zfs_znode.c revision 226732
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 http://www.opensolaris.org/os/licensing. 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 */ 24 25/* Portions Copyright 2007 Jeremy Teo */ 26/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 27 28#ifdef _KERNEL 29#include <sys/types.h> 30#include <sys/param.h> 31#include <sys/time.h> 32#include <sys/systm.h> 33#include <sys/sysmacros.h> 34#include <sys/resource.h> 35#include <sys/mntent.h> 36#include <sys/u8_textprep.h> 37#include <sys/dsl_dataset.h> 38#include <sys/vfs.h> 39#include <sys/vnode.h> 40#include <sys/file.h> 41#include <sys/kmem.h> 42#include <sys/errno.h> 43#include <sys/unistd.h> 44#include <sys/atomic.h> 45#include <sys/zfs_dir.h> 46#include <sys/zfs_acl.h> 47#include <sys/zfs_ioctl.h> 48#include <sys/zfs_rlock.h> 49#include <sys/zfs_fuid.h> 50#include <sys/dnode.h> 51#include <sys/fs/zfs.h> 52#include <sys/kidmap.h> 53#endif /* _KERNEL */ 54 55#include <sys/dmu.h> 56#include <sys/refcount.h> 57#include <sys/stat.h> 58#include <sys/zap.h> 59#include <sys/zfs_znode.h> 60#include <sys/sa.h> 61#include <sys/zfs_sa.h> 62#include <sys/zfs_stat.h> 63#include <sys/refcount.h> 64 65#include "zfs_prop.h" 66#include "zfs_comutil.h" 67 68/* Used by fstat(1). */ 69SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t), 70 "sizeof(znode_t)"); 71 72/* 73 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 74 * turned on when DEBUG is also defined. 75 */ 76#ifdef DEBUG 77#define ZNODE_STATS 78#endif /* DEBUG */ 79 80#ifdef ZNODE_STATS 81#define ZNODE_STAT_ADD(stat) ((stat)++) 82#else 83#define ZNODE_STAT_ADD(stat) /* nothing */ 84#endif /* ZNODE_STATS */ 85 86/* 87 * Functions needed for userland (ie: libzpool) are not put under 88 * #ifdef_KERNEL; the rest of the functions have dependencies 89 * (such as VFS logic) that will not compile easily in userland. 90 */ 91#ifdef _KERNEL 92/* 93 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 94 * be freed before it can be safely accessed. 95 */ 96krwlock_t zfsvfs_lock; 97 98static kmem_cache_t *znode_cache = NULL; 99 100/*ARGSUSED*/ 101static void 102znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 103{ 104 /* 105 * We should never drop all dbuf refs without first clearing 106 * the eviction callback. 107 */ 108 panic("evicting znode %p\n", user_ptr); 109} 110 111extern struct vop_vector zfs_vnodeops; 112extern struct vop_vector zfs_fifoops; 113extern struct vop_vector zfs_shareops; 114 115/* 116 * XXX: We cannot use this function as a cache constructor, because 117 * there is one global cache for all file systems and we need 118 * to pass vfsp here, which is not possible, because argument 119 * 'cdrarg' is defined at kmem_cache_create() time. 120 */ 121/*ARGSUSED*/ 122static int 123zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 124{ 125 znode_t *zp = buf; 126 vnode_t *vp; 127 vfs_t *vfsp = arg; 128 int error; 129 130 POINTER_INVALIDATE(&zp->z_zfsvfs); 131 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 132 133 if (vfsp != NULL) { 134 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp); 135 if (error != 0 && (kmflags & KM_NOSLEEP)) 136 return (-1); 137 ASSERT(error == 0); 138 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 139 zp->z_vnode = vp; 140 vp->v_data = (caddr_t)zp; 141 VN_LOCK_AREC(vp); 142 VN_LOCK_ASHARE(vp); 143 } else { 144 zp->z_vnode = NULL; 145 } 146 147 list_link_init(&zp->z_link_node); 148 149 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 150 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 151 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 152 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 153 154 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 155 avl_create(&zp->z_range_avl, zfs_range_compare, 156 sizeof (rl_t), offsetof(rl_t, r_node)); 157 158 zp->z_dirlocks = NULL; 159 zp->z_acl_cached = NULL; 160 zp->z_moved = 0; 161 return (0); 162} 163 164/*ARGSUSED*/ 165static void 166zfs_znode_cache_destructor(void *buf, void *arg) 167{ 168 znode_t *zp = buf; 169 170 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 171 ASSERT(ZTOV(zp) == NULL); 172 vn_free(ZTOV(zp)); 173 ASSERT(!list_link_active(&zp->z_link_node)); 174 mutex_destroy(&zp->z_lock); 175 rw_destroy(&zp->z_parent_lock); 176 rw_destroy(&zp->z_name_lock); 177 mutex_destroy(&zp->z_acl_lock); 178 avl_destroy(&zp->z_range_avl); 179 mutex_destroy(&zp->z_range_lock); 180 181 ASSERT(zp->z_dirlocks == NULL); 182 ASSERT(zp->z_acl_cached == NULL); 183} 184 185#ifdef ZNODE_STATS 186static struct { 187 uint64_t zms_zfsvfs_invalid; 188 uint64_t zms_zfsvfs_recheck1; 189 uint64_t zms_zfsvfs_unmounted; 190 uint64_t zms_zfsvfs_recheck2; 191 uint64_t zms_obj_held; 192 uint64_t zms_vnode_locked; 193 uint64_t zms_not_only_dnlc; 194} znode_move_stats; 195#endif /* ZNODE_STATS */ 196 197#ifdef sun 198static void 199zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 200{ 201 vnode_t *vp; 202 203 /* Copy fields. */ 204 nzp->z_zfsvfs = ozp->z_zfsvfs; 205 206 /* Swap vnodes. */ 207 vp = nzp->z_vnode; 208 nzp->z_vnode = ozp->z_vnode; 209 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 210 ZTOV(ozp)->v_data = ozp; 211 ZTOV(nzp)->v_data = nzp; 212 213 nzp->z_id = ozp->z_id; 214 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 215 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 216 nzp->z_unlinked = ozp->z_unlinked; 217 nzp->z_atime_dirty = ozp->z_atime_dirty; 218 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 219 nzp->z_blksz = ozp->z_blksz; 220 nzp->z_seq = ozp->z_seq; 221 nzp->z_mapcnt = ozp->z_mapcnt; 222 nzp->z_gen = ozp->z_gen; 223 nzp->z_sync_cnt = ozp->z_sync_cnt; 224 nzp->z_is_sa = ozp->z_is_sa; 225 nzp->z_sa_hdl = ozp->z_sa_hdl; 226 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 227 nzp->z_links = ozp->z_links; 228 nzp->z_size = ozp->z_size; 229 nzp->z_pflags = ozp->z_pflags; 230 nzp->z_uid = ozp->z_uid; 231 nzp->z_gid = ozp->z_gid; 232 nzp->z_mode = ozp->z_mode; 233 234 /* 235 * Since this is just an idle znode and kmem is already dealing with 236 * memory pressure, release any cached ACL. 237 */ 238 if (ozp->z_acl_cached) { 239 zfs_acl_free(ozp->z_acl_cached); 240 ozp->z_acl_cached = NULL; 241 } 242 243 sa_set_userp(nzp->z_sa_hdl, nzp); 244 245 /* 246 * Invalidate the original znode by clearing fields that provide a 247 * pointer back to the znode. Set the low bit of the vfs pointer to 248 * ensure that zfs_znode_move() recognizes the znode as invalid in any 249 * subsequent callback. 250 */ 251 ozp->z_sa_hdl = NULL; 252 POINTER_INVALIDATE(&ozp->z_zfsvfs); 253 254 /* 255 * Mark the znode. 