fs/proc/root.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/root.c
   4  *
   5  *  Copyright (C) 1991, 1992 Linus Torvalds
   6  *
   7  *  proc root directory handling functions
   8  */
   9
  10 #include <linux/uaccess.h>
  11
  12 #include <linux/errno.h>
  13 #include <linux/time.h>
  14 #include <linux/proc_fs.h>
  15 #include <linux/stat.h>
  16 #include <linux/init.h>
  17 #include <linux/sched.h>
  18 #include <linux/sched/stat.h>
  19 #include <linux/module.h>
  20 #include <linux/bitops.h>
  21 #include <linux/user_namespace.h>
  22 #include <linux/fs_context.h>
  23 #include <linux/mount.h>
  24 #include <linux/pid_namespace.h>
  25 #include <linux/fs_parser.h>
  26 #include <linux/cred.h>
  27 #include <linux/magic.h>
  28 #include <linux/slab.h>
  29
  30 #include "internal.h"
  31
  32 struct proc_fs_context {
  33         struct pid_namespace    *pid_ns;
  34         unsigned int            mask;
  35         int                     hidepid;
  36         int                     gid;
  37 };
  38
  39 enum proc_param {
  40         Opt_gid,
  41         Opt_hidepid,
  42 };
  43
  44 static const struct fs_parameter_spec proc_fs_parameters[] = {
  45         fsparam_u32("gid",      Opt_gid),
  46         fsparam_u32("hidepid",  Opt_hidepid),
  47         {}
  48 };
  49
  50 static int proc_parse_param(struct fs_context *fc, struct fs_parameter *param)
  51 {
  52         struct proc_fs_context *ctx = fc->fs_private;
  53         struct fs_parse_result result;
  54         int opt;
  55
  56         opt = fs_parse(fc, proc_fs_parameters, param, &result);
  57         if (opt < 0)
  58                 return opt;
  59
  60         switch (opt) {
  61         case Opt_gid:
  62                 ctx->gid = result.uint_32;
  63                 break;
  64
  65         case Opt_hidepid:
  66                 ctx->hidepid = result.uint_32;
  67                 if (ctx->hidepid < HIDEPID_OFF ||
  68                     ctx->hidepid > HIDEPID_INVISIBLE)
  69                         return invalfc(fc, "hidepid value must be between 0 and 2.\n");
  70                 break;
  71
  72         default:
  73                 return -EINVAL;
  74         }
  75
  76         ctx->mask |= 1 << opt;
  77         return 0;
  78 }
  79
  80 static void proc_apply_options(struct super_block *s,
  81                                struct fs_context *fc,
  82                                struct pid_namespace *pid_ns,
  83                                struct user_namespace *user_ns)
  84 {
  85         struct proc_fs_context *ctx = fc->fs_private;
  86
  87         if (ctx->mask & (1 << Opt_gid))
  88                 pid_ns->pid_gid = make_kgid(user_ns, ctx->gid);
  89         if (ctx->mask & (1 << Opt_hidepid))
  90                 pid_ns->hide_pid = ctx->hidepid;
  91 }
  92
  93 static int proc_fill_super(struct super_block *s, struct fs_context *fc)
  94 {
  95         struct pid_namespace *pid_ns = get_pid_ns(s->s_fs_info);
  96         struct inode *root_inode;
  97         int ret;
  98
  99         proc_apply_options(s, fc, pid_ns, current_user_ns());
 100
 101         /* User space would break if executables or devices appear on proc */
 102         s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV;
 103         s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC;
 104         s->s_blocksize = 1024;
 105         s->s_blocksize_bits = 10;
 106         s->s_magic = PROC_SUPER_MAGIC;
 107         s->s_op = &proc_sops;
 108         s->s_time_gran = 1;
 109
 110         /*
 111          * procfs isn't actually a stacking filesystem; however, there is
