tmpfs_subr.c revision 269174
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 269174 2014-07-28 01:21:02Z 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 | OBJ_TMPFS_NODE); 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 /* lkflag is ignored, the lock is exclusive */ 552 (void) vn_lock(vp, lkflag | LK_RETRY); 553 554 vp->v_data = node; 555 vp->v_type = node->tn_type; 556 557 /* Type-specific initialization. */ 558 switch (node->tn_type) { 559 case VBLK: 560 /* FALLTHROUGH */ 561 case VCHR: 562 /* FALLTHROUGH */ 563 case VLNK: 564 /* FALLTHROUGH */ 565 case VSOCK: 566 break; 567 case VFIFO: 568 vp->v_op = &tmpfs_fifoop_entries; 569 break; 570 case VREG: 571 object = node->tn_reg.tn_aobj; 572 VM_OBJECT_WLOCK(object); 573 VI_LOCK(vp); 574 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 575 vp->v_object = object; 576 object->un_pager.swp.swp_tmpfs = vp; 577 vm_object_set_flag(object, OBJ_TMPFS); 578 VI_UNLOCK(vp); 579 VM_OBJECT_WUNLOCK(object); 580 break; 581 case VDIR: 582 MPASS(node->tn_dir.tn_parent != NULL); 583 if (node->tn_dir.tn_parent == node) 584 vp->v_vflag |= VV_ROOT; 585 break; 586 587 default: 588 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 589 } 590 if (vp->v_type != VFIFO) 591 VN_LOCK_ASHARE(vp); 592 593 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 594 if (error) 595 vp = NULL; 596 597unlock: 598 TMPFS_NODE_LOCK(node); 599 600 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 601 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 602 node->tn_vnode = vp; 603 604 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 605 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 606 TMPFS_NODE_UNLOCK(node); 607 wakeup((caddr_t) &node->tn_vpstate); 608 } else 609 TMPFS_NODE_UNLOCK(node); 610 611out: 612 *vpp = vp; 613 614#ifdef INVARIANTS 615 if (error == 0) { 616 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 617 TMPFS_NODE_LOCK(node); 618 MPASS(*vpp == node->tn_vnode); 619 TMPFS_NODE_UNLOCK(node); 620 } 621#endif 622 623 return error; 624} 625 626/* 627 * Destroys the association between the vnode vp and the node it 628 * references. 629 */ 630void 631tmpfs_free_vp(struct vnode *vp) 632{ 633 struct tmpfs_node *node; 634 635 node = VP_TO_TMPFS_NODE(vp); 636 637 TMPFS_NODE_ASSERT_LOCKED(node); 638 node->tn_vnode = NULL; 639 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 640 wakeup(&node->tn_vnode); 641 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 642 vp->v_data = NULL; 643} 644 645/* 646 * Allocates a new file of type 'type' and adds it to the parent directory 647 * 'dvp'; this addition is done using the component name given in 'cnp'. 648 * The ownership of the new file is automatically assigned based on the 649 * credentials of the caller (through 'cnp'), the group is set based on 650 * the parent directory and the mode is determined from the 'vap' argument. 651 * If successful, *vpp holds a vnode to the newly created file and zero 652 * is returned. Otherwise *vpp is NULL and the function returns an 653 * appropriate error code. 654 */ 655int 656tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 657 struct componentname *cnp, char *target) 658{ 659 int error; 660 struct tmpfs_dirent *de; 661 struct tmpfs_mount *tmp; 662 struct tmpfs_node *dnode; 663 struct tmpfs_node *node; 664 struct tmpfs_node *parent; 665 666 MPASS(VOP_ISLOCKED(dvp)); 667 MPASS(cnp->cn_flags & HASBUF); 668 669 tmp = VFS_TO_TMPFS(dvp->v_mount); 670 dnode = VP_TO_TMPFS_DIR(dvp); 671 *vpp = NULL; 672 673 /* If the entry we are creating is a directory, we cannot overflow 674 * the number of links of its parent, because it will get a new 675 * link. */ 676 if (vap->va_type == VDIR) { 677 /* Ensure that we do not overflow the maximum number of links 678 * imposed by the system. */ 679 MPASS(dnode->tn_links <= LINK_MAX); 680 if (dnode->tn_links == LINK_MAX) { 681 return (EMLINK); 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 return (error); 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 return (error); 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 return (error); 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 return (0); 718} 719 720static struct tmpfs_dirent * 721tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 722{ 723 struct tmpfs_dirent *de; 724 725 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 726 dc->tdc_tree = de; 727 if (de != NULL && tmpfs_dirent_duphead(de)) 728 de = LIST_FIRST(&de->ud.td_duphead); 729 dc->tdc_current = de; 730 731 return (dc->tdc_current); 732} 733 734static struct tmpfs_dirent * 735tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 736{ 737 struct tmpfs_dirent *de; 738 739 MPASS(dc->tdc_tree != NULL); 740 if (tmpfs_dirent_dup(dc->tdc_current)) { 741 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 742 if (dc->tdc_current != NULL) 743 return (dc->tdc_current); 744 } 745 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 746 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 747 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 748 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 749 MPASS(dc->tdc_current != NULL); 750 } 751 752 return (dc->tdc_current); 753} 754 755/* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 756static struct tmpfs_dirent * 757tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 758{ 759 struct tmpfs_dirent *de, dekey; 760 761 dekey.td_hash = hash; 762 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 763 return (de); 764} 765 766/* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 767static struct tmpfs_dirent * 768tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 769 struct tmpfs_dir_cursor *dc) 770{ 771 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 772 struct tmpfs_dirent *de, dekey; 773 774 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 775 776 if (cookie == node->tn_dir.tn_readdir_lastn && 777 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 778 /* Protect against possible race, tn_readdir_last[pn] 779 * may be updated with only shared vnode lock held. */ 780 if (cookie == tmpfs_dirent_cookie(de)) 781 goto out; 782 } 783 784 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 785 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 786 uh.td_dup.index_entries) { 787 MPASS(tmpfs_dirent_dup(de)); 788 if (de->td_cookie == cookie) 789 goto out; 790 /* dupindex list is sorted. */ 791 if (de->td_cookie < cookie) { 792 de = NULL; 793 goto out; 794 } 795 } 796 MPASS(de == NULL); 797 goto out; 798 } 799 800 MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie); 801 dekey.td_hash = cookie; 802 /* Recover if direntry for cookie was removed */ 803 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 804 dc->tdc_tree = de; 805 dc->tdc_current = de; 806 if (de != NULL && tmpfs_dirent_duphead(de)) { 807 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 808 MPASS(dc->tdc_current != NULL); 809 } 810 return (dc->tdc_current); 811 812out: 813 dc->tdc_tree = de; 814 dc->tdc_current = de; 815 if (de != NULL && tmpfs_dirent_dup(de)) 816 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 817 de->td_hash); 818 return (dc->tdc_current); 819} 820 821/* 822 * Looks for a directory entry in the directory represented by node. 823 * 'cnp' describes the name of the entry to look for. Note that the . 824 * and .. components are not allowed as they do not physically exist 825 * within directories. 826 * 827 * Returns a pointer to the entry when found, otherwise NULL. 828 */ 829struct tmpfs_dirent * 830tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 831 struct componentname *cnp) 832{ 833 struct tmpfs_dir_duphead *duphead; 834 struct tmpfs_dirent *de; 835 uint32_t hash; 836 837 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 838 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 839 cnp->cn_nameptr[1] == '.'))); 840 TMPFS_VALIDATE_DIR(node); 841 842 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 843 de = tmpfs_dir_xlookup_hash(node, hash); 844 if (de != NULL && tmpfs_dirent_duphead(de)) { 845 duphead = &de->ud.td_duphead; 846 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 847 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 848 cnp->cn_namelen)) 849 break; 850 } 851 } else if (de != NULL) { 852 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 853 cnp->cn_namelen)) 854 de = NULL; 855 } 856 if (de != NULL && f != NULL && de->td_node != f) 857 de = NULL; 858 859 return (de); 860} 861 862/* 863 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 864 * list, allocate new cookie value. 