bus: when adding memfds to cache and we shorten them, make sure to unmap the remainder

[elogind.git] / src / libsystemd-bus / bus-kernel.c

diff --git a/src/libsystemd-bus/bus-kernel.c b/src/libsystemd-bus/bus-kernel.c
index 8ef5752b311ce5cd237f88c19f3fedc5f881b998..0ece1f014db6e4df12d5b61277f48726c06d8e68 100644 (file)
--- a/src/libsystemd-bus/bus-kernel.c
+++ b/src/libsystemd-bus/bus-kernel.c
@@ -34,20 +34,7 @@
 #include "bus-kernel.h"
 #include "bus-bloom.h"
 
-#define KDBUS_ITEM_NEXT(item) \
-        (typeof(item))(((uint8_t *)item) + ALIGN8((item)->size))
-
-#define KDBUS_ITEM_FOREACH(item, head)                                          \
-        for (item = (head)->items;                                              \
-             (uint8_t *)(item) < (uint8_t *)(head) + (head)->size;              \
-             item = KDBUS_ITEM_NEXT(item))
-
-#define KDBUS_ITEM_HEADER_SIZE offsetof(struct kdbus_item, data)
-#define KDBUS_ITEM_SIZE(s) ALIGN8((s) + KDBUS_ITEM_HEADER_SIZE)
-
-#define KDBUS_POOL_SIZE (4*1024*1024)
-
-static int parse_unique_name(const char *s, uint64_t *id) {
+int bus_kernel_parse_unique_name(const char *s, uint64_t *id) {
         int r;
 
         assert(s);
@@ -154,6 +141,7 @@ static int bus_message_setup_bloom(sd_bus_message *m, void *bloom) {
                 bloom_add_pair(bloom, "member", m->member);
         if (m->path) {
                 bloom_add_pair(bloom, "path", m->path);
+                bloom_add_pair(bloom, "path-slash-prefix", m->path);
                 bloom_add_prefixes(bloom, "path-slash-prefix", m->path, '/');
         }
 
@@ -217,7 +205,7 @@ static int bus_message_setup_kmsg(sd_bus *b, sd_bus_message *m) {
                 return 0;
 
         if (m->destination) {
-                r = parse_unique_name(m->destination, &unique);
+                r = bus_kernel_parse_unique_name(m->destination, &unique);
                 if (r < 0)
                         return r;
 
@@ -338,6 +326,8 @@ int bus_kernel_take_fd(sd_bus *b) {
         if (b->is_server)
                 return -EINVAL;
 
+        b->use_memfd = 1;
+
         if (!b->kdbus_buffer) {
                 b->kdbus_buffer = mmap(NULL, KDBUS_POOL_SIZE, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
                 if (b->kdbus_buffer == MAP_FAILED) {
@@ -347,15 +337,7 @@ int bus_kernel_take_fd(sd_bus *b) {
         }
 
         hello->size = sizeof(h);
-        hello->conn_flags =
-                KDBUS_HELLO_ACCEPT_FD|
-                KDBUS_HELLO_ATTACH_COMM|
-                KDBUS_HELLO_ATTACH_EXE|
-                KDBUS_HELLO_ATTACH_CMDLINE|
-                KDBUS_HELLO_ATTACH_CGROUP|
-                KDBUS_HELLO_ATTACH_CAPS|
-                KDBUS_HELLO_ATTACH_SECLABEL|
-                KDBUS_HELLO_ATTACH_AUDIT;
+        hello->conn_flags = b->hello_flags;
 
         hello->items[0].type = KDBUS_HELLO_POOL;
         hello->items[0].size = KDBUS_ITEM_HEADER_SIZE + sizeof(struct kdbus_vec);
@@ -380,7 +362,7 @@ int bus_kernel_take_fd(sd_bus *b) {
 
         b->is_kernel = true;
         b->bus_client = true;
-        b->can_fds = true;
+        b->can_fds = !!(hello->conn_flags & KDBUS_HELLO_ACCEPT_FD);
 
         r = bus_start_running(b);
         if (r < 0)
@@ -649,19 +631,21 @@ fail:
 }
 
 int bus_kernel_read_message(sd_bus *bus, sd_bus_message **m) {
+        uint64_t addr;
         struct kdbus_msg *k;
         int r;
 
         assert(bus);
         assert(m);
 
