1 /* $NetBSD: vfs_subr.c,v 1.229 2004/06/19 06:20:02 yamt 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 1.229 2004/06/19 06:20:02 yamt 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 vfsp->vfs_refcount++; 387 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfs_name, MFSNAMELEN); 388 mp->mnt_stat.f_mntonname[0] = '/'; 389 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 390 *mpp = mp; 391 return (0); 392 } 393 394 /* 395 * Lookup a mount point by filesystem identifier. 396 */ 397 struct mount * 398 vfs_getvfs(fsid) 399 fsid_t *fsid; 400 { 401 struct mount *mp; 402 403 simple_lock(&mountlist_slock); 404 CIRCLEQ_FOREACH(mp, &mountlist, mnt_list) { 405 if (mp->mnt_stat.f_fsidx.__fsid_val[0] == fsid->__fsid_val[0] && 406 mp->mnt_stat.f_fsidx.__fsid_val[1] == fsid->__fsid_val[1]) { 407 simple_unlock(&mountlist_slock); 408 return (mp); 409 } 410 } 411 simple_unlock(&mountlist_slock); 412 return ((struct mount *)0); 413 } 414 415 /* 416 * Get a new unique fsid 417 */ 418 void 419 vfs_getnewfsid(mp) 420 struct mount *mp; 421 { 422 static u_short xxxfs_mntid; 423 fsid_t tfsid; 424 int mtype; 425 426 simple_lock(&mntid_slock); 427 mtype = makefstype(mp->mnt_op->vfs_name); 428 mp->mnt_stat.f_fsidx.__fsid_val[0] = makedev(mtype, 0); 429 mp->mnt_stat.f_fsidx.__fsid_val[1] = mtype; 430 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 431 if (xxxfs_mntid == 0) 432 ++xxxfs_mntid; 433 tfsid.__fsid_val[0] = makedev(mtype & 0xff, xxxfs_mntid); 434 tfsid.__fsid_val[1] = mtype; 435 if (!CIRCLEQ_EMPTY(&mountlist)) { 436 while (vfs_getvfs(&tfsid)) { 437 tfsid.__fsid_val[0]++; 438 xxxfs_mntid++; 439 } 440 } 441 mp->mnt_stat.f_fsidx.__fsid_val[0] = tfsid.__fsid_val[0]; 442 mp->mnt_stat.f_fsid = mp->mnt_stat.f_fsidx.__fsid_val[0]; 443 simple_unlock(&mntid_slock); 444 } 445 446 /* 447 * Make a 'unique' number from a mount type name. 448 */ 449 long 450 makefstype(type) 451 const char *type; 452 { 453 long rv; 454 455 for (rv = 0; *type; type++) { 456 rv <<= 2; 457 rv ^= *type; 458 } 459 return rv; 460 } 461 462 463 /* 464 * Set vnode attributes to VNOVAL 465 */ 466 void 467 vattr_null(vap) 468 struct vattr *vap; 469 { 470 471 vap->va_type = VNON; 472 473 /* 474 * Assign individually so that it is safe even if size and 475 * sign of each member are varied. 476 */ 477 vap->va_mode = VNOVAL; 478 vap->va_nlink = VNOVAL; 479 vap->va_uid = VNOVAL; 480 vap->va_gid = VNOVAL; 481 vap->va_fsid = VNOVAL; 482 vap->va_fileid = VNOVAL; 483 vap->va_size = VNOVAL; 484 vap->va_blocksize = VNOVAL; 485 vap->va_atime.tv_sec = 486 vap->va_mtime.tv_sec = 487 vap->va_ctime.tv_sec = 488 vap->va_birthtime.tv_sec = VNOVAL; 489 vap->va_atime.tv_nsec = 490 vap->va_mtime.tv_nsec = 491 vap->va_ctime.tv_nsec = 492 vap->va_birthtime.tv_nsec = VNOVAL; 493 vap->va_gen = VNOVAL; 494 vap->va_flags = VNOVAL; 495 vap->va_rdev = VNOVAL; 496 vap->va_bytes = VNOVAL; 497 vap->va_vaflags = 0; 498 } 499 500 /* 501 * Routines having to do with the management of the vnode table. 502 */ 503 extern int (**dead_vnodeop_p)(void *); 504 long numvnodes; 505 506 /* 507 * Return the next vnode from the free list. 508 */ 509 int 510 getnewvnode(tag, mp, vops, vpp) 511 enum vtagtype tag; 512 struct mount *mp; 513 int (**vops)(void *); 514 struct vnode **vpp; 515 { 516 extern struct uvm_pagerops uvm_vnodeops; 517 struct uvm_object *uobj; 518 struct proc *p = curproc; /* XXX */ 519 static int toggle; 520 struct vnode *vp; 521 int error = 0, tryalloc; 522 523 try_again: 524 if (mp) { 525 /* 526 * Mark filesystem busy while we're creating a vnode. 527 * If unmount is in progress, this will wait; if the 528 * unmount succeeds (only if umount -f), this will 529 * return an error. If the unmount fails, we'll keep 530 * going afterwards. 531 * (This puts the per-mount vnode list logically under 532 * the protection of the vfs_busy lock). 533 */ 534 error = vfs_busy(mp, LK_RECURSEFAIL, 0); 535 if (error && error != EDEADLK) 536 return error; 537 } 538 539 /* 540 * We must choose whether to allocate a new vnode or recycle an 541 * existing one. The criterion for allocating a new one is that 542 * the total number of vnodes is less than the number desired or 543 * there are no vnodes on either free list. Generally we only 544 * want to recycle vnodes that have no buffers associated with 545 * them, so we look first on the vnode_free_list. If it is empty, 546 * we next consider vnodes with referencing buffers on the 547 * vnode_hold_list. The toggle ensures that half the time we 548 * will use a buffer from the vnode_hold_list, and half the time 549 * we will allocate a new one unless the list has grown to twice 550 * the desired size. We are reticent to recycle vnodes from the 551 * vnode_hold_list because we will lose the identity of all its 552 * referencing buffers. 553 */ 554 555 vp = NULL; 556 557 simple_lock(&vnode_free_list_slock); 558 559 toggle ^= 1; 560 if (numvnodes > 2 * desiredvnodes) 561 toggle = 0; 562 563 tryalloc = numvnodes < desiredvnodes || 564 (TAILQ_FIRST(&vnode_free_list) == NULL && 565 (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle)); 566 567 if (tryalloc && 568 (vp = pool_get(&vnode_pool, PR_NOWAIT)) != NULL) { 569 numvnodes++; 570 simple_unlock(&vnode_free_list_slock); 571 memset(vp, 0, sizeof(*vp)); 572 simple_lock_init(&vp->v_interlock); 573 uobj = &vp->v_uobj; 574 uobj->pgops = &uvm_vnodeops; 575 TAILQ_INIT(&uobj->memq); 576 /* 577 * done by memset() above. 578 * uobj->uo_npages = 0; 579 * LIST_INIT(&vp->v_nclist); 580 * LIST_INIT(&vp->v_dnclist); 581 */ 582 } else { 583 vp = getcleanvnode(p); 584 /* 585 * Unless this is a bad time of the month, at most 586 * the first NCPUS items on the free list are 587 * locked, so this is close enough to being empty. 588 */ 589 if (vp == NULLVP) { 590 if (mp && error != EDEADLK) 591 vfs_unbusy(mp); 592 if (tryalloc) { 593 printf("WARNING: unable to allocate new " 594 "vnode, retrying...\n"); 595 (void) tsleep(&lbolt, PRIBIO, "newvn", hz); 596 goto try_again; 597 } 598 tablefull("vnode", "increase kern.maxvnodes or NVNODE"); 599 *vpp = 0; 600 return (ENFILE); 601 } 602 vp->v_flag = 0; 603 vp->v_socket = NULL; 604 #ifdef VERIFIED_EXEC 605 vp->fp_status = FINGERPRINT_INVALID; 606 #endif 607 } 608 vp->v_type = VNON; 609 vp->v_vnlock = &vp->v_lock; 610 lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); 611 KASSERT(LIST_EMPTY(&vp->v_nclist)); 612 KASSERT(LIST_EMPTY(&vp->v_dnclist)); 613 vp->v_tag = tag; 614 vp->v_op = vops; 615 insmntque(vp, mp); 616 *vpp = vp; 617 vp->v_usecount = 1; 618 vp->v_data = 0; 619 simple_lock_init(&vp->v_uobj.vmobjlock); 620 621 /* 622 * initialize uvm_object within vnode. 623 */ 624 625 uobj = &vp->v_uobj; 626 KASSERT(uobj->pgops == &uvm_vnodeops); 627 KASSERT(uobj->uo_npages == 0); 628 KASSERT(TAILQ_FIRST(&uobj->memq) == NULL); 629 vp->v_size = VSIZENOTSET; 630 631 if (mp && error != EDEADLK) 632 vfs_unbusy(mp); 633 return (0); 634 } 635 636 /* 637 * This is really just the reverse of getnewvnode(). Needed for 638 * VFS_VGET functions who may need to push back a vnode in case 639 * of a locking race. 640 */ 641 void 642 ungetnewvnode(vp) 643 struct vnode *vp; 644 { 645 #ifdef DIAGNOSTIC 646 if (vp->v_usecount != 1) 647 panic("ungetnewvnode: busy vnode"); 648 #endif 649 vp->v_usecount--; 650 insmntque(vp, NULL); 651 vp->v_type = VBAD; 652 653 simple_lock(&vp->v_interlock); 654 /* 655 * Insert at head of LRU list 656 */ 657 simple_lock(&vnode_free_list_slock); 658 if (vp->v_holdcnt > 0) 659 TAILQ_INSERT_HEAD(&vnode_hold_list, vp, v_freelist); 660 else 661 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 662 simple_unlock(&vnode_free_list_slock); 663 simple_unlock(&vp->v_interlock); 664 } 665 666 /* 667 * Move a vnode from one mount queue to another. 668 */ 669 void 670 insmntque(vp, mp) 671 struct vnode *vp; 672 struct mount *mp; 673 { 674 675 #ifdef DIAGNOSTIC 676 if ((mp != NULL) && 677 (mp->mnt_iflag & IMNT_UNMOUNT) && 678 !(mp->mnt_flag & MNT_SOFTDEP) && 679 vp->v_tag != VT_VFS) { 680 panic("insmntque into dying filesystem"); 681 } 682 #endif 683 684 simple_lock(&mntvnode_slock); 685 /* 686 * Delete from old mount point vnode list, if on one. 687 */ 688 if (vp->v_mount != NULL) 689 LIST_REMOVE(vp, v_mntvnodes); 690 /* 691 * Insert into list of vnodes for the new mount point, if available. 692 */ 693 if ((vp->v_mount = mp) != NULL) 694 LIST_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); 695 simple_unlock(&mntvnode_slock); 696 } 697 698 /* 699 * Update outstanding I/O count and do wakeup if requested. 700 */ 701 void 702 vwakeup(bp) 703 struct buf *bp; 704 { 705 struct vnode *vp; 706 707 if ((vp = bp->b_vp) != NULL) { 708 /* XXX global lock hack 709 * can't use v_interlock here since this is called 710 * in interrupt context from biodone(). 711 */ 712 simple_lock(&global_v_numoutput_slock); 713 if (--vp->v_numoutput < 0) 714 panic("vwakeup: neg numoutput, vp %p", vp); 715 if ((vp->v_flag & VBWAIT) && vp->v_numoutput <= 0) { 716 vp->v_flag &= ~VBWAIT; 717 wakeup((caddr_t)&vp->v_numoutput); 718 } 719 simple_unlock(&global_v_numoutput_slock); 720 } 721 } 722 723 /* 724 * Flush out and invalidate all buffers associated with a vnode. 725 * Called with the underlying vnode locked, which should prevent new dirty 726 * buffers from being queued. 