zfs_znode.c revision 225153
1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21/* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 */ 24 25/* Portions Copyright 2007 Jeremy Teo */ 26 27#ifdef _KERNEL 28#include <sys/types.h> 29#include <sys/param.h> 30#include <sys/time.h> 31#include <sys/systm.h> 32#include <sys/sysmacros.h> 33#include <sys/resource.h> 34#include <sys/mntent.h> 35#include <sys/u8_textprep.h> 36#include <sys/dsl_dataset.h> 37#include <sys/vfs.h> 38#include <sys/vnode.h> 39#include <sys/file.h> 40#include <sys/kmem.h> 41#include <sys/errno.h> 42#include <sys/unistd.h> 43#include <sys/atomic.h> 44#include <sys/zfs_dir.h> 45#include <sys/zfs_acl.h> 46#include <sys/zfs_ioctl.h> 47#include <sys/zfs_rlock.h> 48#include <sys/zfs_fuid.h> 49#include <sys/dnode.h> 50#include <sys/fs/zfs.h> 51#include <sys/kidmap.h> 52#endif /* _KERNEL */ 53 54#include <sys/dmu.h> 55#include <sys/refcount.h> 56#include <sys/stat.h> 57#include <sys/zap.h> 58#include <sys/zfs_znode.h> 59#include <sys/sa.h> 60#include <sys/zfs_sa.h> 61#include <sys/zfs_stat.h> 62#include <sys/refcount.h> 63 64#include "zfs_prop.h" 65#include "zfs_comutil.h" 66 67/* Used by fstat(1). */ 68SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t), 69 "sizeof(znode_t)"); 70 71/* 72 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 73 * turned on when DEBUG is also defined. 74 */ 75#ifdef DEBUG 76#define ZNODE_STATS 77#endif /* DEBUG */ 78 79#ifdef ZNODE_STATS 80#define ZNODE_STAT_ADD(stat) ((stat)++) 81#else 82#define ZNODE_STAT_ADD(stat) /* nothing */ 83#endif /* ZNODE_STATS */ 84 85/* 86 * Functions needed for userland (ie: libzpool) are not put under 87 * #ifdef_KERNEL; the rest of the functions have dependencies 88 * (such as VFS logic) that will not compile easily in userland. 89 */ 90#ifdef _KERNEL 91/* 92 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 93 * be freed before it can be safely accessed. 94 */ 95krwlock_t zfsvfs_lock; 96 97static kmem_cache_t *znode_cache = NULL; 98 99/*ARGSUSED*/ 100static void 101znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 102{ 103 /* 104 * We should never drop all dbuf refs without first clearing 105 * the eviction callback. 106 */ 107 panic("evicting znode %p\n", user_ptr); 108} 109 110extern struct vop_vector zfs_vnodeops; 111extern struct vop_vector zfs_fifoops; 112extern struct vop_vector zfs_shareops; 113 114/* 115 * XXX: We cannot use this function as a cache constructor, because 116 * there is one global cache for all file systems and we need 117 * to pass vfsp here, which is not possible, because argument 118 * 'cdrarg' is defined at kmem_cache_create() time. 119 */ 120/*ARGSUSED*/ 121static int 122zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 123{ 124 znode_t *zp = buf; 125 vnode_t *vp; 126 vfs_t *vfsp = arg; 127 int error; 128 129 POINTER_INVALIDATE(&zp->z_zfsvfs); 130 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 131 132 if (vfsp != NULL) { 133 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp); 134 if (error != 0 && (kmflags & KM_NOSLEEP)) 135 return (-1); 136 ASSERT(error == 0); 137 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 138 zp->z_vnode = vp; 139 vp->v_data = (caddr_t)zp; 140 VN_LOCK_AREC(vp); 141 VN_LOCK_ASHARE(vp); 142 } else { 143 zp->z_vnode = NULL; 144 } 145 146 list_link_init(&zp->z_link_node); 147 148 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 149 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 150 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 151 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 152 153 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 154 avl_create(&zp->z_range_avl, zfs_range_compare, 155 sizeof (rl_t), offsetof(rl_t, r_node)); 156 157 zp->z_dirlocks = NULL; 158 zp->z_acl_cached = NULL; 159 zp->z_moved = 0; 160 return (0); 161} 162 163/*ARGSUSED*/ 164static void 165zfs_znode_cache_destructor(void *buf, void *arg) 166{ 167 znode_t *zp = buf; 168 169 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 170 ASSERT(ZTOV(zp) == NULL); 171 vn_free(ZTOV(zp)); 172 ASSERT(!list_link_active(&zp->z_link_node)); 173 mutex_destroy(&zp->z_lock); 174 rw_destroy(&zp->z_parent_lock); 175 rw_destroy(&zp->z_name_lock); 176 mutex_destroy(&zp->z_acl_lock); 177 avl_destroy(&zp->z_range_avl); 178 mutex_destroy(&zp->z_range_lock); 179 180 ASSERT(zp->z_dirlocks == NULL); 181 ASSERT(zp->z_acl_cached == NULL); 182} 183 184#ifdef ZNODE_STATS 185static struct { 186 uint64_t zms_zfsvfs_invalid; 187 uint64_t zms_zfsvfs_recheck1; 188 uint64_t zms_zfsvfs_unmounted; 189 uint64_t zms_zfsvfs_recheck2; 190 uint64_t zms_obj_held; 191 uint64_t zms_vnode_locked; 192 uint64_t zms_not_only_dnlc; 193} znode_move_stats; 194#endif /* ZNODE_STATS */ 195 196#ifdef sun 197static void 198zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 199{ 200 vnode_t *vp; 201 202 /* Copy fields. */ 203 nzp->z_zfsvfs = ozp->z_zfsvfs; 204 205 /* Swap vnodes. */ 206 vp = nzp->z_vnode; 207 nzp->z_vnode = ozp->z_vnode; 208 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 209 ZTOV(ozp)->v_data = ozp; 210 ZTOV(nzp)->v_data = nzp; 211 212 nzp->z_id = ozp->z_id; 213 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 214 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 215 nzp->z_unlinked = ozp->z_unlinked; 216 nzp->z_atime_dirty = ozp->z_atime_dirty; 217 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 218 nzp->z_blksz = ozp->z_blksz; 219 nzp->z_seq = ozp->z_seq; 220 nzp->z_mapcnt = ozp->z_mapcnt; 221 nzp->z_gen = ozp->z_gen; 222 nzp->z_sync_cnt = ozp->z_sync_cnt; 223 nzp->z_is_sa = ozp->z_is_sa; 224 nzp->z_sa_hdl = ozp->z_sa_hdl; 225 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 226 nzp->z_links = ozp->z_links; 227 nzp->z_size = ozp->z_size; 228 nzp->z_pflags = ozp->z_pflags; 229 nzp->z_uid = ozp->z_uid; 230 nzp->z_gid = ozp->z_gid; 231 nzp->z_mode = ozp->z_mode; 232 233 /* 234 * Since this is just an idle znode and kmem is already dealing with 235 * memory pressure, release any cached ACL. 236 */ 237 if (ozp->z_acl_cached) { 238 zfs_acl_free(ozp->z_acl_cached); 239 ozp->z_acl_cached = NULL; 240 } 241 242 sa_set_userp(nzp->z_sa_hdl, nzp); 243 244 /* 245 * Invalidate the original znode by clearing fields that provide a 246 * pointer back to the znode. Set the low bit of the vfs pointer to 247 * ensure that zfs_znode_move() recognizes the znode as invalid in any 248 * subsequent callback. 249 */ 250 ozp->z_sa_hdl = NULL; 251 POINTER_INVALIDATE(&ozp->z_zfsvfs); 252 253 /* 254 * Mark the znode. 255 */ 256 nzp->z_moved = 1; 257 ozp->z_moved = (uint8_t)-1; 258} 259 260/*ARGSUSED*/ 261static kmem_cbrc_t 262zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 263{ 264 znode_t *ozp = buf, *nzp = newbuf; 265 zfsvfs_t *zfsvfs; 266 vnode_t *vp; 267 268 /* 269 * The znode is on the file system's list of known znodes if the vfs 270 * pointer is valid. We set the low bit of the vfs pointer when freeing 271 * the znode to invalidate it, and the memory patterns written by kmem 272 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 273 * created znode sets the vfs pointer last of all to indicate that the 274 * znode is known and in a valid state to be moved by this function. 275 */ 276 zfsvfs = ozp->z_zfsvfs; 277 if (!POINTER_IS_VALID(zfsvfs)) { 278 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 279 return (KMEM_CBRC_DONT_KNOW); 280 } 281 282 /* 283 * Close a small window in which it's possible that the filesystem could 284 * be unmounted and freed, and zfsvfs, though valid in the previous 285 * statement, could point to unrelated memory by the time we try to 286 * prevent the filesystem from being unmounted. 