Revision 472e5b056f000a778abb41f1e443de58eb259783 authored by Linus Torvalds on 02 October 2020, 02:14:36 UTC, committed by Linus Torvalds on 02 October 2020, 02:14:36 UTC
The pipe splice code still used the old model of waiting for pipe IO by
using a non-specific "pipe_wait()" that waited for any pipe event to
happen, which depended on all pipe IO being entirely serialized by the
pipe lock.  So by checking the state you were waiting for, and then
adding yourself to the wait queue before dropping the lock, you were
guaranteed to see all the wakeups.

Strictly speaking, the actual wakeups were not done under the lock, but
the pipe_wait() model still worked, because since the waiter held the
lock when checking whether it should sleep, it would always see the
current state, and the wakeup was always done after updating the state.

However, commit 0ddad21d3e99 ("pipe: use exclusive waits when reading or
writing") split the single wait-queue into two, and in the process also
made the "wait for event" code wait for _two_ wait queues, and that then
showed a race with the wakers that were not serialized by the pipe lock.

It's only splice that used that "pipe_wait()" model, so the problem
wasn't obvious, but Josef Bacik reports:

 "I hit a hang with fstest btrfs/187, which does a btrfs send into
  /dev/null. This works by creating a pipe, the write side is given to
  the kernel to write into, and the read side is handed to a thread that
  splices into a file, in this case /dev/null.

  The box that was hung had the write side stuck here [pipe_write] and
  the read side stuck here [splice_from_pipe_next -> pipe_wait].

  [ more details about pipe_wait() scenario ]

  The problem is we're doing the prepare_to_wait, which sets our state
  each time, however we can be woken up either with reads or writes. In
  the case above we race with the WRITER waking us up, and re-set our
  state to INTERRUPTIBLE, and thus never break out of schedule"

Josef had a patch that avoided the issue in pipe_wait() by just making
it set the state only once, but the deeper problem is that pipe_wait()
depends on a level of synchonization by the pipe mutex that it really
shouldn't.  And the whole "wait for any pipe state change" model really
isn't very good to begin with.

So rather than trying to work around things in pipe_wait(), remove that
legacy model of "wait for arbitrary pipe event" entirely, and actually
create functions that wait for the pipe actually being readable or
writable, and can do so without depending on the pipe lock serializing
everything.

Fixes: 0ddad21d3e99 ("pipe: use exclusive waits when reading or writing")
Link: https://lore.kernel.org/linux-fsdevel/bfa88b5ad6f069b2b679316b9e495a970130416c.1601567868.git.josef@toxicpanda.com/
Reported-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-and-tested-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
1 parent 44b6e23
Raw File
salsa20_generic.c
/*
 * Salsa20: Salsa20 stream cipher algorithm
 *
 * Copyright (c) 2007 Tan Swee Heng <thesweeheng@gmail.com>
 *
 * Derived from:
 * - salsa20.c: Public domain C code by Daniel J. Bernstein <djb@cr.yp.to>
 *
 * Salsa20 is a stream cipher candidate in eSTREAM, the ECRYPT Stream
 * Cipher Project. It is designed by Daniel J. Bernstein <djb@cr.yp.to>.
 * More information about eSTREAM and Salsa20 can be found here:
 *   https://www.ecrypt.eu.org/stream/
 *   https://cr.yp.to/snuffle.html
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by the Free
 * Software Foundation; either version 2 of the License, or (at your option)
 * any later version.
 *
 */

#include <asm/unaligned.h>
#include <crypto/internal/skcipher.h>
#include <linux/module.h>

#define SALSA20_IV_SIZE        8
#define SALSA20_MIN_KEY_SIZE  16
#define SALSA20_MAX_KEY_SIZE  32
#define SALSA20_BLOCK_SIZE    64

struct salsa20_ctx {
	u32 initial_state[16];
};

static void salsa20_block(u32 *state, __le32 *stream)
{
	u32 x[16];
	int i;

	memcpy(x, state, sizeof(x));

	for (i = 0; i < 20; i += 2) {
		x[ 4] ^= rol32((x[ 0] + x[12]),  7);
		x[ 8] ^= rol32((x[ 4] + x[ 0]),  9);
		x[12] ^= rol32((x[ 8] + x[ 4]), 13);
		x[ 0] ^= rol32((x[12] + x[ 8]), 18);
		x[ 9] ^= rol32((x[ 5] + x[ 1]),  7);
		x[13] ^= rol32((x[ 9] + x[ 5]),  9);
		x[ 1] ^= rol32((x[13] + x[ 9]), 13);
		x[ 5] ^= rol32((x[ 1] + x[13]), 18);
		x[14] ^= rol32((x[10] + x[ 6]),  7);
		x[ 2] ^= rol32((x[14] + x[10]),  9);
		x[ 6] ^= rol32((x[ 2] + x[14]), 13);
		x[10] ^= rol32((x[ 6] + x[ 2]), 18);
		x[ 3] ^= rol32((x[15] + x[11]),  7);
		x[ 7] ^= rol32((x[ 3] + x[15]),  9);
		x[11] ^= rol32((x[ 7] + x[ 3]), 13);
		x[15] ^= rol32((x[11] + x[ 7]), 18);
		x[ 1] ^= rol32((x[ 0] + x[ 3]),  7);
		x[ 2] ^= rol32((x[ 1] + x[ 0]),  9);
		x[ 3] ^= rol32((x[ 2] + x[ 1]), 13);
		x[ 0] ^= rol32((x[ 3] + x[ 2]), 18);
		x[ 6] ^= rol32((x[ 5] + x[ 4]),  7);
		x[ 7] ^= rol32((x[ 6] + x[ 5]),  9);
		x[ 4] ^= rol32((x[ 7] + x[ 6]), 13);
		x[ 5] ^= rol32((x[ 4] + x[ 7]), 18);
		x[11] ^= rol32((x[10] + x[ 9]),  7);
		x[ 8] ^= rol32((x[11] + x[10]),  9);
		x[ 9] ^= rol32((x[ 8] + x[11]), 13);
		x[10] ^= rol32((x[ 9] + x[ 8]), 18);
		x[12] ^= rol32((x[15] + x[14]),  7);
		x[13] ^= rol32((x[12] + x[15]),  9);
		x[14] ^= rol32((x[13] + x[12]), 13);
		x[15] ^= rol32((x[14] + x[13]), 18);
	}

