1 /* $OpenBSD: vfs_subr.c,v 1.307 2021/10/19 06:26:08 semarie Exp $ */ 2 /* $NetBSD: vfs_subr.c,v 1.53 1996/04/22 01:39:13 christos Exp $ */ 3 4 /* 5 * Copyright (c) 1989, 1993 6 * The Regents of the University of California. All rights reserved. 7 * (c) UNIX System Laboratories, Inc. 8 * All or some portions of this file are derived from material licensed 9 * to the University of California by American Telephone and Telegraph 10 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 11 * the permission of UNIX System Laboratories, Inc. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 38 */ 39 40 /* 41 * External virtual filesystem routines 42 */ 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/proc.h> 47 #include <sys/sysctl.h> 48 #include <sys/mount.h> 49 #include <sys/time.h> 50 #include <sys/fcntl.h> 51 #include <sys/kernel.h> 52 #include <sys/conf.h> 53 #include <sys/vnode.h> 54 #include <sys/lock.h> 55 #include <sys/lockf.h> 56 #include <sys/stat.h> 57 #include <sys/acct.h> 58 #include <sys/namei.h> 59 #include <sys/ucred.h> 60 #include <sys/buf.h> 61 #include <sys/errno.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/syscallargs.h> 65 #include <sys/pool.h> 66 #include <sys/tree.h> 67 #include <sys/specdev.h> 68 #include <sys/atomic.h> 69 70 #include <netinet/in.h> 71 72 #include <uvm/uvm_extern.h> 73 #include <uvm/uvm_vnode.h> 74 75 #include "softraid.h" 76 77 void sr_quiesce(void); 78 79 enum vtype iftovt_tab[16] = { 80 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 81 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 82 }; 83 84 int vttoif_tab[9] = { 85 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 86 S_IFSOCK, S_IFIFO, S_IFMT, 87 }; 88 89 int prtactive = 0; /* 1 => print out reclaim of active vnodes */ 90 int suid_clear = 1; /* 1 => clear SUID / SGID on owner change */ 91 92 /* 93 * Insq/Remq for the vnode usage lists. 94 */ 95 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 96 #define bufremvn(bp) { \ 97 LIST_REMOVE(bp, b_vnbufs); \ 98 LIST_NEXT(bp, b_vnbufs) = NOLIST; \ 99 } 100 101 struct freelst vnode_hold_list; /* list of vnodes referencing buffers */ 102 struct freelst vnode_free_list; /* vnode free list */ 103 104 struct mntlist mountlist; /* mounted filesystem list */ 105 106 void vclean(struct vnode *, int, struct proc *); 107 108 void insmntque(struct vnode *, struct mount *); 109 int getdevvp(dev_t, struct vnode **, enum vtype); 110 111 int vfs_hang_addrlist(struct mount *, struct netexport *, 112 struct export_args *); 113 int vfs_free_netcred(struct radix_node *, void *, u_int); 114 void vfs_free_addrlist(struct netexport *); 115 void vputonfreelist(struct vnode *); 116 117 int vflush_vnode(struct vnode *, void *); 118 int maxvnodes; 119 120 struct mutex vnode_mtx = MUTEX_INITIALIZER(IPL_BIO); 121 122 void vfs_unmountall(void); 123 124 #ifdef DEBUG 125 void printlockedvnodes(void); 126 #endif 127 128 struct pool vnode_pool; 129 struct pool uvm_vnode_pool; 130 131 static inline int rb_buf_compare(const struct buf *b1, const struct buf *b2); 132 RBT_GENERATE(buf_rb_bufs, buf, b_rbbufs, rb_buf_compare); 133 134 static inline int 135 rb_buf_compare(const struct buf *b1, const struct buf *b2) 136 { 137 if (b1->b_lblkno < b2->b_lblkno) 138 return(-1); 139 if (b1->b_lblkno > b2->b_lblkno) 140 return(1); 141 return(0); 142 } 143 144 /* 145 * Initialize the vnode management data structures. 146 */ 147 void 148 vntblinit(void) 149 { 150 /* buffer cache may need a vnode for each buffer */ 151 maxvnodes = 2 * initialvnodes; 152 pool_init(&vnode_pool, sizeof(struct vnode), 0, IPL_NONE, 153 PR_WAITOK, "vnodes", NULL); 154 pool_init(&uvm_vnode_pool, sizeof(struct uvm_vnode), 0, IPL_NONE, 155 PR_WAITOK, "uvmvnodes", NULL); 156 TAILQ_INIT(&vnode_hold_list); 157 TAILQ_INIT(&vnode_free_list); 158 TAILQ_INIT(&mountlist); 159 /* 160 * Initialize the filesystem syncer. 161 */ 162 vn_initialize_syncerd(); 163 164 #ifdef NFSSERVER 165 rn_init(sizeof(struct sockaddr_in)); 166 #endif /* NFSSERVER */ 167 } 168 169 /* 170 * Allocate a mount point. 171 * 172 * The returned mount point is marked as busy. 173 */ 174 struct mount * 175 vfs_mount_alloc(struct vnode *vp, struct vfsconf *vfsp) 176 { 177 struct mount *mp; 178 179 mp = malloc(sizeof(*mp), M_MOUNT, M_WAITOK|M_ZERO); 180 rw_init_flags(&mp->mnt_lock, "vfslock", RWL_IS_VNODE); 181 (void)vfs_busy(mp, VB_READ|VB_NOWAIT); 182 183 TAILQ_INIT(&mp->mnt_vnodelist); 184 mp->mnt_vnodecovered = vp; 185 186 atomic_inc_int(&vfsp->vfc_refcount); 187 mp->mnt_vfc = vfsp; 188 mp->mnt_op = vfsp->vfc_vfsops; 189 mp->mnt_flag = vfsp->vfc_flags; 190 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 191 192 return (mp); 193 } 194 195 /* 196 * Release a mount point. 197 */ 198 void 199 vfs_mount_free(struct mount *mp) 200 { 201 atomic_dec_int(&mp->mnt_vfc->vfc_refcount); 202 free(mp, M_MOUNT, sizeof(*mp)); 203 } 204 205 /* 206 * Mark a mount point as busy. Used to synchronize access and to delay 207 * unmounting. 208 * 209 * Default behaviour is to attempt getting a READ lock and in case of an 210 * ongoing unmount, to wait for it to finish and then return failure. 211 */ 212 int 213 vfs_busy(struct mount *mp, int flags) 214 { 215 int rwflags = 0; 216 217 if (flags & VB_WRITE) 218 rwflags |= RW_WRITE; 219 else 220 rwflags |= RW_READ; 221 222 if (flags & VB_WAIT) 223 rwflags |= RW_SLEEPFAIL; 224 else 225 rwflags |= RW_NOSLEEP; 226 227 #ifdef WITNESS 228 if (flags & VB_DUPOK) 229 rwflags |= RW_DUPOK; 230 #endif 231 232 if (rw_enter(&mp->mnt_lock, rwflags)) 233 return (EBUSY); 234 235 return (0); 236 } 237 238 /* 239 * Free a busy file system 240 */ 241 void 242 vfs_unbusy(struct mount *mp) 243 { 244 rw_exit(&mp->mnt_lock); 245 } 246 247 int 248 vfs_isbusy(struct mount *mp) 249 { 250 if (RWLOCK_OWNER(&mp->mnt_lock) > 0) 251 return (1); 252 else 253 return (0); 254 } 255 256 /* 257 * Lookup a filesystem type, and if found allocate and initialize 258 * a mount structure for it. 259 * 260 * Devname is usually updated by mount(8) after booting. 261 */ 262 int 263 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp) 264 { 265 struct vfsconf *vfsp; 266 struct mount *mp; 267 268 vfsp = vfs_byname(fstypename); 269 if (vfsp == NULL) 270 return (ENODEV); 271 mp = vfs_mount_alloc(NULLVP, vfsp); 272 mp->mnt_flag |= MNT_RDONLY; 273 mp->mnt_stat.f_mntonname[0] = '/'; 274 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN, 0); 275 copystr(devname, mp->mnt_stat.f_mntfromspec, MNAMELEN, 0); 276 *mpp = mp; 277 return (0); 278 } 279 280 /* 281 * Lookup a mount point by filesystem identifier. 282 */ 283 struct mount * 284 vfs_getvfs(fsid_t *fsid) 285 { 286 struct mount *mp; 287 288 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 289 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 290 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 291 return (mp); 292 } 293 } 294 295 return (NULL); 296 } 297 298 299 /* 300 * Get a new unique fsid 301 */ 302 void 303 vfs_getnewfsid(struct mount *mp) 304 { 305 static u_short xxxfs_mntid; 306 307 fsid_t tfsid; 308 int mtype; 309 310 mtype = mp->mnt_vfc->vfc_typenum; 311 mp->mnt_stat.f_fsid.val[0] = makedev(nblkdev + mtype, 0); 312 mp->mnt_stat.f_fsid.val[1] = mtype; 313 if (xxxfs_mntid == 0) 314 ++xxxfs_mntid; 315 tfsid.val[0] = makedev(nblkdev + mtype, xxxfs_mntid); 316 tfsid.val[1] = mtype; 317 if (!TAILQ_EMPTY(&mountlist)) { 318 while (vfs_getvfs(&tfsid)) { 319 tfsid.val[0]++; 320 xxxfs_mntid++; 321 } 322 } 323 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 324 } 325 326 /* 327 * Set vnode attributes to VNOVAL 328 */ 329 void 330 vattr_null(struct vattr *vap) 331 { 332 333 vap->va_type = VNON; 334 /* 335 * Don't get fancy: u_quad_t = u_int = VNOVAL leaves the u_quad_t 336 * with 2^31-1 instead of 2^64-1. Just write'm out and let 337 * the compiler do its job. 338 */ 339 vap->va_mode = VNOVAL; 340 vap->va_nlink = VNOVAL; 341 vap->va_uid = VNOVAL; 342 vap->va_gid = VNOVAL; 343 vap->va_fsid = VNOVAL; 344 vap->va_fileid = VNOVAL; 345 vap->va_size = VNOVAL; 346 vap->va_blocksize = VNOVAL; 347 vap->va_atime.tv_sec = VNOVAL; 348 vap->va_atime.tv_nsec = VNOVAL; 349 vap->va_mtime.tv_sec = VNOVAL; 350 vap->va_mtime.tv_nsec = VNOVAL; 351 vap->va_ctime.tv_sec = VNOVAL; 352 vap->va_ctime.tv_nsec = VNOVAL; 353 vap->va_gen = VNOVAL; 354 vap->va_flags = VNOVAL; 355 vap->va_rdev = VNOVAL; 356 vap->va_bytes = VNOVAL; 357 vap->va_filerev = VNOVAL; 358 vap->va_vaflags = 0; 359 } 360 361 /* 362 * Routines having to do with the management of the vnode table. 