287 */ 288 rw_enter(&zfsvfs_lock, RW_WRITER); 289 if (zfsvfs != ozp->z_zfsvfs) { 290 rw_exit(&zfsvfs_lock); 291 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 292 return (KMEM_CBRC_DONT_KNOW); 293 } 294 295 /* 296 * If the znode is still valid, then so is the file system. We know that 297 * no valid file system can be freed while we hold zfsvfs_lock, so we 298 * can safely ensure that the filesystem is not and will not be 299 * unmounted. The next statement is equivalent to ZFS_ENTER(). 300 */ 301 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 302 if (zfsvfs->z_unmounted) { 303 ZFS_EXIT(zfsvfs); 304 rw_exit(&zfsvfs_lock); 305 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 306 return (KMEM_CBRC_DONT_KNOW); 307 } 308 rw_exit(&zfsvfs_lock); 309 310 mutex_enter(&zfsvfs->z_znodes_lock); 311 /* 312 * Recheck the vfs pointer in case the znode was removed just before 313 * acquiring the lock. 314 */ 315 if (zfsvfs != ozp->z_zfsvfs) { 316 mutex_exit(&zfsvfs->z_znodes_lock); 317 ZFS_EXIT(zfsvfs); 318 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 319 return (KMEM_CBRC_DONT_KNOW); 320 } 321 322 /* 323 * At this point we know that as long as we hold z_znodes_lock, the 324 * znode cannot be freed and fields within the znode can be safely 325 * accessed. Now, prevent a race with zfs_zget(). 326 */ 327 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 328 mutex_exit(&zfsvfs->z_znodes_lock); 329 ZFS_EXIT(zfsvfs); 330 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 331 return (KMEM_CBRC_LATER); 332 } 333 334 vp = ZTOV(ozp); 335 if (mutex_tryenter(&vp->v_lock) == 0) { 336 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 337 mutex_exit(&zfsvfs->z_znodes_lock); 338 ZFS_EXIT(zfsvfs); 339 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 340 return (KMEM_CBRC_LATER); 341 } 342 343 /* Only move znodes that are referenced _only_ by the DNLC. */ 344 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 345 mutex_exit(&vp->v_lock); 346 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 347 mutex_exit(&zfsvfs->z_znodes_lock); 348 ZFS_EXIT(zfsvfs); 349 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 350 return (KMEM_CBRC_LATER); 351 } 352 353 /* 354 * The znode is known and in a valid state to move. We're holding the 355 * locks needed to execute the critical section. 356 */ 357 zfs_znode_move_impl(ozp, nzp); 358 mutex_exit(&vp->v_lock); 359 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 360 361 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 362 mutex_exit(&zfsvfs->z_znodes_lock); 363 ZFS_EXIT(zfsvfs); 364 365 return (KMEM_CBRC_YES); 366} 367#endif /* sun */ 368 369void 370zfs_znode_init(void) 371{ 372 /* 373 * Initialize zcache 374 */ 375 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 376 ASSERT(znode_cache == NULL); 377 znode_cache = kmem_cache_create("zfs_znode_cache", 378 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL, 379 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 380 kmem_cache_set_move(znode_cache, zfs_znode_move); 381} 382 383void 384zfs_znode_fini(void) 385{ 386#ifdef sun 387 /* 388 * Cleanup vfs & vnode ops 389 */ 390 zfs_remove_op_tables(); 391#endif /* sun */ 392 393 /* 394 * Cleanup zcache 395 */ 396 if (znode_cache) 397 kmem_cache_destroy(znode_cache); 398 znode_cache = NULL; 399 rw_destroy(&zfsvfs_lock); 400} 401 402#ifdef sun 403struct vnodeops *zfs_dvnodeops; 404struct vnodeops *zfs_fvnodeops; 405struct vnodeops *zfs_symvnodeops; 406struct vnodeops *zfs_xdvnodeops; 407struct vnodeops *zfs_evnodeops; 408struct vnodeops *zfs_sharevnodeops; 409 410void 411zfs_remove_op_tables() 412{ 413 /* 414 * Remove vfs ops 415 */ 416 ASSERT(zfsfstype); 417 (void) vfs_freevfsops_by_type(zfsfstype); 418 zfsfstype = 0; 419 420 /* 421 * Remove vnode ops 422 */ 423 if (zfs_dvnodeops) 424 vn_freevnodeops(zfs_dvnodeops); 425 if (zfs_fvnodeops) 426 vn_freevnodeops(zfs_fvnodeops); 427 if (zfs_symvnodeops) 428 vn_freevnodeops(zfs_symvnodeops); 429 if (zfs_xdvnodeops) 430 vn_freevnodeops(zfs_xdvnodeops); 431 if (zfs_evnodeops) 432 vn_freevnodeops(zfs_evnodeops); 433 if (zfs_sharevnodeops) 434 vn_freevnodeops(zfs_sharevnodeops); 435 436 zfs_dvnodeops = NULL; 437 zfs_fvnodeops = NULL; 438 zfs_symvnodeops = NULL; 439 zfs_xdvnodeops = NULL; 440 zfs_evnodeops = NULL; 441 zfs_sharevnodeops = NULL; 442} 443 444extern const fs_operation_def_t zfs_dvnodeops_template[]; 445extern const fs_operation_def_t zfs_fvnodeops_template[]; 446extern const fs_operation_def_t zfs_xdvnodeops_template[]; 447extern const fs_operation_def_t zfs_symvnodeops_template[]; 448extern const fs_operation_def_t zfs_evnodeops_template[]; 449extern const fs_operation_def_t zfs_sharevnodeops_template[]; 450 451int 452zfs_create_op_tables() 453{ 454 int error; 455 456 /* 457 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 458 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 459 * In this case we just return as the ops vectors are already set up. 460 */ 461 if (zfs_dvnodeops) 462 return (0); 463 464 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 465 &zfs_dvnodeops); 466 if (error) 467 return (error); 468 469 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 470 &zfs_fvnodeops); 471 if (error) 472 return (error); 473 474 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 475 &zfs_symvnodeops); 476 if (error) 477 return (error); 478 479 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 480 &zfs_xdvnodeops); 481 if (error) 482 return (error); 483 484 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 485 &zfs_evnodeops); 486 if (error) 487 return (error); 488 489 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 490 &zfs_sharevnodeops); 491 492 return (error); 493} 494#endif /* sun */ 495 496int 497zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 498{ 499 zfs_acl_ids_t acl_ids; 500 vattr_t vattr; 501 znode_t *sharezp; 502 vnode_t *vp, vnode; 503 znode_t *zp; 504 int error; 505 506 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 507 vattr.va_type = VDIR; 508 vattr.va_mode = S_IFDIR|0555; 509 vattr.va_uid = crgetuid(kcred); 510 vattr.va_gid = crgetgid(kcred); 511 512 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 513 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0); 514 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 515 sharezp->z_moved = 0; 516 sharezp->z_unlinked = 0; 517 sharezp->z_atime_dirty = 0; 518 sharezp->z_zfsvfs = zfsvfs; 519 sharezp->z_is_sa = zfsvfs->z_use_sa; 520 521 sharezp->z_vnode = &vnode; 522 vnode.v_data = sharezp; 523 524 vp = ZTOV(sharezp); 525 vp->v_type = VDIR; 526 527 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 528 kcred, NULL, &acl_ids)); 529 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 530 ASSERT3P(zp, ==, sharezp); 531 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 532 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 533 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 534 zfsvfs->z_shares_dir = sharezp->z_id; 535 536 zfs_acl_ids_free(&acl_ids); 537 ZTOV(sharezp)->v_data = NULL; 538 ZTOV(sharezp)->v_count = 0; 539 ZTOV(sharezp)->v_holdcnt = 0; 540 zp->z_vnode = NULL; 541 sa_handle_destroy(sharezp->z_sa_hdl); 542 sharezp->z_vnode = NULL; 543 kmem_cache_free(znode_cache, sharezp); 544 545 return (error); 546} 547 548/* 549 * define a couple of values we need available 550 * for both 64 and 32 bit environments. 