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 atomic_subtract_long(&tmp->tm_pages_used, uobj->size); 354 KASSERT((uobj->flags & OBJ_TMPFS) == 0, 355 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj)); 356 vm_object_deallocate(uobj); 357 } 358 break; 359 360 default: 361 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 362 } 363 364 free_unr(tmp->tm_ino_unr, node->tn_id); 365 uma_zfree(tmp->tm_node_pool, node); 366 TMPFS_LOCK(tmp); 367 tmpfs_free_tmp(tmp); 368 return (true); 369 } 370 371 static __inline uint32_t 372 tmpfs_dirent_hash(const char *name, u_int len) 373 { 374 uint32_t hash; 375 376 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK; 377 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP 378 hash &= 0xf; 379 #endif 380 if (hash < TMPFS_DIRCOOKIE_MIN) 381 hash += TMPFS_DIRCOOKIE_MIN; 382 383 return (hash); 384 } 385 386 static __inline off_t 387 tmpfs_dirent_cookie(struct tmpfs_dirent *de) 388 { 389 if (de == NULL) 390 return (TMPFS_DIRCOOKIE_EOF); 391 392 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN); 393 394 return (de->td_cookie); 395 } 396 397 static __inline boolean_t 398 tmpfs_dirent_dup(struct tmpfs_dirent *de) 399 { 400 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0); 401 } 402 403 static __inline boolean_t 404 tmpfs_dirent_duphead(struct tmpfs_dirent *de) 405 { 406 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0); 407 } 408 409 void 410 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen) 411 { 412 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen); 413 memcpy(de->ud.td_name, name, namelen); 414 de->td_namelen = namelen; 415 } 416 417 /* 418 * Allocates a new directory entry for the node node with a name of name. 419 * The new directory entry is returned in *de. 420 * 421 * The link count of node is increased by one to reflect the new object 422 * referencing it. 423 * 424 * Returns zero on success or an appropriate error code on failure. 425 */ 426 int 427 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 428 const char *name, u_int len, struct tmpfs_dirent **de) 429 { 430 struct tmpfs_dirent *nde; 431 432 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK); 433 nde->td_node = node; 434 if (name != NULL) { 435 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 436 tmpfs_dirent_init(nde, name, len); 437 } else 438 nde->td_namelen = 0; 439 if (node != NULL) 440 node->tn_links++; 441 442 *de = nde; 443 444 return 0; 445 } 446 447 /* 448 * Frees a directory entry. It is the caller's responsibility to destroy 449 * the node referenced by it if needed. 450 * 451 * The link count of node is decreased by one to reflect the removal of an 452 * object that referenced it. This only happens if 'node_exists' is true; 453 * otherwise the function will not access the node referred to by the 454 * directory entry, as it may already have been released from the outside. 455 */ 456 void 457 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 458 { 459 struct tmpfs_node *node; 460 461 node = de->td_node; 462 if (node != NULL) { 463 MPASS(node->tn_links > 0); 464 node->tn_links--; 465 } 466 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL) 467 free(de->ud.td_name, M_TMPFSNAME); 468 uma_zfree(tmp->tm_dirent_pool, de); 469 } 470 471 void 472 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj) 473 { 474 475 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject"); 476 if (vp->v_type != VREG || obj == NULL) 477 return; 478 479 VM_OBJECT_WLOCK(obj); 480 VI_LOCK(vp); 481 vm_object_clear_flag(obj, OBJ_TMPFS); 482 obj->un_pager.swp.swp_tmpfs = NULL; 483 VI_UNLOCK(vp); 484 VM_OBJECT_WUNLOCK(obj); 485 } 486 487 /* 488 * Need to clear v_object for insmntque failure. 489 */ 490 static void 491 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg) 492 { 493 494 tmpfs_destroy_vobject(vp, vp->v_object); 495 vp->v_object = NULL; 496 vp->v_data = NULL; 497 vp->v_op = &dead_vnodeops; 498 vgone(vp); 499 vput(vp); 500 } 501 502 /* 503 * Allocates a new vnode for the node node or returns a new reference to 504 * an existing one if the node had already a vnode referencing it. The 505 * resulting locked vnode is returned in *vpp. 506 * 507 * Returns zero on success or an appropriate error code on failure. 508 */ 509 int 510 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 511 struct vnode **vpp) 512 { 513 struct vnode *vp; 514 struct tmpfs_mount *tm; 515 vm_object_t object; 516 int error; 517 518 error = 0; 519 tm = VFS_TO_TMPFS(mp); 520 TMPFS_NODE_LOCK(node); 521 tmpfs_ref_node_locked(node); 522 loop: 523 TMPFS_NODE_ASSERT_LOCKED(node); 524 if ((vp = node->tn_vnode) != NULL) { 525 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 526 VI_LOCK(vp); 527 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) || 528 ((vp->v_iflag & VI_DOOMED) != 0 && 529 (lkflag & LK_NOWAIT) != 0)) { 530 VI_UNLOCK(vp); 531 TMPFS_NODE_UNLOCK(node); 532 error = ENOENT; 533 vp = NULL; 534 goto out; 535 } 536 if ((vp->v_iflag & VI_DOOMED) != 0) { 537 VI_UNLOCK(vp); 538 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM; 539 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) { 540 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 541 0, "tmpfsE", 0); 542 } 543 goto loop; 544 } 545 TMPFS_NODE_UNLOCK(node); 546 error = vget(vp, lkflag | LK_INTERLOCK, curthread); 547 if (error == ENOENT) { 548 TMPFS_NODE_LOCK(node); 549 goto loop; 550 } 551 if (error != 0) { 552 vp = NULL; 553 goto out; 554 } 555 556 /* 557 * Make sure the vnode is still there after 558 * getting the interlock to avoid racing a free. 559 */ 560 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 561 vput(vp); 562 TMPFS_NODE_LOCK(node); 563 goto loop; 564 } 565 566 goto out; 567 } 568 569 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 570 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 571 TMPFS_NODE_UNLOCK(node); 572 error = ENOENT; 573 vp = NULL; 574 goto out; 575 } 576 577 /* 578 * otherwise lock the vp list while we call getnewvnode 579 * since that can block. 