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/8.2/sys/ufs/ffs/ffs_alloc.c 204375 2010-02-26 21:49:11Z mckusick $");
   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, u_int cg, ufs2_daddr_t bpref,
   92                                   int size);
   93 
   94 static ufs2_daddr_t ffs_alloccg(struct inode *, u_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 *, u_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 *, u_int, ufs2_daddr_t,
  105                     int, int);
  106 static void     ffs_fserr(struct fs *, ino_t, char *);
  107 static ufs2_daddr_t     ffs_hashalloc
  108                 (struct inode *, u_int, ufs2_daddr_t, int, allocfcn_t *);
  109 static ufs2_daddr_t ffs_nodealloccg(struct inode *, u_int, ufs2_daddr_t, int);
  110 static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int);
  111 static int      ffs_reallocblks_ufs1(struct vop_reallocblks_args *);
  112 static int      ffs_reallocblks_ufs2(struct vop_reallocblks_args *);
  113 
  114 /*
  115  * Allocate a block in the filesystem.
  116  *
  117  * The size of the requested block is given, which must be some
  118  * multiple of fs_fsize and <= fs_bsize.
  119  * A preference may be optionally specified. If a preference is given
  120  * the following hierarchy is used to allocate a block:
  121  *   1) allocate the requested block.
  122  *   2) allocate a rotationally optimal block in the same cylinder.
  123  *   3) allocate a block in the same cylinder group.
  124  *   4) quadradically rehash into other cylinder groups, until an
  125  *      available block is located.
  126  * If no block preference is given the following hierarchy is used
  127  * to allocate a block:
  128  *   1) allocate a block in the cylinder group that contains the
  129  *      inode for the file.
  130  *   2) quadradically rehash into other cylinder groups, until an
  131  *      available block is located.
  132  */
  133 int
  134 ffs_alloc(ip, lbn, bpref, size, flags, cred, bnp)
  135         struct inode *ip;
  136         ufs2_daddr_t lbn, bpref;
  137         int size, flags;
  138         struct ucred *cred;
  139         ufs2_daddr_t *bnp;
  140 {
  141         struct fs *fs;
  142         struct ufsmount *ump;
  143         ufs2_daddr_t bno;
  144         u_int cg, reclaimed;
  145         static struct timeval lastfail;
  146         static int curfail;
  147         int64_t delta;
  148 #ifdef QUOTA
  149         int error;
  150 #endif
  151 
  152         *bnp = 0;
  153         fs = ip->i_fs;
  154         ump = ip->i_ump;
  155         mtx_assert(UFS_MTX(ump), MA_OWNED);
  156 #ifdef INVARIANTS
  157         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
  158                 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n",
  159                     devtoname(ip->i_dev), (long)fs->fs_bsize, size,
  160                     fs->fs_fsmnt);
  161                 panic("ffs_alloc: bad size");
  162         }
  163         if (cred == NOCRED)
  164                 panic("ffs_alloc: missing credential");
  165 #endif /* INVARIANTS */
  166         reclaimed = 0;
  167 retry:
  168 #ifdef QUOTA
  169         UFS_UNLOCK(ump);
  170         error = chkdq(ip, btodb(size), cred, 0);
  171         if (error)
  172                 return (error);
  173         UFS_LOCK(ump);
  174 #endif
  175         if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
  176                 goto nospace;
  177         if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) &&
  178             freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
  179                 goto nospace;
  180         if (bpref >= fs->fs_size)
  181                 bpref = 0;
  182         if (bpref == 0)
  183                 cg = ino_to_cg(fs, ip->i_number);
  184         else
  185                 cg = dtog(fs, bpref);
  186         bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
  187         if (bno > 0) {
  188                 delta = btodb(size);
  189                 if (ip->i_flag & IN_SPACECOUNTED) {
  190                         UFS_LOCK(ump);
  191                         fs->fs_pendingblocks += delta;
  192                         UFS_UNLOCK(ump);
  193                 }
  194                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  195                 if (flags & IO_EXT)
  196                         ip->i_flag |= IN_CHANGE;
  197                 else
  198                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
  199                 *bnp = bno;
  200                 return (0);
  201         }
  202 nospace:
  203 #ifdef QUOTA
  204         UFS_UNLOCK(ump);
  205         /*
  206          * Restore user's disk quota because allocation failed.
  207          */
  208         (void) chkdq(ip, -btodb(size), cred, FORCE);
  209         UFS_LOCK(ump);
  210 #endif
  211         if (fs->fs_pendingblocks > 0 && reclaimed == 0) {
  212                 reclaimed = 1;
  213                 softdep_request_cleanup(fs, ITOV(ip));
  214                 goto retry;
  215         }
  216         UFS_UNLOCK(ump);
  217         if (ppsratecheck(&lastfail, &curfail, 1)) {
  218                 ffs_fserr(fs, ip->i_number, "filesystem full");
  219                 uprintf("\n%s: write failed, filesystem is full\n",
  220                     fs->fs_fsmnt);
  221         }
  222         return (ENOSPC);
  223 }
  224 
  225 /*
  226  * Reallocate a fragment to a bigger size
  227  *
  228  * The number and size of the old block is given, and a preference
  229  * and new size is also specified. The allocator attempts to extend
  230  * the original block. Failing that, the regular block allocator is
  231  * invoked to get an appropriate block.
  232  */
  233 int
  234 ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, flags, cred, bpp)
  235         struct inode *ip;
  236         ufs2_daddr_t lbprev;
  237         ufs2_daddr_t bprev;
  238         ufs2_daddr_t bpref;
  239         int osize, nsize, flags;
  240         struct ucred *cred;
  241         struct buf **bpp;
  242 {
  243         struct vnode *vp;
  244         struct fs *fs;
  245         struct buf *bp;
  246         struct ufsmount *ump;
  247         u_int cg, request, reclaimed;
  248         int error;
  249         ufs2_daddr_t bno;
  250         static struct timeval lastfail;
  251         static int curfail;
  252         int64_t delta;
  253 
  254         *bpp = 0;
  255         vp = ITOV(ip);
  256         fs = ip->i_fs;
  257         bp = NULL;
  258         ump = ip->i_ump;
  259         mtx_assert(UFS_MTX(ump), MA_OWNED);
  260 #ifdef INVARIANTS
  261         if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
  262                 panic("ffs_realloccg: allocation on suspended filesystem");
  263         if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
  264             (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
  265                 printf(
  266                 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
  267                     devtoname(ip->i_dev), (long)fs->fs_bsize, osize,
  268                     nsize, fs->fs_fsmnt);
  269                 panic("ffs_realloccg: bad size");
  270         }
  271         if (cred == NOCRED)
  272                 panic("ffs_realloccg: missing credential");
  273 #endif /* INVARIANTS */
  274         reclaimed = 0;
  275 retry:
  276         if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE, 0) &&
  277             freespace(fs, fs->fs_minfree) -  numfrags(fs, nsize - osize) < 0) {
  278                 goto nospace;
  279         }
  280         if (bprev == 0) {
  281                 printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n",
  282                     devtoname(ip->i_dev), (long)fs->fs_bsize, (intmax_t)bprev,
  283                     fs->fs_fsmnt);
  284                 panic("ffs_realloccg: bad bprev");
  285         }
  286         UFS_UNLOCK(ump);
  287         /*
  288          * Allocate the extra space in the buffer.
  289          */
  290         error = bread(vp, lbprev, osize, NOCRED, &bp);
  291         if (error) {
  292                 brelse(bp);
  293                 return (error);
  294         }
  295 
  296         if (bp->b_blkno == bp->b_lblkno) {
  297                 if (lbprev >= NDADDR)
  298                         panic("ffs_realloccg: lbprev out of range");
  299                 bp->b_blkno = fsbtodb(fs, bprev);
  300         }
  301 
  302 #ifdef QUOTA
  303         error = chkdq(ip, btodb(nsize - osize), cred, 0);
  304         if (error) {
  305                 brelse(bp);
  306                 return (error);
  307         }
  308 #endif
  309         /*
  310          * Check for extension in the existing location.
  311          */
  312         cg = dtog(fs, bprev);
  313         UFS_LOCK(ump);
  314         bno = ffs_fragextend(ip, cg, bprev, osize, nsize);
  315         if (bno) {
  316                 if (bp->b_blkno != fsbtodb(fs, bno))
  317                         panic("ffs_realloccg: bad blockno");
  318                 delta = btodb(nsize - osize);
  319                 if (ip->i_flag & IN_SPACECOUNTED) {
  320                         UFS_LOCK(ump);
  321                         fs->fs_pendingblocks += delta;
  322                         UFS_UNLOCK(ump);
  323                 }
  324                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  325                 if (flags & IO_EXT)
  326                         ip->i_flag |= IN_CHANGE;
  327                 else
  328                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
  329                 allocbuf(bp, nsize);
  330                 bp->b_flags |= B_DONE;
  331                 bzero(bp->b_data + osize, nsize - osize);
  332                 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO)
  333                         vfs_bio_set_valid(bp, osize, nsize - osize);
  334                 *bpp = bp;
  335                 return (0);
  336         }
  337         /*
  338          * Allocate a new disk location.
  339          */
  340         if (bpref >= fs->fs_size)
  341                 bpref = 0;
  342         switch ((int)fs->fs_optim) {
  343         case FS_OPTSPACE:
  344                 /*
  345                  * Allocate an exact sized fragment. Although this makes
  346                  * best use of space, we will waste time relocating it if
  347                  * the file continues to grow. If the fragmentation is
  348                  * less than half of the minimum free reserve, we choose
  349                  * to begin optimizing for time.
  350                  */
  351                 request = nsize;
  352                 if (fs->fs_minfree <= 5 ||
  353                     fs->fs_cstotal.cs_nffree >
  354                     (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100))
  355                         break;
  356                 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
  357                         fs->fs_fsmnt);
  358                 fs->fs_optim = FS_OPTTIME;
  359                 break;
  360         case FS_OPTTIME:
  361                 /*
  362                  * At this point we have discovered a file that is trying to
  363                  * grow a small fragment to a larger fragment. To save time,
  364                  * we allocate a full sized block, then free the unused portion.
  365                  * If the file continues to grow, the `ffs_fragextend' call
  366                  * above will be able to grow it in place without further
  367                  * copying. If aberrant programs cause disk fragmentation to
  368                  * grow within 2% of the free reserve, we choose to begin
  369                  * optimizing for space.
  370                  */
  371                 request = fs->fs_bsize;
  372                 if (fs->fs_cstotal.cs_nffree <
  373                     (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100)
  374                         break;
  375                 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
  376                         fs->fs_fsmnt);
  377                 fs->fs_optim = FS_OPTSPACE;
  378                 break;
  379         default:
  380                 printf("dev = %s, optim = %ld, fs = %s\n",
  381                     devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt);
  382                 panic("ffs_realloccg: bad optim");
  383                 /* NOTREACHED */
  384         }
  385         bno = ffs_hashalloc(ip, cg, bpref, request, ffs_alloccg);
  386         if (bno > 0) {
  387                 bp->b_blkno = fsbtodb(fs, bno);
  388                 if (!DOINGSOFTDEP(vp))
  389                         ffs_blkfree(ump, fs, ip->i_devvp, bprev, (long)osize,
  390                             ip->i_number);
  391                 if (nsize < request)
  392                         ffs_blkfree(ump, fs, ip->i_devvp,
  393                             bno + numfrags(fs, nsize),
  394                             (long)(request - nsize), ip->i_number);
  395                 delta = btodb(nsize - osize);
  396                 if (ip->i_flag & IN_SPACECOUNTED) {
  397                         UFS_LOCK(ump);
  398                         fs->fs_pendingblocks += delta;
  399                         UFS_UNLOCK(ump);
  400                 }
  401                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  402                 if (flags & IO_EXT)
  403                         ip->i_flag |= IN_CHANGE;
  404                 else
  405                         ip->i_flag |= IN_CHANGE | IN_UPDATE;
  406                 allocbuf(bp, nsize);
  407                 bp->b_flags |= B_DONE;
  408                 bzero(bp->b_data + osize, nsize - osize);
  409                 if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO)
  410                         vfs_bio_set_valid(bp, osize, nsize - osize);
  411                 *bpp = bp;
  412                 return (0);
  413         }
  414 #ifdef QUOTA
  415         UFS_UNLOCK(ump);
  416         /*
  417          * Restore user's disk quota because allocation failed.
