wiplukeshu/9p-read-multi-iovec

8 files changed, 112 insertions, 42 deletions

diff --git a/lib9p/9p.generated.c b/lib9p/9p.generated.c
index 6726084..868d936 100644
--- a/lib9p/9p.generated.c
+++ b/lib9p/9p.generated.c
@@ -2259,7 +2259,11 @@ static void unmarshal_Rread([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *net_
 	net_offset += 1;
 	UNMARSHAL_U16LE(ctx, out->tag);
 	UNMARSHAL_U32LE(ctx, out->count);
-	UNMARSHAL_BYTES(ctx, out->data, out->count);
+	out->data.iov = ctx->extra;
+	ctx->extra += sizeof(out->data.iov[0]);
+	out->data.iov->iov_base = &ctx->net_bytes[ctx->net_offset];
+	out->data.iov->iov_len = out->count;
+	out->data.iovcnt = 1;
 }
 
 static void unmarshal_Twrite([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *net_bytes, void *out_buf) {
@@ -2272,7 +2276,11 @@ static void unmarshal_Twrite([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *net
 	UNMARSHAL_U32LE(ctx, out->fid);
 	UNMARSHAL_U64LE(ctx, out->offset);
 	UNMARSHAL_U32LE(ctx, out->count);
-	UNMARSHAL_BYTES(ctx, out->data, out->count);
+	out->data.iov = ctx->extra;
+	ctx->extra += sizeof(out->data.iov[0]);
+	out->data.iov->iov_base = &ctx->net_bytes[ctx->net_offset];
+	out->data.iov->iov_len = out->count;
+	out->data.iovcnt = 1;
 }
 
 static void unmarshal_Rwrite([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *net_bytes, void *out_buf) {
@@ -2772,7 +2780,11 @@ static void unmarshal_Rreaddir([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *n
 	net_offset += 1;
 	UNMARSHAL_U16LE(ctx, out->tag);
 	UNMARSHAL_U32LE(ctx, out->count);
-	UNMARSHAL_BYTES(ctx, out->data, out->count);
+	out->data.iov = ctx->extra;
+	ctx->extra += sizeof(out->data.iov[0]);
+	out->data.iov->iov_base = &ctx->net_bytes[ctx->net_offset];
+	out->data.iov->iov_len = out->count;
+	out->data.iovcnt = 1;
 }
 
 static void unmarshal_Tfsync([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *net_bytes, void *out_buf) {
@@ -2993,7 +3005,11 @@ static void unmarshal_Rsread([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *net
 	net_offset += 1;
 	UNMARSHAL_U16LE(ctx, out->tag);
 	UNMARSHAL_U32LE(ctx, out->count);
-	UNMARSHAL_BYTES(ctx, out->data, out->count);
+	out->data.iov = ctx->extra;
+	ctx->extra += sizeof(out->data.iov[0]);
+	out->data.iov->iov_base = &ctx->net_bytes[ctx->net_offset];
+	out->data.iov->iov_len = out->count;
+	out->data.iovcnt = 1;
 }
 
 static void unmarshal_Tswrite([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *net_bytes, void *out_buf) {
@@ -3012,7 +3028,11 @@ static void unmarshal_Tswrite([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *ne
 		UNMARSHAL_BYTES(ctx, out->wname[i].utf8, out->wname[i].len);
 	}
 	UNMARSHAL_U32LE(ctx, out->count);
-	UNMARSHAL_BYTES(ctx, out->data, out->count);
+	out->data.iov = ctx->extra;
+	ctx->extra += sizeof(out->data.iov[0]);
+	out->data.iov->iov_base = &ctx->net_bytes[ctx->net_offset];
+	out->data.iov->iov_len = out->count;
+	out->data.iovcnt = 1;
 }
 
 static void unmarshal_Rswrite([[gnu::unused]] struct lib9p_ctx *ctx, uint8_t *net_bytes, void *out_buf) {
@@ -3491,7 +3511,8 @@ static bool marshal_Rread(struct lib9p_ctx *ctx, struct lib9p_msg_Rread *val, st
 	MARSHAL_U8LE(ctx, 117);
 	MARSHAL_U16LE(ctx, val->tag);
 	MARSHAL_U32LE(ctx, val->count);
-	MARSHAL_BYTES_ZEROCOPY(ctx, val->data, val->count);
+	for (int iov_i = 0; iov_i < val->data->iovcnt; iov_i++)
+		MARSHAL_BYTES_ZEROCOPY(ctx, val->data->iov[iov_i].iov_base, val->data->iov[iov_i].iov_len);
 	return false;
 }
 