580 */ 581 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 582 node->tn_vpstate |= TMPFS_VNODE_WANT; 583 error = msleep((caddr_t) &node->tn_vpstate, 584 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0); 585 if (error != 0) 586 goto out; 587 goto loop; 588 } else 589 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 590 591 TMPFS_NODE_UNLOCK(node); 592 593 /* Get a new vnode and associate it with our node. */ 594 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ? 595 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp); 596 if (error != 0) 597 goto unlock; 598 MPASS(vp != NULL); 599 600 /* lkflag is ignored, the lock is exclusive */ 601 (void) vn_lock(vp, lkflag | LK_RETRY); 602 603 vp->v_data = node; 604 vp->v_type = node->tn_type; 605 606 /* Type-specific initialization. */ 607 switch (node->tn_type) { 608 case VBLK: 609 /* FALLTHROUGH */ 610 case VCHR: 611 /* FALLTHROUGH */ 612 case VLNK: 613 /* FALLTHROUGH */ 614 case VSOCK: 615 break; 616 case VFIFO: 617 vp->v_op = &tmpfs_fifoop_entries; 618 break; 619 case VREG: 620 object = node->tn_reg.tn_aobj; 621 VM_OBJECT_WLOCK(object); 622 VI_LOCK(vp); 623 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 624 vp->v_object = object; 625 object->un_pager.swp.swp_tmpfs = vp; 626 vm_object_set_flag(object, OBJ_TMPFS); 627 VI_UNLOCK(vp); 628 VM_OBJECT_WUNLOCK(object); 629 break; 630 case VDIR: 631 MPASS(node->tn_dir.tn_parent != NULL); 632 if (node->tn_dir.tn_parent == node) 633 vp->v_vflag |= VV_ROOT; 634 break; 635 636 default: 637 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 638 } 639 if (vp->v_type != VFIFO) 640 VN_LOCK_ASHARE(vp); 641 642 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 643 if (error != 0) 644 vp = NULL; 645 646 unlock: 647 TMPFS_NODE_LOCK(node); 648 649 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 650 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 651 node->tn_vnode = vp; 652 653 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 654 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 655 TMPFS_NODE_UNLOCK(node); 656 wakeup((caddr_t) &node->tn_vpstate); 657 } else 658 TMPFS_NODE_UNLOCK(node); 659 660 out: 661 if (error == 0) { 662 *vpp = vp; 663 664 #ifdef INVARIANTS 665 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 666 TMPFS_NODE_LOCK(node); 667 MPASS(*vpp == node->tn_vnode); 668 TMPFS_NODE_UNLOCK(node); 669 #endif 670 } 671 tmpfs_free_node(tm, node); 672 673 return (error); 674 } 675 676 /* 677 * Destroys the association between the vnode vp and the node it 678 * references. 679 */ 680 void 681 tmpfs_free_vp(struct vnode *vp) 682 { 683 struct tmpfs_node *node; 684 685 node = VP_TO_TMPFS_NODE(vp); 686 687 TMPFS_NODE_ASSERT_LOCKED(node); 688 node->tn_vnode = NULL; 689 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 690 wakeup(&node->tn_vnode); 691 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 692 vp->v_data = NULL; 693 } 694 695 /* 696 * Allocates a new file of type 'type' and adds it to the parent directory 697 * 'dvp'; this addition is done using the component name given in 'cnp'. 698 * The ownership of the new file is automatically assigned based on the 699 * credentials of the caller (through 'cnp'), the group is set based on 700 * the parent directory and the mode is determined from the 'vap' argument. 701 * If successful, *vpp holds a vnode to the newly created file and zero 702 * is returned. Otherwise *vpp is NULL and the function returns an 703 * appropriate error code. 704 */ 705 int 706 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 707 struct componentname *cnp, char *target) 708 { 709 int error; 710 struct tmpfs_dirent *de; 711 struct tmpfs_mount *tmp; 712 struct tmpfs_node *dnode; 713 struct tmpfs_node *node; 714 struct tmpfs_node *parent; 715 716 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file"); 717 MPASS(cnp->cn_flags & HASBUF); 718 719 tmp = VFS_TO_TMPFS(dvp->v_mount); 720 dnode = VP_TO_TMPFS_DIR(dvp); 721 *vpp = NULL; 722 723 /* If the entry we are creating is a directory, we cannot overflow 724 * the number of links of its parent, because it will get a new 725 * link. */ 726 if (vap->va_type == VDIR) { 727 /* Ensure that we do not overflow the maximum number of links 728 * imposed by the system. */ 729 MPASS(dnode->tn_links <= LINK_MAX); 730 if (dnode->tn_links == LINK_MAX) { 731 return (EMLINK); 732 } 733 734 parent = dnode; 735 MPASS(parent != NULL); 736 } else 737 parent = NULL; 738 739 /* Allocate a node that represents the new file. */ 740 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type, 741 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent, 742 target, vap->va_rdev, &node); 743 if (error != 0) 744 return (error); 745 746 /* Allocate a directory entry that points to