kernel/capability.c

   1 /*
   2  * linux/kernel/capability.c
   3  *
   4  * Copyright (C) 1997  Andrew Main <zefram@fysh.org>
   5  *
   6  * Integrated into 2.1.97+,  Andrew G. Morgan <morgan@kernel.org>
   7  * 30 May 2002: Cleanup, Robert M. Love <rml@tech9.net>
   8  */
   9
  10 #include <linux/audit.h>
  11 #include <linux/capability.h>
  12 #include <linux/mm.h>
  13 #include <linux/export.h>
  14 #include <linux/security.h>
  15 #include <linux/syscalls.h>
  16 #include <linux/pid_namespace.h>
  17 #include <linux/user_namespace.h>
  18 #include <asm/uaccess.h>
  19
  20 /*
  21  * Leveraged for setting/resetting capabilities
  22  */
  23
  24 const kernel_cap_t __cap_empty_set = CAP_EMPTY_SET;
  25
  26 EXPORT_SYMBOL(__cap_empty_set);
  27
  28 int file_caps_enabled = 1;
  29
  30 static int __init file_caps_disable(char *str)
  31 {
  32         file_caps_enabled = 0;
  33         return 1;
  34 }
  35 __setup("no_file_caps", file_caps_disable);
  36
  37 /*
  38  * More recent versions of libcap are available from:
  39  *
  40  *   http://www.kernel.org/pub/linux/libs/security/linux-privs/
  41  */
  42
  43 static void warn_legacy_capability_use(void)
  44 {
  45         static int warned;
  46         if (!warned) {
  47                 char name[sizeof(current->comm)];
  48
  49                 printk(KERN_INFO "warning: `%s' uses 32-bit capabilities"
  50                        " (legacy support in use)\n",
  51                        get_task_comm(name, current));
  52                 warned = 1;
  53         }
  54 }
  55
  56 /*
  57  * Version 2 capabilities worked fine, but the linux/capability.h file
  58  * that accompanied their introduction encouraged their use without
  59  * the necessary user-space source code changes. As such, we have
  60  * created a version 3 with equivalent functionality to version 2, but
  61  * with a header change to protect legacy source code from using
  62  * version 2 when it wanted to use version 1. If your system has code
  63  * that trips the following warning, it is using version 2 specific
  64  * capabilities and may be doing so insecurely.
  65  *
  66  * The remedy is to either upgrade your version of libcap (to 2.10+,
  67  * if the application is linked against it), or recompile your
  68  * application with modern kernel headers and this warning will go
  69  * away.
  70  */
  71
  72 static void warn_deprecated_v2(void)
  73 {
  74         static int warned;
  75
  76         if (!warned) {
  77                 char name[sizeof(current->comm)];
  78
  79                 printk(KERN_INFO "warning: `%s' uses deprecated v2"
  80                        " capabilities in a way that may be insecure.\n",
  81                        get_task_comm(name, current));
  82                 warned = 1;
  83         }
  84 }
  85
  86 /*
  87  * Version check. Return the number of u32s in each capability flag
  88  * array, or a negative value on error.
  89  */
  90 static int cap_validate_magic(cap_user_header_t header, unsigned *tocopy)
  91 {
  92         __u32 version;
  93
  94         if (get_user(version, &header->version))
  95                 return -EFAULT;
  96
  97         switch (version) {
  98         case _LINUX_CAPABILITY_VERSION_1:
  99                 warn_legacy_capability_use();
 100                 *tocopy = _LINUX_CAPABILITY_U32S_1;
 101                 break;
 102         case _LINUX_CAPABILITY_VERSION_2:
 103                 warn_deprecated_v2();
 104                 /*
 105                  * fall through - v3 is otherwise equivalent to v2.
