| 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 |
#include <linux/errno.h> |
| 10 |
#include <linux/time.h> |
| 11 |
#include <linux/proc_fs.h> |
| 12 |
#include <linux/stat.h> |
| 13 |
#include <linux/init.h> |
| 14 |
#include <linux/sched.h> |
| 15 |
#include <linux/sched/stat.h> |
| 16 |
#include <linux/module.h> |
| 17 |
#include <linux/bitops.h> |
| 18 |
#include <linux/user_namespace.h> |
| 19 |
#include <linux/fs_context.h> |
| 20 |
#include <linux/mount.h> |
| 21 |
#include <linux/pid_namespace.h> |
| 22 |
#include <linux/fs_parser.h> |
| 23 |
#include <linux/cred.h> |
| 24 |
#include <linux/magic.h> |
| 25 |
#include <linux/slab.h> |
| 26 |
|
| 27 |
#include "internal.h" |
| 28 |
|
| 29 |
struct proc_fs_context { |
| 30 |
struct pid_namespace *pid_ns; |
| 31 |
unsigned int mask; |
| 32 |
enum proc_hidepid hidepid; |
| 33 |
int gid; |
| 34 |
enum proc_pidonly pidonly; |
| 35 |
}; |
| 36 |
|
| 37 |
enum proc_param { |
| 38 |
Opt_gid, |
| 39 |
Opt_hidepid, |
| 40 |
Opt_subset, |
| 41 |
Opt_pidns, |
| 42 |
}; |
| 43 |
|
| 44 |
static const struct fs_parameter_spec proc_fs_parameters[] = { |
| 45 |
fsparam_u32("gid", Opt_gid), |
| 46 |
fsparam_string("hidepid", Opt_hidepid), |
| 47 |
fsparam_string("subset", Opt_subset), |
| 48 |
fsparam_file_or_string("pidns", Opt_pidns), |
| 49 |
{} |
| 50 |
}; |
| 51 |
|
| 52 |
static inline int valid_hidepid(unsigned int value) |
| 53 |
{ |
| 54 |
return (value == HIDEPID_OFF || |
| 55 |
value == HIDEPID_NO_ACCESS || |
| 56 |
value == HIDEPID_INVISIBLE || |
| 57 |
value == HIDEPID_NOT_PTRACEABLE); |
| 58 |
} |
| 59 |
|
| 60 |
static int proc_parse_hidepid_param(struct fs_context *fc, struct fs_parameter *param) |
| 61 |
{ |
| 62 |
struct proc_fs_context *ctx = fc->fs_private; |
| 63 |
struct fs_parameter_spec hidepid_u32_spec = fsparam_u32("hidepid", Opt_hidepid); |
| 64 |
struct fs_parse_result result; |
| 65 |
int base = (unsigned long)hidepid_u32_spec.data; |
| 66 |
|
| 67 |
if (param->type != fs_value_is_string) |
| 68 |
return invalf(fc, "proc: unexpected type of hidepid value\n"); |
| 69 |
|
| 70 |
if (!kstrtouint(param->string, base, &result.uint_32)) { |
| 71 |
if (!valid_hidepid(result.uint_32)) |
| 72 |
return invalf(fc, "proc: unknown value of hidepid - %s\n", param->string); |
| 73 |
ctx->hidepid = result.uint_32; |
| 74 |
return 0; |
| 75 |
} |
| 76 |
|
| 77 |
if (!strcmp(param->string, "off")) |
| 78 |
ctx->hidepid = HIDEPID_OFF; |
| 79 |
else if (!strcmp(param->string, "noaccess")) |
| 80 |
ctx->hidepid = HIDEPID_NO_ACCESS; |
| 81 |
else if (!strcmp(param->string, "invisible")) |
| 82 |
ctx->hidepid = HIDEPID_INVISIBLE; |
| 83 |
else if (!strcmp(param->string, "ptraceable")) |
| 84 |
ctx->hidepid = HIDEPID_NOT_PTRACEABLE; |
| 85 |
else |
| 86 |
return invalf(fc, "proc: unknown value of hidepid - %s\n", param->string); |
| 87 |
|
| 88 |
return 0; |
| 89 |
} |
| 90 |
|
| 91 |
static int proc_parse_subset_param(struct fs_context *fc, char *value) |
| 92 |
{ |
| 93 |
struct proc_fs_context *ctx = fc->fs_private; |
| 94 |
|
| 95 |
