The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/vfs_subr.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 3. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)vfs_subr.c  8.31 (Berkeley) 5/26/95
   35  * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
   36  */
   37 
   38 /*
   39  * External virtual filesystem routines
   40  */
   41 #include "opt_ddb.h"
   42 
   43 #include <sys/param.h>
   44 #include <sys/systm.h>
   45 #include <sys/buf.h>
   46 #include <sys/conf.h>
   47 #include <sys/dirent.h>
   48 #include <sys/domain.h>
   49 #include <sys/eventhandler.h>
   50 #include <sys/fcntl.h>
   51 #include <sys/file.h>
   52 #include <sys/kernel.h>
   53 #include <sys/kthread.h>
   54 #include <sys/malloc.h>
   55 #include <sys/mbuf.h>
   56 #include <sys/mount.h>
   57 #include <sys/priv.h>
   58 #include <sys/proc.h>
   59 #include <sys/reboot.h>
   60 #include <sys/socket.h>
   61 #include <sys/stat.h>
   62 #include <sys/sysctl.h>
   63 #include <sys/syslog.h>
   64 #include <sys/unistd.h>
   65 #include <sys/vmmeter.h>
   66 #include <sys/vnode.h>
   67 
   68 #include <machine/limits.h>
   69 
   70 #include <vm/vm.h>
   71 #include <vm/vm_object.h>
   72 #include <vm/vm_extern.h>
   73 #include <vm/vm_kern.h>
   74 #include <vm/pmap.h>
   75 #include <vm/vm_map.h>
   76 #include <vm/vm_page.h>
   77 #include <vm/vm_pager.h>
   78 #include <vm/vnode_pager.h>
   79 #include <vm/vm_zone.h>
   80 
   81 #include <sys/buf2.h>
   82 #include <sys/thread2.h>
   83 #include <sys/sysref2.h>
   84 #include <sys/mplock2.h>
   85 
   86 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
   87 
   88 int numvnodes;
   89 SYSCTL_INT(_debug, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
   90     "Number of vnodes allocated");
   91 int verbose_reclaims;
   92 SYSCTL_INT(_debug, OID_AUTO, verbose_reclaims, CTLFLAG_RD, &verbose_reclaims, 0,
   93     "Output filename of reclaimed vnode(s)");
   94 
   95 enum vtype iftovt_tab[16] = {
   96         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
   97         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
   98 };
   99 int vttoif_tab[9] = {
  100         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
  101         S_IFSOCK, S_IFIFO, S_IFMT,
  102 };
  103 
  104 static int reassignbufcalls;
  105 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls,
  106     0, "Number of times buffers have been reassigned to the proper list");
  107 
  108 static int check_buf_overlap = 2;       /* invasive check */
  109 SYSCTL_INT(_vfs, OID_AUTO, check_buf_overlap, CTLFLAG_RW, &check_buf_overlap,
  110     0, "Enable overlapping buffer checks");
  111 
  112 int     nfs_mount_type = -1;
  113 static struct lwkt_token spechash_token;
  114 struct nfs_public nfs_pub;      /* publicly exported FS */
  115 
  116 int desiredvnodes;
  117 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 
  118                 &desiredvnodes, 0, "Maximum number of vnodes");
  119 
  120 static void     vfs_free_addrlist (struct netexport *nep);
  121 static int      vfs_free_netcred (struct radix_node *rn, void *w);
  122 static int      vfs_hang_addrlist (struct mount *mp, struct netexport *nep,
  123                                        const struct export_args *argp);
  124 
  125 /*
  126  * Red black tree functions
  127  */
  128 static int rb_buf_compare(struct buf *b1, struct buf *b2);
  129 RB_GENERATE2(buf_rb_tree, buf, b_rbnode, rb_buf_compare, off_t, b_loffset);
  130 RB_GENERATE2(buf_rb_hash, buf, b_rbhash, rb_buf_compare, off_t, b_loffset);
  131 
  132 static int
  133 rb_buf_compare(struct buf *b1, struct buf *b2)
  134 {
  135         if (b1->b_loffset < b2->b_loffset)
  136                 return(-1);
  137         if (b1->b_loffset > b2->b_loffset)
  138                 return(1);
  139         return(0);
  140 }
  141 
  142 /*
  143  * Initialize the vnode management data structures. 
  144  *
  145  * Called from vfsinit()
  146  */
  147 void
  148 vfs_subr_init(void)
  149 {
  150         int factor1;
  151         int factor2;
  152 
  153         /*
  154          * Desiredvnodes is kern.maxvnodes.  We want to scale it 
  155          * according to available system memory but we may also have
  156          * to limit it based on available KVM, which is capped on 32 bit
  157          * systems, to ~80K vnodes or so.
  158          *
  159          * WARNING!  For machines with 64-256M of ram we have to be sure
  160          *           that the default limit scales down well due to HAMMER
  161          *           taking up significantly more memory per-vnode vs UFS.
  162          *           We want around ~5800 on a 128M machine.
  163          */
  164         factor1 = 20 * (sizeof(struct vm_object) + sizeof(struct vnode));
  165         factor2 = 25 * (sizeof(struct vm_object) + sizeof(struct vnode));
  166         desiredvnodes =
  167                 imin((int64_t)vmstats.v_page_count * PAGE_SIZE / factor1,
  168                      KvaSize / factor2);
  169         desiredvnodes = imax(desiredvnodes, maxproc * 8);
  170 
  171         lwkt_token_init(&spechash_token, "spechash");
  172 }
  173 
  174 /*
  175  * Knob to control the precision of file timestamps:
  176  *
  177  *   0 = seconds only; nanoseconds zeroed.
  178  *   1 = seconds and nanoseconds, accurate within 1/HZ.
  179  *   2 = seconds and nanoseconds, truncated to microseconds.
  180  * >=3 = seconds and nanoseconds, maximum precision.
  181  */
  182 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
  183 
  184 static int timestamp_precision = TSP_SEC;
  185 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
  186                 &timestamp_precision, 0, "Precision of file timestamps");
  187 
  188 /*
  189  * Get a current timestamp.
  190  *
  191  * MPSAFE
  192  */
  193 void
  194 vfs_timestamp(struct timespec *tsp)
  195 {
  196         struct timeval tv;
  197 
  198         switch (timestamp_precision) {
  199         case TSP_SEC:
  200                 tsp->tv_sec = time_second;
  201                 tsp->tv_nsec = 0;
  202                 break;
  203         case TSP_HZ:
  204                 getnanotime(tsp);
  205                 break;
  206         case TSP_USEC:
  207                 microtime(&tv);
  208                 TIMEVAL_TO_TIMESPEC(&tv, tsp);
  209                 break;
  210         case TSP_NSEC:
  211         default:
  212                 nanotime(tsp);
  213                 break;
  214         }
  215 }
  216 
  217 /*
  218  * Set vnode attributes to VNOVAL
  219  */
  220 void
  221 vattr_null(struct vattr *vap)
  222 {
  223         vap->va_type = VNON;
  224         vap->va_size = VNOVAL;
  225         vap->va_bytes = VNOVAL;
  226         vap->va_mode = VNOVAL;
  227         vap->va_nlink = VNOVAL;
  228         vap->va_uid = VNOVAL;
  229         vap->va_gid = VNOVAL;
  230         vap->va_fsid = VNOVAL;
  231         vap->va_fileid = VNOVAL;
  232         vap->va_blocksize = VNOVAL;
  233         vap->va_rmajor = VNOVAL;
  234         vap->va_rminor = VNOVAL;
  235         vap->va_atime.tv_sec = VNOVAL;
  236         vap->va_atime.tv_nsec = VNOVAL;
  237         vap->va_mtime.tv_sec = VNOVAL;
  238         vap->va_mtime.tv_nsec = VNOVAL;
  239         vap->va_ctime.tv_sec = VNOVAL;
  240         vap->va_ctime.tv_nsec = VNOVAL;
  241         vap->va_flags = VNOVAL;
  242         vap->va_gen = VNOVAL;
  243         vap->va_vaflags = 0;
  244         /* va_*_uuid fields are only valid if related flags are set */
  245 }
  246 
  247 /*
  248  * Flush out and invalidate all buffers associated with a vnode.
  249  *
  250  * vp must be locked.
  251  */
  252 static int vinvalbuf_bp(struct buf *bp, void *data);
  253 
  254 struct vinvalbuf_bp_info {
  255         struct vnode *vp;
  256         int slptimeo;
  257         int lkflags;
  258         int flags;
  259         int clean;
  260 };
  261 
  262 int
  263 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
  264 {
  265         struct vinvalbuf_bp_info info;
  266         vm_object_t object;
  267         int error;
  268 
  269         lwkt_gettoken(&vp->v_token);
  270 
  271         /*
  272          * If we are being asked to save, call fsync to ensure that the inode
  273          * is updated.
  274          */
  275         if (flags & V_SAVE) {
  276                 error = bio_track_wait(&vp->v_track_write, slpflag, slptimeo);
  277                 if (error)
  278                         goto done;
  279                 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
  280                         if ((error = VOP_FSYNC(vp, MNT_WAIT, 0)) != 0)
  281                                 goto done;
  282 #if 0
  283                         /*
  284                          * Dirty bufs may be left or generated via races
  285                          * in circumstances where vinvalbuf() is called on
  286                          * a vnode not undergoing reclamation.   Only
  287                          * panic if we are trying to reclaim the vnode.
  288                          */
  289                         if ((vp->v_flag & VRECLAIMED) &&
  290                             (bio_track_active(&vp->v_track_write) ||
  291                             !RB_EMPTY(&vp->v_rbdirty_tree))) {
  292                                 panic("vinvalbuf: dirty bufs");
  293                         }
  294 #endif
  295                 }
  296         }
  297         info.slptimeo = slptimeo;
  298         info.lkflags = LK_EXCLUSIVE | LK_SLEEPFAIL;
  299         if (slpflag & PCATCH)
  300                 info.lkflags |= LK_PCATCH;
  301         info.flags = flags;
  302         info.vp = vp;
  303 
  304         /*
  305          * Flush the buffer cache until nothing is left, wait for all I/O
  306          * to complete.  At least one pass is required.  We might block
  307          * in the pip code so we have to re-check.  Order is important.
  308          */
  309         do {
  310                 /*
  311                  * Flush buffer cache
  312                  */
  313                 if (!RB_EMPTY(&vp->v_rbclean_tree)) {
  314                         info.clean = 1;
  315                         error = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree,
  316                                         NULL, vinvalbuf_bp, &info);
  317                 }
  318                 if (!RB_EMPTY(&vp->v_rbdirty_tree)) {
  319                         info.clean = 0;
  320                         error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
  321                                         NULL, vinvalbuf_bp, &info);
  322                 }
  323 
  324                 /*
  325                  * Wait for I/O completion.
