf_get_user(db); 968 if (hdl != NULL) { 969 zp = sa_get_userdata(hdl); 970 971 /* 972 * Since "SA" does immediate eviction we 973 * should never find a sa handle that doesn't 974 * know about the znode. 975 */ 976 ASSERT3P(zp, !=, NULL); 977 ASSERT3U(zp->z_id, ==, obj_num); 978 if (zp->z_unlinked) { 979 err = SET_ERROR(ENOENT); 980 } else { 981 vp = ZTOV(zp); 982 /* 983 * Don't let the vnode disappear after 984 * ZFS_OBJ_HOLD_EXIT. 985 */ 986 VN_HOLD(vp); 987 *zpp = zp; 988 err = 0; 989 } 990 991 sa_buf_rele(db, NULL); 992 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 993 994 if (err) { 995 getnewvnode_drop_reserve(); 996 return (err); 997 } 998 999 locked = VOP_ISLOCKED(vp); 1000 VI_LOCK(vp); 1001 if (VN_IS_DOOMED(vp) && locked != LK_EXCLUSIVE) { 1002 /* 1003 * The vnode is doomed and this thread doesn't 1004 * hold the exclusive lock on it, so the vnode 1005 * must be being reclaimed by another thread. 1006 * Otherwise the doomed vnode is being reclaimed 1007 * by this thread and zfs_zget is called from 1008 * ZIL internals. 1009 */ 1010 VI_UNLOCK(vp); 1011 1012 /* 1013 * XXX vrele() locks the vnode when the last reference 1014 * is dropped. Although in this case the vnode is 1015 * doomed / dead and so no inactivation is required, 1016 * the vnode lock is still acquired. That could result 1017 * in a LOR with z_teardown_lock if another thread holds 1018 * the vnode's lock and tries to take z_teardown_lock. 1019 * But that is only possible if the other thread peforms 1020 * a ZFS vnode operation on the vnode. That either 1021 * should not happen if the vnode is dead or the thread 1022 * should also have a reference to the vnode and thus 1023 * our reference is not last. 1024 */ 1025 VN_RELE(vp); 1026 goto again; 1027 } 1028 VI_UNLOCK(vp); 1029 getnewvnode_drop_reserve(); 1030 return (err); 1031 } 1032 1033 /* 1034 * Not found create new znode/vnode 1035 * but only if file exists. 1036 * 1037 * There is a small window where zfs_vget() could 1038 * find this object while a file create is still in 1039 * progress. This is checked for in zfs_znode_alloc() 1040 * 1041 * if zfs_znode_alloc() fails it will drop the hold on the 1042 * bonus buffer. 1043 */ 1044 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1045 doi.doi_bonus_type, NULL); 1046 if (zp == NULL) { 1047 err = SET_ERROR(ENOENT); 1048 } else { 1049 *zpp = zp; 1050 } 1051 if (err == 0) { 1052 vnode_t *vp = ZTOV(zp); 1053 1054 err = insmntque(vp, zfsvfs->z_vfs); 1055 if (err == 0) { 1056 vp->v_hash = obj_num; 1057 VOP_UNLOCK1(vp); 1058 } else { 1059 zp->z_vnode = NULL; 1060 zfs_znode_dmu_fini(zp); 1061 zfs_znode_free(zp); 1062 *zpp = NULL; 1063 } 1064 } 1065 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1066 getnewvnode_drop_reserve(); 1067 return (err); 1068 } 1069 1070 int 1071 zfs_rezget(znode_t *zp) 1072 { 1073 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1074 dmu_object_info_t doi; 1075 dmu_buf_t *db; 1076 vnode_t *vp; 1077 uint64_t obj_num = zp->z_id; 1078 uint64_t mode, size; 1079 sa_bulk_attr_t bulk[8]; 1080 int err; 1081 int count = 0; 1082 uint64_t gen; 1083 1084 /* 1085 * Remove cached pages before reloading the znode, so that they are not 1086 * lingering after we run into any error. Ideally, we should vgone() 1087 * the vnode in case of error, but currently we cannot do that 1088 * because of the LOR between the vnode lock and z_teardown_lock. 1089 * So, instead, we have to "doom" the znode in the illumos style. 