Re: [PATCH 1/2] ceph: Implement readv/preadv for sync operation.

["Yan, Zheng" <zheng.z.yan@intel.com>]

On 09/05/2013 08:28 AM, majianpeng wrote:
>> On 09/03/2013 04:52 PM, majianpeng wrote:
>>> For readv/preadv sync-operatoin, ceph only do the first iov.
>>> It don't think other iovs.Now implement this.
>>>
>>> Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
>>> ---
>>>  fs/ceph/file.c | 175 ++++++++++++++++++++++++++++++++++++++++-----------------
>>>  1 file changed, 123 insertions(+), 52 deletions(-)
>>>
>>> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
>>> index 3de8982..7d6a3ee 100644
>>> --- a/fs/ceph/file.c
>>> +++ b/fs/ceph/file.c
>>> @@ -408,51 +408,95 @@ more:
>>>   *
>>>   * If the read spans object boundary, just do multiple reads.
>>>   */
>>> -static ssize_t ceph_sync_read(struct file *file, char __user *data,
>>> -			      unsigned len, loff_t *poff, int *checkeof)
>>> +static ssize_t ceph_sync_read(struct kiocb *iocb, struct iovec *iov,
>>> +			      unsigned long nr_segs, int *checkeof)
>>>  {
>>> +	struct file *file = iocb->ki_filp;
>>>  	struct inode *inode = file_inode(file);
>>>  	struct page **pages;
>>> -	u64 off = *poff;
>>> -	int num_pages, ret;
>>> +	u64 off = iocb->ki_pos;
>>> +	int num_pages, ret, i;
>>>  
>>> -	dout("sync_read on file %p %llu~%u %s\n", file, off, len,
>>> +	dout("sync_read on file %p %llu~%u %s\n", file, off,
>>> +	     (unsigned)iocb->ki_left,
>>>  	     (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
>>> -
>>> -	if (file->f_flags & O_DIRECT) {
>>> -		num_pages = calc_pages_for((unsigned long)data, len);
>>> -		pages = ceph_get_direct_page_vector(data, num_pages, true);
>>> -	} else {
>>> -		num_pages = calc_pages_for(off, len);
>>> -		pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
>>> -	}
>>> -	if (IS_ERR(pages))
>>> -		return PTR_ERR(pages);
>>> -
>>>  	/*
>>>  	 * flush any page cache pages in this range.  this
>>>  	 * will make concurrent normal and sync io slow,
>>>  	 * but it will at least behave sensibly when they are
>>>  	 * in sequence.
>>>  	 */
>>> -	ret = filemap_write_and_wait(inode->i_mapping);
>>> +	ret = filemap_write_and_wait_range(inode->i_mapping, off,
>>> +						off + iocb->ki_left);
>>>  	if (ret < 0)
>>> -		goto done;
>>> -
>>> -	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
>>> -			   file->f_flags & O_DIRECT,
>>> -			   (unsigned long)data & ~PAGE_MASK);
>>> +		return ret;
>>>  
>>> -	if (ret >= 0 && (file->f_flags & O_DIRECT) == 0)
>>> -		ret = ceph_copy_page_vector_to_user(pages, data, off, ret);
>>> -	if (ret >= 0)
>>> -		*poff = off + ret;
>>> +	if (file->f_flags & O_DIRECT) {
>>> +		for (i = 0; i < nr_segs; i++) {
>>> +			void __user *data = iov[i].iov_base;
>>> +			size_t len = iov[i].iov_len;
>>> +
>>> +			num_pages = calc_pages_for((unsigned long)data, len);
>>> +			pages = ceph_get_direct_page_vector(data,
>>> +							    num_pages, true);
>>> +			if (IS_ERR(pages))
>>> +				return PTR_ERR(pages);
>>> +
>>> +			ret = striped_read(inode, off, len,
>>> +					   pages, num_pages, checkeof,
>>> +					   1, (unsigned long)data & ~PAGE_MASK);
>>> +			ceph_put_page_vector(pages, num_pages, true);
>>> +
>>> +			if (ret <= 0)
>>> +				break;
>>> +			off += ret;
>>> +			if (ret < len)
>>> +				break;
>>> +		}
>>> +		if (off > iocb->ki_pos) {
>>> +			ret = off - iocb->ki_pos;
>>> +			iocb->ki_pos = off;
>>> +			iocb->ki_left -= ret;
>>> +		}
>>> +	} else {
>>> +		size_t len = iocb->ki_left;
>>>  
>>> -done:
>>> -	if (file->f_flags & O_DIRECT)
>>> -		ceph_put_page_vector(pages, num_pages, true);
>>> -	else
>>> +		num_pages = calc_pages_for(off, len);
>>> +		pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
>>> +		if (IS_ERR(pages))
>>> +			return PTR_ERR(pages);
>>> +		ret = striped_read(inode, off, len, pages,
>>> +					num_pages, checkeof, 0, 0);
>>> +		len = ret;
>>> +		if (len) {
>>> +			int i, l, k = 0;
>>> +			size_t left = len;
>>> +
>>> +			for (i = 0; i < nr_segs && left; i++) {
>>> +				void __user *data = iov[i].iov_base;
>>> +				l = min(left, iov[i].iov_len);
>>> +				ret = ceph_copy_page_vector_to_user(&pages[k],
>>> +								    data, off,
>>> +								    l);
>>> +				if (ret > 0) {
>>> +					left -= ret;
>>> +					off += ret;
>>> +					k = calc_pages_for(iocb->ki_pos,
>>> +							   len - left + 1) - 1;
>>> +					BUG_ON(k >= num_pages && left);
>>> +				} else
>>> +					break;
>>> +			}
>>> +
>>> +			if (left == 0) {
>>> +				iocb->ki_pos += len;
>>> +				iocb->ki_left -= len;
>>> +				ret = len;
>>> +			}
>>
>> If user program provides invalid buffer to readv syscall. 'left' can be larger than 0.
>> For this case, we should return size of data that were successfully copied
> Ok, we should add this code, before "if (left == 0)"
> left = len - left;
> 
>>
>>> +		}
>>>  		ceph_release_page_vector(pages, num_pages);
>>> +	}
>>> +
>>>  	dout("sync_read result %d\n", ret);
>>>  	return ret;
>>>  }
>>> @@ -647,55 +691,82 @@ static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov,
>>>  {
>>>  	struct file *filp = iocb->ki_filp;
>>>  	struct ceph_file_info *fi = filp->private_data;
>>> -	loff_t *ppos = &iocb->ki_pos;
>>> -	size_t len = iov->iov_len;
>>> +	size_t len = iocb->ki_left;
>>>  	struct inode *inode = file_inode(filp);
>>>  	struct ceph_inode_info *ci = ceph_inode(inode);
>>> -	void __user *base = iov->iov_base;
>>>  	ssize_t ret;
>>>  	int want, got = 0;
>>>  	int checkeof = 0, read = 0;
>>>  
>>> +
>>>  	dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
>>>  	     inode, ceph_vinop(inode), pos, (unsigned)len, inode);
>>> -again:
>>> +
>>>  	if (fi->fmode & CEPH_FILE_MODE_LAZY)
>>>  		want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
>>>  	else
>>>  		want = CEPH_CAP_FILE_CACHE;
>>>  	ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
>>>  	if (ret < 0)
>>> -		goto out;
>>> +		return ret;
>>> +
>>>  	dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
>>>  	     inode, ceph_vinop(inode), pos, (unsigned)len,
>>>  	     ceph_cap_string(got));
>>>  
>>>  	if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
>>>  	    (iocb->ki_filp->f_flags & O_DIRECT) ||
>>> -	    (fi->flags & CEPH_F_SYNC))
>>> +	    (fi->flags & CEPH_F_SYNC)) {
>>> +		unsigned long curr_seg = 0;
>>> +		struct iovec *iov_clone;
>>> +
>>> +		iov_clone = kmalloc(nr_segs * sizeof(struct iovec), GFP_KERNEL);
>>> +		if (iov_clone == NULL) {
>>> +			ret = -ENOMEM;
>>> +			goto out;
>>> +		}
>>> +		memcpy(iov_clone, iov, nr_segs * sizeof(struct iovec));
>>
>> why do we need to clone the iov ?
>>
> I noticed in patch2,you asked the same question.
> The reason is i need modify iov,but the iov is const.So i only clone it.
> Or should you have better method?
> 

