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 = swap_pager_alloc(NULL, 0, 163 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 /* 445 * VMIO is mandatory. Tmpfs also supports KVABIO 446 * for its tmpfs_strategy(). 447 */ 448 vsetflags(vp, VKVABIO); 449 vinitvmio(vp, node->tn_size, TMPFS_BLKSIZE, -1); 450 break; 451 case VLNK: 452 break; 453 case VFIFO: 454 vp->v_ops = &mp->mnt_vn_fifo_ops; 455 break; 456 case VDIR: 457 break; 458 459 default: 460 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 461 } 462 463 464 unlock: 465 TMPFS_NODE_LOCK(node); 466 467 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 468 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 469 node->tn_vnode = vp; 470 471 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 472 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 473 TMPFS_NODE_UNLOCK(node); 474 wakeup(&node->tn_vpstate); 475 } else { 476 TMPFS_NODE_UNLOCK(node); 477 } 478 479 out: 480 *vpp = vp; 481 482 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp))); 483 #ifdef INVARIANTS 484 TMPFS_NODE_LOCK(node); 485 KKASSERT(*vpp == node->tn_vnode); 486 TMPFS_NODE_UNLOCK(node); 487 #endif 488 489 return error; 490 } 491 492 /* --------------------------------------------------------------------- */ 493 494 /* 495 * Destroys the association between the vnode vp and the node it 496 * references. 497 */ 498 void 499 tmpfs_free_vp(struct vnode *vp) 500 { 501 struct tmpfs_node *node; 502 503 node = VP_TO_TMPFS_NODE(vp); 504 505 TMPFS_NODE_LOCK(node); 506 KKASSERT(lockinuse(TMPFS_NODE_MTX(node))); 507 node->tn_vnode = NULL; 508 vp->v_data = NULL; 509 TMPFS_NODE_UNLOCK(node); 510 } 511 512 /* --------------------------------------------------------------------- */ 513 514 /* 515 * Allocates a new file of type 'type' and adds it to the parent directory 516 * 'dvp'; this addition is done using the component name given in 'cnp'. 517 * The ownership of the new file is automatically assigned based on the 518 * credentials of the caller (through 'cnp'), the group is set based on 519 * the parent directory and the mode is determined from the 'vap' argument. 520 * If successful, *vpp holds a vnode to the newly created file and zero 521 * is returned. Otherwise *vpp is NULL and the function returns an 522 * appropriate error code. 523 */ 524 int 525 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 526 struct namecache *ncp, struct ucred *cred, char *target) 527 { 528 int error; 529 struct tmpfs_dirent *de; 530 struct tmpfs_mount *tmp; 531 struct tmpfs_node *dnode; 532 struct tmpfs_node *node; 533 534 tmp = VFS_TO_TMPFS(dvp->v_mount); 535 dnode = VP_TO_TMPFS_DIR(dvp); 536 *vpp = NULL; 537 538 /* 539 * If the directory was removed but a process was CD'd into it, 540 * we do not allow any more file/dir creation within it. Otherwise 541 * we will lose track of it. 542 */ 543 KKASSERT(dnode->tn_type == VDIR); 544 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL) 545 return ENOENT; 546 547 /* 548 * Make sure the link count does not overflow. 549 */ 550 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX) 551 return EMLINK; 552 553 /* Allocate a node that represents the new file. */ 554 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid, 555 dnode->tn_gid, vap->va_mode, target, 556 vap->va_rmajor, vap->va_rminor, &node); 557 if (error != 0) 558 return error; 559 TMPFS_NODE_LOCK(node); 560 561 /* Allocate a directory entry that points to the new file. */ 562 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de); 563 if (error != 0) { 564 tmpfs_free_node(tmp, node); 565 /* eats node lock */ 566 return error; 