1 /* $NetBSD: tmpfs_subr.c,v 1.21 2006/06/07 22:33:39 kardel Exp $ */ 2 3 /* 4 * Copyright (c) 2005, 2006 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 44 #include <sys/cdefs.h> 45 __KERNEL_RCSID(0, "$NetBSD: tmpfs_subr.c,v 1.21 2006/06/07 22:33:39 kardel Exp $"); 46 47 #include <sys/param.h> 48 #include <sys/dirent.h> 49 #include <sys/event.h> 50 #include <sys/malloc.h> 51 #include <sys/mount.h> 52 #include <sys/namei.h> 53 #include <sys/time.h> 54 #include <sys/stat.h> 55 #include <sys/systm.h> 56 #include <sys/swap.h> 57 #include <sys/vnode.h> 58 #include <sys/kauth.h> 59 60 #include <uvm/uvm.h> 61 62 #include <miscfs/specfs/specdev.h> 63 #include <fs/tmpfs/tmpfs.h> 64 #include <fs/tmpfs/tmpfs_fifoops.h> 65 #include <fs/tmpfs/tmpfs_specops.h> 66 #include <fs/tmpfs/tmpfs_vnops.h> 67 68 /* --------------------------------------------------------------------- */ 69 70 /* 71 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 72 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 73 * using the credentials of the process 'p'. 74 * 75 * If the node type is set to 'VDIR', then the parent parameter must point 76 * to the parent directory of the node being created. It may only be NULL 77 * while allocating the root node. 78 * 79 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 80 * specifies the device the node represents. 81 * 82 * If the node type is set to 'VLNK', then the parameter target specifies 83 * the file name of the target file for the symbolic link that is being 84 * created. 85 * 86 * Note that new nodes are retrieved from the available list if it has 87 * items or, if it is empty, from the node pool as long as there is enough 88 * space to create them. 89 * 90 * Returns zero on success or an appropriate error code on failure. 91 */ 92 int 93 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 94 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 95 char *target, dev_t rdev, struct proc *p, struct tmpfs_node **node) 96 { 97 struct tmpfs_node *nnode; 98 99 /* If the root directory of the 'tmp' file system is not yet 100 * allocated, this must be the request to do it. */ 101 KASSERT(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 102 103 KASSERT(IFF(type == VLNK, target != NULL)); 104 KASSERT(IFF(type == VBLK || type == VCHR, rdev != VNOVAL)); 105 106 KASSERT(uid != VNOVAL && gid != VNOVAL && mode != VNOVAL); 107 108 nnode = NULL; 109 if (LIST_EMPTY(&tmp->tm_nodes_avail)) { 110 KASSERT(tmp->tm_nodes_last <= tmp->tm_nodes_max); 111 if (tmp->tm_nodes_last == tmp->tm_nodes_max) 112 return ENOSPC; 113 114 nnode = 115 (struct tmpfs_node *)TMPFS_POOL_GET(&tmp->tm_node_pool, 0); 116 if (nnode == NULL) 117 return ENOSPC; 118 nnode->tn_id = tmp->tm_nodes_last++; 119 nnode->tn_gen = 0; 120 } else { 121 nnode = LIST_FIRST(&tmp->tm_nodes_avail); 122 LIST_REMOVE(nnode, tn_entries); 123 nnode->tn_gen++; 124 } 125 KASSERT(nnode != NULL); 126 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 127 128 /* Generic initialization. */ 129 nnode->tn_type = type; 130 nnode->tn_size = 0; 131 nnode->tn_status = 0; 132 nnode->tn_flags = 0; 133 nnode->tn_links = 0; 134 getnanotime(&nnode->tn_atime); 135 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 136 nnode->tn_atime; 137 nnode->tn_uid = uid; 138 nnode->tn_gid = gid; 139 nnode->tn_mode = mode; 140 nnode->tn_lockf = NULL; 141 nnode->tn_vnode = NULL; 142 143 /* Type-specific initialization. */ 144 switch (nnode->tn_type) { 145 case VBLK: 146 case VCHR: 147 nnode->tn_spec.tn_dev.tn_rdev = rdev; 148 break; 149 150 case VDIR: 151 TAILQ_INIT(&nnode->tn_spec.tn_dir.tn_dir); 152 nnode->tn_spec.tn_dir.tn_parent = 153 (parent == NULL) ? nnode : parent; 154 nnode->tn_spec.tn_dir.tn_readdir_lastn = 0; 155 nnode->tn_spec.tn_dir.tn_readdir_lastp = NULL; 156 nnode->tn_links++; 157 nnode->tn_spec.tn_dir.tn_parent->tn_links++; 158 break; 159 160 case