tmpfs_subr.c revision 269164
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 269164 2014-07-28 00:43:42Z 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 if (vp->v_type != VREG || obj == NULL) 432 return; 433 434 VM_OBJECT_WLOCK(obj); 435 VI_LOCK(vp); 436 vm_object_clear_flag(obj, OBJ_TMPFS); 437 obj->un_pager.swp.swp_tmpfs = NULL; 438 VI_UNLOCK(vp); 439 VM_OBJECT_WUNLOCK(obj); 440} 441 442/* 443 * Need to clear v_object for insmntque failure. 444 */ 445static void 446tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg) 447{ 448 449 tmpfs_destroy_vobject(vp, vp->v_object); 450 vp->v_object = NULL; 451 vp->v_data = NULL; 452 vp->v_op = &dead_vnodeops; 453 vgone(vp); 454 vput(vp); 455} 456 457/* 458 * Allocates a new vnode for the node node or returns a new reference to 459 * an existing one if the node had already a vnode referencing it. The 460 * resulting locked vnode is returned in *vpp. 461 * 462 * Returns zero on success or an appropriate error code on failure. 463 */ 464int 465tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 466 struct vnode **vpp) 467{ 468 struct vnode *vp; 469 vm_object_t object; 470 int error; 471 472 error = 0; 473loop: 474 TMPFS_NODE_LOCK(node); 475loop1: 476 if ((vp = node->tn_vnode) != NULL) { 477 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 478 VI_LOCK(vp); 479 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) || 480 ((vp->v_iflag & VI_DOOMED) != 0 && 481 (lkflag & LK_NOWAIT) != 0)) { 482 VI_UNLOCK(vp); 483 TMPFS_NODE_UNLOCK(node); 484 error = ENOENT; 485 vp = NULL; 486 goto out; 487 } 488 if ((vp->v_iflag & VI_DOOMED) != 0) { 489 VI_UNLOCK(vp); 490 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM; 491 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) { 492 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 493 0, "tmpfsE", 0); 494 } 495 goto loop1; 496 } 497 TMPFS_NODE_UNLOCK(node); 498 error = vget(vp, lkflag | LK_INTERLOCK, curthread); 499 if (error == ENOENT) 500 goto loop; 501 if (error != 0) { 502 vp = NULL; 503 goto out; 504 } 505 506 /* 507 * Make sure the vnode is still there after 508 * getting the interlock to avoid racing a free. 509 */ 510 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 511 vput(vp); 512 goto loop; 513 } 514 515 goto out; 516 } 517 518 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 519 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 520 TMPFS_NODE_UNLOCK(node); 521 error = ENOENT; 522 vp = NULL; 523 goto out; 524 } 525 526 /* 527 * otherwise lock the vp list while we call getnewvnode 528 * since that can block. 529 */ 530 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 531 node->tn_vpstate |= TMPFS_VNODE_WANT; 532 error = msleep((caddr_t) &node->tn_vpstate, 533 TMPFS_NODE_MTX(node), PDROP | PCATCH, 534 "tmpfs_alloc_vp", 0); 535 if (error) 536 return error; 537 538 goto loop; 539 } else 540 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 541 542 TMPFS_NODE_UNLOCK(node); 543 544 /* Get a new vnode and associate it with our node. */ 545 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); 546 if (error != 0) 547 goto unlock; 548 MPASS(vp != NULL); 549 550 (void) vn_lock(vp, lkflag | LK_RETRY); 551 552 vp->v_data = node; 553 vp->v_type = node->tn_type; 554 555 /* Type-specific initialization. */ 556 switch (node->tn_type) { 557 case VBLK: 558 /* FALLTHROUGH */ 559 case VCHR: 560 /* FALLTHROUGH */ 561 case VLNK: 562 /* FALLTHROUGH */ 563 case VSOCK: 564 break; 565 case VFIFO: 566 vp->v_op = &tmpfs_fifoop_entries; 567 break; 568 case VREG: 569 object = node->tn_reg.tn_aobj; 570 VM_OBJECT_WLOCK(object); 571 VI_LOCK(vp); 572 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 573 vp->v_object = object; 574 object->un_pager.swp.swp_tmpfs = vp; 575 vm_object_set_flag(object, OBJ_TMPFS); 576 VI_UNLOCK(vp); 577 VM_OBJECT_WUNLOCK(object); 578 break; 579 case VDIR: 580 MPASS(node->tn_dir.tn_parent != NULL); 581 if (node->tn_dir.tn_parent == node) 582 vp->v_vflag |= VV_ROOT; 583 break; 584 585 default: 586 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 587 } 588 if (vp->v_type != VFIFO) 589 VN_LOCK_ASHARE(vp); 590 591 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 592 if (error) 593 vp = NULL; 594 595unlock: 596 TMPFS_NODE_LOCK(node); 597 598 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 599 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 600 node->tn_vnode = vp; 601 602 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 603 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 604 TMPFS_NODE_UNLOCK(node); 605 wakeup((caddr_t) &node->tn_vpstate); 606 } else 607 TMPFS_NODE_UNLOCK(node); 608 609out: 610 *vpp = vp; 611 612#ifdef INVARIANTS 613 if (error == 0) { 614 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 615 TMPFS_NODE_LOCK(node); 616 MPASS(*vpp == node->tn_vnode); 617 TMPFS_NODE_UNLOCK(node); 618 } 619#endif 620 621 return error; 622} 623 624/* 625 * Destroys the association between the vnode vp and the node it 626 * references. 