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 if ((vp = node->tn_vnode) != NULL) { 483 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 484 VI_LOCK(vp); 485 TMPFS_NODE_UNLOCK(node); 486 error = vget(vp, lkflag | LK_INTERLOCK, curthread); 487 if (error != 0) { 488 vp = NULL; 489 goto out; 490 } 491 492 /* 493 * Make sure the vnode is still there after 494 * getting the interlock to avoid racing a free. 495 */ 496 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 497 vput(vp); 498 goto loop; 499 } 500 501 goto out; 502 } 503 504 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 505 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 506 TMPFS_NODE_UNLOCK(node); 507 error = ENOENT; 508 vp = NULL; 509 goto out; 510 } 511 512 /* 513 * otherwise lock the vp list while we call getnewvnode 514 * since that can block. 515 */ 516 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 517 node->tn_vpstate |= TMPFS_VNODE_WANT; 518 error = msleep((caddr_t) &node->tn_vpstate, 519 TMPFS_NODE_MTX(node), PDROP | PCATCH, 520 "tmpfs_alloc_vp", 0); 521 if (error) 522 return error; 523 524 goto loop; 525 } else 526 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 527 528 TMPFS_NODE_UNLOCK(node); 529 530 /* Get a new vnode and associate it with our node. */ 531 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); 532 if (error != 0) 533 goto unlock; 534 MPASS(vp != NULL); 535 536 (void) vn_lock(vp, lkflag | LK_RETRY); 537 538 vp->v_data = node; 539 vp->v_type = node->tn_type; 540 541 /* Type-specific initialization. */ 542 switch (node->tn_type) { 543 case VBLK: 544 /* FALLTHROUGH */ 545 case VCHR: 546 /* FALLTHROUGH */ 547 case VLNK: 548 /* FALLTHROUGH */ 549 case VSOCK: 550 break; 551 case VFIFO: 552 vp->v_op = &tmpfs_fifoop_entries; 553 break; 554 case VREG: 555 object = node->tn_reg.tn_aobj; 556 VM_OBJECT_WLOCK(object); 557 VI_LOCK(vp); 558 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 559 vp->v_object = object; 560 object->un_pager.swp.swp_tmpfs = vp; 561 vm_object_set_flag(object, OBJ_TMPFS); 562 VI_UNLOCK(vp); 563 VM_OBJECT_WUNLOCK(object); 564 break; 565 case VDIR: 566 MPASS(node->tn_dir.tn_parent != NULL); 567 if (node->tn_dir.tn_parent == node) 568 vp->v_vflag |= VV_ROOT; 569 break; 570 571 default: 572 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 573 } 574 575 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 576 if (error) 577 vp = NULL; 578 579 unlock: 580 TMPFS_NODE_LOCK(node); 581 582 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 583 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 584 node->tn_vnode = vp; 585 586 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 587 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 588 TMPFS_NODE_UNLOCK(node); 589 wakeup((caddr_t) &node->tn_vpstate); 590 } else 591 TMPFS_NODE_UNLOCK(node); 592 593 out: 594 *vpp = vp; 595 596 #ifdef INVARIANTS 597 if (error == 0) { 598 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 599 TMPFS_NODE_LOCK(node); 600 MPASS(*vpp == node->tn_vnode); 601 TMPFS_NODE_UNLOCK(node); 602 } 603 #endif 604 605 return error; 606 } 607 608 /* --------------------------------------------------------------------- */ 609 610 /* 611 * Destroys the association between the vnode vp and the node it 612 * references. 613 */ 614 void 615 tmpfs_free_vp(struct vnode *vp) 616 { 617 struct tmpfs_node *node; 618 619 node = VP_TO_TMPFS_NODE(vp); 620 621 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED); 622 node->tn_vnode = NULL; 623 vp->v_data = NULL; 624 } 625 626 /* --------------------------------------------------------------------- */ 627 628 /* 629 * Allocates a new file of type 'type' and adds it to the parent directory 630 * 'dvp'; this addition is done using the component name given in 'cnp'. 631 * The ownership of the new file is automatically assigned based on the 632 * credentials of the caller (through 'cnp'), the group is set based on 633 * the parent directory and the mode is determined from the 'vap' argument. 