/*
* This is called from our getattr.
*/
int ocfs2_inode_revalidate(struct dentry *dentry)
{
struct inode *inode = dentry->d_inode;
int status = 0;
mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
if (!inode) {
mlog(0, "eep, no inode!\n");
status = -ENOENT;
goto bail;
}
spin_lock(&OCFS2_I(inode)->ip_lock);
if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
spin_unlock(&OCFS2_I(inode)->ip_lock);
mlog(0, "inode deleted!\n");
status = -ENOENT;
goto bail;
}
spin_unlock(&OCFS2_I(inode)->ip_lock);
/* Let ocfs2_meta_lock do the work of updating our struct
* inode for us. */
status = ocfs2_meta_lock(inode, NULL, 0);
if (status < 0) {
if (status != -ENOENT)
mlog_errno(status);
goto bail;
}
ocfs2_meta_unlock(inode, 0);
bail:
mlog_exit(status);
return status;
}
/*
* ocfs2_readdir()
*
*/
int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
{
int error = 0;
unsigned long offset, blk;
int i, num, stored;
struct buffer_head * bh, * tmp;
struct ocfs2_dir_entry * de;
int err;
struct inode *inode = filp->f_dentry->d_inode;
struct super_block * sb = inode->i_sb;
int have_disk_lock = 0;
mlog_entry("dirino=%"MLFu64"\n", OCFS2_I(inode)->ip_blkno);
stored = 0;
bh = NULL;
error = ocfs2_meta_lock(inode, NULL, NULL, 0);
if (error < 0) {
if (error != -ENOENT)
mlog_errno(error);
/* we haven't got any yet, so propagate the error. */
stored = error;
goto bail;
}
have_disk_lock = 1;
offset = filp->f_pos & (sb->s_blocksize - 1);
while (!error && !stored && filp->f_pos < i_size_read(inode)) {
blk = (filp->f_pos) >> sb->s_blocksize_bits;
bh = ocfs2_bread(inode, blk, &err, 0);
if (!bh) {
mlog(ML_ERROR, "directory #%"MLFu64" contains a hole "
"at offset %lld\n",
OCFS2_I(inode)->ip_blkno,
filp->f_pos);
filp->f_pos += sb->s_blocksize - offset;
continue;
}
/*
* Do the readahead (8k)
*/
if (!offset) {
for (i = 16 >> (sb->s_blocksize_bits - 9), num = 0;
i > 0; i--) {
tmp = ocfs2_bread(inode, ++blk, &err, 1);
if (tmp)
brelse(tmp);
}
}
revalidate:
/* If the dir block has changed since the last call to
* readdir(2), then we might be pointing to an invalid
* dirent right now. Scan from the start of the block
* to make sure. */
if (filp->f_version != inode->i_version) {
for (i = 0; i < sb->s_blocksize && i < offset; ) {
de = (struct ocfs2_dir_entry *) (bh->b_data + i);
/* It's too expensive to do a full
* dirent test each time round this
* loop, but we do have to test at
* least that it is non-zero. A
* failure will be detected in the
* dirent test below. */
if (le16_to_cpu(de->rec_len) <
OCFS2_DIR_REC_LEN(1))
break;
i += le16_to_cpu(de->rec_len);
}
offset = i;
filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
| offset;
filp->f_version = inode->i_version;
}
while (!error && filp->f_pos < i_size_read(inode)
&& offset < sb->s_blocksize) {
de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
/* On error, skip the f_pos to the
next block. */
filp->f_pos = (filp->f_pos |
(sb->s_blocksize - 1)) + 1;
brelse(bh);
goto bail;
}
offset += le16_to_cpu(de->rec_len);
if (le64_to_cpu(de->inode)) {
/* We might block in the next section
* if the data destination is
* currently swapped out. So, use a
* version stamp to detect whether or
* not the directory has been modified
* during the copy operation.
