zfs_znode.c revision 169325
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 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26/* Portions Copyright 2007 Jeremy Teo */ 27 28#pragma ident "%Z%%M% %I% %E% SMI" 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/vfs.h> 39#include <sys/vnode.h> 40#include <sys/file.h> 41#include <sys/kmem.h> 42#include <sys/cmn_err.h> 43#include <sys/errno.h> 44#include <sys/unistd.h> 45#include <sys/atomic.h> 46#include <sys/zfs_dir.h> 47#include <sys/zfs_acl.h> 48#include <sys/zfs_ioctl.h> 49#include <sys/zfs_rlock.h> 50#include <sys/fs/zfs.h> 51#endif /* _KERNEL */ 52 53#include <sys/dmu.h> 54#include <sys/refcount.h> 55#include <sys/stat.h> 56#include <sys/zap.h> 57#include <sys/zfs_znode.h> 58#include <sys/refcount.h> 59 60/* 61 * Functions needed for userland (ie: libzpool) are not put under 62 * #ifdef_KERNEL; the rest of the functions have dependencies 63 * (such as VFS logic) that will not compile easily in userland. 64 */ 65#ifdef _KERNEL 66struct kmem_cache *znode_cache = NULL; 67 68/*ARGSUSED*/ 69static void 70znode_pageout_func(dmu_buf_t *dbuf, void *user_ptr) 71{ 72 znode_t *zp = user_ptr; 73 vnode_t *vp; 74 75 mutex_enter(&zp->z_lock); 76 vp = ZTOV(zp); 77 if (vp == NULL) { 78 mutex_exit(&zp->z_lock); 79 zfs_znode_free(zp); 80 } else if (vp->v_count == 0) { 81 ZTOV(zp) = NULL; 82 vhold(vp); 83 mutex_exit(&zp->z_lock); 84 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread); 85 vrecycle(vp, curthread); 86 VOP_UNLOCK(vp, 0, curthread); 87 vdrop(vp); 88 zfs_znode_free(zp); 89 } else { 90 /* signal force unmount that this znode can be freed */ 91 zp->z_dbuf = NULL; 92 mutex_exit(&zp->z_lock); 93 } 94} 95 96extern struct vop_vector zfs_vnodeops; 97extern struct vop_vector zfs_fifoops; 98 99/* 100 * XXX: We cannot use this function as a cache constructor, because 101 * there is one global cache for all file systems and we need 102 * to pass vfsp here, which is not possible, because argument 103 * 'cdrarg' is defined at kmem_cache_create() time. 104 */ 105static int 106zfs_znode_cache_constructor(void *buf, void *cdrarg, int kmflags) 107{ 108 znode_t *zp = buf; 109 vnode_t *vp; 110 vfs_t *vfsp = cdrarg; 111 int error; 112 113 if (cdrarg != NULL) { 114 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp); 115 ASSERT(error == 0); 116 zp->z_vnode = vp; 117 vp->v_data = (caddr_t)zp; 118 vhold(vp); 119 vp->v_vnlock->lk_flags |= LK_CANRECURSE; 120 vp->v_vnlock->lk_flags &= ~LK_NOSHARE; 121 } else { 122 zp->z_vnode = NULL; 123 } 124 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 125 rw_init(&zp->z_map_lock, NULL, RW_DEFAULT, NULL); 126 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 127 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 128 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 129 130 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 131 avl_create(&zp->z_range_avl, zfs_range_compare, 132 sizeof (rl_t), offsetof(rl_t, r_node)); 133 134 zp->z_dbuf_held = 0; 135 zp->z_dirlocks = 0; 136 zp->z_lockf = NULL; 137 return (0); 138} 139 140/*ARGSUSED*/ 141static void 142zfs_znode_cache_destructor(void *buf, void *cdarg) 143{ 144 znode_t *zp = buf; 145 146 ASSERT(zp->z_dirlocks == 0); 147 mutex_destroy(&zp->z_lock); 148 rw_destroy(&zp->z_map_lock); 149 rw_destroy(&zp->z_parent_lock); 150 rw_destroy(&zp->z_name_lock); 151 mutex_destroy(&zp->z_acl_lock); 152 mutex_destroy(&zp->z_range_lock); 153 avl_destroy(&zp->z_range_avl); 154 155 