zfs_znode.c revision 210470
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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26/* Portions Copyright 2007 Jeremy Teo */ 27 28#ifdef _KERNEL 29#include <sys/types.h> 30#include <sys/param.h> 31#include <sys/time.h> 32#include <sys/systm.h> 33#include <sys/sysmacros.h> 34#include <sys/resource.h> 35#include <sys/mntent.h> 36#include <sys/u8_textprep.h> 37#include <sys/dsl_dataset.h> 38#include <sys/vfs.h> 39#include <sys/vnode.h> 40#include <sys/file.h> 41#include <sys/kmem.h> 42#include <sys/errno.h> 43#include <sys/unistd.h> 44#include <sys/atomic.h> 45#include <sys/zfs_dir.h> 46#include <sys/zfs_acl.h> 47#include <sys/zfs_ioctl.h> 48#include <sys/zfs_rlock.h> 49#include <sys/zfs_fuid.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/refcount.h> 60 61#include "zfs_prop.h" 62 63/* Used by fstat(1). */ 64SYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t), 65 "sizeof(znode_t)"); 66 67/* 68 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 69 * turned on when DEBUG is also defined. 70 */ 71#ifdef DEBUG 72#define ZNODE_STATS 73#endif /* DEBUG */ 74 75#ifdef ZNODE_STATS 76#define ZNODE_STAT_ADD(stat) ((stat)++) 77#else 78#define ZNODE_STAT_ADD(stat) /* nothing */ 79#endif /* ZNODE_STATS */ 80 81#define POINTER_IS_VALID(p) (!((uintptr_t)(p) & 0x3)) 82#define POINTER_INVALIDATE(pp) (*(pp) = (void *)((uintptr_t)(*(pp)) | 0x1)) 83 84/* 85 * Functions needed for userland (ie: libzpool) are not put under 86 * #ifdef_KERNEL; the rest of the functions have dependencies 87 * (such as VFS logic) that will not compile easily in userland. 88 */ 89#ifdef _KERNEL 90/* 91 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 92 * be freed before it can be safely accessed. 93 */ 94krwlock_t zfsvfs_lock; 95 96static kmem_cache_t *znode_cache = NULL; 97 98/*ARGSUSED*/ 99static void 100znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 101{ 102#if 1 /* XXXPJD: From OpenSolaris. */ 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#else /* XXXPJD */ 109 znode_t *zp = user_ptr; 110 vnode_t *vp; 111 112 mutex_enter(&zp->z_lock); 113 zp->z_dbuf = NULL; 114 vp = ZTOV(zp); 115 if (vp == NULL) { 116 mutex_exit(&zp->z_lock); 117 zfs_znode_free(zp); 118 } else if (vp->v_count == 0) { 119 zp->z_vnode = NULL; 120 vhold(vp); 121 mutex_exit(&zp->z_lock); 122 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread); 123 vrecycle(vp, curthread); 124 VOP_UNLOCK(vp, 0); 125 vdrop(vp); 126 zfs_znode_free(zp); 127 } else { 128 mutex_exit(&zp->z_lock); 129 } 130#endif 131} 132 133extern struct vop_vector zfs_vnodeops; 134extern struct vop_vector zfs_fifoops; 135extern struct vop_vector zfs_shareops; 136 137/* 138 * XXX: We cannot use this function as a cache constructor, because 139 * there is one global cache for all file systems and we need 140 * to pass vfsp here, which is not possible, because argument 141 * 'cdrarg' is defined at kmem_cache_create() time. 142 */ 143static int 144zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 145{ 146 znode_t *zp = buf; 147 vnode_t *vp; 148 vfs_t *vfsp = arg; 149 int error; 150 151 POINTER_INVALIDATE(&zp->z_zfsvfs); 152 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 153 154 if (vfsp != NULL) { 155 error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp); 156 if (error != 0 && (kmflags & KM_NOSLEEP)) 157 return (-1); 158 ASSERT(error == 0); 159 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 160 zp->z_vnode = vp; 161 vp->v_data = (caddr_t)zp; 162 VN_LOCK_AREC(vp); 163 } else { 164 zp->z_vnode = NULL; 165 } 166 167 list_link_init(&zp->z_link_node); 168 169 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 170 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 171 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 172 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 173 174 mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 175 avl_create(&zp->z_range_avl, zfs_range_compare, 176 sizeof (rl_t), offsetof(rl_t, r_node)); 177 178 zp->z_dbuf = NULL; 179 zp->z_dirlocks = NULL; 180 return (0); 181} 182 183/*ARGSUSED*/ 184static void 185zfs_znode_cache_destructor(void *buf, void *arg) 186{ 187 znode_t *zp = buf; 188 189 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 190 ASSERT(ZTOV(zp) == NULL); 191 vn_free(ZTOV(zp)); 192 ASSERT(!list_link_active(&zp->z_link_node)); 193 mutex_destroy(&zp->z_lock); 194 rw_destroy(&zp->z_parent_lock); 195 rw_destroy(&zp->z_name_lock); 196 mutex_destroy(&zp->z_acl_lock); 197 avl_destroy(&zp->z_range_avl); 198 mutex_destroy(&zp->z_range_lock); 199 200 ASSERT(zp->z_dbuf == NULL); 201 ASSERT(zp->z_dirlocks == NULL); 202} 203 204#ifdef ZNODE_STATS 205static struct { 206 uint64_t zms_zfsvfs_invalid; 207 uint64_t zms_zfsvfs_recheck1; 208 uint64_t zms_zfsvfs_unmounted; 209 uint64_t zms_zfsvfs_recheck2; 210 uint64_t zms_obj_held; 211 uint64_t zms_vnode_locked; 212 uint64_t zms_not_only_dnlc; 213} znode_move_stats; 214#endif /* ZNODE_STATS */ 215 216#if defined(sun) 217static void 218zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 219{ 220 vnode_t *vp; 221 222 /* Copy fields. */ 223 nzp->z_zfsvfs = ozp->z_zfsvfs; 224 225 /* Swap vnodes. */ 226 vp = nzp->z_vnode; 227 nzp->z_vnode = ozp->z_vnode; 228 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 229 ZTOV(ozp)->v_data = ozp; 230 ZTOV(nzp)->v_data = nzp; 231 232 nzp->z_id = ozp->z_id; 233 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 234 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 235 nzp->z_unlinked = ozp->z_unlinked; 236 nzp->z_atime_dirty = ozp->z_atime_dirty; 237 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 238 nzp->z_blksz = ozp->z_blksz; 239 nzp->z_seq = ozp->z_seq; 240 nzp->z_mapcnt = ozp->z_mapcnt; 241 nzp->z_last_itx = ozp->z_last_itx; 242 nzp->z_gen = ozp->z_gen; 243 nzp->z_sync_cnt = ozp->z_sync_cnt; 244 nzp->z_phys = ozp->z_phys; 245 nzp->z_dbuf = ozp->z_dbuf; 246 247 /* Update back pointers. */ 248 (void) dmu_buf_update_user(nzp->z_dbuf, ozp, nzp, &nzp->z_phys, 249 znode_evict_error); 250 251 /* 252 * Invalidate the original znode by clearing fields that provide a 253 * pointer back to the znode. Set the low bit of the vfs pointer to 254 * ensure that zfs_znode_move() recognizes the znode as invalid in any 255 * subsequent callback. 