zfs_znode.c revision 297077
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) 2012, 2014 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/dmu_objset.h> 58#include <sys/refcount.h> 59#include <sys/stat.h> 60#include <sys/zap.h> 61#include <sys/zfs_znode.h> 62#include <sys/sa.h> 63#include <sys/zfs_sa.h> 64#include <sys/zfs_stat.h> 65#include <sys/refcount.h> 66 67#include "zfs_prop.h" 68#include "zfs_comutil.h" 69 70/* Used by fstat(1). */ 71SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 72 SYSCTL_NULL_INT_PTR, sizeof(znode_t), "sizeof(znode_t)"); 73 74/* 75 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 76 * turned on when DEBUG is also defined. 77 */ 78#ifdef DEBUG 79#define ZNODE_STATS 80#endif /* DEBUG */ 81 82#ifdef ZNODE_STATS 83#define ZNODE_STAT_ADD(stat) ((stat)++) 84#else 85#define ZNODE_STAT_ADD(stat) /* nothing */ 86#endif /* ZNODE_STATS */ 87 88/* 89 * Functions needed for userland (ie: libzpool) are not put under 90 * #ifdef_KERNEL; the rest of the functions have dependencies 91 * (such as VFS logic) that will not compile easily in userland. 92 */ 93#ifdef _KERNEL 94/* 95 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 96 * be freed before it can be safely accessed. 97 */ 98krwlock_t zfsvfs_lock; 99 100static kmem_cache_t *znode_cache = NULL; 101 102/*ARGSUSED*/ 103static void 104znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 105{ 106 /* 107 * We should never drop all dbuf refs without first clearing 108 * the eviction callback. 109 */ 110 panic("evicting znode %p\n", user_ptr); 111} 112 113extern struct vop_vector zfs_vnodeops; 114extern struct vop_vector zfs_fifoops; 115extern struct vop_vector zfs_shareops; 116 117static int 118zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 119{ 120 znode_t *zp = buf; 121 122 POINTER_INVALIDATE(&zp->z_zfsvfs); 123 124 list_link_init(&zp->z_link_node); 125 126 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 127 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 128 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 129 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 130 131 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 132 avl_create(&zp->z_range_avl, zfs_range_compare, 133 sizeof (rl_t), offsetof(rl_t, r_node)); 134 135 zp->z_dirlocks = NULL; 136 zp->z_acl_cached = NULL; 137 zp->z_vnode = NULL; 138 zp->z_moved = 0; 139 return (0); 140} 141 142/*ARGSUSED*/ 143static void 144zfs_znode_cache_destructor(void *buf, void *arg) 145{ 146 znode_t *zp = buf; 147 148 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 149 ASSERT(ZTOV(zp) == NULL); 150 vn_free(ZTOV(zp)); 151 ASSERT(!list_link_active(&zp->z_link_node)); 152 mutex_destroy(&zp->z_lock); 153 rw_destroy(&zp->z_parent_lock); 154 rw_destroy(&zp->z_name_lock); 155 mutex_destroy(&zp->z_acl_lock); 156 avl_destroy(&zp->z_range_avl); 157 mutex_destroy(&zp->z_range_lock); 158 159 ASSERT(zp->z_dirlocks == NULL); 160 ASSERT(zp->z_acl_cached == NULL); 161} 162 163#ifdef ZNODE_STATS 164static struct { 165 uint64_t zms_zfsvfs_invalid; 166 uint64_t zms_zfsvfs_recheck1; 167 uint64_t zms_zfsvfs_unmounted; 168 uint64_t zms_zfsvfs_recheck2; 169 uint64_t zms_obj_held; 170 uint64_t zms_vnode_locked; 171 uint64_t zms_not_only_dnlc; 172} znode_move_stats; 173#endif /* ZNODE_STATS */ 174 175#ifdef illumos 176static void 177zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 178{ 179 vnode_t *vp; 180 181 /* Copy fields. */ 182 nzp->z_zfsvfs = ozp->z_zfsvfs; 183 184 /* Swap vnodes. */ 185 vp = nzp->z_vnode; 186 nzp->z_vnode = ozp->z_vnode; 187 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 188 ZTOV(ozp)->v_data = ozp; 189 ZTOV(nzp)->v_data = nzp; 190 191 nzp->z_id = ozp->z_id; 192 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 193 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 194 nzp->z_unlinked = ozp->z_unlinked; 195 nzp->z_atime_dirty = ozp->z_atime_dirty; 196 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 197 nzp->z_blksz = ozp->z_blksz; 198 nzp->z_seq = ozp->z_seq; 199 nzp->z_mapcnt = ozp->z_mapcnt; 200 nzp->z_gen = ozp->z_gen; 201 nzp->z_sync_cnt = ozp->z_sync_cnt; 202 nzp->z_is_sa = ozp->z_is_sa; 203 nzp->z_sa_hdl = ozp->z_sa_hdl; 204 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 205 nzp->z_links = ozp->z_links; 206 nzp->z_size = ozp->z_size; 207 nzp->z_pflags = ozp->z_pflags; 208 nzp->z_uid = ozp->z_uid; 209 nzp->z_gid = ozp->z_gid; 210 nzp->z_mode = ozp->z_mode; 211 212 /* 213 * Since this is just an idle znode and kmem is already dealing with 214 * memory pressure, release any cached ACL. 215 */ 216 if (ozp->z_acl_cached) { 217 zfs_acl_free(ozp->z_acl_cached); 218 ozp->z_acl_cached = NULL; 219 } 220 221 sa_set_userp(nzp->z_sa_hdl, nzp); 222 223 /* 224 * Invalidate the original znode by clearing fields that provide a 225 * pointer back to the znode. Set the low bit of the vfs pointer to 226 * ensure that zfs_znode_move() recognizes the znode as invalid in any 227 * subsequent callback. 228 */ 229 ozp->z_sa_hdl = NULL; 230 POINTER_INVALIDATE(&ozp->z_zfsvfs); 231 232 /* 233 * Mark the znode. 234 */ 235 nzp->z_moved = 1; 236 ozp->z_moved = (uint8_t)-1; 237} 238 239/*ARGSUSED*/ 240static kmem_cbrc_t 241zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 242{ 243 znode_t *ozp = buf, *nzp = newbuf; 244 zfsvfs_t *zfsvfs; 245 vnode_t *vp; 246 247 /* 248 * The znode is on the file system's list of known znodes if the vfs 249 * pointer is valid. We set the low bit of the vfs pointer when freeing 250 * the znode to invalidate it, and the memory patterns written by kmem 251 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 252 * created znode sets the vfs pointer last of all to indicate that the 253 * znode is known and in a valid state to be moved by this function. 254 */ 255 zfsvfs = ozp->z_zfsvfs; 256 if (!POINTER_IS_VALID(zfsvfs)) { 257 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 258 return (KMEM_CBRC_DONT_KNOW); 259 } 260 261 /* 262 * Close a small window in which it's possible that the filesystem could 263 * be unmounted and freed, and zfsvfs, though valid in the previous 264 * statement, could point to unrelated memory by the time we try to 265 * prevent the filesystem from being unmounted. 266 */ 267 rw_enter(&zfsvfs_lock, RW_WRITER); 268 if (zfsvfs != ozp->z_zfsvfs) { 269 rw_exit(&zfsvfs_lock); 270 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 271 return (KMEM_CBRC_DONT_KNOW); 272 } 273 274 /* 275 * If the znode is still valid, then so is the file system. We know that 276 * no valid file system can be freed while we hold zfsvfs_lock, so we 277 * can safely ensure that the filesystem is not and will not be 278 * unmounted. The next statement is equivalent to ZFS_ENTER(). 279 */ 280 rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 281 if (zfsvfs->z_unmounted) { 282 ZFS_EXIT(zfsvfs); 283 rw_exit(&zfsvfs_lock); 284 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 285 return (KMEM_CBRC_DONT_KNOW); 286 } 287 rw_exit(&zfsvfs_lock); 288 289 mutex_enter(&zfsvfs->z_znodes_lock); 290 /* 291 * Recheck the vfs pointer in case the znode was removed just before 292 * acquiring the lock. 293 */ 294 if (zfsvfs != ozp->z_zfsvfs) { 295 mutex_exit(&zfsvfs->z_znodes_lock); 296 ZFS_EXIT(zfsvfs); 297 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 298 return (KMEM_CBRC_DONT_KNOW); 299 } 300 301 /* 302 * At this point we know that as long as we hold z_znodes_lock, the 303 * znode cannot be freed and fields within the znode can be safely 304 * accessed. Now, prevent a race with zfs_zget(). 