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/ratelimit.h>
 265#include <linux/syscalls.h>
 266#include <linux/completion.h>
 267#include <linux/uuid.h>
 268#include <crypto/chacha20.h>
 269
 270#include <asm/processor.h>
 271#include <linux/uaccess.h>
 272#include <asm/irq.h>
 273#include <asm/irq_regs.h>
 274#include <asm/io.h>
 275
 276#define CREATE_TRACE_POINTS
 277#include <trace/events/random.h>
 278
 279/* #define ADD_INTERRUPT_BENCH */
 280
 281/*
 282 * Configuration information
 283 */
 284#define INPUT_POOL_SHIFT        12
 285#define INPUT_POOL_WORDS        (1 << (INPUT_POOL_SHIFT-5))
 286#define OUTPUT_POOL_SHIFT       10
 287#define OUTPUT_POOL_WORDS       (1 << (OUTPUT_POOL_SHIFT-5))
 288#define SEC_XFER_SIZE           512
 289#define EXTRACT_SIZE            10
 290
 291
 292#define LONGS(x) (((x) + sizeof(unsigned long) - 1)/sizeof(unsigned long))
 293
 294/*
 295 * To allow fractional bits to be tracked, the entropy_count field is
 296 * denominated in units of 1/8th bits.
 297 *
 298 * 2*(ENTROPY_SHIFT + log2(poolbits)) must <= 31, or the multiply in
 299 * credit_entropy_bits() needs to be 64 bits wide.
 300 */
 301#define ENTROPY_SHIFT 3
 302#define ENTROPY_BITS(r) ((r)->entropy_count >> ENTROPY_SHIFT)
 303
 304/*
 305 * The minimum number of bits of entropy before we wake up a read on
 306 * /dev/random.  Should be enough to do a significant reseed.
 307 */
 308static int random_read_wakeup_bits = 64;
 309
 310/*
 311 * If the entropy count falls under this number of bits, then we
 312 * should wake up processes which are selecting or polling on write
 313 * access to /dev/random.
 314 */
 315static int random_write_wakeup_bits = 28 * OUTPUT_POOL_WORDS;
 316
 317/*
 318 * Originally, we used a primitive polynomial of degree .poolwords
 319 * over GF(2).  The taps for various sizes are defined below.  They
 320 * were chosen to be evenly spaced except for the last tap, which is 1
 321 * to get the twisting happening as fast as possible.
 322 *
 323 * For the purposes of better mixing, we use the CRC-32 polynomial as
 324 * well to make a (modified) twisted Generalized Feedback Shift
 325 * Register.  (See M. Matsumoto & Y. Kurita, 1992.  Twisted GFSR
 326 * generators.  ACM Transactions on Modeling and Computer Simulation
 327 * 2(3):179-194.  Also see M. Matsumoto & Y. Kurita, 1994.  Twisted
 328 * GFSR generators II.  ACM Transactions on Modeling and Computer
 329 * Simulation 4:254-266)
 330 *
 331 * Thanks to Colin Plumb for suggesting this.
 332 *
 333 * The mixing operation is much less sensitive than the output hash,
 334 * where we use SHA-1.  All that we want of mixing operation is that
 335 * it be a good non-cryptographic hash; i.e. it not produce collisions
 336 * when fed "random" data of the sort we expect to see.  As long as
 337 * the pool state differs for different inputs, we have preserved the
 338 * input entropy and done a good job.  The fact that an intelligent
 339 * attacker can construct inputs that will produce controlled
 340 * alterations to the pool's state is not important because we don't
 341 * consider such inputs to contribute any randomness.  The only
 342 * property we need with respect to them is that the attacker can't
 343 * increase his/her knowledge of the pool's state.  Since all
 344 * additions are reversible (knowing the final state and the input,
 345 * you can reconstruct the initial state), if an attacker has any
 346 * uncertainty about the initial state, he/she can only shuffle that
 347 * uncertainty about, but never cause any collisions (which would
 348 * decrease the uncertainty).
 349 *
 350 * Our mixing functions were analyzed by Lacharme, Roeck, Strubel, and
 351 * Videau in their paper, "The Linux Pseudorandom Number Generator
 352 * Revisited" (see: http://eprint.iacr.org/2012/251.pdf).  In their
 353 * paper, they point out that we are not using a true Twisted GFSR,
 354 * since Matsumoto & Kurita used a trinomial feedback polynomial (that
 355 * is, with only three taps, instead of the six that we are using).
 356 * As a result, the resulting polynomial is neither primitive nor
 357 * irreducible, and hence does not have a maximal period over
 358 * GF(2**32).  They suggest a slight change to the generator
 359 * polynomial which improves the resulting TGFSR polynomial to be
 360 * irreducible, which we have made here.
 361 */
 362static struct poolinfo {
 363        int poolbitshift, poolwords, poolbytes, poolbits, poolfracbits;
 364#define S(x) ilog2(x)+5, (x), (x)*4, (x)*32, (x) << (ENTROPY_SHIFT+5)
 365        int tap1, tap2, tap3, tap4, tap5;
 366} poolinfo_table[] = {
 367        /* was: x^128 + x^103 + x^76 + x^51 +x^25 + x + 1 */
 368        /* x^128 + x^104 + x^76 + x^51 +x^25 + x + 1 */
 369        { S(128),       104,    76,     51,     25,     1 },
 370        /* was: x^32 + x^26 + x^20 + x^14 + x^7 + x + 1 */
 371        /* x^32 + x^26 + x^19 + x^14 + x^7 + x + 1 */
 372        { S(32),        26,     19,     14,     7,      1 },
 373#if 0
 374        /* x^2048 + x^1638 + x^1231 + x^819 + x^411 + x + 1  -- 115 */
 375        { S(2048),      1638,   1231,   819,    411,    1 },
 376
 377        /* x^1024 + x^817 + x^615 + x^412 + x^204 + x + 1 -- 290 */
 378        { S(1024),      817,    615,    412,    204,    1 },
 379
 380        /* x^1024 + x^819 + x^616 + x^410 + x^207 + x^2 + 1 -- 115 */
 381        { S(1024),      819,    616,    410,    207,    2 },
 382
 383        /* x^512 + x^411 + x^308 + x^208 + x^104 + x + 1 -- 225 */
 384        { S(512),       411,    308,    208,    104,    1 },
 385
 386        /* x^512 + x^409 + x^307 + x^206 + x^102 + x^2 + 1 -- 95 */
 387        { S(512),       409,    307,    206,    102,    2 },
 388        /* x^512 + x^409 + x^309 + x^205 + x^103 + x^2 + 1 -- 95 */
 389        { S(512),       409,    309,    205,    103,    2 },
 390
 391        /* x^256 + x^205 + x^155 + x^101 + x^52 + x + 1 -- 125 */
 392        { S(256),       205,    155,    101,    52,     1 },
 393
 394        /* x^128 + x^103 + x^78 + x^51 + x^27 + x^2 + 1 -- 70 */
 395        { S(128),       103,    78,     51,     27,     2 },
 396
 397        /* x^64 + x^52 + x^39 + x^26 + x^14 + x + 1 -- 15 */
 398        { S(64),        52,     39,     26,     14,     1 },
 399#endif
 400};
 401
 402/*
 403 * Static global variables
 404 */
 405static DECLARE_WAIT_QUEUE_HEAD(random_read_wait);
 406static DECLARE_WAIT_QUEUE_HEAD(random_write_wait);
 407static struct fasync_struct *fasync;
 408
 409static DEFINE_SPINLOCK(random_ready_list_lock);
 410static LIST_HEAD(random_ready_list);
 411
 412struct crng_state {
 413        __u32           state[16];
 414        unsigned long   init_time;
 415        spinlock_t      lock;
 416};
 417
 418struct crng_state primary_crng = {
 419        .lock = __SPIN_LOCK_UNLOCKED(primary_crng.lock),
 420};
 421
 422/*
 423 * crng_init =  0 --> Uninitialized
 424 *              1 --> Initialized
 425 *              2 --> Initialized from input_pool
 426 *
 427 * crng_init is protected by primary_crng->lock, and only increases
 428 * its value (from 0->1->2).
 429 */
 430static int crng_init = 0;
 431#define crng_ready() (likely(crng_init > 1))
 432static int crng_init_cnt = 0;
 433static unsigned long crng_global_init_time = 0;
 434#define CRNG_INIT_CNT_THRESH (2*CHACHA20_KEY_SIZE)
 435static void _extract_crng(struct crng_state *crng,
 436                          __u32 out[CHACHA20_BLOCK_WORDS]);
 437static void _crng_backtrack_protect(struct crng_state *crng,
 438                                    __u32 tmp[CHACHA20_BLOCK_WORDS], int used);
 439static void process_random_ready_list(void);
 440static void _get_random_bytes(void *buf, int nbytes);
 441
 442static struct ratelimit_state unseeded_warning =
 443        RATELIMIT_STATE_INIT("warn_unseeded_randomness", HZ, 3);
 444static struct ratelimit_state urandom_warning =
 445        RATELIMIT_STATE_INIT("warn_urandom_randomness", HZ, 3);
 446
 447static int ratelimit_disable __read_mostly;
 448
 449module_param_named(ratelimit_disable, ratelimit_disable, int, 0644);
 450MODULE_PARM_DESC(ratelimit_disable, "Disable random ratelimit suppression");
 451
 452/**********************************************************************
 453 *
 454 * OS independent entropy store.   Here are the functions which handle
 455 * storing entropy in an entropy pool.
