tmpfs_subr.c revision 313093
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: stable/10/sys/fs/tmpfs/tmpfs_subr.c 313093 2017-02-02 13:37:00Z kib $"); 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/rwlock.h> 46#include <sys/stat.h> 47#include <sys/systm.h> 48#include <sys/sysctl.h> 49#include <sys/vnode.h> 50#include <sys/vmmeter.h> 51 52#include <vm/vm.h> 53#include <vm/vm_param.h> 54#include <vm/vm_object.h> 55#include <vm/vm_page.h> 56#include <vm/vm_pageout.h> 57#include <vm/vm_pager.h> 58#include <vm/vm_extern.h> 59 60#include <fs/tmpfs/tmpfs.h> 61#include <fs/tmpfs/tmpfs_fifoops.h> 62#include <fs/tmpfs/tmpfs_vnops.h> 63 64SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system"); 65 66static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED; 67 68static int 69sysctl_mem_reserved(SYSCTL_HANDLER_ARGS) 70{ 71 int error; 72 long pages, bytes; 73 74 pages = *(long *)arg1; 75 bytes = pages * PAGE_SIZE; 76 77 error = sysctl_handle_long(oidp, &bytes, 0, req); 78 if (error || !req->newptr) 79 return (error); 80 81 pages = bytes / PAGE_SIZE; 82 if (pages < TMPFS_PAGES_MINRESERVED) 83 return (EINVAL); 84 85 *(long *)arg1 = pages; 86 return (0); 87} 88 89SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW, 90 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L", 91 "Amount of available memory and swap below which tmpfs growth stops"); 92 93static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a, 94 struct tmpfs_dirent *b); 95RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 96 97size_t 98tmpfs_mem_avail(void) 99{ 100 vm_ooffset_t avail; 101 102 avail = swap_pager_avail + cnt.v_free_count + cnt.v_cache_count - 103 tmpfs_pages_reserved; 104 if (__predict_false(avail < 0)) 105 avail = 0; 106 return (avail); 107} 108 109size_t 110tmpfs_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 121static size_t 122tmpfs_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 134void 135tmpfs_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 143void 144tmpfs_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 */ 176int 177tmpfs_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 */ 295void 296tmpfs_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 307bool 308tmpfs_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 371static __inline uint32_t 372tmpfs_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 386static __inline off_t 387tmpfs_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 397static __inline boolean_t 398tmpfs_dirent_dup(struct tmpfs_dirent *de) 399{ 400 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0); 401} 402 403static __inline boolean_t 404tmpfs_dirent_duphead(struct tmpfs_dirent *de) 405{ 406 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0); 407} 408 409void 410tmpfs_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 */ 426int 427tmpfs_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 */ 456void 457tmpfs_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 471void 472tmpfs_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 */ 490static void 491tmpfs_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 */ 509int 510tmpfs_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); 522loop: 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, &tmpfs_vnodeop_entries, &vp); 595 if (error != 0) 596 goto unlock; 597 MPASS(vp != NULL); 598 599 /* lkflag is ignored, the lock is exclusive */ 600 (void) vn_lock(vp, lkflag | LK_RETRY); 601 602 vp->v_data = node; 603 vp->v_type = node->tn_type; 604 605 /* Type-specific initialization. */ 606 switch (node->tn_type) { 607 case VBLK: 608 /* FALLTHROUGH */ 609 case VCHR: 610 /* FALLTHROUGH */ 611 case VLNK: 612 /* FALLTHROUGH */ 613 case VSOCK: 614 break; 615 case VFIFO: 616 vp->v_op = &tmpfs_fifoop_entries; 617 break; 618 case VREG: 619 object = node->tn_reg.tn_aobj; 620 VM_OBJECT_WLOCK(object); 621 VI_LOCK(vp); 622 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 623 vp->v_object = object; 624 object->un_pager.swp.swp_tmpfs = vp; 625 vm_object_set_flag(object, OBJ_TMPFS); 626 VI_UNLOCK(vp); 627 VM_OBJECT_WUNLOCK(object); 628 break; 629 case VDIR: 630 MPASS(node->tn_dir.tn_parent != NULL); 631 if (node->tn_dir.tn_parent == node) 632 vp->v_vflag |= VV_ROOT; 633 break; 634 635 default: 636 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 637 } 638 if (vp->v_type != VFIFO) 639 VN_LOCK_ASHARE(vp); 640 641 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 642 if (error != 0) 643 vp = NULL; 644 645unlock: 646 TMPFS_NODE_LOCK(node); 647 648 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 649 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 650 node->tn_vnode = vp; 651 652 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 653 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 654 TMPFS_NODE_UNLOCK(node); 655 wakeup((caddr_t) &node->tn_vpstate); 656 } else 657 TMPFS_NODE_UNLOCK(node); 658 659out: 660 if (error == 0) { 661 *vpp = vp; 662 663#ifdef INVARIANTS 664 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 665 TMPFS_NODE_LOCK(node); 666 MPASS(*vpp == node->tn_vnode); 667 TMPFS_NODE_UNLOCK(node); 668#endif 669 } 670 tmpfs_free_node(tm, node); 671 672 return (error); 673} 674 675/* 676 * Destroys the association between the vnode vp and the node it 677 * references. 