@@ -3512,7 +3533,8 @@ static bool marshal_Twrite(struct lib9p_ctx *ctx, struct lib9p_msg_Twrite *val,
 	MARSHAL_U32LE(ctx, val->fid);
 	MARSHAL_U64LE(ctx, val->offset);
 	MARSHAL_U32LE(ctx, val->count);
-	MARSHAL_BYTES_ZEROCOPY(ctx, val->data, val->count);
+	for (int iov_i = 0; iov_i < val->data->iovcnt; iov_i++)
+		MARSHAL_BYTES_ZEROCOPY(ctx, val->data->iov[iov_i].iov_base, val->data->iov[iov_i].iov_len);
 	return false;
 }
 
@@ -4326,7 +4348,8 @@ static bool marshal_Rreaddir(struct lib9p_ctx *ctx, struct lib9p_msg_Rreaddir *v
 	MARSHAL_U8LE(ctx, 41);
 	MARSHAL_U16LE(ctx, val->tag);
 	MARSHAL_U32LE(ctx, val->count);
-	MARSHAL_BYTES_ZEROCOPY(ctx, val->data, val->count);
+	for (int iov_i = 0; iov_i < val->data->iovcnt; iov_i++)
+		MARSHAL_BYTES_ZEROCOPY(ctx, val->data->iov[iov_i].iov_base, val->data->iov[iov_i].iov_len);
 	return false;
 }
 
@@ -4692,7 +4715,8 @@ static bool marshal_Rsread(struct lib9p_ctx *ctx, struct lib9p_msg_Rsread *val,
 	MARSHAL_U8LE(ctx, 153);
 	MARSHAL_U16LE(ctx, val->tag);
 	MARSHAL_U32LE(ctx, val->count);
-	MARSHAL_BYTES_ZEROCOPY(ctx, val->data, val->count);
+	for (int iov_i = 0; iov_i < val->data->iovcnt; iov_i++)
+		MARSHAL_BYTES_ZEROCOPY(ctx, val->data->iov[iov_i].iov_base, val->data->iov[iov_i].iov_len);
 	return false;
 }
 
@@ -4720,7 +4744,8 @@ static bool marshal_Tswrite(struct lib9p_ctx *ctx, struct lib9p_msg_Tswrite *val
 		MARSHAL_BYTES_ZEROCOPY(ctx, val->wname[i].utf8, val->wname[i].len);
 	}
 	MARSHAL_U32LE(ctx, val->count);
-	MARSHAL_BYTES_ZEROCOPY(ctx, val->data, val->count);
+	for (int iov_i = 0; iov_i < val->data->iovcnt; iov_i++)
+		MARSHAL_BYTES_ZEROCOPY(ctx, val->data->iov[iov_i].iov_base, val->data->iov[iov_i].iov_len);
 	return false;
 }
 
diff --git a/lib9p/include/lib9p/9p.generated.h b/lib9p/include/lib9p/9p.generated.h
index 7a50537..e28b1d3 100644
--- a/lib9p/include/lib9p/9p.generated.h
+++ b/lib9p/include/lib9p/9p.generated.h
@@ -8,6 +8,11 @@
 
 #include <libhw/generic/net.h> /* for struct iovec */
 
+struct _lib9p_iovec_list {
+	struct iovec *iov;
+	int           iovcnt;
+};
+
 /* config *********************************************************************/
 
 #include "config.h"
@@ -156,6 +161,7 @@ enum lib9p_msg_type { /* uint8_t */
 
 /* payload types **************************************************************/
 
+
 #if CONFIG_9P_ENABLE_9P2000 || CONFIG_9P_ENABLE_9P2000_L || CONFIG_9P_ENABLE_9P2000_e || CONFIG_9P_ENABLE_9P2000_p9p || CONFIG_9P_ENABLE_9P2000_u
 /* size = 2 ; max_iov = 1 ; max_copy = 2 */
 typedef uint16_t lib9p_tag_t;
@@ -549,9 +555,9 @@ struct lib9p_msg_Rflush {
 