*/
unsigned long version = filp->f_version;
unsigned char d_type = DT_UNKNOWN;
if (de->file_type < OCFS2_FT_MAX)
d_type = ocfs2_filetype_table[de->file_type];
error = filldir(dirent, de->name,
de->name_len,
filp->f_pos,
ino_from_blkno(sb, le64_to_cpu(de->inode)),
d_type);
if (error)
break;
if (version != filp->f_version)
goto revalidate;
stored ++;
}
filp->f_pos += le16_to_cpu(de->rec_len);
}
offset = 0;
brelse(bh);
}
stored = 0;
bail:
if (have_disk_lock)
ocfs2_meta_unlock(inode, 0);
mlog_exit(stored);
return stored;
}
void ocfs2_delete_inode(struct inode *inode)
{
int wipe, status;
sigset_t blocked, oldset;
struct buffer_head *di_bh = NULL;
mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
if (is_bad_inode(inode)) {
mlog(0, "Skipping delete of bad inode\n");
goto bail;
}
if (!ocfs2_inode_is_valid_to_delete(inode)) {
/* It's probably not necessary to truncate_inode_pages
* here but we do it for safety anyway (it will most
* likely be a no-op anyway) */
ocfs2_cleanup_delete_inode(inode, 0);
goto bail;
}
/* We want to block signals in delete_inode as the lock and
* messaging paths may return us -ERESTARTSYS. Which would
* cause us to exit early, resulting in inodes being orphaned
* forever. */
sigfillset(&blocked);
status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
if (status < 0) {
mlog_errno(status);
ocfs2_cleanup_delete_inode(inode, 1);
goto bail;
}
/* Lock down the inode. This gives us an up to date view of
* it's metadata (for verification), and allows us to
* serialize delete_inode votes.
*
* Even though we might be doing a truncate, we don't take the
* allocation lock here as it won't be needed - nobody will
* have the file open.
*/
status = ocfs2_meta_lock(inode, &di_bh, 1);
if (status < 0) {
if (status != -ENOENT)
mlog_errno(status);
ocfs2_cleanup_delete_inode(inode, 0);
goto bail_unblock;
}
/* Query the cluster. This will be the final decision made
* before we go ahead and wipe the inode. */
status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
if (!wipe || status < 0) {
/* Error and inode busy vote both mean we won't be
* removing the inode, so they take almost the same
* path. */
if (status < 0)
mlog_errno(status);
/* Someone in the cluster has voted to not wipe this
* inode, or it was never completely orphaned. Write
* out the pages and exit now. */
ocfs2_cleanup_delete_inode(inode, 1);
goto bail_unlock_inode;
}
ocfs2_cleanup_delete_inode(inode, 0);
status = ocfs2_wipe_inode(inode, di_bh);
if (status < 0) {
if (status != -EDEADLK)
mlog_errno(status);
goto bail_unlock_inode;
}
/*
* Mark the inode as successfully deleted.
*
* This is important for ocfs2_clear_inode() as it will check
* this flag and skip any checkpointing work
*
* ocfs2_stuff_meta_lvb() also uses this flag to invalidate
* the LVB for other nodes.
