tmpfs_subr.c revision 269170
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 269170 2014-07-28 01:08:43Z 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 64struct tmpfs_dir_cursor { 65 struct tmpfs_dirent *tdc_current; 66 struct tmpfs_dirent *tdc_tree; 67}; 68 69SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system"); 70 71static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED; 72 73static int 74sysctl_mem_reserved(SYSCTL_HANDLER_ARGS) 75{ 76 int error; 77 long pages, bytes; 78 79 pages = *(long *)arg1; 80 bytes = pages * PAGE_SIZE; 81 82 error = sysctl_handle_long(oidp, &bytes, 0, req); 83 if (error || !req->newptr) 84 return (error); 85 86 pages = bytes / PAGE_SIZE; 87 if (pages < TMPFS_PAGES_MINRESERVED) 88 return (EINVAL); 89 90 *(long *)arg1 = pages; 91 return (0); 92} 93 94SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW, 95 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L", 96 "Amount of available memory and swap below which tmpfs growth stops"); 97 98static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a, 99 struct tmpfs_dirent *b); 100RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 101 102size_t 103tmpfs_mem_avail(void) 104{ 105 vm_ooffset_t avail; 106 107 avail = swap_pager_avail + cnt.v_free_count + cnt.v_cache_count - 108 tmpfs_pages_reserved; 109 if (__predict_false(avail < 0)) 110 avail = 0; 111 return (avail); 112} 113 114size_t 115tmpfs_pages_used(struct tmpfs_mount *tmp) 116{ 117 const size_t node_size = sizeof(struct tmpfs_node) + 118 sizeof(struct tmpfs_dirent); 119 size_t meta_pages; 120 121 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size, 122 PAGE_SIZE); 123 return (meta_pages + tmp->tm_pages_used); 124} 125 126static size_t 127tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages) 128{ 129 if (tmpfs_mem_avail() < req_pages) 130 return (0); 131 132 if (tmp->tm_pages_max != SIZE_MAX && 133 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp)) 134 return (0); 135 136 return (1); 137} 138 139/* 140 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 141 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 142 * using the credentials of the process 'p'. 143 * 144 * If the node type is set to 'VDIR', then the parent parameter must point 145 * to the parent directory of the node being created. It may only be NULL 146 * while allocating the root node. 147 * 148 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 149 * specifies the device the node represents. 150 * 151 * If the node type is set to 'VLNK', then the parameter target specifies 152 * the file name of the target file for the symbolic link that is being 153 * created. 154 * 155 * Note that new nodes are retrieved from the available list if it has 156 * items or, if it is empty, from the node pool as long as there is enough 157 * space to create them. 158 * 159 * Returns zero on success or an appropriate error code on failure. 160 */ 161int 162tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 163 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 164 char *target, dev_t rdev, struct tmpfs_node **node) 165{ 166 struct tmpfs_node *nnode; 167 vm_object_t obj; 168 169 /* If the root directory of the 'tmp' file system is not yet 170 * allocated, this must be the request to do it. */ 171 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 172 173 MPASS(IFF(type == VLNK, target != NULL)); 174 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL)); 175 176 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 177 return (ENOSPC); 178 if (tmpfs_pages_check_avail(tmp, 1) == 0) 179 return (ENOSPC); 180 181 nnode = (struct tmpfs_node *)uma_zalloc_arg( 182 tmp->tm_node_pool, tmp, M_WAITOK); 183 184 /* Generic initialization. */ 185 nnode->tn_type = type; 186 vfs_timestamp(&nnode->tn_atime); 187 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 188 nnode->tn_atime; 189 nnode->tn_uid = uid; 190 nnode->tn_gid = gid; 191 nnode->tn_mode = mode; 192 nnode->tn_id = alloc_unr(tmp->tm_ino_unr); 193 194 /* Type-specific initialization. */ 195 switch (nnode->tn_type) { 196 case VBLK: 197 case VCHR: 198 nnode->tn_rdev = rdev; 199 break; 200 201 case VDIR: 202 RB_INIT(&nnode->tn_dir.tn_dirhead); 203 LIST_INIT(&nnode->tn_dir.tn_dupindex); 204 MPASS(parent != nnode); 205 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL)); 206 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent; 207 nnode->tn_dir.tn_readdir_lastn = 0; 208 nnode->tn_dir.tn_readdir_lastp = NULL; 209 nnode->tn_links++; 210 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent); 211 nnode->tn_dir.tn_parent->tn_links++; 212 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent); 213 break; 214 215 case VFIFO: 216 /* FALLTHROUGH */ 217 case VSOCK: 218 break; 219 220 case VLNK: 221 MPASS(strlen(target) < MAXPATHLEN); 222 nnode->tn_size = strlen(target); 223 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME, 224 M_WAITOK); 225 memcpy(nnode->tn_link, target, nnode->tn_size); 226 break; 227 228 case VREG: 229 obj = nnode->tn_reg.tn_aobj = 230 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0, 231 NULL /* XXXKIB - tmpfs needs swap reservation */); 232 VM_OBJECT_WLOCK(obj); 233 /* OBJ_TMPFS is set together with the setting of vp->v_object */ 234 vm_object_set_flag(obj, OBJ_NOSPLIT); 235 vm_object_clear_flag(obj, OBJ_ONEMAPPING); 236 VM_OBJECT_WUNLOCK(obj); 237 break; 238 239 default: 240 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 241 } 242 243 TMPFS_LOCK(tmp); 244 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 245 tmp->tm_nodes_inuse++; 246 TMPFS_UNLOCK(tmp); 247 248 *node = nnode; 249 return 0; 250} 251 252/* 253 * Destroys the node pointed to by node from the file system 'tmp'. 254 * If the node does not belong to the given mount point, the results are 255 * unpredicted. 256 * 257 * If the node references a directory; no entries are allowed because 258 * their removal could need a recursive algorithm, something forbidden in 259 * kernel space. Furthermore, there is not need to provide such 260 * functionality (recursive removal) because the only primitives offered 261 * to the user are the removal of empty directories and the deletion of 262 * individual files. 