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/priv.h> 40 #include <sys/proc.h> 41 #include <sys/stat.h> 42 #include <sys/systm.h> 43 #include <sys/vnode.h> 44 #include <sys/vmmeter.h> 45 46 #include <vm/vm.h> 47 #include <vm/vm_object.h> 48 #include <vm/vm_page.h> 49 #include <vm/vm_pager.h> 50 #include <vm/vm_extern.h> 51 #include <vm/vm_pageout.h> 52 #include <vm/vm_page2.h> 53 54 #include <vfs/tmpfs/tmpfs.h> 55 #include <vfs/tmpfs/tmpfs_vnops.h> 56 57 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *); 58 59 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a, 60 struct tmpfs_dirent *b); 61 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare); 62 63 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, 64 struct tmpfs_dirent *b); 65 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent, 66 rb_cookienode, tmpfs_dirtree_compare_cookie); 67 68 69 /* --------------------------------------------------------------------- */ 70 71 /* 72 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 73 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 74 * using the credentials of the process 'p'. 75 * 76 * If the node type is set to 'VDIR', then the parent parameter must point 77 * to the parent directory of the node being created. It may only be NULL 78 * while allocating the root node. 79 * 80 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 81 * specifies the device the node represents. 82 * 83 * If the node type is set to 'VLNK', then the parameter target specifies 84 * the file name of the target file for the symbolic link that is being 85 * created. 86 * 87 * Note that new nodes are retrieved from the available list if it has 88 * items or, if it is empty, from the node pool as long as there is enough 89 * space to create them. 90 * 91 * Returns zero on success or an appropriate error code on failure. 92 */ 93 int 94 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 95 uid_t uid, gid_t gid, mode_t mode, 96 char *target, int rmajor, int rminor, 97 struct tmpfs_node **node) 98 { 99 struct tmpfs_node *nnode; 100 struct timespec ts; 101 dev_t rdev; 102 103 KKASSERT(IFF(type == VLNK, target != NULL)); 104 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL)); 105 106 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 107 return (ENOSPC); 108 109 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK); 110 if (nnode == NULL) 111 return (ENOSPC); 112 113 /* Generic initialization. */ 114 nnode->tn_type = type; 115 vfs_timestamp(&ts); 116 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime 117 = ts.tv_sec; 118 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec 119 = ts.tv_nsec; 120 nnode->tn_uid = uid; 121 nnode->tn_gid = gid; 122 nnode->tn_mode = mode; 123 nnode->tn_id = tmpfs_fetch_ino(tmp); 124 nnode->tn_advlock.init_done = 0; 125 KKASSERT(nnode->tn_links == 0); 126 127 /* Type-specific initialization. */ 128 switch (nnode->tn_type) { 129 case VBLK: 130 case VCHR: 131 rdev = makeudev(rmajor, rminor); 132 if (rdev == NOUDEV) { 133 objcache_put(tmp->tm_node_pool, nnode); 134 return(EINVAL); 135 } 136 nnode->tn_rdev = rdev; 137 break; 138 139 case VDIR: 140 RB_INIT(&nnode->tn_dir.tn_dirtree); 141 RB_INIT(&nnode->tn_dir.tn_cookietree); 142 nnode->tn_dir.tn_parent = NULL; 143 nnode->tn_size = 0; 144 break; 145 146 case VFIFO: 147 /* FALLTHROUGH */ 148 case VSOCK: 149 break; 150 151 case VLNK: 152 nnode->tn_size = strlen(target); 153 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone, 154 M_WAITOK | M_NULLOK); 155 if (nnode->tn_link == NULL) { 156 objcache_put(tmp->tm_node_pool, nnode); 157 return (ENOSPC); 158 } 159 bcopy(target, nnode->tn_link, nnode->tn_size); 160 nnode->tn_link[nnode->tn_size] = '\0'; 161 break; 162 163 case VREG: 164 nnode->tn_reg.tn_aobj = swap_pager_alloc(NULL, 0, 165 VM_PROT_DEFAULT, 0); 166 nnode->tn_reg.tn_aobj_pages = 0; 167 nnode->tn_size = 0; 168 vm_object_set_flag(nnode->tn_reg.tn_aobj, OBJ_NOPAGEIN); 169 break; 170 171 default: 172 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 173 } 174 175 TMPFS_NODE_LOCK(nnode); 176 TMPFS_LOCK(tmp); 177 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 178 tmp->tm_nodes_inuse++; 179 TMPFS_UNLOCK(tmp); 180 TMPFS_NODE_UNLOCK(nnode); 181 182 *node = nnode; 183 return 0; 184 } 185 186 /* --------------------------------------------------------------------- */ 187 188 /* 189 * Destroys the node pointed to by node from the file system 'tmp'. 190 * If the node does not belong to the given mount point, the results are 191 * unpredicted. 192 * 193 * If the node references a directory; no entries are allowed because 194 * their removal could need a recursive algorithm, something forbidden in 195 * kernel space. Furthermore, there is not need to provide such 196 * functionality (recursive removal) because the only primitives offered 197 * to the user are the removal of empty directories and the deletion of 198 * individual files. 