VFIFO: 161 /* FALLTHROUGH */ 162 case VSOCK: 163 break; 164 165 case VLNK: 166 KASSERT(strlen(target) < MAXPATHLEN); 167 nnode->tn_size = strlen(target); 168 nnode->tn_spec.tn_lnk.tn_link = 169 tmpfs_str_pool_get(&tmp->tm_str_pool, nnode->tn_size, 0); 170 if (nnode->tn_spec.tn_lnk.tn_link == NULL) { 171 nnode->tn_type = VNON; 172 tmpfs_free_node(tmp, nnode); 173 return ENOSPC; 174 } 175 memcpy(nnode->tn_spec.tn_lnk.tn_link, target, nnode->tn_size); 176 break; 177 178 case VREG: 179 nnode->tn_spec.tn_reg.tn_aobj = 180 uao_create(INT32_MAX - PAGE_SIZE, 0); 181 nnode->tn_spec.tn_reg.tn_aobj_pages = 0; 182 break; 183 184 default: 185 KASSERT(0); 186 } 187 188 *node = nnode; 189 return 0; 190 } 191 192 /* --------------------------------------------------------------------- */ 193 194 /* 195 * Destroys the node pointed to by node from the file system 'tmp'. 196 * If the node does not belong to the given mount point, the results are 197 * unpredicted. 198 * 199 * If the node references a directory; no entries are allowed because 200 * their removal could need a recursive algorithm, something forbidden in 201 * kernel space. Furthermore, there is not need to provide such 202 * functionality (recursive removal) because the only primitives offered 203 * to the user are the removal of empty directories and the deletion of 204 * individual files. 205 * 206 * Note that nodes are not really deleted; in fact, when a node has been 207 * allocated, it cannot be deleted during the whole life of the file 208 * system. Instead, they are moved to the available list and remain there 209 * until reused. 210 */ 211 void 212 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 213 { 214 ino_t id; 215 unsigned long gen; 216 size_t pages; 217 218 switch (node->tn_type) { 219 case VNON: 220 /* Do not do anything. VNON is provided to let the 221 * allocation routine clean itself easily by avoiding 222 * duplicating code in it. */ 223 /* FALLTHROUGH */ 224 case VBLK: 225 /* FALLTHROUGH */ 226 case VCHR: 227 /* FALLTHROUGH */ 228 case VDIR: 229 /* FALLTHROUGH */ 230 case VFIFO: 231 /* FALLTHROUGH */ 232 case VSOCK: 233 pages = 0; 234 break; 235 236 case VLNK: 237 tmpfs_str_pool_put(&tmp->tm_str_pool, 238 node->tn_spec.tn_lnk.tn_link, node->tn_size); 239 pages = 0; 240 break; 241 242 case VREG: 243 if (node->tn_spec.tn_reg.tn_aobj != NULL) 244 uao_detach(node->tn_spec.tn_reg.tn_aobj); 245 pages = node->tn_spec.tn_reg.tn_aobj_pages; 246 break; 247 248 default: 249 KASSERT(0); 250 pages = 0; /* Shut up gcc when !DIAGNOSTIC. */ 251 break; 252 } 253 254 tmp->tm_pages_used -= pages; 255 256 LIST_REMOVE(node, tn_entries); 257 id = node->tn_id; 258 gen = node->tn_gen; 259 memset(node, 0, sizeof(struct tmpfs_node)); 260 node->tn_id = id; 261 node->tn_type = VNON; 262 node->tn_gen = gen; 263 LIST_INSERT_HEAD(&tmp->tm_nodes_avail, node, tn_entries); 264 } 265 266 /* --------------------------------------------------------------------- */ 267 268 /* 269 * Allocates a new directory entry for the node node with a name of name. 270 * The new directory entry is returned in *de. 271 * 272 * The link count of node is increased by one to reflect the new object 273 * referencing it. 274 * 275 * Returns zero on success or an appropriate error code on failure. 276 */ 277 int 278 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 279 const char *name, uint16_t len, struct tmpfs_dirent **de) 280 { 281 struct tmpfs_dirent *nde; 282 283 nde = (struct tmpfs_dirent *)TMPFS_POOL_GET(&tmp->tm_dirent_pool, 0); 284 if (nde == NULL) 285 return ENOSPC; 286 287 nde->td_name = tmpfs_str_pool_get(&tmp->tm_str_pool, len, 0); 288 if (nde->td_name == NULL) { 289 TMPFS_POOL_PUT(&tmp->tm_dirent_pool, nde); 290 return ENOSPC; 291 } 292 nde->td_namelen = len; 293 memcpy(nde->td_name, name, len); 294 nde->td_node = node; 295 296 node->tn_links++; 297 *de = nde; 298 299 return 0; 300 } 301 302 /* --------------------------------------------------------------------- */ 303 304 /* 305 * Frees a directory entry. It is the caller's responsibility to destroy 306 * the node referenced by it if needed. 