263 * 264 * Note that nodes are not really deleted; in fact, when a node has been 265 * allocated, it cannot be deleted during the whole life of the file 266 * system. Instead, they are moved to the available list and remain there 267 * until reused. 268 */ 269void 270tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 271{ 272 vm_object_t uobj; 273 274#ifdef INVARIANTS 275 TMPFS_NODE_LOCK(node); 276 MPASS(node->tn_vnode == NULL); 277 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 278 TMPFS_NODE_UNLOCK(node); 279#endif 280 281 TMPFS_LOCK(tmp); 282 LIST_REMOVE(node, tn_entries); 283 tmp->tm_nodes_inuse--; 284 TMPFS_UNLOCK(tmp); 285 286 switch (node->tn_type) { 287 case VNON: 288 /* Do not do anything. VNON is provided to let the 289 * allocation routine clean itself easily by avoiding 290 * duplicating code in it. */ 291 /* FALLTHROUGH */ 292 case VBLK: 293 /* FALLTHROUGH */ 294 case VCHR: 295 /* FALLTHROUGH */ 296 case VDIR: 297 /* FALLTHROUGH */ 298 case VFIFO: 299 /* FALLTHROUGH */ 300 case VSOCK: 301 break; 302 303 case VLNK: 304 free(node->tn_link, M_TMPFSNAME); 305 break; 306 307 case VREG: 308 uobj = node->tn_reg.tn_aobj; 309 if (uobj != NULL) { 310 TMPFS_LOCK(tmp); 311 tmp->tm_pages_used -= uobj->size; 312 TMPFS_UNLOCK(tmp); 313 KASSERT((uobj->flags & OBJ_TMPFS) == 0, 314 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj)); 315 vm_object_deallocate(uobj); 316 } 317 break; 318 319 default: 320 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 321 } 322 323 free_unr(tmp->tm_ino_unr, node->tn_id); 324 uma_zfree(tmp->tm_node_pool, node); 325} 326 327static __inline uint32_t 328tmpfs_dirent_hash(const char *name, u_int len) 329{ 330 uint32_t hash; 331 332 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK; 333#ifdef TMPFS_DEBUG_DIRCOOKIE_DUP 334 hash &= 0xf; 335#endif 336 if (hash < TMPFS_DIRCOOKIE_MIN) 337 hash += TMPFS_DIRCOOKIE_MIN; 338 339 return (hash); 340} 341 342static __inline off_t 343tmpfs_dirent_cookie(struct tmpfs_dirent *de) 344{ 345 if (de == NULL) 346 return (TMPFS_DIRCOOKIE_EOF); 347 348 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN); 349 350 return (de->td_cookie); 351} 352 353static __inline boolean_t 354tmpfs_dirent_dup(struct tmpfs_dirent *de) 355{ 356 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0); 357} 358 359static __inline boolean_t 360tmpfs_dirent_duphead(struct tmpfs_dirent *de) 361{ 362 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0); 363} 364 365void 366tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen) 367{ 368 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen); 369 memcpy(de->ud.td_name, name, namelen); 370 de->td_namelen = namelen; 371} 372 373/* 374 * Allocates a new directory entry for the node node with a name of name. 375 * The new directory entry is returned in *de. 376 * 377 * The link count of node is increased by one to reflect the new object 378 * referencing it. 379 * 380 * Returns zero on success or an appropriate error code on failure. 381 */ 382int 383tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 384 const char *name, u_int len, struct tmpfs_dirent **de) 385{ 386 struct tmpfs_dirent *nde; 387 388 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK); 389 nde->td_node = node; 390 if (name != NULL) { 391 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 392 tmpfs_dirent_init(nde, name, len); 393 } else 394 nde->td_namelen = 0; 395 if (node != NULL) 396 node->tn_links++; 397 398 *de = nde; 399 400 return 0; 401} 402 403/* 404 * Frees a directory entry. It is the caller's responsibility to destroy 405 * the node referenced by it if needed. 406 * 407 * The link count of node is decreased by one to reflect the removal of an 408 * object that referenced it. This only happens if 'node_exists' is true; 409 * otherwise the function will not access the node referred to by the 410 * directory entry, as it may already have been released from the outside. 411 */ 412void 413tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 414{ 415 struct tmpfs_node *node; 416 417 node = de->td_node; 418 if (node != NULL) { 419 MPASS(node->tn_links > 0); 420 node->tn_links--; 421 } 422 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL) 423 free(de->ud.td_name, M_TMPFSNAME); 424 uma_zfree(tmp->tm_dirent_pool, de); 425} 426 427void 428tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj) 429{ 430 431 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject"); 432 if (vp->v_type != VREG || obj == NULL) 433 return; 434 435 VM_OBJECT_WLOCK(obj); 436 VI_LOCK(vp); 437 vm_object_clear_flag(obj, OBJ_TMPFS); 438 obj->un_pager.swp.swp_tmpfs = NULL; 439 VI_UNLOCK(vp); 440 VM_OBJECT_WUNLOCK(obj); 441} 442 443/* 444 * Need to clear v_object for insmntque failure. 445 */ 446static void 447tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg) 448{ 449 450 tmpfs_destroy_vobject(vp, vp->v_object); 451 vp->v_object = NULL; 452 vp->v_data = NULL; 453 vp->v_op = &dead_vnodeops; 454 vgone(vp); 455 vput(vp); 456} 457 458/* 459 * Allocates a new vnode for the node node or returns a new reference to 460 * an existing one if the node had already a vnode referencing it. The 461 * resulting locked vnode is returned in *vpp. 462 * 463 * Returns zero on success or an appropriate error code on failure. 