[Qemu-devel] [PATCH v10 0/3] linux-user: A set of miscellaneous patches

[Qemu-devel] [PATCH v10 0/3] linux-user: A set of miscellaneous patches

[Aleksandar Markovic]

From: Aleksandar Markovic <amarkovic@wavecomp.com>

This is a collection of misc patches for Linux user that I recently
accumulated from variuous sources. All of them originate from problems
observed on mips target. However, these changes actually affect and fix
problems on multiple targets.

v9->v10:

  - improved commit messages for patches 2 and 3

v8->v9:

  - fixed build error on some systems related to SOL_ALG

v7->v8:

  - added a patch on setsockopt() option SOL_ALG

v6->v7:

  - fixed a build error for older kernels related to the patch on
    setsockopt() options
  - removed four patches that on the meantime got accepted into the
    main source tree

v5->v6:

  - fixed a mistake in patch #4
  - improved commit messages in patches #4 and #6

v4->v5:

  - added the patch on statx() support
  - improved the patch on IPV6_<ADD|DROP>_MEMBERSHIP to take into
    account the possibility of different names for a field
  - minor corrections in commit messages

v3->v4:

  - improved commit messages (fixed some typos, improved relevance)

v2->v3:

  - updated and improved commit messages
  - added IPV6_DROP_MEMBERSHIP support to the patch on setsockopt()'s
    option

v1->v2:

  - added the patch on setsockopt()'s option IPV6_ADD_MEMBERSHIP
  - improved the commit me

Aleksandar Rikalo (1):
  linux-user: Add support for statx() syscall

Neng Chen (1):
  linux-user: Add support for setsockopt() options
    IPV6_<ADD|DROP>_MEMBERSHIP

Yunqiang Su (1):
  linux-user: Add support for setsockopt() option SOL_ALG

 linux-user/syscall.c      | 193 +++++++++++++++++++++++++++++++++++++++++++++-
 linux-user/syscall_defs.h |  37 +++++++++
 2 files changed, 229 insertions(+), 1 deletion(-)

-- 
2.7.4

[Qemu-devel] [PATCH v10 1/3] linux-user: Add support for setsockopt() options IPV6_<ADD|DROP>_MEMBERSHIP

[Aleksandar Markovic]

From: Neng Chen <nchen@wavecomp.com>

Add support for options IPV6_ADD_MEMBERSHIP and IPV6_DROP_MEMPEMBERSHIP
of the syscall setsockopt(). These options control membership in
multicast groups. Their argument is a pointer to a struct ipv6_mreq,
which is in turn defined in IP v6 header netinet/in.h as:

 struct ipv6_mreq {
     /* IPv6 multicast address of group */
     struct  in6_addr  ipv6mr_multiaddr;
     /* local IPv6 address of interface */
     int     ipv6mr_interface;
 };

...whereas its definition in kernel's include/uapi/linux/in6.h is:

 #if __UAPI_DEF_IPV6_MREQ
 struct ipv6_mreq {
     /* IPv6 multicast address of group */
         struct  in6_addr ipv6mr_multiaddr;
     /* local IPv6 address of interface */
     int     ipv6mr_ifindex;
 };
 #endif

The first field of ipv6_mreq has the same name ("ipv6mr_multiaddr")
and type ("in6_addr") in both cases. Moreover, the in6_addr structure
consists of fields that are always big-endian (on host of any endian),
therefore the ipv6_mreq's field ipv6mr_multiaddr doesn't need any
endian conversion.

The second field of ipv6_mreq may, however, depending on the build
environment, have different names. This is the reason why the lines
"#if __UAPI_DEF_IPV6_MREQ" and "#if defined(__UAPI_DEF_IPV6_MREQ)"
are used in this patch - to establish the right choice for the field
name. Also, endian conversion is needed for this field, since it is
of type "int".

Signed-off-by: Neng Chen <nchen@wavecomp.com>
Signed-off-by: Aleksandar Markovic <amarkovic@wavecomp.com>
---
 linux-user/syscall.c | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 5e29e67..dde6889 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -1921,6 +1921,33 @@ static abi_long do_setsockopt(int sockfd, int level, int optname,
                                        &pki, sizeof(pki)));
             break;
         }
+        case IPV6_ADD_MEMBERSHIP:
+        case IPV6_DROP_MEMBERSHIP:
+        {
+            struct ipv6_mreq ipv6mreq;
+
+            if (optlen < sizeof(ipv6mreq)) {
+                return -TARGET_EINVAL;
+            }
+
+            if (copy_from_user(&ipv6mreq, optval_addr, sizeof(ipv6mreq))) {
+                return -TARGET_EFAULT;
+            }
+
+#if defined(__UAPI_DEF_IPV6_MREQ)
+#if __UAPI_DEF_IPV6_MREQ
+            ipv6mreq.ipv6mr_ifindex = tswap32(ipv6mreq.ipv6mr_ifindex);
+#else
+            ipv6mreq.ipv6mr_interface = tswap32(ipv6mreq.ipv6mr_interface);
+#endif /* __UAPI_DEF_IVP6_MREQ */
+#else
+            ipv6mreq.ipv6mr_interface = tswap32(ipv6mreq.ipv6mr_interface);
+#endif /* defined (__UAPI_DEF_IPV6_MREQ) */
+
+            ret = get_errno(setsockopt(sockfd, level, optname,
+                                       &ipv6mreq, sizeof(ipv6mreq)));
+            break;
+        }
         default:
             goto unimplemented;
         }
-- 
2.7.4

Re: [Qemu-devel] [PATCH v10 1/3] linux-user: Add support for setsockopt() options IPV6_<ADD|DROP>_MEMBERSHIP

