Skip to contentSkip to conversation

ceph: Implement readv/preadv for sync operation.

PATCH Superseded majianpeng 2 participants 5 messages Sep 6, 2013 – Sep 9, 2013

Subsystems (2)

v2 of 7 Superseded Latest is v7 →

lore
Share & tools
lore

majianpeng

author
#0 lore raw

PATCH v2 1/2 ceph: Implement readv/preadv for sync operation.

For readv/preadv sync-operatoin, ceph only do the first iov. It don't think other iovs.Now implement this.

V2:
-add generic_segment_checks
-using struct iov_iter replace cloning the iovs.
-return previous successfully copied if ceph_copy_page_vector_to_user
met error.

---
 fs/ceph/file.c | 174 ++++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 123 insertions(+), 51 deletions(-)
fs/ceph/file.c +123 -51
index 3de8982..1c28c52 100644
@@ -408,51 +408,109 @@ 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 iov_iter *i,
+				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, n;
 
-	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;
+		return ret;
 
-	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
-			   file->f_flags & O_DIRECT,
-			   (unsigned long)data & ~PAGE_MASK);
+	if (file->f_flags & O_DIRECT) {
+		for (n = 0; n < i->nr_segs; n++) {
+			void __user *data = i->iov[n].iov_base;
+			size_t len = i->iov[n].iov_len;
+
+			if (n == 0) {
+				len -=  i->iov_offset;
+				data += i->iov_offset;
+			}
+
+			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 (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 (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 l, k = 0;
+			size_t left = len;
+
+			for (n = 0; n < i->nr_segs && left; n++) {
+				void __user *data = i->iov[n].iov_base;
+				l = min(left, i->iov[n].iov_len);
+
+				if (n == 0) {
+					data += i->iov_offset;
+					l = min(i->iov[0].iov_len - i->iov_offset,
+						left);
+				}
+
+				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;
+			}
+
+			len -= left;
+			if (len > 0) {
+				iocb->ki_pos += len;
+				iocb->ki_left -= len;
+				ret = len;
+			}
+		}
 		ceph_release_page_vector(pages, num_pages);
+	}
+
 	dout("sync_read result %d\n", ret);
 	return ret;
 }
@@ -647,55 +705,69 @@ 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 = 0;
 	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:
+
+	ret = generic_segment_checks(iov, &nr_segs, &len, VERIFY_WRITE);
+	if (ret)
+		return ret;
+
 	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)) {
+		struct iov_iter i;
+
+		iocb->ki_left = len;
+		iov_iter_init(&i, iov, nr_segs, len, 0);
+again:
 		/* hmm, this isn't really async... */
-		ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
-	else
+		ret = ceph_sync_read(iocb, &i, &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) {
+				dout("sync_read hit hole, ppos %lld < size %lld"
+				     ", reading more\n", iocb->ki_pos,
+				     inode->i_size);
+
+				read += ret;
+				iov_iter_advance(&i, ret);
+				checkeof = 0;
+				goto again;
+			}
+		}
+
+	} 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;
 
Signature · 1 line
1.8.1.2
#1 lore raw
Quoted 133 lines · majianpeng

For readv/preadv sync-operatoin, ceph only do the first iov. It don't think other iovs.Now implement this.

V2:
-add generic_segment_checks
-using struct iov_iter replace cloning the iovs.
-return previous successfully copied if ceph_copy_page_vector_to_user
met error.

---
 fs/ceph/file.c | 174 ++++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 123 insertions(+), 51 deletions(-)
fs/ceph/file.c +67 -29
index 3de8982..1c28c52 100644
@@ -408,51 +408,109 @@ 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 iov_iter *i,
+				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, n;
 
-	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;
+		return ret;
 