634 * If successful, *vpp holds a vnode to the newly created file and zero 635 * is returned. Otherwise *vpp is NULL and the function returns an 636 * appropriate error code. 637 */ 638 int 639 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 640 struct componentname *cnp, char *target) 641 { 642 int error; 643 struct tmpfs_dirent *de; 644 struct tmpfs_mount *tmp; 645 struct tmpfs_node *dnode; 646 struct tmpfs_node *node; 647 struct tmpfs_node *parent; 648 649 MPASS(VOP_ISLOCKED(dvp)); 650 MPASS(cnp->cn_flags & HASBUF); 651 652 tmp = VFS_TO_TMPFS(dvp->v_mount); 653 dnode = VP_TO_TMPFS_DIR(dvp); 654 *vpp = NULL; 655 656 /* If the entry we are creating is a directory, we cannot overflow 657 * the number of links of its parent, because it will get a new 658 * link. */ 659 if (vap->va_type == VDIR) { 660 /* Ensure that we do not overflow the maximum number of links 661 * imposed by the system. */ 662 MPASS(dnode->tn_links <= LINK_MAX); 663 if (dnode->tn_links == LINK_MAX) { 664 error = EMLINK; 665 goto out; 666 } 667 668 parent = dnode; 669 MPASS(parent != NULL); 670 } else 671 parent = NULL; 672 673 /* Allocate a node that represents the new file. */ 674 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid, 675 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node); 676 if (error != 0) 677 goto out; 678 679 /* Allocate a directory entry that points to the new file. */ 680 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 681 &de); 682 if (error != 0) { 683 tmpfs_free_node(tmp, node); 684 goto out; 685 } 686 687 /* Allocate a vnode for the new file. */ 688 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 689 if (error != 0) { 690 tmpfs_free_dirent(tmp, de); 691 tmpfs_free_node(tmp, node); 692 goto out; 693 } 694 695 /* Now that all required items are allocated, we can proceed to 696 * insert the new node into the directory, an operation that 697 * cannot fail. */ 698 if (cnp->cn_flags & ISWHITEOUT) 699 tmpfs_dir_whiteout_remove(dvp, cnp); 700 tmpfs_dir_attach(dvp, de); 701 702 out: 703 704 return error; 705 } 706 707 /* --------------------------------------------------------------------- */ 708 709 static struct tmpfs_dirent * 710 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 711 { 712 struct tmpfs_dirent *de; 713 714 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 715 dc->tdc_tree = de; 716 if (de != NULL && tmpfs_dirent_duphead(de)) 717 de = LIST_FIRST(&de->ud.td_duphead); 718 dc->tdc_current = de; 719 720 return (dc->tdc_current); 721 } 722 723 static struct tmpfs_dirent * 724 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 725 { 726 struct tmpfs_dirent *de; 727 728 MPASS(dc->tdc_tree != NULL); 729 if (tmpfs_dirent_dup(dc->tdc_current)) { 730 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 731 if (dc->tdc_current != NULL) 732 return (dc->tdc_current); 733 } 734 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 735 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 736 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 737 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 738 MPASS(dc->tdc_current != NULL); 739 } 740 741 return (dc->tdc_current); 742 } 743 744 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 745 static struct tmpfs_dirent * 746 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 747 { 748 struct tmpfs_dirent *de, dekey; 749 750 dekey.td_hash = hash; 751 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 752 return (de); 753 } 754 755 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 756 static struct tmpfs_dirent * 757 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 758 struct tmpfs_dir_cursor *dc) 759 { 760 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 761 struct tmpfs_dirent *de, dekey; 762 763 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 764 765 if (cookie == node->tn_dir.tn_readdir_lastn && 766 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 767 /* Protect against possible race, tn_readdir_last[pn] 768 * may be updated with only shared vnode lock held. */ 769 if (cookie == tmpfs_dirent_cookie(de)) 770 goto out; 771 } 772 773 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 774 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 775 uh.td_dup.index_entries) { 776 MPASS(tmpfs_dirent_dup(de)); 777 if (de->td_cookie == cookie) 778 goto out; 779 /* dupindex list is sorted. */ 780 if (de->td_cookie < cookie) { 781 de = NULL; 782 goto out; 783 } 784 } 785 MPASS(de == NULL); 786 goto out; 787 } 788 789 MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie); 790 dekey.td_hash = cookie; 791 /* Recover if direntry for cookie was removed */ 792 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 793 dc->tdc_tree = de; 794 dc->tdc_current = de; 795 if (de != NULL && tmpfs_dirent_duphead(de)) { 796 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 797 MPASS(dc->tdc_current != NULL); 798 } 799 return (dc->tdc_current); 800 801 out: 802 dc->tdc_tree = de; 803 dc->tdc_current = de; 804 if (de != NULL && tmpfs_dirent_dup(de)) 805 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 806 de->td_hash); 807 return (dc->tdc_current); 808 } 809 810 /* 811 * Looks for a directory entry in the directory represented by node. 812 * 'cnp' describes the name of the entry to look for. Note that the . 