zfs_znode.c revision 224252
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 27#ifdef _KERNEL 28#include <sys/types.h> 29#include <sys/param.h> 30#include <sys/time.h> 31#include <sys/systm.h> 32#include <sys/sysmacros.h> 33#include <sys/resource.h> 34#include <sys/mntent.h> 35#include <sys/u8_textprep.h> 36#include <sys/dsl_dataset.h> 37#include <sys/vfs.h> 38#include <sys/vnode.h> 39#include <sys/file.h> 40#include <sys/kmem.h> 41#include <sys/errno.h> 42#include <sys/unistd.h> 43#include <sys/atomic.h> 44#include <sys/zfs_dir.h> 45#include <sys/zfs_acl.h> 46#include <sys/zfs_ioctl.h> 47#include <sys/zfs_rlock.h> 48#include <sys/zfs_fuid.h> 49#include <sys/dnode.h> 50#include <sys/fs/zfs.h> 51#include <sys/kidmap.h> 52#endif /* _KERNEL */ 53 54#include <sys/dmu.h> 55#include <sys/refcount.h> 56#include <sys/stat.h> 57#include <sys/zap.h> 58#include <sys/zfs_znode.h> 59#include <sys/sa.h> 60#include <sys/zfs_sa.h> 61#include <sys/zfs_stat.h> 62#include <sys/refcount.h> 63 64#include "zfs_prop.h" 65#include "zfs_comutil.h" 66 67/* Used by fstat(1). */ 68SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t), 69 "sizeof(znode_t)"); 70 71/* 72 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 73 * turned on when DEBUG is also defined. 74 */ 75#ifdef DEBUG 76#define ZNODE_STATS 77#endif /* DEBUG */ 78 79#ifdef ZNODE_STATS 80#define ZNODE_STAT_ADD(stat) ((stat)++) 81#else 82#define ZNODE_STAT_ADD(stat) /* nothing */ 83#endif /* ZNODE_STATS */ 84 85/* 86 * Functions needed for userland (ie: libzpool) are not put under 87 * #ifdef_KERNEL; the rest of the functions have dependencies 88 * (such as VFS logic) that will not compile easily in userland. 89 */ 90#ifdef _KERNEL 91/* 92 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 93 * be freed before it can be safely accessed. 94 */ 95krwlock_t zfsvfs_lock; 96 97static kmem_cache_t *znode_cache = NULL; 98 99/*ARGSUSED*/ 100static void 101znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 102{ 103 /* 104 * We should never drop all dbuf refs without first clearing 105 * the eviction callback. 106 */ 107 panic("evicting znode %p\n", user_ptr); 108} 109 110extern struct vop_vector zfs_vnodeops; 111extern struct vop_vector zfs_fifoops; 112extern struct vop_vector zfs_shareops; 113 114/* 115 * XXX: We cannot use this function as a cache constructor, because 116 * there is one global cache for all file systems and we need 117 * to pass vfsp here, which is not possible, because argument 118 * 'cdrarg' is defined at kmem_cache_create() time. 119 */ 120/*ARGSUSED*/ 121static int 122zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 123{ 124 znode_t *zp = buf; 125 vnode_t *vp; 126 vfs_t *vfsp = arg; 127 int error; 128 129 POINTER_INVALIDATE(&zp->z_zfsvfs); 130 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 131 132 if (vfsp != NULL) { 133 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp); 134 if (error != 0 && (kmflags & KM_NOSLEEP)) 135 return (-1); 136 ASSERT(error == 0); 137 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 138 zp->z_vnode = vp; 139 vp->v_data = (caddr_t)zp; 140 VN_LOCK_AREC(vp); 141 VN_LOCK_ASHARE(vp); 142 } else { 143 zp->z_vnode = NULL; 144 } 145 146 list_link_init(&zp->z_link_node); 147 148 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 149 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 150 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 151 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 152 153 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 154 avl_create(&zp->z_range_avl, zfs_range_compare, 155 sizeof (rl_t), offsetof(rl_t, r_node)); 156 157 zp->z_dirlocks = NULL; 158 zp->z_acl_cached = NULL; 159 zp->z_moved = 0; 160 return (0); 161} 162 163/*ARGSUSED*/ 164static void 165zfs_znode_cache_destructor(void *buf, void *arg) 166{ 167 znode_t *zp = buf; 168 169 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 170 ASSERT(ZTOV(zp) == NULL); 171 vn_free(ZTOV(zp)); 172 ASSERT(!list_link_active(&zp->z_link_node)); 173 mutex_destroy(&zp->z_lock); 174 rw_destroy(&zp->z_parent_lock); 175 rw_destroy(&zp->z_name_lock); 176 mutex_destroy(&zp->z_acl_lock); 177 avl_destroy(&zp->z_range_avl); 178 mutex_destroy(&zp->z_range_lock); 179 180 ASSERT(zp->z_dirlocks == NULL); 181 ASSERT(zp->z_acl_cached == NULL); 182} 183 184#ifdef ZNODE_STATS 185static struct { 186 uint64_t zms_zfsvfs_invalid; 187 uint64_t zms_zfsvfs_recheck1; 188 uint64_t zms_zfsvfs_unmounted; 189 uint64_t zms_zfsvfs_recheck2; 190 uint64_t zms_obj_held; 191 uint64_t zms_vnode_locked; 192 uint64_t zms_not_only_dnlc; 193} znode_move_stats; 194#endif /* ZNODE_STATS */ 195 196#ifdef sun 197static void 198zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 199{ 200 vnode_t *vp; 201 202 /* Copy fields. */ 203 nzp->z_zfsvfs = ozp->z_zfsvfs; 204 205 /* Swap vnodes. */ 206 vp = nzp->z_vnode; 207 nzp->z_vnode = ozp->z_vnode; 208 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 209 ZTOV(ozp)->v_data = ozp; 210 ZTOV(nzp)->v_data = nzp; 211 212 nzp->z_id = ozp->z_id; 213 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 214 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 215 nzp->z_unlinked = ozp->z_unlinked; 216 nzp->z_atime_dirty = ozp->z_atime_dirty; 217 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 218 nzp->z_blksz = ozp->z_blksz; 219 nzp->z_seq = ozp->z_seq; 220 nzp->z_mapcnt = ozp->z_mapcnt; 221 nzp->z_gen = ozp->z_gen; 222 nzp->z_sync_cnt = ozp->z_sync_cnt; 223 nzp->z_is_sa = ozp->z_is_sa; 224 nzp->z_sa_hdl = ozp->z_sa_hdl; 225 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 226 nzp->z_links = ozp->z_links; 227 nzp->z_size = ozp->z_size; 228 nzp->z_pflags = ozp->z_pflags; 229 nzp->z_uid = ozp->z_uid; 230 nzp->z_gid = ozp->z_gid; 231 nzp->z_mode = ozp->z_mode; 232 233 /* 234 * Since this is just an idle znode and kmem is already dealing with 235 * memory pressure, release any cached ACL. 236 */ 237 if (ozp->z_acl_cached) { 238 zfs_acl_free(ozp->z_acl_cached); 239 ozp->z_acl_cached = NULL; 240 } 241 242 sa_set_userp(nzp->z_sa_hdl, nzp); 243 244 /* 245 * Invalidate the original znode by clearing fields that provide a 246 * pointer back to the znode. Set the low bit of the vfs pointer to 247 * ensure that zfs_znode_move() recognizes the znode as invalid in any 248 * subsequent callback. 249 */ 250 ozp->z_sa_hdl = NULL; 251 POINTER_INVALIDATE(&ozp->z_zfsvfs); 252 253 /* 254 * Mark the znode. 255 */ 256 nzp->z_moved = 1; 257 ozp->z_moved = (uint8_t)-1; 258} 259 260/*ARGSUSED*/ 261static kmem_cbrc_t 262zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 263{ 264 znode_t *ozp = buf, *nzp = newbuf; 265 zfsvfs_t *zfsvfs; 266 vnode_t *vp; 267 268 /* 269 * The znode is on the file system's list of known znodes if the vfs 270 * pointer is valid. We set the low bit of the vfs pointer when freeing 271 * the znode to invalidate it, and the memory patterns written by kmem 272 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 273 * created znode sets the vfs pointer last of all to indicate that the 274 * znode is known and in a valid state to be moved by this function. 275 */ 276 zfsvfs = ozp->z_zfsvfs; 277 if (!POINTER_IS_VALID(zfsvfs)) { 278 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 279 return (KMEM_CBRC_DONT_KNOW); 280 } 281 282 /* 283 * Close a small window in which it's possible that the filesystem could 284 * be unmounted and freed, and zfsvfs, though valid in the previous 285 * statement, could point to unrelated memory by the time we try to 286 * prevent the filesystem from being unmounted. 