727 */ 728 int 729 vinvalbuf(vp, flags, cred, p, slpflag, slptimeo) 730 struct vnode *vp; 731 int flags; 732 struct ucred *cred; 733 struct proc *p; 734 int slpflag, slptimeo; 735 { 736 struct buf *bp, *nbp; 737 int s, error; 738 int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO | 739 (flags & V_SAVE ? PGO_CLEANIT : 0); 740 741 /* XXXUBC this doesn't look at flags or slp* */ 742 simple_lock(&vp->v_interlock); 743 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 744 if (error) { 745 return error; 746 } 747 748 if (flags & V_SAVE) { 749 error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0, p); 750 if (error) 751 return (error); 752 #ifdef DIAGNOSTIC 753 s = splbio(); 754 if (vp->v_numoutput > 0 || !LIST_EMPTY(&vp->v_dirtyblkhd)) 755 panic("vinvalbuf: dirty bufs, vp %p", vp); 756 splx(s); 757 #endif 758 } 759 760 s = splbio(); 761 762 restart: 763 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 764 nbp = LIST_NEXT(bp, b_vnbufs); 765 simple_lock(&bp->b_interlock); 766 if (bp->b_flags & B_BUSY) { 767 bp->b_flags |= B_WANTED; 768 error = ltsleep((caddr_t)bp, 769 slpflag | (PRIBIO + 1) | PNORELOCK, 770 "vinvalbuf", slptimeo, &bp->b_interlock); 771 if (error) { 772 splx(s); 773 return (error); 774 } 775 goto restart; 776 } 777 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 778 simple_unlock(&bp->b_interlock); 779 brelse(bp); 780 } 781 782 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 783 nbp = LIST_NEXT(bp, b_vnbufs); 784 simple_lock(&bp->b_interlock); 785 if (bp->b_flags & B_BUSY) { 786 bp->b_flags |= B_WANTED; 787 error = ltsleep((caddr_t)bp, 788 slpflag | (PRIBIO + 1) | PNORELOCK, 789 "vinvalbuf", slptimeo, &bp->b_interlock); 790 if (error) { 791 splx(s); 792 return (error); 793 } 794 goto restart; 795 } 796 /* 797 * XXX Since there are no node locks for NFS, I believe 798 * there is a slight chance that a delayed write will 799 * occur while sleeping just above, so check for it. 800 */ 801 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 802 #ifdef DEBUG 803 printf("buffer still DELWRI\n"); 804 #endif 805 bp->b_flags |= B_BUSY | B_VFLUSH; 806 simple_unlock(&bp->b_interlock); 807 VOP_BWRITE(bp); 808 goto restart; 809 } 810 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 811 simple_unlock(&bp->b_interlock); 812 brelse(bp); 813 } 814 815 #ifdef DIAGNOSTIC 816 if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd)) 817 panic("vinvalbuf: flush failed, vp %p", vp); 818 #endif 819 820 splx(s); 821 822 return (0); 823 } 824 825 /* 826 * Destroy any in core blocks past the truncation length. 827 * Called with the underlying vnode locked, which should prevent new dirty 828 * buffers from being queued. 829 */ 830 int 831 vtruncbuf(vp, lbn, slpflag, slptimeo) 832 struct vnode *vp; 833 daddr_t lbn; 834 int slpflag, slptimeo; 835 { 836 struct buf *bp, *nbp; 837 int s, error; 838 voff_t off; 839 840 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 841 simple_lock(&vp->v_interlock); 842 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 843 if (error) { 844 return error; 845 } 846 847 s = splbio(); 848 849 restart: 850 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 851 nbp = LIST_NEXT(bp, b_vnbufs); 852 if (bp->b_lblkno < lbn) 853 continue; 854 simple_lock(&bp->b_interlock); 855 if (bp->b_flags & B_BUSY) { 856 bp->b_flags |= B_WANTED; 857 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 858 "vtruncbuf", slptimeo, &bp->b_interlock); 859 if (error) { 860 splx(s); 861 return (error); 862 } 863 goto restart; 864 } 865 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 866 simple_unlock(&bp->b_interlock); 867 brelse(bp); 868 } 869 870 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 871 nbp = LIST_NEXT(bp, b_vnbufs); 872 if (bp->b_lblkno < lbn) 873 continue; 874 simple_lock(&bp->b_interlock); 875 if (bp->b_flags & B_BUSY) { 876 bp->b_flags |= B_WANTED; 877 error = ltsleep(bp, slpflag | (PRIBIO + 1) | PNORELOCK, 878 "vtruncbuf", slptimeo, &bp->b_interlock); 879 if (error) { 880 splx(s); 881 return (error); 882 } 883 goto restart; 884 } 885 bp->b_flags |= B_BUSY | B_INVAL | B_VFLUSH; 886 simple_unlock(&bp->b_interlock); 887 brelse(bp); 888 } 889 890 splx(s); 891 892 return (0); 893 } 894 895 void 896 vflushbuf(vp, sync) 897 struct vnode *vp; 898 int sync; 899 { 900 struct buf *bp, *nbp; 901 int flags = PGO_CLEANIT | PGO_ALLPAGES | (sync ? PGO_SYNCIO : 0); 902 int s; 903 904 simple_lock(&vp->v_interlock); 905 (void) VOP_PUTPAGES(vp, 0, 0, flags); 906 907 loop: 908 s = splbio(); 909 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 910 nbp = LIST_NEXT(bp, b_vnbufs); 911 simple_lock(&bp->b_interlock); 912 if ((bp->b_flags & B_BUSY)) { 913 simple_unlock(&bp->b_interlock); 914 continue; 915 } 916 if ((bp->b_flags & B_DELWRI) == 0) 917 panic("vflushbuf: not dirty, bp %p", bp); 918 bp->b_flags |= B_BUSY | B_VFLUSH; 919 simple_unlock(&bp->b_interlock); 920 splx(s); 921 /* 922 * Wait for I/O associated with indirect blocks to complete, 923 * since there is no way to quickly wait for them below. 924 */ 925 if (bp->b_vp == vp || sync == 0) 926 (void) bawrite(bp); 927 else 928 (void) bwrite(bp); 929 goto loop; 930 } 931 if (sync == 0) { 932 splx(s); 933 return; 934 } 935 simple_lock(&global_v_numoutput_slock); 936 while (vp->v_numoutput) { 937 vp->v_flag |= VBWAIT; 938 ltsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "vflushbuf", 0, 939 &global_v_numoutput_slock); 940 } 941 simple_unlock(&global_v_numoutput_slock); 942 splx(s); 943 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) { 944 vprint("vflushbuf: dirty", vp); 945 goto loop; 946 } 947 } 948 949 /* 950 * Associate a buffer with a vnode. 951 */ 952 void 953 bgetvp(vp, bp) 954 struct vnode *vp; 955 struct buf *bp; 956 { 957 int s; 958 959 if (bp->b_vp) 960 panic("bgetvp: not free, bp %p", bp); 961 VHOLD(vp); 962 s = splbio(); 963 bp->b_vp = vp; 964 if (vp->v_type == VBLK || vp->v_type == VCHR) 965 bp->b_dev = vp->v_rdev; 966 else 967 bp->b_dev = NODEV; 968 /* 969 * Insert onto list for new vnode. 970 */ 971 bufinsvn(bp, &vp->v_cleanblkhd); 972 splx(s); 973 } 974 975 /* 976 * Disassociate a buffer from a vnode. 977 */ 978 void 979 brelvp(bp) 980 struct buf *bp; 981 { 982 struct vnode *vp; 983 int s; 984 985 if (bp->b_vp == NULL) 986 panic("brelvp: vp NULL, bp %p", bp); 987 988 s = splbio(); 989 vp = bp->b_vp; 990 /* 991 * Delete from old vnode list, if on one. 992 */ 993 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 994 bufremvn(bp); 995 996 if (TAILQ_EMPTY(&vp->v_uobj.memq) && (vp->v_flag & VONWORKLST) && 997 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 998 vp->v_flag &= ~VONWORKLST; 999 LIST_REMOVE(vp, v_synclist); 1000 } 1001 1002 bp->b_vp = NULL; 1003 HOLDRELE(vp); 1004 splx(s); 1005 } 1006 1007 /* 1008 * Reassign a buffer from one vnode to another. 1009 * Used to assign file specific control information 1010 * (indirect blocks) to the vnode to which they belong. 1011 * 1012 * This function must be called at splbio(). 1013 */ 1014 void 1015 reassignbuf(bp, newvp) 1016 struct buf *bp; 1017 struct vnode *newvp; 1018 { 1019 struct buflists *listheadp; 1020 int delay; 1021 1022 /* 1023 * Delete from old vnode list, if on one. 1024 */ 1025 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 1026 bufremvn(bp); 1027 /* 1028 * If dirty, put on list of dirty buffers; 1029 * otherwise insert onto list of clean buffers. 1030 */ 1031 if ((bp->b_flags & B_DELWRI) == 0) { 1032 listheadp = &newvp->v_cleanblkhd; 1033 if (TAILQ_EMPTY(&newvp->v_uobj.memq) && 1034 (newvp->v_flag & VONWORKLST) && 1035 LIST_FIRST(&newvp->v_dirtyblkhd) == NULL) { 1036 newvp->v_flag &= ~VONWORKLST; 1037 LIST_REMOVE(newvp, v_synclist); 1038 } 1039 } else { 1040 listheadp = &newvp->v_dirtyblkhd; 1041 if ((newvp->v_flag & VONWORKLST) == 0) { 1042 switch (newvp->v_type) { 1043 case VDIR: 1044 delay = dirdelay; 1045 break; 1046 case VBLK: 1047 if (newvp->v_specmountpoint != NULL) { 1048 delay = metadelay; 1049 break; 1050 } 1051 /* fall through */ 1052 default: 1053 delay = filedelay; 1054 break; 1055 } 1056 if (!newvp->v_mount || 1057 (newvp->v_mount->mnt_flag & MNT_ASYNC) == 0) 1058 vn_syncer_add_to_worklist(newvp, delay); 1059 } 1060 } 1061 bufinsvn(bp, listheadp); 1062 } 1063 1064 /* 1065 * Create a vnode for a block device. 1066 * Used for root filesystem and swap areas. 1067 * Also used for memory file system special devices. 1068 */ 1069 int 1070 bdevvp(dev, vpp) 1071 dev_t dev; 1072 struct vnode **vpp; 1073 { 1074 1075 return (getdevvp(dev, vpp, VBLK)); 1076 } 1077 1078 /* 1079 * Create a vnode for a character device. 1080 * Used for kernfs and some console handling. 1081 */ 1082 int 1083 cdevvp(dev, vpp) 1084 dev_t dev; 1085 struct vnode **vpp; 1086 { 1087 1088 return (getdevvp(dev, vpp, VCHR)); 1089 } 1090 1091 /* 1092 * Create a vnode for a device. 1093 * Used by bdevvp (block device) for root file system etc., 1094 * and by cdevvp (character device) for console and kernfs. 1095 */ 1096 int 1097 getdevvp(dev, vpp, type) 1098 dev_t dev; 1099 struct vnode **vpp; 1100 enum vtype type; 1101 { 1102 struct vnode *vp; 1103 struct vnode *nvp; 1104 int error; 1105 1106 if (dev == NODEV) { 1107 *vpp = NULLVP; 1108 return (0); 1109 } 1110 error = getnewvnode(VT_NON, NULL, spec_vnodeop_p, &nvp); 1111 if (error) { 1112 *vpp = NULLVP; 1113 return (error); 1114 } 1115 vp = nvp; 1116 vp->v_type = type; 1117 if ((nvp = checkalias(vp, dev, NULL)) != 0) { 1118 vput(vp); 1119 vp = nvp; 1120 } 1121 *vpp = vp; 1122 return (0); 1123 } 1124 1125 /* 1126 * Check to see if the new vnode represents a special device 1127 * for which we already have a vnode (either because of 1128 * bdevvp() or because of a different vnode representing 1129 * the same block device). If such an alias exists, deallocate 1130 * the existing contents and return the aliased vnode. The 1131 * caller is responsible for filling it with its new contents. 