-        r = ioctl(bus->input_fd, KDBUS_CMD_MSG_RECV, &k);
+        r = ioctl(bus->input_fd, KDBUS_CMD_MSG_RECV, &addr);
         if (r < 0) {
                 if (errno == EAGAIN)
                         return 0;
 
                 return -errno;
         }
+        k = UINT64_TO_PTR(addr);
 
         r = bus_kernel_make_message(bus, k, m);
         if (r <= 0)
@@ -723,6 +707,7 @@ int bus_kernel_create(const char *name, char **s) {
 
 int bus_kernel_pop_memfd(sd_bus *bus, void **address, size_t *size) {
         struct memfd_cache *c;
+        int fd;
 
         assert(address);
         assert(size);
@@ -730,8 +715,12 @@ int bus_kernel_pop_memfd(sd_bus *bus, void **address, size_t *size) {
         if (!bus || !bus->is_kernel)
                 return -ENOTSUP;
 
+        assert_se(pthread_mutex_lock(&bus->memfd_cache_mutex) >= 0);
+
         if (bus->n_memfd_cache <= 0) {
-                int fd, r;
+                int r;
+
+                assert_se(pthread_mutex_unlock(&bus->memfd_cache_mutex) >= 0);
 
                 r = ioctl(bus->input_fd, KDBUS_CMD_MEMFD_NEW, &fd);
                 if (r < 0)
@@ -749,23 +738,38 @@ int bus_kernel_pop_memfd(sd_bus *bus, void **address, size_t *size) {
 
         *address = c->address;
         *size = c->size;
+        fd = c->fd;
+
+        assert_se(pthread_mutex_unlock(&bus->memfd_cache_mutex) >= 0);
+
+        return fd;
+}
+
+static void close_and_munmap(int fd, void *address, size_t size) {
+        if (size > 0)
+                assert_se(munmap(address, PAGE_ALIGN(size)) >= 0);
 
-        return c->fd;
+        close_nointr_nofail(fd);
 }
 
 void bus_kernel_push_memfd(sd_bus *bus, int fd, void *address, size_t size) {
         struct memfd_cache *c;
+        uint64_t max_sz = PAGE_ALIGN(MEMFD_CACHE_ITEM_SIZE_MAX);
 
         assert(fd >= 0);
         assert(size == 0 || address);
 
-        if (!bus || !bus->is_kernel ||
-            bus->n_memfd_cache >= ELEMENTSOF(bus->memfd_cache)) {
+        if (!bus || !bus->is_kernel) {
+                close_and_munmap(fd, address, size);
+                return;
+        }
+
+        assert_se(pthread_mutex_lock(&bus->memfd_cache_mutex) >= 0);
 
-                if (size > 0)
-                        assert_se(munmap(address, PAGE_ALIGN(size)) == 0);
+        if (bus->n_memfd_cache >= ELEMENTSOF(bus->memfd_cache)) {
+                assert_se(pthread_mutex_unlock(&bus->memfd_cache_mutex) >= 0);
 
-                close_nointr_nofail(fd);
+                close_and_munmap(fd, address, size);
                 return;
         }
 
@@ -774,14 +778,14 @@ void bus_kernel_push_memfd(sd_bus *bus, int fd, void *address, size_t size) {
         c->address = address;
 
         /* If overly long, let's return a bit to the OS */
-        if (size > MEMFD_CACHE_ITEM_SIZE_MAX) {
-                uint64_t sz = MEMFD_CACHE_ITEM_SIZE_MAX;
-
-                ioctl(bus->input_fd, KDBUS_CMD_MEMFD_SIZE_SET, &sz);
-
-                c->size = MEMFD_CACHE_ITEM_SIZE_MAX;
+        if (size > max_sz) {
+                assert_se(ioctl(fd, KDBUS_CMD_MEMFD_SIZE_SET, &max_sz) >= 0);
+                assert_se(munmap((uint8_t*) address + max_sz, PAGE_ALIGN(size - max_sz)) >= 0);
+                c->size = max_sz;
         } else
                 c->size = size;
+
+        assert_se(pthread_mutex_unlock(&bus->memfd_cache_mutex) >= 0);
 }
 
 void bus_kernel_flush_memfd(sd_bus *b) {
@@ -789,10 +793,6 @@ void bus_kernel_flush_memfd(sd_bus *b) {
 
         assert(b);
 
-        for (i = 0; i < b->n_memfd_cache; i++) {
-                if (b->memfd_cache[i].size > 0)
-                        assert_se(munmap(b->memfd_cache[i].address, PAGE_ALIGN(b->memfd_cache[i].size)) == 0);
-
-                close_nointr_nofail(b->memfd_cache[i].fd);
-        }
+        for (i = 0; i < b->n_memfd_cache; i++)
+                close_and_munmap(b->memfd_cache[i].fd, b->memfd_cache[i].address, b->memfd_cache[i].size);
 }