/*
* Allocate a particular inode with a particular version number, freeing
* any previous versions of this inode that may have gone before.
* Used by the roll-forward code.
*
* XXX this function does not have appropriate locking to be used on a live fs;
* XXX but something similar could probably be used for an "undelete" call.
*
* Called with the Ifile inode locked.
*/
int
lfs_rf_valloc(struct lfs *fs, ino_t ino, int vers, struct lwp *l,
struct vnode **vpp)
{
IFILE *ifp;
struct buf *bp, *cbp;
struct vnode *vp;
struct inode *ip;
ino_t tino, oldnext;
int error;
CLEANERINFO *cip;
ASSERT_SEGLOCK(fs); /* XXX it doesn't, really */
/*
* First, just try a vget. If the version number is the one we want,
* we don't have to do anything else. If the version number is wrong,
* take appropriate action.
*/
error = VFS_VGET(fs->lfs_ivnode->v_mount, ino, &vp);
if (error == 0) {
DLOG((DLOG_RF, "lfs_rf_valloc[1]: ino %d vp %p\n", ino, vp));
*vpp = vp;
ip = VTOI(vp);
if (ip->i_gen == vers)
return 0;
else if (ip->i_gen < vers) {
lfs_truncate(vp, (off_t)0, 0, NOCRED);
ip->i_gen = ip->i_ffs1_gen = vers;
LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
return 0;
} else {
DLOG((DLOG_RF, "ino %d: sought version %d, got %d\n",
ino, vers, ip->i_ffs1_gen));
vput(vp);
*vpp = NULLVP;
return EEXIST;
}
}
/*
* The inode is not in use. Find it on the free list.
*/
/* If the Ifile is too short to contain this inum, extend it */
while (VTOI(fs->lfs_ivnode)->i_size <= (ino /
fs->lfs_ifpb + fs->lfs_cleansz + fs->lfs_segtabsz)
<< fs->lfs_bshift) {
lfs_extend_ifile(fs, NOCRED);
}
LFS_IENTRY(ifp, fs, ino, bp);
oldnext = ifp->if_nextfree;
ifp->if_version = vers;
brelse(bp, 0);
LFS_GET_HEADFREE(fs, cip, cbp, &ino);
if (ino) {
LFS_PUT_HEADFREE(fs, cip, cbp, oldnext);
} else {
tino = ino;
while (1) {
LFS_IENTRY(ifp, fs, tino, bp);
if (ifp->if_nextfree == ino ||
ifp->if_nextfree == LFS_UNUSED_INUM)
break;
tino = ifp->if_nextfree;
brelse(bp, 0);
}
if (ifp->if_nextfree == LFS_UNUSED_INUM) {
brelse(bp, 0);
return ENOENT;
}
ifp->if_nextfree = oldnext;
LFS_BWRITE_LOG(bp);
}
error = lfs_ialloc(fs, fs->lfs_ivnode, ino, vers, &vp);
if (error == 0) {
/*
* Make it VREG so we can put blocks on it. We will change
* this later if it turns out to be some other kind of file.
*/
ip = VTOI(vp);
ip->i_mode = ip->i_ffs1_mode = IFREG;
ip->i_nlink = ip->i_ffs1_nlink = 1;
ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p, &vp);
ip = VTOI(vp);
DLOG((DLOG_RF, "lfs_rf_valloc: ino %d vp %p\n", ino, vp));
/* The dirop-nature of this vnode is past */
lfs_unmark_vnode(vp);
(void)lfs_vunref(vp);
vp->v_uflag &= ~VU_DIROP;
mutex_enter(&lfs_lock);
--lfs_dirvcount;
--fs->lfs_dirvcount;
TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain);
wakeup(&lfs_dirvcount);
wakeup(&fs->lfs_dirvcount);
mutex_exit(&lfs_lock);
}
*vpp = vp;
return error;
}
int
lfs_markv(struct proc *p, fsid_t *fsidp, BLOCK_INFO *blkiov,
int blkcnt)
{
BLOCK_INFO *blkp;
IFILE *ifp;
struct buf *bp;
struct inode *ip = NULL;
struct lfs *fs;
struct mount *mntp;
struct vnode *vp = NULL;
ino_t lastino;
daddr_t b_daddr, v_daddr;
int cnt, error;
int do_again = 0;
int numrefed = 0;
ino_t maxino;
size_t obsize;
/* number of blocks/inodes that we have already bwrite'ed */
int nblkwritten, ninowritten;
if ((mntp = vfs_getvfs(fsidp)) == NULL)
return (ENOENT);
fs = VFSTOULFS(mntp)->um_lfs;
if (fs->lfs_ronly)
return EROFS;
maxino = (lfs_fragstoblks(fs, VTOI(fs->lfs_ivnode)->i_ffs1_blocks) -
fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb;
cnt = blkcnt;
if ((error = vfs_busy(mntp, NULL)) != 0)
return (error);
/*
* This seglock is just to prevent the fact that we might have to sleep
* from allowing the possibility that our blocks might become
* invalid.