1132 */ 1133 struct vnode * 1134 checkalias(nvp, nvp_rdev, mp) 1135 struct vnode *nvp; 1136 dev_t nvp_rdev; 1137 struct mount *mp; 1138 { 1139 struct proc *p = curproc; /* XXX */ 1140 struct vnode *vp; 1141 struct vnode **vpp; 1142 1143 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 1144 return (NULLVP); 1145 1146 vpp = &speclisth[SPECHASH(nvp_rdev)]; 1147 loop: 1148 simple_lock(&spechash_slock); 1149 for (vp = *vpp; vp; vp = vp->v_specnext) { 1150 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) 1151 continue; 1152 /* 1153 * Alias, but not in use, so flush it out. 1154 */ 1155 simple_lock(&vp->v_interlock); 1156 if (vp->v_usecount == 0) { 1157 simple_unlock(&spechash_slock); 1158 vgonel(vp, p); 1159 goto loop; 1160 } 1161 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK | LK_NOWAIT)) { 1162 simple_unlock(&spechash_slock); 1163 goto loop; 1164 } 1165 break; 1166 } 1167 if (vp == NULL || vp->v_tag != VT_NON || vp->v_type != VBLK) { 1168 MALLOC(nvp->v_specinfo, struct specinfo *, 1169 sizeof(struct specinfo), M_VNODE, M_NOWAIT); 1170 /* XXX Erg. */ 1171 if (nvp->v_specinfo == NULL) { 1172 simple_unlock(&spechash_slock); 1173 uvm_wait("checkalias"); 1174 goto loop; 1175 } 1176 1177 nvp->v_rdev = nvp_rdev; 1178 nvp->v_hashchain = vpp; 1179 nvp->v_specnext = *vpp; 1180 nvp->v_specmountpoint = NULL; 1181 simple_unlock(&spechash_slock); 1182 nvp->v_speclockf = NULL; 1183 simple_lock_init(&nvp->v_spec_cow_slock); 1184 SLIST_INIT(&nvp->v_spec_cow_head); 1185 nvp->v_spec_cow_req = 0; 1186 nvp->v_spec_cow_count = 0; 1187 1188 *vpp = nvp; 1189 if (vp != NULLVP) { 1190 nvp->v_flag |= VALIASED; 1191 vp->v_flag |= VALIASED; 1192 vput(vp); 1193 } 1194 return (NULLVP); 1195 } 1196 simple_unlock(&spechash_slock); 1197 VOP_UNLOCK(vp, 0); 1198 simple_lock(&vp->v_interlock); 1199 vclean(vp, 0, p); 1200 vp->v_op = nvp->v_op; 1201 vp->v_tag = nvp->v_tag; 1202 vp->v_vnlock = &vp->v_lock; 1203 lockinit(vp->v_vnlock, PVFS, "vnlock", 0, 0); 1204 nvp->v_type = VNON; 1205 insmntque(vp, mp); 1206 return (vp); 1207 } 1208 1209 /* 1210 * Grab a particular vnode from the free list, increment its 1211 * reference count and lock it. If the vnode lock bit is set the 1212 * vnode is being eliminated in vgone. In that case, we can not 1213 * grab the vnode, so the process is awakened when the transition is 1214 * completed, and an error returned to indicate that the vnode is no 1215 * longer usable (possibly having been changed to a new file system type). 1216 */ 1217 int 1218 vget(vp, flags) 1219 struct vnode *vp; 1220 int flags; 1221 { 1222 int error; 1223 1224 /* 1225 * If the vnode is in the process of being cleaned out for 1226 * another use, we wait for the cleaning to finish and then 1227 * return failure. Cleaning is determined by checking that 1228 * the VXLOCK flag is set. 1229 */ 1230 1231 if ((flags & LK_INTERLOCK) == 0) 1232 simple_lock(&vp->v_interlock); 1233 if (vp->v_flag & VXLOCK) { 1234 if (flags & LK_NOWAIT) { 1235 simple_unlock(&vp->v_interlock); 1236 return EBUSY; 1237 } 1238 vp->v_flag |= VXWANT; 1239 ltsleep(vp, PINOD|PNORELOCK, "vget", 0, &vp->v_interlock); 1240 return (ENOENT); 1241 } 1242 if (vp->v_usecount == 0) { 1243 simple_lock(&vnode_free_list_slock); 1244 if (vp->v_holdcnt > 0) 1245 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 1246 else 1247 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1248 simple_unlock(&vnode_free_list_slock); 1249 } 1250 vp->v_usecount++; 1251 #ifdef DIAGNOSTIC 1252 if (vp->v_usecount == 0) { 1253 vprint("vget", vp); 1254 panic("vget: usecount overflow, vp %p", vp); 1255 } 1256 #endif 1257 if (flags & LK_TYPE_MASK) { 1258 if ((error = vn_lock(vp, flags | LK_INTERLOCK))) { 1259 /* 1260 * must expand vrele here because we do not want 1261 * to call VOP_INACTIVE if the reference count 1262 * drops back to zero since it was never really 1263 * active. We must remove it from the free list 1264 * before sleeping so that multiple processes do 1265 * not try to recycle it. 1266 */ 1267 simple_lock(&vp->v_interlock); 1268 vp->v_usecount--; 1269 if (vp->v_usecount > 0) { 1270 simple_unlock(&vp->v_interlock); 1271 return (error); 1272 } 1273 /* 1274 * insert at tail of LRU list 1275 */ 1276 simple_lock(&vnode_free_list_slock); 1277 if (vp->v_holdcnt > 0) 1278 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, 1279 v_freelist); 1280 else 1281 TAILQ_INSERT_TAIL(&vnode_free_list, vp, 1282 v_freelist); 1283 simple_unlock(&vnode_free_list_slock); 1284 simple_unlock(&vp->v_interlock); 1285 } 1286 return (error); 1287 } 1288 simple_unlock(&vp->v_interlock); 1289 return (0); 1290 } 1291 1292 /* 1293 * vput(), just unlock and vrele() 1294 */ 1295 void 1296 vput(vp) 1297 struct vnode *vp; 1298 { 1299 struct proc *p = curproc; /* XXX */ 1300 1301 #ifdef DIAGNOSTIC 1302 if (vp == NULL) 1303 panic("vput: null vp"); 1304 #endif 1305 simple_lock(&vp->v_interlock); 1306 vp->v_usecount--; 1307 if (vp->v_usecount > 0) { 1308 simple_unlock(&vp->v_interlock); 1309 VOP_UNLOCK(vp, 0); 1310 return; 1311 } 1312 #ifdef DIAGNOSTIC 1313 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1314 vprint("vput: bad ref count", vp); 1315 panic("vput: ref cnt"); 1316 } 1317 #endif 1318 /* 1319 * Insert at tail of LRU list. 1320 */ 1321 simple_lock(&vnode_free_list_slock); 1322 if (vp->v_holdcnt > 0) 1323 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1324 else 1325 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1326 simple_unlock(&vnode_free_list_slock); 1327 if (vp->v_flag & VEXECMAP) { 1328 uvmexp.execpages -= vp->v_uobj.uo_npages; 1329 uvmexp.filepages += vp->v_uobj.uo_npages; 1330 } 1331 vp->v_flag &= ~(VTEXT|VEXECMAP); 1332 simple_unlock(&vp->v_interlock); 1333 VOP_INACTIVE(vp, p); 1334 } 1335 1336 /* 1337 * Vnode release. 1338 * If count drops to zero, call inactive routine and return to freelist. 1339 */ 1340 void 1341 vrele(vp) 1342 struct vnode *vp; 1343 { 1344 struct proc *p = curproc; /* XXX */ 1345 1346 #ifdef DIAGNOSTIC 1347 if (vp == NULL) 1348 panic("vrele: null vp"); 1349 #endif 1350 simple_lock(&vp->v_interlock); 1351 vp->v_usecount--; 1352 if (vp->v_usecount > 0) { 1353 simple_unlock(&vp->v_interlock); 1354 return; 1355 } 1356 #ifdef DIAGNOSTIC 1357 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1358 vprint("vrele: bad ref count", vp); 1359 panic("vrele: ref cnt vp %p", vp); 1360 } 1361 #endif 1362 /* 1363 * Insert at tail of LRU list. 1364 */ 1365 simple_lock(&vnode_free_list_slock); 1366 if (vp->v_holdcnt > 0) 1367 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1368 else 1369 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1370 simple_unlock(&vnode_free_list_slock); 1371 if (vp->v_flag & VEXECMAP) { 1372 uvmexp.execpages -= vp->v_uobj.uo_npages; 1373 uvmexp.filepages += vp->v_uobj.uo_npages; 1374 } 1375 vp->v_flag &= ~(VTEXT|VEXECMAP); 1376 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) 1377 VOP_INACTIVE(vp, p); 1378 } 1379 1380 #ifdef DIAGNOSTIC 1381 /* 1382 * Page or buffer structure gets a reference. 1383 */ 1384 void 1385 vholdl(vp) 1386 struct vnode *vp; 1387 { 1388 1389 /* 1390 * If it is on the freelist and the hold count is currently 1391 * zero, move it to the hold list. The test of the back 1392 * pointer and the use reference count of zero is because 1393 * it will be removed from a free list by getnewvnode, 1394 * but will not have its reference count incremented until 1395 * after calling vgone. If the reference count were 1396 * incremented first, vgone would (incorrectly) try to 1397 * close the previous instance of the underlying object. 1398 * So, the back pointer is explicitly set to `0xdeadb' in 1399 * getnewvnode after removing it from a freelist to ensure 1400 * that we do not try to move it here. 1401 */ 1402 if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) && 1403 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 1404 simple_lock(&vnode_free_list_slock); 1405 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1406 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 1407 simple_unlock(&vnode_free_list_slock); 1408 } 1409 vp->v_holdcnt++; 1410 } 1411 1412 /* 1413 * Page or buffer structure frees a reference. 1414 */ 1415 void 1416 holdrelel(vp) 1417 struct vnode *vp; 1418 { 1419 1420 if (vp->v_holdcnt <= 0) 1421 panic("holdrelel: holdcnt vp %p", vp); 1422 vp->v_holdcnt--; 1423 1424 /* 1425 * If it is on the holdlist and the hold count drops to 1426 * zero, move it to the free list. The test of the back 1427 * pointer and the use reference count of zero is because 1428 * it will be removed from a free list by getnewvnode, 1429 * but will not have its reference count incremented until 1430 * after calling vgone. If the reference count were 1431 * incremented first, vgone would (incorrectly) try to 1432 * close the previous instance of the underlying object. 1433 * So, the back pointer is explicitly set to `0xdeadb' in 1434 * getnewvnode after removing it from a freelist to ensure 1435 * that we do not try to move it here. 1436 */ 1437 1438 if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb) && 1439 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 1440 simple_lock(&vnode_free_list_slock); 1441 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 1442 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1443 simple_unlock(&vnode_free_list_slock); 1444 } 1445 } 1446 1447 /* 1448 * Vnode reference. 