1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2018 by Delphix. All rights reserved. 24 * Copyright (c) 2014 Integros [integros.com] 25 */ 26 27 /* Portions Copyright 2007 Jeremy Teo */ 28 29 #ifdef _KERNEL 30 #include <sys/types.h> 31 #include <sys/param.h> 32 #include <sys/time.h> 33 #include <sys/systm.h> 34 #include <sys/sysmacros.h> 35 #include <sys/resource.h> 36 #include <sys/mntent.h> 37 #include <sys/mkdev.h> 38 #include <sys/u8_textprep.h> 39 #include <sys/dsl_dataset.h> 40 #include <sys/vfs.h> 41 #include <sys/vfs_opreg.h> 42 #include <sys/vnode.h> 43 #include <sys/file.h> 44 #include <sys/kmem.h> 45 #include <sys/errno.h> 46 #include <sys/unistd.h> 47 #include <sys/mode.h> 48 #include <sys/atomic.h> 49 #include <vm/pvn.h> 50 #include "fs/fs_subr.h" 51 #include <sys/zfs_dir.h> 52 #include <sys/zfs_acl.h> 53 #include <sys/zfs_ioctl.h> 54 #include <sys/zfs_rlock.h> 55 #include <sys/zfs_fuid.h> 56 #include <sys/dnode.h> 57 #include <sys/fs/zfs.h> 58 #include <sys/kidmap.h> 59 #endif /* _KERNEL */ 60 61 #include <sys/dmu.h> 62 #include <sys/dmu_objset.h> 63 #include <sys/refcount.h> 64 #include <sys/stat.h> 65 #include <sys/zap.h> 66 #include <sys/zfs_znode.h> 67 #include <sys/sa.h> 68 #include <sys/zfs_sa.h> 69 #include <sys/zfs_stat.h> 70 71 #include "zfs_prop.h" 72 #include "zfs_comutil.h" 73 74 /* 75 * Define ZNODE_STATS to turn on statistic gathering. By default, it is only 76 * turned on when DEBUG is also defined. 77 */ 78 #ifdef DEBUG 79 #define ZNODE_STATS 80 #endif /* DEBUG */ 81 82 #ifdef ZNODE_STATS 83 #define ZNODE_STAT_ADD(stat) ((stat)++) 84 #else 85 #define ZNODE_STAT_ADD(stat) /* nothing */ 86 #endif /* ZNODE_STATS */ 87 88 /* 89 * Functions needed for userland (ie: libzpool) are not put under 90 * #ifdef_KERNEL; the rest of the functions have dependencies 91 * (such as VFS logic) that will not compile easily in userland. 92 */ 93 #ifdef _KERNEL 94 /* 95 * Needed to close a small window in zfs_znode_move() that allows the zfsvfs to 96 * be freed before it can be safely accessed. 97 */ 98 krwlock_t zfsvfs_lock; 99 100 static kmem_cache_t *znode_cache = NULL; 101 102 /*ARGSUSED*/ 103 static void 104 znode_evict_error(dmu_buf_t *dbuf, void *user_ptr) 105 { 106 /* 107 * We should never drop all dbuf refs without first clearing 108 * the eviction callback. 109 */ 110 panic("evicting znode %p\n", user_ptr); 111 } 112 113 /* 114 * This callback is invoked when acquiring a RL_WRITER or RL_APPEND lock on 115 * z_rangelock. It will modify the offset and length of the lock to reflect 116 * znode-specific information, and convert RL_APPEND to RL_WRITER. This is 117 * called with the rangelock_t's rl_lock held, which avoids races. 118 */ 119 static void 120 zfs_rangelock_cb(locked_range_t *new, void *arg) 121 { 122 znode_t *zp = arg; 123 124 /* 125 * If in append mode, convert to writer and lock starting at the 126 * current end of file. 127 */ 128 if (new->lr_type == RL_APPEND) { 129 new->lr_offset = zp->z_size; 130 new->lr_type = RL_WRITER; 131 } 132 133 /* 134 * If we need to grow the block size then lock the whole file range. 135 */ 136 uint64_t end_size = MAX(zp->z_size, new->lr_offset + new->lr_length); 137 if (end_size > zp->z_blksz && (!ISP2(zp->z_blksz) || 138 zp->z_blksz < zp->z_zfsvfs->z_max_blksz)) { 139 new->lr_offset = 0; 140 new->lr_length = UINT64_MAX; 141 } 142 } 143 144 /*ARGSUSED*/ 145 static int 146 zfs_znode_cache_constructor(void *buf, void *arg, int kmflags) 147 { 148 znode_t *zp = buf; 149 150 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 151 152 zp->z_vnode = vn_alloc(kmflags); 153 if (zp->z_vnode == NULL) { 154 return (-1); 155 } 156 ZTOV(zp)->v_data = zp; 157 158 list_link_init(&zp->z_link_node); 159 160 mutex_init(&zp->z_lock, NULL, MUTEX_DEFAULT, NULL); 161 rw_init(&zp->z_parent_lock, NULL, RW_DEFAULT, NULL); 162 rw_init(&zp->z_name_lock, NULL, RW_DEFAULT, NULL); 163 mutex_init(&zp->z_acl_lock, NULL, MUTEX_DEFAULT, NULL); 164 165 rangelock_init(&zp->z_rangelock, zfs_rangelock_cb, zp); 166 167 zp->z_dirlocks = NULL; 168 zp->z_acl_cached = NULL; 169 zp->z_moved = 0; 170 return (0); 171 } 172 173 /*ARGSUSED*/ 174 static void 175 zfs_znode_cache_destructor(void *buf, void *arg) 176 { 177 znode_t *zp = buf; 178 179 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 180 ASSERT(ZTOV(zp)->v_data == zp); 181 vn_free(ZTOV(zp)); 182 ASSERT(!list_link_active(&zp->z_link_node)); 183 mutex_destroy(&zp->z_lock); 184 rw_destroy(&zp->z_parent_lock); 185 rw_destroy(&zp->z_name_lock); 186 mutex_destroy(&zp->z_acl_lock); 187 rangelock_fini(&zp->z_rangelock); 188 189 ASSERT(zp->z_dirlocks == NULL); 190 ASSERT(zp->z_acl_cached == NULL); 191 } 192 193 #ifdef ZNODE_STATS 194 static struct { 195 uint64_t zms_zfsvfs_invalid; 196 uint64_t zms_zfsvfs_recheck1; 197 uint64_t zms_zfsvfs_unmounted; 198 uint64_t zms_zfsvfs_recheck2; 199 uint64_t zms_obj_held; 200 uint64_t zms_vnode_locked; 201 uint64_t zms_not_only_dnlc; 202 } znode_move_stats; 203 #endif /* ZNODE_STATS */ 204 205 static void 206 zfs_znode_move_impl(znode_t *ozp, znode_t *nzp) 207 { 208 vnode_t *vp; 209 210 /* Copy fields. */ 211 nzp->z_zfsvfs = ozp->z_zfsvfs; 212 213 /* Swap vnodes. */ 214 vp = nzp->z_vnode; 215 nzp->z_vnode = ozp->z_vnode; 216 ozp->z_vnode = vp; /* let destructor free the overwritten vnode */ 217 ZTOV(ozp)->v_data = ozp; 218 ZTOV(nzp)->v_data = nzp; 219 220 nzp->z_id = ozp->z_id; 221 ASSERT(ozp->z_dirlocks == NULL); /* znode not in use */ 222 nzp->z_unlinked = ozp->z_unlinked; 223 nzp->z_atime_dirty = ozp->z_atime_dirty; 224 nzp->z_zn_prefetch = ozp->z_zn_prefetch; 225 nzp->z_blksz = ozp->z_blksz; 226 nzp->z_seq = ozp->z_seq; 227 nzp->z_mapcnt = ozp->z_mapcnt; 228 nzp->z_gen = ozp->z_gen; 229 nzp->z_sync_cnt = ozp->z_sync_cnt; 230 nzp->z_is_sa = ozp->z_is_sa; 231 nzp->z_sa_hdl = ozp->z_sa_hdl; 232 bcopy(ozp->z_atime, nzp->z_atime, sizeof (uint64_t) * 2); 233 nzp->z_links = ozp->z_links; 234 nzp->z_size = ozp->z_size; 235 nzp->z_pflags = ozp->z_pflags; 236 nzp->z_uid = ozp->z_uid; 237 nzp->z_gid = ozp->z_gid; 238 nzp->z_mode = ozp->z_mode; 239 240 /* 241 * Since this is just an idle znode and kmem is already dealing with 242 * memory pressure, release any cached ACL. 243 */ 244 if (ozp->z_acl_cached) { 245 zfs_acl_free(ozp->z_acl_cached); 246 ozp->z_acl_cached = NULL; 247 } 248 249 sa_set_userp(nzp->z_sa_hdl, nzp); 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_sa_hdl = NULL; 258 POINTER_INVALIDATE(&ozp->z_zfsvfs); 259 260 /* 261 * Mark the znode. 262 */ 263 nzp->z_moved = 1; 264 ozp->z_moved = (uint8_t)-1; 265 } 266 267 /*ARGSUSED*/ 268 static kmem_cbrc_t 269 zfs_znode_move(void *buf, void *newbuf, size_t size, void *arg) 270 { 271 znode_t *ozp = buf, *nzp = newbuf; 272 zfsvfs_t *zfsvfs; 273 vnode_t *vp; 274 275 /* 276 * The znode is on the file system's list of known znodes if the vfs 277 * pointer is valid. We set the low bit of the vfs pointer when freeing 278 * the znode to invalidate it, and the memory patterns written by kmem 279 * (baddcafe and deadbeef) set at least one of the two low bits. A newly 280 * created znode sets the vfs pointer last of all to indicate that the 281 * znode is known and in a valid state to be moved by this function. 282 */ 283 zfsvfs = ozp->z_zfsvfs; 284 if (!POINTER_IS_VALID(zfsvfs)) { 285 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_invalid); 286 return (KMEM_CBRC_DONT_KNOW); 287 } 288 289 /* 290 * Close a small window in which it's possible that the filesystem could 291 * be unmounted and freed, and zfsvfs, though valid in the previous 292 * statement, could point to unrelated memory by the time we try to 293 * prevent the filesystem from being unmounted. 