autofs_vnops.c revision 296918
1/*- 2 * Copyright (c) 2014 The FreeBSD Foundation 3 * All rights reserved. 4 * 5 * This software was developed by Edward Tomasz Napierala under sponsorship 6 * from the FreeBSD Foundation. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 */ 30 31#include <sys/cdefs.h> 32__FBSDID("$FreeBSD: head/sys/fs/autofs/autofs_vnops.c 296918 2016-03-15 20:42:36Z trasz $"); 33 34#include <sys/param.h> 35#include <sys/kernel.h> 36#include <sys/condvar.h> 37#include <sys/dirent.h> 38#include <sys/fcntl.h> 39#include <sys/lock.h> 40#include <sys/mount.h> 41#include <sys/mutex.h> 42#include <sys/namei.h> 43#include <sys/signalvar.h> 44#include <sys/stat.h> 45#include <sys/systm.h> 46#include <sys/taskqueue.h> 47#include <sys/vnode.h> 48#include <machine/atomic.h> 49#include <vm/uma.h> 50 51#include <fs/autofs/autofs.h> 52 53static int autofs_trigger_vn(struct vnode *vp, const char *path, 54 int pathlen, struct vnode **newvp); 55 56extern struct autofs_softc *autofs_softc; 57 58static int 59autofs_access(struct vop_access_args *ap) 60{ 61 62 /* 63 * Nothing to do here; the only kind of access control 64 * needed is in autofs_mkdir(). 65 */ 66 67 return (0); 68} 69 70static int 71autofs_getattr(struct vop_getattr_args *ap) 72{ 73 struct vnode *vp, *newvp; 74 struct autofs_node *anp; 75 struct mount *mp; 76 struct vattr *vap; 77 int error; 78 79 vp = ap->a_vp; 80 anp = vp->v_data; 81 mp = vp->v_mount; 82 vap = ap->a_vap; 83 84 KASSERT(ap->a_vp->v_type == VDIR, ("!VDIR")); 85 86 /* 87 * The reason we must do this is that some tree-walking software, 88 * namely fts(3), assumes that stat(".") results will not change 89 * between chdir("subdir") and chdir(".."), and fails with ENOENT 90 * otherwise. 91 */ 92 if (autofs_mount_on_stat && autofs_cached(anp, NULL, 0) == false && 93 autofs_ignore_thread(curthread) == false) { 94 error = autofs_trigger_vn(vp, "", 0, &newvp); 95 if (error != 0) 96 return (error); 97 98 if (newvp != NULL) { 99 error = VOP_GETATTR(newvp, ap->a_vap, 100 ap->a_cred); 101 vput(newvp); 102 return (error); 103 } 104 } 105 106 vap->va_type = VDIR; 107 vap->va_mode = 0755; 108 vap->va_nlink = 3; /* XXX */ 109 vap->va_uid = 0; 110 vap->va_gid = 0; 111 vap->va_rdev = NODEV; 112 vap->va_fsid = mp->mnt_stat.f_fsid.val[0]; 113 vap->va_fileid = anp->an_fileno; 114 vap->va_size = S_BLKSIZE; 115 vap->va_blocksize = S_BLKSIZE; 116 vap->va_mtime = anp->an_ctime; 117 vap->va_atime = anp->an_ctime; 118 vap->va_ctime = anp->an_ctime; 119 vap->va_birthtime = anp->an_ctime; 120 vap->va_gen = 0; 121 vap->va_flags = 0; 122 vap->va_rdev = 0; 123 vap->va_bytes = S_BLKSIZE; 124 vap->va_filerev = 0; 125 vap->va_spare = 0; 126 127 return (0); 128} 129 130/* 131 * Unlock the vnode, request automountd(8) action, and then lock it back. 132 * If anything got mounted on top of the vnode, return the new filesystem's 133 * root vnode in 'newvp', locked. 134 */ 135static int 136autofs_trigger_vn(struct vnode *vp, const char *path, int pathlen, 137 struct vnode **newvp) 138{ 139 struct autofs_node *anp; 140 struct autofs_mount *amp; 141 int error, lock_flags; 142 143 anp = vp->v_data; 144 amp = VFSTOAUTOFS(vp->v_mount); 145 146 /* 147 * Release the vnode lock, so that other operations, in partcular 148 * mounting a filesystem on top of it, can proceed. Increase use 149 * count, to prevent the vnode from being deallocated and to prevent 150 * filesystem from being unmounted. 