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 three exported interfaces; the first is one designed to
 105 * be used from within the kernel:
 106 *
 107 *      void get_random_bytes(void *buf, int nbytes);
 108 *
 109 * This interface will return the requested number of random bytes,
 110 * and place it in the requested buffer.
 111 *
 112 * The two other interfaces are two character devices /dev/random and
 113 * /dev/urandom.  /dev/random is suitable for use when very high
 114 * quality randomness is desired (for example, for key generation or
 115 * one-time pads), as it will only return a maximum of the number of
 116 * bits of randomness (as estimated by the random number generator)
 117 * contained in the entropy pool.
 118 *
 119 * The /dev/urandom device does not have this limit, and will return
 120 * as many bytes as are requested.  As more and more random bytes are
 121 * requested without giving time for the entropy pool to recharge,
 122 * this will result in random numbers that are merely cryptographically
 123 * strong.  For many applications, however, this is acceptable.
 124 *
 125 * Exported interfaces ---- input
 126 * ==============================
 127 *
 128 * The current exported interfaces for gathering environmental noise
 129 * from the devices are:
 130 *
 131 *      void add_device_randomness(const void *buf, unsigned int size);
 132 *      void add_input_randomness(unsigned int type, unsigned int code,
 133 *                                unsigned int value);
 134 *      void add_interrupt_randomness(int irq, int irq_flags);
 135 *      void add_disk_randomness(struct gendisk *disk);
 136 *
 137 * add_device_randomness() is for adding data to the random pool that
 138 * is likely to differ between two devices (or possibly even per boot).
 139 * This would be things like MAC addresses or serial numbers, or the
 140 * read-out of the RTC. This does *not* add any actual entropy to the
 141 * pool, but it initializes the pool to different values for devices
 142 * that might otherwise be identical and have very little entropy
 143 * available to them (particularly common in the embedded world).
 144 *
 145 * add_input_randomness() uses the input layer interrupt timing, as well as
 146 * the event type information from the hardware.
 147 *
 148 * add_interrupt_randomness() uses the interrupt timing as random
 149 * inputs to the entropy pool. Using the cycle counters and the irq source
 150 * as inputs, it feeds the randomness roughly once a second.
 151 *
 152 * add_disk_randomness() uses what amounts to the seek time of block
 153 * layer request events, on a per-disk_devt basis, as input to the
 154 * entropy pool. Note that high-speed solid state drives with very low
 155 * seek times do not make for good sources of entropy, as their seek
 156 * times are usually fairly consistent.
 157 *
 158 * All of these routines try to estimate how many bits of randomness a
 159 * particular randomness source.  They do this by keeping track of the
 160 * first and second order deltas of the event timings.
 161 *
 162 * Ensuring unpredictability at system startup
 163 * ============================================
 164 *
 165 * When any operating system starts up, it will go through a sequence
 166 * of actions that are fairly predictable by an adversary, especially
 167 * if the start-up does not involve interaction with a human operator.
 168 * This reduces the actual number of bits of unpredictability in the
 169 * entropy pool below the value in entropy_count.  In order to
 170 * counteract this effect, it helps to carry information in the
 171 * entropy pool across shut-downs and start-ups.  To do this, put the
 172 * following lines an appropriate script which is run during the boot
 173 * sequence:
 174 *
 175 *      echo "Initializing random number generator..."
 176 *      random_seed=/var/run/random-seed
 177 *      # Carry a random seed from start-up to start-up
 178 *      # Load and then save the whole entropy pool
 179 *      if [ -f $random_seed ]; then
 180 *              cat $random_seed >/dev/urandom
 181 *      else
 182 *              touch $random_seed
 183 *      fi
 184 *      chmod 600 $random_seed
 185 *      dd if=/dev/urandom of=$random_seed count=1 bs=512
 186 *
 187 * and the following lines in an appropriate script which is run as
 188 * the system is shutdown:
 189 *
 190 *      # Carry a random seed from shut-down to start-up
 191 *      # Save the whole entropy pool
 192 *      echo "Saving random seed..."
 193 *      random_seed=/var/run/random-seed
 194 *      touch $random_seed
 195 *      chmod 600 $random_seed
 196 *      dd if=/dev/urandom of=$random_seed count=1 bs=512
 197 *
 198 * For example, on most modern systems using the System V init
 199 * scripts, such code fragments would be found in
 200 * /etc/rc.d/init.d/random.  On older Linux systems, the correct script
 201 * location might be in /etc/rcb.d/rc.local or /etc/rc.d/rc.0.
 202 *
 203 * Effectively, these commands cause the contents of the entropy pool
 204 * to be saved at shut-down time and reloaded into the entropy pool at
 205 * start-up.  (The 'dd' in the addition to the bootup script is to
 206 * make sure that /etc/random-seed is different for every start-up,
 207 * even if the system crashes without executing rc.0.)  Even with
 208 * complete knowledge of the start-up activities, predicting the state
 209 * of the entropy pool requires knowledge of the previous history of
 210 * the system.
 211 *
 212 * Configuring the /dev/random driver under Linux
 213 * ==============================================
 214 *
 215 * The /dev/random driver under Linux uses minor numbers 8 and 9 of
 216 * the /dev/mem major number (#1).  So if your system does not have
 217 * /dev/random and /dev/urandom created already, they can be created
 218 * by using the commands:
 219 *
 220 *      mknod /dev/random c 1 8
 221 *      mknod /dev/urandom c 1 9
 222 *
 223 * Acknowledgements:
 224 * =================
 225 *
 226 * Ideas for constructing this random number generator were derived
 227 * from Pretty Good Privacy's random number generator, and from private
 228 * discussions with Phil Karn.  Colin Plumb provided a faster random
 229 * number generator, which speed up the mixing function of the entropy
 230 * pool, taken from PGPfone.  Dale Worley has also contributed many
 231 * useful ideas and suggestions to improve this driver.
 232 *
 233 * Any flaws in the design are solely my responsibility, and should
 234 * not be attributed to the Phil, Colin, or any of authors of PGP.
 235 *
 236 * Further background information on this topic may be obtained from
 237 * RFC 1750, "Randomness Recommendations for Security", by Donald
 238 * Eastlake, Steve Crocker, and Jeff Schiller.
 239 */
 240
 241#include <linux/utsname.h>
 242#include <linux/module.h>
 243#include <linux/kernel.h>
 244#include <linux/major.h>
 245#include <linux/string.h>
 246#include <linux/fcntl.h>
 247#include <linux/slab.h>
 248#include <linux/random.h>
 249#include <linux/poll.h>
 250#include <linux/init.h>
 251#include <linux/fs.h>
 252#include <linux/genhd.h>
 253#include <linux/interrupt.h>
 254#include <linux/mm.h>
 255#include <linux/nodemask.h>
 256#include <linux/spinlock.h>
 257#include <linux/kthread.h>
 258#include <linux/percpu.h>
 259#include <linux/cryptohash.h>
 260#include <linux/fips.h>
 261#include <linux/ptrace.h>
 262#include <linux/workqueue.h>
 263#include <linux/irq.h>
 264#include <linux/syscalls.h>
 265#include <linux/completion.h>
 266#include <linux/uuid.h>
 267#include <crypto/chacha20.h>
 268
 269#include <asm/processor.h>
 270#include <linux/uaccess.h>
 271#include <asm/irq.h>
 272#include <asm/irq_regs.h>
 273#include <asm/io.h>
 274
 275#define CREATE_TRACE_POINTS
 276#include <trace/events/random.h>
 277
 278/* #define ADD_INTERRUPT_BENCH */
 279
 280/*
 281 * Configuration information
 282 */
 283#define INPUT_POOL_SHIFT        12
 284#define INPUT_POOL_WORDS        (1 << (INPUT_POOL_SHIFT-5))
 285#define OUTPUT_POOL_SHIFT       10
 286#define OUTPUT_POOL_WORDS       (1 << (OUTPUT_POOL_SHIFT-5))
 287#define SEC_XFER_SIZE           512
 288#define EXTRACT_SIZE            10
 289
 290
 291#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
 292
 293/*
 294 * To allow fractional bits to be tracked, the entropy_count field is
 295 * denominated in units of 1/8th bits.
 296 *
 297 * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
 298 * credit_entropy_bits() needs to be 64 bits wide.
 299 */
 300#define ENTROPY_SHIFT 3
 301#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
 302
 303/*
 304 * The minimum number of bits of entropy before we wake up a read on
 305 * /dev/random.  Should be enough to do a significant reseed.
 306 */
 307static int random_read_wakeup_bits = 64;
 308
 309/*
 310 * If the entropy count falls under this number of bits, then we
 311 * should wake up processes which are selecting or polling on write
 312 * access to /dev/random.
 313 */
 314static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
 315
 316/*
 317 * Originally, we used a primitive polynomial of degree .poolwords
 318 * over GF(2).  The taps for various sizes are defined below.  They
 319 * were chosen to be evenly spaced except for the last tap, which is 1
 320 * to get the twisting happening as fast as possible.
