linux/drivers/char/random.c
<<
>>
Prefs
   1/*
   2 * random.c -- A strong random number generator
   3 *
   4 * Copyright (C) 2017 Jason A. Donenfeld <Jason@zx2c4.com>. All
   5 * Rights Reserved.
   6 *
   7 * Copyright Matt Mackall <mpm@selenic.com>, 2003, 2004, 2005
   8 *
   9 * Copyright Theodore Ts'o, 1994, 1995, 1996, 1997, 1998, 1999.  All
  10 * rights reserved.
  11 *
  12 * Redistribution and use in source and binary forms, with or without
  13 * modification, are permitted provided that the following conditions
  14 * are met:
  15 * 1. Redistributions of source code must retain the above copyright
  16 *    notice, and the entire permission notice in its entirety,
  17 *    including the disclaimer of warranties.
  18 * 2. Redistributions in binary form must reproduce the above copyright
  19 *    notice, this list of conditions and the following disclaimer in the
  20 *    documentation and/or other materials provided with the distribution.
  21 * 3. The name of the author may not be used to endorse or promote
  22 *    products derived from this software without specific prior
  23 *    written permission.
  24 *
  25 * ALTERNATIVELY, this product may be distributed under the terms of
  26 * the GNU General Public License, in which case the provisions of the GPL are
  27 * required INSTEAD OF the above restrictions.  (This clause is
  28 * necessary due to a potential bad interaction between the GPL and
  29 * the restrictions contained in a BSD-style copyright.)
  30 *
  31 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  32 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  33 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ALL OF
  34 * WHICH ARE HEREBY DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE
  35 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  36 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT
  37 * OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  38 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  39 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  40 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
  41 * USE OF THIS SOFTWARE, EVEN IF NOT ADVISED OF THE POSSIBILITY OF SUCH
  42 * DAMAGE.
  43 */
  44
  45/*
  46 * (now, with legal B.S. out of the way.....)
  47 *
  48 * This routine gathers environmental noise from device drivers, etc.,
  49 * and returns good random numbers, suitable for cryptographic use.
  50 * Besides the obvious cryptographic uses, these numbers are also good
  51 * for seeding TCP sequence numbers, and other places where it is
  52 * desirable to have numbers which are not only random, but hard to
  53 * predict by an attacker.
  54 *
  55 * Theory of operation
  56 * ===================
  57 *
  58 * Computers are very predictable devices.  Hence it is extremely hard
  59 * to produce truly random numbers on a computer --- as opposed to
  60 * pseudo-random numbers, which can easily generated by using a
  61 * algorithm.  Unfortunately, it is very easy for attackers to guess
  62 * the sequence of pseudo-random number generators, and for some
  63 * applications this is not acceptable.  So instead, we must try to
  64 * gather "environmental noise" from the computer's environment, which
  65 * must be hard for outside attackers to observe, and use that to
  66 * generate random numbers.  In a Unix environment, this is best done
  67 * from inside the kernel.
  68 *
  69 * Sources of randomness from the environment include inter-keyboard
  70 * timings, inter-interrupt timings from some interrupts, and other
  71 * events which are both (a) non-deterministic and (b) hard for an
  72 * outside observer to measure.  Randomness from these sources are
  73 * added to an "entropy pool", which is mixed using a CRC-like function.
  74 * This is not cryptographically strong, but it is adequate assuming
  75 * the randomness is not chosen maliciously, and it is fast enough that
  76 * the overhead of doing it on every interrupt is very reasonable.
  77 * As random bytes are mixed into the entropy pool, the routines keep
  78 * an *estimate* of how many bits of randomness have been stored into
  79 * the random number generator's internal state.
  80 *
  81 * When random bytes are desired, they are obtained by taking the SHA
  82 * hash of the contents of the "entropy pool".  The SHA hash avoids
  83 * exposing the internal state of the entropy pool.  It is believed to
  84 * be computationally infeasible to derive any useful information
  85 * about the input of SHA from its output.  Even if it is possible to
  86 * analyze SHA in some clever way, as long as the amount of data
  87 * returned from the generator is less than the inherent entropy in
  88 * the pool, the output data is totally unpredictable.  For this
  89 * reason, the routine decreases its internal estimate of how many
  90 * bits of "true randomness" are contained in the entropy pool as it
  91 * outputs random numbers.
  92 *
  93 * If this estimate goes to zero, the routine can still generate
  94 * random numbers; however, an attacker may (at least in theory) be
  95 * able to infer the future output of the generator from prior
  96 * outputs.  This requires successful cryptanalysis of SHA, which is
  97 * not believed to be feasible, but there is a remote possibility.
  98 * Nonetheless, these numbers should be useful for the vast majority
  99 * of purposes.
 100 *
 101 * Exported interfaces ---- output
 102 * ===============================
 103 *
 104 * There are four exported interfaces; two for use within the kernel,
 105 * and two or use from userspace.
 106 *
 107 * Exported interfaces ---- userspace output
 108 * -----------------------------------------
 109 *
 110 * The userspace interfaces are two character devices /dev/random and
 111 * /dev/urandom.  /dev/random is suitable for use when very high
 112 * quality randomness is desired (for example, for key generation or
 113 * one-time pads), as it will only return a maximum of the number of
 114 * bits of randomness (as estimated by the random number generator)
 115 * contained in the entropy pool.
 116 *
 117 * The /dev/urandom device does not have this limit, and will return
 118 * as many bytes as are requested.  As more and more random bytes are
 119 * requested without giving time for the entropy pool to recharge,
 120 * this will result in random numbers that are merely cryptographically
 121 * strong.  For many applications, however, this is acceptable.
 122 *
 123 * Exported interfaces ---- kernel output
 124 * --------------------------------------
 125 *
 126 * The primary kernel interface is
 127 *
 128 *      void get_random_bytes(void *buf, int nbytes);
 129 *
 130 * This interface will return the requested number of random bytes,
 131 * and place it in the requested buffer.  This is equivalent to a
 132 * read from /dev/urandom.
 133 *
 134 * For less critical applications, there are the functions:
 135 *
 136 *      u32 get_random_u32()
 137 *      u64 get_random_u64()
 138 *      unsigned int get_random_int()
 139 *      unsigned long get_random_long()
 140 *
 141 * These are produced by a cryptographic RNG seeded from get_random_bytes,
 142 * and so do not deplete the entropy pool as much.  These are recommended
 143 * for most in-kernel operations *if the result is going to be stored in
 144 * the kernel*.
 145 *
 146 * Specifically, the get_random_int() family do not attempt to do
 147 * "anti-backtracking".  If you capture the state of the kernel (e.g.
 148 * by snapshotting the VM), you can figure out previous get_random_int()
 149 * return values.  But if the value is stored in the kernel anyway,
 150 * this is not a problem.
 151 *
 152 * It *is* safe to expose get_random_int() output to attackers (e.g. as
 153 * network cookies); given outputs 1..n, it's not feasible to predict
 154 * outputs 0 or n+1.  The only concern is an attacker who breaks into
 155 * the kernel later; the get_random_int() engine is not reseeded as
 156 * often as the get_random_bytes() one.
 157 *
 158 * get_random_bytes() is needed for keys that need to stay secret after
 159 * they are erased from the kernel.  For example, any key that will
 160 * be wrapped and stored encrypted.  And session encryption keys: we'd
 161 * like to know that after the session is closed and the keys erased,
 162 * the plaintext is unrecoverable to someone who recorded the ciphertext.
 163 *
 164 * But for network ports/cookies, stack canaries, PRNG seeds, address
 165 * space layout randomization, session *authentication* keys, or other
 166 * applications where the sensitive data is stored in the kernel in
 167 * plaintext for as long as it's sensitive, the get_random_int() family
 168 * is just fine.
 169 *
 170 * Consider ASLR.  We want to keep the address space secret from an
 171 * outside attacker while the process is running, but once the address
 172 * space is torn down, it's of no use to an attacker any more.  And it's
 173 * stored in kernel data structures as long as it's alive, so worrying
 174 * about an attacker's ability to extrapolate it from the get_random_int()
 175 * CRNG is silly.
 176 *
 177 * Even some cryptographic keys are safe to generate with get_random_int().
 178 * In particular, keys for SipHash are generally fine.  Here, knowledge
 179 * of the key authorizes you to do something to a kernel object (inject
 180 * packets to a network connection, or flood a hash table), and the
 181 * key is stored with the object being protected.  Once it goes away,
 182 * we no longer care if anyone knows the key.
 183 *
 184 * prandom_u32()
 185 * -------------
 186 *
 187 * For even weaker applications, see the pseudorandom generator
 188 * prandom_u32(), prandom_max(), and prandom_bytes().  If the random
 189 * numbers aren't security-critical at all, these are *far* cheaper.
 190 * Useful for self-tests, random error simulation, randomized backoffs,
 191 * and any other application where you trust that nobody is trying to
 192 * maliciously mess with you by guessing the "random" numbers.
 193 *
 194 * Exported interfaces ---- input
 195 * ==============================
 196 *
 197 * The current exported interfaces for gathering environmental noise
 198 * from the devices are:
 199 *
 200 *      void add_device_randomness(const void *buf, unsigned int size);
 201 *      void add_input_randomness(unsigned int type, unsigned int code,
 202 *                                unsigned int value);
 203 *      void add_interrupt_randomness(int irq, int irq_flags);
 204 *      void add_disk_randomness(struct gendisk *disk);
 205 *
 206 * add_device_randomness() is for adding data to the random pool that
 207 * is likely to differ between two devices (or possibly even per boot).
 208 * This would be things like MAC addresses or serial numbers, or the
 209 * read-out of the RTC. This does *not* add any actual entropy to the
 210 * pool, but it initializes the pool to different values for devices
 211 * that might otherwise be identical and have very little entropy
 212 * available to them (particularly common in the embedded world).
 213 *
 214 * add_input_randomness() uses the input layer interrupt timing, as well as
 215 * the event type information from the hardware.
 216 *
 217 * add_interrupt_randomness() uses the interrupt timing as random
 218 * inputs to the entropy pool. Using the cycle counters and the irq source
 219 * as inputs, it feeds the randomness roughly once a second.
