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/resourcevar.h> 38 #include <sys/mntent.h> 39 #include <sys/u8_textprep.h> 40 #include <sys/dsl_dataset.h> 41 #include <sys/vfs.h> 42 #include <sys/vnode.h> 43 #include <sys/file.h> 44 #include <sys/kmem.h> 45 #include <sys/errno.h> 46 #include <sys/unistd.h> 47 #include <sys/atomic.h> 48 #include <sys/zfs_dir.h> 49 #include <sys/zfs_acl.h> 50 #include <sys/zfs_ioctl.h> 51 #include <sys/zfs_rlock.h> 52 #include <sys/zfs_fuid.h> 53 #include <sys/dnode.h> 54 #include <sys/fs/zfs.h> 55 #endif /* _KERNEL */ 56 57 #include <sys/dmu.h> 58 #include <sys/dmu_objset.h> 59 #include <sys/dmu_tx.h> 60 #include <sys/zfs_refcount.h> 61 #include <sys/stat.h> 62 #include <sys/zap.h> 63 #include <sys/zfs_znode.h> 64 #include <sys/sa.h> 65 #include <sys/zfs_sa.h> 66 #include <sys/zfs_stat.h> 67 68 #include "zfs_prop.h" 69 #include "zfs_comutil.h" 70 71 /* Used by fstat(1). */ 72 SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 73 SYSCTL_NULL_INT_PTR, sizeof (znode_t), "sizeof(znode_t)"); 74 75 /* 76 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 77 * turned on when DEBUG is also defined. 78 */ 79 #ifdef ZFS_DEBUG 80 #define ZNODE_STATS 81 #endif /* DEBUG */ 82 83 #ifdef ZNODE_STATS 84 #define ZNODE_STAT_ADD(stat) ((stat)++) 85 #else 86 #define ZNODE_STAT_ADD(stat) /* nothing */ 87 #endif /* ZNODE_STATS */ 88 89 /* 90 * Functions needed for userland (ie: libzpool) are not put under 91 * #ifdef_KERNEL; the rest of the functions have dependencies 92 * (such as VFS logic) that will not compile easily in userland. 93 */ 94 #ifdef _KERNEL 95 #if !defined(KMEM_DEBUG) && __FreeBSD_version >= 1300102 96 #define _ZFS_USE_SMR 97 static uma_zone_t znode_uma_zone; 98 #else 99 static kmem_cache_t *znode_cache = NULL; 100 #endif 101 102 extern struct vop_vector zfs_vnodeops; 103 extern struct vop_vector zfs_fifoops; 104 extern struct vop_vector zfs_shareops; 105 106 107 /* 108 * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on 109 * z_rangelock. It will modify the offset and length of the lock to reflect 110 * znode-specific information, and convert RL_APPEND to RL_WRITER. This is 111 * called with the rangelock_t's rl_lock held, which avoids races. 112 */ 113 static void 114 zfs_rangelock_cb(zfs_locked_range_t *new, void *arg) 115 { 116 znode_t *zp = arg; 117 118 /* 119 * If in append mode, convert to writer and lock starting at the 120 * current end of file. 121 */ 122 if (new->lr_type == RL_APPEND) { 123 new->lr_offset = zp->z_size; 124 new->lr_type = RL_WRITER; 125 } 126 127 /* 128 * If we need to grow the block size then lock the whole file range. 129 */ 130 uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length); 131 if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) || 132 zp->z_blksz < ZTOZSB(zp)->z_max_blksz)) { 133 new->lr_offset = 0; 134 new->lr_length = UINT64_MAX; 135 } 136 } 137 138 static int 139 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 140 { 141 znode_t *zp = buf; 142 143 POINTER_INVALIDATE(&zp->z_zfsvfs); 144 145 list_link_init(&zp->z_link_node); 146 147 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 148 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 149 rw_init(&zp->z_xattr_lock, NULL, RW_DEFAULT, NULL); 150 151 zfs_rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp); 152 153 zp->z_acl_cached = NULL; 154 zp->z_xattr_cached = NULL; 155 zp->z_xattr_parent = 0; 156 zp->z_vnode = NULL; 157 zp->z_sync_writes_cnt = 0; 158 zp->z_async_writes_cnt = 0; 159 160 return (0); 161 } 162 163 static void 164 zfs_znode_cache_destructor(void *buf, void *arg) 165 { 166 (void) arg; 167 znode_t *zp = buf; 168 169 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 170 ASSERT3P(zp->z_vnode, ==, NULL); 171 ASSERT(!list_link_active(&zp->z_link_node)); 172 mutex_destroy(&zp->z_lock); 173 mutex_destroy(&zp->z_acl_lock); 174 rw_destroy(&zp->z_xattr_lock); 175 zfs_rangelock_fini(&zp->z_rangelock); 176 177 ASSERT3P(zp->z_acl_cached, ==, NULL); 178 ASSERT3P(zp->z_xattr_cached, ==, NULL); 179 180 ASSERT0(atomic_load_32(&zp->z_sync_writes_cnt)); 181 ASSERT0(atomic_load_32(&zp->z_async_writes_cnt)); 182 } 183 184 185 #ifdef _ZFS_USE_SMR 186 VFS_SMR_DECLARE; 187 188 static int 189 zfs_znode_cache_constructor_smr(void *mem, int size __unused, void *private, 190 int flags) 191 { 192 return (zfs_znode_cache_constructor(mem, private, flags)); 193 } 194 195 static void 196 zfs_znode_cache_destructor_smr(void *mem, int size __unused, void *private) 197 { 198 zfs_znode_cache_destructor(mem, private); 199 } 200 201 void 202 zfs_znode_init(void) 203 { 204 /* 205 * Initialize zcache 206 */ 207 ASSERT3P(znode_uma_zone, ==, NULL); 208 znode_uma_zone = uma_zcreate("zfs_znode_cache", 209 sizeof (znode_t), zfs_znode_cache_constructor_smr, 210 zfs_znode_cache_destructor_smr, NULL, NULL, 0, 0); 211 VFS_SMR_ZONE_SET(znode_uma_zone); 212 } 213 214 static znode_t * 215 zfs_znode_alloc_kmem(int flags) 216 { 217 return (uma_zalloc_smr(znode_uma_zone, flags)); 218 } 219 220 static void 221 zfs_znode_free_kmem(znode_t *zp) 222 { 223 if (zp->z_xattr_cached) { 224 nvlist_free(zp->z_xattr_cached); 225 zp->z_xattr_cached = NULL; 226 } 227 uma_zfree_smr(znode_uma_zone, zp); 228 } 229 #else 230 void 231 zfs_znode_init(void) 232 { 233 /* 234 * Initialize zcache 235 */ 236 ASSERT3P(znode_cache, ==, NULL); 237 znode_cache = kmem_cache_create("zfs_znode_cache", 238 sizeof (znode_t), 0, zfs_znode_cache_constructor, 239 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 240 } 241 242 static znode_t * 243 zfs_znode_alloc_kmem(int flags) 244 { 245 return (kmem_cache_alloc(znode_cache, flags)); 246 } 247 248 static void 249 zfs_znode_free_kmem(znode_t *zp) 250 { 251 if (zp->z_xattr_cached) { 252 nvlist_free(zp->z_xattr_cached); 253 zp->z_xattr_cached = NULL; 254 } 255 kmem_cache_free(znode_cache, zp); 256 } 257 #endif 258 259 void 260 zfs_znode_fini(void) 261 { 262 /* 263 * Cleanup zcache 264 */ 265 #ifdef _ZFS_USE_SMR 266 if (znode_uma_zone) { 267 uma_zdestroy(znode_uma_zone); 268 znode_uma_zone = NULL; 269 } 270 #else 271 if (znode_cache) { 272 kmem_cache_destroy(znode_cache); 273 znode_cache = NULL; 274 } 275 #endif 276 } 277 278 279 static int 280 zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 281 { 282 zfs_acl_ids_t acl_ids; 283 vattr_t vattr; 284 znode_t *sharezp; 285 znode_t *zp; 286 int error; 287 288 vattr.va_mask = AT_MODE|AT_UID|AT_GID; 289 vattr.va_type = VDIR; 290 vattr.va_mode = S_IFDIR|0555; 291 vattr.va_uid = crgetuid(kcred); 292 vattr.va_gid = crgetgid(kcred); 293 294 sharezp = zfs_znode_alloc_kmem(KM_SLEEP); 295 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 296 sharezp->z_unlinked = 0; 297 sharezp->z_atime_dirty = 0; 298 sharezp->z_zfsvfs = zfsvfs; 299 sharezp->z_is_sa = zfsvfs->z_use_sa; 300 301 VERIFY0(zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 302 kcred, NULL, &acl_ids, NULL)); 303 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 304 ASSERT3P(zp, ==, sharezp); 305 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 306 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 307 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 308 zfsvfs->z_shares_dir = sharezp->z_id; 309 310 zfs_acl_ids_free(&acl_ids); 311 sa_handle_destroy(sharezp->z_sa_hdl); 312 zfs_znode_free_kmem(sharezp); 313 314 return (error); 315 } 316 317 /* 318 * define a couple of values we need available 319 * for both 64 and 32 bit environments. 