464 */ 465int 466tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 467 struct vnode **vpp) 468{ 469 struct vnode *vp; 470 vm_object_t object; 471 int error; 472 473 error = 0; 474loop: 475 TMPFS_NODE_LOCK(node); 476loop1: 477 if ((vp = node->tn_vnode) != NULL) { 478 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 479 VI_LOCK(vp); 480 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) || 481 ((vp->v_iflag & VI_DOOMED) != 0 && 482 (lkflag & LK_NOWAIT) != 0)) { 483 VI_UNLOCK(vp); 484 TMPFS_NODE_UNLOCK(node); 485 error = ENOENT; 486 vp = NULL; 487 goto out; 488 } 489 if ((vp->v_iflag & VI_DOOMED) != 0) { 490 VI_UNLOCK(vp); 491 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM; 492 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) { 493 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 494 0, "tmpfsE", 0); 495 } 496 goto loop1; 497 } 498 TMPFS_NODE_UNLOCK(node); 499 error = vget(vp, lkflag | LK_INTERLOCK, curthread); 500 if (error == ENOENT) 501 goto loop; 502 if (error != 0) { 503 vp = NULL; 504 goto out; 505 } 506 507 /* 508 * Make sure the vnode is still there after 509 * getting the interlock to avoid racing a free. 510 */ 511 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 512 vput(vp); 513 goto loop; 514 } 515 516 goto out; 517 } 518 519 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 520 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 521 TMPFS_NODE_UNLOCK(node); 522 error = ENOENT; 523 vp = NULL; 524 goto out; 525 } 526 527 /* 528 * otherwise lock the vp list while we call getnewvnode 529 * since that can block. 530 */ 531 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 532 node->tn_vpstate |= TMPFS_VNODE_WANT; 533 error = msleep((caddr_t) &node->tn_vpstate, 534 TMPFS_NODE_MTX(node), PDROP | PCATCH, 535 "tmpfs_alloc_vp", 0); 536 if (error) 537 return error; 538 539 goto loop; 540 } else 541 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 542 543 TMPFS_NODE_UNLOCK(node); 544 545 /* Get a new vnode and associate it with our node. */ 546 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); 547 if (error != 0) 548 goto unlock; 549 MPASS(vp != NULL); 550 551 (void) vn_lock(vp, lkflag | LK_RETRY); 552 553 vp->v_data = node; 554 vp->v_type = node->tn_type; 555 556 /* Type-specific initialization. */ 557 switch (node->tn_type) { 558 case VBLK: 559 /* FALLTHROUGH */ 560 case VCHR: 561 /* FALLTHROUGH */ 562 case VLNK: 563 /* FALLTHROUGH */ 564 case VSOCK: 565 break; 566 case VFIFO: 567 vp->v_op = &tmpfs_fifoop_entries; 568 break; 569 case VREG: 570 object = node->tn_reg.tn_aobj; 571 VM_OBJECT_WLOCK(object); 572 VI_LOCK(vp); 573 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 574 vp->v_object = object; 575 object->un_pager.swp.swp_tmpfs = vp; 576 vm_object_set_flag(object, OBJ_TMPFS); 577 VI_UNLOCK(vp); 578 VM_OBJECT_WUNLOCK(object); 579 break; 580 case VDIR: 581 MPASS(node->tn_dir.tn_parent != NULL); 582 if (node->tn_dir.tn_parent == node) 583 vp->v_vflag |= VV_ROOT; 584 break; 585 586 default: 587 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 588 } 589 if (vp->v_type != VFIFO) 590 VN_LOCK_ASHARE(vp); 591 592 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 593 if (error) 594 vp = NULL; 595 596unlock: 597 TMPFS_NODE_LOCK(node); 598 599 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 600 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 601 node->tn_vnode = vp; 602 603 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 604 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 605 TMPFS_NODE_UNLOCK(node); 606 wakeup((caddr_t) &node->tn_vpstate); 607 } else 608 TMPFS_NODE_UNLOCK(node); 609 610out: 611 *vpp = vp; 612 613#ifdef INVARIANTS 614 if (error == 0) { 615 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 616 TMPFS_NODE_LOCK(node); 617 MPASS(*vpp == node->tn_vnode); 618 TMPFS_NODE_UNLOCK(node); 619 } 620#endif 621 622 return error; 623} 624 625/* 626 * Destroys the association between the vnode vp and the node it 627 * references. 628 */ 629void 630tmpfs_free_vp(struct vnode *vp) 631{ 632 struct tmpfs_node *node; 633 634 node = VP_TO_TMPFS_NODE(vp); 635 636 TMPFS_NODE_ASSERT_LOCKED(node); 637 node->tn_vnode = NULL; 638 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 639 wakeup(&node->tn_vnode); 640 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 641 vp->v_data = NULL; 642} 643 644/* 645 * Allocates a new file of type 'type' and adds it to the parent directory 646 * 'dvp'; this addition is done using the component name given in 'cnp'. 647 * The ownership of the new file is automatically assigned based on the 648 * credentials of the caller (through 'cnp'), the group is set based on 649 * the parent directory and the mode is determined from the 'vap' argument. 650 * If successful, *vpp holds a vnode to the newly created file and zero 651 * is returned. Otherwise *vpp is NULL and the function returns an 652 * appropriate error code. 653 */ 654int 655tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 656 struct componentname *cnp, char *target) 657{ 658 int error; 659 struct tmpfs_dirent *de; 660 struct tmpfs_mount *tmp; 661 struct tmpfs_node *dnode; 662 struct tmpfs_node *node; 663 struct tmpfs_node *parent; 664 665 MPASS(VOP_ISLOCKED(dvp)); 666 MPASS(cnp->cn_flags & HASBUF); 667 668 tmp = VFS_TO_TMPFS(dvp->v_mount); 669 dnode = VP_TO_TMPFS_DIR(dvp); 670 *vpp = NULL; 671 672 /* If the entry we are creating is a directory, we cannot overflow 673 * the number of links of its parent, because it will get a new 674 * link. */ 675 if (vap->va_type == VDIR) { 676 /* Ensure that we do not overflow the maximum number of links 677 * imposed by the system. */ 678 MPASS(dnode->tn_links <= LINK_MAX); 679 if (dnode->tn_links == LINK_MAX) { 680 return (EMLINK); 681 } 682 683 parent = dnode; 684 MPASS(parent != NULL); 685 } else 686 parent = NULL; 687 688 /* Allocate a node that represents the new file. */ 689 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid, 690 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node); 691 if (error != 0) 692 return (error); 693 694 /* Allocate a directory entry that points to the new file. */ 695 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 696 &de); 697 if (error != 0) { 698 tmpfs_free_node(tmp, node); 699 return (error); 700 } 701 702 /* Allocate a vnode for the new file. */ 703 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 704 if (error != 0) { 705 tmpfs_free_dirent(tmp, de); 706 tmpfs_free_node(tmp, node); 707 return (error); 708 } 709 710 /* Now that all required items are allocated, we can proceed to 711 * insert the new node into the directory, an operation that 712 * cannot fail. */ 713 if (cnp->cn_flags & ISWHITEOUT) 714 tmpfs_dir_whiteout_remove(dvp, cnp); 715 tmpfs_dir_attach(dvp, de); 716 return (0); 717} 718 719static struct tmpfs_dirent * 720tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 721{ 722 struct tmpfs_dirent *de; 723 724 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 725 dc->tdc_tree = de; 726 if (de != NULL && tmpfs_dirent_duphead(de)) 727 de = LIST_FIRST(&de->ud.td_duphead); 728 dc->tdc_current = de; 729 730 return (dc->tdc_current); 731} 732 733static struct tmpfs_dirent * 734tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 735{ 736 struct tmpfs_dirent *de; 737 738 MPASS(dc->tdc_tree != NULL); 739 if (tmpfs_dirent_dup(dc->tdc_current)) { 740 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 741 if (dc->tdc_current != NULL) 742 return (dc->tdc_current); 743 } 744 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 745 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 746 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 747 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 748 MPASS(dc->tdc_current != NULL); 749 } 750 751 return (dc->tdc_current); 752} 753 754/* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 755static struct tmpfs_dirent * 756tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 757{ 758 struct tmpfs_dirent *de, dekey; 759 760 dekey.td_hash = hash; 761 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 762 return (de); 763} 764 765/* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 766static struct tmpfs_dirent * 767tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 768 struct tmpfs_dir_cursor *dc) 769{ 770 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 771 struct tmpfs_dirent *de, dekey; 772 773 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 774 775 if (cookie == node->tn_dir.tn_readdir_lastn && 776 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 777 /* Protect against possible race, tn_readdir_last[pn] 778 * may be updated with only shared vnode lock held. */ 779 if (cookie == tmpfs_dirent_cookie(de)) 780 goto out; 781 } 782 783 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 784 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 785 uh.td_dup.index_entries) { 786 MPASS(tmpfs_dirent_dup(de)); 787 if (de->td_cookie == cookie) 788 goto out; 789 /* dupindex list is sorted. */ 790 if (de->td_cookie < cookie) { 791 de = NULL; 792 goto out; 793 } 794 } 795 MPASS(de == NULL); 796 goto out; 797 } 798 799 MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie); 800 dekey.td_hash = cookie; 801 /* Recover if direntry for cookie was removed */ 802 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 803 dc->tdc_tree = de; 804 dc->tdc_current = de; 805 if (de != NULL && tmpfs_dirent_duphead(de)) { 806 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 807 MPASS(dc->tdc_current != NULL); 808 } 809 return (dc->tdc_current); 810 811out: 812 dc->tdc_tree = de; 813 dc->tdc_current = de; 814 if (de != NULL && tmpfs_dirent_dup(de)) 815 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 816 de->td_hash); 817 return (dc->tdc_current); 818} 819 820/* 821 * Looks for a directory entry in the directory represented by node. 822 * 'cnp' describes the name of the entry to look for. Note that the . 823 * and .. components are not allowed as they do not physically exist 824 * within directories. 825 * 826 * Returns a pointer to the entry when found, otherwise NULL. 827 */ 828struct tmpfs_dirent * 829tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 830 struct componentname *cnp) 831{ 832 struct tmpfs_dir_duphead *duphead; 833 struct tmpfs_dirent *de; 834 uint32_t hash; 835 836 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 837 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 838 cnp->cn_nameptr[1] == '.'))); 839 TMPFS_VALIDATE_DIR(node); 840 841 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 842 de = tmpfs_dir_xlookup_hash(node, hash); 843 if (de != NULL && tmpfs_dirent_duphead(de)) { 844 duphead = &de->ud.td_duphead; 845 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 846 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 847 cnp->cn_namelen)) 848 break; 849 } 850 } else if (de != NULL) { 851 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 852 cnp->cn_namelen)) 853 de = NULL; 854 } 855 if (de != NULL && f != NULL && de->td_node != f) 856 de = NULL; 857 858 return (de); 859} 860 861/* 862 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 863 * list, allocate new cookie value. 864 */ 865static void 866tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 867 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 868{ 869 struct tmpfs_dir_duphead *dupindex; 870 struct tmpfs_dirent *de, *pde; 871 872 dupindex = &dnode->tn_dir.tn_dupindex; 873 de = LIST_FIRST(dupindex); 874 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 875 if (de == NULL) 876 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 877 else 878 nde->td_cookie = de->td_cookie + 1; 879 MPASS(tmpfs_dirent_dup(nde)); 880 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 881 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 882 return; 883 } 884 885 /* 886 * Cookie numbers are near exhaustion. Scan dupindex list for unused 887 * numbers. dupindex list is sorted in descending order. Keep it so 888 * after inserting nde. 889 */ 890 while (1) { 891 pde = de; 892 de = LIST_NEXT(de, uh.td_dup.index_entries); 893 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 894 /* 895 * Last element of the index doesn't have minimal cookie 896 * value, use it. 897 */ 898 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 899 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 900 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 901 return; 902 } else if (de == NULL) { 903 /* 904 * We are so lucky have 2^30 hash duplicates in single 905 * directory :) Return largest possible cookie value. 906 * It should be fine except possible issues with 907 * VOP_READDIR restart. 