*/
OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
bail_unlock_inode:
ocfs2_meta_unlock(inode, 1);
brelse(di_bh);
bail_unblock:
status = sigprocmask(SIG_SETMASK, &oldset, NULL);
if (status < 0)
mlog_errno(status);
bail:
clear_inode(inode);
mlog_exit_void();
}
static int ocfs2_wipe_inode(struct inode *inode,
struct buffer_head *di_bh)
{
int status, orphaned_slot;
struct inode *orphan_dir_inode = NULL;
struct buffer_head *orphan_dir_bh = NULL;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct ocfs2_dinode *di;
di = (struct ocfs2_dinode *) di_bh->b_data;
orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
if (status)
return status;
orphan_dir_inode = ocfs2_get_system_file_inode(osb,
ORPHAN_DIR_SYSTEM_INODE,
orphaned_slot);
if (!orphan_dir_inode) {
status = -EEXIST;
mlog_errno(status);
goto bail;
}
/* Lock the orphan dir. The lock will be held for the entire
* delete_inode operation. We do this now to avoid races with
* recovery completion on other nodes. */
mutex_lock(&orphan_dir_inode->i_mutex);
status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
if (status < 0) {
mutex_unlock(&orphan_dir_inode->i_mutex);
mlog_errno(status);
goto bail;
}
/* we do this while holding the orphan dir lock because we
* don't want recovery being run from another node to vote for
* an inode delete on us -- this will result in two nodes
* truncating the same file! */
status = ocfs2_truncate_for_delete(osb, inode, di_bh);
if (status < 0) {
mlog_errno(status);
goto bail_unlock_dir;
}
status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
orphan_dir_bh);
if (status < 0)
mlog_errno(status);
bail_unlock_dir:
ocfs2_meta_unlock(orphan_dir_inode, 1);
mutex_unlock(&orphan_dir_inode->i_mutex);
brelse(orphan_dir_bh);
bail:
iput(orphan_dir_inode);
ocfs2_signal_wipe_completion(osb, orphaned_slot);
return status;
}
static int ocfs2_remove_inode(struct inode *inode,
struct buffer_head *di_bh,
struct inode *orphan_dir_inode,
struct buffer_head *orphan_dir_bh)
{
int status;
struct inode *inode_alloc_inode = NULL;
struct buffer_head *inode_alloc_bh = NULL;
handle_t *handle;
struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
inode_alloc_inode =
ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
le16_to_cpu(di->i_suballoc_slot));
if (!inode_alloc_inode) {
status = -EEXIST;
mlog_errno(status);
goto bail;
}
mutex_lock(&inode_alloc_inode->i_mutex);
status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1);
if (status < 0) {
mutex_unlock(&inode_alloc_inode->i_mutex);
mlog_errno(status);
goto bail;
}
handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
if (IS_ERR(handle)) {
status = PTR_ERR(handle);
mlog_errno(status);
goto bail_unlock;
}
status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
orphan_dir_bh);
if (status < 0) {
mlog_errno(status);
goto bail_commit;
}
/* set the inodes dtime */
status = ocfs2_journal_access(handle, inode, di_bh,
OCFS2_JOURNAL_ACCESS_WRITE);
if (status < 0) {
mlog_errno(status);
goto bail_commit;
}
di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
status = ocfs2_journal_dirty(handle, di_bh);
if (status < 0) {
mlog_errno(status);
goto bail_commit;
}
ocfs2_remove_from_cache(inode, di_bh);
status = ocfs2_free_dinode(handle, inode_alloc_inode,
inode_alloc_bh, di);
if (status < 0)
mlog_errno(status);
bail_commit:
ocfs2_commit_trans(osb, handle);
bail_unlock:
ocfs2_meta_unlock(inode_alloc_inode, 1);
mutex_unlock(&inode_alloc_inode->i_mutex);
brelse(inode_alloc_bh);
bail:
iput(inode_alloc_inode);
return status;
}
static int ocfs2_read_locked_inode(struct inode *inode,
struct ocfs2_find_inode_args *args)
{
struct super_block *sb;
struct ocfs2_super *osb;
struct ocfs2_dinode *fe;
struct buffer_head *bh = NULL;
int status, can_lock;
u32 generation = 0;
mlog_entry("(0x%p, 0x%p)\n", inode, args);
status = -EINVAL;
if (inode == NULL || inode->i_sb == NULL) {
mlog(ML_ERROR, "bad inode\n");
return status;
}
sb = inode->i_sb;
osb = OCFS2_SB(sb);
if (!args) {
mlog(ML_ERROR, "bad inode args\n");
make_bad_inode(inode);
return status;
}
/*
* To improve performance of cold-cache inode stats, we take
* the cluster lock here if possible.
*
* Generally, OCFS2 never trusts the contents of an inode
* unless it's holding a cluster lock, so taking it here isn't
* a correctness issue as much as it is a performance
* improvement.