320 */ 321 #ifndef NBITSMINOR64 322 #define NBITSMINOR64 32 323 #endif 324 #ifndef MAXMAJ64 325 #define MAXMAJ64 0xffffffffUL 326 #endif 327 #ifndef MAXMIN64 328 #define MAXMIN64 0xffffffffUL 329 #endif 330 331 /* 332 * Create special expldev for ZFS private use. 333 * Can't use standard expldev since it doesn't do 334 * what we want. The standard expldev() takes a 335 * dev32_t in LP64 and expands it to a long dev_t. 336 * We need an interface that takes a dev32_t in ILP32 337 * and expands it to a long dev_t. 338 */ 339 static uint64_t 340 zfs_expldev(dev_t dev) 341 { 342 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 343 } 344 /* 345 * Special cmpldev for ZFS private use. 346 * Can't use standard cmpldev since it takes 347 * a long dev_t and compresses it to dev32_t in 348 * LP64. We need to do a compaction of a long dev_t 349 * to a dev32_t in ILP32. 350 */ 351 dev_t 352 zfs_cmpldev(uint64_t dev) 353 { 354 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 355 } 356 357 static void 358 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 359 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 360 { 361 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 362 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 363 364 ASSERT3P(zp->z_sa_hdl, ==, NULL); 365 ASSERT3P(zp->z_acl_cached, ==, NULL); 366 if (sa_hdl == NULL) { 367 VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, zp, 368 SA_HDL_SHARED, &zp->z_sa_hdl)); 369 } else { 370 zp->z_sa_hdl = sa_hdl; 371 sa_set_userp(sa_hdl, zp); 372 } 373 374 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 375 376 /* 377 * Slap on VROOT if we are the root znode unless we are the root 378 * node of a snapshot mounted under .zfs. 379 */ 380 if (zp->z_id == zfsvfs->z_root && zfsvfs->z_parent == zfsvfs) 381 ZTOV(zp)->v_flag |= VROOT; 382 383 vn_exists(ZTOV(zp)); 384 } 385 386 void 387 zfs_znode_dmu_fini(znode_t *zp) 388 { 389 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 390 ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zp->z_zfsvfs)); 391 392 sa_handle_destroy(zp->z_sa_hdl); 393 zp->z_sa_hdl = NULL; 394 } 395 396 static void 397 zfs_vnode_forget(vnode_t *vp) 398 { 399 400 /* copied from insmntque_stddtr */ 401 vp->v_data = NULL; 402 vp->v_op = &dead_vnodeops; 403 vgone(vp); 404 vput(vp); 405 } 406 407 /* 408 * Construct a new znode/vnode and initialize. 409 * 410 * This does not do a call to dmu_set_user() that is 411 * up to the caller to do, in case you don't want to 412 * return the znode 413 */ 414 static znode_t * 415 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 416 dmu_object_type_t obj_type, sa_handle_t *hdl) 417 { 418 znode_t *zp; 419 vnode_t *vp; 420 uint64_t mode; 421 uint64_t parent; 422 #ifdef notyet 423 uint64_t mtime[2], ctime[2]; 424 #endif 425 uint64_t projid = ZFS_DEFAULT_PROJID; 426 sa_bulk_attr_t bulk[9]; 427 int count = 0; 428 int error; 429 430 zp = zfs_znode_alloc_kmem(KM_SLEEP); 431 432 #ifndef _ZFS_USE_SMR 433 KASSERT((zfsvfs->z_parent->z_vfs->mnt_kern_flag & MNTK_FPLOOKUP) == 0, 434 ("%s: fast path lookup enabled without smr", __func__)); 435 #endif 436 437 #if __FreeBSD_version >= 1300076 438 KASSERT(curthread->td_vp_reserved != NULL, 439 ("zfs_znode_alloc: getnewvnode without any vnodes reserved")); 440 #else 441 KASSERT(curthread->td_vp_reserv > 0, 442 ("zfs_znode_alloc: getnewvnode without any vnodes reserved")); 443 #endif 444 error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp); 445 if (error != 0) { 446 zfs_znode_free_kmem(zp); 447 return (NULL); 448 } 449 zp->z_vnode = vp; 450 vp->v_data = zp; 451 452 /* 453 * Acquire the vnode lock before any possible interaction with the 454 * outside world. Specifically, there is an error path that calls 455 * zfs_vnode_forget() and the vnode should be exclusively locked. 456 */ 457 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 458 459 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 460 461 zp->z_sa_hdl = NULL; 462 zp->z_unlinked = 0; 463 zp->z_atime_dirty = 0; 464 zp->z_mapcnt = 0; 465 zp->z_id = db->db_object; 466 zp->z_blksz = blksz; 467 zp->z_seq = 0x7A4653; 468 zp->z_sync_cnt = 0; 469 zp->z_sync_writes_cnt = 0; 470 zp->z_async_writes_cnt = 0; 471 #if __FreeBSD_version >= 1300139 472 atomic_store_ptr(&zp->z_cached_symlink, NULL); 473 #endif 474 475 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 476 477 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 478 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 479 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 480 &zp->z_size, 8); 481 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 482 &zp->z_links, 8); 483 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 484 &zp->z_pflags, 8); 485 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 486 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 487 &zp->z_atime, 16); 488 #ifdef notyet 489 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 490 &mtime, 16); 491 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 492 &ctime, 16); 493 #endif 494 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 495 &zp->z_uid, 8); 496 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 497 &zp->z_gid, 8); 498 499 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0 || 500 (dmu_objset_projectquota_enabled(zfsvfs->z_os) && 501 (zp->z_pflags & ZFS_PROJID) && 502 sa_lookup(zp->z_sa_hdl, SA_ZPL_PROJID(zfsvfs), &projid, 8) != 0)) { 503 if (hdl == NULL) 504 sa_handle_destroy(zp->z_sa_hdl); 505 zfs_vnode_forget(vp); 506 zp->z_vnode = NULL; 507 zfs_znode_free_kmem(zp); 508 return (NULL); 509 } 510 511 zp->z_projid = projid; 512 zp->z_mode = mode; 513 514 /* Cache the xattr parent id */ 515 if (zp->z_pflags & ZFS_XATTR) 516 zp->z_xattr_parent = parent; 517 518 vp->v_type = IFTOVT((mode_t)mode); 519 520 switch (vp->v_type) { 521 case VDIR: 522 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 523 break; 524 case VFIFO: 525 vp->v_op = &zfs_fifoops; 526 break; 527 case VREG: 528 if (parent == zfsvfs->z_shares_dir) { 529 ASSERT0(zp->z_uid); 530 ASSERT0(zp->z_gid); 531 vp->v_op = &zfs_shareops; 532 } 533 break; 534 default: 535 break; 536 } 537 538 mutex_enter(&zfsvfs->z_znodes_lock); 539 list_insert_tail(&zfsvfs->z_all_znodes, zp); 540 zfsvfs->z_nr_znodes++; 541 zp->z_zfsvfs = zfsvfs; 542 mutex_exit(&zfsvfs->z_znodes_lock); 543 544 #if __FreeBSD_version >= 1400077 545 vn_set_state(vp, VSTATE_CONSTRUCTED); 546 #endif 547 VN_LOCK_AREC(vp); 548 if (vp->v_type != VFIFO) 549 VN_LOCK_ASHARE(vp); 550 551 return (zp); 552 } 553 554 static uint64_t empty_xattr; 555 static uint64_t pad[4]; 556 static zfs_acl_phys_t acl_phys; 557 /* 558 * Create a new DMU object to hold a zfs znode. 559 * 560 * IN: dzp - parent directory for new znode 561 * vap - file attributes for new znode 562 * tx - dmu transaction id for zap operations 563 * cr - credentials of caller 564 * flag - flags: 565 * IS_ROOT_NODE - new object will be root 566 * IS_XATTR - new object is an attribute 567 * bonuslen - length of bonus buffer 568 * setaclp - File/Dir initial ACL 569 * fuidp - Tracks fuid allocation. 570 * 571 * OUT: zpp - allocated znode 572 * 573 */ 574 void 575 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 576 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 577 { 578 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 579 uint64_t mode, size, links, parent, pflags; 580 uint64_t dzp_pflags = 0; 581 uint64_t rdev = 0; 582 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 583 dmu_buf_t *db; 584 timestruc_t now; 585 uint64_t gen, obj; 586 int bonuslen; 587 int dnodesize; 588 sa_handle_t *sa_hdl; 589 dmu_object_type_t obj_type; 590 sa_bulk_attr_t *sa_attrs; 591 int cnt = 0; 592 zfs_acl_locator_cb_t locate = { 0 }; 593 594 ASSERT3P(vap, !