/* VOP_BWRITE ULFS_NIADDR+2 times */
int
lfs_balloc(struct vnode *vp, off_t startoffset, int iosize, kauth_cred_t cred,
int flags, struct buf **bpp)
{
int offset;
daddr_t daddr, idaddr;
struct buf *ibp, *bp;
struct inode *ip;
struct lfs *fs;
struct indir indirs[ULFS_NIADDR+2], *idp;
daddr_t lbn, lastblock;
int bcount;
int error, frags, i, nsize, osize, num;
ip = VTOI(vp);
fs = ip->i_lfs;
offset = lfs_blkoff(fs, startoffset);
KASSERT(iosize <= lfs_sb_getbsize(fs));
lbn = lfs_lblkno(fs, startoffset);
/* (void)lfs_check(vp, lbn, 0); */
ASSERT_MAYBE_SEGLOCK(fs);
/*
* Three cases: it's a block beyond the end of file, it's a block in
* the file that may or may not have been assigned a disk address or
* we're writing an entire block.
*
* Note, if the daddr is UNWRITTEN, the block already exists in
* the cache (it was read or written earlier). If so, make sure
* we don't count it as a new block or zero out its contents. If
* it did not, make sure we allocate any necessary indirect
* blocks.
*
* If we are writing a block beyond the end of the file, we need to
* check if the old last block was a fragment. If it was, we need
* to rewrite it.
*/
if (bpp)
*bpp = NULL;
/* Check for block beyond end of file and fragment extension needed. */
lastblock = lfs_lblkno(fs, ip->i_size);
if (lastblock < ULFS_NDADDR && lastblock < lbn) {
osize = lfs_blksize(fs, ip, lastblock);
if (osize < lfs_sb_getbsize(fs) && osize > 0) {
if ((error = lfs_fragextend(vp, osize, lfs_sb_getbsize(fs),
lastblock,
(bpp ? &bp : NULL), cred)))
return (error);
ip->i_size = (lastblock + 1) * lfs_sb_getbsize(fs);
lfs_dino_setsize(fs, ip->i_din, ip->i_size);
uvm_vnp_setsize(vp, ip->i_size);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (bpp)
(void) VOP_BWRITE(bp->b_vp, bp);
}
}
/*
* If the block we are writing is a direct block, it's the last
* block in the file, and offset + iosize is less than a full
* block, we can write one or more fragments. There are two cases:
* the block is brand new and we should allocate it the correct
* size or it already exists and contains some fragments and
* may need to extend it.
*/
if (lbn < ULFS_NDADDR && lfs_lblkno(fs, ip->i_size) <= lbn) {
osize = lfs_blksize(fs, ip, lbn);
nsize = lfs_fragroundup(fs, offset + iosize);
if (lfs_lblktosize(fs, lbn) >= ip->i_size) {
/* Brand new block or fragment */
frags = lfs_numfrags(fs, nsize);
if (!ISSPACE(fs, frags, cred))
return ENOSPC;
if (bpp) {
*bpp = bp = getblk(vp, lbn, nsize, 0, 0);
bp->b_blkno = UNWRITTEN;
if (flags & B_CLRBUF)
clrbuf(bp);
}
ip->i_lfs_effnblks += frags;
mutex_enter(&lfs_lock);
lfs_sb_subbfree(fs, frags);
mutex_exit(&lfs_lock);
lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
} else {
if (nsize <= osize) {
/* No need to extend */
if (bpp && (error = bread(vp, lbn, osize,
0, &bp)))
return error;
} else {
/* Extend existing block */
if ((error =
lfs_fragextend(vp, osize, nsize, lbn,
(bpp ? &bp : NULL), cred)))
return error;
}
if (bpp)
*bpp = bp;
}
return 0;
}
error = ulfs_bmaparray(vp, lbn, &daddr, &indirs[0], &num, NULL, NULL);
if (error)
return (error);
KASSERT(daddr <= LFS_MAX_DADDR(fs));
/*
* Do byte accounting all at once, so we can gracefully fail *before*
* we start assigning blocks.
