zfs_znode.c revision 240955
1295016Sjkim/* 2110010Smarkm * CDDL HEADER START 3110010Smarkm * 4160819Ssimon * The contents of this file are subject to the terms of the 5110010Smarkm * Common Development and Distribution License (the "License"). 6110010Smarkm * You may not use this file except in compliance with the License. 7110010Smarkm * 8110010Smarkm * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9110010Smarkm * or http://www.opensolaris.org/os/licensing. 10110010Smarkm * See the License for the specific language governing permissions 11110010Smarkm * and limitations under the License. 12110010Smarkm * 13110010Smarkm * When distributing Covered Code, include this CDDL HEADER in each 14110010Smarkm * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15110010Smarkm * If applicable, add the following below this CDDL HEADER, with the 16110010Smarkm * fields enclosed by brackets "[]" replaced with your own identifying 17110010Smarkm * information: Portions Copyright [yyyy] [name of copyright owner] 18110010Smarkm * 19110010Smarkm * CDDL HEADER END 20215698Ssimon */ 21215698Ssimon/* 22215698Ssimon * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23215698Ssimon * Copyright (c) 2012 by Delphix. All rights reserved. 24215698Ssimon */ 25110010Smarkm 26110010Smarkm/* Portions Copyright 2007 Jeremy Teo */ 27110010Smarkm/* Portions Copyright 2011 Martin Matuska <mm@FreeBSD.org> */ 28110010Smarkm 29110010Smarkm#ifdef _KERNEL 30110010Smarkm#include <sys/types.h> 31110010Smarkm#include <sys/param.h> 32110010Smarkm#include <sys/time.h> 33110010Smarkm#include <sys/systm.h> 34110010Smarkm#include <sys/sysmacros.h> 35110010Smarkm#include <sys/resource.h> 36110010Smarkm#include <sys/mntent.h> 37110010Smarkm#include <sys/u8_textprep.h> 38110010Smarkm#include <sys/dsl_dataset.h> 39110010Smarkm#include <sys/vfs.h> 40110010Smarkm#include <sys/vnode.h> 41276864Sjkim#include <sys/file.h> 42276864Sjkim#include <sys/kmem.h> 43110010Smarkm#include <sys/errno.h> 44110010Smarkm#include <sys/unistd.h> 45215698Ssimon#include <sys/atomic.h> 46215698Ssimon#include <sys/zfs_dir.h> 47215698Ssimon#include <sys/zfs_acl.h> 48215698Ssimon#include <sys/zfs_ioctl.h> 49160819Ssimon#include <sys/zfs_rlock.h> 50215698Ssimon#include <sys/zfs_fuid.h> 51160819Ssimon#include <sys/dnode.h> 52160819Ssimon#include <sys/fs/zfs.h> 53276864Sjkim#include <sys/kidmap.h> 54276864Sjkim#endif /* _KERNEL */ 55276864Sjkim 56110010Smarkm#include <sys/dmu.h> 57276864Sjkim#include <sys/refcount.h> 58276864Sjkim#include <sys/stat.h> 59276864Sjkim#include <sys/zap.h> 60276864Sjkim#include <sys/zfs_znode.h> 61276864Sjkim#include <sys/sa.h> 62276864Sjkim#include <sys/zfs_sa.h> 63215698Ssimon#include <sys/zfs_stat.h> 64276864Sjkim#include <sys/refcount.h> 65276864Sjkim 66276864Sjkim#include "zfs_prop.h" 67276864Sjkim#include "zfs_comutil.h" 68276864Sjkim 69215698Ssimon/* Used by fstat(1). */ 70276864SjkimSYSCTL_INT(_debug_sizeof, OID_AUTO, znode, CTLFLAG_RD, 0, sizeof(znode_t), 71110010Smarkm "sizeof(znode_t)"); 72110010Smarkm 73110010Smarkm/* 74110010Smarkm * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 75110010Smarkm * turned on when DEBUG is also defined. 76110010Smarkm */ 77110010Smarkm#ifdef DEBUG 78110010Smarkm#define ZNODE_STATS 79110010Smarkm#endif /* DEBUG */ 80110010Smarkm 81110010Smarkm#ifdef ZNODE_STATS 82110010Smarkm#define ZNODE_STAT_ADD(stat) ((stat)++) 83110010Smarkm#else 84110010Smarkm#define ZNODE_STAT_ADD(stat) /* nothing */ 85110010Smarkm#endif /* ZNODE_STATS */ 86110010Smarkm 87110010Smarkm/* 88110010Smarkm * Functions needed for userland (ie: libzpool) are not put under 89110010Smarkm * #ifdef_KERNEL; the rest of the functions have dependencies 90110010Smarkm * (such as VFS logic) that will not compile easily in userland. 91110010Smarkm */ 92110010Smarkm#ifdef _KERNEL 93110010Smarkm/* 94110010Smarkm * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 95110010Smarkm * be freed before it can be safely accessed. 96110010Smarkm */ 97110010Smarkmkrwlock_t zfsvfs_lock; 98110010Smarkm 99110010Smarkmstatic kmem_cache_t *znode_cache = NULL; 100110010Smarkm 101110010Smarkm/*ARGSUSED*/ 102110010Smarkmstatic void 103110010Smarkmznode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 104110010Smarkm{ 105110010Smarkm /* 106110010Smarkm * We should never drop all dbuf refs without first clearing 107110010Smarkm * the eviction callback. 108110010Smarkm */ 109110010Smarkm panic("evicting znode %p\n", user_ptr); 110110010Smarkm} 111110010Smarkm 112110010Smarkmextern struct vop_vector zfs_vnodeops; 113110010Smarkmextern struct vop_vector zfs_fifoops; 114110010Smarkmextern struct vop_vector zfs_shareops; 115110010Smarkm 116110010Smarkm/* 117110010Smarkm * XXX: We cannot use this function as a cache constructor, because 118110010Smarkm * there is one global cache for all file systems and we need 119110010Smarkm * to pass vfsp here, which is not possible, because argument 120110010Smarkm * 'cdrarg' is defined at kmem_cache_create() time. 121110010Smarkm */ 122110010Smarkm/*ARGSUSED*/ 123110010Smarkmstatic int 124110010Smarkmzfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 125110010Smarkm{ 126110010Smarkm znode_t *zp = buf; 127110010Smarkm vnode_t *vp; 128110010Smarkm vfs_t *vfsp = arg; 129110010Smarkm int error; 130110010Smarkm 131110010Smarkm POINTER_INVALIDATE(&zp->z_zfsvfs); 132110010Smarkm ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 133160819Ssimon 134110010Smarkm if (vfsp != NULL) { 135110010Smarkm error = getnewvnode("zfs", vfsp, &zfs_vnodeops, &vp); 136296317Sdelphij if (error != 0 && (kmflags & KM_NOSLEEP)) 137215698Ssimon return (-1); 138215698Ssimon ASSERT(error == 0); 139215698Ssimon vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 140215698Ssimon zp->z_vnode = vp; 141110010Smarkm vp->v_data = (caddr_t)zp; 142160819Ssimon VN_LOCK_AREC(vp); 143110010Smarkm VN_LOCK_ASHARE(vp); 144110010Smarkm } else { 145110010Smarkm zp->z_vnode = NULL; 146110010Smarkm } 147215698Ssimon 148110010Smarkm list_link_init(&zp->z_link_node); 149110010Smarkm 150110010Smarkm mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 151110010Smarkm rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 152110010Smarkm rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 153110010Smarkm mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 154110010Smarkm 155110010Smarkm mutex_init(&zp->z_range_lock, NULL, MUTEX_DEFAULT, NULL); 156110010Smarkm avl_create(&zp->z_range_avl, zfs_range_compare, 157110010Smarkm sizeof (rl_t), offsetof(rl_t, r_node)); 158110010Smarkm 159276864Sjkim zp->z_dirlocks = NULL; 160160819Ssimon zp->z_acl_cached = NULL; 161160819Ssimon zp->z_moved = 0; 162110010Smarkm return (0); 163110010Smarkm} 164110010Smarkm 165110010Smarkm/*ARGSUSED*/ 166110010Smarkmstatic void 167110010Smarkmzfs_znode_cache_destructor(void *buf, void *arg) 168110010Smarkm{ 169110010Smarkm znode_t *zp = buf; 170160819Ssimon 171110010Smarkm ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 172110010Smarkm ASSERT(ZTOV(zp) == NULL); 173110010Smarkm vn_free(ZTOV(zp)); 174110010Smarkm ASSERT(!list_link_active(&zp->z_link_node)); 175110010Smarkm mutex_destroy(&zp->z_lock); 176110010Smarkm rw_destroy(&zp->z_parent_lock); 177237657Sjkim rw_destroy(&zp->z_name_lock); 178237657Sjkim mutex_destroy(&zp->z_acl_lock); 179237657Sjkim avl_destroy(&zp->z_range_avl); 180237657Sjkim mutex_destroy(&zp->z_range_lock); 181237657Sjkim 182237657Sjkim ASSERT(zp->z_dirlocks == NULL); 183237657Sjkim ASSERT(zp->z_acl_cached == NULL); 184237657Sjkim} 185237657Sjkim 186237657Sjkim#ifdef ZNODE_STATS 187110010Smarkmstatic struct { 188110010Smarkm uint64_t zms_zfsvfs_invalid; 189110010Smarkm uint64_t zms_zfsvfs_recheck1; 190160819Ssimon uint64_t zms_zfsvfs_unmounted; 191110010Smarkm uint64_t zms_zfsvfs_recheck2; 192110010Smarkm uint64_t zms_obj_held; 193160819Ssimon uint64_t zms_vnode_locked; 194110010Smarkm uint64_t zms_not_only_dnlc; 195110010Smarkm} znode_move_stats; 196110010Smarkm#endif /* ZNODE_STATS */ 197160819Ssimon 198160819Ssimon#ifdef sun 199160819Ssimonstatic void 200160819Ssimonzfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 