287 */ 288 rw_enter(&zfsvfs_lock, RW_WRITER); 289 if (zfsvfs != ozp->z_zfsvfs) { 290 rw_exit(&zfsvfs_lock); 291 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 292 return (KMEM_CBRC_DONT_KNOW); 293 } 294 295 /* 296 * If the znode is still valid, then so is the file system. We know that 297 * no valid file system can be freed while we hold zfsvfs_lock, so we 298 * can safely ensure that the filesystem is not and will not be 299 * unmounted. The next statement is equivalent to ZFS_ENTER(). 300 */ 301 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 302 if (zfsvfs->z_unmounted) { 303 ZFS_EXIT(zfsvfs); 304 rw_exit(&zfsvfs_lock); 305 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 306 return (KMEM_CBRC_DONT_KNOW); 307 } 308 rw_exit(&zfsvfs_lock); 309 310 mutex_enter(&zfsvfs->z_znodes_lock); 311 /* 312 * Recheck the vfs pointer in case the znode was removed just before 313 * acquiring the lock. 314 */ 315 if (zfsvfs != ozp->z_zfsvfs) { 316 mutex_exit(&zfsvfs->z_znodes_lock); 317 ZFS_EXIT(zfsvfs); 318 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 319 return (KMEM_CBRC_DONT_KNOW); 320 } 321 322 /* 323 * At this point we know that as long as we hold z_znodes_lock, the 324 * znode cannot be freed and fields within the znode can be safely 325 * accessed. Now, prevent a race with zfs_zget(). 326 */ 327 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 328 mutex_exit(&zfsvfs->z_znodes_lock); 329 ZFS_EXIT(zfsvfs); 330 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 331 return (KMEM_CBRC_LATER); 332 } 333 334 vp = ZTOV(ozp); 335 if (mutex_tryenter(&vp->v_lock) == 0) { 336 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 337 mutex_exit(&zfsvfs->z_znodes_lock); 338 ZFS_EXIT(zfsvfs); 339 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 340 return (KMEM_CBRC_LATER); 341 } 342 343 /* Only move znodes that are referenced _only_ by the DNLC. */ 344 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 345 mutex_exit(&vp->v_lock); 346 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 347 mutex_exit(&zfsvfs->z_znodes_lock); 348 ZFS_EXIT(zfsvfs); 349 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 350 return (KMEM_CBRC_LATER); 351 } 352 353 /* 354 * The znode is known and in a valid state to move. We're holding the 355 * locks needed to execute the critical section. 356 */ 357 zfs_znode_move_impl(ozp, nzp); 358 mutex_exit(&vp->v_lock); 359 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 360 361 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 362 mutex_exit(&zfsvfs->z_znodes_lock); 363 ZFS_EXIT(zfsvfs); 364 365 return (KMEM_CBRC_YES); 366} 367#endif /* sun */ 368 369void 370zfs_znode_init(void) 371{ 372 /* 373 * Initialize zcache 374 */ 375 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 376 ASSERT(znode_cache == NULL); 377 znode_cache = kmem_cache_create("zfs_znode_cache", 378 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL, 379 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 380 kmem_cache_set_move(znode_cache, zfs_znode_move); 381} 382 383void 384zfs_znode_fini(void) 385{ 386#ifdef sun 387 /* 388 * Cleanup vfs & vnode ops 389 */ 390 zfs_remove_op_tables(); 391#endif /* sun */ 392 393 /* 394 * Cleanup zcache 395 */ 396 if (znode_cache) 397 kmem_cache_destroy(znode_cache); 398 znode_cache = NULL; 399 rw_destroy(&zfsvfs_lock); 400} 401 402#ifdef sun 403struct vnodeops *zfs_dvnodeops; 404struct vnodeops *zfs_fvnodeops; 405struct vnodeops *zfs_symvnodeops; 406struct vnodeops *zfs_xdvnodeops; 407struct vnodeops *zfs_evnodeops; 408struct vnodeops *zfs_sharevnodeops; 409 410void 411zfs_remove_op_tables() 412{ 413 /* 414 * Remove vfs ops 415 */ 416 ASSERT(zfsfstype); 417 (void) vfs_freevfsops_by_type(zfsfstype); 418 zfsfstype = 0; 419 420 /* 421 * Remove vnode ops 422 */ 423 if (zfs_dvnodeops) 424 vn_freevnodeops(zfs_dvnodeops); 425 if (zfs_fvnodeops) 426 vn_freevnodeops(zfs_fvnodeops); 427 if (zfs_symvnodeops) 428 vn_freevnodeops(zfs_symvnodeops); 429 if (zfs_xdvnodeops) 430 vn_freevnodeops(zfs_xdvnodeops); 431 if (zfs_evnodeops) 432 vn_freevnodeops(zfs_evnodeops); 433 if (zfs_sharevnodeops) 434 vn_freevnodeops(zfs_sharevnodeops); 435 436 zfs_dvnodeops = NULL; 437 zfs_fvnodeops = NULL; 438 zfs_symvnodeops = NULL; 439 zfs_xdvnodeops = NULL; 440 zfs_evnodeops = NULL; 441 zfs_sharevnodeops = NULL; 442} 443 444extern const fs_operation_def_t zfs_dvnodeops_template[]; 445extern const fs_operation_def_t zfs_fvnodeops_template[]; 446extern const fs_operation_def_t zfs_xdvnodeops_template[]; 447extern const fs_operation_def_t zfs_symvnodeops_template[]; 448extern const fs_operation_def_t zfs_evnodeops_template[]; 449extern const fs_operation_def_t zfs_sharevnodeops_template[]; 450 451int 452zfs_create_op_tables() 453{ 454 int error; 455 456 /* 457 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 458 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 459 * In this case we just return as the ops vectors are already set up. 460 */ 461 if (zfs_dvnodeops) 462 return (0); 463 464 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 465 &zfs_dvnodeops); 466 if (error) 467 return (error); 468 469 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 470 &zfs_fvnodeops); 471 if (error) 472 return (error); 473 474 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 475 &zfs_symvnodeops); 476 if (error) 477 return (error); 478 479 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 480 &zfs_xdvnodeops); 481 if (error) 482 return (error); 483 484 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 485 &zfs_evnodeops); 486 if (error) 487 return (error); 488 489 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 490 &zfs_sharevnodeops); 491 492 return (error); 493} 494#endif /* sun */ 495 496int 497zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 498{ 499 zfs_acl_ids_t acl_ids; 500 vattr_t vattr; 501 znode_t *sharezp; 502 vnode_t *vp, vnode; 503 znode_t *zp; 504 int error; 505 506 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 507 vattr.va_type = VDIR; 508 vattr.va_mode = S_IFDIR|0555; 509 vattr.va_uid = crgetuid(kcred); 510 vattr.va_gid = crgetgid(kcred); 511 512 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 513 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0); 514 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 515 sharezp->z_moved = 0; 516 sharezp->z_unlinked = 0; 517 sharezp->z_atime_dirty = 0; 518 sharezp->z_zfsvfs = zfsvfs; 519 sharezp->z_is_sa = zfsvfs->z_use_sa; 520 521 sharezp->z_vnode = &vnode; 522 vnode.v_data = sharezp; 523 524 vp = ZTOV(sharezp); 525 vp->v_type = VDIR; 526 527 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 528 kcred, NULL, &acl_ids)); 529 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 530 ASSERT3P(zp, ==, sharezp); 531 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 532 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 533 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 534 zfsvfs->z_shares_dir = sharezp->z_id; 535 536 zfs_acl_ids_free(&acl_ids); 537 ZTOV(sharezp)->v_data = NULL; 538 ZTOV(sharezp)->v_count = 0; 539 ZTOV(sharezp)->v_holdcnt = 0; 540 zp->z_vnode = NULL; 541 sa_handle_destroy(sharezp->z_sa_hdl); 542 sharezp->z_vnode = NULL; 543 kmem_cache_free(znode_cache, sharezp); 544 545 return (error); 546} 547 548/* 549 * define a couple of values we need available 550 * for both 64 and 32 bit environments. 