zfs_znode.c revision 263397
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 * Copyright (c) 2013 by Delphix. All rights reserved. 24 */ 25 26/* Portions Copyright 2007 Jeremy Teo */ 27/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 28 29#ifdef _KERNEL 30#include <sys/types.h> 31#include <sys/param.h> 32#include <sys/time.h> 33#include <sys/systm.h> 34#include <sys/sysmacros.h> 35#include <sys/resource.h> 36#include <sys/mntent.h> 37#include <sys/u8_textprep.h> 38#include <sys/dsl_dataset.h> 39#include <sys/vfs.h> 40#include <sys/vnode.h> 41#include <sys/file.h> 42#include <sys/kmem.h> 43#include <sys/errno.h> 44#include <sys/unistd.h> 45#include <sys/atomic.h> 46#include <sys/zfs_dir.h> 47#include <sys/zfs_acl.h> 48#include <sys/zfs_ioctl.h> 49#include <sys/zfs_rlock.h> 50#include <sys/zfs_fuid.h> 51#include <sys/dnode.h> 52#include <sys/fs/zfs.h> 53#include <sys/kidmap.h> 54#endif /* _KERNEL */ 55 56#include <sys/dmu.h> 57#include <sys/refcount.h> 58#include <sys/stat.h> 59#include <sys/zap.h> 60#include <sys/zfs_znode.h> 61#include <sys/sa.h> 62#include <sys/zfs_sa.h> 63#include <sys/zfs_stat.h> 64#include <sys/refcount.h> 65 66#include "zfs_prop.h" 67#include "zfs_comutil.h" 68 69/* Used by fstat(1). */ 70SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t), 71 "sizeof(znode_t)"); 72 73/* 74 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 75 * turned on when DEBUG is also defined. 76 */ 77#ifdef DEBUG 78#define ZNODE_STATS 79#endif /* DEBUG */ 80 81#ifdef ZNODE_STATS 82#define ZNODE_STAT_ADD(stat) ((stat)++) 83#else 84#define ZNODE_STAT_ADD(stat) /* nothing */ 85#endif /* ZNODE_STATS */ 86 87/* 88 * Functions needed for userland (ie: libzpool) are not put under 89 * #ifdef_KERNEL; the rest of the functions have dependencies 90 * (such as VFS logic) that will not compile easily in userland. 91 */ 92#ifdef _KERNEL 93/* 94 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 95 * be freed before it can be safely accessed. 96 */ 97krwlock_t zfsvfs_lock; 98 99static kmem_cache_t *znode_cache = NULL; 100 101/*ARGSUSED*/ 102static void 103znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 104{ 105 /* 106 * We should never drop all dbuf refs without first clearing 107 * the eviction callback. 108 */ 109 panic("evicting znode %p\n", user_ptr); 110} 111 112extern struct vop_vector zfs_vnodeops; 113extern struct vop_vector zfs_fifoops; 114extern struct vop_vector zfs_shareops; 115 116static int 117zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 118{ 119 znode_t *zp = buf; 120 121 POINTER_INVALIDATE(&zp->z_zfsvfs); 122 123 list_link_init(&zp->z_link_node); 124 125 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 126 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 127 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 128 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 129 130 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 131 avl_create(&zp->z_range_avl, zfs_range_compare, 132 sizeof (rl_t), offsetof(rl_t, r_node)); 133 134 zp->z_dirlocks = NULL; 135 zp->z_acl_cached = NULL; 136 zp->z_vnode = NULL; 137 zp->z_moved = 0; 138 return (0); 139} 140 141/*ARGSUSED*/ 142static void 143zfs_znode_cache_destructor(void *buf, void *arg) 144{ 145 znode_t *zp = buf; 146 147 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 148 ASSERT(ZTOV(zp) == NULL); 149 vn_free(ZTOV(zp)); 150 ASSERT(!list_link_active(&zp->z_link_node)); 151 mutex_destroy(&zp->z_lock); 152 rw_destroy(&zp->z_parent_lock); 153 rw_destroy(&zp->z_name_lock); 154 mutex_destroy(&zp->z_acl_lock); 155 avl_destroy(&zp->z_range_avl); 156 mutex_destroy(&zp->z_range_lock); 157 158 ASSERT(zp->z_dirlocks == NULL); 159 ASSERT(zp->z_acl_cached == NULL); 160} 161 162#ifdef ZNODE_STATS 163static struct { 164 uint64_t zms_zfsvfs_invalid; 165 uint64_t zms_zfsvfs_recheck1; 166 uint64_t zms_zfsvfs_unmounted; 167 uint64_t zms_zfsvfs_recheck2; 168 uint64_t zms_obj_held; 169 uint64_t zms_vnode_locked; 170 uint64_t zms_not_only_dnlc; 171} znode_move_stats; 172#endif /* ZNODE_STATS */ 173 174#ifdef sun 175static void 176zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 177{ 178 vnode_t *vp; 179 180 /* Copy fields. */ 181 nzp->z_zfsvfs = ozp->z_zfsvfs; 182 183 /* Swap vnodes. */ 184 vp = nzp->z_vnode; 185 nzp->z_vnode = ozp->z_vnode; 186 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 187 ZTOV(ozp)->v_data = ozp; 188 ZTOV(nzp)->v_data = nzp; 189 190 nzp->z_id = ozp->z_id; 191 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 192 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 193 nzp->z_unlinked = ozp->z_unlinked; 194 nzp->z_atime_dirty = ozp->z_atime_dirty; 195 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 196 nzp->z_blksz = ozp->z_blksz; 197 nzp->z_seq = ozp->z_seq; 198 nzp->z_mapcnt = ozp->z_mapcnt; 199 nzp->z_gen = ozp->z_gen; 200 nzp->z_sync_cnt = ozp->z_sync_cnt; 201 nzp->z_is_sa = ozp->z_is_sa; 202 nzp->z_sa_hdl = ozp->z_sa_hdl; 203 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 204 nzp->z_links = ozp->z_links; 205 nzp->z_size = ozp->z_size; 206 nzp->z_pflags = ozp->z_pflags; 207 nzp->z_uid = ozp->z_uid; 208 nzp->z_gid = ozp->z_gid; 209 nzp->z_mode = ozp->z_mode; 210 211 /* 212 * Since this is just an idle znode and kmem is already dealing with 213 * memory pressure, release any cached ACL. 214 */ 215 if (ozp->z_acl_cached) { 216 zfs_acl_free(ozp->z_acl_cached); 217 ozp->z_acl_cached = NULL; 218 } 219 220 sa_set_userp(nzp->z_sa_hdl, nzp); 221 222 /* 223 * Invalidate the original znode by clearing fields that provide a 224 * pointer back to the znode. Set the low bit of the vfs pointer to 225 * ensure that zfs_znode_move() recognizes the znode as invalid in any 226 * subsequent callback. 227 */ 228 ozp->z_sa_hdl = NULL; 229 POINTER_INVALIDATE(&ozp->z_zfsvfs); 230 231 /* 232 * Mark the znode. 233 */ 234 nzp->z_moved = 1; 235 ozp->z_moved = (uint8_t)-1; 236} 237 238/*ARGSUSED*/ 239static kmem_cbrc_t 240zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 241{ 242 znode_t *ozp = buf, *nzp = newbuf; 243 zfsvfs_t *zfsvfs; 244 vnode_t *vp; 245 246 /* 247 * The znode is on the file system's list of known znodes if the vfs 248 * pointer is valid. We set the low bit of the vfs pointer when freeing 249 * the znode to invalidate it, and the memory patterns written by kmem 250 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 251 * created znode sets the vfs pointer last of all to indicate that the 252 * znode is known and in a valid state to be moved by this function. 253 */ 254 zfsvfs = ozp->z_zfsvfs; 255 if (!POINTER_IS_VALID(zfsvfs)) { 256 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 257 return (KMEM_CBRC_DONT_KNOW); 258 } 259 260 /* 261 * Close a small window in which it's possible that the filesystem could 262 * be unmounted and freed, and zfsvfs, though valid in the previous 263 * statement, could point to unrelated memory by the time we try to 264 * prevent the filesystem from being unmounted. 265 */ 266 rw_enter(&zfsvfs_lock, RW_WRITER); 267 if (zfsvfs != ozp->z_zfsvfs) { 268 rw_exit(&zfsvfs_lock); 269 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 270 return (KMEM_CBRC_DONT_KNOW); 271 } 272 273 /* 274 * If the znode is still valid, then so is the file system. We know that 275 * no valid file system can be freed while we hold zfsvfs_lock, so we 276 * can safely ensure that the filesystem is not and will not be 277 * unmounted. The next statement is equivalent to ZFS_ENTER(). 