  418          */
  419         (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
  420         UFS_LOCK(ump);
  421 #endif
  422 nospace:
  423         /*
  424          * no space available
  425          */
  426         if (fs->fs_pendingblocks > 0 && reclaimed == 0) {
  427                 reclaimed = 1;
  428                 softdep_request_cleanup(fs, vp);
  429                 UFS_UNLOCK(ump);
  430                 if (bp) {
  431                         brelse(bp);
  432                         bp = NULL;
  433                 }
  434                 UFS_LOCK(ump);
  435                 goto retry;
  436         }
  437         UFS_UNLOCK(ump);
  438         if (bp)
  439                 brelse(bp);
  440         if (ppsratecheck(&lastfail, &curfail, 1)) {
  441                 ffs_fserr(fs, ip->i_number, "filesystem full");
  442                 uprintf("\n%s: write failed, filesystem is full\n",
  443                     fs->fs_fsmnt);
  444         }
  445         return (ENOSPC);
  446 }
  447 
  448 /*
  449  * Reallocate a sequence of blocks into a contiguous sequence of blocks.
  450  *
  451  * The vnode and an array of buffer pointers for a range of sequential
  452  * logical blocks to be made contiguous is given. The allocator attempts
  453  * to find a range of sequential blocks starting as close as possible
  454  * from the end of the allocation for the logical block immediately
  455  * preceding the current range. If successful, the physical block numbers
  456  * in the buffer pointers and in the inode are changed to reflect the new
  457  * allocation. If unsuccessful, the allocation is left unchanged. The
  458  * success in doing the reallocation is returned. Note that the error
  459  * return is not reflected back to the user. Rather the previous block
  460  * allocation will be used.
  461  */
  462 
  463 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
  464 
  465 static int doasyncfree = 1;
  466 SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
  467 
  468 static int doreallocblks = 1;
  469 SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
  470 
  471 #ifdef DEBUG
  472 static volatile int prtrealloc = 0;
  473 #endif
  474 
  475 int
  476 ffs_reallocblks(ap)
  477         struct vop_reallocblks_args /* {
  478                 struct vnode *a_vp;
  479                 struct cluster_save *a_buflist;
  480         } */ *ap;
  481 {
  482 
  483         if (doreallocblks == 0)
  484                 return (ENOSPC);
  485         if (VTOI(ap->a_vp)->i_ump->um_fstype == UFS1)
  486                 return (ffs_reallocblks_ufs1(ap));
  487         return (ffs_reallocblks_ufs2(ap));
  488 }
  489         
  490 static int
  491 ffs_reallocblks_ufs1(ap)
  492         struct vop_reallocblks_args /* {
  493                 struct vnode *a_vp;
  494                 struct cluster_save *a_buflist;
  495         } */ *ap;
  496 {
  497         struct fs *fs;
  498         struct inode *ip;
  499         struct vnode *vp;
  500         struct buf *sbp, *ebp;
  501         ufs1_daddr_t *bap, *sbap, *ebap = 0;
  502         struct cluster_save *buflist;
  503         struct ufsmount *ump;
  504         ufs_lbn_t start_lbn, end_lbn;
  505         ufs1_daddr_t soff, newblk, blkno;
  506         ufs2_daddr_t pref;
  507         struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
  508         int i, len, start_lvl, end_lvl, ssize;
  509 
  510         vp = ap->a_vp;
  511         ip = VTOI(vp);
  512         fs = ip->i_fs;
  513         ump = ip->i_ump;
  514         if (fs->fs_contigsumsize <= 0)
  515                 return (ENOSPC);
  516         buflist = ap->a_buflist;
  517         len = buflist->bs_nchildren;
  518         start_lbn = buflist->bs_children[0]->b_lblkno;
  519         end_lbn = start_lbn + len - 1;
  520 #ifdef INVARIANTS
  521         for (i = 0; i < len; i++)
  522                 if (!ffs_checkblk(ip,
  523                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  524                         panic("ffs_reallocblks: unallocated block 1");
  525         for (i = 1; i < len; i++)
  526                 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
  527                         panic("ffs_reallocblks: non-logical cluster");
  528         blkno = buflist->bs_children[0]->b_blkno;
  529         ssize = fsbtodb(fs, fs->fs_frag);
  530         for (i = 1; i < len - 1; i++)
  531                 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
  532                         panic("ffs_reallocblks: non-physical cluster %d", i);
  533 #endif
  534         /*
  535          * If the latest allocation is in a new cylinder group, assume that
  536          * the filesystem has decided to move and do not force it back to
  537          * the previous cylinder group.
  538          */
  539         if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
  540             dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
  541                 return (ENOSPC);
  542         if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
  543             ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
  544                 return (ENOSPC);
  545         /*
  546          * Get the starting offset and block map for the first block.
  547          */
  548         if (start_lvl == 0) {
  549                 sbap = &ip->i_din1->di_db[0];
  550                 soff = start_lbn;
  551         } else {
  552                 idp = &start_ap[start_lvl - 1];
  553                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
  554                         brelse(sbp);
  555                         return (ENOSPC);
  556                 }
  557                 sbap = (ufs1_daddr_t *)sbp->b_data;
  558                 soff = idp->in_off;
  559         }
  560         /*
  561          * If the block range spans two block maps, get the second map.
  562          */
  563         if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
  564                 ssize = len;
  565         } else {
  566 #ifdef INVARIANTS
  567                 if (start_lvl > 0 &&
  568                     start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
  569                         panic("ffs_reallocblk: start == end");
  570 #endif
  571                 ssize = len - (idp->in_off + 1);
  572                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
  573                         goto fail;
  574                 ebap = (ufs1_daddr_t *)ebp->b_data;
  575         }
  576         /*
  577          * Find the preferred location for the cluster.
  578          */
  579         UFS_LOCK(ump);
  580         pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap);
  581         /*
  582          * Search the block map looking for an allocation of the desired size.
  583          */
  584         if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
  585             len, ffs_clusteralloc)) == 0) {
  586                 UFS_UNLOCK(ump);
  587                 goto fail;
  588         }
  589         /*
  590          * We have found a new contiguous block.
  591          *
  592          * First we have to replace the old block pointers with the new
  593          * block pointers in the inode and indirect blocks associated
  594          * with the file.
  595          */
  596 #ifdef DEBUG
  597         if (prtrealloc)
  598                 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
  599                     (intmax_t)start_lbn, (intmax_t)end_lbn);
  600 #endif
  601         blkno = newblk;
  602         for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
  603                 if (i == ssize) {
  604                         bap = ebap;
  605                         soff = -i;
  606                 }
  607 #ifdef INVARIANTS
  608                 if (!ffs_checkblk(ip,
  609                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  610                         panic("ffs_reallocblks: unallocated block 2");
  611                 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
  612                         panic("ffs_reallocblks: alloc mismatch");
  613 #endif
  614 #ifdef DEBUG
  615                 if (prtrealloc)
  616                         printf(" %d,", *bap);
  617 #endif
  618                 if (DOINGSOFTDEP(vp)) {
  619                         if (sbap == &ip->i_din1->di_db[0] && i < ssize)
  620                                 softdep_setup_allocdirect(ip, start_lbn + i,
  621                                     blkno, *bap, fs->fs_bsize, fs->fs_bsize,
  622                                     buflist->bs_children[i]);
  623                         else
  624                                 softdep_setup_allocindir_page(ip, start_lbn + i,
  625                                     i < ssize ? sbp : ebp, soff + i, blkno,
  626                                     *bap, buflist->bs_children[i]);
  627                 }
  628                 *bap++ = blkno;
  629         }
  630         /*
  631          * Next we must write out the modified inode and indirect blocks.
  632          * For strict correctness, the writes should be synchronous since
  633          * the old block values may have been written to disk. In practise
  634          * they are almost never written, but if we are concerned about
  635          * strict correctness, the `doasyncfree' flag should be set to zero.
  636          *
  637          * The test on `doasyncfree' should be changed to test a flag
  638          * that shows whether the associated buffers and inodes have
  639          * been written. The flag should be set when the cluster is
  640          * started and cleared whenever the buffer or inode is flushed.
  641          * We can then check below to see if it is set, and do the
  642          * synchronous write only when it has been cleared.
  643          */
  644         if (sbap != &ip->i_din1->di_db[0]) {
  645                 if (doasyncfree)
  646                         bdwrite(sbp);
  647                 else
  648                         bwrite(sbp);
  649         } else {
  650                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  651                 if (!doasyncfree)
  652                         ffs_update(vp, 1);
  653         }
  654         if (ssize < len) {
  655                 if (doasyncfree)
  656                         bdwrite(ebp);
  657                 else
  658                         bwrite(ebp);
  659         }
  660         /*
  661          * Last, free the old blocks and assign the new blocks to the buffers.
  662          */
  663 #ifdef DEBUG
  664         if (prtrealloc)
  665                 printf("\n\tnew:");
  666 #endif
  667         for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
  668                 if (!DOINGSOFTDEP(vp))
  669                         ffs_blkfree(ump, fs, ip->i_devvp,
  670                             dbtofsb(fs, buflist->bs_children[i]->b_blkno),
  671                             fs->fs_bsize, ip->i_number);
  672                 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
  673 #ifdef INVARIANTS
  674                 if (!ffs_checkblk(ip,
  675                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  676                         panic("ffs_reallocblks: unallocated block 3");
  677 #endif
  678 #ifdef DEBUG
  679                 if (prtrealloc)
  680                         printf(" %d,", blkno);
  681 #endif
  682         }
  683 #ifdef DEBUG
  684         if (prtrealloc) {
  685                 prtrealloc--;
  686                 printf("\n");
  687         }
  688 #endif
  689         return (0);
  690 
  691 fail:
  692         if (ssize < len)
  693                 brelse(ebp);
  694         if (sbap != &ip->i_din1->di_db[0])
  695                 brelse(sbp);
  696         return (ENOSPC);
  697 }
  698 
  699 static int
  700 ffs_reallocblks_ufs2(ap)
  701         struct vop_reallocblks_args /* {
  702                 struct vnode *a_vp;
  703                 struct cluster_save *a_buflist;
  704         } */ *ap;
  705 {
  706         struct fs *fs;
  707         struct inode *ip;
  708         struct vnode *vp;
  709         struct buf *sbp, *ebp;
  710         ufs2_daddr_t *bap, *sbap, *ebap = 0;
  711         struct cluster_save *buflist;
  712         struct ufsmount *ump;
  713         ufs_lbn_t start_lbn, end_lbn;
  714         ufs2_daddr_t soff, newblk, blkno, pref;
  715         struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
  716         int i, len, start_lvl, end_lvl, ssize;
  717 
  718         vp = ap->a_vp;
  719         ip = VTOI(vp);
  720         fs = ip->i_fs;
  721         ump = ip->i_ump;
  722         if (fs->fs_contigsumsize <= 0)
  723                 return (ENOSPC);
  724         buflist = ap->a_buflist;
  725         len = buflist->bs_nchildren;
  726         start_lbn = buflist->bs_children[0]->b_lblkno;
  727         end_lbn = start_lbn + len - 1;
  728 #ifdef INVARIANTS
  729         for (i = 0; i < len; i++)
  730                 if (!ffs_checkblk(ip,
  731                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  732                         panic("ffs_reallocblks: unallocated block 1");
  733         for (i = 1; i < len; i++)
  734                 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
  735                         panic("ffs_reallocblks: non-logical cluster");
  736         blkno = buflist->bs_children[0]->b_blkno;
  737         ssize = fsbtodb(fs, fs->fs_frag);
  738         for (i = 1; i < len - 1; i++)
  739                 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
  740                         panic("ffs_reallocblks: non-physical cluster %d", i);
  741 #endif
  742         /*
  743          * If the latest allocation is in a new cylinder group, assume that
  744          * the filesystem has decided to move and do not force it back to
  745          * the previous cylinder group.
