1 /* $NetBSD: tmpfs_subr.c,v 1.18 2006/02/16 14:57:50 jmmv 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 1.18 2006/02/16 14:57:50 jmmv 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 59 #include <uvm/uvm.h> 60 61 #include <miscfs/specfs/specdev.h> 62 #include <fs/tmpfs/tmpfs.h> 63 #include <fs/tmpfs/tmpfs_fifoops.h> 64 #include <fs/tmpfs/tmpfs_specops.h> 65 #include <fs/tmpfs/tmpfs_vnops.h> 66 67 /* --------------------------------------------------------------------- */ 68 69 /* 70 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 71 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 72 * using the credentials of the process 'p'. 73 * 74 * If the node type is set to 'VDIR', then the parent parameter must point 75 * to the parent directory of the node being created. It may only be NULL 76 * while allocating the root node. 77 * 78 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 79 * specifies the device the node represents. 80 * 81 * If the node type is set to 'VLNK', then the parameter target specifies 82 * the file name of the target file for the symbolic link that is being 83 * created. 84 * 85 * Note that new nodes are retrieved from the available list if it has 86 * items or, if it is empty, from the node pool as long as there is enough 87 * space to create them. 88 * 89 * Returns zero on success or an appropriate error code on failure. 90 */ 91 int 92 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type, 93 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 94 char *target, dev_t rdev, struct proc *p, struct tmpfs_node **node) 95 { 96 struct tmpfs_node *nnode; 97 98 /* If the root directory of the 'tmp' file system is not yet 99 * allocated, this must be the request to do it. */ 100 KASSERT(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 101 102 KASSERT(IFF(type == VLNK, target != NULL)); 103 KASSERT(IFF(type == VBLK || type == VCHR, rdev != VNOVAL)); 104 105 KASSERT(uid != VNOVAL && gid != VNOVAL && mode != VNOVAL); 106 107 nnode = NULL; 108 if (LIST_EMPTY(&tmp->tm_nodes_avail)) { 109 KASSERT(tmp->tm_nodes_last <= tmp->tm_nodes_max); 110 if (tmp->tm_nodes_last == tmp->tm_nodes_max) 111 return ENOSPC; 112 113 nnode = 114 (struct tmpfs_node *)TMPFS_POOL_GET(&tmp->tm_node_pool, 0); 115 if (nnode == NULL) 116 return ENOSPC; 117 nnode->tn_id = tmp->tm_nodes_last++; 118 nnode->tn_gen = 0; 119 } else { 120 nnode = LIST_FIRST(&tmp->tm_nodes_avail); 121 LIST_REMOVE(nnode, tn_entries); 122 nnode->tn_gen++; 123 } 124 KASSERT(nnode != NULL); 125 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 126 127 /* Generic initialization. */ 128 nnode->tn_type = type; 129 nnode->tn_size = 0; 130 nnode->tn_status = 0; 131 nnode->tn_flags = 0; 132 nnode->tn_links = 0; 133 (void)nanotime(&nnode->tn_atime); 134 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 135 nnode->tn_atime; 136 nnode->tn_uid = uid; 137 nnode->tn_gid = gid; 138 nnode->tn_mode = mode; 139 nnode->tn_lockf = NULL; 140 nnode->tn_vnode = NULL; 141 142 /* Type-specific initialization. */ 143 switch (nnode->tn_type) { 144 case VBLK: 145 case VCHR: 146 nnode->tn_spec.tn_dev.tn_rdev = rdev; 147 break; 148 149 case VDIR: 150 TAILQ_INIT(&nnode->tn_spec.tn_dir.tn_dir); 151 nnode->tn_spec.tn_dir.tn_parent = 152 (parent == NULL) ? nnode : parent; 153 nnode->tn_spec.tn_dir.tn_readdir_lastn = 0; 154 nnode->tn_spec.tn_dir.tn_readdir_lastp = NULL; 155 nnode->tn_links++; 156 nnode->tn_spec.tn_dir.tn_parent->tn_links++; 157 break; 158 159 case VFIFO: 160 /* FALLTHROUGH */ 161 case VSOCK: 162 break; 163 164 case VLNK: 165 KASSERT(strlen(target) < MAXPATHLEN); 166 nnode->tn_size = strlen(target); 167 nnode->tn_spec.tn_lnk.tn_link = 168 tmpfs_str_pool_get(&tmp->tm_str_pool, nnode->tn_size, 0); 169 if (nnode->tn_spec.tn_lnk.tn_link == NULL) { 170 nnode->tn_type = VNON; 171 tmpfs_free_node(tmp, nnode); 172 return ENOSPC; 173 } 174 memcpy(nnode->tn_spec.tn_lnk.tn_link, target, nnode->tn_size); 175 break; 176 177 case VREG: 178 nnode->tn_spec.tn_reg.tn_aobj = 179 uao_create(INT32_MAX - PAGE_SIZE, 0); 180 nnode->tn_spec.tn_reg.tn_aobj_pages = 0; 181 break; 182 183 default: 184 KASSERT(0); 185 } 186 187 *node = nnode; 188 return 0; 189 } 190 191 /* --------------------------------------------------------------------- */ 192 193 /* 194 * Destroys the node pointed to by node from the file system 'tmp'. 