tmpfs_subr.c revision 312809
1/* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */ 2 3/*- 4 * Copyright (c) 2005 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 9 * 2005 program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33/* 34 * Efficient memory file system supporting functions. 35 */ 36#include <sys/cdefs.h> 37__FBSDID("$FreeBSD: stable/10/sys/fs/tmpfs/tmpfs_subr.c 312809 2017-01-26 11:04:27Z kib $"); 38 39#include <sys/param.h> 40#include <sys/fnv_hash.h> 41#include <sys/lock.h> 42#include <sys/namei.h> 43#include <sys/priv.h> 44#include <sys/proc.h> 45#include <sys/rwlock.h> 46#include <sys/stat.h> 47#include <sys/systm.h> 48#include <sys/sysctl.h> 49#include <sys/vnode.h> 50#include <sys/vmmeter.h> 51 52#include <vm/vm.h> 53#include <vm/vm_param.h> 54#include <vm/vm_object.h> 55#include <vm/vm_page.h> 56#include <vm/vm_pageout.h> 57#include <vm/vm_pager.h> 58#include <vm/vm_extern.h> 59 60#include <fs/tmpfs/tmpfs.h> 61#include <fs/tmpfs/tmpfs_fifoops.h> 62#include <fs/tmpfs/tmpfs_vnops.h> 63 64SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system"); 65 66static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED; 67 68static int 69sysctl_mem_reserved(SYSCTL_HANDLER_ARGS) 70{ 71 int error; 72 long pages, bytes; 73 74 pages = *(long *)arg1; 75 bytes = pages * PAGE_SIZE; 76 77 error = sysctl_handle_long(oidp, &bytes, 0, req); 78 if (error || !req->newptr) 79 return (error); 80 81 pages = bytes / PAGE_SIZE; 82 if (pages < TMPFS_PAGES_MINRESERVED) 83 return (EINVAL); 84 85 *(long *)arg1 = pages; 86 return (0); 87} 88 89SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW, 90 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L", 91 "Amount of available memory and swap below which tmpfs growth stops"); 92 93static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a, 94 struct tmpfs_dirent *b); 95RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 96 97size_t 98tmpfs_mem_avail(void) 99{ 100 vm_ooffset_t avail; 101 102 avail = swap_pager_avail + cnt.v_free_count + cnt.v_cache_count - 103 tmpfs_pages_reserved; 104 if (__predict_false(avail < 0)) 105 avail = 0; 106 return (avail); 107} 108 109size_t 110tmpfs_pages_used(struct tmpfs_mount *tmp) 111{ 112 const size_t node_size = sizeof(struct tmpfs_node) + 113 sizeof(struct tmpfs_dirent); 114 size_t meta_pages; 115 116 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size, 117 PAGE_SIZE); 118 return (meta_pages + tmp->tm_pages_used); 119} 120 121static size_t 122tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages) 123{ 124 if (tmpfs_mem_avail() < req_pages) 125 return (0); 126 127 if (tmp->tm_pages_max != ULONG_MAX && 128 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp)) 129 return (0); 130 131 return (1); 132} 133 134/* 135 * Allocates a new node of type 'type' inside the 'tmp' mount point, with 136 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode', 137 * using the credentials of the process 'p'. 138 * 139 * If the node type is set to 'VDIR', then the parent parameter must point 140 * to the parent directory of the node being created. It may only be NULL 141 * while allocating the root node. 142 * 143 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter 144 * specifies the device the node represents. 145 * 146 * If the node type is set to 'VLNK', then the parameter target specifies 147 * the file name of the target file for the symbolic link that is being 148 * created. 149 * 150 * Note that new nodes are retrieved from the available list if it has 151 * items or, if it is empty, from the node pool as long as there is enough 152 * space to create them. 153 * 154 * Returns zero on success or an appropriate error code on failure. 155 */ 156int 157tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type, 158 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent, 159 char *target, dev_t rdev, struct tmpfs_node **node) 160{ 161 struct tmpfs_node *nnode; 162 vm_object_t obj; 163 164 /* If the root directory of the 'tmp' file system is not yet 165 * allocated, this must be the request to do it. */ 166 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR)); 167 KASSERT(tmp->tm_root == NULL || mp->mnt_writeopcount > 0, 168 ("creating node not under vn_start_write")); 169 170 MPASS(IFF(type == VLNK, target != NULL)); 171 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL)); 172 173 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) 174 return (ENOSPC); 175 if (tmpfs_pages_check_avail(tmp, 1) == 0) 176 return (ENOSPC); 177 178 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { 179 /* 180 * When a new tmpfs node is created for fully 181 * constructed mount point, there must be a parent 182 * node, which vnode is locked exclusively. As 183 * consequence, if the unmount is executing in 184 * parallel, vflush() cannot reclaim the parent vnode. 185 * Due to this, the check for MNTK_UNMOUNT flag is not 186 * racy: if we did not see MNTK_UNMOUNT flag, then tmp 187 * cannot be destroyed until node construction is 188 * finished and the parent vnode unlocked. 189 * 190 * Tmpfs does not need to instantiate new nodes during 191 * unmount. 192 */ 193 return (EBUSY); 194 } 195 196 nnode = (struct tmpfs_node *)uma_zalloc_arg(tmp->tm_node_pool, tmp, 197 M_WAITOK); 198 199 /* Generic initialization. */ 200 nnode->tn_type = type; 201 vfs_timestamp(&nnode->tn_atime); 202 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime = 203 nnode->tn_atime; 204 nnode->tn_uid = uid; 205 nnode->tn_gid = gid; 206 nnode->tn_mode = mode; 207 nnode->tn_id = alloc_unr(tmp->tm_ino_unr); 208 209 /* Type-specific initialization. */ 210 switch (nnode->tn_type) { 211 case VBLK: 212 case VCHR: 213 nnode->tn_rdev = rdev; 214 break; 215 216 case VDIR: 217 RB_INIT(&nnode->tn_dir.tn_dirhead); 218 LIST_INIT(&nnode->tn_dir.tn_dupindex); 219 MPASS(parent != nnode); 220 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL)); 221 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent; 222 nnode->tn_dir.tn_readdir_lastn = 0; 223 nnode->tn_dir.tn_readdir_lastp = NULL; 224 nnode->tn_links++; 225 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent); 226 nnode->tn_dir.tn_parent->tn_links++; 227 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent); 228 break; 229 230 case VFIFO: 231 /* FALLTHROUGH */ 232 case VSOCK: 233 break; 234 235 case VLNK: 236 MPASS(strlen(target) < MAXPATHLEN); 237 nnode->tn_size = strlen(target); 238 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME, 239 M_WAITOK); 240 memcpy(nnode->tn_link, target, nnode->tn_size); 241 break; 242 243 case VREG: 244 obj = nnode->tn_reg.tn_aobj = 245 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0, 246 NULL /* XXXKIB - tmpfs needs swap reservation */); 247 VM_OBJECT_WLOCK(obj); 248 /* OBJ_TMPFS is set together with the setting of vp->v_object */ 249 vm_object_set_flag(obj, OBJ_NOSPLIT | OBJ_TMPFS_NODE); 250 vm_object_clear_flag(obj, OBJ_ONEMAPPING); 251 VM_OBJECT_WUNLOCK(obj); 252 break; 253 254 default: 255 panic("tmpfs_alloc_node: type %p %d", nnode, 256 (int)nnode->tn_type); 257 } 258 259 TMPFS_LOCK(tmp); 260 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries); 261 tmp->tm_nodes_inuse++; 262 TMPFS_UNLOCK(tmp); 263 264 *node = nnode; 265 return (0); 266} 267 268/* 269 * Destroys the node pointed to by node from the file system 'tmp'. 