-	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
-			   file->f_flags & O_DIRECT,
-			   (unsigned long)data & ~PAGE_MASK);
+	if (file->f_flags & O_DIRECT) {
+		for (n = 0; n < i->nr_segs; n++) {
+			void __user *data = i->iov[n].iov_base;
+			size_t len = i->iov[n].iov_len;
+
+			if (n == 0) {
+				len -=  i->iov_offset;
+				data += i->iov_offset;
+			}
+
+			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 (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 (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 l, k = 0;
+			size_t left = len;
+
+			for (n = 0; n < i->nr_segs && left; n++) {
+				void __user *data = i->iov[n].iov_base;
+				l = min(left, i->iov[n].iov_len);
+
+				if (n == 0) {
+					data += i->iov_offset;
+					l = min(i->iov[0].iov_len - i->iov_offset,
+						left);
+				}
+
+				ret = ceph_copy_page_vector_to_user(&pages[k],
+								    data, off,

fs/ceph/file.c more: · line 502

Quoted 10 lines
fs/ceph/file.c +10 -0
+								    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;
+			}

I think it's better to call iov_iter_advance() here instead of in ceph_aio_read(), and change the code to something like:
----
while (iov_iter_count(&i) && left > 0) {
void __user *data = i->iov->iov_base + i->iov_offset;
l = min(left, i->iov->iov_len - i->iov_offset);

ret = ceph_copy_page_vector_to_user(&pages[k], data, off, l);

   if (ret > 0) {
     iov_iter_advance(&i, ret);
     left -= ret;
     off += ret;
     k = calc_pages_for(iocb->ki_pos, len - left + 1) - 1;
     BUG_ON(k >= num_pages && left);
   } else
     break;
}

rest change looks good.

Regards
Yan, Zheng

Quoted 107 lines
fs/ceph/file.c +46 -22
+
+			len -= left;
+			if (len > 0) {
+				iocb->ki_pos += len;
+				iocb->ki_left -= len;
+				ret = len;
+			}
+		}
 		ceph_release_page_vector(pages, num_pages);
+	}
+
 	dout("sync_read result %d\n", ret);
 	return ret;
 }
@@ -647,55 +705,69 @@ 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 = 0;
 	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:
+
+	ret = generic_segment_checks(iov, &nr_segs, &len, VERIFY_WRITE);
+	if (ret)
+		return ret;
+
 	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)) {
+		struct iov_iter i;
+
+		iocb->ki_left = len;
+		iov_iter_init(&i, iov, nr_segs, len, 0);
+again:
 		/* hmm, this isn't really async... */
-		ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
-	else
+		ret = ceph_sync_read(iocb, &i, &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) {
+				dout("sync_read hit hole, ppos %lld < size %lld"
+				     ", reading more\n", iocb->ki_pos,
+				     inode->i_size);
+
+				read += ret;
+				iov_iter_advance(&i, ret);
+				checkeof = 0;
+				goto again;
+			}
+		}
+
+	} 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;
 
author
#2 lore raw
Quoted 164 lines · majianpeng
Quoted 133 lines

For readv/preadv sync-operatoin, ceph only do the first iov. It don't think other iovs.Now implement this.

V2:
-add generic_segment_checks
-using struct iov_iter replace cloning the iovs.
-return previous successfully copied if ceph_copy_page_vector_to_user
met error.

---
 fs/ceph/file.c | 174 ++++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 123 insertions(+), 51 deletions(-)
fs/ceph/file.c +67 -29
index 3de8982..1c28c52 100644
@@ -408,51 +408,109 @@ 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 iov_iter *i,
+				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, n;
 
-	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;
+		return ret;
 
-	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
-			   file->f_flags & O_DIRECT,
-			   (unsigned long)data & ~PAGE_MASK);
+	if (file->f_flags & O_DIRECT) {
+		for (n = 0; n < i->nr_segs; n++) {
+			void __user *data = i->iov[n].iov_base;
+			size_t len = i->iov[n].iov_len;
+
+			if (n == 0) {
+				len -=  i->iov_offset;
+				data += i->iov_offset;
+			}
+
+			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 (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 (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 l, k = 0;
+			size_t left = len;
+
+			for (n = 0; n < i->nr_segs && left; n++) {
+				void __user *data = i->iov[n].iov_base;
+				l = min(left, i->iov[n].iov_len);
+
+				if (n == 0) {
+					data += i->iov_offset;
+					l = min(i->iov[0].iov_len - i->iov_offset,
+						left);
+				}
+
+				ret = ceph_copy_page_vector_to_user(&pages[k],
+								    data, off,

fs/ceph/file.c more: · line 502

Quoted 10 lines
fs/ceph/file.c +10 -0
+								    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;
+			}

