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 37 #include <sys/kernel.h> 38 #include <sys/param.h> 39 #include <sys/namei.h> 40 #include <sys/priv.h> 41 #include <sys/proc.h> 42 #include <sys/stat.h> 43 #include <sys/systm.h> 44 #include <sys/vnode.h> 45 #include <sys/vmmeter.h> 46 47 #include <vm/vm.h> 48 #include <vm/vm_object.h> 49 #include <vm/vm_page.h> 50 #include <vm/vm_pager.h> 51 #include <vm/vm_extern.h> 52 53 #include <vfs/tmpfs/tmpfs.h> 54 #include <vfs/tmpfs/tmpfs_vnops.h> 55 56 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *); 57 58 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a, 59 struct tmpfs_dirent *b); 60 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare); 61 62 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, 63 struct tmpfs_dirent *b); 64 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent, 65 rb_cookienode, tmpfs_dirtree_compare_cookie); 66 67 68 /* --------------------------------------------------------------------- */ 69 70 /* 71 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 72 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 73 * using the credentials of the process 'p'. 74 * 75 * If the node type is set to 'VDIR', then the parent parameter must point 76 * to the parent directory of the node being created. It may only be NULL 77 * while allocating the root node. 78 * 79 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 80 * specifies the device the node represents. 81 * 82 * If the node type is set to 'VLNK', then the parameter target specifies 83 * the file name of the target file for the symbolic link that is being 84 * created. 85 * 86 * Note that new nodes are retrieved from the available list if it has 87 * items or, if it is empty, from the node pool as long as there is enough 88 * space to create them. 89 * 90 * Returns zero on success or an appropriate error code on failure. 91 */ 92 int 93 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 94 uid_t uid, gid_t gid, mode_t mode, 95 char *target, int rmajor, int rminor, 96 struct tmpfs_node **node) 97 { 98 struct tmpfs_node *nnode; 99 struct timespec ts; 100 udev_t rdev; 101 102 KKASSERT(IFF(type == VLNK, target != NULL)); 103 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL)); 104 105 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 106 return (ENOSPC); 107 108 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK); 109 if (nnode == NULL) 110 return (ENOSPC); 111 112 /* Generic initialization. */ 113 nnode->tn_type = type; 114 vfs_timestamp(&ts); 115 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime 116 = ts.tv_sec; 117 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec 118 = ts.tv_nsec; 119 nnode->tn_uid = uid; 120 nnode->tn_gid = gid; 121 nnode->tn_mode = mode; 122 nnode->tn_id = tmpfs_fetch_ino(tmp); 123 nnode->tn_advlock.init_done = 0; 124 KKASSERT(nnode->tn_links == 0); 125 126 /* Type-specific initialization. */ 127 switch (nnode->tn_type) { 128 case VBLK: 129 case VCHR: 130 rdev = makeudev(rmajor, rminor); 131 if (rdev == NOUDEV) { 132 objcache_put(tmp->tm_node_pool, nnode); 133 return(EINVAL); 134 } 135 nnode->tn_rdev = rdev; 136 break; 137 138 case VDIR: 139 RB_INIT(&nnode->tn_dir.tn_dirtree); 140 RB_INIT(&nnode->tn_dir.tn_cookietree); 141 nnode->tn_size = 0; 142 break; 143 144 case VFIFO: 145 /* FALLTHROUGH */ 146 case VSOCK: 147 break; 148 149 case VLNK: 150 nnode->tn_size = strlen(target); 151 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone, 152 M_WAITOK | M_NULLOK); 153 if (nnode->tn_link == NULL) { 154 objcache_put(tmp->tm_node_pool, nnode); 155 return (ENOSPC); 156 } 157 bcopy(target, nnode->tn_link, nnode->tn_size); 158 nnode->tn_link[nnode->tn_size] = '\0'; 159 break; 160 161 case VREG: 162 nnode->tn_reg.tn_aobj = 163 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0); 164 nnode->tn_reg.tn_aobj_pages = 0; 165 nnode->tn_size = 0; 166 vm_object_set_flag(nnode->tn_reg.tn_aobj, OBJ_NOPAGEIN); 167 break; 168 169 default: 170 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 171 } 172 173 TMPFS_NODE_LOCK(nnode); 174 TMPFS_LOCK(tmp); 175 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 176 tmp->tm_nodes_inuse++; 177 TMPFS_UNLOCK(tmp); 178 TMPFS_NODE_UNLOCK(nnode); 179 180 *node = nnode; 181 return 0; 182 } 183 184 /* --------------------------------------------------------------------- */ 185 186 /* 187 * Destroys the node pointed to by node from the file system 'tmp'. 188 * If the node does not belong to the given mount point, the results are 189 * unpredicted. 