static int
ext2_ext_balloc(struct inode *ip, uint32_t lbn, int size,
struct ucred *cred, struct buf **bpp, int flags)
{
struct m_ext2fs *fs;
struct buf *bp = NULL;
struct vnode *vp = ITOV(ip);
daddr_t newblk;
int osize, nsize, blks, error, allocated;
fs = ip->i_e2fs;
blks = howmany(size, fs->e2fs_bsize);
error = ext4_ext_get_blocks(ip, lbn, blks, cred, NULL, &allocated, &newblk);
if (error)
return (error);
if (allocated) {
if (ip->i_size < (lbn + 1) * fs->e2fs_bsize)
nsize = fragroundup(fs, size);
else
nsize = fs->e2fs_bsize;
bp = getblk(vp, lbn, nsize, 0, 0, 0);
if(!bp)
return (EIO);
bp->b_blkno = fsbtodb(fs, newblk);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
} else {
if (ip->i_size >= (lbn + 1) * fs->e2fs_bsize) {
error = bread(vp, lbn, fs->e2fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, newblk);
*bpp = bp;
return (0);
}
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize)
error = bread(vp, lbn, osize, NOCRED, &bp);
else
error = bread(vp, lbn, fs->e2fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, newblk);
}
*bpp = bp;
return (error);
}
/*
* Read data to a buf, including read-ahead if we find this to be beneficial.
* cluster_read replaces bread.
*/
int
cluster_read(struct vnode *vp, u_quad_t filesize, daddr_t lblkno, long size,
struct ucred *cred, long totread, int seqcount, int gbflags,
struct buf **bpp)
{
struct buf *bp, *rbp, *reqbp;
struct bufobj *bo;
daddr_t blkno, origblkno;
int maxra, racluster;
int error, ncontig;
int i;
error = 0;
bo = &vp->v_bufobj;
if (!unmapped_buf_allowed)
gbflags &= ~GB_UNMAPPED;
/*
* Try to limit the amount of read-ahead by a few
* ad-hoc parameters. This needs work!!!
*/
racluster = vp->v_mount->mnt_iosize_max / size;
maxra = seqcount;
maxra = min(read_max, maxra);
maxra = min(nbuf/8, maxra);
if (((u_quad_t)(lblkno + maxra + 1) * size) > filesize)
maxra = (filesize / size) - lblkno;
/*
* get the requested block
*/
*bpp = reqbp = bp = getblk(vp, lblkno, size, 0, 0, gbflags);
origblkno = lblkno;
/*
* if it is in the cache, then check to see if the reads have been
* sequential. If they have, then try some read-ahead, otherwise
* back-off on prospective read-aheads.
*/
if (bp->b_flags & B_CACHE) {
if (!seqcount) {
return 0;
} else if ((bp->b_flags & B_RAM) == 0) {
return 0;
} else {
bp->b_flags &= ~B_RAM;
BO_RLOCK(bo);
for (i = 1; i < maxra; i++) {
/*
* Stop if the buffer does not exist or it
* is invalid (about to go away?)
*/
rbp = gbincore(&vp->v_bufobj, lblkno+i);
if (rbp == NULL || (rbp->b_flags & B_INVAL))
break;
/*
* Set another read-ahead mark so we know
* to check again. (If we can lock the
* buffer without waiting)
*/
if ((((i % racluster) == (racluster - 1)) ||
(i == (maxra - 1)))
&& (0 == BUF_LOCK(rbp,
LK_EXCLUSIVE | LK_NOWAIT, NULL))) {
rbp->b_flags |= B_RAM;
BUF_UNLOCK(rbp);
}
}
BO_RUNLOCK(bo);
if (i >= maxra) {
return 0;
}
lblkno += i;
}
reqbp = bp = NULL;
/*
* If it isn't in the cache, then get a chunk from
* disk if sequential, otherwise just get the block.
*/
} else {
off_t firstread = bp->b_offset;
int nblks;
long minread;
KASSERT(bp->b_offset != NOOFFSET,
("cluster_read: no buffer offset"));
ncontig = 0;
/*
* Adjust totread if needed
*/
minread = read_min * size;
if (minread > totread)
totread = minread;
/*
* Compute the total number of blocks that we should read
* synchronously.
*/
if (firstread + totread > filesize)
totread = filesize - firstread;
nblks = howmany(totread, size);
if (nblks > racluster)
nblks = racluster;
/*
* Now compute the number of contiguous blocks.
*/
if (nblks > 1) {
error = VOP_BMAP(vp, lblkno, NULL,
&blkno, &ncontig, NULL);
/*
* If this failed to map just do the original block.
*/
if (error || blkno == -1)
ncontig = 0;
}
/*
* If we have contiguous data available do a cluster
* otherwise just read the requested block.
*/
if (ncontig) {
/* Account for our first block. */
ncontig = min(ncontig + 1, nblks);
if (ncontig < nblks)
nblks = ncontig;
bp = cluster_rbuild(vp, filesize, lblkno,
blkno, size, nblks, gbflags, bp);
lblkno += (bp->b_bufsize / size);
} else {
bp->b_flags |= B_RAM;
bp->b_iocmd = BIO_READ;
lblkno += 1;
}
}
/*
* handle the synchronous read so that it is available ASAP.
*/
if (bp) {
if ((bp->b_flags & B_CLUSTER) == 0) {
vfs_busy_pages(bp, 0);
}
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
if ((bp->b_flags & B_ASYNC) || bp->b_iodone != NULL)
BUF_KERNPROC(bp);
bp->b_iooffset = dbtob(bp->b_blkno);
bstrategy(bp);
curthread->td_ru.ru_inblock++;
}
/*
* If we have been doing sequential I/O, then do some read-ahead.
*/
while (lblkno < (origblkno + maxra)) {
error = VOP_BMAP(vp, lblkno, NULL, &blkno, &ncontig, NULL);
if (error)
break;
if (blkno == -1)
break;
/*
* We could throttle ncontig here by maxra but we might as
* well read the data if it is contiguous. We're throttled
* by racluster anyway.
*/
if (ncontig) {
ncontig = min(ncontig + 1, racluster);
rbp = cluster_rbuild(vp, filesize, lblkno, blkno,
size, ncontig, gbflags, NULL);
lblkno += (rbp->b_bufsize / size);
if (rbp->b_flags & B_DELWRI) {
bqrelse(rbp);
continue;
}
} else {
rbp = getblk(vp, lblkno, size, 0, 0, gbflags);
lblkno += 1;
if (rbp->b_flags & B_DELWRI) {
bqrelse(rbp);
continue;
}
rbp->b_flags |= B_ASYNC | B_RAM;
rbp->b_iocmd = BIO_READ;
rbp->b_blkno = blkno;
}
if (rbp->b_flags & B_CACHE) {
rbp->b_flags &= ~B_ASYNC;
bqrelse(rbp);
continue;
}
if ((rbp->b_flags & B_CLUSTER) == 0) {
vfs_busy_pages(rbp, 0);
}
rbp->b_flags &= ~B_INVAL;
rbp->b_ioflags &= ~BIO_ERROR;
if ((rbp->b_flags & B_ASYNC) || rbp->b_iodone != NULL)
BUF_KERNPROC(rbp);
rbp->b_iooffset = dbtob(rbp->b_blkno);
bstrategy(rbp);
curthread->td_ru.ru_inblock++;
}
if (reqbp)
return (bufwait(reqbp));
else
return (error);
}
/*
* Check the super block and its summary info.
*/
static int
checksb(int listerr)
{
caddr_t err;
/*
* When the fs check is successfully completed, the alternate super
* block at sblk.b_bno will be overwritten by ckfini() with the
* repaired super block.
*/
sblk.b_bno = bflag ? bflag : (SBOFF / dev_bsize);
sblk.b_size = SBSIZE;
/*
* Sanity-check some of the values we are going to use later
* in allocation requests.
*/
if (sblock.fs_cstotal.cs_ndir < 1 ||
sblock.fs_cstotal.cs_ndir > sblock.fs_ncg * sblock.fs_ipg) {
if (verbose)
(void) printf(
"Found %d directories, should be between 1 and %d inclusive.\n",
sblock.fs_cstotal.cs_ndir,
sblock.fs_ncg * sblock.fs_ipg);
err = "NUMBER OF DIRECTORIES OUT OF RANGE";
goto failedsb;
}
if (sblock.fs_nrpos <= 0 || sblock.fs_postbloff < 0 ||
sblock.fs_cpc < 0 ||
(sblock.fs_postbloff +
(sblock.fs_nrpos * sblock.fs_cpc * sizeof (short))) >
sblock.fs_sbsize) {
err = "ROTATIONAL POSITION TABLE SIZE OUT OF RANGE";
goto failedsb;
}
if (sblock.fs_cssize !=
fragroundup(&sblock, sblock.fs_ncg * sizeof (struct csum))) {
err = "SIZE OF CYLINDER GROUP SUMMARY AREA WRONG";
goto failedsb;
}
if (sblock.fs_inopb != (sblock.fs_bsize / sizeof (struct dinode))) {
err = "INOPB NONSENSICAL RELATIVE TO BSIZE";
goto failedsb;
}
if (sblock.fs_bsize > MAXBSIZE) {
err = "BLOCK SIZE LARGER THAN MAXIMUM SUPPORTED";
goto failedsb;
}
if (sblock.fs_bsize != (sblock.fs_frag * sblock.fs_fsize)) {
err = "FRAGS PER BLOCK OR FRAG SIZE WRONG";
goto failedsb;
}
if (sblock.fs_dsize >= sblock.fs_size) {
err = "NUMBER OF DATA BLOCKS OUT OF RANGE";
goto failedsb;
}
#if 0
if (sblock.fs_size >
(sblock.fs_nsect * sblock.fs_ntrak * sblock.fs_ncyl)) {
err = "FILESYSTEM SIZE LARGER THAN DEVICE";
goto failedsb;
}
#endif
/*
* Check that the number of inodes per group isn't less than or
* equal to zero. Also makes sure it isn't more than the
* maximum number mkfs enforces.
*/
if (sblock.fs_ipg <= 0 || sblock.fs_ipg > MAXIpG) {
err = "INODES PER GROUP OUT OF RANGE";
goto failedsb;
}
if (sblock.fs_cgsize > sblock.fs_bsize) {
err = "CG HEADER LARGER THAN ONE BLOCK";
goto failedsb;
}
/*
* Set all possible fields that could differ, then do check
* of whole super block against an alternate super block.
* When an alternate super-block is specified this check is skipped.
*/
(void) getblk(&asblk, cgsblock(&sblock, sblock.fs_ncg - 1),
(size_t)sblock.fs_sbsize);
if (asblk.b_errs != 0) {
brelse(&asblk);
return (0);
}
if (bflag != 0) {
/*
* Invalidate clean flag and state information.
* Note that we couldn't return until after the
* above getblk(), because we're going to want to
* update asblk when everything's done.
*/
sblock.fs_clean = FSACTIVE;
sblock.fs_state = (long)sblock.fs_time;
sblock.fs_reclaim = 0;
sbdirty();
havesb = 1;
return (1);
}
altsblock.fs_link = sblock.fs_link;
altsblock.fs_rolled = sblock.fs_rolled;
altsblock.fs_time = sblock.fs_time;
altsblock.fs_state = sblock.fs_state;
altsblock.fs_cstotal = sblock.fs_cstotal;
altsblock.fs_cgrotor = sblock.fs_cgrotor;
altsblock.fs_fmod = sblock.fs_fmod;
altsblock.fs_clean = sblock.fs_clean;
altsblock.fs_ronly = sblock.fs_ronly;
altsblock.fs_flags = sblock.fs_flags;
altsblock.fs_maxcontig = sblock.fs_maxcontig;
altsblock.fs_minfree = sblock.fs_minfree;
altsblock.fs_optim = sblock.fs_optim;
altsblock.fs_rotdelay = sblock.fs_rotdelay;
altsblock.fs_maxbpg = sblock.fs_maxbpg;
altsblock.fs_logbno = sblock.fs_logbno;
altsblock.fs_reclaim = sblock.fs_reclaim;
altsblock.fs_si = sblock.fs_si;
(void) memmove((void *)altsblock.fs_fsmnt, (void *)sblock.fs_fsmnt,
sizeof (sblock.fs_fsmnt));
/*
* The following should not have to be copied.
*/
(void) memmove((void *)altsblock.fs_u.fs_csp_pad,
(void *)sblock.fs_u.fs_csp_pad, sizeof (sblock.fs_u.fs_csp_pad));
altsblock.fs_fsbtodb = sblock.fs_fsbtodb;
altsblock.fs_npsect = sblock.fs_npsect;
altsblock.fs_nrpos = sblock.fs_nrpos;
if (memcmp((void *)&sblock, (void *)&altsblock,
(size_t)sblock.fs_sbsize) != 0) {
err = "BAD VALUES IN SUPER BLOCK";
goto failedsb;
}
havesb = 1;
return (1);
failedsb:
badsb(listerr, err);
return (0);
}
/*
* Indirect blocks are now on the vnode for the file. They are given negative
* logical block numbers. Indirect blocks are addressed by the negative
* address of the first data block to which they point. Double indirect blocks
* are addressed by one less than the address of the first indirect block to
* which they point. Triple indirect blocks are addressed by one less than
* the address of the first double indirect block to which they point.
*
* ufs_bmaparray does the bmap conversion, and if requested returns the
* array of logical blocks which must be traversed to get to a block.
* Each entry contains the offset into that block that gets you to the
* next block and the disk address of the block (if it is assigned).
*/
int
ufs_bmaparray(struct vnode *vp, daddr64_t bn, daddr64_t *bnp, struct indir *ap,
int *nump, int *runp)
{
struct inode *ip;
struct buf *bp;
struct ufsmount *ump;
struct mount *mp;
struct vnode *devvp;
struct indir a[NIADDR+1], *xap;
daddr64_t daddr, metalbn;
int error, maxrun = 0, num;
ip = VTOI(vp);
mp = vp->v_mount;
ump = VFSTOUFS(mp);
#ifdef DIAGNOSTIC
if ((ap != NULL && nump == NULL) || (ap == NULL && nump != NULL))
panic("ufs_bmaparray: invalid arguments");
#endif
if (runp) {
/*
* XXX
* If MAXBSIZE is the largest transfer the disks can handle,
* we probably want maxrun to be 1 block less so that we
* don't create a block larger than the device can handle.
*/
*runp = 0;
maxrun = MAXBSIZE / mp->mnt_stat.f_iosize - 1;
}
xap = ap == NULL ? a : ap;
if (!nump)
nump = #
if ((error = ufs_getlbns(vp, bn, xap, nump)) != 0)
return (error);
num = *nump;
if (num == 0) {
*bnp = blkptrtodb(ump, DIP(ip, db[bn]));
if (*bnp == 0)
*bnp = -1;
else if (runp)
for (++bn; bn < NDADDR && *runp < maxrun &&
is_sequential(ump, DIP(ip, db[bn - 1]),
DIP(ip, db[bn]));
++bn, ++*runp);
return (0);
}
/* Get disk address out of indirect block array */
daddr = DIP(ip, ib[xap->in_off]);
devvp = VFSTOUFS(vp->v_mount)->um_devvp;
for (bp = NULL, ++xap; --num; ++xap) {
/*
* Exit the loop if there is no disk address assigned yet and
* the indirect block isn't in the cache, or if we were
* looking for an indirect block and we've found it.
*/
metalbn = xap->in_lbn;
if ((daddr == 0 && !incore(vp, metalbn)) || metalbn == bn)
break;
/*
* If we get here, we've either got the block in the cache
* or we have a disk address for it, go fetch it.
*/
if (bp)
brelse(bp);
xap->in_exists = 1;
bp = getblk(vp, metalbn, mp->mnt_stat.f_iosize, 0, 0);
if (bp->b_flags & (B_DONE | B_DELWRI)) {
;
}
#ifdef DIAGNOSTIC
else if (!daddr)
panic("ufs_bmaparray: indirect block not in cache");
#endif
else {
bp->b_blkno = blkptrtodb(ump, daddr);
bp->b_flags |= B_READ;
bcstats.pendingreads++;
bcstats.numreads++;
VOP_STRATEGY(bp);
curproc->p_ru.ru_inblock++; /* XXX */
if ((error = biowait(bp)) != 0) {
brelse(bp);
return (error);
}
}
#ifdef FFS2
if (ip->i_ump->um_fstype == UM_UFS2) {
daddr = ((int64_t *)bp->b_data)[xap->in_off];
if (num == 1 && daddr && runp)
for (bn = xap->in_off + 1;
bn < MNINDIR(ump) && *runp < maxrun &&
is_sequential(ump,
((int64_t *)bp->b_data)[bn - 1],
((int64_t *)bp->b_data)[bn]);
++bn, ++*runp);
continue;
}
#endif /* FFS2 */
daddr = ((int32_t *)bp->b_data)[xap->in_off];
if (num == 1 && daddr && runp)
for (bn = xap->in_off + 1;
bn < MNINDIR(ump) && *runp < maxrun &&
is_sequential(ump,
((int32_t *)bp->b_data)[bn - 1],
((int32_t *)bp->b_data)[bn]);
++bn, ++*runp);
}
if (bp)
brelse(bp);
daddr = blkptrtodb(ump, daddr);
*bnp = daddr == 0 ? -1 : daddr;
return (0);
}
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS2. Above is
* the allocation strategy for UFS1.
*/
int
ffs_balloc_ufs2(vnode *vp, off_t startoffset, int size,
Ucred *cred, int flags, Buf **bpp)
{
int error = 0;
print("HARVEY TODO: %s\n", __func__);
#if 0
struct inode *ip;
struct ufs2_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[UFS_NIADDR + 2];
ufs2_daddr_t nb, newb, *bap, pref;
ufs2_daddr_t *allocib, *blkp, *allocblk, allociblk[UFS_NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[UFS_NIADDR + 1];
int deallocated, osize, nsize, num, i, error;
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int gbflags, reclaimed;
ip = VTOI(vp);
dp = ip->i_din2;
fs = ITOFS(ip);
ump = ITOUMP(ip);
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs2: blk too big");
*bpp = nil;
if (lbn < 0)
return (EFBIG);
gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0;
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* Check for allocating external data.
