tmpfs_subr.c revision 313092
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 313092 2017-02-02 13:36:07Z 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 VNON: 336 /* Do not do anything. VNON is provided to let the 337 * allocation routine clean itself easily by avoiding 338 * duplicating code in it. */ 339 /* FALLTHROUGH */ 340 case VBLK: 341 /* FALLTHROUGH */ 342 case VCHR: 343 /* FALLTHROUGH */ 344 case VDIR: 345 /* FALLTHROUGH */ 346 case VFIFO: 347 /* FALLTHROUGH */ 348 case VSOCK: 349 break; 350 351 case VLNK: 352 free(node->tn_link, M_TMPFSNAME); 353 break; 354 355 case VREG: 356 uobj = node->tn_reg.tn_aobj; 357 if (uobj != NULL) { 358 atomic_subtract_long(&tmp->tm_pages_used, uobj->size); 359 KASSERT((uobj->flags & OBJ_TMPFS) == 0, 360 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj)); 361 vm_object_deallocate(uobj); 362 } 363 break; 364 365 default: 366 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 367 } 368 369 free_unr(tmp->tm_ino_unr, node->tn_id); 370 uma_zfree(tmp->tm_node_pool, node); 371 TMPFS_LOCK(tmp); 372 tmpfs_free_tmp(tmp); 373 return (true); 374} 375 376static __inline uint32_t 377tmpfs_dirent_hash(const char *name, u_int len) 378{ 379 uint32_t hash; 380 381 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK; 382#ifdef TMPFS_DEBUG_DIRCOOKIE_DUP 383 hash &= 0xf; 384#endif 385 if (hash < TMPFS_DIRCOOKIE_MIN) 386 hash += TMPFS_DIRCOOKIE_MIN; 387 388 return (hash); 389} 390 391static __inline off_t 392tmpfs_dirent_cookie(struct tmpfs_dirent *de) 393{ 394 if (de == NULL) 395 return (TMPFS_DIRCOOKIE_EOF); 396 397 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN); 398 399 return (de->td_cookie); 400} 401 402static __inline boolean_t 403tmpfs_dirent_dup(struct tmpfs_dirent *de) 404{ 405 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0); 406} 407 408static __inline boolean_t 409tmpfs_dirent_duphead(struct tmpfs_dirent *de) 410{ 411 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0); 412} 413 414void 415tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen) 416{ 417 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen); 418 memcpy(de->ud.td_name, name, namelen); 419 de->td_namelen = namelen; 420} 421 422/* 423 * Allocates a new directory entry for the node node with a name of name. 424 * The new directory entry is returned in *de. 425 * 426 * The link count of node is increased by one to reflect the new object 427 * referencing it. 428 * 429 * Returns zero on success or an appropriate error code on failure. 430 */ 431int 432tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 433 const char *name, u_int len, struct tmpfs_dirent **de) 434{ 435 struct tmpfs_dirent *nde; 436 437 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK); 438 nde->td_node = node; 439 if (name != NULL) { 440 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 441 tmpfs_dirent_init(nde, name, len); 442 } else 443 nde->td_namelen = 0; 444 if (node != NULL) 445 node->tn_links++; 446 447 *de = nde; 448 449 return 0; 450} 451 452/* 453 * Frees a directory entry. It is the caller's responsibility to destroy 454 * the node referenced by it if needed. 455 * 456 * The link count of node is decreased by one to reflect the removal of an 457 * object that referenced it. This only happens if 'node_exists' is true; 458 * otherwise the function will not access the node referred to by the 459 * directory entry, as it may already have been released from the outside. 460 */ 461void 462tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 463{ 464 struct tmpfs_node *node; 465 466 node = de->td_node; 467 if (node != NULL) { 468 MPASS(node->tn_links > 0); 469 node->tn_links--; 470 } 471 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL) 472 free(de->ud.td_name, M_TMPFSNAME); 473 uma_zfree(tmp->tm_dirent_pool, de); 474} 475 476void 477tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj) 478{ 479 480 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject"); 481 if (vp->v_type != VREG || obj == NULL) 482 return; 483 484 VM_OBJECT_WLOCK(obj); 485 VI_LOCK(vp); 486 vm_object_clear_flag(obj, OBJ_TMPFS); 487 obj->un_pager.swp.swp_tmpfs = NULL; 488 VI_UNLOCK(vp); 489 VM_OBJECT_WUNLOCK(obj); 490} 491 492/* 493 * Need to clear v_object for insmntque failure. 