*/
frags = fs->um_seqinc;
bcount = 0;
if (daddr == UNASSIGNED) {
bcount = frags;
}
for (i = 1; i < num; ++i) {
if (!indirs[i].in_exists) {
bcount += frags;
}
}
if (ISSPACE(fs, bcount, cred)) {
mutex_enter(&lfs_lock);
lfs_sb_subbfree(fs, bcount);
mutex_exit(&lfs_lock);
ip->i_lfs_effnblks += bcount;
} else {
return ENOSPC;
}
if (daddr == UNASSIGNED) {
if (num > 0 && lfs_dino_getib(fs, ip->i_din, indirs[0].in_off) == 0) {
lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, UNWRITTEN);
}
/*
* Create new indirect blocks if necessary
*/
if (num > 1) {
idaddr = lfs_dino_getib(fs, ip->i_din, indirs[0].in_off);
for (i = 1; i < num; ++i) {
ibp = getblk(vp, indirs[i].in_lbn,
lfs_sb_getbsize(fs), 0,0);
if (!indirs[i].in_exists) {
clrbuf(ibp);
ibp->b_blkno = UNWRITTEN;
} else if (!(ibp->b_oflags & (BO_DELWRI | BO_DONE))) {
ibp->b_blkno = LFS_FSBTODB(fs, idaddr);
ibp->b_flags |= B_READ;
VOP_STRATEGY(vp, ibp);
biowait(ibp);
}
/*
* This block exists, but the next one may not.
* If that is the case mark it UNWRITTEN to keep
* the accounting straight.
*/
/* XXX ondisk32 */
if (((int32_t *)ibp->b_data)[indirs[i].in_off] == 0)
((int32_t *)ibp->b_data)[indirs[i].in_off] =
UNWRITTEN;
/* XXX ondisk32 */
idaddr = ((int32_t *)ibp->b_data)[indirs[i].in_off];
#ifdef DEBUG
if (vp == fs->lfs_ivnode) {
LFS_ENTER_LOG("balloc", __FILE__,
__LINE__, indirs[i].in_lbn,
ibp->b_flags, curproc->p_pid);
}
#endif
if ((error = VOP_BWRITE(ibp->b_vp, ibp)))
return error;
}
}
}
/*
* Get the existing block from the cache, if requested.
*/
if (bpp)
*bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn), 0, 0);
/*
* Do accounting on blocks that represent pages.
*/
if (!bpp)
lfs_register_block(vp, lbn);
/*
* The block we are writing may be a brand new block
* in which case we need to do accounting.
*
* We can tell a truly new block because ulfs_bmaparray will say
* it is UNASSIGNED. Once we allocate it we will assign it the
* disk address UNWRITTEN.
*/
if (daddr == UNASSIGNED) {
if (bpp) {
if (flags & B_CLRBUF)
clrbuf(bp);
/* Note the new address */
bp->b_blkno = UNWRITTEN;
}
switch (num) {
case 0:
lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
break;
case 1:
lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, UNWRITTEN);
break;
default:
idp = &indirs[num - 1];
if (bread(vp, idp->in_lbn, lfs_sb_getbsize(fs),
B_MODIFY, &ibp))
panic("lfs_balloc: bread bno %lld",
(long long)idp->in_lbn);
/* XXX ondisk32 */
((int32_t *)ibp->b_data)[idp->in_off] = UNWRITTEN;
#ifdef DEBUG
if (vp == fs->lfs_ivnode) {
LFS_ENTER_LOG("balloc", __FILE__,
__LINE__, idp->in_lbn,
ibp->b_flags, curproc->p_pid);
}
#endif
VOP_BWRITE(ibp->b_vp, ibp);
}
} else if (bpp && !(bp->b_oflags & (BO_DONE|BO_DELWRI))) {
/*
* Not a brand new block, also not in the cache;
* read it in from disk.
*/
if (iosize == lfs_sb_getbsize(fs))
/* Optimization: I/O is unnecessary. */
bp->b_blkno = daddr;
else {
/*
* We need to read the block to preserve the
* existing bytes.