*/
if (flags & IO_EXT) {
if (lbn >= UFS_NXADDR)
return (EFBIG);
/*
* If the next write will extend the data into a new block,
* and the data is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, dp->di_extsize);
if (lastlbn < lbn) {
nb = lastlbn;
osize = sblksize(fs, dp->di_extsize, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, -1 - nb,
dp->di_extb[nb],
ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
&dp->di_extb[0]), osize,
(int)fs->fs_bsize, flags, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, nb,
dbtofsb(fs, bp->b_blkno),
dp->di_extb[nb],
fs->fs_bsize, osize, bp);
dp->di_extsize = smalllblktosize(fs, nb + 1);
dp->di_extb[nb] = dbtofsb(fs, bp->b_blkno);
bp->b_xflags |= BX_ALTDATA;
ip->i_flag |= IN_CHANGE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* All blocks are direct blocks
*/
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs2: BA_METAONLY for ext block");
nb = dp->di_extb[lbn];
if (nb != 0 && dp->di_extsize >= smalllblktosize(fs, lbn + 1)) {
error = bread_gb(vp, -1 - lbn, fs->fs_bsize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
bp->b_xflags |= BX_ALTDATA;
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, dp->di_extsize));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread_gb(vp, -1 - lbn, osize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
bp->b_xflags |= BX_ALTDATA;
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, -1 - lbn,
dp->di_extb[lbn],
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_extb[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
bp->b_xflags |= BX_ALTDATA;
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (dp->di_extsize < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs2(ip, lbn, (int)lbn, &dp->di_extb[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, -1 - lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
bp->b_xflags |= BX_ALTDATA;
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocext(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_extb[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE;
*bpp = bp;
return (0);
}
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < UFS_NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs2(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize,
flags, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno),
dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first UFS_NDADDR blocks are direct blocks
*/
if (lbn < UFS_NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs2: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread_gb(vp, lbn, fs->fs_bsize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread_gb(vp, lbn, osize, NOCRED,
gbflags, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs2(ip, lbn, (int)lbn,
&dp->di_db[0]), nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, gbflags);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs2: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = nil;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs2(ip, lbn, -indirs[0].in_off - 1,
(ufs2_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
pref = newb + fs->fs_frag;
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0,
GB_UNMAPPED);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip,
UFS_NDADDR + indirs[0].in_off, newb, 0,
fs->fs_bsize, 0, bp);
bdwrite(bp);
} else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
} else {
if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs2_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
/*
* If parent indirect has just been allocated, try to cluster
* immediately following it.
*/
if (pref == 0)
pref = ffs_blkpref_ufs2(ip, lbn, i - num - 1,
(ufs2_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
pref = newb + fs->fs_frag;
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0,
GB_UNMAPPED);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else if ((flags & IO_SYNC) == 0 && DOINGASYNC(vp)) {
if (nbp->b_bufsize == fs->fs_bsize)
nbp->b_flags |= B_CLUSTEROK;
bdwrite(nbp);
} else {
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == nil && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
/*
* If allocating metadata at the front of the cylinder
* group and parent indirect block has just been allocated,
* then cluster next to it if it is the first indirect in
* the file. Otherwise it has been allocated in the metadata
* area, so we want to find our own place out in the data area.
*/
if (pref == 0 || (lbn > UFS_NDADDR && fs->fs_metaspace != 0))
pref = ffs_blkpref_ufs2(ip, lbn, indirs[i].in_off,
&bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (DOINGSOFTDEP(vp) && ++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
MPASS(allocblk < allociblk + nitems(allociblk));
MPASS(lbns_remfree < lbns + nitems(lbns));
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, gbflags);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
/*
* If requested clear invalid portions of the buffer. If we
* have to do a read-before-write (typical if BA_CLRBUF is set),
* try to do some read-ahead in the sequential case to reduce
* the number of I/O transactions.
*/
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount != 0 &&
(vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0 &&
!(vm_page_count_severe() || buf_dirty_count_severe())) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, gbflags, &nbp);
} else {
error = bread_gb(vp, lbn, (int)fs->fs_bsize,
NOCRED, gbflags, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
/*
* Balloc defines the structure of file system storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
*/
int
ext2fs_balloc(struct inode *ip, daddr_t bn, int size,
kauth_cred_t cred, struct buf **bpp, int flags)
{
struct m_ext2fs *fs;
daddr_t nb;
struct buf *bp, *nbp;
struct vnode *vp = ITOV(ip);
struct indir indirs[EXT2FS_NIADDR + 2];
daddr_t newb, lbn, pref;
int32_t *bap; /* XXX ondisk32 */
int num, i, error;
u_int deallocated;
daddr_t *blkp, *allocblk, allociblk[EXT2FS_NIADDR + 1];
int32_t *allocib; /* XXX ondisk32 */
int unwindidx = -1;
UVMHIST_FUNC("ext2fs_balloc"); UVMHIST_CALLED(ubchist);
UVMHIST_LOG(ubchist, "bn 0x%x", bn,0,0,0);
if (bpp != NULL) {
*bpp = NULL;
}
if (bn < 0)
return (EFBIG);
fs = ip->i_e2fs;
lbn = bn;
/*
* The first EXT2FS_NDADDR blocks are direct blocks
*/
if (bn < EXT2FS_NDADDR) {
/* XXX ondisk32 */
nb = fs2h32(ip->i_e2fs_blocks[bn]);
if (nb != 0) {
/*
* the block is already allocated, just read it.
*/
if (bpp != NULL) {
error = bread(vp, bn, fs->e2fs_bsize, NOCRED,
B_MODIFY, &bp);
if (error) {
return (error);
}
*bpp = bp;
}
return (0);
}
/*
* allocate a new direct block.
*/
error = ext2fs_alloc(ip, bn,
ext2fs_blkpref(ip, bn, bn, &ip->i_e2fs_blocks[0]),
cred, &newb);
if (error)
return (error);
ip->i_e2fs_last_lblk = lbn;
ip->i_e2fs_last_blk = newb;
/* XXX ondisk32 */
ip->i_e2fs_blocks[bn] = h2fs32((int32_t)newb);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp != NULL) {
bp = getblk(vp, bn, fs->e2fs_bsize, 0, 0);
bp->b_blkno = EXT2_FSBTODB(fs, newb);
if (flags & B_CLRBUF)
clrbuf(bp);
*bpp = bp;
}
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, bn, indirs, &num)) != 0)
return(error);
#ifdef DIAGNOSTIC
if (num < 1)
panic ("ext2fs_balloc: ufs_getlbns returned indirect block\n");
#endif
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
/* XXX ondisk32 */
nb = fs2h32(ip->i_e2fs_blocks[EXT2FS_NDADDR + indirs[0].in_off]);
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ext2fs_blkpref(ip, lbn, 0, (int32_t *)0);
error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
if (error)
return (error);
nb = newb;
*allocblk++ = nb;
ip->i_e2fs_last_blk = newb;
bp = getblk(vp, indirs[1].in_lbn, fs->e2fs_bsize, 0, 0);
bp->b_blkno = EXT2_FSBTODB(fs, newb);
clrbuf(bp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(bp)) != 0)
goto fail;
unwindidx = 0;
allocib = &ip->i_e2fs_blocks[EXT2FS_NDADDR + indirs[0].in_off];
/* XXX ondisk32 */
*allocib = h2fs32((int32_t)newb);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->e2fs_bsize, NOCRED, 0, &bp);
if (error) {
goto fail;
}
bap = (int32_t *)bp->b_data; /* XXX ondisk32 */
nb = fs2h32(bap[indirs[i].in_off]);
if (i == num)
break;
i++;
if (nb != 0) {
brelse(bp, 0);
continue;
}
pref = ext2fs_blkpref(ip, lbn, 0, (int32_t *)0);
error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
if (error) {
brelse(bp, 0);
goto fail;
}
nb = newb;
*allocblk++ = nb;
ip->i_e2fs_last_blk = newb;
nbp = getblk(vp, indirs[i].in_lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = EXT2_FSBTODB(fs, nb);
clrbuf(nbp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp, 0);
goto fail;
}
if (unwindidx < 0)
unwindidx = i - 1;
/* XXX ondisk32 */
bap[indirs[i - 1].in_off] = h2fs32((int32_t)nb);
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
pref = ext2fs_blkpref(ip, lbn, indirs[num].in_off, &bap[0]);
error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
if (error) {
brelse(bp, 0);
goto fail;
}
nb = newb;
*allocblk++ = nb;
ip->i_e2fs_last_lblk = lbn;
ip->i_e2fs_last_blk = newb;
/* XXX ondisk32 */
bap[indirs[num].in_off] = h2fs32((int32_t)nb);
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
if (bpp != NULL) {
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = EXT2_FSBTODB(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
*bpp = nbp;
}
return (0);
}
brelse(bp, 0);
if (bpp != NULL) {
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->e2fs_bsize, NOCRED,
B_MODIFY, &nbp);
if (error) {
goto fail;
}
} else {
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
nbp->b_blkno = EXT2_FSBTODB(fs, nb);
}
*bpp = nbp;
}
return (0);
fail:
/*
* If we have failed part way through block allocation, we
* have to deallocate any indirect blocks that we have allocated.
*/
for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
ext2fs_blkfree(ip, *blkp);
deallocated += fs->e2fs_bsize;
}
if (unwindidx >= 0) {
if (unwindidx == 0) {
*allocib = 0;
} else {
int r;
r = bread(vp, indirs[unwindidx].in_lbn,
(int)fs->e2fs_bsize, NOCRED, B_MODIFY, &bp);
if (r) {
panic("Could not unwind indirect block, error %d", r);
} else {
bap = (int32_t *)bp->b_data; /* XXX ondisk32 */
bap[indirs[unwindidx].in_off] = 0;
if (flags & B_SYNC)
bwrite(bp);
else
bdwrite(bp);
}
}
for (i = unwindidx + 1; i <= num; i++) {
bp = getblk(vp, indirs[i].in_lbn, (int)fs->e2fs_bsize,
0, 0);
brelse(bp, BC_INVAL);
}
}
if (deallocated) {
ext2fs_setnblock(ip, ext2fs_nblock(ip) - btodb(deallocated));
ip->i_e2fs_flags |= IN_CHANGE | IN_UPDATE;
}
return error;
}
static int
prclin(void) /* Build up line; return 1 for eof */
{
uint8_t *pbptr; /* -> xxprev->bdata[bchars] for speed */
int inindent = 040000; /* Inside indenting w/s, mask matches mode */
/* */
for (pbchars = linpos = 0, pbptr = xxprev->bdata; pbchars < BUFMAX; linpos++)
{
if (state != 0 && state != 1)
{ /* Else assume state 2 */
if (!(linpos % tabsiz)) /* Reached tabstop (multiple of 8) */
state = 0; /* Fall through */
else
goto endswitch; /* switch(state) case 2 */
} /* if(state!=0&&state!=1) */
if (!bytes && getblk()) /* Eof, getblk writes data */
return 1;
thisch = *bufpos++;
bytes--;
/* When reading in binary, try to guess good places to break "lines"
* (which are arbitrary anyway) unless requested not to (for w/f to use
* the least amount of memory). (LATER)
* We also try not to break strings of printable characters. */
if (binary)
{
if (thisch == '\n' || (thisch == 0 && pbchars > BLKCAP / 2) ||
(pbchars >= BLKCAP - 2 && (prevch < ' ' || prevch > '\176')))
{
*pbptr++ = prevch = thisch;
pbchars++; /* Store thisch */
return 0; /* Finish this line */
} /* if(thisch=='\n'||(thisch==0&&... */
} /* if(binary) */
else
/* For Ascii files, now process significant characters: Newline, Return & Tab */
{
if (inindent && thisch != '\t' && thisch != ' ')
inindent = 0;
if (thisch < ' ' || state) /* One of the following tests might pass */
{
if (thisch == '\r')
{
/* If doing a DOS read and already holding on to a Cr, store it and stay in
* state 1. Else, move to state 1 and store nothing */
if (xxmode & 1 && state != 1) /* DOS read, not holding Cr */
{
state = 1; /* Holding a Cr */
continue; /* for(pbchars=linpos... (don't store Cr) */
} /* if(xxmode&1) */
linpos = -1; /* Back at line start */
goto endswitch; /* switch(state) cases 0 & 1 */
} /* if(thisch=='\r') */
if (thisch == '\n')
{
if (state == 1) /* Holding Cr */
state = 0; /* Discard Cr */
if (state != 0) /* Unexpected */
{
visbel();
printf("\r\nUnexpected state %d at Nl\r\n", state);
} /* if (state != 0) */
return 0; /* Have line */
} /* if(thisch=='\n') */
else if (state == 1) /* Holding on to Cr 2B stored now */
{
bytes++, bufpos--; /* Put back char just read */
state = 0;
thisch = '\r';
linpos = -1; /* Back at line start */
} /* else if(state==1) */
/* Only check for tab character after return has been checked for */
if (thisch == '\t' || thisch == '\b')
{
/* +tr || +l in ldg w/s */
if ((xxmode & 4) || (xxmode & 040000 & inindent))
{
if (thisch == '\t')
{
thisch = ' '; /* Store spaces */
state = 2; /* Until on next 8-char boundary */
} /* if(thisch=='\t' */
else if (linpos > 0) /* thisch must be '\b' */
linpos -= 2;
} /* if(xxmode&4) */
} /* if(thisch=='\t'||thisch=='\b') */
} /* if(thisch<' '||state) */
} /* if(binary) else */
endswitch:
*pbptr++ = prevch = thisch;
pbchars++;
} /* for(pbchars=linpos=0,pbptr=xxprev->bdata;pbchars<BUFMAX;linpos++) */
/*
* Dropped out of loop: we have a full line. If the file is binary, this
* is the norm, just write away the chars read.
* Otherwise warn user, drain chars until EOL...
*/
state = 0; /* No longer holding on to Cr */
if (!binary && lngwrn)
{
/* If the very next ch is Nl, we haven't actually misssed anything, */
/* so defer outputting a warning here. */
bool tried_next_ch = false;
for (;;)
{
if (!bytes && getblk())
{
if (!tried_next_ch)
puts("Long line found!\r");
return 1;
} /* if (!bytes && getblk()) */
thisch = *bufpos++;
bytes--;
if (thisch == '\n') /* Eol */
break;
if (!tried_next_ch)
{
tried_next_ch = true;
puts("Long line found!\r");
} /* if (!tried_next_ch) */
} /* for(;;) */
} /* if(!binary&&lngwrn) */
return 0;
} /* static int prclin(void) */
errcode_t journal_find_head(journal_t *journal)
{
unsigned int next_commit_ID;
blk64_t next_log_block, head_block;
int err;
journal_superblock_t *sb;
journal_header_t *tmp;
struct buffer_head *bh;
unsigned int sequence;
int blocktype;
/*
* First thing is to establish what we expect to find in the log
* (in terms of transaction IDs), and where (in terms of log
* block offsets): query the superblock.
*/
sb = journal->j_superblock;
next_commit_ID = ext2fs_be32_to_cpu(sb->s_sequence);
next_log_block = ext2fs_be32_to_cpu(sb->s_start);
head_block = next_log_block;
if (next_log_block == 0)
return 0;
bh = getblk(journal->j_dev, 0, journal->j_blocksize);
if (bh == NULL)
return ENOMEM;
/*
* Now we walk through the log, transaction by transaction,
* making sure that each transaction has a commit block in the
* expected place. Each complete transaction gets replayed back
* into the main filesystem.
*/
while (1) {
dbg_printf("Scanning for sequence ID %u at %lu/%lu\n",
next_commit_ID, (unsigned long)next_log_block,
journal->j_last);
/* Skip over each chunk of the transaction looking
* either the next descriptor block or the final commit
* record. */
err = journal_bmap(journal, next_log_block, &bh->b_blocknr);
if (err)
goto err;
mark_buffer_uptodate(bh, 0);
ll_rw_block(READ, 1, &bh);
err = bh->b_err;
if (err)
goto err;
next_log_block++;
wrap(journal, next_log_block);
/* What kind of buffer is it?