678 */ 679void 680tmpfs_free_vp(struct vnode *vp) 681{ 682 struct tmpfs_node *node; 683 684 node = VP_TO_TMPFS_NODE(vp); 685 686 TMPFS_NODE_ASSERT_LOCKED(node); 687 node->tn_vnode = NULL; 688 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 689 wakeup(&node->tn_vnode); 690 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 691 vp->v_data = NULL; 692} 693 694/* 695 * Allocates a new file of type 'type' and adds it to the parent directory 696 * 'dvp'; this addition is done using the component name given in 'cnp'. 697 * The ownership of the new file is automatically assigned based on the 698 * credentials of the caller (through 'cnp'), the group is set based on 699 * the parent directory and the mode is determined from the 'vap' argument. 700 * If successful, *vpp holds a vnode to the newly created file and zero 701 * is returned. Otherwise *vpp is NULL and the function returns an 702 * appropriate error code. 703 */ 704int 705tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 706 struct componentname *cnp, char *target) 707{ 708 int error; 709 struct tmpfs_dirent *de; 710 struct tmpfs_mount *tmp; 711 struct tmpfs_node *dnode; 712 struct tmpfs_node *node; 713 struct tmpfs_node *parent; 714 715 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file"); 716 MPASS(cnp->cn_flags & HASBUF); 717 718 tmp = VFS_TO_TMPFS(dvp->v_mount); 719 dnode = VP_TO_TMPFS_DIR(dvp); 720 *vpp = NULL; 721 722 /* If the entry we are creating is a directory, we cannot overflow 723 * the number of links of its parent, because it will get a new 724 * link. */ 725 if (vap->va_type == VDIR) { 726 /* Ensure that we do not overflow the maximum number of links 727 * imposed by the system. */ 728 MPASS(dnode->tn_links <= LINK_MAX); 729 if (dnode->tn_links == LINK_MAX) { 730 return (EMLINK); 731 } 732 733 parent = dnode; 734 MPASS(parent != NULL); 735 } else 736 parent = NULL; 737 738 /* Allocate a node that represents the new file. */ 739 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type, 740 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent, 741 target, vap->va_rdev, &node); 742 if (error != 0) 743 return (error); 744 745 /* Allocate a directory entry that points to the new file. */ 746 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 747 &de); 748 if (error != 0) { 749 tmpfs_free_node(tmp, node); 750 return (error); 751 } 752 753 /* Allocate a vnode for the new file. */ 754 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 755 if (error != 0) { 756 tmpfs_free_dirent(tmp, de); 757 tmpfs_free_node(tmp, node); 758 return (error); 759 } 760 761 /* Now that all required items are allocated, we can proceed to 762 * insert the new node into the directory, an operation that 763 * cannot fail. */ 764 if (cnp->cn_flags & ISWHITEOUT) 765 tmpfs_dir_whiteout_remove(dvp, cnp); 766 tmpfs_dir_attach(dvp, de); 767 return (0); 768} 769 770struct tmpfs_dirent * 771tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 772{ 773 struct tmpfs_dirent *de; 774 775 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 776 dc->tdc_tree = de; 777 if (de != NULL && tmpfs_dirent_duphead(de)) 778 de = LIST_FIRST(&de->ud.td_duphead); 779 dc->tdc_current = de; 780 781 return (dc->tdc_current); 782} 783 784struct tmpfs_dirent * 785tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 786{ 787 struct tmpfs_dirent *de; 788 789 MPASS(dc->tdc_tree != NULL); 790 if (tmpfs_dirent_dup(dc->tdc_current)) { 791 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 792 if (dc->tdc_current != NULL) 793 return (dc->tdc_current); 794 } 795 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 796 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 797 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 798 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 799 MPASS(dc->tdc_current != NULL); 800 } 801 802 return (dc->tdc_current); 803} 804 805/* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 806static struct tmpfs_dirent * 807tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 808{ 809 struct tmpfs_dirent *de, dekey; 810 811 dekey.td_hash = hash; 812 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 813 return (de); 814} 815 816/* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 817static struct tmpfs_dirent * 818tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 819 struct tmpfs_dir_cursor *dc) 820{ 821 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 822 struct tmpfs_dirent *de, dekey; 823 824 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 825 826 if (cookie == node->tn_dir.tn_readdir_lastn && 827 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 