494 */ 495static void 496tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg) 497{ 498 499 tmpfs_destroy_vobject(vp, vp->v_object); 500 vp->v_object = NULL; 501 vp->v_data = NULL; 502 vp->v_op = &dead_vnodeops; 503 vgone(vp); 504 vput(vp); 505} 506 507/* 508 * Allocates a new vnode for the node node or returns a new reference to 509 * an existing one if the node had already a vnode referencing it. The 510 * resulting locked vnode is returned in *vpp. 511 * 512 * Returns zero on success or an appropriate error code on failure. 513 */ 514int 515tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 516 struct vnode **vpp) 517{ 518 struct vnode *vp; 519 struct tmpfs_mount *tm; 520 vm_object_t object; 521 int error; 522 523 error = 0; 524 tm = VFS_TO_TMPFS(mp); 525 TMPFS_NODE_LOCK(node); 526 tmpfs_ref_node_locked(node); 527loop: 528 TMPFS_NODE_ASSERT_LOCKED(node); 529 if ((vp = node->tn_vnode) != NULL) { 530 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 531 VI_LOCK(vp); 532 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) || 533 ((vp->v_iflag & VI_DOOMED) != 0 && 534 (lkflag & LK_NOWAIT) != 0)) { 535 VI_UNLOCK(vp); 536 TMPFS_NODE_UNLOCK(node); 537 error = ENOENT; 538 vp = NULL; 539 goto out; 540 } 541 if ((vp->v_iflag & VI_DOOMED) != 0) { 542 VI_UNLOCK(vp); 543 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM; 544 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) { 545 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 546 0, "tmpfsE", 0); 547 } 548 goto loop; 549 } 550 TMPFS_NODE_UNLOCK(node); 551 error = vget(vp, lkflag | LK_INTERLOCK, curthread); 552 if (error == ENOENT) { 553 TMPFS_NODE_LOCK(node); 554 goto loop; 555 } 556 if (error != 0) { 557 vp = NULL; 558 goto out; 559 } 560 561 /* 562 * Make sure the vnode is still there after 563 * getting the interlock to avoid racing a free. 564 */ 565 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 566 vput(vp); 567 TMPFS_NODE_LOCK(node); 568 goto loop; 569 } 570 571 goto out; 572 } 573 574 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 575 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 576 TMPFS_NODE_UNLOCK(node); 577 error = ENOENT; 578 vp = NULL; 579 goto out; 580 } 581 582 /* 583 * otherwise lock the vp list while we call getnewvnode 584 * since that can block. 585 */ 586 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 587 node->tn_vpstate |= TMPFS_VNODE_WANT; 588 error = msleep((caddr_t) &node->tn_vpstate, 589 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0); 590 if (error != 0) 591 goto out; 592 goto loop; 593 } else 594 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 595 596 TMPFS_NODE_UNLOCK(node); 597 598 /* Get a new vnode and associate it with our node. */ 599 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); 600 if (error != 0) 601 goto unlock; 602 MPASS(vp != NULL); 603 604 /* lkflag is ignored, the lock is exclusive */ 605 (void) vn_lock(vp, lkflag | LK_RETRY); 606 607 vp->v_data = node; 608 vp->v_type = node->tn_type; 609 610 /* Type-specific initialization. */ 611 switch (node->tn_type) { 612 case VBLK: 613 /* FALLTHROUGH */ 614 case VCHR: 615 /* FALLTHROUGH */ 616 case VLNK: 617 /* FALLTHROUGH */ 618 case VSOCK: 619 break; 620 case VFIFO: 621 vp->v_op = &tmpfs_fifoop_entries; 622 break; 623 case VREG: 624 object = node->tn_reg.tn_aobj; 625 VM_OBJECT_WLOCK(object); 626 VI_LOCK(vp); 627 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 628 vp->v_object = object; 629 object->un_pager.swp.swp_tmpfs = vp; 630 vm_object_set_flag(object, OBJ_TMPFS); 631 VI_UNLOCK(vp); 632 VM_OBJECT_WUNLOCK(object); 633 break; 634 case VDIR: 635 MPASS(node->tn_dir.tn_parent != NULL); 636 if (node->tn_dir.tn_parent == node) 637 vp->v_vflag |= VV_ROOT; 638 break; 639 640 default: 641 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 642 } 643 if (vp->v_type != VFIFO) 644 VN_LOCK_ASHARE(vp); 645 646 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 647 if (error != 0) 648 vp = NULL; 649 650unlock: 651 TMPFS_NODE_LOCK(node); 652 653 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 654 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 655 node->tn_vnode = vp; 656 657 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 658 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 659 TMPFS_NODE_UNLOCK(node); 660 wakeup((caddr_t) &node->tn_vpstate); 661 } else 662 TMPFS_NODE_UNLOCK(node); 663 664out: 665 if (error == 0) { 666 *vpp = vp; 667 668#ifdef INVARIANTS 669 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 670 TMPFS_NODE_LOCK(node); 671 MPASS(*vpp == node->tn_vnode); 672 TMPFS_NODE_UNLOCK(node); 673#endif 674 } 675 tmpfs_free_node(tm, node); 676 677 return (error); 678} 679 680/* 681 * Destroys the association between the vnode vp and the node it 682 * references. 