 456 *
 457 **********************************************************************/
 458
 459struct entropy_store;
 460struct entropy_store {
 461        /* read-only data: */
 462        const struct poolinfo *poolinfo;
 463        __u32 *pool;
 464        const char *name;
 465        struct entropy_store *pull;
 466        struct work_struct push_work;
 467
 468        /* read-write data: */
 469        unsigned long last_pulled;
 470        spinlock_t lock;
 471        unsigned short add_ptr;
 472        unsigned short input_rotate;
 473        int entropy_count;
 474        int entropy_total;
 475        unsigned int initialized:1;
 476        unsigned int last_data_init:1;
 477        __u8 last_data[EXTRACT_SIZE];
 478};
 479
 480static ssize_t extract_entropy(struct entropy_store *r, void *buf,
 481                               size_t nbytes, int min, int rsvd);
 482static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
 483                                size_t nbytes, int fips);
 484
 485static void crng_reseed(struct crng_state *crng, struct entropy_store *r);
 486static void push_to_pool(struct work_struct *work);
 487static __u32 input_pool_data[INPUT_POOL_WORDS] __latent_entropy;
 488static __u32 blocking_pool_data[OUTPUT_POOL_WORDS] __latent_entropy;
 489
 490static struct entropy_store input_pool = {
 491        .poolinfo = &poolinfo_table[0],
 492        .name = "input",
 493        .lock = __SPIN_LOCK_UNLOCKED(input_pool.lock),
 494        .pool = input_pool_data
 495};
 496
 497static struct entropy_store blocking_pool = {
 498        .poolinfo = &poolinfo_table[1],
 499        .name = "blocking",
 500        .pull = &input_pool,
 501        .lock = __SPIN_LOCK_UNLOCKED(blocking_pool.lock),
 502        .pool = blocking_pool_data,
 503        .push_work = __WORK_INITIALIZER(blocking_pool.push_work,
 504                                        push_to_pool),
 505};
 506
 507static __u32 const twist_table[8] = {
 508        0x00000000, 0x3b6e20c8, 0x76dc4190, 0x4db26158,
 509        0xedb88320, 0xd6d6a3e8, 0x9b64c2b0, 0xa00ae278 };
 510
 511/*
 512 * This function adds bytes into the entropy "pool".  It does not
 513 * update the entropy estimate.  The caller should call
 514 * credit_entropy_bits if this is appropriate.
 515 *
 516 * The pool is stirred with a primitive polynomial of the appropriate
 517 * degree, and then twisted.  We twist by three bits at a time because
 518 * it's cheap to do so and helps slightly in the expected case where
 519 * the entropy is concentrated in the low-order bits.
 520 */
 521static void _mix_pool_bytes(struct entropy_store *r, const void *in,
 522                            int nbytes)
 523{
 524        unsigned long i, tap1, tap2, tap3, tap4, tap5;
 525        int input_rotate;
 526        int wordmask = r->poolinfo->poolwords - 1;
 527        const char *bytes = in;
 528        __u32 w;
 529
 530        tap1 = r->poolinfo->tap1;
 531        tap2 = r->poolinfo->tap2;
 532        tap3 = r->poolinfo->tap3;
 533        tap4 = r->poolinfo->tap4;
 534        tap5 = r->poolinfo->tap5;
 535
 536        input_rotate = r->input_rotate;
 537        i = r->add_ptr;
 538
 539        /* mix one byte at a time to simplify size handling and churn faster */
 540        while (nbytes--) {
 541                w = rol32(*bytes++, input_rotate);
 542                i = (i - 1) & wordmask;
 543
 544                /* XOR in the various taps */
 545                w ^= r->pool[i];
 546                w ^= r->pool[(i + tap1) & wordmask];
 547                w ^= r->pool[(i + tap2) & wordmask];
 548                w ^= r->pool[(i + tap3) & wordmask];
 549                w ^= r->pool[(i + tap4) & wordmask];
 550                w ^= r->pool[(i + tap5) & wordmask];
 551
 552                /* Mix the result back in with a twist */
 553                r->pool[i] = (w >> 3) ^ twist_table[w & 7];
 554
 555                /*
 556                 * Normally, we add 7 bits of rotation to the pool.
 557                 * At the beginning of the pool, add an extra 7 bits
 558                 * rotation, so that successive passes spread the
 559                 * input bits across the pool evenly.
 560                 */
 561                input_rotate = (input_rotate + (i ? 7 : 14)) & 31;
 562        }
 563
 564        r->input_rotate = input_rotate;
 565        r->add_ptr = i;
 566}
 567
 568static void __mix_pool_bytes(struct entropy_store *r, const void *in,
 569                             int nbytes)
 570{
 571        trace_mix_pool_bytes_nolock(r->name, nbytes, _RET_IP_);
 572        _mix_pool_bytes(r, in, nbytes);
 573}
 574
 575static void mix_pool_bytes(struct entropy_store *r, const void *in,
 576                           int nbytes)
 577{
 578        unsigned long flags;
 579
 580        trace_mix_pool_bytes(r->name, nbytes, _RET_IP_);
 581        spin_lock_irqsave(&r->lock, flags);
 582        _mix_pool_bytes(r, in, nbytes);
 583        spin_unlock_irqrestore(&r->lock, flags);
 584}
 585
 586struct fast_pool {
 587        __u32           pool[4];
 588        unsigned long   last;
 589        unsigned short  reg_idx;
 590        unsigned char   count;
 591};
 592
 593/*
 594 * This is a fast mixing routine used by the interrupt randomness
 595 * collector.  It's hardcoded for an 128 bit pool and assumes that any
 596 * locks that might be needed are taken by the caller.
 597 */
 598static void fast_mix(struct fast_pool *f)
 599{
 600        __u32 a = f->pool[0],   b = f->pool[1];
 601        __u32 c = f->pool[2],   d = f->pool[3];
 602
 603        a += b;                 c += d;
 604        b = rol32(b, 6);        d = rol32(d, 27);
 605        d ^= a;                 b ^= c;
 606
 607        a += b;                 c += d;
 608        b = rol32(b, 16);       d = rol32(d, 14);
 609        d ^= a;                 b ^= c;
 610
 611        a += b;                 c += d;
 612        b = rol32(b, 6);        d = rol32(d, 27);
 613        d ^= a;                 b ^= c;
 614
 615        a += b;                 c += d;
 616        b = rol32(b, 16);       d = rol32(d, 14);
 617        d ^= a;                 b ^= c;
 618
 619        f->pool[0] = a;  f->pool[1] = b;
 620        f->pool[2] = c;  f->pool[3] = d;
 621        f->count++;
 622}
 623
 624static void process_random_ready_list(void)
 625{
 626        unsigned long flags;
 627        struct random_ready_callback *rdy, *tmp;
 628
 629        spin_lock_irqsave(&random_ready_list_lock, flags);
 630        list_for_each_entry_safe(rdy, tmp, &random_ready_list, list) {
 631                struct module *owner = rdy->owner;
 632
 633                list_del_init(&rdy->list);
 634                rdy->func(rdy);
 635                module_put(owner);
 636        }
 637        spin_unlock_irqrestore(&random_ready_list_lock, flags);
 638}
 639
 640/*
 641 * Credit (or debit) the entropy store with n bits of entropy.
 642 * Use credit_entropy_bits_safe() if the value comes from userspace
 643 * or otherwise should be checked for extreme values.
 644 */
 645static void credit_entropy_bits(struct entropy_store *r, int nbits)
 646{
 647        int entropy_count, orig;
 648        const int pool_size = r->poolinfo->poolfracbits;
 649        int nfrac = nbits << ENTROPY_SHIFT;
 650
 651        if (!nbits)
 652                return;
 653
 654retry:
 655        entropy_count = orig = READ_ONCE(r->entropy_count);
 656        if (nfrac < 0) {
 657                /* Debit */
 658                entropy_count += nfrac;
 659        } else {
 660                /*
 661                 * Credit: we have to account for the possibility of
 662                 * overwriting already present entropy.  Even in the
 663                 * ideal case of pure Shannon entropy, new contributions
 664                 * approach the full value asymptotically:
 665                 *
 666                 * entropy <- entropy + (pool_size - entropy) *
 667                 *      (1 - exp(-add_entropy/pool_size))
 668                 *
 669                 * For add_entropy <= pool_size/2 then
 670                 * (1 - exp(-add_entropy/pool_size)) >=
 671                 *    (add_entropy/pool_size)*0.7869...
 672                 * so we can approximate the exponential with
 673                 * 3/4*add_entropy/pool_size and still be on the
 674                 * safe side by adding at most pool_size/2 at a time.
 675                 *
 676                 * The use of pool_size-2 in the while statement is to
 677                 * prevent rounding artifacts from making the loop
 678                 * arbitrarily long; this limits the loop to log2(pool_size)*2
 679                 * turns no matter how large nbits is.
 680                 */
 681                int pnfrac = nfrac;
 682                const int s = r->poolinfo->poolbitshift + ENTROPY_SHIFT + 2;
 683                /* The +2 corresponds to the /4 in the denominator */
 684
 685                do {
 686                        unsigned int anfrac = min(pnfrac, pool_size/2);
 687                        unsigned int add =
 688                                ((pool_size - entropy_count)*anfrac*3) >> s;
 689
 690                        entropy_count += add;
 691                        pnfrac -= anfrac;
 692                } while (unlikely(entropy_count < pool_size-2 && pnfrac));
 693        }
 694
 695        if (unlikely(entropy_count < 0)) {
 696                pr_warn("random: negative entropy/overflow: pool %s count %d\n",
 697                        r->name, entropy_count);
 698                WARN_ON(1);
 699                entropy_count = 0;
 700        } else if (entropy_count > pool_size)
 701                entropy_count = pool_size;
 702        if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
 703                goto retry;
 704
 705        r->entropy_total += nbits;
 706        if (!r->initialized && r->entropy_total > 128) {
 707                r->initialized = 1;
 708                r->entropy_total = 0;
 709        }
 710
 711        trace_credit_entropy_bits(r->name, nbits,
 712                                  entropy_count >> ENTROPY_SHIFT,
 713                                  r->entropy_total, _RET_IP_);
 714
 715        if (r == &input_pool) {
 716                int entropy_bits = entropy_count >> ENTROPY_SHIFT;
 717
 718                if (crng_init < 2 && entropy_bits >= 128) {
 719                        crng_reseed(&primary_crng, r);
 720                        entropy_bits = r->entropy_count >> ENTROPY_SHIFT;
 721                }
 722
 723                /* should we wake readers? */
 724                if (entropy_bits >= random_read_wakeup_bits &&
 725                    wq_has_sleeper(&random_read_wait)) {
 726                        wake_up_interruptible(&random_read_wait);
 727                        kill_fasync(&fasync, SIGIO, POLL_IN);
 728                }
 729                /* If the input pool is getting full, send some
 730                 * entropy to the blocking pool until it is 75% full.
