1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */ 2 3 /*- 4 * SPDX-License-Identifier: BSD-2-Clause 5 * 6 * Copyright (c) 2005 The NetBSD Foundation, Inc. 7 * All rights reserved. 8 * 9 * This code is derived from software contributed to The NetBSD Foundation 10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 11 * 2005 program. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 32 * POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 /* 36 * Efficient memory file system supporting functions. 37 */ 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/dirent.h> 42 #include <sys/fnv_hash.h> 43 #include <sys/lock.h> 44 #include <sys/limits.h> 45 #include <sys/mount.h> 46 #include <sys/namei.h> 47 #include <sys/priv.h> 48 #include <sys/proc.h> 49 #include <sys/random.h> 50 #include <sys/refcount.h> 51 #include <sys/rwlock.h> 52 #include <sys/smr.h> 53 #include <sys/stat.h> 54 #include <sys/sysctl.h> 55 #include <sys/user.h> 56 #include <sys/vnode.h> 57 #include <sys/vmmeter.h> 58 59 #include <vm/vm.h> 60 #include <vm/vm_param.h> 61 #include <vm/vm_object.h> 62 #include <vm/vm_page.h> 63 #include <vm/vm_pageout.h> 64 #include <vm/vm_pager.h> 65 #include <vm/vm_extern.h> 66 #include <vm/swap_pager.h> 67 #include <vm/uma.h> 68 69 #include <fs/tmpfs/tmpfs.h> 70 #include <fs/tmpfs/tmpfs_fifoops.h> 71 #include <fs/tmpfs/tmpfs_vnops.h> 72 73 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 74 "tmpfs file system"); 75 76 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED; 77 static long tmpfs_pages_avail_init; 78 static int tmpfs_mem_percent = TMPFS_MEM_PERCENT; 79 static void tmpfs_set_reserve_from_percent(void); 80 81 MALLOC_DEFINE(M_TMPFSDIR, "tmpfs dir", "tmpfs dirent structure"); 82 static uma_zone_t tmpfs_node_pool; 83 VFS_SMR_DECLARE; 84 85 int tmpfs_pager_type = -1; 86 87 static vm_object_t 88 tmpfs_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, 89 vm_ooffset_t offset, struct ucred *cred) 90 { 91 vm_object_t object; 92 93 MPASS(handle == NULL); 94 MPASS(offset == 0); 95 object = vm_object_allocate_dyn(tmpfs_pager_type, size, 96 OBJ_COLORED | OBJ_SWAP); 97 if (!swap_pager_init_object(object, NULL, NULL, size, 0)) { 98 vm_object_deallocate(object); 99 object = NULL; 100 } 101 return (object); 102 } 103 104 /* 105 * Make sure tmpfs vnodes with writable mappings can be found on the lazy list. 106 * 107 * This allows for periodic mtime updates while only scanning vnodes which are 108 * plausibly dirty, see tmpfs_update_mtime_lazy. 109 */ 110 static void 111 tmpfs_pager_writecount_recalc(vm_object_t object, vm_offset_t old, 112 vm_offset_t new) 113 { 114 struct vnode *vp; 115 116 VM_OBJECT_ASSERT_WLOCKED(object); 117 118 vp = VM_TO_TMPFS_VP(object); 119 120 /* 121 * Forced unmount? 122 */ 123 if (vp == NULL || vp->v_object == NULL) { 124 KASSERT((object->flags & OBJ_TMPFS_VREF) == 0, 125 ("object %p with OBJ_TMPFS_VREF but without vnode", 126 object)); 127 VM_OBJECT_WUNLOCK(object); 128 return; 129 } 130 131 if (old == 0) { 132 VNASSERT((object->flags & OBJ_TMPFS_VREF) == 0, vp, 133 ("object without writable mappings has a reference")); 134 VNPASS(vp->v_usecount > 0, vp); 135 } else { 136 VNASSERT((object->flags & OBJ_TMPFS_VREF) != 0, vp, 137 ("object with writable mappings does not " 138 "have a reference")); 139 } 140 141 if (old == new) { 142 VM_OBJECT_WUNLOCK(object); 143 return; 144 } 145 146 if (new == 0) { 147 vm_object_clear_flag(object, OBJ_TMPFS_VREF); 148 VM_OBJECT_WUNLOCK(object); 149 vrele(vp); 150 } else { 151 if ((object->flags & OBJ_TMPFS_VREF) == 0) { 152 vref(vp); 153 vlazy(vp); 154 vm_object_set_flag(object, OBJ_TMPFS_VREF); 155 } 156 VM_OBJECT_WUNLOCK(object); 157 } 158 } 159 160 static void 161 tmpfs_pager_update_writecount(vm_object_t object, vm_offset_t start, 162 vm_offset_t end) 163 { 164 vm_offset_t new, old; 165 166 VM_OBJECT_WLOCK(object); 167 KASSERT((object->flags & OBJ_ANON) == 0, 168 ("%s: object %p with OBJ_ANON", __func__, object)); 169 old = object->un_pager.swp.writemappings; 170 object->un_pager.swp.writemappings += (vm_ooffset_t)end - start; 171 new = object->un_pager.swp.writemappings; 172 tmpfs_pager_writecount_recalc(object, old, new); 173 VM_OBJECT_ASSERT_UNLOCKED(object); 174 } 175 176 static void 177 tmpfs_pager_release_writecount(vm_object_t object, vm_offset_t start, 178 vm_offset_t end) 179 { 180 vm_offset_t new, old; 181 182 VM_OBJECT_WLOCK(object); 183 KASSERT((object->flags & OBJ_ANON) == 0, 184 ("%s: object %p with OBJ_ANON", __func__, object)); 185 old = object->un_pager.swp.writemappings; 186 KASSERT(old >= (vm_ooffset_t)end - start, 187 ("tmpfs obj %p writecount %jx dec %jx", object, (uintmax_t)old, 188 (uintmax_t)((vm_ooffset_t)end - start))); 189 object->un_pager.swp.writemappings -= (vm_ooffset_t)end - start; 190 new = object->un_pager.swp.writemappings; 191 tmpfs_pager_writecount_recalc(object, old, new); 192 VM_OBJECT_ASSERT_UNLOCKED(object); 193 } 194 195 static void 196 tmpfs_pager_getvp(vm_object_t object, struct vnode **vpp, bool *vp_heldp) 197 { 198 struct vnode *vp; 199 200 /* 201 * Tmpfs VREG node, which was reclaimed, has tmpfs_pager_type 202 * type. In this case there is no v_writecount to adjust. 203 */ 204 if (vp_heldp != NULL) 205 VM_OBJECT_RLOCK(object); 206 else 207 VM_OBJECT_ASSERT_LOCKED(object); 208 if ((object->flags & OBJ_TMPFS) != 0) { 209 vp = VM_TO_TMPFS_VP(object); 210 if (vp != NULL) { 211 *vpp = vp; 212 if (vp_heldp != NULL) { 213 vhold(vp); 214 *vp_heldp = true; 215 } 216 } 217 } 218 if (vp_heldp != NULL) 219 VM_OBJECT_RUNLOCK(object); 220 } 221 222 static void 223 tmpfs_pager_freespace(vm_object_t obj, vm_pindex_t start, vm_size_t size) 224 { 225 struct tmpfs_node *node; 226 struct tmpfs_mount *tm; 227 vm_size_t c; 228 229 swap_pager_freespace(obj, start, size, &c); 230 if ((obj->flags & OBJ_TMPFS) == 0 || c == 0) 231 return; 232 233 node = obj->un_pager.swp.swp_priv; 234 MPASS(node->tn_type == VREG); 235 tm = node->tn_reg.tn_tmp; 236 237 KASSERT(tm->tm_pages_used >= c, 238 ("tmpfs tm %p pages %jd free %jd", tm, 239 (uintmax_t)tm->tm_pages_used, (uintmax_t)c)); 240 atomic_add_long(&tm->tm_pages_used, -c); 241 KASSERT(node->tn_reg.tn_pages >= c, 242 ("tmpfs node %p pages %jd free %jd", node, 243 (uintmax_t)node->tn_reg.tn_pages, (uintmax_t)c)); 244 node->tn_reg.tn_pages -= c; 245 } 246 247 static void 248 tmpfs_page_inserted(vm_object_t obj, vm_page_t m) 249 { 250 struct tmpfs_node *node; 251 struct tmpfs_mount *tm; 252 253 if ((obj->flags & OBJ_TMPFS) == 0) 254 return; 255 256 node = obj->un_pager.swp.swp_priv; 257 MPASS(node->tn_type == VREG); 258 tm = node->tn_reg.tn_tmp; 259 260 if (!vm_pager_has_page(obj, m->pindex, NULL, NULL)) { 261 atomic_add_long(&tm->tm_pages_used, 1); 262 node->tn_reg.tn_pages += 1; 263 } 264 } 265 266 static void 267 tmpfs_page_removed(vm_object_t obj, vm_page_t m) 268 { 269 struct tmpfs_node *node; 270 struct tmpfs_mount *tm; 271 272 if ((obj->flags & OBJ_TMPFS) == 0) 273 return; 274 275 node = obj->un_pager.swp.swp_priv; 276 MPASS(node->tn_type == VREG); 277 tm = node->tn_reg.tn_tmp; 278 279 if (!vm_pager_has_page(obj, m->pindex, NULL, NULL)) { 280 KASSERT(tm->tm_pages_used >= 1, 281 ("tmpfs tm %p pages %jd free 1", tm, 282 (uintmax_t)tm->tm_pages_used)); 283 atomic_add_long(&tm->tm_pages_used, -1); 284 KASSERT(node->tn_reg.tn_pages >= 1, 285 ("tmpfs node %p pages %jd free 1", node, 286 (uintmax_t)node->tn_reg.tn_pages)); 287 node->tn_reg.tn_pages -= 1; 288 } 289 } 290 291 static boolean_t 292 tmpfs_can_alloc_page(vm_object_t obj, vm_pindex_t pindex) 293 { 294 struct tmpfs_mount *tm; 295 296 tm = VM_TO_TMPFS_MP(obj); 297 if (tm == NULL || vm_pager_has_page(obj, pindex, NULL, NULL) || 298 tm->tm_pages_max == 0) 299 return (true); 300 if (tm->tm_pages_max == ULONG_MAX) 301 return (tmpfs_mem_avail() >= 1); 302 return (tm->tm_pages_max > atomic_load_long(&tm->tm_pages_used)); 303 } 304 305 struct pagerops tmpfs_pager_ops = { 306 .pgo_kvme_type = KVME_TYPE_VNODE, 307 .pgo_alloc = tmpfs_pager_alloc, 308 .pgo_set_writeable_dirty = vm_object_set_writeable_dirty_, 309 .pgo_update_writecount = tmpfs_pager_update_writecount, 310 .pgo_release_writecount = tmpfs_pager_release_writecount, 311 .pgo_mightbedirty = vm_object_mightbedirty_, 312 .pgo_getvp = tmpfs_pager_getvp, 313 .pgo_freespace = tmpfs_pager_freespace, 314 .pgo_page_inserted = tmpfs_page_inserted, 315 .pgo_page_removed = tmpfs_page_removed, 316 .pgo_can_alloc_page = tmpfs_can_alloc_page, 317 }; 318 319 static int 320 tmpfs_node_ctor(void *mem, int size, void *arg, int flags) 