627 */ 628void 629tmpfs_free_vp(struct vnode *vp) 630{ 631 struct tmpfs_node *node; 632 633 node = VP_TO_TMPFS_NODE(vp); 634 635 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED); 636 node->tn_vnode = NULL; 637 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 638 wakeup(&node->tn_vnode); 639 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 640 vp->v_data = NULL; 641} 642 643/* 644 * Allocates a new file of type 'type' and adds it to the parent directory 645 * 'dvp'; this addition is done using the component name given in 'cnp'. 646 * The ownership of the new file is automatically assigned based on the 647 * credentials of the caller (through 'cnp'), the group is set based on 648 * the parent directory and the mode is determined from the 'vap' argument. 649 * If successful, *vpp holds a vnode to the newly created file and zero 650 * is returned. Otherwise *vpp is NULL and the function returns an 651 * appropriate error code. 652 */ 653int 654tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 655 struct componentname *cnp, char *target) 656{ 657 int error; 658 struct tmpfs_dirent *de; 659 struct tmpfs_mount *tmp; 660 struct tmpfs_node *dnode; 661 struct tmpfs_node *node; 662 struct tmpfs_node *parent; 663 664 MPASS(VOP_ISLOCKED(dvp)); 665 MPASS(cnp->cn_flags & HASBUF); 666 667 tmp = VFS_TO_TMPFS(dvp->v_mount); 668 dnode = VP_TO_TMPFS_DIR(dvp); 669 *vpp = NULL; 670 671 /* If the entry we are creating is a directory, we cannot overflow 672 * the number of links of its parent, because it will get a new 673 * link. */ 674 if (vap->va_type == VDIR) { 675 /* Ensure that we do not overflow the maximum number of links 676 * imposed by the system. */ 677 MPASS(dnode->tn_links <= LINK_MAX); 678 if (dnode->tn_links == LINK_MAX) { 679 error = EMLINK; 680 goto out; 681 } 682 683 parent = dnode; 684 MPASS(parent != NULL); 685 } else 686 parent = NULL; 687 688 /* Allocate a node that represents the new file. */ 689 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid, 690 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node); 691 if (error != 0) 692 goto out; 693 694 /* Allocate a directory entry that points to the new file. */ 695 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 696 &de); 697 if (error != 0) { 698 tmpfs_free_node(tmp, node); 699 goto out; 700 } 701 702 /* Allocate a vnode for the new file. */ 703 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 704 if (error != 0) { 705 tmpfs_free_dirent(tmp, de); 706 tmpfs_free_node(tmp, node); 707 goto out; 708 } 709 710 /* Now that all required items are allocated, we can proceed to 711 * insert the new node into the directory, an operation that 712 * cannot fail. */ 713 if (cnp->cn_flags & ISWHITEOUT) 714 tmpfs_dir_whiteout_remove(dvp, cnp); 715 tmpfs_dir_attach(dvp, de); 716 717out: 718 719 return error; 720} 721 722static struct tmpfs_dirent * 723tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 724{ 725 struct tmpfs_dirent *de; 726 727 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 728 dc->tdc_tree = de; 729 if (de != NULL && tmpfs_dirent_duphead(de)) 730 de = LIST_FIRST(&de->ud.td_duphead); 731 dc->tdc_current = de; 732 733 return (dc->tdc_current); 734} 735 736static struct tmpfs_dirent * 737tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 738{ 739 struct tmpfs_dirent *de; 740 741 MPASS(dc->tdc_tree != NULL); 742 if (tmpfs_dirent_dup(dc->tdc_current)) { 743 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 744 if (dc->tdc_current != NULL) 745 return (dc->tdc_current); 746 } 747 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 748 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 749 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 750 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 751 MPASS(dc->tdc_current != NULL); 752 } 753 754 return (dc->tdc_current); 755} 756 757/* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 758static struct tmpfs_dirent * 759tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 760{ 761 struct tmpfs_dirent *de, dekey; 762 763 dekey.td_hash = hash; 764 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 765 return (de); 766} 767 768/* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 769static struct tmpfs_dirent * 770tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 771 struct tmpfs_dir_cursor *dc) 772{ 773 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 774 struct tmpfs_dirent *de, dekey; 775 776 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 