/*
* Release blocks associated with the inode ip and stored in the indirect
* block bn. Blocks are free'd in LIFO order up to (but not including)
* lastbn. If level is greater than SINGLE, the block is an indirect block
* and recursive calls to indirtrunc must be used to cleanse other indirect
* blocks.
*
* NB: triple indirect blocks are untested.
*/
static int
lfs_indirtrunc(struct inode *ip, daddr_t lbn, daddr_t dbn,
daddr_t lastbn, int level, daddr_t *countp,
daddr_t *rcountp, long *lastsegp, size_t *bcp)
{
int i;
struct buf *bp;
struct lfs *fs = ip->i_lfs;
int32_t *bap; /* XXX ondisk32 */
struct vnode *vp;
daddr_t nb, nlbn, last;
int32_t *copy = NULL; /* XXX ondisk32 */
daddr_t blkcount, rblkcount, factor;
int nblocks;
daddr_t blocksreleased = 0, real_released = 0;
int error = 0, allerror = 0;
ASSERT_SEGLOCK(fs);
/*
* Calculate index in current block of last
* block to be kept. -1 indicates the entire
* block so we need not calculate the index.
*/
factor = 1;
for (i = SINGLE; i < level; i++)
factor *= LFS_NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = lfs_btofsb(fs, lfs_sb_getbsize(fs));
/*
* Get buffer of block pointers, zero those entries corresponding
* to blocks to be free'd, and update on disk copy first. Since
* double(triple) indirect before single(double) indirect, calls
* to bmap on these blocks will fail. However, we already have
* the on disk address, so we have to set the b_blkno field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lbn, lfs_sb_getbsize(fs), 0, 0);
if (bp->b_oflags & (BO_DONE | BO_DELWRI)) {
/* Braces must be here in case trace evaluates to nothing. */
trace(TR_BREADHIT, pack(vp, lfs_sb_getbsize(fs)), lbn);
} else {
trace(TR_BREADMISS, pack(vp, lfs_sb_getbsize(fs)), lbn);
curlwp->l_ru.ru_inblock++; /* pay for read */
bp->b_flags |= B_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("lfs_indirtrunc: bad buffer size");
bp->b_blkno = LFS_FSBTODB(fs, dbn);
VOP_STRATEGY(vp, bp);
error = biowait(bp);
}
if (error) {
brelse(bp, 0);
*countp = *rcountp = 0;
return (error);
}
bap = (int32_t *)bp->b_data; /* XXX ondisk32 */
if (lastbn >= 0) {
copy = lfs_malloc(fs, lfs_sb_getbsize(fs), LFS_NB_IBLOCK);
memcpy((void *)copy, (void *)bap, lfs_sb_getbsize(fs));
memset((void *)&bap[last + 1], 0,
/* XXX ondisk32 */
(u_int)(LFS_NINDIR(fs) - (last + 1)) * sizeof (int32_t));
error = VOP_BWRITE(bp->b_vp, bp);
if (error)
allerror = error;
bap = copy;
}
/*
* Recursively free totally unused blocks.
*/
for (i = LFS_NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = bap[i];
if (nb == 0)
continue;
if (level > SINGLE) {
error = lfs_indirtrunc(ip, nlbn, nb,
(daddr_t)-1, level - 1,
&blkcount, &rblkcount,
lastsegp, bcp);
if (error)
allerror = error;
blocksreleased += blkcount;
real_released += rblkcount;
}
lfs_blkfree(fs, ip, nb, lfs_sb_getbsize(fs), lastsegp, bcp);
if (bap[i] > 0)
real_released += nblocks;
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = bap[i];
if (nb != 0) {
error = lfs_indirtrunc(ip, nlbn, nb,
last, level - 1, &blkcount,
&rblkcount, lastsegp, bcp);
if (error)
allerror = error;
real_released += rblkcount;
blocksreleased += blkcount;
}
}
if (copy != NULL) {
lfs_free(fs, copy, LFS_NB_IBLOCK);
} else {
mutex_enter(&bufcache_lock);
if (bp->b_oflags & BO_DELWRI) {
LFS_UNLOCK_BUF(bp);
lfs_sb_addavail(fs, lfs_btofsb(fs, bp->b_bcount));
wakeup(&fs->lfs_availsleep);
}
brelsel(bp, BC_INVAL);
mutex_exit(&bufcache_lock);
}
*countp = blocksreleased;
*rcountp = real_released;
return (allerror);
}
/*
* Actually mark the segment clean.
* Must be called with the segment lock held.
*/
int
lfs_do_segclean(struct lfs *fs, unsigned long segnum)
{
extern int lfs_dostats;
struct buf *bp;
CLEANERINFO *cip;
SEGUSE *sup;
if (lfs_dtosn(fs, lfs_sb_getcurseg(fs)) == segnum) {
return (EBUSY);
}
LFS_SEGENTRY(sup, fs, segnum, bp);
if (sup->su_nbytes) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" %d live bytes\n", segnum, sup->su_nbytes));
brelse(bp, 0);
return (EBUSY);
}
if (sup->su_flags & SEGUSE_ACTIVE) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" segment is active\n", segnum));
brelse(bp, 0);
return (EBUSY);
}
if (!(sup->su_flags & SEGUSE_DIRTY)) {
DLOG((DLOG_CLEAN, "lfs_segclean: not cleaning segment %lu:"
" segment is already clean\n", segnum));
brelse(bp, 0);
return (EALREADY);
}
lfs_sb_addavail(fs, lfs_segtod(fs, 1));
if (sup->su_flags & SEGUSE_SUPERBLOCK)
lfs_sb_subavail(fs, lfs_btofsb(fs, LFS_SBPAD));
if (lfs_sb_getversion(fs) > 1 && segnum == 0 &&
lfs_sb_gets0addr(fs) < lfs_btofsb(fs, LFS_LABELPAD))
lfs_sb_subavail(fs, lfs_btofsb(fs, LFS_LABELPAD) - lfs_sb_gets0addr(fs));
mutex_enter(&lfs_lock);
lfs_sb_addbfree(fs, sup->su_nsums * lfs_btofsb(fs, lfs_sb_getsumsize(fs)) +
lfs_btofsb(fs, sup->su_ninos * lfs_sb_getibsize(fs)));
lfs_sb_subdmeta(fs, sup->su_nsums * lfs_btofsb(fs, lfs_sb_getsumsize(fs)) +
lfs_btofsb(fs, sup->su_ninos * lfs_sb_getibsize(fs)));
if (lfs_sb_getdmeta(fs) < 0)
lfs_sb_setdmeta(fs, 0);
mutex_exit(&lfs_lock);
sup->su_flags &= ~SEGUSE_DIRTY;
LFS_WRITESEGENTRY(sup, fs, segnum, bp);
LFS_CLEANERINFO(cip, fs, bp);
lfs_ci_shiftdirtytoclean(fs, cip, 1);
lfs_sb_setnclean(fs, lfs_ci_getclean(fs, cip));
mutex_enter(&lfs_lock);
lfs_ci_setbfree(fs, cip, lfs_sb_getbfree(fs));
lfs_ci_setavail(fs, cip, lfs_sb_getavail(fs)
- fs->lfs_ravail - fs->lfs_favail);
wakeup(&fs->lfs_availsleep);
mutex_exit(&lfs_lock);
(void) LFS_BWRITE_LOG(bp);
if (lfs_dostats)
++lfs_stats.segs_reclaimed;
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);
}