[Laurent Vivier]

On 07/06/2019 12:35, Aleksandar Markovic wrote:
> From: Neng Chen <nchen@wavecomp.com>
> 
> Add support for options IPV6_ADD_MEMBERSHIP and IPV6_DROP_MEMPEMBERSHIP
> of the syscall setsockopt(). These options control membership in
> multicast groups. Their argument is a pointer to a struct ipv6_mreq,
> which is in turn defined in IP v6 header netinet/in.h as:
> 
>  struct ipv6_mreq {
>      /* IPv6 multicast address of group */
>      struct  in6_addr  ipv6mr_multiaddr;
>      /* local IPv6 address of interface */
>      int     ipv6mr_interface;
>  };
> 
> ...whereas its definition in kernel's include/uapi/linux/in6.h is:
> 
>  #if __UAPI_DEF_IPV6_MREQ
>  struct ipv6_mreq {
>      /* IPv6 multicast address of group */
>          struct  in6_addr ipv6mr_multiaddr;
>      /* local IPv6 address of interface */
>      int     ipv6mr_ifindex;
>  };
>  #endif
> 
> The first field of ipv6_mreq has the same name ("ipv6mr_multiaddr")
> and type ("in6_addr") in both cases. Moreover, the in6_addr structure
> consists of fields that are always big-endian (on host of any endian),
> therefore the ipv6_mreq's field ipv6mr_multiaddr doesn't need any
> endian conversion.
> 
> The second field of ipv6_mreq may, however, depending on the build
> environment, have different names. This is the reason why the lines
> "#if __UAPI_DEF_IPV6_MREQ" and "#if defined(__UAPI_DEF_IPV6_MREQ)"
> are used in this patch - to establish the right choice for the field
> name. Also, endian conversion is needed for this field, since it is
> of type "int".
> 
> Signed-off-by: Neng Chen <nchen@wavecomp.com>
> Signed-off-by: Aleksandar Markovic <amarkovic@wavecomp.com>
> ---
>  linux-user/syscall.c | 27 +++++++++++++++++++++++++++
>  1 file changed, 27 insertions(+)
> 
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index 5e29e67..dde6889 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
> @@ -1921,6 +1921,33 @@ static abi_long do_setsockopt(int sockfd, int level, int optname,
>                                         &pki, sizeof(pki)));
>              break;
>          }
> +        case IPV6_ADD_MEMBERSHIP:
> +        case IPV6_DROP_MEMBERSHIP:
> +        {
> +            struct ipv6_mreq ipv6mreq;
> +
> +            if (optlen < sizeof(ipv6mreq)) {
> +                return -TARGET_EINVAL;
> +            }
> +
> +            if (copy_from_user(&ipv6mreq, optval_addr, sizeof(ipv6mreq))) {
> +                return -TARGET_EFAULT;
> +            }
> +
> +#if defined(__UAPI_DEF_IPV6_MREQ)
> +#if __UAPI_DEF_IPV6_MREQ
> +            ipv6mreq.ipv6mr_ifindex = tswap32(ipv6mreq.ipv6mr_ifindex);
> +#else
> +            ipv6mreq.ipv6mr_interface = tswap32(ipv6mreq.ipv6mr_interface);
> +#endif /* __UAPI_DEF_IVP6_MREQ */
> +#else
> +            ipv6mreq.ipv6mr_interface = tswap32(ipv6mreq.ipv6mr_interface);
> +#endif /* defined (__UAPI_DEF_IPV6_MREQ) */
> +
> +            ret = get_errno(setsockopt(sockfd, level, optname,
> +                                       &ipv6mreq, sizeof(ipv6mreq)));
> +            break;
> +        }
>          default:
>              goto unimplemented;
>          }
> 

It becomes complicated...

I think the first version of your patch using only ipv6mr_interface is
the correct one:

- POSIX defines ipv6mr_interface [1]

- __UAPI_DEF_IVP6_MREQ appears in kernel headers with v3.12

  cfd280c91253 net: sync some IP headers with glibc

- without __UAPI_DEF_IVP6_MREQ kernel defines ipv6mr_ifindex

and in cfd280c91253 it is explained:

   "If you include the kernel headers first you get those,
    and if you include the glibc headers first you get those,
    and the following patch arranges a coordination and
    synchronization between the two."

So before 3.12, a program can't include both netinet/in.h and linux/in6.h.

In linux-user/syscall.c, we only include netinet/in.h (glibc) and not
linux/in6.h (kernel-headers), so ipv6mr_interface is the one to use.

I found debian/wheezy (2013) is the most recent debian distro without
the definition of __UAPI_DEF_IPV6_MREQ, but it doesn't have gcc-4.8 and
thus QEMU can't be built.

Thanks,
Laurent

[1]
http://pubs.opengroup.org/onlinepubs/009695399/basedefs/netinet/in.h.html

[Qemu-devel] [PATCH v10 2/3] linux-user: Add support for setsockopt() option SOL_ALG

[Aleksandar Markovic]

From: Yunqiang Su <ysu@wavecomp.com>

Add support for options SOL_ALG of the syscall setsockopt(). This
option is used in relation to Linux kernel Crypto API, and allows
a user to set additional information for the cipher operation via
syscall setsockopt(). The field "optname" must be one of the
following:

  - ALG_SET_KEY – seting the key
  - ALG_SET_AEAD_AUTHSIZE – set the authentication tag size

SOL_ALG is relatively newer setsockopt() option. Therefore, the
code that handles SOL_ALG is enclosed in "ifdef" so that the build
does not fail for older kernels that do not contain support for
SOL_ALG. "ifdef" also contains check if ALG_SET_KEY and
ALG_SET_AEAD_AUTHSIZE are defined.