151 */ 152 lock_flags = VOP_ISLOCKED(vp); 153 vref(vp); 154 VOP_UNLOCK(vp, 0); 155 156 sx_xlock(&autofs_softc->sc_lock); 157 158 /* 159 * XXX: Workaround for mounting the same thing multiple times; revisit. 160 */ 161 if (vp->v_mountedhere != NULL) { 162 error = 0; 163 goto mounted; 164 } 165 166 error = autofs_trigger(anp, path, pathlen); 167mounted: 168 sx_xunlock(&autofs_softc->sc_lock); 169 vn_lock(vp, lock_flags | LK_RETRY); 170 vunref(vp); 171 if ((vp->v_iflag & VI_DOOMED) != 0) { 172 AUTOFS_DEBUG("VI_DOOMED"); 173 return (ENOENT); 174 } 175 176 if (error != 0) 177 return (error); 178 179 if (vp->v_mountedhere == NULL) { 180 *newvp = NULL; 181 return (0); 182 } else { 183 /* 184 * If the operation that succeeded was mount, then mark 185 * the node as non-cached. Otherwise, if someone unmounts 186 * the filesystem before the cache times out, we will fail 187 * to trigger. 188 */ 189 anp->an_cached = false; 190 } 191 192 error = VFS_ROOT(vp->v_mountedhere, lock_flags, newvp); 193 if (error != 0) { 194 AUTOFS_WARN("VFS_ROOT() failed with error %d", error); 195 return (error); 196 } 197 198 return (0); 199} 200 201static int 202autofs_vget_callback(struct mount *mp, void *arg, int flags, 203 struct vnode **vpp) 204{ 205 206 207 return (autofs_node_vn(arg, mp, flags, vpp)); 208} 209 210static int 211autofs_lookup(struct vop_lookup_args *ap) 212{ 213 struct vnode *dvp, *newvp, **vpp; 214 struct mount *mp; 215 struct autofs_mount *amp; 216 struct autofs_node *anp, *child; 217 struct componentname *cnp; 218 int error; 219 220 dvp = ap->a_dvp; 221 vpp = ap->a_vpp; 222 mp = dvp->v_mount; 223 amp = VFSTOAUTOFS(mp); 224 anp = dvp->v_data; 225 cnp = ap->a_cnp; 226 227 if (cnp->cn_flags & ISDOTDOT) { 228 KASSERT(anp->an_parent != NULL, ("NULL parent")); 229 /* 230 * Note that in this case, dvp is the child vnode, and we 231 * are looking up the parent vnode - exactly reverse from 232 * normal operation. Unlocking dvp requires some rather 233 * tricky unlock/relock dance to prevent mp from being freed; 234 * use vn_vget_ino_gen() which takes care of all that. 235 */ 236 error = vn_vget_ino_gen(dvp, autofs_vget_callback, 237 anp->an_parent, cnp->cn_lkflags, vpp); 238 if (error != 0) { 239 AUTOFS_WARN("vn_vget_ino_gen() failed with error %d", 240 error); 241 return (error); 242 } 243 return (error); 244 } 245 246 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') { 247 vref(dvp); 248 *vpp = dvp; 249 250 return (0); 251 } 252 253 if (autofs_cached(anp, cnp->cn_nameptr, cnp->cn_namelen) == false && 254 autofs_ignore_thread(cnp->cn_thread) == false) { 255 error = autofs_trigger_vn(dvp, 256 cnp->cn_nameptr, cnp->cn_namelen, &newvp); 257 if (error != 0) 258 return (error); 259 260 if (newvp != NULL) { 261 /* 262 * The target filesystem got automounted. 263 * Let the lookup(9) go around with the same 264 * path component. 265 */ 266 vput(newvp); 267 return (ERELOOKUP); 268 } 269 } 270 271 AUTOFS_SLOCK(amp); 272 error = autofs_node_find(anp, cnp->cn_nameptr, cnp->cn_namelen, &child); 273 if (error != 0) { 274 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) { 275 AUTOFS_SUNLOCK(amp); 276 return (EJUSTRETURN); 277 } 278 279 AUTOFS_SUNLOCK(amp); 280 return (ENOENT); 281 } 282 283 /* 284 * XXX: Dropping the node here is ok, because we never remove nodes. 