 106                  */
 107         case _LINUX_CAPABILITY_VERSION_3:
 108                 *tocopy = _LINUX_CAPABILITY_U32S_3;
 109                 break;
 110         default:
 111                 if (put_user((u32)_KERNEL_CAPABILITY_VERSION, &header->version))
 112                         return -EFAULT;
 113                 return -EINVAL;
 114         }
 115
 116         return 0;
 117 }
 118
 119 /*
 120  * The only thing that can change the capabilities of the current
 121  * process is the current process. As such, we can't be in this code
 122  * at the same time as we are in the process of setting capabilities
 123  * in this process. The net result is that we can limit our use of
 124  * locks to when we are reading the caps of another process.
 125  */
 126 static inline int cap_get_target_pid(pid_t pid, kernel_cap_t *pEp,
 127                                      kernel_cap_t *pIp, kernel_cap_t *pPp)
 128 {
 129         int ret;
 130
 131         if (pid && (pid != task_pid_vnr(current))) {
 132                 struct task_struct *target;
 133
 134                 rcu_read_lock();
 135
 136                 target = find_task_by_vpid(pid);
 137                 if (!target)
 138                         ret = -ESRCH;
 139                 else
 140                         ret = security_capget(target, pEp, pIp, pPp);
 141
 142                 rcu_read_unlock();
 143         } else
 144                 ret = security_capget(current, pEp, pIp, pPp);
 145
 146         return ret;
 147 }
 148
 149 /**
 150  * sys_capget - get the capabilities of a given process.
 151  * @header: pointer to struct that contains capability version and
 152  *      target pid data
 153  * @dataptr: pointer to struct that contains the effective, permitted,
 154  *      and inheritable capabilities that are returned
 155  *
 156  * Returns 0 on success and < 0 on error.
 157  */
 158 SYSCALL_DEFINE2(capget, cap_user_header_t, header, cap_user_data_t, dataptr)
 159 {
 160         int ret = 0;
 161         pid_t pid;
 162         unsigned tocopy;
 163         kernel_cap_t pE, pI, pP;
 164
 165         ret = cap_validate_magic(header, &tocopy);
 166         if ((dataptr == NULL) || (ret != 0))
 167                 return ((dataptr == NULL) && (ret == -EINVAL)) ? 0 : ret;
 168
 169         if (get_user(pid, &header->pid))
 170                 return -EFAULT;
 171
 172         if (pid < 0)
 173                 return -EINVAL;
 174
 175         ret = cap_get_target_pid(pid, &pE, &pI, &pP);
 176         if (!ret) {
 177                 struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
 178                 unsigned i;
 179
 180                 for (i = 0; i < tocopy; i++) {
 181                         kdata[i].effective = pE.cap[i];
 182                         kdata[i].permitted = pP.cap[i];
 183                         kdata[i].inheritable = pI.cap[i];
 184                 }
 185
 186                 /*
 187                  * Note, in the case, tocopy < _KERNEL_CAPABILITY_U32S,
 188                  * we silently drop the upper capabilities here. This
 189                  * has the effect of making older libcap
 190                  * implementations implicitly drop upper capability
 191                  * bits when they perform a: capget/modify/capset
 192                  * sequence.
 193                  *
 194                  * This behavior is considered fail-safe
 195                  * behavior. Upgrading the application to a newer
 196                  * version of libcap will enable access to the newer
 197                  * capabilities.
 198                  *
 199                  * An alternative would be to return an error here
 200                  * (-ERANGE), but that causes legacy applications to
 201                  * unexpectidly fail; the capget/modify/capset aborts
 202                  * before modification is attempted and the application
 203                  * fails.
 204                  */
 205                 if (copy_to_user(dataptr, kdata, tocopy
 206                                  * sizeof(struct __user_cap_data_struct))) {
 207                         return -EFAULT;
 208                 }
 209         }
 210
 211         return ret;
 212 }
 213
 214 /**
 215  * sys_capset - set capabilities for a process or (*) a group of processes
 216  * @header: pointer to struct that contains capability version and
 217  *      target pid data
 218  * @data: pointer to struct that contains the effective, permitted,
 219  *      and inheritable capabilities
 220  *
 221  * Set capabilities for the current process only.  The ability to any other
 222  * process(es) has been deprecated and removed.
