[7/8] os-posix: Provide new -runasid option

This allows the caller to specify a uid and gid to use, even if there
is no corresponding password entry.  This will be useful in certain
Xen configurations.

Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
---
v3: Error messages fixed.  Thanks to Peter Maydell and Ross Lagerwall.
v2: Coding style fixes.
---
 os-posix.c      | 49 ++++++++++++++++++++++++++++++++++++++++---------
 qemu-options.hx | 12 ++++++++++++
 2 files changed, 52 insertions(+), 9 deletions(-)

Sorry for the slooooow response.

Ian Jackson <ian.jackson@eu.citrix.com> writes:

> This allows the caller to specify a uid and gid to use, even if there
> is no corresponding password entry.  This will be useful in certain
> Xen configurations.
>
> Signed-off-by: Ian Jackson <Ian.Jackson@eu.citrix.com>
> ---
> v3: Error messages fixed.  Thanks to Peter Maydell and Ross Lagerwall.
> v2: Coding style fixes.
> ---
>  os-posix.c      | 49 ++++++++++++++++++++++++++++++++++++++++---------
>  qemu-options.hx | 12 ++++++++++++
>  2 files changed, 52 insertions(+), 9 deletions(-)
>
> diff --git a/os-posix.c b/os-posix.c
> index 92e9d85..6cc5868 100644
> --- a/os-posix.c
> +++ b/os-posix.c
> @@ -43,6 +43,8 @@
>  #endif
>  
>  static struct passwd *user_pwd;
> +static uid_t user_uid = (uid_t)-1;
> +static gid_t user_gid = (gid_t)-1;
>  static const char *chroot_dir;
>  static int daemonize;
>  static int daemon_pipe;
> @@ -134,6 +136,9 @@ void os_set_proc_name(const char *s)
>   */
>  void os_parse_cmd_args(int index, const char *optarg)
>  {
> +    unsigned long lv;
> +    char *ep;
> +    int rc;
>      switch (index) {
>  #ifdef CONFIG_SLIRP
>      case QEMU_OPTION_smb:
> @@ -150,6 +155,22 @@ void os_parse_cmd_args(int index, const char *optarg)
>              exit(1);
>          }
>          break;
> +    case QEMU_OPTION_runasid:
> +        errno = 0;
> +        lv = strtoul(optarg, &ep, 0); /* can't qemu_strtoul, want *ep=='.' */

I'm afraid I can't see why qemu_strtoul() wouldn't work here.  Can you
enlighten me?

> +        user_uid = lv; /* overflow here is ID in C99 */

I don't get the comment.  You check for overflow the obvious way below.
Sure you need a comment?

> +        if (errno || *ep != '.' || user_uid != lv || user_uid == (uid_t)-1) {
> +            fprintf(stderr, "Could not obtain uid from \"%s\"", optarg);
> +            exit(1);
> +        }
> +        lv = 0;
> +        rc = qemu_strtoul(ep + 1, 0, 0, &lv);
> +        user_gid = lv; /* overflow here is ID in C99 */

Likewise.