*
* There are three times when taking the lock is not a good idea:
*
* 1) During startup, before we have initialized the DLM.
*
* 2) If we are reading certain system files which never get
* cluster locks (local alloc, truncate log).
*
* 3) If the process doing the iget() is responsible for
* orphan dir recovery. We're holding the orphan dir lock and
* can get into a deadlock with another process on another
* node in ->delete_inode().
*
* #1 and #2 can be simply solved by never taking the lock
* here for system files (which are the only type we read
* during mount). It's a heavier approach, but our main
* concern is user-accesible files anyway.
*
* #3 works itself out because we'll eventually take the
* cluster lock before trusting anything anyway.
*/
can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
&& !ocfs2_mount_local(osb);
/*
* To maintain backwards compatibility with older versions of
* ocfs2-tools, we still store the generation value for system
* files. The only ones that actually matter to userspace are
* the journals, but it's easier and inexpensive to just flag
* all system files similarly.
*/
if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
generation = osb->fs_generation;
ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
OCFS2_LOCK_TYPE_META,
generation, inode);
ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
OCFS2_LOCK_TYPE_OPEN,
0, inode);
if (can_lock) {
status = ocfs2_open_lock(inode);
if (status) {
make_bad_inode(inode);
mlog_errno(status);
return status;
}
status = ocfs2_meta_lock(inode, NULL, 0);
if (status) {
make_bad_inode(inode);
mlog_errno(status);
return status;
}
}
if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
status = ocfs2_try_open_lock(inode, 0);
if (status) {
make_bad_inode(inode);
return status;
}
}
status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
can_lock ? inode : NULL);
if (status < 0) {
mlog_errno(status);
goto bail;
}
status = -EINVAL;
fe = (struct ocfs2_dinode *) bh->b_data;
if (!OCFS2_IS_VALID_DINODE(fe)) {
mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
(unsigned long long)args->fi_blkno, 7,
fe->i_signature);
goto bail;
}
/*
* This is a code bug. Right now the caller needs to
* understand whether it is asking for a system file inode or
* not so the proper lock names can be built.
*/
mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
"Inode %llu: system file state is ambigous\n",
(unsigned long long)args->fi_blkno);
if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
S_ISBLK(le16_to_cpu(fe->i_mode)))
inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
if (ocfs2_populate_inode(inode, fe, 0) < 0)
goto bail;
BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
status = 0;
bail:
if (can_lock)
ocfs2_meta_unlock(inode, 0);
if (status < 0)
make_bad_inode(inode);
if (args && bh)
brelse(bh);
mlog_exit(status);
return status;
}
int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
{
int status = 0, size_change;
struct inode *inode = dentry->d_inode;
struct super_block *sb = inode->i_sb;
struct ocfs2_super *osb = OCFS2_SB(sb);
struct buffer_head *bh = NULL;
handle_t *handle = NULL;
mlog_entry("(0x%p, '%.*s')\n", dentry,
dentry->d_name.len, dentry->d_name.name);
if (attr->ia_valid & ATTR_MODE)
mlog(0, "mode change: %d\n", attr->ia_mode);
if (attr->ia_valid & ATTR_UID)
mlog(0, "uid change: %d\n", attr->ia_uid);
if (attr->ia_valid & ATTR_GID)
mlog(0, "gid change: %d\n", attr->ia_gid);
if (attr->ia_valid & ATTR_SIZE)
mlog(0, "size change...\n");
if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
mlog(0, "time change...\n");