Signed-off-by: Yunqiang Su <ysu@wavecomp.com>
Signed-off-by: Aleksandar Markovic <amarkovic@wavecomp.com>
---
 linux-user/syscall.c | 31 +++++++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index dde6889..82c08b6 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -103,6 +103,7 @@
 #include <linux/blkpg.h>
 #include <netpacket/packet.h>
 #include <linux/netlink.h>
+#include <linux/if_alg.h>
 #include "linux_loop.h"
 #include "uname.h"
 
@@ -1998,6 +1999,36 @@ static abi_long do_setsockopt(int sockfd, int level, int optname,
             goto unimplemented;
         }
         break;
+#if defined(SOL_ALG) && defined(ALG_SET_KEY) && defined(ALG_SET_AEAD_AUTHSIZE)
+    case SOL_ALG:
+        switch (optname) {
+        case ALG_SET_KEY:
+        {
+            char *alg_key = g_malloc(optlen);
+
+            if (!alg_key) {
+                return -TARGET_ENOMEM;
+            }
+            if (copy_from_user(alg_key, optval_addr, optlen)) {
+                g_free(alg_key);
+                return -TARGET_EFAULT;
+            }
+            ret = get_errno(setsockopt(sockfd, level, optname,
+                                       alg_key, optlen));
+            g_free(alg_key);
+            break;
+        }
+        case ALG_SET_AEAD_AUTHSIZE:
+        {
+            ret = get_errno(setsockopt(sockfd, level, optname,
+                                       NULL, optlen));
+            break;
+        }
+        default:
+            goto unimplemented;
+        }
+        break;
+#endif
     case TARGET_SOL_SOCKET:
         switch (optname) {
         case TARGET_SO_RCVTIMEO:
-- 
2.7.4

Re: [Qemu-devel] [PATCH v10 2/3] linux-user: Add support for setsockopt() option SOL_ALG

[Laurent Vivier]

Le 07/06/2019 à 12:35, Aleksandar Markovic a écrit :
> From: Yunqiang Su <ysu@wavecomp.com>
> 
> Add support for options SOL_ALG of the syscall setsockopt(). This
> option is used in relation to Linux kernel Crypto API, and allows
> a user to set additional information for the cipher operation via
> syscall setsockopt(). The field "optname" must be one of the
> following:
> 
>   - ALG_SET_KEY – seting the key
>   - ALG_SET_AEAD_AUTHSIZE – set the authentication tag size
> 
> SOL_ALG is relatively newer setsockopt() option. Therefore, the
> code that handles SOL_ALG is enclosed in "ifdef" so that the build
> does not fail for older kernels that do not contain support for
> SOL_ALG. "ifdef" also contains check if ALG_SET_KEY and
> ALG_SET_AEAD_AUTHSIZE are defined.
> 
> Signed-off-by: Yunqiang Su <ysu@wavecomp.com>
> Signed-off-by: Aleksandar Markovic <amarkovic@wavecomp.com>
> ---
>  linux-user/syscall.c | 31 +++++++++++++++++++++++++++++++
>  1 file changed, 31 insertions(+)
> 
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index dde6889..82c08b6 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
> @@ -103,6 +103,7 @@
>  #include <linux/blkpg.h>
>  #include <netpacket/packet.h>
>  #include <linux/netlink.h>
> +#include <linux/if_alg.h>
>  #include "linux_loop.h"
>  #include "uname.h"
>  
> @@ -1998,6 +1999,36 @@ static abi_long do_setsockopt(int sockfd, int level, int optname,
>              goto unimplemented;
>          }
>          break;
> +#if defined(SOL_ALG) && defined(ALG_SET_KEY) && defined(ALG_SET_AEAD_AUTHSIZE)
> +    case SOL_ALG:
> +        switch (optname) {
> +        case ALG_SET_KEY:
> +        {
> +            char *alg_key = g_malloc(optlen);
> +
> +            if (!alg_key) {
> +                return -TARGET_ENOMEM;
> +            }
> +            if (copy_from_user(alg_key, optval_addr, optlen)) {
> +                g_free(alg_key);
> +                return -TARGET_EFAULT;
> +            }
> +            ret = get_errno(setsockopt(sockfd, level, optname,
> +                                       alg_key, optlen));
> +            g_free(alg_key);
> +            break;
> +        }
> +        case ALG_SET_AEAD_AUTHSIZE:
> +        {
> +            ret = get_errno(setsockopt(sockfd, level, optname,
> +                                       NULL, optlen));
> +            break;
> +        }
> +        default:
> +            goto unimplemented;
> +        }
> +        break;
> +#endif
>      case TARGET_SOL_SOCKET:
>          switch (optname) {
>          case TARGET_SO_RCVTIMEO:
> 

Reviewed-by: Laurent Vivier <laurent@vivier.eu>

[Qemu-devel] [PATCH v10 3/3] linux-user: Add support for statx() syscall

[Aleksandar Markovic]

From: Aleksandar Rikalo <arikalo@wavecomp.com>

Implement support for translation of system call statx().

The implementation is based on "best effort" approach: if host is
capable of executing statx(), host statx() is used. If not, the
implementation includes invoking other (more mature) system calls
(from the same 'stat' family) on the host side to achieve as close
as possible functionality.

