1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 2005 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 9 * 2005 program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Efficient memory file system supporting functions. 35 */ 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 #include <sys/param.h> 40 #include <sys/namei.h> 41 #include <sys/priv.h> 42 #include <sys/proc.h> 43 #include <sys/stat.h> 44 #include <sys/systm.h> 45 #include <sys/vnode.h> 46 #include <sys/vmmeter.h> 47 48 #include <vm/vm.h> 49 #include <vm/vm_object.h> 50 #include <vm/vm_page.h> 51 #include <vm/vm_pager.h> 52 #include <vm/vm_extern.h> 53 54 #include <fs/tmpfs/tmpfs.h> 55 #include <fs/tmpfs/tmpfs_fifoops.h> 56 #include <fs/tmpfs/tmpfs_vnops.h> 57 58 /* --------------------------------------------------------------------- */ 59 60 /* 61 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 62 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 63 * using the credentials of the process 'p'. 64 * 65 * If the node type is set to 'VDIR', then the parent parameter must point 66 * to the parent directory of the node being created. It may only be NULL 67 * while allocating the root node. 68 * 69 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 70 * specifies the device the node represents. 71 * 72 * If the node type is set to 'VLNK', then the parameter target specifies 73 * the file name of the target file for the symbolic link that is being 74 * created. 75 * 76 * Note that new nodes are retrieved from the available list if it has 77 * items or, if it is empty, from the node pool as long as there is enough 78 * space to create them. 79 * 80 * Returns zero on success or an appropriate error code on failure. 81 */ 82 int 83 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 84 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 85 char *target, dev_t rdev, struct tmpfs_node **node) 86 { 87 struct tmpfs_node *nnode; 88 89 /* If the root directory of the 'tmp' file system is not yet 90 * allocated, this must be the request to do it. */ 91 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 92 93 MPASS(IFF(type == VLNK, target != NULL)); 94 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL)); 95 96 if (tmp->tm_nodes_inuse > tmp->tm_nodes_max) 97 return (ENOSPC); 98 99 nnode = (struct tmpfs_node *)uma_zalloc_arg( 100 tmp->tm_node_pool, tmp, M_WAITOK); 101 102 /* Generic initialization. */ 103 nnode->tn_type = type; 104 vfs_timestamp(&nnode->tn_atime); 105 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 106 nnode->tn_atime; 107 nnode->tn_uid = uid; 108 nnode->tn_gid = gid; 109 nnode->tn_mode = mode; 110 nnode->tn_id = alloc_unr(tmp->tm_ino_unr); 111 112 /* Type-specific initialization. */ 113 switch (nnode->tn_type) { 114 case VBLK: 115 case VCHR: 116 nnode->tn_rdev = rdev; 117 break; 118 119 case VDIR: 120 TAILQ_INIT(&nnode->tn_dir.tn_dirhead); 121 MPASS(parent != nnode); 122 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL)); 123 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent; 124 nnode->tn_dir.tn_readdir_lastn = 0; 125 nnode->tn_dir.tn_readdir_lastp = NULL; 126 nnode->tn_links++; 127 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent); 128 nnode->tn_dir.tn_parent->tn_links++; 129 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent); 130 break; 131 132 case VFIFO: 133 /* FALLTHROUGH */ 134 case VSOCK: 135 break; 136 137 case VLNK: 138 MPASS(strlen(target) < MAXPATHLEN); 139 nnode->tn_size = strlen(target); 140 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME, 141 M_WAITOK); 142 memcpy(nnode->tn_link, target, nnode->tn_size); 143 break; 144 145 case VREG: 146 nnode->tn_reg.tn_aobj = 147 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0, 148 NULL /* XXXKIB - tmpfs needs swap reservation */); 149 nnode->tn_reg.tn_aobj_pages = 0; 150 break; 151 152 default: 153 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 154 } 155 156 TMPFS_LOCK(tmp); 157 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 158 tmp->tm_nodes_inuse++; 159 TMPFS_UNLOCK(tmp); 160 161 *node = nnode; 162 return 0; 163 } 164 165 /* --------------------------------------------------------------------- */ 166 167 /* 168 * Destroys the node pointed to by node from the file system 'tmp'. 169 * If the node does not belong to the given mount point, the results are 170 * unpredicted. 