  326                  */
  327                 bio_track_wait(&vp->v_track_write, 0, 0);
  328                 if ((object = vp->v_object) != NULL)
  329                         refcount_wait(&object->paging_in_progress, "vnvlbx");
  330         } while (bio_track_active(&vp->v_track_write) ||
  331                  !RB_EMPTY(&vp->v_rbclean_tree) ||
  332                  !RB_EMPTY(&vp->v_rbdirty_tree));
  333 
  334         /*
  335          * Destroy the copy in the VM cache, too.
  336          */
  337         if ((object = vp->v_object) != NULL) {
  338                 vm_object_page_remove(object, 0, 0,
  339                         (flags & V_SAVE) ? TRUE : FALSE);
  340         }
  341 
  342         if (!RB_EMPTY(&vp->v_rbdirty_tree) || !RB_EMPTY(&vp->v_rbclean_tree))
  343                 panic("vinvalbuf: flush failed");
  344         if (!RB_EMPTY(&vp->v_rbhash_tree))
  345                 panic("vinvalbuf: flush failed, buffers still present");
  346         error = 0;
  347 done:
  348         lwkt_reltoken(&vp->v_token);
  349         return (error);
  350 }
  351 
  352 static int
  353 vinvalbuf_bp(struct buf *bp, void *data)
  354 {
  355         struct vinvalbuf_bp_info *info = data;
  356         int error;
  357 
  358         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
  359                 atomic_add_int(&bp->b_refs, 1);
  360                 error = BUF_TIMELOCK(bp, info->lkflags,
  361                                      "vinvalbuf", info->slptimeo);
  362                 atomic_subtract_int(&bp->b_refs, 1);
  363                 if (error == 0) {
  364                         BUF_UNLOCK(bp);
  365                         error = ENOLCK;
  366                 }
  367                 if (error == ENOLCK)
  368                         return(0);
  369                 return (-error);
  370         }
  371         KKASSERT(bp->b_vp == info->vp);
  372 
  373         /*
  374          * Must check clean/dirty status after successfully locking as
  375          * it may race.
  376          */
  377         if ((info->clean && (bp->b_flags & B_DELWRI)) ||
  378             (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0)) {
  379                 BUF_UNLOCK(bp);
  380                 return(0);
  381         }
  382 
  383         /*
  384          * NOTE:  NO B_LOCKED CHECK.  Also no buf_checkwrite()
  385          * check.  This code will write out the buffer, period.
  386          */
  387         bremfree(bp);
  388         if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
  389             (info->flags & V_SAVE)) {
  390                 cluster_awrite(bp);
  391         } else if (info->flags & V_SAVE) {
  392                 /*
  393                  * Cannot set B_NOCACHE on a clean buffer as this will
  394                  * destroy the VM backing store which might actually
  395                  * be dirty (and unsynchronized).
  396                  */
  397                 bp->b_flags |= (B_INVAL | B_RELBUF);
  398                 brelse(bp);
  399         } else {
  400                 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
  401                 brelse(bp);
  402         }
  403         return(0);
  404 }
  405 
  406 /*
  407  * Truncate a file's buffer and pages to a specified length.  This
  408  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
  409  * sync activity.
  410  *
  411  * The vnode must be locked.
  412  */
  413 static int vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data);
  414 static int vtruncbuf_bp_trunc(struct buf *bp, void *data);
  415 static int vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data);
  416 static int vtruncbuf_bp_metasync(struct buf *bp, void *data);
  417 
  418 struct vtruncbuf_info {
  419         struct vnode *vp;
  420         off_t   truncloffset;
  421         int     clean;
  422 };
  423 
  424 int
  425 vtruncbuf(struct vnode *vp, off_t length, int blksize)
  426 {
  427         struct vtruncbuf_info info;
  428         const char *filename;
  429         int count;
  430 
  431         /*
  432          * Round up to the *next* block, then destroy the buffers in question.  
  433          * Since we are only removing some of the buffers we must rely on the
  434          * scan count to determine whether a loop is necessary.
  435          */
  436         if ((count = (int)(length % blksize)) != 0)
  437                 info.truncloffset = length + (blksize - count);
  438         else
  439                 info.truncloffset = length;
  440         info.vp = vp;
  441 
  442         lwkt_gettoken(&vp->v_token);
  443         do {
  444                 info.clean = 1;
  445                 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 
  446                                 vtruncbuf_bp_trunc_cmp,
  447                                 vtruncbuf_bp_trunc, &info);
  448                 info.clean = 0;
  449                 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
  450                                 vtruncbuf_bp_trunc_cmp,
  451                                 vtruncbuf_bp_trunc, &info);
  452         } while(count);
  453 
  454         /*
  455          * For safety, fsync any remaining metadata if the file is not being
  456          * truncated to 0.  Since the metadata does not represent the entire
  457          * dirty list we have to rely on the hit count to ensure that we get
  458          * all of it.
  459          */
  460         if (length > 0) {
  461                 do {
  462                         count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
  463                                         vtruncbuf_bp_metasync_cmp,
  464                                         vtruncbuf_bp_metasync, &info);
  465                 } while (count);
  466         }
  467 
  468         /*
  469          * Clean out any left over VM backing store.
  470          *
  471          * It is possible to have in-progress I/O from buffers that were
  472          * not part of the truncation.  This should not happen if we
  473          * are truncating to 0-length.
  474          */
  475         vnode_pager_setsize(vp, length);
  476         bio_track_wait(&vp->v_track_write, 0, 0);
  477 
  478         /*
  479          * Debugging only
  480          */
  481         spin_lock(&vp->v_spin);
  482         filename = TAILQ_FIRST(&vp->v_namecache) ?
  483                    TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
  484         spin_unlock(&vp->v_spin);
  485 
  486         /*
  487          * Make sure no buffers were instantiated while we were trying
  488          * to clean out the remaining VM pages.  This could occur due
  489          * to busy dirty VM pages being flushed out to disk.
  490          */
  491         do {
  492                 info.clean = 1;
  493                 count = RB_SCAN(buf_rb_tree, &vp->v_rbclean_tree, 
  494                                 vtruncbuf_bp_trunc_cmp,
  495                                 vtruncbuf_bp_trunc, &info);
  496                 info.clean = 0;
  497                 count += RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree,
  498                                 vtruncbuf_bp_trunc_cmp,
  499                                 vtruncbuf_bp_trunc, &info);
  500                 if (count) {
  501                         kprintf("Warning: vtruncbuf():  Had to re-clean %d "
  502                                "left over buffers in %s\n", count, filename);
  503                 }
  504         } while(count);
  505 
  506         lwkt_reltoken(&vp->v_token);
  507 
  508         return (0);
  509 }
  510 
  511 /*
  512  * The callback buffer is beyond the new file EOF and must be destroyed.
  513  * Note that the compare function must conform to the RB_SCAN's requirements.
  514  */
  515 static
  516 int
  517 vtruncbuf_bp_trunc_cmp(struct buf *bp, void *data)
  518 {
  519         struct vtruncbuf_info *info = data;
  520 
  521         if (bp->b_loffset >= info->truncloffset)
  522                 return(0);
  523         return(-1);
  524 }
  525 
  526 static 
  527 int 
  528 vtruncbuf_bp_trunc(struct buf *bp, void *data)
  529 {
  530         struct vtruncbuf_info *info = data;
  531 
  532         /*
  533          * Do not try to use a buffer we cannot immediately lock, but sleep
  534          * anyway to prevent a livelock.  The code will loop until all buffers
  535          * can be acted upon.
  536          *
  537          * We must always revalidate the buffer after locking it to deal
  538          * with MP races.
  539          */
  540         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
  541                 atomic_add_int(&bp->b_refs, 1);
  542                 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
  543                         BUF_UNLOCK(bp);
  544                 atomic_subtract_int(&bp->b_refs, 1);
  545         } else if ((info->clean && (bp->b_flags & B_DELWRI)) ||
  546                    (info->clean == 0 && (bp->b_flags & B_DELWRI) == 0) ||
  547                    bp->b_vp != info->vp ||
  548                    vtruncbuf_bp_trunc_cmp(bp, data)) {
  549                 BUF_UNLOCK(bp);
  550         } else {
  551                 bremfree(bp);
  552                 bp->b_flags |= (B_INVAL | B_RELBUF | B_NOCACHE);
  553                 brelse(bp);
  554         }
  555         return(1);
  556 }
  557 
  558 /*
  559  * Fsync all meta-data after truncating a file to be non-zero.  Only metadata
  560  * blocks (with a negative loffset) are scanned.
  561  * Note that the compare function must conform to the RB_SCAN's requirements.
  562  */
  563 static int
  564 vtruncbuf_bp_metasync_cmp(struct buf *bp, void *data __unused)
  565 {
  566         if (bp->b_loffset < 0)
  567                 return(0);
  568         return(1);
  569 }
  570 
  571 static int
  572 vtruncbuf_bp_metasync(struct buf *bp, void *data)
  573 {
  574         struct vtruncbuf_info *info = data;
  575 
  576         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
  577                 atomic_add_int(&bp->b_refs, 1);
  578                 if (BUF_LOCK(bp, LK_EXCLUSIVE|LK_SLEEPFAIL) == 0)
  579                         BUF_UNLOCK(bp);
  580                 atomic_subtract_int(&bp->b_refs, 1);
  581         } else if ((bp->b_flags & B_DELWRI) == 0 ||
  582                    bp->b_vp != info->vp ||
  583                    vtruncbuf_bp_metasync_cmp(bp, data)) {
  584                 BUF_UNLOCK(bp);
  585         } else {
  586                 bremfree(bp);
  587                 if (bp->b_vp == info->vp)
  588                         bawrite(bp);
  589                 else
  590                         bwrite(bp);
  591         }
  592         return(1);
  593 }
  594 
  595 /*
  596  * vfsync - implements a multipass fsync on a file which understands
  597  * dependancies and meta-data.  The passed vnode must be locked.  The 
  598  * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
  599  *
  600  * When fsyncing data asynchronously just do one consolidated pass starting
  601  * with the most negative block number.  This may not get all the data due
  602  * to dependancies.
  603  *
  604  * When fsyncing data synchronously do a data pass, then a metadata pass,
  605  * then do additional data+metadata passes to try to get all the data out.
  606  */
  607 static int vfsync_wait_output(struct vnode *vp, 
  608                             int (*waitoutput)(struct vnode *, struct thread *));
  609 static int vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused);
  610 static int vfsync_data_only_cmp(struct buf *bp, void *data);
  611 static int vfsync_meta_only_cmp(struct buf *bp, void *data);
  612 static int vfsync_lazy_range_cmp(struct buf *bp, void *data);
  613 static int vfsync_bp(struct buf *bp, void *data);
  614 
  615 struct vfsync_info {
  616         struct vnode *vp;
  617         int synchronous;
  618         int syncdeps;
  619         int lazycount;
  620         int lazylimit;
  621         int skippedbufs;
  622         int (*checkdef)(struct buf *);
  623         int (*cmpfunc)(struct buf *, void *);
  624 };
  625 
  626 int
  627 vfsync(struct vnode *vp, int waitfor, int passes,
  628         int (*checkdef)(struct buf *),
  629         int (*waitoutput)(struct vnode *, struct thread *))
  630 {
  631         struct vfsync_info info;
  632         int error;
  633 
  634         bzero(&info, sizeof(info));
  635         info.vp = vp;
  636         if ((info.checkdef = checkdef) == NULL)
  637                 info.syncdeps = 1;
  638 
  639         lwkt_gettoken(&vp->v_token);
  640 
  641         switch(waitfor) {
  642         case MNT_LAZY | MNT_NOWAIT:
  643         case MNT_LAZY:
  644                 /*
  645                  * Lazy (filesystem syncer typ) Asynchronous plus limit the
  646                  * number of data (not meta) pages we try to flush to 1MB.