*/
bp->b_blkno = daddr;
bp->b_flags |= B_READ;
VOP_STRATEGY(vp, bp);
return (biowait(bp));
}
}
return (0);
}
int
lfs_truncate(struct vnode *ovp, off_t length, int ioflag, kauth_cred_t cred)
{
daddr_t lastblock;
struct inode *oip = VTOI(ovp);
daddr_t bn, lbn, lastiblock[ULFS_NIADDR], indir_lbn[ULFS_NIADDR];
/* XXX ondisk32 */
int32_t newblks[ULFS_NDADDR + ULFS_NIADDR];
struct lfs *fs;
struct buf *bp;
int offset, size, level;
daddr_t count, rcount;
daddr_t blocksreleased = 0, real_released = 0;
int i, nblocks;
int aflags, error, allerror = 0;
off_t osize;
long lastseg;
size_t bc;
int obufsize, odb;
int usepc;
if (ovp->v_type == VCHR || ovp->v_type == VBLK ||
ovp->v_type == VFIFO || ovp->v_type == VSOCK) {
KASSERT(oip->i_size == 0);
return 0;
}
if (length < 0)
return (EINVAL);
/*
* Just return and not update modification times.
*/
if (oip->i_size == length) {
/* still do a uvm_vnp_setsize() as writesize may be larger */
uvm_vnp_setsize(ovp, length);
return (0);
}
fs = oip->i_lfs;
if (ovp->v_type == VLNK &&
(oip->i_size < fs->um_maxsymlinklen ||
(fs->um_maxsymlinklen == 0 &&
oip->i_ffs1_blocks == 0))) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("lfs_truncate: partial truncate of symlink");
#endif
memset((char *)SHORTLINK(oip), 0, (u_int)oip->i_size);
oip->i_size = oip->i_ffs1_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (lfs_update(ovp, NULL, NULL, 0));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (lfs_update(ovp, NULL, NULL, 0));
}
lfs_imtime(fs);
osize = oip->i_size;
usepc = (ovp->v_type == VREG && ovp != fs->lfs_ivnode);
ASSERT_NO_SEGLOCK(fs);
/*
* Lengthen the size of the file. We must ensure that the
* last byte of the file is allocated. Since the smallest
* value of osize is 0, length will be at least 1.
*/
if (osize < length) {
if (length > fs->um_maxfilesize)
return (EFBIG);
aflags = B_CLRBUF;
if (ioflag & IO_SYNC)
aflags |= B_SYNC;
if (usepc) {
if (lfs_lblkno(fs, osize) < ULFS_NDADDR &&
lfs_lblkno(fs, osize) != lfs_lblkno(fs, length) &&
lfs_blkroundup(fs, osize) != osize) {
off_t eob;
eob = lfs_blkroundup(fs, osize);
uvm_vnp_setwritesize(ovp, eob);
error = ulfs_balloc_range(ovp, osize,
eob - osize, cred, aflags);
if (error) {
(void) lfs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
if (ioflag & IO_SYNC) {
mutex_enter(ovp->v_interlock);
VOP_PUTPAGES(ovp,
trunc_page(osize & lfs_sb_getbmask(fs)),
round_page(eob),
PGO_CLEANIT | PGO_SYNCIO);
}
}
uvm_vnp_setwritesize(ovp, length);
error = ulfs_balloc_range(ovp, length - 1, 1, cred,
aflags);
if (error) {
(void) lfs_truncate(ovp, osize,
ioflag & IO_SYNC, cred);
return error;
}
uvm_vnp_setsize(ovp, length);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
KASSERT(ovp->v_size == oip->i_size);
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
return (lfs_update(ovp, NULL, NULL, 0));
} else {
error = lfs_reserve(fs, ovp, NULL,
lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
if (error)
return (error);
error = lfs_balloc(ovp, length - 1, 1, cred,
aflags, &bp);
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (ULFS_NIADDR + 2) << lfs_sb_getbshift(fs)));
if (error)
return (error);
oip->i_ffs1_size = oip->i_size = length;
uvm_vnp_setsize(ovp, length);
(void) VOP_BWRITE(bp->b_vp, bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
return (lfs_update(ovp, NULL, NULL, 0));
}
}
if ((error = lfs_reserve(fs, ovp, NULL,
lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)))) != 0)
return (error);
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundary, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever becomes accessible again because
* of subsequent file growth. Directories however are not
* zero'ed as they should grow back initialized to empty.
