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/ufs/ffs/ffs_alloc.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2002 Networks Associates Technology, Inc.
    3  * All rights reserved.
    4  *
    5  * This software was developed for the FreeBSD Project by Marshall
    6  * Kirk McKusick and Network Associates Laboratories, the Security
    7  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
    8  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
    9  * research program
   10  *
   11  * Redistribution and use in source and binary forms, with or without
   12  * modification, are permitted provided that the following conditions
   13  * are met:
   14  * 1. Redistributions of source code must retain the above copyright
   15  *    notice, this list of conditions and the following disclaimer.
   16  * 2. Redistributions in binary form must reproduce the above copyright
   17  *    notice, this list of conditions and the following disclaimer in the
   18  *    documentation and/or other materials provided with the distribution.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  *
   32  * Copyright (c) 1982, 1986, 1989, 1993
   33  *      The Regents of the University of California.  All rights reserved.
   34  *
   35  * Redistribution and use in source and binary forms, with or without
   36  * modification, are permitted provided that the following conditions
   37  * are met:
   38  * 1. Redistributions of source code must retain the above copyright
   39  *    notice, this list of conditions and the following disclaimer.
   40  * 2. Redistributions in binary form must reproduce the above copyright
   41  *    notice, this list of conditions and the following disclaimer in the
   42  *    documentation and/or other materials provided with the distribution.
   43  * 4. Neither the name of the University nor the names of its contributors
   44  *    may be used to endorse or promote products derived from this software
   45  *    without specific prior written permission.
   46  *
   47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   57  * SUCH DAMAGE.
   58  *
   59  *      @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95
   60  */
   61 
   62 #include <sys/cdefs.h>
   63 __FBSDID("$FreeBSD: releng/8.0/sys/ufs/ffs/ffs_alloc.c 192260 2009-05-17 20:26:00Z alc $");
   64 
   65 #include "opt_quota.h"
   66 
   67 #include <sys/param.h>
   68 #include <sys/systm.h>
   69 #include <sys/bio.h>
   70 #include <sys/buf.h>
   71 #include <sys/conf.h>
   72 #include <sys/file.h>
   73 #include <sys/filedesc.h>
   74 #include <sys/priv.h>
   75 #include <sys/proc.h>
   76 #include <sys/vnode.h>
   77 #include <sys/mount.h>
   78 #include <sys/kernel.h>
   79 #include <sys/sysctl.h>
   80 #include <sys/syslog.h>
   81 
   82 #include <ufs/ufs/extattr.h>
   83 #include <ufs/ufs/quota.h>
   84 #include <ufs/ufs/inode.h>
   85 #include <ufs/ufs/ufs_extern.h>
   86 #include <ufs/ufs/ufsmount.h>
   87 
   88 #include <ufs/ffs/fs.h>
   89 #include <ufs/ffs/ffs_extern.h>
   90 
   91 typedef ufs2_daddr_t allocfcn_t(struct inode *ip, int cg, ufs2_daddr_t bpref,
   92                                   int size);
   93 
   94 static ufs2_daddr_t ffs_alloccg(struct inode *, int, ufs2_daddr_t, int);
   95 static ufs2_daddr_t
   96               ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t);
   97 #ifdef INVARIANTS
   98 static int      ffs_checkblk(struct inode *, ufs2_daddr_t, long);
   99 #endif
  100 static ufs2_daddr_t ffs_clusteralloc(struct inode *, int, ufs2_daddr_t, int);
  101 static void     ffs_clusteracct(struct ufsmount *, struct fs *, struct cg *,
  102                     ufs1_daddr_t, int);
  103 static ino_t    ffs_dirpref(struct inode *);
  104 static ufs2_daddr_t ffs_fragextend(struct inode *, int, ufs2_daddr_t, int, int);
  105 static void     ffs_fserr(struct fs *, ino_t, char *);
  106 static ufs2_daddr_t     ffs_hashalloc
  107                 (struct inode *, int, ufs2_daddr_t, int, allocfcn_t *);
  108 static ufs2_daddr_t ffs_nodealloccg(struct inode *, int, ufs2_daddr_t, int);
  109 static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int);
  110 static int      ffs_reallocblks_ufs1(struct vop_reallocblks_args *);
  111 static int      ffs_reallocblks_ufs2(struct vop_reallocblks_args *);
  112 
  113 /*
  114  * Allocate a block in the filesystem.
  115  *
  116  * The size of the requested block is given, which must be some
  117  * multiple of fs_fsize and <= fs_bsize.
  118  * A preference may be optionally specified. If a preference is given
  119  * the following hierarchy is used to allocate a block:
  120  *   1) allocate the requested block.
  121  *   2) allocate a rotationally optimal block in the same cylinder.
  122  *   3) allocate a block in the same cylinder group.
  123  *   4) quadradically rehash into other cylinder groups, until an
  124  *      available block is located.
  125  * If no block preference is given the following hierarchy is used
  126  * to allocate a block:
  127  *   1) allocate a block in the cylinder group that contains the
  128  *      inode for the file.
  129  *   2) quadradically rehash into other cylinder groups, until an
  130  *      available block is located.
  131  */
  132 int
  133 ffs_alloc(ip, lbn, bpref, size, flags, cred, bnp)
  134         struct inode *ip;
  135         ufs2_daddr_t lbn, bpref;
  136         int size, flags;
  137         struct ucred *cred;
  138         ufs2_daddr_t *bnp;
  139 {
  140         struct fs *fs;
  141         struct ufsmount *ump;
  142         ufs2_daddr_t bno;
  143         int cg, reclaimed;
  144         static struct timeval lastfail;
  145         static int curfail;
  146         int64_t delta;
  147 #ifdef QUOTA
  148         int error;
  149 #endif
  150 
  151         *bnp = 0;
  152         fs = ip->i_fs;
  153         ump = ip->i_ump;
  154         mtx_assert(UFS_MTX(ump), MA_OWNED);
  155 #ifdef INVARIANTS
  156         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
  157                 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n",
  158                     devtoname(ip->i_dev), (long)fs->fs_bsize, size,
  159                     fs->fs_fsmnt);
  160                 panic("ffs_alloc: bad size");
  161         }
  162         if (cred == NOCRED)
  163                 panic("ffs_alloc: missing credential");
  164 #endif /* INVARIANTS */
  165         reclaimed = 0;
  166 retry:
  167 #ifdef QUOTA
  168         UFS_UNLOCK(ump);
  169         error = chkdq(ip, btodb(size), cred, 0);
  170         if (error)
  171                 return (error);
  172         UFS_LOCK(ump);
  173 #endif
  174         if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
  175                 goto nospace;
  176         if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) &&
  177             freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
  178                 goto nospace;
  179         if (bpref >= fs->fs_size)
  180                 bpref = 0;
  181         if (bpref == 0)
  182                 cg = ino_to_cg(fs, ip->i_number);
  183         else
  184                 cg = dtog(fs, bpref);
  185         bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
  186         if (bno > 0) {
  187                 delta = btodb(size);
  188                 if (ip->i_flag & IN_SPACECOUNTED) {
  189                         UFS_LOCK(ump);
  190                         fs->fs_pendingblocks += delta;
  191                         UFS_UNLOCK(ump);
  192                 }
  193                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  194                 if (flags & IO_EXT)
  195                         ip->i_flag |= IN_CHANGE;
  196                 else
  197                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
  198                 *bnp = bno;
  199                 return (0);
  200         }
  201 nospace:
  202 #ifdef QUOTA
  203         UFS_UNLOCK(ump);
  204         /*
  205          * Restore user's disk quota because allocation failed.
  206          */
  207         (void) chkdq(ip, -btodb(size), cred, FORCE);
  208         UFS_LOCK(ump);
  209 #endif
  210         if (fs->fs_pendingblocks > 0 && reclaimed == 0) {
  211                 reclaimed = 1;
  212                 softdep_request_cleanup(fs, ITOV(ip));
  213                 goto retry;
  214         }
  215         UFS_UNLOCK(ump);
  216         if (ppsratecheck(&lastfail, &curfail, 1)) {
  217                 ffs_fserr(fs, ip->i_number, "filesystem full");
  218                 uprintf("\n%s: write failed, filesystem is full\n",
  219                     fs->fs_fsmnt);
  220         }
  221         return (ENOSPC);
  222 }
  223 
  224 /*
  225  * Reallocate a fragment to a bigger size
  226  *
  227  * The number and size of the old block is given, and a preference
  228  * and new size is also specified. The allocator attempts to extend
  229  * the original block. Failing that, the regular block allocator is
  230  * invoked to get an appropriate block.
  231  */
  232 int
  233 ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, flags, cred, bpp)
  234         struct inode *ip;
  235         ufs2_daddr_t lbprev;
  236         ufs2_daddr_t bprev;
  237         ufs2_daddr_t bpref;
  238         int osize, nsize, flags;
  239         struct ucred *cred;
  240         struct buf **bpp;
  241 {
  242         struct vnode *vp;
  243         struct fs *fs;
  244         struct buf *bp;
  245         struct ufsmount *ump;
  246         int cg, request, error, reclaimed;
  247         ufs2_daddr_t bno;
  248         static struct timeval lastfail;
  249         static int curfail;
  250         int64_t delta;
  251 
  252         *bpp = 0;
  253         vp = ITOV(ip);
  254         fs = ip->i_fs;
  255         bp = NULL;
  256         ump = ip->i_ump;
  257         mtx_assert(UFS_MTX(ump), MA_OWNED);
  258 #ifdef INVARIANTS
  259         if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
  260                 panic("ffs_realloccg: allocation on suspended filesystem");
  261         if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
  262             (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
  263                 printf(
  264                 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
  265                     devtoname(ip->i_dev), (long)fs->fs_bsize, osize,
  266                     nsize, fs->fs_fsmnt);
  267                 panic("ffs_realloccg: bad size");
  268         }
  269         if (cred == NOCRED)
  270                 panic("ffs_realloccg: missing credential");
  271 #endif /* INVARIANTS */
  272         reclaimed = 0;
  273 retry:
  274         if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) &&
  275             freespace(fs, fs->fs_minfree) -  numfrags(fs, nsize - osize) < 0) {
  276                 goto nospace;
  277         }
  278         if (bprev == 0) {
  279                 printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n",
  280                     devtoname(ip->i_dev), (long)fs->fs_bsize, (intmax_t)bprev,
  281                     fs->fs_fsmnt);
  282                 panic("ffs_realloccg: bad bprev");
  283         }
  284         UFS_UNLOCK(ump);
  285         /*
  286          * Allocate the extra space in the buffer.
  287          */
  288         error = bread(vp, lbprev, osize, NOCRED, &bp);
  289         if (error) {
  290                 brelse(bp);
  291                 return (error);
  292         }
  293 
  294         if (bp->b_blkno == bp->b_lblkno) {
  295                 if (lbprev >= NDADDR)
  296                         panic("ffs_realloccg: lbprev out of range");
  297                 bp->b_blkno = fsbtodb(fs, bprev);
  298         }
  299 
  300 #ifdef QUOTA
  301         error = chkdq(ip, btodb(nsize - osize), cred, 0);
  302         if (error) {
  303                 brelse(bp);
  304                 return (error);
  305         }
  306 #endif
  307         /*
  308          * Check for extension in the existing location.
  309          */
  310         cg = dtog(fs, bprev);
  311         UFS_LOCK(ump);
  312         bno = ffs_fragextend(ip, cg, bprev, osize, nsize);
  313         if (bno) {
  314                 if (bp->b_blkno != fsbtodb(fs, bno))
  315                         panic("ffs_realloccg: bad blockno");
  316                 delta = btodb(nsize - osize);
  317                 if (ip->i_flag & IN_SPACECOUNTED) {
  318                         UFS_LOCK(ump);
  319                         fs->fs_pendingblocks += delta;
  320                         UFS_UNLOCK(ump);
  321                 }
  322                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  323                 if (flags & IO_EXT)
  324                         ip->i_flag |= IN_CHANGE;
  325                 else
  326                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
  327                 allocbuf(bp, nsize);
  328                 bp->b_flags |= B_DONE;
  329                 bzero(bp->b_data + osize, nsize - osize);
  330                 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO)
  331                         vfs_bio_set_valid(bp, osize, nsize - osize);
  332                 *bpp = bp;
  333                 return (0);
  334         }
  335         /*
  336          * Allocate a new disk location.
