/*
* ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size,
* struct ucred *a_cred, int a_flags, struct buf *a_bpp)
*
* 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.
*
* NOTE: B_CLRBUF - this flag tells balloc to clear invalid portions
* of the buffer. However, any dirty bits will override missing
* valid bits. This case occurs when writable mmaps are truncated
* and then extended.
*/
int
ffs_balloc(struct vop_balloc_args *ap)
{
struct inode *ip;
ufs_daddr_t lbn;
int size;
struct ucred *cred;
int flags;
struct fs *fs;
ufs_daddr_t nb;
struct buf *bp, *nbp, *dbp;
struct vnode *vp;
struct indir indirs[NIADDR + 2];
ufs_daddr_t newb, *bap, pref;
int deallocated, osize, nsize, num, i, error;
ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx;
int seqcount;
vp = ap->a_vp;
ip = VTOI(vp);
fs = ip->i_fs;
lbn = lblkno(fs, ap->a_startoffset);
size = blkoff(fs, ap->a_startoffset) + ap->a_size;
if (size > fs->fs_bsize)
panic("ffs_balloc: blk too big");
*ap->a_bpp = NULL;
if (lbn < 0)
return (EFBIG);
cred = ap->a_cred;
flags = ap->a_flags;
/*
* The vnode must be locked for us to be able to safely mess
* around with the inode.
*/
if (vn_islocked(vp) != LK_EXCLUSIVE) {
panic("ffs_balloc: vnode %p not exclusively locked!", vp);
}
/*
* 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.
*/
nb = lblkno(fs, ip->i_size);
if (nb < NDADDR && nb < lbn) {
/*
* The filesize prior to this write can fit in direct
* blocks (ex. fragmentation is possibly done)
* we are now extending the file write beyond
* the block which has end of the file prior to this write.
*/
osize = blksize(fs, ip, nb);
/*
* osize gives disk allocated size in the last block. It is
* either in fragments or a file system block size.
*/
if (osize < fs->fs_bsize && osize > 0) {
/* A few fragments are already allocated, since the
* current extends beyond this block allocated the
* complete block as fragments are on in last block.
*/
error = ffs_realloccg(ip, nb,
ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
osize, (int)fs->fs_bsize, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dofftofsb(fs, bp->b_bio2.bio_offset),
ip->i_db[nb], fs->fs_bsize, osize, bp);
/* adjust the inode size, we just grew */
ip->i_size = smalllblktosize(fs, nb + 1);
ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
/* bp is already released here */
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
*ap->a_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, lblktodoff(fs, lbn),
osize, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
} else {
/*
* NOTE: ffs_realloccg() issues a bread().
*/
error = ffs_realloccg(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn,
&ip->i_db[0]), osize, nsize, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dofftofsb(fs, bp->b_bio2.bio_offset),
nb, nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
nsize, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0);
bp->b_bio2.bio_offset = fsbtodoff(fs, newb);
if (flags & B_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*ap->a_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 DIAGNOSTIC
if (num < 1)
panic ("ffs_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Get a handle on the data block buffer before working through
* indirect blocks to avoid a deadlock between the VM system holding
* a locked VM page and issuing a BMAP (which tries to lock the
* indirect blocks), and the filesystem holding a locked indirect
* block and then trying to read a data block (which tries to lock
* the underlying VM pages).
*/
dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0);
/*
* Setup undo history
*/
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
unwindidx = -1;
/*
* Fetch the first indirect block directly from the inode, allocating
* one if necessary.
*/
--num;
nb = ip->i_ib[indirs[0].in_off];
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, 0, NULL);
/*
* If the filesystem has run out of space we can skip the
* full fsync/undo of the main [fail] case since no undo
* history has been built yet. Hence the goto fail2.
*/
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
cred, &newb)) != 0)
goto fail2;
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn),
fs->fs_bsize, 0, 0);
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &ip->i_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
if (pref == 0)
pref = ffs_blkpref(ip, lbn, 0, NULL);
if ((error =
ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn),
fs->fs_bsize, 0, 0);
nbp->b_bio2.bio_offset = fsbtodoff(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 {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary. We have already
* called getblk() on the data block buffer, dbp. If we have to
* allocate it and B_CLRBUF has been set the inference is an intention
* to zero out the related disk blocks, so we do not have to issue
* a read. Instead we simply call vfs_bio_clrbuf(). If B_CLRBUF is
* not set the caller intends to overwrite the entire contents of the
* buffer and we don't waste time trying to clean up the contents.
*
* bp references the current indirect block. When allocating,
* the block must be updated.
*/
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip,
lbn, pref, (int)fs->fs_bsize, cred, &newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
if (flags & B_CLRBUF)
vfs_bio_clrbuf(dbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, dbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
*ap->a_bpp = dbp;
return (0);
}
brelse(bp);
/*
* At this point all related indirect blocks have been allocated
* if necessary and released. bp is no longer valid. dbp holds
* our getblk()'d data block.
*
* XXX we previously performed a cluster_read operation here.
*/
if (flags & B_CLRBUF) {
/*
* If B_CLRBUF is set we must validate the invalid portions
* of the buffer. This typically requires a read-before-
* write. The strategy call will fill in bio_offset in that
* case.
*
* If we hit this case we do a cluster read if possible
* since nearby data blocks are likely to be accessed soon
* too.
*/
if ((dbp->b_flags & B_CACHE) == 0) {
bqrelse(dbp);
seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
if (seqcount &&
(vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, (off_t)ip->i_size,
lblktodoff(fs, lbn),
(int)fs->fs_bsize,
fs->fs_bsize,
seqcount * BKVASIZE,
&dbp);
} else {
error = bread(vp, lblktodoff(fs, lbn),
(int)fs->fs_bsize, &dbp);
}
if (error)
goto fail;
} else {
/*
* Find a suitable location for the journal in the filesystem.
*
* Our strategy here is to look for a contiguous block of free space
* at least "logfile" MB in size (plus room for any indirect blocks).
* We start at the middle of the filesystem and check each cylinder
* group working outwards. If "logfile" MB is not available as a
* single contigous chunk, then return the address and size of the
* largest chunk found.
*
* XXX
* At what stage does the search fail? Is if the largest space we could
* find is less than a quarter the requested space reasonable? If the
* search fails entirely, return a block address if "0" it indicate this.