551 */ 552#ifndef NBITSMINOR64 553#define NBITSMINOR64 32 554#endif 555#ifndef MAXMAJ64 556#define MAXMAJ64 0xffffffffUL 557#endif 558#ifndef MAXMIN64 559#define MAXMIN64 0xffffffffUL 560#endif 561 562/* 563 * Create special expldev for ZFS private use. 564 * Can't use standard expldev since it doesn't do 565 * what we want. The standard expldev() takes a 566 * dev32_t in LP64 and expands it to a long dev_t. 567 * We need an interface that takes a dev32_t in ILP32 568 * and expands it to a long dev_t. 569 */ 570static uint64_t 571zfs_expldev(dev_t dev) 572{ 573 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 574} 575/* 576 * Special cmpldev for ZFS private use. 577 * Can't use standard cmpldev since it takes 578 * a long dev_t and compresses it to dev32_t in 579 * LP64. We need to do a compaction of a long dev_t 580 * to a dev32_t in ILP32. 581 */ 582dev_t 583zfs_cmpldev(uint64_t dev) 584{ 585 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 586} 587 588static void 589zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 590 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 591{ 592 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 593 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 594 595 mutex_enter(&zp->z_lock); 596 597 ASSERT(zp->z_sa_hdl == NULL); 598 ASSERT(zp->z_acl_cached == NULL); 599 if (sa_hdl == NULL) { 600 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 601 SA_HDL_SHARED, &zp->z_sa_hdl)); 602 } else { 603 zp->z_sa_hdl = sa_hdl; 604 sa_set_userp(sa_hdl, zp); 605 } 606 607 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 608 609 /* 610 * Slap on VROOT if we are the root znode 611 */ 612 if (zp->z_id == zfsvfs->z_root) 613 ZTOV(zp)->v_flag |= VROOT; 614 615 mutex_exit(&zp->z_lock); 616 vn_exists(ZTOV(zp)); 617} 618 619void 620zfs_znode_dmu_fini(znode_t *zp) 621{ 622 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 623 zp->z_unlinked || 624 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 625 626 sa_handle_destroy(zp->z_sa_hdl); 627 zp->z_sa_hdl = NULL; 628} 629 630static void 631zfs_vnode_forget(vnode_t *vp) 632{ 633 634 VOP_UNLOCK(vp, 0); 635 VI_LOCK(vp); 636 vp->v_usecount--; 637 vp->v_iflag |= VI_DOOMED; 638 vp->v_data = NULL; 639 vdropl(vp); 640} 641 642/* 643 * Construct a new znode/vnode and intialize. 644 * 645 * This does not do a call to dmu_set_user() that is 646 * up to the caller to do, in case you don't want to 647 * return the znode 648 */ 649static znode_t * 650zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 651 dmu_object_type_t obj_type, sa_handle_t *hdl) 652{ 653 znode_t *zp; 654 vnode_t *vp; 655 uint64_t mode; 656 uint64_t parent; 657 sa_bulk_attr_t bulk[9]; 658 int count = 0; 659 660 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 661 zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0); 662 663 ASSERT(zp->z_dirlocks == NULL); 664 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 665 zp->z_moved = 0; 666 667 /* 668 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 669 * the zfs_znode_move() callback. 670 */ 671 zp->z_sa_hdl = NULL; 672 zp->z_unlinked = 0; 673 zp->z_atime_dirty = 0; 674 zp->z_mapcnt = 0; 675 zp->z_id = db->db_object; 676 zp->z_blksz = blksz; 677 zp->z_seq = 0x7A4653; 678 zp->z_sync_cnt = 0; 679 680 vp = ZTOV(zp); 681 682 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 683 684 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 685 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 686 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 687 &zp->z_size, 8); 688 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 689 &zp->z_links, 8); 690 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 691 &zp->z_pflags, 8); 692 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 694 &zp->z_atime, 16); 695 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 696 &zp->z_uid, 8); 697 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 698 &zp->z_gid, 8); 699 700 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 701 if (hdl == NULL) 702 sa_handle_destroy(zp->z_sa_hdl); 703 zfs_vnode_forget(vp); 704 zp->z_vnode = NULL; 705 kmem_cache_free(znode_cache, zp); 706 return (NULL); 707 } 708 709 zp->z_mode = mode; 710 711 vp->v_type = IFTOVT((mode_t)mode); 712 713 switch (vp->v_type) { 714 case VDIR: 715 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 716 break; 717#ifdef sun 718 case VBLK: 719 case VCHR: 720 { 721 uint64_t rdev; 722 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 723 &rdev, sizeof (rdev)) == 0); 724 725 vp->v_rdev = zfs_cmpldev(rdev); 726 } 727 break; 728#endif /* sun */ 729 case VFIFO: 730#ifdef sun 731 case VSOCK: 732 case VDOOR: 733#endif /* sun */ 734 vp->v_op = &zfs_fifoops; 735 break; 736 case VREG: 737 if (parent == zfsvfs->z_shares_dir) { 738 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 739 vp->v_op = &zfs_shareops; 740 } 741 break; 742#ifdef sun 743 case VLNK: 744 vn_setops(vp, zfs_symvnodeops); 745 break; 746 default: 747 vn_setops(vp, zfs_evnodeops); 748 break; 749#endif /* sun */ 750 } 751 if (vp->v_type != VFIFO) 752 VN_LOCK_ASHARE(vp); 753 754 mutex_enter(&zfsvfs->z_znodes_lock); 755 list_insert_tail(&zfsvfs->z_all_znodes, zp); 756 membar_producer(); 757 /* 758 * Everything else must be valid before assigning z_zfsvfs makes the 759 * znode eligible for zfs_znode_move(). 760 */ 761 zp->z_zfsvfs = zfsvfs; 762 mutex_exit(&zfsvfs->z_znodes_lock); 763 764 VFS_HOLD(zfsvfs->z_vfs); 765 return (zp); 766} 767 768static uint64_t empty_xattr; 769static uint64_t pad[4]; 770static zfs_acl_phys_t acl_phys; 771/* 772 * Create a new DMU object to hold a zfs znode. 773 * 774 * IN: dzp - parent directory for new znode 775 * vap - file attributes for new znode 776 * tx - dmu transaction id for zap operations 777 * cr - credentials of caller 778 * flag - flags: 779 * IS_ROOT_NODE - new object will be root 780 * IS_XATTR - new object is an attribute 781 * bonuslen - length of bonus buffer 782 * setaclp - File/Dir initial ACL 783 * fuidp - Tracks fuid allocation. 784 * 785 * OUT: zpp - allocated znode 786 * 787 */ 788void 789zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 790 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 791{ 792 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 793 uint64_t mode, size, links, parent, pflags; 794 uint64_t dzp_pflags = 0; 795 uint64_t rdev = 0; 796 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 797 dmu_buf_t *db; 798 timestruc_t now; 799 uint64_t gen, obj; 800 int err; 801 int bonuslen; 802 sa_handle_t *sa_hdl; 803 dmu_object_type_t obj_type; 804 sa_bulk_attr_t sa_attrs[ZPL_END]; 805 int cnt = 0; 806 zfs_acl_locator_cb_t locate = { 0 }; 807 808 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 809 810 if (zfsvfs->z_replay) { 811 obj = vap->va_nodeid; 812 now = vap->va_ctime; /* see zfs_replay_create() */ 813 gen = vap->va_nblocks; /* ditto */ 814 } else { 815 obj = 0; 816 gethrestime(&now); 817 gen = dmu_tx_get_txg(tx); 818 } 819 820 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 821 bonuslen = (obj_type == DMU_OT_SA) ? 822 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 823 824 /* 825 * Create a new DMU object. 826 */ 827 /* 828 * There's currently no mechanism for pre-reading the blocks that will 829 * be needed to allocate a new object, so we accept the small chance 830 * that there will be an i/o error and we will fail one of the 831 * assertions below. 