270 * If the node references a directory, no entries are allowed. 271 */ 272void 273tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node) 274{ 275 vm_object_t uobj; 276 277#ifdef INVARIANTS 278 TMPFS_NODE_LOCK(node); 279 MPASS(node->tn_vnode == NULL); 280 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0); 281 TMPFS_NODE_UNLOCK(node); 282#endif 283 284 TMPFS_LOCK(tmp); 285 LIST_REMOVE(node, tn_entries); 286 tmp->tm_nodes_inuse--; 287 TMPFS_UNLOCK(tmp); 288 289 switch (node->tn_type) { 290 case VNON: 291 /* Do not do anything. VNON is provided to let the 292 * allocation routine clean itself easily by avoiding 293 * duplicating code in it. */ 294 /* FALLTHROUGH */ 295 case VBLK: 296 /* FALLTHROUGH */ 297 case VCHR: 298 /* FALLTHROUGH */ 299 case VDIR: 300 /* FALLTHROUGH */ 301 case VFIFO: 302 /* FALLTHROUGH */ 303 case VSOCK: 304 break; 305 306 case VLNK: 307 free(node->tn_link, M_TMPFSNAME); 308 break; 309 310 case VREG: 311 uobj = node->tn_reg.tn_aobj; 312 if (uobj != NULL) { 313 atomic_subtract_long(&tmp->tm_pages_used, uobj->size); 314 KASSERT((uobj->flags & OBJ_TMPFS) == 0, 315 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj)); 316 vm_object_deallocate(uobj); 317 } 318 break; 319 320 default: 321 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type); 322 } 323 324 free_unr(tmp->tm_ino_unr, node->tn_id); 325 uma_zfree(tmp->tm_node_pool, node); 326} 327 328static __inline uint32_t 329tmpfs_dirent_hash(const char *name, u_int len) 330{ 331 uint32_t hash; 332 333 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK; 334#ifdef TMPFS_DEBUG_DIRCOOKIE_DUP 335 hash &= 0xf; 336#endif 337 if (hash < TMPFS_DIRCOOKIE_MIN) 338 hash += TMPFS_DIRCOOKIE_MIN; 339 340 return (hash); 341} 342 343static __inline off_t 344tmpfs_dirent_cookie(struct tmpfs_dirent *de) 345{ 346 if (de == NULL) 347 return (TMPFS_DIRCOOKIE_EOF); 348 349 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN); 350 351 return (de->td_cookie); 352} 353 354static __inline boolean_t 355tmpfs_dirent_dup(struct tmpfs_dirent *de) 356{ 357 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0); 358} 359 360static __inline boolean_t 361tmpfs_dirent_duphead(struct tmpfs_dirent *de) 362{ 363 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0); 364} 365 366void 367tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen) 368{ 369 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen); 370 memcpy(de->ud.td_name, name, namelen); 371 de->td_namelen = namelen; 372} 373 374/* 375 * Allocates a new directory entry for the node node with a name of name. 376 * The new directory entry is returned in *de. 377 * 378 * The link count of node is increased by one to reflect the new object 379 * referencing it. 380 * 381 * Returns zero on success or an appropriate error code on failure. 382 */ 383int 384tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node, 385 const char *name, u_int len, struct tmpfs_dirent **de) 386{ 387 struct tmpfs_dirent *nde; 388 389 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK); 390 nde->td_node = node; 391 if (name != NULL) { 392 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK); 393 tmpfs_dirent_init(nde, name, len); 394 } else 395 nde->td_namelen = 0; 396 if (node != NULL) 397 node->tn_links++; 398 399 *de = nde; 400 401 return 0; 402} 403 404/* 405 * Frees a directory entry. It is the caller's responsibility to destroy 406 * the node referenced by it if needed. 407 * 408 * The link count of node is decreased by one to reflect the removal of an 409 * object that referenced it. This only happens if 'node_exists' is true; 410 * otherwise the function will not access the node referred to by the 411 * directory entry, as it may already have been released from the outside. 412 */ 413void 414tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de) 415{ 416 struct tmpfs_node *node; 417 418 node = de->td_node; 419 if (node != NULL) { 420 MPASS(node->tn_links > 0); 421 node->tn_links--; 422 } 423 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL) 424 free(de->ud.td_name, M_TMPFSNAME); 425 uma_zfree(tmp->tm_dirent_pool, de); 426} 427 428void 429tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj) 430{ 431 432 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject"); 433 if (vp->v_type != VREG || obj == NULL) 434 return; 435 436 VM_OBJECT_WLOCK(obj); 437 VI_LOCK(vp); 438 vm_object_clear_flag(obj, OBJ_TMPFS); 439 obj->un_pager.swp.swp_tmpfs = NULL; 440 VI_UNLOCK(vp); 441 VM_OBJECT_WUNLOCK(obj); 442} 443 444/* 445 * Need to clear v_object for insmntque failure. 446 */ 447static void 448tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg) 449{ 450 451 tmpfs_destroy_vobject(vp, vp->v_object); 452 vp->v_object = NULL; 453 vp->v_data = NULL; 454 vp->v_op = &dead_vnodeops; 455 vgone(vp); 456 vput(vp); 457} 458 459/* 460 * Allocates a new vnode for the node node or returns a new reference to 461 * an existing one if the node had already a vnode referencing it. The 462 * resulting locked vnode is returned in *vpp. 463 * 464 * Returns zero on success or an appropriate error code on failure. 