I think it's better to call iov_iter_advance() here instead of in ceph_aio_read(), and change the code to something like:
----
while (iov_iter_count(&i) && left > 0) {
void __user *data = i->iov->iov_base + i->iov_offset;
l = min(left, i->iov->iov_len - i->iov_offset);

ret = ceph_copy_page_vector_to_user(&pages[k], data, off, l);

  if (ret > 0) {
    iov_iter_advance(&i, ret);
    left -= ret;
    off += ret;
    k = calc_pages_for(iocb->ki_pos, len - left + 1) - 1;
    BUG_ON(k >= num_pages && left);
  } else
    break;
}
Quoted 5 lines

rest change looks good.

Regards
Yan, Zheng

Thanks!
Jianpeng Ma

Subject:ceph: Implement readv/preadv for sync operation.

For readv/preadv sync-operatoin, ceph only do the first iov. It don't think other iovs.Now implement this.

V2:
-add generic_segment_checks
-using struct iov_iter replace cloning the iovs.
-return previous successfully copied if ceph_copy_page_vector_to_user
met error.

---
 fs/ceph/file.c | 171 ++++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 120 insertions(+), 51 deletions(-)
fs/ceph/file.c +120 -51
index 3de8982..78018c9 100644
@@ -408,51 +408,106 @@ 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 iov_iter *i,
+				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, n;
 
-	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;
+		return ret;
 
-	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
-			   file->f_flags & O_DIRECT,
-			   (unsigned long)data & ~PAGE_MASK);
+	if (file->f_flags & O_DIRECT) {
+		for (n = 0; n < i->nr_segs; n++) {
+			void __user *data = i->iov[n].iov_base;
+			size_t len = i->iov[n].iov_len;
+
+			if (n == 0) {
+				len -=  i->iov_offset;
+				data += i->iov_offset;
+			}
+
+			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 (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 (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 l, k = 0;
+			size_t left = len;
+
+			while (left) {
+				void __user *data = i->iov[0].iov_base
+							+ i->iov_offset;
+				l = min(i->iov[0].iov_len - i->iov_offset,
+					len);
+
+				ret = ceph_copy_page_vector_to_user(&pages[k],
+								    data, off,
+								    l);
+				if (ret > 0) {
+					iov_iter_advance(i, ret);
+					left -= ret;
+					off += ret;
+					k = calc_pages_for(iocb->ki_pos,
+							   len - left + 1) - 1;
+					BUG_ON(k >= num_pages && left);
+				} else
+					break;
+			}
+
+			len -= left;
+			if (len > 0) {
+				iocb->ki_pos += len;
+				iocb->ki_left -= len;
+				ret = len;
+			}
+		}
 		ceph_release_page_vector(pages, num_pages);
+	}
+
 	dout("sync_read result %d\n", ret);
 	return ret;
 }
@@ -647,55 +702,69 @@ 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 = 0;
 	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:
+
+	ret = generic_segment_checks(iov, &nr_segs, &len, VERIFY_WRITE);
+	if (ret)
+		return ret;
+
 	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)) {
+		struct iov_iter i;
+
+		iocb->ki_left = len;
+		iov_iter_init(&i, iov, nr_segs, len, 0);
+again:
 		/* hmm, this isn't really async... */
-		ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
-	else
+		ret = ceph_sync_read(iocb, &i, &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) {
+				dout("sync_read hit hole, ppos %lld < size %lld"
+				     ", reading more\n", iocb->ki_pos,
+				     inode->i_size);
+
+				read += ret;
+				iov_iter_advance(&i, ret);
+				checkeof = 0;
+				goto again;
+			}
+		}
+
+	} 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;
 
Signature · 1 line
1.8.1.2
#3 lore raw
Quoted 382 lines · majianpeng
Quoted 164 lines · majianpeng
Quoted 133 lines

For readv/preadv sync-operatoin, ceph only do the first iov. It don't think other iovs.Now implement this.