201{ 202 vnode_t *vp; 203 204 /* Copy fields. */ 205 nzp->z_zfsvfs = ozp->z_zfsvfs; 206 207 /* Swap vnodes. */ 208 vp = nzp->z_vnode; 209 nzp->z_vnode = ozp->z_vnode; 210 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 211 ZTOV(ozp)->v_data = ozp; 212 ZTOV(nzp)->v_data = nzp; 213 214 nzp->z_id = ozp->z_id; 215 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 216 ASSERT(avl_numnodes(&ozp->z_range_avl) == 0); 217 nzp->z_unlinked = ozp->z_unlinked; 218 nzp->z_atime_dirty = ozp->z_atime_dirty; 219 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 220 nzp->z_blksz = ozp->z_blksz; 221 nzp->z_seq = ozp->z_seq; 222 nzp->z_mapcnt = ozp->z_mapcnt; 223 nzp->z_gen = ozp->z_gen; 224 nzp->z_sync_cnt = ozp->z_sync_cnt; 225 nzp->z_is_sa = ozp->z_is_sa; 226 nzp->z_sa_hdl = ozp->z_sa_hdl; 227 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 228 nzp->z_links = ozp->z_links; 229 nzp->z_size = ozp->z_size; 230 nzp->z_pflags = ozp->z_pflags; 231 nzp->z_uid = ozp->z_uid; 232 nzp->z_gid = ozp->z_gid; 233 nzp->z_mode = ozp->z_mode; 234 235 /* 236 * Since this is just an idle znode and kmem is already dealing with 237 * memory pressure, release any cached ACL. 238 */ 239 if (ozp->z_acl_cached) { 240 zfs_acl_free(ozp->z_acl_cached); 241 ozp->z_acl_cached = NULL; 242 } 243 244 sa_set_userp(nzp->z_sa_hdl, nzp); 245 246 /* 247 * Invalidate the original znode by clearing fields that provide a 248 * pointer back to the znode. Set the low bit of the vfs pointer to 249 * ensure that zfs_znode_move() recognizes the znode as invalid in any 250 * subsequent callback. 251 */ 252 ozp->z_sa_hdl = NULL; 253 POINTER_INVALIDATE(&ozp->z_zfsvfs); 254 255 /* 256 * Mark the znode. 257 */ 258 nzp->z_moved = 1; 259 ozp->z_moved = (uint8_t)-1; 260} 261 262/*ARGSUSED*/ 263static kmem_cbrc_t 264zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 265{ 266 znode_t *ozp = buf, *nzp = newbuf; 267 zfsvfs_t *zfsvfs; 268 vnode_t *vp; 269 270 /* 271 * The znode is on the file system's list of known znodes if the vfs 272 * pointer is valid. We set the low bit of the vfs pointer when freeing 273 * the znode to invalidate it, and the memory patterns written by kmem 274 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 275 * created znode sets the vfs pointer last of all to indicate that the 276 * znode is known and in a valid state to be moved by this function. 277 */ 278 zfsvfs = ozp->z_zfsvfs; 279 if (!POINTER_IS_VALID(zfsvfs)) { 280 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 281 return (KMEM_CBRC_DONT_KNOW); 282 } 283 284 /* 285 * Close a small window in which it's possible that the filesystem could 286 * be unmounted and freed, and zfsvfs, though valid in the previous 287 * statement, could point to unrelated memory by the time we try to 288 * prevent the filesystem from being unmounted. 289 */ 290 rw_enter(&zfsvfs_lock, RW_WRITER); 291 if (zfsvfs != ozp->z_zfsvfs) { 292 rw_exit(&zfsvfs_lock); 293 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 294 return (KMEM_CBRC_DONT_KNOW); 295 } 296 297 /* 298 * If the znode is still valid, then so is the file system. We know that 299 * no valid file system can be freed while we hold zfsvfs_lock, so we 300 * can safely ensure that the filesystem is not and will not be 301 * unmounted. The next statement is equivalent to ZFS_ENTER(). 302 */ 303 rrw_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 304 if (zfsvfs->z_unmounted) { 305 ZFS_EXIT(zfsvfs); 306 rw_exit(&zfsvfs_lock); 307 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 308 return (KMEM_CBRC_DONT_KNOW); 309 } 310 rw_exit(&zfsvfs_lock); 311 312 mutex_enter(&zfsvfs->z_znodes_lock); 313 /* 314 * Recheck the vfs pointer in case the znode was removed just before 315 * acquiring the lock. 316 */ 317 if (zfsvfs != ozp->z_zfsvfs) { 318 mutex_exit(&zfsvfs->z_znodes_lock); 319 ZFS_EXIT(zfsvfs); 320 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 321 return (KMEM_CBRC_DONT_KNOW); 322 } 323 324 /* 325 * At this point we know that as long as we hold z_znodes_lock, the 326 * znode cannot be freed and fields within the znode can be safely 327 * accessed. Now, prevent a race with zfs_zget(). 328 */ 329 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 330 mutex_exit(&zfsvfs->z_znodes_lock); 331 ZFS_EXIT(zfsvfs); 332 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 333 return (KMEM_CBRC_LATER); 334 } 335 336 vp = ZTOV(ozp); 337 if (mutex_tryenter(&vp->v_lock) == 0) { 338 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 339 mutex_exit(&zfsvfs->z_znodes_lock); 340 ZFS_EXIT(zfsvfs); 341 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 342 return (KMEM_CBRC_LATER); 343 } 344 345 /* Only move znodes that are referenced _only_ by the DNLC. */ 346 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 347 mutex_exit(&vp->v_lock); 348 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 349 mutex_exit(&zfsvfs->z_znodes_lock); 350 ZFS_EXIT(zfsvfs); 351 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 352 return (KMEM_CBRC_LATER); 353 } 354 355 /* 356 * The znode is known and in a valid state to move. We're holding the 357 * locks needed to execute the critical section. 358 */ 359 zfs_znode_move_impl(ozp, nzp); 360 mutex_exit(&vp->v_lock); 361 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 362 363 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 364 mutex_exit(&zfsvfs->z_znodes_lock); 365 ZFS_EXIT(zfsvfs); 366 367 return (KMEM_CBRC_YES); 368} 369#endif /* sun */ 370 371void 372zfs_znode_init(void) 373{ 374 /* 375 * Initialize zcache 376 */ 377 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 378 ASSERT(znode_cache == NULL); 379 znode_cache = kmem_cache_create("zfs_znode_cache", 380 sizeof (znode_t), 0, /* zfs_znode_cache_constructor */ NULL, 381 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 382 kmem_cache_set_move(znode_cache, zfs_znode_move); 383} 384 385void 386zfs_znode_fini(void) 387{ 388#ifdef sun 389 /* 390 * Cleanup vfs & vnode ops 391 */ 392 zfs_remove_op_tables(); 393#endif /* sun */ 394 395 /* 396 * Cleanup zcache 397 */ 398 if (znode_cache) 399 kmem_cache_destroy(znode_cache); 400 znode_cache = NULL; 401 rw_destroy(&zfsvfs_lock); 402} 403 404#ifdef sun 405struct vnodeops *zfs_dvnodeops; 406struct vnodeops *zfs_fvnodeops; 407struct vnodeops *zfs_symvnodeops; 408struct vnodeops *zfs_xdvnodeops; 409struct vnodeops *zfs_evnodeops; 410struct vnodeops *zfs_sharevnodeops; 411 412void 413zfs_remove_op_tables() 414{ 415 /* 416 * Remove vfs ops 417 */ 418 ASSERT(zfsfstype); 419 (void) vfs_freevfsops_by_type(zfsfstype); 420 zfsfstype = 0; 421 422 /* 423 * Remove vnode ops 424 */ 425 if (zfs_dvnodeops) 426 vn_freevnodeops(zfs_dvnodeops); 427 if (zfs_fvnodeops) 428 vn_freevnodeops(zfs_fvnodeops); 429 if (zfs_symvnodeops) 430 vn_freevnodeops(zfs_symvnodeops); 431 if (zfs_xdvnodeops) 432 vn_freevnodeops(zfs_xdvnodeops); 433 if (zfs_evnodeops) 434 vn_freevnodeops(zfs_evnodeops); 435 if (zfs_sharevnodeops) 436 vn_freevnodeops(zfs_sharevnodeops); 437 438 zfs_dvnodeops = NULL; 439 zfs_fvnodeops = NULL; 440 zfs_symvnodeops = NULL; 441 zfs_xdvnodeops = NULL; 442 zfs_evnodeops = NULL; 443 zfs_sharevnodeops = NULL; 444} 445 446extern const fs_operation_def_t zfs_dvnodeops_template[]; 447extern const fs_operation_def_t zfs_fvnodeops_template[]; 448extern const fs_operation_def_t zfs_xdvnodeops_template[]; 449extern const fs_operation_def_t zfs_symvnodeops_template[]; 450extern const fs_operation_def_t zfs_evnodeops_template[]; 451extern const fs_operation_def_t zfs_sharevnodeops_template[]; 452 453int 454zfs_create_op_tables() 455{ 456 int error; 457 458 /* 459 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 460 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 461 * In this case we just return as the ops vectors are already set up. 462 */ 463 if (zfs_dvnodeops) 464 return (0); 465 466 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 467 &zfs_dvnodeops); 468 if (error) 469 return (error); 470 471 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 472 &zfs_fvnodeops); 473 if (error) 474 return (error); 475 476 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 