1 /* $NetBSD: vfs_subr.c,v 1.230 2004/07/01 10:03:29 hannken Exp $ */ 2 3 /*- 4 * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center. 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 * Copyright (c) 1989, 1993 42 * The Regents of the University of California. All rights reserved. 43 * (c) UNIX System Laboratories, Inc. 44 * All or some portions of this file are derived from material licensed 45 * to the University of California by American Telephone and Telegraph 46 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 47 * the permission of UNIX System Laboratories, Inc. 48 * 49 * Redistribution and use in source and binary forms, with or without 50 * modification, are permitted provided that the following conditions 51 * are met: 52 * 1. Redistributions of source code must retain the above copyright 53 * notice, this list of conditions and the following disclaimer. 54 * 2. Redistributions in binary form must reproduce the above copyright 55 * notice, this list of conditions and the following disclaimer in the 56 * documentation and/or other materials provided with the distribution. 57 * 3. Neither the name of the University nor the names of its contributors 58 * may be used to endorse or promote products derived from this software 59 * without specific prior written permission. 60 * 61 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 62 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 63 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 64 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 65 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 66 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 67 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 68 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 69 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 70 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 71 * SUCH DAMAGE. 72 * 73 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 74 */ 75 76 /* 77 * External virtual filesystem routines 78 */ 79 80 #include <sys/cdefs.h> 81 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.230 2004/07/01 10:03:29 hannken Exp $"); 82 83 #include "opt_inet.h" 84 #include "opt_ddb.h" 85 #include "opt_compat_netbsd.h" 86 #include "opt_compat_43.h" 87 88 #include <sys/param.h> 89 #include <sys/systm.h> 90 #include <sys/proc.h> 91 #include <sys/kernel.h> 92 #include <sys/mount.h> 93 #include <sys/time.h> 94 #include <sys/event.h> 95 #include <sys/fcntl.h> 96 #include <sys/vnode.h> 97 #include <sys/stat.h> 98 #include <sys/namei.h> 99 #include <sys/ucred.h> 100 #include <sys/buf.h> 101 #include <sys/errno.h> 102 #include <sys/malloc.h> 103 #include <sys/domain.h> 104 #include <sys/mbuf.h> 105 #include <sys/sa.h> 106 #include <sys/syscallargs.h> 107 #include <sys/device.h> 108 #include <sys/dirent.h> 109 #include <sys/filedesc.h> 110 111 #include <miscfs/specfs/specdev.h> 112 #include <miscfs/genfs/genfs.h> 113 #include <miscfs/syncfs/syncfs.h> 114 115 #include <netinet/in.h> 116 117 #include <uvm/uvm.h> 118 #include <uvm/uvm_ddb.h> 119 120 #include <netinet/in.h> 121 122 #include <sys/sysctl.h> 123 124 const enum vtype iftovt_tab[16] = { 125 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 126 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 127 }; 128 const int vttoif_tab[9] = { 129 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 130 S_IFSOCK, S_IFIFO, S_IFMT, 131 }; 132 133 int doforce = 1; /* 1 => permit forcible unmounting */ 134 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 135 136 extern int dovfsusermount; /* 1 => permit any user to mount filesystems */ 137 138 /* 139 * Insq/Remq for the vnode usage lists. 140 */ 141 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 142 #define bufremvn(bp) { \ 143 LIST_REMOVE(bp, b_vnbufs); \ 144 (bp)->b_vnbufs.le_next = NOLIST; \ 145 } 146 /* TAILQ_HEAD(freelst, vnode) vnode_free_list = vnode free list (in vnode.h) */ 147 struct freelst vnode_free_list = TAILQ_HEAD_INITIALIZER(vnode_free_list); 148 struct freelst vnode_hold_list = TAILQ_HEAD_INITIALIZER(vnode_hold_list); 149 150 struct mntlist mountlist = /* mounted filesystem list */ 151 CIRCLEQ_HEAD_INITIALIZER(mountlist); 152 struct vfs_list_head vfs_list = /* vfs list */ 153 LIST_HEAD_INITIALIZER(vfs_list); 154 155 struct nfs_public nfs_pub; /* publicly exported FS */ 156 157 struct simplelock mountlist_slock = SIMPLELOCK_INITIALIZER; 158 static struct simplelock mntid_slock = SIMPLELOCK_INITIALIZER; 159 struct simplelock mntvnode_slock = SIMPLELOCK_INITIALIZER; 160 struct simplelock vnode_free_list_slock = SIMPLELOCK_INITIALIZER; 161 struct simplelock spechash_slock = SIMPLELOCK_INITIALIZER; 162 163 /* XXX - gross; single global lock to protect v_numoutput */ 164 struct simplelock global_v_numoutput_slock = SIMPLELOCK_INITIALIZER; 165 166 /* 167 * These define the root filesystem and device. 168 */ 169 struct mount *rootfs; 170 struct vnode *rootvnode; 171 struct device *root_device; /* root device */ 172 173 POOL_INIT(vnode_pool, sizeof(struct vnode), 0, 0, 0, "vnodepl", 174 &pool_allocator_nointr); 175 176 MALLOC_DEFINE(M_VNODE, "vnodes", "Dynamically allocated vnodes"); 177 178 /* 179 * Local declarations. 180 */ 181 void insmntque(struct vnode *, struct mount *); 182 int getdevvp(dev_t, struct vnode **, enum vtype); 183 void vgoneall(struct vnode *); 184 185 void vclean(struct vnode *, int, struct proc *); 186 187 static int vfs_hang_addrlist(struct mount *, struct netexport *, 188 struct export_args *); 189 static int vfs_free_netcred(struct radix_node *, void *); 190 static void vfs_free_addrlist(struct netexport *); 191 static struct vnode *getcleanvnode(struct proc *); 192 193 #ifdef DEBUG 194 void printlockedvnodes(void); 195 #endif 196 197 /* 198 * Initialize the vnode management data structures. 199 */ 200 void 201 vntblinit() 202 { 203 204 /* 205 * Initialize the filesystem syncer. 206 */ 207 vn_initialize_syncerd(); 208 } 209 210 int 211 vfs_drainvnodes(long target, struct proc *p) 212 { 213 214 simple_lock(&vnode_free_list_slock); 215 while (numvnodes > target) { 216 struct vnode *vp; 217 218 vp = getcleanvnode(p); 219 if (vp == NULL) 220 return EBUSY; /* give up */ 221 pool_put(&vnode_pool, vp); 222 simple_lock(&vnode_free_list_slock); 223 numvnodes--; 224 } 225 simple_unlock(&vnode_free_list_slock); 226 227 return 0; 228 } 229 230 /* 231 * grab a vnode from freelist and clean it. 232 */ 233 struct vnode * 234 getcleanvnode(p) 235 struct proc *p; 236 { 237 struct vnode *vp; 238 struct mount *mp; 239 struct freelst *listhd; 240 241 LOCK_ASSERT(simple_lock_held(&vnode_free_list_slock)); 242 243 listhd = &vnode_free_list; 244 try_nextlist: 245 TAILQ_FOREACH(vp, listhd, v_freelist) { 246 if (!simple_lock_try(&vp->v_interlock)) 247 continue; 248 /* 249 * as our lwp might hold the underlying vnode locked, 250 * don't try to reclaim the VLAYER vnode if it's locked. 251 */ 252 if ((vp->v_flag & VXLOCK) == 0 && 253 ((vp->v_flag & VLAYER) == 0 || VOP_ISLOCKED(vp) == 0)) { 254 if (vn_start_write(vp, &mp, V_NOWAIT) == 0) 255 break; 256 } 257 mp = NULL; 258 simple_unlock(&vp->v_interlock); 259 } 260 261 if (vp == NULLVP) { 262 if (listhd == &vnode_free_list) { 263 listhd = &vnode_hold_list; 264 goto try_nextlist; 265 } 266 simple_unlock(&vnode_free_list_slock); 267 return NULLVP; 268 } 269 270 if (vp->v_usecount) 271 panic("free vnode isn't, vp %p", vp); 272 TAILQ_REMOVE(listhd, vp, v_freelist); 273 /* see comment on why 0xdeadb is set at end of vgone (below) */ 274 vp->v_freelist.tqe_prev = (struct vnode **)0xdeadb; 275 simple_unlock(&vnode_free_list_slock); 276 vp->v_lease = NULL; 277 278 if (vp->v_type != VBAD) 279 vgonel(vp, p); 280 else 281 simple_unlock(&vp->v_interlock); 282 vn_finished_write(mp, 0); 283 #ifdef DIAGNOSTIC 284 if (vp->v_data || vp->v_uobj.uo_npages || 285 TAILQ_FIRST(&vp->v_uobj.memq)) 286 panic("cleaned vnode isn't, vp %p", vp); 287 if (vp->v_numoutput) 288 panic("clean vnode has pending I/O's, vp %p", vp); 289 #endif 290 KASSERT((vp->v_flag & VONWORKLST) == 0); 291 292 return vp; 293 } 294 295 /* 296 * Mark a mount point as busy. Used to synchronize access and to delay 297 * unmounting. Interlock is not released on failure. 298 */ 299 int 300 vfs_busy(mp, flags, interlkp) 301 struct mount *mp; 302 int flags; 303 struct simplelock *interlkp; 304 { 305 int lkflags; 306 307 while (mp->mnt_iflag & IMNT_UNMOUNT) { 308 int gone, n; 309 310 if (flags & LK_NOWAIT) 311 return (ENOENT); 312 if ((flags & LK_RECURSEFAIL) && mp->mnt_unmounter != NULL 313 && mp->mnt_unmounter == curproc) 314 return (EDEADLK); 315 if (interlkp) 316 simple_unlock(interlkp); 317 /* 318 * Since all busy locks are shared except the exclusive 319 * lock granted when unmounting, the only place that a 320 * wakeup needs to be done is at the release of the 321 * exclusive lock at the end of dounmount. 322 */ 323 simple_lock(&mp->mnt_slock); 324 mp->mnt_wcnt++; 325 ltsleep((caddr_t)mp, PVFS, "vfs_busy", 0, &mp->mnt_slock); 326 n = --mp->mnt_wcnt; 327 simple_unlock(&mp->mnt_slock); 328 gone = mp->mnt_iflag & IMNT_GONE; 329 330 if (n == 0) 331 wakeup(&mp->mnt_wcnt); 332 if (interlkp) 333 simple_lock(interlkp); 334 if (gone) 335 return (ENOENT); 336 } 337 lkflags = LK_SHARED; 338 if (interlkp) 339 lkflags |= LK_INTERLOCK; 340 if (lockmgr(&mp->mnt_lock, lkflags, interlkp)) 341 panic("vfs_busy: unexpected lock failure"); 342 return (0); 343 } 344 345 /* 346 * Free a busy filesystem. 347 */ 348 void 349 vfs_unbusy(mp) 350 struct mount *mp; 351 { 352 353 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL); 354 } 355 356 /* 357 * Lookup a filesystem type, and if found allocate and initialize 358 * a mount structure for it. 359 * 360 * Devname is usually updated by mount(8) after booting. 361 */ 362 int 363 vfs_rootmountalloc(fstypename, devname, mpp) 364 char *fstypename; 365 char *devname; 366 struct mount **mpp; 367 { 368 struct vfsops *vfsp = NULL; 369 struct mount *mp; 370 371 LIST_FOREACH(vfsp, &vfs_list, vfs_list) 372 if (!strncmp(vfsp->vfs_name, fstypename, MFSNAMELEN)) 373 break; 374 375 if (vfsp == NULL) 376 return (ENODEV); 377 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); 378 memset((char *)mp, 0, (u_long)sizeof(struct mount)); 379 lockinit(&mp->mnt_lock, PVFS, "vfslock", 0, 0); 380 simple_lock_init(&mp->mnt_slock); 381 (void)vfs_busy(mp, LK_NOWAIT, 0); 382 LIST_INIT(&mp->mnt_vnodelist); 383 mp->mnt_op = vfsp; 384 mp->mnt_flag = MNT_RDONLY; 385 mp->mnt_vnodecovered = NULLVP; 386 mp->mnt_leaf = mp; 387 vfsp->vfs_refcount++; 388 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, MFSNAMELEN); 389 mp->mnt_stat.f_mntonname[0] = '/'; 390 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 391 *mpp = mp; 392 return (0); 393 } 394 395 /* 396 * Lookup a mount point by filesystem identifier. 