static int
vn_readwrite_io(struct vn_softc * vn, struct buf * bp, vfs_context_t ctx)
{
int error = 0;
char * iov_base;
caddr_t vaddr;
if (buf_map(bp, &vaddr))
panic("vn device: buf_map failed");
iov_base = (char *)vaddr;
if (vn->sc_shadow_vp == NULL) {
user_ssize_t temp_resid;
error = file_io(vn->sc_vp, ctx,
buf_flags(bp) & B_READ ? UIO_READ : UIO_WRITE,
iov_base,
(off_t)buf_blkno(bp) * vn->sc_secsize,
buf_resid(bp), &temp_resid);
buf_setresid(bp, temp_resid);
}
else {
if (buf_flags(bp) & B_READ)
error = shadow_read(vn, bp, iov_base, ctx);
else
error = shadow_write(vn, bp, iov_base, ctx);
}
buf_unmap(bp);
return (error);
}
static int
vnop_strategy_9p(struct vnop_strategy_args *ap)
{
mount_t mp;
struct buf *bp;
node_9p *np;
caddr_t addr;
uio_t uio;
int e, flags;
TRACE();
bp = ap->a_bp;
np = NTO9P(buf_vnode(bp));
flags = buf_flags(bp);
uio = NULL;
addr = NULL;
mp = vnode_mount(buf_vnode(bp));
if (mp == NULL)
return ENXIO;
if ((e=buf_map(bp, &addr)))
goto error;
uio = uio_create(1, buf_blkno(bp) * vfs_statfs(mp)->f_bsize, UIO_SYSSPACE,
ISSET(flags, B_READ)? UIO_READ: UIO_WRITE);
if (uio == NULL) {
e = ENOMEM;
goto error;
}
uio_addiov(uio, CAST_USER_ADDR_T(addr), buf_count(bp));
if (ISSET(flags, B_READ)) {
if((e=nread_9p(np, uio)))
goto error;
/* zero the rest of the page if we reached EOF */
if (uio_resid(uio) > 0) {
bzero(addr+buf_count(bp)-uio_resid(uio), uio_resid(uio));
uio_update(uio, uio_resid(uio));
}
} else {
if ((e=nwrite_9p(np, uio)))
goto error;
}
buf_setresid(bp, uio_resid(uio));
error:
if (uio)
uio_free(uio);
if (addr)
buf_unmap(bp);
buf_seterror(bp, e);
buf_biodone(bp);
return e;
}
static int
shadow_write(struct vn_softc * vn, struct buf * bp, char * base,
vfs_context_t ctx)
{
u_int32_t blocksize = vn->sc_secsize;
int error = 0;
u_int32_t offset;
boolean_t shadow_grew;
u_int32_t resid;
u_int32_t start = 0;
offset = buf_blkno(bp);
resid = buf_resid(bp) / blocksize;
while (resid > 0) {
user_ssize_t temp_resid;
u_int32_t this_offset;
u_int32_t this_resid;
shadow_grew = shadow_map_write(vn->sc_shadow_map,
offset, resid,
&this_offset, &this_resid);
if (shadow_grew) {
#if 0
off_t size;
/* truncate the file to its new length before write */
size = (off_t)shadow_map_shadow_size(vn->sc_shadow_map)
* blocksize;
vnode_setsize(vn->sc_shadow_vp, size, IO_SYNC, ctx);
#endif
}
error = file_io(vn->sc_shadow_vp, ctx, UIO_WRITE,
base + start,
(off_t)this_offset * blocksize,
(user_ssize_t)this_resid * blocksize,
&temp_resid);
if (error) {
break;
}
this_resid -= (temp_resid / blocksize);
if (this_resid == 0) {
printf("vn device: shadow_write zero length write\n");
break;
}
resid -= this_resid;
offset += this_resid;
start += this_resid * blocksize;
}
buf_setresid(bp, resid * blocksize);
return (error);
}
static int
shadow_read(struct vn_softc * vn, struct buf * bp, char * base,
vfs_context_t ctx)
{
u_int32_t blocksize = vn->sc_secsize;
int error = 0;
u_int32_t offset;
boolean_t read_shadow;
u_int32_t resid;
u_int32_t start = 0;
offset = buf_blkno(bp);
resid = buf_resid(bp) / blocksize;
while (resid > 0) {
user_ssize_t temp_resid;
u_int32_t this_offset;
u_int32_t this_resid;
struct vnode * vp;
read_shadow = shadow_map_read(vn->sc_shadow_map,
offset, resid,
&this_offset, &this_resid);
if (read_shadow) {
vp = vn->sc_shadow_vp;
}
else {
vp = vn->sc_vp;
}
error = file_io(vp, ctx, UIO_READ, base + start,
(off_t)this_offset * blocksize,
