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("$FreeBSD$"); 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 64 struct tmpfs_dir_cursor { 65 struct tmpfs_dirent *tdc_current; 66 struct tmpfs_dirent *tdc_tree; 67 }; 68 69 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system"); 70 71 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED; 72 73 static int 74 sysctl_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 94 SYSCTL_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 98 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a, 99 struct tmpfs_dirent *b); 100 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 101 102 size_t 103 tmpfs_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 } 113 114 size_t 115 tmpfs_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 } 125 126 static size_t 127 tmpfs_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 /* --------------------------------------------------------------------- */ 140 141 /* 142 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 143 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 144 * using the credentials of the process 'p'. 145 * 146 * If the node type is set to 'VDIR', then the parent parameter must point 147 * to the parent directory of the node being created. It may only be NULL 148 * while allocating the root node. 149 * 150 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 151 * specifies the device the node represents. 152 * 153 * If the node type is set to 'VLNK', then the parameter target specifies 154 * the file name of the target file for the symbolic link that is being 155 * created. 156 * 157 * Note that new nodes are retrieved from the available list if it has 158 * items or, if it is empty, from the node pool as long as there is enough 159 * space to create them. 160 * 161 * Returns zero on success or an appropriate error code on failure. 162 */ 163 int 164 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 165 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 166 char *target, dev_t rdev, struct tmpfs_node **node) 167 { 168 struct tmpfs_node *nnode; 169 vm_object_t obj; 170 171 /* If the root directory of the 'tmp' file system is not yet 172 * allocated, this must be the request to do it. */ 173 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 174 175 MPASS(IFF(type == VLNK, target != NULL)); 176 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL)); 177 178 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 179 return (ENOSPC); 180 if (tmpfs_pages_check_avail(tmp, 1) == 0) 181 return (ENOSPC); 182 183 nnode = (struct tmpfs_node *)uma_zalloc_arg( 184 tmp->tm_node_pool, tmp, M_WAITOK); 185 186 /* Generic initialization. */ 187 nnode->tn_type = type; 188 vfs_timestamp(&nnode->tn_atime); 189 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 190 nnode->tn_atime; 191 nnode->tn_uid = uid; 192 nnode->tn_gid = gid; 193 nnode->tn_mode = mode; 194 nnode->tn_id = alloc_unr(tmp->tm_ino_unr); 195 196 /* Type-specific initialization. */ 197 switch (nnode->tn_type) { 198 case VBLK: 199 case VCHR: 200 nnode->tn_rdev = rdev; 201 break; 202 203 case VDIR: 204 RB_INIT(&nnode->tn_dir.tn_dirhead); 205 LIST_INIT(&nnode->tn_dir.tn_dupindex); 206 MPASS(parent != nnode); 207 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL)); 208 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent; 209 nnode->tn_dir.tn_readdir_lastn = 0; 210 nnode->tn_dir.tn_readdir_lastp = NULL; 211 nnode->tn_links++; 212 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent); 213 nnode->tn_dir.tn_parent->tn_links++; 214 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent); 215 break; 216 217 case VFIFO: 218 /* FALLTHROUGH */ 219 case VSOCK: 220 break; 221 222 case VLNK: 223 MPASS(strlen(target) < MAXPATHLEN); 224 nnode->tn_size = strlen(target); 225 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME, 226 M_WAITOK); 227 memcpy(nnode->tn_link, target, nnode->tn_size); 228 break; 229 230 case VREG: 231 obj = nnode->tn_reg.tn_aobj = 232 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0, 233 NULL /* XXXKIB - tmpfs needs swap reservation */); 234 VM_OBJECT_WLOCK(obj); 235 /* OBJ_TMPFS is set together with the setting of vp->v_object */ 236 vm_object_set_flag(obj, OBJ_NOSPLIT); 237 vm_object_clear_flag(obj, OBJ_ONEMAPPING); 238 VM_OBJECT_WUNLOCK(obj); 239 break; 240 241 default: 242 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 243 } 244 245 TMPFS_LOCK(tmp); 246 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 247 tmp->tm_nodes_inuse++; 248 TMPFS_UNLOCK(tmp); 249 250 *node = nnode; 251 return 0; 252 } 253 254 /* --------------------------------------------------------------------- */ 255 256 /* 257 * Destroys the node pointed to by node from the file system 'tmp'. 258 * If the node does not belong to the given mount point, the results are 259 * unpredicted. 260 * 261 * If the node references a directory; no entries are allowed because 262 * their removal could need a recursive algorithm, something forbidden in 263 * kernel space. Furthermore, there is not need to provide such 264 * functionality (recursive removal) because the only primitives offered 265 * to the user are the removal of empty directories and the deletion of 266 * individual files. 267 * 268 * Note that nodes are not really deleted; in fact, when a node has been 269 * allocated, it cannot be deleted during the whole life of the file 270 * system. Instead, they are moved to the available list and remain there 271 * until reused. 