397 */ 398 struct mount * 399 vfs_getvfs(fsid) 400 fsid_t *fsid; 401 { 402 struct mount *mp; 403 404 simple_lock(&mountlist_slock); 405 CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) { 406 if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] && 407 mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) { 408 simple_unlock(&mountlist_slock); 409 return (mp); 410 } 411 } 412 simple_unlock(&mountlist_slock); 413 return ((struct mount *)0); 414 } 415 416 /* 417 * Get a new unique fsid 418 */ 419 void 420 vfs_getnewfsid(mp) 421 struct mount *mp; 422 { 423 static u_short xxxfs_mntid; 424 fsid_t tfsid; 425 int mtype; 426 427 simple_lock(&mntid_slock); 428 mtype = makefstype(mp->mnt_op->vfs_name); 429 mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0); 430 mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype; 431 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 432 if (xxxfs_mntid == 0) 433 ++xxxfs_mntid; 434 tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid); 435 tfsid.__fsid_val[1] = mtype; 436 if (!CIRCLEQ_EMPTY(&mountlist)) { 437 while (vfs_getvfs(&tfsid)) { 438 tfsid.__fsid_val[0]++; 439 xxxfs_mntid++; 440 } 441 } 442 mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0]; 443 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 444 simple_unlock(&mntid_slock); 445 } 446 447 /* 448 * Make a 'unique' number from a mount type name. 449 */ 450 long 451 makefstype(type) 452 const char *type; 453 { 454 long rv; 455 456 for (rv = 0; *type; type++) { 457 rv <<= 2; 458 rv ^= *type; 459 } 460 return rv; 461 } 462 463 464 /* 465 * Set vnode attributes to VNOVAL 466 */ 467 void 468 vattr_null(vap) 469 struct vattr *vap; 470 { 471 472 vap->va_type = VNON; 473 474 /* 475 * Assign individually so that it is safe even if size and 476 * sign of each member are varied. 477 */ 478 vap->va_mode = VNOVAL; 479 vap->va_nlink = VNOVAL; 480 vap->va_uid = VNOVAL; 481 vap->va_gid = VNOVAL; 482 vap->va_fsid = VNOVAL; 483 vap->va_fileid = VNOVAL; 484 vap->va_size = VNOVAL; 485 vap->va_blocksize = VNOVAL; 486 vap->va_atime.tv_sec = 487 vap->va_mtime.tv_sec = 488 vap->va_ctime.tv_sec = 489 vap->va_birthtime.tv_sec = VNOVAL; 490 vap->va_atime.tv_nsec = 491 vap->va_mtime.tv_nsec = 492 vap->va_ctime.tv_nsec = 493 vap->va_birthtime.tv_nsec = VNOVAL; 494 vap->va_gen = VNOVAL; 495 vap->va_flags = VNOVAL; 496 vap->va_rdev = VNOVAL; 497 vap->va_bytes = VNOVAL; 498 vap->va_vaflags = 0; 499 } 500 501 /* 502 * Routines having to do with the management of the vnode table. 503 */ 504 extern int (**dead_vnodeop_p)(void *); 505 long numvnodes; 506 507 /* 508 * Return the next vnode from the free list. 509 */ 510 int 511 getnewvnode(tag, mp, vops, vpp) 512 enum vtagtype tag; 513 struct mount *mp; 514 int (**vops)(void *); 515 struct vnode **vpp; 516 { 517 extern struct uvm_pagerops uvm_vnodeops; 518 struct uvm_object *uobj; 519 struct proc *p = curproc; /* XXX */ 520 static int toggle; 521 struct vnode *vp; 522 int error = 0, tryalloc; 523 524 try_again: 525 if (mp) { 526 /* 527 * Mark filesystem busy while we're creating a vnode. 528 * If unmount is in progress, this will wait; if the 529 * unmount succeeds (only if umount -f), this will 530 * return an error. If the unmount fails, we'll keep 531 * going afterwards. 532 * (This puts the per-mount vnode list logically under 533 * the protection of the vfs_busy lock). 534 */ 535 error = vfs_busy(mp, LK_RECURSEFAIL, 0); 536 if (error && error != EDEADLK) 537 return error; 538 } 539 540 /* 541 * We must choose whether to allocate a new vnode or recycle an 542 * existing one. The criterion for allocating a new one is that 543 * the total number of vnodes is less than the number desired or 544 * there are no vnodes on either free list. Generally we only 545 * want to recycle vnodes that have no buffers associated with 546 * them, so we look first on the vnode_free_list. If it is empty, 547 * we next consider vnodes with referencing buffers on the 548 * vnode_hold_list. The toggle ensures that half the time we 549 * will use a buffer from the vnode_hold_list, and half the time 550 * we will allocate a new one unless the list has grown to twice 551 * the desired size. We are reticent to recycle vnodes from the 552 * vnode_hold_list because we will lose the identity of all its 553 * referencing buffers. 554 */ 555 556 vp = NULL; 557 558 simple_lock(&vnode_free_list_slock); 559 560 toggle ^= 1; 561 if (numvnodes > 2 * desiredvnodes) 562 toggle = 0; 563 564 tryalloc = numvnodes < desiredvnodes || 565 (TAILQ_FIRST(&vnode_free_list) == NULL && 566 (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle)); 567 568 if (tryalloc && 569 (vp = pool_get(&vnode_pool, PR_NOWAIT)) != NULL) { 570 numvnodes++; 571 simple_unlock(&vnode_free_list_slock); 572 memset(vp, 0, sizeof(*vp)); 573 simple_lock_init(&vp->v_interlock); 574 uobj = &vp->v_uobj; 575 uobj->pgops = &uvm_vnodeops; 576 TAILQ_INIT(&uobj->memq); 577 /* 578 * done by memset() above. 579 * uobj->uo_npages = 0; 580 * LIST_INIT(&vp->v_nclist); 581 * LIST_INIT(&vp->v_dnclist); 582 */ 583 } else { 584 vp = getcleanvnode(p); 585 /* 586 * Unless this is a bad time of the month, at most 587 * the first NCPUS items on the free list are 588 * locked, so this is close enough to being empty. 589 */ 590 if (vp == NULLVP) { 591 if (mp && error != EDEADLK) 592 vfs_unbusy(mp); 593 if (tryalloc) { 594 printf("WARNING: unable to allocate new " 595 "vnode, retrying...\n"); 596 (void) tsleep(&lbolt, PRIBIO, "newvn", hz); 597 goto try_again; 598 } 599 tablefull("vnode", "increase kern.maxvnodes or NVNODE"); 600 *vpp = 0; 601 return (ENFILE); 602 } 603 vp->v_flag = 0; 604 vp->v_socket = NULL; 605 #ifdef VERIFIED_EXEC 606 vp->fp_status = FINGERPRINT_INVALID; 607 #endif 608 } 609 vp->v_type = VNON; 610 vp->v_vnlock = &vp->v_lock; 611 lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); 612 KASSERT(LIST_EMPTY(&vp->v_nclist)); 613 KASSERT(LIST_EMPTY(&vp->v_dnclist)); 614 vp->v_tag = tag; 615 vp->v_op = vops; 616 insmntque(vp, mp); 617 *vpp = vp; 618 vp->v_usecount = 1; 619 vp->v_data = 0; 620 simple_lock_init(&vp->v_uobj.vmobjlock); 621 622 /* 623 * initialize uvm_object within vnode. 624 */ 625 626 uobj = &vp->v_uobj; 627 KASSERT(uobj->pgops == &uvm_vnodeops); 628 KASSERT(uobj->uo_npages == 0); 629 KASSERT(TAILQ_FIRST(&uobj->memq) == NULL); 630 vp->v_size = VSIZENOTSET; 631 632 if (mp && error != EDEADLK) 633 vfs_unbusy(mp); 634 return (0); 635 } 636 637 /* 638 * This is really just the reverse of getnewvnode(). Needed for 639 * VFS_VGET functions who may need to push back a vnode in case 640 * of a locking race. 641 */ 642 void 643 ungetnewvnode(vp) 644 struct vnode *vp; 645 { 646 #ifdef DIAGNOSTIC 647 if (vp->v_usecount != 1) 648 panic("ungetnewvnode: busy vnode"); 649 #endif 650 vp->v_usecount--; 651 insmntque(vp, NULL); 652 vp->v_type = VBAD; 653 654 simple_lock(&vp->v_interlock); 655 /* 656 * Insert at head of LRU list 657 */ 658 simple_lock(&vnode_free_list_slock); 659 if (vp->v_holdcnt > 0) 660 TAILQ_INSERT_HEAD(&vnode_hold_list, vp, v_freelist); 661 else 662 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 663 simple_unlock(&vnode_free_list_slock); 664 simple_unlock(&vp->v_interlock); 665 } 666 667 /* 668 * Move a vnode from one mount queue to another. 669 */ 670 void 671 insmntque(vp, mp) 672 struct vnode *vp; 673 struct mount *mp; 674 { 675 676 #ifdef DIAGNOSTIC 677 if ((mp != NULL) && 678 (mp->mnt_iflag & IMNT_UNMOUNT) && 679 !(mp->mnt_flag & MNT_SOFTDEP) && 680 vp->v_tag != VT_VFS) { 681 panic("insmntque into dying filesystem"); 682 } 683 #endif 684 685 simple_lock(&mntvnode_slock); 686 /* 687 * Delete from old mount point vnode list, if on one. 688 */ 689 if (vp->v_mount != NULL) 690 LIST_REMOVE(vp, v_mntvnodes); 691 /* 692 * Insert into list of vnodes for the new mount point, if available. 693 */ 694 if ((vp->v_mount = mp) != NULL) 695 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 696 simple_unlock(&mntvnode_slock); 697 } 698 699 /* 700 * Update outstanding I/O count and do wakeup if requested. 701 */ 702 void 703 vwakeup(bp) 704 struct buf *bp; 705 { 706 struct vnode *vp; 707 708 if ((vp = bp->b_vp) != NULL) { 709 /* XXX global lock hack 710 * can't use v_interlock here since this is called 711 * in interrupt context from biodone(). 712 */ 713 simple_lock(&global_v_numoutput_slock); 714 if (--vp->v_numoutput < 0) 715 panic("vwakeup: neg numoutput, vp %p", vp); 716 if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { 717 vp->v_flag &= ~VBWAIT; 718 wakeup((caddr_t)&vp->v_numoutput); 719 } 720 simple_unlock(&global_v_numoutput_slock); 721 } 722 } 723 724 /* 725 * Flush out and invalidate all buffers associated with a vnode. 726 * Called with the underlying vnode locked, which should prevent new dirty 727 * buffers from being queued. 728 */ 729 int 730 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 731 struct vnode *vp; 732 int flags; 733 struct ucred *cred; 734 struct proc *p; 735 int slpflag, slptimeo; 736 { 737 struct buf *bp, *nbp; 738 int s, error; 739 int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO | 740 (flags & V_SAVE ? PGO_CLEANIT : 0); 741 742 /* XXXUBC this doesn't look at flags or slp* */ 743 simple_lock(&vp->v_interlock); 744 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 745 if (error) { 746 return error; 747 } 748 749 if (flags & V_SAVE) { 750 error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p); 751 if (error) 752 return (error); 753 #ifdef DIAGNOSTIC 754 s = splbio(); 755 if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd)) 756 panic("vinvalbuf: dirty bufs, vp %p", vp); 757 splx(s); 758 #endif 759 } 760 761 s = splbio(); 762 763 restart: 764 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 765 nbp = LIST_NEXT(bp, b_vnbufs); 766 simple_lock(&bp->b_interlock); 767 if (bp->b_flags & B_BUSY) { 768 bp->b_flags |= B_WANTED; 769 error = ltsleep((caddr_t)bp, 770 slpflag | (PRIBIO + 1) | PNORELOCK, 771 "vinvalbuf", slptimeo, &bp->b_interlock); 772 if (error) { 773 splx(s); 774 return (error); 775 } 776 goto restart; 777 } 778 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 779 simple_unlock(&bp->b_interlock); 780 brelse(bp); 781 } 782 783 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 784 nbp = LIST_NEXT(bp, b_vnbufs); 785 simple_lock(&bp->b_interlock); 786 if (bp->b_flags & B_BUSY) { 787 bp->b_flags |= B_WANTED; 788 error = ltsleep((caddr_t)bp, 789 slpflag | (PRIBIO + 1) | PNORELOCK, 790 "vinvalbuf", slptimeo, &bp->b_interlock); 791 if (error) { 792 splx(s); 793 return (error); 794 } 795 goto restart; 796 } 797 /* 798 * XXX Since there are no node locks for NFS, I believe 799 * there is a slight chance that a delayed write will 800 * occur while sleeping just above, so check for it. 801 */ 802 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 803 #ifdef DEBUG 804 printf("buffer still DELWRI\n"); 805 #endif 806 bp->b_flags |= B_BUSY | B_VFLUSH; 807 simple_unlock(&bp->b_interlock); 808 VOP_BWRITE(bp); 809 goto restart; 810 } 811 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 812 simple_unlock(&bp->b_interlock); 813 brelse(bp); 814 } 815 816 #ifdef DIAGNOSTIC 817 if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd)) 818 panic("vinvalbuf: flush failed, vp %p", vp); 819 #endif 820 821 splx(s); 822 823 return (0); 824 } 825 826 /* 827 * Destroy any in core blocks past the truncation length. 