199 * 200 * Note that nodes are not really deleted; in fact, when a node has been 201 * allocated, it cannot be deleted during the whole life of the file 202 * system. Instead, they are moved to the available list and remain there 203 * until reused. 204 * 205 * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it. 206 */ 207 void 208 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 209 { 210 vm_pindex_t pages = 0; 211 212 #ifdef INVARIANTS 213 TMPFS_ASSERT_ELOCKED(node); 214 KKASSERT(node->tn_vnode == NULL); 215 #endif 216 217 TMPFS_LOCK(tmp); 218 LIST_REMOVE(node, tn_entries); 219 tmp->tm_nodes_inuse--; 220 TMPFS_UNLOCK(tmp); 221 TMPFS_NODE_UNLOCK(node); /* Caller has this lock */ 222 223 switch (node->tn_type) { 224 case VNON: 225 /* Do not do anything. VNON is provided to let the 226 * allocation routine clean itself easily by avoiding 227 * duplicating code in it. */ 228 /* FALLTHROUGH */ 229 case VBLK: 230 /* FALLTHROUGH */ 231 case VCHR: 232 /* FALLTHROUGH */ 233 break; 234 case VDIR: 235 /* 236 * The parent link can be NULL if this is the root 237 * node or if it is a directory node that was rmdir'd. 238 * 239 * XXX what if node is a directory which still contains 240 * directory entries (e.g. due to a forced umount) ? 241 */ 242 node->tn_size = 0; 243 KKASSERT(node->tn_dir.tn_parent == NULL); 244 245 /* 246 * If the root node is being destroyed don't leave a 247 * dangling pointer in tmpfs_mount. 248 */ 249 if (node == tmp->tm_root) 250 tmp->tm_root = NULL; 251 break; 252 case VFIFO: 253 /* FALLTHROUGH */ 254 case VSOCK: 255 break; 256 257 case VLNK: 258 kfree(node->tn_link, tmp->tm_name_zone); 259 node->tn_link = NULL; 260 node->tn_size = 0; 261 break; 262 263 case VREG: 264 if (node->tn_reg.tn_aobj != NULL) 265 vm_object_deallocate(node->tn_reg.tn_aobj); 266 node->tn_reg.tn_aobj = NULL; 267 pages = node->tn_reg.tn_aobj_pages; 268 break; 269 270 default: 271 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 272 } 273 274 /* 275 * Clean up fields for the next allocation. The objcache only ctors 276 * new allocations. 277 */ 278 tmpfs_node_ctor(node, NULL, 0); 279 objcache_put(tmp->tm_node_pool, node); 280 /* node is now invalid */ 281 282 if (pages) 283 atomic_add_long(&tmp->tm_pages_used, -(long)pages); 284 } 285 286 /* --------------------------------------------------------------------- */ 287 288 /* 289 * Allocates a new directory entry for the node node with a name of name. 290 * The new directory entry is returned in *de. 291 * 292 * The link count of node is increased by one to reflect the new object 293 * referencing it. 294 * 295 * Returns zero on success or an appropriate error code on failure. 296 */ 297 int 298 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 299 const char *name, uint16_t len, struct tmpfs_dirent **de) 300 { 301 struct tmpfs_dirent *nde; 302 303 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK); 304 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK); 305 if (nde->td_name == NULL) { 306 objcache_put(tmp->tm_dirent_pool, nde); 307 *de = NULL; 308 return (ENOSPC); 309 } 310 nde->td_namelen = len; 311 bcopy(name, nde->td_name, len); 312 nde->td_name[len] = '\0'; 313 314 nde->td_node = node; 315 316 atomic_add_int(&node->tn_links, 1); 317 318 *de = nde; 319 320 return 0; 321 } 322 323 /* --------------------------------------------------------------------- */ 324 325 /* 326 * Frees a directory entry. It is the caller's responsibility to destroy 327 * the node referenced by it if needed. 328 * 329 * The link count of node is decreased by one to reflect the removal of an 330 * object that referenced it. This only happens if 'node_exists' is true; 331 * otherwise the function will not access the node referred to by the 332 * directory entry, as it may already have been released from the outside. 333 */ 334 void 335 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 336 { 337 struct tmpfs_node *node; 338 339 node = de->td_node; 340 341 KKASSERT(node->tn_links > 0); 342 atomic_add_int(&node->tn_links, -1); 343 344 kfree(de->td_name, tmp->tm_name_zone); 345 de->td_namelen = 0; 346 de->td_name = NULL; 347 de->td_node = NULL; 348 objcache_put(tmp->tm_dirent_pool, de); 349 } 350 351 /* --------------------------------------------------------------------- */ 352 353 /* 354 * Allocates a new vnode for the node node or returns a new reference to 355 * an existing one if the node had already a vnode referencing it. The 356 * resulting locked vnode is returned in *vpp. 357 * 358 * Returns zero on success or an appropriate error code on failure. 359 * 360 * The caller must ensure that node cannot go away (usually by holding 361 * the related directory entry). 362 * 363 * If dnode is non-NULL this routine avoids deadlocking against it but 364 * can return EAGAIN. Caller must try again. The dnode lock will cycle 365 * in this case, it remains locked on return in all cases. dnode must 366 * be shared-locked. 