Commit 7632e2494a5b8bfea41e7fbfaa0fc324e2178932

Carlos Martín Nieto 2011-06-13T23:01:12

Correctly handle network input Add a parameter to git_pkt_parse_line to tell it how much data you have in your buffer. If the buffer is too short, it returns an error saying so. Adapt the git transport to use this and fix the offset calculation. Add the GIT_ESHORTBUFFER error code.

diff --git a/include/git2/errors.h b/include/git2/errors.h
index 334b9ed..03c74e8 100644
--- a/include/git2/errors.h
+++ b/include/git2/errors.h
@@ -128,6 +128,9 @@ typedef enum {
 
 	/** The path pattern and string did not match */
 	GIT_ENOMATCH = -31,
+
+	/**  The buffer is too short to satisfy the request */
+	GIT_ESHORTBUFFER = -32,
 } git_error;
 
 /**
diff --git a/include/git2/pkt.h b/include/git2/pkt.h
index d0ffa55..79c5828 100644
--- a/include/git2/pkt.h
+++ b/include/git2/pkt.h
@@ -55,5 +55,5 @@ struct git_pkt_ref {
  * Create a git protocol request.
  */
 int git_pkt_gen_proto(char **out, int *outlen, const char *url);
-int git_pkt_parse_line(git_pkt **head, const char *line, const char **out);
+int git_pkt_parse_line(git_pkt **head, const char *line, const char **out, unsigned int len);
 void git_pkt_free(git_pkt *pkt);
diff --git a/src/pkt.c b/src/pkt.c
index f77bc30..17edf32 100644
--- a/src/pkt.c
+++ b/src/pkt.c
@@ -30,6 +30,10 @@
 #include "common.h"
 #include "util.h"
 
+#include <ctype.h>
+
+#define PKT_LEN_SIZE 4
+
 static int flush_pkt(git_pkt **out)
 {
 	git_pkt *pkt;
@@ -50,8 +54,7 @@ static int flush_pkt(git_pkt **out)
 int ref_pkt(git_pkt **out, const char *line, size_t len)
 {
 	git_pkt_ref *pkt;
-	int error;
-	size_t name_len;
+	int error, has_caps = 0;
 
 	pkt = git__malloc(sizeof(git_pkt_ref));
 	if (pkt == NULL)
@@ -75,30 +78,22 @@ int ref_pkt(git_pkt **out, const char *line, size_t len)
 	line += GIT_OID_HEXSZ + 1;
 	len -= (GIT_OID_HEXSZ + 1);
 
-	name_len = min(strlen(line), len);
-	if (line[name_len - 1] == '\n')
-		--name_len;
+	if (strlen(line) < len)
+		has_caps = 1;
+
+	if (line[len - 1] == '\n')
+		--len;
 
-	pkt->head.name = git__strndup(line, name_len);
+	pkt->head.name = git__malloc(len + 1);
 	if (pkt->head.name == NULL) {
 		error = GIT_ENOMEM;
 		goto out;
 	}
+	memcpy(pkt->head.name, line, len);
+	pkt->head.name[len] = '\0';
 
-	/* Try to get the capabilities */
-	line += name_len + 1; /* + \0 */
-	len -= (name_len + 1);
-	if (line[len - 1] == '\n')
-		--len;
-
-	if (len > 0) { /* capatilities */
-		pkt->capabilities = git__malloc(len);
-		if (pkt->capabilities == NULL) {
-			error = GIT_ENOMEM;
-			goto out;
-		}
-
-		memcpy(pkt->capabilities, line, len);
+	if (has_caps) {
+		pkt->capabilities = strchr(pkt->head.name, '\0') + 1;
 	}
 
 out:
@@ -110,6 +105,29 @@ out:
 	return error;
 }
 
+static unsigned int parse_len(const char *line)
+{
+	char num[PKT_LEN_SIZE + 1];
+	int i, error;
+	long len;
+	const char *num_end;
+
+	memcpy(num, line, PKT_LEN_SIZE);
+	num[PKT_LEN_SIZE] = '\0';
+
+	for (i = 0; i < PKT_LEN_SIZE; ++i) {
+		if (!isxdigit(num[i]))
+			return GIT_ENOTNUM;
+	}
+
+	error = git__strtol32(&len, num, &num_end, 16);
+	if (error < GIT_SUCCESS) {
+		return error;
+	}
+
+	return (unsigned int) len;
+}
+
 /*
  * As per the documentation, the syntax is:
  *
@@ -123,35 +141,34 @@ out:
  * in ASCII hexadecimal (including itself)
  */
 
-int git_pkt_parse_line(git_pkt **head, const char *line, const char **out)
+int git_pkt_parse_line(git_pkt **head, const char *line, const char **out, unsigned int bufflen)
 {
 	int error = GIT_SUCCESS;
-	long int len;
-	const int num_len = 4;
-	char *num;
-	const char *num_end;
+	unsigned int len;
 