363 */ 364 long numvnodes; 365 366 /* 367 * Return the next vnode from the free list. 368 */ 369 int 370 getnewvnode(enum vtagtype tag, struct mount *mp, const struct vops *vops, 371 struct vnode **vpp) 372 { 373 struct proc *p = curproc; 374 struct freelst *listhd; 375 static int toggle; 376 struct vnode *vp; 377 int s; 378 379 /* 380 * allow maxvnodes to increase if the buffer cache itself 381 * is big enough to justify it. (we don't shrink it ever) 382 */ 383 maxvnodes = maxvnodes < bcstats.numbufs ? bcstats.numbufs 384 : maxvnodes; 385 386 /* 387 * We must choose whether to allocate a new vnode or recycle an 388 * existing one. The criterion for allocating a new one is that 389 * the total number of vnodes is less than the number desired or 390 * there are no vnodes on either free list. Generally we only 391 * want to recycle vnodes that have no buffers associated with 392 * them, so we look first on the vnode_free_list. If it is empty, 393 * we next consider vnodes with referencing buffers on the 394 * vnode_hold_list. The toggle ensures that half the time we 395 * will use a buffer from the vnode_hold_list, and half the time 396 * we will allocate a new one unless the list has grown to twice 397 * the desired size. We are reticent to recycle vnodes from the 398 * vnode_hold_list because we will lose the identity of all its 399 * referencing buffers. 400 */ 401 toggle ^= 1; 402 if (numvnodes / 2 > maxvnodes) 403 toggle = 0; 404 405 s = splbio(); 406 if ((numvnodes < maxvnodes) || 407 ((TAILQ_FIRST(listhd = &vnode_free_list) == NULL) && 408 ((TAILQ_FIRST(listhd = &vnode_hold_list) == NULL) || toggle))) { 409 splx(s); 410 vp = pool_get(&vnode_pool, PR_WAITOK | PR_ZERO); 411 vp->v_uvm = pool_get(&uvm_vnode_pool, PR_WAITOK | PR_ZERO); 412 vp->v_uvm->u_vnode = vp; 413 RBT_INIT(buf_rb_bufs, &vp->v_bufs_tree); 414 cache_tree_init(&vp->v_nc_tree); 415 TAILQ_INIT(&vp->v_cache_dst); 416 numvnodes++; 417 } else { 418 TAILQ_FOREACH(vp, listhd, v_freelist) { 419 if (VOP_ISLOCKED(vp) == 0) 420 break; 421 } 422 /* 423 * Unless this is a bad time of the month, at most 424 * the first NCPUS items on the free list are 425 * locked, so this is close enough to being empty. 426 */ 427 if (vp == NULL) { 428 splx(s); 429 tablefull("vnode"); 430 *vpp = 0; 431 return (ENFILE); 432 } 433 434 #ifdef DIAGNOSTIC 435 if (vp->v_usecount) { 436 vprint("free vnode", vp); 437 panic("free vnode isn't"); 438 } 439 #endif 440 441 TAILQ_REMOVE(listhd, vp, v_freelist); 442 vp->v_bioflag &= ~VBIOONFREELIST; 443 splx(s); 444 445 if (vp->v_type != VBAD) 446 vgonel(vp, p); 447 #ifdef DIAGNOSTIC 448 if (vp->v_data) { 449 vprint("cleaned vnode", vp); 450 panic("cleaned vnode isn't"); 451 } 452 s = splbio(); 453 if (vp->v_numoutput) 454 panic("Clean vnode has pending I/O's"); 455 splx(s); 456 #endif 457 vp->v_flag = 0; 458 vp->v_socket = NULL; 459 } 460 cache_purge(vp); 461 vp->v_type = VNON; 462 vp->v_tag = tag; 463 vp->v_op = vops; 464 insmntque(vp, mp); 465 *vpp = vp; 466 vp->v_usecount = 1; 467 vp->v_data = 0; 468 return (0); 469 } 470 471 /* 472 * Move a vnode from one mount queue to another. 473 */ 474 void 475 insmntque(struct vnode *vp, struct mount *mp) 476 { 477 /* 478 * Delete from old mount point vnode list, if on one. 479 */ 480 if (vp->v_mount != NULL) 481 TAILQ_REMOVE(&vp->v_mount->mnt_vnodelist, vp, v_mntvnodes); 482 /* 483 * Insert into list of vnodes for the new mount point, if available. 484 */ 485 if ((vp->v_mount = mp) != NULL) 486 TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes); 487 } 488 489 /* 490 * Create a vnode for a block device. 491 * Used for root filesystem, argdev, and swap areas. 492 * Also used for memory file system special devices. 493 */ 494 int 495 bdevvp(dev_t dev, struct vnode **vpp) 496 { 497 return (getdevvp(dev, vpp, VBLK)); 498 } 499 500 /* 501 * Create a vnode for a character device. 502 * Used for console handling. 503 */ 504 int 505 cdevvp(dev_t dev, struct vnode **vpp) 506 { 507 return (getdevvp(dev, vpp, VCHR)); 508 } 509 510 /* 511 * Create a vnode for a device. 512 * Used by bdevvp (block device) for root file system etc., 513 * and by cdevvp (character device) for console. 514 */ 515 int 516 getdevvp(dev_t dev, struct vnode **vpp, enum vtype type) 517 { 518 struct vnode *vp; 519 struct vnode *nvp; 520 int error; 521 522 if (dev == NODEV) { 523 *vpp = NULLVP; 524 return (0); 525 } 526 error = getnewvnode(VT_NON, NULL, &spec_vops, &nvp); 527 if (error) { 528 *vpp = NULLVP; 529 return (error); 530 } 531 vp = nvp; 532 vp->v_type = type; 533 if ((nvp = checkalias(vp, dev, NULL)) != 0) { 534 vput(vp); 535 vp = nvp; 536 } 537 if (vp->v_type == VCHR && cdevsw[major(vp->v_rdev)].d_type == D_TTY) 538 vp->v_flag |= VISTTY; 539 *vpp = vp; 540 return (0); 541 } 542 543 /* 544 * Check to see if the new vnode represents a special device 545 * for which we already have a vnode (either because of 546 * bdevvp() or because of a different vnode representing 547 * the same block device). If such an alias exists, deallocate 548 * the existing contents and return the aliased vnode. The 549 * caller is responsible for filling it with its new contents. 550 */ 551 struct vnode * 552 checkalias(struct vnode *nvp, dev_t nvp_rdev, struct mount *mp) 553 { 554 struct proc *p = curproc; 555 struct vnode *vp; 556 struct vnodechain *vchain; 557 558 if (nvp->v_type != VBLK && nvp->v_type != VCHR) 559 return (NULLVP); 560 561 vchain = &speclisth[SPECHASH(nvp_rdev)]; 562 loop: 563 SLIST_FOREACH(vp, vchain, v_specnext) { 564 if (nvp_rdev != vp->v_rdev || nvp->v_type != vp->v_type) { 565 continue; 566 } 567 /* 568 * Alias, but not in use, so flush it out. 569 */ 570 if (vp->v_usecount == 0) { 571 vgonel(vp, p); 572 goto loop; 573 } 574 if (vget(vp, LK_EXCLUSIVE)) { 575 goto loop; 576 } 577 break; 578 } 579 580 /* 581 * Common case is actually in the if statement 582 */ 583 if (vp == NULL || !(vp->v_tag == VT_NON && vp->v_type == VBLK)) { 584 nvp->v_specinfo = malloc(sizeof(struct specinfo), M_VNODE, 585 M_WAITOK); 586 nvp->v_rdev = nvp_rdev; 587 nvp->v_hashchain = vchain; 588 nvp->v_specmountpoint = NULL; 589 nvp->v_speclockf = NULL; 590 nvp->v_specbitmap = NULL; 591 if (nvp->v_type == VCHR && 592 (cdevsw[major(nvp_rdev)].d_flags & D_CLONE) && 593 (minor(nvp_rdev) >> CLONE_SHIFT == 0)) { 594 if (vp != NULLVP) 595 nvp->v_specbitmap = vp->v_specbitmap; 596 else 597 nvp->v_specbitmap = malloc(CLONE_MAPSZ, 598 M_VNODE, M_WAITOK | M_ZERO); 599 } 600 SLIST_INSERT_HEAD(vchain, nvp, v_specnext); 601 if (vp != NULLVP) { 602 nvp->v_flag |= VALIASED; 603 vp->v_flag |= VALIASED; 604 vput(vp); 605 } 606 return (NULLVP); 607 } 608 609 /* 610 * This code is the uncommon case. It is called in case 611 * we found an alias that was VT_NON && vtype of VBLK 612 * This means we found a block device that was created 613 * using bdevvp. 