From: Richard Henderson <r...@twiddle.net>
If the interp_prefix is a complete chroot, it may have a *lot* of files.
Setting up the cache for this is quite expensive.
For the most part, we can use the *at versions of various syscalls to
attempt the operation in the prefix. For the few cases that remain,
use faccessat and create the full path on demand.
Cc: Eric Blake <ebl...@redhat.com>
Cc: Peter Maydell <peter.mayd...@linaro.org>
Signed-off-by: Richard Henderson <r...@twiddle.net>
---
Changes since v3 (Dec 29 2017):
* Use DO/WHILE as the control construct; wrap it in a macro.
* Introduce linux_user_path to handle the cases *at syscalls
do not cover.
Changes since v2 (Dec 4 2017):
* Use IF as the control construct instead of SWITCH.
Changes since v1 (Nov 2016):
* Require interp_dirfd set before trying the *at path.
r~
---
linux-user/qemu.h | 15 ++++++
linux-user/elfload.c | 5 +-
linux-user/main.c | 36 +++++++++++++-
linux-user/syscall.c | 130 +++++++++++++++++++++++++++++++++++----------------
4 files changed, 141 insertions(+), 45 deletions(-)
diff --git a/linux-user/qemu.h b/linux-user/qemu.h
index 4edd7d0c08..5b621f26e0 100644
--- a/linux-user/qemu.h
+++ b/linux-user/qemu.h
@@ -437,8 +437,23 @@ void mmap_fork_start(void);
void mmap_fork_end(int child);
/* main.c */
+extern int interp_dirfd;
extern unsigned long guest_stack_size;
+#define CHOOSE_INTERP(RET, PATH, OPENAT_EXPR, NORMAL_EXPR) \
+ do { \
+ if (interp_dirfd >= 0 && PATH[0] == '/') { \
+ RET = OPENAT_EXPR; \
+ if (!(RET < 0 && errno == ENOENT)) { \
+ break; \
+ } \
+ } \
+ RET = NORMAL_EXPR; \
+ } while (0)
+
+const char *linux_user_path(const char *);
+#define path(x) linux_user_path(x)
+
/* user access */
#define VERIFY_READ 0
diff --git a/linux-user/elfload.c b/linux-user/elfload.c
index 32a47674e6..1fb097e30d 100644
--- a/linux-user/elfload.c
+++ b/linux-user/elfload.c
@@ -6,7 +6,6 @@
#include "qemu.h"
#include "disas/disas.h"
-#include "qemu/path.h"
#ifdef _ARCH_PPC64
#undef ARCH_DLINFO
@@ -2204,7 +2203,9 @@ static void load_elf_interp(const char *filename, struct
image_info *info,
{
int fd, retval;
- fd = open(path(filename), O_RDONLY);
+ CHOOSE_INTERP(fd, filename,
+ openat(interp_dirfd, filename + 1, O_RDONLY),
+ open(filename, O_RDONLY));
if (fd < 0) {
goto exit_perror;
}
diff --git a/linux-user/main.c b/linux-user/main.c
index 2140465709..8f4087d508 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -23,7 +23,6 @@
#include "qapi/error.h"
#include "qemu.h"
-#include "qemu/path.h"
#include "qemu/config-file.h"
#include "qemu/cutils.h"
#include "qemu/help_option.h"
@@ -98,8 +97,41 @@ unsigned long reserved_va;
static void usage(int exitcode);
static const char *interp_prefix = CONFIG_QEMU_INTERP_PREFIX;
+int interp_dirfd;
const char *qemu_uname_release;
+const char *linux_user_path(const char *pathname)
+{
+ static THREAD size_t save_len;
+ static THREAD char *save_buf;