 220 *
 221 * add_disk_randomness() uses what amounts to the seek time of block
 222 * layer request events, on a per-disk_devt basis, as input to the
 223 * entropy pool. Note that high-speed solid state drives with very low
 224 * seek times do not make for good sources of entropy, as their seek
 225 * times are usually fairly consistent.
 226 *
 227 * All of these routines try to estimate how many bits of randomness a
 228 * particular randomness source.  They do this by keeping track of the
 229 * first and second order deltas of the event timings.
 230 *
 231 * Ensuring unpredictability at system startup
 232 * ============================================
 233 *
 234 * When any operating system starts up, it will go through a sequence
 235 * of actions that are fairly predictable by an adversary, especially
 236 * if the start-up does not involve interaction with a human operator.
 237 * This reduces the actual number of bits of unpredictability in the
 238 * entropy pool below the value in entropy_count.  In order to
 239 * counteract this effect, it helps to carry information in the
 240 * entropy pool across shut-downs and start-ups.  To do this, put the
 241 * following lines an appropriate script which is run during the boot
 242 * sequence:
 243 *
 244 *      echo "Initializing random number generator..."
 245 *      random_seed=/var/run/random-seed
 246 *      # Carry a random seed from start-up to start-up
 247 *      # Load and then save the whole entropy pool
 248 *      if [ -f $random_seed ]; then
 249 *              cat $random_seed >/dev/urandom
 250 *      else
 251 *              touch $random_seed
 252 *      fi
 253 *      chmod 600 $random_seed
 254 *      dd if=/dev/urandom of=$random_seed count=1 bs=512
 255 *
 256 * and the following lines in an appropriate script which is run as
 257 * the system is shutdown:
 258 *
 259 *      # Carry a random seed from shut-down to start-up
 260 *      # Save the whole entropy pool
 261 *      echo "Saving random seed..."
 262 *      random_seed=/var/run/random-seed
 263 *      touch $random_seed
 264 *      chmod 600 $random_seed
 265 *      dd if=/dev/urandom of=$random_seed count=1 bs=512
 266 *
 267 * For example, on most modern systems using the System V init
 268 * scripts, such code fragments would be found in
 269 * /etc/rc.d/init.d/random.  On older Linux systems, the correct script
 270 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
 271 *
 272 * Effectively, these commands cause the contents of the entropy pool
 273 * to be saved at shut-down time and reloaded into the entropy pool at
 274 * start-up.  (The 'dd' in the addition to the bootup script is to
 275 * make sure that /etc/random-seed is different for every start-up,
 276 * even if the system crashes without executing rc.0.)  Even with
 277 * complete knowledge of the start-up activities, predicting the state
 278 * of the entropy pool requires knowledge of the previous history of
 279 * the system.
 280 *
 281 * Configuring the /dev/random driver under Linux
 282 * ==============================================
 283 *
 284 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
 285 * the /dev/mem major number (#1).  So if your system does not have
 286 * /dev/random and /dev/urandom created already, they can be created
 287 * by using the commands:
 288 *
 289 *      mknod /dev/random c 1 8
 290 *      mknod /dev/urandom c 1 9
 291 *
 292 * Acknowledgements:
 293 * =================
 294 *
 295 * Ideas for constructing this random number generator were derived
 296 * from Pretty Good Privacy's random number generator, and from private
 297 * discussions with Phil Karn.  Colin Plumb provided a faster random
 298 * number generator, which speed up the mixing function of the entropy
 299 * pool, taken from PGPfone.  Dale Worley has also contributed many
 300 * useful ideas and suggestions to improve this driver.
 301 *
 302 * Any flaws in the design are solely my responsibility, and should
 303 * not be attributed to the Phil, Colin, or any of authors of PGP.
 304 *
 305 * Further background information on this topic may be obtained from
 306 * RFC 1750, "Randomness Recommendations for Security", by Donald
 307 * Eastlake, Steve Crocker, and Jeff Schiller.
 308 */
 309
 310#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 311
 312#include <linux/utsname.h>
 313#include <linux/module.h>
 314#include <linux/kernel.h>
 315#include <linux/major.h>
 316#include <linux/string.h>
 317#include <linux/fcntl.h>
 318#include <linux/slab.h>
 319#include <linux/random.h>
 320#include <linux/poll.h>
 321#include <linux/init.h>
 322#include <linux/fs.h>
 323#include <linux/genhd.h>
 324#include <linux/interrupt.h>
 325#include <linux/mm.h>
 326#include <linux/nodemask.h>
 327#include <linux/spinlock.h>
 328#include <linux/kthread.h>
 329#include <linux/percpu.h>
 330#include <linux/cryptohash.h>
 331#include <linux/fips.h>
 332#include <linux/ptrace.h>
 333#include <linux/workqueue.h>
 334#include <linux/irq.h>
 335#include <linux/ratelimit.h>
 336#include <linux/syscalls.h>
 337#include <linux/completion.h>
 338#include <linux/uuid.h>
 339#include <crypto/chacha.h>
 340
 341#include <asm/processor.h>
 342#include <linux/uaccess.h>
 343#include <asm/irq.h>
 344#include <asm/irq_regs.h>
 345#include <asm/io.h>
 346
 347#define CREATE_TRACE_POINTS
 348#include <trace/events/random.h>
 349
 350/* #define ADD_INTERRUPT_BENCH */
 351
 352/*
 353 * Configuration information
 354 */
 355#define INPUT_POOL_SHIFT        12
 356#define INPUT_POOL_WORDS        (1 << (INPUT_POOL_SHIFT-5))
 357#define OUTPUT_POOL_SHIFT       10
 358#define OUTPUT_POOL_WORDS       (1 << (OUTPUT_POOL_SHIFT-5))
 359#define EXTRACT_SIZE            10
 360
 361
 362#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
 363
 364/*
 365 * To allow fractional bits to be tracked, the entropy_count field is
 366 * denominated in units of 1/8th bits.
 367 *
 368 * 2*(ENTROPY_SHIFT + poolbitshift) must <= 31, or the multiply in
 369 * credit_entropy_bits() needs to be 64 bits wide.
 370 */
 371#define ENTROPY_SHIFT 3
 372#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
 373
 374/*
 375 * If the entropy count falls under this number of bits, then we
 376 * should wake up processes which are selecting or polling on write
 377 * access to /dev/random.
 378 */
 379static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
 380
 381/*
 382 * Originally, we used a primitive polynomial of degree .poolwords
 383 * over GF(2).  The taps for various sizes are defined below.  They
 384 * were chosen to be evenly spaced except for the last tap, which is 1
 385 * to get the twisting happening as fast as possible.
 386 *
 387 * For the purposes of better mixing, we use the CRC-32 polynomial as
 388 * well to make a (modified) twisted Generalized Feedback Shift
 389 * Register.  (See M. Matsumoto & Y. Kurita, 1992.  Twisted GFSR
 390 * generators.  ACM Transactions on Modeling and Computer Simulation
 391 * 2(3):179-194.  Also see M. Matsumoto & Y. Kurita, 1994.  Twisted
 392 * GFSR generators II.  ACM Transactions on Modeling and Computer
 393 * Simulation 4:254-266)
 394 *
 395 * Thanks to Colin Plumb for suggesting this.
 396 *
 397 * The mixing operation is much less sensitive than the output hash,
 398 * where we use SHA-1.  All that we want of mixing operation is that
 399 * it be a good non-cryptographic hash; i.e. it not produce collisions
 400 * when fed "random" data of the sort we expect to see.  As long as
 401 * the pool state differs for different inputs, we have preserved the
 402 * input entropy and done a good job.  The fact that an intelligent
 403 * attacker can construct inputs that will produce controlled
 404 * alterations to the pool's state is not important because we don't
 405 * consider such inputs to contribute any randomness.  The only
 406 * property we need with respect to them is that the attacker can't
 407 * increase his/her knowledge of the pool's state.  Since all
 408 * additions are reversible (knowing the final state and the input,
 409 * you can reconstruct the initial state), if an attacker has any
 410 * uncertainty about the initial state, he/she can only shuffle that
 411 * uncertainty about, but never cause any collisions (which would
 412 * decrease the uncertainty).
 413 *
 414 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
 415 * Videau in their paper, "The Linux Pseudorandom Number Generator
 416 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf).  In their
 417 * paper, they point out that we are not using a true Twisted GFSR,
 418 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
 419 * is, with only three taps, instead of the six that we are using).
 420 * As a result, the resulting polynomial is neither primitive nor
 421 * irreducible, and hence does not have a maximal period over
 422 * GF(2**32).  They suggest a slight change to the generator
 423 * polynomial which improves the resulting TGFSR polynomial to be
 424 * irreducible, which we have made here.
 425 */
 426static const struct poolinfo {
 427        int poolbitshift, poolwords, poolbytes, poolfracbits;
 428#define S(x) ilog2(x)+5, (x), (x)*4, (x) << (ENTROPY_SHIFT+5)
 429        int tap1, tap2, tap3, tap4, tap5;
 430} poolinfo_table[] = {
 431        /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
 432        /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
 433        { S(128),       104,    76,     51,     25,     1 },
 434};
 435
 436/*
 437 * Static global variables
 438 */
 439static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
 440static struct fasync_struct *fasync;
 441
 442static DEFINE_SPINLOCK(random_ready_list_lock);
 443static LIST_HEAD(random_ready_list);
 444
 445struct crng_state {
 446        __u32           state[16];
 447        unsigned long   init_time;
 448        spinlock_t      lock;
 449};
 450
 451static struct crng_state primary_crng = {
 452        .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
 453};
 454
 455/*
 456 * crng_init =  0 --> Uninitialized
 457 *              1 --> Initialized
 458 *              2 --> Initialized from input_pool
 459 *
 460 * crng_init is protected by primary_crng->lock, and only increases
 461 * its value (from 0->1->2).
 462 */
 463static int crng_init = 0;
 464#define crng_ready() (likely(crng_init > 1))
 465static int crng_init_cnt = 0;
 466static unsigned long crng_global_init_time = 0;
 467#define CRNG_INIT_CNT_THRESH (2*CHACHA_KEY_SIZE)
 468static void _extract_crng(struct crng_state *crng, __u8 out[CHACHA_BLOCK_SIZE]);
 469static void _crng_backtrack_protect(struct crng_state *crng,
 470                                    __u8 tmp[CHACHA_BLOCK_SIZE], int used);
 471static void process_random_ready_list(void);
 472static void _get_random_bytes(void *buf, int nbytes);
 473
 474static struct ratelimit_state unseeded_warning =
 475        RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
 476static struct ratelimit_state urandom_warning =
 477        RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
 478
 479static int ratelimit_disable __read_mostly;
 480
 481module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
 482MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
 483
 484/**********************************************************************
 485 *
 486 * OS independent entropy store.   Here are the functions which handle
 487 * storing entropy in an entropy pool.