#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
| ATTR_GID | ATTR_UID | ATTR_MODE)
if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
return 0;
}
status = inode_change_ok(inode, attr);
if (status)
return status;
size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
if (size_change) {
status = ocfs2_rw_lock(inode, 1);
if (status < 0) {
mlog_errno(status);
goto bail;
}
}
status = ocfs2_meta_lock(inode, &bh, 1);
if (status < 0) {
if (status != -ENOENT)
mlog_errno(status);
goto bail_unlock_rw;
}
if (size_change && attr->ia_size != i_size_read(inode)) {
if (i_size_read(inode) > attr->ia_size)
status = ocfs2_truncate_file(inode, bh, attr->ia_size);
else
status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
if (status < 0) {
if (status != -ENOSPC)
mlog_errno(status);
status = -ENOSPC;
goto bail_unlock;
}
}
handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
if (IS_ERR(handle)) {
status = PTR_ERR(handle);
mlog_errno(status);
goto bail_unlock;
}
status = inode_setattr(inode, attr);
if (status < 0) {
mlog_errno(status);
goto bail_commit;
}
status = ocfs2_mark_inode_dirty(handle, inode, bh);
if (status < 0)
mlog_errno(status);
bail_commit:
ocfs2_commit_trans(osb, handle);
bail_unlock:
ocfs2_meta_unlock(inode, 1);
bail_unlock_rw:
if (size_change)
ocfs2_rw_unlock(inode, 1);
bail:
if (bh)
brelse(bh);
mlog_exit(status);
return status;
}
static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
loff_t *ppos,
size_t count,
int appending,
int *direct_io)
{
int ret = 0, meta_level = appending;
struct inode *inode = dentry->d_inode;
u32 clusters;
loff_t newsize, saved_pos;
/*
* We sample i_size under a read level meta lock to see if our write
* is extending the file, if it is we back off and get a write level
* meta lock.
*/
for(;;) {
ret = ocfs2_meta_lock(inode, NULL, meta_level);
if (ret < 0) {
meta_level = -1;
mlog_errno(ret);
goto out;
}
/* Clear suid / sgid if necessary. We do this here
* instead of later in the write path because
* remove_suid() calls ->setattr without any hint that
* we may have already done our cluster locking. Since
* ocfs2_setattr() *must* take cluster locks to
* proceeed, this will lead us to recursively lock the
* inode. There's also the dinode i_size state which
* can be lost via setattr during extending writes (we
* set inode->i_size at the end of a write. */
if (should_remove_suid(dentry)) {
if (meta_level == 0) {
ocfs2_meta_unlock(inode, meta_level);
meta_level = 1;
continue;
}
ret = ocfs2_write_remove_suid(inode);
if (ret < 0) {
mlog_errno(ret);
goto out_unlock;
}
}
/* work on a copy of ppos until we're sure that we won't have
* to recalculate it due to relocking. */
if (appending) {
saved_pos = i_size_read(inode);
mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
} else {
saved_pos = *ppos;
}
if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
loff_t end = saved_pos + count;
/*
* Skip the O_DIRECT checks if we don't need
* them.
*/
if (!direct_io || !(*direct_io))
break;
/*
* Allowing concurrent direct writes means
* i_size changes wouldn't be synchronized, so
* one node could wind up truncating another
* nodes writes.
*/
if (end > i_size_read(inode)) {
*direct_io = 0;
break;
}
/*
* We don't fill holes during direct io, so
* check for them here. If any are found, the
* caller will have to retake some cluster
* locks and initiate the io as buffered.
*/
ret = ocfs2_check_range_for_holes(inode, saved_pos,
count);
if (ret == 1) {
*direct_io = 0;
ret = 0;
} else if (ret < 0)
mlog_errno(ret);
break;
}
/*
* The rest of this loop is concerned with legacy file
* systems which don't support sparse files.
