/*
* Return buffer with the contents of block "offset" from the beginning of
* directory "ip". If "res" is non-zero, fill it in with a pointer to the
* remaining space in the directory.
*/
int
cd9660_blkatoff(vnode_t vp, off_t offset, char **res, buf_t *bpp)
{
struct iso_node *ip;
register struct iso_mnt *imp;
buf_t bp;
daddr_t lbn;
int bsize, error;
ip = VTOI(vp);
imp = ip->i_mnt;
lbn = lblkno(imp, offset);
bsize = blksize(imp, ip, lbn);
if ((bsize != imp->im_sector_size) &&
(offset & (imp->im_sector_size - 1)) == 0) {
bsize = imp->im_sector_size;
}
if ( (error = (int)buf_bread(vp, (daddr64_t)((unsigned)lbn), bsize, NOCRED, &bp)) ) {
buf_brelse(bp);
*bpp = NULL;
return (error);
}
if (res)
*res = (char *)0 + buf_dataptr(bp) + blkoff(imp, offset);
*bpp = bp;
return (0);
}
static int ReadMultipleNodes( BTScanState *theScanStatePtr )
{
int myErr = E_NONE;
BTreeControlBlockPtr myBTreeCBPtr;
daddr64_t myPhyBlockNum;
u_int32_t myBufferSize;
struct vnode * myDevPtr;
unsigned int myBlockRun;
u_int32_t myBlocksInBufferCount;
// release old buffer if we have one
if ( theScanStatePtr->bufferPtr != NULL )
{
buf_markinvalid(theScanStatePtr->bufferPtr);
buf_brelse( theScanStatePtr->bufferPtr );
theScanStatePtr->bufferPtr = NULL;
theScanStatePtr->currentNodePtr = NULL;
}
myBTreeCBPtr = theScanStatePtr->btcb;
// map logical block in catalog btree file to physical block on volume
myErr = hfs_bmap(myBTreeCBPtr->fileRefNum, theScanStatePtr->nodeNum,
&myDevPtr, &myPhyBlockNum, &myBlockRun);
if ( myErr != E_NONE )
{
goto ExitThisRoutine;
}
// bmap block run gives us the remaining number of valid blocks (number of blocks
// minus the first). so if there are 10 valid blocks our run number will be 9.
// blocks, in our case is the same as nodes (both are 4K)
myBlocksInBufferCount = (theScanStatePtr->bufferSize / myBTreeCBPtr->nodeSize );
myBufferSize = theScanStatePtr->bufferSize;
if ( (myBlockRun + 1) < myBlocksInBufferCount )
{
myBufferSize = (myBlockRun + 1) * myBTreeCBPtr->nodeSize;
}
// now read blocks from the device
myErr = (int)buf_bread(myDevPtr,
myPhyBlockNum,
myBufferSize,
NOCRED,
&theScanStatePtr->bufferPtr );
if ( myErr != E_NONE )
{
goto ExitThisRoutine;
}
theScanStatePtr->nodesLeftInBuffer = buf_count(theScanStatePtr->bufferPtr) / theScanStatePtr->btcb->nodeSize;
theScanStatePtr->currentNodePtr = (BTNodeDescriptor *) buf_dataptr(theScanStatePtr->bufferPtr);
ExitThisRoutine:
return myErr;
} /* ReadMultipleNodes */
/*
;_______________________________________________________________________
;
; Routine: ReadBitmapBlock
;
; Function: Read in a bitmap block corresponding to a given allocation
; block (bit). Return a pointer to the bitmap block.
;
; Inputs:
; vcb -- Pointer to ExtendedVCB
; bit -- Allocation block whose bitmap block is desired
;
; Outputs:
; buffer -- Pointer to bitmap block corresonding to "block"
; blockRef
;_______________________________________________________________________
*/
static OSErr ReadBitmapBlock(
ExtendedVCB *vcb,
u_int32_t bit,
u_int32_t **buffer,
u_int32_t *blockRef)
{
OSErr err;
struct buf *bp = NULL;
struct vnode *vp = NULL;
daddr64_t block;
u_int32_t blockSize;
/*
* volume bitmap blocks are protected by the allocation file lock
*/
REQUIRE_FILE_LOCK(vcb->hfs_allocation_vp, false);
blockSize = (u_int32_t)vcb->vcbVBMIOSize;
block = (daddr64_t)(bit / (blockSize * kBitsPerByte));
if (vcb->vcbSigWord == kHFSPlusSigWord) {
vp = vcb->hfs_allocation_vp; /* use allocation file vnode */
} else /* hfs */ {
vp = VCBTOHFS(vcb)->hfs_devvp; /* use device I/O vnode */
block += vcb->vcbVBMSt; /* map to physical block */
}
err = (int)buf_meta_bread(vp, block, blockSize, NOCRED, &bp);
if (bp) {
if (err) {
buf_brelse(bp);
*blockRef = 0;
*buffer = NULL;
} else {
*blockRef = (u_int32_t)bp;
*buffer = (u_int32_t *)buf_dataptr(bp);
}
}
return err;
}
/*
* Convert a component of a pathname into a pointer to a locked inode.
