1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 2005 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 9 * 2005 program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Efficient memory file system supporting functions. 35 */ 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include <sys/param.h> 40 #include <sys/fnv_hash.h> 41 #include <sys/lock.h> 42 #include <sys/namei.h> 43 #include <sys/priv.h> 44 #include <sys/proc.h> 45 #include <sys/random.h> 46 #include <sys/rwlock.h> 47 #include <sys/stat.h> 48 #include <sys/systm.h> 49 #include <sys/sysctl.h> 50 #include <sys/vnode.h> 51 #include <sys/vmmeter.h> 52 53 #include <vm/vm.h> 54 #include <vm/vm_param.h> 55 #include <vm/vm_object.h> 56 #include <vm/vm_page.h> 57 #include <vm/vm_pageout.h> 58 #include <vm/vm_pager.h> 59 #include <vm/vm_extern.h> 60 61 #include <fs/tmpfs/tmpfs.h> 62 #include <fs/tmpfs/tmpfs_fifoops.h> 63 #include <fs/tmpfs/tmpfs_vnops.h> 64 65 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system"); 66 67 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED; 68 69 static int 70 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS) 71 { 72 int error; 73 long pages, bytes; 74 75 pages = *(long *)arg1; 76 bytes = pages * PAGE_SIZE; 77 78 error = sysctl_handle_long(oidp, &bytes, 0, req); 79 if (error || !req->newptr) 80 return (error); 81 82 pages = bytes / PAGE_SIZE; 83 if (pages < TMPFS_PAGES_MINRESERVED) 84 return (EINVAL); 85 86 *(long *)arg1 = pages; 87 return (0); 88 } 89 90 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW, 91 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L", 92 "Amount of available memory and swap below which tmpfs growth stops"); 93 94 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a, 95 struct tmpfs_dirent *b); 96 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 97 98 size_t 99 tmpfs_mem_avail(void) 100 { 101 vm_ooffset_t avail; 102 103 avail = swap_pager_avail + vm_cnt.v_free_count - tmpfs_pages_reserved; 104 if (__predict_false(avail < 0)) 105 avail = 0; 106 return (avail); 107 } 108 109 size_t 110 tmpfs_pages_used(struct tmpfs_mount *tmp) 111 { 112 const size_t node_size = sizeof(struct tmpfs_node) + 113 sizeof(struct tmpfs_dirent); 114 size_t meta_pages; 115 116 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size, 117 PAGE_SIZE); 118 return (meta_pages + tmp->tm_pages_used); 119 } 120 121 static size_t 122 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages) 123 { 124 if (tmpfs_mem_avail() < req_pages) 125 return (0); 126 127 if (tmp->tm_pages_max != ULONG_MAX && 128 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp)) 129 return (0); 130 131 return (1); 132 } 133 134 void 135 tmpfs_ref_node(struct tmpfs_node *node) 136 { 137 138 TMPFS_NODE_LOCK(node); 139 tmpfs_ref_node_locked(node); 140 TMPFS_NODE_UNLOCK(node); 141 } 142 143 void 144 tmpfs_ref_node_locked(struct tmpfs_node *node) 145 { 146 147 TMPFS_NODE_ASSERT_LOCKED(node); 148 KASSERT(node->tn_refcount > 0, ("node %p zero refcount", node)); 149 KASSERT(node->tn_refcount < UINT_MAX, ("node %p refcount %u", node, 150 node->tn_refcount)); 151 node->tn_refcount++; 152 } 153 154 /* 155 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 156 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 157 * using the credentials of the process 'p'. 158 * 159 * If the node type is set to 'VDIR', then the parent parameter must point 160 * to the parent directory of the node being created. It may only be NULL 161 * while allocating the root node. 162 * 163 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 164 * specifies the device the node represents. 165 * 166 * If the node type is set to 'VLNK', then the parameter target specifies 167 * the file name of the target file for the symbolic link that is being 168 * created. 169 * 170 * Note that new nodes are retrieved from the available list if it has 171 * items or, if it is empty, from the node pool as long as there is enough 172 * space to create them. 173 * 174 * Returns zero on success or an appropriate error code on failure. 175 */ 176 int 177 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type, 178 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 179 char *target, dev_t rdev, struct tmpfs_node **node) 180 { 181 struct tmpfs_node *nnode; 182 vm_object_t obj; 183 184 /* If the root directory of the 'tmp' file system is not yet 185 * allocated, this must be the request to do it. */ 186 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 187 KASSERT(tmp->tm_root == NULL || mp->mnt_writeopcount > 0, 188 ("creating node not under vn_start_write")); 189 190 MPASS(IFF(type == VLNK, target != NULL)); 191 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL)); 192 193 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 194 return (ENOSPC); 195 if (tmpfs_pages_check_avail(tmp, 1) == 0) 196 return (ENOSPC); 197 198 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 199 /* 200 * When a new tmpfs node is created for fully 201 * constructed mount point, there must be a parent 202 * node, which vnode is locked exclusively. As 203 * consequence, if the unmount is executing in 204 * parallel, vflush() cannot reclaim the parent vnode. 205 * Due to this, the check for MNTK_UNMOUNT flag is not 206 * racy: if we did not see MNTK_UNMOUNT flag, then tmp 207 * cannot be destroyed until node construction is 208 * finished and the parent vnode unlocked. 209 * 210 * Tmpfs does not need to instantiate new nodes during 211 * unmount. 212 */ 213 return (EBUSY); 214 } 215 216 nnode = (struct tmpfs_node *)uma_zalloc_arg(tmp->tm_node_pool, tmp, 217 M_WAITOK); 218 219 /* Generic initialization. */ 220 nnode->tn_type = type; 221 vfs_timestamp(&nnode->tn_atime); 222 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 223 nnode->tn_atime; 224 nnode->tn_uid = uid; 225 nnode->tn_gid = gid; 226 nnode->tn_mode = mode; 227 nnode->tn_id = alloc_unr(tmp->tm_ino_unr); 228 nnode->tn_refcount = 1; 229 230 /* Type-specific initialization. */ 231 switch (nnode->tn_type) { 232 case VBLK: 233 case VCHR: 234 nnode->tn_rdev = rdev; 235 break; 236 237 case VDIR: 238 RB_INIT(&nnode->tn_dir.tn_dirhead); 239 LIST_INIT(&nnode->tn_dir.tn_dupindex); 240 MPASS(parent != nnode); 241 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL)); 242 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent; 243 nnode->tn_dir.tn_readdir_lastn = 0; 244 nnode->tn_dir.tn_readdir_lastp = NULL; 245 nnode->tn_links++; 246 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent); 247 nnode->tn_dir.tn_parent->tn_links++; 