  337          */
  338         if (bpref >= fs->fs_size)
  339                 bpref = 0;
  340         switch ((int)fs->fs_optim) {
  341         case FS_OPTSPACE:
  342                 /*
  343                  * Allocate an exact sized fragment. Although this makes
  344                  * best use of space, we will waste time relocating it if
  345                  * the file continues to grow. If the fragmentation is
  346                  * less than half of the minimum free reserve, we choose
  347                  * to begin optimizing for time.
  348                  */
  349                 request = nsize;
  350                 if (fs->fs_minfree <= 5 ||
  351                     fs->fs_cstotal.cs_nffree >
  352                     (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100))
  353                         break;
  354                 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
  355                         fs->fs_fsmnt);
  356                 fs->fs_optim = FS_OPTTIME;
  357                 break;
  358         case FS_OPTTIME:
  359                 /*
  360                  * At this point we have discovered a file that is trying to
  361                  * grow a small fragment to a larger fragment. To save time,
  362                  * we allocate a full sized block, then free the unused portion.
  363                  * If the file continues to grow, the `ffs_fragextend' call
  364                  * above will be able to grow it in place without further
  365                  * copying. If aberrant programs cause disk fragmentation to
  366                  * grow within 2% of the free reserve, we choose to begin
  367                  * optimizing for space.
  368                  */
  369                 request = fs->fs_bsize;
  370                 if (fs->fs_cstotal.cs_nffree <
  371                     (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100)
  372                         break;
  373                 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
  374                         fs->fs_fsmnt);
  375                 fs->fs_optim = FS_OPTSPACE;
  376                 break;
  377         default:
  378                 printf("dev = %s, optim = %ld, fs = %s\n",
  379                     devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt);
  380                 panic("ffs_realloccg: bad optim");
  381                 /* NOTREACHED */
  382         }
  383         bno = ffs_hashalloc(ip, cg, bpref, request, ffs_alloccg);
  384         if (bno > 0) {
  385                 bp->b_blkno = fsbtodb(fs, bno);
  386                 if (!DOINGSOFTDEP(vp))
  387                         ffs_blkfree(ump, fs, ip->i_devvp, bprev, (long)osize,
  388                             ip->i_number);
  389                 if (nsize < request)
  390                         ffs_blkfree(ump, fs, ip->i_devvp,
  391                             bno + numfrags(fs, nsize),
  392                             (long)(request - nsize), ip->i_number);
  393                 delta = btodb(nsize - osize);
  394                 if (ip->i_flag & IN_SPACECOUNTED) {
  395                         UFS_LOCK(ump);
  396                         fs->fs_pendingblocks += delta;
  397                         UFS_UNLOCK(ump);
  398                 }
  399                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  400                 if (flags & IO_EXT)
  401                         ip->i_flag |= IN_CHANGE;
  402                 else
  403                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
  404                 allocbuf(bp, nsize);
  405                 bp->b_flags |= B_DONE;
  406                 bzero(bp->b_data + osize, nsize - osize);
  407                 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO)
  408                         vfs_bio_set_valid(bp, osize, nsize - osize);
  409                 *bpp = bp;
  410                 return (0);
  411         }
  412 #ifdef QUOTA
  413         UFS_UNLOCK(ump);
  414         /*
  415          * Restore user's disk quota because allocation failed.
  416          */
  417         (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
  418         UFS_LOCK(ump);
  419 #endif
  420 nospace:
  421         /*
  422          * no space available
  423          */
  424         if (fs->fs_pendingblocks > 0 && reclaimed == 0) {
  425                 reclaimed = 1;
  426                 softdep_request_cleanup(fs, vp);
  427                 UFS_UNLOCK(ump);
  428                 if (bp)
  429                         brelse(bp);
  430                 UFS_LOCK(ump);
  431                 goto retry;
  432         }
  433         UFS_UNLOCK(ump);
  434         if (bp)
  435                 brelse(bp);
  436         if (ppsratecheck(&lastfail, &curfail, 1)) {
  437                 ffs_fserr(fs, ip->i_number, "filesystem full");
  438                 uprintf("\n%s: write failed, filesystem is full\n",
  439                     fs->fs_fsmnt);
  440         }
  441         return (ENOSPC);
  442 }
  443 
  444 /*
  445  * Reallocate a sequence of blocks into a contiguous sequence of blocks.
  446  *
  447  * The vnode and an array of buffer pointers for a range of sequential
  448  * logical blocks to be made contiguous is given. The allocator attempts
  449  * to find a range of sequential blocks starting as close as possible
  450  * from the end of the allocation for the logical block immediately
  451  * preceding the current range. If successful, the physical block numbers
  452  * in the buffer pointers and in the inode are changed to reflect the new
  453  * allocation. If unsuccessful, the allocation is left unchanged. The
  454  * success in doing the reallocation is returned. Note that the error
  455  * return is not reflected back to the user. Rather the previous block
  456  * allocation will be used.
  457  */
  458 
  459 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
  460 
  461 static int doasyncfree = 1;
  462 SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
  463 
  464 static int doreallocblks = 1;
  465 SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
  466 
  467 #ifdef DEBUG
  468 static volatile int prtrealloc = 0;
  469 #endif
  470 
  471 int
  472 ffs_reallocblks(ap)
  473         struct vop_reallocblks_args /* {
  474                 struct vnode *a_vp;
  475                 struct cluster_save *a_buflist;
  476         } */ *ap;
  477 {
  478 
  479         if (doreallocblks == 0)
  480                 return (ENOSPC);
  481         if (VTOI(ap->a_vp)->i_ump->um_fstype == UFS1)
  482                 return (ffs_reallocblks_ufs1(ap));
  483         return (ffs_reallocblks_ufs2(ap));
  484 }
  485         
  486 static int
  487 ffs_reallocblks_ufs1(ap)
  488         struct vop_reallocblks_args /* {
  489                 struct vnode *a_vp;
  490                 struct cluster_save *a_buflist;
  491         } */ *ap;
  492 {
  493         struct fs *fs;
  494         struct inode *ip;
  495         struct vnode *vp;
  496         struct buf *sbp, *ebp;
  497         ufs1_daddr_t *bap, *sbap, *ebap = 0;
  498         struct cluster_save *buflist;
  499         struct ufsmount *ump;
  500         ufs_lbn_t start_lbn, end_lbn;
  501         ufs1_daddr_t soff, newblk, blkno;
  502         ufs2_daddr_t pref;
  503         struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
  504         int i, len, start_lvl, end_lvl, ssize;
  505 
  506         vp = ap->a_vp;
  507         ip = VTOI(vp);
  508         fs = ip->i_fs;
  509         ump = ip->i_ump;
  510         if (fs->fs_contigsumsize <= 0)
  511                 return (ENOSPC);
  512         buflist = ap->a_buflist;
  513         len = buflist->bs_nchildren;
  514         start_lbn = buflist->bs_children[0]->b_lblkno;
  515         end_lbn = start_lbn + len - 1;
  516 #ifdef INVARIANTS
  517         for (i = 0; i < len; i++)
  518                 if (!ffs_checkblk(ip,
  519                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  520                         panic("ffs_reallocblks: unallocated block 1");
  521         for (i = 1; i < len; i++)
  522                 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
  523                         panic("ffs_reallocblks: non-logical cluster");
  524         blkno = buflist->bs_children[0]->b_blkno;
  525         ssize = fsbtodb(fs, fs->fs_frag);
  526         for (i = 1; i < len - 1; i++)
  527                 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
  528                         panic("ffs_reallocblks: non-physical cluster %d", i);
  529 #endif
  530         /*
  531          * If the latest allocation is in a new cylinder group, assume that
  532          * the filesystem has decided to move and do not force it back to
  533          * the previous cylinder group.
  534          */
  535         if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
  536             dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
  537                 return (ENOSPC);
  538         if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
  539             ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
  540                 return (ENOSPC);
  541         /*
  542          * Get the starting offset and block map for the first block.
  543          */
  544         if (start_lvl == 0) {
  545                 sbap = &ip->i_din1->di_db[0];
  546                 soff = start_lbn;
  547         } else {
  548                 idp = &start_ap[start_lvl - 1];
  549                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
  550                         brelse(sbp);
  551                         return (ENOSPC);
  552                 }
  553                 sbap = (ufs1_daddr_t *)sbp->b_data;
  554                 soff = idp->in_off;
  555         }
  556         /*
  557          * If the block range spans two block maps, get the second map.
  558          */
  559         if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
  560                 ssize = len;
  561         } else {
  562 #ifdef INVARIANTS
  563                 if (start_lvl > 0 &&
  564                     start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
  565                         panic("ffs_reallocblk: start == end");
  566 #endif
  567                 ssize = len - (idp->in_off + 1);
  568                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
  569                         goto fail;
  570                 ebap = (ufs1_daddr_t *)ebp->b_data;
  571         }
  572         /*
  573          * Find the preferred location for the cluster.
  574          */
  575         UFS_LOCK(ump);
  576         pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap);
  577         /*
  578          * Search the block map looking for an allocation of the desired size.
  579          */
  580         if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
  581             len, ffs_clusteralloc)) == 0) {
  582                 UFS_UNLOCK(ump);
  583                 goto fail;
  584         }
  585         /*
  586          * We have found a new contiguous block.
  587          *
  588          * First we have to replace the old block pointers with the new
  589          * block pointers in the inode and indirect blocks associated
  590          * with the file.
  591          */
  592 #ifdef DEBUG
  593         if (prtrealloc)
  594                 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
  595                     (intmax_t)start_lbn, (intmax_t)end_lbn);
  596 #endif
  597         blkno = newblk;
  598         for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
  599                 if (i == ssize) {
  600                         bap = ebap;
  601                         soff = -i;
  602                 }
  603 #ifdef INVARIANTS
  604                 if (!ffs_checkblk(ip,
  605                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  606                         panic("ffs_reallocblks: unallocated block 2");
  607                 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
  608                         panic("ffs_reallocblks: alloc mismatch");
  609 #endif
  610 #ifdef DEBUG
  611                 if (prtrealloc)
  612                         printf(" %d,", *bap);
  613 #endif
  614                 if (DOINGSOFTDEP(vp)) {
  615                         if (sbap == &ip->i_din1->di_db[0] && i < ssize)
  616                                 softdep_setup_allocdirect(ip, start_lbn + i,
  617                                     blkno, *bap, fs->fs_bsize, fs->fs_bsize,
  618                                     buflist->bs_children[i]);
  619                         else
  620                                 softdep_setup_allocindir_page(ip, start_lbn + i,
  621                                     i < ssize ? sbp : ebp, soff + i, blkno,
  622                                     *bap, buflist->bs_children[i]);
  623                 }
  624                 *bap++ = blkno;
  625         }
  626         /*
  627          * Next we must write out the modified inode and indirect blocks.
  628          * For strict correctness, the writes should be synchronous since
  629          * the old block values may have been written to disk. In practise
  630          * they are almost never written, but if we are concerned about
  631          * strict correctness, the `doasyncfree' flag should be set to zero.
  632          *
  633          * The test on `doasyncfree' should be changed to test a flag
  634          * that shows whether the associated buffers and inodes have
  635          * been written. The flag should be set when the cluster is
  636          * started and cleared whenever the buffer or inode is flushed.
  637          * We can then check below to see if it is set, and do the
  638          * synchronous write only when it has been cleared.
  639          */
  640         if (sbap != &ip->i_din1->di_db[0]) {
  641                 if (doasyncfree)
  642                         bdwrite(sbp);
  643                 else
  644                         bwrite(sbp);
  645         } else {
  646                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  647                 if (!doasyncfree)
  648                         ffs_update(vp, 1);
  649         }
  650         if (ssize < len) {
  651                 if (doasyncfree)
  652                         bdwrite(ebp);
  653                 else
  654                         bwrite(ebp);
  655         }
  656         /*
  657          * Last, free the old blocks and assign the new blocks to the buffers.
