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/6.4/sys/ufs/ffs/ffs_alloc.c 176542 2008-02-25 09:30:00Z obrien $");
   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/proc.h>
   75 #include <sys/vnode.h>
   76 #include <sys/mount.h>
   77 #include <sys/kernel.h>
   78 #include <sys/sysctl.h>
   79 #include <sys/syslog.h>
   80 
   81 #include <ufs/ufs/extattr.h>
   82 #include <ufs/ufs/quota.h>
   83 #include <ufs/ufs/inode.h>
   84 #include <ufs/ufs/ufs_extern.h>
   85 #include <ufs/ufs/ufsmount.h>
   86 
   87 #include <ufs/ffs/fs.h>
   88 #include <ufs/ffs/ffs_extern.h>
   89 
   90 typedef ufs2_daddr_t allocfcn_t(struct inode *ip, int cg, ufs2_daddr_t bpref,
   91                                   int size);
   92 
   93 static ufs2_daddr_t ffs_alloccg(struct inode *, int, ufs2_daddr_t, int);
   94 static ufs2_daddr_t
   95               ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t);
   96 #ifdef INVARIANTS
   97 static int      ffs_checkblk(struct inode *, ufs2_daddr_t, long);
   98 #endif
   99 static ufs2_daddr_t ffs_clusteralloc(struct inode *, int, ufs2_daddr_t, int);
  100 static void     ffs_clusteracct(struct ufsmount *, struct fs *, struct cg *,
  101                     ufs1_daddr_t, int);
  102 static ino_t    ffs_dirpref(struct inode *);
  103 static ufs2_daddr_t ffs_fragextend(struct inode *, int, ufs2_daddr_t, int, int);
  104 static void     ffs_fserr(struct fs *, ino_t, char *);
  105 static ufs2_daddr_t     ffs_hashalloc
  106                 (struct inode *, int, ufs2_daddr_t, int, allocfcn_t *);
  107 static ufs2_daddr_t ffs_nodealloccg(struct inode *, int, ufs2_daddr_t, int);
  108 static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int);
  109 static int      ffs_reallocblks_ufs1(struct vop_reallocblks_args *);
  110 static int      ffs_reallocblks_ufs2(struct vop_reallocblks_args *);
  111 
  112 /*
  113  * Allocate a block in the filesystem.
  114  *
  115  * The size of the requested block is given, which must be some
  116  * multiple of fs_fsize and <= fs_bsize.
  117  * A preference may be optionally specified. If a preference is given
  118  * the following hierarchy is used to allocate a block:
  119  *   1) allocate the requested block.
  120  *   2) allocate a rotationally optimal block in the same cylinder.
  121  *   3) allocate a block in the same cylinder group.
  122  *   4) quadradically rehash into other cylinder groups, until an
  123  *      available block is located.
  124  * If no block preference is given the following hierarchy is used
  125  * to allocate a block:
  126  *   1) allocate a block in the cylinder group that contains the
  127  *      inode for the file.
  128  *   2) quadradically rehash into other cylinder groups, until an
  129  *      available block is located.
  130  */
  131 int
  132 ffs_alloc(ip, lbn, bpref, size, cred, bnp)
  133         struct inode *ip;
  134         ufs2_daddr_t lbn, bpref;
  135         int size;
  136         struct ucred *cred;
  137         ufs2_daddr_t *bnp;
  138 {
  139         struct fs *fs;
  140         struct ufsmount *ump;
  141         ufs2_daddr_t bno;
  142         int cg, reclaimed;
  143         static struct timeval lastfail;
  144         static int curfail;
  145         int64_t delta;
  146 #ifdef QUOTA
  147         int error;
  148 #endif
  149 
  150         *bnp = 0;
  151         fs = ip->i_fs;
  152         ump = ip->i_ump;
  153         mtx_assert(UFS_MTX(ump), MA_OWNED);
  154 #ifdef INVARIANTS
  155         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
  156                 printf("dev = %s, bsize = %ld, size = %d, fs = %s\n",
  157                     devtoname(ip->i_dev), (long)fs->fs_bsize, size,
  158                     fs->fs_fsmnt);
  159                 panic("ffs_alloc: bad size");
  160         }
  161         if (cred == NOCRED)
  162                 panic("ffs_alloc: missing credential");
  163 #endif /* INVARIANTS */
  164         reclaimed = 0;
  165 retry:
  166 #ifdef QUOTA
  167         UFS_UNLOCK(ump);
  168         error = chkdq(ip, btodb(size), cred, 0);
  169         if (error)
  170                 return (error);
  171         UFS_LOCK(ump);
  172 #endif
  173         if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
  174                 goto nospace;
  175         if (suser_cred(cred, SUSER_ALLOWJAIL) &&
  176             freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0)
  177                 goto nospace;
  178         if (bpref >= fs->fs_size)
  179                 bpref = 0;
  180         if (bpref == 0)
  181                 cg = ino_to_cg(fs, ip->i_number);
  182         else
  183                 cg = dtog(fs, bpref);
  184         bno = ffs_hashalloc(ip, cg, bpref, size, ffs_alloccg);
  185         if (bno > 0) {
  186                 delta = btodb(size);
  187                 if (ip->i_flag & IN_SPACECOUNTED) {
  188                         UFS_LOCK(ump);
  189                         fs->fs_pendingblocks += delta;
  190                         UFS_UNLOCK(ump);
  191                 }
  192                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  193                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  194                 *bnp = bno;
  195                 return (0);
  196         }
  197 nospace:
  198 #ifdef QUOTA
  199         UFS_UNLOCK(ump);
  200         /*
  201          * Restore user's disk quota because allocation failed.
  202          */
  203         (void) chkdq(ip, -btodb(size), cred, FORCE);
  204         UFS_LOCK(ump);
  205 #endif
  206         if (fs->fs_pendingblocks > 0 && reclaimed == 0) {
  207                 reclaimed = 1;
  208                 softdep_request_cleanup(fs, ITOV(ip));
  209                 goto retry;
  210         }
  211         UFS_UNLOCK(ump);
  212         if (ppsratecheck(&lastfail, &curfail, 1)) {
  213                 ffs_fserr(fs, ip->i_number, "filesystem full");
  214                 uprintf("\n%s: write failed, filesystem is full\n",
  215                     fs->fs_fsmnt);
  216         }
  217         return (ENOSPC);
  218 }
  219 
  220 /*
  221  * Reallocate a fragment to a bigger size
  222  *
  223  * The number and size of the old block is given, and a preference
  224  * and new size is also specified. The allocator attempts to extend
  225  * the original block. Failing that, the regular block allocator is
  226  * invoked to get an appropriate block.
  227  */
  228 int
  229 ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, cred, bpp)
  230         struct inode *ip;
  231         ufs2_daddr_t lbprev;
  232         ufs2_daddr_t bprev;
  233         ufs2_daddr_t bpref;
  234         int osize, nsize;
  235         struct ucred *cred;
  236         struct buf **bpp;
  237 {
  238         struct vnode *vp;
  239         struct fs *fs;
  240         struct buf *bp;
  241         struct ufsmount *ump;
  242         int cg, request, error, reclaimed;
  243         ufs2_daddr_t bno;
  244         static struct timeval lastfail;
  245         static int curfail;
  246         int64_t delta;
  247 
  248         *bpp = 0;
  249         vp = ITOV(ip);
  250         fs = ip->i_fs;
  251         bp = NULL;
  252         ump = ip->i_ump;
  253         mtx_assert(UFS_MTX(ump), MA_OWNED);
  254 #ifdef INVARIANTS
  255         if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
  256                 panic("ffs_realloccg: allocation on suspended filesystem");
  257         if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 ||
  258             (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) {
  259                 printf(
  260                 "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n",
  261                     devtoname(ip->i_dev), (long)fs->fs_bsize, osize,
  262                     nsize, fs->fs_fsmnt);
  263                 panic("ffs_realloccg: bad size");
  264         }
  265         if (cred == NOCRED)
  266                 panic("ffs_realloccg: missing credential");
  267 #endif /* INVARIANTS */
  268         reclaimed = 0;
  269 retry:
  270         if (suser_cred(cred, SUSER_ALLOWJAIL) &&
  271             freespace(fs, fs->fs_minfree) -  numfrags(fs, nsize - osize) < 0) {
  272                 goto nospace;
  273         }
  274         if (bprev == 0) {
  275                 printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n",
  276                     devtoname(ip->i_dev), (long)fs->fs_bsize, (intmax_t)bprev,
  277                     fs->fs_fsmnt);
  278                 panic("ffs_realloccg: bad bprev");
  279         }
  280         UFS_UNLOCK(ump);
  281         /*
  282          * Allocate the extra space in the buffer.
  283          */
  284         error = bread(vp, lbprev, osize, NOCRED, &bp);
  285         if (error) {
  286                 brelse(bp);
  287                 return (error);
  288         }
  289 
  290         if (bp->b_blkno == bp->b_lblkno) {
  291                 if (lbprev >= NDADDR)
  292                         panic("ffs_realloccg: lbprev out of range");
  293                 bp->b_blkno = fsbtodb(fs, bprev);
  294         }
  295 
  296 #ifdef QUOTA
  297         error = chkdq(ip, btodb(nsize - osize), cred, 0);
  298         if (error) {
  299                 brelse(bp);
  300                 return (error);
  301         }
  302 #endif
  303         /*
  304          * Check for extension in the existing location.
  305          */
  306         cg = dtog(fs, bprev);
  307         UFS_LOCK(ump);
  308         bno = ffs_fragextend(ip, cg, bprev, osize, nsize);
  309         if (bno) {
  310                 if (bp->b_blkno != fsbtodb(fs, bno))
  311                         panic("ffs_realloccg: bad blockno");
  312                 delta = btodb(nsize - osize);
  313                 if (ip->i_flag & IN_SPACECOUNTED) {
  314                         UFS_LOCK(ump);
  315                         fs->fs_pendingblocks += delta;
  316                         UFS_UNLOCK(ump);
  317                 }
  318                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  319                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  320                 allocbuf(bp, nsize);
  321                 bp->b_flags |= B_DONE;
  322                 if ((bp->b_flags & (B_MALLOC | B_VMIO)) != B_VMIO)
  323                         bzero((char *)bp->b_data + osize, nsize - osize);
  324                 else
  325                         vfs_bio_clrbuf(bp);
  326                 *bpp = bp;
  327                 return (0);
  328         }
  329         /*
  330          * Allocate a new disk location.
  331          */
  332         if (bpref >= fs->fs_size)
  333                 bpref = 0;
  334         switch ((int)fs->fs_optim) {
  335         case FS_OPTSPACE:
  336                 /*
  337                  * Allocate an exact sized fragment. Although this makes
  338                  * best use of space, we will waste time relocating it if
  339                  * the file continues to grow. If the fragmentation is
  340                  * less than half of the minimum free reserve, we choose
  341                  * to begin optimizing for time.
  342                  */
  343                 request = nsize;
  344                 if (fs->fs_minfree <= 5 ||
  345                     fs->fs_cstotal.cs_nffree >
  346                     (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100))
  347                         break;
  348                 log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n",
  349                         fs->fs_fsmnt);
  350                 fs->fs_optim = FS_OPTTIME;
  351                 break;
  352         case FS_OPTTIME:
  353                 /*
  354                  * At this point we have discovered a file that is trying to
  355                  * grow a small fragment to a larger fragment. To save time,
  356                  * we allocate a full sized block, then free the unused portion.
  357                  * If the file continues to grow, the `ffs_fragextend' call
  358                  * above will be able to grow it in place without further
  359                  * copying. If aberrant programs cause disk fragmentation to
  360                  * grow within 2% of the free reserve, we choose to begin
  361                  * optimizing for space.