256 */ 257 nzp->z_moved = 1; 258 ozp->z_moved = (uint8_t)-1; 259} 260 261/*ARGSUSED*/ 262static kmem_cbrc_t 263zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 264{ 265 znode_t *ozp = buf, *nzp = newbuf; 266 zfsvfs_t *zfsvfs; 267 vnode_t *vp; 268 269 /* 270 * The znode is on the file system's list of known znodes if the vfs 271 * pointer is valid. We set the low bit of the vfs pointer when freeing 272 * the znode to invalidate it, and the memory patterns written by kmem 273 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 274 * created znode sets the vfs pointer last of all to indicate that the 275 * znode is known and in a valid state to be moved by this function. 276 */ 277 zfsvfs = ozp->z_zfsvfs; 278 if (!POINTER_IS_VALID(zfsvfs)) { 279 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 280 return (KMEM_CBRC_DONT_KNOW); 281 } 282 283 /* 284 * Close a small window in which it's possible that the filesystem could 285 * be unmounted and freed, and zfsvfs, though valid in the previous 286 * statement, could point to unrelated memory by the time we try to 287 * prevent the filesystem from being unmounted. 288 */ 289 rw_enter(&zfsvfs_lock, RW_WRITER); 290 if (zfsvfs != ozp->z_zfsvfs) { 291 rw_exit(&zfsvfs_lock); 292 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 293 return (KMEM_CBRC_DONT_KNOW); 294 } 295 296 /* 297 * If the znode is still valid, then so is the file system. We know that 298 * no valid file system can be freed while we hold zfsvfs_lock, so we 299 * can safely ensure that the filesystem is not and will not be 300 * unmounted. The next statement is equivalent to ZFS_ENTER(). 301 */ 302 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 303 if (zfsvfs->z_unmounted) { 304 ZFS_EXIT(zfsvfs); 305 rw_exit(&zfsvfs_lock); 306 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 307 return (KMEM_CBRC_DONT_KNOW); 308 } 309 rw_exit(&zfsvfs_lock); 310 311 mutex_enter(&zfsvfs->z_znodes_lock); 312 /* 313 * Recheck the vfs pointer in case the znode was removed just before 314 * acquiring the lock. 315 */ 316 if (zfsvfs != ozp->z_zfsvfs) { 317 mutex_exit(&zfsvfs->z_znodes_lock); 318 ZFS_EXIT(zfsvfs); 319 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 320 return (KMEM_CBRC_DONT_KNOW); 321 } 322 323 /* 324 * At this point we know that as long as we hold z_znodes_lock, the 325 * znode cannot be freed and fields within the znode can be safely 326 * accessed. Now, prevent a race with zfs_zget(). 327 */ 328 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 329 mutex_exit(&zfsvfs->z_znodes_lock); 330 ZFS_EXIT(zfsvfs); 331 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 332 return (KMEM_CBRC_LATER); 333 } 334 335 vp = ZTOV(ozp); 336 if (mutex_tryenter(&vp->v_lock) == 0) { 337 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 338 mutex_exit(&zfsvfs->z_znodes_lock); 339 ZFS_EXIT(zfsvfs); 340 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 341 return (KMEM_CBRC_LATER); 342 } 343 344 /* Only move znodes that are referenced _only_ by the DNLC. */ 345 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 346 mutex_exit(&vp->v_lock); 347 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 348 mutex_exit(&zfsvfs->z_znodes_lock); 349 ZFS_EXIT(zfsvfs); 350 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 351 return (KMEM_CBRC_LATER); 352 } 353 354 /* 355 * The znode is known and in a valid state to move. We're holding the 356 * locks needed to execute the critical section. 357 */ 358 zfs_znode_move_impl(ozp, nzp); 359 mutex_exit(&vp->v_lock); 360 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 361 362 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 363 mutex_exit(&zfsvfs->z_znodes_lock); 364 ZFS_EXIT(zfsvfs); 365 366 return (KMEM_CBRC_YES); 367} 368#endif /* sun */ 369 370void 371zfs_znode_init(void) 372{ 373 /* 374 * Initialize zcache 375 */ 376 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 377 ASSERT(znode_cache == NULL); 378 znode_cache = kmem_cache_create("zfs_znode_cache", 379 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL, 380 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 381 kmem_cache_set_move(znode_cache, zfs_znode_move); 382} 383 384void 385zfs_znode_fini(void) 386{ 387#ifdef sun 388 /* 389 * Cleanup vfs & vnode ops 390 */ 391 zfs_remove_op_tables(); 392#endif /* sun */ 393 394 /* 395 * Cleanup zcache 396 */ 397 if (znode_cache) 398 kmem_cache_destroy(znode_cache); 399 znode_cache = NULL; 400 rw_destroy(&zfsvfs_lock); 401} 402 403#ifdef sun 404struct vnodeops *zfs_dvnodeops; 405struct vnodeops *zfs_fvnodeops; 406struct vnodeops *zfs_symvnodeops; 407struct vnodeops *zfs_xdvnodeops; 408struct vnodeops *zfs_evnodeops; 409struct vnodeops *zfs_sharevnodeops; 410 411void 412zfs_remove_op_tables() 413{ 414 /* 415 * Remove vfs ops 416 */ 417 ASSERT(zfsfstype); 418 (void) vfs_freevfsops_by_type(zfsfstype); 419 zfsfstype = 0; 420 421 /* 422 * Remove vnode ops 423 */ 424 if (zfs_dvnodeops) 425 vn_freevnodeops(zfs_dvnodeops); 426 if (zfs_fvnodeops) 427 vn_freevnodeops(zfs_fvnodeops); 428 if (zfs_symvnodeops) 429 vn_freevnodeops(zfs_symvnodeops); 430 if (zfs_xdvnodeops) 431 vn_freevnodeops(zfs_xdvnodeops); 432 if (zfs_evnodeops) 433 vn_freevnodeops(zfs_evnodeops); 434 if (zfs_sharevnodeops) 435 vn_freevnodeops(zfs_sharevnodeops); 436 437 zfs_dvnodeops = NULL; 438 zfs_fvnodeops = NULL; 439 zfs_symvnodeops = NULL; 440 zfs_xdvnodeops = NULL; 441 zfs_evnodeops = NULL; 442 zfs_sharevnodeops = NULL; 443} 444 445extern const fs_operation_def_t zfs_dvnodeops_template[]; 446extern const fs_operation_def_t zfs_fvnodeops_template[]; 447extern const fs_operation_def_t zfs_xdvnodeops_template[]; 448extern const fs_operation_def_t zfs_symvnodeops_template[]; 449extern const fs_operation_def_t zfs_evnodeops_template[]; 450extern const fs_operation_def_t zfs_sharevnodeops_template[]; 451 452int 453zfs_create_op_tables() 454{ 455 int error; 456 457 /* 458 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 459 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 460 * In this case we just return as the ops vectors are already set up. 461 */ 462 if (zfs_dvnodeops) 463 return (0); 464 465 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 466 &zfs_dvnodeops); 467 if (error) 468 return (error); 469 470 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 471 &zfs_fvnodeops); 472 if (error) 473 return (error); 474 475 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 476 &zfs_symvnodeops); 477 if (error) 478 return (error); 479 480 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 481 &zfs_xdvnodeops); 482 if (error) 483 return (error); 484 485 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 486 &zfs_evnodeops); 487 if (error) 488 return (error); 489 490 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 491 &zfs_sharevnodeops); 492 493 return (error); 494} 495#endif /* sun */ 496 497int 498zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 499{ 500 zfs_acl_ids_t acl_ids; 501 vattr_t vattr; 502 znode_t *sharezp; 503 vnode_t *vp, vnode; 504 znode_t *zp; 505 int error; 506 507 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 508 vattr.va_type = VDIR; 509 vattr.va_mode = S_IFDIR|0555; 510 vattr.va_uid = crgetuid(kcred); 511 vattr.va_gid = crgetgid(kcred); 512 513 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 514 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0); 515 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 516 sharezp->z_moved = 0; 517 sharezp->z_unlinked = 0; 518 sharezp->z_atime_dirty = 0; 519 sharezp->z_zfsvfs = zfsvfs; 520 sharezp->z_is_sa = zfsvfs->z_use_sa; 521 522 sharezp->z_vnode = &vnode; 523 vnode.v_data = sharezp; 524 525 vp = ZTOV(sharezp); 526 vp->v_type = VDIR; 527 528 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 529 kcred, NULL, &acl_ids)); 530 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 531 ASSERT3P(zp, ==, sharezp); 532 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 533 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 534 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 535 zfsvfs->z_shares_dir = sharezp->z_id; 536 537 zfs_acl_ids_free(&acl_ids); 538 ZTOV(sharezp)->v_data = NULL; 539 ZTOV(sharezp)->v_count = 0; 540 ZTOV(sharezp)->v_holdcnt = 0; 541 zp->z_vnode = NULL; 542 sa_handle_destroy(sharezp->z_sa_hdl); 543 sharezp->z_vnode = NULL; 544 kmem_cache_free(znode_cache, sharezp); 545 546 return (error); 547} 548 549/* 550 * define a couple of values we need available 551 * for both 64 and 32 bit environments. 552 */ 553#ifndef NBITSMINOR64 554#define NBITSMINOR64 32 555#endif 556#ifndef MAXMAJ64 557#define MAXMAJ64 0xffffffffUL 558#endif 559#ifndef MAXMIN64 560#define MAXMIN64 0xffffffffUL 561#endif 562 563/* 564 * Create special expldev for ZFS private use. 565 * Can't use standard expldev since it doesn't do 566 * what we want. The standard expldev() takes a 567 * dev32_t in LP64 and expands it to a long dev_t. 568 * We need an interface that takes a dev32_t in ILP32 569 * and expands it to a long dev_t. 570 */ 571static uint64_t 572zfs_expldev(dev_t dev) 573{ 574 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 575} 576/* 577 * Special cmpldev for ZFS private use. 578 * Can't use standard cmpldev since it takes 579 * a long dev_t and compresses it to dev32_t in 580 * LP64. We need to do a compaction of a long dev_t 581 * to a dev32_t in ILP32. 582 */ 583dev_t 584zfs_cmpldev(uint64_t dev) 585{ 586 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 587} 588 589static void 590zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 591 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 592{ 593 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 594 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 595 596 mutex_enter(&zp->z_lock); 597 598 ASSERT(zp->z_sa_hdl == NULL); 599 ASSERT(zp->z_acl_cached == NULL); 600 if (sa_hdl == NULL) { 601 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 602 SA_HDL_SHARED, &zp->z_sa_hdl)); 603 } else { 604 zp->z_sa_hdl = sa_hdl; 605 sa_set_userp(sa_hdl, zp); 606 } 607 608 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 609 610 /* 611 * Slap on VROOT if we are the root znode 612 */ 613 if (zp->z_id == zfsvfs->z_root) 614 ZTOV(zp)->v_flag |= VROOT; 615 616 mutex_exit(&zp->z_lock); 617 vn_exists(ZTOV(zp)); 618} 619 620void 621zfs_znode_dmu_fini(znode_t *zp) 622{ 623 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 624 zp->z_unlinked || 625 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 626 627 sa_handle_destroy(zp->z_sa_hdl); 628 zp->z_sa_hdl = NULL; 629} 630 631static void 632zfs_vnode_forget(vnode_t *vp) 633{ 634 635 VOP_UNLOCK(vp, 0); 636 VI_LOCK(vp); 637 vp->v_usecount--; 638 vp->v_iflag |= VI_DOOMED; 639 vp->v_data = NULL; 640 vdropl(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 ASSERT3U(err, ==, 0); 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 ASSERT3U(err, ==, 0); 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 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 859 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 860 861 /* 862 * If this is the root, fix up the half-initialized parent pointer 863 * to reference the just-allocated physical data area. 864 */ 865 if (flag & IS_ROOT_NODE) { 866 dzp->z_id = obj; 867 } else { 868 dzp_pflags = dzp->z_pflags; 869 } 870 871 /* 872 * If parent is an xattr, so am I. 873 */ 874 if (dzp_pflags & ZFS_XATTR) { 875 flag |= IS_XATTR; 876 } 877 878 if (zfsvfs->z_use_fuids) 879 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 880 else 881 pflags = 0; 882 883 if (vap->va_type == VDIR) { 884 size = 2; /* contents ("." and "..") */ 885 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 886 } else { 887 size = links = 0; 888 } 889 890 if (vap->va_type == VBLK || vap->va_type == VCHR) { 891 rdev = zfs_expldev(vap->va_rdev); 892 } 893 894 parent = dzp->z_id; 895 mode = acl_ids->z_mode; 896 if (flag & IS_XATTR) 897 pflags |= ZFS_XATTR; 898 899 /* 900 * No execs denied will be deterimed when zfs_mode_compute() is called. 901 */ 902 pflags |= acl_ids->z_aclp->z_hints & 903 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 904 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 905 906 ZFS_TIME_ENCODE(&now, crtime); 907 ZFS_TIME_ENCODE(&now, ctime); 908 909 if (vap->va_mask & AT_ATIME) { 910 ZFS_TIME_ENCODE(&vap->va_atime, atime); 911 } else { 912 ZFS_TIME_ENCODE(&now, atime); 913 } 914 915 if (vap->va_mask & AT_MTIME) { 916 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 917 } else { 918 ZFS_TIME_ENCODE(&now, mtime); 919 } 920 921 /* Now add in all of the "SA" attributes */ 922 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 923 &sa_hdl)); 924 925 /* 926 * Setup the array of attributes to be replaced/set on the new file 927 * 928 * order for DMU_OT_ZNODE is critical since it needs to be constructed 929 * in the old znode_phys_t format. Don't change this ordering 930 */ 931 932 if (obj_type == DMU_OT_ZNODE) { 933 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 934 NULL, &atime, 16); 935 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 936 NULL, &mtime, 16); 937 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 938 NULL, &ctime, 16); 939 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 940 NULL, &crtime, 16); 941 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 942 NULL, &gen, 8); 943 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 944 NULL, &mode, 8); 945 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 946 NULL, &size, 8); 947 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 948 NULL, &parent, 8); 949 } else { 950 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 951 NULL, &mode, 8); 952 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 953 NULL, &size, 8); 954 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 