828 /* Protect against possible race, tn_readdir_last[pn] 829 * may be updated with only shared vnode lock held. */ 830 if (cookie == tmpfs_dirent_cookie(de)) 831 goto out; 832 } 833 834 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 835 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 836 uh.td_dup.index_entries) { 837 MPASS(tmpfs_dirent_dup(de)); 838 if (de->td_cookie == cookie) 839 goto out; 840 /* dupindex list is sorted. */ 841 if (de->td_cookie < cookie) { 842 de = NULL; 843 goto out; 844 } 845 } 846 MPASS(de == NULL); 847 goto out; 848 } 849 850 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) { 851 de = NULL; 852 } else { 853 dekey.td_hash = cookie; 854 /* Recover if direntry for cookie was removed */ 855 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 856 } 857 dc->tdc_tree = de; 858 dc->tdc_current = de; 859 if (de != NULL && tmpfs_dirent_duphead(de)) { 860 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 861 MPASS(dc->tdc_current != NULL); 862 } 863 return (dc->tdc_current); 864 865out: 866 dc->tdc_tree = de; 867 dc->tdc_current = de; 868 if (de != NULL && tmpfs_dirent_dup(de)) 869 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 870 de->td_hash); 871 return (dc->tdc_current); 872} 873 874/* 875 * Looks for a directory entry in the directory represented by node. 876 * 'cnp' describes the name of the entry to look for. Note that the . 877 * and .. components are not allowed as they do not physically exist 878 * within directories. 879 * 880 * Returns a pointer to the entry when found, otherwise NULL. 881 */ 882struct tmpfs_dirent * 883tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 884 struct componentname *cnp) 885{ 886 struct tmpfs_dir_duphead *duphead; 887 struct tmpfs_dirent *de; 888 uint32_t hash; 889 890 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 891 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 892 cnp->cn_nameptr[1] == '.'))); 893 TMPFS_VALIDATE_DIR(node); 894 895 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 896 de = tmpfs_dir_xlookup_hash(node, hash); 897 if (de != NULL && tmpfs_dirent_duphead(de)) { 898 duphead = &de->ud.td_duphead; 899 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 900 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 901 cnp->cn_namelen)) 902 break; 903 } 904 } else if (de != NULL) { 905 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 906 cnp->cn_namelen)) 907 de = NULL; 908 } 909 if (de != NULL && f != NULL && de->td_node != f) 910 de = NULL; 911 912 return (de); 913} 914 915/* 916 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 917 * list, allocate new cookie value. 918 */ 919static void 920tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 921 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 922{ 923 struct tmpfs_dir_duphead *dupindex; 924 struct tmpfs_dirent *de, *pde; 925 926 dupindex = &dnode->tn_dir.tn_dupindex; 927 de = LIST_FIRST(dupindex); 928 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 929 if (de == NULL) 930 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 931 else 932 nde->td_cookie = de->td_cookie + 1; 933 MPASS(tmpfs_dirent_dup(nde)); 934 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 935 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 936 return; 937 } 938 939 /* 940 * Cookie numbers are near exhaustion. Scan dupindex list for unused 941 * numbers. dupindex list is sorted in descending order. Keep it so 942 * after inserting nde. 943 */ 944 while (1) { 945 pde = de; 946 de = LIST_NEXT(de, uh.td_dup.index_entries); 947 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 948 /* 949 * Last element of the index doesn't have minimal cookie 950 * value, use it. 951 */ 952 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 953 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 954 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 955 return; 956 } else if (de == NULL) { 957 /* 958 * We are so lucky have 2^30 hash duplicates in single 959 * directory :) Return largest possible cookie value. 960 * It should be fine except possible issues with 961 * VOP_READDIR restart. 962 */ 963 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 964 LIST_INSERT_HEAD(dupindex, nde, 965 uh.td_dup.index_entries); 966 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 967 return; 968 } 969 if (de->td_cookie + 1 == pde->td_cookie || 970 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 971 continue; /* No hole or invalid cookie. */ 972 nde->td_cookie = de->td_cookie + 1; 973 MPASS(tmpfs_dirent_dup(nde)); 974 MPASS(pde->td_cookie > nde->td_cookie); 975 MPASS(nde->td_cookie > de->td_cookie); 976 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 977 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 978 return; 979 }; 980} 981 982/* 983 * Attaches the directory entry de to the directory represented by vp. 984 * Note that this does not change the link count of the node pointed by 985 * the directory entry, as this is done by tmpfs_alloc_dirent. 