321 { 322 struct tmpfs_node *node; 323 324 node = mem; 325 node->tn_gen++; 326 node->tn_size = 0; 327 node->tn_status = 0; 328 node->tn_accessed = false; 329 node->tn_flags = 0; 330 node->tn_links = 0; 331 node->tn_vnode = NULL; 332 node->tn_vpstate = 0; 333 return (0); 334 } 335 336 static void 337 tmpfs_node_dtor(void *mem, int size, void *arg) 338 { 339 struct tmpfs_node *node; 340 341 node = mem; 342 node->tn_type = VNON; 343 } 344 345 static int 346 tmpfs_node_init(void *mem, int size, int flags) 347 { 348 struct tmpfs_node *node; 349 350 node = mem; 351 node->tn_id = 0; 352 mtx_init(&node->tn_interlock, "tmpfsni", NULL, MTX_DEF); 353 node->tn_gen = arc4random(); 354 return (0); 355 } 356 357 static void 358 tmpfs_node_fini(void *mem, int size) 359 { 360 struct tmpfs_node *node; 361 362 node = mem; 363 mtx_destroy(&node->tn_interlock); 364 } 365 366 int 367 tmpfs_subr_init(void) 368 { 369 tmpfs_pager_type = vm_pager_alloc_dyn_type(&tmpfs_pager_ops, 370 OBJT_SWAP); 371 if (tmpfs_pager_type == -1) 372 return (EINVAL); 373 tmpfs_node_pool = uma_zcreate("TMPFS node", 374 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor, 375 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0); 376 VFS_SMR_ZONE_SET(tmpfs_node_pool); 377 378 tmpfs_pages_avail_init = tmpfs_mem_avail(); 379 tmpfs_set_reserve_from_percent(); 380 return (0); 381 } 382 383 void 384 tmpfs_subr_uninit(void) 385 { 386 if (tmpfs_pager_type != -1) 387 vm_pager_free_dyn_type(tmpfs_pager_type); 388 tmpfs_pager_type = -1; 389 uma_zdestroy(tmpfs_node_pool); 390 } 391 392 static int 393 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS) 394 { 395 int error; 396 long pages, bytes; 397 398 pages = *(long *)arg1; 399 bytes = pages * PAGE_SIZE; 400 401 error = sysctl_handle_long(oidp, &bytes, 0, req); 402 if (error || !req->newptr) 403 return (error); 404 405 pages = bytes / PAGE_SIZE; 406 if (pages < TMPFS_PAGES_MINRESERVED) 407 return (EINVAL); 408 409 *(long *)arg1 = pages; 410 return (0); 411 } 412 413 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, 414 CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &tmpfs_pages_reserved, 0, 415 sysctl_mem_reserved, "L", 416 "Amount of available memory and swap below which tmpfs growth stops"); 417 418 static int 419 sysctl_mem_percent(SYSCTL_HANDLER_ARGS) 420 { 421 int error, percent; 422 423 percent = *(int *)arg1; 424 error = sysctl_handle_int(oidp, &percent, 0, req); 425 if (error || !req->newptr) 426 return (error); 427 428 if ((unsigned) percent > 100) 429 return (EINVAL); 430 431 *(long *)arg1 = percent; 432 tmpfs_set_reserve_from_percent(); 433 return (0); 434 } 435 436 static void 437 tmpfs_set_reserve_from_percent(void) 438 { 439 size_t reserved; 440 441 reserved = tmpfs_pages_avail_init * (100 - tmpfs_mem_percent) / 100; 442 tmpfs_pages_reserved = max(reserved, TMPFS_PAGES_MINRESERVED); 443 } 444 445 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_percent, 446 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &tmpfs_mem_percent, 0, 447 sysctl_mem_percent, "I", 448 "Percent of available memory that can be used if no size limit"); 449 450 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a, 451 struct tmpfs_dirent *b); 452 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 453 454 size_t 455 tmpfs_mem_avail(void) 456 { 457 size_t avail; 458 long reserved; 459 460 avail = swap_pager_avail + vm_free_count(); 461 reserved = atomic_load_long(&tmpfs_pages_reserved); 462 if (__predict_false(avail < reserved)) 463 return (0); 464 return (avail - reserved); 465 } 466 467 size_t 468 tmpfs_pages_used(struct tmpfs_mount *tmp) 469 { 470 const size_t node_size = sizeof(struct tmpfs_node) + 471 sizeof(struct tmpfs_dirent); 472 size_t meta_pages; 473 474 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size, 475 PAGE_SIZE); 476 return (meta_pages + tmp->tm_pages_used); 477 } 478 479 bool 480 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages) 481 { 482 if (tmpfs_mem_avail() < req_pages) 483 return (false); 484 485 if (tmp->tm_pages_max != ULONG_MAX && 486 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp)) 487 return (false); 488 489 return (true); 490 } 491 492 static int 493 tmpfs_partial_page_invalidate(vm_object_t object, vm_pindex_t idx, int base, 494 int end, boolean_t ignerr) 495 { 496 vm_page_t m; 497 int rv, error; 498 499 VM_OBJECT_ASSERT_WLOCKED(object); 500 KASSERT(base >= 0, ("%s: base %d", __func__, base)); 501 KASSERT(end - base <= PAGE_SIZE, ("%s: base %d end %d", __func__, base, 502 end)); 503 error = 0; 504 505 retry: 506 m = vm_page_grab(object, idx, VM_ALLOC_NOCREAT); 507 if (m != NULL) { 508 MPASS(vm_page_all_valid(m)); 509 } else if (vm_pager_has_page(object, idx, NULL, NULL)) { 510 m = vm_page_alloc(object, idx, VM_ALLOC_NORMAL | 511 VM_ALLOC_WAITFAIL); 512 if (m == NULL) 513 goto retry; 514 vm_object_pip_add(object, 1); 515 VM_OBJECT_WUNLOCK(object); 516 rv = vm_pager_get_pages(object, &m, 1, NULL, NULL); 517 VM_OBJECT_WLOCK(object); 518 vm_object_pip_wakeup(object); 519 if (rv == VM_PAGER_OK) { 520 /* 521 * Since the page was not resident, and therefore not 522 * recently accessed, immediately enqueue it for 523 * asynchronous laundering. The current operation is 524 * not regarded as an access. 525 */ 526 vm_page_launder(m); 527 } else { 528 vm_page_free(m); 529 m = NULL; 530 if (!ignerr) 531 error = EIO; 532 } 533 } 534 if (m != NULL) { 535 pmap_zero_page_area(m, base, end - base); 536 vm_page_set_dirty(m); 537 vm_page_xunbusy(m); 538 } 539 540 return (error); 541 } 542 543 void 544 tmpfs_ref_node(struct tmpfs_node *node) 545 { 546 #ifdef INVARIANTS 547 u_int old; 548 549 old = 550 #endif 551 refcount_acquire(&node->tn_refcount); 552 #ifdef INVARIANTS 553 KASSERT(old > 0, ("node %p zero refcount", node)); 554 #endif 555 } 556 557 /* 558 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 559 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 560 * using the credentials of the process 'p'. 561 * 562 * If the node type is set to 'VDIR', then the parent parameter must point 563 * to the parent directory of the node being created. It may only be NULL 564 * while allocating the root node. 565 * 566 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 567 * specifies the device the node represents. 568 * 569 * If the node type is set to 'VLNK', then the parameter target specifies 570 * the file name of the target file for the symbolic link that is being 571 * created. 572 * 573 * Note that new nodes are retrieved from the available list if it has 574 * items or, if it is empty, from the node pool as long as there is enough 575 * space to create them. 576 * 577 * Returns zero on success or an appropriate error code on failure. 578 */ 579 int 580 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, __enum_uint8(vtype) type, 581 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 582 const char *target, dev_t rdev, struct tmpfs_node **node) 583 { 584 struct tmpfs_node *nnode; 585 char *symlink; 586 char symlink_smr; 587 588 /* If the root directory of the 'tmp' file system is not yet 589 * allocated, this must be the request to do it. */ 590 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 591 592 MPASS((type == VLNK) ^ (target == NULL)); 593 MPASS((type == VBLK || type == VCHR) ^ (rdev == VNOVAL)); 594 595 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 596 return (ENOSPC); 597 if (!tmpfs_pages_check_avail(tmp, 1)) 598 return (ENOSPC); 599 600 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 601 /* 602 * When a new tmpfs node is created for fully 603 * constructed mount point, there must be a parent 604 * node, which vnode is locked exclusively. As 605 * consequence, if the unmount is executing in 606 * parallel, vflush() cannot reclaim the parent vnode. 607 * Due to this, the check for MNTK_UNMOUNT flag is not 608 * racy: if we did not see MNTK_UNMOUNT flag, then tmp 609 * cannot be destroyed until node construction is 610 * finished and the parent vnode unlocked. 611 * 612 * Tmpfs does not need to instantiate new nodes during 613 * unmount. 