the new file. */ 747 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 748 &de); 749 if (error != 0) { 750 tmpfs_free_node(tmp, node); 751 return (error); 752 } 753 754 /* Allocate a vnode for the new file. */ 755 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 756 if (error != 0) { 757 tmpfs_free_dirent(tmp, de); 758 tmpfs_free_node(tmp, node); 759 return (error); 760 } 761 762 /* Now that all required items are allocated, we can proceed to 763 * insert the new node into the directory, an operation that 764 * cannot fail. */ 765 if (cnp->cn_flags & ISWHITEOUT) 766 tmpfs_dir_whiteout_remove(dvp, cnp); 767 tmpfs_dir_attach(dvp, de); 768 return (0); 769 } 770 771 struct tmpfs_dirent * 772 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 773 { 774 struct tmpfs_dirent *de; 775 776 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 777 dc->tdc_tree = de; 778 if (de != NULL && tmpfs_dirent_duphead(de)) 779 de = LIST_FIRST(&de->ud.td_duphead); 780 dc->tdc_current = de; 781 782 return (dc->tdc_current); 783 } 784 785 struct tmpfs_dirent * 786 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 787 { 788 struct tmpfs_dirent *de; 789 790 MPASS(dc->tdc_tree != NULL); 791 if (tmpfs_dirent_dup(dc->tdc_current)) { 792 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 793 if (dc->tdc_current != NULL) 794 return (dc->tdc_current); 795 } 796 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 797 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 798 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 799 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 800 MPASS(dc->tdc_current != NULL); 801 } 802 803 return (dc->tdc_current); 804 } 805 806 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 807 static struct tmpfs_dirent * 808 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 809 { 810 struct tmpfs_dirent *de, dekey; 811 812 dekey.td_hash = hash; 813 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 814 return (de); 815 } 816 817 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 818 static struct tmpfs_dirent * 819 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 820 struct tmpfs_dir_cursor *dc) 821 { 822 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 823 struct tmpfs_dirent *de, dekey; 824 825 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 826 827 if (cookie == node->tn_dir.tn_readdir_lastn && 828 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 829 /* Protect against possible race, tn_readdir_last[pn] 830 * may be updated with only shared vnode lock held. */ 831 if (cookie == tmpfs_dirent_cookie(de)) 832 goto out; 833 } 834 835 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 836 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 837 uh.td_dup.index_entries) { 838 MPASS(tmpfs_dirent_dup(de)); 839 if (de->td_cookie == cookie) 840 goto out; 841 /* dupindex list is sorted. */ 842 if (de->td_cookie < cookie) { 843 de = NULL; 844 goto out; 845 } 846 } 847 MPASS(de == NULL); 848 goto out; 849 } 850 851 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) { 852 de = NULL; 853 } else { 854 dekey.td_hash = cookie; 855 /* Recover if direntry for cookie was removed */ 856 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 857 } 858 dc->tdc_tree = de; 859 dc->tdc_current = de; 860 if (de != NULL && tmpfs_dirent_duphead(de)) { 861 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 862 MPASS(dc->tdc_current != NULL); 863 } 864 return (dc->tdc_current); 865 866 out: 867 dc->tdc_tree = de; 868 dc->tdc_current = de; 869 if (de != NULL && tmpfs_dirent_dup(de)) 870 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 871 de->td_hash); 872 return (dc->tdc_current); 873 } 874 875 /* 876 * Looks for a directory entry in the directory represented by node. 877 * 'cnp' describes the name of the entry to look for. Note that the . 878 * and .. components are not allowed as they do not physically exist 879 * within directories. 880 * 881 * Returns a pointer to the entry when found, otherwise NULL. 882 */ 883 struct tmpfs_dirent * 884 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 885 struct componentname *cnp) 886 { 887 struct tmpfs_dir_duphead *duphead; 888 struct tmpfs_dirent *de; 889 uint32_t hash; 890 891 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 892 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 893 cnp->cn_nameptr[1] == '.'))); 894 TMPFS_VALIDATE_DIR(node); 895 896 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 897 de = tmpfs_dir_xlookup_hash(node, hash); 898 if (de != NULL && tmpfs_dirent_duphead(de)) { 899 duphead = &de->ud.td_duphead; 900 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 901 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 902 cnp->cn_namelen)) 903 break; 904 } 905 } else if (de != NULL) { 906 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 907 cnp->cn_namelen)) 908 de = NULL; 909 } 910 if (de != NULL && f != NULL && de->td_node != f) 911 de = NULL; 912 913 return (de); 914 } 915 916 /* 917 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 918 * list, allocate new cookie value. 