 731                 */
 732                if (entropy_bits > random_write_wakeup_bits &&
 733                    r->initialized &&
 734                    r->entropy_total >= 2*random_read_wakeup_bits) {
 735                        struct entropy_store *other = &blocking_pool;
 736
 737                        if (other->entropy_count <=
 738                            3 * other->poolinfo->poolfracbits / 4) {
 739                                schedule_work(&other->push_work);
 740                                r->entropy_total = 0;
 741                        }
 742                }
 743        }
 744}
 745
 746static int credit_entropy_bits_safe(struct entropy_store *r, int nbits)
 747{
 748        const int nbits_max = r->poolinfo->poolwords * 32;
 749
 750        if (nbits < 0)
 751                return -EINVAL;
 752
 753        /* Cap the value to avoid overflows */
 754        nbits = min(nbits,  nbits_max);
 755
 756        credit_entropy_bits(r, nbits);
 757        return 0;
 758}
 759
 760/*********************************************************************
 761 *
 762 * CRNG using CHACHA20
 763 *
 764 *********************************************************************/
 765
 766#define CRNG_RESEED_INTERVAL (300*HZ)
 767
 768static DECLARE_WAIT_QUEUE_HEAD(crng_init_wait);
 769
 770#ifdef CONFIG_NUMA
 771/*
 772 * Hack to deal with crazy userspace progams when they are all trying
 773 * to access /dev/urandom in parallel.  The programs are almost
 774 * certainly doing something terribly wrong, but we'll work around
 775 * their brain damage.
 776 */
 777static struct crng_state **crng_node_pool __read_mostly;
 778#endif
 779
 780static void invalidate_batched_entropy(void);
 781
 782static void crng_initialize(struct crng_state *crng)
 783{
 784        int             i;
 785        unsigned long   rv;
 786
 787        memcpy(&crng->state[0], "expand 32-byte k", 16);
 788        if (crng == &primary_crng)
 789                _extract_entropy(&input_pool, &crng->state[4],
 790                                 sizeof(__u32) * 12, 0);
 791        else
 792                _get_random_bytes(&crng->state[4], sizeof(__u32) * 12);
 793        for (i = 4; i < 16; i++) {
 794                if (!arch_get_random_seed_long(&rv) &&
 795                    !arch_get_random_long(&rv))
 796                        rv = random_get_entropy();
 797                crng->state[i] ^= rv;
 798        }
 799        crng->init_time = jiffies - CRNG_RESEED_INTERVAL - 1;
 800}
 801
 802#ifdef CONFIG_NUMA
 803static void do_numa_crng_init(struct work_struct *work)
 804{
 805        int i;
 806        struct crng_state *crng;
 807        struct crng_state **pool;
 808
 809        pool = kcalloc(nr_node_ids, sizeof(*pool), GFP_KERNEL|__GFP_NOFAIL);
 810        for_each_online_node(i) {
 811                crng = kmalloc_node(sizeof(struct crng_state),
 812                                    GFP_KERNEL | __GFP_NOFAIL, i);
 813                spin_lock_init(&crng->lock);
 814                crng_initialize(crng);
 815                pool[i] = crng;
 816        }
 817        mb();
 818        if (cmpxchg(&crng_node_pool, NULL, pool)) {
 819                for_each_node(i)
 820                        kfree(pool[i]);
 821                kfree(pool);
 822        }
 823}
 824
 825static DECLARE_WORK(numa_crng_init_work, do_numa_crng_init);
 826
 827static void numa_crng_init(void)
 828{
 829        schedule_work(&numa_crng_init_work);
 830}
 831#else
 832static void numa_crng_init(void) {}
 833#endif
 834
 835/*
 836 * crng_fast_load() can be called by code in the interrupt service
 837 * path.  So we can't afford to dilly-dally.
 838 */
 839static int crng_fast_load(const char *cp, size_t len)
 840{
 841        unsigned long flags;
 842        char *p;
 843
 844        if (!spin_trylock_irqsave(&primary_crng.lock, flags))
 845                return 0;
 846        if (crng_init != 0) {
 847                spin_unlock_irqrestore(&primary_crng.lock, flags);
 848                return 0;
 849        }
 850        p = (unsigned char *) &primary_crng.state[4];
 851        while (len > 0 && crng_init_cnt < CRNG_INIT_CNT_THRESH) {
 852                p[crng_init_cnt % CHACHA20_KEY_SIZE] ^= *cp;
 853                cp++; crng_init_cnt++; len--;
 854        }
 855        spin_unlock_irqrestore(&primary_crng.lock, flags);
 856        if (crng_init_cnt >= CRNG_INIT_CNT_THRESH) {
 857                invalidate_batched_entropy();
 858                crng_init = 1;
 859                wake_up_interruptible(&crng_init_wait);
 860                pr_notice("random: fast init done\n");
 861        }
 862        return 1;
 863}
 864
 865/*
 866 * crng_slow_load() is called by add_device_randomness, which has two
 867 * attributes.  (1) We can't trust the buffer passed to it is
 868 * guaranteed to be unpredictable (so it might not have any entropy at
 869 * all), and (2) it doesn't have the performance constraints of
 870 * crng_fast_load().
 871 *
 872 * So we do something more comprehensive which is guaranteed to touch
 873 * all of the primary_crng's state, and which uses a LFSR with a
 874 * period of 255 as part of the mixing algorithm.  Finally, we do
 875 * *not* advance crng_init_cnt since buffer we may get may be something
 876 * like a fixed DMI table (for example), which might very well be
 877 * unique to the machine, but is otherwise unvarying.
 878 */
 879static int crng_slow_load(const char *cp, size_t len)
 880{
 881        unsigned long           flags;
 882        static unsigned char    lfsr = 1;
 883        unsigned char           tmp;
 884        unsigned                i, max = CHACHA20_KEY_SIZE;
 885        const char *            src_buf = cp;
 886        char *                  dest_buf = (char *) &primary_crng.state[4];
 887
 888        if (!spin_trylock_irqsave(&primary_crng.lock, flags))
 889                return 0;
 890        if (crng_init != 0) {
 891                spin_unlock_irqrestore(&primary_crng.lock, flags);
 892                return 0;
 893        }
 894        if (len > max)
 895                max = len;
 896
 897        for (i = 0; i < max ; i++) {
 898                tmp = lfsr;
 899                lfsr >>= 1;
 900                if (tmp & 1)
 901                        lfsr ^= 0xE1;
 902                tmp = dest_buf[i % CHACHA20_KEY_SIZE];
 903                dest_buf[i % CHACHA20_KEY_SIZE] ^= src_buf[i % len] ^ lfsr;
 904                lfsr += (tmp << 3) | (tmp >> 5);
 905        }
 906        spin_unlock_irqrestore(&primary_crng.lock, flags);
 907        return 1;
 908}
 909
 910static void crng_reseed(struct crng_state *crng, struct entropy_store *r)
 911{
 912        unsigned long   flags;
 913        int             i, num;
 914        union {
 915                __u32   block[CHACHA20_BLOCK_WORDS];
 916                __u32   key[8];
 917        } buf;
 918
 919        if (r) {
 920                num = extract_entropy(r, &buf, 32, 16, 0);
 921                if (num == 0)
 922                        return;
 923        } else {
 924                _extract_crng(&primary_crng, buf.block);
 925                _crng_backtrack_protect(&primary_crng, buf.block,
 926                                        CHACHA20_KEY_SIZE);
 927        }
 928        spin_lock_irqsave(&crng->lock, flags);
 929        for (i = 0; i < 8; i++) {
 930                unsigned long   rv;
 931                if (!arch_get_random_seed_long(&rv) &&
 932                    !arch_get_random_long(&rv))
 933                        rv = random_get_entropy();
 934                crng->state[i+4] ^= buf.key[i] ^ rv;
 935        }
 936        memzero_explicit(&buf, sizeof(buf));
 937        crng->init_time = jiffies;
 938        spin_unlock_irqrestore(&crng->lock, flags);
 939        if (crng == &primary_crng && crng_init < 2) {
 940                invalidate_batched_entropy();
 941                numa_crng_init();
 942                crng_init = 2;
 943                process_random_ready_list();
 944                wake_up_interruptible(&crng_init_wait);
 945                pr_notice("random: crng init done\n");
 946                if (unseeded_warning.missed) {
 947                        pr_notice("random: %d get_random_xx warning(s) missed "
 948                                  "due to ratelimiting\n",
 949                                  unseeded_warning.missed);
 950                        unseeded_warning.missed = 0;
 951                }
 952                if (urandom_warning.missed) {
 953                        pr_notice("random: %d urandom warning(s) missed "
 954                                  "due to ratelimiting\n",
 955                                  urandom_warning.missed);
 956                        urandom_warning.missed = 0;
 957                }
 958        }
 959}
 960
 961static void _extract_crng(struct crng_state *crng,
 962                          __u32 out[CHACHA20_BLOCK_WORDS])
 963{
 964        unsigned long v, flags;
 965
 966        if (crng_ready() &&
 967            (time_after(crng_global_init_time, crng->init_time) ||
 968             time_after(jiffies, crng->init_time + CRNG_RESEED_INTERVAL)))
 969                crng_reseed(crng, crng == &primary_crng ? &input_pool : NULL);
 970        spin_lock_irqsave(&crng->lock, flags);
 971        if (arch_get_random_long(&v))
 972                crng->state[14] ^= v;
 973        chacha20_block(&crng->state[0], out);
 974        if (crng->state[12] == 0)
 975                crng->state[13]++;
 976        spin_unlock_irqrestore(&crng->lock, flags);
 977}
 978
 979static void extract_crng(__u32 out[CHACHA20_BLOCK_WORDS])
 980{
 981        struct crng_state *crng = NULL;
 982
 983#ifdef CONFIG_NUMA
 984        if (crng_node_pool)
 985                crng = crng_node_pool[numa_node_id()];
 986        if (crng == NULL)
 987#endif
 988                crng = &primary_crng;
 989        _extract_crng(crng, out);
 990}
 991
 992/*
 993 * Use the leftover bytes from the CRNG block output (if there is
 994 * enough) to mutate the CRNG key to provide backtracking protection.