> +        if (rc || user_gid != lv || user_gid == (gid_t)-1) {
> +            fprintf(stderr, "Could not obtain gid from \"%s\"", optarg);
> +            exit(1);
> +        }
> +        break;
>      case QEMU_OPTION_chroot:
>          chroot_dir = optarg;
>          break;
> @@ -166,18 +187,28 @@ void os_parse_cmd_args(int index, const char *optarg)
>  
>  static void change_process_uid(void)
>  {
> -    if (user_pwd) {
> -        if (setgid(user_pwd->pw_gid) < 0) {
> -            fprintf(stderr, "Failed to setgid(%d)\n", user_pwd->pw_gid);
> +    if (user_pwd || user_uid != (uid_t)-1) {
> +        gid_t intended_gid = user_pwd ? user_pwd->pw_gid : user_gid;
> +        uid_t intended_uid = user_pwd ? user_pwd->pw_uid : user_uid;
> +        if (setgid(intended_gid) < 0) {
> +            fprintf(stderr, "Failed to setgid(%d)\n", intended_gid);
>              exit(1);
>          }
> -        if (initgroups(user_pwd->pw_name, user_pwd->pw_gid) < 0) {
> -            fprintf(stderr, "Failed to initgroups(\"%s\", %d)\n",
> -                    user_pwd->pw_name, user_pwd->pw_gid);
> -            exit(1);
> +        if (user_pwd) {
> +            if (initgroups(user_pwd->pw_name, user_pwd->pw_gid) < 0) {
> +                fprintf(stderr, "Failed to initgroups(\"%s\", %d)\n",
> +                        user_pwd->pw_name, user_pwd->pw_gid);
> +                exit(1);
> +            }
> +        } else {
> +            if (setgroups(1, &user_gid) < 0) {
> +                fprintf(stderr, "Failed to setgroups(1, [%d])",
> +                        user_gid);
> +                exit(1);
> +            }
>          }
> -        if (setuid(user_pwd->pw_uid) < 0) {
> -            fprintf(stderr, "Failed to setuid(%d)\n", user_pwd->pw_uid);
> +        if (setuid(intended_uid) < 0) {
> +            fprintf(stderr, "Failed to setuid(%d)\n", intended_uid);
>              exit(1);
>          }
>          if (setuid(0) != -1) {
> diff --git a/qemu-options.hx b/qemu-options.hx
> index 9f6e2ad..34a5329 100644
> --- a/qemu-options.hx
> +++ b/qemu-options.hx
> @@ -3968,6 +3968,18 @@ Immediately before starting guest execution, drop root privileges, switching
>  to the specified user.
>  ETEXI
>  
> +#ifndef _WIN32
> +DEF("runasid", HAS_ARG, QEMU_OPTION_runasid, \
> +    "-runasid uid.gid     change to numeric uid and gid just before starting the VM\n",
> +    QEMU_ARCH_ALL)
> +#endif
> +STEXI
> +@item -runasid @var{uid}.@var{gid}

Didn't we agree on using ':' instead of '.' as separator?

Sure we need yet another option?  Why can't we compatibly extend -runas?

If we truly need both, the rationale belongs into the commit message,
and we need to consider how they interact.  I'd recommend left-to-right
semantics, i.e. if you specify both, the last one wins.  Not what your
current code does, if I read it correctly.

> +@findex -runasid
> +Immediately before starting guest execution, drop root privileges, switching
> +to the specified uid and gid.
> +ETEXI
> +
>  DEF("prom-env", HAS_ARG, QEMU_OPTION_prom_env,
>      "-prom-env variable=value\n"
>      "                set OpenBIOS nvram variables\n",

Hi.  Thanks for the (re)-review.

Markus Armbruster writes ("Re: [Qemu-devel] [PATCH 7/8] os-posix: Provide new -runasid option"):
> Ian Jackson <ian.jackson@eu.citrix.com> writes:
> > +    case QEMU_OPTION_runasid:
> > +        errno = 0;
> > +        lv = strtoul(optarg, &ep, 0); /* can't qemu_strtoul, want *ep=='.' */
> 
> I'm afraid I can't see why qemu_strtoul() wouldn't work here.  Can you
> enlighten me?

qemu_strtoul fails (returns an error) if the delimiter (that is, the
first character which is not processed as digit by strtoul) is not
'\0'.  Normally this is desirable, because it correctly rejects
strings like "123blather".  But here we are trying to process a string
whose first non-digit is ':', and we will handle the tail part
explicitly.

It would be possible to use strchr and then to write a '\0' over the
':' but I dislike that processing style (and it is forbidden by the
const annotations on os_parse_cmd_args etc.)

> > +        user_uid = lv; /* overflow here is ID in C99 */
> 
> I don't get the comment.  You check for overflow the obvious way below.
> Sure you need a comment?

This relates to overflow in the assignment, only.  lv is an unsigned
long and user_uid is a uid_t (which may be smaller).  Normally, signed
integer overflow is UB, but this is not the case when converting from
another integer type.

