tmpfs_vnops.c revision 171070
1/* $NetBSD: tmpfs_vnops.c,v 1.35 2007/01/04 15:42:37 elad Exp $ */ 2 3/* 4 * Copyright (c) 2005 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 9 * 2005 program. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40/* 41 * tmpfs vnode interface. 42 */ 43#include <sys/cdefs.h> 44__FBSDID("$FreeBSD: head/sys/fs/tmpfs/tmpfs_vnops.c 171070 2007-06-28 02:39:31Z delphij $"); 45 46#include <sys/param.h> 47#include <sys/fcntl.h> 48#include <sys/lockf.h> 49#include <sys/namei.h> 50#include <sys/priv.h> 51#include <sys/proc.h> 52#include <sys/resourcevar.h> 53#include <sys/stat.h> 54#include <sys/systm.h> 55#include <sys/unistd.h> 56#include <sys/vnode.h> 57 58#include <vm/vm.h> 59#include <vm/vm_object.h> 60#include <vm/vm_page.h> 61#include <vm/vm_pager.h> 62#include <sys/sched.h> 63#include <sys/sf_buf.h> 64#include <machine/_inttypes.h> 65 66#include <fs/fifofs/fifo.h> 67#include <fs/tmpfs/tmpfs_vnops.h> 68#include <fs/tmpfs/tmpfs.h> 69 70/* --------------------------------------------------------------------- */ 71 72static int 73tmpfs_lookup(struct vop_cachedlookup_args *v) 74{ 75 struct vnode *dvp = v->a_dvp; 76 struct vnode **vpp = v->a_vpp; 77 struct componentname *cnp = v->a_cnp; 78 struct thread *td = cnp->cn_thread; 79 80 int error; 81 struct tmpfs_dirent *de; 82 struct tmpfs_node *dnode; 83 84 dnode = VP_TO_TMPFS_DIR(dvp); 85 *vpp = NULLVP; 86 87 /* Check accessibility of requested node as a first step. */ 88 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td); 89 if (error != 0) 90 goto out; 91 92 /* We cannot be requesting the parent directory of the root node. */ 93 MPASS(IMPLIES(dnode->tn_type == VDIR && 94 dnode->tn_dir.tn_parent == dnode, 95 !(cnp->cn_flags & ISDOTDOT))); 96 97 if (cnp->cn_flags & ISDOTDOT) { 98 VOP_UNLOCK(dvp, 0, td); 99 100 /* Allocate a new vnode on the matching entry. */ 101 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent, vpp, td); 102 103 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td); 104 105 dnode->tn_dir.tn_parent->tn_lookup_dirent = NULL; 106 } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 107 VREF(dvp); 108 *vpp = dvp; 109 dnode->tn_lookup_dirent = NULL; 110 error = 0; 111 } else { 112 de = tmpfs_dir_lookup(dnode, cnp); 113 if (de == NULL) { 114 /* The entry was not found in the directory. 115 * This is OK if we are creating or renaming an 116 * entry and are working on the last component of 117 * the path name. */ 118 if ((cnp->cn_flags & ISLASTCN) && 119 (cnp->cn_nameiop == CREATE || \ 120 cnp->cn_nameiop == RENAME)) { 121 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, 122 cnp->cn_thread); 123 if (error != 0) 124 goto out; 125 126 /* Keep the component name in the buffer for 127 * future uses. */ 128 cnp->cn_flags |= SAVENAME; 129 130 error = EJUSTRETURN; 131 } else 132 error = ENOENT; 133 } else { 134 struct tmpfs_node *tnode; 135 136 /* The entry was found, so get its associated 137 * tmpfs_node. */ 138 tnode = de->td_node; 139 140 /* If we are not at the last path component and 141 * found a non-directory or non-link entry (which 142 * may itself be pointing to a directory), raise 143 * an error. */ 144 if ((tnode->tn_type != VDIR && 145 tnode->tn_type != VLNK) && 146 !(cnp->cn_flags & ISLASTCN)) { 147 error = ENOTDIR; 148 goto out; 149 } 150 151 /* If we are deleting or renaming the entry, keep 152 * track of its tmpfs_dirent so that it can be 153 * easily deleted later. */ 154 if ((cnp->cn_flags & ISLASTCN) && 155 (cnp->cn_nameiop == DELETE || 156 cnp->cn_nameiop == RENAME)) { 157 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, 158 cnp->cn_thread); 159 if (error != 0) 160 goto out; 161 162 /* Allocate a new vnode on the matching entry. */ 163 error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp, td); 164 if (error != 0) 165 goto out; 166 167 if ((dnode->tn_mode & S_ISTXT) && 168 VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) && 169 VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) { 170 error = EPERM; 171 vput(*vpp); 172 *vpp = NULL; 173 goto out; 174 } 175 tnode->tn_lookup_dirent = de; 176 cnp->cn_flags |= SAVENAME; 177 } 178 else 179 error = tmpfs_alloc_vp(dvp->v_mount, tnode, vpp, td); 180 181 } 182 } 183 184 /* Store the result of this lookup in the cache. Avoid this if the 185 * request was for creation, as it does not improve timings on 186 * emprical tests. */ 187 