/*
* The last lbn argument is the current block on which I/O is being
* performed. Check to see that it doesn't fall in the middle of
* the current block (if last_bp == NULL).
*/
void
cluster_wbuild(struct vnode *vp, struct buf *last_bp, long size,
daddr64_t start_lbn, int len, daddr64_t lbn)
{
struct buf *bp;
#ifdef DIAGNOSTIC
if (size != vp->v_mount->mnt_stat.f_iosize)
panic("cluster_wbuild: size %ld != filesize %ld",
size, vp->v_mount->mnt_stat.f_iosize);
#endif
redo:
while ((!incore(vp, start_lbn) || start_lbn == lbn) && len) {
++start_lbn;
--len;
}
/* Get more memory for current buffer */
if (len <= 1) {
if (last_bp) {
bawrite(last_bp);
} else if (len) {
bp = getblk(vp, start_lbn, size, 0, 0);
/*
* The buffer could have already been flushed out of
* the cache. If that has happened, we'll get a new
* buffer here with random data, just drop it.
*/
if ((bp->b_flags & B_DELWRI) == 0)
brelse(bp);
else
bawrite(bp);
}
return;
}
bp = getblk(vp, start_lbn, size, 0, 0);
if (!(bp->b_flags & B_DELWRI)) {
++start_lbn;
--len;
brelse(bp);
goto redo;
}
++start_lbn;
--len;
bawrite(bp);
goto redo;
}
/*
* Delayed write.
*
* The buffer is marked dirty, but is not queued for I/O.
* This routine should be used when the buffer is expected
* to be modified again soon, typically a small write that
* partially fills a buffer.
*
* NB: magnetic tapes cannot be delayed; they must be
* written in the order that the writes are requested.
*
* Described in Leffler, et al. (pp. 208-213).
*/
void
bdwrite(struct buf *bp)
{
int s;
/*
* If the block hasn't been seen before:
* (1) Mark it as having been seen,
* (2) Charge for the write.
* (3) Make sure it's on its vnode's correct block list,
* (4) If a buffer is rewritten, move it to end of dirty list
*/
if (!ISSET(bp->b_flags, B_DELWRI)) {
SET(bp->b_flags, B_DELWRI);
s = splbio();
reassignbuf(bp);
splx(s);
curproc->p_stats->p_ru.ru_oublock++; /* XXX */
}
/* If this is a tape block, write the block now. */
if (major(bp->b_dev) < nblkdev &&
bdevsw[major(bp->b_dev)].d_type == D_TAPE) {
bawrite(bp);
return;
}
/* Otherwise, the "write" is done, so mark and release the buffer. */
CLR(bp->b_flags, B_NEEDCOMMIT);
SET(bp->b_flags, B_DONE);
brelse(bp);
}
int
ext2fs_cgupdate(struct ufsmount *mp, int waitfor)
{
printf("In file: %s, fun: %s,lineno: %d\n",__FILE__, __func__, __LINE__);
struct m_ext2fs *fs = mp->um_e2fs;
struct buf *bp;
int i, error = 0, allerror = 0;
allerror = ext2fs_sbupdate(mp, waitfor);
for (i = 0; i < fs->e2fs_ngdb; i++) {
bp = getblk(mp->um_devvp, EXT2_FSBTODB(fs,
fs->e2fs.e2fs_first_dblock +
1 /* superblock */ + i), fs->e2fs_bsize, 0, 0);
e2fs_cgsave(&fs->e2fs_gd[
i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
(struct ext2_gd *)bp->b_data, fs->e2fs_bsize);
if (waitfor == MNT_WAIT)
error = bwrite(bp);
else
bawrite(bp);
}
if (!allerror && error)
allerror = error;
return (allerror);
}
/*
* If we have an FSInfo block, update it.
*/
static int
msdosfs_fsiflush(struct msdosfsmount *pmp, int waitfor)
{
struct fsinfo *fp;
struct buf *bp;
int error;
MSDOSFS_LOCK_MP(pmp);
if (pmp->pm_fsinfo == 0 || (pmp->pm_flags & MSDOSFS_FSIMOD) == 0) {
error = 0;
goto unlock;
}
error = bread(pmp->pm_devvp, pmp->pm_fsinfo, pmp->pm_BytesPerSec,
NOCRED, &bp);
if (error != 0) {
brelse(bp);
goto unlock;
}
fp = (struct fsinfo *)bp->b_data;
putulong(fp->fsinfree, pmp->pm_freeclustercount);
putulong(fp->fsinxtfree, pmp->pm_nxtfree);
pmp->pm_flags &= ~MSDOSFS_FSIMOD;
if (waitfor == MNT_WAIT)
error = bwrite(bp);
else
bawrite(bp);
unlock:
MSDOSFS_UNLOCK_MP(pmp);
return (error);
}
int
ext2_cgupdate(struct ext2mount *mp, int waitfor)
{
struct m_ext2fs *fs = mp->um_e2fs;
struct buf *bp;
int i, error = 0, allerror = 0;
allerror = ext2_sbupdate(mp, waitfor);
for (i = 0; i < fs->e2fs_gdbcount; i++) {
bp = getblk(mp->um_devvp, fsbtodb(fs,
fs->e2fs->e2fs_first_dblock +
1 /* superblock */ + i), fs->e2fs_bsize, 0, 0, 0);
e2fs_cgsave(&fs->e2fs_gd[
i * fs->e2fs_bsize / sizeof(struct ext2_gd)],
(struct ext2_gd *)bp->b_data, fs->e2fs_bsize);
if (waitfor == MNT_WAIT)
error = bwrite(bp);
else
bawrite(bp);
}
if (!allerror && error)
allerror = error;
return (allerror);
}
blkno_t blk_alloc(uint16_t devno)
{
fsptr dev;
blkno_t newno;
blkno_t *buf;
uint8_t *mbuf;
int j;
if(baddev(dev = getdev(devno)))
goto corrupt2;
if(dev->s_nfree <= 0 || dev->s_nfree > FILESYS_TABSIZE)
goto corrupt;
newno = dev->s_free[--dev->s_nfree];
if(!newno)
{
if(dev->s_tfree != 0)
goto corrupt;
udata.u_error = ENOSPC;
++dev->s_nfree;
return(0);
}
/* See if we must refill the s_free array */
if(!dev->s_nfree)
{
buf =(blkno_t *)bread(devno, newno, 0);
dev->s_nfree = buf[0];
for(j=0; j < FILESYS_TABSIZE; j++)
{
dev->s_free[j] = buf[j+1];
}
brelse((char *)buf);
}
validblk(devno, newno);
if(!dev->s_tfree)
goto corrupt;
--dev->s_tfree;
/* Zero out the new block */
mbuf = bread(devno, newno, 2);
memset(mbuf, 0, 512);
bawrite(mbuf);
return newno;
corrupt:
kputs("blk_alloc: corrupt\n");
dev->s_mounted = 1;
corrupt2:
udata.u_error = ENOSPC;
return 0;
}
/*
* Buffer cleaning daemon.
*/
void
buf_daemon(struct proc *p)
{
int s;
struct buf *bp;
struct timeval starttime, timediff;
cleanerproc = curproc;
for (;;) {
if (numdirtypages < hidirtypages) {
tsleep(&bd_req, PRIBIO - 7, "cleaner", 0);
}
starttime = time;
s = splbio();
while ((bp = TAILQ_FIRST(&bufqueues[BQ_DIRTY]))) {
bremfree(bp);
SET(bp->b_flags, B_BUSY);
splx(s);
if (ISSET(bp->b_flags, B_INVAL)) {
brelse(bp);
s = splbio();
continue;
}
#ifdef DIAGNOSTIC
if (!ISSET(bp->b_flags, B_DELWRI))
panic("Clean buffer on BQ_DIRTY");
#endif
if (LIST_FIRST(&bp->b_dep) != NULL &&
!ISSET(bp->b_flags, B_DEFERRED) &&
buf_countdeps(bp, 0, 1)) {
SET(bp->b_flags, B_DEFERRED);
s = splbio();
numfreepages += btoc(bp->b_bufsize);
numdirtypages += btoc(bp->b_bufsize);
binstailfree(bp, &bufqueues[BQ_DIRTY]);
CLR(bp->b_flags, B_BUSY);
continue;
}
bawrite(bp);
if (numdirtypages < lodirtypages)
break;
/* Never allow processing to run for more than 1 sec */
timersub(&time, &starttime, &timediff);
if (timediff.tv_sec)
break;
s = splbio();
}
}
}
/*
* Synch buffers associated with a block device
*/
int
spec_fsync(void *v)
{
struct vop_fsync_args *ap = v;
struct vnode *vp = ap->a_vp;
struct buf *bp;
struct buf *nbp;
int s;
if (vp->v_type == VCHR)
return (0);
#ifdef WAPBL
if (vp->v_type == VBLK &&
vp->v_specmountpoint != NULL &&
vp->v_specmountpoint->mnt_wapbl != NULL) {
int error = ffs_wapbl_fsync_vfs(vp, ap->a_waitfor);
if (error)
return (error);
}
#endif
/*
* Flush all dirty buffers associated with a block device.
*/
loop:
s = splbio();
for (bp = LIST_FIRST(&vp->v_dirtyblkhd);
bp != LIST_END(&vp->v_dirtyblkhd); bp = nbp) {
nbp = LIST_NEXT(bp, b_vnbufs);
if ((bp->b_flags & B_BUSY))
continue;
if ((bp->b_flags & B_DELWRI) == 0)
panic("spec_fsync: not dirty");
bremfree(bp);
buf_acquire(bp);
splx(s);
bawrite(bp);
goto loop;
}
if (ap->a_waitfor == MNT_WAIT) {
vwaitforio (vp, 0, "spec_fsync", 0);
#ifdef DIAGNOSTIC
if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
splx(s);
vprint("spec_fsync: dirty", vp);
goto loop;
}
#endif
}
splx(s);
return (0);
}
/*
* Write a superblock and associated information back to disk.
*/
int
ext2fs_sbupdate(struct ufsmount *mp, int waitfor)
{
struct m_ext2fs *fs = mp->um_e2fs;
struct buf *bp;
int error = 0;
bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0);
e2fs_sbsave(&fs->e2fs, (struct ext2fs *) bp->b_data);
if (waitfor == MNT_WAIT)
error = bwrite(bp);
else
bawrite(bp);
return (error);
}
/*
* Write a superblock and associated information back to disk.
