sd-bus: properly support passing memfds around with offset/size parameters

4 files changed, 75 insertions, 40 deletions

diff --git a/src/libsystemd/sd-bus/bus-message.c b/src/libsystemd/sd-bus/bus-message.c
index ad417c0f10..06d8d770fd 100644
--- a/src/libsystemd/sd-bus/bus-message.c
+++ b/src/libsystemd/sd-bus/bus-message.c
@@ -69,13 +69,13 @@ static void message_free_part(sd_bus_message *m, struct bus_body_part *part) {
                         bus_kernel_push_memfd(m->bus, part->memfd, part->data, part->mapped, part->allocated);
                 else {
                         if (part->mapped > 0)
-                                assert_se(munmap(part->data, part->mapped) == 0);
+                                assert_se(munmap(part->mmap_begin, part->mapped) == 0);
 
                         safe_close(part->memfd);
                 }
 
         } else if (part->munmap_this)
-                munmap(part->data, part->mapped);
+                munmap(part->mmap_begin, part->mapped);
         else if (part->free_this)
                 free(part->data);
 
@@ -1126,17 +1126,18 @@ static int part_make_space(
 
                         psz = PAGE_ALIGN(sz > 0 ? sz : 1);
                         if (part->mapped <= 0)
-                                n = mmap(NULL, psz, PROT_READ|PROT_WRITE, MAP_SHARED, part->memfd, part->memfd_offset);
+                                n = mmap(NULL, psz, PROT_READ|PROT_WRITE, MAP_SHARED, part->memfd, 0);
                         else
-                                n = mremap(part->data, part->mapped, psz, MREMAP_MAYMOVE);
+                                n = mremap(part->mmap_begin, part->mapped, psz, MREMAP_MAYMOVE);
 
                         if (n == MAP_FAILED) {
                                 m->poisoned = true;
                                 return -errno;
                         }
 
+                        part->mmap_begin = part->data = n;
                         part->mapped = psz;
-                        part->data = n;
+                        part->memfd_offset = 0;
                 }
 
                 part->munmap_this = true;
@@ -2503,10 +2504,11 @@ _public_ int sd_bus_message_append_array_space(
         return 0;
 }
 
-_public_ int sd_bus_message_append_array(sd_bus_message *m,
-                                         char type,
-                                         const void *ptr,
-                                         size_t size) {
+_public_ int sd_bus_message_append_array(
+                sd_bus_message *m,
+                char type,
+                const void *ptr,
+                size_t size) {
         int r;
         void *p;
 
@@ -2562,26 +2564,26 @@ _public_ int sd_bus_message_append_array_iovec(
         return 0;
 }
 
-_public_ int sd_bus_message_append_array_memfd(sd_bus_message *m,
-                                               char type,
-                                               int memfd) {
+_public_ int sd_bus_message_append_array_memfd(
+                sd_bus_message *m,
+                char type,
+                int memfd,
+                uint64_t offset,
+                uint64_t size) {
+
         _cleanup_close_ int copy_fd = -1;
         struct bus_body_part *part;
         ssize_t align, sz;
-        uint64_t size;
+        uint64_t real_size;
         void *a;
         int r;
 
-        if (!m)
-                return -EINVAL;
-        if (memfd < 0)
-                return -EINVAL;
-        if (m->sealed)
-                return -EPERM;
-        if (!bus_type_is_trivial(type))
-                return -EINVAL;
-        if (m->poisoned)
-                return -ESTALE;
+        assert_return(m, -EINVAL);
+        assert_return(memfd >= 0, -EINVAL);
+        assert_return(bus_type_is_trivial(type), -EINVAL);
+        assert_return(size > 0, -EINVAL);
+        assert_return(!m->sealed, -EPERM);
+        assert_return(!m->poisoned, -ESTALE);
 
         r = memfd_set_sealed(memfd);
         if (r < 0)
@@ -2591,16 +2593,24 @@ _public_ int sd_bus_message_append_array_memfd(sd_bus_message *m,
         if (copy_fd < 0)
                 return copy_fd;
 
-        r = memfd_get_size(memfd, &size);
+        r = memfd_get_size(memfd, &real_size);
         if (r < 0)
                 return r;
 
+        if (offset == 0 && size == (uint64_t) -1)
+                size = real_size;
+        else if (offset + size > real_size)
+                return -EMSGSIZE;
+
         align = bus_type_get_alignment(type);
         sz = bus_type_get_size(type);
 
         assert_se(align > 0);
         assert_se(sz > 0);
 
