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 #include <sys/malloc.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 #include <vm/vm_pageout.h> 53 #include <vm/vm_page2.h> 54 55 #include <vfs/tmpfs/tmpfs.h> 56 #include <vfs/tmpfs/tmpfs_vnops.h> 57 58 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *); 59 60 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a, 61 struct tmpfs_dirent *b); 62 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare); 63 64 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, 65 struct tmpfs_dirent *b); 66 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent, 67 rb_cookienode, tmpfs_dirtree_compare_cookie); 68 69 70 /* --------------------------------------------------------------------- */ 71 72 /* 73 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 74 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 75 * using the credentials of the process 'p'. 76 * 77 * If the node type is set to 'VDIR', then the parent parameter must point 78 * to the parent directory of the node being created. It may only be NULL 79 * while allocating the root node. 80 * 81 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 82 * specifies the device the node represents. 83 * 84 * If the node type is set to 'VLNK', then the parameter target specifies 85 * the file name of the target file for the symbolic link that is being 86 * created. 87 * 88 * Note that new nodes are retrieved from the available list if it has 89 * items or, if it is empty, from the node pool as long as there is enough 90 * space to create them. 91 * 92 * Returns zero on success or an appropriate error code on failure. 93 */ 94 int 95 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 96 uid_t uid, gid_t gid, mode_t mode, 97 char *target, int rmajor, int rminor, 98 struct tmpfs_node **node) 99 { 100 struct tmpfs_node *nnode; 101 struct timespec ts; 102 dev_t rdev; 103 104 KKASSERT(IFF(type == VLNK, target != NULL)); 105 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL)); 106 107 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 108 return (ENOSPC); 109 110 nnode = kmalloc_obj(sizeof(struct tmpfs_node), tmp->tm_node_zone, 111 M_WAITOK | M_ZERO | M_NULLOK); 112 if (nnode == NULL) 113 return (ENOSPC); 114 tmpfs_node_init(nnode); 115 116 /* Generic initialization. */ 117 nnode->tn_type = type; 118 vfs_timestamp(&ts); 119 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime 120 = ts.tv_sec; 121 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec 122 = ts.tv_nsec; 123 nnode->tn_uid = uid; 124 nnode->tn_gid = gid; 125 nnode->tn_mode = mode; 126 nnode->tn_id = tmpfs_fetch_ino(tmp); 127 nnode->tn_advlock.init_done = 0; 128 KKASSERT(nnode->tn_links == 0); 129 130 /* Type-specific initialization. */ 131 switch (nnode->tn_type) { 132 case VBLK: 133 case VCHR: 134 rdev = makeudev(rmajor, rminor); 135 if (rdev == NOUDEV) { 136 tmpfs_node_uninit(nnode); 137 kfree_obj(nnode, tmp->tm_node_zone); 138 return(EINVAL); 139 } 140 nnode->tn_rdev = rdev; 141 break; 142 143 case VDIR: 144 RB_INIT(&nnode->tn_dir.tn_dirtree); 145 RB_INIT(&nnode->tn_dir.tn_cookietree); 146 nnode->tn_dir.tn_parent = NULL; 147 nnode->tn_size = 0; 148 break; 149 150 case VFIFO: 151 /* FALLTHROUGH */ 152 case VSOCK: 153 break; 154 155 case VLNK: 156 nnode->tn_size = strlen(target); 157 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone, 158 M_WAITOK | M_NULLOK); 159 if (nnode->tn_link == NULL) { 160 tmpfs_node_uninit(nnode); 161 kfree_obj(nnode, tmp->tm_node_zone); 162 return (ENOSPC); 163 } 164 bcopy(target, nnode->tn_link, nnode->tn_size); 165 nnode->tn_link[nnode->tn_size] = '\0'; 166 break; 167 168 case VREG: 169 nnode->tn_reg.tn_aobj = swap_pager_alloc(NULL, 0, 170 VM_PROT_DEFAULT, 0); 171 nnode->tn_reg.tn_aobj_pages = 0; 172 nnode->tn_size = 0; 173 vm_object_set_flag(nnode->tn_reg.tn_aobj, OBJ_NOPAGEIN); 174 break; 175 176 default: 177 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 178 } 179 180 TMPFS_NODE_LOCK(nnode); 181 TMPFS_LOCK(tmp); 182 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 183 tmp->tm_nodes_inuse++; 184 TMPFS_UNLOCK(tmp); 185 TMPFS_NODE_UNLOCK(nnode); 186 187 *node = nnode; 188 return 0; 189 } 190 191 /* --------------------------------------------------------------------- */ 192 193 /* 194 * Destroys the node pointed to by node from the file system 'tmp'. 195 * If the node does not belong to the given mount point, the results are 196 * unpredicted. 197 * 198 * If the node references a directory; no entries are allowed because 199 * their removal could need a recursive algorithm, something forbidden in 200 * kernel space. Furthermore, there is not need to provide such 201 * functionality (recursive removal) because the only primitives offered 202 * to the user are the removal of empty directories and the deletion of 203 * individual files. 