614 * An example of such a vnode is the root partition device vnode 615 * created in ffs_mountroot. 616 * 617 * The vnodes created by bdevvp should not be aliased (why?). 618 */ 619 620 VOP_UNLOCK(vp); 621 vclean(vp, 0, p); 622 vp->v_op = nvp->v_op; 623 vp->v_tag = nvp->v_tag; 624 nvp->v_type = VNON; 625 insmntque(vp, mp); 626 return (vp); 627 } 628 629 /* 630 * Grab a particular vnode from the free list, increment its 631 * reference count and lock it. If the vnode lock bit is set, 632 * the vnode is being eliminated in vgone. In that case, we 633 * cannot grab it, so the process is awakened when the 634 * transition is completed, and an error code is returned to 635 * indicate that the vnode is no longer usable, possibly 636 * having been changed to a new file system type. 637 */ 638 int 639 vget(struct vnode *vp, int flags) 640 { 641 int error, s, onfreelist; 642 643 /* 644 * If the vnode is in the process of being cleaned out for 645 * another use, we wait for the cleaning to finish and then 646 * return failure. Cleaning is determined by checking that 647 * the VXLOCK flag is set. 648 */ 649 mtx_enter(&vnode_mtx); 650 if (vp->v_lflag & VXLOCK) { 651 if (flags & LK_NOWAIT) { 652 mtx_leave(&vnode_mtx); 653 return (EBUSY); 654 } 655 656 vp->v_lflag |= VXWANT; 657 msleep_nsec(vp, &vnode_mtx, PINOD, "vget", INFSLP); 658 mtx_leave(&vnode_mtx); 659 return (ENOENT); 660 } 661 mtx_leave(&vnode_mtx); 662 663 onfreelist = vp->v_bioflag & VBIOONFREELIST; 664 if (vp->v_usecount == 0 && onfreelist) { 665 s = splbio(); 666 if (vp->v_holdcnt > 0) 667 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 668 else 669 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 670 vp->v_bioflag &= ~VBIOONFREELIST; 671 splx(s); 672 } 673 674 vp->v_usecount++; 675 if (flags & LK_TYPE_MASK) { 676 if ((error = vn_lock(vp, flags)) != 0) { 677 vp->v_usecount--; 678 if (vp->v_usecount == 0 && onfreelist) 679 vputonfreelist(vp); 680 } 681 return (error); 682 } 683 684 return (0); 685 } 686 687 688 /* Vnode reference. */ 689 void 690 vref(struct vnode *vp) 691 { 692 #ifdef DIAGNOSTIC 693 if (vp->v_usecount == 0) 694 panic("vref used where vget required"); 695 if (vp->v_type == VNON) 696 panic("vref on a VNON vnode"); 697 #endif 698 vp->v_usecount++; 699 } 700 701 void 702 vputonfreelist(struct vnode *vp) 703 { 704 int s; 705 struct freelst *lst; 706 707 s = splbio(); 708 #ifdef DIAGNOSTIC 709 if (vp->v_usecount != 0) 710 panic("Use count is not zero!"); 711 712 /* 713 * If the hold count is still positive, one or many threads could still 714 * be waiting on the vnode lock inside uvn_io(). 715 */ 716 if (vp->v_holdcnt == 0 && vp->v_lockcount != 0) 717 panic("%s: lock count is not zero", __func__); 718 719 if (vp->v_bioflag & VBIOONFREELIST) { 720 vprint("vnode already on free list: ", vp); 721 panic("vnode already on free list"); 722 } 723 #endif 724 725 vp->v_bioflag |= VBIOONFREELIST; 726 vp->v_bioflag &= ~VBIOERROR; 727 728 if (vp->v_holdcnt > 0) 729 lst = &vnode_hold_list; 730 else 731 lst = &vnode_free_list; 732 733 if (vp->v_type == VBAD) 734 TAILQ_INSERT_HEAD(lst, vp, v_freelist); 735 else 736 TAILQ_INSERT_TAIL(lst, vp, v_freelist); 737 738 splx(s); 739 } 740 741 /* 742 * vput(), just unlock and vrele() 743 */ 744 void 745 vput(struct vnode *vp) 746 { 747 struct proc *p = curproc; 748 749 #ifdef DIAGNOSTIC 750 if (vp == NULL) 751 panic("vput: null vp"); 752 #endif 753 754 #ifdef DIAGNOSTIC 755 if (vp->v_usecount == 0) { 756 vprint("vput: bad ref count", vp); 757 panic("vput: ref cnt"); 758 } 759 #endif 760 vp->v_usecount--; 761 KASSERT(vp->v_usecount > 0 || vp->v_uvcount == 0); 762 if (vp->v_usecount > 0) { 763 VOP_UNLOCK(vp); 764 return; 765 } 766 767 #ifdef DIAGNOSTIC 768 if (vp->v_writecount != 0) { 769 vprint("vput: bad writecount", vp); 770 panic("vput: v_writecount != 0"); 771 } 772 #endif 773 774 VOP_INACTIVE(vp, p); 775 776 if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST)) 777 vputonfreelist(vp); 778 } 779 780 /* 781 * Vnode release - use for active VNODES. 782 * If count drops to zero, call inactive routine and return to freelist. 783 * Returns 0 if it did not sleep. 784 */ 785 int 786 vrele(struct vnode *vp) 787 { 788 struct proc *p = curproc; 789 790 #ifdef DIAGNOSTIC 791 if (vp == NULL) 792 panic("vrele: null vp"); 793 #endif 794 #ifdef DIAGNOSTIC 795 if (vp->v_usecount == 0) { 796 vprint("vrele: bad ref count", vp); 797 panic("vrele: ref cnt"); 798 } 799 #endif 800 vp->v_usecount--; 801 if (vp->v_usecount > 0) { 802 return (0); 803 } 804 805 #ifdef DIAGNOSTIC 806 if (vp->v_writecount != 0) { 807 vprint("vrele: bad writecount", vp); 808 panic("vrele: v_writecount != 0"); 809 } 810 #endif 811 812 if (vn_lock(vp, LK_EXCLUSIVE)) { 813 #ifdef DIAGNOSTIC 814 vprint("vrele: cannot lock", vp); 815 #endif 816 return (1); 817 } 818 819 VOP_INACTIVE(vp, p); 820 821 if (vp->v_usecount == 0 && !(vp->v_bioflag & VBIOONFREELIST)) 822 vputonfreelist(vp); 823 return (1); 824 } 825 826 /* Page or buffer structure gets a reference. */ 827 void 828 vhold(struct vnode *vp) 829 { 830 /* 831 * If it is on the freelist and the hold count is currently 832 * zero, move it to the hold list. 833 */ 834 if ((vp->v_bioflag & VBIOONFREELIST) && 835 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 836 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 837 TAILQ_INSERT_TAIL(&vnode_hold_list, vp, v_freelist); 838 } 839 vp->v_holdcnt++; 840 } 841 842 /* Lose interest in a vnode. */ 843 void 844 vdrop(struct vnode *vp) 845 { 846 #ifdef DIAGNOSTIC 847 if (vp->v_holdcnt == 0) 848 panic("vdrop: zero holdcnt"); 849 #endif 850 851 vp->v_holdcnt--; 852 853 /* 854 * If it is on the holdlist and the hold count drops to 855 * zero, move it to the free list. 856 */ 857 if ((vp->v_bioflag & VBIOONFREELIST) && 858 vp->v_holdcnt == 0 && vp->v_usecount == 0) { 859 TAILQ_REMOVE(&vnode_hold_list, vp, v_freelist); 860 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 861 } 862 } 863 864 /* 865 * Remove any vnodes in the vnode table belonging to mount point mp. 866 * 867 * If MNT_NOFORCE is specified, there should not be any active ones, 868 * return error if any are found (nb: this is a user error, not a 869 * system error). If MNT_FORCE is specified, detach any active vnodes 870 * that are found. 871 */ 872 #ifdef DEBUG_SYSCTL 873 int busyprt = 0; /* print out busy vnodes */ 874 struct ctldebug debug_vfs_busyprt = { "vfs_busyprt", &busyprt }; 875 #endif 876 877 int 878 vfs_mount_foreach_vnode(struct mount *mp, 879 int (*func)(struct vnode *, void *), void *arg) { 880 struct vnode *vp, *nvp; 881 int error = 0; 882 883 loop: 884 TAILQ_FOREACH_SAFE(vp , &mp->mnt_vnodelist, v_mntvnodes, nvp) { 885 if (vp->v_mount != mp) 886 goto loop; 887 888 error = func(vp, arg); 889 890 if (error != 0) 891 break; 892 } 893 894 return (error); 895 } 896 897 struct vflush_args { 898 struct vnode *skipvp; 899 int busy; 900 int flags; 901 }; 902 903 int 904 vflush_vnode(struct vnode *vp, void *arg) 905 { 906 struct vflush_args *va = arg; 907 struct proc *p = curproc; 908 909 if (vp == va->skipvp) { 910 return (0); 911 } 912 913 if ((va->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 914 return (0); 915 } 916 917 /* 918 * If WRITECLOSE is set, only flush out regular file 919 * vnodes open for writing. 