  647                  * A non-zero return means that lazy limit was reached.
  648                  */
  649                 info.lazylimit = 1024 * 1024;
  650                 info.syncdeps = 1;
  651                 info.cmpfunc = vfsync_lazy_range_cmp;
  652                 error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 
  653                                 vfsync_lazy_range_cmp, vfsync_bp, &info);
  654                 info.cmpfunc = vfsync_meta_only_cmp;
  655                 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, 
  656                         vfsync_meta_only_cmp, vfsync_bp, &info);
  657                 if (error == 0)
  658                         vp->v_lazyw = 0;
  659                 else if (!RB_EMPTY(&vp->v_rbdirty_tree))
  660                         vn_syncer_add(vp, 1);
  661                 error = 0;
  662                 break;
  663         case MNT_NOWAIT:
  664                 /*
  665                  * Asynchronous.  Do a data-only pass and a meta-only pass.
  666                  */
  667                 info.syncdeps = 1;
  668                 info.cmpfunc = vfsync_data_only_cmp;
  669                 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp, 
  670                         vfsync_bp, &info);
  671                 info.cmpfunc = vfsync_meta_only_cmp;
  672                 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_meta_only_cmp, 
  673                         vfsync_bp, &info);
  674                 error = 0;
  675                 break;
  676         default:
  677                 /*
  678                  * Synchronous.  Do a data-only pass, then a meta-data+data
  679                  * pass, then additional integrated passes to try to get
  680                  * all the dependancies flushed.
  681                  */
  682                 info.cmpfunc = vfsync_data_only_cmp;
  683                 RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, vfsync_data_only_cmp,
  684                         vfsync_bp, &info);
  685                 error = vfsync_wait_output(vp, waitoutput);
  686                 if (error == 0) {
  687                         info.skippedbufs = 0;
  688                         info.cmpfunc = vfsync_dummy_cmp;
  689                         RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
  690                                 vfsync_bp, &info);
  691                         error = vfsync_wait_output(vp, waitoutput);
  692                         if (info.skippedbufs) {
  693                                 kprintf("Warning: vfsync skipped %d dirty "
  694                                         "bufs in pass2!\n", info.skippedbufs);
  695                         }
  696                 }
  697                 while (error == 0 && passes > 0 &&
  698                        !RB_EMPTY(&vp->v_rbdirty_tree)
  699                 ) {
  700                         if (--passes == 0) {
  701                                 info.synchronous = 1;
  702                                 info.syncdeps = 1;
  703                         }
  704                         info.cmpfunc = vfsync_dummy_cmp;
  705                         error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL,
  706                                         vfsync_bp, &info);
  707                         if (error < 0)
  708                                 error = -error;
  709                         info.syncdeps = 1;
  710                         if (error == 0)
  711                                 error = vfsync_wait_output(vp, waitoutput);
  712                 }
  713                 break;
  714         }
  715         lwkt_reltoken(&vp->v_token);
  716         return(error);
  717 }
  718 
  719 static int
  720 vfsync_wait_output(struct vnode *vp,
  721                    int (*waitoutput)(struct vnode *, struct thread *))
  722 {
  723         int error;
  724 
  725         error = bio_track_wait(&vp->v_track_write, 0, 0);
  726         if (waitoutput)
  727                 error = waitoutput(vp, curthread);
  728         return(error);
  729 }
  730 
  731 static int
  732 vfsync_dummy_cmp(struct buf *bp __unused, void *data __unused)
  733 {
  734         return(0);
  735 }
  736 
  737 static int
  738 vfsync_data_only_cmp(struct buf *bp, void *data)
  739 {
  740         if (bp->b_loffset < 0)
  741                 return(-1);
  742         return(0);
  743 }
  744 
  745 static int
  746 vfsync_meta_only_cmp(struct buf *bp, void *data)
  747 {
  748         if (bp->b_loffset < 0)
  749                 return(0);
  750         return(1);
  751 }
  752 
  753 static int
  754 vfsync_lazy_range_cmp(struct buf *bp, void *data)
  755 {
  756         struct vfsync_info *info = data;
  757 
  758         if (bp->b_loffset < info->vp->v_lazyw)
  759                 return(-1);
  760         return(0);
  761 }
  762 
  763 static int
  764 vfsync_bp(struct buf *bp, void *data)
  765 {
  766         struct vfsync_info *info = data;
  767         struct vnode *vp = info->vp;
  768         int error;
  769 
  770         /*
  771          * Ignore buffers that we cannot immediately lock.
  772          */
  773         if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
  774                 ++info->skippedbufs;
  775                 return(0);
  776         }
  777 
  778         /*
  779          * We must revalidate the buffer after locking.
  780          */
  781         if ((bp->b_flags & B_DELWRI) == 0 ||
  782             bp->b_vp != info->vp ||
  783             info->cmpfunc(bp, data)) {
  784                 BUF_UNLOCK(bp);
  785                 return(0);
  786         }
  787 
  788         /*
  789          * If syncdeps is not set we do not try to write buffers which have
  790          * dependancies.
  791          */
  792         if (!info->synchronous && info->syncdeps == 0 && info->checkdef(bp)) {
  793                 BUF_UNLOCK(bp);
  794                 return(0);
  795         }
  796 
  797         /*
  798          * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
  799          * has been written but an additional handshake with the device
  800          * is required before we can dispose of the buffer.  We have no idea
  801          * how to do this so we have to skip these buffers.
  802          */
  803         if (bp->b_flags & B_NEEDCOMMIT) {
  804                 BUF_UNLOCK(bp);
  805                 return(0);
  806         }
  807 
  808         /*
  809          * Ask bioops if it is ok to sync.  If not the VFS may have
  810          * set B_LOCKED so we have to cycle the buffer.
  811          */
  812         if (LIST_FIRST(&bp->b_dep) != NULL && buf_checkwrite(bp)) {
  813                 bremfree(bp);
  814                 brelse(bp);
  815                 return(0);
  816         }
  817 
  818         if (info->synchronous) {
  819                 /*
  820                  * Synchronous flushing.  An error may be returned.
  821                  */
  822                 bremfree(bp);
  823                 error = bwrite(bp);
  824         } else { 
  825                 /*
  826                  * Asynchronous flushing.  A negative return value simply
  827                  * stops the scan and is not considered an error.  We use
  828                  * this to support limited MNT_LAZY flushes.
  829                  */
  830                 vp->v_lazyw = bp->b_loffset;
  831                 bremfree(bp);
  832                 info->lazycount += cluster_awrite(bp);
  833                 waitrunningbufspace();
  834                 vm_wait_nominal();
  835                 if (info->lazylimit && info->lazycount >= info->lazylimit)
  836                         error = 1;
  837                 else
  838                         error = 0;
  839         }
  840         return(-error);
  841 }
  842 
  843 /*
  844  * Associate a buffer with a vnode.
  845  *
  846  * MPSAFE
  847  */
  848 int
  849 bgetvp(struct vnode *vp, struct buf *bp, int testsize)
  850 {
  851         KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
  852         KKASSERT((bp->b_flags & (B_HASHED|B_DELWRI|B_VNCLEAN|B_VNDIRTY)) == 0);
  853 
  854         /*
  855          * Insert onto list for new vnode.
  856          */
  857         lwkt_gettoken(&vp->v_token);
  858 
  859         if (buf_rb_hash_RB_INSERT(&vp->v_rbhash_tree, bp)) {
  860                 lwkt_reltoken(&vp->v_token);
  861                 return (EEXIST);
  862         }
  863 
  864         /*
  865          * Diagnostics (mainly for HAMMER debugging).  Check for
  866          * overlapping buffers.
  867          */
  868         if (check_buf_overlap) {
  869                 struct buf *bx;
  870                 bx = buf_rb_hash_RB_PREV(bp);
  871                 if (bx) {
  872                         if (bx->b_loffset + bx->b_bufsize > bp->b_loffset) {
  873                                 kprintf("bgetvp: overlapl %016jx/%d %016jx "
  874                                         "bx %p bp %p\n",
  875                                         (intmax_t)bx->b_loffset,
  876                                         bx->b_bufsize,
  877                                         (intmax_t)bp->b_loffset,
  878                                         bx, bp);
  879                                 if (check_buf_overlap > 1)
  880                                         panic("bgetvp - overlapping buffer");
  881                         }
  882                 }
  883                 bx = buf_rb_hash_RB_NEXT(bp);
  884                 if (bx) {
  885                         if (bp->b_loffset + testsize > bx->b_loffset) {
  886                                 kprintf("bgetvp: overlapr %016jx/%d %016jx "
  887                                         "bp %p bx %p\n",
  888                                         (intmax_t)bp->b_loffset,
  889                                         testsize,
  890                                         (intmax_t)bx->b_loffset,
  891                                         bp, bx);
  892                                 if (check_buf_overlap > 1)
  893                                         panic("bgetvp - overlapping buffer");
  894                         }
  895                 }
  896         }
  897         bp->b_vp = vp;
  898         bp->b_flags |= B_HASHED;
  899         bp->b_flags |= B_VNCLEAN;
  900         if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp))
  901                 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp, bp);
  902         /*vhold(vp);*/
  903         lwkt_reltoken(&vp->v_token);
  904         return(0);
  905 }
  906 
  907 /*
  908  * Disassociate a buffer from a vnode.
  909  *
  910  * MPSAFE
  911  */
  912 void
  913 brelvp(struct buf *bp)
  914 {
  915         struct vnode *vp;
  916 
  917         KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
  918 
  919         /*
  920          * Delete from old vnode list, if on one.
  921          */
  922         vp = bp->b_vp;
  923         lwkt_gettoken(&vp->v_token);
  924         if (bp->b_flags & (B_VNDIRTY | B_VNCLEAN)) {
  925                 if (bp->b_flags & B_VNDIRTY)
  926                         buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
  927                 else
  928                         buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
  929                 bp->b_flags &= ~(B_VNDIRTY | B_VNCLEAN);
  930         }
  931         if (bp->b_flags & B_HASHED) {
  932                 buf_rb_hash_RB_REMOVE(&vp->v_rbhash_tree, bp);
  933                 bp->b_flags &= ~B_HASHED;
  934         }
  935 
  936         /*
  937          * Only remove from synclist when no dirty buffers are left AND
  938          * the VFS has not flagged the vnode's inode as being dirty.