908 */ 909 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 910 LIST_INSERT_HEAD(dupindex, nde, 911 uh.td_dup.index_entries); 912 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 913 return; 914 } 915 if (de->td_cookie + 1 == pde->td_cookie || 916 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 917 continue; /* No hole or invalid cookie. */ 918 nde->td_cookie = de->td_cookie + 1; 919 MPASS(tmpfs_dirent_dup(nde)); 920 MPASS(pde->td_cookie > nde->td_cookie); 921 MPASS(nde->td_cookie > de->td_cookie); 922 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 923 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 924 return; 925 }; 926} 927 928/* 929 * Attaches the directory entry de to the directory represented by vp. 930 * Note that this does not change the link count of the node pointed by 931 * the directory entry, as this is done by tmpfs_alloc_dirent. 932 */ 933void 934tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 935{ 936 struct tmpfs_node *dnode; 937 struct tmpfs_dirent *xde, *nde; 938 939 ASSERT_VOP_ELOCKED(vp, __func__); 940 MPASS(de->td_namelen > 0); 941 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 942 MPASS(de->td_cookie == de->td_hash); 943 944 dnode = VP_TO_TMPFS_DIR(vp); 945 dnode->tn_dir.tn_readdir_lastn = 0; 946 dnode->tn_dir.tn_readdir_lastp = NULL; 947 948 MPASS(!tmpfs_dirent_dup(de)); 949 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 950 if (xde != NULL && tmpfs_dirent_duphead(xde)) 951 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 952 else if (xde != NULL) { 953 /* 954 * Allocate new duphead. Swap xde with duphead to avoid 955 * adding/removing elements with the same hash. 956 */ 957 MPASS(!tmpfs_dirent_dup(xde)); 958 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 959 &nde); 960 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 961 memcpy(nde, xde, sizeof(*xde)); 962 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 963 LIST_INIT(&xde->ud.td_duphead); 964 xde->td_namelen = 0; 965 xde->td_node = NULL; 966 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 967 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 968 } 969 dnode->tn_size += sizeof(struct tmpfs_dirent); 970 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 971 TMPFS_NODE_MODIFIED; 972} 973 974/* 975 * Detaches the directory entry de from the directory represented by vp. 976 * Note that this does not change the link count of the node pointed by 977 * the directory entry, as this is done by tmpfs_free_dirent. 978 */ 979void 980tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 981{ 982 struct tmpfs_mount *tmp; 983 struct tmpfs_dir *head; 984 struct tmpfs_node *dnode; 985 struct tmpfs_dirent *xde; 986 987 ASSERT_VOP_ELOCKED(vp, __func__); 988 989 dnode = VP_TO_TMPFS_DIR(vp); 990 head = &dnode->tn_dir.tn_dirhead; 991 dnode->tn_dir.tn_readdir_lastn = 0; 992 dnode->tn_dir.tn_readdir_lastp = NULL; 993 994 if (tmpfs_dirent_dup(de)) { 995 /* Remove duphead if de was last entry. */ 996 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 997 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 998 MPASS(tmpfs_dirent_duphead(xde)); 999 } else 1000 xde = NULL; 1001 LIST_REMOVE(de, uh.td_dup.entries); 1002 LIST_REMOVE(de, uh.td_dup.index_entries); 1003 if (xde != NULL) { 1004 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1005 RB_REMOVE(tmpfs_dir, head, xde); 1006 tmp = VFS_TO_TMPFS(vp->v_mount); 1007 MPASS(xde->td_node == NULL); 1008 tmpfs_free_dirent(tmp, xde); 1009 } 1010 } 1011 } else 1012 RB_REMOVE(tmpfs_dir, head, de); 1013 1014 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1015 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1016 TMPFS_NODE_MODIFIED; 1017} 1018 1019void 1020tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1021{ 1022 struct tmpfs_dirent *de, *dde, *nde; 1023 1024 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1025 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1026 /* Node may already be destroyed. */ 1027 de->td_node = NULL; 1028 if (tmpfs_dirent_duphead(de)) { 1029 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1030 LIST_REMOVE(dde, uh.td_dup.entries); 1031 dde->td_node = NULL; 1032 tmpfs_free_dirent(tmp, dde); 1033 } 1034 } 1035 tmpfs_free_dirent(tmp, de); 1036 } 1037} 1038 1039/* 1040 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1041 * directory and returns it in the uio space. The function returns 0 1042 * on success, -1 if there was not enough space in the uio structure to 1043 * hold the directory entry or an appropriate error code if another 1044 * error happens. 1045 */ 1046static int 1047tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1048{ 1049 int error; 1050 struct dirent dent; 1051 1052 TMPFS_VALIDATE_DIR(node); 1053 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1054 1055 dent.d_fileno = node->tn_id; 1056 dent.d_type = DT_DIR; 1057 dent.d_namlen = 1; 1058 dent.d_name[0] = '.'; 1059 dent.d_name[1] = '\0'; 1060 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1061 1062 if (dent.d_reclen > uio->uio_resid) 1063 error = EJUSTRETURN; 1064 else 1065 error = uiomove(&dent, dent.d_reclen, uio); 1066 1067 node->tn_status |= TMPFS_NODE_ACCESSED; 1068 1069 return error; 1070} 1071 1072/* 1073 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1074 * directory and returns it in the uio space. The function returns 0 1075 * on success, -1 if there was not enough space in the uio structure to 1076 * hold the directory entry or an appropriate error code if another 1077 * error happens. 1078 */ 1079static int 1080tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1081{ 1082 int error; 1083 struct dirent dent; 1084 1085 TMPFS_VALIDATE_DIR(node); 1086 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1087 1088 /* 1089 * Return ENOENT if the current node is already removed. 