256 */ 257 ozp->z_dbuf = NULL; 258 POINTER_INVALIDATE(&ozp->z_zfsvfs); 259} 260 261/*ARGSUSED*/ 262static kmem_cbrc_t 263zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 264{ 265 znode_t *ozp = buf, *nzp = newbuf; 266 zfsvfs_t *zfsvfs; 267 vnode_t *vp; 268 269 /* 270 * The znode is on the file system's list of known znodes if the vfs 271 * pointer is valid. We set the low bit of the vfs pointer when freeing 272 * the znode to invalidate it, and the memory patterns written by kmem 273 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 274 * created znode sets the vfs pointer last of all to indicate that the 275 * znode is known and in a valid state to be moved by this function. 276 */ 277 zfsvfs = ozp->z_zfsvfs; 278 if (!POINTER_IS_VALID(zfsvfs)) { 279 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 280 return (KMEM_CBRC_DONT_KNOW); 281 } 282 283 /* 284 * Close a small window in which it's possible that the filesystem could 285 * be unmounted and freed, and zfsvfs, though valid in the previous 286 * statement, could point to unrelated memory by the time we try to 287 * prevent the filesystem from being unmounted. 288 */ 289 rw_enter(&zfsvfs_lock, RW_WRITER); 290 if (zfsvfs != ozp->z_zfsvfs) { 291 rw_exit(&zfsvfs_lock); 292 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 293 return (KMEM_CBRC_DONT_KNOW); 294 } 295 296 /* 297 * If the znode is still valid, then so is the file system. We know that 298 * no valid file system can be freed while we hold zfsvfs_lock, so we 299 * can safely ensure that the filesystem is not and will not be 300 * unmounted. The next statement is equivalent to ZFS_ENTER(). 301 */ 302 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 303 if (zfsvfs->z_unmounted) { 304 ZFS_EXIT(zfsvfs); 305 rw_exit(&zfsvfs_lock); 306 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 307 return (KMEM_CBRC_DONT_KNOW); 308 } 309 rw_exit(&zfsvfs_lock); 310 311 mutex_enter(&zfsvfs->z_znodes_lock); 312 /* 313 * Recheck the vfs pointer in case the znode was removed just before 314 * acquiring the lock. 315 */ 316 if (zfsvfs != ozp->z_zfsvfs) { 317 mutex_exit(&zfsvfs->z_znodes_lock); 318 ZFS_EXIT(zfsvfs); 319 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 320 return (KMEM_CBRC_DONT_KNOW); 321 } 322 323 /* 324 * At this point we know that as long as we hold z_znodes_lock, the 325 * znode cannot be freed and fields within the znode can be safely 326 * accessed. Now, prevent a race with zfs_zget(). 327 */ 328 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 329 mutex_exit(&zfsvfs->z_znodes_lock); 330 ZFS_EXIT(zfsvfs); 331 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 332 return (KMEM_CBRC_LATER); 333 } 334 335 vp = ZTOV(ozp); 336 if (mutex_tryenter(&vp->v_lock) == 0) { 337 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 338 mutex_exit(&zfsvfs->z_znodes_lock); 339 ZFS_EXIT(zfsvfs); 340 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 341 return (KMEM_CBRC_LATER); 342 } 343 344 /* Only move znodes that are referenced _only_ by the DNLC. */ 345 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 346 mutex_exit(&vp->v_lock); 347 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 348 mutex_exit(&zfsvfs->z_znodes_lock); 349 ZFS_EXIT(zfsvfs); 350 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 351 return (KMEM_CBRC_LATER); 352 } 353 354 /* 355 * The znode is known and in a valid state to move. We're holding the 356 * locks needed to execute the critical section. 357 */ 358 zfs_znode_move_impl(ozp, nzp); 359 mutex_exit(&vp->v_lock); 360 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 361 362 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 363 mutex_exit(&zfsvfs->z_znodes_lock); 364 ZFS_EXIT(zfsvfs); 365 366 return (KMEM_CBRC_YES); 367} 368#endif /* sun */ 369 370void 371zfs_znode_init(void) 372{ 373 /* 374 * Initialize zcache 375 */ 376 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 377 ASSERT(znode_cache == NULL); 378 znode_cache = kmem_cache_create("zfs_znode_cache", 379 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL, 380 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 381#if defined(sun) 382 kmem_cache_set_move(znode_cache, zfs_znode_move); 383#endif 384} 385 386void 387zfs_znode_fini(void) 388{ 389 /* 390 * Cleanup zcache 391 */ 392 if (znode_cache) 393 kmem_cache_destroy(znode_cache); 394 znode_cache = NULL; 395 rw_destroy(&zfsvfs_lock); 396} 397 398int 399zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 400{ 401 zfs_acl_ids_t acl_ids; 402 vattr_t vattr; 403 znode_t *sharezp; 404 vnode_t *vp, vnode; 405 znode_t *zp; 406 int error; 407 408 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 409 vattr.va_type = VDIR; 410 vattr.va_mode = S_IFDIR|0555; 411 vattr.va_uid = crgetuid(kcred); 412 vattr.va_gid = crgetgid(kcred); 413 414 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 415 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0); 416 sharezp->z_unlinked = 0; 417 sharezp->z_atime_dirty = 0; 418 sharezp->z_zfsvfs = zfsvfs; 419 420 sharezp->z_vnode = &vnode; 421 vnode.v_data = sharezp; 422 423 vp = ZTOV(sharezp); 424 vp->v_type = VDIR; 425 426 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 427 kcred, NULL, &acl_ids)); 428 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, 429 &zp, 0, &acl_ids); 430 ASSERT3P(zp, ==, sharezp); 431 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 432 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 433 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 434 zfsvfs->z_shares_dir = sharezp->z_id; 435 436 zfs_acl_ids_free(&acl_ids); 437 ZTOV(sharezp)->v_data = NULL; 438 ZTOV(sharezp)->v_count = 0; 439 ZTOV(sharezp)->v_holdcnt = 0; 440 zp->z_vnode = NULL; 441 sharezp->z_vnode = NULL; 442 dmu_buf_rele(sharezp->z_dbuf, NULL); 443 sharezp->z_dbuf = NULL; 444 kmem_cache_free(znode_cache, sharezp); 445 446 return (error); 447} 448 449/* 450 * define a couple of values we need available 451 * for both 64 and 32 bit environments. 452 */ 453#ifndef NBITSMINOR64 454#define NBITSMINOR64 32 455#endif 456#ifndef MAXMAJ64 457#define MAXMAJ64 0xffffffffUL 458#endif 459#ifndef MAXMIN64 460#define MAXMIN64 0xffffffffUL 461#endif 462 463/* 464 * Create special expldev for ZFS private use. 465 * Can't use standard expldev since it doesn't do 466 * what we want. The standard expldev() takes a 467 * dev32_t in LP64 and expands it to a long dev_t. 468 * We need an interface that takes a dev32_t in ILP32 469 * and expands it to a long dev_t. 470 */ 471static uint64_t 472zfs_expldev(dev_t dev) 473{ 474 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 475} 476/* 477 * Special cmpldev for ZFS private use. 478 * Can't use standard cmpldev since it takes 479 * a long dev_t and compresses it to dev32_t in 480 * LP64. We need to do a compaction of a long dev_t 481 * to a dev32_t in ILP32. 482 */ 483dev_t 484zfs_cmpldev(uint64_t dev) 485{ 486 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 487} 488 489static void 490zfs_znode_dmu_init(zfsvfs_t *zfsvfs, znode_t *zp, dmu_buf_t *db) 491{ 492 znode_t *nzp; 493 494 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 495 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 496 497 mutex_enter(&zp->z_lock); 498 499 ASSERT(zp->z_dbuf == NULL); 500 zp->z_dbuf = db; 501 nzp = dmu_buf_set_user_ie(db, zp, &zp->z_phys, znode_evict_error); 502 503 /* 504 * there should be no 505 * concurrent zgets on this object. 506 */ 507 if (nzp != NULL) 508 panic("existing znode %p for dbuf %p", (void *)nzp, (void *)db); 509 510 /* 511 * Slap on VROOT if we are the root znode 512 */ 513 if (zp->z_id == zfsvfs->z_root) 514 ZTOV(zp)->v_flag |= VROOT; 515 516 mutex_exit(&zp->z_lock); 517 vn_exists(ZTOV(zp)); 518} 519 520void 521zfs_znode_dmu_fini(znode_t *zp) 522{ 523 dmu_buf_t *db = zp->z_dbuf; 524 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 525 zp->z_unlinked || 526 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 527 ASSERT(zp->z_dbuf != NULL); 528 zp->z_dbuf = NULL; 529 VERIFY(zp == dmu_buf_update_user(db, zp, NULL, NULL, NULL)); 530 dmu_buf_rele(db, NULL); 531} 532 533/* 534 * Construct a new znode/vnode and intialize. 535 * 536 * This does not do a call to dmu_set_user() that is 537 * up to the caller to do, in case you don't want to 538 * return the znode 539 */ 540static znode_t * 541zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz) 542{ 543 znode_t *zp; 544 vnode_t *vp; 545 546 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 547 zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0); 548 549 ASSERT(zp->z_dirlocks == NULL); 550 ASSERT(zp->z_dbuf == NULL); 551 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 552 553 /* 554 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 555 * the zfs_znode_move() callback. 556 */ 557 zp->z_phys = NULL; 558 zp->z_unlinked = 0; 559 zp->z_atime_dirty = 0; 560 zp->z_mapcnt = 0; 561 zp->z_last_itx = 0; 562 zp->z_id = db->db_object; 563 zp->z_blksz = blksz; 564 zp->z_seq = 0x7A4653; 565 zp->z_sync_cnt = 0; 566 567 vp = ZTOV(zp); 568#ifdef TODO 569 vn_reinit(vp); 570#endif 571 572 zfs_znode_dmu_init(zfsvfs, zp, db); 573 574 zp->z_gen = zp->z_phys->zp_gen; 575 576#if 0 577 if (vp == NULL) 578 return (zp); 579#endif 580 581 vp->v_type = IFTOVT((mode_t)zp->z_phys->zp_mode); 582 switch (vp->v_type) { 583 case VDIR: 584 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 585 break; 586 case VFIFO: 587 vp->v_op = &zfs_fifoops; 588 break; 589 case VREG: 590 if (zp->z_phys->zp_parent == zfsvfs->z_shares_dir) { 591 vp->v_op = &zfs_shareops; 592 } 593 break; 594 } 595 if (vp->v_type != VFIFO) 596 VN_LOCK_ASHARE(vp); 597 598 mutex_enter(&zfsvfs->z_znodes_lock); 599 list_insert_tail(&zfsvfs->z_all_znodes, zp); 600 membar_producer(); 601 /* 602 * Everything else must be valid before assigning z_zfsvfs makes the 603 * znode eligible for zfs_znode_move(). 604 */ 605 zp->z_zfsvfs = zfsvfs; 606 mutex_exit(&zfsvfs->z_znodes_lock); 607 608 VFS_HOLD(zfsvfs->z_vfs); 609 return (zp); 610} 611 612/* 613 * Create a new DMU object to hold a zfs znode. 614 * 615 * IN: dzp - parent directory for new znode 616 * vap - file attributes for new znode 617 * tx - dmu transaction id for zap operations 618 * cr - credentials of caller 619 * flag - flags: 620 * IS_ROOT_NODE - new object will be root 621 * IS_XATTR - new object is an attribute 622 * bonuslen - length of bonus buffer 623 * setaclp - File/Dir initial ACL 624 * fuidp - Tracks fuid allocation. 