tmpfs_vnops.c revision 233851
1/* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 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 * 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 * tmpfs vnode interface. 35 */ 36#include <sys/cdefs.h> 37__FBSDID("$FreeBSD: stable/9/sys/fs/tmpfs/tmpfs_vnops.c 233851 2012-04-03 19:34:00Z gleb $"); 38 39#include <sys/param.h> 40#include <sys/fcntl.h> 41#include <sys/lockf.h> 42#include <sys/namei.h> 43#include <sys/priv.h> 44#include <sys/proc.h> 45#include <sys/sched.h> 46#include <sys/sf_buf.h> 47#include <sys/stat.h> 48#include <sys/systm.h> 49#include <sys/sysctl.h> 50#include <sys/unistd.h> 51#include <sys/vnode.h> 52 53#include <vm/vm.h> 54#include <vm/vm_object.h> 55#include <vm/vm_page.h> 56#include <vm/vm_pager.h> 57 58#include <machine/_inttypes.h> 59 60#include <fs/fifofs/fifo.h> 61#include <fs/tmpfs/tmpfs_vnops.h> 62#include <fs/tmpfs/tmpfs.h> 63 64SYSCTL_DECL(_vfs_tmpfs); 65 66static volatile int tmpfs_rename_restarts; 67SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD, 68 __DEVOLATILE(int *, &tmpfs_rename_restarts), 0, 69 "Times rename had to restart due to lock contention"); 70 71/* --------------------------------------------------------------------- */ 72 73static int 74tmpfs_lookup(struct vop_cachedlookup_args *v) 75{ 76 struct vnode *dvp = v->a_dvp; 77 struct vnode **vpp = v->a_vpp; 78 struct componentname *cnp = v->a_cnp; 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, cnp->cn_thread); 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 TMPFS_ASSERT_LOCKED(dnode); 98 if (dnode->tn_dir.tn_parent == NULL) { 99 error = ENOENT; 100 goto out; 101 } 102 if (cnp->cn_flags & ISDOTDOT) { 103 int ltype = 0; 104 105 ltype = VOP_ISLOCKED(dvp); 106 vhold(dvp); 107 VOP_UNLOCK(dvp, 0); 108 /* Allocate a new vnode on the matching entry. */ 109 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent, 110 cnp->cn_lkflags, vpp); 111 112 vn_lock(dvp, ltype | LK_RETRY); 113 vdrop(dvp); 114 } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 115 VREF(dvp); 116 *vpp = dvp; 117 error = 0; 118 } else { 119 de = tmpfs_dir_lookup(dnode, NULL, cnp); 120 if (de != NULL && de->td_node == NULL) 121 cnp->cn_flags |= ISWHITEOUT; 122 if (de == NULL || de->td_node == NULL) { 123 /* The entry was not found in the directory. 124 * This is OK if we are creating or renaming an 125 * entry and are working on the last component of 126 * the path name. */ 127 if ((cnp->cn_flags & ISLASTCN) && 128 (cnp->cn_nameiop == CREATE || \ 129 cnp->cn_nameiop == RENAME || 130 (cnp->cn_nameiop == DELETE && 131 cnp->cn_flags & DOWHITEOUT && 132 cnp->cn_flags & ISWHITEOUT))) { 133 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, 134 cnp->cn_thread); 135 if (error != 0) 136 goto out; 137 138 /* Keep the component name in the buffer for 139 * future uses. */ 140 cnp->cn_flags |= SAVENAME; 141 142 error = EJUSTRETURN; 143 } else 144 error = ENOENT; 145 } else { 146 struct tmpfs_node *tnode; 147 148 /* The entry was found, so get its associated 149 * tmpfs_node. */ 150 tnode = de->td_node; 151 152 /* If we are not at the last path component and 153 * found a non-directory or non-link entry (which 154 * may itself be pointing to a directory), raise 155 * an error. */ 156 if ((tnode->tn_type != VDIR && 157 tnode->tn_type != VLNK) && 158 !(cnp->cn_flags & ISLASTCN)) { 159 error = ENOTDIR; 160 goto out; 161 } 162 163 /* If we are deleting or renaming the entry, keep 164 * track of its tmpfs_dirent so that it can be 165 * easily deleted later. */ 166 if ((cnp->cn_flags & ISLASTCN) && 167 (cnp->cn_nameiop == DELETE || 168 cnp->cn_nameiop == RENAME)) { 169 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, 170 cnp->cn_thread); 171 if (error != 0) 172 goto out; 173 174 /* Allocate a new vnode on the matching entry. */ 175 error = tmpfs_alloc_vp(dvp->v_mount, tnode, 176 cnp->cn_lkflags, vpp); 177 if (error != 0) 178 goto out; 179 180 if ((dnode->tn_mode & S_ISTXT) && 181 VOP_ACCESS(dvp, VADMIN, cnp->cn_cred, cnp->cn_thread) && 182 VOP_ACCESS(*vpp, VADMIN, cnp->cn_cred, cnp->cn_thread)) { 183 error = EPERM; 184 vput(*vpp); 185 *vpp = NULL; 186 goto out; 187 } 188 cnp->cn_flags |= SAVENAME; 189 } else { 190 error = tmpfs_alloc_vp(dvp->v_mount, tnode, 191 cnp->cn_lkflags, vpp); 192 } 193 } 194 } 195 196 /* Store the result of this lookup in the cache. Avoid this if the 197 * request was for creation, as it does not improve timings on 198 * emprical tests. */ 199 if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE) 200 cache_enter(dvp, *vpp, cnp); 201 202out: 203 /* If there were no errors, *vpp cannot be null and it must be 204 * locked. */ 205 MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp))); 206 207 return error; 208} 209 210/* --------------------------------------------------------------------- */ 211 212static int 213tmpfs_create(struct vop_create_args *v) 214{ 215 struct vnode *dvp = v->a_dvp; 216 struct vnode **vpp = v->a_vpp; 217 struct componentname *cnp = v->a_cnp; 218 struct vattr *vap = v->a_vap; 219 220 MPASS(vap->va_type == VREG || vap->va_type == VSOCK); 221 222 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); 223} 224/* --------------------------------------------------------------------- */ 225 226static int 227tmpfs_mknod(struct vop_mknod_args *v) 228{ 229 struct vnode *dvp = v->a_dvp; 230 struct vnode **vpp = v->a_vpp; 231 struct componentname *cnp = v->a_cnp; 232 struct vattr *vap = v->a_vap; 233 234 if (vap->va_type != VBLK && vap->va_type != VCHR && 235 vap->va_type != VFIFO) 236 return EINVAL; 237 238 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); 239} 240 241/* --------------------------------------------------------------------- */ 