477 &zfs_symvnodeops); 478 if (error) 479 return (error); 480 481 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 482 &zfs_xdvnodeops); 483 if (error) 484 return (error); 485 486 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 487 &zfs_evnodeops); 488 if (error) 489 return (error); 490 491 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 492 &zfs_sharevnodeops); 493 494 return (error); 495} 496#endif /* sun */ 497 498int 499zfs_create_share_dir(zfsvfs_t *zfsvfs, dmu_tx_t *tx) 500{ 501 zfs_acl_ids_t acl_ids; 502 vattr_t vattr; 503 znode_t *sharezp; 504 vnode_t *vp, vnode; 505 znode_t *zp; 506 int error; 507 508 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 509 vattr.va_type = VDIR; 510 vattr.va_mode = S_IFDIR|0555; 511 vattr.va_uid = crgetuid(kcred); 512 vattr.va_gid = crgetgid(kcred); 513 514 sharezp = kmem_cache_alloc(znode_cache, KM_SLEEP); 515 zfs_znode_cache_constructor(sharezp, zfsvfs->z_parent->z_vfs, 0); 516 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 517 sharezp->z_moved = 0; 518 sharezp->z_unlinked = 0; 519 sharezp->z_atime_dirty = 0; 520 sharezp->z_zfsvfs = zfsvfs; 521 sharezp->z_is_sa = zfsvfs->z_use_sa; 522 523 sharezp->z_vnode = &vnode; 524 vnode.v_data = sharezp; 525 526 vp = ZTOV(sharezp); 527 vp->v_type = VDIR; 528 529 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 530 kcred, NULL, &acl_ids)); 531 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 532 ASSERT3P(zp, ==, sharezp); 533 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 534 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 535 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 536 zfsvfs->z_shares_dir = sharezp->z_id; 537 538 zfs_acl_ids_free(&acl_ids); 539 ZTOV(sharezp)->v_data = NULL; 540 ZTOV(sharezp)->v_count = 0; 541 ZTOV(sharezp)->v_holdcnt = 0; 542 zp->z_vnode = NULL; 543 sa_handle_destroy(sharezp->z_sa_hdl); 544 sharezp->z_vnode = NULL; 545 kmem_cache_free(znode_cache, sharezp); 546 547 return (error); 548} 549 550/* 551 * define a couple of values we need available 552 * for both 64 and 32 bit environments. 553 */ 554#ifndef NBITSMINOR64 555#define NBITSMINOR64 32 556#endif 557#ifndef MAXMAJ64 558#define MAXMAJ64 0xffffffffUL 559#endif 560#ifndef MAXMIN64 561#define MAXMIN64 0xffffffffUL 562#endif 563 564/* 565 * Create special expldev for ZFS private use. 566 * Can't use standard expldev since it doesn't do 567 * what we want. The standard expldev() takes a 568 * dev32_t in LP64 and expands it to a long dev_t. 569 * We need an interface that takes a dev32_t in ILP32 570 * and expands it to a long dev_t. 571 */ 572static uint64_t 573zfs_expldev(dev_t dev) 574{ 575 return (((uint64_t)major(dev) << NBITSMINOR64) | minor(dev)); 576} 577/* 578 * Special cmpldev for ZFS private use. 579 * Can't use standard cmpldev since it takes 580 * a long dev_t and compresses it to dev32_t in 581 * LP64. We need to do a compaction of a long dev_t 582 * to a dev32_t in ILP32. 583 */ 584dev_t 585zfs_cmpldev(uint64_t dev) 586{ 587 return (makedev((dev >> NBITSMINOR64), (dev & MAXMIN64))); 588} 589 590static void 591zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 592 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 593{ 594 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 595 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 596 597 mutex_enter(&zp->z_lock); 598 599 ASSERT(zp->z_sa_hdl == NULL); 600 ASSERT(zp->z_acl_cached == NULL); 601 if (sa_hdl == NULL) { 602 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 603 SA_HDL_SHARED, &zp->z_sa_hdl)); 604 } else { 605 zp->z_sa_hdl = sa_hdl; 606 sa_set_userp(sa_hdl, zp); 607 } 608 609 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 610 611 /* 612 * Slap on VROOT if we are the root znode 613 */ 614 if (zp->z_id == zfsvfs->z_root) 615 ZTOV(zp)->v_flag |= VROOT; 616 617 mutex_exit(&zp->z_lock); 618 vn_exists(ZTOV(zp)); 619} 620 621void 622zfs_znode_dmu_fini(znode_t *zp) 623{ 624 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 625 zp->z_unlinked || 626 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 627 628 sa_handle_destroy(zp->z_sa_hdl); 629 zp->z_sa_hdl = NULL; 630} 631 632static void 633zfs_vnode_forget(vnode_t *vp) 634{ 635 636 VOP_UNLOCK(vp, 0); 637 VI_LOCK(vp); 638 vp->v_usecount--; 639 vp->v_iflag |= VI_DOOMED; 640 vp->v_data = NULL; 641 vdropl(vp); 642} 643 644/* 645 * Construct a new znode/vnode and intialize. 646 * 647 * This does not do a call to dmu_set_user() that is 648 * up to the caller to do, in case you don't want to 649 * return the znode 650 */ 651static znode_t * 652zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 653 dmu_object_type_t obj_type, sa_handle_t *hdl) 654{ 655 znode_t *zp; 656 vnode_t *vp; 657 uint64_t mode; 658 uint64_t parent; 659 sa_bulk_attr_t bulk[9]; 660 int count = 0; 661 662 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 663 zfs_znode_cache_constructor(zp, zfsvfs->z_parent->z_vfs, 0); 664 665 ASSERT(zp->z_dirlocks == NULL); 666 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 667 zp->z_moved = 0; 668 669 /* 670 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 671 * the zfs_znode_move() callback. 672 */ 673 zp->z_sa_hdl = NULL; 674 zp->z_unlinked = 0; 675 zp->z_atime_dirty = 0; 676 zp->z_mapcnt = 0; 677 zp->z_id = db->db_object; 678 zp->z_blksz = blksz; 679 zp->z_seq = 0x7A4653; 680 zp->z_sync_cnt = 0; 681 682 vp = ZTOV(zp); 683 684 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 685 686 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 687 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 688 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 689 &zp->z_size, 8); 690 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 691 &zp->z_links, 8); 692 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 693 &zp->z_pflags, 8); 694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 695 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 696 &zp->z_atime, 16); 697 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 698 &zp->z_uid, 8); 699 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 700 &zp->z_gid, 8); 701 702 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 703 if (hdl == NULL) 704 sa_handle_destroy(zp->z_sa_hdl); 705 zfs_vnode_forget(vp); 706 zp->z_vnode = NULL; 707 kmem_cache_free(znode_cache, zp); 708 return (NULL); 709 } 710 711 zp->z_mode = mode; 712 713 vp->v_type = IFTOVT((mode_t)mode); 714 715 switch (vp->v_type) { 716 case VDIR: 717 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 718 break; 719#ifdef sun 720 case VBLK: 721 case VCHR: 722 { 723 uint64_t rdev; 724 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 725 &rdev, sizeof (rdev)) == 0); 726 727 vp->v_rdev = zfs_cmpldev(rdev); 728 } 729 break; 730#endif /* sun */ 731 case VFIFO: 732#ifdef sun 733 case VSOCK: 734 case VDOOR: 735#endif /* sun */ 736 vp->v_op = &zfs_fifoops; 737 break; 738 case VREG: 739 if (parent == zfsvfs->z_shares_dir) { 740 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 741 vp->v_op = &zfs_shareops; 742 } 743 break; 744#ifdef sun 745 case VLNK: 746 vn_setops(vp, zfs_symvnodeops); 747 break; 748 default: 749 vn_setops(vp, zfs_evnodeops); 750 break; 751#endif /* sun */ 752 } 753 if (vp->v_type != VFIFO) 754 VN_LOCK_ASHARE(vp); 755 756 mutex_enter(&zfsvfs->z_znodes_lock); 757 list_insert_tail(&zfsvfs->z_all_znodes, zp); 758 membar_producer(); 759 /* 760 * Everything else must be valid before assigning z_zfsvfs makes the 761 * znode eligible for zfs_znode_move(). 762 */ 763 zp->z_zfsvfs = zfsvfs; 764 mutex_exit(&zfsvfs->z_znodes_lock); 765 766 VFS_HOLD(zfsvfs->z_vfs); 767 return (zp); 768} 769 770static uint64_t empty_xattr; 771static uint64_t pad[4]; 772static zfs_acl_phys_t acl_phys; 773/* 774 * Create a new DMU object to hold a zfs znode. 775 * 776 * IN: dzp - parent directory for new znode 777 * vap - file attributes for new znode 778 * tx - dmu transaction id for zap operations 779 * cr - credentials of caller 780 * flag - flags: 781 * IS_ROOT_NODE - new object will be root 782 * IS_XATTR - new object is an attribute 783 * bonuslen - length of bonus buffer 784 * setaclp - File/Dir initial ACL 785 * fuidp - Tracks fuid allocation. 