777 778 if (cookie == node->tn_dir.tn_readdir_lastn && 779 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 780 /* Protect against possible race, tn_readdir_last[pn] 781 * may be updated with only shared vnode lock held. */ 782 if (cookie == tmpfs_dirent_cookie(de)) 783 goto out; 784 } 785 786 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 787 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 788 uh.td_dup.index_entries) { 789 MPASS(tmpfs_dirent_dup(de)); 790 if (de->td_cookie == cookie) 791 goto out; 792 /* dupindex list is sorted. */ 793 if (de->td_cookie < cookie) { 794 de = NULL; 795 goto out; 796 } 797 } 798 MPASS(de == NULL); 799 goto out; 800 } 801 802 MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie); 803 dekey.td_hash = cookie; 804 /* Recover if direntry for cookie was removed */ 805 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 806 dc->tdc_tree = de; 807 dc->tdc_current = de; 808 if (de != NULL && tmpfs_dirent_duphead(de)) { 809 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 810 MPASS(dc->tdc_current != NULL); 811 } 812 return (dc->tdc_current); 813 814out: 815 dc->tdc_tree = de; 816 dc->tdc_current = de; 817 if (de != NULL && tmpfs_dirent_dup(de)) 818 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 819 de->td_hash); 820 return (dc->tdc_current); 821} 822 823/* 824 * Looks for a directory entry in the directory represented by node. 825 * 'cnp' describes the name of the entry to look for. Note that the . 826 * and .. components are not allowed as they do not physically exist 827 * within directories. 828 * 829 * Returns a pointer to the entry when found, otherwise NULL. 830 */ 831struct tmpfs_dirent * 832tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 833 struct componentname *cnp) 834{ 835 struct tmpfs_dir_duphead *duphead; 836 struct tmpfs_dirent *de; 837 uint32_t hash; 838 839 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 840 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 841 cnp->cn_nameptr[1] == '.'))); 842 TMPFS_VALIDATE_DIR(node); 843 844 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 845 de = tmpfs_dir_xlookup_hash(node, hash); 846 if (de != NULL && tmpfs_dirent_duphead(de)) { 847 duphead = &de->ud.td_duphead; 848 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 849 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 850 cnp->cn_namelen)) 851 break; 852 } 853 } else if (de != NULL) { 854 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 855 cnp->cn_namelen)) 856 de = NULL; 857 } 858 if (de != NULL && f != NULL && de->td_node != f) 859 de = NULL; 860 861 return (de); 862} 863 864/* 865 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 866 * list, allocate new cookie value. 867 */ 868static void 869tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 870 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 871{ 872 struct tmpfs_dir_duphead *dupindex; 873 struct tmpfs_dirent *de, *pde; 874 875 dupindex = &dnode->tn_dir.tn_dupindex; 876 de = LIST_FIRST(dupindex); 877 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 878 if (de == NULL) 879 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 880 else 881 nde->td_cookie = de->td_cookie + 1; 882 MPASS(tmpfs_dirent_dup(nde)); 883 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 884 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 885 return; 886 } 887 888 /* 889 * Cookie numbers are near exhaustion. Scan dupindex list for unused 890 * numbers. dupindex list is sorted in descending order. Keep it so 891 * after inserting nde. 892 */ 893 while (1) { 894 pde = de; 895 de = LIST_NEXT(de, uh.td_dup.index_entries); 896 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 897 /* 898 * Last element of the index doesn't have minimal cookie 899 * value, use it. 900 */ 901 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 902 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 903 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 904 return; 905 } else if (de == NULL) { 906 /* 907 * We are so lucky have 2^30 hash duplicates in single 908 * directory :) Return largest possible cookie value. 909 * It should be fine except possible issues with 910 * VOP_READDIR restart. 911 */ 912 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 913 LIST_INSERT_HEAD(dupindex, nde, 914 uh.td_dup.index_entries); 915 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 916 return; 917 } 918 if (de->td_cookie + 1 == pde->td_cookie || 919 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 920 continue; /* No hole or invalid cookie. */ 921 nde->td_cookie = de->td_cookie + 1; 922 MPASS(tmpfs_dirent_dup(nde)); 923 MPASS(pde->td_cookie > nde->td_cookie); 924 MPASS(nde->td_cookie > de->td_cookie); 925 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 926 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 927 return; 928 }; 929} 930 931/* 932 * Attaches the directory entry de to the directory represented by vp. 933 * Note that this does not change the link count of the node pointed by 934 * the directory entry, as this is done by tmpfs_alloc_dirent. 