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#include <linux/utsname.h>
 311#include <linux/module.h>
 312#include <linux/kernel.h>
 313#include <linux/major.h>
 314#include <linux/string.h>
 315#include <linux/fcntl.h>
 316#include <linux/slab.h>
 317#include <linux/random.h>
 318#include <linux/poll.h>
 319#include <linux/init.h>
 320#include <linux/fs.h>
 321#include <linux/genhd.h>
 322#include <linux/interrupt.h>
 323#include <linux/mm.h>
 324#include <linux/nodemask.h>
 325#include <linux/spinlock.h>
 326#include <linux/kthread.h>
 327#include <linux/percpu.h>
 328#include <linux/cryptohash.h>
 329#include <linux/fips.h>
 330#include <linux/ptrace.h>
 331#include <linux/workqueue.h>
 332#include <linux/irq.h>
 333#include <linux/ratelimit.h>
 334#include <linux/syscalls.h>
 335#include <linux/completion.h>
 336#include <linux/uuid.h>
 337#include <crypto/chacha.h>
 338
 339#include <asm/processor.h>
 340#include <linux/uaccess.h>
 341#include <asm/irq.h>
 342#include <asm/irq_regs.h>
 343#include <asm/io.h>
 344
 345#define CREATE_TRACE_POINTS
 346#include <trace/events/random.h>
 347
 348/* #define ADD_INTERRUPT_BENCH */
 349
 350/*
 351 * Configuration information
 352 */
 353#define INPUT_POOL_SHIFT        12
 354#define INPUT_POOL_WORDS        (1 << (INPUT_POOL_SHIFT-5))
 355#define OUTPUT_POOL_SHIFT       10
 356#define OUTPUT_POOL_WORDS       (1 << (OUTPUT_POOL_SHIFT-5))
 357#define SEC_XFER_SIZE           512
 358#define EXTRACT_SIZE            10
 359
 360
 361#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
 362
 363/*
 364 * To allow fractional bits to be tracked, the entropy_count field is
 365 * denominated in units of 1/8th bits.
 366 *
 367 * 2*(ENTROPY_SHIFT + poolbitshift) must <= 31, or the multiply in
 368 * credit_entropy_bits() needs to be 64 bits wide.
 369 */
 370#define ENTROPY_SHIFT 3
 371#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
 372
 373/*
 374 * The minimum number of bits of entropy before we wake up a read on
 375 * /dev/random.  Should be enough to do a significant reseed.
 376 */
 377static int random_read_wakeup_bits = 64;
 378
 379/*
 380 * If the entropy count falls under this number of bits, then we
 381 * should wake up processes which are selecting or polling on write
 382 * access to /dev/random.
 383 */
 384static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
 385
 386/*
 387 * Originally, we used a primitive polynomial of degree .poolwords
 388 * over GF(2).  The taps for various sizes are defined below.  They
 389 * were chosen to be evenly spaced except for the last tap, which is 1
 390 * to get the twisting happening as fast as possible.
 391 *
 392 * For the purposes of better mixing, we use the CRC-32 polynomial as
 393 * well to make a (modified) twisted Generalized Feedback Shift
 394 * Register.  (See M. Matsumoto & Y. Kurita, 1992.  Twisted GFSR
 395 * generators.  ACM Transactions on Modeling and Computer Simulation
 396 * 2(3):179-194.  Also see M. Matsumoto & Y. Kurita, 1994.  Twisted
 397 * GFSR generators II.  ACM Transactions on Modeling and Computer
 398 * Simulation 4:254-266)
 399 *
 400 * Thanks to Colin Plumb for suggesting this.
 401 *
 402 * The mixing operation is much less sensitive than the output hash,
 403 * where we use SHA-1.  All that we want of mixing operation is that
 404 * it be a good non-cryptographic hash; i.e. it not produce collisions
 405 * when fed "random" data of the sort we expect to see.  As long as
 406 * the pool state differs for different inputs, we have preserved the
 407 * input entropy and done a good job.  The fact that an intelligent
 408 * attacker can construct inputs that will produce controlled
 409 * alterations to the pool's state is not important because we don't
 410 * consider such inputs to contribute any randomness.  The only
 411 * property we need with respect to them is that the attacker can't
 412 * increase his/her knowledge of the pool's state.  Since all
 413 * additions are reversible (knowing the final state and the input,
 414 * you can reconstruct the initial state), if an attacker has any
 415 * uncertainty about the initial state, he/she can only shuffle that
 416 * uncertainty about, but never cause any collisions (which would
 417 * decrease the uncertainty).
 418 *
 419 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
 420 * Videau in their paper, "The Linux Pseudorandom Number Generator
 421 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf).  In their
 422 * paper, they point out that we are not using a true Twisted GFSR,
 423 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
 424 * is, with only three taps, instead of the six that we are using).
 425 * As a result, the resulting polynomial is neither primitive nor
 426 * irreducible, and hence does not have a maximal period over
 427 * GF(2**32).  They suggest a slight change to the generator
 428 * polynomial which improves the resulting TGFSR polynomial to be
 429 * irreducible, which we have made here.
 430 */
 431static const struct poolinfo {
 432        int poolbitshift, poolwords, poolbytes, poolfracbits;
 433#define S(x) ilog2(x)+5, (x), (x)*4, (x) << (ENTROPY_SHIFT+5)
 434        int tap1, tap2, tap3, tap4, tap5;
 435} poolinfo_table[] = {
 436        /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
 437        /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
 438        { S(128),       104,    76,     51,     25,     1 },
 439        /* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
 440        /* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
 441        { S(32),        26,     19,     14,     7,      1 },
 442#if 0
 443        /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1  -- 115 */
 444        { S(2048),      1638,   1231,   819,    411,    1 },
 445
 446        /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
 447        { S(1024),      817,    615,    412,    204,    1 },
 448
 449        /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
 450        { S(1024),      819,    616,    410,    207,    2 },
 451
 452        /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
 453        { S(512),       411,    308,    208,    104,    1 },
 454
 455        /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
 456        { S(512),       409,    307,    206,    102,    2 },
 457        /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
 458        { S(512),       409,    309,    205,    103,    2 },
 459
 460        /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
 461        { S(256),       205,    155,    101,    52,     1 },
 462
 463        /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
 464        { S(128),       103,    78,     51,     27,     2 },
 465
 466        /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
 467        { S(64),        52,     39,     26,     14,     1 },
 468#endif
 469};
 470
 471/*
 472 * Static global variables
 473 */
 474static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
 475static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
 476static struct fasync_struct *fasync;
 477
 478static DEFINE_SPINLOCK(random_ready_list_lock);
 479static LIST_HEAD(random_ready_list);
 480
 481struct crng_state {
 482        __u32           state[16];
 483        unsigned long   init_time;
 484        spinlock_t      lock;
 485};
 486
 487static struct crng_state primary_crng = {
 488        .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
 489};
 490
 491/*
 492 * crng_init =  0 --> Uninitialized
 493 *              1 --> Initialized
 494 *              2 --> Initialized from input_pool
 495 *
 496 * crng_init is protected by primary_crng->lock, and only increases
 497 * its value (from 0->1->2).
 498 */
 499static int crng_init = 0;
 500#define crng_ready() (likely(crng_init > 1))
 501static int crng_init_cnt = 0;
 502static unsigned long crng_global_init_time = 0;
 503#define CRNG_INIT_CNT_THRESH (2*CHACHA_KEY_SIZE)
 504static void _extract_crng(struct crng_state *crng, __u8 out[CHACHA_BLOCK_SIZE]);
 505static void _crng_backtrack_protect(struct crng_state *crng,
 506                                    __u8 tmp[CHACHA_BLOCK_SIZE], int used);
 507static void process_random_ready_list(void);
 508static void _get_random_bytes(void *buf, int nbytes);
 509
 510static struct ratelimit_state unseeded_warning =
 511        RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
 512static struct ratelimit_state urandom_warning =
 513        RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
 514
 515static int ratelimit_disable __read_mostly;
 516
 517module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
 518MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
 519
 520/**********************************************************************
 521 *
 522 * OS independent entropy store.   Here are the functions which handle
 523 * storing entropy in an entropy pool.
 524 *
 525 **********************************************************************/
 526
 527struct entropy_store;
 528struct entropy_store {
 529        /* read-only data: */
 530        const struct poolinfo *poolinfo;
 531        __u32 *pool;
 532        const char *name;
 533        struct entropy_store *pull;
 534        struct work_struct push_work;
 535
 536        /* read-write data: */
 537        unsigned long last_pulled;
 538        spinlock_t lock;
 539        unsigned short add_ptr;
 540        unsigned short input_rotate;
 541        int entropy_count;
 542        unsigned int initialized:1;
 543        unsigned int last_data_init:1;
 544        __u8 last_data[EXTRACT_SIZE];
 545};
 546
 547static ssize_t extract_entropy(struct entropy_store *r, void *buf,
 548                               size_t nbytes, int min, int rsvd);
 549static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
 550                                size_t nbytes, int fips);
 551
 552static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
 553static void push_to_pool(struct work_struct *work);
 554static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
 555static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
 556
 557static struct entropy_store input_pool = {
 558        .poolinfo = &poolinfo_table[0],
 559        .name = "input",
 560        .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
 561        .pool = input_pool_data
 562};
 563
 564static struct entropy_store blocking_pool = {
 565        .poolinfo = &poolinfo_table[1],
 566        .name = "blocking",
 567        .pull = &input_pool,
 568        .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
 569        .pool = blocking_pool_data,
 570        .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
 571                                        push_to_pool),
 572};
 573
 574static __u32 const twist_table[8] = {
 575        0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
 576        0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
 577
 578/*
 579 * This function adds bytes into the entropy "pool".  It does not
 580 * update the entropy estimate.  The caller should call
 581 * credit_entropy_bits if this is appropriate.