171 * 172 * If the node references a directory; no entries are allowed because 173 * their removal could need a recursive algorithm, something forbidden in 174 * kernel space. Furthermore, there is not need to provide such 175 * functionality (recursive removal) because the only primitives offered 176 * to the user are the removal of empty directories and the deletion of 177 * individual files. 178 * 179 * Note that nodes are not really deleted; in fact, when a node has been 180 * allocated, it cannot be deleted during the whole life of the file 181 * system. Instead, they are moved to the available list and remain there 182 * until reused. 183 */ 184 void 185 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 186 { 187 size_t pages = 0; 188 189 #ifdef INVARIANTS 190 TMPFS_NODE_LOCK(node); 191 MPASS(node->tn_vnode == NULL); 192 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 193 TMPFS_NODE_UNLOCK(node); 194 #endif 195 196 TMPFS_LOCK(tmp); 197 LIST_REMOVE(node, tn_entries); 198 tmp->tm_nodes_inuse--; 199 TMPFS_UNLOCK(tmp); 200 201 switch (node->tn_type) { 202 case VNON: 203 /* Do not do anything. VNON is provided to let the 204 * allocation routine clean itself easily by avoiding 205 * duplicating code in it. */ 206 /* FALLTHROUGH */ 207 case VBLK: 208 /* FALLTHROUGH */ 209 case VCHR: 210 /* FALLTHROUGH */ 211 case VDIR: 212 /* FALLTHROUGH */ 213 case VFIFO: 214 /* FALLTHROUGH */ 215 case VSOCK: 216 break; 217 218 case VLNK: 219 free(node->tn_link, M_TMPFSNAME); 220 break; 221 222 case VREG: 223 if (node->tn_reg.tn_aobj != NULL) 224 vm_object_deallocate(node->tn_reg.tn_aobj); 225 pages = node->tn_reg.tn_aobj_pages; 226 break; 227 228 default: 229 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 230 } 231 232 free_unr(tmp->tm_ino_unr, node->tn_id); 233 uma_zfree(tmp->tm_node_pool, node); 234 235 TMPFS_LOCK(tmp); 236 tmp->tm_pages_used -= pages; 237 TMPFS_UNLOCK(tmp); 238 } 239 240 /* --------------------------------------------------------------------- */ 241 242 /* 243 * Allocates a new directory entry for the node node with a name of name. 244 * The new directory entry is returned in *de. 245 * 246 * The link count of node is increased by one to reflect the new object 247 * referencing it. 248 * 249 * Returns zero on success or an appropriate error code on failure. 250 */ 251 int 252 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 253 const char *name, uint16_t len, struct tmpfs_dirent **de) 254 { 255 struct tmpfs_dirent *nde; 256 257 nde = (struct tmpfs_dirent *)uma_zalloc( 258 tmp->tm_dirent_pool, M_WAITOK); 259 nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 260 nde->td_namelen = len; 261 memcpy(nde->td_name, name, len); 262 263 nde->td_node = node; 264 node->tn_links++; 265 266 *de = nde; 267 268 return 0; 269 } 270 271 /* --------------------------------------------------------------------- */ 272 273 /* 274 * Frees a directory entry. It is the caller's responsibility to destroy 275 * the node referenced by it if needed. 276 * 277 * The link count of node is decreased by one to reflect the removal of an 278 * object that referenced it. This only happens if 'node_exists' is true; 279 * otherwise the function will not access the node referred to by the 280 * directory entry, as it may already have been released from the outside. 281 */ 282 void 283 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de, 284 boolean_t node_exists) 285 { 286 if (node_exists) { 287 struct tmpfs_node *node; 288 289 node = de->td_node; 290 291 MPASS(node->tn_links > 0); 292 node->tn_links--; 293 } 294 295 free(de->td_name, M_TMPFSNAME); 296 uma_zfree(tmp->tm_dirent_pool, de); 297 } 298 299 /* --------------------------------------------------------------------- */ 300 301 /* 302 * Allocates a new vnode for the node node or returns a new reference to 303 * an existing one if the node had already a vnode referencing it. The 304 * resulting locked vnode is returned in *vpp. 305 * 306 * Returns zero on success or an appropriate error code on failure. 307 */ 308 int 309 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 310 struct vnode **vpp) 311 { 312 int error = 0; 313 struct vnode *vp; 314 315 loop: 316 TMPFS_NODE_LOCK(node); 317 if ((vp = node->tn_vnode) != NULL) { 318 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 319 VI_LOCK(vp); 320 TMPFS_NODE_UNLOCK(node); 321 vholdl(vp); 322 (void) vget(vp, lkflag | LK_INTERLOCK | LK_RETRY, curthread); 323 vdrop(vp); 324 325 /* 326 * Make sure the vnode is still there after 327 * getting the interlock to avoid racing a free. 328 */ 329 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 330 vput(vp); 331 goto loop; 332 } 333 334 goto out; 335 } 336 337 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 338 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 339 TMPFS_NODE_UNLOCK(node); 340 error = ENOENT; 341 vp = NULL; 342 goto out; 343 } 344 345 /* 346 * otherwise lock the vp list while we call getnewvnode 347 * since that can block. 