  658          */
  659 #ifdef DEBUG
  660         if (prtrealloc)
  661                 printf("\n\tnew:");
  662 #endif
  663         for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
  664                 if (!DOINGSOFTDEP(vp))
  665                         ffs_blkfree(ump, fs, ip->i_devvp,
  666                             dbtofsb(fs, buflist->bs_children[i]->b_blkno),
  667                             fs->fs_bsize, ip->i_number);
  668                 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
  669 #ifdef INVARIANTS
  670                 if (!ffs_checkblk(ip,
  671                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  672                         panic("ffs_reallocblks: unallocated block 3");
  673 #endif
  674 #ifdef DEBUG
  675                 if (prtrealloc)
  676                         printf(" %d,", blkno);
  677 #endif
  678         }
  679 #ifdef DEBUG
  680         if (prtrealloc) {
  681                 prtrealloc--;
  682                 printf("\n");
  683         }
  684 #endif
  685         return (0);
  686 
  687 fail:
  688         if (ssize < len)
  689                 brelse(ebp);
  690         if (sbap != &ip->i_din1->di_db[0])
  691                 brelse(sbp);
  692         return (ENOSPC);
  693 }
  694 
  695 static int
  696 ffs_reallocblks_ufs2(ap)
  697         struct vop_reallocblks_args /* {
  698                 struct vnode *a_vp;
  699                 struct cluster_save *a_buflist;
  700         } */ *ap;
  701 {
  702         struct fs *fs;
  703         struct inode *ip;
  704         struct vnode *vp;
  705         struct buf *sbp, *ebp;
  706         ufs2_daddr_t *bap, *sbap, *ebap = 0;
  707         struct cluster_save *buflist;
  708         struct ufsmount *ump;
  709         ufs_lbn_t start_lbn, end_lbn;
  710         ufs2_daddr_t soff, newblk, blkno, pref;
  711         struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
  712         int i, len, start_lvl, end_lvl, ssize;
  713 
  714         vp = ap->a_vp;
  715         ip = VTOI(vp);
  716         fs = ip->i_fs;
  717         ump = ip->i_ump;
  718         if (fs->fs_contigsumsize <= 0)
  719                 return (ENOSPC);
  720         buflist = ap->a_buflist;
  721         len = buflist->bs_nchildren;
  722         start_lbn = buflist->bs_children[0]->b_lblkno;
  723         end_lbn = start_lbn + len - 1;
  724 #ifdef INVARIANTS
  725         for (i = 0; i < len; i++)
  726                 if (!ffs_checkblk(ip,
  727                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  728                         panic("ffs_reallocblks: unallocated block 1");
  729         for (i = 1; i < len; i++)
  730                 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
  731                         panic("ffs_reallocblks: non-logical cluster");
  732         blkno = buflist->bs_children[0]->b_blkno;
  733         ssize = fsbtodb(fs, fs->fs_frag);
  734         for (i = 1; i < len - 1; i++)
  735                 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
  736                         panic("ffs_reallocblks: non-physical cluster %d", i);
  737 #endif
  738         /*
  739          * If the latest allocation is in a new cylinder group, assume that
  740          * the filesystem has decided to move and do not force it back to
  741          * the previous cylinder group.
  742          */
  743         if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
  744             dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
  745                 return (ENOSPC);
  746         if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
  747             ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
  748                 return (ENOSPC);
  749         /*
  750          * Get the starting offset and block map for the first block.
  751          */
  752         if (start_lvl == 0) {
  753                 sbap = &ip->i_din2->di_db[0];
  754                 soff = start_lbn;
  755         } else {
  756                 idp = &start_ap[start_lvl - 1];
  757                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
  758                         brelse(sbp);
  759                         return (ENOSPC);
  760                 }
  761                 sbap = (ufs2_daddr_t *)sbp->b_data;
  762                 soff = idp->in_off;
  763         }
  764         /*
  765          * If the block range spans two block maps, get the second map.
  766          */
  767         if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
  768                 ssize = len;
  769         } else {
  770 #ifdef INVARIANTS
  771                 if (start_lvl > 0 &&
  772                     start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
  773                         panic("ffs_reallocblk: start == end");
  774 #endif
  775                 ssize = len - (idp->in_off + 1);
  776                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
  777                         goto fail;
  778                 ebap = (ufs2_daddr_t *)ebp->b_data;
  779         }
  780         /*
  781          * Find the preferred location for the cluster.
  782          */
  783         UFS_LOCK(ump);
  784         pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap);
  785         /*
  786          * Search the block map looking for an allocation of the desired size.
  787          */
  788         if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
  789             len, ffs_clusteralloc)) == 0) {
  790                 UFS_UNLOCK(ump);
  791                 goto fail;
  792         }
  793         /*
  794          * We have found a new contiguous block.
  795          *
  796          * First we have to replace the old block pointers with the new
  797          * block pointers in the inode and indirect blocks associated
  798          * with the file.
  799          */
  800 #ifdef DEBUG
  801         if (prtrealloc)
  802                 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
  803                     (intmax_t)start_lbn, (intmax_t)end_lbn);
  804 #endif
  805         blkno = newblk;
  806         for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
  807                 if (i == ssize) {
  808                         bap = ebap;
  809                         soff = -i;
  810                 }
  811 #ifdef INVARIANTS
  812                 if (!ffs_checkblk(ip,
  813                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  814                         panic("ffs_reallocblks: unallocated block 2");
  815                 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
  816                         panic("ffs_reallocblks: alloc mismatch");
  817 #endif
  818 #ifdef DEBUG
  819                 if (prtrealloc)
  820                         printf(" %jd,", (intmax_t)*bap);
  821 #endif
  822                 if (DOINGSOFTDEP(vp)) {
  823                         if (sbap == &ip->i_din2->di_db[0] && i < ssize)
  824                                 softdep_setup_allocdirect(ip, start_lbn + i,
  825                                     blkno, *bap, fs->fs_bsize, fs->fs_bsize,
  826                                     buflist->bs_children[i]);
  827                         else
  828                                 softdep_setup_allocindir_page(ip, start_lbn + i,
  829                                     i < ssize ? sbp : ebp, soff + i, blkno,
  830                                     *bap, buflist->bs_children[i]);
  831                 }
  832                 *bap++ = blkno;
  833         }
  834         /*
  835          * Next we must write out the modified inode and indirect blocks.
  836          * For strict correctness, the writes should be synchronous since
  837          * the old block values may have been written to disk. In practise
  838          * they are almost never written, but if we are concerned about
  839          * strict correctness, the `doasyncfree' flag should be set to zero.
  840          *
  841          * The test on `doasyncfree' should be changed to test a flag
  842          * that shows whether the associated buffers and inodes have
  843          * been written. The flag should be set when the cluster is
  844          * started and cleared whenever the buffer or inode is flushed.
  845          * We can then check below to see if it is set, and do the
  846          * synchronous write only when it has been cleared.
  847          */
  848         if (sbap != &ip->i_din2->di_db[0]) {
  849                 if (doasyncfree)
  850                         bdwrite(sbp);
  851                 else
  852                         bwrite(sbp);
  853         } else {
  854                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  855                 if (!doasyncfree)
  856                         ffs_update(vp, 1);
  857         }
  858         if (ssize < len) {
  859                 if (doasyncfree)
  860                         bdwrite(ebp);
  861                 else
  862                         bwrite(ebp);
  863         }
  864         /*
  865          * Last, free the old blocks and assign the new blocks to the buffers.
  866          */
  867 #ifdef DEBUG
  868         if (prtrealloc)
  869                 printf("\n\tnew:");
  870 #endif
  871         for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
  872                 if (!DOINGSOFTDEP(vp))
  873                         ffs_blkfree(ump, fs, ip->i_devvp,
  874                             dbtofsb(fs, buflist->bs_children[i]->b_blkno),
  875                             fs->fs_bsize, ip->i_number);
  876                 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
  877 #ifdef INVARIANTS
  878                 if (!ffs_checkblk(ip,
  879                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  880                         panic("ffs_reallocblks: unallocated block 3");
  881 #endif
  882 #ifdef DEBUG
  883                 if (prtrealloc)
  884                         printf(" %jd,", (intmax_t)blkno);
  885 #endif
  886         }
  887 #ifdef DEBUG
  888         if (prtrealloc) {
  889                 prtrealloc--;
  890                 printf("\n");
  891         }
  892 #endif
  893         return (0);
  894 
  895 fail:
  896         if (ssize < len)
  897                 brelse(ebp);
  898         if (sbap != &ip->i_din2->di_db[0])
  899                 brelse(sbp);
  900         return (ENOSPC);
  901 }
  902 
  903 /*
  904  * Allocate an inode in the filesystem.
  905  *
  906  * If allocating a directory, use ffs_dirpref to select the inode.
  907  * If allocating in a directory, the following hierarchy is followed:
  908  *   1) allocate the preferred inode.
  909  *   2) allocate an inode in the same cylinder group.
  910  *   3) quadradically rehash into other cylinder groups, until an
  911  *      available inode is located.
  912  * If no inode preference is given the following hierarchy is used
  913  * to allocate an inode:
  914  *   1) allocate an inode in cylinder group 0.
  915  *   2) quadradically rehash into other cylinder groups, until an
  916  *      available inode is located.
  917  */
  918 int
  919 ffs_valloc(pvp, mode, cred, vpp)
  920         struct vnode *pvp;
  921         int mode;
  922         struct ucred *cred;
  923         struct vnode **vpp;
  924 {
  925         struct inode *pip;
  926         struct fs *fs;
  927         struct inode *ip;
  928         struct timespec ts;
  929         struct ufsmount *ump;
  930         ino_t ino, ipref;
  931         int cg, error, error1;
  932         static struct timeval lastfail;
  933         static int curfail;
  934 
  935         *vpp = NULL;
  936         pip = VTOI(pvp);
  937         fs = pip->i_fs;
  938         ump = pip->i_ump;
  939 
  940         UFS_LOCK(ump);
  941         if (fs->fs_cstotal.cs_nifree == 0)
  942                 goto noinodes;
  943 
  944         if ((mode & IFMT) == IFDIR)
  945                 ipref = ffs_dirpref(pip);
  946         else
  947                 ipref = pip->i_number;
  948         if (ipref >= fs->fs_ncg * fs->fs_ipg)
  949                 ipref = 0;
  950         cg = ino_to_cg(fs, ipref);
  951         /*
  952          * Track number of dirs created one after another
  953          * in a same cg without intervening by files.
  954          */
  955         if ((mode & IFMT) == IFDIR) {
  956                 if (fs->fs_contigdirs[cg] < 255)
  957                         fs->fs_contigdirs[cg]++;
  958         } else {
  959                 if (fs->fs_contigdirs[cg] > 0)
  960                         fs->fs_contigdirs[cg]--;
  961         }
  962         ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode,
  963                                         (allocfcn_t *)ffs_nodealloccg);
  964         if (ino == 0)
  965                 goto noinodes;
  966         error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
  967         if (error) {
  968                 error1 = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp,
  969                     FFSV_FORCEINSMQ);
  970                 ffs_vfree(pvp, ino, mode);
  971                 if (error1 == 0) {
  972                         ip = VTOI(*vpp);
  973                         if (ip->i_mode)
  974                                 goto dup_alloc;
  975                         ip->i_flag |= IN_MODIFIED;
  976                         vput(*vpp);
  977                 }
  978                 return (error);
  979         }
  980         ip = VTOI(*vpp);
  981         if (ip->i_mode) {
  982 dup_alloc:
  983                 printf("mode = 0%o, inum = %lu, fs = %s\n",
  984                     ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
  985                 panic("ffs_valloc: dup alloc");
  986         }
  987         if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) {  /* XXX */
  988                 printf("free inode %s/%lu had %ld blocks\n",
  989                     fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks));
  990                 DIP_SET(ip, i_blocks, 0);
  991         }
  992         ip->i_flags = 0;
  993         DIP_SET(ip, i_flags, 0);
  994         /*
  995          * Set up a new generation number for this inode.
  996          */
  997         if (ip->i_gen == 0 || ++ip->i_gen == 0)
  998                 ip->i_gen = arc4random() / 2 + 1;
  999         DIP_SET(ip, i_gen, ip->i_gen);
 1000         if (fs->fs_magic == FS_UFS2_MAGIC) {
 1001                 vfs_timestamp(&ts);
 1002                 ip->i_din2->di_birthtime = ts.tv_sec;
 1003                 ip->i_din2->di_birthnsec = ts.tv_nsec;
 1004         }
 1005         ip->i_flag = 0;
 1006         vnode_destroy_vobject(*vpp);
 1007         (*vpp)->v_type = VNON;
 1008         if (fs->fs_magic == FS_UFS2_MAGIC)
 1009                 (*vpp)->v_op = &ffs_vnodeops2;
 1010         else
 1011                 (*vpp)->v_op = &ffs_vnodeops1;
 1012         return (0);
 1013 noinodes:
 1014         UFS_UNLOCK(ump);
 1015         if (ppsratecheck(&lastfail, &curfail, 1)) {
 1016                 ffs_fserr(fs, pip->i_number, "out of inodes");
 1017                 uprintf("\n%s: create/symlink failed, no inodes free\n",
 1018                     fs->fs_fsmnt);
 1019         }
 1020         return (ENOSPC);
 1021 }
 1022 
 1023 /*
 1024  * Find a cylinder group to place a directory.