 582 *
 583 * The pool is stirred with a primitive polynomial of the appropriate
 584 * degree, and then twisted.  We twist by three bits at a time because
 585 * it's cheap to do so and helps slightly in the expected case where
 586 * the entropy is concentrated in the low-order bits.
 587 */
 588static void _mix_pool_bytes(struct entropy_store *r, const void *in,
 589                            int nbytes)
 590{
 591        unsigned long i, tap1, tap2, tap3, tap4, tap5;
 592        int input_rotate;
 593        int wordmask = r->poolinfo->poolwords - 1;
 594        const char *bytes = in;
 595        __u32 w;
 596
 597        tap1 = r->poolinfo->tap1;
 598        tap2 = r->poolinfo->tap2;
 599        tap3 = r->poolinfo->tap3;
 600        tap4 = r->poolinfo->tap4;
 601        tap5 = r->poolinfo->tap5;
 602
 603        input_rotate = r->input_rotate;
 604        i = r->add_ptr;
 605
 606        /* mix one byte at a time to simplify size handling and churn faster */
 607        while (nbytes--) {
 608                w = rol32(*bytes++, input_rotate);
 609                i = (i - 1) & wordmask;
 610
 611                /* XOR in the various taps */
 612                w ^= r->pool[i];
 613                w ^= r->pool[(i + tap1) & wordmask];
 614                w ^= r->pool[(i + tap2) & wordmask];
 615                w ^= r->pool[(i + tap3) & wordmask];
 616                w ^= r->pool[(i + tap4) & wordmask];
 617                w ^= r->pool[(i + tap5) & wordmask];
 618
 619                /* Mix the result back in with a twist */
 620                r->pool[i] = (w >> 3) ^ twist_table[w & 7];
 621
 622                /*
 623                 * Normally, we add 7 bits of rotation to the pool.
 624                 * At the beginning of the pool, add an extra 7 bits
 625                 * rotation, so that successive passes spread the
 626                 * input bits across the pool evenly.
 627                 */
 628                input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
 629        }
 630
 631        r->input_rotate = input_rotate;
 632        r->add_ptr = i;
 633}
 634
 635static void __mix_pool_bytes(struct entropy_store *r, const void *in,
 636                             int nbytes)
 637{
 638        trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
 639        _mix_pool_bytes(r, in, nbytes);
 640}
 641
 642static void mix_pool_bytes(struct entropy_store *r, const void *in,
 643                           int nbytes)
 644{
 645        unsigned long flags;
 646
 647        trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
 648        spin_lock_irqsave(&r->lock, flags);
 649        _mix_pool_bytes(r, in, nbytes);
 650        spin_unlock_irqrestore(&r->lock, flags);
 651}
 652
 653struct fast_pool {
 654        __u32           pool[4];
 655        unsigned long   last;
 656        unsigned short  reg_idx;
 657        unsigned char   count;
 658};
 659
 660/*
 661 * This is a fast mixing routine used by the interrupt randomness
 662 * collector.  It's hardcoded for an 128 bit pool and assumes that any
 663 * locks that might be needed are taken by the caller.
 664 */
 665static void fast_mix(struct fast_pool *f)
 666{
 667        __u32 a = f->pool[0],   b = f->pool[1];
 668        __u32 c = f->pool[2],   d = f->pool[3];
 669
 670        a += b;                 c += d;
 671        b = rol32(b, 6);        d = rol32(d, 27);
 672        d ^= a;                 b ^= c;
 673
 674        a += b;                 c += d;
 675        b = rol32(b, 16);       d = rol32(d, 14);
 676        d ^= a;                 b ^= c;
 677
 678        a += b;                 c += d;
 679        b = rol32(b, 6);        d = rol32(d, 27);
 680        d ^= a;                 b ^= c;
 681
 682        a += b;                 c += d;
 683        b = rol32(b, 16);       d = rol32(d, 14);
 684        d ^= a;                 b ^= c;
 685
 686        f->pool[0] = a;  f->pool[1] = b;
 687        f->pool[2] = c;  f->pool[3] = d;
 688        f->count++;
 689}
 690
 691static void process_random_ready_list(void)
 692{
 693        unsigned long flags;
 694        struct random_ready_callback *rdy, *tmp;
 695
 696        spin_lock_irqsave(&random_ready_list_lock, flags);
 697        list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
 698                struct module *owner = rdy->owner;
 699
 700                list_del_init(&rdy->list);
 701                rdy->func(rdy);
 702                module_put(owner);
 703        }
 704        spin_unlock_irqrestore(&random_ready_list_lock, flags);
 705}
 706
 707/*
 708 * Credit (or debit) the entropy store with n bits of entropy.
 709 * Use credit_entropy_bits_safe() if the value comes from userspace
 710 * or otherwise should be checked for extreme values.
 711 */
 712static void credit_entropy_bits(struct entropy_store *r, int nbits)
 713{
 714        int entropy_count, orig, has_initialized = 0;
 715        const int pool_size = r->poolinfo->poolfracbits;
 716        int nfrac = nbits << ENTROPY_SHIFT;
 717
 718        if (!nbits)
 719                return;
 720
 721retry:
 722        entropy_count = orig = READ_ONCE(r->entropy_count);
 723        if (nfrac < 0) {
 724                /* Debit */
 725                entropy_count += nfrac;
 726        } else {
 727                /*
 728                 * Credit: we have to account for the possibility of
 729                 * overwriting already present entropy.  Even in the
 730                 * ideal case of pure Shannon entropy, new contributions
 731                 * approach the full value asymptotically:
 732                 *
 733                 * entropy <- entropy + (pool_size - entropy) *
 734                 *      (1 - exp(-add_entropy/pool_size))
 735                 *
 736                 * For add_entropy <= pool_size/2 then
 737                 * (1 - exp(-add_entropy/pool_size)) >=
 738                 *    (add_entropy/pool_size)*0.7869...
 739                 * so we can approximate the exponential with
 740                 * 3/4*add_entropy/pool_size and still be on the
 741                 * safe side by adding at most pool_size/2 at a time.
 742                 *
 743                 * The use of pool_size-2 in the while statement is to
 744                 * prevent rounding artifacts from making the loop
 745                 * arbitrarily long; this limits the loop to log2(pool_size)*2
 746                 * turns no matter how large nbits is.
 747                 */
 748                int pnfrac = nfrac;
 749                const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
 750                /* The +2 corresponds to the /4 in the denominator */
 751
 752                do {
 753                        unsigned int anfrac = min(pnfrac, pool_size/2);
 754                        unsigned int add =
 755                                ((pool_size - entropy_count)*anfrac*3) >> s;
 756
 757                        entropy_count += add;
 758                        pnfrac -= anfrac;
 759                } while (unlikely(entropy_count < pool_size-2 && pnfrac));
 760        }
 761
 762        if (unlikely(entropy_count < 0)) {
 763                pr_warn("random: negative entropy/overflow: pool %s count %d\n",
 764                        r->name, entropy_count);
 765                WARN_ON(1);
 766                entropy_count = 0;
 767        } else if (entropy_count > pool_size)
 768                entropy_count = pool_size;
 769        if ((r == &blocking_pool) && !r->initialized &&
 770            (entropy_count >> ENTROPY_SHIFT) > 128)
 771                has_initialized = 1;
 772        if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
 773                goto retry;
 774
 775        if (has_initialized) {
 776                r->initialized = 1;
 777                wake_up_interruptible(&random_read_wait);
 778                kill_fasync(&fasync, SIGIO, POLL_IN);
 779        }
 780
 781        trace_credit_entropy_bits(r->name, nbits,
 782                                  entropy_count >> ENTROPY_SHIFT, _RET_IP_);
 783
 784        if (r == &input_pool) {
 785                int entropy_bits = entropy_count >> ENTROPY_SHIFT;
 786                struct entropy_store *other = &blocking_pool;
 787
 788                if (crng_init < 2) {
 789                        if (entropy_bits < 128)
 790                                return;
 791                        crng_reseed(&primary_crng, r);
 792                        entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
 793                }
 794
 795                /* initialize the blocking pool if necessary */
 796                if (entropy_bits >= random_read_wakeup_bits &&
 797                    !other->initialized) {
 798                        schedule_work(&other->push_work);
 799                        return;
 800                }
 801
 802                /* should we wake readers? */
 803                if (entropy_bits >= random_read_wakeup_bits &&
 804                    wq_has_sleeper(&random_read_wait)) {
 805                        wake_up_interruptible(&random_read_wait);
 806                        kill_fasync(&fasync, SIGIO, POLL_IN);
 807                }
 808                /* If the input pool is getting full, and the blocking
 809                 * pool has room, send some entropy to the blocking
 810                 * pool.