465 */ 466int 467tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag, 468 struct vnode **vpp) 469{ 470 struct vnode *vp; 471 vm_object_t object; 472 int error; 473 474 error = 0; 475loop: 476 TMPFS_NODE_LOCK(node); 477loop1: 478 if ((vp = node->tn_vnode) != NULL) { 479 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0); 480 VI_LOCK(vp); 481 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) || 482 ((vp->v_iflag & VI_DOOMED) != 0 && 483 (lkflag & LK_NOWAIT) != 0)) { 484 VI_UNLOCK(vp); 485 TMPFS_NODE_UNLOCK(node); 486 error = ENOENT; 487 vp = NULL; 488 goto out; 489 } 490 if ((vp->v_iflag & VI_DOOMED) != 0) { 491 VI_UNLOCK(vp); 492 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM; 493 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) { 494 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node), 495 0, "tmpfsE", 0); 496 } 497 goto loop1; 498 } 499 TMPFS_NODE_UNLOCK(node); 500 error = vget(vp, lkflag | LK_INTERLOCK, curthread); 501 if (error == ENOENT) 502 goto loop; 503 if (error != 0) { 504 vp = NULL; 505 goto out; 506 } 507 508 /* 509 * Make sure the vnode is still there after 510 * getting the interlock to avoid racing a free. 511 */ 512 if (node->tn_vnode == NULL || node->tn_vnode != vp) { 513 vput(vp); 514 goto loop; 515 } 516 517 goto out; 518 } 519 520 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) || 521 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) { 522 TMPFS_NODE_UNLOCK(node); 523 error = ENOENT; 524 vp = NULL; 525 goto out; 526 } 527 528 /* 529 * otherwise lock the vp list while we call getnewvnode 530 * since that can block. 531 */ 532 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) { 533 node->tn_vpstate |= TMPFS_VNODE_WANT; 534 error = msleep((caddr_t) &node->tn_vpstate, 535 TMPFS_NODE_MTX(node), PDROP | PCATCH, 536 "tmpfs_alloc_vp", 0); 537 if (error) 538 return error; 539 540 goto loop; 541 } else 542 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING; 543 544 TMPFS_NODE_UNLOCK(node); 545 546 /* Get a new vnode and associate it with our node. */ 547 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp); 548 if (error != 0) 549 goto unlock; 550 MPASS(vp != NULL); 551 552 /* lkflag is ignored, the lock is exclusive */ 553 (void) vn_lock(vp, lkflag | LK_RETRY); 554 555 vp->v_data = node; 556 vp->v_type = node->tn_type; 557 558 /* Type-specific initialization. */ 559 switch (node->tn_type) { 560 case VBLK: 561 /* FALLTHROUGH */ 562 case VCHR: 563 /* FALLTHROUGH */ 564 case VLNK: 565 /* FALLTHROUGH */ 566 case VSOCK: 567 break; 568 case VFIFO: 569 vp->v_op = &tmpfs_fifoop_entries; 570 break; 571 case VREG: 572 object = node->tn_reg.tn_aobj; 573 VM_OBJECT_WLOCK(object); 574 VI_LOCK(vp); 575 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs")); 576 vp->v_object = object; 577 object->un_pager.swp.swp_tmpfs = vp; 578 vm_object_set_flag(object, OBJ_TMPFS); 579 VI_UNLOCK(vp); 580 VM_OBJECT_WUNLOCK(object); 581 break; 582 case VDIR: 583 MPASS(node->tn_dir.tn_parent != NULL); 584 if (node->tn_dir.tn_parent == node) 585 vp->v_vflag |= VV_ROOT; 586 break; 587 588 default: 589 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type); 590 } 591 if (vp->v_type != VFIFO) 592 VN_LOCK_ASHARE(vp); 593 594 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL); 595 if (error != 0) 596 vp = NULL; 597 598unlock: 599 TMPFS_NODE_LOCK(node); 600 601 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING); 602 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING; 603 node->tn_vnode = vp; 604 605 if (node->tn_vpstate & TMPFS_VNODE_WANT) { 606 node->tn_vpstate &= ~TMPFS_VNODE_WANT; 607 TMPFS_NODE_UNLOCK(node); 608 wakeup((caddr_t) &node->tn_vpstate); 609 } else 610 TMPFS_NODE_UNLOCK(node); 611 612out: 613 *vpp = vp; 614 615#ifdef INVARIANTS 616 if (error == 0) { 617 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp)); 618 TMPFS_NODE_LOCK(node); 619 MPASS(*vpp == node->tn_vnode); 620 TMPFS_NODE_UNLOCK(node); 621 } 622#endif 623 624 return (error); 625} 626 627/* 628 * Destroys the association between the vnode vp and the node it 629 * references. 630 */ 631void 632tmpfs_free_vp(struct vnode *vp) 633{ 634 struct tmpfs_node *node; 635 636 node = VP_TO_TMPFS_NODE(vp); 637 638 TMPFS_NODE_ASSERT_LOCKED(node); 639 node->tn_vnode = NULL; 640 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) 641 wakeup(&node->tn_vnode); 642 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM; 643 vp->v_data = NULL; 644} 645 646/* 647 * Allocates a new file of type 'type' and adds it to the parent directory 648 * 'dvp'; this addition is done using the component name given in 'cnp'. 649 * The ownership of the new file is automatically assigned based on the 650 * credentials of the caller (through 'cnp'), the group is set based on 651 * the parent directory and the mode is determined from the 'vap' argument. 652 * If successful, *vpp holds a vnode to the newly created file and zero 653 * is returned. Otherwise *vpp is NULL and the function returns an 654 * appropriate error code. 655 */ 656int 657tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap, 658 struct componentname *cnp, char *target) 659{ 660 int error; 661 struct tmpfs_dirent *de; 662 struct tmpfs_mount *tmp; 663 struct tmpfs_node *dnode; 664 struct tmpfs_node *node; 665 struct tmpfs_node *parent; 666 667 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file"); 668 MPASS(cnp->cn_flags & HASBUF); 669 670 tmp = VFS_TO_TMPFS(dvp->v_mount); 671 dnode = VP_TO_TMPFS_DIR(dvp); 672 *vpp = NULL; 673 674 /* If the entry we are creating is a directory, we cannot overflow 675 * the number of links of its parent, because it will get a new 676 * link. */ 677 if (vap->va_type == VDIR) { 678 /* Ensure that we do not overflow the maximum number of links 679 * imposed by the system. */ 680 MPASS(dnode->tn_links <= LINK_MAX); 681 if (dnode->tn_links == LINK_MAX) { 682 return (EMLINK); 683 } 684 685 parent = dnode; 686 MPASS(parent != NULL); 687 } else 688 parent = NULL; 689 690 /* Allocate a node that represents the new file. */ 691 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type, 692 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent, 693 target, vap->va_rdev, &node); 694 if (error != 0) 695 return (error); 696 697 /* Allocate a directory entry that points to the new file. */ 698 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen, 699 &de); 700 if (error != 0) { 701 tmpfs_free_node(tmp, node); 702 return (error); 703 } 704 705 /* Allocate a vnode for the new file. */ 706 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp); 707 if (error != 0) { 708 tmpfs_free_dirent(tmp, de); 709 tmpfs_free_node(tmp, node); 710 return (error); 711 } 712 713 /* Now that all required items are allocated, we