 /* min_size = 11 ; exp_size = 8,203 ; max_size = 2,147,483,658 ; max_iov = 2 ; max_copy = 11 */
 struct lib9p_msg_Rread {
-	lib9p_tag_t               tag;
-	uint32_t                  count;
-	[[gnu::nonstring]] char  *data;
+	lib9p_tag_t           tag;
+	uint32_t              count;
+	struct _lib9p_iovec   data;
 };
 
 /* size = 11 ; max_iov = 1 ; max_copy = 11 */
@@ -602,9 +608,9 @@ struct lib9p_msg_Rxattrcreate {
 
 /* min_size = 11 ; exp_size = 8,203 ; max_size = 4,294,967,306 (warning: >UINT32_MAX) ; max_iov = 2 ; max_copy = 11 */
 struct lib9p_msg_Rreaddir {
-	lib9p_tag_t               tag;
-	uint32_t                  count;
-	[[gnu::nonstring]] char  *data;
+	lib9p_tag_t           tag;
+	uint32_t              count;
+	struct _lib9p_iovec   data;
 };
 
 /* size = 7 ; max_iov = 1 ; max_copy = 7 */
@@ -642,9 +648,9 @@ struct lib9p_msg_Rsession {
 
 /* min_size = 11 ; exp_size = 8,203 ; max_size = 4,294,967,306 (warning: >UINT32_MAX) ; max_iov = 2 ; max_copy = 11 */
 struct lib9p_msg_Rsread {
-	lib9p_tag_t               tag;
-	uint32_t                  count;
-	[[gnu::nonstring]] char  *data;
+	lib9p_tag_t           tag;
+	uint32_t              count;
+	struct _lib9p_iovec   data;
 };
 
 /* size = 11 ; max_iov = 1 ; max_copy = 11 */
@@ -665,11 +671,11 @@ struct lib9p_msg_Tread {
 
 /* min_size = 23 ; exp_size = 8,215 ; max_size = 2,147,483,670 ; max_iov = 2 ; max_copy = 23 */
 struct lib9p_msg_Twrite {
-	lib9p_tag_t               tag;
-	lib9p_fid_t               fid;
-	uint64_t                  offset;
-	uint32_t                  count;
-	[[gnu::nonstring]] char  *data;
+	lib9p_tag_t           tag;
+	lib9p_fid_t           fid;
+	uint64_t              offset;
+	uint32_t              count;
+	struct _lib9p_iovec   data;
 };
 
 /* size = 11 ; max_iov = 1 ; max_copy = 11 */
@@ -809,12 +815,12 @@ struct lib9p_msg_Tsread {
 
 /* min_size = 17 ; exp_size = 8,673 ; max_size = 8,589,934,607 (warning: >UINT32_MAX) ; max_iov = 2 + (CONFIG_9P_MAX_9P2000_e_WELEM * 2) ; max_copy = 17 + (CONFIG_9P_MAX_9P2000_e_WELEM * 2) */
 struct lib9p_msg_Tswrite {
-	lib9p_tag_t               tag;
-	uint32_t                  fid;
-	uint16_t                  nwname;
-	struct lib9p_s           *wname;
-	uint32_t                  count;
-	[[gnu::nonstring]] char  *data;
+	lib9p_tag_t           tag;
+	uint32_t              fid;
+	uint16_t              nwname;
+	struct lib9p_s       *wname;
+	uint32_t              count;
+	struct _lib9p_iovec   data;
 };
 