1090 * 1091 * Ignore invalid pages during the scan. This is to avoid deadlocks 1092 * between page busying and the teardown lock, as pages are busied prior 1093 * to a VOP_GETPAGES operation, which acquires the teardown read lock. 1094 * Such pages will be invalid and can safely be skipped here. 1095 */ 1096 vp = ZTOV(zp); 1097 #if __FreeBSD_version >= 1400042 1098 vn_pages_remove_valid(vp, 0, 0); 1099 #else 1100 vn_pages_remove(vp, 0, 0); 1101 #endif 1102 1103 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1104 1105 mutex_enter(&zp->z_acl_lock); 1106 if (zp->z_acl_cached) { 1107 zfs_acl_free(zp->z_acl_cached); 1108 zp->z_acl_cached = NULL; 1109 } 1110 mutex_exit(&zp->z_acl_lock); 1111 1112 rw_enter(&zp->z_xattr_lock, RW_WRITER); 1113 if (zp->z_xattr_cached) { 1114 nvlist_free(zp->z_xattr_cached); 1115 zp->z_xattr_cached = NULL; 1116 } 1117 rw_exit(&zp->z_xattr_lock); 1118 1119 ASSERT3P(zp->z_sa_hdl, ==, NULL); 1120 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1121 if (err) { 1122 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1123 return (err); 1124 } 1125 1126 dmu_object_info_from_db(db, &doi); 1127 if (doi.doi_bonus_type != DMU_OT_SA && 1128 (doi.doi_bonus_type != DMU_OT_ZNODE || 1129 (doi.doi_bonus_type == DMU_OT_ZNODE && 1130 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1131 sa_buf_rele(db, NULL); 1132 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1133 return (SET_ERROR(EINVAL)); 1134 } 1135 1136 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1137 size = zp->z_size; 1138 1139 /* reload cached values */ 1140 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1141 &gen, sizeof (gen)); 1142 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1143 &zp->z_size, sizeof (zp->z_size)); 1144 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1145 &zp->z_links, sizeof (zp->z_links)); 1146 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1147 &zp->z_pflags, sizeof (zp->z_pflags)); 1148 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1149 &zp->z_atime, sizeof (zp->z_atime)); 1150 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1151 &zp->z_uid, sizeof (zp->z_uid)); 1152 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1153 &zp->z_gid, sizeof (zp->z_gid)); 1154 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1155 &mode, sizeof (mode)); 1156 1157 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1158 zfs_znode_dmu_fini(zp); 1159 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1160 return (SET_ERROR(EIO)); 1161 } 1162 1163 zp->z_mode = mode; 1164 1165 if (gen != zp->z_gen) { 1166 zfs_znode_dmu_fini(zp); 1167 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1168 return (SET_ERROR(EIO)); 1169 } 1170 1171 /* 1172 * It is highly improbable but still quite possible that two 1173 * objects in different datasets are created with the same 1174 * object numbers and in transaction groups with the same 1175 * numbers. znodes corresponding to those objects would 1176 * have the same z_id and z_gen, but their other attributes 1177 * may be different. 1178 * zfs recv -F may replace one of such objects with the other. 1179 * As a result file properties recorded in the replaced 1180 * object's vnode may no longer match the received object's 1181 * properties. At present the only cached property is the 1182 * files type recorded in v_type. 1183 * So, handle this case by leaving the old vnode and znode 1184 * disassociated from the actual object. A new vnode and a 1185 * znode will be created if the object is accessed 1186 * (e.g. via a look-up). The old vnode and znode will be 1187 * recycled when the last vnode reference is dropped. 1188 */ 1189 if (vp->v_type != IFTOVT((mode_t)zp->z_mode)) { 1190 zfs_znode_dmu_fini(zp); 1191 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1192 return (SET_ERROR(EIO)); 1193 } 1194 1195 /* 1196 * If the file has zero links, then it has been unlinked on the send 1197 * side and it must be in the received unlinked set. 1198 * We call zfs_znode_dmu_fini() now to prevent any accesses to the 1199 * stale data and to prevent automatically removal of the file in 1200 * zfs_zinactive(). The file will be removed either when it is removed 1201 * on the send side and the next incremental stream is received or 1202 * when the unlinked set gets processed. 