 488 *
 489 **********************************************************************/
 490
 491struct entropy_store;
 492struct entropy_store {
 493        /* read-only data: */
 494        const struct poolinfo *poolinfo;
 495        __u32 *pool;
 496        const char *name;
 497
 498        /* read-write data: */
 499        spinlock_t lock;
 500        unsigned short add_ptr;
 501        unsigned short input_rotate;
 502        int entropy_count;
 503        unsigned int initialized:1;
 504        unsigned int last_data_init:1;
 505        __u8 last_data[EXTRACT_SIZE];
 506};
 507
 508static ssize_t extract_entropy(struct entropy_store *r, void *buf,
 509                               size_t nbytes, int min, int rsvd);
 510static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
 511                                size_t nbytes, int fips);
 512
 513static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
 514static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
 515
 516static struct entropy_store input_pool = {
 517        .poolinfo = &poolinfo_table[0],
 518        .name = "input",
 519        .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
 520        .pool = input_pool_data
 521};
 522
 523static __u32 const twist_table[8] = {
 524        0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
 525        0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
 526
 527/*
 528 * This function adds bytes into the entropy "pool".  It does not
 529 * update the entropy estimate.  The caller should call
 530 * credit_entropy_bits if this is appropriate.
 531 *
 532 * The pool is stirred with a primitive polynomial of the appropriate
 533 * degree, and then twisted.  We twist by three bits at a time because
 534 * it's cheap to do so and helps slightly in the expected case where
 535 * the entropy is concentrated in the low-order bits.
 536 */
 537static void _mix_pool_bytes(struct entropy_store *r, const void *in,
 538                            int nbytes)
 539{
 540        unsigned long i, tap1, tap2, tap3, tap4, tap5;
 541        int input_rotate;
 542        int wordmask = r->poolinfo->poolwords - 1;
 543        const char *bytes = in;
 544        __u32 w;
 545
 546        tap1 = r->poolinfo->tap1;
 547        tap2 = r->poolinfo->tap2;
 548        tap3 = r->poolinfo->tap3;
 549        tap4 = r->poolinfo->tap4;
 550        tap5 = r->poolinfo->tap5;
 551
 552        input_rotate = r->input_rotate;
 553        i = r->add_ptr;
 554
 555        /* mix one byte at a time to simplify size handling and churn faster */
 556        while (nbytes--) {
 557                w = rol32(*bytes++, input_rotate);
 558                i = (i - 1) & wordmask;
 559
 560                /* XOR in the various taps */
 561                w ^= r->pool[i];
 562                w ^= r->pool[(i + tap1) & wordmask];
 563                w ^= r->pool[(i + tap2) & wordmask];
 564                w ^= r->pool[(i + tap3) & wordmask];
 565                w ^= r->pool[(i + tap4) & wordmask];
 566                w ^= r->pool[(i + tap5) & wordmask];
 567
 568                /* Mix the result back in with a twist */
 569                r->pool[i] = (w >> 3) ^ twist_table[w & 7];
 570
 571                /*
 572                 * Normally, we add 7 bits of rotation to the pool.
 573                 * At the beginning of the pool, add an extra 7 bits
 574                 * rotation, so that successive passes spread the
 575                 * input bits across the pool evenly.
 576                 */
 577                input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
 578        }
 579
 580        r->input_rotate = input_rotate;
 581        r->add_ptr = i;
 582}
 583
 584static void __mix_pool_bytes(struct entropy_store *r, const void *in,
 585                             int nbytes)
 586{
 587        trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
 588        _mix_pool_bytes(r, in, nbytes);
 589}
 590
 591static void mix_pool_bytes(struct entropy_store *r, const void *in,
 592                           int nbytes)
 593{
 594        unsigned long flags;
 595
 596        trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
 597        spin_lock_irqsave(&r->lock, flags);
 598        _mix_pool_bytes(r, in, nbytes);
 599        spin_unlock_irqrestore(&r->lock, flags);
 600}
 601
 602struct fast_pool {
 603        __u32           pool[4];
 604        unsigned long   last;
 605        unsigned short  reg_idx;
 606        unsigned char   count;
 607};
 608
 609/*
 610 * This is a fast mixing routine used by the interrupt randomness
 611 * collector.  It's hardcoded for an 128 bit pool and assumes that any
 612 * locks that might be needed are taken by the caller.
 613 */
 614static void fast_mix(struct fast_pool *f)
 615{
 616        __u32 a = f->pool[0],   b = f->pool[1];
 617        __u32 c = f->pool[2],   d = f->pool[3];
 618
 619        a += b;                 c += d;
 620        b = rol32(b, 6);        d = rol32(d, 27);
 621        d ^= a;                 b ^= c;
 622
 623        a += b;                 c += d;
 624        b = rol32(b, 16);       d = rol32(d, 14);
 625        d ^= a;                 b ^= c;
 626
 627        a += b;                 c += d;
 628        b = rol32(b, 6);        d = rol32(d, 27);
 629        d ^= a;                 b ^= c;
 630
 631        a += b;                 c += d;
 632        b = rol32(b, 16);       d = rol32(d, 14);
 633        d ^= a;                 b ^= c;
 634
 635        f->pool[0] = a;  f->pool[1] = b;
 636        f->pool[2] = c;  f->pool[3] = d;
 637        f->count++;
 638}
 639
 640static void process_random_ready_list(void)
 641{
 642        unsigned long flags;
 643        struct random_ready_callback *rdy, *tmp;
 644
 645        spin_lock_irqsave(&random_ready_list_lock, flags);
 646        list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
 647                struct module *owner = rdy->owner;
 648
 649                list_del_init(&rdy->list);
 650                rdy->func(rdy);
 651                module_put(owner);
 652        }
 653        spin_unlock_irqrestore(&random_ready_list_lock, flags);
 654}
 655
 656/*
 657 * Credit (or debit) the entropy store with n bits of entropy.
 658 * Use credit_entropy_bits_safe() if the value comes from userspace
 659 * or otherwise should be checked for extreme values.
 660 */
 661static void credit_entropy_bits(struct entropy_store *r, int nbits)
 662{
 663        int entropy_count, orig, has_initialized = 0;
 664        const int pool_size = r->poolinfo->poolfracbits;
 665        int nfrac = nbits << ENTROPY_SHIFT;
 666
 667        if (!nbits)
 668                return;
 669
 670retry:
 671        entropy_count = orig = READ_ONCE(r->entropy_count);
 672        if (nfrac < 0) {
 673                /* Debit */
 674                entropy_count += nfrac;
 675        } else {
 676                /*
 677                 * Credit: we have to account for the possibility of
 678                 * overwriting already present entropy.  Even in the
 679                 * ideal case of pure Shannon entropy, new contributions
 680                 * approach the full value asymptotically:
 681                 *
 682                 * entropy <- entropy + (pool_size - entropy) *
 683                 *      (1 - exp(-add_entropy/pool_size))
 684                 *
 685                 * For add_entropy <= pool_size/2 then
 686                 * (1 - exp(-add_entropy/pool_size)) >=
 687                 *    (add_entropy/pool_size)*0.7869...
 688                 * so we can approximate the exponential with
 689                 * 3/4*add_entropy/pool_size and still be on the
 690                 * safe side by adding at most pool_size/2 at a time.
 691                 *
 692                 * The use of pool_size-2 in the while statement is to
 693                 * prevent rounding artifacts from making the loop
 694                 * arbitrarily long; this limits the loop to log2(pool_size)*2
 695                 * turns no matter how large nbits is.
 696                 */
 697                int pnfrac = nfrac;
 698                const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
 699                /* The +2 corresponds to the /4 in the denominator */
 700
 701                do {
 702                        unsigned int anfrac = min(pnfrac, pool_size/2);
 703                        unsigned int add =
 704                                ((pool_size - entropy_count)*anfrac*3) >> s;
 705
 706                        entropy_count += add;
 707                        pnfrac -= anfrac;
 708                } while (unlikely(entropy_count < pool_size-2 && pnfrac));
 709        }
 710
 711        if (WARN_ON(entropy_count < 0)) {
 712                pr_warn("negative entropy/overflow: pool %s count %d\n",
 713                        r->name, entropy_count);
 714                entropy_count = 0;
 715        } else if (entropy_count > pool_size)
 716                entropy_count = pool_size;
 717        if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
 718                goto retry;
 719
 720        if (has_initialized) {
 721                r->initialized = 1;
 722                kill_fasync(&fasync, SIGIO, POLL_IN);
 723        }
 724
 725        trace_credit_entropy_bits(r->name, nbits,
 726                                  entropy_count >> ENTROPY_SHIFT, _RET_IP_);
 727
 728        if (r == &input_pool) {
 729                int entropy_bits = entropy_count >> ENTROPY_SHIFT;
 730
 731                if (crng_init < 2) {
 732                        if (entropy_bits < 128)
 733                                return;
 734                        crng_reseed(&primary_crng, r);
 735                        entropy_bits = ENTROPY_BITS(r);
 736                }
 737        }
 738}
 739
 740static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
 741{
 742        const int nbits_max = r->poolinfo->poolwords * 32;
 743
 744        if (nbits < 0)
 745                return -EINVAL;
 746
 747        /* Cap the value to avoid overflows */
 748        nbits = min(nbits,  nbits_max);
 749
 750        credit_entropy_bits(r, nbits);
 751        return 0;
 752}
 753
 754/*********************************************************************
 755 *
 756 * CRNG using CHACHA20
 757 *
 758 *********************************************************************/
 759
 760#define CRNG_RESEED_INTERVAL (300*HZ)
 761
 762static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
 763
 764#ifdef CONFIG_NUMA
 765/*
 766 * Hack to deal with crazy userspace progams when they are all trying
 767 * to access /dev/urandom in parallel.  The programs are almost
 768 * certainly doing something terribly wrong, but we'll work around
 769 * their brain damage.