use iov iterator. see generic_file_buffered_write()

Regards
Yan, Zheng

>> Regards
>> Yan, Zheng
>>
>>> +again:
>>>  		/* hmm, this isn't really async... */
>>> -		ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
>>> -	else
>>> -		ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
>>> +		ret = ceph_sync_read(iocb, &iov_clone[curr_seg],
>>> +					nr_segs - curr_seg, &checkeof);
>>> +
>>> +		if (checkeof && ret >= 0) {
>>> +			int statret = ceph_do_getattr(inode,
>>> +						      CEPH_STAT_CAP_SIZE);
>>> +
>>> +			/* hit EOF or hole? */
>>> +			if (statret == 0 && iocb->ki_pos < inode->i_size &&
>>> +				iocb->ki_left) {
>>> +				size_t tmp = 0;
>>> +				dout("%s sync_read hit hole, ppos %lld < size %lld"
>>> +				     ", reading more\n", __func__, iocb->ki_pos,
>>> +				     inode->i_size);
>>> +
>>> +				read += ret;
>>> +				for (; curr_seg < nr_segs; curr_seg++) {
>>> +					if ((tmp + iov_clone[curr_seg].iov_len)
>>> +						> ret)
>>> +						break;
>>> +					tmp += iov_clone[curr_seg].iov_len;
>>> +				}
>>> +
>>> +				BUG_ON(curr_seg == nr_segs);
>>> +				iov_clone[curr_seg].iov_base += ret - tmp;
>>> +				iov_clone[curr_seg].iov_len -= ret - tmp;
>>> +				checkeof = 0;
>>> +				goto again;
>>> +			}
>>> +		}
>>> +		kfree(iov_clone);
>>>  
>>> +	} else
>>> +		ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
>>>  out:
>>>  	dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
>>>  	     inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
>>>  	ceph_put_cap_refs(ci, got);
>>>  
>>> -	if (checkeof && ret >= 0) {
>>> -		int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
>>> -
>>> -		/* hit EOF or hole? */
>>> -		if (statret == 0 && *ppos < inode->i_size) {
>>> -			dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos, inode->i_size);
>>> -			read += ret;
>>> -			base += ret;
>>> -			len -= ret;
>>> -			checkeof = 0;
>>> -			goto again;
>>> -		}
>>> -	}
>>>  	if (ret >= 0)
>>>  		ret += read;
>>>  
>>>
>>
>