986 */ 987void 988tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 989{ 990 struct tmpfs_node *dnode; 991 struct tmpfs_dirent *xde, *nde; 992 993 ASSERT_VOP_ELOCKED(vp, __func__); 994 MPASS(de->td_namelen > 0); 995 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 996 MPASS(de->td_cookie == de->td_hash); 997 998 dnode = VP_TO_TMPFS_DIR(vp); 999 dnode->tn_dir.tn_readdir_lastn = 0; 1000 dnode->tn_dir.tn_readdir_lastp = NULL; 1001 1002 MPASS(!tmpfs_dirent_dup(de)); 1003 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1004 if (xde != NULL && tmpfs_dirent_duphead(xde)) 1005 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1006 else if (xde != NULL) { 1007 /* 1008 * Allocate new duphead. Swap xde with duphead to avoid 1009 * adding/removing elements with the same hash. 1010 */ 1011 MPASS(!tmpfs_dirent_dup(xde)); 1012 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 1013 &nde); 1014 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 1015 memcpy(nde, xde, sizeof(*xde)); 1016 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 1017 LIST_INIT(&xde->ud.td_duphead); 1018 xde->td_namelen = 0; 1019 xde->td_node = NULL; 1020 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 1021 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1022 } 1023 dnode->tn_size += sizeof(struct tmpfs_dirent); 1024 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1025 TMPFS_NODE_MODIFIED; 1026 tmpfs_update(vp); 1027} 1028 1029/* 1030 * Detaches the directory entry de from the directory represented by vp. 1031 * Note that this does not change the link count of the node pointed by 1032 * the directory entry, as this is done by tmpfs_free_dirent. 1033 */ 1034void 1035tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 1036{ 1037 struct tmpfs_mount *tmp; 1038 struct tmpfs_dir *head; 1039 struct tmpfs_node *dnode; 1040 struct tmpfs_dirent *xde; 1041 1042 ASSERT_VOP_ELOCKED(vp, __func__); 1043 1044 dnode = VP_TO_TMPFS_DIR(vp); 1045 head = &dnode->tn_dir.tn_dirhead; 1046 dnode->tn_dir.tn_readdir_lastn = 0; 1047 dnode->tn_dir.tn_readdir_lastp = NULL; 1048 1049 if (tmpfs_dirent_dup(de)) { 1050 /* Remove duphead if de was last entry. */ 1051 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1052 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1053 MPASS(tmpfs_dirent_duphead(xde)); 1054 } else 1055 xde = NULL; 1056 LIST_REMOVE(de, uh.td_dup.entries); 1057 LIST_REMOVE(de, uh.td_dup.index_entries); 1058 if (xde != NULL) { 1059 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1060 RB_REMOVE(tmpfs_dir, head, xde); 1061 tmp = VFS_TO_TMPFS(vp->v_mount); 1062 MPASS(xde->td_node == NULL); 1063 tmpfs_free_dirent(tmp, xde); 1064 } 1065 } 1066 de->td_cookie = de->td_hash; 1067 } else 1068 RB_REMOVE(tmpfs_dir, head, de); 1069 1070 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1071 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1072 TMPFS_NODE_MODIFIED; 1073 tmpfs_update(vp); 1074} 1075 1076void 1077tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1078{ 1079 struct tmpfs_dirent *de, *dde, *nde; 1080 1081 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1082 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1083 /* Node may already be destroyed. */ 1084 de->td_node = NULL; 1085 if (tmpfs_dirent_duphead(de)) { 1086 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1087 LIST_REMOVE(dde, uh.td_dup.entries); 1088 dde->td_node = NULL; 1089 tmpfs_free_dirent(tmp, dde); 1090 } 1091 } 1092 tmpfs_free_dirent(tmp, de); 1093 } 1094} 1095 1096/* 1097 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1098 * directory and returns it in the uio space. The function returns 0 1099 * on success, -1 if there was not enough space in the uio structure to 1100 * hold the directory entry or an appropriate error code if another 1101 * error happens. 1102 */ 1103static int 1104tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1105{ 1106 int error; 1107 struct dirent dent; 1108 1109 TMPFS_VALIDATE_DIR(node); 1110 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1111 1112 dent.d_fileno = node->tn_id; 1113 dent.d_type = DT_DIR; 1114 dent.d_namlen = 1; 1115 dent.d_name[0] = '.'; 1116 dent.d_name[1] = '\0'; 1117 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1118 1119 if (dent.d_reclen > uio->uio_resid) 1120 error = EJUSTRETURN; 1121 else 1122 error = uiomove(&dent, dent.d_reclen, uio); 1123 1124 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1125 1126 return (error); 1127} 1128 1129/* 1130 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1131 * directory and returns it in the uio space. The function returns 0 1132 * on success, -1 if there was not enough space in the uio structure to 1133 * hold the directory entry or an appropriate error code if another 1134 * error happens. 