tmpfs_subr.c revision 269168
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 269168 2014-07-28 01:04:33Z 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 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED); 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 error = EMLINK; 681 goto out; 682 } 683 684 parent = dnode; 685 MPASS(parent != NULL); 686 } else 687 parent = NULL; 688 689 /* Allocate a node that represents the new file. */ 690 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid, 691 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node); 692 if (error != 0) 693 goto out; 694 695 /* Allocate a directory entry that points to the new file. */ 696 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 697 &de); 698 if (error != 0) { 699 tmpfs_free_node(tmp, node); 700 goto out; 701 } 702 703 /* Allocate a vnode for the new file. */ 704 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 705 if (error != 0) { 706 tmpfs_free_dirent(tmp, de); 707 tmpfs_free_node(tmp, node); 708 goto out; 709 } 710 711 /* Now that all required items are allocated, we can proceed to 712 * insert the new node into the directory, an operation that 713 * cannot fail. */ 714 if (cnp->cn_flags & ISWHITEOUT) 715 tmpfs_dir_whiteout_remove(dvp, cnp); 716 tmpfs_dir_attach(dvp, de); 717 718out: 719 720 return error; 721} 722 723static struct tmpfs_dirent * 724tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 725{ 726 struct tmpfs_dirent *de; 727 728 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 729 dc->tdc_tree = de; 730 if (de != NULL && tmpfs_dirent_duphead(de)) 731 de = LIST_FIRST(&de->ud.td_duphead); 732 dc->tdc_current = de; 733 734 return (dc->tdc_current); 735} 736 737static struct tmpfs_dirent * 738tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 739{ 740 struct tmpfs_dirent *de; 741 742 MPASS(dc->tdc_tree != NULL); 743 if (tmpfs_dirent_dup(dc->tdc_current)) { 744 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 745 if (dc->tdc_current != NULL) 746 return (dc->tdc_current); 747 } 748 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 749 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 750 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 751 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 752 MPASS(dc->tdc_current != NULL); 753 } 754 755 return (dc->tdc_current); 756} 757 758/* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 759static struct tmpfs_dirent * 760tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 761{ 762 struct tmpfs_dirent *de, dekey; 763 764 dekey.td_hash = hash; 765 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 766 return (de); 767} 768 769/* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 770static struct tmpfs_dirent * 771tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 772 struct tmpfs_dir_cursor *dc) 773{ 774 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 775 struct tmpfs_dirent *de, dekey; 776 777 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 778 779 if (cookie == node->tn_dir.tn_readdir_lastn && 780 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 781 /* Protect against possible race, tn_readdir_last[pn] 782 * may be updated with only shared vnode lock held. */ 783 if (cookie == tmpfs_dirent_cookie(de)) 784 goto out; 785 } 786 787 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 788 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 789 uh.td_dup.index_entries) { 790 MPASS(tmpfs_dirent_dup(de)); 791 if (de->td_cookie == cookie) 792 goto out; 793 /* dupindex list is sorted. */ 794 if (de->td_cookie < cookie) { 795 de = NULL; 796 goto out; 797 } 798 } 799 MPASS(de == NULL); 800 goto out; 801 } 802 803 MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie); 804 dekey.td_hash = cookie; 805 /* Recover if direntry for cookie was removed */ 806 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 807 dc->tdc_tree = de; 808 dc->tdc_current = de; 809 if (de != NULL && tmpfs_dirent_duphead(de)) { 810 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 811 MPASS(dc->tdc_current != NULL); 812 } 813 return (dc->tdc_current); 814 815out: 816 dc->tdc_tree = de; 817 dc->tdc_current = de; 818 if (de != NULL && tmpfs_dirent_dup(de)) 819 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 820 de->td_hash); 821 return (dc->tdc_current); 822} 823 824/* 825 * Looks for a directory entry in the directory represented by node. 826 * 'cnp' describes the name of the entry to look for. Note that the . 