551 */ 552#ifndef NBITSMINOR64 553#define NBITSMINOR64 32 554#endif 555#ifndef MAXMAJ64 556#define MAXMAJ64 0xffffffffUL 557#endif 558#ifndef MAXMIN64 559#define MAXMIN64 0xffffffffUL 560#endif 561 562/* 563 * Create special expldev for ZFS private use. 564 * Can't use standard expldev since it doesn't do 565 * what we want. The standard expldev() takes a 566 * dev32_t in LP64 and expands it to a long dev_t. 567 * We need an interface that takes a dev32_t in ILP32 568 * and expands it to a long dev_t. 569 */ 570static uint64_t 571zfs_expldev(dev_t dev) 572{ 573 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 574} 575/* 576 * Special cmpldev for ZFS private use. 577 * Can't use standard cmpldev since it takes 578 * a long dev_t and compresses it to dev32_t in 579 * LP64. We need to do a compaction of a long dev_t 580 * to a dev32_t in ILP32. 581 */ 582dev_t 583zfs_cmpldev(uint64_t dev) 584{ 585 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 586} 587 588static void 589zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 590 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 591{ 592 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 593 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 594 595 mutex_enter(&zp->z_lock); 596 597 ASSERT(zp->z_sa_hdl == NULL); 598 ASSERT(zp->z_acl_cached == NULL); 599 if (sa_hdl == NULL) { 600 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 601 SA_HDL_SHARED, &zp->z_sa_hdl)); 602 } else { 603 zp->z_sa_hdl = sa_hdl; 604 sa_set_userp(sa_hdl, zp); 605 } 606 607 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 608 609 /* 610 * Slap on VROOT if we are the root znode 611 */ 612 if (zp->z_id == zfsvfs->z_root) 613 ZTOV(zp)->v_flag |= VROOT; 614 615 mutex_exit(&zp->z_lock); 616 vn_exists(ZTOV(zp)); 617} 618 619void 620zfs_znode_dmu_fini(znode_t *zp) 621{ 622 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 623 zp->z_unlinked || 624 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 625 626 sa_handle_destroy(zp->z_sa_hdl); 627 zp->z_sa_hdl = NULL; 628} 629 630/* 631 * Construct a new znode/vnode and intialize. 632 * 633 * This does not do a call to dmu_set_user() that is 634 * up to the caller to do, in case you don't want to 635 * return the znode 636 */ 637static znode_t * 638zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 639 dmu_object_type_t obj_type, sa_handle_t *hdl) 640{ 641 znode_t *zp; 642 vnode_t *vp; 643 uint64_t mode; 644 uint64_t parent; 645 sa_bulk_attr_t bulk[9]; 646 int count = 0; 647 648 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 649 zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0); 650 651 ASSERT(zp->z_dirlocks == NULL); 652 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 653 zp->z_moved = 0; 654 655 /* 656 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 657 * the zfs_znode_move() callback. 658 */ 659 zp->z_sa_hdl = NULL; 660 zp->z_unlinked = 0; 661 zp->z_atime_dirty = 0; 662 zp->z_mapcnt = 0; 663 zp->z_id = db->db_object; 664 zp->z_blksz = blksz; 665 zp->z_seq = 0x7A4653; 666 zp->z_sync_cnt = 0; 667 668 vp = ZTOV(zp); 669 670 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 671 672 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 673 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 674 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 675 &zp->z_size, 8); 676 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 677 &zp->z_links, 8); 678 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 679 &zp->z_pflags, 8); 680 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 681 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 682 &zp->z_atime, 16); 683 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 684 &zp->z_uid, 8); 685 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 686 &zp->z_gid, 8); 687 688 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 689 if (hdl == NULL) 690 sa_handle_destroy(zp->z_sa_hdl); 691 kmem_cache_free(znode_cache, zp); 692 return (NULL); 693 } 694 695 zp->z_mode = mode; 696 697 vp->v_type = IFTOVT((mode_t)mode); 698 699 switch (vp->v_type) { 700 case VDIR: 701 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 702 break; 703#ifdef sun 704 case VBLK: 705 case VCHR: 706 { 707 uint64_t rdev; 708 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 709 &rdev, sizeof (rdev)) == 0); 710 711 vp->v_rdev = zfs_cmpldev(rdev); 712 } 713 break; 714#endif /* sun */ 715 case VFIFO: 716#ifdef sun 717 case VSOCK: 718 case VDOOR: 719#endif /* sun */ 720 vp->v_op = &zfs_fifoops; 721 break; 722 case VREG: 723 if (parent == zfsvfs->z_shares_dir) { 724 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 725 vp->v_op = &zfs_shareops; 726 } 727 break; 728#ifdef sun 729 case VLNK: 730 vn_setops(vp, zfs_symvnodeops); 731 break; 732 default: 733 vn_setops(vp, zfs_evnodeops); 734 break; 735#endif /* sun */ 736 } 737 if (vp->v_type != VFIFO) 738 VN_LOCK_ASHARE(vp); 739 740 mutex_enter(&zfsvfs->z_znodes_lock); 741 list_insert_tail(&zfsvfs->z_all_znodes, zp); 742 membar_producer(); 743 /* 744 * Everything else must be valid before assigning z_zfsvfs makes the 745 * znode eligible for zfs_znode_move(). 746 */ 747 zp->z_zfsvfs = zfsvfs; 748 mutex_exit(&zfsvfs->z_znodes_lock); 749 750 VFS_HOLD(zfsvfs->z_vfs); 751 return (zp); 752} 753 754static uint64_t empty_xattr; 755static uint64_t pad[4]; 756static zfs_acl_phys_t acl_phys; 757/* 758 * Create a new DMU object to hold a zfs znode. 759 * 760 * IN: dzp - parent directory for new znode 761 * vap - file attributes for new znode 762 * tx - dmu transaction id for zap operations 763 * cr - credentials of caller 764 * flag - flags: 765 * IS_ROOT_NODE - new object will be root 766 * IS_XATTR - new object is an attribute 767 * bonuslen - length of bonus buffer 768 * setaclp - File/Dir initial ACL 769 * fuidp - Tracks fuid allocation. 770 * 771 * OUT: zpp - allocated znode 772 * 773 */ 774void 775zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 776 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 777{ 778 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 779 uint64_t mode, size, links, parent, pflags; 780 uint64_t dzp_pflags = 0; 781 uint64_t rdev = 0; 782 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 783 dmu_buf_t *db; 784 timestruc_t now; 785 uint64_t gen, obj; 786 int err; 787 int bonuslen; 788 sa_handle_t *sa_hdl; 789 dmu_object_type_t obj_type; 790 sa_bulk_attr_t sa_attrs[ZPL_END]; 791 int cnt = 0; 792 zfs_acl_locator_cb_t locate = { 0 }; 793 794 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 795 796 if (zfsvfs->z_replay) { 797 obj = vap->va_nodeid; 798 now = vap->va_ctime; /* see zfs_replay_create() */ 799 gen = vap->va_nblocks; /* ditto */ 800 } else { 801 obj = 0; 802 gethrestime(&now); 803 gen = dmu_tx_get_txg(tx); 804 } 805 806 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 807 bonuslen = (obj_type == DMU_OT_SA) ? 808 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 809 810 /* 811 * Create a new DMU object. 812 */ 813 /* 814 * There's currently no mechanism for pre-reading the blocks that will 815 * be needed to allocate a new object, so we accept the small chance 816 * that there will be an i/o error and we will fail one of the 817 * assertions below. 