919 */ 920 static void 921 tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 922 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 923 { 924 struct tmpfs_dir_duphead *dupindex; 925 struct tmpfs_dirent *de, *pde; 926 927 dupindex = &dnode->tn_dir.tn_dupindex; 928 de = LIST_FIRST(dupindex); 929 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 930 if (de == NULL) 931 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 932 else 933 nde->td_cookie = de->td_cookie + 1; 934 MPASS(tmpfs_dirent_dup(nde)); 935 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 936 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 937 return; 938 } 939 940 /* 941 * Cookie numbers are near exhaustion. Scan dupindex list for unused 942 * numbers. dupindex list is sorted in descending order. Keep it so 943 * after inserting nde. 944 */ 945 while (1) { 946 pde = de; 947 de = LIST_NEXT(de, uh.td_dup.index_entries); 948 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 949 /* 950 * Last element of the index doesn't have minimal cookie 951 * value, use it. 952 */ 953 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 954 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 955 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 956 return; 957 } else if (de == NULL) { 958 /* 959 * We are so lucky have 2^30 hash duplicates in single 960 * directory :) Return largest possible cookie value. 961 * It should be fine except possible issues with 962 * VOP_READDIR restart. 963 */ 964 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 965 LIST_INSERT_HEAD(dupindex, nde, 966 uh.td_dup.index_entries); 967 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 968 return; 969 } 970 if (de->td_cookie + 1 == pde->td_cookie || 971 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 972 continue; /* No hole or invalid cookie. */ 973 nde->td_cookie = de->td_cookie + 1; 974 MPASS(tmpfs_dirent_dup(nde)); 975 MPASS(pde->td_cookie > nde->td_cookie); 976 MPASS(nde->td_cookie > de->td_cookie); 977 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 978 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 979 return; 980 } 981 } 982 983 /* 984 * Attaches the directory entry de to the directory represented by vp. 985 * Note that this does not change the link count of the node pointed by 986 * the directory entry, as this is done by tmpfs_alloc_dirent. 987 */ 988 void 989 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 990 { 991 struct tmpfs_node *dnode; 992 struct tmpfs_dirent *xde, *nde; 993 994 ASSERT_VOP_ELOCKED(vp, __func__); 995 MPASS(de->td_namelen > 0); 996 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 997 MPASS(de->td_cookie == de->td_hash); 998 999 dnode = VP_TO_TMPFS_DIR(vp); 1000 dnode->tn_dir.tn_readdir_lastn = 0; 1001 dnode->tn_dir.tn_readdir_lastp = NULL; 1002 1003 MPASS(!tmpfs_dirent_dup(de)); 1004 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1005 if (xde != NULL && tmpfs_dirent_duphead(xde)) 1006 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1007 else if (xde != NULL) { 1008 /* 1009 * Allocate new duphead. Swap xde with duphead to avoid 1010 * adding/removing elements with the same hash. 1011 */ 1012 MPASS(!tmpfs_dirent_dup(xde)); 1013 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 1014 &nde); 1015 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 1016 memcpy(nde, xde, sizeof(*xde)); 1017 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 1018 LIST_INIT(&xde->ud.td_duphead); 1019 xde->td_namelen = 0; 1020 xde->td_node = NULL; 1021 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 1022 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1023 } 1024 dnode->tn_size += sizeof(struct tmpfs_dirent); 1025 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1026 TMPFS_NODE_MODIFIED; 1027 tmpfs_update(vp); 1028 } 1029 1030 /* 1031 * Detaches the directory entry de from the directory represented by vp. 1032 * Note that this does not change the link count of the node pointed by 1033 * the directory entry, as this is done by tmpfs_free_dirent. 1034 */ 1035 void 1036 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 1037 { 1038 struct tmpfs_mount *tmp; 1039 struct tmpfs_dir *head; 1040 struct tmpfs_node *dnode; 1041 struct tmpfs_dirent *xde; 1042 1043 ASSERT_VOP_ELOCKED(vp, __func__); 1044 1045 dnode = VP_TO_TMPFS_DIR(vp); 1046 head = &dnode->tn_dir.tn_dirhead; 1047 dnode->tn_dir.tn_readdir_lastn = 0; 1048 dnode->tn_dir.tn_readdir_lastp = NULL; 1049 1050 if (tmpfs_dirent_dup(de)) { 1051 /* Remove duphead if de was last entry. */ 1052 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1053 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1054 MPASS(tmpfs_dirent_duphead(xde)); 1055 } else 1056 xde = NULL; 1057 LIST_REMOVE(de, uh.td_dup.entries); 1058 LIST_REMOVE(de, uh.td_dup.index_entries); 1059 if (xde != NULL) { 1060 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1061 RB_REMOVE(tmpfs_dir, head, xde); 1062 tmp = VFS_TO_TMPFS(vp->v_mount); 1063 MPASS(xde->td_node == NULL); 1064 tmpfs_free_dirent(tmp, xde); 1065 } 1066 } 1067 de->td_cookie = de->td_hash; 1068 } else 1069 RB_REMOVE(tmpfs_dir, head, de); 1070 1071 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1072 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1073 TMPFS_NODE_MODIFIED; 1074 tmpfs_update(vp); 1075 } 1076 1077 void 1078 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1079 { 1080 struct tmpfs_dirent *de, *dde, *nde; 1081 1082 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1083 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1084 /* Node may already be destroyed. */ 1085 de->td_node = NULL; 1086 if (tmpfs_dirent_duphead(de)) { 1087 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1088 LIST_REMOVE(dde, uh.td_dup.entries); 1089 dde->td_node = NULL; 1090 tmpfs_free_dirent(tmp, dde); 1091 } 1092 } 1093 tmpfs_free_dirent(tmp, de); 1094 } 1095 } 1096 1097 /* 1098 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1099 * directory and returns it in the uio space. The function returns 0 1100 * on success, -1 if there was not enough space in the uio structure to 1101 * hold the directory entry or an appropriate error code if another 1102 * error happens. 