567 } 568 569 /* Allocate a vnode for the new file. */ 570 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 571 if (error != 0) { 572 tmpfs_free_dirent(tmp, de); 573 tmpfs_free_node(tmp, node); 574 /* eats node lock */ 575 return error; 576 } 577 578 /* 579 * Now that all required items are allocated, we can proceed to 580 * insert the new node into the directory, an operation that 581 * cannot fail. 582 */ 583 tmpfs_dir_attach(dnode, de); 584 TMPFS_NODE_UNLOCK(node); 585 586 return error; 587 } 588 589 /* --------------------------------------------------------------------- */ 590 591 /* 592 * Attaches the directory entry de to the directory represented by dnode. 593 * Note that this does not change the link count of the node pointed by 594 * the directory entry, as this is done by tmpfs_alloc_dirent. 595 */ 596 void 597 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 598 { 599 struct tmpfs_node *node = de->td_node; 600 601 TMPFS_NODE_LOCK(dnode); 602 if (node && node->tn_type == VDIR) { 603 TMPFS_NODE_LOCK(node); 604 ++node->tn_links; 605 node->tn_status |= TMPFS_NODE_CHANGED; 606 node->tn_dir.tn_parent = dnode; 607 ++dnode->tn_links; 608 TMPFS_NODE_UNLOCK(node); 609 } 610 RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 611 RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 612 dnode->tn_size += sizeof(struct tmpfs_dirent); 613 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 614 TMPFS_NODE_MODIFIED; 615 TMPFS_NODE_UNLOCK(dnode); 616 } 617 618 /* --------------------------------------------------------------------- */ 619 620 /* 621 * Detaches the directory entry de from the directory represented by dnode. 622 * Note that this does not change the link count of the node pointed by 623 * the directory entry, as this is done by tmpfs_free_dirent. 624 */ 625 void 626 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 627 { 628 struct tmpfs_node *node = de->td_node; 629 630 TMPFS_NODE_LOCK(dnode); 631 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 632 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 633 dnode->tn_size -= sizeof(struct tmpfs_dirent); 634 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 635 TMPFS_NODE_MODIFIED; 636 TMPFS_NODE_UNLOCK(dnode); 637 638 /* 639 * Clean out the tn_parent pointer immediately when removing a 640 * directory. 641 * 642 * Removal of the parent linkage also cleans out the extra tn_links 643 * count we had on both node and dnode. 644 * 645 * node can be NULL (typ during a forced umount), in which case 646 * the mount code is dealing with the linkages from a linked list 647 * scan. 648 */ 649 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) { 650 TMPFS_NODE_LOCK(dnode); 651 TMPFS_NODE_LOCK(node); 652 KKASSERT(node->tn_dir.tn_parent == dnode); 653 dnode->tn_links--; 654 node->tn_links--; 655 node->tn_dir.tn_parent = NULL; 656 TMPFS_NODE_UNLOCK(node); 657 TMPFS_NODE_UNLOCK(dnode); 658 } 659 } 660 661 /* --------------------------------------------------------------------- */ 662 663 /* 664 * Looks for a directory entry in the directory represented by node. 665 * 'ncp' describes the name of the entry to look for. Note that the . 666 * and .. components are not allowed as they do not physically exist 667 * within directories. 