ASSERT(zp->z_dbuf_held == 0); 156} 157 158void 159zfs_znode_init(void) 160{ 161 /* 162 * Initialize zcache 163 */ 164 ASSERT(znode_cache == NULL); 165 znode_cache = kmem_cache_create("zfs_znode_cache", 166 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL, 167 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 168} 169 170void 171zfs_znode_fini(void) 172{ 173 /* 174 * Cleanup zcache 175 */ 176 if (znode_cache) 177 kmem_cache_destroy(znode_cache); 178 znode_cache = NULL; 179} 180 181/* 182 * zfs_init_fs - Initialize the zfsvfs struct and the file system 183 * incore "master" object. Verify version compatibility. 184 */ 185int 186zfs_init_fs(zfsvfs_t *zfsvfs, znode_t **zpp, cred_t *cr) 187{ 188 objset_t *os = zfsvfs->z_os; 189 uint64_t version = ZPL_VERSION; 190 int i, error; 191 dmu_object_info_t doi; 192 uint64_t fsid_guid; 193 194 *zpp = NULL; 195 196 /* 197 * XXX - hack to auto-create the pool root filesystem at 198 * the first attempted mount. 199 */ 200 if (dmu_object_info(os, MASTER_NODE_OBJ, &doi) == ENOENT) { 201 dmu_tx_t *tx = dmu_tx_create(os); 202 203 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL); /* master */ 204 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, TRUE, NULL); /* del queue */ 205 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); /* root node */ 206 error = dmu_tx_assign(tx, TXG_WAIT); 207 ASSERT3U(error, ==, 0); 208 zfs_create_fs(os, cr, tx); 209 dmu_tx_commit(tx); 210 } 211 212 error = zap_lookup(os, MASTER_NODE_OBJ, ZPL_VERSION_OBJ, 8, 1, 213 &version); 214 if (error) { 215 return (error); 216 } else if (version != ZPL_VERSION) { 217 (void) printf("Mismatched versions: File system " 218 "is version %lld on-disk format, which is " 219 "incompatible with this software version %lld!", 220 (u_longlong_t)version, ZPL_VERSION); 221 return (ENOTSUP); 222 } 223 224 /* 225 * The fsid is 64 bits, composed of an 8-bit fs type, which 226 * separates our fsid from any other filesystem types, and a 227 * 56-bit objset unique ID. The objset unique ID is unique to 228 * all objsets open on this system, provided by unique_create(). 229 * The 8-bit fs type must be put in the low bits of fsid[1] 230 * because that's where other Solaris filesystems put it. 231 */ 232 fsid_guid = dmu_objset_fsid_guid(os); 233 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0); 234 zfsvfs->z_vfs->vfs_fsid.val[0] = fsid_guid; 235 zfsvfs->z_vfs->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) | 236 zfsvfs->z_vfs->mnt_vfc->vfc_typenum & 0xFF; 237 238 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, 239 &zfsvfs->z_root); 240 if (error) 241 return (error); 242 ASSERT(zfsvfs->z_root != 0); 243 244 /* 245 * Create the per mount vop tables. 246 */ 247 248 /* 249 * Initialize zget mutex's 250 */ 251 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 252 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 253 254 error = zfs_zget(zfsvfs, zfsvfs->z_root, zpp); 255 if (error) 256 return (error); 257 ASSERT3U((*zpp)->z_id, ==, zfsvfs->z_root); 258 259 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, 260 &zfsvfs->z_unlinkedobj); 261 if (error) 262 return (error); 263 264 return (0); 265} 266 267/* 268 * define a couple of values we need available 269 * for both 64 and 32 bit environments. 