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) 188 cache_enter(dvp, *vpp, cnp); 189 190out: 191 /* If there were no errors, *vpp cannot be null and it must be 192 * locked. */ 193 MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp, td))); 194 195 return error; 196} 197 198/* --------------------------------------------------------------------- */ 199 200static int 201tmpfs_create(struct vop_create_args *v) 202{ 203 struct vnode *dvp = v->a_dvp; 204 struct vnode **vpp = v->a_vpp; 205 struct componentname *cnp = v->a_cnp; 206 struct vattr *vap = v->a_vap; 207 208 MPASS(vap->va_type == VREG || vap->va_type == VSOCK); 209 210 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); 211} 212/* --------------------------------------------------------------------- */ 213 214static int 215tmpfs_mknod(struct vop_mknod_args *v) 216{ 217 struct vnode *dvp = v->a_dvp; 218 struct vnode **vpp = v->a_vpp; 219 struct componentname *cnp = v->a_cnp; 220 struct vattr *vap = v->a_vap; 221 222 if (vap->va_type != VBLK && vap->va_type != VCHR && 223 vap->va_type != VFIFO) 224 return EINVAL; 225 226 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); 227} 228 229/* --------------------------------------------------------------------- */ 230 231static int 232tmpfs_open(struct vop_open_args *v) 233{ 234 struct vnode *vp = v->a_vp; 235 int mode = v->a_mode; 236 237 int error; 238 struct tmpfs_node *node; 239 240 MPASS(VOP_ISLOCKED(vp, v->a_td)); 241 242 node = VP_TO_TMPFS_NODE(vp); 243 244 /* The file is still active but all its names have been removed 245 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as 246 * it is about to die. */ 247 if (node->tn_links < 1) 248 return (ENOENT); 249 250 /* If the file is marked append-only, deny write requests. */ 251 if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE) 252 error = EPERM; 253 else { 254 error = 0; 255 vnode_create_vobject(vp, node->tn_size, v->a_td); 256 } 257 258 MPASS(VOP_ISLOCKED(vp, v->a_td)); 259 return error; 260} 261 262/* --------------------------------------------------------------------- */ 263 264static int 265tmpfs_close(struct vop_close_args *v) 266{ 267 struct vnode *vp = v->a_vp; 268 269 struct tmpfs_node *node; 270 271 MPASS(VOP_ISLOCKED(vp, v->a_td)); 272 273 node = VP_TO_TMPFS_NODE(vp); 274 275 if (node->tn_links > 0) { 276 /* Update node times. No need to do it if the node has 277 * been deleted, because it will vanish after we return. */ 278 tmpfs_update(vp); 279 } 280 281 return 0; 282} 283 284/* --------------------------------------------------------------------- */ 285 286int 287tmpfs_access(struct vop_access_args *v) 288{ 289 struct vnode *vp = v->a_vp; 290 int mode = v->a_mode; 291 struct ucred *cred = v->a_cred; 292 293 int error; 294 struct tmpfs_node *node; 295 296 MPASS(VOP_ISLOCKED(vp, v->a_td)); 297 298 node = VP_TO_TMPFS_NODE(vp); 299 300 switch (vp->v_type) { 301 case VDIR: 302 /* FALLTHROUGH */ 303 case VLNK: 304 /* FALLTHROUGH */ 305 case VREG: 306 if (mode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) { 307 error = EROFS; 308 goto out; 309 } 310 break; 311 312 case VBLK: 313 /* FALLTHROUGH */ 314 case VCHR: 315 /* FALLTHROUGH */ 316 case VSOCK: 317 /* FALLTHROUGH */ 318 case VFIFO: 319 break; 320 321 default: 322 error = EINVAL; 323 goto out; 324 } 325 326 if (mode & VWRITE && node->tn_flags & IMMUTABLE) { 327 error = EPERM; 328 goto out; 329 } 330 331 error = vaccess(vp->v_type, node->tn_mode, node->tn_uid, 332 node->tn_gid, mode, cred, NULL); 333 334out: 335 MPASS(VOP_ISLOCKED(vp, v->a_td)); 336 337 return error; 338} 339 340/* --------------------------------------------------------------------- */ 341 342int 343tmpfs_getattr(struct vop_getattr_args *v) 344{ 345 struct vnode *vp = v->a_vp; 346 struct vattr *vap = v->a_vap; 347 348 struct tmpfs_node *node; 349 350 node = VP_TO_TMPFS_NODE(vp); 351 352 VATTR_NULL(vap); 353 354 tmpfs_update(vp); 355 356 vap->va_type = vp->v_type; 357 vap->va_mode = node->tn_mode; 358 vap->va_nlink = node->tn_links; 359 vap->va_uid = node->tn_uid; 360 vap->va_gid = node->tn_gid; 361 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 362 vap->va_fileid = node->tn_id; 363 vap->va_size = node->tn_size; 364 vap->va_blocksize = PAGE_SIZE; 365 vap->va_atime = node->tn_atime; 366 vap->va_mtime = node->tn_mtime; 367 vap->va_ctime = node->tn_ctime; 368 vap->va_birthtime = node->tn_birthtime; 369 vap->va_gen = node->tn_gen; 370 vap->va_flags = node->tn_flags; 371 vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ? 