zfs_znode.c revision 219089
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 case VFIFO: 704 vp->v_op = &zfs_fifoops; 705 break; 706 case VREG: 707 if (parent == zfsvfs->z_shares_dir) { 708 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 709 vp->v_op = &zfs_shareops; 710 } 711 break; 712 } 713 if (vp->v_type != VFIFO) 714 VN_LOCK_ASHARE(vp); 715 716 mutex_enter(&zfsvfs->z_znodes_lock); 717 list_insert_tail(&zfsvfs->z_all_znodes, zp); 718 membar_producer(); 719 /* 720 * Everything else must be valid before assigning z_zfsvfs makes the 721 * znode eligible for zfs_znode_move(). 722 */ 723 zp->z_zfsvfs = zfsvfs; 724 mutex_exit(&zfsvfs->z_znodes_lock); 725 726 VFS_HOLD(zfsvfs->z_vfs); 727 return (zp); 728} 729 730static uint64_t empty_xattr; 731static uint64_t pad[4]; 732static zfs_acl_phys_t acl_phys; 733/* 734 * Create a new DMU object to hold a zfs znode. 735 * 736 * IN: dzp - parent directory for new znode 737 * vap - file attributes for new znode 738 * tx - dmu transaction id for zap operations 739 * cr - credentials of caller 740 * flag - flags: 741 * IS_ROOT_NODE - new object will be root 742 * IS_XATTR - new object is an attribute 743 * bonuslen - length of bonus buffer 744 * setaclp - File/Dir initial ACL 745 * fuidp - Tracks fuid allocation. 746 * 747 * OUT: zpp - allocated znode 748 * 749 */ 750void 751zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 752 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 753{ 754 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 755 uint64_t mode, size, links, parent, pflags; 756 uint64_t dzp_pflags = 0; 757 uint64_t rdev = 0; 758 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 759 dmu_buf_t *db; 760 timestruc_t now; 761 uint64_t gen, obj; 762 int err; 763 int bonuslen; 764 sa_handle_t *sa_hdl; 765 dmu_object_type_t obj_type; 766 sa_bulk_attr_t sa_attrs[ZPL_END]; 767 int cnt = 0; 768 zfs_acl_locator_cb_t locate = { 0 }; 769 770 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 771 772 if (zfsvfs->z_replay) { 773 obj = vap->va_nodeid; 774 now = vap->va_ctime; /* see zfs_replay_create() */ 775 gen = vap->va_nblocks; /* ditto */ 776 } else { 777 obj = 0; 778 gethrestime(&now); 779 gen = dmu_tx_get_txg(tx); 780 } 781 782 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 783 bonuslen = (obj_type == DMU_OT_SA) ? 784 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 785 786 /* 787 * Create a new DMU object. 788 */ 789 /* 790 * There's currently no mechanism for pre-reading the blocks that will 791 * be needed to allocate a new object, so we accept the small chance 792 * that there will be an i/o error and we will fail one of the 793 * assertions below. 794 */ 795 if (vap->va_type == VDIR) { 796 if (zfsvfs->z_replay) { 797 err = zap_create_claim_norm(zfsvfs->z_os, obj, 798 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 799 obj_type, bonuslen, tx); 800 ASSERT3U(err, ==, 0); 801 } else { 802 obj = zap_create_norm(zfsvfs->z_os, 803 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 804 obj_type, bonuslen, tx); 805 } 806 } else { 807 if (zfsvfs->z_replay) { 808 err = dmu_object_claim(zfsvfs->z_os, obj, 809 DMU_OT_PLAIN_FILE_CONTENTS, 0, 810 obj_type, bonuslen, tx); 811 ASSERT3U(err, ==, 0); 812 } else { 813 obj = dmu_object_alloc(zfsvfs->z_os, 814 DMU_OT_PLAIN_FILE_CONTENTS, 0, 815 obj_type, bonuslen, tx); 816 } 817 } 818 819 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 820 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 821 822 /* 823 * If this is the root, fix up the half-initialized parent pointer 824 * to reference the just-allocated physical data area. 825 */ 826 if (flag & IS_ROOT_NODE) { 827 dzp->z_id = obj; 828 } else { 829 dzp_pflags = dzp->z_pflags; 830 } 831 832 /* 833 * If parent is an xattr, so am I. 834 */ 835 if (dzp_pflags & ZFS_XATTR) { 836 flag |= IS_XATTR; 837 } 838 839 if (zfsvfs->z_use_fuids) 840 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 841 else 842 pflags = 0; 843 844 if (vap->va_type == VDIR) { 845 size = 2; /* contents ("." and "..") */ 846 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 847 } else { 848 size = links = 0; 849 } 850 851 if (vap->va_type == VBLK || vap->va_type == VCHR) { 852 rdev = zfs_expldev(vap->va_rdev); 853 } 854 855 parent = dzp->z_id; 856 mode = acl_ids->z_mode; 857 if (flag & IS_XATTR) 858 pflags |= ZFS_XATTR; 859 860 /* 861 * No execs denied will be deterimed when zfs_mode_compute() is called. 