955 NULL, &gen, 8); 956 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 957 &acl_ids->z_fuid, 8); 958 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 959 &acl_ids->z_fgid, 8); 960 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 961 NULL, &parent, 8); 962 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 963 NULL, &pflags, 8); 964 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 965 NULL, &atime, 16); 966 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 967 NULL, &mtime, 16); 968 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 969 NULL, &ctime, 16); 970 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 971 NULL, &crtime, 16); 972 } 973 974 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 975 976 if (obj_type == DMU_OT_ZNODE) { 977 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 978 &empty_xattr, 8); 979 } 980 if (obj_type == DMU_OT_ZNODE || 981 (vap->va_type == VBLK || vap->va_type == VCHR)) { 982 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 983 NULL, &rdev, 8); 984 985 } 986 if (obj_type == DMU_OT_ZNODE) { 987 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 988 NULL, &pflags, 8); 989 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 990 &acl_ids->z_fuid, 8); 991 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 992 &acl_ids->z_fgid, 8); 993 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 994 sizeof (uint64_t) * 4); 995 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 996 &acl_phys, sizeof (zfs_acl_phys_t)); 997 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 998 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 999 &acl_ids->z_aclp->z_acl_count, 8); 1000 locate.cb_aclp = acl_ids->z_aclp; 1001 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 1002 zfs_acl_data_locator, &locate, 1003 acl_ids->z_aclp->z_acl_bytes); 1004 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 1005 acl_ids->z_fuid, acl_ids->z_fgid); 1006 } 1007 1008 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 1009 1010 if (!(flag & IS_ROOT_NODE)) { 1011 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 1012 ASSERT(*zpp != NULL); 1013 } else { 1014 /* 1015 * If we are creating the root node, the "parent" we 1016 * passed in is the znode for the root. 1017 */ 1018 *zpp = dzp; 1019 1020 (*zpp)->z_sa_hdl = sa_hdl; 1021 } 1022 1023 (*zpp)->z_pflags = pflags; 1024 (*zpp)->z_mode = mode; 1025 1026 if (vap->va_mask & AT_XVATTR) 1027 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 1028 1029 if (obj_type == DMU_OT_ZNODE || 1030 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1031 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx); 1032 ASSERT3P(err, ==, 0); 1033 } 1034 if (!(flag & IS_ROOT_NODE)) { 1035 vnode_t *vp; 1036 1037 vp = ZTOV(*zpp); 1038 vp->v_vflag |= VV_FORCEINSMQ; 1039 err = insmntque(vp, zfsvfs->z_vfs); 1040 vp->v_vflag &= ~VV_FORCEINSMQ; 1041 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1042 } 1043 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1044} 1045 1046/* 1047 * zfs_xvattr_set only updates the in-core attributes 1048 * it is assumed the caller will be doing an sa_bulk_update 1049 * to push the changes out 1050 */ 1051void 1052zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1053{ 1054 xoptattr_t *xoap; 1055 1056 xoap = xva_getxoptattr(xvap); 1057 ASSERT(xoap); 1058 1059 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1060 uint64_t times[2]; 1061 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1062 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1063 ×, sizeof (times), tx); 1064 XVA_SET_RTN(xvap, XAT_CREATETIME); 1065 } 1066 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1067 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1068 zp->z_pflags, tx); 1069 XVA_SET_RTN(xvap, XAT_READONLY); 1070 } 1071 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1072 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1073 zp->z_pflags, tx); 1074 XVA_SET_RTN(xvap, XAT_HIDDEN); 1075 } 1076 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1077 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1078 zp->z_pflags, tx); 1079 XVA_SET_RTN(xvap, XAT_SYSTEM); 1080 } 1081 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1082 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1083 zp->z_pflags, tx); 1084 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1085 } 1086 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1087 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1088 zp->z_pflags, tx); 1089 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1090 } 1091 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1092 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1093 zp->z_pflags, tx); 1094 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1095 } 1096 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1097 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1098 zp->z_pflags, tx); 1099 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1100 } 1101 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1102 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1103 zp->z_pflags, tx); 1104 XVA_SET_RTN(xvap, XAT_NODUMP); 1105 } 1106 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1107 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1108 zp->z_pflags, tx); 1109 XVA_SET_RTN(xvap, XAT_OPAQUE); 1110 } 1111 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1112 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1113 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1114 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1115 } 1116 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1117 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1118 zp->z_pflags, tx); 1119 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1120 } 1121 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1122 zfs_sa_set_scanstamp(zp, xvap, tx); 1123 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1124 } 1125 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1126 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1127 zp->z_pflags, tx); 1128 XVA_SET_RTN(xvap, XAT_REPARSE); 1129 } 1130 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1131 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1132 zp->z_pflags, tx); 1133 XVA_SET_RTN(xvap, XAT_OFFLINE); 1134 } 1135 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1136 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1137 zp->z_pflags, tx); 1138 XVA_SET_RTN(xvap, XAT_SPARSE); 1139 } 1140} 1141 1142int 1143zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1144{ 1145 dmu_object_info_t doi; 1146 dmu_buf_t *db; 1147 znode_t *zp; 1148 int err; 1149 sa_handle_t *hdl; 1150 int first = 1; 1151 1152 *zpp = NULL; 1153 1154again: 1155 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1156 1157 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1158 if (err) { 1159 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1160 return (err); 1161 } 1162 1163 dmu_object_info_from_db(db, &doi); 1164 if (doi.doi_bonus_type != DMU_OT_SA && 1165 (doi.doi_bonus_type != DMU_OT_ZNODE || 1166 (doi.doi_bonus_type == DMU_OT_ZNODE && 1167 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1168 sa_buf_rele(db, NULL); 1169 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1170 return (EINVAL); 1171 } 1172 1173 hdl = dmu_buf_get_user(db); 1174 if (hdl != NULL) { 1175 zp = sa_get_userdata(hdl); 1176 1177 1178 /* 1179 * Since "SA" does immediate eviction we 1180 * should never find a sa handle that doesn't 1181 * know about the znode. 