+        if (offset % align != 0)
+                return -EINVAL;
+
         if (size % sz != 0)
                 return -EINVAL;
 
@@ -2620,7 +2630,7 @@ _public_ int sd_bus_message_append_array_memfd(sd_bus_message *m,
                 return -ENOMEM;
 
         part->memfd = copy_fd;
-        part->memfd_offset = 0;
+        part->memfd_offset = offset;
         part->sealed = true;
         part->size = size;
         copy_fd = -1;
@@ -2631,16 +2641,22 @@ _public_ int sd_bus_message_append_array_memfd(sd_bus_message *m,
         return sd_bus_message_close_container(m);
 }
 
-_public_ int sd_bus_message_append_string_memfd(sd_bus_message *m, int memfd) {
+_public_ int sd_bus_message_append_string_memfd(
+                sd_bus_message *m,
+                int memfd,
+                uint64_t offset,
+                uint64_t size) {
+
         _cleanup_close_ int copy_fd = -1;
         struct bus_body_part *part;
         struct bus_container *c;
-        uint64_t size;
+        uint64_t real_size;
         void *a;
         int r;
 
         assert_return(m, -EINVAL);
         assert_return(memfd >= 0, -EINVAL);
+        assert_return(size > 0, -EINVAL);
         assert_return(!m->sealed, -EPERM);
         assert_return(!m->poisoned, -ESTALE);
 
@@ -2652,10 +2668,15 @@ _public_ int sd_bus_message_append_string_memfd(sd_bus_message *m, int memfd) {
         if (copy_fd < 0)
                 return copy_fd;
 
-        r = memfd_get_size(memfd, &size);
+        r = memfd_get_size(memfd, &real_size);
         if (r < 0)
                 return r;
 
+        if (offset == 0 && size == (uint64_t) -1)
+                size = real_size;
+        else if (offset + size > real_size)
+                return -EMSGSIZE;
+
         /* We require this to be NUL terminated */
         if (size == 0)
                 return -EINVAL;
@@ -2696,7 +2717,7 @@ _public_ int sd_bus_message_append_string_memfd(sd_bus_message *m, int memfd) {
                 return -ENOMEM;
 
         part->memfd = copy_fd;
-        part->memfd_offset = 0;
+        part->memfd_offset = offset;
         part->sealed = true;
         part->size = size;
         copy_fd = -1;
@@ -2860,7 +2881,7 @@ int bus_message_seal(sd_bus_message *m, uint64_t cookie, usec_t timeout) {
 
 int bus_body_part_map(struct bus_body_part *part) {
         void *p;
-        size_t psz;
+        size_t psz, shift;
 
         assert_se(part);
 
@@ -2877,10 +2898,11 @@ int bus_body_part_map(struct bus_body_part *part) {
                 return 0;
         }
 
-        psz = PAGE_ALIGN(part->size);
+        shift = part->memfd_offset - ((part->memfd_offset / page_size()) * page_size());
+        psz = PAGE_ALIGN(part->size + shift);
 
         if (part->memfd >= 0)
-                p = mmap(NULL, psz, PROT_READ, MAP_PRIVATE, part->memfd, part->memfd_offset);
+                p = mmap(NULL, psz, PROT_READ, MAP_PRIVATE, part->memfd, part->memfd_offset - shift);
         else if (part->is_zero)
                 p = mmap(NULL, psz, PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
         else
@@ -2890,7 +2912,8 @@ int bus_body_part_map(struct bus_body_part *part) {
                 return -errno;
 
         part->mapped = psz;
-        part->data = p;
+        part->mmap_begin = p;
+        part->data = (uint8_t*) p + shift;
         part->munmap_this = true;
 
         return 0;
@@ -2903,14 +2926,15 @@ void bus_body_part_unmap(struct bus_body_part *part) {
         if (part->memfd < 0)
                 return;
 
-        if (!part->data)
+        if (!part->mmap_begin)
                 return;
 
         if (!part->munmap_this)
                 return;
 
-        assert_se(munmap(part->data, part->mapped) == 0);
+        assert_se(munmap(part->mmap_begin, part->mapped) == 0);
 