818 */ 819 if (vap->va_type == VDIR) { 820 if (zfsvfs->z_replay) { 821 err = zap_create_claim_norm(zfsvfs->z_os, obj, 822 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 823 obj_type, bonuslen, tx); 824 ASSERT3U(err, ==, 0); 825 } else { 826 obj = zap_create_norm(zfsvfs->z_os, 827 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 828 obj_type, bonuslen, tx); 829 } 830 } else { 831 if (zfsvfs->z_replay) { 832 err = dmu_object_claim(zfsvfs->z_os, obj, 833 DMU_OT_PLAIN_FILE_CONTENTS, 0, 834 obj_type, bonuslen, tx); 835 ASSERT3U(err, ==, 0); 836 } else { 837 obj = dmu_object_alloc(zfsvfs->z_os, 838 DMU_OT_PLAIN_FILE_CONTENTS, 0, 839 obj_type, bonuslen, tx); 840 } 841 } 842 843 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 844 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 845 846 /* 847 * If this is the root, fix up the half-initialized parent pointer 848 * to reference the just-allocated physical data area. 849 */ 850 if (flag & IS_ROOT_NODE) { 851 dzp->z_id = obj; 852 } else { 853 dzp_pflags = dzp->z_pflags; 854 } 855 856 /* 857 * If parent is an xattr, so am I. 858 */ 859 if (dzp_pflags & ZFS_XATTR) { 860 flag |= IS_XATTR; 861 } 862 863 if (zfsvfs->z_use_fuids) 864 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 865 else 866 pflags = 0; 867 868 if (vap->va_type == VDIR) { 869 size = 2; /* contents ("." and "..") */ 870 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 871 } else { 872 size = links = 0; 873 } 874 875 if (vap->va_type == VBLK || vap->va_type == VCHR) { 876 rdev = zfs_expldev(vap->va_rdev); 877 } 878 879 parent = dzp->z_id; 880 mode = acl_ids->z_mode; 881 if (flag & IS_XATTR) 882 pflags |= ZFS_XATTR; 883 884 /* 885 * No execs denied will be deterimed when zfs_mode_compute() is called. 886 */ 887 pflags |= acl_ids->z_aclp->z_hints & 888 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 889 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 890 891 ZFS_TIME_ENCODE(&now, crtime); 892 ZFS_TIME_ENCODE(&now, ctime); 893 894 if (vap->va_mask & AT_ATIME) { 895 ZFS_TIME_ENCODE(&vap->va_atime, atime); 896 } else { 897 ZFS_TIME_ENCODE(&now, atime); 898 } 899 900 if (vap->va_mask & AT_MTIME) { 901 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 902 } else { 903 ZFS_TIME_ENCODE(&now, mtime); 904 } 905 906 /* Now add in all of the "SA" attributes */ 907 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 908 &sa_hdl)); 909 910 /* 911 * Setup the array of attributes to be replaced/set on the new file 912 * 913 * order for DMU_OT_ZNODE is critical since it needs to be constructed 914 * in the old znode_phys_t format. Don't change this ordering 915 */ 916 917 if (obj_type == DMU_OT_ZNODE) { 918 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 919 NULL, &atime, 16); 920 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 921 NULL, &mtime, 16); 922 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 923 NULL, &ctime, 16); 924 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 925 NULL, &crtime, 16); 926 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 927 NULL, &gen, 8); 928 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 929 NULL, &mode, 8); 930 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 931 NULL, &size, 8); 932 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 933 NULL, &parent, 8); 934 } else { 935 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 936 NULL, &mode, 8); 937 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 938 NULL, &size, 8); 939 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 940 NULL, &gen, 8); 941 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 942 &acl_ids->z_fuid, 8); 943 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 944 &acl_ids->z_fgid, 8); 945 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 946 NULL, &parent, 8); 947 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 948 NULL, &pflags, 8); 949 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 950 NULL, &atime, 16); 951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 952 NULL, &mtime, 16); 953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 954 NULL, &ctime, 16); 955 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 956 NULL, &crtime, 16); 957 } 958 959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 960 961 if (obj_type == DMU_OT_ZNODE) { 962 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 963 &empty_xattr, 8); 964 } 965 if (obj_type == DMU_OT_ZNODE || 966 (vap->va_type == VBLK || vap->va_type == VCHR)) { 967 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 968 NULL, &rdev, 8); 969 970 } 971 if (obj_type == DMU_OT_ZNODE) { 972 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 973 NULL, &pflags, 8); 974 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 975 &acl_ids->z_fuid, 8); 976 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 977 &acl_ids->z_fgid, 8); 978 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 979 sizeof (uint64_t) * 4); 980 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 981 &acl_phys, sizeof (zfs_acl_phys_t)); 982 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 983 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 984 &acl_ids->z_aclp->z_acl_count, 8); 985 locate.cb_aclp = acl_ids->z_aclp; 986 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 987 zfs_acl_data_locator, &locate, 988 acl_ids->z_aclp->z_acl_bytes); 989 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 990 acl_ids->z_fuid, acl_ids->z_fgid); 991 } 992 993 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 994 995 if (!(flag & IS_ROOT_NODE)) { 996 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 997 ASSERT(*zpp != NULL); 998 } else { 999 /* 1000 * If we are creating the root node, the "parent" we 1001 * passed in is the znode for the root. 1002 */ 1003 *zpp = dzp; 1004 1005 (*zpp)->z_sa_hdl = sa_hdl; 1006 } 1007 1008 (*zpp)->z_pflags = pflags; 1009 (*zpp)->z_mode = mode; 1010 1011 if (vap->va_mask & AT_XVATTR) 1012 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 1013 1014 if (obj_type == DMU_OT_ZNODE || 1015 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1016 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx); 1017 ASSERT3P(err, ==, 0); 1018 } 1019 if (!(flag & IS_ROOT_NODE)) { 1020 vnode_t *vp; 1021 1022 vp = ZTOV(*zpp); 1023 vp->v_vflag |= VV_FORCEINSMQ; 1024 err = insmntque(vp, zfsvfs->z_vfs); 1025 vp->v_vflag &= ~VV_FORCEINSMQ; 1026 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1027 } 1028 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1029} 1030 1031/* 1032 * zfs_xvattr_set only updates the in-core attributes 1033 * it is assumed the caller will be doing an sa_bulk_update 1034 * to push the changes out 1035 */ 1036void 1037zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1038{ 1039 xoptattr_t *xoap; 1040 1041 xoap = xva_getxoptattr(xvap); 1042 ASSERT(xoap); 1043 1044 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1045 uint64_t times[2]; 1046 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1047 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1048 ×, sizeof (times), tx); 1049 XVA_SET_RTN(xvap, XAT_CREATETIME); 1050 } 1051 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1052 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1053 zp->z_pflags, tx); 1054 XVA_SET_RTN(xvap, XAT_READONLY); 1055 } 1056 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1057 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1058 zp->z_pflags, tx); 1059 XVA_SET_RTN(xvap, XAT_HIDDEN); 1060 } 1061 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1062 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1063 zp->z_pflags, tx); 1064 XVA_SET_RTN(xvap, XAT_SYSTEM); 1065 } 1066 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1067 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1068 zp->z_pflags, tx); 1069 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1070 } 1071 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1072 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1073 zp->z_pflags, tx); 1074 