  746          */
  747         if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
  748             dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
  749                 return (ENOSPC);
  750         if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
  751             ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
  752                 return (ENOSPC);
  753         /*
  754          * Get the starting offset and block map for the first block.
  755          */
  756         if (start_lvl == 0) {
  757                 sbap = &ip->i_din2->di_db[0];
  758                 soff = start_lbn;
  759         } else {
  760                 idp = &start_ap[start_lvl - 1];
  761                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
  762                         brelse(sbp);
  763                         return (ENOSPC);
  764                 }
  765                 sbap = (ufs2_daddr_t *)sbp->b_data;
  766                 soff = idp->in_off;
  767         }
  768         /*
  769          * If the block range spans two block maps, get the second map.
  770          */
  771         if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
  772                 ssize = len;
  773         } else {
  774 #ifdef INVARIANTS
  775                 if (start_lvl > 0 &&
  776                     start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
  777                         panic("ffs_reallocblk: start == end");
  778 #endif
  779                 ssize = len - (idp->in_off + 1);
  780                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
  781                         goto fail;
  782                 ebap = (ufs2_daddr_t *)ebp->b_data;
  783         }
  784         /*
  785          * Find the preferred location for the cluster.
  786          */
  787         UFS_LOCK(ump);
  788         pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap);
  789         /*
  790          * Search the block map looking for an allocation of the desired size.
  791          */
  792         if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
  793             len, ffs_clusteralloc)) == 0) {
  794                 UFS_UNLOCK(ump);
  795                 goto fail;
  796         }
  797         /*
  798          * We have found a new contiguous block.
  799          *
  800          * First we have to replace the old block pointers with the new
  801          * block pointers in the inode and indirect blocks associated
  802          * with the file.
  803          */
  804 #ifdef DEBUG
  805         if (prtrealloc)
  806                 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
  807                     (intmax_t)start_lbn, (intmax_t)end_lbn);
  808 #endif
  809         blkno = newblk;
  810         for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
  811                 if (i == ssize) {
  812                         bap = ebap;
  813                         soff = -i;
  814                 }
  815 #ifdef INVARIANTS
  816                 if (!ffs_checkblk(ip,
  817                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  818                         panic("ffs_reallocblks: unallocated block 2");
  819                 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
  820                         panic("ffs_reallocblks: alloc mismatch");
  821 #endif
  822 #ifdef DEBUG
  823                 if (prtrealloc)
  824                         printf(" %jd,", (intmax_t)*bap);
  825 #endif
  826                 if (DOINGSOFTDEP(vp)) {
  827                         if (sbap == &ip->i_din2->di_db[0] && i < ssize)
  828                                 softdep_setup_allocdirect(ip, start_lbn + i,
  829                                     blkno, *bap, fs->fs_bsize, fs->fs_bsize,
  830                                     buflist->bs_children[i]);
  831                         else
  832                                 softdep_setup_allocindir_page(ip, start_lbn + i,
  833                                     i < ssize ? sbp : ebp, soff + i, blkno,
  834                                     *bap, buflist->bs_children[i]);
  835                 }
  836                 *bap++ = blkno;
  837         }
  838         /*
  839          * Next we must write out the modified inode and indirect blocks.
  840          * For strict correctness, the writes should be synchronous since
  841          * the old block values may have been written to disk. In practise
  842          * they are almost never written, but if we are concerned about
  843          * strict correctness, the `doasyncfree' flag should be set to zero.
  844          *
  845          * The test on `doasyncfree' should be changed to test a flag
  846          * that shows whether the associated buffers and inodes have
  847          * been written. The flag should be set when the cluster is
  848          * started and cleared whenever the buffer or inode is flushed.
  849          * We can then check below to see if it is set, and do the
  850          * synchronous write only when it has been cleared.
  851          */
  852         if (sbap != &ip->i_din2->di_db[0]) {
  853                 if (doasyncfree)
  854                         bdwrite(sbp);
  855                 else
  856                         bwrite(sbp);
  857         } else {
  858                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  859                 if (!doasyncfree)
  860                         ffs_update(vp, 1);
  861         }
  862         if (ssize < len) {
  863                 if (doasyncfree)
  864                         bdwrite(ebp);
  865                 else
  866                         bwrite(ebp);
  867         }
  868         /*
  869          * Last, free the old blocks and assign the new blocks to the buffers.
  870          */
  871 #ifdef DEBUG
  872         if (prtrealloc)
  873                 printf("\n\tnew:");
  874 #endif
  875         for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
  876                 if (!DOINGSOFTDEP(vp))
  877                         ffs_blkfree(ump, fs, ip->i_devvp,
  878                             dbtofsb(fs, buflist->bs_children[i]->b_blkno),
  879                             fs->fs_bsize, ip->i_number);
  880                 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
  881 #ifdef INVARIANTS
  882                 if (!ffs_checkblk(ip,
  883                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  884                         panic("ffs_reallocblks: unallocated block 3");
  885 #endif
  886 #ifdef DEBUG
  887                 if (prtrealloc)
  888                         printf(" %jd,", (intmax_t)blkno);
  889 #endif
  890         }
  891 #ifdef DEBUG
  892         if (prtrealloc) {
  893                 prtrealloc--;
  894                 printf("\n");
  895         }
  896 #endif
  897         return (0);
  898 
  899 fail:
  900         if (ssize < len)
  901                 brelse(ebp);
  902         if (sbap != &ip->i_din2->di_db[0])
  903                 brelse(sbp);
  904         return (ENOSPC);
  905 }
  906 
  907 /*
  908  * Allocate an inode in the filesystem.
  909  *
  910  * If allocating a directory, use ffs_dirpref to select the inode.
  911  * If allocating in a directory, the following hierarchy is followed:
  912  *   1) allocate the preferred inode.
  913  *   2) allocate an inode in the same cylinder group.
  914  *   3) quadradically rehash into other cylinder groups, until an
  915  *      available inode is located.
  916  * If no inode preference is given the following hierarchy is used
  917  * to allocate an inode:
  918  *   1) allocate an inode in cylinder group 0.
  919  *   2) quadradically rehash into other cylinder groups, until an
  920  *      available inode is located.
  921  */
  922 int
  923 ffs_valloc(pvp, mode, cred, vpp)
  924         struct vnode *pvp;
  925         int mode;
  926         struct ucred *cred;
  927         struct vnode **vpp;
  928 {
  929         struct inode *pip;
  930         struct fs *fs;
  931         struct inode *ip;
  932         struct timespec ts;
  933         struct ufsmount *ump;
  934         ino_t ino, ipref;
  935         u_int cg;
  936         int error, error1;
  937         static struct timeval lastfail;
  938         static int curfail;
  939 
  940         *vpp = NULL;
  941         pip = VTOI(pvp);
  942         fs = pip->i_fs;
  943         ump = pip->i_ump;
  944 
  945         UFS_LOCK(ump);
  946         if (fs->fs_cstotal.cs_nifree == 0)
  947                 goto noinodes;
  948 
  949         if ((mode & IFMT) == IFDIR)
  950                 ipref = ffs_dirpref(pip);
  951         else
  952                 ipref = pip->i_number;
  953         if (ipref >= fs->fs_ncg * fs->fs_ipg)
  954                 ipref = 0;
  955         cg = ino_to_cg(fs, ipref);
  956         /*
  957          * Track number of dirs created one after another
  958          * in a same cg without intervening by files.
  959          */
  960         if ((mode & IFMT) == IFDIR) {
  961                 if (fs->fs_contigdirs[cg] < 255)
  962                         fs->fs_contigdirs[cg]++;
  963         } else {
  964                 if (fs->fs_contigdirs[cg] > 0)
  965                         fs->fs_contigdirs[cg]--;
  966         }
  967         ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode,
  968                                         (allocfcn_t *)ffs_nodealloccg);
  969         if (ino == 0)
  970                 goto noinodes;
  971         error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
  972         if (error) {
  973                 error1 = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp,
  974                     FFSV_FORCEINSMQ);
  975                 ffs_vfree(pvp, ino, mode);
  976                 if (error1 == 0) {
  977                         ip = VTOI(*vpp);
  978                         if (ip->i_mode)
  979                                 goto dup_alloc;
  980                         ip->i_flag |= IN_MODIFIED;
  981                         vput(*vpp);
  982                 }
  983                 return (error);
  984         }
  985         ip = VTOI(*vpp);
  986         if (ip->i_mode) {
  987 dup_alloc:
  988                 printf("mode = 0%o, inum = %lu, fs = %s\n",
  989                     ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
  990                 panic("ffs_valloc: dup alloc");
  991         }
  992         if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) {  /* XXX */
  993                 printf("free inode %s/%lu had %ld blocks\n",
  994                     fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks));
  995                 DIP_SET(ip, i_blocks, 0);
  996         }
  997         ip->i_flags = 0;
  998         DIP_SET(ip, i_flags, 0);
  999         /*
 1000          * Set up a new generation number for this inode.
 1001          */
 1002         if (ip->i_gen == 0 || ++ip->i_gen == 0)
 1003                 ip->i_gen = arc4random() / 2 + 1;
 1004         DIP_SET(ip, i_gen, ip->i_gen);
 1005         if (fs->fs_magic == FS_UFS2_MAGIC) {
 1006                 vfs_timestamp(&ts);
 1007                 ip->i_din2->di_birthtime = ts.tv_sec;
 1008                 ip->i_din2->di_birthnsec = ts.tv_nsec;
 1009         }
 1010         ip->i_flag = 0;
 1011         vnode_destroy_vobject(*vpp);
 1012         (*vpp)->v_type = VNON;
 1013         if (fs->fs_magic == FS_UFS2_MAGIC)
 1014                 (*vpp)->v_op = &ffs_vnodeops2;
 1015         else
 1016                 (*vpp)->v_op = &ffs_vnodeops1;
 1017         return (0);
 1018 noinodes:
 1019         UFS_UNLOCK(ump);
 1020         if (ppsratecheck(&lastfail, &curfail, 1)) {
 1021                 ffs_fserr(fs, pip->i_number, "out of inodes");
 1022                 uprintf("\n%s: create/symlink failed, no inodes free\n",
 1023                     fs->fs_fsmnt);
 1024         }
 1025         return (ENOSPC);
 1026 }
 1027 
 1028 /*
 1029  * Find a cylinder group to place a directory.
 1030  *
 1031  * The policy implemented by this algorithm is to allocate a
 1032  * directory inode in the same cylinder group as its parent
 1033  * directory, but also to reserve space for its files inodes
 1034  * and data. Restrict the number of directories which may be
 1035  * allocated one after another in the same cylinder group
 1036  * without intervening allocation of files.
 1037  *
 1038  * If we allocate a first level directory then force allocation
 1039  * in another cylinder group.
 1040  */
 1041 static ino_t
 1042 ffs_dirpref(pip)
 1043         struct inode *pip;
 1044 {
 1045         struct fs *fs;
 1046         u_int cg, prefcg, dirsize, cgsize;
 1047         u_int avgifree, avgbfree, avgndir, curdirsize;
 1048         u_int minifree, minbfree, maxndir;
 1049         u_int mincg, minndir;
 1050         u_int maxcontigdirs;
 1051 
 1052         mtx_assert(UFS_MTX(pip->i_ump), MA_OWNED);
 1053         fs = pip->i_fs;
 1054 
 1055         avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
 1056         avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1057         avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
 1058 
 1059         /*
 1060          * Force allocation in another cg if creating a first level dir.
 1061          */
 1062         ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref");
 1063         if (ITOV(pip)->v_vflag & VV_ROOT) {
 1064                 prefcg = arc4random() % fs->fs_ncg;
 1065                 mincg = prefcg;
 1066                 minndir = fs->fs_ipg;
 1067                 for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1068                         if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
 1069                             fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
 1070                             fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1071                                 mincg = cg;
 1072                                 minndir = fs->fs_cs(fs, cg).cs_ndir;
 1073                         }
 1074                 for (cg = 0; cg < prefcg; cg++)
 1075                         if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
 1076                             fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
 1077                             fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1078                                 mincg = cg;
 1079                                 minndir = fs->fs_cs(fs, cg).cs_ndir;
 1080                         }
 1081                 return ((ino_t)(fs->fs_ipg * mincg));
 1082         }
 1083 
 1084         /*
 1085          * Count various limits which used for
 1086          * optimal allocation of a directory inode.