270 */ 271#ifndef NBITSMINOR64 272#define NBITSMINOR64 32 273#endif 274#ifndef MAXMAJ64 275#define MAXMAJ64 0xffffffffUL 276#endif 277#ifndef MAXMIN64 278#define MAXMIN64 0xffffffffUL 279#endif 280#ifndef major 281#define major(x) ((int)(((u_int)(x) >> 8)&0xff)) /* major number */ 282#endif 283#ifndef minor 284#define minor(x) ((int)((x)&0xffff00ff)) /* minor number */ 285#endif 286 287/* 288 * Create special expldev for ZFS private use. 289 * Can't use standard expldev since it doesn't do 290 * what we want. The standard expldev() takes a 291 * dev32_t in LP64 and expands it to a long dev_t. 292 * We need an interface that takes a dev32_t in ILP32 293 * and expands it to a long dev_t. 294 */ 295static uint64_t 296zfs_expldev(dev_t dev) 297{ 298 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 299} 300/* 301 * Special cmpldev for ZFS private use. 302 * Can't use standard cmpldev since it takes 303 * a long dev_t and compresses it to dev32_t in 304 * LP64. We need to do a compaction of a long dev_t 305 * to a dev32_t in ILP32. 306 */ 307dev_t 308zfs_cmpldev(uint64_t dev) 309{ 310 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 311} 312 313/* 314 * Construct a new znode/vnode and intialize. 315 * 316 * This does not do a call to dmu_set_user() that is 317 * up to the caller to do, in case you don't want to 318 * return the znode 319 */ 320static znode_t * 321zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, uint64_t obj_num, int blksz) 322{ 323 znode_t *zp; 324 vnode_t *vp; 325 int error; 326 327 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 328 zfs_znode_cache_constructor(zp, zfsvfs->z_vfs, 0); 329 330 ASSERT(zp->z_dirlocks == NULL); 331 332 zp->z_phys = db->db_data; 333 zp->z_zfsvfs = zfsvfs; 334 zp->z_unlinked = 0; 335 zp->z_atime_dirty = 0; 336 zp->z_dbuf_held = 0; 337 zp->z_mapcnt = 0; 338 zp->z_last_itx = 0; 339 zp->z_dbuf = db; 340 zp->z_id = obj_num; 341 zp->z_blksz = blksz; 342 zp->z_seq = 0x7A4653; 343 zp->z_sync_cnt = 0; 344 345 mutex_enter(&zfsvfs->z_znodes_lock); 346 list_insert_tail(&zfsvfs->z_all_znodes, zp); 347 mutex_exit(&zfsvfs->z_znodes_lock); 348 349 vp = ZTOV(zp); 350 if (vp == NULL) 351 return (zp); 352 353 error = insmntque(vp, zfsvfs->z_vfs); 354 KASSERT(error == 0, ("insmntque() failed: error %d", error)); 355 356 vp->v_type = IFTOVT((mode_t)zp->z_phys->zp_mode); 357 switch (vp->v_type) { 358 case VDIR: 359 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 360 break; 361 case VFIFO: 362 vp->v_op = &zfs_fifoops; 363 break; 364 } 365 366 return (zp); 367} 368 369static void 370zfs_znode_dmu_init(znode_t *zp) 371{ 372 znode_t *nzp; 373 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 374 dmu_buf_t *db = zp->z_dbuf; 375 376 mutex_enter(&zp->z_lock); 377 378 nzp = dmu_buf_set_user_ie(db, zp, &zp->z_phys, znode_pageout_func); 379 380 /* 381 * there should be no 382 * concurrent zgets on this object. 383 */ 384 ASSERT3P(nzp, ==, NULL); 385 386 /* 387 * Slap on VROOT if we are the root znode 388 */ 389 if (zp->z_id == zfsvfs->z_root) { 390 ZTOV(zp)->v_flag |= VROOT; 391 } 392 393 ASSERT(zp->z_dbuf_held == 0); 394 zp->z_dbuf_held = 1; 395 VFS_HOLD(zfsvfs->z_vfs); 396 mutex_exit(&zp->z_lock); 397} 398 399/* 400 * Create a new DMU object to hold a zfs znode. 