242 243static int 244tmpfs_open(struct vop_open_args *v) 245{ 246 struct vnode *vp = v->a_vp; 247 int mode = v->a_mode; 248 249 int error; 250 struct tmpfs_node *node; 251 252 MPASS(VOP_ISLOCKED(vp)); 253 254 node = VP_TO_TMPFS_NODE(vp); 255 256 /* The file is still active but all its names have been removed 257 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as 258 * it is about to die. */ 259 if (node->tn_links < 1) 260 return (ENOENT); 261 262 /* If the file is marked append-only, deny write requests. */ 263 if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE) 264 error = EPERM; 265 else { 266 error = 0; 267 vnode_create_vobject(vp, node->tn_size, v->a_td); 268 } 269 270 MPASS(VOP_ISLOCKED(vp)); 271 return error; 272} 273 274/* --------------------------------------------------------------------- */ 275 276static int 277tmpfs_close(struct vop_close_args *v) 278{ 279 struct vnode *vp = v->a_vp; 280 281 MPASS(VOP_ISLOCKED(vp)); 282 283 /* Update node times. */ 284 tmpfs_update(vp); 285 286 return (0); 287} 288 289/* --------------------------------------------------------------------- */ 290 291int 292tmpfs_access(struct vop_access_args *v) 293{ 294 struct vnode *vp = v->a_vp; 295 accmode_t accmode = v->a_accmode; 296 struct ucred *cred = v->a_cred; 297 298 int error; 299 struct tmpfs_node *node; 300 301 MPASS(VOP_ISLOCKED(vp)); 302 303 node = VP_TO_TMPFS_NODE(vp); 304 305 switch (vp->v_type) { 306 case VDIR: 307 /* FALLTHROUGH */ 308 case VLNK: 309 /* FALLTHROUGH */ 310 case VREG: 311 if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) { 312 error = EROFS; 313 goto out; 314 } 315 break; 316 317 case VBLK: 318 /* FALLTHROUGH */ 319 case VCHR: 320 /* FALLTHROUGH */ 321 case VSOCK: 322 /* FALLTHROUGH */ 323 case VFIFO: 324 break; 325 326 default: 327 error = EINVAL; 328 goto out; 329 } 330 331 if (accmode & VWRITE && node->tn_flags & IMMUTABLE) { 332 error = EPERM; 333 goto out; 334 } 335 336 error = vaccess(vp->v_type, node->tn_mode, node->tn_uid, 337 node->tn_gid, accmode, cred, NULL); 338 339out: 340 MPASS(VOP_ISLOCKED(vp)); 341 342 return error; 343} 344 345/* --------------------------------------------------------------------- */ 346 347int 348tmpfs_getattr(struct vop_getattr_args *v) 349{ 350 struct vnode *vp = v->a_vp; 351 struct vattr *vap = v->a_vap; 352 353 struct tmpfs_node *node; 354 355 node = VP_TO_TMPFS_NODE(vp); 356 357 tmpfs_update(vp); 358 359 vap->va_type = vp->v_type; 360 vap->va_mode = node->tn_mode; 361 vap->va_nlink = node->tn_links; 362 vap->va_uid = node->tn_uid; 363 vap->va_gid = node->tn_gid; 364 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 365 vap->va_fileid = node->tn_id; 366 vap->va_size = node->tn_size; 367 vap->va_blocksize = PAGE_SIZE; 368 vap->va_atime = node->tn_atime; 369 vap->va_mtime = node->tn_mtime; 370 vap->va_ctime = node->tn_ctime; 371 vap->va_birthtime = node->tn_birthtime; 372 vap->va_gen = node->tn_gen; 373 vap->va_flags = node->tn_flags; 374 vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ? 375 node->tn_rdev : NODEV; 376 vap->va_bytes = round_page(node->tn_size); 377 vap->va_filerev = 0; 378 379 return 0; 380} 381 382/* --------------------------------------------------------------------- */ 383 384/* XXX Should this operation be atomic? I think it should, but code in 385 * XXX other places (e.g., ufs) doesn't seem to be... */ 386int 387tmpfs_setattr(struct vop_setattr_args *v) 388{ 389 struct vnode *vp = v->a_vp; 390 struct vattr *vap = v->a_vap; 391 struct ucred *cred = v->a_cred; 392 struct thread *td = curthread; 393 394 int error; 395 396 MPASS(VOP_ISLOCKED(vp)); 397 398 error = 0; 399 400 /* Abort if any unsettable attribute is given. */ 401 if (vap->va_type != VNON || 402 vap->va_nlink != VNOVAL || 403 vap->va_fsid != VNOVAL || 404 vap->va_fileid != VNOVAL || 405 vap->va_blocksize != VNOVAL || 406 vap->va_gen != VNOVAL || 407 vap->va_rdev != VNOVAL || 408 vap->va_bytes != VNOVAL) 409 error = EINVAL; 410 411 if (error == 0 && (vap->va_flags != VNOVAL)) 412 error = tmpfs_chflags(vp, vap->va_flags, cred, td); 413 414 if (error == 0 && (vap->va_size != VNOVAL)) 415 error = tmpfs_chsize(vp, vap->va_size, cred, td); 416 417 if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL)) 418 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td); 419 420 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) 421 error = tmpfs_chmod(vp, vap->va_mode, cred, td); 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, td); 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)); 438 439 return error; 440} 441 442/* --------------------------------------------------------------------- */ 443static int 444tmpfs_nocacheread(vm_object_t tobj, vm_pindex_t idx, 445 vm_offset_t offset, size_t tlen, struct uio *uio) 446{ 447 vm_page_t m; 448 int error; 449 450 VM_OBJECT_LOCK(tobj); 451 vm_object_pip_add(tobj, 1); 452 m = vm_page_grab(tobj, idx, VM_ALLOC_WIRED | 453 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 454 if (m->valid != VM_PAGE_BITS_ALL) { 455 if (vm_pager_has_page(tobj, idx, NULL, NULL)) { 456 error = vm_pager_get_pages(tobj, &m, 1, 0); 457 if (error != 0) { 458 printf("tmpfs get pages from pager error [read]\n"); 459 goto out; 460 } 461 } else 462 vm_page_zero_invalid(m, TRUE); 463 } 464 VM_OBJECT_UNLOCK(tobj); 465 error = uiomove_fromphys(&m, offset, tlen, uio); 466 VM_OBJECT_LOCK(tobj); 467out: 468 vm_page_lock(m); 469 vm_page_unwire(m, TRUE); 470 vm_page_unlock(m); 471 vm_page_wakeup(m); 472 vm_object_pip_subtract(tobj, 1); 473 VM_OBJECT_UNLOCK(tobj); 474 475 return (error); 476} 477 478static __inline int 479tmpfs_nocacheread_buf(vm_object_t tobj, vm_pindex_t idx, 480 vm_offset_t offset, size_t