827 * and .. components are not allowed as they do not physically exist 828 * within directories. 829 * 830 * Returns a pointer to the entry when found, otherwise NULL. 831 */ 832struct tmpfs_dirent * 833tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 834 struct componentname *cnp) 835{ 836 struct tmpfs_dir_duphead *duphead; 837 struct tmpfs_dirent *de; 838 uint32_t hash; 839 840 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 841 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 842 cnp->cn_nameptr[1] == '.'))); 843 TMPFS_VALIDATE_DIR(node); 844 845 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 846 de = tmpfs_dir_xlookup_hash(node, hash); 847 if (de != NULL && tmpfs_dirent_duphead(de)) { 848 duphead = &de->ud.td_duphead; 849 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 850 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 851 cnp->cn_namelen)) 852 break; 853 } 854 } else if (de != NULL) { 855 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 856 cnp->cn_namelen)) 857 de = NULL; 858 } 859 if (de != NULL && f != NULL && de->td_node != f) 860 de = NULL; 861 862 return (de); 863} 864 865/* 866 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 867 * list, allocate new cookie value. 868 */ 869static void 870tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 871 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 872{ 873 struct tmpfs_dir_duphead *dupindex; 874 struct tmpfs_dirent *de, *pde; 875 876 dupindex = &dnode->tn_dir.tn_dupindex; 877 de = LIST_FIRST(dupindex); 878 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 879 if (de == NULL) 880 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 881 else 882 nde->td_cookie = de->td_cookie + 1; 883 MPASS(tmpfs_dirent_dup(nde)); 884 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 885 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 886 return; 887 } 888 889 /* 890 * Cookie numbers are near exhaustion. Scan dupindex list for unused 891 * numbers. dupindex list is sorted in descending order. Keep it so 892 * after inserting nde. 893 */ 894 while (1) { 895 pde = de; 896 de = LIST_NEXT(de, uh.td_dup.index_entries); 897 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 898 /* 899 * Last element of the index doesn't have minimal cookie 900 * value, use it. 901 */ 902 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 903 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 904 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 905 return; 906 } else if (de == NULL) { 907 /* 908 * We are so lucky have 2^30 hash duplicates in single 909 * directory :) Return largest possible cookie value. 910 * It should be fine except possible issues with 911 * VOP_READDIR restart. 912 */ 913 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 914 LIST_INSERT_HEAD(dupindex, nde, 915 uh.td_dup.index_entries); 916 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 917 return; 918 } 919 if (de->td_cookie + 1 == pde->td_cookie || 920 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 921 continue; /* No hole or invalid cookie. */ 922 nde->td_cookie = de->td_cookie + 1; 923 MPASS(tmpfs_dirent_dup(nde)); 924 MPASS(pde->td_cookie > nde->td_cookie); 925 MPASS(nde->td_cookie > de->td_cookie); 926 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 927 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 928 return; 929 }; 930} 931 932/* 933 * Attaches the directory entry de to the directory represented by vp. 934 * Note that this does not change the link count of the node pointed by 935 * the directory entry, as this is done by tmpfs_alloc_dirent. 