 770 */
 771static struct crng_state **crng_node_pool __read_mostly;
 772#endif
 773
 774static void invalidate_batched_entropy(void);
 775static void numa_crng_init(void);
 776
 777static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
 778static int __init parse_trust_cpu(char *arg)
 779{
 780        return kstrtobool(arg, &trust_cpu);
 781}
 782early_param("random.trust_cpu", parse_trust_cpu);
 783
 784static void crng_initialize(struct crng_state *crng)
 785{
 786        int             i;
 787        int             arch_init = 1;
 788        unsigned long   rv;
 789
 790        memcpy(&crng->state[0], "expand 32-byte k", 16);
 791        if (crng == &primary_crng)
 792                _extract_entropy(&input_pool, &crng->state[4],
 793                                 sizeof(__u32) * 12, 0);
 794        else
 795                _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
 796        for (i = 4; i < 16; i++) {
 797                if (!arch_get_random_seed_long(&rv) &&
 798                    !arch_get_random_long(&rv)) {
 799                        rv = random_get_entropy();
 800                        arch_init = 0;
 801                }
 802                crng->state[i] ^= rv;
 803        }
 804        if (trust_cpu && arch_init && crng == &primary_crng) {
 805                invalidate_batched_entropy();
 806                numa_crng_init();
 807                crng_init = 2;
 808                pr_notice("crng done (trusting CPU's manufacturer)\n");
 809        }
 810        crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
 811}
 812
 813#ifdef CONFIG_NUMA
 814static void do_numa_crng_init(struct work_struct *work)
 815{
 816        int i;
 817        struct crng_state *crng;
 818        struct crng_state **pool;
 819
 820        pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
 821        for_each_online_node(i) {
 822                crng = kmalloc_node(sizeof(struct crng_state),
 823                                    GFP_KERNEL | __GFP_NOFAIL, i);
 824                spin_lock_init(&crng->lock);
 825                crng_initialize(crng);
 826                pool[i] = crng;
 827        }
 828        mb();
 829        if (cmpxchg(&crng_node_pool, NULL, pool)) {
 830                for_each_node(i)
 831                        kfree(pool[i]);
 832                kfree(pool);
 833        }
 834}
 835
 836static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
 837
 838static void numa_crng_init(void)
 839{
 840        schedule_work(&numa_crng_init_work);
 841}
 842#else
 843static void numa_crng_init(void) {}
 844#endif
 845
 846/*
 847 * crng_fast_load() can be called by code in the interrupt service
 848 * path.  So we can't afford to dilly-dally.
 849 */
 850static int crng_fast_load(const char *cp, size_t len)
 851{
 852        unsigned long flags;
 853        char *p;
 854
 855        if (!spin_trylock_irqsave(&primary_crng.lock, flags))
 856                return 0;
 857        if (crng_init != 0) {
 858                spin_unlock_irqrestore(&primary_crng.lock, flags);
 859                return 0;
 860        }
 861        p = (unsigned char *) &primary_crng.state[4];
 862        while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
 863                p[crng_init_cnt % CHACHA_KEY_SIZE] ^= *cp;
 864                cp++; crng_init_cnt++; len--;
 865        }
 866        spin_unlock_irqrestore(&primary_crng.lock, flags);
 867        if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
 868                invalidate_batched_entropy();
 869                crng_init = 1;
 870                pr_notice("fast init done\n");
 871        }
 872        return 1;
 873}
 874
 875/*
 876 * crng_slow_load() is called by add_device_randomness, which has two
 877 * attributes.  (1) We can't trust the buffer passed to it is
 878 * guaranteed to be unpredictable (so it might not have any entropy at
 879 * all), and (2) it doesn't have the performance constraints of
 880 * crng_fast_load().
 881 *
 882 * So we do something more comprehensive which is guaranteed to touch
 883 * all of the primary_crng's state, and which uses a LFSR with a
 884 * period of 255 as part of the mixing algorithm.  Finally, we do
 885 * *not* advance crng_init_cnt since buffer we may get may be something
 886 * like a fixed DMI table (for example), which might very well be
 887 * unique to the machine, but is otherwise unvarying.
 888 */
 889static int crng_slow_load(const char *cp, size_t len)
 890{
 891        unsigned long           flags;
 892        static unsigned char    lfsr = 1;
 893        unsigned char           tmp;
 894        unsigned                i, max = CHACHA_KEY_SIZE;
 895        const char *            src_buf = cp;
 896        char *                  dest_buf = (char *) &primary_crng.state[4];
 897
 898        if (!spin_trylock_irqsave(&primary_crng.lock, flags))
 899                return 0;
 900        if (crng_init != 0) {
 901                spin_unlock_irqrestore(&primary_crng.lock, flags);
 902                return 0;
 903        }
 904        if (len > max)
 905                max = len;
 906
 907        for (i = 0; i < max ; i++) {
 908                tmp = lfsr;
 909                lfsr >>= 1;
 910                if (tmp & 1)
 911                        lfsr ^= 0xE1;
 912                tmp = dest_buf[i % CHACHA_KEY_SIZE];
 913                dest_buf[i % CHACHA_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
 914                lfsr += (tmp << 3) | (tmp >> 5);
 915        }
 916        spin_unlock_irqrestore(&primary_crng.lock, flags);
 917        return 1;
 918}
 919
 920static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
 921{
 922        unsigned long   flags;
 923        int             i, num;
 924        union {
 925                __u8    block[CHACHA_BLOCK_SIZE];
 926                __u32   key[8];
 927        } buf;
 928
 929        if (r) {
 930                num = extract_entropy(r, &buf, 32, 16, 0);
 931                if (num == 0)
 932                        return;
 933        } else {
 934                _extract_crng(&primary_crng, buf.block);
 935                _crng_backtrack_protect(&primary_crng, buf.block,
 936                                        CHACHA_KEY_SIZE);
 937        }
 938        spin_lock_irqsave(&crng->lock, flags);
 939        for (i = 0; i < 8; i++) {
 940                unsigned long   rv;
 941                if (!arch_get_random_seed_long(&rv) &&
 942                    !arch_get_random_long(&rv))
 943                        rv = random_get_entropy();
 944                crng->state[i+4] ^= buf.key[i] ^ rv;
 945        }
 946        memzero_explicit(&buf, sizeof(buf));
 947        crng->init_time = jiffies;
 948        spin_unlock_irqrestore(&crng->lock, flags);
 949        if (crng == &primary_crng && crng_init < 2) {
 950                invalidate_batched_entropy();
 951                numa_crng_init();
 952                crng_init = 2;
 953                process_random_ready_list();
 954                wake_up_interruptible(&crng_init_wait);
 955                kill_fasync(&fasync, SIGIO, POLL_IN);
 956                pr_notice("crng init done\n");
 957                if (unseeded_warning.missed) {
 958                        pr_notice("%d get_random_xx warning(s) missed due to ratelimiting\n",
 959                                  unseeded_warning.missed);
 960                        unseeded_warning.missed = 0;
 961                }
 962                if (urandom_warning.missed) {
 963                        pr_notice("%d urandom warning(s) missed due to ratelimiting\n",
 964                                  urandom_warning.missed);
 965                        urandom_warning.missed = 0;
 966                }
 967        }
 968}
 969
 970static void _extract_crng(struct crng_state *crng,
 971                          __u8 out[CHACHA_BLOCK_SIZE])
 972{
 973        unsigned long v, flags;
 974
 975        if (crng_ready() &&
 976            (time_after(crng_global_init_time, crng->init_time) ||
 977             time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL)))
 978                crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
 979        spin_lock_irqsave(&crng->lock, flags);
 980        if (arch_get_random_long(&v))
 981                crng->state[14] ^= v;
 982        chacha20_block(&crng->state[0], out);
 983        if (crng->state[12] == 0)
 984                crng->state[13]++;
 985        spin_unlock_irqrestore(&crng->lock, flags);
 986}
 987
 988static void extract_crng(__u8 out[CHACHA_BLOCK_SIZE])
 989{
 990        struct crng_state *crng = NULL;
 991
 992#ifdef CONFIG_NUMA
 993        if (crng_node_pool)
 994                crng = crng_node_pool[numa_node_id()];
 995        if (crng == NULL)
 996#endif
 997                crng = &primary_crng;
 998        _extract_crng(crng, out);
 999}
1000
1001/*
1002 * Use the leftover bytes from the CRNG block output (if there is
1003 * enough) to mutate the CRNG key to provide backtracking protection.
1004 */
1005static void _crng_backtrack_protect(struct crng_state *crng,
1006                                    __u8 tmp[CHACHA_BLOCK_SIZE], int used)
1007{
1008        unsigned long   flags;
1009        __u32           *s, *d;
1010        int             i;
1011
1012        used = round_up(used, sizeof(__u32));
1013        if (used + CHACHA_KEY_SIZE > CHACHA_BLOCK_SIZE) {
1014                extract_crng(tmp);
1015                used = 0;
1016        }
1017        spin_lock_irqsave(&crng->lock, flags);
1018        s = (__u32 *) &tmp[used];
1019        d = &crng->state[4];
1020        for (i=0; i < 8; i++)
1021                *d++ ^= *s++;
1022        spin_unlock_irqrestore(&crng->lock, flags);
1023}
1024
1025static void crng_backtrack_protect(__u8 tmp[CHACHA_BLOCK_SIZE], int used)
1026{
1027        struct crng_state *crng = NULL;
1028
1029#ifdef CONFIG_NUMA
1030        if (crng_node_pool)
1031                crng = crng_node_pool[numa_node_id()];
1032        if (crng == NULL)
1033#endif
1034                crng = &primary_crng;
1035        _crng_backtrack_protect(crng, tmp, used);
1036}
1037
1038static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1039{
1040        ssize_t ret = 0, i = CHACHA_BLOCK_SIZE;
1041        __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1042        int large_request = (nbytes > 256);
1043
1044        while (nbytes) {
1045                if (large_request && need_resched()) {
1046                        if (signal_pending(current)) {
1047                                if (ret == 0)
1048                                        ret = -ERESTARTSYS;
1049                                break;
1050                        }
1051                        schedule();
1052                }
1053
1054                extract_crng(tmp);
1055                i = min_t(int, nbytes, CHACHA_BLOCK_SIZE);
1056                if (copy_to_user(buf, tmp, i)) {
1057                        ret = -EFAULT;
1058                        break;
1059                }
1060
1061                nbytes -= i;
1062                buf += i;
1063                ret += i;
1064        }
1065        crng_backtrack_protect(tmp, i);
1066
1067        /* Wipe data just written to memory */
1068        memzero_explicit(tmp, sizeof(tmp));
1069
1070        return ret;
1071}
1072
1073
1074/*********************************************************************
1075 *
1076 * Entropy input management
1077 *
1078 *********************************************************************/
1079
1080/* There is one of these per entropy source */
1081struct timer_rand_state {
1082        cycles_t last_time;
1083        long last_delta, last_delta2;
1084};
1085
1086#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1087
1088/*
1089 * Add device- or boot-specific data to the input pool to help
1090 * initialize it.