305 */ 306 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 307 mutex_exit(&zfsvfs->z_znodes_lock); 308 ZFS_EXIT(zfsvfs); 309 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 310 return (KMEM_CBRC_LATER); 311 } 312 313 vp = ZTOV(ozp); 314 if (mutex_tryenter(&vp->v_lock) == 0) { 315 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 316 mutex_exit(&zfsvfs->z_znodes_lock); 317 ZFS_EXIT(zfsvfs); 318 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 319 return (KMEM_CBRC_LATER); 320 } 321 322 /* Only move znodes that are referenced _only_ by the DNLC. */ 323 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 324 mutex_exit(&vp->v_lock); 325 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 326 mutex_exit(&zfsvfs->z_znodes_lock); 327 ZFS_EXIT(zfsvfs); 328 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 329 return (KMEM_CBRC_LATER); 330 } 331 332 /* 333 * The znode is known and in a valid state to move. We're holding the 334 * locks needed to execute the critical section. 335 */ 336 zfs_znode_move_impl(ozp, nzp); 337 mutex_exit(&vp->v_lock); 338 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 339 340 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 341 mutex_exit(&zfsvfs->z_znodes_lock); 342 ZFS_EXIT(zfsvfs); 343 344 return (KMEM_CBRC_YES); 345} 346#endif /* illumos */ 347 348void 349zfs_znode_init(void) 350{ 351 /* 352 * Initialize zcache 353 */ 354 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 355 ASSERT(znode_cache == NULL); 356 znode_cache = kmem_cache_create("zfs_znode_cache", 357 sizeof (znode_t), 0, zfs_znode_cache_constructor, 358 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 359 kmem_cache_set_move(znode_cache, zfs_znode_move); 360} 361 362void 363zfs_znode_fini(void) 364{ 365#ifdef illumos 366 /* 367 * Cleanup vfs & vnode ops 368 */ 369 zfs_remove_op_tables(); 370#endif 371 372 /* 373 * Cleanup zcache 374 */ 375 if (znode_cache) 376 kmem_cache_destroy(znode_cache); 377 znode_cache = NULL; 378 rw_destroy(&zfsvfs_lock); 379} 380 381#ifdef illumos 382struct vnodeops *zfs_dvnodeops; 383struct vnodeops *zfs_fvnodeops; 384struct vnodeops *zfs_symvnodeops; 385struct vnodeops *zfs_xdvnodeops; 386struct vnodeops *zfs_evnodeops; 387struct vnodeops *zfs_sharevnodeops; 388 389void 390zfs_remove_op_tables() 391{ 392 /* 393 * Remove vfs ops 394 */ 395 ASSERT(zfsfstype); 396 (void) vfs_freevfsops_by_type(zfsfstype); 397 zfsfstype = 0; 398 399 /* 400 * Remove vnode ops 401 */ 402 if (zfs_dvnodeops) 403 vn_freevnodeops(zfs_dvnodeops); 404 if (zfs_fvnodeops) 405 vn_freevnodeops(zfs_fvnodeops); 406 if (zfs_symvnodeops) 407 vn_freevnodeops(zfs_symvnodeops); 408 if (zfs_xdvnodeops) 409 vn_freevnodeops(zfs_xdvnodeops); 410 if (zfs_evnodeops) 411 vn_freevnodeops(zfs_evnodeops); 412 if (zfs_sharevnodeops) 413 vn_freevnodeops(zfs_sharevnodeops); 414 415 zfs_dvnodeops = NULL; 416 zfs_fvnodeops = NULL; 417 zfs_symvnodeops = NULL; 418 zfs_xdvnodeops = NULL; 419 zfs_evnodeops = NULL; 420 zfs_sharevnodeops = NULL; 421} 422 423extern const fs_operation_def_t zfs_dvnodeops_template[]; 424extern const fs_operation_def_t zfs_fvnodeops_template[]; 425extern const fs_operation_def_t zfs_xdvnodeops_template[]; 426extern const fs_operation_def_t zfs_symvnodeops_template[]; 427extern const fs_operation_def_t zfs_evnodeops_template[]; 428extern const fs_operation_def_t zfs_sharevnodeops_template[]; 429 430int 431zfs_create_op_tables() 432{ 433 int error; 434 435 /* 436 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 437 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 438 * In this case we just return as the ops vectors are already set up. 439 */ 440 if (zfs_dvnodeops) 441 return (0); 442 443 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 444 &zfs_dvnodeops); 445 if (error) 446 return (error); 447 448 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 449 &zfs_fvnodeops); 450 if (error) 451 return (error); 452 453 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 454 &zfs_symvnodeops); 455 if (error) 456 return (error); 457 458 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 459 &zfs_xdvnodeops); 460 if (error) 461 return (error); 462 463 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 464 &zfs_evnodeops); 465 if (error) 466 return (error); 467 468 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 469 &zfs_sharevnodeops); 470 471 return (error); 472} 473#endif /* illumos */ 474 475int 476zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 477{ 478 zfs_acl_ids_t acl_ids; 479 vattr_t vattr; 480 znode_t *sharezp; 481 znode_t *zp; 482 int error; 483 484 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 485 vattr.va_type = VDIR; 486 vattr.va_mode = S_IFDIR|0555; 487 vattr.va_uid = crgetuid(kcred); 488 vattr.va_gid = crgetgid(kcred); 489 490 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 491 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 492 sharezp->z_moved = 0; 493 sharezp->z_unlinked = 0; 494 sharezp->z_atime_dirty = 0; 495 sharezp->z_zfsvfs = zfsvfs; 496 sharezp->z_is_sa = zfsvfs->z_use_sa; 497 498 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 499 kcred, NULL, &acl_ids)); 500 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 501 ASSERT3P(zp, ==, sharezp); 502 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 503 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 504 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 505 zfsvfs->z_shares_dir = sharezp->z_id; 506 507 zfs_acl_ids_free(&acl_ids); 508 sa_handle_destroy(sharezp->z_sa_hdl); 509 kmem_cache_free(znode_cache, sharezp); 510 511 return (error); 512} 513 514/* 515 * define a couple of values we need available 516 * for both 64 and 32 bit environments. 517 */ 518#ifndef NBITSMINOR64 519#define NBITSMINOR64 32 520#endif 521#ifndef MAXMAJ64 522#define MAXMAJ64 0xffffffffUL 523#endif 524#ifndef MAXMIN64 525#define MAXMIN64 0xffffffffUL 526#endif 527 528/* 529 * Create special expldev for ZFS private use. 530 * Can't use standard expldev since it doesn't do 531 * what we want. The standard expldev() takes a 532 * dev32_t in LP64 and expands it to a long dev_t. 533 * We need an interface that takes a dev32_t in ILP32 534 * and expands it to a long dev_t. 535 */ 536static uint64_t 537zfs_expldev(dev_t dev) 538{ 539 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 540} 541/* 542 * Special cmpldev for ZFS private use. 543 * Can't use standard cmpldev since it takes 544 * a long dev_t and compresses it to dev32_t in 545 * LP64. We need to do a compaction of a long dev_t 546 * to a dev32_t in ILP32. 547 */ 548dev_t 549zfs_cmpldev(uint64_t dev) 550{ 551 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 552} 553 554static void 555zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 556 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 557{ 558 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 559 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 560 561 mutex_enter(&zp->z_lock); 562 563 ASSERT(zp->z_sa_hdl == NULL); 564 ASSERT(zp->z_acl_cached == NULL); 565 if (sa_hdl == NULL) { 566 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 567 SA_HDL_SHARED, &zp->z_sa_hdl)); 568 } else { 569 zp->z_sa_hdl = sa_hdl; 570 sa_set_userp(sa_hdl, zp); 571 } 572 573 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 574 575 /* 576 * Slap on VROOT if we are the root znode 577 */ 578 if (zp->z_id == zfsvfs->z_root) 579 ZTOV(zp)->v_flag |= VROOT; 580 581 mutex_exit(&zp->z_lock); 582 vn_exists(ZTOV(zp)); 583} 584 585void 586zfs_znode_dmu_fini(znode_t *zp) 587{ 588 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 589 zp->z_unlinked || 590 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 591 592 sa_handle_destroy(zp->z_sa_hdl); 593 zp->z_sa_hdl = NULL; 594} 595 596static void 597zfs_vnode_forget(vnode_t *vp) 598{ 599 600 /* copied from insmntque_stddtr */ 601 vp->v_data = NULL; 602 vp->v_op = &dead_vnodeops; 603 vgone(vp); 604 vput(vp); 605} 606 607/* 608 * Construct a new znode/vnode and intialize. 