190 * 191 * If the node references a directory; no entries are allowed because 192 * their removal could need a recursive algorithm, something forbidden in 193 * kernel space. Furthermore, there is not need to provide such 194 * functionality (recursive removal) because the only primitives offered 195 * to the user are the removal of empty directories and the deletion of 196 * individual files. 197 * 198 * Note that nodes are not really deleted; in fact, when a node has been 199 * allocated, it cannot be deleted during the whole life of the file 200 * system. Instead, they are moved to the available list and remain there 201 * until reused. 202 * 203 * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it. 204 */ 205 void 206 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 207 { 208 vm_pindex_t pages = 0; 209 210 #ifdef INVARIANTS 211 TMPFS_ASSERT_ELOCKED(node); 212 KKASSERT(node->tn_vnode == NULL); 213 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 214 #endif 215 216 TMPFS_LOCK(tmp); 217 LIST_REMOVE(node, tn_entries); 218 tmp->tm_nodes_inuse--; 219 TMPFS_UNLOCK(tmp); 220 TMPFS_NODE_UNLOCK(node); /* Caller has this lock */ 221 222 switch (node->tn_type) { 223 case VNON: 224 /* Do not do anything. VNON is provided to let the 225 * allocation routine clean itself easily by avoiding 226 * duplicating code in it. */ 227 /* FALLTHROUGH */ 228 case VBLK: 229 /* FALLTHROUGH */ 230 case VCHR: 231 /* FALLTHROUGH */ 232 break; 233 case VDIR: 234 /* 235 * The parent link can be NULL if this is the root 236 * node or if it is a directory node that was rmdir'd. 237 * 238 * XXX what if node is a directory which still contains 239 * directory entries (e.g. due to a forced umount) ? 240 */ 241 node->tn_size = 0; 242 KKASSERT(node->tn_dir.tn_parent == NULL); 243 244 /* 245 * If the root node is being destroyed don't leave a 246 * dangling pointer in tmpfs_mount. 247 */ 248 if (node == tmp->tm_root) 249 tmp->tm_root = NULL; 250 break; 251 case VFIFO: 252 /* FALLTHROUGH */ 253 case VSOCK: 254 break; 255 256 case VLNK: 257 kfree(node->tn_link, tmp->tm_name_zone); 258 node->tn_link = NULL; 259 node->tn_size = 0; 260 break; 261 262 case VREG: 263 if (node->tn_reg.tn_aobj != NULL) 264 vm_object_deallocate(node->tn_reg.tn_aobj); 265 node->tn_reg.tn_aobj = NULL; 266 pages = node->tn_reg.tn_aobj_pages; 267 break; 268 269 default: 270 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 271 } 272 273 /* 274 * Clean up fields for the next allocation. The objcache only ctors 275 * new allocations. 276 */ 277 tmpfs_node_ctor(node, NULL, 0); 278 objcache_put(tmp->tm_node_pool, node); 279 /* node is now invalid */ 280 281 if (pages) 282 atomic_add_long(&tmp->tm_pages_used, -(long)pages); 283 } 284 285 /* --------------------------------------------------------------------- */ 286 287 /* 288 * Allocates a new directory entry for the node node with a name of name. 289 * The new directory entry is returned in *de. 290 * 291 * The link count of node is increased by one to reflect the new object 292 * referencing it. 293 * 294 * Returns zero on success or an appropriate error code on failure. 295 */ 296 int 297 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 298 const char *name, uint16_t len, struct tmpfs_dirent **de) 299 { 300 struct tmpfs_dirent *nde; 301 302 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK); 303 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK); 304 if (nde->td_name == NULL) { 305 objcache_put(tmp->tm_dirent_pool, nde); 306 *de = NULL; 307 return (ENOSPC); 308 } 309 nde->td_namelen = len; 310 bcopy(name, nde->td_name, len); 311 nde->td_name[len] = '\0'; 312 313 nde->td_node = node; 314 315 TMPFS_NODE_LOCK(node); 316 ++node->tn_links; 317 TMPFS_NODE_UNLOCK(node); 318 319 *de = nde; 320 321 return 0; 322 } 323 324 /* --------------------------------------------------------------------- */ 325 326 /* 327 * Frees a directory entry. It is the caller's responsibility to destroy 328 * the node referenced by it if needed. 329 * 330 * The link count of node is decreased by one to reflect the removal of an 331 * object that referenced it. This only happens if 'node_exists' is true; 332 * otherwise the function will not access the node referred to by the 333 * directory entry, as it may already have been released from the outside. 