*/
offset = lfs_blkoff(fs, length);
lastseg = -1;
bc = 0;
if (ovp != fs->lfs_ivnode)
lfs_seglock(fs, SEGM_PROT);
if (offset == 0) {
oip->i_size = oip->i_ffs1_size = length;
} else if (!usepc) {
lbn = lfs_lblkno(fs, length);
aflags = B_CLRBUF;
if (ioflag & IO_SYNC)
aflags |= B_SYNC;
error = lfs_balloc(ovp, length - 1, 1, cred, aflags, &bp);
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
obufsize = bp->b_bufsize;
odb = lfs_btofsb(fs, bp->b_bcount);
oip->i_size = oip->i_ffs1_size = length;
size = lfs_blksize(fs, oip, lbn);
if (ovp->v_type != VDIR)
memset((char *)bp->b_data + offset, 0,
(u_int)(size - offset));
allocbuf(bp, size, 1);
if ((bp->b_flags & B_LOCKED) != 0 && bp->b_iodone == NULL) {
mutex_enter(&lfs_lock);
locked_queue_bytes -= obufsize - bp->b_bufsize;
mutex_exit(&lfs_lock);
}
if (bp->b_oflags & BO_DELWRI) {
lfs_sb_addavail(fs, odb - lfs_btofsb(fs, size));
/* XXX shouldn't this wake up on lfs_availsleep? */
}
(void) VOP_BWRITE(bp->b_vp, bp);
} else { /* vp->v_type == VREG && length < osize && offset != 0 */
/*
* When truncating a regular file down to a non-block-aligned
* size, we must zero the part of last block which is past
* the new EOF. We must synchronously flush the zeroed pages
* to disk since the new pages will be invalidated as soon
* as we inform the VM system of the new, smaller size.
* We must do this before acquiring the GLOCK, since fetching
* the pages will acquire the GLOCK internally.
* So there is a window where another thread could see a whole
* zeroed page past EOF, but that's life.
*/
daddr_t xlbn;
voff_t eoz;
aflags = ioflag & IO_SYNC ? B_SYNC : 0;
error = ulfs_balloc_range(ovp, length - 1, 1, cred, aflags);
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
xlbn = lfs_lblkno(fs, length);
size = lfs_blksize(fs, oip, xlbn);
eoz = MIN(lfs_lblktosize(fs, xlbn) + size, osize);
ubc_zerorange(&ovp->v_uobj, length, eoz - length,
UBC_UNMAP_FLAG(ovp));
if (round_page(eoz) > round_page(length)) {
mutex_enter(ovp->v_interlock);
error = VOP_PUTPAGES(ovp, round_page(length),
round_page(eoz),
PGO_CLEANIT | PGO_DEACTIVATE |
((ioflag & IO_SYNC) ? PGO_SYNCIO : 0));
if (error) {
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
goto errout;
}
}
}
genfs_node_wrlock(ovp);
oip->i_size = oip->i_ffs1_size = length;
uvm_vnp_setsize(ovp, length);
/*
* Calculate index into inode's block list of
* last direct and indirect blocks (if any)
* which we want to keep. Lastblock is -1 when
* the file is truncated to 0.
*/
/* Avoid sign overflow - XXX assumes that off_t is a quad_t. */
if (length > QUAD_MAX - lfs_sb_getbsize(fs))
lastblock = lfs_lblkno(fs, QUAD_MAX - lfs_sb_getbsize(fs));
else
lastblock = lfs_lblkno(fs, length + lfs_sb_getbsize(fs) - 1) - 1;
lastiblock[SINGLE] = lastblock - ULFS_NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - LFS_NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs);
nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
/*
* Record changed file and block pointers before we start
* freeing blocks. lastiblock values are also normalized to -1
* for calls to lfs_indirtrunc below.