 811                 */
 812                if (!work_pending(&other->push_work) &&
 813                    (ENTROPY_BITS(r) > 6 * r->poolinfo->poolbytes) &&
 814                    (ENTROPY_BITS(other) <= 6 * other->poolinfo->poolbytes))
 815                        schedule_work(&other->push_work);
 816        }
 817}
 818
 819static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
 820{
 821        const int nbits_max = r->poolinfo->poolwords * 32;
 822
 823        if (nbits < 0)
 824                return -EINVAL;
 825
 826        /* Cap the value to avoid overflows */
 827        nbits = min(nbits,  nbits_max);
 828
 829        credit_entropy_bits(r, nbits);
 830        return 0;
 831}
 832
 833/*********************************************************************
 834 *
 835 * CRNG using CHACHA20
 836 *
 837 *********************************************************************/
 838
 839#define CRNG_RESEED_INTERVAL (300*HZ)
 840
 841static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
 842
 843#ifdef CONFIG_NUMA
 844/*
 845 * Hack to deal with crazy userspace progams when they are all trying
 846 * to access /dev/urandom in parallel.  The programs are almost
 847 * certainly doing something terribly wrong, but we'll work around
 848 * their brain damage.
 849 */
 850static struct crng_state **crng_node_pool __read_mostly;
 851#endif
 852
 853static void invalidate_batched_entropy(void);
 854static void numa_crng_init(void);
 855
 856static bool trust_cpu __ro_after_init = IS_ENABLED(CONFIG_RANDOM_TRUST_CPU);
 857static int __init parse_trust_cpu(char *arg)
 858{
 859        return kstrtobool(arg, &trust_cpu);
 860}
 861early_param("random.trust_cpu", parse_trust_cpu);
 862
 863static void crng_initialize(struct crng_state *crng)
 864{
 865        int             i;
 866        int             arch_init = 1;
 867        unsigned long   rv;
 868
 869        memcpy(&crng->state[0], "expand 32-byte k", 16);
 870        if (crng == &primary_crng)
 871                _extract_entropy(&input_pool, &crng->state[4],
 872                                 sizeof(__u32) * 12, 0);
 873        else
 874                _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
 875        for (i = 4; i < 16; i++) {
 876                if (!arch_get_random_seed_long(&rv) &&
 877                    !arch_get_random_long(&rv)) {
 878                        rv = random_get_entropy();
 879                        arch_init = 0;
 880                }
 881                crng->state[i] ^= rv;
 882        }
 883        if (trust_cpu && arch_init && crng == &primary_crng) {
 884                invalidate_batched_entropy();
 885                numa_crng_init();
 886                crng_init = 2;
 887                pr_notice("random: crng done (trusting CPU's manufacturer)\n");
 888        }
 889        crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
 890}
 891
 892#ifdef CONFIG_NUMA
 893static void do_numa_crng_init(struct work_struct *work)
 894{
 895        int i;
 896        struct crng_state *crng;
 897        struct crng_state **pool;
 898
 899        pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
 900        for_each_online_node(i) {
 901                crng = kmalloc_node(sizeof(struct crng_state),
 902                                    GFP_KERNEL | __GFP_NOFAIL, i);
 903                spin_lock_init(&crng->lock);
 904                crng_initialize(crng);
 905                pool[i] = crng;
 906        }
 907        mb();
 908        if (cmpxchg(&crng_node_pool, NULL, pool)) {
 909                for_each_node(i)
 910                        kfree(pool[i]);
 911                kfree(pool);
 912        }
 913}
 914
 915static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
 916
 917static void numa_crng_init(void)
 918{
 919        schedule_work(&numa_crng_init_work);
 920}
 921#else
 922static void numa_crng_init(void) {}
 923#endif
 924
 925/*
 926 * crng_fast_load() can be called by code in the interrupt service
 927 * path.  So we can't afford to dilly-dally.
 928 */
 929static int crng_fast_load(const char *cp, size_t len)
 930{
 931        unsigned long flags;
 932        char *p;
 933
 934        if (!spin_trylock_irqsave(&primary_crng.lock, flags))
 935                return 0;
 936        if (crng_init != 0) {
 937                spin_unlock_irqrestore(&primary_crng.lock, flags);
 938                return 0;
 939        }
 940        p = (unsigned char *) &primary_crng.state[4];
 941        while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
 942                p[crng_init_cnt % CHACHA_KEY_SIZE] ^= *cp;
 943                cp++; crng_init_cnt++; len--;
 944        }
 945        spin_unlock_irqrestore(&primary_crng.lock, flags);
 946        if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
 947                invalidate_batched_entropy();
 948                crng_init = 1;
 949                wake_up_interruptible(&crng_init_wait);
 950                pr_notice("random: fast init done\n");
 951        }
 952        return 1;
 953}
 954
 955/*
 956 * crng_slow_load() is called by add_device_randomness, which has two
 957 * attributes.  (1) We can't trust the buffer passed to it is
 958 * guaranteed to be unpredictable (so it might not have any entropy at
 959 * all), and (2) it doesn't have the performance constraints of
 960 * crng_fast_load().
 961 *
 962 * So we do something more comprehensive which is guaranteed to touch
 963 * all of the primary_crng's state, and which uses a LFSR with a
 964 * period of 255 as part of the mixing algorithm.  Finally, we do
 965 * *not* advance crng_init_cnt since buffer we may get may be something
 966 * like a fixed DMI table (for example), which might very well be
 967 * unique to the machine, but is otherwise unvarying.
 968 */
 969static int crng_slow_load(const char *cp, size_t len)
 970{
 971        unsigned long           flags;
 972        static unsigned char    lfsr = 1;
 973        unsigned char           tmp;
 974        unsigned                i, max = CHACHA_KEY_SIZE;
 975        const char *            src_buf = cp;
 976        char *                  dest_buf = (char *) &primary_crng.state[4];
 977
 978        if (!spin_trylock_irqsave(&primary_crng.lock, flags))
 979                return 0;
 980        if (crng_init != 0) {
 981                spin_unlock_irqrestore(&primary_crng.lock, flags);
 982                return 0;
 983        }
 984        if (len > max)
 985                max = len;
 986
 987        for (i = 0; i < max ; i++) {
 988                tmp = lfsr;
 989                lfsr >>= 1;
 990                if (tmp & 1)
 991                        lfsr ^= 0xE1;
 992                tmp = dest_buf[i % CHACHA_KEY_SIZE];
 993                dest_buf[i % CHACHA_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
 994                lfsr += (tmp << 3) | (tmp >> 5);
 995        }
 996        spin_unlock_irqrestore(&primary_crng.lock, flags);
 997        return 1;
 998}
 999
1000static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
1001{
1002        unsigned long   flags;
1003        int             i, num;
1004        union {
1005                __u8    block[CHACHA_BLOCK_SIZE];
1006                __u32   key[8];
1007        } buf;
1008
1009        if (r) {
1010                num = extract_entropy(r, &buf, 32, 16, 0);
1011                if (num == 0)
1012                        return;
1013        } else {
1014                _extract_crng(&primary_crng, buf.block);
1015                _crng_backtrack_protect(&primary_crng, buf.block,
1016                                        CHACHA_KEY_SIZE);
1017        }
1018        spin_lock_irqsave(&crng->lock, flags);
1019        for (i = 0; i < 8; i++) {
1020                unsigned long   rv;
1021                if (!arch_get_random_seed_long(&rv) &&
1022                    !arch_get_random_long(&rv))
1023                        rv = random_get_entropy();
1024                crng->state[i+4] ^= buf.key[i] ^ rv;
1025        }
1026        memzero_explicit(&buf, sizeof(buf));
1027        crng->init_time = jiffies;
1028        spin_unlock_irqrestore(&crng->lock, flags);
1029        if (crng == &primary_crng && crng_init < 2) {
1030                invalidate_batched_entropy();
1031                numa_crng_init();
1032                crng_init = 2;
1033                process_random_ready_list();
1034                wake_up_interruptible(&crng_init_wait);
1035                pr_notice("random: crng init done\n");
1036                if (unseeded_warning.missed) {
1037                        pr_notice("random: %d get_random_xx warning(s) missed "
1038                                  "due to ratelimiting\n",
1039                                  unseeded_warning.missed);
1040                        unseeded_warning.missed = 0;
1041                }
1042                if (urandom_warning.missed) {
1043                        pr_notice("random: %d urandom warning(s) missed "
1044                                  "due to ratelimiting\n",
1045                                  urandom_warning.missed);
1046                        urandom_warning.missed = 0;
1047                }
1048        }
1049}
1050
1051static void _extract_crng(struct crng_state *crng,
1052                          __u8 out[CHACHA_BLOCK_SIZE])
1053{
1054        unsigned long v, flags;
1055
1056        if (crng_ready() &&
1057            (time_after(crng_global_init_time, crng->init_time) ||
1058             time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL)))
1059                crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
1060        spin_lock_irqsave(&crng->lock, flags);
1061        if (arch_get_random_long(&v))
1062                crng->state[14] ^= v;
1063        chacha20_block(&crng->state[0], out);
1064        if (crng->state[12] == 0)
1065                crng->state[13]++;
1066        spin_unlock_irqrestore(&crng->lock, flags);
1067}
1068
1069static void extract_crng(__u8 out[CHACHA_BLOCK_SIZE])
1070{
1071        struct crng_state *crng = NULL;
1072
1073#ifdef CONFIG_NUMA
1074        if (crng_node_pool)
1075                crng = crng_node_pool[numa_node_id()];
1076        if (crng == NULL)
1077#endif
1078                crng = &primary_crng;
1079        _extract_crng(crng, out);
1080}
1081
1082/*
1083 * Use the leftover bytes from the CRNG block output (if there is
1084 * enough) to mutate the CRNG key to provide backtracking protection.