*/
void
wapbl_find_log_start(struct mount *mp, struct vnode *vp, off_t logsize,
daddr_t *addr, daddr_t *indir_addr, size_t *size)
{
struct ufsmount *ump = VFSTOUFS(mp);
struct fs *fs = ump->um_fs;
struct vnode *devvp = ump->um_devvp;
struct cg *cgp;
struct buf *bp;
uint8_t *blksfree;
daddr_t blkno, best_addr, start_addr;
daddr_t desired_blks, min_desired_blks;
daddr_t freeblks, best_blks;
int bpcg, cg, error, fixedsize, indir_blks, n, s;
#ifdef FFS_EI
const int needswap = UFS_FSNEEDSWAP(fs);
#endif
if (logsize == 0) {
fixedsize = 0; /* We can adjust the size if tight */
logsize = lfragtosize(fs, fs->fs_dsize) /
UFS_WAPBL_JOURNAL_SCALE;
DPRINTF("suggested log size = %lld\n", logsize);
logsize = max(logsize, UFS_WAPBL_MIN_JOURNAL_SIZE);
logsize = min(logsize, UFS_WAPBL_MAX_JOURNAL_SIZE);
DPRINTF("adjusted log size = %lld\n", logsize);
} else {
fixedsize = 1;
DPRINTF("fixed log size = %lld\n", logsize);
}
desired_blks = logsize / fs->fs_bsize;
DPRINTF("desired blocks = %lld\n", desired_blks);
/* add in number of indirect blocks needed */
indir_blks = 0;
if (desired_blks >= NDADDR) {
struct indir indirs[NIADDR + 2];
int num;
error = ufs_getlbns(vp, desired_blks, indirs, &num);
if (error) {
printf("%s: ufs_getlbns failed, error %d!\n",
__func__, error);
goto bad;
}
switch (num) {
case 2:
indir_blks = 1; /* 1st level indirect */
break;
case 3:
indir_blks = 1 + /* 1st level indirect */
1 + /* 2nd level indirect */
indirs[1].in_off + 1; /* extra 1st level indirect */
break;
default:
printf("%s: unexpected numlevels %d from ufs_getlbns\n",
__func__, num);
*size = 0;
goto bad;
}
desired_blks += indir_blks;
}
DPRINTF("desired blocks = %lld (including indirect)\n",
desired_blks);
/*
* If a specific size wasn't requested, allow for a smaller log
* if we're really tight for space...
*/
min_desired_blks = desired_blks;
if (!fixedsize)
min_desired_blks = desired_blks / 4;
/* Look at number of blocks per CG. If it's too small, bail early. */
bpcg = fragstoblks(fs, fs->fs_fpg);
if (min_desired_blks > bpcg) {
printf("ffs_wapbl: cylinder group size of %lld MB "
" is not big enough for journal\n",
lblktosize(fs, bpcg) / (1024 * 1024));
goto bad;
}
/*
* Start with the middle cylinder group, and search outwards in
* both directions until we either find the requested log size
* or reach the start/end of the file system. If we reach the
* start/end without finding enough space for the full requested
* log size, use the largest extent found if it is large enough
* to satisfy the our minimum size.
*
* XXX
* Can we just use the cluster contigsum stuff (esp on UFS2)
* here to simplify this search code?
*/
best_addr = 0;
best_blks = 0;
for (cg = fs->fs_ncg / 2, s = 0, n = 1;
best_blks < desired_blks && cg >= 0 && cg < fs->fs_ncg;
s++, n = -n, cg += n * s) {
DPRINTF("check cg %d of %d\n", cg, fs->fs_ncg);
error = bread(devvp, fsbtodb(fs, cgtod(fs, cg)),
fs->fs_cgsize, &bp);
if (error) {
continue;
}
cgp = (struct cg *)bp->b_data;
if (!cg_chkmagic(cgp)) {
brelse(bp);
continue;
}
blksfree = cg_blksfree(cgp);
for (blkno = 0; blkno < bpcg;) {
/* look for next free block */
/* XXX use scanc() and fragtbl[] here? */
for (; blkno < bpcg - min_desired_blks; blkno++)
if (ffs_isblock(fs, blksfree, blkno))
break;
/* past end of search space in this CG? */
if (blkno >= bpcg - min_desired_blks)
break;
/* count how many free blocks in this extent */
start_addr = blkno;
for (freeblks = 0; blkno < bpcg; blkno++, freeblks++)
if (!ffs_isblock(fs, blksfree, blkno))
break;
if (freeblks > best_blks) {
best_blks = freeblks;
best_addr = blkstofrags(fs, start_addr) +
cgbase(fs, cg);
if (freeblks >= desired_blks) {
DPRINTF("found len %lld"
" at offset %lld in gc\n",
freeblks, start_addr);
break;
}
}
}
brelse(bp);
}
DPRINTF("best found len = %lld, wanted %lld"
" at addr %lld\n", best_blks, desired_blks, best_addr);
if (best_blks < min_desired_blks) {
*addr = 0;
*indir_addr = 0;
} else {
/* put indirect blocks at start, and data blocks after */
*addr = best_addr + blkstofrags(fs, indir_blks);
*indir_addr = best_addr;
}
*size = min(desired_blks, best_blks) - indir_blks;
return;
bad:
*addr = 0;
*indir_addr = 0;
*size = 0;
return;
}
static int
ffs_balloc_ufs1(struct inode *ip, off_t offset, int bufsize, struct buf **bpp)
{
makefs_daddr_t lbn, lastlbn;
int size;
int32_t nb;
struct buf *bp, *nbp;
struct fs *fs = ip->i_fs;
struct indir indirs[UFS_NIADDR + 2];
makefs_daddr_t newb, pref;
int32_t *bap;
int osize, nsize, num, i, error;
int32_t *allocblk, allociblk[UFS_NIADDR + 1];
int32_t *allocib;
const int needswap = UFS_FSNEEDSWAP(fs);
lbn = lblkno(fs, offset);
size = blkoff(fs, offset) + bufsize;
if (bpp != NULL) {
*bpp = NULL;
}
assert(size <= fs->fs_bsize);
if (lbn < 0)
return (EFBIG);
/*
* 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_ffs1_size);
if (lastlbn < UFS_NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
warnx("need to ffs_realloccg; not supported!");
abort();
}
}
/*
* The first UFS_NDADDR blocks are direct blocks
*/
if (lbn < UFS_NDADDR) {
nb = ufs_rw32(ip->i_ffs1_db[lbn], needswap);
if (nb != 0 && ip->i_ffs1_size >=
(uint64_t)lblktosize(fs, lbn + 1)) {
/*
* The block is an already-allocated direct block
* and the file already extends past this block,
* thus this must be a whole block.
* Just read the block (if requested).
*/
if (bpp != NULL) {
error = bread(ip->i_devvp, lbn, fs->fs_bsize,
NULL, bpp);
if (error) {
brelse(*bpp);
return (error);
}
}
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_ffs1_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
/*
* The existing block is already
* at least as big as we want.
* Just read the block (if requested).
*/
if (bpp != NULL) {
error = bread(ip->i_devvp, lbn, osize,
NULL, bpp);
if (error) {
brelse(*bpp);
return (error);
}
}
return 0;
} else {
warnx("need to ffs_realloccg; not supported!");
abort();
}
} else {
/*
* the block was not previously allocated,
* allocate a new block or fragment.