*/
newsize = count + saved_pos;
mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
(long long) saved_pos, (long long) newsize,
(long long) i_size_read(inode));
/* No need for a higher level metadata lock if we're
* never going past i_size. */
if (newsize <= i_size_read(inode))
break;
if (meta_level == 0) {
ocfs2_meta_unlock(inode, meta_level);
meta_level = 1;
continue;
}
spin_lock(&OCFS2_I(inode)->ip_lock);
clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
OCFS2_I(inode)->ip_clusters;
spin_unlock(&OCFS2_I(inode)->ip_lock);
mlog(0, "Writing at EOF, may need more allocation: "
"i_size = %lld, newsize = %lld, need %u clusters\n",
(long long) i_size_read(inode), (long long) newsize,
clusters);
/* We only want to continue the rest of this loop if
* our extend will actually require more
* allocation. */
if (!clusters)
break;
ret = ocfs2_extend_file(inode, NULL, newsize, count);
if (ret < 0) {
if (ret != -ENOSPC)
mlog_errno(ret);
goto out_unlock;
}
break;
}
if (appending)
*ppos = saved_pos;
out_unlock:
ocfs2_meta_unlock(inode, meta_level);
out:
return ret;
}
/*
* ocfs2_readdir()
*
*/
int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
{
int error = 0;
unsigned long offset, blk, last_ra_blk = 0;
int i, stored;
struct buffer_head * bh, * tmp;
struct ocfs2_dir_entry * de;
int err;
struct inode *inode = filp->f_path.dentry->d_inode;
struct super_block * sb = inode->i_sb;
unsigned int ra_sectors = 16;
int lock_level = 0;
mlog_entry("dirino=%llu\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno);
stored = 0;
bh = NULL;
error = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
if (lock_level && error >= 0) {
/* We release EX lock which used to update atime
* and get PR lock again to reduce contention
* on commonly accessed directories. */
ocfs2_meta_unlock(inode, 1);
lock_level = 0;
error = ocfs2_meta_lock(inode, NULL, 0);
}
if (error < 0) {
if (error != -ENOENT)
mlog_errno(error);
/* we haven't got any yet, so propagate the error. */
stored = error;
goto bail_nolock;
}
offset = filp->f_pos & (sb->s_blocksize - 1);
while (!error && !stored && filp->f_pos < i_size_read(inode)) {
blk = (filp->f_pos) >> sb->s_blocksize_bits;
bh = ocfs2_bread(inode, blk, &err, 0);
if (!bh) {
mlog(ML_ERROR,
"directory #%llu contains a hole at offset %lld\n",
(unsigned long long)OCFS2_I(inode)->ip_blkno,
filp->f_pos);
filp->f_pos += sb->s_blocksize - offset;
continue;
}
/* The idea here is to begin with 8k read-ahead and to stay
* 4k ahead of our current position.
*
* TODO: Use the pagecache for this. We just need to
* make sure it's cluster-safe... */
if (!last_ra_blk
|| (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
i > 0; i--) {
tmp = ocfs2_bread(inode, ++blk, &err, 1);
if (tmp)
brelse(tmp);
}
last_ra_blk = blk;
ra_sectors = 8;
}
revalidate:
/* If the dir block has changed since the last call to
* readdir(2), then we might be pointing to an invalid
* dirent right now. Scan from the start of the block
* to make sure. */
if (filp->f_version != inode->i_version) {
for (i = 0; i < sb->s_blocksize && i < offset; ) {
de = (struct ocfs2_dir_entry *) (bh->b_data + i);
/* It's too expensive to do a full
* dirent test each time round this
* loop, but we do have to test at
* least that it is non-zero. A
* failure will be detected in the
* dirent test below. */
if (le16_to_cpu(de->rec_len) <
OCFS2_DIR_REC_LEN(1))
break;
i += le16_to_cpu(de->rec_len);
}
offset = i;
filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
| offset;
filp->f_version = inode->i_version;
}
while (!error && filp->f_pos < i_size_read(inode)
&& offset < sb->s_blocksize) {
de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
/* On error, skip the f_pos to the
next block. */
filp->f_pos = (filp->f_pos |
(sb->s_blocksize - 1)) + 1;
brelse(bh);
goto bail;
}
offset += le16_to_cpu(de->rec_len);
if (le64_to_cpu(de->inode)) {
/* We might block in the next section
* if the data destination is
* currently swapped out. So, use a
* version stamp to detect whether or
* not the directory has been modified
* during the copy operation.