=, NULL); 595 ASSERT3U((vap->va_mask & AT_MODE), ==, AT_MODE); 596 597 if (zfsvfs->z_replay) { 598 obj = vap->va_nodeid; 599 now = vap->va_ctime; /* see zfs_replay_create() */ 600 gen = vap->va_nblocks; /* ditto */ 601 dnodesize = vap->va_fsid; /* ditto */ 602 } else { 603 obj = 0; 604 vfs_timestamp(&now); 605 gen = dmu_tx_get_txg(tx); 606 dnodesize = dmu_objset_dnodesize(zfsvfs->z_os); 607 } 608 609 if (dnodesize == 0) 610 dnodesize = DNODE_MIN_SIZE; 611 612 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 613 bonuslen = (obj_type == DMU_OT_SA) ? 614 DN_BONUS_SIZE(dnodesize) : ZFS_OLD_ZNODE_PHYS_SIZE; 615 616 /* 617 * Create a new DMU object. 618 */ 619 /* 620 * There's currently no mechanism for pre-reading the blocks that will 621 * be needed to allocate a new object, so we accept the small chance 622 * that there will be an i/o error and we will fail one of the 623 * assertions below. 624 */ 625 if (vap->va_type == VDIR) { 626 if (zfsvfs->z_replay) { 627 VERIFY0(zap_create_claim_norm_dnsize(zfsvfs->z_os, obj, 628 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 629 obj_type, bonuslen, dnodesize, tx)); 630 } else { 631 obj = zap_create_norm_dnsize(zfsvfs->z_os, 632 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 633 obj_type, bonuslen, dnodesize, tx); 634 } 635 } else { 636 if (zfsvfs->z_replay) { 637 VERIFY0(dmu_object_claim_dnsize(zfsvfs->z_os, obj, 638 DMU_OT_PLAIN_FILE_CONTENTS, 0, 639 obj_type, bonuslen, dnodesize, tx)); 640 } else { 641 obj = dmu_object_alloc_dnsize(zfsvfs->z_os, 642 DMU_OT_PLAIN_FILE_CONTENTS, 0, 643 obj_type, bonuslen, dnodesize, tx); 644 } 645 } 646 647 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 648 VERIFY0(sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 649 650 /* 651 * If this is the root, fix up the half-initialized parent pointer 652 * to reference the just-allocated physical data area. 653 */ 654 if (flag & IS_ROOT_NODE) { 655 dzp->z_id = obj; 656 } else { 657 dzp_pflags = dzp->z_pflags; 658 } 659 660 /* 661 * If parent is an xattr, so am I. 662 */ 663 if (dzp_pflags & ZFS_XATTR) { 664 flag |= IS_XATTR; 665 } 666 667 if (zfsvfs->z_use_fuids) 668 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 669 else 670 pflags = 0; 671 672 if (vap->va_type == VDIR) { 673 size = 2; /* contents ("." and "..") */ 674 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 675 } else { 676 size = links = 0; 677 } 678 679 if (vap->va_type == VBLK || vap->va_type == VCHR) { 680 rdev = zfs_expldev(vap->va_rdev); 681 } 682 683 parent = dzp->z_id; 684 mode = acl_ids->z_mode; 685 if (flag & IS_XATTR) 686 pflags |= ZFS_XATTR; 687 688 /* 689 * No execs denied will be determined when zfs_mode_compute() is called. 690 */ 691 pflags |= acl_ids->z_aclp->z_hints & 692 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 693 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 694 695 ZFS_TIME_ENCODE(&now, crtime); 696 ZFS_TIME_ENCODE(&now, ctime); 697 698 if (vap->va_mask & AT_ATIME) { 699 ZFS_TIME_ENCODE(&vap->va_atime, atime); 700 } else { 701 ZFS_TIME_ENCODE(&now, atime); 702 } 703 704 if (vap->va_mask & AT_MTIME) { 705 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 706 } else { 707 ZFS_TIME_ENCODE(&now, mtime); 708 } 709 710 /* Now add in all of the "SA" attributes */ 711 VERIFY0(sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 712 &sa_hdl)); 713 714 /* 715 * Setup the array of attributes to be replaced/set on the new file 716 * 717 * order for DMU_OT_ZNODE is critical since it needs to be constructed 718 * in the old znode_phys_t format. Don't change this ordering 719 */ 720 sa_attrs = kmem_alloc(sizeof (sa_bulk_attr_t) * ZPL_END, KM_SLEEP); 721 722 if (obj_type == DMU_OT_ZNODE) { 723 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 724 NULL, &atime, 16); 725 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 726 NULL, &mtime, 16); 727 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 728 NULL, &ctime, 16); 729 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 730 NULL, &crtime, 16); 731 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 732 NULL, &gen, 8); 733 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 734 NULL, &mode, 8); 735 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 736 NULL, &size, 8); 737 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 738 NULL, &parent, 8); 739 } else { 740 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 741 NULL, &mode, 8); 742 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 743 NULL, &size, 8); 744 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 745 NULL, &gen, 8); 746 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), 747 NULL, &acl_ids->z_fuid, 8); 748 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), 749 NULL, &acl_ids->z_fgid, 8); 750 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 751 NULL, &parent, 8); 752 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 753 NULL, &pflags, 8); 754 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 755 NULL, &atime, 16); 756 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 757 NULL, &mtime, 16); 758 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 759 NULL, &ctime, 16); 760 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 761 NULL, &crtime, 16); 762 } 763 764 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 765 766 if (obj_type == DMU_OT_ZNODE) { 767 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 768 &empty_xattr, 8); 769 } 770 if (obj_type == DMU_OT_ZNODE || 771 (vap->va_type == VBLK || vap->va_type == VCHR)) { 772 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 773 NULL, &rdev, 8); 774 775 } 776 if (obj_type == DMU_OT_ZNODE) { 777 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 778 NULL, &pflags, 8); 779 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 780 &acl_ids->z_fuid, 8); 781 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 782 &acl_ids->z_fgid, 8); 783 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 784 sizeof (uint64_t) * 4); 785 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 786 &acl_phys, sizeof (zfs_acl_phys_t)); 787 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 788 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 789 &acl_ids->z_aclp->z_acl_count, 8); 790 locate.cb_aclp = acl_ids->z_aclp; 791 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 792 zfs_acl_data_locator, &locate, 793 acl_ids->z_aclp->z_acl_bytes); 794 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 795 acl_ids->z_fuid, acl_ids->z_fgid); 796 } 797 798 VERIFY0(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx)); 799 800 if (!(flag & IS_ROOT_NODE)) { 801 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 802 ASSERT3P(*zpp, !=, NULL); 803 } else { 804 /* 805 * If we are creating the root node, the "parent" we 806 * passed in is the znode for the root. 807 */ 808 *zpp = dzp; 809 810 (*zpp)->z_sa_hdl = sa_hdl; 811 } 812 813 (*zpp)->z_pflags = pflags; 814 (*zpp)->z_mode = mode; 815 (*zpp)->z_dnodesize = dnodesize; 816 817 if (vap->va_mask & AT_XVATTR) 818 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 819 820 if (obj_type == DMU_OT_ZNODE || 821 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 822 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx)); 823 } 824 if (!