334 */ 335 void 336 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 337 { 338 struct tmpfs_node *node; 339 340 node = de->td_node; 341 342 TMPFS_NODE_LOCK(node); 343 TMPFS_ASSERT_ELOCKED(node); 344 KKASSERT(node->tn_links > 0); 345 node->tn_links--; 346 TMPFS_NODE_UNLOCK(node); 347 348 kfree(de->td_name, tmp->tm_name_zone); 349 de->td_namelen = 0; 350 de->td_name = NULL; 351 de->td_node = NULL; 352 objcache_put(tmp->tm_dirent_pool, de); 353 } 354 355 /* --------------------------------------------------------------------- */ 356 357 /* 358 * Allocates a new vnode for the node node or returns a new reference to 359 * an existing one if the node had already a vnode referencing it. The 360 * resulting locked vnode is returned in *vpp. 361 * 362 * Returns zero on success or an appropriate error code on failure. 363 */ 364 int 365 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 366 struct vnode **vpp) 367 { 368 int error = 0; 369 struct vnode *vp; 370 371 loop: 372 /* 373 * Interlocked extraction from node. This can race many things. 374 * We have to get a soft reference on the vnode while we hold 375 * the node locked, then acquire it properly and check for races. 376 */ 377 TMPFS_NODE_LOCK(node); 378 if ((vp = node->tn_vnode) != NULL) { 379 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 380 vhold(vp); 381 TMPFS_NODE_UNLOCK(node); 382 383 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) { 384 vdrop(vp); 385 goto loop; 386 } 387 if (node->tn_vnode != vp) { 388 vput(vp); 389 vdrop(vp); 390 goto loop; 391 } 392 vdrop(vp); 393 goto out; 394 } 395 /* vp is NULL */ 396 397 /* 398 * This should never happen. 399 */ 400 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) { 401 TMPFS_NODE_UNLOCK(node); 402 error = ENOENT; 403 goto out; 404 } 405 406 /* 407 * Interlock against other calls to tmpfs_alloc_vp() trying to 408 * allocate and assign a vp to node. 409 */ 410 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 411 node->tn_vpstate |= TMPFS_VNODE_WANT; 412 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH, 413 "tmpfs_alloc_vp", 0); 414 TMPFS_NODE_UNLOCK(node); 415 if (error) 416 return error; 417 goto loop; 418 } 419 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 420 TMPFS_NODE_UNLOCK(node); 421 422 /* 423 * Allocate a new vnode (may block). The ALLOCATING flag should 424 * prevent a race against someone else assigning node->tn_vnode. 425 */ 426 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE); 427 if (error != 0) 428 goto unlock; 429 430 KKASSERT(node->tn_vnode == NULL); 431 KKASSERT(vp != NULL); 432 vp->v_data = node; 433 vp->v_type = node->tn_type; 434 435 /* Type-specific initialization. */ 436 switch (node->tn_type) { 437 case VBLK: 438 /* FALLTHROUGH */ 439 case VCHR: 440 /* FALLTHROUGH */ 441 case VSOCK: 442 break; 443 case VREG: 444 vinitvmio(vp, node->tn_size, TMPFS_BLKSIZE, -1); 445 break; 446 case VLNK: 447 break; 448 case VFIFO: 449 vp->v_ops = &mp->mnt_vn_fifo_ops; 450 break; 451 case VDIR: 452 break; 453 454 default: 455 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 456 } 457 458 459 unlock: 460 TMPFS_NODE_LOCK(node); 461 462 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 463 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 464 node->tn_vnode = vp; 465 466 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 467 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 468 TMPFS_NODE_UNLOCK(node); 469 wakeup(&node->tn_vpstate); 470 } else { 471 TMPFS_NODE_UNLOCK(node); 472 } 473 474 out: 475 *vpp = vp; 476 477 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp))); 478 #ifdef INVARIANTS 479 TMPFS_NODE_LOCK(node); 480 KKASSERT(*vpp == node->tn_vnode); 481 TMPFS_NODE_UNLOCK(node); 482 #endif 483 484 return error; 485 } 486 487 /* --------------------------------------------------------------------- */ 488 489 /* 490 * Destroys the association between the vnode vp and the node it 491 * references. 492 */ 493 void 494 tmpfs_free_vp(struct vnode *vp) 495 { 496 struct tmpfs_node *node; 497 498 node = VP_TO_TMPFS_NODE(vp); 499 500 TMPFS_NODE_LOCK(node); 501 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0); 502 node->tn_vnode = NULL; 503 vp->v_data = NULL; 504 TMPFS_NODE_UNLOCK(node); 505 } 506 507 /* --------------------------------------------------------------------- */ 508 509 /* 510 * Allocates a new file of type 'type' and adds it to the parent directory 511 * 'dvp'; this addition is done using the component name given in 'cnp'. 512 * The ownership of the new file is automatically assigned based on the 513 * credentials of the caller (through 'cnp'), the group is set based on 514 * the parent directory and the mode is determined from the 'vap' argument. 