307 * 308 * The link count of node is decreased by one to reflect the removal of an 309 * object that referenced it. This only happens if 'node_exists' is true; 310 * otherwise the function will not access the node referred to by the 311 * directory entry, as it may already have been released from the outside. 312 */ 313 void 314 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de, 315 boolean_t node_exists) 316 { 317 if (node_exists) { 318 struct tmpfs_node *node; 319 320 node = de->td_node; 321 322 KASSERT(node->tn_links > 0); 323 node->tn_links--; 324 } 325 326 tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name, de->td_namelen); 327 TMPFS_POOL_PUT(&tmp->tm_dirent_pool, de); 328 } 329 330 /* --------------------------------------------------------------------- */ 331 332 /* 333 * Allocates a new vnode for the node node or returns a new reference to 334 * an existing one if the node had already a vnode referencing it. The 335 * resulting locked vnode is returned in *vpp. 336 * 337 * Returns zero on success or an appropriate error code on failure. 338 */ 339 int 340 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, struct vnode **vpp) 341 { 342 int error; 343 struct vnode *nvp; 344 struct vnode *vp; 345 346 vp = NULL; 347 348 if (node->tn_vnode != NULL) { 349 vp = node->tn_vnode; 350 vget(vp, LK_EXCLUSIVE | LK_RETRY); 351 error = 0; 352 goto out; 353 } 354 355 /* Get a new vnode and associate it with our node. */ 356 error = getnewvnode(VT_TMPFS, mp, tmpfs_vnodeop_p, &vp); 357 if (error != 0) 358 goto out; 359 KASSERT(vp != NULL); 360 361 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 362 if (error != 0) { 363 vp->v_data = NULL; 364 ungetnewvnode(vp); 365 vp = NULL; 366 goto out; 367 } 368 369 vp->v_data = node; 370 vp->v_type = node->tn_type; 371 372 /* Type-specific initialization. */ 373 switch (node->tn_type) { 374 case VBLK: 375 /* FALLTHROUGH */ 376 case VCHR: 377 vp->v_op = tmpfs_specop_p; 378 nvp = checkalias(vp, node->tn_spec.tn_dev.tn_rdev, mp); 379 if (nvp != NULL) { 380 /* Discard unneeded vnode, but save its inode. */ 381 nvp->v_data = vp->v_data; 382 vp->v_data = NULL; 383 384 /* XXX spec_vnodeops has no locking, so we have to 385 * do it explicitly. */ 386 VOP_UNLOCK(vp, 0); 387 vp->v_op = spec_vnodeop_p; 388 vp->v_flag &= ~VLOCKSWORK; 389 vrele(vp); 390 vgone(vp); 391 392 /* Reinitialize aliased node. */ 393 vp = nvp; 394 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 395 if (error != 0) { 396 vp->v_data = NULL; 397 vp = NULL; 398 goto out; 399 } 400 } 401 break; 402 403 case VDIR: 404 vp->v_flag = node->tn_spec.tn_dir.tn_parent == node ? VROOT : 0; 405 break; 406 407 case VFIFO: 408 vp->v_op = tmpfs_fifoop_p; 409 break; 410 411 case VLNK: 412 /* FALLTHROUGH */ 413 case VREG: 414 /* FALLTHROUGH */ 415 case VSOCK: 416 break; 417 418 default: 419 KASSERT(0); 420 } 421 422 uvm_vnp_setsize(vp, node->tn_size); 423 424 error = 0; 425 426 out: 427 *vpp = node->tn_vnode = vp; 428 429 KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp))); 430 KASSERT(*vpp == node->tn_vnode); 431 432 return error; 433 } 434 435 /* --------------------------------------------------------------------- */ 436 437 /* 438 * Destroys the association between the vnode vp and the node it 439 * references. 440 */ 441 void 442 tmpfs_free_vp(struct vnode *vp) 443 { 444 struct tmpfs_node *node; 445 446 node = VP_TO_TMPFS_NODE(vp); 447 448 node->tn_vnode = NULL; 449 vp->v_data = NULL; 450 } 451 452 /* --------------------------------------------------------------------- */ 453 454 /* 455 * Allocates a new file of type 'type' and adds it to the parent directory 456 * 'dvp'; this addition is done using the component name given in 'cnp'. 