V2:
-add generic_segment_checks
-using struct iov_iter replace cloning the iovs.
-return previous successfully copied if ceph_copy_page_vector_to_user
met error.

---
 fs/ceph/file.c | 174 ++++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 123 insertions(+), 51 deletions(-)
fs/ceph/file.c +67 -29
index 3de8982..1c28c52 100644
@@ -408,51 +408,109 @@ 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 iov_iter *i,
+				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, n;
 
-	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;
+		return ret;
 
-	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
-			   file->f_flags & O_DIRECT,
-			   (unsigned long)data & ~PAGE_MASK);
+	if (file->f_flags & O_DIRECT) {
+		for (n = 0; n < i->nr_segs; n++) {
+			void __user *data = i->iov[n].iov_base;
+			size_t len = i->iov[n].iov_len;
+
+			if (n == 0) {
+				len -=  i->iov_offset;
+				data += i->iov_offset;
+			}
+
+			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 (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 (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 l, k = 0;
+			size_t left = len;
+
+			for (n = 0; n < i->nr_segs && left; n++) {
+				void __user *data = i->iov[n].iov_base;
+				l = min(left, i->iov[n].iov_len);
+
+				if (n == 0) {
+					data += i->iov_offset;
+					l = min(i->iov[0].iov_len - i->iov_offset,
+						left);
+				}
+
+				ret = ceph_copy_page_vector_to_user(&pages[k],
+								    data, off,

fs/ceph/file.c more: · line 502

Quoted 10 lines
fs/ceph/file.c +10 -0
+								    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;
+			}

I think it's better to call iov_iter_advance() here instead of in ceph_aio_read(), and change the code to something like:
----
while (iov_iter_count(&i) && left > 0) {
void __user *data = i->iov->iov_base + i->iov_offset;
l = min(left, i->iov->iov_len - i->iov_offset);

ret = ceph_copy_page_vector_to_user(&pages[k], data, off, l);

  if (ret > 0) {
    iov_iter_advance(&i, ret);
    left -= ret;
    off += ret;
    k = calc_pages_for(iocb->ki_pos, len - left + 1) - 1;
    BUG_ON(k >= num_pages && left);
  } else
    break;
}
Quoted 5 lines

rest change looks good.

Regards
Yan, Zheng

Thanks!
Jianpeng Ma

Subject:ceph: Implement readv/preadv for sync operation.

For readv/preadv sync-operatoin, ceph only do the first iov. It don't think other iovs.Now implement this.

V2:
-add generic_segment_checks
-using struct iov_iter replace cloning the iovs.
-return previous successfully copied if ceph_copy_page_vector_to_user
met error.

---
 fs/ceph/file.c | 171 ++++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 120 insertions(+), 51 deletions(-)
fs/ceph/file.c +104 -37
index 3de8982..78018c9 100644
@@ -408,51 +408,106 @@ 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 iov_iter *i,
+				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, n;
 
-	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;
+		return ret;
 