348 */ 349 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 350 node->tn_vpstate |= TMPFS_VNODE_WANT; 351 error = msleep((caddr_t) &node->tn_vpstate, 352 TMPFS_NODE_MTX(node), PDROP | PCATCH, 353 "tmpfs_alloc_vp", 0); 354 if (error) 355 return error; 356 357 goto loop; 358 } else 359 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 360 361 TMPFS_NODE_UNLOCK(node); 362 363 /* Get a new vnode and associate it with our node. */ 364 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); 365 if (error != 0) 366 goto unlock; 367 MPASS(vp != NULL); 368 369 (void) vn_lock(vp, lkflag | LK_RETRY); 370 371 vp->v_data = node; 372 vp->v_type = node->tn_type; 373 374 /* Type-specific initialization. */ 375 switch (node->tn_type) { 376 case VBLK: 377 /* FALLTHROUGH */ 378 case VCHR: 379 /* FALLTHROUGH */ 380 case VLNK: 381 /* FALLTHROUGH */ 382 case VREG: 383 /* FALLTHROUGH */ 384 case VSOCK: 385 break; 386 case VFIFO: 387 vp->v_op = &tmpfs_fifoop_entries; 388 break; 389 case VDIR: 390 MPASS(node->tn_dir.tn_parent != NULL); 391 if (node->tn_dir.tn_parent == node) 392 vp->v_vflag |= VV_ROOT; 393 break; 394 395 default: 396 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 397 } 398 399 vnode_pager_setsize(vp, node->tn_size); 400 error = insmntque(vp, mp); 401 if (error) 402 vp = NULL; 403 404 unlock: 405 TMPFS_NODE_LOCK(node); 406 407 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 408 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 409 node->tn_vnode = vp; 410 411 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 412 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 413 TMPFS_NODE_UNLOCK(node); 414 wakeup((caddr_t) &node->tn_vpstate); 415 } else 416 TMPFS_NODE_UNLOCK(node); 417 418 out: 419 *vpp = vp; 420 421 MPASS(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp))); 422 #ifdef INVARIANTS 423 TMPFS_NODE_LOCK(node); 424 MPASS(*vpp == node->tn_vnode); 425 TMPFS_NODE_UNLOCK(node); 426 #endif 427 428 return error; 429 } 430 431 /* --------------------------------------------------------------------- */ 432 433 /* 434 * Destroys the association between the vnode vp and the node it 435 * references. 436 */ 437 void 438 tmpfs_free_vp(struct vnode *vp) 439 { 440 struct tmpfs_node *node; 441 442 node = VP_TO_TMPFS_NODE(vp); 443 444 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED); 445 node->tn_vnode = NULL; 446 vp->v_data = NULL; 447 } 448 449 /* --------------------------------------------------------------------- */ 450 451 /* 452 * Allocates a new file of type 'type' and adds it to the parent directory 453 * 'dvp'; this addition is done using the component name given in 'cnp'. 454 * The ownership of the new file is automatically assigned based on the 455 * credentials of the caller (through 'cnp'), the group is set based on 456 * the parent directory and the mode is determined from the 'vap' argument. 457 * If successful, *vpp holds a vnode to the newly created file and zero 458 * is returned. Otherwise *vpp is NULL and the function returns an 459 * appropriate error code. 460 */ 461 int 462 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 463 struct componentname *cnp, char *target) 464 { 465 int error; 466 struct tmpfs_dirent *de; 467 struct tmpfs_mount *tmp; 468 struct tmpfs_node *dnode; 469 struct tmpfs_node *node; 470 struct tmpfs_node *parent; 471 472 MPASS(VOP_ISLOCKED(dvp)); 473 MPASS(cnp->cn_flags & HASBUF); 474 475 tmp = VFS_TO_TMPFS(dvp->v_mount); 476 dnode = VP_TO_TMPFS_DIR(dvp); 477 *vpp = NULL; 478 479 /* If the entry we are creating is a directory, we cannot overflow 480 * the number of links of its parent, because it will get a new 481 * link. */ 482 if (vap->va_type == VDIR) { 483 /* Ensure that we do not overflow the maximum number of links 484 * imposed by the system. */ 485 MPASS(dnode->tn_links <= LINK_MAX); 486 if (dnode->tn_links == LINK_MAX) { 487 error = EMLINK; 488 goto out; 489 } 490 491 parent = dnode; 492 MPASS(parent != NULL); 493 } else 494 parent = NULL; 495 496 /* Allocate a node that represents the new file. */ 497 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid, 498 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node); 