*
* If it is a descriptor block, check that it has the
* expected sequence number. Otherwise, we're all done
* here. */
tmp = (journal_header_t *)bh->b_data;
if (tmp->h_magic != ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER)) {
dbg_printf("JBD2: wrong magic 0x%x\n", tmp->h_magic);
goto err;
}
blocktype = ext2fs_be32_to_cpu(tmp->h_blocktype);
sequence = ext2fs_be32_to_cpu(tmp->h_sequence);
dbg_printf("Found magic %d, sequence %d\n",
blocktype, sequence);
if (sequence != next_commit_ID) {
dbg_printf("JBD2: Wrong sequence %d (wanted %d)\n",
sequence, next_commit_ID);
goto err;
}
/* OK, we have a valid descriptor block which matches
* all of the sequence number checks. What are we going
* to do with it? That depends on the pass... */
switch (blocktype) {
case JFS_DESCRIPTOR_BLOCK:
next_log_block += count_tags(journal, bh->b_data);
wrap(journal, next_log_block);
continue;
case JFS_COMMIT_BLOCK:
head_block = next_log_block;
next_commit_ID++;
continue;
case JFS_REVOKE_BLOCK:
continue;
default:
dbg_printf("Unrecognised magic %d, end of scan.\n",
blocktype);
err = -EINVAL;
goto err;
}
}
err:
if (err == 0) {
dbg_printf("head seq=%d blk=%llu\n", next_commit_ID,
head_block);
journal->j_transaction_sequence = next_commit_ID;
journal->j_head = head_block;
}
brelse(bh);
return err;
}
static errcode_t journal_commit_trans(journal_transaction_t *trans)
{
struct buffer_head *bh, *cbh = NULL;
struct commit_header *commit;
#ifdef HAVE_SYS_TIME_H
struct timeval tv;
#endif
errcode_t err;
JOURNAL_CHECK_TRANS_MAGIC(trans);
if ((trans->flags & J_TRANS_COMMITTED) ||
!(trans->flags & J_TRANS_OPEN))
return EXT2_ET_INVALID_ARGUMENT;
bh = getblk(trans->journal->j_dev, 0, trans->journal->j_blocksize);
if (bh == NULL)
return ENOMEM;
/* write the descriptor block header */
commit = (struct commit_header *)bh->b_data;
commit->h_magic = ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER);
commit->h_blocktype = ext2fs_cpu_to_be32(JFS_COMMIT_BLOCK);
commit->h_sequence = ext2fs_cpu_to_be32(trans->tid);
if (JFS_HAS_COMPAT_FEATURE(trans->journal,
JFS_FEATURE_COMPAT_CHECKSUM)) {
__u32 csum_v1 = ~0;
blk64_t cblk;
cbh = getblk(trans->journal->j_dev, 0,
trans->journal->j_blocksize);
if (cbh == NULL) {
err = ENOMEM;
goto error;
}
for (cblk = trans->start; cblk < trans->block; cblk++) {
err = journal_bmap(trans->journal, cblk,
&cbh->b_blocknr);
if (err)
goto error;
mark_buffer_uptodate(cbh, 0);
ll_rw_block(READ, 1, &cbh);
err = cbh->b_err;
if (err)
goto error;
csum_v1 = ext2fs_crc32_be(csum_v1,
(unsigned char const *)cbh->b_data,
cbh->b_size);
}
commit->h_chksum_type = JFS_CRC32_CHKSUM;
commit->h_chksum_size = JFS_CRC32_CHKSUM_SIZE;
commit->h_chksum[0] = ext2fs_cpu_to_be32(csum_v1);
} else {
commit->h_chksum_type = 0;
commit->h_chksum_size = 0;
commit->h_chksum[0] = 0;
}
#ifdef HAVE_SYS_TIME_H
gettimeofday(&tv, NULL);
commit->h_commit_sec = ext2fs_cpu_to_be32(tv.tv_sec);
commit->h_commit_nsec = ext2fs_cpu_to_be32(tv.tv_usec * 1000);
#else
commit->h_commit_sec = 0;
commit->h_commit_nsec = 0;
#endif
/* Write block */
jbd2_commit_block_csum_set(trans->journal, bh);
err = journal_bmap(trans->journal, trans->block, &bh->b_blocknr);
if (err)
goto error;
dbg_printf("Writing commit block at %llu:%llu\n", trans->block,
bh->b_blocknr);
mark_buffer_dirty(bh);
ll_rw_block(WRITE, 1, &bh);
err = bh->b_err;
if (err)
goto error;
trans->flags |= J_TRANS_COMMITTED;
trans->flags &= ~J_TRANS_OPEN;
trans->block++;
trans->fs->super->s_feature_incompat |= EXT3_FEATURE_INCOMPAT_RECOVER;
ext2fs_mark_super_dirty(trans->fs);
error:
if (cbh)
brelse(cbh);
brelse(bh);
return err;
}
static errcode_t journal_add_revoke_to_trans(journal_transaction_t *trans,
blk64_t *revoke_list,
size_t revoke_len)
{
journal_revoke_header_t *jrb;
void *buf;
size_t i, offset;
blk64_t curr_blk;
int sz, csum_size = 0;
struct buffer_head *bh;
errcode_t err;
JOURNAL_CHECK_TRANS_MAGIC(trans);
if ((trans->flags & J_TRANS_COMMITTED) ||
!(trans->flags & J_TRANS_OPEN))
return EXT2_ET_INVALID_ARGUMENT;
if (revoke_len == 0)
return 0;
/* Do we need to leave space at the end for a checksum? */
if (journal_has_csum_v2or3(trans->journal))
csum_size = sizeof(struct journal_revoke_tail);
curr_blk = trans->block;
bh = getblk(trans->journal->j_dev, curr_blk,
trans->journal->j_blocksize);
if (bh == NULL)
return ENOMEM;
jrb = buf = bh->b_data;
jrb->r_header.h_magic = ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER);
jrb->r_header.h_blocktype = ext2fs_cpu_to_be32(JFS_REVOKE_BLOCK);
jrb->r_header.h_sequence = ext2fs_cpu_to_be32(trans->tid);
offset = sizeof(*jrb);
if (JFS_HAS_INCOMPAT_FEATURE(trans->journal,
JFS_FEATURE_INCOMPAT_64BIT))
sz = 8;
else
sz = 4;
for (i = 0; i < revoke_len; i++) {
/* Block full, write to journal */
if (offset + sz > trans->journal->j_blocksize - csum_size) {
jrb->r_count = ext2fs_cpu_to_be32(offset);
jbd2_revoke_csum_set(trans->journal, bh);
err = journal_bmap(trans->journal, curr_blk,
&bh->b_blocknr);
if (err)
goto error;
dbg_printf("Writing revoke block at %llu:%llu\n",
curr_blk, bh->b_blocknr);
mark_buffer_dirty(bh);
ll_rw_block(WRITE, 1, &bh);
err = bh->b_err;
if (err)
goto error;
offset = sizeof(*jrb);
curr_blk++;
}
if (revoke_list[i] >=
ext2fs_blocks_count(trans->journal->j_fs_dev->k_fs->super)) {
err = EXT2_ET_BAD_BLOCK_NUM;
goto error;
}
if (JFS_HAS_INCOMPAT_FEATURE(trans->journal,
JFS_FEATURE_INCOMPAT_64BIT))
*((__u64 *)(&((char *)buf)[offset])) =
ext2fs_cpu_to_be64(revoke_list[i]);
else
*((__u32 *)(&((char *)buf)[offset])) =
ext2fs_cpu_to_be32(revoke_list[i]);
offset += sz;
}
if (offset > 0) {
jrb->r_count = ext2fs_cpu_to_be32(offset);
jbd2_revoke_csum_set(trans->journal, bh);
err = journal_bmap(trans->journal, curr_blk, &bh->b_blocknr);
if (err)
goto error;
dbg_printf("Writing revoke block at %llu:%llu\n",
curr_blk, bh->b_blocknr);
mark_buffer_dirty(bh);
ll_rw_block(WRITE, 1, &bh);
err = bh->b_err;
if (err)
goto error;
curr_blk++;
}
error:
trans->block = curr_blk;
brelse(bh);
return err;
}
static errcode_t journal_add_blocks_to_trans(journal_transaction_t *trans,
blk64_t *block_list, size_t block_len,
FILE *fp)
{
blk64_t curr_blk, jdb_blk;
size_t i, j;
int csum_size = 0;
journal_header_t *jdb;
journal_block_tag_t *jdbt;
int tag_bytes;
void *buf = NULL, *jdb_buf = NULL;
struct buffer_head *bh = NULL, *data_bh;
errcode_t err;
JOURNAL_CHECK_TRANS_MAGIC(trans);
if ((trans->flags & J_TRANS_COMMITTED) ||
!(trans->flags & J_TRANS_OPEN))
return EXT2_ET_INVALID_ARGUMENT;
if (block_len == 0)
return 0;
/* Do we need to leave space at the end for a checksum? */
if (journal_has_csum_v2or3(trans->journal))
csum_size = sizeof(struct journal_block_tail);
curr_blk = jdb_blk = trans->block;
data_bh = getblk(trans->journal->j_dev, curr_blk,
trans->journal->j_blocksize);
if (data_bh == NULL)
return ENOMEM;
buf = data_bh->b_data;
/* write the descriptor block header */
bh = getblk(trans->journal->j_dev, curr_blk,
trans->journal->j_blocksize);
if (bh == NULL) {
err = ENOMEM;
goto error;
}
jdb = jdb_buf = bh->b_data;
jdb->h_magic = ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER);
jdb->h_blocktype = ext2fs_cpu_to_be32(JFS_DESCRIPTOR_BLOCK);
jdb->h_sequence = ext2fs_cpu_to_be32(trans->tid);
jdbt = (journal_block_tag_t *)(jdb + 1);
curr_blk++;
for (i = 0; i < block_len; i++) {
j = fread(data_bh->b_data, trans->journal->j_blocksize, 1, fp);
if (j != 1) {
err = errno;
goto error;
}
tag_bytes = journal_tag_bytes(trans->journal);
/* No space left in descriptor block, write it out */
if ((char *)jdbt + tag_bytes >
(char *)jdb_buf + trans->journal->j_blocksize - csum_size) {
jbd2_descr_block_csum_set(trans->journal, bh);
err = journal_bmap(trans->journal, jdb_blk,
&bh->b_blocknr);
if (err)
goto error;
dbg_printf("Writing descriptor block at %llu:%llu\n",
jdb_blk, bh->b_blocknr);
mark_buffer_dirty(bh);
ll_rw_block(WRITE, 1, &bh);
err = bh->b_err;
if (err)
goto error;
jdbt = (journal_block_tag_t *)(jdb + 1);
jdb_blk = curr_blk;
curr_blk++;
}
if (block_list[i] >=
ext2fs_blocks_count(trans->journal->j_fs_dev->k_fs->super)) {
err = EXT2_ET_BAD_BLOCK_NUM;
goto error;
}
/* Fill out the block tag */
jdbt->t_blocknr = ext2fs_cpu_to_be32(block_list[i] & 0xFFFFFFFF);
jdbt->t_flags = 0;
if (jdbt != (journal_block_tag_t *)(jdb + 1))
jdbt->t_flags |= ext2fs_cpu_to_be16(JFS_FLAG_SAME_UUID);
else {
memcpy(jdbt + tag_bytes,
trans->journal->j_superblock->s_uuid,
sizeof(trans->journal->j_superblock->s_uuid));
tag_bytes += 16;
}
if (i == block_len - 1)
jdbt->t_flags |= ext2fs_cpu_to_be16(JFS_FLAG_LAST_TAG);
if (*((__u32 *)buf) == ext2fs_cpu_to_be32(JFS_MAGIC_NUMBER)) {
*((__u32 *)buf) = 0;
jdbt->t_flags |= ext2fs_cpu_to_be16(JFS_FLAG_ESCAPE);
}
if (JFS_HAS_INCOMPAT_FEATURE(trans->journal,
JFS_FEATURE_INCOMPAT_64BIT))
jdbt->t_blocknr_high = ext2fs_cpu_to_be32(block_list[i] >> 32);
jbd2_block_tag_csum_set(trans->journal, jdbt, data_bh,
trans->tid);
/* Write the data block */
err = journal_bmap(trans->journal, curr_blk,
&data_bh->b_blocknr);
if (err)
goto error;
dbg_printf("Writing data block %llu at %llu:%llu tag %d\n",
block_list[i], curr_blk, data_bh->b_blocknr,
tag_bytes);
mark_buffer_dirty(data_bh);
ll_rw_block(WRITE, 1, &data_bh);
err = data_bh->b_err;
if (err)
goto error;
curr_blk++;
jdbt = (journal_block_tag_t *)(((char *)jdbt) + tag_bytes);
}
/* Write out the last descriptor block */
if (jdbt != (journal_block_tag_t *)(jdb + 1)) {
jbd2_descr_block_csum_set(trans->journal, bh);
err = journal_bmap(trans->journal, jdb_blk, &bh->b_blocknr);
if (err)
goto error;
dbg_printf("Writing descriptor block at %llu:%llu\n",
jdb_blk, bh->b_blocknr);
mark_buffer_dirty(bh);
ll_rw_block(WRITE, 1, &bh);
err = bh->b_err;
if (err)
goto error;
}
error:
trans->block = curr_blk;
if (bh)
brelse(bh);
brelse(data_bh);
return err;
}
int
xfs_write_file(xfs_inode_t *xip, struct uio *uio, int ioflag)
{
struct buf *bp;
//struct thread *td;
daddr_t lbn;
off_t osize = 0;
off_t offset= 0;
int blkoffset, error, resid, xfersize;
int fsblocksize;
int seqcount;
xfs_iomap_t iomap;
int maps = 1;
xfs_vnode_t *xvp = XFS_ITOV(xip);
struct vnode *vp = xvp->v_vnode;
xfs_mount_t *mp = (&xip->i_iocore)->io_mount;
seqcount = ioflag >> IO_SEQSHIFT;
memset(&iomap,0,sizeof(xfs_iomap_t));
/*
* Maybe this should be above the vnode op call, but so long as
* file servers have no limits, I don't think it matters.
*/
#if 0
td = uio->uio_td;
if (vn_rlimit_fsize(vp, uio, uio->uio_td))
return (EFBIG);
#endif
resid = uio->uio_resid;
offset = uio->uio_offset;
osize = xip->i_d.di_size;
/* xfs bmap wants bytes for both offset and size */
XVOP_BMAP(xvp,
uio->uio_offset,
uio->uio_resid,
BMAPI_WRITE|BMAPI_DIRECT,
&iomap, &maps, error);
if(error) {
printf("XVOP_BMAP failed\n");
goto error;
}
for (error = 0; uio->uio_resid > 0;) {
lbn = XFS_B_TO_FSBT(mp, offset);
blkoffset = XFS_B_FSB_OFFSET(mp, offset);
xfersize = mp->m_sb.sb_blocksize - blkoffset;
fsblocksize = mp->m_sb.sb_blocksize;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
/*
* getblk sets buf by blkno * bo->bo_bsize
* bo_bsize is set from the mnt point fsize
* so we call getblk in the case using fsblocks
* not basic blocks
*/
bp = getblk(vp, lbn, fsblocksize, 0, 0, 0);
if(!bp) {
printf("getblk failed\n");
error = EINVAL;
break;
}
if (!(bp->b_flags & B_CACHE) && fsblocksize > xfersize)
vfs_bio_clrbuf(bp);
if (offset + xfersize > xip->i_d.di_size) {
xip->i_d.di_size = offset + xfersize;
vnode_pager_setsize(vp, offset + fsblocksize);
}
/* move the offset for the next itteration of the loop */
offset += xfersize;
error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
if ((ioflag & IO_VMIO) &&
(LIST_FIRST(&bp->b_dep) == NULL)) /* in ext2fs? */
bp->b_flags |= B_RELBUF;
/* force to full direct for now */
bp->b_flags |= B_DIRECT;
/* and sync ... the delay path is not pushing data out */
ioflag |= IO_SYNC;
if (ioflag & IO_SYNC) {
(void)bwrite(bp);
} else if (0 /* RMC xfersize + blkoffset == fs->s_frag_size */) {
if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
bp->b_flags |= B_CLUSTEROK;
cluster_write(vp, bp, osize, seqcount);
} else {
bawrite(bp);
}
} else {
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
if (error || xfersize == 0)
break;
}
/*
* If we successfully wrote any data, and we are not the superuser
* we clear the setuid and setgid bits as a precaution against
* tampering.
*/
#if 0
if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
ip->i_mode &= ~(ISUID | ISGID);
#endif
if (error) {
if (ioflag & IO_UNIT) {
#if 0
(void)ext2_truncate(vp, osize,
ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
#endif
uio->uio_offset -= resid - uio->uio_resid;
uio->uio_resid = resid;
}
} else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
/* Update the vnode here? */
}
error:
return error;
}
/*
* Release blocks associated with the inode ip and stored in the indirect
* block bn. Blocks are free'd in LIFO order up to (but not including)
* lastbn. If level is greater than SINGLE, the block is an indirect block
* and recursive calls to indirtrunc must be used to cleanse other indirect
* blocks.
*
* NB: triple indirect blocks are untested.
*/
static int
lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
daddr_t lastbn, int level, daddr_t *countp,
daddr_t *rcountp, long *lastsegp, size_t *bcp)
{
int i;
struct buf *bp;
struct lfs *fs = ip->i_lfs;
int32_t *bap; /* XXX ondisk32 */
struct vnode *vp;
daddr_t nb, nlbn, last;
int32_t *copy = NULL; /* XXX ondisk32 */
daddr_t blkcount, rblkcount, factor;
int nblocks;
daddr_t blocksreleased = 0, real_released = 0;
int error = 0, allerror = 0;
ASSERT_SEGLOCK(fs);
/*
* Calculate index in current block of last
* block to be kept. -1 indicates the entire
* block so we need not calculate the index.
*/
factor = 1;
for (i = SINGLE; i < level; i++)
factor *= LFS_NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
/*
* Get buffer of block pointers, zero those entries corresponding
* to blocks to be free'd, and update on disk copy first. Since
* double(triple) indirect before single(double) indirect, calls
* to bmap on these blocks will fail. However, we already have
* the on disk address, so we have to set the b_blkno field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lbn, lfs_sb_getbsize(fs), 0, 0);
if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
/* Braces must be here in case trace evaluates to nothing. */
trace(TR_BREADHIT, pack(vp, lfs_sb_getbsize(fs)), lbn);
} else {
trace(TR_BREADMISS, pack(vp, lfs_sb_getbsize(fs)), lbn);
curlwp->l_ru.ru_inblock++; /* pay for read */
bp->b_flags |= B_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("lfs_indirtrunc: bad buffer size");
bp->b_blkno = LFS_FSBTODB(fs, dbn);
VOP_STRATEGY(vp, bp);
error = biowait(bp);
}
if (error) {
brelse(bp, 0);
*countp = *rcountp = 0;
return (error);
}
bap = (int32_t *)bp->b_data; /* XXX ondisk32 */
if (lastbn >= 0) {
copy = lfs_malloc(fs, lfs_sb_getbsize(fs), LFS_NB_IBLOCK);
memcpy((void *)copy, (void *)bap, lfs_sb_getbsize(fs));
memset((void *)&bap[last + 1], 0,
/* XXX ondisk32 */
(u_int)(LFS_NINDIR(fs) - (last + 1)) * sizeof (int32_t));
error = VOP_BWRITE(bp->b_vp, bp);
if (error)
allerror = error;
bap = copy;
}
/*
* Recursively free totally unused blocks.