  362                  */
  363                 request = fs->fs_bsize;
  364                 if (fs->fs_cstotal.cs_nffree <
  365                     (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100)
  366                         break;
  367                 log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n",
  368                         fs->fs_fsmnt);
  369                 fs->fs_optim = FS_OPTSPACE;
  370                 break;
  371         default:
  372                 printf("dev = %s, optim = %ld, fs = %s\n",
  373                     devtoname(ip->i_dev), (long)fs->fs_optim, fs->fs_fsmnt);
  374                 panic("ffs_realloccg: bad optim");
  375                 /* NOTREACHED */
  376         }
  377         bno = ffs_hashalloc(ip, cg, bpref, request, ffs_alloccg);
  378         if (bno > 0) {
  379                 bp->b_blkno = fsbtodb(fs, bno);
  380                 if (!DOINGSOFTDEP(vp))
  381                         ffs_blkfree(ump, fs, ip->i_devvp, bprev, (long)osize,
  382                             ip->i_number);
  383                 if (nsize < request)
  384                         ffs_blkfree(ump, fs, ip->i_devvp,
  385                             bno + numfrags(fs, nsize),
  386                             (long)(request - nsize), ip->i_number);
  387                 delta = btodb(nsize - osize);
  388                 if (ip->i_flag & IN_SPACECOUNTED) {
  389                         UFS_LOCK(ump);
  390                         fs->fs_pendingblocks += delta;
  391                         UFS_UNLOCK(ump);
  392                 }
  393                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta);
  394                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  395                 allocbuf(bp, nsize);
  396                 bp->b_flags |= B_DONE;
  397                 if ((bp->b_flags & (B_MALLOC | B_VMIO)) != B_VMIO)
  398                         bzero((char *)bp->b_data + osize, nsize - osize);
  399                 else
  400                         vfs_bio_clrbuf(bp);
  401                 *bpp = bp;
  402                 return (0);
  403         }
  404 #ifdef QUOTA
  405         UFS_UNLOCK(ump);
  406         /*
  407          * Restore user's disk quota because allocation failed.
  408          */
  409         (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
  410         UFS_LOCK(ump);
  411 #endif
  412 nospace:
  413         /*
  414          * no space available
  415          */
  416         if (fs->fs_pendingblocks > 0 && reclaimed == 0) {
  417                 reclaimed = 1;
  418                 softdep_request_cleanup(fs, vp);
  419                 UFS_UNLOCK(ump);
  420                 if (bp)
  421                         brelse(bp);
  422                 UFS_LOCK(ump);
  423                 goto retry;
  424         }
  425         UFS_UNLOCK(ump);
  426         if (bp)
  427                 brelse(bp);
  428         if (ppsratecheck(&lastfail, &curfail, 1)) {
  429                 ffs_fserr(fs, ip->i_number, "filesystem full");
  430                 uprintf("\n%s: write failed, filesystem is full\n",
  431                     fs->fs_fsmnt);
  432         }
  433         return (ENOSPC);
  434 }
  435 
  436 /*
  437  * Reallocate a sequence of blocks into a contiguous sequence of blocks.
  438  *
  439  * The vnode and an array of buffer pointers for a range of sequential
  440  * logical blocks to be made contiguous is given. The allocator attempts
  441  * to find a range of sequential blocks starting as close as possible
  442  * from the end of the allocation for the logical block immediately
  443  * preceding the current range. If successful, the physical block numbers
  444  * in the buffer pointers and in the inode are changed to reflect the new
  445  * allocation. If unsuccessful, the allocation is left unchanged. The
  446  * success in doing the reallocation is returned. Note that the error
  447  * return is not reflected back to the user. Rather the previous block
  448  * allocation will be used.
  449  */
  450 
  451 SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW, 0, "FFS filesystem");
  452 
  453 static int doasyncfree = 1;
  454 SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "");
  455 
  456 static int doreallocblks = 1;
  457 SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "");
  458 
  459 #ifdef DEBUG
  460 static volatile int prtrealloc = 0;
  461 #endif
  462 
  463 int
  464 ffs_reallocblks(ap)
  465         struct vop_reallocblks_args /* {
  466                 struct vnode *a_vp;
  467                 struct cluster_save *a_buflist;
  468         } */ *ap;
  469 {
  470 
  471         if (doreallocblks == 0)
  472                 return (ENOSPC);
  473         if (VTOI(ap->a_vp)->i_ump->um_fstype == UFS1)
  474                 return (ffs_reallocblks_ufs1(ap));
  475         return (ffs_reallocblks_ufs2(ap));
  476 }
  477         
  478 static int
  479 ffs_reallocblks_ufs1(ap)
  480         struct vop_reallocblks_args /* {
  481                 struct vnode *a_vp;
  482                 struct cluster_save *a_buflist;
  483         } */ *ap;
  484 {
  485         struct fs *fs;
  486         struct inode *ip;
  487         struct vnode *vp;
  488         struct buf *sbp, *ebp;
  489         ufs1_daddr_t *bap, *sbap, *ebap = 0;
  490         struct cluster_save *buflist;
  491         struct ufsmount *ump;
  492         ufs_lbn_t start_lbn, end_lbn;
  493         ufs1_daddr_t soff, newblk, blkno;
  494         ufs2_daddr_t pref;
  495         struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
  496         int i, len, start_lvl, end_lvl, ssize;
  497 
  498         vp = ap->a_vp;
  499         ip = VTOI(vp);
  500         fs = ip->i_fs;
  501         ump = ip->i_ump;
  502         if (fs->fs_contigsumsize <= 0)
  503                 return (ENOSPC);
  504         buflist = ap->a_buflist;
  505         len = buflist->bs_nchildren;
  506         start_lbn = buflist->bs_children[0]->b_lblkno;
  507         end_lbn = start_lbn + len - 1;
  508 #ifdef INVARIANTS
  509         for (i = 0; i < len; i++)
  510                 if (!ffs_checkblk(ip,
  511                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  512                         panic("ffs_reallocblks: unallocated block 1");
  513         for (i = 1; i < len; i++)
  514                 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
  515                         panic("ffs_reallocblks: non-logical cluster");
  516         blkno = buflist->bs_children[0]->b_blkno;
  517         ssize = fsbtodb(fs, fs->fs_frag);
  518         for (i = 1; i < len - 1; i++)
  519                 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
  520                         panic("ffs_reallocblks: non-physical cluster %d", i);
  521 #endif
  522         /*
  523          * If the latest allocation is in a new cylinder group, assume that
  524          * the filesystem has decided to move and do not force it back to
  525          * the previous cylinder group.
  526          */
  527         if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
  528             dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
  529                 return (ENOSPC);
  530         if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
  531             ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
  532                 return (ENOSPC);
  533         /*
  534          * Get the starting offset and block map for the first block.
  535          */
  536         if (start_lvl == 0) {
  537                 sbap = &ip->i_din1->di_db[0];
  538                 soff = start_lbn;
  539         } else {
  540                 idp = &start_ap[start_lvl - 1];
  541                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
  542                         brelse(sbp);
  543                         return (ENOSPC);
  544                 }
  545                 sbap = (ufs1_daddr_t *)sbp->b_data;
  546                 soff = idp->in_off;
  547         }
  548         /*
  549          * If the block range spans two block maps, get the second map.
  550          */
  551         if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
  552                 ssize = len;
  553         } else {
  554 #ifdef INVARIANTS
  555                 if (start_lvl > 0 &&
  556                     start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
  557                         panic("ffs_reallocblk: start == end");
  558 #endif
  559                 ssize = len - (idp->in_off + 1);
  560                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
  561                         goto fail;
  562                 ebap = (ufs1_daddr_t *)ebp->b_data;
  563         }
  564         /*
  565          * Find the preferred location for the cluster.
  566          */
  567         UFS_LOCK(ump);
  568         pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap);
  569         /*
  570          * Search the block map looking for an allocation of the desired size.
  571          */
  572         if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
  573             len, ffs_clusteralloc)) == 0) {
  574                 UFS_UNLOCK(ump);
  575                 goto fail;
  576         }
  577         /*
  578          * We have found a new contiguous block.
  579          *
  580          * First we have to replace the old block pointers with the new
  581          * block pointers in the inode and indirect blocks associated
  582          * with the file.
  583          */
  584 #ifdef DEBUG
  585         if (prtrealloc)
  586                 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
  587                     (intmax_t)start_lbn, (intmax_t)end_lbn);
  588 #endif
  589         blkno = newblk;
  590         for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
  591                 if (i == ssize) {
  592                         bap = ebap;
  593                         soff = -i;
  594                 }
  595 #ifdef INVARIANTS
  596                 if (!ffs_checkblk(ip,
  597                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  598                         panic("ffs_reallocblks: unallocated block 2");
  599                 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
  600                         panic("ffs_reallocblks: alloc mismatch");
  601 #endif
  602 #ifdef DEBUG
  603                 if (prtrealloc)
  604                         printf(" %d,", *bap);
  605 #endif
  606                 if (DOINGSOFTDEP(vp)) {
  607                         if (sbap == &ip->i_din1->di_db[0] && i < ssize)
  608                                 softdep_setup_allocdirect(ip, start_lbn + i,
  609                                     blkno, *bap, fs->fs_bsize, fs->fs_bsize,
  610                                     buflist->bs_children[i]);
  611                         else
  612                                 softdep_setup_allocindir_page(ip, start_lbn + i,
  613                                     i < ssize ? sbp : ebp, soff + i, blkno,
  614                                     *bap, buflist->bs_children[i]);
  615                 }
  616                 *bap++ = blkno;
  617         }
  618         /*
  619          * Next we must write out the modified inode and indirect blocks.
  620          * For strict correctness, the writes should be synchronous since
  621          * the old block values may have been written to disk. In practise
  622          * they are almost never written, but if we are concerned about
  623          * strict correctness, the `doasyncfree' flag should be set to zero.
  624          *
  625          * The test on `doasyncfree' should be changed to test a flag
  626          * that shows whether the associated buffers and inodes have
  627          * been written. The flag should be set when the cluster is
  628          * started and cleared whenever the buffer or inode is flushed.
  629          * We can then check below to see if it is set, and do the
  630          * synchronous write only when it has been cleared.
  631          */
  632         if (sbap != &ip->i_din1->di_db[0]) {
  633                 if (doasyncfree)
  634                         bdwrite(sbp);
  635                 else
  636                         bwrite(sbp);
  637         } else {
  638                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  639                 if (!doasyncfree)
  640                         ffs_update(vp, 1);
  641         }
  642         if (ssize < len) {
  643                 if (doasyncfree)
  644                         bdwrite(ebp);
  645                 else
  646                         bwrite(ebp);
  647         }
  648         /*
  649          * Last, free the old blocks and assign the new blocks to the buffers.
  650          */
  651 #ifdef DEBUG
  652         if (prtrealloc)
  653                 printf("\n\tnew:");
  654 #endif
  655         for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
  656                 if (!DOINGSOFTDEP(vp))
  657                         ffs_blkfree(ump, fs, ip->i_devvp,
  658                             dbtofsb(fs, buflist->bs_children[i]->b_blkno),
  659                             fs->fs_bsize, ip->i_number);
  660                 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
  661 #ifdef INVARIANTS
  662                 if (!ffs_checkblk(ip,
  663                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  664                         panic("ffs_reallocblks: unallocated block 3");
  665 #endif
  666 #ifdef DEBUG
  667                 if (prtrealloc)
  668                         printf(" %d,", blkno);
  669 #endif
  670         }
  671 #ifdef DEBUG
  672         if (prtrealloc) {
  673                 prtrealloc--;
  674                 printf("\n");
  675         }
  676 #endif
  677         return (0);
  678 
  679 fail:
  680         if (ssize < len)
  681                 brelse(ebp);
  682         if (sbap != &ip->i_din1->di_db[0])
  683                 brelse(sbp);
  684         return (ENOSPC);
  685 }
  686 
  687 static int
  688 ffs_reallocblks_ufs2(ap)
  689         struct vop_reallocblks_args /* {
  690                 struct vnode *a_vp;
  691                 struct cluster_save *a_buflist;
  692         } */ *ap;
  693 {
  694         struct fs *fs;
  695         struct inode *ip;
  696         struct vnode *vp;
  697         struct buf *sbp, *ebp;
  698         ufs2_daddr_t *bap, *sbap, *ebap = 0;
  699         struct cluster_save *buflist;
  700         struct ufsmount *ump;
  701         ufs_lbn_t start_lbn, end_lbn;
  702         ufs2_daddr_t soff, newblk, blkno, pref;
  703         struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp;
  704         int i, len, start_lvl, end_lvl, ssize;
  705 
  706         vp = ap->a_vp;
  707         ip = VTOI(vp);
  708         fs = ip->i_fs;
  709         ump = ip->i_ump;
  710         if (fs->fs_contigsumsize <= 0)
  711                 return (ENOSPC);
  712         buflist = ap->a_buflist;
  713         len = buflist->bs_nchildren;
  714         start_lbn = buflist->bs_children[0]->b_lblkno;
  715         end_lbn = start_lbn + len - 1;
  716 #ifdef INVARIANTS
  717         for (i = 0; i < len; i++)
  718                 if (!ffs_checkblk(ip,
  719                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  720                         panic("ffs_reallocblks: unallocated block 1");
  721         for (i = 1; i < len; i++)
  722                 if (buflist->bs_children[i]->b_lblkno != start_lbn + i)
  723                         panic("ffs_reallocblks: non-logical cluster");
  724         blkno = buflist->bs_children[0]->b_blkno;
  725         ssize = fsbtodb(fs, fs->fs_frag);
  726         for (i = 1; i < len - 1; i++)
  727                 if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize))
  728                         panic("ffs_reallocblks: non-physical cluster %d", i);
  729 #endif
  730         /*
  731          * If the latest allocation is in a new cylinder group, assume that
  732          * the filesystem has decided to move and do not force it back to
  733          * the previous cylinder group.
