zfs_znode.c revision 249195
11573Srgrimes/* 21573Srgrimes * CDDL HEADER START 31573Srgrimes * 41573Srgrimes * The contents of this file are subject to the terms of the 51573Srgrimes * Common Development and Distribution License (the "License"). 61573Srgrimes * You may not use this file except in compliance with the License. 71573Srgrimes * 81573Srgrimes * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 91573Srgrimes * or http://www.opensolaris.org/os/licensing. 101573Srgrimes * See the License for the specific language governing permissions 111573Srgrimes * and limitations under the License. 121573Srgrimes * 131573Srgrimes * When distributing Covered Code, include this CDDL HEADER in each 141573Srgrimes * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 151573Srgrimes * If applicable, add the following below this CDDL HEADER, with the 161573Srgrimes * fields enclosed by brackets "[]" replaced with your own identifying 171573Srgrimes * information: Portions Copyright [yyyy] [name of copyright owner] 181573Srgrimes * 191573Srgrimes * CDDL HEADER END 201573Srgrimes */ 211573Srgrimes/* 221573Srgrimes * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 231573Srgrimes * Copyright (c) 2013 by Delphix. All rights reserved. 241573Srgrimes */ 251573Srgrimes 261573Srgrimes/* Portions Copyright 2007 Jeremy Teo */ 271573Srgrimes/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 281573Srgrimes 291573Srgrimes#ifdef _KERNEL 301573Srgrimes#include <sys/types.h> 311573Srgrimes#include <sys/param.h> 321573Srgrimes#include <sys/time.h> 331573Srgrimes#include <sys/systm.h> 341573Srgrimes#include <sys/sysmacros.h> 3530511Sache#include <sys/resource.h> 361573Srgrimes#include <sys/mntent.h> 3730511Sache#include <sys/u8_textprep.h> 3830511Sache#include <sys/dsl_dataset.h> 3930511Sache#include <sys/vfs.h> 401573Srgrimes#include <sys/vnode.h> 411573Srgrimes#include <sys/file.h> 4230479Sache#include <sys/kmem.h> 431573Srgrimes#include <sys/errno.h> 4425862Speter#include <sys/unistd.h> 451573Srgrimes#include <sys/atomic.h> 461573Srgrimes#include <sys/zfs_dir.h> 471573Srgrimes#include <sys/zfs_acl.h> 481573Srgrimes#include <sys/zfs_ioctl.h> 4925862Speter#include <sys/zfs_rlock.h> 5013545Sjulian#include <sys/zfs_fuid.h> 5125862Speter#include <sys/dnode.h> 5225862Speter#include <sys/fs/zfs.h> 5325862Speter#include <sys/kidmap.h> 5430443Speter#endif /* _KERNEL */ 5525862Speter 561573Srgrimes#include <sys/dmu.h> 57#include <sys/refcount.h> 58#include <sys/stat.h> 59#include <sys/zap.h> 60#include <sys/zfs_znode.h> 61#include <sys/sa.h> 62#include <sys/zfs_sa.h> 63#include <sys/zfs_stat.h> 64#include <sys/refcount.h> 65 66#include "zfs_prop.h" 67#include "zfs_comutil.h" 68 69/* Used by fstat(1). */ 70SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t), 71 "sizeof(znode_t)"); 72 73/* 74 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 75 * turned on when DEBUG is also defined. 76 */ 77#ifdef DEBUG 78#define ZNODE_STATS 79#endif /* DEBUG */ 80 81#ifdef ZNODE_STATS 82#define ZNODE_STAT_ADD(stat) ((stat)++) 83#else 84#define ZNODE_STAT_ADD(stat) /* nothing */ 85#endif /* ZNODE_STATS */ 86 87/* 88 * Functions needed for userland (ie: libzpool) are not put under 89 * #ifdef_KERNEL; the rest of the functions have dependencies 90 * (such as VFS logic) that will not compile easily in userland. 91 */ 92#ifdef _KERNEL 93/* 94 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 95 * be freed before it can be safely accessed. 96 */ 97krwlock_t zfsvfs_lock; 98 99static kmem_cache_t *znode_cache = NULL; 100 101/*ARGSUSED*/ 102static void 103znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 104{ 105 /* 106 * We should never drop all dbuf refs without first clearing 107 * the eviction callback. 108 */ 109 panic("evicting znode %p\n", user_ptr); 110} 111 112extern struct vop_vector zfs_vnodeops; 113extern struct vop_vector zfs_fifoops; 114extern struct vop_vector zfs_shareops; 115 116/* 117 * XXX: We cannot use this function as a cache constructor, because 118 * there is one global cache for all file systems and we need 119 * to pass vfsp here, which is not possible, because argument 120 * 'cdrarg' is defined at kmem_cache_create() time. 121 */ 122/*ARGSUSED*/ 123static int 124zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 125{ 126 znode_t *zp = buf; 127 vnode_t *vp; 128 vfs_t *vfsp = arg; 129 int error; 130 131 POINTER_INVALIDATE(&zp->z_zfsvfs); 132 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 133 134 if (vfsp != NULL) { 135 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp); 136 if (error != 0 && (kmflags & KM_NOSLEEP)) 137 return (-1); 138 ASSERT(error == 0); 139 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 140 zp->z_vnode = vp; 141 vp->v_data = (caddr_t)zp; 142 VN_LOCK_AREC(vp); 143 VN_LOCK_ASHARE(vp); 144 } else { 145 zp->z_vnode = NULL; 146 } 147 148 list_link_init(&zp->z_link_node); 149 150 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 151 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 152 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 153 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 154 155 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 156 avl_create(&zp->z_range_avl, zfs_range_compare, 157 sizeof (rl_t), offsetof(rl_t, r_node)); 158 159 zp->z_dirlocks = NULL; 160 zp->z_acl_cached = NULL; 161 zp->z_moved = 0; 162 return (0); 163} 164 165/*ARGSUSED*/ 166static void 167zfs_znode_cache_destructor(void *buf, void *arg) 168{ 169 znode_t *zp = buf; 170 171 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 172 ASSERT(ZTOV(zp) == NULL); 173 vn_free(ZTOV(zp)); 174 ASSERT(!list_link_active(&zp->z_link_node)); 175 mutex_destroy(&zp->z_lock); 176 rw_destroy(&zp->z_parent_lock); 177 rw_destroy(&zp->z_name_lock); 178 mutex_destroy(&zp->z_acl_lock); 179 avl_destroy(&zp->z_range_avl); 180 mutex_destroy(&zp->z_range_lock); 181 182 ASSERT(zp->z_dirlocks == NULL); 183 ASSERT(zp->z_acl_cached == NULL); 184} 185 186#ifdef ZNODE_STATS 187static struct { 188 uint64_t zms_zfsvfs_invalid; 189 uint64_t zms_zfsvfs_recheck1; 190 uint64_t zms_zfsvfs_unmounted; 191 uint64_t zms_zfsvfs_recheck2; 192 uint64_t zms_obj_held; 193 uint64_t zms_vnode_locked; 194 uint64_t zms_not_only_dnlc; 195} znode_move_stats; 196#endif /* ZNODE_STATS */ 197 198#ifdef sun 199static void 200zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 201{ 202 vnode_t *vp; 203 204 /* Copy fields. */ 205 nzp->z_zfsvfs = ozp->z_zfsvfs; 206 207 /* Swap vnodes. */ 208 vp = nzp->z_vnode; 209 nzp->z_vnode = ozp->z_vnode; 210 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 211 ZTOV(ozp)->v_data = ozp; 212 ZTOV(nzp)->v_data = nzp; 213 214 nzp->z_id = ozp->z_id; 215 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 216 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 217 nzp->z_unlinked = ozp->z_unlinked; 218 nzp->z_atime_dirty = ozp->z_atime_dirty; 219 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 220 nzp->z_blksz = ozp->z_blksz; 221 nzp->z_seq = ozp->z_seq; 222 nzp->z_mapcnt = ozp->z_mapcnt; 223 nzp->z_gen = ozp->z_gen; 224 nzp->z_sync_cnt = ozp->z_sync_cnt; 225 nzp->z_is_sa = ozp->z_is_sa; 226 nzp->z_sa_hdl = ozp->z_sa_hdl; 227 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 228 nzp->z_links = ozp->z_links; 229 nzp->z_size = ozp->z_size; 230 nzp->z_pflags = ozp->z_pflags; 231 nzp->z_uid = ozp->z_uid; 232 nzp->z_gid = ozp->z_gid; 233 nzp->z_mode = ozp->z_mode; 234 235 /* 236 * Since this is just an idle znode and kmem is already dealing with 237 * memory pressure, release any cached ACL. 238 */ 239 if (ozp->z_acl_cached) { 240 zfs_acl_free(ozp->z_acl_cached); 241 ozp->z_acl_cached = NULL; 242 } 243 244 sa_set_userp(nzp->z_sa_hdl, nzp); 245 246 /* 247 * Invalidate the original znode by clearing fields that provide a 248 * pointer back to the znode. Set the low bit of the vfs pointer to 249 * ensure that zfs_znode_move() recognizes the znode as invalid in any 250 * subsequent callback. 