367 */ 368 int 369 tmpfs_alloc_vp(struct mount *mp, 370 struct tmpfs_node *dnode, struct tmpfs_node *node, int lkflag, 371 struct vnode **vpp) 372 { 373 int error = 0; 374 struct vnode *vp; 375 376 loop: 377 vp = NULL; 378 if (node->tn_vnode == NULL) { 379 error = getnewvnode(VT_TMPFS, mp, &vp, 380 VLKTIMEOUT, LK_CANRECURSE); 381 if (error) 382 goto out; 383 } 384 385 /* 386 * Interlocked extraction from node. This can race many things. 387 * We have to get a soft reference on the vnode while we hold 388 * the node locked, then acquire it properly and check for races. 389 */ 390 TMPFS_NODE_LOCK(node); 391 if (node->tn_vnode) { 392 if (vp) { 393 vp->v_type = VBAD; 394 vx_put(vp); 395 } 396 vp = node->tn_vnode; 397 398 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 399 vhold(vp); 400 TMPFS_NODE_UNLOCK(node); 401 402 if (dnode) { 403 /* 404 * Special-case handling to avoid deadlocking against 405 * dnode. This case has been validated and occurs 406 * every so often during synth builds. 407 */ 408 if (vget(vp, (lkflag & ~LK_RETRY) | 409 LK_NOWAIT | 410 LK_EXCLUSIVE) != 0) { 411 TMPFS_NODE_UNLOCK(dnode); 412 if (vget(vp, (lkflag & ~LK_RETRY) | 413 LK_SLEEPFAIL | 414 LK_EXCLUSIVE) == 0) { 415 vn_unlock(vp); 416 } 417 vdrop(vp); 418 TMPFS_NODE_LOCK_SH(dnode); 419 420 return EAGAIN; 421 } 422 } else { 423 /* 424 * Normal path 425 */ 426 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) { 427 vdrop(vp); 428 goto loop; 429 } 430 } 431 if (node->tn_vnode != vp) { 432 vput(vp); 433 vdrop(vp); 434 goto loop; 435 } 436 vdrop(vp); 437 goto out; 438 } 439 440 /* 441 * We need to assign node->tn_vnode. If vp is NULL, loop up to 442 * allocate the vp. This can happen due to SMP races. 443 */ 444 if (vp == NULL) { 445 TMPFS_NODE_UNLOCK(node); 446 goto loop; 447 } 448 449 /* 450 * This should never happen. 451 */ 452 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) { 453 TMPFS_NODE_UNLOCK(node); 454 vp->v_type = VBAD; 455 vx_put(vp); 456 error = ENOENT; 457 goto out; 458 } 459 460 KKASSERT(node->tn_vnode == NULL); 461 KKASSERT(vp != NULL); 462 vp->v_data = node; 463 vp->v_type = node->tn_type; 464 465 /* Type-specific initialization. */ 466 switch (node->tn_type) { 467 case VBLK: 468 /* FALLTHROUGH */ 469 case VCHR: 470 /* FALLTHROUGH */ 471 case VSOCK: 472 break; 473 case VREG: 474 /* 475 * VMIO is mandatory. Tmpfs also supports KVABIO 476 * for its tmpfs_strategy(). 477 */ 478 vsetflags(vp, VKVABIO); 479 vinitvmio(vp, node->tn_size, node->tn_blksize, -1); 480 break; 481 case VLNK: 482 break; 483 case VFIFO: 484 vp->v_ops = &mp->mnt_vn_fifo_ops; 485 break; 486 case VDIR: 487 break; 488 489 default: 490 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 491 } 492 493 node->tn_vnode = vp; 494 TMPFS_NODE_UNLOCK(node); 495 496 vx_downgrade(vp); 497 out: 498 *vpp = vp; 499 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp))); 500 501 return error; 502 } 503 504 /* --------------------------------------------------------------------- */ 505 506 /* 507 * Allocates a new file of type 'type' and adds it to the parent directory 508 * 'dvp'; this addition is done using the component name given in 'cnp'. 509 * The ownership of the new file is automatically assigned based on the 510 * credentials of the caller (through 'cnp'), the group is set based on 511 * the parent directory and the mode is determined from the 'vap' argument. 512 * If successful, *vpp holds a vnode to the newly created file and zero 513 * is returned. Otherwise *vpp is NULL and the function returns an 514 * appropriate error code. 515 */ 516 int 517 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 518 struct namecache *ncp, struct ucred *cred, char *target) 519 { 520 int error; 521 struct tmpfs_dirent *de; 522 struct tmpfs_mount *tmp; 523 struct tmpfs_node *dnode; 524 struct tmpfs_node *node; 525 526 tmp = VFS_TO_TMPFS(dvp->v_mount); 527 dnode = VP_TO_TMPFS_DIR(dvp); 528 *vpp = NULL; 529 530 TMPFS_NODE_LOCK(dnode); 531 532 /* 533 * If the directory was removed but a process was CD'd into it, 534 * we do not allow any more file/dir creation within it. Otherwise 535 * we will lose track of it. 536 */ 537 KKASSERT(dnode->tn_type == VDIR); 538 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL) { 539 TMPFS_NODE_UNLOCK(dnode); 540 return ENOENT; 541 } 542 543 /* 544 * Make sure the link count does not overflow. 