832 */ 833 if (vap->va_type == VDIR) { 834 if (zfsvfs->z_replay) { 835 err = zap_create_claim_norm(zfsvfs->z_os, obj, 836 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 837 obj_type, bonuslen, tx); 838 ASSERT3U(err, ==, 0); 839 } else { 840 obj = zap_create_norm(zfsvfs->z_os, 841 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 842 obj_type, bonuslen, tx); 843 } 844 } else { 845 if (zfsvfs->z_replay) { 846 err = dmu_object_claim(zfsvfs->z_os, obj, 847 DMU_OT_PLAIN_FILE_CONTENTS, 0, 848 obj_type, bonuslen, tx); 849 ASSERT3U(err, ==, 0); 850 } else { 851 obj = dmu_object_alloc(zfsvfs->z_os, 852 DMU_OT_PLAIN_FILE_CONTENTS, 0, 853 obj_type, bonuslen, tx); 854 } 855 } 856 857 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 858 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 859 860 /* 861 * If this is the root, fix up the half-initialized parent pointer 862 * to reference the just-allocated physical data area. 863 */ 864 if (flag & IS_ROOT_NODE) { 865 dzp->z_id = obj; 866 } else { 867 dzp_pflags = dzp->z_pflags; 868 } 869 870 /* 871 * If parent is an xattr, so am I. 872 */ 873 if (dzp_pflags & ZFS_XATTR) { 874 flag |= IS_XATTR; 875 } 876 877 if (zfsvfs->z_use_fuids) 878 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 879 else 880 pflags = 0; 881 882 if (vap->va_type == VDIR) { 883 size = 2; /* contents ("." and "..") */ 884 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 885 } else { 886 size = links = 0; 887 } 888 889 if (vap->va_type == VBLK || vap->va_type == VCHR) { 890 rdev = zfs_expldev(vap->va_rdev); 891 } 892 893 parent = dzp->z_id; 894 mode = acl_ids->z_mode; 895 if (flag & IS_XATTR) 896 pflags |= ZFS_XATTR; 897 898 /* 899 * No execs denied will be deterimed when zfs_mode_compute() is called. 900 */ 901 pflags |= acl_ids->z_aclp->z_hints & 902 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 903 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 904 905 ZFS_TIME_ENCODE(&now, crtime); 906 ZFS_TIME_ENCODE(&now, ctime); 907 908 if (vap->va_mask & AT_ATIME) { 909 ZFS_TIME_ENCODE(&vap->va_atime, atime); 910 } else { 911 ZFS_TIME_ENCODE(&now, atime); 912 } 913 914 if (vap->va_mask & AT_MTIME) { 915 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 916 } else { 917 ZFS_TIME_ENCODE(&now, mtime); 918 } 919 920 /* Now add in all of the "SA" attributes */ 921 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 922 &sa_hdl)); 923 924 /* 925 * Setup the array of attributes to be replaced/set on the new file 926 * 927 * order for DMU_OT_ZNODE is critical since it needs to be constructed 928 * in the old znode_phys_t format. Don't change this ordering 929 */ 930 931 if (obj_type == DMU_OT_ZNODE) { 932 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 933 NULL, &atime, 16); 934 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 935 NULL, &mtime, 16); 936 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 937 NULL, &ctime, 16); 938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 939 NULL, &crtime, 16); 940 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 941 NULL, &gen, 8); 942 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 943 NULL, &mode, 8); 944 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 945 NULL, &size, 8); 946 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 947 NULL, &parent, 8); 948 } else { 949 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 950 NULL, &mode, 8); 951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 952 NULL, &size, 8); 953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 954 NULL, &gen, 8); 955 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 956 &acl_ids->z_fuid, 8); 957 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 958 &acl_ids->z_fgid, 8); 959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 960 NULL, &parent, 8); 961 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 962 NULL, &pflags, 8); 963 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 964 NULL, &atime, 16); 965 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 966 NULL, &mtime, 16); 967 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 968 NULL, &ctime, 16); 969 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 970 NULL, &crtime, 16); 971 } 972 973 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 974 975 if (obj_type == DMU_OT_ZNODE) { 976 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 977 &empty_xattr, 8); 978 } 979 if (obj_type == DMU_OT_ZNODE || 980 (vap->va_type == VBLK || vap->va_type == VCHR)) { 981 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 982 NULL, &rdev, 8); 983 984 } 985 if (obj_type == DMU_OT_ZNODE) { 986 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 987 NULL, &pflags, 8); 988 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 989 &acl_ids->z_fuid, 8); 990 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 991 &acl_ids->z_fgid, 8); 992 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 993 sizeof (uint64_t) * 4); 994 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 995 &acl_phys, sizeof (zfs_acl_phys_t)); 996 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 997 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 998 &acl_ids->z_aclp->z_acl_count, 8); 999 locate.cb_aclp = acl_ids->z_aclp; 1000 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 1001 zfs_acl_data_locator, &locate, 1002 acl_ids->z_aclp->z_acl_bytes); 1003 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 1004 acl_ids->z_fuid, acl_ids->z_fgid); 1005 } 1006 1007 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 1008 1009 if (!(flag & IS_ROOT_NODE)) { 1010 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 1011 ASSERT(*zpp != NULL); 1012 } else { 1013 /* 1014 * If we are creating the root node, the "parent" we 1015 * passed in is the znode for the root. 1016 */ 1017 *zpp = dzp; 1018 1019 (*zpp)->z_sa_hdl = sa_hdl; 1020 } 1021 1022 (*zpp)->z_pflags = pflags; 1023 (*zpp)->z_mode = mode; 1024 1025 if (vap->va_mask & AT_XVATTR) 1026 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 1027 1028 if (obj_type == DMU_OT_ZNODE || 1029 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1030 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx); 1031 ASSERT3P(err, ==, 0); 1032 } 1033 if (!(flag & IS_ROOT_NODE)) { 1034 vnode_t *vp; 1035 1036 vp = ZTOV(*zpp); 1037 vp->v_vflag |= VV_FORCEINSMQ; 1038 err = insmntque(vp, zfsvfs->z_vfs); 1039 vp->v_vflag &= ~VV_FORCEINSMQ; 1040 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1041 } 1042 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1043} 1044 1045/* 1046 * zfs_xvattr_set only updates the in-core attributes 1047 * it is assumed the caller will be doing an sa_bulk_update 1048 * to push the changes out 1049 */ 1050void 1051zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1052{ 1053 xoptattr_t *xoap; 1054 1055 xoap = xva_getxoptattr(xvap); 1056 ASSERT(xoap); 1057 1058 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1059 uint64_t times[2]; 1060 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1061 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1062 ×, sizeof (times), tx); 1063 XVA_SET_RTN(xvap, XAT_CREATETIME); 1064 } 1065 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1066 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1067 zp->z_pflags, tx); 1068 XVA_SET_RTN(xvap, XAT_READONLY); 1069 } 1070 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1071 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1072 zp->z_pflags, tx); 1073 XVA_SET_RTN(xvap, XAT_HIDDEN); 1074 } 1075 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1076 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1077 zp->z_pflags, tx); 1078 XVA_SET_RTN(xvap, XAT_SYSTEM); 1079 } 1080 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1081 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1082 zp->z_pflags, tx); 1083 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1084 } 