1203 */ 1204 zp->z_unlinked = (zp->z_links == 0); 1205 if (zp->z_unlinked) { 1206 zfs_znode_dmu_fini(zp); 1207 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1208 return (0); 1209 } 1210 1211 zp->z_blksz = doi.doi_data_block_size; 1212 if (zp->z_size != size) 1213 vnode_pager_setsize(vp, zp->z_size); 1214 1215 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1216 1217 return (0); 1218 } 1219 1220 void 1221 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1222 { 1223 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1224 objset_t *os = zfsvfs->z_os; 1225 uint64_t obj = zp->z_id; 1226 uint64_t acl_obj = zfs_external_acl(zp); 1227 1228 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1229 if (acl_obj) { 1230 VERIFY(!zp->z_is_sa); 1231 VERIFY0(dmu_object_free(os, acl_obj, tx)); 1232 } 1233 VERIFY0(dmu_object_free(os, obj, tx)); 1234 zfs_znode_dmu_fini(zp); 1235 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1236 zfs_znode_free(zp); 1237 } 1238 1239 void 1240 zfs_zinactive(znode_t *zp) 1241 { 1242 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1243 uint64_t z_id = zp->z_id; 1244 1245 ASSERT3P(zp->z_sa_hdl, !=, NULL); 1246 1247 /* 1248 * Don't allow a zfs_zget() while were trying to release this znode 1249 */ 1250 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1251 1252 /* 1253 * If this was the last reference to a file with no links, remove 1254 * the file from the file system unless the file system is mounted 1255 * read-only. That can happen, for example, if the file system was 1256 * originally read-write, the file was opened, then unlinked and 1257 * the file system was made read-only before the file was finally 1258 * closed. The file will remain in the unlinked set. 1259 */ 1260 if (zp->z_unlinked) { 1261 ASSERT(!zfsvfs->z_issnap); 1262 if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) { 1263 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1264 zfs_rmnode(zp); 1265 return; 1266 } 1267 } 1268 1269 zfs_znode_dmu_fini(zp); 1270 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1271 zfs_znode_free(zp); 1272 } 1273 1274 void 1275 zfs_znode_free(znode_t *zp) 1276 { 1277 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1278 #if __FreeBSD_version >= 1300139 1279 char *symlink; 1280 #endif 1281 1282 ASSERT3P(zp->z_sa_hdl, ==, NULL); 1283 zp->z_vnode = NULL; 1284 mutex_enter(&zfsvfs->z_znodes_lock); 1285 POINTER_INVALIDATE(&zp->z_zfsvfs); 1286 list_remove(&zfsvfs->z_all_znodes, zp); 1287 zfsvfs->z_nr_znodes--; 1288 mutex_exit(&zfsvfs->z_znodes_lock); 1289 1290 #if __FreeBSD_version >= 1300139 1291 symlink = atomic_load_ptr(&zp->z_cached_symlink); 1292 if (symlink != NULL) { 1293 atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink, 1294 (uintptr_t)NULL); 1295 cache_symlink_free(symlink, strlen(symlink) + 1); 1296 } 1297 #endif 1298 1299 if (zp->z_acl_cached) { 1300 zfs_acl_free(zp->z_acl_cached); 1301 zp->z_acl_cached = NULL; 1302 } 1303 1304 zfs_znode_free_kmem(zp); 1305 } 1306 1307 void 1308 zfs_tstamp_update_setup_ext(znode_t *zp, uint_t flag, uint64_t mtime[2], 1309 uint64_t ctime[2], boolean_t have_tx) 1310 { 1311 timestruc_t now; 1312 1313 vfs_timestamp(&now); 1314 1315 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1316 zp->z_atime_dirty = 0; 1317 zp->z_seq++; 1318 } else { 1319 zp->z_atime_dirty = 1; 1320 } 1321 1322 if (flag & AT_ATIME) { 1323 ZFS_TIME_ENCODE(&now, zp->z_atime); 1324 } 1325 1326 if (flag & AT_MTIME) { 1327 ZFS_TIME_ENCODE(&now, mtime); 1328 if (zp->z_zfsvfs->z_use_fuids) { 1329 zp->z_pflags |= (ZFS_ARCHIVE | 1330 ZFS_AV_MODIFIED); 1331 } 1332 } 1333 1334 if (flag & AT_CTIME) { 1335 ZFS_TIME_ENCODE(&now, ctime); 1336 if (zp->z_zfsvfs->z_use_fuids) 1337 zp->z_pflags |= ZFS_ARCHIVE; 1338 } 1339 } 1340 1341 1342 void 1343 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1344 uint64_t ctime[2]) 1345 { 1346 zfs_tstamp_update_setup_ext(zp, flag, mtime, ctime, B_TRUE); 1347 } 1348 /* 1349 * Grow the block size for a file. 1350 * 1351 * IN: zp - znode of file to free data in. 1352 * size - requested block size 1353 * tx - open transaction. 1354 * 1355 * NOTE: this function assumes that the znode is write locked. 