*/
if (ip->i_ffs1_size < (uint64_t)lblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs1(ip, lbn, (int)lbn,
&ip->i_ffs1_db[0]),
nsize, &newb);
if (error)
return (error);
if (bpp != NULL) {
bp = getblk(ip->i_devvp, lbn, nsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
clrbuf(bp);
*bpp = bp;
}
}
ip->i_ffs1_db[lbn] = ufs_rw32((int32_t)newb, needswap);
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(ip, lbn, indirs, &num)) != 0)
return (error);
if (num < 1) {
warnx("ffs_balloc: ufs_getlbns returned indirect block");
abort();
}
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = ufs_rw32(ip->i_ffs1_ib[indirs[0].in_off], needswap);
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ffs_blkpref_ufs1(ip, lbn, 0, (int32_t *)0);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, &newb);
if (error)
return error;
nb = newb;
*allocblk++ = nb;
bp = getblk(ip->i_devvp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
clrbuf(bp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(bp)) != 0)
return error;
allocib = &ip->i_ffs1_ib[indirs[0].in_off];
*allocib = ufs_rw32((int32_t)nb, needswap);
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(ip->i_devvp, indirs[i].in_lbn, fs->fs_bsize,
NULL, &bp);
if (error) {
brelse(bp);
return error;
}
bap = (int32_t *)bp->b_data;
nb = ufs_rw32(bap[indirs[i].in_off], needswap);
if (i == num)
break;
i++;
if (nb != 0) {
brelse(bp);
continue;
}
if (pref == 0)
pref = ffs_blkpref_ufs1(ip, lbn, 0, (int32_t *)0);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, &newb);
if (error) {
brelse(bp);
return error;
}
nb = newb;
*allocblk++ = nb;
nbp = getblk(ip->i_devvp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
return error;
}
bap[indirs[i - 1].in_off] = ufs_rw32(nb, needswap);
bwrite(bp);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
pref = ffs_blkpref_ufs1(ip, lbn, indirs[num].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, &newb);
if (error) {
brelse(bp);
return error;
}
nb = newb;
*allocblk++ = nb;
if (bpp != NULL) {
nbp = getblk(ip->i_devvp, lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
*bpp = nbp;
}
bap[indirs[num].in_off] = ufs_rw32(nb, needswap);
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
bwrite(bp);
return (0);
}
brelse(bp);
if (bpp != NULL) {
error = bread(ip->i_devvp, lbn, (int)fs->fs_bsize, NULL, &nbp);
if (error) {
brelse(nbp);
return error;
}
*bpp = nbp;
}
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.
*/
ffs_balloc(
register struct inode *ip,
register ufs_daddr_t lbn,
int size,
kauth_cred_t cred,
struct buf **bpp,
int flags,
int * blk_alloc)
{
register struct fs *fs;
register ufs_daddr_t nb;
struct buf *bp, *nbp;
struct vnode *vp = ITOV(ip);
struct indir indirs[NIADDR + 2];
ufs_daddr_t newb, *bap, pref;
int deallocated, osize, nsize, num, i, error;
ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
int devBlockSize=0;
int alloc_buffer = 1;
struct mount *mp=vp->v_mount;
#if REV_ENDIAN_FS
int rev_endian=(mp->mnt_flag & MNT_REVEND);
#endif /* REV_ENDIAN_FS */
*bpp = NULL;
if (lbn < 0)
return (EFBIG);
fs = ip->i_fs;
if (flags & B_NOBUFF)
alloc_buffer = 0;
if (blk_alloc)
*blk_alloc = 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.
*/
nb = lblkno(fs, ip->i_size);
if (nb < NDADDR && nb < lbn) {
/* the filesize prior to this write can fit in direct
* blocks (ie. fragmentaion is possibly done)
* we are now extending the file write beyond
* the block which has end of file prior to this write
*/
osize = blksize(fs, ip, nb);
/* osize gives disk allocated size in the last block. It is
* either in fragments or a file system block size */
if (osize < fs->fs_bsize && osize > 0) {
/* few fragments are already allocated,since the
* current extends beyond this block
* allocate the complete block as fragments are only
* in last block
*/
error = ffs_realloccg(ip, nb,
ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
osize, (int)fs->fs_bsize, cred, &bp);
if (error)
return (error);
/* adjust the inode size we just grew */
/* it is in nb+1 as nb starts from 0 */
ip->i_size = (nb + 1) * fs->fs_bsize;
ubc_setsize(vp, (off_t)ip->i_size);
ip->i_db[nb] = dbtofsb(fs, (ufs_daddr_t)buf_blkno(bp));
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if ((flags & B_SYNC) || (!alloc_buffer)) {
if (!alloc_buffer)
buf_setflags(bp, B_NOCACHE);
buf_bwrite(bp);
} else
buf_bdwrite(bp);
/* note that bp is already released here */
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_db[lbn];
if (nb != 0 && ip->i_size >= (lbn + 1) * fs->fs_bsize) {
if (alloc_buffer) {
error = (int)buf_bread(vp, (daddr64_t)((unsigned)lbn), fs->fs_bsize, NOCRED, &bp);
if (error) {
buf_brelse(bp);
return (error);
}
*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) {
if (alloc_buffer) {
error = (int)buf_bread(vp, (daddr64_t)((unsigned)lbn), osize, NOCRED, &bp);
if (error) {
buf_brelse(bp);
return (error);
}
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
else {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (0);
}
} else {
error = ffs_realloccg(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn,
&ip->i_db[0]), osize, nsize, cred, &bp);
if (error)
return (error);
ip->i_db[lbn] = dbtofsb(fs, (ufs_daddr_t)buf_blkno(bp));
ip->i_flag |= IN_CHANGE | IN_UPDATE;
/* adjust the inode size we just grew */
ip->i_size = (lbn * fs->fs_bsize) + size;
ubc_setsize(vp, (off_t)ip->i_size);
if (!alloc_buffer) {
buf_setflags(bp, B_NOCACHE);
if (flags & B_SYNC)
buf_bwrite(bp);
else
buf_bdwrite(bp);
} else
*bpp = bp;
return (0);
}
} else {
if (ip->i_size < (lbn + 1) * fs->fs_bsize)
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
nsize, cred, &newb);
if (error)
return (error);
if (alloc_buffer) {
bp = buf_getblk(vp, (daddr64_t)((unsigned)lbn), nsize, 0, 0, BLK_WRITE);
buf_setblkno(bp, (daddr64_t)((unsigned)fsbtodb(fs, newb)));
if (flags & B_CLRBUF)
buf_clear(bp);
}
ip->i_db[lbn] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (blk_alloc) {
*blk_alloc = nsize;
}
if (alloc_buffer)
*bpp = bp;
return (0);
}
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if (error = ufs_getlbns(vp, lbn, indirs, &num))
return(error);
#if DIAGNOSTIC
if (num < 1)
panic ("ffs_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = ip->i_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
if (error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
cred, &newb))
return (error);
nb = newb;
*allocblk++ = nb;
bp = buf_getblk(vp, (daddr64_t)((unsigned)(indirs[1].in_lbn)), fs->fs_bsize, 0, 0, BLK_META);
buf_setblkno(bp, (daddr64_t)((unsigned)fsbtodb(fs, nb)));
buf_clear(bp);
/*
* Write synchronously conditional on mount flags.
*/
if ((vp)->v_mount->mnt_flag & MNT_ASYNC) {
error = 0;
buf_bdwrite(bp);
} else if ((error = buf_bwrite(bp)) != 0) {
goto fail;
}
allocib = &ip->i_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = (int)buf_meta_bread(vp, (daddr64_t)((unsigned)(indirs[i].in_lbn)), (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
buf_brelse(bp);
goto fail;
}
bap = (ufs_daddr_t *)buf_dataptr(bp);
#if REV_ENDIAN_FS
if (rev_endian)
nb = OSSwapInt32(bap[indirs[i].in_off]);
else {
#endif /* REV_ENDIAN_FS */
nb = bap[indirs[i].in_off];
#if REV_ENDIAN_FS
}
#endif /* REV_ENDIAN_FS */
if (i == num)
break;
i += 1;
if (nb != 0) {
buf_brelse(bp);
continue;
}
if (pref == 0)
pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
if (error =
ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) {
buf_brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
nbp = buf_getblk(vp, (daddr64_t)((unsigned)(indirs[i].in_lbn)), fs->fs_bsize, 0, 0, BLK_META);
buf_setblkno(nbp, (daddr64_t)((unsigned)fsbtodb(fs, nb)));
buf_clear(nbp);
/*
* Write synchronously conditional on mount flags.