tlen, void *buf) 481{ 482 struct uio uio; 483 struct iovec iov; 484 485 uio.uio_iovcnt = 1; 486 uio.uio_iov = &iov; 487 iov.iov_base = buf; 488 iov.iov_len = tlen; 489 490 uio.uio_offset = 0; 491 uio.uio_resid = tlen; 492 uio.uio_rw = UIO_READ; 493 uio.uio_segflg = UIO_SYSSPACE; 494 uio.uio_td = curthread; 495 496 return (tmpfs_nocacheread(tobj, idx, offset, tlen, &uio)); 497} 498 499static int 500tmpfs_mappedread(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio) 501{ 502 struct sf_buf *sf; 503 vm_pindex_t idx; 504 vm_page_t m; 505 vm_offset_t offset; 506 off_t addr; 507 size_t tlen; 508 char *ma; 509 int error; 510 511 addr = uio->uio_offset; 512 idx = OFF_TO_IDX(addr); 513 offset = addr & PAGE_MASK; 514 tlen = MIN(PAGE_SIZE - offset, len); 515 516 if ((vobj == NULL) || 517 (vobj->resident_page_count == 0 && vobj->cache == NULL)) 518 goto nocache; 519 520 VM_OBJECT_LOCK(vobj); 521lookupvpg: 522 if (((m = vm_page_lookup(vobj, idx)) != NULL) && 523 vm_page_is_valid(m, offset, tlen)) { 524 if ((m->oflags & VPO_BUSY) != 0) { 525 /* 526 * Reference the page before unlocking and sleeping so 527 * that the page daemon is less likely to reclaim it. 528 */ 529 vm_page_reference(m); 530 vm_page_sleep(m, "tmfsmr"); 531 goto lookupvpg; 532 } 533 vm_page_busy(m); 534 VM_OBJECT_UNLOCK(vobj); 535 error = uiomove_fromphys(&m, offset, tlen, uio); 536 VM_OBJECT_LOCK(vobj); 537 vm_page_wakeup(m); 538 VM_OBJECT_UNLOCK(vobj); 539 return (error); 540 } else if (m != NULL && uio->uio_segflg == UIO_NOCOPY) { 541 KASSERT(offset == 0, 542 ("unexpected offset in tmpfs_mappedread for sendfile")); 543 if ((m->oflags & VPO_BUSY) != 0) { 544 /* 545 * Reference the page before unlocking and sleeping so 546 * that the page daemon is less likely to reclaim it. 547 */ 548 vm_page_reference(m); 549 vm_page_sleep(m, "tmfsmr"); 550 goto lookupvpg; 551 } 552 vm_page_busy(m); 553 VM_OBJECT_UNLOCK(vobj); 554 sched_pin(); 555 sf = sf_buf_alloc(m, SFB_CPUPRIVATE); 556 ma = (char *)sf_buf_kva(sf); 557 error = tmpfs_nocacheread_buf(tobj, idx, 0, tlen, ma); 558 if (error == 0) { 559 if (tlen != PAGE_SIZE) 560 bzero(ma + tlen, PAGE_SIZE - tlen); 561 uio->uio_offset += tlen; 562 uio->uio_resid -= tlen; 563 } 564 sf_buf_free(sf); 565 sched_unpin(); 566 VM_OBJECT_LOCK(vobj); 567 if (error == 0) 568 m->valid = VM_PAGE_BITS_ALL; 569 vm_page_wakeup(m); 570 VM_OBJECT_UNLOCK(vobj); 571 return (error); 572 } 573 VM_OBJECT_UNLOCK(vobj); 574nocache: 575 error = tmpfs_nocacheread(tobj, idx, offset, tlen, uio); 576 577 return (error); 578} 579 580static int 581tmpfs_read(struct vop_read_args *v) 582{ 583 struct vnode *vp = v->a_vp; 584 struct uio *uio = v->a_uio; 585 586 struct tmpfs_node *node; 587 vm_object_t uobj; 588 size_t len; 589 int resid; 590 591 int error = 0; 592 593 node = VP_TO_TMPFS_NODE(vp); 594 595 if (vp->v_type != VREG) { 596 error = EISDIR; 597 goto out; 598 } 599 600 if (uio->uio_offset < 0) { 601 error = EINVAL; 602 goto out; 603 } 604 605 node->tn_status |= TMPFS_NODE_ACCESSED; 606 607 uobj = node->tn_reg.tn_aobj; 608 while ((resid = uio->uio_resid) > 0) { 609 error = 0; 610 if (node->tn_size <= uio->uio_offset) 611 break; 612 len = MIN(node->tn_size - uio->uio_offset, resid); 613 if (len == 0) 614 break; 615 error = tmpfs_mappedread(vp->v_object, uobj, len, uio); 616 if ((error != 0) || (resid == uio->uio_resid)) 617 break; 618 } 619 620out: 621 622 return error; 623} 624 625/* --------------------------------------------------------------------- */ 626 627static int 628tmpfs_mappedwrite(vm_object_t vobj, vm_object_t tobj, size_t len, struct uio *uio) 629{ 630 vm_pindex_t idx; 631 vm_page_t vpg, tpg; 632 vm_offset_t offset; 633 off_t addr; 634 size_t tlen; 635 int error; 636 637 error = 0; 638 639 addr = uio->uio_offset; 640 idx = OFF_TO_IDX(addr); 641 offset = addr & PAGE_MASK; 642 tlen = MIN(PAGE_SIZE - offset, len); 643 644 if ((vobj == NULL) || 645 (vobj->resident_page_count == 0 && vobj->cache == NULL)) { 646 vpg = NULL; 647 goto nocache; 648 } 649 650 VM_OBJECT_LOCK(vobj); 651lookupvpg: 652 if (((vpg = vm_page_lookup(vobj, idx)) != NULL) && 653 vm_page_is_valid(vpg, offset, tlen)) { 654 if ((vpg->oflags & VPO_BUSY) != 0) { 655 /* 656 * Reference the page before unlocking and sleeping so 657 * that the page daemon is less likely to reclaim it. 658 */ 659 vm_page_reference(vpg); 660 vm_page_sleep(vpg, "tmfsmw"); 661 goto lookupvpg; 662 } 663 vm_page_busy(vpg); 664 vm_page_undirty(vpg); 665 VM_OBJECT_UNLOCK(vobj); 666 error = uiomove_fromphys(&vpg, offset, tlen, uio); 667 } else { 668 if (__predict_false(vobj->cache != NULL)) 669 vm_page_cache_free(vobj, idx, idx + 1); 670 VM_OBJECT_UNLOCK(vobj); 671 vpg = NULL; 672 } 673nocache: 674 VM_OBJECT_LOCK(tobj); 675 vm_object_pip_add(tobj, 1); 676 tpg = vm_page_grab(tobj, idx, VM_ALLOC_WIRED | 677 VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 678 if (tpg->valid != VM_PAGE_BITS_ALL) { 679 if (vm_pager_has_page(tobj, idx, NULL, NULL)) { 680 error = vm_pager_get_pages(tobj, &tpg, 1, 0); 681 if (error != 0) { 682 printf("tmpfs get pages from pager error [write]\n"); 683 goto out; 684 } 685 } else 686 vm_page_zero_invalid(tpg, TRUE); 687 } 688 VM_OBJECT_UNLOCK(tobj); 689 if (vpg == NULL) 690 error = uiomove_fromphys(&tpg, offset, tlen, uio); 691 else { 692 KASSERT(vpg->valid == VM_PAGE_BITS_ALL, ("parts of vpg invalid")); 693 pmap_copy_page(vpg, tpg); 694 } 695 VM_OBJECT_LOCK(tobj); 696out: 697 if (vobj != NULL) 698 VM_OBJECT_LOCK(vobj); 699 if (error == 0) { 700 KASSERT(tpg->valid == VM_PAGE_BITS_ALL, 701 ("parts of tpg invalid")); 702 vm_page_dirty(tpg); 703 } 704 vm_page_lock(tpg); 705 vm_page_unwire(tpg, TRUE); 706 vm_page_unlock(tpg); 707 vm_page_wakeup(tpg); 708 