401 * 402 * IN: dzp - parent directory for new znode 403 * vap - file attributes for new znode 404 * tx - dmu transaction id for zap operations 405 * cr - credentials of caller 406 * flag - flags: 407 * IS_ROOT_NODE - new object will be root 408 * IS_XATTR - new object is an attribute 409 * IS_REPLAY - intent log replay 410 * 411 * OUT: oid - ID of created object 412 * 413 */ 414void 415zfs_mknode(znode_t *dzp, vattr_t *vap, uint64_t *oid, dmu_tx_t *tx, cred_t *cr, 416 uint_t flag, znode_t **zpp, int bonuslen) 417{ 418 dmu_buf_t *dbp; 419 znode_phys_t *pzp; 420 znode_t *zp; 421 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 422 timestruc_t now; 423 uint64_t gen; 424 int err; 425 426 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 427 428 if (zfsvfs->z_assign >= TXG_INITIAL) { /* ZIL replay */ 429 *oid = vap->va_nodeid; 430 flag |= IS_REPLAY; 431 now = vap->va_ctime; /* see zfs_replay_create() */ 432 gen = vap->va_nblocks; /* ditto */ 433 } else { 434 *oid = 0; 435 gethrestime(&now); 436 gen = dmu_tx_get_txg(tx); 437 } 438 439 /* 440 * Create a new DMU object. 441 */ 442 /* 443 * There's currently no mechanism for pre-reading the blocks that will 444 * be to needed allocate a new object, so we accept the small chance 445 * that there will be an i/o error and we will fail one of the 446 * assertions below. 447 */ 448 if (vap->va_type == VDIR) { 449 if (flag & IS_REPLAY) { 450 err = zap_create_claim(zfsvfs->z_os, *oid, 451 DMU_OT_DIRECTORY_CONTENTS, 452 DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx); 453 ASSERT3U(err, ==, 0); 454 } else { 455 *oid = zap_create(zfsvfs->z_os, 456 DMU_OT_DIRECTORY_CONTENTS, 457 DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx); 458 } 459 } else { 460 if (flag & IS_REPLAY) { 461 err = dmu_object_claim(zfsvfs->z_os, *oid, 462 DMU_OT_PLAIN_FILE_CONTENTS, 0, 463 DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx); 464 ASSERT3U(err, ==, 0); 465 } else { 466 *oid = dmu_object_alloc(zfsvfs->z_os, 467 DMU_OT_PLAIN_FILE_CONTENTS, 0, 468 DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx); 469 } 470 } 471 VERIFY(0 == dmu_bonus_hold(zfsvfs->z_os, *oid, NULL, &dbp)); 472 dmu_buf_will_dirty(dbp, tx); 473 474 /* 475 * Initialize the znode physical data to zero. 476 */ 477 ASSERT(dbp->db_size >= sizeof (znode_phys_t)); 478 bzero(dbp->db_data, dbp->db_size); 479 pzp = dbp->db_data; 480 481 /* 482 * If this is the root, fix up the half-initialized parent pointer 483 * to reference the just-allocated physical data area. 484 */ 485 if (flag & IS_ROOT_NODE) { 486 dzp->z_phys = pzp; 487 dzp->z_id = *oid; 488 } 489 490 /* 491 * If parent is an xattr, so am I. 492 */ 493 if (dzp->z_phys->zp_flags & ZFS_XATTR) 494 flag |= IS_XATTR; 495 496 if (vap->va_type == VBLK || vap->va_type == VCHR) { 497 pzp->zp_rdev = zfs_expldev(vap->va_rdev); 498 } 499 500 if (vap->va_type == VDIR) { 501 pzp->zp_size = 2; /* contents ("." and "..") */ 502 pzp->zp_links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 503 } 504 505 pzp->zp_parent = dzp->z_id; 506 if (flag & IS_XATTR) 507 pzp->zp_flags |= ZFS_XATTR; 508 509 pzp->zp_gen = gen; 510 511 ZFS_TIME_ENCODE(&now, pzp->zp_crtime); 512 ZFS_TIME_ENCODE(&now, pzp->zp_ctime); 513 514 if (vap->va_mask & AT_ATIME) { 515 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 516 } else { 517 ZFS_TIME_ENCODE(&now, pzp->zp_atime); 518 } 519 520 if (vap->va_mask & AT_MTIME) { 521 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 522 } else { 523 ZFS_TIME_ENCODE(&now, pzp->zp_mtime); 524 } 525 526 pzp->zp_mode = MAKEIMODE(vap->va_type, vap->va_mode); 527 zp = zfs_znode_alloc(zfsvfs, dbp, *oid, 0); 528 529 zfs_perm_init(zp, dzp, flag, vap, tx, cr); 530 531 if (zpp) { 532 kmutex_t *hash_mtx = ZFS_OBJ_MUTEX(zp); 533 534 mutex_enter(hash_mtx); 535 zfs_znode_dmu_init(zp); 536 mutex_exit(hash_mtx); 537 538 *zpp = zp; 539 } else { 540 if (ZTOV(zp) != NULL) 541 