 112          * too much magic going on inside it to permit stacking things on
 113          * top of it
 114          */
 115         s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
 116
 117         /* procfs dentries and inodes don't require IO to create */
 118         s->s_shrink.seeks = 0;
 119
 120         pde_get(&proc_root);
 121         root_inode = proc_get_inode(s, &proc_root);
 122         if (!root_inode) {
 123                 pr_err("proc_fill_super: get root inode failed\n");
 124                 return -ENOMEM;
 125         }
 126
 127         s->s_root = d_make_root(root_inode);
 128         if (!s->s_root) {
 129                 pr_err("proc_fill_super: allocate dentry failed\n");
 130                 return -ENOMEM;
 131         }
 132
 133         ret = proc_setup_self(s);
 134         if (ret) {
 135                 return ret;
 136         }
 137         return proc_setup_thread_self(s);
 138 }
 139
 140 static int proc_reconfigure(struct fs_context *fc)
 141 {
 142         struct super_block *sb = fc->root->d_sb;
 143         struct pid_namespace *pid = sb->s_fs_info;
 144
 145         sync_filesystem(sb);
 146
 147         proc_apply_options(sb, fc, pid, current_user_ns());
 148         return 0;
 149 }
 150
 151 static int proc_get_tree(struct fs_context *fc)
 152 {
 153         struct proc_fs_context *ctx = fc->fs_private;
 154
 155         return get_tree_keyed(fc, proc_fill_super, ctx->pid_ns);
 156 }
 157
 158 static void proc_fs_context_free(struct fs_context *fc)
 159 {
 160         struct proc_fs_context *ctx = fc->fs_private;
 161
 162         put_pid_ns(ctx->pid_ns);
 163         kfree(ctx);
 164 }
 165
 166 static const struct fs_context_operations proc_fs_context_ops = {
 167         .free           = proc_fs_context_free,
 168         .parse_param    = proc_parse_param,
 169         .get_tree       = proc_get_tree,
 170         .reconfigure    = proc_reconfigure,
 171 };
 172
 173 static int proc_init_fs_context(struct fs_context *fc)
 174 {
 175         struct proc_fs_context *ctx;
 176
 177         ctx = kzalloc(sizeof(struct proc_fs_context), GFP_KERNEL);
 178         if (!ctx)
 179                 return -ENOMEM;
 180
 181         ctx->pid_ns = get_pid_ns(task_active_pid_ns(current));
 182         put_user_ns(fc->user_ns);
 183         fc->user_ns = get_user_ns(ctx->pid_ns->user_ns);
 184         fc->fs_private = ctx;
 185         fc->ops = &proc_fs_context_ops;
 186         return 0;
 187 }
 188
 189 static void proc_kill_sb(struct super_block *sb)
 190 {
 191         struct pid_namespace *ns;
 192
 193         ns = (struct pid_namespace *)sb->s_fs_info;
 194         if (ns->proc_self)
 195                 dput(ns->proc_self);
 196         if (ns->proc_thread_self)
 197                 dput(ns->proc_thread_self);
 198         kill_anon_super(sb);
 199         put_pid_ns(ns);
 200 }
 201
 202 static struct file_system_type proc_fs_type = {
 203         .name                   = "proc",
 204         .init_fs_context        = proc_init_fs_context,
 205         .parameters             = proc_fs_parameters,
 206         .kill_sb                = proc_kill_sb,
 207         .fs_flags               = FS_USERNS_MOUNT | FS_DISALLOW_NOTIFY_PERM,
 208 };
 209
 210 void __init proc_root_init(void)
 211 {
 212         proc_init_kmemcache();
 213         set_proc_pid_nlink();
 214         proc_self_init();
 215         proc_thread_self_init();
 216         proc_symlink("mounts", NULL, "self/mounts");
 217
 218         proc_net_init();
 219         proc_mkdir("fs", NULL);
 220         proc_mkdir("driver", NULL);