while (value) { |
| 96 |
char *ptr = strchr(value, ','); |
| 97 |
|
| 98 |
if (ptr != NULL) |
| 99 |
*ptr++ = '\0'; |
| 100 |
|
| 101 |
if (*value != '\0') { |
| 102 |
if (!strcmp(value, "pid")) { |
| 103 |
ctx->pidonly = PROC_PIDONLY_ON; |
| 104 |
} else { |
| 105 |
return invalf(fc, "proc: unsupported subset option - %s\n", value); |
| 106 |
} |
| 107 |
} |
| 108 |
value = ptr; |
| 109 |
} |
| 110 |
|
| 111 |
return 0; |
| 112 |
} |
| 113 |
|
| 114 |
#ifdef CONFIG_PID_NS |
| 115 |
static int proc_parse_pidns_param(struct fs_context *fc, |
| 116 |
struct fs_parameter *param, |
| 117 |
struct fs_parse_result *result) |
| 118 |
{ |
| 119 |
struct proc_fs_context *ctx = fc->fs_private; |
| 120 |
struct pid_namespace *target, *active = task_active_pid_ns(current); |
| 121 |
struct ns_common *ns; |
| 122 |
struct file *ns_filp __free(fput) = NULL; |
| 123 |
|
| 124 |
switch (param->type) { |
| 125 |
case fs_value_is_file: |
| 126 |
/* came through fsconfig, steal the file reference */ |
| 127 |
ns_filp = no_free_ptr(param->file); |
| 128 |
break; |
| 129 |
case fs_value_is_string: |
| 130 |
ns_filp = filp_open(param->string, O_RDONLY, 0); |
| 131 |
break; |
| 132 |
default: |
| 133 |
WARN_ON_ONCE(true); |
| 134 |
break; |
| 135 |
} |
| 136 |
if (!ns_filp) |
| 137 |
ns_filp = ERR_PTR(-EBADF); |
| 138 |
if (IS_ERR(ns_filp)) { |
| 139 |
errorfc(fc, "could not get file from pidns argument"); |
| 140 |
return PTR_ERR(ns_filp); |
| 141 |
} |
| 142 |
|
| 143 |
if (!proc_ns_file(ns_filp)) |
| 144 |
return invalfc(fc, "pidns argument is not an nsfs file"); |
| 145 |
ns = get_proc_ns(file_inode(ns_filp)); |
| 146 |
if (ns->ns_type != CLONE_NEWPID) |
| 147 |
return invalfc(fc, "pidns argument is not a pidns file"); |
| 148 |
target = container_of(ns, struct pid_namespace, ns); |
| 149 |
|
| 150 |
/* |
| 151 |
* pidns= is shorthand for joining the pidns to get a fsopen fd, so the |
| 152 |
* permission model should be the same as pidns_install(). |
| 153 |
*/ |
| 154 |
if (!ns_capable(target->user_ns, CAP_SYS_ADMIN)) { |
| 155 |
errorfc(fc, "insufficient permissions to set pidns"); |
| 156 |
return -EPERM; |
| 157 |
} |
| 158 |
if (!pidns_is_ancestor(target, active)) |
| 159 |
return invalfc(fc, "cannot set pidns to non-descendant pidns"); |
| 160 |
|
| 161 |
put_pid_ns(ctx->pid_ns); |
| 162 |
ctx->pid_ns = get_pid_ns(target); |
| 163 |
put_user_ns(fc->user_ns); |
| 164 |
fc->user_ns = get_user_ns(ctx->pid_ns->user_ns); |
| 165 |
return 0; |
| 166 |
} |
| 167 |
#endif /* CONFIG_PID_NS */ |
| 168 |
|
| 169 |
static int proc_parse_param(struct fs_context *fc, struct fs_parameter *param) |
| 170 |
{ |
| 171 |
struct proc_fs_context *ctx = fc->fs_private; |
| 172 |
struct fs_parse_result result; |
| 173 |
int opt, err; |
| 174 |
|
| 175 |
opt = fs_parse(fc, proc_fs_parameters, param, &result); |
| 176 |
if (opt < 0) |
| 177 |
return opt; |
| 178 |
|
| 179 |
switch (opt) { |
| 180 |
case Opt_gid: |
| 181 |
ctx->gid = result.uint_32; |
| 182 |
break; |
| 183 |
|
| 184 |
case Opt_hidepid: |
| 185 |
err = proc_parse_hidepid_param(fc, param); |