204 * 205 * Note that nodes are not really deleted; in fact, when a node has been 206 * allocated, it cannot be deleted during the whole life of the file 207 * system. Instead, they are moved to the available list and remain there 208 * until reused. 209 * 210 * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it. 211 */ 212 void 213 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 214 { 215 vm_pindex_t pages = 0; 216 217 #ifdef INVARIANTS 218 TMPFS_ASSERT_ELOCKED(node); 219 KKASSERT(node->tn_vnode == NULL); 220 #endif 221 TMPFS_LOCK(tmp); 222 LIST_REMOVE(node, tn_entries); 223 tmp->tm_nodes_inuse--; 224 TMPFS_UNLOCK(tmp); 225 TMPFS_NODE_UNLOCK(node); /* Caller has this lock */ 226 227 switch (node->tn_type) { 228 case VNON: 229 /* Do not do anything. VNON is provided to let the 230 * allocation routine clean itself easily by avoiding 231 * duplicating code in it. */ 232 /* FALLTHROUGH */ 233 case VBLK: 234 /* FALLTHROUGH */ 235 case VCHR: 236 /* FALLTHROUGH */ 237 break; 238 case VDIR: 239 /* 240 * The parent link can be NULL if this is the root 241 * node or if it is a directory node that was rmdir'd. 242 * 243 * XXX what if node is a directory which still contains 244 * directory entries (e.g. due to a forced umount) ? 245 */ 246 node->tn_size = 0; 247 KKASSERT(node->tn_dir.tn_parent == NULL); 248 249 /* 250 * If the root node is being destroyed don't leave a 251 * dangling pointer in tmpfs_mount. 252 */ 253 if (node == tmp->tm_root) 254 tmp->tm_root = NULL; 255 break; 256 case VFIFO: 257 /* FALLTHROUGH */ 258 case VSOCK: 259 break; 260 261 case VLNK: 262 kfree(node->tn_link, tmp->tm_name_zone); 263 node->tn_link = NULL; 264 node->tn_size = 0; 265 break; 266 267 case VREG: 268 if (node->tn_reg.tn_aobj != NULL) 269 vm_object_deallocate(node->tn_reg.tn_aobj); 270 node->tn_reg.tn_aobj = NULL; 271 pages = node->tn_reg.tn_aobj_pages; 272 break; 273 274 default: 275 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 276 } 277 278 /* 279 * Clean up fields as a safety before destroying the entry. 280 */ 281 tmpfs_node_uninit(node); 282 kfree_obj(node, tmp->tm_node_zone); 283 /* node is now invalid */ 284 285 if (pages) 286 atomic_add_long(&tmp->tm_pages_used, -(long)pages); 287 } 288 289 /* --------------------------------------------------------------------- */ 290 291 /* 292 * Allocates a new directory entry for the node node with a name of name. 293 * The new directory entry is returned in *de. 294 * 295 * The link count of node is increased by one to reflect the new object 296 * referencing it. 297 * 298 * Returns zero on success or an appropriate error code on failure. 299 */ 300 int 301 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 302 const char *name, uint16_t len, struct tmpfs_dirent **de) 303 { 304 struct tmpfs_dirent *nde; 305 306 nde = kmalloc_obj(sizeof(struct tmpfs_dirent), 307 tmp->tm_dirent_zone, M_WAITOK); 308 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK); 309 if (nde->td_name == NULL) { 310 kfree_obj(nde, tmp->tm_dirent_zone); 311 *de = NULL; 312 return (ENOSPC); 313 } 314 nde->td_namelen = len; 315 bcopy(name, nde->td_name, len); 316 nde->td_name[len] = '\0'; 317 318 nde->td_node = node; 319 320 atomic_add_int(&node->tn_links, 1); 321 322 *de = nde; 323 324 return 0; 325 } 326 327 /* --------------------------------------------------------------------- */ 328 329 /* 330 * Frees a directory entry. It is the caller's responsibility to destroy 331 * the node referenced by it if needed. 332 * 333 * The link count of node is decreased by one to reflect the removal of an 334 * object that referenced it. This only happens if 'node_exists' is true; 335 * otherwise the function will not access the node referred to by the 336 * directory entry, as it may already have been released from the outside. 337 */ 338 void 339 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 340 { 341 struct tmpfs_node *node; 342 343 node = de->td_node; 344 345 KKASSERT(node->tn_links > 0); 346 atomic_add_int(&node->tn_links, -1); 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 kfree_obj(de, tmp->tm_dirent_zone); 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 * The caller must ensure that node cannot go away (usually by holding 365 * the related directory entry). 366 * 367 * If dnode is non-NULL this routine avoids deadlocking against it but 368 * can return EAGAIN. Caller must try again. The dnode lock will cycle 369 * in this case, it remains locked on return in all cases. dnode must 370 * be shared-locked. 371 */ 372 int 373 tmpfs_alloc_vp(struct mount *mp, 374 struct tmpfs_node *dnode, struct tmpfs_node *node, int lkflag, 375 struct vnode **vpp) 376 { 377 int error = 0; 378 struct vnode *vp; 379 380 loop: 381 vp = NULL; 382 if (node->tn_vnode == NULL) { 383 error = getnewvnode(VT_TMPFS, mp, &vp, 384 VLKTIMEOUT, LK_CANRECURSE); 385 if (error) 386 goto out; 387 } 388 389 /* 390 * Interlocked extraction from node. This can race many things. 