609 * 610 * This does not do a call to dmu_set_user() that is 611 * up to the caller to do, in case you don't want to 612 * return the znode 613 */ 614static znode_t * 615zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 616 dmu_object_type_t obj_type, sa_handle_t *hdl) 617{ 618 znode_t *zp; 619 vnode_t *vp; 620 uint64_t mode; 621 uint64_t parent; 622 sa_bulk_attr_t bulk[9]; 623 int count = 0; 624 int error; 625 626 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 627 628 KASSERT(curthread->td_vp_reserv > 0, 629 ("zfs_znode_alloc: getnewvnode without any vnodes reserved")); 630 error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp); 631 if (error != 0) { 632 kmem_cache_free(znode_cache, zp); 633 return (NULL); 634 } 635 zp->z_vnode = vp; 636 vp->v_data = zp; 637 638 ASSERT(zp->z_dirlocks == NULL); 639 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 640 zp->z_moved = 0; 641 642 /* 643 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 644 * the zfs_znode_move() callback. 645 */ 646 zp->z_sa_hdl = NULL; 647 zp->z_unlinked = 0; 648 zp->z_atime_dirty = 0; 649 zp->z_mapcnt = 0; 650 zp->z_id = db->db_object; 651 zp->z_blksz = blksz; 652 zp->z_seq = 0x7A4653; 653 zp->z_sync_cnt = 0; 654 655 vp = ZTOV(zp); 656 657 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 658 659 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 660 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 661 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 662 &zp->z_size, 8); 663 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 664 &zp->z_links, 8); 665 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 666 &zp->z_pflags, 8); 667 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 668 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 669 &zp->z_atime, 16); 670 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 671 &zp->z_uid, 8); 672 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 673 &zp->z_gid, 8); 674 675 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 676 if (hdl == NULL) 677 sa_handle_destroy(zp->z_sa_hdl); 678 zfs_vnode_forget(vp); 679 zp->z_vnode = NULL; 680 kmem_cache_free(znode_cache, zp); 681 return (NULL); 682 } 683 684 zp->z_mode = mode; 685 686 vp->v_type = IFTOVT((mode_t)mode); 687 688 switch (vp->v_type) { 689 case VDIR: 690 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 691 break; 692#ifdef illumos 693 case VBLK: 694 case VCHR: 695 { 696 uint64_t rdev; 697 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 698 &rdev, sizeof (rdev)) == 0); 699 700 vp->v_rdev = zfs_cmpldev(rdev); 701 } 702 break; 703#endif 704 case VFIFO: 705#ifdef illumos 706 case VSOCK: 707 case VDOOR: 708#endif 709 vp->v_op = &zfs_fifoops; 710 break; 711 case VREG: 712 if (parent == zfsvfs->z_shares_dir) { 713 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 714 vp->v_op = &zfs_shareops; 715 } 716 break; 717#ifdef illumos 718 case VLNK: 719 vn_setops(vp, zfs_symvnodeops); 720 break; 721 default: 722 vn_setops(vp, zfs_evnodeops); 723 break; 724#endif 725 } 726 727 mutex_enter(&zfsvfs->z_znodes_lock); 728 list_insert_tail(&zfsvfs->z_all_znodes, zp); 729 membar_producer(); 730 /* 731 * Everything else must be valid before assigning z_zfsvfs makes the 732 * znode eligible for zfs_znode_move(). 733 */ 734 zp->z_zfsvfs = zfsvfs; 735 mutex_exit(&zfsvfs->z_znodes_lock); 736 737 /* 738 * Acquire vnode lock before making it available to the world. 739 */ 740 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 741 VN_LOCK_AREC(vp); 742 if (vp->v_type != VFIFO) 743 VN_LOCK_ASHARE(vp); 744 745 VFS_HOLD(zfsvfs->z_vfs); 746 return (zp); 747} 748 749static uint64_t empty_xattr; 750static uint64_t pad[4]; 751static zfs_acl_phys_t acl_phys; 752/* 753 * Create a new DMU object to hold a zfs znode. 754 * 755 * IN: dzp - parent directory for new znode 756 * vap - file attributes for new znode 757 * tx - dmu transaction id for zap operations 758 * cr - credentials of caller 759 * flag - flags: 760 * IS_ROOT_NODE - new object will be root 761 * IS_XATTR - new object is an attribute 762 * bonuslen - length of bonus buffer 763 * setaclp - File/Dir initial ACL 764 * fuidp - Tracks fuid allocation. 765 * 766 * OUT: zpp - allocated znode 767 * 768 */ 769void 770zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 771 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 772{ 773 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 774 uint64_t mode, size, links, parent, pflags; 775 uint64_t dzp_pflags = 0; 776 uint64_t rdev = 0; 777 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 778 dmu_buf_t *db; 779 timestruc_t now; 780 uint64_t gen, obj; 781 int err; 782 int bonuslen; 783 sa_handle_t *sa_hdl; 784 dmu_object_type_t obj_type; 785 sa_bulk_attr_t sa_attrs[ZPL_END]; 786 int cnt = 0; 787 zfs_acl_locator_cb_t locate = { 0 }; 788 789 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 790 791 if (zfsvfs->z_replay) { 792 obj = vap->va_nodeid; 793 now = vap->va_ctime; /* see zfs_replay_create() */ 794 gen = vap->va_nblocks; /* ditto */ 795 } else { 796 obj = 0; 797 gethrestime(&now); 798 gen = dmu_tx_get_txg(tx); 799 } 800 801 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 802 bonuslen = (obj_type == DMU_OT_SA) ? 803 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 804 805 /* 806 * Create a new DMU object. 807 */ 808 /* 809 * There's currently no mechanism for pre-reading the blocks that will 810 * be needed to allocate a new object, so we accept the small chance 811 * that there will be an i/o error and we will fail one of the 812 * assertions below. 813 */ 814 if (vap->va_type == VDIR) { 815 if (zfsvfs->z_replay) { 816 VERIFY0(zap_create_claim_norm(zfsvfs->z_os, obj, 817 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 818 obj_type, bonuslen, tx)); 819 } else { 820 obj = zap_create_norm(zfsvfs->z_os, 821 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 822 obj_type, bonuslen, tx); 823 } 824 } else { 825 if (zfsvfs->z_replay) { 826 VERIFY0(dmu_object_claim(zfsvfs->z_os, obj, 827 DMU_OT_PLAIN_FILE_CONTENTS, 0, 828 obj_type, bonuslen, tx)); 829 } else { 830 obj = dmu_object_alloc(zfsvfs->z_os, 831 DMU_OT_PLAIN_FILE_CONTENTS, 0, 832 obj_type, bonuslen, tx); 833 } 834 } 835 836 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 837 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 838 839 /* 840 * If this is the root, fix up the half-initialized parent pointer 841 * to reference the just-allocated physical data area. 842 */ 843 if (flag & IS_ROOT_NODE) { 844 dzp->z_id = obj; 845 } else { 846 dzp_pflags = dzp->z_pflags; 847 } 848 849 /* 850 * If parent is an xattr, so am I. 851 */ 852 if (dzp_pflags & ZFS_XATTR) { 853 flag |= IS_XATTR; 854 } 855 856 if (zfsvfs->z_use_fuids) 857 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 858 else 859 pflags = 0; 860 861 if (vap->va_type == VDIR) { 862 size = 2; /* contents ("." and "..") */ 863 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 864 } else { 865 size = links = 0; 866 } 867 868 if (vap->va_type == VBLK || vap->va_type == VCHR) { 869 rdev = zfs_expldev(vap->va_rdev); 870 } 871 872 parent = dzp->z_id; 873 mode = acl_ids->z_mode; 874 if (flag & IS_XATTR) 875 pflags |= ZFS_XATTR; 876 877 /* 878 * No execs denied will be deterimed when zfs_mode_compute() is called. 