1103 */ 1104 static int 1105 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1106 { 1107 int error; 1108 struct dirent dent; 1109 1110 TMPFS_VALIDATE_DIR(node); 1111 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1112 1113 dent.d_fileno = node->tn_id; 1114 dent.d_type = DT_DIR; 1115 dent.d_namlen = 1; 1116 dent.d_name[0] = '.'; 1117 dent.d_name[1] = '\0'; 1118 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1119 1120 if (dent.d_reclen > uio->uio_resid) 1121 error = EJUSTRETURN; 1122 else 1123 error = uiomove(&dent, dent.d_reclen, uio); 1124 1125 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1126 1127 return (error); 1128 } 1129 1130 /* 1131 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1132 * directory and returns it in the uio space. The function returns 0 1133 * on success, -1 if there was not enough space in the uio structure to 1134 * hold the directory entry or an appropriate error code if another 1135 * error happens. 1136 */ 1137 static int 1138 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1139 { 1140 int error; 1141 struct dirent dent; 1142 1143 TMPFS_VALIDATE_DIR(node); 1144 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1145 1146 /* 1147 * Return ENOENT if the current node is already removed. 1148 */ 1149 TMPFS_ASSERT_LOCKED(node); 1150 if (node->tn_dir.tn_parent == NULL) 1151 return (ENOENT); 1152 1153 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1154 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1155 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1156 1157 dent.d_type = DT_DIR; 1158 dent.d_namlen = 2; 1159 dent.d_name[0] = '.'; 1160 dent.d_name[1] = '.'; 1161 dent.d_name[2] = '\0'; 1162 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1163 1164 if (dent.d_reclen > uio->uio_resid) 1165 error = EJUSTRETURN; 1166 else 1167 error = uiomove(&dent, dent.d_reclen, uio); 1168 1169 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1170 1171 return (error); 1172 } 1173 1174 /* 1175 * Helper function for tmpfs_readdir. Returns as much directory entries 1176 * as can fit in the uio space. The read starts at uio->uio_offset. 1177 * The function returns 0 on success, -1 if there was not enough space 1178 * in the uio structure to hold the directory entry or an appropriate 1179 * error code if another error happens. 1180 */ 1181 int 1182 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1183 u_long *cookies, int *ncookies) 1184 { 1185 struct tmpfs_dir_cursor dc; 1186 struct tmpfs_dirent *de; 1187 off_t off; 1188 int error; 1189 1190 TMPFS_VALIDATE_DIR(node); 1191 1192 off = 0; 1193 1194 /* 1195 * Lookup the node from the current offset. The starting offset of 1196 * 0 will lookup both '.' and '..', and then the first real entry, 1197 * or EOF if there are none. Then find all entries for the dir that 1198 * fit into the buffer. Once no more entries are found (de == NULL), 1199 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1200 * call to return 0. 1201 */ 1202 switch (uio->uio_offset) { 1203 case TMPFS_DIRCOOKIE_DOT: 1204 error = tmpfs_dir_getdotdent(node, uio); 1205 if (error != 0) 1206 return (error); 1207 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1208 if (cookies != NULL) 1209 cookies[(*ncookies)++] = off = uio->uio_offset; 1210 /* FALLTHROUGH */ 1211 case TMPFS_DIRCOOKIE_DOTDOT: 1212 error = tmpfs_dir_getdotdotdent(node, uio); 1213 if (error != 0) 1214 return (error); 1215 de = tmpfs_dir_first(node, &dc); 1216 uio->uio_offset = tmpfs_dirent_cookie(de); 1217 if (cookies != NULL) 1218 cookies[(*ncookies)++] = off = uio->uio_offset; 1219 /* EOF. */ 1220 if (de == NULL) 1221 return (0); 1222 break; 1223 case TMPFS_DIRCOOKIE_EOF: 1224 return (0); 1225 default: 1226 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1227 if (de == NULL) 1228 return (EINVAL); 1229 if (cookies != NULL) 1230 off = tmpfs_dirent_cookie(de); 1231 } 1232 1233 /* Read as much entries as possible; i.e., until we reach the end of 1234 * the directory or we exhaust uio space. */ 1235 do { 1236 struct dirent d; 1237 1238 /* Create a dirent structure representing the current 1239 * tmpfs_node and fill it. */ 1240 if (de->td_node == NULL) { 1241 d.d_fileno = 1; 1242 d.d_type = DT_WHT; 1243 } else { 1244 d.d_fileno = de->td_node->tn_id; 1245 switch (de->td_node->tn_type) { 1246 case VBLK: 1247 d.d_type = DT_BLK; 1248 break; 1249 1250 case VCHR: 1251 d.d_type = DT_CHR; 1252 break; 1253 1254 case VDIR: 1255 d.d_type = DT_DIR; 1256 break; 1257 1258 case VFIFO: 1259 d.d_type = DT_FIFO; 1260 break; 1261 1262 case VLNK: 1263 d.d_type = DT_LNK; 1264 break; 1265 1266 case VREG: 1267 d.d_type = DT_REG; 1268 break; 1269 1270 case VSOCK: 1271 d.d_type = DT_SOCK; 1272 break; 1273 1274 default: 1275 panic("tmpfs_dir_getdents: type %p %d", 1276 de->td_node, (int)de->td_node->tn_type); 1277 } 1278 } 1279 d.d_namlen = de->td_namelen; 