*/
for (i = LFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = bap[i];
if (nb == 0)
continue;
if (level > SINGLE) {
error = lfs_indirtrunc(ip, nlbn, nb,
(daddr_t)-1, level - 1,
&blkcount, &rblkcount,
lastsegp, bcp);
if (error)
allerror = error;
blocksreleased += blkcount;
real_released += rblkcount;
}
lfs_blkfree(fs, ip, nb, lfs_sb_getbsize(fs), lastsegp, bcp);
if (bap[i] > 0)
real_released += nblocks;
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = bap[i];
if (nb != 0) {
error = lfs_indirtrunc(ip, nlbn, nb,
last, level - 1, &blkcount,
&rblkcount, lastsegp, bcp);
if (error)
allerror = error;
real_released += rblkcount;
blocksreleased += blkcount;
}
}
if (copy != NULL) {
lfs_free(fs, copy, LFS_NB_IBLOCK);
} else {
mutex_enter(&bufcache_lock);
if (bp->b_oflags & BO_DELWRI) {
LFS_UNLOCK_BUF(bp);
lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
wakeup(&fs->lfs_availsleep);
}
brelsel(bp, BC_INVAL);
mutex_exit(&bufcache_lock);
}
*countp = blocksreleased;
*rcountp = real_released;
return (allerror);
}
static int trunc_indirect(struct inode * inode, int offset, unsigned long * p)
{
int i, tmp;
struct buffer_head * bh;
struct buffer_head * ind_bh;
unsigned long * ind;
int retry = 0;
#define INDIRECT_BLOCK (DIRECT_BLOCK-offset)
tmp = *p;
if (!tmp)
return 0;
ind_bh = bread(inode->i_dev, tmp, BLOCK_SIZE);
if (tmp != *p) {
brelse(ind_bh);
return 1;
}
if (!ind_bh) {
*p = 0;
return 0;
}
repeat:
for (i = INDIRECT_BLOCK ; i < 256 ; i++) {
if (i < 0)
i = 0;
if (i < INDIRECT_BLOCK)
goto repeat;
ind = i+(unsigned long *) ind_bh->b_data;
tmp = *ind;
if (!tmp)
continue;
bh = getblk(inode->i_dev,tmp,BLOCK_SIZE);
if (i < INDIRECT_BLOCK) {
brelse(bh);
goto repeat;
}
if ((bh && bh->b_count != 1) || tmp != *ind) {
retry = 1;
brelse(bh);
continue;
}
*ind = 0;
mark_buffer_dirty(ind_bh, 1);
brelse(bh);
ext_free_block(inode->i_sb,tmp);
}
ind = (unsigned long *) ind_bh->b_data;
for (i = 0; i < 256; i++)
if (*(ind++))
break;
if (i >= 256)
if (ind_bh->b_count != 1)
retry = 1;
else {
tmp = *p;
*p = 0;
inode->i_dirt = 1;
ext_free_block(inode->i_sb,tmp);
}
brelse(ind_bh);
return retry;
}
static errcode_t ext2fs_get_journal(ext2_filsys fs, journal_t **ret_journal)
{
struct process_block_struct pb;
struct ext2_super_block *sb = fs->super;
struct ext2_super_block jsuper;
struct buffer_head *bh;
struct inode *j_inode = NULL;
struct kdev_s *dev_fs = NULL, *dev_journal;
const char *journal_name = 0;
journal_t *journal = NULL;
errcode_t retval = 0;
io_manager io_ptr = 0;
unsigned long long start = 0;
int ext_journal = 0;
int tried_backup_jnl = 0;
retval = ext2fs_get_memzero(sizeof(journal_t), &journal);
if (retval)
return retval;
retval = ext2fs_get_memzero(2 * sizeof(struct kdev_s), &dev_fs);
if (retval)
goto errout;
dev_journal = dev_fs+1;
dev_fs->k_fs = dev_journal->k_fs = fs;
dev_fs->k_dev = K_DEV_FS;
dev_journal->k_dev = K_DEV_JOURNAL;
journal->j_dev = dev_journal;
journal->j_fs_dev = dev_fs;
journal->j_inode = NULL;
journal->j_blocksize = fs->blocksize;
if (uuid_is_null(sb->s_journal_uuid)) {
if (!sb->s_journal_inum) {
retval = EXT2_ET_BAD_INODE_NUM;
goto errout;
}
retval = ext2fs_get_memzero(sizeof(*j_inode), &j_inode);
if (retval)
goto errout;
j_inode->i_fs = fs;
j_inode->i_ino = sb->s_journal_inum;
retval = ext2fs_read_inode(fs, sb->s_journal_inum,
&j_inode->i_ext2);
if (retval) {
try_backup_journal:
if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
tried_backup_jnl)
goto errout;
memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
EXT2_N_BLOCKS*4);
j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
j_inode->i_ext2.i_links_count = 1;
j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
tried_backup_jnl++;
}
if (!j_inode->i_ext2.i_links_count ||
!LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
retval = EXT2_ET_NO_JOURNAL;
goto try_backup_journal;
}
if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
JFS_MIN_JOURNAL_BLOCKS) {
retval = EXT2_ET_JOURNAL_TOO_SMALL;
goto try_backup_journal;
}
pb.last_block = -1;
retval = ext2fs_block_iterate3(fs, j_inode->i_ino,
BLOCK_FLAG_HOLE, 0,
process_journal_block, &pb);
if ((pb.last_block + 1) * fs->blocksize <
(int) EXT2_I_SIZE(&j_inode->i_ext2)) {
retval = EXT2_ET_JOURNAL_TOO_SMALL;
goto try_backup_journal;
}
if (tried_backup_jnl && (fs->flags & EXT2_FLAG_RW)) {
retval = ext2fs_write_inode(fs, sb->s_journal_inum,
&j_inode->i_ext2);
if (retval)
goto errout;
}
journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
journal->j_blocksize;
#ifdef USE_INODE_IO
retval = ext2fs_inode_io_intern2(fs, sb->s_journal_inum,
&j_inode->i_ext2,
&journal_name);
if (retval)
goto errout;
io_ptr = inode_io_manager;
#else
journal->j_inode = j_inode;
fs->journal_io = fs->io;
retval = (errcode_t)journal_bmap(journal, 0, &start);
if (retval)
goto errout;
#endif
} else {
ext_journal = 1;
if (!fs->journal_name) {
char uuid[37];
blkid_cache blkid;
blkid_get_cache(&blkid, NULL);
uuid_unparse(sb->s_journal_uuid, uuid);
fs->journal_name = blkid_get_devname(blkid,
"UUID", uuid);
if (!fs->journal_name)
fs->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
blkid_put_cache(blkid);
}
journal_name = fs->journal_name;
if (!journal_name) {
retval = EXT2_ET_LOAD_EXT_JOURNAL;
goto errout;
}
jfs_debug(1, "Using journal file %s\n", journal_name);
io_ptr = unix_io_manager;
}
#if 0
test_io_backing_manager = io_ptr;
io_ptr = test_io_manager;
#endif
#ifndef USE_INODE_IO
if (ext_journal)
#endif
{
retval = io_ptr->open(journal_name, fs->flags & EXT2_FLAG_RW,
&fs->journal_io);
}
if (retval)
goto errout;
io_channel_set_blksize(fs->journal_io, fs->blocksize);
if (ext_journal) {
blk64_t maxlen;
start = ext2fs_journal_sb_start(fs->blocksize) - 1;
bh = getblk(dev_journal, start, fs->blocksize);
if (!bh) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
ll_rw_block(READ, 1, &bh);
retval = bh->b_err;
if (retval) {
brelse(bh);
goto errout;
}
memcpy(&jsuper, start ? bh->b_data :
bh->b_data + SUPERBLOCK_OFFSET,
sizeof(jsuper));
#ifdef WORDS_BIGENDIAN
if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
ext2fs_swap_super(&jsuper);
#endif
if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
!ext2fs_has_feature_journal_dev(&jsuper)) {
retval = EXT2_ET_LOAD_EXT_JOURNAL;
brelse(bh);
goto errout;
}
/* Make sure the journal UUID is correct */
if (memcmp(jsuper.s_uuid, fs->super->s_journal_uuid,
sizeof(jsuper.s_uuid))) {
retval = EXT2_ET_LOAD_EXT_JOURNAL;
brelse(bh);
goto errout;
}
/* Check the superblock checksum */
if (ext2fs_has_feature_metadata_csum(&jsuper)) {
struct struct_ext2_filsys fsx;
struct ext2_super_block superx;
void *p;
p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
memcpy(&fsx, fs, sizeof(fsx));
memcpy(&superx, fs->super, sizeof(superx));
fsx.super = &superx;
ext2fs_set_feature_metadata_csum(fsx.super);
if (!ext2fs_superblock_csum_verify(&fsx, p)) {
retval = EXT2_ET_LOAD_EXT_JOURNAL;
brelse(bh);
goto errout;
}
}
brelse(bh);
maxlen = ext2fs_blocks_count(&jsuper);
journal->j_maxlen = (maxlen < 1ULL << 32) ? maxlen :
(1ULL << 32) - 1;
start++;
}
bh = getblk(dev_journal, start, journal->j_blocksize);
if (!bh) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
journal->j_sb_buffer = bh;
journal->j_superblock = (journal_superblock_t *)bh->b_data;
#ifdef USE_INODE_IO
if (j_inode)
ext2fs_free_mem(&j_inode);
#endif
*ret_journal = journal;
return 0;
errout:
if (dev_fs)
ext2fs_free_mem(&dev_fs);
if (j_inode)
ext2fs_free_mem(&j_inode);
if (journal)
ext2fs_free_mem(&journal);
return retval;
}
/*
* hpfs_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
* struct ucred *a_cred)
*/
static int
hpfs_write(struct vop_write_args *ap)
{
struct vnode *vp = ap->a_vp;
struct hpfsnode *hp = VTOHP(vp);
struct uio *uio = ap->a_uio;
struct buf *bp;
u_int xfersz, towrite;
u_int off;
daddr_t lbn, bn;
int runl;
int error = 0;
dprintf(("hpfs_write(0x%x, off: %d resid: %ld, segflg: %d):\n",
hp->h_no, (u_int32_t)uio->uio_offset,
uio->uio_resid, uio->uio_segflg));
if (ap->a_ioflag & IO_APPEND) {
dprintf(("hpfs_write: APPEND mode\n"));
uio->uio_offset = hp->h_fn.fn_size;
}
if (uio->uio_offset + uio->uio_resid > hp->h_fn.fn_size) {
error = hpfs_extend (hp, uio->uio_offset + uio->uio_resid);
if (error) {
kprintf("hpfs_write: hpfs_extend FAILED %d\n", error);
return (error);
}
}
while (uio->uio_resid) {
lbn = uio->uio_offset >> DEV_BSHIFT;
off = uio->uio_offset & (DEV_BSIZE - 1);
dprintf(("hpfs_write: resid: 0x%lx lbn: 0x%x off: 0x%x\n",
uio->uio_resid, lbn, off));
error = hpfs_hpbmap(hp, lbn, &bn, &runl);
if (error)
return (error);
towrite = szmin(off + uio->uio_resid,
min(DFLTPHYS, (runl+1)*DEV_BSIZE));
xfersz = (towrite + DEV_BSIZE - 1) & ~(DEV_BSIZE - 1);
dprintf(("hpfs_write: bn: 0x%x (0x%x) towrite: 0x%x (0x%x)\n",
bn, runl, towrite, xfersz));
/*
* We do not have to issue a read-before-write if the xfer
* size does not cover the whole block.
*
* In the UIO_NOCOPY case, however, we are not overwriting
* anything and must do a read-before-write to fill in
* any missing pieces.
*/
if (off == 0 && towrite == xfersz &&
uio->uio_segflg != UIO_NOCOPY) {
bp = getblk(hp->h_devvp, dbtodoff(bn), xfersz, 0, 0);
clrbuf(bp);
} else {
error = bread(hp->h_devvp, dbtodoff(bn), xfersz, &bp);
if (error) {
brelse(bp);
return (error);
}
}
error = uiomove(bp->b_data + off, (size_t)(towrite - off), uio);
if(error) {
brelse(bp);
return (error);
}
if (ap->a_ioflag & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
dprintf(("hpfs_write: successful\n"));
return (0);
}
static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
{
struct process_block_struct pb;
struct ext2_super_block *sb = ctx->fs->super;
struct ext2_super_block jsuper;
struct problem_context pctx;
struct buffer_head *bh;
struct inode *j_inode = NULL;
struct kdev_s *dev_fs = NULL, *dev_journal;
const char *journal_name = 0;
journal_t *journal = NULL;
errcode_t retval = 0;
io_manager io_ptr = 0;
unsigned long long start = 0;
int ext_journal = 0;
int tried_backup_jnl = 0;
clear_problem_context(&pctx);
journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
if (!journal) {
return EXT2_ET_NO_MEMORY;
}
dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
if (!dev_fs) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
dev_journal = dev_fs+1;
dev_fs->k_ctx = dev_journal->k_ctx = ctx;
dev_fs->k_dev = K_DEV_FS;
dev_journal->k_dev = K_DEV_JOURNAL;
journal->j_dev = dev_journal;
journal->j_fs_dev = dev_fs;
journal->j_inode = NULL;
journal->j_blocksize = ctx->fs->blocksize;
if (uuid_is_null(sb->s_journal_uuid)) {
if (!sb->s_journal_inum) {
retval = EXT2_ET_BAD_INODE_NUM;
goto errout;
}
j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
"journal inode");
if (!j_inode) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
j_inode->i_ctx = ctx;
j_inode->i_ino = sb->s_journal_inum;
if ((retval = ext2fs_read_inode(ctx->fs,
sb->s_journal_inum,
&j_inode->i_ext2))) {
try_backup_journal:
if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
tried_backup_jnl)
goto errout;
memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
EXT2_N_BLOCKS*4);
j_inode->i_ext2.i_size_high = sb->s_jnl_blocks[15];
j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
j_inode->i_ext2.i_links_count = 1;
j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
e2fsck_use_inode_shortcuts(ctx, 1);
ctx->stashed_ino = j_inode->i_ino;
ctx->stashed_inode = &j_inode->i_ext2;
tried_backup_jnl++;
}
if (!j_inode->i_ext2.i_links_count ||
!LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
retval = EXT2_ET_NO_JOURNAL;
goto try_backup_journal;
}
if (EXT2_I_SIZE(&j_inode->i_ext2) / journal->j_blocksize <
JFS_MIN_JOURNAL_BLOCKS) {
retval = EXT2_ET_JOURNAL_TOO_SMALL;
goto try_backup_journal;
}
pb.last_block = -1;
retval = ext2fs_block_iterate3(ctx->fs, j_inode->i_ino,
BLOCK_FLAG_HOLE, 0,
process_journal_block, &pb);
if ((pb.last_block + 1) * ctx->fs->blocksize <
(int) EXT2_I_SIZE(&j_inode->i_ext2)) {
retval = EXT2_ET_JOURNAL_TOO_SMALL;
goto try_backup_journal;
}
if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
&j_inode->i_ext2);
if (retval)
goto errout;
}
journal->j_maxlen = EXT2_I_SIZE(&j_inode->i_ext2) /
journal->j_blocksize;
#ifdef USE_INODE_IO
retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
&j_inode->i_ext2,
&journal_name);
if (retval)
goto errout;
io_ptr = inode_io_manager;
#else
journal->j_inode = j_inode;
ctx->journal_io = ctx->fs->io;
if ((retval = (errcode_t) journal_bmap(journal, 0, &start)) != 0)
goto errout;
#endif
} else {
ext_journal = 1;
if (!ctx->journal_name) {
char uuid[37];
uuid_unparse(sb->s_journal_uuid, uuid);
ctx->journal_name = blkid_get_devname(ctx->blkid,
"UUID", uuid);
if (!ctx->journal_name)
ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
}
journal_name = ctx->journal_name;
if (!journal_name) {
fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
retval = EXT2_ET_LOAD_EXT_JOURNAL;
goto errout;
}
jfs_debug(1, "Using journal file %s\n", journal_name);
io_ptr = unix_io_manager;
}
#if 0
test_io_backing_manager = io_ptr;
io_ptr = test_io_manager;
#endif
#ifndef USE_INODE_IO
if (ext_journal)
#endif
{
int flags = IO_FLAG_RW;
if (!(ctx->mount_flags & EXT2_MF_ISROOT &&
ctx->mount_flags & EXT2_MF_READONLY))
flags |= IO_FLAG_EXCLUSIVE;
if ((ctx->mount_flags & EXT2_MF_READONLY) &&
(ctx->options & E2F_OPT_FORCE))
flags &= ~IO_FLAG_EXCLUSIVE;
retval = io_ptr->open(journal_name, flags,
&ctx->journal_io);
}
if (retval)
goto errout;
io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
if (ext_journal) {
blk64_t maxlen;
start = ext2fs_journal_sb_start(ctx->fs->blocksize) - 1;
bh = getblk(dev_journal, start, ctx->fs->blocksize);
if (!bh) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
ll_rw_block(READ, 1, &bh);
if ((retval = bh->b_err) != 0) {
brelse(bh);
goto errout;
}
memcpy(&jsuper, start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET,
sizeof(jsuper));
#ifdef WORDS_BIGENDIAN
if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
ext2fs_swap_super(&jsuper);
#endif
if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
!(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
retval = EXT2_ET_LOAD_EXT_JOURNAL;
brelse(bh);
goto errout;
}
/* Make sure the journal UUID is correct */
if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
sizeof(jsuper.s_uuid))) {
fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
retval = EXT2_ET_LOAD_EXT_JOURNAL;
brelse(bh);
goto errout;
}
/* Check the superblock checksum */
if (jsuper.s_feature_ro_compat &
EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) {
struct struct_ext2_filsys fsx;
struct ext2_super_block superx;
void *p;
p = start ? bh->b_data : bh->b_data + SUPERBLOCK_OFFSET;
memcpy(&fsx, ctx->fs, sizeof(fsx));
memcpy(&superx, ctx->fs->super, sizeof(superx));
fsx.super = &superx;
fsx.super->s_feature_ro_compat |=
EXT4_FEATURE_RO_COMPAT_METADATA_CSUM;
if (!ext2fs_superblock_csum_verify(&fsx, p) &&
fix_problem(ctx, PR_0_EXT_JOURNAL_SUPER_CSUM_INVALID,
&pctx)) {
ext2fs_superblock_csum_set(&fsx, p);
mark_buffer_dirty(bh);
}
}
brelse(bh);
maxlen = ext2fs_blocks_count(&jsuper);
journal->j_maxlen = (maxlen < 1ULL << 32) ? maxlen : (1ULL << 32) - 1;
start++;
}
if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
journal->j_sb_buffer = bh;
journal->j_superblock = (journal_superblock_t *)bh->b_data;
#ifdef USE_INODE_IO
if (j_inode)
ext2fs_free_mem(&j_inode);
#endif
*ret_journal = journal;
e2fsck_use_inode_shortcuts(ctx, 0);
return 0;
errout:
e2fsck_use_inode_shortcuts(ctx, 0);
if (dev_fs)
ext2fs_free_mem(&dev_fs);
if (j_inode)
ext2fs_free_mem(&j_inode);
if (journal)
ext2fs_free_mem(&journal);
return retval;
}
void
pass1(void)
{
ino_t inumber, inosused, ninosused;
size_t inospace;
struct inostat *info;
int c;
struct inodesc idesc;
daddr_t i, cgd;
u_int8_t *cp;
/*
* Set file system reserved blocks in used block map.