786 * 787 * OUT: zpp - allocated znode 788 * 789 */ 790void 791zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 792 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 793{ 794 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 795 uint64_t mode, size, links, parent, pflags; 796 uint64_t dzp_pflags = 0; 797 uint64_t rdev = 0; 798 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 799 dmu_buf_t *db; 800 timestruc_t now; 801 uint64_t gen, obj; 802 int err; 803 int bonuslen; 804 sa_handle_t *sa_hdl; 805 dmu_object_type_t obj_type; 806 sa_bulk_attr_t sa_attrs[ZPL_END]; 807 int cnt = 0; 808 zfs_acl_locator_cb_t locate = { 0 }; 809 810 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 811 812 if (zfsvfs->z_replay) { 813 obj = vap->va_nodeid; 814 now = vap->va_ctime; /* see zfs_replay_create() */ 815 gen = vap->va_nblocks; /* ditto */ 816 } else { 817 obj = 0; 818 gethrestime(&now); 819 gen = dmu_tx_get_txg(tx); 820 } 821 822 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 823 bonuslen = (obj_type == DMU_OT_SA) ? 824 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 825 826 /* 827 * Create a new DMU object. 828 */ 829 /* 830 * There's currently no mechanism for pre-reading the blocks that will 831 * be needed to allocate a new object, so we accept the small chance 832 * that there will be an i/o error and we will fail one of the 833 * assertions below. 834 */ 835 if (vap->va_type == VDIR) { 836 if (zfsvfs->z_replay) { 837 err = zap_create_claim_norm(zfsvfs->z_os, obj, 838 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 839 obj_type, bonuslen, tx); 840 ASSERT0(err); 841 } else { 842 obj = zap_create_norm(zfsvfs->z_os, 843 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 844 obj_type, bonuslen, tx); 845 } 846 } else { 847 if (zfsvfs->z_replay) { 848 err = dmu_object_claim(zfsvfs->z_os, obj, 849 DMU_OT_PLAIN_FILE_CONTENTS, 0, 850 obj_type, bonuslen, tx); 851 ASSERT0(err); 852 } else { 853 obj = dmu_object_alloc(zfsvfs->z_os, 854 DMU_OT_PLAIN_FILE_CONTENTS, 0, 855 obj_type, bonuslen, tx); 856 } 857 } 858 859 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 860 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 861 862 /* 863 * If this is the root, fix up the half-initialized parent pointer 864 * to reference the just-allocated physical data area. 865 */ 866 if (flag & IS_ROOT_NODE) { 867 dzp->z_id = obj; 868 } else { 869 dzp_pflags = dzp->z_pflags; 870 } 871 872 /* 873 * If parent is an xattr, so am I. 874 */ 875 if (dzp_pflags & ZFS_XATTR) { 876 flag |= IS_XATTR; 877 } 878 879 if (zfsvfs->z_use_fuids) 880 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 881 else 882 pflags = 0; 883 884 if (vap->va_type == VDIR) { 885 size = 2; /* contents ("." and "..") */ 886 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 887 } else { 888 size = links = 0; 889 } 890 891 if (vap->va_type == VBLK || vap->va_type == VCHR) { 892 rdev = zfs_expldev(vap->va_rdev); 893 } 894 895 parent = dzp->z_id; 896 mode = acl_ids->z_mode; 897 if (flag & IS_XATTR) 898 pflags |= ZFS_XATTR; 899 900 /* 901 * No execs denied will be deterimed when zfs_mode_compute() is called. 902 */ 903 pflags |= acl_ids->z_aclp->z_hints & 904 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 905 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 906 907 ZFS_TIME_ENCODE(&now, crtime); 908 ZFS_TIME_ENCODE(&now, ctime); 909 910 if (vap->va_mask & AT_ATIME) { 911 ZFS_TIME_ENCODE(&vap->va_atime, atime); 912 } else { 913 ZFS_TIME_ENCODE(&now, atime); 914 } 915 916 if (vap->va_mask & AT_MTIME) { 917 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 918 } else { 919 ZFS_TIME_ENCODE(&now, mtime); 920 } 921 922 /* Now add in all of the "SA" attributes */ 923 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 924 &sa_hdl)); 925 926 /* 927 * Setup the array of attributes to be replaced/set on the new file 928 * 929 * order for DMU_OT_ZNODE is critical since it needs to be constructed 930 * in the old znode_phys_t format. Don't change this ordering 931 */ 932 933 if (obj_type == DMU_OT_ZNODE) { 934 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 935 NULL, &atime, 16); 936 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 937 NULL, &mtime, 16); 938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 939 NULL, &ctime, 16); 940 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 941 NULL, &crtime, 16); 942 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 943 NULL, &gen, 8); 944 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 945 NULL, &mode, 8); 946 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 947 NULL, &size, 8); 948 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 949 NULL, &parent, 8); 950 } else { 951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 952 NULL, &mode, 8); 953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 954 NULL, &size, 8); 955 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 956 NULL, &gen, 8); 957 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 958 &acl_ids->z_fuid, 8); 959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 960 &acl_ids->z_fgid, 8); 961 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 962 NULL, &parent, 8); 963 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 964 NULL, &pflags, 8); 965 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 966 NULL, &atime, 16); 967 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 968 NULL, &mtime, 16); 969 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 970 NULL, &ctime, 16); 971 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 972 NULL, &crtime, 16); 973 } 974 975 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 976 977 if (obj_type == DMU_OT_ZNODE) { 978 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 979 &empty_xattr, 8); 980 } 981 if (obj_type == DMU_OT_ZNODE || 982 (vap->va_type == VBLK || vap->va_type == VCHR)) { 983 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 984 NULL, &rdev, 8); 985 986 } 987 if (obj_type == DMU_OT_ZNODE) { 988 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 989 NULL, &pflags, 8); 990 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 991 &acl_ids->z_fuid, 8); 992 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 993 &acl_ids->z_fgid, 8); 994 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 995 sizeof (uint64_t) * 4); 996 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 997 &acl_phys, sizeof (zfs_acl_phys_t)); 998 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 999 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 1000 &acl_ids->z_aclp->z_acl_count, 8); 1001 locate.cb_aclp = acl_ids->z_aclp; 1002 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 1003 zfs_acl_data_locator, &locate, 1004 acl_ids->z_aclp->z_acl_bytes); 1005 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 1006 acl_ids->z_fuid, acl_ids->z_fgid); 1007 } 1008 1009 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 1010 1011 if (!(flag & IS_ROOT_NODE)) { 1012 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 1013 ASSERT(*zpp != NULL); 1014 } else { 1015 /* 1016 * If we are creating the root node, the "parent" we 1017 * passed in is the znode for the root. 1018 */ 1019 *zpp = dzp; 1020 1021 (*zpp)->z_sa_hdl = sa_hdl; 1022 } 1023 1024 (*zpp)->z_pflags = pflags; 1025 (*zpp)->z_mode = mode; 1026 1027 if (vap->va_mask & AT_XVATTR) 1028 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 1029 1030 if (obj_type == DMU_OT_ZNODE || 1031 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1032 err = zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx); 1033 ASSERT0(err); 1034 } 1035 if (!