1085 */
1086static void _crng_backtrack_protect(struct crng_state *crng,
1087                                    __u8 tmp[CHACHA_BLOCK_SIZE], int used)
1088{
1089        unsigned long   flags;
1090        __u32           *s, *d;
1091        int             i;
1092
1093        used = round_up(used, sizeof(__u32));
1094        if (used + CHACHA_KEY_SIZE > CHACHA_BLOCK_SIZE) {
1095                extract_crng(tmp);
1096                used = 0;
1097        }
1098        spin_lock_irqsave(&crng->lock, flags);
1099        s = (__u32 *) &tmp[used];
1100        d = &crng->state[4];
1101        for (i=0; i < 8; i++)
1102                *d++ ^= *s++;
1103        spin_unlock_irqrestore(&crng->lock, flags);
1104}
1105
1106static void crng_backtrack_protect(__u8 tmp[CHACHA_BLOCK_SIZE], int used)
1107{
1108        struct crng_state *crng = NULL;
1109
1110#ifdef CONFIG_NUMA
1111        if (crng_node_pool)
1112                crng = crng_node_pool[numa_node_id()];
1113        if (crng == NULL)
1114#endif
1115                crng = &primary_crng;
1116        _crng_backtrack_protect(crng, tmp, used);
1117}
1118
1119static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1120{
1121        ssize_t ret = 0, i = CHACHA_BLOCK_SIZE;
1122        __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1123        int large_request = (nbytes > 256);
1124
1125        while (nbytes) {
1126                if (large_request && need_resched()) {
1127                        if (signal_pending(current)) {
1128                                if (ret == 0)
1129                                        ret = -ERESTARTSYS;
1130                                break;
1131                        }
1132                        schedule();
1133                }
1134
1135                extract_crng(tmp);
1136                i = min_t(int, nbytes, CHACHA_BLOCK_SIZE);
1137                if (copy_to_user(buf, tmp, i)) {
1138                        ret = -EFAULT;
1139                        break;
1140                }
1141
1142                nbytes -= i;
1143                buf += i;
1144                ret += i;
1145        }
1146        crng_backtrack_protect(tmp, i);
1147
1148        /* Wipe data just written to memory */
1149        memzero_explicit(tmp, sizeof(tmp));
1150
1151        return ret;
1152}
1153
1154
1155/*********************************************************************
1156 *
1157 * Entropy input management
1158 *
1159 *********************************************************************/
1160
1161/* There is one of these per entropy source */
1162struct timer_rand_state {
1163        cycles_t last_time;
1164        long last_delta, last_delta2;
1165};
1166
1167#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1168
1169/*
1170 * Add device- or boot-specific data to the input pool to help
1171 * initialize it.
1172 *
1173 * None of this adds any entropy; it is meant to avoid the problem of
1174 * the entropy pool having similar initial state across largely
1175 * identical devices.
1176 */
1177void add_device_randomness(const void *buf, unsigned int size)
1178{
1179        unsigned long time = random_get_entropy() ^ jiffies;
1180        unsigned long flags;
1181
1182        if (!crng_ready() && size)
1183                crng_slow_load(buf, size);
1184
1185        trace_add_device_randomness(size, _RET_IP_);
1186        spin_lock_irqsave(&input_pool.lock, flags);
1187        _mix_pool_bytes(&input_pool, buf, size);
1188        _mix_pool_bytes(&input_pool, &time, sizeof(time));
1189        spin_unlock_irqrestore(&input_pool.lock, flags);
1190}
1191EXPORT_SYMBOL(add_device_randomness);
1192
1193static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1194
1195/*
1196 * This function adds entropy to the entropy "pool" by using timing
1197 * delays.  It uses the timer_rand_state structure to make an estimate
1198 * of how many bits of entropy this call has added to the pool.
1199 *
1200 * The number "num" is also added to the pool - it should somehow describe
1201 * the type of event which just happened.  This is currently 0-255 for
1202 * keyboard scan codes, and 256 upwards for interrupts.
1203 *
1204 */
1205static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1206{
1207        struct entropy_store    *r;
1208        struct {
1209                long jiffies;
1210                unsigned cycles;
1211                unsigned num;
1212        } sample;
1213        long delta, delta2, delta3;
1214
1215        sample.jiffies = jiffies;
1216        sample.cycles = random_get_entropy();
1217        sample.num = num;
1218        r = &input_pool;
1219        mix_pool_bytes(r, &sample, sizeof(sample));
1220
1221        /*
1222         * Calculate number of bits of randomness we probably added.
1223         * We take into account the first, second and third-order deltas
1224         * in order to make our estimate.
1225         */
1226        delta = sample.jiffies - state->last_time;
1227        state->last_time = sample.jiffies;
1228
1229        delta2 = delta - state->last_delta;
1230        state->last_delta = delta;
1231
1232        delta3 = delta2 - state->last_delta2;
1233        state->last_delta2 = delta2;
1234
1235        if (delta < 0)
1236                delta = -delta;
1237        if (delta2 < 0)
1238                delta2 = -delta2;
1239        if (delta3 < 0)
1240                delta3 = -delta3;
1241        if (delta > delta2)
1242                delta = delta2;
1243        if (delta > delta3)
1244                delta = delta3;
1245
1246        /*
1247         * delta is now minimum absolute delta.
1248         * Round down by 1 bit on general principles,
1249         * and limit entropy entimate to 12 bits.
1250         */
1251        credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1252}
1253
1254void add_input_randomness(unsigned int type, unsigned int code,
1255                                 unsigned int value)
1256{
1257        static unsigned char last_value;
1258
1259        /* ignore autorepeat and the like */
1260        if (value == last_value)
1261                return;
1262
1263        last_value = value;
1264        add_timer_randomness(&input_timer_state,
1265                             (type << 4) ^ code ^ (code >> 4) ^ value);
1266        trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1267}
1268EXPORT_SYMBOL_GPL(add_input_randomness);
1269
1270static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1271
1272#ifdef ADD_INTERRUPT_BENCH
1273static unsigned long avg_cycles, avg_deviation;
1274
1275#define AVG_SHIFT 8     /* Exponential average factor k=1/256 */
1276#define FIXED_1_2 (1 << (AVG_SHIFT-1))
1277
1278static void add_interrupt_bench(cycles_t start)
1279{
1280        long delta = random_get_entropy() - start;
1281
1282        /* Use a weighted moving average */
1283        delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1284        avg_cycles += delta;
1285        /* And average deviation */
1286        delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1287        avg_deviation += delta;
1288}
1289#else
1290#define add_interrupt_bench(x)
1291#endif
1292
1293static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1294{
1295        __u32 *ptr = (__u32 *) regs;
1296        unsigned int idx;
1297
1298        if (regs == NULL)
1299                return 0;
1300        idx = READ_ONCE(f->reg_idx);
1301        if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1302                idx = 0;
1303        ptr += idx++;
1304        WRITE_ONCE(f->reg_idx, idx);
1305        return *ptr;
1306}
1307
1308void add_interrupt_randomness(int irq, int irq_flags)
1309{
1310        struct entropy_store    *r;
1311        struct fast_pool        *fast_pool = this_cpu_ptr(&irq_randomness);
1312        struct pt_regs          *regs = get_irq_regs();
1313        unsigned long           now = jiffies;
1314        cycles_t                cycles = random_get_entropy();
1315        __u32                   c_high, j_high;
1316        __u64                   ip;
1317        unsigned long           seed;
1318        int                     credit = 0;
1319
1320        if (cycles == 0)
1321                cycles = get_reg(fast_pool, regs);
1322        c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1323        j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1324        fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1325        fast_pool->pool[1] ^= now ^ c_high;
1326        ip = regs ? instruction_pointer(regs) : _RET_IP_;
1327        fast_pool->pool[2] ^= ip;
1328        fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1329                get_reg(fast_pool, regs);
1330
1331        fast_mix(fast_pool);
1332        add_interrupt_bench(cycles);
1333
1334        if (unlikely(crng_init == 0)) {
1335                if ((fast_pool->count >= 64) &&
1336                    crng_fast_load((char *) fast_pool->pool,
1337                                   sizeof(fast_pool->pool))) {
1338                        fast_pool->count = 0;
1339                        fast_pool->last = now;
1340                }
1341                return;
1342        }
1343
1344        if ((fast_pool->count < 64) &&
1345            !time_after(now, fast_pool->last + HZ))
1346                return;
1347
1348        r = &input_pool;
1349        if (!spin_trylock(&r->lock))
1350                return;
1351
1352        fast_pool->last = now;
1353        __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1354
1355        /*
1356         * If we have architectural seed generator, produce a seed and
1357         * add it to the pool.  For the sake of paranoia don't let the
1358         * architectural seed generator dominate the input from the
1359         * interrupt noise.
1360         */
1361        if (arch_get_random_seed_long(&seed)) {
1362                __mix_pool_bytes(r, &seed, sizeof(seed));
1363                credit = 1;
1364        }
1365        spin_unlock(&r->lock);
1366
1367        fast_pool->count = 0;
1368
1369        /* award one bit for the contents of the fast pool */
1370        credit_entropy_bits(r, credit + 1);
1371}
1372EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1373
1374#ifdef CONFIG_BLOCK
1375void add_disk_randomness(struct gendisk *disk)
1376{
1377        if (!disk || !disk->random)
1378                return;
1379        /* first major is 1, so we get >= 0x200 here */
1380        add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1381        trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1382}
1383EXPORT_SYMBOL_GPL(add_disk_randomness);
1384#endif
1385
1386/*********************************************************************
1387 *
1388 * Entropy extraction routines
1389 *
1390 *********************************************************************/
1391
1392/*
1393 * This utility inline function is responsible for transferring entropy
1394 * from the primary pool to the secondary extraction pool. We make
1395 * sure we pull enough for a 'catastrophic reseed'.
