From 8b9fb1bbec97dddeee68c5060b06ec9d45e168f3 Mon Sep 17 00:00:00 2001
From: Daan De Meyer <daan@amutable.com>
Date: Thu, 1 Oct 2026 09:24:45 +0200
Subject: [PATCH] io-util: add pwritev_full() and pwrite_full()

loop_write() writes one buffer at the file position. To write at an
offset, callers call pwrite() or pwritev() directly and handle short
writes themselves.

Add pwritev_full(), which writes all iovecs at an offset, retries on
EINTR and short writes, splits batches larger than IOV_MAX, and reports
the number of bytes written, also on failure. Add pwrite_full(), which
does the same for one buffer.

(cherry picked from commit 4d36be491a568180dc6336b91b0c496bb8cb6601)
---
 src/basic/io-util.c     | 50 +++++++++++++++++++++++++++++++++++++++++
 src/basic/io-util.h     |  3 +++
 src/test/test-io-util.c | 49 ++++++++++++++++++++++++++++++++++++++++
 3 files changed, 102 insertions(+)

diff --git a/src/basic/io-util.c b/src/basic/io-util.c
index d4a95128c8..0b584a638a 100644
--- a/src/basic/io-util.c
+++ b/src/basic/io-util.c
@@ -11,6 +11,7 @@
 #include "errno-util.h"
 #include "fiber-ops.h"
 #include "io-util.h"
+#include "iovec-util.h"
 #include "time-util.h"
 
 uint32_t poll_events_to_epoll(uint32_t events) {
@@ -249,6 +250,55 @@ int loop_write_full(int fd, const void *buf, size_t nbytes, usec_t timeout) {
         return 0;
 }
 
+int pwritev_full(int fd, struct iovec *iovec, size_t n, uint64_t offset, uint64_t *ret_written) {
+        uint64_t written = 0;
+        int r = 0;
+
+        assert(fd >= 0);
+        assert(iovec || n == 0);
+
+        /* Writes all of the iovecs at the offset. The iovecs are advanced past what was written, and
+         * 'ret_written' tells how much that was, also if the write fails halfway. */
+
+        for (size_t i = 0;;) {
+                size_t m;
+                ssize_t k;
+
+                while (i < n && !iovec_is_set(iovec + i))
+                        i++;
+                if (i >= n)
+                        break;
+
+                m = MIN(n - i, (size_t) IOV_MAX);
+                k = pwritev(fd, iovec + i, m, offset + written);
+                if (k < 0) {
+                        if (errno == EINTR)
+                                continue;
+
+                        r = -errno;
+                        break;
+                }
+                if (k == 0) {
+                        r = -EIO;
+                        break;
+                }
+
+                written += k;
+                (void) iovec_inc_many(iovec + i, m, k);
+        }
+
+        if (ret_written)
+                *ret_written = written;
+        return r;
+}
+
+int pwrite_full(int fd, const void *buf, size_t nbytes, uint64_t offset) {
+        assert(fd >= 0);
+        assert(buf || nbytes == 0);
+
+        return pwritev_full(fd, &IOVEC_MAKE((void*) buf, nbytes), 1, offset, /* ret_written= */ NULL);
+}
+
 int pipe_eof(int fd) {
         int r;
 
diff --git a/src/basic/io-util.h b/src/basic/io-util.h
index 0b3d7fff8d..8ae8e29bb2 100644
--- a/src/basic/io-util.h
+++ b/src/basic/io-util.h
@@ -20,6 +20,9 @@ static inline int loop_write(int fd, const void *buf, size_t nbytes) {
         return loop_write_full(fd, buf, nbytes, 0);
 }
 
+int pwritev_full(int fd, struct iovec *iovec, size_t n, uint64_t offset, uint64_t *ret_written);
+int pwrite_full(int fd, const void *buf, size_t nbytes, uint64_t offset) _nonnull_if_nonzero_(2, 3);
+
 int pipe_eof(int fd);
 
 int ppoll_usec_full(struct pollfd *fds, size_t n_fds, usec_t timeout, const sigset_t *ss) _nonnull_if_nonzero_(1, 2);
diff --git a/src/test/test-io-util.c b/src/test/test-io-util.c
index dbf3495ab9..694cf01885 100644
--- a/src/test/test-io-util.c
+++ b/src/test/test-io-util.c
@@ -6,7 +6,9 @@
 
 #include "fd-util.h"
 #include "io-util.h"
+#include "iovec-util.h"
 #include "tests.h"
+#include "tmpfile-util.h"
 
 static void test_sparse_write_one(int fd, const char *buffer, size_t n) {
         char check[n];
@@ -43,4 +45,51 @@ TEST(sparse_write) {
         test_sparse_write_one(fd, test_e, sizeof(test_e));
 }
 
+TEST(pwritev_full) {
+        _cleanup_close_ int fd = -EBADF;
+        struct iovec iovec[2 + 2 * IOV_MAX];
+        char expected[3 + 2 * IOV_MAX], buf[sizeof(expected)];
+        uint64_t written;
+        size_t n = 0;
+
+        ASSERT_OK(fd = open_tmpfile_unlinkable(NULL, O_RDWR|O_CLOEXEC));
+
+        /* Nothing, or only empty iovecs, writes nothing */
+        ASSERT_OK(pwritev_full(fd, NULL, 0, 0, &written));
+        ASSERT_EQ(written, 0U);
+        iovec[0] = (struct iovec) {};
+        ASSERT_OK(pwritev_full(fd, iovec, 1, 0, &written));
+        ASSERT_EQ(written, 0U);
+
+        /* More iovecs than one pwritev() takes, some of them empty */
+        iovec[n++] = IOVEC_MAKE_STRING("abc");
+        iovec[n++] = (struct iovec) {};
+        memcpy(expected, "abc", 3);
+        for (size_t k = 0; k < 2 * IOV_MAX; k++) {
+                expected[3 + k] = 'a' + k % 26;
+                iovec[n++] = IOVEC_MAKE(expected + 3 + k, 1);
+        }
+
+        ASSERT_OK(pwritev_full(fd, iovec, n, 7, &written));
+        ASSERT_EQ(written, sizeof(expected));
+        ASSERT_EQ(pread(fd, buf, sizeof(buf), 7), (ssize_t) sizeof(buf));
+        ASSERT_EQ(memcmp(buf, expected, sizeof(expected)), 0);
+
+        /* Written iovecs are advanced past the data */
+        FOREACH_ARRAY(i, iovec, n)
+                ASSERT_FALSE(iovec_is_set(i));
+}
+
+TEST(pwrite_full) {
+        _cleanup_close_ int fd = -EBADF;
+        char buf[3];
+
+        ASSERT_OK(fd = open_tmpfile_unlinkable(NULL, O_RDWR|O_CLOEXEC));
+
+        ASSERT_OK(pwrite_full(fd, NULL, 0, 0));
+        ASSERT_OK(pwrite_full(fd, "abc", 3, 5));
+        ASSERT_EQ(pread(fd, buf, sizeof(buf), 5), (ssize_t) sizeof(buf));
+        ASSERT_EQ(memcmp(buf, "abc", 3), 0);
+}
+
 DEFINE_TEST_MAIN(LOG_INFO);