545 */ 546 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX) { 547 TMPFS_NODE_UNLOCK(dnode); 548 return EMLINK; 549 } 550 551 /* Allocate a node that represents the new file. */ 552 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid, 553 dnode->tn_gid, vap->va_mode, target, 554 vap->va_rmajor, vap->va_rminor, &node); 555 if (error != 0) { 556 TMPFS_NODE_UNLOCK(dnode); 557 return error; 558 } 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_NODE_UNLOCK(dnode); 565 tmpfs_free_node(tmp, node); 566 /* eats node lock */ 567 return error; 568 } 569 570 /* Allocate a vnode for the new file. */ 571 error = tmpfs_alloc_vp(dvp->v_mount, NULL, node, LK_EXCLUSIVE, vpp); 572 if (error != 0) { 573 TMPFS_NODE_UNLOCK(dnode); 574 tmpfs_free_dirent(tmp, de); 575 tmpfs_free_node(tmp, node); 576 /* eats node lock */ 577 return error; 578 } 579 580 /* 581 * Now that all required items are allocated, we can proceed to 582 * insert the new node into the directory, an operation that 583 * cannot fail. 584 */ 585 tmpfs_dir_attach_locked(dnode, de); 586 TMPFS_NODE_UNLOCK(dnode); 587 TMPFS_NODE_UNLOCK(node); 588 589 return error; 590 } 591 592 /* --------------------------------------------------------------------- */ 593 594 /* 595 * Attaches the directory entry de to the directory represented by dnode. 596 * Note that this does not change the link count of the node pointed by 597 * the directory entry, as this is done by tmpfs_alloc_dirent. 598 * 599 * dnode must be locked. 600 */ 601 void 602 tmpfs_dir_attach_locked(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 603 { 604 struct tmpfs_node *node = de->td_node; 605 struct tmpfs_dirent *de2; 606 607 if (node && node->tn_type == VDIR) { 608 TMPFS_NODE_LOCK(node); 609 atomic_add_int(&node->tn_links, 1); 610 node->tn_status |= TMPFS_NODE_CHANGED; 611 node->tn_dir.tn_parent = dnode; 612 atomic_add_int(&dnode->tn_links, 1); 613 TMPFS_NODE_UNLOCK(node); 614 } 615 de2 = RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 616 KASSERT(de2 == NULL, 617 ("tmpfs_dir_attach_lockedA: duplicate insertion of %p, has %p\n", 618 de, de2)); 619 de2 = RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 620 KASSERT(de2 == NULL, 621 ("tmpfs_dir_attach_lockedB: duplicate insertion of %p, has %p\n", 622 de, de2)); 623 dnode->tn_size += sizeof(struct tmpfs_dirent); 624 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 625 TMPFS_NODE_MODIFIED; 626 } 627 628 /* --------------------------------------------------------------------- */ 629 630 /* 631 * Detaches the directory entry de from the directory represented by dnode. 632 * Note that this does not change the link count of the node pointed by 633 * the directory entry, as this is done by tmpfs_free_dirent. 634 * 635 * dnode must be locked. 636 */ 637 void 638 tmpfs_dir_detach_locked(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 639 { 640 struct tmpfs_node *node = de->td_node; 641 642 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 643 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 644 dnode->tn_size -= sizeof(struct tmpfs_dirent); 645 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 646 TMPFS_NODE_MODIFIED; 647 648 /* 649 * Clean out the tn_parent pointer immediately when removing a 650 * directory. 651 * 652 * Removal of the parent linkage also cleans out the extra tn_links 653 * count we had on both node and dnode. 654 * 655 * node can be NULL (typ during a forced umount), in which case 656 * the mount code is dealing with the linkages from a linked list 657 * scan. 658 */ 659 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) { 660 TMPFS_NODE_LOCK(node); 661 KKASSERT(node->tn_dir.tn_parent == dnode); 662 atomic_add_int(&dnode->tn_links, -1); 663 atomic_add_int(&node->tn_links, -1); 664 node->tn_dir.tn_parent = NULL; 665 TMPFS_NODE_UNLOCK(node); 666 } 667 } 668 669 /* --------------------------------------------------------------------- */ 670 671 /* 672 * Looks for a directory entry in the directory represented by node. 673 * 'ncp' describes the name of the entry to look for. Note that the . 674 * and .. components are not allowed as they do not physically exist 675 * within directories. 676 * 677 * Returns a pointer to the entry when found, otherwise NULL. 678 * 679 * Caller must hold the node locked (shared ok) 680 */ 681 struct tmpfs_dirent * 682 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 683 struct namecache *ncp) 684 { 685 struct tmpfs_dirent *de; 686 int len = ncp->nc_nlen; 687 struct tmpfs_dirent wanted; 688 689 wanted.td_namelen = len; 690 wanted.td_name = ncp->nc_name; 691 692 TMPFS_VALIDATE_DIR(node); 693 694 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted); 695 696 KASSERT((f == NULL || de == NULL || f == de->td_node), 697 ("tmpfs_dir_lookup: Incorrect node %p %p %p", 698 f, de, (de ? de->td_node : NULL))); 699 700 return de; 701 } 702 703 /* --------------------------------------------------------------------- */ 704 705 /* 706 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 707 * directory and returns it in the uio space. The function returns 0 708 * on success, -1 if there was not enough space in the uio structure to 709 * hold the directory entry or an appropriate error code if another 710 * error happens. 711 */ 712 int 713 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 714 { 715 int error; 716 717 TMPFS_VALIDATE_DIR(node); 718 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 719 720 if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, ".")) 