278 */ 279 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 280 if (zfsvfs->z_unmounted) { 281 ZFS_EXIT(zfsvfs); 282 rw_exit(&zfsvfs_lock); 283 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 284 return (KMEM_CBRC_DONT_KNOW); 285 } 286 rw_exit(&zfsvfs_lock); 287 288 mutex_enter(&zfsvfs->z_znodes_lock); 289 /* 290 * Recheck the vfs pointer in case the znode was removed just before 291 * acquiring the lock. 292 */ 293 if (zfsvfs != ozp->z_zfsvfs) { 294 mutex_exit(&zfsvfs->z_znodes_lock); 295 ZFS_EXIT(zfsvfs); 296 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 297 return (KMEM_CBRC_DONT_KNOW); 298 } 299 300 /* 301 * At this point we know that as long as we hold z_znodes_lock, the 302 * znode cannot be freed and fields within the znode can be safely 303 * accessed. Now, prevent a race with zfs_zget(). 304 */ 305 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 306 mutex_exit(&zfsvfs->z_znodes_lock); 307 ZFS_EXIT(zfsvfs); 308 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 309 return (KMEM_CBRC_LATER); 310 } 311 312 vp = ZTOV(ozp); 313 if (mutex_tryenter(&vp->v_lock) == 0) { 314 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 315 mutex_exit(&zfsvfs->z_znodes_lock); 316 ZFS_EXIT(zfsvfs); 317 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 318 return (KMEM_CBRC_LATER); 319 } 320 321 /* Only move znodes that are referenced _only_ by the DNLC. */ 322 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 323 mutex_exit(&vp->v_lock); 324 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 325 mutex_exit(&zfsvfs->z_znodes_lock); 326 ZFS_EXIT(zfsvfs); 327 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 328 return (KMEM_CBRC_LATER); 329 } 330 331 /* 332 * The znode is known and in a valid state to move. We're holding the 333 * locks needed to execute the critical section. 334 */ 335 zfs_znode_move_impl(ozp, nzp); 336 mutex_exit(&vp->v_lock); 337 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 338 339 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 340 mutex_exit(&zfsvfs->z_znodes_lock); 341 ZFS_EXIT(zfsvfs); 342 343 return (KMEM_CBRC_YES); 344} 345#endif /* sun */ 346 347void 348zfs_znode_init(void) 349{ 350 /* 351 * Initialize zcache 352 */ 353 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 354 ASSERT(znode_cache == NULL); 355 znode_cache = kmem_cache_create("zfs_znode_cache", 356 sizeof (znode_t), 0, zfs_znode_cache_constructor, 357 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 358 kmem_cache_set_move(znode_cache, zfs_znode_move); 359} 360 361void 362zfs_znode_fini(void) 363{ 364#ifdef sun 365 /* 366 * Cleanup vfs & vnode ops 367 */ 368 zfs_remove_op_tables(); 369#endif /* sun */ 370 371 /* 372 * Cleanup zcache 373 */ 374 if (znode_cache) 375 kmem_cache_destroy(znode_cache); 376 znode_cache = NULL; 377 rw_destroy(&zfsvfs_lock); 378} 379 380#ifdef sun 381struct vnodeops *zfs_dvnodeops; 382struct vnodeops *zfs_fvnodeops; 383struct vnodeops *zfs_symvnodeops; 384struct vnodeops *zfs_xdvnodeops; 385struct vnodeops *zfs_evnodeops; 386struct vnodeops *zfs_sharevnodeops; 387 388void 389zfs_remove_op_tables() 390{ 391 /* 392 * Remove vfs ops 393 */ 394 ASSERT(zfsfstype); 395 (void) vfs_freevfsops_by_type(zfsfstype); 396 zfsfstype = 0; 397 398 /* 399 * Remove vnode ops 400 */ 401 if (zfs_dvnodeops) 402 vn_freevnodeops(zfs_dvnodeops); 403 if (zfs_fvnodeops) 404 vn_freevnodeops(zfs_fvnodeops); 405 if (zfs_symvnodeops) 406 vn_freevnodeops(zfs_symvnodeops); 407 if (zfs_xdvnodeops) 408 vn_freevnodeops(zfs_xdvnodeops); 409 if (zfs_evnodeops) 410 vn_freevnodeops(zfs_evnodeops); 411 if (zfs_sharevnodeops) 412 vn_freevnodeops(zfs_sharevnodeops); 413 414 zfs_dvnodeops = NULL; 415 zfs_fvnodeops = NULL; 416 zfs_symvnodeops = NULL; 417 zfs_xdvnodeops = NULL; 418 zfs_evnodeops = NULL; 419 zfs_sharevnodeops = NULL; 420} 421 422extern const fs_operation_def_t zfs_dvnodeops_template[]; 423extern const fs_operation_def_t zfs_fvnodeops_template[]; 424extern const fs_operation_def_t zfs_xdvnodeops_template[]; 425extern const fs_operation_def_t zfs_symvnodeops_template[]; 426extern const fs_operation_def_t zfs_evnodeops_template[]; 427extern const fs_operation_def_t zfs_sharevnodeops_template[]; 428 429int 430zfs_create_op_tables() 431{ 432 int error; 433 434 /* 435 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 436 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 437 * In this case we just return as the ops vectors are already set up. 438 */ 439 if (zfs_dvnodeops) 440 return (0); 441 442 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 443 &zfs_dvnodeops); 444 if (error) 445 return (error); 446 447 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 448 &zfs_fvnodeops); 449 if (error) 450 return (error); 451 452 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 453 &zfs_symvnodeops); 454 if (error) 455 return (error); 456 457 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 458 &zfs_xdvnodeops); 459 if (error) 460 return (error); 461 462 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 463 &zfs_evnodeops); 464 if (error) 465 return (error); 466 467 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 468 &zfs_sharevnodeops); 469 470 return (error); 471} 472#endif /* sun */ 473 474int 475zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 476{ 477 zfs_acl_ids_t acl_ids; 478 vattr_t vattr; 479 znode_t *sharezp; 480 znode_t *zp; 481 int error; 482 483 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 484 vattr.va_type = VDIR; 485 vattr.va_mode = S_IFDIR|0555; 486 vattr.va_uid = crgetuid(kcred); 487 vattr.va_gid = crgetgid(kcred); 488 489 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 490 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 491 sharezp->z_moved = 0; 492 sharezp->z_unlinked = 0; 493 sharezp->z_atime_dirty = 0; 494 sharezp->z_zfsvfs = zfsvfs; 495 sharezp->z_is_sa = zfsvfs->z_use_sa; 496 497 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 498 kcred, NULL, &acl_ids)); 499 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 500 ASSERT3P(zp, ==, sharezp); 501 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 502 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 503 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 504 zfsvfs->z_shares_dir = sharezp->z_id; 505 506 zfs_acl_ids_free(&acl_ids); 507 sa_handle_destroy(sharezp->z_sa_hdl); 508 kmem_cache_free(znode_cache, sharezp); 509 510 return (error); 511} 512 513/* 514 * define a couple of values we need available 515 * for both 64 and 32 bit environments. 516 */ 517#ifndef NBITSMINOR64 518#define NBITSMINOR64 32 519#endif 520#ifndef MAXMAJ64 521#define MAXMAJ64 0xffffffffUL 522#endif 523#ifndef MAXMIN64 524#define MAXMIN64 0xffffffffUL 525#endif 526 527/* 528 * Create special expldev for ZFS private use. 529 * Can't use standard expldev since it doesn't do 530 * what we want. The standard expldev() takes a 531 * dev32_t in LP64 and expands it to a long dev_t. 532 * We need an interface that takes a dev32_t in ILP32 533 * and expands it to a long dev_t. 534 */ 535static uint64_t 536zfs_expldev(dev_t dev) 537{ 538 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 539} 540/* 541 * Special cmpldev for ZFS private use. 542 * Can't use standard cmpldev since it takes 543 * a long dev_t and compresses it to dev32_t in 544 * LP64. We need to do a compaction of a long dev_t 545 * to a dev32_t in ILP32. 