936 */ 937void 938tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 939{ 940 struct tmpfs_node *dnode; 941 struct tmpfs_dirent *xde, *nde; 942 943 ASSERT_VOP_ELOCKED(vp, __func__); 944 MPASS(de->td_namelen > 0); 945 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 946 MPASS(de->td_cookie == de->td_hash); 947 948 dnode = VP_TO_TMPFS_DIR(vp); 949 dnode->tn_dir.tn_readdir_lastn = 0; 950 dnode->tn_dir.tn_readdir_lastp = NULL; 951 952 MPASS(!tmpfs_dirent_dup(de)); 953 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 954 if (xde != NULL && tmpfs_dirent_duphead(xde)) 955 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 956 else if (xde != NULL) { 957 /* 958 * Allocate new duphead. Swap xde with duphead to avoid 959 * adding/removing elements with the same hash. 960 */ 961 MPASS(!tmpfs_dirent_dup(xde)); 962 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 963 &nde); 964 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 965 memcpy(nde, xde, sizeof(*xde)); 966 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 967 LIST_INIT(&xde->ud.td_duphead); 968 xde->td_namelen = 0; 969 xde->td_node = NULL; 970 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 971 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 972 } 973 dnode->tn_size += sizeof(struct tmpfs_dirent); 974 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 975 TMPFS_NODE_MODIFIED; 976} 977 978/* 979 * Detaches the directory entry de from the directory represented by vp. 980 * Note that this does not change the link count of the node pointed by 981 * the directory entry, as this is done by tmpfs_free_dirent. 982 */ 983void 984tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 985{ 986 struct tmpfs_mount *tmp; 987 struct tmpfs_dir *head; 988 struct tmpfs_node *dnode; 989 struct tmpfs_dirent *xde; 990 991 ASSERT_VOP_ELOCKED(vp, __func__); 992 993 dnode = VP_TO_TMPFS_DIR(vp); 994 head = &dnode->tn_dir.tn_dirhead; 995 dnode->tn_dir.tn_readdir_lastn = 0; 996 dnode->tn_dir.tn_readdir_lastp = NULL; 997 998 if (tmpfs_dirent_dup(de)) { 999 /* Remove duphead if de was last entry. */ 1000 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1001 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1002 MPASS(tmpfs_dirent_duphead(xde)); 1003 } else 1004 xde = NULL; 1005 LIST_REMOVE(de, uh.td_dup.entries); 1006 LIST_REMOVE(de, uh.td_dup.index_entries); 1007 if (xde != NULL) { 1008 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1009 RB_REMOVE(tmpfs_dir, head, xde); 1010 tmp = VFS_TO_TMPFS(vp->v_mount); 1011 MPASS(xde->td_node == NULL); 1012 tmpfs_free_dirent(tmp, xde); 1013 } 1014 } 1015 } else 1016 RB_REMOVE(tmpfs_dir, head, de); 1017 1018 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1019 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1020 TMPFS_NODE_MODIFIED; 1021} 1022 1023void 1024tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1025{ 1026 struct tmpfs_dirent *de, *dde, *nde; 1027 1028 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1029 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1030 /* Node may already be destroyed. */ 1031 de->td_node = NULL; 1032 if (tmpfs_dirent_duphead(de)) { 1033 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1034 LIST_REMOVE(dde, uh.td_dup.entries); 1035 dde->td_node = NULL; 1036 tmpfs_free_dirent(tmp, dde); 1037 } 1038 } 1039 tmpfs_free_dirent(tmp, de); 1040 } 1041} 1042 1043/* 1044 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1045 * directory and returns it in the uio space. The function returns 0 1046 * on success, -1 if there was not enough space in the uio structure to 1047 * hold the directory entry or an appropriate error code if another 1048 * error happens. 