 1087          */
 1088         maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
 1089         minifree = avgifree - avgifree / 4;
 1090         if (minifree < 1)
 1091                 minifree = 1;
 1092         minbfree = avgbfree - avgbfree / 4;
 1093         if (minbfree < 1)
 1094                 minbfree = 1;
 1095         cgsize = fs->fs_fsize * fs->fs_fpg;
 1096         dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir;
 1097         curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
 1098         if (dirsize < curdirsize)
 1099                 dirsize = curdirsize;
 1100         if (dirsize <= 0)
 1101                 maxcontigdirs = 0;              /* dirsize overflowed */
 1102         else
 1103                 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255);
 1104         if (fs->fs_avgfpdir > 0)
 1105                 maxcontigdirs = min(maxcontigdirs,
 1106                                     fs->fs_ipg / fs->fs_avgfpdir);
 1107         if (maxcontigdirs == 0)
 1108                 maxcontigdirs = 1;
 1109 
 1110         /*
 1111          * Limit number of dirs in one cg and reserve space for 
 1112          * regular files, but only if we have no deficit in
 1113          * inodes or space.
 1114          */
 1115         prefcg = ino_to_cg(fs, pip->i_number);
 1116         for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1117                 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
 1118                     fs->fs_cs(fs, cg).cs_nifree >= minifree &&
 1119                     fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
 1120                         if (fs->fs_contigdirs[cg] < maxcontigdirs)
 1121                                 return ((ino_t)(fs->fs_ipg * cg));
 1122                 }
 1123         for (cg = 0; cg < prefcg; cg++)
 1124                 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
 1125                     fs->fs_cs(fs, cg).cs_nifree >= minifree &&
 1126                     fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
 1127                         if (fs->fs_contigdirs[cg] < maxcontigdirs)
 1128                                 return ((ino_t)(fs->fs_ipg * cg));
 1129                 }
 1130         /*
 1131          * This is a backstop when we have deficit in space.
 1132          */
 1133         for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1134                 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
 1135                         return ((ino_t)(fs->fs_ipg * cg));
 1136         for (cg = 0; cg < prefcg; cg++)
 1137                 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
 1138                         break;
 1139         return ((ino_t)(fs->fs_ipg * cg));
 1140 }
 1141 
 1142 /*
 1143  * Select the desired position for the next block in a file.  The file is
 1144  * logically divided into sections. The first section is composed of the
 1145  * direct blocks. Each additional section contains fs_maxbpg blocks.
 1146  *
 1147  * If no blocks have been allocated in the first section, the policy is to
 1148  * request a block in the same cylinder group as the inode that describes
 1149  * the file. If no blocks have been allocated in any other section, the
 1150  * policy is to place the section in a cylinder group with a greater than
 1151  * average number of free blocks.  An appropriate cylinder group is found
 1152  * by using a rotor that sweeps the cylinder groups. When a new group of
 1153  * blocks is needed, the sweep begins in the cylinder group following the
 1154  * cylinder group from which the previous allocation was made. The sweep
 1155  * continues until a cylinder group with greater than the average number
 1156  * of free blocks is found. If the allocation is for the first block in an
 1157  * indirect block, the information on the previous allocation is unavailable;
 1158  * here a best guess is made based upon the logical block number being
 1159  * allocated.
 1160  *
 1161  * If a section is already partially allocated, the policy is to
 1162  * contiguously allocate fs_maxcontig blocks. The end of one of these
 1163  * contiguous blocks and the beginning of the next is laid out
 1164  * contiguously if possible.
 1165  */
 1166 ufs2_daddr_t
 1167 ffs_blkpref_ufs1(ip, lbn, indx, bap)
 1168         struct inode *ip;
 1169         ufs_lbn_t lbn;
 1170         int indx;
 1171         ufs1_daddr_t *bap;
 1172 {
 1173         struct fs *fs;
 1174         u_int cg;
 1175         u_int avgbfree, startcg;
 1176 
 1177         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1178         fs = ip->i_fs;
 1179         if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
 1180                 if (lbn < NDADDR + NINDIR(fs)) {
 1181                         cg = ino_to_cg(fs, ip->i_number);
 1182                         return (cgbase(fs, cg) + fs->fs_frag);
 1183                 }
 1184                 /*
 1185                  * Find a cylinder with greater than average number of
 1186                  * unused data blocks.
 1187                  */
 1188                 if (indx == 0 || bap[indx - 1] == 0)
 1189                         startcg =
 1190                             ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
 1191                 else
 1192                         startcg = dtog(fs, bap[indx - 1]) + 1;
 1193                 startcg %= fs->fs_ncg;
 1194                 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1195                 for (cg = startcg; cg < fs->fs_ncg; 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                 for (cg = 0; cg <= startcg; cg++)
 1201                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1202                                 fs->fs_cgrotor = cg;
 1203                                 return (cgbase(fs, cg) + fs->fs_frag);
 1204                         }
 1205                 return (0);
 1206         }
 1207         /*
 1208          * We just always try to lay things out contiguously.
 1209          */
 1210         return (bap[indx - 1] + fs->fs_frag);
 1211 }
 1212 
 1213 /*
 1214  * Same as above, but for UFS2
 1215  */
 1216 ufs2_daddr_t
 1217 ffs_blkpref_ufs2(ip, lbn, indx, bap)
 1218         struct inode *ip;
 1219         ufs_lbn_t lbn;
 1220         int indx;
 1221         ufs2_daddr_t *bap;
 1222 {
 1223         struct fs *fs;
 1224         u_int cg;
 1225         u_int avgbfree, startcg;
 1226 
 1227         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1228         fs = ip->i_fs;
 1229         if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
 1230                 if (lbn < NDADDR + NINDIR(fs)) {
 1231                         cg = ino_to_cg(fs, ip->i_number);
 1232                         return (cgbase(fs, cg) + fs->fs_frag);
 1233                 }
 1234                 /*
 1235                  * Find a cylinder with greater than average number of
 1236                  * unused data blocks.
 1237                  */
 1238                 if (indx == 0 || bap[indx - 1] == 0)
 1239                         startcg =
 1240                             ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
 1241                 else
 1242                         startcg = dtog(fs, bap[indx - 1]) + 1;
 1243                 startcg %= fs->fs_ncg;
 1244                 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1245                 for (cg = startcg; cg < fs->fs_ncg; 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                 for (cg = 0; cg <= startcg; cg++)
 1251                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1252                                 fs->fs_cgrotor = cg;
 1253                                 return (cgbase(fs, cg) + fs->fs_frag);
 1254                         }
 1255                 return (0);
 1256         }
 1257         /*
 1258          * We just always try to lay things out contiguously.
 1259          */
 1260         return (bap[indx - 1] + fs->fs_frag);
 1261 }
 1262 
 1263 /*
 1264  * Implement the cylinder overflow algorithm.
 1265  *
 1266  * The policy implemented by this algorithm is:
 1267  *   1) allocate the block in its requested cylinder group.
 1268  *   2) quadradically rehash on the cylinder group number.
 1269  *   3) brute force search for a free block.
 1270  *
 1271  * Must be called with the UFS lock held.  Will release the lock on success
 1272  * and return with it held on failure.
 1273  */
 1274 /*VARARGS5*/
 1275 static ufs2_daddr_t
 1276 ffs_hashalloc(ip, cg, pref, size, allocator)
 1277         struct inode *ip;
 1278         u_int cg;
 1279         ufs2_daddr_t pref;
 1280         int size;       /* size for data blocks, mode for inodes */
 1281         allocfcn_t *allocator;
 1282 {
 1283         struct fs *fs;
 1284         ufs2_daddr_t result;
 1285         u_int i, icg = cg;
 1286 
 1287         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1288 #ifdef INVARIANTS
 1289         if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
 1290                 panic("ffs_hashalloc: allocation on suspended filesystem");
 1291 #endif
 1292         fs = ip->i_fs;
 1293         /*
 1294          * 1: preferred cylinder group
 1295          */
 1296         result = (*allocator)(ip, cg, pref, size);
 1297         if (result)
 1298                 return (result);
 1299         /*
 1300          * 2: quadratic rehash
 1301          */
 1302         for (i = 1; i < fs->fs_ncg; i *= 2) {
 1303                 cg += i;
 1304                 if (cg >= fs->fs_ncg)
 1305                         cg -= fs->fs_ncg;
 1306                 result = (*allocator)(ip, cg, 0, size);
 1307                 if (result)
 1308                         return (result);
 1309         }
 1310         /*
 1311          * 3: brute force search
 1312          * Note that we start at i == 2, since 0 was checked initially,
 1313          * and 1 is always checked in the quadratic rehash.
 1314          */
 1315         cg = (icg + 2) % fs->fs_ncg;
 1316         for (i = 2; i < fs->fs_ncg; i++) {
 1317                 result = (*allocator)(ip, cg, 0, size);
 1318                 if (result)
 1319                         return (result);
 1320                 cg++;
 1321                 if (cg == fs->fs_ncg)
 1322                         cg = 0;
 1323         }
 1324         return (0);
 1325 }
 1326 
 1327 /*
 1328  * Determine whether a fragment can be extended.
 1329  *
 1330  * Check to see if the necessary fragments are available, and
 1331  * if they are, allocate them.
 1332  */
 1333 static ufs2_daddr_t
 1334 ffs_fragextend(ip, cg, bprev, osize, nsize)
 1335         struct inode *ip;
 1336         u_int cg;
 1337         ufs2_daddr_t bprev;
 1338         int osize, nsize;
 1339 {
 1340         struct fs *fs;
 1341         struct cg *cgp;
 1342         struct buf *bp;
 1343         struct ufsmount *ump;
 1344         int nffree;
 1345         long bno;
 1346         int frags, bbase;
 1347         int i, error;
 1348         u_int8_t *blksfree;
 1349 
 1350         ump = ip->i_ump;
 1351         fs = ip->i_fs;
 1352         if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
 1353                 return (0);
 1354         frags = numfrags(fs, nsize);
 1355         bbase = fragnum(fs, bprev);
 1356         if (bbase > fragnum(fs, (bprev + frags - 1))) {
 1357                 /* cannot extend across a block boundary */
 1358                 return (0);
 1359         }
 1360         UFS_UNLOCK(ump);
 1361         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1362                 (int)fs->fs_cgsize, NOCRED, &bp);
 1363         if (error)
 1364                 goto fail;
 1365         cgp = (struct cg *)bp->b_data;
 1366         if (!cg_chkmagic(cgp))
 1367                 goto fail;
 1368         bp->b_xflags |= BX_BKGRDWRITE;
 1369         cgp->cg_old_time = cgp->cg_time = time_second;
 1370         bno = dtogd(fs, bprev);
 1371         blksfree = cg_blksfree(cgp);
 1372         for (i = numfrags(fs, osize); i < frags; i++)
 1373                 if (isclr(blksfree, bno + i))
 1374                         goto fail;
 1375         /*
 1376          * the current fragment can be extended
 1377          * deduct the count on fragment being extended into
 1378          * increase the count on the remaining fragment (if any)
 1379          * allocate the extended piece
 1380          */
 1381         for (i = frags; i < fs->fs_frag - bbase; i++)
 1382                 if (isclr(blksfree, bno + i))
 1383                         break;
 1384         cgp->cg_frsum[i - numfrags(fs, osize)]--;
 1385         if (i != frags)
 1386                 cgp->cg_frsum[i - frags]++;
 1387         for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) {
 1388                 clrbit(blksfree, bno + i);
 1389                 cgp->cg_cs.cs_nffree--;
 1390                 nffree++;
 1391         }
 1392         UFS_LOCK(ump);
 1393         fs->fs_cstotal.cs_nffree -= nffree;
 1394         fs->fs_cs(fs, cg).cs_nffree -= nffree;
 1395         fs->fs_fmod = 1;
 1396         ACTIVECLEAR(fs, cg);
 1397         UFS_UNLOCK(ump);
 1398         if (DOINGSOFTDEP(ITOV(ip)))
 1399                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev);
 1400         bdwrite(bp);
 1401         return (bprev);
 1402 
 1403 fail:
 1404         brelse(bp);
 1405         UFS_LOCK(ump);
 1406         return (0);
 1407 
 1408 }
 1409 
 1410 /*
 1411  * Determine whether a block can be allocated.