546 */ 547dev_t 548zfs_cmpldev(uint64_t dev) 549{ 550 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 551} 552 553static void 554zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 555 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 556{ 557 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 558 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 559 560 mutex_enter(&zp->z_lock); 561 562 ASSERT(zp->z_sa_hdl == NULL); 563 ASSERT(zp->z_acl_cached == NULL); 564 if (sa_hdl == NULL) { 565 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 566 SA_HDL_SHARED, &zp->z_sa_hdl)); 567 } else { 568 zp->z_sa_hdl = sa_hdl; 569 sa_set_userp(sa_hdl, zp); 570 } 571 572 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 573 574 /* 575 * Slap on VROOT if we are the root znode 576 */ 577 if (zp->z_id == zfsvfs->z_root) 578 ZTOV(zp)->v_flag |= VROOT; 579 580 mutex_exit(&zp->z_lock); 581 vn_exists(ZTOV(zp)); 582} 583 584void 585zfs_znode_dmu_fini(znode_t *zp) 586{ 587 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 588 zp->z_unlinked || 589 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 590 591 sa_handle_destroy(zp->z_sa_hdl); 592 zp->z_sa_hdl = NULL; 593} 594 595static void 596zfs_vnode_forget(vnode_t *vp) 597{ 598 599 /* copied from insmntque_stddtr */ 600 vp->v_data = NULL; 601 vp->v_op = &dead_vnodeops; 602 vgone(vp); 603 vput(vp); 604} 605 606/* 607 * Construct a new znode/vnode and intialize. 608 * 609 * This does not do a call to dmu_set_user() that is 610 * up to the caller to do, in case you don't want to 611 * return the znode 612 */ 613static znode_t * 614zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 615 dmu_object_type_t obj_type, sa_handle_t *hdl) 616{ 617 znode_t *zp; 618 vnode_t *vp; 619 uint64_t mode; 620 uint64_t parent; 621 sa_bulk_attr_t bulk[9]; 622 int count = 0; 623 int error; 624 625 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 626 627 KASSERT(curthread->td_vp_reserv > 0, 628 ("zfs_znode_alloc: getnewvnode without any vnodes reserved")); 629 error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp); 630 if (error != 0) { 631 kmem_cache_free(znode_cache, zp); 632 return (NULL); 633 } 634 zp->z_vnode = vp; 635 vp->v_data = zp; 636 637 ASSERT(zp->z_dirlocks == NULL); 638 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 639 zp->z_moved = 0; 640 641 /* 642 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 643 * the zfs_znode_move() callback. 644 */ 645 zp->z_sa_hdl = NULL; 646 zp->z_unlinked = 0; 647 zp->z_atime_dirty = 0; 648 zp->z_mapcnt = 0; 649 zp->z_id = db->db_object; 650 zp->z_blksz = blksz; 651 zp->z_seq = 0x7A4653; 652 zp->z_sync_cnt = 0; 653 654 vp = ZTOV(zp); 655 656 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 657 658 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 659 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 660 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 661 &zp->z_size, 8); 662 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 663 &zp->z_links, 8); 664 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 665 &zp->z_pflags, 8); 666 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 667 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 668 &zp->z_atime, 16); 669 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 670 &zp->z_uid, 8); 671 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 672 &zp->z_gid, 8); 673 674 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 675 if (hdl == NULL) 676 sa_handle_destroy(zp->z_sa_hdl); 677 zfs_vnode_forget(vp); 678 zp->z_vnode = NULL; 679 kmem_cache_free(znode_cache, zp); 680 return (NULL); 681 } 682 683 zp->z_mode = mode; 684 685 vp->v_type = IFTOVT((mode_t)mode); 686 687 switch (vp->v_type) { 688 case VDIR: 689 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 690 break; 691#ifdef sun 692 case VBLK: 693 case VCHR: 694 { 695 uint64_t rdev; 696 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 697 &rdev, sizeof (rdev)) == 0); 698 699 vp->v_rdev = zfs_cmpldev(rdev); 700 } 701 break; 702#endif /* sun */ 703 case VFIFO: 704#ifdef sun 705 case VSOCK: 706 case VDOOR: 707#endif /* sun */ 708 vp->v_op = &zfs_fifoops; 709 break; 710 case VREG: 711 if (parent == zfsvfs->z_shares_dir) { 712 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 713 vp->v_op = &zfs_shareops; 714 } 715 break; 716#ifdef sun 717 case VLNK: 718 vn_setops(vp, zfs_symvnodeops); 719 break; 720 default: 721 vn_setops(vp, zfs_evnodeops); 722 break; 723#endif /* sun */ 724 } 725 726 mutex_enter(&zfsvfs->z_znodes_lock); 727 list_insert_tail(&zfsvfs->z_all_znodes, zp); 728 membar_producer(); 729 /* 730 * Everything else must be valid before assigning z_zfsvfs makes the 731 * znode eligible for zfs_znode_move(). 732 */ 733 zp->z_zfsvfs = zfsvfs; 734 mutex_exit(&zfsvfs->z_znodes_lock); 735 736 /* 737 * Acquire vnode lock before making it available to the world. 738 */ 739 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 740 VN_LOCK_AREC(vp); 741 if (vp->v_type != VFIFO) 742 VN_LOCK_ASHARE(vp); 743 744 VFS_HOLD(zfsvfs->z_vfs); 745 return (zp); 746} 747 748static uint64_t empty_xattr; 749static uint64_t pad[4]; 750static zfs_acl_phys_t acl_phys; 751/* 752 * Create a new DMU object to hold a zfs znode. 753 * 754 * IN: dzp - parent directory for new znode 755 * vap - file attributes for new znode 756 * tx - dmu transaction id for zap operations 757 * cr - credentials of caller 758 * flag - flags: 759 * IS_ROOT_NODE - new object will be root 760 * IS_XATTR - new object is an attribute 761 * bonuslen - length of bonus buffer 762 * setaclp - File/Dir initial ACL 763 * fuidp - Tracks fuid allocation. 764 * 765 * OUT: zpp - allocated znode 766 * 767 */ 768void 769zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 770 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 771{ 772 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 773 uint64_t mode, size, links, parent, pflags; 774 uint64_t dzp_pflags = 0; 775 uint64_t rdev = 0; 776 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 777 dmu_buf_t *db; 778 timestruc_t now; 779 uint64_t gen, obj; 780 int err; 781 int bonuslen; 782 sa_handle_t *sa_hdl; 783 dmu_object_type_t obj_type; 784 sa_bulk_attr_t sa_attrs[ZPL_END]; 785 int cnt = 0; 786 zfs_acl_locator_cb_t locate = { 0 }; 787 788 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 789 790 if (zfsvfs->z_replay) { 791 obj = vap->va_nodeid; 792 now = vap->va_ctime; /* see zfs_replay_create() */ 793 gen = vap->va_nblocks; /* ditto */ 794 } else { 795 obj = 0; 796 gethrestime(&now); 797 gen = dmu_tx_get_txg(tx); 798 } 799 800 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 801 bonuslen = (obj_type == DMU_OT_SA) ? 802 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 803 804 /* 805 * Create a new DMU object. 806 */ 807 /* 808 * There's currently no mechanism for pre-reading the blocks that will 809 * be needed to allocate a new object, so we accept the small chance 810 * that there will be an i/o error and we will fail one of the 811 * assertions below. 