865 */ 866static void 867tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 868 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 869{ 870 struct tmpfs_dir_duphead *dupindex; 871 struct tmpfs_dirent *de, *pde; 872 873 dupindex = &dnode->tn_dir.tn_dupindex; 874 de = LIST_FIRST(dupindex); 875 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 876 if (de == NULL) 877 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 878 else 879 nde->td_cookie = de->td_cookie + 1; 880 MPASS(tmpfs_dirent_dup(nde)); 881 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 882 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 883 return; 884 } 885 886 /* 887 * Cookie numbers are near exhaustion. Scan dupindex list for unused 888 * numbers. dupindex list is sorted in descending order. Keep it so 889 * after inserting nde. 890 */ 891 while (1) { 892 pde = de; 893 de = LIST_NEXT(de, uh.td_dup.index_entries); 894 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 895 /* 896 * Last element of the index doesn't have minimal cookie 897 * value, use it. 898 */ 899 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 900 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 901 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 902 return; 903 } else if (de == NULL) { 904 /* 905 * We are so lucky have 2^30 hash duplicates in single 906 * directory :) Return largest possible cookie value. 907 * It should be fine except possible issues with 908 * VOP_READDIR restart. 909 */ 910 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 911 LIST_INSERT_HEAD(dupindex, nde, 912 uh.td_dup.index_entries); 913 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 914 return; 915 } 916 if (de->td_cookie + 1 == pde->td_cookie || 917 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 918 continue; /* No hole or invalid cookie. */ 919 nde->td_cookie = de->td_cookie + 1; 920 MPASS(tmpfs_dirent_dup(nde)); 921 MPASS(pde->td_cookie > nde->td_cookie); 922 MPASS(nde->td_cookie > de->td_cookie); 923 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 924 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 925 return; 926 }; 927} 928 929/* 930 * Attaches the directory entry de to the directory represented by vp. 931 * Note that this does not change the link count of the node pointed by 932 * the directory entry, as this is done by tmpfs_alloc_dirent. 933 */ 934void 935tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 936{ 937 struct tmpfs_node *dnode; 938 struct tmpfs_dirent *xde, *nde; 939 940 ASSERT_VOP_ELOCKED(vp, __func__); 941 MPASS(de->td_namelen > 0); 942 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 943 MPASS(de->td_cookie == de->td_hash); 944 945 dnode = VP_TO_TMPFS_DIR(vp); 946 dnode->tn_dir.tn_readdir_lastn = 0; 947 dnode->tn_dir.tn_readdir_lastp = NULL; 948 949 MPASS(!tmpfs_dirent_dup(de)); 950 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 951 if (xde != NULL && tmpfs_dirent_duphead(xde)) 952 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 953 else if (xde != NULL) { 954 /* 955 * Allocate new duphead. Swap xde with duphead to avoid 956 * adding/removing elements with the same hash. 957 */ 958 MPASS(!tmpfs_dirent_dup(xde)); 959 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 960 &nde); 961 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 962 memcpy(nde, xde, sizeof(*xde)); 963 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 964 LIST_INIT(&xde->ud.td_duphead); 965 xde->td_namelen = 0; 966 xde->td_node = NULL; 967 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 968 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 969 } 970 dnode->tn_size += sizeof(struct tmpfs_dirent); 971 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 972 TMPFS_NODE_MODIFIED; 973} 974 975/* 976 * Detaches the directory entry de from the directory represented by vp. 977 * Note that this does not change the link count of the node pointed by 978 * the directory entry, as this is done by tmpfs_free_dirent. 