721 return -1; 722 if (error == 0) 723 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 724 return error; 725 } 726 727 /* --------------------------------------------------------------------- */ 728 729 /* 730 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 731 * directory and returns it in the uio space. The function returns 0 732 * on success, -1 if there was not enough space in the uio structure to 733 * hold the directory entry or an appropriate error code if another 734 * error happens. 735 */ 736 int 737 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 738 struct uio *uio) 739 { 740 int error; 741 ino_t d_ino; 742 743 TMPFS_VALIDATE_DIR(node); 744 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 745 746 if (node->tn_dir.tn_parent) { 747 TMPFS_NODE_LOCK(node); 748 if (node->tn_dir.tn_parent) 749 d_ino = node->tn_dir.tn_parent->tn_id; 750 else 751 d_ino = tmp->tm_root->tn_id; 752 TMPFS_NODE_UNLOCK(node); 753 } else { 754 d_ino = tmp->tm_root->tn_id; 755 } 756 757 if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, "..")) 758 return -1; 759 if (error == 0) { 760 struct tmpfs_dirent *de; 761 de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree); 762 if (de == NULL) 763 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 764 else 765 uio->uio_offset = tmpfs_dircookie(de); 766 } 767 return error; 768 } 769 770 /* --------------------------------------------------------------------- */ 771 772 /* 773 * Lookup a directory entry by its associated cookie. 774 * 775 * Must be called with the directory node locked (shared ok) 776 */ 777 struct lubycookie_info { 778 off_t cookie; 779 struct tmpfs_dirent *de; 780 }; 781 782 static int 783 lubycookie_cmp(struct tmpfs_dirent *de, void *arg) 784 { 785 struct lubycookie_info *info = arg; 786 off_t cookie = tmpfs_dircookie(de); 787 788 if (cookie < info->cookie) 789 return(-1); 790 if (cookie > info->cookie) 791 return(1); 792 return(0); 793 } 794 795 static int 796 lubycookie_callback(struct tmpfs_dirent *de, void *arg) 797 { 798 struct lubycookie_info *info = arg; 799 800 if (tmpfs_dircookie(de) == info->cookie) { 801 info->de = de; 802 return(-1); 803 } 804 return(0); 805 } 806 807 struct tmpfs_dirent * 808 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie) 809 { 810 struct lubycookie_info info; 811 812 info.cookie = cookie; 813 info.de = NULL; 814 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree, 815 lubycookie_cmp, lubycookie_callback, &info); 816 return (info.de); 817 } 818 819 /* --------------------------------------------------------------------- */ 820 821 /* 822 * Helper function for tmpfs_readdir. Returns as much directory entries 823 * as can fit in the uio space. The read starts at uio->uio_offset. 824 * The function returns 0 on success, -1 if there was not enough space 825 * in the uio structure to hold the directory entry or an appropriate 826 * error code if another error happens. 827 * 828 * Caller must hold the node locked (shared ok) 829 */ 830 int 831 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp) 832 { 833 int error; 834 off_t startcookie; 835 struct tmpfs_dirent *de; 836 837 TMPFS_VALIDATE_DIR(node); 838 839 /* 840 * Locate the first directory entry we have to return. We have cached 841 * the last readdir in the node, so use those values if appropriate. 842 * Otherwise do a linear scan to find the requested entry. 843 */ 844 startcookie = uio->uio_offset; 845 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT); 846 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT); 847 848 if (startcookie == TMPFS_DIRCOOKIE_EOF) 849 return 0; 850 851 de = tmpfs_dir_lookupbycookie(node, startcookie); 852 if (de == NULL) 853 return EINVAL; 854 855 /* 856 * Read as much entries as possible; i.e., until we reach the end of 857 * the directory or we exhaust uio space. 858 */ 859 do { 860 ino_t d_ino; 861 uint8_t d_type; 862 863 /* Create a dirent structure representing the current 864 * tmpfs_node and fill it. */ 865 d_ino = de->td_node->tn_id; 866 switch (de->td_node->tn_type) { 867 case VBLK: 868 d_type = DT_BLK; 869 break; 870 871 case VCHR: 872 d_type = DT_CHR; 873 break; 874 875 case VDIR: 876 d_type = DT_DIR; 877 break; 878 879 case VFIFO: 880 d_type = DT_FIFO; 881 break; 882 883 case VLNK: 884 d_type = DT_LNK; 885 break; 886 887 case VREG: 888 d_type = DT_REG; 889 break; 890 891 case VSOCK: 892 d_type = DT_SOCK; 893 break; 894 895 default: 896 panic("tmpfs_dir_getdents: type %p %d", 897 de->td_node, (int)de->td_node->tn_type); 898 } 899 KKASSERT(de->td_namelen < 256); /* 255 + 1 */ 900 901 if (vop_write_dirent(&error, uio, d_ino, d_type, 902 de->td_namelen, de->td_name)) { 903 error = -1; 904 break; 905 } 906 907 (*cntp)++; 908 de = RB_NEXT(tmpfs_dirtree_cookie, 909 node->tn_dir.tn_cookietree, de); 910 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 911 912 /* Update the offset and cache. */ 913 if (de == NULL) { 914 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 915 } else { 916 uio->uio_offset = tmpfs_dircookie(de); 917 } 918 919 return error; 920 } 921 922 /* --------------------------------------------------------------------- */ 923 924 /* 925 * Resizes the aobj associated to the regular file pointed to by vp to 926 * the size newsize. 