391 * We have to get a soft reference on the vnode while we hold 392 * the node locked, then acquire it properly and check for races. 393 */ 394 TMPFS_NODE_LOCK(node); 395 if (node->tn_vnode) { 396 if (vp) { 397 vp->v_type = VBAD; 398 vx_put(vp); 399 } 400 vp = node->tn_vnode; 401 402 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 403 vhold(vp); 404 TMPFS_NODE_UNLOCK(node); 405 406 if (dnode) { 407 /* 408 * Special-case handling to avoid deadlocking against 409 * dnode. This case has been validated and occurs 410 * every so often during synth builds. 411 */ 412 if (vget(vp, (lkflag & ~LK_RETRY) | 413 LK_NOWAIT | 414 LK_EXCLUSIVE) != 0) { 415 TMPFS_NODE_UNLOCK(dnode); 416 if (vget(vp, (lkflag & ~LK_RETRY) | 417 LK_SLEEPFAIL | 418 LK_EXCLUSIVE) == 0) { 419 kprintf("tmpfs: vp %p sleepfail\n", vp); 420 vput(vp); 421 } 422 vdrop(vp); 423 TMPFS_NODE_LOCK_SH(dnode); 424 425 return EAGAIN; 426 } 427 } else { 428 /* 429 * Normal path 430 */ 431 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) { 432 vdrop(vp); 433 goto loop; 434 } 435 } 436 if (node->tn_vnode != vp) { 437 vput(vp); 438 vdrop(vp); 439 goto loop; 440 } 441 vdrop(vp); 442 goto out; 443 } 444 445 /* 446 * We need to assign node->tn_vnode. If vp is NULL, loop up to 447 * allocate the vp. This can happen due to SMP races. 448 */ 449 if (vp == NULL) { 450 TMPFS_NODE_UNLOCK(node); 451 goto loop; 452 } 453 454 /* 455 * This should never happen. 456 */ 457 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) { 458 TMPFS_NODE_UNLOCK(node); 459 vp->v_type = VBAD; 460 vx_put(vp); 461 error = ENOENT; 462 goto out; 463 } 464 465 KKASSERT(node->tn_vnode == NULL); 466 KKASSERT(vp != NULL); 467 vp->v_data = node; 468 vp->v_type = node->tn_type; 469 470 /* Type-specific initialization. */ 471 switch (node->tn_type) { 472 case VBLK: 473 /* FALLTHROUGH */ 474 case VCHR: 475 /* FALLTHROUGH */ 476 case VSOCK: 477 break; 478 case VREG: 479 /* 480 * VMIO is mandatory. Tmpfs also supports KVABIO 481 * for its tmpfs_strategy(). 482 */ 483 vsetflags(vp, VKVABIO); 484 vinitvmio(vp, node->tn_size, node->tn_blksize, -1); 485 break; 486 case VLNK: 487 break; 488 case VFIFO: 489 vp->v_ops = &mp->mnt_vn_fifo_ops; 490 break; 491 case VDIR: 492 break; 493 494 default: 495 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 496 } 497 498 node->tn_vnode = vp; 499 TMPFS_NODE_UNLOCK(node); 500 501 vx_downgrade(vp); 502 out: 503 *vpp = vp; 504 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp))); 505 506 return error; 507 } 508 509 /* --------------------------------------------------------------------- */ 510 511 /* 512 * Allocates a new file of type 'type' and adds it to the parent directory 513 * 'dvp'; this addition is done using the component name given in 'cnp'. 514 * The ownership of the new file is automatically assigned based on the 515 * credentials of the caller (through 'cnp'), the group is set based on 516 * the parent directory and the mode is determined from the 'vap' argument. 517 * If successful, *vpp holds a vnode to the newly created file and zero 518 * is returned. Otherwise *vpp is NULL and the function returns an 519 * appropriate error code. 520 */ 521 int 522 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 523 struct namecache *ncp, struct ucred *cred, char *target) 524 { 525 int error; 526 struct tmpfs_dirent *de; 527 struct tmpfs_mount *tmp; 528 struct tmpfs_node *dnode; 529 struct tmpfs_node *node; 530 531 tmp = VFS_TO_TMPFS(dvp->v_mount); 532 dnode = VP_TO_TMPFS_DIR(dvp); 533 *vpp = NULL; 534 535 TMPFS_NODE_LOCK(dnode); 536 537 /* 538 * If the directory was removed but a process was CD'd into it, 539 * we do not allow any more file/dir creation within it. Otherwise 540 * we will lose track of it. 541 */ 542 KKASSERT(dnode->tn_type == VDIR); 543 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL) { 544 TMPFS_NODE_UNLOCK(dnode); 545 return ENOENT; 546 } 547 548 /* 549 * Make sure the link count does not overflow. 550 */ 551 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX) { 552 TMPFS_NODE_UNLOCK(dnode); 553 return EMLINK; 554 } 555 556 /* Allocate a node that represents the new file. */ 557 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid, 558 dnode->tn_gid, vap->va_mode, target, 559 vap->va_rmajor, vap->va_rminor, &node); 560 if (error != 0) { 561 TMPFS_NODE_UNLOCK(dnode); 562 return error; 563 } 564 TMPFS_NODE_LOCK(node); 565 566 /* Allocate a directory entry that points to the new file. */ 567 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de); 568 if (error != 0) { 569 TMPFS_NODE_UNLOCK(dnode); 570 tmpfs_free_node(tmp, node); 571 /* eats node lock */ 572 return error; 573 } 574 575 /* Allocate a vnode for the new file. */ 576 error = tmpfs_alloc_vp(dvp->v_mount, NULL, node, LK_EXCLUSIVE, vpp); 577 if (error != 0) { 578 TMPFS_NODE_UNLOCK(dnode); 579 tmpfs_free_dirent(tmp, de); 580 tmpfs_free_node(tmp, node); 581 /* eats node lock */ 582 return error; 583 } 584 585 /* 586 * Now that all required items are allocated, we can proceed to 587 * insert the new node into the directory, an operation that 588 * cannot fail. 589 */ 590 tmpfs_dir_attach_locked(dnode, de); 591 TMPFS_NODE_UNLOCK(dnode); 592 TMPFS_NODE_UNLOCK(node); 593 594 return error; 595 } 596 597 /* --------------------------------------------------------------------- */ 598 599 /* 600 * Attaches the directory entry de to the directory represented by dnode. 