1356 */ 1357 void 1358 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1359 { 1360 int error; 1361 u_longlong_t dummy; 1362 1363 if (size <= zp->z_blksz) 1364 return; 1365 /* 1366 * If the file size is already greater than the current blocksize, 1367 * we will not grow. If there is more than one block in a file, 1368 * the blocksize cannot change. 1369 */ 1370 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1371 return; 1372 1373 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1374 size, 0, tx); 1375 1376 if (error == ENOTSUP) 1377 return; 1378 ASSERT0(error); 1379 1380 /* What blocksize did we actually get? */ 1381 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1382 } 1383 1384 /* 1385 * Increase the file length 1386 * 1387 * IN: zp - znode of file to free data in. 1388 * end - new end-of-file 1389 * 1390 * RETURN: 0 on success, error code on failure 1391 */ 1392 static int 1393 zfs_extend(znode_t *zp, uint64_t end) 1394 { 1395 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1396 dmu_tx_t *tx; 1397 zfs_locked_range_t *lr; 1398 uint64_t newblksz; 1399 int error; 1400 1401 /* 1402 * We will change zp_size, lock the whole file. 1403 */ 1404 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER); 1405 1406 /* 1407 * Nothing to do if file already at desired length. 1408 */ 1409 if (end <= zp->z_size) { 1410 zfs_rangelock_exit(lr); 1411 return (0); 1412 } 1413 tx = dmu_tx_create(zfsvfs->z_os); 1414 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1415 zfs_sa_upgrade_txholds(tx, zp); 1416 if (end > zp->z_blksz && 1417 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1418 /* 1419 * We are growing the file past the current block size. 1420 */ 1421 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1422 /* 1423 * File's blocksize is already larger than the 1424 * "recordsize" property. Only let it grow to 1425 * the next power of 2. 1426 */ 1427 ASSERT(!ISP2(zp->z_blksz)); 1428 newblksz = MIN(end, 1 << highbit64(zp->z_blksz)); 1429 } else { 1430 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1431 } 1432 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1433 } else { 1434 newblksz = 0; 1435 } 1436 1437 error = dmu_tx_assign(tx, TXG_WAIT); 1438 if (error) { 1439 dmu_tx_abort(tx); 1440 zfs_rangelock_exit(lr); 1441 return (error); 1442 } 1443 1444 if (newblksz) 1445 zfs_grow_blocksize(zp, newblksz, tx); 1446 1447 zp->z_size = end; 1448 1449 VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1450 &zp->z_size, sizeof (zp->z_size), tx)); 1451 1452 vnode_pager_setsize(ZTOV(zp), end); 1453 1454 zfs_rangelock_exit(lr); 1455 1456 dmu_tx_commit(tx); 1457 1458 return (0); 1459 } 1460 1461 /* 1462 * Free space in a file. 1463 * 1464 * IN: zp - znode of file to free data in. 1465 * off - start of section to free. 1466 * len - length of section to free. 1467 * 1468 * RETURN: 0 on success, error code on failure 1469 */ 1470 static int 1471 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1472 { 1473 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1474 zfs_locked_range_t *lr; 1475 int error; 1476 1477 /* 1478 * Lock the range being freed. 1479 */ 1480 lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER); 1481 1482 /* 1483 * Nothing to do if file already at desired length. 1484 */ 1485 if (off >= zp->z_size) { 1486 zfs_rangelock_exit(lr); 1487 return (0); 1488 } 1489 1490 if (off + len > zp->z_size) 1491 len = zp->z_size - off; 1492 1493 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1494 1495 if (error == 0) { 1496 #if __FreeBSD_version >= 1400032 1497 vnode_pager_purge_range(ZTOV(zp), off, off + len); 1498 #else 1499 /* 1500 * Before __FreeBSD_version 1400032 we cannot free block in the 1501 * middle of a file, but only at the end of a file, so this code 1502 * path should never happen. 