(flag & IS_ROOT_NODE)) { 825 vnode_t *vp = ZTOV(*zpp); 826 vp->v_vflag |= VV_FORCEINSMQ; 827 int err = insmntque(vp, zfsvfs->z_vfs); 828 vp->v_vflag &= ~VV_FORCEINSMQ; 829 (void) err; 830 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 831 } 832 kmem_free(sa_attrs, sizeof (sa_bulk_attr_t) * ZPL_END); 833 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 834 } 835 836 /* 837 * Update in-core attributes. It is assumed the caller will be doing an 838 * sa_bulk_update to push the changes out. 839 */ 840 void 841 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 842 { 843 xoptattr_t *xoap; 844 845 xoap = xva_getxoptattr(xvap); 846 ASSERT3P(xoap, !=, NULL); 847 848 if (zp->z_zfsvfs->z_replay == B_FALSE) { 849 ASSERT_VOP_IN_SEQC(ZTOV(zp)); 850 } 851 852 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 853 uint64_t times[2]; 854 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 855 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 856 ×, sizeof (times), tx); 857 XVA_SET_RTN(xvap, XAT_CREATETIME); 858 } 859 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 860 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 861 zp->z_pflags, tx); 862 XVA_SET_RTN(xvap, XAT_READONLY); 863 } 864 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 865 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 866 zp->z_pflags, tx); 867 XVA_SET_RTN(xvap, XAT_HIDDEN); 868 } 869 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 870 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 871 zp->z_pflags, tx); 872 XVA_SET_RTN(xvap, XAT_SYSTEM); 873 } 874 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 875 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 876 zp->z_pflags, tx); 877 XVA_SET_RTN(xvap, XAT_ARCHIVE); 878 } 879 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 880 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 881 zp->z_pflags, tx); 882 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 883 } 884 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 885 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 886 zp->z_pflags, tx); 887 XVA_SET_RTN(xvap, XAT_NOUNLINK); 888 } 889 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 890 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 891 zp->z_pflags, tx); 892 XVA_SET_RTN(xvap, XAT_APPENDONLY); 893 } 894 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 895 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 896 zp->z_pflags, tx); 897 XVA_SET_RTN(xvap, XAT_NODUMP); 898 } 899 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 900 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 901 zp->z_pflags, tx); 902 XVA_SET_RTN(xvap, XAT_OPAQUE); 903 } 904 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 905 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 906 xoap->xoa_av_quarantined, zp->z_pflags, tx); 907 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 908 } 909 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 910 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 911 zp->z_pflags, tx); 912 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 913 } 914 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 915 zfs_sa_set_scanstamp(zp, xvap, tx); 916 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 917 } 918 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 919 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 920 zp->z_pflags, tx); 921 XVA_SET_RTN(xvap, XAT_REPARSE); 922 } 923 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 924 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 925 zp->z_pflags, tx); 926 XVA_SET_RTN(xvap, XAT_OFFLINE); 927 } 928 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 929 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 930 zp->z_pflags, tx); 931 XVA_SET_RTN(xvap, XAT_SPARSE); 932 } 933 } 934 935 int 936 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 937 { 938 dmu_object_info_t doi; 939 dmu_buf_t *db; 940 znode_t *zp; 941 vnode_t *vp; 942 sa_handle_t *hdl; 943 int locked; 944 int err; 945 946 getnewvnode_reserve_(); 947 again: 948 *zpp = NULL; 949 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 950 951 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 952 if (err) { 953 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 954 getnewvnode_drop_reserve(); 955 return (err); 956 } 957 958 dmu_object_info_from_db(db, &doi); 959 if (doi.doi_bonus_type != DMU_OT_SA && 960 (doi.doi_bonus_type != DMU_OT_ZNODE || 961 (doi.doi_bonus_type == DMU_OT_ZNODE && 962 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 963 sa_buf_rele(db, NULL); 964 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 965 getnewvnode_drop_reserve(); 966 return (SET_ERROR(EINVAL)); 967 } 968 969 hdl = dmu_buf_get_user(db); 970 if (hdl != NULL) { 971 zp = sa_get_userdata(hdl); 972 973 /* 974 * Since "SA" does immediate eviction we 975 * should never find a sa handle that doesn't 976 * know about the znode. 977 */ 978 ASSERT3P(zp, !=, NULL); 979 ASSERT3U(zp->z_id, ==, obj_num); 980 if (zp->z_unlinked) { 981 err = SET_ERROR(ENOENT); 982 } else { 983 vp = ZTOV(zp); 984 /* 985 * Don't let the vnode disappear after 986 * ZFS_OBJ_HOLD_EXIT. 987 */ 988 VN_HOLD(vp); 989 *zpp = zp; 990 err = 0; 991 } 992 993 sa_buf_rele(db, NULL); 994 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 995 996 if (err) { 997 getnewvnode_drop_reserve(); 998 return (err); 999 } 1000 1001 locked = VOP_ISLOCKED(vp); 1002 VI_LOCK(vp); 1003 if (VN_IS_DOOMED(vp) && locked != LK_EXCLUSIVE) { 1004 /* 1005 * The vnode is doomed and this thread doesn't 1006 * hold the exclusive lock on it, so the vnode 1007 * must be being reclaimed by another thread. 1008 * Otherwise the doomed vnode is being reclaimed 1009 * by this thread and zfs_zget is called from 1010 * ZIL internals. 1011 */ 1012 VI_UNLOCK(vp); 1013 1014 /* 1015 * XXX vrele() locks the vnode when the last reference 1016 * is dropped. Although in this case the vnode is 1017 * doomed / dead and so no inactivation is required, 1018 * the vnode lock is still acquired. That could result 1019 * in a LOR with z_teardown_lock if another thread holds 1020 * the vnode's lock and tries to take z_teardown_lock. 1021 * But that is only possible if the other thread peforms 1022 * a ZFS vnode operation on the vnode. That either 1023 * should not happen if the vnode is dead or the thread 1024 * should also have a reference to the vnode and thus 1025 * our reference is not last. 1026 */ 1027 VN_RELE(vp); 1028 goto again; 1029 } 1030 VI_UNLOCK(vp); 1031 getnewvnode_drop_reserve(); 1032 return (err); 1033 } 1034 1035 /* 1036 * Not found create new znode/vnode 1037 * but only if file exists. 1038 * 1039 * There is a small window where zfs_vget() could 1040 * find this object while a file create is still in 1041 * progress. This is checked for in zfs_znode_alloc() 1042 * 1043 * if zfs_znode_alloc() fails it will drop the hold on the 1044 * bonus buffer. 