 321 *
 322 * For the purposes of better mixing, we use the CRC-32 polynomial as
 323 * well to make a (modified) twisted Generalized Feedback Shift
 324 * Register.  (See M. Matsumoto & Y. Kurita, 1992.  Twisted GFSR
 325 * generators.  ACM Transactions on Modeling and Computer Simulation
 326 * 2(3):179-194.  Also see M. Matsumoto & Y. Kurita, 1994.  Twisted
 327 * GFSR generators II.  ACM Transactions on Modeling and Computer
 328 * Simulation 4:254-266)
 329 *
 330 * Thanks to Colin Plumb for suggesting this.
 331 *
 332 * The mixing operation is much less sensitive than the output hash,
 333 * where we use SHA-1.  All that we want of mixing operation is that
 334 * it be a good non-cryptographic hash; i.e. it not produce collisions
 335 * when fed "random" data of the sort we expect to see.  As long as
 336 * the pool state differs for different inputs, we have preserved the
 337 * input entropy and done a good job.  The fact that an intelligent
 338 * attacker can construct inputs that will produce controlled
 339 * alterations to the pool's state is not important because we don't
 340 * consider such inputs to contribute any randomness.  The only
 341 * property we need with respect to them is that the attacker can't
 342 * increase his/her knowledge of the pool's state.  Since all
 343 * additions are reversible (knowing the final state and the input,
 344 * you can reconstruct the initial state), if an attacker has any
 345 * uncertainty about the initial state, he/she can only shuffle that
 346 * uncertainty about, but never cause any collisions (which would
 347 * decrease the uncertainty).
 348 *
 349 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
 350 * Videau in their paper, "The Linux Pseudorandom Number Generator
 351 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf).  In their
 352 * paper, they point out that we are not using a true Twisted GFSR,
 353 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
 354 * is, with only three taps, instead of the six that we are using).
 355 * As a result, the resulting polynomial is neither primitive nor
 356 * irreducible, and hence does not have a maximal period over
 357 * GF(2**32).  They suggest a slight change to the generator
 358 * polynomial which improves the resulting TGFSR polynomial to be
 359 * irreducible, which we have made here.
 360 */
 361static struct poolinfo {
 362        int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
 363#define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
 364        int tap1, tap2, tap3, tap4, tap5;
 365} poolinfo_table[] = {
 366        /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
 367        /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
 368        { S(128),       104,    76,     51,     25,     1 },
 369        /* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
 370        /* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
 371        { S(32),        26,     19,     14,     7,      1 },
 372#if 0
 373        /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1  -- 115 */
 374        { S(2048),      1638,   1231,   819,    411,    1 },
 375
 376        /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
 377        { S(1024),      817,    615,    412,    204,    1 },
 378
 379        /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
 380        { S(1024),      819,    616,    410,    207,    2 },
 381
 382        /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
 383        { S(512),       411,    308,    208,    104,    1 },
 384
 385        /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
 386        { S(512),       409,    307,    206,    102,    2 },
 387        /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
 388        { S(512),       409,    309,    205,    103,    2 },
 389
 390        /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
 391        { S(256),       205,    155,    101,    52,     1 },
 392
 393        /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
 394        { S(128),       103,    78,     51,     27,     2 },
 395
 396        /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
 397        { S(64),        52,     39,     26,     14,     1 },
 398#endif
 399};
 400
 401/*
 402 * Static global variables
 403 */
 404static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
 405static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
 406static struct fasync_struct *fasync;
 407
 408static DEFINE_SPINLOCK(random_ready_list_lock);
 409static LIST_HEAD(random_ready_list);
 410
 411struct crng_state {
 412        __u32           state[16];
 413        unsigned long   init_time;
 414        spinlock_t      lock;
 415};
 416
 417struct crng_state primary_crng = {
 418        .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
 419};
 420
 421/*
 422 * crng_init =  0 --> Uninitialized
 423 *              1 --> Initialized
 424 *              2 --> Initialized from input_pool
 425 *
 426 * crng_init is protected by primary_crng->lock, and only increases
 427 * its value (from 0->1->2).
 428 */
 429static int crng_init = 0;
 430#define crng_ready() (likely(crng_init > 0))
 431static int crng_init_cnt = 0;
 432#define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
 433static void _extract_crng(struct crng_state *crng,
 434                          __u8 out[CHACHA20_BLOCK_SIZE]);
 435static void _crng_backtrack_protect(struct crng_state *crng,
 436                                    __u8 tmp[CHACHA20_BLOCK_SIZE], int used);
 437static void process_random_ready_list(void);
 438static void _get_random_bytes(void *buf, int nbytes);
 439
 440/**********************************************************************
 441 *
 442 * OS independent entropy store.   Here are the functions which handle
 443 * storing entropy in an entropy pool.
 444 *
 445 **********************************************************************/
 446
 447struct entropy_store;
 448struct entropy_store {
 449        /* read-only data: */
 450        const struct poolinfo *poolinfo;
 451        __u32 *pool;
 452        const char *name;
 453        struct entropy_store *pull;
 454        struct work_struct push_work;
 455
 456        /* read-write data: */
 457        unsigned long last_pulled;
 458        spinlock_t lock;
 459        unsigned short add_ptr;
 460        unsigned short input_rotate;
 461        int entropy_count;
 462        int entropy_total;
 463        unsigned int initialized:1;
 464        unsigned int last_data_init:1;
 465        __u8 last_data[EXTRACT_SIZE];
 466};
 467
 468static ssize_t extract_entropy(struct entropy_store *r, void *buf,
 469                               size_t nbytes, int min, int rsvd);
 470static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
 471                                size_t nbytes, int fips);
 472
 473static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
 474static void push_to_pool(struct work_struct *work);
 475static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
 476static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
 477
 478static struct entropy_store input_pool = {
 479        .poolinfo = &poolinfo_table[0],
 480        .name = "input",
 481        .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
 482        .pool = input_pool_data
 483};
 484
 485static struct entropy_store blocking_pool = {
 486        .poolinfo = &poolinfo_table[1],
 487        .name = "blocking",
 488        .pull = &input_pool,
 489        .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
 490        .pool = blocking_pool_data,
 491        .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
 492                                        push_to_pool),
 493};
 494
 495static __u32 const twist_table[8] = {
 496        0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
 497        0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
 498
 499/*
 500 * This function adds bytes into the entropy "pool".  It does not
 501 * update the entropy estimate.  The caller should call
 502 * credit_entropy_bits if this is appropriate.
 503 *
 504 * The pool is stirred with a primitive polynomial of the appropriate
 505 * degree, and then twisted.  We twist by three bits at a time because
 506 * it's cheap to do so and helps slightly in the expected case where
 507 * the entropy is concentrated in the low-order bits.
 508 */
 509static void _mix_pool_bytes(struct entropy_store *r, const void *in,
 510                            int nbytes)
 511{
 512        unsigned long i, tap1, tap2, tap3, tap4, tap5;
 513        int input_rotate;
 514        int wordmask = r->poolinfo->poolwords - 1;
 515        const char *bytes = in;
 516        __u32 w;
 517
 518        tap1 = r->poolinfo->tap1;
 519        tap2 = r->poolinfo->tap2;
 520        tap3 = r->poolinfo->tap3;
 521        tap4 = r->poolinfo->tap4;
 522        tap5 = r->poolinfo->tap5;
 523
 524        input_rotate = r->input_rotate;
 525        i = r->add_ptr;
 526
 527        /* mix one byte at a time to simplify size handling and churn faster */
 528        while (nbytes--) {
 529                w = rol32(*bytes++, input_rotate);
 530                i = (i - 1) & wordmask;
 531
 532                /* XOR in the various taps */
 533                w ^= r->pool[i];
 534                w ^= r->pool[(i + tap1) & wordmask];
 535                w ^= r->pool[(i + tap2) & wordmask];
 536                w ^= r->pool[(i + tap3) & wordmask];
 537                w ^= r->pool[(i + tap4) & wordmask];
 538                w ^= r->pool[(i + tap5) & wordmask];
 539
 540                /* Mix the result back in with a twist */
 541                r->pool[i] = (w >> 3) ^ twist_table[w & 7];
 542
 543                /*
 544                 * Normally, we add 7 bits of rotation to the pool.
 545                 * At the beginning of the pool, add an extra 7 bits
 546                 * rotation, so that successive passes spread the
 547                 * input bits across the pool evenly.
 548                 */
 549                input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
 550        }
 551
 552        r->input_rotate = input_rotate;
 553        r->add_ptr = i;
 554}
 555
 556static void __mix_pool_bytes(struct entropy_store *r, const void *in,
 557                             int nbytes)
 558{
 559        trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
 560        _mix_pool_bytes(r, in, nbytes);
 561}
 562
 563static void mix_pool_bytes(struct entropy_store *r, const void *in,
 564                           int nbytes)
 565{
 566        unsigned long flags;
 567
 568        trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
 569        spin_lock_irqsave(&r->lock, flags);
 570        _mix_pool_bytes(r, in, nbytes);
 571        spin_unlock_irqrestore(&r->lock, flags);
 572}
 573
 574struct fast_pool {
 575        __u32           pool[4];
 576        unsigned long   last;
 577        unsigned short  reg_idx;
 578        unsigned char   count;
 579};
 580
 581/*
 582 * This is a fast mixing routine used by the interrupt randomness
 583 * collector.  It's hardcoded for an 128 bit pool and assumes that any
 584 * locks that might be needed are taken by the caller.