*/
for (c = 0; c < sblock.fs_ncg; c++) {
cgd = cgdmin(&sblock, c);
if (c == 0)
i = cgbase(&sblock, c);
else
i = cgsblock(&sblock, c);
for (; i < cgd; i++)
setbmap(i);
}
i = sblock.fs_csaddr;
cgd = i + howmany(sblock.fs_cssize, sblock.fs_fsize);
for (; i < cgd; i++)
setbmap(i);
/*
* Find all allocated blocks.
*/
memset(&idesc, 0, sizeof(struct inodesc));
idesc.id_type = ADDR;
idesc.id_func = pass1check;
n_files = n_blks = 0;
info_inumber = 0;
info_fn = pass1_info;
for (c = 0; c < sblock.fs_ncg; c++) {
inumber = c * sblock.fs_ipg;
setinodebuf(inumber);
getblk(&cgblk, cgtod(&sblock, c), sblock.fs_cgsize);
if (sblock.fs_magic == FS_UFS2_MAGIC) {
inosused = cgrp.cg_initediblk;
if (inosused > sblock.fs_ipg)
inosused = sblock.fs_ipg;
} else
inosused = sblock.fs_ipg;
/*
* If we are using soft updates, then we can trust the
* cylinder group inode allocation maps to tell us which
* inodes are allocated. We will scan the used inode map
* to find the inodes that are really in use, and then
* read only those inodes in from disk.
*/
if (preen && usedsoftdep) {
cp = &cg_inosused(&cgrp)[(inosused - 1) / CHAR_BIT];
for ( ; inosused > 0; inosused -= CHAR_BIT, cp--) {
if (*cp == 0)
continue;
for (i = 1 << (CHAR_BIT - 1); i > 0; i >>= 1) {
if (*cp & i)
break;
inosused--;
}
break;
}
if (inosused < 0)
inosused = 0;
}
/*
* Allocate inoinfo structures for the allocated inodes.
*/
inostathead[c].il_numalloced = inosused;
if (inosused == 0) {
inostathead[c].il_stat = 0;
continue;
}
info = calloc((unsigned)inosused, sizeof(struct inostat));
inospace = (unsigned)inosused * sizeof(struct inostat);
if (info == NULL)
errexit("cannot alloc %zu bytes for inoinfo", inospace);
inostathead[c].il_stat = info;
/*
* Scan the allocated inodes.
*/
for (i = 0; i < inosused; i++, inumber++) {
info_inumber = inumber;
if (inumber < ROOTINO) {
(void)getnextinode(inumber);
continue;
}
checkinode(inumber, &idesc);
}
lastino += 1;
if (inosused < sblock.fs_ipg || inumber == lastino)
continue;
/*
* If we were not able to determine in advance which inodes
* were in use, then reduce the size of the inoinfo structure
* to the size necessary to describe the inodes that we
* really found.
*/
if (lastino < (c * sblock.fs_ipg))
ninosused = 0;
else
ninosused = lastino - (c * sblock.fs_ipg);
inostathead[c].il_numalloced = ninosused;
if (ninosused == 0) {
free(inostathead[c].il_stat);
inostathead[c].il_stat = 0;
continue;
}
if (ninosused != inosused) {
struct inostat *ninfo;
size_t ninospace;
ninfo = reallocarray(info, ninosused, sizeof(*ninfo));
if (ninfo == NULL) {
pfatal("too many inodes %llu, or out of memory\n",
(unsigned long long)ninosused);
exit(8);
}
ninospace = ninosused * sizeof(*ninfo);
if (ninosused > inosused)
memset(&ninfo[inosused], 0, ninospace - inospace);
inostathead[c].il_stat = ninfo;
}
}
int main()
{
u16 i,iblk;
char *loc;
u32 *l;
GD *gp;
INODE *ip;
prints("What image: ");
gets(kstr);
prints("\r\n");
// Read group descriptor block and find inode table
getblk(2, buf1);
gp = (GD *)buf1;
iblk = (u16)gp->bg_inode_table;
getblk((u16)iblk, buf1); // read first inode block
ip = (INODE *)buf1 + 1; // ip->root inode #2 ( / directory)
// Read through directory entries looking for /boot
i = findentry((u16)ip->i_block[0], "boot");
// get inode
ip = getinode(iblk, i);
i = findentry((u16)ip->i_block[0], kstr);
// Get inode
ip = getinode(iblk, i);
// get single indirect blocks
getblk((u16)ip->i_block[12], buf2);
// Copy file into memory
prints("Loading: ");
setes(0x1000);
loc = 0;
i = 0;
// Get first 12 blocks
while(i < 12 && ip->i_block[i] != 0)
{
getblk((u16)ip->i_block[i], loc);
i++;
loc += 1024;
putc('.');
}
l = buf2;
i = 0;
while(i < 256 && l[i] > 0)
{
getblk((u16)(l[i]), loc);
i++;
loc += 1024;
putc('+');
}
prints("\n\rReady?\n\r");
getc();
return 1;
}
/*
* Indirect blocks are now on the vnode for the file. They are given negative
* logical block numbers. Indirect blocks are addressed by the negative
* address of the first data block to which they point. Double indirect blocks
* are addressed by one less than the address of the first indirect block to
* which they point. Triple indirect blocks are addressed by one less than
* the address of the first double indirect block to which they point.
*
* ext2_bmaparray does the bmap conversion, and if requested returns the
* array of logical blocks which must be traversed to get to a block.
* Each entry contains the offset into that block that gets you to the
* next block and the disk address of the block (if it is assigned).
*/
static
int
ext2_bmaparray(struct vnode *vp, ext2_daddr_t bn, ext2_daddr_t *bnp,
struct indir *ap, int *nump, int *runp, int *runb)
{
struct inode *ip;
struct buf *bp;
struct ext2_mount *ump;
struct mount *mp;
struct ext2_sb_info *fs;
struct indir a[NIADDR+1], *xap;
ext2_daddr_t daddr;
long metalbn;
int error, maxrun, num;
ip = VTOI(vp);
mp = vp->v_mount;
ump = VFSTOEXT2(mp);
fs = ip->i_e2fs;
#ifdef DIAGNOSTIC
if ((ap != NULL && nump == NULL) || (ap == NULL && nump != NULL))
panic("ext2_bmaparray: invalid arguments");
#endif
if (runp) {
*runp = 0;
}
if (runb) {
*runb = 0;
}
maxrun = mp->mnt_iosize_max / mp->mnt_stat.f_iosize - 1;
xap = ap == NULL ? a : ap;
if (!nump)
nump = #
error = ext2_getlbns(vp, bn, xap, nump);
if (error)
return (error);
num = *nump;
if (num == 0) {
*bnp = blkptrtodb(ump, ip->i_db[bn]);
if (*bnp == 0)
*bnp = -1;
else if (runp) {
daddr_t bnb = bn;
for (++bn; bn < NDADDR && *runp < maxrun &&
is_sequential(ump, ip->i_db[bn - 1], ip->i_db[bn]);
++bn, ++*runp);
bn = bnb;
if (runb && (bn > 0)) {
for (--bn; (bn >= 0) && (*runb < maxrun) &&
is_sequential(ump, ip->i_db[bn],
ip->i_db[bn+1]);
--bn, ++*runb);
}
}
return (0);
}
/* Get disk address out of indirect block array */
daddr = ip->i_ib[xap->in_off];
for (bp = NULL, ++xap; --num; ++xap) {
/*
* Exit the loop if there is no disk address assigned yet and
* the indirect block isn't in the cache, or if we were
* looking for an indirect block and we've found it.
*/
metalbn = xap->in_lbn;
if ((daddr == 0 &&
!findblk(vp, dbtodoff(fs, metalbn), FINDBLK_TEST)) ||
metalbn == bn) {
break;
}
/*
* If we get here, we've either got the block in the cache
* or we have a disk address for it, go fetch it.
*/
if (bp)
bqrelse(bp);
xap->in_exists = 1;
bp = getblk(vp, lblktodoff(fs, metalbn),
mp->mnt_stat.f_iosize, 0, 0);
if ((bp->b_flags & B_CACHE) == 0) {
#ifdef DIAGNOSTIC
if (!daddr)
panic("ext2_bmaparray: indirect block not in cache");
#endif
/*
* This runs through ext2_strategy using bio2 to
* cache the disk offset, then comes back through
* bio1. So we want to wait on bio1
*/
bp->b_bio1.bio_done = biodone_sync;
bp->b_bio1.bio_flags |= BIO_SYNC;
bp->b_bio2.bio_offset = fsbtodoff(fs, daddr);
bp->b_flags &= ~(B_INVAL|B_ERROR);
bp->b_cmd = BUF_CMD_READ;
vfs_busy_pages(bp->b_vp, bp);
vn_strategy(bp->b_vp, &bp->b_bio1);
error = biowait(&bp->b_bio1, "biord");
if (error) {
brelse(bp);
return (error);
}
}
daddr = ((ext2_daddr_t *)bp->b_data)[xap->in_off];
if (num == 1 && daddr && runp) {
for (bn = xap->in_off + 1;
bn < MNINDIR(ump) && *runp < maxrun &&
is_sequential(ump,
((ext2_daddr_t *)bp->b_data)[bn - 1],
((ext2_daddr_t *)bp->b_data)[bn]);
++bn, ++*runp);
bn = xap->in_off;
if (runb && bn) {
for(--bn; bn >= 0 && *runb < maxrun &&
is_sequential(ump, ((daddr_t *)bp->b_data)[bn],
((daddr_t *)bp->b_data)[bn+1]);
--bn, ++*runb);
}
}
}
if (bp)
bqrelse(bp);
daddr = blkptrtodb(ump, daddr);
*bnp = daddr == 0 ? -1 : daddr;
return (0);
}
/* VOP_BWRITE ULFS_NIADDR+2 times */
int
lfs_balloc(struct vnode *vp, off_t startoffset, int iosize, kauth_cred_t cred,
int flags, struct buf **bpp)
{
int offset;
daddr_t daddr, idaddr;
struct buf *ibp, *bp;
struct inode *ip;
struct lfs *fs;
struct indir indirs[ULFS_NIADDR+2], *idp;
daddr_t lbn, lastblock;
int bcount;
int error, frags, i, nsize, osize, num;
ip = VTOI(vp);
fs = ip->i_lfs;
offset = lfs_blkoff(fs, startoffset);
KASSERT(iosize <= lfs_sb_getbsize(fs));
lbn = lfs_lblkno(fs, startoffset);
/* (void)lfs_check(vp, lbn, 0); */
ASSERT_MAYBE_SEGLOCK(fs);
/*
* Three cases: it's a block beyond the end of file, it's a block in
* the file that may or may not have been assigned a disk address or
* we're writing an entire block.
*
* Note, if the daddr is UNWRITTEN, the block already exists in
* the cache (it was read or written earlier). If so, make sure
* we don't count it as a new block or zero out its contents. If
* it did not, make sure we allocate any necessary indirect
* blocks.
*
* If we are writing a block beyond the end of the file, we need to
* check if the old last block was a fragment. If it was, we need
* to rewrite it.
*/
if (bpp)
*bpp = NULL;
/* Check for block beyond end of file and fragment extension needed. */
lastblock = lfs_lblkno(fs, ip->i_size);
if (lastblock < ULFS_NDADDR && lastblock < lbn) {
osize = lfs_blksize(fs, ip, lastblock);
if (osize < lfs_sb_getbsize(fs) && osize > 0) {
if ((error = lfs_fragextend(vp, osize, lfs_sb_getbsize(fs),
lastblock,
(bpp ? &bp : NULL), cred)))
return (error);
ip->i_size = (lastblock + 1) * lfs_sb_getbsize(fs);
lfs_dino_setsize(fs, ip->i_din, ip->i_size);
uvm_vnp_setsize(vp, ip->i_size);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp)
(void) VOP_BWRITE(bp->b_vp, bp);
}
}
/*
* If the block we are writing is a direct block, it's the last
* block in the file, and offset + iosize is less than a full
* block, we can write one or more fragments. There are two cases:
* the block is brand new and we should allocate it the correct
* size or it already exists and contains some fragments and
* may need to extend it.
*/
if (lbn < ULFS_NDADDR && lfs_lblkno(fs, ip->i_size) <= lbn) {
osize = lfs_blksize(fs, ip, lbn);
nsize = lfs_fragroundup(fs, offset + iosize);
if (lfs_lblktosize(fs, lbn) >= ip->i_size) {
/* Brand new block or fragment */
frags = lfs_numfrags(fs, nsize);
if (!ISSPACE(fs, frags, cred))
return ENOSPC;
if (bpp) {
*bpp = bp = getblk(vp, lbn, nsize, 0, 0);
bp->b_blkno = UNWRITTEN;
if (flags & B_CLRBUF)
clrbuf(bp);
}
ip->i_lfs_effnblks += frags;
mutex_enter(&lfs_lock);
lfs_sb_subbfree(fs, frags);
mutex_exit(&lfs_lock);
lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
} else {
if (nsize <= osize) {
/* No need to extend */
if (bpp && (error = bread(vp, lbn, osize,
0, &bp)))
return error;
} else {
/* Extend existing block */
if ((error =
lfs_fragextend(vp, osize, nsize, lbn,
(bpp ? &bp : NULL), cred)))
return error;
}
if (bpp)
*bpp = bp;
}
return 0;
}
error = ulfs_bmaparray(vp, lbn, &daddr, &indirs[0], &num, NULL, NULL);
if (error)
return (error);
KASSERT(daddr <= LFS_MAX_DADDR(fs));
/*
* Do byte accounting all at once, so we can gracefully fail *before*
* we start assigning blocks.
*/
frags = fs->um_seqinc;
bcount = 0;
if (daddr == UNASSIGNED) {
bcount = frags;
}
for (i = 1; i < num; ++i) {
if (!indirs[i].in_exists) {
bcount += frags;
}
}
if (ISSPACE(fs, bcount, cred)) {
mutex_enter(&lfs_lock);
lfs_sb_subbfree(fs, bcount);
mutex_exit(&lfs_lock);
ip->i_lfs_effnblks += bcount;
} else {
return ENOSPC;
}
if (daddr == UNASSIGNED) {
if (num > 0 && lfs_dino_getib(fs, ip->i_din, indirs[0].in_off) == 0) {
lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, UNWRITTEN);
}
/*
* Create new indirect blocks if necessary
*/
if (num > 1) {
idaddr = lfs_dino_getib(fs, ip->i_din, indirs[0].in_off);
for (i = 1; i < num; ++i) {
ibp = getblk(vp, indirs[i].in_lbn,
lfs_sb_getbsize(fs), 0,0);
if (!indirs[i].in_exists) {
clrbuf(ibp);
ibp->b_blkno = UNWRITTEN;
} else if (!(ibp->b_oflags & (BO_DELWRI | BO_DONE))) {
ibp->b_blkno = LFS_FSBTODB(fs, idaddr);
ibp->b_flags |= B_READ;
VOP_STRATEGY(vp, ibp);
biowait(ibp);
}
/*
* This block exists, but the next one may not.
* If that is the case mark it UNWRITTEN to keep
* the accounting straight.
*/
/* XXX ondisk32 */
if (((int32_t *)ibp->b_data)[indirs[i].in_off] == 0)
((int32_t *)ibp->b_data)[indirs[i].in_off] =
UNWRITTEN;
/* XXX ondisk32 */
idaddr = ((int32_t *)ibp->b_data)[indirs[i].in_off];
#ifdef DEBUG
if (vp == fs->lfs_ivnode) {
LFS_ENTER_LOG("balloc", __FILE__,
__LINE__, indirs[i].in_lbn,
ibp->b_flags, curproc->p_pid);
}
#endif
if ((error = VOP_BWRITE(ibp->b_vp, ibp)))
return error;
}
}
}
/*
* Get the existing block from the cache, if requested.
*/
if (bpp)
*bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn), 0, 0);
/*
* Do accounting on blocks that represent pages.
*/
if (!bpp)
lfs_register_block(vp, lbn);
/*
* The block we are writing may be a brand new block
* in which case we need to do accounting.
*
* We can tell a truly new block because ulfs_bmaparray will say
* it is UNASSIGNED. Once we allocate it we will assign it the
* disk address UNWRITTEN.
*/
if (daddr == UNASSIGNED) {
if (bpp) {
if (flags & B_CLRBUF)
clrbuf(bp);
/* Note the new address */
bp->b_blkno = UNWRITTEN;
}
switch (num) {
case 0:
lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
break;
case 1:
lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, UNWRITTEN);
break;
default:
idp = &indirs[num - 1];
if (bread(vp, idp->in_lbn, lfs_sb_getbsize(fs),
B_MODIFY, &ibp))
panic("lfs_balloc: bread bno %lld",
(long long)idp->in_lbn);
/* XXX ondisk32 */
((int32_t *)ibp->b_data)[idp->in_off] = UNWRITTEN;
#ifdef DEBUG
if (vp == fs->lfs_ivnode) {
LFS_ENTER_LOG("balloc", __FILE__,
__LINE__, idp->in_lbn,
ibp->b_flags, curproc->p_pid);
}
#endif
VOP_BWRITE(ibp->b_vp, ibp);
}
} else if (bpp && !(bp->b_oflags & (BO_DONE|BO_DELWRI))) {
/*
* Not a brand new block, also not in the cache;
* read it in from disk.
*/
if (iosize == lfs_sb_getbsize(fs))
/* Optimization: I/O is unnecessary. */
bp->b_blkno = daddr;
else {
/*
* We need to read the block to preserve the
* existing bytes.