862 */ 863 pflags |= acl_ids->z_aclp->z_hints & 864 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 865 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 866 867 ZFS_TIME_ENCODE(&now, crtime); 868 ZFS_TIME_ENCODE(&now, ctime); 869 870 if (vap->va_mask & AT_ATIME) { 871 ZFS_TIME_ENCODE(&vap->va_atime, atime); 872 } else { 873 ZFS_TIME_ENCODE(&now, atime); 874 } 875 876 if (vap->va_mask & AT_MTIME) { 877 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 878 } else { 879 ZFS_TIME_ENCODE(&now, mtime); 880 } 881 882 /* Now add in all of the "SA" attributes */ 883 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 884 &sa_hdl)); 885 886 /* 887 * Setup the array of attributes to be replaced/set on the new file 888 * 889 * order for DMU_OT_ZNODE is critical since it needs to be constructed 890 * in the old znode_phys_t format. Don't change this ordering 891 */ 892 893 if (obj_type == DMU_OT_ZNODE) { 894 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 895 NULL, &atime, 16); 896 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 897 NULL, &mtime, 16); 898 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 899 NULL, &ctime, 16); 900 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 901 NULL, &crtime, 16); 902 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 903 NULL, &gen, 8); 904 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 905 NULL, &mode, 8); 906 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 907 NULL, &size, 8); 908 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 909 NULL, &parent, 8); 910 } else { 911 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 912 NULL, &mode, 8); 913 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 914 NULL, &size, 8); 915 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 916 NULL, &gen, 8); 917 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 918 &acl_ids->z_fuid, 8); 919 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 920 &acl_ids->z_fgid, 8); 921 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 922 NULL, &parent, 8); 923 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 924 NULL, &pflags, 8); 925 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 926 NULL, &atime, 16); 927 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 928 NULL, &mtime, 16); 929 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 930 NULL, &ctime, 16); 931 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 932 NULL, &crtime, 16); 933 } 934 935 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 936 937 if (obj_type == DMU_OT_ZNODE) { 938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 939 &empty_xattr, 8); 940 } 941 if (obj_type == DMU_OT_ZNODE || 942 (vap->va_type == VBLK || vap->va_type == VCHR)) { 943 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 944 NULL, &rdev, 8); 945 946 } 947 if (obj_type == DMU_OT_ZNODE) { 948 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 949 NULL, &pflags, 8); 950 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 951 &acl_ids->z_fuid, 8); 952 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 953 &acl_ids->z_fgid, 8); 954 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 955 sizeof (uint64_t) * 4); 956 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 957 &acl_phys, sizeof (zfs_acl_phys_t)); 958 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 960 &acl_ids->z_aclp->z_acl_count, 8); 961 locate.cb_aclp = acl_ids->z_aclp; 962 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 963 zfs_acl_data_locator, &locate, 964 acl_ids->z_aclp->z_acl_bytes); 965 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 966 acl_ids->z_fuid, acl_ids->z_fgid); 967 } 968 969 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 970 971 if (!(flag & IS_ROOT_NODE)) { 972 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 973 ASSERT(*zpp != NULL); 974 } else { 975 /* 976 * If we are creating the root node, the "parent" we 977 * passed in is the znode for the root. 978 */ 979 *zpp = dzp; 980 981 (*zpp)->z_sa_hdl = sa_hdl; 982 } 983 984 (*zpp)->z_pflags = pflags; 985 (*zpp)->z_mode = mode; 986 987 if (vap->va_mask & AT_XVATTR) 988 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 989 990 if (obj_type == DMU_OT_ZNODE || 991 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 992 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx); 993 ASSERT3P(err, ==, 0); 994 } 995 if (!(flag & IS_ROOT_NODE)) { 996 vnode_t *vp; 997 998 vp = ZTOV(*zpp); 999 vp->v_vflag |= VV_FORCEINSMQ; 1000 err = insmntque(vp, zfsvfs->z_vfs); 1001 vp->v_vflag &= ~VV_FORCEINSMQ; 1002 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1003 } 1004 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1005} 1006 1007/* 1008 * zfs_xvattr_set only updates the in-core attributes 1009 * it is assumed the caller will be doing an sa_bulk_update 1010 * to push the changes out 1011 */ 1012void 1013zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1014{ 1015 xoptattr_t *xoap; 1016 1017 xoap = xva_getxoptattr(xvap); 1018 ASSERT(xoap); 1019 1020 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1021 uint64_t times[2]; 1022 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1023 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1024 ×, sizeof (times), tx); 1025 XVA_SET_RTN(xvap, XAT_CREATETIME); 1026 } 1027 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1028 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1029 zp->z_pflags, tx); 1030 XVA_SET_RTN(xvap, XAT_READONLY); 1031 } 1032 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1033 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1034 zp->z_pflags, tx); 1035 XVA_SET_RTN(xvap, XAT_HIDDEN); 1036 } 1037 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1038 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1039 