make stdio open, read, and write operations cancellation points

it should be noted that only the actual underlying buffer flush and
fill operations are cancellable, not reads from or writes to the
buffer. this behavior is compatible with POSIX, which makes all
cancellation points in stdio optional, and it achieves the goal of
allowing cancellation of a thread that's "stuck" on IO (due to a
non-responsive socket/pipe peer, slow/stuck hardware, etc.) without
imposing any measurable performance cost.

3 files changed, 28 insertions, 5 deletions

diff --git a/src/stdio/__stdio_read.c b/src/stdio/__stdio_read.c
index 218bd88d..ee17a576 100644
--- a/src/stdio/__stdio_read.c
+++ b/src/stdio/__stdio_read.c
@@ -1,4 +1,11 @@
 #include "stdio_impl.h"
+#include <pthread.h>
+
+static void cleanup(void *p)
+{
+	FILE *f = p;
+	if (!f->lockcount) __unlockfile(f);
+}
 
 size_t __stdio_read(FILE *f, unsigned char *buf, size_t len)
 {
@@ -8,7 +15,9 @@ size_t __stdio_read(FILE *f, unsigned char *buf, size_t len)
 	};
 	ssize_t cnt;
 
-	cnt = syscall(SYS_readv, f->fd, iov, 2);
+	pthread_cleanup_push(cleanup, f);
+	cnt = syscall_cp(SYS_readv, f->fd, iov, 2);
+	pthread_cleanup_pop(0);
 	if (cnt <= 0) {
 		f->flags |= F_EOF ^ ((F_ERR^F_EOF) & cnt);
 		f->rpos = f->rend = 0;
diff --git a/src/stdio/__stdio_write.c b/src/stdio/__stdio_write.c
index 63d9c858..dd97cf68 100644
--- a/src/stdio/__stdio_write.c
+++ b/src/stdio/__stdio_write.c
@@ -1,4 +1,11 @@
 #include "stdio_impl.h"
+#include <pthread.h>
+
+static void cleanup(void *p)
+{
+	FILE *f = p;
+	if (!f->lockcount) __unlockfile(f);
+}
 
 size_t __stdio_write(FILE *f, const unsigned char *buf, size_t len)
 {
@@ -10,10 +17,14 @@ size_t __stdio_write(FILE *f, const unsigned char *buf, size_t len)
 	size_t rem = iov[0].iov_len + iov[1].iov_len;
 	int iovcnt = 2;
 	ssize_t cnt;
-	f->wpos = f->wbase;
 	for (;;) {
-		cnt = syscall(SYS_writev, f->fd, iov, iovcnt);
-		if (cnt == rem) return len;
+		pthread_cleanup_push(cleanup, f);
+		cnt = syscall_cp(SYS_writev, f->fd, iov, iovcnt);
+		pthread_cleanup_pop(0);
+		if (cnt == rem) {
+			f->wpos = f->wbase = f->buf;
+			return len;
+		}
 		if (cnt < 0) {
 			f->wpos = f->wbase = f->wend = 0;
 			f->flags |= F_ERR;
@@ -21,8 +32,11 @@ size_t __stdio_write(FILE *f, const unsigned char *buf, size_t len)
 		}
 		rem -= cnt;
 		if (cnt > iov[0].iov_len) {
+			f->wpos = f->wbase = f->buf;
 			cnt -= iov[0].iov_len;
 			iov++; iovcnt--;
+		} else if (iovcnt == 2) {
+			f->wbase += cnt;
 		}
 		iov[0].iov_base = (char *)iov[0].iov_base + cnt;
 		iov[0].iov_len -= cnt;
diff --git a/src/stdio/fopen.c b/src/stdio/fopen.c
index 469de6f0..084cc73c 100644
--- a/src/stdio/fopen.c
+++ b/src/stdio/fopen.c
@@ -21,7 +21,7 @@ FILE *fopen(const char *filename, const char *mode)
 	if (*mode == 'w') flags |= O_TRUNC;
 	if (*mode == 'a') flags |= O_APPEND;
 
-	fd = syscall(SYS_open, filename, flags|O_LARGEFILE, 0666);
+	fd = syscall_cp(SYS_open, filename, flags|O_LARGEFILE, 0666);
 	if (fd < 0) return 0;
 
 	f = __fdopen(fd, mode);