1449 */ 1450 void 1451 vref(vp) 1452 struct vnode *vp; 1453 { 1454 1455 simple_lock(&vp->v_interlock); 1456 if (vp->v_usecount <= 0) 1457 panic("vref used where vget required, vp %p", vp); 1458 vp->v_usecount++; 1459 #ifdef DIAGNOSTIC 1460 if (vp->v_usecount == 0) { 1461 vprint("vref", vp); 1462 panic("vref: usecount overflow, vp %p", vp); 1463 } 1464 #endif 1465 simple_unlock(&vp->v_interlock); 1466 } 1467 #endif /* DIAGNOSTIC */ 1468 1469 /* 1470 * Remove any vnodes in the vnode table belonging to mount point mp. 1471 * 1472 * If FORCECLOSE is not specified, there should not be any active ones, 1473 * return error if any are found (nb: this is a user error, not a 1474 * system error). If FORCECLOSE is specified, detach any active vnodes 1475 * that are found. 1476 * 1477 * If WRITECLOSE is set, only flush out regular file vnodes open for 1478 * writing. 1479 * 1480 * SKIPSYSTEM causes any vnodes marked V_SYSTEM to be skipped. 1481 */ 1482 #ifdef DEBUG 1483 int busyprt = 0; /* print out busy vnodes */ 1484 struct ctldebug debug1 = { "busyprt", &busyprt }; 1485 #endif 1486 1487 int 1488 vflush(mp, skipvp, flags) 1489 struct mount *mp; 1490 struct vnode *skipvp; 1491 int flags; 1492 { 1493 struct proc *p = curproc; /* XXX */ 1494 struct vnode *vp, *nvp; 1495 int busy = 0; 1496 1497 simple_lock(&mntvnode_slock); 1498 loop: 1499 for (vp = LIST_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) { 1500 if (vp->v_mount != mp) 1501 goto loop; 1502 nvp = LIST_NEXT(vp, v_mntvnodes); 1503 /* 1504 * Skip over a selected vnode. 1505 */ 1506 if (vp == skipvp) 1507 continue; 1508 simple_lock(&vp->v_interlock); 1509 /* 1510 * Skip over a vnodes marked VSYSTEM. 1511 */ 1512 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1513 simple_unlock(&vp->v_interlock); 1514 continue; 1515 } 1516 /* 1517 * If WRITECLOSE is set, only flush out regular file 1518 * vnodes open for writing. 1519 */ 1520 if ((flags & WRITECLOSE) && 1521 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1522 simple_unlock(&vp->v_interlock); 1523 continue; 1524 } 1525 /* 1526 * With v_usecount == 0, all we need to do is clear 1527 * out the vnode data structures and we are done. 1528 */ 1529 if (vp->v_usecount == 0) { 1530 simple_unlock(&mntvnode_slock); 1531 vgonel(vp, p); 1532 simple_lock(&mntvnode_slock); 1533 continue; 1534 } 1535 /* 1536 * If FORCECLOSE is set, forcibly close the vnode. 1537 * For block or character devices, revert to an 1538 * anonymous device. For all other files, just kill them. 1539 */ 1540 if (flags & FORCECLOSE) { 1541 simple_unlock(&mntvnode_slock); 1542 if (vp->v_type != VBLK && vp->v_type != VCHR) { 1543 vgonel(vp, p); 1544 } else { 1545 vclean(vp, 0, p); 1546 vp->v_op = spec_vnodeop_p; 1547 insmntque(vp, (struct mount *)0); 1548 } 1549 simple_lock(&mntvnode_slock); 1550 continue; 1551 } 1552 #ifdef DEBUG 1553 if (busyprt) 1554 vprint("vflush: busy vnode", vp); 1555 #endif 1556 simple_unlock(&vp->v_interlock); 1557 busy++; 1558 } 1559 simple_unlock(&mntvnode_slock); 1560 if (busy) 1561 return (EBUSY); 1562 return (0); 1563 } 1564 1565 /* 1566 * Disassociate the underlying file system from a vnode. 1567 */ 1568 void 1569 vclean(vp, flags, p) 1570 struct vnode *vp; 1571 int flags; 1572 struct proc *p; 1573 { 1574 struct mount *mp; 1575 int active; 1576 1577 LOCK_ASSERT(simple_lock_held(&vp->v_interlock)); 1578 1579 /* 1580 * Check to see if the vnode is in use. 1581 * If so we have to reference it before we clean it out 1582 * so that its count cannot fall to zero and generate a 1583 * race against ourselves to recycle it. 1584 */ 1585 1586 if ((active = vp->v_usecount) != 0) { 1587 vp->v_usecount++; 1588 #ifdef DIAGNOSTIC 1589 if (vp->v_usecount == 0) { 1590 vprint("vclean", vp); 1591 panic("vclean: usecount overflow"); 1592 } 1593 #endif 1594 } 1595 1596 /* 1597 * Prevent the vnode from being recycled or 1598 * brought into use while we clean it out. 1599 */ 1600 if (vp->v_flag & VXLOCK) 1601 panic("vclean: deadlock, vp %p", vp); 1602 vp->v_flag |= VXLOCK; 1603 if (vp->v_flag & VEXECMAP) { 1604 uvmexp.execpages -= vp->v_uobj.uo_npages; 1605 uvmexp.filepages += vp->v_uobj.uo_npages; 1606 } 1607 vp->v_flag &= ~(VTEXT|VEXECMAP); 1608 1609 /* 1610 * Even if the count is zero, the VOP_INACTIVE routine may still 1611 * have the object locked while it cleans it out. The VOP_LOCK 1612 * ensures that the VOP_INACTIVE routine is done with its work. 1613 * For active vnodes, it ensures that no other activity can 1614 * occur while the underlying object is being cleaned out. 1615 */ 1616 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK); 1617 1618 /* 1619 * Clean out any cached data associated with the vnode. 1620 */ 1621 if (flags & DOCLOSE) { 1622 int error; 1623 vn_start_write(vp, &mp, V_WAIT | V_LOWER); 1624 error = vinvalbuf(vp, V_SAVE, NOCRED, p, 0, 0); 1625 vn_finished_write(mp, V_LOWER); 1626 if (error) 1627 error = vinvalbuf(vp, 0, NOCRED, p, 0, 0); 1628 KASSERT(error == 0); 1629 KASSERT((vp->v_flag & VONWORKLST) == 0); 1630 } 1631 LOCK_ASSERT(!simple_lock_held(&vp->v_interlock)); 1632 1633 /* 1634 * If purging an active vnode, it must be closed and 1635 * deactivated before being reclaimed. Note that the 1636 * VOP_INACTIVE will unlock the vnode. 1637 */ 1638 if (active) { 1639 if (flags & DOCLOSE) 1640 VOP_CLOSE(vp, FNONBLOCK, NOCRED, NULL); 1641 VOP_INACTIVE(vp, p); 1642 } else { 1643 /* 1644 * Any other processes trying to obtain this lock must first 1645 * wait for VXLOCK to clear, then call the new lock operation. 1646 */ 1647 VOP_UNLOCK(vp, 0); 1648 } 1649 /* 1650 * Reclaim the vnode. 1651 */ 1652 if (VOP_RECLAIM(vp, p)) 1653 panic("vclean: cannot reclaim, vp %p", vp); 1654 if (active) { 1655 /* 1656 * Inline copy of vrele() since VOP_INACTIVE 1657 * has already been called. 1658 */ 1659 simple_lock(&vp->v_interlock); 1660 if (--vp->v_usecount <= 0) { 1661 #ifdef DIAGNOSTIC 1662 if (vp->v_usecount < 0 || vp->v_writecount != 0) { 1663 vprint("vclean: bad ref count", vp); 1664 panic("vclean: ref cnt"); 1665 } 1666 #endif 1667 /* 1668 * Insert at tail of LRU list. 1669 */ 1670 1671 simple_unlock(&vp->v_interlock); 1672 simple_lock(&vnode_free_list_slock); 1673 #ifdef DIAGNOSTIC 1674 if (vp->v_holdcnt > 0) 1675 panic("vclean: not clean, vp %p", vp); 1676 #endif 1677 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1678 simple_unlock(&vnode_free_list_slock); 1679 } else 1680 simple_unlock(&vp->v_interlock); 1681 } 1682 1683 KASSERT(vp->v_uobj.uo_npages == 0); 1684 cache_purge(vp); 1685 1686 /* 1687 * Done with purge, notify sleepers of the grim news. 1688 */ 1689 vp->v_op = dead_vnodeop_p; 1690 vp->v_tag = VT_NON; 1691 simple_lock(&vp->v_interlock); 1692 VN_KNOTE(vp, NOTE_REVOKE); /* FreeBSD has this in vn_pollgone() */ 1693 vp->v_flag &= ~VXLOCK; 1694 if (vp->v_flag & VXWANT) { 1695 vp->v_flag &= ~VXWANT; 1696 simple_unlock(&vp->v_interlock); 1697 wakeup((caddr_t)vp); 1698 } else 1699 simple_unlock(&vp->v_interlock); 1700 } 1701 1702 /* 1703 * Recycle an unused vnode to the front of the free list. 1704 * Release the passed interlock if the vnode will be recycled. 1705 */ 1706 int 1707 vrecycle(vp, inter_lkp, p) 1708 struct vnode *vp; 1709 struct simplelock *inter_lkp; 1710 struct proc *p; 1711 { 1712 1713 simple_lock(&vp->v_interlock); 1714 if (vp->v_usecount == 0) { 1715 if (inter_lkp) 1716 simple_unlock(inter_lkp); 1717 vgonel(vp, p); 1718 return (1); 1719 } 1720 simple_unlock(&vp->v_interlock); 1721 return (0); 1722 } 1723 1724 /* 1725 * Eliminate all activity associated with a vnode 1726 * in preparation for reuse. 1727 */ 1728 void 1729 vgone(vp) 1730 struct vnode *vp; 1731 { 1732 struct proc *p = curproc; /* XXX */ 1733 1734 simple_lock(&vp->v_interlock); 1735 vgonel(vp, p); 1736 } 1737 1738 /* 1739 * vgone, with the vp interlock held. 1740 */ 1741 void 1742 vgonel(vp, p) 1743 struct vnode *vp; 1744 struct proc *p; 1745 { 1746 struct vnode *vq; 1747 struct vnode *vx; 1748 1749 LOCK_ASSERT(simple_lock_held(&vp->v_interlock)); 1750 1751 /* 1752 * If a vgone (or vclean) is already in progress, 1753 * wait until it is done and return. 1754 */ 1755 1756 if (vp->v_flag & VXLOCK) { 1757 vp->v_flag |= VXWANT; 1758 ltsleep(vp, PINOD | PNORELOCK, "vgone", 0, &vp->v_interlock); 1759 return; 1760 } 1761 1762 /* 1763 * Clean out the filesystem specific data. 1764 */ 1765 1766 vclean(vp, DOCLOSE, p); 1767 KASSERT((vp->v_flag & VONWORKLST) == 0); 1768 1769 /* 1770 * Delete from old mount point vnode list, if on one. 1771 */ 1772 1773 if (vp->v_mount != NULL) 1774 insmntque(vp, (struct mount *)0); 1775 1776 /* 1777 * If special device, remove it from special device alias list. 1778 * if it is on one. 1779 */ 1780 1781 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1782 simple_lock(&spechash_slock); 1783 if (vp->v_hashchain != NULL) { 1784 if (*vp->v_hashchain == vp) { 1785 *vp->v_hashchain = vp->v_specnext; 1786 } else { 1787 for (vq = *vp->v_hashchain; vq; 1788 vq = vq->v_specnext) { 1789 if (vq->v_specnext != vp) 1790 continue; 1791 vq->v_specnext = vp->v_specnext; 1792 break; 1793 } 1794 if (vq == NULL) 1795 panic("missing bdev"); 1796 } 1797 if (vp->v_flag & VALIASED) { 1798 vx = NULL; 1799 for (vq = *vp->v_hashchain; vq; 1800 vq = vq->v_specnext) { 1801 if (vq->v_rdev != vp->v_rdev || 1802 vq->v_type != vp->v_type) 1803 continue; 1804 if (vx) 1805 break; 1806 vx = vq; 1807 } 1808 if (vx == NULL) 1809 panic("missing alias"); 1810 if (vq == NULL) 1811 vx->v_flag &= ~VALIASED; 1812 vp->v_flag &= ~VALIASED; 1813 } 1814 } 1815 simple_unlock(&spechash_slock); 1816 FREE(vp->v_specinfo, M_VNODE); 1817 vp->v_specinfo = NULL; 1818 } 1819 1820 /* 1821 * The test of the back pointer and the reference count of 1822 * zero is because it will be removed from the free list by 1823 * getcleanvnode, but will not have its reference count 1824 * incremented until after calling vgone. If the reference 1825 * count were incremented first, vgone would (incorrectly) 1826 * try to close the previous instance of the underlying object. 