813 * and .. components are not allowed as they do not physically exist 814 * within directories. 815 * 816 * Returns a pointer to the entry when found, otherwise NULL. 817 */ 818 struct tmpfs_dirent * 819 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 820 struct componentname *cnp) 821 { 822 struct tmpfs_dir_duphead *duphead; 823 struct tmpfs_dirent *de; 824 uint32_t hash; 825 826 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 827 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 828 cnp->cn_nameptr[1] == '.'))); 829 TMPFS_VALIDATE_DIR(node); 830 831 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 832 de = tmpfs_dir_xlookup_hash(node, hash); 833 if (de != NULL && tmpfs_dirent_duphead(de)) { 834 duphead = &de->ud.td_duphead; 835 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 836 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 837 cnp->cn_namelen)) 838 break; 839 } 840 } else if (de != NULL) { 841 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 842 cnp->cn_namelen)) 843 de = NULL; 844 } 845 if (de != NULL && f != NULL && de->td_node != f) 846 de = NULL; 847 848 return (de); 849 } 850 851 /* 852 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 853 * list, allocate new cookie value. 854 */ 855 static void 856 tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 857 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 858 { 859 struct tmpfs_dir_duphead *dupindex; 860 struct tmpfs_dirent *de, *pde; 861 862 dupindex = &dnode->tn_dir.tn_dupindex; 863 de = LIST_FIRST(dupindex); 864 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 865 if (de == NULL) 866 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 867 else 868 nde->td_cookie = de->td_cookie + 1; 869 MPASS(tmpfs_dirent_dup(nde)); 870 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 871 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 872 return; 873 } 874 875 /* 876 * Cookie numbers are near exhaustion. Scan dupindex list for unused 877 * numbers. dupindex list is sorted in descending order. Keep it so 878 * after inserting nde. 879 */ 880 while (1) { 881 pde = de; 882 de = LIST_NEXT(de, uh.td_dup.index_entries); 883 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 884 /* 885 * Last element of the index doesn't have minimal cookie 886 * value, use it. 887 */ 888 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 889 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 890 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 891 return; 892 } else if (de == NULL) { 893 /* 894 * We are so lucky have 2^30 hash duplicates in single 895 * directory :) Return largest possible cookie value. 896 * It should be fine except possible issues with 897 * VOP_READDIR restart. 898 */ 899 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 900 LIST_INSERT_HEAD(dupindex, nde, 901 uh.td_dup.index_entries); 902 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 903 return; 904 } 905 if (de->td_cookie + 1 == pde->td_cookie || 906 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 907 continue; /* No hole or invalid cookie. */ 908 nde->td_cookie = de->td_cookie + 1; 909 MPASS(tmpfs_dirent_dup(nde)); 910 MPASS(pde->td_cookie > nde->td_cookie); 911 MPASS(nde->td_cookie > de->td_cookie); 912 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 913 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 914 return; 915 }; 916 } 917 918 /* 919 * Attaches the directory entry de to the directory represented by vp. 920 * Note that this does not change the link count of the node pointed by 921 * the directory entry, as this is done by tmpfs_alloc_dirent. 922 */ 923 void 924 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 925 { 926 struct tmpfs_node *dnode; 927 struct tmpfs_dirent *xde, *nde; 928 929 ASSERT_VOP_ELOCKED(vp, __func__); 930 MPASS(de->td_namelen > 0); 931 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 932 MPASS(de->td_cookie == de->td_hash); 933 934 dnode = VP_TO_TMPFS_DIR(vp); 935 dnode->tn_dir.tn_readdir_lastn = 0; 936 dnode->tn_dir.tn_readdir_lastp = NULL; 937 938 MPASS(!tmpfs_dirent_dup(de)); 939 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 940 if (xde != NULL && tmpfs_dirent_duphead(xde)) 941 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 942 else if (xde != NULL) { 943 /* 944 * Allocate new duphead. Swap xde with duphead to avoid 945 * adding/removing elements with the same hash. 946 */ 947 MPASS(!tmpfs_dirent_dup(xde)); 948 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 949 &nde); 950 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 