1503 */ 1504 vnode_pager_setsize(ZTOV(zp), off); 1505 #endif 1506 } 1507 1508 zfs_rangelock_exit(lr); 1509 1510 return (error); 1511 } 1512 1513 /* 1514 * Truncate a file 1515 * 1516 * IN: zp - znode of file to free data in. 1517 * end - new end-of-file. 1518 * 1519 * RETURN: 0 on success, error code on failure 1520 */ 1521 static int 1522 zfs_trunc(znode_t *zp, uint64_t end) 1523 { 1524 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1525 vnode_t *vp = ZTOV(zp); 1526 dmu_tx_t *tx; 1527 zfs_locked_range_t *lr; 1528 int error; 1529 sa_bulk_attr_t bulk[2]; 1530 int count = 0; 1531 1532 /* 1533 * We will change zp_size, lock the whole file. 1534 */ 1535 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER); 1536 1537 /* 1538 * Nothing to do if file already at desired length. 1539 */ 1540 if (end >= zp->z_size) { 1541 zfs_rangelock_exit(lr); 1542 return (0); 1543 } 1544 1545 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, 1546 DMU_OBJECT_END); 1547 if (error) { 1548 zfs_rangelock_exit(lr); 1549 return (error); 1550 } 1551 tx = dmu_tx_create(zfsvfs->z_os); 1552 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1553 zfs_sa_upgrade_txholds(tx, zp); 1554 dmu_tx_mark_netfree(tx); 1555 error = dmu_tx_assign(tx, TXG_WAIT); 1556 if (error) { 1557 dmu_tx_abort(tx); 1558 zfs_rangelock_exit(lr); 1559 return (error); 1560 } 1561 1562 zp->z_size = end; 1563 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1564 NULL, &zp->z_size, sizeof (zp->z_size)); 1565 1566 if (end == 0) { 1567 zp->z_pflags &= ~ZFS_SPARSE; 1568 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1569 NULL, &zp->z_pflags, 8); 1570 } 1571 VERIFY0(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx)); 1572 1573 dmu_tx_commit(tx); 1574 1575 /* 1576 * Clear any mapped pages in the truncated region. This has to 1577 * happen outside of the transaction to avoid the possibility of 1578 * a deadlock with someone trying to push a page that we are 1579 * about to invalidate. 1580 */ 1581 vnode_pager_setsize(vp, end); 1582 1583 zfs_rangelock_exit(lr); 1584 1585 return (0); 1586 } 1587 1588 /* 1589 * Free space in a file 1590 * 1591 * IN: zp - znode of file to free data in. 1592 * off - start of range 1593 * len - end of range (0 => EOF) 1594 * flag - current file open mode flags. 1595 * log - TRUE if this action should be logged 1596 * 1597 * RETURN: 0 on success, error code on failure 1598 */ 1599 int 1600 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1601 { 1602 dmu_tx_t *tx; 1603 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1604 zilog_t *zilog = zfsvfs->z_log; 1605 uint64_t mode; 1606 uint64_t mtime[2], ctime[2]; 1607 sa_bulk_attr_t bulk[3]; 1608 int count = 0; 1609 int error; 1610 1611 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1612 sizeof (mode))) != 0) 1613 return (error); 1614 1615 if (off > zp->z_size) { 1616 error = zfs_extend(zp, off+len); 1617 if (error == 0 && log) 1618 goto log; 1619 else 1620 return (error); 1621 } 1622 1623 if (len == 0) { 1624 error = zfs_trunc(zp, off); 1625 } else { 1626 if ((error = zfs_free_range(zp, off, len)) == 0 && 1627 off + len > zp->z_size) 1628 error = zfs_extend(zp, off+len); 1629 } 1630 if (error || !log) 1631 return (error); 1632 log: 1633 tx = dmu_tx_create(zfsvfs->z_os); 1634 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1635 zfs_sa_upgrade_txholds(tx, zp); 1636 error = dmu_tx_assign(tx, TXG_WAIT); 1637 if (error) { 1638 dmu_tx_abort(tx); 1639 return (error); 1640 } 1641 1642 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1643 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1644 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1645 NULL, &zp->z_pflags, 8); 1646 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); 1647 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1648 ASSERT0(error); 1649 1650 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1651 1652 dmu_tx_commit(tx); 1653 return (0); 1654 } 1655 1656 void 1657 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1658 { 1659 uint64_t moid, obj, sa_obj, version; 1660 uint64_t sense = ZFS_CASE_SENSITIVE; 1661 uint64_t norm = 0; 1662 nvpair_t *elem; 1663 int error; 1664 int i; 1665 znode_t *rootzp = NULL; 1666 zfsvfs_t *zfsvfs; 1667 vattr_t vattr; 1668 znode_t *zp; 1669 zfs_acl_ids_t acl_ids; 1670 1671 /* 1672 * First attempt to create master node. 1673 */ 1674 /* 1675 * In an empty objset, there are no blocks to read and thus 1676 * there can be no i/o errors (which we assert below). 