+ size_t len, prefix_len, path_len;
+ int e;
+
+ /* Only consider absolute paths. */
+ if (pathname[0] != '/' || interp_dirfd < 0) {
+ return pathname;
+ }
+
+ /* Test if the path within interp_dir exists. */
+ e = faccessat(interp_dirfd, pathname + 1, F_OK, AT_SYMLINK_NOFOLLOW);
+ if (e < 0 && errno != ENOENT) {
+ return pathname;
+ }
+
+ /* It does -- form the new absolute path. */
+ prefix_len = strlen(interp_prefix);
+ path_len = strlen(pathname) + 1;
+ len = prefix_len + path_len;
+ if (len <= save_len) {
+ save_len = len;
+ save_buf = realloc(save_buf, len);
+ }
+ memcpy(save_buf, interp_prefix, prefix_len);
+ memcpy(save_buf + prefix_len, pathname, path_len);
+
+ return save_buf;
+}
+
/* XXX: on x86 MAP_GROWSDOWN only works if ESP <= address + 32, so
we allocate a bigger stack. Need a better solution, for example
by remapping the process stack directly at the right place */
@@ -4319,7 +4351,7 @@ int main(int argc, char **argv, char **envp)
memset(&bprm, 0, sizeof (bprm));
/* Scan interp_prefix dir for replacement files. */
- init_paths(interp_prefix);
+ interp_dirfd = open(interp_prefix, O_CLOEXEC | O_DIRECTORY | O_PATH);
init_qemu_uname_release();
diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 74378947f0..ebd41fcab4 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -19,7 +19,6 @@
#define _ATFILE_SOURCE
#include "qemu/osdep.h"
#include "qemu/cutils.h"
-#include "qemu/path.h"
#include <elf.h>
#include <endian.h>
#include <grp.h>
@@ -7263,7 +7262,10 @@ static abi_long do_name_to_handle_at(abi_long dirfd,
abi_long pathname,
fh = g_malloc0(total_size);
fh->handle_bytes = size;
- ret = get_errno(name_to_handle_at(dirfd, path(name), fh, &mid, flags));
+ CHOOSE_INTERP(ret, name,
+ name_to_handle_at(interp_dirfd, name + 1, fh, &mid, flags),
+ name_to_handle_at(dirfd, name, fh, &mid, flags));
+ ret = get_errno(ret);
unlock_user(name, pathname, 0);
/* man name_to_handle_at(2):
@@ -7639,6 +7641,7 @@ static int do_openat(void *cpu_env, int dirfd, const char
*pathname, int flags,
#endif
{ NULL, NULL, NULL }
};
+ int ret;
if (is_proc_myself(pathname, "exe")) {
int execfd = qemu_getauxval(AT_EXECFD);
@@ -7678,7 +7681,10 @@ static int do_openat(void *cpu_env, int dirfd, const
char *pathname, int flags,
return fd;
}
- return safe_openat(dirfd, path(pathname), flags, mode);
+ CHOOSE_INTERP(ret, pathname,
+ safe_openat(interp_dirfd, pathname + 1, flags, mode),
+ safe_openat(dirfd, pathname, flags, mode));
+ return ret;
}
#define TIMER_MAGIC 0x0caf0000
@@ -7831,6 +7837,7 @@ abi_long do_syscall(void *cpu_env, int num, abi_long arg1,
struct stat st;
struct statfs stfs;
void *p;
+ char *fn;
#if defined(DEBUG_ERESTARTSYS)
/* Debug-only code for exercising the syscall-restart code paths
@@ -8362,10 +8369,14 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
} else {
tvp = NULL;
}
- if (!(p = lock_user_string(arg2)))
+ if (!(fn = lock_user_string(arg2))) {
goto efault;