1827 * So, the back pointer is explicitly set to `0xdeadb' in 1828 * getnewvnode after removing it from the freelist to ensure 1829 * that we do not try to move it here. 1830 */ 1831 1832 vp->v_type = VBAD; 1833 if (vp->v_usecount == 0) { 1834 boolean_t dofree; 1835 1836 simple_lock(&vnode_free_list_slock); 1837 if (vp->v_holdcnt > 0) 1838 panic("vgonel: not clean, vp %p", vp); 1839 /* 1840 * if it isn't on the freelist, we're called by getcleanvnode 1841 * and vnode is being re-used. otherwise, we'll free it. 1842 */ 1843 dofree = vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb; 1844 if (dofree) { 1845 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1846 numvnodes--; 1847 } 1848 simple_unlock(&vnode_free_list_slock); 1849 if (dofree) 1850 pool_put(&vnode_pool, vp); 1851 } 1852 } 1853 1854 /* 1855 * Lookup a vnode by device number. 1856 */ 1857 int 1858 vfinddev(dev, type, vpp) 1859 dev_t dev; 1860 enum vtype type; 1861 struct vnode **vpp; 1862 { 1863 struct vnode *vp; 1864 int rc = 0; 1865 1866 simple_lock(&spechash_slock); 1867 for (vp = speclisth[SPECHASH(dev)]; vp; vp = vp->v_specnext) { 1868 if (dev != vp->v_rdev || type != vp->v_type) 1869 continue; 1870 *vpp = vp; 1871 rc = 1; 1872 break; 1873 } 1874 simple_unlock(&spechash_slock); 1875 return (rc); 1876 } 1877 1878 /* 1879 * Revoke all the vnodes corresponding to the specified minor number 1880 * range (endpoints inclusive) of the specified major. 1881 */ 1882 void 1883 vdevgone(maj, minl, minh, type) 1884 int maj, minl, minh; 1885 enum vtype type; 1886 { 1887 struct vnode *vp; 1888 int mn; 1889 1890 for (mn = minl; mn <= minh; mn++) 1891 if (vfinddev(makedev(maj, mn), type, &vp)) 1892 VOP_REVOKE(vp, REVOKEALL); 1893 } 1894 1895 /* 1896 * Calculate the total number of references to a special device. 1897 */ 1898 int 1899 vcount(vp) 1900 struct vnode *vp; 1901 { 1902 struct vnode *vq, *vnext; 1903 int count; 1904 1905 loop: 1906 if ((vp->v_flag & VALIASED) == 0) 1907 return (vp->v_usecount); 1908 simple_lock(&spechash_slock); 1909 for (count = 0, vq = *vp->v_hashchain; vq; vq = vnext) { 1910 vnext = vq->v_specnext; 1911 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1912 continue; 1913 /* 1914 * Alias, but not in use, so flush it out. 1915 */ 1916 if (vq->v_usecount == 0 && vq != vp && 1917 (vq->v_flag & VXLOCK) == 0) { 1918 simple_unlock(&spechash_slock); 1919 vgone(vq); 1920 goto loop; 1921 } 1922 count += vq->v_usecount; 1923 } 1924 simple_unlock(&spechash_slock); 1925 return (count); 1926 } 1927 1928 /* 1929 * Print out a description of a vnode. 1930 */ 1931 const char * const vnode_types[] = { 1932 "VNON", 1933 "VREG", 1934 "VDIR", 1935 "VBLK", 1936 "VCHR", 1937 "VLNK", 1938 "VSOCK", 1939 "VFIFO", 1940 "VBAD" 1941 }; 1942 1943 void 1944 vprint(label, vp) 1945 char *label; 1946 struct vnode *vp; 1947 { 1948 char buf[96]; 1949 1950 if (label != NULL) 1951 printf("%s: ", label); 1952 printf("tag %d type %s, usecount %d, writecount %ld, refcount %ld,", 1953 vp->v_tag, vnode_types[vp->v_type], 1954 vp->v_usecount, vp->v_writecount, vp->v_holdcnt); 1955 buf[0] = '\0'; 1956 if (vp->v_flag & VROOT) 1957 strlcat(buf, "|VROOT", sizeof(buf)); 1958 if (vp->v_flag & VTEXT) 1959 strlcat(buf, "|VTEXT", sizeof(buf)); 1960 if (vp->v_flag & VEXECMAP) 1961 strlcat(buf, "|VEXECMAP", sizeof(buf)); 1962 if (vp->v_flag & VSYSTEM) 1963 strlcat(buf, "|VSYSTEM", sizeof(buf)); 1964 if (vp->v_flag & VXLOCK) 1965 strlcat(buf, "|VXLOCK", sizeof(buf)); 1966 if (vp->v_flag & VXWANT) 1967 strlcat(buf, "|VXWANT", sizeof(buf)); 1968 if (vp->v_flag & VBWAIT) 1969 strlcat(buf, "|VBWAIT", sizeof(buf)); 1970 if (vp->v_flag & VALIASED) 1971 strlcat(buf, "|VALIASED", sizeof(buf)); 1972 if (buf[0] != '\0') 1973 printf(" flags (%s)", &buf[1]); 1974 if (vp->v_data == NULL) { 1975 printf("\n"); 1976 } else { 1977 printf("\n\t"); 1978 VOP_PRINT(vp); 1979 } 1980 } 1981 1982 #ifdef DEBUG 1983 /* 1984 * List all of the locked vnodes in the system. 1985 * Called when debugging the kernel. 1986 */ 1987 void 1988 printlockedvnodes() 1989 { 1990 struct mount *mp, *nmp; 1991 struct vnode *vp; 1992 1993 printf("Locked vnodes\n"); 1994 simple_lock(&mountlist_slock); 1995 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 1996 mp = nmp) { 1997 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 1998 nmp = CIRCLEQ_NEXT(mp, mnt_list); 1999 continue; 2000 } 2001 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2002 if (VOP_ISLOCKED(vp)) 2003 vprint(NULL, vp); 2004 } 2005 simple_lock(&mountlist_slock); 2006 nmp = CIRCLEQ_NEXT(mp, mnt_list); 2007 vfs_unbusy(mp); 2008 } 2009 simple_unlock(&mountlist_slock); 2010 } 2011 #endif 2012 2013 /* 2014 * sysctl helper routine for vfs.generic.conf lookups. 2015 */ 2016 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 2017 static int 2018 sysctl_vfs_generic_conf(SYSCTLFN_ARGS) 2019 { 2020 struct vfsconf vfc; 2021 extern const char * const mountcompatnames[]; 2022 extern int nmountcompatnames; 2023 struct sysctlnode node; 2024 struct vfsops *vfsp; 2025 u_int vfsnum; 2026 2027 if (namelen != 1) 2028 return (ENOTDIR); 2029 vfsnum = name[0]; 2030 if (vfsnum >= nmountcompatnames || 2031 mountcompatnames[vfsnum] == NULL) 2032 return (EOPNOTSUPP); 2033 vfsp = vfs_getopsbyname(mountcompatnames[vfsnum]); 2034 if (vfsp == NULL) 2035 return (EOPNOTSUPP); 2036 2037 vfc.vfc_vfsops = vfsp; 2038 strncpy(vfc.vfc_name, vfsp->vfs_name, MFSNAMELEN); 2039 vfc.vfc_typenum = vfsnum; 2040 vfc.vfc_refcount = vfsp->vfs_refcount; 2041 vfc.vfc_flags = 0; 2042 vfc.vfc_mountroot = vfsp->vfs_mountroot; 2043 vfc.vfc_next = NULL; 2044 2045 node = *rnode; 2046 node.sysctl_data = &vfc; 2047 return (sysctl_lookup(SYSCTLFN_CALL(&node))); 2048 } 2049 #endif 2050 2051 /* 2052 * sysctl helper routine to return list of supported fstypes 2053 */ 2054 static int 2055 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS) 2056 { 2057 char buf[MFSNAMELEN]; 2058 char *where = oldp; 2059 struct vfsops *v; 2060 size_t needed, left, slen; 2061 int error, first; 2062 2063 if (newp != NULL) 2064 return (EPERM); 2065 if (namelen != 0) 2066 return (EINVAL); 2067 2068 first = 1; 2069 error = 0; 2070 needed = 0; 2071 left = *oldlenp; 2072 2073 LIST_FOREACH(v, &vfs_list, vfs_list) { 2074 if (where == NULL) 2075 needed += strlen(v->vfs_name) + 1; 2076 else { 2077 memset(buf, 0, sizeof(buf)); 2078 if (first) { 2079 strncpy(buf, v->vfs_name, sizeof(buf)); 2080 first = 0; 2081 } else { 2082 buf[0] = ' '; 2083 strncpy(buf + 1, v->vfs_name, sizeof(buf) - 1); 2084 } 2085 buf[sizeof(buf)-1] = '\0'; 2086 slen = strlen(buf); 2087 if (left < slen + 1) 2088 break; 2089 /* +1 to copy out the trailing NUL byte */ 2090 error = copyout(buf, where, slen + 1); 2091 if (error) 2092 break; 2093 where += slen; 2094 needed += slen; 2095 left -= slen; 2096 } 2097 } 2098 *oldlenp = needed; 2099 return (error); 2100 } 2101 2102 /* 2103 * Top level filesystem related information gathering. 2104 */ 2105 SYSCTL_SETUP(sysctl_vfs_setup, "sysctl vfs subtree setup") 2106 { 2107 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 2108 extern int nmountcompatnames; 2109 #endif 2110 2111 sysctl_createv(clog, 0, NULL, NULL, 2112 CTLFLAG_PERMANENT, 2113 CTLTYPE_NODE, "vfs", NULL, 2114 NULL, 0, NULL, 0, 2115 CTL_VFS, CTL_EOL); 2116 sysctl_createv(clog, 0, NULL, NULL, 2117 CTLFLAG_PERMANENT, 2118 CTLTYPE_NODE, "generic", 2119 SYSCTL_DESCR("Non-specific vfs related information"), 2120 NULL, 0, NULL, 0, 2121 CTL_VFS, VFS_GENERIC, CTL_EOL); 2122 2123 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 2124 sysctl_createv(clog, 0, NULL, NULL, 2125 CTLFLAG_PERMANENT|CTLFLAG_IMMEDIATE, 2126 CTLTYPE_INT, "maxtypenum", 2127 SYSCTL_DESCR("Highest valid filesystem type number"), 2128 NULL, nmountcompatnames, NULL, 0, 2129 CTL_VFS, VFS_GENERIC, VFS_MAXTYPENUM, CTL_EOL); 2130 #endif 2131 sysctl_createv(clog, 0, NULL, NULL, 2132 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2133 CTLTYPE_INT, "usermount", 2134 SYSCTL_DESCR("Whether unprivileged users may mount " 2135 "filesystems"), 2136 NULL, 0, &dovfsusermount, 0, 2137 CTL_VFS, VFS_GENERIC, VFS_USERMOUNT, CTL_EOL); 2138 sysctl_createv(clog, 0, NULL, NULL, 2139 CTLFLAG_PERMANENT, 2140 CTLTYPE_STRING, "fstypes", 2141 SYSCTL_DESCR("List of file systems present"), 2142 sysctl_vfs_generic_fstypes, 0, NULL, 0, 2143 CTL_VFS, VFS_GENERIC, CTL_CREATE, CTL_EOL); 2144 #if defined(COMPAT_09) || defined(COMPAT_43) || defined(COMPAT_44) 2145 sysctl_createv(clog, 0, NULL, NULL, 2146 CTLFLAG_PERMANENT, 2147 CTLTYPE_STRUCT, "conf", 2148 SYSCTL_DESCR("Filesystem configuration information"), 2149 sysctl_vfs_generic_conf, 0, NULL, 2150 sizeof(struct vfsconf), 2151 CTL_VFS, VFS_GENERIC, VFS_CONF, CTL_EOL); 2152 #endif 2153 } 2154 2155 2156 int kinfo_vdebug = 1; 2157 int kinfo_vgetfailed; 2158 #define KINFO_VNODESLOP 10 2159 /* 2160 * Dump vnode list (via sysctl). 2161 * Copyout address of vnode followed by vnode. 2162 */ 2163 /* ARGSUSED */ 2164 int 2165 sysctl_kern_vnode(SYSCTLFN_ARGS) 2166 { 2167 char *where = oldp; 2168 size_t *sizep = oldlenp; 2169 struct mount *mp, *nmp; 2170 struct vnode *nvp, *vp; 2171 char *bp = where, *savebp; 2172 char *ewhere; 2173 int error; 2174 2175 if (namelen != 0) 2176 return (EOPNOTSUPP); 2177 if (newp != NULL) 2178 return (EPERM); 2179 2180 #define VPTRSZ sizeof(struct vnode *) 2181 #define VNODESZ sizeof(struct vnode) 2182 if (where == NULL) { 2183 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 2184 return (0); 2185 } 2186 ewhere = where + *sizep; 2187 2188 simple_lock(&mountlist_slock); 2189 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist; 2190 mp = nmp) { 2191 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) { 2192 nmp = CIRCLEQ_NEXT(mp, mnt_list); 2193 continue; 2194 } 2195 savebp = bp; 2196 again: 2197 simple_lock(&mntvnode_slock); 2198 for (vp = LIST_FIRST(&mp->mnt_vnodelist); 2199 vp != NULL; 2200 vp = nvp) { 2201 /* 2202 * Check that the vp is still associated with 2203 * this filesystem. RACE: could have been 2204 * recycled onto the same filesystem. 