920 */ 921 if ((va->flags & WRITECLOSE) && 922 (vp->v_writecount == 0 || vp->v_type != VREG)) { 923 return (0); 924 } 925 926 /* 927 * With v_usecount == 0, all we need to do is clear 928 * out the vnode data structures and we are done. 929 */ 930 if (vp->v_usecount == 0) { 931 vgonel(vp, p); 932 return (0); 933 } 934 935 /* 936 * If FORCECLOSE is set, forcibly close the vnode. 937 * For block or character devices, revert to an 938 * anonymous device. For all other files, just kill them. 939 */ 940 if (va->flags & FORCECLOSE) { 941 if (vp->v_type != VBLK && vp->v_type != VCHR) { 942 vgonel(vp, p); 943 } else { 944 vclean(vp, 0, p); 945 vp->v_op = &spec_vops; 946 insmntque(vp, NULL); 947 } 948 return (0); 949 } 950 951 /* 952 * If set, this is allowed to ignore vnodes which don't 953 * have changes pending to disk. 954 * XXX Might be nice to check per-fs "inode" flags, but 955 * generally the filesystem is sync'd already, right? 956 */ 957 if ((va->flags & IGNORECLEAN) && 958 LIST_EMPTY(&vp->v_dirtyblkhd)) 959 return (0); 960 961 #ifdef DEBUG_SYSCTL 962 if (busyprt) 963 vprint("vflush: busy vnode", vp); 964 #endif 965 va->busy++; 966 return (0); 967 } 968 969 int 970 vflush(struct mount *mp, struct vnode *skipvp, int flags) 971 { 972 struct vflush_args va; 973 va.skipvp = skipvp; 974 va.busy = 0; 975 va.flags = flags; 976 977 vfs_mount_foreach_vnode(mp, vflush_vnode, &va); 978 979 if (va.busy) 980 return (EBUSY); 981 return (0); 982 } 983 984 /* 985 * Disassociate the underlying file system from a vnode. 986 */ 987 void 988 vclean(struct vnode *vp, int flags, struct proc *p) 989 { 990 int active, do_wakeup = 0; 991 992 /* 993 * Check to see if the vnode is in use. 994 * If so we have to reference it before we clean it out 995 * so that its count cannot fall to zero and generate a 996 * race against ourselves to recycle it. 997 */ 998 if ((active = vp->v_usecount) != 0) 999 vp->v_usecount++; 1000 1001 /* 1002 * Prevent the vnode from being recycled or 1003 * brought into use while we clean it out. 1004 */ 1005 mtx_enter(&vnode_mtx); 1006 if (vp->v_lflag & VXLOCK) 1007 panic("vclean: deadlock"); 1008 vp->v_lflag |= VXLOCK; 1009 1010 if (vp->v_lockcount > 0) { 1011 /* 1012 * Ensure that any thread currently waiting on the same lock has 1013 * observed that the vnode is about to be exclusively locked 1014 * before continuing. 1015 */ 1016 msleep_nsec(&vp->v_lockcount, &vnode_mtx, PINOD, "vop_lock", 1017 INFSLP); 1018 KASSERT(vp->v_lockcount == 0); 1019 } 1020 mtx_leave(&vnode_mtx); 1021 1022 /* 1023 * Even if the count is zero, the VOP_INACTIVE routine may still 1024 * have the object locked while it cleans it out. The VOP_LOCK 1025 * ensures that the VOP_INACTIVE routine is done with its work. 1026 * For active vnodes, it ensures that no other activity can 1027 * occur while the underlying object is being cleaned out. 1028 */ 1029 VOP_LOCK(vp, LK_EXCLUSIVE | LK_DRAIN); 1030 1031 /* 1032 * Clean out any VM data associated with the vnode. 1033 */ 1034 uvm_vnp_terminate(vp); 1035 /* 1036 * Clean out any buffers associated with the vnode. 1037 */ 1038 if (flags & DOCLOSE) 1039 vinvalbuf(vp, V_SAVE, NOCRED, p, 0, INFSLP); 1040 /* 1041 * If purging an active vnode, it must be closed and 1042 * deactivated before being reclaimed. Note that the 1043 * VOP_INACTIVE will unlock the vnode 1044 */ 1045 if (active) { 1046 if (flags & DOCLOSE) 1047 VOP_CLOSE(vp, FNONBLOCK, NOCRED, p); 1048 VOP_INACTIVE(vp, p); 1049 } else { 1050 /* 1051 * Any other processes trying to obtain this lock must first 1052 * wait for VXLOCK to clear, then call the new lock operation. 1053 */ 1054 VOP_UNLOCK(vp); 1055 } 1056 1057 /* 1058 * Reclaim the vnode. 1059 */ 1060 if (VOP_RECLAIM(vp, p)) 1061 panic("vclean: cannot reclaim"); 1062 if (active) { 1063 vp->v_usecount--; 1064 if (vp->v_usecount == 0) { 1065 if (vp->v_holdcnt > 0) 1066 panic("vclean: not clean"); 1067 vputonfreelist(vp); 1068 } 1069 } 1070 cache_purge(vp); 1071 1072 /* 1073 * Done with purge, notify sleepers of the grim news. 1074 */ 1075 vp->v_op = &dead_vops; 1076 VN_KNOTE(vp, NOTE_REVOKE); 1077 vp->v_tag = VT_NON; 1078 mtx_enter(&vnode_mtx); 1079 vp->v_lflag &= ~VXLOCK; 1080 if (vp->v_lflag & VXWANT) { 1081 vp->v_lflag &= ~VXWANT; 1082 do_wakeup = 1; 1083 } 1084 mtx_leave(&vnode_mtx); 1085 if (do_wakeup) 1086 wakeup(vp); 1087 } 1088 1089 /* 1090 * Recycle an unused vnode to the front of the free list. 1091 */ 1092 int 1093 vrecycle(struct vnode *vp, struct proc *p) 1094 { 1095 if (vp->v_usecount == 0) { 1096 vgonel(vp, p); 1097 return (1); 1098 } 1099 return (0); 1100 } 1101 1102 /* 1103 * Eliminate all activity associated with a vnode 1104 * in preparation for reuse. 1105 */ 1106 void 1107 vgone(struct vnode *vp) 1108 { 1109 struct proc *p = curproc; 1110 vgonel(vp, p); 1111 } 1112 1113 /* 1114 * vgone, with struct proc. 1115 */ 1116 void 1117 vgonel(struct vnode *vp, struct proc *p) 1118 { 1119 struct vnode *vq; 1120 struct vnode *vx; 1121 1122 KASSERT(vp->v_uvcount == 0); 1123 1124 /* 1125 * If a vgone (or vclean) is already in progress, 1126 * wait until it is done and return. 1127 */ 1128 mtx_enter(&vnode_mtx); 1129 if (vp->v_lflag & VXLOCK) { 1130 vp->v_lflag |= VXWANT; 1131 msleep_nsec(vp, &vnode_mtx, PINOD, "vgone", INFSLP); 1132 mtx_leave(&vnode_mtx); 1133 return; 1134 } 1135 mtx_leave(&vnode_mtx); 1136 1137 /* 1138 * Clean out the filesystem specific data. 1139 */ 1140 vclean(vp, DOCLOSE, p); 1141 /* 1142 * Delete from old mount point vnode list, if on one. 1143 */ 1144 if (vp->v_mount != NULL) 1145 insmntque(vp, NULL); 1146 /* 1147 * If special device, remove it from special device alias list 1148 * if it is on one. 1149 */ 1150 if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_specinfo != 0) { 1151 if ((vp->v_flag & VALIASED) == 0 && vp->v_type == VCHR && 1152 (cdevsw[major(vp->v_rdev)].d_flags & D_CLONE) && 1153 (minor(vp->v_rdev) >> CLONE_SHIFT == 0)) { 1154 free(vp->v_specbitmap, M_VNODE, CLONE_MAPSZ); 1155 } 1156 SLIST_REMOVE(vp->v_hashchain, vp, vnode, v_specnext); 1157 if (vp->v_flag & VALIASED) { 1158 vx = NULL; 1159 SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) { 1160 if (vq->v_rdev != vp->v_rdev || 1161 vq->v_type != vp->v_type) 1162 continue; 1163 if (vx) 1164 break; 1165 vx = vq; 1166 } 1167 if (vx == NULL) 1168 panic("missing alias"); 1169 if (vq == NULL) 1170 vx->v_flag &= ~VALIASED; 1171 vp->v_flag &= ~VALIASED; 1172 } 1173 lf_purgelocks(&vp->v_speclockf); 1174 free(vp->v_specinfo, M_VNODE, sizeof(struct specinfo)); 1175 vp->v_specinfo = NULL; 1176 } 1177 /* 1178 * If it is on the freelist and not already at the head, 1179 * move it to the head of the list. 1180 */ 1181 vp->v_type = VBAD; 1182 1183 /* 1184 * Move onto the free list, unless we were called from 1185 * getnewvnode and we're not on any free list 1186 */ 1187 if (vp->v_usecount == 0 && 1188 (vp->v_bioflag & VBIOONFREELIST)) { 1189 int s; 1190 1191 s = splbio(); 1192 1193 if (vp->v_holdcnt > 0) 1194 panic("vgonel: not clean"); 1195 1196 if (TAILQ_FIRST(&vnode_free_list) != vp) { 1197 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 1198 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1199 } 1200 splx(s); 1201 } 1202 } 1203 1204 /* 1205 * Lookup a vnode by device number. 1206 */ 1207 int 1208 vfinddev(dev_t dev, enum vtype type, struct vnode **vpp) 1209 { 1210 struct vnode *vp; 1211 int rc =0; 1212 1213 SLIST_FOREACH(vp, &speclisth[SPECHASH(dev)], v_specnext) { 1214 if (dev != vp->v_rdev || type != vp->v_type) 1215 continue; 1216 *vpp = vp; 1217 rc = 1; 1218 break; 1219 } 1220 return (rc); 1221 } 1222 1223 /* 1224 * Revoke all the vnodes corresponding to the specified minor number 1225 * range (endpoints inclusive) of the specified major. 