*/
bp->b_blkno = daddr;
bp->b_flags |= B_READ;
VOP_STRATEGY(vp, bp);
return (biowait(bp));
}
}
return (0);
}
static int do_one_pass(journal_t *journal, struct recovery_info *info,
enum passtype pass)
{
unsigned int first_commit_ID, next_commit_ID;
unsigned long next_log_block;
int err, success = 0;
journal_superblock_t * sb;
journal_header_t * tmp;
struct buffer_head * bh;
unsigned int sequence;
int blocktype;
/* Precompute the maximum metadata descriptors in a descriptor block */
int MAX_BLOCKS_PER_DESC;
MAX_BLOCKS_PER_DESC = ((journal->j_blocksize-sizeof(journal_header_t))
/ sizeof(journal_block_tag_t));
/*
* First thing is to establish what we expect to find in the log
* (in terms of transaction IDs), and where (in terms of log
* block offsets): query the superblock.
*/
sb = journal->j_superblock;
next_commit_ID = ntohl(sb->s_sequence);
next_log_block = ntohl(sb->s_start);
first_commit_ID = next_commit_ID;
if (pass == PASS_SCAN)
info->start_transaction = first_commit_ID;
jfs_debug(1, "Starting recovery pass %d\n", pass);
/*
* Now we walk through the log, transaction by transaction,
* making sure that each transaction has a commit block in the
* expected place. Each complete transaction gets replayed back
* into the main filesystem.
*/
while (1) {
int flags;
char * tagp;
journal_block_tag_t * tag;
struct buffer_head * obh;
struct buffer_head * nbh;
/* If we already know where to stop the log traversal,
* check right now that we haven't gone past the end of
* the log. */
if (pass != PASS_SCAN)
if (tid_geq(next_commit_ID, info->end_transaction))
break;
jfs_debug(2, "Scanning for sequence ID %u at %lu/%lu\n",
next_commit_ID, next_log_block, journal->j_last);
/* Skip over each chunk of the transaction looking
* either the next descriptor block or the final commit
* record. */
jfs_debug(3, "JFS: checking block %ld\n", next_log_block);
err = jread(&bh, journal, next_log_block);
if (err)
goto failed;
next_log_block++;
wrap(journal, next_log_block);
/* What kind of buffer is it?
*
* If it is a descriptor block, check that it has the
* expected sequence number. Otherwise, we're all done
* here. */
tmp = (journal_header_t *) bh->b_data;
if (tmp->h_magic != htonl(JFS_MAGIC_NUMBER)) {
brelse(bh);
break;
}
blocktype = ntohl(tmp->h_blocktype);
sequence = ntohl(tmp->h_sequence);
jfs_debug(3, "Found magic %d, sequence %d\n",
blocktype, sequence);
if (sequence != next_commit_ID) {
brelse(bh);
break;
}
/* OK, we have a valid descriptor block which matches
* all of the sequence number checks. What are we going
* to do with it? That depends on the pass... */
switch(blocktype) {
case JFS_DESCRIPTOR_BLOCK:
/* If it is a valid descriptor block, replay it
* in pass REPLAY; otherwise, just skip over the
* blocks it describes. */
if (pass != PASS_REPLAY) {
next_log_block += count_tags(bh, journal->j_blocksize);
wrap(journal, next_log_block);
brelse(bh);
continue;
}
/* A descriptor block: we can now write all of
* the data blocks. Yay, useful work is finally
* getting done here! */
tagp = &bh->b_data[sizeof(journal_header_t)];
while ((tagp - bh->b_data +sizeof(journal_block_tag_t))
<= journal->j_blocksize) {
unsigned long io_block;
tag = (journal_block_tag_t *) tagp;
flags = ntohl(tag->t_flags);
io_block = next_log_block++;
wrap(journal, next_log_block);
err = jread(&obh, journal, io_block);
if (err) {
/* Recover what we can, but
* report failure at the end. */
success = err;
printk (KERN_ERR
"JFS: IO error %d recovering "
"block %ld in log\n",
err, io_block);
} else {
unsigned long blocknr;
J_ASSERT(obh != NULL);
blocknr = ntohl(tag->t_blocknr);
/* If the block has been
* revoked, then we're all done
* here. */
if (journal_test_revoke
(journal, blocknr,
next_commit_ID)) {
brelse(obh);
++info->nr_revoke_hits;
goto skip_write;
}
/* Find a buffer for the new
* data being restored */
nbh = getblk(journal->j_dev, blocknr,
journal->j_blocksize);
if (nbh == NULL) {
printk(KERN_ERR
"JFS: Out of memory "
"during recovery.\n");
err = -ENOMEM;
brelse(bh);
brelse(obh);
goto failed;
}
memcpy(nbh->b_data, obh->b_data,
journal->j_blocksize);
if (flags & JFS_FLAG_ESCAPE) {
* ((unsigned int *) bh->b_data) = htonl(JFS_MAGIC_NUMBER);
}
mark_buffer_dirty(nbh, 1);
mark_buffer_uptodate(nbh, 1);
++info->nr_replays;
/* ll_rw_block(WRITE, 1, &nbh); */
brelse(obh);
brelse(nbh);
}
skip_write:
tagp += sizeof(journal_block_tag_t);
if (!(flags & JFS_FLAG_SAME_UUID))
tagp += 16;
if (flags & JFS_FLAG_LAST_TAG)
break;
}
brelse(bh);
continue;
case JFS_COMMIT_BLOCK:
/* Found an expected commit block: not much to
* do other than move on to the next sequence
* number. */
brelse(bh);
next_commit_ID++;
continue;
case JFS_REVOKE_BLOCK:
/* If we aren't in the REVOKE pass, then we can
* just skip over this block. */
if (pass != PASS_REVOKE) {
brelse(bh);
continue;
}
err = scan_revoke_records(journal, bh,
next_commit_ID, info);
brelse(bh);
if (err)
goto failed;
continue;
default:
jfs_debug(3, "Unrecognised magic %d, end of scan.\n",
blocktype);
goto done;
}
}
done:
/*
* We broke out of the log scan loop: either we came to the
* known end of the log or we found an unexpected block in the
* log. If the latter happened, then we know that the "current"
* transaction marks the end of the valid log.
*/
if (pass == PASS_SCAN)
info->end_transaction = next_commit_ID;
else {
/* It's really bad news if different passes end up at
* different places (but possible due to IO errors). */
if (info->end_transaction != next_commit_ID) {
printk (KERN_ERR "JFS: recovery pass %d ended at "
"transaction %u, expected %u\n",
pass, next_commit_ID, info->end_transaction);
if (!success)
success = -EIO;
}
}
return success;
failed:
return err;
}
/*
* Release blocks associated with the inode ip and stored in the indirect
* block bn. Blocks are free'd in LIFO order up to (but not including)
* lastbn. If level is greater than SINGLE, the block is an indirect block
* and recursive calls to indirtrunc must be used to cleanse other indirect
* blocks.
*
* NB: triple indirect blocks are untested.
*/
int
ffs_indirtrunc(struct inode *ip, daddr64_t lbn, daddr64_t dbn,
daddr64_t lastbn, int level, long *countp)
{
int i;
struct buf *bp;
struct fs *fs = ip->i_fs;
struct vnode *vp;
void *copy = NULL;
daddr64_t nb, nlbn, last;
long blkcount, factor;
int nblocks, blocksreleased = 0;
int error = 0, allerror = 0;
int32_t *bap1 = NULL;
#ifdef FFS2
int64_t *bap2 = NULL;
#endif
/*
* Calculate index in current block of last
* block to be kept. -1 indicates the entire
* block so we need not calculate the index.
*/
factor = 1;
for (i = SINGLE; i < level; i++)
factor *= NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = btodb(fs->fs_bsize);
/*
* Get buffer of block pointers, zero those entries corresponding
* to blocks to be free'd, and update on disk copy first. Since
* double(triple) indirect before single(double) indirect, calls
* to bmap on these blocks will fail. However, we already have
* the on disk address, so we have to set the b_blkno field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lbn, (int)fs->fs_bsize, 0, 0);
if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
curproc->p_stats->p_ru.ru_inblock++; /* pay for read */
bcstats.pendingreads++;
bcstats.numreads++;
bp->b_flags |= B_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("ffs_indirtrunc: bad buffer size");
bp->b_blkno = dbn;
VOP_STRATEGY(bp);
error = biowait(bp);
}
if (error) {
brelse(bp);
*countp = 0;
return (error);
}
#ifdef FFS2
if (ip->i_ump->um_fstype == UM_UFS2)
bap2 = (int64_t *)bp->b_data;
else
#endif
bap1 = (int32_t *)bp->b_data;
if (lastbn != -1) {
copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
bcopy(bp->b_data, copy, (u_int) fs->fs_bsize);
for (i = last + 1; i < NINDIR(fs); i++)
BAP_ASSIGN(ip, i, 0);
if (!DOINGASYNC(vp)) {
error = bwrite(bp);
if (error)
allerror = error;
} else {
bawrite(bp);
}
#ifdef FFS2
if (ip->i_ump->um_fstype == UM_UFS2)
bap2 = (int64_t *)copy;
else
#endif
bap1 = (int32_t *)copy;
}
/*
* Recursively free totally unused blocks.
*/
for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = BAP(ip, i);
if (nb == 0)
continue;
if (level > SINGLE) {
error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
(daddr64_t)-1, level - 1,
&blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
ffs_blkfree(ip, nb, fs->fs_bsize);
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = BAP(ip, i);
if (nb != 0) {
error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
last, level - 1, &blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
}
if (copy != NULL) {
free(copy, M_TEMP);
} else {
bp->b_flags |= B_INVAL;
brelse(bp);
}
*countp = blocksreleased;
return (allerror);
}
/*
* Read in the super block and its summary info.
*/
int
readsb(int listerr)
{
off_t super;
int bad, ret;
struct fs *fs;
super = bflag ? bflag * dev_bsize : -1;
readcnt[sblk.b_type]++;
if ((ret = sbget(fsreadfd, &fs, super)) != 0) {
switch (ret) {
case EINVAL:
fprintf(stderr, "The previous newfs operation "
"on this volume did not complete.\nYou must "
"complete newfs before using this volume.\n");
exit(11);
case ENOENT:
if (bflag)
fprintf(stderr, "%jd is not a file system "
"superblock\n", super / dev_bsize);
else
fprintf(stderr, "Cannot find file system "
"superblock\n");
return (0);
case EIO:
default:
fprintf(stderr, "I/O error reading %jd\n",
super / dev_bsize);
return (0);
}
}
memcpy(&sblock, fs, fs->fs_sbsize);
free(fs);
/*
* Compute block size that the file system is based on,
* according to fsbtodb, and adjust superblock block number
* so we can tell if this is an alternate later.
*/
dev_bsize = sblock.fs_fsize / fsbtodb(&sblock, 1);
sblk.b_bno = sblock.fs_sblockactualloc / dev_bsize;
sblk.b_size = SBLOCKSIZE;
/*
* Compare all fields that should not differ in alternate super block.
* When an alternate super-block is specified this check is skipped.
*/
if (bflag)
goto out;
getblk(&asblk, cgsblock(&sblock, sblock.fs_ncg - 1), sblock.fs_sbsize);
if (asblk.b_errs)
return (0);
bad = 0;
#define CHK(x, y) \
if (altsblock.x != sblock.x) { \
bad++; \
if (listerr && debug) \
printf("SUPER BLOCK VS ALTERNATE MISMATCH %s: " y " vs " y "\n", \
#x, (intmax_t)sblock.x, (intmax_t)altsblock.x); \
}
CHK(fs_sblkno, "%jd");
CHK(fs_cblkno, "%jd");
CHK(fs_iblkno, "%jd");
CHK(fs_dblkno, "%jd");
CHK(fs_ncg, "%jd");
CHK(fs_bsize, "%jd");
CHK(fs_fsize, "%jd");
CHK(fs_frag, "%jd");
CHK(fs_bmask, "%#jx");
CHK(fs_fmask, "%#jx");
CHK(fs_bshift, "%jd");
CHK(fs_fshift, "%jd");
CHK(fs_fragshift, "%jd");
CHK(fs_fsbtodb, "%jd");
CHK(fs_sbsize, "%jd");
CHK(fs_nindir, "%jd");
CHK(fs_inopb, "%jd");
CHK(fs_cssize, "%jd");
CHK(fs_ipg, "%jd");
CHK(fs_fpg, "%jd");
CHK(fs_magic, "%#jx");
#undef CHK
if (bad) {
if (listerr == 0)
return (0);
if (preen)
printf("%s: ", cdevname);
printf(
"VALUES IN SUPER BLOCK LSB=%jd DISAGREE WITH THOSE IN\n"
"LAST ALTERNATE LSB=%jd\n",
sblk.b_bno, asblk.b_bno);
if (reply("IGNORE ALTERNATE SUPER BLOCK") == 0)
return (0);
}
out:
/*
* If not yet done, update UFS1 superblock with new wider fields.
*/
if (sblock.fs_magic == FS_UFS1_MAGIC &&
sblock.fs_maxbsize != sblock.fs_bsize) {
sblock.fs_maxbsize = sblock.fs_bsize;
sblock.fs_time = sblock.fs_old_time;
sblock.fs_size = sblock.fs_old_size;
sblock.fs_dsize = sblock.fs_old_dsize;
sblock.fs_csaddr = sblock.fs_old_csaddr;
sblock.fs_cstotal.cs_ndir = sblock.fs_old_cstotal.cs_ndir;
sblock.fs_cstotal.cs_nbfree = sblock.fs_old_cstotal.cs_nbfree;
sblock.fs_cstotal.cs_nifree = sblock.fs_old_cstotal.cs_nifree;
sblock.fs_cstotal.cs_nffree = sblock.fs_old_cstotal.cs_nffree;
}
havesb = 1;
return (1);
}
/*
* If blocks are contiguous on disk, use this to provide clustered
* read ahead. We will read as many blocks as possible sequentially
* and then parcel them up into logical blocks in the buffer hash table.
*/
static struct buf *
cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn,
daddr_t blkno, long size, int run, int gbflags, struct buf *fbp)
{
struct buf *bp, *tbp;
daddr_t bn;
off_t off;
long tinc, tsize;
int i, inc, j, k, toff;
KASSERT(size == vp->v_mount->mnt_stat.f_iosize,
("cluster_rbuild: size %ld != f_iosize %jd\n",
size, (intmax_t)vp->v_mount->mnt_stat.f_iosize));
/*
* avoid a division
*/
while ((u_quad_t) size * (lbn + run) > filesize) {
--run;
}
if (fbp) {
tbp = fbp;
tbp->b_iocmd = BIO_READ;
} else {
tbp = getblk(vp, lbn, size, 0, 0, gbflags);
if (tbp->b_flags & B_CACHE)
return tbp;
tbp->b_flags |= B_ASYNC | B_RAM;
tbp->b_iocmd = BIO_READ;
}
tbp->b_blkno = blkno;
if( (tbp->b_flags & B_MALLOC) ||
((tbp->b_flags & B_VMIO) == 0) || (run <= 1) )
return tbp;
bp = trypbuf(&cluster_pbuf_freecnt);
if (bp == 0)
return tbp;
/*
* We are synthesizing a buffer out of vm_page_t's, but
* if the block size is not page aligned then the starting
* address may not be either. Inherit the b_data offset
* from the original buffer.
*/
bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO;
if ((gbflags & GB_UNMAPPED) != 0) {
bp->b_data = unmapped_buf;
} else {
bp->b_data = (char *)((vm_offset_t)bp->b_data |
((vm_offset_t)tbp->b_data & PAGE_MASK));
}
bp->b_iocmd = BIO_READ;
bp->b_iodone = cluster_callback;
bp->b_blkno = blkno;
bp->b_lblkno = lbn;
bp->b_offset = tbp->b_offset;
KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset"));
pbgetvp(vp, bp);
TAILQ_INIT(&bp->b_cluster.cluster_head);
bp->b_bcount = 0;
bp->b_bufsize = 0;
bp->b_npages = 0;
inc = btodb(size);
for (bn = blkno, i = 0; i < run; ++i, bn += inc) {
if (i == 0) {
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
vfs_drain_busy_pages(tbp);
vm_object_pip_add(tbp->b_bufobj->bo_object,
tbp->b_npages);
for (k = 0; k < tbp->b_npages; k++)
vm_page_sbusy(tbp->b_pages[k]);
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
} else {
if ((bp->b_npages * PAGE_SIZE) +
round_page(size) > vp->v_mount->mnt_iosize_max) {
break;
}
tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT |
(gbflags & GB_UNMAPPED));
/* Don't wait around for locked bufs. */
if (tbp == NULL)
break;
/*
* Stop scanning if the buffer is fully valid
* (marked B_CACHE), or locked (may be doing a
* background write), or if the buffer is not
* VMIO backed. The clustering code can only deal
* with VMIO-backed buffers. The bo lock is not
* required for the BKGRDINPROG check since it
* can not be set without the buf lock.
*/
if ((tbp->b_vflags & BV_BKGRDINPROG) ||
(tbp->b_flags & B_CACHE) ||
(tbp->b_flags & B_VMIO) == 0) {
bqrelse(tbp);
break;
}
/*
* The buffer must be completely invalid in order to
* take part in the cluster. If it is partially valid
* then we stop.
*/
off = tbp->b_offset;
tsize = size;
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
for (j = 0; tsize > 0; j++) {
toff = off & PAGE_MASK;
tinc = tsize;
if (toff + tinc > PAGE_SIZE)
tinc = PAGE_SIZE - toff;
VM_OBJECT_ASSERT_WLOCKED(tbp->b_pages[j]->object);
if ((tbp->b_pages[j]->valid &
vm_page_bits(toff, tinc)) != 0)
break;
if (vm_page_xbusied(tbp->b_pages[j]))
break;
vm_object_pip_add(tbp->b_bufobj->bo_object, 1);
vm_page_sbusy(tbp->b_pages[j]);
off += tinc;
tsize -= tinc;
}
if (tsize > 0) {
clean_sbusy:
vm_object_pip_add(tbp->b_bufobj->bo_object, -j);
for (k = 0; k < j; k++)
vm_page_sunbusy(tbp->b_pages[k]);
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
bqrelse(tbp);
break;
}
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
/*
* Set a read-ahead mark as appropriate
*/
if ((fbp && (i == 1)) || (i == (run - 1)))
tbp->b_flags |= B_RAM;
/*
* Set the buffer up for an async read (XXX should
* we do this only if we do not wind up brelse()ing?).