 995 */
 996static void _crng_backtrack_protect(struct crng_state *crng,
 997                                    __u32 tmp[CHACHA20_BLOCK_WORDS], int used)
 998{
 999        unsigned long   flags;
1000        __u32           *s, *d;
1001        int             i;
1002
1003        used = round_up(used, sizeof(__u32));
1004        if (used + CHACHA20_KEY_SIZE > CHACHA20_BLOCK_SIZE) {
1005                extract_crng(tmp);
1006                used = 0;
1007        }
1008        spin_lock_irqsave(&crng->lock, flags);
1009        s = &tmp[used / sizeof(__u32)];
1010        d = &crng->state[4];
1011        for (i=0; i < 8; i++)
1012                *d++ ^= *s++;
1013        spin_unlock_irqrestore(&crng->lock, flags);
1014}
1015
1016static void crng_backtrack_protect(__u32 tmp[CHACHA20_BLOCK_WORDS], int used)
1017{
1018        struct crng_state *crng = NULL;
1019
1020#ifdef CONFIG_NUMA
1021        if (crng_node_pool)
1022                crng = crng_node_pool[numa_node_id()];
1023        if (crng == NULL)
1024#endif
1025                crng = &primary_crng;
1026        _crng_backtrack_protect(crng, tmp, used);
1027}
1028
1029static ssize_t extract_crng_user(void __user *buf, size_t nbytes)
1030{
1031        ssize_t ret = 0, i = CHACHA20_BLOCK_SIZE;
1032        __u32 tmp[CHACHA20_BLOCK_WORDS];
1033        int large_request = (nbytes > 256);
1034
1035        while (nbytes) {
1036                if (large_request && need_resched()) {
1037                        if (signal_pending(current)) {
1038                                if (ret == 0)
1039                                        ret = -ERESTARTSYS;
1040                                break;
1041                        }
1042                        schedule();
1043                }
1044
1045                extract_crng(tmp);
1046                i = min_t(int, nbytes, CHACHA20_BLOCK_SIZE);
1047                if (copy_to_user(buf, tmp, i)) {
1048                        ret = -EFAULT;
1049                        break;
1050                }
1051
1052                nbytes -= i;
1053                buf += i;
1054                ret += i;
1055        }
1056        crng_backtrack_protect(tmp, i);
1057
1058        /* Wipe data just written to memory */
1059        memzero_explicit(tmp, sizeof(tmp));
1060
1061        return ret;
1062}
1063
1064
1065/*********************************************************************
1066 *
1067 * Entropy input management
1068 *
1069 *********************************************************************/
1070
1071/* There is one of these per entropy source */
1072struct timer_rand_state {
1073        cycles_t last_time;
1074        long last_delta, last_delta2;
1075};
1076
1077#define INIT_TIMER_RAND_STATE { INITIAL_JIFFIES, };
1078
1079/*
1080 * Add device- or boot-specific data to the input pool to help
1081 * initialize it.
1082 *
1083 * None of this adds any entropy; it is meant to avoid the problem of
1084 * the entropy pool having similar initial state across largely
1085 * identical devices.
1086 */
1087void add_device_randomness(const void *buf, unsigned int size)
1088{
1089        unsigned long time = random_get_entropy() ^ jiffies;
1090        unsigned long flags;
1091
1092        if (!crng_ready() && size)
1093                crng_slow_load(buf, size);
1094
1095        trace_add_device_randomness(size, _RET_IP_);
1096        spin_lock_irqsave(&input_pool.lock, flags);
1097        _mix_pool_bytes(&input_pool, buf, size);
1098        _mix_pool_bytes(&input_pool, &time, sizeof(time));
1099        spin_unlock_irqrestore(&input_pool.lock, flags);
1100}
1101EXPORT_SYMBOL(add_device_randomness);
1102
1103static struct timer_rand_state input_timer_state = INIT_TIMER_RAND_STATE;
1104
1105/*
1106 * This function adds entropy to the entropy "pool" by using timing
1107 * delays.  It uses the timer_rand_state structure to make an estimate
1108 * of how many bits of entropy this call has added to the pool.
1109 *
1110 * The number "num" is also added to the pool - it should somehow describe
1111 * the type of event which just happened.  This is currently 0-255 for
1112 * keyboard scan codes, and 256 upwards for interrupts.
1113 *
1114 */
1115static void add_timer_randomness(struct timer_rand_state *state, unsigned num)
1116{
1117        struct entropy_store    *r;
1118        struct {
1119                long jiffies;
1120                unsigned cycles;
1121                unsigned num;
1122        } sample;
1123        long delta, delta2, delta3;
1124
1125        preempt_disable();
1126
1127        sample.jiffies = jiffies;
1128        sample.cycles = random_get_entropy();
1129        sample.num = num;
1130        r = &input_pool;
1131        mix_pool_bytes(r, &sample, sizeof(sample));
1132
1133        /*
1134         * Calculate number of bits of randomness we probably added.
1135         * We take into account the first, second and third-order deltas
1136         * in order to make our estimate.
1137         */
1138        delta = sample.jiffies - state->last_time;
1139        state->last_time = sample.jiffies;
1140
1141        delta2 = delta - state->last_delta;
1142        state->last_delta = delta;
1143
1144        delta3 = delta2 - state->last_delta2;
1145        state->last_delta2 = delta2;
1146
1147        if (delta < 0)
1148                delta = -delta;
1149        if (delta2 < 0)
1150                delta2 = -delta2;
1151        if (delta3 < 0)
1152                delta3 = -delta3;
1153        if (delta > delta2)
1154                delta = delta2;
1155        if (delta > delta3)
1156                delta = delta3;
1157
1158        /*
1159         * delta is now minimum absolute delta.
1160         * Round down by 1 bit on general principles,
1161         * and limit entropy entimate to 12 bits.
1162         */
1163        credit_entropy_bits(r, min_t(int, fls(delta>>1), 11));
1164
1165        preempt_enable();
1166}
1167
1168void add_input_randomness(unsigned int type, unsigned int code,
1169                                 unsigned int value)
1170{
1171        static unsigned char last_value;
1172
1173        /* ignore autorepeat and the like */
1174        if (value == last_value)
1175                return;
1176
1177        last_value = value;
1178        add_timer_randomness(&input_timer_state,
1179                             (type << 4) ^ code ^ (code >> 4) ^ value);
1180        trace_add_input_randomness(ENTROPY_BITS(&input_pool));
1181}
1182EXPORT_SYMBOL_GPL(add_input_randomness);
1183
1184static DEFINE_PER_CPU(struct fast_pool, irq_randomness);
1185
1186#ifdef ADD_INTERRUPT_BENCH
1187static unsigned long avg_cycles, avg_deviation;
1188
1189#define AVG_SHIFT 8     /* Exponential average factor k=1/256 */
1190#define FIXED_1_2 (1 << (AVG_SHIFT-1))
1191
1192static void add_interrupt_bench(cycles_t start)
1193{
1194        long delta = random_get_entropy() - start;
1195
1196        /* Use a weighted moving average */
1197        delta = delta - ((avg_cycles + FIXED_1_2) >> AVG_SHIFT);
1198        avg_cycles += delta;
1199        /* And average deviation */
1200        delta = abs(delta) - ((avg_deviation + FIXED_1_2) >> AVG_SHIFT);
1201        avg_deviation += delta;
1202}
1203#else
1204#define add_interrupt_bench(x)
1205#endif
1206
1207static __u32 get_reg(struct fast_pool *f, struct pt_regs *regs)
1208{
1209        __u32 *ptr = (__u32 *) regs;
1210        unsigned int idx;
1211
1212        if (regs == NULL)
1213                return 0;
1214        idx = READ_ONCE(f->reg_idx);
1215        if (idx >= sizeof(struct pt_regs) / sizeof(__u32))
1216                idx = 0;
1217        ptr += idx++;
1218        WRITE_ONCE(f->reg_idx, idx);
1219        return *ptr;
1220}
1221
1222void add_interrupt_randomness(int irq, int irq_flags)
1223{
1224        struct entropy_store    *r;
1225        struct fast_pool        *fast_pool = this_cpu_ptr(&irq_randomness);
1226        struct pt_regs          *regs = get_irq_regs();
1227        unsigned long           now = jiffies;
1228        cycles_t                cycles = random_get_entropy();
1229        __u32                   c_high, j_high;
1230        __u64                   ip;
1231        unsigned long           seed;
1232        int                     credit = 0;
1233
1234        if (cycles == 0)
1235                cycles = get_reg(fast_pool, regs);
1236        c_high = (sizeof(cycles) > 4) ? cycles >> 32 : 0;
1237        j_high = (sizeof(now) > 4) ? now >> 32 : 0;
1238        fast_pool->pool[0] ^= cycles ^ j_high ^ irq;
1239        fast_pool->pool[1] ^= now ^ c_high;
1240        ip = regs ? instruction_pointer(regs) : _RET_IP_;
1241        fast_pool->pool[2] ^= ip;
1242        fast_pool->pool[3] ^= (sizeof(ip) > 4) ? ip >> 32 :
1243                get_reg(fast_pool, regs);
1244
1245        fast_mix(fast_pool);
1246        add_interrupt_bench(cycles);
1247
1248        if (unlikely(crng_init == 0)) {
1249                if ((fast_pool->count >= 64) &&
1250                    crng_fast_load((char *) fast_pool->pool,
1251                                   sizeof(fast_pool->pool))) {
1252                        fast_pool->count = 0;
1253                        fast_pool->last = now;
1254                }
1255                return;
1256        }
1257
1258        if ((fast_pool->count < 64) &&
1259            !time_after(now, fast_pool->last + HZ))
1260                return;
1261
1262        r = &input_pool;
1263        if (!spin_trylock(&r->lock))
1264                return;
1265
1266        fast_pool->last = now;
1267        __mix_pool_bytes(r, &fast_pool->pool, sizeof(fast_pool->pool));
1268
1269        /*
1270         * If we have architectural seed generator, produce a seed and
1271         * add it to the pool.  For the sake of paranoia don't let the
1272         * architectural seed generator dominate the input from the
1273         * interrupt noise.