499 if (error != 0) 500 goto out; 501 502 /* Allocate a directory entry that points to the new file. */ 503 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 504 &de); 505 if (error != 0) { 506 tmpfs_free_node(tmp, node); 507 goto out; 508 } 509 510 /* Allocate a vnode for the new file. */ 511 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 512 if (error != 0) { 513 tmpfs_free_dirent(tmp, de, TRUE); 514 tmpfs_free_node(tmp, node); 515 goto out; 516 } 517 518 /* Now that all required items are allocated, we can proceed to 519 * insert the new node into the directory, an operation that 520 * cannot fail. */ 521 tmpfs_dir_attach(dvp, de); 522 523 out: 524 525 return error; 526 } 527 528 /* --------------------------------------------------------------------- */ 529 530 /* 531 * Attaches the directory entry de to the directory represented by vp. 532 * Note that this does not change the link count of the node pointed by 533 * the directory entry, as this is done by tmpfs_alloc_dirent. 534 */ 535 void 536 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 537 { 538 struct tmpfs_node *dnode; 539 540 ASSERT_VOP_ELOCKED(vp, __func__); 541 dnode = VP_TO_TMPFS_DIR(vp); 542 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries); 543 dnode->tn_size += sizeof(struct tmpfs_dirent); 544 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 545 TMPFS_NODE_MODIFIED; 546 } 547 548 /* --------------------------------------------------------------------- */ 549 550 /* 551 * Detaches the directory entry de from the directory represented by vp. 552 * Note that this does not change the link count of the node pointed by 553 * the directory entry, as this is done by tmpfs_free_dirent. 554 */ 555 void 556 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 557 { 558 struct tmpfs_node *dnode; 559 560 ASSERT_VOP_ELOCKED(vp, __func__); 561 dnode = VP_TO_TMPFS_DIR(vp); 562 563 if (dnode->tn_dir.tn_readdir_lastp == de) { 564 dnode->tn_dir.tn_readdir_lastn = 0; 565 dnode->tn_dir.tn_readdir_lastp = NULL; 566 } 567 568 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries); 569 dnode->tn_size -= sizeof(struct tmpfs_dirent); 570 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 571 TMPFS_NODE_MODIFIED; 572 } 573 574 /* --------------------------------------------------------------------- */ 575 576 /* 577 * Looks for a directory entry in the directory represented by node. 578 * 'cnp' describes the name of the entry to look for. Note that the . 579 * and .. components are not allowed as they do not physically exist 580 * within directories. 581 * 582 * Returns a pointer to the entry when found, otherwise NULL. 583 */ 584 struct tmpfs_dirent * 585 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 586 struct componentname *cnp) 587 { 588 boolean_t found; 589 struct tmpfs_dirent *de; 590 591 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 592 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 593 cnp->cn_nameptr[1] == '.'))); 594 TMPFS_VALIDATE_DIR(node); 595 596 found = 0; 597 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) { 598 if (f != NULL && de->td_node != f) 599 continue; 600 MPASS(cnp->cn_namelen < 0xffff); 601 if (de->td_namelen == (uint16_t)cnp->cn_namelen && 602 bcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) { 603 found = 1; 604 break; 605 } 606 } 607 node->tn_status |= TMPFS_NODE_ACCESSED; 608 609 return found ? de : NULL; 610 } 611 612 /* --------------------------------------------------------------------- */ 613 614 /* 615 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 616 * directory and returns it in the uio space. The function returns 0 617 * on success, -1 if there was not enough space in the uio structure to 618 * hold the directory entry or an appropriate error code if another 619 * error happens. 620 */ 621 int 622 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 623 { 624 int error; 625 struct dirent dent; 626 627 TMPFS_VALIDATE_DIR(node); 628 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 629 630 dent.d_fileno = node->tn_id; 631 dent.d_type = DT_DIR; 632 dent.d_namlen = 1; 633 dent.d_name[0] = '.'; 634 dent.d_name[1] = '\0'; 635 dent.d_reclen = GENERIC_DIRSIZ(&dent); 636 637 if (dent.d_reclen > uio->uio_resid) 638 error = -1; 639 else { 640 error = uiomove(&dent, dent.d_reclen, uio); 641 if (error == 0) 642 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 643 } 644 645 node->tn_status |= TMPFS_NODE_ACCESSED; 646 647 return error; 648 } 649 650 /* --------------------------------------------------------------------- */ 651 652 /* 653 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 654 * directory and returns it in the uio space. The function returns 0 655 * on success, -1 if there was not enough space in the uio structure to 656 * hold the directory entry or an appropriate error code if another 657 * error happens. 