1226 */ 1227 void 1228 vdevgone(int maj, int minl, int minh, enum vtype type) 1229 { 1230 struct vnode *vp; 1231 int mn; 1232 1233 for (mn = minl; mn <= minh; mn++) 1234 if (vfinddev(makedev(maj, mn), type, &vp)) 1235 VOP_REVOKE(vp, REVOKEALL); 1236 } 1237 1238 /* 1239 * Calculate the total number of references to a special device. 1240 */ 1241 int 1242 vcount(struct vnode *vp) 1243 { 1244 struct vnode *vq; 1245 int count; 1246 1247 loop: 1248 if ((vp->v_flag & VALIASED) == 0) 1249 return (vp->v_usecount); 1250 count = 0; 1251 SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) { 1252 if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) 1253 continue; 1254 /* 1255 * Alias, but not in use, so flush it out. 1256 */ 1257 if (vq->v_usecount == 0 && vq != vp) { 1258 vgone(vq); 1259 goto loop; 1260 } 1261 count += vq->v_usecount; 1262 } 1263 return (count); 1264 } 1265 1266 #if defined(DEBUG) || defined(DIAGNOSTIC) 1267 /* 1268 * Print out a description of a vnode. 1269 */ 1270 static char *typename[] = 1271 { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" }; 1272 1273 void 1274 vprint(char *label, struct vnode *vp) 1275 { 1276 char buf[64]; 1277 1278 if (label != NULL) 1279 printf("%s: ", label); 1280 printf("%p, type %s, use %u, write %u, hold %u,", 1281 vp, typename[vp->v_type], vp->v_usecount, vp->v_writecount, 1282 vp->v_holdcnt); 1283 buf[0] = '\0'; 1284 if (vp->v_flag & VROOT) 1285 strlcat(buf, "|VROOT", sizeof buf); 1286 if (vp->v_flag & VTEXT) 1287 strlcat(buf, "|VTEXT", sizeof buf); 1288 if (vp->v_flag & VSYSTEM) 1289 strlcat(buf, "|VSYSTEM", sizeof buf); 1290 if (vp->v_lflag & VXLOCK) 1291 strlcat(buf, "|VXLOCK", sizeof buf); 1292 if (vp->v_lflag & VXWANT) 1293 strlcat(buf, "|VXWANT", sizeof buf); 1294 if (vp->v_bioflag & VBIOWAIT) 1295 strlcat(buf, "|VBIOWAIT", sizeof buf); 1296 if (vp->v_bioflag & VBIOONFREELIST) 1297 strlcat(buf, "|VBIOONFREELIST", sizeof buf); 1298 if (vp->v_bioflag & VBIOONSYNCLIST) 1299 strlcat(buf, "|VBIOONSYNCLIST", sizeof buf); 1300 if (vp->v_flag & VALIASED) 1301 strlcat(buf, "|VALIASED", sizeof buf); 1302 if (buf[0] != '\0') 1303 printf(" flags (%s)", &buf[1]); 1304 if (vp->v_data == NULL) { 1305 printf("\n"); 1306 } else { 1307 printf("\n\t"); 1308 VOP_PRINT(vp); 1309 } 1310 } 1311 #endif /* DEBUG || DIAGNOSTIC */ 1312 1313 #ifdef DEBUG 1314 /* 1315 * List all of the locked vnodes in the system. 1316 * Called when debugging the kernel. 1317 */ 1318 void 1319 printlockedvnodes(void) 1320 { 1321 struct mount *mp; 1322 struct vnode *vp; 1323 1324 printf("Locked vnodes\n"); 1325 1326 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1327 if (vfs_busy(mp, VB_READ|VB_NOWAIT)) 1328 continue; 1329 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 1330 if (VOP_ISLOCKED(vp)) 1331 vprint(NULL, vp); 1332 } 1333 vfs_unbusy(mp); 1334 } 1335 1336 } 1337 #endif 1338 1339 /* 1340 * Top level filesystem related information gathering. 1341 */ 1342 int 1343 vfs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp, 1344 size_t newlen, struct proc *p) 1345 { 1346 struct vfsconf *vfsp, *tmpvfsp; 1347 int ret; 1348 1349 /* all sysctl names at this level are at least name and field */ 1350 if (namelen < 2) 1351 return (ENOTDIR); /* overloaded */ 1352 1353 if (name[0] != VFS_GENERIC) { 1354 vfsp = vfs_bytypenum(name[0]); 1355 if (vfsp == NULL || vfsp->vfc_vfsops->vfs_sysctl == NULL) 1356 return (EOPNOTSUPP); 1357 1358 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, 1359 oldp, oldlenp, newp, newlen, p)); 1360 } 1361 1362 switch (name[1]) { 1363 case VFS_MAXTYPENUM: 1364 return (sysctl_rdint(oldp, oldlenp, newp, maxvfsconf)); 1365 1366 case VFS_CONF: 1367 if (namelen < 3) 1368 return (ENOTDIR); /* overloaded */ 1369 1370 vfsp = vfs_bytypenum(name[2]); 1371 if (vfsp == NULL) 1372 return (EOPNOTSUPP); 1373 1374 /* Make a copy, clear out kernel pointers */ 1375 tmpvfsp = malloc(sizeof(*tmpvfsp), M_TEMP, M_WAITOK|M_ZERO); 1376 memcpy(tmpvfsp, vfsp, sizeof(*tmpvfsp)); 1377 tmpvfsp->vfc_vfsops = NULL; 1378 1379 ret = sysctl_rdstruct(oldp, oldlenp, newp, tmpvfsp, 1380 sizeof(struct vfsconf)); 1381 1382 free(tmpvfsp, M_TEMP, sizeof(*tmpvfsp)); 1383 return (ret); 1384 case VFS_BCACHESTAT: /* buffer cache statistics */ 1385 ret = sysctl_rdstruct(oldp, oldlenp, newp, &bcstats, 1386 sizeof(struct bcachestats)); 1387 return(ret); 1388 } 1389 return (EOPNOTSUPP); 1390 } 1391 1392 /* 1393 * Check to see if a filesystem is mounted on a block device. 1394 */ 1395 int 1396 vfs_mountedon(struct vnode *vp) 1397 { 1398 struct vnode *vq; 1399 int error = 0; 1400 1401 if (vp->v_specmountpoint != NULL) 1402 return (EBUSY); 1403 if (vp->v_flag & VALIASED) { 1404 SLIST_FOREACH(vq, vp->v_hashchain, v_specnext) { 1405 if (vq->v_rdev != vp->v_rdev || 1406 vq->v_type != vp->v_type) 1407 continue; 1408 if (vq->v_specmountpoint != NULL) { 1409 error = EBUSY; 1410 break; 1411 } 1412 } 1413 } 1414 return (error); 1415 } 1416 1417 #ifdef NFSSERVER 1418 /* 1419 * Build hash lists of net addresses and hang them off the mount point. 1420 * Called by vfs_export() to set up the lists of export addresses. 1421 */ 1422 int 1423 vfs_hang_addrlist(struct mount *mp, struct netexport *nep, 1424 struct export_args *argp) 1425 { 1426 struct netcred *np; 1427 struct radix_node_head *rnh; 1428 int nplen, i; 1429 struct radix_node *rn; 1430 struct sockaddr *saddr, *smask = 0; 1431 int error; 1432 1433 if (argp->ex_addrlen == 0) { 1434 if (mp->mnt_flag & MNT_DEFEXPORTED) 1435 return (EPERM); 1436 np = &nep->ne_defexported; 1437 /* fill in the kernel's ucred from userspace's xucred */ 1438 if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon))) 1439 return (error); 1440 mp->mnt_flag |= MNT_DEFEXPORTED; 1441 goto finish; 1442 } 1443 if (argp->ex_addrlen > MLEN || argp->ex_masklen > MLEN || 1444 argp->ex_addrlen < 0 || argp->ex_masklen < 0) 1445 return (EINVAL); 1446 nplen = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen; 1447 np = (struct netcred *)malloc(nplen, M_NETADDR, M_WAITOK|M_ZERO); 1448 np->netc_len = nplen; 1449 saddr = (struct sockaddr *)(np + 1); 1450 error = copyin(argp->ex_addr, saddr, argp->ex_addrlen); 1451 if (error) 1452 goto out; 1453 if (saddr->sa_len > argp->ex_addrlen) 1454 saddr->sa_len = argp->ex_addrlen; 1455 if (argp->ex_masklen) { 1456 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen); 1457 error = copyin(argp->ex_mask, smask, argp->ex_masklen); 1458 if (error) 1459 goto out; 1460 if (smask->sa_len > argp->ex_masklen) 1461 smask->sa_len = argp->ex_masklen; 1462 } 1463 /* fill in the kernel's ucred from userspace's xucred */ 1464 if ((error = crfromxucred(&np->netc_anon, &argp->ex_anon))) 1465 goto out; 1466 i = saddr->sa_family; 1467 switch (i) { 1468 case AF_INET: 1469 if ((rnh = nep->ne_rtable_inet) == NULL) { 1470 if (!rn_inithead((void **)&nep->ne_rtable_inet, 1471 offsetof(struct sockaddr_in, sin_addr))) { 1472 error = ENOBUFS; 1473 goto out; 1474 } 1475 rnh = nep->ne_rtable_inet; 1476 } 1477 break; 1478 default: 1479 error = EINVAL; 1480 goto out; 1481 } 1482 rn = rn_addroute(saddr, smask, rnh, np->netc_rnodes, 0); 1483 if (rn == 0 || np != (struct netcred *)rn) { /* already exists */ 1484 error = EPERM; 1485 goto out; 1486 } 1487 finish: 1488 np->netc_exflags = argp->ex_flags; 1489 return (0); 1490 out: 1491 free(np, M_NETADDR, np->netc_len); 1492 return (error); 1493 } 1494 1495 int 1496 vfs_free_netcred(struct radix_node *rn, void *w, u_int id) 1497 { 1498 struct radix_node_head *rnh = (struct radix_node_head *)w; 1499 struct netcred * np = (struct netcred *)rn; 1500 1501 rn_delete(rn->rn_key, rn->rn_mask, rnh, NULL); 1502 free(np, M_NETADDR, np->netc_len); 1503 return (0); 1504 } 1505 1506 /* 1507 * Free the net address hash lists that are hanging off the mount points. 