 #endif /* CONFIG_9P_ENABLE_9P2000_e */
diff --git a/lib9p/include/lib9p/srv.h b/lib9p/include/lib9p/srv.h
index ff5ebdc..ad209b8 100644
--- a/lib9p/include/lib9p/srv.h
+++ b/lib9p/include/lib9p/srv.h
@@ -43,8 +43,6 @@ lo_interface lib9p_srv_dio;
 /* FIXME: I don't like that the pointers returned by stat() and
  * pread() have to remain live after they return.  Perhaps a
  * `respond()`-callback?  But that just reads as gross in C.
- *
- * FIXME: It would be nice if pread() could return more than 1 iovec.
  */
 #define lib9p_srv_file_LO_IFACE                                                \
 	/* resource management **********************************************/ \
@@ -99,7 +97,8 @@ LO_INTERFACE(lib9p_srv_file);
 	LO_FUNC(void                       , pread  , struct lib9p_srv_ctx *,  \
 	                                              uint32_t byte_count,     \
 	                                              uint64_t byte_offset,    \
-	                                              struct iovec *ret)       \
+	                                              struct iovec *ret_iov,   \
+	                                              int *ret_iovcnt)         \
 	LO_FUNC(uint32_t                   , pwrite , struct lib9p_srv_ctx *,  \
 	                                              void *buf,               \
 	                                              uint32_t byte_count,     \
diff --git a/lib9p/protogen/c9util.py b/lib9p/protogen/c9util.py
index e7ad999..10d1c8c 100644
--- a/lib9p/protogen/c9util.py
+++ b/lib9p/protogen/c9util.py
@@ -76,6 +76,8 @@ def typename(typ: idl.Type, parent: idl.StructMember | None = None) -> str:
     match typ:
         case idl.Primitive():
             if typ.value == 1 and parent and parent.cnt:  # SPECIAL (string)
+                if parent.membname == "data":  # SPECIAL (zerocopy)
+                    return f"struct {ident('_iovec')}"
                 return "[[gnu::nonstring]] char"
             return f"uint{typ.value*8}_t"
         case idl.Number():
diff --git a/lib9p/protogen/c_marshal.py b/lib9p/protogen/c_marshal.py
index 74b64f5..87ba50b 100644
--- a/lib9p/protogen/c_marshal.py
+++ b/lib9p/protogen/c_marshal.py
@@ -318,7 +318,14 @@ def gen_c_marshal(versions: set[str], typs: list[idl.UserType]) -> str:
                 if path.root.typname == "stat":  # SPECIAL (stat)
                     ret += f"{'\t'*indent_lvl()}MARSHAL_BYTES(ctx, {path.c_str('val->')[:-3]}, {cnt_path.c_str('val->')});\n"
                 else:
-                    ret += f"{'\t'*indent_lvl()}MARSHAL_BYTES_ZEROCOPY(ctx, {path.c_str('val->')[:-3]}, {cnt_path.c_str('val->')});\n"
+                    if (
+                        c9util.typename(child.typ, child)
+                        == f"struct {c9util.ident("_iovec")}"
+                    ):
+                        ret += f"{'\t'*indent_lvl()}for (int iov_i = 0; iov_i < {path.c_str('val->')[:-3]}->iovcnt; iov_i++)\n"
+                        ret += f"{'\t'*(indent_lvl()+1)}MARSHAL_BYTES_ZEROCOPY(ctx, {path.c_str('val->')[:-3]}->iov[iov_i].iov_base, {path.c_str('val->')[:-3]}->iov[iov_i].iov_len);\n"
+                    else:
+                        ret += f"{'\t'*indent_lvl()}MARSHAL_BYTES_ZEROCOPY(ctx, {path.c_str('val->')[:-3]}, {cnt_path.c_str('val->')});\n"
                 return idlutil.WalkCmd.KEEP_GOING, pop
             loopvar = chr(ord("i") + loopdepth - 1)
             ret += f"{'\t'*indent_lvl()}for ({c9util.typename(child.cnt.typ)} {loopvar} = 0; {loopvar} < {cnt_path.c_str('val->')}; {loopvar}++) {{\n"
diff --git a/lib9p/protogen/c_unmarshal.py b/lib9p/protogen/c_unmarshal.py
index 018d750..f14fb55 100644
--- a/lib9p/protogen/c_unmarshal.py
+++ b/lib9p/protogen/c_unmarshal.py
@@ -95,7 +95,19 @@ def gen_c_unmarshal(versions: set[str], typs: list[idl.UserType]) -> str:
         if child.cnt:
             cnt_path = path.parent().add(child.cnt)
             if child.typ.static_size == 1:  # SPECIAL (zerocopy)
-                ret += f"{'\t'*indent_lvl()}UNMARSHAL_BYTES(ctx, {path.c_str('out->')[:-3]}, {cnt_path.c_str('out->')});\n"
+                if (
+                    c9util.typename(child.typ, child)
+                    == f"struct {c9util.ident('_iovec')}"
+                ):  # SPECIAL(zerocopy)
+                    ret += (
+                        f"{'\t'*indent_lvl()}out->{child.membname}.iov = ctx->extra;\n"
+                    )
+                    ret += f"{'\t'*indent_lvl()}ctx->extra += sizeof(out->{child.membname}.iov[0]);\n"
+                    ret += f"{'\t'*indent_lvl()}out->{child.membname}.iov->iov_base = &ctx->net_bytes[ctx->net_offset];\n"
+                    ret += f"{'\t'*indent_lvl()}out->{child.membname}.iov->iov_len = out->{child.cnt.membname};\n"
+                    ret += f"{'\t'*indent_lvl()}out->{child.membname}.iovcnt = 1;\n"
+                else:
+                    ret += f"{'\t'*indent_lvl()}UNMARSHAL_BYTES(ctx, {path.c_str('out->')[:-3]}, {cnt_path.c_str('out->')});\n"
                 return idlutil.WalkCmd.KEEP_GOING, pop
             ret += f"{'\t'*indent_lvl()}{path.c_str('out->')[:-3]} = extra;\n"
             ret += f"{'\t'*indent_lvl()}extra += sizeof({path.c_str('out->')[:-3]}[0]) * {cnt_path.c_str('out->')};\n"
diff --git a/lib9p/protogen/h.py b/lib9p/protogen/h.py
index 7785ca1..3e743e9 100644
--- a/lib9p/protogen/h.py
+++ b/lib9p/protogen/h.py
@@ -166,6 +166,11 @@ def gen_h(versions: set[str], typs: list[idl.UserType]) -> str:
 #include <stdint.h> /* for uint{{n}}_t types */
 