if (vpg != NULL) 709 vm_page_wakeup(vpg); 710 if (vobj != NULL) 711 VM_OBJECT_UNLOCK(vobj); 712 vm_object_pip_subtract(tobj, 1); 713 VM_OBJECT_UNLOCK(tobj); 714 715 return (error); 716} 717 718static int 719tmpfs_write(struct vop_write_args *v) 720{ 721 struct vnode *vp = v->a_vp; 722 struct uio *uio = v->a_uio; 723 int ioflag = v->a_ioflag; 724 725 boolean_t extended; 726 int error = 0; 727 off_t oldsize; 728 struct tmpfs_node *node; 729 vm_object_t uobj; 730 size_t len; 731 int resid; 732 733 node = VP_TO_TMPFS_NODE(vp); 734 oldsize = node->tn_size; 735 736 if (uio->uio_offset < 0 || vp->v_type != VREG) { 737 error = EINVAL; 738 goto out; 739 } 740 741 if (uio->uio_resid == 0) { 742 error = 0; 743 goto out; 744 } 745 746 if (ioflag & IO_APPEND) 747 uio->uio_offset = node->tn_size; 748 749 if (uio->uio_offset + uio->uio_resid > 750 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) 751 return (EFBIG); 752 753 if (vn_rlimit_fsize(vp, uio, uio->uio_td)) 754 return (EFBIG); 755 756 extended = uio->uio_offset + uio->uio_resid > node->tn_size; 757 if (extended) { 758 error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid); 759 if (error != 0) 760 goto out; 761 } 762 763 uobj = node->tn_reg.tn_aobj; 764 while ((resid = uio->uio_resid) > 0) { 765 if (node->tn_size <= uio->uio_offset) 766 break; 767 len = MIN(node->tn_size - uio->uio_offset, resid); 768 if (len == 0) 769 break; 770 error = tmpfs_mappedwrite(vp->v_object, uobj, len, uio); 771 if ((error != 0) || (resid == uio->uio_resid)) 772 break; 773 } 774 775 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | 776 (extended ? TMPFS_NODE_CHANGED : 0); 777 778 if (node->tn_mode & (S_ISUID | S_ISGID)) { 779 if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0)) 780 node->tn_mode &= ~(S_ISUID | S_ISGID); 781 } 782 783 if (error != 0) 784 (void)tmpfs_reg_resize(vp, oldsize); 785 786out: 787 MPASS(IMPLIES(error == 0, uio->uio_resid == 0)); 788 MPASS(IMPLIES(error != 0, oldsize == node->tn_size)); 789 790 return error; 791} 792 793/* --------------------------------------------------------------------- */ 794 795static int 796tmpfs_fsync(struct vop_fsync_args *v) 797{ 798 struct vnode *vp = v->a_vp; 799 800 MPASS(VOP_ISLOCKED(vp)); 801 802 tmpfs_update(vp); 803 804 return 0; 805} 806 807/* --------------------------------------------------------------------- */ 808 809static int 810tmpfs_remove(struct vop_remove_args *v) 811{ 812 struct vnode *dvp = v->a_dvp; 813 struct vnode *vp = v->a_vp; 814 815 int error; 816 struct tmpfs_dirent *de; 817 struct tmpfs_mount *tmp; 818 struct tmpfs_node *dnode; 819 struct tmpfs_node *node; 820 821 MPASS(VOP_ISLOCKED(dvp)); 822 MPASS(VOP_ISLOCKED(vp)); 823 824 if (vp->v_type == VDIR) { 825 error = EISDIR; 826 goto out; 827 } 828 829 dnode = VP_TO_TMPFS_DIR(dvp); 830 node = VP_TO_TMPFS_NODE(vp); 831 tmp = VFS_TO_TMPFS(vp->v_mount); 832 de = tmpfs_dir_lookup(dnode, node, v->a_cnp); 833 MPASS(de != NULL); 834 835 /* Files marked as immutable or append-only cannot be deleted. */ 836 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || 837 (dnode->tn_flags & APPEND)) { 838 error = EPERM; 839 goto out; 840 } 841 842 /* Remove the entry from the directory; as it is a file, we do not 843 * have to change the number of hard links of the directory. */ 844 tmpfs_dir_detach(dvp, de); 845 if (v->a_cnp->cn_flags & DOWHITEOUT) 846 tmpfs_dir_whiteout_add(dvp, v->a_cnp); 847 848 /* Free the directory entry we just deleted. Note that the node 849 * referred by it will not be removed until the vnode is really 850 * reclaimed. */ 851 tmpfs_free_dirent(tmp, de, TRUE); 852 853 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED; 854 error = 0; 855 856out: 857 858 return error; 859} 860 861/* --------------------------------------------------------------------- */ 862 863static int 864tmpfs_link(struct vop_link_args *v) 865{ 866 struct vnode *dvp = v->a_tdvp; 867 struct vnode *vp = v->a_vp; 868 struct componentname *cnp = v->a_cnp; 869 870 int error; 871 struct tmpfs_dirent *de; 872 struct tmpfs_node *node; 873 874 MPASS(VOP_ISLOCKED(dvp)); 875 MPASS(cnp->cn_flags & HASBUF); 876 MPASS(dvp != vp); /* XXX When can this be false? */ 877 878 node = VP_TO_TMPFS_NODE(vp); 879 880 /* XXX: Why aren't the following two tests done by the caller? */ 881 882 /* Hard links of directories are forbidden. */ 883 if (vp->v_type == VDIR) { 884 error = EPERM; 885 goto out; 886 } 887 888 /* Cannot create cross-device links. */ 889 if (dvp->v_mount != vp->v_mount) { 890 error = EXDEV; 891 goto out; 892 } 893 894 /* Ensure that we do not overflow the maximum number of links imposed 895 * by the system. */ 896 MPASS(node->tn_links <= LINK_MAX); 897 if (node->tn_links == LINK_MAX) { 898 error = EMLINK; 899 goto out; 900 } 901 902 /* We cannot create links of files marked immutable or append-only. */ 903 if (node->tn_flags & (IMMUTABLE | APPEND)) { 904 error = EPERM; 905 goto out; 906 } 907 908 /* Allocate a new directory entry to represent the node. */ 909 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, 910 cnp->cn_nameptr, cnp->cn_namelen, &de); 911 if (error != 0) 912 goto out; 913 914 /* Insert the new directory entry into the appropriate directory. */ 915 if (cnp->cn_flags & ISWHITEOUT) 916 tmpfs_dir_whiteout_remove(dvp, cnp); 917 tmpfs_dir_attach(dvp, de); 918 919 /* vp link count has changed, so update node times. */ 920 node->tn_status |= TMPFS_NODE_CHANGED; 921 tmpfs_update(vp); 922 923 error = 0; 924 925out: 926 return error; 927} 928 929/* --------------------------------------------------------------------- */ 930 931/* 932 * We acquire all but fdvp locks using non-blocking acquisitions. If we 933 * fail to acquire any lock in