*
* It is also important to note here that unless we specify SEGM_CKP,
* any Ifile blocks that we might be asked to clean will never get
* to the disk.
*/
lfs_seglock(fs, SEGM_CLEAN | SEGM_CKP | SEGM_SYNC);
/* Mark blocks/inodes dirty. */
error = 0;
/* these were inside the initialization for the for loop */
v_daddr = LFS_UNUSED_DADDR;
lastino = LFS_UNUSED_INUM;
nblkwritten = ninowritten = 0;
for (blkp = blkiov; cnt--; ++blkp)
{
/* Bounds-check incoming data, avoid panic for failed VGET */
if (blkp->bi_inode <= 0 || blkp->bi_inode >= maxino) {
error = EINVAL;
goto err3;
}
/*
* Get the IFILE entry (only once) and see if the file still
* exists.
*/
if (lastino != blkp->bi_inode) {
/*
* Finish the old file, if there was one. The presence
* of a usable vnode in vp is signaled by a valid v_daddr.
*/
if (v_daddr != LFS_UNUSED_DADDR) {
lfs_vunref(vp);
numrefed--;
}
/*
* Start a new file
*/
lastino = blkp->bi_inode;
if (blkp->bi_inode == LFS_IFILE_INUM)
v_daddr = fs->lfs_idaddr;
else {
LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
/* XXX fix for force write */
v_daddr = ifp->if_daddr;
brelse(bp, 0);
}
if (v_daddr == LFS_UNUSED_DADDR)
continue;
/* Get the vnode/inode. */
error = lfs_fastvget(mntp, blkp->bi_inode, v_daddr,
&vp,
(blkp->bi_lbn == LFS_UNUSED_LBN
? blkp->bi_bp
: NULL));
if (!error) {
numrefed++;
}
if (error) {
DLOG((DLOG_CLEAN, "lfs_markv: lfs_fastvget"
" failed with %d (ino %d, segment %d)\n",
error, blkp->bi_inode,
lfs_dtosn(fs, blkp->bi_daddr)));
/*
* If we got EAGAIN, that means that the
* Inode was locked. This is
* recoverable: just clean the rest of
* this segment, and let the cleaner try
* again with another. (When the
* cleaner runs again, this segment will
* sort high on the list, since it is
* now almost entirely empty.) But, we
* still set v_daddr = LFS_UNUSED_ADDR
* so as not to test this over and over
* again.
*/
if (error == EAGAIN) {
error = 0;
do_again++;
}
#ifdef DIAGNOSTIC
else if (error != ENOENT)
panic("lfs_markv VFS_VGET FAILED");
#endif
/* lastino = LFS_UNUSED_INUM; */
v_daddr = LFS_UNUSED_DADDR;
vp = NULL;
ip = NULL;
continue;
}
ip = VTOI(vp);
ninowritten++;
} else if (v_daddr == LFS_UNUSED_DADDR) {
/*
* This can only happen if the vnode is dead (or
* in any case we can't get it...e.g., it is
* inlocked). Keep going.
*/
continue;
}
/* Past this point we are guaranteed that vp, ip are valid. */
/* Can't clean VU_DIROP directories in case of truncation */
/* XXX - maybe we should mark removed dirs specially? */
if (vp->v_type == VDIR && (vp->v_uflag & VU_DIROP)) {
do_again++;
continue;
}
/* If this BLOCK_INFO didn't contain a block, keep going. */
if (blkp->bi_lbn == LFS_UNUSED_LBN) {
/* XXX need to make sure that the inode gets written in this case */
/* XXX but only write the inode if it's the right one */
if (blkp->bi_inode != LFS_IFILE_INUM) {
LFS_IENTRY(ifp, fs, blkp->bi_inode, bp);
if (ifp->if_daddr == blkp->bi_daddr) {
mutex_enter(&lfs_lock);
LFS_SET_UINO(ip, IN_CLEANING);
mutex_exit(&lfs_lock);
}
brelse(bp, 0);
}
continue;
}
b_daddr = 0;
if (VOP_BMAP(vp, blkp->bi_lbn, NULL, &b_daddr, NULL) ||
LFS_DBTOFSB(fs, b_daddr) != blkp->bi_daddr)
{
if (lfs_dtosn(fs, LFS_DBTOFSB(fs, b_daddr)) ==
lfs_dtosn(fs, blkp->bi_daddr))
{
DLOG((DLOG_CLEAN, "lfs_markv: wrong da same seg: %llx vs %llx\n",
(long long)blkp->bi_daddr, (long long)LFS_DBTOFSB(fs, b_daddr)));
}
do_again++;
continue;
}
/*
* Check block sizes. The blocks being cleaned come from
* disk, so they should have the same size as their on-disk
* counterparts.