1396 */
1397static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1398static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1399{
1400        if (!r->pull ||
1401            r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1402            r->entropy_count > r->poolinfo->poolfracbits)
1403                return;
1404
1405        _xfer_secondary_pool(r, nbytes);
1406}
1407
1408static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1409{
1410        __u32   tmp[OUTPUT_POOL_WORDS];
1411
1412        int bytes = nbytes;
1413
1414        /* pull at least as much as a wakeup */
1415        bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1416        /* but never more than the buffer size */
1417        bytes = min_t(int, bytes, sizeof(tmp));
1418
1419        trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1420                                  ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1421        bytes = extract_entropy(r->pull, tmp, bytes,
1422                                random_read_wakeup_bits / 8, 0);
1423        mix_pool_bytes(r, tmp, bytes);
1424        credit_entropy_bits(r, bytes*8);
1425}
1426
1427/*
1428 * Used as a workqueue function so that when the input pool is getting
1429 * full, we can "spill over" some entropy to the output pools.  That
1430 * way the output pools can store some of the excess entropy instead
1431 * of letting it go to waste.
1432 */
1433static void push_to_pool(struct work_struct *work)
1434{
1435        struct entropy_store *r = container_of(work, struct entropy_store,
1436                                              push_work);
1437        BUG_ON(!r);
1438        _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1439        trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1440                           r->pull->entropy_count >> ENTROPY_SHIFT);
1441}
1442
1443/*
1444 * This function decides how many bytes to actually take from the
1445 * given pool, and also debits the entropy count accordingly.
1446 */
1447static size_t account(struct entropy_store *r, size_t nbytes, int min,
1448                      int reserved)
1449{
1450        int entropy_count, orig, have_bytes;
1451        size_t ibytes, nfrac;
1452
1453        BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1454
1455        /* Can we pull enough? */
1456retry:
1457        entropy_count = orig = READ_ONCE(r->entropy_count);
1458        ibytes = nbytes;
1459        /* never pull more than available */
1460        have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1461
1462        if ((have_bytes -= reserved) < 0)
1463                have_bytes = 0;
1464        ibytes = min_t(size_t, ibytes, have_bytes);
1465        if (ibytes < min)
1466                ibytes = 0;
1467
1468        if (unlikely(entropy_count < 0)) {
1469                pr_warn("random: negative entropy count: pool %s count %d\n",
1470                        r->name, entropy_count);
1471                WARN_ON(1);
1472                entropy_count = 0;
1473        }
1474        nfrac = ibytes << (ENTROPY_SHIFT + 3);
1475        if ((size_t) entropy_count > nfrac)
1476                entropy_count -= nfrac;
1477        else
1478                entropy_count = 0;
1479
1480        if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1481                goto retry;
1482
1483        trace_debit_entropy(r->name, 8 * ibytes);
1484        if (ibytes &&
1485            (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1486                wake_up_interruptible(&random_write_wait);
1487                kill_fasync(&fasync, SIGIO, POLL_OUT);
1488        }
1489
1490        return ibytes;
1491}
1492
1493/*
1494 * This function does the actual extraction for extract_entropy and
1495 * extract_entropy_user.
1496 *
1497 * Note: we assume that .poolwords is a multiple of 16 words.
1498 */
1499static void extract_buf(struct entropy_store *r, __u8 *out)
1500{
1501        int i;
1502        union {
1503                __u32 w[5];
1504                unsigned long l[LONGS(20)];
1505        } hash;
1506        __u32 workspace[SHA_WORKSPACE_WORDS];
1507        unsigned long flags;
1508
1509        /*
1510         * If we have an architectural hardware random number
1511         * generator, use it for SHA's initial vector
1512         */
1513        sha_init(hash.w);
1514        for (i = 0; i < LONGS(20); i++) {
1515                unsigned long v;
1516                if (!arch_get_random_long(&v))
1517                        break;
1518                hash.l[i] = v;
1519        }
1520
1521        /* Generate a hash across the pool, 16 words (512 bits) at a time */
1522        spin_lock_irqsave(&r->lock, flags);
1523        for (i = 0; i < r->poolinfo->poolwords; i += 16)
1524                sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1525
1526        /*
1527         * We mix the hash back into the pool to prevent backtracking
1528         * attacks (where the attacker knows the state of the pool
1529         * plus the current outputs, and attempts to find previous
1530         * ouputs), unless the hash function can be inverted. By
1531         * mixing at least a SHA1 worth of hash data back, we make
1532         * brute-forcing the feedback as hard as brute-forcing the
1533         * hash.
1534         */
1535        __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1536        spin_unlock_irqrestore(&r->lock, flags);
1537
1538        memzero_explicit(workspace, sizeof(workspace));
1539
1540        /*
1541         * In case the hash function has some recognizable output
1542         * pattern, we fold it in half. Thus, we always feed back
1543         * twice as much data as we output.
1544         */
1545        hash.w[0] ^= hash.w[3];
1546        hash.w[1] ^= hash.w[4];
1547        hash.w[2] ^= rol32(hash.w[2], 16);
1548
1549        memcpy(out, &hash, EXTRACT_SIZE);
1550        memzero_explicit(&hash, sizeof(hash));
1551}
1552
1553static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1554                                size_t nbytes, int fips)
1555{
1556        ssize_t ret = 0, i;
1557        __u8 tmp[EXTRACT_SIZE];
1558        unsigned long flags;
1559
1560        while (nbytes) {
1561                extract_buf(r, tmp);
1562
1563                if (fips) {
1564                        spin_lock_irqsave(&r->lock, flags);
1565                        if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1566                                panic("Hardware RNG duplicated output!\n");
1567                        memcpy(r->last_data, tmp, EXTRACT_SIZE);
1568                        spin_unlock_irqrestore(&r->lock, flags);
1569                }
1570                i = min_t(int, nbytes, EXTRACT_SIZE);
1571                memcpy(buf, tmp, i);
1572                nbytes -= i;
1573                buf += i;
1574                ret += i;
1575        }
1576
1577        /* Wipe data just returned from memory */
1578        memzero_explicit(tmp, sizeof(tmp));
1579
1580        return ret;
1581}
1582
1583/*
1584 * This function extracts randomness from the "entropy pool", and
1585 * returns it in a buffer.
1586 *
1587 * The min parameter specifies the minimum amount we can pull before
1588 * failing to avoid races that defeat catastrophic reseeding while the
1589 * reserved parameter indicates how much entropy we must leave in the
1590 * pool after each pull to avoid starving other readers.
1591 */
1592static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1593                                 size_t nbytes, int min, int reserved)
1594{
1595        __u8 tmp[EXTRACT_SIZE];
1596        unsigned long flags;
1597
1598        /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1599        if (fips_enabled) {
1600                spin_lock_irqsave(&r->lock, flags);
1601                if (!r->last_data_init) {
1602                        r->last_data_init = 1;
1603                        spin_unlock_irqrestore(&r->lock, flags);
1604                        trace_extract_entropy(r->name, EXTRACT_SIZE,
1605                                              ENTROPY_BITS(r), _RET_IP_);
1606                        xfer_secondary_pool(r, EXTRACT_SIZE);
1607                        extract_buf(r, tmp);
1608                        spin_lock_irqsave(&r->lock, flags);
1609                        memcpy(r->last_data, tmp, EXTRACT_SIZE);
1610                }
1611                spin_unlock_irqrestore(&r->lock, flags);
1612        }
1613
1614        trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1615        xfer_secondary_pool(r, nbytes);
1616        nbytes = account(r, nbytes, min, reserved);
1617
1618        return _extract_entropy(r, buf, nbytes, fips_enabled);
1619}
1620
1621/*
1622 * This function extracts randomness from the "entropy pool", and
1623 * returns it in a userspace buffer.
1624 */
1625static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1626                                    size_t nbytes)
1627{
1628        ssize_t ret = 0, i;
1629        __u8 tmp[EXTRACT_SIZE];
1630        int large_request = (nbytes > 256);
1631
1632        trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1633        if (!r->initialized && r->pull) {
1634                xfer_secondary_pool(r, ENTROPY_BITS(r->pull)/8);
1635                if (!r->initialized)
1636                        return 0;
1637        }
1638        xfer_secondary_pool(r, nbytes);
1639        nbytes = account(r, nbytes, 0, 0);
1640
1641        while (nbytes) {
1642                if (large_request && need_resched()) {
1643                        if (signal_pending(current)) {
1644                                if (ret == 0)
1645                                        ret = -ERESTARTSYS;
1646                                break;
1647                        }
1648                        schedule();
1649                }
1650
1651                extract_buf(r, tmp);
1652                i = min_t(int, nbytes, EXTRACT_SIZE);
1653                if (copy_to_user(buf, tmp, i)) {
1654                        ret = -EFAULT;
1655                        break;
1656                }
1657
1658                nbytes -= i;
1659                buf += i;
1660                ret += i;
1661        }
1662
1663        /* Wipe data just returned from memory */
1664        memzero_explicit(tmp, sizeof(tmp));
1665
1666        return ret;
1667}
1668
1669#define warn_unseeded_randomness(previous) \
1670        _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1671
1672static void _warn_unseeded_randomness(const char *func_name, void *caller,
1673                                      void **previous)
1674{
1675#ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1676        const bool print_once = false;
1677#else
1678        static bool print_once __read_mostly;
1679#endif
1680
1681        if (print_once ||
1682            crng_ready() ||
1683            (previous && (caller == READ_ONCE(*previous))))
1684                return;
1685        WRITE_ONCE(*previous, caller);
1686#ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1687        print_once = true;
1688#endif
1689        if (__ratelimit(&unseeded_warning))
1690                pr_notice("random: %s called from %pS with crng_init=%d\n",
1691                          func_name, caller, crng_init);
1692}
1693
1694/*
1695 * This function is the exported kernel interface.  It returns some
1696 * number of good random numbers, suitable for key generation, seeding
1697 * TCP sequence numbers, etc.  It does not rely on the hardware random
1698 * number generator.  For random bytes direct from the hardware RNG
1699 * (when available), use get_random_bytes_arch(). In order to ensure
1700 * that the randomness provided by this function is okay, the function
1701 * wait_for_random_bytes() should be called and return 0 at least once
1702 * at any point prior.