*/
unsigned long version = filp->f_version;
unsigned char d_type = DT_UNKNOWN;
if (de->file_type < OCFS2_FT_MAX)
d_type = ocfs2_filetype_table[de->file_type];
error = filldir(dirent, de->name,
de->name_len,
filp->f_pos,
ino_from_blkno(sb, le64_to_cpu(de->inode)),
d_type);
if (error)
break;
if (version != filp->f_version)
goto revalidate;
stored ++;
}
filp->f_pos += le16_to_cpu(de->rec_len);
}
offset = 0;
brelse(bh);
}
stored = 0;
bail:
ocfs2_meta_unlock(inode, lock_level);
bail_nolock:
mlog_exit(stored);
return stored;
}
static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
struct ocfs2_alloc_context *ac,
int type,
u32 slot)
{
int status;
u32 bits_wanted = ac->ac_bits_wanted;
struct inode *alloc_inode;
struct buffer_head *bh = NULL;
struct ocfs2_dinode *fe;
u32 free_bits;
mlog_entry_void();
alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
if (!alloc_inode) {
mlog_errno(-EINVAL);
return -EINVAL;
}
mutex_lock(&alloc_inode->i_mutex);
status = ocfs2_meta_lock(alloc_inode, &bh, 1);
if (status < 0) {
mutex_unlock(&alloc_inode->i_mutex);
iput(alloc_inode);
mlog_errno(status);
return status;
}
ac->ac_inode = alloc_inode;
fe = (struct ocfs2_dinode *) bh->b_data;
if (!OCFS2_IS_VALID_DINODE(fe)) {
OCFS2_RO_ON_INVALID_DINODE(alloc_inode->i_sb, fe);
status = -EIO;
goto bail;
}
if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
(unsigned long long)le64_to_cpu(fe->i_blkno));
status = -EIO;
goto bail;
}
free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
le32_to_cpu(fe->id1.bitmap1.i_used);
if (bits_wanted > free_bits) {
/* cluster bitmap never grows */
if (ocfs2_is_cluster_bitmap(alloc_inode)) {
mlog(0, "Disk Full: wanted=%u, free_bits=%u\n",
bits_wanted, free_bits);
status = -ENOSPC;
goto bail;
}
status = ocfs2_block_group_alloc(osb, alloc_inode, bh);
if (status < 0) {
if (status != -ENOSPC)
mlog_errno(status);
goto bail;
}
atomic_inc(&osb->alloc_stats.bg_extends);
/* You should never ask for this much metadata */
BUG_ON(bits_wanted >
(le32_to_cpu(fe->id1.bitmap1.i_total)
- le32_to_cpu(fe->id1.bitmap1.i_used)));
}
get_bh(bh);
ac->ac_bh = bh;
bail:
if (bh)
brelse(bh);
mlog_exit(status);
return status;
}
/*
* Step 2: By now, we've completed the journal recovery, we've stamped
* a clean local alloc on disk and dropped the node out of the
* recovery map. Dlm locks will no longer stall, so lets clear out the
* main bitmap.
*/
int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
struct ocfs2_dinode *alloc)
{
int status;
struct ocfs2_journal_handle *handle = NULL;
struct buffer_head *main_bm_bh = NULL;
struct inode *main_bm_inode = NULL;
mlog_entry_void();
handle = ocfs2_alloc_handle(osb);
if (!handle) {
status = -ENOMEM;
mlog_errno(status);
goto bail;
}
main_bm_inode = ocfs2_get_system_file_inode(osb,
GLOBAL_BITMAP_SYSTEM_INODE,
OCFS2_INVALID_SLOT);
if (!main_bm_inode) {
status = -EINVAL;
mlog_errno(status);
goto bail;
}
ocfs2_handle_add_inode(handle, main_bm_inode);
status = ocfs2_meta_lock(main_bm_inode, handle, &main_bm_bh, 1);
if (status < 0) {
mlog_errno(status);
goto bail;
}
handle = ocfs2_start_trans(osb, handle, OCFS2_WINDOW_MOVE_CREDITS);
if (IS_ERR(handle)) {
status = PTR_ERR(handle);
handle = NULL;
mlog_errno(status);
goto bail;
}
/* we want the bitmap change to be recorded on disk asap */
ocfs2_handle_set_sync(handle, 1);
status = ocfs2_sync_local_to_main(osb, handle, alloc,
main_bm_inode, main_bm_bh);
if (status < 0)
mlog_errno(status);
bail:
if (handle)
ocfs2_commit_trans(handle);
if (main_bm_bh)
brelse(main_bm_bh);
if (main_bm_inode)
iput(main_bm_inode);
mlog_exit(status);
return status;
}
/*
* return any unused bits to the bitmap and write out a clean
* local_alloc.
