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. Instead, use the *at
versions of various syscalls to attempt the operation in the prefix.
Cc: Eric Blake <ebl...@redhat.com>
Cc: Peter Maydell <peter.mayd...@linaro.org>
Signed-off-by: Richard Henderson <r...@twiddle.net>
---
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 | 1 +
linux-user/elfload.c | 7 +-
linux-user/main.c | 3 +-
linux-user/syscall.c | 176 ++++++++++++++++++++++++++++++++++++++-------------
4 files changed, 141 insertions(+), 46 deletions(-)
diff --git a/linux-user/qemu.h b/linux-user/qemu.h
index 4edd7d0c08..d04924fd2e 100644
--- a/linux-user/qemu.h
+++ b/linux-user/qemu.h
@@ -437,6 +437,7 @@ void mmap_fork_start(void);
void mmap_fork_end(int child);
/* main.c */
+extern int interp_dirfd;
extern unsigned long guest_stack_size;
/* user access */
diff --git a/linux-user/elfload.c b/linux-user/elfload.c
index 20f3d8c2c3..e70eebc7b8 100644
--- a/linux-user/elfload.c
+++ b/linux-user/elfload.c
@@ -2203,7 +2203,12 @@ static void load_elf_interp(const char *filename, struct
image_info *info,
{
int fd, retval;
- fd = open(path(filename), O_RDONLY);
+ if (interp_dirfd < 0
+ || filename[0] != '/'
+ || (fd = openat(interp_dirfd, filename + 1, O_RDONLY),
+ fd < 0 && errno == ENOENT)) {
+ fd = open(filename, O_RDONLY);
+ }
if (fd < 0) {
goto exit_perror;
}
diff --git a/linux-user/main.c b/linux-user/main.c
index 71696ed33d..391cc5ef82 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -98,6 +98,7 @@ 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;
/* XXX: on x86 MAP_GROWSDOWN only works if ESP <= address + 32, so
@@ -4307,7 +4308,7 @@ 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 11c9116c4a..007db808df 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -7228,7 +7228,13 @@ 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));
+ if (interp_dirfd < 0
+ || name[0] != '/'
+ || (ret = name_to_handle_at(interp_dirfd, name + 1, fh, &mid, flags),
+ ret < 0 && errno == ENOENT)) {
+ ret = name_to_handle_at(dirfd, name, fh, &mid, flags);
+ }
+ ret = get_errno(ret);
unlock_user(name, pathname, 0);
/* man name_to_handle_at(2):
@@ -7604,6 +7610,7 @@ 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);
@@ -7643,7 +7650,13 @@ static int do_openat(void *cpu_env, int dirfd, const
char *pathname, int flags,
return fd;
}
- return safe_openat(dirfd, path(pathname), flags, mode);
+ if (interp_dirfd < 0
+ || pathname[0] != '/'
+ || (ret = safe_openat(interp_dirfd, pathname + 1, flags, mode),
+ ret < 0 && errno == ENOENT)) {
+ ret = safe_openat(dirfd, pathname, flags, mode);
+ }
+ return ret;
}
#define TIMER_MAGIC 0x0caf0000
@@ -7729,6 +7742,7 @@ 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
@@ -8260,10 +8274,17 @@ 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);
+ }
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = futimesat(interp_dirfd, fn + 1, tvp),
+ ret < 0 && errno == ENOENT)) {
+ ret = futimesat(arg1, fn, tvp);
+ }
+ ret = get_errno(ret);
+ unlock_user(fn, arg2, 0);
}
break;
#endif
@@ -8277,18 +8298,32 @@ 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);
+ }
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = faccessat(interp_dirfd, fn + 1, arg2, 0),
+ ret < 0 && errno == ENOENT)) {
+ ret = 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);
+ }
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = faccessat(interp_dirfd, fn + 1, arg3, 0),
+ ret < 0 && errno == ENOENT)) {
+ ret = faccessat(arg1, fn, arg3, 0);
+ }
+ ret = get_errno(ret);
+ unlock_user(fn, arg2, 0);
break;