1677 */ 1678 moid = MASTER_NODE_OBJ; 1679 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1680 DMU_OT_NONE, 0, tx); 1681 ASSERT0(error); 1682 1683 /* 1684 * Set starting attributes. 1685 */ 1686 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1687 elem = NULL; 1688 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1689 /* For the moment we expect all zpl props to be uint64_ts */ 1690 uint64_t val; 1691 char *name; 1692 1693 ASSERT3S(nvpair_type(elem), ==, DATA_TYPE_UINT64); 1694 val = fnvpair_value_uint64(elem); 1695 name = nvpair_name(elem); 1696 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1697 if (val < version) 1698 version = val; 1699 } else { 1700 error = zap_update(os, moid, name, 8, 1, &val, tx); 1701 } 1702 ASSERT0(error); 1703 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1704 norm = val; 1705 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1706 sense = val; 1707 } 1708 ASSERT3U(version, !=, 0); 1709 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1710 ASSERT0(error); 1711 1712 /* 1713 * Create zap object used for SA attribute registration 1714 */ 1715 1716 if (version >= ZPL_VERSION_SA) { 1717 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1718 DMU_OT_NONE, 0, tx); 1719 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1720 ASSERT0(error); 1721 } else { 1722 sa_obj = 0; 1723 } 1724 /* 1725 * Create a delete queue. 1726 */ 1727 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1728 1729 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1730 ASSERT0(error); 1731 1732 /* 1733 * Create root znode. Create minimal znode/vnode/zfsvfs 1734 * to allow zfs_mknode to work. 1735 */ 1736 VATTR_NULL(&vattr); 1737 vattr.va_mask = AT_MODE|AT_UID|AT_GID; 1738 vattr.va_type = VDIR; 1739 vattr.va_mode = S_IFDIR|0755; 1740 vattr.va_uid = crgetuid(cr); 1741 vattr.va_gid = crgetgid(cr); 1742 1743 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 1744 1745 rootzp = zfs_znode_alloc_kmem(KM_SLEEP); 1746 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1747 rootzp->z_unlinked = 0; 1748 rootzp->z_atime_dirty = 0; 1749 rootzp->z_is_sa = USE_SA(version, os); 1750 1751 zfsvfs->z_os = os; 1752 zfsvfs->z_parent = zfsvfs; 1753 zfsvfs->z_version = version; 1754 zfsvfs->z_use_fuids = USE_FUIDS(version, os); 1755 zfsvfs->z_use_sa = USE_SA(version, os); 1756 zfsvfs->z_norm = norm; 1757 1758 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1759 &zfsvfs->z_attr_table); 1760 1761 ASSERT0(error); 1762 1763 /* 1764 * Fold case on file systems that are always or sometimes case 1765 * insensitive. 1766 */ 1767 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1768 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 1769 1770 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1771 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 1772 offsetof(znode_t, z_link_node)); 1773 1774 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1775 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1776 1777 rootzp->z_zfsvfs = zfsvfs; 1778 VERIFY0(zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1779 cr, NULL, &acl_ids, NULL)); 1780 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1781 ASSERT3P(zp, ==, rootzp); 1782 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1783 ASSERT0(error); 1784 zfs_acl_ids_free(&acl_ids); 1785 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1786 1787 sa_handle_destroy(rootzp->z_sa_hdl); 1788 zfs_znode_free_kmem(rootzp); 1789 1790 /* 1791 * Create shares directory 1792 */ 1793 1794 error = zfs_create_share_dir(zfsvfs, tx); 1795 1796 ASSERT0(error); 1797 1798 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1799 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1800 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1801 } 1802 #endif /* _KERNEL */ 1803 1804 static int 1805 