195 * If the node does not belong to the given mount point, the results are 196 * unpredicted. 197 * 198 * If the node references a directory; no entries are allowed because 199 * their removal could need a recursive algorithm, something forbidden in 200 * kernel space. Furthermore, there is not need to provide such 201 * functionality (recursive removal) because the only primitives offered 202 * to the user are the removal of empty directories and the deletion of 203 * individual files. 204 * 205 * Note that nodes are not really deleted; in fact, when a node has been 206 * allocated, it cannot be deleted during the whole life of the file 207 * system. Instead, they are moved to the available list and remain there 208 * until reused. 209 */ 210 void 211 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 212 { 213 ino_t id; 214 unsigned long gen; 215 size_t pages; 216 217 switch (node->tn_type) { 218 case VNON: 219 /* Do not do anything. VNON is provided to let the 220 * allocation routine clean itself easily by avoiding 221 * duplicating code in it. */ 222 /* FALLTHROUGH */ 223 case VBLK: 224 /* FALLTHROUGH */ 225 case VCHR: 226 /* FALLTHROUGH */ 227 case VDIR: 228 /* FALLTHROUGH */ 229 case VFIFO: 230 /* FALLTHROUGH */ 231 case VSOCK: 232 pages = 0; 233 break; 234 235 case VLNK: 236 tmpfs_str_pool_put(&tmp->tm_str_pool, 237 node->tn_spec.tn_lnk.tn_link, node->tn_size); 238 pages = 0; 239 break; 240 241 case VREG: 242 if (node->tn_spec.tn_reg.tn_aobj != NULL) 243 uao_detach(node->tn_spec.tn_reg.tn_aobj); 244 pages = node->tn_spec.tn_reg.tn_aobj_pages; 245 break; 246 247 default: 248 KASSERT(0); 249 pages = 0; /* Shut up gcc when !DIAGNOSTIC. */ 250 break; 251 } 252 253 tmp->tm_pages_used -= pages; 254 255 LIST_REMOVE(node, tn_entries); 256 id = node->tn_id; 257 gen = node->tn_gen; 258 memset(node, 0, sizeof(struct tmpfs_node)); 259 node->tn_id = id; 260 node->tn_type = VNON; 261 node->tn_gen = gen; 262 LIST_INSERT_HEAD(&tmp->tm_nodes_avail, node, tn_entries); 263 } 264 265 /* --------------------------------------------------------------------- */ 266 267 /* 268 * Allocates a new directory entry for the node node with a name of name. 269 * The new directory entry is returned in *de. 270 * 271 * The link count of node is increased by one to reflect the new object 272 * referencing it. 273 * 274 * Returns zero on success or an appropriate error code on failure. 275 */ 276 int 277 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 278 const char *name, uint16_t len, struct tmpfs_dirent **de) 279 { 280 struct tmpfs_dirent *nde; 281 282 nde = (struct tmpfs_dirent *)TMPFS_POOL_GET(&tmp->tm_dirent_pool, 0); 283 if (nde == NULL) 284 return ENOSPC; 285 286 nde->td_name = tmpfs_str_pool_get(&tmp->tm_str_pool, len, 0); 287 if (nde->td_name == NULL) { 288 TMPFS_POOL_PUT(&tmp->tm_dirent_pool, nde); 289 return ENOSPC; 290 } 291 nde->td_namelen = len; 292 memcpy(nde->td_name, name, len); 293 nde->td_node = node; 294 295 node->tn_links++; 296 *de = nde; 297 298 return 0; 299 } 300 301 /* --------------------------------------------------------------------- */ 302 303 /* 304 * Frees a directory entry. It is the caller's responsibility to destroy 305 * the node referenced by it if needed. 