*/
if (blkp->bi_lbn >= 0)
obsize = lfs_blksize(fs, ip, blkp->bi_lbn);
else
obsize = fs->lfs_bsize;
/* Check for fragment size change */
if (blkp->bi_lbn >= 0 && blkp->bi_lbn < ULFS_NDADDR) {
obsize = ip->i_lfs_fragsize[blkp->bi_lbn];
}
if (obsize != blkp->bi_size) {
DLOG((DLOG_CLEAN, "lfs_markv: ino %d lbn %lld wrong"
" size (%ld != %d), try again\n",
blkp->bi_inode, (long long)blkp->bi_lbn,
(long) obsize, blkp->bi_size));
do_again++;
continue;
}
/*
* If we get to here, then we are keeping the block. If
* it is an indirect block, we want to actually put it
* in the buffer cache so that it can be updated in the
* finish_meta section. If it's not, we need to
* allocate a fake buffer so that writeseg can perform
* the copyin and write the buffer.
*/
if (ip->i_number != LFS_IFILE_INUM && blkp->bi_lbn >= 0) {
/* Data Block */
bp = lfs_fakebuf(fs, vp, blkp->bi_lbn,
blkp->bi_size, blkp->bi_bp);
/* Pretend we used bread() to get it */
bp->b_blkno = LFS_FSBTODB(fs, blkp->bi_daddr);
} else {
/* Indirect block or ifile */
if (blkp->bi_size != fs->lfs_bsize &&
ip->i_number != LFS_IFILE_INUM)
panic("lfs_markv: partial indirect block?"
" size=%d\n", blkp->bi_size);
bp = getblk(vp, blkp->bi_lbn, blkp->bi_size, 0, 0);
if (!(bp->b_oflags & (BO_DONE|BO_DELWRI))) {
/*
* The block in question was not found
* in the cache; i.e., the block that
* getblk() returned is empty. So, we
* can (and should) copy in the
* contents, because we've already
* determined that this was the right
* version of this block on disk.
*
* And, it can't have changed underneath
* us, because we have the segment lock.
*/
error = copyin(blkp->bi_bp, bp->b_data, blkp->bi_size);
if (error)
goto err2;
}
}
if ((error = lfs_bwrite_ext(bp, BW_CLEAN)) != 0)
goto err2;
nblkwritten++;
/*
* XXX should account indirect blocks and ifile pages as well
*/
if (nblkwritten + lfs_lblkno(fs, ninowritten * sizeof (struct ulfs1_dinode))
> LFS_MARKV_MAX_BLOCKS) {
DLOG((DLOG_CLEAN, "lfs_markv: writing %d blks %d inos\n",
nblkwritten, ninowritten));
lfs_segwrite(mntp, SEGM_CLEAN);
nblkwritten = ninowritten = 0;
}
}
/*
* Finish the old file, if there was one
*/
if (v_daddr != LFS_UNUSED_DADDR) {
lfs_vunref(vp);
numrefed--;
}
#ifdef DIAGNOSTIC
if (numrefed != 0)
panic("lfs_markv: numrefed=%d", numrefed);
#endif
DLOG((DLOG_CLEAN, "lfs_markv: writing %d blks %d inos (check point)\n",
nblkwritten, ninowritten));
/*
* The last write has to be SEGM_SYNC, because of calling semantics.
* It also has to be SEGM_CKP, because otherwise we could write
* over the newly cleaned data contained in a checkpoint, and then
* we'd be unhappy at recovery time.
*/
lfs_segwrite(mntp, SEGM_CLEAN | SEGM_CKP | SEGM_SYNC);
lfs_segunlock(fs);
vfs_unbusy(mntp, false, NULL);
if (error)
return (error);
else if (do_again)
return EAGAIN;
return 0;
err2:
DLOG((DLOG_CLEAN, "lfs_markv err2\n"));
/*
* XXX we're here because copyin() failed.
* XXX it means that we can't trust the cleanerd. too bad.
* XXX how can we recover from this?
*/
err3:
/*
* XXX should do segwrite here anyway?