614 */ 615 return (EBUSY); 616 } 617 if ((mp->mnt_kern_flag & MNT_RDONLY) != 0) 618 return (EROFS); 619 620 nnode = uma_zalloc_smr(tmpfs_node_pool, M_WAITOK); 621 622 /* Generic initialization. */ 623 nnode->tn_type = type; 624 vfs_timestamp(&nnode->tn_atime); 625 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 626 nnode->tn_atime; 627 nnode->tn_uid = uid; 628 nnode->tn_gid = gid; 629 nnode->tn_mode = mode; 630 nnode->tn_id = alloc_unr64(&tmp->tm_ino_unr); 631 nnode->tn_refcount = 1; 632 LIST_INIT(&nnode->tn_extattrs); 633 634 /* Type-specific initialization. */ 635 switch (nnode->tn_type) { 636 case VBLK: 637 case VCHR: 638 nnode->tn_rdev = rdev; 639 break; 640 641 case VDIR: 642 RB_INIT(&nnode->tn_dir.tn_dirhead); 643 LIST_INIT(&nnode->tn_dir.tn_dupindex); 644 MPASS(parent != nnode); 645 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL)); 646 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent; 647 nnode->tn_dir.tn_readdir_lastn = 0; 648 nnode->tn_dir.tn_readdir_lastp = NULL; 649 nnode->tn_links++; 650 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent); 651 nnode->tn_dir.tn_parent->tn_links++; 652 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent); 653 break; 654 655 case VFIFO: 656 /* FALLTHROUGH */ 657 case VSOCK: 658 break; 659 660 case VLNK: 661 MPASS(strlen(target) < MAXPATHLEN); 662 nnode->tn_size = strlen(target); 663 664 symlink = NULL; 665 if (!tmp->tm_nonc) { 666 symlink = cache_symlink_alloc(nnode->tn_size + 1, 667 M_WAITOK); 668 symlink_smr = true; 669 } 670 if (symlink == NULL) { 671 symlink = malloc(nnode->tn_size + 1, M_TMPFSNAME, 672 M_WAITOK); 673 symlink_smr = false; 674 } 675 memcpy(symlink, target, nnode->tn_size + 1); 676 677 /* 678 * Allow safe symlink resolving for lockless lookup. 679 * tmpfs_fplookup_symlink references this comment. 680 * 681 * 1. nnode is not yet visible to the world 682 * 2. both tn_link_target and tn_link_smr get populated 683 * 3. release fence publishes their content 684 * 4. tn_link_target content is immutable until node 685 * destruction, where the pointer gets set to NULL 686 * 5. tn_link_smr is never changed once set 687 * 688 * As a result it is sufficient to issue load consume 689 * on the node pointer to also get the above content 690 * in a stable manner. Worst case tn_link_smr flag 691 * may be set to true despite being stale, while the 692 * target buffer is already cleared out. 693 */ 694 atomic_store_ptr(&nnode->tn_link_target, symlink); 695 atomic_store_char((char *)&nnode->tn_link_smr, symlink_smr); 696 atomic_thread_fence_rel(); 697 break; 698 699 case VREG: 700 nnode->tn_reg.tn_aobj = 701 vm_pager_allocate(tmpfs_pager_type, NULL, 0, 702 VM_PROT_DEFAULT, 0, 703 NULL /* XXXKIB - tmpfs needs swap reservation */); 704 nnode->tn_reg.tn_aobj->un_pager.swp.swp_priv = nnode; 705 vm_object_set_flag(nnode->tn_reg.tn_aobj, OBJ_TMPFS); 706 nnode->tn_reg.tn_tmp = tmp; 707 nnode->tn_reg.tn_pages = 0; 708 break; 709 710 default: 711 panic("tmpfs_alloc_node: type %p %d", nnode, 712 (int)nnode->tn_type); 713 } 714 715 TMPFS_LOCK(tmp); 716 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 717 nnode->tn_attached = true; 718 tmp->tm_nodes_inuse++; 719 tmp->tm_refcount++; 720 TMPFS_UNLOCK(tmp); 721 722 *node = nnode; 723 return (0); 724 } 725 726 /* 727 * Destroys the node pointed to by node from the file system 'tmp'. 728 * If the node references a directory, no entries are allowed. 729 */ 730 void 731 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 732 { 733 if (refcount_release_if_not_last(&node->tn_refcount)) 734 return; 735 736 TMPFS_LOCK(tmp); 737 TMPFS_NODE_LOCK(node); 738 if (!tmpfs_free_node_locked(tmp, node, false)) { 739 TMPFS_NODE_UNLOCK(node); 740 TMPFS_UNLOCK(tmp); 741 } 742 } 743 744 bool 745 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node, 746 bool detach) 747 { 748 struct tmpfs_extattr *ea; 749 vm_object_t uobj; 750 char *symlink; 751 bool last; 752 753 TMPFS_MP_ASSERT_LOCKED(tmp); 754 TMPFS_NODE_ASSERT_LOCKED(node); 755 756 last = refcount_release(&node->tn_refcount); 757 if (node->tn_attached && (detach || last)) { 758 MPASS(tmp->tm_nodes_inuse > 0); 759 tmp->tm_nodes_inuse--; 760 LIST_REMOVE(node, tn_entries); 761 node->tn_attached = false; 762 } 763 if (!last) 764 return (false); 765 766 TMPFS_NODE_UNLOCK(node); 767 768 #ifdef INVARIANTS 769 MPASS(node->tn_vnode == NULL); 770 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 771 772 /* 773 * Make sure this is a node type we can deal with. Everything 774 * is explicitly enumerated without the 'default' clause so 775 * the compiler can throw an error in case a new type is 776 * added. 777 */ 778 switch (node->tn_type) { 779 case VBLK: 780 case VCHR: 781 case VDIR: 782 case VFIFO: 783 case VSOCK: 784 case VLNK: 785 case VREG: 786 break; 787 case VNON: 788 case VBAD: 789 case VMARKER: 790 panic("%s: bad type %d for node %p", __func__, 791 (int)node->tn_type, node); 792 } 793 #endif 794 795 while ((ea = LIST_FIRST(&node->tn_extattrs)) != NULL) { 796 LIST_REMOVE(ea, ea_extattrs); 797 tmpfs_extattr_free(ea); 798 } 799 800 switch (node->tn_type) { 801 case VREG: 802 uobj = node->tn_reg.tn_aobj; 803 node->tn_reg.tn_aobj = NULL; 804 if (uobj != NULL) { 805 VM_OBJECT_WLOCK(uobj); 806 KASSERT((uobj->flags & OBJ_TMPFS) != 0, 807 ("tmpfs node %p uobj %p not tmpfs", node, uobj)); 808 vm_object_clear_flag(uobj, OBJ_TMPFS); 809 KASSERT(tmp->tm_pages_used >= node->tn_reg.tn_pages, 810 ("tmpfs tmp %p node %p pages %jd free %jd", tmp, 811 node, (uintmax_t)tmp->tm_pages_used, 812 (uintmax_t)node->tn_reg.tn_pages)); 813 atomic_add_long(&tmp->tm_pages_used, 814 -node->tn_reg.tn_pages); 815 VM_OBJECT_WUNLOCK(uobj); 816 } 817 tmpfs_free_tmp(tmp); 818 819 /* 820 * vm_object_deallocate() must not be called while 821 * owning tm_allnode_lock, because deallocate might 822 * sleep. Call it after tmpfs_free_tmp() does the 823 * unlock. 824 */ 825 if (uobj != NULL) 826 vm_object_deallocate(uobj); 827 828 break; 829 case VLNK: 830 tmpfs_free_tmp(tmp); 831 832 symlink = node->tn_link_target; 833 atomic_store_ptr(&node->tn_link_target, NULL); 834 if (atomic_load_char(&node->tn_link_smr)) { 835 cache_symlink_free(symlink, node->tn_size + 1); 836 } else { 837 free(symlink, M_TMPFSNAME); 838 } 839 break; 840 default: 841 tmpfs_free_tmp(tmp); 842 break; 843 } 844 845 uma_zfree_smr(tmpfs_node_pool, node); 846 return (true); 847 } 848 849 static __inline uint32_t 850 tmpfs_dirent_hash(const char *name, u_int len) 851 { 852 uint32_t hash; 853 854 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK; 855 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP 856 hash &= 0xf; 857 #endif 858 if (hash < TMPFS_DIRCOOKIE_MIN) 859 hash += TMPFS_DIRCOOKIE_MIN; 860 861 return (hash); 862 } 863 864 static __inline off_t 865 tmpfs_dirent_cookie(struct tmpfs_dirent *de) 866 { 867 if (de == NULL) 868 return (TMPFS_DIRCOOKIE_EOF); 869 870 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN); 871 872 return (de->td_cookie); 873 } 874 875 static __inline boolean_t 876 tmpfs_dirent_dup(struct tmpfs_dirent *de) 877 { 878 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0); 879 } 880 881 static __inline boolean_t 882 tmpfs_dirent_duphead(struct tmpfs_dirent *de) 883 { 884 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0); 885 } 886 887 void 888 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen) 889 { 890 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen); 891 memcpy(de->ud.td_name, name, namelen); 892 de->td_namelen = namelen; 893 } 894 895 /* 896 * Allocates a new directory entry for the node node with a name of name. 897 * The new directory entry is returned in *de. 898 * 899 * The link count of node is increased by one to reflect the new object 900 * referencing it. 901 * 902 * Returns zero on success or an appropriate error code on failure. 903 */ 904 int 905 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 906 const char *name, u_int len, struct tmpfs_dirent **de) 907 { 908 struct tmpfs_dirent *nde; 909 910 nde = malloc(sizeof(*nde), M_TMPFSDIR, M_WAITOK); 911 nde->td_node = node; 912 if (name != NULL) { 913 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 914 tmpfs_dirent_init(nde, name, len); 915 } else 916 nde->td_namelen = 0; 917 if (node != NULL) 918 node->tn_links++; 919 920 *de = nde; 921 922 return (0); 923 } 924 925 /* 926 * Frees a directory entry. It is the caller's responsibility to destroy 927 * the node referenced by it if needed. 928 * 929 * The link count of node is decreased by one to reflect the removal of an 930 * object that referenced it. This only happens if 'node_exists' is true; 931 * otherwise the function will not access the node referred to by the 932 * directory entry, as it may already have been released from the outside. 