*/
int
ext2fs_sbupdate(struct ufsmount *mp, int waitfor)
{
printf("In file: %s, fun: %s,lineno: %d\n",__FILE__, __func__, __LINE__);
struct m_ext2fs *fs = mp->um_e2fs;
struct buf *bp;
int error = 0;
bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0);
e2fs_sbsave(&fs->e2fs, (struct ext2fs*)bp->b_data);
if (waitfor == MNT_WAIT)
error = bwrite(bp);
else
bawrite(bp);
return (error);
}
int
deupdat(struct denode *dep, int waitfor)
{
struct direntry dir;
struct timespec ts;
struct buf *bp;
struct direntry *dirp;
int error;
if (DETOV(dep)->v_mount->mnt_flag & MNT_RDONLY) {
dep->de_flag &= ~(DE_UPDATE | DE_CREATE | DE_ACCESS |
DE_MODIFIED);
return (0);
}
getnanotime(&ts);
DETIMES(dep, &ts, &ts, &ts);
if ((dep->de_flag & DE_MODIFIED) == 0 && waitfor == 0)
return (0);
dep->de_flag &= ~DE_MODIFIED;
if (DETOV(dep)->v_vflag & VV_ROOT)
return (EINVAL);
if (dep->de_refcnt <= 0)
return (0);
error = readde(dep, &bp, &dirp);
if (error)
return (error);
DE_EXTERNALIZE(&dir, dep);
if (bcmp(dirp, &dir, sizeof(dir)) == 0) {
if (waitfor == 0 || (bp->b_flags & B_DELWRI) == 0) {
brelse(bp);
return (0);
}
} else
*dirp = dir;
if ((DETOV(dep)->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0)
bp->b_flags |= B_CLUSTEROK;
if (waitfor)
error = bwrite(bp);
else if (vm_page_count_severe() || buf_dirty_count_severe())
bawrite(bp);
else
bdwrite(bp);
return (error);
}
int
ext2fs_cgupdate(struct ufsmount *mp, int waitfor)
{
struct m_ext2fs *fs = mp->um_e2fs;
struct buf *bp;
int i, error = 0, allerror = 0;
allerror = ext2fs_sbupdate(mp, waitfor);
for (i = 0; i < fs->e2fs_ngdb; i++) {
bp = getblk(mp->um_devvp, fsbtodb(fs, ((fs->e2fs_bsize>1024)?0:1)+i+1),
fs->e2fs_bsize, 0, 0);
e2fs_cgsave(&fs->e2fs_gd[i* fs->e2fs_bsize / sizeof(struct ext2_gd)], (struct ext2_gd*)bp->b_data, fs->e2fs_bsize);
if (waitfor == MNT_WAIT)
error = bwrite(bp);
else
bawrite(bp);
}
if (!allerror && error)
allerror = error;
return (allerror);
}
/*
* Write a superblock and associated information back to disk.
*/
static int
ext2_sbupdate(struct ext2mount *mp, int waitfor)
{
struct m_ext2fs *fs = mp->um_e2fs;
struct ext2fs *es = fs->e2fs;
struct buf *bp;
int error = 0;
bp = getblk(mp->um_devvp, SBLOCK, SBSIZE, 0, 0, 0);
bcopy((caddr_t)es, bp->b_data, (u_int)sizeof(struct ext2fs));
if (waitfor == MNT_WAIT)
error = bwrite(bp);
else
bawrite(bp);
/*
* The buffers for group descriptors, inode bitmaps and block bitmaps
* are not busy at this point and are (hopefully) written by the
* usual sync mechanism. No need to write them here.
*/
return (error);
}
void blk_free(uint16_t devno, blkno_t blk)
{
fsptr dev;
uint8_t *buf;
if(!blk)
return;
if(baddev(dev = getdev(devno)))
return;
validblk(devno, blk);
if(dev->s_nfree == FILESYS_TABSIZE) {
buf = bread(devno, blk, 1);
memcpy(buf, (char *)&(dev->s_nfree), 51*sizeof(int));
bawrite(buf);
dev->s_nfree = 0;
}
++dev->s_tfree;
dev->s_free[(dev->s_nfree)++] = blk;
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp = vp;
ufs_daddr_t lastblock;
struct inode *oip;
ufs_daddr_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
ufs_daddr_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, blocksreleased = 0;
int i;
int aflags, error, allerror;
off_t osize;
oip = VTOI(ovp);
fs = oip->i_fs;
if (length < 0)
return (EINVAL);
if (length > fs->fs_maxfilesize)
return (EFBIG);
if (ovp->v_type == VLNK &&
(oip->i_size < ovp->v_mount->mnt_maxsymlinklen || oip->i_din.di_blocks == 0)) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif /* DIAGNOSTIC */
bzero((char *)&oip->i_shortlink, (uint)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 0));
}
if (fs->fs_ronly)
panic("ffs_truncate: read-only filesystem");
#ifdef QUOTA
error = ufs_getinoquota(oip);
if (error)
return (error);
#endif
ovp->v_lasta = ovp->v_clen = ovp->v_cstart = ovp->v_lastw = 0;
if (DOINGSOFTDEP(ovp)) {
if (length > 0 || softdep_slowdown(ovp)) {
/*
* If a file is only partially truncated, then
* we have to clean up the data structures
* describing the allocation past the truncation
* point. Finding and deallocating those structures
* is a lot of work. Since partial truncation occurs
* rarely, we solve the problem by syncing the file
* so that it will have no data structures left.
*/
if ((error = VOP_FSYNC(ovp, MNT_WAIT, 0)) != 0)
return (error);
} else {
#ifdef QUOTA
(void) ufs_chkdq(oip, -oip->i_blocks, NOCRED, 0);
#endif
softdep_setup_freeblocks(oip, length);
vinvalbuf(ovp, 0, 0, 0);
nvnode_pager_setsize(ovp, 0, fs->fs_bsize, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 0));
}
}
osize = oip->i_size;
/*
* 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.
*
* nvextendbuf() only breads the old buffer. The blocksize
* of the new buffer must be specified so it knows how large
* to make the VM object.
*/
if (osize < length) {
nvextendbuf(vp, osize, length,
blkoffsize(fs, oip, osize), /* oblksize */
blkoffresize(fs, length), /* nblksize */
blkoff(fs, osize),
blkoff(fs, length),
0);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
/* BALLOC will reallocate the fragment at the old EOF */
error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
if (error)
return (error);
oip->i_size = length;
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ffs_update(ovp, 1));
}
/*
* Shorten the size of the file.
*
* NOTE: The block size specified in nvtruncbuf() is the blocksize
* of the buffer containing length prior to any reallocation
* of the block.
*/
allerror = nvtruncbuf(ovp, length, blkoffsize(fs, oip, length),
blkoff(fs, length), 0);
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_size = length;
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = VOP_BALLOC(ovp, length - 1, 1, cred, aflags, &bp);
if (error)
return (error);
/*
* When we are doing soft updates and the UFS_BALLOC
* above fills in a direct block hole with a full sized
* block that will be truncated down to a fragment below,
* we must flush out the block dependency with an FSYNC
* so that we do not get a soft updates inconsistency
* when we create the fragment below.
*
* nvtruncbuf() may have re-dirtied the underlying block
* as part of its truncation zeroing code. To avoid a
* 'locking against myself' panic in the second fsync we
* can simply undirty the bp since the redirtying was
* related to areas of the buffer that we are going to
* throw away anyway, and we will b*write() the remainder
* anyway down below.
*/
if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize) {
bundirty(bp);
error = VOP_FSYNC(ovp, MNT_WAIT, 0);
if (error) {
bdwrite(bp);
return (error);
}
}
oip->i_size = length;
size = blksize(fs, oip, lbn);
#if 0
/* remove - nvtruncbuf deals with this */
if (ovp->v_type != VDIR)
bzero((char *)bp->b_data + offset,
(uint)(size - offset));
#endif
/* Kirk's code has reallocbuf(bp, size, 1) here */
allocbuf(bp, size);
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
}
/*
* 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.
*/
lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ffs_indirtrunc below.
*/
bcopy((caddr_t)&oip->i_db[0], (caddr_t)oldblks, sizeof oldblks);
for (level = TRIPLE; level >= SINGLE; level--)
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
lastiblock[level] = -1;
}
for (i = NDADDR - 1; i > lastblock; i--)
oip->i_db[i] = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
error = ffs_update(ovp, 1);
if (error && allerror == 0)
allerror = error;
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
bcopy((caddr_t)&oip->i_db[0], (caddr_t)newblks, sizeof newblks);
bcopy((caddr_t)oldblks, (caddr_t)&oip->i_db[0], sizeof oldblks);
oip->i_size = osize;
if (error && allerror == 0)
allerror = error;
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ib[level];
if (bn != 0) {
error = ffs_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
ffs_blkfree(oip, bn, fs->fs_bsize);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = oip->i_db[i];
if (bn == 0)
continue;
oip->i_db[i] = 0;
bsize = blksize(fs, oip, i);
ffs_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("ffs_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ffs_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_ib[level])
panic("ffs_truncate1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("ffs_truncate2");
if (length == 0 && !RB_EMPTY(&ovp->v_rbdirty_tree))
panic("ffs_truncate3");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
oip->i_size = length;
oip->i_blocks -= blocksreleased;
if (oip->i_blocks < 0) /* sanity */
oip->i_blocks = 0;
oip->i_flag |= IN_CHANGE;
#ifdef QUOTA
(void) ufs_chkdq(oip, -blocksreleased, NOCRED, 0);
#endif
return (allerror);
}
/*
* Vnode op for write
*/
int
spec_write(void *v)
{
struct vop_write_args *ap = v;
struct vnode *vp = ap->a_vp;
struct uio *uio = ap->a_uio;
struct proc *p = uio->uio_procp;
struct buf *bp;
daddr_t bn, bscale;
int bsize;
struct partinfo dpart;
size_t n;
int on, majordev;
int (*ioctl)(dev_t, u_long, caddr_t, int, struct proc *);
int error = 0;
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("spec_write mode");
if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc)
panic("spec_write proc");
#endif
switch (vp->v_type) {
case VCHR:
VOP_UNLOCK(vp, 0, p);
error = (*cdevsw[major(vp->v_rdev)].d_write)
(vp->v_rdev, uio, ap->a_ioflag);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
return (error);
case VBLK:
if (uio->uio_resid == 0)
return (0);
if (uio->uio_offset < 0)
return (EINVAL);
bsize = BLKDEV_IOSIZE;
if ((majordev = major(vp->v_rdev)) < nblkdev &&
(ioctl = bdevsw[majordev].d_ioctl) != NULL &&
(*ioctl)(vp->v_rdev, DIOCGPART, (caddr_t)&dpart, FREAD, p) == 0) {
u_int32_t frag =
DISKLABELV1_FFS_FRAG(dpart.part->p_fragblock);
u_int32_t fsize =
DISKLABELV1_FFS_FSIZE(dpart.part->p_fragblock);
if (dpart.part->p_fstype == FS_BSDFFS && frag != 0 &&
fsize != 0)
bsize = frag * fsize;
}
bscale = btodb(bsize);
do {
bn = btodb(uio->uio_offset) & ~(bscale - 1);
on = uio->uio_offset % bsize;
n = ulmin((bsize - on), uio->uio_resid);
error = bread(vp, bn, bsize, &bp);
n = ulmin(n, bsize - bp->b_resid);
if (error) {
brelse(bp);
return (error);
}
error = uiomove((char *)bp->b_data + on, n, uio);
if (n + on == bsize)
bawrite(bp);
else
bdwrite(bp);
} while (error == 0 && uio->uio_resid > 0 && n != 0);
return (error);
default:
panic("spec_write type");
}
/* NOTREACHED */
}
/*
* Vnode op for writing.
