Mekhanik evgenii/1411 7

fw/hpack.c

@@ -30,6 +30,41 @@
 
 #include "hpack_tbl.h"
 
+/**
+ * There are to cases when we allocate space for header:
+ * - it->rspace is equal to zero. In this case we allocate
+ *   `len` bytes;
+ * - it->rspace is not equal to zero. This means that we
+ *   already have buffer with rspace bytes available.
+ *   If this buffer is rather big to contains the whole
+ *   header or it is greater then sizeof(TfwStr) we use
+ *   it. (Save delta = len - it->rspace bytes. Later when
+ *   `it->rspace` will be exceeded we use it to allocate
+ *   new chunk).
+ *   Otherwise allocate new buffer (if we have some free
+ *   bytes in old buffer, but is is not enough for the
+ *   new header and is less then sizeof(TfwStr)), we allocate
+ *   new buffer to prevent extra header chunking.
+ */
+#define	BUFFER_GET(len, it)						\
+do {									\
+									\
+									\
+	BUG_ON(!(len));							\
+	if (!(it)->rspace ||						\
+	    (len > (it)->rspace && (it)->rspace < sizeof(TfwStr)))	\
+	{								\
+		(it)->rspace = len;					\
+		(it)->pos = tfw_pool_alloc_not_align((it)->pool, len);	\
+		T_DBG3("%s: get buffer, len=%lu, it->pos=[%p],"		\
+		       " it->pos=%lu\n", __func__, (unsigned long)len,	\
+		       (it)->pos, (unsigned long)(it)->pos);		\
+	} else {							\
+		(it)->to_alloc = len > (it)->rspace ?			\
+			len - (it)->rspace : 0;				\
+	}								\
+} while (0)
+
 #define HP_HDR_NAME(name)						\
 	(&(TfwStr){							\
 		.chunks = &(TfwStr){					\
@@ -278,23 +313,32 @@ do {								\
 	       last - src);					\
 } while (0)
 
-#define	BUFFER_HDR_INIT(length, it)				\
+#define	BUFFER_HDR_INIT(it)					\
 do {								\
 	(it)->hdr.data = (it)->pos;				\
-	(it)->hdr.len = length;					\
+	(it)->hdr.len = 0;					\
 	(it)->next = 0;						\
 } while (0)
 
 #define	BUFFER_NAME_OPEN(length)				\
 do {								\
 	WARN_ON_ONCE(!TFW_STR_EMPTY(&it->hdr));			\
 	if (state & HPACK_FLAGS_HUFFMAN_NAME) {			\
-		BUFFER_GET(length, it);				\
+		/*						\
+		 * Allocate extra 50% bytes. Huffman usually	\
+		 * gives compression benefit about 20 - 60 %.	\
+		 * Since we use extra allocated bytes for the	\
+		 * next header, we can allocate extra 50% bytes	\
+		 * without fear of losing a lot of memory.	\
+		 */						\
+		unsigned long len = length + (length >> 1);	\
+								\
+		BUFFER_GET(len, it);				\
 		if (!it->pos) {					\
 			r = -ENOMEM;				\
 			goto out;				\
 		}						\
-		BUFFER_HDR_INIT(length, it);			\
+		BUFFER_HDR_INIT(it);				\
 	}							\
 } while (0)
 
@@ -306,19 +350,27 @@ do {								\
 		? it->parsed_hdr->nchunks			\
 		: 1;						\
 	if (state & HPACK_FLAGS_HUFFMAN_VALUE) {		\
-		BUFFER_GET(length, it);				\
+		/*						\
+		 * Allocate extra 50% bytes. Huffman usually	\
+		 * gives compression benefit about 20 - 60 %.	\
+		 * Since we use extra allocated bytes for the	\
+		 * next header, we can allocate extra 50% bytes	\
+		 * without fear of losing a lot of memory.	\
+		 */						\
+		unsigned long len = length + (length >> 1);	\
+								\
+		BUFFER_GET(len, it);				\
 		if (!it->pos) {					\
 			r = -ENOMEM;				\
 			goto out;				\
 		}						\
 		if (!TFW_STR_EMPTY(&it->hdr)) {			\
-			r = tfw_hpack_exp_hdr(req->pool, length, \
-					      it); 		\
+			r = tfw_hpack_exp_hdr(req->pool, 0, it); \
 			if (unlikely(r))			\
 				return r;			\
 			it->next = it->hdr.nchunks - 1;		\
 		} else	{					\
-			BUFFER_HDR_INIT(length, it);		\
+			BUFFER_HDR_INIT(it);			\
 		}						\
 	}							\
 } while (0)
@@ -456,41 +508,61 @@ static inline int
 tfw_hpack_huffman_write(char sym, TfwHttpReq *__restrict req)
 {
 	TfwMsgParseIter *it = &req->pit;
+	unsigned long to_alloc;
 	bool np;
 	int r;
 
+#define ADJUST_HDR_LEN(it)						\
+do {									\
+	TfwStr *hdr = &it->hdr;						\
+	TfwStr *last = TFW_STR_LAST(hdr);				\
+									\
+	T_DBG3("%s: add to hdr, hdr->len=%lu, last->len=%lu,"		\
+	       " last->data=%.*s\n", __func__, hdr->len, last->len,	\
+	       (int)last->len, last->data);				\
+									\
+	++hdr->len;							\
+	if (!TFW_STR_PLAIN(hdr))					\
+		++last->len;						\
+} while (0)
+
 	if (it->rspace) {
 		--it->rspace;
 		*it->pos++ = sym;
+		ADJUST_HDR_LEN(it);
 		return 0;
 	}
 