zp->z_pflags, tx); 1040 XVA_SET_RTN(xvap, XAT_SYSTEM); 1041 } 1042 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1043 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1044 zp->z_pflags, tx); 1045 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1046 } 1047 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1048 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1049 zp->z_pflags, tx); 1050 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1051 } 1052 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1053 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1054 zp->z_pflags, tx); 1055 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1056 } 1057 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1058 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1059 zp->z_pflags, tx); 1060 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1061 } 1062 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1063 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1064 zp->z_pflags, tx); 1065 XVA_SET_RTN(xvap, XAT_NODUMP); 1066 } 1067 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1068 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1069 zp->z_pflags, tx); 1070 XVA_SET_RTN(xvap, XAT_OPAQUE); 1071 } 1072 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1073 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1074 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1075 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1076 } 1077 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1078 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1079 zp->z_pflags, tx); 1080 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1081 } 1082 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1083 zfs_sa_set_scanstamp(zp, xvap, tx); 1084 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1085 } 1086 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1087 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1088 zp->z_pflags, tx); 1089 XVA_SET_RTN(xvap, XAT_REPARSE); 1090 } 1091 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1092 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1093 zp->z_pflags, tx); 1094 XVA_SET_RTN(xvap, XAT_OFFLINE); 1095 } 1096 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1097 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1098 zp->z_pflags, tx); 1099 XVA_SET_RTN(xvap, XAT_SPARSE); 1100 } 1101} 1102 1103int 1104zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1105{ 1106 dmu_object_info_t doi; 1107 dmu_buf_t *db; 1108 znode_t *zp; 1109 int err; 1110 sa_handle_t *hdl; 1111 int first = 1; 1112 1113 *zpp = NULL; 1114 1115again: 1116 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1117 1118 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1119 if (err) { 1120 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1121 return (err); 1122 } 1123 1124 dmu_object_info_from_db(db, &doi); 1125 if (doi.doi_bonus_type != DMU_OT_SA && 1126 (doi.doi_bonus_type != DMU_OT_ZNODE || 1127 (doi.doi_bonus_type == DMU_OT_ZNODE && 1128 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1129 sa_buf_rele(db, NULL); 1130 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1131 return (EINVAL); 1132 } 1133 1134 hdl = dmu_buf_get_user(db); 1135 if (hdl != NULL) { 1136 zp = sa_get_userdata(hdl); 1137 1138 1139 /* 1140 * Since "SA" does immediate eviction we 1141 * should never find a sa handle that doesn't 1142 * know about the znode. 1143 */ 1144 1145 ASSERT3P(zp, !=, NULL); 1146 1147 mutex_enter(&zp->z_lock); 1148 ASSERT3U(zp->z_id, ==, obj_num); 1149 if (zp->z_unlinked) { 1150 err = ENOENT; 1151 } else { 1152 vnode_t *vp; 1153 int dying = 0; 1154 1155 vp = ZTOV(zp); 1156 if (vp == NULL) 1157 dying = 1; 1158 else { 1159 VN_HOLD(vp); 1160 if ((vp->v_iflag & VI_DOOMED) != 0) { 1161 dying = 1; 1162 /* 1163 * Don't VN_RELE() vnode here, because 1164 * it can call vn_lock() which creates 1165 * LOR between vnode lock and znode 1166 * lock. We will VN_RELE() the vnode 1167 * after droping znode lock. 1168 */ 1169 } 1170 } 1171 if (dying) { 1172 if (first) { 1173 ZFS_LOG(1, "dying znode detected (zp=%p)", zp); 1174 first = 0; 1175 } 1176 /* 1177 * znode is dying so we can't reuse it, we must 1178 * wait until destruction is completed. 1179 */ 1180 sa_buf_rele(db, NULL); 1181 mutex_exit(&zp->z_lock); 1182 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1183 if (vp != NULL) 1184 VN_RELE(vp); 1185 tsleep(zp, 0, "zcollide", 1); 1186 goto again; 1187 } 1188 *zpp = zp; 1189 err = 0; 1190 } 1191 sa_buf_rele(db, NULL); 1192 mutex_exit(&zp->z_lock); 1193 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1194 return (err); 1195 } 1196 1197 /* 1198 * Not found create new znode/vnode 1199 * but only if file exists. 1200 * 1201 * There is a small window where zfs_vget() could 1202 * find this object while a file create is still in 1203 * progress. This is checked for in zfs_znode_alloc() 1204 * 1205 * if zfs_znode_alloc() fails it will drop the hold on the 1206 * bonus buffer. 