272 */ 273 void 274 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 275 { 276 vm_object_t uobj; 277 278 #ifdef INVARIANTS 279 TMPFS_NODE_LOCK(node); 280 MPASS(node->tn_vnode == NULL); 281 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 282 TMPFS_NODE_UNLOCK(node); 283 #endif 284 285 TMPFS_LOCK(tmp); 286 LIST_REMOVE(node, tn_entries); 287 tmp->tm_nodes_inuse--; 288 TMPFS_UNLOCK(tmp); 289 290 switch (node->tn_type) { 291 case VNON: 292 /* Do not do anything. VNON is provided to let the 293 * allocation routine clean itself easily by avoiding 294 * duplicating code in it. */ 295 /* FALLTHROUGH */ 296 case VBLK: 297 /* FALLTHROUGH */ 298 case VCHR: 299 /* FALLTHROUGH */ 300 case VDIR: 301 /* FALLTHROUGH */ 302 case VFIFO: 303 /* FALLTHROUGH */ 304 case VSOCK: 305 break; 306 307 case VLNK: 308 free(node->tn_link, M_TMPFSNAME); 309 break; 310 311 case VREG: 312 uobj = node->tn_reg.tn_aobj; 313 if (uobj != NULL) { 314 TMPFS_LOCK(tmp); 315 tmp->tm_pages_used -= uobj->size; 316 TMPFS_UNLOCK(tmp); 317 KASSERT((uobj->flags & OBJ_TMPFS) == 0, 318 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj)); 319 vm_object_deallocate(uobj); 320 } 321 break; 322 323 default: 324 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 325 } 326 327 free_unr(tmp->tm_ino_unr, node->tn_id); 328 uma_zfree(tmp->tm_node_pool, node); 329 } 330 331 /* --------------------------------------------------------------------- */ 332 333 static __inline uint32_t 334 tmpfs_dirent_hash(const char *name, u_int len) 335 { 336 uint32_t hash; 337 338 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK; 339 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP 340 hash &= 0xf; 341 #endif 342 if (hash < TMPFS_DIRCOOKIE_MIN) 343 hash += TMPFS_DIRCOOKIE_MIN; 344 345 return (hash); 346 } 347 348 static __inline off_t 349 tmpfs_dirent_cookie(struct tmpfs_dirent *de) 350 { 351 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN); 352 353 return (de->td_cookie); 354 } 355 356 static __inline boolean_t 357 tmpfs_dirent_dup(struct tmpfs_dirent *de) 358 { 359 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0); 360 } 361 362 static __inline boolean_t 363 tmpfs_dirent_duphead(struct tmpfs_dirent *de) 364 { 365 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0); 366 } 367 368 void 369 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen) 370 { 371 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen); 372 memcpy(de->ud.td_name, name, namelen); 373 de->td_namelen = namelen; 374 } 375 376 /* 377 * Allocates a new directory entry for the node node with a name of name. 378 * The new directory entry is returned in *de. 379 * 380 * The link count of node is increased by one to reflect the new object 381 * referencing it. 382 * 383 * Returns zero on success or an appropriate error code on failure. 384 */ 385 int 386 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 387 const char *name, u_int len, struct tmpfs_dirent **de) 388 { 389 struct tmpfs_dirent *nde; 390 391 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK); 392 nde->td_node = node; 393 if (name != NULL) { 394 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 395 tmpfs_dirent_init(nde, name, len); 396 } else 397 nde->td_namelen = 0; 398 if (node != NULL) 399 node->tn_links++; 400 401 *de = nde; 402 403 return 0; 404 } 405 406 /* --------------------------------------------------------------------- */ 407 408 /* 409 * Frees a directory entry. It is the caller's responsibility to destroy 410 * the node referenced by it if needed. 411 * 412 * The link count of node is decreased by one to reflect the removal of an 413 * object that referenced it. This only happens if 'node_exists' is true; 414 * otherwise the function will not access the node referred to by the 415 * directory entry, as it may already have been released from the outside. 416 */ 417 void 418 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 419 { 420 struct tmpfs_node *node; 421 422 node = de->td_node; 423 if (node != NULL) { 424 MPASS(node->tn_links > 0); 425 node->tn_links--; 426 } 427 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL) 428 free(de->ud.td_name, M_TMPFSNAME); 429 uma_zfree(tmp->tm_dirent_pool, de); 430 } 431 432 /* --------------------------------------------------------------------- */ 433 434 void 435 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj) 436 { 437 438 if (vp->v_type != VREG || obj == NULL) 439 return; 440 441 VM_OBJECT_WLOCK(obj); 442 VI_LOCK(vp); 443 vm_object_clear_flag(obj, OBJ_TMPFS); 444 obj->un_pager.swp.swp_tmpfs = NULL; 445 VI_UNLOCK(vp); 446 VM_OBJECT_WUNLOCK(obj); 447 } 448 449 /* 450 * Need to clear v_object for insmntque failure. 451 */ 452 static void 453 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg) 454 { 455 456 tmpfs_destroy_vobject(vp, vp->v_object); 457 vp->v_object = NULL; 458 vp->v_data = NULL; 459 vp->v_op = &dead_vnodeops; 460 vgone(vp); 461 vput(vp); 462 } 463 464 /* 465 * Allocates a new vnode for the node node or returns a new reference to 466 * an existing one if the node had already a vnode referencing it. The 467 * resulting locked vnode is returned in *vpp. 468 * 469 * Returns zero on success or an appropriate error code on failure. 470 */ 471 int 472 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 473 struct vnode **vpp) 474 { 475 struct vnode *vp; 476 vm_object_t object; 477 int error; 478 479 error = 0; 480 loop: 481 TMPFS_NODE_LOCK(node); 482 loop1: 483 if ((vp = node->tn_vnode) != NULL) { 484 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 485 VI_LOCK(vp); 486 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) || 487 ((vp->v_iflag & VI_DOOMED) != 0 && 488 (lkflag & LK_NOWAIT) != 0)) { 489 VI_UNLOCK(vp); 490 TMPFS_NODE_UNLOCK(node); 491 error = ENOENT; 492 vp = NULL; 493 goto out; 494 } 495 if ((vp->v_iflag & VI_DOOMED) != 0) { 496 VI_UNLOCK(vp); 497 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM; 498 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) { 499 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 500 0, "tmpfsE", 0); 501 } 502 goto loop1; 503 } 504 TMPFS_NODE_UNLOCK(node); 505 error = vget(vp, lkflag | LK_INTERLOCK, curthread); 506 if (error == ENOENT) 507 goto loop; 508 if (error != 0) { 509 vp = NULL; 510 goto out; 511 } 512 513 /* 514 * Make sure the vnode is still there after 515 * getting the interlock to avoid racing a free. 516 */ 517 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 518 vput(vp); 519 goto loop; 520 } 521 522 goto out; 523 } 524 525 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 526 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 527 TMPFS_NODE_UNLOCK(node); 528 error = ENOENT; 529 vp = NULL; 530 goto out; 531 } 532 533 /* 534 * otherwise lock the vp list while we call getnewvnode 535 * since that can block. 