372 node->tn_rdev : VNOVAL; 373 vap->va_bytes = round_page(node->tn_size); 374 vap->va_filerev = VNOVAL; 375 vap->va_vaflags = 0; 376 vap->va_spare = VNOVAL; /* XXX */ 377 378 return 0; 379} 380 381/* --------------------------------------------------------------------- */ 382 383/* XXX Should this operation be atomic? I think it should, but code in 384 * XXX other places (e.g., ufs) doesn't seem to be... */ 385int 386tmpfs_setattr(struct vop_setattr_args *v) 387{ 388 struct vnode *vp = v->a_vp; 389 struct vattr *vap = v->a_vap; 390 struct ucred *cred = v->a_cred; 391 struct thread *l = v->a_td; 392 393 int error; 394 395 MPASS(VOP_ISLOCKED(vp, l)); 396 397 error = 0; 398 399 /* Abort if any unsettable attribute is given. */ 400 if (vap->va_type != VNON || 401 vap->va_nlink != VNOVAL || 402 vap->va_fsid != VNOVAL || 403 vap->va_fileid != VNOVAL || 404 vap->va_blocksize != VNOVAL || 405 vap->va_gen != VNOVAL || 406 vap->va_rdev != VNOVAL || 407 vap->va_bytes != VNOVAL) 408 error = EINVAL; 409 410 if (error == 0 && (vap->va_flags != VNOVAL)) 411 error = tmpfs_chflags(vp, vap->va_flags, cred, l); 412 413 if (error == 0 && (vap->va_size != VNOVAL)) 414 error = tmpfs_chsize(vp, vap->va_size, cred, l); 415 416 if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL)) 417 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, 418 l); 419 420 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) 421 error = tmpfs_chmod(vp, vap->va_mode, cred, l); 422 423 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL && 424 vap->va_atime.tv_nsec != VNOVAL) || 425 (vap->va_mtime.tv_sec != VNOVAL && 426 vap->va_mtime.tv_nsec != VNOVAL) || 427 (vap->va_birthtime.tv_sec != VNOVAL && 428 vap->va_birthtime.tv_nsec != VNOVAL))) 429 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime, 430 &vap->va_birthtime, vap->va_vaflags, cred, l); 431 432 /* Update the node times. We give preference to the error codes 433 * generated by this function rather than the ones that may arise 434 * from tmpfs_update. */ 435 tmpfs_update(vp); 436 437 MPASS(VOP_ISLOCKED(vp, l)); 438 439 return error; 440} 441 442/* --------------------------------------------------------------------- */ 443static int 444tmpfs_uio_xfer(struct tmpfs_mount *tmp, struct tmpfs_node *node, 445 struct uio *uio, vm_object_t uobj) 446{ 447 struct sf_buf *sf; 448 vm_pindex_t idx; 449 vm_offset_t d; 450 vm_page_t m; 451 size_t len; 452 int error = 0; 453 454 /* uobj - locked by caller */ 455 456 VM_OBJECT_LOCK(uobj); 457 vm_object_pip_add(uobj, 1); 458 while (error == 0 && uio->uio_resid > 0) { 459 if (node->tn_size <= uio->uio_offset) 460 break; 461 462 len = MIN(node->tn_size - uio->uio_offset, uio->uio_resid); 463 if (len == 0) 464 break; 465 466 idx = OFF_TO_IDX(uio->uio_offset); 467 d = uio->uio_offset - IDX_TO_OFF(idx); 468 len = MIN(len, (PAGE_SIZE - d)); 469 m = vm_page_grab(uobj, idx, VM_ALLOC_WIRED | VM_ALLOC_ZERO | 470 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 471 if (uio->uio_rw == UIO_READ && m->valid != VM_PAGE_BITS_ALL) 472 vm_page_zero_invalid(m, TRUE); 473 VM_OBJECT_UNLOCK(uobj); 474 sched_pin(); 475 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 476 error = uiomove((void *)(sf_buf_kva(sf) + d), len, uio); 477 sf_buf_free(sf); 478 sched_unpin(); 479 VM_OBJECT_LOCK(uobj); 480 vm_page_lock_queues(); 481 if (error == 0 && uio->uio_rw == UIO_WRITE) { 482 vm_page_set_validclean(m, d, len); 483 vm_page_zero_invalid(m, TRUE); 484 vm_page_dirty(m); 485 } 486 vm_page_unwire(m, 0); 487 vm_page_activate(m); 488 vm_page_wakeup(m); 489 vm_page_unlock_queues(); 490 } 491 vm_object_pip_subtract(uobj, 1); 492 VM_OBJECT_UNLOCK(uobj); 493 return error; 494} 495 496static int 497tmpfs_read(struct vop_read_args *v) 498{ 499 struct vnode *vp = v->a_vp; 500 struct uio *uio = v->a_uio; 501 502 struct tmpfs_node *node; 503 vm_object_t uobj; 504 505 int error; 506 507 node = VP_TO_TMPFS_NODE(vp); 508 509 if (vp->v_type != VREG) { 510 error = EISDIR; 511 goto out; 512 } 513 514 if (uio->uio_offset < 0) { 515 error = EINVAL; 516 goto out; 517 } 518 519 node->tn_status |= TMPFS_NODE_ACCESSED; 520 521 uobj = node->tn_reg.tn_aobj; 522 error = tmpfs_uio_xfer(VFS_TO_TMPFS(vp->v_mount), node, uio, uobj); 523 524out: 525 526 return error; 527} 528 529/* --------------------------------------------------------------------- */ 