933 */ 934 void 935 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 936 { 937 struct tmpfs_node *node; 938 939 node = de->td_node; 940 if (node != NULL) { 941 MPASS(node->tn_links > 0); 942 node->tn_links--; 943 } 944 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL) 945 free(de->ud.td_name, M_TMPFSNAME); 946 free(de, M_TMPFSDIR); 947 } 948 949 void 950 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj) 951 { 952 bool want_vrele; 953 954 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject"); 955 if (vp->v_type != VREG || obj == NULL) 956 return; 957 958 VM_OBJECT_WLOCK(obj); 959 VI_LOCK(vp); 960 vp->v_object = NULL; 961 962 /* 963 * May be going through forced unmount. 964 */ 965 want_vrele = false; 966 if ((obj->flags & OBJ_TMPFS_VREF) != 0) { 967 vm_object_clear_flag(obj, OBJ_TMPFS_VREF); 968 want_vrele = true; 969 } 970 971 if (vp->v_writecount < 0) 972 vp->v_writecount = 0; 973 VI_UNLOCK(vp); 974 VM_OBJECT_WUNLOCK(obj); 975 if (want_vrele) { 976 vrele(vp); 977 } 978 } 979 980 /* 981 * Allocates a new vnode for the node node or returns a new reference to 982 * an existing one if the node had already a vnode referencing it. The 983 * resulting locked vnode is returned in *vpp. 984 * 985 * Returns zero on success or an appropriate error code on failure. 986 */ 987 int 988 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 989 struct vnode **vpp) 990 { 991 struct vnode *vp; 992 enum vgetstate vs; 993 struct tmpfs_mount *tm; 994 vm_object_t object; 995 int error; 996 997 error = 0; 998 tm = VFS_TO_TMPFS(mp); 999 TMPFS_NODE_LOCK(node); 1000 tmpfs_ref_node(node); 1001 loop: 1002 TMPFS_NODE_ASSERT_LOCKED(node); 1003 if ((vp = node->tn_vnode) != NULL) { 1004 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 1005 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) || 1006 (VN_IS_DOOMED(vp) && 1007 (lkflag & LK_NOWAIT) != 0)) { 1008 TMPFS_NODE_UNLOCK(node); 1009 error = ENOENT; 1010 vp = NULL; 1011 goto out; 1012 } 1013 if (VN_IS_DOOMED(vp)) { 1014 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM; 1015 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) { 1016 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 1017 0, "tmpfsE", 0); 1018 } 1019 goto loop; 1020 } 1021 vs = vget_prep(vp); 1022 TMPFS_NODE_UNLOCK(node); 1023 error = vget_finish(vp, lkflag, vs); 1024 if (error == ENOENT) { 1025 TMPFS_NODE_LOCK(node); 1026 goto loop; 1027 } 1028 if (error != 0) { 1029 vp = NULL; 1030 goto out; 1031 } 1032 1033 /* 1034 * Make sure the vnode is still there after 1035 * getting the interlock to avoid racing a free. 1036 */ 1037 if (node->tn_vnode != vp) { 1038 vput(vp); 1039 TMPFS_NODE_LOCK(node); 1040 goto loop; 1041 } 1042 1043 goto out; 1044 } 1045 1046 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 1047 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 1048 TMPFS_NODE_UNLOCK(node); 1049 error = ENOENT; 1050 vp = NULL; 1051 goto out; 1052 } 1053 1054 /* 1055 * otherwise lock the vp list while we call getnewvnode 1056 * since that can block. 1057 */ 1058 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 1059 node->tn_vpstate |= TMPFS_VNODE_WANT; 1060 error = msleep((caddr_t) &node->tn_vpstate, 1061 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0); 1062 if (error != 0) 1063 goto out; 1064 goto loop; 1065 } else 1066 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 1067 1068 TMPFS_NODE_UNLOCK(node); 1069 1070 /* Get a new vnode and associate it with our node. */ 1071 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ? 1072 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp); 1073 if (error != 0) 1074 goto unlock; 1075 MPASS(vp != NULL); 1076 1077 /* lkflag is ignored, the lock is exclusive */ 1078 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1079 1080 vp->v_data = node; 1081 vp->v_type = node->tn_type; 1082 1083 /* Type-specific initialization. */ 1084 switch (node->tn_type) { 1085 case VBLK: 1086 /* FALLTHROUGH */ 1087 case VCHR: 1088 /* FALLTHROUGH */ 1089 case VLNK: 1090 /* FALLTHROUGH */ 1091 case VSOCK: 1092 break; 1093 case VFIFO: 1094 vp->v_op = &tmpfs_fifoop_entries; 1095 break; 1096 case VREG: 1097 object = node->tn_reg.tn_aobj; 1098 VM_OBJECT_WLOCK(object); 1099 KASSERT((object->flags & OBJ_TMPFS_VREF) == 0, 1100 ("%s: object %p with OBJ_TMPFS_VREF but without vnode", 1101 __func__, object)); 1102 VI_LOCK(vp); 1103 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 1104 vp->v_object = object; 1105 vn_irflag_set_locked(vp, (tm->tm_pgread ? VIRF_PGREAD : 0) | 1106 VIRF_TEXT_REF); 1107 VI_UNLOCK(vp); 1108 VNASSERT((object->flags & OBJ_TMPFS_VREF) == 0, vp, 1109 ("leaked OBJ_TMPFS_VREF")); 1110 if (object->un_pager.swp.writemappings > 0) { 1111 vrefact(vp); 1112 vlazy(vp); 1113 vm_object_set_flag(object, OBJ_TMPFS_VREF); 1114 } 1115 VM_OBJECT_WUNLOCK(object); 1116 break; 1117 case VDIR: 1118 MPASS(node->tn_dir.tn_parent != NULL); 1119 if (node->tn_dir.tn_parent == node) 1120 vp->v_vflag |= VV_ROOT; 1121 break; 1122 1123 default: 1124 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 1125 } 1126 if (vp->v_type != VFIFO) 1127 VN_LOCK_ASHARE(vp); 1128 1129 error = insmntque1(vp, mp); 1130 if (error != 0) { 1131 /* Need to clear v_object for insmntque failure. */ 1132 tmpfs_destroy_vobject(vp, vp->v_object); 1133 vp->v_object = NULL; 1134 vp->v_data = NULL; 1135 vp->v_op = &dead_vnodeops; 1136 vgone(vp); 1137 vput(vp); 1138 vp = NULL; 1139 } else { 1140 vn_set_state(vp, VSTATE_CONSTRUCTED); 1141 } 1142 1143 unlock: 1144 TMPFS_NODE_LOCK(node); 1145 1146 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 1147 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 1148 node->tn_vnode = vp; 1149 1150 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 1151 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 1152 TMPFS_NODE_UNLOCK(node); 1153 wakeup((caddr_t) &node->tn_vpstate); 1154 } else 1155 TMPFS_NODE_UNLOCK(node); 1156 1157 out: 1158 if (error == 0) { 1159 *vpp = vp; 1160 1161 #ifdef INVARIANTS 1162 MPASS(*vpp != NULL); 1163 ASSERT_VOP_LOCKED(*vpp, __func__); 1164 TMPFS_NODE_LOCK(node); 1165 MPASS(*vpp == node->tn_vnode); 1166 TMPFS_NODE_UNLOCK(node); 1167 #endif 1168 } 1169 tmpfs_free_node(tm, node); 1170 1171 return (error); 1172 } 1173 1174 /* 1175 * Destroys the association between the vnode vp and the node it 1176 * references. 1177 */ 1178 void 1179 tmpfs_free_vp(struct vnode *vp) 1180 { 1181 struct tmpfs_node *node; 1182 1183 node = VP_TO_TMPFS_NODE(vp); 1184 1185 TMPFS_NODE_ASSERT_LOCKED(node); 1186 node->tn_vnode = NULL; 1187 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 1188 wakeup(&node->tn_vnode); 1189 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 1190 vp->v_data = NULL; 1191 } 1192 1193 /* 1194 * Allocates a new file of type 'type' and adds it to the parent directory 1195 * 'dvp'; this addition is done using the component name given in 'cnp'. 1196 * The ownership of the new file is automatically assigned based on the 1197 * credentials of the caller (through 'cnp'), the group is set based on 1198 * the parent directory and the mode is determined from the 'vap' argument. 1199 * If successful, *vpp holds a vnode to the newly created file and zero 1200 * is returned. Otherwise *vpp is NULL and the function returns an 1201 * appropriate error code. 1202 */ 1203 int 1204 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 1205 struct componentname *cnp, const char *target) 1206 { 1207 int error; 1208 struct tmpfs_dirent *de; 1209 struct tmpfs_mount *tmp; 1210 struct tmpfs_node *dnode; 1211 struct tmpfs_node *node; 1212 struct tmpfs_node *parent; 1213 1214 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file"); 1215 1216 tmp = VFS_TO_TMPFS(dvp->v_mount); 1217 dnode = VP_TO_TMPFS_DIR(dvp); 1218 *vpp = NULL; 1219 1220 /* If the entry we are creating is a directory, we cannot overflow 1221 * the number of links of its parent, because it will get a new 1222 * link. */ 1223 if (vap->va_type == VDIR) { 1224 /* Ensure that we do not overflow the maximum number of links 1225 * imposed by the system. */ 1226 MPASS(dnode->tn_links <= TMPFS_LINK_MAX); 1227 if (dnode->tn_links == TMPFS_LINK_MAX) { 1228 return (EMLINK); 1229 } 1230 1231 parent = dnode; 1232 MPASS(parent != NULL); 1233 } else 1234 parent = NULL; 1235 1236 /* Allocate a node that represents the new file. */ 1237 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type, 1238 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent, 1239 target, vap->va_rdev, &node); 1240 if (error != 0) 1241 return (error); 1242 1243 /* Allocate a directory entry that points to the new file. */ 1244 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 1245 &de); 1246 if (error != 0) { 1247 tmpfs_free_node(tmp, node); 1248 