457 * The ownership of the new file is automatically assigned based on the 458 * credentials of the caller (through 'cnp'), the group is set based on 459 * the parent directory and the mode is determined from the 'vap' argument. 460 * If successful, *vpp holds a vnode to the newly created file and zero 461 * is returned. Otherwise *vpp is NULL and the function returns an 462 * appropriate error code. 463 */ 464 int 465 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 466 struct componentname *cnp, char *target) 467 { 468 int error; 469 struct tmpfs_dirent *de; 470 struct tmpfs_mount *tmp; 471 struct tmpfs_node *dnode; 472 struct tmpfs_node *node; 473 struct tmpfs_node *parent; 474 475 KASSERT(VOP_ISLOCKED(dvp)); 476 KASSERT(cnp->cn_flags & HASBUF); 477 478 tmp = VFS_TO_TMPFS(dvp->v_mount); 479 dnode = VP_TO_TMPFS_DIR(dvp); 480 *vpp = NULL; 481 482 /* If the entry we are creating is a directory, we cannot overflow 483 * the number of links of its parent, because it will get a new 484 * link. */ 485 if (vap->va_type == VDIR) { 486 /* Ensure that we do not overflow the maximum number of links 487 * imposed by the system. */ 488 KASSERT(dnode->tn_links <= LINK_MAX); 489 if (dnode->tn_links == LINK_MAX) { 490 error = EMLINK; 491 goto out; 492 } 493 494 parent = dnode; 495 } else 496 parent = NULL; 497 498 /* Allocate a node that represents the new file. */ 499 error = tmpfs_alloc_node(tmp, vap->va_type, kauth_cred_geteuid(cnp->cn_cred), 500 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, 501 cnp->cn_lwp->l_proc, &node); 502 if (error != 0) 503 goto out; 504 505 /* Allocate a directory entry that points to the new file. */ 506 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 507 &de); 508 if (error != 0) { 509 tmpfs_free_node(tmp, node); 510 goto out; 511 } 512 513 /* Allocate a vnode for the new file. */ 514 error = tmpfs_alloc_vp(dvp->v_mount, node, vpp); 515 if (error != 0) { 516 tmpfs_free_dirent(tmp, de, TRUE); 517 tmpfs_free_node(tmp, node); 518 goto out; 519 } 520 521 /* Now that all required items are allocated, we can proceed to 522 * insert the new node into the directory, an operation that 523 * cannot fail. */ 524 tmpfs_dir_attach(dvp, de); 525 VN_KNOTE(dvp, NOTE_WRITE); 526 527 out: 528 if (error != 0 || !(cnp->cn_flags & SAVESTART)) 529 PNBUF_PUT(cnp->cn_pnbuf); 530 vput(dvp); 531 532 KASSERT(!VOP_ISLOCKED(dvp)); 533 KASSERT(IFF(error == 0, *vpp != NULL)); 534 535 return error; 536 } 537 538 /* --------------------------------------------------------------------- */ 539 540 /* 541 * Attaches the directory entry de to the directory represented by vp. 542 * Note that this does not change the link count of the node pointed by 543 * the directory entry, as this is done by tmpfs_alloc_dirent. 544 */ 545 void 546 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 547 { 548 struct tmpfs_node *dnode; 549 550 dnode = VP_TO_TMPFS_DIR(vp); 551 552 TAILQ_INSERT_TAIL(&dnode->tn_spec.tn_dir.tn_dir, de, td_entries); 553 dnode->tn_size += sizeof(struct tmpfs_dirent); 554 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 555 TMPFS_NODE_MODIFIED; 556 uvm_vnp_setsize(vp, dnode->tn_size); 557 } 558 559 /* --------------------------------------------------------------------- */ 560 561 /* 562 * Detaches the directory entry de from the directory represented by vp. 563 * Note that this does not change the link count of the node pointed by 564 * the directory entry, as this is done by tmpfs_free_dirent. 