1085 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1086 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1087 zp->z_pflags, tx); 1088 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1089 } 1090 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1091 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1092 zp->z_pflags, tx); 1093 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1094 } 1095 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1096 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1097 zp->z_pflags, tx); 1098 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1099 } 1100 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1101 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1102 zp->z_pflags, tx); 1103 XVA_SET_RTN(xvap, XAT_NODUMP); 1104 } 1105 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1106 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1107 zp->z_pflags, tx); 1108 XVA_SET_RTN(xvap, XAT_OPAQUE); 1109 } 1110 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1111 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1112 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1113 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1114 } 1115 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1116 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1117 zp->z_pflags, tx); 1118 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1119 } 1120 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1121 zfs_sa_set_scanstamp(zp, xvap, tx); 1122 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1123 } 1124 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1125 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1126 zp->z_pflags, tx); 1127 XVA_SET_RTN(xvap, XAT_REPARSE); 1128 } 1129 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1130 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1131 zp->z_pflags, tx); 1132 XVA_SET_RTN(xvap, XAT_OFFLINE); 1133 } 1134 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1135 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1136 zp->z_pflags, tx); 1137 XVA_SET_RTN(xvap, XAT_SPARSE); 1138 } 1139} 1140 1141int 1142zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1143{ 1144 dmu_object_info_t doi; 1145 dmu_buf_t *db; 1146 znode_t *zp; 1147 int err; 1148 sa_handle_t *hdl; 1149 int first = 1; 1150 1151 *zpp = NULL; 1152 1153again: 1154 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1155 1156 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1157 if (err) { 1158 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1159 return (err); 1160 } 1161 1162 dmu_object_info_from_db(db, &doi); 1163 if (doi.doi_bonus_type != DMU_OT_SA && 1164 (doi.doi_bonus_type != DMU_OT_ZNODE || 1165 (doi.doi_bonus_type == DMU_OT_ZNODE && 1166 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1167 sa_buf_rele(db, NULL); 1168 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1169 return (EINVAL); 1170 } 1171 1172 hdl = dmu_buf_get_user(db); 1173 if (hdl != NULL) { 1174 zp = sa_get_userdata(hdl); 1175 1176 1177 /* 1178 * Since "SA" does immediate eviction we 1179 * should never find a sa handle that doesn't 1180 * know about the znode. 1181 */ 1182 1183 ASSERT3P(zp, !=, NULL); 1184 1185 mutex_enter(&zp->z_lock); 1186 ASSERT3U(zp->z_id, ==, obj_num); 1187 if (zp->z_unlinked) { 1188 err = ENOENT; 1189 } else { 1190 vnode_t *vp; 1191 int dying = 0; 1192 1193 vp = ZTOV(zp); 1194 if (vp == NULL) 1195 dying = 1; 1196 else { 1197 VN_HOLD(vp); 1198 if ((vp->v_iflag & VI_DOOMED) != 0) { 1199 dying = 1; 1200 /* 1201 * Don't VN_RELE() vnode here, because 1202 * it can call vn_lock() which creates 1203 * LOR between vnode lock and znode 1204 * lock. We will VN_RELE() the vnode 1205 * after droping znode lock. 1206 */ 1207 } 1208 } 1209 if (dying) { 1210 if (first) { 1211 ZFS_LOG(1, "dying znode detected (zp=%p)", zp); 1212 first = 0; 1213 } 1214 /* 1215 * znode is dying so we can't reuse it, we must 1216 * wait until destruction is completed. 1217 */ 1218 sa_buf_rele(db, NULL); 1219 mutex_exit(&zp->z_lock); 1220 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1221 if (vp != NULL) 1222 VN_RELE(vp); 1223 tsleep(zp, 0, "zcollide", 1); 1224 goto again; 1225 } 1226 *zpp = zp; 1227 err = 0; 1228 } 1229 sa_buf_rele(db, NULL); 1230 mutex_exit(&zp->z_lock); 1231 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1232 return (err); 1233 } 1234 1235 /* 1236 * Not found create new znode/vnode 1237 * but only if file exists. 1238 * 1239 * There is a small window where zfs_vget() could 1240 * find this object while a file create is still in 1241 * progress. This is checked for in zfs_znode_alloc() 1242 * 1243 * if zfs_znode_alloc() fails it will drop the hold on the 1244 * bonus buffer. 1245 */ 1246 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1247 doi.doi_bonus_type, NULL); 1248 if (zp == NULL) { 1249 err = ENOENT; 1250 } else { 1251 *zpp = zp; 1252 } 1253 if (err == 0) { 1254 vnode_t *vp = ZTOV(zp); 1255 1256 err = insmntque(vp, zfsvfs->z_vfs); 1257 if (err == 0) 1258 VOP_UNLOCK(vp, 0); 1259 else { 1260 zp->z_vnode = NULL; 1261 zfs_znode_dmu_fini(zp); 1262 zfs_znode_free(zp); 1263 *zpp = NULL; 1264 } 1265 } 1266 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1267 return (err); 1268} 1269 1270int 1271zfs_rezget(znode_t *zp) 1272{ 1273 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1274 dmu_object_info_t doi; 1275 dmu_buf_t *db; 1276 uint64_t obj_num = zp->z_id; 1277 uint64_t mode, size; 1278 sa_bulk_attr_t bulk[8]; 1279 int err; 1280 int count = 0; 1281 uint64_t gen; 1282 1283 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1284 1285 mutex_enter(&zp->z_acl_lock); 1286 if (zp->z_acl_cached) { 1287 zfs_acl_free(zp->z_acl_cached); 1288 zp->z_acl_cached = NULL; 1289 } 1290 1291 mutex_exit(&zp->z_acl_lock); 1292 ASSERT(zp->z_sa_hdl == NULL); 1293 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1294 if (err) { 1295 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1296 return (err); 1297 } 1298 1299 dmu_object_info_from_db(db, &doi); 1300 if (doi.doi_bonus_type != DMU_OT_SA && 1301 (doi.doi_bonus_type != DMU_OT_ZNODE || 1302 (doi.doi_bonus_type == DMU_OT_ZNODE && 1303 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1304 sa_buf_rele(db, NULL); 1305 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1306 return (EINVAL); 1307 } 1308 1309 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1310 size = zp->z_size; 1311 1312 /* reload cached values */ 1313 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1314 &gen, sizeof (gen)); 1315 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1316 &zp->z_size, sizeof (zp->z_size)); 1317 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1318 &zp->z_links, sizeof (zp->z_links)); 1319 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1320 &zp->z_pflags, sizeof (zp->z_pflags)); 1321 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1322 &zp->z_atime, sizeof (zp->z_atime)); 1323 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1324 &zp->z_uid, sizeof (zp->z_uid)); 1325 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1326 &zp->z_gid, sizeof (zp->z_gid)); 1327 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1328 &mode, sizeof (mode)); 1329 1330 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1331 zfs_znode_dmu_fini(zp); 1332 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1333 return (EIO); 1334 } 1335 1336 zp->z_mode = mode; 1337 1338 if (gen != zp->z_gen) { 1339 zfs_znode_dmu_fini(zp); 1340 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1341 return (EIO); 1342 } 1343 1344 /* 1345 * XXXPJD: Not sure how is that possible, but under heavy 1346 * zfs recv -F load it happens that z_gen is the same, but 1347 * vnode type is different than znode type. This would mean 1348 * that for example regular file was replaced with directory 1349 * which has the same object number. 