812 */ 813 if (vap->va_type == VDIR) { 814 if (zfsvfs->z_replay) { 815 VERIFY0(zap_create_claim_norm(zfsvfs->z_os, obj, 816 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 817 obj_type, bonuslen, tx)); 818 } else { 819 obj = zap_create_norm(zfsvfs->z_os, 820 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 821 obj_type, bonuslen, tx); 822 } 823 } else { 824 if (zfsvfs->z_replay) { 825 VERIFY0(dmu_object_claim(zfsvfs->z_os, obj, 826 DMU_OT_PLAIN_FILE_CONTENTS, 0, 827 obj_type, bonuslen, tx)); 828 } else { 829 obj = dmu_object_alloc(zfsvfs->z_os, 830 DMU_OT_PLAIN_FILE_CONTENTS, 0, 831 obj_type, bonuslen, tx); 832 } 833 } 834 835 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 836 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 837 838 /* 839 * If this is the root, fix up the half-initialized parent pointer 840 * to reference the just-allocated physical data area. 841 */ 842 if (flag & IS_ROOT_NODE) { 843 dzp->z_id = obj; 844 } else { 845 dzp_pflags = dzp->z_pflags; 846 } 847 848 /* 849 * If parent is an xattr, so am I. 850 */ 851 if (dzp_pflags & ZFS_XATTR) { 852 flag |= IS_XATTR; 853 } 854 855 if (zfsvfs->z_use_fuids) 856 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 857 else 858 pflags = 0; 859 860 if (vap->va_type == VDIR) { 861 size = 2; /* contents ("." and "..") */ 862 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 863 } else { 864 size = links = 0; 865 } 866 867 if (vap->va_type == VBLK || vap->va_type == VCHR) { 868 rdev = zfs_expldev(vap->va_rdev); 869 } 870 871 parent = dzp->z_id; 872 mode = acl_ids->z_mode; 873 if (flag & IS_XATTR) 874 pflags |= ZFS_XATTR; 875 876 /* 877 * No execs denied will be deterimed when zfs_mode_compute() is called. 878 */ 879 pflags |= acl_ids->z_aclp->z_hints & 880 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 881 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 882 883 ZFS_TIME_ENCODE(&now, crtime); 884 ZFS_TIME_ENCODE(&now, ctime); 885 886 if (vap->va_mask & AT_ATIME) { 887 ZFS_TIME_ENCODE(&vap->va_atime, atime); 888 } else { 889 ZFS_TIME_ENCODE(&now, atime); 890 } 891 892 if (vap->va_mask & AT_MTIME) { 893 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 894 } else { 895 ZFS_TIME_ENCODE(&now, mtime); 896 } 897 898 /* Now add in all of the "SA" attributes */ 899 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 900 &sa_hdl)); 901 902 /* 903 * Setup the array of attributes to be replaced/set on the new file 904 * 905 * order for DMU_OT_ZNODE is critical since it needs to be constructed 906 * in the old znode_phys_t format. Don't change this ordering 907 */ 908 909 if (obj_type == DMU_OT_ZNODE) { 910 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 911 NULL, &atime, 16); 912 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 913 NULL, &mtime, 16); 914 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 915 NULL, &ctime, 16); 916 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 917 NULL, &crtime, 16); 918 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 919 NULL, &gen, 8); 920 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 921 NULL, &mode, 8); 922 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 923 NULL, &size, 8); 924 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 925 NULL, &parent, 8); 926 } else { 927 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 928 NULL, &mode, 8); 929 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 930 NULL, &size, 8); 931 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 932 NULL, &gen, 8); 933 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 934 &acl_ids->z_fuid, 8); 935 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 936 &acl_ids->z_fgid, 8); 937 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 938 NULL, &parent, 8); 939 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 940 NULL, &pflags, 8); 941 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 942 NULL, &atime, 16); 943 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 944 NULL, &mtime, 16); 945 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 946 NULL, &ctime, 16); 947 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 948 NULL, &crtime, 16); 949 } 950 951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 952 953 if (obj_type == DMU_OT_ZNODE) { 954 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 955 &empty_xattr, 8); 956 } 957 if (obj_type == DMU_OT_ZNODE || 958 (vap->va_type == VBLK || vap->va_type == VCHR)) { 959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 960 NULL, &rdev, 8); 961 962 } 963 if (obj_type == DMU_OT_ZNODE) { 964 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 965 NULL, &pflags, 8); 966 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 967 &acl_ids->z_fuid, 8); 968 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 969 &acl_ids->z_fgid, 8); 970 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 971 sizeof (uint64_t) * 4); 972 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 973 &acl_phys, sizeof (zfs_acl_phys_t)); 974 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 975 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 976 &acl_ids->z_aclp->z_acl_count, 8); 977 locate.cb_aclp = acl_ids->z_aclp; 978 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 979 zfs_acl_data_locator, &locate, 980 acl_ids->z_aclp->z_acl_bytes); 981 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 982 acl_ids->z_fuid, acl_ids->z_fgid); 983 } 984 985 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 986 987 if (!(flag & IS_ROOT_NODE)) { 988 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 989 ASSERT(*zpp != NULL); 990 } else { 991 /* 992 * If we are creating the root node, the "parent" we 993 * passed in is the znode for the root. 994 */ 995 *zpp = dzp; 996 997 (*zpp)->z_sa_hdl = sa_hdl; 998 } 999 1000 (*zpp)->z_pflags = pflags; 1001 (*zpp)->z_mode = mode; 1002 1003 if (vap->va_mask & AT_XVATTR) 1004 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 1005 1006 if (obj_type == DMU_OT_ZNODE || 1007 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1008 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx)); 1009 } 1010 if (!(flag & IS_ROOT_NODE)) { 1011 vnode_t *vp; 1012 1013 vp = ZTOV(*zpp); 1014 vp->v_vflag |= VV_FORCEINSMQ; 1015 err = insmntque(vp, zfsvfs->z_vfs); 1016 vp->v_vflag &= ~VV_FORCEINSMQ; 1017 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1018 } 1019 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1020} 1021 1022/* 1023 * Update in-core attributes. It is assumed the caller will be doing an 1024 * sa_bulk_update to push the changes out. 1025 */ 1026void 1027zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1028{ 1029 xoptattr_t *xoap; 1030 1031 xoap = xva_getxoptattr(xvap); 1032 ASSERT(xoap); 1033 1034 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1035 uint64_t times[2]; 1036 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1037 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1038 ×, sizeof (times), tx); 1039 XVA_SET_RTN(xvap, XAT_CREATETIME); 1040 } 1041 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1042 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1043 zp->z_pflags, tx); 1044 XVA_SET_RTN(xvap, XAT_READONLY); 1045 } 1046 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1047 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1048 zp->z_pflags, tx); 1049 XVA_SET_RTN(xvap, XAT_HIDDEN); 1050 } 1051 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1052 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1053 zp->z_pflags, tx); 1054 XVA_SET_RTN(xvap, XAT_SYSTEM); 1055 } 1056 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1057 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1058 zp->z_pflags, tx); 1059 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1060 } 1061 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1062 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1063 