Merge branch 'work.mount' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux-2.6-block.git] / fs / proc / root.c
index 621e6ec322ca93032d915519717f995abfeaf266..8b145e7b96614a8da1f5df306fae771539a613a3 100644 (file)
 #include <linux/module.h>
 #include <linux/bitops.h>
 #include <linux/user_namespace.h>
+#include <linux/fs_context.h>
 #include <linux/mount.h>
 #include <linux/pid_namespace.h>
-#include <linux/parser.h>
+#include <linux/fs_parser.h>
 #include <linux/cred.h>
+#include <linux/magic.h>
+#include <linux/slab.h>
 
 #include "internal.h"
 
-enum {
-       Opt_gid, Opt_hidepid, Opt_err,
+struct proc_fs_context {
+       struct pid_namespace    *pid_ns;
+       unsigned int            mask;
+       int                     hidepid;
+       int                     gid;
 };
 
-static const match_table_t tokens = {
-       {Opt_hidepid, "hidepid=%u"},
-       {Opt_gid, "gid=%u"},
-       {Opt_err, NULL},
+enum proc_param {
+       Opt_gid,
+       Opt_hidepid,
 };
 
-int proc_parse_options(char *options, struct pid_namespace *pid)
+static const struct fs_parameter_spec proc_param_specs[] = {
+       fsparam_u32("gid",      Opt_gid),
+       fsparam_u32("hidepid",  Opt_hidepid),
+       {}
+};
+
+static const struct fs_parameter_description proc_fs_parameters = {
+       .name           = "proc",
+       .specs          = proc_param_specs,
+};
+
+static int proc_parse_param(struct fs_context *fc, struct fs_parameter *param)
 {
-       char *p;
-       substring_t args[MAX_OPT_ARGS];
-       int option;
-
-       if (!options)
-               return 1;
-
-       while ((p = strsep(&options, ",")) != NULL) {
-               int token;
-               if (!*p)
-                       continue;
-
-               args[0].to = args[0].from = NULL;
-               token = match_token(p, tokens, args);
-               switch (token) {
-               case Opt_gid:
-                       if (match_int(&args[0], &option))
-                               return 0;
-                       pid->pid_gid = make_kgid(current_user_ns(), option);
-                       break;
-               case Opt_hidepid:
-                       if (match_int(&args[0], &option))
-                               return 0;
-                       if (option < HIDEPID_OFF ||
-                           option > HIDEPID_INVISIBLE) {
-                               pr_err("proc: hidepid value must be between 0 and 2.\n");
-                               return 0;
-                       }
-                       pid->hide_pid = option;
-                       break;
-               default:
-                       pr_err("proc: unrecognized mount option \"%s\" "
-                              "or missing value\n", p);
-                       return 0;
-               }
+       struct proc_fs_context *ctx = fc->fs_private;
+       struct fs_parse_result result;
+       int opt;
+
+       opt = fs_parse(fc, &proc_fs_parameters, param, &result);
+       if (opt < 0)
+               return opt;
+
+       switch (opt) {
+       case Opt_gid:
+               ctx->gid = result.uint_32;
+               break;
+
+       case Opt_hidepid:
+               ctx->hidepid = result.uint_32;
+               if (ctx->hidepid < HIDEPID_OFF ||
+                   ctx->hidepid > HIDEPID_INVISIBLE)
+                       return invalf(fc, "proc: hidepid value must be between 0 and 2.\n");
+               break;
+
+       default:
+               return -EINVAL;
        }
 
-       return 1;
+       ctx->mask |= 1 << opt;
+       return 0;
 }
 
-int proc_remount(struct super_block *sb, int *flags, char *data)
+static void proc_apply_options(struct super_block *s,
+                              struct fs_context *fc,
+                              struct pid_namespace *pid_ns,
+                              struct user_namespace *user_ns)
 {
+       struct proc_fs_context *ctx = fc->fs_private;
+
+       if (ctx->mask & (1 << Opt_gid))
+               pid_ns->pid_gid = make_kgid(user_ns, ctx->gid);
+       if (ctx->mask & (1 << Opt_hidepid))
+               pid_ns->hide_pid = ctx->hidepid;
+}
+
+static int proc_fill_super(struct super_block *s, struct fs_context *fc)
+{
+       struct pid_namespace *pid_ns = get_pid_ns(s->s_fs_info);
+       struct inode *root_inode;
+       int ret;
+
+       proc_apply_options(s, fc, pid_ns, current_user_ns());
+
+       /* User space would break if executables or devices appear on proc */
+       s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
+       s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
+       s->s_blocksize = 1024;
+       s->s_blocksize_bits = 10;
+       s->s_magic = PROC_SUPER_MAGIC;
+       s->s_op = &proc_sops;
+       s->s_time_gran = 1;
+
+       /*
+        * procfs isn't actually a stacking filesystem; however, there is
+        * too much magic going on inside it to permit stacking things on
+        * top of it
+        */
+       s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
+       
+       /* procfs dentries and inodes don't require IO to create */
+       s->s_shrink.seeks = 0;
+
+       pde_get(&proc_root);
+       root_inode = proc_get_inode(s, &proc_root);
+       if (!root_inode) {
+               pr_err("proc_fill_super: get root inode failed\n");
+               return -ENOMEM;
+       }
+
+       s->s_root = d_make_root(root_inode);
+       if (!s->s_root) {
+               pr_err("proc_fill_super: allocate dentry failed\n");
+               return -ENOMEM;
+       }
+
+       ret = proc_setup_self(s);
+       if (ret) {
+               return ret;
+       }
+       return proc_setup_thread_self(s);
+}
+
+static int proc_reconfigure(struct fs_context *fc)
+{
+       struct super_block *sb = fc->root->d_sb;
        struct pid_namespace *pid = sb->s_fs_info;
 