1207 */ 1208 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1209 doi.doi_bonus_type, NULL); 1210 if (zp == NULL) { 1211 err = ENOENT; 1212 } else { 1213 *zpp = zp; 1214 } 1215 if (err == 0) { 1216 vnode_t *vp = ZTOV(zp); 1217 1218 err = insmntque(vp, zfsvfs->z_vfs); 1219 if (err == 0) 1220 VOP_UNLOCK(vp, 0); 1221 else { 1222 zp->z_vnode = NULL; 1223 zfs_znode_dmu_fini(zp); 1224 zfs_znode_free(zp); 1225 *zpp = NULL; 1226 } 1227 } 1228 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1229 return (err); 1230} 1231 1232int 1233zfs_rezget(znode_t *zp) 1234{ 1235 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1236 dmu_object_info_t doi; 1237 dmu_buf_t *db; 1238 uint64_t obj_num = zp->z_id; 1239 uint64_t mode, size; 1240 sa_bulk_attr_t bulk[8]; 1241 int err; 1242 int count = 0; 1243 uint64_t gen; 1244 1245 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1246 1247 mutex_enter(&zp->z_acl_lock); 1248 if (zp->z_acl_cached) { 1249 zfs_acl_free(zp->z_acl_cached); 1250 zp->z_acl_cached = NULL; 1251 } 1252 1253 mutex_exit(&zp->z_acl_lock); 1254 ASSERT(zp->z_sa_hdl == NULL); 1255 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1256 if (err) { 1257 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1258 return (err); 1259 } 1260 1261 dmu_object_info_from_db(db, &doi); 1262 if (doi.doi_bonus_type != DMU_OT_SA && 1263 (doi.doi_bonus_type != DMU_OT_ZNODE || 1264 (doi.doi_bonus_type == DMU_OT_ZNODE && 1265 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1266 sa_buf_rele(db, NULL); 1267 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1268 return (EINVAL); 1269 } 1270 1271 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1272 size = zp->z_size; 1273 1274 /* reload cached values */ 1275 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1276 &gen, sizeof (gen)); 1277 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1278 &zp->z_size, sizeof (zp->z_size)); 1279 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1280 &zp->z_links, sizeof (zp->z_links)); 1281 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1282 &zp->z_pflags, sizeof (zp->z_pflags)); 1283 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1284 &zp->z_atime, sizeof (zp->z_atime)); 1285 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1286 &zp->z_uid, sizeof (zp->z_uid)); 1287 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1288 &zp->z_gid, sizeof (zp->z_gid)); 1289 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1290 &mode, sizeof (mode)); 1291 1292 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1293 zfs_znode_dmu_fini(zp); 1294 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1295 return (EIO); 1296 } 1297 1298 zp->z_mode = mode; 1299 1300 if (gen != zp->z_gen) { 1301 zfs_znode_dmu_fini(zp); 1302 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1303 return (EIO); 1304 } 1305 1306 /* 1307 * XXXPJD: Not sure how is that possible, but under heavy 1308 * zfs recv -F load it happens that z_gen is the same, but 1309 * vnode type is different than znode type. This would mean 1310 * that for example regular file was replaced with directory 1311 * which has the same object number. 1312 */ 1313 if (ZTOV(zp) != NULL && 1314 ZTOV(zp)->v_type != IFTOVT((mode_t)zp->z_mode)) { 1315 zfs_znode_dmu_fini(zp); 1316 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1317 return (EIO); 1318 } 1319 1320 zp->z_unlinked = (zp->z_links == 0); 1321 zp->z_blksz = doi.doi_data_block_size; 1322 if (zp->z_size != size && ZTOV(zp) != NULL) 1323 vnode_pager_setsize(ZTOV(zp), zp->z_size); 1324 1325 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1326 1327 return (0); 1328} 1329 1330void 1331zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1332{ 1333 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1334 objset_t *os = zfsvfs->z_os; 1335 uint64_t obj = zp->z_id; 1336 uint64_t acl_obj = zfs_external_acl(zp); 1337 1338 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1339 if (acl_obj) { 1340 VERIFY(!zp->z_is_sa); 1341 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1342 } 1343 VERIFY(0 == dmu_object_free(os, obj, tx)); 1344 zfs_znode_dmu_fini(zp); 1345 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1346 zfs_znode_free(zp); 1347} 1348 1349void 1350zfs_zinactive(znode_t *zp) 1351{ 1352 vnode_t *vp = ZTOV(zp); 1353 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1354 uint64_t z_id = zp->z_id; 1355 int vfslocked; 1356 1357 ASSERT(zp->z_sa_hdl); 1358 1359 /* 1360 * Don't allow a zfs_zget() while were trying to release this znode 1361 */ 1362 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1363 1364 mutex_enter(&zp->z_lock); 1365 VI_LOCK(vp); 1366 if (vp->v_count > 0) { 1367 /* 1368 * If the hold count is greater than zero, somebody has 1369 * obtained a new reference on this znode while we were 1370 * processing it here, so we are done. 1371 */ 1372 VI_UNLOCK(vp); 1373 mutex_exit(&zp->z_lock); 1374 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1375 return; 1376 } 1377 VI_UNLOCK(vp); 1378 1379 /* 1380 * If this was the last reference to a file with no links, 1381 * remove the file from the file system. 