static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
const struct iovec *iov,
unsigned long nr_segs,
loff_t pos)
{
int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
struct file *filp = iocb->ki_filp;
struct inode *inode = filp->f_path.dentry->d_inode;
mlog_entry("(0x%p, %u, '%.*s')\n", filp,
(unsigned int)nr_segs,
filp->f_path.dentry->d_name.len,
filp->f_path.dentry->d_name.name);
if (!inode) {
ret = -EINVAL;
mlog_errno(ret);
goto bail;
}
/*
* buffered reads protect themselves in ->readpage(). O_DIRECT reads
* need locks to protect pending reads from racing with truncate.
*/
if (filp->f_flags & O_DIRECT) {
down_read(&inode->i_alloc_sem);
have_alloc_sem = 1;
ret = ocfs2_rw_lock(inode, 0);
if (ret < 0) {
mlog_errno(ret);
goto bail;
}
rw_level = 0;
/* communicate with ocfs2_dio_end_io */
ocfs2_iocb_set_rw_locked(iocb, rw_level);
}
/*
* We're fine letting folks race truncates and extending
* writes with read across the cluster, just like they can
* locally. Hence no rw_lock during read.
*
* Take and drop the meta data lock to update inode fields
* like i_size. This allows the checks down below
* generic_file_aio_read() a chance of actually working.
*/
ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
if (ret < 0) {
mlog_errno(ret);
goto bail;
}
ocfs2_meta_unlock(inode, lock_level);
ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
if (ret == -EINVAL)
mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
/* buffered aio wouldn't have proper lock coverage today */
BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
/* see ocfs2_file_aio_write */
if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
rw_level = -1;
have_alloc_sem = 0;
}
bail:
if (have_alloc_sem)
up_read(&inode->i_alloc_sem);
if (rw_level != -1)
ocfs2_rw_unlock(inode, rw_level);
mlog_exit(ret);
return ret;
}
/*
* ocfs2_dio_end_io is called by the dio core when a dio is finished. We're
* particularly interested in the aio/dio case. Like the core uses
* i_alloc_sem, we use the rw_lock DLM lock to protect io on one node from
* truncation on another.
*/
static void ocfs2_dio_end_io(struct kiocb *iocb,
loff_t offset,
ssize_t bytes,
void *private)
{
struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
int level;
/* this io's submitter should not have unlocked this before we could */
BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
ocfs2_iocb_clear_rw_locked(iocb);
level = ocfs2_iocb_rw_locked_level(iocb);
if (!level)
up_read(&inode->i_alloc_sem);
ocfs2_rw_unlock(inode, level);
}
/*
* ocfs2_invalidatepage() and ocfs2_releasepage() are shamelessly stolen
* from ext3. PageChecked() bits have been removed as OCFS2 does not
* do journalled data.
*/
static void ocfs2_invalidatepage(struct page *page, unsigned long offset)
static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
const struct iovec *iov,
unsigned long nr_segs,
loff_t pos)
{
int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
int can_do_direct, sync = 0;
ssize_t written = 0;
size_t ocount; /* original count */
size_t count; /* after file limit checks */
loff_t *ppos = &iocb->ki_pos;
struct file *file = iocb->ki_filp;
struct inode *inode = file->f_path.dentry->d_inode;
mlog_entry("(0x%p, %u, '%.*s')\n", file,
(unsigned int)nr_segs,
file->f_path.dentry->d_name.len,
file->f_path.dentry->d_name.name);
if (iocb->ki_left == 0)
return 0;
ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
if (ret)
return ret;
count = ocount;
vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
appending = file->f_flags & O_APPEND ? 1 : 0;
direct_io = file->f_flags & O_DIRECT ? 1 : 0;
mutex_lock(&inode->i_mutex);
relock:
/* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
if (direct_io) {
down_read(&inode->i_alloc_sem);
have_alloc_sem = 1;
}
/* concurrent O_DIRECT writes are allowed */
rw_level = !direct_io;
ret = ocfs2_rw_lock(inode, rw_level);
if (ret < 0) {
mlog_errno(ret);
goto out_sems;
}
can_do_direct = direct_io;
ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
iocb->ki_left, appending,
&can_do_direct);
if (ret < 0) {
mlog_errno(ret);
goto out;
}
/*
* We can't complete the direct I/O as requested, fall back to
* buffered I/O.
*/
if (direct_io && !can_do_direct) {
ocfs2_rw_unlock(inode, rw_level);
up_read(&inode->i_alloc_sem);
have_alloc_sem = 0;
rw_level = -1;
direct_io = 0;
sync = 1;
goto relock;
}
if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode)))
sync = 1;
/*
* XXX: Is it ok to execute these checks a second time?
*/
ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode));
if (ret)
goto out;
/*
* Set pos so that sync_page_range_nolock() below understands
* where to start from. We might've moved it around via the
* calls above. The range we want to actually sync starts from
* *ppos here.
*
*/
pos = *ppos;
/* communicate with ocfs2_dio_end_io */
ocfs2_iocb_set_rw_locked(iocb, rw_level);
if (direct_io) {
written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
ppos, count, ocount);
if (written < 0) {
ret = written;
goto out_dio;
}
} else {
written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs,
count, written);
if (written < 0) {
ret = written;
if (ret != -EFAULT || ret != -ENOSPC)
mlog_errno(ret);
goto out;
}
}
out_dio:
/* buffered aio wouldn't have proper lock coverage today */
BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
/*
* deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
* function pointer which is called when o_direct io completes so that
* it can unlock our rw lock. (it's the clustered equivalent of
* i_alloc_sem; protects truncate from racing with pending ios).
* Unfortunately there are error cases which call end_io and others
* that don't. so we don't have to unlock the rw_lock if either an
* async dio is going to do it in the future or an end_io after an
* error has already done it.
*/
if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
rw_level = -1;
have_alloc_sem = 0;
}
out:
if (rw_level != -1)
ocfs2_rw_unlock(inode, rw_level);
out_sems:
if (have_alloc_sem)
up_read(&inode->i_alloc_sem);
if (written > 0 && sync) {
ssize_t err;
err = sync_page_range_nolock(inode, file->f_mapping, pos, count);
if (err < 0)
written = err;
}
mutex_unlock(&inode->i_mutex);
mlog_exit(ret);
return written ? written : ret;
}