 585 */
 586static void fast_mix(struct fast_pool *f)
 587{
 588        __u32 a = f->pool[0],   b = f->pool[1];
 589        __u32 c = f->pool[2],   d = f->pool[3];
 590
 591        a += b;                 c += d;
 592        b = rol32(b, 6);        d = rol32(d, 27);
 593        d ^= a;                 b ^= c;
 594
 595        a += b;                 c += d;
 596        b = rol32(b, 16);       d = rol32(d, 14);
 597        d ^= a;                 b ^= c;
 598
 599        a += b;                 c += d;
 600        b = rol32(b, 6);        d = rol32(d, 27);
 601        d ^= a;                 b ^= c;
 602
 603        a += b;                 c += d;
 604        b = rol32(b, 16);       d = rol32(d, 14);
 605        d ^= a;                 b ^= c;
 606
 607        f->pool[0] = a;  f->pool[1] = b;
 608        f->pool[2] = c;  f->pool[3] = d;
 609        f->count++;
 610}
 611
 612static void process_random_ready_list(void)
 613{
 614        unsigned long flags;
 615        struct random_ready_callback *rdy, *tmp;
 616
 617        spin_lock_irqsave(&random_ready_list_lock, flags);
 618        list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
 619                struct module *owner = rdy->owner;
 620
 621                list_del_init(&rdy->list);
 622                rdy->func(rdy);
 623                module_put(owner);
 624        }
 625        spin_unlock_irqrestore(&random_ready_list_lock, flags);
 626}
 627
 628/*
 629 * Credit (or debit) the entropy store with n bits of entropy.
 630 * Use credit_entropy_bits_safe() if the value comes from userspace
 631 * or otherwise should be checked for extreme values.
 632 */
 633static void credit_entropy_bits(struct entropy_store *r, int nbits)
 634{
 635        int entropy_count, orig;
 636        const int pool_size = r->poolinfo->poolfracbits;
 637        int nfrac = nbits << ENTROPY_SHIFT;
 638
 639        if (!nbits)
 640                return;
 641
 642retry:
 643        entropy_count = orig = READ_ONCE(r->entropy_count);
 644        if (nfrac < 0) {
 645                /* Debit */
 646                entropy_count += nfrac;
 647        } else {
 648                /*
 649                 * Credit: we have to account for the possibility of
 650                 * overwriting already present entropy.  Even in the
 651                 * ideal case of pure Shannon entropy, new contributions
 652                 * approach the full value asymptotically:
 653                 *
 654                 * entropy <- entropy + (pool_size - entropy) *
 655                 *      (1 - exp(-add_entropy/pool_size))
 656                 *
 657                 * For add_entropy <= pool_size/2 then
 658                 * (1 - exp(-add_entropy/pool_size)) >=
 659                 *    (add_entropy/pool_size)*0.7869...
 660                 * so we can approximate the exponential with
 661                 * 3/4*add_entropy/pool_size and still be on the
 662                 * safe side by adding at most pool_size/2 at a time.
 663                 *
 664                 * The use of pool_size-2 in the while statement is to
 665                 * prevent rounding artifacts from making the loop
 666                 * arbitrarily long; this limits the loop to log2(pool_size)*2
 667                 * turns no matter how large nbits is.
 668                 */
 669                int pnfrac = nfrac;
 670                const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
 671                /* The +2 corresponds to the /4 in the denominator */
 672
 673                do {
 674                        unsigned int anfrac = min(pnfrac, pool_size/2);
 675                        unsigned int add =
 676                                ((pool_size - entropy_count)*anfrac*3) >> s;
 677
 678                        entropy_count += add;
 679                        pnfrac -= anfrac;
 680                } while (unlikely(entropy_count < pool_size-2 && pnfrac));
 681        }
 682
 683        if (unlikely(entropy_count < 0)) {
 684                pr_warn("random: negative entropy/overflow: pool %s count %d\n",
 685                        r->name, entropy_count);
 686                WARN_ON(1);
 687                entropy_count = 0;
 688        } else if (entropy_count > pool_size)
 689                entropy_count = pool_size;
 690        if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
 691                goto retry;
 692
 693        r->entropy_total += nbits;
 694        if (!r->initialized && r->entropy_total > 128) {
 695                r->initialized = 1;
 696                r->entropy_total = 0;
 697        }
 698
 699        trace_credit_entropy_bits(r->name, nbits,
 700                                  entropy_count >> ENTROPY_SHIFT,
 701                                  r->entropy_total, _RET_IP_);
 702
 703        if (r == &input_pool) {
 704                int entropy_bits = entropy_count >> ENTROPY_SHIFT;
 705
 706                if (crng_init < 2 && entropy_bits >= 128) {
 707                        crng_reseed(&primary_crng, r);
 708                        entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
 709                }
 710
 711                /* should we wake readers? */
 712                if (entropy_bits >= random_read_wakeup_bits) {
 713                        wake_up_interruptible(&random_read_wait);
 714                        kill_fasync(&fasync, SIGIO, POLL_IN);
 715                }
 716                /* If the input pool is getting full, send some
 717                 * entropy to the blocking pool until it is 75% full.
 718                 */
 719                if (entropy_bits > random_write_wakeup_bits &&
 720                    r->initialized &&
 721                    r->entropy_total >= 2*random_read_wakeup_bits) {
 722                        struct entropy_store *other = &blocking_pool;
 723
 724                        if (other->entropy_count <=
 725                            3 * other->poolinfo->poolfracbits / 4) {
 726                                schedule_work(&other->push_work);
 727                                r->entropy_total = 0;
 728                        }
 729                }
 730        }
 731}
 732
 733static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
 734{
 735        const int nbits_max = (int)(~0U >> (ENTROPY_SHIFT + 1));
 736
 737        if (nbits < 0)
 738                return -EINVAL;
 739
 740        /* Cap the value to avoid overflows */
 741        nbits = min(nbits,  nbits_max);
 742
 743        credit_entropy_bits(r, nbits);
 744        return 0;
 745}
 746
 747/*********************************************************************
 748 *
 749 * CRNG using CHACHA20
 750 *
 751 *********************************************************************/
 752
 753#define CRNG_RESEED_INTERVAL (300*HZ)
 754
 755static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
 756
 757#ifdef CONFIG_NUMA
 758/*
 759 * Hack to deal with crazy userspace progams when they are all trying
 760 * to access /dev/urandom in parallel.  The programs are almost
 761 * certainly doing something terribly wrong, but we'll work around
 762 * their brain damage.
 763 */
 764static struct crng_state **crng_node_pool __read_mostly;
 765#endif
 766
 767static void invalidate_batched_entropy(void);
 768
 769static void crng_initialize(struct crng_state *crng)
 770{
 771        int             i;
 772        unsigned long   rv;
 773
 774        memcpy(&crng->state[0], "expand 32-byte k", 16);
 775        if (crng == &primary_crng)
 776                _extract_entropy(&input_pool, &crng->state[4],
 777                                 sizeof(__u32) * 12, 0);
 778        else
 779                _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
 780        for (i = 4; i < 16; i++) {
 781                if (!arch_get_random_seed_long(&rv) &&
 782                    !arch_get_random_long(&rv))
 783                        rv = random_get_entropy();
 784                crng->state[i] ^= rv;
 785        }
 786        crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
 787}
 788
 789static int crng_fast_load(const char *cp, size_t len)
 790{
 791        unsigned long flags;
 792        char *p;
 793
 794        if (!spin_trylock_irqsave(&primary_crng.lock, flags))
 795                return 0;
 796        if (crng_ready()) {
 797                spin_unlock_irqrestore(&primary_crng.lock, flags);
 798                return 0;
 799        }
 800        p = (unsigned char *) &primary_crng.state[4];
 801        while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
 802                p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
 803                cp++; crng_init_cnt++; len--;
 804        }
 805        spin_unlock_irqrestore(&primary_crng.lock, flags);
 806        if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
 807                invalidate_batched_entropy();
 808                crng_init = 1;
 809                wake_up_interruptible(&crng_init_wait);
 810                pr_notice("random: fast init done\n");
 811        }
 812        return 1;
 813}
 814
 815static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
 816{
 817        unsigned long   flags;
 818        int             i, num;
 819        union {
 820                __u8    block[CHACHA20_BLOCK_SIZE];
 821                __u32   key[8];
 822        } buf;
 823
 824        if (r) {
 825                num = extract_entropy(r, &buf, 32, 16, 0);
 826                if (num == 0)
 827                        return;
 828        } else {
 829                _extract_crng(&primary_crng, buf.block);
 830                _crng_backtrack_protect(&primary_crng, buf.block,
 831                                        CHACHA20_KEY_SIZE);
 832        }
 833        spin_lock_irqsave(&primary_crng.lock, flags);
 834        for (i = 0; i < 8; i++) {
 835                unsigned long   rv;
 836                if (!arch_get_random_seed_long(&rv) &&
 837                    !arch_get_random_long(&rv))
 838                        rv = random_get_entropy();
 839                crng->state[i+4] ^= buf.key[i] ^ rv;
 840        }
 841        memzero_explicit(&buf, sizeof(buf));
 842        crng->init_time = jiffies;
 843        spin_unlock_irqrestore(&primary_crng.lock, flags);
 844        if (crng == &primary_crng && crng_init < 2) {
 845                invalidate_batched_entropy();
 846                crng_init = 2;
 847                process_random_ready_list();
 848                wake_up_interruptible(&crng_init_wait);
 849                pr_notice("random: crng init done\n");
 850        }
 851}
 852
 853static void _extract_crng(struct crng_state *crng,
 854                          __u8 out[CHACHA20_BLOCK_SIZE])
 855{
 856        unsigned long v, flags;
 857
 858        if (crng_init > 1 &&
 859            time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL))
 860                crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
 861        spin_lock_irqsave(&crng->lock, flags);
 862        if (arch_get_random_long(&v))
 863                crng->state[14] ^= v;
 864        chacha20_block(&crng->state[0], out);
 865        if (crng->state[12] == 0)
 866                crng->state[13]++;
 867        spin_unlock_irqrestore(&crng->lock, flags);
 868}
 869
 870static void extract_crng(__u8 out[CHACHA20_BLOCK_SIZE])
 871{
 872        struct crng_state *crng = NULL;
 873
 874#ifdef CONFIG_NUMA
 875        if (crng_node_pool)
 876                crng = crng_node_pool[numa_node_id()];
 877        if (crng == NULL)
 878#endif
 879                crng = &primary_crng;
 880        _extract_crng(crng, out);
 881}
 882
 883/*
 884 * Use the leftover bytes from the CRNG block output (if there is
 885 * enough) to mutate the CRNG key to provide backtracking protection.