536 */ 537 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 538 node->tn_vpstate |= TMPFS_VNODE_WANT; 539 error = msleep((caddr_t) &node->tn_vpstate, 540 TMPFS_NODE_MTX(node), PDROP | PCATCH, 541 "tmpfs_alloc_vp", 0); 542 if (error) 543 return error; 544 545 goto loop; 546 } else 547 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 548 549 TMPFS_NODE_UNLOCK(node); 550 551 /* Get a new vnode and associate it with our node. */ 552 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); 553 if (error != 0) 554 goto unlock; 555 MPASS(vp != NULL); 556 557 (void) vn_lock(vp, lkflag | LK_RETRY); 558 559 vp->v_data = node; 560 vp->v_type = node->tn_type; 561 562 /* Type-specific initialization. */ 563 switch (node->tn_type) { 564 case VBLK: 565 /* FALLTHROUGH */ 566 case VCHR: 567 /* FALLTHROUGH */ 568 case VLNK: 569 /* FALLTHROUGH */ 570 case VSOCK: 571 break; 572 case VFIFO: 573 vp->v_op = &tmpfs_fifoop_entries; 574 break; 575 case VREG: 576 object = node->tn_reg.tn_aobj; 577 VM_OBJECT_WLOCK(object); 578 VI_LOCK(vp); 579 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 580 vp->v_object = object; 581 object->un_pager.swp.swp_tmpfs = vp; 582 vm_object_set_flag(object, OBJ_TMPFS); 583 VI_UNLOCK(vp); 584 VM_OBJECT_WUNLOCK(object); 585 break; 586 case VDIR: 587 MPASS(node->tn_dir.tn_parent != NULL); 588 if (node->tn_dir.tn_parent == node) 589 vp->v_vflag |= VV_ROOT; 590 break; 591 592 default: 593 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 594 } 595 596 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 597 if (error) 598 vp = NULL; 599 600 unlock: 601 TMPFS_NODE_LOCK(node); 602 603 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 604 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 605 node->tn_vnode = vp; 606 607 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 608 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 609 TMPFS_NODE_UNLOCK(node); 610 wakeup((caddr_t) &node->tn_vpstate); 611 } else 612 TMPFS_NODE_UNLOCK(node); 613 614 out: 615 *vpp = vp; 616 617 #ifdef INVARIANTS 618 if (error == 0) { 619 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 620 TMPFS_NODE_LOCK(node); 621 MPASS(*vpp == node->tn_vnode); 622 TMPFS_NODE_UNLOCK(node); 623 } 624 #endif 625 626 return error; 627 } 628 629 /* --------------------------------------------------------------------- */ 630 631 /* 632 * Destroys the association between the vnode vp and the node it 633 * references. 634 */ 635 void 636 tmpfs_free_vp(struct vnode *vp) 637 { 638 struct tmpfs_node *node; 639 640 node = VP_TO_TMPFS_NODE(vp); 641 642 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED); 643 node->tn_vnode = NULL; 644 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 645 wakeup(&node->tn_vnode); 646 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 647 vp->v_data = NULL; 648 } 649 650 /* --------------------------------------------------------------------- */ 651 652 /* 653 * Allocates a new file of type 'type' and adds it to the parent directory 654 * 'dvp'; this addition is done using the component name given in 'cnp'. 655 * The ownership of the new file is automatically assigned based on the 656 * credentials of the caller (through 'cnp'), the group is set based on 657 * the parent directory and the mode is determined from the 'vap' argument. 658 * If successful, *vpp holds a vnode to the newly created file and zero 659 * is returned. Otherwise *vpp is NULL and the function returns an 660 * appropriate error code. 661 */ 662 int 663 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 664 struct componentname *cnp, char *target) 665 { 666 int error; 667 struct tmpfs_dirent *de; 668 struct tmpfs_mount *tmp; 669 struct tmpfs_node *dnode; 670 struct tmpfs_node *node; 671 struct tmpfs_node *parent; 672 673 MPASS(VOP_ISLOCKED(dvp)); 674 MPASS(cnp->cn_flags & HASBUF); 675 676 tmp = VFS_TO_TMPFS(dvp->v_mount); 677 dnode = VP_TO_TMPFS_DIR(dvp); 678 *vpp = NULL; 679 680 /* If the entry we are creating is a directory, we cannot overflow 681 * the number of links of its parent, because it will get a new 682 * link. */ 683 if (vap->va_type == VDIR) { 684 /* Ensure that we do not overflow the maximum number of links 685 * imposed by the system. */ 686 MPASS(dnode->tn_links <= LINK_MAX); 687 if (dnode->tn_links == LINK_MAX) { 688 error = EMLINK; 689 goto out; 690 } 691 692 parent = dnode; 693 MPASS(parent != NULL); 694 } else 695 parent = NULL; 696 697 /* Allocate a node that represents the new file. */ 698 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid, 699 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node); 700 if (error != 0) 701 goto out; 702 703 /* Allocate a directory entry that points to the new file. */ 704 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 705 &de); 706 if (error != 0) { 707 tmpfs_free_node(tmp, node); 708 goto out; 709 } 710 711 /* Allocate a vnode for the new file. */ 712 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 713 if (error != 0) { 714 tmpfs_free_dirent(tmp, de); 715 tmpfs_free_node(tmp, node); 716 goto out; 717 } 718 719 /* Now that all required items are allocated, we can proceed to 720 * insert the new node into the directory, an operation that 721 * cannot fail. */ 722 if (cnp->cn_flags & ISWHITEOUT) 723 tmpfs_dir_whiteout_remove(dvp, cnp); 724 tmpfs_dir_attach(dvp, de); 725 726 out: 727 728 return error; 729 } 730 731 /* --------------------------------------------------------------------- */ 732 733 static struct tmpfs_dirent * 734 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 735 { 736 struct tmpfs_dirent *de; 737 738 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 739 dc->tdc_tree = de; 740 if (de != NULL && tmpfs_dirent_duphead(de)) 741 de = LIST_FIRST(&de->ud.td_duphead); 742 dc->tdc_current = de; 743 744 return (dc->tdc_current); 745 } 746 747 static struct tmpfs_dirent * 748 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 749 { 750 struct tmpfs_dirent *de; 751 752 MPASS(dc->tdc_tree != NULL); 753 if (tmpfs_dirent_dup(dc->tdc_current)) { 