| 186 |
if (err) |
| 187 |
return err; |
| 188 |
break; |
| 189 |
|
| 190 |
case Opt_subset: |
| 191 |
err = proc_parse_subset_param(fc, param->string); |
| 192 |
if (err) |
| 193 |
return err; |
| 194 |
break; |
| 195 |
|
| 196 |
case Opt_pidns: |
| 197 |
#ifdef CONFIG_PID_NS |
| 198 |
/* |
| 199 |
* We would have to RCU-protect every proc_pid_ns() or |
| 200 |
* proc_sb_info() access if we allowed this to be reconfigured |
| 201 |
* for an existing procfs instance. Luckily, procfs instances |
| 202 |
* are cheap to create, and mount-beneath would let you |
| 203 |
* atomically replace an instance even with overmounts. |
| 204 |
*/ |
| 205 |
if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { |
| 206 |
errorfc(fc, "cannot reconfigure pidns for existing procfs"); |
| 207 |
return -EBUSY; |
| 208 |
} |
| 209 |
err = proc_parse_pidns_param(fc, param, &result); |
| 210 |
if (err) |
| 211 |
return err; |
| 212 |
break; |
| 213 |
#else |
| 214 |
errorfc(fc, "pidns mount flag not supported on this system"); |
| 215 |
return -EOPNOTSUPP; |
| 216 |
#endif |
| 217 |
|
| 218 |
default: |
| 219 |
return -EINVAL; |
| 220 |
} |
| 221 |
|
| 222 |
ctx->mask |= 1 << opt; |
| 223 |
return 0; |
| 224 |
} |
| 225 |
|
| 226 |
static void proc_apply_options(struct proc_fs_info *fs_info, |
| 227 |
struct fs_context *fc, |
| 228 |
struct user_namespace *user_ns) |
| 229 |
{ |
| 230 |
struct proc_fs_context *ctx = fc->fs_private; |
| 231 |
|
| 232 |
if (ctx->mask & (1 << Opt_gid)) |
| 233 |
fs_info->pid_gid = make_kgid(user_ns, ctx->gid); |
| 234 |
if (ctx->mask & (1 << Opt_hidepid)) |
| 235 |
fs_info->hide_pid = ctx->hidepid; |
| 236 |
if (ctx->mask & (1 << Opt_subset)) |
| 237 |
fs_info->pidonly = ctx->pidonly; |
| 238 |
if (ctx->mask & (1 << Opt_pidns) && |
| 239 |
!WARN_ON_ONCE(fc->purpose == FS_CONTEXT_FOR_RECONFIGURE)) { |
| 240 |
put_pid_ns(fs_info->pid_ns); |
| 241 |
fs_info->pid_ns = get_pid_ns(ctx->pid_ns); |
| 242 |
} |
| 243 |
} |
| 244 |
|
| 245 |
static int proc_fill_super(struct super_block *s, struct fs_context *fc) |
| 246 |
{ |
| 247 |
struct proc_fs_context *ctx = fc->fs_private; |
| 248 |
struct inode *root_inode; |
| 249 |
struct proc_fs_info *fs_info; |
| 250 |
int ret; |
| 251 |
|
| 252 |
fs_info = kzalloc_obj(*fs_info); |
| 253 |
if (!fs_info) |
| 254 |
return -ENOMEM; |
| 255 |
|
| 256 |
fs_info->pid_ns = get_pid_ns(ctx->pid_ns); |
| 257 |
proc_apply_options(fs_info, fc, current_user_ns()); |
| 258 |
|
| 259 |
/* User space would break if executables or devices appear on proc */ |
| 260 |
s->s_iflags |= SB_I_USERNS_VISIBLE | SB_I_NOEXEC | SB_I_NODEV; |
| 261 |
s->s_flags |= SB_NODIRATIME | SB_NOSUID | SB_NOEXEC; |
| 262 |
s->s_blocksize = 1024; |
| 263 |
s->s_blocksize_bits = 10; |
| 264 |
s->s_magic = PROC_SUPER_MAGIC; |
| 265 |
s->s_op = &proc_sops; |
| 266 |
s->s_time_gran = 1; |
| 267 |
s->s_fs_info = fs_info; |
| 268 |
|
| 269 |
/* |
| 270 |
* procfs isn't actually a stacking filesystem; however, there is |
| 271 |
* too much magic going on inside it to permit stacking things on |