2205 */ 2206 if (vp->v_mount != mp) { 2207 simple_unlock(&mntvnode_slock); 2208 if (kinfo_vdebug) 2209 printf("kinfo: vp changed\n"); 2210 bp = savebp; 2211 goto again; 2212 } 2213 nvp = LIST_NEXT(vp, v_mntvnodes); 2214 if (bp + VPTRSZ + VNODESZ > ewhere) { 2215 simple_unlock(&mntvnode_slock); 2216 *sizep = bp - where; 2217 return (ENOMEM); 2218 } 2219 simple_unlock(&mntvnode_slock); 2220 if ((error = copyout((caddr_t)&vp, bp, VPTRSZ)) || 2221 (error = copyout((caddr_t)vp, bp + VPTRSZ, VNODESZ))) 2222 return (error); 2223 bp += VPTRSZ + VNODESZ; 2224 simple_lock(&mntvnode_slock); 2225 } 2226 simple_unlock(&mntvnode_slock); 2227 simple_lock(&mountlist_slock); 2228 nmp = CIRCLEQ_NEXT(mp, mnt_list); 2229 vfs_unbusy(mp); 2230 } 2231 simple_unlock(&mountlist_slock); 2232 2233 *sizep = bp - where; 2234 return (0); 2235 } 2236 2237 /* 2238 * Check to see if a filesystem is mounted on a block device. 2239 */ 2240 int 2241 vfs_mountedon(vp) 2242 struct vnode *vp; 2243 { 2244 struct vnode *vq; 2245 int error = 0; 2246 2247 if (vp->v_specmountpoint != NULL) 2248 return (EBUSY); 2249 if (vp->v_flag & VALIASED) { 2250 simple_lock(&spechash_slock); 2251 for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { 2252 if (vq->v_rdev != vp->v_rdev || 2253 vq->v_type != vp->v_type) 2254 continue; 2255 if (vq->v_specmountpoint != NULL) { 2256 error = EBUSY; 2257 break; 2258 } 2259 } 2260 simple_unlock(&spechash_slock); 2261 } 2262 return (error); 2263 } 2264 2265 static int 2266 sacheck(struct sockaddr *sa) 2267 { 2268 switch (sa->sa_family) { 2269 #ifdef INET 2270 case AF_INET: { 2271 struct sockaddr_in *sin = (struct sockaddr_in *)sa; 2272 char *p = (char *)sin->sin_zero; 2273 size_t i; 2274 2275 if (sin->sin_len != sizeof(*sin)) 2276 return -1; 2277 if (sin->sin_port != 0) 2278 return -1; 2279 for (i = 0; i < sizeof(sin->sin_zero); i++) 2280 if (*p++ != '\0') 2281 return -1; 2282 return 0; 2283 } 2284 #endif 2285 #ifdef INET6 2286 case AF_INET6: { 2287 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sa; 2288 2289 if (sin6->sin6_len != sizeof(*sin6)) 2290 return -1; 2291 if (sin6->sin6_port != 0) 2292 return -1; 2293 return 0; 2294 } 2295 #endif 2296 default: 2297 return -1; 2298 } 2299 } 2300 2301 /* 2302 * Build hash lists of net addresses and hang them off the mount point. 2303 * Called by ufs_mount() to set up the lists of export addresses. 2304 */ 2305 static int 2306 vfs_hang_addrlist(mp, nep, argp) 2307 struct mount *mp; 2308 struct netexport *nep; 2309 struct export_args *argp; 2310 { 2311 struct netcred *np, *enp; 2312 struct radix_node_head *rnh; 2313 int i; 2314 struct sockaddr *saddr, *smask = 0; 2315 struct domain *dom; 2316 int error; 2317 2318 if (argp->ex_addrlen == 0) { 2319 if (mp->mnt_flag & MNT_DEFEXPORTED) 2320 return (EPERM); 2321 np = &nep->ne_defexported; 2322 np->netc_exflags = argp->ex_flags; 2323 crcvt(&np->netc_anon, &argp->ex_anon); 2324 np->netc_anon.cr_ref = 1; 2325 mp->mnt_flag |= MNT_DEFEXPORTED; 2326 return (0); 2327 } 2328 2329 if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN) 2330 return (EINVAL); 2331 2332 i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 2333 np = (struct netcred *)malloc(i, M_NETADDR, M_WAITOK); 2334 memset((caddr_t)np, 0, i); 2335 saddr = (struct sockaddr *)(np + 1); 2336 error = copyin(argp->ex_addr, (caddr_t)saddr, argp->ex_addrlen); 2337 if (error) 2338 goto out; 2339 if (saddr->sa_len > argp->ex_addrlen) 2340 saddr->sa_len = argp->ex_addrlen; 2341 if (sacheck(saddr) == -1) 2342 return EINVAL; 2343 if (argp->ex_masklen) { 2344 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 2345 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen); 2346 if (error) 2347 goto out; 2348 if (smask->sa_len > argp->ex_masklen) 2349 smask->sa_len = argp->ex_masklen; 2350 if (smask->sa_family != saddr->sa_family) 2351 return EINVAL; 2352 if (sacheck(smask) == -1) 2353 return EINVAL; 2354 } 2355 i = saddr->sa_family; 2356 if ((rnh = nep->ne_rtable[i]) == 0) { 2357 /* 2358 * Seems silly to initialize every AF when most are not 2359 * used, do so on demand here 2360 */ 2361 for (dom = domains; dom; dom = dom->dom_next) 2362 if (dom->dom_family == i && dom->dom_rtattach) { 2363 dom->dom_rtattach((void **)&nep->ne_rtable[i], 2364 dom->dom_rtoffset); 2365 break; 2366 } 2367 if ((rnh = nep->ne_rtable[i]) == 0) { 2368 error = ENOBUFS; 2369 goto out; 2370 } 2371 } 2372 2373 enp = (struct netcred *)(*rnh->rnh_addaddr)(saddr, smask, rnh, 2374 np->netc_rnodes); 2375 if (enp != np) { 2376 if (enp == NULL) { 2377 enp = (struct netcred *)(*rnh->rnh_lookup)(saddr, 2378 smask, rnh); 2379 if (enp == NULL) { 2380 error = EPERM; 2381 goto out; 2382 } 2383 } else 2384 enp->netc_refcnt++; 2385 2386 goto check; 2387 } else 2388 enp->netc_refcnt = 1; 2389 2390 np->netc_exflags = argp->ex_flags; 2391 crcvt(&np->netc_anon, &argp->ex_anon); 2392 np->netc_anon.cr_ref = 1; 2393 return 0; 2394 check: 2395 if (enp->netc_exflags != argp->ex_flags || 2396 crcmp(&enp->netc_anon, &argp->ex_anon) != 0) 2397 error = EPERM; 2398 else 2399 error = 0; 2400 out: 2401 free(np, M_NETADDR); 2402 return error; 2403 } 2404 2405 /* ARGSUSED */ 2406 static int 2407 vfs_free_netcred(rn, w) 2408 struct radix_node *rn; 2409 void *w; 2410 { 2411 struct radix_node_head *rnh = (struct radix_node_head *)w; 2412 struct netcred *np = (struct netcred *)(void *)rn; 2413 2414 (*rnh->rnh_deladdr)(rn->rn_key, rn->rn_mask, rnh); 2415 if (--(np->netc_refcnt) <= 0) 2416 free(np, M_NETADDR); 2417 return (0); 2418 } 2419 2420 /* 2421 * Free the net address hash lists that are hanging off the mount points. 2422 */ 2423 static void 2424 vfs_free_addrlist(nep) 2425 struct netexport *nep; 2426 { 2427 int i; 2428 struct radix_node_head *rnh; 2429 2430 for (i = 0; i <= AF_MAX; i++) 2431 if ((rnh = nep->ne_rtable[i]) != NULL) { 2432 (*rnh->rnh_walktree)(rnh, vfs_free_netcred, rnh); 2433 free((caddr_t)rnh, M_RTABLE); 2434 nep->ne_rtable[i] = 0; 2435 } 2436 } 2437 2438 int 2439 vfs_export(mp, nep, argp) 2440 struct mount *mp; 2441 struct netexport *nep; 2442 struct export_args *argp; 2443 { 2444 int error; 2445 2446 if (argp->ex_flags & MNT_DELEXPORT) { 2447 if (mp->mnt_flag & MNT_EXPUBLIC) { 2448 vfs_setpublicfs(NULL, NULL, NULL); 2449 mp->mnt_flag &= ~MNT_EXPUBLIC; 2450 } 2451 vfs_free_addrlist(nep); 2452 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 2453 } 2454 if (argp->ex_flags & MNT_EXPORTED) { 2455 if (argp->ex_flags & MNT_EXPUBLIC) { 2456 if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) 2457 return (error); 2458 mp->mnt_flag |= MNT_EXPUBLIC; 2459 } 2460 if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0) 2461 return (error); 2462 mp->mnt_flag |= MNT_EXPORTED; 2463 } 2464 return (0); 2465 } 2466 2467 /* 2468 * Set the publicly exported filesystem (WebNFS). Currently, only 2469 * one public filesystem is possible in the spec (RFC 2054 and 2055) 2470 */ 2471 int 2472 vfs_setpublicfs(mp, nep, argp) 2473 struct mount *mp; 2474 struct netexport *nep; 2475 struct export_args *argp; 2476 { 2477 int error; 2478 struct vnode *rvp; 2479 char *cp; 2480 2481 /* 2482 * mp == NULL -> invalidate the current info, the FS is 2483 * no longer exported. May be called from either vfs_export 2484 * or unmount, so check if it hasn't already been done. 2485 */ 2486 if (mp == NULL) { 2487 if (nfs_pub.np_valid) { 2488 nfs_pub.np_valid = 0; 2489 if (nfs_pub.np_index != NULL) { 2490 FREE(nfs_pub.np_index, M_TEMP); 2491 nfs_pub.np_index = NULL; 2492 } 2493 } 2494 return (0); 2495 } 2496 2497 /* 2498 * Only one allowed at a time. 2499 */ 2500 if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount) 2501 return (EBUSY); 2502 2503 /* 2504 * Get real filehandle for root of exported FS. 2505 */ 2506 memset((caddr_t)&nfs_pub.np_handle, 0, sizeof(nfs_pub.np_handle)); 2507 nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsidx; 2508 2509 if ((error = VFS_ROOT(mp, &rvp))) 2510 return (error); 2511 2512 if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid))) 2513 return (error); 2514 2515 vput(rvp); 2516 2517 /* 2518 * If an indexfile was specified, pull it in. 2519 */ 2520 if (argp->ex_indexfile != NULL) { 2521 MALLOC(nfs_pub.np_index, char *, MAXNAMLEN + 1, M_TEMP, 2522 M_WAITOK); 2523 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index, 2524 MAXNAMLEN, (size_t *)0); 2525 if (!error) { 2526 /* 2527 * Check for illegal filenames. 2528 */ 2529 for (cp = nfs_pub.np_index; *cp; cp++) { 2530 if (*cp == '/') { 2531 error = EINVAL; 2532 break; 2533 } 2534 } 2535 } 2536 if (error) { 2537 FREE(nfs_pub.np_index, M_TEMP); 2538 return (error); 2539 } 2540 } 2541 2542 nfs_pub.np_mount = mp; 2543 nfs_pub.np_valid = 1; 2544 return (0); 2545 } 2546 2547 struct netcred * 2548 vfs_export_lookup(mp, nep, nam) 2549 struct mount *mp; 2550 struct netexport *nep; 2551 struct mbuf *nam; 2552 { 2553 struct netcred *np; 2554 struct radix_node_head *rnh; 2555 struct sockaddr *saddr; 2556 2557 np = NULL; 2558 if (mp->mnt_flag & MNT_EXPORTED) { 2559 /* 2560 * Lookup in the export list first. 2561 */ 2562 if (nam != NULL) { 2563 saddr = mtod(nam, struct sockaddr *); 2564 rnh = nep->ne_rtable[saddr->sa_family]; 2565 if (rnh != NULL) { 2566 np = (struct netcred *) 2567 (*rnh->rnh_matchaddr)((caddr_t)saddr, 2568 rnh); 2569 if (np && np->netc_rnodes->rn_flags & RNF_ROOT) 2570 np = NULL; 2571 } 2572 } 2573 /* 2574 * If no address match, use the default if it exists. 