754 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 755 if (dc->tdc_current != NULL) 756 return (dc->tdc_current); 757 } 758 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 759 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 760 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 761 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 762 MPASS(dc->tdc_current != NULL); 763 } 764 765 return (dc->tdc_current); 766 } 767 768 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 769 static struct tmpfs_dirent * 770 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 771 { 772 struct tmpfs_dirent *de, dekey; 773 774 dekey.td_hash = hash; 775 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 776 return (de); 777 } 778 779 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 780 static struct tmpfs_dirent * 781 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 782 struct tmpfs_dir_cursor *dc) 783 { 784 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 785 struct tmpfs_dirent *de, dekey; 786 787 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 788 789 if (cookie == node->tn_dir.tn_readdir_lastn && 790 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 791 /* Protect against possible race, tn_readdir_last[pn] 792 * may be updated with only shared vnode lock held. */ 793 if (cookie == tmpfs_dirent_cookie(de)) 794 goto out; 795 } 796 797 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 798 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 799 uh.td_dup.index_entries) { 800 MPASS(tmpfs_dirent_dup(de)); 801 if (de->td_cookie == cookie) 802 goto out; 803 /* dupindex list is sorted. */ 804 if (de->td_cookie < cookie) { 805 de = NULL; 806 goto out; 807 } 808 } 809 MPASS(de == NULL); 810 goto out; 811 } 812 813 MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie); 814 dekey.td_hash = cookie; 815 /* Recover if direntry for cookie was removed */ 816 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 817 dc->tdc_tree = de; 818 dc->tdc_current = de; 819 if (de != NULL && tmpfs_dirent_duphead(de)) { 820 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 821 MPASS(dc->tdc_current != NULL); 822 } 823 return (dc->tdc_current); 824 825 out: 826 dc->tdc_tree = de; 827 dc->tdc_current = de; 828 if (de != NULL && tmpfs_dirent_dup(de)) 829 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 830 de->td_hash); 831 return (dc->tdc_current); 832 } 833 834 /* 835 * Looks for a directory entry in the directory represented by node. 836 * 'cnp' describes the name of the entry to look for. Note that the . 837 * and .. components are not allowed as they do not physically exist 838 * within directories. 839 * 840 * Returns a pointer to the entry when found, otherwise NULL. 841 */ 842 struct tmpfs_dirent * 843 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 844 struct componentname *cnp) 845 { 846 struct tmpfs_dir_duphead *duphead; 847 struct tmpfs_dirent *de; 848 uint32_t hash; 849 850 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 851 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 852 cnp->cn_nameptr[1] == '.'))); 853 TMPFS_VALIDATE_DIR(node); 854 855 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 856 de = tmpfs_dir_xlookup_hash(node, hash); 857 if (de != NULL && tmpfs_dirent_duphead(de)) { 858 duphead = &de->ud.td_duphead; 859 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 860 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 861 cnp->cn_namelen)) 862 break; 863 } 864 } else if (de != NULL) { 865 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 866 cnp->cn_namelen)) 867 de = NULL; 868 } 869 if (de != NULL && f != NULL && de->td_node != f) 870 de = NULL; 871 872 return (de); 873 } 874 875 /* 876 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 877 * list, allocate new cookie value. 878 */ 879 static void 880 tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 881 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 882 { 883 struct tmpfs_dir_duphead *dupindex; 884 struct tmpfs_dirent *de, *pde; 885 886 dupindex = &dnode->tn_dir.tn_dupindex; 887 de = LIST_FIRST(dupindex); 888 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 889 if (de == NULL) 890 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 891 else 892 nde->td_cookie = de->td_cookie + 1; 893 MPASS(tmpfs_dirent_dup(nde)); 894 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 895 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 896 return; 897 } 898 899 /* 900 * Cookie numbers are near exhaustion. Scan dupindex list for unused 901 * numbers. dupindex list is sorted in descending order. Keep it so 902 * after inserting nde. 903 */ 904 while (1) { 905 pde = de; 906 de = LIST_NEXT(de, uh.td_dup.index_entries); 907 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 908 /* 909 * Last element of the index doesn't have minimal cookie 910 * value, use it. 911 */ 912 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 913 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 914 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 915 return; 916 } else if (de == NULL) { 917 /* 918 * We are so lucky have 2^30 hash duplicates in single 919 * directory :) Return largest possible cookie value. 920 * It should be fine except possible issues with 921 * VOP_READDIR restart. 922 */ 923 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 924 LIST_INSERT_HEAD(dupindex, nde, 925 uh.td_dup.index_entries); 926 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 927 return; 928 } 929 if (de->td_cookie + 1 == pde->td_cookie || 930 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 931 continue; /* No hole or invalid cookie. */ 932 nde->td_cookie = de->td_cookie + 1; 933 MPASS(tmpfs_dirent_dup(nde)); 934 MPASS(pde->td_cookie > nde->td_cookie); 935 MPASS(nde->td_cookie > de->td_cookie); 936 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 937 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 938 return; 939 }; 940 } 941 942 /* 943 * Attaches the directory entry de to the directory represented by vp. 944 * Note that this does not change the link count of the node pointed by 945 * the directory entry, as this is done by tmpfs_alloc_dirent. 