601 * Note that this does not change the link count of the node pointed by 602 * the directory entry, as this is done by tmpfs_alloc_dirent. 603 * 604 * dnode must be locked. 605 */ 606 void 607 tmpfs_dir_attach_locked(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 608 { 609 struct tmpfs_node *node = de->td_node; 610 struct tmpfs_dirent *de2; 611 612 if (node && node->tn_type == VDIR) { 613 TMPFS_NODE_LOCK(node); 614 atomic_add_int(&node->tn_links, 1); 615 node->tn_status |= TMPFS_NODE_CHANGED; 616 node->tn_dir.tn_parent = dnode; 617 atomic_add_int(&dnode->tn_links, 1); 618 TMPFS_NODE_UNLOCK(node); 619 } 620 de2 = RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 621 KASSERT(de2 == NULL, 622 ("tmpfs_dir_attach_lockedA: duplicate insertion of %p, has %p\n", 623 de, de2)); 624 de2 = RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 625 KASSERT(de2 == NULL, 626 ("tmpfs_dir_attach_lockedB: duplicate insertion of %p, has %p\n", 627 de, de2)); 628 dnode->tn_size += sizeof(struct tmpfs_dirent); 629 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 630 TMPFS_NODE_MODIFIED; 631 } 632 633 /* --------------------------------------------------------------------- */ 634 635 /* 636 * Detaches the directory entry de from the directory represented by dnode. 637 * Note that this does not change the link count of the node pointed by 638 * the directory entry, as this is done by tmpfs_free_dirent. 639 * 640 * dnode must be locked. 641 */ 642 void 643 tmpfs_dir_detach_locked(struct tmpfs_node *dnode, struct tmpfs_dirent *de) 644 { 645 struct tmpfs_node *node = de->td_node; 646 647 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de); 648 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de); 649 dnode->tn_size -= sizeof(struct tmpfs_dirent); 650 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 651 TMPFS_NODE_MODIFIED; 652 653 /* 654 * Clean out the tn_parent pointer immediately when removing a 655 * directory. 656 * 657 * Removal of the parent linkage also cleans out the extra tn_links 658 * count we had on both node and dnode. 659 * 660 * node can be NULL (typ during a forced umount), in which case 661 * the mount code is dealing with the linkages from a linked list 662 * scan. 663 */ 664 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) { 665 TMPFS_NODE_LOCK(node); 666 KKASSERT(node->tn_dir.tn_parent == dnode); 667 atomic_add_int(&dnode->tn_links, -1); 668 atomic_add_int(&node->tn_links, -1); 669 node->tn_dir.tn_parent = NULL; 670 TMPFS_NODE_UNLOCK(node); 671 } 672 } 673 674 /* --------------------------------------------------------------------- */ 675 676 /* 677 * Looks for a directory entry in the directory represented by node. 678 * 'ncp' describes the name of the entry to look for. Note that the . 679 * and .. components are not allowed as they do not physically exist 680 * within directories. 681 * 682 * Returns a pointer to the entry when found, otherwise NULL. 683 * 684 * Caller must hold the node locked (shared ok) 685 */ 686 struct tmpfs_dirent * 687 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 688 struct namecache *ncp) 689 { 690 struct tmpfs_dirent *de; 691 int len = ncp->nc_nlen; 692 struct tmpfs_dirent wanted; 693 694 wanted.td_namelen = len; 695 wanted.td_name = ncp->nc_name; 696 697 TMPFS_VALIDATE_DIR(node); 698 699 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted); 700 701 KASSERT((f == NULL || de == NULL || f == de->td_node), 702 ("tmpfs_dir_lookup: Incorrect node %p %p %p", 703 f, de, (de ? de->td_node : NULL))); 704 705 return de; 706 } 707 708 /* --------------------------------------------------------------------- */ 709 710 /* 711 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 712 * directory and returns it in the uio space. The function returns 0 713 * on success, -1 if there was not enough space in the uio structure to 714 * hold the directory entry or an appropriate error code if another 715 * error happens. 716 */ 717 int 718 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 719 { 720 int error; 721 722 TMPFS_VALIDATE_DIR(node); 723 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 724 725 if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, ".")) 