#include <linux/highmem.h>
#include <linux/pagemap.h>
#include <linux/uio.h>
+#include <linux/sched.h>
#define MLOG_MASK_PREFIX ML_INODE
#include <cluster/masklog.h>
#include "file.h"
#include "sysfile.h"
#include "inode.h"
+#include "ioctl.h"
#include "journal.h"
#include "mmap.h"
#include "suballoc.h"
/* Some parts of this taken from generic_cont_expand, which turned out
* to be too fragile to do exactly what we need without us having to
- * worry about recursive locking in ->commit_write(). */
+ * worry about recursive locking in ->prepare_write() and
+ * ->commit_write(). */
static int ocfs2_write_zero_page(struct inode *inode,
u64 size)
{
goto out;
}
- ret = ocfs2_prepare_write(NULL, page, offset, offset);
+ ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
if (ret < 0) {
mlog_errno(ret);
goto out_unlock;
}
start_off += sb->s_blocksize;
+
+ /*
+ * Very large extends have the potential to lock up
+ * the cpu for extended periods of time.
+ */
+ cond_resched();
}
out:
return ret;
}
+/*
+ * A tail_to_skip value > 0 indicates that we're being called from
+ * ocfs2_file_aio_write(). This has the following implications:
+ *
+ * - we don't want to update i_size
+ * - di_bh will be NULL, which is fine because it's only used in the
+ * case where we want to update i_size.
+ * - ocfs2_zero_extend() will then only be filling the hole created
+ * between i_size and the start of the write.
+ */
static int ocfs2_extend_file(struct inode *inode,
struct buffer_head *di_bh,
- u64 new_i_size)
+ u64 new_i_size,
+ size_t tail_to_skip)
{
int ret = 0;
u32 clusters_to_add;
+ BUG_ON(!tail_to_skip && !di_bh);
+
/* setattr sometimes calls us like this. */
if (new_i_size == 0)
goto out;
clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
OCFS2_I(inode)->ip_clusters;
+ /*
+ * protect the pages that ocfs2_zero_extend is going to be
+ * pulling into the page cache.. we do this before the
+ * metadata extend so that we don't get into the situation
+ * where we've extended the metadata but can't get the data
+ * lock to zero.
+ */
+ ret = ocfs2_data_lock(inode, 1);
+ if (ret < 0) {
+ mlog_errno(ret);
+ goto out;
+ }
+
if (clusters_to_add) {
ret = ocfs2_extend_allocation(inode, clusters_to_add);
if (ret < 0) {
mlog_errno(ret);
- goto out;
- }
-
- ret = ocfs2_zero_extend(inode, new_i_size);
- if (ret < 0) {
- mlog_errno(ret);
- goto out;
+ goto out_unlock;
}
- }
+ }
- /* No allocation required, we just use this helper to
- * do a trivial update of i_size. */
- ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
+ /*
+ * Call this even if we don't add any clusters to the tree. We
+ * still need to zero the area between the old i_size and the
+ * new i_size.
+ */
+ ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
if (ret < 0) {
mlog_errno(ret);
- goto out;
+ goto out_unlock;
+ }
+
+ if (!tail_to_skip) {
+ /* We're being called from ocfs2_setattr() which wants
+ * us to update i_size */
+ ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
+ if (ret < 0)
+ mlog_errno(ret);
}
+out_unlock:
+ ocfs2_data_unlock(inode, 1);
+
out:
return ret;
}
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);
+ status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
if (status < 0) {
if (status != -ENOSPC)
mlog_errno(status);
}
static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
- const char __user *buf,
- size_t count,
+ const struct iovec *iov,
+ unsigned long nr_segs,
loff_t pos)
{
- struct iovec local_iov = { .iov_base = (void __user *)buf,
- .iov_len = count };
int ret, rw_level = -1, meta_level = -1, have_alloc_sem = 0;
u32 clusters;
struct file *filp = iocb->ki_filp;
struct inode *inode = filp->f_dentry->d_inode;
loff_t newsize, saved_pos;
- mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp, buf,
- (unsigned int)count,
+ mlog_entry("(0x%p, %u, '%.*s')\n", filp,
+ (unsigned int)nr_segs,
filp->f_dentry->d_name.len,
filp->f_dentry->d_name.name);
/* happy write of zero bytes */
- if (count == 0)
+ if (iocb->ki_left == 0)
return 0;
if (!inode) {
} else {
saved_pos = iocb->ki_pos;
}
- newsize = count + saved_pos;
+ newsize = iocb->ki_left + saved_pos;
mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
(long long) saved_pos, (long long) newsize,
if (!clusters)
break;
- ret = ocfs2_extend_allocation(inode, clusters);
+ ret = ocfs2_extend_file(inode, NULL, newsize, iocb->ki_left);
if (ret < 0) {
if (ret != -ENOSPC)
mlog_errno(ret);
goto out;
}
-
- /* Fill any holes which would've been created by this
- * write. If we're O_APPEND, this will wind up
- * (correctly) being a noop. */
- ret = ocfs2_zero_extend(inode, (u64) newsize - count);
- if (ret < 0) {
- mlog_errno(ret);
- goto out;
- }
break;
}
/* communicate with ocfs2_dio_end_io */
ocfs2_iocb_set_rw_locked(iocb);
- ret = generic_file_aio_write_nolock(iocb, &local_iov, 1, &iocb->ki_pos);
+ ret = generic_file_aio_write_nolock(iocb, iov, nr_segs, iocb->ki_pos);
/* buffered aio wouldn't have proper lock coverage today */
BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
}
static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
- char __user *buf,
- size_t count,
+ const struct iovec *iov,
+ unsigned long nr_segs,
loff_t pos)
{
int ret = 0, rw_level = -1, have_alloc_sem = 0;
struct file *filp = iocb->ki_filp;
struct inode *inode = filp->f_dentry->d_inode;
- mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp, buf,
- (unsigned int)count,
+ mlog_entry("(0x%p, %u, '%.*s')\n", filp,
+ (unsigned int)nr_segs,
filp->f_dentry->d_name.len,
filp->f_dentry->d_name.name);
ocfs2_iocb_set_rw_locked(iocb);
}
- ret = generic_file_aio_read(iocb, buf, count, iocb->ki_pos);
+ /*
+ * 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(inode, NULL, NULL, 0);
+ if (ret < 0) {
+ mlog_errno(ret);
+ goto bail;
+ }
+ ocfs2_meta_unlock(inode, 0);
+
+ 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");
.open = ocfs2_file_open,
.aio_read = ocfs2_file_aio_read,
.aio_write = ocfs2_file_aio_write,
+ .ioctl = ocfs2_ioctl,
};
const struct file_operations ocfs2_dops = {
.read = generic_read_dir,
.readdir = ocfs2_readdir,
.fsync = ocfs2_sync_file,
+ .ioctl = ocfs2_ioctl,
};