zfs_znode.c revision 307996
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 * Copyright (c) 2014 Integros [integros.com] 25 */ 26 27/* Portions Copyright 2007 Jeremy Teo */ 28/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 29 30#ifdef _KERNEL 31#include <sys/types.h> 32#include <sys/param.h> 33#include <sys/time.h> 34#include <sys/systm.h> 35#include <sys/sysmacros.h> 36#include <sys/resource.h> 37#include <sys/mntent.h> 38#include <sys/u8_textprep.h> 39#include <sys/dsl_dataset.h> 40#include <sys/vfs.h> 41#include <sys/vnode.h> 42#include <sys/file.h> 43#include <sys/kmem.h> 44#include <sys/errno.h> 45#include <sys/unistd.h> 46#include <sys/atomic.h> 47#include <sys/zfs_dir.h> 48#include <sys/zfs_acl.h> 49#include <sys/zfs_ioctl.h> 50#include <sys/zfs_rlock.h> 51#include <sys/zfs_fuid.h> 52#include <sys/dnode.h> 53#include <sys/fs/zfs.h> 54#include <sys/kidmap.h> 55#endif /* _KERNEL */ 56 57#include <sys/dmu.h> 58#include <sys/dmu_objset.h> 59#include <sys/refcount.h> 60#include <sys/stat.h> 61#include <sys/zap.h> 62#include <sys/zfs_znode.h> 63#include <sys/sa.h> 64#include <sys/zfs_sa.h> 65#include <sys/zfs_stat.h> 66#include <sys/refcount.h> 67 68#include "zfs_prop.h" 69#include "zfs_comutil.h" 70 71/* Used by fstat(1). */ 72SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 73 SYSCTL_NULL_INT_PTR, sizeof(znode_t), "sizeof(znode_t)"); 74 75/* 76 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 77 * turned on when DEBUG is also defined. 78 */ 79#ifdef DEBUG 80#define ZNODE_STATS 81#endif /* DEBUG */ 82 83#ifdef ZNODE_STATS 84#define ZNODE_STAT_ADD(stat) ((stat)++) 85#else 86#define ZNODE_STAT_ADD(stat) /* nothing */ 87#endif /* ZNODE_STATS */ 88 89/* 90 * Functions needed for userland (ie: libzpool) are not put under 91 * #ifdef_KERNEL; the rest of the functions have dependencies 92 * (such as VFS logic) that will not compile easily in userland. 93 */ 94#ifdef _KERNEL 95/* 96 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 97 * be freed before it can be safely accessed. 98 */ 99krwlock_t zfsvfs_lock; 100 101static kmem_cache_t *znode_cache = NULL; 102 103/*ARGSUSED*/ 104static void 105znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 106{ 107 /* 108 * We should never drop all dbuf refs without first clearing 109 * the eviction callback. 110 */ 111 panic("evicting znode %p\n", user_ptr); 112} 113 114extern struct vop_vector zfs_vnodeops; 115extern struct vop_vector zfs_fifoops; 116extern struct vop_vector zfs_shareops; 117 118static int 119zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 120{ 121 znode_t *zp = buf; 122 123 POINTER_INVALIDATE(&zp->z_zfsvfs); 124 125 list_link_init(&zp->z_link_node); 126 127 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 128 129 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 130 avl_create(&zp->z_range_avl, zfs_range_compare, 131 sizeof (rl_t), offsetof(rl_t, r_node)); 132 133 zp->z_acl_cached = NULL; 134 zp->z_vnode = NULL; 135 zp->z_moved = 0; 136 return (0); 137} 138 139/*ARGSUSED*/ 140static void 141zfs_znode_cache_destructor(void *buf, void *arg) 142{ 143 znode_t *zp = buf; 144 145 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 146 ASSERT(ZTOV(zp) == NULL); 147 vn_free(ZTOV(zp)); 148 ASSERT(!list_link_active(&zp->z_link_node)); 149 mutex_destroy(&zp->z_acl_lock); 150 avl_destroy(&zp->z_range_avl); 151 mutex_destroy(&zp->z_range_lock); 152 153 ASSERT(zp->z_acl_cached == NULL); 154} 155 156#ifdef ZNODE_STATS 157static struct { 158 uint64_t zms_zfsvfs_invalid; 159 uint64_t zms_zfsvfs_recheck1; 160 uint64_t zms_zfsvfs_unmounted; 161 uint64_t zms_zfsvfs_recheck2; 162 uint64_t zms_obj_held; 163 uint64_t zms_vnode_locked; 164 uint64_t zms_not_only_dnlc; 165} znode_move_stats; 166#endif /* ZNODE_STATS */ 167 168#ifdef illumos 169static void 170zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 171{ 172 vnode_t *vp; 173 174 /* Copy fields. */ 175 nzp->z_zfsvfs = ozp->z_zfsvfs; 176 177 /* Swap vnodes. */ 178 vp = nzp->z_vnode; 179 nzp->z_vnode = ozp->z_vnode; 180 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 181 ZTOV(ozp)->v_data = ozp; 182 ZTOV(nzp)->v_data = nzp; 183 184 nzp->z_id = ozp->z_id; 185 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 186 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 187 nzp->z_unlinked = ozp->z_unlinked; 188 nzp->z_atime_dirty = ozp->z_atime_dirty; 189 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 190 nzp->z_blksz = ozp->z_blksz; 191 nzp->z_seq = ozp->z_seq; 192 nzp->z_mapcnt = ozp->z_mapcnt; 193 nzp->z_gen = ozp->z_gen; 194 nzp->z_sync_cnt = ozp->z_sync_cnt; 195 nzp->z_is_sa = ozp->z_is_sa; 196 nzp->z_sa_hdl = ozp->z_sa_hdl; 197 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 198 nzp->z_links = ozp->z_links; 199 nzp->z_size = ozp->z_size; 200 nzp->z_pflags = ozp->z_pflags; 201 nzp->z_uid = ozp->z_uid; 202 nzp->z_gid = ozp->z_gid; 203 nzp->z_mode = ozp->z_mode; 204 205 /* 206 * Since this is just an idle znode and kmem is already dealing with 207 * memory pressure, release any cached ACL. 208 */ 209 if (ozp->z_acl_cached) { 210 zfs_acl_free(ozp->z_acl_cached); 211 ozp->z_acl_cached = NULL; 212 } 213 214 sa_set_userp(nzp->z_sa_hdl, nzp); 215 216 /* 217 * Invalidate the original znode by clearing fields that provide a 218 * pointer back to the znode. Set the low bit of the vfs pointer to 219 * ensure that zfs_znode_move() recognizes the znode as invalid in any 220 * subsequent callback. 221 */ 222 ozp->z_sa_hdl = NULL; 223 POINTER_INVALIDATE(&ozp->z_zfsvfs); 224 225 /* 226 * Mark the znode. 