658 */ 659 int 660 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 661 { 662 int error; 663 struct dirent dent; 664 665 TMPFS_VALIDATE_DIR(node); 666 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 667 668 /* 669 * Return ENOENT if the current node is already removed. 670 */ 671 TMPFS_ASSERT_LOCKED(node); 672 if (node->tn_dir.tn_parent == NULL) { 673 return (ENOENT); 674 } 675 676 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 677 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 678 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 679 680 dent.d_type = DT_DIR; 681 dent.d_namlen = 2; 682 dent.d_name[0] = '.'; 683 dent.d_name[1] = '.'; 684 dent.d_name[2] = '\0'; 685 dent.d_reclen = GENERIC_DIRSIZ(&dent); 686 687 if (dent.d_reclen > uio->uio_resid) 688 error = -1; 689 else { 690 error = uiomove(&dent, dent.d_reclen, uio); 691 if (error == 0) { 692 struct tmpfs_dirent *de; 693 694 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead); 695 if (de == NULL) 696 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 697 else 698 uio->uio_offset = tmpfs_dircookie(de); 699 } 700 } 701 702 node->tn_status |= TMPFS_NODE_ACCESSED; 703 704 return error; 705 } 706 707 /* --------------------------------------------------------------------- */ 708 709 /* 710 * Lookup a directory entry by its associated cookie. 711 */ 712 struct tmpfs_dirent * 713 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie) 714 { 715 struct tmpfs_dirent *de; 716 717 if (cookie == node->tn_dir.tn_readdir_lastn && 718 node->tn_dir.tn_readdir_lastp != NULL) { 719 return node->tn_dir.tn_readdir_lastp; 720 } 721 722 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) { 723 if (tmpfs_dircookie(de) == cookie) { 724 break; 725 } 726 } 727 728 return de; 729 } 730 731 /* --------------------------------------------------------------------- */ 732 733 /* 734 * Helper function for tmpfs_readdir. Returns as much directory entries 735 * as can fit in the uio space. The read starts at uio->uio_offset. 736 * The function returns 0 on success, -1 if there was not enough space 737 * in the uio structure to hold the directory entry or an appropriate 738 * error code if another error happens. 739 */ 740 int 741 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp) 742 { 743 int error; 744 off_t startcookie; 745 struct tmpfs_dirent *de; 746 747 TMPFS_VALIDATE_DIR(node); 748 749 /* Locate the first directory entry we have to return. We have cached 750 * the last readdir in the node, so use those values if appropriate. 751 * Otherwise do a linear scan to find the requested entry. */ 752 startcookie = uio->uio_offset; 753 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT); 754 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT); 755 if (startcookie == TMPFS_DIRCOOKIE_EOF) { 756 return 0; 757 } else { 758 de = tmpfs_dir_lookupbycookie(node, startcookie); 759 } 760 if (de == NULL) { 761 return EINVAL; 762 } 763 764 /* Read as much entries as possible; i.e., until we reach the end of 765 * the directory or we exhaust uio space. */ 766 do { 767 struct dirent d; 768 769 /* Create a dirent structure representing the current 770 * tmpfs_node and fill it. */ 771 d.d_fileno = de->td_node->tn_id; 772 switch (de->td_node->tn_type) { 773 case VBLK: 774 d.d_type = DT_BLK; 775 break; 776 777 case VCHR: 778 d.d_type = DT_CHR; 779 break; 780 781 case VDIR: 782 d.d_type = DT_DIR; 783 break; 784 785 case VFIFO: 786 d.d_type = DT_FIFO; 787 break; 788 789 case VLNK: 790 d.d_type = DT_LNK; 791 break; 792 793 case VREG: 794 d.d_type = DT_REG; 795 break; 796 797 case VSOCK: 798 d.d_type = DT_SOCK; 799 break; 800 801 default: 802 panic("tmpfs_dir_getdents: type %p %d", 803 de->td_node, (int)de->td_node->tn_type); 804 } 805 d.d_namlen = de->td_namelen; 806 MPASS(de->td_namelen < sizeof(d.d_name)); 807 (void)memcpy(d.d_name, de->td_name, de->td_namelen); 808 d.d_name[de->td_namelen] = '\0'; 809 d.d_reclen = GENERIC_DIRSIZ(&d); 810 811 /* Stop reading if the directory entry we are treating is 812 * bigger than the amount of data that can be returned. */ 813 if (d.d_reclen > uio->uio_resid) { 814 error = -1; 815 break; 816 } 817 818 /* Copy the new dirent structure into the output buffer and 819 * advance pointers. */ 820 error = uiomove(&d, d.d_reclen, uio); 821 822 (*cntp)++; 823 de = TAILQ_NEXT(de, td_entries); 824 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 825 826 /* Update the offset and cache. */ 827 if (de == NULL) { 828 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 829 node->tn_dir.tn_readdir_lastn = 0; 830 node->tn_dir.tn_readdir_lastp = NULL; 831 } else { 832 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de); 833 node->tn_dir.tn_readdir_lastp = de; 834 } 835 836 node->tn_status |= TMPFS_NODE_ACCESSED; 837 return error; 838 } 839 840 /* --------------------------------------------------------------------- */ 841 842 /* 843 * Resizes the aobj associated to the regular file pointed to by vp to 844 * the size newsize. 'vp' must point to a vnode that represents a regular 845 * file. 'newsize' must be positive. 846 * 847 * Returns zero on success or an appropriate error code on failure. 848 */ 849 int 850 tmpfs_reg_resize(struct vnode *vp, off_t newsize) 851 { 852 int error; 853 size_t newpages, oldpages; 854 struct tmpfs_mount *tmp; 855 struct tmpfs_node *node; 856 off_t oldsize; 857 858 MPASS(vp->v_type == VREG); 859 MPASS(newsize >= 0); 860 861 node = VP_TO_TMPFS_NODE(vp); 862 tmp = VFS_TO_TMPFS(vp->v_mount); 863 864 /* Convert the old and new sizes to the number of pages needed to 865 * store them. It may happen that we do not need to do anything 866 * because the last allocated page can accommodate the change on 867 * its own. */ 868 oldsize = node->tn_size; 869 oldpages = round_page(oldsize) / PAGE_SIZE; 870 MPASS(oldpages == node->tn_reg.tn_aobj_pages); 871 newpages = round_page(newsize) / PAGE_SIZE; 872 873 if (newpages > oldpages && 874 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) { 875 error = ENOSPC; 876 goto out; 877 } 878 879 node->tn_reg.tn_aobj_pages = newpages; 880 881 TMPFS_LOCK(tmp); 882 tmp->tm_pages_used += (newpages - oldpages); 883 TMPFS_UNLOCK(tmp); 884 885 node->tn_size = newsize; 886 vnode_pager_setsize(vp, newsize); 887 if (newsize < oldsize) { 888 size_t zerolen = round_page(newsize) - newsize; 889 vm_object_t uobj = node->tn_reg.tn_aobj; 890 vm_page_t m; 891 892 /* 893 * free "backing store" 894 */ 895 VM_OBJECT_LOCK(uobj); 896 if (newpages < oldpages) { 897 swap_pager_freespace(uobj, 898 newpages, oldpages - newpages); 899 vm_object_page_remove(uobj, 900 OFF_TO_IDX(newsize + PAGE_MASK), 0, FALSE); 901 } 902 903 /* 904 * zero out the truncated part of the last page. 905 */ 906 907 if (zerolen > 0) { 908 m = vm_page_grab(uobj, OFF_TO_IDX(newsize), 909 VM_ALLOC_NOBUSY | VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 910 pmap_zero_page_area(m, PAGE_SIZE - zerolen, 911 zerolen); 912 } 913 VM_OBJECT_UNLOCK(uobj); 914 915 } 916 917 error = 0; 918 919 out: 920 return error; 921 } 922 923 /* --------------------------------------------------------------------- */ 924 925 /* 926 * Change flags of the given vnode. 927 * Caller should execute tmpfs_update on vp after a successful execution. 928 * The vnode must be locked on entry and remain locked on exit. 929 */ 930 int 931 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p) 932 { 933 int error; 934 struct tmpfs_node *node; 935 936 MPASS(VOP_ISLOCKED(vp)); 937 938 node = VP_TO_TMPFS_NODE(vp); 939 940 /* Disallow this operation if the file system is mounted read-only. */ 941 if (vp->v_mount->mnt_flag & MNT_RDONLY) 942 return EROFS; 943 944 /* 945 * Callers may only modify the file flags on objects they 946 * have VADMIN rights for. 947 */ 948 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 949 return (error); 950 /* 951 * Unprivileged processes are not permitted to unset system 952 * flags, or modify flags if any system flags are set. 953 */ 954 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 955 if (node->tn_flags 956 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 957 error = securelevel_gt(cred, 0); 958 if (error) 959 return (error); 960 } 961 /* Snapshot flag cannot be set or cleared */ 962 if (((flags & SF_SNAPSHOT) != 0 && 963 (node->tn_flags & SF_SNAPSHOT) == 0) || 964 ((flags & SF_SNAPSHOT) == 0 && 965 (node->tn_flags & SF_SNAPSHOT) != 0)) 966 return (EPERM); 967 node->tn_flags = flags; 968 } else { 969 if (node->tn_flags 970 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 971 (flags & UF_SETTABLE) != flags) 972 return (EPERM); 973 node->tn_flags &= SF_SETTABLE; 974 node->tn_flags |= (flags & UF_SETTABLE); 975 } 976 node->tn_status |= TMPFS_NODE_CHANGED; 977 978 MPASS(VOP_ISLOCKED(vp)); 979 980 return 0; 981 } 982 983 /* --------------------------------------------------------------------- */ 984 985 /* 986 * Change access mode on the given vnode. 