935 */ 936void 937tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 938{ 939 struct tmpfs_node *dnode; 940 struct tmpfs_dirent *xde, *nde; 941 942 ASSERT_VOP_ELOCKED(vp, __func__); 943 MPASS(de->td_namelen > 0); 944 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 945 MPASS(de->td_cookie == de->td_hash); 946 947 dnode = VP_TO_TMPFS_DIR(vp); 948 dnode->tn_dir.tn_readdir_lastn = 0; 949 dnode->tn_dir.tn_readdir_lastp = NULL; 950 951 MPASS(!tmpfs_dirent_dup(de)); 952 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 953 if (xde != NULL && tmpfs_dirent_duphead(xde)) 954 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 955 else if (xde != NULL) { 956 /* 957 * Allocate new duphead. Swap xde with duphead to avoid 958 * adding/removing elements with the same hash. 959 */ 960 MPASS(!tmpfs_dirent_dup(xde)); 961 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 962 &nde); 963 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 964 memcpy(nde, xde, sizeof(*xde)); 965 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 966 LIST_INIT(&xde->ud.td_duphead); 967 xde->td_namelen = 0; 968 xde->td_node = NULL; 969 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 970 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 971 } 972 dnode->tn_size += sizeof(struct tmpfs_dirent); 973 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 974 TMPFS_NODE_MODIFIED; 975} 976 977/* 978 * Detaches the directory entry de from the directory represented by vp. 979 * Note that this does not change the link count of the node pointed by 980 * the directory entry, as this is done by tmpfs_free_dirent. 981 */ 982void 983tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 984{ 985 struct tmpfs_mount *tmp; 986 struct tmpfs_dir *head; 987 struct tmpfs_node *dnode; 988 struct tmpfs_dirent *xde; 989 990 ASSERT_VOP_ELOCKED(vp, __func__); 991 992 dnode = VP_TO_TMPFS_DIR(vp); 993 head = &dnode->tn_dir.tn_dirhead; 994 dnode->tn_dir.tn_readdir_lastn = 0; 995 dnode->tn_dir.tn_readdir_lastp = NULL; 996 997 if (tmpfs_dirent_dup(de)) { 998 /* Remove duphead if de was last entry. */ 999 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1000 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1001 MPASS(tmpfs_dirent_duphead(xde)); 1002 } else 1003 xde = NULL; 1004 LIST_REMOVE(de, uh.td_dup.entries); 1005 LIST_REMOVE(de, uh.td_dup.index_entries); 1006 if (xde != NULL) { 1007 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1008 RB_REMOVE(tmpfs_dir, head, xde); 1009 tmp = VFS_TO_TMPFS(vp->v_mount); 1010 MPASS(xde->td_node == NULL); 1011 tmpfs_free_dirent(tmp, xde); 1012 } 1013 } 1014 } else 1015 RB_REMOVE(tmpfs_dir, head, de); 1016 1017 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1018 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1019 TMPFS_NODE_MODIFIED; 1020} 1021 1022void 1023tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1024{ 1025 struct tmpfs_dirent *de, *dde, *nde; 1026 1027 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1028 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1029 /* Node may already be destroyed. */ 1030 de->td_node = NULL; 1031 if (tmpfs_dirent_duphead(de)) { 1032 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1033 LIST_REMOVE(dde, uh.td_dup.entries); 1034 dde->td_node = NULL; 1035 tmpfs_free_dirent(tmp, dde); 1036 } 1037 } 1038 tmpfs_free_dirent(tmp, de); 1039 } 1040} 1041 1042/* 1043 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1044 * directory and returns it in the uio space. The function returns 0 1045 * on success, -1 if there was not enough space in the uio structure to 1046 * hold the directory entry or an appropriate error code if another 1047 * error happens. 1048 */ 1049static int 1050tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1051{ 1052 int error; 1053 struct dirent dent; 1054 1055 TMPFS_VALIDATE_DIR(node); 1056 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1057 1058 dent.d_fileno = node->tn_id; 1059 dent.d_type = DT_DIR; 1060 dent.d_namlen = 1; 1061 dent.d_name[0] = '.'; 1062 dent.d_name[1] = '\0'; 1063 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1064 1065 if (dent.d_reclen > uio->uio_resid) 1066 error = EJUSTRETURN; 1067 else 1068 error = uiomove(&dent, dent.d_reclen, uio); 1069 1070 node->tn_status |= TMPFS_NODE_ACCESSED; 1071 1072 return error; 1073} 1074 1075/* 1076 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1077 * directory and returns it in the uio space. The function returns 0 1078 * on success, -1 if there was not enough space in the uio structure to 1079 * hold the directory entry or an appropriate error code if another 1080 * error happens. 