the path we will drop all held locks, 934 * acquire the new lock in a blocking fashion, and then release it and 935 * restart the rename. This acquire/release step ensures that we do not 936 * spin on a lock waiting for release. On error release all vnode locks 937 * and decrement references the way tmpfs_rename() would do. 938 */ 939static int 940tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp, 941 struct vnode *tdvp, struct vnode **tvpp, 942 struct componentname *fcnp, struct componentname *tcnp) 943{ 944 struct vnode *nvp; 945 struct mount *mp; 946 struct tmpfs_dirent *de; 947 int error, restarts = 0; 948 949 VOP_UNLOCK(tdvp, 0); 950 if (*tvpp != NULL && *tvpp != tdvp) 951 VOP_UNLOCK(*tvpp, 0); 952 mp = fdvp->v_mount; 953 954relock: 955 restarts += 1; 956 error = vn_lock(fdvp, LK_EXCLUSIVE); 957 if (error) 958 goto releout; 959 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 960 VOP_UNLOCK(fdvp, 0); 961 error = vn_lock(tdvp, LK_EXCLUSIVE); 962 if (error) 963 goto releout; 964 VOP_UNLOCK(tdvp, 0); 965 goto relock; 966 } 967 /* 968 * Re-resolve fvp to be certain it still exists and fetch the 969 * correct vnode. 970 */ 971 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp); 972 if (de == NULL) { 973 VOP_UNLOCK(fdvp, 0); 974 VOP_UNLOCK(tdvp, 0); 975 if ((fcnp->cn_flags & ISDOTDOT) != 0 || 976 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) 977 error = EINVAL; 978 else 979 error = ENOENT; 980 goto releout; 981 } 982 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp); 983 if (error != 0) { 984 VOP_UNLOCK(fdvp, 0); 985 VOP_UNLOCK(tdvp, 0); 986 if (error != EBUSY) 987 goto releout; 988 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp); 989 if (error != 0) 990 goto releout; 991 VOP_UNLOCK(nvp, 0); 992 /* 993 * Concurrent rename race. 994 */ 995 if (nvp == tdvp) { 996 vrele(nvp); 997 error = EINVAL; 998 goto releout; 999 } 1000 vrele(*fvpp); 1001 *fvpp = nvp; 1002 goto relock; 1003 } 1004 vrele(*fvpp); 1005 *fvpp = nvp; 1006 VOP_UNLOCK(*fvpp, 0); 1007 /* 1008 * Re-resolve tvp and acquire the vnode lock if present. 1009 */ 1010 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp); 1011 /* 1012 * If tvp disappeared we just carry on. 1013 */ 1014 if (de == NULL && *tvpp != NULL) { 1015 vrele(*tvpp); 1016 *tvpp = NULL; 1017 } 1018 /* 1019 * Get the tvp ino if the lookup succeeded. We may have to restart 1020 * if the non-blocking acquire fails. 1021 */ 1022 if (de != NULL) { 1023 nvp = NULL; 1024 error = tmpfs_alloc_vp(mp, de->td_node, 1025 LK_EXCLUSIVE | LK_NOWAIT, &nvp); 1026 if (*tvpp != NULL) 1027 vrele(*tvpp); 1028 *tvpp = nvp; 1029 if (error != 0) { 1030 VOP_UNLOCK(fdvp, 0); 1031 VOP_UNLOCK(tdvp, 0); 1032 if (error != EBUSY) 1033 goto releout; 1034 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, 1035 &nvp); 1036 if (error != 0) 1037 goto releout; 1038 VOP_UNLOCK(nvp, 0); 1039 /* 1040 * fdvp contains fvp, thus tvp (=fdvp) is not empty. 1041 */ 1042 if (nvp == fdvp) { 1043 error = ENOTEMPTY; 1044 goto releout; 1045 } 1046 goto relock; 1047 } 1048 } 1049 tmpfs_rename_restarts += restarts; 1050 1051 return (0); 1052 1053releout: 1054 vrele(fdvp); 1055 vrele(*fvpp); 1056 vrele(tdvp); 1057 if (*tvpp != NULL) 1058 vrele(*tvpp); 1059 tmpfs_rename_restarts += restarts; 1060 1061 return (error); 1062} 1063 1064static int 1065tmpfs_rename(struct vop_rename_args *v) 1066{ 1067 struct vnode *fdvp = v->a_fdvp; 1068 struct vnode *fvp = v->a_fvp; 1069 struct componentname *fcnp = v->a_fcnp; 1070 struct vnode *tdvp = v->a_tdvp; 1071 struct vnode *tvp = v->a_tvp; 1072 struct componentname *tcnp = v->a_tcnp; 1073 struct mount *mp = NULL; 1074 1075 char *newname; 1076 int error; 1077 struct tmpfs_dirent *de; 1078 struct tmpfs_mount *tmp; 1079 struct tmpfs_node *fdnode; 1080 struct tmpfs_node *fnode; 1081 struct tmpfs_node *tnode; 1082 struct tmpfs_node *tdnode; 1083 1084 MPASS(VOP_ISLOCKED(tdvp)); 1085 MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp))); 1086 MPASS(fcnp->cn_flags & HASBUF); 1087 MPASS(tcnp->cn_flags & HASBUF); 1088 1089 /* Disallow cross-device renames. 1090 * XXX Why isn't this done by the caller? */ 1091 if (fvp->v_mount != tdvp->v_mount || 1092 (tvp != NULL && fvp->v_mount != tvp->v_mount)) { 1093 error = EXDEV; 1094 goto out; 1095 } 1096 1097 /* If source and target are the same file, there is nothing to do. */ 1098 if (fvp == tvp) { 1099 error = 0; 1100 goto out; 1101 } 1102 1103 /* If we need to move the directory between entries, lock the 1104 * source so that we can safely operate on it. */ 1105 if (fdvp != tdvp && fdvp != tvp) { 1106 if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 1107 mp = tdvp->v_mount; 1108 error = vfs_busy(mp, 0); 1109 if (error != 0) { 1110 mp = NULL; 1111 goto out; 1112 } 1113 error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp, 1114 fcnp, tcnp); 1115 if (error != 0) { 1116 vfs_unbusy(mp); 1117 return (error); 1118 } 1119 ASSERT_VOP_ELOCKED(fdvp, 1120 "tmpfs_rename: fdvp not locked"); 1121 ASSERT_VOP_ELOCKED(tdvp, 1122 "tmpfs_rename: tdvp not locked"); 1123 if (tvp != NULL) 1124 ASSERT_VOP_ELOCKED(tvp, 1125 "tmpfs_rename: tvp not locked"); 1126 if (fvp == tvp) { 1127 error = 0; 1128 goto out_locked; 1129 } 1130 } 1131 } 1132 1133 tmp = VFS_TO_TMPFS(tdvp->v_mount); 1134 tdnode = VP_TO_TMPFS_DIR(tdvp); 1135 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp); 1136 fdnode = VP_TO_TMPFS_DIR(fdvp); 1137 fnode = VP_TO_TMPFS_NODE(fvp); 1138 de = tmpfs_dir_lookup(fdnode, fnode, fcnp); 1139 1140 /* Entry can disappear before we lock fdvp, 1141 * also avoid manipulating '.' and '..' entries. */ 1142 if (de == NULL) { 1143 if ((fcnp->cn_flags & ISDOTDOT) != 0 || 1144 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.')) 