*/
if (v_daddr != LFS_UNUSED_DADDR) {
lfs_vunref(vp);
--numrefed;
}
lfs_segunlock(fs);
vfs_unbusy(mntp, false, NULL);
#ifdef DIAGNOSTIC
if (numrefed != 0)
panic("lfs_markv: numrefed=%d", numrefed);
#endif
return (error);
}
static int
update_inoblk(struct lfs *fs, daddr_t offset, kauth_cred_t cred,
struct lwp *l)
{
struct vnode *devvp, *vp;
struct inode *ip;
struct ufs1_dinode *dip;
struct buf *dbp, *ibp;
int error;
daddr_t daddr;
IFILE *ifp;
SEGUSE *sup;
devvp = VTOI(fs->lfs_ivnode)->i_devvp;
/*
* Get the inode, update times and perms.
* DO NOT update disk blocks, we do that separately.
*/
error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize,
cred, 0, &dbp);
if (error) {
DLOG((DLOG_RF, "update_inoblk: bread returned %d\n", error));
return error;
}
dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs);
while (--dip >= (struct ufs1_dinode *)dbp->b_data) {
if (dip->di_inumber > LFS_IFILE_INUM) {
error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
l, &vp);
if (error) {
DLOG((DLOG_RF, "update_inoblk: lfs_rf_valloc"
" returned %d\n", error));
continue;
}
ip = VTOI(vp);
if (dip->di_size != ip->i_size)
lfs_truncate(vp, dip->di_size, 0, NOCRED);
/* Get mode, link count, size, and times */
memcpy(ip->i_din.ffs1_din, dip,
offsetof(struct ufs1_dinode, di_db[0]));
/* Then the rest, except di_blocks */
ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
ip->i_mode = ip->i_ffs1_mode;
ip->i_nlink = ip->i_ffs1_nlink;
ip->i_size = ip->i_ffs1_size;
LFS_SET_UINO(ip, IN_CHANGE | IN_UPDATE);
/* Re-initialize to get type right */
ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
&vp);
vput(vp);
/* Record change in location */
LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
daddr = ifp->if_daddr;
ifp->if_daddr = dbtofsb(fs, dbp->b_blkno);
error = LFS_BWRITE_LOG(ibp); /* Ifile */
/* And do segment accounting */
if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) {
if (daddr > 0) {
LFS_SEGENTRY(sup, fs, dtosn(fs, daddr),
ibp);
sup->su_nbytes -= sizeof (struct ufs1_dinode);
LFS_WRITESEGENTRY(sup, fs,
dtosn(fs, daddr),
ibp);
}
LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
ibp);
sup->su_nbytes += sizeof (struct ufs1_dinode);
LFS_WRITESEGENTRY(sup, fs,
dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
ibp);
}
}
}
void
lfs_itimes(struct inode *ip, const struct timespec *acc,
const struct timespec *mod, const struct timespec *cre)
{
#ifdef _KERNEL
struct timespec now;
KASSERT(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY));
vfs_timestamp(&now);
#endif
if (ip->i_flag & IN_ACCESS) {
#ifdef _KERNEL
if (acc == NULL)
acc = &now;
#endif
ip->i_ffs1_atime = acc->tv_sec;
ip->i_ffs1_atimensec = acc->tv_nsec;
if (ip->i_lfs->lfs_is64 || lfs_sb_getversion(ip->i_lfs) > 1) {
struct lfs *fs = ip->i_lfs;
struct buf *ibp;
IFILE *ifp;
LFS_IENTRY(ifp, ip->i_lfs, ip->i_number, ibp);
lfs_if_setatime_sec(fs, ifp, acc->tv_sec);
lfs_if_setatime_nsec(fs, ifp, acc->tv_nsec);
LFS_BWRITE_LOG(ibp);
mutex_enter(&lfs_lock);
fs->lfs_flags |= LFS_IFDIRTY;
mutex_exit(&lfs_lock);
} else {
mutex_enter(&lfs_lock);
LFS_SET_UINO(ip, IN_ACCESSED);
mutex_exit(&lfs_lock);
}
}
if (ip->i_flag & (IN_CHANGE | IN_UPDATE | IN_MODIFY)) {
if (ip->i_flag & (IN_UPDATE | IN_MODIFY)) {
#ifdef _KERNEL
if (mod == NULL)
mod = &now;
#endif
ip->i_ffs1_mtime = mod->tv_sec;
ip->i_ffs1_mtimensec = mod->tv_nsec;
ip->i_modrev++;
}
if (ip->i_flag & (IN_CHANGE | IN_MODIFY)) {
#ifdef _KERNEL
if (cre == NULL)
cre = &now;
#endif
ip->i_ffs1_ctime = cre->tv_sec;
ip->i_ffs1_ctimensec = cre->tv_nsec;
}
mutex_enter(&lfs_lock);
if (ip->i_flag & (IN_CHANGE | IN_UPDATE))
LFS_SET_UINO(ip, IN_MODIFIED);
if (ip->i_flag & IN_MODIFY)
LFS_SET_UINO(ip, IN_ACCESSED);
mutex_exit(&lfs_lock);
}
ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY);
}