828 * Called with the underlying vnode locked, which should prevent new dirty 829 * buffers from being queued. 830 */ 831 int 832 vtruncbuf(vp, lbn, slpflag, slptimeo) 833 struct vnode *vp; 834 daddr_t lbn; 835 int slpflag, slptimeo; 836 { 837 struct buf *bp, *nbp; 838 int s, error; 839 voff_t off; 840 841 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 842 simple_lock(&vp->v_interlock); 843 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 844 if (error) { 845 return error; 846 } 847 848 s = splbio(); 849 850 restart: 851 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 852 nbp = LIST_NEXT(bp, b_vnbufs); 853 if (bp->b_lblkno < lbn) 854 continue; 855 simple_lock(&bp->b_interlock); 856 if (bp->b_flags & B_BUSY) { 857 bp->b_flags |= B_WANTED; 858 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 859 "vtruncbuf", slptimeo, &bp->b_interlock); 860 if (error) { 861 splx(s); 862 return (error); 863 } 864 goto restart; 865 } 866 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 867 simple_unlock(&bp->b_interlock); 868 brelse(bp); 869 } 870 871 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 872 nbp = LIST_NEXT(bp, b_vnbufs); 873 if (bp->b_lblkno < lbn) 874 continue; 875 simple_lock(&bp->b_interlock); 876 if (bp->b_flags & B_BUSY) { 877 bp->b_flags |= B_WANTED; 878 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 879 "vtruncbuf", slptimeo, &bp->b_interlock); 880 if (error) { 881 splx(s); 882 return (error); 883 } 884 goto restart; 885 } 886 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 887 simple_unlock(&bp->b_interlock); 888 brelse(bp); 889 } 890 891 splx(s); 892 893 return (0); 894 } 895 896 void 897 vflushbuf(vp, sync) 898 struct vnode *vp; 899 int sync; 900 { 901 struct buf *bp, *nbp; 902 int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0); 903 int s; 904 905 simple_lock(&vp->v_interlock); 906 (void) VOP_PUTPAGES(vp, 0, 0, flags); 907 908 loop: 909 s = splbio(); 910 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 911 nbp = LIST_NEXT(bp, b_vnbufs); 912 simple_lock(&bp->b_interlock); 913 if ((bp->b_flags & B_BUSY)) { 914 simple_unlock(&bp->b_interlock); 915 continue; 916 } 917 if ((bp->b_flags & B_DELWRI) == 0) 918 panic("vflushbuf: not dirty, bp %p", bp); 919 bp->b_flags |= B_BUSY | B_VFLUSH; 920 simple_unlock(&bp->b_interlock); 921 splx(s); 922 /* 923 * Wait for I/O associated with indirect blocks to complete, 924 * since there is no way to quickly wait for them below. 925 */ 926 if (bp->b_vp == vp || sync == 0) 927 (void) bawrite(bp); 928 else 929 (void) bwrite(bp); 930 goto loop; 931 } 932 if (sync == 0) { 933 splx(s); 934 return; 935 } 936 simple_lock(&global_v_numoutput_slock); 937 while (vp->v_numoutput) { 938 vp->v_flag |= VBWAIT; 939 ltsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0, 940 &global_v_numoutput_slock); 941 } 942 simple_unlock(&global_v_numoutput_slock); 943 splx(s); 944 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) { 945 vprint("vflushbuf: dirty", vp); 946 goto loop; 947 } 948 } 949 950 /* 951 * Associate a buffer with a vnode. 952 */ 953 void 954 bgetvp(vp, bp) 955 struct vnode *vp; 956 struct buf *bp; 957 { 958 int s; 959 960 if (bp->b_vp) 961 panic("bgetvp: not free, bp %p", bp); 962 VHOLD(vp); 963 s = splbio(); 964 bp->b_vp = vp; 965 if (vp->v_type == VBLK || vp->v_type == VCHR) 966 bp->b_dev = vp->v_rdev; 967 else 968 bp->b_dev = NODEV; 969 /* 970 * Insert onto list for new vnode. 971 */ 972 bufinsvn(bp, &vp->v_cleanblkhd); 973 splx(s); 974 } 975 976 /* 977 * Disassociate a buffer from a vnode. 978 */ 979 void 980 brelvp(bp) 981 struct buf *bp; 982 { 983 struct vnode *vp; 984 int s; 985 986 if (bp->b_vp == NULL) 987 panic("brelvp: vp NULL, bp %p", bp); 988 989 s = splbio(); 990 vp = bp->b_vp; 991 /* 992 * Delete from old vnode list, if on one. 993 */ 994 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 995 bufremvn(bp); 996 997 if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_flag & VONWORKLST) && 998 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 999 vp->v_flag &= ~VONWORKLST; 1000 LIST_REMOVE(vp, v_synclist); 1001 } 1002 1003 bp->b_vp = NULL; 1004 HOLDRELE(vp); 1005 splx(s); 1006 } 1007 1008 /* 1009 * Reassign a buffer from one vnode to another. 1010 * Used to assign file specific control information 1011 * (indirect blocks) to the vnode to which they belong. 1012 * 1013 * This function must be called at splbio(). 1014 */ 1015 void 1016 reassignbuf(bp, newvp) 1017 struct buf *bp; 1018 struct vnode *newvp; 1019 { 1020 struct buflists *listheadp; 1021 int delay; 1022 1023 /* 1024 * Delete from old vnode list, if on one. 1025 */ 1026 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 1027 bufremvn(bp); 1028 /* 1029 * If dirty, put on list of dirty buffers; 1030 * otherwise insert onto list of clean buffers. 1031 */ 1032 if ((bp->b_flags & B_DELWRI) == 0) { 1033 listheadp = &newvp->v_cleanblkhd; 1034 if (TAILQ_EMPTY(&newvp->v_uobj.memq) && 1035 (newvp->v_flag & VONWORKLST) && 1036 LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) { 1037 newvp->v_flag &= ~VONWORKLST; 1038 LIST_REMOVE(newvp, v_synclist); 1039 } 1040 } else { 1041 listheadp = &newvp->v_dirtyblkhd; 1042 if ((newvp->v_flag & VONWORKLST) == 0) { 1043 switch (newvp->v_type) { 1044 case VDIR: 1045 delay = dirdelay; 1046 break; 1047 case VBLK: 1048 if (newvp->v_specmountpoint != NULL) { 1049 delay = metadelay; 1050 break; 1051 } 1052 /* fall through */ 1053 default: 1054 delay = filedelay; 1055 break; 1056 } 1057 if (!newvp->v_mount || 1058 (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0) 1059 vn_syncer_add_to_worklist(newvp, delay); 1060 } 1061 } 1062 bufinsvn(bp, listheadp); 1063 } 1064 1065 /* 1066 * Create a vnode for a block device. 1067 * Used for root filesystem and swap areas. 1068 * Also used for memory file system special devices. 1069 */ 1070 int 1071 bdevvp(dev, vpp) 1072 dev_t dev; 1073 struct vnode **vpp; 1074 { 1075 1076 return (getdevvp(dev, vpp, VBLK)); 1077 } 1078 1079 /* 1080 * Create a vnode for a character device. 1081 * Used for kernfs and some console handling. 1082 */ 1083 int 1084 cdevvp(dev, vpp) 1085 dev_t dev; 1086 struct vnode **vpp; 1087 { 1088 1089 return (getdevvp(dev, vpp, VCHR)); 1090 } 1091 1092 /* 1093 * Create a vnode for a device. 1094 * Used by bdevvp (block device) for root file system etc., 1095 * and by cdevvp (character device) for console and kernfs. 1096 */ 1097 int 1098 getdevvp(dev, vpp, type) 1099 dev_t dev; 1100 struct vnode **vpp; 1101 enum vtype type; 1102 { 1103 struct vnode *vp; 1104 struct vnode *nvp; 1105 int error; 1106 1107 if (dev == NODEV) { 1108 *vpp = NULLVP; 1109 return (0); 1110 } 1111 error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp); 1112 if (error) { 1113 *vpp = NULLVP; 1114 return (error); 1115 } 1116 vp = nvp; 1117 vp->v_type = type; 1118 if ((nvp = checkalias(vp, dev, NULL)) != 0) { 1119 vput(vp); 1120 vp = nvp; 1121 } 1122 *vpp = vp; 1123 return (0); 1124 } 1125 1126 /* 1127 * Check to see if the new vnode represents a special device 1128 * for which we already have a vnode (either because of 1129 * bdevvp() or because of a different vnode representing 1130 * the same block device). If such an alias exists, deallocate 1131 * the existing contents and return the aliased vnode. The 1132 * caller is responsible for filling it with its new contents. 1133 */ 1134 struct vnode * 1135 checkalias(nvp, nvp_rdev, mp) 1136 struct vnode *nvp; 1137 dev_t nvp_rdev; 1138 struct mount *mp; 1139 { 1140 struct proc *p = curproc; /* XXX */ 1141 struct vnode *vp; 1142 struct vnode **vpp; 1143 1144 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1145 return (NULLVP); 1146 1147 vpp = &speclisth[SPECHASH(nvp_rdev)]; 1148 loop: 1149 simple_lock(&spechash_slock); 1150 for (vp = *vpp; vp; vp = vp->v_specnext) { 1151 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 1152 continue; 1153 /* 1154 * Alias, but not in use, so flush it out. 1155 */ 1156 simple_lock(&vp->v_interlock); 1157 if (vp->v_usecount == 0) { 1158 simple_unlock(&spechash_slock); 1159 vgonel(vp, p); 1160 goto loop; 1161 } 1162 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_NOWAIT)) { 1163 simple_unlock(&spechash_slock); 1164 goto loop; 1165 } 1166 break; 1167 } 1168 if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) { 1169 MALLOC(nvp->v_specinfo, struct specinfo *, 1170 sizeof(struct specinfo), M_VNODE, M_NOWAIT); 1171 /* XXX Erg. */ 1172 if (nvp->v_specinfo == NULL) { 1173 simple_unlock(&spechash_slock); 1174 uvm_wait("checkalias"); 1175 goto loop; 1176 } 1177 1178 nvp->v_rdev = nvp_rdev; 1179 nvp->v_hashchain = vpp; 1180 nvp->v_specnext = *vpp; 1181 nvp->v_specmountpoint = NULL; 1182 simple_unlock(&spechash_slock); 1183 nvp->v_speclockf = NULL; 1184 simple_lock_init(&nvp->v_spec_cow_slock); 1185 SLIST_INIT(&nvp->v_spec_cow_head); 1186 nvp->v_spec_cow_req = 0; 1187 nvp->v_spec_cow_count = 0; 1188 1189 *vpp = nvp; 1190 if (vp != NULLVP) { 1191 nvp->v_flag |= VALIASED; 1192 vp->v_flag |= VALIASED; 1193 vput(vp); 1194 } 1195 return (NULLVP); 1196 } 1197 simple_unlock(&spechash_slock); 1198 VOP_UNLOCK(vp, 0); 1199 simple_lock(&vp->v_interlock); 1200 vclean(vp, 0, p); 1201 vp->v_op = nvp->v_op; 1202 vp->v_tag = nvp->v_tag; 1203 vp->v_vnlock = &vp->v_lock; 1204 lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); 1205 nvp->v_type = VNON; 1206 insmntque(vp, mp); 1207 return (vp); 1208 } 1209 1210 /* 1211 * Grab a particular vnode from the free list, increment its 1212 * reference count and lock it. If the vnode lock bit is set the 1213 * vnode is being eliminated in vgone. In that case, we can not 1214 * grab the vnode, so the process is awakened when the transition is 1215 * completed, and an error returned to indicate that the vnode is no 1216 * longer usable (possibly having been changed to a new file system type). 