Skip to contentSkip to conversation

SMB2.1 multicredit requests for writing

PATCH In review Pavel Shilovsky 2 participants 14 messages Jun 23, 2014 – Jun 26, 2014 4 messages not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover SMB2.1 multicredit requests for writing

The goal of this patchset is to add large MTU/multicredits support for writing from the CIFS client to servers that support SMB 2.1 and higher versions of the protocol.

First three patches simplify the writepages code. Patches #4 and #5 both fix the problem in writepages when a server changes wsize value after reconnects from the client. Patches #6 and #7 simplify the iovec write code, #8 fixes the wsize changing problem there. Finally, patch #9 introduces multicredits support for writing.

Pavel Shilovsky (9):
CIFS: Separate page processing from writepages
CIFS: Separate page sending from writepages
CIFS: Separate pages initialization from writepages
CIFS: Fix wsize usage in writepages
CIFS: Fix cifs_writev_requeue when wsize changes
CIFS: Separate filling pages from iovec write
CIFS: Separate writing from iovec write
CIFS: Fix wsize usage in iovec write
CIFS: Use multicredits for writes

 fs/cifs/cifsglob.h      |   17 ++
 fs/cifs/cifsproto.h     |    3 +
 fs/cifs/cifssmb.c       |   84 ++++++--
 fs/cifs/file.c          |  540 ++++++++++++++++++++++++++++-------------------
 fs/cifs/smb1ops.c       |    8 +
 fs/cifs/smb2ops.c       |   62 +++++-
 fs/cifs/smb2pdu.c       |   24 ++-
 fs/cifs/smb2transport.c |    5 +
 fs/cifs/transport.c     |   25 ++-
 9 files changed, 517 insertions(+), 251 deletions(-)
Signature · 1 line
1.7.10.4

A message that is not in the archive (sent off-list or not yet received): 1403535517-8301-1-git-send-email-pshilovsky-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org

#2 lore raw

PATCH 1/9 CIFS: Separate page processing from writepages

---
 fs/cifs/file.c |  154 ++++++++++++++++++++++++++++++--------------------------
 1 file changed, 83 insertions(+), 71 deletions(-)
fs/cifs/file.c +83 -71
index 208f56e..216959e 100644
@@ -1878,6 +1878,86 @@ static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
 	return rc;
 }
 
+static unsigned int
+wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
+		    struct address_space *mapping,
+		    struct writeback_control *wbc,
+		    pgoff_t end, pgoff_t *index, pgoff_t *next, bool *done)
+{
+	unsigned int nr_pages = 0, i;
+	struct page *page;
+
+	for (i = 0; i < found_pages; i++) {
+		page = wdata->pages[i];
+		/*
+		 * At this point we hold neither mapping->tree_lock nor
+		 * lock on the page itself: the page may be truncated or
+		 * invalidated (changing page->mapping to NULL), or even
+		 * swizzled back from swapper_space to tmpfs file
+		 * mapping
+		 */
+
+		if (nr_pages == 0)
+			lock_page(page);
+		else if (!trylock_page(page))
+			break;
+
+		if (unlikely(page->mapping != mapping)) {
+			unlock_page(page);
+			break;
+		}
+
+		if (!wbc->range_cyclic && page->index > end) {
+			*done = true;
+			unlock_page(page);
+			break;
+		}
+
+		if (*next && (page->index != *next)) {
+			/* Not next consecutive page */
+			unlock_page(page);
+			break;
+		}
+
+		if (wbc->sync_mode != WB_SYNC_NONE)
+			wait_on_page_writeback(page);
+
+		if (PageWriteback(page) ||
+				!clear_page_dirty_for_io(page)) {
+			unlock_page(page);
+			break;
+		}
+
+		/*
+		 * This actually clears the dirty bit in the radix tree.
+		 * See cifs_writepage() for more commentary.
+		 */
+		set_page_writeback(page);
+		if (page_offset(page) >= i_size_read(mapping->host)) {
+			*done = true;
+			unlock_page(page);
+			end_page_writeback(page);
+			break;
+		}
+
+		wdata->pages[i] = page;
+		*next = page->index + 1;
+		++nr_pages;
+	}
+
+	/* reset index to refind any pages skipped */
+	if (nr_pages == 0)
+		*index = wdata->pages[0]->index + 1;
+
+	/* put any pages we aren't going to use */
+	for (i = nr_pages; i < found_pages; i++) {
+		page_cache_release(wdata->pages[i]);
+		wdata->pages[i] = NULL;
+	}
+
+	return nr_pages;
+}
+
 static int cifs_writepages(struct address_space *mapping,
 			   struct writeback_control *wbc)
 {
@@ -1886,7 +1966,6 @@ static int cifs_writepages(struct address_space *mapping,
 	pgoff_t end, index;
 	struct cifs_writedata *wdata;
 	struct TCP_Server_Info *server;
-	struct page *page;
 	int rc = 0;
 
 	/*
@@ -1908,7 +1987,7 @@ static int cifs_writepages(struct address_space *mapping,
 	}
 retry:
 	while (!done && index <= end) {
-		unsigned int i, nr_pages, found_pages;
+		unsigned int nr_pages, found_pages;
 		pgoff_t next = 0, tofind;
 		struct page **pages;
 
@@ -1944,75 +2023,8 @@ retry:
 			break;
 		}
 
-		nr_pages = 0;
-		for (i = 0; i < found_pages; i++) {
-			page = wdata->pages[i];
-			/*
-			 * At this point we hold neither mapping->tree_lock nor
-			 * lock on the page itself: the page may be truncated or
-			 * invalidated (changing page->mapping to NULL), or even
-			 * swizzled back from swapper_space to tmpfs file
-			 * mapping
-			 */
-
-			if (nr_pages == 0)
-				lock_page(page);
-			else if (!trylock_page(page))
-				break;
-
-			if (unlikely(page->mapping != mapping)) {
-				unlock_page(page);
-				break;
-			}
-
-			if (!wbc->range_cyclic && page->index > end) {
-				done = true;
-				unlock_page(page);
-				break;
-			}
-
-			if (next && (page->index != next)) {
-				/* Not next consecutive page */
-				unlock_page(page);
-				break;
-			}
-
-			if (wbc->sync_mode != WB_SYNC_NONE)
-				wait_on_page_writeback(page);
-
-			if (PageWriteback(page) ||
-					!clear_page_dirty_for_io(page)) {
-				unlock_page(page);
-				break;
-			}
-
-			/*
-			 * This actually clears the dirty bit in the radix tree.
-			 * See cifs_writepage() for more commentary.
-			 */
-			set_page_writeback(page);
-
-			if (page_offset(page) >= i_size_read(mapping->host)) {
-				done = true;
-				unlock_page(page);
-				end_page_writeback(page);
-				break;
-			}
-
-			wdata->pages[i] = page;
-			next = page->index + 1;
-			++nr_pages;
-		}
-
-		/* reset index to refind any pages skipped */
-		if (nr_pages == 0)
-			index = wdata->pages[0]->index + 1;
-
-		/* put any pages we aren't going to use */
-		for (i = nr_pages; i < found_pages; i++) {
-			page_cache_release(wdata->pages[i]);
-			wdata->pages[i] = NULL;
-		}
+		nr_pages = wdata_prepare_pages(wdata, found_pages, mapping, wbc,
+					       end, &index, &next, &done);
 
 		/* nothing to write? */
 		if (nr_pages == 0) {
Signature · 1 line
1.7.10.4
#3 lore raw

PATCH 2/9 CIFS: Separate page sending from writepages

In reply to a message not in the archive
---
 fs/cifs/file.c |   98 ++++++++++++++++++++++++++++++--------------------------
 1 file changed, 53 insertions(+), 45 deletions(-)
fs/cifs/file.c +53 -45
index 216959e..f143a4f 100644
@@ -1958,6 +1958,58 @@ wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
 	return nr_pages;
 }
 