1274         */
1275        if (arch_get_random_seed_long(&seed)) {
1276                __mix_pool_bytes(r, &seed, sizeof(seed));
1277                credit = 1;
1278        }
1279        spin_unlock(&r->lock);
1280
1281        fast_pool->count = 0;
1282
1283        /* award one bit for the contents of the fast pool */
1284        credit_entropy_bits(r, credit + 1);
1285}
1286EXPORT_SYMBOL_GPL(add_interrupt_randomness);
1287
1288#ifdef CONFIG_BLOCK
1289void add_disk_randomness(struct gendisk *disk)
1290{
1291        if (!disk || !disk->random)
1292                return;
1293        /* first major is 1, so we get >= 0x200 here */
1294        add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
1295        trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
1296}
1297EXPORT_SYMBOL_GPL(add_disk_randomness);
1298#endif
1299
1300/*********************************************************************
1301 *
1302 * Entropy extraction routines
1303 *
1304 *********************************************************************/
1305
1306/*
1307 * This utility inline function is responsible for transferring entropy
1308 * from the primary pool to the secondary extraction pool. We make
1309 * sure we pull enough for a 'catastrophic reseed'.
1310 */
1311static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes);
1312static void xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1313{
1314        if (!r->pull ||
1315            r->entropy_count >= (nbytes << (ENTROPY_SHIFT + 3)) ||
1316            r->entropy_count > r->poolinfo->poolfracbits)
1317                return;
1318
1319        _xfer_secondary_pool(r, nbytes);
1320}
1321
1322static void _xfer_secondary_pool(struct entropy_store *r, size_t nbytes)
1323{
1324        __u32   tmp[OUTPUT_POOL_WORDS];
1325
1326        int bytes = nbytes;
1327
1328        /* pull at least as much as a wakeup */
1329        bytes = max_t(int, bytes, random_read_wakeup_bits / 8);
1330        /* but never more than the buffer size */
1331        bytes = min_t(int, bytes, sizeof(tmp));
1332
1333        trace_xfer_secondary_pool(r->name, bytes * 8, nbytes * 8,
1334                                  ENTROPY_BITS(r), ENTROPY_BITS(r->pull));
1335        bytes = extract_entropy(r->pull, tmp, bytes,
1336                                random_read_wakeup_bits / 8, 0);
1337        mix_pool_bytes(r, tmp, bytes);
1338        credit_entropy_bits(r, bytes*8);
1339}
1340
1341/*
1342 * Used as a workqueue function so that when the input pool is getting
1343 * full, we can "spill over" some entropy to the output pools.  That
1344 * way the output pools can store some of the excess entropy instead
1345 * of letting it go to waste.
1346 */
1347static void push_to_pool(struct work_struct *work)
1348{
1349        struct entropy_store *r = container_of(work, struct entropy_store,
1350                                              push_work);
1351        BUG_ON(!r);
1352        _xfer_secondary_pool(r, random_read_wakeup_bits/8);
1353        trace_push_to_pool(r->name, r->entropy_count >> ENTROPY_SHIFT,
1354                           r->pull->entropy_count >> ENTROPY_SHIFT);
1355}
1356
1357/*
1358 * This function decides how many bytes to actually take from the
1359 * given pool, and also debits the entropy count accordingly.
1360 */
1361static size_t account(struct entropy_store *r, size_t nbytes, int min,
1362                      int reserved)
1363{
1364        int entropy_count, orig, have_bytes;
1365        size_t ibytes, nfrac;
1366
1367        BUG_ON(r->entropy_count > r->poolinfo->poolfracbits);
1368
1369        /* Can we pull enough? */
1370retry:
1371        entropy_count = orig = READ_ONCE(r->entropy_count);
1372        ibytes = nbytes;
1373        /* never pull more than available */
1374        have_bytes = entropy_count >> (ENTROPY_SHIFT + 3);
1375
1376        if ((have_bytes -= reserved) < 0)
1377                have_bytes = 0;
1378        ibytes = min_t(size_t, ibytes, have_bytes);
1379        if (ibytes < min)
1380                ibytes = 0;
1381
1382        if (unlikely(entropy_count < 0)) {
1383                pr_warn("random: negative entropy count: pool %s count %d\n",
1384                        r->name, entropy_count);
1385                WARN_ON(1);
1386                entropy_count = 0;
1387        }
1388        nfrac = ibytes << (ENTROPY_SHIFT + 3);
1389        if ((size_t) entropy_count > nfrac)
1390                entropy_count -= nfrac;
1391        else
1392                entropy_count = 0;
1393
1394        if (cmpxchg(&r->entropy_count, orig, entropy_count) != orig)
1395                goto retry;
1396
1397        trace_debit_entropy(r->name, 8 * ibytes);
1398        if (ibytes &&
1399            (r->entropy_count >> ENTROPY_SHIFT) < random_write_wakeup_bits) {
1400                wake_up_interruptible(&random_write_wait);
1401                kill_fasync(&fasync, SIGIO, POLL_OUT);
1402        }
1403
1404        return ibytes;
1405}
1406
1407/*
1408 * This function does the actual extraction for extract_entropy and
1409 * extract_entropy_user.
1410 *
1411 * Note: we assume that .poolwords is a multiple of 16 words.
1412 */
1413static void extract_buf(struct entropy_store *r, __u8 *out)
1414{
1415        int i;
1416        union {
1417                __u32 w[5];
1418                unsigned long l[LONGS(20)];
1419        } hash;
1420        __u32 workspace[SHA_WORKSPACE_WORDS];
1421        unsigned long flags;
1422
1423        /*
1424         * If we have an architectural hardware random number
1425         * generator, use it for SHA's initial vector
1426         */
1427        sha_init(hash.w);
1428        for (i = 0; i < LONGS(20); i++) {
1429                unsigned long v;
1430                if (!arch_get_random_long(&v))
1431                        break;
1432                hash.l[i] = v;
1433        }
1434
1435        /* Generate a hash across the pool, 16 words (512 bits) at a time */
1436        spin_lock_irqsave(&r->lock, flags);
1437        for (i = 0; i < r->poolinfo->poolwords; i += 16)
1438                sha_transform(hash.w, (__u8 *)(r->pool + i), workspace);
1439
1440        /*
1441         * We mix the hash back into the pool to prevent backtracking
1442         * attacks (where the attacker knows the state of the pool
1443         * plus the current outputs, and attempts to find previous
1444         * ouputs), unless the hash function can be inverted. By
1445         * mixing at least a SHA1 worth of hash data back, we make
1446         * brute-forcing the feedback as hard as brute-forcing the
1447         * hash.
1448         */
1449        __mix_pool_bytes(r, hash.w, sizeof(hash.w));
1450        spin_unlock_irqrestore(&r->lock, flags);
1451
1452        memzero_explicit(workspace, sizeof(workspace));
1453
1454        /*
1455         * In case the hash function has some recognizable output
1456         * pattern, we fold it in half. Thus, we always feed back
1457         * twice as much data as we output.
1458         */
1459        hash.w[0] ^= hash.w[3];
1460        hash.w[1] ^= hash.w[4];
1461        hash.w[2] ^= rol32(hash.w[2], 16);
1462
1463        memcpy(out, &hash, EXTRACT_SIZE);
1464        memzero_explicit(&hash, sizeof(hash));
1465}
1466
1467static ssize_t _extract_entropy(struct entropy_store *r, void *buf,
1468                                size_t nbytes, int fips)
1469{
1470        ssize_t ret = 0, i;
1471        __u8 tmp[EXTRACT_SIZE];
1472        unsigned long flags;
1473
1474        while (nbytes) {
1475                extract_buf(r, tmp);
1476
1477                if (fips) {
1478                        spin_lock_irqsave(&r->lock, flags);
1479                        if (!memcmp(tmp, r->last_data, EXTRACT_SIZE))
1480                                panic("Hardware RNG duplicated output!\n");
1481                        memcpy(r->last_data, tmp, EXTRACT_SIZE);
1482                        spin_unlock_irqrestore(&r->lock, flags);
1483                }
1484                i = min_t(int, nbytes, EXTRACT_SIZE);
1485                memcpy(buf, tmp, i);
1486                nbytes -= i;
1487                buf += i;
1488                ret += i;
1489        }
1490
1491        /* Wipe data just returned from memory */
1492        memzero_explicit(tmp, sizeof(tmp));
1493
1494        return ret;
1495}
1496
1497/*
1498 * This function extracts randomness from the "entropy pool", and
1499 * returns it in a buffer.
1500 *
1501 * The min parameter specifies the minimum amount we can pull before
1502 * failing to avoid races that defeat catastrophic reseeding while the
1503 * reserved parameter indicates how much entropy we must leave in the
1504 * pool after each pull to avoid starving other readers.