251 */ 252 ozp->z_sa_hdl = NULL; 253 POINTER_INVALIDATE(&ozp->z_zfsvfs); 254 255 /* 256 * Mark the znode. 257 */ 258 nzp->z_moved = 1; 259 ozp->z_moved = (uint8_t)-1; 260} 261 262/*ARGSUSED*/ 263static kmem_cbrc_t 264zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 265{ 266 znode_t *ozp = buf, *nzp = newbuf; 267 zfsvfs_t *zfsvfs; 268 vnode_t *vp; 269 270 /* 271 * The znode is on the file system's list of known znodes if the vfs 272 * pointer is valid. We set the low bit of the vfs pointer when freeing 273 * the znode to invalidate it, and the memory patterns written by kmem 274 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 275 * created znode sets the vfs pointer last of all to indicate that the 276 * znode is known and in a valid state to be moved by this function. 277 */ 278 zfsvfs = ozp->z_zfsvfs; 279 if (!POINTER_IS_VALID(zfsvfs)) { 280 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 281 return (KMEM_CBRC_DONT_KNOW); 282 } 283 284 /* 285 * Close a small window in which it's possible that the filesystem could 286 * be unmounted and freed, and zfsvfs, though valid in the previous 287 * statement, could point to unrelated memory by the time we try to 288 * prevent the filesystem from being unmounted. 289 */ 290 rw_enter(&zfsvfs_lock, RW_WRITER); 291 if (zfsvfs != ozp->z_zfsvfs) { 292 rw_exit(&zfsvfs_lock); 293 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 294 return (KMEM_CBRC_DONT_KNOW); 295 } 296 297 /* 298 * If the znode is still valid, then so is the file system. We know that 299 * no valid file system can be freed while we hold zfsvfs_lock, so we 300 * can safely ensure that the filesystem is not and will not be 301 * unmounted. The next statement is equivalent to ZFS_ENTER(). 302 */ 303 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 304 if (zfsvfs->z_unmounted) { 305 ZFS_EXIT(zfsvfs); 306 rw_exit(&zfsvfs_lock); 307 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 308 return (KMEM_CBRC_DONT_KNOW); 309 } 310 rw_exit(&zfsvfs_lock); 311 312 mutex_enter(&zfsvfs->z_znodes_lock); 313 /* 314 * Recheck the vfs pointer in case the znode was removed just before 315 * acquiring the lock. 316 */ 317 if (zfsvfs != ozp->z_zfsvfs) { 318 mutex_exit(&zfsvfs->z_znodes_lock); 319 ZFS_EXIT(zfsvfs); 320 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 321 return (KMEM_CBRC_DONT_KNOW); 322 } 323 324 /* 325 * At this point we know that as long as we hold z_znodes_lock, the 326 * znode cannot be freed and fields within the znode can be safely 327 * accessed. Now, prevent a race with zfs_zget(). 328 */ 329 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 330 mutex_exit(&zfsvfs->z_znodes_lock); 331 ZFS_EXIT(zfsvfs); 332 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 333 return (KMEM_CBRC_LATER); 334 } 335 336 vp = ZTOV(ozp); 337 if (mutex_tryenter(&vp->v_lock) == 0) { 338 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 339 mutex_exit(&zfsvfs->z_znodes_lock); 340 ZFS_EXIT(zfsvfs); 341 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 342 return (KMEM_CBRC_LATER); 343 } 344 345 /* Only move znodes that are referenced _only_ by the DNLC. */ 346 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 347 mutex_exit(&vp->v_lock); 348 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 349 mutex_exit(&zfsvfs->z_znodes_lock); 350 ZFS_EXIT(zfsvfs); 351 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 352 return (KMEM_CBRC_LATER); 353 } 354 355 /* 356 * The znode is known and in a valid state to move. We're holding the 357 * locks needed to execute the critical section. 358 */ 359 zfs_znode_move_impl(ozp, nzp); 360 mutex_exit(&vp->v_lock); 361 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 362 363 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 364 mutex_exit(&zfsvfs->z_znodes_lock); 365 ZFS_EXIT(zfsvfs); 366 367 return (KMEM_CBRC_YES); 368} 369#endif /* sun */ 370 371void 372zfs_znode_init(void) 373{ 374 /* 375 * Initialize zcache 376 */ 377 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 378 ASSERT(znode_cache == NULL); 379 znode_cache = kmem_cache_create("zfs_znode_cache", 380 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL, 381 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 382 kmem_cache_set_move(znode_cache, zfs_znode_move); 383} 384 385void 386zfs_znode_fini(void) 387{ 388#ifdef sun 389 /* 390 * Cleanup vfs & vnode ops 391 */ 392 zfs_remove_op_tables(); 393#endif /* sun */ 394 395 /* 396 * Cleanup zcache 397 */ 398 if (znode_cache) 399 kmem_cache_destroy(znode_cache); 400 znode_cache = NULL; 401 rw_destroy(&zfsvfs_lock); 402} 403 404#ifdef sun 405struct vnodeops *zfs_dvnodeops; 406struct vnodeops *zfs_fvnodeops; 407struct vnodeops *zfs_symvnodeops; 408struct vnodeops *zfs_xdvnodeops; 409struct vnodeops *zfs_evnodeops; 410struct vnodeops *zfs_sharevnodeops; 411 412void 413zfs_remove_op_tables() 414{ 415 /* 416 * Remove vfs ops 417 */ 418 ASSERT(zfsfstype); 419 (void) vfs_freevfsops_by_type(zfsfstype); 420 zfsfstype = 0; 421 422 /* 423 * Remove vnode ops 424 */ 425 if (zfs_dvnodeops) 426 vn_freevnodeops(zfs_dvnodeops); 427 if (zfs_fvnodeops) 428 vn_freevnodeops(zfs_fvnodeops); 429 if (zfs_symvnodeops) 430 vn_freevnodeops(zfs_symvnodeops); 431 if (zfs_xdvnodeops) 432 vn_freevnodeops(zfs_xdvnodeops); 433 if (zfs_evnodeops) 434 vn_freevnodeops(zfs_evnodeops); 435 if (zfs_sharevnodeops) 436 vn_freevnodeops(zfs_sharevnodeops); 437 438 zfs_dvnodeops = NULL; 439 zfs_fvnodeops = NULL; 440 zfs_symvnodeops = NULL; 441 zfs_xdvnodeops = NULL; 442 zfs_evnodeops = NULL; 443 zfs_sharevnodeops = NULL; 444} 445 446extern const fs_operation_def_t zfs_dvnodeops_template[]; 447extern const fs_operation_def_t zfs_fvnodeops_template[]; 448extern const fs_operation_def_t zfs_xdvnodeops_template[]; 449extern const fs_operation_def_t zfs_symvnodeops_template[]; 450extern const fs_operation_def_t zfs_evnodeops_template[]; 451extern const fs_operation_def_t zfs_sharevnodeops_template[]; 452 453int 454zfs_create_op_tables() 455{ 456 int error; 457 458 /* 459 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 460 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 461 * In this case we just return as the ops vectors are already set up. 462 */ 463 if (zfs_dvnodeops) 464 return (0); 465 466 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 467 &zfs_dvnodeops); 468 if (error) 469 return (error); 470 471 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 472 &zfs_fvnodeops); 473 if (error) 474 return (error); 475 476 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 477 &zfs_symvnodeops); 478 if (error) 479 return (error); 480 481 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 482 &zfs_xdvnodeops); 483 if (error) 484 return (error); 485 486 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 487 &zfs_evnodeops); 488 if (error) 489 return (error); 490 491 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 492 &zfs_sharevnodeops); 493 494 return (error); 495} 496#endif /* sun */ 497 498int 499zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 500{ 501 zfs_acl_ids_t acl_ids; 502 vattr_t vattr; 503 znode_t *sharezp; 504 vnode_t *vp, vnode; 505 znode_t *zp; 506 int error; 507 508 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 509 vattr.va_type = VDIR; 510 vattr.va_mode = S_IFDIR|0555; 511 vattr.va_uid = crgetuid(kcred); 512 vattr.va_gid = crgetgid(kcred); 513 514 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 515 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0); 516 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 517 sharezp->z_moved = 0; 518 sharezp->z_unlinked = 0; 519 sharezp->z_atime_dirty = 0; 520 sharezp->z_zfsvfs = zfsvfs; 521 sharezp->z_is_sa = zfsvfs->z_use_sa; 522 523 sharezp->z_vnode = &vnode; 524 vnode.v_data = sharezp; 525 526 vp = ZTOV(sharezp); 527 vp->v_type = VDIR; 528 529 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 530 kcred, NULL, &acl_ids)); 531 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 532 ASSERT3P(zp, ==, sharezp); 533 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 534 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 535 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 536 zfsvfs->z_shares_dir = sharezp->z_id; 537 538 zfs_acl_ids_free(&acl_ids); 539 ZTOV(sharezp)->v_data = NULL; 540 ZTOV(sharezp)->v_count = 0; 541 ZTOV(sharezp)->v_holdcnt = 0; 542 zp->z_vnode = NULL; 543 sa_handle_destroy(sharezp->z_sa_hdl); 544 sharezp->z_vnode = NULL; 545 kmem_cache_free(znode_cache, sharezp); 546 547 return (error); 548} 549 550/* 551 * define a couple of values we need available 552 * for both 64 and 32 bit environments. 