* This is a very central and rather complicated routine.
* If the file system is not maintained in a strict tree hierarchy,
* this can result in a deadlock situation (see comments in code below).
*
* The flag argument is LOOKUP, CREATE, RENAME, or DELETE depending on
* whether the name is to be looked up, created, renamed, or deleted.
* When CREATE, RENAME, or DELETE is specified, information usable in
* creating, renaming, or deleting a directory entry may be calculated.
* If flag has LOCKPARENT or'ed into it and the target of the pathname
* exists, lookup returns both the target and its parent directory locked.
* When creating or renaming and LOCKPARENT is specified, the target may
* not be ".". When deleting and LOCKPARENT is specified, the target may
* be "."., but the caller must check to ensure it does an vrele and iput
* instead of two iputs.
*
* Overall outline of ufs_lookup:
*
* check accessibility of directory
* look for name in cache, if found, then if at end of path
* and deleting or creating, drop it, else return name
* search for name in directory, to found or notfound
* notfound:
* if creating, return locked directory, leaving info on available slots
* else return error
* found:
* if at end of path and deleting, return information to allow delete
* if at end of path and rewriting (RENAME and LOCKPARENT), lock target
* inode and return info to allow rewrite
* if not at end, add name to cache; if at end and neither creating
* nor deleting, add name to cache
*
* NOTE: (LOOKUP | LOCKPARENT) currently returns the parent inode unlocked.
*/
int
cd9660_lookup(struct vnop_lookup_args *ap)
{
register struct vnode *vdp; /* vnode for directory being searched */
register struct iso_node *dp; /* inode for directory being searched */
register struct iso_mnt *imp; /* file system that directory is in */
struct buf *bp; /* a buffer of directory entries */
struct iso_directory_record *ep = NULL;/* the current directory entry */
int entryoffsetinblock; /* offset of ep in bp's buffer */
int saveoffset = 0; /* offset of last directory entry in dir */
int numdirpasses; /* strategy for directory search */
doff_t endsearch; /* offset to end directory search */
struct vnode *pdp; /* saved dp during symlink work */
struct vnode *tdp; /* returned by cd9660_vget_internal */
u_long bmask; /* block offset mask */
int lockparent; /* 1 => lockparent flag is set */
int wantparent; /* 1 => wantparent or lockparent flag */
int wantassoc;
int error;
ino_t ino = 0;
int reclen;
u_short namelen;
int isoflags;
char altname[ISO_RRIP_NAMEMAX];
int res;
int len;
char *name;
struct vnode **vpp = ap->a_vpp;
struct componentname *cnp = ap->a_cnp;
int flags = cnp->cn_flags;
int nameiop = cnp->cn_nameiop;
vfs_context_t ctx = cnp->cn_context;
size_t altlen;
bp = NULL;
*vpp = NULL;
vdp = ap->a_dvp;
dp = VTOI(vdp);
imp = dp->i_mnt;
lockparent = flags & LOCKPARENT;
wantparent = flags & (LOCKPARENT|WANTPARENT);
wantassoc = 0;
/*
* We now have a segment name to search for, and a directory to search.
*
* Before tediously performing a linear scan of the directory,
* check the name cache to see if the directory/name pair
* we are looking for is known already.