*/
int
ffs_write(void *v)
{
struct vop_write_args *ap = v;
struct vnode *vp;
struct uio *uio;
struct inode *ip;
struct fs *fs;
struct buf *bp;
struct proc *p;
daddr64_t lbn;
off_t osize;
int blkoffset, error, extended, flags, ioflag, resid, size, xfersize;
extended = 0;
ioflag = ap->a_ioflag;
uio = ap->a_uio;
vp = ap->a_vp;
ip = VTOI(vp);
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("ffs_write: mode");
#endif
/*
* If writing 0 bytes, succeed and do not change
* update time or file offset (standards compliance)
*/
if (uio->uio_resid == 0)
return (0);
switch (vp->v_type) {
case VREG:
if (ioflag & IO_APPEND)
uio->uio_offset = DIP(ip, size);
if ((DIP(ip, flags) & APPEND) && uio->uio_offset != DIP(ip, size))
return (EPERM);
/* FALLTHROUGH */
case VLNK:
break;
case VDIR:
if ((ioflag & IO_SYNC) == 0)
panic("ffs_write: nonsync dir write");
break;
default:
panic("ffs_write: type");
}
fs = ip->i_fs;
if (uio->uio_offset < 0 ||
(u_int64_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
return (EFBIG);
/*
* Maybe this should be above the vnode op call, but so long as
* file servers have no limits, I don't think it matters.
*/
p = uio->uio_procp;
if (vp->v_type == VREG && p && !(ioflag & IO_NOLIMIT) &&
uio->uio_offset + uio->uio_resid >
p->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
psignal(p, SIGXFSZ);
return (EFBIG);
}
resid = uio->uio_resid;
osize = DIP(ip, size);
flags = ioflag & IO_SYNC ? B_SYNC : 0;
for (error = 0; uio->uio_resid > 0;) {
lbn = lblkno(fs, uio->uio_offset);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->fs_bsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (fs->fs_bsize > xfersize)
flags |= B_CLRBUF;
else
flags &= ~B_CLRBUF;
if ((error = UFS_BUF_ALLOC(ip, uio->uio_offset, xfersize,
ap->a_cred, flags, &bp)) != 0)
break;
if (uio->uio_offset + xfersize > DIP(ip, size)) {
DIP_ASSIGN(ip, size, uio->uio_offset + xfersize);
uvm_vnp_setsize(vp, DIP(ip, size));
extended = 1;
}
(void)uvm_vnp_uncache(vp);
size = blksize(fs, ip, lbn) - bp->b_resid;
if (size < xfersize)
xfersize = size;
error =
uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
if (error != 0)
bzero((char *)bp->b_data + blkoffset, xfersize);
if (ioflag & IO_SYNC)
(void)bwrite(bp);
else if (xfersize + blkoffset == fs->fs_bsize) {
if (doclusterwrite)
cluster_write(bp, &ip->i_ci, DIP(ip, size));
else
bawrite(bp);
} else
bdwrite(bp);
if (error || xfersize == 0)
break;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* If we successfully wrote any data, and we are not the superuser
* we clear the setuid and setgid bits as a precaution against
* tampering.
*/
if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
DIP_ASSIGN(ip, mode, DIP(ip, mode) & ~(ISUID | ISGID));
if (resid > uio->uio_resid)
VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
if (error) {
if (ioflag & IO_UNIT) {
(void)UFS_TRUNCATE(ip, osize,
ioflag & IO_SYNC, ap->a_cred);
uio->uio_offset -= resid - uio->uio_resid;
uio->uio_resid = resid;
}
} else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
error = UFS_UPDATE(ip, MNT_WAIT);
}
return (error);
}
static int
tmpfs_write (struct vop_write_args *ap)
{
struct buf *bp;
struct vnode *vp = ap->a_vp;
struct uio *uio = ap->a_uio;
struct thread *td = uio->uio_td;
struct tmpfs_node *node;
boolean_t extended;
off_t oldsize;
int error;
off_t base_offset;
size_t offset;
size_t len;
struct rlimit limit;
int trivial = 0;
int kflags = 0;
error = 0;
if (uio->uio_resid == 0) {
return error;
}
node = VP_TO_TMPFS_NODE(vp);
if (vp->v_type != VREG)
return (EINVAL);
lwkt_gettoken(&vp->v_mount->mnt_token);
oldsize = node->tn_size;
if (ap->a_ioflag & IO_APPEND)
uio->uio_offset = node->tn_size;
/*
* Check for illegal write offsets.
*/
if (uio->uio_offset + uio->uio_resid >
VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
lwkt_reltoken(&vp->v_mount->mnt_token);
return (EFBIG);
}
if (vp->v_type == VREG && td != NULL) {
error = kern_getrlimit(RLIMIT_FSIZE, &limit);
if (error != 0) {
lwkt_reltoken(&vp->v_mount->mnt_token);
return error;
}
if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
ksignal(td->td_proc, SIGXFSZ);
lwkt_reltoken(&vp->v_mount->mnt_token);
return (EFBIG);
}
}
/*
* Extend the file's size if necessary
*/
extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
while (uio->uio_resid > 0) {
/*
* Use buffer cache I/O (via tmpfs_strategy)
*/
offset = (size_t)uio->uio_offset & BMASK;
base_offset = (off_t)uio->uio_offset - offset;
len = BSIZE - offset;
if (len > uio->uio_resid)
len = uio->uio_resid;
if ((uio->uio_offset + len) > node->tn_size) {
trivial = (uio->uio_offset <= node->tn_size);
error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
if (error)
break;
}
/*
* Read to fill in any gaps. Theoretically we could
* optimize this if the write covers the entire buffer
* and is not a UIO_NOCOPY write, however this can lead
* to a security violation exposing random kernel memory
* (whatever junk was in the backing VM pages before).
*
* So just use bread() to do the right thing.
*/
error = bread(vp, base_offset, BSIZE, &bp);
error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio);
if (error) {
kprintf("tmpfs_write uiomove error %d\n", error);
brelse(bp);
break;
}
if (uio->uio_offset > node->tn_size) {
node->tn_size = uio->uio_offset;
kflags |= NOTE_EXTEND;
}
kflags |= NOTE_WRITE;
/*
* Always try to flush the page if the request is coming
* from the pageout daemon (IO_ASYNC), else buwrite() the
* buffer.
*
* buwrite() dirties the underlying VM pages instead of
* dirtying the buffer, releasing the buffer as a clean
* buffer. This allows tmpfs to use essentially all
* available memory to cache file data. If we used bdwrite()
* the buffer cache would wind up flushing the data to
* swap too quickly.
*/
bp->b_flags |= B_AGE;
if (ap->a_ioflag & IO_ASYNC) {
bawrite(bp);
} else {
buwrite(bp);
}
if (bp->b_error) {
kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
break;
}
}
if (error) {
if (extended) {
(void)tmpfs_reg_resize(vp, oldsize, trivial);
kflags &= ~NOTE_EXTEND;
}
goto done;
}
/*
* Currently we don't set the mtime on files modified via mmap()
* because we can't tell the difference between those modifications
* and an attempt by the pageout daemon to flush tmpfs pages to
* swap.
*
* This is because in order to defer flushes as long as possible
* buwrite() works by marking the underlying VM pages dirty in
* order to be able to dispose of the buffer cache buffer without
* flushing it.
*/
TMPFS_NODE_LOCK(node);
if (uio->uio_segflg != UIO_NOCOPY)
node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED;
if (extended)
node->tn_status |= TMPFS_NODE_CHANGED;
if (node->tn_mode & (S_ISUID | S_ISGID)) {
if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
node->tn_mode &= ~(S_ISUID | S_ISGID);
}
TMPFS_NODE_UNLOCK(node);
done:
tmpfs_knote(vp, kflags);
lwkt_reltoken(&vp->v_mount->mnt_token);
return(error);
}
/*
* Write a superblock and associated information back to disk.
*/
int
ffs_sbupdate(struct ufsmount *mp, int waitfor)
{
struct fs *dfs, *fs = mp->um_fs;
struct buf *bp;
int blks;
caddr_t space;
int i, size, error, allerror = 0;
/*
* First write back the summary information.
*/
blks = howmany(fs->fs_cssize, fs->fs_fsize);
space = (caddr_t)fs->fs_csp;
for (i = 0; i < blks; i += fs->fs_frag) {
size = fs->fs_bsize;
if (i + fs->fs_frag > blks)
size = (blks - i) * fs->fs_fsize;
bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
size, 0, 0);
bcopy(space, bp->b_data, (u_int)size);
space += size;
if (waitfor != MNT_WAIT)
bawrite(bp);
else if ((error = bwrite(bp)))
allerror = error;
}
/*
* Now write back the superblock itself. If any errors occurred
* up to this point, then fail so that the superblock avoids
* being written out as clean.
*/
if (allerror)
return (allerror);
bp = getblk(mp->um_devvp, SBOFF >> (fs->fs_fshift - fs->fs_fsbtodb),
(int)fs->fs_sbsize, 0, 0);
fs->fs_fmod = 0;
fs->fs_time = time_second;
bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
/* Restore compatibility to old file systems. XXX */
dfs = (struct fs *)bp->b_data; /* XXX */
if (fs->fs_postblformat == FS_42POSTBLFMT) /* XXX */
dfs->fs_nrpos = -1; /* XXX */
if (fs->fs_inodefmt < FS_44INODEFMT) { /* XXX */
int32_t *lp, tmp; /* XXX */
/* XXX */
lp = (int32_t *)&dfs->fs_qbmask; /* XXX */
tmp = lp[4]; /* XXX */
for (i = 4; i > 0; i--) /* XXX */
lp[i] = lp[i-1]; /* XXX */
lp[0] = tmp; /* XXX */
} /* XXX */
dfs->fs_maxfilesize = mp->um_savedmaxfilesize; /* XXX */
ffs1_compat_write(dfs, mp);
if (waitfor != MNT_WAIT)
bawrite(bp);
else if ((error = bwrite(bp)))
allerror = error;
return (allerror);
}
/*
* 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.
*/
int
ffs_indirtrunc(struct inode *ip, daddr64_t lbn, daddr64_t dbn,
daddr64_t lastbn, int level, long *countp)
{
int i;
struct buf *bp;
struct fs *fs = ip->i_fs;
struct vnode *vp;
void *copy = NULL;
daddr64_t nb, nlbn, last;
long blkcount, factor;
int nblocks, blocksreleased = 0;
int error = 0, allerror = 0;
int32_t *bap1 = NULL;
#ifdef FFS2
int64_t *bap2 = NULL;
#endif
/*
* 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 *= NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = btodb(fs->fs_bsize);
/*
* 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, (int)fs->fs_bsize, 0, 0);
if (!(bp->b_flags & (B_DONE | B_DELWRI))) {
curproc->p_ru.ru_inblock++; /* pay for read */
bcstats.pendingreads++;
bcstats.numreads++;
bp->b_flags |= B_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("ffs_indirtrunc: bad buffer size");
bp->b_blkno = dbn;
VOP_STRATEGY(bp);
error = biowait(bp);
}
if (error) {
brelse(bp);
*countp = 0;
return (error);
}
#ifdef FFS2
if (ip->i_ump->um_fstype == UM_UFS2)
bap2 = (int64_t *)bp->b_data;
else
#endif
bap1 = (int32_t *)bp->b_data;
if (lastbn != -1) {
copy = malloc(fs->fs_bsize, M_TEMP, M_WAITOK);
bcopy(bp->b_data, copy, (u_int) fs->fs_bsize);
for (i = last + 1; i < NINDIR(fs); i++)
BAP_ASSIGN(ip, i, 0);
if (!DOINGASYNC(vp)) {
error = bwrite(bp);
if (error)
allerror = error;
} else {
bawrite(bp);
}
#ifdef FFS2
if (ip->i_ump->um_fstype == UM_UFS2)
bap2 = (int64_t *)copy;
else
#endif
bap1 = (int32_t *)copy;
}
/*
* Recursively free totally unused blocks.