* Set the block number if it isn't set, otherwise
* if it is make sure it matches the block number we
* expect.
*/
tbp->b_flags |= B_ASYNC;
tbp->b_iocmd = BIO_READ;
if (tbp->b_blkno == tbp->b_lblkno) {
tbp->b_blkno = bn;
} else if (tbp->b_blkno != bn) {
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
goto clean_sbusy;
}
}
/*
* XXX fbp from caller may not be B_ASYNC, but we are going
* to biodone() it in cluster_callback() anyway
*/
BUF_KERNPROC(tbp);
TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head,
tbp, b_cluster.cluster_entry);
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
for (j = 0; j < tbp->b_npages; j += 1) {
vm_page_t m;
m = tbp->b_pages[j];
if ((bp->b_npages == 0) ||
(bp->b_pages[bp->b_npages-1] != m)) {
bp->b_pages[bp->b_npages] = m;
bp->b_npages++;
}
if (m->valid == VM_PAGE_BITS_ALL)
tbp->b_pages[j] = bogus_page;
}
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
/*
* Don't inherit tbp->b_bufsize as it may be larger due to
* a non-page-aligned size. Instead just aggregate using
* 'size'.
*/
if (tbp->b_bcount != size)
printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size);
if (tbp->b_bufsize != size)
printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size);
bp->b_bcount += size;
bp->b_bufsize += size;
}
/*
* Fully valid pages in the cluster are already good and do not need
* to be re-read from disk. Replace the page with bogus_page
*/
VM_OBJECT_WLOCK(bp->b_bufobj->bo_object);
for (j = 0; j < bp->b_npages; j++) {
VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[j]->object);
if (bp->b_pages[j]->valid == VM_PAGE_BITS_ALL)
bp->b_pages[j] = bogus_page;
}
VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object);
if (bp->b_bufsize > bp->b_kvasize)
panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n",
bp->b_bufsize, bp->b_kvasize);
if (buf_mapped(bp)) {
pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
(vm_page_t *)bp->b_pages, bp->b_npages);
}
return (bp);
}
static errcode_t e2fsck_get_journal(e2fsck_t ctx, journal_t **ret_journal)
{
struct process_block_struct pb;
struct ext2_super_block *sb = ctx->fs->super;
struct ext2_super_block jsuper;
struct problem_context pctx;
struct buffer_head *bh;
struct inode *j_inode = NULL;
struct kdev_s *dev_fs = NULL, *dev_journal;
const char *journal_name = 0;
journal_t *journal = NULL;
errcode_t retval = 0;
io_manager io_ptr = 0;
unsigned long start = 0;
int ext_journal = 0;
int tried_backup_jnl = 0;
clear_problem_context(&pctx);
journal = e2fsck_allocate_memory(ctx, sizeof(journal_t), "journal");
if (!journal) {
return EXT2_ET_NO_MEMORY;
}
dev_fs = e2fsck_allocate_memory(ctx, 2*sizeof(struct kdev_s), "kdev");
if (!dev_fs) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
dev_journal = dev_fs+1;
dev_fs->k_ctx = dev_journal->k_ctx = ctx;
dev_fs->k_dev = K_DEV_FS;
dev_journal->k_dev = K_DEV_JOURNAL;
journal->j_dev = dev_journal;
journal->j_fs_dev = dev_fs;
journal->j_inode = NULL;
journal->j_blocksize = ctx->fs->blocksize;
if (uuid_is_null(sb->s_journal_uuid)) {
if (!sb->s_journal_inum) {
retval = EXT2_ET_BAD_INODE_NUM;
goto errout;
}
j_inode = e2fsck_allocate_memory(ctx, sizeof(*j_inode),
"journal inode");
if (!j_inode) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
j_inode->i_ctx = ctx;
j_inode->i_ino = sb->s_journal_inum;
if ((retval = ext2fs_read_inode(ctx->fs,
sb->s_journal_inum,
&j_inode->i_ext2))) {
try_backup_journal:
if (sb->s_jnl_backup_type != EXT3_JNL_BACKUP_BLOCKS ||
tried_backup_jnl)
goto errout;
memset(&j_inode->i_ext2, 0, sizeof(struct ext2_inode));
memcpy(&j_inode->i_ext2.i_block[0], sb->s_jnl_blocks,
EXT2_N_BLOCKS*4);
j_inode->i_ext2.i_size = sb->s_jnl_blocks[16];
j_inode->i_ext2.i_links_count = 1;
j_inode->i_ext2.i_mode = LINUX_S_IFREG | 0600;
e2fsck_use_inode_shortcuts(ctx, 1);
ctx->stashed_ino = j_inode->i_ino;
ctx->stashed_inode = &j_inode->i_ext2;
tried_backup_jnl++;
}
if (!j_inode->i_ext2.i_links_count ||
!LINUX_S_ISREG(j_inode->i_ext2.i_mode)) {
retval = EXT2_ET_NO_JOURNAL;
goto try_backup_journal;
}
if (j_inode->i_ext2.i_size / journal->j_blocksize <
JFS_MIN_JOURNAL_BLOCKS) {
retval = EXT2_ET_JOURNAL_TOO_SMALL;
goto try_backup_journal;
}
pb.last_block = -1;
retval = ext2fs_block_iterate2(ctx->fs, j_inode->i_ino,
BLOCK_FLAG_HOLE, 0,
process_journal_block, &pb);
if ((pb.last_block+1) * ctx->fs->blocksize <
j_inode->i_ext2.i_size) {
retval = EXT2_ET_JOURNAL_TOO_SMALL;
goto try_backup_journal;
}
if (tried_backup_jnl && !(ctx->options & E2F_OPT_READONLY)) {
retval = ext2fs_write_inode(ctx->fs, sb->s_journal_inum,
&j_inode->i_ext2);
if (retval)
goto errout;
}
journal->j_maxlen = j_inode->i_ext2.i_size / journal->j_blocksize;
#ifdef USE_INODE_IO
retval = ext2fs_inode_io_intern2(ctx->fs, sb->s_journal_inum,
&j_inode->i_ext2,
&journal_name);
if (retval)
goto errout;
io_ptr = inode_io_manager;
#else
journal->j_inode = j_inode;
ctx->journal_io = ctx->fs->io;
if ((retval = journal_bmap(journal, 0, &start)) != 0)
goto errout;
#endif
} else {
ext_journal = 1;
if (!ctx->journal_name) {
char uuid[37];
uuid_unparse(sb->s_journal_uuid, uuid);
ctx->journal_name = blkid_get_devname(ctx->blkid,
"UUID", uuid);
if (!ctx->journal_name)
ctx->journal_name = blkid_devno_to_devname(sb->s_journal_dev);
}
journal_name = ctx->journal_name;
if (!journal_name) {
fix_problem(ctx, PR_0_CANT_FIND_JOURNAL, &pctx);
retval = EXT2_ET_LOAD_EXT_JOURNAL;
goto errout;
}
jfs_debug(1, "Using journal file %s\n", journal_name);
io_ptr = unix_io_manager;
}
#if 0
test_io_backing_manager = io_ptr;
io_ptr = test_io_manager;
#endif
#ifndef USE_INODE_IO
if (ext_journal)
#endif
retval = io_ptr->open(journal_name, IO_FLAG_RW,
&ctx->journal_io);
if (retval)
goto errout;
io_channel_set_blksize(ctx->journal_io, ctx->fs->blocksize);
if (ext_journal) {
if (ctx->fs->blocksize == 1024)
start = 1;
bh = getblk(dev_journal, start, ctx->fs->blocksize);
if (!bh) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
ll_rw_block(READ, 1, &bh);
if ((retval = bh->b_err) != 0) {
brelse(bh);
goto errout;
}
memcpy(&jsuper, start ? bh->b_data : bh->b_data + 1024,
sizeof(jsuper));
brelse(bh);
#ifdef EXT2FS_ENABLE_SWAPFS
if (jsuper.s_magic == ext2fs_swab16(EXT2_SUPER_MAGIC))
ext2fs_swap_super(&jsuper);
#endif
if (jsuper.s_magic != EXT2_SUPER_MAGIC ||
!(jsuper.s_feature_incompat & EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
fix_problem(ctx, PR_0_EXT_JOURNAL_BAD_SUPER, &pctx);
retval = EXT2_ET_LOAD_EXT_JOURNAL;
goto errout;
}
/* Make sure the journal UUID is correct */
if (memcmp(jsuper.s_uuid, ctx->fs->super->s_journal_uuid,
sizeof(jsuper.s_uuid))) {
fix_problem(ctx, PR_0_JOURNAL_BAD_UUID, &pctx);
retval = EXT2_ET_LOAD_EXT_JOURNAL;
goto errout;
}
journal->j_maxlen = jsuper.s_blocks_count;
start++;
}
if (!(bh = getblk(dev_journal, start, journal->j_blocksize))) {
retval = EXT2_ET_NO_MEMORY;
goto errout;
}
journal->j_sb_buffer = bh;
journal->j_superblock = (journal_superblock_t *)bh->b_data;
#ifdef USE_INODE_IO
if (j_inode)
ext2fs_free_mem(&j_inode);
#endif
*ret_journal = journal;
e2fsck_use_inode_shortcuts(ctx, 0);
return 0;
errout:
e2fsck_use_inode_shortcuts(ctx, 0);
if (dev_fs)
ext2fs_free_mem(&dev_fs);
if (j_inode)
ext2fs_free_mem(&j_inode);
if (journal)
ext2fs_free_mem(&journal);
return retval;
}
/*
* Write a superblock and associated information back to disk.
*/
int
ffs_sbupdate(struct ufsmount *mp, int waitfor)
{
struct fs *dfs, *fs = mp->um_fs;
struct buf *bp;
int blks;
caddr_t space;
int i, size, error, allerror = 0;
/*
* First write back the summary information.
*/
blks = howmany(fs->fs_cssize, fs->fs_fsize);
space = (caddr_t)fs->fs_csp;
for (i = 0; i < blks; i += fs->fs_frag) {
size = fs->fs_bsize;
if (i + fs->fs_frag > blks)
size = (blks - i) * fs->fs_fsize;
bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
size, 0, 0);
memcpy(bp->b_data, space, size);
space += size;
if (waitfor != MNT_WAIT)
bawrite(bp);
else if ((error = bwrite(bp)))
allerror = error;
}
/*
* Now write back the superblock itself. If any errors occurred
* up to this point, then fail so that the superblock avoids
* being written out as clean.
*/
if (allerror) {
return (allerror);
}
bp = getblk(mp->um_devvp,
fs->fs_sblockloc >> (fs->fs_fshift - fs->fs_fsbtodb),
(int)fs->fs_sbsize, 0, 0);
fs->fs_fmod = 0;
fs->fs_time = time_second;
memcpy(bp->b_data, fs, fs->fs_sbsize);
/* Restore compatibility to old file systems. XXX */
dfs = (struct fs *)bp->b_data; /* XXX */
if (fs->fs_postblformat == FS_42POSTBLFMT) /* XXX */
dfs->fs_nrpos = -1; /* XXX */
if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */
int32_t *lp, tmp; /* XXX */
/* XXX */
lp = (int32_t *)&dfs->fs_qbmask; /* XXX */
tmp = lp[4]; /* XXX */
for (i = 4; i > 0; i--) /* XXX */
lp[i] = lp[i-1]; /* XXX */
lp[0] = tmp; /* XXX */
} /* XXX */
dfs->fs_maxfilesize = mp->um_savedmaxfilesize; /* XXX */
ffs1_compat_write(dfs, mp);
if (waitfor != MNT_WAIT)
bawrite(bp);
else if ((error = bwrite(bp)))
allerror = error;
return (allerror);
}
//
// set up start journal block and journal size
// make journal unreplayable by kernel replay routine
//
void reset_journal (struct super_block * s)
{
int i ;
struct buffer_head *bh ;
int done = 0;
int len;
int start;
/* first block of journal */
s->u.reiserfs_sb.s_rs->s_journal_block = get_journal_start (s);
start = s->u.reiserfs_sb.s_rs->s_journal_block;
/* journal size */
s->u.reiserfs_sb.s_rs->s_orig_journal_size = get_journal_size (s);
len = s->u.reiserfs_sb.s_rs->s_orig_journal_size + 1;
printf ("Resetting journal - "); fflush (stdout);
for (i = 0 ; i < len ; i++) {
print_how_far (&done, len);
bh = getblk (s->s_dev, start + i, s->s_blocksize) ;
memset(bh->b_data, 0, s->s_blocksize) ;
mark_buffer_dirty(bh,0) ;
mark_buffer_uptodate(bh,0) ;
bwrite (bh);
brelse(bh) ;
}
printf ("\n"); fflush (stdout);
#if 0 /* need better way to make journal unreplayable */
/* have journal_read to replay nothing: look for first non-desc
block and set j_first_unflushed_offset to it */
{
int offset;
struct buffer_head * bh, *jh_bh;
struct reiserfs_journal_header * j_head;
struct reiserfs_journal_desc * desc;
jh_bh = bread (s->s_dev, s->u.reiserfs_sb.s_rs->s_journal_block + s->u.reiserfs_sb.s_rs->s_orig_journal_size,
s->s_blocksize);
j_head = (struct reiserfs_journal_header *)(jh_bh->b_data);
for (offset = 0; offset < s->u.reiserfs_sb.s_rs->s_orig_journal_size; offset ++) {
bh = bread (s->s_dev, s->u.reiserfs_sb.s_rs->s_journal_block + offset, s->s_blocksize);
desc = (struct reiserfs_journal_desc *)((bh)->b_data);
if (memcmp(desc->j_magic, JOURNAL_DESC_MAGIC, 8)) {
/* not desc block found */
j_head->j_first_unflushed_offset = offset;
brelse (bh);
break;
}
brelse (bh);
}
mark_buffer_uptodate (jh_bh, 1);
mark_buffer_dirty (jh_bh, 1);
bwrite (jh_bh);
brelse (jh_bh);
}
#endif
}
void
pass5(void)
{
int c, blk, frags, basesize, sumsize, mapsize, cssize;
int inomapsize, blkmapsize;
struct fs *fs = sblock;
daddr_t dbase, dmax;
daddr_t d;
long i, j, k;
struct csum *cs;
struct csum_total cstotal;
struct inodesc idesc[4];
char buf[MAXBSIZE];
struct cg *newcg = (struct cg *)buf;
struct ocg *ocg = (struct ocg *)buf;
struct cg *cg = cgrp, *ncg;
struct inostat *info;
u_int32_t ncgsize;
inoinfo(WINO)->ino_state = USTATE;
memset(newcg, 0, (size_t)fs->fs_cgsize);
newcg->cg_niblk = fs->fs_ipg;
if (cvtlevel >= 3) {
if (fs->fs_maxcontig < 2 && fs->fs_contigsumsize > 0) {
if (preen)
pwarn("DELETING CLUSTERING MAPS\n");
if (preen || reply("DELETE CLUSTERING MAPS")) {
fs->fs_contigsumsize = 0;
doinglevel1 = 1;
sbdirty();
}
}
if (fs->fs_maxcontig > 1) {
const char *doit = NULL;
if (fs->fs_contigsumsize < 1) {
doit = "CREAT";
} else if (fs->fs_contigsumsize < fs->fs_maxcontig &&
fs->fs_contigsumsize < FS_MAXCONTIG) {
doit = "EXPAND";
}
if (doit) {
i = fs->fs_contigsumsize;
fs->fs_contigsumsize =
MIN(fs->fs_maxcontig, FS_MAXCONTIG);
if (CGSIZE(fs) > fs->fs_bsize) {
pwarn("CANNOT %s CLUSTER MAPS\n", doit);
fs->fs_contigsumsize = i;
} else if (preen ||
reply("CREATE CLUSTER MAPS")) {
if (preen)
pwarn("%sING CLUSTER MAPS\n",
doit);
ncgsize = fragroundup(fs, CGSIZE(fs));
ncg = realloc(cgrp, ncgsize);
if (ncg == NULL)
errexit(
"cannot reallocate cg space");
cg = cgrp = ncg;
fs->fs_cgsize = ncgsize;
doinglevel1 = 1;
sbdirty();
}
}
}
}
basesize = &newcg->cg_space[0] - (u_char *)(&newcg->cg_firstfield);
cssize = (u_char *)&cstotal.cs_spare[0] - (u_char *)&cstotal.cs_ndir;
sumsize = 0;
if (is_ufs2) {
newcg->cg_iusedoff = basesize;
} else {
/*
* We reserve the space for the old rotation summary
* tables for the benefit of old kernels, but do not
* maintain them in modern kernels. In time, they can
* go away.
*/
newcg->cg_old_btotoff = basesize;
newcg->cg_old_boff = newcg->cg_old_btotoff +
fs->fs_old_cpg * sizeof(int32_t);
newcg->cg_iusedoff = newcg->cg_old_boff +
fs->fs_old_cpg * fs->fs_old_nrpos * sizeof(u_int16_t);
memset(&newcg->cg_space[0], 0, newcg->cg_iusedoff - basesize);
}
inomapsize = howmany(fs->fs_ipg, CHAR_BIT);
newcg->cg_freeoff = newcg->cg_iusedoff + inomapsize;
blkmapsize = howmany(fs->fs_fpg, CHAR_BIT);
newcg->cg_nextfreeoff = newcg->cg_freeoff + blkmapsize;
if (fs->fs_contigsumsize > 0) {
newcg->cg_clustersumoff = newcg->cg_nextfreeoff -
sizeof(u_int32_t);
if (isappleufs) {
/* Apple PR2216969 gives rationale for this change.