1382 */ 1383 if (zp->z_unlinked) { 1384 mutex_exit(&zp->z_lock); 1385 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1386 ASSERT(vp->v_count == 0); 1387 vrecycle(vp, curthread); 1388 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs); 1389 zfs_rmnode(zp); 1390 VFS_UNLOCK_GIANT(vfslocked); 1391 return; 1392 } 1393 1394 mutex_exit(&zp->z_lock); 1395 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1396} 1397 1398void 1399zfs_znode_free(znode_t *zp) 1400{ 1401 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1402 1403 ASSERT(ZTOV(zp) == NULL); 1404 ASSERT(zp->z_sa_hdl == NULL); 1405 mutex_enter(&zfsvfs->z_znodes_lock); 1406 POINTER_INVALIDATE(&zp->z_zfsvfs); 1407 list_remove(&zfsvfs->z_all_znodes, zp); 1408 mutex_exit(&zfsvfs->z_znodes_lock); 1409 1410 if (zp->z_acl_cached) { 1411 zfs_acl_free(zp->z_acl_cached); 1412 zp->z_acl_cached = NULL; 1413 } 1414 1415 kmem_cache_free(znode_cache, zp); 1416 1417 VFS_RELE(zfsvfs->z_vfs); 1418} 1419 1420void 1421zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1422 uint64_t ctime[2], boolean_t have_tx) 1423{ 1424 timestruc_t now; 1425 1426 gethrestime(&now); 1427 1428 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1429 zp->z_atime_dirty = 0; 1430 zp->z_seq++; 1431 } else { 1432 zp->z_atime_dirty = 1; 1433 } 1434 1435 if (flag & AT_ATIME) { 1436 ZFS_TIME_ENCODE(&now, zp->z_atime); 1437 } 1438 1439 if (flag & AT_MTIME) { 1440 ZFS_TIME_ENCODE(&now, mtime); 1441 if (zp->z_zfsvfs->z_use_fuids) { 1442 zp->z_pflags |= (ZFS_ARCHIVE | 1443 ZFS_AV_MODIFIED); 1444 } 1445 } 1446 1447 if (flag & AT_CTIME) { 1448 ZFS_TIME_ENCODE(&now, ctime); 1449 if (zp->z_zfsvfs->z_use_fuids) 1450 zp->z_pflags |= ZFS_ARCHIVE; 1451 } 1452} 1453 1454/* 1455 * Grow the block size for a file. 1456 * 1457 * IN: zp - znode of file to free data in. 1458 * size - requested block size 1459 * tx - open transaction. 1460 * 1461 * NOTE: this function assumes that the znode is write locked. 1462 */ 1463void 1464zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1465{ 1466 int error; 1467 u_longlong_t dummy; 1468 1469 if (size <= zp->z_blksz) 1470 return; 1471 /* 1472 * If the file size is already greater than the current blocksize, 1473 * we will not grow. If there is more than one block in a file, 1474 * the blocksize cannot change. 1475 */ 1476 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1477 return; 1478 1479 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1480 size, 0, tx); 1481 1482 if (error == ENOTSUP) 1483 return; 1484 ASSERT3U(error, ==, 0); 1485 1486 /* What blocksize did we actually get? */ 1487 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1488} 1489 1490#ifdef sun 1491/* 1492 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1493 * be calling back into the fs for a putpage(). E.g.: when truncating 1494 * a file, the pages being "thrown away* don't need to be written out. 1495 */ 1496/* ARGSUSED */ 1497static int 1498zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1499 int flags, cred_t *cr) 1500{ 1501 ASSERT(0); 1502 return (0); 1503} 1504#endif /* sun */ 1505 1506/* 1507 * Increase the file length 1508 * 1509 * IN: zp - znode of file to free data in. 1510 * end - new end-of-file 1511 * 1512 * RETURN: 0 if success 1513 * error code if failure 1514 */ 1515static int 1516zfs_extend(znode_t *zp, uint64_t end) 1517{ 1518 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1519 dmu_tx_t *tx; 1520 rl_t *rl; 1521 uint64_t newblksz; 1522 int error; 1523 1524 /* 1525 * We will change zp_size, lock the whole file. 1526 */ 1527 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1528 1529 /* 1530 * Nothing to do if file already at desired length. 1531 */ 1532 if (end <= zp->z_size) { 1533 zfs_range_unlock(rl); 1534 return (0); 1535 } 1536top: 1537 tx = dmu_tx_create(zfsvfs->z_os); 1538 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1539 zfs_sa_upgrade_txholds(tx, zp); 1540 if (end > zp->z_blksz && 1541 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1542 /* 1543 * We are growing the file past the current block size. 1544 */ 1545 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1546 ASSERT(!ISP2(zp->z_blksz)); 1547 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1548 } else { 1549 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1550 } 1551 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1552 } else { 1553 newblksz = 0; 1554 } 1555 1556 error = dmu_tx_assign(tx, TXG_NOWAIT); 1557 if (error) { 1558 if (error == ERESTART) { 1559 dmu_tx_wait(tx); 1560 dmu_tx_abort(tx); 1561 goto top; 1562 } 1563 dmu_tx_abort(tx); 1564 zfs_range_unlock(rl); 1565 return (error); 1566 } 1567 1568 if (newblksz) 1569 zfs_grow_blocksize(zp, newblksz, tx); 1570 1571 zp->z_size = end; 1572 1573 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1574 &zp->z_size, sizeof (zp->z_size), tx)); 1575 1576 vnode_pager_setsize(ZTOV(zp), end); 1577 1578 zfs_range_unlock(rl); 1579 1580 dmu_tx_commit(tx); 1581 1582 return (0); 1583} 1584 1585/* 1586 * Free space in a file. 1587 * 1588 * IN: zp - znode of file to free data in. 1589 * off - start of section to free. 1590 * len - length of section to free. 1591 * 1592 * RETURN: 0 if success 1593 * error code if failure 1594 */ 1595static int 1596zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1597{ 1598 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1599 rl_t *rl; 1600 int error; 1601 1602 /* 1603 * Lock the range being freed. 