946 */ 947 void 948 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 949 { 950 struct tmpfs_node *dnode; 951 struct tmpfs_dirent *xde, *nde; 952 953 ASSERT_VOP_ELOCKED(vp, __func__); 954 MPASS(de->td_namelen > 0); 955 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 956 MPASS(de->td_cookie == de->td_hash); 957 958 dnode = VP_TO_TMPFS_DIR(vp); 959 dnode->tn_dir.tn_readdir_lastn = 0; 960 dnode->tn_dir.tn_readdir_lastp = NULL; 961 962 MPASS(!tmpfs_dirent_dup(de)); 963 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 964 if (xde != NULL && tmpfs_dirent_duphead(xde)) 965 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 966 else if (xde != NULL) { 967 /* 968 * Allocate new duphead. Swap xde with duphead to avoid 969 * adding/removing elements with the same hash. 970 */ 971 MPASS(!tmpfs_dirent_dup(xde)); 972 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 973 &nde); 974 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 975 memcpy(nde, xde, sizeof(*xde)); 976 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 977 LIST_INIT(&xde->ud.td_duphead); 978 xde->td_namelen = 0; 979 xde->td_node = NULL; 980 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 981 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 982 } 983 dnode->tn_size += sizeof(struct tmpfs_dirent); 984 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 985 TMPFS_NODE_MODIFIED; 986 } 987 988 /* --------------------------------------------------------------------- */ 989 990 /* 991 * Detaches the directory entry de from the directory represented by vp. 992 * Note that this does not change the link count of the node pointed by 993 * the directory entry, as this is done by tmpfs_free_dirent. 994 */ 995 void 996 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 997 { 998 struct tmpfs_mount *tmp; 999 struct tmpfs_dir *head; 1000 struct tmpfs_node *dnode; 1001 struct tmpfs_dirent *xde; 1002 1003 ASSERT_VOP_ELOCKED(vp, __func__); 1004 1005 dnode = VP_TO_TMPFS_DIR(vp); 1006 head = &dnode->tn_dir.tn_dirhead; 1007 dnode->tn_dir.tn_readdir_lastn = 0; 1008 dnode->tn_dir.tn_readdir_lastp = NULL; 1009 1010 if (tmpfs_dirent_dup(de)) { 1011 /* Remove duphead if de was last entry. */ 1012 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1013 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1014 MPASS(tmpfs_dirent_duphead(xde)); 1015 } else 1016 xde = NULL; 1017 LIST_REMOVE(de, uh.td_dup.entries); 1018 LIST_REMOVE(de, uh.td_dup.index_entries); 1019 if (xde != NULL) { 1020 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1021 RB_REMOVE(tmpfs_dir, head, xde); 1022 tmp = VFS_TO_TMPFS(vp->v_mount); 1023 MPASS(xde->td_node == NULL); 1024 tmpfs_free_dirent(tmp, xde); 1025 } 1026 } 1027 } else 1028 RB_REMOVE(tmpfs_dir, head, de); 1029 1030 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1031 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1032 TMPFS_NODE_MODIFIED; 1033 } 1034 1035 void 1036 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1037 { 1038 struct tmpfs_dirent *de, *dde, *nde; 1039 1040 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1041 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1042 /* Node may already be destroyed. */ 1043 de->td_node = NULL; 1044 if (tmpfs_dirent_duphead(de)) { 1045 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1046 LIST_REMOVE(dde, uh.td_dup.entries); 1047 dde->td_node = NULL; 1048 tmpfs_free_dirent(tmp, dde); 1049 } 1050 } 1051 tmpfs_free_dirent(tmp, de); 1052 } 1053 } 1054 1055 /* --------------------------------------------------------------------- */ 1056 1057 /* 1058 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1059 * directory and returns it in the uio space. The function returns 0 1060 * on success, -1 if there was not enough space in the uio structure to 1061 * hold the directory entry or an appropriate error code if another 1062 * error happens. 1063 */ 1064 static int 1065 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1066 { 1067 int error; 1068 struct dirent dent; 1069 1070 TMPFS_VALIDATE_DIR(node); 1071 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1072 1073 dent.d_fileno = node->tn_id; 1074 dent.d_type = DT_DIR; 1075 dent.d_namlen = 1; 1076 dent.d_name[0] = '.'; 1077 dent.d_name[1] = '\0'; 1078 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1079 1080 if (dent.d_reclen > uio->uio_resid) 1081 error = EJUSTRETURN; 1082 else 1083 error = uiomove(&dent, dent.d_reclen, uio); 1084 1085 node->tn_status |= TMPFS_NODE_ACCESSED; 1086 1087 return error; 1088 } 1089 1090 /* --------------------------------------------------------------------- */ 1091 1092 /* 1093 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1094 * directory and returns it in the uio space. The function returns 0 1095 * on success, -1 if there was not enough space in the uio structure to 1096 * hold the directory entry or an appropriate error code if another 1097 * error happens. 1098 */ 1099 static int 1100 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1101 { 1102 int error; 1103 struct dirent dent; 1104 1105 TMPFS_VALIDATE_DIR(node); 1106 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1107 1108 /* 1109 * Return ENOENT if the current node is already removed. 1110 */ 1111 TMPFS_ASSERT_LOCKED(node); 1112 if (node->tn_dir.tn_parent == NULL) { 1113 return (ENOENT); 1114 } 1115 1116 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1117 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1118 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1119 1120 dent.d_type = DT_DIR; 1121 dent.d_namlen = 2; 1122 dent.d_name[0] = '.'; 1123 dent.d_name[1] = '.'; 1124 dent.d_name[2] = '\0'; 1125 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1126 1127 if (dent.d_reclen > uio->uio_resid) 1128 error = EJUSTRETURN; 1129 else 1130 error = uiomove(&dent, dent.d_reclen, uio); 1131 1132 node->tn_status |= TMPFS_NODE_ACCESSED; 1133 1134 return error; 1135 } 1136 1137 /* --------------------------------------------------------------------- */ 1138 1139 /* 1140 * Helper function for tmpfs_readdir. Returns as much directory entries 1141 * as can fit in the uio space. The read starts at uio->uio_offset. 