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1075 } 1076 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1077 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1078 zp->z_pflags, tx); 1079 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1080 } 1081 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1082 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1083 zp->z_pflags, tx); 1084 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1085 } 1086 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1087 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1088 zp->z_pflags, tx); 1089 XVA_SET_RTN(xvap, XAT_NODUMP); 1090 } 1091 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1092 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1093 zp->z_pflags, tx); 1094 XVA_SET_RTN(xvap, XAT_OPAQUE); 1095 } 1096 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1097 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1098 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1099 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1100 } 1101 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1102 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1103 zp->z_pflags, tx); 1104 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1105 } 1106 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1107 zfs_sa_set_scanstamp(zp, xvap, tx); 1108 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1109 } 1110 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1111 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1112 zp->z_pflags, tx); 1113 XVA_SET_RTN(xvap, XAT_REPARSE); 1114 } 1115 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1116 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1117 zp->z_pflags, tx); 1118 XVA_SET_RTN(xvap, XAT_OFFLINE); 1119 } 1120 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1121 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1122 zp->z_pflags, tx); 1123 XVA_SET_RTN(xvap, XAT_SPARSE); 1124 } 1125} 1126 1127int 1128zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1129{ 1130 dmu_object_info_t doi; 1131 dmu_buf_t *db; 1132 znode_t *zp; 1133 int err; 1134 sa_handle_t *hdl; 1135 int first = 1; 1136 1137 *zpp = NULL; 1138 1139again: 1140 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1141 1142 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1143 if (err) { 1144 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1145 return (err); 1146 } 1147 1148 dmu_object_info_from_db(db, &doi); 1149 if (doi.doi_bonus_type != DMU_OT_SA && 1150 (doi.doi_bonus_type != DMU_OT_ZNODE || 1151 (doi.doi_bonus_type == DMU_OT_ZNODE && 1152 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1153 sa_buf_rele(db, NULL); 1154 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1155 return (EINVAL); 1156 } 1157 1158 hdl = dmu_buf_get_user(db); 1159 if (hdl != NULL) { 1160 zp = sa_get_userdata(hdl); 1161 1162 1163 /* 1164 * Since "SA" does immediate eviction we 1165 * should never find a sa handle that doesn't 1166 * know about the znode. 1167 */ 1168 1169 ASSERT3P(zp, !=, NULL); 1170 1171 mutex_enter(&zp->z_lock); 1172 ASSERT3U(zp->z_id, ==, obj_num); 1173 if (zp->z_unlinked) { 1174 err = ENOENT; 1175 } else { 1176 vnode_t *vp; 1177 int dying = 0; 1178 1179 vp = ZTOV(zp); 1180 if (vp == NULL) 1181 dying = 1; 1182 else { 1183 VN_HOLD(vp); 1184 if ((vp->v_iflag & VI_DOOMED) != 0) { 1185 dying = 1; 1186 /* 1187 * Don't VN_RELE() vnode here, because 1188 * it can call vn_lock() which creates 1189 * LOR between vnode lock and znode 1190 * lock. We will VN_RELE() the vnode 1191 * after droping znode lock. 1192 */ 1193 } 1194 } 1195 if (dying) { 1196 if (first) { 1197 ZFS_LOG(1, "dying znode detected (zp=%p)", zp); 1198 first = 0; 1199 } 1200 /* 1201 * znode is dying so we can't reuse it, we must 1202 * wait until destruction is completed. 1203 */ 1204 sa_buf_rele(db, NULL); 1205 mutex_exit(&zp->z_lock); 1206 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1207 if (vp != NULL) 1208 VN_RELE(vp); 1209 tsleep(zp, 0, "zcollide", 1); 1210 goto again; 1211 } 1212 *zpp = zp; 1213 err = 0; 1214 } 1215 sa_buf_rele(db, NULL); 1216 mutex_exit(&zp->z_lock); 1217 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1218 return (err); 1219 } 1220 1221 /* 1222 * Not found create new znode/vnode 1223 * but only if file exists. 1224 * 1225 * There is a small window where zfs_vget() could 1226 * find this object while a file create is still in 1227 * progress. This is checked for in zfs_znode_alloc() 1228 * 1229 * if zfs_znode_alloc() fails it will drop the hold on the 1230 * bonus buffer. 1231 */ 1232 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1233 doi.doi_bonus_type, NULL); 1234 if (zp == NULL) { 1235 err = ENOENT; 1236 } else { 1237 *zpp = zp; 1238 } 1239 if (err == 0) { 1240 vnode_t *vp = ZTOV(zp); 1241 1242 err = insmntque(vp, zfsvfs->z_vfs); 1243 if (err == 0) 1244 VOP_UNLOCK(vp, 0); 1245 else { 1246 zp->z_vnode = NULL; 1247 zfs_znode_dmu_fini(zp); 1248 zfs_znode_free(zp); 1249 *zpp = NULL; 1250 } 1251 } 1252 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1253 return (err); 1254} 1255 1256int 1257zfs_rezget(znode_t *zp) 1258{ 1259 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1260 dmu_object_info_t doi; 1261 dmu_buf_t *db; 1262 uint64_t obj_num = zp->z_id; 1263 uint64_t mode, size; 1264 sa_bulk_attr_t bulk[8]; 1265 int err; 1266 int count = 0; 1267 uint64_t gen; 1268 1269 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1270 1271 mutex_enter(&zp->z_acl_lock); 1272 if (zp->z_acl_cached) { 1273 zfs_acl_free(zp->z_acl_cached); 1274 zp->z_acl_cached = NULL; 1275 } 1276 1277 mutex_exit(&zp->z_acl_lock); 1278 ASSERT(zp->z_sa_hdl == NULL); 1279 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1280 if (err) { 1281 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1282 return (err); 1283 } 1284 1285 dmu_object_info_from_db(db, &doi); 1286 if (doi.doi_bonus_type != DMU_OT_SA && 1287 (doi.doi_bonus_type != DMU_OT_ZNODE || 1288 (doi.doi_bonus_type == DMU_OT_ZNODE && 1289 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1290 sa_buf_rele(db, NULL); 1291 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1292 return (EINVAL); 1293 } 1294 1295 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1296 size = zp->z_size; 1297 1298 /* reload cached values */ 1299 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1300 &gen, sizeof (gen)); 1301 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1302 &zp->z_size, sizeof (zp->z_size)); 1303 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1304 &zp->z_links, sizeof (zp->z_links)); 1305 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1306 &zp->z_pflags, sizeof (zp->z_pflags)); 1307 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1308 &zp->z_atime, sizeof (zp->z_atime)); 1309 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1310 &zp->z_uid, sizeof (zp->z_uid)); 1311 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1312 &zp->z_gid, sizeof (zp->z_gid)); 1313 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1314 &mode, sizeof (mode)); 1315 1316 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1317 zfs_znode_dmu_fini(zp); 1318 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1319 return (EIO); 1320 } 1321 1322 zp->z_mode = mode; 1323 1324 if (gen != zp->z_gen) { 1325 zfs_znode_dmu_fini(zp); 1326 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1327 return (EIO); 1328 } 1329 1330 /* 1331 * XXXPJD: Not sure how is that possible, but under heavy 1332 * zfs recv -F load it happens that z_gen is the same, but 1333 * vnode type is different than znode type. This would mean 1334 * that for example regular file was replaced with directory 1335 * which has the same object number. 