1604 */ 1605 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1606 1607 /* 1608 * Nothing to do if file already at desired length. 1609 */ 1610 if (off >= zp->z_size) { 1611 zfs_range_unlock(rl); 1612 return (0); 1613 } 1614 1615 if (off + len > zp->z_size) 1616 len = zp->z_size - off; 1617 1618 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1619 1620 if (error == 0) { 1621 /* 1622 * In FreeBSD we cannot free block in the middle of a file, 1623 * but only at the end of a file, so this code path should 1624 * never happen. 1625 */ 1626 vnode_pager_setsize(ZTOV(zp), off); 1627 } 1628 1629 zfs_range_unlock(rl); 1630 1631 return (error); 1632} 1633 1634/* 1635 * Truncate a file 1636 * 1637 * IN: zp - znode of file to free data in. 1638 * end - new end-of-file. 1639 * 1640 * RETURN: 0 if success 1641 * error code if failure 1642 */ 1643static int 1644zfs_trunc(znode_t *zp, uint64_t end) 1645{ 1646 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1647 vnode_t *vp = ZTOV(zp); 1648 dmu_tx_t *tx; 1649 rl_t *rl; 1650 int error; 1651 sa_bulk_attr_t bulk[2]; 1652 int count = 0; 1653 1654 /* 1655 * We will change zp_size, lock the whole file. 1656 */ 1657 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1658 1659 /* 1660 * Nothing to do if file already at desired length. 1661 */ 1662 if (end >= zp->z_size) { 1663 zfs_range_unlock(rl); 1664 return (0); 1665 } 1666 1667 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1668 if (error) { 1669 zfs_range_unlock(rl); 1670 return (error); 1671 } 1672top: 1673 tx = dmu_tx_create(zfsvfs->z_os); 1674 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1675 zfs_sa_upgrade_txholds(tx, zp); 1676 error = dmu_tx_assign(tx, TXG_NOWAIT); 1677 if (error) { 1678 if (error == ERESTART) { 1679 dmu_tx_wait(tx); 1680 dmu_tx_abort(tx); 1681 goto top; 1682 } 1683 dmu_tx_abort(tx); 1684 zfs_range_unlock(rl); 1685 return (error); 1686 } 1687 1688 zp->z_size = end; 1689 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1690 NULL, &zp->z_size, sizeof (zp->z_size)); 1691 1692 if (end == 0) { 1693 zp->z_pflags &= ~ZFS_SPARSE; 1694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1695 NULL, &zp->z_pflags, 8); 1696 } 1697 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1698 1699 dmu_tx_commit(tx); 1700 1701 /* 1702 * Clear any mapped pages in the truncated region. This has to 1703 * happen outside of the transaction to avoid the possibility of 1704 * a deadlock with someone trying to push a page that we are 1705 * about to invalidate. 1706 */ 1707 vnode_pager_setsize(vp, end); 1708 1709 zfs_range_unlock(rl); 1710 1711 return (0); 1712} 1713 1714/* 1715 * Free space in a file 1716 * 1717 * IN: zp - znode of file to free data in. 1718 * off - start of range 1719 * len - end of range (0 => EOF) 1720 * flag - current file open mode flags. 1721 * log - TRUE if this action should be logged 1722 * 1723 * RETURN: 0 if success 1724 * error code if failure 1725 */ 1726int 1727zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1728{ 1729 vnode_t *vp = ZTOV(zp); 1730 dmu_tx_t *tx; 1731 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1732 zilog_t *zilog = zfsvfs->z_log; 1733 uint64_t mode; 1734 uint64_t mtime[2], ctime[2]; 1735 sa_bulk_attr_t bulk[3]; 1736 int count = 0; 1737 int error; 1738 1739 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1740 sizeof (mode))) != 0) 1741 return (error); 1742 1743 if (off > zp->z_size) { 1744 error = zfs_extend(zp, off+len); 1745 if (error == 0 && log) 1746 goto log; 1747 else 1748 return (error); 1749 } 1750 1751 /* 1752 * Check for any locks in the region to be freed. 1753 */ 1754 1755 if (MANDLOCK(vp, (mode_t)mode)) { 1756 uint64_t length = (len ? len : zp->z_size - off); 1757 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1758 return (error); 1759 } 1760 1761 if (len == 0) { 1762 error = zfs_trunc(zp, off); 1763 } else { 1764 if ((error = zfs_free_range(zp, off, len)) == 0 && 1765 off + len > zp->z_size) 1766 error = zfs_extend(zp, off+len); 1767 } 1768 if (error || !log) 1769 return (error); 1770log: 1771 tx = dmu_tx_create(zfsvfs->z_os); 1772 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1773 zfs_sa_upgrade_txholds(tx, zp); 1774 error = dmu_tx_assign(tx, TXG_NOWAIT); 1775 if (error) { 1776 if (error == ERESTART) { 1777 dmu_tx_wait(tx); 1778 dmu_tx_abort(tx); 1779 goto log; 1780 } 1781 dmu_tx_abort(tx); 1782 return (error); 1783 } 1784 1785 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1786 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1787 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1788 NULL, &zp->z_pflags, 8); 1789 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1790 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1791 ASSERT(error == 0); 1792 1793 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1794 1795 dmu_tx_commit(tx); 1796 return (0); 1797} 1798 1799void 1800zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1801{ 1802 zfsvfs_t zfsvfs; 1803 uint64_t moid, obj, sa_obj, version; 1804 uint64_t sense = ZFS_CASE_SENSITIVE; 1805 uint64_t norm = 0; 1806 nvpair_t *elem; 1807 int error; 1808 int i; 1809 znode_t *rootzp = NULL; 1810 vnode_t vnode; 1811 vattr_t vattr; 1812 znode_t *zp; 1813 zfs_acl_ids_t acl_ids; 1814 1815 /* 1816 * First attempt to create master node. 1817 */ 1818 /* 1819 * In an empty objset, there are no blocks to read and thus 1820 * there can be no i/o errors (which we assert below). 