306 * 307 * The link count of node is decreased by one to reflect the removal of an 308 * object that referenced it. This only happens if 'node_exists' is true; 309 * otherwise the function will not access the node referred to by the 310 * directory entry, as it may already have been released from the outside. 311 */ 312 void 313 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de, 314 boolean_t node_exists) 315 { 316 if (node_exists) { 317 struct tmpfs_node *node; 318 319 node = de->td_node; 320 321 KASSERT(node->tn_links > 0); 322 node->tn_links--; 323 } 324 325 tmpfs_str_pool_put(&tmp->tm_str_pool, de->td_name, de->td_namelen); 326 TMPFS_POOL_PUT(&tmp->tm_dirent_pool, de); 327 } 328 329 /* --------------------------------------------------------------------- */ 330 331 /* 332 * Allocates a new vnode for the node node or returns a new reference to 333 * an existing one if the node had already a vnode referencing it. The 334 * resulting locked vnode is returned in *vpp. 335 * 336 * Returns zero on success or an appropriate error code on failure. 337 */ 338 int 339 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, struct vnode **vpp) 340 { 341 int error; 342 struct vnode *nvp; 343 struct vnode *vp; 344 345 vp = NULL; 346 347 if (node->tn_vnode != NULL) { 348 vp = node->tn_vnode; 349 vget(vp, LK_EXCLUSIVE | LK_RETRY); 350 error = 0; 351 goto out; 352 } 353 354 /* Get a new vnode and associate it with our node. */ 355 error = getnewvnode(VT_TMPFS, mp, tmpfs_vnodeop_p, &vp); 356 if (error != 0) 357 goto out; 358 KASSERT(vp != NULL); 359 360 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 361 if (error != 0) { 362 vp->v_data = NULL; 363 ungetnewvnode(vp); 364 vp = NULL; 365 goto out; 366 } 367 368 vp->v_data = node; 369 vp->v_type = node->tn_type; 370 371 /* Type-specific initialization. */ 372 switch (node->tn_type) { 373 case VBLK: 374 /* FALLTHROUGH */ 375 case VCHR: 376 vp->v_op = tmpfs_specop_p; 377 nvp = checkalias(vp, node->tn_spec.tn_dev.tn_rdev, mp); 378 if (nvp != NULL) { 379 /* Discard unneeded vnode, but save its inode. */ 380 nvp->v_data = vp->v_data; 381 vp->v_data = NULL; 382 383 /* XXX spec_vnodeops has no locking, so we have to 384 * do it explicitly. */ 385 VOP_UNLOCK(vp, 0); 386 vp->v_op = spec_vnodeop_p; 387 vp->v_flag &= ~VLOCKSWORK; 388 vrele(vp); 389 vgone(vp); 390 391 /* Reinitialize aliased node. */ 392 vp = nvp; 393 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 394 if (error != 0) { 395 vp->v_data = NULL; 396 vp = NULL; 397 goto out; 398 } 399 } 400 break; 401 402 case VDIR: 403 vp->v_flag = node->tn_spec.tn_dir.tn_parent == node ? VROOT : 0; 404 break; 405 406 case VFIFO: 407 vp->v_op = tmpfs_fifoop_p; 408 break; 409 410 case VLNK: 411 /* FALLTHROUGH */ 412 case VREG: 413 /* FALLTHROUGH */ 414 case VSOCK: 415 break; 416 417 default: 418 KASSERT(0); 419 } 420 421 uvm_vnp_setsize(vp, node->tn_size); 422 423 error = 0; 424 425 out: 426 *vpp = node->tn_vnode = vp; 427 428 KASSERT(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp))); 429 KASSERT(*vpp == node->tn_vnode); 430 431 return error; 432 } 433 434 /* --------------------------------------------------------------------- */ 435 436 /* 437 * Destroys the association between the vnode vp and the node it 438 * references. 439 */ 440 void 441 tmpfs_free_vp(struct vnode *vp) 442 { 443 struct tmpfs_node *node; 444 445 node = VP_TO_TMPFS_NODE(vp); 446 447 node->tn_vnode = NULL; 448 vp->v_data = NULL; 449 } 450 451 /* --------------------------------------------------------------------- */ 452 453 /* 454 * Allocates a new file of type 'type' and adds it to the parent directory 455 * 'dvp'; this addition is done using the component name given in 'cnp'. 456 * The ownership of the new file is automatically assigned based on the 457 * credentials of the caller (through 'cnp'), the group is set based on 458 * the parent directory and the mode is determined from the 'vap' argument. 