-	if (!(it->pos = tfw_pool_alloc_not_align_np(it->pool, 1, &np)))
-		return -ENOMEM;
+	to_alloc = it->to_alloc ? it->to_alloc : 1;
+	if (to_alloc > 1)
+		it->rspace = to_alloc - 1;
 
-	*it->pos = sym;
+	it->pos = tfw_pool_alloc_not_align_np(it->pool, to_alloc, &np);
+	if (!it->pos)
+		return -ENOMEM;
 
 	T_DBG3("%s: it->rspace=%lu, sym=%c, np=%d\n", __func__,
 	       it->rspace, sym, np);
 
-	if (!np) {
-		TfwStr *hdr = &it->hdr;
-		TfwStr *last = TFW_STR_LAST(hdr);
-
-		T_DBG3("%s: add to hdr, hdr->len=%lu, last->len=%lu,"
-		       " last->data=%.*s\n", __func__, hdr->len, last->len,
-		       (int)last->len, last->data);
-
-		++hdr->len;
-		if (!TFW_STR_PLAIN(hdr))
-			++last->len;
+	/*
+	 * If the new page was allocated or it->pos points to
+	 * the buffer, which was allocated to store previously
+	 * decoded header (it->to_alloc != 0) we should expand
+	 * header.
+	 */
+	if (!np && !it->to_alloc) {
+		*it->pos++ = sym;
+		ADJUST_HDR_LEN(it);
 		return 0;
 	}
 
 	r = tfw_hpack_exp_hdr(req->pool, 1, it);
 	if (unlikely(r))
 		return r;
 
+	*it->pos++ = sym;
+	it->to_alloc = 0;
+
 	return 0;
 }
 
@@ -536,9 +608,8 @@ huffman_decode_tail(TfwHPack *__restrict hp, TfwHttpReq *__restrict req,
 			 */
 			if (likely(offset == 0)) {
 				if ((i ^ (HT_EOS_HIGH >> 1)) <
-				    (1U << -hp->curr)) {
-					return 0;
-				}
+				    (1U << -hp->curr))
+					return T_OK;
 			}
 			/*
 			 * The first condition here equivalent to the
@@ -562,11 +633,12 @@ huffman_decode_tail(TfwHPack *__restrict hp, TfwHttpReq *__restrict req,
 		}
 	}
 	if (likely(offset == 0)) {
-		if ((i ^ (HT_EOS_HIGH >> 1)) < (1U << -hp->curr)) {
+		if ((i ^ (HT_EOS_HIGH >> 1)) < (1U << -hp->curr))
 			return T_OK;
-		}
 	}
 	return T_COMPRESSION;
+
+#undef ADJUST_EXTRA_RSPACE
 }
 
 static int

fw/hpack.h

@@ -296,17 +296,6 @@ typedef struct {
 	TfwStr			hdr_data;
 } TfwDecodeCacheIter;
 
-#define	BUFFER_GET(len, it)					\
-do {								\
-	BUG_ON(!(len));						\
-	WARN_ON_ONCE((it)->rspace);				\
-	(it)->rspace = len;					\
-	(it)->pos = tfw_pool_alloc_not_align((it)->pool, len);	\
-	T_DBG3("%s: get buffer, len=%lu, it->pos=[%p],"		\
-	       " it->pos=%lu\n", __func__, (unsigned long)len,	\
-	       (it)->pos, (unsigned long)(it)->pos);		\
-} while (0)
-
 void write_int(unsigned long index, unsigned short max, unsigned short mask,
 	       TfwHPackInt *__restrict res_idx);
 int tfw_hpack_init(TfwHPack *__restrict hp, unsigned int htbl_sz);

fw/msg.h

@@ -67,14 +67,16 @@ typedef struct {
  * @pool	- allocation pool for target buffer of decoded headers;
  * @parsed_hdr	- pointer to the message header which is currently processed;
  * @hdrs_len	- accumulated length of message's decoded and parsed headers;
+ * @rspace	- space remained in the allocated chunk;
+ * @pos		- pointer to the currently allocated chunk of decoded headers'
+ *		  buffer;
  * @hdrs_cnt	- count of all headers from message headers block;
  * @__off	- offset for iterator reinitializing before next processing
  *		  stage;
  * @hdr		- descriptor of currently decoded header in target buffer;
- * @pos		- pointer to the currently allocated chunk of decoded headers'
- *		  buffer;
- * @rspace	- space remained in the allocated chunk;
  * @nm_len	- length of the decoded header's name;
+ * @to_alloc	- count of bytes which, should be allocated, after rspace
+ *		  will be exceeded;
  * @nm_num	- chunks number of the decoded header's name;
  * @tag		- tag of currently processed decoded header.
  * @next	- number of chunk which points to the decoded header part
@@ -84,12 +86,13 @@ typedef struct {
 	TfwPool		*pool;
 	TfwStr		*parsed_hdr;
 	unsigned long	hdrs_len;
+	unsigned long	rspace;
+	char		*pos;
 	unsigned int	hdrs_cnt;
 	char		__off[0];
 	TfwStr		hdr;
-	char		*pos;
-	unsigned long	rspace;
 	unsigned long	nm_len;
+	unsigned long	to_alloc;
 	unsigned int	nm_num;
 	unsigned int	tag;
 	unsigned int	next;