248 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent); 249 break; 250 251 case VFIFO: 252 /* FALLTHROUGH */ 253 case VSOCK: 254 break; 255 256 case VLNK: 257 MPASS(strlen(target) < MAXPATHLEN); 258 nnode->tn_size = strlen(target); 259 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME, 260 M_WAITOK); 261 memcpy(nnode->tn_link, target, nnode->tn_size); 262 break; 263 264 case VREG: 265 obj = nnode->tn_reg.tn_aobj = 266 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0, 267 NULL /* XXXKIB - tmpfs needs swap reservation */); 268 VM_OBJECT_WLOCK(obj); 269 /* OBJ_TMPFS is set together with the setting of vp->v_object */ 270 vm_object_set_flag(obj, OBJ_NOSPLIT | OBJ_TMPFS_NODE); 271 vm_object_clear_flag(obj, OBJ_ONEMAPPING); 272 VM_OBJECT_WUNLOCK(obj); 273 break; 274 275 default: 276 panic("tmpfs_alloc_node: type %p %d", nnode, 277 (int)nnode->tn_type); 278 } 279 280 TMPFS_LOCK(tmp); 281 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 282 nnode->tn_attached = true; 283 tmp->tm_nodes_inuse++; 284 tmp->tm_refcount++; 285 TMPFS_UNLOCK(tmp); 286 287 *node = nnode; 288 return (0); 289 } 290 291 /* 292 * Destroys the node pointed to by node from the file system 'tmp'. 293 * If the node references a directory, no entries are allowed. 294 */ 295 void 296 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 297 { 298 299 TMPFS_LOCK(tmp); 300 TMPFS_NODE_LOCK(node); 301 if (!tmpfs_free_node_locked(tmp, node, false)) { 302 TMPFS_NODE_UNLOCK(node); 303 TMPFS_UNLOCK(tmp); 304 } 305 } 306 307 bool 308 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node, 309 bool detach) 310 { 311 vm_object_t uobj; 312 313 TMPFS_MP_ASSERT_LOCKED(tmp); 314 TMPFS_NODE_ASSERT_LOCKED(node); 315 KASSERT(node->tn_refcount > 0, ("node %p refcount zero", node)); 316 317 node->tn_refcount--; 318 if (node->tn_attached && (detach || node->tn_refcount == 0)) { 319 MPASS(tmp->tm_nodes_inuse > 0); 320 tmp->tm_nodes_inuse--; 321 LIST_REMOVE(node, tn_entries); 322 node->tn_attached = false; 323 } 324 if (node->tn_refcount > 0) 325 return (false); 326 327 #ifdef INVARIANTS 328 MPASS(node->tn_vnode == NULL); 329 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 330 #endif 331 TMPFS_NODE_UNLOCK(node); 332 TMPFS_UNLOCK(tmp); 333 334 switch (node->tn_type) { 335 case VBLK: 336 /* FALLTHROUGH */ 337 case VCHR: 338 /* FALLTHROUGH */ 339 case VDIR: 340 /* FALLTHROUGH */ 341 case VFIFO: 342 /* FALLTHROUGH */ 343 case VSOCK: 344 break; 345 346 case VLNK: 347 free(node->tn_link, M_TMPFSNAME); 348 break; 349 350 case VREG: 351 uobj = node->tn_reg.tn_aobj; 352 if (uobj != NULL) { 353 if (uobj->size != 0) 354 atomic_subtract_long(&tmp->tm_pages_used, uobj->size); 355 KASSERT((uobj->flags & OBJ_TMPFS) == 0, 356 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj)); 357 vm_object_deallocate(uobj); 358 } 359 break; 360 361 default: 362 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 363 } 364 365 free_unr(tmp->tm_ino_unr, node->tn_id); 366 uma_zfree(tmp->tm_node_pool, node); 367 TMPFS_LOCK(tmp); 368 tmpfs_free_tmp(tmp); 369 return (true); 370 } 371 372 static __inline uint32_t 373 tmpfs_dirent_hash(const char *name, u_int len) 374 { 375 uint32_t hash; 376 377 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK; 378 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP 379 hash &= 0xf; 380 #endif 381 if (hash < TMPFS_DIRCOOKIE_MIN) 382 hash += TMPFS_DIRCOOKIE_MIN; 383 384 return (hash); 385 } 386 387 static __inline off_t 388 tmpfs_dirent_cookie(struct tmpfs_dirent *de) 389 { 390 if (de == NULL) 391 return (TMPFS_DIRCOOKIE_EOF); 392 393 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN); 394 395 return (de->td_cookie); 396 } 397 398 static __inline boolean_t 399 tmpfs_dirent_dup(struct tmpfs_dirent *de) 400 { 401 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0); 402 } 403 404 static __inline boolean_t 405 tmpfs_dirent_duphead(struct tmpfs_dirent *de) 406 { 407 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0); 408 } 409 410 void 411 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen) 412 { 413 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen); 414 memcpy(de->ud.td_name, name, namelen); 415 de->td_namelen = namelen; 416 } 417 418 /* 419 * Allocates a new directory entry for the node node with a name of name. 420 * The new directory entry is returned in *de. 421 * 422 * The link count of node is increased by one to reflect the new object 423 * referencing it. 424 * 425 * Returns zero on success or an appropriate error code on failure. 426 */ 427 int 428 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 429 const char *name, u_int len, struct tmpfs_dirent **de) 430 { 431 struct tmpfs_dirent *nde; 432 433 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK); 434 nde->td_node = node; 435 if (name != NULL) { 436 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 437 tmpfs_dirent_init(nde, name, len); 438 } else 439 nde->td_namelen = 0; 440 if (node != NULL) 441 node->tn_links++; 442 443 *de = nde; 444 445 return 0; 446 } 447 448 /* 449 * Frees a directory entry. It is the caller's responsibility to destroy 450 * the node referenced by it if needed. 451 * 452 * The link count of node is decreased by one to reflect the removal of an 453 * object that referenced it. This only happens if 'node_exists' is true; 454 * otherwise the function will not access the node referred to by the 455 * directory entry, as it may already have been released from the outside. 456 */ 457 void 458 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 459 { 460 struct tmpfs_node *node; 461 462 node = de->td_node; 463 if (node != NULL) { 464 MPASS(node->tn_links > 0); 465 node->tn_links--; 466 } 467 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL) 468 free(de->ud.td_name, M_TMPFSNAME); 469 uma_zfree(tmp->tm_dirent_pool, de); 470 } 471 472 void 473 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj) 474 { 475 476 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject"); 477 if (vp->v_type != VREG || obj == NULL) 478 return; 479 480 VM_OBJECT_WLOCK(obj); 481 VI_LOCK(vp); 482 vm_object_clear_flag(obj, OBJ_TMPFS); 483 obj->un_pager.swp.swp_tmpfs = NULL; 484 VI_UNLOCK(vp); 485 VM_OBJECT_WUNLOCK(obj); 486 } 487 488 /* 489 * Need to clear v_object for insmntque failure. 