+static int
+wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
+		 struct address_space *mapping, struct writeback_control *wbc)
+{
+	int rc = 0;
+	struct TCP_Server_Info *server;
+	unsigned int i;
+
+	wdata->sync_mode = wbc->sync_mode;
+	wdata->nr_pages = nr_pages;
+	wdata->offset = page_offset(wdata->pages[0]);
+	wdata->pagesz = PAGE_CACHE_SIZE;
+	wdata->tailsz = min(i_size_read(mapping->host) -
+			page_offset(wdata->pages[nr_pages - 1]),
+			(loff_t)PAGE_CACHE_SIZE);
+	wdata->bytes = ((nr_pages - 1) * PAGE_CACHE_SIZE) + wdata->tailsz;
+
+	do {
+		if (wdata->cfile != NULL)
+			cifsFileInfo_put(wdata->cfile);
+		wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
+		if (!wdata->cfile) {
+			cifs_dbg(VFS, "No writable handles for inode\n");
+			rc = -EBADF;
+			break;
+		}
+		wdata->pid = wdata->cfile->pid;
+		server = tlink_tcon(wdata->cfile->tlink)->ses->server;
+		rc = server->ops->async_writev(wdata, cifs_writedata_release);
+	} while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
+
+	for (i = 0; i < nr_pages; ++i)
+		unlock_page(wdata->pages[i]);
+
+	if (!rc)
+		return rc;
+
+	/* send failure -- clean up the mess */
+	for (i = 0; i < nr_pages; ++i) {
+		if (rc == -EAGAIN)
+			redirty_page_for_writepage(wbc, wdata->pages[i]);
+		else
+			SetPageError(wdata->pages[i]);
+		end_page_writeback(wdata->pages[i]);
+		page_cache_release(wdata->pages[i]);
+	}
+	if (rc != -EAGAIN)
+		mapping_set_error(mapping, rc);
+
+	return rc;
+}
+
 static int cifs_writepages(struct address_space *mapping,
 			   struct writeback_control *wbc)
 {
@@ -1965,7 +2017,6 @@ static int cifs_writepages(struct address_space *mapping,
 	bool done = false, scanned = false, range_whole = false;
 	pgoff_t end, index;
 	struct cifs_writedata *wdata;
-	struct TCP_Server_Info *server;
 	int rc = 0;
 
 	/*
@@ -2032,50 +2083,7 @@ retry:
 			continue;
 		}
 
-		wdata->sync_mode = wbc->sync_mode;
-		wdata->nr_pages = nr_pages;
-		wdata->offset = page_offset(wdata->pages[0]);
-		wdata->pagesz = PAGE_CACHE_SIZE;
-		wdata->tailsz =
-			min(i_size_read(mapping->host) -
-			    page_offset(wdata->pages[nr_pages - 1]),
-			    (loff_t)PAGE_CACHE_SIZE);
-		wdata->bytes = ((nr_pages - 1) * PAGE_CACHE_SIZE) +
-					wdata->tailsz;
-
-		do {
-			if (wdata->cfile != NULL)
-				cifsFileInfo_put(wdata->cfile);
-			wdata->cfile = find_writable_file(CIFS_I(mapping->host),
-							  false);
-			if (!wdata->cfile) {
-				cifs_dbg(VFS, "No writable handles for inode\n");
-				rc = -EBADF;
-				break;
-			}
-			wdata->pid = wdata->cfile->pid;
-			server = tlink_tcon(wdata->cfile->tlink)->ses->server;
-			rc = server->ops->async_writev(wdata,
-							cifs_writedata_release);
-		} while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
-
-		for (i = 0; i < nr_pages; ++i)
-			unlock_page(wdata->pages[i]);
-
-		/* send failure -- clean up the mess */
-		if (rc != 0) {
-			for (i = 0; i < nr_pages; ++i) {
-				if (rc == -EAGAIN)
-					redirty_page_for_writepage(wbc,
-							   wdata->pages[i]);
-				else
-					SetPageError(wdata->pages[i]);
-				end_page_writeback(wdata->pages[i]);
-				page_cache_release(wdata->pages[i]);
-			}
-			if (rc != -EAGAIN)
-				mapping_set_error(mapping, rc);
-		}
+		rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
 		kref_put(&wdata->refcount, cifs_writedata_release);
 
 		wbc->nr_to_write -= nr_pages;
Signature · 1 line
1.7.10.4
#4 lore raw

PATCH 3/9 CIFS: Separate pages initialization from writepages

In reply to a message not in the archive
---
 fs/cifs/file.c |   56 ++++++++++++++++++++++++++++++++++++--------------------
 1 file changed, 36 insertions(+), 20 deletions(-)
fs/cifs/file.c +36 -20
index f143a4f..2c7fbd7 100644
@@ -1878,6 +1878,40 @@ static int cifs_partialpagewrite(struct page *page, unsigned from, unsigned to)
 	return rc;
 }
 
+static struct cifs_writedata *
+wdata_alloc_and_fillpages(pgoff_t tofind, struct address_space *mapping,
+			  pgoff_t end, pgoff_t *index,
+			  unsigned int *found_pages)
+{
+	unsigned int nr_pages;
+	struct page **pages;
+	struct cifs_writedata *wdata;
+
+	wdata = cifs_writedata_alloc((unsigned int)tofind,
+				     cifs_writev_complete);
+	if (!wdata)
+		return NULL;
+
+	/*
+	 * find_get_pages_tag seems to return a max of 256 on each
+	 * iteration, so we must call it several times in order to
+	 * fill the array or the wsize is effectively limited to
+	 * 256 * PAGE_CACHE_SIZE.
+	 */
+	*found_pages = 0;
+	pages = wdata->pages;
+	do {
+		nr_pages = find_get_pages_tag(mapping, index,
+					      PAGECACHE_TAG_DIRTY, tofind,
+					      pages);
+		*found_pages += nr_pages;
+		tofind -= nr_pages;
+		pages += nr_pages;
+	} while (nr_pages && tofind && *index <= end);
+
+	return wdata;
+}
+
 static unsigned int
 wdata_prepare_pages(struct cifs_writedata *wdata, unsigned int found_pages,
 		    struct address_space *mapping,
@@ -2040,35 +2074,17 @@ retry:
 	while (!done && index <= end) {
 		unsigned int nr_pages, found_pages;
 		pgoff_t next = 0, tofind;
-		struct page **pages;
 
 		tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
 				end - index) + 1;
 
-		wdata = cifs_writedata_alloc((unsigned int)tofind,
-					     cifs_writev_complete);
+		wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
+						  &found_pages);
 		if (!wdata) {
 			rc = -ENOMEM;
 			break;
 		}
 
-		/*
-		 * find_get_pages_tag seems to return a max of 256 on each
-		 * iteration, so we must call it several times in order to
-		 * fill the array or the wsize is effectively limited to
-		 * 256 * PAGE_CACHE_SIZE.
-		 */
-		found_pages = 0;
-		pages = wdata->pages;
-		do {
-			nr_pages = find_get_pages_tag(mapping, &index,
-							PAGECACHE_TAG_DIRTY,
-							tofind, pages);
-			found_pages += nr_pages;
-			tofind -= nr_pages;
-			pages += nr_pages;
-		} while (nr_pages && tofind && index <= end);
-
 		if (found_pages == 0) {
 			kref_put(&wdata->refcount, cifs_writedata_release);
 			break;
Signature · 1 line
1.7.10.4
#5 lore raw

PATCH 4/9 CIFS: Fix wsize usage in writepages

In reply to a message not in the archive

If a server change maximum buffer size for write (wsize) requests on reconnect we can fail on repeating with a big size buffer on -EAGAIN error in writepages. Fix this by checking wsize all the time before repeating request in writepages.