227 */ 228 nzp->z_moved = 1; 229 ozp->z_moved = (uint8_t)-1; 230} 231 232/*ARGSUSED*/ 233static kmem_cbrc_t 234zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 235{ 236 znode_t *ozp = buf, *nzp = newbuf; 237 zfsvfs_t *zfsvfs; 238 vnode_t *vp; 239 240 /* 241 * The znode is on the file system's list of known znodes if the vfs 242 * pointer is valid. We set the low bit of the vfs pointer when freeing 243 * the znode to invalidate it, and the memory patterns written by kmem 244 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 245 * created znode sets the vfs pointer last of all to indicate that the 246 * znode is known and in a valid state to be moved by this function. 247 */ 248 zfsvfs = ozp->z_zfsvfs; 249 if (!POINTER_IS_VALID(zfsvfs)) { 250 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 251 return (KMEM_CBRC_DONT_KNOW); 252 } 253 254 /* 255 * Close a small window in which it's possible that the filesystem could 256 * be unmounted and freed, and zfsvfs, though valid in the previous 257 * statement, could point to unrelated memory by the time we try to 258 * prevent the filesystem from being unmounted. 259 */ 260 rw_enter(&zfsvfs_lock, RW_WRITER); 261 if (zfsvfs != ozp->z_zfsvfs) { 262 rw_exit(&zfsvfs_lock); 263 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 264 return (KMEM_CBRC_DONT_KNOW); 265 } 266 267 /* 268 * If the znode is still valid, then so is the file system. We know that 269 * no valid file system can be freed while we hold zfsvfs_lock, so we 270 * can safely ensure that the filesystem is not and will not be 271 * unmounted. The next statement is equivalent to ZFS_ENTER(). 272 */ 273 rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 274 if (zfsvfs->z_unmounted) { 275 ZFS_EXIT(zfsvfs); 276 rw_exit(&zfsvfs_lock); 277 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 278 return (KMEM_CBRC_DONT_KNOW); 279 } 280 rw_exit(&zfsvfs_lock); 281 282 mutex_enter(&zfsvfs->z_znodes_lock); 283 /* 284 * Recheck the vfs pointer in case the znode was removed just before 285 * acquiring the lock. 286 */ 287 if (zfsvfs != ozp->z_zfsvfs) { 288 mutex_exit(&zfsvfs->z_znodes_lock); 289 ZFS_EXIT(zfsvfs); 290 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 291 return (KMEM_CBRC_DONT_KNOW); 292 } 293 294 /* 295 * At this point we know that as long as we hold z_znodes_lock, the 296 * znode cannot be freed and fields within the znode can be safely 297 * accessed. Now, prevent a race with zfs_zget(). 298 */ 299 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 300 mutex_exit(&zfsvfs->z_znodes_lock); 301 ZFS_EXIT(zfsvfs); 302 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 303 return (KMEM_CBRC_LATER); 304 } 305 306 vp = ZTOV(ozp); 307 if (mutex_tryenter(&vp->v_lock) == 0) { 308 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 309 mutex_exit(&zfsvfs->z_znodes_lock); 310 ZFS_EXIT(zfsvfs); 311 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 312 return (KMEM_CBRC_LATER); 313 } 314 315 /* Only move znodes that are referenced _only_ by the DNLC. */ 316 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 317 mutex_exit(&vp->v_lock); 318 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 319 mutex_exit(&zfsvfs->z_znodes_lock); 320 ZFS_EXIT(zfsvfs); 321 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 322 return (KMEM_CBRC_LATER); 323 } 324 325 /* 326 * The znode is known and in a valid state to move. We're holding the 327 * locks needed to execute the critical section. 328 */ 329 zfs_znode_move_impl(ozp, nzp); 330 mutex_exit(&vp->v_lock); 331 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 332 333 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 334 mutex_exit(&zfsvfs->z_znodes_lock); 335 ZFS_EXIT(zfsvfs); 336 337 return (KMEM_CBRC_YES); 338} 339#endif /* illumos */ 340 341void 342zfs_znode_init(void) 343{ 344 /* 345 * Initialize zcache 346 */ 347 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 348 ASSERT(znode_cache == NULL); 349 znode_cache = kmem_cache_create("zfs_znode_cache", 350 sizeof (znode_t), 0, zfs_znode_cache_constructor, 351 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 352 kmem_cache_set_move(znode_cache, zfs_znode_move); 353} 354 355void 356zfs_znode_fini(void) 357{ 358#ifdef illumos 359 /* 360 * Cleanup vfs & vnode ops 361 */ 362 zfs_remove_op_tables(); 363#endif 364 365 /* 366 * Cleanup zcache 367 */ 368 if (znode_cache) 369 kmem_cache_destroy(znode_cache); 370 znode_cache = NULL; 371 rw_destroy(&zfsvfs_lock); 372} 373 374#ifdef illumos 375struct vnodeops *zfs_dvnodeops; 376struct vnodeops *zfs_fvnodeops; 377struct vnodeops *zfs_symvnodeops; 378struct vnodeops *zfs_xdvnodeops; 379struct vnodeops *zfs_evnodeops; 380struct vnodeops *zfs_sharevnodeops; 381 382void 383zfs_remove_op_tables() 384{ 385 /* 386 * Remove vfs ops 387 */ 388 ASSERT(zfsfstype); 389 (void) vfs_freevfsops_by_type(zfsfstype); 390 zfsfstype = 0; 391 392 /* 393 * Remove vnode ops 394 */ 395 if (zfs_dvnodeops) 396 vn_freevnodeops(zfs_dvnodeops); 397 if (zfs_fvnodeops) 398 vn_freevnodeops(zfs_fvnodeops); 399 if (zfs_symvnodeops) 400 vn_freevnodeops(zfs_symvnodeops); 401 if (zfs_xdvnodeops) 402 vn_freevnodeops(zfs_xdvnodeops); 403 if (zfs_evnodeops) 404 vn_freevnodeops(zfs_evnodeops); 405 if (zfs_sharevnodeops) 406 vn_freevnodeops(zfs_sharevnodeops); 407 408 zfs_dvnodeops = NULL; 409 zfs_fvnodeops = NULL; 410 zfs_symvnodeops = NULL; 411 zfs_xdvnodeops = NULL; 412 zfs_evnodeops = NULL; 413 zfs_sharevnodeops = NULL; 414} 415 416extern const fs_operation_def_t zfs_dvnodeops_template[]; 417extern const fs_operation_def_t zfs_fvnodeops_template[]; 418extern const fs_operation_def_t