(user_ssize_t)this_resid * blocksize,
&temp_resid);
if (error) {
break;
}
this_resid -= (temp_resid / blocksize);
if (this_resid == 0) {
printf("vn device: shadow_read zero length read\n");
break;
}
resid -= this_resid;
offset += this_resid;
start += this_resid * blocksize;
}
buf_setresid(bp, resid * blocksize);
return (error);
}
static void mdevstrategy(struct buf *bp) {
unsigned int left, lop, csize;
vm_offset_t vaddr, blkoff;
int devid;
addr64_t paddr, fvaddr;
ppnum_t pp;
devid = minor(buf_device(bp)); /* Get minor device number */
if ((mdev[devid].mdFlags & mdInited) == 0) { /* Have we actually been defined yet? */
buf_seterror(bp, ENXIO);
buf_biodone(bp);
return;
}
buf_setresid(bp, buf_count(bp)); /* Set byte count */
blkoff = buf_blkno(bp) * mdev[devid].mdSecsize; /* Get offset into file */
/*
* Note that reading past end is an error, but reading at end is an EOF. For these
* we just return with resid == count.
*/
if (blkoff >= (mdev[devid].mdSize << 12)) { /* Are they trying to read/write at/after end? */
if(blkoff != (mdev[devid].mdSize << 12)) { /* Are we trying to read after EOF? */
buf_seterror(bp, EINVAL); /* Yeah, this is an error */
}
buf_biodone(bp); /* Return */
return;
}
if ((blkoff + buf_count(bp)) > (mdev[devid].mdSize << 12)) { /* Will this read go past end? */
buf_setcount(bp, ((mdev[devid].mdSize << 12) - blkoff)); /* Yes, trim to max */
}
/*
* make sure the buffer's data area is
* accessible
*/
if (buf_map(bp, (caddr_t *)&vaddr))
panic("ramstrategy: buf_map failed\n");
fvaddr = (mdev[devid].mdBase << 12) + blkoff; /* Point to offset into ram disk */
if (buf_flags(bp) & B_READ) { /* Is this a read? */
if(!(mdev[devid].mdFlags & mdPhys)) { /* Physical mapped disk? */
bcopy((void *)((uintptr_t)fvaddr),
(void *)vaddr, (size_t)buf_count(bp)); /* This is virtual, just get the data */
}
else {
left = buf_count(bp); /* Init the amount left to copy */
while(left) { /* Go until it is all copied */
lop = min((4096 - (vaddr & 4095)), (4096 - (fvaddr & 4095))); /* Get smallest amount left on sink and source */
csize = min(lop, left); /* Don't move more than we need to */
pp = pmap_find_phys(kernel_pmap, (addr64_t)((uintptr_t)vaddr)); /* Get the sink physical address */
if(!pp) { /* Not found, what gives? */
panic("mdevstrategy: sink address %016llX not mapped\n", (addr64_t)((uintptr_t)vaddr));
}
paddr = (addr64_t)(((addr64_t)pp << 12) | (addr64_t)(vaddr & 4095)); /* Get actual address */
bcopy_phys(fvaddr, paddr, csize); /* Copy this on in */
mapping_set_mod(paddr >> 12); /* Make sure we know that it is modified */
left = left - csize; /* Calculate what is left */
vaddr = vaddr + csize; /* Move to next sink address */
fvaddr = fvaddr + csize; /* Bump to next physical address */
}
}
}
else { /* This is a write */
if(!(mdev[devid].mdFlags & mdPhys)) { /* Physical mapped disk? */
static void
vnstrategy(struct buf *bp)
{
struct vn_softc *vn;
int error = 0;
long sz; /* in sc_secsize chunks */
daddr64_t blk_num;
struct vnode * shadow_vp = NULL;
struct vnode * vp = NULL;
struct vfs_context context;
vn = vn_table + vnunit(buf_device(bp));
if ((vn->sc_flags & VNF_INITED) == 0) {
error = ENXIO;
goto done;
}
context.vc_thread = current_thread();
context.vc_ucred = vn->sc_cred;
buf_setresid(bp, buf_count(bp));
/*
* Check for required alignment. Transfers must be a valid
* multiple of the sector size.