1182 */ 1183 1184 ASSERT3P(zp, !=, NULL); 1185 1186 mutex_enter(&zp->z_lock); 1187 ASSERT3U(zp->z_id, ==, obj_num); 1188 if (zp->z_unlinked) { 1189 err = ENOENT; 1190 } else { 1191 vnode_t *vp; 1192 int dying = 0; 1193 1194 vp = ZTOV(zp); 1195 if (vp == NULL) 1196 dying = 1; 1197 else { 1198 VN_HOLD(vp); 1199 if ((vp->v_iflag & VI_DOOMED) != 0) { 1200 dying = 1; 1201 /* 1202 * Don't VN_RELE() vnode here, because 1203 * it can call vn_lock() which creates 1204 * LOR between vnode lock and znode 1205 * lock. We will VN_RELE() the vnode 1206 * after droping znode lock. 1207 */ 1208 } 1209 } 1210 if (dying) { 1211 if (first) { 1212 ZFS_LOG(1, "dying znode detected (zp=%p)", zp); 1213 first = 0; 1214 } 1215 /* 1216 * znode is dying so we can't reuse it, we must 1217 * wait until destruction is completed. 1218 */ 1219 sa_buf_rele(db, NULL); 1220 mutex_exit(&zp->z_lock); 1221 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1222 if (vp != NULL) 1223 VN_RELE(vp); 1224 tsleep(zp, 0, "zcollide", 1); 1225 goto again; 1226 } 1227 *zpp = zp; 1228 err = 0; 1229 } 1230 sa_buf_rele(db, NULL); 1231 mutex_exit(&zp->z_lock); 1232 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 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 = 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 return (err); 1269} 1270 1271int 1272zfs_rezget(znode_t *zp) 1273{ 1274 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1275 dmu_object_info_t doi; 1276 dmu_buf_t *db; 1277 vnode_t *vp; 1278 uint64_t obj_num = zp->z_id; 1279 uint64_t mode, size; 1280 sa_bulk_attr_t bulk[8]; 1281 int err; 1282 int count = 0; 1283 uint64_t gen; 1284 1285 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1286 1287 mutex_enter(&zp->z_acl_lock); 1288 if (zp->z_acl_cached) { 1289 zfs_acl_free(zp->z_acl_cached); 1290 zp->z_acl_cached = NULL; 1291 } 1292 1293 mutex_exit(&zp->z_acl_lock); 1294 ASSERT(zp->z_sa_hdl == NULL); 1295 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1296 if (err) { 1297 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1298 return (err); 1299 } 1300 1301 dmu_object_info_from_db(db, &doi); 1302 if (doi.doi_bonus_type != DMU_OT_SA && 1303 (doi.doi_bonus_type != DMU_OT_ZNODE || 1304 (doi.doi_bonus_type == DMU_OT_ZNODE && 1305 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1306 sa_buf_rele(db, NULL); 1307 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1308 return (EINVAL); 1309 } 1310 1311 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1312 size = zp->z_size; 1313 1314 /* reload cached values */ 1315 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1316 &gen, sizeof (gen)); 1317 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1318 &zp->z_size, sizeof (zp->z_size)); 1319 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1320 &zp->z_links, sizeof (zp->z_links)); 1321 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1322 &zp->z_pflags, sizeof (zp->z_pflags)); 1323 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1324 &zp->z_atime, sizeof (zp->z_atime)); 1325 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1326 &zp->z_uid, sizeof (zp->z_uid)); 1327 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1328 &zp->z_gid, sizeof (zp->z_gid)); 1329 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1330 &mode, sizeof (mode)); 1331 1332 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1333 zfs_znode_dmu_fini(zp); 1334 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1335 return (EIO); 1336 } 1337 1338 zp->z_mode = mode; 1339 1340 if (gen != zp->z_gen) { 1341 zfs_znode_dmu_fini(zp); 1342 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1343 return (EIO); 1344 } 1345 1346 /* 1347 * XXXPJD: Not sure how is that possible, but under heavy 1348 * zfs recv -F load it happens that z_gen is the same, but 1349 * vnode type is different than znode type. This would mean 1350 * that for example regular file was replaced with directory 1351 * which has the same object number. 1352 */ 1353 vp = ZTOV(zp); 1354 if (vp != NULL && 1355 vp->v_type != IFTOVT((mode_t)zp->z_mode)) { 1356 zfs_znode_dmu_fini(zp); 1357 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1358 return (EIO); 1359 } 1360 1361 zp->z_unlinked = (zp->z_links == 0); 1362 zp->z_blksz = doi.doi_data_block_size; 1363 if (vp != NULL) { 1364 vn_pages_remove(vp, 0, 0); 1365 if (zp->z_size != size) 1366 vnode_pager_setsize(vp, zp->z_size); 1367 } 1368 1369 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1370 1371 return (0); 1372} 1373 1374void 1375zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1376{ 1377 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1378 objset_t *os = zfsvfs->z_os; 1379 uint64_t obj = zp->z_id; 1380 uint64_t acl_obj = zfs_external_acl(zp); 1381 1382 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1383 if (acl_obj) { 1384 VERIFY(!zp->z_is_sa); 1385 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1386 } 1387 VERIFY(0 == dmu_object_free(os, obj, tx)); 1388 zfs_znode_dmu_fini(zp); 1389 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1390 zfs_znode_free(zp); 1391} 1392 1393void 1394zfs_zinactive(znode_t *zp) 1395{ 1396 vnode_t *vp = ZTOV(zp); 1397 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1398 uint64_t z_id = zp->z_id; 1399 int vfslocked; 1400 1401 ASSERT(zp->z_sa_hdl); 1402 1403 /* 1404 * Don't allow a zfs_zget() while were trying to release this znode 1405 */ 1406 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1407 1408 mutex_enter(&zp->z_lock); 1409 VI_LOCK(vp); 1410 if (vp->v_count > 0) { 1411 /* 1412 * If the hold count is greater than zero, somebody has 1413 * obtained a new reference on this znode while we were 1414 * processing it here, so we are done. 1415 */ 1416 VI_UNLOCK(vp); 1417 mutex_exit(&zp->z_lock); 1418 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1419 return; 1420 } 1421 VI_UNLOCK(vp); 1422 1423 /* 1424 * If this was the last reference to a file with no links, 1425 * remove the file from the file system. 