294 */ 295 rw_enter(&zfsvfs_lock, RW_WRITER); 296 if (zfsvfs != ozp->z_zfsvfs) { 297 rw_exit(&zfsvfs_lock); 298 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck1); 299 return (KMEM_CBRC_DONT_KNOW); 300 } 301 302 /* 303 * If the znode is still valid, then so is the file system. We know that 304 * no valid file system can be freed while we hold zfsvfs_lock, so we 305 * can safely ensure that the filesystem is not and will not be 306 * unmounted. The next statement is equivalent to ZFS_ENTER(). 307 */ 308 rrm_enter(&zfsvfs->z_teardown_lock, RW_READER, FTAG); 309 if (zfsvfs->z_unmounted) { 310 ZFS_EXIT(zfsvfs); 311 rw_exit(&zfsvfs_lock); 312 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_unmounted); 313 return (KMEM_CBRC_DONT_KNOW); 314 } 315 rw_exit(&zfsvfs_lock); 316 317 mutex_enter(&zfsvfs->z_znodes_lock); 318 /* 319 * Recheck the vfs pointer in case the znode was removed just before 320 * acquiring the lock. 321 */ 322 if (zfsvfs != ozp->z_zfsvfs) { 323 mutex_exit(&zfsvfs->z_znodes_lock); 324 ZFS_EXIT(zfsvfs); 325 ZNODE_STAT_ADD(znode_move_stats.zms_zfsvfs_recheck2); 326 return (KMEM_CBRC_DONT_KNOW); 327 } 328 329 /* 330 * At this point we know that as long as we hold z_znodes_lock, the 331 * znode cannot be freed and fields within the znode can be safely 332 * accessed. Now, prevent a race with zfs_zget(). 333 */ 334 if (ZFS_OBJ_HOLD_TRYENTER(zfsvfs, ozp->z_id) == 0) { 335 mutex_exit(&zfsvfs->z_znodes_lock); 336 ZFS_EXIT(zfsvfs); 337 ZNODE_STAT_ADD(znode_move_stats.zms_obj_held); 338 return (KMEM_CBRC_LATER); 339 } 340 341 vp = ZTOV(ozp); 342 if (mutex_tryenter(&vp->v_lock) == 0) { 343 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 344 mutex_exit(&zfsvfs->z_znodes_lock); 345 ZFS_EXIT(zfsvfs); 346 ZNODE_STAT_ADD(znode_move_stats.zms_vnode_locked); 347 return (KMEM_CBRC_LATER); 348 } 349 350 /* Only move znodes that are referenced _only_ by the DNLC. */ 351 if (vp->v_count != 1 || !vn_in_dnlc(vp)) { 352 mutex_exit(&vp->v_lock); 353 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 354 mutex_exit(&zfsvfs->z_znodes_lock); 355 ZFS_EXIT(zfsvfs); 356 ZNODE_STAT_ADD(znode_move_stats.zms_not_only_dnlc); 357 return (KMEM_CBRC_LATER); 358 } 359 360 /* 361 * The znode is known and in a valid state to move. We're holding the 362 * locks needed to execute the critical section. 363 */ 364 zfs_znode_move_impl(ozp, nzp); 365 mutex_exit(&vp->v_lock); 366 ZFS_OBJ_HOLD_EXIT(zfsvfs, ozp->z_id); 367 368 list_link_replace(&ozp->z_link_node, &nzp->z_link_node); 369 mutex_exit(&zfsvfs->z_znodes_lock); 370 ZFS_EXIT(zfsvfs); 371 372 return (KMEM_CBRC_YES); 373 } 374 375 void 376 zfs_znode_init(void) 377 { 378 /* 379 * Initialize zcache 380 */ 381 rw_init(&zfsvfs_lock, NULL, RW_DEFAULT, NULL); 382 ASSERT(znode_cache == NULL); 383 znode_cache = kmem_cache_create("zfs_znode_cache", 384 sizeof (znode_t), 0, zfs_znode_cache_constructor, 385 zfs_znode_cache_destructor, NULL, NULL, NULL, 0); 386 kmem_cache_set_move(znode_cache, zfs_znode_move); 387 } 388 389 void 390 zfs_znode_fini(void) 391 { 392 /* 393 * Cleanup vfs & vnode ops 394 */ 395 zfs_remove_op_tables(); 396 397 /* 398 * Cleanup zcache 399 */ 400 if (znode_cache) 401 kmem_cache_destroy(znode_cache); 402 znode_cache = NULL; 403 rw_destroy(&zfsvfs_lock); 404 } 405 406 struct vnodeops *zfs_dvnodeops; 407 struct vnodeops *zfs_fvnodeops; 408 struct vnodeops *zfs_symvnodeops; 409 struct vnodeops *zfs_xdvnodeops; 410 struct vnodeops *zfs_evnodeops; 411 struct vnodeops *zfs_sharevnodeops; 412 413 void 414 zfs_remove_op_tables() 415 { 416 /* 417 * Remove vfs ops 418 */ 419 ASSERT(zfsfstype); 420 (void) vfs_freevfsops_by_type(zfsfstype); 421 zfsfstype = 0; 422 423 /* 424 * Remove vnode ops 425 */ 426 if (zfs_dvnodeops) 427 vn_freevnodeops(zfs_dvnodeops); 428 if (zfs_fvnodeops) 429 vn_freevnodeops(zfs_fvnodeops); 430 if (zfs_symvnodeops) 431 vn_freevnodeops(zfs_symvnodeops); 432 if (zfs_xdvnodeops) 433 vn_freevnodeops(zfs_xdvnodeops); 434 if (zfs_evnodeops) 435 vn_freevnodeops(zfs_evnodeops); 436 if (zfs_sharevnodeops) 437 vn_freevnodeops(zfs_sharevnodeops); 438 439 zfs_dvnodeops = NULL; 440 zfs_fvnodeops = NULL; 441 zfs_symvnodeops = NULL; 442 zfs_xdvnodeops = NULL; 443 zfs_evnodeops = NULL; 444 zfs_sharevnodeops = NULL; 445 } 446 447 extern const fs_operation_def_t zfs_dvnodeops_template[]; 448 extern const fs_operation_def_t zfs_fvnodeops_template[]; 449 extern const fs_operation_def_t zfs_xdvnodeops_template[]; 450 extern const fs_operation_def_t zfs_symvnodeops_template[]; 451 extern const fs_operation_def_t zfs_evnodeops_template[]; 452 extern const fs_operation_def_t zfs_sharevnodeops_template[]; 453 454 int 455 zfs_create_op_tables() 456 { 457 int error; 458 459 /* 460 * zfs_dvnodeops can be set if mod_remove() calls mod_installfs() 461 * due to a failure to remove the the 2nd modlinkage (zfs_modldrv). 462 * In this case we just return as the ops vectors are already set up. 463 */ 464 if (zfs_dvnodeops) 465 return (0); 466 467 error = vn_make_ops(MNTTYPE_ZFS, zfs_dvnodeops_template, 468 &zfs_dvnodeops); 469 if (error) 470 return (error); 471 472 error = vn_make_ops(MNTTYPE_ZFS, zfs_fvnodeops_template, 473 &zfs_fvnodeops); 474 if (error) 475 return (error); 476 477 error = vn_make_ops(MNTTYPE_ZFS, zfs_symvnodeops_template, 478 &zfs_symvnodeops); 479 if (error) 480 return (error); 481 482 error = vn_make_ops(MNTTYPE_ZFS, zfs_xdvnodeops_template, 483 &zfs_xdvnodeops); 484 if (error) 485 return (error); 486 487 error = vn_make_ops(MNTTYPE_ZFS, zfs_evnodeops_template, 488 &zfs_evnodeops); 489 if (error) 490 return (error); 491 492 error = vn_make_ops(MNTTYPE_ZFS, zfs_sharevnodeops_template, 493 &zfs_sharevnodeops); 494 495 return (error); 496 } 497 498 int 499 zfs_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; 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 ASSERT(!POINTER_IS_VALID(sharezp->z_zfsvfs)); 516 sharezp->z_moved = 0; 517 sharezp->z_unlinked = 0; 518 sharezp->z_atime_dirty = 0; 519 sharezp->z_zfsvfs = zfsvfs; 520 sharezp->z_is_sa = zfsvfs->z_use_sa; 521 522 vp = ZTOV(sharezp); 523 vn_reinit(vp); 524 vp->v_type = VDIR; 525 526 VERIFY(0 == zfs_acl_ids_create(sharezp, IS_ROOT_NODE, &vattr, 527 kcred, NULL, &acl_ids)); 528 zfs_mknode(sharezp, &vattr, tx, kcred, IS_ROOT_NODE, &zp, &acl_ids); 529 ASSERT3P(zp, ==, sharezp); 530 ASSERT(!vn_in_dnlc(ZTOV(sharezp))); /* not valid to move */ 531 POINTER_INVALIDATE(&sharezp->z_zfsvfs); 532 error = zap_add(zfsvfs->z_os, MASTER_NODE_OBJ, 533 ZFS_SHARES_DIR, 8, 1, &sharezp->z_id, tx); 534 zfsvfs->z_shares_dir = sharezp->z_id; 535 536 zfs_acl_ids_free(&acl_ids); 537 ZTOV(sharezp)->v_count = 0; 538 sa_handle_destroy(sharezp->z_sa_hdl); 539 kmem_cache_free(znode_cache, sharezp); 540 541 return (error); 542 } 543 544 /* 545 * define a couple of values we need available 546 * for both 64 and 32 bit environments. 547 */ 548 #ifndef NBITSMINOR64 549 #define NBITSMINOR64 32 550 #endif 551 #ifndef MAXMAJ64 552 #define MAXMAJ64 0xffffffffUL 553 #endif 554 #ifndef MAXMIN64 555 #define MAXMIN64 0xffffffffUL 556 #endif 557 558 /* 559 * Create special expldev for ZFS private use. 560 * Can't use standard expldev since it doesn't do 561 * what we want. The standard expldev() takes a 562 * dev32_t in LP64 and expands it to a long dev_t. 563 * We need an interface that takes a dev32_t in ILP32 564 * and expands it to a long dev_t. 565 */ 566 static uint64_t 567 zfs_expldev(dev_t dev) 568 { 569 #ifndef _LP64 570 major_t major = (major_t)dev >> NBITSMINOR32 & MAXMAJ32; 571 return (((uint64_t)major << NBITSMINOR64) | 572 ((minor_t)dev & MAXMIN32)); 573 #else 574 return (dev); 575 #endif 576 } 577 578 /* 579 * Special cmpldev for ZFS private use. 580 * Can't use standard cmpldev since it takes 581 * a long dev_t and compresses it to dev32_t in 582 * LP64. We need to do a compaction of a long dev_t 583 * to a dev32_t in ILP32. 