(flag & IS_ROOT_NODE)) { 1036 vnode_t *vp; 1037 1038 vp = ZTOV(*zpp); 1039 vp->v_vflag |= VV_FORCEINSMQ; 1040 err = insmntque(vp, zfsvfs->z_vfs); 1041 vp->v_vflag &= ~VV_FORCEINSMQ; 1042 KASSERT(err == 0, ("insmntque() failed: error %d", err)); 1043 } 1044 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1045} 1046 1047/* 1048 * zfs_xvattr_set only updates the in-core attributes 1049 * it is assumed the caller will be doing an sa_bulk_update 1050 * to push the changes out 1051 */ 1052void 1053zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1054{ 1055 xoptattr_t *xoap; 1056 1057 xoap = xva_getxoptattr(xvap); 1058 ASSERT(xoap); 1059 1060 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1061 uint64_t times[2]; 1062 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1063 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1064 ×, sizeof (times), tx); 1065 XVA_SET_RTN(xvap, XAT_CREATETIME); 1066 } 1067 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1068 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1069 zp->z_pflags, tx); 1070 XVA_SET_RTN(xvap, XAT_READONLY); 1071 } 1072 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1073 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1074 zp->z_pflags, tx); 1075 XVA_SET_RTN(xvap, XAT_HIDDEN); 1076 } 1077 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1078 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1079 zp->z_pflags, tx); 1080 XVA_SET_RTN(xvap, XAT_SYSTEM); 1081 } 1082 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1083 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1084 zp->z_pflags, tx); 1085 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1086 } 1087 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1088 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1089 zp->z_pflags, tx); 1090 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1091 } 1092 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1093 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1094 zp->z_pflags, tx); 1095 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1096 } 1097 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1098 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1099 zp->z_pflags, tx); 1100 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1101 } 1102 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1103 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1104 zp->z_pflags, tx); 1105 XVA_SET_RTN(xvap, XAT_NODUMP); 1106 } 1107 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1108 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1109 zp->z_pflags, tx); 1110 XVA_SET_RTN(xvap, XAT_OPAQUE); 1111 } 1112 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1113 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1114 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1115 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1116 } 1117 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1118 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1119 zp->z_pflags, tx); 1120 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1121 } 1122 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1123 zfs_sa_set_scanstamp(zp, xvap, tx); 1124 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1125 } 1126 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1127 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1128 zp->z_pflags, tx); 1129 XVA_SET_RTN(xvap, XAT_REPARSE); 1130 } 1131 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1132 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1133 zp->z_pflags, tx); 1134 XVA_SET_RTN(xvap, XAT_OFFLINE); 1135 } 1136 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1137 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1138 zp->z_pflags, tx); 1139 XVA_SET_RTN(xvap, XAT_SPARSE); 1140 } 1141} 1142 1143int 1144zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1145{ 1146 dmu_object_info_t doi; 1147 dmu_buf_t *db; 1148 znode_t *zp; 1149 int err; 1150 sa_handle_t *hdl; 1151 int first = 1; 1152 1153 *zpp = NULL; 1154 1155again: 1156 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1157 1158 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1159 if (err) { 1160 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1161 return (err); 1162 } 1163 1164 dmu_object_info_from_db(db, &doi); 1165 if (doi.doi_bonus_type != DMU_OT_SA && 1166 (doi.doi_bonus_type != DMU_OT_ZNODE || 1167 (doi.doi_bonus_type == DMU_OT_ZNODE && 1168 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1169 sa_buf_rele(db, NULL); 1170 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1171 return (EINVAL); 1172 } 1173 1174 hdl = dmu_buf_get_user(db); 1175 if (hdl != NULL) { 1176 zp = sa_get_userdata(hdl); 1177 1178 1179 /* 1180 * Since "SA" does immediate eviction we 1181 * should never find a sa handle that doesn't 1182 * know about the znode. 1183 */ 1184 1185 ASSERT3P(zp, !=, NULL); 1186 1187 mutex_enter(&zp->z_lock); 1188 ASSERT3U(zp->z_id, ==, obj_num); 1189 if (zp->z_unlinked) { 1190 err = ENOENT; 1191 } else { 1192 vnode_t *vp; 1193 int dying = 0; 1194 1195 vp = ZTOV(zp); 1196 if (vp == NULL) 1197 dying = 1; 1198 else { 1199 VN_HOLD(vp); 1200 if ((vp->v_iflag & VI_DOOMED) != 0) { 1201 dying = 1; 1202 /* 1203 * Don't VN_RELE() vnode here, because 1204 * it can call vn_lock() which creates 1205 * LOR between vnode lock and znode 1206 * lock. We will VN_RELE() the vnode 1207 * after droping znode lock. 1208 */ 1209 } 1210 } 1211 if (dying) { 1212 if (first) { 1213 ZFS_LOG(1, "dying znode detected (zp=%p)", zp); 1214 first = 0; 1215 } 1216 /* 1217 * znode is dying so we can't reuse it, we must 1218 * wait until destruction is completed. 1219 */ 1220 sa_buf_rele(db, NULL); 1221 mutex_exit(&zp->z_lock); 1222 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1223 if (vp != NULL) 1224 VN_RELE(vp); 1225 tsleep(zp, 0, "zcollide", 1); 1226 goto again; 1227 } 1228 *zpp = zp; 1229 err = 0; 1230 } 1231 sa_buf_rele(db, NULL); 1232 mutex_exit(&zp->z_lock); 1233 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1234 return (err); 1235 } 1236 1237 /* 1238 * Not found create new znode/vnode 1239 * but only if file exists. 1240 * 1241 * There is a small window where zfs_vget() could 1242 * find this object while a file create is still in 1243 * progress. This is checked for in zfs_znode_alloc() 1244 * 1245 * if zfs_znode_alloc() fails it will drop the hold on the 1246 * bonus buffer. 1247 */ 1248 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1249 doi.doi_bonus_type, NULL); 1250 if (zp == NULL) { 1251 err = ENOENT; 1252 } else { 1253 *zpp = zp; 1254 } 1255 if (err == 0) { 1256 vnode_t *vp = ZTOV(zp); 1257 1258 err = insmntque(vp, zfsvfs->z_vfs); 1259 if (err == 0) 1260 VOP_UNLOCK(vp, 0); 1261 else { 1262 zp->z_vnode = NULL; 1263 zfs_znode_dmu_fini(zp); 1264 zfs_znode_free(zp); 1265 *zpp = NULL; 1266 } 1267 } 1268 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1269 return (err); 1270} 1271 1272int 1273zfs_rezget(znode_t *zp) 1274{ 1275 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1276 dmu_object_info_t doi; 1277 dmu_buf_t *db; 1278 vnode_t *vp; 1279 uint64_t obj_num = zp->z_id; 1280 uint64_t mode, size; 1281 sa_bulk_attr_t bulk[8]; 1282 int err; 1283 int count = 0; 1284 uint64_t gen; 1285 1286 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1287 1288 mutex_enter(&zp->z_acl_lock); 1289 if (zp->z_acl_cached) { 1290 zfs_acl_free(zp->z_acl_cached); 1291 zp->z_acl_cached = NULL; 1292 } 1293 1294 mutex_exit(&zp->z_acl_lock); 1295 ASSERT(zp->z_sa_hdl == NULL); 1296 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1297 if (err) { 1298 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1299 return (err); 1300 } 1301 1302 dmu_object_info_from_db(db, &doi); 1303 if (doi.doi_bonus_type != DMU_OT_SA && 1304 (doi.doi_bonus_type != DMU_OT_ZNODE || 1305 (doi.doi_bonus_type == DMU_OT_ZNODE && 1306 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1307 sa_buf_rele(db, NULL); 1308 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1309 return (EINVAL); 1310 } 1311 1312 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1313 size = zp->z_size; 1314 1315 /* reload cached values */ 1316 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1317 &gen, sizeof (gen)); 1318 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1319 &zp->z_size, sizeof (zp->z_size)); 1320 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1321 &zp->z_links, sizeof (zp->z_links)); 1322 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1323 &zp->z_pflags, sizeof (zp->z_pflags)); 1324 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1325 &zp->z_atime, sizeof (zp->z_atime)); 1326 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1327 &zp->z_uid, sizeof (zp->z_uid)); 1328 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1329 &zp->z_gid, sizeof (zp->z_gid)); 1330 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1331 &mode, sizeof (mode)); 1332 1333 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1334 zfs_znode_dmu_fini(zp); 1335 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1336 return (EIO); 1337 } 1338 1339 zp->z_mode = mode; 1340 1341 if (gen != zp->z_gen) { 1342 zfs_znode_dmu_fini(zp); 1343 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1344 return (EIO); 1345 } 1346 1347 /* 1348 * XXXPJD: Not sure how is that possible, but under heavy 1349 * zfs recv -F load it happens that z_gen is the same, but 1350 * vnode type is different than znode type. This would mean 1351 * that for example regular file was replaced with directory 1352 * which has the same object number. 