 221         proc_create_mount_point("fs/nfsd"); /* somewhere for the nfsd filesystem to be mounted */
 222 #if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE)
 223         /* just give it a mountpoint */
 224         proc_create_mount_point("openprom");
 225 #endif
 226         proc_tty_init();
 227         proc_mkdir("bus", NULL);
 228         proc_sys_init();
 229
 230         register_filesystem(&proc_fs_type);
 231 }
 232
 233 static int proc_root_getattr(const struct path *path, struct kstat *stat,
 234                              u32 request_mask, unsigned int query_flags)
 235 {
 236         generic_fillattr(d_inode(path->dentry), stat);
 237         stat->nlink = proc_root.nlink + nr_processes();
 238         return 0;
 239 }
 240
 241 static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags)
 242 {
 243         if (!proc_pid_lookup(dentry, flags))
 244                 return NULL;
 245
 246         return proc_lookup(dir, dentry, flags);
 247 }
 248
 249 static int proc_root_readdir(struct file *file, struct dir_context *ctx)
 250 {
 251         if (ctx->pos < FIRST_PROCESS_ENTRY) {
 252                 int error = proc_readdir(file, ctx);
 253                 if (unlikely(error <= 0))
 254                         return error;
 255                 ctx->pos = FIRST_PROCESS_ENTRY;
 256         }
 257
 258         return proc_pid_readdir(file, ctx);
 259 }
 260
 261 /*
 262  * The root /proc directory is special, as it has the
 263  * <pid> directories. Thus we don't use the generic
 264  * directory handling functions for that..
 265  */
 266 static const struct file_operations proc_root_operations = {
 267         .read            = generic_read_dir,
 268         .iterate_shared  = proc_root_readdir,
 269         .llseek         = generic_file_llseek,
 270 };
 271
 272 /*
 273  * proc root can do almost nothing..
 274  */
 275 static const struct inode_operations proc_root_inode_operations = {
 276         .lookup         = proc_root_lookup,
 277         .getattr        = proc_root_getattr,
 278 };
 279
 280 /*
 281  * This is the root "inode" in the /proc tree..
 282  */
 283 struct proc_dir_entry proc_root = {
 284         .low_ino        = PROC_ROOT_INO,
 285         .namelen        = 5,
 286         .mode           = S_IFDIR | S_IRUGO | S_IXUGO,
 287         .nlink          = 2,
 288         .refcnt         = REFCOUNT_INIT(1),
 289         .proc_iops      = &proc_root_inode_operations,
 290         .proc_dir_ops   = &proc_root_operations,
 291         .parent         = &proc_root,
 292         .subdir         = RB_ROOT,
 293         .name           = "/proc",
 294 };
 295
 296 int pid_ns_prepare_proc(struct pid_namespace *ns)
 297 {
 298         struct proc_fs_context *ctx;
 299         struct fs_context *fc;
 300         struct vfsmount *mnt;
 301
 302         fc = fs_context_for_mount(&proc_fs_type, SB_KERNMOUNT);
 303         if (IS_ERR(fc))
 304                 return PTR_ERR(fc);
 305
 306         if (fc->user_ns != ns->user_ns) {
 307                 put_user_ns(fc->user_ns);
 308                 fc->user_ns = get_user_ns(ns->user_ns);
 309         }
 310
 311         ctx = fc->fs_private;
 312         if (ctx->pid_ns != ns) {
 313                 put_pid_ns(ctx->pid_ns);
 314                 get_pid_ns(ns);
 315                 ctx->pid_ns = ns;
 316         }
 317
 318         mnt = fc_mount(fc);
 319         put_fs_context(fc);
 320         if (IS_ERR(mnt))
 321                 return PTR_ERR(mnt);
 322
 323         ns->proc_mnt = mnt;
 324         return 0;
 325 }
 326
 327 void pid_ns_release_proc(struct pid_namespace *ns)
 328 {
 329         kern_unmount(ns->proc_mnt);
 330 }