*/
for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = BAP(ip, i);
if (nb == 0)
continue;
if (level > SINGLE) {
error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
(daddr64_t)-1, level - 1,
&blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
ffs_blkfree(ip, nb, fs->fs_bsize);
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = BAP(ip, i);
if (nb != 0) {
error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
last, level - 1, &blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
}
if (copy != NULL) {
free(copy, M_TEMP);
} else {
bp->b_flags |= B_INVAL;
brelse(bp);
}
*countp = blocksreleased;
return (allerror);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ffs_truncate(struct inode *oip, off_t length, int flags, struct ucred *cred)
{
struct vnode *ovp;
daddr64_t lastblock;
daddr64_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
daddr64_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct fs *fs;
struct buf *bp;
int offset, size, level;
long count, nblocks, vflags, blocksreleased = 0;
int i, aflags, error, allerror, indirect = 0;
off_t osize;
extern int num_indirdep;
extern int max_indirdep;
if (length < 0)
return (EINVAL);
ovp = ITOV(oip);
if (ovp->v_type != VREG &&
ovp->v_type != VDIR &&
ovp->v_type != VLNK)
return (0);
if (DIP(oip, size) == length)
return (0);
if (ovp->v_type == VLNK &&
(DIP(oip, size) < ovp->v_mount->mnt_maxsymlinklen ||
(ovp->v_mount->mnt_maxsymlinklen == 0 &&
oip->i_din1->di_blocks == 0))) {
#ifdef DIAGNOSTIC
if (length != 0)
panic("ffs_truncate: partial truncate of symlink");
#endif
memset(SHORTLINK(oip), 0, (size_t) DIP(oip, size));
DIP_ASSIGN(oip, size, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (UFS_UPDATE(oip, MNT_WAIT));
}
if ((error = getinoquota(oip)) != 0)
return (error);
uvm_vnp_setsize(ovp, length);
oip->i_ci.ci_lasta = oip->i_ci.ci_clen
= oip->i_ci.ci_cstart = oip->i_ci.ci_lastw = 0;
if (DOINGSOFTDEP(ovp)) {
if (length > 0 || softdep_slowdown(ovp)) {
/*
* If a file is only partially truncated, then
* we have to clean up the data structures
* describing the allocation past the truncation
* point. Finding and deallocating those structures
* is a lot of work. Since partial truncation occurs
* rarely, we solve the problem by syncing the file
* so that it will have no data structures left.
*/
if ((error = VOP_FSYNC(ovp, cred, MNT_WAIT)) != 0)
return (error);
} else {
(void)ufs_quota_free_blocks(oip, DIP(oip, blocks),
NOCRED);
softdep_setup_freeblocks(oip, length);
(void) vinvalbuf(ovp, 0, cred, curproc, 0, 0);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (UFS_UPDATE(oip, 0));
}
}
fs = oip->i_fs;
osize = DIP(oip, size);
/*
* 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->fs_maxfilesize)
return (EFBIG);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = UFS_BUF_ALLOC(oip, length - 1, 1,
cred, aflags, &bp);
if (error)
return (error);
if (bp->b_lblkno >= NDADDR)
indirect = 1;
DIP_ASSIGN(oip, size, length);
uvm_vnp_setsize(ovp, length);
(void) uvm_vnp_uncache(ovp);
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
error = UFS_UPDATE(oip, MNT_WAIT);
if (DOINGSOFTDEP(ovp) && num_indirdep > max_indirdep)
if (indirect) {
/*
* If the number of pending indirect block
* dependencies is sufficiently close to the
* maximum number of simultaneously mappable
* buffers force a sync on the vnode to prevent
* buffer cache exhaustion.
*/
VOP_FSYNC(ovp, curproc->p_ucred, MNT_WAIT);
}
return (error);
}
uvm_vnp_setsize(ovp, length);
/*
* 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 = blkoff(fs, length);
if (offset == 0) {
DIP_ASSIGN(oip, size, length);
} else {
lbn = lblkno(fs, length);
aflags = B_CLRBUF;
if (flags & IO_SYNC)
aflags |= B_SYNC;
error = UFS_BUF_ALLOC(oip, length - 1, 1,
cred, aflags, &bp);
if (error)
return (error);
/*
* When we are doing soft updates and the UFS_BALLOC
* above fills in a direct block hole with a full sized
* block that will be truncated down to a fragment below,
* we must flush out the block dependency with an FSYNC
* so that we do not get a soft updates inconsistency
* when we create the fragment below.
*/
if (DOINGSOFTDEP(ovp) && lbn < NDADDR &&
fragroundup(fs, blkoff(fs, length)) < fs->fs_bsize &&
(error = VOP_FSYNC(ovp, cred, MNT_WAIT)) != 0)
return (error);
DIP_ASSIGN(oip, size, length);
size = blksize(fs, oip, lbn);
(void) uvm_vnp_uncache(ovp);
if (ovp->v_type != VDIR)
bzero((char *)bp->b_data + offset,
(u_int)(size - offset));
bp->b_bcount = size;
if (aflags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
}
/*
* 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.
*/
lastblock = lblkno(fs, length + fs->fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ffs_indirtrunc below.
*/
for (level = TRIPLE; level >= SINGLE; level--) {
oldblks[NDADDR + level] = DIP(oip, ib[level]);
if (lastiblock[level] < 0) {
DIP_ASSIGN(oip, ib[level], 0);
lastiblock[level] = -1;
}
}
for (i = 0; i < NDADDR; i++) {
oldblks[i] = DIP(oip, db[i]);
if (i > lastblock)
DIP_ASSIGN(oip, db[i], 0);
}
oip->i_flag |= IN_CHANGE | IN_UPDATE;
if ((error = UFS_UPDATE(oip, MNT_WAIT)) != 0)
allerror = error;
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
for (i = 0; i < NDADDR; i++) {
newblks[i] = DIP(oip, db[i]);
DIP_ASSIGN(oip, db[i], oldblks[i]);
}
for (i = 0; i < NIADDR; i++) {
newblks[NDADDR + i] = DIP(oip, ib[i]);
DIP_ASSIGN(oip, ib[i], oldblks[NDADDR + i]);
}
DIP_ASSIGN(oip, size, osize);
vflags = ((length > 0) ? V_SAVE : 0) | V_SAVEMETA;
allerror = vinvalbuf(ovp, vflags, cred, curproc, 0, 0);
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = DIP(oip, ib[level]);
if (bn != 0) {
error = ffs_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
DIP_ASSIGN(oip, ib[level], 0);
ffs_blkfree(oip, bn, fs->fs_bsize);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = DIP(oip, db[i]);
if (bn == 0)
continue;
DIP_ASSIGN(oip, db[i], 0);
bsize = blksize(fs, oip, i);
ffs_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = DIP(oip, db[lastblock]);
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
DIP_ASSIGN(oip, size, length);
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("ffs_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ffs_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef DIAGNOSTIC
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != DIP(oip, ib[level]))
panic("ffs_truncate1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != DIP(oip, db[i]))
panic("ffs_truncate2");
#endif /* DIAGNOSTIC */
/*
* Put back the real size.
*/
DIP_ASSIGN(oip, size, length);
DIP_ADD(oip, blocks, -blocksreleased);
oip->i_flag |= IN_CHANGE;
(void)ufs_quota_free_blocks(oip, blocksreleased, NOCRED);
return (allerror);
}
/*
* Vnode op for writing.
*/
int
ext2fs_write(void *v)
{
struct vop_write_args *ap = v;
struct vnode *vp;
struct uio *uio;
struct inode *ip;
struct m_ext2fs *fs;
struct buf *bp;
int32_t lbn;
off_t osize;
int blkoffset, error, flags, ioflag, size, xfersize;
ssize_t resid, overrun;
ioflag = ap->a_ioflag;
uio = ap->a_uio;
vp = ap->a_vp;
ip = VTOI(vp);
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("%s: mode", "ext2fs_write");
#endif
/*
* If writing 0 bytes, succeed and do not change
* update time or file offset (standards compliance)
*/
if (uio->uio_resid == 0)
return (0);
switch (vp->v_type) {
case VREG:
if (ioflag & IO_APPEND)
uio->uio_offset = ext2fs_size(ip);
if ((ip->i_e2fs_flags & EXT2_APPEND) &&
uio->uio_offset != ext2fs_size(ip))
return (EPERM);
/* FALLTHROUGH */
case VLNK:
break;
case VDIR:
if ((ioflag & IO_SYNC) == 0)
panic("%s: nonsync dir write", "ext2fs_write");
break;
default:
panic("%s: type", "ext2fs_write");
}
fs = ip->i_e2fs;
if (uio->uio_offset < 0 ||
(u_int64_t)uio->uio_offset + uio->uio_resid >
((u_int64_t)0x80000000 * fs->e2fs_bsize - 1))
return (EFBIG);
/* do the filesize rlimit check */
if ((error = vn_fsizechk(vp, uio, ioflag, &overrun)))
return (error);
resid = uio->uio_resid;
osize = ext2fs_size(ip);
flags = ioflag & IO_SYNC ? B_SYNC : 0;
for (error = 0; uio->uio_resid > 0;) {
lbn = lblkno(fs, uio->uio_offset);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->e2fs_bsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (fs->e2fs_bsize > xfersize)
flags |= B_CLRBUF;
else
flags &= ~B_CLRBUF;
error = ext2fs_buf_alloc(ip,
lbn, blkoffset + xfersize, ap->a_cred, &bp, flags);
if (error)
break;
if (uio->uio_offset + xfersize > ext2fs_size(ip)) {
error = ext2fs_setsize(ip, uio->uio_offset + xfersize);
if (error)
break;
uvm_vnp_setsize(vp, ip->i_e2fs_size);
}
uvm_vnp_uncache(vp);
size = fs->e2fs_bsize - bp->b_resid;
if (size < xfersize)
xfersize = size;
error =
uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
if (ioflag & IO_SYNC)
(void)bwrite(bp);
else if (xfersize + blkoffset == fs->e2fs_bsize) {
if (doclusterwrite)
cluster_write(bp, &ip->i_ci, ext2fs_size(ip));
else
bawrite(bp);
} else
bdwrite(bp);
if (error || xfersize == 0)
break;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* If we successfully wrote any data, and we are not the superuser
* we clear the setuid and setgid bits as a precaution against
* tampering.