---
 fs/cifs/file.c |   25 ++++++++++++++-----------
 1 file changed, 14 insertions(+), 11 deletions(-)
fs/cifs/file.c +14 -11
index 2c7fbd7..e9dc57a 100644
@@ -2009,19 +2009,17 @@ wdata_send_pages(struct cifs_writedata *wdata, unsigned int nr_pages,
 			(loff_t)PAGE_CACHE_SIZE);
 	wdata->bytes = ((nr_pages - 1) * PAGE_CACHE_SIZE) + wdata->tailsz;
 
-	do {
-		if (wdata->cfile != NULL)
-			cifsFileInfo_put(wdata->cfile);
-		wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
-		if (!wdata->cfile) {
-			cifs_dbg(VFS, "No writable handles for inode\n");
-			rc = -EBADF;
-			break;
-		}
+	if (wdata->cfile != NULL)
+		cifsFileInfo_put(wdata->cfile);
+	wdata->cfile = find_writable_file(CIFS_I(mapping->host), false);
+	if (!wdata->cfile) {
+		cifs_dbg(VFS, "No writable handles for inode\n");
+		rc = -EBADF;
+	} else {
 		wdata->pid = wdata->cfile->pid;
 		server = tlink_tcon(wdata->cfile->tlink)->ses->server;
 		rc = server->ops->async_writev(wdata, cifs_writedata_release);
-	} while (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN);
+	}
 
 	for (i = 0; i < nr_pages; ++i)
 		unlock_page(wdata->pages[i]);
@@ -2073,7 +2071,7 @@ static int cifs_writepages(struct address_space *mapping,
 retry:
 	while (!done && index <= end) {
 		unsigned int nr_pages, found_pages;
-		pgoff_t next = 0, tofind;
+		pgoff_t next = 0, tofind, saved_index = index;
 
 		tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
 				end - index) + 1;
@@ -2102,6 +2100,11 @@ retry:
 		rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
 		kref_put(&wdata->refcount, cifs_writedata_release);
 
+		if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
+			index = saved_index;
+			continue;
+		}
+
 		wbc->nr_to_write -= nr_pages;
 		if (wbc->nr_to_write <= 0)
 			done = true;
Signature · 1 line
1.7.10.4
#6 lore raw

PATCH 5/9 CIFS: Fix cifs_writev_requeue when wsize changes

In reply to a message not in the archive

If wsize changes on reconnect we need to use new writedata structure that for retrying.

---
 fs/cifs/cifsglob.h |    2 ++
 fs/cifs/cifssmb.c  |   84 ++++++++++++++++++++++++++++++++++++++++++----------
 fs/cifs/smb1ops.c  |    7 +++++
 fs/cifs/smb2ops.c  |   10 +++++++
 4 files changed, 87 insertions(+), 16 deletions(-)
fs/cifs/cifsglob.h +2 -0
index de6aed8..8c53f20 100644
@@ -404,6 +404,8 @@ struct smb_version_operations {
 			const struct cifs_fid *, u32 *);
 	int (*set_acl)(struct cifs_ntsd *, __u32, struct inode *, const char *,
 			int);
+	/* writepages retry size */
+	unsigned int (*wp_retry_size)(struct inode *);
 };
 
 struct smb_version_values {
fs/cifs/cifssmb.c +68 -16

95 lines not on this page · raw message

fs/cifs/smb1ops.c +7 -0
index d1fdfa8..8a96342 100644
@@ -1009,6 +1009,12 @@ cifs_is_read_op(__u32 oplock)
 	return oplock == OPLOCK_READ;
 }
 
+static unsigned int
+cifs_wp_retry_size(struct inode *inode)
+{
+	return CIFS_SB(inode->i_sb)->wsize;
+}
+
 struct smb_version_operations smb1_operations = {
 	.send_cancel = send_nt_cancel,
 	.compare_fids = cifs_compare_fids,
@@ -1078,6 +1084,7 @@ struct smb_version_operations smb1_operations = {
 	.query_mf_symlink = cifs_query_mf_symlink,
 	.create_mf_symlink = cifs_create_mf_symlink,
 	.is_read_op = cifs_is_read_op,
+	.wp_retry_size = cifs_wp_retry_size,
 #ifdef CONFIG_CIFS_XATTR
 	.query_all_EAs = CIFSSMBQAllEAs,
 	.set_EA = CIFSSMBSetEA,
fs/cifs/smb2ops.c +10 -0
index 787844b..e35ce5b 100644
@@ -1104,6 +1104,13 @@ smb3_parse_lease_buf(void *buf, unsigned int *epoch)
 	return le32_to_cpu(lc->lcontext.LeaseState);
 }
 
+static unsigned int
+smb2_wp_retry_size(struct inode *inode)
+{
+	return min_t(unsigned int, CIFS_SB(inode->i_sb)->wsize,
+		     SMB2_MAX_BUFFER_SIZE);
+}
+
 struct smb_version_operations smb20_operations = {
 	.compare_fids = smb2_compare_fids,
 	.setup_request = smb2_setup_request,
@@ -1177,6 +1184,7 @@ struct smb_version_operations smb20_operations = {
 	.create_lease_buf = smb2_create_lease_buf,
 	.parse_lease_buf = smb2_parse_lease_buf,
 	.clone_range = smb2_clone_range,
+	.wp_retry_size = smb2_wp_retry_size,
 };
 
 struct smb_version_operations smb21_operations = {
@@ -1252,6 +1260,7 @@ struct smb_version_operations smb21_operations = {
 	.create_lease_buf = smb2_create_lease_buf,
 	.parse_lease_buf = smb2_parse_lease_buf,
 	.clone_range = smb2_clone_range,
+	.wp_retry_size = smb2_wp_retry_size,
 };
 
 struct smb_version_operations smb30_operations = {
@@ -1330,6 +1339,7 @@ struct smb_version_operations smb30_operations = {
 	.parse_lease_buf = smb3_parse_lease_buf,
 	.clone_range = smb2_clone_range,
 	.validate_negotiate = smb3_validate_negotiate,
+	.wp_retry_size = smb2_wp_retry_size,
 };
 
 struct smb_version_values smb20_values = {
Signature · 1 line
1.7.10.4
#7 lore raw

PATCH 6/9 CIFS: Separate filling pages from iovec write

In reply to a message not in the archive
---
 fs/cifs/file.c |   88 ++++++++++++++++++++++++++++++++------------------------
 1 file changed, 50 insertions(+), 38 deletions(-)
fs/cifs/file.c +50 -38
index e9dc57a..89d647b 100644
@@ -2423,12 +2423,59 @@ cifs_uncached_retry_writev(struct cifs_writedata *wdata)
 	return rc;
 }
 
+static int
+wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *it,
+		      size_t *len, unsigned long nr_pages)
+{
+	int rc = 0;
+	size_t save_len, copied, bytes, cur_len = *len;
+	unsigned long i;
+
+	save_len = cur_len;
+	for (i = 0; i < nr_pages; i++) {
+		bytes = min_t(const size_t, cur_len, PAGE_SIZE);
+		copied = iov_iter_copy_from_user(wdata->pages[i], it,
+						 0, bytes);
+		cur_len -= copied;
+		iov_iter_advance(it, copied);
+		/*
+		 * If we didn't copy as much as we expected, then that
+		 * may mean we trod into an unmapped area. Stop copying
+		 * at that point. On the next pass through the big
+		 * loop, we'll likely end up getting a zero-length
+		 * write and bailing out of it.
+		 */
+		if (copied < bytes)
+			break;
+	}
+	cur_len = save_len - cur_len;
+	*len = cur_len;
+
+	/*
+	 * If we have no data to send, then that probably means that
+	 * the copy above failed altogether. That's most likely because
+	 * the address in the iovec was bogus. Return -EFAULT and let
+	 * the caller free anything we allocated and bail out.
+	 */
+	if (!cur_len)
+		return -EFAULT;
+
+	/*
+	 * i + 1 now represents the number of pages we actually used in
+	 * the copy phase above. Bring nr_pages down to that, and free
+	 * any pages that we didn't use.
+	 */
+	for ( ; nr_pages > i + 1; nr_pages--)
+		put_page(wdata->pages[nr_pages - 1]);
+	return rc;
+}
+
 static ssize_t
 cifs_iovec_write(struct file *file, const struct iovec *iov,
 		 unsigned long nr_segs, loff_t *poffset)
 {
 	unsigned long nr_pages, i;
-	size_t bytes, copied, len, cur_len;
+	size_t len, cur_len;
 	ssize_t total_written = 0;
 	loff_t offset;
 	struct iov_iter it;
@@ -2465,8 +2512,6 @@ cifs_iovec_write(struct file *file, const struct iovec *iov,
 
 	iov_iter_init(&it, iov, nr_segs, len, 0);
 	do {
-		size_t save_len;
-
 		nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
 		wdata = cifs_writedata_alloc(nr_pages,
 					     cifs_uncached_writev_complete);
@@ -2481,47 +2526,14 @@ cifs_iovec_write(struct file *file, const struct iovec *iov,
 			break;
 		}
 