Support for statx() in kernel and glibc was, however, introduced
at different points of time (the difference is more than a year):

  - kernel: Linux 4.11 (30 April 2017)
  - glibc: glibc 2.28 (1 Aug 2018)

In this patch, the availability of statx() support is established
via __NR_statx (if it is defined, statx() is considered available).
This coincedes with statx() introduction in kernel.

However, the structure statx definition may not be available for hosts
with glibc older than 2.28 (it is, by design, to be defined in one of
glibc headers), even though the full statx() functionality may be
supported in kernel, if the kernel is not older than 4.11. Hence,
a structure "target_statx" is defined in this patch, to remove that
dependency on glibc headers, and to use statx() functionality as soon
as the host kernel is capable of supporting it. Such structure statx
definition is used for both target and host structures statx (of
course, this doesn't mean the endian arrangement is the same on
target and host, and endian conversion is done in all necessary
cases).

Signed-off-by: Aleksandar Rikalo <arikalo@wavecomp.com>
Signed-off-by: Aleksandar Markovic <amarkovic@wavecomp.com>
---
 linux-user/syscall.c      | 135 +++++++++++++++++++++++++++++++++++++++++++++-
 linux-user/syscall_defs.h |  37 +++++++++++++
 2 files changed, 171 insertions(+), 1 deletion(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 82c08b6..b78ed45 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -43,6 +43,7 @@
 #include <sys/times.h>
 #include <sys/shm.h>
 #include <sys/sem.h>
+#include <sys/stat.h>
 #include <sys/statfs.h>
 #include <utime.h>
 #include <sys/sysinfo.h>
@@ -6526,6 +6527,48 @@ static inline abi_long host_to_target_stat64(void *cpu_env,
 }
 #endif
 
+#if defined(TARGET_NR_statx) && defined(__NR_statx)
+static inline abi_long host_to_target_statx(struct target_statx *host_stx,
+                                            abi_ulong target_addr)
+{
+    struct target_statx *target_stx;
+
+    if (!lock_user_struct(VERIFY_WRITE, target_stx, target_addr,  0)) {
+        return -TARGET_EFAULT;
+    }
+    memset(target_stx, 0, sizeof(*target_stx));
+
+    __put_user(host_stx->stx_mask, &target_stx->stx_mask);
+    __put_user(host_stx->stx_blksize, &target_stx->stx_blksize);
+    __put_user(host_stx->stx_attributes, &target_stx->stx_attributes);
+    __put_user(host_stx->stx_nlink, &target_stx->stx_nlink);
+    __put_user(host_stx->stx_uid, &target_stx->stx_uid);
+    __put_user(host_stx->stx_gid, &target_stx->stx_gid);
+    __put_user(host_stx->stx_mode, &target_stx->stx_mode);
+    __put_user(host_stx->stx_ino, &target_stx->stx_ino);
+    __put_user(host_stx->stx_size, &target_stx->stx_size);
+    __put_user(host_stx->stx_blocks, &target_stx->stx_blocks);
+    __put_user(host_stx->stx_attributes_mask, &target_stx->stx_attributes_mask);
+    __put_user(host_stx->stx_atime.tv_sec, &target_stx->stx_atime.tv_sec);
+    __put_user(host_stx->stx_atime.tv_nsec, &target_stx->stx_atime.tv_nsec);
+    __put_user(host_stx->stx_btime.tv_sec, &target_stx->stx_atime.tv_sec);
+    __put_user(host_stx->stx_btime.tv_nsec, &target_stx->stx_atime.tv_nsec);
+    __put_user(host_stx->stx_ctime.tv_sec, &target_stx->stx_atime.tv_sec);
+    __put_user(host_stx->stx_ctime.tv_nsec, &target_stx->stx_atime.tv_nsec);
+    __put_user(host_stx->stx_mtime.tv_sec, &target_stx->stx_atime.tv_sec);
+    __put_user(host_stx->stx_mtime.tv_nsec, &target_stx->stx_atime.tv_nsec);
+    __put_user(host_stx->stx_rdev_major, &target_stx->stx_rdev_major);
+    __put_user(host_stx->stx_rdev_minor, &target_stx->stx_rdev_minor);
+    __put_user(host_stx->stx_dev_major, &target_stx->stx_dev_major);
+    __put_user(host_stx->stx_dev_minor, &target_stx->stx_dev_minor);
+
+    unlock_user_struct(target_stx, target_addr, 1);
+
+    return 0;
+}
+#endif
+
+
 /* ??? Using host futex calls even when target atomic operations
    are not really atomic probably breaks things.  However implementing
    futexes locally would make futexes shared between multiple processes
@@ -7104,7 +7147,8 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
     abi_long ret;
 #if defined(TARGET_NR_stat) || defined(TARGET_NR_stat64) \
     || defined(TARGET_NR_lstat) || defined(TARGET_NR_lstat64) \
-    || defined(TARGET_NR_fstat) || defined(TARGET_NR_fstat64)
+    || defined(TARGET_NR_fstat) || defined(TARGET_NR_fstat64) \
+    || defined(TARGET_NR_statx)
     struct stat st;
 #endif
 #if defined(TARGET_NR_statfs) || defined(TARGET_NR_statfs64) \
@@ -10182,6 +10226,95 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
             ret = host_to_target_stat64(cpu_env, arg3, &st);
         return ret;
 #endif
+#if defined(TARGET_NR_statx)
+    case TARGET_NR_statx:
+        {
+            struct target_statx *target_stx;
+            int dirfd = arg1;
+            int flags = arg3;
+
+            p = lock_user_string(arg2);
+            if (p == NULL) {
+                return -TARGET_EFAULT;
+            }
+#if defined(__NR_statx)
+            {
+                /*
+                 * It is assumed that struct statx is arhitecture independent
+                 */
+                struct target_statx host_stx;
+                int mask = arg4;
+
+                ret = get_errno(syscall(__NR_statx, dirfd, p, flags, mask,
+                                        &host_stx));
+                if (!is_error(ret)) {
+                    if (host_to_target_statx(&host_stx, arg5) != 0) {
+                        unlock_user(p, arg2, 0);
+                        return -TARGET_EFAULT;
+                    }
+                }
+
+                if (ret != TARGET_ENOSYS) {
+                    unlock_user(p, arg2, 0);
+                    return ret;
+                }
+            }
+#endif
+            if ((p == NULL) || (*((char *)p) == 0)) {
+                /*
+                 * By file descriptor
+                 */
+                if (flags & AT_EMPTY_PATH) {
+                    unlock_user(p, arg2, 0);
+                    return -TARGET_ENOENT;
+                }
+                ret = get_errno(fstat(dirfd, &st));
+            } else if (*((char *)p) == '/') {
+                /*
+                 * By absolute pathname
+                 */
+                ret = get_errno(stat(path(p), &st));
+            } else {
+                if (dirfd == AT_FDCWD) {
+                    /*
+                     * By pathname relative to the current working directory
+                     */
+                    ret = get_errno(stat(path(p), &st));
+                } else {
+                    /*
+                     * By pathname relative to the directory referred to by
+                     * the file descriptor 'dirfd'
+                     */
+                    ret = get_errno(fstatat(dirfd, path(p), &st, flags));
+                }
+            }
+            unlock_user(p, arg2, 0);
+
+            if (!is_error(ret)) {
+                if (!lock_user_struct(VERIFY_WRITE, target_stx, arg5, 0)) {
+                    return -TARGET_EFAULT;
+                }
+                memset(target_stx, 0, sizeof(*target_stx));
+                __put_user(major(st.st_dev), &target_stx->stx_dev_major);
+                __put_user(minor(st.st_dev), &target_stx->stx_dev_minor);
+                __put_user(st.st_ino, &target_stx->stx_ino);
+                __put_user(st.st_mode, &target_stx->stx_mode);
+                __put_user(st.st_uid, &target_stx->stx_uid);
+                __put_user(st.st_gid, &target_stx->stx_gid);
+                __put_user(st.st_nlink, &target_stx->stx_nlink);
+                __put_user(major(st.st_rdev), &target_stx->stx_rdev_major);
+                __put_user(minor(st.st_rdev), &target_stx->stx_rdev_minor);
+                __put_user(st.st_size, &target_stx->stx_size);
+                __put_user(st.st_blksize, &target_stx->stx_blksize);
+                __put_user(st.st_blocks, &target_stx->stx_blocks);
+                __put_user(st.st_atime, &target_stx->stx_atime.tv_sec);
+                __put_user(st.st_mtime, &target_stx->stx_mtime.tv_sec);
+                __put_user(st.st_ctime, &target_stx->stx_ctime.tv_sec);
+                unlock_user_struct(target_stx, arg5, 1);
+            }
+        }
+        return ret;
+#endif
 #ifdef TARGET_NR_lchown
     case TARGET_NR_lchown:
         if (!(p = lock_user_string(arg1)))
diff --git a/linux-user/syscall_defs.h b/linux-user/syscall_defs.h
index 7f141f6..170c4dd 100644
--- a/linux-user/syscall_defs.h
+++ b/linux-user/syscall_defs.h
@@ -2536,4 +2536,41 @@ struct target_user_cap_data {
 /* Return size of the log buffer */
 #define TARGET_SYSLOG_ACTION_SIZE_BUFFER   10
 