1350 */ 1351 if (ZTOV(zp) != NULL && 1352 ZTOV(zp)->v_type != IFTOVT((mode_t)zp->z_mode)) { 1353 zfs_znode_dmu_fini(zp); 1354 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1355 return (EIO); 1356 } 1357 1358 zp->z_unlinked = (zp->z_links == 0); 1359 zp->z_blksz = doi.doi_data_block_size; 1360 if (zp->z_size != size && ZTOV(zp) != NULL) 1361 vnode_pager_setsize(ZTOV(zp), zp->z_size); 1362 1363 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1364 1365 return (0); 1366} 1367 1368void 1369zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1370{ 1371 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1372 objset_t *os = zfsvfs->z_os; 1373 uint64_t obj = zp->z_id; 1374 uint64_t acl_obj = zfs_external_acl(zp); 1375 1376 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1377 if (acl_obj) { 1378 VERIFY(!zp->z_is_sa); 1379 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1380 } 1381 VERIFY(0 == dmu_object_free(os, obj, tx)); 1382 zfs_znode_dmu_fini(zp); 1383 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1384 zfs_znode_free(zp); 1385} 1386 1387void 1388zfs_zinactive(znode_t *zp) 1389{ 1390 vnode_t *vp = ZTOV(zp); 1391 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1392 uint64_t z_id = zp->z_id; 1393 int vfslocked; 1394 1395 ASSERT(zp->z_sa_hdl); 1396 1397 /* 1398 * Don't allow a zfs_zget() while were trying to release this znode 1399 */ 1400 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1401 1402 mutex_enter(&zp->z_lock); 1403 VI_LOCK(vp); 1404 if (vp->v_count > 0) { 1405 /* 1406 * If the hold count is greater than zero, somebody has 1407 * obtained a new reference on this znode while we were 1408 * processing it here, so we are done. 1409 */ 1410 VI_UNLOCK(vp); 1411 mutex_exit(&zp->z_lock); 1412 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1413 return; 1414 } 1415 VI_UNLOCK(vp); 1416 1417 /* 1418 * If this was the last reference to a file with no links, 1419 * remove the file from the file system. 1420 */ 1421 if (zp->z_unlinked) { 1422 mutex_exit(&zp->z_lock); 1423 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1424 ASSERT(vp->v_count == 0); 1425 vrecycle(vp, curthread); 1426 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs); 1427 zfs_rmnode(zp); 1428 VFS_UNLOCK_GIANT(vfslocked); 1429 return; 1430 } 1431 1432 mutex_exit(&zp->z_lock); 1433 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1434} 1435 1436void 1437zfs_znode_free(znode_t *zp) 1438{ 1439 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1440 1441 ASSERT(ZTOV(zp) == NULL); 1442 ASSERT(zp->z_sa_hdl == NULL); 1443 mutex_enter(&zfsvfs->z_znodes_lock); 1444 POINTER_INVALIDATE(&zp->z_zfsvfs); 1445 list_remove(&zfsvfs->z_all_znodes, zp); 1446 mutex_exit(&zfsvfs->z_znodes_lock); 1447 1448 if (zp->z_acl_cached) { 1449 zfs_acl_free(zp->z_acl_cached); 1450 zp->z_acl_cached = NULL; 1451 } 1452 1453 kmem_cache_free(znode_cache, zp); 1454 1455 VFS_RELE(zfsvfs->z_vfs); 1456} 1457 1458void 1459zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1460 uint64_t ctime[2], boolean_t have_tx) 1461{ 1462 timestruc_t now; 1463 1464 gethrestime(&now); 1465 1466 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1467 zp->z_atime_dirty = 0; 1468 zp->z_seq++; 1469 } else { 1470 zp->z_atime_dirty = 1; 1471 } 1472 1473 if (flag & AT_ATIME) { 1474 ZFS_TIME_ENCODE(&now, zp->z_atime); 1475 } 1476 1477 if (flag & AT_MTIME) { 1478 ZFS_TIME_ENCODE(&now, mtime); 1479 if (zp->z_zfsvfs->z_use_fuids) { 1480 zp->z_pflags |= (ZFS_ARCHIVE | 1481 ZFS_AV_MODIFIED); 1482 } 1483 } 1484 1485 if (flag & AT_CTIME) { 1486 ZFS_TIME_ENCODE(&now, ctime); 1487 if (zp->z_zfsvfs->z_use_fuids) 1488 zp->z_pflags |= ZFS_ARCHIVE; 1489 } 1490} 1491 1492/* 1493 * Grow the block size for a file. 1494 * 1495 * IN: zp - znode of file to free data in. 1496 * size - requested block size 1497 * tx - open transaction. 1498 * 1499 * NOTE: this function assumes that the znode is write locked. 1500 */ 1501void 1502zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1503{ 1504 int error; 1505 u_longlong_t dummy; 1506 1507 if (size <= zp->z_blksz) 1508 return; 1509 /* 1510 * If the file size is already greater than the current blocksize, 1511 * we will not grow. If there is more than one block in a file, 1512 * the blocksize cannot change. 1513 */ 1514 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1515 return; 1516 1517 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1518 size, 0, tx); 1519 1520 if (error == ENOTSUP) 1521 return; 1522 ASSERT3U(error, ==, 0); 1523 1524 /* What blocksize did we actually get? */ 1525 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1526} 1527 1528#ifdef sun 1529/* 1530 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1531 * be calling back into the fs for a putpage(). E.g.: when truncating 1532 * a file, the pages being "thrown away* don't need to be written out. 1533 */ 1534/* ARGSUSED */ 1535static int 1536zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1537 int flags, cred_t *cr) 1538{ 1539 ASSERT(0); 1540 return (0); 1541} 1542#endif /* sun */ 1543 1544/* 1545 * Increase the file length 1546 * 1547 * IN: zp - znode of file to free data in. 1548 * end - new end-of-file 1549 * 1550 * RETURN: 0 if success 1551 * error code if failure 1552 */ 1553static int 1554zfs_extend(znode_t *zp, uint64_t end) 1555{ 1556 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1557 dmu_tx_t *tx; 1558 rl_t *rl; 1559 uint64_t newblksz; 1560 int error; 1561 1562 /* 1563 * We will change zp_size, lock the whole file. 1564 */ 1565 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1566 1567 /* 1568 * Nothing to do if file already at desired length. 1569 */ 1570 if (end <= zp->z_size) { 1571 zfs_range_unlock(rl); 1572 return (0); 1573 } 1574top: 1575 tx = dmu_tx_create(zfsvfs->z_os); 1576 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1577 zfs_sa_upgrade_txholds(tx, zp); 1578 if (end > zp->z_blksz && 1579 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1580 /* 1581 * We are growing the file past the current block size. 1582 */ 1583 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1584 ASSERT(!ISP2(zp->z_blksz)); 1585 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1586 } else { 1587 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1588 } 1589 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1590 } else { 1591 newblksz = 0; 1592 } 1593 1594 error = dmu_tx_assign(tx, TXG_NOWAIT); 1595 if (error) { 1596 if (error == ERESTART) { 1597 dmu_tx_wait(tx); 1598 dmu_tx_abort(tx); 1599 goto top; 1600 } 1601 dmu_tx_abort(tx); 1602 zfs_range_unlock(rl); 1603 return (error); 1604 } 1605 1606 if (newblksz) 1607 zfs_grow_blocksize(zp, newblksz, tx); 1608 1609 zp->z_size = end; 1610 1611 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1612 &zp->z_size, sizeof (zp->z_size), tx)); 1613 1614 vnode_pager_setsize(ZTOV(zp), end); 1615 1616 zfs_range_unlock(rl); 1617 1618 dmu_tx_commit(tx); 1619 1620 return (0); 1621} 1622 1623/* 1624 * Free space in a file. 1625 * 1626 * IN: zp - znode of file to free data in. 1627 * off - start of section to free. 1628 * len - length of section to free. 1629 * 1630 * RETURN: 0 if success 1631 * error code if failure 1632 */ 1633static int 1634zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1635{ 1636 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1637 rl_t *rl; 1638 int error; 1639 1640 /* 1641 * Lock the range being freed. 