fbarray: support --no-shconf mode
authorAnatoly Burakov <anatoly.burakov@intel.com>
Fri, 13 Jul 2018 12:47:57 +0000 (13:47 +0100)
committerThomas Monjalon <thomas@monjalon.net>
Fri, 13 Jul 2018 13:32:05 +0000 (15:32 +0200)
When using --no-shconf option, the expectation is that no multiprocess
will be supported as no shared files are created. However, fbarray still
creates some shared files that prevent multiple processes with the same
prefix from starting.

Fix this by avoiding creating shared files whenever noshconf option is
specified. Since virtual areas we get from eal_get_virtual_area() are
read-only, remap them as writable.

Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
lib/librte_eal/common/eal_common_fbarray.c

index 977174c..43caf3c 100644 (file)
@@ -705,39 +705,52 @@ rte_fbarray_init(struct rte_fbarray *arr, const char *name, unsigned int len,
        if (data == NULL)
                goto fail;
 
-       eal_get_fbarray_path(path, sizeof(path), name);
+       if (internal_config.no_shconf) {
+               /* remap virtual area as writable */
+               void *new_data = mmap(data, mmap_len, PROT_READ | PROT_WRITE,
+                               MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+               if (new_data == MAP_FAILED) {
+                       RTE_LOG(DEBUG, EAL, "%s(): couldn't remap anonymous memory: %s\n",
+                                       __func__, strerror(errno));
+                       goto fail;
+               }
+       } else {
+               eal_get_fbarray_path(path, sizeof(path), name);
 
-       /*
-        * Each fbarray is unique to process namespace, i.e. the filename
-        * depends on process prefix. Try to take out a lock and see if we
-        * succeed. If we don't, someone else is using it already.
-        */
-       fd = open(path, O_CREAT | O_RDWR, 0600);
-       if (fd < 0) {
-               RTE_LOG(DEBUG, EAL, "%s(): couldn't open %s: %s\n", __func__,
-                               path, strerror(errno));
-               rte_errno = errno;
-               goto fail;
-       } else if (flock(fd, LOCK_EX | LOCK_NB)) {
-               RTE_LOG(DEBUG, EAL, "%s(): couldn't lock %s: %s\n", __func__,
-                               path, strerror(errno));
-               rte_errno = EBUSY;
-               goto fail;
-       }
+               /*
+                * Each fbarray is unique to process namespace, i.e. the
+                * filename depends on process prefix. Try to take out a lock
+                * and see if we succeed. If we don't, someone else is using it
+                * already.
+                */
+               fd = open(path, O_CREAT | O_RDWR, 0600);
+               if (fd < 0) {
+                       RTE_LOG(DEBUG, EAL, "%s(): couldn't open %s: %s\n",
+                                       __func__, path, strerror(errno));
+                       rte_errno = errno;
+                       goto fail;
+               } else if (flock(fd, LOCK_EX | LOCK_NB)) {
+                       RTE_LOG(DEBUG, EAL, "%s(): couldn't lock %s: %s\n",
+                                       __func__, path, strerror(errno));
+                       rte_errno = EBUSY;
+                       goto fail;
+               }
 
-       /* take out a non-exclusive lock, so that other processes could still
-        * attach to it, but no other process could reinitialize it.
-        */
-       if (flock(fd, LOCK_SH | LOCK_NB)) {
-               rte_errno = errno;
-               goto fail;
-       }
+               /* take out a non-exclusive lock, so that other processes could
+                * still attach to it, but no other process could reinitialize
+                * it.
+                */
+               if (flock(fd, LOCK_SH | LOCK_NB)) {
+                       rte_errno = errno;
+                       goto fail;
+               }
 
-       if (resize_and_map(fd, data, mmap_len))
-               goto fail;
+               if (resize_and_map(fd, data, mmap_len))
+                       goto fail;
 
-       /* we've mmap'ed the file, we can now close the fd */
-       close(fd);
+               /* we've mmap'ed the file, we can now close the fd */
+               close(fd);
+       }
 
        /* initialize the data */
        memset(data, 0, mmap_len);