 223  *
 224  * The restrictions on setting capabilities are specified as:
 225  *
 226  * I: any raised capabilities must be a subset of the old permitted
 227  * P: any raised capabilities must be a subset of the old permitted
 228  * E: must be set to a subset of new permitted
 229  *
 230  * Returns 0 on success and < 0 on error.
 231  */
 232 SYSCALL_DEFINE2(capset, cap_user_header_t, header, const cap_user_data_t, data)
 233 {
 234         struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
 235         unsigned i, tocopy, copybytes;
 236         kernel_cap_t inheritable, permitted, effective;
 237         struct cred *new;
 238         int ret;
 239         pid_t pid;
 240
 241         ret = cap_validate_magic(header, &tocopy);
 242         if (ret != 0)
 243                 return ret;
 244
 245         if (get_user(pid, &header->pid))
 246                 return -EFAULT;
 247
 248         /* may only affect current now */
 249         if (pid != 0 && pid != task_pid_vnr(current))
 250                 return -EPERM;
 251
 252         copybytes = tocopy * sizeof(struct __user_cap_data_struct);
 253         if (copybytes > sizeof(kdata))
 254                 return -EFAULT;
 255
 256         if (copy_from_user(&kdata, data, copybytes))
 257                 return -EFAULT;
 258
 259         for (i = 0; i < tocopy; i++) {
 260                 effective.cap[i] = kdata[i].effective;
 261                 permitted.cap[i] = kdata[i].permitted;
 262                 inheritable.cap[i] = kdata[i].inheritable;
 263         }
 264         while (i < _KERNEL_CAPABILITY_U32S) {
 265                 effective.cap[i] = 0;
 266                 permitted.cap[i] = 0;
 267                 inheritable.cap[i] = 0;
 268                 i++;
 269         }
 270
 271         new = prepare_creds();
 272         if (!new)
 273                 return -ENOMEM;
 274
 275         ret = security_capset(new, current_cred(),
 276                               &effective, &inheritable, &permitted);
 277         if (ret < 0)
 278                 goto error;
 279
 280         audit_log_capset(pid, new, current_cred());
 281
 282         return commit_creds(new);
 283
 284 error:
 285         abort_creds(new);
 286         return ret;
 287 }
 288
 289 /**
 290  * has_ns_capability - Does a task have a capability in a specific user ns
 291  * @t: The task in question
 292  * @ns: target user namespace
 293  * @cap: The capability to be tested for
 294  *
 295  * Return true if the specified task has the given superior capability
 296  * currently in effect to the specified user namespace, false if not.
 297  *
 298  * Note that this does not set PF_SUPERPRIV on the task.
 299  */
 300 bool has_ns_capability(struct task_struct *t,
 301                        struct user_namespace *ns, int cap)
 302 {
 303         int ret;
 304
 305         rcu_read_lock();
 306         ret = security_capable(__task_cred(t), ns, cap);
 307         rcu_read_unlock();
 308
 309         return (ret == 0);
 310 }
 311
 312 /**
 313  * has_capability - Does a task have a capability in init_user_ns
 314  * @t: The task in question
 315  * @cap: The capability to be tested for
 316  *
 317  * Return true if the specified task has the given superior capability
 318  * currently in effect to the initial user namespace, false if not.
 319  *
 320  * Note that this does not set PF_SUPERPRIV on the task.
 321  */
 322 bool has_capability(struct task_struct *t, int cap)
 323 {
 324         return has_ns_capability(t, &init_user_ns, cap);
 325 }
 326
 327 /**
 328  * has_ns_capability_noaudit - Does a task have a capability (unaudited)
 329  * in a specific user ns.
 330  * @t: The task in question
 331  * @ns: target user namespace
 332  * @cap: The capability to be tested for
 333  *
 334  * Return true if the specified task has the given superior capability
 335  * currently in effect to the specified user namespace, false if not.
 336  * Do not write an audit message for the check.
 337  *
 338  * Note that this does not set PF_SUPERPRIV on the task.