683 */ 684void 685tmpfs_free_vp(struct vnode *vp) 686{ 687 struct tmpfs_node *node; 688 689 node = VP_TO_TMPFS_NODE(vp); 690 691 TMPFS_NODE_ASSERT_LOCKED(node); 692 node->tn_vnode = NULL; 693 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 694 wakeup(&node->tn_vnode); 695 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 696 vp->v_data = NULL; 697} 698 699/* 700 * Allocates a new file of type 'type' and adds it to the parent directory 701 * 'dvp'; this addition is done using the component name given in 'cnp'. 702 * The ownership of the new file is automatically assigned based on the 703 * credentials of the caller (through 'cnp'), the group is set based on 704 * the parent directory and the mode is determined from the 'vap' argument. 705 * If successful, *vpp holds a vnode to the newly created file and zero 706 * is returned. Otherwise *vpp is NULL and the function returns an 707 * appropriate error code. 708 */ 709int 710tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 711 struct componentname *cnp, char *target) 712{ 713 int error; 714 struct tmpfs_dirent *de; 715 struct tmpfs_mount *tmp; 716 struct tmpfs_node *dnode; 717 struct tmpfs_node *node; 718 struct tmpfs_node *parent; 719 720 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file"); 721 MPASS(cnp->cn_flags & HASBUF); 722 723 tmp = VFS_TO_TMPFS(dvp->v_mount); 724 dnode = VP_TO_TMPFS_DIR(dvp); 725 *vpp = NULL; 726 727 /* If the entry we are creating is a directory, we cannot overflow 728 * the number of links of its parent, because it will get a new 729 * link. */ 730 if (vap->va_type == VDIR) { 731 /* Ensure that we do not overflow the maximum number of links 732 * imposed by the system. */ 733 MPASS(dnode->tn_links <= LINK_MAX); 734 if (dnode->tn_links == LINK_MAX) { 735 return (EMLINK); 736 } 737 738 parent = dnode; 739 MPASS(parent != NULL); 740 } else 741 parent = NULL; 742 743 /* Allocate a node that represents the new file. */ 744 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type, 745 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent, 746 target, vap->va_rdev, &node); 747 if (error != 0) 748 return (error); 749 750 /* Allocate a directory entry that points to the new file. */ 751 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 752 &de); 753 if (error != 0) { 754 tmpfs_free_node(tmp, node); 755 return (error); 756 } 757 758 /* Allocate a vnode for the new file. */ 759 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 760 if (error != 0) { 761 tmpfs_free_dirent(tmp, de); 762 tmpfs_free_node(tmp, node); 763 return (error); 764 } 765 766 /* Now that all required items are allocated, we can proceed to 767 * insert the new node into the directory, an operation that 768 * cannot fail. */ 769 if (cnp->cn_flags & ISWHITEOUT) 770 tmpfs_dir_whiteout_remove(dvp, cnp); 771 tmpfs_dir_attach(dvp, de); 772 return (0); 773} 774 775struct tmpfs_dirent * 776tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 777{ 778 struct tmpfs_dirent *de; 779 780 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 781 dc->tdc_tree = de; 782 if (de != NULL && tmpfs_dirent_duphead(de)) 783 de = LIST_FIRST(&de->ud.td_duphead); 784 dc->tdc_current = de; 785 786 return (dc->tdc_current); 787} 788 789struct tmpfs_dirent * 790tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 791{ 792 struct tmpfs_dirent *de; 793 794 MPASS(dc->tdc_tree != NULL); 795 if (tmpfs_dirent_dup(dc->tdc_current)) { 796 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 797 if (dc->tdc_current != NULL) 798 return (dc->tdc_current); 799 } 800 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 801 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 802 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 