-		save_len = cur_len;
-		for (i = 0; i < nr_pages; i++) {
-			bytes = min_t(const size_t, cur_len, PAGE_SIZE);
-			copied = iov_iter_copy_from_user(wdata->pages[i], &it,
-							 0, bytes);
-			cur_len -= copied;
-			iov_iter_advance(&it, copied);
-			/*
-			 * If we didn't copy as much as we expected, then that
-			 * may mean we trod into an unmapped area. Stop copying
-			 * at that point. On the next pass through the big
-			 * loop, we'll likely end up getting a zero-length
-			 * write and bailing out of it.
-			 */
-			if (copied < bytes)
-				break;
-		}
-		cur_len = save_len - cur_len;
-
-		/*
-		 * If we have no data to send, then that probably means that
-		 * the copy above failed altogether. That's most likely because
-		 * the address in the iovec was bogus. Set the rc to -EFAULT,
-		 * free anything we allocated and bail out.
-		 */
-		if (!cur_len) {
+		rc = wdata_fill_from_iovec(wdata, &it, &cur_len, nr_pages);
+		if (rc) {
 			for (i = 0; i < nr_pages; i++)
 				put_page(wdata->pages[i]);
 			kfree(wdata);
-			rc = -EFAULT;
 			break;
 		}
 
-		/*
-		 * i + 1 now represents the number of pages we actually used in
-		 * the copy phase above. Bring nr_pages down to that, and free
-		 * any pages that we didn't use.
-		 */
-		for ( ; nr_pages > i + 1; nr_pages--)
-			put_page(wdata->pages[nr_pages - 1]);
-
 		wdata->sync_mode = WB_SYNC_ALL;
 		wdata->nr_pages = nr_pages;
 		wdata->offset = (__u64)offset;
Signature · 1 line
1.7.10.4

A message that is not in the archive (sent off-list or not yet received): 1403535517-8301-7-git-send-email-pshilovsky-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org

#9 lore raw

Re: [PATCH 6/9] CIFS: Separate filling pages from iovec write

FYI - merge conflict in current cifs-2.6.git for-next (and current mainline kernel) due to Al Viro's read_iter and write_iter changes.

Presumably due to changeset
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/fs/cifs?id=16b9057804c02e2d351e9c8f606e909b43cbd9e7

Would you rebase on current cifs-2.6.git or current mainline from this patch on?

Quoted 136 lines · Pavel Shilovsky
---
 fs/cifs/file.c |   88 ++++++++++++++++++++++++++++++++------------------------
 1 file changed, 50 insertions(+), 38 deletions(-)
fs/cifs/file.c +50 -39
index e9dc57a..89d647b 100644
@@ -2423,12 +2423,59 @@ cifs_uncached_retry_writev(struct cifs_writedata *wdata)
        return rc;
 }
 
+static int
+wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *it,
+                     size_t *len, unsigned long nr_pages)
+{
+       int rc = 0;
+       size_t save_len, copied, bytes, cur_len = *len;
+       unsigned long i;
+
+       save_len = cur_len;
+       for (i = 0; i < nr_pages; i++) {
+               bytes = min_t(const size_t, cur_len, PAGE_SIZE);
+               copied = iov_iter_copy_from_user(wdata->pages[i], it,
+                                                0, bytes);
+               cur_len -= copied;
+               iov_iter_advance(it, copied);
+               /*
+                * If we didn't copy as much as we expected, then that
+                * may mean we trod into an unmapped area. Stop copying
+                * at that point. On the next pass through the big
+                * loop, we'll likely end up getting a zero-length
+                * write and bailing out of it.
+                */
+               if (copied < bytes)
+                       break;
+       }
+       cur_len = save_len - cur_len;
+       *len = cur_len;
+
+       /*
+        * If we have no data to send, then that probably means that
+        * the copy above failed altogether. That's most likely because
+        * the address in the iovec was bogus. Return -EFAULT and let
+        * the caller free anything we allocated and bail out.
+        */
+       if (!cur_len)
+               return -EFAULT;
+
+       /*
+        * i + 1 now represents the number of pages we actually used in
+        * the copy phase above. Bring nr_pages down to that, and free
+        * any pages that we didn't use.
+        */
+       for ( ; nr_pages > i + 1; nr_pages--)
+               put_page(wdata->pages[nr_pages - 1]);
+       return rc;
+}
+
 static ssize_t
 cifs_iovec_write(struct file *file, const struct iovec *iov,
                 unsigned long nr_segs, loff_t *poffset)
 {
        unsigned long nr_pages, i;
-       size_t bytes, copied, len, cur_len;
+       size_t len, cur_len;
        ssize_t total_written = 0;
        loff_t offset;
        struct iov_iter it;
@@ -2465,8 +2512,6 @@ cifs_iovec_write(struct file *file, const struct iovec *iov,
 
        iov_iter_init(&it, iov, nr_segs, len, 0);
        do {
-               size_t save_len;
-
                nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
                wdata = cifs_writedata_alloc(nr_pages,
                                             cifs_uncached_writev_complete);
@@ -2481,47 +2526,14 @@ cifs_iovec_write(struct file *file, const struct iovec *iov,
                        break;
                }
 
-               save_len = cur_len;
-               for (i = 0; i < nr_pages; i++) {
-                       bytes = min_t(const size_t, cur_len, PAGE_SIZE);
-                       copied = iov_iter_copy_from_user(wdata->pages[i], &it,
-                                                        0, bytes);
-                       cur_len -= copied;
-                       iov_iter_advance(&it, copied);
-                       /*
-                        * If we didn't copy as much as we expected, then that
-                        * may mean we trod into an unmapped area. Stop copying
-                        * at that point. On the next pass through the big
-                        * loop, we'll likely end up getting a zero-length
-                        * write and bailing out of it.
-                        */
-                       if (copied < bytes)
-                               break;
-               }
-               cur_len = save_len - cur_len;
-
-               /*
-                * If we have no data to send, then that probably means that
-                * the copy above failed altogether. That's most likely because
-                * the address in the iovec was bogus. Set the rc to -EFAULT,
-                * free anything we allocated and bail out.
-                */
-               if (!cur_len) {
+               rc = wdata_fill_from_iovec(wdata, &it, &cur_len, nr_pages);
+               if (rc) {
                        for (i = 0; i < nr_pages; i++)
                                put_page(wdata->pages[i]);
                        kfree(wdata);
-                       rc = -EFAULT;
                        break;
                }
 
-               /*
-                * i + 1 now represents the number of pages we actually used in
-                * the copy phase above. Bring nr_pages down to that, and free
-                * any pages that we didn't use.
-                */
-               for ( ; nr_pages > i + 1; nr_pages--)
-                       put_page(wdata->pages[nr_pages - 1]);
-
                wdata->sync_mode = WB_SYNC_ALL;
                wdata->nr_pages = nr_pages;
                wdata->offset = (__u64)offset;
--

1.7.10.4

--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 3 lines
Thanks,

Steve

A message that is not in the archive (sent off-list or not yet received): CAH2r5mvK-co524mw8yEsA3-EbzpeRG2rO+FurGJ5CThbwQdLYg-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

#11 lore raw

Re: [PATCH 6/9] CIFS: Separate filling pages from iovec write

2014-06-26 9:46 GMT+04:00 Steve French <[email protected]>:

Quoted 7 lines

FYI - merge conflict in current cifs-2.6.git for-next (and current mainline kernel) due to Al Viro's read_iter and write_iter changes.

Presumably due to changeset
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/fs/cifs?id=16b9057804c02e2d351e9c8f606e909b43cbd9e7

Would you rebase on current cifs-2.6.git or current mainline from this patch on?

Yes, I started to do it but noticed that the existing code (without my patches) has data coherency problem on reconnect during reading when mounting with cache=none. Now, I am investigating this.

Signature · 2 lines
Best regards,
Pavel Shilovsky.