can proceed to 714 * insert the new node into the directory, an operation that 715 * cannot fail. */ 716 if (cnp->cn_flags & ISWHITEOUT) 717 tmpfs_dir_whiteout_remove(dvp, cnp); 718 tmpfs_dir_attach(dvp, de); 719 return (0); 720} 721 722struct tmpfs_dirent * 723tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 724{ 725 struct tmpfs_dirent *de; 726 727 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead); 728 dc->tdc_tree = de; 729 if (de != NULL && tmpfs_dirent_duphead(de)) 730 de = LIST_FIRST(&de->ud.td_duphead); 731 dc->tdc_current = de; 732 733 return (dc->tdc_current); 734} 735 736struct tmpfs_dirent * 737tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc) 738{ 739 struct tmpfs_dirent *de; 740 741 MPASS(dc->tdc_tree != NULL); 742 if (tmpfs_dirent_dup(dc->tdc_current)) { 743 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries); 744 if (dc->tdc_current != NULL) 745 return (dc->tdc_current); 746 } 747 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir, 748 &dnode->tn_dir.tn_dirhead, dc->tdc_tree); 749 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) { 750 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 751 MPASS(dc->tdc_current != NULL); 752 } 753 754 return (dc->tdc_current); 755} 756 757/* Lookup directory entry in RB-Tree. Function may return duphead entry. */ 758static struct tmpfs_dirent * 759tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash) 760{ 761 struct tmpfs_dirent *de, dekey; 762 763 dekey.td_hash = hash; 764 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey); 765 return (de); 766} 767 768/* Lookup directory entry by cookie, initialize directory cursor accordingly. */ 769static struct tmpfs_dirent * 770tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie, 771 struct tmpfs_dir_cursor *dc) 772{ 773 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead; 774 struct tmpfs_dirent *de, dekey; 775 776 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN); 777 778 if (cookie == node->tn_dir.tn_readdir_lastn && 779 (de = node->tn_dir.tn_readdir_lastp) != NULL) { 780 /* Protect against possible race, tn_readdir_last[pn] 781 * may be updated with only shared vnode lock held. */ 782 if (cookie == tmpfs_dirent_cookie(de)) 783 goto out; 784 } 785 786 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) { 787 LIST_FOREACH(de, &node->tn_dir.tn_dupindex, 788 uh.td_dup.index_entries) { 789 MPASS(tmpfs_dirent_dup(de)); 790 if (de->td_cookie == cookie) 791 goto out; 792 /* dupindex list is sorted. */ 793 if (de->td_cookie < cookie) { 794 de = NULL; 795 goto out; 796 } 797 } 798 MPASS(de == NULL); 799 goto out; 800 } 801 802 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) { 803 de = NULL; 804 } else { 805 dekey.td_hash = cookie; 806 /* Recover if direntry for cookie was removed */ 807 de = RB_NFIND(tmpfs_dir, dirhead, &dekey); 808 } 809 dc->tdc_tree = de; 810 dc->tdc_current = de; 811 if (de != NULL && tmpfs_dirent_duphead(de)) { 812 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead); 813 MPASS(dc->tdc_current != NULL); 814 } 815 return (dc->tdc_current); 816 817out: 818 dc->tdc_tree = de; 819 dc->tdc_current = de; 820 if (de != NULL && tmpfs_dirent_dup(de)) 821 dc->tdc_tree = tmpfs_dir_xlookup_hash(node, 822 de->td_hash); 823 return (dc->tdc_current); 824} 825 826/* 827 * Looks for a directory entry in the directory represented by node. 828 * 'cnp' describes the name of the entry to look for. Note that the . 829 * and .. components are not allowed as they do not physically exist 830 * within directories. 831 * 832 * Returns a pointer to the entry when found, otherwise NULL. 833 */ 834struct tmpfs_dirent * 835tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f, 836 struct componentname *cnp) 837{ 838 struct tmpfs_dir_duphead *duphead; 839 struct tmpfs_dirent *de; 840 uint32_t hash; 841 842 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.')); 843 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' && 844 cnp->cn_nameptr[1] == '.'))); 845 TMPFS_VALIDATE_DIR(node); 846 847 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen); 848 de = tmpfs_dir_xlookup_hash(node, hash); 849 if (de != NULL && tmpfs_dirent_duphead(de)) { 850 duphead = &de->ud.td_duphead; 851 LIST_FOREACH(de, duphead, uh.td_dup.entries) { 852 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 853 cnp->cn_namelen)) 854 break; 855 } 856 } else if (de != NULL) { 857 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr, 858 cnp->cn_namelen)) 859 de = NULL; 860 } 861 if (de != NULL && f != NULL && de->td_node != f) 862 de = NULL; 863 864 return (de); 865} 866 867/* 868 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex 869 * list, allocate new cookie value. 870 */ 871static void 872tmpfs_dir_attach_dup(struct tmpfs_node *dnode, 873 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde) 874{ 875 struct tmpfs_dir_duphead *dupindex; 876 struct tmpfs_dirent *de, *pde; 877 878 dupindex = &dnode->tn_dir.tn_dupindex; 879 de = LIST_FIRST(dupindex); 880 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) { 881 if (de == NULL) 882 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 883 else 884 nde->td_cookie = de->td_cookie + 1; 885 MPASS(tmpfs_dirent_dup(nde)); 886 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries); 887 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 888 return; 889 } 890 891 /* 892 * Cookie numbers are near exhaustion. Scan dupindex list for unused 893 * numbers. dupindex list is sorted in descending order. Keep it so 894 * after inserting nde. 895 */ 896 while (1) { 897 pde = de; 898 de = LIST_NEXT(de, uh.td_dup.index_entries); 899 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) { 900 /* 901 * Last element of the index doesn't have minimal cookie 902 * value, use it. 903 */ 904 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN; 905 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries); 906 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 907 return; 908 } else if (de == NULL) { 909 /* 910 * We are so lucky have 2^30 hash duplicates in single 911 * directory :) Return largest possible cookie value. 912 * It should be fine except possible issues with 913 * VOP_READDIR restart. 