1049 */ 1050static int 1051tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1052{ 1053 int error; 1054 struct dirent dent; 1055 1056 TMPFS_VALIDATE_DIR(node); 1057 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1058 1059 dent.d_fileno = node->tn_id; 1060 dent.d_type = DT_DIR; 1061 dent.d_namlen = 1; 1062 dent.d_name[0] = '.'; 1063 dent.d_name[1] = '\0'; 1064 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1065 1066 if (dent.d_reclen > uio->uio_resid) 1067 error = EJUSTRETURN; 1068 else 1069 error = uiomove(&dent, dent.d_reclen, uio); 1070 1071 node->tn_status |= TMPFS_NODE_ACCESSED; 1072 1073 return error; 1074} 1075 1076/* 1077 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1078 * directory and returns it in the uio space. The function returns 0 1079 * on success, -1 if there was not enough space in the uio structure to 1080 * hold the directory entry or an appropriate error code if another 1081 * error happens. 1082 */ 1083static int 1084tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1085{ 1086 int error; 1087 struct dirent dent; 1088 1089 TMPFS_VALIDATE_DIR(node); 1090 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1091 1092 /* 1093 * Return ENOENT if the current node is already removed. 1094 */ 1095 TMPFS_ASSERT_LOCKED(node); 1096 if (node->tn_dir.tn_parent == NULL) { 1097 return (ENOENT); 1098 } 1099 1100 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1101 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1102 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1103 1104 dent.d_type = DT_DIR; 1105 dent.d_namlen = 2; 1106 dent.d_name[0] = '.'; 1107 dent.d_name[1] = '.'; 1108 dent.d_name[2] = '\0'; 1109 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1110 1111 if (dent.d_reclen > uio->uio_resid) 1112 error = EJUSTRETURN; 1113 else 1114 error = uiomove(&dent, dent.d_reclen, uio); 1115 1116 node->tn_status |= TMPFS_NODE_ACCESSED; 1117 1118 return error; 1119} 1120 1121/* 1122 * Helper function for tmpfs_readdir. Returns as much directory entries 1123 * as can fit in the uio space. The read starts at uio->uio_offset. 1124 * The function returns 0 on success, -1 if there was not enough space 1125 * in the uio structure to hold the directory entry or an appropriate 1126 * error code if another error happens. 1127 */ 1128int 1129tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1130 u_long *cookies, int *ncookies) 1131{ 1132 struct tmpfs_dir_cursor dc; 1133 struct tmpfs_dirent *de; 1134 off_t off; 1135 int error; 1136 1137 TMPFS_VALIDATE_DIR(node); 1138 1139 off = 0; 1140 1141 /* 1142 * Lookup the node from the current offset. The starting offset of 1143 * 0 will lookup both '.' and '..', and then the first real entry, 1144 * or EOF if there are none. Then find all entries for the dir that 1145 * fit into the buffer. Once no more entries are found (de == NULL), 1146 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1147 * call to return 0. 1148 */ 1149 switch (uio->uio_offset) { 1150 case TMPFS_DIRCOOKIE_DOT: 1151 error = tmpfs_dir_getdotdent(node, uio); 1152 if (error != 0) 1153 return (error); 1154 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1155 if (cookies != NULL) 1156 cookies[(*ncookies)++] = off = uio->uio_offset; 1157 /* FALLTHROUGH */ 1158 case TMPFS_DIRCOOKIE_DOTDOT: 1159 error = tmpfs_dir_getdotdotdent(node, uio); 1160 if (error != 0) 1161 return (error); 1162 de = tmpfs_dir_first(node, &dc); 1163 uio->uio_offset = tmpfs_dirent_cookie(de); 1164 if (cookies != NULL) 1165 cookies[(*ncookies)++] = off = uio->uio_offset; 1166 /* EOF. */ 1167 if (de == NULL) 1168 return (0); 1169 break; 1170 case TMPFS_DIRCOOKIE_EOF: 1171 return (0); 1172 default: 1173 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1174 if (de == NULL) 1175 return (EINVAL); 1176 if (cookies != NULL) 1177 off = tmpfs_dirent_cookie(de); 1178 } 1179 1180 /* Read as much entries as possible; i.e., until we reach the end of 1181 * the directory or we