726 return -1; 727 if (error == 0) 728 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 729 return error; 730 } 731 732 /* --------------------------------------------------------------------- */ 733 734 /* 735 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 736 * directory and returns it in the uio space. The function returns 0 737 * on success, -1 if there was not enough space in the uio structure to 738 * hold the directory entry or an appropriate error code if another 739 * error happens. 740 */ 741 int 742 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 743 struct uio *uio) 744 { 745 int error; 746 ino_t d_ino; 747 748 TMPFS_VALIDATE_DIR(node); 749 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 750 751 if (node->tn_dir.tn_parent) { 752 TMPFS_NODE_LOCK(node); 753 if (node->tn_dir.tn_parent) 754 d_ino = node->tn_dir.tn_parent->tn_id; 755 else 756 d_ino = tmp->tm_root->tn_id; 757 TMPFS_NODE_UNLOCK(node); 758 } else { 759 d_ino = tmp->tm_root->tn_id; 760 } 761 762 if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, "..")) 763 return -1; 764 if (error == 0) { 765 struct tmpfs_dirent *de; 766 de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree); 767 if (de == NULL) 768 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 769 else 770 uio->uio_offset = tmpfs_dircookie(de); 771 } 772 return error; 773 } 774 775 /* --------------------------------------------------------------------- */ 776 777 /* 778 * Lookup a directory entry by its associated cookie. 779 * 780 * Must be called with the directory node locked (shared ok) 781 */ 782 struct lubycookie_info { 783 off_t cookie; 784 struct tmpfs_dirent *de; 785 }; 786 787 static int 788 lubycookie_cmp(struct tmpfs_dirent *de, void *arg) 789 { 790 struct lubycookie_info *info = arg; 791 off_t cookie = tmpfs_dircookie(de); 792 793 if (cookie < info->cookie) 794 return(-1); 795 if (cookie > info->cookie) 796 return(1); 797 return(0); 798 } 799 800 static int 801 lubycookie_callback(struct tmpfs_dirent *de, void *arg) 802 { 803 struct lubycookie_info *info = arg; 804 805 if (tmpfs_dircookie(de) == info->cookie) { 806 info->de = de; 807 return(-1); 808 } 809 return(0); 810 } 811 812 struct tmpfs_dirent * 813 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie) 814 { 815 struct lubycookie_info info; 816 817 info.cookie = cookie; 818 info.de = NULL; 819 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree, 820 lubycookie_cmp, lubycookie_callback, &info); 821 return (info.de); 822 } 823 824 /* --------------------------------------------------------------------- */ 825 826 /* 827 * Helper function for tmpfs_readdir. Returns as much directory entries 828 * as can fit in the uio space. The read starts at uio->uio_offset. 829 * The function returns 0 on success, -1 if there was not enough space 830 * in the uio structure to hold the directory entry or an appropriate 831 * error code if another error happens. 832 * 833 * Caller must hold the node locked (shared ok) 834 */ 835 int 836 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp) 837 { 838 int error; 839 off_t startcookie; 840 struct tmpfs_dirent *de; 841 842 TMPFS_VALIDATE_DIR(node); 843 844 /* 845 * Locate the first directory entry we have to return. We have cached 846 * the last readdir in the node, so use those values if appropriate. 847 * Otherwise do a linear scan to find the requested entry. 848 */ 849 startcookie = uio->uio_offset; 850 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT); 851 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT); 852 853 if (startcookie == TMPFS_DIRCOOKIE_EOF) 854 return 0; 855 856 de = tmpfs_dir_lookupbycookie(node, startcookie); 857 if (de == NULL) 858 return EINVAL; 859 860 /* 861 * Read as much entries as possible; i.e., until we reach the end of 862 * the directory or we exhaust uio space. 863 */ 864 do { 865 ino_t d_ino; 866 uint8_t d_type; 867 868 /* Create a dirent structure representing the current 869 * tmpfs_node and fill it. */ 870 d_ino = de->td_node->tn_id; 871 switch (de->td_node->tn_type) { 872 case VBLK: 873 d_type = DT_BLK; 874 break; 875 876 case VCHR: 877 d_type = DT_CHR; 878 break; 879 880 case VDIR: 881 d_type = DT_DIR; 882 break; 883 884 case VFIFO: 885 d_type = DT_FIFO; 886 break; 887 888 case VLNK: 889 d_type = DT_LNK; 890 break; 891 892 case VREG: 893 d_type = DT_REG; 894 break; 895 896 case VSOCK: 897 d_type = DT_SOCK; 898 break; 899 900 default: 901 panic("tmpfs_dir_getdents: type %p %d", 902 de->td_node, (int)de->td_node->tn_type); 903 } 904 KKASSERT(de->td_namelen < 256); /* 255 + 1 */ 905 906 if (vop_write_dirent(&error, uio, d_ino, d_type, 907 de->td_namelen, de->td_name)) { 908 error = -1; 909 break; 910 } 911 912 (*cntp)++; 913 de = RB_NEXT(tmpfs_dirtree_cookie, 914 node->tn_dir.tn_cookietree, de); 915 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 916 917 /* Update the offset and cache. */ 918 if (de == NULL) { 919 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 920 } else { 921 uio->uio_offset = tmpfs_dircookie(de); 922 } 923 924 return error; 925 } 926 927 /* --------------------------------------------------------------------- */ 928 929 /* 930 * Resizes the aobj associated to the regular file pointed to by vp to 931 * the size newsize. 