*/
if ((error = cache_lookup(vdp, vpp, cnp))) {
if (error == ENOENT)
return (error);
return (0);
}
len = cnp->cn_namelen;
name = cnp->cn_nameptr;
altname[0] = '\0';
/*
* A "._" prefix means, we are looking for an associated file
*/
if (imp->iso_ftype != ISO_FTYPE_RRIP &&
*name == ASSOCCHAR1 && *(name+1) == ASSOCCHAR2) {
wantassoc = 1;
len -= 2;
name += 2;
}
/*
* Decode search name into UCS-2 (Unicode)
*/
if ((imp->iso_ftype == ISO_FTYPE_JOLIET) &&
!((len == 1 && *name == '.') || (flags & ISDOTDOT))) {
int flags1 = UTF_PRECOMPOSED;
(void) utf8_decodestr(name, len, (u_int16_t*) altname, &altlen,
sizeof(altname), 0, flags1);
name = altname;
len = altlen;
}
/*
* If there is cached information on a previous search of
* this directory, pick up where we last left off.
* We cache only lookups as these are the most common
* and have the greatest payoff. Caching CREATE has little
* benefit as it usually must search the entire directory
* to determine that the entry does not exist. Caching the
* location of the last DELETE or RENAME has not reduced
* profiling time and hence has been removed in the interest
* of simplicity.
*/
bmask = imp->im_sector_size - 1;
if (nameiop != LOOKUP || dp->i_diroff == 0 ||
dp->i_diroff > dp->i_size) {
entryoffsetinblock = 0;
dp->i_offset = 0;
numdirpasses = 1;
} else {
dp->i_offset = dp->i_diroff;
if ((entryoffsetinblock = dp->i_offset & bmask) &&
(error = cd9660_blkatoff(vdp, SECTOFF(imp, dp->i_offset), NULL, &bp)))
return (error);
numdirpasses = 2;
iso_nchstats.ncs_2passes++;
}
endsearch = dp->i_size;
searchloop:
while (dp->i_offset < endsearch) {
/*
* If offset is on a block boundary,
* read the next directory block.
* Release previous if it exists.
*/
if ((dp->i_offset & bmask) == 0) {
if (bp != NULL)
buf_brelse(bp);
if ( (error = cd9660_blkatoff(vdp, SECTOFF(imp,dp->i_offset), NULL, &bp)) )
return (error);
entryoffsetinblock = 0;
}
/*
* Get pointer to next entry.
*/
ep = (struct iso_directory_record *)
((char *)0 + buf_dataptr(bp) + entryoffsetinblock);
reclen = isonum_711(ep->length);
if (reclen == 0) {
/* skip to next block, if any */
dp->i_offset =
(dp->i_offset & ~bmask) + imp->im_sector_size;
continue;
}
if (reclen < ISO_DIRECTORY_RECORD_SIZE) {
/* illegal entry, stop */
break;
}
if (entryoffsetinblock + reclen > imp->im_sector_size) {
/* entries are not allowed to cross sector boundaries */
break;
}
namelen = isonum_711(ep->name_len);
isoflags = isonum_711(ep->flags);
if (reclen < ISO_DIRECTORY_RECORD_SIZE + namelen)
/* illegal entry, stop */
break;
/*
* Check for a name match.
*/
if (imp->iso_ftype == ISO_FTYPE_RRIP) {
if (isoflags & directoryBit)
ino = isodirino(ep, imp);
else
ino = ((daddr_t)buf_blkno(bp) << imp->im_bshift) + entryoffsetinblock;
dp->i_ino = ino;
cd9660_rrip_getname(ep,altname,&namelen,&dp->i_ino,imp);
if (namelen == cnp->cn_namelen
&& !bcmp(name,altname,namelen))
goto found;
ino = 0;
} else {
if ((!(isoflags & associatedBit)) == !wantassoc) {
if ((len == 1
&& *name == '.')
|| (flags & ISDOTDOT)) {
if (namelen == 1
&& ep->name[0] == ((flags & ISDOTDOT) ? 1 : 0)) {
/*
* Save directory entry's inode number and
* release directory buffer.