1217 */ 1218 int 1219 vget(vp, flags) 1220 struct vnode *vp; 1221 int flags; 1222 { 1223 int error; 1224 1225 /* 1226 * If the vnode is in the process of being cleaned out for 1227 * another use, we wait for the cleaning to finish and then 1228 * return failure. Cleaning is determined by checking that 1229 * the VXLOCK flag is set. 1230 */ 1231 1232 if ((flags & LK_INTERLOCK) == 0) 1233 simple_lock(&vp->v_interlock); 1234 if (vp->v_flag & VXLOCK) { 1235 if (flags & LK_NOWAIT) { 1236 simple_unlock(&vp->v_interlock); 1237 return EBUSY; 1238 } 1239 vp->v_flag |= VXWANT; 1240 ltsleep(vp, PINOD|PNORELOCK, "vget", 0, &vp->v_interlock); 1241 return (ENOENT); 1242 } 1243 if (vp->v_usecount == 0) { 1244 simple_lock(&vnode_free_list_slock); 1245 if (vp->v_holdcnt > 0) 1246 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 1247 else 1248 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1249 simple_unlock(&vnode_free_list_slock); 1250 } 1251 vp->v_usecount++; 1252 #ifdef DIAGNOSTIC 1253 if (vp->v_usecount == 0) { 1254 vprint("vget", vp); 1255 panic("vget: usecount overflow, vp %p", vp); 1256 } 1257 #endif 1258 if (flags & LK_TYPE_MASK) { 1259 if ((error = vn_lock(vp, flags | LK_INTERLOCK))) { 1260 /* 1261 * must expand vrele here because we do not want 1262 * to call VOP_INACTIVE if the reference count 1263 * drops back to zero since it was never really 1264 * active. We must remove it from the free list 1265 * before sleeping so that multiple processes do 1266 * not try to recycle it. 1267 */ 1268 simple_lock(&vp->v_interlock); 1269 vp->v_usecount--; 1270 if (vp->v_usecount > 0) { 1271 simple_unlock(&vp->v_interlock); 1272 return (error); 1273 } 1274 /* 1275 * insert at tail of LRU list 1276 */ 1277 simple_lock(&vnode_free_list_slock); 1278 if (vp->v_holdcnt > 0) 1279 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, 1280 v_freelist); 1281 else 1282 TAILQ_INSERT_TAIL(&vnode_free_list, vp, 1283 v_freelist); 1284 simple_unlock(&vnode_free_list_slock); 1285 simple_unlock(&vp->v_interlock); 1286 } 1287 return (error); 1288 } 1289 simple_unlock(&vp->v_interlock); 1290 return (0); 1291 } 1292 1293 /* 1294 * vput(), just unlock and vrele() 1295 */ 1296 void 1297 vput(vp) 1298 struct vnode *vp; 1299 { 1300 struct proc *p = curproc; /* XXX */ 1301 1302 #ifdef DIAGNOSTIC 1303 if (vp == NULL) 1304 panic("vput: null vp"); 1305 #endif 1306 simple_lock(&vp->v_interlock); 1307 vp->v_usecount--; 1308 if (vp->v_usecount > 0) { 1309 simple_unlock(&vp->v_interlock); 1310 VOP_UNLOCK(vp, 0); 1311 return; 1312 } 1313 #ifdef DIAGNOSTIC 1314 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1315 vprint("vput: bad ref count", vp); 1316 panic("vput: ref cnt"); 1317 } 1318 #endif 1319 /* 1320 * Insert at tail of LRU list. 1321 */ 1322 simple_lock(&vnode_free_list_slock); 1323 if (vp->v_holdcnt > 0) 1324 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1325 else 1326 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1327 simple_unlock(&vnode_free_list_slock); 1328 if (vp->v_flag & VEXECMAP) { 1329 uvmexp.execpages -= vp->v_uobj.uo_npages; 1330 uvmexp.filepages += vp->v_uobj.uo_npages; 1331 } 1332 vp->v_flag &= ~(VTEXT|VEXECMAP); 1333 simple_unlock(&vp->v_interlock); 1334 VOP_INACTIVE(vp, p); 1335 } 1336 1337 /* 1338 * Vnode release. 1339 * If count drops to zero, call inactive routine and return to freelist. 1340 */ 1341 void 1342 vrele(vp) 1343 struct vnode *vp; 1344 { 1345 struct proc *p = curproc; /* XXX */ 1346 1347 #ifdef DIAGNOSTIC 1348 if (vp == NULL) 1349 panic("vrele: null vp"); 1350 #endif 1351 simple_lock(&vp->v_interlock); 1352 vp->v_usecount--; 1353 if (vp->v_usecount > 0) { 1354 simple_unlock(&vp->v_interlock); 1355 return; 1356 } 1357 #ifdef DIAGNOSTIC 1358 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1359 vprint("vrele: bad ref count", vp); 1360 panic("vrele: ref cnt vp %p", vp); 1361 } 1362 #endif 1363 /* 1364 * Insert at tail of LRU list. 1365 */ 1366 simple_lock(&vnode_free_list_slock); 1367 if (vp->v_holdcnt > 0) 1368 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1369 else 1370 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1371 simple_unlock(&vnode_free_list_slock); 1372 if (vp->v_flag & VEXECMAP) { 1373 uvmexp.execpages -= vp->v_uobj.uo_npages; 1374 uvmexp.filepages += vp->v_uobj.uo_npages; 1375 } 1376 vp->v_flag &= ~(VTEXT|VEXECMAP); 1377 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) 1378 VOP_INACTIVE(vp, p); 1379 } 1380 1381 #ifdef DIAGNOSTIC 1382 /* 1383 * Page or buffer structure gets a reference. 1384 */ 1385 void 1386 vholdl(vp) 1387 struct vnode *vp; 1388 { 1389 1390 /* 1391 * If it is on the freelist and the hold count is currently 1392 * zero, move it to the hold list. The test of the back 1393 * pointer and the use reference count of zero is because 1394 * it will be removed from a free list by getnewvnode, 1395 * but will not have its reference count incremented until 1396 * after calling vgone. If the reference count were 1397 * incremented first, vgone would (incorrectly) try to 1398 * close the previous instance of the underlying object. 1399 * So, the back pointer is explicitly set to `0xdeadb' in 1400 * getnewvnode after removing it from a freelist to ensure 1401 * that we do not try to move it here. 1402 */ 1403 if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) && 1404 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 1405 simple_lock(&vnode_free_list_slock); 1406 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1407 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1408 simple_unlock(&vnode_free_list_slock); 1409 } 1410 vp->v_holdcnt++; 1411 } 1412 1413 /* 1414 * Page or buffer structure frees a reference. 1415 */ 1416 void 1417 holdrelel(vp) 1418 struct vnode *vp; 1419 { 1420 1421 if (vp->v_holdcnt <= 0) 1422 panic("holdrelel: holdcnt vp %p", vp); 1423 vp->v_holdcnt--; 1424 1425 /* 1426 * If it is on the holdlist and the hold count drops to 1427 * zero, move it to the free list. The test of the back 1428 * pointer and the use reference count of zero is because 1429 * it will be removed from a free list by getnewvnode, 1430 * but will not have its reference count incremented until 1431 * after calling vgone. If the reference count were 1432 * incremented first, vgone would (incorrectly) try to 1433 * close the previous instance of the underlying object. 1434 * So, the back pointer is explicitly set to `0xdeadb' in 1435 * getnewvnode after removing it from a freelist to ensure 1436 * that we do not try to move it here. 1437 */ 1438 1439 if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) && 1440 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 1441 simple_lock(&vnode_free_list_slock); 1442 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 1443 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1444 simple_unlock(&vnode_free_list_slock); 1445 } 1446 } 1447 1448 /* 1449 * Vnode reference. 1450 */ 1451 void 1452 vref(vp) 1453 struct vnode *vp; 1454 { 1455 1456 simple_lock(&vp->v_interlock); 1457 if (vp->v_usecount <= 0) 1458 panic("vref used where vget required, vp %p", vp); 1459 vp->v_usecount++; 1460 #ifdef DIAGNOSTIC 1461 if (vp->v_usecount == 0) { 1462 vprint("vref", vp); 1463 panic("vref: usecount overflow, vp %p", vp); 1464 } 1465 #endif 1466 simple_unlock(&vp->v_interlock); 1467 } 1468 #endif /* DIAGNOSTIC */ 1469 1470 /* 1471 * Remove any vnodes in the vnode table belonging to mount point mp. 1472 * 1473 * If FORCECLOSE is not specified, there should not be any active ones, 1474 * return error if any are found (nb: this is a user error, not a 1475 * system error). If FORCECLOSE is specified, detach any active vnodes 1476 * that are found. 1477 * 1478 * If WRITECLOSE is set, only flush out regular file vnodes open for 1479 * writing. 1480 * 1481 * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped. 1482 */ 1483 #ifdef DEBUG 1484 int busyprt = 0; /* print out busy vnodes */ 1485 struct ctldebug debug1 = { "busyprt", &busyprt }; 1486 #endif 1487 1488 int 1489 vflush(mp, skipvp, flags) 1490 struct mount *mp; 1491 struct vnode *skipvp; 1492 int flags; 1493 { 1494 struct proc *p = curproc; /* XXX */ 1495 struct vnode *vp, *nvp; 1496 int busy = 0; 1497 1498 simple_lock(&mntvnode_slock); 1499 loop: 1500 for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) { 1501 if (vp->v_mount != mp) 1502 goto loop; 1503 nvp = LIST_NEXT(vp, v_mntvnodes); 1504 /* 1505 * Skip over a selected vnode. 1506 */ 1507 if (vp == skipvp) 1508 continue; 1509 simple_lock(&vp->v_interlock); 1510 /* 1511 * Skip over a vnodes marked VSYSTEM. 1512 */ 1513 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1514 simple_unlock(&vp->v_interlock); 1515 continue; 1516 } 1517 /* 1518 * If WRITECLOSE is set, only flush out regular file 1519 * vnodes open for writing. 1520 */ 1521 if ((flags & WRITECLOSE) && 1522 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1523 simple_unlock(&vp->v_interlock); 1524 continue; 1525 } 1526 /* 1527 * With v_usecount == 0, all we need to do is clear 1528 * out the vnode data structures and we are done. 1529 */ 1530 if (vp->v_usecount == 0) { 1531 simple_unlock(&mntvnode_slock); 1532 vgonel(vp, p); 1533 simple_lock(&mntvnode_slock); 1534 continue; 1535 } 1536 /* 1537 * If FORCECLOSE is set, forcibly close the vnode. 1538 * For block or character devices, revert to an 1539 * anonymous device. For all other files, just kill them. 1540 */ 1541 if (flags & FORCECLOSE) { 1542 simple_unlock(&mntvnode_slock); 1543 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1544 vgonel(vp, p); 1545 } else { 1546 vclean(vp, 0, p); 1547 vp->v_op = spec_vnodeop_p; 1548 insmntque(vp, (struct mount *)0); 1549 } 1550 simple_lock(&mntvnode_slock); 1551 continue; 1552 } 1553 #ifdef DEBUG 1554 if (busyprt) 1555 vprint("vflush: busy vnode", vp); 1556 #endif 1557 simple_unlock(&vp->v_interlock); 1558 busy++; 1559 } 1560 simple_unlock(&mntvnode_slock); 1561 if (busy) 1562 return (EBUSY); 1563 return (0); 1564 } 1565 1566 /* 1567 * Disassociate the underlying file system from a vnode. 1568 */ 1569 void 1570 vclean(vp, flags, p) 1571 struct vnode *vp; 1572 int flags; 1573 struct proc *p; 1574 { 1575 struct mount *mp; 1576 int active; 1577 1578 LOCK_ASSERT(simple_lock_held(&vp->v_interlock)); 1579 1580 /* 1581 * Check to see if the vnode is in use. 1582 * If so we have to reference it before we clean it out 1583 * so that its count cannot fall to zero and generate a 1584 * race against ourselves to recycle it. 1585 */ 1586 1587 if ((active = vp->v_usecount) != 0) { 1588 vp->v_usecount++; 1589 #ifdef DIAGNOSTIC 1590 if (vp->v_usecount == 0) { 1591 vprint("vclean", vp); 1592 panic("vclean: usecount overflow"); 1593 } 1594 #endif 1595 } 1596 1597 /* 1598 * Prevent the vnode from being recycled or 1599 * brought into use while we clean it out. 1600 */ 1601 if (vp->v_flag & VXLOCK) 1602 panic("vclean: deadlock, vp %p", vp); 1603 vp->v_flag |= VXLOCK; 1604 if (vp->v_flag & VEXECMAP) { 1605 uvmexp.execpages -= vp->v_uobj.uo_npages; 1606 uvmexp.filepages += vp->v_uobj.uo_npages; 1607 } 1608 vp->v_flag &= ~(VTEXT|VEXECMAP); 1609 1610 /* 1611 * Even if the count is zero, the VOP_INACTIVE routine may still 1612 * have the object locked while it cleans it out. The VOP_LOCK 1613 * ensures that the VOP_INACTIVE routine is done with its work. 1614 * For active vnodes, it ensures that no other activity can 1615 * occur while the underlying object is being cleaned out. 1616 */ 1617 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK); 1618 1619 /* 1620 * Clean out any cached data associated with the vnode. 