879 */ 880 pflags |= acl_ids->z_aclp->z_hints & 881 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 882 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 883 884 ZFS_TIME_ENCODE(&now, crtime); 885 ZFS_TIME_ENCODE(&now, ctime); 886 887 if (vap->va_mask & AT_ATIME) { 888 ZFS_TIME_ENCODE(&vap->va_atime, atime); 889 } else { 890 ZFS_TIME_ENCODE(&now, atime); 891 } 892 893 if (vap->va_mask & AT_MTIME) { 894 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 895 } else { 896 ZFS_TIME_ENCODE(&now, mtime); 897 } 898 899 /* Now add in all of the "SA" attributes */ 900 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 901 &sa_hdl)); 902 903 /* 904 * Setup the array of attributes to be replaced/set on the new file 905 * 906 * order for DMU_OT_ZNODE is critical since it needs to be constructed 907 * in the old znode_phys_t format. Don't change this ordering 908 */ 909 910 if (obj_type == DMU_OT_ZNODE) { 911 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 912 NULL, &atime, 16); 913 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 914 NULL, &mtime, 16); 915 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 916 NULL, &ctime, 16); 917 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 918 NULL, &crtime, 16); 919 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 920 NULL, &gen, 8); 921 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 922 NULL, &mode, 8); 923 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 924 NULL, &size, 8); 925 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 926 NULL, &parent, 8); 927 } else { 928 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 929 NULL, &mode, 8); 930 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 931 NULL, &size, 8); 932 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 933 NULL, &gen, 8); 934 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 935 &acl_ids->z_fuid, 8); 936 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 937 &acl_ids->z_fgid, 8); 938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 939 NULL, &parent, 8); 940 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 941 NULL, &pflags, 8); 942 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 943 NULL, &atime, 16); 944 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 945 NULL, &mtime, 16); 946 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 947 NULL, &ctime, 16); 948 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 949 NULL, &crtime, 16); 950 } 951 952 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 953 954 if (obj_type == DMU_OT_ZNODE) { 955 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 956 &empty_xattr, 8); 957 } 958 if (obj_type == DMU_OT_ZNODE || 959 (vap->va_type == VBLK || vap->va_type == VCHR)) { 960 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 961 NULL, &rdev, 8); 962 963 } 964 if (obj_type == DMU_OT_ZNODE) { 965 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 966 NULL, &pflags, 8); 967 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 968 &acl_ids->z_fuid, 8); 969 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 970 &acl_ids->z_fgid, 8); 971 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 972 sizeof (uint64_t) * 4); 973 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 974 &acl_phys, sizeof (zfs_acl_phys_t)); 975 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 976 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 977 &acl_ids->z_aclp->z_acl_count, 8); 978 locate.cb_aclp = acl_ids->z_aclp; 979 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 980 zfs_acl_data_locator, &locate, 981 acl_ids->z_aclp->z_acl_bytes); 982 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 983 acl_ids->z_fuid, acl_ids->z_fgid); 984 } 985 986 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 987 988 if (!(flag & IS_ROOT_NODE)) { 989 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 990 ASSERT(*zpp != NULL); 991 } else { 992 /* 993 * If we are creating the root node, the "parent" we 994 * passed in is the znode for the root. 995 */ 996 *zpp = dzp; 997 998 (*zpp)->z_sa_hdl = sa_hdl; 999 } 1000 1001 (*zpp)->z_pflags = pflags; 1002 (*zpp)->z_mode = mode; 1003 1004 if (vap->va_mask & AT_XVATTR) 1005 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 1006 1007 if (obj_type == DMU_OT_ZNODE || 1008 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1009 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx)); 1010 } 1011 if (!(flag & IS_ROOT_NODE)) { 1012 vnode_t *vp; 1013 1014 vp = ZTOV(*zpp); 1015 vp->v_vflag |= VV_FORCEINSMQ; 1016 err = insmntque(vp, zfsvfs->z_vfs); 1017 vp->v_vflag &= ~VV_FORCEINSMQ; 1018 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1019 } 1020 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1021} 1022 1023/* 1024 * Update in-core attributes. It is assumed the caller will be doing an 1025 * sa_bulk_update to push the changes out. 1026 */ 1027void 1028zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1029{ 1030 xoptattr_t *xoap; 1031 1032 xoap = xva_getxoptattr(xvap); 1033 ASSERT(xoap); 1034 1035 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1036 uint64_t times[2]; 1037 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1038 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1039 ×, sizeof (times), tx); 1040 XVA_SET_RTN(xvap, XAT_CREATETIME); 1041 } 1042 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1043 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1044 zp->z_pflags, tx); 1045 XVA_SET_RTN(xvap, XAT_READONLY); 1046 } 1047 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1048 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1049 zp->z_pflags, tx); 1050 XVA_SET_RTN(xvap, XAT_HIDDEN); 1051 } 1052 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1053 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1054 zp->z_pflags, tx); 1055 XVA_SET_RTN(xvap, XAT_SYSTEM); 1056 } 1057 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1058 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1059 zp->z_pflags, tx); 1060 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1061 } 1062 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1063 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1064 zp->z_pflags, tx); 1065 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1066 } 1067 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1068 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1069 zp->z_pflags, tx); 1070 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1071 } 1072 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1073 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1074 zp->z_pflags, tx); 1075 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1076 } 1077 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1078 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1079 zp->z_pflags, tx); 1080 XVA_SET_RTN(xvap, XAT_NODUMP); 1081 } 1082 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1083 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1084 zp->z_pflags, tx); 1085 XVA_SET_RTN(xvap, XAT_OPAQUE); 1086 } 1087 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1088 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1089 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1090 