530 531static int 532tmpfs_write(struct vop_write_args *v) 533{ 534 struct vnode *vp = v->a_vp; 535 struct uio *uio = v->a_uio; 536 int ioflag = v->a_ioflag; 537 struct thread *td = uio->uio_td; 538 539 boolean_t extended; 540 int error; 541 off_t oldsize; 542 struct tmpfs_node *node; 543 vm_object_t uobj; 544 545 node = VP_TO_TMPFS_NODE(vp); 546 oldsize = node->tn_size; 547 548 if (uio->uio_offset < 0 || vp->v_type != VREG) { 549 error = EINVAL; 550 goto out; 551 } 552 553 if (uio->uio_resid == 0) { 554 error = 0; 555 goto out; 556 } 557 558 if (ioflag & IO_APPEND) 559 uio->uio_offset = node->tn_size; 560 561 if (uio->uio_offset + uio->uio_resid > 562 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 563 return (EFBIG); 564 565 if (vp->v_type == VREG && td != NULL) { 566 PROC_LOCK(td->td_proc); 567 if (uio->uio_offset + uio->uio_resid > 568 lim_cur(td->td_proc, RLIMIT_FSIZE)) { 569 psignal(td->td_proc, SIGXFSZ); 570 PROC_UNLOCK(td->td_proc); 571 return (EFBIG); 572 } 573 PROC_UNLOCK(td->td_proc); 574 } 575 576 extended = uio->uio_offset + uio->uio_resid > node->tn_size; 577 if (extended) { 578 error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid); 579 if (error != 0) 580 goto out; 581 } 582 583 uobj = node->tn_reg.tn_aobj; 584 error = tmpfs_uio_xfer(VFS_TO_TMPFS(vp->v_mount), node, uio, uobj); 585 586 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 587 (extended ? TMPFS_NODE_CHANGED : 0); 588 589 if (node->tn_mode & (S_ISUID | S_ISGID)) { 590 if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0)) 591 node->tn_mode &= ~(S_ISUID | S_ISGID); 592 } 593 594 if (error != 0) 595 (void)tmpfs_reg_resize(vp, oldsize); 596 597out: 598 MPASS(IMPLIES(error == 0, uio->uio_resid == 0)); 599 MPASS(IMPLIES(error != 0, oldsize == node->tn_size)); 600 601 return error; 602} 603 604/* --------------------------------------------------------------------- */ 605 606static int 607tmpfs_fsync(struct vop_fsync_args *v) 608{ 609 struct vnode *vp = v->a_vp; 610 611 MPASS(VOP_ISLOCKED(vp, v->a_td)); 612 613 tmpfs_update(vp); 614 615 return 0; 616} 617 618/* --------------------------------------------------------------------- */ 619 620static int 621tmpfs_remove(struct vop_remove_args *v) 622{ 623 struct vnode *dvp = v->a_dvp; 624 struct vnode *vp = v->a_vp; 625 626 int error; 627 struct tmpfs_dirent *de; 628 struct tmpfs_mount *tmp; 629 struct tmpfs_node *dnode; 630 struct tmpfs_node *node; 631 632 MPASS(VOP_ISLOCKED(dvp, v->a_cnp->cn_thread)); 633 MPASS(VOP_ISLOCKED(vp, v->a_cnp->cn_thread)); 634 635 if (vp->v_type == VDIR) { 636 error = EISDIR; 637 goto out; 638 } 639 640 dnode = VP_TO_TMPFS_DIR(dvp); 641 node = VP_TO_TMPFS_NODE(vp); 642 tmp = VFS_TO_TMPFS(vp->v_mount); 643 de = node->tn_lookup_dirent; 644 MPASS(de != NULL); 645 646 /* Files marked as immutable or append-only cannot be deleted. */ 647 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || 648 (dnode->tn_flags & APPEND)) { 649 error = EPERM; 650 goto out; 651 } 652 653 /* Remove the entry from the directory; as it is a file, we do not 654 * have to change the number of hard links of the directory. */ 655 tmpfs_dir_detach(dvp, de); 656 657 /* Free the directory entry we just deleted. Note that the node 658 * referred by it will not be removed until the vnode is really 659 * reclaimed. */ 660 tmpfs_free_dirent(tmp, de, TRUE); 661 662 if (node->tn_links > 0) 663 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 664 TMPFS_NODE_MODIFIED; 665 error = 0; 666 667out: 668 669 return error; 670} 671 672/* --------------------------------------------------------------------- */ 673 674static int 675tmpfs_link(struct vop_link_args *v) 676{ 677 struct vnode *dvp = v->a_tdvp; 678 struct vnode *vp = v->a_vp; 679 struct componentname *cnp = v->a_cnp; 680 681 int error; 682 struct tmpfs_dirent *de; 683 struct tmpfs_node *dnode; 684 struct tmpfs_node *node; 685 686 MPASS(VOP_ISLOCKED(dvp, cnp->cn_thread)); 687 MPASS(cnp->cn_flags & HASBUF); 688 MPASS(dvp != vp); /* XXX When can this be false? */ 689 690 dnode = VP_TO_TMPFS_DIR(dvp); 691 node = VP_TO_TMPFS_NODE(vp); 692 693 /* XXX: Why aren't the following two tests done by the caller? */ 694 695 /* Hard links of directories are forbidden. */ 696 if (vp->v_type == VDIR) { 697 error = EPERM; 698 goto out; 699 } 700 701 /* Cannot create cross-device links. */ 702 if (dvp->v_mount != vp->v_mount) { 703 