 339  */
 340 bool has_ns_capability_noaudit(struct task_struct *t,
 341                                struct user_namespace *ns, int cap)
 342 {
 343         int ret;
 344
 345         rcu_read_lock();
 346         ret = security_capable_noaudit(__task_cred(t), ns, cap);
 347         rcu_read_unlock();
 348
 349         return (ret == 0);
 350 }
 351
 352 /**
 353  * has_capability_noaudit - Does a task have a capability (unaudited) in the
 354  * initial user ns
 355  * @t: The task in question
 356  * @cap: The capability to be tested for
 357  *
 358  * Return true if the specified task has the given superior capability
 359  * currently in effect to init_user_ns, false if not.  Don't write an
 360  * audit message for the check.
 361  *
 362  * Note that this does not set PF_SUPERPRIV on the task.
 363  */
 364 bool has_capability_noaudit(struct task_struct *t, int cap)
 365 {
 366         return has_ns_capability_noaudit(t, &init_user_ns, cap);
 367 }
 368
 369 /**
 370  * ns_capable - Determine if the current task has a superior capability in effect
 371  * @ns:  The usernamespace we want the capability in
 372  * @cap: The capability to be tested for
 373  *
 374  * Return true if the current task has the given superior capability currently
 375  * available for use, false if not.
 376  *
 377  * This sets PF_SUPERPRIV on the task if the capability is available on the
 378  * assumption that it's about to be used.
 379  */
 380 bool ns_capable(struct user_namespace *ns, int cap)
 381 {
 382         if (unlikely(!cap_valid(cap))) {
 383                 printk(KERN_CRIT "capable() called with invalid cap=%u\n", cap);
 384                 BUG();
 385         }
 386
 387         if (security_capable(current_cred(), ns, cap) == 0) {
 388                 current->flags |= PF_SUPERPRIV;
 389                 return true;
 390         }
 391         return false;
 392 }
 393 EXPORT_SYMBOL(ns_capable);
 394
 395 /**
 396  * file_ns_capable - Determine if the file's opener had a capability in effect
 397  * @file:  The file we want to check
 398  * @ns:  The usernamespace we want the capability in
 399  * @cap: The capability to be tested for
 400  *
 401  * Return true if task that opened the file had a capability in effect
 402  * when the file was opened.
 403  *
 404  * This does not set PF_SUPERPRIV because the caller may not
 405  * actually be privileged.
 406  */
 407 bool file_ns_capable(const struct file *file, struct user_namespace *ns, int cap)
 408 {
 409         if (WARN_ON_ONCE(!cap_valid(cap)))
 410                 return false;
 411
 412         if (security_capable(file->f_cred, ns, cap) == 0)
 413                 return true;
 414
 415         return false;
 416 }
 417 EXPORT_SYMBOL(file_ns_capable);
 418
 419 /**
 420  * capable - Determine if the current task has a superior capability in effect
 421  * @cap: The capability to be tested for
 422  *
 423  * Return true if the current task has the given superior capability currently
 424  * available for use, false if not.
 425  *
 426  * This sets PF_SUPERPRIV on the task if the capability is available on the
 427  * assumption that it's about to be used.
 428  */
 429 bool capable(int cap)
 430 {
 431         return ns_capable(&init_user_ns, cap);
 432 }
 433 EXPORT_SYMBOL(capable);
 434
 435 /**
 436  * capable_wrt_inode_uidgid - Check nsown_capable and uid and gid mapped
 437  * @inode: The inode in question
 438  * @cap: The capability in question
 439  *
 440  * Return true if the current task has the given capability targeted at
 441  * its own user namespace and that the given inode's uid and gid are
 442  * mapped into the current user namespace.
 443  */
 444 bool capable_wrt_inode_uidgid(const struct inode *inode, int cap)
 445 {
 446         struct user_namespace *ns = current_user_ns();
 447
 448         return ns_capable(ns, cap) && kuid_has_mapping(ns, inode->i_uid) &&
 449                 kgid_has_mapping(ns, inode->i_gid);
 450 }
 451 EXPORT_SYMBOL(capable_wrt_inode_uidgid);