 886 */
 887static void _crng_backtrack_protect(struct crng_state *crng,
 888                                    __u8 tmp[CHACHA20_BLOCK_SIZE], int used)
 889{
 890        unsigned long   flags;
 891        __u32           *s, *d;
 892        int             i;
 893
 894        used = round_up(used, sizeof(__u32));
 895        if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
 896                extract_crng(tmp);
 897                used = 0;
 898        }
 899        spin_lock_irqsave(&crng->lock, flags);
 900        s = (__u32 *) &tmp[used];
 901        d = &crng->state[4];
 902        for (i=0; i < 8; i++)
 903                *d++ ^= *s++;
 904        spin_unlock_irqrestore(&crng->lock, flags);
 905}
 906
 907static void crng_backtrack_protect(__u8 tmp[CHACHA20_BLOCK_SIZE], int used)
 908{
 909        struct crng_state *crng = NULL;
 910
 911#ifdef CONFIG_NUMA
 912        if (crng_node_pool)
 913                crng = crng_node_pool[numa_node_id()];
 914        if (crng == NULL)
 915#endif
 916                crng = &primary_crng;
 917        _crng_backtrack_protect(crng, tmp, used);
 918}
 919
 920static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
 921{
 922        ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
 923        __u8 tmp[CHACHA20_BLOCK_SIZE];
 924        int large_request = (nbytes > 256);
 925
 926        while (nbytes) {
 927                if (large_request && need_resched()) {
 928                        if (signal_pending(current)) {
 929                                if (ret == 0)
 930                                        ret = -ERESTARTSYS;
 931                                break;
 932                        }
 933                        schedule();
 934                }
 935
 936                extract_crng(tmp);
 937                i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
 938                if (copy_to_user(buf, tmp, i)) {
 939                        ret = -EFAULT;
 940                        break;
 941                }
 942
 943                nbytes -= i;
 944                buf += i;
 945                ret += i;
 946        }
 947        crng_backtrack_protect(tmp, i);
 948
 949        /* Wipe data just written to memory */
 950        memzero_explicit(tmp, sizeof(tmp));
 951
 952        return ret;
 953}
 954
 955
 956/*********************************************************************
 957 *
 958 * Entropy input management
 959 *
 960 *********************************************************************/
 961
 962/* There is one of these per entropy source */
 963struct timer_rand_state {
 964        cycles_t last_time;
 965        long last_delta, last_delta2;
 966        unsigned dont_count_entropy:1;
 967};
 968
 969#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
 970
 971/*
 972 * Add device- or boot-specific data to the input pool to help
 973 * initialize it.
 974 *
 975 * None of this adds any entropy; it is meant to avoid the problem of
 976 * the entropy pool having similar initial state across largely
 977 * identical devices.
 978 */
 979void add_device_randomness(const void *buf, unsigned int size)
 980{
 981        unsigned long time = random_get_entropy() ^ jiffies;
 982        unsigned long flags;
 983
 984        if (!crng_ready()) {
 985                crng_fast_load(buf, size);
 986                return;
 987        }
 988
 989        trace_add_device_randomness(size, _RET_IP_);
 990        spin_lock_irqsave(&input_pool.lock, flags);
 991        _mix_pool_bytes(&input_pool, buf, size);
 992        _mix_pool_bytes(&input_pool, &time, sizeof(time));
 993        spin_unlock_irqrestore(&input_pool.lock, flags);
 994}
 995EXPORT_SYMBOL(add_device_randomness);
 996
 997static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
 998
 999/*
1000 * This function adds entropy to the entropy "pool" by using timing
1001 * delays.  It uses the timer_rand_state structure to make an estimate
1002 * of how many bits of entropy this call has added to the pool.
1003 *
1004 * The number "num" is also added to the pool - it should somehow describe
1005 * the type of event which just happened.  This is currently 0-255 for
1006 * keyboard scan codes, and 256 upwards for interrupts.
1007 *
1008 */
1009static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1010{
1011        struct entropy_store    *r;
1012        struct {
1013                long jiffies;
1014                unsigned cycles;
1015                unsigned num;
1016        } sample;
1017        long delta, delta2, delta3;
1018
1019        preempt_disable();
1020
1021        sample.jiffies = jiffies;
1022        sample.cycles = random_get_entropy();
1023        sample.num = num;
1024        r = &input_pool;
1025        mix_pool_bytes(r, &sample, sizeof(sample));
1026
1027        /*
1028         * Calculate number of bits of randomness we probably added.
1029         * We take into account the first, second and third-order deltas
1030         * in order to make our estimate.
1031         */
1032
1033        if (!state->dont_count_entropy) {
1034                delta = sample.jiffies - state->last_time;
1035                state->last_time = sample.jiffies;
1036
1037                delta2 = delta - state->last_delta;
1038                state->last_delta = delta;
1039
1040                delta3 = delta2 - state->last_delta2;
1041                state->last_delta2 = delta2;
1042
1043                if (delta < 0)
1044                        delta = -delta;
1045                if (delta2 < 0)
1046                        delta2 = -delta2;
1047                if (delta3 < 0)
1048                        delta3 = -delta3;
1049                if (delta > delta2)
1050                        delta = delta2;
1051                if (delta > delta3)
1052                        delta = delta3;
1053
1054                /*
1055                 * delta is now minimum absolute delta.
1056                 * Round down by 1 bit on general principles,
1057                 * and limit entropy entimate to 12 bits.
1058                 */
1059                credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1060        }
1061        preempt_enable();
1062}
1063
1064void add_input_randomness(unsigned int type, unsigned int code,
1065                                 unsigned int value)
1066{
1067        static unsigned char last_value;
1068
1069        /* ignore autorepeat and the like */
1070        if (value == last_value)
1071                return;
1072
1073        last_value = value;
1074        add_timer_randomness(&input_timer_state,
1075                             (type << 4) ^ code ^ (code >> 4) ^ value);
1076        trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1077}
1078EXPORT_SYMBOL_GPL(add_input_randomness);
1079
1080static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1081
1082#ifdef ADD_INTERRUPT_BENCH
1083static unsigned long avg_cycles, avg_deviation;
1084
1085#define AVG_SHIFT 8     /* Exponential average factor k=1/256 */
1086#define FIXED_1_2 (1 << (AVG_SHIFT-1))
1087
1088static void add_interrupt_bench(cycles_t start)
1089{
1090        long delta = random_get_entropy() - start;
1091
1092        /* Use a weighted moving average */
1093        delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1094        avg_cycles += delta;
1095        /* And average deviation */
1096        delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1097        avg_deviation += delta;
1098}
1099#else
1100#define add_interrupt_bench(x)
1101#endif
1102
1103static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1104{
1105        __u32 *ptr = (__u32 *) regs;
1106        unsigned int idx;
1107
1108        if (regs == NULL)
1109                return 0;
1110        idx = READ_ONCE(f->reg_idx);
1111        if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1112                idx = 0;
1113        ptr += idx++;
1114        WRITE_ONCE(f->reg_idx, idx);
1115        return *ptr;
1116}
1117
1118void add_interrupt_randomness(int irq, int irq_flags)
1119{
1120        struct entropy_store    *r;
1121        struct fast_pool        *fast_pool = this_cpu_ptr(&irq_randomness);
1122        struct pt_regs          *regs = get_irq_regs();
1123        unsigned long           now = jiffies;
1124        cycles_t                cycles = random_get_entropy();
1125        __u32                   c_high, j_high;
1126        __u64                   ip;
1127        unsigned long           seed;
1128        int                     credit = 0;
1129
1130        if (cycles == 0)
1131                cycles = get_reg(fast_pool, regs);
1132        c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1133        j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1134        fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1135        fast_pool->pool[1] ^= now ^ c_high;
1136        ip = regs ? instruction_pointer(regs) : _RET_IP_;
1137        fast_pool->pool[2] ^= ip;
1138        fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1139                get_reg(fast_pool, regs);
1140
1141        fast_mix(fast_pool);
1142        add_interrupt_bench(cycles);
1143
1144        if (!crng_ready()) {
1145                if ((fast_pool->count >= 64) &&
1146                    crng_fast_load((char *) fast_pool->pool,
1147                                   sizeof(fast_pool->pool))) {
1148                        fast_pool->count = 0;
1149                        fast_pool->last = now;
1150                }
1151                return;
1152        }
1153
1154        if ((fast_pool->count < 64) &&
1155            !time_after(now, fast_pool->last + HZ))
1156                return;
1157
1158        r = &input_pool;
1159        if (!spin_trylock(&r->lock))
1160                return;
1161
1162        fast_pool->last = now;
1163        __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1164
1165        /*
1166         * If we have architectural seed generator, produce a seed and
1167         * add it to the pool.  For the sake of paranoia don't let the
1168         * architectural seed generator dominate the input from the
1169         * interrupt noise.