        sync_filesystem(sb);
-       return !proc_parse_options(data, pid);
+
+       proc_apply_options(sb, fc, pid, current_user_ns());
+       return 0;
 }
 
-static struct dentry *proc_mount(struct file_system_type *fs_type,
-       int flags, const char *dev_name, void *data)
+static int proc_get_tree(struct fs_context *fc)
 {
-       struct pid_namespace *ns;
+       struct proc_fs_context *ctx = fc->fs_private;
 
-       if (flags & SB_KERNMOUNT) {
-               ns = data;
-               data = NULL;
-       } else {
-               ns = task_active_pid_ns(current);
-       }
+       put_user_ns(fc->user_ns);
+       fc->user_ns = get_user_ns(ctx->pid_ns->user_ns);
+       fc->s_fs_info = ctx->pid_ns;
+       return vfs_get_super(fc, vfs_get_keyed_super, proc_fill_super);
+}
 
-       return mount_ns(fs_type, flags, data, ns, ns->user_ns, proc_fill_super);
+static void proc_fs_context_free(struct fs_context *fc)
+{
+       struct proc_fs_context *ctx = fc->fs_private;
+
+       if (ctx->pid_ns)
+               put_pid_ns(ctx->pid_ns);
+       kfree(ctx);
+}
+
+static const struct fs_context_operations proc_fs_context_ops = {
+       .free           = proc_fs_context_free,
+       .parse_param    = proc_parse_param,
+       .get_tree       = proc_get_tree,
+       .reconfigure    = proc_reconfigure,
+};
+
+static int proc_init_fs_context(struct fs_context *fc)
+{
+       struct proc_fs_context *ctx;
+
+       ctx = kzalloc(sizeof(struct proc_fs_context), GFP_KERNEL);
+       if (!ctx)
+               return -ENOMEM;
+
+       ctx->pid_ns = get_pid_ns(task_active_pid_ns(current));
+       fc->fs_private = ctx;
+       fc->ops = &proc_fs_context_ops;
+       return 0;
 }
 
 static void proc_kill_sb(struct super_block *sb)
@@ -115,10 +207,11 @@ static void proc_kill_sb(struct super_block *sb)
 }
 
 static struct file_system_type proc_fs_type = {
-       .name           = "proc",
-       .mount          = proc_mount,
-       .kill_sb        = proc_kill_sb,
-       .fs_flags       = FS_USERNS_MOUNT,
+       .name                   = "proc",
+       .init_fs_context        = proc_init_fs_context,
+       .parameters             = &proc_fs_parameters,
+       .kill_sb                = proc_kill_sb,
+       .fs_flags               = FS_USERNS_MOUNT,
 };
 
 void __init proc_root_init(void)
@@ -156,7 +249,7 @@ static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentr
 {
        if (!proc_pid_lookup(dentry, flags))
                return NULL;
-       
+
        return proc_lookup(dir, dentry, flags);
 }
 
@@ -209,9 +302,28 @@ struct proc_dir_entry proc_root = {
 
 int pid_ns_prepare_proc(struct pid_namespace *ns)
 {
+       struct proc_fs_context *ctx;
+       struct fs_context *fc;
        struct vfsmount *mnt;
 
-       mnt = kern_mount_data(&proc_fs_type, ns);
+       fc = fs_context_for_mount(&proc_fs_type, SB_KERNMOUNT);
+       if (IS_ERR(fc))
+               return PTR_ERR(fc);
+
+       if (fc->user_ns != ns->user_ns) {
+               put_user_ns(fc->user_ns);
+               fc->user_ns = get_user_ns(ns->user_ns);
+       }
+
+       ctx = fc->fs_private;
+       if (ctx->pid_ns != ns) {
+               put_pid_ns(ctx->pid_ns);
+               get_pid_ns(ns);
+               ctx->pid_ns = ns;
+       }
+
+       mnt = fc_mount(fc);
+       put_fs_context(fc);
        if (IS_ERR(mnt))
                return PTR_ERR(mnt);