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 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Efficient memory file system supporting functions. 42 */ 43 #include <sys/cdefs.h> 44 __FBSDID("$FreeBSD$"); 45 46 #include <sys/param.h> 47 #include <sys/namei.h> 48 #include <sys/priv.h> 49 #include <sys/proc.h> 50 #include <sys/stat.h> 51 #include <sys/systm.h> 52 #include <sys/vnode.h> 53 #include <sys/vmmeter.h> 54 55 #include <vm/vm.h> 56 #include <vm/vm_object.h> 57 #include <vm/vm_page.h> 58 #include <vm/vm_pager.h> 59 #include <vm/vm_extern.h> 60 61 #include <fs/tmpfs/tmpfs.h> 62 #include <fs/tmpfs/tmpfs_fifoops.h> 63 #include <fs/tmpfs/tmpfs_vnops.h> 64 65 /* --------------------------------------------------------------------- */ 66 67 /* 68 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 69 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 70 * using the credentials of the process 'p'. 71 * 72 * If the node type is set to 'VDIR', then the parent parameter must point 73 * to the parent directory of the node being created. It may only be NULL 74 * while allocating the root node. 75 * 76 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 77 * specifies the device the node represents. 78 * 79 * If the node type is set to 'VLNK', then the parameter target specifies 80 * the file name of the target file for the symbolic link that is being 81 * created. 82 * 83 * Note that new nodes are retrieved from the available list if it has 84 * items or, if it is empty, from the node pool as long as there is enough 85 * space to create them. 86 * 87 * Returns zero on success or an appropriate error code on failure. 88 */ 89 int 90 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 91 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 92 char *target, dev_t rdev, struct thread *p, struct tmpfs_node **node) 93 { 94 struct tmpfs_node *nnode; 95 96 /* If the root directory of the 'tmp' file system is not yet 97 * allocated, this must be the request to do it. */ 98 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 99 100 MPASS(IFF(type == VLNK, target != NULL)); 101 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL)); 102 103 if (tmp->tm_nodes_inuse > tmp->tm_nodes_max) 104 return (ENOSPC); 105 106 nnode = (struct tmpfs_node *)uma_zalloc_arg( 107 tmp->tm_node_pool, tmp, M_WAITOK); 108 109 /* Generic initialization. */ 110 nnode->tn_type = type; 111 vfs_timestamp(&nnode->tn_atime); 112 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 113 nnode->tn_atime; 114 nnode->tn_uid = uid; 115 nnode->tn_gid = gid; 116 nnode->tn_mode = mode; 117 nnode->tn_id = alloc_unr(tmp->tm_ino_unr); 118 119 /* Type-specific initialization. */ 120 switch (nnode->tn_type) { 121 case VBLK: 122 case VCHR: 123 nnode->tn_rdev = rdev; 124 break; 125 126 case VDIR: 127 TAILQ_INIT(&nnode->tn_dir.tn_dirhead); 128 MPASS(parent != nnode); 129 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL)); 130 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent; 131 nnode->tn_dir.tn_readdir_lastn = 0; 132 nnode->tn_dir.tn_readdir_lastp = NULL; 133 nnode->tn_links++; 134 nnode->tn_dir.tn_parent->tn_links++; 135 break; 136 137 case VFIFO: 138 /* FALLTHROUGH */ 139 case VSOCK: 140 break; 141 142 case VLNK: 143 MPASS(strlen(target) < MAXPATHLEN); 144 nnode->tn_size = strlen(target); 145 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME, 146 M_WAITOK); 147 memcpy(nnode->tn_link, target, nnode->tn_size); 148 break; 149 150 case VREG: 151 nnode->tn_reg.tn_aobj = 152 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0); 153 nnode->tn_reg.tn_aobj_pages = 0; 154 break; 155 156 default: 157 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type); 158 } 159 160 TMPFS_LOCK(tmp); 161 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 162 tmp->tm_nodes_inuse++; 163 TMPFS_UNLOCK(tmp); 164 165 *node = nnode; 166 return 0; 167 } 168 169 /* --------------------------------------------------------------------- */ 170 171 /* 172 * Destroys the node pointed to by node from the file system 'tmp'. 