1336 */ 1337 if (ZTOV(zp) != NULL && 1338 ZTOV(zp)->v_type != IFTOVT((mode_t)zp->z_mode)) { 1339 zfs_znode_dmu_fini(zp); 1340 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1341 return (EIO); 1342 } 1343 1344 zp->z_unlinked = (zp->z_links == 0); 1345 zp->z_blksz = doi.doi_data_block_size; 1346 if (zp->z_size != size && ZTOV(zp) != NULL) 1347 vnode_pager_setsize(ZTOV(zp), zp->z_size); 1348 1349 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1350 1351 return (0); 1352} 1353 1354void 1355zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1356{ 1357 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1358 objset_t *os = zfsvfs->z_os; 1359 uint64_t obj = zp->z_id; 1360 uint64_t acl_obj = zfs_external_acl(zp); 1361 1362 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1363 if (acl_obj) { 1364 VERIFY(!zp->z_is_sa); 1365 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1366 } 1367 VERIFY(0 == dmu_object_free(os, obj, tx)); 1368 zfs_znode_dmu_fini(zp); 1369 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1370 zfs_znode_free(zp); 1371} 1372 1373void 1374zfs_zinactive(znode_t *zp) 1375{ 1376 vnode_t *vp = ZTOV(zp); 1377 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1378 uint64_t z_id = zp->z_id; 1379 int vfslocked; 1380 1381 ASSERT(zp->z_sa_hdl); 1382 1383 /* 1384 * Don't allow a zfs_zget() while were trying to release this znode 1385 */ 1386 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1387 1388 mutex_enter(&zp->z_lock); 1389 VI_LOCK(vp); 1390 if (vp->v_count > 0) { 1391 /* 1392 * If the hold count is greater than zero, somebody has 1393 * obtained a new reference on this znode while we were 1394 * processing it here, so we are done. 1395 */ 1396 VI_UNLOCK(vp); 1397 mutex_exit(&zp->z_lock); 1398 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1399 return; 1400 } 1401 VI_UNLOCK(vp); 1402 1403 /* 1404 * If this was the last reference to a file with no links, 1405 * remove the file from the file system. 1406 */ 1407 if (zp->z_unlinked) { 1408 mutex_exit(&zp->z_lock); 1409 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1410 ASSERT(vp->v_count == 0); 1411 vrecycle(vp, curthread); 1412 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs); 1413 zfs_rmnode(zp); 1414 VFS_UNLOCK_GIANT(vfslocked); 1415 return; 1416 } 1417 1418 mutex_exit(&zp->z_lock); 1419 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1420} 1421 1422void 1423zfs_znode_free(znode_t *zp) 1424{ 1425 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1426 1427 ASSERT(ZTOV(zp) == NULL); 1428 ASSERT(zp->z_sa_hdl == NULL); 1429 mutex_enter(&zfsvfs->z_znodes_lock); 1430 POINTER_INVALIDATE(&zp->z_zfsvfs); 1431 list_remove(&zfsvfs->z_all_znodes, zp); 1432 mutex_exit(&zfsvfs->z_znodes_lock); 1433 1434 if (zp->z_acl_cached) { 1435 zfs_acl_free(zp->z_acl_cached); 1436 zp->z_acl_cached = NULL; 1437 } 1438 1439 kmem_cache_free(znode_cache, zp); 1440 1441 VFS_RELE(zfsvfs->z_vfs); 1442} 1443 1444void 1445zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1446 uint64_t ctime[2], boolean_t have_tx) 1447{ 1448 timestruc_t now; 1449 1450 gethrestime(&now); 1451 1452 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1453 zp->z_atime_dirty = 0; 1454 zp->z_seq++; 1455 } else { 1456 zp->z_atime_dirty = 1; 1457 } 1458 1459 if (flag & AT_ATIME) { 1460 ZFS_TIME_ENCODE(&now, zp->z_atime); 1461 } 1462 1463 if (flag & AT_MTIME) { 1464 ZFS_TIME_ENCODE(&now, mtime); 1465 if (zp->z_zfsvfs->z_use_fuids) { 1466 zp->z_pflags |= (ZFS_ARCHIVE | 1467 ZFS_AV_MODIFIED); 1468 } 1469 } 1470 1471 if (flag & AT_CTIME) { 1472 ZFS_TIME_ENCODE(&now, ctime); 1473 if (zp->z_zfsvfs->z_use_fuids) 1474 zp->z_pflags |= ZFS_ARCHIVE; 1475 } 1476} 1477 1478/* 1479 * Grow the block size for a file. 1480 * 1481 * IN: zp - znode of file to free data in. 1482 * size - requested block size 1483 * tx - open transaction. 1484 * 1485 * NOTE: this function assumes that the znode is write locked. 1486 */ 1487void 1488zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1489{ 1490 int error; 1491 u_longlong_t dummy; 1492 1493 if (size <= zp->z_blksz) 1494 return; 1495 /* 1496 * If the file size is already greater than the current blocksize, 1497 * we will not grow. If there is more than one block in a file, 1498 * the blocksize cannot change. 1499 */ 1500 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1501 return; 1502 1503 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1504 size, 0, tx); 1505 1506 if (error == ENOTSUP) 1507 return; 1508 ASSERT3U(error, ==, 0); 1509 1510 /* What blocksize did we actually get? */ 1511 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1512} 1513 1514#ifdef sun 1515/* 1516 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1517 * be calling back into the fs for a putpage(). E.g.: when truncating 1518 * a file, the pages being "thrown away* don't need to be written out. 1519 */ 1520/* ARGSUSED */ 1521static int 1522zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1523 int flags, cred_t *cr) 1524{ 1525 ASSERT(0); 1526 return (0); 1527} 1528#endif /* sun */ 1529 1530/* 1531 * Increase the file length 1532 * 1533 * IN: zp - znode of file to free data in. 1534 * end - new end-of-file 1535 * 1536 * RETURN: 0 if success 1537 * error code if failure 1538 */ 1539static int 1540zfs_extend(znode_t *zp, uint64_t end) 1541{ 1542 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1543 dmu_tx_t *tx; 1544 rl_t *rl; 1545 uint64_t newblksz; 1546 int error; 1547 1548 /* 1549 * We will change zp_size, lock the whole file. 1550 */ 1551 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1552 1553 /* 1554 * Nothing to do if file already at desired length. 1555 */ 1556 if (end <= zp->z_size) { 1557 zfs_range_unlock(rl); 1558 return (0); 1559 } 1560top: 1561 tx = dmu_tx_create(zfsvfs->z_os); 1562 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1563 zfs_sa_upgrade_txholds(tx, zp); 1564 if (end > zp->z_blksz && 1565 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1566 /* 1567 * We are growing the file past the current block size. 1568 */ 1569 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1570 ASSERT(!ISP2(zp->z_blksz)); 1571 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1572 } else { 1573 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1574 } 1575 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1576 } else { 1577 newblksz = 0; 1578 } 1579 1580 error = dmu_tx_assign(tx, TXG_NOWAIT); 1581 if (error) { 1582 if (error == ERESTART) { 1583 dmu_tx_wait(tx); 1584 dmu_tx_abort(tx); 1585 goto top; 1586 } 1587 dmu_tx_abort(tx); 1588 zfs_range_unlock(rl); 1589 return (error); 1590 } 1591 1592 if (newblksz) 1593 zfs_grow_blocksize(zp, newblksz, tx); 1594 1595 zp->z_size = end; 1596 1597 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1598 &zp->z_size, sizeof (zp->z_size), tx)); 1599 1600 vnode_pager_setsize(ZTOV(zp), end); 1601 1602 zfs_range_unlock(rl); 1603 1604 dmu_tx_commit(tx); 1605 1606 return (0); 1607} 1608 1609/* 1610 * Free space in a file. 1611 * 1612 * IN: zp - znode of file to free data in. 1613 * off - start of section to free. 1614 * len - length of section to free. 1615 * 1616 * RETURN: 0 if success 1617 * error code if failure 1618 */ 1619static int 1620zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1621{ 1622 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1623 rl_t *rl; 1624 int error; 1625 1626 /* 1627 * Lock the range being freed. 