1426 */ 1427 if (zp->z_unlinked) { 1428 mutex_exit(&zp->z_lock); 1429 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1430 ASSERT(vp->v_count == 0); 1431 vrecycle(vp, curthread); 1432 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs); 1433 zfs_rmnode(zp); 1434 VFS_UNLOCK_GIANT(vfslocked); 1435 return; 1436 } 1437 1438 mutex_exit(&zp->z_lock); 1439 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1440} 1441 1442void 1443zfs_znode_free(znode_t *zp) 1444{ 1445 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1446 1447 ASSERT(ZTOV(zp) == NULL); 1448 ASSERT(zp->z_sa_hdl == NULL); 1449 mutex_enter(&zfsvfs->z_znodes_lock); 1450 POINTER_INVALIDATE(&zp->z_zfsvfs); 1451 list_remove(&zfsvfs->z_all_znodes, zp); 1452 mutex_exit(&zfsvfs->z_znodes_lock); 1453 1454 if (zp->z_acl_cached) { 1455 zfs_acl_free(zp->z_acl_cached); 1456 zp->z_acl_cached = NULL; 1457 } 1458 1459 kmem_cache_free(znode_cache, zp); 1460 1461 VFS_RELE(zfsvfs->z_vfs); 1462} 1463 1464void 1465zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1466 uint64_t ctime[2], boolean_t have_tx) 1467{ 1468 timestruc_t now; 1469 1470 gethrestime(&now); 1471 1472 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1473 zp->z_atime_dirty = 0; 1474 zp->z_seq++; 1475 } else { 1476 zp->z_atime_dirty = 1; 1477 } 1478 1479 if (flag & AT_ATIME) { 1480 ZFS_TIME_ENCODE(&now, zp->z_atime); 1481 } 1482 1483 if (flag & AT_MTIME) { 1484 ZFS_TIME_ENCODE(&now, mtime); 1485 if (zp->z_zfsvfs->z_use_fuids) { 1486 zp->z_pflags |= (ZFS_ARCHIVE | 1487 ZFS_AV_MODIFIED); 1488 } 1489 } 1490 1491 if (flag & AT_CTIME) { 1492 ZFS_TIME_ENCODE(&now, ctime); 1493 if (zp->z_zfsvfs->z_use_fuids) 1494 zp->z_pflags |= ZFS_ARCHIVE; 1495 } 1496} 1497 1498/* 1499 * Grow the block size for a file. 1500 * 1501 * IN: zp - znode of file to free data in. 1502 * size - requested block size 1503 * tx - open transaction. 1504 * 1505 * NOTE: this function assumes that the znode is write locked. 1506 */ 1507void 1508zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1509{ 1510 int error; 1511 u_longlong_t dummy; 1512 1513 if (size <= zp->z_blksz) 1514 return; 1515 /* 1516 * If the file size is already greater than the current blocksize, 1517 * we will not grow. If there is more than one block in a file, 1518 * the blocksize cannot change. 1519 */ 1520 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1521 return; 1522 1523 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1524 size, 0, tx); 1525 1526 if (error == ENOTSUP) 1527 return; 1528 ASSERT3U(error, ==, 0); 1529 1530 /* What blocksize did we actually get? */ 1531 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1532} 1533 1534#ifdef sun 1535/* 1536 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1537 * be calling back into the fs for a putpage(). E.g.: when truncating 1538 * a file, the pages being "thrown away* don't need to be written out. 1539 */ 1540/* ARGSUSED */ 1541static int 1542zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1543 int flags, cred_t *cr) 1544{ 1545 ASSERT(0); 1546 return (0); 1547} 1548#endif /* sun */ 1549 1550/* 1551 * Increase the file length 1552 * 1553 * IN: zp - znode of file to free data in. 1554 * end - new end-of-file 1555 * 1556 * RETURN: 0 if success 1557 * error code if failure 1558 */ 1559static int 1560zfs_extend(znode_t *zp, uint64_t end) 1561{ 1562 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1563 dmu_tx_t *tx; 1564 rl_t *rl; 1565 uint64_t newblksz; 1566 int error; 1567 1568 /* 1569 * We will change zp_size, lock the whole file. 1570 */ 1571 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1572 1573 /* 1574 * Nothing to do if file already at desired length. 1575 */ 1576 if (end <= zp->z_size) { 1577 zfs_range_unlock(rl); 1578 return (0); 1579 } 1580top: 1581 tx = dmu_tx_create(zfsvfs->z_os); 1582 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1583 zfs_sa_upgrade_txholds(tx, zp); 1584 if (end > zp->z_blksz && 1585 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1586 /* 1587 * We are growing the file past the current block size. 1588 */ 1589 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1590 ASSERT(!ISP2(zp->z_blksz)); 1591 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1592 } else { 1593 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1594 } 1595 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1596 } else { 1597 newblksz = 0; 1598 } 1599 1600 error = dmu_tx_assign(tx, TXG_NOWAIT); 1601 if (error) { 1602 if (error == ERESTART) { 1603 dmu_tx_wait(tx); 1604 dmu_tx_abort(tx); 1605 goto top; 1606 } 1607 dmu_tx_abort(tx); 1608 zfs_range_unlock(rl); 1609 return (error); 1610 } 1611 1612 if (newblksz) 1613 zfs_grow_blocksize(zp, newblksz, tx); 1614 1615 zp->z_size = end; 1616 1617 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1618 &zp->z_size, sizeof (zp->z_size), tx)); 1619 1620 vnode_pager_setsize(ZTOV(zp), end); 1621 1622 zfs_range_unlock(rl); 1623 1624 dmu_tx_commit(tx); 1625 1626 return (0); 1627} 1628 1629/* 1630 * Free space in a file. 1631 * 1632 * IN: zp - znode of file to free data in. 1633 * off - start of section to free. 1634 * len - length of section to free. 1635 * 1636 * RETURN: 0 if success 1637 * error code if failure 1638 */ 1639static int 1640zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1641{ 1642 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1643 rl_t *rl; 1644 int error; 1645 1646 /* 1647 * Lock the range being freed. 1648 */ 1649 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1650 1651 /* 1652 * Nothing to do if file already at desired length. 1653 */ 1654 if (off >= zp->z_size) { 1655 zfs_range_unlock(rl); 1656 return (0); 1657 } 1658 1659 if (off + len > zp->z_size) 1660 len = zp->z_size - off; 1661 1662 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1663 1664 if (error == 0) { 1665 /* 1666 * In FreeBSD we cannot free block in the middle of a file, 1667 * but only at the end of a file, so this code path should 1668 * never happen. 1669 */ 1670 vnode_pager_setsize(ZTOV(zp), off); 1671 } 1672 1673 zfs_range_unlock(rl); 1674 1675 return (error); 1676} 1677 1678/* 1679 * Truncate a file 1680 * 1681 * IN: zp - znode of file to free data in. 1682 * end - new end-of-file. 1683 * 1684 * RETURN: 0 if success 1685 * error code if failure 1686 */ 1687static int 1688zfs_trunc(znode_t *zp, uint64_t end) 1689{ 1690 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1691 vnode_t *vp = ZTOV(zp); 1692 dmu_tx_t *tx; 1693 rl_t *rl; 1694 int error; 1695 sa_bulk_attr_t bulk[2]; 1696 int count = 0; 1697 1698 /* 1699 * We will change zp_size, lock the whole file. 1700 */ 1701 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1702 1703 /* 1704 * Nothing to do if file already at desired length. 