*/
blk_num = buf_blkno(bp);
if (buf_count(bp) % vn->sc_secsize != 0) {
error = EINVAL;
goto done;
}
sz = howmany(buf_count(bp), vn->sc_secsize);
/*
* If out of bounds return an error. If at the EOF point,
* simply read or write less.
*/
if (blk_num >= 0 && (u_int64_t)blk_num >= vn->sc_size) {
if (blk_num > 0 && (u_int64_t)blk_num > vn->sc_size) {
error = EINVAL;
}
goto done;
}
/*
* If the request crosses EOF, truncate the request.
*/
if ((blk_num + sz) > 0 && ((u_int64_t)(blk_num + sz)) > vn->sc_size) {
buf_setcount(bp, (vn->sc_size - blk_num) * vn->sc_secsize);
buf_setresid(bp, buf_count(bp));
}
vp = vn->sc_vp;
if (vp == NULL) {
error = ENXIO;
goto done;
}
error = vnode_getwithvid(vp, vn->sc_vid);
if (error != 0) {
/* the vnode is no longer available, abort */
error = ENXIO;
vnclear(vn, &context);
goto done;
}
shadow_vp = vn->sc_shadow_vp;
if (shadow_vp != NULL) {
error = vnode_getwithvid(shadow_vp,
vn->sc_shadow_vid);
if (error != 0) {
/* the vnode is no longer available, abort */
error = ENXIO;
vnode_put(vn->sc_vp);
vnclear(vn, &context);
goto done;
}
}
error = vn_readwrite_io(vn, bp, &context);
vnode_put(vp);
if (shadow_vp != NULL) {
vnode_put(shadow_vp);
}
done:
if (error) {
buf_seterror(bp, error);
}
buf_biodone(bp);
return;
}
/*
* 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);
}
int
spec_strategy(struct vnop_strategy_args *ap)
{
buf_t bp;
int bflags;
int policy;
dev_t bdev;
uthread_t ut;
size_t devbsdunit;
mount_t mp;
bp = ap->a_bp;
bdev = buf_device(bp);
bflags = buf_flags(bp);
mp = buf_vnode(bp)->v_mount;
if (kdebug_enable) {
int code = 0;
if (bflags & B_READ)
code |= DKIO_READ;
if (bflags & B_ASYNC)
code |= DKIO_ASYNC;
if (bflags & B_META)
code |= DKIO_META;
else if (bflags & B_PAGEIO)
code |= DKIO_PAGING;
KERNEL_DEBUG_CONSTANT(FSDBG_CODE(DBG_DKRW, code) | DBG_FUNC_NONE,
bp, bdev, (int)buf_blkno(bp), buf_count(bp), 0);
}
if (((bflags & (B_IOSTREAMING | B_PAGEIO | B_READ)) == (B_PAGEIO | B_READ)) &&
mp && (mp->mnt_kern_flag & MNTK_ROOTDEV))
hard_throttle_on_root = 1;
if (mp != NULL)
devbsdunit = mp->mnt_devbsdunit;
else
devbsdunit = LOWPRI_MAX_NUM_DEV - 1;
throttle_info_update(&_throttle_io_info[devbsdunit], bflags);
if ((policy = throttle_get_io_policy(&ut)) == IOPOL_THROTTLE) {
bp->b_flags |= B_THROTTLED_IO;
}
if ((bflags & B_READ) == 0) {
microuptime(&_throttle_io_info[devbsdunit].last_IO_timestamp);
if (mp) {
INCR_PENDING_IO(buf_count(bp), mp->mnt_pending_write_size);
}
} else if (mp) {
INCR_PENDING_IO(buf_count(bp), mp->mnt_pending_read_size);
}
(*bdevsw[major(bdev)].d_strategy)(bp);
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);
}
__private_extern__
int
fuse_internal_strategy(vnode_t vp, buf_t bp)
{
size_t biosize;
size_t chunksize;
size_t respsize;
int mapped = FALSE;
int mode;
int op;
int vtype = vnode_vtype(vp);
int err = 0;
caddr_t bufdat;
off_t left;
off_t offset;
int32_t bflags = buf_flags(bp);
fufh_type_t fufh_type;
struct fuse_dispatcher fdi;
struct fuse_data *data;
struct fuse_vnode_data *fvdat = VTOFUD(vp);
struct fuse_filehandle *fufh = NULL;
mount_t mp = vnode_mount(vp);
data = fuse_get_mpdata(mp);
biosize = data->blocksize;
if (!(vtype == VREG || vtype == VDIR)) {
return ENOTSUP;
}
if (bflags & B_READ) {
mode = FREAD;
fufh_type = FUFH_RDONLY; /* FUFH_RDWR will also do */
} else {
mode = FWRITE;
fufh_type = FUFH_WRONLY; /* FUFH_RDWR will also do */
}
if (fvdat->flag & FN_CREATING) {
fuse_lck_mtx_lock(fvdat->createlock);
if (fvdat->flag & FN_CREATING) {
(void)fuse_msleep(fvdat->creator, fvdat->createlock,
PDROP | PINOD | PCATCH, "fuse_internal_strategy",
NULL);
} else {
fuse_lck_mtx_unlock(fvdat->createlock);
}
}
fufh = &(fvdat->fufh[fufh_type]);
if (!FUFH_IS_VALID(fufh)) {
fufh_type = FUFH_RDWR;
fufh = &(fvdat->fufh[fufh_type]);
if (!FUFH_IS_VALID(fufh)) {
fufh = NULL;
} else {
/* We've successfully fallen back to FUFH_RDWR. */
}
}
if (!fufh) {
if (mode == FREAD) {
fufh_type = FUFH_RDONLY;
} else {
fufh_type = FUFH_RDWR;
}
/*
* Lets NOT do the filehandle preflight check here.