584 */ 585 dev_t 586 zfs_cmpldev(uint64_t dev) 587 { 588 #ifndef _LP64 589 minor_t minor = (minor_t)dev & MAXMIN64; 590 major_t major = (major_t)(dev >> NBITSMINOR64) & MAXMAJ64; 591 592 if (major > MAXMAJ32 || minor > MAXMIN32) 593 return (NODEV32); 594 595 return (((dev32_t)major << NBITSMINOR32) | minor); 596 #else 597 return (dev); 598 #endif 599 } 600 601 static void 602 zfs_znode_sa_init(zfsvfs_t *zfsvfs, znode_t *zp, 603 dmu_buf_t *db, dmu_object_type_t obj_type, sa_handle_t *sa_hdl) 604 { 605 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs) || (zfsvfs == zp->z_zfsvfs)); 606 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zfsvfs, zp->z_id))); 607 608 mutex_enter(&zp->z_lock); 609 610 ASSERT(zp->z_sa_hdl == NULL); 611 ASSERT(zp->z_acl_cached == NULL); 612 if (sa_hdl == NULL) { 613 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, zp, 614 SA_HDL_SHARED, &zp->z_sa_hdl)); 615 } else { 616 zp->z_sa_hdl = sa_hdl; 617 sa_set_userp(sa_hdl, zp); 618 } 619 620 zp->z_is_sa = (obj_type == DMU_OT_SA) ? B_TRUE : B_FALSE; 621 622 /* 623 * Slap on VROOT if we are the root znode 624 */ 625 if (zp->z_id == zfsvfs->z_root) 626 ZTOV(zp)->v_flag |= VROOT; 627 628 mutex_exit(&zp->z_lock); 629 vn_exists(ZTOV(zp)); 630 } 631 632 void 633 zfs_znode_dmu_fini(znode_t *zp) 634 { 635 ASSERT(MUTEX_HELD(ZFS_OBJ_MUTEX(zp->z_zfsvfs, zp->z_id)) || 636 zp->z_unlinked || 637 RW_WRITE_HELD(&zp->z_zfsvfs->z_teardown_inactive_lock)); 638 639 sa_handle_destroy(zp->z_sa_hdl); 640 zp->z_sa_hdl = NULL; 641 } 642 643 /* 644 * Construct a new znode/vnode and intialize. 645 * 646 * This does not do a call to dmu_set_user() that is 647 * up to the caller to do, in case you don't want to 648 * return the znode 649 */ 650 static znode_t * 651 zfs_znode_alloc(zfsvfs_t *zfsvfs, dmu_buf_t *db, int blksz, 652 dmu_object_type_t obj_type, sa_handle_t *hdl) 653 { 654 znode_t *zp; 655 vnode_t *vp; 656 uint64_t mode; 657 uint64_t parent; 658 sa_bulk_attr_t bulk[9]; 659 int count = 0; 660 661 zp = kmem_cache_alloc(znode_cache, KM_SLEEP); 662 663 ASSERT(zp->z_dirlocks == NULL); 664 ASSERT(!POINTER_IS_VALID(zp->z_zfsvfs)); 665 zp->z_moved = 0; 666 667 /* 668 * Defer setting z_zfsvfs until the znode is ready to be a candidate for 669 * the zfs_znode_move() callback. 670 */ 671 zp->z_sa_hdl = NULL; 672 zp->z_unlinked = 0; 673 zp->z_atime_dirty = 0; 674 zp->z_mapcnt = 0; 675 zp->z_id = db->db_object; 676 zp->z_blksz = blksz; 677 zp->z_seq = 0x7A4653; 678 zp->z_sync_cnt = 0; 679 680 vp = ZTOV(zp); 681 vn_reinit(vp); 682 683 zfs_znode_sa_init(zfsvfs, zp, db, obj_type, hdl); 684 685 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 686 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, &zp->z_gen, 8); 687 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 688 &zp->z_size, 8); 689 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 690 &zp->z_links, 8); 691 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 692 &zp->z_pflags, 8); 693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, &parent, 8); 694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 695 &zp->z_atime, 16); 696 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 697 &zp->z_uid, 8); 698 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 699 &zp->z_gid, 8); 700 701 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count) != 0 || zp->z_gen == 0) { 702 if (hdl == NULL) 703 sa_handle_destroy(zp->z_sa_hdl); 704 kmem_cache_free(znode_cache, zp); 705 return (NULL); 706 } 707 708 zp->z_mode = mode; 709 vp->v_vfsp = zfsvfs->z_parent->z_vfs; 710 711 vp->v_type = IFTOVT((mode_t)mode); 712 713 switch (vp->v_type) { 714 case VDIR: 715 if (zp->z_pflags & ZFS_XATTR) { 716 vn_setops(vp, zfs_xdvnodeops); 717 vp->v_flag |= V_XATTRDIR; 718 } else { 719 vn_setops(vp, zfs_dvnodeops); 720 } 721 zp->z_zn_prefetch = B_TRUE; /* z_prefetch default is enabled */ 722 break; 723 case VBLK: 724 case VCHR: 725 { 726 uint64_t rdev; 727 VERIFY(sa_lookup(zp->z_sa_hdl, SA_ZPL_RDEV(zfsvfs), 728 &rdev, sizeof (rdev)) == 0); 729 730 vp->v_rdev = zfs_cmpldev(rdev); 731 } 732 /*FALLTHROUGH*/ 733 case VFIFO: 734 case VSOCK: 735 case VDOOR: 736 vn_setops(vp, zfs_fvnodeops); 737 break; 738 case VREG: 739 vp->v_flag |= VMODSORT; 740 if (parent == zfsvfs->z_shares_dir) { 741 ASSERT(zp->z_uid == 0 && zp->z_gid == 0); 742 vn_setops(vp, zfs_sharevnodeops); 743 } else { 744 vn_setops(vp, zfs_fvnodeops); 745 } 746 break; 747 case VLNK: 748 vn_setops(vp, zfs_symvnodeops); 749 break; 750 default: 751 vn_setops(vp, zfs_evnodeops); 752 break; 753 } 754 755 mutex_enter(&zfsvfs->z_znodes_lock); 756 list_insert_tail(&zfsvfs->z_all_znodes, zp); 757 membar_producer(); 758 /* 759 * Everything else must be valid before assigning z_zfsvfs makes the 760 * znode eligible for zfs_znode_move(). 761 */ 762 zp->z_zfsvfs = zfsvfs; 763 mutex_exit(&zfsvfs->z_znodes_lock); 764 765 VFS_HOLD(zfsvfs->z_vfs); 766 return (zp); 767 } 768 769 static uint64_t empty_xattr; 770 static uint64_t pad[4]; 771 static zfs_acl_phys_t acl_phys; 772 /* 773 * Create a new DMU object to hold a zfs znode. 774 * 775 * IN: dzp - parent directory for new znode 776 * vap - file attributes for new znode 777 * tx - dmu transaction id for zap operations 778 * cr - credentials of caller 779 * flag - flags: 780 * IS_ROOT_NODE - new object will be root 781 * IS_XATTR - new object is an attribute 782 * bonuslen - length of bonus buffer 783 * setaclp - File/Dir initial ACL 784 * fuidp - Tracks fuid allocation. 785 * 786 * OUT: zpp - allocated znode 787 * 788 */ 789 void 790 zfs_mknode(znode_t *dzp, vattr_t *vap, dmu_tx_t *tx, cred_t *cr, 791 uint_t flag, znode_t **zpp, zfs_acl_ids_t *acl_ids) 792 { 793 uint64_t crtime[2], atime[2], mtime[2], ctime[2]; 794 uint64_t mode, size, links, parent, pflags; 795 uint64_t dzp_pflags = 0; 796 uint64_t rdev = 0; 797 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 798 dmu_buf_t *db; 799 timestruc_t now; 800 uint64_t gen, obj; 801 int bonuslen; 802 sa_handle_t *sa_hdl; 803 dmu_object_type_t obj_type; 804 sa_bulk_attr_t sa_attrs[ZPL_END]; 805 int cnt = 0; 806 zfs_acl_locator_cb_t locate = { 0 }; 807 808 ASSERT(vap && (vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE)); 809 810 if (zfsvfs->z_replay) { 811 obj = vap->va_nodeid; 812 now = vap->va_ctime; /* see zfs_replay_create() */ 813 gen = vap->va_nblocks; /* ditto */ 814 } else { 815 obj = 0; 816 gethrestime(&now); 817 gen = dmu_tx_get_txg(tx); 818 } 819 820 obj_type = zfsvfs->z_use_sa ? DMU_OT_SA : DMU_OT_ZNODE; 821 bonuslen = (obj_type == DMU_OT_SA) ? 822 DN_MAX_BONUSLEN : ZFS_OLD_ZNODE_PHYS_SIZE; 823 824 /* 825 * Create a new DMU object. 826 */ 827 /* 828 * There's currently no mechanism for pre-reading the blocks that will 829 * be needed to allocate a new object, so we accept the small chance 830 * that there will be an i/o error and we will fail one of the 831 * assertions below. 832 */ 833 if (vap->va_type == VDIR) { 834 if (zfsvfs->z_replay) { 835 VERIFY0(zap_create_claim_norm(zfsvfs->z_os, obj, 836 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 837 obj_type, bonuslen, tx)); 838 } else { 839 obj = zap_create_norm(zfsvfs->z_os, 840 zfsvfs->z_norm, DMU_OT_DIRECTORY_CONTENTS, 841 obj_type, bonuslen, tx); 842 } 843 } else { 844 if (zfsvfs->z_replay) { 845 VERIFY0(dmu_object_claim(zfsvfs->z_os, obj, 846 DMU_OT_PLAIN_FILE_CONTENTS, 0, 847 obj_type, bonuslen, tx)); 848 } else { 849 obj = dmu_object_alloc(zfsvfs->z_os, 850 DMU_OT_PLAIN_FILE_CONTENTS, 0, 851 obj_type, bonuslen, tx); 852 } 853 } 854 855 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 856 VERIFY(0 == sa_buf_hold(zfsvfs->z_os, obj, NULL, &db)); 857 858 /* 859 * If this is the root, fix up the half-initialized parent pointer 860 * to reference the just-allocated physical data area. 861 */ 862 if (flag & IS_ROOT_NODE) { 863 dzp->z_id = obj; 864 } else { 865 dzp_pflags = dzp->z_pflags; 866 } 867 868 /* 869 * If parent is an xattr, so am I. 870 */ 871 if (dzp_pflags & ZFS_XATTR) { 872 flag |= IS_XATTR; 873 } 874 875 if (zfsvfs->z_use_fuids) 876 pflags = ZFS_ARCHIVE | ZFS_AV_MODIFIED; 877 else 878 pflags = 0; 879 880 if (vap->va_type == VDIR) { 881 size = 2; /* contents ("." and "..") */ 882 links = (flag & (IS_ROOT_NODE | IS_XATTR)) ? 