2575 */ 2576 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 2577 np = &nep->ne_defexported; 2578 } 2579 return (np); 2580 } 2581 2582 /* 2583 * Do the usual access checking. 2584 * file_mode, uid and gid are from the vnode in question, 2585 * while acc_mode and cred are from the VOP_ACCESS parameter list 2586 */ 2587 int 2588 vaccess(type, file_mode, uid, gid, acc_mode, cred) 2589 enum vtype type; 2590 mode_t file_mode; 2591 uid_t uid; 2592 gid_t gid; 2593 mode_t acc_mode; 2594 struct ucred *cred; 2595 { 2596 mode_t mask; 2597 2598 /* 2599 * Super-user always gets read/write access, but execute access depends 2600 * on at least one execute bit being set. 2601 */ 2602 if (cred->cr_uid == 0) { 2603 if ((acc_mode & VEXEC) && type != VDIR && 2604 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 2605 return (EACCES); 2606 return (0); 2607 } 2608 2609 mask = 0; 2610 2611 /* Otherwise, check the owner. */ 2612 if (cred->cr_uid == uid) { 2613 if (acc_mode & VEXEC) 2614 mask |= S_IXUSR; 2615 if (acc_mode & VREAD) 2616 mask |= S_IRUSR; 2617 if (acc_mode & VWRITE) 2618 mask |= S_IWUSR; 2619 return ((file_mode & mask) == mask ? 0 : EACCES); 2620 } 2621 2622 /* Otherwise, check the groups. */ 2623 if (cred->cr_gid == gid || groupmember(gid, cred)) { 2624 if (acc_mode & VEXEC) 2625 mask |= S_IXGRP; 2626 if (acc_mode & VREAD) 2627 mask |= S_IRGRP; 2628 if (acc_mode & VWRITE) 2629 mask |= S_IWGRP; 2630 return ((file_mode & mask) == mask ? 0 : EACCES); 2631 } 2632 2633 /* Otherwise, check everyone else. */ 2634 if (acc_mode & VEXEC) 2635 mask |= S_IXOTH; 2636 if (acc_mode & VREAD) 2637 mask |= S_IROTH; 2638 if (acc_mode & VWRITE) 2639 mask |= S_IWOTH; 2640 return ((file_mode & mask) == mask ? 0 : EACCES); 2641 } 2642 2643 /* 2644 * Unmount all file systems. 2645 * We traverse the list in reverse order under the assumption that doing so 2646 * will avoid needing to worry about dependencies. 2647 */ 2648 void 2649 vfs_unmountall(p) 2650 struct proc *p; 2651 { 2652 struct mount *mp, *nmp; 2653 int allerror, error; 2654 2655 for (allerror = 0, 2656 mp = mountlist.cqh_last; mp != (void *)&mountlist; mp = nmp) { 2657 nmp = mp->mnt_list.cqe_prev; 2658 #ifdef DEBUG 2659 printf("unmounting %s (%s)...\n", 2660 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_mntfromname); 2661 #endif 2662 /* 2663 * XXX Freeze syncer. Must do this before locking the 2664 * mount point. See dounmount() for details. 2665 */ 2666 lockmgr(&syncer_lock, LK_EXCLUSIVE, NULL); 2667 if (vfs_busy(mp, 0, 0)) { 2668 lockmgr(&syncer_lock, LK_RELEASE, NULL); 2669 continue; 2670 } 2671 if ((error = dounmount(mp, MNT_FORCE, p)) != 0) { 2672 printf("unmount of %s failed with error %d\n", 2673 mp->mnt_stat.f_mntonname, error); 2674 allerror = 1; 2675 } 2676 } 2677 if (allerror) 2678 printf("WARNING: some file systems would not unmount\n"); 2679 } 2680 2681 extern struct simplelock bqueue_slock; /* XXX */ 2682 2683 /* 2684 * Sync and unmount file systems before shutting down. 2685 */ 2686 void 2687 vfs_shutdown() 2688 { 2689 struct lwp *l = curlwp; 2690 struct proc *p; 2691 2692 /* XXX we're certainly not running in proc0's context! */ 2693 if (l == NULL || (p = l->l_proc) == NULL) 2694 p = &proc0; 2695 2696 printf("syncing disks... "); 2697 2698 /* remove user process from run queue */ 2699 suspendsched(); 2700 (void) spl0(); 2701 2702 /* avoid coming back this way again if we panic. */ 2703 doing_shutdown = 1; 2704 2705 sys_sync(l, NULL, NULL); 2706 2707 /* Wait for sync to finish. */ 2708 if (buf_syncwait() != 0) { 2709 #if defined(DDB) && defined(DEBUG_HALT_BUSY) 2710 Debugger(); 2711 #endif 2712 printf("giving up\n"); 2713 return; 2714 } else 2715 printf("done\n"); 2716 2717 /* 2718 * If we've panic'd, don't make the situation potentially 2719 * worse by unmounting the file systems. 2720 */ 2721 if (panicstr != NULL) 2722 return; 2723 2724 /* Release inodes held by texts before update. */ 2725 #ifdef notdef 2726 vnshutdown(); 2727 #endif 2728 /* Unmount file systems. */ 2729 vfs_unmountall(p); 2730 } 2731 2732 /* 2733 * Mount the root file system. If the operator didn't specify a 2734 * file system to use, try all possible file systems until one 2735 * succeeds. 2736 */ 2737 int 2738 vfs_mountroot() 2739 { 2740 struct vfsops *v; 2741 2742 if (root_device == NULL) 2743 panic("vfs_mountroot: root device unknown"); 2744 2745 switch (root_device->dv_class) { 2746 case DV_IFNET: 2747 if (rootdev != NODEV) 2748 panic("vfs_mountroot: rootdev set for DV_IFNET " 2749 "(0x%08x -> %d,%d)", rootdev, 2750 major(rootdev), minor(rootdev)); 2751 break; 2752 2753 case DV_DISK: 2754 if (rootdev == NODEV) 2755 panic("vfs_mountroot: rootdev not set for DV_DISK"); 2756 break; 2757 2758 default: 2759 printf("%s: inappropriate for root file system\n", 2760 root_device->dv_xname); 2761 return (ENODEV); 2762 } 2763 2764 /* 2765 * If user specified a file system, use it. 2766 */ 2767 if (mountroot != NULL) 2768 return ((*mountroot)()); 2769 2770 /* 2771 * Try each file system currently configured into the kernel. 2772 */ 2773 LIST_FOREACH(v, &vfs_list, vfs_list) { 2774 if (v->vfs_mountroot == NULL) 2775 continue; 2776 #ifdef DEBUG 2777 aprint_normal("mountroot: trying %s...\n", v->vfs_name); 2778 #endif 2779 if ((*v->vfs_mountroot)() == 0) { 2780 aprint_normal("root file system type: %s\n", 2781 v->vfs_name); 2782 break; 2783 } 2784 } 2785 2786 if (v == NULL) { 2787 printf("no file system for %s", root_device->dv_xname); 2788 if (root_device->dv_class == DV_DISK) 2789 printf(" (dev 0x%x)", rootdev); 2790 printf("\n"); 2791 return (EFTYPE); 2792 } 2793 return (0); 2794 } 2795 2796 /* 2797 * Given a file system name, look up the vfsops for that 2798 * file system, or return NULL if file system isn't present 2799 * in the kernel. 2800 */ 2801 struct vfsops * 2802 vfs_getopsbyname(name) 2803 const char *name; 2804 { 2805 struct vfsops *v; 2806 2807 LIST_FOREACH(v, &vfs_list, vfs_list) { 2808 if (strcmp(v->vfs_name, name) == 0) 2809 break; 2810 } 2811 2812 return (v); 2813 } 2814 2815 /* 2816 * Establish a file system and initialize it. 2817 */ 2818 int 2819 vfs_attach(vfs) 2820 struct vfsops *vfs; 2821 { 2822 struct vfsops *v; 2823 int error = 0; 2824 2825 2826 /* 2827 * Make sure this file system doesn't already exist. 2828 */ 2829 LIST_FOREACH(v, &vfs_list, vfs_list) { 2830 if (strcmp(vfs->vfs_name, v->vfs_name) == 0) { 2831 error = EEXIST; 2832 goto out; 2833 } 2834 } 2835 2836 /* 2837 * Initialize the vnode operations for this file system. 2838 */ 2839 vfs_opv_init(vfs->vfs_opv_descs); 2840 2841 /* 2842 * Now initialize the file system itself. 2843 */ 2844 (*vfs->vfs_init)(); 2845 2846 /* 2847 * ...and link it into the kernel's list. 2848 */ 2849 LIST_INSERT_HEAD(&vfs_list, vfs, vfs_list); 2850 2851 /* 2852 * Sanity: make sure the reference count is 0. 2853 */ 2854 vfs->vfs_refcount = 0; 2855 2856 out: 2857 return (error); 2858 } 2859 2860 /* 2861 * Remove a file system from the kernel. 2862 */ 2863 int 2864 vfs_detach(vfs) 2865 struct vfsops *vfs; 2866 { 2867 struct vfsops *v; 2868 2869 /* 2870 * Make sure no one is using the filesystem. 2871 */ 2872 if (vfs->vfs_refcount != 0) 2873 return (EBUSY); 2874 2875 /* 2876 * ...and remove it from the kernel's list. 2877 */ 2878 LIST_FOREACH(v, &vfs_list, vfs_list) { 2879 if (v == vfs) { 2880 LIST_REMOVE(v, vfs_list); 2881 break; 2882 } 2883 } 2884 2885 if (v == NULL) 2886 return (ESRCH); 2887 2888 /* 2889 * Now run the file system-specific cleanups. 2890 */ 2891 (*vfs->vfs_done)(); 2892 2893 /* 2894 * Free the vnode operations vector. 2895 */ 2896 vfs_opv_free(vfs->vfs_opv_descs); 2897 return (0); 2898 } 2899 2900 void 2901 vfs_reinit(void) 2902 { 2903 struct vfsops *vfs; 2904 2905 LIST_FOREACH(vfs, &vfs_list, vfs_list) { 2906 if (vfs->vfs_reinit) { 2907 (*vfs->vfs_reinit)(); 2908 } 2909 } 2910 } 2911 2912 /* 2913 * Request a filesystem to suspend write operations. 2914 */ 2915 int 2916 vfs_write_suspend(struct mount *mp, int slpflag, int slptimeo) 2917 { 2918 struct proc *p = curproc; /* XXX */ 2919 int error; 2920 2921 while ((mp->mnt_iflag & IMNT_SUSPEND)) { 2922 if (slptimeo < 0) 2923 return EWOULDBLOCK; 2924 error = tsleep(&mp->mnt_flag, slpflag, "suspwt1", slptimeo); 2925 if (error) 2926 return error; 2927 } 2928 mp->mnt_iflag |= IMNT_SUSPEND; 2929 2930 simple_lock(&mp->mnt_slock); 2931 if (mp->mnt_writeopcountupper > 0) 2932 ltsleep(&mp->mnt_writeopcountupper, PUSER - 1, "suspwt", 2933 0, &mp->mnt_slock); 2934 simple_unlock(&mp->mnt_slock); 2935 2936 error = VFS_SYNC(mp, MNT_WAIT, p->p_ucred, p); 2937 if (error) { 2938 vfs_write_resume(mp); 2939 return error; 2940 } 2941 mp->mnt_iflag |= IMNT_SUSPENDLOW; 2942 2943 simple_lock(&mp->mnt_slock); 2944 if (mp->mnt_writeopcountlower > 0) 2945 ltsleep(&mp->mnt_writeopcountlower, PUSER - 1, "suspwt", 2946 0, &mp->mnt_slock); 2947 mp->mnt_iflag |= IMNT_SUSPENDED; 2948 simple_unlock(&mp->mnt_slock); 2949 2950 return 0; 2951 } 2952 2953 /* 2954 * Request a filesystem to resume write operations. 