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1091 } 1092 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1093 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1094 zp->z_pflags, tx); 1095 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1096 } 1097 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1098 zfs_sa_set_scanstamp(zp, xvap, tx); 1099 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1100 } 1101 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1102 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1103 zp->z_pflags, tx); 1104 XVA_SET_RTN(xvap, XAT_REPARSE); 1105 } 1106 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1107 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1108 zp->z_pflags, tx); 1109 XVA_SET_RTN(xvap, XAT_OFFLINE); 1110 } 1111 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1112 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1113 zp->z_pflags, tx); 1114 XVA_SET_RTN(xvap, XAT_SPARSE); 1115 } 1116} 1117 1118int 1119zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1120{ 1121 dmu_object_info_t doi; 1122 dmu_buf_t *db; 1123 znode_t *zp; 1124 vnode_t *vp; 1125 sa_handle_t *hdl; 1126 struct thread *td; 1127 int locked; 1128 int err; 1129 1130 td = curthread; 1131 getnewvnode_reserve(1); 1132again: 1133 *zpp = NULL; 1134 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1135 1136 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1137 if (err) { 1138 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1139 getnewvnode_drop_reserve(); 1140 return (err); 1141 } 1142 1143 dmu_object_info_from_db(db, &doi); 1144 if (doi.doi_bonus_type != DMU_OT_SA && 1145 (doi.doi_bonus_type != DMU_OT_ZNODE || 1146 (doi.doi_bonus_type == DMU_OT_ZNODE && 1147 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1148 sa_buf_rele(db, NULL); 1149 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1150#ifdef __FreeBSD__ 1151 getnewvnode_drop_reserve(); 1152#endif 1153 return (SET_ERROR(EINVAL)); 1154 } 1155 1156 hdl = dmu_buf_get_user(db); 1157 if (hdl != NULL) { 1158 zp = sa_get_userdata(hdl); 1159 1160 1161 /* 1162 * Since "SA" does immediate eviction we 1163 * should never find a sa handle that doesn't 1164 * know about the znode. 1165 */ 1166 1167 ASSERT3P(zp, !=, NULL); 1168 1169 mutex_enter(&zp->z_lock); 1170 ASSERT3U(zp->z_id, ==, obj_num); 1171 if (zp->z_unlinked) { 1172 err = SET_ERROR(ENOENT); 1173 } else { 1174 vp = ZTOV(zp); 1175 *zpp = zp; 1176 err = 0; 1177 } 1178 sa_buf_rele(db, NULL); 1179 1180 /* Don't let the vnode disappear after ZFS_OBJ_HOLD_EXIT. */ 1181 if (err == 0) 1182 VN_HOLD(vp); 1183 1184 mutex_exit(&zp->z_lock); 1185 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1186 1187 if (err == 0) { 1188 locked = VOP_ISLOCKED(vp); 1189 VI_LOCK(vp); 1190 if ((vp->v_iflag & VI_DOOMED) != 0 && 1191 locked != LK_EXCLUSIVE) { 1192 /* 1193 * The vnode is doomed and this thread doesn't 1194 * hold the exclusive lock on it, so the vnode 1195 * must be being reclaimed by another thread. 1196 * Otherwise the doomed vnode is being reclaimed 1197 * by this thread and zfs_zget is called from 1198 * ZIL internals. 1199 */ 1200 VI_UNLOCK(vp); 1201 VN_RELE(vp); 1202 goto again; 1203 } 1204 VI_UNLOCK(vp); 1205 } 1206 getnewvnode_drop_reserve(); 1207 return (err); 1208 } 1209 1210 /* 1211 * Not found create new znode/vnode 1212 * but only if file exists. 1213 * 1214 * There is a small window where zfs_vget() could 1215 * find this object while a file create is still in 1216 * progress. This is checked for in zfs_znode_alloc() 1217 * 1218 * if zfs_znode_alloc() fails it will drop the hold on the 1219 * bonus buffer. 1220 */ 1221 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1222 doi.doi_bonus_type, NULL); 1223 if (zp == NULL) { 1224 err = SET_ERROR(ENOENT); 1225 } else { 1226 *zpp = zp; 1227 } 1228 if (err == 0) { 1229 vnode_t *vp = ZTOV(zp); 1230 1231 err = insmntque(vp, zfsvfs->z_vfs); 1232 if (err == 0) { 1233 vp->v_hash = obj_num; 1234 VOP_UNLOCK(vp, 0); 1235 } else { 1236 zp->z_vnode = NULL; 1237 zfs_znode_dmu_fini(zp); 1238 zfs_znode_free(zp); 1239 *zpp = NULL; 1240 } 1241 } 1242 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1243 getnewvnode_drop_reserve(); 1244 return (err); 1245} 1246 1247int 1248zfs_rezget(znode_t *zp) 1249{ 1250 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1251 dmu_object_info_t doi; 1252 dmu_buf_t *db; 1253 vnode_t *vp; 1254 uint64_t obj_num = zp->z_id; 1255 uint64_t mode, size; 1256 sa_bulk_attr_t bulk[8]; 1257 int err; 1258 int count = 0; 1259 uint64_t gen; 1260 1261 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1262 1263 mutex_enter(&zp->z_acl_lock); 1264 if (zp->z_acl_cached) { 1265 zfs_acl_free(zp->z_acl_cached); 1266 zp->z_acl_cached = NULL; 1267 } 1268 1269 mutex_exit(&zp->z_acl_lock); 1270 ASSERT(zp->z_sa_hdl == NULL); 1271 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1272 if (err) { 1273 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1274 return (err); 1275 } 1276 1277 dmu_object_info_from_db(db, &doi); 1278 if (doi.doi_bonus_type != DMU_OT_SA && 1279 (doi.doi_bonus_type != DMU_OT_ZNODE || 1280 (doi.doi_bonus_type == DMU_OT_ZNODE && 1281 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1282 sa_buf_rele(db, NULL); 1283 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1284 return (SET_ERROR(EINVAL)); 1285 } 1286 1287 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1288 size = zp->z_size; 1289 1290 /* reload cached values */ 1291 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1292 &gen, sizeof (gen)); 1293 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1294 &zp->z_size, sizeof (zp->z_size)); 1295 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1296 &zp->z_links, sizeof (zp->z_links)); 1297 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1298 &zp->z_pflags, sizeof (zp->z_pflags)); 1299 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1300 &zp->z_atime, sizeof (zp->z_atime)); 1301 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1302 &zp->z_uid, sizeof (zp->z_uid)); 1303 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1304 &zp->z_gid, sizeof (zp->z_gid)); 1305 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1306 &mode, sizeof (mode)); 1307 1308 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1309 zfs_znode_dmu_fini(zp); 1310 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1311 return (SET_ERROR(EIO)); 1312 } 1313 1314 zp->z_mode = mode; 1315 1316 if (gen != zp->z_gen) { 1317 zfs_znode_dmu_fini(zp); 1318 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1319 return (SET_ERROR(EIO)); 1320 } 1321 1322 /* 1323 * It is highly improbable but still quite possible that two 1324 * objects in different datasets are created with the same 1325 * object numbers and in transaction groups with the same 1326 * numbers. znodes corresponding to those objects would 1327 * have the same z_id and z_gen, but their other attributes 1328 * may be different. 1329 * zfs recv -F may replace one of such objects with the other. 1330 * As a result file properties recorded in the replaced 1331 * object's vnode may no longer match the received object's 1332 * properties. At present the only cached property is the 1333 * files type recorded in v_type. 1334 * So, handle this case by leaving the old vnode and znode 1335 * disassociated from the actual object. A new vnode and a 1336 * znode will be created if the object is accessed 1337 * (e.g. via a look-up). The old vnode and znode will be 1338 * recycled when the last vnode reference is dropped. 