515 * If successful, *vpp holds a vnode to the newly created file and zero 516 * is returned. Otherwise *vpp is NULL and the function returns an 517 * appropriate error code. 518 */ 519 int 520 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 521 struct namecache *ncp, struct ucred *cred, char *target) 522 { 523 int error; 524 struct tmpfs_dirent *de; 525 struct tmpfs_mount *tmp; 526 struct tmpfs_node *dnode; 527 struct tmpfs_node *node; 528 529 tmp = VFS_TO_TMPFS(dvp->v_mount); 530 dnode = VP_TO_TMPFS_DIR(dvp); 531 *vpp = NULL; 532 533 /* 534 * If the directory was removed but a process was CD'd into it, 535 * we do not allow any more file/dir creation within it. Otherwise 536 * we will lose track of it. 537 */ 538 KKASSERT(dnode->tn_type == VDIR); 539 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL) 540 return ENOENT; 541 542 /* 543 * Make sure the link count does not overflow. 544 */ 545 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX) 546 return EMLINK; 547 548 /* Allocate a node that represents the new file. */ 549 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid, 550 dnode->tn_gid, vap->va_mode, target, 551 vap->va_rmajor, vap->va_rminor, &node); 552 if (error != 0) 553 return error; 554 TMPFS_NODE_LOCK(node); 555 556 /* Allocate a directory entry that points to the new file. */ 557 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de); 558 if (error != 0) { 559 tmpfs_free_node(tmp, node); 560 /* eats node lock */ 561 return error; 562 } 563 564 /* Allocate a vnode for the new file. */ 565 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 566 if (error != 0) { 567 tmpfs_free_dirent(tmp, de); 568 tmpfs_free_node(tmp, node); 569 /* eats node lock */ 570 return error; 571 } 572 573 /* 574 * Now that all required items are allocated, we can proceed to 575 * insert the new node into the directory, an operation that 576 * cannot fail. 577 */ 578 tmpfs_dir_attach(dnode, de); 579 TMPFS_NODE_UNLOCK(node); 580 581 return error; 582 } 583 584 /* --------------------------------------------------------------------- */ 585 586 /* 587 * Attaches the directory entry de to the directory represented by dnode. 588 * Note that this does not change the link count of the node pointed by 589 * the directory entry, as this is done by tmpfs_alloc_dirent. 590 */ 591 void 592 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 593 { 594 struct tmpfs_node *node = de->td_node; 595 596 TMPFS_NODE_LOCK(dnode); 597 if (node && node->tn_type == VDIR) { 598 TMPFS_NODE_LOCK(node); 599 ++node->tn_links; 600 node->tn_status |= TMPFS_NODE_CHANGED; 601 node->tn_dir.tn_parent = dnode; 602 ++dnode->tn_links; 603 TMPFS_NODE_UNLOCK(node); 604 } 605 RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 606 RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 607 dnode->tn_size += sizeof(struct tmpfs_dirent); 608 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 609 TMPFS_NODE_MODIFIED; 610 TMPFS_NODE_UNLOCK(dnode); 611 } 612 613 /* --------------------------------------------------------------------- */ 614 615 /* 616 * Detaches the directory entry de from the directory represented by dnode. 617 * Note that this does not change the link count of the node pointed by 618 * the directory entry, as this is done by tmpfs_free_dirent. 619 */ 620 void 621 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 622 { 623 struct tmpfs_node *node = de->td_node; 624 625 TMPFS_NODE_LOCK(dnode); 626 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 627 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 628 dnode->tn_size -= sizeof(struct tmpfs_dirent); 629 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 630 TMPFS_NODE_MODIFIED; 631 TMPFS_NODE_UNLOCK(dnode); 632 633 /* 634 * Clean out the tn_parent pointer immediately when removing a 635 * directory. 636 * 637 * Removal of the parent linkage also cleans out the extra tn_links 638 * count we had on both node and dnode. 639 * 640 * node can be NULL (typ during a forced umount), in which case 641 * the mount code is dealing with the linkages from a linked list 642 * scan. 643 */ 644 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) { 645 TMPFS_NODE_LOCK(dnode); 646 TMPFS_NODE_LOCK(node); 647 KKASSERT(node->tn_dir.tn_parent == dnode); 648 dnode->tn_links--; 649 node->tn_links--; 650 node->tn_dir.tn_parent = NULL; 651 TMPFS_NODE_UNLOCK(node); 652 TMPFS_NODE_UNLOCK(dnode); 653 } 654 } 655 656 /* --------------------------------------------------------------------- */ 657 658 /* 659 * Looks for a directory entry in the directory represented by node. 660 * 'ncp' describes the name of the entry to look for. Note that the . 661 * and .. components are not allowed as they do not physically exist 662 * within directories. 