*/
err = fuse_filehandle_get(vp, NULL, fufh_type, 0 /* mode */);
if (!err) {
fufh = &(fvdat->fufh[fufh_type]);
FUFH_AUX_INC(fufh);
/* We've created a NEW fufh of type fufh_type. open_count is 1. */
}
} else { /* good fufh */
FUSE_OSAddAtomic(1, (SInt32 *)&fuse_fh_reuse_count);
/* We're using an existing fufh of type fufh_type. */
}
if (err) {
/* A more typical error case. */
if ((err == ENOTCONN) || fuse_isdeadfs(vp)) {
buf_seterror(bp, EIO);
buf_biodone(bp);
return EIO;
}
IOLog("MacFUSE: strategy failed to get fh "
"(vtype=%d, fufh_type=%d, err=%d)\n", vtype, fufh_type, err);
if (!vfs_issynchronous(mp)) {
IOLog("MacFUSE: asynchronous write failed!\n");
}
buf_seterror(bp, EIO);
buf_biodone(bp);
return EIO;
}
if (!fufh) {
panic("MacFUSE: tried everything but still no fufh");
/* NOTREACHED */
}
#define B_INVAL 0x00040000 /* Does not contain valid info. */
#define B_ERROR 0x00080000 /* I/O error occurred. */
if (bflags & B_INVAL) {
IOLog("MacFUSE: buffer does not contain valid information\n");
}
if (bflags & B_ERROR) {
IOLog("MacFUSE: an I/O error has occured\n");
}
if (buf_count(bp) == 0) {
return 0;
}
fdisp_init(&fdi, 0);
if (mode == FREAD) {
struct fuse_read_in *fri;
buf_setresid(bp, buf_count(bp));
offset = (off_t)((off_t)buf_blkno(bp) * biosize);
if (offset >= fvdat->filesize) {
/* Trying to read at/after EOF? */
if (offset != fvdat->filesize) {
/* Trying to read after EOF? */
buf_seterror(bp, EINVAL);
}
buf_biodone(bp);
return 0;
}
/* Note that we just made sure that offset < fvdat->filesize. */
if ((offset + buf_count(bp)) > fvdat->filesize) {
/* Trimming read */
buf_setcount(bp, (uint32_t)(fvdat->filesize - offset));
}
if (buf_map(bp, &bufdat)) {
IOLog("MacFUSE: failed to map buffer in strategy\n");
return EFAULT;
} else {
mapped = TRUE;
}
while (buf_resid(bp) > 0) {
chunksize = min((size_t)buf_resid(bp), data->iosize);
fdi.iosize = sizeof(*fri);
op = FUSE_READ;
if (vtype == VDIR) {
op = FUSE_READDIR;
}
fdisp_make_vp(&fdi, op, vp, (vfs_context_t)0);
fri = fdi.indata;
fri->fh = fufh->fh_id;
/*
* Historical note:
*
* fri->offset = ((off_t)(buf_blkno(bp))) * biosize;
*
* This wasn't being incremented!?