2 : 1; 883 } else { 884 size = links = 0; 885 } 886 887 if (vap->va_type == VBLK || vap->va_type == VCHR) { 888 rdev = zfs_expldev(vap->va_rdev); 889 } 890 891 parent = dzp->z_id; 892 mode = acl_ids->z_mode; 893 if (flag & IS_XATTR) 894 pflags |= ZFS_XATTR; 895 896 /* 897 * No execs denied will be deterimed when zfs_mode_compute() is called. 898 */ 899 pflags |= acl_ids->z_aclp->z_hints & 900 (ZFS_ACL_TRIVIAL|ZFS_INHERIT_ACE|ZFS_ACL_AUTO_INHERIT| 901 ZFS_ACL_DEFAULTED|ZFS_ACL_PROTECTED); 902 903 ZFS_TIME_ENCODE(&now, crtime); 904 ZFS_TIME_ENCODE(&now, ctime); 905 906 if (vap->va_mask & AT_ATIME) { 907 ZFS_TIME_ENCODE(&vap->va_atime, atime); 908 } else { 909 ZFS_TIME_ENCODE(&now, atime); 910 } 911 912 if (vap->va_mask & AT_MTIME) { 913 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 914 } else { 915 ZFS_TIME_ENCODE(&now, mtime); 916 } 917 918 /* Now add in all of the "SA" attributes */ 919 VERIFY(0 == sa_handle_get_from_db(zfsvfs->z_os, db, NULL, SA_HDL_SHARED, 920 &sa_hdl)); 921 922 /* 923 * Setup the array of attributes to be replaced/set on the new file 924 * 925 * order for DMU_OT_ZNODE is critical since it needs to be constructed 926 * in the old znode_phys_t format. Don't change this ordering 927 */ 928 929 if (obj_type == DMU_OT_ZNODE) { 930 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 931 NULL, &atime, 16); 932 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 933 NULL, &mtime, 16); 934 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 935 NULL, &ctime, 16); 936 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 937 NULL, &crtime, 16); 938 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 939 NULL, &gen, 8); 940 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 941 NULL, &mode, 8); 942 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 943 NULL, &size, 8); 944 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 945 NULL, &parent, 8); 946 } else { 947 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MODE(zfsvfs), 948 NULL, &mode, 8); 949 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_SIZE(zfsvfs), 950 NULL, &size, 8); 951 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GEN(zfsvfs), 952 NULL, &gen, 8); 953 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 954 &acl_ids->z_fuid, 8); 955 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 956 &acl_ids->z_fgid, 8); 957 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PARENT(zfsvfs), 958 NULL, &parent, 8); 959 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 960 NULL, &pflags, 8); 961 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ATIME(zfsvfs), 962 NULL, &atime, 16); 963 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_MTIME(zfsvfs), 964 NULL, &mtime, 16); 965 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CTIME(zfsvfs), 966 NULL, &ctime, 16); 967 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_CRTIME(zfsvfs), 968 NULL, &crtime, 16); 969 } 970 971 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_LINKS(zfsvfs), NULL, &links, 8); 972 973 if (obj_type == DMU_OT_ZNODE) { 974 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_XATTR(zfsvfs), NULL, 975 &empty_xattr, 8); 976 } 977 if (obj_type == DMU_OT_ZNODE || 978 (vap->va_type == VBLK || vap->va_type == VCHR)) { 979 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_RDEV(zfsvfs), 980 NULL, &rdev, 8); 981 982 } 983 if (obj_type == DMU_OT_ZNODE) { 984 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_FLAGS(zfsvfs), 985 NULL, &pflags, 8); 986 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_UID(zfsvfs), NULL, 987 &acl_ids->z_fuid, 8); 988 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_GID(zfsvfs), NULL, 989 &acl_ids->z_fgid, 8); 990 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_PAD(zfsvfs), NULL, pad, 991 sizeof (uint64_t) * 4); 992 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_ZNODE_ACL(zfsvfs), NULL, 993 &acl_phys, sizeof (zfs_acl_phys_t)); 994 } else if (acl_ids->z_aclp->z_version >= ZFS_ACL_VERSION_FUID) { 995 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_COUNT(zfsvfs), NULL, 996 &acl_ids->z_aclp->z_acl_count, 8); 997 locate.cb_aclp = acl_ids->z_aclp; 998 SA_ADD_BULK_ATTR(sa_attrs, cnt, SA_ZPL_DACL_ACES(zfsvfs), 999 zfs_acl_data_locator, &locate, 1000 acl_ids->z_aclp->z_acl_bytes); 1001 mode = zfs_mode_compute(mode, acl_ids->z_aclp, &pflags, 1002 acl_ids->z_fuid, acl_ids->z_fgid); 1003 } 1004 1005 VERIFY(sa_replace_all_by_template(sa_hdl, sa_attrs, cnt, tx) == 0); 1006 1007 if (!(flag & IS_ROOT_NODE)) { 1008 *zpp = zfs_znode_alloc(zfsvfs, db, 0, obj_type, sa_hdl); 1009 ASSERT(*zpp != NULL); 1010 } else { 1011 /* 1012 * If we are creating the root node, the "parent" we 1013 * passed in is the znode for the root. 1014 */ 1015 *zpp = dzp; 1016 1017 (*zpp)->z_sa_hdl = sa_hdl; 1018 } 1019 1020 (*zpp)->z_pflags = pflags; 1021 (*zpp)->z_mode = mode; 1022 1023 if (vap->va_mask & AT_XVATTR) 1024 zfs_xvattr_set(*zpp, (xvattr_t *)vap, tx); 1025 1026 if (obj_type == DMU_OT_ZNODE || 1027 acl_ids->z_aclp->z_version < ZFS_ACL_VERSION_FUID) { 1028 VERIFY0(zfs_aclset_common(*zpp, acl_ids->z_aclp, cr, tx)); 1029 } 1030 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1031 } 1032 1033 /* 1034 * Update in-core attributes. It is assumed the caller will be doing an 1035 * sa_bulk_update to push the changes out. 1036 */ 1037 void 1038 zfs_xvattr_set(znode_t *zp, xvattr_t *xvap, dmu_tx_t *tx) 1039 { 1040 xoptattr_t *xoap; 1041 1042 xoap = xva_getxoptattr(xvap); 1043 ASSERT(xoap); 1044 1045 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 1046 uint64_t times[2]; 1047 ZFS_TIME_ENCODE(&xoap->xoa_createtime, times); 1048 (void) sa_update(zp->z_sa_hdl, SA_ZPL_CRTIME(zp->z_zfsvfs), 1049 ×, sizeof (times), tx); 1050 XVA_SET_RTN(xvap, XAT_CREATETIME); 1051 } 1052 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 1053 ZFS_ATTR_SET(zp, ZFS_READONLY, xoap->xoa_readonly, 1054 zp->z_pflags, tx); 1055 XVA_SET_RTN(xvap, XAT_READONLY); 1056 } 1057 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 1058 ZFS_ATTR_SET(zp, ZFS_HIDDEN, xoap->xoa_hidden, 1059 zp->z_pflags, tx); 1060 XVA_SET_RTN(xvap, XAT_HIDDEN); 1061 } 1062 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 1063 ZFS_ATTR_SET(zp, ZFS_SYSTEM, xoap->xoa_system, 1064 zp->z_pflags, tx); 1065 XVA_SET_RTN(xvap, XAT_SYSTEM); 1066 } 1067 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 1068 ZFS_ATTR_SET(zp, ZFS_ARCHIVE, xoap->xoa_archive, 1069 zp->z_pflags, tx); 1070 XVA_SET_RTN(xvap, XAT_ARCHIVE); 1071 } 1072 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 1073 ZFS_ATTR_SET(zp, ZFS_IMMUTABLE, xoap->xoa_immutable, 1074 zp->z_pflags, tx); 1075 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 1076 } 1077 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 1078 ZFS_ATTR_SET(zp, ZFS_NOUNLINK, xoap->xoa_nounlink, 1079 zp->z_pflags, tx); 1080 XVA_SET_RTN(xvap, XAT_NOUNLINK); 1081 } 1082 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 1083 ZFS_ATTR_SET(zp, ZFS_APPENDONLY, xoap->xoa_appendonly, 1084 zp->z_pflags, tx); 1085 XVA_SET_RTN(xvap, XAT_APPENDONLY); 1086 } 1087 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 1088 ZFS_ATTR_SET(zp, ZFS_NODUMP, xoap->xoa_nodump, 1089 zp->z_pflags, tx); 