668 * 669 * Returns a pointer to the entry when found, otherwise NULL. 670 * 671 * Caller must hold the node locked (shared ok) 672 */ 673 struct tmpfs_dirent * 674 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 675 struct namecache *ncp) 676 { 677 struct tmpfs_dirent *de; 678 int len = ncp->nc_nlen; 679 struct tmpfs_dirent wanted; 680 681 wanted.td_namelen = len; 682 wanted.td_name = ncp->nc_name; 683 684 TMPFS_VALIDATE_DIR(node); 685 686 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted); 687 688 KKASSERT(f == NULL || f == de->td_node); 689 690 return de; 691 } 692 693 /* --------------------------------------------------------------------- */ 694 695 /* 696 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 697 * directory and returns it in the uio space. The function returns 0 698 * on success, -1 if there was not enough space in the uio structure to 699 * hold the directory entry or an appropriate error code if another 700 * error happens. 701 */ 702 int 703 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 704 { 705 int error; 706 707 TMPFS_VALIDATE_DIR(node); 708 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 709 710 if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, ".")) 711 return -1; 712 if (error == 0) 713 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 714 return error; 715 } 716 717 /* --------------------------------------------------------------------- */ 718 719 /* 720 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 721 * directory and returns it in the uio space. The function returns 0 722 * on success, -1 if there was not enough space in the uio structure to 723 * hold the directory entry or an appropriate error code if another 724 * error happens. 725 */ 726 int 727 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 728 struct uio *uio) 729 { 730 int error; 731 ino_t d_ino; 732 733 TMPFS_VALIDATE_DIR(node); 734 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 735 736 if (node->tn_dir.tn_parent) { 737 TMPFS_NODE_LOCK(node); 738 if (node->tn_dir.tn_parent) 739 d_ino = node->tn_dir.tn_parent->tn_id; 740 else 741 d_ino = tmp->tm_root->tn_id; 742 TMPFS_NODE_UNLOCK(node); 743 } else { 744 d_ino = tmp->tm_root->tn_id; 745 } 746 747 if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, "..")) 748 return -1; 749 if (error == 0) { 750 struct tmpfs_dirent *de; 751 de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree); 752 if (de == NULL) 753 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 754 else 755 uio->uio_offset = tmpfs_dircookie(de); 756 } 757 return error; 758 } 759 760 /* --------------------------------------------------------------------- */ 761 762 /* 763 * Lookup a directory entry by its associated cookie. 764 * 765 * Must be called with the directory node locked (shared ok) 766 */ 767 struct lubycookie_info { 768 off_t cookie; 769 struct tmpfs_dirent *de; 770 }; 771 772 static int 773 lubycookie_cmp(struct tmpfs_dirent *de, void *arg) 774 { 775 struct lubycookie_info *info = arg; 776 off_t cookie = tmpfs_dircookie(de); 777 778 if (cookie < info->cookie) 779 return(-1); 780 if (cookie > info->cookie) 781 return(1); 782 return(0); 783 } 784 785 static int 786 lubycookie_callback(struct tmpfs_dirent *de, void *arg) 787 { 788 struct lubycookie_info *info = arg; 789 790 if (tmpfs_dircookie(de) == info->cookie) { 791 info->de = de; 792 return(-1); 793 } 794 return(0); 795 } 796 797 struct tmpfs_dirent * 798 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie) 799 { 800 struct lubycookie_info info; 801 802 info.cookie = cookie; 803 info.de = NULL; 804 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree, 805 lubycookie_cmp, lubycookie_callback, &info); 806 return (info.de); 807 } 808 809 /* --------------------------------------------------------------------- */ 810 811 /* 812 * Helper function for tmpfs_readdir. Returns as much directory entries 813 * as can fit in the uio space. The read starts at uio->uio_offset. 