490 */ 491 static void 492 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg) 493 { 494 495 tmpfs_destroy_vobject(vp, vp->v_object); 496 vp->v_object = NULL; 497 vp->v_data = NULL; 498 vp->v_op = &dead_vnodeops; 499 vgone(vp); 500 vput(vp); 501 } 502 503 /* 504 * Allocates a new vnode for the node node or returns a new reference to 505 * an existing one if the node had already a vnode referencing it. The 506 * resulting locked vnode is returned in *vpp. 507 * 508 * Returns zero on success or an appropriate error code on failure. 509 */ 510 int 511 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 512 struct vnode **vpp) 513 { 514 struct vnode *vp; 515 struct tmpfs_mount *tm; 516 vm_object_t object; 517 int error; 518 519 error = 0; 520 tm = VFS_TO_TMPFS(mp); 521 TMPFS_NODE_LOCK(node); 522 tmpfs_ref_node_locked(node); 523 loop: 524 TMPFS_NODE_ASSERT_LOCKED(node); 525 if ((vp = node->tn_vnode) != NULL) { 526 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 527 VI_LOCK(vp); 528 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) || 529 ((vp->v_iflag & VI_DOOMED) != 0 && 530 (lkflag & LK_NOWAIT) != 0)) { 531 VI_UNLOCK(vp); 532 TMPFS_NODE_UNLOCK(node); 533 error = ENOENT; 534 vp = NULL; 535 goto out; 536 } 537 if ((vp->v_iflag & VI_DOOMED) != 0) { 538 VI_UNLOCK(vp); 539 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM; 540 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) { 541 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 542 0, "tmpfsE", 0); 543 } 544 goto loop; 545 } 546 TMPFS_NODE_UNLOCK(node); 547 error = vget(vp, lkflag | LK_INTERLOCK, curthread); 548 if (error == ENOENT) { 549 TMPFS_NODE_LOCK(node); 550 goto loop; 551 } 552 if (error != 0) { 553 vp = NULL; 554 goto out; 555 } 556 557 /* 558 * Make sure the vnode is still there after 559 * getting the interlock to avoid racing a free. 560 */ 561 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 562 vput(vp); 563 TMPFS_NODE_LOCK(node); 564 goto loop; 565 } 566 567 goto out; 568 } 569 570 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 571 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 572 TMPFS_NODE_UNLOCK(node); 573 error = ENOENT; 574 vp = NULL; 575 goto out; 576 } 577 578 /* 579 * otherwise lock the vp list while we call getnewvnode 580 * since that can block. 581 */ 582 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 583 node->tn_vpstate |= TMPFS_VNODE_WANT; 584 error = msleep((caddr_t) &node->tn_vpstate, 585 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0); 586 if (error != 0) 587 goto out; 588 goto loop; 589 } else 590 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 591 592 TMPFS_NODE_UNLOCK(node); 593 594 /* Get a new vnode and associate it with our node. */ 595 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ? 596 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp); 597 if (error != 0) 598 goto unlock; 599 MPASS(vp != NULL); 600 601 /* lkflag is ignored, the lock is exclusive */ 602 (void) vn_lock(vp, lkflag | LK_RETRY); 603 604 vp->v_data = node; 605 vp->v_type = node->tn_type; 606 607 /* Type-specific initialization. */ 608 switch (node->tn_type) { 609 case VBLK: 610 /* FALLTHROUGH */ 611 case VCHR: 612 /* FALLTHROUGH */ 613 case VLNK: 614 /* FALLTHROUGH */ 615 case VSOCK: 616 break; 617 case VFIFO: 618 vp->v_op = &tmpfs_fifoop_entries; 619 break; 620 case VREG: 621 object = node->tn_reg.tn_aobj; 622 VM_OBJECT_WLOCK(object); 623 VI_LOCK(vp); 624 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 625 vp->v_object = object; 626 object->un_pager.swp.swp_tmpfs = vp; 627 vm_object_set_flag(object, OBJ_TMPFS); 628 VI_UNLOCK(vp); 629 VM_OBJECT_WUNLOCK(object); 630 break; 631 case VDIR: 632 MPASS(node->tn_dir.tn_parent != NULL); 633 if (node->tn_dir.tn_parent == node) 634 vp->v_vflag |= VV_ROOT; 635 break; 636 637 default: 638 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 639 } 640 if (vp->v_type != VFIFO) 641 VN_LOCK_ASHARE(vp); 642 643 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 644 if (error != 0) 645 vp = NULL; 646 647 unlock: 648 TMPFS_NODE_LOCK(node); 649 650 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 651 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 652 node->tn_vnode = vp; 653 654 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 655 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 656 TMPFS_NODE_UNLOCK(node); 657 wakeup((caddr_t) &node->tn_vpstate); 658 } else 659 TMPFS_NODE_UNLOCK(node); 660 661 out: 662 if (error == 0) { 663 *vpp = vp; 664 665 #ifdef INVARIANTS 666 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 667 TMPFS_NODE_LOCK(node); 668 MPASS(*vpp == node->tn_vnode); 669 TMPFS_NODE_UNLOCK(node); 670 #endif 671 } 672 tmpfs_free_node(tm, node); 673 674 return (error); 675 } 676 677 /* 678 * Destroys the association between the vnode vp and the node it 679 * references. 680 */ 681 void 682 tmpfs_free_vp(struct vnode *vp) 683 { 684 struct tmpfs_node *node; 685 686 node = VP_TO_TMPFS_NODE(vp); 687 688 TMPFS_NODE_ASSERT_LOCKED(node); 689 node->tn_vnode = NULL; 690 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 691 wakeup(&node->tn_vnode); 692 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 693 vp->v_data = NULL; 694 } 695 696 /* 697 * Allocates a new file of type 'type' and adds it to the parent directory 698 * 'dvp'; this addition is done using the component name given in 'cnp'. 699 * The ownership of the new file is automatically assigned based on the 700 * credentials of the caller (through 'cnp'), the group is set based on 701 * the parent directory and the mode is determined from the 'vap' argument. 702 * If successful, *vpp holds a vnode to the newly created file and zero 703 * is returned. Otherwise *vpp is NULL and the function returns an 704 * appropriate error code. 705 */ 706 int 707 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 708 struct componentname *cnp, char *target) 709 { 710 int error; 711 struct tmpfs_dirent *de; 712 struct tmpfs_mount *tmp; 713 struct tmpfs_node *dnode; 714 struct tmpfs_node *node; 715 struct tmpfs_node *parent; 716 717 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file"); 718 MPASS(cnp->cn_flags & HASBUF); 719 720 tmp = VFS_TO_TMPFS(dvp->v_mount); 721 dnode = VP_TO_TMPFS_DIR(dvp); 722 *vpp = NULL; 723 724 /* If the entry we are creating is a directory, we cannot overflow 725 * the number of links of its parent, because it will get a new 726 * link. */ 727 if (vap->va_type == VDIR) { 728 /* Ensure that we do not overflow the maximum number of links 729 * imposed by the system. */ 730 MPASS(dnode->tn_links <= LINK_MAX); 731 if (dnode->tn_links == LINK_MAX) { 732 return (EMLINK); 733 } 734 735 parent = dnode; 736 MPASS(parent != NULL); 737 } else 738 parent = NULL; 739 740 /* Allocate a node that represents the new