zfs_xdvnodeops_template[]; 419extern const fs_operation_def_t zfs_symvnodeops_template[]; 420extern const fs_operation_def_t zfs_evnodeops_template[]; 421extern const fs_operation_def_t zfs_sharevnodeops_template[]; 422 423int 424zfs_create_op_tables() 425{ 426 int error; 427 428 /* 429 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 430 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 431 * In this case we just return as the ops vectors are already set up. 432 */ 433 if (zfs_dvnodeops) 434 return (0); 435 436 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 437 &zfs_dvnodeops); 438 if (error) 439 return (error); 440 441 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 442 &zfs_fvnodeops); 443 if (error) 444 return (error); 445 446 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 447 &zfs_symvnodeops); 448 if (error) 449 return (error); 450 451 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 452 &zfs_xdvnodeops); 453 if (error) 454 return (error); 455 456 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 457 &zfs_evnodeops); 458 if (error) 459 return (error); 460 461 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 462 &zfs_sharevnodeops); 463 464 return (error); 465} 466#endif /* illumos */ 467 468int 469zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 470{ 471 zfs_acl_ids_t acl_ids; 472 vattr_t vattr; 473 znode_t *sharezp; 474 znode_t *zp; 475 int error; 476 477 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 478 vattr.va_type = VDIR; 479 vattr.va_mode = S_IFDIR|0555; 480 vattr.va_uid = crgetuid(kcred); 481 vattr.va_gid = crgetgid(kcred); 482 483 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 484 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 485 sharezp->z_moved = 0; 486 sharezp->z_unlinked = 0; 487 sharezp->z_atime_dirty = 0; 488 sharezp->z_zfsvfs = zfsvfs; 489 sharezp->z_is_sa = zfsvfs->z_use_sa; 490 491 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 492 kcred, NULL, &acl_ids)); 493 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 494 ASSERT3P(zp, ==, sharezp); 495 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 496 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 497 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 498 zfsvfs->z_shares_dir = sharezp->z_id; 499 500 zfs_acl_ids_free(&acl_ids); 501 sa_handle_destroy(sharezp->z_sa_hdl); 502 kmem_cache_free(znode_cache, sharezp); 503 504 return (error); 505} 506 507/* 508 * define a couple of values we need available 509 * for both 64 and 32 bit environments. 510 */ 511#ifndef NBITSMINOR64 512#define NBITSMINOR64 32 513#endif 514#ifndef MAXMAJ64 515#define MAXMAJ64 0xffffffffUL 516#endif 517#ifndef MAXMIN64 518#define MAXMIN64 0xffffffffUL 519#endif 520 521/* 522 * Create special expldev for ZFS private use. 523 * Can't use standard expldev since it doesn't do 524 * what we want. The standard expldev() takes a 525 * dev32_t in LP64 and expands it to a long dev_t. 526 * We need an interface that takes a dev32_t in ILP32 527 * and expands it to a long dev_t. 528 */ 529static uint64_t 530zfs_expldev(dev_t dev) 531{ 532 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 533} 534/* 535 * Special cmpldev for ZFS private use. 536 * Can't use standard cmpldev since it takes 537 * a long dev_t and compresses it to dev32_t in 538 * LP64. We need to do a compaction of a long dev_t 539 * to a dev32_t in ILP32. 540 */ 541dev_t 542zfs_cmpldev(uint64_t dev) 543{ 544 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 545} 546 547static void 548zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 549 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 550{ 551 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 552 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 553 554 ASSERT(zp->z_sa_hdl == NULL); 555 ASSERT(zp->z_acl_cached == NULL); 556 if (sa_hdl == NULL) { 557 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 558 SA_HDL_SHARED, &zp->z_sa_hdl)); 559 } else { 560 zp->z_sa_hdl = sa_hdl; 561 sa_set_userp(sa_hdl, zp); 562 } 563 564 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 565 566 /* 567 * Slap on VROOT if we are the root znode unless we are the root 568 * node of a snapshot mounted under .zfs. 569 */ 570 if (zp->z_id == zfsvfs->z_root && zfsvfs->z_parent == zfsvfs) 571 ZTOV(zp)->v_flag |= VROOT; 572 573 vn_exists(ZTOV(zp)); 574} 575 576void 577zfs_znode_dmu_fini(znode_t *zp) 578{ 579 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 580 zp->z_unlinked || 581 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 582 583 sa_handle_destroy(zp->z_sa_hdl); 584 zp->z_sa_hdl = NULL; 585} 586 587static void 588zfs_vnode_forget(vnode_t *vp) 589{ 590 591 /* copied from insmntque_stddtr */ 592 vp->v_data = NULL; 593 vp->v_op = &dead_vnodeops; 594 vgone(vp); 595 vput(vp); 596} 597 598/* 599 * Construct a new znode/vnode and intialize. 600 * 601 * This does not do a call to dmu_set_user() that is 602 * up to the caller to do, in case you don't want to 603 * return the znode 604 */ 605static znode_t * 606zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 607 dmu_object_type_t obj_type, sa_handle_t *hdl) 608{ 609 znode_t *zp; 610 vnode_t *vp; 611 uint64_t mode; 612 uint64_t parent; 613 sa_bulk_attr_t bulk[9]; 614 int count = 0; 615 int error; 616 617 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 618 619 KASSERT(curthread->td_vp_reserv > 0, 620 ("zfs_znode_alloc: getnewvnode without any vnodes reserved")); 621 error = getnewvnode("zfs", zfsvfs->z_parent->z_vfs, &zfs_vnodeops, &vp); 622 if (error != 0) { 623 kmem_cache_free(znode_cache, zp); 624 return (NULL); 625 } 626 zp->z_vnode = vp; 627 vp->v_data = zp; 628 629 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 630 zp->z_moved = 0; 631 632 /* 633 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 634 * the zfs_znode_move() callback. 