-	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
-			   file->f_flags & O_DIRECT,
-			   (unsigned long)data & ~PAGE_MASK);
+	if (file->f_flags & O_DIRECT) {
+		for (n = 0; n < i->nr_segs; n++) {
+			void __user *data = i->iov[n].iov_base;
+			size_t len = i->iov[n].iov_len;
+
+			if (n == 0) {
+				len -=  i->iov_offset;
+				data += i->iov_offset;
+			}
+
+			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 (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 (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 l, k = 0;
+			size_t left = len;
+
+			while (left) {
+				void __user *data = i->iov[0].iov_base
+							+ i->iov_offset;
+				l = min(i->iov[0].iov_len - i->iov_offset,
+					len);
+
+				ret = ceph_copy_page_vector_to_user(&pages[k],
+								    data, off,
+								    l);
+				if (ret > 0) {
+					iov_iter_advance(i, ret);
+					left -= ret;
+					off += ret;
+					k = calc_pages_for(iocb->ki_pos,
+							   len - left + 1) - 1;
+					BUG_ON(k >= num_pages && left);
+				} else
+					break;
+			}
+
+			len -= left;
+			if (len > 0) {
+				iocb->ki_pos += len;
+				iocb->ki_left -= len;
+				ret = len;
+			}
+		}
 		ceph_release_page_vector(pages, num_pages);
+	}
+
 	dout("sync_read result %d\n", ret);
 	return ret;
 }
@@ -647,55 +702,69 @@ 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 = 0;
 	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:
+
+	ret = generic_segment_checks(iov, &nr_segs, &len, VERIFY_WRITE);
+	if (ret)
+		return ret;
+
 	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)) {
+		struct iov_iter i;
+
+		iocb->ki_left = len;
+		iov_iter_init(&i, iov, nr_segs, len, 0);
+again:
 		/* hmm, this isn't really async... */
-		ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
-	else
+		ret = ceph_sync_read(iocb, &i, &checkeof);
+
+		if (checkeof && ret >= 0) {
+			int statret = ceph_do_getattr(inode,
+						      CEPH_STAT_CAP_SIZE);

fs/ceph/file.c static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov, · line 755

Quoted 10 lines
fs/ceph/file.c +10 -0
+
+			/* hit EOF or hole? */
+			if (statret == 0 && iocb->ki_pos < inode->i_size &&
+				iocb->ki_left) {
+				dout("sync_read hit hole, ppos %lld < size %lld"
+				     ", reading more\n", iocb->ki_pos,
+				     inode->i_size);
+
+				read += ret;
+				iov_iter_advance(&i, ret);

I think this "iov_iter_advance" is superfluous. Other than this, your patch looks good.

By the way, your email is base64 encoded. please make your email client use plain text instead.

Regards
Yan, Zheng

Quoted 30 lines
fs/ceph/file.c +6 -14
+				checkeof = 0;
+				goto again;
+			}
+		}
+
+	} 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;
 
author
#4 lore raw
Quoted 388 lines · majianpeng
Quoted 387 lines
Quoted 164 lines · majianpeng
Quoted 133 lines

For readv/preadv sync-operatoin, ceph only do the first iov. It don't think other iovs.Now implement this.

V2:
-add generic_segment_checks
-using struct iov_iter replace cloning the iovs.
-return previous successfully copied if ceph_copy_page_vector_to_user
met error.

---
 fs/ceph/file.c | 174 ++++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 123 insertions(+), 51 deletions(-)
fs/ceph/file.c +67 -29
index 3de8982..1c28c52 100644
@@ -408,51 +408,109 @@ 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 iov_iter *i,
+				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, n;
 
-	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;
+		return ret;
 
-	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
-			   file->f_flags & O_DIRECT,
-			   (unsigned long)data & ~PAGE_MASK);
+	if (file->f_flags & O_DIRECT) {
+		for (n = 0; n < i->nr_segs; n++) {
+			void __user *data = i->iov[n].iov_base;
+			size_t len = i->iov[n].iov_len;
+
+			if (n == 0) {
+				len -=  i->iov_offset;
+				data += i->iov_offset;
+			}
+
+			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 (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 (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 l, k = 0;
+			size_t left = len;
+
+			for (n = 0; n < i->nr_segs && left; n++) {
+				void __user *data = i->iov[n].iov_base;
+				l = min(left, i->iov[n].iov_len);
+
+				if (n == 0) {
+					data += i->iov_offset;
+					l = min(i->iov[0].iov_len - i->iov_offset,
+						left);
+				}
+
+				ret = ceph_copy_page_vector_to_user(&pages[k],
+								    data, off,

fs/ceph/file.c more: · line 502

Quoted 10 lines
fs/ceph/file.c +10 -0
+								    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;
+			}

I think it's better to call iov_iter_advance() here instead of in ceph_aio_read(), and change the code to something like:
----
while (iov_iter_count(&i) && left > 0) {
void __user *data = i->iov->iov_base + i->iov_offset;
l = min(left, i->iov->iov_len - i->iov_offset);

ret = ceph_copy_page_vector_to_user(&pages[k], data, off, l);

  if (ret > 0) {
    iov_iter_advance(&i, ret);
    left -= ret;
    off += ret;
    k = calc_pages_for(iocb->ki_pos, len - left + 1) - 1;
    BUG_ON(k >= num_pages && left);
  } else
    break;
}
Quoted 5 lines

rest change looks good.