zp->z_pflags, tx); 1064 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1065 } 1066 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1067 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1068 zp->z_pflags, tx); 1069 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1070 } 1071 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1072 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1073 zp->z_pflags, tx); 1074 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1075 } 1076 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1077 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1078 zp->z_pflags, tx); 1079 XVA_SET_RTN(xvap, XAT_NODUMP); 1080 } 1081 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1082 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1083 zp->z_pflags, tx); 1084 XVA_SET_RTN(xvap, XAT_OPAQUE); 1085 } 1086 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1087 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1088 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1089 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1090 } 1091 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1092 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1093 zp->z_pflags, tx); 1094 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1095 } 1096 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1097 zfs_sa_set_scanstamp(zp, xvap, tx); 1098 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1099 } 1100 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1101 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1102 zp->z_pflags, tx); 1103 XVA_SET_RTN(xvap, XAT_REPARSE); 1104 } 1105 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1106 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1107 zp->z_pflags, tx); 1108 XVA_SET_RTN(xvap, XAT_OFFLINE); 1109 } 1110 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1111 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1112 zp->z_pflags, tx); 1113 XVA_SET_RTN(xvap, XAT_SPARSE); 1114 } 1115} 1116 1117int 1118zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1119{ 1120 dmu_object_info_t doi; 1121 dmu_buf_t *db; 1122 znode_t *zp; 1123 vnode_t *vp; 1124 sa_handle_t *hdl; 1125 struct thread *td; 1126 int locked; 1127 int err; 1128 1129 td = curthread; 1130 getnewvnode_reserve(1); 1131again: 1132 *zpp = NULL; 1133 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1134 1135 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1136 if (err) { 1137 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1138 getnewvnode_drop_reserve(); 1139 return (err); 1140 } 1141 1142 dmu_object_info_from_db(db, &doi); 1143 if (doi.doi_bonus_type != DMU_OT_SA && 1144 (doi.doi_bonus_type != DMU_OT_ZNODE || 1145 (doi.doi_bonus_type == DMU_OT_ZNODE && 1146 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1147 sa_buf_rele(db, NULL); 1148 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1149#ifdef __FreeBSD__ 1150 getnewvnode_drop_reserve(); 1151#endif 1152 return (SET_ERROR(EINVAL)); 1153 } 1154 1155 hdl = dmu_buf_get_user(db); 1156 if (hdl != NULL) { 1157 zp = sa_get_userdata(hdl); 1158 1159 1160 /* 1161 * Since "SA" does immediate eviction we 1162 * should never find a sa handle that doesn't 1163 * know about the znode. 1164 */ 1165 1166 ASSERT3P(zp, !=, NULL); 1167 1168 mutex_enter(&zp->z_lock); 1169 ASSERT3U(zp->z_id, ==, obj_num); 1170 if (zp->z_unlinked) { 1171 err = SET_ERROR(ENOENT); 1172 } else { 1173 vp = ZTOV(zp); 1174 *zpp = zp; 1175 err = 0; 1176 } 1177 sa_buf_rele(db, NULL); 1178 1179 /* Don't let the vnode disappear after ZFS_OBJ_HOLD_EXIT. */ 1180 if (err == 0) 1181 VN_HOLD(vp); 1182 1183 mutex_exit(&zp->z_lock); 1184 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1185 1186 if (err == 0) { 1187 locked = VOP_ISLOCKED(vp); 1188 VI_LOCK(vp); 1189 if ((vp->v_iflag & VI_DOOMED) != 0 && 1190 locked != LK_EXCLUSIVE) { 1191 /* 1192 * The vnode is doomed and this thread doesn't 1193 * hold the exclusive lock on it, so the vnode 1194 * must be being reclaimed by another thread. 1195 * Otherwise the doomed vnode is being reclaimed 1196 * by this thread and zfs_zget is called from 1197 * ZIL internals. 1198 */ 1199 VI_UNLOCK(vp); 1200 VN_RELE(vp); 1201 goto again; 1202 } 1203 VI_UNLOCK(vp); 1204 } 1205 getnewvnode_drop_reserve(); 1206 return (err); 1207 } 1208 1209 /* 1210 * Not found create new znode/vnode 1211 * but only if file exists. 1212 * 1213 * There is a small window where zfs_vget() could 1214 * find this object while a file create is still in 1215 * progress. This is checked for in zfs_znode_alloc() 1216 * 1217 * if zfs_znode_alloc() fails it will drop the hold on the 1218 * bonus buffer. 1219 */ 1220 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1221 doi.doi_bonus_type, NULL); 1222 if (zp == NULL) { 1223 err = SET_ERROR(ENOENT); 1224 } else { 1225 *zpp = zp; 1226 } 1227 if (err == 0) { 1228 vnode_t *vp = ZTOV(zp); 1229 1230 err = insmntque(vp, zfsvfs->z_vfs); 1231 if (err == 0) 1232 VOP_UNLOCK(vp, 0); 1233 else { 1234 zp->z_vnode = NULL; 1235 zfs_znode_dmu_fini(zp); 1236 zfs_znode_free(zp); 1237 *zpp = NULL; 1238 } 1239 } 1240 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1241 getnewvnode_drop_reserve(); 1242 return (err); 1243} 1244 1245int 1246zfs_rezget(znode_t *zp) 1247{ 1248 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1249 dmu_object_info_t doi; 1250 dmu_buf_t *db; 1251 vnode_t *vp; 1252 uint64_t obj_num = zp->z_id; 1253 uint64_t mode, size; 1254 sa_bulk_attr_t bulk[8]; 1255 int err; 1256 int count = 0; 1257 uint64_t gen; 1258 1259 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1260 1261 mutex_enter(&zp->z_acl_lock); 1262 if (zp->z_acl_cached) { 1263 zfs_acl_free(zp->z_acl_cached); 1264 zp->z_acl_cached = NULL; 1265 } 1266 1267 mutex_exit(&zp->z_acl_lock); 1268 ASSERT(zp->z_sa_hdl == NULL); 1269 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1270 if (err) { 1271 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1272 return (err); 1273 } 1274 1275 dmu_object_info_from_db(db, &doi); 1276 if (doi.doi_bonus_type != DMU_OT_SA && 1277 (doi.doi_bonus_type != DMU_OT_ZNODE || 1278 (doi.doi_bonus_type == DMU_OT_ZNODE && 1279 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1280 sa_buf_rele(db, NULL); 1281 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1282 return (SET_ERROR(EINVAL)); 1283 } 1284 1285 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1286 size = zp->z_size; 1287 1288 /* reload cached values */ 1289 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1290 &gen, sizeof (gen)); 1291 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1292 &zp->z_size, sizeof (zp->z_size)); 1293 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1294 &zp->z_links, sizeof (zp->z_links)); 1295 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1296 &zp->z_pflags, sizeof (zp->z_pflags)); 1297 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1298 &zp->z_atime, sizeof (zp->z_atime)); 1299 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1300 &zp->z_uid, sizeof (zp->z_uid)); 1301 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1302 &zp->z_gid, sizeof (zp->z_gid)); 1303 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1304 &mode, sizeof (mode)); 1305 1306 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1307 zfs_znode_dmu_fini(zp); 1308 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1309 return (SET_ERROR(EIO)); 1310 } 1311 1312 zp->z_mode = mode; 1313 1314 if (gen != zp->z_gen) { 1315 zfs_znode_dmu_fini(zp); 1316 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1317 return (SET_ERROR(EIO)); 1318 } 1319 1320 /* 1321 * XXXPJD: Not sure how is that possible, but under heavy 1322 * zfs recv -F load it happens that z_gen is the same, but 1323 * vnode type is different than znode type. This would mean 1324 * that for example regular file was replaced with directory 1325 * which has the same object number. 