285 */ 286 AUTOFS_SUNLOCK(amp); 287 288 error = autofs_node_vn(child, mp, cnp->cn_lkflags, vpp); 289 if (error != 0) { 290 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) 291 return (EJUSTRETURN); 292 293 return (error); 294 } 295 296 return (0); 297} 298 299static int 300autofs_mkdir(struct vop_mkdir_args *ap) 301{ 302 struct vnode *vp; 303 struct autofs_node *anp; 304 struct autofs_mount *amp; 305 struct autofs_node *child; 306 int error; 307 308 vp = ap->a_dvp; 309 anp = vp->v_data; 310 amp = VFSTOAUTOFS(vp->v_mount); 311 312 /* 313 * Do not allow mkdir() if the calling thread is not 314 * automountd(8) descendant. 315 */ 316 if (autofs_ignore_thread(curthread) == false) 317 return (EPERM); 318 319 AUTOFS_XLOCK(amp); 320 error = autofs_node_new(anp, amp, ap->a_cnp->cn_nameptr, 321 ap->a_cnp->cn_namelen, &child); 322 if (error != 0) { 323 AUTOFS_XUNLOCK(amp); 324 return (error); 325 } 326 AUTOFS_XUNLOCK(amp); 327 328 error = autofs_node_vn(child, vp->v_mount, LK_EXCLUSIVE, ap->a_vpp); 329 330 return (error); 331} 332 333/* 334 * Write out a single 'struct dirent', based on 'name' and 'fileno' arguments. 335 */ 336static int 337autofs_readdir_one(struct uio *uio, const char *name, int fileno, 338 size_t *reclenp) 339{ 340 struct dirent dirent; 341 size_t namlen, padded_namlen, reclen; 342 int error; 343 344 namlen = strlen(name); 345 padded_namlen = roundup2(namlen + 1, __alignof(struct dirent)); 346 KASSERT(padded_namlen <= MAXNAMLEN, ("%zd > MAXNAMLEN", padded_namlen)); 347 reclen = offsetof(struct dirent, d_name) + padded_namlen; 348 349 if (reclenp != NULL) 350 *reclenp = reclen; 351 352 if (uio == NULL) 353 return (0); 354 355 if (uio->uio_resid < reclen) 356 return (EINVAL); 357 358 dirent.d_fileno = fileno; 359 dirent.d_reclen = reclen; 360 dirent.d_type = DT_DIR; 361 dirent.d_namlen = namlen; 362 memcpy(dirent.d_name, name, namlen); 363 memset(dirent.d_name + namlen, 0, padded_namlen - namlen); 364 error = uiomove(&dirent, reclen, uio); 365 366 return (error); 367} 368 369static size_t 370autofs_dirent_reclen(const char *name) 371{ 372 size_t reclen; 373 int error; 374 375 error = autofs_readdir_one(NULL, name, -1, &reclen); 376 KASSERT(error == 0, ("autofs_readdir_one() failed")); 377 378 return (reclen); 379} 380 381static int 382autofs_readdir(struct vop_readdir_args *ap) 383{ 384 struct vnode *vp, *newvp; 385 struct autofs_mount *amp; 386 struct autofs_node *anp, *child; 387 struct uio *uio; 388 size_t reclen, reclens; 389 ssize_t initial_resid; 390 int error; 391 392 vp = ap->a_vp; 393 amp = VFSTOAUTOFS(vp->v_mount); 394 anp = vp->v_data; 395 uio = ap->a_uio; 396 initial_resid = ap->a_uio->uio_resid; 397 398 KASSERT(vp->v_type == VDIR, ("!VDIR")); 399 400 if (autofs_cached(anp, NULL, 0) == false && 401 autofs_ignore_thread(curthread) == false) { 402 error = autofs_trigger_vn(vp, "", 0, &newvp); 403 if (error != 0) 404 return (error); 405 406 if (newvp != NULL) { 407 error = VOP_READDIR(newvp, ap->a_uio, ap->a_cred, 408 ap->a_eofflag, ap->a_ncookies, ap->a_cookies); 409 vput(newvp); 410 return (error); 411 } 412 } 413 414 if (uio->uio_offset < 0) 415 return (EINVAL); 416 417 if (ap->a_eofflag != NULL) 418 *ap->a_eofflag = FALSE; 419 420 /* 421 * Write out the directory entry for ".". This is conditional 422 * on the current offset into the directory; same applies to the 423 * other two cases below. 