/*
* NAME: dbUpdatePMap()
*
* FUNCTION: update the allocation state (free or allocate) of the
* specified block range in the persistent block allocation map.
*
* the blocks will be updated in the persistent map one
* dmap at a time.
*
* PARAMETERS:
* ipbmap - pointer to in-core inode for the block map.
* free - 'true' if block range is to be freed from the persistent
* map; 'false' if it is to be allocated.
* blkno - starting block number of the range.
* nblocks - number of contiguous blocks in the range.
* tblk - transaction block;
*
* RETURN VALUES:
* 0 - success
* -EIO - i/o error
*/
int
dbUpdatePMap(struct inode *ipbmap,
int free, s64 blkno, s64 nblocks, struct tblock * tblk)
{
int nblks, dbitno, wbitno, rbits;
int word, nbits, nwords;
struct bmap *bmp = JFS_SBI(ipbmap->i_sb)->bmap;
s64 lblkno, rem, lastlblkno;
u32 mask;
struct dmap *dp;
struct metapage *mp;
struct jfs_log *log;
int lsn, difft, diffp;
unsigned long flags;
/* the blocks better be within the mapsize. */
if (blkno + nblocks > bmp->db_mapsize) {
printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
(unsigned long long) blkno,
(unsigned long long) nblocks);
jfs_error(ipbmap->i_sb,
"dbUpdatePMap: blocks are outside the map");
return -EIO;
}
/* compute delta of transaction lsn from log syncpt */
lsn = tblk->lsn;
log = (struct jfs_log *) JFS_SBI(tblk->sb)->log;
logdiff(difft, lsn, log);
/*
* update the block state a dmap at a time.
*/
mp = NULL;
lastlblkno = 0;
for (rem = nblocks; rem > 0; rem -= nblks, blkno += nblks) {
/* get the buffer for the current dmap. */
lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
if (lblkno != lastlblkno) {
if (mp) {
write_metapage(mp);
}
mp = read_metapage(bmp->db_ipbmap, lblkno, PSIZE,
0);
if (mp == NULL)
return -EIO;
metapage_wait_for_io(mp);
}
dp = (struct dmap *) mp->data;
/* determine the bit number and word within the dmap of
* the starting block. also determine how many blocks
* are to be updated within this dmap.
*/
dbitno = blkno & (BPERDMAP - 1);
word = dbitno >> L2DBWORD;
nblks = min(rem, (s64)BPERDMAP - dbitno);
/* update the bits of the dmap words. the first and last
* words may only have a subset of their bits updated. if
* this is the case, we'll work against that word (i.e.
* partial first and/or last) only in a single pass. a
* single pass will also be used to update all words that
* are to have all their bits updated.
*/
for (rbits = nblks; rbits > 0;
rbits -= nbits, dbitno += nbits) {
/* determine the bit number within the word and
* the number of bits within the word.
*/
wbitno = dbitno & (DBWORD - 1);
nbits = min(rbits, DBWORD - wbitno);
/* check if only part of the word is to be updated. */
if (nbits < DBWORD) {
/* update (free or allocate) the bits
* in this word.
*/
mask =
(ONES << (DBWORD - nbits) >> wbitno);
if (free)
dp->pmap[word] &=
cpu_to_le32(~mask);
else
dp->pmap[word] |=
cpu_to_le32(mask);
word += 1;
} else {
/* one or more words are to have all
* their bits updated. determine how
* many words and how many bits.
*/
nwords = rbits >> L2DBWORD;
nbits = nwords << L2DBWORD;
/* update (free or allocate) the bits
* in these words.
*/
if (free)
memset(&dp->pmap[word], 0,
nwords * 4);
else
memset(&dp->pmap[word], (int) ONES,
nwords * 4);
word += nwords;
}
}
/*
* NAME: jfs_rename
*
* FUNCTION: rename a file or directory
*/
static int jfs_rename(struct inode *old_dir, struct dentry *old_dentry,
struct inode *new_dir, struct dentry *new_dentry)
{
struct btstack btstack;
ino_t ino;
struct component_name new_dname;
struct inode *new_ip;
struct component_name old_dname;
struct inode *old_ip;
int rc;
tid_t tid;
struct tlock *tlck;
struct dt_lock *dtlck;
struct lv *lv;
int ipcount;
struct inode *iplist[4];
struct tblock *tblk;
s64 new_size = 0;
int commit_flag;
jfs_info("jfs_rename: %s %s", old_dentry->d_name.name,
new_dentry->d_name.name);
dquot_initialize(old_dir);
dquot_initialize(new_dir);
old_ip = old_dentry->d_inode;
new_ip = new_dentry->d_inode;
if ((rc = get_UCSname(&old_dname, old_dentry)))
goto out1;
if ((rc = get_UCSname(&new_dname, new_dentry)))
goto out2;
/*
* Make sure source inode number is what we think it is
*/
rc = dtSearch(old_dir, &old_dname, &ino, &btstack, JFS_LOOKUP);
if (rc || (ino != old_ip->i_ino)) {
rc = -ENOENT;
goto out3;
}
/*
* Make sure dest inode number (if any) is what we think it is
*/
rc = dtSearch(new_dir, &new_dname, &ino, &btstack, JFS_LOOKUP);
if (!rc) {
if ((!new_ip) || (ino != new_ip->i_ino)) {
rc = -ESTALE;
goto out3;
}
} else if (rc != -ENOENT)
goto out3;
else if (new_ip) {
/* no entry exists, but one was expected */
rc = -ESTALE;
goto out3;
}
if (S_ISDIR(old_ip->i_mode)) {
if (new_ip) {
if (!dtEmpty(new_ip)) {
rc = -ENOTEMPTY;
goto out3;
}
} else if ((new_dir != old_dir) &&
(new_dir->i_nlink == JFS_LINK_MAX)) {
rc = -EMLINK;
goto out3;
}
} else if (new_ip) {
IWRITE_LOCK(new_ip, RDWRLOCK_NORMAL);
/* Init inode for quota operations. */
dquot_initialize(new_ip);
}
/*
* The real work starts here
*/
tid = txBegin(new_dir->i_sb, 0);
/*
* How do we know the locking is safe from deadlocks?