1280 MPASS(de->td_namelen < sizeof(d.d_name)); 1281 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1282 d.d_name[de->td_namelen] = '\0'; 1283 d.d_reclen = GENERIC_DIRSIZ(&d); 1284 1285 /* Stop reading if the directory entry we are treating is 1286 * bigger than the amount of data that can be returned. */ 1287 if (d.d_reclen > uio->uio_resid) { 1288 error = EJUSTRETURN; 1289 break; 1290 } 1291 1292 /* Copy the new dirent structure into the output buffer and 1293 * advance pointers. */ 1294 error = uiomove(&d, d.d_reclen, uio); 1295 if (error == 0) { 1296 de = tmpfs_dir_next(node, &dc); 1297 if (cookies != NULL) { 1298 off = tmpfs_dirent_cookie(de); 1299 MPASS(*ncookies < maxcookies); 1300 cookies[(*ncookies)++] = off; 1301 } 1302 } 1303 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1304 1305 /* Skip setting off when using cookies as it is already done above. */ 1306 if (cookies == NULL) 1307 off = tmpfs_dirent_cookie(de); 1308 1309 /* Update the offset and cache. */ 1310 uio->uio_offset = off; 1311 node->tn_dir.tn_readdir_lastn = off; 1312 node->tn_dir.tn_readdir_lastp = de; 1313 1314 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1315 return error; 1316 } 1317 1318 int 1319 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1320 { 1321 struct tmpfs_dirent *de; 1322 int error; 1323 1324 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1325 cnp->cn_nameptr, cnp->cn_namelen, &de); 1326 if (error != 0) 1327 return (error); 1328 tmpfs_dir_attach(dvp, de); 1329 return (0); 1330 } 1331 1332 void 1333 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1334 { 1335 struct tmpfs_dirent *de; 1336 1337 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1338 MPASS(de != NULL && de->td_node == NULL); 1339 tmpfs_dir_detach(dvp, de); 1340 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1341 } 1342 1343 /* 1344 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1345 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1346 * 'newsize' must be positive. 1347 * 1348 * Returns zero on success or an appropriate error code on failure. 1349 */ 1350 int 1351 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1352 { 1353 struct tmpfs_mount *tmp; 1354 struct tmpfs_node *node; 1355 vm_object_t uobj; 1356 vm_page_t m; 1357 vm_pindex_t idx, newpages, oldpages; 1358 off_t oldsize; 1359 int base, rv; 1360 1361 MPASS(vp->v_type == VREG); 1362 MPASS(newsize >= 0); 1363 1364 node = VP_TO_TMPFS_NODE(vp); 1365 uobj = node->tn_reg.tn_aobj; 1366 tmp = VFS_TO_TMPFS(vp->v_mount); 1367 1368 /* 1369 * Convert the old and new sizes to the number of pages needed to 1370 * store them. It may happen that we do not need to do anything 1371 * because the last allocated page can accommodate the change on 1372 * its own. 1373 */ 1374 oldsize = node->tn_size; 1375 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1376 MPASS(oldpages == uobj->size); 1377 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1378 if (newpages > oldpages && 1379 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1380 return (ENOSPC); 1381 1382 VM_OBJECT_WLOCK(uobj); 1383 if (newsize < oldsize) { 1384 /* 1385 * Zero the truncated part of the last page. 1386 */ 1387 base = newsize & PAGE_MASK; 1388 if (base != 0) { 1389 idx = OFF_TO_IDX(newsize); 1390 retry: 1391 m = vm_page_lookup(uobj, idx); 1392 if (m != NULL) { 1393 if (vm_page_sleep_if_busy(m, "tmfssz")) 1394 goto retry; 1395 MPASS(m->valid == VM_PAGE_BITS_ALL); 1396 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1397 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1398 if (m == NULL) { 1399 VM_OBJECT_WUNLOCK(uobj); 1400 VM_WAIT; 1401 VM_OBJECT_WLOCK(uobj); 1402 goto retry; 1403 } 1404 rv = vm_pager_get_pages(uobj, &m, 1, NULL, 1405 NULL); 1406 vm_page_lock(m); 1407 if (rv == VM_PAGER_OK) { 1408 /* 1409 * Since the page was not resident, 1410 * and therefore not recently 1411 * accessed, immediately enqueue it 1412 * for asynchronous laundering. The 1413 * current operation is not regarded 1414 * as an access. 1415 */ 1416 vm_page_launder(m); 1417 vm_page_unlock(m); 1418 vm_page_xunbusy(m); 1419 } else { 1420 vm_page_free(m); 1421 vm_page_unlock(m); 1422 if (ignerr) 1423 m = NULL; 1424 else { 1425 VM_OBJECT_WUNLOCK(uobj); 1426 return (EIO); 1427 } 1428 } 1429 } 1430 if (m != NULL) { 1431 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1432 vm_page_dirty(m); 1433 vm_pager_page_unswapped(m); 1434 } 1435 } 1436 1437 /* 1438 * Release any swap space and free any whole pages. 1439 */ 1440 if (newpages < oldpages) { 1441 swap_pager_freespace(uobj, newpages, oldpages - 1442 newpages); 1443 vm_object_page_remove(uobj, newpages, 0, 0); 1444 } 1445 } 1446 uobj->size = newpages; 1447 VM_OBJECT_WUNLOCK(uobj); 1448 1449 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages); 1450 1451 node->tn_size = newsize; 1452 return (0); 1453 } 1454 1455 void 1456 tmpfs_check_mtime(struct vnode *vp) 1457 { 1458 struct tmpfs_node *node; 1459 struct vm_object *obj; 1460 1461 ASSERT_VOP_ELOCKED(vp, "check_mtime"); 1462 if (vp->v_type != VREG) 1463 return; 1464 obj = vp->v_object; 1465 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) == 1466 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj")); 1467 /* unlocked read */ 1468 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) { 1469 VM_OBJECT_WLOCK(obj); 1470 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) { 1471 obj->flags &= ~OBJ_TMPFS_DIRTY; 1472 node = VP_TO_TMPFS_NODE(vp); 1473 node->tn_status |= TMPFS_NODE_MODIFIED | 1474 TMPFS_NODE_CHANGED; 1475 } 1476 VM_OBJECT_WUNLOCK(obj); 1477 } 1478 } 1479 1480 /* 1481 * Change flags of the given vnode. 