553 */ 554#ifndef NBITSMINOR64 555#define NBITSMINOR64 32 556#endif 557#ifndef MAXMAJ64 558#define MAXMAJ64 0xffffffffUL 559#endif 560#ifndef MAXMIN64 561#define MAXMIN64 0xffffffffUL 562#endif 563 564/* 565 * Create special expldev for ZFS private use. 566 * Can't use standard expldev since it doesn't do 567 * what we want. The standard expldev() takes a 568 * dev32_t in LP64 and expands it to a long dev_t. 569 * We need an interface that takes a dev32_t in ILP32 570 * and expands it to a long dev_t. 571 */ 572static uint64_t 573zfs_expldev(dev_t dev) 574{ 575 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 576} 577/* 578 * Special cmpldev for ZFS private use. 579 * Can't use standard cmpldev since it takes 580 * a long dev_t and compresses it to dev32_t in 581 * LP64. We need to do a compaction of a long dev_t 582 * to a dev32_t in ILP32. 583 */ 584dev_t 585zfs_cmpldev(uint64_t dev) 586{ 587 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 588} 589 590static void 591zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 592 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 593{ 594 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 595 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 596 597 mutex_enter(&zp->z_lock); 598 599 ASSERT(zp->z_sa_hdl == NULL); 600 ASSERT(zp->z_acl_cached == NULL); 601 if (sa_hdl == NULL) { 602 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 603 SA_HDL_SHARED, &zp->z_sa_hdl)); 604 } else { 605 zp->z_sa_hdl = sa_hdl; 606 sa_set_userp(sa_hdl, zp); 607 } 608 609 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 610 611 /* 612 * Slap on VROOT if we are the root znode 613 */ 614 if (zp->z_id == zfsvfs->z_root) 615 ZTOV(zp)->v_flag |= VROOT; 616 617 mutex_exit(&zp->z_lock); 618 vn_exists(ZTOV(zp)); 619} 620 621void 622zfs_znode_dmu_fini(znode_t *zp) 623{ 624 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 625 zp->z_unlinked || 626 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 627 628 sa_handle_destroy(zp->z_sa_hdl); 629 zp->z_sa_hdl = NULL; 630} 631 632static void 633zfs_vnode_forget(vnode_t *vp) 634{ 635 636 /* copied from insmntque_stddtr */ 637 vp->v_data = NULL; 638 vp->v_op = &dead_vnodeops; 639 vgone(vp); 640 vput(vp); 641} 642 643/* 644 * Construct a new znode/vnode and intialize. 645 * 646 * This does not do a call to dmu_set_user() that is 647 * up to the caller to do, in case you don't want to 648 * return the znode 649 */ 650static znode_t * 651zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 652 dmu_object_type_t obj_type, sa_handle_t *hdl) 653{ 654 znode_t *zp; 655 vnode_t *vp; 656 uint64_t mode; 657 uint64_t parent; 658 sa_bulk_attr_t bulk[9]; 659 int count = 0; 660 661 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 662 zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0); 663 664 ASSERT(zp->z_dirlocks == NULL); 665 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 666 zp->z_moved = 0; 667 668 /* 669 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 670 * the zfs_znode_move() callback. 671 */ 672 zp->z_sa_hdl = NULL; 673 zp->z_unlinked = 0; 674 zp->z_atime_dirty = 0; 675 zp->z_mapcnt = 0; 676 zp->z_id = db->db_object; 677 zp->z_blksz = blksz; 678 zp->z_seq = 0x7A4653; 679 zp->z_sync_cnt = 0; 680 681 vp = ZTOV(zp); 682 683 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 684 685 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 686 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 687 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 688 &zp->z_size, 8); 689 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 690 &zp->z_links, 8); 691 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 692 &zp->z_pflags, 8); 693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 695 &zp->z_atime, 16); 696 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 697 &zp->z_uid, 8); 698 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 699 &zp->z_gid, 8); 700 701 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 702 if (hdl == NULL) 703 sa_handle_destroy(zp->z_sa_hdl); 704 zfs_vnode_forget(vp); 705 zp->z_vnode = NULL; 706 kmem_cache_free(znode_cache, zp); 707 return (NULL); 708 } 709 710 zp->z_mode = mode; 711 712 vp->v_type = IFTOVT((mode_t)mode); 713 714 switch (vp->v_type) { 715 case VDIR: 716 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 717 break; 718#ifdef sun 719 case VBLK: 720 case VCHR: 721 { 722 uint64_t rdev; 723 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 724 &rdev, sizeof (rdev)) == 0); 725 726 vp->v_rdev = zfs_cmpldev(rdev); 727 } 728 break; 729#endif /* sun */ 730 case VFIFO: 731#ifdef sun 732 case VSOCK: 733 case VDOOR: 734#endif /* sun */ 735 vp->v_op = &zfs_fifoops; 736 break; 737 case VREG: 738 if (parent == zfsvfs->z_shares_dir) { 739 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 740 vp->v_op = &zfs_shareops; 741 } 742 break; 743#ifdef sun 744 case VLNK: 745 vn_setops(vp, zfs_symvnodeops); 746 break; 747 default: 748 vn_setops(vp, zfs_evnodeops); 749 break; 750#endif /* sun */ 751 } 752 if (vp->v_type != VFIFO) 753 VN_LOCK_ASHARE(vp); 754 755 mutex_enter(&zfsvfs->z_znodes_lock); 756 list_insert_tail(&zfsvfs->z_all_znodes, zp); 757 membar_producer(); 758 /* 759 * Everything else must be valid before assigning z_zfsvfs makes the 760 * znode eligible for zfs_znode_move(). 761 */ 762 zp->z_zfsvfs = zfsvfs; 763 mutex_exit(&zfsvfs->z_znodes_lock); 764 765 VFS_HOLD(zfsvfs->z_vfs); 766 return (zp); 767} 768 769static uint64_t empty_xattr; 770static uint64_t pad[4]; 771static zfs_acl_phys_t acl_phys; 772/* 773 * Create a new DMU object to hold a zfs znode. 774 * 775 * IN: dzp - parent directory for new znode 776 * vap - file attributes for new znode 777 * tx - dmu transaction id for zap operations 778 * cr - credentials of caller 779 * flag - flags: 780 * IS_ROOT_NODE - new object will be root 781 * IS_XATTR - new object is an attribute 782 * bonuslen - length of bonus buffer 783 * setaclp - File/Dir initial ACL 784 * fuidp - Tracks fuid allocation. 785 * 786 * OUT: zpp - allocated znode 787 * 788 */ 789void 790zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 791 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 792{ 793 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 794 uint64_t mode, size, links, parent, pflags; 795 uint64_t dzp_pflags = 0; 796 uint64_t rdev = 0; 797 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 798 dmu_buf_t *db; 799 timestruc_t now; 800 uint64_t gen, obj; 801 int err; 802 int bonuslen; 803 sa_handle_t *sa_hdl; 804 dmu_object_type_t obj_type; 805 sa_bulk_attr_t sa_attrs[ZPL_END]; 806 int cnt = 0; 807 zfs_acl_locator_cb_t locate = { 0 }; 808 809 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 810 811 if (zfsvfs->z_replay) { 812 obj = vap->va_nodeid; 813 now = vap->va_ctime; /* see zfs_replay_create() */ 814 gen = vap->va_nblocks; /* ditto */ 815 } else { 816 obj = 0; 817 gethrestime(&now); 818 gen = dmu_tx_get_txg(tx); 819 } 820 821 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 822 bonuslen = (obj_type == DMU_OT_SA) ? 