error = EXDEV; 704 goto out; 705 } 706 707 /* Ensure that we do not overflow the maximum number of links imposed 708 * by the system. */ 709 MPASS(node->tn_links <= LINK_MAX); 710 if (node->tn_links == LINK_MAX) { 711 error = EMLINK; 712 goto out; 713 } 714 715 /* We cannot create links of files marked immutable or append-only. */ 716 if (node->tn_flags & (IMMUTABLE | APPEND)) { 717 error = EPERM; 718 goto out; 719 } 720 721 /* Allocate a new directory entry to represent the node. */ 722 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, 723 cnp->cn_nameptr, cnp->cn_namelen, &de); 724 if (error != 0) 725 goto out; 726 727 /* Insert the new directory entry into the appropriate directory. */ 728 tmpfs_dir_attach(dvp, de); 729 730 /* vp link count has changed, so update node times. */ 731 node->tn_status |= TMPFS_NODE_CHANGED; 732 tmpfs_update(vp); 733 734 error = 0; 735 736out: 737 return error; 738} 739 740/* --------------------------------------------------------------------- */ 741 742static int 743tmpfs_rename(struct vop_rename_args *v) 744{ 745 struct vnode *fdvp = v->a_fdvp; 746 struct vnode *fvp = v->a_fvp; 747 struct componentname *fcnp = v->a_fcnp; 748 struct vnode *tdvp = v->a_tdvp; 749 struct vnode *tvp = v->a_tvp; 750 struct componentname *tcnp = v->a_tcnp; 751 struct tmpfs_node *tnode = 0; /* pacify gcc */ 752 753 char *newname; 754 int error; 755 struct tmpfs_dirent *de; 756 struct tmpfs_mount *tmp; 757 struct tmpfs_node *fdnode; 758 struct tmpfs_node *fnode; 759 struct tmpfs_node *tdnode; 760 761 MPASS(VOP_ISLOCKED(tdvp, tcnp->cn_thread)); 762 MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp, tcnp->cn_thread))); 763 MPASS(fcnp->cn_flags & HASBUF); 764 MPASS(tcnp->cn_flags & HASBUF); 765 766 fdnode = VP_TO_TMPFS_DIR(fdvp); 767 fnode = VP_TO_TMPFS_NODE(fvp); 768 de = fnode->tn_lookup_dirent; 769 770 /* Disallow cross-device renames. 771 * XXX Why isn't this done by the caller? */ 772 if (fvp->v_mount != tdvp->v_mount || 773 (tvp != NULL && fvp->v_mount != tvp->v_mount)) { 774 error = EXDEV; 775 goto out; 776 } 777 778 tmp = VFS_TO_TMPFS(tdvp->v_mount); 779 tdnode = VP_TO_TMPFS_DIR(tdvp); 780 781 /* If source and target are the same file, there is nothing to do. */ 782 if (fvp == tvp) { 783 error = 0; 784 goto out; 785 } 786 787 /* Avoid manipulating '.' and '..' entries. */ 788 if (de == NULL) { 789 MPASS(fvp->v_type == VDIR); 790 error = EINVAL; 791 goto out; 792 } 793 MPASS(de->td_node == fnode); 794 795 /* If re-naming a directory to another preexisting directory 796 * ensure that the target directory is empty so that its 797 * removal causes no side effects. 798 * Kern_rename gurantees the destination to be a directory 799 * if the source is one. */ 800 if (tvp != NULL) { 801 tnode = VP_TO_TMPFS_NODE(tvp); 802 803 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 804 (tdnode->tn_flags & (APPEND | IMMUTABLE))) { 805 error = EPERM; 806 goto out; 807 } 808 809 if ((de->td_node->tn_type == VDIR) && (tnode->tn_size > 0)) { 810 error = ENOTEMPTY; 811 goto out; 812 } 813 } 814 815 /* If we need to move the directory between entries, lock the 816 * source so that we can safely operate on it. */ 817 if (fdnode != tdnode) { 818 error = vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY, tcnp->cn_thread); 819 if (error != 0) 820 goto out; 821 } 822 823 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) 824 || (fdnode->tn_flags & (APPEND | IMMUTABLE))) { 825 error = EPERM; 826 goto out_locked; 827 } 828 829 /* Ensure that we have enough memory to hold the new name, if it 830 * has to be changed. */ 831 if (fcnp->cn_namelen != tcnp->cn_namelen || 832 memcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) { 833 newname = tmpfs_str_zone_alloc(&tmp->tm_str_pool, M_WAITOK, 834 tcnp->cn_namelen); 835 if (newname == NULL) { 836 error = ENOSPC; 837 goto out_locked; 838 } 839 } else 840 newname = NULL; 841 842 /* If the node is being moved to another directory, we have to do 843 * the move. */ 844 if (fdnode != tdnode) { 845 /* In case we are moving a directory, we have to adjust its 846 * parent to point to the new parent. */ 847 if (de->td_node->tn_type == VDIR) { 848 struct tmpfs_node *n; 849 850 /* Ensure the target directory is not a child of the 851 * directory being moved. Otherwise, we'd end up 852 * with stale nodes. */ 853 n = tdnode; 854 while (n != n->tn_dir.tn_parent) { 855 if (n == fnode) { 856 error = EINVAL; 857 if (newname != NULL) 858 tmpfs_str_zone_free(&tmp->tm_str_pool, 859 newname, tcnp->cn_namelen); 860 goto out_locked; 861 } 862 n = n->tn_dir.tn_parent; 863 } 864 865 /* Adjust the parent pointer. */ 866 TMPFS_VALIDATE_DIR(fnode); 867 de->td_node->tn_dir.tn_parent = tdnode; 868 869 /* As a result of changing the target of the '..' 