+        part->mmap_begin = NULL;
         part->data = NULL;
         part->mapped = 0;
         part->munmap_this = false;
diff --git a/src/libsystemd/sd-bus/bus-message.h b/src/libsystemd/sd-bus/bus-message.h
index db5f90c5de..4dd280dcf0 100644
--- a/src/libsystemd/sd-bus/bus-message.h
+++ b/src/libsystemd/sd-bus/bus-message.h
@@ -55,10 +55,11 @@ struct bus_container {
 struct bus_body_part {
         struct bus_body_part *next;
         void *data;
+        void *mmap_begin;
         size_t size;
         size_t mapped;
         size_t allocated;
-        size_t memfd_offset;
+        uint64_t memfd_offset;
         int memfd;
         bool free_this:1;
         bool munmap_this:1;
diff --git a/src/libsystemd/sd-bus/test-bus-zero-copy.c b/src/libsystemd/sd-bus/test-bus-zero-copy.c
index 2cc671b16d..a054f74bf3 100644
--- a/src/libsystemd/sd-bus/test-bus-zero-copy.c
+++ b/src/libsystemd/sd-bus/test-bus-zero-copy.c
@@ -48,6 +48,7 @@ int main(int argc, char *argv[]) {
         uint32_t u32;
         size_t i, l;
         char *s;
+        _cleanup_close_ int sfd = -1;
 
         log_set_max_level(LOG_DEBUG);
 
@@ -107,7 +108,7 @@ int main(int argc, char *argv[]) {
         assert_se(r >= 0);
         assert_se(sz == STRING_SIZE);
 
-        r = sd_bus_message_append_string_memfd(m, f);
+        r = sd_bus_message_append_string_memfd(m, f, 0, (uint64_t) -1);
         assert_se(r >= 0);
 
         close(f);
@@ -124,7 +125,7 @@ int main(int argc, char *argv[]) {
         assert_se(r >= 0);
         assert_se(sz == SECOND_ARRAY);
 
-        r = sd_bus_message_append_array_memfd(m, 'y', f);
+        r = sd_bus_message_append_array_memfd(m, 'y', f, 0, (uint64_t) -1);
         assert_se(r >= 0);
 
         close(f);
@@ -135,6 +136,11 @@ int main(int argc, char *argv[]) {
         r = sd_bus_message_append(m, "u", 4711);
         assert_se(r >= 0);
 
+        assert_se((sfd = memfd_new_and_map(NULL, 6, (void**) &p)) >= 0);
+        memcpy(p, "abcd\0", 6);
+        munmap(p, 6);
+        assert_se(sd_bus_message_append_string_memfd(m, sfd, 1, 4) >= 0);
+
         r = bus_message_seal(m, 55, 99*USEC_PER_SEC);
         assert_se(r >= 0);
 
@@ -188,6 +194,10 @@ int main(int argc, char *argv[]) {
         assert_se(r > 0);
         assert_se(u32 == 4711);
 
+        r = sd_bus_message_read(m, "s", &s);
+        assert_se(r > 0);
+        assert_se(streq_ptr(s, "bcd"));
+
         sd_bus_message_unref(m);
 
         sd_bus_unref(a);
diff --git a/src/systemd/sd-bus.h b/src/systemd/sd-bus.h
index f5fd37a662..3636d66fbb 100644
--- a/src/systemd/sd-bus.h
+++ b/src/systemd/sd-bus.h
@@ -261,10 +261,10 @@ int sd_bus_message_append_basic(sd_bus_message *m, char type, const void *p);
 int sd_bus_message_append_array(sd_bus_message *m, char type, const void *ptr, size_t size);
 int sd_bus_message_append_array_space(sd_bus_message *m, char type, size_t size, void **ptr);
 int sd_bus_message_append_array_iovec(sd_bus_message *m, char type, const struct iovec *iov, unsigned n);
-int sd_bus_message_append_array_memfd(sd_bus_message *m, char type, int memfd);
+int sd_bus_message_append_array_memfd(sd_bus_message *m, char type, int memfd, uint64_t offset, uint64_t size);
 int sd_bus_message_append_string_space(sd_bus_message *m, size_t size, char **s);
 int sd_bus_message_append_string_iovec(sd_bus_message *m, const struct iovec *iov, unsigned n);
-int sd_bus_message_append_string_memfd(sd_bus_message *m, int memfd);
+int sd_bus_message_append_string_memfd(sd_bus_message *m, int memfd, uint64_t offset, uint64_t size);
 int sd_bus_message_append_strv(sd_bus_message *m, char **l);
 int sd_bus_message_open_container(sd_bus_message *m, char type, const char *contents);
 int sd_bus_message_close_container(sd_bus_message *m);