1045 */ 1046 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1047 doi.doi_bonus_type, NULL); 1048 if (zp == NULL) { 1049 err = SET_ERROR(ENOENT); 1050 } else { 1051 *zpp = zp; 1052 } 1053 if (err == 0) { 1054 vnode_t *vp = ZTOV(zp); 1055 1056 err = insmntque(vp, zfsvfs->z_vfs); 1057 if (err == 0) { 1058 vp->v_hash = obj_num; 1059 VOP_UNLOCK1(vp); 1060 } else { 1061 zp->z_vnode = NULL; 1062 zfs_znode_dmu_fini(zp); 1063 zfs_znode_free(zp); 1064 *zpp = NULL; 1065 } 1066 } 1067 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1068 getnewvnode_drop_reserve(); 1069 return (err); 1070 } 1071 1072 int 1073 zfs_rezget(znode_t *zp) 1074 { 1075 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1076 dmu_object_info_t doi; 1077 dmu_buf_t *db; 1078 vnode_t *vp; 1079 uint64_t obj_num = zp->z_id; 1080 uint64_t mode, size; 1081 sa_bulk_attr_t bulk[8]; 1082 int err; 1083 int count = 0; 1084 uint64_t gen; 1085 1086 /* 1087 * Remove cached pages before reloading the znode, so that they are not 1088 * lingering after we run into any error. Ideally, we should vgone() 1089 * the vnode in case of error, but currently we cannot do that 1090 * because of the LOR between the vnode lock and z_teardown_lock. 1091 * So, instead, we have to "doom" the znode in the illumos style. 1092 * 1093 * Ignore invalid pages during the scan. This is to avoid deadlocks 1094 * between page busying and the teardown lock, as pages are busied prior 1095 * to a VOP_GETPAGES operation, which acquires the teardown read lock. 1096 * Such pages will be invalid and can safely be skipped here. 1097 */ 1098 vp = ZTOV(zp); 1099 #if __FreeBSD_version >= 1400042 1100 vn_pages_remove_valid(vp, 0, 0); 1101 #else 1102 vn_pages_remove(vp, 0, 0); 1103 #endif 1104 1105 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1106 1107 mutex_enter(&zp->z_acl_lock); 1108 if (zp->z_acl_cached) { 1109 zfs_acl_free(zp->z_acl_cached); 1110 zp->z_acl_cached = NULL; 1111 } 1112 mutex_exit(&zp->z_acl_lock); 1113 1114 rw_enter(&zp->z_xattr_lock, RW_WRITER); 1115 if (zp->z_xattr_cached) { 1116 nvlist_free(zp->z_xattr_cached); 1117 zp->z_xattr_cached = NULL; 1118 } 1119 rw_exit(&zp->z_xattr_lock); 1120 1121 ASSERT3P(zp->z_sa_hdl, ==, NULL); 1122 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1123 if (err) { 1124 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1125 return (err); 1126 } 1127 1128 dmu_object_info_from_db(db, &doi); 1129 if (doi.doi_bonus_type != DMU_OT_SA && 1130 (doi.doi_bonus_type != DMU_OT_ZNODE || 1131 (doi.doi_bonus_type == DMU_OT_ZNODE && 1132 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1133 sa_buf_rele(db, NULL); 1134 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1135 return (SET_ERROR(EINVAL)); 1136 } 1137 1138 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1139 size = zp->z_size; 1140 1141 /* reload cached values */ 1142 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1143 &gen, sizeof (gen)); 1144 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1145 &zp->z_size, sizeof (zp->z_size)); 1146 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1147 &zp->z_links, sizeof (zp->z_links)); 1148 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1149 &zp->z_pflags, sizeof (zp->z_pflags)); 1150 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1151 &zp->z_atime, sizeof (zp->z_atime)); 1152 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1153 &zp->z_uid, sizeof (zp->z_uid)); 1154 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1155 &zp->z_gid, sizeof (zp->z_gid)); 1156 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1157 &mode, sizeof (mode)); 1158 1159 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1160 zfs_znode_dmu_fini(zp); 1161 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1162 return (SET_ERROR(EIO)); 1163 } 1164 1165 zp->z_mode = mode; 1166 1167 if (gen != zp->z_gen) { 1168 zfs_znode_dmu_fini(zp); 1169 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1170 return (SET_ERROR(EIO)); 1171 } 1172 1173 /* 1174 * It is highly improbable but still quite possible that two 1175 * objects in different datasets are created with the same 1176 * object numbers and in transaction groups with the same 1177 * numbers. znodes corresponding to those objects would 1178 * have the same z_id and z_gen, but their other attributes 1179 * may be different. 1180 * zfs recv -F may replace one of such objects with the other. 1181 * As a result file properties recorded in the replaced 1182 * object's vnode may no longer match the received object's 1183 * properties. At present the only cached property is the 1184 * files type recorded in v_type. 1185 * So, handle this case by leaving the old vnode and znode 1186 * disassociated from the actual object. A new vnode and a 1187 * znode will be created if the object is accessed 1188 * (e.g. via a look-up). The old vnode and znode will be 1189 * recycled when the last vnode reference is dropped. 1190 */ 1191 if (vp->v_type != IFTOVT((mode_t)zp->z_mode)) { 1192 zfs_znode_dmu_fini(zp); 1193 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1194 return (SET_ERROR(EIO)); 1195 } 1196 1197 /* 1198 * If the file has zero links, then it has been unlinked on the send 1199 * side and it must be in the received unlinked set. 1200 * We call zfs_znode_dmu_fini() now to prevent any accesses to the 1201 * stale data and to prevent automatically removal of the file in 1202 * zfs_zinactive(). The file will be removed either when it is removed 1203 * on the send side and the next incremental stream is received or 1204 * when the unlinked set gets processed. 1205 */ 1206 zp->z_unlinked = (zp->z_links == 0); 1207 if (zp->z_unlinked) { 1208 zfs_znode_dmu_fini(zp); 1209 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1210 return (0); 1211 } 1212 1213 zp->z_blksz = doi.doi_data_block_size; 1214 if (zp->z_size != size) 1215 vnode_pager_setsize(vp, zp->z_size); 1216 1217 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1218 1219 return (0); 1220 } 1221 1222 void 1223 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1224 { 1225 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1226 objset_t *os = zfsvfs->z_os; 1227 uint64_t obj = zp->z_id; 1228 uint64_t acl_obj = zfs_external_acl(zp); 1229 1230 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1231 if (acl_obj) { 1232 VERIFY(!zp->z_is_sa); 1233 VERIFY0(dmu_object_free(os, acl_obj, tx)); 1234 } 1235 VERIFY0(dmu_object_free(os, obj, tx)); 1236 zfs_znode_dmu_fini(zp); 1237 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1238 zfs_znode_free(zp); 1239 } 1240 1241 void 1242 zfs_zinactive(znode_t *zp) 1243 { 1244 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1245 uint64_t z_id = zp->z_id; 1246 1247 ASSERT3P(zp->z_sa_hdl, !=, NULL); 1248 1249 /* 1250 * Don't allow a zfs_zget() while were trying to release this znode 1251 */ 1252 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1253 1254 /* 1255 * If this was the last reference to a file with no links, remove 1256 * the file from the file system unless the file system is mounted 1257 * read-only. That can happen, for example, if the file system was 1258 * originally read-write, the file was opened, then unlinked and 1259 * the file system was made read-only before the file was finally 1260 * closed. The file will remain in the unlinked set. 1261 */ 1262 if (zp->z_unlinked) { 1263 ASSERT(!zfsvfs->z_issnap); 1264 if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) { 1265 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1266 zfs_rmnode(zp); 1267 return; 1268 } 1269 } 1270 1271 zfs_znode_dmu_fini(zp); 1272 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1273 zfs_znode_free(zp); 1274 } 1275 1276 void 1277 zfs_znode_free(znode_t *zp) 1278 { 1279 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1280 #if __FreeBSD_version >= 1300139 1281 char *symlink; 1282 #endif 1283 1284 ASSERT3P(zp->z_sa_hdl, ==, NULL); 1285 zp->z_vnode = NULL; 1286 mutex_enter(&zfsvfs->z_znodes_lock); 1287 