*/
if ((vp)->v_mount->mnt_flag & MNT_ASYNC) {
error = 0;
buf_bdwrite(nbp);
} else if (error = buf_bwrite(nbp)) {
buf_brelse(bp);
goto fail;
}
#if REV_ENDIAN_FS
if (rev_endian)
bap[indirs[i - 1].in_off] = OSSwapInt32(nb);
else {
#endif /* REV_ENDIAN_FS */
bap[indirs[i - 1].in_off] = nb;
#if REV_ENDIAN_FS
}
#endif /* REV_ENDIAN_FS */
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
buf_bwrite(bp);
} else {
buf_bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
if (error = ffs_alloc(ip,
lbn, pref, (int)fs->fs_bsize, cred, &newb)) {
buf_brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
#if REV_ENDIAN_FS
if (rev_endian)
bap[indirs[i].in_off] = OSSwapInt32(nb);
else {
#endif /* REV_ENDIAN_FS */
bap[indirs[i].in_off] = nb;
#if REV_ENDIAN_FS
}
#endif /* REV_ENDIAN_FS */
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if ((flags & B_SYNC)) {
buf_bwrite(bp);
} else {
buf_bdwrite(bp);
}
if(alloc_buffer ) {
nbp = buf_getblk(vp, (daddr64_t)((unsigned)lbn), fs->fs_bsize, 0, 0, BLK_WRITE);
buf_setblkno(nbp, (daddr64_t)((unsigned)fsbtodb(fs, nb)));
if (flags & B_CLRBUF)
buf_clear(nbp);
}
if (blk_alloc) {
*blk_alloc = fs->fs_bsize;
}
if(alloc_buffer)
*bpp = nbp;
return (0);
}
buf_brelse(bp);
if (alloc_buffer) {
if (flags & B_CLRBUF) {
error = (int)buf_bread(vp, (daddr64_t)((unsigned)lbn), (int)fs->fs_bsize, NOCRED, &nbp);
if (error) {
buf_brelse(nbp);
goto fail;
}
} else {
nbp = buf_getblk(vp, (daddr64_t)((unsigned)lbn), fs->fs_bsize, 0, 0, BLK_WRITE);
buf_setblkno(nbp, (daddr64_t)((unsigned)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++) {
ffs_blkfree(ip, *blkp, fs->fs_bsize);
deallocated += fs->fs_bsize;
}
if (allocib != NULL)
*allocib = 0;
if (deallocated) {
devBlockSize = vfs_devblocksize(mp);
#if QUOTA
/*
* Restore user's disk quota because allocation failed.
*/
(void) chkdq(ip, (int64_t)-deallocated, cred, FORCE);
#endif /* QUOTA */
ip->i_blocks -= btodb(deallocated, devBlockSize);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
return (error);
}
/*
* 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.
* This is the allocation strategy for UFS1. Below is
* the allocation strategy for UFS2.
*/
int
ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
struct ucred *cred, int flags, struct buf **bpp)
{
struct inode *ip;
struct ufs1_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
ufs1_daddr_t nb;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[NIADDR + 2];
int deallocated, osize, nsize, num, i, error;
ufs2_daddr_t newb;
ufs1_daddr_t *bap, pref;
ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int reclaimed;
ip = VTOI(vp);
dp = ip->i_din1;
fs = ip->i_fs;
ump = ip->i_ump;
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs1: blk too big");
*bpp = NULL;
if (flags & IO_EXT)
return (EOPNOTSUPP);
if (lbn < 0)
return (EFBIG);
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* 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 < 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_ufs1(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 NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lbn, fs->fs_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 {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs1(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_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, 0);
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_ufs1: 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 = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_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);
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
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 = (ufs1_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 (pref == 0)
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (++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;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
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 {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && 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);
pref = ffs_blkpref_ufs1(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 (++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;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, 0);
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 (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->fs_bsize, NOCRED,
MAXBSIZE, seqcount, &nbp);
} else {
error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &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.
* 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 {
/*
* ffs_blkalloc allocates a disk block for ffs_pageout(), as a consequence
* it does no buf_breads (that could lead to deadblock as the page may be already
* marked busy as it is being paged out. Also important to note that we are not
* growing the file in pageouts. So ip->i_size cannot increase by this call
* due to the way UBC works.
* This code is derived from ffs_balloc and many cases of that are dealt
* in ffs_balloc are not applicable here
* Do not call with B_CLRBUF flags as this should only be called only
* from pageouts
*/
ffs_blkalloc(
struct inode *ip,
ufs_daddr_t lbn,
int size,
kauth_cred_t cred,
int flags)
{
register struct fs *fs;
register ufs_daddr_t nb;
struct buf *bp, *nbp;
struct vnode *vp = ITOV(ip);
struct indir indirs[NIADDR + 2];
ufs_daddr_t newb, *bap, pref;
int deallocated, osize, nsize, num, i, error;
ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
int devBlockSize=0;
struct mount *mp=vp->v_mount;
#if REV_ENDIAN_FS
int rev_endian=(mp->mnt_flag & MNT_REVEND);
#endif /* REV_ENDIAN_FS */
fs = ip->i_fs;
if(size > fs->fs_bsize)
panic("ffs_blkalloc: too large for allocation");
/*
* 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.
*/
nb = lblkno(fs, ip->i_size);
if (nb < NDADDR && nb < lbn) {
panic("ffs_blkalloc():cannot extend file: i_size %d, lbn %d", ip->i_size, lbn);
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_db[lbn];
if (nb != 0 && ip->i_size >= (lbn + 1) * fs->fs_bsize) {
/* TBD: trivial case; the block is already allocated */
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) {
panic("ffs_allocblk: trying to extend a fragment");
}
return(0);
} else {
if (ip->i_size < (lbn + 1) * fs->fs_bsize)
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
nsize, cred, &newb);
if (error)
return (error);
ip->i_db[lbn] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (0);
}
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if (error = ufs_getlbns(vp, lbn, indirs, &num))
return(error);
if(num == 0) {
panic("ffs_blkalloc: file with direct blocks only");
}
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = ip->i_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
if (error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
cred, &newb))
return (error);
nb = newb;
*allocblk++ = nb;
bp = buf_getblk(vp, (daddr64_t)((unsigned)(indirs[1].in_lbn)), fs->fs_bsize, 0, 0, BLK_META);
buf_setblkno(bp, (daddr64_t)((unsigned)fsbtodb(fs, nb)));
buf_clear(bp);
/*
* Write synchronously conditional on mount flags.