459 * If successful, *vpp holds a vnode to the newly created file and zero 460 * is returned. Otherwise *vpp is NULL and the function returns an 461 * appropriate error code. 462 */ 463 int 464 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 465 struct componentname *cnp, char *target) 466 { 467 int error; 468 struct tmpfs_dirent *de; 469 struct tmpfs_mount *tmp; 470 struct tmpfs_node *dnode; 471 struct tmpfs_node *node; 472 struct tmpfs_node *parent; 473 474 KASSERT(VOP_ISLOCKED(dvp)); 475 KASSERT(cnp->cn_flags & HASBUF); 476 477 tmp = VFS_TO_TMPFS(dvp->v_mount); 478 dnode = VP_TO_TMPFS_DIR(dvp); 479 *vpp = NULL; 480 481 /* If the entry we are creating is a directory, we cannot overflow 482 * the number of links of its parent, because it will get a new 483 * link. */ 484 if (vap->va_type == VDIR) { 485 /* Ensure that we do not overflow the maximum number of links 486 * imposed by the system. */ 487 KASSERT(dnode->tn_links <= LINK_MAX); 488 if (dnode->tn_links == LINK_MAX) { 489 error = EMLINK; 490 goto out; 491 } 492 493 parent = dnode; 494 } else 495 parent = NULL; 496 497 /* Allocate a node that represents the new file. */ 498 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid, 499 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, 500 cnp->cn_lwp->l_proc, &node); 501 if (error != 0) 502 goto out; 503 504 /* Allocate a directory entry that points to the new file. */ 505 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 506 &de); 507 if (error != 0) { 508 tmpfs_free_node(tmp, node); 509 goto out; 510 } 511 512 /* Allocate a vnode for the new file. */ 513 error = tmpfs_alloc_vp(dvp->v_mount, node, vpp); 514 if (error != 0) { 515 tmpfs_free_dirent(tmp, de, TRUE); 516 tmpfs_free_node(tmp, node); 517 goto out; 518 } 519 520 /* Now that all required items are allocated, we can proceed to 521 * insert the new node into the directory, an operation that 522 * cannot fail. */ 523 tmpfs_dir_attach(dvp, de); 524 VN_KNOTE(dvp, NOTE_WRITE); 525 526 out: 527 if (error != 0 || !(cnp->cn_flags & SAVESTART)) 528 PNBUF_PUT(cnp->cn_pnbuf); 529 vput(dvp); 530 531 KASSERT(!VOP_ISLOCKED(dvp)); 532 KASSERT(IFF(error == 0, *vpp != NULL)); 533 534 return error; 535 } 536 537 /* --------------------------------------------------------------------- */ 538 539 /* 540 * Attaches the directory entry de to the directory represented by vp. 541 * Note that this does not change the link count of the node pointed by 542 * the directory entry, as this is done by tmpfs_alloc_dirent. 543 */ 544 void 545 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 546 { 547 struct tmpfs_node *dnode; 548 549 dnode = VP_TO_TMPFS_DIR(vp); 550 551 TAILQ_INSERT_TAIL(&dnode->tn_spec.tn_dir.tn_dir, de, td_entries); 552 dnode->tn_size += sizeof(struct tmpfs_dirent); 553 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 554 TMPFS_NODE_MODIFIED; 555 uvm_vnp_setsize(vp, dnode->tn_size); 556 } 557 558 /* --------------------------------------------------------------------- */ 559 560 /* 561 * Detaches the directory entry de from the directory represented by vp. 562 * Note that this does not change the link count of the node pointed by 563 * the directory entry, as this is done by tmpfs_free_dirent. 564 */ 565 void 566 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 567 { 568 struct tmpfs_node *dnode; 569 570 KASSERT(VOP_ISLOCKED(vp)); 571 572 dnode = VP_TO_TMPFS_DIR(vp); 573 574 if (dnode->tn_spec.tn_dir.tn_readdir_lastp == de) { 575 dnode->tn_spec.tn_dir.tn_readdir_lastn = 0; 576 dnode->tn_spec.tn_dir.tn_readdir_lastp = NULL; 577 } 578 579 TAILQ_REMOVE(&dnode->tn_spec.tn_dir.tn_dir, de, td_entries); 580 dnode->tn_size -= sizeof(struct tmpfs_dirent); 581 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 582 TMPFS_NODE_MODIFIED; 583 uvm_vnp_setsize(vp, dnode->tn_size); 584 } 585 586 /* --------------------------------------------------------------------- */ 587 588 /* 589 * Looks for a directory entry in the directory represented by node. 590 * 'cnp' describes the name of the entry to look for. Note that the . 