file. */ 741 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type, 742 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent, 743 target, vap->va_rdev, &node); 744 if (error != 0) 745 return (error); 746 747 /* Allocate a directory entry that points to the new file. */ 748 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 749 &de); 750 if (error != 0) { 751 tmpfs_free_node(tmp, node); 752 return (error); 753 } 754 755 /* Allocate a vnode for the new file. */ 756 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 757 if (error != 0) { 758 tmpfs_free_dirent(tmp, de); 759 tmpfs_free_node(tmp, node); 760 return (error); 761 } 762 763 /* Now that all required items are allocated, we can proceed to 764 * insert the new node into the directory, an operation that 765 * cannot fail. */ 766 if (cnp->cn_flags & ISWHITEOUT) 767 tmpfs_dir_whiteout_remove(dvp, cnp); 768 tmpfs_dir_attach(dvp, de); 769 return (0); 770 } 771 772 struct tmpfs_dirent * 773 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 774 { 775 struct tmpfs_dirent *de; 776 777 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 778 dc->tdc_tree = de; 779 if (de != NULL && tmpfs_dirent_duphead(de)) 780 de = LIST_FIRST(&de->ud.td_duphead); 781 dc->tdc_current = de; 782 783 return (dc->tdc_current); 784 } 785 786 struct tmpfs_dirent * 787 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 788 { 789 struct tmpfs_dirent *de; 790 791 MPASS(dc->tdc_tree != NULL); 792 if (tmpfs_dirent_dup(dc->tdc_current)) { 793 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 794 if (dc->tdc_current != NULL) 795 return (dc->tdc_current); 796 } 797 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 798 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 799 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 800 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 801 MPASS(dc->tdc_current != NULL); 802 } 803 804 return (dc->tdc_current); 805 } 806 807 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 808 static struct tmpfs_dirent * 809 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 810 { 811 struct tmpfs_dirent *de, dekey; 812 813 dekey.td_hash = hash; 814 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 815 return (de); 816 } 817 818 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 819 static struct tmpfs_dirent * 820 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 821 struct tmpfs_dir_cursor *dc) 822 { 823 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 824 struct tmpfs_dirent *de, dekey; 825 826 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 827 828 if (cookie == node->tn_dir.tn_readdir_lastn && 829 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 830 /* Protect against possible race, tn_readdir_last[pn] 831 * may be updated with only shared vnode lock held. */ 832 if (cookie == tmpfs_dirent_cookie(de)) 833 goto out; 834 } 835 836 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 837 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 838 uh.td_dup.index_entries) { 839 MPASS(tmpfs_dirent_dup(de)); 840 if (de->td_cookie == cookie) 841 goto out; 842 /* dupindex list is sorted. */ 843 if (de->td_cookie < cookie) { 844 de = NULL; 845 goto out; 846 } 847 } 848 MPASS(de == NULL); 849 goto out; 850 } 851 852 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) { 853 de = NULL; 854 } else { 855 dekey.td_hash = cookie; 856 /* Recover if direntry for cookie was removed */ 857 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 858 } 859 dc->tdc_tree = de; 860 dc->tdc_current = de; 861 if (de != NULL && tmpfs_dirent_duphead(de)) { 862 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 863 MPASS(dc->tdc_current != NULL); 864 } 865 return (dc->tdc_current); 866 867 out: 868 dc->tdc_tree = de; 869 dc->tdc_current = de; 870 if (de != NULL && tmpfs_dirent_dup(de)) 871 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 872 de->td_hash); 873 return (dc->tdc_current); 874 } 875 876 /* 877 * Looks for a directory entry in the directory represented by node. 878 * 'cnp' describes the name of the entry to look for. Note that the . 879 * and .. components are not allowed as they do not physically exist 880 * within directories. 881 * 882 * Returns a pointer to the entry when found, otherwise NULL. 883 */ 884 struct tmpfs_dirent * 885 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 886 struct componentname *cnp) 887 { 888 struct tmpfs_dir_duphead *duphead; 889 struct tmpfs_dirent *de; 890 uint32_t hash; 891 892 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 893 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 894 cnp->cn_nameptr[1] == '.'))); 895 TMPFS_VALIDATE_DIR(node); 896 897 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 898 de = tmpfs_dir_xlookup_hash(node, hash); 899 if (de != NULL && tmpfs_dirent_duphead(de)) { 900 duphead = &de->ud.td_duphead; 901 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 902 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 903 cnp->cn_namelen)) 904 break; 905 } 906 } else if (de != NULL) { 907 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 908 cnp->cn_namelen)) 909 de = NULL; 910 } 911 if (de != NULL && f != NULL && de->td_node != f) 912 de = NULL; 913 914 return (de); 915 } 916 917 /* 918 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 919 * list, allocate new cookie value. 920 */ 921 static void 922 tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 923 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 924 { 925 struct tmpfs_dir_duphead *dupindex; 926 struct tmpfs_dirent *de, *pde; 927 928 dupindex = &dnode->tn_dir.tn_dupindex; 929 de = LIST_FIRST(dupindex); 930 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 931 if (de == NULL) 932 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 933 else 934 nde->td_cookie = de->td_cookie + 1; 935 MPASS(tmpfs_dirent_dup(nde)); 936 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 937 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 938 return; 939 } 940 941 /* 942 * Cookie numbers are near exhaustion. Scan dupindex list for unused 943 * numbers. dupindex list is sorted in descending order. Keep it so 944 * after inserting nde. 945 */ 946 while (1) { 947 pde = de; 948 de = LIST_NEXT(de, uh.td_dup.index_entries); 949 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 950 /* 951 * Last element of the index doesn't have minimal cookie 952 * value, use it. 953 */ 954 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 955 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 956 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 957 return; 958 } else if (de == NULL) { 959 /* 960 * We are so lucky have 2^30 hash duplicates in single 961 * directory :) Return largest possible cookie value. 