1081 */ 1082static int 1083tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1084{ 1085 int error; 1086 struct dirent dent; 1087 1088 TMPFS_VALIDATE_DIR(node); 1089 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1090 1091 /* 1092 * Return ENOENT if the current node is already removed. 1093 */ 1094 TMPFS_ASSERT_LOCKED(node); 1095 if (node->tn_dir.tn_parent == NULL) { 1096 return (ENOENT); 1097 } 1098 1099 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1100 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1101 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1102 1103 dent.d_type = DT_DIR; 1104 dent.d_namlen = 2; 1105 dent.d_name[0] = '.'; 1106 dent.d_name[1] = '.'; 1107 dent.d_name[2] = '\0'; 1108 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1109 1110 if (dent.d_reclen > uio->uio_resid) 1111 error = EJUSTRETURN; 1112 else 1113 error = uiomove(&dent, dent.d_reclen, uio); 1114 1115 node->tn_status |= TMPFS_NODE_ACCESSED; 1116 1117 return error; 1118} 1119 1120/* 1121 * Helper function for tmpfs_readdir. Returns as much directory entries 1122 * as can fit in the uio space. The read starts at uio->uio_offset. 1123 * The function returns 0 on success, -1 if there was not enough space 1124 * in the uio structure to hold the directory entry or an appropriate 1125 * error code if another error happens. 1126 */ 1127int 1128tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1129 u_long *cookies, int *ncookies) 1130{ 1131 struct tmpfs_dir_cursor dc; 1132 struct tmpfs_dirent *de; 1133 off_t off; 1134 int error; 1135 1136 TMPFS_VALIDATE_DIR(node); 1137 1138 off = 0; 1139 1140 /* 1141 * Lookup the node from the current offset. The starting offset of 1142 * 0 will lookup both '.' and '..', and then the first real entry, 1143 * or EOF if there are none. Then find all entries for the dir that 1144 * fit into the buffer. Once no more entries are found (de == NULL), 1145 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1146 * call to return 0. 1147 */ 1148 switch (uio->uio_offset) { 1149 case TMPFS_DIRCOOKIE_DOT: 1150 error = tmpfs_dir_getdotdent(node, uio); 1151 if (error != 0) 1152 return (error); 1153 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1154 if (cookies != NULL) 1155 cookies[(*ncookies)++] = off = uio->uio_offset; 1156 /* FALLTHROUGH */ 1157 case TMPFS_DIRCOOKIE_DOTDOT: 1158 error = tmpfs_dir_getdotdotdent(node, uio); 1159 if (error != 0) 1160 return (error); 1161 de = tmpfs_dir_first(node, &dc); 1162 uio->uio_offset = tmpfs_dirent_cookie(de); 1163 if (cookies != NULL) 1164 cookies[(*ncookies)++] = off = uio->uio_offset; 1165 /* EOF. */ 1166 if (de == NULL) 1167 return (0); 1168 break; 1169 case TMPFS_DIRCOOKIE_EOF: 1170 return (0); 1171 default: 1172 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1173 if (de == NULL) 1174 return (EINVAL); 1175 if (cookies != NULL) 1176 off = tmpfs_dirent_cookie(de); 1177 } 1178 1179 /* Read as much entries as possible; i.e., until we reach the end of 1180 * the directory or we exhaust uio space. */ 1181 do { 1182 struct dirent d; 1183 1184 /* Create a dirent structure representing the current 1185 * tmpfs_node and fill it. */ 1186 if (de->td_node == NULL) { 1187 d.d_fileno = 1; 1188 d.d_type = DT_WHT; 1189 } else { 1190 d.d_fileno = de->td_node->tn_id; 1191 switch (de->td_node->tn_type) { 1192 case VBLK: 1193 d.d_type = DT_BLK; 1194 break; 1195 1196 case VCHR: 1197 d.d_type = DT_CHR; 1198 break; 1199 1200 case VDIR: 1201 d.d_type = DT_DIR; 1202 break; 1203 1204 case VFIFO: 1205 d.d_type = DT_FIFO; 1206 break; 1207 1208 case VLNK: 1209 d.d_type = DT_LNK; 1210 break; 1211 1212 case VREG: 1213 d.d_type = DT_REG; 1214 break; 1215 1216 case VSOCK: 1217 d.d_type = DT_SOCK; 1218 break; 1219 1220 default: 1221 panic("tmpfs_dir_getdents: type %p %d", 1222 de->td_node, (int)de->td_node->tn_type); 1223 } 1224 } 1225 d.d_namlen = de->td_namelen; 1226 MPASS(de->td_namelen < sizeof(d.d_name)); 1227 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1228 d.d_name[de->td_namelen] = '\0'; 1229 d.d_reclen = GENERIC_DIRSIZ(&d); 1230 1231 /* Stop reading if the directory entry we are treating is 1232 * bigger than the amount of data that can be returned. */ 1233 if (d.d_reclen > uio->uio_resid) { 1234 error = EJUSTRETURN; 1235 break; 1236 } 1237 1238 /* Copy the new dirent structure into the output buffer and 1239 * advance pointers. */ 1240 error = uiomove(&d, d.d_reclen, uio); 1241 if (error == 0) { 1242 de = tmpfs_dir_next(node, &dc); 1243 if (cookies != NULL) { 1244 