 1025  *
 1026  * The policy implemented by this algorithm is to allocate a
 1027  * directory inode in the same cylinder group as its parent
 1028  * directory, but also to reserve space for its files inodes
 1029  * and data. Restrict the number of directories which may be
 1030  * allocated one after another in the same cylinder group
 1031  * without intervening allocation of files.
 1032  *
 1033  * If we allocate a first level directory then force allocation
 1034  * in another cylinder group.
 1035  */
 1036 static ino_t
 1037 ffs_dirpref(pip)
 1038         struct inode *pip;
 1039 {
 1040         struct fs *fs;
 1041         int cg, prefcg, dirsize, cgsize;
 1042         int avgifree, avgbfree, avgndir, curdirsize;
 1043         int minifree, minbfree, maxndir;
 1044         int mincg, minndir;
 1045         int maxcontigdirs;
 1046 
 1047         mtx_assert(UFS_MTX(pip->i_ump), MA_OWNED);
 1048         fs = pip->i_fs;
 1049 
 1050         avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
 1051         avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1052         avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
 1053 
 1054         /*
 1055          * Force allocation in another cg if creating a first level dir.
 1056          */
 1057         ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref");
 1058         if (ITOV(pip)->v_vflag & VV_ROOT) {
 1059                 prefcg = arc4random() % fs->fs_ncg;
 1060                 mincg = prefcg;
 1061                 minndir = fs->fs_ipg;
 1062                 for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1063                         if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
 1064                             fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
 1065                             fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1066                                 mincg = cg;
 1067                                 minndir = fs->fs_cs(fs, cg).cs_ndir;
 1068                         }
 1069                 for (cg = 0; cg < prefcg; cg++)
 1070                         if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
 1071                             fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
 1072                             fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1073                                 mincg = cg;
 1074                                 minndir = fs->fs_cs(fs, cg).cs_ndir;
 1075                         }
 1076                 return ((ino_t)(fs->fs_ipg * mincg));
 1077         }
 1078 
 1079         /*
 1080          * Count various limits which used for
 1081          * optimal allocation of a directory inode.
 1082          */
 1083         maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
 1084         minifree = avgifree - avgifree / 4;
 1085         if (minifree < 1)
 1086                 minifree = 1;
 1087         minbfree = avgbfree - avgbfree / 4;
 1088         if (minbfree < 1)
 1089                 minbfree = 1;
 1090         cgsize = fs->fs_fsize * fs->fs_fpg;
 1091         dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir;
 1092         curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
 1093         if (dirsize < curdirsize)
 1094                 dirsize = curdirsize;
 1095         if (dirsize <= 0)
 1096                 maxcontigdirs = 0;              /* dirsize overflowed */
 1097         else
 1098                 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255);
 1099         if (fs->fs_avgfpdir > 0)
 1100                 maxcontigdirs = min(maxcontigdirs,
 1101                                     fs->fs_ipg / fs->fs_avgfpdir);
 1102         if (maxcontigdirs == 0)
 1103                 maxcontigdirs = 1;
 1104 
 1105         /*
 1106          * Limit number of dirs in one cg and reserve space for 
 1107          * regular files, but only if we have no deficit in
 1108          * inodes or space.
 1109          */
 1110         prefcg = ino_to_cg(fs, pip->i_number);
 1111         for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1112                 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
 1113                     fs->fs_cs(fs, cg).cs_nifree >= minifree &&
 1114                     fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
 1115                         if (fs->fs_contigdirs[cg] < maxcontigdirs)
 1116                                 return ((ino_t)(fs->fs_ipg * cg));
 1117                 }
 1118         for (cg = 0; cg < prefcg; cg++)
 1119                 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
 1120                     fs->fs_cs(fs, cg).cs_nifree >= minifree &&
 1121                     fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
 1122                         if (fs->fs_contigdirs[cg] < maxcontigdirs)
 1123                                 return ((ino_t)(fs->fs_ipg * cg));
 1124                 }
 1125         /*
 1126          * This is a backstop when we have deficit in space.
 1127          */
 1128         for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1129                 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
 1130                         return ((ino_t)(fs->fs_ipg * cg));
 1131         for (cg = 0; cg < prefcg; cg++)
 1132                 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
 1133                         break;
 1134         return ((ino_t)(fs->fs_ipg * cg));
 1135 }
 1136 
 1137 /*
 1138  * Select the desired position for the next block in a file.  The file is
 1139  * logically divided into sections. The first section is composed of the
 1140  * direct blocks. Each additional section contains fs_maxbpg blocks.
 1141  *
 1142  * If no blocks have been allocated in the first section, the policy is to
 1143  * request a block in the same cylinder group as the inode that describes
 1144  * the file. If no blocks have been allocated in any other section, the
 1145  * policy is to place the section in a cylinder group with a greater than
 1146  * average number of free blocks.  An appropriate cylinder group is found
 1147  * by using a rotor that sweeps the cylinder groups. When a new group of
 1148  * blocks is needed, the sweep begins in the cylinder group following the
 1149  * cylinder group from which the previous allocation was made. The sweep
 1150  * continues until a cylinder group with greater than the average number
 1151  * of free blocks is found. If the allocation is for the first block in an
 1152  * indirect block, the information on the previous allocation is unavailable;
 1153  * here a best guess is made based upon the logical block number being
 1154  * allocated.
 1155  *
 1156  * If a section is already partially allocated, the policy is to
 1157  * contiguously allocate fs_maxcontig blocks. The end of one of these
 1158  * contiguous blocks and the beginning of the next is laid out
 1159  * contiguously if possible.
 1160  */
 1161 ufs2_daddr_t
 1162 ffs_blkpref_ufs1(ip, lbn, indx, bap)
 1163         struct inode *ip;
 1164         ufs_lbn_t lbn;
 1165         int indx;
 1166         ufs1_daddr_t *bap;
 1167 {
 1168         struct fs *fs;
 1169         int cg;
 1170         int avgbfree, startcg;
 1171 
 1172         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1173         fs = ip->i_fs;
 1174         if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
 1175                 if (lbn < NDADDR + NINDIR(fs)) {
 1176                         cg = ino_to_cg(fs, ip->i_number);
 1177                         return (cgbase(fs, cg) + fs->fs_frag);
 1178                 }
 1179                 /*
 1180                  * Find a cylinder with greater than average number of
 1181                  * unused data blocks.
 1182                  */
 1183                 if (indx == 0 || bap[indx - 1] == 0)
 1184                         startcg =
 1185                             ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
 1186                 else
 1187                         startcg = dtog(fs, bap[indx - 1]) + 1;
 1188                 startcg %= fs->fs_ncg;
 1189                 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1190                 for (cg = startcg; cg < fs->fs_ncg; cg++)
 1191                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1192                                 fs->fs_cgrotor = cg;
 1193                                 return (cgbase(fs, cg) + fs->fs_frag);
 1194                         }
 1195                 for (cg = 0; cg <= startcg; cg++)
 1196                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1197                                 fs->fs_cgrotor = cg;
 1198                                 return (cgbase(fs, cg) + fs->fs_frag);
 1199                         }
 1200                 return (0);
 1201         }
 1202         /*
 1203          * We just always try to lay things out contiguously.
 1204          */
 1205         return (bap[indx - 1] + fs->fs_frag);
 1206 }
 1207 
 1208 /*
 1209  * Same as above, but for UFS2
 1210  */
 1211 ufs2_daddr_t
 1212 ffs_blkpref_ufs2(ip, lbn, indx, bap)
 1213         struct inode *ip;
 1214         ufs_lbn_t lbn;
 1215         int indx;
 1216         ufs2_daddr_t *bap;
 1217 {
 1218         struct fs *fs;
 1219         int cg;
 1220         int avgbfree, startcg;
 1221 
 1222         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1223         fs = ip->i_fs;
 1224         if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
 1225                 if (lbn < NDADDR + NINDIR(fs)) {
 1226                         cg = ino_to_cg(fs, ip->i_number);
 1227                         return (cgbase(fs, cg) + fs->fs_frag);
 1228                 }
 1229                 /*
 1230                  * Find a cylinder with greater than average number of
 1231                  * unused data blocks.
 1232                  */
 1233                 if (indx == 0 || bap[indx - 1] == 0)
 1234                         startcg =
 1235                             ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
 1236                 else
 1237                         startcg = dtog(fs, bap[indx - 1]) + 1;
 1238                 startcg %= fs->fs_ncg;
 1239                 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1240                 for (cg = startcg; cg < fs->fs_ncg; cg++)
 1241                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1242                                 fs->fs_cgrotor = cg;
 1243                                 return (cgbase(fs, cg) + fs->fs_frag);
 1244                         }
 1245                 for (cg = 0; cg <= startcg; cg++)
 1246                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1247                                 fs->fs_cgrotor = cg;
 1248                                 return (cgbase(fs, cg) + fs->fs_frag);
 1249                         }
 1250                 return (0);
 1251         }
 1252         /*
 1253          * We just always try to lay things out contiguously.
 1254          */
 1255         return (bap[indx - 1] + fs->fs_frag);
 1256 }
 1257 
 1258 /*
 1259  * Implement the cylinder overflow algorithm.
 1260  *
 1261  * The policy implemented by this algorithm is:
 1262  *   1) allocate the block in its requested cylinder group.
 1263  *   2) quadradically rehash on the cylinder group number.
 1264  *   3) brute force search for a free block.
 1265  *
 1266  * Must be called with the UFS lock held.  Will release the lock on success
 1267  * and return with it held on failure.
 1268  */
 1269 /*VARARGS5*/
 1270 static ufs2_daddr_t
 1271 ffs_hashalloc(ip, cg, pref, size, allocator)
 1272         struct inode *ip;
 1273         int cg;
 1274         ufs2_daddr_t pref;
 1275         int size;       /* size for data blocks, mode for inodes */
 1276         allocfcn_t *allocator;
 1277 {
 1278         struct fs *fs;
 1279         ufs2_daddr_t result;
 1280         int i, icg = cg;
 1281 
 1282         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1283 #ifdef INVARIANTS
 1284         if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
 1285                 panic("ffs_hashalloc: allocation on suspended filesystem");
 1286 #endif
 1287         fs = ip->i_fs;
 1288         /*
 1289          * 1: preferred cylinder group
 1290          */
 1291         result = (*allocator)(ip, cg, pref, size);
 1292         if (result)
 1293                 return (result);
 1294         /*
 1295          * 2: quadratic rehash
 1296          */
 1297         for (i = 1; i < fs->fs_ncg; i *= 2) {
 1298                 cg += i;
 1299                 if (cg >= fs->fs_ncg)
 1300                         cg -= fs->fs_ncg;
 1301                 result = (*allocator)(ip, cg, 0, size);
 1302                 if (result)
 1303                         return (result);
 1304         }
 1305         /*
 1306          * 3: brute force search
 1307          * Note that we start at i == 2, since 0 was checked initially,
 1308          * and 1 is always checked in the quadratic rehash.
 1309          */
 1310         cg = (icg + 2) % fs->fs_ncg;
 1311         for (i = 2; i < fs->fs_ncg; i++) {
 1312                 result = (*allocator)(ip, cg, 0, size);
 1313                 if (result)
 1314                         return (result);
 1315                 cg++;
 1316                 if (cg == fs->fs_ncg)
 1317                         cg = 0;
 1318         }
 1319         return (0);
 1320 }
 1321 
 1322 /*
 1323  * Determine whether a fragment can be extended.
 1324  *
 1325  * Check to see if the necessary fragments are available, and
 1326  * if they are, allocate them.