1142 * The function returns 0 on success, -1 if there was not enough space 1143 * in the uio structure to hold the directory entry or an appropriate 1144 * error code if another error happens. 1145 */ 1146 int 1147 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int cnt, 1148 u_long *cookies, int *ncookies) 1149 { 1150 struct tmpfs_dir_cursor dc; 1151 struct tmpfs_dirent *de; 1152 off_t off; 1153 int error; 1154 1155 TMPFS_VALIDATE_DIR(node); 1156 1157 off = 0; 1158 switch (uio->uio_offset) { 1159 case TMPFS_DIRCOOKIE_DOT: 1160 error = tmpfs_dir_getdotdent(node, uio); 1161 if (error != 0) 1162 return (error); 1163 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1164 if (cnt != 0) 1165 cookies[(*ncookies)++] = off = uio->uio_offset; 1166 case TMPFS_DIRCOOKIE_DOTDOT: 1167 error = tmpfs_dir_getdotdotdent(node, uio); 1168 if (error != 0) 1169 return (error); 1170 de = tmpfs_dir_first(node, &dc); 1171 if (de == NULL) 1172 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 1173 else 1174 uio->uio_offset = tmpfs_dirent_cookie(de); 1175 if (cnt != 0) 1176 cookies[(*ncookies)++] = off = uio->uio_offset; 1177 if (de == NULL) 1178 return (0); 1179 break; 1180 case TMPFS_DIRCOOKIE_EOF: 1181 return (0); 1182 default: 1183 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1184 if (de == NULL) 1185 return (EINVAL); 1186 if (cnt != 0) 1187 off = tmpfs_dirent_cookie(de); 1188 } 1189 1190 /* Read as much entries as possible; i.e., until we reach the end of 1191 * the directory or we exhaust uio space. */ 1192 do { 1193 struct dirent d; 1194 1195 /* Create a dirent structure representing the current 1196 * tmpfs_node and fill it. */ 1197 if (de->td_node == NULL) { 1198 d.d_fileno = 1; 1199 d.d_type = DT_WHT; 1200 } else { 1201 d.d_fileno = de->td_node->tn_id; 1202 switch (de->td_node->tn_type) { 1203 case VBLK: 1204 d.d_type = DT_BLK; 1205 break; 1206 1207 case VCHR: 1208 d.d_type = DT_CHR; 1209 break; 1210 1211 case VDIR: 1212 d.d_type = DT_DIR; 1213 break; 1214 1215 case VFIFO: 1216 d.d_type = DT_FIFO; 1217 break; 1218 1219 case VLNK: 1220 d.d_type = DT_LNK; 1221 break; 1222 1223 case VREG: 1224 d.d_type = DT_REG; 1225 break; 1226 1227 case VSOCK: 1228 d.d_type = DT_SOCK; 1229 break; 1230 1231 default: 1232 panic("tmpfs_dir_getdents: type %p %d", 1233 de->td_node, (int)de->td_node->tn_type); 1234 } 1235 } 1236 d.d_namlen = de->td_namelen; 1237 MPASS(de->td_namelen < sizeof(d.d_name)); 1238 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1239 d.d_name[de->td_namelen] = '\0'; 1240 d.d_reclen = GENERIC_DIRSIZ(&d); 1241 1242 /* Stop reading if the directory entry we are treating is 1243 * bigger than the amount of data that can be returned. */ 1244 if (d.d_reclen > uio->uio_resid) { 1245 error = EJUSTRETURN; 1246 break; 1247 } 1248 1249 /* Copy the new dirent structure into the output buffer and 1250 * advance pointers. */ 1251 error = uiomove(&d, d.d_reclen, uio); 1252 if (error == 0) { 1253 de = tmpfs_dir_next(node, &dc); 1254 if (cnt != 0) { 1255 if (de == NULL) 1256 off = TMPFS_DIRCOOKIE_EOF; 1257 else 1258 off = tmpfs_dirent_cookie(de); 1259 MPASS(*ncookies < cnt); 1260 cookies[(*ncookies)++] = off; 1261 } 1262 } 1263 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1264 1265 /* Update the offset and cache. */ 1266 if (cnt == 0) { 1267 if (de == NULL) 1268 off = TMPFS_DIRCOOKIE_EOF; 1269 else 1270 off = tmpfs_dirent_cookie(de); 1271 } 1272 1273 uio->uio_offset = off; 1274 node->tn_dir.tn_readdir_lastn = off; 1275 node->tn_dir.tn_readdir_lastp = de; 1276 1277 node->tn_status |= TMPFS_NODE_ACCESSED; 1278 return error; 1279 } 1280 1281 int 1282 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1283 { 1284 struct tmpfs_dirent *de; 1285 int error; 1286 1287 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1288 cnp->cn_nameptr, cnp->cn_namelen, &de); 1289 if (error != 0) 1290 return (error); 1291 tmpfs_dir_attach(dvp, de); 1292 return (0); 1293 } 1294 1295 void 1296 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1297 { 1298 struct tmpfs_dirent *de; 1299 1300 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1301 MPASS(de != NULL && de->td_node == NULL); 1302 tmpfs_dir_detach(dvp, de); 1303 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1304 } 1305 1306 /* --------------------------------------------------------------------- */ 1307 1308 /* 1309 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1310 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1311 * 'newsize' must be positive. 1312 * 1313 * Returns zero on success or an appropriate error code on failure. 1314 */ 1315 int 1316 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1317 { 1318 struct tmpfs_mount *tmp; 1319 struct tmpfs_node *node; 1320 vm_object_t uobj; 1321 vm_page_t m, ma[1]; 1322 vm_pindex_t idx, newpages, oldpages; 1323 off_t oldsize; 1324 int base, rv; 1325 1326 MPASS(vp->v_type == VREG); 1327 MPASS(newsize >= 0); 1328 1329 node = VP_TO_TMPFS_NODE(vp); 1330 uobj = node->tn_reg.tn_aobj; 1331 tmp = VFS_TO_TMPFS(vp->v_mount); 1332 1333 /* 1334 * Convert the old and new sizes to the number of pages needed to 1335 * store them. It may happen that we do not need to do anything 1336 * because the last allocated page can accommodate the change on 1337 * its own. 1338 */ 1339 oldsize = node->tn_size; 1340 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1341 MPASS(oldpages == uobj->size); 1342 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1343 if (newpages > oldpages && 1344 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1345 return (ENOSPC); 1346 1347 VM_OBJECT_WLOCK(uobj); 1348 if (newsize < oldsize) { 1349 /* 1350 * Zero the truncated part of the last page. 1351 */ 1352 base = newsize & PAGE_MASK; 1353 if (base != 0) { 1354 idx = OFF_TO_IDX(newsize); 1355 retry: 1356 m = vm_page_lookup(uobj, idx); 1357 if (m != NULL) { 1358 if (vm_page_sleep_if_busy(m, "tmfssz")) 1359 goto retry; 1360 MPASS(m->valid == VM_PAGE_BITS_ALL); 1361 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1362 