diff options
Diffstat (limited to 'src/user/app/netstack/fs.c')
-rw-r--r-- | src/user/app/netstack/fs.c | 50 |
1 files changed, 25 insertions, 25 deletions
diff --git a/src/user/app/netstack/fs.c b/src/user/app/netstack/fs.c index ad6c23c..6d51c35 100644 --- a/src/user/app/netstack/fs.c +++ b/src/user/app/netstack/fs.c @@ -43,14 +43,14 @@ struct handle { size_t readcap; } tcp; bool dead; - handle_t reqh; + hid_t reqh; }; static void tcp_listen_callback(struct tcp_conn *c, void *arg) { struct handle *h = arg; h->tcp.c = c; - _syscall_fs_respond(h->reqh, h, 0, 0); + _sys_fs_respond(h->reqh, h, 0, 0); h->reqh = -1; } @@ -63,7 +63,7 @@ static void tcp_recv_callback(void *arg) { h->tcp.readcap = sizeof buf; size_t len = tcpc_tryread(h->tcp.c, buf, h->tcp.readcap); if (len > 0) { - _syscall_fs_respond(h->reqh, buf, len, 0); + _sys_fs_respond(h->reqh, buf, len, 0); h->reqh = -1; } } @@ -73,7 +73,7 @@ static void tcp_close_callback(void *arg) { struct handle *h = arg; h->dead = true; if (h->reqh >= 0) { - _syscall_fs_respond(h->reqh, NULL, -ECONNRESET, 0); + _sys_fs_respond(h->reqh, NULL, -ECONNRESET, 0); h->reqh = -1; return; } @@ -82,14 +82,14 @@ static void tcp_close_callback(void *arg) { static void udp_listen_callback(struct udp_conn *c, void *arg) { struct handle *h = arg; h->udp.c = c; - _syscall_fs_respond(h->reqh, h, 0, 0); + _sys_fs_respond(h->reqh, h, 0, 0); h->reqh = -1; } static void udp_recv_callback(const void *buf, size_t len, void *arg) { struct handle *h = arg; if (h->reqh >= 0) { - _syscall_fs_respond(h->reqh, buf, len, 0); + _sys_fs_respond(h->reqh, buf, len, 0); h->reqh = -1; return; } @@ -106,14 +106,14 @@ static void udp_recv_callback(const void *buf, size_t len, void *arg) { } } -static void recv_enqueue(struct handle *h, handle_t reqh, size_t readcap) { +static void recv_enqueue(struct handle *h, hid_t reqh, size_t readcap) { if (h->reqh > 0) { // TODO queue - _syscall_fs_respond(reqh, NULL, -1, 0); + _sys_fs_respond(reqh, NULL, -1, 0); return; } if (h->type == H_UDP && h->udp.rx) { - _syscall_fs_respond(reqh, h->udp.rx->buf, h->udp.rx->len, 0); + _sys_fs_respond(reqh, h->udp.rx->buf, h->udp.rx->len, 0); h->udp.rx = h->udp.rx->next; free(h->udp.rx); return; @@ -125,8 +125,8 @@ static void recv_enqueue(struct handle *h, handle_t reqh, size_t readcap) { } } -static void fs_open(handle_t reqh, char *path, int flags) { -#define respond(buf, val) do{ _syscall_fs_respond(reqh, buf, val, 0); return; }while(0) +static void fs_open(hid_t reqh, char *path, int flags) { +#define respond(buf, val) do{ _sys_fs_respond(reqh, buf, val, 0); return; }while(0) struct handle *h; if (*path != '/') respond(NULL, -1); path++; @@ -241,7 +241,7 @@ void fs_thread(void *arg) { (void)arg; char *buf = malloc(buflen); for (;;) { struct ufs_request res; - handle_t reqh = _syscall_fs_wait(buf, buflen, &res); + hid_t reqh = _sys_fs_wait(buf, buflen, &res); if (reqh < 0) break; struct handle *h = res.id; long ret; @@ -251,12 +251,12 @@ void fs_thread(void *arg) { (void)arg; buf[res.len] = '\0'; fs_open(reqh, buf, res.flags); } else { - _syscall_fs_respond(reqh, NULL, -1, 0); + _sys_fs_respond(reqh, NULL, -1, 0); } break; case VFSOP_READ: if (h->dead) { - _syscall_fs_respond(reqh, NULL, -ECONNRESET, 0); + _sys_fs_respond(reqh, NULL, -ECONNRESET, 0); break; } switch (h->type) { @@ -275,44 +275,44 @@ void fs_thread(void *arg) { (void)arg; arp_fsread(reqh, res.offset); break; default: - _syscall_fs_respond(reqh, NULL, -1, 0); + _sys_fs_respond(reqh, NULL, -1, 0); } break; case VFSOP_WRITE: if (h->dead) { - _syscall_fs_respond(reqh, NULL, -ECONNRESET, 0); + _sys_fs_respond(reqh, NULL, -ECONNRESET, 0); break; } switch (h->type) { case H_ETHER: - ret = _syscall_write(state.raw_h, buf, res.len, 0, 0); - _syscall_fs_respond(reqh, NULL, ret, 0); + ret = _sys_write(state.raw_h, buf, res.len, 0, 0); + _sys_fs_respond(reqh, NULL, ret, 0); break; case H_TCP: tcpc_send(h->tcp.c, buf, res.len); - _syscall_fs_respond(reqh, NULL, res.len, 0); + _sys_fs_respond(reqh, NULL, res.len, 0); break; case H_UDP: udpc_send(h->udp.c, buf, res.len); - _syscall_fs_respond(reqh, NULL, res.len, 0); + _sys_fs_respond(reqh, NULL, res.len, 0); break; case H_ARP: - _syscall_fs_respond(reqh, NULL, arp_fswrite(buf, res.len, res.offset), 0); + _sys_fs_respond(reqh, NULL, arp_fswrite(buf, res.len, res.offset), 0); break; default: - _syscall_fs_respond(reqh, NULL, -1, 0); + _sys_fs_respond(reqh, NULL, -1, 0); } break; case VFSOP_CLOSE: // TODO remove entries in queue - // TODO why does close even have _syscall_fs_respond? + // TODO why does close even have _sys_fs_respond? if (h->type == H_TCP) tcpc_close(h->tcp.c); if (h->type == H_UDP) udpc_close(h->udp.c); free(h); - _syscall_fs_respond(reqh, NULL, -1, 0); + _sys_fs_respond(reqh, NULL, -1, 0); break; default: - _syscall_fs_respond(reqh, NULL, -1, 0); + _sys_fs_respond(reqh, NULL, -1, 0); break; } } |