'vp' must point to a vnode that represents a regular 932 * file. 'newsize' must be positive. 933 * 934 * pass NVEXTF_TRIVIAL when buf content will be overwritten, otherwise set 0 935 * to be zero filled. 936 * 937 * Returns zero on success or an appropriate error code on failure. 938 * 939 * Caller must hold the node exclusively locked. 940 */ 941 int 942 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial) 943 { 944 int error; 945 vm_pindex_t newpages, oldpages; 946 struct tmpfs_mount *tmp; 947 struct tmpfs_node *node; 948 off_t oldsize; 949 int nvextflags; 950 951 #ifdef INVARIANTS 952 KKASSERT(vp->v_type == VREG); 953 KKASSERT(newsize >= 0); 954 #endif 955 956 node = VP_TO_TMPFS_NODE(vp); 957 tmp = VFS_TO_TMPFS(vp->v_mount); 958 959 /* 960 * Convert the old and new sizes to the number of pages needed to 961 * store them. It may happen that we do not need to do anything 962 * because the last allocated page can accommodate the change on 963 * its own. 964 */ 965 oldsize = node->tn_size; 966 oldpages = round_page64(oldsize) / PAGE_SIZE; 967 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages); 968 newpages = round_page64(newsize) / PAGE_SIZE; 969 970 if (newpages > oldpages && 971 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) { 972 error = ENOSPC; 973 goto out; 974 } 975 node->tn_reg.tn_aobj_pages = newpages; 976 node->tn_size = newsize; 977 978 if (newpages != oldpages) 979 atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages)); 980 981 /* 982 * nvextflags to pass along for bdwrite() vs buwrite(), this is 983 * so tmpfs activity doesn't eat memory being freed by the pageout 984 * daemon. 985 */ 986 if (vm_pages_needed || vm_paging_start(0) || 987 tmpfs_bufcache_mode >= 2) { 988 nvextflags = 0; 989 } else { 990 nvextflags = NVEXTF_BUWRITE; 991 } 992 993 994 /* 995 * When adjusting the vnode filesize and its VM object we must 996 * also adjust our backing VM object (aobj). The blocksize 997 * used must match the block sized we use for the buffer cache. 998 * 999 * The backing VM object may contain VM pages as well as swap 1000 * assignments if we previously renamed main object pages into 1001 * it during deactivation. 1002 * 1003 * To make things easier tmpfs uses a blksize in multiples of 1004 * PAGE_SIZE, and will only increase the blksize as a small file 1005 * increases in size. Once a file has exceeded TMPFS_BLKSIZE (16KB), 1006 * the blksize is maxed out. Truncating the file does not reduce 1007 * the blksize. 1008 */ 1009 if (newsize < oldsize) { 1010 vm_pindex_t osize; 1011 vm_pindex_t nsize; 1012 vm_object_t aobj; 1013 1014 error = nvtruncbuf(vp, newsize, node->tn_blksize, 1015 -1, nvextflags); 1016 aobj = node->tn_reg.tn_aobj; 1017 if (aobj) { 1018 osize = aobj->size; 1019 nsize = vp->v_object->size; 1020 if (nsize < osize) { 1021 aobj->size = osize; 1022 swap_pager_freespace(aobj, nsize, 1023 osize - nsize); 1024 vm_object_page_remove(aobj, nsize, osize, 1025 FALSE); 1026 } 1027 } 1028 } else { 1029 vm_object_t aobj; 1030 int nblksize; 1031 1032 /* 1033 * The first (and only the first) buffer in the file is resized 1034 * in multiples of PAGE_SIZE, up to TMPFS_BLKSIZE. 1035 */ 1036 nblksize = node->tn_blksize; 1037 while (nblksize < TMPFS_BLKSIZE && 1038 nblksize < newsize) { 1039 nblksize += PAGE_SIZE; 1040 } 1041 1042 if (trivial) 1043 nvextflags |= NVEXTF_TRIVIAL; 1044 1045 error = nvextendbuf(vp, oldsize, newsize, 1046 node->tn_blksize, nblksize, 1047 -1, -1, nvextflags); 1048 node->tn_blksize = nblksize; 1049 aobj = node->tn_reg.tn_aobj; 1050 if (aobj) 1051 aobj->size = vp->v_object->size; 1052 } 1053 1054 out: 1055 return error; 1056 } 1057 1058 /* --------------------------------------------------------------------- */ 1059 1060 /* 1061 * Change flags of the given vnode. 1062 * Caller should execute tmpfs_update on vp after a successful execution. 1063 * The vnode must be locked on entry and remain locked on exit. 1064 */ 1065 int 1066 tmpfs_chflags(struct vnode *vp, u_long vaflags, struct ucred *cred) 1067 { 1068 int error; 1069 struct tmpfs_node *node; 1070 int flags; 1071 1072 KKASSERT(vn_islocked(vp)); 1073 1074 node = VP_TO_TMPFS_NODE(vp); 1075 flags = node->tn_flags; 1076 1077 /* Disallow this operation if the file system is mounted read-only. */ 1078 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1079 return EROFS; 1080 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred); 1081 1082 /* Actually change the flags on the node itself */ 1083 if (error == 0) { 1084 TMPFS_NODE_LOCK(node); 1085 node->tn_flags = flags; 1086 node->tn_status |= TMPFS_NODE_CHANGED; 1087 TMPFS_NODE_UNLOCK(node); 1088 } 1089 1090 KKASSERT(vn_islocked(vp)); 1091 1092 return error; 1093 } 1094 1095 /* --------------------------------------------------------------------- */ 1096 1097 /* 1098 * Change access mode on the given vnode. 