979 */ 980void 981tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 982{ 983 struct tmpfs_mount *tmp; 984 struct tmpfs_dir *head; 985 struct tmpfs_node *dnode; 986 struct tmpfs_dirent *xde; 987 988 ASSERT_VOP_ELOCKED(vp, __func__); 989 990 dnode = VP_TO_TMPFS_DIR(vp); 991 head = &dnode->tn_dir.tn_dirhead; 992 dnode->tn_dir.tn_readdir_lastn = 0; 993 dnode->tn_dir.tn_readdir_lastp = NULL; 994 995 if (tmpfs_dirent_dup(de)) { 996 /* Remove duphead if de was last entry. */ 997 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 998 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 999 MPASS(tmpfs_dirent_duphead(xde)); 1000 } else 1001 xde = NULL; 1002 LIST_REMOVE(de, uh.td_dup.entries); 1003 LIST_REMOVE(de, uh.td_dup.index_entries); 1004 if (xde != NULL) { 1005 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1006 RB_REMOVE(tmpfs_dir, head, xde); 1007 tmp = VFS_TO_TMPFS(vp->v_mount); 1008 MPASS(xde->td_node == NULL); 1009 tmpfs_free_dirent(tmp, xde); 1010 } 1011 } 1012 } else 1013 RB_REMOVE(tmpfs_dir, head, de); 1014 1015 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1016 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1017 TMPFS_NODE_MODIFIED; 1018} 1019 1020void 1021tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1022{ 1023 struct tmpfs_dirent *de, *dde, *nde; 1024 1025 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1026 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1027 /* Node may already be destroyed. */ 1028 de->td_node = NULL; 1029 if (tmpfs_dirent_duphead(de)) { 1030 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1031 LIST_REMOVE(dde, uh.td_dup.entries); 1032 dde->td_node = NULL; 1033 tmpfs_free_dirent(tmp, dde); 1034 } 1035 } 1036 tmpfs_free_dirent(tmp, de); 1037 } 1038} 1039 1040/* 1041 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1042 * directory and returns it in the uio space. The function returns 0 1043 * on success, -1 if there was not enough space in the uio structure to 1044 * hold the directory entry or an appropriate error code if another 1045 * error happens. 1046 */ 1047static int 1048tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1049{ 1050 int error; 1051 struct dirent dent; 1052 1053 TMPFS_VALIDATE_DIR(node); 1054 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1055 1056 dent.d_fileno = node->tn_id; 1057 dent.d_type = DT_DIR; 1058 dent.d_namlen = 1; 1059 dent.d_name[0] = '.'; 1060 dent.d_name[1] = '\0'; 1061 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1062 1063 if (dent.d_reclen > uio->uio_resid) 1064 error = EJUSTRETURN; 1065 else 1066 error = uiomove(&dent, dent.d_reclen, uio); 1067 1068 node->tn_status |= TMPFS_NODE_ACCESSED; 1069 1070 return error; 1071} 1072 1073/* 1074 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1075 * directory and returns it in the uio space. The function returns 0 1076 * on success, -1 if there was not enough space in the uio structure to 1077 * hold the directory entry or an appropriate error code if another 1078 * error happens. 1079 */ 1080static int 1081tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1082{ 1083 int error; 1084 struct dirent dent; 1085 1086 TMPFS_VALIDATE_DIR(node); 1087 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1088 1089 /* 1090 * Return ENOENT if the current node is already removed. 1091 */ 1092 TMPFS_ASSERT_LOCKED(node); 1093 if (node->tn_dir.tn_parent == NULL) { 1094 return (ENOENT); 1095 } 1096 1097 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1098 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1099 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1100 1101 dent.d_type = DT_DIR; 1102 dent.d_namlen = 2; 1103 dent.d_name[0] = '.'; 1104 dent.d_name[1] = '.'; 1105 dent.d_name[2] = '\0'; 1106 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1107 1108 if (dent.d_reclen > uio->uio_resid) 1109 error = EJUSTRETURN; 1110 else 1111 error = uiomove(&dent, dent.d_reclen, uio); 1112 1113 node->tn_status |= TMPFS_NODE_ACCESSED; 1114 1115 return error; 1116} 1117 1118/* 1119 * Helper function for tmpfs_readdir. Returns as much directory entries 1120 * as can fit in the uio space. The read starts at uio->uio_offset. 1121 * The function returns 0 on success, -1 if there was not enough space 1122 * in the uio structure to hold the directory entry or an appropriate 1123 * error code if another error happens. 