962 * It should be fine except possible issues with 963 * VOP_READDIR restart. 964 */ 965 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 966 LIST_INSERT_HEAD(dupindex, nde, 967 uh.td_dup.index_entries); 968 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 969 return; 970 } 971 if (de->td_cookie + 1 == pde->td_cookie || 972 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 973 continue; /* No hole or invalid cookie. */ 974 nde->td_cookie = de->td_cookie + 1; 975 MPASS(tmpfs_dirent_dup(nde)); 976 MPASS(pde->td_cookie > nde->td_cookie); 977 MPASS(nde->td_cookie > de->td_cookie); 978 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 979 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 980 return; 981 } 982 } 983 984 /* 985 * Attaches the directory entry de to the directory represented by vp. 986 * Note that this does not change the link count of the node pointed by 987 * the directory entry, as this is done by tmpfs_alloc_dirent. 988 */ 989 void 990 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 991 { 992 struct tmpfs_node *dnode; 993 struct tmpfs_dirent *xde, *nde; 994 995 ASSERT_VOP_ELOCKED(vp, __func__); 996 MPASS(de->td_namelen > 0); 997 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 998 MPASS(de->td_cookie == de->td_hash); 999 1000 dnode = VP_TO_TMPFS_DIR(vp); 1001 dnode->tn_dir.tn_readdir_lastn = 0; 1002 dnode->tn_dir.tn_readdir_lastp = NULL; 1003 1004 MPASS(!tmpfs_dirent_dup(de)); 1005 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1006 if (xde != NULL && tmpfs_dirent_duphead(xde)) 1007 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1008 else if (xde != NULL) { 1009 /* 1010 * Allocate new duphead. Swap xde with duphead to avoid 1011 * adding/removing elements with the same hash. 1012 */ 1013 MPASS(!tmpfs_dirent_dup(xde)); 1014 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 1015 &nde); 1016 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 1017 memcpy(nde, xde, sizeof(*xde)); 1018 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 1019 LIST_INIT(&xde->ud.td_duphead); 1020 xde->td_namelen = 0; 1021 xde->td_node = NULL; 1022 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 1023 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1024 } 1025 dnode->tn_size += sizeof(struct tmpfs_dirent); 1026 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1027 TMPFS_NODE_MODIFIED; 1028 tmpfs_update(vp); 1029 } 1030 1031 /* 1032 * Detaches the directory entry de from the directory represented by vp. 1033 * Note that this does not change the link count of the node pointed by 1034 * the directory entry, as this is done by tmpfs_free_dirent. 1035 */ 1036 void 1037 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 1038 { 1039 struct tmpfs_mount *tmp; 1040 struct tmpfs_dir *head; 1041 struct tmpfs_node *dnode; 1042 struct tmpfs_dirent *xde; 1043 1044 ASSERT_VOP_ELOCKED(vp, __func__); 1045 1046 dnode = VP_TO_TMPFS_DIR(vp); 1047 head = &dnode->tn_dir.tn_dirhead; 1048 dnode->tn_dir.tn_readdir_lastn = 0; 1049 dnode->tn_dir.tn_readdir_lastp = NULL; 1050 1051 if (tmpfs_dirent_dup(de)) { 1052 /* Remove duphead if de was last entry. */ 1053 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1054 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1055 MPASS(tmpfs_dirent_duphead(xde)); 1056 } else 1057 xde = NULL; 1058 LIST_REMOVE(de, uh.td_dup.entries); 1059 LIST_REMOVE(de, uh.td_dup.index_entries); 1060 if (xde != NULL) { 1061 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1062 RB_REMOVE(tmpfs_dir, head, xde); 1063 tmp = VFS_TO_TMPFS(vp->v_mount); 1064 MPASS(xde->td_node == NULL); 1065 tmpfs_free_dirent(tmp, xde); 1066 } 1067 } 1068 de->td_cookie = de->td_hash; 1069 } else 1070 RB_REMOVE(tmpfs_dir, head, de); 1071 1072 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1073 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1074 TMPFS_NODE_MODIFIED; 1075 tmpfs_update(vp); 1076 } 1077 1078 void 1079 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1080 { 1081 struct tmpfs_dirent *de, *dde, *nde; 1082 1083 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1084 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1085 /* Node may already be destroyed. */ 1086 de->td_node = NULL; 1087 if (tmpfs_dirent_duphead(de)) { 1088 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1089 LIST_REMOVE(dde, uh.td_dup.entries); 1090 dde->td_node = NULL; 1091 tmpfs_free_dirent(tmp, dde); 1092 } 1093 } 1094 tmpfs_free_dirent(tmp, de); 1095 } 1096 } 1097 1098 /* 1099 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1100 * directory and returns it in the uio space. The function returns 0 1101 * on success, -1 if there was not enough space in the uio structure to 1102 * hold the directory entry or an appropriate error code if another 1103 * error happens. 1104 */ 1105 static int 1106 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1107 { 1108 int error; 1109 struct dirent dent; 1110 1111 TMPFS_VALIDATE_DIR(node); 1112 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1113 1114 dent.d_fileno = node->tn_id; 1115 dent.d_type = DT_DIR; 1116 dent.d_namlen = 1; 1117 dent.d_name[0] = '.'; 1118 dent.d_name[1] = '\0'; 1119 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1120 1121 if (dent.d_reclen > uio->uio_resid) 1122 error = EJUSTRETURN; 1123 else 1124 error = uiomove(&dent, dent.d_reclen, uio); 1125 1126 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1127 1128 return (error); 1129 } 1130 1131 /* 1132 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1133 * directory and returns it in the uio space. The function returns 0 1134 * on success, -1 if there was not enough space in the uio structure to 1135 * hold the directory entry or an appropriate error code if another 1136 * error happens. 1137 */ 1138 static int 1139 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1140 { 1141 int error; 1142 struct dirent dent; 1143 1144 TMPFS_VALIDATE_DIR(node); 1145 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1146 1147 /* 1148 * Return ENOENT if the current node is already removed. 