Re: [PATCH] Revert "drivers: char: mem: Check {read,write}_kmem() addresses"

From: Greg KH
Date: Wed Jan 04 2017 - 04:21:54 EST


On Tue, Jan 03, 2017 at 09:50:52PM +0100, Jason A. Donenfeld wrote:
> This reverts commit 148a1bc84398039e2b96ff78678c4d9a67f81452.
>
> This commit was intended to fix a problem, but it did not do so
> correctly, resulting in /dev/kmem being entirely broken. So, revert this
> commit until a different solution is found. The following PoC shows
> breakage bisecting to this commit. On working systems, it returns 0 and
> prints the last message. On broken systems containing this commit, it
> exits returning EIO.
>
> #define _FILE_OFFSET_BITS 64
> #include <unistd.h>
> #include <fcntl.h>
> #include <stdio.h>
> #include <string.h>
> #include <stdlib.h>
> #include <errno.h>
> #include <sys/types.h>
> #include <sys/socket.h>
> #include <netinet/in.h>
>
> int main(int argc, char *argv[])
> {
> int fd, sock;
> FILE *f;
> char line[256], *ptr;
> unsigned long addr = 0;
> struct sockaddr_in saddr = { .sin_family = AF_INET, .sin_port = 55193 };
>
> sock = socket(AF_INET, SOCK_DGRAM, 0);
> if (sock < 0) {
> perror("socket");
> return errno;
> }
> if (bind(sock, (struct sockaddr *)&saddr, sizeof(saddr)) < 0) {
> perror("bind");
> return errno;
> }
> f = fopen("/proc/net/udp", "r");
> if (!f) {
> perror("fopen(/proc/net/udp)");
> return errno;
> }
> if (!fgets(line, 256, f) || !fgets(line, 256, f)) {
> perror("fgets");
> return errno;
> }
> fclose(f);
> ptr = line + strlen(line) - 1;
> while (*--ptr == ' ');
> while (*--ptr != ' ');
> while (*(--ptr - 1) != ' ');
> addr = strtoul(ptr, NULL, 16);
>
> printf("Attempting to read from skbuff at 0x%lx\n", addr);
>
> fd = open("/dev/kmem", O_RDONLY);
> if (fd < 0) {
> perror("open(/dev/kmem)");
> return errno;
> }
> if (lseek(fd, addr, SEEK_SET) == (off_t)-1) {
> perror("lseek");
> return errno;
> }
> if (read(fd, &addr, sizeof(addr)) != sizeof(addr)) {
> perror("read");
> return errno;
> }
>
> printf("It worked, reading the value %lx!\n", addr);
> close(fd);
> close(sock);
> return 0;
> }
>
> Signed-off-by: Jason A. Donenfeld <Jason@xxxxxxxxx>
> Cc: Robin Murphy <robin.murphy@xxxxxxx>
> Cc: Kefeng Wang <wangkefeng.wang@xxxxxxxxxx>
> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> ---
> drivers/char/mem.c | 6 ------
> 1 file changed, 6 deletions(-)
>
> diff --git a/drivers/char/mem.c b/drivers/char/mem.c
> index 5bb1985ec484..a33163dbb913 100644
> --- a/drivers/char/mem.c
> +++ b/drivers/char/mem.c
> @@ -381,9 +381,6 @@ static ssize_t read_kmem(struct file *file, char __user *buf,
> char *kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
> int err = 0;
>
> - if (!pfn_valid(PFN_DOWN(p)))
> - return -EIO;
> -
> read = 0;
> if (p < (unsigned long) high_memory) {
> low_count = count;
> @@ -512,9 +509,6 @@ static ssize_t write_kmem(struct file *file, const char __user *buf,
> char *kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
> int err = 0;
>
> - if (!pfn_valid(PFN_DOWN(p)))
> - return -EIO;
> -
> if (p < (unsigned long) high_memory) {
> unsigned long to_write = min_t(unsigned long, count,
> (unsigned long)high_memory - p);

Ick, I hate /dev/kmem...

Robin and Kefeng, any comments? Reverting seems correct at this point
in time, any objection?

thanks,

greg k-h