625 * 626 * OUT: zpp - allocated znode 627 * 628 */ 629void 630zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 631 uint_t flag, znode_t **zpp, int bonuslen, zfs_acl_ids_t *acl_ids) 632{ 633 dmu_buf_t *db; 634 znode_phys_t *pzp; 635 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 636 timestruc_t now; 637 uint64_t gen, obj; 638 int err; 639 640 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 641 642 if (zfsvfs->z_replay) { 643 obj = vap->va_nodeid; 644 now = vap->va_ctime; /* see zfs_replay_create() */ 645 gen = vap->va_nblocks; /* ditto */ 646 } else { 647 obj = 0; 648 gethrestime(&now); 649 gen = dmu_tx_get_txg(tx); 650 } 651 652 /* 653 * Create a new DMU object. 654 */ 655 /* 656 * There's currently no mechanism for pre-reading the blocks that will 657 * be to needed allocate a new object, so we accept the small chance 658 * that there will be an i/o error and we will fail one of the 659 * assertions below. 660 */ 661 if (vap->va_type == VDIR) { 662 if (zfsvfs->z_replay) { 663 err = zap_create_claim_norm(zfsvfs->z_os, obj, 664 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 665 DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx); 666 ASSERT3U(err, ==, 0); 667 } else { 668 obj = zap_create_norm(zfsvfs->z_os, 669 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 670 DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx); 671 } 672 } else { 673 if (zfsvfs->z_replay) { 674 err = dmu_object_claim(zfsvfs->z_os, obj, 675 DMU_OT_PLAIN_FILE_CONTENTS, 0, 676 DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx); 677 ASSERT3U(err, ==, 0); 678 } else { 679 obj = dmu_object_alloc(zfsvfs->z_os, 680 DMU_OT_PLAIN_FILE_CONTENTS, 0, 681 DMU_OT_ZNODE, sizeof (znode_phys_t) + bonuslen, tx); 682 } 683 } 684 685 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 686 VERIFY(0 == dmu_bonus_hold(zfsvfs->z_os, obj, NULL, &db)); 687 dmu_buf_will_dirty(db, tx); 688 689 /* 690 * Initialize the znode physical data to zero. 691 */ 692 ASSERT(db->db_size >= sizeof (znode_phys_t)); 693 bzero(db->db_data, db->db_size); 694 pzp = db->db_data; 695 696 /* 697 * If this is the root, fix up the half-initialized parent pointer 698 * to reference the just-allocated physical data area. 699 */ 700 if (flag & IS_ROOT_NODE) { 701 dzp->z_dbuf = db; 702 dzp->z_phys = pzp; 703 dzp->z_id = obj; 704 } 705 706 /* 707 * If parent is an xattr, so am I. 708 */ 709 if (dzp->z_phys->zp_flags & ZFS_XATTR) 710 flag |= IS_XATTR; 711 712 if (vap->va_type == VBLK || vap->va_type == VCHR) { 713 pzp->zp_rdev = zfs_expldev(vap->va_rdev); 714 } 715 716 if (zfsvfs->z_use_fuids) 717 pzp->zp_flags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 718 719 if (vap->va_type == VDIR) { 720 pzp->zp_size = 2; /* contents ("." and "..") */ 721 pzp->zp_links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 722 } 723 724 pzp->zp_parent = dzp->z_id; 725 if (flag & IS_XATTR) 726 pzp->zp_flags |= ZFS_XATTR; 727 728 pzp->zp_gen = gen; 729 730 ZFS_TIME_ENCODE(&now, pzp->zp_crtime); 731 ZFS_TIME_ENCODE(&now, pzp->zp_ctime); 732 733 if (vap->va_mask & AT_ATIME) { 734 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 735 } else { 736 ZFS_TIME_ENCODE(&now, pzp->zp_atime); 737 } 738 739 if (vap->va_mask & AT_MTIME) { 740 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 741 } else { 742 ZFS_TIME_ENCODE(&now, pzp->zp_mtime); 743 } 744 745 pzp->zp_mode = MAKEIMODE(vap->va_type, vap->va_mode); 746 if (!(flag & IS_ROOT_NODE)) { 747 *zpp = zfs_znode_alloc(zfsvfs, db, 0); 748 } else { 749 /* 750 * If we are creating the root node, the "parent" we 751 * passed in is the znode for the root. 752 */ 753 *zpp = dzp; 754 } 755 pzp->zp_uid = acl_ids->z_fuid; 756 pzp->zp_gid = acl_ids->z_fgid; 757 pzp->zp_mode = acl_ids->z_mode; 758 VERIFY(0 == zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx)); 759 if (vap->va_mask & AT_XVATTR) 760 zfs_xvattr_set(*zpp, (xvattr_t *)vap); 761 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 762 if (!(flag & IS_ROOT_NODE)) { 763 vnode_t *vp; 764 765 vp = ZTOV(*zpp); 766 vp->v_vflag |= VV_FORCEINSMQ; 767 err = insmntque(vp, zfsvfs->z_vfs); 768 vp->v_vflag &= ~VV_FORCEINSMQ; 769 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 770 } 771} 772 773void 774zfs_xvattr_set(znode_t *zp, xvattr_t *xvap) 775{ 776 xoptattr_t *xoap; 777 778 xoap = xva_getxoptattr(xvap); 779 ASSERT(xoap); 780 781 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 782 ZFS_TIME_ENCODE(&xoap->xoa_createtime, zp->z_phys->zp_crtime); 783 XVA_SET_RTN(xvap, XAT_CREATETIME); 784 } 785 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 786 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly); 787 XVA_SET_RTN(xvap, XAT_READONLY); 788 } 789 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 790 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden); 791 XVA_SET_RTN(xvap, XAT_HIDDEN); 792 } 793 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 794 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system); 795 XVA_SET_RTN(xvap, XAT_SYSTEM); 796 } 797 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 798 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive); 799 XVA_SET_RTN(xvap, XAT_ARCHIVE); 800 } 801 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 802 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable); 803 