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1806 { 1807 uint64_t sa_obj = 0; 1808 int error; 1809 1810 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1811 if (error != 0 && error != ENOENT) 1812 return (error); 1813 1814 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1815 return (error); 1816 } 1817 1818 static int 1819 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1820 dmu_buf_t **db, const void *tag) 1821 { 1822 dmu_object_info_t doi; 1823 int error; 1824 1825 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 1826 return (error); 1827 1828 dmu_object_info_from_db(*db, &doi); 1829 if ((doi.doi_bonus_type != DMU_OT_SA && 1830 doi.doi_bonus_type != DMU_OT_ZNODE) || 1831 (doi.doi_bonus_type == DMU_OT_ZNODE && 1832 doi.doi_bonus_size < sizeof (znode_phys_t))) { 1833 sa_buf_rele(*db, tag); 1834 return (SET_ERROR(ENOTSUP)); 1835 } 1836 1837 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 1838 if (error != 0) { 1839 sa_buf_rele(*db, tag); 1840 return (error); 1841 } 1842 1843 return (0); 1844 } 1845 1846 static void 1847 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, const void *tag) 1848 { 1849 sa_handle_destroy(hdl); 1850 sa_buf_rele(db, tag); 1851 } 1852 1853 /* 1854 * Given an object number, return its parent object number and whether 1855 * or not the object is an extended attribute directory. 1856 */ 1857 static int 1858 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table, 1859 uint64_t *pobjp, int *is_xattrdir) 1860 { 1861 uint64_t parent; 1862 uint64_t pflags; 1863 uint64_t mode; 1864 uint64_t parent_mode; 1865 sa_bulk_attr_t bulk[3]; 1866 sa_handle_t *sa_hdl; 1867 dmu_buf_t *sa_db; 1868 int count = 0; 1869 int error; 1870 1871 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 1872 &parent, sizeof (parent)); 1873 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 1874 &pflags, sizeof (pflags)); 1875 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 1876 &mode, sizeof (mode)); 1877 1878 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 1879 return (error); 1880 1881 /* 1882 * When a link is removed its parent pointer is not changed and will 1883 * be invalid. There are two cases where a link is removed but the 1884 * file stays around, when it goes to the delete queue and when there 1885 * are additional links. 1886 */ 1887 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG); 1888 if (error != 0) 1889 return (error); 1890 1891 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode)); 1892 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 1893 if (error != 0) 1894 return (error); 1895 1896 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 1897 1898 /* 1899 * Extended attributes can be applied to files, directories, etc. 1900 * Otherwise the parent must be a directory. 1901 */ 1902 if (!*is_xattrdir && !S_ISDIR(parent_mode)) 1903 return (SET_ERROR(EINVAL)); 1904 1905 *pobjp = parent; 1906 1907 return (0); 1908 } 1909 1910 /* 1911 * Given an object number, return some zpl level statistics 1912 */ 1913 static int 1914 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 1915 zfs_stat_t *sb) 1916 { 1917 sa_bulk_attr_t bulk[4]; 1918 int count = 0; 1919 1920 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 1921 &sb->zs_mode, sizeof (sb->zs_mode)); 1922 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 1923 &sb->zs_gen, sizeof (sb->zs_gen)); 1924 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 1925 &sb->zs_links, sizeof (sb->zs_links)); 1926 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 1927 &sb->zs_ctime, sizeof (sb->zs_ctime)); 1928 1929 return (sa_bulk_lookup(hdl, bulk, count)); 1930 } 1931 1932 static int 1933 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 1934 sa_attr_type_t *sa_table, char *buf, int len) 1935 { 1936 sa_handle_t *sa_hdl; 1937 sa_handle_t *prevhdl = NULL; 1938 dmu_buf_t *prevdb = NULL; 1939 dmu_buf_t *sa_db = NULL; 1940 char *path = buf + len - 1; 1941 int error; 1942 1943 *path = '\0'; 1944 sa_hdl = hdl; 1945 1946 uint64_t deleteq_obj; 1947 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ, 1948 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj)); 1949 error = zap_lookup_int(osp, deleteq_obj, obj); 1950 if (error == 0) { 1951 return (ESTALE); 1952 } else if (error != ENOENT) { 1953 return (error); 1954 } 1955 1956 for (;;) { 1957 uint64_t pobj; 1958 char component[MAXNAMELEN + 2]; 1959 size_t complen; 1960 int is_xattrdir; 1961 1962 if (prevdb) { 1963 ASSERT3P(prevhdl, !