803 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 804 MPASS(dc->tdc_current != NULL); 805 } 806 807 return (dc->tdc_current); 808} 809 810/* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 811static struct tmpfs_dirent * 812tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 813{ 814 struct tmpfs_dirent *de, dekey; 815 816 dekey.td_hash = hash; 817 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 818 return (de); 819} 820 821/* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 822static struct tmpfs_dirent * 823tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 824 struct tmpfs_dir_cursor *dc) 825{ 826 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 827 struct tmpfs_dirent *de, dekey; 828 829 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 830 831 if (cookie == node->tn_dir.tn_readdir_lastn && 832 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 833 /* Protect against possible race, tn_readdir_last[pn] 834 * may be updated with only shared vnode lock held. */ 835 if (cookie == tmpfs_dirent_cookie(de)) 836 goto out; 837 } 838 839 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 840 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 841 uh.td_dup.index_entries) { 842 MPASS(tmpfs_dirent_dup(de)); 843 if (de->td_cookie == cookie) 844 goto out; 845 /* dupindex list is sorted. */ 846 if (de->td_cookie < cookie) { 847 de = NULL; 848 goto out; 849 } 850 } 851 MPASS(de == NULL); 852 goto out; 853 } 854 855 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) { 856 de = NULL; 857 } else { 858 dekey.td_hash = cookie; 859 /* Recover if direntry for cookie was removed */ 860 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 861 } 862 dc->tdc_tree = de; 863 dc->tdc_current = de; 864 if (de != NULL && tmpfs_dirent_duphead(de)) { 865 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 866 MPASS(dc->tdc_current != NULL); 867 } 868 return (dc->tdc_current); 869 870out: 871 dc->tdc_tree = de; 872 dc->tdc_current = de; 873 if (de != NULL && tmpfs_dirent_dup(de)) 874 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 875 de->td_hash); 876 return (dc->tdc_current); 877} 878 879/* 880 * Looks for a directory entry in the directory represented by node. 881 * 'cnp' describes the name of the entry to look for. Note that the . 882 * and .. components are not allowed as they do not physically exist 883 * within directories. 884 * 885 * Returns a pointer to the entry when found, otherwise NULL. 886 */ 887struct tmpfs_dirent * 888tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 889 struct componentname *cnp) 890{ 891 struct tmpfs_dir_duphead *duphead; 892 struct tmpfs_dirent *de; 893 uint32_t hash; 894 895 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 896 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 897 cnp->cn_nameptr[1] == '.'))); 898 TMPFS_VALIDATE_DIR(node); 899 900 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 901 de = tmpfs_dir_xlookup_hash(node, hash); 902 if (de != NULL && tmpfs_dirent_duphead(de)) { 903 duphead = &de->ud.td_duphead; 904 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 905 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 906 cnp->cn_namelen)) 907 break; 908 } 909 } else if (de != NULL) { 910 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 911 cnp->cn_namelen)) 912 de = NULL; 913 } 914 if (de != NULL && f != NULL && de->td_node != f) 915 de = NULL; 916 917 return (de); 918} 919 920/* 921 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 922 * list, allocate new cookie value. 923 */ 924static void 925tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 926 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 927{ 928 struct tmpfs_dir_duphead *dupindex; 929 struct tmpfs_dirent *de, *pde; 930 931 dupindex = &dnode->tn_dir.tn_dupindex; 932 de = LIST_FIRST(dupindex); 933 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 934 if (de == NULL) 935 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 936 else 937 nde->td_cookie = de->td_cookie + 1; 938 MPASS(tmpfs_dirent_dup(nde)); 939 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 940 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 941 return; 942 } 943 944 /* 945 * Cookie numbers are