A message that is not in the archive (sent off-list or not yet received): CAKywueQao8-Oqi_OwM8yKu63LcYUL8jCd4h7kTt6y8_ZJ7YPwQ-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org

#13 lore raw

Re: [PATCH 6/9] CIFS: Separate filling pages from iovec write

2014-06-26 13:33 GMT+04:00 Pavel Shilovsky <[email protected]>:

Quoted 4 lines

2014-06-26 9:46 GMT+04:00 Steve French <[email protected]>:

Quoted 3 lines

FYI - merge conflict in current cifs-2.6.git for-next (and current mainline kernel) due to Al Viro's read_iter and write_iter changes.

Quoted 8 lines
Quoted 4 lines

Presumably due to changeset
https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/fs/cifs?id=16b9057804c02e2d351e9c8f606e909b43cbd9e7

Would you rebase on current cifs-2.6.git or current mainline from this patch on?

Yes, I started to do it but noticed that the existing code (without my patches) has data coherency problem on reconnect during reading when mounting with cache=none. Now, I am investigating this.

Figured it out:

started to receive read data (16 iovs are expected):

Jun 26 14:46:03 adventure kernel: [11874.820889]
/home/piastry/git/cifs-2.6/fs/cifs/cifssmb.c: cifs_readv_receive:
mid=4938 offset=25308416 bytes=65536
Jun 26 14:46:03 adventure kernel: [11874.820893]
/home/piastry/git/cifs-2.6/fs/cifs/cifssmb.c: cifs_readv_receive:
total_read=84 data_offset=84
Jun 26 14:46:03 adventure kernel: [11874.820895]
/home/piastry/git/cifs-2.6/fs/cifs/cifssmb.c: 0: iov_base=f43e5340
iov_len=84
Jun 26 14:46:03 adventure kernel: [11874.820898]
/home/piastry/git/cifs-2.6/fs/cifs/file.c: 0: iov_base=ffd10000
iov_len=4096
Jun 26 14:46:03 adventure kernel: [11874.821739]
/home/piastry/git/cifs-2.6/fs/cifs/file.c: 1: iov_base=ffd11000
iov_len=4096
Jun 26 14:46:03 adventure kernel: [11874.821745]
/home/piastry/git/cifs-2.6/fs/cifs/file.c: 2: iov_base=ffd12000
iov_len=4096
Jun 26 14:46:03 adventure kernel: [11874.821750]
/home/piastry/git/cifs-2.6/fs/cifs/file.c: 3: iov_base=ffd13000
iov_len=4096
Jun 26 14:46:03 adventure kernel: [11874.823222]
/home/piastry/git/cifs-2.6/fs/cifs/file.c: 4: iov_base=ffd14000
iov_len=4096
Jun 26 14:46:03 adventure kernel: [11874.823229]
/home/piastry/git/cifs-2.6/fs/cifs/file.c: 5: iov_base=ffd15000
iov_len=4096
Jun 26 14:46:03 adventure kernel: [11874.850629]
/home/piastry/git/cifs-2.6/fs/cifs/file.c: 6: iov_base=ffd16000
iov_len=4096
Jun 26 14:46:03 adventure kernel: [11874.850647]
/home/piastry/git/cifs-2.6/fs/cifs/file.c: 7: iov_base=ffd17000
iov_len=4096
Jun 26 14:46:03 adventure kernel: [11874.850653]

/home/piastry/git/cifs-2.6/fs/cifs/file.c: 8: iov_base=ffd18000 iov_len=4096

stop the link here (got only 8 iovs) ...

Jun 26 14:47:26 adventure kernel: [11957.724068]
/home/piastry/git/cifs-2.6/fs/cifs/smb2pdu.c: In echo request
Jun 26 14:47:26 adventure kernel: [11957.724078]
/home/piastry/git/cifs-2.6/fs/cifs/transport.c: Sending smb:
smb_len=68
Jun 26 14:48:09 adventure kernel: [12000.850119] CIFS VFS: Server
192.168.1.34 has not responded in 120 seconds. Reconnecting...
Jun 26 14:48:09 adventure kernel: [12000.850125]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: Reconnecting tcp session
Jun 26 14:48:09 adventure kernel: [12000.850128]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: cifs_reconnect: marking
sessions and tcons for reconnect
Jun 26 14:48:09 adventure kernel: [12000.850131]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: cifs_reconnect: tearing
down socket
Jun 26 14:48:09 adventure kernel: [12000.850133]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: State: 0x3 Flags: 0x0
Jun 26 14:48:09 adventure kernel: [12000.850186]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: Post shutdown state: 0x3
Flags: 0x0
Jun 26 14:48:09 adventure kernel: [12000.850195]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: cifs_reconnect: moving
mids to private list
Jun 26 14:48:09 adventure kernel: [12000.850197]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: cifs_reconnect: issuing
mid callbacks
Jun 26 14:48:09 adventure kernel: [12000.850219]

/home/piastry/git/cifs-2.6/fs/cifs/smb2pdu.c: smb2_readv_callback:
mid=4938 state=8 result=0 bytes=65536

state=8 means -EAGAIN error that is set to rdata->result

Jun 26 14:49:09 adventure kernel: [12060.888067]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: Socket created
Jun 26 14:49:09 adventure kernel: [12060.888074]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: sndbuf 16384 rcvbuf
87380 rcvtimeo 0x6d6
Jun 26 14:49:12 adventure kernel: [12063.898765]
/home/piastry/git/cifs-2.6/fs/cifs/cifssmb.c: total_read=32852
buflen=65620 remaining=65536
Jun 26 14:49:12 adventure kernel: [12063.898811]
/home/piastry/git/cifs-2.6/fs/cifs/smb2pdu.c: Negotiate protocol
Jun 26 14:49:12 adventure kernel: [12063.898819]
/home/piastry/git/cifs-2.6/fs/cifs/transport.c: Sending smb:
smb_len=102
Jun 26 14:49:47 adventure kernel: [12098.271408]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: Received no data or
error: expecting 16052
Jun 26 14:49:47 adventure kernel: [12098.271408] got -104
Jun 26 14:49:47 adventure kernel: [12098.271413]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: Reconnecting tcp session
Jun 26 14:49:47 adventure kernel: [12098.271417]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: cifs_reconnect: marking
sessions and tcons for reconnect
Jun 26 14:49:47 adventure kernel: [12098.271420]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: cifs_reconnect: tearing
down socket
Jun 26 14:49:47 adventure kernel: [12098.271422]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: State: 0x3 Flags: 0x0
Jun 26 14:49:47 adventure kernel: [12098.271425]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: Post shutdown state: 0x3
Flags: 0x0
Jun 26 14:49:47 adventure kernel: [12098.271436]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: cifs_reconnect: moving
mids to private list
Jun 26 14:49:47 adventure kernel: [12098.271439]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: cifs_reconnect: issuing
mid callbacks
Jun 26 14:49:47 adventure kernel: [12098.271456]
/home/piastry/git/cifs-2.6/fs/cifs/transport.c: cifs_sync_mid_result:
cmd=0 mid=0 state=8
Jun 26 14:49:47 adventure kernel: [12098.271458]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: Socket created
Jun 26 14:49:47 adventure kernel: [12098.271461]
/home/piastry/git/cifs-2.6/fs/cifs/connect.c: sndbuf 16384 rcvbuf
87380 rcvtimeo 0x6d6
Jun 26 14:49:47 adventure kernel: [12098.271465]
/home/piastry/git/cifs-2.6/fs/cifs/misc.c: Null buffer passed to
cifs_small_buf_release
Jun 26 14:49:47 adventure kernel: [12098.271491]
/home/piastry/git/cifs-2.6/fs/cifs/smb2pdu.c: smb2_async_readv:
offset=25439488 bytes=65536
Jun 26 14:49:57 adventure kernel: [12108.268065]
/home/piastry/git/cifs-2.6/fs/cifs/smb2pdu.c: Negotiate protocol
Jun 26 14:49:57 adventure kernel: [12108.268075]

/home/piastry/git/cifs-2.6/fs/cifs/transport.c: Sending smb:
smb_len=102

Then connection is establed and we resend read request ...