off = tmpfs_dirent_cookie(de); 1245 MPASS(*ncookies < maxcookies); 1246 cookies[(*ncookies)++] = off; 1247 } 1248 } 1249 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1250 1251 /* Skip setting off when using cookies as it is already done above. */ 1252 if (cookies == NULL) 1253 off = tmpfs_dirent_cookie(de); 1254 1255 /* Update the offset and cache. */ 1256 uio->uio_offset = off; 1257 node->tn_dir.tn_readdir_lastn = off; 1258 node->tn_dir.tn_readdir_lastp = de; 1259 1260 node->tn_status |= TMPFS_NODE_ACCESSED; 1261 return error; 1262} 1263 1264int 1265tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1266{ 1267 struct tmpfs_dirent *de; 1268 int error; 1269 1270 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1271 cnp->cn_nameptr, cnp->cn_namelen, &de); 1272 if (error != 0) 1273 return (error); 1274 tmpfs_dir_attach(dvp, de); 1275 return (0); 1276} 1277 1278void 1279tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1280{ 1281 struct tmpfs_dirent *de; 1282 1283 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1284 MPASS(de != NULL && de->td_node == NULL); 1285 tmpfs_dir_detach(dvp, de); 1286 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1287} 1288 1289/* 1290 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1291 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1292 * 'newsize' must be positive. 1293 * 1294 * Returns zero on success or an appropriate error code on failure. 1295 */ 1296int 1297tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1298{ 1299 struct tmpfs_mount *tmp; 1300 struct tmpfs_node *node; 1301 vm_object_t uobj; 1302 vm_page_t m, ma[1]; 1303 vm_pindex_t idx, newpages, oldpages; 1304 off_t oldsize; 1305 int base, rv; 1306 1307 MPASS(vp->v_type == VREG); 1308 MPASS(newsize >= 0); 1309 1310 node = VP_TO_TMPFS_NODE(vp); 1311 uobj = node->tn_reg.tn_aobj; 1312 tmp = VFS_TO_TMPFS(vp->v_mount); 1313 1314 /* 1315 * Convert the old and new sizes to the number of pages needed to 1316 * store them. It may happen that we do not need to do anything 1317 * because the last allocated page can accommodate the change on 1318 * its own. 1319 */ 1320 oldsize = node->tn_size; 1321 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1322 MPASS(oldpages == uobj->size); 1323 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1324 if (newpages > oldpages && 1325 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1326 return (ENOSPC); 1327 1328 VM_OBJECT_WLOCK(uobj); 1329 if (newsize < oldsize) { 1330 /* 1331 * Zero the truncated part of the last page. 1332 */ 1333 base = newsize & PAGE_MASK; 1334 if (base != 0) { 1335 idx = OFF_TO_IDX(newsize); 1336retry: 1337 m = vm_page_lookup(uobj, idx); 1338 if (m != NULL) { 1339 if (vm_page_sleep_if_busy(m, "tmfssz")) 1340 goto retry; 1341 MPASS(m->valid == VM_PAGE_BITS_ALL); 1342 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1343 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1344 if (m == NULL) { 1345 VM_OBJECT_WUNLOCK(uobj); 1346 VM_WAIT; 1347 VM_OBJECT_WLOCK(uobj); 1348 goto retry; 1349 } else if (m->valid != VM_PAGE_BITS_ALL) { 1350 ma[0] = m; 1351 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1352 m = vm_page_lookup(uobj, idx); 1353 } else 1354 /* A cached page was reactivated. */ 1355 rv = VM_PAGER_OK; 1356 vm_page_lock(m); 1357 if (rv == VM_PAGER_OK) { 1358 vm_page_deactivate(m); 1359 vm_page_unlock(m); 1360 vm_page_xunbusy(m); 1361 } else { 1362 vm_page_free(m); 1363 vm_page_unlock(m); 1364 if (ignerr) 1365 m = NULL; 1366 else { 1367 VM_OBJECT_WUNLOCK(uobj); 1368 return (EIO); 1369 } 1370 } 1371 } 1372 if (m != NULL) { 1373 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1374 vm_page_dirty(m); 1375 vm_pager_page_unswapped(m); 1376 } 1377 } 1378 1379 /* 1380 * Release any swap space and free any whole pages. 1381 */ 1382 if (newpages < oldpages) { 1383 swap_pager_freespace(uobj, newpages, oldpages - 1384 newpages); 1385 vm_object_page_remove(uobj, newpages, 0, 0); 1386 } 1387 } 1388 uobj->size = newpages; 1389 VM_OBJECT_WUNLOCK(uobj); 1390 1391 TMPFS_LOCK(tmp); 1392 tmp->tm_pages_used += (newpages - oldpages); 1393 TMPFS_UNLOCK(tmp); 1394 1395 node->tn_size = newsize; 1396 return (0); 1397} 1398 1399/* 1400 * Change flags of the given vnode. 1401 * Caller should execute tmpfs_update on vp after a successful execution. 