1705 */ 1706 if (end >= zp->z_size) { 1707 zfs_range_unlock(rl); 1708 return (0); 1709 } 1710 1711 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1712 if (error) { 1713 zfs_range_unlock(rl); 1714 return (error); 1715 } 1716top: 1717 tx = dmu_tx_create(zfsvfs->z_os); 1718 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1719 zfs_sa_upgrade_txholds(tx, zp); 1720 error = dmu_tx_assign(tx, TXG_NOWAIT); 1721 if (error) { 1722 if (error == ERESTART) { 1723 dmu_tx_wait(tx); 1724 dmu_tx_abort(tx); 1725 goto top; 1726 } 1727 dmu_tx_abort(tx); 1728 zfs_range_unlock(rl); 1729 return (error); 1730 } 1731 1732 zp->z_size = end; 1733 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1734 NULL, &zp->z_size, sizeof (zp->z_size)); 1735 1736 if (end == 0) { 1737 zp->z_pflags &= ~ZFS_SPARSE; 1738 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1739 NULL, &zp->z_pflags, 8); 1740 } 1741 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1742 1743 dmu_tx_commit(tx); 1744 1745 /* 1746 * Clear any mapped pages in the truncated region. This has to 1747 * happen outside of the transaction to avoid the possibility of 1748 * a deadlock with someone trying to push a page that we are 1749 * about to invalidate. 1750 */ 1751 vnode_pager_setsize(vp, end); 1752 1753 zfs_range_unlock(rl); 1754 1755 return (0); 1756} 1757 1758/* 1759 * Free space in a file 1760 * 1761 * IN: zp - znode of file to free data in. 1762 * off - start of range 1763 * len - end of range (0 => EOF) 1764 * flag - current file open mode flags. 1765 * log - TRUE if this action should be logged 1766 * 1767 * RETURN: 0 if success 1768 * error code if failure 1769 */ 1770int 1771zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1772{ 1773 vnode_t *vp = ZTOV(zp); 1774 dmu_tx_t *tx; 1775 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1776 zilog_t *zilog = zfsvfs->z_log; 1777 uint64_t mode; 1778 uint64_t mtime[2], ctime[2]; 1779 sa_bulk_attr_t bulk[3]; 1780 int count = 0; 1781 int error; 1782 1783 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1784 sizeof (mode))) != 0) 1785 return (error); 1786 1787 if (off > zp->z_size) { 1788 error = zfs_extend(zp, off+len); 1789 if (error == 0 && log) 1790 goto log; 1791 else 1792 return (error); 1793 } 1794 1795 /* 1796 * Check for any locks in the region to be freed. 1797 */ 1798 1799 if (MANDLOCK(vp, (mode_t)mode)) { 1800 uint64_t length = (len ? len : zp->z_size - off); 1801 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1802 return (error); 1803 } 1804 1805 if (len == 0) { 1806 error = zfs_trunc(zp, off); 1807 } else { 1808 if ((error = zfs_free_range(zp, off, len)) == 0 && 1809 off + len > zp->z_size) 1810 error = zfs_extend(zp, off+len); 1811 } 1812 if (error || !log) 1813 return (error); 1814log: 1815 tx = dmu_tx_create(zfsvfs->z_os); 1816 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1817 zfs_sa_upgrade_txholds(tx, zp); 1818 error = dmu_tx_assign(tx, TXG_NOWAIT); 1819 if (error) { 1820 if (error == ERESTART) { 1821 dmu_tx_wait(tx); 1822 dmu_tx_abort(tx); 1823 goto log; 1824 } 1825 dmu_tx_abort(tx); 1826 return (error); 1827 } 1828 1829 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1830 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1831 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1832 NULL, &zp->z_pflags, 8); 1833 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1834 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1835 ASSERT(error == 0); 1836 1837 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1838 1839 dmu_tx_commit(tx); 1840 return (0); 1841} 1842 1843void 1844zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1845{ 1846 zfsvfs_t zfsvfs; 1847 uint64_t moid, obj, sa_obj, version; 1848 uint64_t sense = ZFS_CASE_SENSITIVE; 1849 uint64_t norm = 0; 1850 nvpair_t *elem; 1851 int error; 1852 int i; 1853 znode_t *rootzp = NULL; 1854 vnode_t vnode; 1855 vattr_t vattr; 1856 znode_t *zp; 1857 zfs_acl_ids_t acl_ids; 1858 1859 /* 1860 * First attempt to create master node. 1861 */ 1862 /* 1863 * In an empty objset, there are no blocks to read and thus 1864 * there can be no i/o errors (which we assert below). 1865 */ 1866 moid = MASTER_NODE_OBJ; 1867 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1868 DMU_OT_NONE, 0, tx); 1869 ASSERT(error == 0); 1870 1871 /* 1872 * Set starting attributes. 1873 */ 1874 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1875 elem = NULL; 1876 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1877 /* For the moment we expect all zpl props to be uint64_ts */ 1878 uint64_t val; 1879 char *name; 1880 1881 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1882 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1883 name = nvpair_name(elem); 1884 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1885 if (val < version) 1886 version = val; 1887 } else { 1888 error = zap_update(os, moid, name, 8, 1, &val, tx); 1889 } 1890 ASSERT(error == 0); 1891 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1892 norm = val; 1893 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1894 sense = val; 1895 } 1896 ASSERT(version != 0); 1897 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1898 1899 /* 1900 * Create zap object used for SA attribute registration 1901 */ 1902 1903 if (version >= ZPL_VERSION_SA) { 1904 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1905 DMU_OT_NONE, 0, tx); 1906 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1907 ASSERT(error == 0); 1908 } else { 1909 sa_obj = 0; 1910 } 1911 /* 1912 * Create a delete queue. 1913 */ 1914 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1915 1916 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1917 ASSERT(error == 0); 1918 1919 /* 1920 * Create root znode. Create minimal znode/vnode/zfsvfs 1921 * to allow zfs_mknode to work. 1922 */ 1923 VATTR_NULL(&vattr); 1924 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1925 vattr.va_type = VDIR; 1926 vattr.va_mode = S_IFDIR|0755; 1927 vattr.va_uid = crgetuid(cr); 1928 vattr.va_gid = crgetgid(cr); 1929 1930 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1931 1932 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1933 zfs_znode_cache_constructor(rootzp, NULL, 0); 1934 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1935 rootzp->z_moved = 0; 1936 rootzp->z_unlinked = 0; 1937 rootzp->z_atime_dirty = 0; 1938 rootzp->z_is_sa = USE_SA(version, os); 1939 1940 vnode.v_type = VDIR; 1941 vnode.v_data = rootzp; 1942 rootzp->z_vnode = &vnode; 1943 1944 zfsvfs.z_os = os; 1945 zfsvfs.z_parent = &zfsvfs; 1946 zfsvfs.z_version = version; 1947 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1948 zfsvfs.z_use_sa = USE_SA(version, os); 1949 zfsvfs.z_norm = norm; 1950 1951 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1952 &zfsvfs.z_attr_table); 1953 1954 ASSERT(error == 0); 1955 1956 /* 1957 * Fold case on file systems that are always or sometimes case 1958 * insensitive. 