987 * Caller should execute tmpfs_update on vp after a successful execution. 988 * The vnode must be locked on entry and remain locked on exit. 989 */ 990 int 991 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 992 { 993 int error; 994 struct tmpfs_node *node; 995 996 MPASS(VOP_ISLOCKED(vp)); 997 998 node = VP_TO_TMPFS_NODE(vp); 999 1000 /* Disallow this operation if the file system is mounted read-only. */ 1001 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1002 return EROFS; 1003 1004 /* Immutable or append-only files cannot be modified, either. */ 1005 if (node->tn_flags & (IMMUTABLE | APPEND)) 1006 return EPERM; 1007 1008 /* 1009 * To modify the permissions on a file, must possess VADMIN 1010 * for that file. 1011 */ 1012 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1013 return (error); 1014 1015 /* 1016 * Privileged processes may set the sticky bit on non-directories, 1017 * as well as set the setgid bit on a file with a group that the 1018 * process is not a member of. 1019 */ 1020 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1021 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1022 return (EFTYPE); 1023 } 1024 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1025 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1026 if (error) 1027 return (error); 1028 } 1029 1030 1031 node->tn_mode &= ~ALLPERMS; 1032 node->tn_mode |= mode & ALLPERMS; 1033 1034 node->tn_status |= TMPFS_NODE_CHANGED; 1035 1036 MPASS(VOP_ISLOCKED(vp)); 1037 1038 return 0; 1039 } 1040 1041 /* --------------------------------------------------------------------- */ 1042 1043 /* 1044 * Change ownership of the given vnode. At least one of uid or gid must 1045 * be different than VNOVAL. If one is set to that value, the attribute 1046 * is unchanged. 1047 * Caller should execute tmpfs_update on vp after a successful execution. 1048 * The vnode must be locked on entry and remain locked on exit. 1049 */ 1050 int 1051 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1052 struct thread *p) 1053 { 1054 int error; 1055 struct tmpfs_node *node; 1056 uid_t ouid; 1057 gid_t ogid; 1058 1059 MPASS(VOP_ISLOCKED(vp)); 1060 1061 node = VP_TO_TMPFS_NODE(vp); 1062 1063 /* Assign default values if they are unknown. */ 1064 MPASS(uid != VNOVAL || gid != VNOVAL); 1065 if (uid == VNOVAL) 1066 uid = node->tn_uid; 1067 if (gid == VNOVAL) 1068 gid = node->tn_gid; 1069 MPASS(uid != VNOVAL && gid != VNOVAL); 1070 1071 /* Disallow this operation if the file system is mounted read-only. */ 1072 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1073 return EROFS; 1074 1075 /* Immutable or append-only files cannot be modified, either. */ 1076 if (node->tn_flags & (IMMUTABLE | APPEND)) 1077 return EPERM; 1078 1079 /* 1080 * To modify the ownership of a file, must possess VADMIN for that 1081 * file. 1082 */ 1083 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1084 return (error); 1085 1086 /* 1087 * To change the owner of a file, or change the group of a file to a 1088 * group of which we are not a member, the caller must have 1089 * privilege. 1090 */ 1091 if ((uid != node->tn_uid || 1092 (gid != node->tn_gid && !groupmember(gid, cred))) && 1093 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1094 return (error); 1095 1096 ogid = node->tn_gid; 1097 ouid = node->tn_uid; 1098 1099 node->tn_uid = uid; 1100 node->tn_gid = gid; 1101 1102 node->tn_status |= TMPFS_NODE_CHANGED; 1103 1104 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1105 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1106 node->tn_mode &= ~(S_ISUID | S_ISGID); 1107 } 1108 1109 MPASS(VOP_ISLOCKED(vp)); 1110 1111 return 0; 1112 } 1113 1114 /* --------------------------------------------------------------------- */ 1115 1116 /* 1117 * Change size of the given vnode. 1118 * Caller should execute tmpfs_update on vp after a successful execution. 