1091 *
1092 * None of this adds any entropy; it is meant to avoid the problem of
1093 * the entropy pool having similar initial state across largely
1094 * identical devices.
1095 */
1096void add_device_randomness(const void *buf, unsigned int size)
1097{
1098        unsigned long time = random_get_entropy() ^ jiffies;
1099        unsigned long flags;
1100
1101        if (!crng_ready() && size)
1102                crng_slow_load(buf, size);
1103
1104        trace_add_device_randomness(size, _RET_IP_);
1105        spin_lock_irqsave(&input_pool.lock, flags);
1106        _mix_pool_bytes(&input_pool, buf, size);
1107        _mix_pool_bytes(&input_pool, &time, sizeof(time));
1108        spin_unlock_irqrestore(&input_pool.lock, flags);
1109}
1110EXPORT_SYMBOL(add_device_randomness);
1111
1112static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1113
1114/*
1115 * This function adds entropy to the entropy "pool" by using timing
1116 * delays.  It uses the timer_rand_state structure to make an estimate
1117 * of how many bits of entropy this call has added to the pool.
1118 *
1119 * The number "num" is also added to the pool - it should somehow describe
1120 * the type of event which just happened.  This is currently 0-255 for
1121 * keyboard scan codes, and 256 upwards for interrupts.
1122 *
1123 */
1124static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1125{
1126        struct entropy_store    *r;
1127        struct {
1128                long jiffies;
1129                unsigned cycles;
1130                unsigned num;
1131        } sample;
1132        long delta, delta2, delta3;
1133
1134        sample.jiffies = jiffies;
1135        sample.cycles = random_get_entropy();
1136        sample.num = num;
1137        r = &input_pool;
1138        mix_pool_bytes(r, &sample, sizeof(sample));
1139
1140        /*
1141         * Calculate number of bits of randomness we probably added.
1142         * We take into account the first, second and third-order deltas
1143         * in order to make our estimate.
1144         */
1145        delta = sample.jiffies - state->last_time;
1146        state->last_time = sample.jiffies;
1147
1148        delta2 = delta - state->last_delta;
1149        state->last_delta = delta;
1150
1151        delta3 = delta2 - state->last_delta2;
1152        state->last_delta2 = delta2;
1153
1154        if (delta < 0)
1155                delta = -delta;
1156        if (delta2 < 0)
1157                delta2 = -delta2;
1158        if (delta3 < 0)
1159                delta3 = -delta3;
1160        if (delta > delta2)
1161                delta = delta2;
1162        if (delta > delta3)
1163                delta = delta3;
1164
1165        /*
1166         * delta is now minimum absolute delta.
1167         * Round down by 1 bit on general principles,
1168         * and limit entropy estimate to 12 bits.
1169         */
1170        credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1171}
1172
1173void add_input_randomness(unsigned int type, unsigned int code,
1174                                 unsigned int value)
1175{
1176        static unsigned char last_value;
1177
1178        /* ignore autorepeat and the like */
1179        if (value == last_value)
1180                return;
1181
1182        last_value = value;
1183        add_timer_randomness(&input_timer_state,
1184                             (type << 4) ^ code ^ (code >> 4) ^ value);
1185        trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1186}
1187EXPORT_SYMBOL_GPL(add_input_randomness);
1188
1189static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1190
1191#ifdef ADD_INTERRUPT_BENCH
1192static unsigned long avg_cycles, avg_deviation;
1193
1194#define AVG_SHIFT 8     /* Exponential average factor k=1/256 */
1195#define FIXED_1_2 (1 << (AVG_SHIFT-1))
1196
1197static void add_interrupt_bench(cycles_t start)
1198{
1199        long delta = random_get_entropy() - start;
1200
1201        /* Use a weighted moving average */
1202        delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1203        avg_cycles += delta;
1204        /* And average deviation */
1205        delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1206        avg_deviation += delta;
1207}
1208#else
1209#define add_interrupt_bench(x)
1210#endif
1211
1212static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1213{
1214        __u32 *ptr = (__u32 *) regs;
1215        unsigned int idx;
1216
1217        if (regs == NULL)
1218                return 0;
1219        idx = READ_ONCE(f->reg_idx);
1220        if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1221                idx = 0;
1222        ptr += idx++;
1223        WRITE_ONCE(f->reg_idx, idx);
1224        return *ptr;
1225}
1226
1227void add_interrupt_randomness(int irq, int irq_flags)
1228{
1229        struct entropy_store    *r;
1230        struct fast_pool        *fast_pool = this_cpu_ptr(&irq_randomness);
1231        struct pt_regs          *regs = get_irq_regs();
1232        unsigned long           now = jiffies;
1233        cycles_t                cycles = random_get_entropy();
1234        __u32                   c_high, j_high;
1235        __u64                   ip;
1236        unsigned long           seed;
1237        int                     credit = 0;
1238
1239        if (cycles == 0)
1240                cycles = get_reg(fast_pool, regs);
1241        c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1242        j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1243        fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1244        fast_pool->pool[1] ^= now ^ c_high;
1245        ip = regs ? instruction_pointer(regs) : _RET_IP_;
1246        fast_pool->pool[2] ^= ip;
1247        fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1248                get_reg(fast_pool, regs);
1249
1250        fast_mix(fast_pool);
1251        add_interrupt_bench(cycles);
1252
1253        if (unlikely(crng_init == 0)) {
1254                if ((fast_pool->count >= 64) &&
1255                    crng_fast_load((char *) fast_pool->pool,
1256                                   sizeof(fast_pool->pool))) {
1257                        fast_pool->count = 0;
1258                        fast_pool->last = now;
1259                }
1260                return;
1261        }
1262
1263        if ((fast_pool->count < 64) &&
1264            !time_after(now, fast_pool->last + HZ))
1265                return;
1266
1267        r = &input_pool;
1268        if (!spin_trylock(&r->lock))
1269                return;
1270
1271        fast_pool->last = now;
1272        __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1273
1274        /*
1275         * If we have architectural seed generator, produce a seed and
1276         * add it to the pool.  For the sake of paranoia don't let the
1277         * architectural seed generator dominate the input from the
1278         * interrupt noise.
1279         */
1280        if (arch_get_random_seed_long(&seed)) {
1281                __mix_pool_bytes(r, &seed, sizeof(seed));
1282                credit = 1;
1283        }
1284        spin_unlock(&r->lock);
1285
1286        fast_pool->count = 0;
1287
1288        /* award one bit for the contents of the fast pool */
1289        credit_entropy_bits(r, credit + 1);
1290}
1291EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1292
1293#ifdef CONFIG_BLOCK
1294void add_disk_randomness(struct gendisk *disk)
1295{
1296        if (!disk || !disk->random)
1297                return;
1298        /* first major is 1, so we get >= 0x200 here */
1299        add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1300        trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1301}
1302EXPORT_SYMBOL_GPL(add_disk_randomness);
1303#endif
1304
1305/*********************************************************************
1306 *
1307 * Entropy extraction routines
1308 *
1309 *********************************************************************/
1310
1311/*
1312 * This function decides how many bytes to actually take from the
1313 * given pool, and also debits the entropy count accordingly.
1314 */
1315static size_t account(struct entropy_store *r, size_t nbytes, int min,
1316                      int reserved)
1317{
1318        int entropy_count, orig, have_bytes;
1319        size_t ibytes, nfrac;
1320
1321        BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1322
1323        /* Can we pull enough? */
1324retry:
1325        entropy_count = orig = READ_ONCE(r->entropy_count);
1326        ibytes = nbytes;
1327        /* never pull more than available */
1328        have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1329
1330        if ((have_bytes -= reserved) < 0)
1331                have_bytes = 0;
1332        ibytes = min_t(size_t, ibytes, have_bytes);
1333        if (ibytes < min)
1334                ibytes = 0;
1335
1336        if (WARN_ON(entropy_count < 0)) {
1337                pr_warn("negative entropy count: pool %s count %d\n",
1338                        r->name, entropy_count);
1339                entropy_count = 0;
1340        }
1341        nfrac = ibytes << (ENTROPY_SHIFT + 3);
1342        if ((size_t) entropy_count > nfrac)
1343                entropy_count -= nfrac;
1344        else
1345                entropy_count = 0;
1346
1347        if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1348                goto retry;
1349
1350        trace_debit_entropy(r->name, 8 * ibytes);
1351        if (ibytes && ENTROPY_BITS(r) < random_write_wakeup_bits) {
1352                wake_up_interruptible(&random_write_wait);
1353                kill_fasync(&fasync, SIGIO, POLL_OUT);
1354        }
1355
1356        return ibytes;
1357}
1358
1359/*
1360 * This function does the actual extraction for extract_entropy and
1361 * extract_entropy_user.
1362 *
1363 * Note: we assume that .poolwords is a multiple of 16 words.
1364 */
1365static void extract_buf(struct entropy_store *r, __u8 *out)
1366{
1367        int i;
1368        union {
1369                __u32 w[5];
1370                unsigned long l[LONGS(20)];
1371        } hash;
1372        __u32 workspace[SHA_WORKSPACE_WORDS];
1373        unsigned long flags;
1374
1375        /*
1376         * If we have an architectural hardware random number
1377         * generator, use it for SHA's initial vector
1378         */
1379        sha_init(hash.w);
1380        for (i = 0; i < LONGS(20); i++) {
1381                unsigned long v;
1382                if (!arch_get_random_long(&v))
1383                        break;
1384                hash.l[i] = v;
1385        }
1386
1387        /* Generate a hash across the pool, 16 words (512 bits) at a time */
1388        spin_lock_irqsave(&r->lock, flags);
1389        for (i = 0; i < r->poolinfo->poolwords; i += 16)
1390                sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1391
1392        /*
1393         * We mix the hash back into the pool to prevent backtracking
1394         * attacks (where the attacker knows the state of the pool
1395         * plus the current outputs, and attempts to find previous
1396         * ouputs), unless the hash function can be inverted. By
1397         * mixing at least a SHA1 worth of hash data back, we make
1398         * brute-forcing the feedback as hard as brute-forcing the
1399         * hash.