173 * If the node does not belong to the given mount point, the results are 174 * unpredicted. 175 * 176 * If the node references a directory; no entries are allowed because 177 * their removal could need a recursive algorithm, something forbidden in 178 * kernel space. Furthermore, there is not need to provide such 179 * functionality (recursive removal) because the only primitives offered 180 * to the user are the removal of empty directories and the deletion of 181 * individual files. 182 * 183 * Note that nodes are not really deleted; in fact, when a node has been 184 * allocated, it cannot be deleted during the whole life of the file 185 * system. Instead, they are moved to the available list and remain there 186 * until reused. 187 */ 188 void 189 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 190 { 191 size_t pages = 0; 192 193 #ifdef INVARIANTS 194 TMPFS_NODE_LOCK(node); 195 MPASS(node->tn_vnode == NULL); 196 TMPFS_NODE_UNLOCK(node); 197 #endif 198 199 TMPFS_LOCK(tmp); 200 LIST_REMOVE(node, tn_entries); 201 tmp->tm_nodes_inuse--; 202 TMPFS_UNLOCK(tmp); 203 204 switch (node->tn_type) { 205 case VNON: 206 /* Do not do anything. VNON is provided to let the 207 * allocation routine clean itself easily by avoiding 208 * duplicating code in it. */ 209 /* FALLTHROUGH */ 210 case VBLK: 211 /* FALLTHROUGH */ 212 case VCHR: 213 /* FALLTHROUGH */ 214 case VDIR: 215 /* FALLTHROUGH */ 216 case VFIFO: 217 /* FALLTHROUGH */ 218 case VSOCK: 219 break; 220 221 case VLNK: 222 free(node->tn_link, M_TMPFSNAME); 223 break; 224 225 case VREG: 226 if (node->tn_reg.tn_aobj != NULL) 227 vm_object_deallocate(node->tn_reg.tn_aobj); 228 pages = node->tn_reg.tn_aobj_pages; 229 break; 230 231 default: 232 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 233 } 234 235 free_unr(tmp->tm_ino_unr, node->tn_id); 236 uma_zfree(tmp->tm_node_pool, node); 237 238 TMPFS_LOCK(tmp); 239 tmp->tm_pages_used -= pages; 240 TMPFS_UNLOCK(tmp); 241 } 242 243 /* --------------------------------------------------------------------- */ 244 245 /* 246 * Allocates a new directory entry for the node node with a name of name. 247 * The new directory entry is returned in *de. 248 * 249 * The link count of node is increased by one to reflect the new object 250 * referencing it. 251 * 252 * Returns zero on success or an appropriate error code on failure. 253 */ 254 int 255 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 256 const char *name, uint16_t len, struct tmpfs_dirent **de) 257 { 258 struct tmpfs_dirent *nde; 259 260 nde = (struct tmpfs_dirent *)uma_zalloc( 261 tmp->tm_dirent_pool, M_WAITOK); 262 nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 263 nde->td_namelen = len; 264 memcpy(nde->td_name, name, len); 265 266 nde->td_node = node; 267 node->tn_links++; 268 269 *de = nde; 270 271 return 0; 272 } 273 274 /* --------------------------------------------------------------------- */ 275 276 /* 277 * Frees a directory entry. It is the caller's responsibility to destroy 278 * the node referenced by it if needed. 279 * 280 * The link count of node is decreased by one to reflect the removal of an 281 * object that referenced it. This only happens if 'node_exists' is true; 282 * otherwise the function will not access the node referred to by the 283 * directory entry, as it may already have been released from the outside. 