1119 * The vnode must be locked on entry and remain locked on exit. 1120 */ 1121 int 1122 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1123 struct thread *p) 1124 { 1125 int error; 1126 struct tmpfs_node *node; 1127 1128 MPASS(VOP_ISLOCKED(vp)); 1129 1130 node = VP_TO_TMPFS_NODE(vp); 1131 1132 /* Decide whether this is a valid operation based on the file type. */ 1133 error = 0; 1134 switch (vp->v_type) { 1135 case VDIR: 1136 return EISDIR; 1137 1138 case VREG: 1139 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1140 return EROFS; 1141 break; 1142 1143 case VBLK: 1144 /* FALLTHROUGH */ 1145 case VCHR: 1146 /* FALLTHROUGH */ 1147 case VFIFO: 1148 /* Allow modifications of special files even if in the file 1149 * system is mounted read-only (we are not modifying the 1150 * files themselves, but the objects they represent). */ 1151 return 0; 1152 1153 default: 1154 /* Anything else is unsupported. */ 1155 return EOPNOTSUPP; 1156 } 1157 1158 /* Immutable or append-only files cannot be modified, either. */ 1159 if (node->tn_flags & (IMMUTABLE | APPEND)) 1160 return EPERM; 1161 1162 error = tmpfs_truncate(vp, size); 1163 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1164 * for us, as will update tn_status; no need to do that here. */ 1165 1166 MPASS(VOP_ISLOCKED(vp)); 1167 1168 return error; 1169 } 1170 1171 /* --------------------------------------------------------------------- */ 1172 1173 /* 1174 * Change access and modification times of the given vnode. 1175 * Caller should execute tmpfs_update on vp after a successful execution. 1176 * The vnode must be locked on entry and remain locked on exit. 1177 */ 1178 int 1179 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1180 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l) 1181 { 1182 int error; 1183 struct tmpfs_node *node; 1184 1185 MPASS(VOP_ISLOCKED(vp)); 1186 1187 node = VP_TO_TMPFS_NODE(vp); 1188 1189 /* Disallow this operation if the file system is mounted read-only. */ 1190 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1191 return EROFS; 1192 1193 /* Immutable or append-only files cannot be modified, either. */ 1194 if (node->tn_flags & (IMMUTABLE | APPEND)) 1195 return EPERM; 1196 1197 /* Determine if the user have proper privilege to update time. */ 1198 if (vaflags & VA_UTIMES_NULL) { 1199 error = VOP_ACCESS(vp, VADMIN, cred, l); 1200 if (error) 1201 error = VOP_ACCESS(vp, VWRITE, cred, l); 1202 } else 1203 error = VOP_ACCESS(vp, VADMIN, cred, l); 1204 if (error) 1205 return (error); 1206 1207 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1208 node->tn_status |= TMPFS_NODE_ACCESSED; 1209 1210 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) 1211 node->tn_status |= TMPFS_NODE_MODIFIED; 1212 1213 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL) 1214 node->tn_status |= TMPFS_NODE_MODIFIED; 1215 1216 tmpfs_itimes(vp, atime, mtime); 1217 1218 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL) 1219 node->tn_birthtime = *birthtime; 1220 MPASS(VOP_ISLOCKED(vp)); 1221 1222 return 0; 1223 } 1224 1225 /* --------------------------------------------------------------------- */ 1226 /* Sync timestamps */ 1227 void 1228 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1229 const struct timespec *mod) 1230 { 1231 struct tmpfs_node *node; 1232 struct timespec now; 1233 1234 node = VP_TO_TMPFS_NODE(vp); 1235 1236 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1237 TMPFS_NODE_CHANGED)) == 0) 1238 return; 1239 1240 vfs_timestamp(&now); 1241 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1242 if (acc == NULL) 1243 acc = &now; 1244 node->tn_atime = *acc; 1245 } 1246 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1247 if (mod == NULL) 1248 mod = &now; 1249 node->tn_mtime = *mod; 1250 } 1251 if (node->tn_status & TMPFS_NODE_CHANGED) { 1252 node->tn_ctime = now; 1253 } 1254 node->tn_status &= 1255 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1256 } 1257 1258 /* --------------------------------------------------------------------- */ 1259 1260 void 1261 tmpfs_update(struct vnode *vp) 1262 { 1263 1264 tmpfs_itimes(vp, NULL, NULL); 1265 } 1266 1267 /* --------------------------------------------------------------------- */ 1268 1269 int 1270 tmpfs_truncate(struct vnode *vp, off_t length) 1271 { 1272 int error; 1273 struct tmpfs_node *node; 1274 1275 node = VP_TO_TMPFS_NODE(vp); 1276 1277 if (length < 0) { 1278 error = EINVAL; 1279 goto out; 1280 } 1281 1282 if (node->tn_size == length) { 1283 error = 0; 1284 goto out; 1285 } 1286 1287 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1288 return (EFBIG); 1289 1290 error = tmpfs_reg_resize(vp, length); 1291 if (error == 0) { 1292 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1293 } 1294 1295 out: 1296 tmpfs_update(vp); 1297 1298 return error; 1299 } 1300