return (error); 1249 } 1250 1251 /* Allocate a vnode for the new file. */ 1252 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 1253 if (error != 0) { 1254 tmpfs_free_dirent(tmp, de); 1255 tmpfs_free_node(tmp, node); 1256 return (error); 1257 } 1258 1259 /* Now that all required items are allocated, we can proceed to 1260 * insert the new node into the directory, an operation that 1261 * cannot fail. */ 1262 if (cnp->cn_flags & ISWHITEOUT) 1263 tmpfs_dir_whiteout_remove(dvp, cnp); 1264 tmpfs_dir_attach(dvp, de); 1265 return (0); 1266 } 1267 1268 struct tmpfs_dirent * 1269 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 1270 { 1271 struct tmpfs_dirent *de; 1272 1273 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 1274 dc->tdc_tree = de; 1275 if (de != NULL && tmpfs_dirent_duphead(de)) 1276 de = LIST_FIRST(&de->ud.td_duphead); 1277 dc->tdc_current = de; 1278 1279 return (dc->tdc_current); 1280 } 1281 1282 struct tmpfs_dirent * 1283 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 1284 { 1285 struct tmpfs_dirent *de; 1286 1287 MPASS(dc->tdc_tree != NULL); 1288 if (tmpfs_dirent_dup(dc->tdc_current)) { 1289 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 1290 if (dc->tdc_current != NULL) 1291 return (dc->tdc_current); 1292 } 1293 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 1294 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 1295 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 1296 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 1297 MPASS(dc->tdc_current != NULL); 1298 } 1299 1300 return (dc->tdc_current); 1301 } 1302 1303 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 1304 static struct tmpfs_dirent * 1305 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 1306 { 1307 struct tmpfs_dirent *de, dekey; 1308 1309 dekey.td_hash = hash; 1310 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 1311 return (de); 1312 } 1313 1314 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 1315 static struct tmpfs_dirent * 1316 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 1317 struct tmpfs_dir_cursor *dc) 1318 { 1319 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 1320 struct tmpfs_dirent *de, dekey; 1321 1322 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 1323 1324 if (cookie == node->tn_dir.tn_readdir_lastn && 1325 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 1326 /* Protect against possible race, tn_readdir_last[pn] 1327 * may be updated with only shared vnode lock held. */ 1328 if (cookie == tmpfs_dirent_cookie(de)) 1329 goto out; 1330 } 1331 1332 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 1333 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 1334 uh.td_dup.index_entries) { 1335 MPASS(tmpfs_dirent_dup(de)); 1336 if (de->td_cookie == cookie) 1337 goto out; 1338 /* dupindex list is sorted. */ 1339 if (de->td_cookie < cookie) { 1340 de = NULL; 1341 goto out; 1342 } 1343 } 1344 MPASS(de == NULL); 1345 goto out; 1346 } 1347 1348 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) { 1349 de = NULL; 1350 } else { 1351 dekey.td_hash = cookie; 1352 /* Recover if direntry for cookie was removed */ 1353 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 1354 } 1355 dc->tdc_tree = de; 1356 dc->tdc_current = de; 1357 if (de != NULL && tmpfs_dirent_duphead(de)) { 1358 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 1359 MPASS(dc->tdc_current != NULL); 1360 } 1361 return (dc->tdc_current); 1362 1363 out: 1364 dc->tdc_tree = de; 1365 dc->tdc_current = de; 1366 if (de != NULL && tmpfs_dirent_dup(de)) 1367 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 1368 de->td_hash); 1369 return (dc->tdc_current); 1370 } 1371 1372 /* 1373 * Looks for a directory entry in the directory represented by node. 1374 * 'cnp' describes the name of the entry to look for. Note that the . 1375 * and .. components are not allowed as they do not physically exist 1376 * within directories. 1377 * 1378 * Returns a pointer to the entry when found, otherwise NULL. 1379 */ 1380 struct tmpfs_dirent * 1381 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 1382 struct componentname *cnp) 1383 { 1384 struct tmpfs_dir_duphead *duphead; 1385 struct tmpfs_dirent *de; 1386 uint32_t hash; 1387 1388 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 1389 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 1390 cnp->cn_nameptr[1] == '.'))); 1391 TMPFS_VALIDATE_DIR(node); 1392 1393 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 1394 de = tmpfs_dir_xlookup_hash(node, hash); 1395 if (de != NULL && tmpfs_dirent_duphead(de)) { 1396 duphead = &de->ud.td_duphead; 1397 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 1398 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 1399 cnp->cn_namelen)) 1400 break; 1401 } 1402 } else if (de != NULL) { 1403 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 1404 cnp->cn_namelen)) 1405 de = NULL; 1406 } 1407 if (de != NULL && f != NULL && de->td_node != f) 1408 de = NULL; 1409 1410 return (de); 1411 } 1412 1413 /* 1414 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 1415 * list, allocate new cookie value. 1416 */ 1417 static void 1418 tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 1419 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 1420 { 1421 struct tmpfs_dir_duphead *dupindex; 1422 struct tmpfs_dirent *de, *pde; 1423 1424 dupindex = &dnode->tn_dir.tn_dupindex; 1425 de = LIST_FIRST(dupindex); 1426 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 1427 if (de == NULL) 1428 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 1429 else 1430 nde->td_cookie = de->td_cookie + 1; 1431 MPASS(tmpfs_dirent_dup(nde)); 1432 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 1433 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 1434 return; 1435 } 1436 1437 /* 1438 * Cookie numbers are near exhaustion. Scan dupindex list for unused 1439 * numbers. dupindex list is sorted in descending order. Keep it so 1440 * after inserting nde. 1441 */ 1442 while (1) { 1443 pde = de; 1444 de = LIST_NEXT(de, uh.td_dup.index_entries); 1445 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 1446 /* 1447 * Last element of the index doesn't have minimal cookie 1448 * value, use it. 1449 */ 1450 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 1451 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 1452 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 1453 return; 1454 } else if (de == NULL) { 1455 /* 1456 * We are so lucky have 2^30 hash duplicates in single 1457 * directory :) Return largest possible cookie value. 1458 * It should be fine except possible issues with 1459 * VOP_READDIR restart. 1460 */ 1461 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 1462 LIST_INSERT_HEAD(dupindex, nde, 1463 uh.td_dup.index_entries); 1464 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 1465 return; 1466 } 1467 if (de->td_cookie + 1 == pde->td_cookie || 1468 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 1469 continue; /* No hole or invalid cookie. */ 1470 nde->td_cookie = de->td_cookie + 1; 1471 MPASS(tmpfs_dirent_dup(nde)); 1472 MPASS(pde->td_cookie > nde->td_cookie); 1473 MPASS(nde->td_cookie > de->td_cookie); 1474 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 1475 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 1476 return; 1477 } 1478 } 1479 1480 /* 1481 * Attaches the directory entry de to the directory represented by vp. 1482 * Note that this does not change the link count of the node pointed by 1483 * the directory entry, as this is done by tmpfs_alloc_dirent. 1484 */ 1485 void 1486 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 1487 { 1488 struct tmpfs_node *dnode; 1489 struct tmpfs_dirent *xde, *nde; 1490 1491 ASSERT_VOP_ELOCKED(vp, __func__); 1492 MPASS(de->td_namelen > 0); 1493 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 1494 MPASS(de->td_cookie == de->td_hash); 1495 1496 dnode = VP_TO_TMPFS_DIR(vp); 1497 dnode->tn_dir.tn_readdir_lastn = 0; 1498 dnode->tn_dir.tn_readdir_lastp = NULL; 1499 1500 MPASS(!tmpfs_dirent_dup(de)); 1501 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1502 if (xde != NULL && tmpfs_dirent_duphead(xde)) 1503 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1504 else if (xde != NULL) { 1505 /* 1506 * Allocate new duphead. Swap xde with duphead to avoid 1507 * adding/removing elements with the same hash. 1508 */ 1509 MPASS(!tmpfs_dirent_dup(xde)); 1510 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 1511 &nde); 1512 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 1513 memcpy(nde, xde, sizeof(*xde)); 1514 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 1515 LIST_INIT(&xde->ud.td_duphead); 1516 xde->td_namelen = 0; 1517 xde->td_node = NULL; 1518 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 1519 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1520 } 1521 dnode->tn_size += sizeof(struct tmpfs_dirent); 1522 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1523 dnode->tn_accessed = true; 1524 tmpfs_update(vp); 1525 } 1526 1527 /* 1528 * Detaches the directory entry de from the directory represented by vp. 1529 * Note that this does not change the link count of the node pointed by 1530 * the directory entry, as this is done by tmpfs_free_dirent. 