*/
dp->i_ino = isodirino(ep, imp);
goto found;
}
if (namelen != 1
|| ep->name[0] != 0)
goto notfound;
} else if (imp->iso_ftype != ISO_FTYPE_JOLIET && !(res = isofncmp(name, len, ep->name, namelen))) {
if ( isoflags & directoryBit )
ino = isodirino(ep, imp);
else
ino = ((daddr_t)buf_blkno(bp) << imp->im_bshift) + entryoffsetinblock;
saveoffset = dp->i_offset;
} else if (imp->iso_ftype == ISO_FTYPE_JOLIET && !(res = ucsfncmp((u_int16_t*)name, len,
(u_int16_t*) ep->name, namelen))) {
if ( isoflags & directoryBit )
ino = isodirino(ep, imp);
else
ino = ((daddr_t)buf_blkno(bp) << imp->im_bshift) + entryoffsetinblock;
saveoffset = dp->i_offset;
} else if (ino)
goto foundino;
#ifdef NOSORTBUG /* On some CDs directory entries are not sorted correctly */
else if (res < 0)
goto notfound;
else if (res > 0 && numdirpasses == 2)
numdirpasses++;
#endif
}
}
dp->i_offset += reclen;
entryoffsetinblock += reclen;
} /* endwhile */
if (ino) {
foundino:
dp->i_ino = ino;
if (saveoffset != dp->i_offset) {
if (lblkno(imp, dp->i_offset) !=
lblkno(imp, saveoffset)) {
if (bp != NULL)
buf_brelse(bp);
if ( (error = cd9660_blkatoff(vdp, SECTOFF(imp, saveoffset), NULL, &bp)) )
return (error);
}
entryoffsetinblock = saveoffset & bmask;
ep = (struct iso_directory_record *)
((char *)0 + buf_dataptr(bp) + entryoffsetinblock);
dp->i_offset = saveoffset;
}
goto found;
}
notfound:
/*
* If we started in the middle of the directory and failed
* to find our target, we must check the beginning as well.
*/
if (numdirpasses == 2) {
numdirpasses--;
dp->i_offset = 0;
endsearch = dp->i_diroff;
goto searchloop;
}
if (bp != NULL)
buf_brelse(bp);
/*
* Insert name into cache (as non-existent) if appropriate.
*/
if (cnp->cn_flags & MAKEENTRY)
cache_enter(vdp, *vpp, cnp);
return (ENOENT);
found:
if (numdirpasses == 2)
iso_nchstats.ncs_pass2++;
/*
* Found component in pathname.
* If the final component of path name, save information
* in the cache as to where the entry was found.
*/
if ((flags & ISLASTCN) && nameiop == LOOKUP)
dp->i_diroff = dp->i_offset;
/*
* Step through the translation in the name. We do not `iput' the
* directory because we may need it again if a symbolic link
* is relative to the current directory. Instead we save it
* unlocked as "pdp". We must get the target inode before unlocking
* the directory to insure that the inode will not be removed
* before we get it. We prevent deadlock by always fetching
* inodes from the root, moving down the directory tree. Thus
* when following backward pointers ".." we must unlock the
* parent directory before getting the requested directory.
* There is a potential race condition here if both the current
* and parent directories are removed before the `iget' for the
* inode associated with ".." returns. We hope that this occurs
* infrequently since we cannot avoid this race condition without
* implementing a sophisticated deadlock detection algorithm.
* Note also that this simple deadlock detection scheme will not
* work if the file system has any hard links other than ".."
* that point backwards in the directory structure.
*/
pdp = vdp;
/*
* If ino is different from dp->i_ino,
* it's a relocated directory.
*/
if (flags & ISDOTDOT) {
error = cd9660_vget_internal(vnode_mount(vdp), dp->i_ino, &tdp, NULL, NULL,
dp->i_ino != ino, ep,
vfs_context_proc(ctx));
VTOI(tdp)->i_parent = VTOI(pdp)->i_number;
buf_brelse(bp);
*vpp = tdp;
} else if (dp->i_number == dp->i_ino) {
buf_brelse(bp);
vnode_get(vdp); /* we want ourself, ie "." */
*vpp = vdp;
} else {
error = cd9660_vget_internal(vnode_mount(vdp), dp->i_ino, &tdp, vdp, cnp,
dp->i_ino != ino, ep, vfs_context_proc(ctx));
/* save parent inode number */
VTOI(tdp)->i_parent = VTOI(pdp)->i_number;
buf_brelse(bp);
if (error)
return (error);
*vpp = tdp;
}
return (0);
}
/*
* Vnode op for write
*/
int
spec_write(struct vnop_write_args *ap)
{
struct vnode *vp = ap->a_vp;
struct uio *uio = ap->a_uio;
struct buf *bp;
daddr64_t bn;
int bsize, blkmask, bscale;
int io_sync;
int devBlockSize=0;
int n, on;
int error = 0;
dev_t dev;
#if DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("spec_write mode");
if (UIO_SEG_IS_USER_SPACE(uio->uio_segflg))
panic("spec_write proc");
#endif
switch (vp->v_type) {
case VCHR:
error = (*cdevsw[major(vp->v_rdev)].d_write)
(vp->v_rdev, uio, ap->a_ioflag);
return (error);
case VBLK:
if (uio_resid(uio) == 0)
return (0);
if (uio->uio_offset < 0)
return (EINVAL);
io_sync = (ap->a_ioflag & IO_SYNC);
dev = (vp->v_rdev);
devBlockSize = vp->v_specsize;
if (devBlockSize > PAGE_SIZE)
return(EINVAL);
bscale = PAGE_SIZE / devBlockSize;
blkmask = bscale - 1;
bsize = bscale * devBlockSize;
do {
bn = (daddr64_t)((uio->uio_offset / devBlockSize) &~ blkmask);
on = uio->uio_offset % bsize;
n = min((unsigned)(bsize - on), uio_resid(uio));
/*
* Use buf_getblk() as an optimization IFF:
*
* 1) We are reading exactly a block on a block
* aligned boundary
* 2) We know the size of the device from spec_open
* 3) The read doesn't span the end of the device
*
* Otherwise, we fall back on buf_bread().