284 */ 285 void 286 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de, 287 boolean_t node_exists) 288 { 289 if (node_exists) { 290 struct tmpfs_node *node; 291 292 node = de->td_node; 293 294 MPASS(node->tn_links > 0); 295 node->tn_links--; 296 } 297 298 free(de->td_name, M_TMPFSNAME); 299 uma_zfree(tmp->tm_dirent_pool, de); 300 } 301 302 /* --------------------------------------------------------------------- */ 303 304 /* 305 * Allocates a new vnode for the node node or returns a new reference to 306 * an existing one if the node had already a vnode referencing it. The 307 * resulting locked vnode is returned in *vpp. 308 * 309 * Returns zero on success or an appropriate error code on failure. 310 */ 311 int 312 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 313 struct vnode **vpp, struct thread *td) 314 { 315 int error = 0; 316 struct vnode *vp; 317 318 loop: 319 TMPFS_NODE_LOCK(node); 320 if ((vp = node->tn_vnode) != NULL) { 321 VI_LOCK(vp); 322 TMPFS_NODE_UNLOCK(node); 323 vholdl(vp); 324 (void) vget(vp, lkflag | LK_INTERLOCK | LK_RETRY, td); 325 vdrop(vp); 326 327 /* 328 * Make sure the vnode is still there after 329 * getting the interlock to avoid racing a free. 330 */ 331 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 332 vput(vp); 333 goto loop; 334 } 335 336 goto out; 337 } 338 339 /* 340 * otherwise lock the vp list while we call getnewvnode 341 * since that can block. 342 */ 343 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 344 node->tn_vpstate |= TMPFS_VNODE_WANT; 345 error = msleep((caddr_t) &node->tn_vpstate, 346 TMPFS_NODE_MTX(node), PDROP | PCATCH, 347 "tmpfs_alloc_vp", 0); 348 if (error) 349 return error; 350 351 goto loop; 352 } else 353 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 354 355 TMPFS_NODE_UNLOCK(node); 356 357 /* Get a new vnode and associate it with our node. */ 358 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); 359 if (error != 0) 360 goto unlock; 361 MPASS(vp != NULL); 362 363 (void) vn_lock(vp, lkflag | LK_RETRY); 364 365 vp->v_data = node; 366 vp->v_type = node->tn_type; 367 368 /* Type-specific initialization. */ 369 switch (node->tn_type) { 370 case VBLK: 371 /* FALLTHROUGH */ 372 case VCHR: 373 /* FALLTHROUGH */ 374 case VLNK: 375 /* FALLTHROUGH */ 376 case VREG: 377 /* FALLTHROUGH */ 378 case VSOCK: 379 break; 380 case VFIFO: 381 vp->v_op = &tmpfs_fifoop_entries; 382 break; 383 case VDIR: 384 if (node->tn_dir.tn_parent == node) 385 vp->v_vflag |= VV_ROOT; 386 break; 387 388 default: 389 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 390 } 391 392 vnode_pager_setsize(vp, node->tn_size); 393 error = insmntque(vp, mp); 394 if (error) { 395 vgone(vp); 396 vput(vp); 397 vp = NULL; 398 } 399 400 unlock: 401 TMPFS_NODE_LOCK(node); 402 403 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 404 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 405 node->tn_vnode = vp; 406 407 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 408 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 409 TMPFS_NODE_UNLOCK(node); 410 wakeup((caddr_t) &node->tn_vpstate); 411 } else 412 TMPFS_NODE_UNLOCK(node); 413 414 out: 415 *vpp = vp; 416 417 MPASS(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp, td))); 418 #ifdef INVARIANTS 419 TMPFS_NODE_LOCK(node); 420 MPASS(*vpp == node->tn_vnode); 421 TMPFS_NODE_UNLOCK(node); 422 #endif 423 424 return error; 425 } 426 427 /* --------------------------------------------------------------------- */ 428 429 /* 430 * Destroys the association between the vnode vp and the node it 431 * references. 