635 */ 636 zp->z_sa_hdl = NULL; 637 zp->z_unlinked = 0; 638 zp->z_atime_dirty = 0; 639 zp->z_mapcnt = 0; 640 zp->z_id = db->db_object; 641 zp->z_blksz = blksz; 642 zp->z_seq = 0x7A4653; 643 zp->z_sync_cnt = 0; 644 645 vp = ZTOV(zp); 646 647 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 648 649 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 650 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 651 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 652 &zp->z_size, 8); 653 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 654 &zp->z_links, 8); 655 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 656 &zp->z_pflags, 8); 657 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 658 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 659 &zp->z_atime, 16); 660 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 661 &zp->z_uid, 8); 662 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 663 &zp->z_gid, 8); 664 665 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 666 if (hdl == NULL) 667 sa_handle_destroy(zp->z_sa_hdl); 668 zfs_vnode_forget(vp); 669 zp->z_vnode = NULL; 670 kmem_cache_free(znode_cache, zp); 671 return (NULL); 672 } 673 674 zp->z_mode = mode; 675 676 vp->v_type = IFTOVT((mode_t)mode); 677 678 switch (vp->v_type) { 679 case VDIR: 680 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 681 break; 682#ifdef illumos 683 case VBLK: 684 case VCHR: 685 { 686 uint64_t rdev; 687 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 688 &rdev, sizeof (rdev)) == 0); 689 690 vp->v_rdev = zfs_cmpldev(rdev); 691 } 692 break; 693#endif 694 case VFIFO: 695#ifdef illumos 696 case VSOCK: 697 case VDOOR: 698#endif 699 vp->v_op = &zfs_fifoops; 700 break; 701 case VREG: 702 if (parent == zfsvfs->z_shares_dir) { 703 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 704 vp->v_op = &zfs_shareops; 705 } 706 break; 707#ifdef illumos 708 case VLNK: 709 vn_setops(vp, zfs_symvnodeops); 710 break; 711 default: 712 vn_setops(vp, zfs_evnodeops); 713 break; 714#endif 715 } 716 717 mutex_enter(&zfsvfs->z_znodes_lock); 718 list_insert_tail(&zfsvfs->z_all_znodes, zp); 719 membar_producer(); 720 /* 721 * Everything else must be valid before assigning z_zfsvfs makes the 722 * znode eligible for zfs_znode_move(). 723 */ 724 zp->z_zfsvfs = zfsvfs; 725 mutex_exit(&zfsvfs->z_znodes_lock); 726 727 /* 728 * Acquire vnode lock before making it available to the world. 729 */ 730#ifdef DIAGNOSTIC 731 vop_lock1_t *orig_lock = vp->v_op->vop_lock1; 732 vp->v_op->vop_lock1 = vop_stdlock; 733 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 734 vp->v_op->vop_lock1 = orig_lock; 735#else 736 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 737#endif 738 VN_LOCK_AREC(vp); 739 if (vp->v_type != VFIFO) 740 VN_LOCK_ASHARE(vp); 741 742#ifdef illumos 743 VFS_HOLD(zfsvfs->z_vfs); 744#endif 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 vfs_timestamp(&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 * Since "SA" does immediate eviction we 1161 * should never find a sa handle that doesn't 1162 * know about the znode. 1163 */ 1164 ASSERT3P(zp, !=, NULL); 1165 ASSERT3U(zp->z_id, ==, obj_num); 1166 *zpp = zp; 1167 vp = ZTOV(zp); 1168 1169 /* Don't let the vnode disappear after ZFS_OBJ_HOLD_EXIT. */ 1170 VN_HOLD(vp); 1171 1172 sa_buf_rele(db, NULL); 1173 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1174 1175 locked = VOP_ISLOCKED(vp); 1176 VI_LOCK(vp); 1177 if ((vp->v_iflag & VI_DOOMED) != 0 && 1178 locked != LK_EXCLUSIVE) { 1179 /* 1180 * The vnode is doomed and this thread doesn't 1181 * hold the exclusive lock on it, so the vnode 1182 * must be being reclaimed by another thread. 1183 * Otherwise the doomed vnode is being reclaimed 1184 * by this thread and zfs_zget is called from 1185 * ZIL internals. 1186 */ 1187 VI_UNLOCK(vp); 1188 1189 /* 1190 * XXX vrele() locks the vnode when the last reference 1191 * is dropped. Although in this case the vnode is 1192 * doomed / dead and so no inactivation is required, 1193 * the vnode lock is still acquired. That could result 1194 * in a LOR with z_teardown_lock if another thread holds 1195 * the vnode's lock and tries to take z_teardown_lock. 1196 * But that is only possible if the other thread peforms 1197 * a ZFS vnode operation on the vnode. That either 1198 * should not happen if the vnode is dead or the thread 1199 * should also have a refrence to the vnode and thus 1200 * our reference is not last. 1201 */ 1202 VN_RELE(vp); 1203 goto again; 1204 } 1205 VI_UNLOCK(vp); 1206 getnewvnode_drop_reserve(); 1207 return (0); 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->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 vn_pages_remove(vp, 0, 0); 1350 if (zp->z_size != size) 1351 vnode_pager_setsize(vp, zp->z_size); 1352 1353 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1354 1355 return (0); 1356} 1357 1358void 1359zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1360{ 1361 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1362 objset_t *os = zfsvfs->z_os; 1363 uint64_t obj = zp->z_id; 1364 uint64_t acl_obj = zfs_external_acl(zp); 1365 1366 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1367 if (acl_obj) { 1368 VERIFY(!zp->z_is_sa); 1369 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1370 } 1371 VERIFY(0 == dmu_object_free(os, obj, tx)); 1372 zfs_znode_dmu_fini(zp); 1373 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1374 zfs_znode_free(zp); 1375} 1376 1377void 1378zfs_zinactive(znode_t *zp) 1379{ 1380 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1381 uint64_t z_id = zp->z_id; 1382 1383 ASSERT(zp->z_sa_hdl); 1384 1385 /* 1386 * Don't allow a zfs_zget() while were trying to release this znode 1387 */ 1388 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1389 1390 /* 1391 * If this was the last reference to a file with no links, 1392 * remove the file from the file system. 