1090 */ 1091 TMPFS_ASSERT_LOCKED(node); 1092 if (node->tn_dir.tn_parent == NULL) { 1093 return (ENOENT); 1094 } 1095 1096 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1097 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1098 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1099 1100 dent.d_type = DT_DIR; 1101 dent.d_namlen = 2; 1102 dent.d_name[0] = '.'; 1103 dent.d_name[1] = '.'; 1104 dent.d_name[2] = '\0'; 1105 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1106 1107 if (dent.d_reclen > uio->uio_resid) 1108 error = EJUSTRETURN; 1109 else 1110 error = uiomove(&dent, dent.d_reclen, uio); 1111 1112 node->tn_status |= TMPFS_NODE_ACCESSED; 1113 1114 return error; 1115} 1116 1117/* 1118 * Helper function for tmpfs_readdir. Returns as much directory entries 1119 * as can fit in the uio space. The read starts at uio->uio_offset. 1120 * The function returns 0 on success, -1 if there was not enough space 1121 * in the uio structure to hold the directory entry or an appropriate 1122 * error code if another error happens. 1123 */ 1124int 1125tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1126 u_long *cookies, int *ncookies) 1127{ 1128 struct tmpfs_dir_cursor dc; 1129 struct tmpfs_dirent *de; 1130 off_t off; 1131 int error; 1132 1133 TMPFS_VALIDATE_DIR(node); 1134 1135 off = 0; 1136 1137 /* 1138 * Lookup the node from the current offset. The starting offset of 1139 * 0 will lookup both '.' and '..', and then the first real entry, 1140 * or EOF if there are none. Then find all entries for the dir that 1141 * fit into the buffer. Once no more entries are found (de == NULL), 1142 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1143 * call to return 0. 1144 */ 1145 switch (uio->uio_offset) { 1146 case TMPFS_DIRCOOKIE_DOT: 1147 error = tmpfs_dir_getdotdent(node, uio); 1148 if (error != 0) 1149 return (error); 1150 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1151 if (cookies != NULL) 1152 cookies[(*ncookies)++] = off = uio->uio_offset; 1153 /* FALLTHROUGH */ 1154 case TMPFS_DIRCOOKIE_DOTDOT: 1155 error = tmpfs_dir_getdotdotdent(node, uio); 1156 if (error != 0) 1157 return (error); 1158 de = tmpfs_dir_first(node, &dc); 1159 uio->uio_offset = tmpfs_dirent_cookie(de); 1160 if (cookies != NULL) 1161 cookies[(*ncookies)++] = off = uio->uio_offset; 1162 /* EOF. */ 1163 if (de == NULL) 1164 return (0); 1165 break; 1166 case TMPFS_DIRCOOKIE_EOF: 1167 return (0); 1168 default: 1169 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1170 if (de == NULL) 1171 return (EINVAL); 1172 if (cookies != NULL) 1173 off = tmpfs_dirent_cookie(de); 1174 } 1175 1176 /* Read as much entries as possible; i.e., until we reach the end of 1177 * the directory or we exhaust uio space. */ 1178 do { 1179 struct dirent d; 1180 1181 /* Create a dirent structure representing the current 1182 * tmpfs_node and fill it. */ 1183 if (de->td_node == NULL) { 1184 d.d_fileno = 1; 1185 d.d_type = DT_WHT; 1186 } else { 1187 d.d_fileno = de->td_node->tn_id; 1188 switch (de->td_node->tn_type) { 1189 case VBLK: 1190 d.d_type = DT_BLK; 1191 break; 1192 1193 case VCHR: 1194 d.d_type = DT_CHR; 1195 break; 1196 1197 case VDIR: 1198 d.d_type = DT_DIR; 1199 break; 1200 1201 case VFIFO: 1202 d.d_type = DT_FIFO; 1203 break; 1204 1205 case VLNK: 1206 d.d_type = DT_LNK; 1207 break; 1208 1209 case VREG: 1210 d.d_type = DT_REG; 1211 break; 1212 1213 case VSOCK: 1214 d.d_type = DT_SOCK; 1215 break; 1216 1217 default: 1218 panic("tmpfs_dir_getdents: type %p %d", 1219 de->td_node, (int)de->td_node->tn_type); 1220 } 1221 } 1222 d.d_namlen = de->td_namelen; 1223 MPASS(de->td_namelen < sizeof(d.d_name)); 1224 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1225 d.d_name[de->td_namelen] = '\0'; 1226 d.d_reclen = GENERIC_DIRSIZ(&d); 1227 1228 /* Stop reading if the directory entry we are treating is 1229 * bigger than the amount of data that can be returned. */ 1230 if (d.d_reclen > uio->uio_resid) { 1231 error = EJUSTRETURN; 1232 break; 1233 } 1234 1235 /* Copy the new dirent structure into the output buffer and 1236 * advance pointers. */ 1237 error = uiomove(&d, d.d_reclen, uio); 1238 if (error == 0) { 1239 de = tmpfs_dir_next(node, &dc); 1240 if (cookies != NULL) { 1241 off = tmpfs_dirent_cookie(de); 1242 MPASS(*ncookies < maxcookies); 1243 cookies[(*ncookies)++] = off; 1244 } 1245 } 1246 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1247 1248 /* Skip setting off when using cookies as it is already done above. */ 1249 if (cookies == NULL) 1250 off = tmpfs_dirent_cookie(de); 1251 1252 /* Update the offset and cache. */ 1253 uio->uio_offset = off; 1254 node->tn_dir.tn_readdir_lastn = off; 1255 node->tn_dir.tn_readdir_lastp = de; 1256 1257 node->tn_status |= TMPFS_NODE_ACCESSED; 1258 return error; 1259} 1260 1261int 1262tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1263{ 1264 struct tmpfs_dirent *de; 1265 int error; 1266 1267 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1268 cnp->cn_nameptr, cnp->cn_namelen, &de); 1269 if (error != 0) 1270 return (error); 1271 tmpfs_dir_attach(dvp, de); 1272 return (0); 1273} 1274 1275void 1276tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1277{ 1278 struct tmpfs_dirent *de; 1279 1280 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1281 MPASS(de != NULL && de->td_node == NULL); 1282 tmpfs_dir_detach(dvp, de); 1283 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1284} 1285 1286/* 1287 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1288 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1289 * 'newsize' must be positive. 1290 * 1291 * Returns zero on success or an appropriate error code on failure. 