*/
fri->offset = offset;
fri->size = (typeof(fri->size))chunksize;
fdi.tick->tk_aw_type = FT_A_BUF;
fdi.tick->tk_aw_bufdata = bufdat;
if ((err = fdisp_wait_answ(&fdi))) {
/* There was a problem with reading. */
goto out;
}
respsize = fdi.tick->tk_aw_bufsize;
if (respsize < 0) { /* Cannot really happen... */
err = EIO;
goto out;
}
buf_setresid(bp, (uint32_t)(buf_resid(bp) - respsize));
bufdat += respsize;
offset += respsize;
/* Did we hit EOF before being done? */
if ((respsize == 0) && (buf_resid(bp) > 0)) {
/*
* Historical note:
* If we don't get enough data, just fill the rest with zeros.
* In NFS context, this would mean a hole in the file.
*/
/* Zero-pad the incomplete buffer. */
bzero(bufdat, buf_resid(bp));
buf_setresid(bp, 0);
break;
}
} /* while (buf_resid(bp) > 0) */
} else {
/* write */
struct fuse_write_in *fwi;
struct fuse_write_out *fwo;
int merr = 0;
off_t diff;
if (buf_map(bp, &bufdat)) {
IOLog("MacFUSE: failed to map buffer in strategy\n");
return EFAULT;
} else {
mapped = TRUE;
}
/* Write begin */
buf_setresid(bp, buf_count(bp));
offset = (off_t)((off_t)buf_blkno(bp) * biosize);
/* XXX: TBD -- Check here for extension (writing past end) */
left = buf_count(bp);
while (left) {
fdi.iosize = sizeof(*fwi);
op = FUSE_WRITE;
fdisp_make_vp(&fdi, op, vp, (vfs_context_t)0);
chunksize = min((size_t)left, data->iosize);
fwi = fdi.indata;
fwi->fh = fufh->fh_id;
fwi->offset = offset;
fwi->size = (typeof(fwi->size))chunksize;
fdi.tick->tk_ms_type = FT_M_BUF;
fdi.tick->tk_ms_bufdata = bufdat;
fdi.tick->tk_ms_bufsize = chunksize;
/* About to write <chunksize> at <offset> */
if ((err = fdisp_wait_answ(&fdi))) {
merr = 1;
break;
}
fwo = fdi.answ;
diff = chunksize - fwo->size;
if (diff < 0) {
err = EINVAL;
break;
}
left -= fwo->size;
bufdat += fwo->size;
offset += fwo->size;
buf_setresid(bp, buf_resid(bp) - fwo->size);
}
if (merr) {
goto out;
}
}
if (fdi.tick) {
fuse_ticket_drop(fdi.tick);
} else {
/* No ticket upon leaving */
}
out:
if (err) {
buf_seterror(bp, err);
}
if (mapped == TRUE) {
buf_unmap(bp);
}
buf_biodone(bp);
return err;
}
__private_extern__
errno_t
fuse_internal_strategy_buf(struct vnop_strategy_args *ap)
{
int32_t bflags;
upl_t bupl;
daddr64_t blkno, lblkno;
int bmap_flags;
buf_t bp = ap->a_bp;
vnode_t vp = buf_vnode(bp);
int vtype = vnode_vtype(vp);
struct fuse_data *data;
if (!vp || vtype == VCHR || vtype == VBLK) {
panic("MacFUSE: buf_strategy: b_vp == NULL || vtype == VCHR | VBLK\n");
}
bflags = buf_flags(bp);
if (bflags & B_READ) {
bmap_flags = VNODE_READ;
} else {
bmap_flags = VNODE_WRITE;
}
bupl = buf_upl(bp);
blkno = buf_blkno(bp);
lblkno = buf_lblkno(bp);
if (!(bflags & B_CLUSTER)) {
if (bupl) {
return cluster_bp(bp);
}
if (blkno == lblkno) {
off_t f_offset;
size_t contig_bytes;
data = fuse_get_mpdata(vnode_mount(vp));
// Still think this is a kludge?
f_offset = lblkno * data->blocksize;
blkno = f_offset / data->blocksize;
buf_setblkno(bp, blkno);
contig_bytes = buf_count(bp);
if (blkno == -1) {
buf_clear(bp);
}
/*
* Our "device" is always /all contiguous/. We don't wanna be
* doing things like:
*
* ...
* else if ((long)contig_bytes < buf_count(bp)) {
* ret = buf_strategy_fragmented(devvp, bp, f_offset,
* contig_bytes));
* return ret;
* }
*/
}
if (blkno == -1) {
buf_biodone(bp);
return 0;
}
}
// Issue the I/O
return fuse_internal_strategy(vp, bp);
}