2955 */ 2956 void 2957 vfs_write_resume(struct mount *mp) 2958 { 2959 2960 if ((mp->mnt_iflag & IMNT_SUSPEND) == 0) 2961 return; 2962 mp->mnt_iflag &= ~(IMNT_SUSPEND | IMNT_SUSPENDLOW | IMNT_SUSPENDED); 2963 wakeup(&mp->mnt_flag); 2964 } 2965 2966 void 2967 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp) 2968 { 2969 const struct statvfs *mbp; 2970 2971 if (sbp == (mbp = &mp->mnt_stat)) 2972 return; 2973 2974 (void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx)); 2975 sbp->f_fsid = mbp->f_fsid; 2976 sbp->f_owner = mbp->f_owner; 2977 sbp->f_flag = mbp->f_flag; 2978 sbp->f_syncwrites = mbp->f_syncwrites; 2979 sbp->f_asyncwrites = mbp->f_asyncwrites; 2980 sbp->f_syncreads = mbp->f_syncreads; 2981 sbp->f_asyncreads = mbp->f_asyncreads; 2982 (void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare)); 2983 (void)memcpy(sbp->f_fstypename, mbp->f_fstypename, 2984 sizeof(sbp->f_fstypename)); 2985 (void)memcpy(sbp->f_mntonname, mbp->f_mntonname, 2986 sizeof(sbp->f_mntonname)); 2987 (void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, 2988 sizeof(sbp->f_mntfromname)); 2989 } 2990 2991 int 2992 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom, 2993 struct mount *mp, struct proc *p) 2994 { 2995 int error; 2996 size_t size; 2997 struct statvfs *sfs = &mp->mnt_stat; 2998 int (*fun)(const void *, void *, size_t, size_t *); 2999 3000 (void)strncpy(mp->mnt_stat.f_fstypename, mp->mnt_op->vfs_name, 3001 sizeof(mp->mnt_stat.f_fstypename)); 3002 3003 if (onp) { 3004 struct cwdinfo *cwdi = p->p_cwdi; 3005 fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr; 3006 if (cwdi->cwdi_rdir != NULL) { 3007 size_t len; 3008 char *bp; 3009 char *path = malloc(MAXPATHLEN, M_TEMP, M_WAITOK); 3010 3011 if (!path) /* XXX can't happen with M_WAITOK */ 3012 return ENOMEM; 3013 3014 bp = path + MAXPATHLEN; 3015 *--bp = '\0'; 3016 error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp, 3017 path, MAXPATHLEN / 2, 0, p); 3018 if (error) { 3019 free(path, M_TEMP); 3020 return error; 3021 } 3022 3023 len = strlen(bp); 3024 if (len > sizeof(sfs->f_mntonname) - 1) 3025 len = sizeof(sfs->f_mntonname) - 1; 3026 (void)strncpy(sfs->f_mntonname, bp, len); 3027 free(path, M_TEMP); 3028 3029 if (len < sizeof(sfs->f_mntonname) - 1) { 3030 error = (*fun)(onp, &sfs->f_mntonname[len], 3031 sizeof(sfs->f_mntonname) - len - 1, &size); 3032 if (error) 3033 return error; 3034 size += len; 3035 } else { 3036 size = len; 3037 } 3038 } else { 3039 error = (*fun)(onp, &sfs->f_mntonname, 3040 sizeof(sfs->f_mntonname) - 1, &size); 3041 if (error) 3042 return error; 3043 } 3044 (void)memset(sfs->f_mntonname + size, 0, 3045 sizeof(sfs->f_mntonname) - size); 3046 } 3047 3048 if (fromp) { 3049 fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr; 3050 error = (*fun)(fromp, sfs->f_mntfromname, 3051 sizeof(sfs->f_mntfromname) - 1, &size); 3052 if (error) 3053 return error; 3054 (void)memset(sfs->f_mntfromname + size, 0, 3055 sizeof(sfs->f_mntfromname) - size); 3056 } 3057 return 0; 3058 } 3059 3060 #ifdef DDB 3061 const char buf_flagbits[] = 3062 "\20\1AGE\2NEEDCOMMIT\3ASYNC\4BAD\5BUSY\6SCANNED\7CALL\10DELWRI" 3063 "\11DIRTY\12DONE\13EINTR\14ERROR\15GATHERED\16INVAL\17LOCKED\20NOCACHE" 3064 "\21ORDERED\22CACHE\23PHYS\24RAW\25READ\26TAPE\30WANTED" 3065 "\32XXX\33VFLUSH"; 3066 3067 void 3068 vfs_buf_print(bp, full, pr) 3069 struct buf *bp; 3070 int full; 3071 void (*pr)(const char *, ...); 3072 { 3073 char buf[1024]; 3074 3075 (*pr)(" vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" dev 0x%x\n", 3076 bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_dev); 3077 3078 bitmask_snprintf(bp->b_flags, buf_flagbits, buf, sizeof(buf)); 3079 (*pr)(" error %d flags 0x%s\n", bp->b_error, buf); 3080 3081 (*pr)(" bufsize 0x%lx bcount 0x%lx resid 0x%lx\n", 3082 bp->b_bufsize, bp->b_bcount, bp->b_resid); 3083 (*pr)(" data %p saveaddr %p dep %p\n", 3084 bp->b_data, bp->b_saveaddr, LIST_FIRST(&bp->b_dep)); 3085 (*pr)(" iodone %p\n", bp->b_iodone); 3086 } 3087 3088 3089 const char vnode_flagbits[] = 3090 "\20\1ROOT\2TEXT\3SYSTEM\4ISTTY\5EXECMAP" 3091 "\11XLOCK\12XWANT\13BWAIT\14ALIASED" 3092 "\15DIROP\16LAYER\17ONWORKLIST\20DIRTY"; 3093 3094 const char * const vnode_tags[] = { 3095 "VT_NON", 3096 "VT_UFS", 3097 "VT_NFS", 3098 "VT_MFS", 3099 "VT_MSDOSFS", 3100 "VT_LFS", 3101 "VT_LOFS", 3102 "VT_FDESC", 3103 "VT_PORTAL", 3104 "VT_NULL", 3105 "VT_UMAP", 3106 "VT_KERNFS", 3107 "VT_PROCFS", 3108 "VT_AFS", 3109 "VT_ISOFS", 3110 "VT_UNION", 3111 "VT_ADOSFS", 3112 "VT_EXT2FS", 3113 "VT_CODA", 3114 "VT_FILECORE", 3115 "VT_NTFS", 3116 "VT_VFS", 3117 "VT_OVERLAY", 3118 "VT_SMBFS" 3119 }; 3120 3121 void 3122 vfs_vnode_print(vp, full, pr) 3123 struct vnode *vp; 3124 int full; 3125 void (*pr)(const char *, ...); 3126 { 3127 char buf[256]; 3128 const char *vtype, *vtag; 3129 3130 uvm_object_printit(&vp->v_uobj, full, pr); 3131 bitmask_snprintf(vp->v_flag, vnode_flagbits, buf, sizeof(buf)); 3132 (*pr)("\nVNODE flags %s\n", buf); 3133 (*pr)("mp %p numoutput %d size 0x%llx\n", 3134 vp->v_mount, vp->v_numoutput, vp->v_size); 3135 3136 (*pr)("data %p usecount %d writecount %ld holdcnt %ld numoutput %d\n", 3137 vp->v_data, vp->v_usecount, vp->v_writecount, 3138 vp->v_holdcnt, vp->v_numoutput); 3139 3140 vtype = (vp->v_type >= 0 && 3141 vp->v_type < sizeof(vnode_types) / sizeof(vnode_types[0])) ? 3142 vnode_types[vp->v_type] : "UNKNOWN"; 3143 vtag = (vp->v_tag >= 0 && 3144 vp->v_tag < sizeof(vnode_tags) / sizeof(vnode_tags[0])) ? 3145 vnode_tags[vp->v_tag] : "UNKNOWN"; 3146 3147 (*pr)("type %s(%d) tag %s(%d) mount %p typedata %p\n", 3148 vtype, vp->v_type, vtag, vp->v_tag, 3149 vp->v_mount, vp->v_mountedhere); 3150 3151 if (full) { 3152 struct buf *bp; 3153 3154 (*pr)("clean bufs:\n"); 3155 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) { 3156 (*pr)(" bp %p\n", bp); 3157 vfs_buf_print(bp, full, pr); 3158 } 3159 3160 (*pr)("dirty bufs:\n"); 3161 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 3162 (*pr)(" bp %p\n", bp); 3163 vfs_buf_print(bp, full, pr); 3164 } 3165 } 3166 } 3167 3168 void 3169 vfs_mount_print(mp, full, pr) 3170 struct mount *mp; 3171 int full; 3172 void (*pr)(const char *, ...); 3173 { 3174 char sbuf[256]; 3175 3176 (*pr)("vnodecovered = %p syncer = %p data = %p\n", 3177 mp->mnt_vnodecovered,mp->mnt_syncer,mp->mnt_data); 3178 3179 (*pr)("fs_bshift %d dev_bshift = %d maxsymlinklen = %d\n", 3180 mp->mnt_fs_bshift,mp->mnt_dev_bshift,mp->mnt_maxsymlinklen); 3181 3182 bitmask_snprintf(mp->mnt_flag, __MNT_FLAG_BITS, sbuf, sizeof(sbuf)); 3183 (*pr)("flag = %s\n", sbuf); 3184 3185 bitmask_snprintf(mp->mnt_iflag, __IMNT_FLAG_BITS, sbuf, sizeof(sbuf)); 3186 (*pr)("iflag = %s\n", sbuf); 3187 3188 /* XXX use lockmgr_printinfo */ 3189 if (mp->mnt_lock.lk_sharecount) 3190 (*pr)(" lock type %s: SHARED (count %d)", mp->mnt_lock.lk_wmesg, 3191 mp->mnt_lock.lk_sharecount); 3192 else if (mp->mnt_lock.lk_flags & LK_HAVE_EXCL) { 3193 (*pr)(" lock type %s: EXCL (count %d) by ", 3194 mp->mnt_lock.lk_wmesg, mp->mnt_lock.lk_exclusivecount); 3195 if (mp->mnt_lock.lk_flags & LK_SPIN) 3196 (*pr)("processor %lu", mp->mnt_lock.lk_cpu); 3197 else 3198 (*pr)("pid %d.%d", mp->mnt_lock.lk_lockholder, 3199 mp->mnt_lock.lk_locklwp); 3200 } else 3201 (*pr)(" not locked"); 3202 if ((mp->mnt_lock.lk_flags & LK_SPIN) == 0 && mp->mnt_lock.lk_waitcount > 0) 3203 (*pr)(" with %d pending", mp->mnt_lock.lk_waitcount); 3204 3205 (*pr)("\n"); 3206 3207 if (mp->mnt_unmounter) { 3208 (*pr)("unmounter pid = %d ",mp->mnt_unmounter->p_pid); 3209 } 3210 (*pr)("wcnt = %d, writeopcountupper = %d, writeopcountupper = %d\n", 3211 mp->mnt_wcnt,mp->mnt_writeopcountupper,mp->mnt_writeopcountlower); 3212 3213 (*pr)("statvfs cache:\n"); 3214 (*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize); 3215 (*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize); 3216 (*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize); 3217 3218 (*pr)("\tblocks = "PRIu64"\n",mp->mnt_stat.f_blocks); 3219 (*pr)("\tbfree = "PRIu64"\n",mp->mnt_stat.f_bfree); 3220 (*pr)("\tbavail = "PRIu64"\n",mp->mnt_stat.f_bavail); 3221 (*pr)("\tbresvd = "PRIu64"\n",mp->mnt_stat.f_bresvd); 3222 3223 (*pr)("\tfiles = "PRIu64"\n",mp->mnt_stat.f_files); 3224 (*pr)("\tffree = "PRIu64"\n",mp->mnt_stat.f_ffree); 3225 (*pr)("\tfavail = "PRIu64"\n",mp->mnt_stat.f_favail); 3226 (*pr)("\tfresvd = "PRIu64"\n",mp->mnt_stat.f_fresvd); 3227 3228 (*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n", 3229 mp->mnt_stat.f_fsidx.__fsid_val[0], 3230 mp->mnt_stat.f_fsidx.__fsid_val[1]); 3231 3232 (*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner); 3233 (*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax); 3234 3235 bitmask_snprintf(mp->mnt_stat.f_flag, __MNT_FLAG_BITS, sbuf, 3236 sizeof(sbuf)); 3237 (*pr)("\tflag = %s\n",sbuf); 3238 (*pr)("\tsyncwrites = " PRIu64 "\n",mp->mnt_stat.f_syncwrites); 3239 (*pr)("\tasyncwrites = " PRIu64 "\n",mp->mnt_stat.f_asyncwrites); 3240 (*pr)("\tsyncreads = " PRIu64 "\n",mp->mnt_stat.f_syncreads); 3241 (*pr)("\tasyncreads = " PRIu64 "\n",mp->mnt_stat.f_asyncreads); 3242 (*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename); 3243 (*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname); 3244 (*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname); 3245 3246 { 3247 int cnt = 0; 3248 struct vnode *vp; 3249 (*pr)("locked vnodes ="); 3250 /* XXX would take mountlist lock, except ddb may not have context */ 3251 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 3252 if (VOP_ISLOCKED(vp)) { 3253 if ((++cnt % 6) == 0) { 3254 (*pr)(" %p,\n\t", vp); 3255 } else { 3256 (*pr)(" %p,", vp); 3257 } 3258 } 3259 } 3260 (*pr)("\n"); 3261 } 3262 3263 if (full) { 3264 int cnt = 0; 3265 struct vnode *vp; 3266 (*pr)("all vnodes ="); 3267 /* XXX would take mountlist lock, except ddb may not have context */ 3268 LIST_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 3269 if (!LIST_NEXT(vp, v_mntvnodes)) { 3270 (*pr)(" %p", vp); 3271 } else if ((++cnt % 6) == 0) { 3272 (*pr)(" %p,\n\t", vp); 3273 } else { 3274 (*pr)(" %p,", vp); 3275 } 3276 } 3277 (*pr)("\n", vp); 3278 } 3279 } 3280 3281 #endif 3282