  939          */
  940         if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) == VONWORKLST &&
  941             RB_EMPTY(&vp->v_rbdirty_tree)) {
  942                 vn_syncer_remove(vp);
  943         }
  944         bp->b_vp = NULL;
  945 
  946         lwkt_reltoken(&vp->v_token);
  947 
  948         /*vdrop(vp);*/
  949 }
  950 
  951 /*
  952  * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
  953  * This routine is called when the state of the B_DELWRI bit is changed.
  954  *
  955  * Must be called with vp->v_token held.
  956  * MPSAFE
  957  */
  958 void
  959 reassignbuf(struct buf *bp)
  960 {
  961         struct vnode *vp = bp->b_vp;
  962         int delay;
  963 
  964         ASSERT_LWKT_TOKEN_HELD(&vp->v_token);
  965         ++reassignbufcalls;
  966 
  967         /*
  968          * B_PAGING flagged buffers cannot be reassigned because their vp
  969          * is not fully linked in.
  970          */
  971         if (bp->b_flags & B_PAGING)
  972                 panic("cannot reassign paging buffer");
  973 
  974         if (bp->b_flags & B_DELWRI) {
  975                 /*
  976                  * Move to the dirty list, add the vnode to the worklist
  977                  */
  978                 if (bp->b_flags & B_VNCLEAN) {
  979                         buf_rb_tree_RB_REMOVE(&vp->v_rbclean_tree, bp);
  980                         bp->b_flags &= ~B_VNCLEAN;
  981                 }
  982                 if ((bp->b_flags & B_VNDIRTY) == 0) {
  983                         if (buf_rb_tree_RB_INSERT(&vp->v_rbdirty_tree, bp)) {
  984                                 panic("reassignbuf: dup lblk vp %p bp %p",
  985                                       vp, bp);
  986                         }
  987                         bp->b_flags |= B_VNDIRTY;
  988                 }
  989                 if ((vp->v_flag & VONWORKLST) == 0) {
  990                         switch (vp->v_type) {
  991                         case VDIR:
  992                                 delay = dirdelay;
  993                                 break;
  994                         case VCHR:
  995                         case VBLK:
  996                                 if (vp->v_rdev && 
  997                                     vp->v_rdev->si_mountpoint != NULL) {
  998                                         delay = metadelay;
  999                                         break;
 1000                                 }
 1001                                 /* fall through */
 1002                         default:
 1003                                 delay = filedelay;
 1004                         }
 1005                         vn_syncer_add(vp, delay);
 1006                 }
 1007         } else {
 1008                 /*
 1009                  * Move to the clean list, remove the vnode from the worklist
 1010                  * if no dirty blocks remain.
 1011                  */
 1012                 if (bp->b_flags & B_VNDIRTY) {
 1013                         buf_rb_tree_RB_REMOVE(&vp->v_rbdirty_tree, bp);
 1014                         bp->b_flags &= ~B_VNDIRTY;
 1015                 }
 1016                 if ((bp->b_flags & B_VNCLEAN) == 0) {
 1017                         if (buf_rb_tree_RB_INSERT(&vp->v_rbclean_tree, bp)) {
 1018                                 panic("reassignbuf: dup lblk vp %p bp %p",
 1019                                       vp, bp);
 1020                         }
 1021                         bp->b_flags |= B_VNCLEAN;
 1022                 }
 1023 
 1024                 /*
 1025                  * Only remove from synclist when no dirty buffers are left
 1026                  * AND the VFS has not flagged the vnode's inode as being
 1027                  * dirty.
 1028                  */
 1029                 if ((vp->v_flag & (VONWORKLST | VISDIRTY | VOBJDIRTY)) ==
 1030                      VONWORKLST &&
 1031                     RB_EMPTY(&vp->v_rbdirty_tree)) {
 1032                         vn_syncer_remove(vp);
 1033                 }
 1034         }
 1035 }
 1036 
 1037 /*
 1038  * Create a vnode for a block device.  Used for mounting the root file
 1039  * system.
 1040  *
 1041  * A vref()'d vnode is returned.
 1042  */
 1043 extern struct vop_ops *devfs_vnode_dev_vops_p;
 1044 int
 1045 bdevvp(cdev_t dev, struct vnode **vpp)
 1046 {
 1047         struct vnode *vp;
 1048         struct vnode *nvp;
 1049         int error;
 1050 
 1051         if (dev == NULL) {
 1052                 *vpp = NULLVP;
 1053                 return (ENXIO);
 1054         }
 1055         error = getspecialvnode(VT_NON, NULL, &devfs_vnode_dev_vops_p,
 1056                                 &nvp, 0, 0);
 1057         if (error) {
 1058                 *vpp = NULLVP;
 1059                 return (error);
 1060         }
 1061         vp = nvp;
 1062         vp->v_type = VCHR;
 1063 #if 0
 1064         vp->v_rdev = dev;
 1065 #endif
 1066         v_associate_rdev(vp, dev);
 1067         vp->v_umajor = dev->si_umajor;
 1068         vp->v_uminor = dev->si_uminor;
 1069         vx_unlock(vp);
 1070         *vpp = vp;
 1071         return (0);
 1072 }
 1073 
 1074 int
 1075 v_associate_rdev(struct vnode *vp, cdev_t dev)
 1076 {
 1077         if (dev == NULL)
 1078                 return(ENXIO);
 1079         if (dev_is_good(dev) == 0)
 1080                 return(ENXIO);
 1081         KKASSERT(vp->v_rdev == NULL);
 1082         vp->v_rdev = reference_dev(dev);
 1083         lwkt_gettoken(&spechash_token);
 1084         SLIST_INSERT_HEAD(&dev->si_hlist, vp, v_cdevnext);
 1085         lwkt_reltoken(&spechash_token);
 1086         return(0);
 1087 }
 1088 
 1089 void
 1090 v_release_rdev(struct vnode *vp)
 1091 {
 1092         cdev_t dev;
 1093 
 1094         if ((dev = vp->v_rdev) != NULL) {
 1095                 lwkt_gettoken(&spechash_token);
 1096                 SLIST_REMOVE(&dev->si_hlist, vp, vnode, v_cdevnext);
 1097                 vp->v_rdev = NULL;
 1098                 release_dev(dev);
 1099                 lwkt_reltoken(&spechash_token);
 1100         }
 1101 }
 1102 
 1103 /*
 1104  * Add a vnode to the alias list hung off the cdev_t.  We only associate
 1105  * the device number with the vnode.  The actual device is not associated
 1106  * until the vnode is opened (usually in spec_open()), and will be 
 1107  * disassociated on last close.
 1108  */
 1109 void
 1110 addaliasu(struct vnode *nvp, int x, int y)
 1111 {
 1112         if (nvp->v_type != VBLK && nvp->v_type != VCHR)
 1113                 panic("addaliasu on non-special vnode");
 1114         nvp->v_umajor = x;
 1115         nvp->v_uminor = y;
 1116 }
 1117 
 1118 /*
 1119  * Simple call that a filesystem can make to try to get rid of a
 1120  * vnode.  It will fail if anyone is referencing the vnode (including
 1121  * the caller).
 1122  *
 1123  * The filesystem can check whether its in-memory inode structure still
 1124  * references the vp on return.
 1125  *
 1126  * May only be called if the vnode is in a known state (i.e. being prevented
 1127  * from being deallocated by some other condition such as a vfs inode hold).
 1128  */
 1129 void
 1130 vclean_unlocked(struct vnode *vp)
 1131 {
 1132         vx_get(vp);
 1133         if (VREFCNT(vp) <= 0)
 1134                 vgone_vxlocked(vp);
 1135         vx_put(vp);
 1136 }
 1137 
 1138 /*
 1139  * Disassociate a vnode from its underlying filesystem. 
 1140  *
 1141  * The vnode must be VX locked and referenced.  In all normal situations
 1142  * there are no active references.  If vclean_vxlocked() is called while
 1143  * there are active references, the vnode is being ripped out and we have
 1144  * to call VOP_CLOSE() as appropriate before we can reclaim it.
 1145  */
 1146 void
 1147 vclean_vxlocked(struct vnode *vp, int flags)
 1148 {
 1149         int active;
 1150         int n;
 1151         vm_object_t object;
 1152         struct namecache *ncp;
 1153 
 1154         /*
 1155          * If the vnode has already been reclaimed we have nothing to do.
 1156          */
 1157         if (vp->v_flag & VRECLAIMED)
 1158                 return;
 1159 
 1160         /*
 1161          * Set flag to interlock operation, flag finalization to ensure
 1162          * that the vnode winds up on the inactive list, and set v_act to 0.
 1163          */
 1164         vsetflags(vp, VRECLAIMED);
 1165         atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
 1166         vp->v_act = 0;
 1167 
 1168         if (verbose_reclaims) {
 1169                 if ((ncp = TAILQ_FIRST(&vp->v_namecache)) != NULL)
 1170                         kprintf("Debug: reclaim %p %s\n", vp, ncp->nc_name);
 1171         }
 1172 
 1173         /*
 1174          * Scrap the vfs cache
 1175          */
 1176         while (cache_inval_vp(vp, 0) != 0) {
 1177                 kprintf("Warning: vnode %p clean/cache_resolution "
 1178                         "race detected\n", vp);
 1179                 tsleep(vp, 0, "vclninv", 2);
 1180         }
 1181 
 1182         /*
 1183          * Check to see if the vnode is in use. If so we have to reference it
 1184          * before we clean it out so that its count cannot fall to zero and
 1185          * generate a race against ourselves to recycle it.
 1186          */
 1187         active = (VREFCNT(vp) > 0);
 1188 
 1189         /*
 1190          * Clean out any buffers associated with the vnode and destroy its
 1191          * object, if it has one. 
 1192          */
 1193         vinvalbuf(vp, V_SAVE, 0, 0);
 1194         KKASSERT(lockcountnb(&vp->v_lock) == 1);
 1195 
 1196         /*
 1197          * If purging an active vnode (typically during a forced unmount
 1198          * or reboot), it must be closed and deactivated before being
 1199          * reclaimed.  This isn't really all that safe, but what can
 1200          * we do? XXX.
 1201          *
 1202          * Note that neither of these routines unlocks the vnode.
 1203          */
 1204         if (active && (flags & DOCLOSE)) {
 1205                 while ((n = vp->v_opencount) != 0) {
 1206                         if (vp->v_writecount)
 1207                                 VOP_CLOSE(vp, FWRITE|FNONBLOCK);
 1208                         else
 1209                                 VOP_CLOSE(vp, FNONBLOCK);
 1210                         if (vp->v_opencount == n) {
 1211                                 kprintf("Warning: unable to force-close"
 1212                                        " vnode %p\n", vp);
 1213                                 break;
 1214                         }
 1215                 }
 1216         }
 1217 
 1218         /*
 1219          * If the vnode has not been deactivated, deactivated it.  Deactivation
 1220          * can create new buffers and VM pages so we have to call vinvalbuf()
 1221          * again to make sure they all get flushed.