951 memcpy(nde, xde, sizeof(*xde)); 952 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 953 LIST_INIT(&xde->ud.td_duphead); 954 xde->td_namelen = 0; 955 xde->td_node = NULL; 956 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 957 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 958 } 959 dnode->tn_size += sizeof(struct tmpfs_dirent); 960 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 961 TMPFS_NODE_MODIFIED; 962 } 963 964 /* --------------------------------------------------------------------- */ 965 966 /* 967 * Detaches the directory entry de from the directory represented by vp. 968 * Note that this does not change the link count of the node pointed by 969 * the directory entry, as this is done by tmpfs_free_dirent. 970 */ 971 void 972 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 973 { 974 struct tmpfs_mount *tmp; 975 struct tmpfs_dir *head; 976 struct tmpfs_node *dnode; 977 struct tmpfs_dirent *xde; 978 979 ASSERT_VOP_ELOCKED(vp, __func__); 980 981 dnode = VP_TO_TMPFS_DIR(vp); 982 head = &dnode->tn_dir.tn_dirhead; 983 dnode->tn_dir.tn_readdir_lastn = 0; 984 dnode->tn_dir.tn_readdir_lastp = NULL; 985 986 if (tmpfs_dirent_dup(de)) { 987 /* Remove duphead if de was last entry. */ 988 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 989 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 990 MPASS(tmpfs_dirent_duphead(xde)); 991 } else 992 xde = NULL; 993 LIST_REMOVE(de, uh.td_dup.entries); 994 LIST_REMOVE(de, uh.td_dup.index_entries); 995 if (xde != NULL) { 996 if (LIST_EMPTY(&xde->ud.td_duphead)) { 997 RB_REMOVE(tmpfs_dir, head, xde); 998 tmp = VFS_TO_TMPFS(vp->v_mount); 999 MPASS(xde->td_node == NULL); 1000 tmpfs_free_dirent(tmp, xde); 1001 } 1002 } 1003 } else 1004 RB_REMOVE(tmpfs_dir, head, de); 1005 1006 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1007 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1008 TMPFS_NODE_MODIFIED; 1009 } 1010 1011 void 1012 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1013 { 1014 struct tmpfs_dirent *de, *dde, *nde; 1015 1016 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1017 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1018 /* Node may already be destroyed. */ 1019 de->td_node = NULL; 1020 if (tmpfs_dirent_duphead(de)) { 1021 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1022 LIST_REMOVE(dde, uh.td_dup.entries); 1023 dde->td_node = NULL; 1024 tmpfs_free_dirent(tmp, dde); 1025 } 1026 } 1027 tmpfs_free_dirent(tmp, de); 1028 } 1029 } 1030 1031 /* --------------------------------------------------------------------- */ 1032 1033 /* 1034 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1035 * directory and returns it in the uio space. The function returns 0 1036 * on success, -1 if there was not enough space in the uio structure to 1037 * hold the directory entry or an appropriate error code if another 1038 * error happens. 1039 */ 1040 static int 1041 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1042 { 1043 int error; 1044 struct dirent dent; 1045 1046 TMPFS_VALIDATE_DIR(node); 1047 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1048 1049 dent.d_fileno = node->tn_id; 1050 dent.d_type = DT_DIR; 1051 dent.d_namlen = 1; 1052 dent.d_name[0] = '.'; 1053 dent.d_name[1] = '\0'; 1054 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1055 1056 if (dent.d_reclen > uio->uio_resid) 1057 error = EJUSTRETURN; 1058 else 1059 error = uiomove(&dent, dent.d_reclen, uio); 1060 1061 node->tn_status |= TMPFS_NODE_ACCESSED; 1062 1063 return error; 1064 } 1065 1066 /* --------------------------------------------------------------------- */ 1067 1068 /* 1069 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1070 * directory and returns it in the uio space. The function returns 0 1071 * on success, -1 if there was not enough space in the uio structure to 1072 * hold the directory entry or an appropriate error code if another 1073 * error happens. 1074 */ 1075 static int 1076 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1077 { 1078 int error; 1079 struct dirent dent; 1080 1081 TMPFS_VALIDATE_DIR(node); 1082 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1083 1084 /* 1085 * Return ENOENT if the current node is already removed. 