565 */ 566 void 567 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 568 { 569 struct tmpfs_node *dnode; 570 571 KASSERT(VOP_ISLOCKED(vp)); 572 573 dnode = VP_TO_TMPFS_DIR(vp); 574 575 if (dnode->tn_spec.tn_dir.tn_readdir_lastp == de) { 576 dnode->tn_spec.tn_dir.tn_readdir_lastn = 0; 577 dnode->tn_spec.tn_dir.tn_readdir_lastp = NULL; 578 } 579 580 TAILQ_REMOVE(&dnode->tn_spec.tn_dir.tn_dir, de, td_entries); 581 dnode->tn_size -= sizeof(struct tmpfs_dirent); 582 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 583 TMPFS_NODE_MODIFIED; 584 uvm_vnp_setsize(vp, dnode->tn_size); 585 } 586 587 /* --------------------------------------------------------------------- */ 588 589 /* 590 * Looks for a directory entry in the directory represented by node. 591 * 'cnp' describes the name of the entry to look for. Note that the . 592 * and .. components are not allowed as they do not physically exist 593 * within directories. 594 * 595 * Returns a pointer to the entry when found, otherwise NULL. 596 */ 597 struct tmpfs_dirent * 598 tmpfs_dir_lookup(struct tmpfs_node *node, struct componentname *cnp) 599 { 600 boolean_t found; 601 struct tmpfs_dirent *de; 602 603 KASSERT(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 604 KASSERT(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 605 cnp->cn_nameptr[1] == '.'))); 606 TMPFS_VALIDATE_DIR(node); 607 608 node->tn_status |= TMPFS_NODE_ACCESSED; 609 610 found = 0; 611 TAILQ_FOREACH(de, &node->tn_spec.tn_dir.tn_dir, td_entries) { 612 KASSERT(cnp->cn_namelen < 0xffff); 613 if (de->td_namelen == (uint16_t)cnp->cn_namelen && 614 memcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) { 615 found = 1; 616 break; 617 } 618 } 619 620 return found ? de : NULL; 621 } 622 623 /* --------------------------------------------------------------------- */ 624 625 /* 626 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 627 * directory and returns it in the uio space. The function returns 0 628 * on success, -1 if there was not enough space in the uio structure to 629 * hold the directory entry or an appropriate error code if another 630 * error happens. 631 */ 632 int 633 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 634 { 635 int error; 636 struct dirent dent; 637 638 TMPFS_VALIDATE_DIR(node); 639 KASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 640 641 dent.d_fileno = node->tn_id; 642 dent.d_type = DT_DIR; 643 dent.d_namlen = 1; 644 dent.d_name[0] = '.'; 645 dent.d_name[1] = '\0'; 646 dent.d_reclen = _DIRENT_SIZE(&dent); 647 648 if (dent.d_reclen > uio->uio_resid) 649 error = -1; 650 else { 651 error = uiomove(&dent, dent.d_reclen, uio); 652 if (error == 0) 653 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 654 } 655 656 node->tn_status |= TMPFS_NODE_ACCESSED; 657 658 return error; 659 } 660 661 /* --------------------------------------------------------------------- */ 662 663 /* 664 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 665 * directory and returns it in the uio space. The function returns 0 666 * on success, -1 if there was not enough space in the uio structure to 667 * hold the directory entry or an appropriate error code if another 668 * error happens. 669 */ 670 int 671 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 672 { 673 int error; 674 struct dirent dent; 675 676 TMPFS_VALIDATE_DIR(node); 677 KASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 678 679 dent.d_fileno = node->tn_spec.tn_dir.tn_parent->tn_id; 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 = _DIRENT_SIZE(&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_spec.tn_dir.tn_dir); 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_spec.tn_dir.tn_readdir_lastn && 718 node->tn_spec.tn_dir.tn_readdir_lastp != NULL) { 719 return node->tn_spec.tn_dir.tn_readdir_lastp; 720 } 721 722 TAILQ_FOREACH(de, &node->tn_spec.tn_dir.tn_dir, 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 KASSERT(startcookie != TMPFS_DIRCOOKIE_DOT); 754 KASSERT(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 KASSERT(0); 803 } 804 d.d_namlen = de->td_namelen; 805 KASSERT(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 = _DIRENT_SIZE(&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_spec.tn_dir.tn_readdir_lastn = 0; 829 node->tn_spec.tn_dir.tn_readdir_lastp = NULL; 830 } else { 831 node->tn_spec.tn_dir.tn_readdir_lastn = uio->uio_offset = 832 TMPFS_DIRCOOKIE(de); 833 node->tn_spec.tn_dir.tn_readdir_lastp = de; 834 } 835 836 node->tn_status |= TMPFS_NODE_ACCESSED; 837 838 return error; 839 } 840 841 /* --------------------------------------------------------------------- */ 842 843 /* 844 * Resizes the aobj associated to the regular file pointed to by vp to 845 * the size newsize. 