1339 */ 1340 vp = ZTOV(zp); 1341 if (vp != NULL && vp->v_type != IFTOVT((mode_t)zp->z_mode)) { 1342 zfs_znode_dmu_fini(zp); 1343 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1344 return (EIO); 1345 } 1346 1347 zp->z_unlinked = (zp->z_links == 0); 1348 zp->z_blksz = doi.doi_data_block_size; 1349 if (vp != NULL) { 1350 vn_pages_remove(vp, 0, 0); 1351 if (zp->z_size != size) 1352 vnode_pager_setsize(vp, zp->z_size); 1353 } 1354 1355 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1356 1357 return (0); 1358} 1359 1360void 1361zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1362{ 1363 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1364 objset_t *os = zfsvfs->z_os; 1365 uint64_t obj = zp->z_id; 1366 uint64_t acl_obj = zfs_external_acl(zp); 1367 1368 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1369 if (acl_obj) { 1370 VERIFY(!zp->z_is_sa); 1371 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1372 } 1373 VERIFY(0 == dmu_object_free(os, obj, tx)); 1374 zfs_znode_dmu_fini(zp); 1375 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1376 zfs_znode_free(zp); 1377} 1378 1379void 1380zfs_zinactive(znode_t *zp) 1381{ 1382 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1383 uint64_t z_id = zp->z_id; 1384 1385 ASSERT(zp->z_sa_hdl); 1386 1387 /* 1388 * Don't allow a zfs_zget() while were trying to release this znode 1389 */ 1390 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1391 1392 mutex_enter(&zp->z_lock); 1393 1394 /* 1395 * If this was the last reference to a file with no links, 1396 * remove the file from the file system. 1397 */ 1398 if (zp->z_unlinked) { 1399 mutex_exit(&zp->z_lock); 1400 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1401 zfs_rmnode(zp); 1402 return; 1403 } 1404 1405 mutex_exit(&zp->z_lock); 1406 zfs_znode_dmu_fini(zp); 1407 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1408 zfs_znode_free(zp); 1409} 1410 1411void 1412zfs_znode_free(znode_t *zp) 1413{ 1414 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1415 1416 ASSERT(zp->z_sa_hdl == NULL); 1417 zp->z_vnode = NULL; 1418 mutex_enter(&zfsvfs->z_znodes_lock); 1419 POINTER_INVALIDATE(&zp->z_zfsvfs); 1420 list_remove(&zfsvfs->z_all_znodes, zp); 1421 mutex_exit(&zfsvfs->z_znodes_lock); 1422 1423 if (zp->z_acl_cached) { 1424 zfs_acl_free(zp->z_acl_cached); 1425 zp->z_acl_cached = NULL; 1426 } 1427 1428 kmem_cache_free(znode_cache, zp); 1429 1430 VFS_RELE(zfsvfs->z_vfs); 1431} 1432 1433void 1434zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1435 uint64_t ctime[2], boolean_t have_tx) 1436{ 1437 timestruc_t now; 1438 1439 gethrestime(&now); 1440 1441 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1442 zp->z_atime_dirty = 0; 1443 zp->z_seq++; 1444 } else { 1445 zp->z_atime_dirty = 1; 1446 } 1447 1448 if (flag & AT_ATIME) { 1449 ZFS_TIME_ENCODE(&now, zp->z_atime); 1450 } 1451 1452 if (flag & AT_MTIME) { 1453 ZFS_TIME_ENCODE(&now, mtime); 1454 if (zp->z_zfsvfs->z_use_fuids) { 1455 zp->z_pflags |= (ZFS_ARCHIVE | 1456 ZFS_AV_MODIFIED); 1457 } 1458 } 1459 1460 if (flag & AT_CTIME) { 1461 ZFS_TIME_ENCODE(&now, ctime); 1462 if (zp->z_zfsvfs->z_use_fuids) 1463 zp->z_pflags |= ZFS_ARCHIVE; 1464 } 1465} 1466 1467/* 1468 * Grow the block size for a file. 1469 * 1470 * IN: zp - znode of file to free data in. 1471 * size - requested block size 1472 * tx - open transaction. 1473 * 1474 * NOTE: this function assumes that the znode is write locked. 1475 */ 1476void 1477zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1478{ 1479 int error; 1480 u_longlong_t dummy; 1481 1482 if (size <= zp->z_blksz) 1483 return; 1484 /* 1485 * If the file size is already greater than the current blocksize, 1486 * we will not grow. If there is more than one block in a file, 1487 * the blocksize cannot change. 1488 */ 1489 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1490 return; 1491 1492 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1493 size, 0, tx); 1494 1495 if (error == ENOTSUP) 1496 return; 1497 ASSERT0(error); 1498 1499 /* What blocksize did we actually get? */ 1500 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1501} 1502 1503#ifdef illumos 1504/* 1505 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1506 * be calling back into the fs for a putpage(). E.g.: when truncating 1507 * a file, the pages being "thrown away* don't need to be written out. 1508 */ 1509/* ARGSUSED */ 1510static int 1511zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1512 int flags, cred_t *cr) 1513{ 1514 ASSERT(0); 1515 return (0); 1516} 1517#endif 1518 1519/* 1520 * Increase the file length 1521 * 1522 * IN: zp - znode of file to free data in. 1523 * end - new end-of-file 1524 * 1525 * RETURN: 0 on success, error code on failure 1526 */ 1527static int 1528zfs_extend(znode_t *zp, uint64_t end) 1529{ 1530 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1531 dmu_tx_t *tx; 1532 rl_t *rl; 1533 uint64_t newblksz; 1534 int error; 1535 1536 /* 1537 * We will change zp_size, lock the whole file. 1538 */ 1539 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1540 1541 /* 1542 * Nothing to do if file already at desired length. 1543 */ 1544 if (end <= zp->z_size) { 1545 zfs_range_unlock(rl); 1546 return (0); 1547 } 1548 tx = dmu_tx_create(zfsvfs->z_os); 1549 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1550 zfs_sa_upgrade_txholds(tx, zp); 1551 if (end > zp->z_blksz && 1552 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1553 /* 1554 * We are growing the file past the current block size. 1555 */ 1556 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1557 /* 1558 * File's blocksize is already larger than the 1559 * "recordsize" property. Only let it grow to 1560 * the next power of 2. 1561 */ 1562 ASSERT(!ISP2(zp->z_blksz)); 1563 newblksz = MIN(end, 1 << highbit64(zp->z_blksz)); 1564 } else { 1565 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1566 } 1567 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1568 } else { 1569 newblksz = 0; 1570 } 1571 1572 error = dmu_tx_assign(tx, TXG_WAIT); 1573 if (error) { 1574 dmu_tx_abort(tx); 1575 zfs_range_unlock(rl); 1576 return (error); 1577 } 1578 1579 if (newblksz) 1580 zfs_grow_blocksize(zp, newblksz, tx); 1581 1582 zp->z_size = end; 1583 1584 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1585 &zp->z_size, sizeof (zp->z_size), tx)); 1586 1587 vnode_pager_setsize(ZTOV(zp), end); 1588 1589 zfs_range_unlock(rl); 1590 1591 dmu_tx_commit(tx); 1592 1593 return (0); 1594} 1595 1596/* 1597 * Free space in a file. 1598 * 1599 * IN: zp - znode of file to free data in. 1600 * off - start of section to free. 1601 * len - length of section to free. 1602 * 1603 * RETURN: 0 on success, error code on failure 1604 */ 1605static int 1606zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1607{ 1608 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1609 rl_t *rl; 1610 int error; 1611 1612 /* 1613 * Lock the range being freed. 1614 */ 1615 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1616 1617 /* 1618 * Nothing to do if file already at desired length. 1619 */ 1620 if (off >= zp->z_size) { 1621 zfs_range_unlock(rl); 1622 return (0); 1623 } 1624 1625 if (off + len > zp->z_size) 1626 len = zp->z_size - off; 1627 1628 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1629 1630 if (error == 0) { 1631 /* 1632 * In FreeBSD we cannot free block in the middle of a file, 1633 * but only at the end of a file, so this code path should 1634 * never happen. 