+struct target_statx_timestamp {
+   int64_t tv_sec;
+   uint32_t tv_nsec;
+   int32_t __reserved;
+};
+
+struct target_statx {
+   /* 0x00 */
+   uint32_t stx_mask;       /* What results were written [uncond] */
+   uint32_t stx_blksize;    /* Preferred general I/O size [uncond] */
+   uint64_t stx_attributes; /* Flags conveying information about the file */
+   /* 0x10 */
+   uint32_t stx_nlink;      /* Number of hard links */
+   uint32_t stx_uid;        /* User ID of owner */
+   uint32_t stx_gid;        /* Group ID of owner */
+   uint16_t stx_mode;       /* File mode */
+   uint16_t __spare0[1];
+   /* 0x20 */
+   uint64_t stx_ino;        /* Inode number */
+   uint64_t stx_size;       /* File size */
+   uint64_t stx_blocks;     /* Number of 512-byte blocks allocated */
+   uint64_t stx_attributes_mask; /* Mask to show what is supported */
+   /* 0x40 */
+   struct target_statx_timestamp  stx_atime;  /* Last access time */
+   struct target_statx_timestamp  stx_btime;  /* File creation time */
+   struct target_statx_timestamp  stx_ctime;  /* Last attribute change time */
+   struct target_statx_timestamp  stx_mtime;  /* Last data modification time */
+   /* 0x80 */
+   uint32_t stx_rdev_major;   /* Device ID of special file [if bdev/cdev] */
+   uint32_t stx_rdev_minor;
+   uint32_t stx_dev_major; /* ID of device containing file [uncond] */
+   uint32_t stx_dev_minor;
+   /* 0x90 */
+   uint64_t __spare2[14];  /* Spare space for future expansion */
+   /* 0x100 */
+};
+
 #endif
-- 
2.7.4

Re: [Qemu-devel] [PATCH v10 3/3] linux-user: Add support for statx() syscall

[Laurent Vivier]