 1327  */
 1328 static ufs2_daddr_t
 1329 ffs_fragextend(ip, cg, bprev, osize, nsize)
 1330         struct inode *ip;
 1331         int cg;
 1332         ufs2_daddr_t bprev;
 1333         int osize, nsize;
 1334 {
 1335         struct fs *fs;
 1336         struct cg *cgp;
 1337         struct buf *bp;
 1338         struct ufsmount *ump;
 1339         int nffree;
 1340         long bno;
 1341         int frags, bbase;
 1342         int i, error;
 1343         u_int8_t *blksfree;
 1344 
 1345         ump = ip->i_ump;
 1346         fs = ip->i_fs;
 1347         if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
 1348                 return (0);
 1349         frags = numfrags(fs, nsize);
 1350         bbase = fragnum(fs, bprev);
 1351         if (bbase > fragnum(fs, (bprev + frags - 1))) {
 1352                 /* cannot extend across a block boundary */
 1353                 return (0);
 1354         }
 1355         UFS_UNLOCK(ump);
 1356         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1357                 (int)fs->fs_cgsize, NOCRED, &bp);
 1358         if (error)
 1359                 goto fail;
 1360         cgp = (struct cg *)bp->b_data;
 1361         if (!cg_chkmagic(cgp))
 1362                 goto fail;
 1363         bp->b_xflags |= BX_BKGRDWRITE;
 1364         cgp->cg_old_time = cgp->cg_time = time_second;
 1365         bno = dtogd(fs, bprev);
 1366         blksfree = cg_blksfree(cgp);
 1367         for (i = numfrags(fs, osize); i < frags; i++)
 1368                 if (isclr(blksfree, bno + i))
 1369                         goto fail;
 1370         /*
 1371          * the current fragment can be extended
 1372          * deduct the count on fragment being extended into
 1373          * increase the count on the remaining fragment (if any)
 1374          * allocate the extended piece
 1375          */
 1376         for (i = frags; i < fs->fs_frag - bbase; i++)
 1377                 if (isclr(blksfree, bno + i))
 1378                         break;
 1379         cgp->cg_frsum[i - numfrags(fs, osize)]--;
 1380         if (i != frags)
 1381                 cgp->cg_frsum[i - frags]++;
 1382         for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) {
 1383                 clrbit(blksfree, bno + i);
 1384                 cgp->cg_cs.cs_nffree--;
 1385                 nffree++;
 1386         }
 1387         UFS_LOCK(ump);
 1388         fs->fs_cstotal.cs_nffree -= nffree;
 1389         fs->fs_cs(fs, cg).cs_nffree -= nffree;
 1390         fs->fs_fmod = 1;
 1391         ACTIVECLEAR(fs, cg);
 1392         UFS_UNLOCK(ump);
 1393         if (DOINGSOFTDEP(ITOV(ip)))
 1394                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev);
 1395         bdwrite(bp);
 1396         return (bprev);
 1397 
 1398 fail:
 1399         brelse(bp);
 1400         UFS_LOCK(ump);
 1401         return (0);
 1402 
 1403 }
 1404 
 1405 /*
 1406  * Determine whether a block can be allocated.
 1407  *
 1408  * Check to see if a block of the appropriate size is available,
 1409  * and if it is, allocate it.
 1410  */
 1411 static ufs2_daddr_t
 1412 ffs_alloccg(ip, cg, bpref, size)
 1413         struct inode *ip;
 1414         int cg;
 1415         ufs2_daddr_t bpref;
 1416         int size;
 1417 {
 1418         struct fs *fs;
 1419         struct cg *cgp;
 1420         struct buf *bp;
 1421         struct ufsmount *ump;
 1422         ufs1_daddr_t bno;
 1423         ufs2_daddr_t blkno;
 1424         int i, allocsiz, error, frags;
 1425         u_int8_t *blksfree;
 1426 
 1427         ump = ip->i_ump;
 1428         fs = ip->i_fs;
 1429         if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
 1430                 return (0);
 1431         UFS_UNLOCK(ump);
 1432         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1433                 (int)fs->fs_cgsize, NOCRED, &bp);
 1434         if (error)
 1435                 goto fail;
 1436         cgp = (struct cg *)bp->b_data;
 1437         if (!cg_chkmagic(cgp) ||
 1438             (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize))
 1439                 goto fail;
 1440         bp->b_xflags |= BX_BKGRDWRITE;
 1441         cgp->cg_old_time = cgp->cg_time = time_second;
 1442         if (size == fs->fs_bsize) {
 1443                 UFS_LOCK(ump);
 1444                 blkno = ffs_alloccgblk(ip, bp, bpref);
 1445                 ACTIVECLEAR(fs, cg);
 1446                 UFS_UNLOCK(ump);
 1447                 bdwrite(bp);
 1448                 return (blkno);
 1449         }
 1450         /*
 1451          * check to see if any fragments are already available
 1452          * allocsiz is the size which will be allocated, hacking
 1453          * it down to a smaller size if necessary
 1454          */
 1455         blksfree = cg_blksfree(cgp);
 1456         frags = numfrags(fs, size);
 1457         for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
 1458                 if (cgp->cg_frsum[allocsiz] != 0)
 1459                         break;
 1460         if (allocsiz == fs->fs_frag) {
 1461                 /*
 1462                  * no fragments were available, so a block will be
 1463                  * allocated, and hacked up
 1464                  */
 1465                 if (cgp->cg_cs.cs_nbfree == 0)
 1466                         goto fail;
 1467                 UFS_LOCK(ump);
 1468                 blkno = ffs_alloccgblk(ip, bp, bpref);
 1469                 bno = dtogd(fs, blkno);
 1470                 for (i = frags; i < fs->fs_frag; i++)
 1471                         setbit(blksfree, bno + i);
 1472                 i = fs->fs_frag - frags;
 1473                 cgp->cg_cs.cs_nffree += i;
 1474                 fs->fs_cstotal.cs_nffree += i;
 1475                 fs->fs_cs(fs, cg).cs_nffree += i;
 1476                 fs->fs_fmod = 1;
 1477                 cgp->cg_frsum[i]++;
 1478                 ACTIVECLEAR(fs, cg);
 1479                 UFS_UNLOCK(ump);
 1480                 bdwrite(bp);
 1481                 return (blkno);
 1482         }
 1483         bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
 1484         if (bno < 0)
 1485                 goto fail;
 1486         for (i = 0; i < frags; i++)
 1487                 clrbit(blksfree, bno + i);
 1488         cgp->cg_cs.cs_nffree -= frags;
 1489         cgp->cg_frsum[allocsiz]--;
 1490         if (frags != allocsiz)
 1491                 cgp->cg_frsum[allocsiz - frags]++;
 1492         UFS_LOCK(ump);
 1493         fs->fs_cstotal.cs_nffree -= frags;
 1494         fs->fs_cs(fs, cg).cs_nffree -= frags;
 1495         fs->fs_fmod = 1;
 1496         blkno = cgbase(fs, cg) + bno;
 1497         ACTIVECLEAR(fs, cg);
 1498         UFS_UNLOCK(ump);
 1499         if (DOINGSOFTDEP(ITOV(ip)))
 1500                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno);
 1501         bdwrite(bp);
 1502         return (blkno);
 1503 
 1504 fail:
 1505         brelse(bp);
 1506         UFS_LOCK(ump);
 1507         return (0);
 1508 }
 1509 
 1510 /*
 1511  * Allocate a block in a cylinder group.
 1512  *
 1513  * This algorithm implements the following policy:
 1514  *   1) allocate the requested block.
 1515  *   2) allocate a rotationally optimal block in the same cylinder.
 1516  *   3) allocate the next available block on the block rotor for the
 1517  *      specified cylinder group.
 1518  * Note that this routine only allocates fs_bsize blocks; these
 1519  * blocks may be fragmented by the routine that allocates them.
 1520  */
 1521 static ufs2_daddr_t
 1522 ffs_alloccgblk(ip, bp, bpref)
 1523         struct inode *ip;
 1524         struct buf *bp;
 1525         ufs2_daddr_t bpref;
 1526 {
 1527         struct fs *fs;
 1528         struct cg *cgp;
 1529         struct ufsmount *ump;
 1530         ufs1_daddr_t bno;
 1531         ufs2_daddr_t blkno;
 1532         u_int8_t *blksfree;
 1533 
 1534         fs = ip->i_fs;
 1535         ump = ip->i_ump;
 1536         mtx_assert(UFS_MTX(ump), MA_OWNED);
 1537         cgp = (struct cg *)bp->b_data;
 1538         blksfree = cg_blksfree(cgp);
 1539         if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
 1540                 bpref = cgp->cg_rotor;
 1541         } else {
 1542                 bpref = blknum(fs, bpref);
 1543                 bno = dtogd(fs, bpref);
 1544                 /*
 1545                  * if the requested block is available, use it
 1546                  */
 1547                 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno)))
 1548                         goto gotit;
 1549         }
 1550         /*
 1551          * Take the next available block in this cylinder group.
 1552          */
 1553         bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
 1554         if (bno < 0)
 1555                 return (0);
 1556         cgp->cg_rotor = bno;
 1557 gotit:
 1558         blkno = fragstoblks(fs, bno);
 1559         ffs_clrblock(fs, blksfree, (long)blkno);
 1560         ffs_clusteracct(ump, fs, cgp, blkno, -1);
 1561         cgp->cg_cs.cs_nbfree--;
 1562         fs->fs_cstotal.cs_nbfree--;
 1563         fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
 1564         fs->fs_fmod = 1;
 1565         blkno = cgbase(fs, cgp->cg_cgx) + bno;
 1566         /* XXX Fixme. */
 1567         UFS_UNLOCK(ump);
 1568         if (DOINGSOFTDEP(ITOV(ip)))
 1569                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno);
 1570         UFS_LOCK(ump);
 1571         return (blkno);
 1572 }
 1573 
 1574 /*
 1575  * Determine whether a cluster can be allocated.
 1576  *
 1577  * We do not currently check for optimal rotational layout if there
 1578  * are multiple choices in the same cylinder group. Instead we just
 1579  * take the first one that we find following bpref.
 1580  */
 1581 static ufs2_daddr_t
 1582 ffs_clusteralloc(ip, cg, bpref, len)
 1583         struct inode *ip;
 1584         int cg;
 1585         ufs2_daddr_t bpref;
 1586         int len;
 1587 {
 1588         struct fs *fs;
 1589         struct cg *cgp;
 1590         struct buf *bp;
 1591         struct ufsmount *ump;
 1592         int i, run, bit, map, got;
 1593         ufs2_daddr_t bno;
 1594         u_char *mapp;
 1595         int32_t *lp;
 1596         u_int8_t *blksfree;
 1597 
 1598         fs = ip->i_fs;
 1599         ump = ip->i_ump;
 1600         if (fs->fs_maxcluster[cg] < len)
 1601                 return (0);
 1602         UFS_UNLOCK(ump);
 1603         if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
 1604             NOCRED, &bp))
 1605                 goto fail_lock;
 1606         cgp = (struct cg *)bp->b_data;
 1607         if (!cg_chkmagic(cgp))
 1608                 goto fail_lock;
 1609         bp->b_xflags |= BX_BKGRDWRITE;
 1610         /*
 1611          * Check to see if a cluster of the needed size (or bigger) is
 1612          * available in this cylinder group.
 1613          */
 1614         lp = &cg_clustersum(cgp)[len];
 1615         for (i = len; i <= fs->fs_contigsumsize; i++)
 1616                 if (*lp++ > 0)
 1617                         break;
 1618         if (i > fs->fs_contigsumsize) {
 1619                 /*
 1620                  * This is the first time looking for a cluster in this
 1621                  * cylinder group. Update the cluster summary information
 1622                  * to reflect the true maximum sized cluster so that
 1623                  * future cluster allocation requests can avoid reading
 1624                  * the cylinder group map only to find no clusters.
 1625                  */
 1626                 lp = &cg_clustersum(cgp)[len - 1];
 1627                 for (i = len - 1; i > 0; i--)
 1628                         if (*lp-- > 0)
 1629                                 break;
 1630                 UFS_LOCK(ump);
 1631                 fs->fs_maxcluster[cg] = i;
 1632                 goto fail;
 1633         }
 1634         /*
 1635          * Search the cluster map to find a big enough cluster.
 1636          * We take the first one that we find, even if it is larger
 1637          * than we need as we prefer to get one close to the previous
 1638          * block allocation. We do not search before the current
 1639          * preference point as we do not want to allocate a block
 1640          * that is allocated before the previous one (as we will
 1641          * then have to wait for another pass of the elevator
 1642          * algorithm before it will be read). We prefer to fail and
 1643          * be recalled to try an allocation in the next cylinder group.
 1644          */
 1645         if (dtog(fs, bpref) != cg)
 1646                 bpref = 0;
 1647         else
 1648                 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
 1649         mapp = &cg_clustersfree(cgp)[bpref / NBBY];
 1650         map = *mapp++;
 1651         bit = 1 << (bpref % NBBY);
 1652         for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
 1653                 if ((map & bit) == 0) {
 1654                         run = 0;
 1655                 } else {
 1656                         run++;
 1657                         if (run == len)
 1658                                 break;
 1659                 }
 1660                 if ((got & (NBBY - 1)) != (NBBY - 1)) {
 1661                         bit <<= 1;
 1662                 } else {
 1663                         map = *mapp++;
 1664                         bit = 1;
 1665                 }
 1666         }
 1667         if (got >= cgp->cg_nclusterblks)
 1668                 goto fail_lock;
 1669         /*
 1670          * Allocate the cluster that we have found.
