Merge branch 'upstream-add-cap-devs-to-list-command' into 'master'

Add support to "list" command to print /dev based capabilities

See merge request nvidia/container-toolkit/libnvidia-container!39

Author: klueska

pkg/deb/libnvidia-container@[email protected]

@@ -10,6 +10,8 @@ libnvidia-container.so.1 libnvidia-container1 #MINVER#
  nvc_context_new@NVC_1.0 1.2.0~rc.3
  nvc_device_info_free@NVC_1.0 1.2.0~rc.3
  nvc_device_info_new@NVC_1.0 1.2.0~rc.3
+ nvc_nvcaps_style@NVC_1.0 1.2.0~rc.3
+ nvc_nvcaps_device_from_proc_path@NVC_1.0 1.2.0~rc.3
  nvc_device_mount@NVC_1.0 1.2.0~rc.3
  nvc_mig_device_access_caps_mount@NVC_1.0 1.2.0~rc.3
  nvc_mig_config_global_caps_mount@NVC_1.0 1.2.0~rc.3

src/cli/cli.h

@@ -69,6 +69,9 @@ struct devices {
 int new_devices(struct error *err, const struct nvc_device_info *dev, struct devices *d);
 void free_devices(struct devices *d);
 
+int print_nvcaps_device_from_proc_file(struct nvc_context *, const char*, const char*);
+int print_all_mig_minor_devices(const struct nvc_device_node *);
+
 int select_devices(
     struct error *err,
     char *devs,

src/cli/common.c

@@ -2,6 +2,8 @@
  * Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
  */
 
+#include <sys/sysmacros.h>
+
 #include <inttypes.h>
 #include <string.h>
 
@@ -464,3 +466,50 @@ free_devices(struct devices *d)
         free(d->migs);
         memset(d, 0, sizeof(*d));
 }
+
+int
+print_nvcaps_device_from_proc_file(struct nvc_context *ctx, const char* cap_dir, const char* cap_file)
+{
+        char cap_path[PATH_MAX];
+        struct nvc_device_node node;
+
+        if (path_join(NULL, cap_path, cap_dir, cap_file) < 0)
+                return (-1);
+        if (nvc_nvcaps_device_from_proc_path(ctx, cap_path, &node) < 0)
+                return (-1);
+
+        printf("%s\n", node.path);
+        free(node.path);
+
+        return (0);
+}
+
+int
+print_all_mig_minor_devices(const struct nvc_device_node *node)
+{
+        unsigned int gpu_minor = 0;
+        unsigned int mig_minor = 0;
+        char line[PATH_MAX];
+        char dummy[PATH_MAX];
+        FILE *fp;
+        int rv = -1;
+
+        if ((fp = fopen(NV_CAPS_MIG_MINORS_PATH, "r")) == NULL) {
+            goto fail;
+        }
+
+        line[PATH_MAX - 1] = '\0';
+        while (fgets(line, PATH_MAX - 1, fp)) {
+                if (sscanf(line, "gpu%u%s %u", &gpu_minor, dummy, &mig_minor) != 3)
+                        continue;
+                if (gpu_minor != minor(node->id))
+                        continue;
+                printf(NV_CAPS_DEVICE_PATH "\n", mig_minor);
+        }
+
+        rv = 0;
+
+fail:
+        fclose(fp);
+        return (rv);
+}

src/cli/list.c

@@ -188,22 +188,38 @@ list_command(const struct context *ctx)
                         for (size_t i = 0; i < devices.nmigs; ++i) {
                                 printf("%s/%s\n", devices.migs[i]->gi_caps_path, NV_MIG_ACCESS_FILE);
                                 printf("%s/%s\n", devices.migs[i]->ci_caps_path, NV_MIG_ACCESS_FILE);
+                                if (nvc_nvcaps_style() == NVC_NVCAPS_STYLE_DEV) {
+                                        print_nvcaps_device_from_proc_file(nvc, devices.migs[i]->gi_caps_path, NV_MIG_ACCESS_FILE);
+                                        print_nvcaps_device_from_proc_file(nvc, devices.migs[i]->ci_caps_path, NV_MIG_ACCESS_FILE);
+                                }
                         }
                 }
         }
 