- ret = get_errno(futimesat(arg1, path(p), tvp));
- unlock_user(p, arg2, 0);
+ }
+ CHOOSE_INTERP(ret, fn,
+ futimesat(interp_dirfd, fn + 1, tvp),
+ futimesat(arg1, fn, tvp));
+ ret = get_errno(ret);
+ unlock_user(fn, arg2, 0);
}
break;
#endif
@@ -8379,18 +8390,26 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
#endif
#ifdef TARGET_NR_access
case TARGET_NR_access:
- if (!(p = lock_user_string(arg1)))
+ if (!(fn = lock_user_string(arg1))) {
goto efault;
- ret = get_errno(access(path(p), arg2));
- unlock_user(p, arg1, 0);
+ }
+ CHOOSE_INTERP(ret, fn,
+ faccessat(interp_dirfd, fn + 1, arg2, 0),
+ access(fn, arg2));
+ ret = get_errno(ret);
+ unlock_user(fn, arg1, 0);
break;
#endif
#if defined(TARGET_NR_faccessat) && defined(__NR_faccessat)
case TARGET_NR_faccessat:
- if (!(p = lock_user_string(arg2)))
+ if (!(fn = lock_user_string(arg2))) {
goto efault;
- ret = get_errno(faccessat(arg1, p, arg3, 0));
- unlock_user(p, arg2, 0);
+ }
+ CHOOSE_INTERP(ret, fn,
+ faccessat(interp_dirfd, fn + 1, arg3, 0),
+ faccessat(arg1, fn, arg3, 0));
+ ret = get_errno(ret);
+ unlock_user(fn, arg2, 0);
break;
#endif
#ifdef TARGET_NR_nice /* not on alpha */
@@ -9307,14 +9326,14 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
case TARGET_NR_readlink:
{
void *p2;
- p = lock_user_string(arg1);
+ fn = lock_user_string(arg1);
p2 = lock_user(VERIFY_WRITE, arg2, arg3, 0);
- if (!p || !p2) {
+ if (!fn || !p2) {
ret = -TARGET_EFAULT;
} else if (!arg3) {
/* Short circuit this for the magic exe check. */
ret = -TARGET_EINVAL;
- } else if (is_proc_myself((const char *)p, "exe")) {
+ } else if (is_proc_myself(fn, "exe")) {
char real[PATH_MAX], *temp;
temp = realpath(exec_path, real);
/* Return value is # of bytes that we wrote to the buffer. */
@@ -9328,10 +9347,13 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
memcpy(p2, real, ret);
}
} else {
- ret = get_errno(readlink(path(p), p2, arg3));
+ CHOOSE_INTERP(ret, fn,
+ readlinkat(interp_dirfd, fn + 1, p2, arg3),
+ readlink(fn, p2, arg3));
+ ret = get_errno(ret);
}
unlock_user(p2, arg2, ret);
- unlock_user(p, arg1, 0);
+ unlock_user(fn, arg1, 0);
}
break;
#endif
@@ -9339,20 +9361,23 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
case TARGET_NR_readlinkat:
{
void *p2;
- p = lock_user_string(arg2);
+ fn = lock_user_string(arg2);
p2 = lock_user(VERIFY_WRITE, arg3, arg4, 0);
- if (!p || !p2) {
+ if (!fn || !p2) {
ret = -TARGET_EFAULT;
- } else if (is_proc_myself((const char *)p, "exe")) {
+ } else if (is_proc_myself(fn, "exe")) {
char real[PATH_MAX], *temp;
temp = realpath(exec_path, real);
ret = temp == NULL ? get_errno(-1) : strlen(real) ;
snprintf((char *)p2, arg4, "%s", real);
} else {
- ret = get_errno(readlinkat(arg1, path(p), p2, arg4));
+ CHOOSE_INTERP(ret, fn,
+ readlinkat(interp_dirfd, fn + 1, p2, arg4),
+ readlinkat(arg1, fn, p2, arg4));
+ ret = get_errno(ret);
}
unlock_user(p2, arg3, ret);
- unlock_user(p, arg2, 0);