823 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 824 825 /* 826 * Create a new DMU object. 827 */ 828 /* 829 * There's currently no mechanism for pre-reading the blocks that will 830 * be needed to allocate a new object, so we accept the small chance 831 * that there will be an i/o error and we will fail one of the 832 * assertions below. 833 */ 834 if (vap->va_type == VDIR) { 835 if (zfsvfs->z_replay) { 836 err = zap_create_claim_norm(zfsvfs->z_os, obj, 837 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 838 obj_type, bonuslen, tx); 839 ASSERT0(err); 840 } else { 841 obj = zap_create_norm(zfsvfs->z_os, 842 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 843 obj_type, bonuslen, tx); 844 } 845 } else { 846 if (zfsvfs->z_replay) { 847 err = dmu_object_claim(zfsvfs->z_os, obj, 848 DMU_OT_PLAIN_FILE_CONTENTS, 0, 849 obj_type, bonuslen, tx); 850 ASSERT0(err); 851 } else { 852 obj = dmu_object_alloc(zfsvfs->z_os, 853 DMU_OT_PLAIN_FILE_CONTENTS, 0, 854 obj_type, bonuslen, tx); 855 } 856 } 857 858 getnewvnode_reserve(1); 859 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 860 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 861 862 /* 863 * If this is the root, fix up the half-initialized parent pointer 864 * to reference the just-allocated physical data area. 865 */ 866 if (flag & IS_ROOT_NODE) { 867 dzp->z_id = obj; 868 } else { 869 dzp_pflags = dzp->z_pflags; 870 } 871 872 /* 873 * If parent is an xattr, so am I. 874 */ 875 if (dzp_pflags & ZFS_XATTR) { 876 flag |= IS_XATTR; 877 } 878 879 if (zfsvfs->z_use_fuids) 880 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 881 else 882 pflags = 0; 883 884 if (vap->va_type == VDIR) { 885 size = 2; /* contents ("." and "..") */ 886 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 887 } else { 888 size = links = 0; 889 } 890 891 if (vap->va_type == VBLK || vap->va_type == VCHR) { 892 rdev = zfs_expldev(vap->va_rdev); 893 } 894 895 parent = dzp->z_id; 896 mode = acl_ids->z_mode; 897 if (flag & IS_XATTR) 898 pflags |= ZFS_XATTR; 899 900 /* 901 * No execs denied will be deterimed when zfs_mode_compute() is called. 902 */ 903 pflags |= acl_ids->z_aclp->z_hints & 904 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 905 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 906 907 ZFS_TIME_ENCODE(&now, crtime); 908 ZFS_TIME_ENCODE(&now, ctime); 909 910 if (vap->va_mask & AT_ATIME) { 911 ZFS_TIME_ENCODE(&vap->va_atime, atime); 912 } else { 913 ZFS_TIME_ENCODE(&now, atime); 914 } 915 916 if (vap->va_mask & AT_MTIME) { 917 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 918 } else { 919 ZFS_TIME_ENCODE(&now, mtime); 920 } 921 922 /* Now add in all of the "SA" attributes */ 923 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 924 &sa_hdl)); 925 926 /* 927 * Setup the array of attributes to be replaced/set on the new file 928 * 929 * order for DMU_OT_ZNODE is critical since it needs to be constructed 930 * in the old znode_phys_t format. Don't change this ordering 931 */ 932 933 if (obj_type == DMU_OT_ZNODE) { 934 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 935 NULL, &atime, 16); 936 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 937 NULL, &mtime, 16); 938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 939 NULL, &ctime, 16); 940 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 941 NULL, &crtime, 16); 942 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 943 NULL, &gen, 8); 944 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 945 NULL, &mode, 8); 946 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 947 NULL, &size, 8); 948 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 949 NULL, &parent, 8); 950 } else { 951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 952 NULL, &mode, 8); 953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 954 NULL, &size, 8); 955 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 956 NULL, &gen, 8); 957 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 958 &acl_ids->z_fuid, 8); 959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 960 &acl_ids->z_fgid, 8); 961 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 962 NULL, &parent, 8); 963 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 964 NULL, &pflags, 8); 965 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 966 NULL, &atime, 16); 967 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 968 NULL, &mtime, 16); 969 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 970 NULL, &ctime, 16); 971 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 972 NULL, &crtime, 16); 973 } 974 975 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 976 977 if (obj_type == DMU_OT_ZNODE) { 978 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 979 &empty_xattr, 8); 980 } 981 if (obj_type == DMU_OT_ZNODE || 982 (vap->va_type == VBLK || vap->va_type == VCHR)) { 983 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 984 NULL, &rdev, 8); 985 986 } 987 if (obj_type == DMU_OT_ZNODE) { 988 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 989 NULL, &pflags, 8); 990 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 991 &acl_ids->z_fuid, 8); 992 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 993 &acl_ids->z_fgid, 8); 994 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 995 sizeof (uint64_t) * 4); 996 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 997 &acl_phys, sizeof (zfs_acl_phys_t)); 998 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 999 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 1000 &acl_ids->z_aclp->z_acl_count, 8); 1001 locate.cb_aclp = acl_ids->z_aclp; 1002 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 1003 zfs_acl_data_locator, &locate, 1004 acl_ids->z_aclp->z_acl_bytes); 1005 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 1006 acl_ids->z_fuid, acl_ids->z_fgid); 1007 } 1008 1009 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 1010 1011 if (!(flag & IS_ROOT_NODE)) { 1012 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 1013 ASSERT(*zpp != NULL); 1014 } else { 1015 /* 1016 * If we are creating the root node, the "parent" we 1017 * passed in is the znode for the root. 1018 */ 1019 *zpp = dzp; 1020 1021 (*zpp)->z_sa_hdl = sa_hdl; 1022 } 1023 1024 (*zpp)->z_pflags = pflags; 1025 (*zpp)->z_mode = mode; 1026 1027 if (vap->va_mask & AT_XVATTR) 1028 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 1029 1030 if (obj_type == DMU_OT_ZNODE || 1031 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1032 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx); 1033 ASSERT0(err); 1034 } 1035 if (!