1635 */ 1636 vnode_pager_setsize(ZTOV(zp), off); 1637 } 1638 1639 zfs_range_unlock(rl); 1640 1641 return (error); 1642} 1643 1644/* 1645 * Truncate a file 1646 * 1647 * IN: zp - znode of file to free data in. 1648 * end - new end-of-file. 1649 * 1650 * RETURN: 0 on success, error code on failure 1651 */ 1652static int 1653zfs_trunc(znode_t *zp, uint64_t end) 1654{ 1655 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1656 vnode_t *vp = ZTOV(zp); 1657 dmu_tx_t *tx; 1658 rl_t *rl; 1659 int error; 1660 sa_bulk_attr_t bulk[2]; 1661 int count = 0; 1662 1663 /* 1664 * We will change zp_size, lock the whole file. 1665 */ 1666 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1667 1668 /* 1669 * Nothing to do if file already at desired length. 1670 */ 1671 if (end >= zp->z_size) { 1672 zfs_range_unlock(rl); 1673 return (0); 1674 } 1675 1676 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1677 if (error) { 1678 zfs_range_unlock(rl); 1679 return (error); 1680 } 1681 tx = dmu_tx_create(zfsvfs->z_os); 1682 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1683 zfs_sa_upgrade_txholds(tx, zp); 1684 dmu_tx_mark_netfree(tx); 1685 error = dmu_tx_assign(tx, TXG_WAIT); 1686 if (error) { 1687 dmu_tx_abort(tx); 1688 zfs_range_unlock(rl); 1689 return (error); 1690 } 1691 1692 zp->z_size = end; 1693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1694 NULL, &zp->z_size, sizeof (zp->z_size)); 1695 1696 if (end == 0) { 1697 zp->z_pflags &= ~ZFS_SPARSE; 1698 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1699 NULL, &zp->z_pflags, 8); 1700 } 1701 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1702 1703 dmu_tx_commit(tx); 1704 1705 /* 1706 * Clear any mapped pages in the truncated region. This has to 1707 * happen outside of the transaction to avoid the possibility of 1708 * a deadlock with someone trying to push a page that we are 1709 * about to invalidate. 1710 */ 1711 vnode_pager_setsize(vp, end); 1712 1713 zfs_range_unlock(rl); 1714 1715 return (0); 1716} 1717 1718/* 1719 * Free space in a file 1720 * 1721 * IN: zp - znode of file to free data in. 1722 * off - start of range 1723 * len - end of range (0 => EOF) 1724 * flag - current file open mode flags. 1725 * log - TRUE if this action should be logged 1726 * 1727 * RETURN: 0 on success, error code on failure 1728 */ 1729int 1730zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1731{ 1732 vnode_t *vp = ZTOV(zp); 1733 dmu_tx_t *tx; 1734 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1735 zilog_t *zilog = zfsvfs->z_log; 1736 uint64_t mode; 1737 uint64_t mtime[2], ctime[2]; 1738 sa_bulk_attr_t bulk[3]; 1739 int count = 0; 1740 int error; 1741 1742 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1743 sizeof (mode))) != 0) 1744 return (error); 1745 1746 if (off > zp->z_size) { 1747 error = zfs_extend(zp, off+len); 1748 if (error == 0 && log) 1749 goto log; 1750 else 1751 return (error); 1752 } 1753 1754 /* 1755 * Check for any locks in the region to be freed. 1756 */ 1757 1758 if (MANDLOCK(vp, (mode_t)mode)) { 1759 uint64_t length = (len ? len : zp->z_size - off); 1760 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1761 return (error); 1762 } 1763 1764 if (len == 0) { 1765 error = zfs_trunc(zp, off); 1766 } else { 1767 if ((error = zfs_free_range(zp, off, len)) == 0 && 1768 off + len > zp->z_size) 1769 error = zfs_extend(zp, off+len); 1770 } 1771 if (error || !log) 1772 return (error); 1773log: 1774 tx = dmu_tx_create(zfsvfs->z_os); 1775 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1776 zfs_sa_upgrade_txholds(tx, zp); 1777 error = dmu_tx_assign(tx, TXG_WAIT); 1778 if (error) { 1779 dmu_tx_abort(tx); 1780 return (error); 1781 } 1782 1783 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1784 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1785 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1786 NULL, &zp->z_pflags, 8); 1787 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1788 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1789 ASSERT(error == 0); 1790 1791 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1792 1793 dmu_tx_commit(tx); 1794 return (0); 1795} 1796 1797void 1798zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1799{ 1800 uint64_t moid, obj, sa_obj, version; 1801 uint64_t sense = ZFS_CASE_SENSITIVE; 1802 uint64_t norm = 0; 1803 nvpair_t *elem; 1804 int error; 1805 int i; 1806 znode_t *rootzp = NULL; 1807 zfsvfs_t *zfsvfs; 1808 vattr_t vattr; 1809 znode_t *zp; 1810 zfs_acl_ids_t acl_ids; 1811 1812 /* 1813 * First attempt to create master node. 1814 */ 1815 /* 1816 * In an empty objset, there are no blocks to read and thus 1817 * there can be no i/o errors (which we assert below). 1818 */ 1819 moid = MASTER_NODE_OBJ; 1820 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1821 DMU_OT_NONE, 0, tx); 1822 ASSERT(error == 0); 1823 1824 /* 1825 * Set starting attributes. 1826 */ 1827 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1828 elem = NULL; 1829 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1830 /* For the moment we expect all zpl props to be uint64_ts */ 1831 uint64_t val; 1832 char *name; 1833 1834 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1835 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1836 name = nvpair_name(elem); 1837 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1838 if (val < version) 1839 version = val; 1840 } else { 1841 error = zap_update(os, moid, name, 8, 1, &val, tx); 1842 } 1843 ASSERT(error == 0); 1844 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1845 norm = val; 1846 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1847 sense = val; 1848 } 1849 ASSERT(version != 0); 1850 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1851 1852 /* 1853 * Create zap object used for SA attribute registration 1854 */ 1855 1856 if (version >= ZPL_VERSION_SA) { 1857 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1858 DMU_OT_NONE, 0, tx); 1859 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1860 ASSERT(error == 0); 1861 } else { 1862 sa_obj = 0; 1863 } 1864 /* 1865 * Create a delete queue. 1866 */ 1867 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1868 1869 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1870 ASSERT(error == 0); 1871 1872 /* 1873 * Create root znode. Create minimal znode/vnode/zfsvfs 1874 * to allow zfs_mknode to work. 1875 */ 1876 VATTR_NULL(&vattr); 1877 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1878 vattr.va_type = VDIR; 1879 vattr.va_mode = S_IFDIR|0755; 1880 vattr.va_uid = crgetuid(cr); 1881 vattr.va_gid = crgetgid(cr); 1882 1883 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 1884 1885 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1886 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1887 rootzp->z_moved = 0; 1888 rootzp->z_unlinked = 0; 1889 rootzp->z_atime_dirty = 0; 1890 rootzp->z_is_sa = USE_SA(version, os); 1891 1892 zfsvfs->z_os = os; 1893 zfsvfs->z_parent = zfsvfs; 1894 zfsvfs->z_version = version; 1895 zfsvfs->z_use_fuids = USE_FUIDS(version, os); 1896 zfsvfs->z_use_sa = USE_SA(version, os); 1897 zfsvfs->z_norm = norm; 1898 1899 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1900 &zfsvfs->z_attr_table); 1901 1902 ASSERT(error == 0); 1903 1904 /* 1905 * Fold case on file systems that are always or sometimes case 1906 * insensitive. 