+ unlock_user(fn, arg2, 0);
}
break;
#endif
@@ -9780,18 +9805,26 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
break;
#ifdef TARGET_NR_stat
case TARGET_NR_stat:
- if (!(p = lock_user_string(arg1)))
+ if (!(fn = lock_user_string(arg1))) {
goto efault;
- ret = get_errno(stat(path(p), &st));
- unlock_user(p, arg1, 0);
+ }
+ CHOOSE_INTERP(ret, fn,
+ fstatat(interp_dirfd, fn + 1, &st, 0),
+ stat(fn, &st));
+ ret = get_errno(ret);
+ unlock_user(fn, arg1, 0);
goto do_stat;
#endif
#ifdef TARGET_NR_lstat
case TARGET_NR_lstat:
- if (!(p = lock_user_string(arg1)))
+ if (!(fn = lock_user_string(arg1))) {
goto efault;
- ret = get_errno(lstat(path(p), &st));
- unlock_user(p, arg1, 0);
+ }
+ CHOOSE_INTERP(ret, fn,
+ fstatat(interp_dirfd, fn + 1, &st, AT_SYMLINK_NOFOLLOW),
+ lstat(fn, &st));
+ ret = get_errno(ret);
+ unlock_user(fn, arg1, 0);
goto do_stat;
#endif
case TARGET_NR_fstat:
@@ -10886,20 +10919,28 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
#endif
#ifdef TARGET_NR_stat64
case TARGET_NR_stat64:
- if (!(p = lock_user_string(arg1)))
+ if (!(fn = lock_user_string(arg1))) {
goto efault;
- ret = get_errno(stat(path(p), &st));
- unlock_user(p, arg1, 0);
+ }
+ CHOOSE_INTERP(ret, fn,
+ fstatat(interp_dirfd, fn + 1, &st, 0),
+ stat(fn, &st));
+ ret = get_errno(ret);
+ unlock_user(fn, arg1, 0);
if (!is_error(ret))
ret = host_to_target_stat64(cpu_env, arg2, &st);
break;
#endif
#ifdef TARGET_NR_lstat64
case TARGET_NR_lstat64:
- if (!(p = lock_user_string(arg1)))
+ if (!(fn = lock_user_string(arg1))) {
goto efault;
- ret = get_errno(lstat(path(p), &st));
- unlock_user(p, arg1, 0);
+ }
+ CHOOSE_INTERP(ret, fn,
+ fstatat(interp_dirfd, fn + 1, &st, AT_SYMLINK_NOFOLLOW),
+ lstat(fn, &st));
+ ret = get_errno(ret);
+ unlock_user(fn, arg1, 0);
if (!is_error(ret))
ret = host_to_target_stat64(cpu_env, arg2, &st);
break;
@@ -10918,9 +10959,14 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
#ifdef TARGET_NR_newfstatat
case TARGET_NR_newfstatat:
#endif
- if (!(p = lock_user_string(arg2)))
+ if (!(fn = lock_user_string(arg2))) {
goto efault;
- ret = get_errno(fstatat(arg1, path(p), &st, arg4));
+ }
+ CHOOSE_INTERP(ret, fn,
+ fstatat(interp_dirfd, fn + 1, &st, arg4),
+ fstatat(arg1, fn, &st, arg4));
+ ret = get_errno(ret);
+ unlock_user(fn, arg2, 0);
if (!is_error(ret))
ret = host_to_target_stat64(cpu_env, arg3, &st);
break;
@@ -11917,12 +11963,14 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
if (!arg2)
ret = get_errno(sys_utimensat(arg1, NULL, tsp, arg4));
else {
- if (!(p = lock_user_string(arg2))) {
- ret = -TARGET_EFAULT;
- goto fail;
+ if (!(fn = lock_user_string(arg2))) {
+ goto efault;
}
- ret = get_errno(sys_utimensat(arg1, path(p), tsp, arg4));
- unlock_user(p, arg2, 0);
+ CHOOSE_INTERP(ret, fn,
+ sys_utimensat(interp_dirfd, fn + 1, tsp, arg4),
+ sys_utimensat(arg1, fn, tsp, arg4));
+ ret = get_errno(ret);
+ unlock_user(fn, arg2, 0);
}
}
break;
--
2.14.3