  734          */
  735         if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) !=
  736             dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno)))
  737                 return (ENOSPC);
  738         if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) ||
  739             ufs_getlbns(vp, end_lbn, end_ap, &end_lvl))
  740                 return (ENOSPC);
  741         /*
  742          * Get the starting offset and block map for the first block.
  743          */
  744         if (start_lvl == 0) {
  745                 sbap = &ip->i_din2->di_db[0];
  746                 soff = start_lbn;
  747         } else {
  748                 idp = &start_ap[start_lvl - 1];
  749                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) {
  750                         brelse(sbp);
  751                         return (ENOSPC);
  752                 }
  753                 sbap = (ufs2_daddr_t *)sbp->b_data;
  754                 soff = idp->in_off;
  755         }
  756         /*
  757          * If the block range spans two block maps, get the second map.
  758          */
  759         if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) {
  760                 ssize = len;
  761         } else {
  762 #ifdef INVARIANTS
  763                 if (start_lvl > 0 &&
  764                     start_ap[start_lvl - 1].in_lbn == idp->in_lbn)
  765                         panic("ffs_reallocblk: start == end");
  766 #endif
  767                 ssize = len - (idp->in_off + 1);
  768                 if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp))
  769                         goto fail;
  770                 ebap = (ufs2_daddr_t *)ebp->b_data;
  771         }
  772         /*
  773          * Find the preferred location for the cluster.
  774          */
  775         UFS_LOCK(ump);
  776         pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap);
  777         /*
  778          * Search the block map looking for an allocation of the desired size.
  779          */
  780         if ((newblk = ffs_hashalloc(ip, dtog(fs, pref), pref,
  781             len, ffs_clusteralloc)) == 0) {
  782                 UFS_UNLOCK(ump);
  783                 goto fail;
  784         }
  785         /*
  786          * We have found a new contiguous block.
  787          *
  788          * First we have to replace the old block pointers with the new
  789          * block pointers in the inode and indirect blocks associated
  790          * with the file.
  791          */
  792 #ifdef DEBUG
  793         if (prtrealloc)
  794                 printf("realloc: ino %d, lbns %jd-%jd\n\told:", ip->i_number,
  795                     (intmax_t)start_lbn, (intmax_t)end_lbn);
  796 #endif
  797         blkno = newblk;
  798         for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) {
  799                 if (i == ssize) {
  800                         bap = ebap;
  801                         soff = -i;
  802                 }
  803 #ifdef INVARIANTS
  804                 if (!ffs_checkblk(ip,
  805                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  806                         panic("ffs_reallocblks: unallocated block 2");
  807                 if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap)
  808                         panic("ffs_reallocblks: alloc mismatch");
  809 #endif
  810 #ifdef DEBUG
  811                 if (prtrealloc)
  812                         printf(" %jd,", (intmax_t)*bap);
  813 #endif
  814                 if (DOINGSOFTDEP(vp)) {
  815                         if (sbap == &ip->i_din2->di_db[0] && i < ssize)
  816                                 softdep_setup_allocdirect(ip, start_lbn + i,
  817                                     blkno, *bap, fs->fs_bsize, fs->fs_bsize,
  818                                     buflist->bs_children[i]);
  819                         else
  820                                 softdep_setup_allocindir_page(ip, start_lbn + i,
  821                                     i < ssize ? sbp : ebp, soff + i, blkno,
  822                                     *bap, buflist->bs_children[i]);
  823                 }
  824                 *bap++ = blkno;
  825         }
  826         /*
  827          * Next we must write out the modified inode and indirect blocks.
  828          * For strict correctness, the writes should be synchronous since
  829          * the old block values may have been written to disk. In practise
  830          * they are almost never written, but if we are concerned about
  831          * strict correctness, the `doasyncfree' flag should be set to zero.
  832          *
  833          * The test on `doasyncfree' should be changed to test a flag
  834          * that shows whether the associated buffers and inodes have
  835          * been written. The flag should be set when the cluster is
  836          * started and cleared whenever the buffer or inode is flushed.
  837          * We can then check below to see if it is set, and do the
  838          * synchronous write only when it has been cleared.
  839          */
  840         if (sbap != &ip->i_din2->di_db[0]) {
  841                 if (doasyncfree)
  842                         bdwrite(sbp);
  843                 else
  844                         bwrite(sbp);
  845         } else {
  846                 ip->i_flag |= IN_CHANGE | IN_UPDATE;
  847                 if (!doasyncfree)
  848                         ffs_update(vp, 1);
  849         }
  850         if (ssize < len) {
  851                 if (doasyncfree)
  852                         bdwrite(ebp);
  853                 else
  854                         bwrite(ebp);
  855         }
  856         /*
  857          * Last, free the old blocks and assign the new blocks to the buffers.
  858          */
  859 #ifdef DEBUG
  860         if (prtrealloc)
  861                 printf("\n\tnew:");
  862 #endif
  863         for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) {
  864                 if (!DOINGSOFTDEP(vp))
  865                         ffs_blkfree(ump, fs, ip->i_devvp,
  866                             dbtofsb(fs, buflist->bs_children[i]->b_blkno),
  867                             fs->fs_bsize, ip->i_number);
  868                 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno);
  869 #ifdef INVARIANTS
  870                 if (!ffs_checkblk(ip,
  871                    dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize))
  872                         panic("ffs_reallocblks: unallocated block 3");
  873 #endif
  874 #ifdef DEBUG
  875                 if (prtrealloc)
  876                         printf(" %jd,", (intmax_t)blkno);
  877 #endif
  878         }
  879 #ifdef DEBUG
  880         if (prtrealloc) {
  881                 prtrealloc--;
  882                 printf("\n");
  883         }
  884 #endif
  885         return (0);
  886 
  887 fail:
  888         if (ssize < len)
  889                 brelse(ebp);
  890         if (sbap != &ip->i_din2->di_db[0])
  891                 brelse(sbp);
  892         return (ENOSPC);
  893 }
  894 
  895 /*
  896  * Allocate an inode in the filesystem.
  897  *
  898  * If allocating a directory, use ffs_dirpref to select the inode.
  899  * If allocating in a directory, the following hierarchy is followed:
  900  *   1) allocate the preferred inode.
  901  *   2) allocate an inode in the same cylinder group.
  902  *   3) quadradically rehash into other cylinder groups, until an
  903  *      available inode is located.
  904  * If no inode preference is given the following hierarchy is used
  905  * to allocate an inode:
  906  *   1) allocate an inode in cylinder group 0.
  907  *   2) quadradically rehash into other cylinder groups, until an
  908  *      available inode is located.
  909  */
  910 int
  911 ffs_valloc(pvp, mode, cred, vpp)
  912         struct vnode *pvp;
  913         int mode;
  914         struct ucred *cred;
  915         struct vnode **vpp;
  916 {
  917         struct inode *pip;
  918         struct fs *fs;
  919         struct inode *ip;
  920         struct timespec ts;
  921         struct ufsmount *ump;
  922         ino_t ino, ipref;
  923         int cg, error;
  924         static struct timeval lastfail;
  925         static int curfail;
  926 
  927         *vpp = NULL;
  928         pip = VTOI(pvp);
  929         fs = pip->i_fs;
  930         ump = pip->i_ump;
  931 
  932         UFS_LOCK(ump);
  933         if (fs->fs_cstotal.cs_nifree == 0)
  934                 goto noinodes;
  935 
  936         if ((mode & IFMT) == IFDIR)
  937                 ipref = ffs_dirpref(pip);
  938         else
  939                 ipref = pip->i_number;
  940         if (ipref >= fs->fs_ncg * fs->fs_ipg)
  941                 ipref = 0;
  942         cg = ino_to_cg(fs, ipref);
  943         /*
  944          * Track number of dirs created one after another
  945          * in a same cg without intervening by files.
  946          */
  947         if ((mode & IFMT) == IFDIR) {
  948                 if (fs->fs_contigdirs[cg] < 255)
  949                         fs->fs_contigdirs[cg]++;
  950         } else {
  951                 if (fs->fs_contigdirs[cg] > 0)
  952                         fs->fs_contigdirs[cg]--;
  953         }
  954         ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode,
  955                                         (allocfcn_t *)ffs_nodealloccg);
  956         if (ino == 0)
  957                 goto noinodes;
  958         error = ffs_vget(pvp->v_mount, ino, LK_EXCLUSIVE, vpp);
  959         if (error) {
  960                 ffs_vfree(pvp, ino, mode);
  961                 return (error);
  962         }
  963         ip = VTOI(*vpp);
  964         if (ip->i_mode) {
  965                 printf("mode = 0%o, inum = %lu, fs = %s\n",
  966                     ip->i_mode, (u_long)ip->i_number, fs->fs_fsmnt);
  967                 panic("ffs_valloc: dup alloc");
  968         }
  969         if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) {  /* XXX */
  970                 printf("free inode %s/%lu had %ld blocks\n",
  971                     fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks));
  972                 DIP_SET(ip, i_blocks, 0);
  973         }
  974         ip->i_flags = 0;
  975         DIP_SET(ip, i_flags, 0);
  976         /*
  977          * Set up a new generation number for this inode.
  978          */
  979         if (ip->i_gen == 0 || ++ip->i_gen == 0)
  980                 ip->i_gen = arc4random() / 2 + 1;
  981         DIP_SET(ip, i_gen, ip->i_gen);
  982         if (fs->fs_magic == FS_UFS2_MAGIC) {
  983                 vfs_timestamp(&ts);
  984                 ip->i_din2->di_birthtime = ts.tv_sec;
  985                 ip->i_din2->di_birthnsec = ts.tv_nsec;
  986         }
  987         ip->i_flag = 0;
  988         vnode_destroy_vobject(*vpp);
  989         (*vpp)->v_type = VNON;
  990         if (fs->fs_magic == FS_UFS2_MAGIC)
  991                 (*vpp)->v_op = &ffs_vnodeops2;
  992         else
  993                 (*vpp)->v_op = &ffs_vnodeops1;
  994         return (0);
  995 noinodes:
  996         UFS_UNLOCK(ump);
  997         if (ppsratecheck(&lastfail, &curfail, 1)) {
  998                 ffs_fserr(fs, pip->i_number, "out of inodes");
  999                 uprintf("\n%s: create/symlink failed, no inodes free\n",
 1000                     fs->fs_fsmnt);
 1001         }
 1002         return (ENOSPC);
 1003 }
 1004 
 1005 /*
 1006  * Find a cylinder group to place a directory.
 1007  *
 1008  * The policy implemented by this algorithm is to allocate a
 1009  * directory inode in the same cylinder group as its parent
 1010  * directory, but also to reserve space for its files inodes
 1011  * and data. Restrict the number of directories which may be
 1012  * allocated one after another in the same cylinder group
 1013  * without intervening allocation of files.
 1014  *
 1015  * If we allocate a first level directory then force allocation
 1016  * in another cylinder group.
 1017  */
 1018 static ino_t
 1019 ffs_dirpref(pip)
 1020         struct inode *pip;
 1021 {
 1022         struct fs *fs;
 1023         int cg, prefcg, dirsize, cgsize;
 1024         int avgifree, avgbfree, avgndir, curdirsize;
 1025         int minifree, minbfree, maxndir;
 1026         int mincg, minndir;
 1027         int maxcontigdirs;
 1028 
 1029         mtx_assert(UFS_MTX(pip->i_ump), MA_OWNED);
 1030         fs = pip->i_fs;
 1031 
 1032         avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
 1033         avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1034         avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
 1035 
 1036         /*
 1037          * Force allocation in another cg if creating a first level dir.