(flag & IS_ROOT_NODE)) { 1036 vnode_t *vp; 1037 1038 vp = ZTOV(*zpp); 1039 vp->v_vflag |= VV_FORCEINSMQ; 1040 err = insmntque(vp, zfsvfs->z_vfs); 1041 vp->v_vflag &= ~VV_FORCEINSMQ; 1042 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1043 } 1044 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1045 getnewvnode_drop_reserve(); 1046} 1047 1048/* 1049 * zfs_xvattr_set only updates the in-core attributes 1050 * it is assumed the caller will be doing an sa_bulk_update 1051 * to push the changes out 1052 */ 1053void 1054zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1055{ 1056 xoptattr_t *xoap; 1057 1058 xoap = xva_getxoptattr(xvap); 1059 ASSERT(xoap); 1060 1061 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1062 uint64_t times[2]; 1063 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1064 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1065 ×, sizeof (times), tx); 1066 XVA_SET_RTN(xvap, XAT_CREATETIME); 1067 } 1068 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1069 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1070 zp->z_pflags, tx); 1071 XVA_SET_RTN(xvap, XAT_READONLY); 1072 } 1073 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1074 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1075 zp->z_pflags, tx); 1076 XVA_SET_RTN(xvap, XAT_HIDDEN); 1077 } 1078 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1079 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1080 zp->z_pflags, tx); 1081 XVA_SET_RTN(xvap, XAT_SYSTEM); 1082 } 1083 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1084 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1085 zp->z_pflags, tx); 1086 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1087 } 1088 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1089 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1090 zp->z_pflags, tx); 1091 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1092 } 1093 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1094 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1095 zp->z_pflags, tx); 1096 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1097 } 1098 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1099 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1100 zp->z_pflags, tx); 1101 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1102 } 1103 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1104 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1105 zp->z_pflags, tx); 1106 XVA_SET_RTN(xvap, XAT_NODUMP); 1107 } 1108 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1109 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1110 zp->z_pflags, tx); 1111 XVA_SET_RTN(xvap, XAT_OPAQUE); 1112 } 1113 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1114 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1115 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1116 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1117 } 1118 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1119 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1120 zp->z_pflags, tx); 1121 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1122 } 1123 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1124 zfs_sa_set_scanstamp(zp, xvap, tx); 1125 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1126 } 1127 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1128 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1129 zp->z_pflags, tx); 1130 XVA_SET_RTN(xvap, XAT_REPARSE); 1131 } 1132 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1133 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1134 zp->z_pflags, tx); 1135 XVA_SET_RTN(xvap, XAT_OFFLINE); 1136 } 1137 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1138 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1139 zp->z_pflags, tx); 1140 XVA_SET_RTN(xvap, XAT_SPARSE); 1141 } 1142} 1143 1144int 1145zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1146{ 1147 dmu_object_info_t doi; 1148 dmu_buf_t *db; 1149 znode_t *zp; 1150 vnode_t *vp; 1151 sa_handle_t *hdl; 1152 struct thread *td; 1153 int locked; 1154 int err; 1155 1156 td = curthread; 1157 getnewvnode_reserve(1); 1158again: 1159 *zpp = NULL; 1160 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1161 1162 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1163 if (err) { 1164 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1165 getnewvnode_drop_reserve(); 1166 return (err); 1167 } 1168 1169 dmu_object_info_from_db(db, &doi); 1170 if (doi.doi_bonus_type != DMU_OT_SA && 1171 (doi.doi_bonus_type != DMU_OT_ZNODE || 1172 (doi.doi_bonus_type == DMU_OT_ZNODE && 1173 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1174 sa_buf_rele(db, NULL); 1175 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1176#ifdef __FreeBSD__ 1177 getnewvnode_drop_reserve(); 1178#endif 1179 return (SET_ERROR(EINVAL)); 1180 } 1181 1182 hdl = dmu_buf_get_user(db); 1183 if (hdl != NULL) { 1184 zp = sa_get_userdata(hdl); 1185 1186 1187 /* 1188 * Since "SA" does immediate eviction we 1189 * should never find a sa handle that doesn't 1190 * know about the znode. 1191 */ 1192 1193 ASSERT3P(zp, !=, NULL); 1194 1195 mutex_enter(&zp->z_lock); 1196 ASSERT3U(zp->z_id, ==, obj_num); 1197 if (zp->z_unlinked) { 1198 err = SET_ERROR(ENOENT); 1199 } else { 1200 vp = ZTOV(zp); 1201 *zpp = zp; 1202 err = 0; 1203 } 1204 sa_buf_rele(db, NULL); 1205 1206 /* Don't let the vnode disappear after ZFS_OBJ_HOLD_EXIT. */ 1207 if (err == 0) 1208 VN_HOLD(vp); 1209 1210 mutex_exit(&zp->z_lock); 1211 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1212 1213 if (err == 0) { 1214 locked = VOP_ISLOCKED(vp); 1215 VI_LOCK(vp); 1216 if ((vp->v_iflag & VI_DOOMED) != 0 && 1217 locked != LK_EXCLUSIVE) { 1218 /* 1219 * The vnode is doomed and this thread doesn't 1220 * hold the exclusive lock on it, so the vnode 1221 * must be being reclaimed by another thread. 1222 * Otherwise the doomed vnode is being reclaimed 1223 * by this thread and zfs_zget is called from 1224 * ZIL internals. 1225 */ 1226 VI_UNLOCK(vp); 1227 VN_RELE(vp); 1228 goto again; 1229 } 1230 VI_UNLOCK(vp); 1231 } 1232 getnewvnode_drop_reserve(); 1233 return (err); 1234 } 1235 1236 /* 1237 * Not found create new znode/vnode 1238 * but only if file exists. 1239 * 1240 * There is a small window where zfs_vget() could 1241 * find this object while a file create is still in 1242 * progress. This is checked for in zfs_znode_alloc() 1243 * 1244 * if zfs_znode_alloc() fails it will drop the hold on the 1245 * bonus buffer. 1246 */ 1247 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1248 doi.doi_bonus_type, NULL); 1249 if (zp == NULL) { 1250 err = SET_ERROR(ENOENT); 1251 } else { 1252 *zpp = zp; 1253 } 1254 if (err == 0) { 1255 vnode_t *vp = ZTOV(zp); 1256 1257 err = insmntque(vp, zfsvfs->z_vfs); 1258 if (err == 0) 1259 VOP_UNLOCK(vp, 0); 1260 else { 1261 zp->z_vnode = NULL; 1262 zfs_znode_dmu_fini(zp); 1263 zfs_znode_free(zp); 1264 *zpp = NULL; 1265 } 1266 } 1267 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1268 getnewvnode_drop_reserve(); 1269 return (err); 1270} 1271 1272int 1273zfs_rezget(znode_t *zp) 1274{ 1275 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1276 dmu_object_info_t doi; 1277 dmu_buf_t *db; 1278 vnode_t *vp; 1279 uint64_t obj_num = zp->z_id; 1280 uint64_t mode, size; 1281 sa_bulk_attr_t bulk[8]; 1282 int err; 1283 int count = 0; 1284 uint64_t gen; 1285 1286 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1287 1288 mutex_enter(&zp->z_acl_lock); 1289 if (zp->z_acl_cached) { 1290 zfs_acl_free(zp->z_acl_cached); 1291 zp->z_acl_cached = NULL; 1292 } 1293 1294 mutex_exit(&zp->z_acl_lock); 1295 ASSERT(zp->z_sa_hdl == NULL); 1296 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1297 if (err) { 1298 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1299 return (err); 1300 } 1301 1302 dmu_object_info_from_db(db, &doi); 1303 if (doi.doi_bonus_type != DMU_OT_SA && 1304 (doi.doi_bonus_type != DMU_OT_ZNODE || 1305 (doi.doi_bonus_type == DMU_OT_ZNODE && 1306 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1307 sa_buf_rele(db, NULL); 1308 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1309 return (SET_ERROR(EINVAL)); 1310 } 1311 1312 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1313 size = zp->z_size; 1314 1315 /* reload cached values */ 1316 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1317 &gen, sizeof (gen)); 1318 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1319 &zp->z_size, sizeof (zp->z_size)); 1320 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1321 &zp->z_links, sizeof (zp->z_links)); 1322 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1323 &zp->z_pflags, sizeof (zp->z_pflags)); 1324 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1325 &zp->z_atime, sizeof (zp->z_atime)); 1326 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1327 &zp->z_uid, sizeof (zp->z_uid)); 1328 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1329 &zp->z_gid, sizeof (zp->z_gid)); 1330 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1331 &mode, sizeof (mode)); 1332 1333 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1334 zfs_znode_dmu_fini(zp); 1335 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1336 return (SET_ERROR(EIO)); 1337 } 1338 1339 zp->z_mode = mode; 1340 1341 if (gen != zp->z_gen) { 1342 zfs_znode_dmu_fini(zp); 1343 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1344 return (SET_ERROR(EIO)); 1345 } 1346 1347 /* 1348 * XXXPJD: Not sure how is that possible, but under heavy 1349 * zfs recv -F load it happens that z_gen is the same, but 1350 * vnode type is different than znode type. This would mean 1351 * that for example regular file was replaced with directory 1352 * which has the same object number. 