814 * The function returns 0 on success, -1 if there was not enough space 815 * in the uio structure to hold the directory entry or an appropriate 816 * error code if another error happens. 817 * 818 * Caller must hold the node locked (shared ok) 819 */ 820 int 821 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp) 822 { 823 int error; 824 off_t startcookie; 825 struct tmpfs_dirent *de; 826 827 TMPFS_VALIDATE_DIR(node); 828 829 /* 830 * Locate the first directory entry we have to return. We have cached 831 * the last readdir in the node, so use those values if appropriate. 832 * Otherwise do a linear scan to find the requested entry. 833 */ 834 startcookie = uio->uio_offset; 835 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT); 836 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT); 837 838 if (startcookie == TMPFS_DIRCOOKIE_EOF) 839 return 0; 840 841 de = tmpfs_dir_lookupbycookie(node, startcookie); 842 if (de == NULL) 843 return EINVAL; 844 845 /* 846 * Read as much entries as possible; i.e., until we reach the end of 847 * the directory or we exhaust uio space. 848 */ 849 do { 850 ino_t d_ino; 851 uint8_t d_type; 852 853 /* Create a dirent structure representing the current 854 * tmpfs_node and fill it. */ 855 d_ino = de->td_node->tn_id; 856 switch (de->td_node->tn_type) { 857 case VBLK: 858 d_type = DT_BLK; 859 break; 860 861 case VCHR: 862 d_type = DT_CHR; 863 break; 864 865 case VDIR: 866 d_type = DT_DIR; 867 break; 868 869 case VFIFO: 870 d_type = DT_FIFO; 871 break; 872 873 case VLNK: 874 d_type = DT_LNK; 875 break; 876 877 case VREG: 878 d_type = DT_REG; 879 break; 880 881 case VSOCK: 882 d_type = DT_SOCK; 883 break; 884 885 default: 886 panic("tmpfs_dir_getdents: type %p %d", 887 de->td_node, (int)de->td_node->tn_type); 888 } 889 KKASSERT(de->td_namelen < 256); /* 255 + 1 */ 890 891 if (vop_write_dirent(&error, uio, d_ino, d_type, 892 de->td_namelen, de->td_name)) { 893 error = -1; 894 break; 895 } 896 897 (*cntp)++; 898 de = RB_NEXT(tmpfs_dirtree_cookie, 899 node->tn_dir.tn_cookietree, de); 900 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 901 902 /* Update the offset and cache. */ 903 if (de == NULL) { 904 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 905 } else { 906 uio->uio_offset = tmpfs_dircookie(de); 907 } 908 909 return error; 910 } 911 912 /* --------------------------------------------------------------------- */ 913 914 /* 915 * Resizes the aobj associated to the regular file pointed to by vp to 916 * the size newsize. 'vp' must point to a vnode that represents a regular 917 * file. 'newsize' must be positive. 918 * 919 * pass trivial as 1 when buf content will be overwritten, otherwise set 0 920 * to be zero filled. 921 * 922 * Returns zero on success or an appropriate error code on failure. 923 * 924 * Caller must hold the node exclusively locked. 925 */ 926 int 927 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial) 928 { 929 int error; 930 vm_pindex_t newpages, oldpages; 931 struct tmpfs_mount *tmp; 932 struct tmpfs_node *node; 933 off_t oldsize; 934 935 #ifdef INVARIANTS 936 KKASSERT(vp->v_type == VREG); 937 KKASSERT(newsize >= 0); 938 #endif 939 940 node = VP_TO_TMPFS_NODE(vp); 941 tmp = VFS_TO_TMPFS(vp->v_mount); 942 943 /* 944 * Convert the old and new sizes to the number of pages needed to 945 * store them. It may happen that we do not need to do anything 946 * because the last allocated page can accommodate the change on 947 * its own. 948 */ 949 oldsize = node->tn_size; 950 oldpages = round_page64(oldsize) / PAGE_SIZE; 951 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages); 952 newpages = round_page64(newsize) / PAGE_SIZE; 953 954 if (newpages > oldpages && 955 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) { 956 error = ENOSPC; 957 goto out; 958 } 959 node->tn_reg.tn_aobj_pages = newpages; 960 node->tn_size = newsize; 961 962 if (newpages != oldpages) 963 atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages)); 964 965 /* 966 * When adjusting the vnode filesize and its VM object we must 967 * also adjust our backing VM object (aobj). The blocksize 968 * used must match the block sized we use for the buffer cache. 