1393 */ 1394 if (zp->z_unlinked) { 1395 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1396 zfs_rmnode(zp); 1397 return; 1398 } 1399 1400 zfs_znode_dmu_fini(zp); 1401 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1402 zfs_znode_free(zp); 1403} 1404 1405void 1406zfs_znode_free(znode_t *zp) 1407{ 1408 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1409 1410 ASSERT(zp->z_sa_hdl == NULL); 1411 zp->z_vnode = NULL; 1412 mutex_enter(&zfsvfs->z_znodes_lock); 1413 POINTER_INVALIDATE(&zp->z_zfsvfs); 1414 list_remove(&zfsvfs->z_all_znodes, zp); 1415 mutex_exit(&zfsvfs->z_znodes_lock); 1416 1417 if (zp->z_acl_cached) { 1418 zfs_acl_free(zp->z_acl_cached); 1419 zp->z_acl_cached = NULL; 1420 } 1421 1422 kmem_cache_free(znode_cache, zp); 1423 1424#ifdef illumos 1425 VFS_RELE(zfsvfs->z_vfs); 1426#endif 1427} 1428 1429void 1430zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1431 uint64_t ctime[2], boolean_t have_tx) 1432{ 1433 timestruc_t now; 1434 1435 vfs_timestamp(&now); 1436 1437 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1438 zp->z_atime_dirty = 0; 1439 zp->z_seq++; 1440 } else { 1441 zp->z_atime_dirty = 1; 1442 } 1443 1444 if (flag & AT_ATIME) { 1445 ZFS_TIME_ENCODE(&now, zp->z_atime); 1446 } 1447 1448 if (flag & AT_MTIME) { 1449 ZFS_TIME_ENCODE(&now, mtime); 1450 if (zp->z_zfsvfs->z_use_fuids) { 1451 zp->z_pflags |= (ZFS_ARCHIVE | 1452 ZFS_AV_MODIFIED); 1453 } 1454 } 1455 1456 if (flag & AT_CTIME) { 1457 ZFS_TIME_ENCODE(&now, ctime); 1458 if (zp->z_zfsvfs->z_use_fuids) 1459 zp->z_pflags |= ZFS_ARCHIVE; 1460 } 1461} 1462 1463/* 1464 * Grow the block size for a file. 1465 * 1466 * IN: zp - znode of file to free data in. 1467 * size - requested block size 1468 * tx - open transaction. 1469 * 1470 * NOTE: this function assumes that the znode is write locked. 1471 */ 1472void 1473zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1474{ 1475 int error; 1476 u_longlong_t dummy; 1477 1478 if (size <= zp->z_blksz) 1479 return; 1480 /* 1481 * If the file size is already greater than the current blocksize, 1482 * we will not grow. If there is more than one block in a file, 1483 * the blocksize cannot change. 1484 */ 1485 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1486 return; 1487 1488 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1489 size, 0, tx); 1490 1491 if (error == ENOTSUP) 1492 return; 1493 ASSERT0(error); 1494 1495 /* What blocksize did we actually get? */ 1496 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1497} 1498 1499#ifdef illumos 1500/* 1501 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1502 * be calling back into the fs for a putpage(). E.g.: when truncating 1503 * a file, the pages being "thrown away* don't need to be written out. 1504 */ 1505/* ARGSUSED */ 1506static int 1507zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1508 int flags, cred_t *cr) 1509{ 1510 ASSERT(0); 1511 return (0); 1512} 1513#endif 1514 1515/* 1516 * Increase the file length 1517 * 1518 * IN: zp - znode of file to free data in. 1519 * end - new end-of-file 1520 * 1521 * RETURN: 0 on success, error code on failure 1522 */ 1523static int 1524zfs_extend(znode_t *zp, uint64_t end) 1525{ 1526 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1527 dmu_tx_t *tx; 1528 rl_t *rl; 1529 uint64_t newblksz; 1530 int error; 1531 1532 /* 1533 * We will change zp_size, lock the whole file. 1534 */ 1535 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1536 1537 /* 1538 * Nothing to do if file already at desired length. 1539 */ 1540 if (end <= zp->z_size) { 1541 zfs_range_unlock(rl); 1542 return (0); 1543 } 1544 tx = dmu_tx_create(zfsvfs->z_os); 1545 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1546 zfs_sa_upgrade_txholds(tx, zp); 1547 if (end > zp->z_blksz && 1548 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1549 /* 1550 * We are growing the file past the current block size. 1551 */ 1552 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1553 /* 1554 * File's blocksize is already larger than the 1555 * "recordsize" property. Only let it grow to 1556 * the next power of 2. 1557 */ 1558 ASSERT(!ISP2(zp->z_blksz)); 1559 newblksz = MIN(end, 1 << highbit64(zp->z_blksz)); 1560 } else { 1561 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1562 } 1563 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1564 } else { 1565 newblksz = 0; 1566 } 1567 1568 error = dmu_tx_assign(tx, TXG_WAIT); 1569 if (error) { 1570 dmu_tx_abort(tx); 1571 zfs_range_unlock(rl); 1572 return (error); 1573 } 1574 1575 if (newblksz) 1576 zfs_grow_blocksize(zp, newblksz, tx); 1577 1578 zp->z_size = end; 1579 1580 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1581 &zp->z_size, sizeof (zp->z_size), tx)); 1582 1583 vnode_pager_setsize(ZTOV(zp), end); 1584 1585 zfs_range_unlock(rl); 1586 1587 dmu_tx_commit(tx); 1588 1589 return (0); 1590} 1591 1592/* 1593 * Free space in a file. 1594 * 1595 * IN: zp - znode of file to free data in. 1596 * off - start of section to free. 1597 * len - length of section to free. 1598 * 1599 * RETURN: 0 on success, error code on failure 1600 */ 1601static int 1602zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1603{ 1604 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1605 rl_t *rl; 1606 int error; 1607 1608 /* 1609 * Lock the range being freed. 