         /* List the files required for MIG configuration of the visible devices */
         if (mig_config_devices.all && mig_config_devices.ngpus) {
                 printf("%s/%s\n", NV_MIG_CAPS_PATH, NV_MIG_CONFIG_FILE);
+                if (nvc_nvcaps_style() == NVC_NVCAPS_STYLE_DEV)
+                        print_nvcaps_device_from_proc_file(nvc, NV_MIG_CAPS_PATH, NV_MIG_CONFIG_FILE);
                 for (size_t i = 0; i < mig_config_devices.ngpus; ++i) {
                         printf("%s\n", mig_config_devices.gpus[i]->mig_caps_path);
+                        if (nvc_nvcaps_style() == NVC_NVCAPS_STYLE_DEV) {
+                                printf("%s\n", NV_CAPS_DEVICE_DIR);
+                                print_all_mig_minor_devices(&mig_config_devices.gpus[i]->node);
+                        }
                 }
         }
         /* List the files required for MIG monitoring of the visible devices */
         if (mig_monitor_devices.all && mig_monitor_devices.ngpus) {
                 printf("%s/%s\n", NV_MIG_CAPS_PATH, NV_MIG_MONITOR_FILE);
+                if (nvc_nvcaps_style() == NVC_NVCAPS_STYLE_DEV)
+                        print_nvcaps_device_from_proc_file(nvc, NV_MIG_CAPS_PATH, NV_MIG_MONITOR_FILE);
                 for (size_t i = 0; i < mig_monitor_devices.ngpus; ++i) {
                         printf("%s\n", mig_monitor_devices.gpus[i]->mig_caps_path);
+                        if (nvc_nvcaps_style() == NVC_NVCAPS_STYLE_DEV) {
+                                printf("%s\n", NV_CAPS_DEVICE_DIR);
+                                print_all_mig_minor_devices(&mig_monitor_devices.gpus[i]->node);
+                        }
                 }
         }
 

src/libnvidia-container.lds

@@ -33,6 +33,8 @@ VERSION {
             nvc_device_info_free;
             nvc_driver_mount;
             nvc_device_mount;
+            nvc_nvcaps_style;
+            nvc_nvcaps_device_from_proc_path;
             nvc_mig_device_access_caps_mount;
             nvc_mig_config_global_caps_mount;
             nvc_mig_monitor_global_caps_mount;

src/nvc.h

@@ -20,6 +20,10 @@ extern "C" {
 
 #define NVC_ARG_MAX 256
 
+#define NVC_NVCAPS_STYLE_NONE 0
+#define NVC_NVCAPS_STYLE_PROC 1
+#define NVC_NVCAPS_STYLE_DEV  2
+
 struct nvc_context;
 struct nvc_container;
 
@@ -121,6 +125,10 @@ void nvc_driver_info_free(struct nvc_driver_info *);
 struct nvc_device_info *nvc_device_info_new(struct nvc_context *, const char *);
 void nvc_device_info_free(struct nvc_device_info *);
 
+int nvc_nvcaps_style(void);
+
+int nvc_nvcaps_device_from_proc_path(struct nvc_context *, const char *, struct nvc_device_node *);
+
 int nvc_driver_mount(struct nvc_context *, const struct nvc_container *, const struct nvc_driver_info *);
 
 int nvc_device_mount(struct nvc_context *, const struct nvc_container *, const struct nvc_device *);

src/nvc_info.c

@@ -6,6 +6,7 @@
 
 #include <errno.h>
 #include <limits.h>
+#include <nvidia-modprobe-utils.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -788,3 +789,39 @@ nvc_device_info_free(struct nvc_device_info *info)
         free(info->gpus);
         free(info);
 }
+
+int
+nvc_nvcaps_style(void)
+{
+        if (nvidia_get_chardev_major(NV_CAPS_MODULE_NAME) >= 0)
+                return NVC_NVCAPS_STYLE_DEV;
+        if (file_exists(NULL, NV_PROC_DRIVER_CAPS) >= 0)
+                return NVC_NVCAPS_STYLE_PROC;
+        return NVC_NVCAPS_STYLE_NONE;
+}
+
+int
+nvc_nvcaps_device_from_proc_path(struct nvc_context *ctx, const char *cap_path, struct nvc_device_node *node)
+{
+        char abs_cap_path[PATH_MAX];
+        char dev_name[PATH_MAX];
+        int major, minor;
+        int rv = -1;
+
+        if (path_join(&ctx->err, abs_cap_path, ctx->cfg.root, cap_path) < 0)
+                goto fail;
+
+        if (nvidia_cap_get_device_file_attrs(abs_cap_path, &major, &minor, dev_name) == 0) {
+                error_set(&ctx->err, "unable to get cap device attributes: %s", cap_path);
+                goto fail;
+        }
+
+        if ((node->path = xstrdup(&ctx->err, dev_name)) == NULL)
+                goto fail;
+        node->id = makedev((unsigned int)major, (unsigned int)minor);
+
+        rv = 0;
+
+fail:
+        return (rv);
+}