'vp' must point to a vnode that represents a regular 927 * file. 'newsize' must be positive. 928 * 929 * pass NVEXTF_TRIVIAL when buf content will be overwritten, otherwise set 0 930 * to be zero filled. 931 * 932 * Returns zero on success or an appropriate error code on failure. 933 * 934 * Caller must hold the node exclusively locked. 935 */ 936 int 937 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial) 938 { 939 int error; 940 vm_pindex_t newpages, oldpages; 941 struct tmpfs_mount *tmp; 942 struct tmpfs_node *node; 943 off_t oldsize; 944 int nvextflags; 945 946 #ifdef INVARIANTS 947 KKASSERT(vp->v_type == VREG); 948 KKASSERT(newsize >= 0); 949 #endif 950 951 node = VP_TO_TMPFS_NODE(vp); 952 tmp = VFS_TO_TMPFS(vp->v_mount); 953 954 /* 955 * Convert the old and new sizes to the number of pages needed to 956 * store them. It may happen that we do not need to do anything 957 * because the last allocated page can accommodate the change on 958 * its own. 959 */ 960 oldsize = node->tn_size; 961 oldpages = round_page64(oldsize) / PAGE_SIZE; 962 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages); 963 newpages = round_page64(newsize) / PAGE_SIZE; 964 965 if (newpages > oldpages && 966 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) { 967 error = ENOSPC; 968 goto out; 969 } 970 node->tn_reg.tn_aobj_pages = newpages; 971 node->tn_size = newsize; 972 973 if (newpages != oldpages) 974 atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages)); 975 976 /* 977 * nvextflags to pass along for bdwrite() vs buwrite() 978 */ 979 if (vm_pages_needed || vm_paging_needed(0) || 980 tmpfs_bufcache_mode >= 2) { 981 nvextflags = 0; 982 } else { 983 nvextflags = NVEXTF_BUWRITE; 984 } 985 986 987 /* 988 * When adjusting the vnode filesize and its VM object we must 989 * also adjust our backing VM object (aobj). The blocksize 990 * used must match the block sized we use for the buffer cache. 991 * 992 * The backing VM object may contain VM pages as well as swap 993 * assignments if we previously renamed main object pages into 994 * it during deactivation. 995 * 996 * To make things easier tmpfs uses a blksize in multiples of 997 * PAGE_SIZE, and will only increase the blksize as a small file 998 * increases in size. Once a file has exceeded TMPFS_BLKSIZE (16KB), 999 * the blksize is maxed out. Truncating the file does not reduce 1000 * the blksize. 1001 */ 1002 if (newsize < oldsize) { 1003 vm_pindex_t osize; 1004 vm_pindex_t nsize; 1005 vm_object_t aobj; 1006 1007 error = nvtruncbuf(vp, newsize, node->tn_blksize, 1008 -1, nvextflags); 1009 aobj = node->tn_reg.tn_aobj; 1010 if (aobj) { 1011 osize = aobj->size; 1012 nsize = vp->v_object->size; 1013 if (nsize < osize) { 1014 aobj->size = osize; 1015 swap_pager_freespace(aobj, nsize, 1016 osize - nsize); 1017 vm_object_page_remove(aobj, nsize, osize, 1018 FALSE); 1019 } 1020 } 1021 } else { 1022 vm_object_t aobj; 1023 int nblksize; 1024 1025 /* 1026 * The first (and only the first) buffer in the file is resized 1027 * in multiples of PAGE_SIZE, up to TMPFS_BLKSIZE. 1028 */ 1029 nblksize = node->tn_blksize; 1030 while (nblksize < TMPFS_BLKSIZE && 1031 nblksize < newsize) { 1032 nblksize += PAGE_SIZE; 1033 } 1034 1035 if (trivial) 1036 nvextflags |= NVEXTF_TRIVIAL; 1037 1038 error = nvextendbuf(vp, oldsize, newsize, 1039 node->tn_blksize, nblksize, 1040 -1, -1, nvextflags); 1041 node->tn_blksize = nblksize; 1042 aobj = node->tn_reg.tn_aobj; 1043 if (aobj) 1044 aobj->size = vp->v_object->size; 1045 } 1046 1047 out: 1048 return error; 1049 } 1050 1051 /* --------------------------------------------------------------------- */ 1052 1053 /* 1054 * Change flags of the given vnode. 1055 * Caller should execute tmpfs_update on vp after a successful execution. 1056 * The vnode must be locked on entry and remain locked on exit. 1057 */ 1058 int 1059 tmpfs_chflags(struct vnode *vp, u_long vaflags, struct ucred *cred) 1060 { 1061 int error; 1062 struct tmpfs_node *node; 1063 int flags; 1064 1065 KKASSERT(vn_islocked(vp)); 1066 1067 node = VP_TO_TMPFS_NODE(vp); 1068 flags = node->tn_flags; 1069 1070 /* Disallow this operation if the file system is mounted read-only. */ 1071 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1072 return EROFS; 1073 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred); 1074 1075 /* Actually change the flags on the node itself */ 1076 if (error == 0) { 1077 TMPFS_NODE_LOCK(node); 1078 node->tn_flags = flags; 1079 node->tn_status |= TMPFS_NODE_CHANGED; 1080 TMPFS_NODE_UNLOCK(node); 1081 } 1082 1083 KKASSERT(vn_islocked(vp)); 1084 1085 return error; 1086 } 1087 1088 /* --------------------------------------------------------------------- */ 1089 1090 /* 1091 * Change access mode on the given vnode. 