Jun 26 14:49:58 adventure kernel: [12108.958891]
/home/piastry/git/cifs-2.6/fs/cifs/smb2pdu.c: smb2_async_readv:
offset=25308416 bytes=32768 (!!!!!!!!!!!!!!!!!!!!)
Jun 26 14:49:58 adventure kernel: [12108.958897]

/home/piastry/git/cifs-2.6/fs/cifs/transport.c: Sending smb:
smb_len=113

... with wrong length (32768 instead of 65536 as it was before). The rdata->bytes field is set to this value here:
1515 >-------length = rdata->read_into_pages(server, rdata, data_len); 1516 >-------if (length < 0)
1517 >------->-------return length;
1518
1519 >-------server->total_read += length;
1520 >-------rdata->bytes = length;

So, we got short read, update rdata->bytes and then retry the request with this new length value. I suggest we need to add a check like this (tested and ok):

if (rdata->result != -EAGAIN)

        rdata->bytes = length;

Thoughts?

Signature · 2 lines
Best regards,
Pavel Shilovsky.
#14 lore raw

PATCH 7/9 CIFS: Separate writing from iovec write

In reply to a message not in the archive
---
 fs/cifs/file.c |   75 +++++++++++++++++++++++++++++++++-----------------------
 1 file changed, 44 insertions(+), 31 deletions(-)
fs/cifs/file.c +44 -31
index 89d647b..887c18c 100644
@@ -2470,41 +2470,19 @@ wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *it,
 	return rc;
 }
 
-static ssize_t
-cifs_iovec_write(struct file *file, const struct iovec *iov,
-		 unsigned long nr_segs, loff_t *poffset)
+static int
+cifs_write_from_iovec(loff_t offset, size_t len, const struct iovec *iov,
+		      unsigned long nr_segs, struct cifsFileInfo *open_file,
+		      struct cifs_sb_info *cifs_sb,
+		      struct list_head *wdata_list)
 {
+	int rc = 0;
+	size_t cur_len;
 	unsigned long nr_pages, i;
-	size_t len, cur_len;
-	ssize_t total_written = 0;
-	loff_t offset;
+	struct cifs_writedata *wdata;
 	struct iov_iter it;
-	struct cifsFileInfo *open_file;
-	struct cifs_tcon *tcon;
-	struct cifs_sb_info *cifs_sb;
-	struct cifs_writedata *wdata, *tmp;
-	struct list_head wdata_list;
-	int rc;
 	pid_t pid;
 
-	len = iov_length(iov, nr_segs);
-	if (!len)
-		return 0;
-
-	rc = generic_write_checks(file, poffset, &len, 0);
-	if (rc)
-		return rc;
-
-	INIT_LIST_HEAD(&wdata_list);
-	cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
-	open_file = file->private_data;
-	tcon = tlink_tcon(open_file->tlink);
-
-	if (!tcon->ses->server->ops->async_writev)
-		return -ENOSYS;
-
-	offset = *poffset;
-
 	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
 		pid = open_file->pid;
 	else
@@ -2549,11 +2527,46 @@ cifs_iovec_write(struct file *file, const struct iovec *iov,
 			break;
 		}
 
-		list_add_tail(&wdata->list, &wdata_list);
+		list_add_tail(&wdata->list, wdata_list);
 		offset += cur_len;
 		len -= cur_len;
 	} while (len > 0);
 
+	return rc;
+}
+
+static ssize_t
+cifs_iovec_write(struct file *file, const struct iovec *iov,
+		 unsigned long nr_segs, loff_t *poffset)
+{
+	size_t len;
+	ssize_t total_written = 0;
+	struct cifsFileInfo *open_file;
+	struct cifs_tcon *tcon;
+	struct cifs_sb_info *cifs_sb;
+	struct cifs_writedata *wdata, *tmp;
+	struct list_head wdata_list;
+	int rc;
+
+	len = iov_length(iov, nr_segs);
+	if (!len)
+		return 0;
+
+	rc = generic_write_checks(file, poffset, &len, 0);
+	if (rc)
+		return rc;
+
+	INIT_LIST_HEAD(&wdata_list);
+	cifs_sb = CIFS_SB(file->f_path.dentry->d_sb);
+	open_file = file->private_data;
+	tcon = tlink_tcon(open_file->tlink);
+
+	if (!tcon->ses->server->ops->async_writev)
+		return -ENOSYS;
+
+	rc = cifs_write_from_iovec(*poffset, len, iov, nr_segs, open_file,
+				   cifs_sb, &wdata_list);
+
 	/*
 	 * If at least one write was successfully sent, then discard any rc
 	 * value from the later writes. If the other write succeeds, then
Signature · 1 line
1.7.10.4
#15 lore raw

PATCH 8/9 CIFS: Fix wsize usage in iovec write

In reply to a message not in the archive

If a server change maximum buffer size for write (wsize) requests on reconnect we can fail on repeating with a big size buffer on -EAGAIN error in iovec write. Fix this by checking wsize all the time before repeating request in iovec write.

---
 fs/cifs/file.c |   62 +++++++++++++++++++++++++++++++-------------------------
 1 file changed, 34 insertions(+), 28 deletions(-)
fs/cifs/file.c +34 -28
index 887c18c..4ba59cd 100644
@@ -2401,28 +2401,6 @@ cifs_uncached_writev_complete(struct work_struct *work)
 	kref_put(&wdata->refcount, cifs_uncached_writedata_release);
 }
 
-/* attempt to send write to server, retry on any -EAGAIN errors */
-static int
-cifs_uncached_retry_writev(struct cifs_writedata *wdata)
-{
-	int rc;
-	struct TCP_Server_Info *server;
-
-	server = tlink_tcon(wdata->cfile->tlink)->ses->server;
-
-	do {
-		if (wdata->cfile->invalidHandle) {
-			rc = cifs_reopen_file(wdata->cfile, false);
-			if (rc != 0)
-				continue;
-		}
-		rc = server->ops->async_writev(wdata,
-					       cifs_uncached_writedata_release);
-	} while (rc == -EAGAIN);
-
-	return rc;
-}
-
 static int
 wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *it,
 		      size_t *len, unsigned long nr_pages)
@@ -2472,23 +2450,28 @@ wdata_fill_from_iovec(struct cifs_writedata *wdata, struct iov_iter *it,
 
 static int
 cifs_write_from_iovec(loff_t offset, size_t len, const struct iovec *iov,
-		      unsigned long nr_segs, struct cifsFileInfo *open_file,
+		      unsigned long nr_segs, loff_t written,
+		      struct cifsFileInfo *open_file,
 		      struct cifs_sb_info *cifs_sb,
 		      struct list_head *wdata_list)
 {
 	int rc = 0;
-	size_t cur_len;
+	size_t cur_len, saved_len = len;
 	unsigned long nr_pages, i;
 	struct cifs_writedata *wdata;
 	struct iov_iter it;
+	loff_t saved_offset = offset;
 	pid_t pid;
+	struct TCP_Server_Info *server;
 
 	if (cifs_sb->mnt_cifs_flags & CIFS_MOUNT_RWPIDFORWARD)
 		pid = open_file->pid;
 	else
 		pid = current->tgid;
 
-	iov_iter_init(&it, iov, nr_segs, len, 0);
+	server = tlink_tcon(open_file->tlink)->ses->server;
+
+	iov_iter_init(&it, iov, nr_segs, len, written);
 	do {
 		nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
 		wdata = cifs_writedata_alloc(nr_pages,
@@ -2520,10 +2503,19 @@ cifs_write_from_iovec(loff_t offset, size_t len, const struct iovec *iov,
 		wdata->bytes = cur_len;
 		wdata->pagesz = PAGE_SIZE;
 		wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
-		rc = cifs_uncached_retry_writev(wdata);
+
+		if (!wdata->cfile->invalidHandle ||
+		    !cifs_reopen_file(wdata->cfile, false))
+			rc = server->ops->async_writev(wdata,
+					cifs_uncached_writedata_release);
 		if (rc) {
 			kref_put(&wdata->refcount,
 				 cifs_uncached_writedata_release);
+			if (rc == -EAGAIN) {
+				iov_iter_init(&it, iov, nr_segs, saved_len,
+					      written + offset - saved_offset);
+				continue;
+			}
 			break;
 		}
 
@@ -2564,7 +2556,7 @@ cifs_iovec_write(struct file *file, const struct iovec *iov,
 	if (!tcon->ses->server->ops->async_writev)
 		return -ENOSYS;
 