969 * 970 * The backing VM object may contain VM pages as well as swap 971 * assignments if we previously renamed main object pages into 972 * it during deactivation. 973 */ 974 if (newsize < oldsize) { 975 vm_pindex_t osize; 976 vm_pindex_t nsize; 977 vm_object_t aobj; 978 979 error = nvtruncbuf(vp, newsize, TMPFS_BLKSIZE, -1, 0); 980 aobj = node->tn_reg.tn_aobj; 981 if (aobj) { 982 osize = aobj->size; 983 nsize = vp->v_object->size; 984 if (nsize < osize) { 985 aobj->size = osize; 986 swap_pager_freespace(aobj, nsize, 987 osize - nsize); 988 vm_object_page_remove(aobj, nsize, osize, 989 FALSE); 990 } 991 } 992 } else { 993 vm_object_t aobj; 994 995 error = nvextendbuf(vp, oldsize, newsize, 996 TMPFS_BLKSIZE, TMPFS_BLKSIZE, 997 -1, -1, trivial); 998 aobj = node->tn_reg.tn_aobj; 999 if (aobj) 1000 aobj->size = vp->v_object->size; 1001 } 1002 1003 out: 1004 return error; 1005 } 1006 1007 /* --------------------------------------------------------------------- */ 1008 1009 /* 1010 * Change flags of the given vnode. 1011 * Caller should execute tmpfs_update on vp after a successful execution. 1012 * The vnode must be locked on entry and remain locked on exit. 1013 */ 1014 int 1015 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred) 1016 { 1017 int error; 1018 struct tmpfs_node *node; 1019 int flags; 1020 1021 KKASSERT(vn_islocked(vp)); 1022 1023 node = VP_TO_TMPFS_NODE(vp); 1024 flags = node->tn_flags; 1025 1026 /* Disallow this operation if the file system is mounted read-only. */ 1027 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1028 return EROFS; 1029 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred); 1030 1031 /* Actually change the flags on the node itself */ 1032 if (error == 0) { 1033 TMPFS_NODE_LOCK(node); 1034 node->tn_flags = flags; 1035 node->tn_status |= TMPFS_NODE_CHANGED; 1036 TMPFS_NODE_UNLOCK(node); 1037 } 1038 1039 KKASSERT(vn_islocked(vp)); 1040 1041 return error; 1042 } 1043 1044 /* --------------------------------------------------------------------- */ 1045 1046 /* 1047 * Change access mode on the given vnode. 1048 * Caller should execute tmpfs_update on vp after a successful execution. 1049 * The vnode must be locked on entry and remain locked on exit. 1050 */ 1051 int 1052 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred) 1053 { 1054 struct tmpfs_node *node; 1055 mode_t cur_mode; 1056 int error; 1057 1058 KKASSERT(vn_islocked(vp)); 1059 1060 node = VP_TO_TMPFS_NODE(vp); 1061 1062 /* Disallow this operation if the file system is mounted read-only. */ 1063 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1064 return EROFS; 1065 1066 /* Immutable or append-only files cannot be modified, either. */ 1067 if (node->tn_flags & (IMMUTABLE | APPEND)) 1068 return EPERM; 1069 1070 cur_mode = node->tn_mode; 1071 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid, 1072 &cur_mode); 1073 1074 if (error == 0 && 1075 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) { 1076 TMPFS_NODE_LOCK(node); 1077 node->tn_mode &= ~ALLPERMS; 1078 node->tn_mode |= cur_mode & ALLPERMS; 1079 1080 node->tn_status |= TMPFS_NODE_CHANGED; 1081 TMPFS_NODE_UNLOCK(node); 1082 } 1083 1084 KKASSERT(vn_islocked(vp)); 1085 1086 return 0; 1087 } 1088 1089 /* --------------------------------------------------------------------- */ 1090 1091 /* 1092 * Change ownership of the given vnode. At least one of uid or gid must 1093 * be different than VNOVAL. If one is set to that value, the attribute 1094 * is unchanged. 