1959 */ 1960 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1961 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1962 1963 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1964 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1965 offsetof(znode_t, z_link_node)); 1966 1967 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1968 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1969 1970 rootzp->z_zfsvfs = &zfsvfs; 1971 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1972 cr, NULL, &acl_ids)); 1973 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1974 ASSERT3P(zp, ==, rootzp); 1975 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1976 ASSERT(error == 0); 1977 zfs_acl_ids_free(&acl_ids); 1978 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1979 1980 sa_handle_destroy(rootzp->z_sa_hdl); 1981 rootzp->z_vnode = NULL; 1982 kmem_cache_free(znode_cache, rootzp); 1983 1984 /* 1985 * Create shares directory 1986 */ 1987 1988 error = zfs_create_share_dir(&zfsvfs, tx); 1989 1990 ASSERT(error == 0); 1991 1992 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1993 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1994} 1995 1996#endif /* _KERNEL */ 1997 1998static int 1999zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 2000{ 2001 uint64_t sa_obj = 0; 2002 int error; 2003 2004 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 2005 if (error != 0 && error != ENOENT) 2006 return (error); 2007 2008 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 2009 return (error); 2010} 2011 2012static int 2013zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 2014 dmu_buf_t **db, void *tag) 2015{ 2016 dmu_object_info_t doi; 2017 int error; 2018 2019 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 2020 return (error); 2021 2022 dmu_object_info_from_db(*db, &doi); 2023 if ((doi.doi_bonus_type != DMU_OT_SA && 2024 doi.doi_bonus_type != DMU_OT_ZNODE) || 2025 doi.doi_bonus_type == DMU_OT_ZNODE && 2026 doi.doi_bonus_size < sizeof (znode_phys_t)) { 2027 sa_buf_rele(*db, tag); 2028 return (ENOTSUP); 2029 } 2030 2031 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 2032 if (error != 0) { 2033 sa_buf_rele(*db, tag); 2034 return (error); 2035 } 2036 2037 return (0); 2038} 2039 2040void 2041zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 2042{ 2043 sa_handle_destroy(hdl); 2044 sa_buf_rele(db, tag); 2045} 2046 2047/* 2048 * Given an object number, return its parent object number and whether 2049 * or not the object is an extended attribute directory. 2050 */ 2051static int 2052zfs_obj_to_pobj(sa_handle_t *hdl, sa_attr_type_t *sa_table, uint64_t *pobjp, 2053 int *is_xattrdir) 2054{ 2055 uint64_t parent; 2056 uint64_t pflags; 2057 uint64_t mode; 2058 sa_bulk_attr_t bulk[3]; 2059 int count = 0; 2060 int error; 2061 2062 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2063 &parent, sizeof (parent)); 2064 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2065 &pflags, sizeof (pflags)); 2066 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2067 &mode, sizeof (mode)); 2068 2069 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2070 return (error); 2071 2072 *pobjp = parent; 2073 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2074 2075 return (0); 2076} 2077 2078/* 2079 * Given an object number, return some zpl level statistics 2080 */ 2081static int 2082zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2083 zfs_stat_t *sb) 2084{ 2085 sa_bulk_attr_t bulk[4]; 2086 int count = 0; 2087 2088 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2089 &sb->zs_mode, sizeof (sb->zs_mode)); 2090 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2091 &sb->zs_gen, sizeof (sb->zs_gen)); 2092 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2093 &sb->zs_links, sizeof (sb->zs_links)); 2094 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2095 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2096 2097 return (sa_bulk_lookup(hdl, bulk, count)); 2098} 2099 2100static int 2101zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2102 sa_attr_type_t *sa_table, char *buf, int len) 2103{ 2104 sa_handle_t *sa_hdl; 2105 sa_handle_t *prevhdl = NULL; 2106 dmu_buf_t *prevdb = NULL; 2107 dmu_buf_t *sa_db = NULL; 2108 char *path = buf + len - 1; 2109 int error; 2110 2111 *path = '\0'; 2112 sa_hdl = hdl; 2113 2114 for (;;) { 2115 uint64_t pobj; 2116 char component[MAXNAMELEN + 2]; 2117 size_t complen; 2118 int is_xattrdir; 2119 2120 if (prevdb) 2121 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2122 2123 if ((error = zfs_obj_to_pobj(sa_hdl, sa_table, &pobj, 2124 &is_xattrdir)) != 0) 2125 break; 2126 2127 if (pobj == obj) { 2128 if (path[0] != '/') 2129 *--path = '/'; 2130 break; 2131 } 2132 2133 component[0] = '/'; 2134 if (is_xattrdir) { 2135 (void) sprintf(component + 1, "<xattrdir>"); 2136 } else { 2137 error = zap_value_search(osp, pobj, obj, 2138 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2139 if (error != 0) 2140 break; 2141 } 2142 2143 complen = strlen(component); 2144 path -= complen; 2145 ASSERT(path >= buf); 2146 bcopy(component, path, complen); 2147 obj = pobj; 2148 2149 if (sa_hdl != hdl) { 2150 prevhdl = sa_hdl; 2151 prevdb = sa_db; 2152 } 2153 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2154 if (error != 0) { 2155 sa_hdl = prevhdl; 2156 sa_db = prevdb; 2157 break; 2158 } 2159 } 2160 2161 if (sa_hdl != NULL && sa_hdl != hdl) { 2162 ASSERT(sa_db != NULL); 2163 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2164 } 2165 2166 if (error == 0) 2167 (void) memmove(buf, path, buf + len - path); 2168 2169 return (error); 2170} 2171 2172int 2173zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2174{ 2175 sa_attr_type_t *sa_table; 2176 sa_handle_t *hdl; 2177 dmu_buf_t *db; 2178 int error; 2179 2180 error = zfs_sa_setup(osp, &sa_table); 2181 if (error != 0) 2182 return (error); 2183 2184 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2185 if (error != 0) 2186 return (error); 2187 2188 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2189 2190 zfs_release_sa_handle(hdl, db, FTAG); 2191 return (error); 2192} 2193 2194int 2195zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2196 char *buf, int len) 2197{ 2198 char *path = buf + len - 1; 2199 sa_attr_type_t *sa_table; 2200 sa_handle_t *hdl; 2201 dmu_buf_t *db; 2202 int error; 2203 2204 *path = '\0'; 2205 2206 error = zfs_sa_setup(osp, &sa_table); 2207 if (error != 0) 2208 return (error); 2209 2210 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2211 if (error != 0) 2212 return (error); 2213 2214 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2215 if (error != 0) { 2216 zfs_release_sa_handle(hdl, db, FTAG); 2217 return (error); 2218 } 2219 2220 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2221 2222 zfs_release_sa_handle(hdl, db, FTAG); 2223 return (error); 2224} 2225