* I believe they were mistaken, but we need to
* duplicate it for compatibility. -- [email protected]
*/
newcg->cg_clustersumoff += sizeof(u_int32_t);
}
newcg->cg_clustersumoff =
roundup(newcg->cg_clustersumoff, sizeof(u_int32_t));
newcg->cg_clusteroff = newcg->cg_clustersumoff +
(fs->fs_contigsumsize + 1) * sizeof(u_int32_t);
newcg->cg_nextfreeoff = newcg->cg_clusteroff +
howmany(fragstoblks(fs, fs->fs_fpg), CHAR_BIT);
}
newcg->cg_magic = CG_MAGIC;
mapsize = newcg->cg_nextfreeoff - newcg->cg_iusedoff;
if (!is_ufs2 && ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
switch ((int)fs->fs_old_postblformat) {
case FS_42POSTBLFMT:
basesize = (char *)(&ocg->cg_btot[0]) -
(char *)(&ocg->cg_firstfield);
sumsize = &ocg->cg_iused[0] - (u_int8_t *)(&ocg->cg_btot[0]);
mapsize = &ocg->cg_free[howmany(fs->fs_fpg, NBBY)] -
(u_char *)&ocg->cg_iused[0];
blkmapsize = howmany(fs->fs_fpg, NBBY);
inomapsize = &ocg->cg_free[0] - (u_char *)&ocg->cg_iused[0];
ocg->cg_magic = CG_MAGIC;
newcg->cg_magic = 0;
break;
case FS_DYNAMICPOSTBLFMT:
sumsize = newcg->cg_iusedoff - newcg->cg_old_btotoff;
break;
default:
errexit("UNKNOWN ROTATIONAL TABLE FORMAT %d",
fs->fs_old_postblformat);
}
}
memset(&idesc[0], 0, sizeof idesc);
for (i = 0; i < 4; i++) {
idesc[i].id_type = ADDR;
if (!is_ufs2 && doinglevel2)
idesc[i].id_fix = FIX;
}
memset(&cstotal, 0, sizeof(struct csum_total));
dmax = blknum(fs, fs->fs_size + fs->fs_frag - 1);
for (d = fs->fs_size; d < dmax; d++)
setbmap(d);
for (c = 0; c < fs->fs_ncg; c++) {
if (got_siginfo) {
fprintf(stderr,
"%s: phase 5: cyl group %d of %d (%d%%)\n",
cdevname(), c, fs->fs_ncg,
c * 100 / fs->fs_ncg);
got_siginfo = 0;
}
#ifdef PROGRESS
progress_bar(cdevname(), preen ? NULL : "phase 5",
c, fs->fs_ncg);
#endif /* PROGRESS */
getblk(&cgblk, cgtod(fs, c), fs->fs_cgsize);
memcpy(cg, cgblk.b_un.b_cg, fs->fs_cgsize);
if((doswap && !needswap) || (!doswap && needswap))
ffs_cg_swap(cgblk.b_un.b_cg, cg, sblock);
if (!doinglevel1 && !cg_chkmagic(cg, 0))
pfatal("CG %d: PASS5: BAD MAGIC NUMBER\n", c);
if(doswap)
cgdirty();
/*
* While we have the disk head where we want it,
* write back the superblock to the spare at this
* cylinder group.
*/
if ((cvtlevel && sblk.b_dirty) || doswap) {
bwrite(fswritefd, sblk.b_un.b_buf,
fsbtodb(sblock, cgsblock(sblock, c)),
sblock->fs_sbsize);
} else {
/*
* Read in the current alternate superblock,
* and compare it to the master. If it's
* wrong, fix it up.
*/
getblk(&asblk, cgsblock(sblock, c), sblock->fs_sbsize);
if (asblk.b_errs)
pfatal("CG %d: UNABLE TO READ ALTERNATE "
"SUPERBLK\n", c);
else {
memmove(altsblock, asblk.b_un.b_fs,
sblock->fs_sbsize);
if (needswap)
ffs_sb_swap(asblk.b_un.b_fs, altsblock);
}
sb_oldfscompat_write(sblock, sblocksave);
if ((asblk.b_errs || cmpsblks(sblock, altsblock)) &&
dofix(&idesc[3],
"ALTERNATE SUPERBLK(S) ARE INCORRECT")) {
bwrite(fswritefd, sblk.b_un.b_buf,
fsbtodb(sblock, cgsblock(sblock, c)),
sblock->fs_sbsize);
}
sb_oldfscompat_read(sblock, 0);
}
dbase = cgbase(fs, c);
dmax = dbase + fs->fs_fpg;
if (dmax > fs->fs_size)
dmax = fs->fs_size;
if (is_ufs2 || (fs->fs_old_flags & FS_FLAGS_UPDATED))
newcg->cg_time = cg->cg_time;
newcg->cg_old_time = cg->cg_old_time;
newcg->cg_cgx = c;
newcg->cg_ndblk = dmax - dbase;
if (!is_ufs2) {
if (c == fs->fs_ncg - 1) {
/* Avoid fighting old fsck for this value. Its never used
* outside of this check anyway.
*/
if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)
newcg->cg_old_ncyl = fs->fs_old_ncyl % fs->fs_old_cpg;
else
newcg->cg_old_ncyl = howmany(newcg->cg_ndblk,
fs->fs_fpg / fs->fs_old_cpg);
} else
newcg->cg_old_ncyl = fs->fs_old_cpg;
newcg->cg_old_niblk = fs->fs_ipg;
newcg->cg_niblk = 0;
}
if (fs->fs_contigsumsize > 0)
newcg->cg_nclusterblks = newcg->cg_ndblk / fs->fs_frag;
newcg->cg_cs.cs_ndir = 0;
newcg->cg_cs.cs_nffree = 0;
newcg->cg_cs.cs_nbfree = 0;
newcg->cg_cs.cs_nifree = fs->fs_ipg;
if (cg->cg_rotor >= 0 && cg->cg_rotor < newcg->cg_ndblk)
newcg->cg_rotor = cg->cg_rotor;
else
newcg->cg_rotor = 0;
if (cg->cg_frotor >= 0 && cg->cg_frotor < newcg->cg_ndblk)
newcg->cg_frotor = cg->cg_frotor;
else
newcg->cg_frotor = 0;
if (cg->cg_irotor >= 0 && cg->cg_irotor < fs->fs_ipg)
newcg->cg_irotor = cg->cg_irotor;
else
newcg->cg_irotor = 0;
if (!is_ufs2) {
newcg->cg_initediblk = 0;
} else {
if ((unsigned)cg->cg_initediblk > fs->fs_ipg)
newcg->cg_initediblk = fs->fs_ipg;
else
newcg->cg_initediblk = cg->cg_initediblk;
}
memset(&newcg->cg_frsum[0], 0, sizeof newcg->cg_frsum);
memset(&old_cg_blktot(newcg, 0)[0], 0, (size_t)(sumsize));
memset(cg_inosused(newcg, 0), 0, (size_t)(mapsize));
if (!is_ufs2 && ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) &&
fs->fs_old_postblformat == FS_42POSTBLFMT)
ocg->cg_magic = CG_MAGIC;
j = fs->fs_ipg * c;
for (i = 0; i < fs->fs_ipg; j++, i++) {
info = inoinfo(j);
switch (info->ino_state) {
case USTATE:
break;
case DSTATE:
case DCLEAR:
case DFOUND:
newcg->cg_cs.cs_ndir++;
/* fall through */
case FSTATE:
case FCLEAR:
newcg->cg_cs.cs_nifree--;
setbit(cg_inosused(newcg, 0), i);
break;
default:
if (j < ROOTINO)
break;
errexit("BAD STATE %d FOR INODE I=%ld",
info->ino_state, (long)j);
}
}
if (c == 0)
for (i = 0; i < ROOTINO; i++) {
setbit(cg_inosused(newcg, 0), i);
newcg->cg_cs.cs_nifree--;
}
for (i = 0, d = dbase;
d < dmax;
d += fs->fs_frag, i += fs->fs_frag) {
frags = 0;
for (j = 0; j < fs->fs_frag; j++) {
if (testbmap(d + j))
continue;
setbit(cg_blksfree(newcg, 0), i + j);
frags++;
}
if (frags == fs->fs_frag) {
newcg->cg_cs.cs_nbfree++;
if (sumsize) {
j = old_cbtocylno(fs, i);
old_cg_blktot(newcg, 0)[j]++;
old_cg_blks(fs, newcg, j, 0)[old_cbtorpos(fs, i)]++;
}
if (fs->fs_contigsumsize > 0)
setbit(cg_clustersfree(newcg, 0),
fragstoblks(fs, i));
} else if (frags > 0) {
newcg->cg_cs.cs_nffree += frags;
blk = blkmap(fs, cg_blksfree(newcg, 0), i);
ffs_fragacct(fs, blk, newcg->cg_frsum, 1, 0);
}
}
if (fs->fs_contigsumsize > 0) {
int32_t *sump = cg_clustersum(newcg, 0);
u_char *mapp = cg_clustersfree(newcg, 0);
int map = *mapp++;
int bit = 1;
int run = 0;
for (i = 0; i < newcg->cg_nclusterblks; i++) {
if ((map & bit) != 0) {
run++;
} else if (run != 0) {
if (run > fs->fs_contigsumsize)
run = fs->fs_contigsumsize;
sump[run]++;
run = 0;
}
if ((i & (NBBY - 1)) != (NBBY - 1)) {
bit <<= 1;
} else {
map = *mapp++;
bit = 1;
}
}
if (run != 0) {
if (run > fs->fs_contigsumsize)
run = fs->fs_contigsumsize;
sump[run]++;
}
}
cstotal.cs_nffree += newcg->cg_cs.cs_nffree;
cstotal.cs_nbfree += newcg->cg_cs.cs_nbfree;
cstotal.cs_nifree += newcg->cg_cs.cs_nifree;
cstotal.cs_ndir += newcg->cg_cs.cs_ndir;
cs = &fs->fs_cs(fs, c);
if (memcmp(&newcg->cg_cs, cs, sizeof *cs) != 0) {
if (debug) {
printf("cg %d: nffree: %d/%d nbfree %d/%d"
" nifree %d/%d ndir %d/%d\n",
c, cs->cs_nffree,newcg->cg_cs.cs_nffree,
cs->cs_nbfree,newcg->cg_cs.cs_nbfree,
cs->cs_nifree,newcg->cg_cs.cs_nifree,
cs->cs_ndir,newcg->cg_cs.cs_ndir);
}
if (dofix(&idesc[0], "FREE BLK COUNT(S) WRONG IN SUPERBLK")) {
memmove(cs, &newcg->cg_cs, sizeof *cs);
sbdirty();
} else
markclean = 0;
}
if (doinglevel1) {
memmove(cg, newcg, (size_t)fs->fs_cgsize);
cgdirty();
continue;
}
if ((memcmp(newcg, cg, basesize) != 0) ||
(memcmp(&old_cg_blktot(newcg, 0)[0],
&old_cg_blktot(cg, 0)[0], sumsize) != 0)) {
if (dofix(&idesc[2], "SUMMARY INFORMATION BAD")) {
memmove(cg, newcg, (size_t)basesize);
memmove(&old_cg_blktot(cg, 0)[0],
&old_cg_blktot(newcg, 0)[0], (size_t)sumsize);
cgdirty();
} else
markclean = 0;
}
if (usedsoftdep) {
for (i = 0; i < inomapsize; i++) {
j = cg_inosused(newcg, 0)[i];
if ((cg_inosused(cg, 0)[i] & j) == j)
continue;
for (k = 0; k < NBBY; k++) {
if ((j & (1 << k)) == 0)
continue;
if (cg_inosused(cg, 0)[i] & (1 << k))
continue;
pwarn("ALLOCATED INODE %ld "
"MARKED FREE\n",
c * fs->fs_ipg + i * 8 + k);
}
}
for (i = 0; i < blkmapsize; i++) {
j = cg_blksfree(cg, 0)[i];
if ((cg_blksfree(newcg, 0)[i] & j) == j)
continue;
for (k = 0; k < NBBY; k++) {
if ((j & (1 << k)) == 0)
continue;
if (cg_inosused(cg, 0)[i] & (1 << k))
continue;
pwarn("ALLOCATED FRAG %ld "
"MARKED FREE\n",
c * fs->fs_fpg + i * 8 + k);
}
}
}
if (memcmp(cg_inosused(newcg, 0), cg_inosused(cg, 0), mapsize)
!= 0 && dofix(&idesc[1], "BLK(S) MISSING IN BIT MAPS")) {
memmove(cg_inosused(cg, 0), cg_inosused(newcg, 0),
(size_t)mapsize);
cgdirty();
}
}
if (memcmp(&cstotal, &fs->fs_cstotal, cssize) != 0) {
if (debug) {
printf("total: nffree: %lld/%lld nbfree %lld/%lld"
" nifree %lld/%lld ndir %lld/%lld\n",
(long long int)fs->fs_cstotal.cs_nffree,
(long long int)cstotal.cs_nffree,
(long long int)fs->fs_cstotal.cs_nbfree,
(long long int)cstotal.cs_nbfree,
(long long int)fs->fs_cstotal.cs_nifree,
(long long int)cstotal.cs_nifree,
(long long int)fs->fs_cstotal.cs_ndir,
(long long int)cstotal.cs_ndir);
}
if (dofix(&idesc[0], "FREE BLK COUNT(S) WRONG IN SUPERBLK")) {
memmove(&fs->fs_cstotal, &cstotal, sizeof cstotal);
fs->fs_ronly = 0;
fs->fs_fmod = 0;
sbdirty();
} else
markclean = 0;
}
#ifdef PROGRESS
if (!preen)
progress_done();
#endif /* PROGRESS */
}
/*
* Balloc defines the structure of filesystem storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
*/
int
ext2_balloc(struct inode *ip, e2fs_lbn_t lbn, int size, struct ucred *cred,
struct buf **bpp, int flags)
{
struct m_ext2fs *fs;
struct ext2mount *ump;
struct buf *bp, *nbp;
struct vnode *vp = ITOV(ip);
struct indir indirs[EXT2_NIADDR + 2];
e4fs_daddr_t nb, newb;
e2fs_daddr_t *bap, pref;
int osize, nsize, num, i, error;
*bpp = NULL;
if (lbn < 0)
return (EFBIG);
fs = ip->i_e2fs;
ump = ip->i_ump;
/*
* check if this is a sequential block allocation.
* If so, increment next_alloc fields to allow ext2_blkpref
* to make a good guess
*/
if (lbn == ip->i_next_alloc_block + 1) {
ip->i_next_alloc_block++;
ip->i_next_alloc_goal++;
}
if (ip->i_flag & IN_E4EXTENTS)
return (ext2_ext_balloc(ip, lbn, size, cred, bpp, flags));
/*
* The first EXT2_NDADDR blocks are direct blocks
*/
if (lbn < EXT2_NDADDR) {
nb = ip->i_db[lbn];
/*
* no new block is to be allocated, and no need to expand
* the file
*/
if (nb != 0 && ip->i_size >= (lbn + 1) * fs->e2fs_bsize) {
error = bread(vp, lbn, fs->e2fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lbn, osize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
/*
* Godmar thinks: this shouldn't happen w/o
* fragments
*/
printf("nsize %d(%d) > osize %d(%d) nb %d\n",
(int)nsize, (int)size, (int)osize,
(int)ip->i_size, (int)nb);
panic(
"ext2_balloc: Something is terribly wrong");
/*
* please note there haven't been any changes from here on -
* FFS seems to work.
*/
}
} else {
if (ip->i_size < (lbn + 1) * fs->e2fs_bsize)
nsize = fragroundup(fs, size);
else
nsize = fs->e2fs_bsize;
EXT2_LOCK(ump);
error = ext2_alloc(ip, lbn,
ext2_blkpref(ip, lbn, (int)lbn, &ip->i_db[0], 0),
nsize, cred, &newb);
if (error)
return (error);
/*
* If the newly allocated block exceeds 32-bit limit,
* we can not use it in file block maps.
*/
if (newb > UINT_MAX)
return (EFBIG);
bp = getblk(vp, lbn, nsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
}
ip->i_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ext2_getlbns(vp, lbn, indirs, &num)) != 0)
return (error);
#ifdef INVARIANTS
if (num < 1)
panic("ext2_balloc: ext2_getlbns returned indirect block");
#endif
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = ip->i_ib[indirs[0].in_off];
if (nb == 0) {
EXT2_LOCK(ump);
pref = ext2_blkpref(ip, lbn, indirs[0].in_off +
EXT2_NDIR_BLOCKS, &ip->i_db[0], 0);
if ((error = ext2_alloc(ip, lbn, pref, fs->e2fs_bsize, cred,
&newb)))
return (error);
if (newb > UINT_MAX)
return (EFBIG);
nb = newb;
bp = getblk(vp, indirs[1].in_lbn, fs->e2fs_bsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
vfs_bio_clrbuf(bp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(bp)) != 0) {
ext2_blkfree(ip, nb, fs->e2fs_bsize);
return (error);
}
ip->i_ib[indirs[0].in_off] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->e2fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bap = (e2fs_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
EXT2_LOCK(ump);
if (pref == 0)
pref = ext2_blkpref(ip, lbn, indirs[i].in_off, bap,
bp->b_lblkno);
error = ext2_alloc(ip, lbn, pref, (int)fs->e2fs_bsize, cred, &newb);
if (error) {
brelse(bp);
return (error);
}
if (newb > UINT_MAX)
return (EFBIG);
nb = newb;
nbp = getblk(vp, indirs[i].in_lbn, fs->e2fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
ext2_blkfree(ip, nb, fs->e2fs_bsize);
EXT2_UNLOCK(ump);
brelse(bp);
return (error);
}
bap[indirs[i - 1].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
EXT2_LOCK(ump);
pref = ext2_blkpref(ip, lbn, indirs[i].in_off, &bap[0],
bp->b_lblkno);
if ((error = ext2_alloc(ip,
lbn, pref, (int)fs->e2fs_bsize, cred, &newb)) != 0) {
brelse(bp);
return (error);
}
if (newb > UINT_MAX)
return (EFBIG);
nb = newb;
nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
*bpp = nbp;
return (0);
}
brelse(bp);
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->e2fs_bsize, NOCRED,
MAXBSIZE, seqcount, 0, &nbp);
} else {
error = bread(vp, lbn, (int)fs->e2fs_bsize, NOCRED, &nbp);
}
if (error) {
brelse(nbp);
return (error);
}
} else {