1508 */ 1509 void 1510 vfs_free_addrlist(struct netexport *nep) 1511 { 1512 struct radix_node_head *rnh; 1513 1514 if ((rnh = nep->ne_rtable_inet) != NULL) { 1515 rn_walktree(rnh, vfs_free_netcred, rnh); 1516 free(rnh, M_RTABLE, sizeof(*rnh)); 1517 nep->ne_rtable_inet = NULL; 1518 } 1519 } 1520 #endif /* NFSSERVER */ 1521 1522 int 1523 vfs_export(struct mount *mp, struct netexport *nep, struct export_args *argp) 1524 { 1525 #ifdef NFSSERVER 1526 int error; 1527 1528 if (argp->ex_flags & MNT_DELEXPORT) { 1529 vfs_free_addrlist(nep); 1530 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); 1531 } 1532 if (argp->ex_flags & MNT_EXPORTED) { 1533 if ((error = vfs_hang_addrlist(mp, nep, argp)) != 0) 1534 return (error); 1535 mp->mnt_flag |= MNT_EXPORTED; 1536 } 1537 return (0); 1538 #else 1539 return (ENOTSUP); 1540 #endif /* NFSSERVER */ 1541 } 1542 1543 struct netcred * 1544 vfs_export_lookup(struct mount *mp, struct netexport *nep, struct mbuf *nam) 1545 { 1546 #ifdef NFSSERVER 1547 struct netcred *np; 1548 struct radix_node_head *rnh; 1549 struct sockaddr *saddr; 1550 1551 np = NULL; 1552 if (mp->mnt_flag & MNT_EXPORTED) { 1553 /* 1554 * Lookup in the export list first. 1555 */ 1556 if (nam != NULL) { 1557 saddr = mtod(nam, struct sockaddr *); 1558 switch(saddr->sa_family) { 1559 case AF_INET: 1560 rnh = nep->ne_rtable_inet; 1561 break; 1562 default: 1563 rnh = NULL; 1564 break; 1565 } 1566 if (rnh != NULL) 1567 np = (struct netcred *)rn_match(saddr, rnh); 1568 } 1569 /* 1570 * If no address match, use the default if it exists. 1571 */ 1572 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED) 1573 np = &nep->ne_defexported; 1574 } 1575 return (np); 1576 #else 1577 return (NULL); 1578 #endif /* NFSSERVER */ 1579 } 1580 1581 /* 1582 * Do the usual access checking. 1583 * file_mode, uid and gid are from the vnode in question, 1584 * while acc_mode and cred are from the VOP_ACCESS parameter list 1585 */ 1586 int 1587 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1588 mode_t acc_mode, struct ucred *cred) 1589 { 1590 mode_t mask; 1591 1592 /* User id 0 always gets read/write access. */ 1593 if (cred->cr_uid == 0) { 1594 /* For VEXEC, at least one of the execute bits must be set. */ 1595 if ((acc_mode & VEXEC) && type != VDIR && 1596 (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0) 1597 return EACCES; 1598 return 0; 1599 } 1600 1601 mask = 0; 1602 1603 /* Otherwise, check the owner. */ 1604 if (cred->cr_uid == uid) { 1605 if (acc_mode & VEXEC) 1606 mask |= S_IXUSR; 1607 if (acc_mode & VREAD) 1608 mask |= S_IRUSR; 1609 if (acc_mode & VWRITE) 1610 mask |= S_IWUSR; 1611 return (file_mode & mask) == mask ? 0 : EACCES; 1612 } 1613 1614 /* Otherwise, check the groups. */ 1615 if (groupmember(gid, cred)) { 1616 if (acc_mode & VEXEC) 1617 mask |= S_IXGRP; 1618 if (acc_mode & VREAD) 1619 mask |= S_IRGRP; 1620 if (acc_mode & VWRITE) 1621 mask |= S_IWGRP; 1622 return (file_mode & mask) == mask ? 0 : EACCES; 1623 } 1624 1625 /* Otherwise, check everyone else. */ 1626 if (acc_mode & VEXEC) 1627 mask |= S_IXOTH; 1628 if (acc_mode & VREAD) 1629 mask |= S_IROTH; 1630 if (acc_mode & VWRITE) 1631 mask |= S_IWOTH; 1632 return (file_mode & mask) == mask ? 0 : EACCES; 1633 } 1634 1635 int 1636 vnoperm(struct vnode *vp) 1637 { 1638 if (vp->v_flag & VROOT || vp->v_mount == NULL) 1639 return 0; 1640 1641 return (vp->v_mount->mnt_flag & MNT_NOPERM); 1642 } 1643 1644 struct rwlock vfs_stall_lock = RWLOCK_INITIALIZER("vfs_stall"); 1645 unsigned int vfs_stalling = 0; 1646 1647 int 1648 vfs_stall(struct proc *p, int stall) 1649 { 1650 struct mount *mp; 1651 int allerror = 0, error; 1652 1653 if (stall) { 1654 atomic_inc_int(&vfs_stalling); 1655 rw_enter_write(&vfs_stall_lock); 1656 } 1657 1658 /* 1659 * The loop variable mp is protected by vfs_busy() so that it cannot 1660 * be unmounted while VFS_SYNC() sleeps. Traverse forward to keep the 1661 * lock order consistent with dounmount(). 1662 */ 1663 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1664 if (stall) { 1665 error = vfs_busy(mp, VB_WRITE|VB_WAIT|VB_DUPOK); 1666 if (error) { 1667 printf("%s: busy\n", mp->mnt_stat.f_mntonname); 1668 allerror = error; 1669 continue; 1670 } 1671 uvm_vnp_sync(mp); 1672 error = VFS_SYNC(mp, MNT_WAIT, stall, p->p_ucred, p); 1673 if (error) { 1674 printf("%s: failed to sync\n", 1675 mp->mnt_stat.f_mntonname); 1676 vfs_unbusy(mp); 1677 allerror = error; 1678 continue; 1679 } 1680 mp->mnt_flag |= MNT_STALLED; 1681 } else { 1682 if (mp->mnt_flag & MNT_STALLED) { 1683 vfs_unbusy(mp); 1684 mp->mnt_flag &= ~MNT_STALLED; 1685 } 1686 } 1687 } 1688 1689 if (!stall) { 1690 rw_exit_write(&vfs_stall_lock); 1691 atomic_dec_int(&vfs_stalling); 1692 } 1693 1694 return (allerror); 1695 } 1696 1697 void 1698 vfs_stall_barrier(void) 1699 { 1700 if (__predict_false(vfs_stalling)) { 1701 rw_enter_read(&vfs_stall_lock); 1702 rw_exit_read(&vfs_stall_lock); 1703 } 1704 } 1705 1706 /* 1707 * Unmount all file systems. 1708 * We traverse the list in reverse order under the assumption that doing so 1709 * will avoid needing to worry about dependencies. 1710 */ 1711 void 1712 vfs_unmountall(void) 1713 { 1714 struct mount *mp, *nmp; 1715 int allerror, error, again = 1; 1716 1717 retry: 1718 allerror = 0; 1719 TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, nmp) { 1720 if (vfs_busy(mp, VB_WRITE|VB_NOWAIT)) 1721 continue; 1722 /* XXX Here is a race, the next pointer is not locked. */ 1723 if ((error = dounmount(mp, MNT_FORCE, curproc)) != 0) { 1724 printf("unmount of %s failed with error %d\n", 1725 mp->mnt_stat.f_mntonname, error); 1726 allerror = 1; 1727 } 1728 } 1729 1730 if (allerror) { 1731 printf("WARNING: some file systems would not unmount\n"); 1732 if (again) { 1733 printf("retrying\n"); 1734 again = 0; 1735 goto retry; 1736 } 1737 } 1738 } 1739 1740 /* 1741 * Sync and unmount file systems before shutting down. 1742 */ 1743 void 1744 vfs_shutdown(struct proc *p) 1745 { 1746 #ifdef ACCOUNTING 1747 acct_shutdown(); 1748 #endif 1749 1750 printf("syncing disks..."); 1751 1752 if (panicstr == 0) { 1753 /* Sync before unmount, in case we hang on something. */ 1754 sys_sync(p, NULL, NULL); 1755 vfs_unmountall(); 1756 } 1757 1758 #if NSOFTRAID > 0 1759 sr_quiesce(); 1760 #endif 1761 1762 if (vfs_syncwait(p, 1)) 1763 printf(" giving up\n"); 1764 else 1765 printf(" done\n"); 1766 } 1767 1768 /* 1769 * perform sync() operation and wait for buffers to flush. 1770 */ 1771 int 1772 vfs_syncwait(struct proc *p, int verbose) 1773 { 1774 struct buf *bp; 1775 int iter, nbusy, dcount, s; 1776 #ifdef MULTIPROCESSOR 1777 int hold_count; 1778 #endif 1779 1780 sys_sync(p, NULL, NULL); 1781 1782 /* Wait for sync to finish. */ 1783 dcount = 10000; 1784 for (iter = 0; iter < 20; iter++) { 1785 nbusy = 0; 1786 LIST_FOREACH(bp, &bufhead, b_list) { 1787 if ((bp->b_flags & (B_BUSY|B_INVAL|B_READ)) == B_BUSY) 1788 nbusy++; 1789 /* 1790 * With soft updates, some buffers that are 1791 * written will be remarked as dirty until other 1792 * buffers are written. 