1821 */ 1822 moid = MASTER_NODE_OBJ; 1823 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1824 DMU_OT_NONE, 0, tx); 1825 ASSERT(error == 0); 1826 1827 /* 1828 * Set starting attributes. 1829 */ 1830 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1831 elem = NULL; 1832 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1833 /* For the moment we expect all zpl props to be uint64_ts */ 1834 uint64_t val; 1835 char *name; 1836 1837 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1838 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1839 name = nvpair_name(elem); 1840 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1841 if (val < version) 1842 version = val; 1843 } else { 1844 error = zap_update(os, moid, name, 8, 1, &val, tx); 1845 } 1846 ASSERT(error == 0); 1847 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1848 norm = val; 1849 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1850 sense = val; 1851 } 1852 ASSERT(version != 0); 1853 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1854 1855 /* 1856 * Create zap object used for SA attribute registration 1857 */ 1858 1859 if (version >= ZPL_VERSION_SA) { 1860 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1861 DMU_OT_NONE, 0, tx); 1862 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1863 ASSERT(error == 0); 1864 } else { 1865 sa_obj = 0; 1866 } 1867 /* 1868 * Create a delete queue. 1869 */ 1870 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1871 1872 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1873 ASSERT(error == 0); 1874 1875 /* 1876 * Create root znode. Create minimal znode/vnode/zfsvfs 1877 * to allow zfs_mknode to work. 1878 */ 1879 VATTR_NULL(&vattr); 1880 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1881 vattr.va_type = VDIR; 1882 vattr.va_mode = S_IFDIR|0755; 1883 vattr.va_uid = crgetuid(cr); 1884 vattr.va_gid = crgetgid(cr); 1885 1886 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1887 1888 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1889 zfs_znode_cache_constructor(rootzp, NULL, 0); 1890 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1891 rootzp->z_moved = 0; 1892 rootzp->z_unlinked = 0; 1893 rootzp->z_atime_dirty = 0; 1894 rootzp->z_is_sa = USE_SA(version, os); 1895 1896 vnode.v_type = VDIR; 1897 vnode.v_data = rootzp; 1898 rootzp->z_vnode = &vnode; 1899 1900 zfsvfs.z_os = os; 1901 zfsvfs.z_parent = &zfsvfs; 1902 zfsvfs.z_version = version; 1903 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1904 zfsvfs.z_use_sa = USE_SA(version, os); 1905 zfsvfs.z_norm = norm; 1906 1907 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1908 &zfsvfs.z_attr_table); 1909 1910 ASSERT(error == 0); 1911 1912 /* 1913 * Fold case on file systems that are always or sometimes case 1914 * insensitive. 1915 */ 1916 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1917 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1918 1919 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1920 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1921 offsetof(znode_t, z_link_node)); 1922 1923 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1924 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1925 1926 rootzp->z_zfsvfs = &zfsvfs; 1927 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1928 cr, NULL, &acl_ids)); 1929 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1930 ASSERT3P(zp, ==, rootzp); 1931 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1932 ASSERT(error == 0); 1933 zfs_acl_ids_free(&acl_ids); 1934 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1935 1936 sa_handle_destroy(rootzp->z_sa_hdl); 1937 rootzp->z_vnode = NULL; 1938 kmem_cache_free(znode_cache, rootzp); 1939 1940 /* 1941 * Create shares directory 1942 */ 1943 1944 error = zfs_create_share_dir(&zfsvfs, tx); 1945 1946 ASSERT(error == 0); 1947 1948 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1949 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1950} 1951 1952#endif /* _KERNEL */ 1953 1954static int 1955zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1956{ 1957 uint64_t sa_obj = 0; 1958 int error; 1959 1960 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1961 if (error != 0 && error != ENOENT) 1962 return (error); 1963 1964 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1965 return (error); 1966} 1967 1968static int 1969zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1970 dmu_buf_t **db, void *tag) 1971{ 1972 dmu_object_info_t doi; 1973 int error; 1974 1975 