'vp' must point to a vnode that represents a regular 846 * file. 'newsize' must be positive. 847 * 848 * Returns zero on success or an appropriate error code on failure. 849 */ 850 int 851 tmpfs_reg_resize(struct vnode *vp, off_t newsize) 852 { 853 int error; 854 size_t newpages, oldpages; 855 struct tmpfs_mount *tmp; 856 struct tmpfs_node *node; 857 off_t oldsize; 858 859 KASSERT(vp->v_type == VREG); 860 KASSERT(newsize >= 0); 861 862 node = VP_TO_TMPFS_NODE(vp); 863 tmp = VFS_TO_TMPFS(vp->v_mount); 864 865 /* Convert the old and new sizes to the number of pages needed to 866 * store them. It may happen that we do not need to do anything 867 * because the last allocated page can accommodate the change on 868 * its own. */ 869 oldsize = node->tn_size; 870 oldpages = round_page(oldsize) / PAGE_SIZE; 871 KASSERT(oldpages == node->tn_spec.tn_reg.tn_aobj_pages); 872 newpages = round_page(newsize) / PAGE_SIZE; 873 874 if (newpages > oldpages && 875 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) { 876 error = ENOSPC; 877 goto out; 878 } 879 880 node->tn_spec.tn_reg.tn_aobj_pages = newpages; 881 882 tmp->tm_pages_used += (newpages - oldpages); 883 node->tn_size = newsize; 884 uvm_vnp_setsize(vp, newsize); 885 if (newsize < oldsize) { 886 int zerolen = MIN(round_page(newsize), node->tn_size) - newsize; 887 888 /* 889 * free "backing store" 890 */ 891 892 if (newpages < oldpages) { 893 struct uvm_object *uobj; 894 895 uobj = node->tn_spec.tn_reg.tn_aobj; 896 897 simple_lock(&uobj->vmobjlock); 898 uao_dropswap_range(uobj, newpages, oldpages); 899 simple_unlock(&uobj->vmobjlock); 900 } 901 902 /* 903 * zero out the truncated part of the last page. 904 */ 905 906 uvm_vnp_zerorange(vp, newsize, zerolen); 907 } 908 909 error = 0; 910 911 out: 912 return error; 913 } 914 915 /* --------------------------------------------------------------------- */ 916 917 /* 918 * Returns information about the number of available memory pages, 919 * including physical and virtual ones. 920 * 921 * If 'total' is TRUE, the value returned is the total amount of memory 922 * pages configured for the system (either in use or free). 923 * If it is FALSE, the value returned is the amount of free memory pages. 924 * 925 * Remember to remove TMPFS_PAGES_RESERVED from the returned value to avoid 926 * excessive memory usage. 927 * 928 */ 929 size_t 930 tmpfs_mem_info(boolean_t total) 931 { 932 size_t size; 933 934 size = 0; 935 size += uvmexp.swpgavail; 936 if (!total) { 937 size -= uvmexp.swpgonly; 938 } 939 size += uvmexp.free; 940 size += uvmexp.filepages; 941 if (size > uvmexp.wired) { 942 size -= uvmexp.wired; 943 } else { 944 size = 0; 945 } 946 947 return size; 948 } 949 950 /* --------------------------------------------------------------------- */ 951 952 /* 953 * Change flags of the given vnode. 954 * Caller should execute tmpfs_update on vp after a successful execution. 