ZTOV(zp)->v_count = 0; 542 dmu_buf_rele(dbp, NULL); 543 zfs_znode_free(zp); 544 } 545} 546 547int 548zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 549{ 550 dmu_object_info_t doi; 551 dmu_buf_t *db; 552 znode_t *zp; 553 vnode_t *vp; 554 int err; 555 556 *zpp = NULL; 557 558 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 559 560 err = dmu_bonus_hold(zfsvfs->z_os, obj_num, NULL, &db); 561 if (err) { 562 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 563 return (err); 564 } 565 566 dmu_object_info_from_db(db, &doi); 567 if (doi.doi_bonus_type != DMU_OT_ZNODE || 568 doi.doi_bonus_size < sizeof (znode_phys_t)) { 569 dmu_buf_rele(db, NULL); 570 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 571 return (EINVAL); 572 } 573 574 ASSERT(db->db_object == obj_num); 575 ASSERT(db->db_offset == -1); 576 ASSERT(db->db_data != NULL); 577 578 zp = dmu_buf_get_user(db); 579 580 if (zp != NULL) { 581 mutex_enter(&zp->z_lock); 582 583 ASSERT3U(zp->z_id, ==, obj_num); 584 if (zp->z_unlinked) { 585 dmu_buf_rele(db, NULL); 586 mutex_exit(&zp->z_lock); 587 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 588 return (ENOENT); 589 } else if (zp->z_dbuf_held) { 590 dmu_buf_rele(db, NULL); 591 } else { 592 zp->z_dbuf_held = 1; 593 VFS_HOLD(zfsvfs->z_vfs); 594 } 595 596 if (ZTOV(zp) != NULL) 597 VN_HOLD(ZTOV(zp)); 598 else { 599 err = getnewvnode("zfs", zfsvfs->z_vfs, &zfs_vnodeops, 600 &zp->z_vnode); 601 ASSERT(err == 0); 602 vp = ZTOV(zp); 603 vp->v_data = (caddr_t)zp; 604 vhold(vp); 605 vp->v_vnlock->lk_flags |= LK_CANRECURSE; 606 vp->v_vnlock->lk_flags &= ~LK_NOSHARE; 607 vp->v_type = IFTOVT((mode_t)zp->z_phys->zp_mode); 608 if (vp->v_type == VDIR) 609 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 610 err = insmntque(vp, zfsvfs->z_vfs); 611 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 612 } 613 mutex_exit(&zp->z_lock); 614 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 615 *zpp = zp; 616 return (0); 617 } 618 619 /* 620 * Not found create new znode/vnode 621 */ 622 zp = zfs_znode_alloc(zfsvfs, db, obj_num, doi.doi_data_block_size); 623 ASSERT3U(zp->z_id, ==, obj_num); 624 zfs_znode_dmu_init(zp); 625 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 626 *zpp = zp; 627 return (0); 628} 629 630void 631zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 632{ 633 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 634 int error; 635 636 ZFS_OBJ_HOLD_ENTER(zfsvfs, zp->z_id); 637 if (zp->z_phys->zp_acl.z_acl_extern_obj) { 638 error = dmu_object_free(zfsvfs->z_os, 639 zp->z_phys->zp_acl.z_acl_extern_obj, tx); 640 ASSERT3U(error, ==, 0); 641 } 642 error = dmu_object_free(zfsvfs->z_os, zp->z_id, tx); 643 ASSERT3U(error, ==, 0); 644 zp->z_dbuf_held = 0; 645 ZFS_OBJ_HOLD_EXIT(zfsvfs, zp->z_id); 646 dmu_buf_rele(zp->z_dbuf, NULL); 647} 648 649void 650zfs_zinactive(znode_t *zp) 651{ 652 vnode_t *vp = ZTOV(zp); 653 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 654 uint64_t z_id = zp->z_id; 655 656 ASSERT(zp->z_dbuf_held && zp->z_phys); 657 658 /* 659 * Don't allow a zfs_zget() while were trying to release this znode 660 */ 661 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 662 663 mutex_enter(&zp->z_lock); 664 VI_LOCK(vp); 665 if (vp->v_count > 0) { 666 /* 667 * If the hold count is greater than zero, somebody has 668 * obtained a new reference on this znode while we were 669 * processing it here, so we are done. 670 */ 671 VI_UNLOCK(vp); 672 mutex_exit(&zp->z_lock); 673 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 674 return; 675 } 676 VI_UNLOCK(vp); 677 678 /* 679 * If this was the last reference to a file with no links, 680 * remove the file from the file system. 