1531 */ 1532 void 1533 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 1534 { 1535 struct tmpfs_mount *tmp; 1536 struct tmpfs_dir *head; 1537 struct tmpfs_node *dnode; 1538 struct tmpfs_dirent *xde; 1539 1540 ASSERT_VOP_ELOCKED(vp, __func__); 1541 1542 dnode = VP_TO_TMPFS_DIR(vp); 1543 head = &dnode->tn_dir.tn_dirhead; 1544 dnode->tn_dir.tn_readdir_lastn = 0; 1545 dnode->tn_dir.tn_readdir_lastp = NULL; 1546 1547 if (tmpfs_dirent_dup(de)) { 1548 /* Remove duphead if de was last entry. */ 1549 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1550 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1551 MPASS(tmpfs_dirent_duphead(xde)); 1552 } else 1553 xde = NULL; 1554 LIST_REMOVE(de, uh.td_dup.entries); 1555 LIST_REMOVE(de, uh.td_dup.index_entries); 1556 if (xde != NULL) { 1557 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1558 RB_REMOVE(tmpfs_dir, head, xde); 1559 tmp = VFS_TO_TMPFS(vp->v_mount); 1560 MPASS(xde->td_node == NULL); 1561 tmpfs_free_dirent(tmp, xde); 1562 } 1563 } 1564 de->td_cookie = de->td_hash; 1565 } else 1566 RB_REMOVE(tmpfs_dir, head, de); 1567 1568 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1569 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1570 dnode->tn_accessed = true; 1571 tmpfs_update(vp); 1572 } 1573 1574 void 1575 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1576 { 1577 struct tmpfs_dirent *de, *dde, *nde; 1578 1579 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1580 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1581 /* Node may already be destroyed. */ 1582 de->td_node = NULL; 1583 if (tmpfs_dirent_duphead(de)) { 1584 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1585 LIST_REMOVE(dde, uh.td_dup.entries); 1586 dde->td_node = NULL; 1587 tmpfs_free_dirent(tmp, dde); 1588 } 1589 } 1590 tmpfs_free_dirent(tmp, de); 1591 } 1592 } 1593 1594 /* 1595 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1596 * directory and returns it in the uio space. The function returns 0 1597 * on success, -1 if there was not enough space in the uio structure to 1598 * hold the directory entry or an appropriate error code if another 1599 * error happens. 1600 */ 1601 static int 1602 tmpfs_dir_getdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node, 1603 struct uio *uio) 1604 { 1605 int error; 1606 struct dirent dent; 1607 1608 TMPFS_VALIDATE_DIR(node); 1609 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1610 1611 dent.d_fileno = node->tn_id; 1612 dent.d_off = TMPFS_DIRCOOKIE_DOTDOT; 1613 dent.d_type = DT_DIR; 1614 dent.d_namlen = 1; 1615 dent.d_name[0] = '.'; 1616 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1617 dirent_terminate(&dent); 1618 1619 if (dent.d_reclen > uio->uio_resid) 1620 error = EJUSTRETURN; 1621 else 1622 error = uiomove(&dent, dent.d_reclen, uio); 1623 1624 tmpfs_set_accessed(tm, node); 1625 1626 return (error); 1627 } 1628 1629 /* 1630 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1631 * directory and returns it in the uio space. The function returns 0 1632 * on success, -1 if there was not enough space in the uio structure to 1633 * hold the directory entry or an appropriate error code if another 1634 * error happens. 1635 */ 1636 static int 1637 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node, 1638 struct uio *uio, off_t next) 1639 { 1640 struct tmpfs_node *parent; 1641 struct dirent dent; 1642 int error; 1643 1644 TMPFS_VALIDATE_DIR(node); 1645 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1646 1647 /* 1648 * Return ENOENT if the current node is already removed. 1649 */ 1650 TMPFS_ASSERT_LOCKED(node); 1651 parent = node->tn_dir.tn_parent; 1652 if (parent == NULL) 1653 return (ENOENT); 1654 1655 dent.d_fileno = parent->tn_id; 1656 dent.d_off = next; 1657 dent.d_type = DT_DIR; 1658 dent.d_namlen = 2; 1659 dent.d_name[0] = '.'; 1660 dent.d_name[1] = '.'; 1661 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1662 dirent_terminate(&dent); 1663 1664 if (dent.d_reclen > uio->uio_resid) 1665 error = EJUSTRETURN; 1666 else 1667 error = uiomove(&dent, dent.d_reclen, uio); 1668 1669 tmpfs_set_accessed(tm, node); 1670 1671 return (error); 1672 } 1673 1674 /* 1675 * Helper function for tmpfs_readdir. Returns as much directory entries 1676 * as can fit in the uio space. The read starts at uio->uio_offset. 1677 * The function returns 0 on success, -1 if there was not enough space 1678 * in the uio structure to hold the directory entry or an appropriate 1679 * error code if another error happens. 1680 */ 1681 int 1682 tmpfs_dir_getdents(struct tmpfs_mount *tm, struct tmpfs_node *node, 1683 struct uio *uio, int maxcookies, uint64_t *cookies, int *ncookies) 1684 { 1685 struct tmpfs_dir_cursor dc; 1686 struct tmpfs_dirent *de, *nde; 1687 off_t off; 1688 int error; 1689 1690 TMPFS_VALIDATE_DIR(node); 1691 1692 off = 0; 1693 1694 /* 1695 * Lookup the node from the current offset. The starting offset of 1696 * 0 will lookup both '.' and '..', and then the first real entry, 1697 * or EOF if there are none. Then find all entries for the dir that 1698 * fit into the buffer. Once no more entries are found (de == NULL), 1699 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1700 * call to return 0. 1701 */ 1702 switch (uio->uio_offset) { 1703 case TMPFS_DIRCOOKIE_DOT: 1704 error = tmpfs_dir_getdotdent(tm, node, uio); 1705 if (error != 0) 1706 return (error); 1707 uio->uio_offset = off = TMPFS_DIRCOOKIE_DOTDOT; 1708 if (cookies != NULL) 1709 cookies[(*ncookies)++] = off; 1710 /* FALLTHROUGH */ 1711 case TMPFS_DIRCOOKIE_DOTDOT: 1712 de = tmpfs_dir_first(node, &dc); 1713 off = tmpfs_dirent_cookie(de); 1714 error = tmpfs_dir_getdotdotdent(tm, node, uio, off); 1715 if (error != 0) 1716 return (error); 1717 uio->uio_offset = off; 1718 if (cookies != NULL) 1719 cookies[(*ncookies)++] = off; 1720 /* EOF. */ 1721 if (de == NULL) 1722 return (0); 1723 break; 1724 case TMPFS_DIRCOOKIE_EOF: 1725 return (0); 1726 default: 1727 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1728 if (de == NULL) 1729 return (EINVAL); 1730 if (cookies != NULL) 1731 off = tmpfs_dirent_cookie(de); 1732 } 1733 1734 /* 1735 * Read as much entries as possible; i.e., until we reach the end of the 1736 * directory or we exhaust uio space. 1737 */ 1738 do { 1739 struct dirent d; 1740 1741 /* 1742 * Create a dirent structure representing the current tmpfs_node 1743 * and fill it. 1744 */ 1745 if (de->td_node == NULL) { 1746 d.d_fileno = 1; 1747 d.d_type = DT_WHT; 1748 } else { 1749 d.d_fileno = de->td_node->tn_id; 1750 switch (de->td_node->tn_type) { 1751 case VBLK: 1752 d.d_type = DT_BLK; 1753 break; 1754 1755 case VCHR: 1756 d.d_type = DT_CHR; 1757 break; 1758 1759 case VDIR: 1760 d.d_type = DT_DIR; 1761 break; 1762 1763 case VFIFO: 1764 d.d_type = DT_FIFO; 1765 break; 1766 1767 case VLNK: 1768 d.d_type = DT_LNK; 1769 break; 1770 1771 case VREG: 1772 d.d_type = DT_REG; 1773 break; 1774 1775 case VSOCK: 1776 d.d_type = DT_SOCK; 1777 break; 1778 1779 default: 1780 panic("tmpfs_dir_getdents: type %p %d", 1781 de->td_node, (int)de->td_node->tn_type); 1782 } 1783 } 1784 d.d_namlen = de->td_namelen; 1785 MPASS(de->td_namelen < sizeof(d.d_name)); 1786 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1787 d.d_reclen = GENERIC_DIRSIZ(&d); 1788 1789 /* 1790 * Stop reading if the directory entry we are treating is bigger 1791 * than the amount of data that can be returned. 1792 */ 1793 if (d.d_reclen > uio->uio_resid) { 1794 error = EJUSTRETURN; 1795 break; 1796 } 1797 1798 nde = tmpfs_dir_next(node, &dc); 1799 d.d_off = tmpfs_dirent_cookie(nde); 1800 dirent_terminate(&d); 1801 1802 /* 1803 * Copy the new dirent structure into the output buffer and 1804 * advance pointers. 