1086 */ 1087 TMPFS_ASSERT_LOCKED(node); 1088 if (node->tn_dir.tn_parent == NULL) { 1089 return (ENOENT); 1090 } 1091 1092 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1093 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1094 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1095 1096 dent.d_type = DT_DIR; 1097 dent.d_namlen = 2; 1098 dent.d_name[0] = '.'; 1099 dent.d_name[1] = '.'; 1100 dent.d_name[2] = '\0'; 1101 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1102 1103 if (dent.d_reclen > uio->uio_resid) 1104 error = EJUSTRETURN; 1105 else 1106 error = uiomove(&dent, dent.d_reclen, uio); 1107 1108 node->tn_status |= TMPFS_NODE_ACCESSED; 1109 1110 return error; 1111 } 1112 1113 /* --------------------------------------------------------------------- */ 1114 1115 /* 1116 * Helper function for tmpfs_readdir. Returns as much directory entries 1117 * as can fit in the uio space. The read starts at uio->uio_offset. 1118 * The function returns 0 on success, -1 if there was not enough space 1119 * in the uio structure to hold the directory entry or an appropriate 1120 * error code if another error happens. 1121 */ 1122 int 1123 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int cnt, 1124 u_long *cookies, int *ncookies) 1125 { 1126 struct tmpfs_dir_cursor dc; 1127 struct tmpfs_dirent *de; 1128 off_t off; 1129 int error; 1130 1131 TMPFS_VALIDATE_DIR(node); 1132 1133 off = 0; 1134 switch (uio->uio_offset) { 1135 case TMPFS_DIRCOOKIE_DOT: 1136 error = tmpfs_dir_getdotdent(node, uio); 1137 if (error != 0) 1138 return (error); 1139 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1140 if (cnt != 0) 1141 cookies[(*ncookies)++] = off = uio->uio_offset; 1142 case TMPFS_DIRCOOKIE_DOTDOT: 1143 error = tmpfs_dir_getdotdotdent(node, uio); 1144 if (error != 0) 1145 return (error); 1146 de = tmpfs_dir_first(node, &dc); 1147 if (de == NULL) 1148 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 1149 else 1150 uio->uio_offset = tmpfs_dirent_cookie(de); 1151 if (cnt != 0) 1152 cookies[(*ncookies)++] = off = uio->uio_offset; 1153 if (de == NULL) 1154 return (0); 1155 break; 1156 case TMPFS_DIRCOOKIE_EOF: 1157 return (0); 1158 default: 1159 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1160 if (de == NULL) 1161 return (EINVAL); 1162 if (cnt != 0) 1163 off = tmpfs_dirent_cookie(de); 1164 } 1165 1166 /* Read as much entries as possible; i.e., until we reach the end of 1167 * the directory or we exhaust uio space. */ 1168 do { 1169 struct dirent d; 1170 1171 /* Create a dirent structure representing the current 1172 * tmpfs_node and fill it. */ 1173 if (de->td_node == NULL) { 1174 d.d_fileno = 1; 1175 d.d_type = DT_WHT; 1176 } else { 1177 d.d_fileno = de->td_node->tn_id; 1178 switch (de->td_node->tn_type) { 1179 case VBLK: 1180 d.d_type = DT_BLK; 1181 break; 1182 1183 case VCHR: 1184 d.d_type = DT_CHR; 1185 break; 1186 1187 case VDIR: 1188 d.d_type = DT_DIR; 1189 break; 1190 1191 case VFIFO: 1192 d.d_type = DT_FIFO; 1193 break; 1194 1195 case VLNK: 1196 d.d_type = DT_LNK; 1197 break; 1198 1199 case VREG: 1200 d.d_type = DT_REG; 1201 break; 1202 1203 case VSOCK: 1204 d.d_type = DT_SOCK; 1205 break; 1206 1207 default: 1208 panic("tmpfs_dir_getdents: type %p %d", 1209 de->td_node, (int)de->td_node->tn_type); 1210 } 1211 } 1212 d.d_namlen = de->td_namelen; 1213 MPASS(de->td_namelen < sizeof(d.d_name)); 1214 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1215 d.d_name[de->td_namelen] = '\0'; 1216 d.d_reclen = GENERIC_DIRSIZ(&d); 1217 1218 /* Stop reading if the directory entry we are treating is 1219 * bigger than the amount of data that can be returned. */ 1220 if (d.d_reclen > uio->uio_resid) { 1221 error = EJUSTRETURN; 1222 break; 1223 } 1224 1225 /* Copy the new dirent structure into the output buffer and 1226 * advance pointers. */ 1227 error = uiomove(&d, d.d_reclen, uio); 1228 if (error == 0) { 1229 de = tmpfs_dir_next(node, &dc); 1230 if (cnt != 0) { 1231 if (de == NULL) 1232 off = TMPFS_DIRCOOKIE_EOF; 1233 else 1234 off = tmpfs_dirent_cookie(de); 1235 MPASS(*ncookies < cnt); 1236 cookies[(*ncookies)++] = off; 1237 } 1238 } 1239 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1240 1241 /* Update the offset and cache. */ 1242 if (cnt == 0) { 1243 if (de == NULL) 1244 off = TMPFS_DIRCOOKIE_EOF; 1245 else 1246 off = tmpfs_dirent_cookie(de); 1247 } 1248 1249 uio->uio_offset = off; 1250 node->tn_dir.tn_readdir_lastn = off; 1251 node->tn_dir.tn_readdir_lastp = de; 1252 1253 node->tn_status |= TMPFS_NODE_ACCESSED; 1254 return error; 1255 } 1256 1257 int 1258 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1259 { 1260 struct tmpfs_dirent *de; 1261 int error; 1262 1263 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1264 cnp->cn_nameptr, cnp->cn_namelen, &de); 1265 if (error != 0) 1266 return (error); 1267 tmpfs_dir_attach(dvp, de); 1268 return (0); 1269 } 1270 1271 void 1272 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1273 { 1274 struct tmpfs_dirent *de; 1275 1276 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1277 MPASS(de != NULL && de->td_node == NULL); 1278 tmpfs_dir_detach(dvp, de); 1279 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1280 } 1281 1282 /* --------------------------------------------------------------------- */ 1283 1284 /* 1285 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1286 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1287 * 'newsize' must be positive. 