 1038          */
 1039         ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref");
 1040         if (ITOV(pip)->v_vflag & VV_ROOT) {
 1041                 prefcg = arc4random() % fs->fs_ncg;
 1042                 mincg = prefcg;
 1043                 minndir = fs->fs_ipg;
 1044                 for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1045                         if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
 1046                             fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
 1047                             fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1048                                 mincg = cg;
 1049                                 minndir = fs->fs_cs(fs, cg).cs_ndir;
 1050                         }
 1051                 for (cg = 0; cg < prefcg; cg++)
 1052                         if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
 1053                             fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
 1054                             fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1055                                 mincg = cg;
 1056                                 minndir = fs->fs_cs(fs, cg).cs_ndir;
 1057                         }
 1058                 return ((ino_t)(fs->fs_ipg * mincg));
 1059         }
 1060 
 1061         /*
 1062          * Count various limits which used for
 1063          * optimal allocation of a directory inode.
 1064          */
 1065         maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
 1066         minifree = avgifree - avgifree / 4;
 1067         if (minifree < 1)
 1068                 minifree = 1;
 1069         minbfree = avgbfree - avgbfree / 4;
 1070         if (minbfree < 1)
 1071                 minbfree = 1;
 1072         cgsize = fs->fs_fsize * fs->fs_fpg;
 1073         dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir;
 1074         curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0;
 1075         if (dirsize < curdirsize)
 1076                 dirsize = curdirsize;
 1077         if (dirsize <= 0)
 1078                 maxcontigdirs = 0;              /* dirsize overflowed */
 1079         else
 1080                 maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255);
 1081         if (fs->fs_avgfpdir > 0)
 1082                 maxcontigdirs = min(maxcontigdirs,
 1083                                     fs->fs_ipg / fs->fs_avgfpdir);
 1084         if (maxcontigdirs == 0)
 1085                 maxcontigdirs = 1;
 1086 
 1087         /*
 1088          * Limit number of dirs in one cg and reserve space for 
 1089          * regular files, but only if we have no deficit in
 1090          * inodes or space.
 1091          */
 1092         prefcg = ino_to_cg(fs, pip->i_number);
 1093         for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1094                 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
 1095                     fs->fs_cs(fs, cg).cs_nifree >= minifree &&
 1096                     fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
 1097                         if (fs->fs_contigdirs[cg] < maxcontigdirs)
 1098                                 return ((ino_t)(fs->fs_ipg * cg));
 1099                 }
 1100         for (cg = 0; cg < prefcg; cg++)
 1101                 if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
 1102                     fs->fs_cs(fs, cg).cs_nifree >= minifree &&
 1103                     fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
 1104                         if (fs->fs_contigdirs[cg] < maxcontigdirs)
 1105                                 return ((ino_t)(fs->fs_ipg * cg));
 1106                 }
 1107         /*
 1108          * This is a backstop when we have deficit in space.
 1109          */
 1110         for (cg = prefcg; cg < fs->fs_ncg; cg++)
 1111                 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
 1112                         return ((ino_t)(fs->fs_ipg * cg));
 1113         for (cg = 0; cg < prefcg; cg++)
 1114                 if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
 1115                         break;
 1116         return ((ino_t)(fs->fs_ipg * cg));
 1117 }
 1118 
 1119 /*
 1120  * Select the desired position for the next block in a file.  The file is
 1121  * logically divided into sections. The first section is composed of the
 1122  * direct blocks. Each additional section contains fs_maxbpg blocks.
 1123  *
 1124  * If no blocks have been allocated in the first section, the policy is to
 1125  * request a block in the same cylinder group as the inode that describes
 1126  * the file. If no blocks have been allocated in any other section, the
 1127  * policy is to place the section in a cylinder group with a greater than
 1128  * average number of free blocks.  An appropriate cylinder group is found
 1129  * by using a rotor that sweeps the cylinder groups. When a new group of
 1130  * blocks is needed, the sweep begins in the cylinder group following the
 1131  * cylinder group from which the previous allocation was made. The sweep
 1132  * continues until a cylinder group with greater than the average number
 1133  * of free blocks is found. If the allocation is for the first block in an
 1134  * indirect block, the information on the previous allocation is unavailable;
 1135  * here a best guess is made based upon the logical block number being
 1136  * allocated.
 1137  *
 1138  * If a section is already partially allocated, the policy is to
 1139  * contiguously allocate fs_maxcontig blocks. The end of one of these
 1140  * contiguous blocks and the beginning of the next is laid out
 1141  * contiguously if possible.
 1142  */
 1143 ufs2_daddr_t
 1144 ffs_blkpref_ufs1(ip, lbn, indx, bap)
 1145         struct inode *ip;
 1146         ufs_lbn_t lbn;
 1147         int indx;
 1148         ufs1_daddr_t *bap;
 1149 {
 1150         struct fs *fs;
 1151         int cg;
 1152         int avgbfree, startcg;
 1153 
 1154         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1155         fs = ip->i_fs;
 1156         if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
 1157                 if (lbn < NDADDR + NINDIR(fs)) {
 1158                         cg = ino_to_cg(fs, ip->i_number);
 1159                         return (cgbase(fs, cg) + fs->fs_frag);
 1160                 }
 1161                 /*
 1162                  * Find a cylinder with greater than average number of
 1163                  * unused data blocks.
 1164                  */
 1165                 if (indx == 0 || bap[indx - 1] == 0)
 1166                         startcg =
 1167                             ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
 1168                 else
 1169                         startcg = dtog(fs, bap[indx - 1]) + 1;
 1170                 startcg %= fs->fs_ncg;
 1171                 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1172                 for (cg = startcg; cg < fs->fs_ncg; cg++)
 1173                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1174                                 fs->fs_cgrotor = cg;
 1175                                 return (cgbase(fs, cg) + fs->fs_frag);
 1176                         }
 1177                 for (cg = 0; cg <= startcg; cg++)
 1178                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1179                                 fs->fs_cgrotor = cg;
 1180                                 return (cgbase(fs, cg) + fs->fs_frag);
 1181                         }
 1182                 return (0);
 1183         }
 1184         /*
 1185          * We just always try to lay things out contiguously.
 1186          */
 1187         return (bap[indx - 1] + fs->fs_frag);
 1188 }
 1189 
 1190 /*
 1191  * Same as above, but for UFS2
 1192  */
 1193 ufs2_daddr_t
 1194 ffs_blkpref_ufs2(ip, lbn, indx, bap)
 1195         struct inode *ip;
 1196         ufs_lbn_t lbn;
 1197         int indx;
 1198         ufs2_daddr_t *bap;
 1199 {
 1200         struct fs *fs;
 1201         int cg;
 1202         int avgbfree, startcg;
 1203 
 1204         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1205         fs = ip->i_fs;
 1206         if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
 1207                 if (lbn < NDADDR + NINDIR(fs)) {
 1208                         cg = ino_to_cg(fs, ip->i_number);
 1209                         return (cgbase(fs, cg) + fs->fs_frag);
 1210                 }
 1211                 /*
 1212                  * Find a cylinder with greater than average number of
 1213                  * unused data blocks.
 1214                  */
 1215                 if (indx == 0 || bap[indx - 1] == 0)
 1216                         startcg =
 1217                             ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
 1218                 else
 1219                         startcg = dtog(fs, bap[indx - 1]) + 1;
 1220                 startcg %= fs->fs_ncg;
 1221                 avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
 1222                 for (cg = startcg; cg < fs->fs_ncg; cg++)
 1223                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1224                                 fs->fs_cgrotor = cg;
 1225                                 return (cgbase(fs, cg) + fs->fs_frag);
 1226                         }
 1227                 for (cg = 0; cg <= startcg; cg++)
 1228                         if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
 1229                                 fs->fs_cgrotor = cg;
 1230                                 return (cgbase(fs, cg) + fs->fs_frag);
 1231                         }
 1232                 return (0);
 1233         }
 1234         /*
 1235          * We just always try to lay things out contiguously.
 1236          */
 1237         return (bap[indx - 1] + fs->fs_frag);
 1238 }
 1239 
 1240 /*
 1241  * Implement the cylinder overflow algorithm.
 1242  *
 1243  * The policy implemented by this algorithm is:
 1244  *   1) allocate the block in its requested cylinder group.
 1245  *   2) quadradically rehash on the cylinder group number.
 1246  *   3) brute force search for a free block.
 1247  *
 1248  * Must be called with the UFS lock held.  Will release the lock on success
 1249  * and return with it held on failure.
 1250  */
 1251 /*VARARGS5*/
 1252 static ufs2_daddr_t
 1253 ffs_hashalloc(ip, cg, pref, size, allocator)
 1254         struct inode *ip;
 1255         int cg;
 1256         ufs2_daddr_t pref;
 1257         int size;       /* size for data blocks, mode for inodes */
 1258         allocfcn_t *allocator;
 1259 {
 1260         struct fs *fs;
 1261         ufs2_daddr_t result;
 1262         int i, icg = cg;
 1263 
 1264         mtx_assert(UFS_MTX(ip->i_ump), MA_OWNED);
 1265 #ifdef INVARIANTS
 1266         if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED)
 1267                 panic("ffs_hashalloc: allocation on suspended filesystem");
 1268 #endif
 1269         fs = ip->i_fs;
 1270         /*
 1271          * 1: preferred cylinder group
 1272          */
 1273         result = (*allocator)(ip, cg, pref, size);
 1274         if (result)
 1275                 return (result);
 1276         /*
 1277          * 2: quadratic rehash
 1278          */
 1279         for (i = 1; i < fs->fs_ncg; i *= 2) {
 1280                 cg += i;
 1281                 if (cg >= fs->fs_ncg)
 1282                         cg -= fs->fs_ncg;
 1283                 result = (*allocator)(ip, cg, 0, size);
 1284                 if (result)
 1285                         return (result);
 1286         }
 1287         /*
 1288          * 3: brute force search
 1289          * Note that we start at i == 2, since 0 was checked initially,
 1290          * and 1 is always checked in the quadratic rehash.
 1291          */
 1292         cg = (icg + 2) % fs->fs_ncg;
 1293         for (i = 2; i < fs->fs_ncg; i++) {
 1294                 result = (*allocator)(ip, cg, 0, size);
 1295                 if (result)
 1296                         return (result);
 1297                 cg++;
 1298                 if (cg == fs->fs_ncg)
 1299                         cg = 0;
 1300         }
 1301         return (0);
 1302 }
 1303 
 1304 /*
 1305  * Determine whether a fragment can be extended.
 1306  *
 1307  * Check to see if the necessary fragments are available, and
 1308  * if they are, allocate them.
 1309  */
 1310 static ufs2_daddr_t
 1311 ffs_fragextend(ip, cg, bprev, osize, nsize)
 1312         struct inode *ip;
 1313         int cg;
 1314         ufs2_daddr_t bprev;
 1315         int osize, nsize;
 1316 {
 1317         struct fs *fs;
 1318         struct cg *cgp;
 1319         struct buf *bp;
 1320         struct ufsmount *ump;
 1321         int nffree;
 1322         long bno;
 1323         int frags, bbase;
 1324         int i, error;
 1325         u_int8_t *blksfree;
 1326 
 1327         ump = ip->i_ump;
 1328         fs = ip->i_fs;
 1329         if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize))
 1330                 return (0);
 1331         frags = numfrags(fs, nsize);
 1332         bbase = fragnum(fs, bprev);
 1333         if (bbase > fragnum(fs, (bprev + frags - 1))) {
 1334                 /* cannot extend across a block boundary */
 1335                 return (0);
 1336         }
 1337         UFS_UNLOCK(ump);
 1338         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1339                 (int)fs->fs_cgsize, NOCRED, &bp);
 1340         if (error)
 1341                 goto fail;
 1342         cgp = (struct cg *)bp->b_data;
 1343         if (!cg_chkmagic(cgp))
 1344                 goto fail;
 1345         bp->b_xflags |= BX_BKGRDWRITE;
 1346         cgp->cg_old_time = cgp->cg_time = time_second;
 1347         bno = dtogd(fs, bprev);
 1348         blksfree = cg_blksfree(cgp);
 1349         for (i = numfrags(fs, osize); i < frags; i++)
 1350                 if (isclr(blksfree, bno + i))
 1351                         goto fail;
 1352         /*
 1353          * the current fragment can be extended
 1354          * deduct the count on fragment being extended into
 1355          * increase the count on the remaining fragment (if any)
 1356          * allocate the extended piece
 1357          */
 1358         for (i = frags; i < fs->fs_frag - bbase; i++)
 1359                 if (isclr(blksfree, bno + i))
 1360                         break;
 1361         cgp->cg_frsum[i - numfrags(fs, osize)]--;
 1362         if (i != frags)
 1363                 cgp->cg_frsum[i - frags]++;
 1364         for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) {
 1365                 clrbit(blksfree, bno + i);
 1366                 cgp->cg_cs.cs_nffree--;
 1367                 nffree++;
 1368         }
 1369         UFS_LOCK(ump);
 1370         fs->fs_cstotal.cs_nffree -= nffree;
 1371         fs->fs_cs(fs, cg).cs_nffree -= nffree;
 1372         fs->fs_fmod = 1;
 1373         ACTIVECLEAR(fs, cg);
 1374         UFS_UNLOCK(ump);
 1375         if (DOINGSOFTDEP(ITOV(ip)))
 1376                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev);
 1377         bdwrite(bp);
 1378         return (bprev);
 1379 
 1380 fail:
 1381         brelse(bp);
 1382         UFS_LOCK(ump);
 1383         return (0);
 1384 
 1385 }
 1386 
 1387 /*
 1388  * Determine whether a block can be allocated.