 #include <libhw/generic/net.h> /* for struct iovec */
+
+struct {c9util.ident('_iovec_list')} {{
+\tstruct iovec *iov;
+\tint           iovcnt;
+}};
 """
 
     id2typ: dict[int, idl.Message] = {}
@@ -224,6 +229,7 @@ enum {c9util.ident('version')} {{
 
     ret += """
 /* payload types **************************************************************/
+
 """
 
     def per_version_comment(
@@ -357,7 +363,13 @@ enum {c9util.ident('version')} {{
                     if member.val:
                         continue
                     ret += cutil.ifdef_push(2, c9util.ver_ifdef(member.in_versions))
-                    ret += f"\t{c9util.typename(member.typ, member):<{typewidth}}  {'*' if member.cnt else ' '}{member.membname};\n"
+                    ptr = bool(member.cnt)
+                    if (
+                        c9util.typename(member.typ, member)
+                        == f"struct {c9util.ident('_iovec')}"
+                    ):  # SPECIAL (zerocopy)
+                        ptr = False
+                    ret += f"\t{c9util.typename(member.typ, member):<{typewidth}}  {'*' if ptr else ' '}{member.membname};\n"
                 ret += cutil.ifdef_pop(1)
                 ret += "};\n"
         del typ
diff --git a/lib9p/srv.c b/lib9p/srv.c
index 50d0b78..2c26ffe 100644
--- a/lib9p/srv.c
+++ b/lib9p/srv.c
@@ -980,13 +980,20 @@ static void handle_Tread(struct _lib9p_srv_req *ctx,
 		fidinfo->dir.idx = idx+num;
 		fidinfo->dir.off = req->offset + len;
 	} else {
-		struct iovec iov;
-		LO_CALL(fidinfo->file.io, pread, &ctx->ctx, req->count, req->offset, &iov);
+		LO_CALL(fidinfo->file.io, pread, &ctx->ctx, req->count, req->offset,
+		        &resp->iov, &resp->iovcnt);
 		if (!lib9p_ctx_has_error(&ctx->ctx.basectx)) {
-			resp->count = iov.iov_len;
-			resp->data = iov.iov_base;
-			if (resp->count > req->count)
-				resp->count = req->count;
+			resp->count = 0;
+			assert(resp->iovcnt >= 0);
+			if (resp->iovcnt) {
+				assert(resp->iov);
+				for (int i = 0; i < resp->iovcnt; i++) {
+					assert(resp->iov[i].iov_len == 0 || resp->iov[i].iov_base);
+					bool overflowed = __builtin_add_overflow(resp->count, resp->iov[i].iov_len, &resp->count);
+					assert(!overflowed);
+				}
+			}
+			assert(resp->count <= req->count);
 		}
 	}
 }