1628 */ 1629 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1630 1631 /* 1632 * Nothing to do if file already at desired length. 1633 */ 1634 if (off >= zp->z_size) { 1635 zfs_range_unlock(rl); 1636 return (0); 1637 } 1638 1639 if (off + len > zp->z_size) 1640 len = zp->z_size - off; 1641 1642 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1643 1644 if (error == 0) { 1645 /* 1646 * In FreeBSD we cannot free block in the middle of a file, 1647 * but only at the end of a file, so this code path should 1648 * never happen. 1649 */ 1650 vnode_pager_setsize(ZTOV(zp), off); 1651 } 1652 1653 zfs_range_unlock(rl); 1654 1655 return (error); 1656} 1657 1658/* 1659 * Truncate a file 1660 * 1661 * IN: zp - znode of file to free data in. 1662 * end - new end-of-file. 1663 * 1664 * RETURN: 0 if success 1665 * error code if failure 1666 */ 1667static int 1668zfs_trunc(znode_t *zp, uint64_t end) 1669{ 1670 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1671 vnode_t *vp = ZTOV(zp); 1672 dmu_tx_t *tx; 1673 rl_t *rl; 1674 int error; 1675 sa_bulk_attr_t bulk[2]; 1676 int count = 0; 1677 1678 /* 1679 * We will change zp_size, lock the whole file. 1680 */ 1681 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1682 1683 /* 1684 * Nothing to do if file already at desired length. 1685 */ 1686 if (end >= zp->z_size) { 1687 zfs_range_unlock(rl); 1688 return (0); 1689 } 1690 1691 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1692 if (error) { 1693 zfs_range_unlock(rl); 1694 return (error); 1695 } 1696top: 1697 tx = dmu_tx_create(zfsvfs->z_os); 1698 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1699 zfs_sa_upgrade_txholds(tx, zp); 1700 error = dmu_tx_assign(tx, TXG_NOWAIT); 1701 if (error) { 1702 if (error == ERESTART) { 1703 dmu_tx_wait(tx); 1704 dmu_tx_abort(tx); 1705 goto top; 1706 } 1707 dmu_tx_abort(tx); 1708 zfs_range_unlock(rl); 1709 return (error); 1710 } 1711 1712 zp->z_size = end; 1713 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1714 NULL, &zp->z_size, sizeof (zp->z_size)); 1715 1716 if (end == 0) { 1717 zp->z_pflags &= ~ZFS_SPARSE; 1718 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1719 NULL, &zp->z_pflags, 8); 1720 } 1721 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1722 1723 dmu_tx_commit(tx); 1724 1725 /* 1726 * Clear any mapped pages in the truncated region. This has to 1727 * happen outside of the transaction to avoid the possibility of 1728 * a deadlock with someone trying to push a page that we are 1729 * about to invalidate. 1730 */ 1731 vnode_pager_setsize(vp, end); 1732 1733 zfs_range_unlock(rl); 1734 1735 return (0); 1736} 1737 1738/* 1739 * Free space in a file 1740 * 1741 * IN: zp - znode of file to free data in. 1742 * off - start of range 1743 * len - end of range (0 => EOF) 1744 * flag - current file open mode flags. 1745 * log - TRUE if this action should be logged 1746 * 1747 * RETURN: 0 if success 1748 * error code if failure 1749 */ 1750int 1751zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1752{ 1753 vnode_t *vp = ZTOV(zp); 1754 dmu_tx_t *tx; 1755 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1756 zilog_t *zilog = zfsvfs->z_log; 1757 uint64_t mode; 1758 uint64_t mtime[2], ctime[2]; 1759 sa_bulk_attr_t bulk[3]; 1760 int count = 0; 1761 int error; 1762 1763 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1764 sizeof (mode))) != 0) 1765 return (error); 1766 1767 if (off > zp->z_size) { 1768 error = zfs_extend(zp, off+len); 1769 if (error == 0 && log) 1770 goto log; 1771 else 1772 return (error); 1773 } 1774 1775 /* 1776 * Check for any locks in the region to be freed. 1777 */ 1778 1779 if (MANDLOCK(vp, (mode_t)mode)) { 1780 uint64_t length = (len ? len : zp->z_size - off); 1781 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1782 return (error); 1783 } 1784 1785 if (len == 0) { 1786 error = zfs_trunc(zp, off); 1787 } else { 1788 if ((error = zfs_free_range(zp, off, len)) == 0 && 1789 off + len > zp->z_size) 1790 error = zfs_extend(zp, off+len); 1791 } 1792 if (error || !log) 1793 return (error); 1794log: 1795 tx = dmu_tx_create(zfsvfs->z_os); 1796 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1797 zfs_sa_upgrade_txholds(tx, zp); 1798 error = dmu_tx_assign(tx, TXG_NOWAIT); 1799 if (error) { 1800 if (error == ERESTART) { 1801 dmu_tx_wait(tx); 1802 dmu_tx_abort(tx); 1803 goto log; 1804 } 1805 dmu_tx_abort(tx); 1806 return (error); 1807 } 1808 1809 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1810 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1811 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1812 NULL, &zp->z_pflags, 8); 1813 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1814 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1815 ASSERT(error == 0); 1816 1817 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1818 1819 dmu_tx_commit(tx); 1820 return (0); 1821} 1822 1823void 1824zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1825{ 1826 zfsvfs_t zfsvfs; 1827 uint64_t moid, obj, sa_obj, version; 1828 uint64_t sense = ZFS_CASE_SENSITIVE; 1829 uint64_t norm = 0; 1830 nvpair_t *elem; 1831 int error; 1832 int i; 1833 znode_t *rootzp = NULL; 1834 vnode_t vnode; 1835 vattr_t vattr; 1836 znode_t *zp; 1837 zfs_acl_ids_t acl_ids; 1838 1839 /* 1840 * First attempt to create master node. 1841 */ 1842 /* 1843 * In an empty objset, there are no blocks to read and thus 1844 * there can be no i/o errors (which we assert below). 1845 */ 1846 moid = MASTER_NODE_OBJ; 1847 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1848 DMU_OT_NONE, 0, tx); 1849 ASSERT(error == 0); 1850 1851 /* 1852 * Set starting attributes. 1853 */ 1854 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1855 elem = NULL; 1856 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1857 /* For the moment we expect all zpl props to be uint64_ts */ 1858 uint64_t val; 1859 char *name; 1860 1861 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1862 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1863 name = nvpair_name(elem); 1864 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1865 if (val < version) 1866 version = val; 1867 } else { 1868 error = zap_update(os, moid, name, 8, 1, &val, tx); 1869 } 1870 ASSERT(error == 0); 1871 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1872 norm = val; 1873 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1874 sense = val; 1875 } 1876 ASSERT(version != 0); 1877 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1878 1879 /* 1880 * Create zap object used for SA attribute registration 1881 */ 1882 1883 if (version >= ZPL_VERSION_SA) { 1884 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1885 DMU_OT_NONE, 0, tx); 1886 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1887 ASSERT(error == 0); 1888 } else { 1889 sa_obj = 0; 1890 } 1891 /* 1892 * Create a delete queue. 