near exhaustion. Scan dupindex list for unused 946 * numbers. dupindex list is sorted in descending order. Keep it so 947 * after inserting nde. 948 */ 949 while (1) { 950 pde = de; 951 de = LIST_NEXT(de, uh.td_dup.index_entries); 952 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 953 /* 954 * Last element of the index doesn't have minimal cookie 955 * value, use it. 956 */ 957 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 958 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 959 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 960 return; 961 } else if (de == NULL) { 962 /* 963 * We are so lucky have 2^30 hash duplicates in single 964 * directory :) Return largest possible cookie value. 965 * It should be fine except possible issues with 966 * VOP_READDIR restart. 967 */ 968 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 969 LIST_INSERT_HEAD(dupindex, nde, 970 uh.td_dup.index_entries); 971 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 972 return; 973 } 974 if (de->td_cookie + 1 == pde->td_cookie || 975 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 976 continue; /* No hole or invalid cookie. */ 977 nde->td_cookie = de->td_cookie + 1; 978 MPASS(tmpfs_dirent_dup(nde)); 979 MPASS(pde->td_cookie > nde->td_cookie); 980 MPASS(nde->td_cookie > de->td_cookie); 981 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 982 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 983 return; 984 }; 985} 986 987/* 988 * Attaches the directory entry de to the directory represented by vp. 989 * Note that this does not change the link count of the node pointed by 990 * the directory entry, as this is done by tmpfs_alloc_dirent. 991 */ 992void 993tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 994{ 995 struct tmpfs_node *dnode; 996 struct tmpfs_dirent *xde, *nde; 997 998 ASSERT_VOP_ELOCKED(vp, __func__); 999 MPASS(de->td_namelen > 0); 1000 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 1001 MPASS(de->td_cookie == de->td_hash); 1002 1003 dnode = VP_TO_TMPFS_DIR(vp); 1004 dnode->tn_dir.tn_readdir_lastn = 0; 1005 dnode->tn_dir.tn_readdir_lastp = NULL; 1006 1007 MPASS(!tmpfs_dirent_dup(de)); 1008 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1009 if (xde != NULL && tmpfs_dirent_duphead(xde)) 1010 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1011 else if (xde != NULL) { 1012 /* 1013 * Allocate new duphead. Swap xde with duphead to avoid 1014 * adding/removing elements with the same hash. 1015 */ 1016 MPASS(!tmpfs_dirent_dup(xde)); 1017 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 1018 &nde); 1019 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 1020 memcpy(nde, xde, sizeof(*xde)); 1021 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 1022 LIST_INIT(&xde->ud.td_duphead); 1023 xde->td_namelen = 0; 1024 xde->td_node = NULL; 1025 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 1026 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 1027 } 1028 dnode->tn_size += sizeof(struct tmpfs_dirent); 1029 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1030 TMPFS_NODE_MODIFIED; 1031 tmpfs_update(vp); 1032} 1033 1034/* 1035 * Detaches the directory entry de from the directory represented by vp. 1036 * Note that this does not change the link count of the node pointed by 1037 * the directory entry, as this is done by tmpfs_free_dirent. 1038 */ 1039void 1040tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 1041{ 1042 struct tmpfs_mount *tmp; 1043 struct tmpfs_dir *head; 1044 struct tmpfs_node *dnode; 1045 struct tmpfs_dirent *xde; 1046 1047 ASSERT_VOP_ELOCKED(vp, __func__); 1048 1049 dnode = VP_TO_TMPFS_DIR(vp); 1050 head = &dnode->tn_dir.tn_dirhead; 1051 dnode->tn_dir.tn_readdir_lastn = 0; 1052 dnode->tn_dir.tn_readdir_lastp = NULL; 1053 1054 if (tmpfs_dirent_dup(de)) { 1055 /* Remove duphead if de was last entry. */ 1056 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1057 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1058 MPASS(tmpfs_dirent_duphead(xde)); 1059 } else 1060 xde = NULL; 1061 LIST_REMOVE(de, uh.td_dup.entries); 1062 LIST_REMOVE(de, uh.td_dup.index_entries); 1063 if (xde != NULL) { 1064 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1065 RB_REMOVE(tmpfs_dir, head, xde); 