1135 */ 1136static int 1137tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1138{ 1139 int error; 1140 struct dirent dent; 1141 1142 TMPFS_VALIDATE_DIR(node); 1143 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1144 1145 /* 1146 * Return ENOENT if the current node is already removed. 1147 */ 1148 TMPFS_ASSERT_LOCKED(node); 1149 if (node->tn_dir.tn_parent == NULL) 1150 return (ENOENT); 1151 1152 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1153 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1154 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1155 1156 dent.d_type = DT_DIR; 1157 dent.d_namlen = 2; 1158 dent.d_name[0] = '.'; 1159 dent.d_name[1] = '.'; 1160 dent.d_name[2] = '\0'; 1161 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1162 1163 if (dent.d_reclen > uio->uio_resid) 1164 error = EJUSTRETURN; 1165 else 1166 error = uiomove(&dent, dent.d_reclen, uio); 1167 1168 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1169 1170 return (error); 1171} 1172 1173/* 1174 * Helper function for tmpfs_readdir. Returns as much directory entries 1175 * as can fit in the uio space. The read starts at uio->uio_offset. 1176 * The function returns 0 on success, -1 if there was not enough space 1177 * in the uio structure to hold the directory entry or an appropriate 1178 * error code if another error happens. 1179 */ 1180int 1181tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1182 u_long *cookies, int *ncookies) 1183{ 1184 struct tmpfs_dir_cursor dc; 1185 struct tmpfs_dirent *de; 1186 off_t off; 1187 int error; 1188 1189 TMPFS_VALIDATE_DIR(node); 1190 1191 off = 0; 1192 1193 /* 1194 * Lookup the node from the current offset. The starting offset of 1195 * 0 will lookup both '.' and '..', and then the first real entry, 1196 * or EOF if there are none. Then find all entries for the dir that 1197 * fit into the buffer. Once no more entries are found (de == NULL), 1198 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1199 * call to return 0. 1200 */ 1201 switch (uio->uio_offset) { 1202 case TMPFS_DIRCOOKIE_DOT: 1203 error = tmpfs_dir_getdotdent(node, uio); 1204 if (error != 0) 1205 return (error); 1206 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1207 if (cookies != NULL) 1208 cookies[(*ncookies)++] = off = uio->uio_offset; 1209 /* FALLTHROUGH */ 1210 case TMPFS_DIRCOOKIE_DOTDOT: 1211 error = tmpfs_dir_getdotdotdent(node, uio); 1212 if (error != 0) 1213 return (error); 1214 de = tmpfs_dir_first(node, &dc); 1215 uio->uio_offset = tmpfs_dirent_cookie(de); 1216 if (cookies != NULL) 1217 cookies[(*ncookies)++] = off = uio->uio_offset; 1218 /* EOF. */ 1219 if (de == NULL) 1220 return (0); 1221 break; 1222 case TMPFS_DIRCOOKIE_EOF: 1223 return (0); 1224 default: 1225 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1226 if (de == NULL) 1227 return (EINVAL); 1228 if (cookies != NULL) 1229 off = tmpfs_dirent_cookie(de); 1230 } 1231 1232 /* Read as much entries as possible; i.e., until we reach the end of 1233 * the directory or we exhaust uio space. */ 1234 do { 1235 struct dirent d; 1236 1237 /* Create a dirent structure representing the current 1238 * tmpfs_node and fill it. */ 1239 if (de->td_node == NULL) { 1240 d.d_fileno = 1; 1241 d.d_type = DT_WHT; 1242 } else { 1243 d.d_fileno = de->td_node->tn_id; 1244 switch (de->td_node->tn_type) { 1245 case VBLK: 1246 d.d_type = DT_BLK; 1247 break; 1248 1249 case VCHR: 1250 d.d_type = DT_CHR; 1251 break; 1252 1253 case VDIR: 1254 d.d_type = DT_DIR; 1255 break; 1256 1257 case VFIFO: 1258 d.d_type = DT_FIFO; 1259 break; 1260 1261 case VLNK: 1262 d.d_type = DT_LNK; 1263 break; 1264 1265 case VREG: 1266 d.d_type = DT_REG; 1267 break; 1268 1269 case VSOCK: 1270 d.d_type = DT_SOCK; 1271 break; 1272 1273 default: 1274 panic("tmpfs_dir_getdents: type %p %d", 1275 de->td_node, (int)de->td_node->tn_type); 1276 } 1277 } 1278 d.d_namlen = de->td_namelen; 1279 MPASS(de->td_namelen < sizeof(d.d_name)); 1280 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1281 d.d_name[de->td_namelen] = '\0'; 1282 d.d_reclen = GENERIC_DIRSIZ(&d); 1283 1284 /* Stop reading if the directory entry we are treating is 1285 * bigger than the amount of data that can be returned. */ 1286 if (d.d_reclen > uio->uio_resid) { 1287 error = EJUSTRETURN; 1288 break; 1289 } 1290 1291 /* Copy the new dirent structure into the output buffer and 1292 * advance pointers. */ 1293 error = uiomove(&d, d.d_reclen, uio); 1294 if (error == 0) { 1295 de = tmpfs_dir_next(node, &dc); 1296 if (cookies != NULL) { 1297 off = tmpfs_dirent_cookie(de); 1298 MPASS(*ncookies < maxcookies); 1299 cookies[(*ncookies)++] = off; 1300 } 1301 } 1302 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1303 1304 /* Skip setting off when using cookies as it is already done above. */ 1305 if (cookies == NULL) 1306 off = tmpfs_dirent_cookie(de); 1307 1308 /* Update the offset and cache. */ 1309 uio->uio_offset = off; 1310 node->tn_dir.tn_readdir_lastn = off; 1311 node->tn_dir.tn_readdir_lastp = de; 1312 1313 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1314 