914 */ 915 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX; 916 LIST_INSERT_HEAD(dupindex, nde, 917 uh.td_dup.index_entries); 918 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 919 return; 920 } 921 if (de->td_cookie + 1 == pde->td_cookie || 922 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX) 923 continue; /* No hole or invalid cookie. */ 924 nde->td_cookie = de->td_cookie + 1; 925 MPASS(tmpfs_dirent_dup(nde)); 926 MPASS(pde->td_cookie > nde->td_cookie); 927 MPASS(nde->td_cookie > de->td_cookie); 928 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries); 929 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries); 930 return; 931 }; 932} 933 934/* 935 * Attaches the directory entry de to the directory represented by vp. 936 * Note that this does not change the link count of the node pointed by 937 * the directory entry, as this is done by tmpfs_alloc_dirent. 938 */ 939void 940tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de) 941{ 942 struct tmpfs_node *dnode; 943 struct tmpfs_dirent *xde, *nde; 944 945 ASSERT_VOP_ELOCKED(vp, __func__); 946 MPASS(de->td_namelen > 0); 947 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN); 948 MPASS(de->td_cookie == de->td_hash); 949 950 dnode = VP_TO_TMPFS_DIR(vp); 951 dnode->tn_dir.tn_readdir_lastn = 0; 952 dnode->tn_dir.tn_readdir_lastp = NULL; 953 954 MPASS(!tmpfs_dirent_dup(de)); 955 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 956 if (xde != NULL && tmpfs_dirent_duphead(xde)) 957 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 958 else if (xde != NULL) { 959 /* 960 * Allocate new duphead. Swap xde with duphead to avoid 961 * adding/removing elements with the same hash. 962 */ 963 MPASS(!tmpfs_dirent_dup(xde)); 964 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0, 965 &nde); 966 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */ 967 memcpy(nde, xde, sizeof(*xde)); 968 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD; 969 LIST_INIT(&xde->ud.td_duphead); 970 xde->td_namelen = 0; 971 xde->td_node = NULL; 972 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde); 973 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de); 974 } 975 dnode->tn_size += sizeof(struct tmpfs_dirent); 976 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 977 TMPFS_NODE_MODIFIED; 978 tmpfs_update(vp); 979} 980 981/* 982 * Detaches the directory entry de from the directory represented by vp. 983 * Note that this does not change the link count of the node pointed by 984 * the directory entry, as this is done by tmpfs_free_dirent. 985 */ 986void 987tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de) 988{ 989 struct tmpfs_mount *tmp; 990 struct tmpfs_dir *head; 991 struct tmpfs_node *dnode; 992 struct tmpfs_dirent *xde; 993 994 ASSERT_VOP_ELOCKED(vp, __func__); 995 996 dnode = VP_TO_TMPFS_DIR(vp); 997 head = &dnode->tn_dir.tn_dirhead; 998 dnode->tn_dir.tn_readdir_lastn = 0; 999 dnode->tn_dir.tn_readdir_lastp = NULL; 1000 1001 if (tmpfs_dirent_dup(de)) { 1002 /* Remove duphead if de was last entry. */ 1003 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) { 1004 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash); 1005 MPASS(tmpfs_dirent_duphead(xde)); 1006 } else 1007 xde = NULL; 1008 LIST_REMOVE(de, uh.td_dup.entries); 1009 LIST_REMOVE(de, uh.td_dup.index_entries); 1010 if (xde != NULL) { 1011 if (LIST_EMPTY(&xde->ud.td_duphead)) { 1012 RB_REMOVE(tmpfs_dir, head, xde); 1013 tmp = VFS_TO_TMPFS(vp->v_mount); 1014 MPASS(xde->td_node == NULL); 1015 tmpfs_free_dirent(tmp, xde); 1016 } 1017 } 1018 de->td_cookie = de->td_hash; 1019 } else 1020 RB_REMOVE(tmpfs_dir, head, de); 1021 1022 dnode->tn_size -= sizeof(struct tmpfs_dirent); 1023 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1024 TMPFS_NODE_MODIFIED; 1025 tmpfs_update(vp); 1026} 1027 1028void 1029tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode) 1030{ 1031 struct tmpfs_dirent *de, *dde, *nde; 1032 1033 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) { 1034 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de); 1035 /* Node may already be destroyed. */ 1036 de->td_node = NULL; 1037 if (tmpfs_dirent_duphead(de)) { 1038 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) { 1039 LIST_REMOVE(dde, uh.td_dup.entries); 1040 dde->td_node = NULL; 1041 tmpfs_free_dirent(tmp, dde); 1042 } 1043 } 1044 tmpfs_free_dirent(tmp, de); 1045 } 1046} 1047 1048/* 1049 * Helper function for tmpfs_readdir. Creates a '.' entry for the given 1050 * directory and returns it in the uio space. The function returns 0 1051 * on success, -1 if there was not enough space in the uio structure to 1052 * hold the directory entry or an appropriate error code if another 1053 * error happens. 1054 */ 1055static int 1056tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio) 1057{ 1058 int error; 1059 struct dirent dent; 1060 1061 TMPFS_VALIDATE_DIR(node); 1062 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT); 1063 1064 dent.d_fileno = node->tn_id; 1065 dent.d_type = DT_DIR; 1066 dent.d_namlen = 1; 1067 dent.d_name[0] = '.'; 1068 dent.d_name[1] = '\0'; 1069 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1070 1071 if (dent.d_reclen > uio->uio_resid) 1072 error = EJUSTRETURN; 1073 else 1074 error = uiomove(&dent, dent.d_reclen, uio); 1075 1076 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1077 1078 return (error); 1079} 1080 1081/* 1082 * Helper function for tmpfs_readdir. Creates a '..' entry for the given 1083 * directory and returns it in the uio space. The function returns 0 1084 * on success, -1 if there was not enough space in the uio structure to 1085 * hold the directory entry or an appropriate error code if another 1086 * error happens. 1087 */ 1088static int 1089tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio) 1090{ 1091 int error; 1092 struct dirent dent; 1093 1094 TMPFS_VALIDATE_DIR(node); 1095 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT); 1096 1097 /* 1098 * Return ENOENT if the current node is already removed. 1099 */ 1100 TMPFS_ASSERT_LOCKED(node); 1101 if (node->tn_dir.tn_parent == NULL) 1102 return (ENOENT); 1103 1104 TMPFS_NODE_LOCK(node->tn_dir.tn_parent); 1105 dent.d_fileno = node->tn_dir.tn_parent->tn_id; 1106 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent); 1107 1108 dent.d_type = DT_DIR; 1109 dent.d_namlen = 2; 1110 dent.d_name[0] = '.'; 1111 dent.d_name[1] = '.'; 1112 dent.d_name[2] = '\0'; 1113 dent.d_reclen = GENERIC_DIRSIZ(&dent); 1114 1115 if (dent.d_reclen > uio->uio_resid) 1116 error = EJUSTRETURN; 1117 else 1118 error = uiomove(&dent, dent.d_reclen, uio); 1119 1120 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1121 1122 return (error); 1123} 1124 1125/* 1126 * Helper function for tmpfs_readdir. Returns as much directory entries 1127 * as can fit in the uio space. The read starts at uio->uio_offset. 