1099 * Caller should execute tmpfs_update on vp after a successful execution. 1100 * The vnode must be locked on entry and remain locked on exit. 1101 */ 1102 int 1103 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred) 1104 { 1105 struct tmpfs_node *node; 1106 mode_t cur_mode; 1107 int error; 1108 1109 KKASSERT(vn_islocked(vp)); 1110 1111 node = VP_TO_TMPFS_NODE(vp); 1112 1113 /* Disallow this operation if the file system is mounted read-only. */ 1114 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1115 return EROFS; 1116 1117 /* Immutable or append-only files cannot be modified, either. */ 1118 if (node->tn_flags & (IMMUTABLE | APPEND)) 1119 return EPERM; 1120 1121 cur_mode = node->tn_mode; 1122 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid, 1123 &cur_mode); 1124 1125 if (error == 0 && 1126 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) { 1127 TMPFS_NODE_LOCK(node); 1128 node->tn_mode &= ~ALLPERMS; 1129 node->tn_mode |= cur_mode & ALLPERMS; 1130 1131 node->tn_status |= TMPFS_NODE_CHANGED; 1132 TMPFS_NODE_UNLOCK(node); 1133 } 1134 1135 KKASSERT(vn_islocked(vp)); 1136 1137 return 0; 1138 } 1139 1140 /* --------------------------------------------------------------------- */ 1141 1142 /* 1143 * Change ownership of the given vnode. At least one of uid or gid must 1144 * be different than VNOVAL. If one is set to that value, the attribute 1145 * is unchanged. 1146 * Caller should execute tmpfs_update on vp after a successful execution. 1147 * The vnode must be locked on entry and remain locked on exit. 1148 */ 1149 int 1150 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred) 1151 { 1152 mode_t cur_mode; 1153 uid_t cur_uid; 1154 gid_t cur_gid; 1155 struct tmpfs_node *node; 1156 int error; 1157 1158 KKASSERT(vn_islocked(vp)); 1159 node = VP_TO_TMPFS_NODE(vp); 1160 1161 /* Disallow this operation if the file system is mounted read-only. */ 1162 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1163 return EROFS; 1164 1165 /* Immutable or append-only files cannot be modified, either. */ 1166 if (node->tn_flags & (IMMUTABLE | APPEND)) 1167 return EPERM; 1168 1169 cur_uid = node->tn_uid; 1170 cur_gid = node->tn_gid; 1171 cur_mode = node->tn_mode; 1172 error = vop_helper_chown(vp, uid, gid, cred, 1173 &cur_uid, &cur_gid, &cur_mode); 1174 1175 if (error == 0) { 1176 TMPFS_NODE_LOCK(node); 1177 if (cur_uid != node->tn_uid || 1178 cur_gid != node->tn_gid || 1179 cur_mode != node->tn_mode) { 1180 node->tn_uid = cur_uid; 1181 node->tn_gid = cur_gid; 1182 node->tn_mode = cur_mode; 1183 node->tn_status |= TMPFS_NODE_CHANGED; 1184 } 1185 TMPFS_NODE_UNLOCK(node); 1186 } 1187 1188 return error; 1189 } 1190 1191 /* --------------------------------------------------------------------- */ 1192 1193 /* 1194 * Change size of the given vnode. 1195 * Caller should execute tmpfs_update on vp after a successful execution. 1196 * The vnode must be locked on entry and remain locked on exit. 1197 */ 1198 int 1199 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred) 1200 { 1201 int error; 1202 struct tmpfs_node *node; 1203 1204 KKASSERT(vn_islocked(vp)); 1205 1206 node = VP_TO_TMPFS_NODE(vp); 1207 1208 /* Decide whether this is a valid operation based on the file type. */ 1209 error = 0; 1210 switch (vp->v_type) { 1211 case VDIR: 1212 return EISDIR; 1213 1214 case VREG: 1215 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1216 return EROFS; 1217 break; 1218 1219 case VBLK: 1220 /* FALLTHROUGH */ 1221 case VCHR: 1222 /* FALLTHROUGH */ 1223 case VFIFO: 1224 /* Allow modifications of special files even if in the file 1225 * system is mounted read-only (we are not modifying the 1226 * files themselves, but the objects they represent). */ 1227 return 0; 1228 1229 default: 1230 /* Anything else is unsupported. */ 1231 return EOPNOTSUPP; 1232 } 1233 1234 /* Immutable or append-only files cannot be modified, either. */ 1235 if (node->tn_flags & (IMMUTABLE | APPEND)) 1236 return EPERM; 1237 1238 error = tmpfs_truncate(vp, size); 1239 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1240 * for us, as will update tn_status; no need to do that here. */ 1241 1242 KKASSERT(vn_islocked(vp)); 1243 1244 return error; 1245 } 1246 1247 /* --------------------------------------------------------------------- */ 1248 1249 /* 1250 * Change access and modification times of the given vnode. 1251 * Caller should execute tmpfs_update on vp after a successful execution. 