1090 XVA_SET_RTN(xvap, XAT_NODUMP); 1091 } 1092 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 1093 ZFS_ATTR_SET(zp, ZFS_OPAQUE, xoap->xoa_opaque, 1094 zp->z_pflags, tx); 1095 XVA_SET_RTN(xvap, XAT_OPAQUE); 1096 } 1097 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 1098 ZFS_ATTR_SET(zp, ZFS_AV_QUARANTINED, 1099 xoap->xoa_av_quarantined, zp->z_pflags, tx); 1100 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 1101 } 1102 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 1103 ZFS_ATTR_SET(zp, ZFS_AV_MODIFIED, xoap->xoa_av_modified, 1104 zp->z_pflags, tx); 1105 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 1106 } 1107 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) { 1108 zfs_sa_set_scanstamp(zp, xvap, tx); 1109 XVA_SET_RTN(xvap, XAT_AV_SCANSTAMP); 1110 } 1111 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 1112 ZFS_ATTR_SET(zp, ZFS_REPARSE, xoap->xoa_reparse, 1113 zp->z_pflags, tx); 1114 XVA_SET_RTN(xvap, XAT_REPARSE); 1115 } 1116 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 1117 ZFS_ATTR_SET(zp, ZFS_OFFLINE, xoap->xoa_offline, 1118 zp->z_pflags, tx); 1119 XVA_SET_RTN(xvap, XAT_OFFLINE); 1120 } 1121 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 1122 ZFS_ATTR_SET(zp, ZFS_SPARSE, xoap->xoa_sparse, 1123 zp->z_pflags, tx); 1124 XVA_SET_RTN(xvap, XAT_SPARSE); 1125 } 1126 } 1127 1128 int 1129 zfs_zget(zfsvfs_t *zfsvfs, uint64_t obj_num, znode_t **zpp) 1130 { 1131 dmu_object_info_t doi; 1132 dmu_buf_t *db; 1133 znode_t *zp; 1134 int err; 1135 sa_handle_t *hdl; 1136 1137 *zpp = NULL; 1138 1139 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1140 1141 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1142 if (err) { 1143 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1144 return (err); 1145 } 1146 1147 dmu_object_info_from_db(db, &doi); 1148 if (doi.doi_bonus_type != DMU_OT_SA && 1149 (doi.doi_bonus_type != DMU_OT_ZNODE || 1150 (doi.doi_bonus_type == DMU_OT_ZNODE && 1151 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1152 sa_buf_rele(db, NULL); 1153 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1154 return (SET_ERROR(EINVAL)); 1155 } 1156 1157 hdl = dmu_buf_get_user(db); 1158 if (hdl != NULL) { 1159 zp = sa_get_userdata(hdl); 1160 1161 1162 /* 1163 * Since "SA" does immediate eviction we 1164 * should never find a sa handle that doesn't 1165 * know about the znode. 1166 */ 1167 1168 ASSERT3P(zp, !=, NULL); 1169 1170 mutex_enter(&zp->z_lock); 1171 ASSERT3U(zp->z_id, ==, obj_num); 1172 if (zp->z_unlinked) { 1173 err = SET_ERROR(ENOENT); 1174 } else { 1175 VN_HOLD(ZTOV(zp)); 1176 *zpp = zp; 1177 err = 0; 1178 } 1179 mutex_exit(&zp->z_lock); 1180 sa_buf_rele(db, NULL); 1181 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1182 return (err); 1183 } 1184 1185 /* 1186 * Not found create new znode/vnode 1187 * but only if file exists. 1188 * 1189 * There is a small window where zfs_vget() could 1190 * find this object while a file create is still in 1191 * progress. This is checked for in zfs_znode_alloc() 1192 * 1193 * if zfs_znode_alloc() fails it will drop the hold on the 1194 * bonus buffer. 1195 */ 1196 zp = zfs_znode_alloc(zfsvfs, db, doi.doi_data_block_size, 1197 doi.doi_bonus_type, NULL); 1198 if (zp == NULL) { 1199 err = SET_ERROR(ENOENT); 1200 } else { 1201 *zpp = zp; 1202 } 1203 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1204 return (err); 1205 } 1206 1207 int 1208 zfs_rezget(znode_t *zp) 1209 { 1210 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1211 dmu_object_info_t doi; 1212 dmu_buf_t *db; 1213 uint64_t obj_num = zp->z_id; 1214 uint64_t mode; 1215 sa_bulk_attr_t bulk[8]; 1216 int err; 1217 int count = 0; 1218 uint64_t gen; 1219 1220 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj_num); 1221 1222 mutex_enter(&zp->z_acl_lock); 1223 if (zp->z_acl_cached) { 1224 zfs_acl_free(zp->z_acl_cached); 1225 zp->z_acl_cached = NULL; 1226 } 1227 1228 mutex_exit(&zp->z_acl_lock); 1229 ASSERT(zp->z_sa_hdl == NULL); 1230 err = sa_buf_hold(zfsvfs->z_os, obj_num, NULL, &db); 1231 if (err) { 1232 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1233 return (err); 1234 } 1235 1236 dmu_object_info_from_db(db, &doi); 1237 if (doi.doi_bonus_type != DMU_OT_SA && 1238 (doi.doi_bonus_type != DMU_OT_ZNODE || 1239 (doi.doi_bonus_type == DMU_OT_ZNODE && 1240 doi.doi_bonus_size < sizeof (znode_phys_t)))) { 1241 sa_buf_rele(db, NULL); 1242 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1243 return (SET_ERROR(EINVAL)); 1244 } 1245 1246 zfs_znode_sa_init(zfsvfs, zp, db, doi.doi_bonus_type, NULL); 1247 1248 /* reload cached values */ 1249 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GEN(zfsvfs), NULL, 1250 &gen, sizeof (gen)); 1251 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 1252 &zp->z_size, sizeof (zp->z_size)); 1253 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 1254 &zp->z_links, sizeof (zp->z_links)); 1255 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 1256 &zp->z_pflags, sizeof (zp->z_pflags)); 1257 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 1258 &zp->z_atime, sizeof (zp->z_atime)); 1259 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1260 &zp->z_uid, sizeof (zp->z_uid)); 1261 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1262 &zp->z_gid, sizeof (zp->z_gid)); 1263 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 1264 &mode, sizeof (mode)); 1265 1266 if (sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) { 1267 zfs_znode_dmu_fini(zp); 1268 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1269 return (SET_ERROR(EIO)); 1270 } 1271 1272 zp->z_mode = mode; 1273 1274 if (gen != zp->z_gen) { 1275 zfs_znode_dmu_fini(zp); 1276 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1277 return (SET_ERROR(EIO)); 1278 } 1279 1280 zp->z_blksz = doi.doi_data_block_size; 1281 1282 /* 1283 * If the file has zero links, then it has been unlinked on the send 1284 * side and it must be in the received unlinked set. 1285 * We call zfs_znode_dmu_fini() now to prevent any accesses to the 1286 * stale data and to prevent automatical removal of the file in 1287 * zfs_zinactive(). The file will be removed either when it is removed 1288 * on the send side and the next incremental stream is received or 1289 * when the unlinked set gets processed. 1290 */ 1291 zp->z_unlinked = (zp->z_links == 0); 1292 if (zp->z_unlinked) 1293 zfs_znode_dmu_fini(zp); 1294 1295 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj_num); 1296 1297 return (0); 1298 } 1299 1300 void 1301 zfs_znode_delete(znode_t *zp, dmu_tx_t *tx) 1302 { 1303 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1304 objset_t *os = zfsvfs->z_os; 1305 uint64_t obj = zp->z_id; 1306 uint64_t acl_obj = zfs_external_acl(zp); 1307 1308 ZFS_OBJ_HOLD_ENTER(zfsvfs, obj); 1309 if (acl_obj) { 1310 VERIFY(!zp->z_is_sa); 1311 VERIFY(0 == dmu_object_free(os, acl_obj, tx)); 1312 } 1313 VERIFY(0 == dmu_object_free(os, obj, tx)); 1314 zfs_znode_dmu_fini(zp); 1315 ZFS_OBJ_HOLD_EXIT(zfsvfs, obj); 1316 zfs_znode_free(zp); 1317 } 1318 1319 void 1320 zfs_zinactive(znode_t *zp) 1321 { 1322 vnode_t *vp = ZTOV(zp); 1323 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1324 uint64_t z_id = zp->z_id; 1325 1326 ASSERT(zp->z_sa_hdl); 1327 1328 /* 1329 * Don't allow a zfs_zget() while were trying to release this znode 1330 */ 1331 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 1332 1333 mutex_enter(&zp->z_lock); 1334 mutex_enter(&vp->v_lock); 1335 VN_RELE_LOCKED(vp); 1336 if (vp->v_count > 0 || vn_has_cached_data(vp)) { 1337 /* 1338 * If the hold count is greater than zero, somebody has 1339 * obtained a new reference on this znode while we were 1340 * processing it here, so we are done. If we still have 1341 * mapped pages then we are also done, since we don't 1342 * want to inactivate the znode until the pages get pushed. 