-	rc = cifs_write_from_iovec(*poffset, len, iov, nr_segs, open_file,
+	rc = cifs_write_from_iovec(*poffset, len, iov, nr_segs, 0, open_file,
 				   cifs_sb, &wdata_list);
 
 	/*
@@ -2595,7 +2587,21 @@ restart_loop:
 
 			/* resend call if it's a retryable error */
 			if (rc == -EAGAIN) {
-				rc = cifs_uncached_retry_writev(wdata);
+				struct list_head tmp_list;
+
+				INIT_LIST_HEAD(&tmp_list);
+				list_del_init(&wdata->list);
+
+				rc = cifs_write_from_iovec(wdata->offset,
+					wdata->bytes, iov, nr_segs,
+					wdata->offset - *poffset, open_file,
+					cifs_sb, &tmp_list);
+
+				if (!list_empty(&tmp_list))
+					list_splice(&tmp_list, &wdata_list);
+
+				kref_put(&wdata->refcount,
+					 cifs_uncached_writedata_release);
 				goto restart_loop;
 			}
 		}
Signature · 1 line
1.7.10.4
#16 lore raw

PATCH 9/9 CIFS: Use multicredits for writes

In reply to a message not in the archive

If we negotiate SMB2.1 and higher version of the protocol and a server supports large write buffer size, we need to consume 1 credit per 65536 bytes. So, we need to know how many credits we have and obtain the required number of them before constructing a writedata structure in writepages and iovec write.

---
 fs/cifs/cifsglob.h      |   15 ++++++++++++++
 fs/cifs/cifsproto.h     |    3 +++
 fs/cifs/file.c          |   34 +++++++++++++++++++++++++++----
 fs/cifs/smb1ops.c       |    1 +
 fs/cifs/smb2ops.c       |   52 +++++++++++++++++++++++++++++++++++++++++++++--
 fs/cifs/smb2pdu.c       |   24 ++++++++++++++++++----
 fs/cifs/smb2transport.c |    5 +++++
 fs/cifs/transport.c     |   25 ++++++++++++++++-------
 8 files changed, 142 insertions(+), 17 deletions(-)
fs/cifs/cifsglob.h +15 -0
index 8c53f20..95d6867 100644
@@ -406,6 +406,9 @@ struct smb_version_operations {
 			int);
 	/* writepages retry size */
 	unsigned int (*wp_retry_size)(struct inode *);
+	/* get mtu credits */
+	int (*wait_mtu_credits)(struct TCP_Server_Info *, unsigned int,
+				unsigned int *, unsigned int *);
 };
 
 struct smb_version_values {
@@ -642,6 +645,16 @@ add_credits(struct TCP_Server_Info *server, const unsigned int add,
 }
 
 static inline void
+add_credits_and_wake(struct TCP_Server_Info *server, const unsigned int add,
+		     const int optype)
+{
+	if (add) {
+		server->ops->add_credits(server, add, optype);
+		wake_up(&server->request_q);
+	}
+}
+
+static inline void
 set_credits(struct TCP_Server_Info *server, const int val)
 {
 	server->ops->set_credits(server, val);
@@ -1075,6 +1088,7 @@ struct cifs_writedata {
 	int				result;
 	unsigned int			pagesz;
 	unsigned int			tailsz;
+	unsigned int			credits;
 	unsigned int			nr_pages;
 	struct page			*pages[];
 };
@@ -1400,6 +1414,7 @@ static inline void free_dfs_info_array(struct dfs_info3_param *param,
 #define   CIFS_OBREAK_OP   0x0100    /* oplock break request */
 #define   CIFS_NEG_OP      0x0200    /* negotiate request */
 #define   CIFS_OP_MASK     0x0380    /* mask request type */
+#define   CIFS_HAS_CREDITS 0x0400    /* already has credits */
 
 /* Security Flags: indicate type of session setup needed */
 #define   CIFSSEC_MAY_SIGN	0x00001
fs/cifs/cifsproto.h +3 -0
index ca7980a..009e5cb 100644
@@ -89,6 +89,9 @@ extern struct mid_q_entry *cifs_setup_async_request(struct TCP_Server_Info *,
 						struct smb_rqst *);
 extern int cifs_check_receive(struct mid_q_entry *mid,
 			struct TCP_Server_Info *server, bool log_error);
+extern int cifs_wait_mtu_credits(struct TCP_Server_Info *server,
+				 unsigned int size, unsigned int *num,
+				 unsigned int *credits);
 extern int SendReceive2(const unsigned int /* xid */ , struct cifs_ses *,
 			struct kvec *, int /* nvec to send */,
 			int * /* type of buf returned */ , const int flags);
fs/cifs/file.c +30 -4
index 4ba59cd..d8ff8b6 100644
@@ -2046,6 +2046,7 @@ static int cifs_writepages(struct address_space *mapping,
 			   struct writeback_control *wbc)
 {
 	struct cifs_sb_info *cifs_sb = CIFS_SB(mapping->host->i_sb);
+	struct TCP_Server_Info *server;
 	bool done = false, scanned = false, range_whole = false;
 	pgoff_t end, index;
 	struct cifs_writedata *wdata;
@@ -2068,23 +2069,30 @@ static int cifs_writepages(struct address_space *mapping,
 			range_whole = true;
 		scanned = true;
 	}
+	server = cifs_sb_master_tcon(cifs_sb)->ses->server;
 retry:
 	while (!done && index <= end) {
-		unsigned int nr_pages, found_pages;
+		unsigned int nr_pages, found_pages, wsize, credits;
 		pgoff_t next = 0, tofind, saved_index = index;
 
-		tofind = min((cifs_sb->wsize / PAGE_CACHE_SIZE) - 1,
-				end - index) + 1;
+		rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
+						   &wsize, &credits);
+		if (rc)
+			break;
+
+		tofind = min((wsize / PAGE_CACHE_SIZE) - 1, end - index) + 1;
 
 		wdata = wdata_alloc_and_fillpages(tofind, mapping, end, &index,
 						  &found_pages);
 		if (!wdata) {
 			rc = -ENOMEM;
+			add_credits_and_wake(server, credits, 0);
 			break;
 		}
 
 		if (found_pages == 0) {
 			kref_put(&wdata->refcount, cifs_writedata_release);
+			add_credits_and_wake(server, credits, 0);
 			break;
 		}
 
@@ -2094,10 +2102,16 @@ retry:
 		/* nothing to write? */
 		if (nr_pages == 0) {
 			kref_put(&wdata->refcount, cifs_writedata_release);
+			add_credits_and_wake(server, credits, 0);
 			continue;
 		}
 
+		wdata->credits = credits;
+
 		rc = wdata_send_pages(wdata, nr_pages, mapping, wbc);
+		if (rc)
+			add_credits_and_wake(server, wdata->credits, 0);
+
 		kref_put(&wdata->refcount, cifs_writedata_release);
 
 		if (wbc->sync_mode == WB_SYNC_ALL && rc == -EAGAIN) {
@@ -2473,17 +2487,26 @@ cifs_write_from_iovec(loff_t offset, size_t len, const struct iovec *iov,
 
 	iov_iter_init(&it, iov, nr_segs, len, written);
 	do {
-		nr_pages = get_numpages(cifs_sb->wsize, len, &cur_len);
+		unsigned int wsize, credits;
+
+		rc = server->ops->wait_mtu_credits(server, cifs_sb->wsize,
+						   &wsize, &credits);
+		if (rc)
+			break;
+
+		nr_pages = get_numpages(wsize, len, &cur_len);
 		wdata = cifs_writedata_alloc(nr_pages,
 					     cifs_uncached_writev_complete);
 		if (!wdata) {
 			rc = -ENOMEM;
+			add_credits_and_wake(server, credits, 0);
 			break;
 		}
 
 		rc = cifs_write_allocate_pages(wdata->pages, nr_pages);
 		if (rc) {
 			kfree(wdata);
+			add_credits_and_wake(server, credits, 0);
 			break;
 		}
 