1482 * Caller should execute tmpfs_update on vp after a successful execution. 1483 * The vnode must be locked on entry and remain locked on exit. 1484 */ 1485 int 1486 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1487 struct thread *p) 1488 { 1489 int error; 1490 struct tmpfs_node *node; 1491 1492 ASSERT_VOP_ELOCKED(vp, "chflags"); 1493 1494 node = VP_TO_TMPFS_NODE(vp); 1495 1496 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1497 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1498 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1499 UF_SPARSE | UF_SYSTEM)) != 0) 1500 return (EOPNOTSUPP); 1501 1502 /* Disallow this operation if the file system is mounted read-only. */ 1503 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1504 return EROFS; 1505 1506 /* 1507 * Callers may only modify the file flags on objects they 1508 * have VADMIN rights for. 1509 */ 1510 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1511 return (error); 1512 /* 1513 * Unprivileged processes are not permitted to unset system 1514 * flags, or modify flags if any system flags are set. 1515 */ 1516 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1517 if (node->tn_flags & 1518 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1519 error = securelevel_gt(cred, 0); 1520 if (error) 1521 return (error); 1522 } 1523 } else { 1524 if (node->tn_flags & 1525 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1526 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1527 return (EPERM); 1528 } 1529 node->tn_flags = flags; 1530 node->tn_status |= TMPFS_NODE_CHANGED; 1531 1532 ASSERT_VOP_ELOCKED(vp, "chflags2"); 1533 1534 return (0); 1535 } 1536 1537 /* 1538 * Change access mode on the given vnode. 1539 * Caller should execute tmpfs_update on vp after a successful execution. 1540 * The vnode must be locked on entry and remain locked on exit. 1541 */ 1542 int 1543 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1544 { 1545 int error; 1546 struct tmpfs_node *node; 1547 1548 ASSERT_VOP_ELOCKED(vp, "chmod"); 1549 1550 node = VP_TO_TMPFS_NODE(vp); 1551 1552 /* Disallow this operation if the file system is mounted read-only. */ 1553 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1554 return EROFS; 1555 1556 /* Immutable or append-only files cannot be modified, either. */ 1557 if (node->tn_flags & (IMMUTABLE | APPEND)) 1558 return EPERM; 1559 1560 /* 1561 * To modify the permissions on a file, must possess VADMIN 1562 * for that file. 1563 */ 1564 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1565 return (error); 1566 1567 /* 1568 * Privileged processes may set the sticky bit on non-directories, 1569 * as well as set the setgid bit on a file with a group that the 1570 * process is not a member of. 1571 */ 1572 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1573 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1574 return (EFTYPE); 1575 } 1576 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1577 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1578 if (error) 1579 return (error); 1580 } 1581 1582 1583 node->tn_mode &= ~ALLPERMS; 1584 node->tn_mode |= mode & ALLPERMS; 1585 1586 node->tn_status |= TMPFS_NODE_CHANGED; 1587 1588 ASSERT_VOP_ELOCKED(vp, "chmod2"); 1589 1590 return (0); 1591 } 1592 1593 /* 1594 * Change ownership of the given vnode. At least one of uid or gid must 1595 * be different than VNOVAL. If one is set to that value, the attribute 1596 * is unchanged. 1597 * Caller should execute tmpfs_update on vp after a successful execution. 1598 * The vnode must be locked on entry and remain locked on exit. 1599 */ 1600 int 1601 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1602 struct thread *p) 1603 { 1604 int error; 1605 struct tmpfs_node *node; 1606 uid_t ouid; 1607 gid_t ogid; 1608 1609 ASSERT_VOP_ELOCKED(vp, "chown"); 1610 1611 node = VP_TO_TMPFS_NODE(vp); 1612 1613 /* Assign default values if they are unknown. */ 1614 MPASS(uid != VNOVAL || gid != VNOVAL); 1615 if (uid == VNOVAL) 1616 uid = node->tn_uid; 1617 if (gid == VNOVAL) 1618 gid = node->tn_gid; 1619 MPASS(uid != VNOVAL && gid != VNOVAL); 1620 1621 /* Disallow this operation if the file system is mounted read-only. */ 1622 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1623 return EROFS; 1624 1625 /* Immutable or append-only files cannot be modified, either. */ 1626 if (node->tn_flags & (IMMUTABLE | APPEND)) 1627 return EPERM; 1628 1629 /* 1630 * To modify the ownership of a file, must possess VADMIN for that 1631 * file. 1632 */ 1633 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1634 return (error); 1635 1636 /* 1637 * To change the owner of a file, or change the group of a file to a 1638 * group of which we are not a member, the caller must have 1639 * privilege. 1640 */ 1641 if ((uid != node->tn_uid || 1642 (gid != node->tn_gid && !groupmember(gid, cred))) && 1643 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1644 return (error); 1645 1646 ogid = node->tn_gid; 1647 ouid = node->tn_uid; 1648 1649 node->tn_uid = uid; 1650 node->tn_gid = gid; 1651 1652 node->tn_status |= TMPFS_NODE_CHANGED; 1653 1654 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1655 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1656 node->tn_mode &= ~(S_ISUID | S_ISGID); 1657 } 1658 1659 ASSERT_VOP_ELOCKED(vp, "chown2"); 1660 1661 return (0); 1662 } 1663 1664 /* 1665 * Change size of the given vnode. 