*/
if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
ip->i_e2fs_mode &= ~(ISUID | ISGID);
if (error) {
if (ioflag & IO_UNIT) {
(void)ext2fs_truncate(ip, osize,
ioflag & IO_SYNC, ap->a_cred);
uio->uio_offset -= resid - uio->uio_resid;
uio->uio_resid = resid;
}
} else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
error = ext2fs_update(ip, NULL, NULL, 1);
}
/* correct the result for writes clamped by vn_fsizechk() */
uio->uio_resid += overrun;
return (error);
}
/*
* Synch an open file.
*/
int
ffs_fsync(void *v)
{
struct vop_fsync_args *ap = v;
struct vnode *vp = ap->a_vp;
struct buf *bp, *nbp;
int s, error, passes, skipmeta;
if (vp->v_type == VBLK &&
vp->v_specmountpoint != NULL &&
(vp->v_specmountpoint->mnt_flag & MNT_SOFTDEP))
softdep_fsync_mountdev(vp, ap->a_waitfor);
/*
* Flush all dirty buffers associated with a vnode.
*/
passes = NIADDR + 1;
skipmeta = 0;
if (ap->a_waitfor == MNT_WAIT)
skipmeta = 1;
s = splbio();
loop:
for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp;
bp = LIST_NEXT(bp, b_vnbufs))
bp->b_flags &= ~B_SCANNED;
for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
nbp = LIST_NEXT(bp, b_vnbufs);
/*
* Reasons to skip this buffer: it has already been considered
* on this pass, this pass is the first time through on a
* synchronous flush request and the buffer being considered
* is metadata, the buffer has dependencies that will cause
* it to be redirtied and it has not already been deferred,
* or it is already being written.
*/
if (bp->b_flags & (B_BUSY | B_SCANNED))
continue;
if ((bp->b_flags & B_DELWRI) == 0)
panic("ffs_fsync: not dirty");
if (skipmeta && bp->b_lblkno < 0)
continue;
if (ap->a_waitfor != MNT_WAIT &&
LIST_FIRST(&bp->b_dep) != NULL &&
(bp->b_flags & B_DEFERRED) == 0 &&
buf_countdeps(bp, 0, 1)) {
bp->b_flags |= B_DEFERRED;
continue;
}
bremfree(bp);
buf_acquire(bp);
bp->b_flags |= B_SCANNED;
splx(s);
/*
* On our final pass through, do all I/O synchronously
* so that we can find out if our flush is failing
* because of write errors.
*/
if (passes > 0 || ap->a_waitfor != MNT_WAIT)
(void) bawrite(bp);
else if ((error = bwrite(bp)) != 0)
return (error);
s = splbio();
/*
* Since we may have slept during the I/O, we need
* to start from a known point.
*/
nbp = LIST_FIRST(&vp->v_dirtyblkhd);
}
if (skipmeta) {
skipmeta = 0;
goto loop;
}
if (ap->a_waitfor == MNT_WAIT) {
vwaitforio(vp, 0, "ffs_fsync", 0);
/*
* Ensure that any filesystem metadata associated
* with the vnode has been written.
*/
splx(s);
if ((error = softdep_sync_metadata(ap)) != 0)
return (error);
s = splbio();
if (!LIST_EMPTY(&vp->v_dirtyblkhd)) {
/*
* Block devices associated with filesystems may
* have new I/O requests posted for them even if
* the vnode is locked, so no amount of trying will
* get them clean. Thus we give block devices a
* good effort, then just give up. For all other file
* types, go around and try again until it is clean.
*/
if (passes > 0) {
passes -= 1;
goto loop;
}
#ifdef DIAGNOSTIC
if (vp->v_type != VBLK)
vprint("ffs_fsync: dirty", vp);
#endif
}
}
splx(s);
return (UFS_UPDATE(VTOI(vp), ap->a_waitfor == MNT_WAIT));
}
/*
* ffs_balloc(struct vnode *a_vp, ufs_daddr_t a_lbn, int a_size,
* struct ucred *a_cred, int a_flags, struct buf *a_bpp)
*
* Balloc defines the structure of filesystem storage by allocating
* the physical blocks on a device given the inode and the logical
* block number in a file.
*
* NOTE: B_CLRBUF - this flag tells balloc to clear invalid portions
* of the buffer. However, any dirty bits will override missing
* valid bits. This case occurs when writable mmaps are truncated
* and then extended.
*/
int
ffs_balloc(struct vop_balloc_args *ap)
{
struct inode *ip;
ufs_daddr_t lbn;
int size;
struct ucred *cred;
int flags;
struct fs *fs;
ufs_daddr_t nb;
struct buf *bp, *nbp, *dbp;
struct vnode *vp;
struct indir indirs[NIADDR + 2];
ufs_daddr_t newb, *bap, pref;
int deallocated, osize, nsize, num, i, error;
ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx;
int seqcount;
vp = ap->a_vp;
ip = VTOI(vp);
fs = ip->i_fs;
lbn = lblkno(fs, ap->a_startoffset);
size = blkoff(fs, ap->a_startoffset) + ap->a_size;
if (size > fs->fs_bsize)
panic("ffs_balloc: blk too big");
*ap->a_bpp = NULL;
if (lbn < 0)
return (EFBIG);
cred = ap->a_cred;
flags = ap->a_flags;
/*
* The vnode must be locked for us to be able to safely mess
* around with the inode.
*/
if (vn_islocked(vp) != LK_EXCLUSIVE) {
panic("ffs_balloc: vnode %p not exclusively locked!", vp);
}
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
nb = lblkno(fs, ip->i_size);
if (nb < NDADDR && nb < lbn) {
/*
* The filesize prior to this write can fit in direct
* blocks (ex. fragmentation is possibly done)
* we are now extending the file write beyond
* the block which has end of the file prior to this write.
*/
osize = blksize(fs, ip, nb);
/*
* osize gives disk allocated size in the last block. It is
* either in fragments or a file system block size.
*/
if (osize < fs->fs_bsize && osize > 0) {
/* A few fragments are already allocated, since the
* current extends beyond this block allocated the
* complete block as fragments are on in last block.
*/
error = ffs_realloccg(ip, nb,
ffs_blkpref(ip, nb, (int)nb, &ip->i_db[0]),
osize, (int)fs->fs_bsize, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dofftofsb(fs, bp->b_bio2.bio_offset),
ip->i_db[nb], fs->fs_bsize, osize, bp);
/* adjust the inode size, we just grew */
ip->i_size = smalllblktosize(fs, nb + 1);
ip->i_db[nb] = dofftofsb(fs, bp->b_bio2.bio_offset);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & B_SYNC)
bwrite(bp);
else
bawrite(bp);
/* bp is already released here */
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
nb = ip->i_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lblktodoff(fs, lbn), fs->fs_bsize, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
*ap->a_bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lblktodoff(fs, lbn),
osize, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
} else {
/*
* NOTE: ffs_realloccg() issues a bread().
*/
error = ffs_realloccg(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn,
&ip->i_db[0]), osize, nsize, cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dofftofsb(fs, bp->b_bio2.bio_offset),
nb, nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
error = ffs_alloc(ip, lbn,
ffs_blkpref(ip, lbn, (int)lbn, &ip->i_db[0]),
nsize, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lblktodoff(fs, lbn), nsize, 0, 0);
bp->b_bio2.bio_offset = fsbtodoff(fs, newb);
if (flags & B_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
ip->i_db[lbn] = dofftofsb(fs, bp->b_bio2.bio_offset);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*ap->a_bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef DIAGNOSTIC
if (num < 1)
panic ("ffs_balloc: ufs_bmaparray returned indirect block");
#endif
/*
* Get a handle on the data block buffer before working through
* indirect blocks to avoid a deadlock between the VM system holding
* a locked VM page and issuing a BMAP (which tries to lock the
* indirect blocks), and the filesystem holding a locked indirect
* block and then trying to read a data block (which tries to lock
* the underlying VM pages).
*/
dbp = getblk(vp, lblktodoff(fs, lbn), fs->fs_bsize, 0, 0);
/*
* Setup undo history
*/
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
unwindidx = -1;
/*
* Fetch the first indirect block directly from the inode, allocating
* one if necessary.
*/
--num;
nb = ip->i_ib[indirs[0].in_off];
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, 0, NULL);
/*
* If the filesystem has run out of space we can skip the
* full fsync/undo of the main [fail] case since no undo
* history has been built yet. Hence the goto fail2.
*/
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
cred, &newb)) != 0)
goto fail2;
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, lblktodoff(fs, indirs[1].in_lbn),
fs->fs_bsize, 0, 0);
bp->b_bio2.bio_offset = fsbtodoff(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &ip->i_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
for (i = 1;;) {
error = bread(vp, lblktodoff(fs, indirs[i].in_lbn), (int)fs->fs_bsize, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
if (pref == 0)
pref = ffs_blkpref(ip, lbn, 0, NULL);
if ((error =
ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize, cred, &newb)) != 0) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, lblktodoff(fs, indirs[i].in_lbn),
fs->fs_bsize, 0, 0);
nbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* Get the data block, allocating if necessary. We have already
* called getblk() on the data block buffer, dbp. If we have to
* allocate it and B_CLRBUF has been set the inference is an intention
* to zero out the related disk blocks, so we do not have to issue
* a read. Instead we simply call vfs_bio_clrbuf(). If B_CLRBUF is
* not set the caller intends to overwrite the entire contents of the
* buffer and we don't waste time trying to clean up the contents.
*
* bp references the current indirect block. When allocating,
* the block must be updated.
*/
if (nb == 0) {
pref = ffs_blkpref(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip,
lbn, pref, (int)fs->fs_bsize, cred, &newb);
if (error) {
brelse(bp);
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
dbp->b_bio2.bio_offset = fsbtodoff(fs, nb);
if (flags & B_CLRBUF)
vfs_bio_clrbuf(dbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, dbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & B_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
*ap->a_bpp = dbp;
return (0);
}
brelse(bp);
/*
* At this point all related indirect blocks have been allocated
* if necessary and released. bp is no longer valid. dbp holds
* our getblk()'d data block.
*
* XXX we previously performed a cluster_read operation here.
*/
if (flags & B_CLRBUF) {
/*
* If B_CLRBUF is set we must validate the invalid portions
* of the buffer. This typically requires a read-before-
* write. The strategy call will fill in bio_offset in that
* case.
*
* If we hit this case we do a cluster read if possible
* since nearby data blocks are likely to be accessed soon
* too.
*/
if ((dbp->b_flags & B_CACHE) == 0) {
bqrelse(dbp);
seqcount = (flags & B_SEQMASK) >> B_SEQSHIFT;
if (seqcount &&
(vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, (off_t)ip->i_size,
lblktodoff(fs, lbn),
(int)fs->fs_bsize,
fs->fs_bsize,
seqcount * BKVASIZE,
&dbp);
} else {
error = bread(vp, lblktodoff(fs, lbn),
(int)fs->fs_bsize, &dbp);
}
if (error)
goto fail;
} else {
/*
* 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
ffs_indirtrunc(struct inode *ip, ufs_daddr_t lbn, ufs_daddr_t dbn,
ufs_daddr_t lastbn, int level, long *countp)
{
int i;
struct buf *bp;
struct fs *fs = ip->i_fs;
ufs_daddr_t *bap;
struct vnode *vp;
ufs_daddr_t *copy = NULL, nb, nlbn, last;
long blkcount, factor;
int nblocks, blocksreleased = 0;
int error = 0, allerror = 0;
/*
* 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 *= NINDIR(fs);
last = lastbn;
if (lastbn > 0)
last /= factor;
nblocks = btodb(fs->fs_bsize);
/*
* 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 bio_offset field
* explicitly instead of letting bread do everything for us.