1128 * The function returns 0 on success, -1 if there was not enough space 1129 * in the uio structure to hold the directory entry or an appropriate 1130 * error code if another error happens. 1131 */ 1132int 1133tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies, 1134 u_long *cookies, int *ncookies) 1135{ 1136 struct tmpfs_dir_cursor dc; 1137 struct tmpfs_dirent *de; 1138 off_t off; 1139 int error; 1140 1141 TMPFS_VALIDATE_DIR(node); 1142 1143 off = 0; 1144 1145 /* 1146 * Lookup the node from the current offset. The starting offset of 1147 * 0 will lookup both '.' and '..', and then the first real entry, 1148 * or EOF if there are none. Then find all entries for the dir that 1149 * fit into the buffer. Once no more entries are found (de == NULL), 1150 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next 1151 * call to return 0. 1152 */ 1153 switch (uio->uio_offset) { 1154 case TMPFS_DIRCOOKIE_DOT: 1155 error = tmpfs_dir_getdotdent(node, uio); 1156 if (error != 0) 1157 return (error); 1158 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT; 1159 if (cookies != NULL) 1160 cookies[(*ncookies)++] = off = uio->uio_offset; 1161 /* FALLTHROUGH */ 1162 case TMPFS_DIRCOOKIE_DOTDOT: 1163 error = tmpfs_dir_getdotdotdent(node, uio); 1164 if (error != 0) 1165 return (error); 1166 de = tmpfs_dir_first(node, &dc); 1167 uio->uio_offset = tmpfs_dirent_cookie(de); 1168 if (cookies != NULL) 1169 cookies[(*ncookies)++] = off = uio->uio_offset; 1170 /* EOF. */ 1171 if (de == NULL) 1172 return (0); 1173 break; 1174 case TMPFS_DIRCOOKIE_EOF: 1175 return (0); 1176 default: 1177 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc); 1178 if (de == NULL) 1179 return (EINVAL); 1180 if (cookies != NULL) 1181 off = tmpfs_dirent_cookie(de); 1182 } 1183 1184 /* Read as much entries as possible; i.e., until we reach the end of 1185 * the directory or we exhaust uio space. */ 1186 do { 1187 struct dirent d; 1188 1189 /* Create a dirent structure representing the current 1190 * tmpfs_node and fill it. */ 1191 if (de->td_node == NULL) { 1192 d.d_fileno = 1; 1193 d.d_type = DT_WHT; 1194 } else { 1195 d.d_fileno = de->td_node->tn_id; 1196 switch (de->td_node->tn_type) { 1197 case VBLK: 1198 d.d_type = DT_BLK; 1199 break; 1200 1201 case VCHR: 1202 d.d_type = DT_CHR; 1203 break; 1204 1205 case VDIR: 1206 d.d_type = DT_DIR; 1207 break; 1208 1209 case VFIFO: 1210 d.d_type = DT_FIFO; 1211 break; 1212 1213 case VLNK: 1214 d.d_type = DT_LNK; 1215 break; 1216 1217 case VREG: 1218 d.d_type = DT_REG; 1219 break; 1220 1221 case VSOCK: 1222 d.d_type = DT_SOCK; 1223 break; 1224 1225 default: 1226 panic("tmpfs_dir_getdents: type %p %d", 1227 de->td_node, (int)de->td_node->tn_type); 1228 } 1229 } 1230 d.d_namlen = de->td_namelen; 1231 MPASS(de->td_namelen < sizeof(d.d_name)); 1232 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen); 1233 d.d_name[de->td_namelen] = '\0'; 1234 d.d_reclen = GENERIC_DIRSIZ(&d); 1235 1236 /* Stop reading if the directory entry we are treating is 1237 * bigger than the amount of data that can be returned. */ 1238 if (d.d_reclen > uio->uio_resid) { 1239 error = EJUSTRETURN; 1240 break; 1241 } 1242 1243 /* Copy the new dirent structure into the output buffer and 1244 * advance pointers. */ 1245 error = uiomove(&d, d.d_reclen, uio); 1246 if (error == 0) { 1247 de = tmpfs_dir_next(node, &dc); 1248 if (cookies != NULL) { 1249 off = tmpfs_dirent_cookie(de); 1250 MPASS(*ncookies < maxcookies); 1251 cookies[(*ncookies)++] = off; 1252 } 1253 } 1254 } while (error == 0 && uio->uio_resid > 0 && de != NULL); 1255 1256 /* Skip setting off when using cookies as it is already done above. */ 1257 if (cookies == NULL) 1258 off = tmpfs_dirent_cookie(de); 1259 1260 /* Update the offset and cache. */ 1261 uio->uio_offset = off; 1262 node->tn_dir.tn_readdir_lastn = off; 1263 node->tn_dir.tn_readdir_lastp = de; 1264 1265 tmpfs_set_status(node, TMPFS_NODE_ACCESSED); 1266 return error; 1267} 1268 1269int 1270tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp) 1271{ 1272 struct tmpfs_dirent *de; 1273 int error; 1274 1275 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL, 1276 cnp->cn_nameptr, cnp->cn_namelen, &de); 1277 if (error != 0) 1278 return (error); 1279 tmpfs_dir_attach(dvp, de); 1280 return (0); 1281} 1282 1283void 1284tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp) 1285{ 1286 struct tmpfs_dirent *de; 1287 1288 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1289 MPASS(de != NULL && de->td_node == NULL); 1290 tmpfs_dir_detach(dvp, de); 1291 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de); 1292} 1293 1294/* 1295 * Resizes the aobj associated with the regular file pointed to by 'vp' to the 1296 * size 'newsize'. 'vp' must point to a vnode that represents a regular file. 1297 * 'newsize' must be positive. 1298 * 1299 * Returns zero on success or an appropriate error code on failure. 1300 */ 1301int 1302tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr) 1303{ 1304 struct tmpfs_mount *tmp; 1305 struct tmpfs_node *node; 1306 vm_object_t uobj; 1307 vm_page_t m, ma[1]; 1308 vm_pindex_t idx, newpages, oldpages; 1309 off_t oldsize; 1310 int base, rv; 1311 1312 MPASS(vp->v_type == VREG); 1313 MPASS(newsize >= 0); 1314 1315 node = VP_TO_TMPFS_NODE(vp); 1316 uobj = node->tn_reg.tn_aobj; 1317 tmp = VFS_TO_TMPFS(vp->v_mount); 1318 1319 /* 1320 * Convert the old and new sizes to the number of pages needed to 1321 * store them. It may happen that we do not need to do anything 1322 * because the last allocated page can accommodate the change on 1323 * its own. 1324 */ 1325 oldsize = node->tn_size; 1326 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK); 1327 MPASS(oldpages == uobj->size); 1328 newpages = OFF_TO_IDX(newsize + PAGE_MASK); 1329 if (newpages > oldpages && 1330 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0) 1331 return (ENOSPC); 1332 1333 VM_OBJECT_WLOCK(uobj); 1334 if (newsize < oldsize) { 1335 /* 1336 * Zero the truncated part of the last page. 1337 */ 1338 base = newsize & PAGE_MASK; 1339 if (base != 0) { 1340 idx = OFF_TO_IDX(newsize); 1341retry: 1342 m = vm_page_lookup(uobj, idx); 1343 if (m != NULL) { 1344 if (vm_page_sleep_if_busy(m, "tmfssz")) 1345 goto retry; 1346 MPASS(m->valid == VM_PAGE_BITS_ALL); 1347 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) { 1348 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL); 1349 if (m == NULL) { 1350 VM_OBJECT_WUNLOCK(uobj); 1351 VM_WAIT; 1352 VM_OBJECT_WLOCK(uobj); 1353 goto retry; 1354 } else if (m->valid != VM_PAGE_BITS_ALL) { 1355 ma[0] = m; 1356 rv = vm_pager_get_pages(uobj, ma, 1, 0); 1357 m = vm_page_lookup(uobj, idx); 1358 } else 1359 /* A cached page was reactivated. */ 1360 rv = VM_PAGER_OK; 1361 vm_page_lock(m); 1362 if (rv == VM_PAGER_OK) { 1363 vm_page_deactivate(m); 1364 vm_page_unlock(m); 1365 vm_page_xunbusy(m); 1366 } else { 1367 vm_page_free(m); 1368 vm_page_unlock(m); 1369 if (ignerr) 1370 m = NULL; 1371 else { 1372 VM_OBJECT_WUNLOCK(uobj); 1373 return (EIO); 1374 } 1375 } 1376 } 1377 if (m != NULL) { 1378 pmap_zero_page_area(m, base, PAGE_SIZE - base); 1379 vm_page_dirty(m); 1380 vm_pager_page_unswapped(m); 1381 } 1382 } 1383 1384 /* 1385 * Release any swap space and free any whole pages. 1386 */ 1387 if (newpages < oldpages) { 1388 swap_pager_freespace(uobj, newpages, oldpages - 1389 newpages); 1390 vm_object_page_remove(uobj, newpages, 0, 0); 1391 } 1392 } 1393 uobj->size = newpages; 1394 VM_OBJECT_WUNLOCK(uobj); 1395 1396 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages); 1397 1398 node->tn_size = newsize; 1399 return (0); 1400} 1401 1402void 1403tmpfs_check_mtime(struct vnode *vp) 1404{ 1405 struct tmpfs_node *node; 1406 struct vm_object *obj; 1407 1408 ASSERT_VOP_ELOCKED(vp, "check_mtime"); 1409 if (vp->v_type != VREG) 1410 return; 1411 obj = vp->v_object; 1412 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) == 1413 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj")); 1414 /* unlocked read */ 1415 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) { 1416 VM_OBJECT_WLOCK(obj); 1417 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) { 1418 obj->flags &= ~OBJ_TMPFS_DIRTY; 1419 node = VP_TO_TMPFS_NODE(vp); 1420 node->tn_status |= TMPFS_NODE_MODIFIED | 1421 TMPFS_NODE_CHANGED; 1422 } 1423 VM_OBJECT_WUNLOCK(obj); 1424 } 1425} 1426 1427/* 1428 * Change flags of the given vnode. 1429 * Caller should execute tmpfs_update on vp after a successful execution. 1430 * The vnode must be locked on entry and remain locked on exit. 1431 */ 1432int 1433tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred, 1434 struct thread *p) 1435{ 1436 int error; 1437 struct tmpfs_node *node; 1438 1439 ASSERT_VOP_ELOCKED(vp, "chflags"); 1440 1441 node = VP_TO_TMPFS_NODE(vp); 1442 1443 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK | 1444 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP | 1445 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE | 1446 UF_SPARSE | UF_SYSTEM)) != 0) 1447 return (EOPNOTSUPP); 1448 1449 /* Disallow this operation if the file system is mounted read-only. */ 1450 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1451 return EROFS; 1452 1453 /* 1454 * Callers may only modify the file flags on objects they 1455 * have VADMIN rights for. 1456 */ 1457 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1458 return (error); 1459 /* 1460 * Unprivileged processes are not permitted to unset system 1461 * flags, or modify flags if any system flags are set. 1462 */ 1463 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { 1464 if (node->tn_flags & 1465 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) { 1466 error = securelevel_gt(cred, 0); 1467 if (error) 1468 return (error); 1469 } 1470 } else { 1471 if (node->tn_flags & 1472 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) || 1473 ((flags ^ node->tn_flags) & SF_SETTABLE)) 1474 return (EPERM); 1475 } 1476 node->tn_flags = flags; 1477 node->tn_status |= TMPFS_NODE_CHANGED; 1478 1479 ASSERT_VOP_ELOCKED(vp, "chflags2"); 1480 1481 return (0); 1482} 1483 1484/* 1485 * Change access mode on the given vnode. 1486 * Caller should execute tmpfs_update on vp after a successful execution. 1487 * The vnode must be locked on entry and remain locked on exit. 1488 */ 1489int 1490tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p) 1491{ 1492 int error; 1493 struct tmpfs_node *node; 1494 1495 ASSERT_VOP_ELOCKED(vp, "chmod"); 1496 1497 node = VP_TO_TMPFS_NODE(vp); 1498 1499 /* Disallow this operation if the file system is mounted read-only. */ 1500 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1501 return EROFS; 1502 1503 /* Immutable or append-only files cannot be modified, either. */ 1504 if (node->tn_flags & (IMMUTABLE | APPEND)) 1505 return EPERM; 1506 1507 /* 1508 * To modify the permissions on a file, must possess VADMIN 1509 * for that file. 1510 */ 1511 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1512 return (error); 1513 1514 /* 1515 * Privileged processes may set the sticky bit on non-directories, 1516 * as well as set the setgid bit on a file with a group that the 1517 * process is not a member of. 1518 */ 1519 if (vp->v_type != VDIR && (mode & S_ISTXT)) { 1520 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0)) 1521 return (EFTYPE); 1522 } 1523 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) { 1524 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); 1525 if (error) 1526 return (error); 1527 } 1528 1529 1530 node->tn_mode &= ~ALLPERMS; 1531 node->tn_mode |= mode & ALLPERMS; 1532 1533 node->tn_status |= TMPFS_NODE_CHANGED; 1534 1535 ASSERT_VOP_ELOCKED(vp, "chmod2"); 1536 1537 return (0); 1538} 1539 1540/* 1541 * Change ownership of the given vnode. At least one of uid or gid must 1542 * be different than VNOVAL. If one is set to that value, the attribute 1543 * is unchanged. 1544 * Caller should execute tmpfs_update on vp after a successful execution. 