exhaust uio space. */ 1182 do { 1183 struct dirent d; 1184 1185 /* Create a dirent structure representing the current 1186 * tmpfs_node and fill it. */ 1187 if (de->td_node == NULL) { 1188 d.d_fileno = 1; 1189 d.d_type = DT_WHT; 1190 } else { 1191 d.d_fileno = de->td_node->tn_id; 1192 switch (de->td_node->tn_type) { 1193 case VBLK: 1194 d.d_type = DT_BLK; 1195 break; 1196 1197 case VCHR: 1198 d.d_type = DT_CHR; 1199 break; 1200 1201 case VDIR: 1202 d.d_type = DT_DIR; 1203 break; 1204 1205 case VFIFO: 1206 d.d_type = DT_FIFO; 1207 break; 1208 1209 case VLNK: 1210 d.d_type = DT_LNK; 1211 break; 1212 1213 case VREG: 1214 d.d_type = DT_REG; 1215 break; 1216 1217 case VSOCK: 1218 d.d_type = DT_SOCK; 1219 break; 1220 1221 default: 1222 panic("tmpfs_dir_getdents: type %p %d", 1223 de->td_node, (int)de->td_node->tn_type); 1224 } 1225 } 1226 d.d_namlen = de->td_namelen; 1227 MPASS(de->td_namelen < sizeof(d.d_name)); 1228 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1229 d.d_name[de->td_namelen] = '\0'; 1230 d.d_reclen = GENERIC_DIRSIZ(&d); 1231 1232 /* Stop reading if the directory entry we are treating is 1233 * bigger than the amount of data that can be returned. */ 1234 if (d.d_reclen > uio->uio_resid) { 1235 error = EJUSTRETURN; 1236 break; 1237 } 1238 1239 /* Copy the new dirent structure into the output buffer and 1240 * advance pointers. */ 1241 error = uiomove(&d, d.d_reclen, uio); 1242 if (error == 0) { 1243 de = tmpfs_dir_next(node, &dc); 1244 if (cookies != NULL) { 1245 off = tmpfs_dirent_cookie(de); 1246 MPASS(*ncookies < maxcookies); 1247 cookies[(*ncookies)++] = off; 1248 } 1249 } 1250 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1251 1252 /* Skip setting off when using cookies as it is already done above. */ 1253 if (cookies == NULL) 1254 off = tmpfs_dirent_cookie(de); 1255 1256 /* Update the offset and cache. */ 1257 uio->uio_offset = off; 1258 node->tn_dir.tn_readdir_lastn = off; 1259 node->tn_dir.tn_readdir_lastp = de; 1260 1261 node->tn_status |= TMPFS_NODE_ACCESSED; 1262 return error; 1263} 1264 1265int 1266tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1267{ 1268 struct tmpfs_dirent *de; 1269 int error; 1270 1271 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1272 cnp->cn_nameptr, cnp->cn_namelen, &de); 1273 if (error != 0) 1274 return (error); 1275 tmpfs_dir_attach(dvp, de); 1276 return (0); 1277} 1278 1279void 1280tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1281{ 1282 struct tmpfs_dirent *de; 1283 1284 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1285 MPASS(de != NULL && de->td_node == NULL); 1286 tmpfs_dir_detach(dvp, de); 1287 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1288} 1289 1290/* 1291 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1292 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1293 * 'newsize' must be positive. 1294 * 1295 * Returns zero on success or an appropriate error code on failure. 1296 */ 1297int 1298tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1299{ 1300 struct tmpfs_mount *tmp; 1301 struct tmpfs_node *node; 1302 vm_object_t uobj; 1303 vm_page_t m, ma[1]; 1304 vm_pindex_t idx, newpages, oldpages; 1305 off_t oldsize; 1306 int base, rv; 1307 1308 MPASS(vp->v_type == VREG); 1309 MPASS(newsize >= 0); 1310 1311 node = VP_TO_TMPFS_NODE(vp); 1312 uobj = node->tn_reg.tn_aobj; 1313 tmp = VFS_TO_TMPFS(vp->v_mount); 1314 1315 /* 1316 * Convert the old and new sizes to the number of pages needed to 1317 * store them. It may happen that we do not need to do anything 1318 * because the last allocated page can accommodate the change on 1319 * its own. 1320 */ 1321 oldsize = node->tn_size; 1322 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1323 MPASS(oldpages == uobj->size); 1324 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1325 if (newpages > oldpages && 1326 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1327 return (ENOSPC); 1328 1329 