Skip to contentSkip to conversation

af_alg broken in 3.12

PATCH New Torsten Duwe 1 participant 3 messages Feb 1, 2017 – Feb 14, 2017

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi Herbert,

you sent a backport of 6de62f15b581f920ade22d758f4c338311c2f0d4 to be included in the 3.12 branch (as b2a0707817d3dec83652bb460a7775613058ae), but this leaves af_alg broken for unkeyed hash functions:

f382cd5ac26674877143fa7d9c0ea23c6640e706 (3.12 just before your commit) :

socket(PF_ALG, SOCK_SEQPACKET, 0) = 3
bind(3, {sa_family=AF_ALG, sa_data="hash\0\0\0\0\0\0\0\0\0\0"}, 88) = 0
accept(3, 0, NULL) = 4
write(4, "abc", 3) = 3
read(4, "\220\1P\230<\322O\260\326\226?}(\341\177r", 16) = 16

and with b2a0707817d3dec83652bb460a7775613058ae applied:

socket(PF_ALG, SOCK_SEQPACKET, 0) = 3
bind(3, {sa_family=AF_ALG, sa_data="hash\0\0\0\0\0\0\0\0\0\0"}, 88) = 0

accept(3, 0, NULL)                      = 4
write(4, "abc", 3)                      = -1 ENOKEY (Required key not available)
read(4, 0x7ffebeba0e30, 16)             = -1 ENOKEY (Required key not available)

Mainline has meanwhile seen many fixes to this change; can you suggest an elegant and crisp backport for these as well?

TIA,
Torsten

#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <linux/if_alg.h>
#include <errno.h>

int main(void)
{
int opfd;
int tfmfd;

  struct sockaddr_alg sa = {
    .salg_family = AF_ALG,
    .salg_type = "hash",
    .salg_name = "md5"

};
char buf[20];
int i;

  tfmfd = socket(AF_ALG, SOCK_SEQPACKET, 0);

  bind(tfmfd, (struct sockaddr *)&sa, sizeof(sa));

  opfd = accept(tfmfd, NULL, 0);

  write(opfd, "abc", 3);
  if (read(opfd, buf, 16) == -1) {
    printf("ERROR: %d\n", errno);
    exit(-1);
  }

  for (i = 0; i < 16; i++) {
    printf("%02x", (unsigned char)buf[i]);
  }
  printf("\n");

  close(opfd);
  close(tfmfd);

return 0;
}

#1 lore raw
Quoted 5 lines · Torsten Duwe

Hi Herbert,

you sent a backport of 6de62f15b581f920ade22d758f4c338311c2f0d4 to be included in the 3.12 branch (as b2a0707817d3dec83652bb460a7775613058ae), but this leaves af_alg broken for unkeyed hash functions:

[...]

Quoted 3 lines

Mainline has meanwhile seen many fixes to this change; can you suggest an elegant and crisp backport for these as well?

If Herbert does not have a better idea, I suggest to back out this change and fix dynamically allocated key structures for the individual algorithms instead, for the older branches.

Any other suggestions?

Torsten

#2 lore raw

PATCH 1/1 Fix af_alg in 3.12

Quoted 4 lines · Torsten Duwe

If Herbert does not have a better idea, I suggest to back out this change and fix dynamically allocated key structures for the individual algorithms instead, for the older branches.

So, the solution IMHO is to revert

b2a0707817d3dec83652bb460a7775613058aedd
f382cd5ac26674877143fa7d9c0ea23c6640e706
c25e22ff51d3bebf579a054aecbaa98c81149c02
1e3f8a31f01e5967fcf413d72832ce41aa4efd1d
79adba68c32883c6559dc80040e97c35e208c7f1
82a0aa2c08de674191cf5e99b649af145c5ade25

that is the whole sequence from 82a0aa2c08de674191..b2a0707817d3dec836 backwards sans b9da7c51a11a2e7 which is a separate, unrelated fix.

This series aimed to fix the whole class of problems and creates and fixes its own ABI breakage. But since that problem class contains exactly 1 element, fix instead the one broken algorithm with a dynamically allocated member in its context that gets initialised during setkey.

crypto/lrw.c +6 -0
index 6f9908a..4c9257d 100644
@@ -150,6 +150,9 @@ static int crypt(struct blkcipher_desc *d,
 	u8 *wsrc;
 	u8 *wdst;
 
+	if (!(ctx->table.table))
+		return -ENOKEY;
+
 	err = blkcipher_walk_virt(d, w);
 	if (!(avail = w->nbytes))
 		return err;
@@ -229,6 +232,9 @@ int lrw_crypt(struct blkcipher_desc *desc, struct scatterlist *sdst,
 
 	BUG_ON(max_blks < 1);
 
+	if (!ctx->table)
+		return -ENOKEY;
+
 	blkcipher_walk_init(&walk, sdst, ssrc, nbytes);
 
 	err = blkcipher_walk_virt(desc, &walk);