1402 * The vnode must be locked on entry and remain locked on exit. 1403 */ 1404int 1405tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1406 struct thread *p) 1407{ 1408 int error; 1409 struct tmpfs_node *node; 1410 1411 MPASS(VOP_ISLOCKED(vp)); 1412 1413 node = VP_TO_TMPFS_NODE(vp); 1414 1415 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1416 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1417 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1418 UF_SPARSE | UF_SYSTEM)) != 0) 1419 return (EOPNOTSUPP); 1420 1421 /* Disallow this operation if the file system is mounted read-only. */ 1422 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1423 return EROFS; 1424 1425 /* 1426 * Callers may only modify the file flags on objects they 1427 * have VADMIN rights for. 1428 */ 1429 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1430 return (error); 1431 /* 1432 * Unprivileged processes are not permitted to unset system 1433 * flags, or modify flags if any system flags are set. 1434 */ 1435 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1436 if (node->tn_flags & 1437 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1438 error = securelevel_gt(cred, 0); 1439 if (error) 1440 return (error); 1441 } 1442 } else { 1443 if (node->tn_flags & 1444 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1445 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1446 return (EPERM); 1447 } 1448 node->tn_flags = flags; 1449 node->tn_status |= TMPFS_NODE_CHANGED; 1450 1451 MPASS(VOP_ISLOCKED(vp)); 1452 1453 return 0; 1454} 1455 1456/* 1457 * Change access mode on the given vnode. 1458 * Caller should execute tmpfs_update on vp after a successful execution. 1459 * The vnode must be locked on entry and remain locked on exit. 1460 */ 1461int 1462tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1463{ 1464 int error; 1465 struct tmpfs_node *node; 1466 1467 MPASS(VOP_ISLOCKED(vp)); 1468 1469 node = VP_TO_TMPFS_NODE(vp); 1470 1471 /* Disallow this operation if the file system is mounted read-only. */ 1472 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1473 return EROFS; 1474 1475 /* Immutable or append-only files cannot be modified, either. */ 1476 if (node->tn_flags & (IMMUTABLE | APPEND)) 1477 return EPERM; 1478 1479 /* 1480 * To modify the permissions on a file, must possess VADMIN 1481 * for that file. 1482 */ 1483 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1484 return (error); 1485 1486 /* 1487 * Privileged processes may set the sticky bit on non-directories, 1488 * as well as set the setgid bit on a file with a group that the 1489 * process is not a member of. 1490 */ 1491 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1492 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1493 return (EFTYPE); 1494 } 1495 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1496 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1497 if (error) 1498 return (error); 1499 } 1500 1501 1502 node->tn_mode &= ~ALLPERMS; 1503 node->tn_mode |= mode & ALLPERMS; 1504 1505 node->tn_status |= TMPFS_NODE_CHANGED; 1506 1507 MPASS(VOP_ISLOCKED(vp)); 1508 1509 return 0; 1510} 1511 1512/* 1513 * Change ownership of the given vnode. At least one of uid or gid must 1514 * be different than VNOVAL. If one is set to that value, the attribute 1515 * is unchanged. 1516 * Caller should execute tmpfs_update on vp after a successful execution. 1517 * The vnode must be locked on entry and remain locked on exit. 1518 */ 1519int 1520tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1521 struct thread *p) 1522{ 1523 int error; 1524 struct tmpfs_node *node; 1525 uid_t ouid; 1526 gid_t ogid; 1527 1528 MPASS(VOP_ISLOCKED(vp)); 1529 1530 node = VP_TO_TMPFS_NODE(vp); 1531 1532 /* Assign default values if they are unknown. */ 1533 MPASS(uid != VNOVAL || gid != VNOVAL); 1534 if (uid == VNOVAL) 1535 uid = node->tn_uid; 1536 if (gid == VNOVAL) 1537 gid = node->tn_gid; 1538 MPASS(uid != VNOVAL && gid != VNOVAL); 1539 1540 /* Disallow this operation if the file system is mounted read-only. */ 1541 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1542 return EROFS; 1543 1544 /* Immutable or append-only files cannot be modified, either. */ 1545 if (node->tn_flags & (IMMUTABLE | APPEND)) 1546 return EPERM; 1547 1548 /* 1549 * To modify the ownership of a file, must possess VADMIN for that 1550 * file. 1551 */ 1552 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1553 return (error); 1554 1555 /* 1556 * To change the owner of a file, or change the group of a file to a 1557 * group of which we are not a member, the caller must have 1558 * privilege. 1559 */ 1560 if ((uid != node->tn_uid || 1561 (gid != node->tn_gid && !groupmember(gid, cred))) && 1562 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1563 return (error); 1564 1565 ogid = node->tn_gid; 1566 ouid = node->tn_uid; 1567 1568 node->tn_uid = uid; 1569 node->tn_gid = gid; 1570 1571 node->tn_status |= TMPFS_NODE_CHANGED; 1572 1573 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1574 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1575 node->tn_mode &= ~(S_ISUID | S_ISGID); 1576 } 1577 1578 MPASS(VOP_ISLOCKED(vp)); 1579 1580 return 0; 1581} 1582 1583/* 1584 * Change size of the given vnode. 