663 * 664 * Returns a pointer to the entry when found, otherwise NULL. 665 * 666 * Caller must hold the node locked (shared ok) 667 */ 668 struct tmpfs_dirent * 669 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 670 struct namecache *ncp) 671 { 672 struct tmpfs_dirent *de; 673 int len = ncp->nc_nlen; 674 struct tmpfs_dirent wanted; 675 676 wanted.td_namelen = len; 677 wanted.td_name = ncp->nc_name; 678 679 TMPFS_VALIDATE_DIR(node); 680 681 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted); 682 683 KKASSERT(f == NULL || f == de->td_node); 684 685 return de; 686 } 687 688 /* --------------------------------------------------------------------- */ 689 690 /* 691 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 692 * directory and returns it in the uio space. The function returns 0 693 * on success, -1 if there was not enough space in the uio structure to 694 * hold the directory entry or an appropriate error code if another 695 * error happens. 696 */ 697 int 698 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 699 { 700 int error; 701 702 TMPFS_VALIDATE_DIR(node); 703 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 704 705 if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, ".")) 706 return -1; 707 if (error == 0) 708 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 709 return error; 710 } 711 712 /* --------------------------------------------------------------------- */ 713 714 /* 715 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 716 * directory and returns it in the uio space. The function returns 0 717 * on success, -1 if there was not enough space in the uio structure to 718 * hold the directory entry or an appropriate error code if another 719 * error happens. 720 */ 721 int 722 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 723 struct uio *uio) 724 { 725 int error; 726 ino_t d_ino; 727 728 TMPFS_VALIDATE_DIR(node); 729 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 730 731 if (node->tn_dir.tn_parent) { 732 TMPFS_NODE_LOCK(node); 733 if (node->tn_dir.tn_parent) 734 d_ino = node->tn_dir.tn_parent->tn_id; 735 else 736 d_ino = tmp->tm_root->tn_id; 737 TMPFS_NODE_UNLOCK(node); 738 } else { 739 d_ino = tmp->tm_root->tn_id; 740 } 741 742 if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, "..")) 743 return -1; 744 if (error == 0) { 745 struct tmpfs_dirent *de; 746 de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree); 747 if (de == NULL) 748 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 749 else 750 uio->uio_offset = tmpfs_dircookie(de); 751 } 752 return error; 753 } 754 755 /* --------------------------------------------------------------------- */ 756 757 /* 758 * Lookup a directory entry by its associated cookie. 759 * 760 * Must be called with the directory node locked (shared ok) 761 */ 762 struct lubycookie_info { 763 off_t cookie; 764 struct tmpfs_dirent *de; 765 }; 766 767 static int 768 lubycookie_cmp(struct tmpfs_dirent *de, void *arg) 769 { 770 struct lubycookie_info *info = arg; 771 off_t cookie = tmpfs_dircookie(de); 772 773 if (cookie < info->cookie) 774 return(-1); 775 if (cookie > info->cookie) 776 return(1); 777 return(0); 778 } 779 780 static int 781 lubycookie_callback(struct tmpfs_dirent *de, void *arg) 782 { 783 struct lubycookie_info *info = arg; 784 785 if (tmpfs_dircookie(de) == info->cookie) { 786 info->de = de; 787 return(-1); 788 } 789 return(0); 790 } 791 792 struct tmpfs_dirent * 793 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie) 794 { 795 struct lubycookie_info info; 796 797 info.cookie = cookie; 798 info.de = NULL; 799 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree, 800 lubycookie_cmp, lubycookie_callback, &info); 801 return (info.de); 802 } 803 804 /* --------------------------------------------------------------------- */ 805 806 /* 807 * Helper function for tmpfs_readdir. Returns as much directory entries 808 * as can fit in the uio space. The read starts at uio->uio_offset. 809 * The function returns 0 on success, -1 if there was not enough space 810 * in the uio structure to hold the directory entry or an appropriate 811 * error code if another error happens. 812 * 813 * Caller must hold the node locked (shared ok) 814 */ 815 int 816 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp) 817 { 818 int error; 819 off_t startcookie; 820 struct tmpfs_dirent *de; 821 822 TMPFS_VALIDATE_DIR(node); 823 824 /* 825 * Locate the first directory entry we have to return. We have cached 826 * the last readdir in the node, so use those values if appropriate. 