1124 */ 1125int 1126tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1127 u_long *cookies, int *ncookies) 1128{ 1129 struct tmpfs_dir_cursor dc; 1130 struct tmpfs_dirent *de; 1131 off_t off; 1132 int error; 1133 1134 TMPFS_VALIDATE_DIR(node); 1135 1136 off = 0; 1137 1138 /* 1139 * Lookup the node from the current offset. The starting offset of 1140 * 0 will lookup both '.' and '..', and then the first real entry, 1141 * or EOF if there are none. Then find all entries for the dir that 1142 * fit into the buffer. Once no more entries are found (de == NULL), 1143 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1144 * call to return 0. 1145 */ 1146 switch (uio->uio_offset) { 1147 case TMPFS_DIRCOOKIE_DOT: 1148 error = tmpfs_dir_getdotdent(node, uio); 1149 if (error != 0) 1150 return (error); 1151 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1152 if (cookies != NULL) 1153 cookies[(*ncookies)++] = off = uio->uio_offset; 1154 /* FALLTHROUGH */ 1155 case TMPFS_DIRCOOKIE_DOTDOT: 1156 error = tmpfs_dir_getdotdotdent(node, uio); 1157 if (error != 0) 1158 return (error); 1159 de = tmpfs_dir_first(node, &dc); 1160 uio->uio_offset = tmpfs_dirent_cookie(de); 1161 if (cookies != NULL) 1162 cookies[(*ncookies)++] = off = uio->uio_offset; 1163 /* EOF. */ 1164 if (de == NULL) 1165 return (0); 1166 break; 1167 case TMPFS_DIRCOOKIE_EOF: 1168 return (0); 1169 default: 1170 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1171 if (de == NULL) 1172 return (EINVAL); 1173 if (cookies != NULL) 1174 off = tmpfs_dirent_cookie(de); 1175 } 1176 1177 /* Read as much entries as possible; i.e., until we reach the end of 1178 * the directory or we exhaust uio space. */ 1179 do { 1180 struct dirent d; 1181 1182 /* Create a dirent structure representing the current 1183 * tmpfs_node and fill it. */ 1184 if (de->td_node == NULL) { 1185 d.d_fileno = 1; 1186 d.d_type = DT_WHT; 1187 } else { 1188 d.d_fileno = de->td_node->tn_id; 1189 switch (de->td_node->tn_type) { 1190 case VBLK: 1191 d.d_type = DT_BLK; 1192 break; 1193 1194 case VCHR: 1195 d.d_type = DT_CHR; 1196 break; 1197 1198 case VDIR: 1199 d.d_type = DT_DIR; 1200 break; 1201 1202 case VFIFO: 1203 d.d_type = DT_FIFO; 1204 break; 1205 1206 case VLNK: 1207 d.d_type = DT_LNK; 1208 break; 1209 1210 case VREG: 1211 d.d_type = DT_REG; 1212 break; 1213 1214 case VSOCK: 1215 d.d_type = DT_SOCK; 1216 break; 1217 1218 default: 1219 panic("tmpfs_dir_getdents: type %p %d", 1220 de->td_node, (int)de->td_node->tn_type); 1221 } 1222 } 1223 d.d_namlen = de->td_namelen; 1224 MPASS(de->td_namelen < sizeof(d.d_name)); 1225 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1226 d.d_name[de->td_namelen] = '\0'; 1227 d.d_reclen = GENERIC_DIRSIZ(&d); 1228 1229 /* Stop reading if the directory entry we are treating is 1230 * bigger than the amount of data that can be returned. */ 1231 if (d.d_reclen > uio->uio_resid) { 1232 error = EJUSTRETURN; 1233 break; 1234 } 1235 1236 /* Copy the new dirent structure into the output buffer and 1237 * advance pointers. */ 1238 error = uiomove(&d, d.d_reclen, uio); 1239 if (error == 0) { 1240 de = tmpfs_dir_next(node, &dc); 1241 if (cookies != NULL) { 1242 off = tmpfs_dirent_cookie(de); 1243 MPASS(*ncookies < maxcookies); 1244 cookies[(*ncookies)++] = off; 1245 } 1246 } 1247 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1248 1249 /* Skip setting off when using cookies as it is already done above. */ 1250 if (cookies == NULL) 1251 off = tmpfs_dirent_cookie(de); 1252 1253 /* Update the offset and cache. */ 1254 uio->uio_offset = off; 1255 node->tn_dir.tn_readdir_lastn = off; 1256 node->tn_dir.tn_readdir_lastp = de; 1257 1258 node->tn_status |= TMPFS_NODE_ACCESSED; 1259 return error; 1260} 1261 1262int 1263tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1264{ 1265 struct tmpfs_dirent *de; 1266 int error; 1267 1268 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1269 cnp->cn_nameptr, cnp->cn_namelen, &de); 1270 if (error != 0) 1271 return (error); 1272 tmpfs_dir_attach(dvp, de); 1273 return (0); 1274} 1275 1276void 1277tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1278{ 1279 struct tmpfs_dirent *de; 1280 1281 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1282 MPASS(de != NULL && de->td_node == NULL); 1283 tmpfs_dir_detach(dvp, de); 1284 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1285} 1286 1287/* 1288 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1289 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1290 * 'newsize' must be positive. 