1326 */ 1327 vp = ZTOV(zp); 1328 if (vp != NULL && 1329 vp->v_type != IFTOVT((mode_t)zp->z_mode)) { 1330 zfs_znode_dmu_fini(zp); 1331 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1332 return (EIO); 1333 } 1334 1335 zp->z_unlinked = (zp->z_links == 0); 1336 zp->z_blksz = doi.doi_data_block_size; 1337 if (vp != NULL) { 1338 vn_pages_remove(vp, 0, 0); 1339 if (zp->z_size != size) 1340 vnode_pager_setsize(vp, zp->z_size); 1341 } 1342 1343 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1344 1345 return (0); 1346} 1347 1348void 1349zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1350{ 1351 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1352 objset_t *os = zfsvfs->z_os; 1353 uint64_t obj = zp->z_id; 1354 uint64_t acl_obj = zfs_external_acl(zp); 1355 1356 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1357 if (acl_obj) { 1358 VERIFY(!zp->z_is_sa); 1359 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1360 } 1361 VERIFY(0 == dmu_object_free(os, obj, tx)); 1362 zfs_znode_dmu_fini(zp); 1363 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1364 zfs_znode_free(zp); 1365} 1366 1367void 1368zfs_zinactive(znode_t *zp) 1369{ 1370 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1371 uint64_t z_id = zp->z_id; 1372 1373 ASSERT(zp->z_sa_hdl); 1374 1375 /* 1376 * Don't allow a zfs_zget() while were trying to release this znode 1377 */ 1378 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1379 1380 mutex_enter(&zp->z_lock); 1381 1382 /* 1383 * If this was the last reference to a file with no links, 1384 * remove the file from the file system. 1385 */ 1386 if (zp->z_unlinked) { 1387 mutex_exit(&zp->z_lock); 1388 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1389 zfs_rmnode(zp); 1390 return; 1391 } 1392 1393 mutex_exit(&zp->z_lock); 1394 zfs_znode_dmu_fini(zp); 1395 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1396 zfs_znode_free(zp); 1397} 1398 1399void 1400zfs_znode_free(znode_t *zp) 1401{ 1402 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1403 1404 ASSERT(zp->z_sa_hdl == NULL); 1405 zp->z_vnode = NULL; 1406 mutex_enter(&zfsvfs->z_znodes_lock); 1407 POINTER_INVALIDATE(&zp->z_zfsvfs); 1408 list_remove(&zfsvfs->z_all_znodes, zp); 1409 mutex_exit(&zfsvfs->z_znodes_lock); 1410 1411 if (zp->z_acl_cached) { 1412 zfs_acl_free(zp->z_acl_cached); 1413 zp->z_acl_cached = NULL; 1414 } 1415 1416 kmem_cache_free(znode_cache, zp); 1417 1418 VFS_RELE(zfsvfs->z_vfs); 1419} 1420 1421void 1422zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1423 uint64_t ctime[2], boolean_t have_tx) 1424{ 1425 timestruc_t now; 1426 1427 gethrestime(&now); 1428 1429 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1430 zp->z_atime_dirty = 0; 1431 zp->z_seq++; 1432 } else { 1433 zp->z_atime_dirty = 1; 1434 } 1435 1436 if (flag & AT_ATIME) { 1437 ZFS_TIME_ENCODE(&now, zp->z_atime); 1438 } 1439 1440 if (flag & AT_MTIME) { 1441 ZFS_TIME_ENCODE(&now, mtime); 1442 if (zp->z_zfsvfs->z_use_fuids) { 1443 zp->z_pflags |= (ZFS_ARCHIVE | 1444 ZFS_AV_MODIFIED); 1445 } 1446 } 1447 1448 if (flag & AT_CTIME) { 1449 ZFS_TIME_ENCODE(&now, ctime); 1450 if (zp->z_zfsvfs->z_use_fuids) 1451 zp->z_pflags |= ZFS_ARCHIVE; 1452 } 1453} 1454 1455/* 1456 * Grow the block size for a file. 1457 * 1458 * IN: zp - znode of file to free data in. 1459 * size - requested block size 1460 * tx - open transaction. 1461 * 1462 * NOTE: this function assumes that the znode is write locked. 1463 */ 1464void 1465zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1466{ 1467 int error; 1468 u_longlong_t dummy; 1469 1470 if (size <= zp->z_blksz) 1471 return; 1472 /* 1473 * If the file size is already greater than the current blocksize, 1474 * we will not grow. If there is more than one block in a file, 1475 * the blocksize cannot change. 1476 */ 1477 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1478 return; 1479 1480 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1481 size, 0, tx); 1482 1483 if (error == ENOTSUP) 1484 return; 1485 ASSERT0(error); 1486 1487 /* What blocksize did we actually get? */ 1488 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1489} 1490 1491#ifdef sun 1492/* 1493 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1494 * be calling back into the fs for a putpage(). E.g.: when truncating 1495 * a file, the pages being "thrown away* don't need to be written out. 1496 */ 1497/* ARGSUSED */ 1498static int 1499zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1500 int flags, cred_t *cr) 1501{ 1502 ASSERT(0); 1503 return (0); 1504} 1505#endif /* sun */ 1506 1507/* 1508 * Increase the file length 1509 * 1510 * IN: zp - znode of file to free data in. 1511 * end - new end-of-file 1512 * 1513 * RETURN: 0 on success, error code on 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 } 1536 tx = dmu_tx_create(zfsvfs->z_os); 1537 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1538 zfs_sa_upgrade_txholds(tx, zp); 1539 if (end > zp->z_blksz && 1540 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1541 /* 1542 * We are growing the file past the current block size. 1543 */ 1544 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1545 ASSERT(!ISP2(zp->z_blksz)); 1546 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1547 } else { 1548 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1549 } 1550 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1551 } else { 1552 newblksz = 0; 1553 } 1554 1555 error = dmu_tx_assign(tx, TXG_WAIT); 1556 if (error) { 1557 dmu_tx_abort(tx); 1558 zfs_range_unlock(rl); 1559 return (error); 1560 } 1561 1562 if (newblksz) 1563 zfs_grow_blocksize(zp, newblksz, tx); 1564 1565 zp->z_size = end; 1566 1567 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1568 &zp->z_size, sizeof (zp->z_size), tx)); 1569 1570 vnode_pager_setsize(ZTOV(zp), end); 1571 1572 zfs_range_unlock(rl); 1573 1574 dmu_tx_commit(tx); 1575 1576 return (0); 1577} 1578 1579/* 1580 * Free space in a file. 1581 * 1582 * IN: zp - znode of file to free data in. 1583 * off - start of section to free. 1584 * len - length of section to free. 1585 * 1586 * RETURN: 0 on success, error code on failure 1587 */ 1588static int 1589zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1590{ 1591 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1592 rl_t *rl; 1593 int error; 1594 1595 /* 1596 * Lock the range being freed. 1597 */ 1598 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1599 1600 /* 1601 * Nothing to do if file already at desired length. 1602 */ 1603 if (off >= zp->z_size) { 1604 zfs_range_unlock(rl); 1605 return (0); 1606 } 1607 1608 if (off + len > zp->z_size) 1609 len = zp->z_size - off; 1610 1611 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1612 1613 if (error == 0) { 1614 /* 1615 * In FreeBSD we cannot free block in the middle of a file, 1616 * but only at the end of a file, so this code path should 1617 * never happen. 