424 */ 425 if (uio->uio_offset == 0) { 426 error = autofs_readdir_one(uio, ".", anp->an_fileno, &reclen); 427 if (error != 0) 428 goto out; 429 } 430 reclens = autofs_dirent_reclen("."); 431 432 /* 433 * Write out the directory entry for "..". 434 */ 435 if (uio->uio_offset <= reclens) { 436 if (uio->uio_offset != reclens) 437 return (EINVAL); 438 if (anp->an_parent == NULL) { 439 error = autofs_readdir_one(uio, "..", 440 anp->an_fileno, &reclen); 441 } else { 442 error = autofs_readdir_one(uio, "..", 443 anp->an_parent->an_fileno, &reclen); 444 } 445 if (error != 0) 446 goto out; 447 } 448 449 reclens += autofs_dirent_reclen(".."); 450 451 /* 452 * Write out the directory entries for subdirectories. 453 */ 454 AUTOFS_SLOCK(amp); 455 TAILQ_FOREACH(child, &anp->an_children, an_next) { 456 /* 457 * Check the offset to skip entries returned by previous 458 * calls to getdents(). 459 */ 460 if (uio->uio_offset > reclens) { 461 reclens += autofs_dirent_reclen(child->an_name); 462 continue; 463 } 464 465 /* 466 * Prevent seeking into the middle of dirent. 467 */ 468 if (uio->uio_offset != reclens) { 469 AUTOFS_SUNLOCK(amp); 470 return (EINVAL); 471 } 472 473 error = autofs_readdir_one(uio, child->an_name, 474 child->an_fileno, &reclen); 475 reclens += reclen; 476 if (error != 0) { 477 AUTOFS_SUNLOCK(amp); 478 goto out; 479 } 480 } 481 AUTOFS_SUNLOCK(amp); 482 483 if (ap->a_eofflag != NULL) 484 *ap->a_eofflag = TRUE; 485 486 return (0); 487 488out: 489 /* 490 * Return error if the initial buffer was too small to do anything. 491 */ 492 if (uio->uio_resid == initial_resid) 493 return (error); 494 495 /* 496 * Don't return an error if we managed to copy out some entries. 497 */ 498 if (uio->uio_resid < reclen) 499 return (0); 500 501 return (error); 502} 503 504static int 505autofs_reclaim(struct vop_reclaim_args *ap) 506{ 507 struct vnode *vp; 508 struct autofs_node *anp; 509 510 vp = ap->a_vp; 511 anp = vp->v_data; 512 513 /* 514 * We do not free autofs_node here; instead we are 515 * destroying them in autofs_node_delete(). 516 */ 517 sx_xlock(&anp->an_vnode_lock); 518 anp->an_vnode = NULL; 519 vp->v_data = NULL; 520 sx_xunlock(&anp->an_vnode_lock); 521 522 return (0); 523} 524 525struct vop_vector autofs_vnodeops = { 526 .vop_default = &default_vnodeops, 527 528 .vop_access = autofs_access, 529 .vop_lookup = autofs_lookup, 530 .vop_create = VOP_EOPNOTSUPP, 531 .vop_getattr = autofs_getattr, 532 .vop_link = VOP_EOPNOTSUPP, 533 .vop_mkdir = autofs_mkdir, 534 .vop_mknod = VOP_EOPNOTSUPP, 535 .vop_read = VOP_EOPNOTSUPP, 536 .vop_readdir = autofs_readdir, 537 .vop_remove = VOP_EOPNOTSUPP, 538 .vop_rename = VOP_EOPNOTSUPP, 539 .vop_rmdir = VOP_EOPNOTSUPP, 540 .vop_setattr = VOP_EOPNOTSUPP, 541 .vop_symlink = VOP_EOPNOTSUPP, 542 .vop_write = VOP_EOPNOTSUPP, 543 .vop_reclaim = autofs_reclaim, 544}; 545 546int 547autofs_node_new(struct autofs_node *parent, struct autofs_mount *amp, 548 const char *name, int namelen, struct autofs_node **anpp) 549{ 550 struct autofs_node *anp; 551 552 if (parent != NULL) { 553 AUTOFS_ASSERT_XLOCKED(parent->an_mount); 554 555 KASSERT(autofs_node_find(parent, name, namelen, NULL) == ENOENT, 556 ("node \"%s\" already exists", name)); 557 } 558 559 anp = uma_zalloc(autofs_node_zone, M_WAITOK | M_ZERO); 560 if (namelen >= 0) 561 anp->an_name = strndup(name, namelen, M_AUTOFS); 562 else 563 anp->an_name = strdup(name, M_AUTOFS); 564 anp->an_fileno = atomic_fetchadd_int(&->am_last_fileno, 1); 565 callout_init(&anp->an_callout, 1); 566 /* 567 * The reason for SX_NOWITNESS here is that witness(4) 568 * cannot tell vnodes apart, so the following perfectly 569 * valid lock order... 