1092 * Caller should execute tmpfs_update on vp after a successful execution. 1093 * The vnode must be locked on entry and remain locked on exit. 1094 */ 1095 int 1096 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred) 1097 { 1098 struct tmpfs_node *node; 1099 mode_t cur_mode; 1100 int error; 1101 1102 KKASSERT(vn_islocked(vp)); 1103 1104 node = VP_TO_TMPFS_NODE(vp); 1105 1106 /* Disallow this operation if the file system is mounted read-only. */ 1107 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1108 return EROFS; 1109 1110 /* Immutable or append-only files cannot be modified, either. */ 1111 if (node->tn_flags & (IMMUTABLE | APPEND)) 1112 return EPERM; 1113 1114 cur_mode = node->tn_mode; 1115 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid, 1116 &cur_mode); 1117 1118 if (error == 0 && 1119 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) { 1120 TMPFS_NODE_LOCK(node); 1121 node->tn_mode &= ~ALLPERMS; 1122 node->tn_mode |= cur_mode & ALLPERMS; 1123 1124 node->tn_status |= TMPFS_NODE_CHANGED; 1125 TMPFS_NODE_UNLOCK(node); 1126 } 1127 1128 KKASSERT(vn_islocked(vp)); 1129 1130 return 0; 1131 } 1132 1133 /* --------------------------------------------------------------------- */ 1134 1135 /* 1136 * Change ownership of the given vnode. At least one of uid or gid must 1137 * be different than VNOVAL. If one is set to that value, the attribute 1138 * is unchanged. 1139 * Caller should execute tmpfs_update on vp after a successful execution. 1140 * The vnode must be locked on entry and remain locked on exit. 1141 */ 1142 int 1143 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 1144 { 1145 mode_t cur_mode; 1146 uid_t cur_uid; 1147 gid_t cur_gid; 1148 struct tmpfs_node *node; 1149 int error; 1150 1151 KKASSERT(vn_islocked(vp)); 1152 node = VP_TO_TMPFS_NODE(vp); 1153 1154 /* Disallow this operation if the file system is mounted read-only. */ 1155 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1156 return EROFS; 1157 1158 /* Immutable or append-only files cannot be modified, either. */ 1159 if (node->tn_flags & (IMMUTABLE | APPEND)) 1160 return EPERM; 1161 1162 cur_uid = node->tn_uid; 1163 cur_gid = node->tn_gid; 1164 cur_mode = node->tn_mode; 1165 error = vop_helper_chown(vp, uid, gid, cred, 1166 &cur_uid, &cur_gid, &cur_mode); 1167 1168 if (error == 0) { 1169 TMPFS_NODE_LOCK(node); 1170 if (cur_uid != node->tn_uid || 1171 cur_gid != node->tn_gid || 1172 cur_mode != node->tn_mode) { 1173 node->tn_uid = cur_uid; 1174 node->tn_gid = cur_gid; 1175 node->tn_mode = cur_mode; 1176 node->tn_status |= TMPFS_NODE_CHANGED; 1177 } 1178 TMPFS_NODE_UNLOCK(node); 1179 } 1180 1181 return error; 1182 } 1183 1184 /* --------------------------------------------------------------------- */ 1185 1186 /* 1187 * Change size of the given vnode. 1188 * Caller should execute tmpfs_update on vp after a successful execution. 1189 * The vnode must be locked on entry and remain locked on exit. 1190 */ 1191 int 1192 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred) 1193 { 1194 int error; 1195 struct tmpfs_node *node; 1196 1197 KKASSERT(vn_islocked(vp)); 1198 1199 node = VP_TO_TMPFS_NODE(vp); 1200 1201 /* Decide whether this is a valid operation based on the file type. */ 1202 error = 0; 1203 switch (vp->v_type) { 1204 case VDIR: 1205 return EISDIR; 1206 1207 case VREG: 1208 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1209 return EROFS; 1210 break; 1211 1212 case VBLK: 1213 /* FALLTHROUGH */ 1214 case VCHR: 1215 /* FALLTHROUGH */ 1216 case VFIFO: 1217 /* Allow modifications of special files even if in the file 1218 * system is mounted read-only (we are not modifying the 1219 * files themselves, but the objects they represent). */ 1220 return 0; 1221 1222 default: 1223 /* Anything else is unsupported. */ 1224 return EOPNOTSUPP; 1225 } 1226 1227 /* Immutable or append-only files cannot be modified, either. */ 1228 if (node->tn_flags & (IMMUTABLE | APPEND)) 1229 return EPERM; 1230 1231 error = tmpfs_truncate(vp, size); 1232 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1233 * for us, as will update tn_status; no need to do that here. */ 1234 1235 KKASSERT(vn_islocked(vp)); 1236 1237 return error; 1238 } 1239 1240 /* --------------------------------------------------------------------- */ 1241 1242 /* 1243 * Change access and modification times of the given vnode. 1244 * Caller should execute tmpfs_update on vp after a successful execution. 