1353 */ 1354 vp = ZTOV(zp); 1355 if (vp != NULL && 1356 vp->v_type != IFTOVT((mode_t)zp->z_mode)) { 1357 zfs_znode_dmu_fini(zp); 1358 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1359 return (EIO); 1360 } 1361 1362 zp->z_unlinked = (zp->z_links == 0); 1363 zp->z_blksz = doi.doi_data_block_size; 1364 if (vp != NULL) { 1365 vn_pages_remove(vp, 0, 0); 1366 if (zp->z_size != size) 1367 vnode_pager_setsize(vp, zp->z_size); 1368 } 1369 1370 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1371 1372 return (0); 1373} 1374 1375void 1376zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1377{ 1378 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1379 objset_t *os = zfsvfs->z_os; 1380 uint64_t obj = zp->z_id; 1381 uint64_t acl_obj = zfs_external_acl(zp); 1382 1383 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1384 if (acl_obj) { 1385 VERIFY(!zp->z_is_sa); 1386 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1387 } 1388 VERIFY(0 == dmu_object_free(os, obj, tx)); 1389 zfs_znode_dmu_fini(zp); 1390 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1391 zfs_znode_free(zp); 1392} 1393 1394void 1395zfs_zinactive(znode_t *zp) 1396{ 1397 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1398 uint64_t z_id = zp->z_id; 1399 1400 ASSERT(zp->z_sa_hdl); 1401 1402 /* 1403 * Don't allow a zfs_zget() while were trying to release this znode 1404 */ 1405 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1406 1407 mutex_enter(&zp->z_lock); 1408 1409 /* 1410 * If this was the last reference to a file with no links, 1411 * remove the file from the file system. 1412 */ 1413 if (zp->z_unlinked) { 1414 mutex_exit(&zp->z_lock); 1415 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1416 zfs_rmnode(zp); 1417 return; 1418 } 1419 1420 mutex_exit(&zp->z_lock); 1421 zfs_znode_dmu_fini(zp); 1422 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1423 zfs_znode_free(zp); 1424} 1425 1426void 1427zfs_znode_free(znode_t *zp) 1428{ 1429 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1430 1431 ASSERT(zp->z_sa_hdl == NULL); 1432 zp->z_vnode = NULL; 1433 mutex_enter(&zfsvfs->z_znodes_lock); 1434 POINTER_INVALIDATE(&zp->z_zfsvfs); 1435 list_remove(&zfsvfs->z_all_znodes, zp); 1436 mutex_exit(&zfsvfs->z_znodes_lock); 1437 1438 if (zp->z_acl_cached) { 1439 zfs_acl_free(zp->z_acl_cached); 1440 zp->z_acl_cached = NULL; 1441 } 1442 1443 kmem_cache_free(znode_cache, zp); 1444 1445 VFS_RELE(zfsvfs->z_vfs); 1446} 1447 1448void 1449zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1450 uint64_t ctime[2], boolean_t have_tx) 1451{ 1452 timestruc_t now; 1453 1454 gethrestime(&now); 1455 1456 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1457 zp->z_atime_dirty = 0; 1458 zp->z_seq++; 1459 } else { 1460 zp->z_atime_dirty = 1; 1461 } 1462 1463 if (flag & AT_ATIME) { 1464 ZFS_TIME_ENCODE(&now, zp->z_atime); 1465 } 1466 1467 if (flag & AT_MTIME) { 1468 ZFS_TIME_ENCODE(&now, mtime); 1469 if (zp->z_zfsvfs->z_use_fuids) { 1470 zp->z_pflags |= (ZFS_ARCHIVE | 1471 ZFS_AV_MODIFIED); 1472 } 1473 } 1474 1475 if (flag & AT_CTIME) { 1476 ZFS_TIME_ENCODE(&now, ctime); 1477 if (zp->z_zfsvfs->z_use_fuids) 1478 zp->z_pflags |= ZFS_ARCHIVE; 1479 } 1480} 1481 1482/* 1483 * Grow the block size for a file. 1484 * 1485 * IN: zp - znode of file to free data in. 1486 * size - requested block size 1487 * tx - open transaction. 1488 * 1489 * NOTE: this function assumes that the znode is write locked. 1490 */ 1491void 1492zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1493{ 1494 int error; 1495 u_longlong_t dummy; 1496 1497 if (size <= zp->z_blksz) 1498 return; 1499 /* 1500 * If the file size is already greater than the current blocksize, 1501 * we will not grow. If there is more than one block in a file, 1502 * the blocksize cannot change. 1503 */ 1504 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1505 return; 1506 1507 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1508 size, 0, tx); 1509 1510 if (error == ENOTSUP) 1511 return; 1512 ASSERT0(error); 1513 1514 /* What blocksize did we actually get? */ 1515 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1516} 1517 1518#ifdef sun 1519/* 1520 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1521 * be calling back into the fs for a putpage(). E.g.: when truncating 1522 * a file, the pages being "thrown away* don't need to be written out. 1523 */ 1524/* ARGSUSED */ 1525static int 1526zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1527 int flags, cred_t *cr) 1528{ 1529 ASSERT(0); 1530 return (0); 1531} 1532#endif /* sun */ 1533 1534/* 1535 * Increase the file length 1536 * 1537 * IN: zp - znode of file to free data in. 1538 * end - new end-of-file 1539 * 1540 * RETURN: 0 if success 1541 * error code if failure 1542 */ 1543static int 1544zfs_extend(znode_t *zp, uint64_t end) 1545{ 1546 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1547 dmu_tx_t *tx; 1548 rl_t *rl; 1549 uint64_t newblksz; 1550 int error; 1551 1552 /* 1553 * We will change zp_size, lock the whole file. 1554 */ 1555 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1556 1557 /* 1558 * Nothing to do if file already at desired length. 1559 */ 1560 if (end <= zp->z_size) { 1561 zfs_range_unlock(rl); 1562 return (0); 1563 } 1564top: 1565 tx = dmu_tx_create(zfsvfs->z_os); 1566 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1567 zfs_sa_upgrade_txholds(tx, zp); 1568 if (end > zp->z_blksz && 1569 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1570 /* 1571 * We are growing the file past the current block size. 1572 */ 1573 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1574 ASSERT(!ISP2(zp->z_blksz)); 1575 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1576 } else { 1577 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1578 } 1579 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1580 } else { 1581 newblksz = 0; 1582 } 1583 1584 error = dmu_tx_assign(tx, TXG_NOWAIT); 1585 if (error) { 1586 if (error == ERESTART) { 1587 dmu_tx_wait(tx); 1588 dmu_tx_abort(tx); 1589 goto top; 1590 } 1591 dmu_tx_abort(tx); 1592 zfs_range_unlock(rl); 1593 return (error); 1594 } 1595 1596 if (newblksz) 1597 zfs_grow_blocksize(zp, newblksz, tx); 1598 1599 zp->z_size = end; 1600 1601 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1602 &zp->z_size, sizeof (zp->z_size), tx)); 1603 1604 vnode_pager_setsize(ZTOV(zp), end); 1605 1606 zfs_range_unlock(rl); 1607 1608 dmu_tx_commit(tx); 1609 1610 return (0); 1611} 1612 1613/* 1614 * Free space in a file. 1615 * 1616 * IN: zp - znode of file to free data in. 1617 * off - start of section to free. 1618 * len - length of section to free. 1619 * 1620 * RETURN: 0 if success 1621 * error code if failure 1622 */ 1623static int 1624zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1625{ 1626 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1627 rl_t *rl; 1628 int error; 1629 1630 /* 1631 * Lock the range being freed. 1632 */ 1633 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1634 1635 /* 1636 * Nothing to do if file already at desired length. 