-	num = git__strndup(line, num_len);
-	if (num == NULL)
-		return GIT_ENOMEM;
+	/* Not even enough for the length */
+	if (bufflen > 0 && bufflen < PKT_LEN_SIZE)
+		return GIT_ESHORTBUFFER;
 
-	error = git__strtol32(&len, num, &num_end, 16);
-	if (error < GIT_SUCCESS) {
-		free(num);
+	error = parse_len(line);
+	if (error < GIT_SUCCESS)
 		return git__throw(error, "Failed to parse pkt length");
-	}
-	if (num_end - num != num_len) {
-		free(num);
-		return git__throw(GIT_EOBJCORRUPTED, "Wrong pkt length");
-	}
-	free(num);
 
-	line += num_len;
+	len = error;
+
+	/*
+	 * If we were given a buffer length, then make sure there is
+	 * enough in the buffer to satisfy this line
+	 */
+	if (bufflen > 0 && bufflen < len)
+		return GIT_ESHORTBUFFER;
+
+	line += PKT_LEN_SIZE;
 	/*
 	 * TODO: How do we deal with empty lines? Try again? with the next
 	 * line?
 	 */
-	if (len == 4) {
+	if (len == PKT_LEN_SIZE) {
 		*out = line;
 		return GIT_SUCCESS;
 	}
@@ -161,7 +178,7 @@ int git_pkt_parse_line(git_pkt **head, const char *line, const char **out)
 		return flush_pkt(head);
 	}
 
-	len -= num_len; /* the length includes the space for the length */
+	len -= PKT_LEN_SIZE; /* the encoded length includes its own size */
 
 	/*
 	 * For now, we're just going to assume we're parsing references
@@ -177,7 +194,6 @@ void git_pkt_free(git_pkt *pkt)
 {
 	if(pkt->type == GIT_PKT_REF) {
 		git_pkt_ref *p = (git_pkt_ref *) pkt;
-		free(p->capabilities);
 		free(p->head.name);
 	}
 
diff --git a/src/transport_git.c b/src/transport_git.c
index ed72031..e5c7b1d 100644
--- a/src/transport_git.c
+++ b/src/transport_git.c
@@ -157,46 +157,58 @@ static int store_refs(git_priv *priv)
 	int s = priv->socket;
 	git_vector *refs = &priv->refs;
 	int error = GIT_SUCCESS;
-	char buffer[1024] = {0};
+	char buffer[1024];
 	const char *line_end, *ptr;
 	int off = 0, ret;
+	unsigned int bufflen = 0;
 	git_pkt *pkt;
 
+	memset(buffer, 0x0, sizeof(buffer));
+
 	while (1) {
-		ret = recv(s, buffer, sizeof(buffer) - off, 0);
+		ret = recv(s, buffer + off, sizeof(buffer) - off, 0);
 		if (ret < 0)
 			return git__throw(GIT_EOSERR, "Failed to receive data");
-		if (ret == 0)
-			puts("got zero!");
+		if (ret == 0) /* Orderly shutdown, so exit */
+			return GIT_SUCCESS;
 
+		bufflen += ret;
 		ptr = buffer;
 		while (1) {
-			error = git_pkt_parse_line(&pkt, ptr, &line_end);
+			if (bufflen == 0)
+				break;
+			error = git_pkt_parse_line(&pkt, ptr, &line_end, bufflen);
 			/*
-			 * An error here can mean that the input in corrupt or
-			 * (more likely) that the input hasn't arrived yet.
-			 *
-			 * FIXME: Add actual error checking. Right now we just ask
-			 * for more input.
+			 * If the error is GIT_ESHORTBUFFER, it means the buffer
+			 * isn't long enough to satisfy the request. Break out and
+			 * wait for more input.
+			 * On any other error, fail.
 			 */
-			if (error < GIT_SUCCESS)
+			if (error == GIT_ESHORTBUFFER) {
+				line_end = ptr;
 				break;
+			}
+			if (error < GIT_SUCCESS) {
+				return error;
+			}
 
 			error = git_vector_insert(refs, pkt);
 			if (error < GIT_SUCCESS)
 				return error;
 
-			if (pkt->type != GIT_PKT_REF)
+			if (pkt->type == GIT_PKT_FLUSH)
 				return GIT_SUCCESS;
 
+			bufflen -= line_end - ptr;
 			ptr = line_end;
 		}
 
 		/*
 		 * Move the rest to the start of the buffer
 		 */
-		memmove(buffer, line_end, sizeof(buffer) - (line_end - buffer));
-		off = ret - (line_end - buffer);
+		memmove(buffer, line_end, bufflen);
+		off = bufflen;
+		memset(buffer + off, 0x0, sizeof(buffer) - off);
 	}
 
 	return error;
@@ -271,6 +283,8 @@ static int git_close(git_transport *transport)
 	int s = priv->socket;
 	int error;
 
+	/* FIXME:  We probably want to send a flush pkt back */
+
 	error = close(s);
 	if (error < 0)
 		error = git__throw(GIT_EOSERR, "Failed to close socket");