1149 */ 1150 TMPFS_ASSERT_LOCKED(node); 1151 if (node->tn_dir.tn_parent == NULL) 1152 return (ENOENT); 1153 1154 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1155 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1156 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1157 1158 dent.d_type = DT_DIR; 1159 dent.d_namlen = 2; 1160 dent.d_name[0] = '.'; 1161 dent.d_name[1] = '.'; 1162 dent.d_name[2] = '\0'; 1163 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1164 1165 if (dent.d_reclen > uio->uio_resid) 1166 error = EJUSTRETURN; 1167 else 1168 error = uiomove(&dent, dent.d_reclen, uio); 1169 1170 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1171 1172 return (error); 1173 } 1174 1175 /* 1176 * Helper function for tmpfs_readdir. Returns as much directory entries 1177 * as can fit in the uio space. The read starts at uio->uio_offset. 1178 * The function returns 0 on success, -1 if there was not enough space 1179 * in the uio structure to hold the directory entry or an appropriate 1180 * error code if another error happens. 1181 */ 1182 int 1183 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1184 u_long *cookies, int *ncookies) 1185 { 1186 struct tmpfs_dir_cursor dc; 1187 struct tmpfs_dirent *de; 1188 off_t off; 1189 int error; 1190 1191 TMPFS_VALIDATE_DIR(node); 1192 1193 off = 0; 1194 1195 /* 1196 * Lookup the node from the current offset. The starting offset of 1197 * 0 will lookup both '.' and '..', and then the first real entry, 1198 * or EOF if there are none. Then find all entries for the dir that 1199 * fit into the buffer. Once no more entries are found (de == NULL), 1200 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1201 * call to return 0. 1202 */ 1203 switch (uio->uio_offset) { 1204 case TMPFS_DIRCOOKIE_DOT: 1205 error = tmpfs_dir_getdotdent(node, uio); 1206 if (error != 0) 1207 return (error); 1208 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1209 if (cookies != NULL) 1210 cookies[(*ncookies)++] = off = uio->uio_offset; 1211 /* FALLTHROUGH */ 1212 case TMPFS_DIRCOOKIE_DOTDOT: 1213 error = tmpfs_dir_getdotdotdent(node, uio); 1214 if (error != 0) 1215 return (error); 1216 de = tmpfs_dir_first(node, &dc); 1217 uio->uio_offset = tmpfs_dirent_cookie(de); 1218 if (cookies != NULL) 1219 cookies[(*ncookies)++] = off = uio->uio_offset; 1220 /* EOF. */ 1221 if (de == NULL) 1222 return (0); 1223 break; 1224 case TMPFS_DIRCOOKIE_EOF: 1225 return (0); 1226 default: 1227 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1228 if (de == NULL) 1229 return (EINVAL); 1230 if (cookies != NULL) 1231 off = tmpfs_dirent_cookie(de); 1232 } 1233 1234 /* Read as much entries as possible; i.e., until we reach the end of 1235 * the directory or we exhaust uio space. */ 1236 do { 1237 struct dirent d; 1238 1239 /* Create a dirent structure representing the current 1240 * tmpfs_node and fill it. */ 1241 if (de->td_node == NULL) { 1242 d.d_fileno = 1; 1243 d.d_type = DT_WHT; 1244 } else { 1245 d.d_fileno = de->td_node->tn_id; 1246 switch (de->td_node->tn_type) { 1247 case VBLK: 1248 d.d_type = DT_BLK; 1249 break; 1250 1251 case VCHR: 1252 d.d_type = DT_CHR; 1253 break; 1254 1255 case VDIR: 1256 d.d_type = DT_DIR; 1257 break; 1258 1259 case VFIFO: 1260 d.d_type = DT_FIFO; 1261 break; 1262 1263 case VLNK: 1264 d.d_type = DT_LNK; 1265 break; 1266 1267 case VREG: 1268 d.d_type = DT_REG; 1269 break; 1270 1271 case VSOCK: 1272 d.d_type = DT_SOCK; 1273 break; 1274 1275 default: 1276 panic("tmpfs_dir_getdents: type %p %d", 1277 de->td_node, (int)de->td_node->tn_type); 1278 } 1279 } 1280 d.d_namlen = de->td_namelen; 1281 MPASS(de->td_namelen < sizeof(d.d_name)); 1282 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1283 d.d_name[de->td_namelen] = '\0'; 1284 d.d_reclen = GENERIC_DIRSIZ(&d); 1285 1286 /* Stop reading if the directory entry we are treating is 1287 * bigger than the amount of data that can be returned. */ 1288 if (d.d_reclen > uio->uio_resid) { 1289 error = EJUSTRETURN; 1290 break; 1291 } 1292 1293 /* Copy the new dirent structure into the output buffer and 1294 * advance pointers. */ 1295 error = uiomove(&d, d.d_reclen, uio); 1296 if (error == 0) { 1297 de = tmpfs_dir_next(node, &dc); 1298 if (cookies != NULL) { 1299 off = tmpfs_dirent_cookie(de); 1300 MPASS(*ncookies < maxcookies); 1301 cookies[(*ncookies)++] = off; 1302 } 1303 } 1304 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1305 1306 /* Skip setting off when using cookies as it is already done above. */ 1307 if (cookies == NULL) 1308 off = tmpfs_dirent_cookie(de); 1309 1310 /* Update the offset and cache. */ 1311 uio->uio_offset = off; 1312 node->tn_dir.tn_readdir_lastn = off; 1313 node->tn_dir.tn_readdir_lastp = de; 1314 1315 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1316 return error; 1317 } 1318 1319 int 1320 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1321 { 1322 struct tmpfs_dirent *de; 1323 int error; 1324 1325 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1326 cnp->cn_nameptr, cnp->cn_namelen, &de); 1327 if (error != 0) 1328 return (error); 1329 tmpfs_dir_attach(dvp, de); 1330 return (0); 1331 } 1332 1333 void 1334 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1335 { 1336 struct tmpfs_dirent *de; 1337 1338 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1339 MPASS(de != NULL && de->td_node == NULL); 1340 tmpfs_dir_detach(dvp, de); 1341 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1342 } 1343 1344 /* 1345 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1346 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1347 * 'newsize' must be positive. 1348 * 1349 * Returns zero on success or an appropriate error code on failure. 1350 */ 1351 int 1352 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1353 { 1354 struct tmpfs_mount *tmp; 1355 struct tmpfs_node *node; 1356 vm_object_t uobj; 1357 vm_page_t m; 1358 vm_pindex_t idx, newpages, oldpages; 1359 off_t oldsize; 1360 int base, rv; 1361 1362 MPASS(vp->v_type == VREG); 1363 MPASS(newsize >= 0); 1364 1365 node = VP_TO_TMPFS_NODE(vp); 1366 uobj = node->tn_reg.tn_aobj; 1367 tmp = VFS_TO_TMPFS(vp->v_mount); 1368 1369 /* 1370 * Convert the old and new sizes to the number of pages needed to 1371 * store them. It may happen that we do not need to do anything 1372 * because the last allocated page can accommodate the change on 1373 * its own. 1374 */ 1375 oldsize = node->tn_size; 1376 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1377 MPASS(oldpages == uobj->size); 1378 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1379 1380 if (__predict_true(newpages == oldpages && newsize >= oldsize)) { 1381 node->tn_size = newsize; 1382 return (0); 1383 } 1384 1385 if (newpages > oldpages && 1386 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1387 return (ENOSPC); 1388 1389 VM_OBJECT_WLOCK(uobj); 1390 if (newsize < oldsize) { 1391 /* 1392 * Zero the truncated part of the last page. 1393 */ 1394 base = newsize & PAGE_MASK; 1395 if (base != 0) { 1396 idx = OFF_TO_IDX(newsize); 1397 retry: 1398 m = vm_page_lookup(uobj, idx); 1399 if (m != NULL) { 1400 if (vm_page_sleep_if_busy(m, "tmfssz")) 1401 goto retry; 1402 MPASS(m->valid == VM_PAGE_BITS_ALL); 1403 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1404 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1405 if (m == NULL) { 1406 VM_OBJECT_WUNLOCK(uobj); 1407 VM_WAIT; 1408 VM_OBJECT_WLOCK(uobj); 1409 goto retry; 1410 } 1411 rv = vm_pager_get_pages(uobj, &m, 1, NULL, 1412 NULL); 1413 vm_page_lock(m); 1414 if (rv == VM_PAGER_OK) { 1415 /* 1416 * Since the page was not resident, 1417 * and therefore not recently 1418 * accessed, immediately enqueue it 1419 * for asynchronous laundering. The 1420 * current operation is not regarded 1421 * as an access. 