591 * and .. components are not allowed as they do not physically exist 592 * within directories. 593 * 594 * Returns a pointer to the entry when found, otherwise NULL. 595 */ 596 struct tmpfs_dirent * 597 tmpfs_dir_lookup(struct tmpfs_node *node, struct componentname *cnp) 598 { 599 boolean_t found; 600 struct tmpfs_dirent *de; 601 602 KASSERT(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 603 KASSERT(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 604 cnp->cn_nameptr[1] == '.'))); 605 TMPFS_VALIDATE_DIR(node); 606 607 node->tn_status |= TMPFS_NODE_ACCESSED; 608 609 found = 0; 610 TAILQ_FOREACH(de, &node->tn_spec.tn_dir.tn_dir, td_entries) { 611 KASSERT(cnp->cn_namelen < 0xffff); 612 if (de->td_namelen == (uint16_t)cnp->cn_namelen && 613 memcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) { 614 found = 1; 615 break; 616 } 617 } 618 619 return found ? de : 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 KASSERT(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 = _DIRENT_SIZE(&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 KASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 677 678 dent.d_fileno = node->tn_spec.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 = _DIRENT_SIZE(&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_spec.tn_dir.tn_dir); 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_spec.tn_dir.tn_readdir_lastn && 717 node->tn_spec.tn_dir.tn_readdir_lastp != NULL) { 718 return node->tn_spec.tn_dir.tn_readdir_lastp; 719 } 720 721 TAILQ_FOREACH(de, &node->tn_spec.tn_dir.tn_dir, 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 KASSERT(startcookie != TMPFS_DIRCOOKIE_DOT); 753 KASSERT(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 KASSERT(0); 802 } 803 d.d_namlen = de->td_namelen; 804 KASSERT(de->td_namelen < sizeof(d.d_name)); 805 (void)memcpy(d.d_name, de->td_name, de->td_namelen); 806 d.d_name[de->td_namelen] = '\0'; 807 d.d_reclen = _DIRENT_SIZE(&d); 808 809 /* Stop reading if the directory entry we are treating is 810 * bigger than the amount of data that can be returned. */ 811 if (d.d_reclen > uio->uio_resid) { 812 error = -1; 813 break; 814 } 815 816 /* Copy the new dirent structure into the output buffer and 817 * advance pointers. */ 818 error = uiomove(&d, d.d_reclen, uio); 819 820 (*cntp)++; 821 de = TAILQ_NEXT(de, td_entries); 822 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 823 824 /* Update the offset and cache. */ 825 if (de == NULL) { 826 uio->uio_offset = TMPFS_DIRCOOKIE_EOF; 827 node->tn_spec.tn_dir.tn_readdir_lastn = 0; 828 node->tn_spec.tn_dir.tn_readdir_lastp = NULL; 829 } else { 830 node->tn_spec.tn_dir.tn_readdir_lastn = uio->uio_offset = 831 TMPFS_DIRCOOKIE(de); 832 node->tn_spec.tn_dir.tn_readdir_lastp = de; 833 } 834 835 node->tn_status |= TMPFS_NODE_ACCESSED; 836 837 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 KASSERT(vp->v_type == VREG); 859 KASSERT(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 KASSERT(oldpages == node->tn_spec.