1703 */
1704static void _get_random_bytes(void *buf, int nbytes)
1705{
1706        __u8 tmp[CHACHA_BLOCK_SIZE] __aligned(4);
1707
1708        trace_get_random_bytes(nbytes, _RET_IP_);
1709
1710        while (nbytes >= CHACHA_BLOCK_SIZE) {
1711                extract_crng(buf);
1712                buf += CHACHA_BLOCK_SIZE;
1713                nbytes -= CHACHA_BLOCK_SIZE;
1714        }
1715
1716        if (nbytes > 0) {
1717                extract_crng(tmp);
1718                memcpy(buf, tmp, nbytes);
1719                crng_backtrack_protect(tmp, nbytes);
1720        } else
1721                crng_backtrack_protect(tmp, CHACHA_BLOCK_SIZE);
1722        memzero_explicit(tmp, sizeof(tmp));
1723}
1724
1725void get_random_bytes(void *buf, int nbytes)
1726{
1727        static void *previous;
1728
1729        warn_unseeded_randomness(&previous);
1730        _get_random_bytes(buf, nbytes);
1731}
1732EXPORT_SYMBOL(get_random_bytes);
1733
1734
1735/*
1736 * Each time the timer fires, we expect that we got an unpredictable
1737 * jump in the cycle counter. Even if the timer is running on another
1738 * CPU, the timer activity will be touching the stack of the CPU that is
1739 * generating entropy..
1740 *
1741 * Note that we don't re-arm the timer in the timer itself - we are
1742 * happy to be scheduled away, since that just makes the load more
1743 * complex, but we do not want the timer to keep ticking unless the
1744 * entropy loop is running.
1745 *
1746 * So the re-arming always happens in the entropy loop itself.
1747 */
1748static void entropy_timer(struct timer_list *t)
1749{
1750        credit_entropy_bits(&input_pool, 1);
1751}
1752
1753/*
1754 * If we have an actual cycle counter, see if we can
1755 * generate enough entropy with timing noise
1756 */
1757static void try_to_generate_entropy(void)
1758{
1759        struct {
1760                unsigned long now;
1761                struct timer_list timer;
1762        } stack;
1763
1764        stack.now = random_get_entropy();
1765
1766        /* Slow counter - or none. Don't even bother */
1767        if (stack.now == random_get_entropy())
1768                return;
1769
1770        timer_setup_on_stack(&stack.timer, entropy_timer, 0);
1771        while (!crng_ready()) {
1772                if (!timer_pending(&stack.timer))
1773                        mod_timer(&stack.timer, jiffies+1);
1774                mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1775                schedule();
1776                stack.now = random_get_entropy();
1777        }
1778
1779        del_timer_sync(&stack.timer);
1780        destroy_timer_on_stack(&stack.timer);
1781        mix_pool_bytes(&input_pool, &stack.now, sizeof(stack.now));
1782}
1783
1784/*
1785 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1786 * cryptographically secure random numbers. This applies to: the /dev/urandom
1787 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1788 * family of functions. Using any of these functions without first calling
1789 * this function forfeits the guarantee of security.
1790 *
1791 * Returns: 0 if the urandom pool has been seeded.
1792 *          -ERESTARTSYS if the function was interrupted by a signal.
1793 */
1794int wait_for_random_bytes(void)
1795{
1796        if (likely(crng_ready()))
1797                return 0;
1798
1799        do {
1800                int ret;
1801                ret = wait_event_interruptible_timeout(crng_init_wait, crng_ready(), HZ);
1802                if (ret)
1803                        return ret > 0 ? 0 : ret;
1804
1805                try_to_generate_entropy();
1806        } while (!crng_ready());
1807
1808        return 0;
1809}
1810EXPORT_SYMBOL(wait_for_random_bytes);
1811
1812/*
1813 * Returns whether or not the urandom pool has been seeded and thus guaranteed
1814 * to supply cryptographically secure random numbers. This applies to: the
1815 * /dev/urandom device, the get_random_bytes function, and the get_random_{u32,
1816 * ,u64,int,long} family of functions.
1817 *
1818 * Returns: true if the urandom pool has been seeded.
1819 *          false if the urandom pool has not been seeded.
1820 */
1821bool rng_is_initialized(void)
1822{
1823        return crng_ready();
1824}
1825EXPORT_SYMBOL(rng_is_initialized);
1826
1827/*
1828 * Add a callback function that will be invoked when the nonblocking
1829 * pool is initialised.
1830 *
1831 * returns: 0 if callback is successfully added
1832 *          -EALREADY if pool is already initialised (callback not called)
1833 *          -ENOENT if module for callback is not alive
1834 */
1835int add_random_ready_callback(struct random_ready_callback *rdy)
1836{
1837        struct module *owner;
1838        unsigned long flags;
1839        int err = -EALREADY;
1840
1841        if (crng_ready())
1842                return err;
1843
1844        owner = rdy->owner;
1845        if (!try_module_get(owner))
1846                return -ENOENT;
1847
1848        spin_lock_irqsave(&random_ready_list_lock, flags);
1849        if (crng_ready())
1850                goto out;
1851
1852        owner = NULL;
1853
1854        list_add(&rdy->list, &random_ready_list);
1855        err = 0;
1856
1857out:
1858        spin_unlock_irqrestore(&random_ready_list_lock, flags);
1859
1860        module_put(owner);
1861
1862        return err;
1863}
1864EXPORT_SYMBOL(add_random_ready_callback);
1865
1866/*
1867 * Delete a previously registered readiness callback function.
1868 */
1869void del_random_ready_callback(struct random_ready_callback *rdy)
1870{
1871        unsigned long flags;
1872        struct module *owner = NULL;
1873
1874        spin_lock_irqsave(&random_ready_list_lock, flags);
1875        if (!list_empty(&rdy->list)) {
1876                list_del_init(&rdy->list);
1877                owner = rdy->owner;
1878        }
1879        spin_unlock_irqrestore(&random_ready_list_lock, flags);
1880
1881        module_put(owner);
1882}
1883EXPORT_SYMBOL(del_random_ready_callback);
1884
1885/*
1886 * This function will use the architecture-specific hardware random
1887 * number generator if it is available.  The arch-specific hw RNG will
1888 * almost certainly be faster than what we can do in software, but it
1889 * is impossible to verify that it is implemented securely (as
1890 * opposed, to, say, the AES encryption of a sequence number using a
1891 * key known by the NSA).  So it's useful if we need the speed, but
1892 * only if we're willing to trust the hardware manufacturer not to
1893 * have put in a back door.
1894 *
1895 * Return number of bytes filled in.
1896 */
1897int __must_check get_random_bytes_arch(void *buf, int nbytes)
1898{
1899        int left = nbytes;
1900        char *p = buf;
1901
1902        trace_get_random_bytes_arch(left, _RET_IP_);
1903        while (left) {
1904                unsigned long v;
1905                int chunk = min_t(int, left, sizeof(unsigned long));
1906
1907                if (!arch_get_random_long(&v))
1908                        break;
1909
1910                memcpy(p, &v, chunk);
1911                p += chunk;
1912                left -= chunk;
1913        }
1914
1915        return nbytes - left;
1916}
1917EXPORT_SYMBOL(get_random_bytes_arch);
1918
1919/*
1920 * init_std_data - initialize pool with system data
1921 *
1922 * @r: pool to initialize
1923 *
1924 * This function clears the pool's entropy count and mixes some system
1925 * data into the pool to prepare it for use. The pool is not cleared
1926 * as that can only decrease the entropy in the pool.
1927 */
1928static void __init init_std_data(struct entropy_store *r)
1929{
1930        int i;
1931        ktime_t now = ktime_get_real();
1932        unsigned long rv;
1933
1934        r->last_pulled = jiffies;
1935        mix_pool_bytes(r, &now, sizeof(now));
1936        for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1937                if (!arch_get_random_seed_long(&rv) &&
1938                    !arch_get_random_long(&rv))
1939                        rv = random_get_entropy();
1940                mix_pool_bytes(r, &rv, sizeof(rv));
1941        }
1942        mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1943}
1944
1945/*
1946 * Note that setup_arch() may call add_device_randomness()
1947 * long before we get here. This allows seeding of the pools
1948 * with some platform dependent data very early in the boot
1949 * process. But it limits our options here. We must use
1950 * statically allocated structures that already have all
1951 * initializations complete at compile time. We should also
1952 * take care not to overwrite the precious per platform data
1953 * we were given.