1288 * 1289 * Returns zero on success or an appropriate error code on failure. 1290 */ 1291 int 1292 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1293 { 1294 struct tmpfs_mount *tmp; 1295 struct tmpfs_node *node; 1296 vm_object_t uobj; 1297 vm_page_t m, ma[1]; 1298 vm_pindex_t idx, newpages, oldpages; 1299 off_t oldsize; 1300 int base, rv; 1301 1302 MPASS(vp->v_type == VREG); 1303 MPASS(newsize >= 0); 1304 1305 node = VP_TO_TMPFS_NODE(vp); 1306 uobj = node->tn_reg.tn_aobj; 1307 tmp = VFS_TO_TMPFS(vp->v_mount); 1308 1309 /* 1310 * Convert the old and new sizes to the number of pages needed to 1311 * store them. It may happen that we do not need to do anything 1312 * because the last allocated page can accommodate the change on 1313 * its own. 1314 */ 1315 oldsize = node->tn_size; 1316 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1317 MPASS(oldpages == uobj->size); 1318 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1319 if (newpages > oldpages && 1320 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1321 return (ENOSPC); 1322 1323 VM_OBJECT_WLOCK(uobj); 1324 if (newsize < oldsize) { 1325 /* 1326 * Zero the truncated part of the last page. 1327 */ 1328 base = newsize & PAGE_MASK; 1329 if (base != 0) { 1330 idx = OFF_TO_IDX(newsize); 1331 retry: 1332 m = vm_page_lookup(uobj, idx); 1333 if (m != NULL) { 1334 if ((m->oflags & VPO_BUSY) != 0 || 1335 m->busy != 0) { 1336 vm_page_sleep(m, "tmfssz"); 1337 goto retry; 1338 } 1339 MPASS(m->valid == VM_PAGE_BITS_ALL); 1340 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1341 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1342 if (m == NULL) { 1343 VM_OBJECT_WUNLOCK(uobj); 1344 VM_WAIT; 1345 VM_OBJECT_WLOCK(uobj); 1346 goto retry; 1347 } else if (m->valid != VM_PAGE_BITS_ALL) { 1348 ma[0] = m; 1349 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1350 m = vm_page_lookup(uobj, idx); 1351 } else 1352 /* A cached page was reactivated. */ 1353 rv = VM_PAGER_OK; 1354 vm_page_lock(m); 1355 if (rv == VM_PAGER_OK) { 1356 vm_page_deactivate(m); 1357 vm_page_unlock(m); 1358 vm_page_wakeup(m); 1359 } else { 1360 vm_page_free(m); 1361 vm_page_unlock(m); 1362 if (ignerr) 1363 m = NULL; 1364 else { 1365 VM_OBJECT_WUNLOCK(uobj); 1366 return (EIO); 1367 } 1368 } 1369 } 1370 if (m != NULL) { 1371 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1372 vm_page_dirty(m); 1373 vm_pager_page_unswapped(m); 1374 } 1375 } 1376 1377 /* 1378 * Release any swap space and free any whole pages. 1379 */ 1380 if (newpages < oldpages) { 1381 swap_pager_freespace(uobj, newpages, oldpages - 1382 newpages); 1383 vm_object_page_remove(uobj, newpages, 0, 0); 1384 } 1385 } 1386 uobj->size = newpages; 1387 VM_OBJECT_WUNLOCK(uobj); 1388 1389 TMPFS_LOCK(tmp); 1390 tmp->tm_pages_used += (newpages - oldpages); 1391 TMPFS_UNLOCK(tmp); 1392 1393 node->tn_size = newsize; 1394 return (0); 1395 } 1396 1397 /* --------------------------------------------------------------------- */ 1398 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 */ 1404 int 1405 tmpfs_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 & ~(UF_NODUMP | UF_IMMUTABLE | UF_APPEND | UF_OPAQUE | 1416 UF_NOUNLINK | SF_ARCHIVED | SF_IMMUTABLE | SF_APPEND | 1417 SF_NOUNLINK)) != 0) 1418 return (EOPNOTSUPP); 1419 1420 /* Disallow this operation if the file system is mounted read-only. */ 1421 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1422 return EROFS; 1423 1424 /* 1425 * Callers may only modify the file flags on objects they 1426 * have VADMIN rights for. 1427 */ 1428 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1429 return (error); 1430 /* 1431 * Unprivileged processes are not permitted to unset system 1432 * flags, or modify flags if any system flags are set. 1433 */ 1434 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1435 if (node->tn_flags & 1436 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1437 error = securelevel_gt(cred, 0); 1438 if (error) 1439 return (error); 1440 } 1441 } else { 1442 if (node->tn_flags & 1443 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1444 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1445 return (EPERM); 1446 } 1447 node->tn_flags = flags; 1448 node->tn_status |= TMPFS_NODE_CHANGED; 1449 1450 MPASS(VOP_ISLOCKED(vp)); 1451 1452 return 0; 1453 } 1454 1455 /* --------------------------------------------------------------------- */ 1456 1457 /* 1458 * Change access mode on the given vnode. 