955 * The vnode must be locked on entry and remain locked on exit. 956 */ 957 int 958 tmpfs_chflags(struct vnode *vp, int flags, kauth_cred_t cred, struct proc *p) 959 { 960 int error; 961 struct tmpfs_node *node; 962 963 KASSERT(VOP_ISLOCKED(vp)); 964 965 node = VP_TO_TMPFS_NODE(vp); 966 967 /* Disallow this operation if the file system is mounted read-only. */ 968 if (vp->v_mount->mnt_flag & MNT_RDONLY) 969 return EROFS; 970 971 /* XXX: The following comes from UFS code, and can be found in 972 * several other file systems. Shouldn't this be centralized 973 * somewhere? */ 974 if (kauth_cred_geteuid(cred) != node->tn_uid && 975 (error = kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 976 &p->p_acflag))) 977 return error; 978 if (kauth_cred_geteuid(cred) == 0) { 979 /* The super-user is only allowed to change flags if the file 980 * wasn't protected before and the securelevel is zero. */ 981 if ((node->tn_flags & (SF_IMMUTABLE | SF_APPEND)) && 982 securelevel > 0) 983 return EPERM; 984 node->tn_flags = flags; 985 } else { 986 /* Regular users can change flags provided they only want to 987 * change user-specific ones, not those reserved for the 988 * super-user. */ 989 if ((node->tn_flags & (SF_IMMUTABLE | SF_APPEND)) || 990 (flags & UF_SETTABLE) != flags) 991 return EPERM; 992 if ((node->tn_flags & SF_SETTABLE) != (flags & SF_SETTABLE)) 993 return EPERM; 994 node->tn_flags &= SF_SETTABLE; 995 node->tn_flags |= (flags & UF_SETTABLE); 996 } 997 998 node->tn_status |= TMPFS_NODE_CHANGED; 999 VN_KNOTE(vp, NOTE_ATTRIB); 1000 1001 KASSERT(VOP_ISLOCKED(vp)); 1002 1003 return 0; 1004 } 1005 1006 /* --------------------------------------------------------------------- */ 1007 1008 /* 1009 * Change access mode on the given vnode. 1010 * Caller should execute tmpfs_update on vp after a successful execution. 1011 * The vnode must be locked on entry and remain locked on exit. 1012 */ 1013 int 1014 tmpfs_chmod(struct vnode *vp, mode_t mode, kauth_cred_t cred, struct proc *p) 1015 { 1016 int error, ismember = 0; 1017 struct tmpfs_node *node; 1018 1019 KASSERT(VOP_ISLOCKED(vp)); 1020 1021 node = VP_TO_TMPFS_NODE(vp); 1022 1023 /* Disallow this operation if the file system is mounted read-only. */ 1024 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1025 return EROFS; 1026 1027 /* Immutable or append-only files cannot be modified, either. */ 1028 if (node->tn_flags & (IMMUTABLE | APPEND)) 1029 return EPERM; 1030 1031 /* XXX: The following comes from UFS code, and can be found in 1032 * several other file systems. Shouldn't this be centralized 1033 * somewhere? */ 1034 if (kauth_cred_geteuid(cred) != node->tn_uid && 1035 (error = kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 1036 &p->p_acflag))) 1037 return error; 1038 if (kauth_cred_geteuid(cred) != 0) { 1039 if (vp->v_type != VDIR && (mode & S_ISTXT)) 1040 return EFTYPE; 1041 1042 if ((kauth_cred_ismember_gid(cred, node->tn_gid, 1043 &ismember) != 0 || !ismember) && (mode & S_ISGID)) 1044 return EPERM; 1045 } 1046 1047 node->tn_mode = (mode & ALLPERMS); 1048 1049 node->tn_status |= TMPFS_NODE_CHANGED; 1050 VN_KNOTE(vp, NOTE_ATTRIB); 1051 1052 KASSERT(VOP_ISLOCKED(vp)); 1053 1054 return 0; 1055 } 1056 1057 /* --------------------------------------------------------------------- */ 1058 1059 /* 1060 * Change ownership of the given vnode. At least one of uid or gid must 1061 * be different than VNOVAL. If one is set to that value, the attribute 1062 * is unchanged. 1063 * Caller should execute tmpfs_update on vp after a successful execution. 1064 * The vnode must be locked on entry and remain locked on exit. 1065 */ 1066 int 1067 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, kauth_cred_t cred, 1068 struct proc *p) 1069 { 1070 int error, ismember = 0; 1071 struct tmpfs_node *node; 1072 1073 KASSERT(VOP_ISLOCKED(vp)); 1074 1075 node = VP_TO_TMPFS_NODE(vp); 1076 1077 /* Assign default values if they are unknown. */ 1078 KASSERT(uid != VNOVAL || gid != VNOVAL); 1079 if (uid == VNOVAL) 1080 uid = node->tn_uid; 1081 if (gid == VNOVAL) 1082 gid = node->tn_gid; 1083 KASSERT(uid != VNOVAL && gid != VNOVAL); 1084 1085 /* Disallow this operation if the file system is mounted read-only. */ 1086 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1087 return EROFS; 1088 1089 /* Immutable or append-only files cannot be modified, either. */ 1090 if (node->tn_flags & (IMMUTABLE | APPEND)) 1091 return EPERM; 1092 1093 /* XXX: The following comes from UFS code, and can be found in 1094 * several other file systems. Shouldn't this be centralized 1095 * somewhere? */ 1096 if ((kauth_cred_geteuid(cred) != node->tn_uid || uid != node->tn_uid || 1097 (gid != node->tn_gid && !