 1671          */
 1672         blksfree = cg_blksfree(cgp);
 1673         for (i = 1; i <= len; i++)
 1674                 if (!ffs_isblock(fs, blksfree, got - run + i))
 1675                         panic("ffs_clusteralloc: map mismatch");
 1676         bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1);
 1677         if (dtog(fs, bno) != cg)
 1678                 panic("ffs_clusteralloc: allocated out of group");
 1679         len = blkstofrags(fs, len);
 1680         UFS_LOCK(ump);
 1681         for (i = 0; i < len; i += fs->fs_frag)
 1682                 if (ffs_alloccgblk(ip, bp, bno + i) != bno + i)
 1683                         panic("ffs_clusteralloc: lost block");
 1684         ACTIVECLEAR(fs, cg);
 1685         UFS_UNLOCK(ump);
 1686         bdwrite(bp);
 1687         return (bno);
 1688 
 1689 fail_lock:
 1690         UFS_LOCK(ump);
 1691 fail:
 1692         brelse(bp);
 1693         return (0);
 1694 }
 1695 
 1696 /*
 1697  * Determine whether an inode can be allocated.
 1698  *
 1699  * Check to see if an inode is available, and if it is,
 1700  * allocate it using the following policy:
 1701  *   1) allocate the requested inode.
 1702  *   2) allocate the next available inode after the requested
 1703  *      inode in the specified cylinder group.
 1704  */
 1705 static ufs2_daddr_t
 1706 ffs_nodealloccg(ip, cg, ipref, mode)
 1707         struct inode *ip;
 1708         int cg;
 1709         ufs2_daddr_t ipref;
 1710         int mode;
 1711 {
 1712         struct fs *fs;
 1713         struct cg *cgp;
 1714         struct buf *bp, *ibp;
 1715         struct ufsmount *ump;
 1716         u_int8_t *inosused;
 1717         struct ufs2_dinode *dp2;
 1718         int error, start, len, loc, map, i;
 1719 
 1720         fs = ip->i_fs;
 1721         ump = ip->i_ump;
 1722         if (fs->fs_cs(fs, cg).cs_nifree == 0)
 1723                 return (0);
 1724         UFS_UNLOCK(ump);
 1725         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1726                 (int)fs->fs_cgsize, NOCRED, &bp);
 1727         if (error) {
 1728                 brelse(bp);
 1729                 UFS_LOCK(ump);
 1730                 return (0);
 1731         }
 1732         cgp = (struct cg *)bp->b_data;
 1733         if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
 1734                 brelse(bp);
 1735                 UFS_LOCK(ump);
 1736                 return (0);
 1737         }
 1738         bp->b_xflags |= BX_BKGRDWRITE;
 1739         cgp->cg_old_time = cgp->cg_time = time_second;
 1740         inosused = cg_inosused(cgp);
 1741         if (ipref) {
 1742                 ipref %= fs->fs_ipg;
 1743                 if (isclr(inosused, ipref))
 1744                         goto gotit;
 1745         }
 1746         start = cgp->cg_irotor / NBBY;
 1747         len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
 1748         loc = skpc(0xff, len, &inosused[start]);
 1749         if (loc == 0) {
 1750                 len = start + 1;
 1751                 start = 0;
 1752                 loc = skpc(0xff, len, &inosused[0]);
 1753                 if (loc == 0) {
 1754                         printf("cg = %d, irotor = %ld, fs = %s\n",
 1755                             cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
 1756                         panic("ffs_nodealloccg: map corrupted");
 1757                         /* NOTREACHED */
 1758                 }
 1759         }
 1760         i = start + len - loc;
 1761         map = inosused[i];
 1762         ipref = i * NBBY;
 1763         for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
 1764                 if ((map & i) == 0) {
 1765                         cgp->cg_irotor = ipref;
 1766                         goto gotit;
 1767                 }
 1768         }
 1769         printf("fs = %s\n", fs->fs_fsmnt);
 1770         panic("ffs_nodealloccg: block not in map");
 1771         /* NOTREACHED */
 1772 gotit:
 1773         /*
 1774          * Check to see if we need to initialize more inodes.
 1775          */
 1776         ibp = NULL;
 1777         if (fs->fs_magic == FS_UFS2_MAGIC &&
 1778             ipref + INOPB(fs) > cgp->cg_initediblk &&
 1779             cgp->cg_initediblk < cgp->cg_niblk) {
 1780                 ibp = getblk(ip->i_devvp, fsbtodb(fs,
 1781                     ino_to_fsba(fs, cg * fs->fs_ipg + cgp->cg_initediblk)),
 1782                     (int)fs->fs_bsize, 0, 0, 0);
 1783                 bzero(ibp->b_data, (int)fs->fs_bsize);
 1784                 dp2 = (struct ufs2_dinode *)(ibp->b_data);
 1785                 for (i = 0; i < INOPB(fs); i++) {
 1786                         dp2->di_gen = arc4random() / 2 + 1;
 1787                         dp2++;
 1788                 }
 1789                 cgp->cg_initediblk += INOPB(fs);
 1790         }
 1791         UFS_LOCK(ump);
 1792         ACTIVECLEAR(fs, cg);
 1793         setbit(inosused, ipref);
 1794         cgp->cg_cs.cs_nifree--;
 1795         fs->fs_cstotal.cs_nifree--;
 1796         fs->fs_cs(fs, cg).cs_nifree--;
 1797         fs->fs_fmod = 1;
 1798         if ((mode & IFMT) == IFDIR) {
 1799                 cgp->cg_cs.cs_ndir++;
 1800                 fs->fs_cstotal.cs_ndir++;
 1801                 fs->fs_cs(fs, cg).cs_ndir++;
 1802         }
 1803         UFS_UNLOCK(ump);
 1804         if (DOINGSOFTDEP(ITOV(ip)))
 1805                 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref);
 1806         bdwrite(bp);
 1807         if (ibp != NULL)
 1808                 bawrite(ibp);
 1809         return (cg * fs->fs_ipg + ipref);
 1810 }
 1811 
 1812 /*
 1813  * check if a block is free
 1814  */
 1815 static int
 1816 ffs_isfreeblock(struct fs *fs, u_char *cp, ufs1_daddr_t h)
 1817 {
 1818 
 1819         switch ((int)fs->fs_frag) {
 1820         case 8:
 1821                 return (cp[h] == 0);
 1822         case 4:
 1823                 return ((cp[h >> 1] & (0x0f << ((h & 0x1) << 2))) == 0);
 1824         case 2:
 1825                 return ((cp[h >> 2] & (0x03 << ((h & 0x3) << 1))) == 0);
 1826         case 1:
 1827                 return ((cp[h >> 3] & (0x01 << (h & 0x7))) == 0);
 1828         default:
 1829                 panic("ffs_isfreeblock");
 1830         }
 1831         return (0);
 1832 }
 1833 
 1834 /*
 1835  * Free a block or fragment.
 1836  *
 1837  * The specified block or fragment is placed back in the
 1838  * free map. If a fragment is deallocated, a possible
 1839  * block reassembly is checked.
 1840  */
 1841 void
 1842 ffs_blkfree(ump, fs, devvp, bno, size, inum)
 1843         struct ufsmount *ump;
 1844         struct fs *fs;
 1845         struct vnode *devvp;
 1846         ufs2_daddr_t bno;
 1847         long size;
 1848         ino_t inum;
 1849 {
 1850         struct cg *cgp;
 1851         struct buf *bp;
 1852         ufs1_daddr_t fragno, cgbno;
 1853         ufs2_daddr_t cgblkno;
 1854         int i, cg, blk, frags, bbase;
 1855         u_int8_t *blksfree;
 1856         struct cdev *dev;
 1857 
 1858         cg = dtog(fs, bno);
 1859         if (devvp->v_type == VREG) {
 1860                 /* devvp is a snapshot */
 1861                 dev = VTOI(devvp)->i_devvp->v_rdev;
 1862                 cgblkno = fragstoblks(fs, cgtod(fs, cg));
 1863         } else {
 1864                 /* devvp is a normal disk device */
 1865                 dev = devvp->v_rdev;
 1866                 cgblkno = fsbtodb(fs, cgtod(fs, cg));
 1867                 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree");
 1868                 if ((devvp->v_vflag & VV_COPYONWRITE) &&
 1869                     ffs_snapblkfree(fs, devvp, bno, size, inum))
 1870                         return;
 1871         }
 1872 #ifdef INVARIANTS
 1873         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
 1874             fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
 1875                 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n",
 1876                     devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize,
 1877                     size, fs->fs_fsmnt);
 1878                 panic("ffs_blkfree: bad size");
 1879         }
 1880 #endif
 1881         if ((u_int)bno >= fs->fs_size) {
 1882                 printf("bad block %jd, ino %lu\n", (intmax_t)bno,
 1883                     (u_long)inum);
 1884                 ffs_fserr(fs, inum, "bad block");
 1885                 return;
 1886         }
 1887         if (bread(devvp, cgblkno, (int)fs->fs_cgsize, NOCRED, &bp)) {
 1888                 brelse(bp);
 1889                 return;
 1890         }
 1891         cgp = (struct cg *)bp->b_data;
 1892         if (!cg_chkmagic(cgp)) {
 1893                 brelse(bp);
 1894                 return;
 1895         }
 1896         bp->b_xflags |= BX_BKGRDWRITE;
 1897         cgp->cg_old_time = cgp->cg_time = time_second;
 1898         cgbno = dtogd(fs, bno);
 1899         blksfree = cg_blksfree(cgp);
 1900         UFS_LOCK(ump);
 1901         if (size == fs->fs_bsize) {
 1902                 fragno = fragstoblks(fs, cgbno);
 1903                 if (!ffs_isfreeblock(fs, blksfree, fragno)) {
 1904                         if (devvp->v_type == VREG) {
 1905                                 UFS_UNLOCK(ump);
 1906                                 /* devvp is a snapshot */
 1907                                 brelse(bp);
 1908                                 return;
 1909                         }
 1910                         printf("dev = %s, block = %jd, fs = %s\n",
 1911                             devtoname(dev), (intmax_t)bno, fs->fs_fsmnt);
 1912                         panic("ffs_blkfree: freeing free block");
 1913                 }
 1914                 ffs_setblock(fs, blksfree, fragno);
 1915                 ffs_clusteracct(ump, fs, cgp, fragno, 1);
 1916                 cgp->cg_cs.cs_nbfree++;
 1917                 fs->fs_cstotal.cs_nbfree++;
 1918                 fs->fs_cs(fs, cg).cs_nbfree++;
 1919         } else {
 1920                 bbase = cgbno - fragnum(fs, cgbno);
 1921                 /*
 1922                  * decrement the counts associated with the old frags
 1923                  */
 1924                 blk = blkmap(fs, blksfree, bbase);
 1925                 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
 1926                 /*
 1927                  * deallocate the fragment
 1928                  */
 1929                 frags = numfrags(fs, size);
 1930                 for (i = 0; i < frags; i++) {
 1931                         if (isset(blksfree, cgbno + i)) {
 1932                                 printf("dev = %s, block = %jd, fs = %s\n",
 1933                                     devtoname(dev), (intmax_t)(bno + i),
 1934                                     fs->fs_fsmnt);
 1935                                 panic("ffs_blkfree: freeing free frag");
 1936                         }
 1937                         setbit(blksfree, cgbno + i);
 1938                 }
 1939                 cgp->cg_cs.cs_nffree += i;
 1940                 fs->fs_cstotal.cs_nffree += i;
 1941                 fs->fs_cs(fs, cg).cs_nffree += i;
 1942                 /*
 1943                  * add back in counts associated with the new frags
 1944                  */
 1945                 blk = blkmap(fs, blksfree, bbase);
 1946                 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
 1947                 /*
 1948                  * if a complete block has been reassembled, account for it
 1949                  */
 1950                 fragno = fragstoblks(fs, bbase);
 1951                 if (ffs_isblock(fs, blksfree, fragno)) {
 1952                         cgp->cg_cs.cs_nffree -= fs->fs_frag;
 1953                         fs->fs_cstotal.cs_nffree -= fs->fs_frag;
 1954                         fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
 1955                         ffs_clusteracct(ump, fs, cgp, fragno, 1);
 1956                         cgp->cg_cs.cs_nbfree++;
 1957                         fs->fs_cstotal.cs_nbfree++;
 1958                         fs->fs_cs(fs, cg).cs_nbfree++;
 1959                 }
 1960         }
 1961         fs->fs_fmod = 1;
 1962         ACTIVECLEAR(fs, cg);
 1963         UFS_UNLOCK(ump);
 1964         bdwrite(bp);
 1965 }
 1966 
 1967 #ifdef INVARIANTS
 1968 /*
 1969  * Verify allocation of a block or fragment. Returns true if block or
 1970  * fragment is allocated, false if it is free.