1292 */ 1293int 1294tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1295{ 1296 struct tmpfs_mount *tmp; 1297 struct tmpfs_node *node; 1298 vm_object_t uobj; 1299 vm_page_t m, ma[1]; 1300 vm_pindex_t idx, newpages, oldpages; 1301 off_t oldsize; 1302 int base, rv; 1303 1304 MPASS(vp->v_type == VREG); 1305 MPASS(newsize >= 0); 1306 1307 node = VP_TO_TMPFS_NODE(vp); 1308 uobj = node->tn_reg.tn_aobj; 1309 tmp = VFS_TO_TMPFS(vp->v_mount); 1310 1311 /* 1312 * Convert the old and new sizes to the number of pages needed to 1313 * store them. It may happen that we do not need to do anything 1314 * because the last allocated page can accommodate the change on 1315 * its own. 1316 */ 1317 oldsize = node->tn_size; 1318 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1319 MPASS(oldpages == uobj->size); 1320 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1321 if (newpages > oldpages && 1322 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1323 return (ENOSPC); 1324 1325 VM_OBJECT_WLOCK(uobj); 1326 if (newsize < oldsize) { 1327 /* 1328 * Zero the truncated part of the last page. 1329 */ 1330 base = newsize & PAGE_MASK; 1331 if (base != 0) { 1332 idx = OFF_TO_IDX(newsize); 1333retry: 1334 m = vm_page_lookup(uobj, idx); 1335 if (m != NULL) { 1336 if (vm_page_sleep_if_busy(m, "tmfssz")) 1337 goto retry; 1338 MPASS(m->valid == VM_PAGE_BITS_ALL); 1339 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1340 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1341 if (m == NULL) { 1342 VM_OBJECT_WUNLOCK(uobj); 1343 VM_WAIT; 1344 VM_OBJECT_WLOCK(uobj); 1345 goto retry; 1346 } else if (m->valid != VM_PAGE_BITS_ALL) { 1347 ma[0] = m; 1348 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1349 m = vm_page_lookup(uobj, idx); 1350 } else 1351 /* A cached page was reactivated. */ 1352 rv = VM_PAGER_OK; 1353 vm_page_lock(m); 1354 if (rv == VM_PAGER_OK) { 1355 vm_page_deactivate(m); 1356 vm_page_unlock(m); 1357 vm_page_xunbusy(m); 1358 } else { 1359 vm_page_free(m); 1360 vm_page_unlock(m); 1361 if (ignerr) 1362 m = NULL; 1363 else { 1364 VM_OBJECT_WUNLOCK(uobj); 1365 return (EIO); 1366 } 1367 } 1368 } 1369 if (m != NULL) { 1370 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1371 vm_page_dirty(m); 1372 vm_pager_page_unswapped(m); 1373 } 1374 } 1375 1376 /* 1377 * Release any swap space and free any whole pages. 1378 */ 1379 if (newpages < oldpages) { 1380 swap_pager_freespace(uobj, newpages, oldpages - 1381 newpages); 1382 vm_object_page_remove(uobj, newpages, 0, 0); 1383 } 1384 } 1385 uobj->size = newpages; 1386 VM_OBJECT_WUNLOCK(uobj); 1387 1388 TMPFS_LOCK(tmp); 1389 tmp->tm_pages_used += (newpages - oldpages); 1390 TMPFS_UNLOCK(tmp); 1391 1392 node->tn_size = newsize; 1393 return (0); 1394} 1395 1396/* 1397 * Change flags of the given vnode. 1398 * Caller should execute tmpfs_update on vp after a successful execution. 1399 * The vnode must be locked on entry and remain locked on exit. 1400 */ 1401int 1402tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1403 struct thread *p) 1404{ 1405 int error; 1406 struct tmpfs_node *node; 1407 1408 MPASS(VOP_ISLOCKED(vp)); 1409 1410 node = VP_TO_TMPFS_NODE(vp); 1411 1412 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1413 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1414 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1415 UF_SPARSE | UF_SYSTEM)) != 0) 1416 return (EOPNOTSUPP); 1417 1418 /* Disallow this operation if the file system is mounted read-only. */ 1419 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1420 return EROFS; 1421 1422 /* 1423 * Callers may only modify the file flags on objects they 1424 * have VADMIN rights for. 1425 */ 1426 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1427 return (error); 1428 /* 1429 * Unprivileged processes are not permitted to unset system 1430 * flags, or modify flags if any system flags are set. 1431 */ 1432 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1433 if (node->tn_flags & 1434 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1435 error = securelevel_gt(cred, 0); 1436 if (error) 1437 return (error); 1438 } 1439 } else { 1440 if (node->tn_flags & 1441 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1442 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1443 return (EPERM); 1444 } 1445 node->tn_flags = flags; 1446 node->tn_status |= TMPFS_NODE_CHANGED; 1447 1448 MPASS(VOP_ISLOCKED(vp)); 1449 1450 return 0; 1451} 1452 1453/* 1454 * Change access mode on the given vnode. 1455 * Caller should execute tmpfs_update on vp after a successful execution. 1456 * The vnode must be locked on entry and remain locked on exit. 1457 */ 1458int 1459tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1460{ 1461 int error; 1462 struct tmpfs_node *node; 1463 1464 MPASS(VOP_ISLOCKED(vp)); 1465 1466 node = VP_TO_TMPFS_NODE(vp); 1467 1468 /* Disallow this operation if the file system is mounted read-only. */ 1469 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1470 return EROFS; 1471 1472 /* Immutable or append-only files cannot be modified, either. */ 1473 if (node->tn_flags & (IMMUTABLE | APPEND)) 1474 return EPERM; 1475 1476 /* 1477 * To modify the permissions on a file, must possess VADMIN 1478 * for that file. 1479 */ 1480 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1481 return (error); 1482 1483 /* 1484 * Privileged processes may set the sticky bit on non-directories, 1485 * as well as set the setgid bit on a file with a group that the 1486 * process is not a member of. 1487 */ 1488 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1489 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1490 return (EFTYPE); 1491 } 1492 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1493 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1494 if (error) 1495 return (error); 1496 } 1497 1498 1499 node->tn_mode &= ~ALLPERMS; 1500 node->tn_mode |= mode & ALLPERMS; 1501 1502 node->tn_status |= TMPFS_NODE_CHANGED; 1503 1504 MPASS(VOP_ISLOCKED(vp)); 1505 1506 return 0; 1507} 1508 1509/* 1510 * Change ownership of the given vnode. At least one of uid or gid must 1511 * be different than VNOVAL. If one is set to that value, the attribute 1512 * is unchanged. 1513 * Caller should execute tmpfs_update on vp after a successful execution. 