*/
if ((vp)->v_mount->mnt_flag & MNT_ASYNC) {
error = 0;
buf_bdwrite(bp);
} else if (error = buf_bwrite(bp)) {
goto fail;
}
allocib = &ip->i_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = (int)buf_meta_bread(vp, (daddr64_t)((unsigned)(indirs[i].in_lbn)), (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
buf_brelse(bp);
goto fail;
}
bap = (ufs_daddr_t *)buf_dataptr(bp);
#if REV_ENDIAN_FS
if (rev_endian)
nb = OSSwapInt32(bap[indirs[i].in_off]);
else {
#endif /* REV_ENDIAN_FS */
nb = bap[indirs[i].in_off];
#if REV_ENDIAN_FS
}
#endif /* REV_ENDIAN_FS */
if (i == num)
break;
i += 1;
if (nb != 0) {
buf_brelse(bp);
continue;
}
if (pref == 0)
pref = ffs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
if (error =
ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) {
buf_brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
nbp = buf_getblk(vp, (daddr64_t)((unsigned)(indirs[i].in_lbn)), fs->fs_bsize, 0, 0, BLK_META);
buf_setblkno(nbp, (daddr64_t)((unsigned)fsbtodb(fs, nb)));
buf_clear(nbp);
/*
* Write synchronously conditional on mount flags.
*/
if ((vp)->v_mount->mnt_flag & MNT_ASYNC) {
error = 0;
buf_bdwrite(nbp);
} else if (error = buf_bwrite(nbp)) {
buf_brelse(bp);
goto fail;
}
#if REV_ENDIAN_FS
if (rev_endian)
bap[indirs[i - 1].in_off] = OSSwapInt32(nb);
else {
#endif /* REV_ENDIAN_FS */
bap[indirs[i - 1].in_off] = nb;
#if REV_ENDIAN_FS
}
#endif /* REV_ENDIAN_FS */
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
buf_bwrite(bp);
} else {
buf_bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
if (error = ffs_alloc(ip,
lbn, pref, (int)fs->fs_bsize, cred, &newb)) {
buf_brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
#if REV_ENDIAN_FS
if (rev_endian)
bap[indirs[i].in_off] = OSSwapInt32(nb);
else {
#endif /* REV_ENDIAN_FS */
bap[indirs[i].in_off] = nb;
#if REV_ENDIAN_FS
}
#endif /* REV_ENDIAN_FS */
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
buf_bwrite(bp);
} else {
buf_bdwrite(bp);
}
return (0);
}
buf_brelse(bp);
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++) {
ffs_blkfree(ip, *blkp, fs->fs_bsize);
deallocated += fs->fs_bsize;
}
if (allocib != NULL)
*allocib = 0;
if (deallocated) {
devBlockSize = vfs_devblocksize(mp);
#if QUOTA
/*
* Restore user's disk quota because allocation failed.
*/
(void) chkdq(ip, (int64_t)-deallocated, cred, FORCE);
#endif /* QUOTA */
ip->i_blocks -= btodb(deallocated, devBlockSize);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
return (error);
}
/*
* 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;
}
/*
* 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);
}
int
ffs_fsync(void *v)
{
struct vop_fsync_args /* {
struct vnode *a_vp;
kauth_cred_t a_cred;
int a_flags;
off_t a_offlo;
off_t a_offhi;
struct lwp *a_l;
} */ *ap = v;
struct buf *bp;
int num, error, i;
struct indir ia[NIADDR + 1];
int bsize;
daddr_t blk_high;
struct vnode *vp;
struct mount *mp;
vp = ap->a_vp;
mp = vp->v_mount;
fstrans_start(mp, FSTRANS_LAZY);
if ((ap->a_offlo == 0 && ap->a_offhi == 0) || (vp->v_type != VREG)) {
error = ffs_full_fsync(vp, ap->a_flags);
goto out;
}
bsize = mp->mnt_stat.f_iosize;
blk_high = ap->a_offhi / bsize;
if (ap->a_offhi % bsize != 0)
blk_high++;
/*
* First, flush all pages in range.
*/
mutex_enter(vp->v_interlock);
error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo),
round_page(ap->a_offhi), PGO_CLEANIT |
((ap->a_flags & FSYNC_WAIT) ? PGO_SYNCIO : 0));
if (error) {
goto out;
}
#ifdef WAPBL
KASSERT(vp->v_type == VREG);
if (mp->mnt_wapbl) {
/*
* Don't bother writing out metadata if the syncer is
* making the request. We will let the sync vnode
* write it out in a single burst through a call to
* VFS_SYNC().
*/
if ((ap->a_flags & (FSYNC_DATAONLY | FSYNC_LAZY)) != 0) {
fstrans_done(mp);
return 0;
}
error = 0;
if (vp->v_tag == VT_UFS && VTOI(vp)->i_flag &
(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY |
IN_MODIFIED | IN_ACCESSED)) {
error = UFS_WAPBL_BEGIN(mp);
if (error) {
fstrans_done(mp);
return error;
}
error = ffs_update(vp, NULL, NULL, UPDATE_CLOSE |
((ap->a_flags & FSYNC_WAIT) ? UPDATE_WAIT : 0));
UFS_WAPBL_END(mp);
}
if (error || (ap->a_flags & FSYNC_NOLOG) != 0) {
fstrans_done(mp);
return error;
}
error = wapbl_flush(mp->mnt_wapbl, 0);
fstrans_done(mp);
return error;
}
#endif /* WAPBL */
/*
* Then, flush indirect blocks.
*/
if (blk_high >= NDADDR) {
error = ufs_getlbns(vp, blk_high, ia, &num);
if (error)
goto out;
mutex_enter(&bufcache_lock);
for (i = 0; i < num; i++) {
if ((bp = incore(vp, ia[i].in_lbn)) == NULL)
continue;
if ((bp->b_cflags & BC_BUSY) != 0 ||
(bp->b_oflags & BO_DELWRI) == 0)
continue;
bp->b_cflags |= BC_BUSY | BC_VFLUSH;
mutex_exit(&bufcache_lock);
bawrite(bp);
mutex_enter(&bufcache_lock);
}
mutex_exit(&bufcache_lock);
}
if (ap->a_flags & FSYNC_WAIT) {
mutex_enter(vp->v_interlock);
while (vp->v_numoutput > 0)
cv_wait(&vp->v_cv, vp->v_interlock);
mutex_exit(vp->v_interlock);
}
error = ffs_update(vp, NULL, NULL, UPDATE_CLOSE |
(((ap->a_flags & (FSYNC_WAIT | FSYNC_DATAONLY)) == FSYNC_WAIT)
? UPDATE_WAIT : 0));
if (error == 0 && ap->a_flags & FSYNC_CACHE) {
int l = 0;
VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE,
curlwp->l_cred);
}
out:
fstrans_done(mp);
return error;
}
int
ufs_bmaparray(
vnode *vp,
ufs2_daddr_t bn,
ufs2_daddr_t *bnp,
Buf *nbp,
int *runp,
int *runb)
{
Buf *bp;
Indir a[UFS_NIADDR+1], *ap;
ufs2_daddr_t daddr;
ufs_lbn_t metalbn;
int error, num, maxrun = 0;
int *nump;
inode *ip = vp->data;
MountPoint *mp = vp->mount;
ufsmount *ump = mp->mnt_data;
if (runp) {
maxrun = mp->mnt_iosize_max / mp->mnt_stat.f_iosize - 1;
*runp = 0;
}
if (runb) {
*runb = 0;
}
ap = a;
nump = #
error = ufs_getlbns(vp, bn, ap, nump);
if (error)
return error;
num = *nump;
if (num == 0) {
if (bn >= 0 && bn < UFS_NDADDR) {
*bnp = blkptrtodb(ump, ip->din2->di_db[bn]);
} else if (bn < 0 && bn >= -UFS_NXADDR) {
*bnp = blkptrtodb(ump, ip->din2->di_extb[-1 - bn]);
if (*bnp == 0)
*bnp = -1;
if (nbp == nil)
panic("ufs_bmaparray: mapping ext data");
// TODO HARVEY Mark ALTDATA?