1353 */ 1354 vp = ZTOV(zp); 1355 if (vp != NULL && 1356 vp->v_type != IFTOVT((mode_t)zp->z_mode)) { 1357 zfs_znode_dmu_fini(zp); 1358 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1359 return (EIO); 1360 } 1361 1362 zp->z_unlinked = (zp->z_links == 0); 1363 zp->z_blksz = doi.doi_data_block_size; 1364 if (vp != NULL) { 1365 vn_pages_remove(vp, 0, 0); 1366 if (zp->z_size != size) 1367 vnode_pager_setsize(vp, zp->z_size); 1368 } 1369 1370 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1371 1372 return (0); 1373} 1374 1375void 1376zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1377{ 1378 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1379 objset_t *os = zfsvfs->z_os; 1380 uint64_t obj = zp->z_id; 1381 uint64_t acl_obj = zfs_external_acl(zp); 1382 1383 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1384 if (acl_obj) { 1385 VERIFY(!zp->z_is_sa); 1386 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1387 } 1388 VERIFY(0 == dmu_object_free(os, obj, tx)); 1389 zfs_znode_dmu_fini(zp); 1390 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1391 zfs_znode_free(zp); 1392} 1393 1394void 1395zfs_zinactive(znode_t *zp) 1396{ 1397 vnode_t *vp = ZTOV(zp); 1398 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1399 uint64_t z_id = zp->z_id; 1400 int vfslocked; 1401 1402 ASSERT(zp->z_sa_hdl); 1403 1404 /* 1405 * Don't allow a zfs_zget() while were trying to release this znode 1406 */ 1407 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1408 1409 mutex_enter(&zp->z_lock); 1410 VI_LOCK(vp); 1411 if (vp->v_count > 0) { 1412 /* 1413 * If the hold count is greater than zero, somebody has 1414 * obtained a new reference on this znode while we were 1415 * processing it here, so we are done. 1416 */ 1417 VI_UNLOCK(vp); 1418 mutex_exit(&zp->z_lock); 1419 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1420 return; 1421 } 1422 VI_UNLOCK(vp); 1423 1424 /* 1425 * If this was the last reference to a file with no links, 1426 * remove the file from the file system. 1427 */ 1428 if (zp->z_unlinked) { 1429 mutex_exit(&zp->z_lock); 1430 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1431 ASSERT(vp->v_count == 0); 1432 vrecycle(vp); 1433 vfslocked = VFS_LOCK_GIANT(zfsvfs->z_vfs); 1434 zfs_rmnode(zp); 1435 VFS_UNLOCK_GIANT(vfslocked); 1436 return; 1437 } 1438 1439 mutex_exit(&zp->z_lock); 1440 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1441} 1442 1443void 1444zfs_znode_free(znode_t *zp) 1445{ 1446 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1447 1448 ASSERT(ZTOV(zp) == NULL); 1449 ASSERT(zp->z_sa_hdl == NULL); 1450 mutex_enter(&zfsvfs->z_znodes_lock); 1451 POINTER_INVALIDATE(&zp->z_zfsvfs); 1452 list_remove(&zfsvfs->z_all_znodes, zp); 1453 mutex_exit(&zfsvfs->z_znodes_lock); 1454 1455 if (zp->z_acl_cached) { 1456 zfs_acl_free(zp->z_acl_cached); 1457 zp->z_acl_cached = NULL; 1458 } 1459 1460 kmem_cache_free(znode_cache, zp); 1461 1462 VFS_RELE(zfsvfs->z_vfs); 1463} 1464 1465void 1466zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1467 uint64_t ctime[2], boolean_t have_tx) 1468{ 1469 timestruc_t now; 1470 1471 gethrestime(&now); 1472 1473 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1474 zp->z_atime_dirty = 0; 1475 zp->z_seq++; 1476 } else { 1477 zp->z_atime_dirty = 1; 1478 } 1479 1480 if (flag & AT_ATIME) { 1481 ZFS_TIME_ENCODE(&now, zp->z_atime); 1482 } 1483 1484 if (flag & AT_MTIME) { 1485 ZFS_TIME_ENCODE(&now, mtime); 1486 if (zp->z_zfsvfs->z_use_fuids) { 1487 zp->z_pflags |= (ZFS_ARCHIVE | 1488 ZFS_AV_MODIFIED); 1489 } 1490 } 1491 1492 if (flag & AT_CTIME) { 1493 ZFS_TIME_ENCODE(&now, ctime); 1494 if (zp->z_zfsvfs->z_use_fuids) 1495 zp->z_pflags |= ZFS_ARCHIVE; 1496 } 1497} 1498 1499/* 1500 * Grow the block size for a file. 1501 * 1502 * IN: zp - znode of file to free data in. 1503 * size - requested block size 1504 * tx - open transaction. 1505 * 1506 * NOTE: this function assumes that the znode is write locked. 1507 */ 1508void 1509zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1510{ 1511 int error; 1512 u_longlong_t dummy; 1513 1514 if (size <= zp->z_blksz) 1515 return; 1516 /* 1517 * If the file size is already greater than the current blocksize, 1518 * we will not grow. If there is more than one block in a file, 1519 * the blocksize cannot change. 1520 */ 1521 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1522 return; 1523 1524 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1525 size, 0, tx); 1526 1527 if (error == ENOTSUP) 1528 return; 1529 ASSERT0(error); 1530 1531 /* What blocksize did we actually get? */ 1532 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1533} 1534 1535#ifdef sun 1536/* 1537 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1538 * be calling back into the fs for a putpage(). E.g.: when truncating 1539 * a file, the pages being "thrown away* don't need to be written out. 1540 */ 1541/* ARGSUSED */ 1542static int 1543zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1544 int flags, cred_t *cr) 1545{ 1546 ASSERT(0); 1547 return (0); 1548} 1549#endif /* sun */ 1550 1551/* 1552 * Increase the file length 1553 * 1554 * IN: zp - znode of file to free data in. 1555 * end - new end-of-file 1556 * 1557 * RETURN: 0 if success 1558 * error code if failure 1559 */ 1560static int 1561zfs_extend(znode_t *zp, uint64_t end) 1562{ 1563 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1564 dmu_tx_t *tx; 1565 rl_t *rl; 1566 uint64_t newblksz; 1567 int error; 1568 1569 /* 1570 * We will change zp_size, lock the whole file. 1571 */ 1572 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1573 1574 /* 1575 * Nothing to do if file already at desired length. 1576 */ 1577 if (end <= zp->z_size) { 1578 zfs_range_unlock(rl); 1579 return (0); 1580 } 1581top: 1582 tx = dmu_tx_create(zfsvfs->z_os); 1583 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1584 zfs_sa_upgrade_txholds(tx, zp); 1585 if (end > zp->z_blksz && 1586 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1587 /* 1588 * We are growing the file past the current block size. 1589 */ 1590 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1591 ASSERT(!ISP2(zp->z_blksz)); 1592 newblksz = MIN(end, SPA_MAXBLOCKSIZE); 1593 } else { 1594 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1595 } 1596 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1597 } else { 1598 newblksz = 0; 1599 } 1600 1601 error = dmu_tx_assign(tx, TXG_NOWAIT); 1602 if (error) { 1603 if (error == ERESTART) { 1604 dmu_tx_wait(tx); 1605 dmu_tx_abort(tx); 1606 goto top; 1607 } 1608 dmu_tx_abort(tx); 1609 zfs_range_unlock(rl); 1610 return (error); 1611 } 1612 1613 if (newblksz) 1614 zfs_grow_blocksize(zp, newblksz, tx); 1615 1616 zp->z_size = end; 1617 1618 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1619 &zp->z_size, sizeof (zp->z_size), tx)); 1620 1621 vnode_pager_setsize(ZTOV(zp), end); 1622 1623 zfs_range_unlock(rl); 1624 1625 dmu_tx_commit(tx); 1626 1627 return (0); 1628} 1629 1630/* 1631 * Free space in a file. 1632 * 1633 * IN: zp - znode of file to free data in. 1634 * off - start of section to free. 1635 * len - length of section to free. 1636 * 1637 * RETURN: 0 if success 1638 * error code if failure 1639 */ 1640static int 1641zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1642{ 1643 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1644 rl_t *rl; 1645 int error; 1646 1647 /* 1648 * Lock the range being freed. 