Regards
Yan, Zheng

Thanks!
Jianpeng Ma

Subject:ceph: Implement readv/preadv for sync operation.

For readv/preadv sync-operatoin, ceph only do the first iov. It don't think other iovs.Now implement this.

V2:
-add generic_segment_checks
-using struct iov_iter replace cloning the iovs.
-return previous successfully copied if ceph_copy_page_vector_to_user
met error.

---
 fs/ceph/file.c | 171 ++++++++++++++++++++++++++++++++++++++++-----------------
 1 file changed, 120 insertions(+), 51 deletions(-)
fs/ceph/file.c +109 -37
index 3de8982..78018c9 100644
@@ -408,51 +408,106 @@ 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 iov_iter *i,
+				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, n;
 
-	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;
+		return ret;
 
-	ret = striped_read(inode, off, len, pages, num_pages, checkeof,
-			   file->f_flags & O_DIRECT,
-			   (unsigned long)data & ~PAGE_MASK);
+	if (file->f_flags & O_DIRECT) {
+		for (n = 0; n < i->nr_segs; n++) {
+			void __user *data = i->iov[n].iov_base;
+			size_t len = i->iov[n].iov_len;
+
+			if (n == 0) {
+				len -=  i->iov_offset;
+				data += i->iov_offset;
+			}
+
+			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 (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 (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 l, k = 0;
+			size_t left = len;
+
+			while (left) {
+				void __user *data = i->iov[0].iov_base
+							+ i->iov_offset;
+				l = min(i->iov[0].iov_len - i->iov_offset,
+					len);
+
+				ret = ceph_copy_page_vector_to_user(&pages[k],
+								    data, off,
+								    l);
+				if (ret > 0) {
+					iov_iter_advance(i, ret);
+					left -= ret;
+					off += ret;
+					k = calc_pages_for(iocb->ki_pos,
+							   len - left + 1) - 1;
+					BUG_ON(k >= num_pages && left);
+				} else
+					break;
+			}
+
+			len -= left;
+			if (len > 0) {
+				iocb->ki_pos += len;
+				iocb->ki_left -= len;
+				ret = len;
+			}
+		}
 		ceph_release_page_vector(pages, num_pages);
+	}
+
 	dout("sync_read result %d\n", ret);
 	return ret;
 }
@@ -647,55 +702,69 @@ 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 = 0;
 	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:
+
+	ret = generic_segment_checks(iov, &nr_segs, &len, VERIFY_WRITE);
+	if (ret)
+		return ret;
+
 	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)) {
+		struct iov_iter i;
+
+		iocb->ki_left = len;
+		iov_iter_init(&i, iov, nr_segs, len, 0);
+again:
 		/* hmm, this isn't really async... */
-		ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
-	else
+		ret = ceph_sync_read(iocb, &i, &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) {
+				dout("sync_read hit hole, ppos %lld < size %lld"
Quoted 9 lines

fs/ceph/file.c static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov, · line 755

Quoted 5 lines
fs/ceph/file.c +5 -0
+				     ", reading more\n", iocb->ki_pos,
+				     inode->i_size);
+
+				read += ret;
+				iov_iter_advance(&i, ret);

I think this "iov_iter_advance" is superfluous. Other than this, your patch looks good.

No, i made a error.In ceph_sync_read, for direct mode, i don't use iov_iter_advance. But for sync mode, i used.So it's error.
I'll check carefully and resend again.

Thanks!
Jianpeng Ma

Quoted 38 lines

By the way, your email is base64 encoded. please make your email client use plain text instead.

Regards
Yan, Zheng

Quoted 30 lines
fs/ceph/file.c +6 -14
+				checkeof = 0;
+				goto again;
+			}
+		}
+
+	} 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;