870 * entry, the link count of the source and target 871 * directories has to be adjusted. */ 872 fdnode->tn_links--; 873 tdnode->tn_links++; 874 } 875 876 /* Do the move: just remove the entry from the source directory 877 * and insert it into the target one. */ 878 tmpfs_dir_detach(fdvp, de); 879 tmpfs_dir_attach(tdvp, de); 880 } 881 882 /* If the name has changed, we need to make it effective by changing 883 * it in the directory entry. */ 884 if (newname != NULL) { 885 MPASS(tcnp->cn_namelen <= MAXNAMLEN); 886 887 tmpfs_str_zone_free(&tmp->tm_str_pool, de->td_name, 888 de->td_namelen); 889 de->td_namelen = (uint16_t)tcnp->cn_namelen; 890 memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen); 891 de->td_name = newname; 892 893 fnode->tn_status |= TMPFS_NODE_CHANGED; 894 tdnode->tn_status |= TMPFS_NODE_MODIFIED; 895 } 896 897 /* If we are overwriting an entry, we have to remove the old one 898 * from the target directory. */ 899 if (tvp != NULL) { 900 /* Remove the old entry from the target directory. */ 901 de = tnode->tn_lookup_dirent; 902 tmpfs_dir_detach(tdvp, de); 903 904 /* Free the directory entry we just deleted. Note that the 905 * node referred by it will not be removed until the vnode is 906 * really reclaimed. */ 907 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE); 908 } 909 910 error = 0; 911 912out_locked: 913 if (fdnode != tdnode) 914 VOP_UNLOCK(fdvp, 0, tcnp->cn_thread); 915 916out: 917 /* Release target nodes. */ 918 /* XXX: I don't understand when tdvp can be the same as tvp, but 919 * other code takes care of this... */ 920 if (tdvp == tvp) 921 vrele(tdvp); 922 else 923 vput(tdvp); 924 if (tvp != NULL) 925 vput(tvp); 926 927 /* Release source nodes. */ 928 vrele(fdvp); 929 vrele(fvp); 930 931 return error; 932} 933 934/* --------------------------------------------------------------------- */ 935 936static int 937tmpfs_mkdir(struct vop_mkdir_args *v) 938{ 939 struct vnode *dvp = v->a_dvp; 940 struct vnode **vpp = v->a_vpp; 941 struct componentname *cnp = v->a_cnp; 942 struct vattr *vap = v->a_vap; 943 944 MPASS(vap->va_type == VDIR); 945 946 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); 947} 948 949/* --------------------------------------------------------------------- */ 950 951static int 952tmpfs_rmdir(struct vop_rmdir_args *v) 953{ 954 struct vnode *dvp = v->a_dvp; 955 struct vnode *vp = v->a_vp; 956 957 int error; 958 struct tmpfs_dirent *de; 959 struct tmpfs_mount *tmp; 960 struct tmpfs_node *dnode; 961 struct tmpfs_node *node; 962 963 MPASS(VOP_ISLOCKED(dvp, v->a_cnp->cn_thread)); 964 MPASS(VOP_ISLOCKED(vp, v->a_cnp->cn_thread)); 965 966 tmp = VFS_TO_TMPFS(dvp->v_mount); 967 dnode = VP_TO_TMPFS_DIR(dvp); 968 node = VP_TO_TMPFS_DIR(vp); 969 970 /* Directories with more than two entries ('.' and '..') cannot be 971 * removed. */ 972 if (node->tn_size > 0) { 973 error = ENOTEMPTY; 974 goto out; 975 } 976 977 if ((dnode->tn_flags & APPEND) 978 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 979 error = EPERM; 980 goto out; 981 } 982 983 /* This invariant holds only if we are not trying to remove "..". 984 * We checked for that above so this is safe now. */ 985 MPASS(node->tn_dir.tn_parent == dnode); 986 987 /* Get the directory entry associated with node (vp). This was 988 * filled by tmpfs_lookup while looking up the entry. */ 989 de = node->tn_lookup_dirent; 990 MPASS(TMPFS_DIRENT_MATCHES(de, 991 v->a_cnp->cn_nameptr, 992 v->a_cnp->cn_namelen)); 993 994 /* Check flags to see if we are allowed to remove the directory. */ 995 if (dnode->tn_flags & APPEND 996 || node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) { 997 error = EPERM; 998 goto out; 999 } 1000 1001 /* Detach the directory entry from the directory (dnode). */ 1002 tmpfs_dir_detach(dvp, de); 1003 1004 node->tn_links--; 1005 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1006 TMPFS_NODE_MODIFIED; 1007 node->tn_dir.tn_parent->tn_links--; 1008 node->tn_dir.tn_parent->tn_status |= TMPFS_NODE_ACCESSED | \ 1009 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1010 