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 1976 return (error); 1977 1978 dmu_object_info_from_db(*db, &doi); 1979 if ((doi.doi_bonus_type != DMU_OT_SA && 1980 doi.doi_bonus_type != DMU_OT_ZNODE) || 1981 doi.doi_bonus_type == DMU_OT_ZNODE && 1982 doi.doi_bonus_size < sizeof (znode_phys_t)) { 1983 sa_buf_rele(*db, tag); 1984 return (ENOTSUP); 1985 } 1986 1987 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 1988 if (error != 0) { 1989 sa_buf_rele(*db, tag); 1990 return (error); 1991 } 1992 1993 return (0); 1994} 1995 1996void 1997zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 1998{ 1999 sa_handle_destroy(hdl); 2000 sa_buf_rele(db, tag); 2001} 2002 2003/* 2004 * Given an object number, return its parent object number and whether 2005 * or not the object is an extended attribute directory. 2006 */ 2007static int 2008zfs_obj_to_pobj(sa_handle_t *hdl, sa_attr_type_t *sa_table, uint64_t *pobjp, 2009 int *is_xattrdir) 2010{ 2011 uint64_t parent; 2012 uint64_t pflags; 2013 uint64_t mode; 2014 sa_bulk_attr_t bulk[3]; 2015 int count = 0; 2016 int error; 2017 2018 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2019 &parent, sizeof (parent)); 2020 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2021 &pflags, sizeof (pflags)); 2022 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2023 &mode, sizeof (mode)); 2024 2025 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2026 return (error); 2027 2028 *pobjp = parent; 2029 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2030 2031 return (0); 2032} 2033 2034/* 2035 * Given an object number, return some zpl level statistics 2036 */ 2037static int 2038zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2039 zfs_stat_t *sb) 2040{ 2041 sa_bulk_attr_t bulk[4]; 2042 int count = 0; 2043 2044 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2045 &sb->zs_mode, sizeof (sb->zs_mode)); 2046 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2047 &sb->zs_gen, sizeof (sb->zs_gen)); 2048 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2049 &sb->zs_links, sizeof (sb->zs_links)); 2050 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2051 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2052 2053 return (sa_bulk_lookup(hdl, bulk, count)); 2054} 2055 2056static int 2057zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2058 sa_attr_type_t *sa_table, char *buf, int len) 2059{ 2060 sa_handle_t *sa_hdl; 2061 sa_handle_t *prevhdl = NULL; 2062 dmu_buf_t *prevdb = NULL; 2063 dmu_buf_t *sa_db = NULL; 2064 char *path = buf + len - 1; 2065 int error; 2066 2067 *path = '\0'; 2068 sa_hdl = hdl; 2069 2070 for (;;) { 2071 uint64_t pobj; 2072 char component[MAXNAMELEN + 2]; 2073 size_t complen; 2074 int is_xattrdir; 2075 2076 if (prevdb) 2077 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2078 2079 if ((error = zfs_obj_to_pobj(sa_hdl, sa_table, &pobj, 2080 &is_xattrdir)) != 0) 2081 break; 2082 2083 if (pobj == obj) { 2084 if (path[0] != '/') 2085 *--path = '/'; 2086 break; 2087 } 2088 2089 component[0] = '/'; 2090 if (is_xattrdir) { 2091 (void) sprintf(component + 1, "<xattrdir>"); 2092 } else { 2093 error = zap_value_search(osp, pobj, obj, 2094 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2095 if (error != 0) 2096 break; 2097 } 2098 2099 complen = strlen(component); 2100 path -= complen; 2101 ASSERT(path >= buf); 2102 bcopy(component, path, complen); 2103 obj = pobj; 2104 2105 if (sa_hdl != hdl) { 2106 prevhdl = sa_hdl; 2107 prevdb = sa_db; 2108 } 2109 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2110 if (error != 0) { 2111 sa_hdl = prevhdl; 2112 sa_db = prevdb; 2113 break; 2114 } 2115 } 2116 2117 if (sa_hdl != NULL && sa_hdl != hdl) { 2118 ASSERT(sa_db != NULL); 2119 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2120 } 2121 2122 if (error == 0) 2123 (void) memmove(buf, path, buf + len - path); 2124 2125 return (error); 2126} 2127 2128int 2129zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2130{ 2131 sa_attr_type_t *sa_table; 2132 sa_handle_t *hdl; 2133 dmu_buf_t *db; 2134 int error; 2135 2136 error = zfs_sa_setup(osp, &sa_table); 2137 if (error != 0) 2138 return (error); 2139 2140 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2141 if (error != 0) 2142 return (error); 2143 2144 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2145 2146 zfs_release_sa_handle(hdl, db, FTAG); 2147 return (error); 2148} 2149 2150int 2151zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2152 char *buf, int len) 2153{ 2154 char *path = buf + len - 1; 2155 sa_attr_type_t *sa_table; 2156 sa_handle_t *hdl; 2157 dmu_buf_t *db; 2158 int error; 2159 2160 *path = '\0'; 2161 2162 error = zfs_sa_setup(osp, &sa_table); 2163 if (error != 0) 2164 return (error); 2165 2166 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2167 if (error != 0) 2168 return (error); 2169 2170 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2171 if (error != 0) { 2172 zfs_release_sa_handle(hdl, db, FTAG); 2173 return (error); 2174 } 2175 2176 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2177 2178 zfs_release_sa_handle(hdl, db, FTAG); 2179 return (error); 2180} 2181