@@ -2492,6 +2515,7 @@ cifs_write_from_iovec(loff_t offset, size_t len, const struct iovec *iov,
 			for (i = 0; i < nr_pages; i++)
 				put_page(wdata->pages[i]);
 			kfree(wdata);
+			add_credits_and_wake(server, credits, 0);
 			break;
 		}
 
@@ -2503,12 +2527,14 @@ cifs_write_from_iovec(loff_t offset, size_t len, const struct iovec *iov,
 		wdata->bytes = cur_len;
 		wdata->pagesz = PAGE_SIZE;
 		wdata->tailsz = cur_len - ((nr_pages - 1) * PAGE_SIZE);
+		wdata->credits = credits;
 
 		if (!wdata->cfile->invalidHandle ||
 		    !cifs_reopen_file(wdata->cfile, false))
 			rc = server->ops->async_writev(wdata,
 					cifs_uncached_writedata_release);
 		if (rc) {
+			add_credits_and_wake(server, wdata->credits, 0);
 			kref_put(&wdata->refcount,
 				 cifs_uncached_writedata_release);
 			if (rc == -EAGAIN) {
fs/cifs/smb1ops.c +1 -0
index 8a96342..5e8c22d 100644
@@ -1025,6 +1025,7 @@ struct smb_version_operations smb1_operations = {
 	.set_credits = cifs_set_credits,
 	.get_credits_field = cifs_get_credits_field,
 	.get_credits = cifs_get_credits,
+	.wait_mtu_credits = cifs_wait_mtu_credits,
 	.get_next_mid = cifs_get_next_mid,
 	.read_data_offset = cifs_read_data_offset,
 	.read_data_length = cifs_read_data_length,
fs/cifs/smb2ops.c +50 -2

82 lines not on this page · raw message

fs/cifs/smb2pdu.c +20 -4
index b0b260d..56bf028 100644
@@ -1906,15 +1906,20 @@ int
 smb2_async_writev(struct cifs_writedata *wdata,
 		  void (*release)(struct kref *kref))
 {
-	int rc = -EACCES;
+	int rc = -EACCES, flags = 0;
 	struct smb2_write_req *req = NULL;
 	struct cifs_tcon *tcon = tlink_tcon(wdata->cfile->tlink);
+	struct TCP_Server_Info *server = tcon->ses->server;
 	struct kvec iov;
 	struct smb_rqst rqst;
 
 	rc = small_smb2_init(SMB2_WRITE, tcon, (void **) &req);
-	if (rc)
+	if (rc) {
+		if (rc == -EAGAIN && wdata->credits)
+			/* credits was reseted by reconnect */
+			wdata->credits = 0;
 		goto async_writev_out;
+	}
 
 	req->hdr.ProcessId = cpu_to_le32(wdata->cfile->pid);
 
@@ -1947,9 +1952,20 @@ smb2_async_writev(struct cifs_writedata *wdata,
 
 	inc_rfc1001_len(&req->hdr, wdata->bytes - 1 /* Buffer */);
 
+	if (wdata->credits) {
+		req->hdr.CreditCharge = cpu_to_le16(DIV_ROUND_UP(wdata->bytes,
+						    SMB2_MAX_BUFFER_SIZE));
+		spin_lock(&server->req_lock);
+		server->credits += wdata->credits -
+					le16_to_cpu(req->hdr.CreditCharge);
+		spin_unlock(&server->req_lock);
+		wake_up(&server->request_q);
+		flags = CIFS_HAS_CREDITS;
+	}
+
 	kref_get(&wdata->refcount);
-	rc = cifs_call_async(tcon->ses->server, &rqst, NULL,
-				smb2_writev_callback, wdata, 0);
+	rc = cifs_call_async(server, &rqst, NULL, smb2_writev_callback, wdata,
+			     flags);
 
 	if (rc) {
 		kref_put(&wdata->refcount, release);
fs/cifs/smb2transport.c +5 -0
index 59c748c..5111e72 100644
@@ -466,7 +466,12 @@ smb2_verify_signature(struct smb_rqst *rqst, struct TCP_Server_Info *server)
 static inline void
 smb2_seq_num_into_buf(struct TCP_Server_Info *server, struct smb2_hdr *hdr)
 {
+	unsigned int i, num = le16_to_cpu(hdr->CreditCharge);
+
 	hdr->MessageId = get_next_mid64(server);
+	/* skip message numbers according to CreditCharge field */
+	for (i = 1; i < num; i++)
+		get_next_mid(server);
 }
 
 static struct mid_q_entry *
fs/cifs/transport.c +18 -7
index 18cd565..0b5d5fc 100644
@@ -448,6 +448,15 @@ wait_for_free_request(struct TCP_Server_Info *server, const int timeout,
 	return wait_for_free_credits(server, timeout, val);
 }
 
+int
+cifs_wait_mtu_credits(struct TCP_Server_Info *server, unsigned int size,
+		      unsigned int *num, unsigned int *credits)
+{
+	*num = size;
+	*credits = 0;
+	return 0;
+}
+
 static int allocate_mid(struct cifs_ses *ses, struct smb_hdr *in_buf,
 			struct mid_q_entry **ppmidQ)
 {
@@ -531,20 +540,23 @@ cifs_call_async(struct TCP_Server_Info *server, struct smb_rqst *rqst,
 {
 	int rc, timeout, optype;
 	struct mid_q_entry *mid;
+	unsigned int credits = 0;
 
 	timeout = flags & CIFS_TIMEOUT_MASK;
 	optype = flags & CIFS_OP_MASK;
 
-	rc = wait_for_free_request(server, timeout, optype);
-	if (rc)
-		return rc;
+	if ((flags & CIFS_HAS_CREDITS) == 0) {
+		rc = wait_for_free_request(server, timeout, optype);
+		if (rc)
+			return rc;
+		credits = 1;
+	}
 
 	mutex_lock(&server->srv_mutex);
 	mid = server->ops->setup_async_request(server, rqst);
 	if (IS_ERR(mid)) {
 		mutex_unlock(&server->srv_mutex);
-		add_credits(server, 1, optype);
-		wake_up(&server->request_q);
+		add_credits_and_wake(server, credits, optype);
 		return PTR_ERR(mid);
 	}
 
@@ -572,8 +584,7 @@ cifs_call_async(struct TCP_Server_Info *server, struct smb_rqst *rqst,
 		return 0;
 
 	cifs_delete_mid(mid);
-	add_credits(server, 1, optype);
-	wake_up(&server->request_q);
+	add_credits_and_wake(server, credits, optype);
 	return rc;
 }
 
Signature · 1 line
1.7.10.4
#17 lore raw
In reply to a message not in the archive

Partially merged (first 5 ie those which merged cleanly) into cifs-2.6.git for-next (for further testing).

Quoted 33 lines · Pavel Shilovsky

The goal of this patchset is to add large MTU/multicredits support for writing from the CIFS client to servers that support SMB 2.1 and higher versions of the protocol.

First three patches simplify the writepages code. Patches #4 and #5 both fix the problem in writepages when a server changes wsize value after reconnects from the client. Patches #6 and #7 simplify the iovec write code, #8 fixes the wsize changing problem there. Finally, patch #9 introduces multicredits support for writing.

Pavel Shilovsky (9):
CIFS: Separate page processing from writepages
CIFS: Separate page sending from writepages
CIFS: Separate pages initialization from writepages
CIFS: Fix wsize usage in writepages
CIFS: Fix cifs_writev_requeue when wsize changes
CIFS: Separate filling pages from iovec write
CIFS: Separate writing from iovec write
CIFS: Fix wsize usage in iovec write
CIFS: Use multicredits for writes

 fs/cifs/cifsglob.h      |   17 ++
 fs/cifs/cifsproto.h     |    3 +
 fs/cifs/cifssmb.c       |   84 ++++++--
 fs/cifs/file.c          |  540 ++++++++++++++++++++++++++++-------------------
 fs/cifs/smb1ops.c       |    8 +
 fs/cifs/smb2ops.c       |   62 +++++-
 fs/cifs/smb2pdu.c       |   24 ++-
 fs/cifs/smb2transport.c |    5 +
 fs/cifs/transport.c     |   25 ++-
 9 files changed, 517 insertions(+), 251 deletions(-)

--
1.7.10.4

--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 3 lines
Thanks,

Steve