1066 tmp = VFS_TO_TMPFS(vp->v_mount); 1067 MPASS(xde->td_node == NULL); 1068 tmpfs_free_dirent(tmp, xde); 1069 } 1070 } 1071 de->td_cookie = de->td_hash; 1072 } else 1073 RB_REMOVE(tmpfs_dir, head, de); 1074 1075 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1076 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1077 TMPFS_NODE_MODIFIED; 1078 tmpfs_update(vp); 1079} 1080 1081void 1082tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1083{ 1084 struct tmpfs_dirent *de, *dde, *nde; 1085 1086 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1087 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1088 /* Node may already be destroyed. */ 1089 de->td_node = NULL; 1090 if (tmpfs_dirent_duphead(de)) { 1091 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1092 LIST_REMOVE(dde, uh.td_dup.entries); 1093 dde->td_node = NULL; 1094 tmpfs_free_dirent(tmp, dde); 1095 } 1096 } 1097 tmpfs_free_dirent(tmp, de); 1098 } 1099} 1100 1101/* 1102 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1103 * directory and returns it in the uio space. The function returns 0 1104 * on success, -1 if there was not enough space in the uio structure to 1105 * hold the directory entry or an appropriate error code if another 1106 * error happens. 1107 */ 1108static int 1109tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1110{ 1111 int error; 1112 struct dirent dent; 1113 1114 TMPFS_VALIDATE_DIR(node); 1115 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1116 1117 dent.d_fileno = node->tn_id; 1118 dent.d_type = DT_DIR; 1119 dent.d_namlen = 1; 1120 dent.d_name[0] = '.'; 1121 dent.d_name[1] = '\0'; 1122 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1123 1124 if (dent.d_reclen > uio->uio_resid) 1125 error = EJUSTRETURN; 1126 else 1127 error = uiomove(&dent, dent.d_reclen, uio); 1128 1129 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1130 1131 return (error); 1132} 1133 1134/* 1135 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1136 * directory and returns it in the uio space. The function returns 0 1137 * on success, -1 if there was not enough space in the uio structure to 1138 * hold the directory entry or an appropriate error code if another 1139 * error happens. 1140 */ 1141static int 1142tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1143{ 1144 int error; 1145 struct dirent dent; 1146 1147 TMPFS_VALIDATE_DIR(node); 1148 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1149 1150 /* 1151 * Return ENOENT if the current node is already removed. 1152 */ 1153 TMPFS_ASSERT_LOCKED(node); 1154 if (node->tn_dir.tn_parent == NULL) 1155 return (ENOENT); 1156 1157 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1158 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1159 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1160 1161 dent.d_type = DT_DIR; 1162 dent.d_namlen = 2; 1163 dent.d_name[0] = '.'; 1164 dent.d_name[1] = '.'; 1165 dent.d_name[2] = '\0'; 1166 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1167 1168 if (dent.d_reclen > uio->uio_resid) 1169 error = EJUSTRETURN; 1170 else 1171 error = uiomove(&dent, dent.d_reclen, uio); 1172 1173 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1174 1175 return (error); 1176} 1177 1178/* 1179 * Helper function for tmpfs_readdir. Returns as much directory entries 1180 * as can fit in the uio space. The read starts at uio->uio_offset. 1181 * The function returns 0 on success, -1 if there was not enough space 1182 * in the uio structure to hold the directory entry or an appropriate 1183 * error code if another error happens. 1184 */ 1185int 1186tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1187 u_long *cookies, int *ncookies) 1188{ 1189 struct tmpfs_dir_cursor dc; 1190 struct tmpfs_dirent *de; 1191 off_t off; 1192 int error; 1193 1194 TMPFS_VALIDATE_DIR(node); 1195 1196 off = 0; 1197 1198 /* 1199 * Lookup the node from the current offset. The starting offset of 1200 * 0 will lookup both '.' and '..', and then the first real entry, 1201 * or EOF if there are none. Then find all entries for the dir that 1202 * fit into the buffer. Once no more entries are found (de == NULL), 1203 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1204 * call to return 0. 