1170         */
1171        if (arch_get_random_seed_long(&seed)) {
1172                __mix_pool_bytes(r, &seed, sizeof(seed));
1173                credit = 1;
1174        }
1175        spin_unlock(&r->lock);
1176
1177        fast_pool->count = 0;
1178
1179        /* award one bit for the contents of the fast pool */
1180        credit_entropy_bits(r, credit + 1);
1181}
1182EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1183
1184#ifdef CONFIG_BLOCK
1185void add_disk_randomness(struct gendisk *disk)
1186{
1187        if (!disk || !disk->random)
1188                return;
1189        /* first major is 1, so we get >= 0x200 here */
1190        add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1191        trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1192}
1193EXPORT_SYMBOL_GPL(add_disk_randomness);
1194#endif
1195
1196/*********************************************************************
1197 *
1198 * Entropy extraction routines
1199 *
1200 *********************************************************************/
1201
1202/*
1203 * This utility inline function is responsible for transferring entropy
1204 * from the primary pool to the secondary extraction pool. We make
1205 * sure we pull enough for a 'catastrophic reseed'.
1206 */
1207static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1208static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1209{
1210        if (!r->pull ||
1211            r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1212            r->entropy_count > r->poolinfo->poolfracbits)
1213                return;
1214
1215        _xfer_secondary_pool(r, nbytes);
1216}
1217
1218static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1219{
1220        __u32   tmp[OUTPUT_POOL_WORDS];
1221
1222        int bytes = nbytes;
1223
1224        /* pull at least as much as a wakeup */
1225        bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1226        /* but never more than the buffer size */
1227        bytes = min_t(int, bytes, sizeof(tmp));
1228
1229        trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1230                                  ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1231        bytes = extract_entropy(r->pull, tmp, bytes,
1232                                random_read_wakeup_bits / 8, 0);
1233        mix_pool_bytes(r, tmp, bytes);
1234        credit_entropy_bits(r, bytes*8);
1235}
1236
1237/*
1238 * Used as a workqueue function so that when the input pool is getting
1239 * full, we can "spill over" some entropy to the output pools.  That
1240 * way the output pools can store some of the excess entropy instead
1241 * of letting it go to waste.
1242 */
1243static void push_to_pool(struct work_struct *work)
1244{
1245        struct entropy_store *r = container_of(work, struct entropy_store,
1246                                              push_work);
1247        BUG_ON(!r);
1248        _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1249        trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1250                           r->pull->entropy_count >> ENTROPY_SHIFT);
1251}
1252
1253/*
1254 * This function decides how many bytes to actually take from the
1255 * given pool, and also debits the entropy count accordingly.
1256 */
1257static size_t account(struct entropy_store *r, size_t nbytes, int min,
1258                      int reserved)
1259{
1260        int entropy_count, orig, have_bytes;
1261        size_t ibytes, nfrac;
1262
1263        BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1264
1265        /* Can we pull enough? */
1266retry:
1267        entropy_count = orig = READ_ONCE(r->entropy_count);
1268        ibytes = nbytes;
1269        /* never pull more than available */
1270        have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1271
1272        if ((have_bytes -= reserved) < 0)
1273                have_bytes = 0;
1274        ibytes = min_t(size_t, ibytes, have_bytes);
1275        if (ibytes < min)
1276                ibytes = 0;
1277
1278        if (unlikely(entropy_count < 0)) {
1279                pr_warn("random: negative entropy count: pool %s count %d\n",
1280                        r->name, entropy_count);
1281                WARN_ON(1);
1282                entropy_count = 0;
1283        }
1284        nfrac = ibytes << (ENTROPY_SHIFT + 3);
1285        if ((size_t) entropy_count > nfrac)
1286                entropy_count -= nfrac;
1287        else
1288                entropy_count = 0;
1289
1290        if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1291                goto retry;
1292
1293        trace_debit_entropy(r->name, 8 * ibytes);
1294        if (ibytes &&
1295            (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1296                wake_up_interruptible(&random_write_wait);
1297                kill_fasync(&fasync, SIGIO, POLL_OUT);
1298        }
1299
1300        return ibytes;
1301}
1302
1303/*
1304 * This function does the actual extraction for extract_entropy and
1305 * extract_entropy_user.
1306 *
1307 * Note: we assume that .poolwords is a multiple of 16 words.
1308 */
1309static void extract_buf(struct entropy_store *r, __u8 *out)
1310{
1311        int i;
1312        union {
1313                __u32 w[5];
1314                unsigned long l[LONGS(20)];
1315        } hash;
1316        __u32 workspace[SHA_WORKSPACE_WORDS];
1317        unsigned long flags;
1318
1319        /*
1320         * If we have an architectural hardware random number
1321         * generator, use it for SHA's initial vector
1322         */
1323        sha_init(hash.w);
1324        for (i = 0; i < LONGS(20); i++) {
1325                unsigned long v;
1326                if (!arch_get_random_long(&v))
1327                        break;
1328                hash.l[i] = v;
1329        }
1330
1331        /* Generate a hash across the pool, 16 words (512 bits) at a time */
1332        spin_lock_irqsave(&r->lock, flags);
1333        for (i = 0; i < r->poolinfo->poolwords; i += 16)
1334                sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1335
1336        /*
1337         * We mix the hash back into the pool to prevent backtracking
1338         * attacks (where the attacker knows the state of the pool
1339         * plus the current outputs, and attempts to find previous
1340         * ouputs), unless the hash function can be inverted. By
1341         * mixing at least a SHA1 worth of hash data back, we make
1342         * brute-forcing the feedback as hard as brute-forcing the
1343         * hash.
1344         */
1345        __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1346        spin_unlock_irqrestore(&r->lock, flags);
1347
1348        memzero_explicit(workspace, sizeof(workspace));
1349
1350        /*
1351         * In case the hash function has some recognizable output
1352         * pattern, we fold it in half. Thus, we always feed back
1353         * twice as much data as we output.
1354         */
1355        hash.w[0] ^= hash.w[3];
1356        hash.w[1] ^= hash.w[4];
1357        hash.w[2] ^= rol32(hash.w[2], 16);
1358
1359        memcpy(out, &hash, EXTRACT_SIZE);
1360        memzero_explicit(&hash, sizeof(hash));
1361}
1362
1363static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1364                                size_t nbytes, int fips)
1365{
1366        ssize_t ret = 0, i;
1367        __u8 tmp[EXTRACT_SIZE];
1368        unsigned long flags;
1369
1370        while (nbytes) {
1371                extract_buf(r, tmp);
1372
1373                if (fips) {
1374                        spin_lock_irqsave(&r->lock, flags);
1375                        if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1376                                panic("Hardware RNG duplicated output!\n");
1377                        memcpy(r->last_data, tmp, EXTRACT_SIZE);
1378                        spin_unlock_irqrestore(&r->lock, flags);
1379                }
1380                i = min_t(int, nbytes, EXTRACT_SIZE);
1381                memcpy(buf, tmp, i);
1382                nbytes -= i;
1383                buf += i;
1384                ret += i;
1385        }
1386
1387        /* Wipe data just returned from memory */
1388        memzero_explicit(tmp, sizeof(tmp));
1389
1390        return ret;
1391}
1392
1393/*
1394 * This function extracts randomness from the "entropy pool", and
1395 * returns it in a buffer.
1396 *
1397 * The min parameter specifies the minimum amount we can pull before
1398 * failing to avoid races that defeat catastrophic reseeding while the
1399 * reserved parameter indicates how much entropy we must leave in the
1400 * pool after each pull to avoid starving other readers.