Le 07/06/2019 à 12:35, Aleksandar Markovic a écrit :
> From: Aleksandar Rikalo <arikalo@wavecomp.com>
> 
> Implement support for translation of system call statx().
> 
> The implementation is based on "best effort" approach: if host is
> capable of executing statx(), host statx() is used. If not, the
> implementation includes invoking other (more mature) system calls
> (from the same 'stat' family) on the host side to achieve as close
> as possible functionality.
> 
> Support for statx() in kernel and glibc was, however, introduced
> at different points of time (the difference is more than a year):
> 
>   - kernel: Linux 4.11 (30 April 2017)
>   - glibc: glibc 2.28 (1 Aug 2018)
> 
> In this patch, the availability of statx() support is established
> via __NR_statx (if it is defined, statx() is considered available).
> This coincedes with statx() introduction in kernel.
> 
> However, the structure statx definition may not be available for hosts
> with glibc older than 2.28 (it is, by design, to be defined in one of
> glibc headers), even though the full statx() functionality may be
> supported in kernel, if the kernel is not older than 4.11. Hence,
> a structure "target_statx" is defined in this patch, to remove that
> dependency on glibc headers, and to use statx() functionality as soon
> as the host kernel is capable of supporting it. Such structure statx
> definition is used for both target and host structures statx (of
> course, this doesn't mean the endian arrangement is the same on
> target and host, and endian conversion is done in all necessary
> cases).
> 
> Signed-off-by: Aleksandar Rikalo <arikalo@wavecomp.com>
> Signed-off-by: Aleksandar Markovic <amarkovic@wavecomp.com>
> ---
>  linux-user/syscall.c      | 135 +++++++++++++++++++++++++++++++++++++++++++++-
>  linux-user/syscall_defs.h |  37 +++++++++++++
>  2 files changed, 171 insertions(+), 1 deletion(-)
> 
> diff --git a/linux-user/syscall.c b/linux-user/syscall.c
> index 82c08b6..b78ed45 100644
> --- a/linux-user/syscall.c
> +++ b/linux-user/syscall.c
...
> @@ -10182,6 +10226,95 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
>              ret = host_to_target_stat64(cpu_env, arg3, &st);
>          return ret;
>  #endif
> +#if defined(TARGET_NR_statx)
> +    case TARGET_NR_statx:
> +        {
> +            struct target_statx *target_stx;
> +            int dirfd = arg1;
> +            int flags = arg3;
> +
> +            p = lock_user_string(arg2);
> +            if (p == NULL) {
> +                return -TARGET_EFAULT;
> +            }
> +#if defined(__NR_statx)
> +            {
> +                /*
> +                 * It is assumed that struct statx is arhitecture independent

s/arhitecture/architecture/

> +                 */
> +                struct target_statx host_stx;
> +                int mask = arg4;
> +
> +                ret = get_errno(syscall(__NR_statx, dirfd, p, flags, mask,
> +                                        &host_stx));

You should define sys_statx() using _syscall5() macro and use it.

> +                if (!is_error(ret)) {
> +                    if (host_to_target_statx(&host_stx, arg5) != 0) {
> +                        unlock_user(p, arg2, 0);
> +                        return -TARGET_EFAULT;
> +                    }
> +                }
> +
> +                if (ret != TARGET_ENOSYS) {

ret != -TARGET_ENOSYS

> +                    unlock_user(p, arg2, 0);
> +                    return ret;
> +                }
> +            }
> +#endif
> +            if ((p == NULL) || (*((char *)p) == 0)) {

You already checked above p is not NULL and exited with -TARGET_EFAULT.

BTW, do we really need to emulate the syscall if it is not available?

I think the user-space application calling statx() should be ready to
receive ENOSYS and define some kinds of fallback (like you do below). So
it should not be done by QEMU.

> +                /*
> +                 * By file descriptor
> +                 */
> +                if (flags & AT_EMPTY_PATH) {
> +                    unlock_user(p, arg2, 0);
> +                    return -TARGET_ENOENT;
> +                }
> +                ret = get_errno(fstat(dirfd, &st));
> +            } else if (*((char *)p) == '/') {
> +                /*
> +                 * By absolute pathname
> +                 */
> +                ret = get_errno(stat(path(p), &st));
> +            } else {
> +                if (dirfd == AT_FDCWD) {
> +                    /*
> +                     * By pathname relative to the current working directory
> +                     */
> +                    ret = get_errno(stat(path(p), &st));

Why do we divide the case in two parts, fstatat() should work here too.

> +                } else {
> +                    /*
> +                     * By pathname relative to the directory referred to by
> +                     * the file descriptor 'dirfd'
> +                     */
> +                    ret = get_errno(fstatat(dirfd, path(p), &st, flags));
> +                }
> +            }
> +            unlock_user(p, arg2, 0);
> +
> +            if (!is_error(ret)) {
> +                if (!lock_user_struct(VERIFY_WRITE, target_stx, arg5, 0)) {
> +                    return -TARGET_EFAULT;
> +                }
> +                memset(target_stx, 0, sizeof(*target_stx));
> +                __put_user(major(st.st_dev), &target_stx->stx_dev_major);
> +                __put_user(minor(st.st_dev), &target_stx->stx_dev_minor);
> +                __put_user(st.st_ino, &target_stx->stx_ino);
> +                __put_user(st.st_mode, &target_stx->stx_mode);
> +                __put_user(st.st_uid, &target_stx->stx_uid);
> +                __put_user(st.st_gid, &target_stx->stx_gid);
> +                __put_user(st.st_nlink, &target_stx->stx_nlink);
> +                __put_user(major(st.st_rdev), &target_stx->stx_rdev_major);
> +                __put_user(minor(st.st_rdev), &target_stx->stx_rdev_minor);
> +                __put_user(st.st_size, &target_stx->stx_size);
> +                __put_user(st.st_blksize, &target_stx->stx_blksize);
> +                __put_user(st.st_blocks, &target_stx->stx_blocks);
> +                __put_user(st.st_atime, &target_stx->stx_atime.tv_sec);
> +                __put_user(st.st_mtime, &target_stx->stx_mtime.tv_sec);
> +                __put_user(st.st_ctime, &target_stx->stx_ctime.tv_sec);
> +                unlock_user_struct(target_stx, arg5, 1);
> +            }
> +        }
> +        return ret;
> +#endif