827 * Otherwise do a linear scan to find the requested entry. 828 */ 829 startcookie = uio->uio_offset; 830 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT); 831 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT); 832 833 if (startcookie == TMPFS_DIRCOOKIE_EOF) 834 return 0; 835 836 de = tmpfs_dir_lookupbycookie(node, startcookie); 837 if (de == NULL) 838 return EINVAL; 839 840 /* 841 * Read as much entries as possible; i.e., until we reach the end of 842 * the directory or we exhaust uio space. 843 */ 844 do { 845 ino_t d_ino; 846 uint8_t d_type; 847 848 /* Create a dirent structure representing the current 849 * tmpfs_node and fill it. */ 850 d_ino = de->td_node->tn_id; 851 switch (de->td_node->tn_type) { 852 case VBLK: 853 d_type = DT_BLK; 854 break; 855 856 case VCHR: 857 d_type = DT_CHR; 858 break; 859 860 case VDIR: 861 d_type = DT_DIR; 862 break; 863 864 case VFIFO: 865 d_type = DT_FIFO; 866 break; 867 868 case VLNK: 869 d_type = DT_LNK; 870 break; 871 872 case VREG: 873 d_type = DT_REG; 874 break; 875 876 case VSOCK: 877 d_type = DT_SOCK; 878 break; 879 880 default: 881 panic("tmpfs_dir_getdents: type %p %d", 882 de->td_node, (int)de->td_node->tn_type); 883 } 884 KKASSERT(de->td_namelen < 256); /* 255 + 1 */ 885 886 if (vop_write_dirent(&error, uio, d_ino, d_type, 887 de->td_namelen, de->td_name)) { 888 error = -1; 889 break; 890 } 891 892 (*cntp)++; 893 de = RB_NEXT(tmpfs_dirtree_cookie, 894 node->tn_dir.tn_cookietree, de); 895 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 896 897 /* Update the offset and cache. */ 898 if (de == NULL) { 899 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 900 } else { 901 uio->uio_offset = tmpfs_dircookie(de); 902 } 903 904 return error; 905 } 906 907 /* --------------------------------------------------------------------- */ 908 909 /* 910 * Resizes the aobj associated to the regular file pointed to by vp to 911 * the size newsize. 'vp' must point to a vnode that represents a regular 912 * file. 'newsize' must be positive. 913 * 914 * pass trivial as 1 when buf content will be overwritten, otherwise set 0 915 * to be zero filled. 916 * 917 * Returns zero on success or an appropriate error code on failure. 918 * 919 * Caller must hold the node exclusively locked. 920 */ 921 int 922 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial) 923 { 924 int error; 925 vm_pindex_t newpages, oldpages; 926 struct tmpfs_mount *tmp; 927 struct tmpfs_node *node; 928 off_t oldsize; 929 930 #ifdef INVARIANTS 931 KKASSERT(vp->v_type == VREG); 932 KKASSERT(newsize >= 0); 933 #endif 934 935 node = VP_TO_TMPFS_NODE(vp); 936 tmp = VFS_TO_TMPFS(vp->v_mount); 937 938 /* 939 * Convert the old and new sizes to the number of pages needed to 940 * store them. It may happen that we do not need to do anything 941 * because the last allocated page can accommodate the change on 942 * its own. 943 */ 944 oldsize = node->tn_size; 945 oldpages = round_page64(oldsize) / PAGE_SIZE; 946 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages); 947 newpages = round_page64(newsize) / PAGE_SIZE; 948 949 if (newpages > oldpages && 950 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) { 951 error = ENOSPC; 952 goto out; 953 } 954 node->tn_reg.tn_aobj_pages = newpages; 955 node->tn_size = newsize; 956 957 if (newpages != oldpages) 958 atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages)); 959 960 /* 961 * When adjusting the vnode filesize and its VM object we must 962 * also adjust our backing VM object (aobj). The blocksize 963 * used must match the block sized we use for the buffer cache. 964 * 965 * The backing VM object contains no VM pages, only swap 966 * assignments. 967 */ 968 if (newsize < oldsize) { 969 vm_pindex_t osize; 970 vm_pindex_t nsize; 971 vm_object_t aobj; 972 973 error = nvtruncbuf(vp, newsize, TMPFS_BLKSIZE, -1, 0); 974 aobj = node->tn_reg.tn_aobj; 975 if (aobj) { 976 osize = aobj->size; 977 nsize = vp->v_object->size; 978 if (nsize < osize) { 979 aobj->size = osize; 980 swap_pager_freespace(aobj, nsize, 981 osize - nsize); 982 } 983 } 984 } else { 985 vm_object_t aobj; 986 987 error = nvextendbuf(vp, oldsize, newsize, 988 TMPFS_BLKSIZE, TMPFS_BLKSIZE, 989 -1, -1, trivial); 990 aobj = node->tn_reg.tn_aobj; 991 if (aobj) 992 aobj->size = vp->v_object->size; 993 } 994 995 out: 996 return error; 997 } 998 999 /* --------------------------------------------------------------------- */ 1000 1001 /* 1002 * Change flags of the given vnode. 1003 * Caller should execute tmpfs_update on vp after a successful execution. 