 1222          *
 1223          * This can occur if a file with a link count of 0 needs to be
 1224          * truncated.
 1225          *
 1226          * If the vnode is already dead don't try to deactivate it.
 1227          */
 1228         if ((vp->v_flag & VINACTIVE) == 0) {
 1229                 vsetflags(vp, VINACTIVE);
 1230                 if (vp->v_mount)
 1231                         VOP_INACTIVE(vp);
 1232                 vinvalbuf(vp, V_SAVE, 0, 0);
 1233         }
 1234         KKASSERT(lockcountnb(&vp->v_lock) == 1);
 1235 
 1236         /*
 1237          * If the vnode has an object, destroy it.
 1238          */
 1239         while ((object = vp->v_object) != NULL) {
 1240                 vm_object_hold(object);
 1241                 if (object == vp->v_object)
 1242                         break;
 1243                 vm_object_drop(object);
 1244         }
 1245 
 1246         if (object != NULL) {
 1247                 if (object->ref_count == 0) {
 1248                         if ((object->flags & OBJ_DEAD) == 0)
 1249                                 vm_object_terminate(object);
 1250                         vm_object_drop(object);
 1251                         vclrflags(vp, VOBJBUF);
 1252                 } else {
 1253                         vm_pager_deallocate(object);
 1254                         vclrflags(vp, VOBJBUF);
 1255                         vm_object_drop(object);
 1256                 }
 1257         }
 1258         KKASSERT((vp->v_flag & VOBJBUF) == 0);
 1259 
 1260         /*
 1261          * Reclaim the vnode if not already dead.
 1262          */
 1263         if (vp->v_mount && VOP_RECLAIM(vp))
 1264                 panic("vclean: cannot reclaim");
 1265 
 1266         /*
 1267          * Done with purge, notify sleepers of the grim news.
 1268          */
 1269         vp->v_ops = &dead_vnode_vops_p;
 1270         vn_gone(vp);
 1271         vp->v_tag = VT_NON;
 1272 
 1273         /*
 1274          * If we are destroying an active vnode, reactivate it now that
 1275          * we have reassociated it with deadfs.  This prevents the system
 1276          * from crashing on the vnode due to it being unexpectedly marked
 1277          * as inactive or reclaimed.
 1278          */
 1279         if (active && (flags & DOCLOSE)) {
 1280                 vclrflags(vp, VINACTIVE | VRECLAIMED);
 1281         }
 1282 }
 1283 
 1284 /*
 1285  * Eliminate all activity associated with the requested vnode
 1286  * and with all vnodes aliased to the requested vnode.
 1287  *
 1288  * The vnode must be referenced but should not be locked.
 1289  */
 1290 int
 1291 vrevoke(struct vnode *vp, struct ucred *cred)
 1292 {
 1293         struct vnode *vq;
 1294         struct vnode *vqn;
 1295         cdev_t dev;
 1296         int error;
 1297 
 1298         /*
 1299          * If the vnode has a device association, scrap all vnodes associated
 1300          * with the device.  Don't let the device disappear on us while we
 1301          * are scrapping the vnodes.
 1302          *
 1303          * The passed vp will probably show up in the list, do not VX lock
 1304          * it twice!
 1305          *
 1306          * Releasing the vnode's rdev here can mess up specfs's call to
 1307          * device close, so don't do it.  The vnode has been disassociated
 1308          * and the device will be closed after the last ref on the related
 1309          * fp goes away (if not still open by e.g. the kernel).
 1310          */
 1311         if (vp->v_type != VCHR) {
 1312                 error = fdrevoke(vp, DTYPE_VNODE, cred);
 1313                 return (error);
 1314         }
 1315         if ((dev = vp->v_rdev) == NULL) {
 1316                 return(0);
 1317         }
 1318         reference_dev(dev);
 1319         lwkt_gettoken(&spechash_token);
 1320 
 1321 restart:
 1322         vqn = SLIST_FIRST(&dev->si_hlist);
 1323         if (vqn)
 1324                 vhold(vqn);
 1325         while ((vq = vqn) != NULL) {
 1326                 if (VREFCNT(vq) > 0) {
 1327                         vref(vq);
 1328                         fdrevoke(vq, DTYPE_VNODE, cred);
 1329                         /*v_release_rdev(vq);*/
 1330                         vrele(vq);
 1331                         if (vq->v_rdev != dev) {
 1332                                 vdrop(vq);
 1333                                 goto restart;
 1334                         }
 1335                 }
 1336                 vqn = SLIST_NEXT(vq, v_cdevnext);
 1337                 if (vqn)
 1338                         vhold(vqn);
 1339                 vdrop(vq);
 1340         }
 1341         lwkt_reltoken(&spechash_token);
 1342         dev_drevoke(dev);
 1343         release_dev(dev);
 1344         return (0);
 1345 }
 1346 
 1347 /*
 1348  * This is called when the object underlying a vnode is being destroyed,
 1349  * such as in a remove().  Try to recycle the vnode immediately if the
 1350  * only active reference is our reference.
 1351  *
 1352  * Directory vnodes in the namecache with children cannot be immediately
 1353  * recycled because numerous VOP_N*() ops require them to be stable.
 1354  *
 1355  * To avoid recursive recycling from VOP_INACTIVE implemenetations this
 1356  * function is a NOP if VRECLAIMED is already set.
 1357  */
 1358 int
 1359 vrecycle(struct vnode *vp)
 1360 {
 1361         if (VREFCNT(vp) <= 1 && (vp->v_flag & VRECLAIMED) == 0) {
 1362                 if (cache_inval_vp_nonblock(vp))
 1363                         return(0);
 1364                 vgone_vxlocked(vp);
 1365                 return (1);
 1366         }
 1367         return (0);
 1368 }
 1369 
 1370 /*
 1371  * Return the maximum I/O size allowed for strategy calls on VP.
 1372  *
 1373  * If vp is VCHR or VBLK we dive the device, otherwise we use
 1374  * the vp's mount info.
 1375  *
 1376  * The returned value is clamped at MAXPHYS as most callers cannot use
 1377  * buffers larger than that size.
 1378  */
 1379 int
 1380 vmaxiosize(struct vnode *vp)
 1381 {
 1382         int maxiosize;
 1383 
 1384         if (vp->v_type == VBLK || vp->v_type == VCHR)
 1385                 maxiosize = vp->v_rdev->si_iosize_max;
 1386         else
 1387                 maxiosize = vp->v_mount->mnt_iosize_max;
 1388 
 1389         if (maxiosize > MAXPHYS)
 1390                 maxiosize = MAXPHYS;
 1391         return (maxiosize);
 1392 }
 1393 
 1394 /*
 1395  * Eliminate all activity associated with a vnode in preparation for
 1396  * destruction.
 1397  *
 1398  * The vnode must be VX locked and refd and will remain VX locked and refd
 1399  * on return.  This routine may be called with the vnode in any state, as
 1400  * long as it is VX locked.  The vnode will be cleaned out and marked
 1401  * VRECLAIMED but will not actually be reused until all existing refs and
 1402  * holds go away.
 1403  *
 1404  * NOTE: This routine may be called on a vnode which has not yet been
 1405  * already been deactivated (VOP_INACTIVE), or on a vnode which has
 1406  * already been reclaimed.
 1407  *
 1408  * This routine is not responsible for placing us back on the freelist. 
 1409  * Instead, it happens automatically when the caller releases the VX lock
 1410  * (assuming there aren't any other references).
 1411  */
 1412 void
 1413 vgone_vxlocked(struct vnode *vp)
 1414 {
 1415         /*
 1416          * assert that the VX lock is held.  This is an absolute requirement
 1417          * now for vgone_vxlocked() to be called.
 1418          */
 1419         KKASSERT(lockcountnb(&vp->v_lock) == 1);
 1420 
 1421         /*
 1422          * Clean out the filesystem specific data and set the VRECLAIMED
 1423          * bit.  Also deactivate the vnode if necessary. 
 1424          *
 1425          * The vnode should have automatically been removed from the syncer
 1426          * list as syncer/dirty flags cleared during the cleaning.
 1427          */
 1428         vclean_vxlocked(vp, DOCLOSE);
 1429         KKASSERT((vp->v_flag & VONWORKLST) == 0);
 1430 
 1431         /*
 1432          * Delete from old mount point vnode list, if on one.
 1433          */
 1434         if (vp->v_mount != NULL) {
 1435                 KKASSERT(vp->v_data == NULL);
 1436                 insmntque(vp, NULL);
 1437         }
 1438 
 1439         /*
 1440          * If special device, remove it from special device alias list
 1441          * if it is on one.  This should normally only occur if a vnode is
 1442          * being revoked as the device should otherwise have been released
 1443          * naturally.
 1444          */
 1445         if ((vp->v_type == VBLK || vp->v_type == VCHR) && vp->v_rdev != NULL) {
 1446                 v_release_rdev(vp);
 1447         }
 1448 
 1449         /*
 1450          * Set us to VBAD
 1451          */
 1452         vp->v_type = VBAD;
 1453 }
 1454 
 1455 /*
 1456  * Lookup a vnode by device number.
 1457  *
 1458  * Returns non-zero and *vpp set to a vref'd vnode on success.
 1459  * Returns zero on failure.
 1460  */
 1461 int
 1462 vfinddev(cdev_t dev, enum vtype type, struct vnode **vpp)
 1463 {
 1464         struct vnode *vp;
 1465 
 1466         lwkt_gettoken(&spechash_token);
 1467         SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
 1468                 if (type == vp->v_type) {
 1469                         *vpp = vp;
 1470                         vref(vp);
 1471                         lwkt_reltoken(&spechash_token);
 1472                         return (1);
 1473                 }
 1474         }
 1475         lwkt_reltoken(&spechash_token);
 1476         return (0);
 1477 }
 1478 
 1479 /*
 1480  * Calculate the total number of references to a special device.  This
 1481  * routine may only be called for VBLK and VCHR vnodes since v_rdev is
 1482  * an overloaded field.  Since udev2dev can now return NULL, we have
 1483  * to check for a NULL v_rdev.
 1484  */
 1485 int
 1486 count_dev(cdev_t dev)
 1487 {
 1488         struct vnode *vp;
 1489         int count = 0;
 1490 
 1491         if (SLIST_FIRST(&dev->si_hlist)) {
 1492                 lwkt_gettoken(&spechash_token);
 1493                 SLIST_FOREACH(vp, &dev->si_hlist, v_cdevnext) {
 1494                         count += vp->v_opencount;
 1495                 }
 1496                 lwkt_reltoken(&spechash_token);
 1497         }
 1498         return(count);
 1499 }
 1500 
 1501 int
 1502 vcount(struct vnode *vp)
 1503 {
 1504         if (vp->v_rdev == NULL)
 1505                 return(0);
 1506         return(count_dev(vp->v_rdev));
 1507 }
 1508 
 1509 /*
 1510  * Initialize VMIO for a vnode.  This routine MUST be called before a
 1511  * VFS can issue buffer cache ops on a vnode.  It is typically called
 1512  * when a vnode is initialized from its inode.