1145 error = EINVAL; 1146 else 1147 error = ENOENT; 1148 goto out_locked; 1149 } 1150 MPASS(de->td_node == fnode); 1151 1152 /* If re-naming a directory to another preexisting directory 1153 * ensure that the target directory is empty so that its 1154 * removal causes no side effects. 1155 * Kern_rename gurantees the destination to be a directory 1156 * if the source is one. */ 1157 if (tvp != NULL) { 1158 MPASS(tnode != NULL); 1159 1160 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1161 (tdnode->tn_flags & (APPEND | IMMUTABLE))) { 1162 error = EPERM; 1163 goto out_locked; 1164 } 1165 1166 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) { 1167 if (tnode->tn_size > 0) { 1168 error = ENOTEMPTY; 1169 goto out_locked; 1170 } 1171 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) { 1172 error = ENOTDIR; 1173 goto out_locked; 1174 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) { 1175 error = EISDIR; 1176 goto out_locked; 1177 } else { 1178 MPASS(fnode->tn_type != VDIR && 1179 tnode->tn_type != VDIR); 1180 } 1181 } 1182 1183 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) 1184 || (fdnode->tn_flags & (APPEND | IMMUTABLE))) { 1185 error = EPERM; 1186 goto out_locked; 1187 } 1188 1189 /* Ensure that we have enough memory to hold the new name, if it 1190 * has to be changed. */ 1191 if (fcnp->cn_namelen != tcnp->cn_namelen || 1192 bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) { 1193 newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK); 1194 } else 1195 newname = NULL; 1196 1197 /* If the node is being moved to another directory, we have to do 1198 * the move. */ 1199 if (fdnode != tdnode) { 1200 /* In case we are moving a directory, we have to adjust its 1201 * parent to point to the new parent. */ 1202 if (de->td_node->tn_type == VDIR) { 1203 struct tmpfs_node *n; 1204 1205 /* Ensure the target directory is not a child of the 1206 * directory being moved. Otherwise, we'd end up 1207 * with stale nodes. */ 1208 n = tdnode; 1209 /* TMPFS_LOCK garanties that no nodes are freed while 1210 * traversing the list. Nodes can only be marked as 1211 * removed: tn_parent == NULL. */ 1212 TMPFS_LOCK(tmp); 1213 TMPFS_NODE_LOCK(n); 1214 while (n != n->tn_dir.tn_parent) { 1215 struct tmpfs_node *parent; 1216 1217 if (n == fnode) { 1218 TMPFS_NODE_UNLOCK(n); 1219 TMPFS_UNLOCK(tmp); 1220 error = EINVAL; 1221 if (newname != NULL) 1222 free(newname, M_TMPFSNAME); 1223 goto out_locked; 1224 } 1225 parent = n->tn_dir.tn_parent; 1226 TMPFS_NODE_UNLOCK(n); 1227 if (parent == NULL) { 1228 n = NULL; 1229 break; 1230 } 1231 TMPFS_NODE_LOCK(parent); 1232 if (parent->tn_dir.tn_parent == NULL) { 1233 TMPFS_NODE_UNLOCK(parent); 1234 n = NULL; 1235 break; 1236 } 1237 n = parent; 1238 } 1239 TMPFS_UNLOCK(tmp); 1240 if (n == NULL) { 1241 error = EINVAL; 1242 if (newname != NULL) 1243 free(newname, M_TMPFSNAME); 1244 goto out_locked; 1245 } 1246 TMPFS_NODE_UNLOCK(n); 1247 1248 /* Adjust the parent pointer. */ 1249 TMPFS_VALIDATE_DIR(fnode); 1250 TMPFS_NODE_LOCK(de->td_node); 1251 de->td_node->tn_dir.tn_parent = tdnode; 1252 TMPFS_NODE_UNLOCK(de->td_node); 1253 1254 /* As a result of changing the target of the '..' 1255 * entry, the link count of the source and target 1256 * directories has to be adjusted. */ 1257 TMPFS_NODE_LOCK(tdnode); 1258 TMPFS_ASSERT_LOCKED(tdnode); 1259 tdnode->tn_links++; 1260 TMPFS_NODE_UNLOCK(tdnode); 1261 1262 TMPFS_NODE_LOCK(fdnode); 1263 TMPFS_ASSERT_LOCKED(fdnode); 1264 fdnode->tn_links--; 1265 TMPFS_NODE_UNLOCK(fdnode); 1266 } 1267 1268 /* Do the move: just remove the entry from the source directory 1269 * and insert it into the target one. */ 1270 tmpfs_dir_detach(fdvp, de); 1271 if (fcnp->cn_flags & DOWHITEOUT) 1272 tmpfs_dir_whiteout_add(fdvp, fcnp); 1273 if (tcnp->cn_flags & ISWHITEOUT) 1274 tmpfs_dir_whiteout_remove(tdvp, tcnp); 1275 tmpfs_dir_attach(tdvp, de); 1276 } 1277 1278 /* If the name has changed, we need to make it effective by changing 1279 * it in the directory entry. */ 1280 if (newname != NULL) { 1281 MPASS(tcnp->cn_namelen <= MAXNAMLEN); 1282 1283 free(de->td_name, M_TMPFSNAME); 1284 de->td_namelen = (uint16_t)tcnp->cn_namelen; 1285 memcpy(newname, tcnp->cn_nameptr, tcnp->cn_namelen); 1286 de->td_name = newname; 1287 1288 fnode->tn_status |= TMPFS_NODE_CHANGED; 1289 tdnode->tn_status |= TMPFS_NODE_MODIFIED; 1290 } 1291 1292 /* If we are overwriting an entry, we have to remove the old one 1293 * from the target directory. */ 1294 if (tvp != NULL) { 1295 /* Remove the old entry from the target directory. */ 1296 de = tmpfs_dir_lookup(tdnode, tnode, tcnp); 1297 tmpfs_dir_detach(tdvp, de); 1298 1299 /* Free the directory entry we just deleted. Note that the 1300 * node referred by it will not be removed until the vnode is 1301 * really reclaimed. */ 1302 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de, TRUE); 1303 } 1304 cache_purge(fvp); 1305 if (tvp != NULL) 1306 cache_purge(tvp); 1307 1308 error = 0; 1309 1310out_locked: 1311 if (fdvp != tdvp && fdvp != tvp) 1312 VOP_UNLOCK(fdvp, 0); 1313 1314out: 1315 /* Release target nodes. */ 1316 /* XXX: I don't understand when tdvp can be the same as tvp, but 1317 * other code takes care of this... */ 1318 if (tdvp == tvp) 1319 vrele(tdvp); 1320 else 1321 vput(tdvp); 1322 if (tvp != NULL) 1323 vput(tvp); 1324 1325 /* Release source nodes. */ 1326 vrele(fdvp); 1327 vrele(fvp); 1328 1329 if (mp != NULL) 1330 vfs_unbusy(mp); 1331 1332 return error; 1333} 1334 1335/* --------------------------------------------------------------------- */ 1336 1337static int 1338tmpfs_mkdir(struct vop_mkdir_args *v) 1339{ 1340 struct vnode *dvp = v->a_dvp; 1341 struct vnode **vpp = v->a_vpp; 1342 