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 804 } 805 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 806 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink); 807 XVA_SET_RTN(xvap, XAT_NOUNLINK); 808 } 809 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 810 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly); 811 XVA_SET_RTN(xvap, XAT_APPENDONLY); 812 } 813 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 814 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump); 815 XVA_SET_RTN(xvap, XAT_NODUMP); 816 } 817 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 818 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque); 819 XVA_SET_RTN(xvap, XAT_OPAQUE); 820 } 821 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 822 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 823 xoap->xoa_av_quarantined); 824 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 825 } 826 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 827 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified); 828 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 829 } 830 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 831 (void) memcpy(zp->z_phys + 1, xoap->xoa_av_scanstamp, 832 sizeof (xoap->xoa_av_scanstamp)); 833 zp->z_phys->zp_flags |= ZFS_BONUS_SCANSTAMP; 834 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 835 } 836} 837 838int 839zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 840{ 841 dmu_object_info_t doi; 842 dmu_buf_t *db; 843 znode_t *zp; 844 vnode_t *vp; 845 int err, first = 1; 846 847 *zpp = NULL; 848again: 849 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 850 851 err = dmu_bonus_hold(zfsvfs->z_os, obj_num, NULL, &db); 852 if (err) { 853 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 854 return (err); 855 } 856 857 dmu_object_info_from_db(db, &doi); 858 if (doi.doi_bonus_type != DMU_OT_ZNODE || 859 doi.doi_bonus_size < sizeof (znode_phys_t)) { 860 dmu_buf_rele(db, NULL); 861 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 862 return (EINVAL); 863 } 864 865 zp = dmu_buf_get_user(db); 866 if (zp != NULL) { 867 mutex_enter(&zp->z_lock); 868 869 /* 870 * Since we do immediate eviction of the z_dbuf, we 871 * should never find a dbuf with a znode that doesn't 872 * know about the dbuf. 873 */ 874 ASSERT3P(zp->z_dbuf, ==, db); 875 ASSERT3U(zp->z_id, ==, obj_num); 876 if (zp->z_unlinked) { 877 err = ENOENT; 878 } else { 879 int dying = 0; 880 881 vp = ZTOV(zp); 882 if (vp == NULL) 883 dying = 1; 884 else { 885 VN_HOLD(vp); 886 if ((vp->v_iflag & VI_DOOMED) != 0) { 887 dying = 1; 888 /* 889 * Don't VN_RELE() vnode here, because 890 * it can call vn_lock() which creates 891 * LOR between vnode lock and znode 892 * lock. We will VN_RELE() the vnode 893 * after droping znode lock. 894 */ 895 } 896 } 897 if (dying) { 898 if (first) { 899 ZFS_LOG(1, "dying znode detected (zp=%p)", zp); 900 first = 0; 901 } 902 /* 903 * znode is dying so we can't reuse it, we must 904 * wait until destruction is completed. 905 */ 906 dmu_buf_rele(db, NULL); 907 mutex_exit(&zp->z_lock); 908 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 909 if (vp != NULL) 910 VN_RELE(vp); 911 tsleep(zp, 0, "zcollide", 1); 912 goto again; 913 } 914 *zpp = zp; 915 err = 0; 916 } 917 dmu_buf_rele(db, NULL); 918 mutex_exit(&zp->z_lock); 919 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 920 return (err); 921 } 922 923 /* 924 * Not found create new znode/vnode 925 * but only if file exists. 926 * 927 * There is a small window where zfs_vget() could 928 * find this object while a file create is still in 929 * progress. Since a gen number can never be zero 930 * we will check that to determine if its an allocated 931 * file. 932 */ 933 934 if (((znode_phys_t *)db->db_data)->zp_gen != 0) { 935 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size); 936 *zpp = zp; 937 vp = ZTOV(zp); 938 vp->v_vflag |= VV_FORCEINSMQ; 939 err = insmntque(vp, zfsvfs->z_vfs); 940 vp->v_vflag &= ~VV_FORCEINSMQ; 941 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 942 VOP_UNLOCK(vp, 0); 943 err = 0; 944 } else { 945 dmu_buf_rele(db, NULL); 946 err = ENOENT; 947 } 948 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 949 return (err); 950} 951 952int 953zfs_rezget(znode_t *zp) 954{ 955 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 956 dmu_object_info_t doi; 957 dmu_buf_t *db; 958 uint64_t obj_num = zp->z_id; 959 int err; 960 961 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 962 963 err = dmu_bonus_hold(zfsvfs->z_os, obj_num, NULL, &db); 964 if (err) { 965 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 966 return (err); 967 } 968 969 dmu_object_info_from_db(db, &doi); 970 if (doi.doi_bonus_type != DMU_OT_ZNODE || 971 doi.doi_bonus_size < sizeof (znode_phys_t)) { 972 dmu_buf_rele(db, NULL); 973 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 974 return (EINVAL); 975 } 976 977 if (((znode_phys_t *)db->db_data)->zp_gen != zp->z_gen) { 978 dmu_buf_rele(db, NULL); 979 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 980 return (EIO); 981 } 982 983 zfs_znode_dmu_init(zfsvfs, zp, db); 984 zp->z_unlinked = (zp->z_phys->zp_links == 0); 985 zp->z_blksz = doi.doi_data_block_size; 986 987 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 988 989 return (0); 990} 991 992void 993zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 994{ 995 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 996 objset_t *os = zfsvfs->z_os; 997 uint64_t obj = zp->z_id; 998 uint64_t acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj; 999 1000 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1001 if (acl_obj) 1002 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1003 VERIFY(0 == dmu_object_free(os, obj, tx)); 1004 zfs_znode_dmu_fini(zp); 1005 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1006 zfs_znode_free(zp); 1007} 1008 1009void 1010zfs_zinactive(znode_t *zp) 1011{ 1012 vnode_t *vp = ZTOV(zp); 1013 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1014 uint64_t z_id = zp->z_id; 1015 int vfslocked; 1016 1017 ASSERT(zp->z_dbuf && zp->z_phys); 1018 1019 /* 1020 * Don't allow a zfs_zget() while were trying to release this znode 1021 */ 1022 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1023 1024 mutex_enter(&zp->z_lock); 1025 VI_LOCK(vp); 1026 if (vp->v_count > 0) { 1027 /* 1028 * If the hold count is greater than zero, somebody has 1029 * obtained a new reference on this znode while we were 1030 * processing it here, so we are done. 1031 */ 1032 VI_UNLOCK(vp); 1033 mutex_exit(&zp->z_lock); 1034 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1035 return; 1036 } 1037 VI_UNLOCK(vp); 1038 1039 /* 1040 * If this was the last reference to a file with no links, 1041 * remove the file from the file system. 1042 */ 1043 if (zp->z_unlinked) { 1044 mutex_exit(&zp->z_lock); 1045 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1046 ASSERT(vp->v_count == 0); 1047 vrecycle(vp, curthread); 1048 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs); 1049 zfs_rmnode(zp); 1050 VFS_UNLOCK_GIANT(vfslocked); 1051 return; 1052 } 1053 mutex_exit(&zp->z_lock); 1054 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1055} 1056 1057void 1058zfs_znode_free(znode_t *zp) 1059{ 1060 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1061 1062 ASSERT(ZTOV(zp) == NULL); 1063 mutex_enter(&zfsvfs->z_znodes_lock); 1064 POINTER_INVALIDATE(&zp->z_zfsvfs); 1065 list_remove(&zfsvfs->z_all_znodes, zp); 1066 mutex_exit(&zfsvfs->z_znodes_lock); 1067 1068 kmem_cache_free(znode_cache, zp); 1069 1070 VFS_RELE(zfsvfs->z_vfs); 1071} 1072 1073void 1074zfs_time_stamper_locked(znode_t *zp, uint_t flag, dmu_tx_t *tx) 1075{ 1076 timestruc_t now; 1077 1078 ASSERT(MUTEX_HELD(&zp->z_lock)); 1079 1080 gethrestime(&now); 1081 1082 if (tx) { 1083 dmu_buf_will_dirty(zp->z_dbuf, tx); 1084 zp->z_atime_dirty = 0; 1085 zp->z_seq++; 1086 } else { 1087 zp->z_atime_dirty = 1; 1088 } 1089 1090 if (flag & AT_ATIME) 1091 ZFS_TIME_ENCODE(&now, zp->z_phys->zp_atime); 1092 1093 if (flag & AT_MTIME) { 1094 ZFS_TIME_ENCODE(&now, zp->z_phys->zp_mtime); 1095 if (zp->z_zfsvfs->z_use_fuids) 1096 zp->z_phys->zp_flags |= (ZFS_ARCHIVE | ZFS_AV_MODIFIED); 1097 } 1098 1099 if (flag & AT_CTIME) { 1100 ZFS_TIME_ENCODE(&now, zp->z_phys->zp_ctime); 1101 if (zp->z_zfsvfs->z_use_fuids) 1102 zp->z_phys->zp_flags |= ZFS_ARCHIVE; 1103 } 1104} 1105 1106/* 1107 * Update the requested znode timestamps with the current time. 1108 * If we are in a transaction, then go ahead and mark the znode 1109 * dirty in the transaction so the timestamps will go to disk. 1110 * Otherwise, we will get pushed next time the znode is updated 1111 * in a transaction, or when this znode eventually goes inactive. 1112 * 1113 * Why is this OK? 1114 * 1 - Only the ACCESS time is ever updated outside of a transaction. 1115 * 2 - Multiple consecutive updates will be collapsed into a single 1116 * znode update by the transaction grouping semantics of the DMU. 1117 */ 1118void 1119zfs_time_stamper(znode_t *zp, uint_t flag, dmu_tx_t *tx) 1120{ 1121 mutex_enter(&zp->z_lock); 1122 zfs_time_stamper_locked(zp, flag, tx); 1123 mutex_exit(&zp->z_lock); 1124} 1125 1126/* 1127 * Grow the block size for a file. 1128 * 1129 * IN: zp - znode of file to free data in. 1130 * size - requested block size 1131 * tx - open transaction. 1132 * 1133 * NOTE: this function assumes that the znode is write locked. 1134 */ 1135void 1136zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1137{ 1138 int error; 1139 u_longlong_t dummy; 1140 1141 if (size <= zp->z_blksz) 1142 return; 1143 /* 1144 * If the file size is already greater than the current blocksize, 1145 * we will not grow. If there is more than one block in a file, 1146 * the blocksize cannot change. 1147 */ 1148 if (zp->z_blksz && zp->z_phys->zp_size > zp->z_blksz) 1149 return; 1150 1151 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1152 size, 0, tx); 1153 if (error == ENOTSUP) 1154 return; 1155 ASSERT3U(error, ==, 0); 1156 1157 /* What blocksize did we actually get? */ 1158 dmu_object_size_from_db(zp->z_dbuf, &zp->z_blksz, &dummy); 1159} 1160 1161/* 1162 * Increase the file length 1163 * 1164 * IN: zp - znode of file to free data in. 1165 * end - new end-of-file 1166 * 1167 * RETURN: 0 if success 1168 * error code if failure 1169 */ 1170static int 1171zfs_extend(znode_t *zp, uint64_t end) 1172{ 1173 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1174 dmu_tx_t *tx; 1175 rl_t *rl; 1176 uint64_t newblksz; 1177 int error; 1178 1179 /* 1180 * We will change zp_size, lock the whole file. 1181 */ 1182 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1183 1184 /* 1185 * Nothing to do if file already at desired length. 