1954 */
1955int __init rand_initialize(void)
1956{
1957        init_std_data(&input_pool);
1958        init_std_data(&blocking_pool);
1959        crng_initialize(&primary_crng);
1960        crng_global_init_time = jiffies;
1961        if (ratelimit_disable) {
1962                urandom_warning.interval = 0;
1963                unseeded_warning.interval = 0;
1964        }
1965        return 0;
1966}
1967
1968#ifdef CONFIG_BLOCK
1969void rand_initialize_disk(struct gendisk *disk)
1970{
1971        struct timer_rand_state *state;
1972
1973        /*
1974         * If kzalloc returns null, we just won't use that entropy
1975         * source.
1976         */
1977        state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1978        if (state) {
1979                state->last_time = INITIAL_JIFFIES;
1980                disk->random = state;
1981        }
1982}
1983#endif
1984
1985static ssize_t
1986_random_read(int nonblock, char __user *buf, size_t nbytes)
1987{
1988        ssize_t n;
1989
1990        if (nbytes == 0)
1991                return 0;
1992
1993        nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1994        while (1) {
1995                n = extract_entropy_user(&blocking_pool, buf, nbytes);
1996                if (n < 0)
1997                        return n;
1998                trace_random_read(n*8, (nbytes-n)*8,
1999                                  ENTROPY_BITS(&blocking_pool),
2000                                  ENTROPY_BITS(&input_pool));
2001                if (n > 0)
2002                        return n;
2003
2004                /* Pool is (near) empty.  Maybe wait and retry. */
2005                if (nonblock)
2006                        return -EAGAIN;
2007
2008                wait_event_interruptible(random_read_wait,
2009                    blocking_pool.initialized &&
2010                    (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits));
2011                if (signal_pending(current))
2012                        return -ERESTARTSYS;
2013        }
2014}
2015
2016static ssize_t
2017random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
2018{
2019        return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
2020}
2021
2022static ssize_t
2023urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
2024{
2025        unsigned long flags;
2026        static int maxwarn = 10;
2027        int ret;
2028
2029        if (!crng_ready() && maxwarn > 0) {
2030                maxwarn--;
2031                if (__ratelimit(&urandom_warning))
2032                        printk(KERN_NOTICE "random: %s: uninitialized "
2033                               "urandom read (%zd bytes read)\n",
2034                               current->comm, nbytes);
2035                spin_lock_irqsave(&primary_crng.lock, flags);
2036                crng_init_cnt = 0;
2037                spin_unlock_irqrestore(&primary_crng.lock, flags);
2038        }
2039        nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
2040        ret = extract_crng_user(buf, nbytes);
2041        trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
2042        return ret;
2043}
2044
2045static __poll_t
2046random_poll(struct file *file, poll_table * wait)
2047{
2048        __poll_t mask;
2049
2050        poll_wait(file, &random_read_wait, wait);
2051        poll_wait(file, &random_write_wait, wait);
2052        mask = 0;
2053        if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
2054                mask |= EPOLLIN | EPOLLRDNORM;
2055        if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
2056                mask |= EPOLLOUT | EPOLLWRNORM;
2057        return mask;
2058}
2059
2060static int
2061write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
2062{
2063        size_t bytes;
2064        __u32 t, buf[16];
2065        const char __user *p = buffer;
2066
2067        while (count > 0) {
2068                int b, i = 0;
2069
2070                bytes = min(count, sizeof(buf));
2071                if (copy_from_user(&buf, p, bytes))
2072                        return -EFAULT;
2073
2074                for (b = bytes ; b > 0 ; b -= sizeof(__u32), i++) {
2075                        if (!arch_get_random_int(&t))
2076                                break;
2077                        buf[i] ^= t;
2078                }
2079
2080                count -= bytes;
2081                p += bytes;
2082
2083                mix_pool_bytes(r, buf, bytes);
2084                cond_resched();
2085        }
2086
2087        return 0;
2088}
2089
2090static ssize_t random_write(struct file *file, const char __user *buffer,
2091                            size_t count, loff_t *ppos)
2092{
2093        size_t ret;
2094
2095        ret = write_pool(&input_pool, buffer, count);
2096        if (ret)
2097                return ret;
2098
2099        return (ssize_t)count;
2100}
2101
2102static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
2103{
2104        int size, ent_count;
2105        int __user *p = (int __user *)arg;
2106        int retval;
2107
2108        switch (cmd) {
2109        case RNDGETENTCNT:
2110                /* inherently racy, no point locking */
2111                ent_count = ENTROPY_BITS(&input_pool);
2112                if (put_user(ent_count, p))
2113                        return -EFAULT;
2114                return 0;
2115        case RNDADDTOENTCNT:
2116                if (!capable(CAP_SYS_ADMIN))
2117                        return -EPERM;
2118                if (get_user(ent_count, p))
2119                        return -EFAULT;
2120                return credit_entropy_bits_safe(&input_pool, ent_count);
2121        case RNDADDENTROPY:
2122                if (!capable(CAP_SYS_ADMIN))
2123                        return -EPERM;
2124                if (get_user(ent_count, p++))
2125                        return -EFAULT;
2126                if (ent_count < 0)
2127                        return -EINVAL;
2128                if (get_user(size, p++))
2129                        return -EFAULT;
2130                retval = write_pool(&input_pool, (const char __user *)p,
2131                                    size);
2132                if (retval < 0)
2133                        return retval;
2134                return credit_entropy_bits_safe(&input_pool, ent_count);
2135        case RNDZAPENTCNT:
2136        case RNDCLEARPOOL:
2137                /*
2138                 * Clear the entropy pool counters. We no longer clear
2139                 * the entropy pool, as that's silly.
2140                 */
2141                if (!capable(CAP_SYS_ADMIN))
2142                        return -EPERM;
2143                input_pool.entropy_count = 0;
2144                blocking_pool.entropy_count = 0;
2145                return 0;
2146        case RNDRESEEDCRNG:
2147                if (!capable(CAP_SYS_ADMIN))
2148                        return -EPERM;
2149                if (crng_init < 2)
2150                        return -ENODATA;
2151                crng_reseed(&primary_crng, NULL);
2152                crng_global_init_time = jiffies - 1;
2153                return 0;
2154        default:
2155                return -EINVAL;
2156        }
2157}
2158
2159static int random_fasync(int fd, struct file *filp, int on)
2160{
2161        return fasync_helper(fd, filp, on, &fasync);
2162}
2163
2164const struct file_operations random_fops = {
2165        .read  = random_read,
2166        .write = random_write,
2167        .poll  = random_poll,
2168        .unlocked_ioctl = random_ioctl,
2169        .compat_ioctl = compat_ptr_ioctl,
2170        .fasync = random_fasync,
2171        .llseek = noop_llseek,
2172};
2173
2174const struct file_operations urandom_fops = {
2175        .read  = urandom_read,
2176        .write = random_write,
2177        .unlocked_ioctl = random_ioctl,
2178        .compat_ioctl = compat_ptr_ioctl,
2179        .fasync = random_fasync,
2180        .llseek = noop_llseek,
2181};
2182
2183SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
2184                unsigned int, flags)
2185{
2186        int ret;
2187
2188        if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
2189                return -EINVAL;
2190
2191        if (count > INT_MAX)
2192                count = INT_MAX;
2193
2194        if (flags & GRND_RANDOM)
2195                return _random_read(flags & GRND_NONBLOCK, buf, count);
2196
2197        if (!crng_ready()) {
2198                if (flags & GRND_NONBLOCK)
2199                        return -EAGAIN;
2200                ret = wait_for_random_bytes();
2201                if (unlikely(ret))
2202                        return ret;
2203        }
2204        return urandom_read(NULL, buf, count, NULL);
2205}
2206
2207/********************************************************************
2208 *
2209 * Sysctl interface
2210 *
2211 ********************************************************************/
2212
2213#ifdef CONFIG_SYSCTL
2214
2215#include <linux/sysctl.h>
2216
2217static int min_read_thresh = 8, min_write_thresh;
2218static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
2219static int max_write_thresh = INPUT_POOL_WORDS * 32;
2220static int random_min_urandom_seed = 60;
2221static char sysctl_bootid[16];
2222
2223/*
2224 * This function is used to return both the bootid UUID, and random
2225 * UUID.  The difference is in whether table->data is NULL; if it is,
2226 * then a new UUID is generated and returned to the user.
2227 *
2228 * If the user accesses this via the proc interface, the UUID will be
2229 * returned as an ASCII string in the standard UUID format; if via the
2230 * sysctl system call, as 16 bytes of binary data.