 1412  *
 1413  * Check to see if a block of the appropriate size is available,
 1414  * and if it is, allocate it.
 1415  */
 1416 static ufs2_daddr_t
 1417 ffs_alloccg(ip, cg, bpref, size)
 1418         struct inode *ip;
 1419         u_int cg;
 1420         ufs2_daddr_t bpref;
 1421         int size;
 1422 {
 1423         struct fs *fs;
 1424         struct cg *cgp;
 1425         struct buf *bp;
 1426         struct ufsmount *ump;
 1427         ufs1_daddr_t bno;
 1428         ufs2_daddr_t blkno;
 1429         int i, allocsiz, error, frags;
 1430         u_int8_t *blksfree;
 1431 
 1432         ump = ip->i_ump;
 1433         fs = ip->i_fs;
 1434         if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
 1435                 return (0);
 1436         UFS_UNLOCK(ump);
 1437         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1438                 (int)fs->fs_cgsize, NOCRED, &bp);
 1439         if (error)
 1440                 goto fail;
 1441         cgp = (struct cg *)bp->b_data;
 1442         if (!cg_chkmagic(cgp) ||
 1443             (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize))
 1444                 goto fail;
 1445         bp->b_xflags |= BX_BKGRDWRITE;
 1446         cgp->cg_old_time = cgp->cg_time = time_second;
 1447         if (size == fs->fs_bsize) {
 1448                 UFS_LOCK(ump);
 1449                 blkno = ffs_alloccgblk(ip, bp, bpref);
 1450                 ACTIVECLEAR(fs, cg);
 1451                 UFS_UNLOCK(ump);
 1452                 bdwrite(bp);
 1453                 return (blkno);
 1454         }
 1455         /*
 1456          * check to see if any fragments are already available
 1457          * allocsiz is the size which will be allocated, hacking
 1458          * it down to a smaller size if necessary
 1459          */
 1460         blksfree = cg_blksfree(cgp);
 1461         frags = numfrags(fs, size);
 1462         for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
 1463                 if (cgp->cg_frsum[allocsiz] != 0)
 1464                         break;
 1465         if (allocsiz == fs->fs_frag) {
 1466                 /*
 1467                  * no fragments were available, so a block will be
 1468                  * allocated, and hacked up
 1469                  */
 1470                 if (cgp->cg_cs.cs_nbfree == 0)
 1471                         goto fail;
 1472                 UFS_LOCK(ump);
 1473                 blkno = ffs_alloccgblk(ip, bp, bpref);
 1474                 bno = dtogd(fs, blkno);
 1475                 for (i = frags; i < fs->fs_frag; i++)
 1476                         setbit(blksfree, bno + i);
 1477                 i = fs->fs_frag - frags;
 1478                 cgp->cg_cs.cs_nffree += i;
 1479                 fs->fs_cstotal.cs_nffree += i;
 1480                 fs->fs_cs(fs, cg).cs_nffree += i;
 1481                 fs->fs_fmod = 1;
 1482                 cgp->cg_frsum[i]++;
 1483                 ACTIVECLEAR(fs, cg);
 1484                 UFS_UNLOCK(ump);
 1485                 bdwrite(bp);
 1486                 return (blkno);
 1487         }
 1488         bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
 1489         if (bno < 0)
 1490                 goto fail;
 1491         for (i = 0; i < frags; i++)
 1492                 clrbit(blksfree, bno + i);
 1493         cgp->cg_cs.cs_nffree -= frags;
 1494         cgp->cg_frsum[allocsiz]--;
 1495         if (frags != allocsiz)
 1496                 cgp->cg_frsum[allocsiz - frags]++;
 1497         UFS_LOCK(ump);
 1498         fs->fs_cstotal.cs_nffree -= frags;
 1499         fs->fs_cs(fs, cg).cs_nffree -= frags;
 1500         fs->fs_fmod = 1;
 1501         blkno = cgbase(fs, cg) + bno;
 1502         ACTIVECLEAR(fs, cg);
 1503         UFS_UNLOCK(ump);
 1504         if (DOINGSOFTDEP(ITOV(ip)))
 1505                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno);
 1506         bdwrite(bp);
 1507         return (blkno);
 1508 
 1509 fail:
 1510         brelse(bp);
 1511         UFS_LOCK(ump);
 1512         return (0);
 1513 }
 1514 
 1515 /*
 1516  * Allocate a block in a cylinder group.
 1517  *
 1518  * This algorithm implements the following policy:
 1519  *   1) allocate the requested block.
 1520  *   2) allocate a rotationally optimal block in the same cylinder.
 1521  *   3) allocate the next available block on the block rotor for the
 1522  *      specified cylinder group.
 1523  * Note that this routine only allocates fs_bsize blocks; these
 1524  * blocks may be fragmented by the routine that allocates them.
 1525  */
 1526 static ufs2_daddr_t
 1527 ffs_alloccgblk(ip, bp, bpref)
 1528         struct inode *ip;
 1529         struct buf *bp;
 1530         ufs2_daddr_t bpref;
 1531 {
 1532         struct fs *fs;
 1533         struct cg *cgp;
 1534         struct ufsmount *ump;
 1535         ufs1_daddr_t bno;
 1536         ufs2_daddr_t blkno;
 1537         u_int8_t *blksfree;
 1538 
 1539         fs = ip->i_fs;
 1540         ump = ip->i_ump;
 1541         mtx_assert(UFS_MTX(ump), MA_OWNED);
 1542         cgp = (struct cg *)bp->b_data;
 1543         blksfree = cg_blksfree(cgp);
 1544         if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
 1545                 bpref = cgp->cg_rotor;
 1546         } else {
 1547                 bpref = blknum(fs, bpref);
 1548                 bno = dtogd(fs, bpref);
 1549                 /*
 1550                  * if the requested block is available, use it
 1551                  */
 1552                 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno)))
 1553                         goto gotit;
 1554         }
 1555         /*
 1556          * Take the next available block in this cylinder group.
 1557          */
 1558         bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
 1559         if (bno < 0)
 1560                 return (0);
 1561         cgp->cg_rotor = bno;
 1562 gotit:
 1563         blkno = fragstoblks(fs, bno);
 1564         ffs_clrblock(fs, blksfree, (long)blkno);
 1565         ffs_clusteracct(ump, fs, cgp, blkno, -1);
 1566         cgp->cg_cs.cs_nbfree--;
 1567         fs->fs_cstotal.cs_nbfree--;
 1568         fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
 1569         fs->fs_fmod = 1;
 1570         blkno = cgbase(fs, cgp->cg_cgx) + bno;
 1571         /* XXX Fixme. */
 1572         UFS_UNLOCK(ump);
 1573         if (DOINGSOFTDEP(ITOV(ip)))
 1574                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno);
 1575         UFS_LOCK(ump);
 1576         return (blkno);
 1577 }
 1578 
 1579 /*
 1580  * Determine whether a cluster can be allocated.
 1581  *
 1582  * We do not currently check for optimal rotational layout if there
 1583  * are multiple choices in the same cylinder group. Instead we just
 1584  * take the first one that we find following bpref.
 1585  */
 1586 static ufs2_daddr_t
 1587 ffs_clusteralloc(ip, cg, bpref, len)
 1588         struct inode *ip;
 1589         u_int cg;
 1590         ufs2_daddr_t bpref;
 1591         int len;
 1592 {
 1593         struct fs *fs;
 1594         struct cg *cgp;
 1595         struct buf *bp;
 1596         struct ufsmount *ump;
 1597         int i, run, bit, map, got;
 1598         ufs2_daddr_t bno;
 1599         u_char *mapp;
 1600         int32_t *lp;
 1601         u_int8_t *blksfree;
 1602 
 1603         fs = ip->i_fs;
 1604         ump = ip->i_ump;
 1605         if (fs->fs_maxcluster[cg] < len)
 1606                 return (0);
 1607         UFS_UNLOCK(ump);
 1608         if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
 1609             NOCRED, &bp))
 1610                 goto fail_lock;
 1611         cgp = (struct cg *)bp->b_data;
 1612         if (!cg_chkmagic(cgp))
 1613                 goto fail_lock;
 1614         bp->b_xflags |= BX_BKGRDWRITE;
 1615         /*
 1616          * Check to see if a cluster of the needed size (or bigger) is
 1617          * available in this cylinder group.
 1618          */
 1619         lp = &cg_clustersum(cgp)[len];
 1620         for (i = len; i <= fs->fs_contigsumsize; i++)
 1621                 if (*lp++ > 0)
 1622                         break;
 1623         if (i > fs->fs_contigsumsize) {
 1624                 /*
 1625                  * This is the first time looking for a cluster in this
 1626                  * cylinder group. Update the cluster summary information
 1627                  * to reflect the true maximum sized cluster so that
 1628                  * future cluster allocation requests can avoid reading
 1629                  * the cylinder group map only to find no clusters.
 1630                  */
 1631                 lp = &cg_clustersum(cgp)[len - 1];
 1632                 for (i = len - 1; i > 0; i--)
 1633                         if (*lp-- > 0)
 1634                                 break;
 1635                 UFS_LOCK(ump);
 1636                 fs->fs_maxcluster[cg] = i;
 1637                 goto fail;
 1638         }
 1639         /*
 1640          * Search the cluster map to find a big enough cluster.
 1641          * We take the first one that we find, even if it is larger
 1642          * than we need as we prefer to get one close to the previous
 1643          * block allocation. We do not search before the current
 1644          * preference point as we do not want to allocate a block
 1645          * that is allocated before the previous one (as we will
 1646          * then have to wait for another pass of the elevator
 1647          * algorithm before it will be read). We prefer to fail and
 1648          * be recalled to try an allocation in the next cylinder group.
 1649          */
 1650         if (dtog(fs, bpref) != cg)
 1651                 bpref = 0;
 1652         else
 1653                 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
 1654         mapp = &cg_clustersfree(cgp)[bpref / NBBY];
 1655         map = *mapp++;
 1656         bit = 1 << (bpref % NBBY);
 1657         for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
 1658                 if ((map & bit) == 0) {
 1659                         run = 0;
 1660                 } else {
 1661                         run++;
 1662                         if (run == len)
 1663                                 break;
 1664                 }
 1665                 if ((got & (NBBY - 1)) != (NBBY - 1)) {
 1666                         bit <<= 1;
 1667                 } else {
 1668                         map = *mapp++;
 1669                         bit = 1;
 1670                 }
 1671         }
 1672         if (got >= cgp->cg_nclusterblks)
 1673                 goto fail_lock;
 1674         /*
 1675          * Allocate the cluster that we have found.
 1676          */
 1677         blksfree = cg_blksfree(cgp);
 1678         for (i = 1; i <= len; i++)
 1679                 if (!ffs_isblock(fs, blksfree, got - run + i))
 1680                         panic("ffs_clusteralloc: map mismatch");
 1681         bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1);
 1682         if (dtog(fs, bno) != cg)
 1683                 panic("ffs_clusteralloc: allocated out of group");
 1684         len = blkstofrags(fs, len);
 1685         UFS_LOCK(ump);
 1686         for (i = 0; i < len; i += fs->fs_frag)
 1687                 if (ffs_alloccgblk(ip, bp, bno + i) != bno + i)
 1688                         panic("ffs_clusteralloc: lost block");
 1689         ACTIVECLEAR(fs, cg);
 1690         UFS_UNLOCK(ump);
 1691         bdwrite(bp);
 1692         return (bno);
 1693 
 1694 fail_lock:
 1695         UFS_LOCK(ump);
 1696 fail:
 1697         brelse(bp);
 1698         return (0);
 1699 }
 1700 
 1701 /*
 1702  * Determine whether an inode can be allocated.
 1703  *
 1704  * Check to see if an inode is available, and if it is,
 1705  * allocate it using the following policy:
 1706  *   1) allocate the requested inode.
 1707  *   2) allocate the next available inode after the requested
 1708  *      inode in the specified cylinder group.