*/
if (n == bsize &&
vp->v_specdevsize != (u_int64_t)0 &&
(uio->uio_offset + (u_int64_t)n) > vp->v_specdevsize) {
/* reduce the size of the read to what is there */
n = (uio->uio_offset + (u_int64_t)n) - vp->v_specdevsize;
}
if (n == bsize)
bp = buf_getblk(vp, bn, bsize, 0, 0, BLK_WRITE);
else
error = (int)buf_bread(vp, bn, bsize, NOCRED, &bp);
/* Translate downstream error for upstream, if needed */
if (!error)
error = (int)buf_error(bp);
if (error) {
buf_brelse(bp);
return (error);
}
n = min(n, bsize - buf_resid(bp));
error = uiomove((char *)0 + buf_dataptr(bp) + on, n, uio);
if (error) {
buf_brelse(bp);
return (error);
}
buf_markaged(bp);
if (io_sync)
error = buf_bwrite(bp);
else {
if ((n + on) == bsize)
error = buf_bawrite(bp);
else
error = buf_bdwrite(bp);
}
} while (error == 0 && uio_resid(uio) > 0 && n != 0);
return (error);
default:
panic("spec_write type");
}
/* NOTREACHED */
return (0);
}
/*
* Vnode op for read
*/
int
spec_read(struct vnop_read_args *ap)
{
struct vnode *vp = ap->a_vp;
struct uio *uio = ap->a_uio;
struct buf *bp;
daddr64_t bn, nextbn;
long bsize, bscale;
int devBlockSize=0;
int n, on;
int error = 0;
dev_t dev;
#if DIAGNOSTIC
if (uio->uio_rw != UIO_READ)
panic("spec_read mode");
if (UIO_SEG_IS_USER_SPACE(uio->uio_segflg))
panic("spec_read proc");
#endif
if (uio_resid(uio) == 0)
return (0);
switch (vp->v_type) {
case VCHR:
error = (*cdevsw[major(vp->v_rdev)].d_read)
(vp->v_rdev, uio, ap->a_ioflag);
return (error);
case VBLK:
if (uio->uio_offset < 0)
return (EINVAL);
dev = vp->v_rdev;
devBlockSize = vp->v_specsize;
if (devBlockSize > PAGE_SIZE)
return (EINVAL);
bscale = PAGE_SIZE / devBlockSize;
bsize = bscale * devBlockSize;
do {
on = uio->uio_offset % bsize;
bn = (daddr64_t)((uio->uio_offset / devBlockSize) &~ (bscale - 1));
if (vp->v_speclastr + bscale == bn) {
nextbn = bn + bscale;
error = buf_breadn(vp, bn, (int)bsize, &nextbn,
(int *)&bsize, 1, NOCRED, &bp);
} else
error = buf_bread(vp, bn, (int)bsize, NOCRED, &bp);
vnode_lock(vp);
vp->v_speclastr = bn;
vnode_unlock(vp);
n = bsize - buf_resid(bp);
if ((on > n) || error) {
if (!error)
error = EINVAL;
buf_brelse(bp);
return (error);
}
n = min((unsigned)(n - on), uio_resid(uio));
error = uiomove((char *)0 + buf_dataptr(bp) + on, n, uio);
if (n + on == bsize)
buf_markaged(bp);
buf_brelse(bp);
} while (error == 0 && uio_resid(uio) > 0 && n != 0);
return (error);
default:
panic("spec_read type");
}
/* NOTREACHED */
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);
}
/*
* 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);
}
int
physio( void (*f_strategy)(buf_t),
buf_t bp,
dev_t dev,
int flags,
u_int (*f_minphys)(buf_t),
struct uio *uio,
int blocksize)
{
struct proc *p = current_proc();
int error, i, buf_allocated, todo, iosize;
int orig_bflags = 0;
int64_t done;
error = 0;
flags &= B_READ | B_WRITE;
buf_allocated = 0;
/*
* [check user read/write access to the data buffer]
*
* Check each iov one by one. Note that we know if we're reading or
* writing, so we ignore the uio's rw parameter. Also note that if
* we're doing a read, that's a *write* to user-space.