1245 * The vnode must be locked on entry and remain locked on exit. 1246 */ 1247 int 1248 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1249 int vaflags, struct ucred *cred) 1250 { 1251 struct tmpfs_node *node; 1252 1253 KKASSERT(vn_islocked(vp)); 1254 1255 node = VP_TO_TMPFS_NODE(vp); 1256 1257 /* Disallow this operation if the file system is mounted read-only. */ 1258 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1259 return EROFS; 1260 1261 /* Immutable or append-only files cannot be modified, either. */ 1262 if (node->tn_flags & (IMMUTABLE | APPEND)) 1263 return EPERM; 1264 1265 TMPFS_NODE_LOCK(node); 1266 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1267 node->tn_status |= TMPFS_NODE_ACCESSED; 1268 1269 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) { 1270 node->tn_status |= TMPFS_NODE_MODIFIED; 1271 vclrflags(vp, VLASTWRITETS); 1272 } 1273 1274 TMPFS_NODE_UNLOCK(node); 1275 1276 tmpfs_itimes(vp, atime, mtime); 1277 1278 KKASSERT(vn_islocked(vp)); 1279 1280 return 0; 1281 } 1282 1283 /* --------------------------------------------------------------------- */ 1284 /* Sync timestamps */ 1285 void 1286 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1287 const struct timespec *mod) 1288 { 1289 struct tmpfs_node *node; 1290 struct timespec now; 1291 1292 node = VP_TO_TMPFS_NODE(vp); 1293 1294 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1295 TMPFS_NODE_CHANGED)) == 0) { 1296 return; 1297 } 1298 1299 vfs_timestamp(&now); 1300 1301 TMPFS_NODE_LOCK(node); 1302 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1303 if (acc == NULL) 1304 acc = &now; 1305 node->tn_atime = acc->tv_sec; 1306 node->tn_atimensec = acc->tv_nsec; 1307 } 1308 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1309 if (mod == NULL) 1310 mod = &now; 1311 node->tn_mtime = mod->tv_sec; 1312 node->tn_mtimensec = mod->tv_nsec; 1313 } 1314 if (node->tn_status & TMPFS_NODE_CHANGED) { 1315 node->tn_ctime = now.tv_sec; 1316 node->tn_ctimensec = now.tv_nsec; 1317 } 1318 1319 node->tn_status &= ~(TMPFS_NODE_ACCESSED | 1320 TMPFS_NODE_MODIFIED | 1321 TMPFS_NODE_CHANGED); 1322 TMPFS_NODE_UNLOCK(node); 1323 } 1324 1325 /* --------------------------------------------------------------------- */ 1326 1327 void 1328 tmpfs_update(struct vnode *vp) 1329 { 1330 tmpfs_itimes(vp, NULL, NULL); 1331 } 1332 1333 /* --------------------------------------------------------------------- */ 1334 1335 /* 1336 * Caller must hold an exclusive node lock. 1337 */ 1338 int 1339 tmpfs_truncate(struct vnode *vp, off_t length) 1340 { 1341 int error; 1342 struct tmpfs_node *node; 1343 1344 node = VP_TO_TMPFS_NODE(vp); 1345 1346 if (length < 0) { 1347 error = EINVAL; 1348 goto out; 1349 } 1350 1351 if (node->tn_size == length) { 1352 error = 0; 1353 goto out; 1354 } 1355 1356 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1357 return (EFBIG); 1358 1359 1360 error = tmpfs_reg_resize(vp, length, 1); 1361 1362 if (error == 0) 1363 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1364 1365 out: 1366 tmpfs_update(vp); 1367 1368 return error; 1369 } 1370 1371 /* --------------------------------------------------------------------- */ 1372 1373 static ino_t 1374 tmpfs_fetch_ino(struct tmpfs_mount *tmp) 1375 { 1376 ino_t ret; 1377 1378 ret = atomic_fetchadd_64(&tmp->tm_ino, 1); 1379 1380 return (ret); 1381 } 1382 1383 static int 1384 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1385 { 1386 if (a->td_namelen > b->td_namelen) 1387 return 1; 1388 else if (a->td_namelen < b->td_namelen) 1389 return -1; 1390 else 1391 return strncmp(a->td_name, b->td_name, a->td_namelen); 1392 } 1393 1394 static int 1395 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1396 { 1397 if (a < b) 1398 return(-1); 1399 if (a > b) 1400 return(1); 1401 return 0; 1402 } 1403 1404 /* 1405 * Lock for rename. The namecache entries for the related terminal files 1406 * are already locked but the directories are not. A directory lock order 1407 * reversal is possible so use a deterministic order. 1408 * 1409 * Generally order path parent-to-child or using a simple pointer comparison. 1410 * Probably not perfect but it should catch most of the cases. 1411 * 1412 * Underlying files must be locked after the related directory. 1413 */ 1414 void 1415 tmpfs_lock4(struct tmpfs_node *node1, struct tmpfs_node *node2, 1416 struct tmpfs_node *node3, struct tmpfs_node *node4) 1417 { 1418 if (node1->tn_dir.tn_parent != node2 && 1419 (node1 < node2 || node2->tn_dir.tn_parent == node1)) { 1420 TMPFS_NODE_LOCK(node1); /* fdir */ 1421 TMPFS_NODE_LOCK(node3); /* ffile */ 1422 TMPFS_NODE_LOCK(node2); /* tdir */ 1423 if (node4) 1424 TMPFS_NODE_LOCK(node4); /* tfile */ 1425 } else { 1426 TMPFS_NODE_LOCK(node2); /* tdir */ 1427 if (node4) 1428 TMPFS_NODE_LOCK(node4); /* tfile */ 1429 TMPFS_NODE_LOCK(node1); /* fdir */ 1430 TMPFS_NODE_LOCK(node3); /* ffile */ 1431 } 1432 } 1433 1434 void 1435 tmpfs_unlock4(struct tmpfs_node *node1, struct tmpfs_node *node2, 1436 struct tmpfs_node *node3, struct tmpfs_node *node4) 1437 { 1438 if (node4) 1439 TMPFS_NODE_UNLOCK(node4); 1440 TMPFS_NODE_UNLOCK(node2); 1441 TMPFS_NODE_UNLOCK(node3); 1442 TMPFS_NODE_UNLOCK(node1); 1443 } 1444