1618 */ 1619 vnode_pager_setsize(ZTOV(zp), off); 1620 } 1621 1622 zfs_range_unlock(rl); 1623 1624 return (error); 1625} 1626 1627/* 1628 * Truncate a file 1629 * 1630 * IN: zp - znode of file to free data in. 1631 * end - new end-of-file. 1632 * 1633 * RETURN: 0 on success, error code on failure 1634 */ 1635static int 1636zfs_trunc(znode_t *zp, uint64_t end) 1637{ 1638 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1639 vnode_t *vp = ZTOV(zp); 1640 dmu_tx_t *tx; 1641 rl_t *rl; 1642 int error; 1643 sa_bulk_attr_t bulk[2]; 1644 int count = 0; 1645 1646 /* 1647 * We will change zp_size, lock the whole file. 1648 */ 1649 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1650 1651 /* 1652 * Nothing to do if file already at desired length. 1653 */ 1654 if (end >= zp->z_size) { 1655 zfs_range_unlock(rl); 1656 return (0); 1657 } 1658 1659 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1660 if (error) { 1661 zfs_range_unlock(rl); 1662 return (error); 1663 } 1664 tx = dmu_tx_create(zfsvfs->z_os); 1665 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1666 zfs_sa_upgrade_txholds(tx, zp); 1667 error = dmu_tx_assign(tx, TXG_WAIT); 1668 if (error) { 1669 dmu_tx_abort(tx); 1670 zfs_range_unlock(rl); 1671 return (error); 1672 } 1673 1674 zp->z_size = end; 1675 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1676 NULL, &zp->z_size, sizeof (zp->z_size)); 1677 1678 if (end == 0) { 1679 zp->z_pflags &= ~ZFS_SPARSE; 1680 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1681 NULL, &zp->z_pflags, 8); 1682 } 1683 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1684 1685 dmu_tx_commit(tx); 1686 1687 /* 1688 * Clear any mapped pages in the truncated region. This has to 1689 * happen outside of the transaction to avoid the possibility of 1690 * a deadlock with someone trying to push a page that we are 1691 * about to invalidate. 1692 */ 1693 vnode_pager_setsize(vp, end); 1694 1695 zfs_range_unlock(rl); 1696 1697 return (0); 1698} 1699 1700/* 1701 * Free space in a file 1702 * 1703 * IN: zp - znode of file to free data in. 1704 * off - start of range 1705 * len - end of range (0 => EOF) 1706 * flag - current file open mode flags. 1707 * log - TRUE if this action should be logged 1708 * 1709 * RETURN: 0 on success, error code on failure 1710 */ 1711int 1712zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1713{ 1714 vnode_t *vp = ZTOV(zp); 1715 dmu_tx_t *tx; 1716 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1717 zilog_t *zilog = zfsvfs->z_log; 1718 uint64_t mode; 1719 uint64_t mtime[2], ctime[2]; 1720 sa_bulk_attr_t bulk[3]; 1721 int count = 0; 1722 int error; 1723 1724 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1725 sizeof (mode))) != 0) 1726 return (error); 1727 1728 if (off > zp->z_size) { 1729 error = zfs_extend(zp, off+len); 1730 if (error == 0 && log) 1731 goto log; 1732 else 1733 return (error); 1734 } 1735 1736 /* 1737 * Check for any locks in the region to be freed. 1738 */ 1739 1740 if (MANDLOCK(vp, (mode_t)mode)) { 1741 uint64_t length = (len ? len : zp->z_size - off); 1742 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1743 return (error); 1744 } 1745 1746 if (len == 0) { 1747 error = zfs_trunc(zp, off); 1748 } else { 1749 if ((error = zfs_free_range(zp, off, len)) == 0 && 1750 off + len > zp->z_size) 1751 error = zfs_extend(zp, off+len); 1752 } 1753 if (error || !log) 1754 return (error); 1755log: 1756 tx = dmu_tx_create(zfsvfs->z_os); 1757 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1758 zfs_sa_upgrade_txholds(tx, zp); 1759 error = dmu_tx_assign(tx, TXG_WAIT); 1760 if (error) { 1761 dmu_tx_abort(tx); 1762 return (error); 1763 } 1764 1765 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1766 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1767 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1768 NULL, &zp->z_pflags, 8); 1769 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1770 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1771 ASSERT(error == 0); 1772 1773 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1774 1775 dmu_tx_commit(tx); 1776 return (0); 1777} 1778 1779void 1780zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1781{ 1782 zfsvfs_t zfsvfs; 1783 uint64_t moid, obj, sa_obj, version; 1784 uint64_t sense = ZFS_CASE_SENSITIVE; 1785 uint64_t norm = 0; 1786 nvpair_t *elem; 1787 int error; 1788 int i; 1789 znode_t *rootzp = NULL; 1790 vattr_t vattr; 1791 znode_t *zp; 1792 zfs_acl_ids_t acl_ids; 1793 1794 /* 1795 * First attempt to create master node. 1796 */ 1797 /* 1798 * In an empty objset, there are no blocks to read and thus 1799 * there can be no i/o errors (which we assert below). 1800 */ 1801 moid = MASTER_NODE_OBJ; 1802 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1803 DMU_OT_NONE, 0, tx); 1804 ASSERT(error == 0); 1805 1806 /* 1807 * Set starting attributes. 1808 */ 1809 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1810 elem = NULL; 1811 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1812 /* For the moment we expect all zpl props to be uint64_ts */ 1813 uint64_t val; 1814 char *name; 1815 1816 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1817 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1818 name = nvpair_name(elem); 1819 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1820 if (val < version) 1821 version = val; 1822 } else { 1823 error = zap_update(os, moid, name, 8, 1, &val, tx); 1824 } 1825 ASSERT(error == 0); 1826 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1827 norm = val; 1828 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1829 sense = val; 1830 } 1831 ASSERT(version != 0); 1832 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1833 1834 /* 1835 * Create zap object used for SA attribute registration 1836 */ 1837 1838 if (version >= ZPL_VERSION_SA) { 1839 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1840 DMU_OT_NONE, 0, tx); 1841 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1842 ASSERT(error == 0); 1843 } else { 1844 sa_obj = 0; 1845 } 1846 /* 1847 * Create a delete queue. 1848 */ 1849 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1850 1851 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1852 ASSERT(error == 0); 1853 1854 /* 1855 * Create root znode. Create minimal znode/vnode/zfsvfs 1856 * to allow zfs_mknode to work. 1857 */ 1858 VATTR_NULL(&vattr); 1859 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1860 vattr.va_type = VDIR; 1861 vattr.va_mode = S_IFDIR|0755; 1862 vattr.va_uid = crgetuid(cr); 1863 vattr.va_gid = crgetgid(cr); 1864 1865 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1866 1867 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1868 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1869 rootzp->z_moved = 0; 1870 rootzp->z_unlinked = 0; 1871 rootzp->z_atime_dirty = 0; 1872 rootzp->z_is_sa = USE_SA(version, os); 1873 1874 zfsvfs.z_os = os; 1875 zfsvfs.z_parent = &zfsvfs; 1876 zfsvfs.z_version = version; 1877 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1878 zfsvfs.z_use_sa = USE_SA(version, os); 1879 zfsvfs.z_norm = norm; 1880 1881 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1882 &zfsvfs.z_attr_table); 1883 1884 ASSERT(error == 0); 1885 1886 /* 1887 * Fold case on file systems that are always or sometimes case 1888 * insensitive. 