*/
for (i = 0; i < uio->uio_iovcnt; i++) {
if (UIO_SEG_IS_USER_SPACE(uio->uio_segflg)) {
user_addr_t base;
user_size_t len;
if (uio_getiov(uio, i, &base, &len) ||
!useracc(base,
len,
(flags == B_READ) ? B_WRITE : B_READ))
return (EFAULT);
}
}
/*
* Make sure we have a buffer, creating one if necessary.
*/
if (bp == NULL) {
bp = buf_alloc((vnode_t)0);
buf_allocated = 1;
} else
orig_bflags = buf_flags(bp);
/*
* at this point we should have a buffer
* that is marked BL_BUSY... we either
* acquired it via buf_alloc, or it was
* passed into us... if it was passed
* in, it needs to already be owned by
* the caller (i.e. BL_BUSY is set)
*/
assert(bp->b_lflags & BL_BUSY);
/*
* [set up the fixed part of the buffer for a transfer]
*/
bp->b_dev = dev;
bp->b_proc = p;
/*
* [mark the buffer busy for physical I/O]
* (i.e. set B_PHYS (because it's an I/O to user
* memory, and B_RAW, because B_RAW is to be
* "Set by physio for raw transfers.", in addition
* to the read/write flag.)
*/
buf_setflags(bp, B_PHYS | B_RAW);
/*
* [while there is data to transfer and no I/O error]
* Note that I/O errors are handled with a 'goto' at the bottom
* of the 'while' loop.
*/
while (uio_resid(uio) > 0) {
if ( (iosize = uio_curriovlen(uio)) > MAXPHYSIO_WIRED)
iosize = MAXPHYSIO_WIRED;
/*
* make sure we're set to issue a fresh I/O
* in the right direction
*/
buf_reset(bp, flags);
/* [set up the buffer for a maximum-sized transfer] */
buf_setblkno(bp, uio_offset(uio) / blocksize);
buf_setcount(bp, iosize);
buf_setdataptr(bp, (uintptr_t)CAST_DOWN(caddr_t, uio_curriovbase(uio)));
/*
* [call f_minphys to bound the tranfer size]
* and remember the amount of data to transfer,
* for later comparison.
*/
(*f_minphys)(bp);
todo = buf_count(bp);
/*
* [lock the part of the user address space involved
* in the transfer]
*/
if(UIO_SEG_IS_USER_SPACE(uio->uio_segflg)) {
error = vslock(CAST_USER_ADDR_T(buf_dataptr(bp)),
(user_size_t)todo);
if (error)
goto done;
}
/* [call f_strategy to start the transfer] */
(*f_strategy)(bp);
/* [wait for the transfer to complete] */
error = (int)buf_biowait(bp);
/*
* [unlock the part of the address space previously
* locked]
*/
if(UIO_SEG_IS_USER_SPACE(uio->uio_segflg))
vsunlock(CAST_USER_ADDR_T(buf_dataptr(bp)),
(user_size_t)todo,
(flags & B_READ));
/*
* [deduct the transfer size from the total number
* of data to transfer]
*/
done = buf_count(bp) - buf_resid(bp);
uio_update(uio, done);
/*
* Now, check for an error.
* Also, handle weird end-of-disk semantics.
*/
if (error || done < todo)
goto done;
}
done:
if (buf_allocated)
buf_free(bp);
else
buf_setflags(bp, orig_bflags);
return (error);
}