1637 */ 1638 if (off >= zp->z_size) { 1639 zfs_range_unlock(rl); 1640 return (0); 1641 } 1642 1643 if (off + len > zp->z_size) 1644 len = zp->z_size - off; 1645 1646 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1647 1648 if (error == 0) { 1649 /* 1650 * In FreeBSD we cannot free block in the middle of a file, 1651 * but only at the end of a file, so this code path should 1652 * never happen. 1653 */ 1654 vnode_pager_setsize(ZTOV(zp), off); 1655 } 1656 1657 zfs_range_unlock(rl); 1658 1659 return (error); 1660} 1661 1662/* 1663 * Truncate a file 1664 * 1665 * IN: zp - znode of file to free data in. 1666 * end - new end-of-file. 1667 * 1668 * RETURN: 0 if success 1669 * error code if failure 1670 */ 1671static int 1672zfs_trunc(znode_t *zp, uint64_t end) 1673{ 1674 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1675 vnode_t *vp = ZTOV(zp); 1676 dmu_tx_t *tx; 1677 rl_t *rl; 1678 int error; 1679 sa_bulk_attr_t bulk[2]; 1680 int count = 0; 1681 1682 /* 1683 * We will change zp_size, lock the whole file. 1684 */ 1685 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1686 1687 /* 1688 * Nothing to do if file already at desired length. 1689 */ 1690 if (end >= zp->z_size) { 1691 zfs_range_unlock(rl); 1692 return (0); 1693 } 1694 1695 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1696 if (error) { 1697 zfs_range_unlock(rl); 1698 return (error); 1699 } 1700top: 1701 tx = dmu_tx_create(zfsvfs->z_os); 1702 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1703 zfs_sa_upgrade_txholds(tx, zp); 1704 error = dmu_tx_assign(tx, TXG_NOWAIT); 1705 if (error) { 1706 if (error == ERESTART) { 1707 dmu_tx_wait(tx); 1708 dmu_tx_abort(tx); 1709 goto top; 1710 } 1711 dmu_tx_abort(tx); 1712 zfs_range_unlock(rl); 1713 return (error); 1714 } 1715 1716 zp->z_size = end; 1717 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1718 NULL, &zp->z_size, sizeof (zp->z_size)); 1719 1720 if (end == 0) { 1721 zp->z_pflags &= ~ZFS_SPARSE; 1722 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1723 NULL, &zp->z_pflags, 8); 1724 } 1725 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1726 1727 dmu_tx_commit(tx); 1728 1729 /* 1730 * Clear any mapped pages in the truncated region. This has to 1731 * happen outside of the transaction to avoid the possibility of 1732 * a deadlock with someone trying to push a page that we are 1733 * about to invalidate. 1734 */ 1735 vnode_pager_setsize(vp, end); 1736 1737 zfs_range_unlock(rl); 1738 1739 return (0); 1740} 1741 1742/* 1743 * Free space in a file 1744 * 1745 * IN: zp - znode of file to free data in. 1746 * off - start of range 1747 * len - end of range (0 => EOF) 1748 * flag - current file open mode flags. 1749 * log - TRUE if this action should be logged 1750 * 1751 * RETURN: 0 if success 1752 * error code if failure 1753 */ 1754int 1755zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1756{ 1757 vnode_t *vp = ZTOV(zp); 1758 dmu_tx_t *tx; 1759 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1760 zilog_t *zilog = zfsvfs->z_log; 1761 uint64_t mode; 1762 uint64_t mtime[2], ctime[2]; 1763 sa_bulk_attr_t bulk[3]; 1764 int count = 0; 1765 int error; 1766 1767 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1768 sizeof (mode))) != 0) 1769 return (error); 1770 1771 if (off > zp->z_size) { 1772 error = zfs_extend(zp, off+len); 1773 if (error == 0 && log) 1774 goto log; 1775 else 1776 return (error); 1777 } 1778 1779 /* 1780 * Check for any locks in the region to be freed. 1781 */ 1782 1783 if (MANDLOCK(vp, (mode_t)mode)) { 1784 uint64_t length = (len ? len : zp->z_size - off); 1785 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1786 return (error); 1787 } 1788 1789 if (len == 0) { 1790 error = zfs_trunc(zp, off); 1791 } else { 1792 if ((error = zfs_free_range(zp, off, len)) == 0 && 1793 off + len > zp->z_size) 1794 error = zfs_extend(zp, off+len); 1795 } 1796 if (error || !log) 1797 return (error); 1798log: 1799 tx = dmu_tx_create(zfsvfs->z_os); 1800 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1801 zfs_sa_upgrade_txholds(tx, zp); 1802 error = dmu_tx_assign(tx, TXG_NOWAIT); 1803 if (error) { 1804 if (error == ERESTART) { 1805 dmu_tx_wait(tx); 1806 dmu_tx_abort(tx); 1807 goto log; 1808 } 1809 dmu_tx_abort(tx); 1810 return (error); 1811 } 1812 1813 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1814 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1815 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1816 NULL, &zp->z_pflags, 8); 1817 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1818 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1819 ASSERT(error == 0); 1820 1821 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1822 1823 dmu_tx_commit(tx); 1824 return (0); 1825} 1826 1827void 1828zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1829{ 1830 zfsvfs_t zfsvfs; 1831 uint64_t moid, obj, sa_obj, version; 1832 uint64_t sense = ZFS_CASE_SENSITIVE; 1833 uint64_t norm = 0; 1834 nvpair_t *elem; 1835 int error; 1836 int i; 1837 znode_t *rootzp = NULL; 1838 vnode_t vnode; 1839 vattr_t vattr; 1840 znode_t *zp; 1841 zfs_acl_ids_t acl_ids; 1842 1843 /* 1844 * First attempt to create master node. 1845 */ 1846 /* 1847 * In an empty objset, there are no blocks to read and thus 1848 * there can be no i/o errors (which we assert below). 1849 */ 1850 moid = MASTER_NODE_OBJ; 1851 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1852 DMU_OT_NONE, 0, tx); 1853 ASSERT(error == 0); 1854 1855 /* 1856 * Set starting attributes. 1857 */ 1858 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1859 elem = NULL; 1860 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1861 /* For the moment we expect all zpl props to be uint64_ts */ 1862 uint64_t val; 1863 char *name; 1864 1865 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1866 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1867 name = nvpair_name(elem); 1868 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1869 if (val < version) 1870 version = val; 1871 } else { 1872 error = zap_update(os, moid, name, 8, 1, &val, tx); 1873 } 1874 ASSERT(error == 0); 1875 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1876 norm = val; 1877 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1878 sense = val; 1879 } 1880 ASSERT(version != 0); 1881 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1882 1883 /* 1884 * Create zap object used for SA attribute registration 1885 */ 1886 1887 if (version >= ZPL_VERSION_SA) { 1888 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1889 DMU_OT_NONE, 0, tx); 1890 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1891 ASSERT(error == 0); 1892 } else { 1893 sa_obj = 0; 1894 } 1895 /* 1896 * Create a delete queue. 1897 */ 1898 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1899 1900 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1901 ASSERT(error == 0); 1902 1903 /* 1904 * Create root znode. Create minimal znode/vnode/zfsvfs 1905 * to allow zfs_mknode to work. 1906 */ 1907 VATTR_NULL(&vattr); 1908 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1909 vattr.va_type = VDIR; 1910 vattr.va_mode = S_IFDIR|0755; 1911 vattr.va_uid = crgetuid(cr); 1912 vattr.va_gid = crgetgid(cr); 1913 1914 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1915 1916 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1917 zfs_znode_cache_constructor(rootzp, NULL, 0); 1918 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1919 rootzp->z_moved = 0; 1920 rootzp->z_unlinked = 0; 1921 rootzp->z_atime_dirty = 0; 1922 rootzp->z_is_sa = USE_SA(version, os); 1923 1924 vnode.v_type = VDIR; 1925 vnode.v_data = rootzp; 1926 rootzp->z_vnode = &vnode; 1927 1928 zfsvfs.z_os = os; 1929 zfsvfs.z_parent = &zfsvfs; 1930 zfsvfs.z_version = version; 1931 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1932 zfsvfs.z_use_sa = USE_SA(version, os); 1933 zfsvfs.z_norm = norm; 1934 1935 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1936 &zfsvfs.z_attr_table); 1937 1938 ASSERT(error == 0); 1939 1940 /* 1941 * Fold case on file systems that are always or sometimes case 1942 * insensitive. 