1422 */ 1423 vm_page_launder(m); 1424 vm_page_unlock(m); 1425 vm_page_xunbusy(m); 1426 } else { 1427 vm_page_free(m); 1428 vm_page_unlock(m); 1429 if (ignerr) 1430 m = NULL; 1431 else { 1432 VM_OBJECT_WUNLOCK(uobj); 1433 return (EIO); 1434 } 1435 } 1436 } 1437 if (m != NULL) { 1438 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1439 vm_page_dirty(m); 1440 vm_pager_page_unswapped(m); 1441 } 1442 } 1443 1444 /* 1445 * Release any swap space and free any whole pages. 1446 */ 1447 if (newpages < oldpages) { 1448 swap_pager_freespace(uobj, newpages, oldpages - 1449 newpages); 1450 vm_object_page_remove(uobj, newpages, 0, 0); 1451 } 1452 } 1453 uobj->size = newpages; 1454 VM_OBJECT_WUNLOCK(uobj); 1455 1456 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages); 1457 1458 node->tn_size = newsize; 1459 return (0); 1460 } 1461 1462 void 1463 tmpfs_check_mtime(struct vnode *vp) 1464 { 1465 struct tmpfs_node *node; 1466 struct vm_object *obj; 1467 1468 ASSERT_VOP_ELOCKED(vp, "check_mtime"); 1469 if (vp->v_type != VREG) 1470 return; 1471 obj = vp->v_object; 1472 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) == 1473 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj")); 1474 /* unlocked read */ 1475 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) { 1476 VM_OBJECT_WLOCK(obj); 1477 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) { 1478 obj->flags &= ~OBJ_TMPFS_DIRTY; 1479 node = VP_TO_TMPFS_NODE(vp); 1480 node->tn_status |= TMPFS_NODE_MODIFIED | 1481 TMPFS_NODE_CHANGED; 1482 } 1483 VM_OBJECT_WUNLOCK(obj); 1484 } 1485 } 1486 1487 /* 1488 * Change flags of the given vnode. 1489 * Caller should execute tmpfs_update on vp after a successful execution. 1490 * The vnode must be locked on entry and remain locked on exit. 1491 */ 1492 int 1493 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1494 struct thread *p) 1495 { 1496 int error; 1497 struct tmpfs_node *node; 1498 1499 ASSERT_VOP_ELOCKED(vp, "chflags"); 1500 1501 node = VP_TO_TMPFS_NODE(vp); 1502 1503 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1504 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1505 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1506 UF_SPARSE | UF_SYSTEM)) != 0) 1507 return (EOPNOTSUPP); 1508 1509 /* Disallow this operation if the file system is mounted read-only. */ 1510 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1511 return EROFS; 1512 1513 /* 1514 * Callers may only modify the file flags on objects they 1515 * have VADMIN rights for. 1516 */ 1517 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1518 return (error); 1519 /* 1520 * Unprivileged processes are not permitted to unset system 1521 * flags, or modify flags if any system flags are set. 1522 */ 1523 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1524 if (node->tn_flags & 1525 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1526 error = securelevel_gt(cred, 0); 1527 if (error) 1528 return (error); 1529 } 1530 } else { 1531 if (node->tn_flags & 1532 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1533 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1534 return (EPERM); 1535 } 1536 node->tn_flags = flags; 1537 node->tn_status |= TMPFS_NODE_CHANGED; 1538 1539 ASSERT_VOP_ELOCKED(vp, "chflags2"); 1540 1541 return (0); 1542 } 1543 1544 /* 1545 * Change access mode on the given vnode. 1546 * Caller should execute tmpfs_update on vp after a successful execution. 1547 * The vnode must be locked on entry and remain locked on exit. 1548 */ 1549 int 1550 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1551 { 1552 int error; 1553 struct tmpfs_node *node; 1554 1555 ASSERT_VOP_ELOCKED(vp, "chmod"); 1556 1557 node = VP_TO_TMPFS_NODE(vp); 1558 1559 /* Disallow this operation if the file system is mounted read-only. */ 1560 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1561 return EROFS; 1562 1563 /* Immutable or append-only files cannot be modified, either. */ 1564 if (node->tn_flags & (IMMUTABLE | APPEND)) 1565 return EPERM; 1566 1567 /* 1568 * To modify the permissions on a file, must possess VADMIN 1569 * for that file. 1570 */ 1571 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1572 return (error); 1573 1574 /* 1575 * Privileged processes may set the sticky bit on non-directories, 1576 * as well as set the setgid bit on a file with a group that the 1577 * process is not a member of. 1578 */ 1579 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1580 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1581 return (EFTYPE); 1582 } 1583 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1584 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1585 if (error) 1586 return (error); 1587 } 1588 1589 1590 node->tn_mode &= ~ALLPERMS; 1591 node->tn_mode |= mode & ALLPERMS; 1592 1593 node->tn_status |= TMPFS_NODE_CHANGED; 1594 1595 ASSERT_VOP_ELOCKED(vp, "chmod2"); 1596 1597 return (0); 1598 } 1599 1600 /* 1601 * Change ownership of the given vnode. At least one of uid or gid must 1602 * be different than VNOVAL. If one is set to that value, the attribute 1603 * is unchanged. 1604 * Caller should execute tmpfs_update on vp after a successful execution. 