 1709  */
 1710 static ufs2_daddr_t
 1711 ffs_nodealloccg(ip, cg, ipref, mode)
 1712         struct inode *ip;
 1713         u_int cg;
 1714         ufs2_daddr_t ipref;
 1715         int mode;
 1716 {
 1717         struct fs *fs;
 1718         struct cg *cgp;
 1719         struct buf *bp, *ibp;
 1720         struct ufsmount *ump;
 1721         u_int8_t *inosused;
 1722         struct ufs2_dinode *dp2;
 1723         int error, start, len, loc, map, i;
 1724 
 1725         fs = ip->i_fs;
 1726         ump = ip->i_ump;
 1727         if (fs->fs_cs(fs, cg).cs_nifree == 0)
 1728                 return (0);
 1729         UFS_UNLOCK(ump);
 1730         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1731                 (int)fs->fs_cgsize, NOCRED, &bp);
 1732         if (error) {
 1733                 brelse(bp);
 1734                 UFS_LOCK(ump);
 1735                 return (0);
 1736         }
 1737         cgp = (struct cg *)bp->b_data;
 1738         if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
 1739                 brelse(bp);
 1740                 UFS_LOCK(ump);
 1741                 return (0);
 1742         }
 1743         bp->b_xflags |= BX_BKGRDWRITE;
 1744         cgp->cg_old_time = cgp->cg_time = time_second;
 1745         inosused = cg_inosused(cgp);
 1746         if (ipref) {
 1747                 ipref %= fs->fs_ipg;
 1748                 if (isclr(inosused, ipref))
 1749                         goto gotit;
 1750         }
 1751         start = cgp->cg_irotor / NBBY;
 1752         len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
 1753         loc = skpc(0xff, len, &inosused[start]);
 1754         if (loc == 0) {
 1755                 len = start + 1;
 1756                 start = 0;
 1757                 loc = skpc(0xff, len, &inosused[0]);
 1758                 if (loc == 0) {
 1759                         printf("cg = %d, irotor = %ld, fs = %s\n",
 1760                             cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
 1761                         panic("ffs_nodealloccg: map corrupted");
 1762                         /* NOTREACHED */
 1763                 }
 1764         }
 1765         i = start + len - loc;
 1766         map = inosused[i];
 1767         ipref = i * NBBY;
 1768         for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
 1769                 if ((map & i) == 0) {
 1770                         cgp->cg_irotor = ipref;
 1771                         goto gotit;
 1772                 }
 1773         }
 1774         printf("fs = %s\n", fs->fs_fsmnt);
 1775         panic("ffs_nodealloccg: block not in map");
 1776         /* NOTREACHED */
 1777 gotit:
 1778         /*
 1779          * Check to see if we need to initialize more inodes.
 1780          */
 1781         ibp = NULL;
 1782         if (fs->fs_magic == FS_UFS2_MAGIC &&
 1783             ipref + INOPB(fs) > cgp->cg_initediblk &&
 1784             cgp->cg_initediblk < cgp->cg_niblk) {
 1785                 ibp = getblk(ip->i_devvp, fsbtodb(fs,
 1786                     ino_to_fsba(fs, cg * fs->fs_ipg + cgp->cg_initediblk)),
 1787                     (int)fs->fs_bsize, 0, 0, 0);
 1788                 bzero(ibp->b_data, (int)fs->fs_bsize);
 1789                 dp2 = (struct ufs2_dinode *)(ibp->b_data);
 1790                 for (i = 0; i < INOPB(fs); i++) {
 1791                         dp2->di_gen = arc4random() / 2 + 1;
 1792                         dp2++;
 1793                 }
 1794                 cgp->cg_initediblk += INOPB(fs);
 1795         }
 1796         UFS_LOCK(ump);
 1797         ACTIVECLEAR(fs, cg);
 1798         setbit(inosused, ipref);
 1799         cgp->cg_cs.cs_nifree--;
 1800         fs->fs_cstotal.cs_nifree--;
 1801         fs->fs_cs(fs, cg).cs_nifree--;
 1802         fs->fs_fmod = 1;
 1803         if ((mode & IFMT) == IFDIR) {
 1804                 cgp->cg_cs.cs_ndir++;
 1805                 fs->fs_cstotal.cs_ndir++;
 1806                 fs->fs_cs(fs, cg).cs_ndir++;
 1807         }
 1808         UFS_UNLOCK(ump);
 1809         if (DOINGSOFTDEP(ITOV(ip)))
 1810                 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref);
 1811         bdwrite(bp);
 1812         if (ibp != NULL)
 1813                 bawrite(ibp);
 1814         return ((ino_t)(cg * fs->fs_ipg + ipref));
 1815 }
 1816 
 1817 /*
 1818  * check if a block is free
 1819  */
 1820 static int
 1821 ffs_isfreeblock(struct fs *fs, u_char *cp, ufs1_daddr_t h)
 1822 {
 1823 
 1824         switch ((int)fs->fs_frag) {
 1825         case 8:
 1826                 return (cp[h] == 0);
 1827         case 4:
 1828                 return ((cp[h >> 1] & (0x0f << ((h & 0x1) << 2))) == 0);
 1829         case 2:
 1830                 return ((cp[h >> 2] & (0x03 << ((h & 0x3) << 1))) == 0);
 1831         case 1:
 1832                 return ((cp[h >> 3] & (0x01 << (h & 0x7))) == 0);
 1833         default:
 1834                 panic("ffs_isfreeblock");
 1835         }
 1836         return (0);
 1837 }
 1838 
 1839 /*
 1840  * Free a block or fragment.
 1841  *
 1842  * The specified block or fragment is placed back in the
 1843  * free map. If a fragment is deallocated, a possible
 1844  * block reassembly is checked.
 1845  */
 1846 void
 1847 ffs_blkfree(ump, fs, devvp, bno, size, inum)
 1848         struct ufsmount *ump;
 1849         struct fs *fs;
 1850         struct vnode *devvp;
 1851         ufs2_daddr_t bno;
 1852         long size;
 1853         ino_t inum;
 1854 {
 1855         struct cg *cgp;
 1856         struct buf *bp;
 1857         ufs1_daddr_t fragno, cgbno;
 1858         ufs2_daddr_t cgblkno;
 1859         int i, blk, frags, bbase;
 1860         u_int cg;
 1861         u_int8_t *blksfree;
 1862         struct cdev *dev;
 1863 
 1864         cg = dtog(fs, bno);
 1865         if (devvp->v_type == VREG) {
 1866                 /* devvp is a snapshot */
 1867                 dev = VTOI(devvp)->i_devvp->v_rdev;
 1868                 cgblkno = fragstoblks(fs, cgtod(fs, cg));
 1869         } else {
 1870                 /* devvp is a normal disk device */
 1871                 dev = devvp->v_rdev;
 1872                 cgblkno = fsbtodb(fs, cgtod(fs, cg));
 1873                 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree");
 1874                 if ((devvp->v_vflag & VV_COPYONWRITE) &&
 1875                     ffs_snapblkfree(fs, devvp, bno, size, inum))
 1876                         return;
 1877         }
 1878 #ifdef INVARIANTS
 1879         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
 1880             fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
 1881                 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n",
 1882                     devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize,
 1883                     size, fs->fs_fsmnt);
 1884                 panic("ffs_blkfree: bad size");
 1885         }
 1886 #endif
 1887         if ((u_int)bno >= fs->fs_size) {
 1888                 printf("bad block %jd, ino %lu\n", (intmax_t)bno,
 1889                     (u_long)inum);
 1890                 ffs_fserr(fs, inum, "bad block");
 1891                 return;
 1892         }
 1893         if (bread(devvp, cgblkno, (int)fs->fs_cgsize, NOCRED, &bp)) {
 1894                 brelse(bp);
 1895                 return;
 1896         }
 1897         cgp = (struct cg *)bp->b_data;
 1898         if (!cg_chkmagic(cgp)) {
 1899                 brelse(bp);
 1900                 return;
 1901         }
 1902         bp->b_xflags |= BX_BKGRDWRITE;
 1903         cgp->cg_old_time = cgp->cg_time = time_second;
 1904         cgbno = dtogd(fs, bno);
 1905         blksfree = cg_blksfree(cgp);
 1906         UFS_LOCK(ump);
 1907         if (size == fs->fs_bsize) {
 1908                 fragno = fragstoblks(fs, cgbno);
 1909                 if (!ffs_isfreeblock(fs, blksfree, fragno)) {
 1910                         if (devvp->v_type == VREG) {
 1911                                 UFS_UNLOCK(ump);
 1912                                 /* devvp is a snapshot */
 1913                                 brelse(bp);
 1914                                 return;
 1915                         }
 1916                         printf("dev = %s, block = %jd, fs = %s\n",
 1917                             devtoname(dev), (intmax_t)bno, fs->fs_fsmnt);
 1918                         panic("ffs_blkfree: freeing free block");
 1919                 }
 1920                 ffs_setblock(fs, blksfree, fragno);
 1921                 ffs_clusteracct(ump, fs, cgp, fragno, 1);
 1922                 cgp->cg_cs.cs_nbfree++;
 1923                 fs->fs_cstotal.cs_nbfree++;
 1924                 fs->fs_cs(fs, cg).cs_nbfree++;
 1925         } else {
 1926                 bbase = cgbno - fragnum(fs, cgbno);
 1927                 /*
 1928                  * decrement the counts associated with the old frags
 1929                  */
 1930                 blk = blkmap(fs, blksfree, bbase);
 1931                 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
 1932                 /*
 1933                  * deallocate the fragment
 1934                  */
 1935                 frags = numfrags(fs, size);
 1936                 for (i = 0; i < frags; i++) {
 1937                         if (isset(blksfree, cgbno + i)) {
 1938                                 printf("dev = %s, block = %jd, fs = %s\n",
 1939                                     devtoname(dev), (intmax_t)(bno + i),
 1940                                     fs->fs_fsmnt);
 1941                                 panic("ffs_blkfree: freeing free frag");
 1942                         }
 1943                         setbit(blksfree, cgbno + i);
 1944                 }
 1945                 cgp->cg_cs.cs_nffree += i;
 1946                 fs->fs_cstotal.cs_nffree += i;
 1947                 fs->fs_cs(fs, cg).cs_nffree += i;
 1948                 /*
 1949                  * add back in counts associated with the new frags
 1950                  */
 1951                 blk = blkmap(fs, blksfree, bbase);
 1952                 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
 1953                 /*
 1954                  * if a complete block has been reassembled, account for it
 1955                  */
 1956                 fragno = fragstoblks(fs, bbase);
 1957                 if (ffs_isblock(fs, blksfree, fragno)) {
 1958                         cgp->cg_cs.cs_nffree -= fs->fs_frag;
 1959                         fs->fs_cstotal.cs_nffree -= fs->fs_frag;
 1960                         fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
 1961                         ffs_clusteracct(ump, fs, cgp, fragno, 1);
 1962                         cgp->cg_cs.cs_nbfree++;
 1963                         fs->fs_cstotal.cs_nbfree++;
 1964                         fs->fs_cs(fs, cg).cs_nbfree++;
 1965                 }
 1966         }
 1967         fs->fs_fmod = 1;
 1968         ACTIVECLEAR(fs, cg);
 1969         UFS_UNLOCK(ump);
 1970         bdwrite(bp);
 1971 }
 1972 
 1973 #ifdef INVARIANTS
 1974 /*
 1975  * Verify allocation of a block or fragment. Returns true if block or
 1976  * fragment is allocated, false if it is free.