 1513  */
 1514 int
 1515 vinitvmio(struct vnode *vp, off_t filesize, int blksize, int boff)
 1516 {
 1517         vm_object_t object;
 1518         int error = 0;
 1519 
 1520         object = vp->v_object;
 1521         if (object) {
 1522                 vm_object_hold(object);
 1523                 KKASSERT(vp->v_object == object);
 1524         }
 1525 
 1526         if (object == NULL) {
 1527                 object = vnode_pager_alloc(vp, filesize, 0, 0, blksize, boff);
 1528 
 1529                 /*
 1530                  * Dereference the reference we just created.  This assumes
 1531                  * that the object is associated with the vp.  Allow it to
 1532                  * have zero refs.  It cannot be destroyed as long as it
 1533                  * is associated with the vnode.
 1534                  */
 1535                 vm_object_hold(object);
 1536                 atomic_add_int(&object->ref_count, -1);
 1537                 vrele(vp);
 1538         } else {
 1539                 KKASSERT((object->flags & OBJ_DEAD) == 0);
 1540         }
 1541         KASSERT(vp->v_object != NULL, ("vinitvmio: NULL object"));
 1542         vsetflags(vp, VOBJBUF);
 1543         vm_object_drop(object);
 1544 
 1545         return (error);
 1546 }
 1547 
 1548 
 1549 /*
 1550  * Print out a description of a vnode.
 1551  */
 1552 static char *typename[] =
 1553 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
 1554 
 1555 void
 1556 vprint(char *label, struct vnode *vp)
 1557 {
 1558         char buf[96];
 1559 
 1560         if (label != NULL)
 1561                 kprintf("%s: %p: ", label, (void *)vp);
 1562         else
 1563                 kprintf("%p: ", (void *)vp);
 1564         kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,",
 1565                 typename[vp->v_type],
 1566                 vp->v_refcnt, vp->v_writecount, vp->v_auxrefs);
 1567         buf[0] = '\0';
 1568         if (vp->v_flag & VROOT)
 1569                 strcat(buf, "|VROOT");
 1570         if (vp->v_flag & VPFSROOT)
 1571                 strcat(buf, "|VPFSROOT");
 1572         if (vp->v_flag & VTEXT)
 1573                 strcat(buf, "|VTEXT");
 1574         if (vp->v_flag & VSYSTEM)
 1575                 strcat(buf, "|VSYSTEM");
 1576         if (vp->v_flag & VOBJBUF)
 1577                 strcat(buf, "|VOBJBUF");
 1578         if (buf[0] != '\0')
 1579                 kprintf(" flags (%s)", &buf[1]);
 1580         if (vp->v_data == NULL) {
 1581                 kprintf("\n");
 1582         } else {
 1583                 kprintf("\n\t");
 1584                 VOP_PRINT(vp);
 1585         }
 1586 }
 1587 
 1588 /*
 1589  * Do the usual access checking.
 1590  * file_mode, uid and gid are from the vnode in question,
 1591  * while acc_mode and cred are from the VOP_ACCESS parameter list
 1592  */
 1593 int
 1594 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid,
 1595     mode_t acc_mode, struct ucred *cred)
 1596 {
 1597         mode_t mask;
 1598         int ismember;
 1599 
 1600         /*
 1601          * Super-user always gets read/write access, but execute access depends
 1602          * on at least one execute bit being set.
 1603          */
 1604         if (priv_check_cred(cred, PRIV_ROOT, 0) == 0) {
 1605                 if ((acc_mode & VEXEC) && type != VDIR &&
 1606                     (file_mode & (S_IXUSR|S_IXGRP|S_IXOTH)) == 0)
 1607                         return (EACCES);
 1608                 return (0);
 1609         }
 1610 
 1611         mask = 0;
 1612 
 1613         /* Otherwise, check the owner. */
 1614         if (cred->cr_uid == uid) {
 1615                 if (acc_mode & VEXEC)
 1616                         mask |= S_IXUSR;
 1617                 if (acc_mode & VREAD)
 1618                         mask |= S_IRUSR;
 1619                 if (acc_mode & VWRITE)
 1620                         mask |= S_IWUSR;
 1621                 return ((file_mode & mask) == mask ? 0 : EACCES);
 1622         }
 1623 
 1624         /* Otherwise, check the groups. */
 1625         ismember = groupmember(gid, cred);
 1626         if (cred->cr_svgid == gid || ismember) {
 1627                 if (acc_mode & VEXEC)
 1628                         mask |= S_IXGRP;
 1629                 if (acc_mode & VREAD)
 1630                         mask |= S_IRGRP;
 1631                 if (acc_mode & VWRITE)
 1632                         mask |= S_IWGRP;
 1633                 return ((file_mode & mask) == mask ? 0 : EACCES);
 1634         }
 1635 
 1636         /* Otherwise, check everyone else. */
 1637         if (acc_mode & VEXEC)
 1638                 mask |= S_IXOTH;
 1639         if (acc_mode & VREAD)
 1640                 mask |= S_IROTH;
 1641         if (acc_mode & VWRITE)
 1642                 mask |= S_IWOTH;
 1643         return ((file_mode & mask) == mask ? 0 : EACCES);
 1644 }
 1645 
 1646 #ifdef DDB
 1647 #include <ddb/ddb.h>
 1648 
 1649 static int db_show_locked_vnodes(struct mount *mp, void *data);
 1650 
 1651 /*
 1652  * List all of the locked vnodes in the system.
 1653  * Called when debugging the kernel.
 1654  */
 1655 DB_SHOW_COMMAND(lockedvnodes, lockedvnodes)
 1656 {
 1657         kprintf("Locked vnodes\n");
 1658         mountlist_scan(db_show_locked_vnodes, NULL, 
 1659                         MNTSCAN_FORWARD|MNTSCAN_NOBUSY);
 1660 }
 1661 
 1662 static int
 1663 db_show_locked_vnodes(struct mount *mp, void *data __unused)
 1664 {
 1665         struct vnode *vp;
 1666 
 1667         TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 1668                 if (vn_islocked(vp))
 1669                         vprint(NULL, vp);
 1670         }
 1671         return(0);
 1672 }
 1673 #endif
 1674 
 1675 /*
 1676  * Top level filesystem related information gathering.
 1677  */
 1678 static int      sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS);
 1679 
 1680 static int
 1681 vfs_sysctl(SYSCTL_HANDLER_ARGS)
 1682 {
 1683         int *name = (int *)arg1 - 1;    /* XXX */
 1684         u_int namelen = arg2 + 1;       /* XXX */
 1685         struct vfsconf *vfsp;
 1686         int maxtypenum;
 1687 
 1688 #if 1 || defined(COMPAT_PRELITE2)
 1689         /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
 1690         if (namelen == 1)
 1691                 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
 1692 #endif
 1693 
 1694 #ifdef notyet
 1695         /* all sysctl names at this level are at least name and field */
 1696         if (namelen < 2)
 1697                 return (ENOTDIR);               /* overloaded */
 1698         if (name[0] != VFS_GENERIC) {
 1699                 vfsp = vfsconf_find_by_typenum(name[0]);
 1700                 if (vfsp == NULL)
 1701                         return (EOPNOTSUPP);
 1702                 return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1,
 1703                     oldp, oldlenp, newp, newlen, p));
 1704         }
 1705 #endif
 1706         switch (name[1]) {
 1707         case VFS_MAXTYPENUM:
 1708                 if (namelen != 2)
 1709                         return (ENOTDIR);
 1710                 maxtypenum = vfsconf_get_maxtypenum();
 1711                 return (SYSCTL_OUT(req, &maxtypenum, sizeof(maxtypenum)));
 1712         case VFS_CONF:
 1713                 if (namelen != 3)
 1714                         return (ENOTDIR);       /* overloaded */
 1715                 vfsp = vfsconf_find_by_typenum(name[2]);
 1716                 if (vfsp == NULL)
 1717                         return (EOPNOTSUPP);
 1718                 return (SYSCTL_OUT(req, vfsp, sizeof *vfsp));
 1719         }
 1720         return (EOPNOTSUPP);
 1721 }
 1722 
 1723 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD, vfs_sysctl,
 1724         "Generic filesystem");
 1725 
 1726 #if 1 || defined(COMPAT_PRELITE2)
 1727 
 1728 static int
 1729 sysctl_ovfs_conf_iter(struct vfsconf *vfsp, void *data)
 1730 {
 1731         int error;
 1732         struct ovfsconf ovfs;
 1733         struct sysctl_req *req = (struct sysctl_req*) data;
 1734 
 1735         bzero(&ovfs, sizeof(ovfs));
 1736         ovfs.vfc_vfsops = vfsp->vfc_vfsops;     /* XXX used as flag */
 1737         strcpy(ovfs.vfc_name, vfsp->vfc_name);
 1738         ovfs.vfc_index = vfsp->vfc_typenum;
 1739         ovfs.vfc_refcount = vfsp->vfc_refcount;
 1740         ovfs.vfc_flags = vfsp->vfc_flags;
 1741         error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
 1742         if (error)
 1743                 return error; /* abort iteration with error code */
 1744         else
 1745                 return 0; /* continue iterating with next element */
 1746 }
 1747 
 1748 static int
 1749 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
 1750 {
 1751         return vfsconf_each(sysctl_ovfs_conf_iter, (void*)req);
 1752 }
 1753 
 1754 #endif /* 1 || COMPAT_PRELITE2 */
 1755 
 1756 /*
 1757  * Check to see if a filesystem is mounted on a block device.
 1758  */
 1759 int
 1760 vfs_mountedon(struct vnode *vp)
 1761 {
 1762         cdev_t dev;
 1763 
 1764         if ((dev = vp->v_rdev) == NULL) {
 1765 /*              if (vp->v_type != VBLK)
 1766                         dev = get_dev(vp->v_uminor, vp->v_umajor); */
 1767         }
 1768         if (dev != NULL && dev->si_mountpoint)
 1769                 return (EBUSY);
 1770         return (0);
 1771 }
 1772 
 1773 /*
 1774  * Unmount all filesystems. The list is traversed in reverse order
 1775  * of mounting to avoid dependencies.
 1776  */
 1777 
 1778 static int vfs_umountall_callback(struct mount *mp, void *data);
 1779 
 1780 void
 1781 vfs_unmountall(void)
 1782 {
 1783         int count;
 1784 
 1785         do {
 1786                 count = mountlist_scan(vfs_umountall_callback, 
 1787                                         NULL, MNTSCAN_REVERSE|MNTSCAN_NOBUSY);
 1788         } while (count);
 1789 }
 1790 
 1791 static
 1792 int
 1793 vfs_umountall_callback(struct mount *mp, void *data)
 1794 {
 1795         int error;
 1796 
 1797         error = dounmount(mp, MNT_FORCE);
 1798         if (error) {
 1799                 mountlist_remove(mp);
 1800                 kprintf("unmount of filesystem mounted from %s failed (", 
 1801                         mp->mnt_stat.f_mntfromname);
 1802                 if (error == EBUSY)
 1803                         kprintf("BUSY)\n");
 1804                 else
 1805                         kprintf("%d)\n", error);
 1806         }
 1807         return(1);
 1808 }
 1809 
 1810 /*
 1811  * Checks the mount flags for parameter mp and put the names comma-separated
 1812  * into a string buffer buf with a size limit specified by len.