return error; 1315} 1316 1317int 1318tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1319{ 1320 struct tmpfs_dirent *de; 1321 int error; 1322 1323 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1324 cnp->cn_nameptr, cnp->cn_namelen, &de); 1325 if (error != 0) 1326 return (error); 1327 tmpfs_dir_attach(dvp, de); 1328 return (0); 1329} 1330 1331void 1332tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1333{ 1334 struct tmpfs_dirent *de; 1335 1336 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1337 MPASS(de != NULL && de->td_node == NULL); 1338 tmpfs_dir_detach(dvp, de); 1339 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1340} 1341 1342/* 1343 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1344 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1345 * 'newsize' must be positive. 1346 * 1347 * Returns zero on success or an appropriate error code on failure. 1348 */ 1349int 1350tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1351{ 1352 struct tmpfs_mount *tmp; 1353 struct tmpfs_node *node; 1354 vm_object_t uobj; 1355 vm_page_t m, ma[1]; 1356 vm_pindex_t idx, newpages, oldpages; 1357 off_t oldsize; 1358 int base, rv; 1359 1360 MPASS(vp->v_type == VREG); 1361 MPASS(newsize >= 0); 1362 1363 node = VP_TO_TMPFS_NODE(vp); 1364 uobj = node->tn_reg.tn_aobj; 1365 tmp = VFS_TO_TMPFS(vp->v_mount); 1366 1367 /* 1368 * Convert the old and new sizes to the number of pages needed to 1369 * store them. It may happen that we do not need to do anything 1370 * because the last allocated page can accommodate the change on 1371 * its own. 1372 */ 1373 oldsize = node->tn_size; 1374 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1375 MPASS(oldpages == uobj->size); 1376 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1377 if (newpages > oldpages && 1378 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1379 return (ENOSPC); 1380 1381 VM_OBJECT_WLOCK(uobj); 1382 if (newsize < oldsize) { 1383 /* 1384 * Zero the truncated part of the last page. 1385 */ 1386 base = newsize & PAGE_MASK; 1387 if (base != 0) { 1388 idx = OFF_TO_IDX(newsize); 1389retry: 1390 m = vm_page_lookup(uobj, idx); 1391 if (m != NULL) { 1392 if (vm_page_sleep_if_busy(m, "tmfssz")) 1393 goto retry; 1394 MPASS(m->valid == VM_PAGE_BITS_ALL); 1395 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1396 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1397 if (m == NULL) { 1398 VM_OBJECT_WUNLOCK(uobj); 1399 VM_WAIT; 1400 VM_OBJECT_WLOCK(uobj); 1401 goto retry; 1402 } else if (m->valid != VM_PAGE_BITS_ALL) { 1403 ma[0] = m; 1404 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1405 m = vm_page_lookup(uobj, idx); 1406 } else 1407 /* A cached page was reactivated. */ 1408 rv = VM_PAGER_OK; 1409 vm_page_lock(m); 1410 if (rv == VM_PAGER_OK) { 1411 vm_page_deactivate(m); 1412 vm_page_unlock(m); 1413 vm_page_xunbusy(m); 1414 } else { 1415 vm_page_free(m); 1416 vm_page_unlock(m); 1417 if (ignerr) 1418 m = NULL; 1419 else { 1420 VM_OBJECT_WUNLOCK(uobj); 1421 return (EIO); 1422 } 1423 } 1424 } 1425 if (m != NULL) { 1426 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1427 vm_page_dirty(m); 1428 vm_pager_page_unswapped(m); 1429 } 1430 } 1431 1432 /* 1433 * Release any swap space and free any whole pages. 1434 */ 1435 if (newpages < oldpages) { 1436 swap_pager_freespace(uobj, newpages, oldpages - 1437 newpages); 1438 vm_object_page_remove(uobj, newpages, 0, 0); 1439 } 1440 } 1441 uobj->size = newpages; 1442 VM_OBJECT_WUNLOCK(uobj); 1443 1444 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages); 1445 1446 node->tn_size = newsize; 1447 return (0); 1448} 1449 1450void 1451tmpfs_check_mtime(struct vnode *vp) 1452{ 1453 struct tmpfs_node *node; 1454 struct vm_object *obj; 1455 1456 ASSERT_VOP_ELOCKED(vp, "check_mtime"); 1457 if (vp->v_type != VREG) 1458 return; 1459 obj = vp->v_object; 1460 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) == 1461 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj")); 1462 /* unlocked read */ 1463 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) { 1464 VM_OBJECT_WLOCK(obj); 1465 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) { 1466 obj->flags &= ~OBJ_TMPFS_DIRTY; 1467 node = VP_TO_TMPFS_NODE(vp); 1468 node->tn_status |= TMPFS_NODE_MODIFIED | 1469 TMPFS_NODE_CHANGED; 1470 } 1471 VM_OBJECT_WUNLOCK(obj); 1472 } 1473} 1474 1475/* 1476 * Change flags of the given vnode. 1477 * Caller should execute tmpfs_update on vp after a successful execution. 