1401 */
1402static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1403                                 size_t nbytes, int min, int reserved)
1404{
1405        __u8 tmp[EXTRACT_SIZE];
1406        unsigned long flags;
1407
1408        /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1409        if (fips_enabled) {
1410                spin_lock_irqsave(&r->lock, flags);
1411                if (!r->last_data_init) {
1412                        r->last_data_init = 1;
1413                        spin_unlock_irqrestore(&r->lock, flags);
1414                        trace_extract_entropy(r->name, EXTRACT_SIZE,
1415                                              ENTROPY_BITS(r), _RET_IP_);
1416                        xfer_secondary_pool(r, EXTRACT_SIZE);
1417                        extract_buf(r, tmp);
1418                        spin_lock_irqsave(&r->lock, flags);
1419                        memcpy(r->last_data, tmp, EXTRACT_SIZE);
1420                }
1421                spin_unlock_irqrestore(&r->lock, flags);
1422        }
1423
1424        trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1425        xfer_secondary_pool(r, nbytes);
1426        nbytes = account(r, nbytes, min, reserved);
1427
1428        return _extract_entropy(r, buf, nbytes, fips_enabled);
1429}
1430
1431/*
1432 * This function extracts randomness from the "entropy pool", and
1433 * returns it in a userspace buffer.
1434 */
1435static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1436                                    size_t nbytes)
1437{
1438        ssize_t ret = 0, i;
1439        __u8 tmp[EXTRACT_SIZE];
1440        int large_request = (nbytes > 256);
1441
1442        trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1443        xfer_secondary_pool(r, nbytes);
1444        nbytes = account(r, nbytes, 0, 0);
1445
1446        while (nbytes) {
1447                if (large_request && need_resched()) {
1448                        if (signal_pending(current)) {
1449                                if (ret == 0)
1450                                        ret = -ERESTARTSYS;
1451                                break;
1452                        }
1453                        schedule();
1454                }
1455
1456                extract_buf(r, tmp);
1457                i = min_t(int, nbytes, EXTRACT_SIZE);
1458                if (copy_to_user(buf, tmp, i)) {
1459                        ret = -EFAULT;
1460                        break;
1461                }
1462
1463                nbytes -= i;
1464                buf += i;
1465                ret += i;
1466        }
1467
1468        /* Wipe data just returned from memory */
1469        memzero_explicit(tmp, sizeof(tmp));
1470
1471        return ret;
1472}
1473
1474#define warn_unseeded_randomness(previous) \
1475        _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1476
1477static void _warn_unseeded_randomness(const char *func_name, void *caller,
1478                                      void **previous)
1479{
1480#ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1481        const bool print_once = false;
1482#else
1483        static bool print_once __read_mostly;
1484#endif
1485
1486        if (print_once ||
1487            crng_ready() ||
1488            (previous && (caller == READ_ONCE(*previous))))
1489                return;
1490        WRITE_ONCE(*previous, caller);
1491#ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1492        print_once = true;
1493#endif
1494        pr_notice("random: %s called from %pS with crng_init=%d\n",
1495                  func_name, caller, crng_init);
1496}
1497
1498/*
1499 * This function is the exported kernel interface.  It returns some
1500 * number of good random numbers, suitable for key generation, seeding
1501 * TCP sequence numbers, etc.  It does not rely on the hardware random
1502 * number generator.  For random bytes direct from the hardware RNG
1503 * (when available), use get_random_bytes_arch(). In order to ensure
1504 * that the randomness provided by this function is okay, the function
1505 * wait_for_random_bytes() should be called and return 0 at least once
1506 * at any point prior.
1507 */
1508static void _get_random_bytes(void *buf, int nbytes)
1509{
1510        __u8 tmp[CHACHA20_BLOCK_SIZE];
1511
1512        trace_get_random_bytes(nbytes, _RET_IP_);
1513
1514        while (nbytes >= CHACHA20_BLOCK_SIZE) {
1515                extract_crng(buf);
1516                buf += CHACHA20_BLOCK_SIZE;
1517                nbytes -= CHACHA20_BLOCK_SIZE;
1518        }
1519
1520        if (nbytes > 0) {
1521                extract_crng(tmp);
1522                memcpy(buf, tmp, nbytes);
1523                crng_backtrack_protect(tmp, nbytes);
1524        } else
1525                crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1526        memzero_explicit(tmp, sizeof(tmp));
1527}
1528
1529void get_random_bytes(void *buf, int nbytes)
1530{
1531        static void *previous;
1532
1533        warn_unseeded_randomness(&previous);
1534        _get_random_bytes(buf, nbytes);
1535}
1536EXPORT_SYMBOL(get_random_bytes);
1537
1538/*
1539 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1540 * cryptographically secure random numbers. This applies to: the /dev/urandom
1541 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1542 * family of functions. Using any of these functions without first calling
1543 * this function forfeits the guarantee of security.
1544 *
1545 * Returns: 0 if the urandom pool has been seeded.
1546 *          -ERESTARTSYS if the function was interrupted by a signal.
1547 */
1548int wait_for_random_bytes(void)
1549{
1550        if (likely(crng_ready()))
1551                return 0;
1552        return wait_event_interruptible(crng_init_wait, crng_ready());
1553}
1554EXPORT_SYMBOL(wait_for_random_bytes);
1555
1556/*
1557 * Add a callback function that will be invoked when the nonblocking
1558 * pool is initialised.
1559 *
1560 * returns: 0 if callback is successfully added
1561 *          -EALREADY if pool is already initialised (callback not called)
1562 *          -ENOENT if module for callback is not alive
1563 */
1564int add_random_ready_callback(struct random_ready_callback *rdy)
1565{
1566        struct module *owner;
1567        unsigned long flags;
1568        int err = -EALREADY;
1569
1570        if (crng_ready())
1571                return err;
1572
1573        owner = rdy->owner;
1574        if (!try_module_get(owner))
1575                return -ENOENT;
1576
1577        spin_lock_irqsave(&random_ready_list_lock, flags);
1578        if (crng_ready())
1579                goto out;
1580
1581        owner = NULL;
1582
1583        list_add(&rdy->list, &random_ready_list);
1584        err = 0;
1585
1586out:
1587        spin_unlock_irqrestore(&random_ready_list_lock, flags);
1588
1589        module_put(owner);
1590
1591        return err;
1592}
1593EXPORT_SYMBOL(add_random_ready_callback);
1594
1595/*
1596 * Delete a previously registered readiness callback function.
1597 */
1598void del_random_ready_callback(struct random_ready_callback *rdy)
1599{
1600        unsigned long flags;
1601        struct module *owner = NULL;
1602
1603        spin_lock_irqsave(&random_ready_list_lock, flags);
1604        if (!list_empty(&rdy->list)) {
1605                list_del_init(&rdy->list);
1606                owner = rdy->owner;
1607        }
1608        spin_unlock_irqrestore(&random_ready_list_lock, flags);
1609
1610        module_put(owner);
1611}
1612EXPORT_SYMBOL(del_random_ready_callback);
1613
1614/*
1615 * This function will use the architecture-specific hardware random
1616 * number generator if it is available.  The arch-specific hw RNG will
1617 * almost certainly be faster than what we can do in software, but it
1618 * is impossible to verify that it is implemented securely (as
1619 * opposed, to, say, the AES encryption of a sequence number using a
1620 * key known by the NSA).  So it's useful if we need the speed, but
1621 * only if we're willing to trust the hardware manufacturer not to
1622 * have put in a back door.
1623 */
1624void get_random_bytes_arch(void *buf, int nbytes)
1625{
1626        char *p = buf;
1627
1628        trace_get_random_bytes_arch(nbytes, _RET_IP_);
1629        while (nbytes) {
1630                unsigned long v;
1631                int chunk = min(nbytes, (int)sizeof(unsigned long));
1632
1633                if (!arch_get_random_long(&v))
1634                        break;
1635                
1636                memcpy(p, &v, chunk);
1637                p += chunk;
1638                nbytes -= chunk;
1639        }
1640
1641        if (nbytes)
1642                get_random_bytes(p, nbytes);
1643}
1644EXPORT_SYMBOL(get_random_bytes_arch);
1645
1646
1647/*
1648 * init_std_data - initialize pool with system data
1649 *
1650 * @r: pool to initialize
1651 *
1652 * This function clears the pool's entropy count and mixes some system
1653 * data into the pool to prepare it for use. The pool is not cleared
1654 * as that can only decrease the entropy in the pool.
1655 */
1656static void init_std_data(struct entropy_store *r)
1657{
1658        int i;
1659        ktime_t now = ktime_get_real();
1660        unsigned long rv;
1661
1662        r->last_pulled = jiffies;
1663        mix_pool_bytes(r, &now, sizeof(now));
1664        for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1665                if (!arch_get_random_seed_long(&rv) &&
1666                    !arch_get_random_long(&rv))
1667                        rv = random_get_entropy();
1668                mix_pool_bytes(r, &rv, sizeof(rv));
1669        }
1670        mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1671}
1672
1673/*
1674 * Note that setup_arch() may call add_device_randomness()
1675 * long before we get here. This allows seeding of the pools
1676 * with some platform dependent data very early in the boot
1677 * process. But it limits our options here. We must use
1678 * statically allocated structures that already have all
1679 * initializations complete at compile time. We should also
1680 * take care not to overwrite the precious per platform data
1681 * we were given.