*
* local_alloc_bh is optional. If not passed, we will simply use the
* one off osb. If you do pass it however, be warned that it *will* be
* returned brelse'd and NULL'd out.*/
void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
{
int status;
struct ocfs2_journal_handle *handle = NULL;
struct inode *local_alloc_inode = NULL;
struct buffer_head *bh = NULL;
struct buffer_head *main_bm_bh = NULL;
struct inode *main_bm_inode = NULL;
struct ocfs2_dinode *alloc_copy = NULL;
struct ocfs2_dinode *alloc = NULL;
mlog_entry_void();
if (osb->local_alloc_state == OCFS2_LA_UNUSED)
goto bail;
local_alloc_inode =
ocfs2_get_system_file_inode(osb,
LOCAL_ALLOC_SYSTEM_INODE,
osb->slot_num);
if (!local_alloc_inode) {
status = -ENOENT;
mlog_errno(status);
goto bail;
}
osb->local_alloc_state = OCFS2_LA_DISABLED;
handle = ocfs2_alloc_handle(osb);
if (!handle) {
status = -ENOMEM;
mlog_errno(status);
goto bail;
}
main_bm_inode = ocfs2_get_system_file_inode(osb,
GLOBAL_BITMAP_SYSTEM_INODE,
OCFS2_INVALID_SLOT);
if (!main_bm_inode) {
status = -EINVAL;
mlog_errno(status);
goto bail;
}
ocfs2_handle_add_inode(handle, main_bm_inode);
status = ocfs2_meta_lock(main_bm_inode, handle, &main_bm_bh, 1);
if (status < 0) {
mlog_errno(status);
goto bail;
}
/* WINDOW_MOVE_CREDITS is a bit heavy... */
handle = ocfs2_start_trans(osb, handle, OCFS2_WINDOW_MOVE_CREDITS);
if (IS_ERR(handle)) {
mlog_errno(PTR_ERR(handle));
handle = NULL;
goto bail;
}
bh = osb->local_alloc_bh;
alloc = (struct ocfs2_dinode *) bh->b_data;
alloc_copy = kmalloc(bh->b_size, GFP_KERNEL);
if (!alloc_copy) {
status = -ENOMEM;
goto bail;
}
memcpy(alloc_copy, alloc, bh->b_size);
status = ocfs2_journal_access(handle, local_alloc_inode, bh,
OCFS2_JOURNAL_ACCESS_WRITE);
if (status < 0) {
mlog_errno(status);
goto bail;
}
ocfs2_clear_local_alloc(alloc);
status = ocfs2_journal_dirty(handle, bh);
if (status < 0) {
mlog_errno(status);
goto bail;
}
brelse(bh);
osb->local_alloc_bh = NULL;
osb->local_alloc_state = OCFS2_LA_UNUSED;
status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
main_bm_inode, main_bm_bh);
if (status < 0)
mlog_errno(status);
bail:
if (handle)
ocfs2_commit_trans(handle);
if (main_bm_bh)
brelse(main_bm_bh);
if (main_bm_inode)
iput(main_bm_inode);
if (local_alloc_inode)
iput(local_alloc_inode);
if (alloc_copy)
kfree(alloc_copy);
mlog_exit_void();
}