 1389  *
 1390  * Check to see if a block of the appropriate size is available,
 1391  * and if it is, allocate it.
 1392  */
 1393 static ufs2_daddr_t
 1394 ffs_alloccg(ip, cg, bpref, size)
 1395         struct inode *ip;
 1396         int cg;
 1397         ufs2_daddr_t bpref;
 1398         int size;
 1399 {
 1400         struct fs *fs;
 1401         struct cg *cgp;
 1402         struct buf *bp;
 1403         struct ufsmount *ump;
 1404         ufs1_daddr_t bno;
 1405         ufs2_daddr_t blkno;
 1406         int i, allocsiz, error, frags;
 1407         u_int8_t *blksfree;
 1408 
 1409         ump = ip->i_ump;
 1410         fs = ip->i_fs;
 1411         if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
 1412                 return (0);
 1413         UFS_UNLOCK(ump);
 1414         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1415                 (int)fs->fs_cgsize, NOCRED, &bp);
 1416         if (error)
 1417                 goto fail;
 1418         cgp = (struct cg *)bp->b_data;
 1419         if (!cg_chkmagic(cgp) ||
 1420             (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize))
 1421                 goto fail;
 1422         bp->b_xflags |= BX_BKGRDWRITE;
 1423         cgp->cg_old_time = cgp->cg_time = time_second;
 1424         if (size == fs->fs_bsize) {
 1425                 UFS_LOCK(ump);
 1426                 blkno = ffs_alloccgblk(ip, bp, bpref);
 1427                 ACTIVECLEAR(fs, cg);
 1428                 UFS_UNLOCK(ump);
 1429                 bdwrite(bp);
 1430                 return (blkno);
 1431         }
 1432         /*
 1433          * check to see if any fragments are already available
 1434          * allocsiz is the size which will be allocated, hacking
 1435          * it down to a smaller size if necessary
 1436          */
 1437         blksfree = cg_blksfree(cgp);
 1438         frags = numfrags(fs, size);
 1439         for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
 1440                 if (cgp->cg_frsum[allocsiz] != 0)
 1441                         break;
 1442         if (allocsiz == fs->fs_frag) {
 1443                 /*
 1444                  * no fragments were available, so a block will be
 1445                  * allocated, and hacked up
 1446                  */
 1447                 if (cgp->cg_cs.cs_nbfree == 0)
 1448                         goto fail;
 1449                 UFS_LOCK(ump);
 1450                 blkno = ffs_alloccgblk(ip, bp, bpref);
 1451                 bno = dtogd(fs, blkno);
 1452                 for (i = frags; i < fs->fs_frag; i++)
 1453                         setbit(blksfree, bno + i);
 1454                 i = fs->fs_frag - frags;
 1455                 cgp->cg_cs.cs_nffree += i;
 1456                 fs->fs_cstotal.cs_nffree += i;
 1457                 fs->fs_cs(fs, cg).cs_nffree += i;
 1458                 fs->fs_fmod = 1;
 1459                 cgp->cg_frsum[i]++;
 1460                 ACTIVECLEAR(fs, cg);
 1461                 UFS_UNLOCK(ump);
 1462                 bdwrite(bp);
 1463                 return (blkno);
 1464         }
 1465         bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
 1466         if (bno < 0)
 1467                 goto fail;
 1468         for (i = 0; i < frags; i++)
 1469                 clrbit(blksfree, bno + i);
 1470         cgp->cg_cs.cs_nffree -= frags;
 1471         cgp->cg_frsum[allocsiz]--;
 1472         if (frags != allocsiz)
 1473                 cgp->cg_frsum[allocsiz - frags]++;
 1474         UFS_LOCK(ump);
 1475         fs->fs_cstotal.cs_nffree -= frags;
 1476         fs->fs_cs(fs, cg).cs_nffree -= frags;
 1477         fs->fs_fmod = 1;
 1478         blkno = cgbase(fs, cg) + bno;
 1479         ACTIVECLEAR(fs, cg);
 1480         UFS_UNLOCK(ump);
 1481         if (DOINGSOFTDEP(ITOV(ip)))
 1482                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno);
 1483         bdwrite(bp);
 1484         return (blkno);
 1485 
 1486 fail:
 1487         brelse(bp);
 1488         UFS_LOCK(ump);
 1489         return (0);
 1490 }
 1491 
 1492 /*
 1493  * Allocate a block in a cylinder group.
 1494  *
 1495  * This algorithm implements the following policy:
 1496  *   1) allocate the requested block.
 1497  *   2) allocate a rotationally optimal block in the same cylinder.
 1498  *   3) allocate the next available block on the block rotor for the
 1499  *      specified cylinder group.
 1500  * Note that this routine only allocates fs_bsize blocks; these
 1501  * blocks may be fragmented by the routine that allocates them.
 1502  */
 1503 static ufs2_daddr_t
 1504 ffs_alloccgblk(ip, bp, bpref)
 1505         struct inode *ip;
 1506         struct buf *bp;
 1507         ufs2_daddr_t bpref;
 1508 {
 1509         struct fs *fs;
 1510         struct cg *cgp;
 1511         struct ufsmount *ump;
 1512         ufs1_daddr_t bno;
 1513         ufs2_daddr_t blkno;
 1514         u_int8_t *blksfree;
 1515 
 1516         fs = ip->i_fs;
 1517         ump = ip->i_ump;
 1518         mtx_assert(UFS_MTX(ump), MA_OWNED);
 1519         cgp = (struct cg *)bp->b_data;
 1520         blksfree = cg_blksfree(cgp);
 1521         if (bpref == 0 || dtog(fs, bpref) != cgp->cg_cgx) {
 1522                 bpref = cgp->cg_rotor;
 1523         } else {
 1524                 bpref = blknum(fs, bpref);
 1525                 bno = dtogd(fs, bpref);
 1526                 /*
 1527                  * if the requested block is available, use it
 1528                  */
 1529                 if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno)))
 1530                         goto gotit;
 1531         }
 1532         /*
 1533          * Take the next available block in this cylinder group.
 1534          */
 1535         bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
 1536         if (bno < 0)
 1537                 return (0);
 1538         cgp->cg_rotor = bno;
 1539 gotit:
 1540         blkno = fragstoblks(fs, bno);
 1541         ffs_clrblock(fs, blksfree, (long)blkno);
 1542         ffs_clusteracct(ump, fs, cgp, blkno, -1);
 1543         cgp->cg_cs.cs_nbfree--;
 1544         fs->fs_cstotal.cs_nbfree--;
 1545         fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--;
 1546         fs->fs_fmod = 1;
 1547         blkno = cgbase(fs, cgp->cg_cgx) + bno;
 1548         /* XXX Fixme. */
 1549         UFS_UNLOCK(ump);
 1550         if (DOINGSOFTDEP(ITOV(ip)))
 1551                 softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno);
 1552         UFS_LOCK(ump);
 1553         return (blkno);
 1554 }
 1555 
 1556 /*
 1557  * Determine whether a cluster can be allocated.
 1558  *
 1559  * We do not currently check for optimal rotational layout if there
 1560  * are multiple choices in the same cylinder group. Instead we just
 1561  * take the first one that we find following bpref.
 1562  */
 1563 static ufs2_daddr_t
 1564 ffs_clusteralloc(ip, cg, bpref, len)
 1565         struct inode *ip;
 1566         int cg;
 1567         ufs2_daddr_t bpref;
 1568         int len;
 1569 {
 1570         struct fs *fs;
 1571         struct cg *cgp;
 1572         struct buf *bp;
 1573         struct ufsmount *ump;
 1574         int i, run, bit, map, got;
 1575         ufs2_daddr_t bno;
 1576         u_char *mapp;
 1577         int32_t *lp;
 1578         u_int8_t *blksfree;
 1579 
 1580         fs = ip->i_fs;
 1581         ump = ip->i_ump;
 1582         if (fs->fs_maxcluster[cg] < len)
 1583                 return (0);
 1584         UFS_UNLOCK(ump);
 1585         if (bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)), (int)fs->fs_cgsize,
 1586             NOCRED, &bp))
 1587                 goto fail_lock;
 1588         cgp = (struct cg *)bp->b_data;
 1589         if (!cg_chkmagic(cgp))
 1590                 goto fail_lock;
 1591         bp->b_xflags |= BX_BKGRDWRITE;
 1592         /*
 1593          * Check to see if a cluster of the needed size (or bigger) is
 1594          * available in this cylinder group.
 1595          */
 1596         lp = &cg_clustersum(cgp)[len];
 1597         for (i = len; i <= fs->fs_contigsumsize; i++)
 1598                 if (*lp++ > 0)
 1599                         break;
 1600         if (i > fs->fs_contigsumsize) {
 1601                 /*
 1602                  * This is the first time looking for a cluster in this
 1603                  * cylinder group. Update the cluster summary information
 1604                  * to reflect the true maximum sized cluster so that
 1605                  * future cluster allocation requests can avoid reading
 1606                  * the cylinder group map only to find no clusters.
 1607                  */
 1608                 lp = &cg_clustersum(cgp)[len - 1];
 1609                 for (i = len - 1; i > 0; i--)
 1610                         if (*lp-- > 0)
 1611                                 break;
 1612                 UFS_LOCK(ump);
 1613                 fs->fs_maxcluster[cg] = i;
 1614                 goto fail;
 1615         }
 1616         /*
 1617          * Search the cluster map to find a big enough cluster.
 1618          * We take the first one that we find, even if it is larger
 1619          * than we need as we prefer to get one close to the previous
 1620          * block allocation. We do not search before the current
 1621          * preference point as we do not want to allocate a block
 1622          * that is allocated before the previous one (as we will
 1623          * then have to wait for another pass of the elevator
 1624          * algorithm before it will be read). We prefer to fail and
 1625          * be recalled to try an allocation in the next cylinder group.
 1626          */
 1627         if (dtog(fs, bpref) != cg)
 1628                 bpref = 0;
 1629         else
 1630                 bpref = fragstoblks(fs, dtogd(fs, blknum(fs, bpref)));
 1631         mapp = &cg_clustersfree(cgp)[bpref / NBBY];
 1632         map = *mapp++;
 1633         bit = 1 << (bpref % NBBY);
 1634         for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) {
 1635                 if ((map & bit) == 0) {
 1636                         run = 0;
 1637                 } else {
 1638                         run++;
 1639                         if (run == len)
 1640                                 break;
 1641                 }
 1642                 if ((got & (NBBY - 1)) != (NBBY - 1)) {
 1643                         bit <<= 1;
 1644                 } else {
 1645                         map = *mapp++;
 1646                         bit = 1;
 1647                 }
 1648         }
 1649         if (got >= cgp->cg_nclusterblks)
 1650                 goto fail_lock;
 1651         /*
 1652          * Allocate the cluster that we have found.
 1653          */
 1654         blksfree = cg_blksfree(cgp);
 1655         for (i = 1; i <= len; i++)
 1656                 if (!ffs_isblock(fs, blksfree, got - run + i))
 1657                         panic("ffs_clusteralloc: map mismatch");
 1658         bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1);
 1659         if (dtog(fs, bno) != cg)
 1660                 panic("ffs_clusteralloc: allocated out of group");
 1661         len = blkstofrags(fs, len);
 1662         UFS_LOCK(ump);
 1663         for (i = 0; i < len; i += fs->fs_frag)
 1664                 if (ffs_alloccgblk(ip, bp, bno + i) != bno + i)
 1665                         panic("ffs_clusteralloc: lost block");
 1666         ACTIVECLEAR(fs, cg);
 1667         UFS_UNLOCK(ump);
 1668         bdwrite(bp);
 1669         return (bno);
 1670 
 1671 fail_lock:
 1672         UFS_LOCK(ump);
 1673 fail:
 1674         brelse(bp);
 1675         return (0);
 1676 }
 1677 
 1678 /*
 1679  * Determine whether an inode can be allocated.