1011 cache_purge(dvp); 1012 cache_purge(vp); 1013 1014 /* Free the directory entry we just deleted. Note that the node 1015 * referred by it will not be removed until the vnode is really 1016 * reclaimed. */ 1017 tmpfs_free_dirent(tmp, de, TRUE); 1018 1019 /* Release the deleted vnode (will destroy the node, notify 1020 * interested parties and clean it from the cache). */ 1021 1022 dnode->tn_status |= TMPFS_NODE_CHANGED; 1023 tmpfs_update(dvp); 1024 1025 error = 0; 1026 1027out: 1028 return error; 1029} 1030 1031/* --------------------------------------------------------------------- */ 1032 1033static int 1034tmpfs_symlink(struct vop_symlink_args *v) 1035{ 1036 struct vnode *dvp = v->a_dvp; 1037 struct vnode **vpp = v->a_vpp; 1038 struct componentname *cnp = v->a_cnp; 1039 struct vattr *vap = v->a_vap; 1040 char *target = v->a_target; 1041 1042#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */ 1043 MPASS(vap->va_type == VLNK); 1044#else 1045 vap->va_type = VLNK; 1046#endif 1047 1048 return tmpfs_alloc_file(dvp, vpp, vap, cnp, target); 1049} 1050 1051/* --------------------------------------------------------------------- */ 1052 1053static int 1054tmpfs_readdir(struct vop_readdir_args *v) 1055{ 1056 struct vnode *vp = v->a_vp; 1057 struct uio *uio = v->a_uio; 1058 int *eofflag = v->a_eofflag; 1059 u_long **cookies = v->a_cookies; 1060 int *ncookies = v->a_ncookies; 1061 1062 int error; 1063 off_t startoff; 1064 off_t cnt; 1065 struct tmpfs_node *node; 1066 1067 /* This operation only makes sense on directory nodes. */ 1068 if (vp->v_type != VDIR) { 1069 error = ENOTDIR; 1070 goto out; 1071 } 1072 1073 node = VP_TO_TMPFS_DIR(vp); 1074 1075 startoff = uio->uio_offset; 1076 1077 cnt = 0; 1078 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) { 1079 error = tmpfs_dir_getdotdent(node, uio); 1080 if (error == -1) { 1081 error = 0; 1082 goto outok; 1083 } else if (error != 0) 1084 goto outok; 1085 cnt++; 1086 } 1087 1088 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) { 1089 error = tmpfs_dir_getdotdotdent(node, uio); 1090 if (error == -1) { 1091 error = 0; 1092 goto outok; 1093 } else if (error != 0) 1094 goto outok; 1095 cnt++; 1096 } 1097 1098 error = tmpfs_dir_getdents(node, uio, &cnt); 1099 if (error == -1) 1100 error = 0; 1101 MPASS(error >= 0); 1102 1103outok: 1104 /* This label assumes that startoff has been 1105 * initialized. If the compiler didn't spit out warnings, we'd 1106 * simply make this one be 'out' and drop 'outok'. */ 1107 1108 if (eofflag != NULL) 1109 *eofflag = 1110 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); 1111 1112 /* Update NFS-related variables. */ 1113 if (error == 0 && cookies != NULL && ncookies != NULL) { 1114 off_t i; 1115 off_t off = startoff; 1116 struct tmpfs_dirent *de = NULL; 1117 1118 *ncookies = cnt; 1119 *cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK); 1120 1121 for (i = 0; i < cnt; i++) { 1122 MPASS(off != TMPFS_DIRCOOKIE_EOF); 1123 if (off == TMPFS_DIRCOOKIE_DOT) { 1124 off = TMPFS_DIRCOOKIE_DOTDOT; 1125 } else { 1126 if (off == TMPFS_DIRCOOKIE_DOTDOT) { 1127 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead); 1128 } else if (de != NULL) { 1129 de = TAILQ_NEXT(de, td_entries); 1130 } else { 1131 de = tmpfs_dir_lookupbycookie(node, 1132 off); 1133 MPASS(de != NULL); 1134 de = TAILQ_NEXT(de, td_entries); 1135 } 1136 if (de == NULL) { 1137 off = TMPFS_DIRCOOKIE_EOF; 1138 } else { 1139 off = TMPFS_DIRCOOKIE(de); 1140 } 1141 } 1142 1143 (*cookies)[i] = off; 1144 } 1145 MPASS(uio->uio_offset == off); 1146 } 1147 1148out: 1149 return error; 1150} 1151 1152/* --------------------------------------------------------------------- */ 1153 1154static int 1155tmpfs_readlink(struct vop_readlink_args *v) 1156{ 1157 struct vnode *vp = v->a_vp; 1158 struct uio *uio = v->a_uio; 1159 1160 int error; 1161 struct tmpfs_node *node; 1162 1163 MPASS(uio->uio_offset == 0); 1164 MPASS(vp->v_type == VLNK); 1165 1166 node = VP_TO_TMPFS_NODE(vp); 1167 1168 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid), 1169 uio); 1170 node->tn_status |= TMPFS_NODE_ACCESSED; 1171 1172 return error; 1173} 1174 1175/* --------------------------------------------------------------------- */ 1176 1177static int 1178tmpfs_inactive(struct vop_inactive_args *v) 1179{ 1180 struct vnode *vp = v->a_vp; 1181 struct thread *l = v->a_td; 1182 1183 struct tmpfs_node *node; 1184 1185 