src/nvc_mount.c

@@ -40,7 +40,6 @@ static int  symlink_libraries(struct error *, const struct nvc_container *, cons
 static void filter_libraries(const struct nvc_driver_info *, char * [], size_t *);
 static int  device_mount_dxcore(struct nvc_context *, const struct nvc_container *);
 static int  device_mount_native(struct nvc_context *, const struct nvc_container *, const struct nvc_device *);
-static int  cap_device_from_path(struct nvc_context *, const char *, struct nvc_device_node *);
 static int  cap_device_mount(struct nvc_context *, const struct nvc_container *, const char *);
 static int  setup_mig_minor_cgroups(struct error *, const struct nvc_container *, int, const struct nvc_device_node *);
 
@@ -643,40 +642,14 @@ device_mount_native(struct nvc_context *ctx, const struct nvc_container *cnt, co
         return (rv);
 }
 
-static int
-cap_device_from_path(struct nvc_context *ctx, const char *cap_path, struct nvc_device_node *node)
-{
-        char abs_cap_path[PATH_MAX];
-        char dev_name[PATH_MAX];
-        int major, minor;
-        int rv = -1;
-
-        if (path_join(&ctx->err, abs_cap_path, ctx->cfg.root, cap_path) < 0)
-                goto fail;
-
-        if (nvidia_cap_get_device_file_attrs(abs_cap_path, &major, &minor, dev_name) == 0) {
-                error_set(&ctx->err, "unable to get cap device attributes: %s", cap_path);
-                goto fail;
-        }
-
-        if ((node->path = xstrdup(&ctx->err, dev_name)) == NULL)
-                goto fail;
-        node->id = makedev((unsigned int)major, (unsigned int)minor);
-
-        rv = 0;
-
-fail:
-        return (rv);
-}
-
 static int
 cap_device_mount(struct nvc_context *ctx, const struct nvc_container *cnt, const char *cap_path)
 {
         char *dev_mnt = NULL;
         struct nvc_device_node node = {0};
         int rv = -1;
 
-        if (cap_device_from_path(ctx, cap_path, &node) < 0)
+        if (nvc_nvcaps_device_from_proc_path(ctx, cap_path, &node) < 0)
                 goto fail;
 
         if (!(cnt->flags & OPT_NO_DEVBIND)) {
@@ -978,7 +951,7 @@ nvc_mig_config_global_caps_mount(struct nvc_context *ctx, const struct nvc_conta
                 if ((dev_mnt = mount_directory(&ctx->err, ctx->cfg.root, cnt, NV_CAPS_DEVICE_DIR)) == NULL)
                     goto fail;
 
-                if (cap_device_from_path(ctx, config, &node) < 0)
+                if (nvc_nvcaps_device_from_proc_path(ctx, config, &node) < 0)
                         goto fail;
 
                 if (!(cnt->flags & OPT_NO_CGROUPS))
@@ -1041,7 +1014,7 @@ nvc_mig_monitor_global_caps_mount(struct nvc_context *ctx, const struct nvc_cont
                 if ((dev_mnt = mount_directory(&ctx->err, ctx->cfg.root, cnt, NV_CAPS_DEVICE_DIR)) == NULL)
                         goto fail;
 
-                if (cap_device_from_path(ctx, monitor, &node) < 0)
+                if (nvc_nvcaps_device_from_proc_path(ctx, monitor, &node) < 0)
                         goto fail;
 
                 if (!(cnt->flags & OPT_NO_CGROUPS))