POINTER_INVALIDATE(&zp->z_zfsvfs); 1288 list_remove(&zfsvfs->z_all_znodes, zp); 1289 zfsvfs->z_nr_znodes--; 1290 mutex_exit(&zfsvfs->z_znodes_lock); 1291 1292 #if __FreeBSD_version >= 1300139 1293 symlink = atomic_load_ptr(&zp->z_cached_symlink); 1294 if (symlink != NULL) { 1295 atomic_store_rel_ptr((uintptr_t *)&zp->z_cached_symlink, 1296 (uintptr_t)NULL); 1297 cache_symlink_free(symlink, strlen(symlink) + 1); 1298 } 1299 #endif 1300 1301 if (zp->z_acl_cached) { 1302 zfs_acl_free(zp->z_acl_cached); 1303 zp->z_acl_cached = NULL; 1304 } 1305 1306 zfs_znode_free_kmem(zp); 1307 } 1308 1309 void 1310 zfs_tstamp_update_setup_ext(znode_t *zp, uint_t flag, uint64_t mtime[2], 1311 uint64_t ctime[2], boolean_t have_tx) 1312 { 1313 timestruc_t now; 1314 1315 vfs_timestamp(&now); 1316 1317 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1318 zp->z_atime_dirty = 0; 1319 zp->z_seq++; 1320 } else { 1321 zp->z_atime_dirty = 1; 1322 } 1323 1324 if (flag & AT_ATIME) { 1325 ZFS_TIME_ENCODE(&now, zp->z_atime); 1326 } 1327 1328 if (flag & AT_MTIME) { 1329 ZFS_TIME_ENCODE(&now, mtime); 1330 if (zp->z_zfsvfs->z_use_fuids) { 1331 zp->z_pflags |= (ZFS_ARCHIVE | 1332 ZFS_AV_MODIFIED); 1333 } 1334 } 1335 1336 if (flag & AT_CTIME) { 1337 ZFS_TIME_ENCODE(&now, ctime); 1338 if (zp->z_zfsvfs->z_use_fuids) 1339 zp->z_pflags |= ZFS_ARCHIVE; 1340 } 1341 } 1342 1343 1344 void 1345 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1346 uint64_t ctime[2]) 1347 { 1348 zfs_tstamp_update_setup_ext(zp, flag, mtime, ctime, B_TRUE); 1349 } 1350 /* 1351 * Grow the block size for a file. 1352 * 1353 * IN: zp - znode of file to free data in. 1354 * size - requested block size 1355 * tx - open transaction. 1356 * 1357 * NOTE: this function assumes that the znode is write locked. 1358 */ 1359 void 1360 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1361 { 1362 int error; 1363 u_longlong_t dummy; 1364 1365 if (size <= zp->z_blksz) 1366 return; 1367 /* 1368 * If the file size is already greater than the current blocksize, 1369 * we will not grow. If there is more than one block in a file, 1370 * the blocksize cannot change. 1371 */ 1372 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1373 return; 1374 1375 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1376 size, 0, tx); 1377 1378 if (error == ENOTSUP) 1379 return; 1380 ASSERT0(error); 1381 1382 /* What blocksize did we actually get? */ 1383 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1384 } 1385 1386 /* 1387 * Increase the file length 1388 * 1389 * IN: zp - znode of file to free data in. 1390 * end - new end-of-file 1391 * 1392 * RETURN: 0 on success, error code on failure 1393 */ 1394 static int 1395 zfs_extend(znode_t *zp, uint64_t end) 1396 { 1397 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1398 dmu_tx_t *tx; 1399 zfs_locked_range_t *lr; 1400 uint64_t newblksz; 1401 int error; 1402 1403 /* 1404 * We will change zp_size, lock the whole file. 1405 */ 1406 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER); 1407 1408 /* 1409 * Nothing to do if file already at desired length. 1410 */ 1411 if (end <= zp->z_size) { 1412 zfs_rangelock_exit(lr); 1413 return (0); 1414 } 1415 tx = dmu_tx_create(zfsvfs->z_os); 1416 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1417 zfs_sa_upgrade_txholds(tx, zp); 1418 if (end > zp->z_blksz && 1419 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1420 /* 1421 * We are growing the file past the current block size. 1422 */ 1423 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1424 /* 1425 * File's blocksize is already larger than the 1426 * "recordsize" property. Only let it grow to 1427 * the next power of 2. 1428 */ 1429 ASSERT(!ISP2(zp->z_blksz)); 1430 newblksz = MIN(end, 1 << highbit64(zp->z_blksz)); 1431 } else { 1432 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1433 } 1434 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1435 } else { 1436 newblksz = 0; 1437 } 1438 1439 error = dmu_tx_assign(tx, TXG_WAIT); 1440 if (error) { 1441 dmu_tx_abort(tx); 1442 zfs_rangelock_exit(lr); 1443 return (error); 1444 } 1445 1446 if (newblksz) 1447 zfs_grow_blocksize(zp, newblksz, tx); 1448 1449 zp->z_size = end; 1450 1451 VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1452 &zp->z_size, sizeof (zp->z_size), tx)); 1453 1454 vnode_pager_setsize(ZTOV(zp), end); 1455 1456 zfs_rangelock_exit(lr); 1457 1458 dmu_tx_commit(tx); 1459 1460 return (0); 1461 } 1462 1463 /* 1464 * Free space in a file. 1465 * 1466 * IN: zp - znode of file to free data in. 1467 * off - start of section to free. 1468 * len - length of section to free. 1469 * 1470 * RETURN: 0 on success, error code on failure 1471 */ 1472 static int 1473 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1474 { 1475 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1476 zfs_locked_range_t *lr; 1477 int error; 1478 1479 /* 1480 * Lock the range being freed. 1481 */ 1482 lr = zfs_rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER); 1483 1484 /* 1485 * Nothing to do if file already at desired length. 1486 */ 1487 if (off >= zp->z_size) { 1488 zfs_rangelock_exit(lr); 1489 return (0); 1490 } 1491 1492 if (off + len > zp->z_size) 1493 len = zp->z_size - off; 1494 1495 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1496 1497 if (error == 0) { 1498 #if __FreeBSD_version >= 1400032 1499 vnode_pager_purge_range(ZTOV(zp), off, off + len); 1500 #else 1501 /* 1502 * Before __FreeBSD_version 1400032 we cannot free block in the 1503 * middle of a file, but only at the end of a file, so this code 1504 * path should never happen. 1505 */ 1506 vnode_pager_setsize(ZTOV(zp), off); 1507 #endif 1508 } 1509 1510 zfs_rangelock_exit(lr); 1511 1512 return (error); 1513 } 1514 1515 /* 1516 * Truncate a file 1517 * 1518 * IN: zp - znode of file to free data in. 1519 * end - new end-of-file. 1520 * 1521 * RETURN: 0 on success, error code on failure 1522 */ 1523 static int 1524 zfs_trunc(znode_t *zp, uint64_t end) 1525 { 1526 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1527 vnode_t *vp = ZTOV(zp); 1528 dmu_tx_t *tx; 1529 zfs_locked_range_t *lr; 1530 int error; 1531 sa_bulk_attr_t bulk[2]; 1532 int count = 0; 1533 1534 /* 1535 * We will change zp_size, lock the whole file. 1536 */ 1537 lr = zfs_rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER); 1538 1539 /* 1540 * Nothing to do if file already at desired length. 1541 */ 1542 if (end >= zp->z_size) { 1543 zfs_rangelock_exit(lr); 1544 return (0); 1545 } 1546 1547 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, 1548 DMU_OBJECT_END); 1549 if (error) { 1550 zfs_rangelock_exit(lr); 1551 return (error); 1552 } 1553 tx = dmu_tx_create(zfsvfs->z_os); 1554 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1555 zfs_sa_upgrade_txholds(tx, zp); 1556 dmu_tx_mark_netfree(tx); 1557 error = dmu_tx_assign(tx, TXG_WAIT); 1558 if (error) { 1559 dmu_tx_abort(tx); 1560 zfs_rangelock_exit(lr); 1561 return (error); 1562 } 1563 1564 zp->z_size = end; 1565 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1566 NULL, &zp->z_size, sizeof (zp->z_size)); 1567 1568 if (end == 0) { 1569 zp->z_pflags &= ~ZFS_SPARSE; 1570 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1571 NULL, &zp->z_pflags, 8); 1572 } 1573 VERIFY0(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx)); 1574 1575 dmu_tx_commit(tx); 1576 1577 /* 1578 * Clear any mapped pages in the truncated region. This has to 1579 * happen outside of the transaction to avoid the possibility of 1580 * a deadlock with someone trying to push a page that we are 1581 * about to invalidate. 