1291 * 1292 * Returns zero on success or an appropriate error code on failure. 1293 */ 1294int 1295tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1296{ 1297 struct tmpfs_mount *tmp; 1298 struct tmpfs_node *node; 1299 vm_object_t uobj; 1300 vm_page_t m, ma[1]; 1301 vm_pindex_t idx, newpages, oldpages; 1302 off_t oldsize; 1303 int base, rv; 1304 1305 MPASS(vp->v_type == VREG); 1306 MPASS(newsize >= 0); 1307 1308 node = VP_TO_TMPFS_NODE(vp); 1309 uobj = node->tn_reg.tn_aobj; 1310 tmp = VFS_TO_TMPFS(vp->v_mount); 1311 1312 /* 1313 * Convert the old and new sizes to the number of pages needed to 1314 * store them. It may happen that we do not need to do anything 1315 * because the last allocated page can accommodate the change on 1316 * its own. 1317 */ 1318 oldsize = node->tn_size; 1319 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1320 MPASS(oldpages == uobj->size); 1321 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1322 if (newpages > oldpages && 1323 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1324 return (ENOSPC); 1325 1326 VM_OBJECT_WLOCK(uobj); 1327 if (newsize < oldsize) { 1328 /* 1329 * Zero the truncated part of the last page. 1330 */ 1331 base = newsize & PAGE_MASK; 1332 if (base != 0) { 1333 idx = OFF_TO_IDX(newsize); 1334retry: 1335 m = vm_page_lookup(uobj, idx); 1336 if (m != NULL) { 1337 if (vm_page_sleep_if_busy(m, "tmfssz")) 1338 goto retry; 1339 MPASS(m->valid == VM_PAGE_BITS_ALL); 1340 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1341 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1342 if (m == NULL) { 1343 VM_OBJECT_WUNLOCK(uobj); 1344 VM_WAIT; 1345 VM_OBJECT_WLOCK(uobj); 1346 goto retry; 1347 } else if (m->valid != VM_PAGE_BITS_ALL) { 1348 ma[0] = m; 1349 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1350 m = vm_page_lookup(uobj, idx); 1351 } else 1352 /* A cached page was reactivated. */ 1353 rv = VM_PAGER_OK; 1354 vm_page_lock(m); 1355 if (rv == VM_PAGER_OK) { 1356 vm_page_deactivate(m); 1357 vm_page_unlock(m); 1358 vm_page_xunbusy(m); 1359 } else { 1360 vm_page_free(m); 1361 vm_page_unlock(m); 1362 if (ignerr) 1363 m = NULL; 1364 else { 1365 VM_OBJECT_WUNLOCK(uobj); 1366 return (EIO); 1367 } 1368 } 1369 } 1370 if (m != NULL) { 1371 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1372 vm_page_dirty(m); 1373 vm_pager_page_unswapped(m); 1374 } 1375 } 1376 1377 /* 1378 * Release any swap space and free any whole pages. 1379 */ 1380 if (newpages < oldpages) { 1381 swap_pager_freespace(uobj, newpages, oldpages - 1382 newpages); 1383 vm_object_page_remove(uobj, newpages, 0, 0); 1384 } 1385 } 1386 uobj->size = newpages; 1387 VM_OBJECT_WUNLOCK(uobj); 1388 1389 TMPFS_LOCK(tmp); 1390 tmp->tm_pages_used += (newpages - oldpages); 1391 TMPFS_UNLOCK(tmp); 1392 1393 node->tn_size = newsize; 1394 return (0); 1395} 1396 1397/* 1398 * Change flags of the given vnode. 1399 * Caller should execute tmpfs_update on vp after a successful execution. 1400 * The vnode must be locked on entry and remain locked on exit. 1401 */ 1402int 1403tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1404 struct thread *p) 1405{ 1406 int error; 1407 struct tmpfs_node *node; 1408 1409 MPASS(VOP_ISLOCKED(vp)); 1410 1411 node = VP_TO_TMPFS_NODE(vp); 1412 1413 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1414 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1415 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1416 UF_SPARSE | UF_SYSTEM)) != 0) 1417 return (EOPNOTSUPP); 1418 1419 /* Disallow this operation if the file system is mounted read-only. */ 1420 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1421 return EROFS; 1422 1423 /* 1424 * Callers may only modify the file flags on objects they 1425 * have VADMIN rights for. 1426 */ 1427 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1428 return (error); 1429 /* 1430 * Unprivileged processes are not permitted to unset system 1431 * flags, or modify flags if any system flags are set. 1432 */ 1433 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1434 if (node->tn_flags & 1435 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1436 error = securelevel_gt(cred, 0); 1437 if (error) 1438 return (error); 1439 } 1440 } else { 1441 if (node->tn_flags & 1442 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1443 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1444 return (EPERM); 1445 } 1446 node->tn_flags = flags; 1447 node->tn_status |= TMPFS_NODE_CHANGED; 1448 1449 MPASS(VOP_ISLOCKED(vp)); 1450 1451 return 0; 1452} 1453 1454/* 1455 * Change access mode on the given vnode. 