1943 */ 1944 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1945 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1946 1947 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1948 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1949 offsetof(znode_t, z_link_node)); 1950 1951 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1952 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1953 1954 rootzp->z_zfsvfs = &zfsvfs; 1955 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1956 cr, NULL, &acl_ids)); 1957 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1958 ASSERT3P(zp, ==, rootzp); 1959 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1960 ASSERT(error == 0); 1961 zfs_acl_ids_free(&acl_ids); 1962 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1963 1964 sa_handle_destroy(rootzp->z_sa_hdl); 1965 rootzp->z_vnode = NULL; 1966 kmem_cache_free(znode_cache, rootzp); 1967 1968 /* 1969 * Create shares directory 1970 */ 1971 1972 error = zfs_create_share_dir(&zfsvfs, tx); 1973 1974 ASSERT(error == 0); 1975 1976 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1977 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1978} 1979 1980#endif /* _KERNEL */ 1981 1982static int 1983zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1984{ 1985 uint64_t sa_obj = 0; 1986 int error; 1987 1988 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1989 if (error != 0 && error != ENOENT) 1990 return (error); 1991 1992 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1993 return (error); 1994} 1995 1996static int 1997zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1998 dmu_buf_t **db, void *tag) 1999{ 2000 dmu_object_info_t doi; 2001 int error; 2002 2003 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 2004 return (error); 2005 2006 dmu_object_info_from_db(*db, &doi); 2007 if ((doi.doi_bonus_type != DMU_OT_SA && 2008 doi.doi_bonus_type != DMU_OT_ZNODE) || 2009 doi.doi_bonus_type == DMU_OT_ZNODE && 2010 doi.doi_bonus_size < sizeof (znode_phys_t)) { 2011 sa_buf_rele(*db, tag); 2012 return (SET_ERROR(ENOTSUP)); 2013 } 2014 2015 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 2016 if (error != 0) { 2017 sa_buf_rele(*db, tag); 2018 return (error); 2019 } 2020 2021 return (0); 2022} 2023 2024void 2025zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 2026{ 2027 sa_handle_destroy(hdl); 2028 sa_buf_rele(db, tag); 2029} 2030 2031/* 2032 * Given an object number, return its parent object number and whether 2033 * or not the object is an extended attribute directory. 2034 */ 2035static int 2036zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table, 2037 uint64_t *pobjp, int *is_xattrdir) 2038{ 2039 uint64_t parent; 2040 uint64_t pflags; 2041 uint64_t mode; 2042 uint64_t parent_mode; 2043 sa_bulk_attr_t bulk[3]; 2044 sa_handle_t *sa_hdl; 2045 dmu_buf_t *sa_db; 2046 int count = 0; 2047 int error; 2048 2049 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2050 &parent, sizeof (parent)); 2051 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2052 &pflags, sizeof (pflags)); 2053 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2054 &mode, sizeof (mode)); 2055 2056 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2057 return (error); 2058 2059 /* 2060 * When a link is removed its parent pointer is not changed and will 2061 * be invalid. There are two cases where a link is removed but the 2062 * file stays around, when it goes to the delete queue and when there 2063 * are additional links. 2064 */ 2065 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG); 2066 if (error != 0) 2067 return (error); 2068 2069 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode)); 2070 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2071 if (error != 0) 2072 return (error); 2073 2074 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2075 2076 /* 2077 * Extended attributes can be applied to files, directories, etc. 2078 * Otherwise the parent must be a directory. 2079 */ 2080 if (!*is_xattrdir && !S_ISDIR(parent_mode)) 2081 return (SET_ERROR(EINVAL)); 2082 2083 *pobjp = parent; 2084 2085 return (0); 2086} 2087 2088/* 2089 * Given an object number, return some zpl level statistics 2090 */ 2091static int 2092zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2093 zfs_stat_t *sb) 2094{ 2095 sa_bulk_attr_t bulk[4]; 2096 int count = 0; 2097 2098 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2099 &sb->zs_mode, sizeof (sb->zs_mode)); 2100 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2101 &sb->zs_gen, sizeof (sb->zs_gen)); 2102 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2103 &sb->zs_links, sizeof (sb->zs_links)); 2104 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2105 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2106 2107 return (sa_bulk_lookup(hdl, bulk, count)); 2108} 2109 2110static int 2111zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2112 sa_attr_type_t *sa_table, char *buf, int len) 2113{ 2114 sa_handle_t *sa_hdl; 2115 sa_handle_t *prevhdl = NULL; 2116 dmu_buf_t *prevdb = NULL; 2117 dmu_buf_t *sa_db = NULL; 2118 char *path = buf + len - 1; 2119 int error; 2120 2121 *path = '\0'; 2122 sa_hdl = hdl; 2123 2124 for (;;) { 2125 uint64_t pobj; 2126 char component[MAXNAMELEN + 2]; 2127 size_t complen; 2128 int is_xattrdir; 2129 2130 if (prevdb) 2131 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2132 2133 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj, 2134 &is_xattrdir)) != 0) 2135 break; 2136 2137 if (pobj == obj) { 2138 if (path[0] != '/') 2139 *--path = '/'; 2140 break; 2141 } 2142 2143 component[0] = '/'; 2144 if (is_xattrdir) { 2145 (void) sprintf(component + 1, "<xattrdir>"); 2146 } else { 2147 error = zap_value_search(osp, pobj, obj, 2148 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2149 if (error != 0) 2150 break; 2151 } 2152 2153 complen = strlen(component); 2154 path -= complen; 2155 ASSERT(path >= buf); 2156 bcopy(component, path, complen); 2157 obj = pobj; 2158 2159 if (sa_hdl != hdl) { 2160 prevhdl = sa_hdl; 2161 prevdb = sa_db; 2162 } 2163 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2164 if (error != 0) { 2165 sa_hdl = prevhdl; 2166 sa_db = prevdb; 2167 break; 2168 } 2169 } 2170 2171 if (sa_hdl != NULL && sa_hdl != hdl) { 2172 ASSERT(sa_db != NULL); 2173 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2174 } 2175 2176 if (error == 0) 2177 (void) memmove(buf, path, buf + len - path); 2178 2179 return (error); 2180} 2181 2182int 2183zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2184{ 2185 sa_attr_type_t *sa_table; 2186 sa_handle_t *hdl; 2187 dmu_buf_t *db; 2188 int error; 2189 2190 error = zfs_sa_setup(osp, &sa_table); 2191 if (error != 0) 2192 return (error); 2193 2194 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2195 if (error != 0) 2196 return (error); 2197 2198 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2199 2200 zfs_release_sa_handle(hdl, db, FTAG); 2201 return (error); 2202} 2203 2204int 2205zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2206 char *buf, int len) 2207{ 2208 char *path = buf + len - 1; 2209 sa_attr_type_t *sa_table; 2210 sa_handle_t *hdl; 2211 dmu_buf_t *db; 2212 int error; 2213 2214 *path = '\0'; 2215 2216 error = zfs_sa_setup(osp, &sa_table); 2217 if (error != 0) 2218 return (error); 2219 2220 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2221 if (error != 0) 2222 return (error); 2223 2224 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2225 if (error != 0) { 2226 zfs_release_sa_handle(hdl, db, FTAG); 2227 return (error); 2228 } 2229 2230 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2231 2232 zfs_release_sa_handle(hdl, db, FTAG); 2233 return (error); 2234} 2235