1343 * 1344 * XXX - if vn_has_cached_data(vp) is true, but count == 0, 1345 * this seems like it would leave the znode hanging with 1346 * no chance to go inactive... 1347 */ 1348 mutex_exit(&vp->v_lock); 1349 mutex_exit(&zp->z_lock); 1350 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1351 return; 1352 } 1353 mutex_exit(&vp->v_lock); 1354 1355 /* 1356 * If this was the last reference to a file with no links, remove 1357 * the file from the file system unless the file system is mounted 1358 * read-only. That can happen, for example, if the file system was 1359 * originally read-write, the file was opened, then unlinked and 1360 * the file system was made read-only before the file was finally 1361 * closed. The file will remain in the unlinked set. 1362 */ 1363 if (zp->z_unlinked) { 1364 ASSERT(!zfsvfs->z_issnap); 1365 if ((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) { 1366 mutex_exit(&zp->z_lock); 1367 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1368 zfs_rmnode(zp); 1369 return; 1370 } 1371 } 1372 1373 mutex_exit(&zp->z_lock); 1374 zfs_znode_dmu_fini(zp); 1375 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 1376 zfs_znode_free(zp); 1377 } 1378 1379 void 1380 zfs_znode_free(znode_t *zp) 1381 { 1382 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1383 1384 vn_invalid(ZTOV(zp)); 1385 1386 ASSERT(ZTOV(zp)->v_count == 0); 1387 1388 mutex_enter(&zfsvfs->z_znodes_lock); 1389 POINTER_INVALIDATE(&zp->z_zfsvfs); 1390 list_remove(&zfsvfs->z_all_znodes, zp); 1391 mutex_exit(&zfsvfs->z_znodes_lock); 1392 1393 if (zp->z_acl_cached) { 1394 zfs_acl_free(zp->z_acl_cached); 1395 zp->z_acl_cached = NULL; 1396 } 1397 1398 kmem_cache_free(znode_cache, zp); 1399 1400 VFS_RELE(zfsvfs->z_vfs); 1401 } 1402 1403 void 1404 zfs_tstamp_update_setup(znode_t *zp, uint_t flag, uint64_t mtime[2], 1405 uint64_t ctime[2], boolean_t have_tx) 1406 { 1407 timestruc_t now; 1408 1409 gethrestime(&now); 1410 1411 if (have_tx) { /* will sa_bulk_update happen really soon? */ 1412 zp->z_atime_dirty = 0; 1413 zp->z_seq++; 1414 } else { 1415 zp->z_atime_dirty = 1; 1416 } 1417 1418 if (flag & AT_ATIME) { 1419 ZFS_TIME_ENCODE(&now, zp->z_atime); 1420 } 1421 1422 if (flag & AT_MTIME) { 1423 ZFS_TIME_ENCODE(&now, mtime); 1424 if (zp->z_zfsvfs->z_use_fuids) { 1425 zp->z_pflags |= (ZFS_ARCHIVE | 1426 ZFS_AV_MODIFIED); 1427 } 1428 } 1429 1430 if (flag & AT_CTIME) { 1431 ZFS_TIME_ENCODE(&now, ctime); 1432 if (zp->z_zfsvfs->z_use_fuids) 1433 zp->z_pflags |= ZFS_ARCHIVE; 1434 } 1435 } 1436 1437 /* 1438 * Grow the block size for a file. 1439 * 1440 * IN: zp - znode of file to free data in. 1441 * size - requested block size 1442 * tx - open transaction. 1443 * 1444 * NOTE: this function assumes that the znode is write locked. 1445 */ 1446 void 1447 zfs_grow_blocksize(znode_t *zp, uint64_t size, dmu_tx_t *tx) 1448 { 1449 int error; 1450 u_longlong_t dummy; 1451 1452 if (size <= zp->z_blksz) 1453 return; 1454 /* 1455 * If the file size is already greater than the current blocksize, 1456 * we will not grow. If there is more than one block in a file, 1457 * the blocksize cannot change. 1458 */ 1459 if (zp->z_blksz && zp->z_size > zp->z_blksz) 1460 return; 1461 1462 error = dmu_object_set_blocksize(zp->z_zfsvfs->z_os, zp->z_id, 1463 size, 0, tx); 1464 1465 if (error == ENOTSUP) 1466 return; 1467 ASSERT0(error); 1468 1469 /* What blocksize did we actually get? */ 1470 dmu_object_size_from_db(sa_get_db(zp->z_sa_hdl), &zp->z_blksz, &dummy); 1471 } 1472 1473 /* 1474 * This is a dummy interface used when pvn_vplist_dirty() should *not* 1475 * be calling back into the fs for a putpage(). E.g.: when truncating 1476 * a file, the pages being "thrown away* don't need to be written out. 1477 */ 1478 /* ARGSUSED */ 1479 static int 1480 zfs_no_putpage(vnode_t *vp, page_t *pp, u_offset_t *offp, size_t *lenp, 1481 int flags, cred_t *cr) 1482 { 1483 ASSERT(0); 1484 return (0); 1485 } 1486 1487 /* 1488 * Increase the file length 1489 * 1490 * IN: zp - znode of file to free data in. 1491 * end - new end-of-file 1492 * 1493 * RETURN: 0 on success, error code on failure 1494 */ 1495 static int 1496 zfs_extend(znode_t *zp, uint64_t end) 1497 { 1498 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1499 dmu_tx_t *tx; 1500 locked_range_t *lr; 1501 uint64_t newblksz; 1502 int error; 1503 1504 /* 1505 * We will change zp_size, lock the whole file. 1506 */ 1507 lr = rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER); 1508 1509 /* 1510 * Nothing to do if file already at desired length. 1511 */ 1512 if (end <= zp->z_size) { 1513 rangelock_exit(lr); 1514 return (0); 1515 } 1516 tx = dmu_tx_create(zfsvfs->z_os); 1517 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1518 zfs_sa_upgrade_txholds(tx, zp); 1519 if (end > zp->z_blksz && 1520 (!ISP2(zp->z_blksz) || zp->z_blksz < zfsvfs->z_max_blksz)) { 1521 /* 1522 * We are growing the file past the current block size. 1523 */ 1524 if (zp->z_blksz > zp->z_zfsvfs->z_max_blksz) { 1525 /* 1526 * File's blocksize is already larger than the 1527 * "recordsize" property. Only let it grow to 1528 * the next power of 2. 1529 */ 1530 ASSERT(!ISP2(zp->z_blksz)); 1531 newblksz = MIN(end, 1 << highbit64(zp->z_blksz)); 1532 } else { 1533 newblksz = MIN(end, zp->z_zfsvfs->z_max_blksz); 1534 } 1535 dmu_tx_hold_write(tx, zp->z_id, 0, newblksz); 1536 } else { 1537 newblksz = 0; 1538 } 1539 1540 error = dmu_tx_assign(tx, TXG_WAIT); 1541 if (error) { 1542 dmu_tx_abort(tx); 1543 rangelock_exit(lr); 1544 return (error); 1545 } 1546 1547 if (newblksz) 1548 zfs_grow_blocksize(zp, newblksz, tx); 1549 1550 zp->z_size = end; 1551 1552 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zp->z_zfsvfs), 1553 &zp->z_size, sizeof (zp->z_size), tx)); 1554 1555 rangelock_exit(lr); 1556 1557 dmu_tx_commit(tx); 1558 1559 return (0); 1560 } 1561 1562 /* 1563 * Free space in a file. 1564 * 1565 * IN: zp - znode of file to free data in. 1566 * off - start of section to free. 1567 * len - length of section to free. 1568 * 1569 * RETURN: 0 on success, error code on failure 1570 */ 1571 static int 1572 zfs_free_range(znode_t *zp, uint64_t off, uint64_t len) 1573 { 1574 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1575 locked_range_t *lr; 1576 int error; 1577 1578 /* 1579 * Lock the range being freed. 1580 */ 1581 lr = rangelock_enter(&zp->z_rangelock, off, len, RL_WRITER); 1582 1583 /* 1584 * Nothing to do if file already at desired length. 1585 */ 1586 if (off >= zp->z_size) { 1587 rangelock_exit(lr); 1588 return (0); 1589 } 1590 1591 if (off + len > zp->z_size) 1592 len = zp->z_size - off; 1593 1594 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, off, len); 1595 1596 rangelock_exit(lr); 1597 1598 return (error); 1599 } 1600 1601 /* 1602 * Truncate a file 1603 * 1604 * IN: zp - znode of file to free data in. 1605 * end - new end-of-file. 1606 * 1607 * RETURN: 0 on success, error code on failure 1608 */ 1609 static int 1610 zfs_trunc(znode_t *zp, uint64_t end) 1611 { 1612 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1613 vnode_t *vp = ZTOV(zp); 1614 dmu_tx_t *tx; 1615 locked_range_t *lr; 1616 int error; 1617 sa_bulk_attr_t bulk[2]; 1618 int count = 0; 1619 1620 /* 1621 * We will change zp_size, lock the whole file. 1622 */ 1623 lr = rangelock_enter(&zp->z_rangelock, 0, UINT64_MAX, RL_WRITER); 1624 1625 /* 1626 * Nothing to do if file already at desired length. 