*/
memcpy((void *)newblks, (void *)&oip->i_ffs1_db[0], sizeof newblks);
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
newblks[ULFS_NDADDR+level] = 0;
lastiblock[level] = -1;
}
for (i = ULFS_NDADDR - 1; i > lastblock; i--)
newblks[i] = 0;
oip->i_size = oip->i_ffs1_size = osize;
error = lfs_vtruncbuf(ovp, lastblock + 1, false, 0);
if (error && !allerror)
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -ULFS_NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - LFS_NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - LFS_NINDIR(fs) * LFS_NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ffs1_ib[level];
if (bn != 0) {
error = lfs_indirtrunc(oip, indir_lbn[level],
bn, lastiblock[level],
level, &count, &rcount,
&lastseg, &bc);
if (error)
allerror = error;
real_released += rcount;
blocksreleased += count;
if (lastiblock[level] < 0) {
if (oip->i_ffs1_ib[level] > 0)
real_released += nblocks;
blocksreleased += nblocks;
oip->i_ffs1_ib[level] = 0;
lfs_blkfree(fs, oip, bn, lfs_sb_getbsize(fs),
&lastseg, &bc);
lfs_deregister_block(ovp, bn);
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = ULFS_NDADDR - 1; i > lastblock; i--) {
long bsize, obsize;
bn = oip->i_ffs1_db[i];
if (bn == 0)
continue;
bsize = lfs_blksize(fs, oip, i);
if (oip->i_ffs1_db[i] > 0) {
/* Check for fragment size changes */
obsize = oip->i_lfs_fragsize[i];
real_released += lfs_btofsb(fs, obsize);
oip->i_lfs_fragsize[i] = 0;
} else
obsize = 0;
blocksreleased += lfs_btofsb(fs, bsize);
oip->i_ffs1_db[i] = 0;
lfs_blkfree(fs, oip, bn, obsize, &lastseg, &bc);
lfs_deregister_block(ovp, bn);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_ffs1_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
#if 0
long olddspace;
#endif
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = lfs_blksize(fs, oip, lastblock);
#if 0
olddspace = oip->i_lfs_fragsize[lastblock];
#endif
oip->i_size = oip->i_ffs1_size = length;
newspace = lfs_blksize(fs, oip, lastblock);
if (newspace == 0)
panic("itrunc: newspace");
if (oldspace - newspace > 0) {
blocksreleased += lfs_btofsb(fs, oldspace - newspace);
}
#if 0
if (bn > 0 && olddspace - newspace > 0) {
/* No segment accounting here, just vnode */
real_released += lfs_btofsb(fs, olddspace - newspace);
}
#endif
}
done:
/* Finish segment accounting corrections */
lfs_update_seguse(fs, oip, lastseg, bc);
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if ((newblks[ULFS_NDADDR + level] == 0) !=
((oip->i_ffs1_ib[level]) == 0)) {
panic("lfs itrunc1");
}
for (i = 0; i < ULFS_NDADDR; i++)
if ((newblks[i] == 0) != (oip->i_ffs1_db[i] == 0)) {
panic("lfs itrunc2");
}
if (length == 0 &&
(!LIST_EMPTY(&ovp->v_cleanblkhd) || !LIST_EMPTY(&ovp->v_dirtyblkhd)))
panic("lfs itrunc3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = oip->i_ffs1_size = length;
oip->i_lfs_effnblks -= blocksreleased;
oip->i_ffs1_blocks -= real_released;
mutex_enter(&lfs_lock);
lfs_sb_addbfree(fs, blocksreleased);
mutex_exit(&lfs_lock);
#ifdef DIAGNOSTIC
if (oip->i_size == 0 &&
(oip->i_ffs1_blocks != 0 || oip->i_lfs_effnblks != 0)) {
printf("lfs_truncate: truncate to 0 but %d blks/%jd effblks\n",
oip->i_ffs1_blocks, (intmax_t)oip->i_lfs_effnblks);
panic("lfs_truncate: persistent blocks");
}
#endif
/*
* If we truncated to zero, take us off the paging queue.
*/
mutex_enter(&lfs_lock);
if (oip->i_size == 0 && oip->i_flags & IN_PAGING) {
oip->i_flags &= ~IN_PAGING;
TAILQ_REMOVE(&fs->lfs_pchainhd, oip, i_lfs_pchain);
}
mutex_exit(&lfs_lock);
oip->i_flag |= IN_CHANGE;
#if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
(void) lfs_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
lfs_reserve(fs, ovp, NULL,
-lfs_btofsb(fs, (2 * ULFS_NIADDR + 3) << lfs_sb_getbshift(fs)));
genfs_node_unlock(ovp);
errout:
oip->i_lfs_hiblk = lfs_lblkno(fs, oip->i_size + lfs_sb_getbsize(fs) - 1) - 1;
if (ovp != fs->lfs_ivnode)
lfs_segunlock(fs);
return (allerror ? allerror : error);
}