432 */ 433 void 434 tmpfs_free_vp(struct vnode *vp) 435 { 436 struct tmpfs_node *node; 437 438 node = VP_TO_TMPFS_NODE(vp); 439 440 TMPFS_NODE_LOCK(node); 441 node->tn_vnode = NULL; 442 vp->v_data = NULL; 443 TMPFS_NODE_UNLOCK(node); 444 } 445 446 /* --------------------------------------------------------------------- */ 447 448 /* 449 * Allocates a new file of type 'type' and adds it to the parent directory 450 * 'dvp'; this addition is done using the component name given in 'cnp'. 451 * The ownership of the new file is automatically assigned based on the 452 * credentials of the caller (through 'cnp'), the group is set based on 453 * the parent directory and the mode is determined from the 'vap' argument. 454 * If successful, *vpp holds a vnode to the newly created file and zero 455 * is returned. Otherwise *vpp is NULL and the function returns an 456 * appropriate error code. 457 */ 458 int 459 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 460 struct componentname *cnp, char *target) 461 { 462 int error; 463 struct tmpfs_dirent *de; 464 struct tmpfs_mount *tmp; 465 struct tmpfs_node *dnode; 466 struct tmpfs_node *node; 467 struct tmpfs_node *parent; 468 469 MPASS(VOP_ISLOCKED(dvp, cnp->cn_thread)); 470 MPASS(cnp->cn_flags & HASBUF); 471 472 tmp = VFS_TO_TMPFS(dvp->v_mount); 473 dnode = VP_TO_TMPFS_DIR(dvp); 474 *vpp = NULL; 475 476 /* If the entry we are creating is a directory, we cannot overflow 477 * the number of links of its parent, because it will get a new 478 * link. */ 479 if (vap->va_type == VDIR) { 480 /* Ensure that we do not overflow the maximum number of links 481 * imposed by the system. */ 482 MPASS(dnode->tn_links <= LINK_MAX); 483 if (dnode->tn_links == LINK_MAX) { 484 error = EMLINK; 485 goto out; 486 } 487 488 parent = dnode; 489 MPASS(parent != NULL); 490 } else 491 parent = NULL; 492 493 /* Allocate a node that represents the new file. */ 494 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid, 495 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, 496 cnp->cn_thread, &node); 497 if (error != 0) 498 goto out; 499 500 /* Allocate a directory entry that points to the new file. */ 501 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 502 &de); 503 if (error != 0) { 504 tmpfs_free_node(tmp, node); 505 goto out; 506 } 507 508 /* Allocate a vnode for the new file. */ 509 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp, 510 cnp->cn_thread); 511 if (error != 0) { 512 tmpfs_free_dirent(tmp, de, TRUE); 513 tmpfs_free_node(tmp, node); 514 goto out; 515 } 516 517 /* Now that all required items are allocated, we can proceed to 518 * insert the new node into the directory, an operation that 519 * cannot fail. */ 520 tmpfs_dir_attach(dvp, de); 521 522 out: 523 524 return error; 525 } 526 527 /* --------------------------------------------------------------------- */ 528 529 /* 530 * Attaches the directory entry de to the directory represented by vp. 531 * Note that this does not change the link count of the node pointed by 532 * the directory entry, as this is done by tmpfs_alloc_dirent. 533 */ 534 void 535 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 536 { 537 struct tmpfs_node *dnode; 538 539 ASSERT_VOP_ELOCKED(vp, __func__); 540 dnode = VP_TO_TMPFS_DIR(vp); 541 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries); 542 dnode->tn_size += sizeof(struct tmpfs_dirent); 543 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 544 TMPFS_NODE_MODIFIED; 545 } 546 547 /* --------------------------------------------------------------------- */ 548 549 /* 550 * Detaches the directory entry de from the directory represented by vp. 551 * Note that this does not change the link count of the node pointed by 552 * the directory entry, as this is done by tmpfs_free_dirent. 