1642 */ 1643 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1644 1645 /* 1646 * Nothing to do if file already at desired length. 1647 */ 1648 if (off >= zp->z_size) { 1649 zfs_range_unlock(rl); 1650 return (0); 1651 } 1652 1653 if (off + len > zp->z_size) 1654 len = zp->z_size - off; 1655 1656 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1657 1658 if (error == 0) { 1659 /* 1660 * In FreeBSD we cannot free block in the middle of a file, 1661 * but only at the end of a file, so this code path should 1662 * never happen. 1663 */ 1664 vnode_pager_setsize(ZTOV(zp), off); 1665 } 1666 1667 zfs_range_unlock(rl); 1668 1669 return (error); 1670} 1671 1672/* 1673 * Truncate a file 1674 * 1675 * IN: zp - znode of file to free data in. 1676 * end - new end-of-file. 1677 * 1678 * RETURN: 0 if success 1679 * error code if failure 1680 */ 1681static int 1682zfs_trunc(znode_t *zp, uint64_t end) 1683{ 1684 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1685 vnode_t *vp = ZTOV(zp); 1686 dmu_tx_t *tx; 1687 rl_t *rl; 1688 int error; 1689 sa_bulk_attr_t bulk[2]; 1690 int count = 0; 1691 1692 /* 1693 * We will change zp_size, lock the whole file. 1694 */ 1695 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1696 1697 /* 1698 * Nothing to do if file already at desired length. 1699 */ 1700 if (end >= zp->z_size) { 1701 zfs_range_unlock(rl); 1702 return (0); 1703 } 1704 1705 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1706 if (error) { 1707 zfs_range_unlock(rl); 1708 return (error); 1709 } 1710top: 1711 tx = dmu_tx_create(zfsvfs->z_os); 1712 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1713 zfs_sa_upgrade_txholds(tx, zp); 1714 error = dmu_tx_assign(tx, TXG_NOWAIT); 1715 if (error) { 1716 if (error == ERESTART) { 1717 dmu_tx_wait(tx); 1718 dmu_tx_abort(tx); 1719 goto top; 1720 } 1721 dmu_tx_abort(tx); 1722 zfs_range_unlock(rl); 1723 return (error); 1724 } 1725 1726 zp->z_size = end; 1727 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1728 NULL, &zp->z_size, sizeof (zp->z_size)); 1729 1730 if (end == 0) { 1731 zp->z_pflags &= ~ZFS_SPARSE; 1732 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1733 NULL, &zp->z_pflags, 8); 1734 } 1735 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1736 1737 dmu_tx_commit(tx); 1738 1739 /* 1740 * Clear any mapped pages in the truncated region. This has to 1741 * happen outside of the transaction to avoid the possibility of 1742 * a deadlock with someone trying to push a page that we are 1743 * about to invalidate. 1744 */ 1745 vnode_pager_setsize(vp, end); 1746 1747 zfs_range_unlock(rl); 1748 1749 return (0); 1750} 1751 1752/* 1753 * Free space in a file 1754 * 1755 * IN: zp - znode of file to free data in. 1756 * off - start of range 1757 * len - end of range (0 => EOF) 1758 * flag - current file open mode flags. 1759 * log - TRUE if this action should be logged 1760 * 1761 * RETURN: 0 if success 1762 * error code if failure 1763 */ 1764int 1765zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1766{ 1767 vnode_t *vp = ZTOV(zp); 1768 dmu_tx_t *tx; 1769 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1770 zilog_t *zilog = zfsvfs->z_log; 1771 uint64_t mode; 1772 uint64_t mtime[2], ctime[2]; 1773 sa_bulk_attr_t bulk[3]; 1774 int count = 0; 1775 int error; 1776 1777 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1778 sizeof (mode))) != 0) 1779 return (error); 1780 1781 if (off > zp->z_size) { 1782 error = zfs_extend(zp, off+len); 1783 if (error == 0 && log) 1784 goto log; 1785 else 1786 return (error); 1787 } 1788 1789 /* 1790 * Check for any locks in the region to be freed. 1791 */ 1792 1793 if (MANDLOCK(vp, (mode_t)mode)) { 1794 uint64_t length = (len ? len : zp->z_size - off); 1795 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1796 return (error); 1797 } 1798 1799 if (len == 0) { 1800 error = zfs_trunc(zp, off); 1801 } else { 1802 if ((error = zfs_free_range(zp, off, len)) == 0 && 1803 off + len > zp->z_size) 1804 error = zfs_extend(zp, off+len); 1805 } 1806 if (error || !log) 1807 return (error); 1808log: 1809 tx = dmu_tx_create(zfsvfs->z_os); 1810 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1811 zfs_sa_upgrade_txholds(tx, zp); 1812 error = dmu_tx_assign(tx, TXG_NOWAIT); 1813 if (error) { 1814 if (error == ERESTART) { 1815 dmu_tx_wait(tx); 1816 dmu_tx_abort(tx); 1817 goto log; 1818 } 1819 dmu_tx_abort(tx); 1820 return (error); 1821 } 1822 1823 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1824 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1825 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1826 NULL, &zp->z_pflags, 8); 1827 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1828 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1829 ASSERT(error == 0); 1830 1831 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1832 1833 dmu_tx_commit(tx); 1834 return (0); 1835} 1836 1837void 1838zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1839{ 1840 zfsvfs_t zfsvfs; 1841 uint64_t moid, obj, sa_obj, version; 1842 uint64_t sense = ZFS_CASE_SENSITIVE; 1843 uint64_t norm = 0; 1844 nvpair_t *elem; 1845 int error; 1846 int i; 1847 znode_t *rootzp = NULL; 1848 vnode_t vnode; 1849 vattr_t vattr; 1850 znode_t *zp; 1851 zfs_acl_ids_t acl_ids; 1852 1853 /* 1854 * First attempt to create master node. 1855 */ 1856 /* 1857 * In an empty objset, there are no blocks to read and thus 1858 * there can be no i/o errors (which we assert below). 1859 */ 1860 moid = MASTER_NODE_OBJ; 1861 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1862 DMU_OT_NONE, 0, tx); 1863 ASSERT(error == 0); 1864 1865 /* 1866 * Set starting attributes. 1867 */ 1868 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1869 elem = NULL; 1870 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1871 /* For the moment we expect all zpl props to be uint64_ts */ 1872 uint64_t val; 1873 char *name; 1874 1875 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1876 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1877 name = nvpair_name(elem); 1878 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1879 if (val < version) 1880 version = val; 1881 } else { 1882 error = zap_update(os, moid, name, 8, 1, &val, tx); 1883 } 1884 ASSERT(error == 0); 1885 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1886 norm = val; 1887 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1888 sense = val; 1889 } 1890 ASSERT(version != 0); 1891 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1892 1893 /* 1894 * Create zap object used for SA attribute registration 1895 */ 1896 1897 if (version >= ZPL_VERSION_SA) { 1898 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1899 DMU_OT_NONE, 0, tx); 1900 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1901 ASSERT(error == 0); 1902 } else { 1903 sa_obj = 0; 1904 } 1905 /* 1906 * Create a delete queue. 