1682 */
1683static int rand_initialize(void)
1684{
1685#ifdef CONFIG_NUMA
1686        int i;
1687        struct crng_state *crng;
1688        struct crng_state **pool;
1689#endif
1690
1691        init_std_data(&input_pool);
1692        init_std_data(&blocking_pool);
1693        crng_initialize(&primary_crng);
1694
1695#ifdef CONFIG_NUMA
1696        pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
1697        for_each_online_node(i) {
1698                crng = kmalloc_node(sizeof(struct crng_state),
1699                                    GFP_KERNEL | __GFP_NOFAIL, i);
1700                spin_lock_init(&crng->lock);
1701                crng_initialize(crng);
1702                pool[i] = crng;
1703        }
1704        mb();
1705        crng_node_pool = pool;
1706#endif
1707        return 0;
1708}
1709early_initcall(rand_initialize);
1710
1711#ifdef CONFIG_BLOCK
1712void rand_initialize_disk(struct gendisk *disk)
1713{
1714        struct timer_rand_state *state;
1715
1716        /*
1717         * If kzalloc returns null, we just won't use that entropy
1718         * source.
1719         */
1720        state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1721        if (state) {
1722                state->last_time = INITIAL_JIFFIES;
1723                disk->random = state;
1724        }
1725}
1726#endif
1727
1728static ssize_t
1729_random_read(int nonblock, char __user *buf, size_t nbytes)
1730{
1731        ssize_t n;
1732
1733        if (nbytes == 0)
1734                return 0;
1735
1736        nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1737        while (1) {
1738                n = extract_entropy_user(&blocking_pool, buf, nbytes);
1739                if (n < 0)
1740                        return n;
1741                trace_random_read(n*8, (nbytes-n)*8,
1742                                  ENTROPY_BITS(&blocking_pool),
1743                                  ENTROPY_BITS(&input_pool));
1744                if (n > 0)
1745                        return n;
1746
1747                /* Pool is (near) empty.  Maybe wait and retry. */
1748                if (nonblock)
1749                        return -EAGAIN;
1750
1751                wait_event_interruptible(random_read_wait,
1752                        ENTROPY_BITS(&input_pool) >=
1753                        random_read_wakeup_bits);
1754                if (signal_pending(current))
1755                        return -ERESTARTSYS;
1756        }
1757}
1758
1759static ssize_t
1760random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1761{
1762        return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1763}
1764
1765static ssize_t
1766urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1767{
1768        unsigned long flags;
1769        static int maxwarn = 10;
1770        int ret;
1771
1772        if (!crng_ready() && maxwarn > 0) {
1773                maxwarn--;
1774                printk(KERN_NOTICE "random: %s: uninitialized urandom read "
1775                       "(%zd bytes read)\n",
1776                       current->comm, nbytes);
1777                spin_lock_irqsave(&primary_crng.lock, flags);
1778                crng_init_cnt = 0;
1779                spin_unlock_irqrestore(&primary_crng.lock, flags);
1780        }
1781        nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1782        ret = extract_crng_user(buf, nbytes);
1783        trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1784        return ret;
1785}
1786
1787static unsigned int
1788random_poll(struct file *file, poll_table * wait)
1789{
1790        unsigned int mask;
1791
1792        poll_wait(file, &random_read_wait, wait);
1793        poll_wait(file, &random_write_wait, wait);
1794        mask = 0;
1795        if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1796                mask |= POLLIN | POLLRDNORM;
1797        if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1798                mask |= POLLOUT | POLLWRNORM;
1799        return mask;
1800}
1801
1802static int
1803write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1804{
1805        size_t bytes;
1806        __u32 buf[16];
1807        const char __user *p = buffer;
1808
1809        while (count > 0) {
1810                bytes = min(count, sizeof(buf));
1811                if (copy_from_user(&buf, p, bytes))
1812                        return -EFAULT;
1813
1814                count -= bytes;
1815                p += bytes;
1816
1817                mix_pool_bytes(r, buf, bytes);
1818                cond_resched();
1819        }
1820
1821        return 0;
1822}
1823
1824static ssize_t random_write(struct file *file, const char __user *buffer,
1825                            size_t count, loff_t *ppos)
1826{
1827        size_t ret;
1828
1829        ret = write_pool(&input_pool, buffer, count);
1830        if (ret)
1831                return ret;
1832
1833        return (ssize_t)count;
1834}
1835
1836static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1837{
1838        int size, ent_count;
1839        int __user *p = (int __user *)arg;
1840        int retval;
1841
1842        switch (cmd) {
1843        case RNDGETENTCNT:
1844                /* inherently racy, no point locking */
1845                ent_count = ENTROPY_BITS(&input_pool);
1846                if (put_user(ent_count, p))
1847                        return -EFAULT;
1848                return 0;
1849        case RNDADDTOENTCNT:
1850                if (!capable(CAP_SYS_ADMIN))
1851                        return -EPERM;
1852                if (get_user(ent_count, p))
1853                        return -EFAULT;
1854                return credit_entropy_bits_safe(&input_pool, ent_count);
1855        case RNDADDENTROPY:
1856                if (!capable(CAP_SYS_ADMIN))
1857                        return -EPERM;
1858                if (get_user(ent_count, p++))
1859                        return -EFAULT;
1860                if (ent_count < 0)
1861                        return -EINVAL;
1862                if (get_user(size, p++))
1863                        return -EFAULT;
1864                retval = write_pool(&input_pool, (const char __user *)p,
1865                                    size);
1866                if (retval < 0)
1867                        return retval;
1868                return credit_entropy_bits_safe(&input_pool, ent_count);
1869        case RNDZAPENTCNT:
1870        case RNDCLEARPOOL:
1871                /*
1872                 * Clear the entropy pool counters. We no longer clear
1873                 * the entropy pool, as that's silly.
1874                 */
1875                if (!capable(CAP_SYS_ADMIN))
1876                        return -EPERM;
1877                input_pool.entropy_count = 0;
1878                blocking_pool.entropy_count = 0;
1879                return 0;
1880        default:
1881                return -EINVAL;
1882        }
1883}
1884
1885static int random_fasync(int fd, struct file *filp, int on)
1886{
1887        return fasync_helper(fd, filp, on, &fasync);
1888}
1889
1890const struct file_operations random_fops = {
1891        .read  = random_read,
1892        .write = random_write,
1893        .poll  = random_poll,
1894        .unlocked_ioctl = random_ioctl,
1895        .fasync = random_fasync,
1896        .llseek = noop_llseek,
1897};
1898
1899const struct file_operations urandom_fops = {
1900        .read  = urandom_read,
1901        .write = random_write,
1902        .unlocked_ioctl = random_ioctl,
1903        .fasync = random_fasync,
1904        .llseek = noop_llseek,
1905};
1906
1907SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
1908                unsigned int, flags)
1909{
1910        int ret;
1911
1912        if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
1913                return -EINVAL;
1914
1915        if (count > INT_MAX)
1916                count = INT_MAX;
1917
1918        if (flags & GRND_RANDOM)
1919                return _random_read(flags & GRND_NONBLOCK, buf, count);
1920
1921        if (!crng_ready()) {
1922                if (flags & GRND_NONBLOCK)
1923                        return -EAGAIN;
1924                ret = wait_for_random_bytes();
1925                if (unlikely(ret))
1926                        return ret;
1927        }
1928        return urandom_read(NULL, buf, count, NULL);
1929}
1930
1931/********************************************************************
1932 *
1933 * Sysctl interface
1934 *
1935 ********************************************************************/
1936
1937#ifdef CONFIG_SYSCTL
1938
1939#include <linux/sysctl.h>
1940
1941static int min_read_thresh = 8, min_write_thresh;
1942static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
1943static int max_write_thresh = INPUT_POOL_WORDS * 32;
1944static int random_min_urandom_seed = 60;
1945static char sysctl_bootid[16];
1946
1947/*
1948 * This function is used to return both the bootid UUID, and random
1949 * UUID.  The difference is in whether table->data is NULL; if it is,
1950 * then a new UUID is generated and returned to the user.
1951 *
1952 * If the user accesses this via the proc interface, the UUID will be
1953 * returned as an ASCII string in the standard UUID format; if via the
1954 * sysctl system call, as 16 bytes of binary data.