(kauth_cred_getegid(cred) == node->tn_gid || 1098 (kauth_cred_ismember_gid(cred, gid, &ismember) == 0 && ismember)))) && 1099 ((error = kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 1100 &p->p_acflag)) != 0)) 1101 return error; 1102 1103 node->tn_uid = uid; 1104 node->tn_gid = gid; 1105 1106 node->tn_status |= TMPFS_NODE_CHANGED; 1107 VN_KNOTE(vp, NOTE_ATTRIB); 1108 1109 KASSERT(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, kauth_cred_t cred, 1123 struct proc *p) 1124 { 1125 int error; 1126 struct tmpfs_node *node; 1127 1128 KASSERT(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 KASSERT(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 int vaflags, kauth_cred_t cred, struct lwp *l) 1181 { 1182 int error; 1183 struct tmpfs_node *node; 1184 1185 KASSERT(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 /* XXX: The following comes from UFS code, and can be found in 1198 * several other file systems. Shouldn't this be centralized 1199 * somewhere? */ 1200 if (kauth_cred_geteuid(cred) != node->tn_uid && 1201 (error = kauth_authorize_generic(cred, KAUTH_GENERIC_ISSUSER, 1202 &l->l_proc->p_acflag)) && 1203 ((vaflags & VA_UTIMES_NULL) == 0 || 1204 (error = VOP_ACCESS(vp, VWRITE, cred, l)))) 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 tmpfs_update(vp, atime, mtime, 0); 1214 1215 KASSERT(VOP_ISLOCKED(vp)); 1216 1217 return 0; 1218 } 1219 1220 /* --------------------------------------------------------------------- */ 1221 1222 /* Sync timestamps */ 1223 void 1224 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1225 const struct timespec *mod) 1226 { 1227 struct timespec now; 1228 struct tmpfs_node *node; 1229 1230 node = VP_TO_TMPFS_NODE(vp); 1231 1232 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1233 TMPFS_NODE_CHANGED)) == 0) 1234 return; 1235 1236 getnanotime(&now); 1237 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1238 if (acc == NULL) 1239 acc = &now; 1240 node->tn_atime = *acc; 1241 } 1242 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1243 if (mod == NULL) 1244 mod = &now; 1245 node->tn_mtime = *mod; 1246 } 1247 if (node->tn_status & TMPFS_NODE_CHANGED) 1248 node->tn_ctime = now; 1249 1250 node->tn_status &= 1251 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1252 } 1253 1254 /* --------------------------------------------------------------------- */ 1255 1256 void 1257 tmpfs_update(struct vnode *vp, const struct timespec *acc, 1258 const struct timespec *mod, int flags) 1259 { 1260 1261 struct tmpfs_node *node; 1262 1263 KASSERT(VOP_ISLOCKED(vp)); 1264 1265 node = VP_TO_TMPFS_NODE(vp); 1266 1267 if (flags & UPDATE_CLOSE) 1268 ; /* XXX Need to do anything special? */ 1269 1270 tmpfs_itimes(vp, acc, mod); 1271 1272 KASSERT(VOP_ISLOCKED(vp)); 1273 } 1274 1275 /* --------------------------------------------------------------------- */ 1276 1277 int 1278 tmpfs_truncate(struct vnode *vp, off_t length) 1279 { 1280 boolean_t extended; 1281 int error; 1282 struct tmpfs_node *node; 1283 1284 node = VP_TO_TMPFS_NODE(vp); 1285 extended = length > node->tn_size; 1286 1287 if (length < 0) { 1288 error = EINVAL; 1289 goto out; 1290 } 1291 1292 if (node->tn_size == length) { 1293 error = 0; 1294 goto out; 1295 } 1296 1297 error = tmpfs_reg_resize(vp, length); 1298 if (error == 0) { 1299 VN_KNOTE(vp, NOTE_ATTRIB | (extended ? NOTE_EXTEND : 0)); 1300 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1301 } 1302 1303 out: 1304 tmpfs_update(vp, NULL, NULL, 0); 1305 1306 return error; 1307 } 1308