* The vfs does the hard part for us. Any time we are taking nested
* commit_mutexes, the vfs already has i_mutex held on the parent.
* Here, the vfs has already taken i_mutex on both old_dir and new_dir.
*/
mutex_lock_nested(&JFS_IP(new_dir)->commit_mutex, COMMIT_MUTEX_PARENT);
mutex_lock_nested(&JFS_IP(old_ip)->commit_mutex, COMMIT_MUTEX_CHILD);
if (old_dir != new_dir)
mutex_lock_nested(&JFS_IP(old_dir)->commit_mutex,
COMMIT_MUTEX_SECOND_PARENT);
if (new_ip) {
mutex_lock_nested(&JFS_IP(new_ip)->commit_mutex,
COMMIT_MUTEX_VICTIM);
/*
* Change existing directory entry to new inode number
*/
ino = new_ip->i_ino;
rc = dtModify(tid, new_dir, &new_dname, &ino,
old_ip->i_ino, JFS_RENAME);
if (rc)
goto out4;
drop_nlink(new_ip);
if (S_ISDIR(new_ip->i_mode)) {
drop_nlink(new_ip);
if (new_ip->i_nlink) {
mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
if (old_dir != new_dir)
mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
if (!S_ISDIR(old_ip->i_mode) && new_ip)
IWRITE_UNLOCK(new_ip);
jfs_error(new_ip->i_sb,
"jfs_rename: new_ip->i_nlink != 0");
return -EIO;
}
tblk = tid_to_tblock(tid);
tblk->xflag |= COMMIT_DELETE;
tblk->u.ip = new_ip;
} else if (new_ip->i_nlink == 0) {
assert(!test_cflag(COMMIT_Nolink, new_ip));
/* free block resources */
if ((new_size = commitZeroLink(tid, new_ip)) < 0) {
txAbort(tid, 1); /* Marks FS Dirty */
rc = new_size;
goto out4;
}
tblk = tid_to_tblock(tid);
tblk->xflag |= COMMIT_DELETE;
tblk->u.ip = new_ip;
} else {
new_ip->i_ctime = CURRENT_TIME;
mark_inode_dirty(new_ip);
}
} else {
/*
* Add new directory entry
*/
rc = dtSearch(new_dir, &new_dname, &ino, &btstack,
JFS_CREATE);
if (rc) {
jfs_err("jfs_rename didn't expect dtSearch to fail "
"w/rc = %d", rc);
goto out4;
}
ino = old_ip->i_ino;
rc = dtInsert(tid, new_dir, &new_dname, &ino, &btstack);
if (rc) {
if (rc == -EIO)
jfs_err("jfs_rename: dtInsert returned -EIO");
goto out4;
}
if (S_ISDIR(old_ip->i_mode))
inc_nlink(new_dir);
}
/*
* Remove old directory entry
*/
ino = old_ip->i_ino;
rc = dtDelete(tid, old_dir, &old_dname, &ino, JFS_REMOVE);
if (rc) {
jfs_err("jfs_rename did not expect dtDelete to return rc = %d",
rc);
txAbort(tid, 1); /* Marks Filesystem dirty */
goto out4;
}
if (S_ISDIR(old_ip->i_mode)) {
drop_nlink(old_dir);
if (old_dir != new_dir) {
/*
* Change inode number of parent for moved directory
*/
JFS_IP(old_ip)->i_dtroot.header.idotdot =
cpu_to_le32(new_dir->i_ino);
/* Linelock header of dtree */
tlck = txLock(tid, old_ip,
(struct metapage *) &JFS_IP(old_ip)->bxflag,
tlckDTREE | tlckBTROOT | tlckRELINK);
dtlck = (struct dt_lock *) & tlck->lock;
ASSERT(dtlck->index == 0);
lv = & dtlck->lv[0];
lv->offset = 0;
lv->length = 1;
dtlck->index++;
}
}
/*
* Update ctime on changed/moved inodes & mark dirty
*/
old_ip->i_ctime = CURRENT_TIME;
mark_inode_dirty(old_ip);
new_dir->i_ctime = new_dir->i_mtime = current_fs_time(new_dir->i_sb);
mark_inode_dirty(new_dir);
/* Build list of inodes modified by this transaction */
ipcount = 0;
iplist[ipcount++] = old_ip;
if (new_ip)
iplist[ipcount++] = new_ip;
iplist[ipcount++] = old_dir;
if (old_dir != new_dir) {
iplist[ipcount++] = new_dir;
old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
mark_inode_dirty(old_dir);
}
/*
* Incomplete truncate of file data can
* result in timing problems unless we synchronously commit the
* transaction.