1649 */ 1650 rl = zfs_range_lock(zp, off, len, RL_WRITER); 1651 1652 /* 1653 * Nothing to do if file already at desired length. 1654 */ 1655 if (off >= zp->z_size) { 1656 zfs_range_unlock(rl); 1657 return (0); 1658 } 1659 1660 if (off + len > zp->z_size) 1661 len = zp->z_size - off; 1662 1663 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1664 1665 if (error == 0) { 1666 /* 1667 * In FreeBSD we cannot free block in the middle of a file, 1668 * but only at the end of a file, so this code path should 1669 * never happen. 1670 */ 1671 vnode_pager_setsize(ZTOV(zp), off); 1672 } 1673 1674 zfs_range_unlock(rl); 1675 1676 return (error); 1677} 1678 1679/* 1680 * Truncate a file 1681 * 1682 * IN: zp - znode of file to free data in. 1683 * end - new end-of-file. 1684 * 1685 * RETURN: 0 if success 1686 * error code if failure 1687 */ 1688static int 1689zfs_trunc(znode_t *zp, uint64_t end) 1690{ 1691 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1692 vnode_t *vp = ZTOV(zp); 1693 dmu_tx_t *tx; 1694 rl_t *rl; 1695 int error; 1696 sa_bulk_attr_t bulk[2]; 1697 int count = 0; 1698 1699 /* 1700 * We will change zp_size, lock the whole file. 1701 */ 1702 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1703 1704 /* 1705 * Nothing to do if file already at desired length. 1706 */ 1707 if (end >= zp->z_size) { 1708 zfs_range_unlock(rl); 1709 return (0); 1710 } 1711 1712 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, -1); 1713 if (error) { 1714 zfs_range_unlock(rl); 1715 return (error); 1716 } 1717top: 1718 tx = dmu_tx_create(zfsvfs->z_os); 1719 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1720 zfs_sa_upgrade_txholds(tx, zp); 1721 error = dmu_tx_assign(tx, TXG_NOWAIT); 1722 if (error) { 1723 if (error == ERESTART) { 1724 dmu_tx_wait(tx); 1725 dmu_tx_abort(tx); 1726 goto top; 1727 } 1728 dmu_tx_abort(tx); 1729 zfs_range_unlock(rl); 1730 return (error); 1731 } 1732 1733 zp->z_size = end; 1734 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1735 NULL, &zp->z_size, sizeof (zp->z_size)); 1736 1737 if (end == 0) { 1738 zp->z_pflags &= ~ZFS_SPARSE; 1739 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1740 NULL, &zp->z_pflags, 8); 1741 } 1742 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1743 1744 dmu_tx_commit(tx); 1745 1746 /* 1747 * Clear any mapped pages in the truncated region. This has to 1748 * happen outside of the transaction to avoid the possibility of 1749 * a deadlock with someone trying to push a page that we are 1750 * about to invalidate. 1751 */ 1752 vnode_pager_setsize(vp, end); 1753 1754 zfs_range_unlock(rl); 1755 1756 return (0); 1757} 1758 1759/* 1760 * Free space in a file 1761 * 1762 * IN: zp - znode of file to free data in. 1763 * off - start of range 1764 * len - end of range (0 => EOF) 1765 * flag - current file open mode flags. 1766 * log - TRUE if this action should be logged 1767 * 1768 * RETURN: 0 if success 1769 * error code if failure 1770 */ 1771int 1772zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1773{ 1774 vnode_t *vp = ZTOV(zp); 1775 dmu_tx_t *tx; 1776 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1777 zilog_t *zilog = zfsvfs->z_log; 1778 uint64_t mode; 1779 uint64_t mtime[2], ctime[2]; 1780 sa_bulk_attr_t bulk[3]; 1781 int count = 0; 1782 int error; 1783 1784 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1785 sizeof (mode))) != 0) 1786 return (error); 1787 1788 if (off > zp->z_size) { 1789 error = zfs_extend(zp, off+len); 1790 if (error == 0 && log) 1791 goto log; 1792 else 1793 return (error); 1794 } 1795 1796 /* 1797 * Check for any locks in the region to be freed. 1798 */ 1799 1800 if (MANDLOCK(vp, (mode_t)mode)) { 1801 uint64_t length = (len ? len : zp->z_size - off); 1802 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1803 return (error); 1804 } 1805 1806 if (len == 0) { 1807 error = zfs_trunc(zp, off); 1808 } else { 1809 if ((error = zfs_free_range(zp, off, len)) == 0 && 1810 off + len > zp->z_size) 1811 error = zfs_extend(zp, off+len); 1812 } 1813 if (error || !log) 1814 return (error); 1815log: 1816 tx = dmu_tx_create(zfsvfs->z_os); 1817 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1818 zfs_sa_upgrade_txholds(tx, zp); 1819 error = dmu_tx_assign(tx, TXG_NOWAIT); 1820 if (error) { 1821 if (error == ERESTART) { 1822 dmu_tx_wait(tx); 1823 dmu_tx_abort(tx); 1824 goto log; 1825 } 1826 dmu_tx_abort(tx); 1827 return (error); 1828 } 1829 1830 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1831 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1832 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1833 NULL, &zp->z_pflags, 8); 1834 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1835 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1836 ASSERT(error == 0); 1837 1838 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1839 1840 dmu_tx_commit(tx); 1841 return (0); 1842} 1843 1844void 1845zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1846{ 1847 zfsvfs_t zfsvfs; 1848 uint64_t moid, obj, sa_obj, version; 1849 uint64_t sense = ZFS_CASE_SENSITIVE; 1850 uint64_t norm = 0; 1851 nvpair_t *elem; 1852 int error; 1853 int i; 1854 znode_t *rootzp = NULL; 1855 vnode_t vnode; 1856 vattr_t vattr; 1857 znode_t *zp; 1858 zfs_acl_ids_t acl_ids; 1859 1860 /* 1861 * First attempt to create master node. 1862 */ 1863 /* 1864 * In an empty objset, there are no blocks to read and thus 1865 * there can be no i/o errors (which we assert below). 1866 */ 1867 moid = MASTER_NODE_OBJ; 1868 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1869 DMU_OT_NONE, 0, tx); 1870 ASSERT(error == 0); 1871 1872 /* 1873 * Set starting attributes. 1874 */ 1875 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1876 elem = NULL; 1877 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1878 /* For the moment we expect all zpl props to be uint64_ts */ 1879 uint64_t val; 1880 char *name; 1881 1882 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1883 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1884 name = nvpair_name(elem); 1885 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1886 if (val < version) 1887 version = val; 1888 } else { 1889 error = zap_update(os, moid, name, 8, 1, &val, tx); 1890 } 1891 ASSERT(error == 0); 1892 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1893 norm = val; 1894 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1895 sense = val; 1896 } 1897 ASSERT(version != 0); 1898 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1899 1900 /* 1901 * Create zap object used for SA attribute registration 1902 */ 1903 1904 if (version >= ZPL_VERSION_SA) { 1905 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1906 DMU_OT_NONE, 0, tx); 1907 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1908 ASSERT(error == 0); 1909 } else { 1910 sa_obj = 0; 1911 } 1912 /* 1913 * Create a delete queue. 1914 */ 1915 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1916 1917 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1918 ASSERT(error == 0); 1919 1920 /* 1921 * Create root znode. Create minimal znode/vnode/zfsvfs 1922 * to allow zfs_mknode to work. 1923 */ 1924 VATTR_NULL(&vattr); 1925 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1926 vattr.va_type = VDIR; 1927 vattr.va_mode = S_IFDIR|0755; 1928 vattr.va_uid = crgetuid(cr); 1929 vattr.va_gid = crgetgid(cr); 1930 1931 bzero(&zfsvfs, sizeof (zfsvfs_t)); 1932 1933 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1934 zfs_znode_cache_constructor(rootzp, NULL, 0); 1935 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1936 rootzp->z_moved = 0; 1937 rootzp->z_unlinked = 0; 1938 rootzp->z_atime_dirty = 0; 1939 rootzp->z_is_sa = USE_SA(version, os); 1940 1941 vnode.v_type = VDIR; 1942 vnode.v_data = rootzp; 1943 rootzp->z_vnode = &vnode; 1944 1945 zfsvfs.z_os = os; 1946 zfsvfs.z_parent = &zfsvfs; 1947 zfsvfs.z_version = version; 1948 zfsvfs.z_use_fuids = USE_FUIDS(version, os); 1949 zfsvfs.z_use_sa = USE_SA(version, os); 1950 zfsvfs.z_norm = norm; 1951 1952 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1953 &zfsvfs.z_attr_table); 1954 1955 ASSERT(error == 0); 1956 1957 /* 1958 * Fold case on file systems that are always or sometimes case 1959 * insensitive. 