1621 */ 1622 if (flags & DOCLOSE) { 1623 int error; 1624 vn_start_write(vp, &mp, V_WAIT | V_LOWER); 1625 error = vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1626 vn_finished_write(mp, V_LOWER); 1627 if (error) 1628 error = vinvalbuf(vp, 0, NOCRED, p, 0, 0); 1629 KASSERT(error == 0); 1630 KASSERT((vp->v_flag & VONWORKLST) == 0); 1631 } 1632 LOCK_ASSERT(!simple_lock_held(&vp->v_interlock)); 1633 1634 /* 1635 * If purging an active vnode, it must be closed and 1636 * deactivated before being reclaimed. Note that the 1637 * VOP_INACTIVE will unlock the vnode. 1638 */ 1639 if (active) { 1640 if (flags & DOCLOSE) 1641 VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL); 1642 VOP_INACTIVE(vp, p); 1643 } else { 1644 /* 1645 * Any other processes trying to obtain this lock must first 1646 * wait for VXLOCK to clear, then call the new lock operation. 1647 */ 1648 VOP_UNLOCK(vp, 0); 1649 } 1650 /* 1651 * Reclaim the vnode. 1652 */ 1653 if (VOP_RECLAIM(vp, p)) 1654 panic("vclean: cannot reclaim, vp %p", vp); 1655 if (active) { 1656 /* 1657 * Inline copy of vrele() since VOP_INACTIVE 1658 * has already been called. 1659 */ 1660 simple_lock(&vp->v_interlock); 1661 if (--vp->v_usecount <= 0) { 1662 #ifdef DIAGNOSTIC 1663 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1664 vprint("vclean: bad ref count", vp); 1665 panic("vclean: ref cnt"); 1666 } 1667 #endif 1668 /* 1669 * Insert at tail of LRU list. 1670 */ 1671 1672 simple_unlock(&vp->v_interlock); 1673 simple_lock(&vnode_free_list_slock); 1674 #ifdef DIAGNOSTIC 1675 if (vp->v_holdcnt > 0) 1676 panic("vclean: not clean, vp %p", vp); 1677 #endif 1678 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1679 simple_unlock(&vnode_free_list_slock); 1680 } else 1681 simple_unlock(&vp->v_interlock); 1682 } 1683 1684 KASSERT(vp->v_uobj.uo_npages == 0); 1685 cache_purge(vp); 1686 1687 /* 1688 * Done with purge, notify sleepers of the grim news. 1689 */ 1690 vp->v_op = dead_vnodeop_p; 1691 vp->v_tag = VT_NON; 1692 simple_lock(&vp->v_interlock); 1693 VN_KNOTE(vp, NOTE_REVOKE); /* FreeBSD has this in vn_pollgone() */ 1694 vp->v_flag &= ~VXLOCK; 1695 if (vp->v_flag & VXWANT) { 1696 vp->v_flag &= ~VXWANT; 1697 simple_unlock(&vp->v_interlock); 1698 wakeup((caddr_t)vp); 1699 } else 1700 simple_unlock(&vp->v_interlock); 1701 } 1702 1703 /* 1704 * Recycle an unused vnode to the front of the free list. 1705 * Release the passed interlock if the vnode will be recycled. 1706 */ 1707 int 1708 vrecycle(vp, inter_lkp, p) 1709 struct vnode *vp; 1710 struct simplelock *inter_lkp; 1711 struct proc *p; 1712 { 1713 1714 simple_lock(&vp->v_interlock); 1715 if (vp->v_usecount == 0) { 1716 if (inter_lkp) 1717 simple_unlock(inter_lkp); 1718 vgonel(vp, p); 1719 return (1); 1720 } 1721 simple_unlock(&vp->v_interlock); 1722 return (0); 1723 } 1724 1725 /* 1726 * Eliminate all activity associated with a vnode 1727 * in preparation for reuse. 1728 */ 1729 void 1730 vgone(vp) 1731 struct vnode *vp; 1732 { 1733 struct proc *p = curproc; /* XXX */ 1734 1735 simple_lock(&vp->v_interlock); 1736 vgonel(vp, p); 1737 } 1738 1739 /* 1740 * vgone, with the vp interlock held. 1741 */ 1742 void 1743 vgonel(vp, p) 1744 struct vnode *vp; 1745 struct proc *p; 1746 { 1747 struct vnode *vq; 1748 struct vnode *vx; 1749 1750 LOCK_ASSERT(simple_lock_held(&vp->v_interlock)); 1751 1752 /* 1753 * If a vgone (or vclean) is already in progress, 1754 * wait until it is done and return. 1755 */ 1756 1757 if (vp->v_flag & VXLOCK) { 1758 vp->v_flag |= VXWANT; 1759 ltsleep(vp, PINOD | PNORELOCK, "vgone", 0, &vp->v_interlock); 1760 return; 1761 } 1762 1763 /* 1764 * Clean out the filesystem specific data. 1765 */ 1766 1767 vclean(vp, DOCLOSE, p); 1768 KASSERT((vp->v_flag & VONWORKLST) == 0); 1769 1770 /* 1771 * Delete from old mount point vnode list, if on one. 1772 */ 1773 1774 if (vp->v_mount != NULL) 1775 insmntque(vp, (struct mount *)0); 1776 1777 /* 1778 * If special device, remove it from special device alias list. 1779 * if it is on one. 1780 */ 1781 1782 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1783 simple_lock(&spechash_slock); 1784 if (vp->v_hashchain != NULL) { 1785 if (*vp->v_hashchain == vp) { 1786 *vp->v_hashchain = vp->v_specnext; 1787 } else { 1788 for (vq = *vp->v_hashchain; vq; 1789 vq = vq->v_specnext) { 1790 if (vq->v_specnext != vp) 1791 continue; 1792 vq->v_specnext = vp->v_specnext; 1793 break; 1794 } 1795 if (vq == NULL) 1796 panic("missing bdev"); 1797 } 1798 if (vp->v_flag & VALIASED) { 1799 vx = NULL; 1800 for (vq = *vp->v_hashchain; vq; 1801 vq = vq->v_specnext) { 1802 if (vq->v_rdev != vp->v_rdev || 1803 vq->v_type != vp->v_type) 1804 continue; 1805 if (vx) 1806 break; 1807 vx = vq; 1808 } 1809 if (vx == NULL) 1810 panic("missing alias"); 1811 if (vq == NULL) 1812 vx->v_flag &= ~VALIASED; 1813 vp->v_flag &= ~VALIASED; 1814 } 1815 } 1816 simple_unlock(&spechash_slock); 1817 FREE(vp->v_specinfo, M_VNODE); 1818 vp->v_specinfo = NULL; 1819 } 1820 1821 /* 1822 * The test of the back pointer and the reference count of 1823 * zero is because it will be removed from the free list by 1824 * getcleanvnode, but will not have its reference count 1825 * incremented until after calling vgone. If the reference 1826 * count were incremented first, vgone would (incorrectly) 1827 * try to close the previous instance of the underlying object. 1828 * So, the back pointer is explicitly set to `0xdeadb' in 1829 * getnewvnode after removing it from the freelist to ensure 1830 * that we do not try to move it here. 1831 */ 1832 1833 vp->v_type = VBAD; 1834 if (vp->v_usecount == 0) { 1835 boolean_t dofree; 1836 1837 simple_lock(&vnode_free_list_slock); 1838 if (vp->v_holdcnt > 0) 1839 panic("vgonel: not clean, vp %p", vp); 1840 /* 1841 * if it isn't on the freelist, we're called by getcleanvnode 1842 * and vnode is being re-used. otherwise, we'll free it. 1843 */ 1844 dofree = vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb; 1845 if (dofree) { 1846 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1847 numvnodes--; 1848 } 1849 simple_unlock(&vnode_free_list_slock); 1850 if (dofree) 1851 pool_put(&vnode_pool, vp); 1852 } 1853 } 1854 1855 /* 1856 * Lookup a vnode by device number. 1857 */ 1858 int 1859 vfinddev(dev, type, vpp) 1860 dev_t dev; 1861 enum vtype type; 1862 struct vnode **vpp; 1863 { 1864 struct vnode *vp; 1865 int rc = 0; 1866 1867 simple_lock(&spechash_slock); 1868 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1869 if (dev != vp->v_rdev || type != vp->v_type) 1870 continue; 1871 *vpp = vp; 1872 rc = 1; 1873 break; 1874 } 1875 simple_unlock(&spechash_slock); 1876 return (rc); 1877 } 1878 1879 /* 1880 * Revoke all the vnodes corresponding to the specified minor number 1881 * range (endpoints inclusive) of the specified major. 1882 */ 1883 void 1884 vdevgone(maj, minl, minh, type) 1885 int maj, minl, minh; 1886 enum vtype type; 1887 { 1888 struct vnode *vp; 1889 int mn; 1890 1891 for (mn = minl; mn <= minh; mn++) 1892 if (vfinddev(makedev(maj, mn), type, &vp)) 1893 VOP_REVOKE(vp, REVOKEALL); 1894 } 1895 1896 /* 1897 * Calculate the total number of references to a special device. 1898 */ 1899 int 1900 vcount(vp) 1901 struct vnode *vp; 1902 { 1903 struct vnode *vq, *vnext; 1904 int count; 1905 1906 loop: 1907 if ((vp->v_flag & VALIASED) == 0) 1908 return (vp->v_usecount); 1909 simple_lock(&spechash_slock); 1910 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1911 vnext = vq->v_specnext; 1912 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1913 continue; 1914 /* 1915 * Alias, but not in use, so flush it out. 1916 */ 1917 if (vq->v_usecount == 0 && vq != vp && 1918 (vq->v_flag & VXLOCK) == 0) { 1919 simple_unlock(&spechash_slock); 1920 vgone(vq); 1921 goto loop; 1922 } 1923 count += vq->v_usecount; 1924 } 1925 simple_unlock(&spechash_slock); 1926 return (count); 1927 } 1928 1929 /* 1930 * Print out a description of a vnode. 1931 */ 1932 const char * const vnode_types[] = { 1933 "VNON", 1934 "VREG", 1935 "VDIR", 1936 "VBLK", 1937 "VCHR", 1938 "VLNK", 1939 "VSOCK", 1940 "VFIFO", 1941 "VBAD" 1942 }; 1943 1944 void 1945 vprint(label, vp) 1946 char *label; 1947 struct vnode *vp; 1948 { 1949 char buf[96]; 1950 1951 if (label != NULL) 1952 printf("%s: ", label); 1953 printf("tag %d type %s, usecount %d, writecount %ld, refcount %ld,", 1954 vp->v_tag, vnode_types[vp->v_type], 1955 vp->v_usecount, vp->v_writecount, vp->v_holdcnt); 1956 buf[0] = '\0'; 1957 if (vp->v_flag & VROOT) 1958 strlcat(buf, "|VROOT", sizeof(buf)); 1959 if (vp->v_flag & VTEXT) 1960 strlcat(buf, "|VTEXT", sizeof(buf)); 1961 if (vp->v_flag & VEXECMAP) 1962 strlcat(buf, "|VEXECMAP", sizeof(buf)); 1963 if (vp->v_flag & VSYSTEM) 1964 strlcat(buf, "|VSYSTEM", sizeof(buf)); 1965 if (vp->v_flag & VXLOCK) 1966 strlcat(buf, "|VXLOCK", sizeof(buf)); 1967 if (vp->v_flag & VXWANT) 1968 strlcat(buf, "|VXWANT", sizeof(buf)); 1969 if (vp->v_flag & VBWAIT) 1970 strlcat(buf, "|VBWAIT", sizeof(buf)); 1971 if (vp->v_flag & VALIASED) 1972 strlcat(buf, "|VALIASED", sizeof(buf)); 1973 if (buf[0] != '\0') 1974 printf(" flags (%s)", &buf[1]); 1975 if (vp->v_data == NULL) { 1976 printf("\n"); 1977 } else { 1978 printf("\n\t"); 1979 VOP_PRINT(vp); 1980 } 1981 } 1982 1983 #ifdef DEBUG 1984 /* 1985 * List all of the locked vnodes in the system. 1986 * Called when debugging the kernel. 1987 */ 1988 void 1989 printlockedvnodes() 1990 { 1991 struct mount *mp, *nmp; 1992 struct vnode *vp; 1993 1994 printf("Locked vnodes\n"); 1995 simple_lock(&mountlist_slock); 1996 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 1997 mp = nmp) { 1998 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 1999 nmp = CIRCLEQ_NEXT(mp, mnt_list); 2000 continue; 2001 } 2002 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2003 if (VOP_ISLOCKED(vp)) 2004 vprint(NULL, vp); 2005 } 2006 simple_lock(&mountlist_slock); 2007 nmp = CIRCLEQ_NEXT(mp, mnt_list); 2008 vfs_unbusy(mp); 2009 } 2010 simple_unlock(&mountlist_slock); 2011 } 2012 #endif 2013 2014 /* 2015 * sysctl helper routine for vfs.generic.conf lookups. 