1793 */ 1794 if (bp->b_flags & B_DELWRI) { 1795 s = splbio(); 1796 bremfree(bp); 1797 buf_acquire(bp); 1798 splx(s); 1799 nbusy++; 1800 bawrite(bp); 1801 if (dcount-- <= 0) { 1802 if (verbose) 1803 printf("softdep "); 1804 return 1; 1805 } 1806 } 1807 } 1808 if (nbusy == 0) 1809 break; 1810 if (verbose) 1811 printf("%d ", nbusy); 1812 #ifdef MULTIPROCESSOR 1813 if (_kernel_lock_held()) 1814 hold_count = __mp_release_all(&kernel_lock); 1815 else 1816 hold_count = 0; 1817 #endif 1818 DELAY(40000 * iter); 1819 #ifdef MULTIPROCESSOR 1820 if (hold_count) 1821 __mp_acquire_count(&kernel_lock, hold_count); 1822 #endif 1823 } 1824 1825 return nbusy; 1826 } 1827 1828 /* 1829 * posix file system related system variables. 1830 */ 1831 int 1832 fs_posix_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, 1833 void *newp, size_t newlen, struct proc *p) 1834 { 1835 /* all sysctl names at this level are terminal */ 1836 if (namelen != 1) 1837 return (ENOTDIR); 1838 1839 switch (name[0]) { 1840 case FS_POSIX_SETUID: 1841 if (newp && securelevel > 0) 1842 return (EPERM); 1843 return(sysctl_int(oldp, oldlenp, newp, newlen, &suid_clear)); 1844 default: 1845 return (EOPNOTSUPP); 1846 } 1847 /* NOTREACHED */ 1848 } 1849 1850 /* 1851 * file system related system variables. 1852 */ 1853 int 1854 fs_sysctl(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp, 1855 size_t newlen, struct proc *p) 1856 { 1857 sysctlfn *fn; 1858 1859 switch (name[0]) { 1860 case FS_POSIX: 1861 fn = fs_posix_sysctl; 1862 break; 1863 default: 1864 return (EOPNOTSUPP); 1865 } 1866 return (*fn)(name + 1, namelen - 1, oldp, oldlenp, newp, newlen, p); 1867 } 1868 1869 1870 /* 1871 * Routines dealing with vnodes and buffers 1872 */ 1873 1874 /* 1875 * Wait for all outstanding I/Os to complete 1876 * 1877 * Manipulates v_numoutput. Must be called at splbio() 1878 */ 1879 int 1880 vwaitforio(struct vnode *vp, int slpflag, char *wmesg, uint64_t timeo) 1881 { 1882 int error = 0; 1883 1884 splassert(IPL_BIO); 1885 1886 while (vp->v_numoutput) { 1887 vp->v_bioflag |= VBIOWAIT; 1888 error = tsleep_nsec(&vp->v_numoutput, 1889 slpflag | (PRIBIO + 1), wmesg, timeo); 1890 if (error) 1891 break; 1892 } 1893 1894 return (error); 1895 } 1896 1897 /* 1898 * Update outstanding I/O count and do wakeup if requested. 1899 * 1900 * Manipulates v_numoutput. Must be called at splbio() 1901 */ 1902 void 1903 vwakeup(struct vnode *vp) 1904 { 1905 splassert(IPL_BIO); 1906 1907 if (vp != NULL) { 1908 if (vp->v_numoutput-- == 0) 1909 panic("vwakeup: neg numoutput"); 1910 if ((vp->v_bioflag & VBIOWAIT) && vp->v_numoutput == 0) { 1911 vp->v_bioflag &= ~VBIOWAIT; 1912 wakeup(&vp->v_numoutput); 1913 } 1914 } 1915 } 1916 1917 /* 1918 * Flush out and invalidate all buffers associated with a vnode. 1919 * Called with the underlying object locked. 1920 */ 1921 int 1922 vinvalbuf(struct vnode *vp, int flags, struct ucred *cred, struct proc *p, 1923 int slpflag, uint64_t slptimeo) 1924 { 1925 struct buf *bp; 1926 struct buf *nbp, *blist; 1927 int s, error; 1928 1929 #ifdef VFSLCKDEBUG 1930 if ((vp->v_op->vop_islocked != nullop) && !VOP_ISLOCKED(vp)) 1931 panic("%s: vp isn't locked, vp %p", __func__, vp); 1932 #endif 1933 1934 if (flags & V_SAVE) { 1935 s = splbio(); 1936 vwaitforio(vp, 0, "vinvalbuf", INFSLP); 1937 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) { 1938 splx(s); 1939 if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, p)) != 0) 1940 return (error); 1941 s = splbio(); 1942 if (vp->v_numoutput > 0 || 1943 !LIST_EMPTY(&vp->v_dirtyblkhd)) 1944 panic("%s: dirty bufs, vp %p", __func__, vp); 1945 } 1946 splx(s); 1947 } 1948 loop: 1949 s = splbio(); 1950 for (;;) { 1951 int count = 0; 1952 if ((blist = LIST_FIRST(&vp->v_cleanblkhd)) && 1953 (flags & V_SAVEMETA)) 1954 while (blist && blist->b_lblkno < 0) 1955 blist = LIST_NEXT(blist, b_vnbufs); 1956 if (blist == NULL && 1957 (blist = LIST_FIRST(&vp->v_dirtyblkhd)) && 1958 (flags & V_SAVEMETA)) 1959 while (blist && blist->b_lblkno < 0) 1960 blist = LIST_NEXT(blist, b_vnbufs); 1961 if (!blist) 1962 break; 1963 1964 for (bp = blist; bp; bp = nbp) { 1965 nbp = LIST_NEXT(bp, b_vnbufs); 1966 if (flags & V_SAVEMETA && bp->b_lblkno < 0) 1967 continue; 1968 if (bp->b_flags & B_BUSY) { 1969 bp->b_flags |= B_WANTED; 1970 error = tsleep_nsec(bp, slpflag | (PRIBIO + 1), 1971 "vinvalbuf", slptimeo); 1972 if (error) { 1973 splx(s); 1974 return (error); 1975 } 1976 break; 1977 } 1978 bremfree(bp); 1979 /* 1980 * XXX Since there are no node locks for NFS, I believe 1981 * there is a slight chance that a delayed write will 1982 * occur while sleeping just above, so check for it. 1983 */ 1984 if ((bp->b_flags & B_DELWRI) && (flags & V_SAVE)) { 1985 buf_acquire(bp); 1986 splx(s); 1987 (void) VOP_BWRITE(bp); 1988 goto loop; 1989 } 1990 buf_acquire_nomap(bp); 1991 bp->b_flags |= B_INVAL; 1992 brelse(bp); 1993 count++; 1994 /* 1995 * XXX Temporary workaround XXX 1996 * 1997 * If this is a gigantisch vnode and we are 1998 * trashing a ton of buffers, drop the lock 1999 * and yield every so often. The longer term 2000 * fix is to add a separate list for these 2001 * invalid buffers so we don't have to do the 2002 * work to free these here. 2003 */ 2004 if (count > 100) { 2005 splx(s); 2006 sched_pause(yield); 2007 goto loop; 2008 } 2009 } 2010 } 2011 if (!(flags & V_SAVEMETA) && 2012 (!LIST_EMPTY(&vp->v_dirtyblkhd) || !LIST_EMPTY(&vp->v_cleanblkhd))) 2013 panic("%s: flush failed, vp %p", __func__, vp); 2014 splx(s); 2015 return (0); 2016 } 2017 2018 void 2019 vflushbuf(struct vnode *vp, int sync) 2020 { 2021 struct buf *bp, *nbp; 2022 int s; 2023 2024 loop: 2025 s = splbio(); 2026 LIST_FOREACH_SAFE(bp, &vp->v_dirtyblkhd, b_vnbufs, nbp) { 2027 if ((bp->b_flags & B_BUSY)) 2028 continue; 2029 if ((bp->b_flags & B_DELWRI) == 0) 2030 panic("vflushbuf: not dirty"); 2031 bremfree(bp); 2032 buf_acquire(bp); 2033 splx(s); 2034 /* 2035 * Wait for I/O associated with indirect blocks to complete, 2036 * since there is no way to quickly wait for them below. 2037 */ 2038 if (bp->b_vp == vp || sync == 0) 2039 (void) bawrite(bp); 2040 else 2041 (void) bwrite(bp); 2042 goto loop; 2043 } 2044 if (sync == 0) { 2045 splx(s); 2046 return; 2047 } 2048 vwaitforio(vp, 0, "vflushbuf", INFSLP); 2049 if (!LIST_EMPTY(&vp->v_dirtyblkhd)) { 2050 splx(s); 2051 #ifdef DIAGNOSTIC 2052 vprint("vflushbuf: dirty", vp); 2053 #endif 2054 goto loop; 2055 } 2056 splx(s); 2057 } 2058 2059 /* 2060 * Associate a buffer with a vnode. 2061 * 2062 * Manipulates buffer vnode queues. Must be called at splbio(). 2063 */ 2064 void 2065 bgetvp(struct vnode *vp, struct buf *bp) 2066 { 2067 splassert(IPL_BIO); 2068 2069 2070 if (bp->b_vp) 2071 panic("bgetvp: not free"); 2072 vhold(vp); 2073 bp->b_vp = vp; 2074 if (vp->v_type == VBLK || vp->v_type == VCHR) 2075 bp->b_dev = vp->v_rdev; 2076 else 2077 bp->b_dev = NODEV; 2078 /* 2079 * Insert onto list for new vnode. 2080 */ 2081 bufinsvn(bp, &vp->v_cleanblkhd); 2082 } 2083 2084 /* 2085 * Disassociate a buffer from a vnode. 2086 * 2087 * Manipulates vnode buffer queues. Must be called at splbio(). 2088 */ 2089 void 2090 brelvp(struct buf *bp) 2091 { 2092 struct vnode *vp; 2093 2094 splassert(IPL_BIO); 2095 2096 if ((vp = bp->b_vp) == (struct vnode *) 0) 2097 panic("brelvp: NULL"); 2098 /* 2099 * Delete from old vnode list, if on one. 2100 */ 2101 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 2102 bufremvn(bp); 2103 if ((vp->v_bioflag & VBIOONSYNCLIST) && 2104 LIST_EMPTY(&vp->v_dirtyblkhd)) { 2105 vp->v_bioflag &= ~VBIOONSYNCLIST; 2106 LIST_REMOVE(vp, v_synclist); 2107 } 2108 bp->b_vp = NULL; 2109 2110 vdrop(vp); 2111 } 2112 2113 /* 2114 * Replaces the current vnode associated with the buffer, if any, 2115 * with a new vnode. 