1582 */ 1583 vnode_pager_setsize(vp, end); 1584 1585 zfs_rangelock_exit(lr); 1586 1587 return (0); 1588 } 1589 1590 /* 1591 * Free space in a file 1592 * 1593 * IN: zp - znode of file to free data in. 1594 * off - start of range 1595 * len - end of range (0 => EOF) 1596 * flag - current file open mode flags. 1597 * log - TRUE if this action should be logged 1598 * 1599 * RETURN: 0 on success, error code on failure 1600 */ 1601 int 1602 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1603 { 1604 dmu_tx_t *tx; 1605 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1606 zilog_t *zilog = zfsvfs->z_log; 1607 uint64_t mode; 1608 uint64_t mtime[2], ctime[2]; 1609 sa_bulk_attr_t bulk[3]; 1610 int count = 0; 1611 int error; 1612 1613 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1614 sizeof (mode))) != 0) 1615 return (error); 1616 1617 if (off > zp->z_size) { 1618 error = zfs_extend(zp, off+len); 1619 if (error == 0 && log) 1620 goto log; 1621 else 1622 return (error); 1623 } 1624 1625 if (len == 0) { 1626 error = zfs_trunc(zp, off); 1627 } else { 1628 if ((error = zfs_free_range(zp, off, len)) == 0 && 1629 off + len > zp->z_size) 1630 error = zfs_extend(zp, off+len); 1631 } 1632 if (error || !log) 1633 return (error); 1634 log: 1635 tx = dmu_tx_create(zfsvfs->z_os); 1636 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1637 zfs_sa_upgrade_txholds(tx, zp); 1638 error = dmu_tx_assign(tx, TXG_WAIT); 1639 if (error) { 1640 dmu_tx_abort(tx); 1641 return (error); 1642 } 1643 1644 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1645 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1646 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1647 NULL, &zp->z_pflags, 8); 1648 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); 1649 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1650 ASSERT0(error); 1651 1652 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1653 1654 dmu_tx_commit(tx); 1655 return (0); 1656 } 1657 1658 void 1659 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1660 { 1661 uint64_t moid, obj, sa_obj, version; 1662 uint64_t sense = ZFS_CASE_SENSITIVE; 1663 uint64_t norm = 0; 1664 nvpair_t *elem; 1665 int error; 1666 int i; 1667 znode_t *rootzp = NULL; 1668 zfsvfs_t *zfsvfs; 1669 vattr_t vattr; 1670 znode_t *zp; 1671 zfs_acl_ids_t acl_ids; 1672 1673 /* 1674 * First attempt to create master node. 1675 */ 1676 /* 1677 * In an empty objset, there are no blocks to read and thus 1678 * there can be no i/o errors (which we assert below). 1679 */ 1680 moid = MASTER_NODE_OBJ; 1681 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1682 DMU_OT_NONE, 0, tx); 1683 ASSERT0(error); 1684 1685 /* 1686 * Set starting attributes. 1687 */ 1688 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1689 elem = NULL; 1690 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1691 /* For the moment we expect all zpl props to be uint64_ts */ 1692 uint64_t val; 1693 const char *name; 1694 1695 ASSERT3S(nvpair_type(elem), ==, DATA_TYPE_UINT64); 1696 val = fnvpair_value_uint64(elem); 1697 name = nvpair_name(elem); 1698 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1699 if (val < version) 1700 version = val; 1701 } else { 1702 error = zap_update(os, moid, name, 8, 1, &val, tx); 1703 } 1704 ASSERT0(error); 1705 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1706 norm = val; 1707 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1708 sense = val; 1709 } 1710 ASSERT3U(version, !=, 0); 1711 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1712 ASSERT0(error); 1713 1714 /* 1715 * Create zap object used for SA attribute registration 1716 */ 1717 1718 if (version >= ZPL_VERSION_SA) { 1719 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1720 DMU_OT_NONE, 0, tx); 1721 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1722 ASSERT0(error); 1723 } else { 1724 sa_obj = 0; 1725 } 1726 /* 1727 * Create a delete queue. 1728 */ 1729 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1730 1731 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1732 ASSERT0(error); 1733 1734 /* 1735 * Create root znode. Create minimal znode/vnode/zfsvfs 1736 * to allow zfs_mknode to work. 1737 */ 1738 VATTR_NULL(&vattr); 1739 vattr.va_mask = AT_MODE|AT_UID|AT_GID; 1740 vattr.va_type = VDIR; 1741 vattr.va_mode = S_IFDIR|0755; 1742 vattr.va_uid = crgetuid(cr); 1743 vattr.va_gid = crgetgid(cr); 1744 1745 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 1746 1747 rootzp = zfs_znode_alloc_kmem(KM_SLEEP); 1748 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1749 rootzp->z_unlinked = 0; 1750 rootzp->z_atime_dirty = 0; 1751 rootzp->z_is_sa = USE_SA(version, os); 1752 1753 zfsvfs->z_os = os; 1754 zfsvfs->z_parent = zfsvfs; 1755 zfsvfs->z_version = version; 1756 zfsvfs->z_use_fuids = USE_FUIDS(version, os); 1757 zfsvfs->z_use_sa = USE_SA(version, os); 1758 zfsvfs->z_norm = norm; 1759 1760 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1761 &zfsvfs->z_attr_table); 1762 1763 ASSERT0(error); 1764 1765 /* 1766 * Fold case on file systems that are always or sometimes case 1767 * insensitive. 1768 */ 1769 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1770 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 1771 1772 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1773 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 1774 offsetof(znode_t, z_link_node)); 1775 1776 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1777 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1778 1779 rootzp->z_zfsvfs = zfsvfs; 1780 VERIFY0(zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1781 cr, NULL, &acl_ids, NULL)); 1782 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1783 ASSERT3P(zp, ==, rootzp); 1784 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1785 ASSERT0(error); 1786 zfs_acl_ids_free(&acl_ids); 1787 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1788 1789 sa_handle_destroy(rootzp->z_sa_hdl); 1790 zfs_znode_free_kmem(rootzp); 1791 1792 /* 1793 * Create shares directory 1794 */ 1795 1796 error = zfs_create_share_dir(zfsvfs, tx); 1797 1798 ASSERT0(error); 1799 1800 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1801 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1802 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1803 } 1804 #endif /* _KERNEL */ 1805 1806 static int 1807 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1808 { 1809 uint64_t sa_obj = 0; 1810 int error; 1811 1812 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1813 if (error != 0 && error != ENOENT) 1814 return (error); 1815 1816 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1817 return (error); 1818 } 1819 1820 static int 1821 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1822 dmu_buf_t **db, const void *tag) 1823 { 1824 dmu_object_info_t doi; 1825 int error; 1826 1827 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 1828 return (error); 1829 1830 dmu_object_info_from_db(*db, &doi); 1831 if ((doi.doi_bonus_type != DMU_OT_SA && 1832 doi.doi_bonus_type != DMU_OT_ZNODE) || 1833 (doi.doi_bonus_type == DMU_OT_ZNODE && 1834 doi.doi_bonus_size < sizeof (znode_phys_t))) { 1835 sa_buf_rele(*db, tag); 1836 return (SET_ERROR(ENOTSUP)); 1837 } 1838 1839 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 1840 if (error != 0) { 1841 sa_buf_rele(*db, tag); 1842 return (error); 1843 } 1844 1845 return (0); 1846 } 1847 1848 static void 1849 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, const void *tag) 1850 { 1851 sa_handle_destroy(hdl); 1852 sa_buf_rele(db, tag); 1853 } 1854 1855 /* 1856 * Given an object number, return its parent object number and whether 1857 * or not the object is an extended attribute directory. 