1186 */ 1187 if (end <= zp->z_phys->zp_size) { 1188 zfs_range_unlock(rl); 1189 return (0); 1190 } 1191top: 1192 tx = dmu_tx_create(zfsvfs->z_os); 1193 dmu_tx_hold_bonus(tx, zp->z_id); 1194 if (end > zp->z_blksz && 1195 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1196 /* 1197 * We are growing the file past the current block size. 1198 */ 1199 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1200 ASSERT(!ISP2(zp->z_blksz)); 1201 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1202 } else { 1203 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1204 } 1205 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1206 } else { 1207 newblksz = 0; 1208 } 1209 1210 error = dmu_tx_assign(tx, TXG_NOWAIT); 1211 if (error) { 1212 if (error == ERESTART) { 1213 dmu_tx_wait(tx); 1214 dmu_tx_abort(tx); 1215 goto top; 1216 } 1217 dmu_tx_abort(tx); 1218 zfs_range_unlock(rl); 1219 return (error); 1220 } 1221 dmu_buf_will_dirty(zp->z_dbuf, tx); 1222 1223 if (newblksz) 1224 zfs_grow_blocksize(zp, newblksz, tx); 1225 1226 zp->z_phys->zp_size = end; 1227 1228 zfs_range_unlock(rl); 1229 1230 dmu_tx_commit(tx); 1231 1232 vnode_pager_setsize(ZTOV(zp), end); 1233 1234 return (0); 1235} 1236 1237/* 1238 * Free space in a file. 1239 * 1240 * IN: zp - znode of file to free data in. 1241 * off - start of section to free. 1242 * len - length of section to free. 1243 * 1244 * RETURN: 0 if success 1245 * error code if failure 1246 */ 1247static int 1248zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1249{ 1250 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1251 rl_t *rl; 1252 int error; 1253 1254 /* 1255 * Lock the range being freed. 1256 */ 1257 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1258 1259 /* 1260 * Nothing to do if file already at desired length. 1261 */ 1262 if (off >= zp->z_phys->zp_size) { 1263 zfs_range_unlock(rl); 1264 return (0); 1265 } 1266 1267 if (off + len > zp->z_phys->zp_size) 1268 len = zp->z_phys->zp_size - off; 1269 1270 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1271 1272 if (error == 0) { 1273 /* 1274 * In FreeBSD we cannot free block in the middle of a file, 1275 * but only at the end of a file. 1276 */ 1277 vnode_pager_setsize(ZTOV(zp), off); 1278 } 1279 1280 zfs_range_unlock(rl); 1281 1282 return (error); 1283} 1284 1285/* 1286 * Truncate a file 1287 * 1288 * IN: zp - znode of file to free data in. 1289 * end - new end-of-file. 1290 * 1291 * RETURN: 0 if success 1292 * error code if failure 1293 */ 1294static int 1295zfs_trunc(znode_t *zp, uint64_t end) 1296{ 1297 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1298 vnode_t *vp = ZTOV(zp); 1299 dmu_tx_t *tx; 1300 rl_t *rl; 1301 int error; 1302 1303 /* 1304 * We will change zp_size, lock the whole file. 1305 */ 1306 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1307 1308 /* 1309 * Nothing to do if file already at desired length. 1310 */ 1311 if (end >= zp->z_phys->zp_size) { 1312 zfs_range_unlock(rl); 1313 return (0); 1314 } 1315 1316 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1317 if (error) { 1318 zfs_range_unlock(rl); 1319 return (error); 1320 } 1321top: 1322 tx = dmu_tx_create(zfsvfs->z_os); 1323 dmu_tx_hold_bonus(tx, zp->z_id); 1324 error = dmu_tx_assign(tx, TXG_NOWAIT); 1325 if (error) { 1326 if (error == ERESTART) { 1327 dmu_tx_wait(tx); 1328 dmu_tx_abort(tx); 1329 goto top; 1330 } 1331 dmu_tx_abort(tx); 1332 zfs_range_unlock(rl); 1333 return (error); 1334 } 1335 dmu_buf_will_dirty(zp->z_dbuf, tx); 1336 1337 zp->z_phys->zp_size = end; 1338 1339 dmu_tx_commit(tx); 1340 1341 /* 1342 * Clear any mapped pages in the truncated region. This has to 1343 * happen outside of the transaction to avoid the possibility of 1344 * a deadlock with someone trying to push a page that we are 1345 * about to invalidate. 1346 */ 1347 vnode_pager_setsize(vp, end); 1348 1349 zfs_range_unlock(rl); 1350 1351 return (0); 1352} 1353 1354/* 1355 * Free space in a file 1356 * 1357 * IN: zp - znode of file to free data in. 1358 * off - start of range 1359 * len - end of range (0 => EOF) 1360 * flag - current file open mode flags. 1361 * log - TRUE if this action should be logged 1362 * 1363 * RETURN: 0 if success 1364 * error code if failure 1365 */ 1366int 1367zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1368{ 1369 vnode_t *vp = ZTOV(zp); 1370 dmu_tx_t *tx; 1371 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1372 zilog_t *zilog = zfsvfs->z_log; 1373 int error; 1374 1375 if (off > zp->z_phys->zp_size) { 1376 error = zfs_extend(zp, off+len); 1377 if (error == 0 && log) 1378 goto log; 1379 else 1380 return (error); 1381 } 1382 1383 if (len == 0) { 1384 error = zfs_trunc(zp, off); 1385 } else { 1386 if ((error = zfs_free_range(zp, off, len)) == 0 && 1387 off + len > zp->z_phys->zp_size) 1388 error = zfs_extend(zp, off+len); 1389 } 1390 if (error || !log) 1391 return (error); 1392log: 1393 tx = dmu_tx_create(zfsvfs->z_os); 1394 dmu_tx_hold_bonus(tx, zp->z_id); 1395 error = dmu_tx_assign(tx, TXG_NOWAIT); 1396 if (error) { 1397 if (error == ERESTART) { 1398 dmu_tx_wait(tx); 1399 dmu_tx_abort(tx); 1400 goto log; 1401 } 1402 dmu_tx_abort(tx); 1403 return (error); 1404 } 1405 1406 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 1407 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1408 1409 dmu_tx_commit(tx); 1410 return (0); 1411} 1412 1413void 1414zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1415{ 1416 zfsvfs_t zfsvfs; 1417 uint64_t moid, obj, version; 1418 uint64_t sense = ZFS_CASE_SENSITIVE; 1419 uint64_t norm = 0; 1420 nvpair_t *elem; 1421 int error; 1422 int i; 1423 znode_t *rootzp = NULL; 1424 vnode_t vnode; 1425 vattr_t vattr; 1426 znode_t *zp; 1427 zfs_acl_ids_t acl_ids; 1428 1429 /* 1430 * First attempt to create master node. 1431 */ 1432 /* 1433 * In an empty objset, there are no blocks to read and thus 1434 * there can be no i/o errors (which we assert below). 