1610 */ 1611 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1612 1613 /* 1614 * Nothing to do if file already at desired length. 1615 */ 1616 if (off >= zp->z_size) { 1617 zfs_range_unlock(rl); 1618 return (0); 1619 } 1620 1621 if (off + len > zp->z_size) 1622 len = zp->z_size - off; 1623 1624 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1625 1626 if (error == 0) { 1627 /* 1628 * In FreeBSD we cannot free block in the middle of a file, 1629 * but only at the end of a file, so this code path should 1630 * never happen. 1631 */ 1632 vnode_pager_setsize(ZTOV(zp), off); 1633 } 1634 1635 zfs_range_unlock(rl); 1636 1637 return (error); 1638} 1639 1640/* 1641 * Truncate a file 1642 * 1643 * IN: zp - znode of file to free data in. 1644 * end - new end-of-file. 1645 * 1646 * RETURN: 0 on success, error code on failure 1647 */ 1648static int 1649zfs_trunc(znode_t *zp, uint64_t end) 1650{ 1651 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1652 vnode_t *vp = ZTOV(zp); 1653 dmu_tx_t *tx; 1654 rl_t *rl; 1655 int error; 1656 sa_bulk_attr_t bulk[2]; 1657 int count = 0; 1658 1659 /* 1660 * We will change zp_size, lock the whole file. 1661 */ 1662 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1663 1664 /* 1665 * Nothing to do if file already at desired length. 1666 */ 1667 if (end >= zp->z_size) { 1668 zfs_range_unlock(rl); 1669 return (0); 1670 } 1671 1672 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1673 if (error) { 1674 zfs_range_unlock(rl); 1675 return (error); 1676 } 1677 tx = dmu_tx_create(zfsvfs->z_os); 1678 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1679 zfs_sa_upgrade_txholds(tx, zp); 1680 dmu_tx_mark_netfree(tx); 1681 error = dmu_tx_assign(tx, TXG_WAIT); 1682 if (error) { 1683 dmu_tx_abort(tx); 1684 zfs_range_unlock(rl); 1685 return (error); 1686 } 1687 1688 zp->z_size = end; 1689 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1690 NULL, &zp->z_size, sizeof (zp->z_size)); 1691 1692 if (end == 0) { 1693 zp->z_pflags &= ~ZFS_SPARSE; 1694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1695 NULL, &zp->z_pflags, 8); 1696 } 1697 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1698 1699 dmu_tx_commit(tx); 1700 1701 /* 1702 * Clear any mapped pages in the truncated region. This has to 1703 * happen outside of the transaction to avoid the possibility of 1704 * a deadlock with someone trying to push a page that we are 1705 * about to invalidate. 1706 */ 1707 vnode_pager_setsize(vp, end); 1708 1709 zfs_range_unlock(rl); 1710 1711 return (0); 1712} 1713 1714/* 1715 * Free space in a file 1716 * 1717 * IN: zp - znode of file to free data in. 1718 * off - start of range 1719 * len - end of range (0 => EOF) 1720 * flag - current file open mode flags. 1721 * log - TRUE if this action should be logged 1722 * 1723 * RETURN: 0 on success, error code on failure 1724 */ 1725int 1726zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1727{ 1728 vnode_t *vp = ZTOV(zp); 1729 dmu_tx_t *tx; 1730 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1731 zilog_t *zilog = zfsvfs->z_log; 1732 uint64_t mode; 1733 uint64_t mtime[2], ctime[2]; 1734 sa_bulk_attr_t bulk[3]; 1735 int count = 0; 1736 int error; 1737 1738 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1739 sizeof (mode))) != 0) 1740 return (error); 1741 1742 if (off > zp->z_size) { 1743 error = zfs_extend(zp, off+len); 1744 if (error == 0 && log) 1745 goto log; 1746 else 1747 return (error); 1748 } 1749 1750 /* 1751 * Check for any locks in the region to be freed. 1752 */ 1753 1754 if (MANDLOCK(vp, (mode_t)mode)) { 1755 uint64_t length = (len ? len : zp->z_size - off); 1756 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1757 return (error); 1758 } 1759 1760 if (len == 0) { 1761 error = zfs_trunc(zp, off); 1762 } else { 1763 if ((error = zfs_free_range(zp, off, len)) == 0 && 1764 off + len > zp->z_size) 1765 error = zfs_extend(zp, off+len); 1766 } 1767 if (error || !log) 1768 return (error); 1769log: 1770 tx = dmu_tx_create(zfsvfs->z_os); 1771 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1772 zfs_sa_upgrade_txholds(tx, zp); 1773 error = dmu_tx_assign(tx, TXG_WAIT); 1774 if (error) { 1775 dmu_tx_abort(tx); 1776 return (error); 1777 } 1778 1779 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1780 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1781 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1782 NULL, &zp->z_pflags, 8); 1783 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1784 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1785 ASSERT(error == 0); 1786 1787 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1788 1789 dmu_tx_commit(tx); 1790 return (0); 1791} 1792 1793void 1794zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1795{ 1796 uint64_t moid, obj, sa_obj, version; 1797 uint64_t sense = ZFS_CASE_SENSITIVE; 1798 uint64_t norm = 0; 1799 nvpair_t *elem; 1800 int error; 1801 int i; 1802 znode_t *rootzp = NULL; 1803 zfsvfs_t *zfsvfs; 1804 vattr_t vattr; 1805 znode_t *zp; 1806 zfs_acl_ids_t acl_ids; 1807 1808 /* 1809 * First attempt to create master node. 1810 */ 1811 /* 1812 * In an empty objset, there are no blocks to read and thus 1813 * there can be no i/o errors (which we assert below). 