1505 */
1506static ssize_t extract_entropy(struct entropy_store *r, void *buf,
1507                                 size_t nbytes, int min, int reserved)
1508{
1509        __u8 tmp[EXTRACT_SIZE];
1510        unsigned long flags;
1511
1512        /* if last_data isn't primed, we need EXTRACT_SIZE extra bytes */
1513        if (fips_enabled) {
1514                spin_lock_irqsave(&r->lock, flags);
1515                if (!r->last_data_init) {
1516                        r->last_data_init = 1;
1517                        spin_unlock_irqrestore(&r->lock, flags);
1518                        trace_extract_entropy(r->name, EXTRACT_SIZE,
1519                                              ENTROPY_BITS(r), _RET_IP_);
1520                        xfer_secondary_pool(r, EXTRACT_SIZE);
1521                        extract_buf(r, tmp);
1522                        spin_lock_irqsave(&r->lock, flags);
1523                        memcpy(r->last_data, tmp, EXTRACT_SIZE);
1524                }
1525                spin_unlock_irqrestore(&r->lock, flags);
1526        }
1527
1528        trace_extract_entropy(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1529        xfer_secondary_pool(r, nbytes);
1530        nbytes = account(r, nbytes, min, reserved);
1531
1532        return _extract_entropy(r, buf, nbytes, fips_enabled);
1533}
1534
1535/*
1536 * This function extracts randomness from the "entropy pool", and
1537 * returns it in a userspace buffer.
1538 */
1539static ssize_t extract_entropy_user(struct entropy_store *r, void __user *buf,
1540                                    size_t nbytes)
1541{
1542        ssize_t ret = 0, i;
1543        __u8 tmp[EXTRACT_SIZE];
1544        int large_request = (nbytes > 256);
1545
1546        trace_extract_entropy_user(r->name, nbytes, ENTROPY_BITS(r), _RET_IP_);
1547        xfer_secondary_pool(r, nbytes);
1548        nbytes = account(r, nbytes, 0, 0);
1549
1550        while (nbytes) {
1551                if (large_request && need_resched()) {
1552                        if (signal_pending(current)) {
1553                                if (ret == 0)
1554                                        ret = -ERESTARTSYS;
1555                                break;
1556                        }
1557                        schedule();
1558                }
1559
1560                extract_buf(r, tmp);
1561                i = min_t(int, nbytes, EXTRACT_SIZE);
1562                if (copy_to_user(buf, tmp, i)) {
1563                        ret = -EFAULT;
1564                        break;
1565                }
1566
1567                nbytes -= i;
1568                buf += i;
1569                ret += i;
1570        }
1571
1572        /* Wipe data just returned from memory */
1573        memzero_explicit(tmp, sizeof(tmp));
1574
1575        return ret;
1576}
1577
1578#define warn_unseeded_randomness(previous) \
1579        _warn_unseeded_randomness(__func__, (void *) _RET_IP_, (previous))
1580
1581static void _warn_unseeded_randomness(const char *func_name, void *caller,
1582                                      void **previous)
1583{
1584#ifdef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1585        const bool print_once = false;
1586#else
1587        static bool print_once __read_mostly;
1588#endif
1589
1590        if (print_once ||
1591            crng_ready() ||
1592            (previous && (caller == READ_ONCE(*previous))))
1593                return;
1594        WRITE_ONCE(*previous, caller);
1595#ifndef CONFIG_WARN_ALL_UNSEEDED_RANDOM
1596        print_once = true;
1597#endif
1598        if (__ratelimit(&unseeded_warning))
1599                pr_notice("random: %s called from %pS with crng_init=%d\n",
1600                          func_name, caller, crng_init);
1601}
1602
1603/*
1604 * This function is the exported kernel interface.  It returns some
1605 * number of good random numbers, suitable for key generation, seeding
1606 * TCP sequence numbers, etc.  It does not rely on the hardware random
1607 * number generator.  For random bytes direct from the hardware RNG
1608 * (when available), use get_random_bytes_arch(). In order to ensure
1609 * that the randomness provided by this function is okay, the function
1610 * wait_for_random_bytes() should be called and return 0 at least once
1611 * at any point prior.
1612 */
1613static void _get_random_bytes(void *buf, int nbytes)
1614{
1615        __u32 tmp[CHACHA20_BLOCK_WORDS];
1616
1617        trace_get_random_bytes(nbytes, _RET_IP_);
1618
1619        while (nbytes >= CHACHA20_BLOCK_SIZE) {
1620                extract_crng(buf);
1621                buf += CHACHA20_BLOCK_SIZE;
1622                nbytes -= CHACHA20_BLOCK_SIZE;
1623        }
1624
1625        if (nbytes > 0) {
1626                extract_crng(tmp);
1627                memcpy(buf, tmp, nbytes);
1628                crng_backtrack_protect(tmp, nbytes);
1629        } else
1630                crng_backtrack_protect(tmp, CHACHA20_BLOCK_SIZE);
1631        memzero_explicit(tmp, sizeof(tmp));
1632}
1633
1634void get_random_bytes(void *buf, int nbytes)
1635{
1636        static void *previous;
1637
1638        warn_unseeded_randomness(&previous);
1639        _get_random_bytes(buf, nbytes);
1640}
1641EXPORT_SYMBOL(get_random_bytes);
1642
1643/*
1644 * Wait for the urandom pool to be seeded and thus guaranteed to supply
1645 * cryptographically secure random numbers. This applies to: the /dev/urandom
1646 * device, the get_random_bytes function, and the get_random_{u32,u64,int,long}
1647 * family of functions. Using any of these functions without first calling
1648 * this function forfeits the guarantee of security.
1649 *
1650 * Returns: 0 if the urandom pool has been seeded.
1651 *          -ERESTARTSYS if the function was interrupted by a signal.
1652 */
1653int wait_for_random_bytes(void)
1654{
1655        if (likely(crng_ready()))
1656                return 0;
1657        return wait_event_interruptible(crng_init_wait, crng_ready());
1658}
1659EXPORT_SYMBOL(wait_for_random_bytes);
1660
1661/*
1662 * Add a callback function that will be invoked when the nonblocking
1663 * pool is initialised.
1664 *
1665 * returns: 0 if callback is successfully added
1666 *          -EALREADY if pool is already initialised (callback not called)
1667 *          -ENOENT if module for callback is not alive
1668 */
1669int add_random_ready_callback(struct random_ready_callback *rdy)
1670{
1671        struct module *owner;
1672        unsigned long flags;
1673        int err = -EALREADY;
1674
1675        if (crng_ready())
1676                return err;
1677
1678        owner = rdy->owner;
1679        if (!try_module_get(owner))
1680                return -ENOENT;
1681
1682        spin_lock_irqsave(&random_ready_list_lock, flags);
1683        if (crng_ready())
1684                goto out;
1685
1686        owner = NULL;
1687
1688        list_add(&rdy->list, &random_ready_list);
1689        err = 0;
1690
1691out:
1692        spin_unlock_irqrestore(&random_ready_list_lock, flags);
1693
1694        module_put(owner);
1695
1696        return err;
1697}
1698EXPORT_SYMBOL(add_random_ready_callback);
1699
1700/*
1701 * Delete a previously registered readiness callback function.
1702 */
1703void del_random_ready_callback(struct random_ready_callback *rdy)
1704{
1705        unsigned long flags;
1706        struct module *owner = NULL;
1707
1708        spin_lock_irqsave(&random_ready_list_lock, flags);
1709        if (!list_empty(&rdy->list)) {
1710                list_del_init(&rdy->list);
1711                owner = rdy->owner;
1712        }
1713        spin_unlock_irqrestore(&random_ready_list_lock, flags);
1714
1715        module_put(owner);
1716}
1717EXPORT_SYMBOL(del_random_ready_callback);
1718
1719/*
1720 * This function will use the architecture-specific hardware random
1721 * number generator if it is available.  The arch-specific hw RNG will
1722 * almost certainly be faster than what we can do in software, but it
1723 * is impossible to verify that it is implemented securely (as
1724 * opposed, to, say, the AES encryption of a sequence number using a
1725 * key known by the NSA).  So it's useful if we need the speed, but
1726 * only if we're willing to trust the hardware manufacturer not to
1727 * have put in a back door.
1728 */
1729void get_random_bytes_arch(void *buf, int nbytes)
1730{
1731        char *p = buf;
1732
1733        trace_get_random_bytes_arch(nbytes, _RET_IP_);
1734        while (nbytes) {
1735                unsigned long v;
1736                int chunk = min(nbytes, (int)sizeof(unsigned long));
1737
1738                if (!arch_get_random_long(&v))
1739                        break;
1740                
1741                memcpy(p, &v, chunk);
1742                p += chunk;
1743                nbytes -= chunk;
1744        }
1745
1746        if (nbytes)
1747                get_random_bytes(p, nbytes);
1748}
1749EXPORT_SYMBOL(get_random_bytes_arch);
1750
1751
1752/*
1753 * init_std_data - initialize pool with system data
1754 *
1755 * @r: pool to initialize
1756 *
1757 * This function clears the pool's entropy count and mixes some system
1758 * data into the pool to prepare it for use. The pool is not cleared
1759 * as that can only decrease the entropy in the pool.
1760 */
1761static void init_std_data(struct entropy_store *r)
1762{
1763        int i;
1764        ktime_t now = ktime_get_real();
1765        unsigned long rv;
1766
1767        r->last_pulled = jiffies;
1768        mix_pool_bytes(r, &now, sizeof(now));
1769        for (i = r->poolinfo->poolbytes; i > 0; i -= sizeof(rv)) {
1770                if (!arch_get_random_seed_long(&rv) &&
1771                    !arch_get_random_long(&rv))
1772                        rv = random_get_entropy();
1773                mix_pool_bytes(r, &rv, sizeof(rv));
1774        }
1775        mix_pool_bytes(r, utsname(), sizeof(*(utsname())));
1776}
1777
1778/*
1779 * Note that setup_arch() may call add_device_randomness()
1780 * long before we get here. This allows seeding of the pools
1781 * with some platform dependent data very early in the boot
1782 * process. But it limits our options here. We must use
1783 * statically allocated structures that already have all
1784 * initializations complete at compile time. We should also
1785 * take care not to overwrite the precious per platform data
1786 * we were given.