1805 */ 1806 error = uiomove(&d, d.d_reclen, uio); 1807 if (error == 0) { 1808 de = nde; 1809 if (cookies != NULL) { 1810 off = tmpfs_dirent_cookie(de); 1811 MPASS(*ncookies < maxcookies); 1812 cookies[(*ncookies)++] = off; 1813 } 1814 } 1815 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1816 1817 /* Skip setting off when using cookies as it is already done above. */ 1818 if (cookies == NULL) 1819 off = tmpfs_dirent_cookie(de); 1820 1821 /* Update the offset and cache. */ 1822 uio->uio_offset = off; 1823 node->tn_dir.tn_readdir_lastn = off; 1824 node->tn_dir.tn_readdir_lastp = de; 1825 1826 tmpfs_set_accessed(tm, node); 1827 return (error); 1828 } 1829 1830 int 1831 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1832 { 1833 struct tmpfs_dirent *de; 1834 int error; 1835 1836 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1837 cnp->cn_nameptr, cnp->cn_namelen, &de); 1838 if (error != 0) 1839 return (error); 1840 tmpfs_dir_attach(dvp, de); 1841 return (0); 1842 } 1843 1844 void 1845 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1846 { 1847 struct tmpfs_dirent *de; 1848 1849 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1850 MPASS(de != NULL && de->td_node == NULL); 1851 tmpfs_dir_detach(dvp, de); 1852 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1853 } 1854 1855 /* 1856 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1857 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1858 * 'newsize' must be positive. 1859 * 1860 * Returns zero on success or an appropriate error code on failure. 1861 */ 1862 int 1863 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1864 { 1865 struct tmpfs_node *node; 1866 vm_object_t uobj; 1867 vm_pindex_t idx, newpages, oldpages; 1868 off_t oldsize; 1869 int base, error; 1870 1871 MPASS(vp->v_type == VREG); 1872 MPASS(newsize >= 0); 1873 1874 node = VP_TO_TMPFS_NODE(vp); 1875 uobj = node->tn_reg.tn_aobj; 1876 1877 /* 1878 * Convert the old and new sizes to the number of pages needed to 1879 * store them. It may happen that we do not need to do anything 1880 * because the last allocated page can accommodate the change on 1881 * its own. 1882 */ 1883 oldsize = node->tn_size; 1884 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1885 MPASS(oldpages == uobj->size); 1886 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1887 1888 if (__predict_true(newpages == oldpages && newsize >= oldsize)) { 1889 node->tn_size = newsize; 1890 return (0); 1891 } 1892 1893 VM_OBJECT_WLOCK(uobj); 1894 if (newsize < oldsize) { 1895 /* 1896 * Zero the truncated part of the last page. 1897 */ 1898 base = newsize & PAGE_MASK; 1899 if (base != 0) { 1900 idx = OFF_TO_IDX(newsize); 1901 error = tmpfs_partial_page_invalidate(uobj, idx, base, 1902 PAGE_SIZE, ignerr); 1903 if (error != 0) { 1904 VM_OBJECT_WUNLOCK(uobj); 1905 return (error); 1906 } 1907 } 1908 1909 /* 1910 * Release any swap space and free any whole pages. 1911 */ 1912 if (newpages < oldpages) 1913 vm_object_page_remove(uobj, newpages, 0, 0); 1914 } 1915 uobj->size = newpages; 1916 VM_OBJECT_WUNLOCK(uobj); 1917 1918 node->tn_size = newsize; 1919 return (0); 1920 } 1921 1922 /* 1923 * Punch hole in the aobj associated with the regular file pointed to by 'vp'. 1924 * Requests completely beyond the end-of-file are converted to no-op. 1925 * 1926 * Returns 0 on success or error code from tmpfs_partial_page_invalidate() on 1927 * failure. 1928 */ 1929 int 1930 tmpfs_reg_punch_hole(struct vnode *vp, off_t *offset, off_t *length) 1931 { 1932 struct tmpfs_node *node; 1933 vm_object_t object; 1934 vm_pindex_t pistart, pi, piend; 1935 int startofs, endofs, end; 1936 off_t off, len; 1937 int error; 1938 1939 KASSERT(*length <= OFF_MAX - *offset, ("%s: offset + length overflows", 1940 __func__)); 1941 node = VP_TO_TMPFS_NODE(vp); 1942 KASSERT(node->tn_type == VREG, ("%s: node is not regular file", 1943 __func__)); 1944 object = node->tn_reg.tn_aobj; 1945 off = *offset; 1946 len = omin(node->tn_size - off, *length); 1947 startofs = off & PAGE_MASK; 1948 endofs = (off + len) & PAGE_MASK; 1949 pistart = OFF_TO_IDX(off); 1950 piend = OFF_TO_IDX(off + len); 1951 pi = OFF_TO_IDX((vm_ooffset_t)off + PAGE_MASK); 1952 error = 0; 1953 1954 /* Handle the case when offset is on or beyond file size. */ 1955 if (len <= 0) { 1956 *length = 0; 1957 return (0); 1958 } 1959 1960 VM_OBJECT_WLOCK(object); 1961 1962 /* 1963 * If there is a partial page at the beginning of the hole-punching 1964 * request, fill the partial page with zeroes. 1965 */ 1966 if (startofs != 0) { 1967 end = pistart != piend ? PAGE_SIZE : endofs; 1968 error = tmpfs_partial_page_invalidate(object, pistart, startofs, 1969 end, FALSE); 1970 if (error != 0) 1971 goto out; 1972 off += end - startofs; 1973 len -= end - startofs; 1974 } 1975 1976 /* 1977 * Toss away the full pages in the affected area. 1978 */ 1979 if (pi < piend) { 1980 vm_object_page_remove(object, pi, piend, 0); 1981 off += IDX_TO_OFF(piend - pi); 1982 len -= IDX_TO_OFF(piend - pi); 1983 } 1984 1985 /* 1986 * If there is a partial page at the end of the hole-punching request, 1987 * fill the partial page with zeroes. 1988 */ 1989 if (endofs != 0 && pistart != piend) { 1990 error = tmpfs_partial_page_invalidate(object, piend, 0, endofs, 1991 FALSE); 1992 if (error != 0) 1993 goto out; 1994 off += endofs; 1995 len -= endofs; 1996 } 1997 1998 out: 1999 VM_OBJECT_WUNLOCK(object); 2000 *offset = off; 2001 *length = len; 2002 return (error); 2003 } 2004 2005 void 2006 tmpfs_check_mtime(struct vnode *vp) 2007 { 2008 struct tmpfs_node *node; 2009 struct vm_object *obj; 2010 2011 ASSERT_VOP_ELOCKED(vp, "check_mtime"); 2012 if (vp->v_type != VREG) 2013 return; 2014 obj = vp->v_object; 2015 KASSERT(obj->type == tmpfs_pager_type && 2016 (obj->flags & (OBJ_SWAP | OBJ_TMPFS)) == 2017 (OBJ_SWAP | OBJ_TMPFS), ("non-tmpfs obj")); 2018 /* unlocked read */ 2019 if (obj->generation != obj->cleangeneration) { 2020 VM_OBJECT_WLOCK(obj); 2021 if (obj->generation != obj->cleangeneration) { 2022 obj->cleangeneration = obj->generation; 2023 node = VP_TO_TMPFS_NODE(vp); 2024 node->tn_status |= TMPFS_NODE_MODIFIED | 2025 TMPFS_NODE_CHANGED; 2026 } 2027 VM_OBJECT_WUNLOCK(obj); 2028 } 2029 } 2030 2031 /* 2032 * Change flags of the given vnode. 2033 * Caller should execute tmpfs_update on vp after a successful execution. 2034 * The vnode must be locked on entry and remain locked on exit. 2035 */ 2036 int 2037 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 2038 struct thread *td) 2039 { 2040 int error; 2041 struct tmpfs_node *node; 2042 2043 ASSERT_VOP_ELOCKED(vp, "chflags"); 2044 2045 node = VP_TO_TMPFS_NODE(vp); 2046 2047 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 2048 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 2049 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 2050 UF_SPARSE | UF_SYSTEM)) != 0) 2051 return (EOPNOTSUPP); 2052 2053 /* Disallow this operation if the file system is mounted read-only. */ 2054 if (vp->v_mount->mnt_flag & MNT_RDONLY) 2055 return (EROFS); 2056 2057 /* 2058 * Callers may only modify the file flags on objects they 2059 * have VADMIN rights for. 2060 */ 2061 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 2062 return (error); 2063 /* 2064 * Unprivileged processes are not permitted to unset system 2065 * flags, or modify flags if any system flags are set. 2066 */ 2067 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) { 2068 if (node->tn_flags & 2069 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 2070 error = securelevel_gt(cred, 0); 2071 if (error) 2072 return (error); 2073 } 2074 } else { 2075 if (node->tn_flags & 2076 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 2077 ((flags ^ node->tn_flags) & SF_SETTABLE)) 2078 return (EPERM); 2079 } 2080 node->tn_flags = flags; 2081 node->tn_status |= TMPFS_NODE_CHANGED; 2082 2083 ASSERT_VOP_ELOCKED(vp, "chflags2"); 2084 2085 return (0); 2086 } 2087 2088 /* 2089 * Change access mode on the given vnode. 2090 * Caller should execute tmpfs_update on vp after a successful execution. 