	for (i = 0; i < 16; i++)
		stream[i] = cpu_to_le32(x[i] + state[i]);

	if (++state[8] == 0)
		state[9]++;
}

static void salsa20_docrypt(u32 *state, u8 *dst, const u8 *src,
			    unsigned int bytes)
{
	__le32 stream[SALSA20_BLOCK_SIZE / sizeof(__le32)];

	while (bytes >= SALSA20_BLOCK_SIZE) {
		salsa20_block(state, stream);
		crypto_xor_cpy(dst, src, (const u8 *)stream,
			       SALSA20_BLOCK_SIZE);
		bytes -= SALSA20_BLOCK_SIZE;
		dst += SALSA20_BLOCK_SIZE;
		src += SALSA20_BLOCK_SIZE;
	}
	if (bytes) {
		salsa20_block(state, stream);
		crypto_xor_cpy(dst, src, (const u8 *)stream, bytes);
	}
}

static void salsa20_init(u32 *state, const struct salsa20_ctx *ctx,
			 const u8 *iv)
{
	memcpy(state, ctx->initial_state, sizeof(ctx->initial_state));
	state[6] = get_unaligned_le32(iv + 0);
	state[7] = get_unaligned_le32(iv + 4);
}

static int salsa20_setkey(struct crypto_skcipher *tfm, const u8 *key,
			  unsigned int keysize)
{
	static const char sigma[16] = "expand 32-byte k";
	static const char tau[16] = "expand 16-byte k";
	struct salsa20_ctx *ctx = crypto_skcipher_ctx(tfm);
	const char *constants;

	if (keysize != SALSA20_MIN_KEY_SIZE &&
	    keysize != SALSA20_MAX_KEY_SIZE)
		return -EINVAL;

	ctx->initial_state[1] = get_unaligned_le32(key + 0);
	ctx->initial_state[2] = get_unaligned_le32(key + 4);
	ctx->initial_state[3] = get_unaligned_le32(key + 8);
	ctx->initial_state[4] = get_unaligned_le32(key + 12);
	if (keysize == 32) { /* recommended */
		key += 16;
		constants = sigma;
	} else { /* keysize == 16 */
		constants = tau;
	}
	ctx->initial_state[11] = get_unaligned_le32(key + 0);
	ctx->initial_state[12] = get_unaligned_le32(key + 4);
	ctx->initial_state[13] = get_unaligned_le32(key + 8);
	ctx->initial_state[14] = get_unaligned_le32(key + 12);
	ctx->initial_state[0]  = get_unaligned_le32(constants + 0);
	ctx->initial_state[5]  = get_unaligned_le32(constants + 4);
	ctx->initial_state[10] = get_unaligned_le32(constants + 8);
	ctx->initial_state[15] = get_unaligned_le32(constants + 12);

	/* space for the nonce; it will be overridden for each request */
	ctx->initial_state[6] = 0;
	ctx->initial_state[7] = 0;

	/* initial block number */
	ctx->initial_state[8] = 0;
	ctx->initial_state[9] = 0;

	return 0;
}

static int salsa20_crypt(struct skcipher_request *req)
{
	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
	const struct salsa20_ctx *ctx = crypto_skcipher_ctx(tfm);
	struct skcipher_walk walk;
	u32 state[16];
	int err;

	err = skcipher_walk_virt(&walk, req, false);

	salsa20_init(state, ctx, req->iv);

	while (walk.nbytes > 0) {
		unsigned int nbytes = walk.nbytes;

		if (nbytes < walk.total)
			nbytes = round_down(nbytes, walk.stride);

		salsa20_docrypt(state, walk.dst.virt.addr, walk.src.virt.addr,
				nbytes);
		err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
	}

	return err;
}

static struct skcipher_alg alg = {
	.base.cra_name		= "salsa20",
	.base.cra_driver_name	= "salsa20-generic",
	.base.cra_priority	= 100,
	.base.cra_blocksize	= 1,
	.base.cra_ctxsize	= sizeof(struct salsa20_ctx),
	.base.cra_module	= THIS_MODULE,

	.min_keysize		= SALSA20_MIN_KEY_SIZE,
	.max_keysize		= SALSA20_MAX_KEY_SIZE,
	.ivsize			= SALSA20_IV_SIZE,
	.chunksize		= SALSA20_BLOCK_SIZE,
	.setkey			= salsa20_setkey,
	.encrypt		= salsa20_crypt,
	.decrypt		= salsa20_crypt,
};

static int __init salsa20_generic_mod_init(void)
{
	return crypto_register_skcipher(&alg);
}

static void __exit salsa20_generic_mod_fini(void)
{
	crypto_unregister_skcipher(&alg);
}

subsys_initcall(salsa20_generic_mod_init);
module_exit(salsa20_generic_mod_fini);

MODULE_LICENSE("GPL");
MODULE_DESCRIPTION ("Salsa20 stream cipher algorithm");
MODULE_ALIAS_CRYPTO("salsa20");
MODULE_ALIAS_CRYPTO("salsa20-generic");
back to top