1787 */
1788static int rand_initialize(void)
1789{
1790        init_std_data(&input_pool);
1791        init_std_data(&blocking_pool);
1792        crng_initialize(&primary_crng);
1793        crng_global_init_time = jiffies;
1794        if (ratelimit_disable) {
1795                urandom_warning.interval = 0;
1796                unseeded_warning.interval = 0;
1797        }
1798        return 0;
1799}
1800early_initcall(rand_initialize);
1801
1802#ifdef CONFIG_BLOCK
1803void rand_initialize_disk(struct gendisk *disk)
1804{
1805        struct timer_rand_state *state;
1806
1807        /*
1808         * If kzalloc returns null, we just won't use that entropy
1809         * source.
1810         */
1811        state = kzalloc(sizeof(struct timer_rand_state), GFP_KERNEL);
1812        if (state) {
1813                state->last_time = INITIAL_JIFFIES;
1814                disk->random = state;
1815        }
1816}
1817#endif
1818
1819static ssize_t
1820_random_read(int nonblock, char __user *buf, size_t nbytes)
1821{
1822        ssize_t n;
1823
1824        if (nbytes == 0)
1825                return 0;
1826
1827        nbytes = min_t(size_t, nbytes, SEC_XFER_SIZE);
1828        while (1) {
1829                n = extract_entropy_user(&blocking_pool, buf, nbytes);
1830                if (n < 0)
1831                        return n;
1832                trace_random_read(n*8, (nbytes-n)*8,
1833                                  ENTROPY_BITS(&blocking_pool),
1834                                  ENTROPY_BITS(&input_pool));
1835                if (n > 0)
1836                        return n;
1837
1838                /* Pool is (near) empty.  Maybe wait and retry. */
1839                if (nonblock)
1840                        return -EAGAIN;
1841
1842                wait_event_interruptible(random_read_wait,
1843                        ENTROPY_BITS(&input_pool) >=
1844                        random_read_wakeup_bits);
1845                if (signal_pending(current))
1846                        return -ERESTARTSYS;
1847        }
1848}
1849
1850static ssize_t
1851random_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1852{
1853        return _random_read(file->f_flags & O_NONBLOCK, buf, nbytes);
1854}
1855
1856static ssize_t
1857urandom_read(struct file *file, char __user *buf, size_t nbytes, loff_t *ppos)
1858{
1859        unsigned long flags;
1860        static int maxwarn = 10;
1861        int ret;
1862
1863        if (!crng_ready() && maxwarn > 0) {
1864                maxwarn--;
1865                if (__ratelimit(&urandom_warning))
1866                        printk(KERN_NOTICE "random: %s: uninitialized "
1867                               "urandom read (%zd bytes read)\n",
1868                               current->comm, nbytes);
1869                spin_lock_irqsave(&primary_crng.lock, flags);
1870                crng_init_cnt = 0;
1871                spin_unlock_irqrestore(&primary_crng.lock, flags);
1872        }
1873        nbytes = min_t(size_t, nbytes, INT_MAX >> (ENTROPY_SHIFT + 3));
1874        ret = extract_crng_user(buf, nbytes);
1875        trace_urandom_read(8 * nbytes, 0, ENTROPY_BITS(&input_pool));
1876        return ret;
1877}
1878
1879static __poll_t
1880random_poll(struct file *file, poll_table * wait)
1881{
1882        __poll_t mask;
1883
1884        poll_wait(file, &random_read_wait, wait);
1885        poll_wait(file, &random_write_wait, wait);
1886        mask = 0;
1887        if (ENTROPY_BITS(&input_pool) >= random_read_wakeup_bits)
1888                mask |= EPOLLIN | EPOLLRDNORM;
1889        if (ENTROPY_BITS(&input_pool) < random_write_wakeup_bits)
1890                mask |= EPOLLOUT | EPOLLWRNORM;
1891        return mask;
1892}
1893
1894static int
1895write_pool(struct entropy_store *r, const char __user *buffer, size_t count)
1896{
1897        size_t bytes;
1898        __u32 buf[16];
1899        const char __user *p = buffer;
1900
1901        while (count > 0) {
1902                bytes = min(count, sizeof(buf));
1903                if (copy_from_user(&buf, p, bytes))
1904                        return -EFAULT;
1905
1906                count -= bytes;
1907                p += bytes;
1908
1909                mix_pool_bytes(r, buf, bytes);
1910                cond_resched();
1911        }
1912
1913        return 0;
1914}
1915
1916static ssize_t random_write(struct file *file, const char __user *buffer,
1917                            size_t count, loff_t *ppos)
1918{
1919        size_t ret;
1920
1921        ret = write_pool(&input_pool, buffer, count);
1922        if (ret)
1923                return ret;
1924
1925        return (ssize_t)count;
1926}
1927
1928static long random_ioctl(struct file *f, unsigned int cmd, unsigned long arg)
1929{
1930        int size, ent_count;
1931        int __user *p = (int __user *)arg;
1932        int retval;
1933
1934        switch (cmd) {
1935        case RNDGETENTCNT:
1936                /* inherently racy, no point locking */
1937                ent_count = ENTROPY_BITS(&input_pool);
1938                if (put_user(ent_count, p))
1939                        return -EFAULT;
1940                return 0;
1941        case RNDADDTOENTCNT:
1942                if (!capable(CAP_SYS_ADMIN))
1943                        return -EPERM;
1944                if (get_user(ent_count, p))
1945                        return -EFAULT;
1946                return credit_entropy_bits_safe(&input_pool, ent_count);
1947        case RNDADDENTROPY:
1948                if (!capable(CAP_SYS_ADMIN))
1949                        return -EPERM;
1950                if (get_user(ent_count, p++))
1951                        return -EFAULT;
1952                if (ent_count < 0)
1953                        return -EINVAL;
1954                if (get_user(size, p++))
1955                        return -EFAULT;
1956                retval = write_pool(&input_pool, (const char __user *)p,
1957                                    size);
1958                if (retval < 0)
1959                        return retval;
1960                return credit_entropy_bits_safe(&input_pool, ent_count);
1961        case RNDZAPENTCNT:
1962        case RNDCLEARPOOL:
1963                /*
1964                 * Clear the entropy pool counters. We no longer clear
1965                 * the entropy pool, as that's silly.
1966                 */
1967                if (!capable(CAP_SYS_ADMIN))
1968                        return -EPERM;
1969                input_pool.entropy_count = 0;
1970                blocking_pool.entropy_count = 0;
1971                return 0;
1972        case RNDRESEEDCRNG:
1973                if (!capable(CAP_SYS_ADMIN))
1974                        return -EPERM;
1975                if (crng_init < 2)
1976                        return -ENODATA;
1977                crng_reseed(&primary_crng, NULL);
1978                crng_global_init_time = jiffies - 1;
1979                return 0;
1980        default:
1981                return -EINVAL;
1982        }
1983}
1984
1985static int random_fasync(int fd, struct file *filp, int on)
1986{
1987        return fasync_helper(fd, filp, on, &fasync);
1988}
1989
1990const struct file_operations random_fops = {
1991        .read  = random_read,
1992        .write = random_write,
1993        .poll  = random_poll,
1994        .unlocked_ioctl = random_ioctl,
1995        .fasync = random_fasync,
1996        .llseek = noop_llseek,
1997};
1998
1999const struct file_operations urandom_fops = {
2000        .read  = urandom_read,
2001        .write = random_write,
2002        .unlocked_ioctl = random_ioctl,
2003        .fasync = random_fasync,
2004        .llseek = noop_llseek,
2005};
2006
2007SYSCALL_DEFINE3(getrandom, char __user *, buf, size_t, count,
2008                unsigned int, flags)
2009{
2010        int ret;
2011
2012        if (flags & ~(GRND_NONBLOCK|GRND_RANDOM))
2013                return -EINVAL;
2014
2015        if (count > INT_MAX)
2016                count = INT_MAX;
2017
2018        if (flags & GRND_RANDOM)
2019                return _random_read(flags & GRND_NONBLOCK, buf, count);
2020
2021        if (!crng_ready()) {
2022                if (flags & GRND_NONBLOCK)
2023                        return -EAGAIN;
2024                ret = wait_for_random_bytes();
2025                if (unlikely(ret))
2026                        return ret;
2027        }
2028        return urandom_read(NULL, buf, count, NULL);
2029}
2030
2031/********************************************************************
2032 *
2033 * Sysctl interface
2034 *
2035 ********************************************************************/
2036
2037#ifdef CONFIG_SYSCTL
2038
2039#include <linux/sysctl.h>
2040
2041static int min_read_thresh = 8, min_write_thresh;
2042static int max_read_thresh = OUTPUT_POOL_WORDS * 32;
2043static int max_write_thresh = INPUT_POOL_WORDS * 32;
2044static int random_min_urandom_seed = 60;
2045static char sysctl_bootid[16];
2046
2047/*
2048 * This function is used to return both the bootid UUID, and random
2049 * UUID.  The difference is in whether table->data is NULL; if it is,
2050 * then a new UUID is generated and returned to the user.
2051 *
2052 * If the user accesses this via the proc interface, the UUID will be
2053 * returned as an ASCII string in the standard UUID format; if via the
2054 * sysctl system call, as 16 bytes of binary data.