struct componentname *cnp = v->a_cnp; 1343 struct vattr *vap = v->a_vap; 1344 1345 MPASS(vap->va_type == VDIR); 1346 1347 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL); 1348} 1349 1350/* --------------------------------------------------------------------- */ 1351 1352static int 1353tmpfs_rmdir(struct vop_rmdir_args *v) 1354{ 1355 struct vnode *dvp = v->a_dvp; 1356 struct vnode *vp = v->a_vp; 1357 1358 int error; 1359 struct tmpfs_dirent *de; 1360 struct tmpfs_mount *tmp; 1361 struct tmpfs_node *dnode; 1362 struct tmpfs_node *node; 1363 1364 MPASS(VOP_ISLOCKED(dvp)); 1365 MPASS(VOP_ISLOCKED(vp)); 1366 1367 tmp = VFS_TO_TMPFS(dvp->v_mount); 1368 dnode = VP_TO_TMPFS_DIR(dvp); 1369 node = VP_TO_TMPFS_DIR(vp); 1370 1371 /* Directories with more than two entries ('.' and '..') cannot be 1372 * removed. */ 1373 if (node->tn_size > 0) { 1374 error = ENOTEMPTY; 1375 goto out; 1376 } 1377 1378 if ((dnode->tn_flags & APPEND) 1379 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1380 error = EPERM; 1381 goto out; 1382 } 1383 1384 /* This invariant holds only if we are not trying to remove "..". 1385 * We checked for that above so this is safe now. */ 1386 MPASS(node->tn_dir.tn_parent == dnode); 1387 1388 /* Get the directory entry associated with node (vp). This was 1389 * filled by tmpfs_lookup while looking up the entry. */ 1390 de = tmpfs_dir_lookup(dnode, node, v->a_cnp); 1391 MPASS(TMPFS_DIRENT_MATCHES(de, 1392 v->a_cnp->cn_nameptr, 1393 v->a_cnp->cn_namelen)); 1394 1395 /* Check flags to see if we are allowed to remove the directory. */ 1396 if (dnode->tn_flags & APPEND 1397 || node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) { 1398 error = EPERM; 1399 goto out; 1400 } 1401 1402 1403 /* Detach the directory entry from the directory (dnode). */ 1404 tmpfs_dir_detach(dvp, de); 1405 if (v->a_cnp->cn_flags & DOWHITEOUT) 1406 tmpfs_dir_whiteout_add(dvp, v->a_cnp); 1407 1408 /* No vnode should be allocated for this entry from this point */ 1409 TMPFS_NODE_LOCK(node); 1410 TMPFS_ASSERT_ELOCKED(node); 1411 node->tn_links--; 1412 node->tn_dir.tn_parent = NULL; 1413 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 1414 TMPFS_NODE_MODIFIED; 1415 1416 TMPFS_NODE_UNLOCK(node); 1417 1418 TMPFS_NODE_LOCK(dnode); 1419 TMPFS_ASSERT_ELOCKED(dnode); 1420 dnode->tn_links--; 1421 dnode->tn_status |= TMPFS_NODE_ACCESSED | \ 1422 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED; 1423 TMPFS_NODE_UNLOCK(dnode); 1424 1425 cache_purge(dvp); 1426 cache_purge(vp); 1427 1428 /* Free the directory entry we just deleted. Note that the node 1429 * referred by it will not be removed until the vnode is really 1430 * reclaimed. */ 1431 tmpfs_free_dirent(tmp, de, TRUE); 1432 1433 /* Release the deleted vnode (will destroy the node, notify 1434 * interested parties and clean it from the cache). */ 1435 1436 dnode->tn_status |= TMPFS_NODE_CHANGED; 1437 tmpfs_update(dvp); 1438 1439 error = 0; 1440 1441out: 1442 return error; 1443} 1444 1445/* --------------------------------------------------------------------- */ 1446 1447static int 1448tmpfs_symlink(struct vop_symlink_args *v) 1449{ 1450 struct vnode *dvp = v->a_dvp; 1451 struct vnode **vpp = v->a_vpp; 1452 struct componentname *cnp = v->a_cnp; 1453 struct vattr *vap = v->a_vap; 1454 char *target = v->a_target; 1455 1456#ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */ 1457 MPASS(vap->va_type == VLNK); 1458#else 1459 vap->va_type = VLNK; 1460#endif 1461 1462 return tmpfs_alloc_file(dvp, vpp, vap, cnp, target); 1463} 1464 1465/* --------------------------------------------------------------------- */ 1466 1467static int 1468tmpfs_readdir(struct vop_readdir_args *v) 1469{ 1470 struct vnode *vp = v->a_vp; 1471 struct uio *uio = v->a_uio; 1472 int *eofflag = v->a_eofflag; 1473 u_long **cookies = v->a_cookies; 1474 int *ncookies = v->a_ncookies; 1475 1476 int error; 1477 off_t startoff; 1478 off_t cnt = 0; 1479 struct tmpfs_node *node; 1480 1481 /* This operation only makes sense on directory nodes. */ 1482 if (vp->v_type != VDIR) 1483 return ENOTDIR; 1484 1485 node = VP_TO_TMPFS_DIR(vp); 1486 1487 startoff = uio->uio_offset; 1488 1489 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) { 1490 error = tmpfs_dir_getdotdent(node, uio); 1491 if (error != 0) 1492 goto outok; 1493 cnt++; 1494 } 1495 1496 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) { 1497 error = tmpfs_dir_getdotdotdent(node, uio); 1498 if (error != 0) 1499 goto outok; 1500 cnt++; 1501 } 1502 1503 error = tmpfs_dir_getdents(node, uio, &cnt); 1504 1505outok: 1506 MPASS(error >= -1); 1507 1508 if (error == -1) 1509 error = (cnt != 0) ? 0 : EINVAL; 1510 1511 if (eofflag != NULL) 1512 *eofflag = 1513 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); 1514 1515 /* Update NFS-related variables. */ 1516 if (error == 0 && cookies != NULL && ncookies != NULL) { 1517 off_t i; 1518 off_t off = startoff; 1519 struct tmpfs_dirent *de = NULL; 1520 1521 *ncookies = cnt; 1522 *cookies = malloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK); 1523 1524 for (i = 0; i < cnt; i++) { 1525 MPASS(off != TMPFS_DIRCOOKIE_EOF); 1526 if (off == TMPFS_DIRCOOKIE_DOT) { 1527 off = TMPFS_DIRCOOKIE_DOTDOT; 1528 } else { 1529 if (off == TMPFS_DIRCOOKIE_DOTDOT) { 1530 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead); 1531 } else if (de != NULL) { 1532 de = TAILQ_NEXT(de, td_entries); 1533 } else { 1534 de = tmpfs_dir_lookupbycookie(node, 1535 off); 1536 MPASS(de != NULL); 1537 de = TAILQ_NEXT(de, td_entries); 1538 } 1539 if (de == NULL) 1540 off = TMPFS_DIRCOOKIE_EOF; 1541 else 1542 off = tmpfs_dircookie(de); 1543 } 1544 1545 (*cookies)[i] = off; 1546 } 1547 MPASS(uio->uio_offset == off); 