 1977  */
 1978 static int
 1979 ffs_checkblk(ip, bno, size)
 1980         struct inode *ip;
 1981         ufs2_daddr_t bno;
 1982         long size;
 1983 {
 1984         struct fs *fs;
 1985         struct cg *cgp;
 1986         struct buf *bp;
 1987         ufs1_daddr_t cgbno;
 1988         int i, error, frags, free;
 1989         u_int8_t *blksfree;
 1990 
 1991         fs = ip->i_fs;
 1992         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
 1993                 printf("bsize = %ld, size = %ld, fs = %s\n",
 1994                     (long)fs->fs_bsize, size, fs->fs_fsmnt);
 1995                 panic("ffs_checkblk: bad size");
 1996         }
 1997         if ((u_int)bno >= fs->fs_size)
 1998                 panic("ffs_checkblk: bad block %jd", (intmax_t)bno);
 1999         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
 2000                 (int)fs->fs_cgsize, NOCRED, &bp);
 2001         if (error)
 2002                 panic("ffs_checkblk: cg bread failed");
 2003         cgp = (struct cg *)bp->b_data;
 2004         if (!cg_chkmagic(cgp))
 2005                 panic("ffs_checkblk: cg magic mismatch");
 2006         bp->b_xflags |= BX_BKGRDWRITE;
 2007         blksfree = cg_blksfree(cgp);
 2008         cgbno = dtogd(fs, bno);
 2009         if (size == fs->fs_bsize) {
 2010                 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno));
 2011         } else {
 2012                 frags = numfrags(fs, size);
 2013                 for (free = 0, i = 0; i < frags; i++)
 2014                         if (isset(blksfree, cgbno + i))
 2015                                 free++;
 2016                 if (free != 0 && free != frags)
 2017                         panic("ffs_checkblk: partially free fragment");
 2018         }
 2019         brelse(bp);
 2020         return (!free);
 2021 }
 2022 #endif /* INVARIANTS */
 2023 
 2024 /*
 2025  * Free an inode.
 2026  */
 2027 int
 2028 ffs_vfree(pvp, ino, mode)
 2029         struct vnode *pvp;
 2030         ino_t ino;
 2031         int mode;
 2032 {
 2033         struct inode *ip;
 2034 
 2035         if (DOINGSOFTDEP(pvp)) {
 2036                 softdep_freefile(pvp, ino, mode);
 2037                 return (0);
 2038         }
 2039         ip = VTOI(pvp);
 2040         return (ffs_freefile(ip->i_ump, ip->i_fs, ip->i_devvp, ino, mode));
 2041 }
 2042 
 2043 /*
 2044  * Do the actual free operation.
 2045  * The specified inode is placed back in the free map.
 2046  */
 2047 int
 2048 ffs_freefile(ump, fs, devvp, ino, mode)
 2049         struct ufsmount *ump;
 2050         struct fs *fs;
 2051         struct vnode *devvp;
 2052         ino_t ino;
 2053         int mode;
 2054 {
 2055         struct cg *cgp;
 2056         struct buf *bp;
 2057         ufs2_daddr_t cgbno;
 2058         int error;
 2059         u_int cg;
 2060         u_int8_t *inosused;
 2061         struct cdev *dev;
 2062 
 2063         cg = ino_to_cg(fs, ino);
 2064         if (devvp->v_type == VREG) {
 2065                 /* devvp is a snapshot */
 2066                 dev = VTOI(devvp)->i_devvp->v_rdev;
 2067                 cgbno = fragstoblks(fs, cgtod(fs, cg));
 2068         } else {
 2069                 /* devvp is a normal disk device */
 2070                 dev = devvp->v_rdev;
 2071                 cgbno = fsbtodb(fs, cgtod(fs, cg));
 2072         }
 2073         if (ino >= fs->fs_ipg * fs->fs_ncg)
 2074                 panic("ffs_freefile: range: dev = %s, ino = %lu, fs = %s",
 2075                     devtoname(dev), (u_long)ino, fs->fs_fsmnt);
 2076         if ((error = bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp))) {
 2077                 brelse(bp);
 2078                 return (error);
 2079         }
 2080         cgp = (struct cg *)bp->b_data;
 2081         if (!cg_chkmagic(cgp)) {
 2082                 brelse(bp);
 2083                 return (0);
 2084         }
 2085         bp->b_xflags |= BX_BKGRDWRITE;
 2086         cgp->cg_old_time = cgp->cg_time = time_second;
 2087         inosused = cg_inosused(cgp);
 2088         ino %= fs->fs_ipg;
 2089         if (isclr(inosused, ino)) {
 2090                 printf("dev = %s, ino = %u, fs = %s\n", devtoname(dev),
 2091                     ino + cg * fs->fs_ipg, fs->fs_fsmnt);
 2092                 if (fs->fs_ronly == 0)
 2093                         panic("ffs_freefile: freeing free inode");
 2094         }
 2095         clrbit(inosused, ino);
 2096         if (ino < cgp->cg_irotor)
 2097                 cgp->cg_irotor = ino;
 2098         cgp->cg_cs.cs_nifree++;
 2099         UFS_LOCK(ump);
 2100         fs->fs_cstotal.cs_nifree++;
 2101         fs->fs_cs(fs, cg).cs_nifree++;
 2102         if ((mode & IFMT) == IFDIR) {
 2103                 cgp->cg_cs.cs_ndir--;
 2104                 fs->fs_cstotal.cs_ndir--;
 2105                 fs->fs_cs(fs, cg).cs_ndir--;
 2106         }
 2107         fs->fs_fmod = 1;
 2108         ACTIVECLEAR(fs, cg);
 2109         UFS_UNLOCK(ump);
 2110         bdwrite(bp);
 2111         return (0);
 2112 }
 2113 
 2114 /*
 2115  * Check to see if a file is free.
 2116  */
 2117 int
 2118 ffs_checkfreefile(fs, devvp, ino)
 2119         struct fs *fs;
 2120         struct vnode *devvp;
 2121         ino_t ino;
 2122 {
 2123         struct cg *cgp;
 2124         struct buf *bp;
 2125         ufs2_daddr_t cgbno;
 2126         int ret;
 2127         u_int cg;
 2128         u_int8_t *inosused;
 2129 
 2130         cg = ino_to_cg(fs, ino);
 2131         if (devvp->v_type == VREG) {
 2132                 /* devvp is a snapshot */
 2133                 cgbno = fragstoblks(fs, cgtod(fs, cg));
 2134         } else {
 2135                 /* devvp is a normal disk device */
 2136                 cgbno = fsbtodb(fs, cgtod(fs, cg));
 2137         }
 2138         if (ino >= fs->fs_ipg * fs->fs_ncg)
 2139                 return (1);
 2140         if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp)) {
 2141                 brelse(bp);
 2142                 return (1);
 2143         }
 2144         cgp = (struct cg *)bp->b_data;
 2145         if (!cg_chkmagic(cgp)) {
 2146                 brelse(bp);
 2147                 return (1);
 2148         }
 2149         inosused = cg_inosused(cgp);
 2150         ino %= fs->fs_ipg;
 2151         ret = isclr(inosused, ino);
 2152         brelse(bp);
 2153         return (ret);
 2154 }
 2155 
 2156 /*
 2157  * Find a block of the specified size in the specified cylinder group.
 2158  *
 2159  * It is a panic if a request is made to find a block if none are
 2160  * available.
 2161  */
 2162 static ufs1_daddr_t
 2163 ffs_mapsearch(fs, cgp, bpref, allocsiz)
 2164         struct fs *fs;
 2165         struct cg *cgp;
 2166         ufs2_daddr_t bpref;
 2167         int allocsiz;
 2168 {
 2169         ufs1_daddr_t bno;
 2170         int start, len, loc, i;
 2171         int blk, field, subfield, pos;
 2172         u_int8_t *blksfree;
 2173 
 2174         /*
 2175          * find the fragment by searching through the free block
 2176          * map for an appropriate bit pattern
 2177          */
 2178         if (bpref)
 2179                 start = dtogd(fs, bpref) / NBBY;
 2180         else
 2181                 start = cgp->cg_frotor / NBBY;
 2182         blksfree = cg_blksfree(cgp);
 2183         len = howmany(fs->fs_fpg, NBBY) - start;
 2184         loc = scanc((u_int)len, (u_char *)&blksfree[start],
 2185                 fragtbl[fs->fs_frag],
 2186                 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
 2187         if (loc == 0) {
 2188                 len = start + 1;
 2189                 start = 0;
 2190                 loc = scanc((u_int)len, (u_char *)&blksfree[0],
 2191                         fragtbl[fs->fs_frag],
 2192                         (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
 2193                 if (loc == 0) {
 2194                         printf("start = %d, len = %d, fs = %s\n",
 2195                             start, len, fs->fs_fsmnt);
 2196                         panic("ffs_alloccg: map corrupted");
 2197                         /* NOTREACHED */
 2198                 }
 2199         }
 2200         bno = (start + len - loc) * NBBY;
 2201         cgp->cg_frotor = bno;
 2202         /*
 2203          * found the byte in the map
 2204          * sift through the bits to find the selected frag
 2205          */
 2206         for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
 2207                 blk = blkmap(fs, blksfree, bno);
 2208                 blk <<= 1;
 2209                 field = around[allocsiz];
 2210                 subfield = inside[allocsiz];
 2211                 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
 2212                         if ((blk & field) == subfield)
 2213                                 return (bno + pos);
 2214                         field <<= 1;
 2215                         subfield <<= 1;
 2216                 }
 2217         }
 2218         printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
 2219         panic("ffs_alloccg: block not in map");
 2220         return (-1);
 2221 }
 2222 
 2223 /*
 2224  * Update the cluster map because of an allocation or free.
 2225  *
 2226  * Cnt == 1 means free; cnt == -1 means allocating.
 2227  */
 2228 void
 2229 ffs_clusteracct(ump, fs, cgp, blkno, cnt)
 2230         struct ufsmount *ump;
 2231         struct fs *fs;
 2232         struct cg *cgp;
 2233         ufs1_daddr_t blkno;
 2234         int cnt;
 2235 {
 2236         int32_t *sump;
 2237         int32_t *lp;
 2238         u_char *freemapp, *mapp;
 2239         int i, start, end, forw, back, map, bit;
 2240 
 2241         mtx_assert(UFS_MTX(ump), MA_OWNED);
 2242 
 2243         if (fs->fs_contigsumsize <= 0)
 2244                 return;
 2245         freemapp = cg_clustersfree(cgp);
 2246         sump = cg_clustersum(cgp);
 2247         /*
 2248          * Allocate or clear the actual block.
 2249          */
 2250         if (cnt > 0)
 2251                 setbit(freemapp, blkno);
 2252         else
 2253                 clrbit(freemapp, blkno);
 2254         /*
 2255          * Find the size of the cluster going forward.
 2256          */
 2257         start = blkno + 1;
 2258         end = start + fs->fs_contigsumsize;
 2259         if (end >= cgp->cg_nclusterblks)
 2260                 end = cgp->cg_nclusterblks;
 2261         mapp = &freemapp[start / NBBY];
 2262         map = *mapp++;
 2263         bit = 1 << (start % NBBY);
 2264         for (i = start; i < end; i++) {
 2265                 if ((map & bit) == 0)
 2266                         break;
 2267                 if ((i & (NBBY - 1)) != (NBBY - 1)) {
 2268                         bit <<= 1;
 2269                 } else {
 2270                         map = *mapp++;
 2271                         bit = 1;
 2272                 }
 2273         }
 2274         forw = i - start;
 2275         /*
 2276          * Find the size of the cluster going backward.
 2277          */
 2278         start = blkno - 1;
 2279         end = start - fs->fs_contigsumsize;
 2280         if (end < 0)
 2281                 end = -1;
 2282         mapp = &freemapp[start / NBBY];
 2283         map = *mapp--;
 2284         bit = 1 << (start % NBBY);
 2285         for (i = start; i > end; i--) {
 2286                 if ((map & bit) == 0)
 2287                         break;
 2288                 if ((i & (NBBY - 1)) != 0) {
 2289                         bit >>= 1;
 2290                 } else {
 2291                         map = *mapp--;
 2292                         bit = 1 << (NBBY - 1);
 2293                 }
 2294         }
 2295         back = start - i;
 2296         /*
 2297          * Account for old cluster and the possibly new forward and
 2298          * back clusters.
 2299          */
 2300         i = back + forw + 1;
 2301         if (i > fs->fs_contigsumsize)
 2302                 i = fs->fs_contigsumsize;
 2303         sump[i] += cnt;
 2304         if (back > 0)
 2305                 sump[back] -= cnt;
 2306         if (forw > 0)
 2307                 sump[forw] -= cnt;
 2308         /*
 2309          * Update cluster summary information.