*/
vp = ITOV(ip);
bp = getblk(vp, lblktodoff(fs, lbn), (int)fs->fs_bsize, 0, 0);
if ((bp->b_flags & B_CACHE) == 0) {
bp->b_flags &= ~(B_ERROR|B_INVAL);
bp->b_cmd = BUF_CMD_READ;
if (bp->b_bcount > bp->b_bufsize)
panic("ffs_indirtrunc: bad buffer size");
/*
* BIO is bio2 which chains back to bio1. We wait
* on bio1.
*/
bp->b_bio2.bio_offset = dbtodoff(fs, dbn);
bp->b_bio1.bio_done = biodone_sync;
bp->b_bio1.bio_flags |= BIO_SYNC;
vfs_busy_pages(vp, bp);
/*
* Access the block device layer using the device vnode
* and the translated block number (bio2) instead of the
* file vnode (vp) and logical block number (bio1).
*
* Even though we are bypassing the vnode layer, we still
* want the vnode state to indicate that an I/O on its behalf
* is in progress.
*/
bio_start_transaction(&bp->b_bio1, &vp->v_track_read);
vn_strategy(ip->i_devvp, &bp->b_bio2);
error = biowait(&bp->b_bio1, "biord");
}
if (error) {
brelse(bp);
*countp = 0;
return (error);
}
bap = (ufs_daddr_t *)bp->b_data;
if (lastbn != -1) {
copy = kmalloc(fs->fs_bsize, M_TEMP, M_WAITOK);
bcopy((caddr_t)bap, (caddr_t)copy, (uint)fs->fs_bsize);
bzero((caddr_t)&bap[last + 1],
(uint)(NINDIR(fs) - (last + 1)) * sizeof (ufs_daddr_t));
if (DOINGASYNC(vp)) {
bawrite(bp);
} else {
error = bwrite(bp);
if (error)
allerror = error;
}
bap = copy;
}
/*
* Recursively free totally unused blocks.
*/
for (i = NINDIR(fs) - 1, nlbn = lbn + 1 - i * factor; i > last;
i--, nlbn += factor) {
nb = bap[i];
if (nb == 0)
continue;
if (level > SINGLE) {
if ((error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
(ufs_daddr_t)-1, level - 1, &blkcount)) != 0)
allerror = error;
blocksreleased += blkcount;
}
ffs_blkfree(ip, nb, fs->fs_bsize);
blocksreleased += nblocks;
}
/*
* Recursively free last partial block.
*/
if (level > SINGLE && lastbn >= 0) {
last = lastbn % factor;
nb = bap[i];
if (nb != 0) {
error = ffs_indirtrunc(ip, nlbn, fsbtodb(fs, nb),
last, level - 1, &blkcount);
if (error)
allerror = error;
blocksreleased += blkcount;
}
}
if (copy != NULL) {
kfree(copy, M_TEMP);
} else {
bp->b_flags |= B_INVAL | B_NOCACHE;
brelse(bp);
}
*countp = blocksreleased;
return (allerror);
}
/*
* Update the access, modified, and inode change times as specified by the
* IN_ACCESS, IN_UPDATE, and IN_CHANGE flags respectively. Write the inode
* to disk if the IN_MODIFIED flag is set (it may be set initially, or by
* the timestamp update). The IN_LAZYMOD flag is set to force a write
* later if not now. The IN_LAZYACCESS is set instead of IN_MODIFIED if the fs
* is currently being suspended (or is suspended) and vnode has been accessed.
* If we write now, then clear IN_MODIFIED, IN_LAZYACCESS and IN_LAZYMOD to
* reflect the presumably successful write, and if waitfor is set, then wait
* for the write to complete.
*/
int
ffs_update (vnode *vp, int waitfor)
{
int error = 0;
print("HARVEY TODO: %s\n", __func__);
#if 0
struct fs *fs;
struct buf *bp;
struct inode *ip;
int flags, error;
ASSERT_VOP_ELOCKED(vp, "ffs_update");
ufs_itimes(vp);
ip = VTOI(vp);
if ((ip->i_flag & IN_MODIFIED) == 0 && waitfor == 0)
return (0);
ip->i_flag &= ~(IN_LAZYACCESS | IN_LAZYMOD | IN_MODIFIED);
fs = ITOFS(ip);
if (fs->fs_ronly && ITOUMP(ip)->um_fsckpid == 0)
return (0);
/*
* If we are updating a snapshot and another process is currently
* writing the buffer containing the inode for this snapshot then
* a deadlock can occur when it tries to check the snapshot to see
* if that block needs to be copied. Thus when updating a snapshot
* we check to see if the buffer is already locked, and if it is
* we drop the snapshot lock until the buffer has been written
* and is available to us. We have to grab a reference to the
* snapshot vnode to prevent it from being removed while we are
* waiting for the buffer.
*/
flags = 0;
if (IS_SNAPSHOT(ip))
flags = GB_LOCK_NOWAIT;
loop:
error = breadn_flags(ITODEVVP(ip),
fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
(int) fs->fs_bsize, 0, 0, 0, NOCRED, flags, &bp);
if (error != 0) {
if (error != EBUSY)
return (error);
KASSERT((IS_SNAPSHOT(ip)), ("EBUSY from non-snapshot"));
/*
* Wait for our inode block to become available.
*
* Hold a reference to the vnode to protect against
* ffs_snapgone(). Since we hold a reference, it can only
* get reclaimed (VI_DOOMED flag) in a forcible downgrade
* or unmount. For an unmount, the entire filesystem will be
* gone, so we cannot attempt to touch anything associated
* with it while the vnode is unlocked; all we can do is
* pause briefly and try again. If when we relock the vnode
* we discover that it has been reclaimed, updating it is no
* longer necessary and we can just return an error.
*/
vref(vp);
VOP_UNLOCK(vp, 0);
pause("ffsupd", 1);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vrele(vp);
if ((vp->v_iflag & VI_DOOMED) != 0)
return (ENOENT);
goto loop;
}
if (DOINGSOFTDEP(vp))
softdep_update_inodeblock(ip, bp, waitfor);
else if (ip->i_effnlink != ip->i_nlink)
panic("ffs_update: bad link cnt");
if (I_IS_UFS1(ip)) {
*((struct ufs1_dinode *)bp->b_data +
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din1;
/* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
random_harvest_queue(&(ip->i_din1), sizeof(ip->i_din1), 1, RANDOM_FS_ATIME);
} else {
*((struct ufs2_dinode *)bp->b_data +
ino_to_fsbo(fs, ip->i_number)) = *ip->i_din2;
/* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
random_harvest_queue(&(ip->i_din2), sizeof(ip->i_din2), 1, RANDOM_FS_ATIME);
}
if (waitfor)
error = bwrite(bp);
else if (vm_page_count_severe() || buf_dirty_count_severe()) {
bawrite(bp);
error = 0;
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
error = 0;
}
#endif // 0
return (error);
}
// ffs文件系统的写入操作
int
ffs_write(void *v)
{
struct vop_write_args *ap = v;
struct vnode *vp;
struct uio *uio;
struct inode *ip;
struct fs *fs;
struct buf *bp;
daddr_t lbn;
off_t osize;
int blkoffset, error, extended, flags, ioflag, size, xfersize;
ssize_t resid, overrun;
extended = 0;
ioflag = ap->a_ioflag;
uio = ap->a_uio;
vp = ap->a_vp;
ip = VTOI(vp);
#ifdef DIAGNOSTIC
if (uio->uio_rw != UIO_WRITE)
panic("ffs_write: mode");
#endif
/*
* If writing 0 bytes, succeed and do not change
* update time or file offset (standards compliance)
*/
if (uio->uio_resid == 0)
return (0);
switch (vp->v_type) {
case VREG:
if (ioflag & IO_APPEND)
uio->uio_offset = DIP(ip, size);
if ((DIP(ip, flags) & APPEND) && uio->uio_offset != DIP(ip, size))
return (EPERM);
/* FALLTHROUGH */
case VLNK:
break;
case VDIR:
if ((ioflag & IO_SYNC) == 0)
panic("ffs_write: nonsync dir write");
break;
default:
panic("ffs_write: type");
}
fs = ip->i_fs;
if (uio->uio_offset < 0 ||
(u_int64_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize)
return (EFBIG);
/* do the filesize rlimit check */
if ((error = vn_fsizechk(vp, uio, ioflag, &overrun)))
return (error);
resid = uio->uio_resid;
osize = DIP(ip, size);
flags = ioflag & IO_SYNC ? B_SYNC : 0;
for (error = 0; uio->uio_resid > 0;) {
lbn = lblkno(fs, uio->uio_offset);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->fs_bsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (fs->fs_bsize > xfersize)
flags |= B_CLRBUF;
else
flags &= ~B_CLRBUF;
if ((error = UFS_BUF_ALLOC(ip, uio->uio_offset, xfersize,
ap->a_cred, flags, &bp)) != 0)
break;
if (uio->uio_offset + xfersize > DIP(ip, size)) {
DIP_ASSIGN(ip, size, uio->uio_offset + xfersize);
uvm_vnp_setsize(vp, DIP(ip, size));
extended = 1;
}
(void)uvm_vnp_uncache(vp);
size = blksize(fs, ip, lbn) - bp->b_resid;
if (size < xfersize)
xfersize = size;
error =
uiomovei(bp->b_data + blkoffset, xfersize, uio);
if (error != 0)
memset(bp->b_data + blkoffset, 0, xfersize);
#if 0
if (ioflag & IO_NOCACHE)
bp->b_flags |= B_NOCACHE;
#endif
if (ioflag & IO_SYNC)
(void)bwrite(bp);
else if (xfersize + blkoffset == fs->fs_bsize) {
if (doclusterwrite)
cluster_write(bp, &ip->i_ci, DIP(ip, size));
else
bawrite(bp);
} else
bdwrite(bp);
if (error || xfersize == 0)
break;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* If we successfully wrote any data, and we are not the superuser
* we clear the setuid and setgid bits as a precaution against
* tampering.
*/
if (resid > uio->uio_resid && ap->a_cred && ap->a_cred->cr_uid != 0)
DIP_ASSIGN(ip, mode, DIP(ip, mode) & ~(ISUID | ISGID));
if (resid > uio->uio_resid)
VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
if (error) {
if (ioflag & IO_UNIT) {
(void)UFS_TRUNCATE(ip, osize,
ioflag & IO_SYNC, ap->a_cred);
uio->uio_offset -= resid - uio->uio_resid;
uio->uio_resid = resid;
}
} else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) {
error = UFS_UPDATE(ip, 1);
}
/* correct the result for writes clamped by vn_fsizechk() */
uio->uio_resid += overrun;
return (error);
}
/*
* Balloc defines the structure of filesystem storage
* by allocating the physical blocks on a device given
* the inode and the logical block number in a file.