1666 * Caller should execute tmpfs_update on vp after a successful execution. 1667 * The vnode must be locked on entry and remain locked on exit. 1668 */ 1669 int 1670 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1671 struct thread *p) 1672 { 1673 int error; 1674 struct tmpfs_node *node; 1675 1676 ASSERT_VOP_ELOCKED(vp, "chsize"); 1677 1678 node = VP_TO_TMPFS_NODE(vp); 1679 1680 /* Decide whether this is a valid operation based on the file type. */ 1681 error = 0; 1682 switch (vp->v_type) { 1683 case VDIR: 1684 return EISDIR; 1685 1686 case VREG: 1687 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1688 return EROFS; 1689 break; 1690 1691 case VBLK: 1692 /* FALLTHROUGH */ 1693 case VCHR: 1694 /* FALLTHROUGH */ 1695 case VFIFO: 1696 /* Allow modifications of special files even if in the file 1697 * system is mounted read-only (we are not modifying the 1698 * files themselves, but the objects they represent). */ 1699 return 0; 1700 1701 default: 1702 /* Anything else is unsupported. */ 1703 return EOPNOTSUPP; 1704 } 1705 1706 /* Immutable or append-only files cannot be modified, either. */ 1707 if (node->tn_flags & (IMMUTABLE | APPEND)) 1708 return EPERM; 1709 1710 error = tmpfs_truncate(vp, size); 1711 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1712 * for us, as will update tn_status; no need to do that here. */ 1713 1714 ASSERT_VOP_ELOCKED(vp, "chsize2"); 1715 1716 return (error); 1717 } 1718 1719 /* 1720 * Change access and modification times of the given vnode. 1721 * Caller should execute tmpfs_update on vp after a successful execution. 1722 * The vnode must be locked on entry and remain locked on exit. 1723 */ 1724 int 1725 tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 1726 struct ucred *cred, struct thread *l) 1727 { 1728 int error; 1729 struct tmpfs_node *node; 1730 1731 ASSERT_VOP_ELOCKED(vp, "chtimes"); 1732 1733 node = VP_TO_TMPFS_NODE(vp); 1734 1735 /* Disallow this operation if the file system is mounted read-only. */ 1736 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1737 return EROFS; 1738 1739 /* Immutable or append-only files cannot be modified, either. */ 1740 if (node->tn_flags & (IMMUTABLE | APPEND)) 1741 return EPERM; 1742 1743 error = vn_utimes_perm(vp, vap, cred, l); 1744 if (error != 0) 1745 return (error); 1746 1747 if (vap->va_atime.tv_sec != VNOVAL) 1748 node->tn_status |= TMPFS_NODE_ACCESSED; 1749 1750 if (vap->va_mtime.tv_sec != VNOVAL) 1751 node->tn_status |= TMPFS_NODE_MODIFIED; 1752 1753 if (vap->va_birthtime.tv_sec != VNOVAL) 1754 node->tn_status |= TMPFS_NODE_MODIFIED; 1755 1756 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1757 1758 if (vap->va_birthtime.tv_sec != VNOVAL) 1759 node->tn_birthtime = vap->va_birthtime; 1760 ASSERT_VOP_ELOCKED(vp, "chtimes2"); 1761 1762 return (0); 1763 } 1764 1765 void 1766 tmpfs_set_status(struct tmpfs_node *node, int status) 1767 { 1768 1769 if ((node->tn_status & status) == status) 1770 return; 1771 TMPFS_NODE_LOCK(node); 1772 node->tn_status |= status; 1773 TMPFS_NODE_UNLOCK(node); 1774 } 1775 1776 /* Sync timestamps */ 1777 void 1778 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1779 const struct timespec *mod) 1780 { 1781 struct tmpfs_node *node; 1782 struct timespec now; 1783 1784 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes"); 1785 node = VP_TO_TMPFS_NODE(vp); 1786 1787 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1788 TMPFS_NODE_CHANGED)) == 0) 1789 return; 1790 1791 vfs_timestamp(&now); 1792 TMPFS_NODE_LOCK(node); 1793 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1794 if (acc == NULL) 1795 acc = &now; 1796 node->tn_atime = *acc; 1797 } 1798 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1799 if (mod == NULL) 1800 mod = &now; 1801 node->tn_mtime = *mod; 1802 } 1803 if (node->tn_status & TMPFS_NODE_CHANGED) 1804 node->tn_ctime = now; 1805 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1806 TMPFS_NODE_CHANGED); 1807 TMPFS_NODE_UNLOCK(node); 1808 1809 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */ 1810 random_harvest_queue(node, sizeof(*node), 1, RANDOM_FS_ATIME); 1811 } 1812 1813 void 1814 tmpfs_update(struct vnode *vp) 1815 { 1816 1817 tmpfs_itimes(vp, NULL, NULL); 1818 } 1819 1820 int 1821 tmpfs_truncate(struct vnode *vp, off_t length) 1822 { 1823 int error; 1824 struct tmpfs_node *node; 1825 1826 node = VP_TO_TMPFS_NODE(vp); 1827 1828 if (length < 0) { 1829 error = EINVAL; 1830 goto out; 1831 } 1832 1833 if (node->tn_size == length) { 1834 error = 0; 1835 goto out; 1836 } 1837 1838 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1839 return (EFBIG); 1840 1841 error = tmpfs_reg_resize(vp, length, FALSE); 1842 if (error == 0) 1843 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1844 1845 out: 1846 tmpfs_update(vp); 1847 1848 return (error); 1849 } 1850 1851 static __inline int 1852 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1853 { 1854 if (a->td_hash > b->td_hash) 1855 return (1); 1856 else if (a->td_hash < b->td_hash) 1857 return (-1); 1858 return (0); 1859 } 1860 1861 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1862