VM_OBJECT_WLOCK(uobj); 1330 if (newsize < oldsize) { 1331 /* 1332 * Zero the truncated part of the last page. 1333 */ 1334 base = newsize & PAGE_MASK; 1335 if (base != 0) { 1336 idx = OFF_TO_IDX(newsize); 1337retry: 1338 m = vm_page_lookup(uobj, idx); 1339 if (m != NULL) { 1340 if (vm_page_sleep_if_busy(m, "tmfssz")) 1341 goto retry; 1342 MPASS(m->valid == VM_PAGE_BITS_ALL); 1343 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1344 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1345 if (m == NULL) { 1346 VM_OBJECT_WUNLOCK(uobj); 1347 VM_WAIT; 1348 VM_OBJECT_WLOCK(uobj); 1349 goto retry; 1350 } else if (m->valid != VM_PAGE_BITS_ALL) { 1351 ma[0] = m; 1352 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1353 m = vm_page_lookup(uobj, idx); 1354 } else 1355 /* A cached page was reactivated. */ 1356 rv = VM_PAGER_OK; 1357 vm_page_lock(m); 1358 if (rv == VM_PAGER_OK) { 1359 vm_page_deactivate(m); 1360 vm_page_unlock(m); 1361 vm_page_xunbusy(m); 1362 } else { 1363 vm_page_free(m); 1364 vm_page_unlock(m); 1365 if (ignerr) 1366 m = NULL; 1367 else { 1368 VM_OBJECT_WUNLOCK(uobj); 1369 return (EIO); 1370 } 1371 } 1372 } 1373 if (m != NULL) { 1374 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1375 vm_page_dirty(m); 1376 vm_pager_page_unswapped(m); 1377 } 1378 } 1379 1380 /* 1381 * Release any swap space and free any whole pages. 1382 */ 1383 if (newpages < oldpages) { 1384 swap_pager_freespace(uobj, newpages, oldpages - 1385 newpages); 1386 vm_object_page_remove(uobj, newpages, 0, 0); 1387 } 1388 } 1389 uobj->size = newpages; 1390 VM_OBJECT_WUNLOCK(uobj); 1391 1392 TMPFS_LOCK(tmp); 1393 tmp->tm_pages_used += (newpages - oldpages); 1394 TMPFS_UNLOCK(tmp); 1395 1396 node->tn_size = newsize; 1397 return (0); 1398} 1399 1400/* 1401 * Change flags of the given vnode. 1402 * Caller should execute tmpfs_update on vp after a successful execution. 1403 * The vnode must be locked on entry and remain locked on exit. 1404 */ 1405int 1406tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1407 struct thread *p) 1408{ 1409 int error; 1410 struct tmpfs_node *node; 1411 1412 MPASS(VOP_ISLOCKED(vp)); 1413 1414 node = VP_TO_TMPFS_NODE(vp); 1415 1416 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1417 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1418 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1419 UF_SPARSE | UF_SYSTEM)) != 0) 1420 return (EOPNOTSUPP); 1421 1422 /* Disallow this operation if the file system is mounted read-only. */ 1423 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1424 return EROFS; 1425 1426 /* 1427 * Callers may only modify the file flags on objects they 1428 * have VADMIN rights for. 1429 */ 1430 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1431 return (error); 1432 /* 1433 * Unprivileged processes are not permitted to unset system 1434 * flags, or modify flags if any system flags are set. 1435 */ 1436 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1437 if (node->tn_flags & 1438 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1439 error = securelevel_gt(cred, 0); 1440 if (error) 1441 return (error); 1442 } 1443 } else { 1444 if (node->tn_flags & 1445 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1446 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1447 return (EPERM); 1448 } 1449 node->tn_flags = flags; 1450 node->tn_status |= TMPFS_NODE_CHANGED; 1451 1452 MPASS(VOP_ISLOCKED(vp)); 1453 1454 return 0; 1455} 1456 1457/* 1458 * Change access mode on the given vnode. 1459 * Caller should execute tmpfs_update on vp after a successful execution. 