| 272 |
* top of it |
| 273 |
*/ |
| 274 |
s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH; |
| 275 |
|
| 276 |
/* procfs dentries and inodes don't require IO to create */ |
| 277 |
s->s_shrink->seeks = 0; |
| 278 |
|
| 279 |
pde_get(&proc_root); |
| 280 |
root_inode = proc_get_inode(s, &proc_root); |
| 281 |
if (!root_inode) { |
| 282 |
pr_err("proc_fill_super: get root inode failed\n"); |
| 283 |
return -ENOMEM; |
| 284 |
} |
| 285 |
|
| 286 |
s->s_root = d_make_root(root_inode); |
| 287 |
if (!s->s_root) { |
| 288 |
pr_err("proc_fill_super: allocate dentry failed\n"); |
| 289 |
return -ENOMEM; |
| 290 |
} |
| 291 |
|
| 292 |
ret = proc_setup_self(s); |
| 293 |
if (ret) { |
| 294 |
return ret; |
| 295 |
} |
| 296 |
return proc_setup_thread_self(s); |
| 297 |
} |
| 298 |
|
| 299 |
static int proc_reconfigure(struct fs_context *fc) |
| 300 |
{ |
| 301 |
struct super_block *sb = fc->root->d_sb; |
| 302 |
struct proc_fs_info *fs_info = proc_sb_info(sb); |
| 303 |
|
| 304 |
sync_filesystem(sb); |
| 305 |
|
| 306 |
proc_apply_options(fs_info, fc, current_user_ns()); |
| 307 |
return 0; |
| 308 |
} |
| 309 |
|
| 310 |
static int proc_get_tree(struct fs_context *fc) |
| 311 |
{ |
| 312 |
return get_tree_nodev(fc, proc_fill_super); |
| 313 |
} |
| 314 |
|
| 315 |
static void proc_fs_context_free(struct fs_context *fc) |
| 316 |
{ |
| 317 |
struct proc_fs_context *ctx = fc->fs_private; |
| 318 |
|
| 319 |
put_pid_ns(ctx->pid_ns); |
| 320 |
kfree(ctx); |
| 321 |
} |
| 322 |
|
| 323 |
static const struct fs_context_operations proc_fs_context_ops = { |
| 324 |
.free = proc_fs_context_free, |
| 325 |
.parse_param = proc_parse_param, |
| 326 |
.get_tree = proc_get_tree, |
| 327 |
.reconfigure = proc_reconfigure, |
| 328 |
}; |
| 329 |
|
| 330 |
static int proc_init_fs_context(struct fs_context *fc) |
| 331 |
{ |
| 332 |
struct proc_fs_context *ctx; |
| 333 |
|
| 334 |
ctx = kzalloc_obj(struct proc_fs_context); |
| 335 |
if (!ctx) |
| 336 |
return -ENOMEM; |
| 337 |
|
| 338 |
ctx->pid_ns = get_pid_ns(task_active_pid_ns(current)); |
| 339 |
put_user_ns(fc->user_ns); |
| 340 |
fc->user_ns = get_user_ns(ctx->pid_ns->user_ns); |
| 341 |
fc->fs_private = ctx; |
| 342 |
fc->ops = &proc_fs_context_ops; |
| 343 |
return 0; |
| 344 |
} |
| 345 |
|
| 346 |
static void proc_kill_sb(struct super_block *sb) |
| 347 |
{ |
| 348 |
struct proc_fs_info *fs_info = proc_sb_info(sb); |
| 349 |
|
| 350 |
kill_anon_super(sb); |
| 351 |
if (fs_info) { |
| 352 |
put_pid_ns(fs_info->pid_ns); |
| 353 |
kfree_rcu(fs_info, rcu); |
| 354 |
} |
| 355 |
} |
| 356 |
|
| 357 |
static struct file_system_type proc_fs_type = { |
| 358 |
.name = "proc", |
| 359 |
.init_fs_context = proc_init_fs_context, |
| 360 |
.parameters = proc_fs_parameters, |
| 361 |
.kill_sb = proc_kill_sb, |
| 362 |
.fs_flags = FS_USERNS_MOUNT | FS_DISALLOW_NOTIFY_PERM, |
| 363 |
}; |
| 364 |
|
| 365 |
void __init proc_root_init(void) |
| 366 |
{ |
| 367 |
proc_init_kmemcache(); |
| 368 |
set_proc_pid_nlink(); |
| 369 |
proc_self_init(); |
| 370 |