681 */ 682 if (zp->z_unlinked) { 683 ZTOV(zp) = NULL; 684 mutex_exit(&zp->z_lock); 685 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 686 ASSERT(vp->v_count == 0); 687 vrecycle(vp, curthread); 688 zfs_rmnode(zp); 689 VFS_RELE(zfsvfs->z_vfs); 690 return; 691 } 692 ASSERT(zp->z_phys); 693 ASSERT(zp->z_dbuf_held); 694 mutex_exit(&zp->z_lock); 695 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 696} 697 698void 699zfs_znode_free(znode_t *zp) 700{ 701 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 702 703 mutex_enter(&zfsvfs->z_znodes_lock); 704 list_remove(&zfsvfs->z_all_znodes, zp); 705 mutex_exit(&zfsvfs->z_znodes_lock); 706 707 kmem_cache_free(znode_cache, zp); 708} 709 710void 711zfs_time_stamper_locked(znode_t *zp, uint_t flag, dmu_tx_t *tx) 712{ 713 timestruc_t now; 714 715 ASSERT(MUTEX_HELD(&zp->z_lock)); 716 717 gethrestime(&now); 718 719 if (tx) { 720 dmu_buf_will_dirty(zp->z_dbuf, tx); 721 zp->z_atime_dirty = 0; 722 zp->z_seq++; 723 } else { 724 zp->z_atime_dirty = 1; 725 } 726 727 if (flag & AT_ATIME) 728 ZFS_TIME_ENCODE(&now, zp->z_phys->zp_atime); 729 730 if (flag & AT_MTIME) 731 ZFS_TIME_ENCODE(&now, zp->z_phys->zp_mtime); 732 733 if (flag & AT_CTIME) 734 ZFS_TIME_ENCODE(&now, zp->z_phys->zp_ctime); 735} 736 737/* 738 * Update the requested znode timestamps with the current time. 739 * If we are in a transaction, then go ahead and mark the znode 740 * dirty in the transaction so the timestamps will go to disk. 741 * Otherwise, we will get pushed next time the znode is updated 742 * in a transaction, or when this znode eventually goes inactive. 743 * 744 * Why is this OK? 745 * 1 - Only the ACCESS time is ever updated outside of a transaction. 746 * 2 - Multiple consecutive updates will be collapsed into a single 747 * znode update by the transaction grouping semantics of the DMU. 748 */ 749void 750zfs_time_stamper(znode_t *zp, uint_t flag, dmu_tx_t *tx) 751{ 752 mutex_enter(&zp->z_lock); 753 zfs_time_stamper_locked(zp, flag, tx); 754 mutex_exit(&zp->z_lock); 755} 756 757/* 758 * Grow the block size for a file. 759 * 760 * IN: zp - znode of file to free data in. 761 * size - requested block size 762 * tx - open transaction. 763 * 764 * NOTE: this function assumes that the znode is write locked. 765 */ 766void 767zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 768{ 769 int error; 770 u_longlong_t dummy; 771 772 if (size <= zp->z_blksz) 773 return; 774 /* 775 * If the file size is already greater than the current blocksize, 776 * we will not grow. If there is more than one block in a file, 777 * the blocksize cannot change. 778 */ 779 if (zp->z_blksz && zp->z_phys->zp_size > zp->z_blksz) 780 return; 781 782 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 783 size, 0, tx); 784 if (error == ENOTSUP) 785 return; 786 ASSERT3U(error, ==, 0); 787 788 /* What blocksize did we actually get? */ 789 dmu_object_size_from_db(zp->z_dbuf, &zp->z_blksz, &dummy); 790} 791 792/* 793 * Free space in a file. 794 * 795 * IN: zp - znode of file to free data in. 796 * off - start of section to free. 797 * len - length of section to free (0 => to EOF). 798 * flag - current file open mode flags. 