* This is the allocation strategy for UFS1. Below is
* the allocation strategy for UFS2.
*/
int
ffs_balloc_ufs1(struct vnode *vp, off_t startoffset, int size,
struct ucred *cred, int flags, struct buf **bpp)
{
struct inode *ip;
struct ufs1_dinode *dp;
ufs_lbn_t lbn, lastlbn;
struct fs *fs;
ufs1_daddr_t nb;
struct buf *bp, *nbp;
struct ufsmount *ump;
struct indir indirs[NIADDR + 2];
int deallocated, osize, nsize, num, i, error;
ufs2_daddr_t newb;
ufs1_daddr_t *bap, pref;
ufs1_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
ufs2_daddr_t *lbns_remfree, lbns[NIADDR + 1];
int unwindidx = -1;
int saved_inbdflush;
static struct timeval lastfail;
static int curfail;
int reclaimed;
ip = VTOI(vp);
dp = ip->i_din1;
fs = ip->i_fs;
ump = ip->i_ump;
lbn = lblkno(fs, startoffset);
size = blkoff(fs, startoffset) + size;
reclaimed = 0;
if (size > fs->fs_bsize)
panic("ffs_balloc_ufs1: blk too big");
*bpp = NULL;
if (flags & IO_EXT)
return (EOPNOTSUPP);
if (lbn < 0)
return (EFBIG);
if (DOINGSOFTDEP(vp))
softdep_prealloc(vp, MNT_WAIT);
/*
* If the next write will extend the file into a new block,
* and the file is currently composed of a fragment
* this fragment has to be extended to be a full block.
*/
lastlbn = lblkno(fs, ip->i_size);
if (lastlbn < NDADDR && lastlbn < lbn) {
nb = lastlbn;
osize = blksize(fs, ip, nb);
if (osize < fs->fs_bsize && osize > 0) {
UFS_LOCK(ump);
error = ffs_realloccg(ip, nb, dp->di_db[nb],
ffs_blkpref_ufs1(ip, lastlbn, (int)nb,
&dp->di_db[0]), osize, (int)fs->fs_bsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, nb,
dbtofsb(fs, bp->b_blkno), dp->di_db[nb],
fs->fs_bsize, osize, bp);
ip->i_size = smalllblktosize(fs, nb + 1);
dp->di_size = ip->i_size;
dp->di_db[nb] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (flags & IO_SYNC)
bwrite(bp);
else
bawrite(bp);
}
}
/*
* The first NDADDR blocks are direct blocks
*/
if (lbn < NDADDR) {
if (flags & BA_METAONLY)
panic("ffs_balloc_ufs1: BA_METAONLY for direct block");
nb = dp->di_db[lbn];
if (nb != 0 && ip->i_size >= smalllblktosize(fs, lbn + 1)) {
error = bread(vp, lbn, fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
*bpp = bp;
return (0);
}
if (nb != 0) {
/*
* Consider need to reallocate a fragment.
*/
osize = fragroundup(fs, blkoff(fs, ip->i_size));
nsize = fragroundup(fs, size);
if (nsize <= osize) {
error = bread(vp, lbn, osize, NOCRED, &bp);
if (error) {
brelse(bp);
return (error);
}
bp->b_blkno = fsbtodb(fs, nb);
} else {
UFS_LOCK(ump);
error = ffs_realloccg(ip, lbn, dp->di_db[lbn],
ffs_blkpref_ufs1(ip, lbn, (int)lbn,
&dp->di_db[0]), osize, nsize, flags,
cred, &bp);
if (error)
return (error);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn,
dbtofsb(fs, bp->b_blkno), nb,
nsize, osize, bp);
}
} else {
if (ip->i_size < smalllblktosize(fs, lbn + 1))
nsize = fragroundup(fs, size);
else
nsize = fs->fs_bsize;
UFS_LOCK(ump);
error = ffs_alloc(ip, lbn,
ffs_blkpref_ufs1(ip, lbn, (int)lbn, &dp->di_db[0]),
nsize, flags, cred, &newb);
if (error)
return (error);
bp = getblk(vp, lbn, nsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, newb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocdirect(ip, lbn, newb, 0,
nsize, 0, bp);
}
dp->di_db[lbn] = dbtofsb(fs, bp->b_blkno);
ip->i_flag |= IN_CHANGE | IN_UPDATE;
*bpp = bp;
return (0);
}
/*
* Determine the number of levels of indirection.
*/
pref = 0;
if ((error = ufs_getlbns(vp, lbn, indirs, &num)) != 0)
return(error);
#ifdef INVARIANTS
if (num < 1)
panic ("ffs_balloc_ufs1: ufs_getlbns returned indirect block");
#endif
saved_inbdflush = curthread_pflags_set(TDP_INBDFLUSH);
/*
* Fetch the first indirect block allocating if necessary.
*/
--num;
nb = dp->di_ib[indirs[0].in_off];
allocib = NULL;
allocblk = allociblk;
lbns_remfree = lbns;
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags, cred, &newb)) != 0) {
curthread_pflags_restore(saved_inbdflush);
return (error);
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[1].in_lbn;
bp = getblk(vp, indirs[1].in_lbn, fs->fs_bsize, 0, 0, 0);
bp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(bp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocdirect(ip, NDADDR + indirs[0].in_off,
newb, 0, fs->fs_bsize, 0, bp);
bdwrite(bp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if (DOINGASYNC(vp))
bdwrite(bp);
else if ((error = bwrite(bp)) != 0)
goto fail;
}
allocib = &dp->di_ib[indirs[0].in_off];
*allocib = nb;
ip->i_flag |= IN_CHANGE | IN_UPDATE;
}
/*
* Fetch through the indirect blocks, allocating as necessary.
*/
retry:
for (i = 1;;) {
error = bread(vp,
indirs[i].in_lbn, (int)fs->fs_bsize, NOCRED, &bp);
if (error) {
brelse(bp);
goto fail;
}
bap = (ufs1_daddr_t *)bp->b_data;
nb = bap[indirs[i].in_off];
if (i == num)
break;
i += 1;
if (nb != 0) {
bqrelse(bp);
continue;
}
UFS_LOCK(ump);
if (pref == 0)
pref = ffs_blkpref_ufs1(ip, lbn, 0, (ufs1_daddr_t *)0);
if ((error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb)) != 0) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = indirs[i].in_lbn;
nbp = getblk(vp, indirs[i].in_lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp)) {
softdep_setup_allocindir_meta(nbp, ip, bp,
indirs[i - 1].in_off, nb);
bdwrite(nbp);
} else {
/*
* Write synchronously so that indirect blocks
* never point at garbage.
*/
if ((error = bwrite(nbp)) != 0) {
brelse(bp);
goto fail;
}
}
bap[indirs[i - 1].in_off] = nb;
if (allocib == NULL && unwindidx < 0)
unwindidx = i - 1;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
}
/*
* If asked only for the indirect block, then return it.
*/
if (flags & BA_METAONLY) {
curthread_pflags_restore(saved_inbdflush);
*bpp = bp;
return (0);
}
/*
* Get the data block, allocating if necessary.
*/
if (nb == 0) {
UFS_LOCK(ump);
pref = ffs_blkpref_ufs1(ip, lbn, indirs[i].in_off, &bap[0]);
error = ffs_alloc(ip, lbn, pref, (int)fs->fs_bsize,
flags | IO_BUFLOCKED, cred, &newb);
if (error) {
brelse(bp);
if (++reclaimed == 1) {
UFS_LOCK(ump);
softdep_request_cleanup(fs, vp, cred,
FLUSH_BLOCKS_WAIT);
UFS_UNLOCK(ump);
goto retry;
}
if (ppsratecheck(&lastfail, &curfail, 1)) {
ffs_fserr(fs, ip->i_number, "filesystem full");
uprintf("\n%s: write failed, filesystem "
"is full\n", fs->fs_fsmnt);
}
goto fail;
}
nb = newb;
*allocblk++ = nb;
*lbns_remfree++ = lbn;
nbp = getblk(vp, lbn, fs->fs_bsize, 0, 0, 0);
nbp->b_blkno = fsbtodb(fs, nb);
if (flags & BA_CLRBUF)
vfs_bio_clrbuf(nbp);
if (DOINGSOFTDEP(vp))
softdep_setup_allocindir_page(ip, lbn, bp,
indirs[i].in_off, nb, 0, nbp);
bap[indirs[i].in_off] = nb;
/*
* If required, write synchronously, otherwise use
* delayed write.
*/
if (flags & IO_SYNC) {
bwrite(bp);
} else {
if (bp->b_bufsize == fs->fs_bsize)
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
curthread_pflags_restore(saved_inbdflush);
*bpp = nbp;
return (0);
}
brelse(bp);
if (flags & BA_CLRBUF) {
int seqcount = (flags & BA_SEQMASK) >> BA_SEQSHIFT;
if (seqcount && (vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) {
error = cluster_read(vp, ip->i_size, lbn,
(int)fs->fs_bsize, NOCRED,
MAXBSIZE, seqcount, &nbp);
} else {
error = bread(vp, lbn, (int)fs->fs_bsize, NOCRED, &nbp);
}
if (error) {
brelse(nbp);
goto fail;
}
} else {
/*
* Vnode op for writing.
*/
static int
ext2_write(struct vop_write_args *ap)
{
struct vnode *vp;
struct uio *uio;
struct inode *ip;
struct m_ext2fs *fs;
struct buf *bp;
daddr_t lbn;
off_t osize;
int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize;
ioflag = ap->a_ioflag;
uio = ap->a_uio;
vp = ap->a_vp;
seqcount = ioflag >> IO_SEQSHIFT;
ip = VTOI(vp);
#ifdef INVARIANTS
if (uio->uio_rw != UIO_WRITE)
panic("%s: mode", "ext2_write");
#endif
switch (vp->v_type) {
case VREG:
if (ioflag & IO_APPEND)
uio->uio_offset = ip->i_size;
if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
return (EPERM);
/* FALLTHROUGH */
case VLNK:
break;
case VDIR:
/* XXX differs from ffs -- this is called from ext2_mkdir(). */
if ((ioflag & IO_SYNC) == 0)
panic("ext2_write: nonsync dir write");
break;
default:
panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp,
vp->v_type, (intmax_t)uio->uio_offset,
(intmax_t)uio->uio_resid);
}
KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0"));
KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0"));
fs = ip->i_e2fs;
if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize)
return (EFBIG);
/*
* Maybe this should be above the vnode op call, but so long as
* file servers have no limits, I don't think it matters.