1205 */ 1206 switch (uio->uio_offset) { 1207 case TMPFS_DIRCOOKIE_DOT: 1208 error = tmpfs_dir_getdotdent(node, uio); 1209 if (error != 0) 1210 return (error); 1211 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1212 if (cookies != NULL) 1213 cookies[(*ncookies)++] = off = uio->uio_offset; 1214 /* FALLTHROUGH */ 1215 case TMPFS_DIRCOOKIE_DOTDOT: 1216 error = tmpfs_dir_getdotdotdent(node, uio); 1217 if (error != 0) 1218 return (error); 1219 de = tmpfs_dir_first(node, &dc); 1220 uio->uio_offset = tmpfs_dirent_cookie(de); 1221 if (cookies != NULL) 1222 cookies[(*ncookies)++] = off = uio->uio_offset; 1223 /* EOF. */ 1224 if (de == NULL) 1225 return (0); 1226 break; 1227 case TMPFS_DIRCOOKIE_EOF: 1228 return (0); 1229 default: 1230 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1231 if (de == NULL) 1232 return (EINVAL); 1233 if (cookies != NULL) 1234 off = tmpfs_dirent_cookie(de); 1235 } 1236 1237 /* Read as much entries as possible; i.e., until we reach the end of 1238 * the directory or we exhaust uio space. */ 1239 do { 1240 struct dirent d; 1241 1242 /* Create a dirent structure representing the current 1243 * tmpfs_node and fill it. */ 1244 if (de->td_node == NULL) { 1245 d.d_fileno = 1; 1246 d.d_type = DT_WHT; 1247 } else { 1248 d.d_fileno = de->td_node->tn_id; 1249 switch (de->td_node->tn_type) { 1250 case VBLK: 1251 d.d_type = DT_BLK; 1252 break; 1253 1254 case VCHR: 1255 d.d_type = DT_CHR; 1256 break; 1257 1258 case VDIR: 1259 d.d_type = DT_DIR; 1260 break; 1261 1262 case VFIFO: 1263 d.d_type = DT_FIFO; 1264 break; 1265 1266 case VLNK: 1267 d.d_type = DT_LNK; 1268 break; 1269 1270 case VREG: 1271 d.d_type = DT_REG; 1272 break; 1273 1274 case VSOCK: 1275 d.d_type = DT_SOCK; 1276 break; 1277 1278 default: 1279 panic("tmpfs_dir_getdents: type %p %d", 1280 de->td_node, (int)de->td_node->tn_type); 1281 } 1282 } 1283 d.d_namlen = de->td_namelen; 1284 MPASS(de->td_namelen < sizeof(d.d_name)); 1285 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1286 d.d_name[de->td_namelen] = '\0'; 1287 d.d_reclen = GENERIC_DIRSIZ(&d); 1288 1289 /* Stop reading if the directory entry we are treating is 1290 * bigger than the amount of data that can be returned. */ 1291 if (d.d_reclen > uio->uio_resid) { 1292 error = EJUSTRETURN; 1293 break; 1294 } 1295 1296 /* Copy the new dirent structure into the output buffer and 1297 * advance pointers. */ 1298 error = uiomove(&d, d.d_reclen, uio); 1299 if (error == 0) { 1300 de = tmpfs_dir_next(node, &dc); 1301 if (cookies != NULL) { 1302 off = tmpfs_dirent_cookie(de); 1303 MPASS(*ncookies < maxcookies); 1304 cookies[(*ncookies)++] = off; 1305 } 1306 } 1307 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1308 1309 /* Skip setting off when using cookies as it is already done above. */ 1310 if (cookies == NULL) 1311 off = tmpfs_dirent_cookie(de); 1312 1313 /* Update the offset and cache. */ 1314 uio->uio_offset = off; 1315 node->tn_dir.tn_readdir_lastn = off; 1316 node->tn_dir.tn_readdir_lastp = de; 1317 1318 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1319 return error; 1320} 1321 1322int 1323tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1324{ 1325 struct tmpfs_dirent *de; 1326 int error; 1327 1328 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1329 cnp->cn_nameptr, cnp->cn_namelen, &de); 1330 if (error != 0) 1331 return (error); 1332 tmpfs_dir_attach(dvp, de); 1333 return (0); 1334} 1335 1336void 1337tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1338{ 1339 struct tmpfs_dirent *de; 1340 1341 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1342 MPASS(de != NULL && de->td_node == NULL); 1343 tmpfs_dir_detach(dvp, de); 1344 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1345} 1346 1347/* 1348 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1349 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1350 * 'newsize' must be positive. 1351 * 1352 * Returns zero on success or an appropriate error code on failure. 