1435 */ 1436 moid = MASTER_NODE_OBJ; 1437 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1438 DMU_OT_NONE, 0, tx); 1439 ASSERT(error == 0); 1440 1441 /* 1442 * Set starting attributes. 1443 */ 1444 if (spa_version(dmu_objset_spa(os)) >= SPA_VERSION_USERSPACE) 1445 version = ZPL_VERSION; 1446 else if (spa_version(dmu_objset_spa(os)) >= SPA_VERSION_FUID) 1447 version = ZPL_VERSION_USERSPACE - 1; 1448 else 1449 version = ZPL_VERSION_FUID - 1; 1450 elem = NULL; 1451 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1452 /* For the moment we expect all zpl props to be uint64_ts */ 1453 uint64_t val; 1454 char *name; 1455 1456 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1457 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1458 name = nvpair_name(elem); 1459 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1460 if (val < version) 1461 version = val; 1462 } else { 1463 error = zap_update(os, moid, name, 8, 1, &val, tx); 1464 } 1465 ASSERT(error == 0); 1466 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1467 norm = val; 1468 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1469 sense = val; 1470 } 1471 ASSERT(version != 0); 1472 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1473 1474 /* 1475 * Create a delete queue. 1476 */ 1477 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1478 1479 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1480 ASSERT(error == 0); 1481 1482 /* 1483 * Create root znode. Create minimal znode/vnode/zfsvfs 1484 * to allow zfs_mknode to work. 1485 */ 1486 VATTR_NULL(&vattr); 1487 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1488 vattr.va_type = VDIR; 1489 vattr.va_mode = S_IFDIR|0755; 1490 vattr.va_uid = crgetuid(cr); 1491 vattr.va_gid = crgetgid(cr); 1492 1493 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1494 zfs_znode_cache_constructor(rootzp, NULL, 0); 1495 rootzp->z_unlinked = 0; 1496 rootzp->z_atime_dirty = 0; 1497 1498 vnode.v_type = VDIR; 1499 vnode.v_data = rootzp; 1500 rootzp->z_vnode = &vnode; 1501 1502 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1503 1504 zfsvfs.z_os = os; 1505 zfsvfs.z_parent = &zfsvfs; 1506 zfsvfs.z_version = version; 1507 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1508 zfsvfs.z_norm = norm; 1509 /* 1510 * Fold case on file systems that are always or sometimes case 1511 * insensitive. 1512 */ 1513 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1514 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1515 1516 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1517 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1518 offsetof(znode_t, z_link_node)); 1519 1520 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1521 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1522 1523 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1524 rootzp->z_zfsvfs = &zfsvfs; 1525 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1526 cr, NULL, &acl_ids)); 1527 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, 0, &acl_ids); 1528 ASSERT3P(zp, ==, rootzp); 1529 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1530 ASSERT(error == 0); 1531 zfs_acl_ids_free(&acl_ids); 1532 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1533 1534 dmu_buf_rele(rootzp->z_dbuf, NULL); 1535 rootzp->z_dbuf = NULL; 1536 rootzp->z_vnode = NULL; 1537 kmem_cache_free(znode_cache, rootzp); 1538 1539 /* 1540 * Create shares directory 1541 */ 1542 1543 error = zfs_create_share_dir(&zfsvfs, tx); 1544 1545 ASSERT(error == 0); 1546 1547 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1548 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1549} 1550 1551#endif /* _KERNEL */ 1552/* 1553 * Given an object number, return its parent object number and whether 1554 * or not the object is an extended attribute directory. 1555 */ 1556static int 1557zfs_obj_to_pobj(objset_t *osp, uint64_t obj, uint64_t *pobjp, int *is_xattrdir) 1558{ 1559 dmu_buf_t *db; 1560 dmu_object_info_t doi; 1561 znode_phys_t *zp; 1562 int error; 1563 1564 if ((error = dmu_bonus_hold(osp, obj, FTAG, &db)) != 0) 1565 return (error); 1566 1567 dmu_object_info_from_db(db, &doi); 1568 if (doi.doi_bonus_type != DMU_OT_ZNODE || 1569 doi.doi_bonus_size < sizeof (znode_phys_t)) { 1570 dmu_buf_rele(db, FTAG); 1571 return (EINVAL); 1572 } 1573 1574 zp = db->db_data; 1575 *pobjp = zp->zp_parent; 1576 *is_xattrdir = ((zp->zp_flags & ZFS_XATTR) != 0) && 1577 S_ISDIR(zp->zp_mode); 1578 dmu_buf_rele(db, FTAG); 1579 1580 return (0); 1581} 1582 1583int 1584zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 1585{ 1586 char *path = buf + len - 1; 1587 int error; 1588 1589 *path = '\0'; 1590 1591 for (;;) { 1592 uint64_t pobj; 1593 char component[MAXNAMELEN + 2]; 1594 size_t complen; 1595 int is_xattrdir; 1596 1597 if ((error = zfs_obj_to_pobj(osp, obj, &pobj, 1598 &is_xattrdir)) != 0) 1599 break; 1600 1601 if (pobj == obj) { 1602 if (path[0] != '/') 1603 *--path = '/'; 1604 break; 1605 } 1606 1607 component[0] = '/'; 1608 if (is_xattrdir) { 1609 (void) sprintf(component + 1, "<xattrdir>"); 1610 } else { 1611 error = zap_value_search(osp, pobj, obj, 1612 ZFS_DIRENT_OBJ(-1ULL), component + 1); 1613 if (error != 0) 1614 break; 1615 } 1616 1617 complen = strlen(component); 1618 path -= complen; 1619 ASSERT(path >= buf); 1620 bcopy(component, path, complen); 1621 obj = pobj; 1622 } 1623 1624 if (error == 0) 1625 (void) memmove(buf, path, buf + len - path); 1626 return (error); 1627} 1628