=, NULL); 1964 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 1965 } 1966 1967 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj, 1968 &is_xattrdir)) != 0) 1969 break; 1970 1971 if (pobj == obj) { 1972 if (path[0] != '/') 1973 *--path = '/'; 1974 break; 1975 } 1976 1977 component[0] = '/'; 1978 if (is_xattrdir) { 1979 (void) sprintf(component + 1, "<xattrdir>"); 1980 } else { 1981 error = zap_value_search(osp, pobj, obj, 1982 ZFS_DIRENT_OBJ(-1ULL), component + 1); 1983 if (error != 0) 1984 break; 1985 } 1986 1987 complen = strlen(component); 1988 path -= complen; 1989 ASSERT3P(path, >=, buf); 1990 memcpy(path, component, complen); 1991 obj = pobj; 1992 1993 if (sa_hdl != hdl) { 1994 prevhdl = sa_hdl; 1995 prevdb = sa_db; 1996 } 1997 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 1998 if (error != 0) { 1999 sa_hdl = prevhdl; 2000 sa_db = prevdb; 2001 break; 2002 } 2003 } 2004 2005 if (sa_hdl != NULL && sa_hdl != hdl) { 2006 ASSERT3P(sa_db, !=, NULL); 2007 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2008 } 2009 2010 if (error == 0) 2011 (void) memmove(buf, path, buf + len - path); 2012 2013 return (error); 2014 } 2015 2016 int 2017 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2018 { 2019 sa_attr_type_t *sa_table; 2020 sa_handle_t *hdl; 2021 dmu_buf_t *db; 2022 int error; 2023 2024 error = zfs_sa_setup(osp, &sa_table); 2025 if (error != 0) 2026 return (error); 2027 2028 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2029 if (error != 0) 2030 return (error); 2031 2032 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2033 2034 zfs_release_sa_handle(hdl, db, FTAG); 2035 return (error); 2036 } 2037 2038 int 2039 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2040 char *buf, int len) 2041 { 2042 char *path = buf + len - 1; 2043 sa_attr_type_t *sa_table; 2044 sa_handle_t *hdl; 2045 dmu_buf_t *db; 2046 int error; 2047 2048 *path = '\0'; 2049 2050 error = zfs_sa_setup(osp, &sa_table); 2051 if (error != 0) 2052 return (error); 2053 2054 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2055 if (error != 0) 2056 return (error); 2057 2058 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2059 if (error != 0) { 2060 zfs_release_sa_handle(hdl, db, FTAG); 2061 return (error); 2062 } 2063 2064 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2065 2066 zfs_release_sa_handle(hdl, db, FTAG); 2067 return (error); 2068 } 2069 2070 2071 void 2072 zfs_znode_update_vfs(znode_t *zp) 2073 { 2074 vm_object_t object; 2075 2076 if ((object = ZTOV(zp)->v_object) == NULL || 2077 zp->z_size == object->un_pager.vnp.vnp_size) 2078 return; 2079 2080 vnode_pager_setsize(ZTOV(zp), zp->z_size); 2081 } 2082 2083 2084 #ifdef _KERNEL 2085 int 2086 zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf) 2087 { 2088 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2089 uint64_t parent; 2090 int is_xattrdir; 2091 int err; 2092 2093 /* Extended attributes should not be visible as regular files. */ 2094 if ((zp->z_pflags & ZFS_XATTR) != 0) 2095 return (SET_ERROR(EINVAL)); 2096 2097 err = zfs_obj_to_pobj(zfsvfs->z_os, zp->z_sa_hdl, zfsvfs->z_attr_table, 2098 &parent, &is_xattrdir); 2099 if (err != 0) 2100 return (err); 2101 ASSERT0(is_xattrdir); 2102 2103 /* No name as this is a root object. */ 2104 if (parent == zp->z_id) 2105 return (SET_ERROR(EINVAL)); 2106 2107 err = zap_value_search(zfsvfs->z_os, parent, zp->z_id, 2108 ZFS_DIRENT_OBJ(-1ULL), buf); 2109 if (err != 0) 2110 return (err); 2111 err = zfs_zget(zfsvfs, parent, dzpp); 2112 return (err); 2113 } 2114 #endif /* _KERNEL */ 2115