2091 * The vnode must be locked on entry and remain locked on exit. 2092 */ 2093 int 2094 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, 2095 struct thread *td) 2096 { 2097 int error; 2098 struct tmpfs_node *node; 2099 mode_t newmode; 2100 2101 ASSERT_VOP_ELOCKED(vp, "chmod"); 2102 ASSERT_VOP_IN_SEQC(vp); 2103 2104 node = VP_TO_TMPFS_NODE(vp); 2105 2106 /* Disallow this operation if the file system is mounted read-only. */ 2107 if (vp->v_mount->mnt_flag & MNT_RDONLY) 2108 return (EROFS); 2109 2110 /* Immutable or append-only files cannot be modified, either. */ 2111 if (node->tn_flags & (IMMUTABLE | APPEND)) 2112 return (EPERM); 2113 2114 /* 2115 * To modify the permissions on a file, must possess VADMIN 2116 * for that file. 2117 */ 2118 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 2119 return (error); 2120 2121 /* 2122 * Privileged processes may set the sticky bit on non-directories, 2123 * as well as set the setgid bit on a file with a group that the 2124 * process is not a member of. 2125 */ 2126 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 2127 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE)) 2128 return (EFTYPE); 2129 } 2130 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 2131 error = priv_check_cred(cred, PRIV_VFS_SETGID); 2132 if (error) 2133 return (error); 2134 } 2135 2136 newmode = node->tn_mode & ~ALLPERMS; 2137 newmode |= mode & ALLPERMS; 2138 atomic_store_short(&node->tn_mode, newmode); 2139 2140 node->tn_status |= TMPFS_NODE_CHANGED; 2141 2142 ASSERT_VOP_ELOCKED(vp, "chmod2"); 2143 2144 return (0); 2145 } 2146 2147 /* 2148 * Change ownership of the given vnode. At least one of uid or gid must 2149 * be different than VNOVAL. If one is set to that value, the attribute 2150 * is unchanged. 2151 * Caller should execute tmpfs_update on vp after a successful execution. 2152 * The vnode must be locked on entry and remain locked on exit. 2153 */ 2154 int 2155 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 2156 struct thread *td) 2157 { 2158 int error; 2159 struct tmpfs_node *node; 2160 uid_t ouid; 2161 gid_t ogid; 2162 mode_t newmode; 2163 2164 ASSERT_VOP_ELOCKED(vp, "chown"); 2165 ASSERT_VOP_IN_SEQC(vp); 2166 2167 node = VP_TO_TMPFS_NODE(vp); 2168 2169 /* Assign default values if they are unknown. */ 2170 MPASS(uid != VNOVAL || gid != VNOVAL); 2171 if (uid == VNOVAL) 2172 uid = node->tn_uid; 2173 if (gid == VNOVAL) 2174 gid = node->tn_gid; 2175 MPASS(uid != VNOVAL && gid != VNOVAL); 2176 2177 /* Disallow this operation if the file system is mounted read-only. */ 2178 if (vp->v_mount->mnt_flag & MNT_RDONLY) 2179 return (EROFS); 2180 2181 /* Immutable or append-only files cannot be modified, either. */ 2182 if (node->tn_flags & (IMMUTABLE | APPEND)) 2183 return (EPERM); 2184 2185 /* 2186 * To modify the ownership of a file, must possess VADMIN for that 2187 * file. 2188 */ 2189 if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) 2190 return (error); 2191 2192 /* 2193 * To change the owner of a file, or change the group of a file to a 2194 * group of which we are not a member, the caller must have 2195 * privilege. 2196 */ 2197 if ((uid != node->tn_uid || 2198 (gid != node->tn_gid && !groupmember(gid, cred))) && 2199 (error = priv_check_cred(cred, PRIV_VFS_CHOWN))) 2200 return (error); 2201 2202 ogid = node->tn_gid; 2203 ouid = node->tn_uid; 2204 2205 node->tn_uid = uid; 2206 node->tn_gid = gid; 2207 2208 node->tn_status |= TMPFS_NODE_CHANGED; 2209 2210 if ((node->tn_mode & (S_ISUID | S_ISGID)) != 0 && 2211 (ouid != uid || ogid != gid)) { 2212 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) { 2213 newmode = node->tn_mode & ~(S_ISUID | S_ISGID); 2214 atomic_store_short(&node->tn_mode, newmode); 2215 } 2216 } 2217 2218 ASSERT_VOP_ELOCKED(vp, "chown2"); 2219 2220 return (0); 2221 } 2222 2223 /* 2224 * Change size of the given vnode. 2225 * Caller should execute tmpfs_update on vp after a successful execution. 2226 * The vnode must be locked on entry and remain locked on exit. 2227 */ 2228 int 2229 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 2230 struct thread *td) 2231 { 2232 int error; 2233 struct tmpfs_node *node; 2234 2235 ASSERT_VOP_ELOCKED(vp, "chsize"); 2236 2237 node = VP_TO_TMPFS_NODE(vp); 2238 2239 /* Decide whether this is a valid operation based on the file type. */ 2240 error = 0; 2241 switch (vp->v_type) { 2242 case VDIR: 2243 return (EISDIR); 2244 2245 case VREG: 2246 if (vp->v_mount->mnt_flag & MNT_RDONLY) 2247 return (EROFS); 2248 break; 2249 2250 case VBLK: 2251 /* FALLTHROUGH */ 2252 case VCHR: 2253 /* FALLTHROUGH */ 2254 case VFIFO: 2255 /* 2256 * Allow modifications of special files even if in the file 2257 * system is mounted read-only (we are not modifying the 2258 * files themselves, but the objects they represent). 2259 */ 2260 return (0); 2261 2262 default: 2263 /* Anything else is unsupported. */ 2264 return (EOPNOTSUPP); 2265 } 2266 2267 /* Immutable or append-only files cannot be modified, either. */ 2268 if (node->tn_flags & (IMMUTABLE | APPEND)) 2269 return (EPERM); 2270 2271 error = vn_rlimit_trunc(size, td); 2272 if (error != 0) 2273 return (error); 2274 2275 error = tmpfs_truncate(vp, size); 2276 /* 2277 * tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 2278 * for us, as will update tn_status; no need to do that here. 2279 */ 2280 2281 ASSERT_VOP_ELOCKED(vp, "chsize2"); 2282 2283 return (error); 2284 } 2285 2286 /* 2287 * Change access and modification times of the given vnode. 2288 * Caller should execute tmpfs_update on vp after a successful execution. 2289 * The vnode must be locked on entry and remain locked on exit. 2290 */ 2291 int 2292 tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 2293 struct ucred *cred, struct thread *td) 2294 { 2295 int error; 2296 struct tmpfs_node *node; 2297 2298 ASSERT_VOP_ELOCKED(vp, "chtimes"); 2299 2300 node = VP_TO_TMPFS_NODE(vp); 2301 2302 /* Disallow this operation if the file system is mounted read-only. */ 2303 if (vp->v_mount->mnt_flag & MNT_RDONLY) 2304 return (EROFS); 2305 2306 /* Immutable or append-only files cannot be modified, either. */ 2307 if (node->tn_flags & (IMMUTABLE | APPEND)) 2308 return (EPERM); 2309 2310 error = vn_utimes_perm(vp, vap, cred, td); 2311 if (error != 0) 2312 return (error); 2313 2314 if (vap->va_atime.tv_sec != VNOVAL) 2315 node->tn_accessed = true; 2316 if (vap->va_mtime.tv_sec != VNOVAL) 2317 node->tn_status |= TMPFS_NODE_MODIFIED; 2318 if (vap->va_birthtime.tv_sec != VNOVAL) 2319 node->tn_status |= TMPFS_NODE_MODIFIED; 2320 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 2321 if (vap->va_birthtime.tv_sec != VNOVAL) 2322 node->tn_birthtime = vap->va_birthtime; 2323 ASSERT_VOP_ELOCKED(vp, "chtimes2"); 2324 2325 return (0); 2326 } 2327 2328 void 2329 tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, int status) 2330 { 2331 2332 if ((node->tn_status & status) == status || tm->tm_ronly) 2333 return; 2334 TMPFS_NODE_LOCK(node); 2335 node->tn_status |= status; 2336 TMPFS_NODE_UNLOCK(node); 2337 } 2338 2339 void 2340 tmpfs_set_accessed(struct tmpfs_mount *tm, struct tmpfs_node *node) 2341 { 2342 if (node->tn_accessed || tm->tm_ronly) 2343 return; 2344 atomic_store_8(&node->tn_accessed, true); 2345 } 2346 2347 /* Sync timestamps */ 2348 void 2349 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 2350 const struct timespec *mod) 2351 { 2352 struct tmpfs_node *node; 2353 struct timespec now; 2354 2355 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes"); 2356 node = VP_TO_TMPFS_NODE(vp); 2357 2358 if (!node->tn_accessed && 2359 (node->tn_status & (TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED)) == 0) 2360 return; 2361 2362 vfs_timestamp(&now); 2363 TMPFS_NODE_LOCK(node); 2364 if (node->tn_accessed) { 2365 if (acc == NULL) 2366 acc = &now; 2367 node->tn_atime = *acc; 2368 } 2369 if (node->tn_status & TMPFS_NODE_MODIFIED) { 2370 if (mod == NULL) 2371 mod = &now; 2372 node->tn_mtime = *mod; 2373 } 2374 if (node->tn_status & TMPFS_NODE_CHANGED) 2375 node->tn_ctime = now; 2376 node->tn_status &= ~(TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 2377 node->tn_accessed = false; 2378 TMPFS_NODE_UNLOCK(node); 2379 2380 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */ 2381 random_harvest_queue(node, sizeof(*node), RANDOM_FS_ATIME); 2382 } 2383 2384 int 2385 tmpfs_truncate(struct vnode *vp, off_t length) 2386 { 2387 struct tmpfs_node *node; 2388 int error; 2389 2390 if (length < 0) 2391 return (EINVAL); 2392 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 2393 return (EFBIG); 2394 2395 node = VP_TO_TMPFS_NODE(vp); 2396 error = node->tn_size == length ? 0 : tmpfs_reg_resize(vp, length, 2397 FALSE); 2398 if (error == 0) 2399 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 2400 tmpfs_update(vp); 2401 2402 return (error); 2403 } 2404 2405 static __inline int 2406 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 2407 { 2408 if (a->td_hash > b->td_hash) 2409 return (1); 2410 else if (a->td_hash < b->td_hash) 2411 return (-1); 2412 return (0); 2413 } 2414 2415 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 2416