799 * 800 * RETURN: 0 if success 801 * error code if failure 802 */ 803int 804zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 805{ 806 vnode_t *vp = ZTOV(zp); 807 dmu_tx_t *tx; 808 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 809 zilog_t *zilog = zfsvfs->z_log; 810 rl_t *rl; 811 uint64_t end = off + len; 812 uint64_t size, new_blksz; 813 int error; 814 815 if (ZTOV(zp)->v_type == VFIFO) 816 return (0); 817 818 /* 819 * If we will change zp_size then lock the whole file, 820 * otherwise just lock the range being freed. 821 */ 822 if (len == 0 || off + len > zp->z_phys->zp_size) { 823 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 824 } else { 825 rl = zfs_range_lock(zp, off, len, RL_WRITER); 826 /* recheck, in case zp_size changed */ 827 if (off + len > zp->z_phys->zp_size) { 828 /* lost race: file size changed, lock whole file */ 829 zfs_range_unlock(rl); 830 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 831 } 832 } 833 834 /* 835 * Nothing to do if file already at desired length. 836 */ 837 size = zp->z_phys->zp_size; 838 if (len == 0 && size == off && off != 0) { 839 zfs_range_unlock(rl); 840 return (0); 841 } 842 843 tx = dmu_tx_create(zfsvfs->z_os); 844 dmu_tx_hold_bonus(tx, zp->z_id); 845 new_blksz = 0; 846 if (end > size && 847 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 848 /* 849 * We are growing the file past the current block size. 850 */ 851 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 852 ASSERT(!ISP2(zp->z_blksz)); 853 new_blksz = MIN(end, SPA_MAXBLOCKSIZE); 854 } else { 855 new_blksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 856 } 857 dmu_tx_hold_write(tx, zp->z_id, 0, MIN(end, new_blksz)); 858 } else if (off < size) { 859 /* 860 * If len == 0, we are truncating the file. 861 */ 862 dmu_tx_hold_free(tx, zp->z_id, off, len ? len : DMU_OBJECT_END); 863 } 864 865 error = dmu_tx_assign(tx, zfsvfs->z_assign); 866 if (error) { 867 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) 868 dmu_tx_wait(tx); 869 dmu_tx_abort(tx); 870 zfs_range_unlock(rl); 871 return (error); 872 } 873 874 if (new_blksz) 875 zfs_grow_blocksize(zp, new_blksz, tx); 876 877 if (end > size || len == 0) 878 zp->z_phys->zp_size = end; 879 880 if (off < size) { 881 objset_t *os = zfsvfs->z_os; 882 uint64_t rlen = len; 883 884 if (len == 0) 885 rlen = -1; 886 else if (end > size) 887 rlen = size - off; 888 VERIFY(0 == dmu_free_range(os, zp->z_id, off, rlen, tx)); 889 } 890 891 if (log) { 892 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 893 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 894 } 895 896 zfs_range_unlock(rl); 897 898 dmu_tx_commit(tx); 899 900 /* 901 * Clear any mapped pages in the truncated region. This has to 902 * happen outside of the transaction to avoid the possibility of 903 * a deadlock with someone trying to push a page that we are 904 * about to invalidate. 905 */ 906 rw_enter(&zp->z_map_lock, RW_WRITER); 907 if (end > size) 908 vnode_pager_setsize(vp, end); 909 else if (len == 0) { 910#if 0 911 error = vtruncbuf(vp, curthread->td_ucred, curthread, end, PAGE_SIZE); 912#else 913 error = vinvalbuf(vp, V_SAVE, curthread, 0, 0); 914 vnode_pager_setsize(vp, end); 915#endif 916 } 917 rw_exit(&zp->z_map_lock); 918 919 return (0); 920} 921 922void 923zfs_create_fs(objset_t *os, cred_t *cr, dmu_tx_t *tx) 924{ 925 zfsvfs_t zfsvfs; 926 uint64_t moid, doid, roid = 0; 927 uint64_t version = ZPL_VERSION; 928 int error; 929 znode_t *rootzp = NULL; 930 vattr_t vattr; 931 932 /* 933 * First attempt to create master node. 934 */ 935 /* 936 * In an empty objset, there are no blocks to read and thus 937 * there can be no i/o errors (which we assert below). 