1095 * Caller should execute tmpfs_update on vp after a successful execution. 1096 * The vnode must be locked on entry and remain locked on exit. 1097 */ 1098 int 1099 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 1100 { 1101 mode_t cur_mode; 1102 uid_t cur_uid; 1103 gid_t cur_gid; 1104 struct tmpfs_node *node; 1105 int error; 1106 1107 KKASSERT(vn_islocked(vp)); 1108 node = VP_TO_TMPFS_NODE(vp); 1109 1110 /* Disallow this operation if the file system is mounted read-only. */ 1111 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1112 return EROFS; 1113 1114 /* Immutable or append-only files cannot be modified, either. */ 1115 if (node->tn_flags & (IMMUTABLE | APPEND)) 1116 return EPERM; 1117 1118 cur_uid = node->tn_uid; 1119 cur_gid = node->tn_gid; 1120 cur_mode = node->tn_mode; 1121 error = vop_helper_chown(vp, uid, gid, cred, 1122 &cur_uid, &cur_gid, &cur_mode); 1123 1124 if (error == 0) { 1125 TMPFS_NODE_LOCK(node); 1126 if (cur_uid != node->tn_uid || 1127 cur_gid != node->tn_gid || 1128 cur_mode != node->tn_mode) { 1129 node->tn_uid = cur_uid; 1130 node->tn_gid = cur_gid; 1131 node->tn_mode = cur_mode; 1132 node->tn_status |= TMPFS_NODE_CHANGED; 1133 } 1134 TMPFS_NODE_UNLOCK(node); 1135 } 1136 1137 return error; 1138 } 1139 1140 /* --------------------------------------------------------------------- */ 1141 1142 /* 1143 * Change size of the given vnode. 1144 * Caller should execute tmpfs_update on vp after a successful execution. 1145 * The vnode must be locked on entry and remain locked on exit. 1146 */ 1147 int 1148 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred) 1149 { 1150 int error; 1151 struct tmpfs_node *node; 1152 1153 KKASSERT(vn_islocked(vp)); 1154 1155 node = VP_TO_TMPFS_NODE(vp); 1156 1157 /* Decide whether this is a valid operation based on the file type. */ 1158 error = 0; 1159 switch (vp->v_type) { 1160 case VDIR: 1161 return EISDIR; 1162 1163 case VREG: 1164 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1165 return EROFS; 1166 break; 1167 1168 case VBLK: 1169 /* FALLTHROUGH */ 1170 case VCHR: 1171 /* FALLTHROUGH */ 1172 case VFIFO: 1173 /* Allow modifications of special files even if in the file 1174 * system is mounted read-only (we are not modifying the 1175 * files themselves, but the objects they represent). */ 1176 return 0; 1177 1178 default: 1179 /* Anything else is unsupported. */ 1180 return EOPNOTSUPP; 1181 } 1182 1183 /* Immutable or append-only files cannot be modified, either. */ 1184 if (node->tn_flags & (IMMUTABLE | APPEND)) 1185 return EPERM; 1186 1187 error = tmpfs_truncate(vp, size); 1188 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1189 * for us, as will update tn_status; no need to do that here. */ 1190 1191 KKASSERT(vn_islocked(vp)); 1192 1193 return error; 1194 } 1195 1196 /* --------------------------------------------------------------------- */ 1197 1198 /* 1199 * Change access and modification times of the given vnode. 1200 * Caller should execute tmpfs_update on vp after a successful execution. 