1548 } 1549 1550 return error; 1551} 1552 1553/* --------------------------------------------------------------------- */ 1554 1555static int 1556tmpfs_readlink(struct vop_readlink_args *v) 1557{ 1558 struct vnode *vp = v->a_vp; 1559 struct uio *uio = v->a_uio; 1560 1561 int error; 1562 struct tmpfs_node *node; 1563 1564 MPASS(uio->uio_offset == 0); 1565 MPASS(vp->v_type == VLNK); 1566 1567 node = VP_TO_TMPFS_NODE(vp); 1568 1569 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid), 1570 uio); 1571 node->tn_status |= TMPFS_NODE_ACCESSED; 1572 1573 return error; 1574} 1575 1576/* --------------------------------------------------------------------- */ 1577 1578static int 1579tmpfs_inactive(struct vop_inactive_args *v) 1580{ 1581 struct vnode *vp = v->a_vp; 1582 struct thread *l = v->a_td; 1583 1584 struct tmpfs_node *node; 1585 1586 MPASS(VOP_ISLOCKED(vp)); 1587 1588 node = VP_TO_TMPFS_NODE(vp); 1589 1590 if (node->tn_links == 0) 1591 vrecycle(vp, l); 1592 1593 return 0; 1594} 1595 1596/* --------------------------------------------------------------------- */ 1597 1598int 1599tmpfs_reclaim(struct vop_reclaim_args *v) 1600{ 1601 struct vnode *vp = v->a_vp; 1602 1603 struct tmpfs_mount *tmp; 1604 struct tmpfs_node *node; 1605 1606 node = VP_TO_TMPFS_NODE(vp); 1607 tmp = VFS_TO_TMPFS(vp->v_mount); 1608 1609 vnode_destroy_vobject(vp); 1610 cache_purge(vp); 1611 1612 TMPFS_NODE_LOCK(node); 1613 TMPFS_ASSERT_ELOCKED(node); 1614 tmpfs_free_vp(vp); 1615 1616 /* If the node referenced by this vnode was deleted by the user, 1617 * we must free its associated data structures (now that the vnode 1618 * is being reclaimed). */ 1619 if (node->tn_links == 0 && 1620 (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) { 1621 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1622 TMPFS_NODE_UNLOCK(node); 1623 tmpfs_free_node(tmp, node); 1624 } else 1625 TMPFS_NODE_UNLOCK(node); 1626 1627 MPASS(vp->v_data == NULL); 1628 return 0; 1629} 1630 1631/* --------------------------------------------------------------------- */ 1632 1633static int 1634tmpfs_print(struct vop_print_args *v) 1635{ 1636 struct vnode *vp = v->a_vp; 1637 1638 struct tmpfs_node *node; 1639 1640 node = VP_TO_TMPFS_NODE(vp); 1641 1642 printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n", 1643 node, node->tn_flags, node->tn_links); 1644 printf("\tmode 0%o, owner %d, group %d, size %" PRIdMAX 1645 ", status 0x%x\n", 1646 node->tn_mode, node->tn_uid, node->tn_gid, 1647 (uintmax_t)node->tn_size, node->tn_status); 1648 1649 if (vp->v_type == VFIFO) 1650 fifo_printinfo(vp); 1651 1652 printf("\n"); 1653 1654 return 0; 1655} 1656 1657/* --------------------------------------------------------------------- */ 1658 1659static int 1660tmpfs_pathconf(struct vop_pathconf_args *v) 1661{ 1662 int name = v->a_name; 1663 register_t *retval = v->a_retval; 1664 1665 int error; 1666 1667 error = 0; 1668 1669 switch (name) { 1670 case _PC_LINK_MAX: 1671 *retval = LINK_MAX; 1672 break; 1673 1674 case _PC_NAME_MAX: 1675 *retval = NAME_MAX; 1676 break; 1677 1678 case _PC_PATH_MAX: 1679 *retval = PATH_MAX; 1680 break; 1681 1682 case _PC_PIPE_BUF: 1683 *retval = PIPE_BUF; 1684 break; 1685 1686 case _PC_CHOWN_RESTRICTED: 1687 *retval = 1; 1688 break; 1689 1690 case _PC_NO_TRUNC: 1691 *retval = 1; 1692 break; 1693 1694 case _PC_SYNC_IO: 1695 *retval = 1; 1696 break; 1697 1698 case _PC_FILESIZEBITS: 1699 *retval = 0; /* XXX Don't know which value should I return. */ 1700 break; 1701 1702 default: 1703 error = EINVAL; 1704 } 1705 1706 return error; 1707} 1708 1709static int 1710tmpfs_vptofh(struct vop_vptofh_args *ap) 1711{ 1712 struct tmpfs_fid *tfhp; 1713 struct tmpfs_node *node; 1714 1715 tfhp = (struct tmpfs_fid *)ap->a_fhp; 1716 node = VP_TO_TMPFS_NODE(ap->a_vp); 1717 1718 tfhp->tf_len = sizeof(struct tmpfs_fid); 1719 tfhp->tf_id = node->tn_id; 1720 tfhp->tf_gen = node->tn_gen; 1721 1722 return (0); 1723} 1724 1725static int 1726tmpfs_whiteout(struct vop_whiteout_args *ap) 1727{ 1728 struct vnode *dvp = ap->a_dvp; 1729 struct componentname *cnp = ap->a_cnp; 1730 struct tmpfs_dirent *de; 1731 1732 switch (ap->a_flags) { 1733 case LOOKUP: 1734 return (0); 1735 case CREATE: 1736 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp); 1737 if (de != NULL) 1738 return (de->td_node == NULL ? 0 : EEXIST); 1739 return (tmpfs_dir_whiteout_add(dvp, cnp)); 1740 case DELETE: 1741 tmpfs_dir_whiteout_remove(dvp, cnp); 1742 return (0); 1743 default: 1744 panic("tmpfs_whiteout: unknown op"); 1745 } 1746} 1747 1748/* --------------------------------------------------------------------- */ 1749 1750/* 1751 * vnode operations vector used for files stored in a tmpfs file system. 1752 */ 1753struct vop_vector tmpfs_vnodeop_entries = { 1754 .vop_default = &default_vnodeops, 1755 .vop_lookup = vfs_cache_lookup, 1756 .vop_cachedlookup = tmpfs_lookup, 1757 .vop_create = tmpfs_create, 1758 .vop_mknod = tmpfs_mknod, 1759 .vop_open = tmpfs_open, 1760 .vop_close = tmpfs_close, 1761 .vop_access = tmpfs_access, 1762 .vop_getattr = tmpfs_getattr, 1763 .vop_setattr = tmpfs_setattr, 1764 .vop_read = tmpfs_read, 1765 .vop_write = tmpfs_write, 1766 .vop_fsync = tmpfs_fsync, 1767 .vop_remove = tmpfs_remove, 1768 .vop_link = tmpfs_link, 1769 .vop_rename = tmpfs_rename, 1770 .vop_mkdir = tmpfs_mkdir, 1771 .vop_rmdir = tmpfs_rmdir, 1772 .vop_symlink = tmpfs_symlink, 1773 .vop_readdir = tmpfs_readdir, 1774 .vop_readlink = tmpfs_readlink, 1775 .vop_inactive = tmpfs_inactive, 1776 .vop_reclaim = tmpfs_reclaim, 1777 .vop_print = tmpfs_print, 1778 .vop_pathconf = tmpfs_pathconf, 1779 .vop_vptofh = tmpfs_vptofh, 1780 .vop_whiteout = tmpfs_whiteout, 1781 .vop_bmap = VOP_EOPNOTSUPP, 1782}; 1783 1784