1545 * The vnode must be locked on entry and remain locked on exit. 1546 */ 1547int 1548tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, 1549 struct thread *p) 1550{ 1551 int error; 1552 struct tmpfs_node *node; 1553 uid_t ouid; 1554 gid_t ogid; 1555 1556 ASSERT_VOP_ELOCKED(vp, "chown"); 1557 1558 node = VP_TO_TMPFS_NODE(vp); 1559 1560 /* Assign default values if they are unknown. */ 1561 MPASS(uid != VNOVAL || gid != VNOVAL); 1562 if (uid == VNOVAL) 1563 uid = node->tn_uid; 1564 if (gid == VNOVAL) 1565 gid = node->tn_gid; 1566 MPASS(uid != VNOVAL && gid != VNOVAL); 1567 1568 /* Disallow this operation if the file system is mounted read-only. */ 1569 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1570 return EROFS; 1571 1572 /* Immutable or append-only files cannot be modified, either. */ 1573 if (node->tn_flags & (IMMUTABLE | APPEND)) 1574 return EPERM; 1575 1576 /* 1577 * To modify the ownership of a file, must possess VADMIN for that 1578 * file. 1579 */ 1580 if ((error = VOP_ACCESS(vp, VADMIN, cred, p))) 1581 return (error); 1582 1583 /* 1584 * To change the owner of a file, or change the group of a file to a 1585 * group of which we are not a member, the caller must have 1586 * privilege. 1587 */ 1588 if ((uid != node->tn_uid || 1589 (gid != node->tn_gid && !groupmember(gid, cred))) && 1590 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0))) 1591 return (error); 1592 1593 ogid = node->tn_gid; 1594 ouid = node->tn_uid; 1595 1596 node->tn_uid = uid; 1597 node->tn_gid = gid; 1598 1599 node->tn_status |= TMPFS_NODE_CHANGED; 1600 1601 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) { 1602 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0)) 1603 node->tn_mode &= ~(S_ISUID | S_ISGID); 1604 } 1605 1606 ASSERT_VOP_ELOCKED(vp, "chown2"); 1607 1608 return (0); 1609} 1610 1611/* 1612 * Change size of the given vnode. 1613 * Caller should execute tmpfs_update on vp after a successful execution. 1614 * The vnode must be locked on entry and remain locked on exit. 1615 */ 1616int 1617tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred, 1618 struct thread *p) 1619{ 1620 int error; 1621 struct tmpfs_node *node; 1622 1623 ASSERT_VOP_ELOCKED(vp, "chsize"); 1624 1625 node = VP_TO_TMPFS_NODE(vp); 1626 1627 /* Decide whether this is a valid operation based on the file type. */ 1628 error = 0; 1629 switch (vp->v_type) { 1630 case VDIR: 1631 return EISDIR; 1632 1633 case VREG: 1634 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1635 return EROFS; 1636 break; 1637 1638 case VBLK: 1639 /* FALLTHROUGH */ 1640 case VCHR: 1641 /* FALLTHROUGH */ 1642 case VFIFO: 1643 /* Allow modifications of special files even if in the file 1644 * system is mounted read-only (we are not modifying the 1645 * files themselves, but the objects they represent). */ 1646 return 0; 1647 1648 default: 1649 /* Anything else is unsupported. */ 1650 return EOPNOTSUPP; 1651 } 1652 1653 /* Immutable or append-only files cannot be modified, either. */ 1654 if (node->tn_flags & (IMMUTABLE | APPEND)) 1655 return EPERM; 1656 1657 error = tmpfs_truncate(vp, size); 1658 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents 1659 * for us, as will update tn_status; no need to do that here. */ 1660 1661 ASSERT_VOP_ELOCKED(vp, "chsize2"); 1662 1663 return (error); 1664} 1665 1666/* 1667 * Change access and modification times of the given vnode. 1668 * Caller should execute tmpfs_update on vp after a successful execution. 1669 * The vnode must be locked on entry and remain locked on exit. 1670 */ 1671int 1672tmpfs_chtimes(struct vnode *vp, struct vattr *vap, 1673 struct ucred *cred, struct thread *l) 1674{ 1675 int error; 1676 struct tmpfs_node *node; 1677 1678 ASSERT_VOP_ELOCKED(vp, "chtimes"); 1679 1680 node = VP_TO_TMPFS_NODE(vp); 1681 1682 /* Disallow this operation if the file system is mounted read-only. */ 1683 if (vp->v_mount->mnt_flag & MNT_RDONLY) 1684 return EROFS; 1685 1686 /* Immutable or append-only files cannot be modified, either. */ 1687 if (node->tn_flags & (IMMUTABLE | APPEND)) 1688 return EPERM; 1689 1690 error = vn_utimes_perm(vp, vap, cred, l); 1691 if (error != 0) 1692 return (error); 1693 1694 if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL) 1695 node->tn_status |= TMPFS_NODE_ACCESSED; 1696 1697 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL) 1698 node->tn_status |= TMPFS_NODE_MODIFIED; 1699 1700 if (vap->va_birthtime.tv_nsec != VNOVAL && 1701 vap->va_birthtime.tv_nsec != VNOVAL) 1702 node->tn_status |= TMPFS_NODE_MODIFIED; 1703 1704 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime); 1705 1706 if (vap->va_birthtime.tv_nsec != VNOVAL && 1707 vap->va_birthtime.tv_nsec != VNOVAL) 1708 node->tn_birthtime = vap->va_birthtime; 1709 ASSERT_VOP_ELOCKED(vp, "chtimes2"); 1710 1711 return (0); 1712} 1713 1714void 1715tmpfs_set_status(struct tmpfs_node *node, int status) 1716{ 1717 1718 if ((node->tn_status & status) == status) 1719 return; 1720 TMPFS_NODE_LOCK(node); 1721 node->tn_status |= status; 1722 TMPFS_NODE_UNLOCK(node); 1723} 1724 1725/* Sync timestamps */ 1726void 1727tmpfs_itimes(struct vnode *vp, const struct timespec *acc, 1728 const struct timespec *mod) 1729{ 1730 struct tmpfs_node *node; 1731 struct timespec now; 1732 1733 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes"); 1734 node = VP_TO_TMPFS_NODE(vp); 1735 1736 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1737 TMPFS_NODE_CHANGED)) == 0) 1738 return; 1739 1740 vfs_timestamp(&now); 1741 TMPFS_NODE_LOCK(node); 1742 if (node->tn_status & TMPFS_NODE_ACCESSED) { 1743 if (acc == NULL) 1744 acc = &now; 1745 node->tn_atime = *acc; 1746 } 1747 if (node->tn_status & TMPFS_NODE_MODIFIED) { 1748 if (mod == NULL) 1749 mod = &now; 1750 node->tn_mtime = *mod; 1751 } 1752 if (node->tn_status & TMPFS_NODE_CHANGED) 1753 node->tn_ctime = now; 1754 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 1755 TMPFS_NODE_CHANGED); 1756 TMPFS_NODE_UNLOCK(node); 1757 1758} 1759 1760void 1761tmpfs_update(struct vnode *vp) 1762{ 1763 1764 tmpfs_itimes(vp, NULL, NULL); 1765} 1766 1767int 1768tmpfs_truncate(struct vnode *vp, off_t length) 1769{ 1770 int error; 1771 struct tmpfs_node *node; 1772 1773 node = VP_TO_TMPFS_NODE(vp); 1774 1775 if (length < 0) { 1776 error = EINVAL; 1777 goto out; 1778 } 1779 1780 if (node->tn_size == length) { 1781 error = 0; 1782 goto out; 1783 } 1784 1785 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 1786 return (EFBIG); 1787 1788 error = tmpfs_reg_resize(vp, length, FALSE); 1789 if (error == 0) 1790 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1791 1792out: 1793 tmpfs_update(vp); 1794 1795 return (error); 1796} 1797 1798static __inline int 1799tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b) 1800{ 1801 if (a->td_hash > b->td_hash) 1802 return (1); 1803 else if (a->td_hash < b->td_hash) 1804 return (-1); 1805 return (0); 1806} 1807 1808RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp); 1809