2016 */ 2017 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 2018 static int 2019 sysctl_vfs_generic_conf(SYSCTLFN_ARGS) 2020 { 2021 struct vfsconf vfc; 2022 extern const char * const mountcompatnames[]; 2023 extern int nmountcompatnames; 2024 struct sysctlnode node; 2025 struct vfsops *vfsp; 2026 u_int vfsnum; 2027 2028 if (namelen != 1) 2029 return (ENOTDIR); 2030 vfsnum = name[0]; 2031 if (vfsnum >= nmountcompatnames || 2032 mountcompatnames[vfsnum] == NULL) 2033 return (EOPNOTSUPP); 2034 vfsp = vfs_getopsbyname(mountcompatnames[vfsnum]); 2035 if (vfsp == NULL) 2036 return (EOPNOTSUPP); 2037 2038 vfc.vfc_vfsops = vfsp; 2039 strncpy(vfc.vfc_name, vfsp->vfs_name, MFSNAMELEN); 2040 vfc.vfc_typenum = vfsnum; 2041 vfc.vfc_refcount = vfsp->vfs_refcount; 2042 vfc.vfc_flags = 0; 2043 vfc.vfc_mountroot = vfsp->vfs_mountroot; 2044 vfc.vfc_next = NULL; 2045 2046 node = *rnode; 2047 node.sysctl_data = &vfc; 2048 return (sysctl_lookup(SYSCTLFN_CALL(&node))); 2049 } 2050 #endif 2051 2052 /* 2053 * sysctl helper routine to return list of supported fstypes 2054 */ 2055 static int 2056 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS) 2057 { 2058 char buf[MFSNAMELEN]; 2059 char *where = oldp; 2060 struct vfsops *v; 2061 size_t needed, left, slen; 2062 int error, first; 2063 2064 if (newp != NULL) 2065 return (EPERM); 2066 if (namelen != 0) 2067 return (EINVAL); 2068 2069 first = 1; 2070 error = 0; 2071 needed = 0; 2072 left = *oldlenp; 2073 2074 LIST_FOREACH(v, &vfs_list, vfs_list) { 2075 if (where == NULL) 2076 needed += strlen(v->vfs_name) + 1; 2077 else { 2078 memset(buf, 0, sizeof(buf)); 2079 if (first) { 2080 strncpy(buf, v->vfs_name, sizeof(buf)); 2081 first = 0; 2082 } else { 2083 buf[0] = ' '; 2084 strncpy(buf + 1, v->vfs_name, sizeof(buf) - 1); 2085 } 2086 buf[sizeof(buf)-1] = '\0'; 2087 slen = strlen(buf); 2088 if (left < slen + 1) 2089 break; 2090 /* +1 to copy out the trailing NUL byte */ 2091 error = copyout(buf, where, slen + 1); 2092 if (error) 2093 break; 2094 where += slen; 2095 needed += slen; 2096 left -= slen; 2097 } 2098 } 2099 *oldlenp = needed; 2100 return (error); 2101 } 2102 2103 /* 2104 * Top level filesystem related information gathering. 2105 */ 2106 SYSCTL_SETUP(sysctl_vfs_setup, "sysctl vfs subtree setup") 2107 { 2108 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 2109 extern int nmountcompatnames; 2110 #endif 2111 2112 sysctl_createv(clog, 0, NULL, NULL, 2113 CTLFLAG_PERMANENT, 2114 CTLTYPE_NODE, "vfs", NULL, 2115 NULL, 0, NULL, 0, 2116 CTL_VFS, CTL_EOL); 2117 sysctl_createv(clog, 0, NULL, NULL, 2118 CTLFLAG_PERMANENT, 2119 CTLTYPE_NODE, "generic", 2120 SYSCTL_DESCR("Non-specific vfs related information"), 2121 NULL, 0, NULL, 0, 2122 CTL_VFS, VFS_GENERIC, CTL_EOL); 2123 2124 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 2125 sysctl_createv(clog, 0, NULL, NULL, 2126 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE, 2127 CTLTYPE_INT, "maxtypenum", 2128 SYSCTL_DESCR("Highest valid filesystem type number"), 2129 NULL, nmountcompatnames, NULL, 0, 2130 CTL_VFS, VFS_GENERIC, VFS_MAXTYPENUM, CTL_EOL); 2131 #endif 2132 sysctl_createv(clog, 0, NULL, NULL, 2133 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2134 CTLTYPE_INT, "usermount", 2135 SYSCTL_DESCR("Whether unprivileged users may mount " 2136 "filesystems"), 2137 NULL, 0, &dovfsusermount, 0, 2138 CTL_VFS, VFS_GENERIC, VFS_USERMOUNT, CTL_EOL); 2139 sysctl_createv(clog, 0, NULL, NULL, 2140 CTLFLAG_PERMANENT, 2141 CTLTYPE_STRING, "fstypes", 2142 SYSCTL_DESCR("List of file systems present"), 2143 sysctl_vfs_generic_fstypes, 0, NULL, 0, 2144 CTL_VFS, VFS_GENERIC, CTL_CREATE, CTL_EOL); 2145 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 2146 sysctl_createv(clog, 0, NULL, NULL, 2147 CTLFLAG_PERMANENT, 2148 CTLTYPE_STRUCT, "conf", 2149 SYSCTL_DESCR("Filesystem configuration information"), 2150 sysctl_vfs_generic_conf, 0, NULL, 2151 sizeof(struct vfsconf), 2152 CTL_VFS, VFS_GENERIC, VFS_CONF, CTL_EOL); 2153 #endif 2154 } 2155 2156 2157 int kinfo_vdebug = 1; 2158 int kinfo_vgetfailed; 2159 #define KINFO_VNODESLOP 10 2160 /* 2161 * Dump vnode list (via sysctl). 2162 * Copyout address of vnode followed by vnode. 2163 */ 2164 /* ARGSUSED */ 2165 int 2166 sysctl_kern_vnode(SYSCTLFN_ARGS) 2167 { 2168 char *where = oldp; 2169 size_t *sizep = oldlenp; 2170 struct mount *mp, *nmp; 2171 struct vnode *nvp, *vp; 2172 char *bp = where, *savebp; 2173 char *ewhere; 2174 int error; 2175 2176 if (namelen != 0) 2177 return (EOPNOTSUPP); 2178 if (newp != NULL) 2179 return (EPERM); 2180 2181 #define VPTRSZ sizeof(struct vnode *) 2182 #define VNODESZ sizeof(struct vnode) 2183 if (where == NULL) { 2184 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 2185 return (0); 2186 } 2187 ewhere = where + *sizep; 2188 2189 simple_lock(&mountlist_slock); 2190 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 2191 mp = nmp) { 2192 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 2193 nmp = CIRCLEQ_NEXT(mp, mnt_list); 2194 continue; 2195 } 2196 savebp = bp; 2197 again: 2198 simple_lock(&mntvnode_slock); 2199 for (vp = LIST_FIRST(&mp->mnt_vnodelist); 2200 vp != NULL; 2201 vp = nvp) { 2202 /* 2203 * Check that the vp is still associated with 2204 * this filesystem. RACE: could have been 2205 * recycled onto the same filesystem. 2206 */ 2207 if (vp->v_mount != mp) { 2208 simple_unlock(&mntvnode_slock); 2209 if (kinfo_vdebug) 2210 printf("kinfo: vp changed\n"); 2211 bp = savebp; 2212 goto again; 2213 } 2214 nvp = LIST_NEXT(vp, v_mntvnodes); 2215 if (bp + VPTRSZ + VNODESZ > ewhere) { 2216 simple_unlock(&mntvnode_slock); 2217 *sizep = bp - where; 2218 return (ENOMEM); 2219 } 2220 simple_unlock(&mntvnode_slock); 2221 if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) || 2222 (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ))) 2223 return (error); 2224 bp += VPTRSZ + VNODESZ; 2225 simple_lock(&mntvnode_slock); 2226 } 2227 simple_unlock(&mntvnode_slock); 2228 simple_lock(&mountlist_slock); 2229 nmp = CIRCLEQ_NEXT(mp, mnt_list); 2230 vfs_unbusy(mp); 2231 } 2232 simple_unlock(&mountlist_slock); 2233 2234 *sizep = bp - where; 2235 return (0); 2236 } 2237 2238 /* 2239 * Check to see if a filesystem is mounted on a block device. 2240 */ 2241 int 2242 vfs_mountedon(vp) 2243 struct vnode *vp; 2244 { 2245 struct vnode *vq; 2246 int error = 0; 2247 2248 if (vp->v_specmountpoint != NULL) 2249 return (EBUSY); 2250 if (vp->v_flag & VALIASED) { 2251 simple_lock(&spechash_slock); 2252 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 2253 if (vq->v_rdev != vp->v_rdev || 2254 vq->v_type != vp->v_type) 2255 continue; 2256 if (vq->v_specmountpoint != NULL) { 2257 error = EBUSY; 2258 break; 2259 } 2260 } 2261 simple_unlock(&spechash_slock); 2262 } 2263 return (error); 2264 } 2265 2266 static int 2267 sacheck(struct sockaddr *sa) 2268 { 2269 switch (sa->sa_family) { 2270 #ifdef INET 2271 case AF_INET: { 2272 struct sockaddr_in *sin = (struct sockaddr_in *)sa; 2273 char *p = (char *)sin->sin_zero; 2274 size_t i; 2275 2276 if (sin->sin_len != sizeof(*sin)) 2277 return -1; 2278 if (sin->sin_port != 0) 2279 return -1; 2280 for (i = 0; i < sizeof(sin->sin_zero); i++) 2281 if (*p++ != '\0') 2282 return -1; 2283 return 0; 2284 } 2285 #endif 2286 #ifdef INET6 2287 case AF_INET6: { 2288 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa; 2289 2290 if (sin6->sin6_len != sizeof(*sin6)) 2291 return -1; 2292 if (sin6->sin6_port != 0) 2293 return -1; 2294 return 0; 2295 } 2296 #endif 2297 default: 2298 return -1; 2299 } 2300 } 2301 2302 /* 2303 * Build hash lists of net addresses and hang them off the mount point. 2304 * Called by ufs_mount() to set up the lists of export addresses. 2305 */ 2306 static int 2307 vfs_hang_addrlist(mp, nep, argp) 2308 struct mount *mp; 2309 struct netexport *nep; 2310 struct export_args *argp; 2311 { 2312 struct netcred *np, *enp; 2313 struct radix_node_head *rnh; 2314 int i; 2315 struct sockaddr *saddr, *smask = 0; 2316 struct domain *dom; 2317 int error; 2318 2319 if (argp->ex_addrlen == 0) { 2320 if (mp->mnt_flag & MNT_DEFEXPORTED) 2321 return (EPERM); 2322 np = &nep->ne_defexported; 2323 np->netc_exflags = argp->ex_flags; 2324 crcvt(&np->netc_anon, &argp->ex_anon); 2325 np->netc_anon.cr_ref = 1; 2326 mp->mnt_flag |= MNT_DEFEXPORTED; 2327 return (0); 2328 } 2329 2330 if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN) 2331 return (EINVAL); 2332 2333 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2334 np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK); 2335 memset((caddr_t)np, 0, i); 2336 saddr = (struct sockaddr *)(np + 1); 2337 error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen); 2338 if (error) 2339 goto out; 2340 if (saddr->sa_len > argp->ex_addrlen) 2341 saddr->sa_len = argp->ex_addrlen; 2342 if (sacheck(saddr) == -1) 2343 return EINVAL; 2344 if (argp->ex_masklen) { 2345 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 2346 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 2347 if (error) 2348 goto out; 2349 if (smask->sa_len > argp->ex_masklen) 2350 smask->sa_len = argp->ex_masklen; 2351 if (smask->sa_family != saddr->sa_family) 2352 return EINVAL; 2353 if (sacheck(smask) == -1) 2354 return EINVAL; 2355 } 2356 i = saddr->sa_family; 2357 if ((rnh = nep->ne_rtable[i]) == 0) { 2358 /* 2359 * Seems silly to initialize every AF when most are not 2360 * used, do so on demand here 2361 */ 2362 for (dom = domains; dom; dom = dom->dom_next) 2363 if (dom->dom_family == i && dom->dom_rtattach) { 2364 dom->dom_rtattach((void **)&nep->ne_rtable[i], 2365 dom->dom_rtoffset); 2366 break; 2367 } 2368 if ((rnh = nep->ne_rtable[i]) == 0) { 2369 error = ENOBUFS; 2370 goto out; 2371 } 2372 } 2373 2374 enp = (struct netcred *)(*rnh->rnh_addaddr)(saddr, smask, rnh, 2375 np->netc_rnodes); 2376 if (enp != np) { 2377 if (enp == NULL) { 2378 enp = (struct netcred *)(*rnh->rnh_lookup)(saddr, 2379 smask, rnh); 2380 if (enp == NULL) { 2381 error = EPERM; 2382 goto out; 2383 } 2384 } else 2385 enp->netc_refcnt++; 2386 2387 goto check; 2388 } else 2389 enp->netc_refcnt = 1; 2390 2391 np->netc_exflags = argp->ex_flags; 2392 crcvt(&np->netc_anon, &argp->ex_anon); 2393 np->netc_anon.cr_ref = 1; 2394 return 0; 2395 check: 2396 if (enp->netc_exflags != argp->ex_flags || 2397 crcmp(&enp->netc_anon, &argp->ex_anon) != 0) 2398 error = EPERM; 2399 else 2400 error = 0; 2401 out: 2402 free(np, M_NETADDR); 2403 return error; 2404 } 2405 2406 /* ARGSUSED */ 2407 static int 2408 vfs_free_netcred(rn, w) 2409 struct radix_node *rn; 2410 void *w; 2411 { 2412 struct radix_node_head *rnh = (struct radix_node_head *)w; 2413 struct netcred *np = (struct netcred *)(void *)rn; 2414 2415 (*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh); 2416 if (--(np->netc_refcnt) <= 0) 2417 free(np, M_NETADDR); 2418 return (0); 2419 } 2420 2421 /* 2422 * Free the net address hash lists that are hanging off the mount points. 