1460 * The vnode must be locked on entry and remain locked on exit. 1461 */ 1462int 1463tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1464{ 1465 int error; 1466 struct tmpfs_node *node; 1467 1468 MPASS(VOP_ISLOCKED(vp)); 1469 1470 node = VP_TO_TMPFS_NODE(vp); 1471 1472 /* Disallow this operation if the file system is mounted read-only. */ 1473 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1474 return EROFS; 1475 1476 /* Immutable or append-only files cannot be modified, either. */ 1477 if (node->tn_flags & (IMMUTABLE | APPEND)) 1478 return EPERM; 1479 1480 /* 1481 * To modify the permissions on a file, must possess VADMIN 1482 * for that file. 1483 */ 1484 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1485 return (error); 1486 1487 /* 1488 * Privileged processes may set the sticky bit on non-directories, 1489 * as well as set the setgid bit on a file with a group that the 1490 * process is not a member of. 1491 */ 1492 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1493 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1494 return (EFTYPE); 1495 } 1496 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1497 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1498 if (error) 1499 return (error); 1500 } 1501 1502 1503 node->tn_mode &= ~ALLPERMS; 1504 node->tn_mode |= mode & ALLPERMS; 1505 1506 node->tn_status |= TMPFS_NODE_CHANGED; 1507 1508 MPASS(VOP_ISLOCKED(vp)); 1509 1510 return 0; 1511} 1512 1513/* 1514 * Change ownership of the given vnode. At least one of uid or gid must 1515 * be different than VNOVAL. If one is set to that value, the attribute 1516 * is unchanged. 1517 * Caller should execute tmpfs_update on vp after a successful execution. 1518 * The vnode must be locked on entry and remain locked on exit. 1519 */ 1520int 1521tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1522 struct thread *p) 1523{ 1524 int error; 1525 struct tmpfs_node *node; 1526 uid_t ouid; 1527 gid_t ogid; 1528 1529 MPASS(VOP_ISLOCKED(vp)); 1530 1531 node = VP_TO_TMPFS_NODE(vp); 1532 1533 /* Assign default values if they are unknown. */ 1534 MPASS(uid != VNOVAL || gid != VNOVAL); 1535 if (uid == VNOVAL) 1536 uid = node->tn_uid; 1537 if (gid == VNOVAL) 1538 gid = node->tn_gid; 1539 MPASS(uid != VNOVAL && gid != VNOVAL); 1540 1541 /* Disallow this operation if the file system is mounted read-only. */ 1542 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1543 return EROFS; 1544 1545 /* Immutable or append-only files cannot be modified, either. */ 1546 if (node->tn_flags & (IMMUTABLE | APPEND)) 1547 return EPERM; 1548 1549 /* 1550 * To modify the ownership of a file, must possess VADMIN for that 1551 * file. 1552 */ 1553 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1554 return (error); 1555 1556 /* 1557 * To change the owner of a file, or change the group of a file to a 1558 * group of which we are not a member, the caller must have 1559 * privilege. 1560 */ 1561 if ((uid != node->tn_uid || 1562 (gid != node->tn_gid && !groupmember(gid, cred))) && 1563 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1564 return (error); 1565 1566 ogid = node->tn_gid; 1567 ouid = node->tn_uid; 1568 1569 node->tn_uid = uid; 1570 node->tn_gid = gid; 1571 1572 node->tn_status |= TMPFS_NODE_CHANGED; 1573 1574 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1575 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1576 node->tn_mode &= ~(S_ISUID | S_ISGID); 1577 } 1578 1579 MPASS(VOP_ISLOCKED(vp)); 1580 1581 return 0; 1582} 1583 1584/* 1585 * Change size of the given vnode. 1586 * Caller should execute tmpfs_update on vp after a successful execution. 1587 * The vnode must be locked on entry and remain locked on exit. 1588 */ 1589int 1590tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1591 struct thread *p) 1592{ 1593 int error; 1594 struct tmpfs_node *node; 1595 1596 MPASS(VOP_ISLOCKED(vp)); 1597 1598 node = VP_TO_TMPFS_NODE(vp); 1599 1600 /* Decide whether this is a valid operation based on the file type. */ 1601 error = 0; 1602 switch (vp->v_type) { 1603 case VDIR: 1604 return EISDIR; 1605 1606 case VREG: 1607 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1608 return EROFS; 1609 break; 1610 1611 case VBLK: 1612 /* FALLTHROUGH */ 1613 case VCHR: 1614 /* FALLTHROUGH */ 1615 case VFIFO: 1616 /* Allow