Thanks,
Laurent

Re: [Qemu-devel] [PATCH v10 3/3] linux-user: Add support for statx() syscall

[Aleksandar Rikalo]

Hi Laurent,

> s/arhitecture/architecture/

Done.

> You should define sys_statx() using _syscall5() macro and use it.

Done.

> ret != -TARGET_ENOSYS

Done.

> You already checked above p is not NULL and exited with -TARGET_EFAULT.

Done.

> BTW, do we really need to emulate the syscall if it is not available?
>
> I think the user-space application calling statx() should be ready to
> receive ENOSYS and define some kinds of fallback (like you do below). So
> it should not be done by QEMU.

nanoMIPS linux port doesn't support any of "stats" but the statx, so there
is no fallback in nanoMIPS user-space applications.

I think, we can expect similar situation for any new linux port.

> Why do we divide the case in two parts, fstatat() should work here too.

fstat() uses file descriptor, but here we have string which represents
file name with absolute path.

All system calls from 'stat' group whose name starts with letter f require
file descriptor as an argument. Whereas remaining system calls require
file name / path as string. In that sense, statx() is a hybrid between
the two, hence the solution I propose.

Aleksandar Rikalo

Re: [Qemu-devel] [PATCH v10 3/3] linux-user: Add support for statx() syscall

[Laurent Vivier]

Le 19/06/2019 à 14:12, Aleksandar Rikalo a écrit :
> Hi Laurent,
...
>> BTW, do we really need to emulate the syscall if it is not available?
>>
>> I think the user-space application calling statx() should be ready to
>> receive ENOSYS and define some kinds of fallback (like you do below). So
>> it should not be done by QEMU.
> 
> nanoMIPS linux port doesn't support any of "stats" but the statx, so there
> is no fallback in nanoMIPS user-space applications.
> 
> I think, we can expect similar situation for any new linux port.

OK, I understand, so I agree, we need the emulation part.

>> Why do we divide the case in two parts, fstatat() should work here too.
> 
> fstat() uses file descriptor, but here we have string which represents
> file name with absolute path.
> 
> All system calls from 'stat' group whose name starts with letter f require
> file descriptor as an argument. Whereas remaining system calls require
> file name / path as string. In that sense, statx() is a hybrid between
> the two, hence the solution I propose.

but fstatat() works like statx(), it accepts file descriptor and path.

So what I proposed is to replace:

+            } else {
+                if (dirfd == AT_FDCWD) {
+                    /*
+                     * By pathname relative to the current working directory
+                     */
+                    ret = get_errno(stat(path(p), &st));
+                    unlock_user(p, arg2, 0);
+                } else {
+                    /*
+                     * By pathname relative to the directory referred to by
+                     * the file descriptor 'dirfd'
+                     */
+                    ret = get_errno(fstatat(dirfd, path(p), &st, flags));
+                    unlock_user(p, arg2, 0);
+                }
+            }

by something like;

+            } else {
+                ret = get_errno(fstatat(dirfd, path(p), &st, flags));
+            }

as fstatat() also accepts AT_FDCWD.

Moreover in kernel vfs_fstatat() calls vfs_statx():

static inline int vfs_fstatat(int dfd, const char __user *filename,
                              struct kstat *stat, int flags)
{
        return vfs_statx(dfd, filename, flags | AT_NO_AUTOMOUNT,
                         stat, STATX_BASIC_STATS);
}

so maybe all the cases can be emulated by fstatat()?

Or did I miss something?

Thanks,
Laurent

Re: [Qemu-devel] [PATCH v10 3/3] linux-user: Add support for statx() syscall

[Jim Wilson]

On 6/7/19 3:35 AM, Aleksandar Markovic wrote:
> Implement support for translation of system call statx().

I also need these patches for 32-bit RISC-V linux user mode support.

glibc ld.so calls statx if fstatat is not supported.  Apparently new 
linux architecture ports aren't allowed to define __ARCH_WANT_NEW_STAT 
which enables fstatat because this is already obsolete.  64-bit RISC-V 
linux does have fstatat, but apparently this was a mistake which we 
can't fix now because the ABI is already frozen.  The 32-bit RISC-V ABI 
is not frozen yet, so it won't have fstatat.  Anyways, without statx, 
ld.so doesn't work, which makes user mode qemu pretty useless, so we do 
need this emulated in qemu to make the 32-bit RISC-V linux user mode 
support work properly.

I started with the August 2018 version of the patch a few weeks ago, and 
just noticed that it has been resubmitted.  I had to modify the patch 
slightly to apply to current sources, and had to fix one bug to make it 
work.  The line
+                if (ret != TARGET_ENOSYS) {
needs to instead be
+                if (ret != -TARGET_ENOSYS) {
I see that Laurent has already pointed that out.

Incidentally, I also have strace patches for statx that work on top of 
this patch, since I didn't see that in the nanomips patch set I started 
with.  That helped me debug the 32-bit RISC-V user mode support.