1004 * The vnode must be locked on entry and remain locked on exit. 1005 */ 1006 int 1007 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred) 1008 { 1009 int error; 1010 struct tmpfs_node *node; 1011 int flags; 1012 1013 KKASSERT(vn_islocked(vp)); 1014 1015 node = VP_TO_TMPFS_NODE(vp); 1016 flags = node->tn_flags; 1017 1018 /* Disallow this operation if the file system is mounted read-only. */ 1019 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1020 return EROFS; 1021 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred); 1022 1023 /* Actually change the flags on the node itself */ 1024 if (error == 0) { 1025 TMPFS_NODE_LOCK(node); 1026 node->tn_flags = flags; 1027 node->tn_status |= TMPFS_NODE_CHANGED; 1028 TMPFS_NODE_UNLOCK(node); 1029 } 1030 1031 KKASSERT(vn_islocked(vp)); 1032 1033 return error; 1034 } 1035 1036 /* --------------------------------------------------------------------- */ 1037 1038 /* 1039 * Change access mode on the given vnode. 1040 * Caller should execute tmpfs_update on vp after a successful execution. 1041 * The vnode must be locked on entry and remain locked on exit. 1042 */ 1043 int 1044 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred) 1045 { 1046 struct tmpfs_node *node; 1047 mode_t cur_mode; 1048 int error; 1049 1050 KKASSERT(vn_islocked(vp)); 1051 1052 node = VP_TO_TMPFS_NODE(vp); 1053 1054 /* Disallow this operation if the file system is mounted read-only. */ 1055 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1056 return EROFS; 1057 1058 /* Immutable or append-only files cannot be modified, either. */ 1059 if (node->tn_flags & (IMMUTABLE | APPEND)) 1060 return EPERM; 1061 1062 cur_mode = node->tn_mode; 1063 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid, 1064 &cur_mode); 1065 1066 if (error == 0 && 1067 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) { 1068 TMPFS_NODE_LOCK(node); 1069 node->tn_mode &= ~ALLPERMS; 1070 node->tn_mode |= cur_mode & ALLPERMS; 1071 1072 node->tn_status |= TMPFS_NODE_CHANGED; 1073 TMPFS_NODE_UNLOCK(node); 1074 } 1075 1076 KKASSERT(vn_islocked(vp)); 1077 1078 return 0; 1079 } 1080 1081 /* --------------------------------------------------------------------- */ 1082 1083 /* 1084 * Change ownership of the given vnode. At least one of uid or gid must 1085 * be different than VNOVAL. If one is set to that value, the attribute 1086 * is unchanged. 1087 * Caller should execute tmpfs_update on vp after a successful execution. 1088 * The vnode must be locked on entry and remain locked on exit. 1089 */ 1090 int 1091 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 1092 { 1093 mode_t cur_mode; 1094 uid_t cur_uid; 1095 gid_t cur_gid; 1096 struct tmpfs_node *node; 1097 int error; 1098 1099 KKASSERT(vn_islocked(vp)); 1100 node = VP_TO_TMPFS_NODE(vp); 1101 1102 /* Disallow this operation if the file system is mounted read-only. */ 1103 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1104 return EROFS; 1105 1106 /* Immutable or append-only files cannot be modified, either. */ 1107 if (node->tn_flags & (IMMUTABLE | APPEND)) 1108 return EPERM; 1109 1110 cur_uid = node->tn_uid; 1111 cur_gid = node->tn_gid; 1112 cur_mode = node->tn_mode; 1113 error = vop_helper_chown(vp, uid, gid, cred, 1114 &cur_uid, &cur_gid, &cur_mode); 1115 1116 if (error == 0) { 1117 TMPFS_NODE_LOCK(node); 1118 if (cur_uid != node->tn_uid || 1119 cur_gid != node->tn_gid || 1120 cur_mode != node->tn_mode) { 1121 node->tn_uid = cur_uid; 1122 node->tn_gid = cur_gid; 1123 node->tn_mode = cur_mode; 1124 node->tn_status |= TMPFS_NODE_CHANGED; 1125 } 1126 TMPFS_NODE_UNLOCK(node); 1127 } 1128 1129 return error; 1130 } 1131 1132 /* --------------------------------------------------------------------- */ 1133 1134 /* 1135 * Change size of the given vnode. 1136 * Caller should execute tmpfs_update on vp after a successful execution. 1137 * The vnode must be locked on entry and remain locked on exit. 1138 */ 1139 int 1140 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred) 1141 { 1142 int error; 1143 struct tmpfs_node *node; 1144 1145 KKASSERT(vn_islocked(vp)); 1146 1147 node = VP_TO_TMPFS_NODE(vp); 1148 1149 /* Decide whether this is a valid operation based on the file type. */ 1150 error = 0; 1151 switch (vp->v_type) { 1152 case VDIR: 1153 return EISDIR; 1154 1155 case VREG: 1156 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1157 return EROFS; 1158 break; 1159 1160 case VBLK: 1161 /* FALLTHROUGH */ 1162 case VCHR: 1163 /* FALLTHROUGH */ 1164 case VFIFO: 1165 /* Allow modifications of special files even if in the file 1166 * system is mounted