2423 */ 2424 static void 2425 vfs_free_addrlist(nep) 2426 struct netexport *nep; 2427 { 2428 int i; 2429 struct radix_node_head *rnh; 2430 2431 for (i = 0; i <= AF_MAX; i++) 2432 if ((rnh = nep->ne_rtable[i]) != NULL) { 2433 (*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh); 2434 free((caddr_t)rnh, M_RTABLE); 2435 nep->ne_rtable[i] = 0; 2436 } 2437 } 2438 2439 int 2440 vfs_export(mp, nep, argp) 2441 struct mount *mp; 2442 struct netexport *nep; 2443 struct export_args *argp; 2444 { 2445 int error; 2446 2447 if (argp->ex_flags & MNT_DELEXPORT) { 2448 if (mp->mnt_flag & MNT_EXPUBLIC) { 2449 vfs_setpublicfs(NULL, NULL, NULL); 2450 mp->mnt_flag &= ~MNT_EXPUBLIC; 2451 } 2452 vfs_free_addrlist(nep); 2453 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2454 } 2455 if (argp->ex_flags & MNT_EXPORTED) { 2456 if (argp->ex_flags & MNT_EXPUBLIC) { 2457 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2458 return (error); 2459 mp->mnt_flag |= MNT_EXPUBLIC; 2460 } 2461 if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0) 2462 return (error); 2463 mp->mnt_flag |= MNT_EXPORTED; 2464 } 2465 return (0); 2466 } 2467 2468 /* 2469 * Set the publicly exported filesystem (WebNFS). Currently, only 2470 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2471 */ 2472 int 2473 vfs_setpublicfs(mp, nep, argp) 2474 struct mount *mp; 2475 struct netexport *nep; 2476 struct export_args *argp; 2477 { 2478 int error; 2479 struct vnode *rvp; 2480 char *cp; 2481 2482 /* 2483 * mp == NULL -> invalidate the current info, the FS is 2484 * no longer exported. May be called from either vfs_export 2485 * or unmount, so check if it hasn't already been done. 2486 */ 2487 if (mp == NULL) { 2488 if (nfs_pub.np_valid) { 2489 nfs_pub.np_valid = 0; 2490 if (nfs_pub.np_index != NULL) { 2491 FREE(nfs_pub.np_index, M_TEMP); 2492 nfs_pub.np_index = NULL; 2493 } 2494 } 2495 return (0); 2496 } 2497 2498 /* 2499 * Only one allowed at a time. 2500 */ 2501 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2502 return (EBUSY); 2503 2504 /* 2505 * Get real filehandle for root of exported FS. 2506 */ 2507 memset((caddr_t)&nfs_pub.np_handle, 0, sizeof(nfs_pub.np_handle)); 2508 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsidx; 2509 2510 if ((error = VFS_ROOT(mp, &rvp))) 2511 return (error); 2512 2513 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2514 return (error); 2515 2516 vput(rvp); 2517 2518 /* 2519 * If an indexfile was specified, pull it in. 2520 */ 2521 if (argp->ex_indexfile != NULL) { 2522 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2523 M_WAITOK); 2524 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2525 MAXNAMLEN, (size_t *)0); 2526 if (!error) { 2527 /* 2528 * Check for illegal filenames. 2529 */ 2530 for (cp = nfs_pub.np_index; *cp; cp++) { 2531 if (*cp == '/') { 2532 error = EINVAL; 2533 break; 2534 } 2535 } 2536 } 2537 if (error) { 2538 FREE(nfs_pub.np_index, M_TEMP); 2539 return (error); 2540 } 2541 } 2542 2543 nfs_pub.np_mount = mp; 2544 nfs_pub.np_valid = 1; 2545 return (0); 2546 } 2547 2548 struct netcred * 2549 vfs_export_lookup(mp, nep, nam) 2550 struct mount *mp; 2551 struct netexport *nep; 2552 struct mbuf *nam; 2553 { 2554 struct netcred *np; 2555 struct radix_node_head *rnh; 2556 struct sockaddr *saddr; 2557 2558 np = NULL; 2559 if (mp->mnt_flag & MNT_EXPORTED) { 2560 /* 2561 * Lookup in the export list first. 2562 */ 2563 if (nam != NULL) { 2564 saddr = mtod(nam, struct sockaddr *); 2565 rnh = nep->ne_rtable[saddr->sa_family]; 2566 if (rnh != NULL) { 2567 np = (struct netcred *) 2568 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2569 rnh); 2570 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2571 np = NULL; 2572 } 2573 } 2574 /* 2575 * If no address match, use the default if it exists. 2576 */ 2577 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2578 np = &nep->ne_defexported; 2579 } 2580 return (np); 2581 } 2582 2583 /* 2584 * Do the usual access checking. 2585 * file_mode, uid and gid are from the vnode in question, 2586 * while acc_mode and cred are from the VOP_ACCESS parameter list 2587 */ 2588 int 2589 vaccess(type, file_mode, uid, gid, acc_mode, cred) 2590 enum vtype type; 2591 mode_t file_mode; 2592 uid_t uid; 2593 gid_t gid; 2594 mode_t acc_mode; 2595 struct ucred *cred; 2596 { 2597 mode_t mask; 2598 2599 /* 2600 * Super-user always gets read/write access, but execute access depends 2601 * on at least one execute bit being set. 2602 */ 2603 if (cred->cr_uid == 0) { 2604 if ((acc_mode & VEXEC) && type != VDIR && 2605 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 2606 return (EACCES); 2607 return (0); 2608 } 2609 2610 mask = 0; 2611 2612 /* Otherwise, check the owner. */ 2613 if (cred->cr_uid == uid) { 2614 if (acc_mode & VEXEC) 2615 mask |= S_IXUSR; 2616 if (acc_mode & VREAD) 2617 mask |= S_IRUSR; 2618 if (acc_mode & VWRITE) 2619 mask |= S_IWUSR; 2620 return ((file_mode & mask) == mask ? 0 : EACCES); 2621 } 2622 2623 /* Otherwise, check the groups. */ 2624 if (cred->cr_gid == gid || groupmember(gid, cred)) { 2625 if (acc_mode & VEXEC) 2626 mask |= S_IXGRP; 2627 if (acc_mode & VREAD) 2628 mask |= S_IRGRP; 2629 if (acc_mode & VWRITE) 2630 mask |= S_IWGRP; 2631 return ((file_mode & mask) == mask ? 0 : EACCES); 2632 } 2633 2634 /* Otherwise, check everyone else. */ 2635 if (acc_mode & VEXEC) 2636 mask |= S_IXOTH; 2637 if (acc_mode & VREAD) 2638 mask |= S_IROTH; 2639 if (acc_mode & VWRITE) 2640 mask |= S_IWOTH; 2641 return ((file_mode & mask) == mask ? 0 : EACCES); 2642 } 2643 2644 /* 2645 * Unmount all file systems. 2646 * We traverse the list in reverse order under the assumption that doing so 2647 * will avoid needing to worry about dependencies. 2648 */ 2649 void 2650 vfs_unmountall(p) 2651 struct proc *p; 2652 { 2653 struct mount *mp, *nmp; 2654 int allerror, error; 2655 2656 for (allerror = 0, 2657 mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 2658 nmp = mp->mnt_list.cqe_prev; 2659 #ifdef DEBUG 2660 printf("unmounting %s (%s)...\n", 2661 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname); 2662 #endif 2663 /* 2664 * XXX Freeze syncer. Must do this before locking the 2665 * mount point. See dounmount() for details. 2666 */ 2667 lockmgr(&syncer_lock, LK_EXCLUSIVE, NULL); 2668 if (vfs_busy(mp, 0, 0)) { 2669 lockmgr(&syncer_lock, LK_RELEASE, NULL); 2670 continue; 2671 } 2672 if ((error = dounmount(mp, MNT_FORCE, p)) != 0) { 2673 printf("unmount of %s failed with error %d\n", 2674 mp->mnt_stat.f_mntonname, error); 2675 allerror = 1; 2676 } 2677 } 2678 if (allerror) 2679 printf("WARNING: some file systems would not unmount\n"); 2680 } 2681 2682 extern struct simplelock bqueue_slock; /* XXX */ 2683 2684 /* 2685 * Sync and unmount file systems before shutting down. 2686 */ 2687 void 2688 vfs_shutdown() 2689 { 2690 struct lwp *l = curlwp; 2691 struct proc *p; 2692 2693 /* XXX we're certainly not running in proc0's context! */ 2694 if (l == NULL || (p = l->l_proc) == NULL) 2695 p = &proc0; 2696 2697 printf("syncing disks... "); 2698 2699 /* remove user process from run queue */ 2700 suspendsched(); 2701 (void) spl0(); 2702 2703 /* avoid coming back this way again if we panic. */ 2704 doing_shutdown = 1; 2705 2706 sys_sync(l, NULL, NULL); 2707 2708 /* Wait for sync to finish. */ 2709 if (buf_syncwait() != 0) { 2710 #if defined(DDB) && defined(DEBUG_HALT_BUSY) 2711 Debugger(); 2712 #endif 2713 printf("giving up\n"); 2714 return; 2715 } else 2716 printf("done\n"); 2717 2718 /* 2719 * If we've panic'd, don't make the situation potentially 2720 * worse by unmounting the file systems. 2721 */ 2722 if (panicstr != NULL) 2723 return; 2724 2725 /* Release inodes held by texts before update. */ 2726 #ifdef notdef 2727 vnshutdown(); 2728 #endif 2729 /* Unmount file systems. */ 2730 vfs_unmountall(p); 2731 } 2732 2733 /* 2734 * Mount the root file system. If the operator didn't specify a 2735 * file system to use, try all possible file systems until one 2736 * succeeds. 2737 */ 2738 int 2739 vfs_mountroot() 2740 { 2741 struct vfsops *v; 2742 2743 if (root_device == NULL) 2744 panic("vfs_mountroot: root device unknown"); 2745 2746 switch (root_device->dv_class) { 2747 case DV_IFNET: 2748 if (rootdev != NODEV) 2749 panic("vfs_mountroot: rootdev set for DV_IFNET " 2750 "(0x%08x -> %d,%d)", rootdev, 2751 major(rootdev), minor(rootdev)); 2752 break; 2753 2754 case DV_DISK: 2755 if (rootdev == NODEV) 2756 panic("vfs_mountroot: rootdev not set for DV_DISK"); 2757 break; 2758 2759 default: 2760 printf("%s: inappropriate for root file system\n", 2761 root_device->dv_xname); 2762 return (ENODEV); 2763 } 2764 2765 /* 2766 * If user specified a file system, use it. 2767 */ 2768 if (mountroot != NULL) 2769 return ((*mountroot)()); 2770 2771 /* 2772 * Try each file system currently configured into the kernel. 2773 */ 2774 LIST_FOREACH(v, &vfs_list, vfs_list) { 2775 if (v->vfs_mountroot == NULL) 2776 continue; 2777 #ifdef DEBUG 2778 aprint_normal("mountroot: trying %s...\n", v->vfs_name); 2779 #endif 2780 if ((*v->vfs_mountroot)() == 0) { 2781 aprint_normal("root file system type: %s\n", 2782 v->vfs_name); 2783 break; 2784 } 2785 } 2786 2787 if (v == NULL) { 2788 printf("no file system for %s", root_device->dv_xname); 2789 if (root_device->dv_class == DV_DISK) 2790 printf(" (dev 0x%x)", rootdev); 2791 printf("\n"); 2792 return (EFTYPE); 2793 } 2794 return (0); 2795 } 2796 2797 /* 2798 * Given a file system name, look up the vfsops for that 2799 * file system, or return NULL if file system isn't present 2800 * in the kernel. 2801 */ 2802 struct vfsops * 2803 vfs_getopsbyname(name) 2804 const char *name; 2805 { 2806 struct vfsops *v; 2807 2808 LIST_FOREACH(v, &vfs_list, vfs_list) { 2809 if (strcmp(v->vfs_name, name) == 0) 2810 break; 2811 } 2812 2813 return (v); 2814 } 2815 2816 /* 2817 * Establish a file system and initialize it. 2818 */ 2819 int 2820 vfs_attach(vfs) 2821 struct vfsops *vfs; 2822 { 2823 struct vfsops *v; 2824 int error = 0; 2825 2826 2827 /* 2828 * Make sure this file system doesn't already exist. 2829 */ 2830 LIST_FOREACH(v, &vfs_list, vfs_list) { 2831 if (strcmp(vfs->vfs_name, v->vfs_name) == 0) { 2832 error = EEXIST; 2833 goto out; 2834 } 2835 } 2836 2837 /* 2838 * Initialize the vnode operations for this file system. 2839 */ 2840 vfs_opv_init(vfs->vfs_opv_descs); 2841 2842 /* 2843 * Now initialize the file system itself. 2844 */ 2845 (*vfs->vfs_init)(); 2846 2847 /* 2848 * ...and link it into the kernel's list. 2849 */ 2850 LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list); 2851 2852 /* 2853 * Sanity: make sure the reference count is 0. 2854 */ 2855 vfs->vfs_refcount = 0; 2856 2857 out: 2858 return (error); 2859 } 2860 2861 /* 2862 * Remove a file system from the kernel. 2863 */ 2864 int 2865 vfs_detach(vfs) 2866 struct vfsops *vfs; 2867 { 2868 struct vfsops *v; 2869 2870 /* 2871 * Make sure no one is using the filesystem. 2872 */ 2873 if (vfs->vfs_refcount != 0) 2874 return (EBUSY); 2875 2876 /* 2877 * ...and remove it from the kernel's list. 2878 */ 2879 LIST_FOREACH(v, &vfs_list, vfs_list) { 2880 if (v == vfs) { 2881 LIST_REMOVE(v, vfs_list); 2882 break; 2883 } 2884 } 2885 2886 if (v == NULL) 2887 return (ESRCH); 2888 2889 /* 2890 * Now run the file system-specific cleanups. 2891 */ 2892 (*vfs->vfs_done)(); 2893 2894 /* 2895 * Free the vnode operations vector. 2896 */ 2897 vfs_opv_free(vfs->vfs_opv_descs); 2898 return (0); 2899 } 2900 2901 void 2902 vfs_reinit(void) 2903 { 2904 struct vfsops *vfs; 2905 2906 LIST_FOREACH(vfs, &vfs_list, vfs_list) { 2907 if (vfs->vfs_reinit) { 2908 (*vfs->vfs_reinit)(); 2909 } 2910 } 2911 } 2912 2913 /* 2914 * Request a filesystem to suspend write operations. 