 1680  *
 1681  * Check to see if an inode is available, and if it is,
 1682  * allocate it using the following policy:
 1683  *   1) allocate the requested inode.
 1684  *   2) allocate the next available inode after the requested
 1685  *      inode in the specified cylinder group.
 1686  */
 1687 static ufs2_daddr_t
 1688 ffs_nodealloccg(ip, cg, ipref, mode)
 1689         struct inode *ip;
 1690         int cg;
 1691         ufs2_daddr_t ipref;
 1692         int mode;
 1693 {
 1694         struct fs *fs;
 1695         struct cg *cgp;
 1696         struct buf *bp, *ibp;
 1697         struct ufsmount *ump;
 1698         u_int8_t *inosused;
 1699         struct ufs2_dinode *dp2;
 1700         int error, start, len, loc, map, i;
 1701 
 1702         fs = ip->i_fs;
 1703         ump = ip->i_ump;
 1704         if (fs->fs_cs(fs, cg).cs_nifree == 0)
 1705                 return (0);
 1706         UFS_UNLOCK(ump);
 1707         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, cg)),
 1708                 (int)fs->fs_cgsize, NOCRED, &bp);
 1709         if (error) {
 1710                 brelse(bp);
 1711                 UFS_LOCK(ump);
 1712                 return (0);
 1713         }
 1714         cgp = (struct cg *)bp->b_data;
 1715         if (!cg_chkmagic(cgp) || cgp->cg_cs.cs_nifree == 0) {
 1716                 brelse(bp);
 1717                 UFS_LOCK(ump);
 1718                 return (0);
 1719         }
 1720         bp->b_xflags |= BX_BKGRDWRITE;
 1721         cgp->cg_old_time = cgp->cg_time = time_second;
 1722         inosused = cg_inosused(cgp);
 1723         if (ipref) {
 1724                 ipref %= fs->fs_ipg;
 1725                 if (isclr(inosused, ipref))
 1726                         goto gotit;
 1727         }
 1728         start = cgp->cg_irotor / NBBY;
 1729         len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY);
 1730         loc = skpc(0xff, len, &inosused[start]);
 1731         if (loc == 0) {
 1732                 len = start + 1;
 1733                 start = 0;
 1734                 loc = skpc(0xff, len, &inosused[0]);
 1735                 if (loc == 0) {
 1736                         printf("cg = %d, irotor = %ld, fs = %s\n",
 1737                             cg, (long)cgp->cg_irotor, fs->fs_fsmnt);
 1738                         panic("ffs_nodealloccg: map corrupted");
 1739                         /* NOTREACHED */
 1740                 }
 1741         }
 1742         i = start + len - loc;
 1743         map = inosused[i];
 1744         ipref = i * NBBY;
 1745         for (i = 1; i < (1 << NBBY); i <<= 1, ipref++) {
 1746                 if ((map & i) == 0) {
 1747                         cgp->cg_irotor = ipref;
 1748                         goto gotit;
 1749                 }
 1750         }
 1751         printf("fs = %s\n", fs->fs_fsmnt);
 1752         panic("ffs_nodealloccg: block not in map");
 1753         /* NOTREACHED */
 1754 gotit:
 1755         /*
 1756          * Check to see if we need to initialize more inodes.
 1757          */
 1758         ibp = NULL;
 1759         if (fs->fs_magic == FS_UFS2_MAGIC &&
 1760             ipref + INOPB(fs) > cgp->cg_initediblk &&
 1761             cgp->cg_initediblk < cgp->cg_niblk) {
 1762                 ibp = getblk(ip->i_devvp, fsbtodb(fs,
 1763                     ino_to_fsba(fs, cg * fs->fs_ipg + cgp->cg_initediblk)),
 1764                     (int)fs->fs_bsize, 0, 0, 0);
 1765                 bzero(ibp->b_data, (int)fs->fs_bsize);
 1766                 dp2 = (struct ufs2_dinode *)(ibp->b_data);
 1767                 for (i = 0; i < INOPB(fs); i++) {
 1768                         dp2->di_gen = arc4random() / 2 + 1;
 1769                         dp2++;
 1770                 }
 1771                 cgp->cg_initediblk += INOPB(fs);
 1772         }
 1773         UFS_LOCK(ump);
 1774         ACTIVECLEAR(fs, cg);
 1775         setbit(inosused, ipref);
 1776         cgp->cg_cs.cs_nifree--;
 1777         fs->fs_cstotal.cs_nifree--;
 1778         fs->fs_cs(fs, cg).cs_nifree--;
 1779         fs->fs_fmod = 1;
 1780         if ((mode & IFMT) == IFDIR) {
 1781                 cgp->cg_cs.cs_ndir++;
 1782                 fs->fs_cstotal.cs_ndir++;
 1783                 fs->fs_cs(fs, cg).cs_ndir++;
 1784         }
 1785         UFS_UNLOCK(ump);
 1786         if (DOINGSOFTDEP(ITOV(ip)))
 1787                 softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref);
 1788         bdwrite(bp);
 1789         if (ibp != NULL)
 1790                 bawrite(ibp);
 1791         return (cg * fs->fs_ipg + ipref);
 1792 }
 1793 
 1794 /*
 1795  * check if a block is free
 1796  */
 1797 static int
 1798 ffs_isfreeblock(struct fs *fs, u_char *cp, ufs1_daddr_t h)
 1799 {
 1800 
 1801         switch ((int)fs->fs_frag) {
 1802         case 8:
 1803                 return (cp[h] == 0);
 1804         case 4:
 1805                 return ((cp[h >> 1] & (0x0f << ((h & 0x1) << 2))) == 0);
 1806         case 2:
 1807                 return ((cp[h >> 2] & (0x03 << ((h & 0x3) << 1))) == 0);
 1808         case 1:
 1809                 return ((cp[h >> 3] & (0x01 << (h & 0x7))) == 0);
 1810         default:
 1811                 panic("ffs_isfreeblock");
 1812         }
 1813         return (0);
 1814 }
 1815 
 1816 /*
 1817  * Free a block or fragment.
 1818  *
 1819  * The specified block or fragment is placed back in the
 1820  * free map. If a fragment is deallocated, a possible
 1821  * block reassembly is checked.
 1822  */
 1823 void
 1824 ffs_blkfree(ump, fs, devvp, bno, size, inum)
 1825         struct ufsmount *ump;
 1826         struct fs *fs;
 1827         struct vnode *devvp;
 1828         ufs2_daddr_t bno;
 1829         long size;
 1830         ino_t inum;
 1831 {
 1832         struct cg *cgp;
 1833         struct buf *bp;
 1834         ufs1_daddr_t fragno, cgbno;
 1835         ufs2_daddr_t cgblkno;
 1836         int i, cg, blk, frags, bbase;
 1837         u_int8_t *blksfree;
 1838         struct cdev *dev;
 1839 
 1840         cg = dtog(fs, bno);
 1841         if (devvp->v_type != VCHR) {
 1842                 /* devvp is a snapshot */
 1843                 dev = VTOI(devvp)->i_devvp->v_rdev;
 1844                 cgblkno = fragstoblks(fs, cgtod(fs, cg));
 1845         } else {
 1846                 /* devvp is a normal disk device */
 1847                 dev = devvp->v_rdev;
 1848                 cgblkno = fsbtodb(fs, cgtod(fs, cg));
 1849                 ASSERT_VOP_LOCKED(devvp, "ffs_blkfree");
 1850                 if ((devvp->v_vflag & VV_COPYONWRITE) &&
 1851                     ffs_snapblkfree(fs, devvp, bno, size, inum))
 1852                         return;
 1853         }
 1854 #ifdef INVARIANTS
 1855         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 ||
 1856             fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) {
 1857                 printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n",
 1858                     devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize,
 1859                     size, fs->fs_fsmnt);
 1860                 panic("ffs_blkfree: bad size");
 1861         }
 1862 #endif
 1863         if ((u_int)bno >= fs->fs_size) {
 1864                 printf("bad block %jd, ino %lu\n", (intmax_t)bno,
 1865                     (u_long)inum);
 1866                 ffs_fserr(fs, inum, "bad block");
 1867                 return;
 1868         }
 1869         if (bread(devvp, cgblkno, (int)fs->fs_cgsize, NOCRED, &bp)) {
 1870                 brelse(bp);
 1871                 return;
 1872         }
 1873         cgp = (struct cg *)bp->b_data;
 1874         if (!cg_chkmagic(cgp)) {
 1875                 brelse(bp);
 1876                 return;
 1877         }
 1878         bp->b_xflags |= BX_BKGRDWRITE;
 1879         cgp->cg_old_time = cgp->cg_time = time_second;
 1880         cgbno = dtogd(fs, bno);
 1881         blksfree = cg_blksfree(cgp);
 1882         UFS_LOCK(ump);
 1883         if (size == fs->fs_bsize) {
 1884                 fragno = fragstoblks(fs, cgbno);
 1885                 if (!ffs_isfreeblock(fs, blksfree, fragno)) {
 1886                         if (devvp->v_type != VCHR) {
 1887                                 UFS_UNLOCK(ump);
 1888                                 /* devvp is a snapshot */
 1889                                 brelse(bp);
 1890                                 return;
 1891                         }
 1892                         printf("dev = %s, block = %jd, fs = %s\n",
 1893                             devtoname(dev), (intmax_t)bno, fs->fs_fsmnt);
 1894                         panic("ffs_blkfree: freeing free block");
 1895                 }
 1896                 ffs_setblock(fs, blksfree, fragno);
 1897                 ffs_clusteracct(ump, fs, cgp, fragno, 1);
 1898                 cgp->cg_cs.cs_nbfree++;
 1899                 fs->fs_cstotal.cs_nbfree++;
 1900                 fs->fs_cs(fs, cg).cs_nbfree++;
 1901         } else {
 1902                 bbase = cgbno - fragnum(fs, cgbno);
 1903                 /*
 1904                  * decrement the counts associated with the old frags
 1905                  */
 1906                 blk = blkmap(fs, blksfree, bbase);
 1907                 ffs_fragacct(fs, blk, cgp->cg_frsum, -1);
 1908                 /*
 1909                  * deallocate the fragment
 1910                  */
 1911                 frags = numfrags(fs, size);
 1912                 for (i = 0; i < frags; i++) {
 1913                         if (isset(blksfree, cgbno + i)) {
 1914                                 printf("dev = %s, block = %jd, fs = %s\n",
 1915                                     devtoname(dev), (intmax_t)(bno + i),
 1916                                     fs->fs_fsmnt);
 1917                                 panic("ffs_blkfree: freeing free frag");
 1918                         }
 1919                         setbit(blksfree, cgbno + i);
 1920                 }
 1921                 cgp->cg_cs.cs_nffree += i;
 1922                 fs->fs_cstotal.cs_nffree += i;
 1923                 fs->fs_cs(fs, cg).cs_nffree += i;
 1924                 /*
 1925                  * add back in counts associated with the new frags
 1926                  */
 1927                 blk = blkmap(fs, blksfree, bbase);
 1928                 ffs_fragacct(fs, blk, cgp->cg_frsum, 1);
 1929                 /*
 1930                  * if a complete block has been reassembled, account for it
 1931                  */
 1932                 fragno = fragstoblks(fs, bbase);
 1933                 if (ffs_isblock(fs, blksfree, fragno)) {
 1934                         cgp->cg_cs.cs_nffree -= fs->fs_frag;
 1935                         fs->fs_cstotal.cs_nffree -= fs->fs_frag;
 1936                         fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
 1937                         ffs_clusteracct(ump, fs, cgp, fragno, 1);
 1938                         cgp->cg_cs.cs_nbfree++;
 1939                         fs->fs_cstotal.cs_nbfree++;
 1940                         fs->fs_cs(fs, cg).cs_nbfree++;
 1941                 }
 1942         }
 1943         fs->fs_fmod = 1;
 1944         ACTIVECLEAR(fs, cg);
 1945         UFS_UNLOCK(ump);
 1946         bdwrite(bp);
 1947 }
 1948 
 1949 #ifdef INVARIANTS
 1950 /*
 1951  * Verify allocation of a block or fragment. Returns true if block or
 1952  * fragment is allocated, false if it is free.