1907 */ 1908 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1909 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 1910 1911 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1912 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 1913 offsetof(znode_t, z_link_node)); 1914 1915 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1916 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1917 1918 rootzp->z_zfsvfs = zfsvfs; 1919 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1920 cr, NULL, &acl_ids)); 1921 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1922 ASSERT3P(zp, ==, rootzp); 1923 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1924 ASSERT(error == 0); 1925 zfs_acl_ids_free(&acl_ids); 1926 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1927 1928 sa_handle_destroy(rootzp->z_sa_hdl); 1929 kmem_cache_free(znode_cache, rootzp); 1930 1931 /* 1932 * Create shares directory 1933 */ 1934 1935 error = zfs_create_share_dir(zfsvfs, tx); 1936 1937 ASSERT(error == 0); 1938 1939 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1940 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1941 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1942} 1943 1944#endif /* _KERNEL */ 1945 1946static int 1947zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1948{ 1949 uint64_t sa_obj = 0; 1950 int error; 1951 1952 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1953 if (error != 0 && error != ENOENT) 1954 return (error); 1955 1956 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1957 return (error); 1958} 1959 1960static int 1961zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1962 dmu_buf_t **db, void *tag) 1963{ 1964 dmu_object_info_t doi; 1965 int error; 1966 1967 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 1968 return (error); 1969 1970 dmu_object_info_from_db(*db, &doi); 1971 if ((doi.doi_bonus_type != DMU_OT_SA && 1972 doi.doi_bonus_type != DMU_OT_ZNODE) || 1973 doi.doi_bonus_type == DMU_OT_ZNODE && 1974 doi.doi_bonus_size < sizeof (znode_phys_t)) { 1975 sa_buf_rele(*db, tag); 1976 return (SET_ERROR(ENOTSUP)); 1977 } 1978 1979 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 1980 if (error != 0) { 1981 sa_buf_rele(*db, tag); 1982 return (error); 1983 } 1984 1985 return (0); 1986} 1987 1988void 1989zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 1990{ 1991 sa_handle_destroy(hdl); 1992 sa_buf_rele(db, tag); 1993} 1994 1995/* 1996 * Given an object number, return its parent object number and whether 1997 * or not the object is an extended attribute directory. 1998 */ 1999static int 2000zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table, 2001 uint64_t *pobjp, int *is_xattrdir) 2002{ 2003 uint64_t parent; 2004 uint64_t pflags; 2005 uint64_t mode; 2006 uint64_t parent_mode; 2007 sa_bulk_attr_t bulk[3]; 2008 sa_handle_t *sa_hdl; 2009 dmu_buf_t *sa_db; 2010 int count = 0; 2011 int error; 2012 2013 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2014 &parent, sizeof (parent)); 2015 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2016 &pflags, sizeof (pflags)); 2017 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2018 &mode, sizeof (mode)); 2019 2020 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2021 return (error); 2022 2023 /* 2024 * When a link is removed its parent pointer is not changed and will 2025 * be invalid. There are two cases where a link is removed but the 2026 * file stays around, when it goes to the delete queue and when there 2027 * are additional links. 2028 */ 2029 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG); 2030 if (error != 0) 2031 return (error); 2032 2033 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode)); 2034 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2035 if (error != 0) 2036 return (error); 2037 2038 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2039 2040 /* 2041 * Extended attributes can be applied to files, directories, etc. 2042 * Otherwise the parent must be a directory. 2043 */ 2044 if (!*is_xattrdir && !S_ISDIR(parent_mode)) 2045 return (SET_ERROR(EINVAL)); 2046 2047 *pobjp = parent; 2048 2049 return (0); 2050} 2051 2052/* 2053 * Given an object number, return some zpl level statistics 2054 */ 2055static int 2056zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2057 zfs_stat_t *sb) 2058{ 2059 sa_bulk_attr_t bulk[4]; 2060 int count = 0; 2061 2062 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2063 &sb->zs_mode, sizeof (sb->zs_mode)); 2064 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2065 &sb->zs_gen, sizeof (sb->zs_gen)); 2066 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2067 &sb->zs_links, sizeof (sb->zs_links)); 2068 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2069 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2070 2071 return (sa_bulk_lookup(hdl, bulk, count)); 2072} 2073 2074static int 2075zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2076 sa_attr_type_t *sa_table, char *buf, int len) 2077{ 2078 sa_handle_t *sa_hdl; 2079 sa_handle_t *prevhdl = NULL; 2080 dmu_buf_t *prevdb = NULL; 2081 dmu_buf_t *sa_db = NULL; 2082 char *path = buf + len - 1; 2083 int error; 2084 2085 *path = '\0'; 2086 sa_hdl = hdl; 2087 2088 for (;;) { 2089 uint64_t pobj; 2090 char component[MAXNAMELEN + 2]; 2091 size_t complen; 2092 int is_xattrdir; 2093 2094 if (prevdb) 2095 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2096 2097 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj, 2098 &is_xattrdir)) != 0) 2099 break; 2100 2101 if (pobj == obj) { 2102 if (path[0] != '/') 2103 *--path = '/'; 2104 break; 2105 } 2106 2107 component[0] = '/'; 2108 if (is_xattrdir) { 2109 (void) sprintf(component + 1, "<xattrdir>"); 2110 } else { 2111 error = zap_value_search(osp, pobj, obj, 2112 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2113 if (error != 0) 2114 break; 2115 } 2116 2117 complen = strlen(component); 2118 path -= complen; 2119 ASSERT(path >= buf); 2120 bcopy(component, path, complen); 2121 obj = pobj; 2122 2123 if (sa_hdl != hdl) { 2124 prevhdl = sa_hdl; 2125 prevdb = sa_db; 2126 } 2127 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2128 if (error != 0) { 2129 sa_hdl = prevhdl; 2130 sa_db = prevdb; 2131 break; 2132 } 2133 } 2134 2135 if (sa_hdl != NULL && sa_hdl != hdl) { 2136 ASSERT(sa_db != NULL); 2137 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2138 } 2139 2140 if (error == 0) 2141 (void) memmove(buf, path, buf + len - path); 2142 2143 return (error); 2144} 2145 2146int 2147zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2148{ 2149 sa_attr_type_t *sa_table; 2150 sa_handle_t *hdl; 2151 dmu_buf_t *db; 2152 int error; 2153 2154 error = zfs_sa_setup(osp, &sa_table); 2155 if (error != 0) 2156 return (error); 2157 2158 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2159 if (error != 0) 2160 return (error); 2161 2162 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2163 2164 zfs_release_sa_handle(hdl, db, FTAG); 2165 return (error); 2166} 2167 2168int 2169zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2170 char *buf, int len) 2171{ 2172 char *path = buf + len - 1; 2173 sa_attr_type_t *sa_table; 2174 sa_handle_t *hdl; 2175 dmu_buf_t *db; 2176 int error; 2177 2178 *path = '\0'; 2179 2180 error = zfs_sa_setup(osp, &sa_table); 2181 if (error != 0) 2182 return (error); 2183 2184 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2185 if (error != 0) 2186 return (error); 2187 2188 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2189 if (error != 0) { 2190 zfs_release_sa_handle(hdl, db, FTAG); 2191 return (error); 2192 } 2193 2194 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2195 2196 zfs_release_sa_handle(hdl, db, FTAG); 2197 return (error); 2198} 2199