1459 * Caller should execute tmpfs_update on vp after a successful execution. 1460 * The vnode must be locked on entry and remain locked on exit. 1461 */ 1462 int 1463 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1464 { 1465 int error; 1466 struct tmpfs_node *node; 1467 1468 MPASS(VOP_ISLOCKED(vp)); 1469 1470 node = VP_TO_TMPFS_NODE(vp); 1471 1472 /* Disallow this operation if the file system is mounted read-only. */ 1473 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1474 return EROFS; 1475 1476 /* Immutable or append-only files cannot be modified, either. */ 1477 if (node->tn_flags & (IMMUTABLE | APPEND)) 1478 return EPERM; 1479 1480 /* 1481 * To modify the permissions on a file, must possess VADMIN 1482 * for that file. 1483 */ 1484 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1485 return (error); 1486 1487 /* 1488 * Privileged processes may set the sticky bit on non-directories, 1489 * as well as set the setgid bit on a file with a group that the 1490 * process is not a member of. 1491 */ 1492 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1493 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1494 return (EFTYPE); 1495 } 1496 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1497 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1498 if (error) 1499 return (error); 1500 } 1501 1502 1503 node->tn_mode &= ~ALLPERMS; 1504 node->tn_mode |= mode & ALLPERMS; 1505 1506 node->tn_status |= TMPFS_NODE_CHANGED; 1507 1508 MPASS(VOP_ISLOCKED(vp)); 1509 1510 return 0; 1511 } 1512 1513 /* --------------------------------------------------------------------- */ 1514 1515 /* 1516 * Change ownership of the given vnode. At least one of uid or gid must 1517 * be different than VNOVAL. If one is set to that value, the attribute 1518 * is unchanged. 1519 * Caller should execute tmpfs_update on vp after a successful execution. 1520 * The vnode must be locked on entry and remain locked on exit. 1521 */ 1522 int 1523 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1524 struct thread *p) 1525 { 1526 int error; 1527 struct tmpfs_node *node; 1528 uid_t ouid; 1529 gid_t ogid; 1530 1531 MPASS(VOP_ISLOCKED(vp)); 1532 1533 node = VP_TO_TMPFS_NODE(vp); 1534 1535 /* Assign default values if they are unknown. */ 1536 MPASS(uid != VNOVAL || gid != VNOVAL); 1537 if (uid == VNOVAL) 1538 uid = node->tn_uid; 1539 if (gid == VNOVAL) 1540 gid = node->tn_gid; 1541 MPASS(uid != VNOVAL && gid != VNOVAL); 1542 1543 /* Disallow this operation if the file system is mounted read-only. */ 1544 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1545 return EROFS; 1546 1547 /* Immutable or append-only files cannot be modified, either. */ 1548 if (node->tn_flags & (IMMUTABLE | APPEND)) 1549 return EPERM; 1550 1551 /* 1552 * To modify the ownership of a file, must possess VADMIN for that 1553 * file. 1554 */ 1555 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1556 return (error); 1557 1558 /* 1559 * To change the owner of a file, or change the group of a file to a 1560 * group of which we are not a member, the caller must have 1561 * privilege. 1562 */ 1563 if ((uid != node->tn_uid || 1564 (gid != node->tn_gid && !groupmember(gid, cred))) && 1565 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1566 return (error); 1567 1568 ogid = node->tn_gid; 1569 ouid = node->tn_uid; 1570 1571 node->tn_uid = uid; 1572 node->tn_gid = gid; 1573 1574 node->tn_status |= TMPFS_NODE_CHANGED; 1575 1576 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1577 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1578 node->tn_mode &= ~(S_ISUID | S_ISGID); 1579 } 1580 1581 MPASS(VOP_ISLOCKED(vp)); 1582 1583 return 0; 1584 } 1585 1586 /* --------------------------------------------------------------------- */ 1587 1588 /* 1589 * Change size of the given vnode. 1590 * Caller should execute tmpfs_update on vp after a successful execution. 