1353 */ 1354int 1355tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1356{ 1357 struct tmpfs_mount *tmp; 1358 struct tmpfs_node *node; 1359 vm_object_t uobj; 1360 vm_page_t m, ma[1]; 1361 vm_pindex_t idx, newpages, oldpages; 1362 off_t oldsize; 1363 int base, rv; 1364 1365 MPASS(vp->v_type == VREG); 1366 MPASS(newsize >= 0); 1367 1368 node = VP_TO_TMPFS_NODE(vp); 1369 uobj = node->tn_reg.tn_aobj; 1370 tmp = VFS_TO_TMPFS(vp->v_mount); 1371 1372 /* 1373 * Convert the old and new sizes to the number of pages needed to 1374 * store them. It may happen that we do not need to do anything 1375 * because the last allocated page can accommodate the change on 1376 * its own. 1377 */ 1378 oldsize = node->tn_size; 1379 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1380 MPASS(oldpages == uobj->size); 1381 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1382 if (newpages > oldpages && 1383 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1384 return (ENOSPC); 1385 1386 VM_OBJECT_WLOCK(uobj); 1387 if (newsize < oldsize) { 1388 /* 1389 * Zero the truncated part of the last page. 1390 */ 1391 base = newsize & PAGE_MASK; 1392 if (base != 0) { 1393 idx = OFF_TO_IDX(newsize); 1394retry: 1395 m = vm_page_lookup(uobj, idx); 1396 if (m != NULL) { 1397 if (vm_page_sleep_if_busy(m, "tmfssz")) 1398 goto retry; 1399 MPASS(m->valid == VM_PAGE_BITS_ALL); 1400 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1401 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1402 if (m == NULL) { 1403 VM_OBJECT_WUNLOCK(uobj); 1404 VM_WAIT; 1405 VM_OBJECT_WLOCK(uobj); 1406 goto retry; 1407 } else if (m->valid != VM_PAGE_BITS_ALL) { 1408 ma[0] = m; 1409 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1410 m = vm_page_lookup(uobj, idx); 1411 } else 1412 /* A cached page was reactivated. */ 1413 rv = VM_PAGER_OK; 1414 vm_page_lock(m); 1415 if (rv == VM_PAGER_OK) { 1416 vm_page_deactivate(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 1455void 1456tmpfs_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 */ 1485int 1486tmpfs_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 */ 1542int 1543tmpfs_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 */ 1600int 1601tmpfs_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 */ 1669int 1670tmpfs_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 */ 1724int 1725tmpfs_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 && vap->va_atime.tv_nsec != VNOVAL) 1748 node->tn_status |= TMPFS_NODE_ACCESSED; 1749 1750 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL) 1751 node->tn_status |= TMPFS_NODE_MODIFIED; 1752 1753 if (vap->va_birthtime.tv_nsec != VNOVAL && 1754 vap->va_birthtime.tv_nsec != VNOVAL) 1755 node->tn_status |= TMPFS_NODE_MODIFIED; 1756 1757 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1758 1759 if (vap->va_birthtime.tv_nsec != VNOVAL && 1760 vap->va_birthtime.tv_nsec != VNOVAL) 1761 node->tn_birthtime = vap->va_birthtime; 1762 ASSERT_VOP_ELOCKED(vp, "chtimes2"); 1763 1764 return (0); 1765} 1766 1767void 1768tmpfs_set_status(struct tmpfs_node *node, int status) 1769{ 1770 1771 if ((node->tn_status & status) == status) 1772 return; 1773 TMPFS_NODE_LOCK(node); 1774 node->tn_status |= status; 1775 TMPFS_NODE_UNLOCK(node); 1776} 1777 1778/* Sync timestamps */ 1779void 1780tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1781 const struct timespec *mod) 1782{ 1783 struct tmpfs_node *node; 1784 struct timespec now; 1785 1786 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes"); 1787 node = VP_TO_TMPFS_NODE(vp); 1788 1789 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1790 TMPFS_NODE_CHANGED)) == 0) 1791 return; 1792 1793 vfs_timestamp(&now); 1794 TMPFS_NODE_LOCK(node); 1795 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1796 if (acc == NULL) 1797 acc = &now; 1798 node->tn_atime = *acc; 1799 } 1800 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1801 if (mod == NULL) 1802 mod = &now; 1803 node->tn_mtime = *mod; 1804 } 1805 if (node->tn_status & TMPFS_NODE_CHANGED) 1806 node->tn_ctime = now; 1807 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1808 TMPFS_NODE_CHANGED); 1809 TMPFS_NODE_UNLOCK(node); 1810 1811} 1812 1813void 1814tmpfs_update(struct vnode *vp) 1815{ 1816 1817 tmpfs_itimes(vp, NULL, NULL); 1818} 1819 1820int 1821tmpfs_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 1845out: 1846 tmpfs_update(vp); 1847 1848 return (error); 1849} 1850 1851static __inline int 1852tmpfs_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 1861RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1862