1627 */ 1628 if (end >= zp->z_size) { 1629 rangelock_exit(lr); 1630 return (0); 1631 } 1632 1633 error = dmu_free_long_range(zfsvfs->z_os, zp->z_id, end, 1634 DMU_OBJECT_END); 1635 if (error) { 1636 rangelock_exit(lr); 1637 return (error); 1638 } 1639 tx = dmu_tx_create(zfsvfs->z_os); 1640 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1641 zfs_sa_upgrade_txholds(tx, zp); 1642 dmu_tx_mark_netfree(tx); 1643 error = dmu_tx_assign(tx, TXG_WAIT); 1644 if (error) { 1645 dmu_tx_abort(tx); 1646 rangelock_exit(lr); 1647 return (error); 1648 } 1649 1650 zp->z_size = end; 1651 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1652 NULL, &zp->z_size, sizeof (zp->z_size)); 1653 1654 if (end == 0) { 1655 zp->z_pflags &= ~ZFS_SPARSE; 1656 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1657 NULL, &zp->z_pflags, 8); 1658 } 1659 VERIFY(sa_bulk_update(zp->z_sa_hdl, bulk, count, tx) == 0); 1660 1661 dmu_tx_commit(tx); 1662 1663 /* 1664 * Clear any mapped pages in the truncated region. This has to 1665 * happen outside of the transaction to avoid the possibility of 1666 * a deadlock with someone trying to push a page that we are 1667 * about to invalidate. 1668 */ 1669 if (vn_has_cached_data(vp)) { 1670 page_t *pp; 1671 uint64_t start = end & PAGEMASK; 1672 int poff = end & PAGEOFFSET; 1673 1674 if (poff != 0 && (pp = page_lookup(vp, start, SE_SHARED))) { 1675 /* 1676 * We need to zero a partial page. 1677 */ 1678 pagezero(pp, poff, PAGESIZE - poff); 1679 start += PAGESIZE; 1680 page_unlock(pp); 1681 } 1682 error = pvn_vplist_dirty(vp, start, zfs_no_putpage, 1683 B_INVAL | B_TRUNC, NULL); 1684 ASSERT(error == 0); 1685 } 1686 1687 rangelock_exit(lr); 1688 1689 return (0); 1690 } 1691 1692 /* 1693 * Free space in a file 1694 * 1695 * IN: zp - znode of file to free data in. 1696 * off - start of range 1697 * len - end of range (0 => EOF) 1698 * flag - current file open mode flags. 1699 * log - TRUE if this action should be logged 1700 * 1701 * RETURN: 0 on success, error code on failure 1702 */ 1703 int 1704 zfs_freesp(znode_t *zp, uint64_t off, uint64_t len, int flag, boolean_t log) 1705 { 1706 vnode_t *vp = ZTOV(zp); 1707 dmu_tx_t *tx; 1708 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1709 zilog_t *zilog = zfsvfs->z_log; 1710 uint64_t mode; 1711 uint64_t mtime[2], ctime[2]; 1712 sa_bulk_attr_t bulk[3]; 1713 int count = 0; 1714 int error; 1715 1716 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), &mode, 1717 sizeof (mode))) != 0) 1718 return (error); 1719 1720 if (off > zp->z_size) { 1721 error = zfs_extend(zp, off+len); 1722 if (error == 0 && log) 1723 goto log; 1724 else 1725 return (error); 1726 } 1727 1728 /* 1729 * Check for any locks in the region to be freed. 1730 */ 1731 1732 if (MANDLOCK(vp, (mode_t)mode)) { 1733 uint64_t length = (len ? len : zp->z_size - off); 1734 if (error = chklock(vp, FWRITE, off, length, flag, NULL)) 1735 return (error); 1736 } 1737 1738 if (len == 0) { 1739 error = zfs_trunc(zp, off); 1740 } else { 1741 if ((error = zfs_free_range(zp, off, len)) == 0 && 1742 off + len > zp->z_size) 1743 error = zfs_extend(zp, off+len); 1744 } 1745 if (error || !log) 1746 return (error); 1747 log: 1748 tx = dmu_tx_create(zfsvfs->z_os); 1749 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1750 zfs_sa_upgrade_txholds(tx, zp); 1751 error = dmu_tx_assign(tx, TXG_WAIT); 1752 if (error) { 1753 dmu_tx_abort(tx); 1754 return (error); 1755 } 1756 1757 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, mtime, 16); 1758 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, ctime, 16); 1759 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1760 NULL, &zp->z_pflags, 8); 1761 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, B_TRUE); 1762 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1763 ASSERT(error == 0); 1764 1765 zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 1766 1767 dmu_tx_commit(tx); 1768 return (0); 1769 } 1770 1771 void 1772 zfs_create_fs(objset_t *os, cred_t *cr, nvlist_t *zplprops, dmu_tx_t *tx) 1773 { 1774 uint64_t moid, obj, sa_obj, version; 1775 uint64_t sense = ZFS_CASE_SENSITIVE; 1776 uint64_t norm = 0; 1777 nvpair_t *elem; 1778 int error; 1779 int i; 1780 znode_t *rootzp = NULL; 1781 zfsvfs_t *zfsvfs; 1782 vnode_t *vp; 1783 vattr_t vattr; 1784 znode_t *zp; 1785 zfs_acl_ids_t acl_ids; 1786 1787 /* 1788 * First attempt to create master node. 1789 */ 1790 /* 1791 * In an empty objset, there are no blocks to read and thus 1792 * there can be no i/o errors (which we assert below). 1793 */ 1794 moid = MASTER_NODE_OBJ; 1795 error = zap_create_claim(os, moid, DMU_OT_MASTER_NODE, 1796 DMU_OT_NONE, 0, tx); 1797 ASSERT(error == 0); 1798 1799 /* 1800 * Set starting attributes. 1801 */ 1802 version = zfs_zpl_version_map(spa_version(dmu_objset_spa(os))); 1803 elem = NULL; 1804 while ((elem = nvlist_next_nvpair(zplprops, elem)) != NULL) { 1805 /* For the moment we expect all zpl props to be uint64_ts */ 1806 uint64_t val; 1807 char *name; 1808 1809 ASSERT(nvpair_type(elem) == DATA_TYPE_UINT64); 1810 VERIFY(nvpair_value_uint64(elem, &val) == 0); 1811 name = nvpair_name(elem); 1812 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_VERSION)) == 0) { 1813 if (val < version) 1814 version = val; 1815 } else { 1816 error = zap_update(os, moid, name, 8, 1, &val, tx); 1817 } 1818 ASSERT(error == 0); 1819 if (strcmp(name, zfs_prop_to_name(ZFS_PROP_NORMALIZE)) == 0) 1820 norm = val; 1821 else if (strcmp(name, zfs_prop_to_name(ZFS_PROP_CASE)) == 0) 1822 sense = val; 1823 } 1824 ASSERT(version != 0); 1825 error = zap_update(os, moid, ZPL_VERSION_STR, 8, 1, &version, tx); 1826 1827 /* 1828 * Create zap object used for SA attribute registration 1829 */ 1830 1831 if (version >= ZPL_VERSION_SA) { 1832 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, 1833 DMU_OT_NONE, 0, tx); 1834 error = zap_add(os, moid, ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); 1835 ASSERT(error == 0); 1836 } else { 1837 sa_obj = 0; 1838 } 1839 /* 1840 * Create a delete queue. 1841 */ 1842 obj = zap_create(os, DMU_OT_UNLINKED_SET, DMU_OT_NONE, 0, tx); 1843 1844 error = zap_add(os, moid, ZFS_UNLINKED_SET, 8, 1, &obj, tx); 1845 ASSERT(error == 0); 1846 1847 /* 1848 * Create root znode. Create minimal znode/vnode/zfsvfs 1849 * to allow zfs_mknode to work. 1850 */ 1851 vattr.va_mask = AT_MODE|AT_UID|AT_GID|AT_TYPE; 1852 vattr.va_type = VDIR; 1853 vattr.va_mode = S_IFDIR|0755; 1854 vattr.va_uid = crgetuid(cr); 1855 vattr.va_gid = crgetgid(cr); 1856 1857 rootzp = kmem_cache_alloc(znode_cache, KM_SLEEP); 1858 ASSERT(!POINTER_IS_VALID(rootzp->z_zfsvfs)); 1859 rootzp->z_moved = 0; 1860 rootzp->z_unlinked = 0; 1861 rootzp->z_atime_dirty = 0; 1862 rootzp->z_is_sa = USE_SA(version, os); 1863 1864 vp = ZTOV(rootzp); 1865 vn_reinit(vp); 1866 vp->v_type = VDIR; 1867 1868 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); 1869 zfsvfs->z_os = os; 1870 zfsvfs->z_parent = zfsvfs; 1871 zfsvfs->z_version = version; 1872 zfsvfs->z_use_fuids = USE_FUIDS(version, os); 1873 zfsvfs->z_use_sa = USE_SA(version, os); 1874 zfsvfs->z_norm = norm; 1875 1876 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, 1877 &zfsvfs->z_attr_table); 1878 1879 ASSERT(error == 0); 1880 1881 /* 1882 * Fold case on file systems that are always or sometimes case 1883 * insensitive. 