*/
if (new_size)
commit_flag = COMMIT_SYNC;
else
commit_flag = 0;
rc = txCommit(tid, ipcount, iplist, commit_flag);
out4:
txEnd(tid);
if (new_ip)
mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
if (old_dir != new_dir)
mutex_unlock(&JFS_IP(old_dir)->commit_mutex);
mutex_unlock(&JFS_IP(old_ip)->commit_mutex);
mutex_unlock(&JFS_IP(new_dir)->commit_mutex);
while (new_size && (rc == 0)) {
tid = txBegin(new_ip->i_sb, 0);
mutex_lock(&JFS_IP(new_ip)->commit_mutex);
new_size = xtTruncate_pmap(tid, new_ip, new_size);
if (new_size < 0) {
txAbort(tid, 1);
rc = new_size;
} else
rc = txCommit(tid, 1, &new_ip, COMMIT_SYNC);
txEnd(tid);
mutex_unlock(&JFS_IP(new_ip)->commit_mutex);
}
if (new_ip && (new_ip->i_nlink == 0))
set_cflag(COMMIT_Nolink, new_ip);
out3:
free_UCSname(&new_dname);
out2:
free_UCSname(&old_dname);
out1:
if (new_ip && !S_ISDIR(new_ip->i_mode))
IWRITE_UNLOCK(new_ip);
/*
* Truncating the directory index table is not guaranteed. It
* may need to be done iteratively
*/
if (test_cflag(COMMIT_Stale, old_dir)) {
if (old_dir->i_size > 1)
jfs_truncate_nolock(old_dir, 0);
clear_cflag(COMMIT_Stale, old_dir);
}
jfs_info("jfs_rename: returning %d", rc);
return rc;
}
/*
* NAME: dbFree()
*
* FUNCTION: free the specified block range from the working block
* allocation map.
*
* the blocks will be free from the working map one dmap
* at a time.
*
* PARAMETERS:
* ip - pointer to in-core inode;
* blkno - starting block number to be freed.
* nblocks - number of blocks to be freed.
*
* RETURN VALUES:
* 0 - success
* -EIO - i/o error
*/
int dbFree(struct inode *ip, s64 blkno, s64 nblocks)
{
struct metapage *mp;
struct dmap *dp;
int nb, rc;
s64 lblkno, rem;
struct inode *ipbmap = JFS_SBI(ip->i_sb)->ipbmap;
struct bmap *bmp = JFS_SBI(ip->i_sb)->bmap;
struct super_block *sb = ipbmap->i_sb;
IREAD_LOCK(ipbmap, RDWRLOCK_DMAP);
/* block to be freed better be within the mapsize. */
if (unlikely((blkno == 0) || (blkno + nblocks > bmp->db_mapsize))) {
IREAD_UNLOCK(ipbmap);
printk(KERN_ERR "blkno = %Lx, nblocks = %Lx\n",
(unsigned long long) blkno,
(unsigned long long) nblocks);
jfs_error(ip->i_sb,
"dbFree: block to be freed is outside the map");
return -EIO;
}
/**
* TRIM the blocks, when mounted with discard option
*/
if (JFS_SBI(sb)->flag & JFS_DISCARD)
if (JFS_SBI(sb)->minblks_trim <= nblocks)
jfs_issue_discard(ipbmap, blkno, nblocks);
/*
* free the blocks a dmap at a time.
*/
mp = NULL;
for (rem = nblocks; rem > 0; rem -= nb, blkno += nb) {
/* release previous dmap if any */
if (mp) {
write_metapage(mp);
}
/* get the buffer for the current dmap. */
lblkno = BLKTODMAP(blkno, bmp->db_l2nbperpage);
mp = read_metapage(ipbmap, lblkno, PSIZE, 0);
if (mp == NULL) {
IREAD_UNLOCK(ipbmap);
return -EIO;
}
dp = (struct dmap *) mp->data;
/* determine the number of blocks to be freed from
* this dmap.
*/
nb = min(rem, BPERDMAP - (blkno & (BPERDMAP - 1)));
/* free the blocks. */
if ((rc = dbFreeDmap(bmp, dp, blkno, nb))) {
jfs_error(ip->i_sb, "dbFree: error in block map\n");
release_metapage(mp);
IREAD_UNLOCK(ipbmap);
return (rc);
}
}
/* write the last buffer. */
write_metapage(mp);
IREAD_UNLOCK(ipbmap);
return (0);
}