I've tested this on Ubuntu 16.04 (no host statx) and Ubuntu 19.10 (with 
host statx) and it worked well for me running the gcc testsuite for a 
riscv32-linux target.  I haven't tried testing the latest version of the 
patch yet.  I can do that if this is helpful.

Jim

Re: [Qemu-devel] [PATCH v10 3/3] linux-user: Add support for statx() syscall

[Aleksandar Markovic]

On Wednesday, June 19, 2019, Jim Wilson <jimw@sifive.com> wrote:

> On 6/7/19 3:35 AM, Aleksandar Markovic wrote:
>
>> Implement support for translation of system call statx().
>>
>
> I also need these patches for 32-bit RISC-V linux user mode support.
>
> glibc ld.so calls statx if fstatat is not supported.  Apparently new linux
> architecture ports aren't allowed to define __ARCH_WANT_NEW_STAT which
> enables fstatat because this is already obsolete.  64-bit RISC-V linux does
> have fstatat, but apparently this was a mistake which we can't fix now
> because the ABI is already frozen.  The 32-bit RISC-V ABI is not frozen
> yet, so it won't have fstatat.  Anyways, without statx, ld.so doesn't work,
> which makes user mode qemu pretty useless, so we do need this emulated in
> qemu to make the 32-bit RISC-V linux user mode support work properly.
>
>
Jim, Aleksandar Rikalo, the author of the patch was about to send Laurent
explanation why this aproach is needed in its current organization, on a
very similar line of reasoning as yours.

I am waiting on him to send a new version of the series. Meanwhile you can
send strace patch to the list, and I can even incude it in my series after
and together with Aleksandar's patch, if you don't object.

Yours,

Aleksandar M.

I started with the August 2018 version of the patch a few weeks ago, and
> just noticed that it has been resubmitted.  I had to modify the patch
> slightly to apply to current sources, and had to fix one bug to make it
> work.  The line
> +                if (ret != TARGET_ENOSYS) {
> needs to instead be
> +                if (ret != -TARGET_ENOSYS) {
> I see that Laurent has already pointed that out.
>
> Incidentally, I also have strace patches for statx that work on top of
> this patch, since I didn't see that in the nanomips patch set I started
> with.  That helped me debug the 32-bit RISC-V user mode support.
>
> I've tested this on Ubuntu 16.04 (no host statx) and Ubuntu 19.10 (with
> host statx) and it worked well for me running the gcc testsuite for a
> riscv32-linux target.  I haven't tried testing the latest version of the
> patch yet.  I can do that if this is helpful.
>
> Jim
>
>

Re: [Qemu-devel] [PATCH v10 3/3] linux-user: Add support for statx() syscall

[Jim Wilson]

On Tue, Jun 18, 2019 at 4:13 PM Aleksandar Markovic
<aleksandar.m.mail@gmail.com> wrote:
> I am waiting on him to send a new version of the series. Meanwhile you can send strace patch to the list, and I can even incude it in my series after and together with Aleksandar's patch, if you don't object.

I submitted it the usual way, so it is on the mailing list now.  If
you want to include it with your patch series that is fine.
https://lists.nongnu.org/archive/html/qemu-devel/2019-06/msg04087.html

Jim

Re: [Qemu-devel] [PATCH v10 0/3] linux-user: A set of miscellaneous patches

[Aleksandar Markovic]

Ping
On Jun 7, 2019 2:21 PM, "Aleksandar Markovic" <aleksandar.markovic@rt-rk.com>
wrote:

> From: Aleksandar Markovic <amarkovic@wavecomp.com>
>
> This is a collection of misc patches for Linux user that I recently
> accumulated from variuous sources. All of them originate from problems
> observed on mips target. However, these changes actually affect and fix
> problems on multiple targets.
>
> v9->v10:
>
>   - improved commit messages for patches 2 and 3
>
> v8->v9:
>
>   - fixed build error on some systems related to SOL_ALG
>
> v7->v8:
>
>   - added a patch on setsockopt() option SOL_ALG
>
> v6->v7:
>
>   - fixed a build error for older kernels related to the patch on
>     setsockopt() options
>   - removed four patches that on the meantime got accepted into the
>     main source tree
>
> v5->v6:
>
>   - fixed a mistake in patch #4
>   - improved commit messages in patches #4 and #6
>
> v4->v5:
>
>   - added the patch on statx() support
>   - improved the patch on IPV6_<ADD|DROP>_MEMBERSHIP to take into
>     account the possibility of different names for a field
>   - minor corrections in commit messages
>
> v3->v4:
>
>   - improved commit messages (fixed some typos, improved relevance)
>
> v2->v3:
>
>   - updated and improved commit messages
>   - added IPV6_DROP_MEMBERSHIP support to the patch on setsockopt()'s
>     option
>
> v1->v2:
>
>   - added the patch on setsockopt()'s option IPV6_ADD_MEMBERSHIP
>   - improved the commit me
>
> Aleksandar Rikalo (1):
>   linux-user: Add support for statx() syscall
>
> Neng Chen (1):
>   linux-user: Add support for setsockopt() options
>     IPV6_<ADD|DROP>_MEMBERSHIP
>
> Yunqiang Su (1):
>   linux-user: Add support for setsockopt() option SOL_ALG
>
>  linux-user/syscall.c      | 193 ++++++++++++++++++++++++++++++
> +++++++++++++++-
>  linux-user/syscall_defs.h |  37 +++++++++
>  2 files changed, 229 insertions(+), 1 deletion(-)
>
> --
> 2.7.4
>
>
>