There are two possible overflows: 1. a string that strtoul cannot get
to fit in an unsigned long, which produces a nonzero errno; and, 2. a
value which fits in an unsigned long but not a uid_t.  In the latter
case, we convert it _back_ into an unsigned long, as an implicit
conversion in this middle test:

> > +        if (errno || *ep != '.' || user_uid != lv || user_uid == (uid_t)-1) {

If that succeeds, we know we can round-trip it so it is valid.  The
remaining check needed is that it doesn't round trip to the sentinel
uid value.

Does that all make sense ?  I'm not sure how much of this to document
in comments.  It's deeply annoying that C is such a puzzle language
here and that therefore such complicated reasoning and
not-immediately-obvious code is needed, to do something so simple.

If you would like me to remove the comment, I can drop it, of course.
I don't really mind.

> > +#ifndef _WIN32
> > +DEF("runasid", HAS_ARG, QEMU_OPTION_runasid, \
> > +    "-runasid uid.gid     change to numeric uid and gid just before starting the VM\n",
> > +    QEMU_ARCH_ALL)
> > +#endif
> > +STEXI
> > +@item -runasid @var{uid}.@var{gid}
> 
> Didn't we agree on using ':' instead of '.' as separator?
> 
> Sure we need yet another option?  Why can't we compatibly extend -runas?

For some reason you are looking at an old version of the patch.  That
may be my fault - there were a few mis-posts.  Sorry about that.

The final version does indeed have ':' and does reuse the -runas
option.

> If we truly need both, the rationale belongs into the commit message,
> and we need to consider how they interact.  I'd recommend left-to-right
> semantics, i.e. if you specify both, the last one wins.  Not what your
> current code does, if I read it correctly.

Happily this is now irrelevant.

I will repost this series after I hear from you about strtoul and the
overflow comment.

Regards,
Ian.

Ian Jackson <ian.jackson@eu.citrix.com> writes:

> Hi.  Thanks for the (re)-review.
>
> Markus Armbruster writes ("Re: [Qemu-devel] [PATCH 7/8] os-posix: Provide new -runasid option"):
>> Ian Jackson <ian.jackson@eu.citrix.com> writes:
>> > +    case QEMU_OPTION_runasid:
>> > +        errno = 0;
>> > +        lv = strtoul(optarg, &ep, 0); /* can't qemu_strtoul, want *ep=='.' */
>> 
>> I'm afraid I can't see why qemu_strtoul() wouldn't work here.  Can you
>> enlighten me?
>
> qemu_strtoul fails (returns an error) if the delimiter (that is, the
> first character which is not processed as digit by strtoul) is not
> '\0'.

It does that *only* when its @endptr argument is null.  Since you pass
&ep, it should work fine here.

>        Normally this is desirable, because it correctly rejects
> strings like "123blather".  But here we are trying to process a string
> whose first non-digit is ':', and we will handle the tail part
> explicitly.
>
> It would be possible to use strchr and then to write a '\0' over the
> ':' but I dislike that processing style (and it is forbidden by the
> const annotations on os_parse_cmd_args etc.)

I'd dislike that, too.

>> > +        user_uid = lv; /* overflow here is ID in C99 */
>> 
>> I don't get the comment.  You check for overflow the obvious way below.
>> Sure you need a comment?
>
> This relates to overflow in the assignment, only.  lv is an unsigned
> long and user_uid is a uid_t (which may be smaller).  Normally, signed
> integer overflow is UB, but this is not the case when converting from
> another integer type.
>
> There are two possible overflows: 1. a string that strtoul cannot get
> to fit in an unsigned long, which produces a nonzero errno; and, 2. a
> value which fits in an unsigned long but not a uid_t.  In the latter
> case, we convert it _back_ into an unsigned long, as an implicit
> conversion in this middle test:
>
>> > +        if (errno || *ep != '.' || user_uid != lv || user_uid == (uid_t)-1) {
>
> If that succeeds, we know we can round-trip it so it is valid.  The
> remaining check needed is that it doesn't round trip to the sentinel
> uid value.
>
> Does that all make sense ?

Your code makes sense to me.  It's just the comment that confuses me.
Does ID mean "implementation-defined behavior"?  That would be wrong:

       6.3.1.3  Signed and unsigned integers

       [#1] When a value with integer type is converted to  another
       integer   type  other  than  _Bool,  if  the  value  can  be
       represented by the new type, it is unchanged.