553 */ 554 void 555 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 556 { 557 struct tmpfs_node *dnode; 558 559 ASSERT_VOP_ELOCKED(vp, __func__); 560 dnode = VP_TO_TMPFS_DIR(vp); 561 562 if (dnode->tn_dir.tn_readdir_lastp == de) { 563 dnode->tn_dir.tn_readdir_lastn = 0; 564 dnode->tn_dir.tn_readdir_lastp = NULL; 565 } 566 567 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries); 568 dnode->tn_size -= sizeof(struct tmpfs_dirent); 569 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 570 TMPFS_NODE_MODIFIED; 571 } 572 573 /* --------------------------------------------------------------------- */ 574 575 /* 576 * Looks for a directory entry in the directory represented by node. 577 * 'cnp' describes the name of the entry to look for. Note that the . 578 * and .. components are not allowed as they do not physically exist 579 * within directories. 580 * 581 * Returns a pointer to the entry when found, otherwise NULL. 582 */ 583 struct tmpfs_dirent * 584 tmpfs_dir_lookup(struct tmpfs_node *node, struct componentname *cnp) 585 { 586 boolean_t found; 587 struct tmpfs_dirent *de; 588 589 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 590 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 591 cnp->cn_nameptr[1] == '.'))); 592 TMPFS_VALIDATE_DIR(node); 593 594 found = 0; 595 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) { 596 MPASS(cnp->cn_namelen < 0xffff); 597 if (de->td_namelen == (uint16_t)cnp->cn_namelen && 598 memcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) { 599 found = 1; 600 break; 601 } 602 } 603 node->tn_status |= TMPFS_NODE_ACCESSED; 604 605 return found ? de : NULL; 606 } 607 608 struct tmpfs_dirent * 609 tmpfs_dir_search(struct tmpfs_node *node, struct tmpfs_node *f) 610 { 611 struct tmpfs_dirent *de; 612 613 TMPFS_VALIDATE_DIR(node); 614 node->tn_status |= TMPFS_NODE_ACCESSED; 615 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) { 616 if (de->td_node == f) 617 return (de); 618 } 619 return (NULL); 620 } 621 622 /* --------------------------------------------------------------------- */ 623 624 /* 625 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 626 * directory and returns it in the uio space. The function returns 0 627 * on success, -1 if there was not enough space in the uio structure to 628 * hold the directory entry or an appropriate error code if another 629 * error happens. 630 */ 631 int 632 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 633 { 634 int error; 635 struct dirent dent; 636 637 TMPFS_VALIDATE_DIR(node); 638 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 639 640 dent.d_fileno = node->tn_id; 641 dent.d_type = DT_DIR; 642 dent.d_namlen = 1; 643 dent.d_name[0] = '.'; 644 dent.d_name[1] = '\0'; 645 dent.d_reclen = GENERIC_DIRSIZ(&dent); 646 647 if (dent.d_reclen > uio->uio_resid) 648 error = -1; 649 else { 650 error = uiomove(&dent, dent.d_reclen, uio); 651 if (error == 0) 652 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 653 } 654 655 node->tn_status |= TMPFS_NODE_ACCESSED; 656 657 return error; 658 } 659 660 /* --------------------------------------------------------------------- */ 661 662 /* 663 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 664 * directory and returns it in the uio space. The function returns 0 665 * on success, -1 if there was not enough space in the uio structure to 666 * hold the directory entry or an appropriate error code if another 667 * error happens. 