1478 * The vnode must be locked on entry and remain locked on exit. 1479 */ 1480int 1481tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1482 struct thread *p) 1483{ 1484 int error; 1485 struct tmpfs_node *node; 1486 1487 ASSERT_VOP_ELOCKED(vp, "chflags"); 1488 1489 node = VP_TO_TMPFS_NODE(vp); 1490 1491 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1492 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1493 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1494 UF_SPARSE | UF_SYSTEM)) != 0) 1495 return (EOPNOTSUPP); 1496 1497 /* Disallow this operation if the file system is mounted read-only. */ 1498 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1499 return EROFS; 1500 1501 /* 1502 * Callers may only modify the file flags on objects they 1503 * have VADMIN rights for. 1504 */ 1505 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1506 return (error); 1507 /* 1508 * Unprivileged processes are not permitted to unset system 1509 * flags, or modify flags if any system flags are set. 1510 */ 1511 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1512 if (node->tn_flags & 1513 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1514 error = securelevel_gt(cred, 0); 1515 if (error) 1516 return (error); 1517 } 1518 } else { 1519 if (node->tn_flags & 1520 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1521 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1522 return (EPERM); 1523 } 1524 node->tn_flags = flags; 1525 node->tn_status |= TMPFS_NODE_CHANGED; 1526 1527 ASSERT_VOP_ELOCKED(vp, "chflags2"); 1528 1529 return (0); 1530} 1531 1532/* 1533 * Change access mode on the given vnode. 1534 * Caller should execute tmpfs_update on vp after a successful execution. 1535 * The vnode must be locked on entry and remain locked on exit. 1536 */ 1537int 1538tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1539{ 1540 int error; 1541 struct tmpfs_node *node; 1542 1543 ASSERT_VOP_ELOCKED(vp, "chmod"); 1544 1545 node = VP_TO_TMPFS_NODE(vp); 1546 1547 /* Disallow this operation if the file system is mounted read-only. */ 1548 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1549 return EROFS; 1550 1551 /* Immutable or append-only files cannot be modified, either. */ 1552 if (node->tn_flags & (IMMUTABLE | APPEND)) 1553 return EPERM; 1554 1555 /* 1556 * To modify the permissions on a file, must possess VADMIN 1557 * for that file. 1558 */ 1559 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1560 return (error); 1561 1562 /* 1563 * Privileged processes may set the sticky bit on non-directories, 1564 * as well as set the setgid bit on a file with a group that the 1565 * process is not a member of. 1566 */ 1567 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1568 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1569 return (EFTYPE); 1570 } 1571 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1572 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1573 if (error) 1574 return (error); 1575 } 1576 1577 1578 node->tn_mode &= ~ALLPERMS; 1579 node->tn_mode |= mode & ALLPERMS; 1580 1581 node->tn_status |= TMPFS_NODE_CHANGED; 1582 1583 ASSERT_VOP_ELOCKED(vp, "chmod2"); 1584 1585 return (0); 1586} 1587 1588/* 1589 * Change ownership of the given vnode. At least one of uid or gid must 1590 * be different than VNOVAL. If one is set to that value, the attribute 1591 * is unchanged. 1592 * Caller should execute tmpfs_update on vp after a successful execution. 1593 * The vnode must be locked on entry and remain locked on exit. 1594 */ 1595int 1596tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1597 struct thread *p) 1598{ 1599 int error; 1600 struct tmpfs_node *node; 1601 uid_t ouid; 1602 gid_t ogid; 1603 1604 ASSERT_VOP_ELOCKED(vp, "chown"); 1605 1606 node = VP_TO_TMPFS_NODE(vp); 1607 1608 /* Assign default values if they are unknown. */ 1609 MPASS(uid != VNOVAL || gid != VNOVAL); 1610 if (uid == VNOVAL) 1611 uid = node->tn_uid; 1612 if (gid == VNOVAL) 1613 gid = node->tn_gid; 1614 MPASS(uid != VNOVAL && gid != VNOVAL); 1615 1616 /* Disallow this operation if the file system is mounted read-only. */ 1617 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1618 return EROFS; 1619 1620 /* Immutable or append-only files cannot be modified, either. */ 1621 if (node->tn_flags & (IMMUTABLE | APPEND)) 1622 return EPERM; 1623 1624 /* 1625 * To modify the ownership of a file, must possess VADMIN for that 1626 * file. 1627 */ 1628 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1629 return (error); 1630 1631 /* 1632 * To change the owner of a file, or change the group of a file to a 1633 * group of which we are not a member, the caller must have 1634 * privilege. 1635 */ 1636 if ((uid != node->tn_uid || 1637 (gid != node->tn_gid && !groupmember(gid, cred))) && 1638 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1639 return (error); 1640 1641 ogid = node->tn_gid; 1642 ouid = node->tn_uid; 1643 1644 node->tn_uid = uid; 1645 node->tn_gid = gid; 1646 1647 node->tn_status |= TMPFS_NODE_CHANGED; 1648 1649 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1650 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1651 node->tn_mode &= ~(S_ISUID | S_ISGID); 1652 } 1653 1654 ASSERT_VOP_ELOCKED(vp, "chown2"); 1655 1656 return (0); 1657} 1658 1659/* 1660 * Change size of the given vnode. 1661 * Caller should execute tmpfs_update on vp after a successful execution. 