1400         */
1401        __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1402        spin_unlock_irqrestore(&r->lock, flags);
1403
1404        memzero_explicit(workspace, sizeof(workspace));
1405
1406        /*
1407         * In case the hash function has some recognizable output
1408         * pattern, we fold it in half. Thus, we always feed back
1409         * twice as much data as we output.
1410         */
1411        hash.w[0] ^= hash.w[3];
1412        hash.w[1] ^= hash.w[4];
1413        hash.w[2] ^= rol32(hash.w[2], 16);
1414
1415        memcpy(out, &hash, EXTRACT_SIZE);
1416        memzero_explicit(&hash, sizeof(hash));
1417}
1418
1419static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1420                                size_t nbytes, int fips)
1421{
1422        ssize_t ret = 0, i;
1423        __u8 tmp[EXTRACT_SIZE];
1424        unsigned long flags;
1425
1426        while (nbytes) {
1427                extract_buf(r, tmp);
1428
1429                if (fips) {
1430                        spin_lock_irqsave(&r->lock, flags);
1431                        if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1432                                panic("Hardware RNG duplicated output!\n");
1433                        memcpy(r->last_data, tmp, EXTRACT_SIZE);
1434                        spin_unlock_irqrestore(&r->lock, flags);
1435                }
1436                i = min_t(int, nbytes, EXTRACT_SIZE);
1437                memcpy(buf, tmp, i);
1438                nbytes -= i;
1439                buf += i;
1440                ret += i;
1441        }
1442
1443        /* Wipe data just returned from memory */
1444        memzero_explicit(tmp, sizeof(tmp));
1445
1446        return ret;
1447}
1448
1449/*
1450 * This function extracts randomness from the "entropy pool", and
1451 * returns it in a buffer.
1452 *
1453 * The min parameter specifies the minimum amount we can pull before
1454 * failing to avoid races that defeat catastrophic reseeding while the
1455 * reserved parameter indicates how much entropy we must leave in the
1456 * pool after each pull to avoid starving other readers.
1457 */
1458static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1459                                 size_t nbytes, int min, int reserved)
1460{
1461        __u8 tmp[EXTRACT_SIZE];
1462        unsigned long flags;
1463
1464        /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1465        if (fips_enabled) {
1466                spin_lock_irqsave(&r->lock, flags);
1467                if (!r->last_data_init) {
1468                        r->last_data_init = 1;
1469                        spin_unlock_irqrestore(&r->lock, flags);
1470                        trace_extract_entropy(r->name, EXTRACT_SIZE,
1471                                              ENTROPY_BITS(r), _RET_IP_);
1472                        extract_buf(r, tmp);
1473                        spin_lock_irqsave(&r->lock, flags);
1474                        memcpy(r->last_data, tmp, EXTRACT_SIZE);
1475                }
1476                spin_unlock_irqrestore(&r->lock, flags);
1477        }
1478
1479        trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1480        nbytes = account(r, nbytes, min, reserved);
1481
1482        return _extract_entropy(r, buf, nbytes, fips_enabled);
1483}
1484
1485#define warn_unseeded_randomness(previous) \
1486        _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1487
1488static void _warn_unseeded_randomness(const char *func_name, void *caller,
1489                                      void **previous)
1490{
1491#ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1492        const bool print_once = false;
1493#else
1494        static bool print_once __read_mostly;
1495#endif
1496
1497        if (print_once ||
1498            crng_ready() ||
1499            (previous && (caller == READ_ONCE(*previous))))
1500                return;
1501        WRITE_ONCE(*previous, caller);
1502#ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1503        print_once = true;
1504#endif
1505        if (__ratelimit(&unseeded_warning))
1506                printk_deferred(KERN_NOTICE "random: %s called from %pS "
1507                                "with crng_init=%d\n", func_name, caller,
1508                                crng_init);
1509}
1510
1511/*
1512 * This function is the exported kernel interface.  It returns some
1513 * number of good random numbers, suitable for key generation, seeding
1514 * TCP sequence numbers, etc.  It does not rely on the hardware random
1515 * number generator.  For random bytes direct from the hardware RNG
1516 * (when available), use get_random_bytes_arch(). In order to ensure
1517 * that the randomness provided by this function is okay, the function
1518 * wait_for_random_bytes() should be called and return 0 at least once
1519 * at any point prior.
1520 */
1521static void _get_random_bytes(void *buf, int nbytes)
1522{
1523        __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1524
1525        trace_get_random_bytes(nbytes, _RET_IP_);
1526
1527        while (nbytes >= CHACHA_BLOCK_SIZE) {
1528                extract_crng(buf);
1529                buf += CHACHA_BLOCK_SIZE;
1530                nbytes -= CHACHA_BLOCK_SIZE;
1531        }
1532
1533        if (nbytes > 0) {
1534                extract_crng(tmp);
1535                memcpy(buf, tmp, nbytes);
1536                crng_backtrack_protect(tmp, nbytes);
1537        } else
1538                crng_backtrack_protect(tmp, CHACHA_BLOCK_SIZE);
1539        memzero_explicit(tmp, sizeof(tmp));
1540}
1541
1542void get_random_bytes(void *buf, int nbytes)
1543{
1544        static void *previous;
1545
1546        warn_unseeded_randomness(&previous);
1547        _get_random_bytes(buf, nbytes);
1548}
1549EXPORT_SYMBOL(get_random_bytes);
1550
1551
1552/*
1553 * Each time the timer fires, we expect that we got an unpredictable
1554 * jump in the cycle counter. Even if the timer is running on another
1555 * CPU, the timer activity will be touching the stack of the CPU that is
1556 * generating entropy..
1557 *
1558 * Note that we don't re-arm the timer in the timer itself - we are
1559 * happy to be scheduled away, since that just makes the load more
1560 * complex, but we do not want the timer to keep ticking unless the
1561 * entropy loop is running.
1562 *
1563 * So the re-arming always happens in the entropy loop itself.
1564 */
1565static void entropy_timer(struct timer_list *t)
1566{
1567        credit_entropy_bits(&input_pool, 1);
1568}
1569
1570/*
1571 * If we have an actual cycle counter, see if we can
1572 * generate enough entropy with timing noise
1573 */
1574static void try_to_generate_entropy(void)
1575{
1576        struct {
1577                unsigned long now;
1578                struct timer_list timer;
1579        } stack;
1580
1581        stack.now = random_get_entropy();
1582
1583        /* Slow counter - or none. Don't even bother */
1584        if (stack.now == random_get_entropy())
1585                return;
1586
1587        timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1588        while (!crng_ready()) {
1589                if (!timer_pending(&stack.timer))
1590                        mod_timer(&stack.timer, jiffies+1);
1591                mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1592                schedule();
1593                stack.now = random_get_entropy();
1594        }
1595
1596        del_timer_sync(&stack.timer);
1597        destroy_timer_on_stack(&stack.timer);
1598        mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1599}
1600
1601/*
1602 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1603 * cryptographically secure random numbers. This applies to: the /dev/urandom
1604 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1605 * family of functions. Using any of these functions without first calling
1606 * this function forfeits the guarantee of security.
1607 *
1608 * Returns: 0 if the urandom pool has been seeded.
1609 *          -ERESTARTSYS if the function was interrupted by a signal.
1610 */
1611int wait_for_random_bytes(void)
1612{
1613        if (likely(crng_ready()))
1614                return 0;
1615
1616        do {
1617                int ret;
1618                ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
1619                if (ret)
1620                        return ret > 0 ? 0 : ret;
1621
1622                try_to_generate_entropy();
1623        } while (!crng_ready());
1624
1625        return 0;
1626}
1627EXPORT_SYMBOL(wait_for_random_bytes);
1628
1629/*
1630 * Returns whether or not the urandom pool has been seeded and thus guaranteed
1631 * to supply cryptographically secure random numbers. This applies to: the
1632 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
1633 * ,u64,int,long} family of functions.
1634 *
1635 * Returns: true if the urandom pool has been seeded.
1636 *          false if the urandom pool has not been seeded.
1637 */
1638bool rng_is_initialized(void)
1639{
1640        return crng_ready();
1641}
1642EXPORT_SYMBOL(rng_is_initialized);
1643
1644/*
1645 * Add a callback function that will be invoked when the nonblocking
1646 * pool is initialised.
1647 *
1648 * returns: 0 if callback is successfully added
1649 *          -EALREADY if pool is already initialised (callback not called)
1650 *          -ENOENT if module for callback is not alive
1651 */
1652int add_random_ready_callback(struct random_ready_callback *rdy)
1653{
1654        struct module *owner;
1655        unsigned long flags;
1656        int err = -EALREADY;
1657
1658        if (crng_ready())
1659                return err;
1660
1661        owner = rdy->owner;
1662        if (!try_module_get(owner))
1663                return -ENOENT;
1664
1665        spin_lock_irqsave(&random_ready_list_lock, flags);
1666        if (crng_ready())
1667                goto out;
1668
1669        owner = NULL;
1670
1671        list_add(&rdy->list, &random_ready_list);
1672        err = 0;
1673
1674out:
1675        spin_unlock_irqrestore(&random_ready_list_lock, flags);
1676
1677        module_put(owner);
1678
1679        return err;
1680}
1681EXPORT_SYMBOL(add_random_ready_callback);
1682
1683/*
1684 * Delete a previously registered readiness callback function.