 1813  *
 1814  * It returns the number of bytes written into buf, and (*errorp) will be
 1815  * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
 1816  * not large enough).  The buffer will be 0-terminated if len was not 0.
 1817  */
 1818 size_t
 1819 vfs_flagstostr(int flags, const struct mountctl_opt *optp,
 1820                char *buf, size_t len, int *errorp)
 1821 {
 1822         static const struct mountctl_opt optnames[] = {
 1823                 { MNT_ASYNC,            "asynchronous" },
 1824                 { MNT_EXPORTED,         "NFS exported" },
 1825                 { MNT_LOCAL,            "local" },
 1826                 { MNT_NOATIME,          "noatime" },
 1827                 { MNT_NODEV,            "nodev" },
 1828                 { MNT_NOEXEC,           "noexec" },
 1829                 { MNT_NOSUID,           "nosuid" },
 1830                 { MNT_NOSYMFOLLOW,      "nosymfollow" },
 1831                 { MNT_QUOTA,            "with-quotas" },
 1832                 { MNT_RDONLY,           "read-only" },
 1833                 { MNT_SYNCHRONOUS,      "synchronous" },
 1834                 { MNT_UNION,            "union" },
 1835                 { MNT_NOCLUSTERR,       "noclusterr" },
 1836                 { MNT_NOCLUSTERW,       "noclusterw" },
 1837                 { MNT_SUIDDIR,          "suiddir" },
 1838                 { MNT_SOFTDEP,          "soft-updates" },
 1839                 { MNT_IGNORE,           "ignore" },
 1840                 { 0,                    NULL}
 1841         };
 1842         int bwritten;
 1843         int bleft;
 1844         int optlen;
 1845         int actsize;
 1846 
 1847         *errorp = 0;
 1848         bwritten = 0;
 1849         bleft = len - 1;        /* leave room for trailing \0 */
 1850 
 1851         /*
 1852          * Checks the size of the string. If it contains
 1853          * any data, then we will append the new flags to
 1854          * it.
 1855          */
 1856         actsize = strlen(buf);
 1857         if (actsize > 0)
 1858                 buf += actsize;
 1859 
 1860         /* Default flags if no flags passed */
 1861         if (optp == NULL)
 1862                 optp = optnames;
 1863 
 1864         if (bleft < 0) {        /* degenerate case, 0-length buffer */
 1865                 *errorp = EINVAL;
 1866                 return(0);
 1867         }
 1868 
 1869         for (; flags && optp->o_opt; ++optp) {
 1870                 if ((flags & optp->o_opt) == 0)
 1871                         continue;
 1872                 optlen = strlen(optp->o_name);
 1873                 if (bwritten || actsize > 0) {
 1874                         if (bleft < 2) {
 1875                                 *errorp = ENOSPC;
 1876                                 break;
 1877                         }
 1878                         buf[bwritten++] = ',';
 1879                         buf[bwritten++] = ' ';
 1880                         bleft -= 2;
 1881                 }
 1882                 if (bleft < optlen) {
 1883                         *errorp = ENOSPC;
 1884                         break;
 1885                 }
 1886                 bcopy(optp->o_name, buf + bwritten, optlen);
 1887                 bwritten += optlen;
 1888                 bleft -= optlen;
 1889                 flags &= ~optp->o_opt;
 1890         }
 1891 
 1892         /*
 1893          * Space already reserved for trailing \0
 1894          */
 1895         buf[bwritten] = 0;
 1896         return (bwritten);
 1897 }
 1898 
 1899 /*
 1900  * Build hash lists of net addresses and hang them off the mount point.
 1901  * Called by ufs_mount() to set up the lists of export addresses.
 1902  */
 1903 static int
 1904 vfs_hang_addrlist(struct mount *mp, struct netexport *nep,
 1905                 const struct export_args *argp)
 1906 {
 1907         struct netcred *np;
 1908         struct radix_node_head *rnh;
 1909         int i;
 1910         struct radix_node *rn;
 1911         struct sockaddr *saddr, *smask = NULL;
 1912         struct domain *dom;
 1913         int error;
 1914 
 1915         if (argp->ex_addrlen == 0) {
 1916                 if (mp->mnt_flag & MNT_DEFEXPORTED)
 1917                         return (EPERM);
 1918                 np = &nep->ne_defexported;
 1919                 np->netc_exflags = argp->ex_flags;
 1920                 np->netc_anon = argp->ex_anon;
 1921                 np->netc_anon.cr_ref = 1;
 1922                 mp->mnt_flag |= MNT_DEFEXPORTED;
 1923                 return (0);
 1924         }
 1925 
 1926         if (argp->ex_addrlen < 0 || argp->ex_addrlen > MLEN)
 1927                 return (EINVAL);
 1928         if (argp->ex_masklen < 0 || argp->ex_masklen > MLEN)
 1929                 return (EINVAL);
 1930 
 1931         i = sizeof(struct netcred) + argp->ex_addrlen + argp->ex_masklen;
 1932         np = (struct netcred *) kmalloc(i, M_NETADDR, M_WAITOK | M_ZERO);
 1933         saddr = (struct sockaddr *) (np + 1);
 1934         if ((error = copyin(argp->ex_addr, (caddr_t) saddr, argp->ex_addrlen)))
 1935                 goto out;
 1936         if (saddr->sa_len > argp->ex_addrlen)
 1937                 saddr->sa_len = argp->ex_addrlen;
 1938         if (argp->ex_masklen) {
 1939                 smask = (struct sockaddr *)((caddr_t)saddr + argp->ex_addrlen);
 1940                 error = copyin(argp->ex_mask, (caddr_t)smask, argp->ex_masklen);
 1941                 if (error)
 1942                         goto out;
 1943                 if (smask->sa_len > argp->ex_masklen)
 1944                         smask->sa_len = argp->ex_masklen;
 1945         }
 1946         i = saddr->sa_family;
 1947         if ((rnh = nep->ne_rtable[i]) == NULL) {
 1948                 /*
 1949                  * Seems silly to initialize every AF when most are not used,
 1950                  * do so on demand here
 1951                  */
 1952                 SLIST_FOREACH(dom, &domains, dom_next)
 1953                         if (dom->dom_family == i && dom->dom_rtattach) {
 1954                                 dom->dom_rtattach((void **) &nep->ne_rtable[i],
 1955                                     dom->dom_rtoffset);
 1956                                 break;
 1957                         }
 1958                 if ((rnh = nep->ne_rtable[i]) == NULL) {
 1959                         error = ENOBUFS;
 1960                         goto out;
 1961                 }
 1962         }
 1963         rn = (*rnh->rnh_addaddr) ((char *) saddr, (char *) smask, rnh,
 1964             np->netc_rnodes);
 1965         if (rn == NULL || np != (struct netcred *) rn) {        /* already exists */
 1966                 error = EPERM;
 1967                 goto out;
 1968         }
 1969         np->netc_exflags = argp->ex_flags;
 1970         np->netc_anon = argp->ex_anon;
 1971         np->netc_anon.cr_ref = 1;
 1972         return (0);
 1973 out:
 1974         kfree(np, M_NETADDR);
 1975         return (error);
 1976 }
 1977 
 1978 /* ARGSUSED */
 1979 static int
 1980 vfs_free_netcred(struct radix_node *rn, void *w)
 1981 {
 1982         struct radix_node_head *rnh = (struct radix_node_head *) w;
 1983 
 1984         (*rnh->rnh_deladdr) (rn->rn_key, rn->rn_mask, rnh);
 1985         kfree((caddr_t) rn, M_NETADDR);
 1986         return (0);
 1987 }
 1988 
 1989 /*
 1990  * Free the net address hash lists that are hanging off the mount points.
 1991  */
 1992 static void
 1993 vfs_free_addrlist(struct netexport *nep)
 1994 {
 1995         int i;
 1996         struct radix_node_head *rnh;
 1997 
 1998         for (i = 0; i <= AF_MAX; i++)
 1999                 if ((rnh = nep->ne_rtable[i])) {
 2000                         (*rnh->rnh_walktree) (rnh, vfs_free_netcred,
 2001                             (caddr_t) rnh);
 2002                         kfree((caddr_t) rnh, M_RTABLE);
 2003                         nep->ne_rtable[i] = 0;
 2004                 }
 2005 }
 2006 
 2007 int
 2008 vfs_export(struct mount *mp, struct netexport *nep,
 2009            const struct export_args *argp)
 2010 {
 2011         int error;
 2012 
 2013         if (argp->ex_flags & MNT_DELEXPORT) {
 2014                 if (mp->mnt_flag & MNT_EXPUBLIC) {
 2015                         vfs_setpublicfs(NULL, NULL, NULL);
 2016                         mp->mnt_flag &= ~MNT_EXPUBLIC;
 2017                 }
 2018                 vfs_free_addrlist(nep);
 2019                 mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED);
 2020         }
 2021         if (argp->ex_flags & MNT_EXPORTED) {
 2022                 if (argp->ex_flags & MNT_EXPUBLIC) {
 2023                         if ((error = vfs_setpublicfs(mp, nep, argp)) != 0)
 2024                                 return (error);
 2025                         mp->mnt_flag |= MNT_EXPUBLIC;
 2026                 }
 2027                 if ((error = vfs_hang_addrlist(mp, nep, argp)))
 2028                         return (error);
 2029                 mp->mnt_flag |= MNT_EXPORTED;
 2030         }
 2031         return (0);
 2032 }
 2033 
 2034 
 2035 /*
 2036  * Set the publicly exported filesystem (WebNFS). Currently, only
 2037  * one public filesystem is possible in the spec (RFC 2054 and 2055)
 2038  */
 2039 int
 2040 vfs_setpublicfs(struct mount *mp, struct netexport *nep,
 2041                 const struct export_args *argp)
 2042 {
 2043         int error;
 2044         struct vnode *rvp;
 2045         char *cp;
 2046 
 2047         /*
 2048          * mp == NULL -> invalidate the current info, the FS is
 2049          * no longer exported. May be called from either vfs_export
 2050          * or unmount, so check if it hasn't already been done.
 2051          */
 2052         if (mp == NULL) {
 2053                 if (nfs_pub.np_valid) {
 2054                         nfs_pub.np_valid = 0;
 2055                         if (nfs_pub.np_index != NULL) {
 2056                                 kfree(nfs_pub.np_index, M_TEMP);
 2057                                 nfs_pub.np_index = NULL;
 2058                         }
 2059                 }
 2060                 return (0);
 2061         }
 2062 
 2063         /*
 2064          * Only one allowed at a time.