1907 */ 1908 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1909 1910 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1911 ASSERT(error == 0); 1912 1913 /* 1914 * Create root znode. Create minimal znode/vnode/zfsvfs 1915 * to allow zfs_mknode to work. 1916 */ 1917 VATTR_NULL(&vattr); 1918 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1919 vattr.va_type = VDIR; 1920 vattr.va_mode = S_IFDIR|0755; 1921 vattr.va_uid = crgetuid(cr); 1922 vattr.va_gid = crgetgid(cr); 1923 1924 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1925 1926 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1927 zfs_znode_cache_constructor(rootzp, NULL, 0); 1928 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1929 rootzp->z_moved = 0; 1930 rootzp->z_unlinked = 0; 1931 rootzp->z_atime_dirty = 0; 1932 rootzp->z_is_sa = USE_SA(version, os); 1933 1934 vnode.v_type = VDIR; 1935 vnode.v_data = rootzp; 1936 rootzp->z_vnode = &vnode; 1937 1938 zfsvfs.z_os = os; 1939 zfsvfs.z_parent = &zfsvfs; 1940 zfsvfs.z_version = version; 1941 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1942 zfsvfs.z_use_sa = USE_SA(version, os); 1943 zfsvfs.z_norm = norm; 1944 1945 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1946 &zfsvfs.z_attr_table); 1947 1948 ASSERT(error == 0); 1949 1950 /* 1951 * Fold case on file systems that are always or sometimes case 1952 * insensitive. 1953 */ 1954 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1955 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1956 1957 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1958 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1959 offsetof(znode_t, z_link_node)); 1960 1961 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1962 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1963 1964 rootzp->z_zfsvfs = &zfsvfs; 1965 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1966 cr, NULL, &acl_ids)); 1967 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1968 ASSERT3P(zp, ==, rootzp); 1969 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1970 ASSERT(error == 0); 1971 zfs_acl_ids_free(&acl_ids); 1972 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1973 1974 sa_handle_destroy(rootzp->z_sa_hdl); 1975 rootzp->z_vnode = NULL; 1976 kmem_cache_free(znode_cache, rootzp); 1977 1978 /* 1979 * Create shares directory 1980 */ 1981 1982 error = zfs_create_share_dir(&zfsvfs, tx); 1983 1984 ASSERT(error == 0); 1985 1986 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1987 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1988} 1989 1990#endif /* _KERNEL */ 1991 1992static int 1993zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1994{ 1995 uint64_t sa_obj = 0; 1996 int error; 1997 1998 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1999 if (error != 0 && error != ENOENT) 2000 return (error); 2001 2002 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 2003 return (error); 2004} 2005 2006static int 2007zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 2008 dmu_buf_t **db, void *tag) 2009{ 2010 dmu_object_info_t doi; 2011 int error; 2012 2013 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 2014 return (error); 2015 2016 dmu_object_info_from_db(*db, &doi); 2017 if ((doi.doi_bonus_type != DMU_OT_SA && 2018 doi.doi_bonus_type != DMU_OT_ZNODE) || 2019 doi.doi_bonus_type == DMU_OT_ZNODE && 2020 doi.doi_bonus_size < sizeof (znode_phys_t)) { 2021 sa_buf_rele(*db, tag); 2022 return (ENOTSUP); 2023 } 2024 2025 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 2026 if (error != 0) { 2027 sa_buf_rele(*db, tag); 2028 return (error); 2029 } 2030 2031 return (0); 2032} 2033 2034void 2035zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 2036{ 2037 sa_handle_destroy(hdl); 2038 sa_buf_rele(db, tag); 2039} 2040 2041/* 2042 * Given an object number, return its parent object number and whether 2043 * or not the object is an extended attribute directory. 2044 */ 2045static int 2046zfs_obj_to_pobj(sa_handle_t *hdl, sa_attr_type_t *sa_table, uint64_t *pobjp, 2047 int *is_xattrdir) 2048{ 2049 uint64_t parent; 2050 uint64_t pflags; 2051 uint64_t mode; 2052 sa_bulk_attr_t bulk[3]; 2053 int count = 0; 2054 int error; 2055 2056 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2057 &parent, sizeof (parent)); 2058 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2059 &pflags, sizeof (pflags)); 2060 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2061 &mode, sizeof (mode)); 2062 2063 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2064 return (error); 2065 2066 *pobjp = parent; 2067 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2068 2069 return (0); 2070} 2071 2072/* 2073 * Given an object number, return some zpl level statistics 2074 */ 2075static int 2076zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2077 zfs_stat_t *sb) 2078{ 2079 sa_bulk_attr_t bulk[4]; 2080 int count = 0; 2081 2082 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2083 &sb->zs_mode, sizeof (sb->zs_mode)); 2084 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2085 &sb->zs_gen, sizeof (sb->zs_gen)); 2086 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2087 &sb->zs_links, sizeof (sb->zs_links)); 2088 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2089 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2090 2091 return (sa_bulk_lookup(hdl, bulk, count)); 2092} 2093 2094static int 2095zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2096 sa_attr_type_t *sa_table, char *buf, int len) 2097{ 2098 sa_handle_t *sa_hdl; 2099 sa_handle_t *prevhdl = NULL; 2100 dmu_buf_t *prevdb = NULL; 2101 dmu_buf_t *sa_db = NULL; 2102 char *path = buf + len - 1; 2103 int error; 2104 2105 *path = '\0'; 2106 sa_hdl = hdl; 2107 2108 for (;;) { 2109 uint64_t pobj; 2110 char component[MAXNAMELEN + 2]; 2111 size_t complen; 2112 int is_xattrdir; 2113 2114 if (prevdb) 2115 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2116 2117 if ((error = zfs_obj_to_pobj(sa_hdl, sa_table, &pobj, 2118 &is_xattrdir)) != 0) 2119 break; 2120 2121 if (pobj == obj) { 2122 if (path[0] != '/') 2123 *--path = '/'; 2124 break; 2125 } 2126 2127 component[0] = '/'; 2128 if (is_xattrdir) { 2129 (void) sprintf(component + 1, "<xattrdir>"); 2130 } else { 2131 error = zap_value_search(osp, pobj, obj, 2132 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2133 if (error != 0) 2134 break; 2135 } 2136 2137 complen = strlen(component); 2138 path -= complen; 2139 ASSERT(path >= buf); 2140 bcopy(component, path, complen); 2141 obj = pobj; 2142 2143 if (sa_hdl != hdl) { 2144 prevhdl = sa_hdl; 2145 prevdb = sa_db; 2146 } 2147 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2148 if (error != 0) { 2149 sa_hdl = prevhdl; 2150 sa_db = prevdb; 2151 break; 2152 } 2153 } 2154 2155 if (sa_hdl != NULL && sa_hdl != hdl) { 2156 ASSERT(sa_db != NULL); 2157 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2158 } 2159 2160 if (error == 0) 2161 (void) memmove(buf, path, buf + len - path); 2162 2163 return (error); 2164} 2165 2166int 2167zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2168{ 2169 sa_attr_type_t *sa_table; 2170 sa_handle_t *hdl; 2171 dmu_buf_t *db; 2172 int error; 2173 2174 error = zfs_sa_setup(osp, &sa_table); 2175 if (error != 0) 2176 return (error); 2177 2178 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2179 if (error != 0) 2180 return (error); 2181 2182 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2183 2184 zfs_release_sa_handle(hdl, db, FTAG); 2185 return (error); 2186} 2187 2188int 2189zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2190 char *buf, int len) 2191{ 2192 char *path = buf + len - 1; 2193 sa_attr_type_t *sa_table; 2194 sa_handle_t *hdl; 2195 dmu_buf_t *db; 2196 int error; 2197 2198 *path = '\0'; 2199 2200 error = zfs_sa_setup(osp, &sa_table); 2201 if (error != 0) 2202 return (error); 2203 2204 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2205 if (error != 0) 2206 return (error); 2207 2208 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2209 if (error != 0) { 2210 zfs_release_sa_handle(hdl, db, FTAG); 2211 return (error); 2212 } 2213 2214 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2215 2216 zfs_release_sa_handle(hdl, db, FTAG); 2217 return (error); 2218} 2219