1955 */
1956static int proc_do_uuid(struct ctl_table *table, int write,
1957                        void __user *buffer, size_t *lenp, loff_t *ppos)
1958{
1959        struct ctl_table fake_table;
1960        unsigned char buf[64], tmp_uuid[16], *uuid;
1961
1962        uuid = table->data;
1963        if (!uuid) {
1964                uuid = tmp_uuid;
1965                generate_random_uuid(uuid);
1966        } else {
1967                static DEFINE_SPINLOCK(bootid_spinlock);
1968
1969                spin_lock(&bootid_spinlock);
1970                if (!uuid[8])
1971                        generate_random_uuid(uuid);
1972                spin_unlock(&bootid_spinlock);
1973        }
1974
1975        sprintf(buf, "%pU", uuid);
1976
1977        fake_table.data = buf;
1978        fake_table.maxlen = sizeof(buf);
1979
1980        return proc_dostring(&fake_table, write, buffer, lenp, ppos);
1981}
1982
1983/*
1984 * Return entropy available scaled to integral bits
1985 */
1986static int proc_do_entropy(struct ctl_table *table, int write,
1987                           void __user *buffer, size_t *lenp, loff_t *ppos)
1988{
1989        struct ctl_table fake_table;
1990        int entropy_count;
1991
1992        entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
1993
1994        fake_table.data = &entropy_count;
1995        fake_table.maxlen = sizeof(entropy_count);
1996
1997        return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
1998}
1999
2000static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2001extern struct ctl_table random_table[];
2002struct ctl_table random_table[] = {
2003        {
2004                .procname       = "poolsize",
2005                .data           = &sysctl_poolsize,
2006                .maxlen         = sizeof(int),
2007                .mode           = 0444,
2008                .proc_handler   = proc_dointvec,
2009        },
2010        {
2011                .procname       = "entropy_avail",
2012                .maxlen         = sizeof(int),
2013                .mode           = 0444,
2014                .proc_handler   = proc_do_entropy,
2015                .data           = &input_pool.entropy_count,
2016        },
2017        {
2018                .procname       = "read_wakeup_threshold",
2019                .data           = &random_read_wakeup_bits,
2020                .maxlen         = sizeof(int),
2021                .mode           = 0644,
2022                .proc_handler   = proc_dointvec_minmax,
2023                .extra1         = &min_read_thresh,
2024                .extra2         = &max_read_thresh,
2025        },
2026        {
2027                .procname       = "write_wakeup_threshold",
2028                .data           = &random_write_wakeup_bits,
2029                .maxlen         = sizeof(int),
2030                .mode           = 0644,
2031                .proc_handler   = proc_dointvec_minmax,
2032                .extra1         = &min_write_thresh,
2033                .extra2         = &max_write_thresh,
2034        },
2035        {
2036                .procname       = "urandom_min_reseed_secs",
2037                .data           = &random_min_urandom_seed,
2038                .maxlen         = sizeof(int),
2039                .mode           = 0644,
2040                .proc_handler   = proc_dointvec,
2041        },
2042        {
2043                .procname       = "boot_id",
2044                .data           = &sysctl_bootid,
2045                .maxlen         = 16,
2046                .mode           = 0444,
2047                .proc_handler   = proc_do_uuid,
2048        },
2049        {
2050                .procname       = "uuid",
2051                .maxlen         = 16,
2052                .mode           = 0444,
2053                .proc_handler   = proc_do_uuid,
2054        },
2055#ifdef ADD_INTERRUPT_BENCH
2056        {
2057                .procname       = "add_interrupt_avg_cycles",
2058                .data           = &avg_cycles,
2059                .maxlen         = sizeof(avg_cycles),
2060                .mode           = 0444,
2061                .proc_handler   = proc_doulongvec_minmax,
2062        },
2063        {
2064                .procname       = "add_interrupt_avg_deviation",
2065                .data           = &avg_deviation,
2066                .maxlen         = sizeof(avg_deviation),
2067                .mode           = 0444,
2068                .proc_handler   = proc_doulongvec_minmax,
2069        },
2070#endif
2071        { }
2072};
2073#endif  /* CONFIG_SYSCTL */
2074
2075struct batched_entropy {
2076        union {
2077                u64 entropy_u64[CHACHA20_BLOCK_SIZE / sizeof(u64)];
2078                u32 entropy_u32[CHACHA20_BLOCK_SIZE / sizeof(u32)];
2079        };
2080        unsigned int position;
2081};
2082static rwlock_t batched_entropy_reset_lock = __RW_LOCK_UNLOCKED(batched_entropy_reset_lock);
2083
2084/*
2085 * Get a random word for internal kernel use only. The quality of the random
2086 * number is either as good as RDRAND or as good as /dev/urandom, with the
2087 * goal of being quite fast and not depleting entropy. In order to ensure
2088 * that the randomness provided by this function is okay, the function
2089 * wait_for_random_bytes() should be called and return 0 at least once
2090 * at any point prior.
2091 */
2092static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64);
2093u64 get_random_u64(void)
2094{
2095        u64 ret;
2096        bool use_lock;
2097        unsigned long flags = 0;
2098        struct batched_entropy *batch;
2099        static void *previous;
2100
2101#if BITS_PER_LONG == 64
2102        if (arch_get_random_long((unsigned long *)&ret))
2103                return ret;
2104#else
2105        if (arch_get_random_long((unsigned long *)&ret) &&
2106            arch_get_random_long((unsigned long *)&ret + 1))
2107            return ret;
2108#endif
2109
2110        warn_unseeded_randomness(&previous);
2111
2112        use_lock = READ_ONCE(crng_init) < 2;
2113        batch = &get_cpu_var(batched_entropy_u64);
2114        if (use_lock)
2115                read_lock_irqsave(&batched_entropy_reset_lock, flags);
2116        if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2117                extract_crng((u8 *)batch->entropy_u64);
2118                batch->position = 0;
2119        }
2120        ret = batch->entropy_u64[batch->position++];
2121        if (use_lock)
2122                read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2123        put_cpu_var(batched_entropy_u64);
2124        return ret;
2125}
2126EXPORT_SYMBOL(get_random_u64);
2127
2128static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32);
2129u32 get_random_u32(void)
2130{
2131        u32 ret;
2132        bool use_lock;
2133        unsigned long flags = 0;
2134        struct batched_entropy *batch;
2135        static void *previous;
2136
2137        if (arch_get_random_int(&ret))
2138                return ret;
2139
2140        warn_unseeded_randomness(&previous);
2141
2142        use_lock = READ_ONCE(crng_init) < 2;
2143        batch = &get_cpu_var(batched_entropy_u32);
2144        if (use_lock)
2145                read_lock_irqsave(&batched_entropy_reset_lock, flags);
2146        if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2147                extract_crng((u8 *)batch->entropy_u32);
2148                batch->position = 0;
2149        }
2150        ret = batch->entropy_u32[batch->position++];
2151        if (use_lock)
2152                read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2153        put_cpu_var(batched_entropy_u32);
2154        return ret;
2155}
2156EXPORT_SYMBOL(get_random_u32);
2157
2158/* It's important to invalidate all potential batched entropy that might
2159 * be stored before the crng is initialized, which we can do lazily by
2160 * simply resetting the counter to zero so that it's re-extracted on the
2161 * next usage. */
2162static void invalidate_batched_entropy(void)
2163{
2164        int cpu;
2165        unsigned long flags;
2166
2167        write_lock_irqsave(&batched_entropy_reset_lock, flags);
2168        for_each_possible_cpu (cpu) {
2169                per_cpu_ptr(&batched_entropy_u32, cpu)->position = 0;
2170                per_cpu_ptr(&batched_entropy_u64, cpu)->position = 0;
2171        }
2172        write_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2173}
2174
2175/**
2176 * randomize_page - Generate a random, page aligned address
2177 * @start:      The smallest acceptable address the caller will take.
2178 * @range:      The size of the area, starting at @start, within which the
2179 *              random address must fall.
2180 *
2181 * If @start + @range would overflow, @range is capped.
2182 *
2183 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2184 * @start was already page aligned.  We now align it regardless.
2185 *
2186 * Return: A page aligned address within [start, start + range).  On error,
2187 * @start is returned.
2188 */
2189unsigned long
2190randomize_page(unsigned long start, unsigned long range)
2191{
2192        if (!PAGE_ALIGNED(start)) {
2193                range -= PAGE_ALIGN(start) - start;
2194                start = PAGE_ALIGN(start);
2195        }
2196
2197        if (start > ULONG_MAX - range)
2198                range = ULONG_MAX - start;
2199
2200        range >>= PAGE_SHIFT;
2201
2202        if (range == 0)
2203                return start;
2204
2205        return start + (get_random_long() % range << PAGE_SHIFT);
2206}
2207
2208/* Interface for in-kernel drivers of true hardware RNGs.
2209 * Those devices may produce endless random bits and will be throttled
2210 * when our pool is full.
2211 */
2212void add_hwgenerator_randomness(const char *buffer, size_t count,
2213                                size_t entropy)
2214{
2215        struct entropy_store *poolp = &input_pool;
2216
2217        if (!crng_ready()) {
2218                crng_fast_load(buffer, count);
2219                return;
2220        }
2221
2222        /* Suspend writing if we're above the trickle threshold.
2223         * We'll be woken up again once below random_write_wakeup_thresh,
2224         * or when the calling thread is about to terminate.
2225         */
2226        wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2227                        ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2228        mix_pool_bytes(poolp, buffer, count);
2229        credit_entropy_bits(poolp, entropy);
2230}
2231EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2232
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.