1814 */ 1815 moid = MASTER_NODE_OBJ; 1816 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1817 DMU_OT_NONE, 0, tx); 1818 ASSERT(error == 0); 1819 1820 /* 1821 * Set starting attributes. 1822 */ 1823 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1824 elem = NULL; 1825 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1826 /* For the moment we expect all zpl props to be uint64_ts */ 1827 uint64_t val; 1828 char *name; 1829 1830 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1831 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1832 name = nvpair_name(elem); 1833 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1834 if (val < version) 1835 version = val; 1836 } else { 1837 error = zap_update(os, moid, name, 8, 1, &val, tx); 1838 } 1839 ASSERT(error == 0); 1840 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1841 norm = val; 1842 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1843 sense = val; 1844 } 1845 ASSERT(version != 0); 1846 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1847 1848 /* 1849 * Create zap object used for SA attribute registration 1850 */ 1851 1852 if (version >= ZPL_VERSION_SA) { 1853 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1854 DMU_OT_NONE, 0, tx); 1855 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1856 ASSERT(error == 0); 1857 } else { 1858 sa_obj = 0; 1859 } 1860 /* 1861 * Create a delete queue. 1862 */ 1863 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1864 1865 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1866 ASSERT(error == 0); 1867 1868 /* 1869 * Create root znode. Create minimal znode/vnode/zfsvfs 1870 * to allow zfs_mknode to work. 1871 */ 1872 VATTR_NULL(&vattr); 1873 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1874 vattr.va_type = VDIR; 1875 vattr.va_mode = S_IFDIR|0755; 1876 vattr.va_uid = crgetuid(cr); 1877 vattr.va_gid = crgetgid(cr); 1878 1879 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 1880 1881 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1882 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1883 rootzp->z_moved = 0; 1884 rootzp->z_unlinked = 0; 1885 rootzp->z_atime_dirty = 0; 1886 rootzp->z_is_sa = USE_SA(version, os); 1887 1888 zfsvfs->z_os = os; 1889 zfsvfs->z_parent = zfsvfs; 1890 zfsvfs->z_version = version; 1891 zfsvfs->z_use_fuids = USE_FUIDS(version, os); 1892 zfsvfs->z_use_sa = USE_SA(version, os); 1893 zfsvfs->z_norm = norm; 1894 1895 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1896 &zfsvfs->z_attr_table); 1897 1898 ASSERT(error == 0); 1899 1900 /* 1901 * Fold case on file systems that are always or sometimes case 1902 * insensitive. 1903 */ 1904 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1905 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 1906 1907 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1908 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 1909 offsetof(znode_t, z_link_node)); 1910 1911 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1912 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1913 1914 rootzp->z_zfsvfs = zfsvfs; 1915 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1916 cr, NULL, &acl_ids)); 1917 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1918 ASSERT3P(zp, ==, rootzp); 1919 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1920 ASSERT(error == 0); 1921 zfs_acl_ids_free(&acl_ids); 1922 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1923 1924 sa_handle_destroy(rootzp->z_sa_hdl); 1925 kmem_cache_free(znode_cache, rootzp); 1926 1927 /* 1928 * Create shares directory 1929 */ 1930 1931 error = zfs_create_share_dir(zfsvfs, tx); 1932 1933 ASSERT(error == 0); 1934 1935 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1936 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1937 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1938} 1939#endif /* _KERNEL */ 1940 1941static int 1942zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1943{ 1944 uint64_t sa_obj = 0; 1945 int error; 1946 1947 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1948 if (error != 0 && error != ENOENT) 1949 return (error); 1950 1951 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1952 return (error); 1953} 1954 1955static int 1956zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1957 dmu_buf_t **db, void *tag) 1958{ 1959 dmu_object_info_t doi; 1960 int error; 1961 1962 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 1963 return (error); 1964 1965 dmu_object_info_from_db(*db, &doi); 1966 if ((doi.doi_bonus_type != DMU_OT_SA && 1967 doi.doi_bonus_type != DMU_OT_ZNODE) || 1968 doi.doi_bonus_type == DMU_OT_ZNODE && 1969 doi.doi_bonus_size < sizeof (znode_phys_t)) { 1970 sa_buf_rele(*db, tag); 1971 return (SET_ERROR(ENOTSUP)); 1972 } 1973 1974 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 1975 if (error != 0) { 1976 sa_buf_rele(*db, tag); 1977 return (error); 1978 } 1979 1980 return (0); 1981} 1982 1983void 1984zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 1985{ 1986 sa_handle_destroy(hdl); 1987 sa_buf_rele(db, tag); 1988} 1989 1990/* 1991 * Given an object number, return its parent object number and whether 1992 * or not the object is an extended attribute directory. 