1201 * The vnode must be locked on entry and remain locked on exit. 1202 */ 1203 int 1204 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1205 int vaflags, struct ucred *cred) 1206 { 1207 struct tmpfs_node *node; 1208 1209 KKASSERT(vn_islocked(vp)); 1210 1211 node = VP_TO_TMPFS_NODE(vp); 1212 1213 /* Disallow this operation if the file system is mounted read-only. */ 1214 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1215 return EROFS; 1216 1217 /* Immutable or append-only files cannot be modified, either. */ 1218 if (node->tn_flags & (IMMUTABLE | APPEND)) 1219 return EPERM; 1220 1221 TMPFS_NODE_LOCK(node); 1222 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1223 node->tn_status |= TMPFS_NODE_ACCESSED; 1224 1225 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) { 1226 node->tn_status |= TMPFS_NODE_MODIFIED; 1227 vclrflags(vp, VLASTWRITETS); 1228 } 1229 1230 TMPFS_NODE_UNLOCK(node); 1231 1232 tmpfs_itimes(vp, atime, mtime); 1233 1234 KKASSERT(vn_islocked(vp)); 1235 1236 return 0; 1237 } 1238 1239 /* --------------------------------------------------------------------- */ 1240 /* Sync timestamps */ 1241 void 1242 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1243 const struct timespec *mod) 1244 { 1245 struct tmpfs_node *node; 1246 struct timespec now; 1247 1248 node = VP_TO_TMPFS_NODE(vp); 1249 1250 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1251 TMPFS_NODE_CHANGED)) == 0) 1252 return; 1253 1254 vfs_timestamp(&now); 1255 1256 TMPFS_NODE_LOCK(node); 1257 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1258 if (acc == NULL) 1259 acc = &now; 1260 node->tn_atime = acc->tv_sec; 1261 node->tn_atimensec = acc->tv_nsec; 1262 } 1263 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1264 if (mod == NULL) 1265 mod = &now; 1266 node->tn_mtime = mod->tv_sec; 1267 node->tn_mtimensec = mod->tv_nsec; 1268 } 1269 if (node->tn_status & TMPFS_NODE_CHANGED) { 1270 node->tn_ctime = now.tv_sec; 1271 node->tn_ctimensec = now.tv_nsec; 1272 } 1273 1274 node->tn_status &= ~(TMPFS_NODE_ACCESSED | 1275 TMPFS_NODE_MODIFIED | 1276 TMPFS_NODE_CHANGED); 1277 TMPFS_NODE_UNLOCK(node); 1278 } 1279 1280 /* --------------------------------------------------------------------- */ 1281 1282 void 1283 tmpfs_update(struct vnode *vp) 1284 { 1285 tmpfs_itimes(vp, NULL, NULL); 1286 } 1287 1288 /* --------------------------------------------------------------------- */ 1289 1290 /* 1291 * Caller must hold an exclusive node lock. 1292 */ 1293 int 1294 tmpfs_truncate(struct vnode *vp, off_t length) 1295 { 1296 int error; 1297 struct tmpfs_node *node; 1298 1299 node = VP_TO_TMPFS_NODE(vp); 1300 1301 if (length < 0) { 1302 error = EINVAL; 1303 goto out; 1304 } 1305 1306 if (node->tn_size == length) { 1307 error = 0; 1308 goto out; 1309 } 1310 1311 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1312 return (EFBIG); 1313 1314 1315 error = tmpfs_reg_resize(vp, length, 1); 1316 1317 if (error == 0) 1318 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1319 1320 out: 1321 tmpfs_update(vp); 1322 1323 return error; 1324 } 1325 1326 /* --------------------------------------------------------------------- */ 1327 1328 static ino_t 1329 tmpfs_fetch_ino(struct tmpfs_mount *tmp) 1330 { 1331 ino_t ret; 1332 1333 TMPFS_LOCK(tmp); 1334 ret = tmp->tm_ino++; 1335 TMPFS_UNLOCK(tmp); 1336 1337 return (ret); 1338 } 1339 1340 static int 1341 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1342 { 1343 if (a->td_namelen > b->td_namelen) 1344 return 1; 1345 else if (a->td_namelen < b->td_namelen) 1346 return -1; 1347 else 1348 return strncmp(a->td_name, b->td_name, a->td_namelen); 1349 } 1350 1351 static int 1352 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1353 { 1354 if (a < b) 1355 return(-1); 1356 if (a > b) 1357 return(1); 1358 return 0; 1359 } 1360