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_spec.tn_reg.tn_aobj_pages = newpages; 880 881 tmp->tm_pages_used += (newpages - oldpages); 882 node->tn_size = newsize; 883 uvm_vnp_setsize(vp, newsize); 884 if (newsize < oldsize) { 885 int zerolen = MIN(round_page(newsize), node->tn_size) - newsize; 886 887 /* 888 * free "backing store" 889 */ 890 891 if (newpages < oldpages) { 892 struct uvm_object *uobj; 893 894 uobj = node->tn_spec.tn_reg.tn_aobj; 895 896 simple_lock(&uobj->vmobjlock); 897 uao_dropswap_range(uobj, newpages, oldpages); 898 simple_unlock(&uobj->vmobjlock); 899 } 900 901 /* 902 * zero out the truncated part of the last page. 903 */ 904 905 uvm_vnp_zerorange(vp, newsize, zerolen); 906 } 907 908 error = 0; 909 910 out: 911 return error; 912 } 913 914 /* --------------------------------------------------------------------- */ 915 916 /* 917 * Returns information about the number of available memory pages, 918 * including physical and virtual ones. 919 * 920 * If 'total' is TRUE, the value returned is the total amount of memory 921 * pages configured for the system (either in use or free). 922 * If it is FALSE, the value returned is the amount of free memory pages. 923 * 924 * Remember to remove TMPFS_PAGES_RESERVED from the returned value to avoid 925 * excessive memory usage. 926 * 927 */ 928 size_t 929 tmpfs_mem_info(boolean_t total) 930 { 931 size_t size; 932 933 size = 0; 934 size += uvmexp.swpgavail; 935 if (!total) { 936 size -= uvmexp.swpgonly; 937 } 938 size += uvmexp.free; 939 size += uvmexp.filepages; 940 if (size > uvmexp.wired) { 941 size -= uvmexp.wired; 942 } else { 943 size = 0; 944 } 945 946 return size; 947 } 948 949 /* --------------------------------------------------------------------- */ 950 951 /* 952 * Change flags of the given vnode. 953 * Caller should execute tmpfs_update on vp after a successful execution. 954 * The vnode must be locked on entry and remain locked on exit. 955 */ 956 int 957 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct proc *p) 958 { 959 int error; 960 struct tmpfs_node *node; 961 962 KASSERT(VOP_ISLOCKED(vp)); 963 964 node = VP_TO_TMPFS_NODE(vp); 965 966 /* Disallow this operation if the file system is mounted read-only. */ 967 if (vp->v_mount->mnt_flag & MNT_RDONLY) 968 return EROFS; 969 970 /* XXX: The following comes from UFS code, and can be found in 971 * several other file systems. Shouldn't this be centralized 972 * somewhere? */ 973 if (cred->cr_uid != node->tn_uid && 974 (error = suser(cred, &p->p_acflag))) 975 return error; 976 if (cred->cr_uid == 0) { 977 /* The super-user is only allowed to change flags if the file 978 * wasn't protected before and the securelevel is zero. */ 979 if ((node->tn_flags & (SF_IMMUTABLE | SF_APPEND)) && 980 securelevel > 0) 981 return EPERM; 982 node->tn_flags = flags; 983 } else { 984 /* Regular users can change flags provided they only want to 985 * change user-specific ones, not those reserved for the 986 * super-user. */ 987 if ((node->tn_flags & (SF_IMMUTABLE | SF_APPEND)) || 988 (flags & UF_SETTABLE) != flags) 989 return EPERM; 990 if ((node->tn_flags & SF_SETTABLE) != (flags & SF_SETTABLE)) 991 return EPERM; 992 node->tn_flags &= SF_SETTABLE; 993 node->tn_flags |= (flags & UF_SETTABLE); 994 } 995 996 node->tn_status |= TMPFS_NODE_CHANGED; 997 VN_KNOTE(vp, NOTE_ATTRIB); 998 999 KASSERT(VOP_ISLOCKED(vp)); 1000 1001 return 0; 1002 } 1003 1004 /* --------------------------------------------------------------------- */ 1005 1006 /* 1007 * Change access mode on the given vnode. 1008 * Caller should execute tmpfs_update on vp after a successful execution. 1009 * The vnode must be locked on entry and remain locked on exit. 1010 */ 1011 int 1012 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct proc *p) 1013 { 1014 int error; 1015 struct tmpfs_node *node; 1016 1017 KASSERT(VOP_ISLOCKED(vp)); 1018 1019 node = VP_TO_TMPFS_NODE(vp); 1020 1021 /* Disallow this operation if the file system is mounted read-only. */ 1022 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1023 return EROFS; 1024 1025 /* Immutable or append-only files cannot be modified, either. */ 1026 if (node->tn_flags & (IMMUTABLE | APPEND)) 1027 return EPERM; 1028 1029 /* XXX: The following comes from UFS code, and can be found in 1030 * several other file systems. Shouldn't this be centralized 1031 * somewhere? */ 1032 if (cred->cr_uid != node->tn_uid && 1033 (error = suser(cred, &p->p_acflag))) 1034 return error; 1035 if (cred->cr_uid != 0) { 1036 if (vp->v_type != VDIR && (mode & S_ISTXT)) 1037 return EFTYPE; 1038 1039 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) 1040 return EPERM; 1041 } 1042 1043 node->tn_mode = (mode & ALLPERMS); 1044 1045 node->tn_status |= TMPFS_NODE_CHANGED; 1046 VN_KNOTE(vp, NOTE_ATTRIB); 1047 1048 KASSERT(VOP_ISLOCKED(vp)); 1049 1050 return 0; 1051 } 1052 1053 /* --------------------------------------------------------------------- */ 1054 1055 /* 1056 * Change ownership of the given vnode. At least one of uid or gid must 1057 * be different than VNOVAL. If one is set to that value, the attribute 1058 * is unchanged. 1059 * Caller should execute tmpfs_update on vp after a successful execution. 1060 * The vnode must be locked on entry and remain locked on exit. 1061 */ 1062 int 1063 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1064 struct proc *p) 1065 { 1066 int error; 1067 struct tmpfs_node *node; 1068 1069 KASSERT(VOP_ISLOCKED(vp)); 1070 1071 node = VP_TO_TMPFS_NODE(vp); 1072 1073 /* Assign default values if they are unknown. */ 1074 KASSERT(uid != VNOVAL || gid != VNOVAL); 1075 if (uid == VNOVAL) 1076 uid = node->tn_uid; 1077 if (gid == VNOVAL) 1078 gid = node->tn_gid; 1079 KASSERT(uid != VNOVAL && gid != VNOVAL); 1080 1081 /* Disallow this operation if the file system is mounted read-only. */ 1082 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1083 return EROFS; 1084 1085 /* Immutable or append-only files cannot be modified, either. */ 1086 if (node->tn_flags & (IMMUTABLE | APPEND)) 1087 return EPERM; 1088 1089 /* XXX: The following comes from UFS code, and can be found in 1090 * several other file systems. Shouldn't this be centralized 1091 * somewhere? */ 1092 if ((cred->cr_uid != node->tn_uid || uid != node->tn_uid || 1093 (gid != node->tn_gid && !(cred->cr_gid == node->tn_gid || 1094 groupmember(gid, cred)))) && 1095 ((error = suser(cred, &p->p_acflag)) != 0)) 1096 return error; 1097 1098 node->tn_uid = uid; 1099 node->tn_gid = gid; 1100 1101 node->tn_status |= TMPFS_NODE_CHANGED; 1102 VN_KNOTE(vp, NOTE_ATTRIB); 1103 1104 KASSERT(VOP_ISLOCKED(vp)); 1105 1106 return 0; 1107 } 1108 1109 /* --------------------------------------------------------------------- */ 1110 1111 /* 1112 * Change size of the given vnode. 1113 * Caller should execute tmpfs_update on vp after a successful execution. 1114 * The vnode must be locked on entry and remain locked on exit. 1115 */ 1116 int 1117 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1118 struct proc *p) 1119 { 1120 int error; 1121 struct tmpfs_node *node; 1122 1123 KASSERT(VOP_ISLOCKED(vp)); 1124 1125 node = VP_TO_TMPFS_NODE(vp); 1126 1127 /* Decide whether this is a valid operation based on the file type. */ 1128 error = 0; 1129 switch (vp->v_type) { 1130 case VDIR: 1131 return EISDIR; 1132 1133 case VREG: 1134 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1135 return EROFS; 1136 break; 1137 1138 case VBLK: 1139 /* FALLTHROUGH */ 1140 case VCHR: 1141 /* FALLTHROUGH */ 1142 case VFIFO: 1143 /* Allow modifications of special files even if in the file 1144 * system is mounted read-only (we are not modifying the 1145 * files themselves, but the objects they represent). */ 1146 return 0; 1147 1148 default: 1149 /* Anything else is unsupported. */ 1150 return EOPNOTSUPP; 1151 } 1152 1153 /* Immutable or append-only files cannot be modified, either. */ 1154 if (node->tn_flags & (IMMUTABLE | APPEND)) 1155 return EPERM; 1156 1157 error = tmpfs_truncate(vp, size); 1158 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1159 * for us, as will update tn_status; no need to do that here. */ 1160 1161 KASSERT(VOP_ISLOCKED(vp)); 1162 1163 return error; 1164 } 1165 1166 /* --------------------------------------------------------------------- */ 1167 1168 /* 1169 * Change access and modification times of the given vnode. 