       [#2] Otherwise, if the new type is unsigned,  the  value  is
       converted  by repeatedly adding or subtracting one more than
       the maximum value that can be represented in  the  new  type
       until the value is in the range of the new type.

>                             I'm not sure how much of this to document
> in comments.  It's deeply annoying that C is such a puzzle language
> here and that therefore such complicated reasoning and
> not-immediately-obvious code is needed, to do something so simple.
>
> If you would like me to remove the comment, I can drop it, of course.
> I don't really mind.

I'd drop it.

You might find this variation of your code slightly more obvious, or
slightly less:

        if (errno || *ep != '.' || (uid_t)lv != lv || user_uid == (uid_t)-1) {
            fprintf(stderr, "Could not obtain uid from \"%s\"", optarg);
            exit(1);
        }
        user_uid = lv;

Your choice.

One more thing: please report errors with error_report().  Here:

            error_report("Could not obtain uid");

No need to quote optarg back at the user, because error_report() already
does.

>> > +#ifndef _WIN32
>> > +DEF("runasid", HAS_ARG, QEMU_OPTION_runasid, \
>> > +    "-runasid uid.gid     change to numeric uid and gid just before starting the VM\n",
>> > +    QEMU_ARCH_ALL)
>> > +#endif
>> > +STEXI
>> > +@item -runasid @var{uid}.@var{gid}
>> 
>> Didn't we agree on using ':' instead of '.' as separator?
>> 
>> Sure we need yet another option?  Why can't we compatibly extend -runas?
>
> For some reason you are looking at an old version of the patch.  That
> may be my fault - there were a few mis-posts.  Sorry about that.

It could be just as well my fault: my inbox spun out of control before
KVM Forum.

> The final version does indeed have ':' and does reuse the -runas
> option.
>
>> If we truly need both, the rationale belongs into the commit message,
>> and we need to consider how they interact.  I'd recommend left-to-right
>> semantics, i.e. if you specify both, the last one wins.  Not what your
>> current code does, if I read it correctly.
>
> Happily this is now irrelevant.
>
> I will repost this series after I hear from you about strtoul and the
> overflow comment.

Thanks!

Markus Armbruster writes ("Re: [Qemu-devel] [PATCH 7/8] os-posix: Provide new -runasid option"):
> Ian Jackson <ian.jackson@eu.citrix.com> writes:
> > qemu_strtoul fails (returns an error) if the delimiter (that is, the
> > first character which is not processed as digit by strtoul) is not
> > '\0'.
> 
> It does that *only* when its @endptr argument is null.  Since you pass
> &ep, it should work fine here.

I have just read it again and you are right.  Sorry.  I will fix this
then.

> > Does that all make sense ?
> 
> Your code makes sense to me.  It's just the comment that confuses me.
> Does ID mean "implementation-defined behavior"?  That would be wrong:

Yes, that's what I meant by ID.

>        6.3.1.3  Signed and unsigned integers
> 
>        [#1] When a value with integer type is converted to  another
>        integer   type  other  than  _Bool,  if  the  value  can  be
>        represented by the new type, it is unchanged.
> 
>        [#2] Otherwise, if the new type is unsigned,  the  value  is
>        converted  by repeatedly adding or subtracting one more than
>        the maximum value that can be represented in  the  new  type
>        until the value is in the range of the new type.

That applies if the new type (uid_t, here) is unsigned.  But I think
uid_t's signedness is not specified, so it might be signed.  (SuS
says, in the section on types.h, only

  Additionally:
     * nlink_t, uid_t, gid_t, and id_t shall be integer types.

C99 6.3.1.3 continues:

         Otherwise, the new type is signed and the value cannot be
         represented in it; either the result is
         implementation-defined or an implementation-defined signal is
         raised.

So the type of uid_t is ID too.

> One more thing: please report errors with error_report().  Here:

>             error_report("Could not obtain uid");
> 
> No need to quote optarg back at the user, because error_report() already
> does.

Ah, I will do that.  Thanks.

Regards,
Ian.