1456 * Caller should execute tmpfs_update on vp after a successful execution. 1457 * The vnode must be locked on entry and remain locked on exit. 1458 */ 1459int 1460tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1461{ 1462 int error; 1463 struct tmpfs_node *node; 1464 1465 MPASS(VOP_ISLOCKED(vp)); 1466 1467 node = VP_TO_TMPFS_NODE(vp); 1468 1469 /* Disallow this operation if the file system is mounted read-only. */ 1470 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1471 return EROFS; 1472 1473 /* Immutable or append-only files cannot be modified, either. */ 1474 if (node->tn_flags & (IMMUTABLE | APPEND)) 1475 return EPERM; 1476 1477 /* 1478 * To modify the permissions on a file, must possess VADMIN 1479 * for that file. 1480 */ 1481 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1482 return (error); 1483 1484 /* 1485 * Privileged processes may set the sticky bit on non-directories, 1486 * as well as set the setgid bit on a file with a group that the 1487 * process is not a member of. 1488 */ 1489 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1490 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1491 return (EFTYPE); 1492 } 1493 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1494 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1495 if (error) 1496 return (error); 1497 } 1498 1499 1500 node->tn_mode &= ~ALLPERMS; 1501 node->tn_mode |= mode & ALLPERMS; 1502 1503 node->tn_status |= TMPFS_NODE_CHANGED; 1504 1505 MPASS(VOP_ISLOCKED(vp)); 1506 1507 return 0; 1508} 1509 1510/* 1511 * Change ownership of the given vnode. At least one of uid or gid must 1512 * be different than VNOVAL. If one is set to that value, the attribute 1513 * is unchanged. 1514 * Caller should execute tmpfs_update on vp after a successful execution. 1515 * The vnode must be locked on entry and remain locked on exit. 1516 */ 1517int 1518tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1519 struct thread *p) 1520{ 1521 int error; 1522 struct tmpfs_node *node; 1523 uid_t ouid; 1524 gid_t ogid; 1525 1526 MPASS(VOP_ISLOCKED(vp)); 1527 1528 node = VP_TO_TMPFS_NODE(vp); 1529 1530 /* Assign default values if they are unknown. */ 1531 MPASS(uid != VNOVAL || gid != VNOVAL); 1532 if (uid == VNOVAL) 1533 uid = node->tn_uid; 1534 if (gid == VNOVAL) 1535 gid = node->tn_gid; 1536 MPASS(uid != VNOVAL && gid != VNOVAL); 1537 1538 /* Disallow this operation if the file system is mounted read-only. */ 1539 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1540 return EROFS; 1541 1542 /* Immutable or append-only files cannot be modified, either. */ 1543 if (node->tn_flags & (IMMUTABLE | APPEND)) 1544 return EPERM; 1545 1546 /* 1547 * To modify the ownership of a file, must possess VADMIN for that 1548 * file. 1549 */ 1550 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1551 return (error); 1552 1553 /* 1554 * To change the owner of a file, or change the group of a file to a 1555 * group of which we are not a member, the caller must have 1556 * privilege. 1557 */ 1558 if ((uid != node->tn_uid || 1559 (gid != node->tn_gid && !groupmember(gid, cred))) && 1560 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1561 return (error); 1562 1563 ogid = node->tn_gid; 1564 ouid = node->tn_uid; 1565 1566 node->tn_uid = uid; 1567 node->tn_gid = gid; 1568 1569 node->tn_status |= TMPFS_NODE_CHANGED; 1570 1571 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1572 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1573 node->tn_mode &= ~(S_ISUID | S_ISGID); 1574 } 1575 1576 MPASS(VOP_ISLOCKED(vp)); 1577 1578 return 0; 1579} 1580 1581/* 1582 * Change size of the given vnode. 1583 * Caller should execute tmpfs_update on vp after a successful execution. 