 1971  */
 1972 static int
 1973 ffs_checkblk(ip, bno, size)
 1974         struct inode *ip;
 1975         ufs2_daddr_t bno;
 1976         long size;
 1977 {
 1978         struct fs *fs;
 1979         struct cg *cgp;
 1980         struct buf *bp;
 1981         ufs1_daddr_t cgbno;
 1982         int i, error, frags, free;
 1983         u_int8_t *blksfree;
 1984 
 1985         fs = ip->i_fs;
 1986         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
 1987                 printf("bsize = %ld, size = %ld, fs = %s\n",
 1988                     (long)fs->fs_bsize, size, fs->fs_fsmnt);
 1989                 panic("ffs_checkblk: bad size");
 1990         }
 1991         if ((u_int)bno >= fs->fs_size)
 1992                 panic("ffs_checkblk: bad block %jd", (intmax_t)bno);
 1993         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
 1994                 (int)fs->fs_cgsize, NOCRED, &bp);
 1995         if (error)
 1996                 panic("ffs_checkblk: cg bread failed");
 1997         cgp = (struct cg *)bp->b_data;
 1998         if (!cg_chkmagic(cgp))
 1999                 panic("ffs_checkblk: cg magic mismatch");
 2000         bp->b_xflags |= BX_BKGRDWRITE;
 2001         blksfree = cg_blksfree(cgp);
 2002         cgbno = dtogd(fs, bno);
 2003         if (size == fs->fs_bsize) {
 2004                 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno));
 2005         } else {
 2006                 frags = numfrags(fs, size);
 2007                 for (free = 0, i = 0; i < frags; i++)
 2008                         if (isset(blksfree, cgbno + i))
 2009                                 free++;
 2010                 if (free != 0 && free != frags)
 2011                         panic("ffs_checkblk: partially free fragment");
 2012         }
 2013         brelse(bp);
 2014         return (!free);
 2015 }
 2016 #endif /* INVARIANTS */
 2017 
 2018 /*
 2019  * Free an inode.
 2020  */
 2021 int
 2022 ffs_vfree(pvp, ino, mode)
 2023         struct vnode *pvp;
 2024         ino_t ino;
 2025         int mode;
 2026 {
 2027         struct inode *ip;
 2028 
 2029         if (DOINGSOFTDEP(pvp)) {
 2030                 softdep_freefile(pvp, ino, mode);
 2031                 return (0);
 2032         }
 2033         ip = VTOI(pvp);
 2034         return (ffs_freefile(ip->i_ump, ip->i_fs, ip->i_devvp, ino, mode));
 2035 }
 2036 
 2037 /*
 2038  * Do the actual free operation.
 2039  * The specified inode is placed back in the free map.
 2040  */
 2041 int
 2042 ffs_freefile(ump, fs, devvp, ino, mode)
 2043         struct ufsmount *ump;
 2044         struct fs *fs;
 2045         struct vnode *devvp;
 2046         ino_t ino;
 2047         int mode;
 2048 {
 2049         struct cg *cgp;
 2050         struct buf *bp;
 2051         ufs2_daddr_t cgbno;
 2052         int error, cg;
 2053         u_int8_t *inosused;
 2054         struct cdev *dev;
 2055 
 2056         cg = ino_to_cg(fs, ino);
 2057         if (devvp->v_type == VREG) {
 2058                 /* devvp is a snapshot */
 2059                 dev = VTOI(devvp)->i_devvp->v_rdev;
 2060                 cgbno = fragstoblks(fs, cgtod(fs, cg));
 2061         } else {
 2062                 /* devvp is a normal disk device */
 2063                 dev = devvp->v_rdev;
 2064                 cgbno = fsbtodb(fs, cgtod(fs, cg));
 2065         }
 2066         if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
 2067                 panic("ffs_freefile: range: dev = %s, ino = %lu, fs = %s",
 2068                     devtoname(dev), (u_long)ino, fs->fs_fsmnt);
 2069         if ((error = bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp))) {
 2070                 brelse(bp);
 2071                 return (error);
 2072         }
 2073         cgp = (struct cg *)bp->b_data;
 2074         if (!cg_chkmagic(cgp)) {
 2075                 brelse(bp);
 2076                 return (0);
 2077         }
 2078         bp->b_xflags |= BX_BKGRDWRITE;
 2079         cgp->cg_old_time = cgp->cg_time = time_second;
 2080         inosused = cg_inosused(cgp);
 2081         ino %= fs->fs_ipg;
 2082         if (isclr(inosused, ino)) {
 2083                 printf("dev = %s, ino = %lu, fs = %s\n", devtoname(dev),
 2084                     (u_long)ino + cg * fs->fs_ipg, fs->fs_fsmnt);
 2085                 if (fs->fs_ronly == 0)
 2086                         panic("ffs_freefile: freeing free inode");
 2087         }
 2088         clrbit(inosused, ino);
 2089         if (ino < cgp->cg_irotor)
 2090                 cgp->cg_irotor = ino;
 2091         cgp->cg_cs.cs_nifree++;
 2092         UFS_LOCK(ump);
 2093         fs->fs_cstotal.cs_nifree++;
 2094         fs->fs_cs(fs, cg).cs_nifree++;
 2095         if ((mode & IFMT) == IFDIR) {
 2096                 cgp->cg_cs.cs_ndir--;
 2097                 fs->fs_cstotal.cs_ndir--;
 2098                 fs->fs_cs(fs, cg).cs_ndir--;
 2099         }
 2100         fs->fs_fmod = 1;
 2101         ACTIVECLEAR(fs, cg);
 2102         UFS_UNLOCK(ump);
 2103         bdwrite(bp);
 2104         return (0);
 2105 }
 2106 
 2107 /*
 2108  * Check to see if a file is free.
 2109  */
 2110 int
 2111 ffs_checkfreefile(fs, devvp, ino)
 2112         struct fs *fs;
 2113         struct vnode *devvp;
 2114         ino_t ino;
 2115 {
 2116         struct cg *cgp;
 2117         struct buf *bp;
 2118         ufs2_daddr_t cgbno;
 2119         int ret, cg;
 2120         u_int8_t *inosused;
 2121 
 2122         cg = ino_to_cg(fs, ino);
 2123         if (devvp->v_type == VREG) {
 2124                 /* devvp is a snapshot */
 2125                 cgbno = fragstoblks(fs, cgtod(fs, cg));
 2126         } else {
 2127                 /* devvp is a normal disk device */
 2128                 cgbno = fsbtodb(fs, cgtod(fs, cg));
 2129         }
 2130         if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
 2131                 return (1);
 2132         if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp)) {
 2133                 brelse(bp);
 2134                 return (1);
 2135         }
 2136         cgp = (struct cg *)bp->b_data;
 2137         if (!cg_chkmagic(cgp)) {
 2138                 brelse(bp);
 2139                 return (1);
 2140         }
 2141         inosused = cg_inosused(cgp);
 2142         ino %= fs->fs_ipg;
 2143         ret = isclr(inosused, ino);
 2144         brelse(bp);
 2145         return (ret);
 2146 }
 2147 
 2148 /*
 2149  * Find a block of the specified size in the specified cylinder group.
 2150  *
 2151  * It is a panic if a request is made to find a block if none are
 2152  * available.
 2153  */
 2154 static ufs1_daddr_t
 2155 ffs_mapsearch(fs, cgp, bpref, allocsiz)
 2156         struct fs *fs;
 2157         struct cg *cgp;
 2158         ufs2_daddr_t bpref;
 2159         int allocsiz;
 2160 {
 2161         ufs1_daddr_t bno;
 2162         int start, len, loc, i;
 2163         int blk, field, subfield, pos;
 2164         u_int8_t *blksfree;
 2165 
 2166         /*
 2167          * find the fragment by searching through the free block
 2168          * map for an appropriate bit pattern
 2169          */
 2170         if (bpref)
 2171                 start = dtogd(fs, bpref) / NBBY;
 2172         else
 2173                 start = cgp->cg_frotor / NBBY;
 2174         blksfree = cg_blksfree(cgp);
 2175         len = howmany(fs->fs_fpg, NBBY) - start;
 2176         loc = scanc((u_int)len, (u_char *)&blksfree[start],
 2177                 fragtbl[fs->fs_frag],
 2178                 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
 2179         if (loc == 0) {
 2180                 len = start + 1;
 2181                 start = 0;
 2182                 loc = scanc((u_int)len, (u_char *)&blksfree[0],
 2183                         fragtbl[fs->fs_frag],
 2184                         (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
 2185                 if (loc == 0) {
 2186                         printf("start = %d, len = %d, fs = %s\n",
 2187                             start, len, fs->fs_fsmnt);
 2188                         panic("ffs_alloccg: map corrupted");
 2189                         /* NOTREACHED */
 2190                 }
 2191         }
 2192         bno = (start + len - loc) * NBBY;
 2193         cgp->cg_frotor = bno;
 2194         /*
 2195          * found the byte in the map
 2196          * sift through the bits to find the selected frag
 2197          */
 2198         for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
 2199                 blk = blkmap(fs, blksfree, bno);
 2200                 blk <<= 1;
 2201                 field = around[allocsiz];
 2202                 subfield = inside[allocsiz];
 2203                 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
 2204                         if ((blk & field) == subfield)
 2205                                 return (bno + pos);
 2206                         field <<= 1;
 2207                         subfield <<= 1;
 2208                 }
 2209         }
 2210         printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
 2211         panic("ffs_alloccg: block not in map");
 2212         return (-1);
 2213 }
 2214 
 2215 /*
 2216  * Update the cluster map because of an allocation or free.
 2217  *
 2218  * Cnt == 1 means free; cnt == -1 means allocating.
 2219  */
 2220 void
 2221 ffs_clusteracct(ump, fs, cgp, blkno, cnt)
 2222         struct ufsmount *ump;
 2223         struct fs *fs;
 2224         struct cg *cgp;
 2225         ufs1_daddr_t blkno;
 2226         int cnt;
 2227 {
 2228         int32_t *sump;
 2229         int32_t *lp;
 2230         u_char *freemapp, *mapp;
 2231         int i, start, end, forw, back, map, bit;
 2232 
 2233         mtx_assert(UFS_MTX(ump), MA_OWNED);
 2234 
 2235         if (fs->fs_contigsumsize <= 0)
 2236                 return;
 2237         freemapp = cg_clustersfree(cgp);
 2238         sump = cg_clustersum(cgp);
 2239         /*
 2240          * Allocate or clear the actual block.
 2241          */
 2242         if (cnt > 0)
 2243                 setbit(freemapp, blkno);
 2244         else
 2245                 clrbit(freemapp, blkno);
 2246         /*
 2247          * Find the size of the cluster going forward.
 2248          */
 2249         start = blkno + 1;
 2250         end = start + fs->fs_contigsumsize;
 2251         if (end >= cgp->cg_nclusterblks)
 2252                 end = cgp->cg_nclusterblks;
 2253         mapp = &freemapp[start / NBBY];
 2254         map = *mapp++;
 2255         bit = 1 << (start % NBBY);
 2256         for (i = start; i < end; i++) {
 2257                 if ((map & bit) == 0)
 2258                         break;
 2259                 if ((i & (NBBY - 1)) != (NBBY - 1)) {
 2260                         bit <<= 1;
 2261                 } else {
 2262                         map = *mapp++;
 2263                         bit = 1;
 2264                 }
 2265         }
 2266         forw = i - start;
 2267         /*
 2268          * Find the size of the cluster going backward.
 2269          */
 2270         start = blkno - 1;
 2271         end = start - fs->fs_contigsumsize;
 2272         if (end < 0)
 2273                 end = -1;
 2274         mapp = &freemapp[start / NBBY];
 2275         map = *mapp--;
 2276         bit = 1 << (start % NBBY);
 2277         for (i = start; i > end; i--) {
 2278                 if ((map & bit) == 0)
 2279                         break;
 2280                 if ((i & (NBBY - 1)) != 0) {
 2281                         bit >>= 1;
 2282                 } else {
 2283                         map = *mapp--;
 2284                         bit = 1 << (NBBY - 1);
 2285                 }
 2286         }
 2287         back = start - i;
 2288         /*
 2289          * Account for old cluster and the possibly new forward and
 2290          * back clusters.