1514 * The vnode must be locked on entry and remain locked on exit. 1515 */ 1516int 1517tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1518 struct thread *p) 1519{ 1520 int error; 1521 struct tmpfs_node *node; 1522 uid_t ouid; 1523 gid_t ogid; 1524 1525 MPASS(VOP_ISLOCKED(vp)); 1526 1527 node = VP_TO_TMPFS_NODE(vp); 1528 1529 /* Assign default values if they are unknown. */ 1530 MPASS(uid != VNOVAL || gid != VNOVAL); 1531 if (uid == VNOVAL) 1532 uid = node->tn_uid; 1533 if (gid == VNOVAL) 1534 gid = node->tn_gid; 1535 MPASS(uid != VNOVAL && gid != VNOVAL); 1536 1537 /* Disallow this operation if the file system is mounted read-only. */ 1538 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1539 return EROFS; 1540 1541 /* Immutable or append-only files cannot be modified, either. */ 1542 if (node->tn_flags & (IMMUTABLE | APPEND)) 1543 return EPERM; 1544 1545 /* 1546 * To modify the ownership of a file, must possess VADMIN for that 1547 * file. 1548 */ 1549 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1550 return (error); 1551 1552 /* 1553 * To change the owner of a file, or change the group of a file to a 1554 * group of which we are not a member, the caller must have 1555 * privilege. 1556 */ 1557 if ((uid != node->tn_uid || 1558 (gid != node->tn_gid && !groupmember(gid, cred))) && 1559 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1560 return (error); 1561 1562 ogid = node->tn_gid; 1563 ouid = node->tn_uid; 1564 1565 node->tn_uid = uid; 1566 node->tn_gid = gid; 1567 1568 node->tn_status |= TMPFS_NODE_CHANGED; 1569 1570 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1571 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1572 node->tn_mode &= ~(S_ISUID | S_ISGID); 1573 } 1574 1575 MPASS(VOP_ISLOCKED(vp)); 1576 1577 return 0; 1578} 1579 1580/* 1581 * Change size of the given vnode. 1582 * Caller should execute tmpfs_update on vp after a successful execution. 1583 * The vnode must be locked on entry and remain locked on exit. 1584 */ 1585int 1586tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1587 struct thread *p) 1588{ 1589 int error; 1590 struct tmpfs_node *node; 1591 1592 MPASS(VOP_ISLOCKED(vp)); 1593 1594 node = VP_TO_TMPFS_NODE(vp); 1595 1596 /* Decide whether this is a valid operation based on the file type. */ 1597 error = 0; 1598 switch (vp->v_type) { 1599 case VDIR: 1600 return EISDIR; 1601 1602 case VREG: 1603 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1604 return EROFS; 1605 break; 1606 1607 case VBLK: 1608 /* FALLTHROUGH */ 1609 case VCHR: 1610 /* FALLTHROUGH */ 1611 case VFIFO: 1612 /* Allow modifications of special files even if in the file 1613 * system is mounted read-only (we are not modifying the 1614 * files themselves, but the objects they represent). */ 1615 return 0; 1616 1617 default: 1618 /* Anything else is unsupported. */ 1619 return EOPNOTSUPP; 1620 } 1621 1622 /* Immutable or append-only files cannot be modified, either. */ 1623 if (node->tn_flags & (IMMUTABLE | APPEND)) 1624 return EPERM; 1625 1626 error = tmpfs_truncate(vp, size); 1627 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1628 * for us, as will update tn_status; no need to do that here. */ 1629 1630 MPASS(VOP_ISLOCKED(vp)); 1631 1632 return error; 1633} 1634 1635/* 1636 * Change access and modification times of the given vnode. 1637 * Caller should execute tmpfs_update on vp after a successful execution. 1638 * The vnode must be locked on entry and remain locked on exit. 1639 */ 1640int 1641tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 1642 struct ucred *cred, struct thread *l) 1643{ 1644 int error; 1645 struct tmpfs_node *node; 1646 1647 MPASS(VOP_ISLOCKED(vp)); 1648 1649 node = VP_TO_TMPFS_NODE(vp); 1650 1651 /* Disallow this operation if the file system is mounted read-only. */ 1652 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1653 return EROFS; 1654 1655 /* Immutable or append-only files cannot be modified, either. */ 1656 if (node->tn_flags & (IMMUTABLE | APPEND)) 1657 return EPERM; 1658 1659 error = vn_utimes_perm(vp, vap, cred, l); 1660 if (error != 0) 1661 return (error); 1662 1663 if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL) 1664 node->tn_status |= TMPFS_NODE_ACCESSED; 1665 1666 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL) 1667 node->tn_status |= TMPFS_NODE_MODIFIED; 1668 1669 if (vap->va_birthtime.tv_nsec != VNOVAL && 1670 vap->va_birthtime.tv_nsec != VNOVAL) 1671 node->tn_status |= TMPFS_NODE_MODIFIED; 1672 1673 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1674 1675 if (vap->va_birthtime.tv_nsec != VNOVAL && 1676 vap->va_birthtime.tv_nsec != VNOVAL) 1677 node->tn_birthtime = vap->va_birthtime; 1678 MPASS(VOP_ISLOCKED(vp)); 1679 1680 return 0; 1681} 1682 1683/* Sync timestamps */ 1684void 1685tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1686 const struct timespec *mod) 1687{ 1688 struct tmpfs_node *node; 1689 struct timespec now; 1690 1691 node = VP_TO_TMPFS_NODE(vp); 1692 1693 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1694 TMPFS_NODE_CHANGED)) == 0) 1695 return; 1696 1697 vfs_timestamp(&now); 1698 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1699 if (acc == NULL) 1700 acc = &now; 1701 node->tn_atime = *acc; 1702 } 1703 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1704 if (mod == NULL) 1705 mod = &now; 1706 node->tn_mtime = *mod; 1707 } 1708 if (node->tn_status & TMPFS_NODE_CHANGED) { 1709 node->tn_ctime = now; 1710 } 1711 node->tn_status &= 1712 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1713} 1714 1715void 1716tmpfs_update(struct vnode *vp) 1717{ 1718 1719 tmpfs_itimes(vp, NULL, NULL); 1720} 1721 1722int 1723tmpfs_truncate(struct vnode *vp, off_t length) 1724{ 1725 int error; 1726 struct tmpfs_node *node; 1727 1728 node = VP_TO_TMPFS_NODE(vp); 1729 1730 if (length < 0) { 1731 error = EINVAL; 1732 goto out; 1733 } 1734 1735 if (node->tn_size == length) { 1736 error = 0; 1737 goto out; 1738 } 1739 1740 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1741 return (EFBIG); 1742 1743 error = tmpfs_reg_resize(vp, length, FALSE); 1744 if (error == 0) { 1745 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1746 } 1747 1748out: 1749 tmpfs_update(vp); 1750 1751 return error; 1752} 1753 1754static __inline int 1755tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1756{ 1757 if (a->td_hash > b->td_hash) 1758 return (1); 1759 else if (a->td_hash < b->td_hash) 1760 return (-1); 1761 return (0); 1762} 1763 1764RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1765