2055 */
2056static int proc_do_uuid(struct ctl_table *table, int write,
2057                        void __user *buffer, size_t *lenp, loff_t *ppos)
2058{
2059        struct ctl_table fake_table;
2060        unsigned char buf[64], tmp_uuid[16], *uuid;
2061
2062        uuid = table->data;
2063        if (!uuid) {
2064                uuid = tmp_uuid;
2065                generate_random_uuid(uuid);
2066        } else {
2067                static DEFINE_SPINLOCK(bootid_spinlock);
2068
2069                spin_lock(&bootid_spinlock);
2070                if (!uuid[8])
2071                        generate_random_uuid(uuid);
2072                spin_unlock(&bootid_spinlock);
2073        }
2074
2075        sprintf(buf, "%pU", uuid);
2076
2077        fake_table.data = buf;
2078        fake_table.maxlen = sizeof(buf);
2079
2080        return proc_dostring(&fake_table, write, buffer, lenp, ppos);
2081}
2082
2083/*
2084 * Return entropy available scaled to integral bits
2085 */
2086static int proc_do_entropy(struct ctl_table *table, int write,
2087                           void __user *buffer, size_t *lenp, loff_t *ppos)
2088{
2089        struct ctl_table fake_table;
2090        int entropy_count;
2091
2092        entropy_count = *(int *)table->data >> ENTROPY_SHIFT;
2093
2094        fake_table.data = &entropy_count;
2095        fake_table.maxlen = sizeof(entropy_count);
2096
2097        return proc_dointvec(&fake_table, write, buffer, lenp, ppos);
2098}
2099
2100static int sysctl_poolsize = INPUT_POOL_WORDS * 32;
2101extern struct ctl_table random_table[];
2102struct ctl_table random_table[] = {
2103        {
2104                .procname       = "poolsize",
2105                .data           = &sysctl_poolsize,
2106                .maxlen         = sizeof(int),
2107                .mode           = 0444,
2108                .proc_handler   = proc_dointvec,
2109        },
2110        {
2111                .procname       = "entropy_avail",
2112                .maxlen         = sizeof(int),
2113                .mode           = 0444,
2114                .proc_handler   = proc_do_entropy,
2115                .data           = &input_pool.entropy_count,
2116        },
2117        {
2118                .procname       = "read_wakeup_threshold",
2119                .data           = &random_read_wakeup_bits,
2120                .maxlen         = sizeof(int),
2121                .mode           = 0644,
2122                .proc_handler   = proc_dointvec_minmax,
2123                .extra1         = &min_read_thresh,
2124                .extra2         = &max_read_thresh,
2125        },
2126        {
2127                .procname       = "write_wakeup_threshold",
2128                .data           = &random_write_wakeup_bits,
2129                .maxlen         = sizeof(int),
2130                .mode           = 0644,
2131                .proc_handler   = proc_dointvec_minmax,
2132                .extra1         = &min_write_thresh,
2133                .extra2         = &max_write_thresh,
2134        },
2135        {
2136                .procname       = "urandom_min_reseed_secs",
2137                .data           = &random_min_urandom_seed,
2138                .maxlen         = sizeof(int),
2139                .mode           = 0644,
2140                .proc_handler   = proc_dointvec,
2141        },
2142        {
2143                .procname       = "boot_id",
2144                .data           = &sysctl_bootid,
2145                .maxlen         = 16,
2146                .mode           = 0444,
2147                .proc_handler   = proc_do_uuid,
2148        },
2149        {
2150                .procname       = "uuid",
2151                .maxlen         = 16,
2152                .mode           = 0444,
2153                .proc_handler   = proc_do_uuid,
2154        },
2155#ifdef ADD_INTERRUPT_BENCH
2156        {
2157                .procname       = "add_interrupt_avg_cycles",
2158                .data           = &avg_cycles,
2159                .maxlen         = sizeof(avg_cycles),
2160                .mode           = 0444,
2161                .proc_handler   = proc_doulongvec_minmax,
2162        },
2163        {
2164                .procname       = "add_interrupt_avg_deviation",
2165                .data           = &avg_deviation,
2166                .maxlen         = sizeof(avg_deviation),
2167                .mode           = 0444,
2168                .proc_handler   = proc_doulongvec_minmax,
2169        },
2170#endif
2171        { }
2172};
2173#endif  /* CONFIG_SYSCTL */
2174
2175struct batched_entropy {
2176        union {
2177                u64 entropy_u64[CHACHA20_BLOCK_SIZE / sizeof(u64)];
2178                u32 entropy_u32[CHACHA20_BLOCK_SIZE / sizeof(u32)];
2179        };
2180        unsigned int position;
2181};
2182static rwlock_t batched_entropy_reset_lock = __RW_LOCK_UNLOCKED(batched_entropy_reset_lock);
2183
2184/*
2185 * Get a random word for internal kernel use only. The quality of the random
2186 * number is either as good as RDRAND or as good as /dev/urandom, with the
2187 * goal of being quite fast and not depleting entropy. In order to ensure
2188 * that the randomness provided by this function is okay, the function
2189 * wait_for_random_bytes() should be called and return 0 at least once
2190 * at any point prior.
2191 */
2192static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u64);
2193u64 get_random_u64(void)
2194{
2195        u64 ret;
2196        bool use_lock;
2197        unsigned long flags = 0;
2198        struct batched_entropy *batch;
2199        static void *previous;
2200
2201#if BITS_PER_LONG == 64
2202        if (arch_get_random_long((unsigned long *)&ret))
2203                return ret;
2204#else
2205        if (arch_get_random_long((unsigned long *)&ret) &&
2206            arch_get_random_long((unsigned long *)&ret + 1))
2207            return ret;
2208#endif
2209
2210        warn_unseeded_randomness(&previous);
2211
2212        use_lock = READ_ONCE(crng_init) < 2;
2213        batch = &get_cpu_var(batched_entropy_u64);
2214        if (use_lock)
2215                read_lock_irqsave(&batched_entropy_reset_lock, flags);
2216        if (batch->position % ARRAY_SIZE(batch->entropy_u64) == 0) {
2217                extract_crng((__u32 *)batch->entropy_u64);
2218                batch->position = 0;
2219        }
2220        ret = batch->entropy_u64[batch->position++];
2221        if (use_lock)
2222                read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2223        put_cpu_var(batched_entropy_u64);
2224        return ret;
2225}
2226EXPORT_SYMBOL(get_random_u64);
2227
2228static DEFINE_PER_CPU(struct batched_entropy, batched_entropy_u32);
2229u32 get_random_u32(void)
2230{
2231        u32 ret;
2232        bool use_lock;
2233        unsigned long flags = 0;
2234        struct batched_entropy *batch;
2235        static void *previous;
2236
2237        if (arch_get_random_int(&ret))
2238                return ret;
2239
2240        warn_unseeded_randomness(&previous);
2241
2242        use_lock = READ_ONCE(crng_init) < 2;
2243        batch = &get_cpu_var(batched_entropy_u32);
2244        if (use_lock)
2245                read_lock_irqsave(&batched_entropy_reset_lock, flags);
2246        if (batch->position % ARRAY_SIZE(batch->entropy_u32) == 0) {
2247                extract_crng(batch->entropy_u32);
2248                batch->position = 0;
2249        }
2250        ret = batch->entropy_u32[batch->position++];
2251        if (use_lock)
2252                read_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2253        put_cpu_var(batched_entropy_u32);
2254        return ret;
2255}
2256EXPORT_SYMBOL(get_random_u32);
2257
2258/* It's important to invalidate all potential batched entropy that might
2259 * be stored before the crng is initialized, which we can do lazily by
2260 * simply resetting the counter to zero so that it's re-extracted on the
2261 * next usage. */
2262static void invalidate_batched_entropy(void)
2263{
2264        int cpu;
2265        unsigned long flags;
2266
2267        write_lock_irqsave(&batched_entropy_reset_lock, flags);
2268        for_each_possible_cpu (cpu) {
2269                per_cpu_ptr(&batched_entropy_u32, cpu)->position = 0;
2270                per_cpu_ptr(&batched_entropy_u64, cpu)->position = 0;
2271        }
2272        write_unlock_irqrestore(&batched_entropy_reset_lock, flags);
2273}
2274
2275/**
2276 * randomize_page - Generate a random, page aligned address
2277 * @start:      The smallest acceptable address the caller will take.
2278 * @range:      The size of the area, starting at @start, within which the
2279 *              random address must fall.
2280 *
2281 * If @start + @range would overflow, @range is capped.
2282 *
2283 * NOTE: Historical use of randomize_range, which this replaces, presumed that
2284 * @start was already page aligned.  We now align it regardless.
2285 *
2286 * Return: A page aligned address within [start, start + range).  On error,
2287 * @start is returned.
2288 */
2289unsigned long
2290randomize_page(unsigned long start, unsigned long range)
2291{
2292        if (!PAGE_ALIGNED(start)) {
2293                range -= PAGE_ALIGN(start) - start;
2294                start = PAGE_ALIGN(start);
2295        }
2296
2297        if (start > ULONG_MAX - range)
2298                range = ULONG_MAX - start;
2299
2300        range >>= PAGE_SHIFT;
2301
2302        if (range == 0)
2303                return start;
2304
2305        return start + (get_random_long() % range << PAGE_SHIFT);
2306}
2307
2308/* Interface for in-kernel drivers of true hardware RNGs.
2309 * Those devices may produce endless random bits and will be throttled
2310 * when our pool is full.
2311 */
2312void add_hwgenerator_randomness(const char *buffer, size_t count,
2313                                size_t entropy)
2314{
2315        struct entropy_store *poolp = &input_pool;
2316
2317        if (unlikely(crng_init == 0)) {
2318                crng_fast_load(buffer, count);
2319                return;
2320        }
2321
2322        /* Suspend writing if we're above the trickle threshold.
2323         * We'll be woken up again once below random_write_wakeup_thresh,
2324         * or when the calling thread is about to terminate.
2325         */
2326        wait_event_interruptible(random_write_wait, kthread_should_stop() ||
2327                        ENTROPY_BITS(&input_pool) <= random_write_wakeup_bits);
2328        mix_pool_bytes(poolp, buffer, count);
2329        credit_entropy_bits(poolp, entropy);
2330}
2331EXPORT_SYMBOL_GPL(add_hwgenerator_randomness);
2332