#endif
#ifdef TARGET_NR_nice /* not on alpha */
@@ -8415,7 +8450,7 @@ abi_long do_syscall(void *cpu_env, int num, abi_long arg1,
} else {
if (!(p = lock_user_string(arg1)))
goto efault;
- ret = get_errno(acct(path(p)));
+ ret = get_errno(acct(p));
unlock_user(p, arg1, 0);
}
break;
@@ -9181,14 +9216,14 @@ 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. */
@@ -9202,10 +9237,16 @@ 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));
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = readlinkat(interp_dirfd, fn + 1, p2, arg3),
+ ret < 0 && errno == ENOENT)) {
+ ret = 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
@@ -9213,20 +9254,26 @@ 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));
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = readlinkat(interp_dirfd, fn + 1, p2, arg4),
+ ret < 0 && errno == ENOENT)) {
+ ret = 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
@@ -9395,7 +9442,7 @@ abi_long do_syscall(void *cpu_env, int num, abi_long arg1,
case TARGET_NR_statfs:
if (!(p = lock_user_string(arg1)))
goto efault;
- ret = get_errno(statfs(path(p), &stfs));
+ ret = get_errno(statfs(p, &stfs));
unlock_user(p, arg1, 0);
convert_statfs:
if (!is_error(ret)) {
@@ -9425,7 +9472,7 @@ abi_long do_syscall(void *cpu_env, int num, abi_long arg1,
case TARGET_NR_statfs64:
if (!(p = lock_user_string(arg1)))
goto efault;
- ret = get_errno(statfs(path(p), &stfs));
+ ret = get_errno(statfs(p, &stfs));
unlock_user(p, arg1, 0);
convert_statfs64:
if (!is_error(ret)) {
@@ -9654,18 +9701,32 @@ 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);
+ }
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = fstatat(interp_dirfd, fn + 1, &st, 0),
+ ret < 0 && errno == ENOENT)) {
+ ret = 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);
+ }
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = fstatat(interp_dirfd, fn + 1, &st, AT_SYMLINK_NOFOLLOW),
+ ret < 0 && errno == ENOENT)) {
+ ret = lstat(fn, &st);
+ }
+ ret = get_errno(ret);
+ unlock_user(fn, arg1, 0);
goto do_stat;
#endif
case TARGET_NR_fstat:
@@ -10745,20 +10806,34 @@ 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);
+ }
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = fstatat(interp_dirfd, fn + 1, &st, 0),
+ ret < 0 && errno == ENOENT)) {
+ ret = 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);
+ }
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = fstatat(interp_dirfd, fn + 1, &st, AT_SYMLINK_NOFOLLOW),
+ ret < 0 && errno == ENOENT)) {
+ ret = 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;
@@ -10777,9 +10852,17 @@ 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));
+ }
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = fstatat(interp_dirfd, fn + 1, &st, arg4),
+ ret < 0 && errno == ENOENT)) {
+ ret = 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;
@@ -11776,12 +11859,17 @@ 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);
+ if (interp_dirfd < 0
+ || fn[0] != '/'
+ || (ret = sys_utimensat(interp_dirfd, fn + 1, tsp, arg4),
+ ret < 0 && errno == ENOENT)) {
+ ret = sys_utimensat(arg1, fn, tsp, arg4);
+ }
+ ret = get_errno(ret);
+ unlock_user(fn, arg2, 0);
}
}
break;
@@ -11811,7 +11899,7 @@ abi_long do_syscall(void *cpu_env, int num, abi_long
arg1,
#if defined(TARGET_NR_inotify_add_watch) && defined(__NR_inotify_add_watch)
case TARGET_NR_inotify_add_watch:
p = lock_user_string(arg2);
- ret = get_errno(sys_inotify_add_watch(arg1, path(p), arg3));
+ ret = get_errno(sys_inotify_add_watch(arg1, p, arg3));
unlock_user(p, arg2, 0);
break;
#endif
--
2.14.3