1662 * The vnode must be locked on entry and remain locked on exit. 1663 */ 1664int 1665tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1666 struct thread *p) 1667{ 1668 int error; 1669 struct tmpfs_node *node; 1670 1671 ASSERT_VOP_ELOCKED(vp, "chsize"); 1672 1673 node = VP_TO_TMPFS_NODE(vp); 1674 1675 /* Decide whether this is a valid operation based on the file type. */ 1676 error = 0; 1677 switch (vp->v_type) { 1678 case VDIR: 1679 return EISDIR; 1680 1681 case VREG: 1682 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1683 return EROFS; 1684 break; 1685 1686 case VBLK: 1687 /* FALLTHROUGH */ 1688 case VCHR: 1689 /* FALLTHROUGH */ 1690 case VFIFO: 1691 /* Allow modifications of special files even if in the file 1692 * system is mounted read-only (we are not modifying the 1693 * files themselves, but the objects they represent). */ 1694 return 0; 1695 1696 default: 1697 /* Anything else is unsupported. */ 1698 return EOPNOTSUPP; 1699 } 1700 1701 /* Immutable or append-only files cannot be modified, either. */ 1702 if (node->tn_flags & (IMMUTABLE | APPEND)) 1703 return EPERM; 1704 1705 error = tmpfs_truncate(vp, size); 1706 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1707 * for us, as will update tn_status; no need to do that here. */ 1708 1709 ASSERT_VOP_ELOCKED(vp, "chsize2"); 1710 1711 return (error); 1712} 1713 1714/* 1715 * Change access and modification times of the given vnode. 1716 * Caller should execute tmpfs_update on vp after a successful execution. 1717 * The vnode must be locked on entry and remain locked on exit. 1718 */ 1719int 1720tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 1721 struct ucred *cred, struct thread *l) 1722{ 1723 int error; 1724 struct tmpfs_node *node; 1725 1726 ASSERT_VOP_ELOCKED(vp, "chtimes"); 1727 1728 node = VP_TO_TMPFS_NODE(vp); 1729 1730 /* Disallow this operation if the file system is mounted read-only. */ 1731 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1732 return EROFS; 1733 1734 /* Immutable or append-only files cannot be modified, either. */ 1735 if (node->tn_flags & (IMMUTABLE | APPEND)) 1736 return EPERM; 1737 1738 error = vn_utimes_perm(vp, vap, cred, l); 1739 if (error != 0) 1740 return (error); 1741 1742 if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL) 1743 node->tn_status |= TMPFS_NODE_ACCESSED; 1744 1745 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL) 1746 node->tn_status |= TMPFS_NODE_MODIFIED; 1747 1748 if (vap->va_birthtime.tv_nsec != VNOVAL && 1749 vap->va_birthtime.tv_nsec != VNOVAL) 1750 node->tn_status |= TMPFS_NODE_MODIFIED; 1751 1752 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1753 1754 if (vap->va_birthtime.tv_nsec != VNOVAL && 1755 vap->va_birthtime.tv_nsec != VNOVAL) 1756 node->tn_birthtime = vap->va_birthtime; 1757 ASSERT_VOP_ELOCKED(vp, "chtimes2"); 1758 1759 return (0); 1760} 1761 1762void 1763tmpfs_set_status(struct tmpfs_node *node, int status) 1764{ 1765 1766 if ((node->tn_status & status) == status) 1767 return; 1768 TMPFS_NODE_LOCK(node); 1769 node->tn_status |= status; 1770 TMPFS_NODE_UNLOCK(node); 1771} 1772 1773/* Sync timestamps */ 1774void 1775tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1776 const struct timespec *mod) 1777{ 1778 struct tmpfs_node *node; 1779 struct timespec now; 1780 1781 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes"); 1782 node = VP_TO_TMPFS_NODE(vp); 1783 1784 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1785 TMPFS_NODE_CHANGED)) == 0) 1786 return; 1787 1788 vfs_timestamp(&now); 1789 TMPFS_NODE_LOCK(node); 1790 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1791 if (acc == NULL) 1792 acc = &now; 1793 node->tn_atime = *acc; 1794 } 1795 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1796 if (mod == NULL) 1797 mod = &now; 1798 node->tn_mtime = *mod; 1799 } 1800 if (node->tn_status & TMPFS_NODE_CHANGED) 1801 node->tn_ctime = now; 1802 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1803 TMPFS_NODE_CHANGED); 1804 TMPFS_NODE_UNLOCK(node); 1805 1806} 1807 1808void 1809tmpfs_update(struct vnode *vp) 1810{ 1811 1812 tmpfs_itimes(vp, NULL, NULL); 1813} 1814 1815int 1816tmpfs_truncate(struct vnode *vp, off_t length) 1817{ 1818 int error; 1819 struct tmpfs_node *node; 1820 1821 node = VP_TO_TMPFS_NODE(vp); 1822 1823 if (length < 0) { 1824 error = EINVAL; 1825 goto out; 1826 } 1827 1828 if (node->tn_size == length) { 1829 error = 0; 1830 goto out; 1831 } 1832 1833 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1834 return (EFBIG); 1835 1836 error = tmpfs_reg_resize(vp, length, FALSE); 1837 if (error == 0) 1838 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1839 1840out: 1841 tmpfs_update(vp); 1842 1843 return (error); 1844} 1845 1846static __inline int 1847tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1848{ 1849 if (a->td_hash > b->td_hash) 1850 return (1); 1851 else if (a->td_hash < b->td_hash) 1852 return (-1); 1853 return (0); 1854} 1855 1856RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1857