1889 */ 1890 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1891 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1892 1893 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1894 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1895 offsetof(znode_t, z_link_node)); 1896 1897 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1898 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1899 1900 rootzp->z_zfsvfs = &zfsvfs; 1901 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1902 cr, NULL, &acl_ids)); 1903 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1904 ASSERT3P(zp, ==, rootzp); 1905 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1906 ASSERT(error == 0); 1907 zfs_acl_ids_free(&acl_ids); 1908 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1909 1910 sa_handle_destroy(rootzp->z_sa_hdl); 1911 kmem_cache_free(znode_cache, rootzp); 1912 1913 /* 1914 * Create shares directory 1915 */ 1916 1917 error = zfs_create_share_dir(&zfsvfs, tx); 1918 1919 ASSERT(error == 0); 1920 1921 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1922 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1923} 1924 1925#endif /* _KERNEL */ 1926 1927static int 1928zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1929{ 1930 uint64_t sa_obj = 0; 1931 int error; 1932 1933 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1934 if (error != 0 && error != ENOENT) 1935 return (error); 1936 1937 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1938 return (error); 1939} 1940 1941static int 1942zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1943 dmu_buf_t **db, void *tag) 1944{ 1945 dmu_object_info_t doi; 1946 int error; 1947 1948 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 1949 return (error); 1950 1951 dmu_object_info_from_db(*db, &doi); 1952 if ((doi.doi_bonus_type != DMU_OT_SA && 1953 doi.doi_bonus_type != DMU_OT_ZNODE) || 1954 doi.doi_bonus_type == DMU_OT_ZNODE && 1955 doi.doi_bonus_size < sizeof (znode_phys_t)) { 1956 sa_buf_rele(*db, tag); 1957 return (SET_ERROR(ENOTSUP)); 1958 } 1959 1960 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 1961 if (error != 0) { 1962 sa_buf_rele(*db, tag); 1963 return (error); 1964 } 1965 1966 return (0); 1967} 1968 1969void 1970zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 1971{ 1972 sa_handle_destroy(hdl); 1973 sa_buf_rele(db, tag); 1974} 1975 1976/* 1977 * Given an object number, return its parent object number and whether 1978 * or not the object is an extended attribute directory. 1979 */ 1980static int 1981zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table, 1982 uint64_t *pobjp, int *is_xattrdir) 1983{ 1984 uint64_t parent; 1985 uint64_t pflags; 1986 uint64_t mode; 1987 uint64_t parent_mode; 1988 sa_bulk_attr_t bulk[3]; 1989 sa_handle_t *sa_hdl; 1990 dmu_buf_t *sa_db; 1991 int count = 0; 1992 int error; 1993 1994 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 1995 &parent, sizeof (parent)); 1996 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 1997 &pflags, sizeof (pflags)); 1998 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 1999 &mode, sizeof (mode)); 2000 2001 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2002 return (error); 2003 2004 /* 2005 * When a link is removed its parent pointer is not changed and will 2006 * be invalid. There are two cases where a link is removed but the 2007 * file stays around, when it goes to the delete queue and when there 2008 * are additional links. 2009 */ 2010 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG); 2011 if (error != 0) 2012 return (error); 2013 2014 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode)); 2015 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2016 if (error != 0) 2017 return (error); 2018 2019 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2020 2021 /* 2022 * Extended attributes can be applied to files, directories, etc. 2023 * Otherwise the parent must be a directory. 2024 */ 2025 if (!*is_xattrdir && !S_ISDIR(parent_mode)) 2026 return (SET_ERROR(EINVAL)); 2027 2028 *pobjp = parent; 2029 2030 return (0); 2031} 2032 2033/* 2034 * Given an object number, return some zpl level statistics 2035 */ 2036static int 2037zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2038 zfs_stat_t *sb) 2039{ 2040 sa_bulk_attr_t bulk[4]; 2041 int count = 0; 2042 2043 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2044 &sb->zs_mode, sizeof (sb->zs_mode)); 2045 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2046 &sb->zs_gen, sizeof (sb->zs_gen)); 2047 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2048 &sb->zs_links, sizeof (sb->zs_links)); 2049 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2050 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2051 2052 return (sa_bulk_lookup(hdl, bulk, count)); 2053} 2054 2055static int 2056zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2057 sa_attr_type_t *sa_table, char *buf, int len) 2058{ 2059 sa_handle_t *sa_hdl; 2060 sa_handle_t *prevhdl = NULL; 2061 dmu_buf_t *prevdb = NULL; 2062 dmu_buf_t *sa_db = NULL; 2063 char *path = buf + len - 1; 2064 int error; 2065 2066 *path = '\0'; 2067 sa_hdl = hdl; 2068 2069 for (;;) { 2070 uint64_t pobj; 2071 char component[MAXNAMELEN + 2]; 2072 size_t complen; 2073 int is_xattrdir; 2074 2075 if (prevdb) 2076 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2077 2078 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj, 2079 &is_xattrdir)) != 0) 2080 break; 2081 2082 if (pobj == obj) { 2083 if (path[0] != '/') 2084 *--path = '/'; 2085 break; 2086 } 2087 2088 component[0] = '/'; 2089 if (is_xattrdir) { 2090 (void) sprintf(component + 1, "<xattrdir>"); 2091 } else { 2092 error = zap_value_search(osp, pobj, obj, 2093 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2094 if (error != 0) 2095 break; 2096 } 2097 2098 complen = strlen(component); 2099 path -= complen; 2100 ASSERT(path >= buf); 2101 bcopy(component, path, complen); 2102 obj = pobj; 2103 2104 if (sa_hdl != hdl) { 2105 prevhdl = sa_hdl; 2106 prevdb = sa_db; 2107 } 2108 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2109 if (error != 0) { 2110 sa_hdl = prevhdl; 2111 sa_db = prevdb; 2112 break; 2113 } 2114 } 2115 2116 if (sa_hdl != NULL && sa_hdl != hdl) { 2117 ASSERT(sa_db != NULL); 2118 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2119 } 2120 2121 if (error == 0) 2122 (void) memmove(buf, path, buf + len - path); 2123 2124 return (error); 2125} 2126 2127int 2128zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2129{ 2130 sa_attr_type_t *sa_table; 2131 sa_handle_t *hdl; 2132 dmu_buf_t *db; 2133 int error; 2134 2135 error = zfs_sa_setup(osp, &sa_table); 2136 if (error != 0) 2137 return (error); 2138 2139 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2140 if (error != 0) 2141 return (error); 2142 2143 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2144 2145 zfs_release_sa_handle(hdl, db, FTAG); 2146 return (error); 2147} 2148 2149int 2150zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2151 char *buf, int len) 2152{ 2153 char *path = buf + len - 1; 2154 sa_attr_type_t *sa_table; 2155 sa_handle_t *hdl; 2156 dmu_buf_t *db; 2157 int error; 2158 2159 *path = '\0'; 2160 2161 error = zfs_sa_setup(osp, &sa_table); 2162 if (error != 0) 2163 return (error); 2164 2165 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2166 if (error != 0) 2167 return (error); 2168 2169 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2170 if (error != 0) { 2171 zfs_release_sa_handle(hdl, db, FTAG); 2172 return (error); 2173 } 2174 2175 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2176 2177 zfs_release_sa_handle(hdl, db, FTAG); 2178 return (error); 2179} 2180