1685 */
1686void del_random_ready_callback(struct random_ready_callback *rdy)
1687{
1688        unsigned long flags;
1689        struct module *owner = NULL;
1690
1691        spin_lock_irqsave(&random_ready_list_lock, flags);
1692        if (!list_empty(&rdy->list)) {
1693                list_del_init(&rdy->list);
1694                owner = rdy->owner;
1695        }
1696        spin_unlock_irqrestore(&random_ready_list_lock, flags);
1697
1698        module_put(owner);
1699}
1700EXPORT_SYMBOL(del_random_ready_callback);
1701
1702/*
1703 * This function will use the architecture-specific hardware random
1704 * number generator if it is available.  The arch-specific hw RNG will
1705 * almost certainly be faster than what we can do in software, but it
1706 * is impossible to verify that it is implemented securely (as
1707 * opposed, to, say, the AES encryption of a sequence number using a
1708 * key known by the NSA).  So it's useful if we need the speed, but
1709 * only if we're willing to trust the hardware manufacturer not to
1710 * have put in a back door.
1711 *
1712 * Return number of bytes filled in.
1713 */
1714int __must_check get_random_bytes_arch(void *buf, int nbytes)
1715{
1716        int left = nbytes;
1717        char *p = buf;
1718
1719        trace_get_random_bytes_arch(left, _RET_IP_);
1720        while (left) {
1721                unsigned long v;
1722                int chunk = min_t(int, left, sizeof(unsigned long));
1723
1724                if (!arch_get_random_long(&v))
1725                        break;
1726
1727                memcpy(p, &v, chunk);
1728                p += chunk;
1729                left -= chunk;
1730        }
1731
1732        return nbytes - left;
1733}
1734EXPORT_SYMBOL(get_random_bytes_arch);
1735
1736/*
1737 * init_std_data - initialize pool with system data
1738 *
1739 * @r: pool to initialize
1740 *
1741 * This function clears the pool's entropy count and mixes some system
1742 * data into the pool to prepare it for use. The pool is not cleared
1743 * as that can only decrease the entropy in the pool.
1744 */
1745static void __init init_std_data(struct entropy_store *r)
1746{
1747        int i;
1748        ktime_t now = ktime_get_real();
1749        unsigned long rv;
1750
1751        mix_pool_bytes(r, &now, sizeof(now));
1752        for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1753                if (!arch_get_random_seed_long(&rv) &&
1754                    !arch_get_random_long(&rv))
1755                        rv = random_get_entropy();
1756                mix_pool_bytes(r, &rv, sizeof(rv));
1757        }
1758        mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1759}
1760
1761/*
1762 * Note that setup_arch() may call add_device_randomness()
1763 * long before we get here. This allows seeding of the pools
1764 * with some platform dependent data very early in the boot
1765 * process. But it limits our options here. We must use
1766 * statically allocated structures that already have all
1767 * initializations complete at compile time. We should also
1768 * take care not to overwrite the precious per platform data
1769 * we were given.
1770 */
1771int __init rand_initialize(void)
1772{
1773        init_std_data(&input_pool);
1774        crng_initialize(&primary_crng);
1775        crng_global_init_time = jiffies;
1776        if (ratelimit_disable) {
1777                urandom_warning.interval = 0;
1778                unseeded_warning.interval = 0;
1779        }
1780        return 0;
1781}
1782
1783#ifdef CONFIG_BLOCK
1784void rand_initialize_disk(struct gendisk *disk)
1785{
1786        struct timer_rand_state *state;
1787
1788        /*
1789         * If kzalloc returns null, we just won't use that entropy
1790         * source.
1791         */
1792        state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1793        if (state) {
1794                state->last_time = INITIAL_JIFFIES;
1795                disk->random = state;
1796        }
1797}
1798#endif
1799
1800static ssize_t
1801urandom_read_nowarn(struct file *file, char __user *buf, size_t nbytes,
1802                    loff_t *ppos)
1803{
1804        int ret;
1805
1806        nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1807        ret = extract_crng_user(buf, nbytes);
1808        trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1809        return ret;
1810}
1811
1812static ssize_t
1813urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1814{
1815        unsigned long flags;
1816        static int maxwarn = 10;
1817
1818        if (!crng_ready() && maxwarn > 0) {
1819                maxwarn--;
1820                if (__ratelimit(&urandom_warning))
1821                        pr_notice("%s: uninitialized urandom read (%zd bytes read)\n",
1822                                  current->comm, nbytes);
1823                spin_lock_irqsave(&primary_crng.lock, flags);
1824                crng_init_cnt = 0;
1825                spin_unlock_irqrestore(&primary_crng.lock, flags);
1826        }
1827
1828        return urandom_read_nowarn(file, buf, nbytes, ppos);
1829}
1830
1831static ssize_t
1832random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1833{
1834        int ret;
1835
1836        ret = wait_for_random_bytes();
1837        if (ret != 0)
1838                return ret;
1839        return urandom_read_nowarn(file, buf, nbytes, ppos);
1840}
1841
1842static __poll_t
1843random_poll(struct file *file, poll_table * wait)
1844{
1845        __poll_t mask;
1846
1847        poll_wait(file, &crng_init_wait, wait);
1848        poll_wait(file, &random_write_wait, wait);
1849        mask = 0;
1850        if (crng_ready())
1851                mask |= EPOLLIN | EPOLLRDNORM;
1852        if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1853                mask |= EPOLLOUT | EPOLLWRNORM;
1854        return mask;
1855}
1856
1857static int
1858write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1859{
1860        size_t bytes;
1861        __u32 t, buf[16];
1862        const char __user *p = buffer;
1863
1864        while (count > 0) {
1865                int b, i = 0;
1866
1867                bytes = min(count, sizeof(buf));
1868                if (copy_from_user(&buf, p, bytes))
1869                        return -EFAULT;
1870
1871                for (b = bytes ; b > 0 ; b -= sizeof(__u32), i++) {
1872                        if (!arch_get_random_int(&t))
1873                                break;
1874                        buf[i] ^= t;
1875                }
1876
1877                count -= bytes;
1878                p += bytes;
1879
1880                mix_pool_bytes(r, buf, bytes);
1881                cond_resched();
1882        }
1883
1884        return 0;
1885}
1886
1887static ssize_t random_write(struct file *file, const char __user *buffer,
1888                            size_t count, loff_t *ppos)
1889{
1890        size_t ret;
1891
1892        ret = write_pool(&input_pool, buffer, count);
1893        if (ret)
1894                return ret;
1895
1896        return (ssize_t)count;
1897}
1898
1899static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1900{
1901        int size, ent_count;
1902        int __user *p = (int __user *)arg;
1903        int retval;
1904
1905        switch (cmd) {
1906        case RNDGETENTCNT:
1907                /* inherently racy, no point locking */
1908                ent_count = ENTROPY_BITS(&input_pool);
1909                if (put_user(ent_count, p))
1910                        return -EFAULT;
1911                return 0;
1912        case RNDADDTOENTCNT:
1913                if (!capable(CAP_SYS_ADMIN))
1914                        return -EPERM;
1915                if (get_user(ent_count, p))
1916                        return -EFAULT;
1917                return credit_entropy_bits_safe(&input_pool, ent_count);
1918        case RNDADDENTROPY:
1919                if (!capable(CAP_SYS_ADMIN))
1920                        return -EPERM;
1921                if (get_user(ent_count, p++))
1922                        return -EFAULT;
1923                if (ent_count < 0)
1924                        return -EINVAL;
1925                if (get_user(size, p++))
1926                        return -EFAULT;
1927                retval = write_pool(&input_pool, (const char __user *)p,
1928                                    size);
1929                if (retval < 0)
1930                        return retval;
1931                return credit_entropy_bits_safe(&input_pool, ent_count);
1932        case RNDZAPENTCNT:
1933        case RNDCLEARPOOL:
1934                /*
1935                 * Clear the entropy pool counters. We no longer clear
1936                 * the entropy pool, as that's silly.
1937                 */
1938                if (!capable(CAP_SYS_ADMIN))
1939                        return -EPERM;
1940                input_pool.entropy_count = 0;
1941                return 0;
1942        case RNDRESEEDCRNG:
1943                if (!capable(CAP_SYS_ADMIN))
1944                        return -EPERM;
1945                if (crng_init < 2)
1946                        return -ENODATA;
1947                crng_reseed(&primary_crng, NULL);
1948                crng_global_init_time = jiffies - 1;
1949                return 0;
1950        default:
1951                return -EINVAL;
1952        }
1953}
1954
1955static int random_fasync(int fd, struct file *filp, int on)
1956{
1957        return fasync_helper(fd, filp, on, &fasync);
1958}
1959
1960const struct file_operations random_fops = {
1961        .read  = random_read,
1962        .write = random_write,
1963        .poll  = random_poll,
1964        .unlocked_ioctl = random_ioctl,
1965        .compat_ioctl = compat_ptr_ioctl,
1966        .fasync = random_fasync,
1967        .llseek = noop_llseek,
1968};
1969
1970const struct file_operations urandom_fops = {
1971        .read  = urandom_read,
1972        .write = random_write,
1973        .unlocked_ioctl = random_ioctl,
1974        .compat_ioctl = compat_ptr_ioctl,
1975        .fasync = random_fasync,
1976        .llseek = noop_llseek,
1977};
1978
1979SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1980                unsigned int, flags)
1981{
1982        int ret;
1983
1984        if (flags & ~(GRND_NONBLOCK|GRND_RANDOM|GRND_INSECURE))
1985                return -EINVAL;
1986
1987        /*
1988         * Requesting insecure and blocking randomness at the same time makes
1989         * no sense.
1990         */
1991        if ((flags & (GRND_INSECURE|GRND_RANDOM)) == (GRND_INSECURE|GRND_RANDOM))
1992                return -EINVAL;
1993
1994        if (count > INT_MAX)
1995                count = INT_MAX;
1996
1997        if (!(flags & GRND_INSECURE) && !crng_ready()) {
1998                if (flags & GRND_NONBLOCK)
1999                        return -EAGAIN;
2000                ret = wait_for_random_bytes();
2001                if (unlikely(ret))
2002                        return ret;
2003        }
2004        return urandom_read_nowarn(NULL, buf, count, NULL);
2005}
2006
2007/********************************************************************
2008 *
2009 * Sysctl interface
2010 *
2011 ********************************************************************/
2012
2013#ifdef CONFIG_SYSCTL
2014
2015#include <linux/sysctl.h>
2016
2017static int min_write_thresh;
2018static int max_write_thresh = INPUT_POOL_WORDS * 32;
2019static int random_min_urandom_seed = 60;
2020static char sysctl_bootid[16];
2021
2022/*
2023 * This function is used to return both the bootid UUID, and random
2024 * UUID.  The difference is in whether table->data is NULL; if it is,
2025 * then a new UUID is generated and returned to the user.