668 */ 669 int 670 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 671 { 672 int error; 673 struct dirent dent; 674 675 TMPFS_VALIDATE_DIR(node); 676 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 677 678 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 679 dent.d_type = DT_DIR; 680 dent.d_namlen = 2; 681 dent.d_name[0] = '.'; 682 dent.d_name[1] = '.'; 683 dent.d_name[2] = '\0'; 684 dent.d_reclen = GENERIC_DIRSIZ(&dent); 685 686 if (dent.d_reclen > uio->uio_resid) 687 error = -1; 688 else { 689 error = uiomove(&dent, dent.d_reclen, uio); 690 if (error == 0) { 691 struct tmpfs_dirent *de; 692 693 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead); 694 if (de == NULL) 695 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 696 else 697 uio->uio_offset = tmpfs_dircookie(de); 698 } 699 } 700 701 node->tn_status |= TMPFS_NODE_ACCESSED; 702 703 return error; 704 } 705 706 /* --------------------------------------------------------------------- */ 707 708 /* 709 * Lookup a directory entry by its associated cookie. 710 */ 711 struct tmpfs_dirent * 712 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie) 713 { 714 struct tmpfs_dirent *de; 715 716 if (cookie == node->tn_dir.tn_readdir_lastn && 717 node->tn_dir.tn_readdir_lastp != NULL) { 718 return node->tn_dir.tn_readdir_lastp; 719 } 720 721 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) { 722 if (tmpfs_dircookie(de) == cookie) { 723 break; 724 } 725 } 726 727 return de; 728 } 729 730 /* --------------------------------------------------------------------- */ 731 732 /* 733 * Helper function for tmpfs_readdir. Returns as much directory entries 734 * as can fit in the uio space. The read starts at uio->uio_offset. 735 * The function returns 0 on success, -1 if there was not enough space 736 * in the uio structure to hold the directory entry or an appropriate 737 * error code if another error happens. 738 */ 739 int 740 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp) 741 { 742 int error; 743 off_t startcookie; 744 struct tmpfs_dirent *de; 745 746 TMPFS_VALIDATE_DIR(node); 747 748 /* Locate the first directory entry we have to return. We have cached 749 * the last readdir in the node, so use those values if appropriate. 750 * Otherwise do a linear scan to find the requested entry. */ 751 startcookie = uio->uio_offset; 752 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT); 753 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT); 754 if (startcookie == TMPFS_DIRCOOKIE_EOF) { 755 return 0; 756 } else { 757 de = tmpfs_dir_lookupbycookie(node, startcookie); 758 } 759 if (de == NULL) { 760 return EINVAL; 761 } 762 763 /* Read as much entries as possible; i.e., until we reach the end of 764 * the directory or we exhaust uio space. */ 765 do { 766 struct dirent d; 767 768 /* Create a dirent structure representing the current 769 * tmpfs_node and fill it. */ 770 d.d_fileno = de->td_node->tn_id; 771 switch (de->td_node->tn_type) { 772 case VBLK: 773 d.d_type = DT_BLK; 774 break; 775 776 case VCHR: 777 d.d_type = DT_CHR; 778 break; 779 780 case VDIR: 781 d.d_type = DT_DIR; 782 break; 783 784 case VFIFO: 785 d.d_type = DT_FIFO; 786 break; 787 788 case VLNK: 789 d.d_type = DT_LNK; 790 break; 791 792 case VREG: 793 d.d_type = DT_REG; 794 break; 795 796 case VSOCK: 797 d.d_type = DT_SOCK; 798 break; 799 800 default: 801 panic("tmpfs_dir_getdents: type %p %d", 802 de->td_node, (int)de->td_node->tn_type); 803 } 804 d.d_namlen = de->td_namelen; 805 MPASS(de->td_namelen < sizeof(d.d_name)); 806 (void)memcpy(d.d_name, de->td_name, de->td_namelen); 807 d.d_name[de->td_namelen] = '\0'; 808 d.d_reclen = GENERIC_DIRSIZ(&d); 809 810 /* Stop reading if the directory entry we are treating is 811 * bigger than the amount of data that can be returned. */ 812 if (d.d_reclen > uio->uio_resid) { 813 error = -1; 814 break; 815 } 816 817 /* Copy the new dirent structure into the output buffer and 818 * advance pointers. */ 819 error = uiomove(&d, d.d_reclen, uio); 820 821 (*cntp)++; 822 de = TAILQ_NEXT(de, td_entries); 823 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 824 825 /* Update the offset and cache. */ 826 if (de == NULL) { 827 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 828 node->tn_dir.tn_readdir_lastn = 0; 829 node->tn_dir.tn_readdir_lastp = NULL; 830 } else { 831 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de); 832 node->tn_dir.tn_readdir_lastp = de; 833 } 834 835 node->tn_status |= TMPFS_NODE_ACCESSED; 836 return error; 837 } 838 839 /* --------------------------------------------------------------------- */ 840 841 /* 842 * Resizes the aobj associated to the regular file pointed to by vp to 843 * the size newsize. 