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1363 if (m == NULL) { 1364 VM_OBJECT_WUNLOCK(uobj); 1365 VM_WAIT; 1366 VM_OBJECT_WLOCK(uobj); 1367 goto retry; 1368 } else if (m->valid != VM_PAGE_BITS_ALL) { 1369 ma[0] = m; 1370 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1371 m = vm_page_lookup(uobj, idx); 1372 } else 1373 /* A cached page was reactivated. */ 1374 rv = VM_PAGER_OK; 1375 vm_page_lock(m); 1376 if (rv == VM_PAGER_OK) { 1377 vm_page_deactivate(m); 1378 vm_page_unlock(m); 1379 vm_page_xunbusy(m); 1380 } else { 1381 vm_page_free(m); 1382 vm_page_unlock(m); 1383 if (ignerr) 1384 m = NULL; 1385 else { 1386 VM_OBJECT_WUNLOCK(uobj); 1387 return (EIO); 1388 } 1389 } 1390 } 1391 if (m != NULL) { 1392 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1393 vm_page_dirty(m); 1394 vm_pager_page_unswapped(m); 1395 } 1396 } 1397 1398 /* 1399 * Release any swap space and free any whole pages. 1400 */ 1401 if (newpages < oldpages) { 1402 swap_pager_freespace(uobj, newpages, oldpages - 1403 newpages); 1404 vm_object_page_remove(uobj, newpages, 0, 0); 1405 } 1406 } 1407 uobj->size = newpages; 1408 VM_OBJECT_WUNLOCK(uobj); 1409 1410 TMPFS_LOCK(tmp); 1411 tmp->tm_pages_used += (newpages - oldpages); 1412 TMPFS_UNLOCK(tmp); 1413 1414 node->tn_size = newsize; 1415 return (0); 1416 } 1417 1418 /* --------------------------------------------------------------------- */ 1419 1420 /* 1421 * Change flags of the given vnode. 1422 * Caller should execute tmpfs_update on vp after a successful execution. 1423 * The vnode must be locked on entry and remain locked on exit. 1424 */ 1425 int 1426 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1427 struct thread *p) 1428 { 1429 int error; 1430 struct tmpfs_node *node; 1431 1432 MPASS(VOP_ISLOCKED(vp)); 1433 1434 node = VP_TO_TMPFS_NODE(vp); 1435 1436 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1437 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1438 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1439 UF_SPARSE | UF_SYSTEM)) != 0) 1440 return (EOPNOTSUPP); 1441 1442 /* Disallow this operation if the file system is mounted read-only. */ 1443 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1444 return EROFS; 1445 1446 /* 1447 * Callers may only modify the file flags on objects they 1448 * have VADMIN rights for. 1449 */ 1450 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1451 return (error); 1452 /* 1453 * Unprivileged processes are not permitted to unset system 1454 * flags, or modify flags if any system flags are set. 1455 */ 1456 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1457 if (node->tn_flags & 1458 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1459 error = securelevel_gt(cred, 0); 1460 if (error) 1461 return (error); 1462 } 1463 } else { 1464 if (node->tn_flags & 1465 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1466 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1467 return (EPERM); 1468 } 1469 node->tn_flags = flags; 1470 node->tn_status |= TMPFS_NODE_CHANGED; 1471 1472 MPASS(VOP_ISLOCKED(vp)); 1473 1474 return 0; 1475 } 1476 1477 /* --------------------------------------------------------------------- */ 1478 1479 /* 1480 * Change access mode on the given vnode. 1481 * Caller should execute tmpfs_update on vp after a successful execution. 1482 * The vnode must be locked on entry and remain locked on exit. 1483 */ 1484 int 1485 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1486 { 1487 int error; 1488 struct tmpfs_node *node; 1489 1490 MPASS(VOP_ISLOCKED(vp)); 1491 1492 node = VP_TO_TMPFS_NODE(vp); 1493 1494 /* Disallow this operation if the file system is mounted read-only. */ 1495 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1496 return EROFS; 1497 1498 /* Immutable or append-only files cannot be modified, either. */ 1499 if (node->tn_flags & (IMMUTABLE | APPEND)) 1500 return EPERM; 1501 1502 /* 1503 * To modify the permissions on a file, must possess VADMIN 1504 * for that file. 1505 */ 1506 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1507 return (error); 1508 1509 /* 1510 * Privileged processes may set the sticky bit on non-directories, 1511 * as well as set the setgid bit on a file with a group that the 1512 * process is not a member of. 1513 */ 1514 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1515 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1516 return (EFTYPE); 1517 } 1518 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1519 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1520 if (error) 1521 return (error); 1522 } 1523 1524 1525 node->tn_mode &= ~ALLPERMS; 1526 node->tn_mode |= mode & ALLPERMS; 1527 1528 node->tn_status |= TMPFS_NODE_CHANGED; 1529 1530 MPASS(VOP_ISLOCKED(vp)); 1531 1532 return 0; 1533 } 1534 1535 /* --------------------------------------------------------------------- */ 1536 1537 /* 1538 * Change ownership of the given vnode. At least one of uid or gid must 1539 * be different than VNOVAL. If one is set to that value, the attribute 1540 * is unchanged. 1541 * Caller should execute tmpfs_update on vp after a successful execution. 1542 * The vnode must be locked on entry and remain locked on exit. 1543 */ 1544 int 1545 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1546 struct thread *p) 1547 { 1548 int error; 1549 struct tmpfs_node *node; 1550 uid_t ouid; 1551 gid_t ogid; 1552 1553 MPASS(VOP_ISLOCKED(vp)); 1554 1555 node = VP_TO_TMPFS_NODE(vp); 1556 1557 /* Assign default values if they are unknown. */ 1558 MPASS(uid != VNOVAL || gid != VNOVAL); 1559 if (uid == VNOVAL) 1560 uid = node->tn_uid; 1561 if (gid == VNOVAL) 1562 gid = node->tn_gid; 1563 MPASS(uid != VNOVAL && gid != VNOVAL); 1564 1565 /* Disallow this operation if the file system is mounted read-only. */ 1566 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1567 return EROFS; 1568 1569 /* Immutable or append-only files cannot be modified, either. */ 1570 if (node->tn_flags & (IMMUTABLE | APPEND)) 1571 return EPERM; 1572 1573 /* 1574 * To modify the ownership of a file, must possess VADMIN for that 1575 * file. 1576 */ 1577 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1578 return (error); 1579 1580 /* 1581 * To change the owner of a file, or change the group of a file to a 1582 * group of which we are not a member, the caller must have 1583 * privilege. 