//nbp->b_xflags |= BX_ALTDATA;
return (0);
} else {
panic("ufs_bmaparray: blkno out of range");
}
/*
* Since this is FFS independent code, we are out of
* scope for the definitions of BLK_NOCOPY and
* BLK_SNAP, but we do know that they will fall in
* the range 1..um_seqinc, so we use that test and
* return a request for a zeroed out buffer if attempts
* are made to read a BLK_NOCOPY or BLK_SNAP block.
*/
if ((ip->i_flags & SF_SNAPSHOT) && ip->din2->di_db[bn] > 0 &&
ip->din2->di_db[bn] < ump->um_seqinc) {
*bnp = -1;
} else if (*bnp == 0) {
if (ip->i_flags & SF_SNAPSHOT)
*bnp = blkptrtodb(ump, bn * ump->um_seqinc);
else
*bnp = -1;
} else if (runp) {
ufs2_daddr_t bnb = bn;
for (++bn; bn < UFS_NDADDR && *runp < maxrun &&
is_sequential(ump, ip->din2->di_db[bn - 1], ip->din2->di_db[bn]);
++bn, ++*runp)
;
bn = bnb;
if (runb && (bn > 0)) {
for (--bn; (bn >= 0) && (*runb < maxrun) &&
is_sequential(ump, ip->din2->di_db[bn],
ip->din2->di_db[bn+1]);
--bn, ++*runb)
;
}
}
return (0);
}
/* Get disk address out of indirect block array */
daddr = ip->din2->di_ib[ap->in_off];
for (bp = nil, ++ap; --num; ++ap) {
/*
* 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 = ap->in_lbn;
// TODO HARVEY Going to have to revisit this when we implement
// writing, so we can read writes before they've been flushed
// to disk.
//if ((daddr == 0 && !incore(&vp->v_bufobj, metalbn)) || metalbn == bn)
// break;
if (daddr == 0 || 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)
releasebuf(bp);
bp = getblk(vp, metalbn, mp->mnt_stat.f_iosize, 0);
// TODO HARVEY Revisit when we manage a cache of Bufs
/*if ((bp->b_flags & B_CACHE) == 0) {
#ifdef INVARIANTS
if (!daddr)
panic("ufs_bmaparray: indirect block not in cache");
#endif
bp->b_blkno = blkptrtodb(ump, daddr);
bp->b_iocmd = BIO_READ;
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
vfs_busy_pages(bp, 0);
bp->b_iooffset = dbtob(bp->b_blkno);
ffs_geom_strategy(bp);
curthread->td_ru.ru_inblock++;
error = bufwait(bp);
if (error) {
brelse(bp);
return (error);
}
}*/
daddr = ((ufs2_daddr_t *)bp->data)[ap->in_off];
if (num == 1 && daddr && runp) {
for (bn = ap->in_off + 1;
bn < ump->um_nindir && *runp < maxrun &&
is_sequential(ump,
((ufs2_daddr_t *)bp->data)[bn - 1],
((ufs2_daddr_t *)bp->data)[bn]);
++bn, ++*runp);
bn = ap->in_off;
if (runb && bn) {
for (--bn; bn >= 0 && *runb < maxrun &&
is_sequential(ump,
((ufs2_daddr_t *)bp->data)[bn],
((ufs2_daddr_t *)bp->data)[bn + 1]);
--bn, ++*runb);
}
}
}
if (bp)
releasebuf(bp);
/*
* Since this is FFS independent code, we are out of scope for the
* definitions of BLK_NOCOPY and BLK_SNAP, but we do know that they
* will fall in the range 1..um_seqinc, so we use that test and
* return a request for a zeroed out buffer if attempts are made
* to read a BLK_NOCOPY or BLK_SNAP block.
*/
if ((ip->i_flags & SF_SNAPSHOT) && daddr > 0 && daddr < ump->um_seqinc) {
*bnp = -1;
return (0);
}
*bnp = blkptrtodb(ump, daddr);
if (*bnp == 0) {
if (ip->i_flags & SF_SNAPSHOT)
*bnp = blkptrtodb(ump, bn * ump->um_seqinc);
else
*bnp = -1;
}
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.
*/
int
ffs1_balloc(struct inode *ip, off_t startoffset, int size, struct ucred *cred,
int flags, struct buf **bpp)
{
daddr_t lbn, nb, newb, pref;
struct fs *fs;
struct buf *bp, *nbp;
struct vnode *vp;
struct proc *p;
struct indir indirs[NIADDR + 2];
int32_t *bap;
int deallocated, osize, nsize, num, i, error;
int32_t *allocib, *blkp, *allocblk, allociblk[NIADDR+1];
int unwindidx = -1;
vp = ITOV(ip);
fs = ip->i_fs;
p = curproc;
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
if (size > fs->fs_bsize)
panic("ffs1_balloc: blk too big");
if (bpp != NULL)
*bpp = NULL;
if (lbn < 0)
return (EFBIG);
/*
* 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.
*/
nb = lblkno(fs, ip->i_ffs1_size);
if (nb < NDADDR && nb < lbn) {
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
error = ffs_realloccg(ip, nb,
ffs1_blkpref(ip, nb, (int)nb, &ip->i_ffs1_db[0]),
osize, (int)fs->fs_bsize, cred, bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb, newb,
ip->i_ffs1_db[nb], fs->fs_bsize, osize,
bpp ? *bpp : NULL);
ip->i_ffs1_size = lblktosize(fs, nb + 1);
uvm_vnp_setsize(vp, ip->i_ffs1_size);
ip->i_ffs1_db[nb] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp != NULL) {
if (flags & B_SYNC)
bwrite(*bpp);
else
bawrite(*bpp);
}
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_ffs1_db[lbn];
if (nb != 0 && ip->i_ffs1_size >= lblktosize(fs, lbn + 1)) {
/*
* The block is an already-allocated direct block
* and the file already extends past this block,
* thus this must be a whole block.
* Just read the block (if requested).
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize, bpp);
if (error) {
brelse(*bpp);
return (error);
}
}
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_ffs1_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
/*
* The existing block is already
* at least as big as we want.
* Just read the block (if requested).
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize,
bpp);
if (error) {
brelse(*bpp);
return (error);
}
(*bpp)->b_bcount = osize;
}
return (0);
} else {
/*
* The existing block is smaller than we
* want, grow it.
*/
error = ffs_realloccg(ip, lbn,
ffs1_blkpref(ip, lbn, (int)lbn,
&ip->i_ffs1_db[0]),
osize, nsize, cred, bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
newb, nb, nsize, osize,
bpp ? *bpp : NULL);
}
} else {
/*
* The block was not previously allocated,
* allocate a new block or fragment.