'vp' must point to a vnode that represents a regular 844 * file. 'newsize' must be positive. 845 * 846 * Returns zero on success or an appropriate error code on failure. 847 */ 848 int 849 tmpfs_reg_resize(struct vnode *vp, off_t newsize) 850 { 851 int error; 852 size_t newpages, oldpages; 853 struct tmpfs_mount *tmp; 854 struct tmpfs_node *node; 855 off_t oldsize; 856 857 MPASS(vp->v_type == VREG); 858 MPASS(newsize >= 0); 859 860 node = VP_TO_TMPFS_NODE(vp); 861 tmp = VFS_TO_TMPFS(vp->v_mount); 862 863 /* Convert the old and new sizes to the number of pages needed to 864 * store them. It may happen that we do not need to do anything 865 * because the last allocated page can accommodate the change on 866 * its own. */ 867 oldsize = node->tn_size; 868 oldpages = round_page(oldsize) / PAGE_SIZE; 869 MPASS(oldpages == node->tn_reg.tn_aobj_pages); 870 newpages = round_page(newsize) / PAGE_SIZE; 871 872 if (newpages > oldpages && 873 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) { 874 error = ENOSPC; 875 goto out; 876 } 877 878 node->tn_reg.tn_aobj_pages = newpages; 879 880 TMPFS_LOCK(tmp); 881 tmp->tm_pages_used += (newpages - oldpages); 882 TMPFS_UNLOCK(tmp); 883 884 node->tn_size = newsize; 885 vnode_pager_setsize(vp, newsize); 886 if (newsize < oldsize) { 887 size_t zerolen = round_page(newsize) - newsize; 888 vm_object_t uobj = node->tn_reg.tn_aobj; 889 vm_page_t m; 890 891 /* 892 * free "backing store" 893 */ 894 VM_OBJECT_LOCK(uobj); 895 if (newpages < oldpages) { 896 swap_pager_freespace(uobj, 897 newpages, oldpages - newpages); 898 vm_object_page_remove(uobj, 899 OFF_TO_IDX(newsize + PAGE_MASK), 0, FALSE); 900 } 901 902 /* 903 * zero out the truncated part of the last page. 904 */ 905 906 if (zerolen > 0) { 907 m = vm_page_grab(uobj, OFF_TO_IDX(newsize), 908 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 909 pmap_zero_page_area(m, PAGE_SIZE - zerolen, 910 zerolen); 911 vm_page_wakeup(m); 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, p)); 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, p)); 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, p)); 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, p)); 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, p)); 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, p)); 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, p)); 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, p)); 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, l)); 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, l)); 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 boolean_t extended; 1273 int error; 1274 struct tmpfs_node *node; 1275 1276 node = VP_TO_TMPFS_NODE(vp); 1277 extended = length > node->tn_size; 1278 1279 if (length < 0) { 1280 error = EINVAL; 1281 goto out; 1282 } 1283 1284 if (node->tn_size == length) { 1285 error = 0; 1286 goto out; 1287 } 1288 1289 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1290 return (EFBIG); 1291 1292 error = tmpfs_reg_resize(vp, length); 1293 if (error == 0) { 1294 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1295 } 1296 1297 out: 1298 tmpfs_update(vp); 1299 1300 return error; 1301 } 1302