1584 */ 1585 if ((uid != node->tn_uid || 1586 (gid != node->tn_gid && !groupmember(gid, cred))) && 1587 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1588 return (error); 1589 1590 ogid = node->tn_gid; 1591 ouid = node->tn_uid; 1592 1593 node->tn_uid = uid; 1594 node->tn_gid = gid; 1595 1596 node->tn_status |= TMPFS_NODE_CHANGED; 1597 1598 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1599 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1600 node->tn_mode &= ~(S_ISUID | S_ISGID); 1601 } 1602 1603 MPASS(VOP_ISLOCKED(vp)); 1604 1605 return 0; 1606 } 1607 1608 /* --------------------------------------------------------------------- */ 1609 1610 /* 1611 * Change size of the given vnode. 1612 * Caller should execute tmpfs_update on vp after a successful execution. 1613 * The vnode must be locked on entry and remain locked on exit. 1614 */ 1615 int 1616 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1617 struct thread *p) 1618 { 1619 int error; 1620 struct tmpfs_node *node; 1621 1622 MPASS(VOP_ISLOCKED(vp)); 1623 1624 node = VP_TO_TMPFS_NODE(vp); 1625 1626 /* Decide whether this is a valid operation based on the file type. */ 1627 error = 0; 1628 switch (vp->v_type) { 1629 case VDIR: 1630 return EISDIR; 1631 1632 case VREG: 1633 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1634 return EROFS; 1635 break; 1636 1637 case VBLK: 1638 /* FALLTHROUGH */ 1639 case VCHR: 1640 /* FALLTHROUGH */ 1641 case VFIFO: 1642 /* Allow modifications of special files even if in the file 1643 * system is mounted read-only (we are not modifying the 1644 * files themselves, but the objects they represent). */ 1645 return 0; 1646 1647 default: 1648 /* Anything else is unsupported. */ 1649 return EOPNOTSUPP; 1650 } 1651 1652 /* Immutable or append-only files cannot be modified, either. */ 1653 if (node->tn_flags & (IMMUTABLE | APPEND)) 1654 return EPERM; 1655 1656 error = tmpfs_truncate(vp, size); 1657 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1658 * for us, as will update tn_status; no need to do that here. */ 1659 1660 MPASS(VOP_ISLOCKED(vp)); 1661 1662 return error; 1663 } 1664 1665 /* --------------------------------------------------------------------- */ 1666 1667 /* 1668 * Change access and modification times of the given vnode. 1669 * Caller should execute tmpfs_update on vp after a successful execution. 1670 * The vnode must be locked on entry and remain locked on exit. 1671 */ 1672 int 1673 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1674 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l) 1675 { 1676 int error; 1677 struct tmpfs_node *node; 1678 1679 MPASS(VOP_ISLOCKED(vp)); 1680 1681 node = VP_TO_TMPFS_NODE(vp); 1682 1683 /* Disallow this operation if the file system is mounted read-only. */ 1684 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1685 return EROFS; 1686 1687 /* Immutable or append-only files cannot be modified, either. */ 1688 if (node->tn_flags & (IMMUTABLE | APPEND)) 1689 return EPERM; 1690 1691 /* Determine if the user have proper privilege to update time. */ 1692 if (vaflags & VA_UTIMES_NULL) { 1693 error = VOP_ACCESS(vp, VADMIN, cred, l); 1694 if (error) 1695 error = VOP_ACCESS(vp, VWRITE, cred, l); 1696 } else 1697 error = VOP_ACCESS(vp, VADMIN, cred, l); 1698 if (error) 1699 return (error); 1700 1701 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1702 node->tn_status |= TMPFS_NODE_ACCESSED; 1703 1704 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) 1705 node->tn_status |= TMPFS_NODE_MODIFIED; 1706 1707 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL) 1708 node->tn_status |= TMPFS_NODE_MODIFIED; 1709 1710 tmpfs_itimes(vp, atime, mtime); 1711 1712 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL) 1713 node->tn_birthtime = *birthtime; 1714 MPASS(VOP_ISLOCKED(vp)); 1715 1716 return 0; 1717 } 1718 1719 /* --------------------------------------------------------------------- */ 1720 /* Sync timestamps */ 1721 void 1722 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1723 const struct timespec *mod) 1724 { 1725 struct tmpfs_node *node; 1726 struct timespec now; 1727 1728 node = VP_TO_TMPFS_NODE(vp); 1729 1730 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1731 TMPFS_NODE_CHANGED)) == 0) 1732 return; 1733 1734 vfs_timestamp(&now); 1735 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1736 if (acc == NULL) 1737 acc = &now; 1738 node->tn_atime = *acc; 1739 } 1740 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1741 if (mod == NULL) 1742 mod = &now; 1743 node->tn_mtime = *mod; 1744 } 1745 if (node->tn_status & TMPFS_NODE_CHANGED) { 1746 node->tn_ctime = now; 1747 } 1748 node->tn_status &= 1749 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1750 } 1751 1752 /* --------------------------------------------------------------------- */ 1753 1754 void 1755 tmpfs_update(struct vnode *vp) 1756 { 1757 1758 tmpfs_itimes(vp, NULL, NULL); 1759 } 1760 1761 /* --------------------------------------------------------------------- */ 1762 1763 int 1764 tmpfs_truncate(struct vnode *vp, off_t length) 1765 { 1766 int error; 1767 struct tmpfs_node *node; 1768 1769 node = VP_TO_TMPFS_NODE(vp); 1770 1771 if (length < 0) { 1772 error = EINVAL; 1773 goto out; 1774 } 1775 1776 if (node->tn_size == length) { 1777 error = 0; 1778 goto out; 1779 } 1780 1781 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1782 return (EFBIG); 1783 1784 error = tmpfs_reg_resize(vp, length, FALSE); 1785 if (error == 0) { 1786 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1787 } 1788 1789 out: 1790 tmpfs_update(vp); 1791 1792 return error; 1793 } 1794 1795 static __inline int 1796 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1797 { 1798 if (a->td_hash > b->td_hash) 1799 return (1); 1800 else if (a->td_hash < b->td_hash) 1801 return (-1); 1802 return (0); 1803 } 1804 1805 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1806