1858 */ 1859 static int 1860 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table, 1861 uint64_t *pobjp, int *is_xattrdir) 1862 { 1863 uint64_t parent; 1864 uint64_t pflags; 1865 uint64_t mode; 1866 uint64_t parent_mode; 1867 sa_bulk_attr_t bulk[3]; 1868 sa_handle_t *sa_hdl; 1869 dmu_buf_t *sa_db; 1870 int count = 0; 1871 int error; 1872 1873 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 1874 &parent, sizeof (parent)); 1875 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 1876 &pflags, sizeof (pflags)); 1877 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 1878 &mode, sizeof (mode)); 1879 1880 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 1881 return (error); 1882 1883 /* 1884 * When a link is removed its parent pointer is not changed and will 1885 * be invalid. There are two cases where a link is removed but the 1886 * file stays around, when it goes to the delete queue and when there 1887 * are additional links. 1888 */ 1889 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG); 1890 if (error != 0) 1891 return (error); 1892 1893 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode)); 1894 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 1895 if (error != 0) 1896 return (error); 1897 1898 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 1899 1900 /* 1901 * Extended attributes can be applied to files, directories, etc. 1902 * Otherwise the parent must be a directory. 1903 */ 1904 if (!*is_xattrdir && !S_ISDIR(parent_mode)) 1905 return (SET_ERROR(EINVAL)); 1906 1907 *pobjp = parent; 1908 1909 return (0); 1910 } 1911 1912 /* 1913 * Given an object number, return some zpl level statistics 1914 */ 1915 static int 1916 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 1917 zfs_stat_t *sb) 1918 { 1919 sa_bulk_attr_t bulk[4]; 1920 int count = 0; 1921 1922 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 1923 &sb->zs_mode, sizeof (sb->zs_mode)); 1924 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 1925 &sb->zs_gen, sizeof (sb->zs_gen)); 1926 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 1927 &sb->zs_links, sizeof (sb->zs_links)); 1928 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 1929 &sb->zs_ctime, sizeof (sb->zs_ctime)); 1930 1931 return (sa_bulk_lookup(hdl, bulk, count)); 1932 } 1933 1934 static int 1935 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 1936 sa_attr_type_t *sa_table, char *buf, int len) 1937 { 1938 sa_handle_t *sa_hdl; 1939 sa_handle_t *prevhdl = NULL; 1940 dmu_buf_t *prevdb = NULL; 1941 dmu_buf_t *sa_db = NULL; 1942 char *path = buf + len - 1; 1943 int error; 1944 1945 *path = '\0'; 1946 sa_hdl = hdl; 1947 1948 uint64_t deleteq_obj; 1949 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ, 1950 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj)); 1951 error = zap_lookup_int(osp, deleteq_obj, obj); 1952 if (error == 0) { 1953 return (ESTALE); 1954 } else if (error != ENOENT) { 1955 return (error); 1956 } 1957 1958 for (;;) { 1959 uint64_t pobj; 1960 char component[MAXNAMELEN + 2]; 1961 size_t complen; 1962 int is_xattrdir; 1963 1964 if (prevdb) { 1965 ASSERT3P(prevhdl, !=, NULL); 1966 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 1967 } 1968 1969 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj, 1970 &is_xattrdir)) != 0) 1971 break; 1972 1973 if (pobj == obj) { 1974 if (path[0] != '/') 1975 *--path = '/'; 1976 break; 1977 } 1978 1979 component[0] = '/'; 1980 if (is_xattrdir) { 1981 (void) sprintf(component + 1, "<xattrdir>"); 1982 } else { 1983 error = zap_value_search(osp, pobj, obj, 1984 ZFS_DIRENT_OBJ(-1ULL), component + 1); 1985 if (error != 0) 1986 break; 1987 } 1988 1989 complen = strlen(component); 1990 path -= complen; 1991 ASSERT3P(path, >=, buf); 1992 memcpy(path, component, complen); 1993 obj = pobj; 1994 1995 if (sa_hdl != hdl) { 1996 prevhdl = sa_hdl; 1997 prevdb = sa_db; 1998 } 1999 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2000 if (error != 0) { 2001 sa_hdl = prevhdl; 2002 sa_db = prevdb; 2003 break; 2004 } 2005 } 2006 2007 if (sa_hdl != NULL && sa_hdl != hdl) { 2008 ASSERT3P(sa_db, !=, NULL); 2009 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2010 } 2011 2012 if (error == 0) 2013 (void) memmove(buf, path, buf + len - path); 2014 2015 return (error); 2016 } 2017 2018 int 2019 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2020 { 2021 sa_attr_type_t *sa_table; 2022 sa_handle_t *hdl; 2023 dmu_buf_t *db; 2024 int error; 2025 2026 error = zfs_sa_setup(osp, &sa_table); 2027 if (error != 0) 2028 return (error); 2029 2030 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2031 if (error != 0) 2032 return (error); 2033 2034 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2035 2036 zfs_release_sa_handle(hdl, db, FTAG); 2037 return (error); 2038 } 2039 2040 int 2041 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2042 char *buf, int len) 2043 { 2044 char *path = buf + len - 1; 2045 sa_attr_type_t *sa_table; 2046 sa_handle_t *hdl; 2047 dmu_buf_t *db; 2048 int error; 2049 2050 *path = '\0'; 2051 2052 error = zfs_sa_setup(osp, &sa_table); 2053 if (error != 0) 2054 return (error); 2055 2056 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2057 if (error != 0) 2058 return (error); 2059 2060 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2061 if (error != 0) { 2062 zfs_release_sa_handle(hdl, db, FTAG); 2063 return (error); 2064 } 2065 2066 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2067 2068 zfs_release_sa_handle(hdl, db, FTAG); 2069 return (error); 2070 } 2071 2072 2073 void 2074 zfs_znode_update_vfs(znode_t *zp) 2075 { 2076 vm_object_t object; 2077 2078 if ((object = ZTOV(zp)->v_object) == NULL || 2079 zp->z_size == object->un_pager.vnp.vnp_size) 2080 return; 2081 2082 vnode_pager_setsize(ZTOV(zp), zp->z_size); 2083 } 2084 2085 2086 #ifdef _KERNEL 2087 int 2088 zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf) 2089 { 2090 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2091 uint64_t parent; 2092 int is_xattrdir; 2093 int err; 2094 2095 /* Extended attributes should not be visible as regular files. */ 2096 if ((zp->z_pflags & ZFS_XATTR) != 0) 2097 return (SET_ERROR(EINVAL)); 2098 2099 err = zfs_obj_to_pobj(zfsvfs->z_os, zp->z_sa_hdl, zfsvfs->z_attr_table, 2100 &parent, &is_xattrdir); 2101 if (err != 0) 2102 return (err); 2103 ASSERT0(is_xattrdir); 2104 2105 /* No name as this is a root object. */ 2106 if (parent == zp->z_id) 2107 return (SET_ERROR(EINVAL)); 2108 2109 err = zap_value_search(zfsvfs->z_os, parent, zp->z_id, 2110 ZFS_DIRENT_OBJ(-1ULL), buf); 2111 if (err != 0) 2112 return (err); 2113 err = zfs_zget(zfsvfs, parent, dzpp); 2114 return (err); 2115 } 2116 #endif /* _KERNEL */ 2117 2118 #ifdef _KERNEL 2119 int 2120 zfs_rlimit_fsize(off_t fsize) 2121 { 2122 struct thread *td = curthread; 2123 off_t lim; 2124 2125 if (td == NULL) 2126 return (0); 2127 2128 lim = lim_cur(td, RLIMIT_FSIZE); 2129 if (__predict_true((uoff_t)fsize <= lim)) 2130 return (0); 2131 2132 /* 2133 * The limit is reached. 2134 */ 2135 PROC_LOCK(td->td_proc); 2136 kern_psignal(td->td_proc, SIGXFSZ); 2137 PROC_UNLOCK(td->td_proc); 2138 2139 return (EFBIG); 2140 } 2141 #endif /* _KERNEL */ 2142