1960 */ 1961 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1962 zfsvfs.z_norm |= U8_TEXTPREP_TOUPPER; 1963 1964 mutex_init(&zfsvfs.z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1965 list_create(&zfsvfs.z_all_znodes, sizeof (znode_t), 1966 offsetof(znode_t, z_link_node)); 1967 1968 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1969 mutex_init(&zfsvfs.z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1970 1971 rootzp->z_zfsvfs = &zfsvfs; 1972 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1973 cr, NULL, &acl_ids)); 1974 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1975 ASSERT3P(zp, ==, rootzp); 1976 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1977 ASSERT(error == 0); 1978 zfs_acl_ids_free(&acl_ids); 1979 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1980 1981 sa_handle_destroy(rootzp->z_sa_hdl); 1982 rootzp->z_vnode = NULL; 1983 kmem_cache_free(znode_cache, rootzp); 1984 1985 /* 1986 * Create shares directory 1987 */ 1988 1989 error = zfs_create_share_dir(&zfsvfs, tx); 1990 1991 ASSERT(error == 0); 1992 1993 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1994 mutex_destroy(&zfsvfs.z_hold_mtx[i]); 1995} 1996 1997#endif /* _KERNEL */ 1998 1999static int 2000zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 2001{ 2002 uint64_t sa_obj = 0; 2003 int error; 2004 2005 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 2006 if (error != 0 && error != ENOENT) 2007 return (error); 2008 2009 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 2010 return (error); 2011} 2012 2013static int 2014zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 2015 dmu_buf_t **db, void *tag) 2016{ 2017 dmu_object_info_t doi; 2018 int error; 2019 2020 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 2021 return (error); 2022 2023 dmu_object_info_from_db(*db, &doi); 2024 if ((doi.doi_bonus_type != DMU_OT_SA && 2025 doi.doi_bonus_type != DMU_OT_ZNODE) || 2026 doi.doi_bonus_type == DMU_OT_ZNODE && 2027 doi.doi_bonus_size < sizeof (znode_phys_t)) { 2028 sa_buf_rele(*db, tag); 2029 return (ENOTSUP); 2030 } 2031 2032 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 2033 if (error != 0) { 2034 sa_buf_rele(*db, tag); 2035 return (error); 2036 } 2037 2038 return (0); 2039} 2040 2041void 2042zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 2043{ 2044 sa_handle_destroy(hdl); 2045 sa_buf_rele(db, tag); 2046} 2047 2048/* 2049 * Given an object number, return its parent object number and whether 2050 * or not the object is an extended attribute directory. 2051 */ 2052static int 2053zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table, 2054 uint64_t *pobjp, int *is_xattrdir) 2055{ 2056 uint64_t parent; 2057 uint64_t pflags; 2058 uint64_t mode; 2059 uint64_t parent_mode; 2060 sa_bulk_attr_t bulk[3]; 2061 sa_handle_t *sa_hdl; 2062 dmu_buf_t *sa_db; 2063 int count = 0; 2064 int error; 2065 2066 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 2067 &parent, sizeof (parent)); 2068 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 2069 &pflags, sizeof (pflags)); 2070 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2071 &mode, sizeof (mode)); 2072 2073 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2074 return (error); 2075 2076 /* 2077 * When a link is removed its parent pointer is not changed and will 2078 * be invalid. There are two cases where a link is removed but the 2079 * file stays around, when it goes to the delete queue and when there 2080 * are additional links. 2081 */ 2082 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG); 2083 if (error != 0) 2084 return (error); 2085 2086 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode)); 2087 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2088 if (error != 0) 2089 return (error); 2090 2091 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2092 2093 /* 2094 * Extended attributes can be applied to files, directories, etc. 2095 * Otherwise the parent must be a directory. 2096 */ 2097 if (!*is_xattrdir && !S_ISDIR(parent_mode)) 2098 return (EINVAL); 2099 2100 *pobjp = parent; 2101 2102 return (0); 2103} 2104 2105/* 2106 * Given an object number, return some zpl level statistics 2107 */ 2108static int 2109zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2110 zfs_stat_t *sb) 2111{ 2112 sa_bulk_attr_t bulk[4]; 2113 int count = 0; 2114 2115 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2116 &sb->zs_mode, sizeof (sb->zs_mode)); 2117 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2118 &sb->zs_gen, sizeof (sb->zs_gen)); 2119 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2120 &sb->zs_links, sizeof (sb->zs_links)); 2121 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2122 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2123 2124 return (sa_bulk_lookup(hdl, bulk, count)); 2125} 2126 2127static int 2128zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2129 sa_attr_type_t *sa_table, char *buf, int len) 2130{ 2131 sa_handle_t *sa_hdl; 2132 sa_handle_t *prevhdl = NULL; 2133 dmu_buf_t *prevdb = NULL; 2134 dmu_buf_t *sa_db = NULL; 2135 char *path = buf + len - 1; 2136 int error; 2137 2138 *path = '\0'; 2139 sa_hdl = hdl; 2140 2141 for (;;) { 2142 uint64_t pobj; 2143 char component[MAXNAMELEN + 2]; 2144 size_t complen; 2145 int is_xattrdir; 2146 2147 if (prevdb) 2148 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2149 2150 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj, 2151 &is_xattrdir)) != 0) 2152 break; 2153 2154 if (pobj == obj) { 2155 if (path[0] != '/') 2156 *--path = '/'; 2157 break; 2158 } 2159 2160 component[0] = '/'; 2161 if (is_xattrdir) { 2162 (void) sprintf(component + 1, "<xattrdir>"); 2163 } else { 2164 error = zap_value_search(osp, pobj, obj, 2165 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2166 if (error != 0) 2167 break; 2168 } 2169 2170 complen = strlen(component); 2171 path -= complen; 2172 ASSERT(path >= buf); 2173 bcopy(component, path, complen); 2174 obj = pobj; 2175 2176 if (sa_hdl != hdl) { 2177 prevhdl = sa_hdl; 2178 prevdb = sa_db; 2179 } 2180 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2181 if (error != 0) { 2182 sa_hdl = prevhdl; 2183 sa_db = prevdb; 2184 break; 2185 } 2186 } 2187 2188 if (sa_hdl != NULL && sa_hdl != hdl) { 2189 ASSERT(sa_db != NULL); 2190 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2191 } 2192 2193 if (error == 0) 2194 (void) memmove(buf, path, buf + len - path); 2195 2196 return (error); 2197} 2198 2199int 2200zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2201{ 2202 sa_attr_type_t *sa_table; 2203 sa_handle_t *hdl; 2204 dmu_buf_t *db; 2205 int error; 2206 2207 error = zfs_sa_setup(osp, &sa_table); 2208 if (error != 0) 2209 return (error); 2210 2211 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2212 if (error != 0) 2213 return (error); 2214 2215 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2216 2217 zfs_release_sa_handle(hdl, db, FTAG); 2218 return (error); 2219} 2220 2221int 2222zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2223 char *buf, int len) 2224{ 2225 char *path = buf + len - 1; 2226 sa_attr_type_t *sa_table; 2227 sa_handle_t *hdl; 2228 dmu_buf_t *db; 2229 int error; 2230 2231 *path = '\0'; 2232 2233 error = zfs_sa_setup(osp, &sa_table); 2234 if (error != 0) 2235 return (error); 2236 2237 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2238 if (error != 0) 2239 return (error); 2240 2241 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2242 if (error != 0) { 2243 zfs_release_sa_handle(hdl, db, FTAG); 2244 return (error); 2245 } 2246 2247 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2248 2249 zfs_release_sa_handle(hdl, db, FTAG); 2250 return (error); 2251} 2252