*/
if (ip->i_ffs1_size < lblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs1_blkpref(ip, lbn, (int)lbn, &ip->i_ffs1_db[0]),
nsize, cred, &newb);
if (error)
return (error);
if (bpp != NULL) {
*bpp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
if (nsize < fs->fs_bsize)
(*bpp)->b_bcount = nsize;
(*bpp)->b_blkno = fsbtodb(fs, newb);
if (flags & B_CLRBUF)
clrbuf(*bpp);
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bpp ? *bpp : NULL);
}
ip->i_ffs1_db[lbn] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef DIAGNOSTIC
if (num < 1)
panic ("ffs1_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = ip->i_ffs1_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ffs1_blkpref(ip, lbn, -indirs[0].in_off - 1, NULL);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
cred, &newb);
if (error)
goto fail;
nb = newb;
*allocblk++ = nb;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &ip->i_ffs1_ib[indirs[0].in_off];
*allocib = nb;
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->fs_bsize, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (int32_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i++;
if (nb != 0) {
brelse(bp);
continue;
}
if (pref == 0)
pref = ffs1_blkpref(ip, lbn, i - num - 1, NULL);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
&newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* 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 = ffs1_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
&newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
if (bpp != NULL) {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
*bpp = nbp;
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, bpp ? *bpp : NULL);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
return (0);
}
brelse(bp);
if (bpp != NULL) {
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->fs_bsize, &nbp);
if (error) {
brelse(nbp);
goto fail;
}
} else {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
}
*bpp = nbp;
}
return (0);
fail:
/*
* If we have failed to allocate any blocks, simply return the error.
* This is the usual case and avoids the need to fsync the file.
*/
if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
return (error);
/*
* If we have failed part way through block allocation, we have to
* deallocate any indirect blocks that we have allocated. We have to
* fsync the file before we start to get rid of all of its
* dependencies so that we do not leave them dangling. We have to sync
* it at the end so that the softdep code does not find any untracked
* changes. Although this is really slow, running out of disk space is
* not expected to be a common occurrence. The error return from fsync
* is ignored as we already have an error to return to the user.
*/
VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p);
for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
ffs_blkfree(ip, *blkp, fs->fs_bsize);
deallocated += fs->fs_bsize;
}
if (allocib != NULL) {
*allocib = 0;
} else if (unwindidx >= 0) {
int r;
r = bread(vp, indirs[unwindidx].in_lbn, (int)fs->fs_bsize, &bp);
if (r)
panic("Could not unwind indirect block, error %d", r);
bap = (int32_t *)bp->b_data;
bap[indirs[unwindidx].in_off] = 0;
if (flags & B_SYNC) {
bwrite(bp);
} else {
bdwrite(bp);
}
}
if (deallocated) {
/*
* Restore user's disk quota because allocation failed.
*/
(void)ufs_quota_free_blocks(ip, btodb(deallocated), cred);
ip->i_ffs1_blocks -= btodb(deallocated);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p);
return (error);
}
int
ffs2_balloc(struct inode *ip, off_t off, int size, struct ucred *cred,
int flags, struct buf **bpp)
{
daddr_t lbn, lastlbn, nb, newb, *blkp;
daddr_t pref, *allocblk, allociblk[NIADDR + 1];
daddr_t *bap, *allocib;
int deallocated, osize, nsize, num, i, error, unwindidx, r;
struct buf *bp, *nbp;
struct indir indirs[NIADDR + 2];
struct fs *fs;
struct vnode *vp;
struct proc *p;
vp = ITOV(ip);
fs = ip->i_fs;
p = curproc;
unwindidx = -1;
lbn = lblkno(fs, off);
size = blkoff(fs, off) + size;
if (size > fs->fs_bsize)
panic("ffs2_balloc: block too big");
if (bpp != NULL)
*bpp = NULL;
if (lbn < 0)
return (EFBIG);
/*
* 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_ffs2_size);
if (lastlbn < NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
error = ffs_realloccg(ip, nb, ffs2_blkpref(ip,
lastlbn, nb, &ip->i_ffs2_db[0]), osize,
(int) fs->fs_bsize, cred, bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb, newb,
ip->i_ffs2_db[nb], fs->fs_bsize, osize,
bpp ? *bpp : NULL);
ip->i_ffs2_size = lblktosize(fs, nb + 1);
uvm_vnp_setsize(vp, ip->i_ffs2_size);
ip->i_ffs2_db[nb] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp) {
if (flags & B_SYNC)
bwrite(*bpp);
else
bawrite(*bpp);
}
}
}
/*
* The first NDADDR blocks are direct.
*/
if (lbn < NDADDR) {
nb = ip->i_ffs2_db[lbn];
if (nb != 0 && ip->i_ffs2_size >= lblktosize(fs, lbn + 1)) {
/*
* The direct block is already allocated and the file
* extends past this block, thus this must be a whole
* block. Just read it, if requested.
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize, bpp);
if (error) {
brelse(*bpp);
return (error);
}
}
return (0);
}
if (nb != 0) {
/*
* Consider the need to allocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_ffs2_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
/*
* The existing block is already at least as
* big as we want. Just read it, if requested.
*/
if (bpp != NULL) {
error = bread(vp, lbn, fs->fs_bsize,
bpp);
if (error) {
brelse(*bpp);
return (error);
}
(*bpp)->b_bcount = osize;
}
return (0);
} else {
/*
* The existing block is smaller than we want,
* grow it.
*/
error = ffs_realloccg(ip, lbn,
ffs2_blkpref(ip, lbn, (int) lbn,
&ip->i_ffs2_db[0]), osize, nsize, cred,
bpp, &newb);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
newb, nb, nsize, osize,
bpp ? *bpp : NULL);
}
} else {
/*
* The block was not previously allocated, allocate a
* new block or fragment.
*/
if (ip->i_ffs2_size < lblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn, ffs2_blkpref(ip, lbn,
(int) lbn, &ip->i_ffs2_db[0]), nsize, cred, &newb);
if (error)
return (error);
if (bpp != NULL) {
bp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
if (nsize < fs->fs_bsize)
bp->b_bcount = nsize;
bp->b_blkno = fsbtodb(fs, newb);
if (flags & B_CLRBUF)
clrbuf(bp);
*bpp = bp;
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bpp ? *bpp : NULL);
}
ip->i_ffs2_db[lbn] = newb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
error = ufs_getlbns(vp, lbn, indirs, &num);
if (error)
return (error);
#ifdef DIAGNOSTIC
if (num < 1)
panic("ffs2_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Fetch the first indirect block allocating it necessary.
*/
--num;
nb = ip->i_ffs2_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
if (nb == 0) {
pref = ffs2_blkpref(ip, lbn, -indirs[0].in_off - 1, NULL);
error = ffs_alloc(ip, lbn, pref, (int) fs->fs_bsize, cred,
&newb);
if (error)
goto fail;
nb = newb;
*allocblk++ = nb;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks never
* point at garbage.
*/
error = bwrite(bp);
if (error)
goto fail;
}
unwindidx = 0;
allocib = &ip->i_ffs2_ib[indirs[0].in_off];
*allocib = nb;
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->fs_bsize, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (int64_t *) bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i++;
if (nb != 0) {
brelse(bp);
continue;
}
if (pref == 0)
pref = ffs2_blkpref(ip, lbn, i - num - 1, NULL);
error = ffs_alloc(ip, lbn, pref, (int) fs->fs_bsize, cred,
&newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks never
* point at garbage.