2116 * 2117 * If an output I/O is pending on the buffer, the old vnode 2118 * I/O count is adjusted. 2119 * 2120 * Ignores vnode buffer queues. Must be called at splbio(). 2121 */ 2122 void 2123 buf_replacevnode(struct buf *bp, struct vnode *newvp) 2124 { 2125 struct vnode *oldvp = bp->b_vp; 2126 2127 splassert(IPL_BIO); 2128 2129 if (oldvp) 2130 brelvp(bp); 2131 2132 if ((bp->b_flags & (B_READ | B_DONE)) == 0) { 2133 newvp->v_numoutput++; /* put it on swapdev */ 2134 vwakeup(oldvp); 2135 } 2136 2137 bgetvp(newvp, bp); 2138 bufremvn(bp); 2139 } 2140 2141 /* 2142 * Used to assign buffers to the appropriate clean or dirty list on 2143 * the vnode and to add newly dirty vnodes to the appropriate 2144 * filesystem syncer list. 2145 * 2146 * Manipulates vnode buffer queues. Must be called at splbio(). 2147 */ 2148 void 2149 reassignbuf(struct buf *bp) 2150 { 2151 struct buflists *listheadp; 2152 int delay; 2153 struct vnode *vp = bp->b_vp; 2154 2155 splassert(IPL_BIO); 2156 2157 /* 2158 * Delete from old vnode list, if on one. 2159 */ 2160 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 2161 bufremvn(bp); 2162 2163 /* 2164 * If dirty, put on list of dirty buffers; 2165 * otherwise insert onto list of clean buffers. 2166 */ 2167 if ((bp->b_flags & B_DELWRI) == 0) { 2168 listheadp = &vp->v_cleanblkhd; 2169 if ((vp->v_bioflag & VBIOONSYNCLIST) && 2170 LIST_EMPTY(&vp->v_dirtyblkhd)) { 2171 vp->v_bioflag &= ~VBIOONSYNCLIST; 2172 LIST_REMOVE(vp, v_synclist); 2173 } 2174 } else { 2175 listheadp = &vp->v_dirtyblkhd; 2176 if ((vp->v_bioflag & VBIOONSYNCLIST) == 0) { 2177 switch (vp->v_type) { 2178 case VDIR: 2179 delay = syncdelay / 2; 2180 break; 2181 case VBLK: 2182 if (vp->v_specmountpoint != NULL) { 2183 delay = syncdelay / 3; 2184 break; 2185 } 2186 /* FALLTHROUGH */ 2187 default: 2188 delay = syncdelay; 2189 } 2190 vn_syncer_add_to_worklist(vp, delay); 2191 } 2192 } 2193 bufinsvn(bp, listheadp); 2194 } 2195 2196 /* 2197 * Check if vnode represents a disk device 2198 */ 2199 int 2200 vn_isdisk(struct vnode *vp, int *errp) 2201 { 2202 if (vp->v_type != VBLK && vp->v_type != VCHR) 2203 return (0); 2204 2205 return (1); 2206 } 2207 2208 #ifdef DDB 2209 #include <machine/db_machdep.h> 2210 #include <ddb/db_interface.h> 2211 2212 void 2213 vfs_buf_print(void *b, int full, 2214 int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2)))) 2215 { 2216 struct buf *bp = b; 2217 2218 (*pr)(" vp %p lblkno 0x%llx blkno 0x%llx dev 0x%x\n" 2219 " proc %p error %d flags %lb\n", 2220 bp->b_vp, (int64_t)bp->b_lblkno, (int64_t)bp->b_blkno, bp->b_dev, 2221 bp->b_proc, bp->b_error, bp->b_flags, B_BITS); 2222 2223 (*pr)(" bufsize 0x%lx bcount 0x%lx resid 0x%lx\n" 2224 " data %p saveaddr %p dep %p iodone %p\n", 2225 bp->b_bufsize, bp->b_bcount, (long)bp->b_resid, 2226 bp->b_data, bp->b_saveaddr, 2227 LIST_FIRST(&bp->b_dep), bp->b_iodone); 2228 2229 (*pr)(" dirty {off 0x%x end 0x%x} valid {off 0x%x end 0x%x}\n", 2230 bp->b_dirtyoff, bp->b_dirtyend, bp->b_validoff, bp->b_validend); 2231 2232 #ifdef FFS_SOFTUPDATES 2233 if (full) 2234 softdep_print(bp, full, pr); 2235 #endif 2236 } 2237 2238 const char *vtypes[] = { VTYPE_NAMES }; 2239 const char *vtags[] = { VTAG_NAMES }; 2240 2241 void 2242 vfs_vnode_print(void *v, int full, 2243 int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2)))) 2244 { 2245 struct vnode *vp = v; 2246 2247 (*pr)("tag %s(%d) type %s(%d) mount %p typedata %p\n", 2248 (u_int)vp->v_tag >= nitems(vtags)? "<unk>":vtags[vp->v_tag], 2249 vp->v_tag, 2250 (u_int)vp->v_type >= nitems(vtypes)? "<unk>":vtypes[vp->v_type], 2251 vp->v_type, vp->v_mount, vp->v_mountedhere); 2252 2253 (*pr)("data %p usecount %d writecount %d holdcnt %d numoutput %d\n", 2254 vp->v_data, vp->v_usecount, vp->v_writecount, 2255 vp->v_holdcnt, vp->v_numoutput); 2256 2257 /* uvm_object_printit(&vp->v_uobj, full, pr); */ 2258 2259 if (full) { 2260 struct buf *bp; 2261 2262 (*pr)("clean bufs:\n"); 2263 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) { 2264 (*pr)(" bp %p\n", bp); 2265 vfs_buf_print(bp, full, pr); 2266 } 2267 2268 (*pr)("dirty bufs:\n"); 2269 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 2270 (*pr)(" bp %p\n", bp); 2271 vfs_buf_print(bp, full, pr); 2272 } 2273 } 2274 } 2275 2276 void 2277 vfs_mount_print(struct mount *mp, int full, 2278 int (*pr)(const char *, ...) __attribute__((__format__(__kprintf__,1,2)))) 2279 { 2280 struct vfsconf *vfc = mp->mnt_vfc; 2281 struct vnode *vp; 2282 int cnt; 2283 2284 (*pr)("flags %b\nvnodecovered %p syncer %p data %p\n", 2285 mp->mnt_flag, MNT_BITS, 2286 mp->mnt_vnodecovered, mp->mnt_syncer, mp->mnt_data); 2287 2288 (*pr)("vfsconf: ops %p name \"%s\" num %d ref %u flags 0x%x\n", 2289 vfc->vfc_vfsops, vfc->vfc_name, vfc->vfc_typenum, 2290 vfc->vfc_refcount, vfc->vfc_flags); 2291 2292 (*pr)("statvfs cache: bsize %x iosize %x\n" 2293 "blocks %llu free %llu avail %lld\n", 2294 mp->mnt_stat.f_bsize, mp->mnt_stat.f_iosize, mp->mnt_stat.f_blocks, 2295 mp->mnt_stat.f_bfree, mp->mnt_stat.f_bavail); 2296 2297 (*pr)(" files %llu ffiles %llu favail %lld\n", mp->mnt_stat.f_files, 2298 mp->mnt_stat.f_ffree, mp->mnt_stat.f_favail); 2299 2300 (*pr)(" f_fsidx {0x%x, 0x%x} owner %u ctime 0x%llx\n", 2301 mp->mnt_stat.f_fsid.val[0], mp->mnt_stat.f_fsid.val[1], 2302 mp->mnt_stat.f_owner, mp->mnt_stat.f_ctime); 2303 2304 (*pr)(" syncwrites %llu asyncwrites = %llu\n", 2305 mp->mnt_stat.f_syncwrites, mp->mnt_stat.f_asyncwrites); 2306 2307 (*pr)(" syncreads %llu asyncreads = %llu\n", 2308 mp->mnt_stat.f_syncreads, mp->mnt_stat.f_asyncreads); 2309 2310 (*pr)(" fstype \"%s\" mnton \"%s\" mntfrom \"%s\" mntspec \"%s\"\n", 2311 mp->mnt_stat.f_fstypename, mp->mnt_stat.f_mntonname, 2312 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntfromspec); 2313 2314 (*pr)("locked vnodes:"); 2315 /* XXX would take mountlist lock, except ddb has no context */ 2316 cnt = 0; 2317 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2318 if (VOP_ISLOCKED(vp)) { 2319 if (cnt == 0) 2320 (*pr)("\n %p", vp); 2321 else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0) 2322 (*pr)(",\n %p", vp); 2323 else 2324 (*pr)(", %p", vp); 2325 cnt++; 2326 } 2327 } 2328 (*pr)("\n"); 2329 2330 if (full) { 2331 (*pr)("all vnodes:"); 2332 /* XXX would take mountlist lock, except ddb has no context */ 2333 cnt = 0; 2334 TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { 2335 if (cnt == 0) 2336 (*pr)("\n %p", vp); 2337 else if ((cnt % (72 / (sizeof(void *) * 2 + 4))) == 0) 2338 (*pr)(",\n %p", vp); 2339 else 2340 (*pr)(", %p", vp); 2341 cnt++; 2342 } 2343 (*pr)("\n"); 2344 } 2345 } 2346 #endif /* DDB */ 2347 2348 void 2349 copy_statfs_info(struct statfs *sbp, const struct mount *mp) 2350 { 2351 const struct statfs *mbp; 2352 2353 strncpy(sbp->f_fstypename, mp->mnt_vfc->vfc_name, MFSNAMELEN); 2354 2355 if (sbp == (mbp = &mp->mnt_stat)) 2356 return; 2357 2358 sbp->f_fsid = mbp->f_fsid; 2359 sbp->f_owner = mbp->f_owner; 2360 sbp->f_flags = mbp->f_flags; 2361 sbp->f_syncwrites = mbp->f_syncwrites; 2362 sbp->f_asyncwrites = mbp->f_asyncwrites; 2363 sbp->f_syncreads = mbp->f_syncreads; 2364 sbp->f_asyncreads = mbp->f_asyncreads; 2365 sbp->f_namemax = mbp->f_namemax; 2366 memcpy(sbp->f_mntonname, mp->mnt_stat.f_mntonname, MNAMELEN); 2367 memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, MNAMELEN); 2368 memcpy(sbp->f_mntfromspec, mp->mnt_stat.f_mntfromspec, MNAMELEN); 2369 memcpy(&sbp->mount_info, &mp->mnt_stat.mount_info, 2370 sizeof(union mount_info)); 2371 } 2372