 1953  */
 1954 static int
 1955 ffs_checkblk(ip, bno, size)
 1956         struct inode *ip;
 1957         ufs2_daddr_t bno;
 1958         long size;
 1959 {
 1960         struct fs *fs;
 1961         struct cg *cgp;
 1962         struct buf *bp;
 1963         ufs1_daddr_t cgbno;
 1964         int i, error, frags, free;
 1965         u_int8_t *blksfree;
 1966 
 1967         fs = ip->i_fs;
 1968         if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) {
 1969                 printf("bsize = %ld, size = %ld, fs = %s\n",
 1970                     (long)fs->fs_bsize, size, fs->fs_fsmnt);
 1971                 panic("ffs_checkblk: bad size");
 1972         }
 1973         if ((u_int)bno >= fs->fs_size)
 1974                 panic("ffs_checkblk: bad block %jd", (intmax_t)bno);
 1975         error = bread(ip->i_devvp, fsbtodb(fs, cgtod(fs, dtog(fs, bno))),
 1976                 (int)fs->fs_cgsize, NOCRED, &bp);
 1977         if (error)
 1978                 panic("ffs_checkblk: cg bread failed");
 1979         cgp = (struct cg *)bp->b_data;
 1980         if (!cg_chkmagic(cgp))
 1981                 panic("ffs_checkblk: cg magic mismatch");
 1982         bp->b_xflags |= BX_BKGRDWRITE;
 1983         blksfree = cg_blksfree(cgp);
 1984         cgbno = dtogd(fs, bno);
 1985         if (size == fs->fs_bsize) {
 1986                 free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno));
 1987         } else {
 1988                 frags = numfrags(fs, size);
 1989                 for (free = 0, i = 0; i < frags; i++)
 1990                         if (isset(blksfree, cgbno + i))
 1991                                 free++;
 1992                 if (free != 0 && free != frags)
 1993                         panic("ffs_checkblk: partially free fragment");
 1994         }
 1995         brelse(bp);
 1996         return (!free);
 1997 }
 1998 #endif /* INVARIANTS */
 1999 
 2000 /*
 2001  * Free an inode.
 2002  */
 2003 int
 2004 ffs_vfree(pvp, ino, mode)
 2005         struct vnode *pvp;
 2006         ino_t ino;
 2007         int mode;
 2008 {
 2009         struct inode *ip;
 2010 
 2011         if (DOINGSOFTDEP(pvp)) {
 2012                 softdep_freefile(pvp, ino, mode);
 2013                 return (0);
 2014         }
 2015         ip = VTOI(pvp);
 2016         return (ffs_freefile(ip->i_ump, ip->i_fs, ip->i_devvp, ino, mode));
 2017 }
 2018 
 2019 /*
 2020  * Do the actual free operation.
 2021  * The specified inode is placed back in the free map.
 2022  */
 2023 int
 2024 ffs_freefile(ump, fs, devvp, ino, mode)
 2025         struct ufsmount *ump;
 2026         struct fs *fs;
 2027         struct vnode *devvp;
 2028         ino_t ino;
 2029         int mode;
 2030 {
 2031         struct cg *cgp;
 2032         struct buf *bp;
 2033         ufs2_daddr_t cgbno;
 2034         int error, cg;
 2035         u_int8_t *inosused;
 2036         struct cdev *dev;
 2037 
 2038         cg = ino_to_cg(fs, ino);
 2039         if (devvp->v_type != VCHR) {
 2040                 /* devvp is a snapshot */
 2041                 dev = VTOI(devvp)->i_devvp->v_rdev;
 2042                 cgbno = fragstoblks(fs, cgtod(fs, cg));
 2043         } else {
 2044                 /* devvp is a normal disk device */
 2045                 dev = devvp->v_rdev;
 2046                 cgbno = fsbtodb(fs, cgtod(fs, cg));
 2047         }
 2048         if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
 2049                 panic("ffs_freefile: range: dev = %s, ino = %lu, fs = %s",
 2050                     devtoname(dev), (u_long)ino, fs->fs_fsmnt);
 2051         if ((error = bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp))) {
 2052                 brelse(bp);
 2053                 return (error);
 2054         }
 2055         cgp = (struct cg *)bp->b_data;
 2056         if (!cg_chkmagic(cgp)) {
 2057                 brelse(bp);
 2058                 return (0);
 2059         }
 2060         bp->b_xflags |= BX_BKGRDWRITE;
 2061         cgp->cg_old_time = cgp->cg_time = time_second;
 2062         inosused = cg_inosused(cgp);
 2063         ino %= fs->fs_ipg;
 2064         if (isclr(inosused, ino)) {
 2065                 printf("dev = %s, ino = %lu, fs = %s\n", devtoname(dev),
 2066                     (u_long)ino + cg * fs->fs_ipg, fs->fs_fsmnt);
 2067                 if (fs->fs_ronly == 0)
 2068                         panic("ffs_freefile: freeing free inode");
 2069         }
 2070         clrbit(inosused, ino);
 2071         if (ino < cgp->cg_irotor)
 2072                 cgp->cg_irotor = ino;
 2073         cgp->cg_cs.cs_nifree++;
 2074         UFS_LOCK(ump);
 2075         fs->fs_cstotal.cs_nifree++;
 2076         fs->fs_cs(fs, cg).cs_nifree++;
 2077         if ((mode & IFMT) == IFDIR) {
 2078                 cgp->cg_cs.cs_ndir--;
 2079                 fs->fs_cstotal.cs_ndir--;
 2080                 fs->fs_cs(fs, cg).cs_ndir--;
 2081         }
 2082         fs->fs_fmod = 1;
 2083         ACTIVECLEAR(fs, cg);
 2084         UFS_UNLOCK(ump);
 2085         bdwrite(bp);
 2086         return (0);
 2087 }
 2088 
 2089 /*
 2090  * Check to see if a file is free.
 2091  */
 2092 int
 2093 ffs_checkfreefile(fs, devvp, ino)
 2094         struct fs *fs;
 2095         struct vnode *devvp;
 2096         ino_t ino;
 2097 {
 2098         struct cg *cgp;
 2099         struct buf *bp;
 2100         ufs2_daddr_t cgbno;
 2101         int ret, cg;
 2102         u_int8_t *inosused;
 2103 
 2104         cg = ino_to_cg(fs, ino);
 2105         if (devvp->v_type != VCHR) {
 2106                 /* devvp is a snapshot */
 2107                 cgbno = fragstoblks(fs, cgtod(fs, cg));
 2108         } else {
 2109                 /* devvp is a normal disk device */
 2110                 cgbno = fsbtodb(fs, cgtod(fs, cg));
 2111         }
 2112         if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
 2113                 return (1);
 2114         if (bread(devvp, cgbno, (int)fs->fs_cgsize, NOCRED, &bp)) {
 2115                 brelse(bp);
 2116                 return (1);
 2117         }
 2118         cgp = (struct cg *)bp->b_data;
 2119         if (!cg_chkmagic(cgp)) {
 2120                 brelse(bp);
 2121                 return (1);
 2122         }
 2123         inosused = cg_inosused(cgp);
 2124         ino %= fs->fs_ipg;
 2125         ret = isclr(inosused, ino);
 2126         brelse(bp);
 2127         return (ret);
 2128 }
 2129 
 2130 /*
 2131  * Find a block of the specified size in the specified cylinder group.
 2132  *
 2133  * It is a panic if a request is made to find a block if none are
 2134  * available.
 2135  */
 2136 static ufs1_daddr_t
 2137 ffs_mapsearch(fs, cgp, bpref, allocsiz)
 2138         struct fs *fs;
 2139         struct cg *cgp;
 2140         ufs2_daddr_t bpref;
 2141         int allocsiz;
 2142 {
 2143         ufs1_daddr_t bno;
 2144         int start, len, loc, i;
 2145         int blk, field, subfield, pos;
 2146         u_int8_t *blksfree;
 2147 
 2148         /*
 2149          * find the fragment by searching through the free block
 2150          * map for an appropriate bit pattern
 2151          */
 2152         if (bpref)
 2153                 start = dtogd(fs, bpref) / NBBY;
 2154         else
 2155                 start = cgp->cg_frotor / NBBY;
 2156         blksfree = cg_blksfree(cgp);
 2157         len = howmany(fs->fs_fpg, NBBY) - start;
 2158         loc = scanc((u_int)len, (u_char *)&blksfree[start],
 2159                 fragtbl[fs->fs_frag],
 2160                 (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
 2161         if (loc == 0) {
 2162                 len = start + 1;
 2163                 start = 0;
 2164                 loc = scanc((u_int)len, (u_char *)&blksfree[0],
 2165                         fragtbl[fs->fs_frag],
 2166                         (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY))));
 2167                 if (loc == 0) {
 2168                         printf("start = %d, len = %d, fs = %s\n",
 2169                             start, len, fs->fs_fsmnt);
 2170                         panic("ffs_alloccg: map corrupted");
 2171                         /* NOTREACHED */
 2172                 }
 2173         }
 2174         bno = (start + len - loc) * NBBY;
 2175         cgp->cg_frotor = bno;
 2176         /*
 2177          * found the byte in the map
 2178          * sift through the bits to find the selected frag
 2179          */
 2180         for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
 2181                 blk = blkmap(fs, blksfree, bno);
 2182                 blk <<= 1;
 2183                 field = around[allocsiz];
 2184                 subfield = inside[allocsiz];
 2185                 for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
 2186                         if ((blk & field) == subfield)
 2187                                 return (bno + pos);
 2188                         field <<= 1;
 2189                         subfield <<= 1;
 2190                 }
 2191         }
 2192         printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt);
 2193         panic("ffs_alloccg: block not in map");
 2194         return (-1);
 2195 }
 2196 
 2197 /*
 2198  * Update the cluster map because of an allocation or free.
 2199  *
 2200  * Cnt == 1 means free; cnt == -1 means allocating.
 2201  */
 2202 void
 2203 ffs_clusteracct(ump, fs, cgp, blkno, cnt)
 2204         struct ufsmount *ump;
 2205         struct fs *fs;
 2206         struct cg *cgp;
 2207         ufs1_daddr_t blkno;
 2208         int cnt;
 2209 {
 2210         int32_t *sump;
 2211         int32_t *lp;
 2212         u_char *freemapp, *mapp;
 2213         int i, start, end, forw, back, map, bit;
 2214 
 2215         mtx_assert(UFS_MTX(ump), MA_OWNED);
 2216 
 2217         if (fs->fs_contigsumsize <= 0)
 2218                 return;
 2219         freemapp = cg_clustersfree(cgp);
 2220         sump = cg_clustersum(cgp);
 2221         /*
 2222          * Allocate or clear the actual block.
 2223          */
 2224         if (cnt > 0)
 2225                 setbit(freemapp, blkno);
 2226         else
 2227                 clrbit(freemapp, blkno);
 2228         /*
 2229          * Find the size of the cluster going forward.
 2230          */
 2231         start = blkno + 1;
 2232         end = start + fs->fs_contigsumsize;
 2233         if (end >= cgp->cg_nclusterblks)
 2234                 end = cgp->cg_nclusterblks;
 2235         mapp = &freemapp[start / NBBY];
 2236         map = *mapp++;
 2237         bit = 1 << (start % NBBY);
 2238         for (i = start; i < end; i++) {
 2239                 if ((map & bit) == 0)
 2240                         break;
 2241                 if ((i & (NBBY - 1)) != (NBBY - 1)) {
 2242                         bit <<= 1;
 2243                 } else {
 2244                         map = *mapp++;
 2245                         bit = 1;
 2246                 }
 2247         }
 2248         forw = i - start;
 2249         /*
 2250          * Find the size of the cluster going backward.
 2251          */
 2252         start = blkno - 1;
 2253         end = start - fs->fs_contigsumsize;
 2254         if (end < 0)
 2255                 end = -1;
 2256         mapp = &freemapp[start / NBBY];
 2257         map = *mapp--;
 2258         bit = 1 << (start % NBBY);
 2259         for (i = start; i > end; i--) {
 2260                 if ((map & bit) == 0)
 2261                         break;
 2262                 if ((i & (NBBY - 1)) != 0) {
 2263                         bit >>= 1;
 2264                 } else {
 2265                         map = *mapp--;
 2266                         bit = 1 << (NBBY - 1);
 2267                 }
 2268         }
 2269         back = start - i;
 2270         /*
 2271          * Account for old cluster and the possibly new forward and
 2272          * back clusters.
 2273          */
 2274         i = back + forw + 1;
 2275         if (i > fs->fs_contigsumsize)
 2276                 i = fs->fs_contigsumsize;
 2277         sump[i] += cnt;
 2278         if (back > 0)
 2279                 sump[back] -= cnt;
 2280         if (forw > 0)
 2281                 sump[forw] -= cnt;
 2282         /*
 2283          * Update cluster summary information.