1591 * The vnode must be locked on entry and remain locked on exit. 1592 */ 1593 int 1594 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1595 struct thread *p) 1596 { 1597 int error; 1598 struct tmpfs_node *node; 1599 1600 MPASS(VOP_ISLOCKED(vp)); 1601 1602 node = VP_TO_TMPFS_NODE(vp); 1603 1604 /* Decide whether this is a valid operation based on the file type. */ 1605 error = 0; 1606 switch (vp->v_type) { 1607 case VDIR: 1608 return EISDIR; 1609 1610 case VREG: 1611 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1612 return EROFS; 1613 break; 1614 1615 case VBLK: 1616 /* FALLTHROUGH */ 1617 case VCHR: 1618 /* FALLTHROUGH */ 1619 case VFIFO: 1620 /* Allow modifications of special files even if in the file 1621 * system is mounted read-only (we are not modifying the 1622 * files themselves, but the objects they represent). */ 1623 return 0; 1624 1625 default: 1626 /* Anything else is unsupported. */ 1627 return EOPNOTSUPP; 1628 } 1629 1630 /* Immutable or append-only files cannot be modified, either. */ 1631 if (node->tn_flags & (IMMUTABLE | APPEND)) 1632 return EPERM; 1633 1634 error = tmpfs_truncate(vp, size); 1635 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1636 * for us, as will update tn_status; no need to do that here. */ 1637 1638 MPASS(VOP_ISLOCKED(vp)); 1639 1640 return error; 1641 } 1642 1643 /* --------------------------------------------------------------------- */ 1644 1645 /* 1646 * Change access and modification times of the given vnode. 1647 * Caller should execute tmpfs_update on vp after a successful execution. 1648 * The vnode must be locked on entry and remain locked on exit. 1649 */ 1650 int 1651 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1652 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l) 1653 { 1654 int error; 1655 struct tmpfs_node *node; 1656 1657 MPASS(VOP_ISLOCKED(vp)); 1658 1659 node = VP_TO_TMPFS_NODE(vp); 1660 1661 /* Disallow this operation if the file system is mounted read-only. */ 1662 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1663 return EROFS; 1664 1665 /* Immutable or append-only files cannot be modified, either. */ 1666 if (node->tn_flags & (IMMUTABLE | APPEND)) 1667 return EPERM; 1668 1669 /* Determine if the user have proper privilege to update time. */ 1670 if (vaflags & VA_UTIMES_NULL) { 1671 error = VOP_ACCESS(vp, VADMIN, cred, l); 1672 if (error) 1673 error = VOP_ACCESS(vp, VWRITE, cred, l); 1674 } else 1675 error = VOP_ACCESS(vp, VADMIN, cred, l); 1676 if (error) 1677 return (error); 1678 1679 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1680 node->tn_status |= TMPFS_NODE_ACCESSED; 1681 1682 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) 1683 node->tn_status |= TMPFS_NODE_MODIFIED; 1684 1685 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL) 1686 node->tn_status |= TMPFS_NODE_MODIFIED; 1687 1688 tmpfs_itimes(vp, atime, mtime); 1689 1690 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL) 1691 node->tn_birthtime = *birthtime; 1692 MPASS(VOP_ISLOCKED(vp)); 1693 1694 return 0; 1695 } 1696 1697 /* --------------------------------------------------------------------- */ 1698 /* Sync timestamps */ 1699 void 1700 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1701 const struct timespec *mod) 1702 { 1703 struct tmpfs_node *node; 1704 struct timespec now; 1705 1706 node = VP_TO_TMPFS_NODE(vp); 1707 1708 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1709 TMPFS_NODE_CHANGED)) == 0) 1710 return; 1711 1712 vfs_timestamp(&now); 1713 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1714 if (acc == NULL) 1715 acc = &now; 1716 node->tn_atime = *acc; 1717 } 1718 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1719 if (mod == NULL) 1720 mod = &now; 1721 node->tn_mtime = *mod; 1722 } 1723 if (node->tn_status & TMPFS_NODE_CHANGED) { 1724 node->tn_ctime = now; 1725 } 1726 node->tn_status &= 1727 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1728 } 1729 1730 /* --------------------------------------------------------------------- */ 1731 1732 void 1733 tmpfs_update(struct vnode *vp) 1734 { 1735 1736 tmpfs_itimes(vp, NULL, NULL); 1737 } 1738 1739 /* --------------------------------------------------------------------- */ 1740 1741 int 1742 tmpfs_truncate(struct vnode *vp, off_t length) 1743 { 1744 int error; 1745 struct tmpfs_node *node; 1746 1747 node = VP_TO_TMPFS_NODE(vp); 1748 1749 if (length < 0) { 1750 error = EINVAL; 1751 goto out; 1752 } 1753 1754 if (node->tn_size == length) { 1755 error = 0; 1756 goto out; 1757 } 1758 1759 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1760 return (EFBIG); 1761 1762 error = tmpfs_reg_resize(vp, length, FALSE); 1763 if (error == 0) { 1764 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1765 } 1766 1767 out: 1768 tmpfs_update(vp); 1769 1770 return error; 1771 } 1772 1773 static __inline int 1774 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1775 { 1776 if (a->td_hash > b->td_hash) 1777 return (1); 1778 else if (a->td_hash < b->td_hash) 1779 return (-1); 1780 return (0); 1781 } 1782 1783 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1784