1893 */ 1894 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1895 1896 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1897 ASSERT(error == 0); 1898 1899 /* 1900 * Create root znode. Create minimal znode/vnode/zfsvfs 1901 * to allow zfs_mknode to work. 1902 */ 1903 VATTR_NULL(&vattr); 1904 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1905 vattr.va_type = VDIR; 1906 vattr.va_mode = S_IFDIR|0755; 1907 vattr.va_uid = crgetuid(cr); 1908 vattr.va_gid = crgetgid(cr); 1909 1910 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1911 1912 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1913 zfs_znode_cache_constructor(rootzp, NULL, 0); 1914 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1915 rootzp->z_moved = 0; 1916 rootzp->z_unlinked = 0; 1917 rootzp->z_atime_dirty = 0; 1918 rootzp->z_is_sa = USE_SA(version, os); 1919 1920 vnode.v_type = VDIR; 1921 vnode.v_data = rootzp; 1922 rootzp->z_vnode = &vnode; 1923 1924 zfsvfs.z_os = os; 1925 zfsvfs.z_parent = &zfsvfs; 1926 zfsvfs.z_version = version; 1927 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1928 zfsvfs.z_use_sa = USE_SA(version, os); 1929 zfsvfs.z_norm = norm; 1930 1931 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1932 &zfsvfs.z_attr_table); 1933 1934 ASSERT(error == 0); 1935 1936 /* 1937 * Fold case on file systems that are always or sometimes case 1938 * insensitive. 1939 */ 1940 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1941 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1942 1943 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1944 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1945 offsetof(znode_t, z_link_node)); 1946 1947 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1948 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1949 1950 rootzp->z_zfsvfs = &zfsvfs; 1951 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1952 cr, NULL, &acl_ids)); 1953 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1954 ASSERT3P(zp, ==, rootzp); 1955 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1956 ASSERT(error == 0); 1957 zfs_acl_ids_free(&acl_ids); 1958 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1959 1960 sa_handle_destroy(rootzp->z_sa_hdl); 1961 rootzp->z_vnode = NULL; 1962 kmem_cache_free(znode_cache, rootzp); 1963 1964 /* 1965 * Create shares directory 1966 */ 1967 1968 error = zfs_create_share_dir(&zfsvfs, tx); 1969 1970 ASSERT(error == 0); 1971 1972 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1973 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1974} 1975 1976#endif /* _KERNEL */ 1977 1978static int 1979zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1980{ 1981 uint64_t sa_obj = 0; 1982 int error; 1983 1984 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1985 if (error != 0 && error != ENOENT) 1986 return (error); 1987 1988 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1989 return (error); 1990} 1991 1992static int 1993zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1994 dmu_buf_t **db, void *tag) 1995{ 1996 dmu_object_info_t doi; 1997 int error; 1998 1999 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 2000 return (error); 2001 2002 dmu_object_info_from_db(*db, &doi); 2003 if ((doi.doi_bonus_type != DMU_OT_SA && 2004 doi.doi_bonus_type != DMU_OT_ZNODE) || 2005 doi.doi_bonus_type == DMU_OT_ZNODE && 2006 doi.doi_bonus_size < sizeof (znode_phys_t)) { 2007 sa_buf_rele(*db, tag); 2008 return (ENOTSUP); 2009 } 2010 2011 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 2012 if (error != 0) { 2013 sa_buf_rele(*db, tag); 2014 return (error); 2015 } 2016 2017 return (0); 2018} 2019 2020void 2021zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 2022{ 2023 sa_handle_destroy(hdl); 2024 sa_buf_rele(db, tag); 2025} 2026 2027/* 2028 * Given an object number, return its parent object number and whether 2029 * or not the object is an extended attribute directory. 2030 */ 2031static int 2032zfs_obj_to_pobj(sa_handle_t *hdl, sa_attr_type_t *sa_table, uint64_t *pobjp, 2033 int *is_xattrdir) 2034{ 2035 uint64_t parent; 2036 uint64_t pflags; 2037 uint64_t mode; 2038 sa_bulk_attr_t bulk[3]; 2039 int count = 0; 2040 int error; 2041 2042 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2043 &parent, sizeof (parent)); 2044 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2045 &pflags, sizeof (pflags)); 2046 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2047 &mode, sizeof (mode)); 2048 2049 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2050 return (error); 2051 2052 *pobjp = parent; 2053 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2054 2055 return (0); 2056} 2057 2058/* 2059 * Given an object number, return some zpl level statistics 2060 */ 2061static int 2062zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2063 zfs_stat_t *sb) 2064{ 2065 sa_bulk_attr_t bulk[4]; 2066 int count = 0; 2067 2068 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2069 &sb->zs_mode, sizeof (sb->zs_mode)); 2070 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2071 &sb->zs_gen, sizeof (sb->zs_gen)); 2072 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2073 &sb->zs_links, sizeof (sb->zs_links)); 2074 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2075 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2076 2077 return (sa_bulk_lookup(hdl, bulk, count)); 2078} 2079 2080static int 2081zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2082 sa_attr_type_t *sa_table, char *buf, int len) 2083{ 2084 sa_handle_t *sa_hdl; 2085 sa_handle_t *prevhdl = NULL; 2086 dmu_buf_t *prevdb = NULL; 2087 dmu_buf_t *sa_db = NULL; 2088 char *path = buf + len - 1; 2089 int error; 2090 2091 *path = '\0'; 2092 sa_hdl = hdl; 2093 2094 for (;;) { 2095 uint64_t pobj; 2096 char component[MAXNAMELEN + 2]; 2097 size_t complen; 2098 int is_xattrdir; 2099 2100 if (prevdb) 2101 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2102 2103 if ((error = zfs_obj_to_pobj(sa_hdl, sa_table, &pobj, 2104 &is_xattrdir)) != 0) 2105 break; 2106 2107 if (pobj == obj) { 2108 if (path[0] != '/') 2109 *--path = '/'; 2110 break; 2111 } 2112 2113 component[0] = '/'; 2114 if (is_xattrdir) { 2115 (void) sprintf(component + 1, "<xattrdir>"); 2116 } else { 2117 error = zap_value_search(osp, pobj, obj, 2118 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2119 if (error != 0) 2120 break; 2121 } 2122 2123 complen = strlen(component); 2124 path -= complen; 2125 ASSERT(path >= buf); 2126 bcopy(component, path, complen); 2127 obj = pobj; 2128 2129 if (sa_hdl != hdl) { 2130 prevhdl = sa_hdl; 2131 prevdb = sa_db; 2132 } 2133 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2134 if (error != 0) { 2135 sa_hdl = prevhdl; 2136 sa_db = prevdb; 2137 break; 2138 } 2139 } 2140 2141 if (sa_hdl != NULL && sa_hdl != hdl) { 2142 ASSERT(sa_db != NULL); 2143 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2144 } 2145 2146 if (error == 0) 2147 (void) memmove(buf, path, buf + len - path); 2148 2149 return (error); 2150} 2151 2152int 2153zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2154{ 2155 sa_attr_type_t *sa_table; 2156 sa_handle_t *hdl; 2157 dmu_buf_t *db; 2158 int error; 2159 2160 error = zfs_sa_setup(osp, &sa_table); 2161 if (error != 0) 2162 return (error); 2163 2164 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2165 if (error != 0) 2166 return (error); 2167 2168 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2169 2170 zfs_release_sa_handle(hdl, db, FTAG); 2171 return (error); 2172} 2173 2174int 2175zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2176 char *buf, int len) 2177{ 2178 char *path = buf + len - 1; 2179 sa_attr_type_t *sa_table; 2180 sa_handle_t *hdl; 2181 dmu_buf_t *db; 2182 int error; 2183 2184 *path = '\0'; 2185 2186 error = zfs_sa_setup(osp, &sa_table); 2187 if (error != 0) 2188 return (error); 2189 2190 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2191 if (error != 0) 2192 return (error); 2193 2194 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2195 if (error != 0) { 2196 zfs_release_sa_handle(hdl, db, FTAG); 2197 return (error); 2198 } 2199 2200 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2201 2202 zfs_release_sa_handle(hdl, db, FTAG); 2203 return (error); 2204} 2205