1585 * Caller should execute tmpfs_update on vp after a successful execution. 1586 * The vnode must be locked on entry and remain locked on exit. 1587 */ 1588int 1589tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1590 struct thread *p) 1591{ 1592 int error; 1593 struct tmpfs_node *node; 1594 1595 MPASS(VOP_ISLOCKED(vp)); 1596 1597 node = VP_TO_TMPFS_NODE(vp); 1598 1599 /* Decide whether this is a valid operation based on the file type. */ 1600 error = 0; 1601 switch (vp->v_type) { 1602 case VDIR: 1603 return EISDIR; 1604 1605 case VREG: 1606 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1607 return EROFS; 1608 break; 1609 1610 case VBLK: 1611 /* FALLTHROUGH */ 1612 case VCHR: 1613 /* FALLTHROUGH */ 1614 case VFIFO: 1615 /* Allow modifications of special files even if in the file 1616 * system is mounted read-only (we are not modifying the 1617 * files themselves, but the objects they represent). */ 1618 return 0; 1619 1620 default: 1621 /* Anything else is unsupported. */ 1622 return EOPNOTSUPP; 1623 } 1624 1625 /* Immutable or append-only files cannot be modified, either. */ 1626 if (node->tn_flags & (IMMUTABLE | APPEND)) 1627 return EPERM; 1628 1629 error = tmpfs_truncate(vp, size); 1630 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1631 * for us, as will update tn_status; no need to do that here. */ 1632 1633 MPASS(VOP_ISLOCKED(vp)); 1634 1635 return error; 1636} 1637 1638/* 1639 * Change access and modification times of the given vnode. 1640 * Caller should execute tmpfs_update on vp after a successful execution. 1641 * The vnode must be locked on entry and remain locked on exit. 1642 */ 1643int 1644tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 1645 struct ucred *cred, struct thread *l) 1646{ 1647 int error; 1648 struct tmpfs_node *node; 1649 1650 MPASS(VOP_ISLOCKED(vp)); 1651 1652 node = VP_TO_TMPFS_NODE(vp); 1653 1654 /* Disallow this operation if the file system is mounted read-only. */ 1655 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1656 return EROFS; 1657 1658 /* Immutable or append-only files cannot be modified, either. */ 1659 if (node->tn_flags & (IMMUTABLE | APPEND)) 1660 return EPERM; 1661 1662 error = vn_utimes_perm(vp, vap, cred, l); 1663 if (error != 0) 1664 return (error); 1665 1666 if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL) 1667 node->tn_status |= TMPFS_NODE_ACCESSED; 1668 1669 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL) 1670 node->tn_status |= TMPFS_NODE_MODIFIED; 1671 1672 if (vap->va_birthtime.tv_nsec != VNOVAL && 1673 vap->va_birthtime.tv_nsec != VNOVAL) 1674 node->tn_status |= TMPFS_NODE_MODIFIED; 1675 1676 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1677 1678 if (vap->va_birthtime.tv_nsec != VNOVAL && 1679 vap->va_birthtime.tv_nsec != VNOVAL) 1680 node->tn_birthtime = vap->va_birthtime; 1681 MPASS(VOP_ISLOCKED(vp)); 1682 1683 return 0; 1684} 1685 1686/* Sync timestamps */ 1687void 1688tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1689 const struct timespec *mod) 1690{ 1691 struct tmpfs_node *node; 1692 struct timespec now; 1693 1694 node = VP_TO_TMPFS_NODE(vp); 1695 1696 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1697 TMPFS_NODE_CHANGED)) == 0) 1698 return; 1699 1700 vfs_timestamp(&now); 1701 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1702 if (acc == NULL) 1703 acc = &now; 1704 node->tn_atime = *acc; 1705 } 1706 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1707 if (mod == NULL) 1708 mod = &now; 1709 node->tn_mtime = *mod; 1710 } 1711 if (node->tn_status & TMPFS_NODE_CHANGED) { 1712 node->tn_ctime = now; 1713 } 1714 node->tn_status &= 1715 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1716} 1717 1718void 1719tmpfs_update(struct vnode *vp) 1720{ 1721 1722 tmpfs_itimes(vp, NULL, NULL); 1723} 1724 1725int 1726tmpfs_truncate(struct vnode *vp, off_t length) 1727{ 1728 int error; 1729 struct tmpfs_node *node; 1730 1731 node = VP_TO_TMPFS_NODE(vp); 1732 1733 if (length < 0) { 1734 error = EINVAL; 1735 goto out; 1736 } 1737 1738 if (node->tn_size == length) { 1739 error = 0; 1740 goto out; 1741 } 1742 1743 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1744 return (EFBIG); 1745 1746 error = tmpfs_reg_resize(vp, length, FALSE); 1747 if (error == 0) { 1748 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1749 } 1750 1751out: 1752 tmpfs_update(vp); 1753 1754 return error; 1755} 1756 1757static __inline int 1758tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1759{ 1760 if (a->td_hash > b->td_hash) 1761 return (1); 1762 else if (a->td_hash < b->td_hash) 1763 return (-1); 1764 return (0); 1765} 1766 1767RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1768