 2310          */
 2311         lp = &sump[fs->fs_contigsumsize];
 2312         for (i = fs->fs_contigsumsize; i > 0; i--)
 2313                 if (*lp-- > 0)
 2314                         break;
 2315         fs->fs_maxcluster[cgp->cg_cgx] = i;
 2316 }
 2317 
 2318 /*
 2319  * Fserr prints the name of a filesystem with an error diagnostic.
 2320  *
 2321  * The form of the error message is:
 2322  *      fs: error message
 2323  */
 2324 static void
 2325 ffs_fserr(fs, inum, cp)
 2326         struct fs *fs;
 2327         ino_t inum;
 2328         char *cp;
 2329 {
 2330         struct thread *td = curthread;  /* XXX */
 2331         struct proc *p = td->td_proc;
 2332 
 2333         log(LOG_ERR, "pid %d (%s), uid %d inumber %d on %s: %s\n",
 2334             p->p_pid, p->p_comm, td->td_ucred->cr_uid, inum, fs->fs_fsmnt, cp);
 2335 }
 2336 
 2337 /*
 2338  * This function provides the capability for the fsck program to
 2339  * update an active filesystem. Eleven operations are provided:
 2340  *
 2341  * adjrefcnt(inode, amt) - adjusts the reference count on the
 2342  *      specified inode by the specified amount. Under normal
 2343  *      operation the count should always go down. Decrementing
 2344  *      the count to zero will cause the inode to be freed.
 2345  * adjblkcnt(inode, amt) - adjust the number of blocks used to
 2346  *      by the specifed amount.
 2347  * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) -
 2348  *      adjust the superblock summary.
 2349  * freedirs(inode, count) - directory inodes [inode..inode + count - 1]
 2350  *      are marked as free. Inodes should never have to be marked
 2351  *      as in use.
 2352  * freefiles(inode, count) - file inodes [inode..inode + count - 1]
 2353  *      are marked as free. Inodes should never have to be marked
 2354  *      as in use.
 2355  * freeblks(blockno, size) - blocks [blockno..blockno + size - 1]
 2356  *      are marked as free. Blocks should never have to be marked
 2357  *      as in use.
 2358  * setflags(flags, set/clear) - the fs_flags field has the specified
 2359  *      flags set (second parameter +1) or cleared (second parameter -1).
 2360  */
 2361 
 2362 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS);
 2363 
 2364 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT,
 2365         0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count");
 2366 
 2367 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR,
 2368         sysctl_ffs_fsck, "Adjust Inode Used Blocks Count");
 2369 
 2370 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR,
 2371         sysctl_ffs_fsck, "Adjust number of directories");
 2372 
 2373 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR,
 2374         sysctl_ffs_fsck, "Adjust number of free blocks");
 2375 
 2376 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR,
 2377         sysctl_ffs_fsck, "Adjust number of free inodes");
 2378 
 2379 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR,
 2380         sysctl_ffs_fsck, "Adjust number of free frags");
 2381 
 2382 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR,
 2383         sysctl_ffs_fsck, "Adjust number of free clusters");
 2384 
 2385 static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR,
 2386         sysctl_ffs_fsck, "Free Range of Directory Inodes");
 2387 
 2388 static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR,
 2389         sysctl_ffs_fsck, "Free Range of File Inodes");
 2390 
 2391 static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR,
 2392         sysctl_ffs_fsck, "Free Range of Blocks");
 2393 
 2394 static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR,
 2395         sysctl_ffs_fsck, "Change Filesystem Flags");
 2396 
 2397 #ifdef DEBUG
 2398 static int fsckcmds = 0;
 2399 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, "");
 2400 #endif /* DEBUG */
 2401 
 2402 static int
 2403 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS)
 2404 {
 2405         struct fsck_cmd cmd;
 2406         struct ufsmount *ump;
 2407         struct vnode *vp;
 2408         struct inode *ip;
 2409         struct mount *mp;
 2410         struct fs *fs;
 2411         ufs2_daddr_t blkno;
 2412         long blkcnt, blksize;
 2413         struct file *fp;
 2414         int filetype, error;
 2415 
 2416         if (req->newlen > sizeof cmd)
 2417                 return (EBADRPC);
 2418         if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0)
 2419                 return (error);
 2420         if (cmd.version != FFS_CMD_VERSION)
 2421                 return (ERPCMISMATCH);
 2422         if ((error = getvnode(curproc->p_fd, cmd.handle, &fp)) != 0)
 2423                 return (error);
 2424         vn_start_write(fp->f_data, &mp, V_WAIT);
 2425         if (mp == 0 || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) {
 2426                 vn_finished_write(mp);
 2427                 fdrop(fp, curthread);
 2428                 return (EINVAL);
 2429         }
 2430         if (mp->mnt_flag & MNT_RDONLY) {
 2431                 vn_finished_write(mp);
 2432                 fdrop(fp, curthread);
 2433                 return (EROFS);
 2434         }
 2435         ump = VFSTOUFS(mp);
 2436         fs = ump->um_fs;
 2437         filetype = IFREG;
 2438 
 2439         switch (oidp->oid_number) {
 2440 
 2441         case FFS_SET_FLAGS:
 2442 #ifdef DEBUG
 2443                 if (fsckcmds)
 2444                         printf("%s: %s flags\n", mp->mnt_stat.f_mntonname,
 2445                             cmd.size > 0 ? "set" : "clear");
 2446 #endif /* DEBUG */
 2447                 if (cmd.size > 0)
 2448                         fs->fs_flags |= (long)cmd.value;
 2449                 else
 2450                         fs->fs_flags &= ~(long)cmd.value;
 2451                 break;
 2452 
 2453         case FFS_ADJ_REFCNT:
 2454 #ifdef DEBUG
 2455                 if (fsckcmds) {
 2456                         printf("%s: adjust inode %jd count by %jd\n",
 2457                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
 2458                             (intmax_t)cmd.size);
 2459                 }
 2460 #endif /* DEBUG */
 2461                 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
 2462                         break;
 2463                 ip = VTOI(vp);
 2464                 ip->i_nlink += cmd.size;
 2465                 DIP_SET(ip, i_nlink, ip->i_nlink);
 2466                 ip->i_effnlink += cmd.size;
 2467                 ip->i_flag |= IN_CHANGE;
 2468                 if (DOINGSOFTDEP(vp))
 2469                         softdep_change_linkcnt(ip);
 2470                 vput(vp);
 2471                 break;
 2472 
 2473         case FFS_ADJ_BLKCNT:
 2474 #ifdef DEBUG
 2475                 if (fsckcmds) {
 2476                         printf("%s: adjust inode %jd block count by %jd\n",
 2477                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
 2478                             (intmax_t)cmd.size);
 2479                 }
 2480 #endif /* DEBUG */
 2481                 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
 2482                         break;
 2483                 ip = VTOI(vp);
 2484                 if (ip->i_flag & IN_SPACECOUNTED) {
 2485                         UFS_LOCK(ump);
 2486                         fs->fs_pendingblocks += cmd.size;
 2487                         UFS_UNLOCK(ump);
 2488                 }
 2489                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size);
 2490                 ip->i_flag |= IN_CHANGE;
 2491                 vput(vp);
 2492                 break;
 2493 
 2494         case FFS_DIR_FREE:
 2495                 filetype = IFDIR;
 2496                 /* fall through */
 2497 
 2498         case FFS_FILE_FREE:
 2499 #ifdef DEBUG
 2500                 if (fsckcmds) {
 2501                         if (cmd.size == 1)
 2502                                 printf("%s: free %s inode %d\n",
 2503                                     mp->mnt_stat.f_mntonname,
 2504                                     filetype == IFDIR ? "directory" : "file",
 2505                                     (ino_t)cmd.value);
 2506                         else
 2507                                 printf("%s: free %s inodes %d-%d\n",
 2508                                     mp->mnt_stat.f_mntonname,
 2509                                     filetype == IFDIR ? "directory" : "file",
 2510                                     (ino_t)cmd.value,
 2511                                     (ino_t)(cmd.value + cmd.size - 1));
 2512                 }
 2513 #endif /* DEBUG */
 2514                 while (cmd.size > 0) {
 2515                         if ((error = ffs_freefile(ump, fs, ump->um_devvp,
 2516                             cmd.value, filetype)))
 2517                                 break;
 2518                         cmd.size -= 1;
 2519                         cmd.value += 1;
 2520                 }
 2521                 break;
 2522 
 2523         case FFS_BLK_FREE:
 2524 #ifdef DEBUG
 2525                 if (fsckcmds) {
 2526                         if (cmd.size == 1)
 2527                                 printf("%s: free block %jd\n",
 2528                                     mp->mnt_stat.f_mntonname,
 2529                                     (intmax_t)cmd.value);
 2530                         else
 2531                                 printf("%s: free blocks %jd-%jd\n",
 2532                                     mp->mnt_stat.f_mntonname, 
 2533                                     (intmax_t)cmd.value,
 2534                                     (intmax_t)cmd.value + cmd.size - 1);
 2535                 }
 2536 #endif /* DEBUG */
 2537                 blkno = cmd.value;
 2538                 blkcnt = cmd.size;
 2539                 blksize = fs->fs_frag - (blkno % fs->fs_frag);
 2540                 while (blkcnt > 0) {
 2541                         if (blksize > blkcnt)
 2542                                 blksize = blkcnt;
 2543                         ffs_blkfree(ump, fs, ump->um_devvp, blkno,
 2544                             blksize * fs->fs_fsize, ROOTINO);
 2545                         blkno += blksize;
 2546                         blkcnt -= blksize;
 2547                         blksize = fs->fs_frag;
 2548                 }
 2549                 break;
 2550 
 2551         /*
 2552          * Adjust superblock summaries.  fsck(8) is expected to
 2553          * submit deltas when necessary.
 2554          */
 2555         case FFS_ADJ_NDIR:
 2556 #ifdef DEBUG
 2557                 if (fsckcmds) {
 2558                         printf("%s: adjust number of directories by %jd\n",
 2559                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2560                 }
 2561 #endif /* DEBUG */
 2562                 fs->fs_cstotal.cs_ndir += cmd.value;
 2563                 break;
 2564         case FFS_ADJ_NBFREE:
 2565 #ifdef DEBUG
 2566                 if (fsckcmds) {
 2567                         printf("%s: adjust number of free blocks by %+jd\n",
 2568                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2569                 }
 2570 #endif /* DEBUG */
 2571                 fs->fs_cstotal.cs_nbfree += cmd.value;
 2572                 break;
 2573         case FFS_ADJ_NIFREE:
 2574 #ifdef DEBUG
 2575                 if (fsckcmds) {
 2576                         printf("%s: adjust number of free inodes by %+jd\n",
 2577                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2578                 }
 2579 #endif /* DEBUG */
 2580                 fs->fs_cstotal.cs_nifree += cmd.value;
 2581                 break;
 2582         case FFS_ADJ_NFFREE:
 2583 #ifdef DEBUG
 2584                 if (fsckcmds) {
 2585                         printf("%s: adjust number of free frags by %+jd\n",
 2586                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2587                 }
 2588 #endif /* DEBUG */
 2589                 fs->fs_cstotal.cs_nffree += cmd.value;
 2590                 break;
 2591         case FFS_ADJ_NUMCLUSTERS:
 2592 #ifdef DEBUG
 2593                 if (fsckcmds) {
 2594                         printf("%s: adjust number of free clusters by %+jd\n",
 2595                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2596                 }
 2597 #endif /* DEBUG */
 2598                 fs->fs_cstotal.cs_numclusters += cmd.value;
 2599                 break;
 2600 
 2601         default:
 2602 #ifdef DEBUG
 2603                 if (fsckcmds) {
 2604                         printf("Invalid request %d from fsck\n",
 2605                             oidp->oid_number);
 2606                 }
 2607 #endif /* DEBUG */
 2608                 error = EINVAL;
 2609                 break;
 2610 
 2611         }
 2612         fdrop(fp, curthread);
 2613         vn_finished_write(mp);
 2614         return (error);
 2615 }

Cache object: 16a1d4150e102a4665f23d5a344ba039


[ 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.