1584 * The vnode must be locked on entry and remain locked on exit. 1585 */ 1586int 1587tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1588 struct thread *p) 1589{ 1590 int error; 1591 struct tmpfs_node *node; 1592 1593 MPASS(VOP_ISLOCKED(vp)); 1594 1595 node = VP_TO_TMPFS_NODE(vp); 1596 1597 /* Decide whether this is a valid operation based on the file type. */ 1598 error = 0; 1599 switch (vp->v_type) { 1600 case VDIR: 1601 return EISDIR; 1602 1603 case VREG: 1604 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1605 return EROFS; 1606 break; 1607 1608 case VBLK: 1609 /* FALLTHROUGH */ 1610 case VCHR: 1611 /* FALLTHROUGH */ 1612 case VFIFO: 1613 /* Allow modifications of special files even if in the file 1614 * system is mounted read-only (we are not modifying the 1615 * files themselves, but the objects they represent). */ 1616 return 0; 1617 1618 default: 1619 /* Anything else is unsupported. */ 1620 return EOPNOTSUPP; 1621 } 1622 1623 /* Immutable or append-only files cannot be modified, either. */ 1624 if (node->tn_flags & (IMMUTABLE | APPEND)) 1625 return EPERM; 1626 1627 error = tmpfs_truncate(vp, size); 1628 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1629 * for us, as will update tn_status; no need to do that here. */ 1630 1631 MPASS(VOP_ISLOCKED(vp)); 1632 1633 return error; 1634} 1635 1636/* 1637 * Change access and modification times of the given vnode. 1638 * Caller should execute tmpfs_update on vp after a successful execution. 1639 * The vnode must be locked on entry and remain locked on exit. 1640 */ 1641int 1642tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 1643 struct ucred *cred, struct thread *l) 1644{ 1645 int error; 1646 struct tmpfs_node *node; 1647 1648 MPASS(VOP_ISLOCKED(vp)); 1649 1650 node = VP_TO_TMPFS_NODE(vp); 1651 1652 /* Disallow this operation if the file system is mounted read-only. */ 1653 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1654 return EROFS; 1655 1656 /* Immutable or append-only files cannot be modified, either. */ 1657 if (node->tn_flags & (IMMUTABLE | APPEND)) 1658 return EPERM; 1659 1660 error = vn_utimes_perm(vp, vap, cred, l); 1661 if (error != 0) 1662 return (error); 1663 1664 if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL) 1665 node->tn_status |= TMPFS_NODE_ACCESSED; 1666 1667 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL) 1668 node->tn_status |= TMPFS_NODE_MODIFIED; 1669 1670 if (vap->va_birthtime.tv_nsec != VNOVAL && 1671 vap->va_birthtime.tv_nsec != VNOVAL) 1672 node->tn_status |= TMPFS_NODE_MODIFIED; 1673 1674 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1675 1676 if (vap->va_birthtime.tv_nsec != VNOVAL && 1677 vap->va_birthtime.tv_nsec != VNOVAL) 1678 node->tn_birthtime = vap->va_birthtime; 1679 MPASS(VOP_ISLOCKED(vp)); 1680 1681 return 0; 1682} 1683 1684/* Sync timestamps */ 1685void 1686tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1687 const struct timespec *mod) 1688{ 1689 struct tmpfs_node *node; 1690 struct timespec now; 1691 1692 node = VP_TO_TMPFS_NODE(vp); 1693 1694 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1695 TMPFS_NODE_CHANGED)) == 0) 1696 return; 1697 1698 vfs_timestamp(&now); 1699 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1700 if (acc == NULL) 1701 acc = &now; 1702 node->tn_atime = *acc; 1703 } 1704 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1705 if (mod == NULL) 1706 mod = &now; 1707 node->tn_mtime = *mod; 1708 } 1709 if (node->tn_status & TMPFS_NODE_CHANGED) { 1710 node->tn_ctime = now; 1711 } 1712 node->tn_status &= 1713 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1714} 1715 1716void 1717tmpfs_update(struct vnode *vp) 1718{ 1719 1720 tmpfs_itimes(vp, NULL, NULL); 1721} 1722 1723int 1724tmpfs_truncate(struct vnode *vp, off_t length) 1725{ 1726 int error; 1727 struct tmpfs_node *node; 1728 1729 node = VP_TO_TMPFS_NODE(vp); 1730 1731 if (length < 0) { 1732 error = EINVAL; 1733 goto out; 1734 } 1735 1736 if (node->tn_size == length) { 1737 error = 0; 1738 goto out; 1739 } 1740 1741 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1742 return (EFBIG); 1743 1744 error = tmpfs_reg_resize(vp, length, FALSE); 1745 if (error == 0) { 1746 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1747 } 1748 1749out: 1750 tmpfs_update(vp); 1751 1752 return error; 1753} 1754 1755static __inline int 1756tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1757{ 1758 if (a->td_hash > b->td_hash) 1759 return (1); 1760 else if (a->td_hash < b->td_hash) 1761 return (-1); 1762 return (0); 1763} 1764 1765RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1766