read-only (we are not modifying the 1167 * files themselves, but the objects they represent). */ 1168 return 0; 1169 1170 default: 1171 /* Anything else is unsupported. */ 1172 return EOPNOTSUPP; 1173 } 1174 1175 /* Immutable or append-only files cannot be modified, either. */ 1176 if (node->tn_flags & (IMMUTABLE | APPEND)) 1177 return EPERM; 1178 1179 error = tmpfs_truncate(vp, size); 1180 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1181 * for us, as will update tn_status; no need to do that here. */ 1182 1183 KKASSERT(vn_islocked(vp)); 1184 1185 return error; 1186 } 1187 1188 /* --------------------------------------------------------------------- */ 1189 1190 /* 1191 * Change access and modification times of the given vnode. 1192 * Caller should execute tmpfs_update on vp after a successful execution. 1193 * The vnode must be locked on entry and remain locked on exit. 1194 */ 1195 int 1196 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1197 int vaflags, struct ucred *cred) 1198 { 1199 struct tmpfs_node *node; 1200 1201 KKASSERT(vn_islocked(vp)); 1202 1203 node = VP_TO_TMPFS_NODE(vp); 1204 1205 /* Disallow this operation if the file system is mounted read-only. */ 1206 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1207 return EROFS; 1208 1209 /* Immutable or append-only files cannot be modified, either. */ 1210 if (node->tn_flags & (IMMUTABLE | APPEND)) 1211 return EPERM; 1212 1213 TMPFS_NODE_LOCK(node); 1214 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1215 node->tn_status |= TMPFS_NODE_ACCESSED; 1216 1217 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) 1218 node->tn_status |= TMPFS_NODE_MODIFIED; 1219 1220 TMPFS_NODE_UNLOCK(node); 1221 1222 tmpfs_itimes(vp, atime, mtime); 1223 1224 KKASSERT(vn_islocked(vp)); 1225 1226 return 0; 1227 } 1228 1229 /* --------------------------------------------------------------------- */ 1230 /* Sync timestamps */ 1231 void 1232 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1233 const struct timespec *mod) 1234 { 1235 struct tmpfs_node *node; 1236 struct timespec now; 1237 1238 node = VP_TO_TMPFS_NODE(vp); 1239 1240 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1241 TMPFS_NODE_CHANGED)) == 0) 1242 return; 1243 1244 vfs_timestamp(&now); 1245 1246 TMPFS_NODE_LOCK(node); 1247 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1248 if (acc == NULL) 1249 acc = &now; 1250 node->tn_atime = acc->tv_sec; 1251 node->tn_atimensec = acc->tv_nsec; 1252 } 1253 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1254 if (mod == NULL) 1255 mod = &now; 1256 node->tn_mtime = mod->tv_sec; 1257 node->tn_mtimensec = mod->tv_nsec; 1258 } 1259 if (node->tn_status & TMPFS_NODE_CHANGED) { 1260 node->tn_ctime = now.tv_sec; 1261 node->tn_ctimensec = now.tv_nsec; 1262 } 1263 node->tn_status &= 1264 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1265 TMPFS_NODE_UNLOCK(node); 1266 } 1267 1268 /* --------------------------------------------------------------------- */ 1269 1270 void 1271 tmpfs_update(struct vnode *vp) 1272 { 1273 tmpfs_itimes(vp, NULL, NULL); 1274 } 1275 1276 /* --------------------------------------------------------------------- */ 1277 1278 /* 1279 * Caller must hold an exclusive node lock. 1280 */ 1281 int 1282 tmpfs_truncate(struct vnode *vp, off_t length) 1283 { 1284 int error; 1285 struct tmpfs_node *node; 1286 1287 node = VP_TO_TMPFS_NODE(vp); 1288 1289 if (length < 0) { 1290 error = EINVAL; 1291 goto out; 1292 } 1293 1294 if (node->tn_size == length) { 1295 error = 0; 1296 goto out; 1297 } 1298 1299 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1300 return (EFBIG); 1301 1302 1303 error = tmpfs_reg_resize(vp, length, 1); 1304 1305 if (error == 0) 1306 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1307 1308 out: 1309 tmpfs_update(vp); 1310 1311 return error; 1312 } 1313 1314 /* --------------------------------------------------------------------- */ 1315 1316 static ino_t 1317 tmpfs_fetch_ino(struct tmpfs_mount *tmp) 1318 { 1319 ino_t ret; 1320 1321 TMPFS_LOCK(tmp); 1322 ret = tmp->tm_ino++; 1323 TMPFS_UNLOCK(tmp); 1324 1325 return (ret); 1326 } 1327 1328 static int 1329 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1330 { 1331 if (a->td_namelen > b->td_namelen) 1332 return 1; 1333 else if (a->td_namelen < b->td_namelen) 1334 return -1; 1335 else 1336 return strncmp(a->td_name, b->td_name, a->td_namelen); 1337 } 1338 1339 static int 1340 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1341 { 1342 if (a < b) 1343 return(-1); 1344 if (a > b) 1345 return(1); 1346 return 0; 1347 } 1348