1252 * The vnode must be locked on entry and remain locked on exit. 1253 */ 1254 int 1255 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1256 int vaflags, struct ucred *cred) 1257 { 1258 struct tmpfs_node *node; 1259 1260 KKASSERT(vn_islocked(vp)); 1261 1262 node = VP_TO_TMPFS_NODE(vp); 1263 1264 /* Disallow this operation if the file system is mounted read-only. */ 1265 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1266 return EROFS; 1267 1268 /* Immutable or append-only files cannot be modified, either. */ 1269 if (node->tn_flags & (IMMUTABLE | APPEND)) 1270 return EPERM; 1271 1272 TMPFS_NODE_LOCK(node); 1273 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1274 node->tn_status |= TMPFS_NODE_ACCESSED; 1275 1276 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) { 1277 node->tn_status |= TMPFS_NODE_MODIFIED; 1278 vclrflags(vp, VLASTWRITETS); 1279 } 1280 1281 TMPFS_NODE_UNLOCK(node); 1282 1283 tmpfs_itimes(vp, atime, mtime); 1284 1285 KKASSERT(vn_islocked(vp)); 1286 1287 return 0; 1288 } 1289 1290 /* --------------------------------------------------------------------- */ 1291 /* Sync timestamps */ 1292 void 1293 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1294 const struct timespec *mod) 1295 { 1296 struct tmpfs_node *node; 1297 struct timespec now; 1298 1299 node = VP_TO_TMPFS_NODE(vp); 1300 1301 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1302 TMPFS_NODE_CHANGED)) == 0) { 1303 return; 1304 } 1305 1306 vfs_timestamp(&now); 1307 1308 TMPFS_NODE_LOCK(node); 1309 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1310 if (acc == NULL) 1311 acc = &now; 1312 node->tn_atime = acc->tv_sec; 1313 node->tn_atimensec = acc->tv_nsec; 1314 } 1315 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1316 if (mod == NULL) 1317 mod = &now; 1318 node->tn_mtime = mod->tv_sec; 1319 node->tn_mtimensec = mod->tv_nsec; 1320 } 1321 if (node->tn_status & TMPFS_NODE_CHANGED) { 1322 node->tn_ctime = now.tv_sec; 1323 node->tn_ctimensec = now.tv_nsec; 1324 } 1325 1326 node->tn_status &= ~(TMPFS_NODE_ACCESSED | 1327 TMPFS_NODE_MODIFIED | 1328 TMPFS_NODE_CHANGED); 1329 TMPFS_NODE_UNLOCK(node); 1330 } 1331 1332 /* --------------------------------------------------------------------- */ 1333 1334 void 1335 tmpfs_update(struct vnode *vp) 1336 { 1337 tmpfs_itimes(vp, NULL, NULL); 1338 } 1339 1340 /* --------------------------------------------------------------------- */ 1341 1342 /* 1343 * Caller must hold an exclusive node lock. 1344 */ 1345 int 1346 tmpfs_truncate(struct vnode *vp, off_t length) 1347 { 1348 int error; 1349 struct tmpfs_node *node; 1350 1351 node = VP_TO_TMPFS_NODE(vp); 1352 1353 if (length < 0) { 1354 error = EINVAL; 1355 goto out; 1356 } 1357 1358 if (node->tn_size == length) { 1359 error = 0; 1360 goto out; 1361 } 1362 1363 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1364 return (EFBIG); 1365 1366 1367 error = tmpfs_reg_resize(vp, length, 1); 1368 1369 if (error == 0) 1370 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1371 1372 out: 1373 tmpfs_update(vp); 1374 1375 return error; 1376 } 1377 1378 /* --------------------------------------------------------------------- */ 1379 1380 static ino_t 1381 tmpfs_fetch_ino(struct tmpfs_mount *tmp) 1382 { 1383 ino_t ret; 1384 1385 ret = atomic_fetchadd_64(&tmp->tm_ino, 1); 1386 1387 return (ret); 1388 } 1389 1390 static int 1391 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1392 { 1393 if (a->td_namelen > b->td_namelen) 1394 return 1; 1395 else if (a->td_namelen < b->td_namelen) 1396 return -1; 1397 else 1398 return strncmp(a->td_name, b->td_name, a->td_namelen); 1399 } 1400 1401 static int 1402 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1403 { 1404 if (a < b) 1405 return(-1); 1406 if (a > b) 1407 return(1); 1408 return 0; 1409 } 1410 1411 /* 1412 * Lock for rename. The namecache entries for the related terminal files 1413 * are already locked but the directories are not. A directory lock order 1414 * reversal is possible so use a deterministic order. 1415 * 1416 * Generally order path parent-to-child or using a simple pointer comparison. 1417 * Probably not perfect but it should catch most of the cases. 1418 * 1419 * Underlying files must be locked after the related directory. 1420 */ 1421 void 1422 tmpfs_lock4(struct tmpfs_node *node1, struct tmpfs_node *node2, 1423 struct tmpfs_node *node3, struct tmpfs_node *node4) 1424 { 1425 if (node1->tn_dir.tn_parent != node2 && 1426 (node1 < node2 || node2->tn_dir.tn_parent == node1)) { 1427 TMPFS_NODE_LOCK(node1); /* fdir */ 1428 TMPFS_NODE_LOCK(node3); /* ffile */ 1429 TMPFS_NODE_LOCK(node2); /* tdir */ 1430 if (node4) 1431 TMPFS_NODE_LOCK(node4); /* tfile */ 1432 } else { 1433 TMPFS_NODE_LOCK(node2); /* tdir */ 1434 if (node4) 1435 TMPFS_NODE_LOCK(node4); /* tfile */ 1436 TMPFS_NODE_LOCK(node1); /* fdir */ 1437 TMPFS_NODE_LOCK(node3); /* ffile */ 1438 } 1439 } 1440 1441 void 1442 tmpfs_unlock4(struct tmpfs_node *node1, struct tmpfs_node *node2, 1443 struct tmpfs_node *node3, struct tmpfs_node *node4) 1444 { 1445 if (node4) 1446 TMPFS_NODE_UNLOCK(node4); 1447 TMPFS_NODE_UNLOCK(node2); 1448 TMPFS_NODE_UNLOCK(node3); 1449 TMPFS_NODE_UNLOCK(node1); 1450 } 1451