570 * 571 * vnode lock A -> autofsvlk B -> vnode lock B 572 * 573 * ... gets reported as a LOR. 574 */ 575 sx_init_flags(&anp->an_vnode_lock, "autofsvlk", SX_NOWITNESS); 576 getnanotime(&anp->an_ctime); 577 anp->an_parent = parent; 578 anp->an_mount = amp; 579 if (parent != NULL) 580 TAILQ_INSERT_TAIL(&parent->an_children, anp, an_next); 581 TAILQ_INIT(&anp->an_children); 582 583 *anpp = anp; 584 return (0); 585} 586 587int 588autofs_node_find(struct autofs_node *parent, const char *name, 589 int namelen, struct autofs_node **anpp) 590{ 591 struct autofs_node *anp; 592 593 AUTOFS_ASSERT_LOCKED(parent->an_mount); 594 595 TAILQ_FOREACH(anp, &parent->an_children, an_next) { 596 if (namelen >= 0) { 597 if (strlen(anp->an_name) != namelen) 598 continue; 599 if (strncmp(anp->an_name, name, namelen) != 0) 600 continue; 601 } else { 602 if (strcmp(anp->an_name, name) != 0) 603 continue; 604 } 605 606 if (anpp != NULL) 607 *anpp = anp; 608 return (0); 609 } 610 611 return (ENOENT); 612} 613 614void 615autofs_node_delete(struct autofs_node *anp) 616{ 617 struct autofs_node *parent; 618 619 AUTOFS_ASSERT_XLOCKED(anp->an_mount); 620 KASSERT(TAILQ_EMPTY(&anp->an_children), ("have children")); 621 622 callout_drain(&anp->an_callout); 623 624 parent = anp->an_parent; 625 if (parent != NULL) 626 TAILQ_REMOVE(&parent->an_children, anp, an_next); 627 sx_destroy(&anp->an_vnode_lock); 628 free(anp->an_name, M_AUTOFS); 629 uma_zfree(autofs_node_zone, anp); 630} 631 632int 633autofs_node_vn(struct autofs_node *anp, struct mount *mp, int flags, 634 struct vnode **vpp) 635{ 636 struct vnode *vp; 637 int error; 638 639 AUTOFS_ASSERT_UNLOCKED(anp->an_mount); 640 641 sx_xlock(&anp->an_vnode_lock); 642 643 vp = anp->an_vnode; 644 if (vp != NULL) { 645 error = vget(vp, flags | LK_RETRY, curthread); 646 if (error != 0) { 647 AUTOFS_WARN("vget failed with error %d", error); 648 sx_xunlock(&anp->an_vnode_lock); 649 return (error); 650 } 651 if (vp->v_iflag & VI_DOOMED) { 652 /* 653 * We got forcibly unmounted. 654 */ 655 AUTOFS_DEBUG("doomed vnode"); 656 sx_xunlock(&anp->an_vnode_lock); 657 vput(vp); 658 659 return (ENOENT); 660 } 661 662 *vpp = vp; 663 sx_xunlock(&anp->an_vnode_lock); 664 return (0); 665 } 666 667 error = getnewvnode("autofs", mp, &autofs_vnodeops, &vp); 668 if (error != 0) { 669 sx_xunlock(&anp->an_vnode_lock); 670 return (error); 671 } 672 673 error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 674 if (error != 0) { 675 sx_xunlock(&anp->an_vnode_lock); 676 vdrop(vp); 677 return (error); 678 } 679 680 vp->v_type = VDIR; 681 if (anp->an_parent == NULL) 682 vp->v_vflag |= VV_ROOT; 683 vp->v_data = anp; 684 685 VN_LOCK_ASHARE(vp); 686 687 error = insmntque(vp, mp); 688 if (error != 0) { 689 AUTOFS_WARN("insmntque() failed with error %d", error); 690 sx_xunlock(&anp->an_vnode_lock); 691 return (error); 692 } 693 694 KASSERT(anp->an_vnode == NULL, ("lost race")); 695 anp->an_vnode = vp; 696 697 sx_xunlock(&anp->an_vnode_lock); 698 699 *vpp = vp; 700 return (0); 701} 702