*/
if (vn_rlimit_fsize(vp, uio, uio->uio_td))
return (EFBIG);
resid = uio->uio_resid;
osize = ip->i_size;
if (seqcount > BA_SEQMAX)
flags = BA_SEQMAX << BA_SEQSHIFT;
else
flags = seqcount << BA_SEQSHIFT;
if ((ioflag & IO_SYNC) && !DOINGASYNC(vp))
flags |= IO_SYNC;
for (error = 0; uio->uio_resid > 0;) {
lbn = lblkno(fs, uio->uio_offset);
blkoffset = blkoff(fs, uio->uio_offset);
xfersize = fs->e2fs_fsize - blkoffset;
if (uio->uio_resid < xfersize)
xfersize = uio->uio_resid;
if (uio->uio_offset + xfersize > ip->i_size)
vnode_pager_setsize(vp, uio->uio_offset + xfersize);
/*
* We must perform a read-before-write if the transfer size
* does not cover the entire buffer.
*/
if (fs->e2fs_bsize > xfersize)
flags |= BA_CLRBUF;
else
flags &= ~BA_CLRBUF;
error = ext2_balloc(ip, lbn, blkoffset + xfersize,
ap->a_cred, &bp, flags);
if (error != 0)
break;
/*
* If the buffer is not valid and we did not clear garbage
* out above, we have to do so here even though the write
* covers the entire buffer in order to avoid a mmap()/write
* race where another process may see the garbage prior to
* the uiomove() for a write replacing it.
*/
if ((bp->b_flags & B_CACHE) == 0 && fs->e2fs_bsize <= xfersize)
vfs_bio_clrbuf(bp);
if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL))
bp->b_flags |= B_NOCACHE;
if (uio->uio_offset + xfersize > ip->i_size)
ip->i_size = uio->uio_offset + xfersize;
size = blksize(fs, ip, lbn) - bp->b_resid;
if (size < xfersize)
xfersize = size;
error =
uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio);
if (ioflag & (IO_VMIO|IO_DIRECT)) {
bp->b_flags |= B_RELBUF;
}
/*
* If IO_SYNC each buffer is written synchronously. Otherwise
* if we have a severe page deficiency write the buffer
* asynchronously. Otherwise try to cluster, and if that
* doesn't do it then either do an async write (if O_DIRECT),
* or a delayed write (if not).
*/
if (ioflag & IO_SYNC) {
(void)bwrite(bp);
} else if (vm_page_count_severe() ||
buf_dirty_count_severe() ||
(ioflag & IO_ASYNC)) {
bp->b_flags |= B_CLUSTEROK;
bawrite(bp);
} else if (xfersize + blkoffset == fs->e2fs_fsize) {
if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) {
bp->b_flags |= B_CLUSTEROK;
cluster_write(vp, bp, ip->i_size, seqcount, 0);
} else {
bawrite(bp);
}
} else if (ioflag & IO_DIRECT) {
bp->b_flags |= B_CLUSTEROK;
bawrite(bp);
} else {
bp->b_flags |= B_CLUSTEROK;
bdwrite(bp);
}
if (error || xfersize == 0)
break;
}
/*
* If we successfully wrote any data, and we are not the superuser
* we clear the setuid and setgid bits as a precaution against
* tampering.
*/
if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid &&
ap->a_cred) {
if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
ip->i_mode &= ~(ISUID | ISGID);
}
if (error) {
if (ioflag & IO_UNIT) {
(void)ext2_truncate(vp, osize,
ioflag & IO_SYNC, ap->a_cred, uio->uio_td);
uio->uio_offset -= resid - uio->uio_resid;
uio->uio_resid = resid;
}
}
if (uio->uio_resid != resid) {
ip->i_flag |= IN_CHANGE | IN_UPDATE;
if (ioflag & IO_SYNC)
error = ext2_update(vp, 1);
}
return (error);
}
/*
* Truncate the inode oip to at most length size, freeing the
* disk blocks.
*/
int
ext2_truncate(struct vnode *vp, off_t length, int flags, struct ucred *cred,
struct thread *td)
{
struct vnode *ovp = vp;
int32_t lastblock;
struct inode *oip;
int32_t bn, lbn, lastiblock[NIADDR], indir_lbn[NIADDR];
uint32_t oldblks[NDADDR + NIADDR], newblks[NDADDR + NIADDR];
struct m_ext2fs *fs;
struct buf *bp;
int offset, size, level;
e4fs_daddr_t count, nblocks, blocksreleased = 0;
int error, i, allerror;
off_t osize;
#ifdef INVARIANTS
struct bufobj *bo;
#endif
oip = VTOI(ovp);
#ifdef INVARIANTS
bo = &ovp->v_bufobj;
#endif
ASSERT_VOP_LOCKED(vp, "ext2_truncate");
if (length < 0)
return (EINVAL);
if (ovp->v_type == VLNK &&
oip->i_size < ovp->v_mount->mnt_maxsymlinklen) {
#ifdef INVARIANTS
if (length != 0)
panic("ext2_truncate: partial truncate of symlink");
#endif
bzero((char *)&oip->i_shortlink, (u_int)oip->i_size);
oip->i_size = 0;
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, 1));
}
if (oip->i_size == length) {
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, 0));
}
fs = oip->i_e2fs;
osize = oip->i_size;
/*
* 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 > oip->i_e2fs->e2fs_maxfilesize)
return (EFBIG);
vnode_pager_setsize(ovp, length);
offset = blkoff(fs, length - 1);
lbn = lblkno(fs, length - 1);
flags |= BA_CLRBUF;
error = ext2_balloc(oip, lbn, offset + 1, cred, &bp, flags);
if (error) {
vnode_pager_setsize(vp, osize);
return (error);
}
oip->i_size = length;
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(ovp))
bdwrite(bp);
else
bawrite(bp);
oip->i_flag |= IN_CHANGE | IN_UPDATE;
return (ext2_update(ovp, !DOINGASYNC(ovp)));
}
/*
* Shorten the size of the file. If the file is not being
* truncated to a block boundry, the contents of the
* partial block following the end of the file must be
* zero'ed in case it ever become accessible again because
* of subsequent file growth.
*/
/* I don't understand the comment above */
offset = blkoff(fs, length);
if (offset == 0) {
oip->i_size = length;
} else {
lbn = lblkno(fs, length);
flags |= BA_CLRBUF;
error = ext2_balloc(oip, lbn, offset, cred, &bp, flags);
if (error)
return (error);
oip->i_size = length;
size = blksize(fs, oip, lbn);
bzero((char *)bp->b_data + offset, (u_int)(size - offset));
allocbuf(bp, size);
if (bp->b_bufsize == fs->e2fs_bsize)
bp->b_flags |= B_CLUSTEROK;
if (flags & IO_SYNC)
bwrite(bp);
else if (DOINGASYNC(ovp))
bdwrite(bp);
else
bawrite(bp);
}
/*
* 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.
*/
lastblock = lblkno(fs, length + fs->e2fs_bsize - 1) - 1;
lastiblock[SINGLE] = lastblock - NDADDR;
lastiblock[DOUBLE] = lastiblock[SINGLE] - NINDIR(fs);
lastiblock[TRIPLE] = lastiblock[DOUBLE] - NINDIR(fs) * NINDIR(fs);
nblocks = btodb(fs->e2fs_bsize);
/*
* Update file and block pointers on disk before we start freeing
* blocks. If we crash before free'ing blocks below, the blocks
* will be returned to the free list. lastiblock values are also
* normalized to -1 for calls to ext2_indirtrunc below.
*/
for (level = TRIPLE; level >= SINGLE; level--) {
oldblks[NDADDR + level] = oip->i_ib[level];
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
lastiblock[level] = -1;
}
}
for (i = 0; i < NDADDR; i++) {
oldblks[i] = oip->i_db[i];
if (i > lastblock)
oip->i_db[i] = 0;
}
oip->i_flag |= IN_CHANGE | IN_UPDATE;
allerror = ext2_update(ovp, !DOINGASYNC(ovp));
/*
* Having written the new inode to disk, save its new configuration
* and put back the old block pointers long enough to process them.
* Note that we save the new block configuration so we can check it
* when we are done.
*/
for (i = 0; i < NDADDR; i++) {
newblks[i] = oip->i_db[i];
oip->i_db[i] = oldblks[i];
}
for (i = 0; i < NIADDR; i++) {
newblks[NDADDR + i] = oip->i_ib[i];
oip->i_ib[i] = oldblks[NDADDR + i];
}
oip->i_size = osize;
error = vtruncbuf(ovp, cred, length, (int)fs->e2fs_bsize);
if (error && (allerror == 0))
allerror = error;
vnode_pager_setsize(ovp, length);
/*
* Indirect blocks first.
*/
indir_lbn[SINGLE] = -NDADDR;
indir_lbn[DOUBLE] = indir_lbn[SINGLE] - NINDIR(fs) - 1;
indir_lbn[TRIPLE] = indir_lbn[DOUBLE] - NINDIR(fs) * NINDIR(fs) - 1;
for (level = TRIPLE; level >= SINGLE; level--) {
bn = oip->i_ib[level];
if (bn != 0) {
error = ext2_indirtrunc(oip, indir_lbn[level],
fsbtodb(fs, bn), lastiblock[level], level, &count);
if (error)
allerror = error;
blocksreleased += count;
if (lastiblock[level] < 0) {
oip->i_ib[level] = 0;
ext2_blkfree(oip, bn, fs->e2fs_fsize);
blocksreleased += nblocks;
}
}
if (lastiblock[level] >= 0)
goto done;
}
/*
* All whole direct blocks or frags.
*/
for (i = NDADDR - 1; i > lastblock; i--) {
long bsize;
bn = oip->i_db[i];
if (bn == 0)
continue;
oip->i_db[i] = 0;
bsize = blksize(fs, oip, i);
ext2_blkfree(oip, bn, bsize);
blocksreleased += btodb(bsize);
}
if (lastblock < 0)
goto done;
/*
* Finally, look for a change in size of the
* last direct block; release any frags.
*/
bn = oip->i_db[lastblock];
if (bn != 0) {
long oldspace, newspace;
/*
* Calculate amount of space we're giving
* back as old block size minus new block size.
*/
oldspace = blksize(fs, oip, lastblock);
oip->i_size = length;
newspace = blksize(fs, oip, lastblock);
if (newspace == 0)
panic("ext2_truncate: newspace");
if (oldspace - newspace > 0) {
/*
* Block number of space to be free'd is
* the old block # plus the number of frags
* required for the storage we're keeping.
*/
bn += numfrags(fs, newspace);
ext2_blkfree(oip, bn, oldspace - newspace);
blocksreleased += btodb(oldspace - newspace);
}
}
done:
#ifdef INVARIANTS
for (level = SINGLE; level <= TRIPLE; level++)
if (newblks[NDADDR + level] != oip->i_ib[level])
panic("itrunc1");
for (i = 0; i < NDADDR; i++)
if (newblks[i] != oip->i_db[i])
panic("itrunc2");
BO_LOCK(bo);
if (length == 0 && (bo->bo_dirty.bv_cnt != 0 ||
bo->bo_clean.bv_cnt != 0))
panic("itrunc3");
BO_UNLOCK(bo);
#endif /* INVARIANTS */
/*
* Put back the real size.
*/
oip->i_size = length;
if (oip->i_blocks >= blocksreleased)
oip->i_blocks -= blocksreleased;
else /* sanity */
oip->i_blocks = 0;
oip->i_flag |= IN_CHANGE;
vnode_pager_setsize(ovp, length);
return (allerror);
}