938 */ 939 moid = MASTER_NODE_OBJ; 940 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 941 DMU_OT_NONE, 0, tx); 942 ASSERT(error == 0); 943 944 /* 945 * Set starting attributes. 946 */ 947 948 error = zap_update(os, moid, ZPL_VERSION_OBJ, 8, 1, &version, tx); 949 ASSERT(error == 0); 950 951 /* 952 * Create a delete queue. 953 */ 954 doid = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 955 956 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &doid, tx); 957 ASSERT(error == 0); 958 959 /* 960 * Create root znode. Create minimal znode/vnode/zfsvfs 961 * to allow zfs_mknode to work. 962 */ 963 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 964 vattr.va_type = VDIR; 965 vattr.va_mode = S_IFDIR|0755; 966 vattr.va_uid = UID_ROOT; 967 vattr.va_gid = GID_WHEEL; 968 969 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 970 zfs_znode_cache_constructor(rootzp, NULL, 0); 971 rootzp->z_zfsvfs = &zfsvfs; 972 rootzp->z_unlinked = 0; 973 rootzp->z_atime_dirty = 0; 974 rootzp->z_dbuf_held = 0; 975 976 bzero(&zfsvfs, sizeof (zfsvfs_t)); 977 978 zfsvfs.z_os = os; 979 zfsvfs.z_assign = TXG_NOWAIT; 980 zfsvfs.z_parent = &zfsvfs; 981 982 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 983 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 984 offsetof(znode_t, z_link_node)); 985 986 zfs_mknode(rootzp, &vattr, &roid, tx, cr, IS_ROOT_NODE, NULL, 0); 987 ASSERT3U(rootzp->z_id, ==, roid); 988 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &roid, tx); 989 ASSERT(error == 0); 990 991 mutex_destroy(&zfsvfs.z_znodes_lock); 992 kmem_cache_free(znode_cache, rootzp); 993} 994#endif /* _KERNEL */ 995 996/* 997 * Given an object number, return its parent object number and whether 998 * or not the object is an extended attribute directory. 999 */ 1000static int 1001zfs_obj_to_pobj(objset_t *osp, uint64_t obj, uint64_t *pobjp, int *is_xattrdir) 1002{ 1003 dmu_buf_t *db; 1004 dmu_object_info_t doi; 1005 znode_phys_t *zp; 1006 int error; 1007 1008 if ((error = dmu_bonus_hold(osp, obj, FTAG, &db)) != 0) 1009 return (error); 1010 1011 dmu_object_info_from_db(db, &doi); 1012 if (doi.doi_bonus_type != DMU_OT_ZNODE || 1013 doi.doi_bonus_size < sizeof (znode_phys_t)) { 1014 dmu_buf_rele(db, FTAG); 1015 return (EINVAL); 1016 } 1017 1018 zp = db->db_data; 1019 *pobjp = zp->zp_parent; 1020 *is_xattrdir = ((zp->zp_flags & ZFS_XATTR) != 0) && 1021 S_ISDIR(zp->zp_mode); 1022 dmu_buf_rele(db, FTAG); 1023 1024 return (0); 1025} 1026 1027int 1028zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 1029{ 1030 char *path = buf + len - 1; 1031 int error; 1032 1033 *path = '\0'; 1034 1035 for (;;) { 1036 uint64_t pobj; 1037 char component[MAXNAMELEN + 2]; 1038 size_t complen; 1039 int is_xattrdir; 1040 1041 if ((error = zfs_obj_to_pobj(osp, obj, &pobj, 1042 &is_xattrdir)) != 0) 1043 break; 1044 1045 if (pobj == obj) { 1046 if (path[0] != '/') 1047 *--path = '/'; 1048 break; 1049 } 1050 1051 component[0] = '/'; 1052 if (is_xattrdir) { 1053 (void) sprintf(component + 1, "<xattrdir>"); 1054 } else { 1055 error = zap_value_search(osp, pobj, obj, component + 1); 1056 if (error != 0) 1057 break; 1058 } 1059 1060 complen = strlen(component); 1061 path -= complen; 1062 ASSERT(path >= buf); 1063 bcopy(component, path, complen); 1064 obj = pobj; 1065 } 1066 1067 if (error == 0) 1068 (void) memmove(buf, path, buf + len - path); 1069 return (error); 1070} 1071