MPASS(VOP_ISLOCKED(vp, l)); 1186 1187 node = VP_TO_TMPFS_NODE(vp); 1188 1189 if (node->tn_links == 0) 1190 vrecycle(vp, l); 1191 1192 return 0; 1193} 1194 1195/* --------------------------------------------------------------------- */ 1196 1197int 1198tmpfs_reclaim(struct vop_reclaim_args *v) 1199{ 1200 struct vnode *vp = v->a_vp; 1201 1202 struct tmpfs_mount *tmp; 1203 struct tmpfs_node *node; 1204 1205 node = VP_TO_TMPFS_NODE(vp); 1206 tmp = VFS_TO_TMPFS(vp->v_mount); 1207 1208 vnode_destroy_vobject(vp); 1209 cache_purge(vp); 1210 tmpfs_free_vp(vp); 1211 1212 /* If the node referenced by this vnode was deleted by the user, 1213 * we must free its associated data structures (now that the vnode 1214 * is being reclaimed). */ 1215 if (node->tn_links == 0) 1216 tmpfs_free_node(tmp, node); 1217 1218 MPASS(vp->v_data == NULL); 1219 return 0; 1220} 1221 1222/* --------------------------------------------------------------------- */ 1223 1224static int 1225tmpfs_print(struct vop_print_args *v) 1226{ 1227 struct vnode *vp = v->a_vp; 1228 1229 struct tmpfs_node *node; 1230 1231 node = VP_TO_TMPFS_NODE(vp); 1232 1233 printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n", 1234 node, node->tn_flags, node->tn_links); 1235 printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX 1236 ", status 0x%x\n", 1237 node->tn_mode, node->tn_uid, node->tn_gid, 1238 (uintmax_t)node->tn_size, node->tn_status); 1239 1240 if (vp->v_type == VFIFO) 1241 fifo_printinfo(vp); 1242 1243 printf("\n"); 1244 1245 return 0; 1246} 1247 1248/* --------------------------------------------------------------------- */ 1249 1250static int 1251tmpfs_pathconf(struct vop_pathconf_args *v) 1252{ 1253 int name = v->a_name; 1254 register_t *retval = v->a_retval; 1255 1256 int error; 1257 1258 error = 0; 1259 1260 switch (name) { 1261 case _PC_LINK_MAX: 1262 *retval = LINK_MAX; 1263 break; 1264 1265 case _PC_NAME_MAX: 1266 *retval = NAME_MAX; 1267 break; 1268 1269 case _PC_PATH_MAX: 1270 *retval = PATH_MAX; 1271 break; 1272 1273 case _PC_PIPE_BUF: 1274 *retval = PIPE_BUF; 1275 break; 1276 1277 case _PC_CHOWN_RESTRICTED: 1278 *retval = 1; 1279 break; 1280 1281 case _PC_NO_TRUNC: 1282 *retval = 1; 1283 break; 1284 1285 case _PC_SYNC_IO: 1286 *retval = 1; 1287 break; 1288 1289 case _PC_FILESIZEBITS: 1290 *retval = 0; /* XXX Don't know which value should I return. */ 1291 break; 1292 1293 default: 1294 error = EINVAL; 1295 } 1296 1297 return error; 1298} 1299 1300/* --------------------------------------------------------------------- */ 1301 1302static int 1303tmpfs_advlock(struct vop_advlock_args *v) 1304{ 1305 struct vnode *vp = v->a_vp; 1306 1307 struct tmpfs_node *node; 1308 1309 node = VP_TO_TMPFS_NODE(vp); 1310 1311 return lf_advlock(v, &node->tn_lockf, node->tn_size); 1312} 1313 1314/* --------------------------------------------------------------------- */ 1315 1316static int 1317tmpfs_vptofh(struct vop_vptofh_args *ap) 1318{ 1319 struct tmpfs_fid *tfhp; 1320 struct tmpfs_node *node; 1321 1322 tfhp = (struct tmpfs_fid *)ap->a_fhp; 1323 node = VP_TO_TMPFS_NODE(ap->a_vp); 1324 1325 tfhp->tf_len = sizeof(struct tmpfs_fid); 1326 tfhp->tf_id = node->tn_id; 1327 tfhp->tf_gen = node->tn_gen; 1328 1329 return (0); 1330} 1331 1332/* --------------------------------------------------------------------- */ 1333 1334/* 1335 * vnode operations vector used for files stored in a tmpfs file system. 1336 */ 1337struct vop_vector tmpfs_vnodeop_entries = { 1338 .vop_default = &default_vnodeops, 1339 .vop_lookup = vfs_cache_lookup, 1340 .vop_cachedlookup = tmpfs_lookup, 1341 .vop_create = tmpfs_create, 1342 .vop_mknod = tmpfs_mknod, 1343 .vop_open = tmpfs_open, 1344 .vop_close = tmpfs_close, 1345 .vop_access = tmpfs_access, 1346 .vop_getattr = tmpfs_getattr, 1347 .vop_setattr = tmpfs_setattr, 1348 .vop_read = tmpfs_read, 1349 .vop_write = tmpfs_write, 1350 .vop_fsync = tmpfs_fsync, 1351 .vop_remove = tmpfs_remove, 1352 .vop_link = tmpfs_link, 1353 .vop_rename = tmpfs_rename, 1354 .vop_mkdir = tmpfs_mkdir, 1355 .vop_rmdir = tmpfs_rmdir, 1356 .vop_symlink = tmpfs_symlink, 1357 .vop_readdir = tmpfs_readdir, 1358 .vop_readlink = tmpfs_readlink, 1359 .vop_inactive = tmpfs_inactive, 1360 .vop_reclaim = tmpfs_reclaim, 1361 .vop_print = tmpfs_print, 1362 .vop_pathconf = tmpfs_pathconf, 1363 .vop_advlock = tmpfs_advlock, 1364 .vop_vptofh = tmpfs_vptofh, 1365 .vop_bmap = VOP_EOPNOTSUPP, 1366}; 1367 1368