1993 */ 1994static int 1995zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table, 1996 uint64_t *pobjp, int *is_xattrdir) 1997{ 1998 uint64_t parent; 1999 uint64_t pflags; 2000 uint64_t mode; 2001 uint64_t parent_mode; 2002 sa_bulk_attr_t bulk[3]; 2003 sa_handle_t *sa_hdl; 2004 dmu_buf_t *sa_db; 2005 int count = 0; 2006 int error; 2007 2008 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2009 &parent, sizeof (parent)); 2010 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2011 &pflags, sizeof (pflags)); 2012 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2013 &mode, sizeof (mode)); 2014 2015 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2016 return (error); 2017 2018 /* 2019 * When a link is removed its parent pointer is not changed and will 2020 * be invalid. There are two cases where a link is removed but the 2021 * file stays around, when it goes to the delete queue and when there 2022 * are additional links. 2023 */ 2024 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG); 2025 if (error != 0) 2026 return (error); 2027 2028 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode)); 2029 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2030 if (error != 0) 2031 return (error); 2032 2033 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2034 2035 /* 2036 * Extended attributes can be applied to files, directories, etc. 2037 * Otherwise the parent must be a directory. 2038 */ 2039 if (!*is_xattrdir && !S_ISDIR(parent_mode)) 2040 return (SET_ERROR(EINVAL)); 2041 2042 *pobjp = parent; 2043 2044 return (0); 2045} 2046 2047/* 2048 * Given an object number, return some zpl level statistics 2049 */ 2050static int 2051zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2052 zfs_stat_t *sb) 2053{ 2054 sa_bulk_attr_t bulk[4]; 2055 int count = 0; 2056 2057 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2058 &sb->zs_mode, sizeof (sb->zs_mode)); 2059 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2060 &sb->zs_gen, sizeof (sb->zs_gen)); 2061 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2062 &sb->zs_links, sizeof (sb->zs_links)); 2063 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2064 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2065 2066 return (sa_bulk_lookup(hdl, bulk, count)); 2067} 2068 2069static int 2070zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2071 sa_attr_type_t *sa_table, char *buf, int len) 2072{ 2073 sa_handle_t *sa_hdl; 2074 sa_handle_t *prevhdl = NULL; 2075 dmu_buf_t *prevdb = NULL; 2076 dmu_buf_t *sa_db = NULL; 2077 char *path = buf + len - 1; 2078 int error; 2079 2080 *path = '\0'; 2081 sa_hdl = hdl; 2082 2083 for (;;) { 2084 uint64_t pobj; 2085 char component[MAXNAMELEN + 2]; 2086 size_t complen; 2087 int is_xattrdir; 2088 2089 if (prevdb) 2090 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2091 2092 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj, 2093 &is_xattrdir)) != 0) 2094 break; 2095 2096 if (pobj == obj) { 2097 if (path[0] != '/') 2098 *--path = '/'; 2099 break; 2100 } 2101 2102 component[0] = '/'; 2103 if (is_xattrdir) { 2104 (void) sprintf(component + 1, "<xattrdir>"); 2105 } else { 2106 error = zap_value_search(osp, pobj, obj, 2107 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2108 if (error != 0) 2109 break; 2110 } 2111 2112 complen = strlen(component); 2113 path -= complen; 2114 ASSERT(path >= buf); 2115 bcopy(component, path, complen); 2116 obj = pobj; 2117 2118 if (sa_hdl != hdl) { 2119 prevhdl = sa_hdl; 2120 prevdb = sa_db; 2121 } 2122 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2123 if (error != 0) { 2124 sa_hdl = prevhdl; 2125 sa_db = prevdb; 2126 break; 2127 } 2128 } 2129 2130 if (sa_hdl != NULL && sa_hdl != hdl) { 2131 ASSERT(sa_db != NULL); 2132 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2133 } 2134 2135 if (error == 0) 2136 (void) memmove(buf, path, buf + len - path); 2137 2138 return (error); 2139} 2140 2141int 2142zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2143{ 2144 sa_attr_type_t *sa_table; 2145 sa_handle_t *hdl; 2146 dmu_buf_t *db; 2147 int error; 2148 2149 error = zfs_sa_setup(osp, &sa_table); 2150 if (error != 0) 2151 return (error); 2152 2153 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2154 if (error != 0) 2155 return (error); 2156 2157 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2158 2159 zfs_release_sa_handle(hdl, db, FTAG); 2160 return (error); 2161} 2162 2163int 2164zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2165 char *buf, int len) 2166{ 2167 char *path = buf + len - 1; 2168 sa_attr_type_t *sa_table; 2169 sa_handle_t *hdl; 2170 dmu_buf_t *db; 2171 int error; 2172 2173 *path = '\0'; 2174 2175 error = zfs_sa_setup(osp, &sa_table); 2176 if (error != 0) 2177 return (error); 2178 2179 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2180 if (error != 0) 2181 return (error); 2182 2183 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2184 if (error != 0) { 2185 zfs_release_sa_handle(hdl, db, FTAG); 2186 return (error); 2187 } 2188 2189 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2190 2191 zfs_release_sa_handle(hdl, db, FTAG); 2192 return (error); 2193} 2194 2195#ifdef _KERNEL 2196int 2197zfs_znode_parent_and_name(znode_t *zp, znode_t **dzpp, char *buf) 2198{ 2199 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2200 uint64_t parent; 2201 int is_xattrdir; 2202 int err; 2203 2204 /* Extended attributes should not be visible as regular files. */ 2205 if ((zp->z_pflags & ZFS_XATTR) != 0) 2206 return (SET_ERROR(EINVAL)); 2207 2208 err = zfs_obj_to_pobj(zfsvfs->z_os, zp->z_sa_hdl, zfsvfs->z_attr_table, 2209 &parent, &is_xattrdir); 2210 if (err != 0) 2211 return (err); 2212 ASSERT0(is_xattrdir); 2213 2214 /* No name as this is a root object. */ 2215 if (parent == zp->z_id) 2216 return (SET_ERROR(EINVAL)); 2217 2218 err = zap_value_search(zfsvfs->z_os, parent, zp->z_id, 2219 ZFS_DIRENT_OBJ(-1ULL), buf); 2220 if (err != 0) 2221 return (err); 2222 err = zfs_zget(zfsvfs, parent, dzpp); 2223 return (err); 2224} 2225#endif /* _KERNEL */ 2226