1605 * The vnode must be locked on entry and remain locked on exit. 1606 */ 1607 int 1608 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1609 struct thread *p) 1610 { 1611 int error; 1612 struct tmpfs_node *node; 1613 uid_t ouid; 1614 gid_t ogid; 1615 1616 ASSERT_VOP_ELOCKED(vp, "chown"); 1617 1618 node = VP_TO_TMPFS_NODE(vp); 1619 1620 /* Assign default values if they are unknown. */ 1621 MPASS(uid != VNOVAL || gid != VNOVAL); 1622 if (uid == VNOVAL) 1623 uid = node->tn_uid; 1624 if (gid == VNOVAL) 1625 gid = node->tn_gid; 1626 MPASS(uid != VNOVAL && gid != VNOVAL); 1627 1628 /* Disallow this operation if the file system is mounted read-only. */ 1629 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1630 return EROFS; 1631 1632 /* Immutable or append-only files cannot be modified, either. */ 1633 if (node->tn_flags & (IMMUTABLE | APPEND)) 1634 return EPERM; 1635 1636 /* 1637 * To modify the ownership of a file, must possess VADMIN for that 1638 * file. 1639 */ 1640 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1641 return (error); 1642 1643 /* 1644 * To change the owner of a file, or change the group of a file to a 1645 * group of which we are not a member, the caller must have 1646 * privilege. 1647 */ 1648 if ((uid != node->tn_uid || 1649 (gid != node->tn_gid && !groupmember(gid, cred))) && 1650 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1651 return (error); 1652 1653 ogid = node->tn_gid; 1654 ouid = node->tn_uid; 1655 1656 node->tn_uid = uid; 1657 node->tn_gid = gid; 1658 1659 node->tn_status |= TMPFS_NODE_CHANGED; 1660 1661 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1662 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1663 node->tn_mode &= ~(S_ISUID | S_ISGID); 1664 } 1665 1666 ASSERT_VOP_ELOCKED(vp, "chown2"); 1667 1668 return (0); 1669 } 1670 1671 /* 1672 * Change size of the given vnode. 1673 * Caller should execute tmpfs_update on vp after a successful execution. 1674 * The vnode must be locked on entry and remain locked on exit. 1675 */ 1676 int 1677 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1678 struct thread *p) 1679 { 1680 int error; 1681 struct tmpfs_node *node; 1682 1683 ASSERT_VOP_ELOCKED(vp, "chsize"); 1684 1685 node = VP_TO_TMPFS_NODE(vp); 1686 1687 /* Decide whether this is a valid operation based on the file type. */ 1688 error = 0; 1689 switch (vp->v_type) { 1690 case VDIR: 1691 return EISDIR; 1692 1693 case VREG: 1694 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1695 return EROFS; 1696 break; 1697 1698 case VBLK: 1699 /* FALLTHROUGH */ 1700 case VCHR: 1701 /* FALLTHROUGH */ 1702 case VFIFO: 1703 /* Allow modifications of special files even if in the file 1704 * system is mounted read-only (we are not modifying the 1705 * files themselves, but the objects they represent). */ 1706 return 0; 1707 1708 default: 1709 /* Anything else is unsupported. */ 1710 return EOPNOTSUPP; 1711 } 1712 1713 /* Immutable or append-only files cannot be modified, either. */ 1714 if (node->tn_flags & (IMMUTABLE | APPEND)) 1715 return EPERM; 1716 1717 error = tmpfs_truncate(vp, size); 1718 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1719 * for us, as will update tn_status; no need to do that here. */ 1720 1721 ASSERT_VOP_ELOCKED(vp, "chsize2"); 1722 1723 return (error); 1724 } 1725 1726 /* 1727 * Change access and modification times of the given vnode. 1728 * Caller should execute tmpfs_update on vp after a successful execution. 1729 * The vnode must be locked on entry and remain locked on exit. 1730 */ 1731 int 1732 tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 1733 struct ucred *cred, struct thread *l) 1734 { 1735 int error; 1736 struct tmpfs_node *node; 1737 1738 ASSERT_VOP_ELOCKED(vp, "chtimes"); 1739 1740 node = VP_TO_TMPFS_NODE(vp); 1741 1742 /* Disallow this operation if the file system is mounted read-only. */ 1743 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1744 return EROFS; 1745 1746 /* Immutable or append-only files cannot be modified, either. */ 1747 if (node->tn_flags & (IMMUTABLE | APPEND)) 1748 return EPERM; 1749 1750 error = vn_utimes_perm(vp, vap, cred, l); 1751 if (error != 0) 1752 return (error); 1753 1754 if (vap->va_atime.tv_sec != VNOVAL) 1755 node->tn_status |= TMPFS_NODE_ACCESSED; 1756 1757 if (vap->va_mtime.tv_sec != VNOVAL) 1758 node->tn_status |= TMPFS_NODE_MODIFIED; 1759 1760 if (vap->va_birthtime.tv_sec != VNOVAL) 1761 node->tn_status |= TMPFS_NODE_MODIFIED; 1762 1763 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1764 1765 if (vap->va_birthtime.tv_sec != VNOVAL) 1766 node->tn_birthtime = vap->va_birthtime; 1767 ASSERT_VOP_ELOCKED(vp, "chtimes2"); 1768 1769 return (0); 1770 } 1771 1772 void 1773 tmpfs_set_status(struct tmpfs_node *node, int status) 1774 { 1775 1776 if ((node->tn_status & status) == status) 1777 return; 1778 TMPFS_NODE_LOCK(node); 1779 node->tn_status |= status; 1780 TMPFS_NODE_UNLOCK(node); 1781 } 1782 1783 /* Sync timestamps */ 1784 void 1785 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1786 const struct timespec *mod) 1787 { 1788 struct tmpfs_node *node; 1789 struct timespec now; 1790 1791 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes"); 1792 node = VP_TO_TMPFS_NODE(vp); 1793 1794 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1795 TMPFS_NODE_CHANGED)) == 0) 1796 return; 1797 1798 vfs_timestamp(&now); 1799 TMPFS_NODE_LOCK(node); 1800 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1801 if (acc == NULL) 1802 acc = &now; 1803 node->tn_atime = *acc; 1804 } 1805 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1806 if (mod == NULL) 1807 mod = &now; 1808 node->tn_mtime = *mod; 1809 } 1810 if (node->tn_status & TMPFS_NODE_CHANGED) 1811 node->tn_ctime = now; 1812 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1813 TMPFS_NODE_CHANGED); 1814 TMPFS_NODE_UNLOCK(node); 1815 1816 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */ 1817 random_harvest_queue(node, sizeof(*node), 1, RANDOM_FS_ATIME); 1818 } 1819 1820 void 1821 tmpfs_update(struct vnode *vp) 1822 { 1823 1824 tmpfs_itimes(vp, NULL, NULL); 1825 } 1826 1827 int 1828 tmpfs_truncate(struct vnode *vp, off_t length) 1829 { 1830 int error; 1831 struct tmpfs_node *node; 1832 1833 node = VP_TO_TMPFS_NODE(vp); 1834 1835 if (length < 0) { 1836 error = EINVAL; 1837 goto out; 1838 } 1839 1840 if (node->tn_size == length) { 1841 error = 0; 1842 goto out; 1843 } 1844 1845 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1846 return (EFBIG); 1847 1848 error = tmpfs_reg_resize(vp, length, FALSE); 1849 if (error == 0) 1850 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1851 1852 out: 1853 tmpfs_update(vp); 1854 1855 return (error); 1856 } 1857 1858 static __inline int 1859 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1860 { 1861 if (a->td_hash > b->td_hash) 1862 return (1); 1863 else if (a->td_hash < b->td_hash) 1864 return (-1); 1865 return (0); 1866 } 1867 1868 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1869