1170 * Caller should execute tmpfs_update on vp after a successful execution. 1171 * The vnode must be locked on entry and remain locked on exit. 1172 */ 1173 int 1174 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime, 1175 int vaflags, struct ucred *cred, struct lwp *l) 1176 { 1177 int error; 1178 struct tmpfs_node *node; 1179 1180 KASSERT(VOP_ISLOCKED(vp)); 1181 1182 node = VP_TO_TMPFS_NODE(vp); 1183 1184 /* Disallow this operation if the file system is mounted read-only. */ 1185 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1186 return EROFS; 1187 1188 /* Immutable or append-only files cannot be modified, either. */ 1189 if (node->tn_flags & (IMMUTABLE | APPEND)) 1190 return EPERM; 1191 1192 /* XXX: The following comes from UFS code, and can be found in 1193 * several other file systems. Shouldn't this be centralized 1194 * somewhere? */ 1195 if (cred->cr_uid != node->tn_uid && 1196 (error = suser(cred, &l->l_proc->p_acflag)) && 1197 ((vaflags & VA_UTIMES_NULL) == 0 || 1198 (error = VOP_ACCESS(vp, VWRITE, cred, l)))) 1199 return error; 1200 1201 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL) 1202 node->tn_status |= TMPFS_NODE_ACCESSED; 1203 1204 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) 1205 node->tn_status |= TMPFS_NODE_MODIFIED; 1206 1207 tmpfs_update(vp, atime, mtime, 0); 1208 1209 KASSERT(VOP_ISLOCKED(vp)); 1210 1211 return 0; 1212 } 1213 1214 /* --------------------------------------------------------------------- */ 1215 1216 /* Sync timestamps */ 1217 void 1218 tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1219 const struct timespec *mod) 1220 { 1221 struct tmpfs_node *node; 1222 const struct timespec *ts = NULL; 1223 struct timespec tsb; 1224 1225 node = VP_TO_TMPFS_NODE(vp); 1226 1227 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1228 TMPFS_NODE_CHANGED)) == 0) 1229 return; 1230 1231 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1232 if (acc == NULL) 1233 acc = ts == NULL ? (ts = nanotime(&tsb)) : ts; 1234 node->tn_atime = *acc; 1235 } 1236 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1237 if (mod == NULL) 1238 mod = ts == NULL ? (ts = nanotime(&tsb)) : ts; 1239 node->tn_mtime = *mod; 1240 } 1241 if (node->tn_status & TMPFS_NODE_CHANGED) { 1242 if (ts == NULL) 1243 ts = nanotime(&tsb); 1244 node->tn_ctime = *ts; 1245 } 1246 node->tn_status &= 1247 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED); 1248 } 1249 1250 /* --------------------------------------------------------------------- */ 1251 1252 void 1253 tmpfs_update(struct vnode *vp, const struct timespec *acc, 1254 const struct timespec *mod, int flags) 1255 { 1256 1257 struct tmpfs_node *node; 1258 1259 KASSERT(VOP_ISLOCKED(vp)); 1260 1261 node = VP_TO_TMPFS_NODE(vp); 1262 1263 if (flags & UPDATE_CLOSE) 1264 ; /* XXX Need to do anything special? */ 1265 1266 tmpfs_itimes(vp, acc, mod); 1267 1268 KASSERT(VOP_ISLOCKED(vp)); 1269 } 1270 1271 /* --------------------------------------------------------------------- */ 1272 1273 int 1274 tmpfs_truncate(struct vnode *vp, off_t length) 1275 { 1276 boolean_t extended; 1277 int error; 1278 struct tmpfs_node *node; 1279 1280 node = VP_TO_TMPFS_NODE(vp); 1281 extended = length > node->tn_size; 1282 1283 if (length < 0) { 1284 error = EINVAL; 1285 goto out; 1286 } 1287 1288 if (node->tn_size == length) { 1289 error = 0; 1290 goto out; 1291 } 1292 1293 error = tmpfs_reg_resize(vp, length); 1294 if (error == 0) { 1295 VN_KNOTE(vp, NOTE_ATTRIB | (extended ? NOTE_EXTEND : 0)); 1296 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1297 } 1298 1299 out: 1300 tmpfs_update(vp, NULL, NULL, 0); 1301 1302 return error; 1303 } 1304