2231 */
2232static int proc_do_uuid(struct ctl_table *table, int write,
2233                        void __user *buffer, size_t *lenp, loff_t *ppos)
2234{
2235        struct ctl_table fake_table;
2236        unsigned char buf[64], tmp_uuid[16], *uuid;
2237
2238        uuid = table->data;
2239        if (!uuid) {
2240                uuid = tmp_uuid;
2241                generate_random_uuid(uuid);
2242        } else {
2243                static DEFINE_SPINLOCK(bootid_spinlock);
2244
2245                spin_lock(&bootid_spinlock);
2246                if (!uuid[8])
2247                        generate_random_uuid(uuid);
2248                spin_unlock(&bootid_spinlock);
2249        }
2250
2251        sprintf(buf, "%pU", uuid);
2252
2253        fake_table.data = buf;
2254        fake_table.maxlen = sizeof(buf);
2255
2256        return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2257}
2258
2259/*
2260 * Return entropy available scaled to integral bits
2261 */
2262static int proc_do_entropy(struct ctl_table *table, int write,
2263                           void __user *buffer, size_t *lenp, loff_t *ppos)
2264{
2265        struct ctl_table fake_table;
2266        int entropy_count;
2267
2268        entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2269
2270        fake_table.data = &entropy_count;
2271        fake_table.maxlen = sizeof(entropy_count);
2272
2273        return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2274}
2275
2276static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2277extern struct ctl_table random_table[];
2278struct ctl_table random_table[] = {
2279        {
2280                .procname       = "poolsize",
2281                .data           = &sysctl_poolsize,
2282                .maxlen         = sizeof(int),
2283                .mode           = 0444,
2284                .proc_handler   = proc_dointvec,
2285        },
2286        {
2287                .procname       = "entropy_avail",
2288                .maxlen         = sizeof(int),
2289                .mode           = 0444,
2290                .proc_handler   = proc_do_entropy,
2291                .data           = &input_pool.entropy_count,
2292        },
2293        {
2294                .procname       = "read_wakeup_threshold",
2295                .data           = &random_read_wakeup_bits,
2296                .maxlen         = sizeof(int),
2297                .mode           = 0644,
2298                .proc_handler   = proc_dointvec_minmax,
2299                .extra1         = &min_read_thresh,
2300                .extra2         = &max_read_thresh,
2301        },
2302        {
2303                .procname       = "write_wakeup_threshold",
2304                .data           = &random_write_wakeup_bits,
2305                .maxlen         = sizeof(int),
2306                .mode           = 0644,
2307                .proc_handler   = proc_dointvec_minmax,
2308                .extra1         = &min_write_thresh,
2309                .extra2         = &max_write_thresh,
2310        },
2311        {
2312                .procname       = "urandom_min_reseed_secs",
2313                .data           = &random_min_urandom_seed,
2314                .maxlen         = sizeof(int),
2315                .mode           = 0644,
2316                .proc_handler   = proc_dointvec,
2317        },
2318        {
2319                .procname       = "boot_id",
2320                .data           = &sysctl_bootid,
2321                .maxlen         = 16,
2322                .mode           = 0444,
2323                .proc_handler   = proc_do_uuid,
2324        },
2325        {
2326                .procname       = "uuid",
2327                .maxlen         = 16,
2328                .mode           = 0444,
2329                .proc_handler   = proc_do_uuid,
2330        },
2331#ifdef ADD_INTERRUPT_BENCH
2332        {
2333                .procname       = "add_interrupt_avg_cycles",
2334                .data           = &avg_cycles,
2335                .maxlen         = sizeof(avg_cycles),
2336                .mode           = 0444,
2337                .proc_handler   = proc_doulongvec_minmax,
2338        },
2339        {
2340                .procname       = "add_interrupt_avg_deviation",
2341                .data           = &avg_deviation,
2342                .maxlen         = sizeof(avg_deviation),
2343                .mode           = 0444,
2344                .proc_handler   = proc_doulongvec_minmax,
2345        },
2346#endif
2347        { }
2348};
2349#endif  /* CONFIG_SYSCTL */
2350
2351struct batched_entropy {
2352        union {
2353                u64 entropy_u64[CHACHA_BLOCK_SIZE / sizeof(u64)];
2354                u32 entropy_u32[CHACHA_BLOCK_SIZE / sizeof(u32)];
2355        };
2356        unsigned int position;
2357        spinlock_t batch_lock;
2358};
2359
2360/*
2361 * Get a random word for internal kernel use only. The quality of the random
2362 * number is either as good as RDRAND or as good as /dev/urandom, with the
2363 * goal of being quite fast and not depleting entropy. In order to ensure
2364 * that the randomness provided by this function is okay, the function
2365 * wait_for_random_bytes() should be called and return 0 at least once
2366 * at any point prior.
2367 */
2368static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64) = {
2369        .batch_lock     = __SPIN_LOCK_UNLOCKED(batched_entropy_u64.lock),
2370};
2371
2372u64 get_random_u64(void)
2373{
2374        u64 ret;
2375        unsigned long flags;
2376        struct batched_entropy *batch;
2377        static void *previous;
2378
2379#if BITS_PER_LONG == 64
2380        if (arch_get_random_long((unsigned long *)&ret))
2381                return ret;
2382#else
2383        if (arch_get_random_long((unsigned long *)&ret) &&
2384            arch_get_random_long((unsigned long *)&ret + 1))
2385            return ret;
2386#endif
2387
2388        warn_unseeded_randomness(&previous);
2389
2390        batch = raw_cpu_ptr(&batched_entropy_u64);
2391        spin_lock_irqsave(&batch->batch_lock, flags);
2392        if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2393                extract_crng((u8 *)batch->entropy_u64);
2394                batch->position = 0;
2395        }
2396        ret = batch->entropy_u64[batch->position++];
2397        spin_unlock_irqrestore(&batch->batch_lock, flags);
2398        return ret;
2399}
2400EXPORT_SYMBOL(get_random_u64);
2401
2402static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32) = {
2403        .batch_lock     = __SPIN_LOCK_UNLOCKED(batched_entropy_u32.lock),
2404};
2405u32 get_random_u32(void)
2406{
2407        u32 ret;
2408        unsigned long flags;
2409        struct batched_entropy *batch;
2410        static void *previous;
2411
2412        if (arch_get_random_int(&ret))
2413                return ret;
2414
2415        warn_unseeded_randomness(&previous);
2416
2417        batch = raw_cpu_ptr(&batched_entropy_u32);
2418        spin_lock_irqsave(&batch->batch_lock, flags);
2419        if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2420                extract_crng((u8 *)batch->entropy_u32);
2421                batch->position = 0;
2422        }
2423        ret = batch->entropy_u32[batch->position++];
2424        spin_unlock_irqrestore(&batch->batch_lock, flags);
2425        return ret;
2426}
2427EXPORT_SYMBOL(get_random_u32);
2428
2429/* It's important to invalidate all potential batched entropy that might
2430 * be stored before the crng is initialized, which we can do lazily by
2431 * simply resetting the counter to zero so that it's re-extracted on the
2432 * next usage. */
2433static void invalidate_batched_entropy(void)
2434{
2435        int cpu;
2436        unsigned long flags;
2437
2438        for_each_possible_cpu (cpu) {
2439                struct batched_entropy *batched_entropy;
2440
2441                batched_entropy = per_cpu_ptr(&batched_entropy_u32, cpu);
2442                spin_lock_irqsave(&batched_entropy->batch_lock, flags);
2443                batched_entropy->position = 0;
2444                spin_unlock(&batched_entropy->batch_lock);
2445
2446                batched_entropy = per_cpu_ptr(&batched_entropy_u64, cpu);
2447                spin_lock(&batched_entropy->batch_lock);
2448                batched_entropy->position = 0;
2449                spin_unlock_irqrestore(&batched_entropy->batch_lock, flags);
2450        }
2451}
2452
2453/**
2454 * randomize_page - Generate a random, page aligned address
2455 * @start:      The smallest acceptable address the caller will take.
2456 * @range:      The size of the area, starting at @start, within which the
2457 *              random address must fall.
2458 *
2459 * If @start + @range would overflow, @range is capped.
2460 *
2461 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2462 * @start was already page aligned.  We now align it regardless.
2463 *
2464 * Return: A page aligned address within [start, start + range).  On error,
2465 * @start is returned.
2466 */
2467unsigned long
2468randomize_page(unsigned long start, unsigned long range)
2469{
2470        if (!PAGE_ALIGNED(start)) {
2471                range -= PAGE_ALIGN(start) - start;
2472                start = PAGE_ALIGN(start);
2473        }
2474
2475        if (start > ULONG_MAX - range)
2476                range = ULONG_MAX - start;
2477
2478        range >>= PAGE_SHIFT;
2479
2480        if (range == 0)
2481                return start;
2482
2483        return start + (get_random_long() % range << PAGE_SHIFT);
2484}
2485
2486/* Interface for in-kernel drivers of true hardware RNGs.
2487 * Those devices may produce endless random bits and will be throttled
2488 * when our pool is full.
2489 */
2490void add_hwgenerator_randomness(const char *buffer, size_t count,
2491                                size_t entropy)
2492{
2493        struct entropy_store *poolp = &input_pool;
2494
2495        if (unlikely(crng_init == 0)) {
2496                crng_fast_load(buffer, count);
2497                return;
2498        }
2499
2500        /* Suspend writing if we're above the trickle threshold.
2501         * We'll be woken up again once below random_write_wakeup_thresh,
2502         * or when the calling thread is about to terminate.
2503         */
2504        wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2505                        ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2506        mix_pool_bytes(poolp, buffer, count);
2507        credit_entropy_bits(poolp, entropy);
2508}
2509EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2510
2511/* Handle random seed passed by bootloader.
2512 * If the seed is trustworthy, it would be regarded as hardware RNGs. Otherwise
2513 * it would be regarded as device data.
2514 * The decision is controlled by CONFIG_RANDOM_TRUST_BOOTLOADER.
2515 */
2516void add_bootloader_randomness(const void *buf, unsigned int size)
2517{
2518        if (IS_ENABLED(CONFIG_RANDOM_TRUST_BOOTLOADER))
2519                add_hwgenerator_randomness(buf, size, size * 8);
2520        else
2521                add_device_randomness(buf, size);
2522}
2523EXPORT_SYMBOL_GPL(add_bootloader_randomness);
2524