*/
error = bwrite(nbp);
if (error) {
brelse(bp);
goto fail;
}
}
if (unwindidx < 0)
unwindidx = i - 1;
bap[indirs[i - 1].in_off] = 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 = ffs2_blkpref(ip, lbn, indirs[num].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred,
&newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
if (bpp != NULL) {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & B_CLRBUF)
clrbuf(nbp);
*bpp = nbp;
}
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[num].in_off, nb, 0, bpp ? *bpp : NULL);
bap[indirs[num].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use delayed
* write.
*/
if (flags & B_SYNC)
bwrite(bp);
else
bdwrite(bp);
return (0);
}
brelse(bp);
if (bpp != NULL) {
if (flags & B_CLRBUF) {
error = bread(vp, lbn, (int)fs->fs_bsize, &nbp);
if (error) {
brelse(nbp);
goto fail;
}
} else {
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
clrbuf(nbp);
}
*bpp = nbp;
}
return (0);
fail:
/*
* If we have failed to allocate any blocks, simply return the error.
* This is the usual case and avoids the need to fsync the file.
*/
if (allocblk == allociblk && allocib == NULL && unwindidx == -1)
return (error);
/*
* If we have failed part way through block allocation, we have to
* deallocate any indirect blocks that we have allocated. We have to
* fsync the file before we start to get rid of all of its
* dependencies so that we do not leave them dangling. We have to sync
* it at the end so that the softdep code does not find any untracked
* changes. Although this is really slow, running out of disk space is
* not expected to be a common occurrence. The error return from fsync
* is ignored as we already have an error to return to the user.
*/
VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p);
if (unwindidx >= 0) {
/*
* First write out any buffers we've created to resolve their
* softdeps. This must be done in reverse order of creation so
* that we resolve the dependencies in one pass.
* Write the cylinder group buffers for these buffers too.
*/
for (i = num; i >= unwindidx; i--) {
if (i == 0)
break;
bp = getblk(vp, indirs[i].in_lbn, (int) fs->fs_bsize,
0, 0);
if (bp->b_flags & B_DELWRI) {
nb = fsbtodb(fs, cgtod(fs, dtog(fs,
dbtofsb(fs, bp->b_blkno))));
bwrite(bp);
bp = getblk(ip->i_devvp, nb,
(int) fs->fs_cgsize, 0, 0);
if (bp->b_flags & B_DELWRI)
bwrite(bp);
else {
bp->b_flags |= B_INVAL;
brelse(bp);
}
} else {
bp->b_flags |= B_INVAL;
brelse(bp);
}
}
if (DOINGSOFTDEP(vp) && unwindidx == 0) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
ffs_update(ip, 1);
}
/*
* Now that any dependencies that we created have been
* resolved, we can undo the partial allocation.
*/
if (unwindidx == 0) {
*allocib = 0;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (DOINGSOFTDEP(vp))
ffs_update(ip, 1);
} else {
r = bread(vp, indirs[unwindidx].in_lbn,
(int)fs->fs_bsize, &bp);
if (r)
panic("ffs2_balloc: unwind failed");
bap = (int64_t *) bp->b_data;
bap[indirs[unwindidx].in_off] = 0;
bwrite(bp);
}
for (i = unwindidx + 1; i <= num; i++) {
bp = getblk(vp, indirs[i].in_lbn, (int)fs->fs_bsize, 0,
0);
bp->b_flags |= B_INVAL;
brelse(bp);
}
}
for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
ffs_blkfree(ip, *blkp, fs->fs_bsize);
deallocated += fs->fs_bsize;
}
if (deallocated) {
/*
* Restore user's disk quota because allocation failed.
*/
(void) ufs_quota_free_blocks(ip, btodb(deallocated), cred);
ip->i_ffs2_blocks -= btodb(deallocated);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
VOP_FSYNC(vp, p->p_ucred, MNT_WAIT, p);
return (error);
}
int
ext2fs_bmaparray(struct vnode *vp,
#undef struct
daddr_t bn, daddr_t *bnp, struct indir *ap,
int *nump, int *runp)
{
struct inode *ip;
struct buf *bp, *cbp;
#define struct
// struct ufsmount *ump;
struct mount *mp;
#undef struct
struct indir a[NIADDR+1], *xap;
daddr_t daddr;
daddr_t metalbn;
int error, maxrun = 0, num;
ip = VTOI(vp);
mp = EXT2_SIMPLE_FILE_SYSTEM_PRIVATE_DATA_FROM_THIS(vp->Filesystem);
// mp = vp->v_mount; !!!! need to fix this badly!
// ump = ip->i_ump; NEED TO DO SOMETHING ABOUT ufsmount
#ifdef DIAGNOSTIC
if ((ap != NULL && nump == NULL) || (ap == NULL && nump != NULL))
panic("ext2fs_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; NEEDS FIX!!!
mp->fs->e2fs_bsize - 1;
}
if (bn >= 0 && bn < NDADDR) {
/* XXX ondisk32 */
*bnp = blkptrtodb(ump, fs2h32(ip->i_e2fs_blocks[bn]));
if (*bnp == 0)
*bnp = -1;
else if (runp)
/* XXX ondisk32 */
for (++bn; bn < NDADDR && *runp < maxrun &&
is_sequential(ump, (daddr_t)fs2h32(ip->i_e2fs_blocks[bn - 1]),
(daddr_t)fs2h32(ip->i_e2fs_blocks[bn]));
++bn, ++*runp);
return (0);
}
xap = ap == NULL ? a : ap;
if (!nump)
nump = #
if ((error = ufs_getlbns(vp, bn, xap, nump)) != 0)
return (error);
num = *nump;
/* Get disk address out of indirect block array */
/* XXX ondisk32 */
daddr = fs2h32(ip->i_e2fs_blocks[NDADDR + xap->in_off]);
#ifdef DIAGNOSTIC
if (num > NIADDR + 1 || num < 1) {
printf("ext2fs_bmaparray: num=%d\n", num);
panic("ext2fs_bmaparray: num");
}
#endif
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 (metalbn == bn)
break;
if (daddr == 0) {
mutex_enter(&bufcache_lock);
cbp = incore(vp, metalbn);
mutex_exit(&bufcache_lock);
if (cbp == NULL)
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, 0);
xap->in_exists = 1;
//!!!!!!!!!!!!!!replaced 3rd param with 1 ftw
bp = getblk(vp, metalbn, 1, 0, 0);
if (bp == NULL) {
/*
* getblk() above returns NULL only iff we are
* pagedaemon. See the implementation of getblk
* for detail.
*/
return (ENOMEM);
}
if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
trace(TR_BREADHIT, pack(vp, size), metalbn);
}
#ifdef DIAGNOSTIC
else if (!daddr)
panic("ext2fs_bmaparry: indirect block not in cache");
#endif
else {
trace(TR_BREADMISS, pack(vp, size), metalbn);
bp->b_blkno = blkptrtodb(ump, daddr);
bp->b_flags |= B_READ;
VOP_STRATEGY(vp, bp);
// curlwp->l_ru.ru_inblock++; *//* XXX */
if ((error = biowait(bp)) != 0) {
brelse(bp, 0);
return (error);
}
}
/* XXX ondisk32 */
daddr = fs2h32(((int32_t *)bp->b_data)[xap->in_off]);
if (num == 1 && daddr && runp)
/* XXX ondisk32 */
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, 0);
daddr = blkptrtodb(ump, daddr);
*bnp = daddr == 0 ? -1 : daddr;
return (0);
}