1884 */ 1885 if (sense == ZFS_CASE_INSENSITIVE || sense == ZFS_CASE_MIXED) 1886 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; 1887 1888 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); 1889 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), 1890 offsetof(znode_t, z_link_node)); 1891 1892 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1893 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); 1894 1895 rootzp->z_zfsvfs = zfsvfs; 1896 VERIFY(0 == zfs_acl_ids_create(rootzp, IS_ROOT_NODE, &vattr, 1897 cr, NULL, &acl_ids)); 1898 zfs_mknode(rootzp, &vattr, tx, cr, IS_ROOT_NODE, &zp, &acl_ids); 1899 ASSERT3P(zp, ==, rootzp); 1900 ASSERT(!vn_in_dnlc(ZTOV(rootzp))); /* not valid to move */ 1901 error = zap_add(os, moid, ZFS_ROOT_OBJ, 8, 1, &rootzp->z_id, tx); 1902 ASSERT(error == 0); 1903 zfs_acl_ids_free(&acl_ids); 1904 POINTER_INVALIDATE(&rootzp->z_zfsvfs); 1905 1906 ZTOV(rootzp)->v_count = 0; 1907 sa_handle_destroy(rootzp->z_sa_hdl); 1908 kmem_cache_free(znode_cache, rootzp); 1909 1910 /* 1911 * Create shares directory 1912 */ 1913 1914 error = zfs_create_share_dir(zfsvfs, tx); 1915 1916 ASSERT(error == 0); 1917 1918 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) 1919 mutex_destroy(&zfsvfs->z_hold_mtx[i]); 1920 kmem_free(zfsvfs, sizeof (zfsvfs_t)); 1921 } 1922 1923 #endif /* _KERNEL */ 1924 1925 static int 1926 zfs_sa_setup(objset_t *osp, sa_attr_type_t **sa_table) 1927 { 1928 uint64_t sa_obj = 0; 1929 int error; 1930 1931 error = zap_lookup(osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, &sa_obj); 1932 if (error != 0 && error != ENOENT) 1933 return (error); 1934 1935 error = sa_setup(osp, sa_obj, zfs_attr_table, ZPL_END, sa_table); 1936 return (error); 1937 } 1938 1939 static int 1940 zfs_grab_sa_handle(objset_t *osp, uint64_t obj, sa_handle_t **hdlp, 1941 dmu_buf_t **db, void *tag) 1942 { 1943 dmu_object_info_t doi; 1944 int error; 1945 1946 if ((error = sa_buf_hold(osp, obj, tag, db)) != 0) 1947 return (error); 1948 1949 dmu_object_info_from_db(*db, &doi); 1950 if ((doi.doi_bonus_type != DMU_OT_SA && 1951 doi.doi_bonus_type != DMU_OT_ZNODE) || 1952 doi.doi_bonus_type == DMU_OT_ZNODE && 1953 doi.doi_bonus_size < sizeof (znode_phys_t)) { 1954 sa_buf_rele(*db, tag); 1955 return (SET_ERROR(ENOTSUP)); 1956 } 1957 1958 error = sa_handle_get(osp, obj, NULL, SA_HDL_PRIVATE, hdlp); 1959 if (error != 0) { 1960 sa_buf_rele(*db, tag); 1961 return (error); 1962 } 1963 1964 return (0); 1965 } 1966 1967 void 1968 zfs_release_sa_handle(sa_handle_t *hdl, dmu_buf_t *db, void *tag) 1969 { 1970 sa_handle_destroy(hdl); 1971 sa_buf_rele(db, tag); 1972 } 1973 1974 /* 1975 * Given an object number, return its parent object number and whether 1976 * or not the object is an extended attribute directory. 1977 */ 1978 static int 1979 zfs_obj_to_pobj(objset_t *osp, sa_handle_t *hdl, sa_attr_type_t *sa_table, 1980 uint64_t *pobjp, int *is_xattrdir) 1981 { 1982 uint64_t parent; 1983 uint64_t pflags; 1984 uint64_t mode; 1985 uint64_t parent_mode; 1986 sa_bulk_attr_t bulk[3]; 1987 sa_handle_t *sa_hdl; 1988 dmu_buf_t *sa_db; 1989 int count = 0; 1990 int error; 1991 1992 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_PARENT], NULL, 1993 &parent, sizeof (parent)); 1994 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_FLAGS], NULL, 1995 &pflags, sizeof (pflags)); 1996 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 1997 &mode, sizeof (mode)); 1998 1999 if ((error = sa_bulk_lookup(hdl, bulk, count)) != 0) 2000 return (error); 2001 2002 /* 2003 * When a link is removed its parent pointer is not changed and will 2004 * be invalid. There are two cases where a link is removed but the 2005 * file stays around, when it goes to the delete queue and when there 2006 * are additional links. 2007 */ 2008 error = zfs_grab_sa_handle(osp, parent, &sa_hdl, &sa_db, FTAG); 2009 if (error != 0) 2010 return (error); 2011 2012 error = sa_lookup(sa_hdl, ZPL_MODE, &parent_mode, sizeof (parent_mode)); 2013 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2014 if (error != 0) 2015 return (error); 2016 2017 *is_xattrdir = ((pflags & ZFS_XATTR) != 0) && S_ISDIR(mode); 2018 2019 /* 2020 * Extended attributes can be applied to files, directories, etc. 2021 * Otherwise the parent must be a directory. 2022 */ 2023 if (!*is_xattrdir && !S_ISDIR(parent_mode)) 2024 return (SET_ERROR(EINVAL)); 2025 2026 *pobjp = parent; 2027 2028 return (0); 2029 } 2030 2031 /* 2032 * Given an object number, return some zpl level statistics 2033 */ 2034 static int 2035 zfs_obj_to_stats_impl(sa_handle_t *hdl, sa_attr_type_t *sa_table, 2036 zfs_stat_t *sb) 2037 { 2038 sa_bulk_attr_t bulk[4]; 2039 int count = 0; 2040 2041 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_MODE], NULL, 2042 &sb->zs_mode, sizeof (sb->zs_mode)); 2043 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_GEN], NULL, 2044 &sb->zs_gen, sizeof (sb->zs_gen)); 2045 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_LINKS], NULL, 2046 &sb->zs_links, sizeof (sb->zs_links)); 2047 SA_ADD_BULK_ATTR(bulk, count, sa_table[ZPL_CTIME], NULL, 2048 &sb->zs_ctime, sizeof (sb->zs_ctime)); 2049 2050 return (sa_bulk_lookup(hdl, bulk, count)); 2051 } 2052 2053 static int 2054 zfs_obj_to_path_impl(objset_t *osp, uint64_t obj, sa_handle_t *hdl, 2055 sa_attr_type_t *sa_table, char *buf, int len) 2056 { 2057 sa_handle_t *sa_hdl; 2058 sa_handle_t *prevhdl = NULL; 2059 dmu_buf_t *prevdb = NULL; 2060 dmu_buf_t *sa_db = NULL; 2061 char *path = buf + len - 1; 2062 int error; 2063 2064 *path = '\0'; 2065 sa_hdl = hdl; 2066 2067 uint64_t deleteq_obj; 2068 VERIFY0(zap_lookup(osp, MASTER_NODE_OBJ, 2069 ZFS_UNLINKED_SET, sizeof (uint64_t), 1, &deleteq_obj)); 2070 error = zap_lookup_int(osp, deleteq_obj, obj); 2071 if (error == 0) { 2072 return (ESTALE); 2073 } else if (error != ENOENT) { 2074 return (error); 2075 } 2076 error = 0; 2077 2078 for (;;) { 2079 uint64_t pobj; 2080 char component[MAXNAMELEN + 2]; 2081 size_t complen; 2082 int is_xattrdir; 2083 2084 if (prevdb) 2085 zfs_release_sa_handle(prevhdl, prevdb, FTAG); 2086 2087 if ((error = zfs_obj_to_pobj(osp, sa_hdl, sa_table, &pobj, 2088 &is_xattrdir)) != 0) 2089 break; 2090 2091 if (pobj == obj) { 2092 if (path[0] != '/') 2093 *--path = '/'; 2094 break; 2095 } 2096 2097 component[0] = '/'; 2098 if (is_xattrdir) { 2099 (void) sprintf(component + 1, "<xattrdir>"); 2100 } else { 2101 error = zap_value_search(osp, pobj, obj, 2102 ZFS_DIRENT_OBJ(-1ULL), component + 1); 2103 if (error != 0) 2104 break; 2105 } 2106 2107 complen = strlen(component); 2108 path -= complen; 2109 ASSERT(path >= buf); 2110 bcopy(component, path, complen); 2111 obj = pobj; 2112 2113 if (sa_hdl != hdl) { 2114 prevhdl = sa_hdl; 2115 prevdb = sa_db; 2116 } 2117 error = zfs_grab_sa_handle(osp, obj, &sa_hdl, &sa_db, FTAG); 2118 if (error != 0) { 2119 sa_hdl = prevhdl; 2120 sa_db = prevdb; 2121 break; 2122 } 2123 } 2124 2125 if (sa_hdl != NULL && sa_hdl != hdl) { 2126 ASSERT(sa_db != NULL); 2127 zfs_release_sa_handle(sa_hdl, sa_db, FTAG); 2128 } 2129 2130 if (error == 0) 2131 (void) memmove(buf, path, buf + len - path); 2132 2133 return (error); 2134 } 2135 2136 int 2137 zfs_obj_to_path(objset_t *osp, uint64_t obj, char *buf, int len) 2138 { 2139 sa_attr_type_t *sa_table; 2140 sa_handle_t *hdl; 2141 dmu_buf_t *db; 2142 int error; 2143 2144 error = zfs_sa_setup(osp, &sa_table); 2145 if (error != 0) 2146 return (error); 2147 2148 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2149 if (error != 0) 2150 return (error); 2151 2152 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2153 2154 zfs_release_sa_handle(hdl, db, FTAG); 2155 return (error); 2156 } 2157 2158 int 2159 zfs_obj_to_stats(objset_t *osp, uint64_t obj, zfs_stat_t *sb, 2160 char *buf, int len) 2161 { 2162 char *path = buf + len - 1; 2163 sa_attr_type_t *sa_table; 2164 sa_handle_t *hdl; 2165 dmu_buf_t *db; 2166 int error; 2167 2168 *path = '\0'; 2169 2170 error = zfs_sa_setup(osp, &sa_table); 2171 if (error != 0) 2172 return (error); 2173 2174 error = zfs_grab_sa_handle(osp, obj, &hdl, &db, FTAG); 2175 if (error != 0) 2176 return (error); 2177 2178 error = zfs_obj_to_stats_impl(hdl, sa_table, sb); 2179 if (error != 0) { 2180 zfs_release_sa_handle(hdl, db, FTAG); 2181 return (error); 2182 } 2183 2184 error = zfs_obj_to_path_impl(osp, obj, hdl, sa_table, buf, len); 2185 2186 zfs_release_sa_handle(hdl, db, FTAG); 2187 return (error); 2188 } 2189