2026 *
2027 * If the user accesses this via the proc interface, the UUID will be
2028 * returned as an ASCII string in the standard UUID format; if via the
2029 * sysctl system call, as 16 bytes of binary data.
2030 */
2031static int proc_do_uuid(struct ctl_table *table, int write,
2032                        void __user *buffer, size_t *lenp, loff_t *ppos)
2033{
2034        struct ctl_table fake_table;
2035        unsigned char buf[64], tmp_uuid[16], *uuid;
2036
2037        uuid = table->data;
2038        if (!uuid) {
2039                uuid = tmp_uuid;
2040                generate_random_uuid(uuid);
2041        } else {
2042                static DEFINE_SPINLOCK(bootid_spinlock);
2043
2044                spin_lock(&bootid_spinlock);
2045                if (!uuid[8])
2046                        generate_random_uuid(uuid);
2047                spin_unlock(&bootid_spinlock);
2048        }
2049
2050        sprintf(buf, "%pU", uuid);
2051
2052        fake_table.data = buf;
2053        fake_table.maxlen = sizeof(buf);
2054
2055        return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2056}
2057
2058/*
2059 * Return entropy available scaled to integral bits
2060 */
2061static int proc_do_entropy(struct ctl_table *table, int write,
2062                           void __user *buffer, size_t *lenp, loff_t *ppos)
2063{
2064        struct ctl_table fake_table;
2065        int entropy_count;
2066
2067        entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2068
2069        fake_table.data = &entropy_count;
2070        fake_table.maxlen = sizeof(entropy_count);
2071
2072        return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2073}
2074
2075static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2076extern struct ctl_table random_table[];
2077struct ctl_table random_table[] = {
2078        {
2079                .procname       = "poolsize",
2080                .data           = &sysctl_poolsize,
2081                .maxlen         = sizeof(int),
2082                .mode           = 0444,
2083                .proc_handler   = proc_dointvec,
2084        },
2085        {
2086                .procname       = "entropy_avail",
2087                .maxlen         = sizeof(int),
2088                .mode           = 0444,
2089                .proc_handler   = proc_do_entropy,
2090                .data           = &input_pool.entropy_count,
2091        },
2092        {
2093                .procname       = "write_wakeup_threshold",
2094                .data           = &random_write_wakeup_bits,
2095                .maxlen         = sizeof(int),
2096                .mode           = 0644,
2097                .proc_handler   = proc_dointvec_minmax,
2098                .extra1         = &min_write_thresh,
2099                .extra2         = &max_write_thresh,
2100        },
2101        {
2102                .procname       = "urandom_min_reseed_secs",
2103                .data           = &random_min_urandom_seed,
2104                .maxlen         = sizeof(int),
2105                .mode           = 0644,
2106                .proc_handler   = proc_dointvec,
2107        },
2108        {
2109                .procname       = "boot_id",
2110                .data           = &sysctl_bootid,
2111                .maxlen         = 16,
2112                .mode           = 0444,
2113                .proc_handler   = proc_do_uuid,
2114        },
2115        {
2116                .procname       = "uuid",
2117                .maxlen         = 16,
2118                .mode           = 0444,
2119                .proc_handler   = proc_do_uuid,
2120        },
2121#ifdef ADD_INTERRUPT_BENCH
2122        {
2123                .procname       = "add_interrupt_avg_cycles",
2124                .data           = &avg_cycles,
2125                .maxlen         = sizeof(avg_cycles),
2126                .mode           = 0444,
2127                .proc_handler   = proc_doulongvec_minmax,
2128        },
2129        {
2130                .procname       = "add_interrupt_avg_deviation",
2131                .data           = &avg_deviation,
2132                .maxlen         = sizeof(avg_deviation),
2133                .mode           = 0444,
2134                .proc_handler   = proc_doulongvec_minmax,
2135        },
2136#endif
2137        { }
2138};
2139#endif  /* CONFIG_SYSCTL */
2140
2141struct batched_entropy {
2142        union {
2143                u64 entropy_u64[CHACHA_BLOCK_SIZE / sizeof(u64)];
2144                u32 entropy_u32[CHACHA_BLOCK_SIZE / sizeof(u32)];
2145        };
2146        unsigned int position;
2147        spinlock_t batch_lock;
2148};
2149
2150/*
2151 * Get a random word for internal kernel use only. The quality of the random
2152 * number is either as good as RDRAND or as good as /dev/urandom, with the
2153 * goal of being quite fast and not depleting entropy. In order to ensure
2154 * that the randomness provided by this function is okay, the function
2155 * wait_for_random_bytes() should be called and return 0 at least once
2156 * at any point prior.
2157 */
2158static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
2159        .batch_lock     = __SPIN_LOCK_UNLOCKED(batched_entropy_u64.lock),
2160};
2161
2162u64 get_random_u64(void)
2163{
2164        u64 ret;
2165        unsigned long flags;
2166        struct batched_entropy *batch;
2167        static void *previous;
2168
2169#if BITS_PER_LONG == 64
2170        if (arch_get_random_long((unsigned long *)&ret))
2171                return ret;
2172#else
2173        if (arch_get_random_long((unsigned long *)&ret) &&
2174            arch_get_random_long((unsigned long *)&ret + 1))
2175            return ret;
2176#endif
2177
2178        warn_unseeded_randomness(&previous);
2179
2180        batch = raw_cpu_ptr(&batched_entropy_u64);
2181        spin_lock_irqsave(&batch->batch_lock, flags);
2182        if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2183                extract_crng((u8 *)batch->entropy_u64);
2184                batch->position = 0;
2185        }
2186        ret = batch->entropy_u64[batch->position++];
2187        spin_unlock_irqrestore(&batch->batch_lock, flags);
2188        return ret;
2189}
2190EXPORT_SYMBOL(get_random_u64);
2191
2192static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
2193        .batch_lock     = __SPIN_LOCK_UNLOCKED(batched_entropy_u32.lock),
2194};
2195u32 get_random_u32(void)
2196{
2197        u32 ret;
2198        unsigned long flags;
2199        struct batched_entropy *batch;
2200        static void *previous;
2201
2202        if (arch_get_random_int(&ret))
2203                return ret;
2204
2205        warn_unseeded_randomness(&previous);
2206
2207        batch = raw_cpu_ptr(&batched_entropy_u32);
2208        spin_lock_irqsave(&batch->batch_lock, flags);
2209        if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2210                extract_crng((u8 *)batch->entropy_u32);
2211                batch->position = 0;
2212        }
2213        ret = batch->entropy_u32[batch->position++];
2214        spin_unlock_irqrestore(&batch->batch_lock, flags);
2215        return ret;
2216}
2217EXPORT_SYMBOL(get_random_u32);
2218
2219/* It's important to invalidate all potential batched entropy that might
2220 * be stored before the crng is initialized, which we can do lazily by
2221 * simply resetting the counter to zero so that it's re-extracted on the
2222 * next usage. */
2223static void invalidate_batched_entropy(void)
2224{
2225        int cpu;
2226        unsigned long flags;
2227
2228        for_each_possible_cpu (cpu) {
2229                struct batched_entropy *batched_entropy;
2230
2231                batched_entropy = per_cpu_ptr(&batched_entropy_u32, cpu);
2232                spin_lock_irqsave(&batched_entropy->batch_lock, flags);
2233                batched_entropy->position = 0;
2234                spin_unlock(&batched_entropy->batch_lock);
2235
2236                batched_entropy = per_cpu_ptr(&batched_entropy_u64, cpu);
2237                spin_lock(&batched_entropy->batch_lock);
2238                batched_entropy->position = 0;
2239                spin_unlock_irqrestore(&batched_entropy->batch_lock, flags);
2240        }
2241}
2242
2243/**
2244 * randomize_page - Generate a random, page aligned address
2245 * @start:      The smallest acceptable address the caller will take.
2246 * @range:      The size of the area, starting at @start, within which the
2247 *              random address must fall.
2248 *
2249 * If @start + @range would overflow, @range is capped.
2250 *
2251 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2252 * @start was already page aligned.  We now align it regardless.
2253 *
2254 * Return: A page aligned address within [start, start + range).  On error,
2255 * @start is returned.
2256 */
2257unsigned long
2258randomize_page(unsigned long start, unsigned long range)
2259{
2260        if (!PAGE_ALIGNED(start)) {
2261                range -= PAGE_ALIGN(start) - start;
2262                start = PAGE_ALIGN(start);
2263        }
2264
2265        if (start > ULONG_MAX - range)
2266                range = ULONG_MAX - start;
2267
2268        range >>= PAGE_SHIFT;
2269
2270        if (range == 0)
2271                return start;
2272
2273        return start + (get_random_long() % range << PAGE_SHIFT);
2274}
2275
2276/* Interface for in-kernel drivers of true hardware RNGs.
2277 * Those devices may produce endless random bits and will be throttled
2278 * when our pool is full.
2279 */
2280void add_hwgenerator_randomness(const char *buffer, size_t count,
2281                                size_t entropy)
2282{
2283        struct entropy_store *poolp = &input_pool;
2284
2285        if (unlikely(crng_init == 0)) {
2286                crng_fast_load(buffer, count);
2287                return;
2288        }
2289
2290        /* Suspend writing if we're above the trickle threshold.
2291         * We'll be woken up again once below random_write_wakeup_thresh,
2292         * or when the calling thread is about to terminate.
2293         */
2294        wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2295                        ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2296        mix_pool_bytes(poolp, buffer, count);
2297        credit_entropy_bits(poolp, entropy);
2298}
2299EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2300
2301/* Handle random seed passed by bootloader.
2302 * If the seed is trustworthy, it would be regarded as hardware RNGs. Otherwise
2303 * it would be regarded as device data.
2304 * The decision is controlled by CONFIG_RANDOM_TRUST_BOOTLOADER.
2305 */
2306void add_bootloader_randomness(const void *buf, unsigned int size)
2307{
2308        if (IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER))
2309                add_hwgenerator_randomness(buf, size, size * 8);
2310        else
2311                add_device_randomness(buf, size);
2312}
2313EXPORT_SYMBOL_GPL(add_bootloader_randomness);
2314