 2284          */
 2285         lp = &sump[fs->fs_contigsumsize];
 2286         for (i = fs->fs_contigsumsize; i > 0; i--)
 2287                 if (*lp-- > 0)
 2288                         break;
 2289         fs->fs_maxcluster[cgp->cg_cgx] = i;
 2290 }
 2291 
 2292 /*
 2293  * Fserr prints the name of a filesystem with an error diagnostic.
 2294  *
 2295  * The form of the error message is:
 2296  *      fs: error message
 2297  */
 2298 static void
 2299 ffs_fserr(fs, inum, cp)
 2300         struct fs *fs;
 2301         ino_t inum;
 2302         char *cp;
 2303 {
 2304         struct thread *td = curthread;  /* XXX */
 2305         struct proc *p = td->td_proc;
 2306 
 2307         log(LOG_ERR, "pid %d (%s), uid %d inumber %d on %s: %s\n",
 2308             p->p_pid, p->p_comm, td->td_ucred->cr_uid, inum, fs->fs_fsmnt, cp);
 2309 }
 2310 
 2311 /*
 2312  * This function provides the capability for the fsck program to
 2313  * update an active filesystem. Eleven operations are provided:
 2314  *
 2315  * adjrefcnt(inode, amt) - adjusts the reference count on the
 2316  *      specified inode by the specified amount. Under normal
 2317  *      operation the count should always go down. Decrementing
 2318  *      the count to zero will cause the inode to be freed.
 2319  * adjblkcnt(inode, amt) - adjust the number of blocks used to
 2320  *      by the specifed amount.
 2321  * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) -
 2322  *      adjust the superblock summary.
 2323  * freedirs(inode, count) - directory inodes [inode..inode + count - 1]
 2324  *      are marked as free. Inodes should never have to be marked
 2325  *      as in use.
 2326  * freefiles(inode, count) - file inodes [inode..inode + count - 1]
 2327  *      are marked as free. Inodes should never have to be marked
 2328  *      as in use.
 2329  * freeblks(blockno, size) - blocks [blockno..blockno + size - 1]
 2330  *      are marked as free. Blocks should never have to be marked
 2331  *      as in use.
 2332  * setflags(flags, set/clear) - the fs_flags field has the specified
 2333  *      flags set (second parameter +1) or cleared (second parameter -1).
 2334  */
 2335 
 2336 static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS);
 2337 
 2338 SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR|CTLTYPE_STRUCT,
 2339         0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count");
 2340 
 2341 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR,
 2342         sysctl_ffs_fsck, "Adjust Inode Used Blocks Count");
 2343 
 2344 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR,
 2345         sysctl_ffs_fsck, "Adjust number of directories");
 2346 
 2347 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR,
 2348         sysctl_ffs_fsck, "Adjust number of free blocks");
 2349 
 2350 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR,
 2351         sysctl_ffs_fsck, "Adjust number of free inodes");
 2352 
 2353 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR,
 2354         sysctl_ffs_fsck, "Adjust number of free frags");
 2355 
 2356 static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR,
 2357         sysctl_ffs_fsck, "Adjust number of free clusters");
 2358 
 2359 static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR,
 2360         sysctl_ffs_fsck, "Free Range of Directory Inodes");
 2361 
 2362 static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR,
 2363         sysctl_ffs_fsck, "Free Range of File Inodes");
 2364 
 2365 static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR,
 2366         sysctl_ffs_fsck, "Free Range of Blocks");
 2367 
 2368 static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR,
 2369         sysctl_ffs_fsck, "Change Filesystem Flags");
 2370 
 2371 #ifdef DEBUG
 2372 static int fsckcmds = 0;
 2373 SYSCTL_INT(_debug, OID_AUTO, fsckcmds, CTLFLAG_RW, &fsckcmds, 0, "");
 2374 #endif /* DEBUG */
 2375 
 2376 static int
 2377 sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS)
 2378 {
 2379         struct fsck_cmd cmd;
 2380         struct ufsmount *ump;
 2381         struct vnode *vp;
 2382         struct inode *ip;
 2383         struct mount *mp;
 2384         struct fs *fs;
 2385         ufs2_daddr_t blkno;
 2386         long blkcnt, blksize;
 2387         struct file *fp;
 2388         int filetype, error;
 2389 
 2390         if (req->newlen > sizeof cmd)
 2391                 return (EBADRPC);
 2392         if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0)
 2393                 return (error);
 2394         if (cmd.version != FFS_CMD_VERSION)
 2395                 return (ERPCMISMATCH);
 2396         if ((error = getvnode(curproc->p_fd, cmd.handle, &fp)) != 0)
 2397                 return (error);
 2398         vn_start_write(fp->f_data, &mp, V_WAIT);
 2399         if (mp == 0 || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) {
 2400                 vn_finished_write(mp);
 2401                 fdrop(fp, curthread);
 2402                 return (EINVAL);
 2403         }
 2404         if (mp->mnt_flag & MNT_RDONLY) {
 2405                 vn_finished_write(mp);
 2406                 fdrop(fp, curthread);
 2407                 return (EROFS);
 2408         }
 2409         ump = VFSTOUFS(mp);
 2410         fs = ump->um_fs;
 2411         filetype = IFREG;
 2412 
 2413         switch (oidp->oid_number) {
 2414 
 2415         case FFS_SET_FLAGS:
 2416 #ifdef DEBUG
 2417                 if (fsckcmds)
 2418                         printf("%s: %s flags\n", mp->mnt_stat.f_mntonname,
 2419                             cmd.size > 0 ? "set" : "clear");
 2420 #endif /* DEBUG */
 2421                 if (cmd.size > 0)
 2422                         fs->fs_flags |= (long)cmd.value;
 2423                 else
 2424                         fs->fs_flags &= ~(long)cmd.value;
 2425                 break;
 2426 
 2427         case FFS_ADJ_REFCNT:
 2428 #ifdef DEBUG
 2429                 if (fsckcmds) {
 2430                         printf("%s: adjust inode %jd count by %jd\n",
 2431                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
 2432                             (intmax_t)cmd.size);
 2433                 }
 2434 #endif /* DEBUG */
 2435                 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
 2436                         break;
 2437                 ip = VTOI(vp);
 2438                 ip->i_nlink += cmd.size;
 2439                 DIP_SET(ip, i_nlink, ip->i_nlink);
 2440                 ip->i_effnlink += cmd.size;
 2441                 ip->i_flag |= IN_CHANGE;
 2442                 if (DOINGSOFTDEP(vp))
 2443                         softdep_change_linkcnt(ip);
 2444                 vput(vp);
 2445                 break;
 2446 
 2447         case FFS_ADJ_BLKCNT:
 2448 #ifdef DEBUG
 2449                 if (fsckcmds) {
 2450                         printf("%s: adjust inode %jd block count by %jd\n",
 2451                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value,
 2452                             (intmax_t)cmd.size);
 2453                 }
 2454 #endif /* DEBUG */
 2455                 if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp)))
 2456                         break;
 2457                 ip = VTOI(vp);
 2458                 if (ip->i_flag & IN_SPACECOUNTED) {
 2459                         UFS_LOCK(ump);
 2460                         fs->fs_pendingblocks += cmd.size;
 2461                         UFS_UNLOCK(ump);
 2462                 }
 2463                 DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size);
 2464                 ip->i_flag |= IN_CHANGE;
 2465                 vput(vp);
 2466                 break;
 2467 
 2468         case FFS_DIR_FREE:
 2469                 filetype = IFDIR;
 2470                 /* fall through */
 2471 
 2472         case FFS_FILE_FREE:
 2473 #ifdef DEBUG
 2474                 if (fsckcmds) {
 2475                         if (cmd.size == 1)
 2476                                 printf("%s: free %s inode %d\n",
 2477                                     mp->mnt_stat.f_mntonname,
 2478                                     filetype == IFDIR ? "directory" : "file",
 2479                                     (ino_t)cmd.value);
 2480                         else
 2481                                 printf("%s: free %s inodes %d-%d\n",
 2482                                     mp->mnt_stat.f_mntonname,
 2483                                     filetype == IFDIR ? "directory" : "file",
 2484                                     (ino_t)cmd.value,
 2485                                     (ino_t)(cmd.value + cmd.size - 1));
 2486                 }
 2487 #endif /* DEBUG */
 2488                 while (cmd.size > 0) {
 2489                         if ((error = ffs_freefile(ump, fs, ump->um_devvp,
 2490                             cmd.value, filetype)))
 2491                                 break;
 2492                         cmd.size -= 1;
 2493                         cmd.value += 1;
 2494                 }
 2495                 break;
 2496 
 2497         case FFS_BLK_FREE:
 2498 #ifdef DEBUG
 2499                 if (fsckcmds) {
 2500                         if (cmd.size == 1)
 2501                                 printf("%s: free block %jd\n",
 2502                                     mp->mnt_stat.f_mntonname,
 2503                                     (intmax_t)cmd.value);
 2504                         else
 2505                                 printf("%s: free blocks %jd-%jd\n",
 2506                                     mp->mnt_stat.f_mntonname, 
 2507                                     (intmax_t)cmd.value,
 2508                                     (intmax_t)cmd.value + cmd.size - 1);
 2509                 }
 2510 #endif /* DEBUG */
 2511                 blkno = cmd.value;
 2512                 blkcnt = cmd.size;
 2513                 blksize = fs->fs_frag - (blkno % fs->fs_frag);
 2514                 while (blkcnt > 0) {
 2515                         if (blksize > blkcnt)
 2516                                 blksize = blkcnt;
 2517                         ffs_blkfree(ump, fs, ump->um_devvp, blkno,
 2518                             blksize * fs->fs_fsize, ROOTINO);
 2519                         blkno += blksize;
 2520                         blkcnt -= blksize;
 2521                         blksize = fs->fs_frag;
 2522                 }
 2523                 break;
 2524 
 2525         /*
 2526          * Adjust superblock summaries.  fsck(8) is expected to
 2527          * submit deltas when necessary.
 2528          */
 2529         case FFS_ADJ_NDIR:
 2530 #ifdef DEBUG
 2531                 if (fsckcmds) {
 2532                         printf("%s: adjust number of directories by %jd\n",
 2533                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2534                 }
 2535 #endif /* DEBUG */
 2536                 fs->fs_cstotal.cs_ndir += cmd.value;
 2537                 break;
 2538         case FFS_ADJ_NBFREE:
 2539 #ifdef DEBUG
 2540                 if (fsckcmds) {
 2541                         printf("%s: adjust number of free blocks by %+jd\n",
 2542                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2543                 }
 2544 #endif /* DEBUG */
 2545                 fs->fs_cstotal.cs_nbfree += cmd.value;
 2546                 break;
 2547         case FFS_ADJ_NIFREE:
 2548 #ifdef DEBUG
 2549                 if (fsckcmds) {
 2550                         printf("%s: adjust number of free inodes by %+jd\n",
 2551                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2552                 }
 2553 #endif /* DEBUG */
 2554                 fs->fs_cstotal.cs_nifree += cmd.value;
 2555                 break;
 2556         case FFS_ADJ_NFFREE:
 2557 #ifdef DEBUG
 2558                 if (fsckcmds) {
 2559                         printf("%s: adjust number of free frags by %+jd\n",
 2560                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2561                 }
 2562 #endif /* DEBUG */
 2563                 fs->fs_cstotal.cs_nffree += cmd.value;
 2564                 break;
 2565         case FFS_ADJ_NUMCLUSTERS:
 2566 #ifdef DEBUG
 2567                 if (fsckcmds) {
 2568                         printf("%s: adjust number of free clusters by %+jd\n",
 2569                             mp->mnt_stat.f_mntonname, (intmax_t)cmd.value);
 2570                 }
 2571 #endif /* DEBUG */
 2572                 fs->fs_cstotal.cs_numclusters += cmd.value;
 2573                 break;
 2574 
 2575         default:
 2576 #ifdef DEBUG
 2577                 if (fsckcmds) {
 2578                         printf("Invalid request %d from fsck\n",
 2579                             oidp->oid_number);
 2580                 }
 2581 #endif /* DEBUG */
 2582                 error = EINVAL;
 2583                 break;
 2584 
 2585         }
 2586         fdrop(fp, curthread);
 2587         vn_finished_write(mp);
 2588         return (error);
 2589 }

Cache object: a1066cce32a62c470396ceb0df30ab3b


[ 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.