2915 */ 2916 int 2917 vfs_write_suspend(struct mount *mp, int slpflag, int slptimeo) 2918 { 2919 struct proc *p = curproc; /* XXX */ 2920 int error; 2921 2922 while ((mp->mnt_iflag & IMNT_SUSPEND)) { 2923 if (slptimeo < 0) 2924 return EWOULDBLOCK; 2925 error = tsleep(&mp->mnt_flag, slpflag, "suspwt1", slptimeo); 2926 if (error) 2927 return error; 2928 } 2929 mp->mnt_iflag |= IMNT_SUSPEND; 2930 2931 simple_lock(&mp->mnt_slock); 2932 if (mp->mnt_writeopcountupper > 0) 2933 ltsleep(&mp->mnt_writeopcountupper, PUSER - 1, "suspwt", 2934 0, &mp->mnt_slock); 2935 simple_unlock(&mp->mnt_slock); 2936 2937 error = VFS_SYNC(mp, MNT_WAIT, p->p_ucred, p); 2938 if (error) { 2939 vfs_write_resume(mp); 2940 return error; 2941 } 2942 mp->mnt_iflag |= IMNT_SUSPENDLOW; 2943 2944 simple_lock(&mp->mnt_slock); 2945 if (mp->mnt_writeopcountlower > 0) 2946 ltsleep(&mp->mnt_writeopcountlower, PUSER - 1, "suspwt", 2947 0, &mp->mnt_slock); 2948 mp->mnt_iflag |= IMNT_SUSPENDED; 2949 simple_unlock(&mp->mnt_slock); 2950 2951 return 0; 2952 } 2953 2954 /* 2955 * Request a filesystem to resume write operations. 2956 */ 2957 void 2958 vfs_write_resume(struct mount *mp) 2959 { 2960 2961 if ((mp->mnt_iflag & IMNT_SUSPEND) == 0) 2962 return; 2963 mp->mnt_iflag &= ~(IMNT_SUSPEND | IMNT_SUSPENDLOW | IMNT_SUSPENDED); 2964 wakeup(&mp->mnt_flag); 2965 } 2966 2967 void 2968 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp) 2969 { 2970 const struct statvfs *mbp; 2971 2972 if (sbp == (mbp = &mp->mnt_stat)) 2973 return; 2974 2975 (void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx)); 2976 sbp->f_fsid = mbp->f_fsid; 2977 sbp->f_owner = mbp->f_owner; 2978 sbp->f_flag = mbp->f_flag; 2979 sbp->f_syncwrites = mbp->f_syncwrites; 2980 sbp->f_asyncwrites = mbp->f_asyncwrites; 2981 sbp->f_syncreads = mbp->f_syncreads; 2982 sbp->f_asyncreads = mbp->f_asyncreads; 2983 (void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare)); 2984 (void)memcpy(sbp->f_fstypename, mbp->f_fstypename, 2985 sizeof(sbp->f_fstypename)); 2986 (void)memcpy(sbp->f_mntonname, mbp->f_mntonname, 2987 sizeof(sbp->f_mntonname)); 2988 (void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, 2989 sizeof(sbp->f_mntfromname)); 2990 } 2991 2992 int 2993 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom, 2994 struct mount *mp, struct proc *p) 2995 { 2996 int error; 2997 size_t size; 2998 struct statvfs *sfs = &mp->mnt_stat; 2999 int (*fun)(const void *, void *, size_t, size_t *); 3000 3001 (void)strncpy(mp->mnt_stat.f_fstypename, mp->mnt_op->vfs_name, 3002 sizeof(mp->mnt_stat.f_fstypename)); 3003 3004 if (onp) { 3005 struct cwdinfo *cwdi = p->p_cwdi; 3006 fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr; 3007 if (cwdi->cwdi_rdir != NULL) { 3008 size_t len; 3009 char *bp; 3010 char *path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 3011 3012 if (!path) /* XXX can't happen with M_WAITOK */ 3013 return ENOMEM; 3014 3015 bp = path + MAXPATHLEN; 3016 *--bp = '\0'; 3017 error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp, 3018 path, MAXPATHLEN / 2, 0, p); 3019 if (error) { 3020 free(path, M_TEMP); 3021 return error; 3022 } 3023 3024 len = strlen(bp); 3025 if (len > sizeof(sfs->f_mntonname) - 1) 3026 len = sizeof(sfs->f_mntonname) - 1; 3027 (void)strncpy(sfs->f_mntonname, bp, len); 3028 free(path, M_TEMP); 3029 3030 if (len < sizeof(sfs->f_mntonname) - 1) { 3031 error = (*fun)(onp, &sfs->f_mntonname[len], 3032 sizeof(sfs->f_mntonname) - len - 1, &size); 3033 if (error) 3034 return error; 3035 size += len; 3036 } else { 3037 size = len; 3038 } 3039 } else { 3040 error = (*fun)(onp, &sfs->f_mntonname, 3041 sizeof(sfs->f_mntonname) - 1, &size); 3042 if (error) 3043 return error; 3044 } 3045 (void)memset(sfs->f_mntonname + size, 0, 3046 sizeof(sfs->f_mntonname) - size); 3047 } 3048 3049 if (fromp) { 3050 fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr; 3051 error = (*fun)(fromp, sfs->f_mntfromname, 3052 sizeof(sfs->f_mntfromname) - 1, &size); 3053 if (error) 3054 return error; 3055 (void)memset(sfs->f_mntfromname + size, 0, 3056 sizeof(sfs->f_mntfromname) - size); 3057 } 3058 return 0; 3059 } 3060 3061 #ifdef DDB 3062 const char buf_flagbits[] = 3063 "\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI" 3064 "\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE" 3065 "\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED" 3066 "\32XXX\33VFLUSH"; 3067 3068 void 3069 vfs_buf_print(bp, full, pr) 3070 struct buf *bp; 3071 int full; 3072 void (*pr)(const char *, ...); 3073 { 3074 char buf[1024]; 3075 3076 (*pr)(" vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" dev 0x%x\n", 3077 bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_dev); 3078 3079 bitmask_snprintf(bp->b_flags, buf_flagbits, buf, sizeof(buf)); 3080 (*pr)(" error %d flags 0x%s\n", bp->b_error, buf); 3081 3082 (*pr)(" bufsize 0x%lx bcount 0x%lx resid 0x%lx\n", 3083 bp->b_bufsize, bp->b_bcount, bp->b_resid); 3084 (*pr)(" data %p saveaddr %p dep %p\n", 3085 bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep)); 3086 (*pr)(" iodone %p\n", bp->b_iodone); 3087 } 3088 3089 3090 const char vnode_flagbits[] = 3091 "\20\1ROOT\2TEXT\3SYSTEM\4ISTTY\5EXECMAP" 3092 "\11XLOCK\12XWANT\13BWAIT\14ALIASED" 3093 "\15DIROP\16LAYER\17ONWORKLIST\20DIRTY"; 3094 3095 const char * const vnode_tags[] = { 3096 "VT_NON", 3097 "VT_UFS", 3098 "VT_NFS", 3099 "VT_MFS", 3100 "VT_MSDOSFS", 3101 "VT_LFS", 3102 "VT_LOFS", 3103 "VT_FDESC", 3104 "VT_PORTAL", 3105 "VT_NULL", 3106 "VT_UMAP", 3107 "VT_KERNFS", 3108 "VT_PROCFS", 3109 "VT_AFS", 3110 "VT_ISOFS", 3111 "VT_UNION", 3112 "VT_ADOSFS", 3113 "VT_EXT2FS", 3114 "VT_CODA", 3115 "VT_FILECORE", 3116 "VT_NTFS", 3117 "VT_VFS", 3118 "VT_OVERLAY", 3119 "VT_SMBFS" 3120 }; 3121 3122 void 3123 vfs_vnode_print(vp, full, pr) 3124 struct vnode *vp; 3125 int full; 3126 void (*pr)(const char *, ...); 3127 { 3128 char buf[256]; 3129 const char *vtype, *vtag; 3130 3131 uvm_object_printit(&vp->v_uobj, full, pr); 3132 bitmask_snprintf(vp->v_flag, vnode_flagbits, buf, sizeof(buf)); 3133 (*pr)("\nVNODE flags %s\n", buf); 3134 (*pr)("mp %p numoutput %d size 0x%llx\n", 3135 vp->v_mount, vp->v_numoutput, vp->v_size); 3136 3137 (*pr)("data %p usecount %d writecount %ld holdcnt %ld numoutput %d\n", 3138 vp->v_data, vp->v_usecount, vp->v_writecount, 3139 vp->v_holdcnt, vp->v_numoutput); 3140 3141 vtype = (vp->v_type >= 0 && 3142 vp->v_type < sizeof(vnode_types) / sizeof(vnode_types[0])) ? 3143 vnode_types[vp->v_type] : "UNKNOWN"; 3144 vtag = (vp->v_tag >= 0 && 3145 vp->v_tag < sizeof(vnode_tags) / sizeof(vnode_tags[0])) ? 3146 vnode_tags[vp->v_tag] : "UNKNOWN"; 3147 3148 (*pr)("type %s(%d) tag %s(%d) mount %p typedata %p\n", 3149 vtype, vp->v_type, vtag, vp->v_tag, 3150 vp->v_mount, vp->v_mountedhere); 3151 3152 if (full) { 3153 struct buf *bp; 3154 3155 (*pr)("clean bufs:\n"); 3156 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) { 3157 (*pr)(" bp %p\n", bp); 3158 vfs_buf_print(bp, full, pr); 3159 } 3160 3161 (*pr)("dirty bufs:\n"); 3162 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 3163 (*pr)(" bp %p\n", bp); 3164 vfs_buf_print(bp, full, pr); 3165 } 3166 } 3167 } 3168 3169 void 3170 vfs_mount_print(mp, full, pr) 3171 struct mount *mp; 3172 int full; 3173 void (*pr)(const char *, ...); 3174 { 3175 char sbuf[256]; 3176 3177 (*pr)("vnodecovered = %p syncer = %p data = %p\n", 3178 mp->mnt_vnodecovered,mp->mnt_syncer,mp->mnt_data); 3179 3180 (*pr)("fs_bshift %d dev_bshift = %d maxsymlinklen = %d\n", 3181 mp->mnt_fs_bshift,mp->mnt_dev_bshift,mp->mnt_maxsymlinklen); 3182 3183 bitmask_snprintf(mp->mnt_flag, __MNT_FLAG_BITS, sbuf, sizeof(sbuf)); 3184 (*pr)("flag = %s\n", sbuf); 3185 3186 bitmask_snprintf(mp->mnt_iflag, __IMNT_FLAG_BITS, sbuf, sizeof(sbuf)); 3187 (*pr)("iflag = %s\n", sbuf); 3188 3189 /* XXX use lockmgr_printinfo */ 3190 if (mp->mnt_lock.lk_sharecount) 3191 (*pr)(" lock type %s: SHARED (count %d)", mp->mnt_lock.lk_wmesg, 3192 mp->mnt_lock.lk_sharecount); 3193 else if (mp->mnt_lock.lk_flags & LK_HAVE_EXCL) { 3194 (*pr)(" lock type %s: EXCL (count %d) by ", 3195 mp->mnt_lock.lk_wmesg, mp->mnt_lock.lk_exclusivecount); 3196 if (mp->mnt_lock.lk_flags & LK_SPIN) 3197 (*pr)("processor %lu", mp->mnt_lock.lk_cpu); 3198 else 3199 (*pr)("pid %d.%d", mp->mnt_lock.lk_lockholder, 3200 mp->mnt_lock.lk_locklwp); 3201 } else 3202 (*pr)(" not locked"); 3203 if ((mp->mnt_lock.lk_flags & LK_SPIN) == 0 && mp->mnt_lock.lk_waitcount > 0) 3204 (*pr)(" with %d pending", mp->mnt_lock.lk_waitcount); 3205 3206 (*pr)("\n"); 3207 3208 if (mp->mnt_unmounter) { 3209 (*pr)("unmounter pid = %d ",mp->mnt_unmounter->p_pid); 3210 } 3211 (*pr)("wcnt = %d, writeopcountupper = %d, writeopcountupper = %d\n", 3212 mp->mnt_wcnt,mp->mnt_writeopcountupper,mp->mnt_writeopcountlower); 3213 3214 (*pr)("statvfs cache:\n"); 3215 (*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize); 3216 (*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize); 3217 (*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize); 3218 3219 (*pr)("\tblocks = "PRIu64"\n",mp->mnt_stat.f_blocks); 3220 (*pr)("\tbfree = "PRIu64"\n",mp->mnt_stat.f_bfree); 3221 (*pr)("\tbavail = "PRIu64"\n",mp->mnt_stat.f_bavail); 3222 (*pr)("\tbresvd = "PRIu64"\n",mp->mnt_stat.f_bresvd); 3223 3224 (*pr)("\tfiles = "PRIu64"\n",mp->mnt_stat.f_files); 3225 (*pr)("\tffree = "PRIu64"\n",mp->mnt_stat.f_ffree); 3226 (*pr)("\tfavail = "PRIu64"\n",mp->mnt_stat.f_favail); 3227 (*pr)("\tfresvd = "PRIu64"\n",mp->mnt_stat.f_fresvd); 3228 3229 (*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n", 3230 mp->mnt_stat.f_fsidx.__fsid_val[0], 3231 mp->mnt_stat.f_fsidx.__fsid_val[1]); 3232 3233 (*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner); 3234 (*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax); 3235 3236 bitmask_snprintf(mp->mnt_stat.f_flag, __MNT_FLAG_BITS, sbuf, 3237 sizeof(sbuf)); 3238 (*pr)("\tflag = %s\n",sbuf); 3239 (*pr)("\tsyncwrites = " PRIu64 "\n",mp->mnt_stat.f_syncwrites); 3240 (*pr)("\tasyncwrites = " PRIu64 "\n",mp->mnt_stat.f_asyncwrites); 3241 (*pr)("\tsyncreads = " PRIu64 "\n",mp->mnt_stat.f_syncreads); 3242 (*pr)("\tasyncreads = " PRIu64 "\n",mp->mnt_stat.f_asyncreads); 3243 (*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename); 3244 (*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname); 3245 (*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname); 3246 3247 { 3248 int cnt = 0; 3249 struct vnode *vp; 3250 (*pr)("locked vnodes ="); 3251 /* XXX would take mountlist lock, except ddb may not have context */ 3252 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 3253 if (VOP_ISLOCKED(vp)) { 3254 if ((++cnt % 6) == 0) { 3255 (*pr)(" %p,\n\t", vp); 3256 } else { 3257 (*pr)(" %p,", vp); 3258 } 3259 } 3260 } 3261 (*pr)("\n"); 3262 } 3263 3264 if (full) { 3265 int cnt = 0; 3266 struct vnode *vp; 3267 (*pr)("all vnodes ="); 3268 /* XXX would take mountlist lock, except ddb may not have context */ 3269 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 3270 if (!LIST_NEXT(vp, v_mntvnodes)) { 3271 (*pr)(" %p", vp); 3272 } else if ((++cnt % 6) == 0) { 3273 (*pr)(" %p,\n\t", vp); 3274 } else { 3275 (*pr)(" %p,", vp); 3276 } 3277 } 3278 (*pr)("\n", vp); 3279 } 3280 } 3281 3282 #endif 3283