 2065          */
 2066         if (nfs_pub.np_valid != 0 && mp != nfs_pub.np_mount)
 2067                 return (EBUSY);
 2068 
 2069         /*
 2070          * Get real filehandle for root of exported FS.
 2071          */
 2072         bzero((caddr_t)&nfs_pub.np_handle, sizeof(nfs_pub.np_handle));
 2073         nfs_pub.np_handle.fh_fsid = mp->mnt_stat.f_fsid;
 2074 
 2075         if ((error = VFS_ROOT(mp, &rvp)))
 2076                 return (error);
 2077 
 2078         if ((error = VFS_VPTOFH(rvp, &nfs_pub.np_handle.fh_fid)))
 2079                 return (error);
 2080 
 2081         vput(rvp);
 2082 
 2083         /*
 2084          * If an indexfile was specified, pull it in.
 2085          */
 2086         if (argp->ex_indexfile != NULL) {
 2087                 int namelen;
 2088 
 2089                 error = vn_get_namelen(rvp, &namelen);
 2090                 if (error)
 2091                         return (error);
 2092                 nfs_pub.np_index = kmalloc(namelen, M_TEMP, M_WAITOK);
 2093                 error = copyinstr(argp->ex_indexfile, nfs_pub.np_index,
 2094                     namelen, NULL);
 2095                 if (!error) {
 2096                         /*
 2097                          * Check for illegal filenames.
 2098                          */
 2099                         for (cp = nfs_pub.np_index; *cp; cp++) {
 2100                                 if (*cp == '/') {
 2101                                         error = EINVAL;
 2102                                         break;
 2103                                 }
 2104                         }
 2105                 }
 2106                 if (error) {
 2107                         kfree(nfs_pub.np_index, M_TEMP);
 2108                         return (error);
 2109                 }
 2110         }
 2111 
 2112         nfs_pub.np_mount = mp;
 2113         nfs_pub.np_valid = 1;
 2114         return (0);
 2115 }
 2116 
 2117 struct netcred *
 2118 vfs_export_lookup(struct mount *mp, struct netexport *nep,
 2119                 struct sockaddr *nam)
 2120 {
 2121         struct netcred *np;
 2122         struct radix_node_head *rnh;
 2123         struct sockaddr *saddr;
 2124 
 2125         np = NULL;
 2126         if (mp->mnt_flag & MNT_EXPORTED) {
 2127                 /*
 2128                  * Lookup in the export list first.
 2129                  */
 2130                 if (nam != NULL) {
 2131                         saddr = nam;
 2132                         rnh = nep->ne_rtable[saddr->sa_family];
 2133                         if (rnh != NULL) {
 2134                                 np = (struct netcred *)
 2135                                         (*rnh->rnh_matchaddr)((char *)saddr,
 2136                                                               rnh);
 2137                                 if (np && np->netc_rnodes->rn_flags & RNF_ROOT)
 2138                                         np = NULL;
 2139                         }
 2140                 }
 2141                 /*
 2142                  * If no address match, use the default if it exists.
 2143                  */
 2144                 if (np == NULL && mp->mnt_flag & MNT_DEFEXPORTED)
 2145                         np = &nep->ne_defexported;
 2146         }
 2147         return (np);
 2148 }
 2149 
 2150 /*
 2151  * perform msync on all vnodes under a mount point.  The mount point must
 2152  * be locked.  This code is also responsible for lazy-freeing unreferenced
 2153  * vnodes whos VM objects no longer contain pages.
 2154  *
 2155  * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
 2156  *
 2157  * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
 2158  * but vnode_pager_putpages() doesn't lock the vnode.  We have to do it
 2159  * way up in this high level function.
 2160  */
 2161 static int vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data);
 2162 static int vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data);
 2163 
 2164 void
 2165 vfs_msync(struct mount *mp, int flags) 
 2166 {
 2167         int vmsc_flags;
 2168 
 2169         /*
 2170          * tmpfs sets this flag to prevent msync(), sync, and the
 2171          * filesystem periodic syncer from trying to flush VM pages
 2172          * to swap.  Only pure memory pressure flushes tmpfs VM pages
 2173          * to swap.
 2174          */
 2175         if (mp->mnt_kern_flag & MNTK_NOMSYNC)
 2176                 return;
 2177 
 2178         /*
 2179          * Ok, scan the vnodes for work.  If the filesystem is using the
 2180          * syncer thread feature we can use vsyncscan() instead of
 2181          * vmntvnodescan(), which is much faster.
 2182          */
 2183         vmsc_flags = VMSC_GETVP;
 2184         if (flags != MNT_WAIT)
 2185                 vmsc_flags |= VMSC_NOWAIT;
 2186 
 2187         if (mp->mnt_kern_flag & MNTK_THR_SYNC) {
 2188                 vsyncscan(mp, vmsc_flags, vfs_msync_scan2,
 2189                           (void *)(intptr_t)flags);
 2190         } else {
 2191                 vmntvnodescan(mp, vmsc_flags,
 2192                               vfs_msync_scan1, vfs_msync_scan2,
 2193                               (void *)(intptr_t)flags);
 2194         }
 2195 }
 2196 
 2197 /*
 2198  * scan1 is a fast pre-check.  There could be hundreds of thousands of
 2199  * vnodes, we cannot afford to do anything heavy weight until we have a
 2200  * fairly good indication that there is work to do.
 2201  */
 2202 static
 2203 int
 2204 vfs_msync_scan1(struct mount *mp, struct vnode *vp, void *data)
 2205 {
 2206         int flags = (int)(intptr_t)data;
 2207 
 2208         if ((vp->v_flag & VRECLAIMED) == 0) {
 2209                 if (vp->v_auxrefs == 0 && VREFCNT(vp) <= 0 &&
 2210                     vp->v_object) {
 2211                         return(0);      /* call scan2 */
 2212                 }
 2213                 if ((mp->mnt_flag & MNT_RDONLY) == 0 &&
 2214                     (vp->v_flag & VOBJDIRTY) &&
 2215                     (flags == MNT_WAIT || vn_islocked(vp) == 0)) {
 2216                         return(0);      /* call scan2 */
 2217                 }
 2218         }
 2219 
 2220         /*
 2221          * do not call scan2, continue the loop
 2222          */
 2223         return(-1);
 2224 }
 2225 
 2226 /*
 2227  * This callback is handed a locked vnode.
 2228  */
 2229 static
 2230 int
 2231 vfs_msync_scan2(struct mount *mp, struct vnode *vp, void *data)
 2232 {
 2233         vm_object_t obj;
 2234         int flags = (int)(intptr_t)data;
 2235 
 2236         if (vp->v_flag & VRECLAIMED)
 2237                 return(0);
 2238 
 2239         if ((mp->mnt_flag & MNT_RDONLY) == 0 && (vp->v_flag & VOBJDIRTY)) {
 2240                 if ((obj = vp->v_object) != NULL) {
 2241                         vm_object_page_clean(obj, 0, 0, 
 2242                          flags == MNT_WAIT ? OBJPC_SYNC : OBJPC_NOSYNC);
 2243                 }
 2244         }
 2245         return(0);
 2246 }
 2247 
 2248 /*
 2249  * Wake up anyone interested in vp because it is being revoked.
 2250  */
 2251 void
 2252 vn_gone(struct vnode *vp)
 2253 {
 2254         lwkt_gettoken(&vp->v_token);
 2255         KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, NOTE_REVOKE);
 2256         lwkt_reltoken(&vp->v_token);
 2257 }
 2258 
 2259 /*
 2260  * extract the cdev_t from a VBLK or VCHR.  The vnode must have been opened
 2261  * (or v_rdev might be NULL).
 2262  */
 2263 cdev_t
 2264 vn_todev(struct vnode *vp)
 2265 {
 2266         if (vp->v_type != VBLK && vp->v_type != VCHR)
 2267                 return (NULL);
 2268         KKASSERT(vp->v_rdev != NULL);
 2269         return (vp->v_rdev);
 2270 }
 2271 
 2272 /*
 2273  * Check if vnode represents a disk device.  The vnode does not need to be
 2274  * opened.
 2275  *
 2276  * MPALMOSTSAFE
 2277  */
 2278 int
 2279 vn_isdisk(struct vnode *vp, int *errp)
 2280 {
 2281         cdev_t dev;
 2282 
 2283         if (vp->v_type != VCHR) {
 2284                 if (errp != NULL)
 2285                         *errp = ENOTBLK;
 2286                 return (0);
 2287         }
 2288 
 2289         dev = vp->v_rdev;
 2290 
 2291         if (dev == NULL) {
 2292                 if (errp != NULL)
 2293                         *errp = ENXIO;
 2294                 return (0);
 2295         }
 2296         if (dev_is_good(dev) == 0) {
 2297                 if (errp != NULL)
 2298                         *errp = ENXIO;
 2299                 return (0);
 2300         }
 2301         if ((dev_dflags(dev) & D_DISK) == 0) {
 2302                 if (errp != NULL)
 2303                         *errp = ENOTBLK;
 2304                 return (0);
 2305         }
 2306         if (errp != NULL)
 2307                 *errp = 0;
 2308         return (1);
 2309 }
 2310 
 2311 int
 2312 vn_get_namelen(struct vnode *vp, int *namelen)
 2313 {
 2314         int error;
 2315         register_t retval[2];
 2316 
 2317         error = VOP_PATHCONF(vp, _PC_NAME_MAX, retval);
 2318         if (error)
 2319                 return (error);
 2320         *namelen = (int)retval[0];
 2321         return (0);
 2322 }
 2323 
 2324 int
 2325 vop_write_dirent(int *error, struct uio *uio, ino_t d_ino, uint8_t d_type, 
 2326                 uint16_t d_namlen, const char *d_name)
 2327 {
 2328         struct dirent *dp;
 2329         size_t len;
 2330 
 2331         len = _DIRENT_RECLEN(d_namlen);
 2332         if (len > uio->uio_resid)
 2333                 return(1);
 2334 
 2335         dp = kmalloc(len, M_TEMP, M_WAITOK | M_ZERO);
 2336 
 2337         dp->d_ino = d_ino;
 2338         dp->d_namlen = d_namlen;
 2339         dp->d_type = d_type;
 2340         bcopy(d_name, dp->d_name, d_namlen);
 2341 
 2342         *error = uiomove((caddr_t)dp, len, uio);
 2343 
 2344         kfree(dp, M_TEMP);
 2345 
 2346         return(0);
 2347 }
 2348 
 2349 void
 2350 vn_mark_atime(struct vnode *vp, struct thread *td)
 2351 {
 2352         struct proc *p = td->td_proc;
 2353         struct ucred *cred = p ? p->p_ucred : proc0.p_ucred;
 2354 
 2355         if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
 2356                 VOP_MARKATIME(vp, cred);
 2357         }
 2358 }

Cache object: 33e21ea0bc28117d2299e985d8a42768


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.