proc_thread_self_init(); |
| 371 |
proc_symlink("mounts", NULL, "self/mounts"); |
| 372 |
|
| 373 |
proc_net_init(); |
| 374 |
proc_mkdir("fs", NULL); |
| 375 |
proc_mkdir("driver", NULL); |
| 376 |
proc_create_mount_point("fs/nfsd"); /* somewhere for the nfsd filesystem to be mounted */ |
| 377 |
#if defined(CONFIG_SUN_OPENPROMFS) || defined(CONFIG_SUN_OPENPROMFS_MODULE) |
| 378 |
/* just give it a mountpoint */ |
| 379 |
proc_create_mount_point("openprom"); |
| 380 |
#endif |
| 381 |
proc_tty_init(); |
| 382 |
proc_mkdir("bus", NULL); |
| 383 |
proc_sys_init(); |
| 384 |
|
| 385 |
/* |
| 386 |
* Last things last. It is not like userspace processes eager |
| 387 |
* to open /proc files exist at this point but register last |
| 388 |
* anyway. |
| 389 |
*/ |
| 390 |
register_filesystem(&proc_fs_type); |
| 391 |
} |
| 392 |
|
| 393 |
static int proc_root_getattr(struct mnt_idmap *idmap, |
| 394 |
const struct path *path, struct kstat *stat, |
| 395 |
u32 request_mask, unsigned int query_flags) |
| 396 |
{ |
| 397 |
generic_fillattr(&nop_mnt_idmap, request_mask, d_inode(path->dentry), |
| 398 |
stat); |
| 399 |
stat->nlink = proc_root.nlink + nr_processes(); |
| 400 |
return 0; |
| 401 |
} |
| 402 |
|
| 403 |
static struct dentry *proc_root_lookup(struct inode * dir, struct dentry * dentry, unsigned int flags) |
| 404 |
{ |
| 405 |
if (!proc_pid_lookup(dentry, flags)) |
| 406 |
return NULL; |
| 407 |
|
| 408 |
return proc_lookup(dir, dentry, flags); |
| 409 |
} |
| 410 |
|
| 411 |
static int proc_root_readdir(struct file *file, struct dir_context *ctx) |
| 412 |
{ |
| 413 |
if (ctx->pos < FIRST_PROCESS_ENTRY) { |
| 414 |
int error = proc_readdir(file, ctx); |
| 415 |
if (unlikely(error <= 0)) |
| 416 |
return error; |
| 417 |
ctx->pos = FIRST_PROCESS_ENTRY; |
| 418 |
} |
| 419 |
|
| 420 |
return proc_pid_readdir(file, ctx); |
| 421 |
} |
| 422 |
|
| 423 |
/* |
| 424 |
* The root /proc directory is special, as it has the |
| 425 |
* <pid> directories. Thus we don't use the generic |
| 426 |
* directory handling functions for that.. |
| 427 |
*/ |
| 428 |
static const struct file_operations proc_root_operations = { |
| 429 |
.read = generic_read_dir, |
| 430 |
.iterate_shared = proc_root_readdir, |
| 431 |
.llseek = generic_file_llseek, |
| 432 |
}; |
| 433 |
|
| 434 |
/* |
| 435 |
* proc root can do almost nothing.. |
| 436 |
*/ |
| 437 |
static const struct inode_operations proc_root_inode_operations = { |
| 438 |
.lookup = proc_root_lookup, |
| 439 |
.getattr = proc_root_getattr, |
| 440 |
}; |
| 441 |
|
| 442 |
/* |
| 443 |
* This is the root "inode" in the /proc tree.. |
| 444 |
*/ |
| 445 |
struct proc_dir_entry proc_root = { |
| 446 |
.low_ino = PROCFS_ROOT_INO, |
| 447 |
.namelen = 5, |
| 448 |
.mode = S_IFDIR | S_IRUGO | S_IXUGO, |
| 449 |
.nlink = 2, |
| 450 |
.refcnt = REFCOUNT_INIT(1), |
| 451 |
.proc_iops = &proc_root_inode_operations, |
| 452 |
.proc_dir_ops = &proc_root_operations, |
| 453 |
.parent = &proc_root, |
| 454 |
.subdir = RB_ROOT, |
| 455 |
.name = "/proc", |
| 456 |
}; |