modifications of special files even if in the file 1617 * system is mounted read-only (we are not modifying the 1618 * files themselves, but the objects they represent). */ 1619 return 0; 1620 1621 default: 1622 /* Anything else is unsupported. */ 1623 return EOPNOTSUPP; 1624 } 1625 1626 /* Immutable or append-only files cannot be modified, either. */ 1627 if (node->tn_flags & (IMMUTABLE | APPEND)) 1628 return EPERM; 1629 1630 error = tmpfs_truncate(vp, size); 1631 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1632 * for us, as will update tn_status; no need to do that here. */ 1633 1634 MPASS(VOP_ISLOCKED(vp)); 1635 1636 return error; 1637} 1638 1639/* 1640 * Change access and modification times of the given vnode. 1641 * Caller should execute tmpfs_update on vp after a successful execution. 1642 * The vnode must be locked on entry and remain locked on exit. 1643 */ 1644int 1645tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 1646 struct ucred *cred, struct thread *l) 1647{ 1648 int error; 1649 struct tmpfs_node *node; 1650 1651 MPASS(VOP_ISLOCKED(vp)); 1652 1653 node = VP_TO_TMPFS_NODE(vp); 1654 1655 /* Disallow this operation if the file system is mounted read-only. */ 1656 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1657 return EROFS; 1658 1659 /* Immutable or append-only files cannot be modified, either. */ 1660 if (node->tn_flags & (IMMUTABLE | APPEND)) 1661 return EPERM; 1662 1663 error = vn_utimes_perm(vp, vap, cred, l); 1664 if (error != 0) 1665 return (error); 1666 1667 if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL) 1668 node->tn_status |= TMPFS_NODE_ACCESSED; 1669 1670 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL) 1671 node->tn_status |= TMPFS_NODE_MODIFIED; 1672 1673 if (vap->va_birthtime.tv_nsec != VNOVAL && 1674 vap->va_birthtime.tv_nsec != VNOVAL) 1675 node->tn_status |= TMPFS_NODE_MODIFIED; 1676 1677 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1678 1679 if (vap->va_birthtime.tv_nsec != VNOVAL && 1680 vap->va_birthtime.tv_nsec != VNOVAL) 1681 node->tn_birthtime = vap->va_birthtime; 1682 MPASS(VOP_ISLOCKED(vp)); 1683 1684 return 0; 1685} 1686 1687/* Sync timestamps */ 1688void 1689tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1690 const struct timespec *mod) 1691{ 1692 struct tmpfs_node *node; 1693 struct timespec now; 1694 1695 node = VP_TO_TMPFS_NODE(vp); 1696 1697 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1698 TMPFS_NODE_CHANGED)) == 0) 1699 return; 1700 1701 vfs_timestamp(&now); 1702 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1703 if (acc == NULL) 1704 acc = &now; 1705 node->tn_atime = *acc; 1706 } 1707 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1708 if (mod == NULL) 1709 mod = &now; 1710 node->tn_mtime = *mod; 1711 } 1712 if (node->tn_status & TMPFS_NODE_CHANGED) { 1713 node->tn_ctime = now; 1714 } 1715 node->tn_status &= 1716 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1717} 1718 1719void 1720tmpfs_update(struct vnode *vp) 1721{ 1722 1723 tmpfs_itimes(vp, NULL, NULL); 1724} 1725 1726int 1727tmpfs_truncate(struct vnode *vp, off_t length) 1728{ 1729 int error; 1730 struct tmpfs_node *node; 1731 1732 node = VP_TO_TMPFS_NODE(vp); 1733 1734 if (length < 0) { 1735 error = EINVAL; 1736 goto out; 1737 } 1738 1739 if (node->tn_size == length) { 1740 error = 0; 1741 goto out; 1742 } 1743 1744 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1745 return (EFBIG); 1746 1747 error = tmpfs_reg_resize(vp, length, FALSE); 1748 if (error == 0) { 1749 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1750 } 1751 1752out: 1753 tmpfs_update(vp); 1754 1755 return error; 1756} 1757 1758static __inline int 1759tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1760{ 1761 if (a->td_hash > b->td_hash) 1762 return (1); 1763 else if (a->td_hash < b->td_hash) 1764 return (-1); 1765 return (0); 1766} 1767 1768RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1769