 2291          */
 2292         i = back + forw + 1;
 2293         if (i > fs->fs_contigsumsize)
 2294                 i = fs->fs_contigsumsize;
 2295         sump[i] += cnt;
 2296         if (back > 0)
 2297                 sump[back] -= cnt;
 2298         if (forw > 0)
 2299                 sump[forw] -= cnt;
 2300         /*
 2301          * Update cluster summary information.
 2302          */
 2303         lp = &sump[fs->fs_contigsumsize];
 2304         for (i = fs->fs_contigsumsize; i > 0; i--)
 2305                 if (*lp-- > 0)
 2306                         break;
 2307         fs->fs_maxcluster[cgp->cg_cgx] = i;
 2308 }
 2309 
 2310 /*
 2311  * Fserr prints the name of a filesystem with an error diagnostic.
 2312  *
 2313  * The form of the error message is:
 2314  *      fs: error message
 2315  */
 2316 static void
 2317 ffs_fserr(fs, inum, cp)
 2318         struct fs *fs;
 2319         ino_t inum;
 2320         char *cp;
 2321 {
 2322         struct thread *td = curthread;  /* XXX */
 2323         struct proc *p = td->td_proc;
 2324 
 2325         log(LOG_ERR, "pid %d (%s), uid %d inumber %d on %s: %s\n",
 2326             p->p_pid, p->p_comm, td->td_ucred->cr_uid, inum, fs->fs_fsmnt, cp);
 2327 }
 2328 
 2329 /*
 2330  * This function provides the capability for the fsck program to
 2331  * update an active filesystem. Eleven operations are provided:
 2332  *
 2333  * adjrefcnt(inode, amt) - adjusts the reference count on the
 2334  *      specified inode by the specified amount. Under normal
 2335  *      operation the count should always go down. Decrementing
 2336  *      the count to zero will cause the inode to be freed.
 2337  * adjblkcnt(inode, amt) - adjust the number of blocks used to
 2338  *      by the specifed amount.
 2339  * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) -
 2340  *      adjust the superblock summary.
 2341  * freedirs(inode, count) - directory inodes [inode..inode + count - 1]
 2342  *      are marked as free. Inodes should never have to be marked
 2343  *      as in use.
 2344  * freefiles(inode, count) - file inodes [inode..inode + count - 1]
 2345  *      are marked as free. Inodes should never have to be marked
 2346  *      as in use.
 2347  * freeblks(blockno, size) - blocks [blockno..blockno + size - 1]
 2348  *      are marked as free. Blocks should never have to be marked
 2349  *      as in use.
 2350  * setflags(flags, set/clear) - the fs_flags field has the specified
 2351  *      flags set (second parameter +1) or cleared (second parameter -1).
 2352  */
 2353 
 2354 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS);
 2355 
 2356 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT,
 2357         0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count");
 2358 
 2359 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR,
 2360         sysctl_ffs_fsck, "Adjust Inode Used Blocks Count");
 2361 
 2362 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR,
 2363         sysctl_ffs_fsck, "Adjust number of directories");
 2364 
 2365 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR,
 2366         sysctl_ffs_fsck, "Adjust number of free blocks");
 2367 
 2368 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR,
 2369         sysctl_ffs_fsck, "Adjust number of free inodes");
 2370 
 2371 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR,
 2372         sysctl_ffs_fsck, "Adjust number of free frags");
 2373 
 2374 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR,
 2375         sysctl_ffs_fsck, "Adjust number of free clusters");
 2376 
 2377 static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR,
 2378         sysctl_ffs_fsck, "Free Range of Directory Inodes");
 2379 
 2380 static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR,
 2381         sysctl_ffs_fsck, "Free Range of File Inodes");
 2382 
 2383 static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR,
 2384         sysctl_ffs_fsck, "Free Range of Blocks");
 2385 
 2386 static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR,
 2387         sysctl_ffs_fsck, "Change Filesystem Flags");
 2388 
 2389 #ifdef DEBUG
 2390 static int fsckcmds = 0;
 2391 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, "");
 2392 #endif /* DEBUG */
 2393 
 2394 static int
 2395 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS)
 2396 {
 2397         struct fsck_cmd cmd;
 2398         struct ufsmount *ump;
 2399         struct vnode *vp;
 2400         struct inode *ip;
 2401         struct mount *mp;
 2402         struct fs *fs;
 2403         ufs2_daddr_t blkno;
 2404         long blkcnt, blksize;
 2405         struct file *fp;
 2406         int filetype, error;
 2407 
 2408         if (req->newlen > sizeof cmd)
 2409                 return (EBADRPC);
 2410         if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0)
 2411                 return (error);
 2412         if (cmd.version != FFS_CMD_VERSION)
 2413                 return (ERPCMISMATCH);
 2414         if ((error = getvnode(curproc->p_fd, cmd.handle, &fp)) != 0)
 2415                 return (error);
 2416         vn_start_write(fp->f_data, &mp, V_WAIT);
 2417         if (mp == 0 || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) {
 2418                 vn_finished_write(mp);
 2419                 fdrop(fp, curthread);
 2420                 return (EINVAL);
 2421         }
 2422         if (mp->mnt_flag & MNT_RDONLY) {
 2423                 vn_finished_write(mp);
 2424                 fdrop(fp, curthread);
 2425                 return (EROFS);
 2426         }
 2427         ump = VFSTOUFS(mp);
 2428         fs = ump->um_fs;
 2429         filetype = IFREG;
 2430 
 2431         switch (oidp->oid_number) {
 2432 
 2433         case FFS_SET_FLAGS:
 2434 #ifdef DEBUG
 2435                 if (fsckcmds)
 2436                         printf("%s: %s flags\n", mp->mnt_stat.f_mntonname,
 2437                             cmd.size > 0 ? "set" : "clear");
 2438 #endif /* DEBUG */
 2439                 if (cmd.size > 0)
 2440                         fs->fs_flags |= (long)cmd.value;
 2441                 else
 2442                         fs->fs_flags &= ~(long)cmd.value;
 2443                 break;
 2444 
 2445         case FFS_ADJ_REFCNT:
 2446 #ifdef DEBUG
 2447                 if (fsckcmds) {
 2448                         printf("%s: adjust inode %jd count by %jd\n",
 2449                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
 2450                             (intmax_t)cmd.size);
 2451                 }
 2452 #endif /* DEBUG */
 2453                 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
 2454                         break;
 2455                 ip = VTOI(vp);
 2456                 ip->i_nlink += cmd.size;
 2457                 DIP_SET(ip, i_nlink, ip->i_nlink);
 2458                 ip->i_effnlink += cmd.size;
 2459                 ip->i_flag |= IN_CHANGE;
 2460                 if (DOINGSOFTDEP(vp))
 2461                         softdep_change_linkcnt(ip);
 2462                 vput(vp);
 2463                 break;
 2464 
 2465         case FFS_ADJ_BLKCNT:
 2466 #ifdef DEBUG
 2467                 if (fsckcmds) {
 2468                         printf("%s: adjust inode %jd block count by %jd\n",
 2469                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
 2470                             (intmax_t)cmd.size);
 2471                 }
 2472 #endif /* DEBUG */
 2473                 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
 2474                         break;
 2475                 ip = VTOI(vp);
 2476                 if (ip->i_flag & IN_SPACECOUNTED) {
 2477                         UFS_LOCK(ump);
 2478                         fs->fs_pendingblocks += cmd.size;
 2479                         UFS_UNLOCK(ump);
 2480                 }
 2481                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size);
 2482                 ip->i_flag |= IN_CHANGE;
 2483                 vput(vp);
 2484                 break;
 2485 
 2486         case FFS_DIR_FREE:
 2487                 filetype = IFDIR;
 2488                 /* fall through */
 2489 
 2490         case FFS_FILE_FREE:
 2491 #ifdef DEBUG
 2492                 if (fsckcmds) {
 2493                         if (cmd.size == 1)
 2494                                 printf("%s: free %s inode %d\n",
 2495                                     mp->mnt_stat.f_mntonname,
 2496                                     filetype == IFDIR ? "directory" : "file",
 2497                                     (ino_t)cmd.value);
 2498                         else
 2499                                 printf("%s: free %s inodes %d-%d\n",
 2500                                     mp->mnt_stat.f_mntonname,
 2501                                     filetype == IFDIR ? "directory" : "file",
 2502                                     (ino_t)cmd.value,
 2503                                     (ino_t)(cmd.value + cmd.size - 1));
 2504                 }
 2505 #endif /* DEBUG */
 2506                 while (cmd.size > 0) {
 2507                         if ((error = ffs_freefile(ump, fs, ump->um_devvp,
 2508                             cmd.value, filetype)))
 2509                                 break;
 2510                         cmd.size -= 1;
 2511                         cmd.value += 1;
 2512                 }
 2513                 break;
 2514 
 2515         case FFS_BLK_FREE:
 2516 #ifdef DEBUG
 2517                 if (fsckcmds) {
 2518                         if (cmd.size == 1)
 2519                                 printf("%s: free block %jd\n",
 2520                                     mp->mnt_stat.f_mntonname,
 2521                                     (intmax_t)cmd.value);
 2522                         else
 2523                                 printf("%s: free blocks %jd-%jd\n",
 2524                                     mp->mnt_stat.f_mntonname, 
 2525                                     (intmax_t)cmd.value,
 2526                                     (intmax_t)cmd.value + cmd.size - 1);
 2527                 }
 2528 #endif /* DEBUG */
 2529                 blkno = cmd.value;
 2530                 blkcnt = cmd.size;
 2531                 blksize = fs->fs_frag - (blkno % fs->fs_frag);
 2532                 while (blkcnt > 0) {
 2533                         if (blksize > blkcnt)
 2534                                 blksize = blkcnt;
 2535                         ffs_blkfree(ump, fs, ump->um_devvp, blkno,
 2536                             blksize * fs->fs_fsize, ROOTINO);
 2537                         blkno += blksize;
 2538                         blkcnt -= blksize;
 2539                         blksize = fs->fs_frag;
 2540                 }
 2541                 break;
 2542 
 2543         /*
 2544          * Adjust superblock summaries.  fsck(8) is expected to
 2545          * submit deltas when necessary.
 2546          */
 2547         case FFS_ADJ_NDIR:
 2548 #ifdef DEBUG
 2549                 if (fsckcmds) {
 2550                         printf("%s: adjust number of directories by %jd\n",
 2551                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2552                 }
 2553 #endif /* DEBUG */
 2554                 fs->fs_cstotal.cs_ndir += cmd.value;
 2555                 break;
 2556         case FFS_ADJ_NBFREE:
 2557 #ifdef DEBUG
 2558                 if (fsckcmds) {
 2559                         printf("%s: adjust number of free blocks by %+jd\n",
 2560                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2561                 }
 2562 #endif /* DEBUG */
 2563                 fs->fs_cstotal.cs_nbfree += cmd.value;
 2564                 break;
 2565         case FFS_ADJ_NIFREE:
 2566 #ifdef DEBUG
 2567                 if (fsckcmds) {
 2568                         printf("%s: adjust number of free inodes by %+jd\n",
 2569                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2570                 }
 2571 #endif /* DEBUG */
 2572                 fs->fs_cstotal.cs_nifree += cmd.value;
 2573                 break;
 2574         case FFS_ADJ_NFFREE:
 2575 #ifdef DEBUG
 2576                 if (fsckcmds) {
 2577                         printf("%s: adjust number of free frags by %+jd\n",
 2578                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2579                 }
 2580 #endif /* DEBUG */
 2581                 fs->fs_cstotal.cs_nffree += cmd.value;
 2582                 break;
 2583         case FFS_ADJ_NUMCLUSTERS:
 2584 #ifdef DEBUG
 2585                 if (fsckcmds) {
 2586                         printf("%s: adjust number of free clusters by %+jd\n",
 2587                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2588                 }
 2589 #endif /* DEBUG */
 2590                 fs->fs_cstotal.cs_numclusters += cmd.value;
 2591                 break;
 2592 
 2593         default:
 2594 #ifdef DEBUG
 2595                 if (fsckcmds) {
 2596                         printf("Invalid request %d from fsck\n",
 2597                             oidp->oid_number);
 2598                 }
 2599 #endif /* DEBUG */
 2600                 error = EINVAL;
 2601                 break;
 2602 
 2603         }
 2604         fdrop(fp, curthread);
 2605         vn_finished_write(mp);
 2606         return (error);
 2607 }

Cache object: 4470ba990fa7c2f432a08e4622cd452c


[ 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.