cosmopolitan/libc/calls/unveil.c
Justine Tunney 957c61cbbf
Release Cosmopolitan v3.3
This change upgrades to GCC 12.3 and GNU binutils 2.42. The GNU linker
appears to have changed things so that only a single de-duplicated str
table is present in the binary, and it gets placed wherever the linker
wants, regardless of what the linker script says. To cope with that we
need to stop using .ident to embed licenses. As such, this change does
significant work to revamp how third party licenses are defined in the
codebase, using `.section .notice,"aR",@progbits`.

This new GCC 12.3 toolchain has support for GNU indirect functions. It
lets us support __target_clones__ for the first time. This is used for
optimizing the performance of libc string functions such as strlen and
friends so far on x86, by ensuring AVX systems favor a second codepath
that uses VEX encoding. It shaves some latency off certain operations.
It's a useful feature to have for scientific computing for the reasons
explained by the test/libcxx/openmp_test.cc example which compiles for
fifteen different microarchitectures. Thanks to the upgrades, it's now
also possible to use newer instruction sets, such as AVX512FP16, VNNI.

Cosmo now uses the %gs register on x86 by default for TLS. Doing it is
helpful for any program that links `cosmo_dlopen()`. Such programs had
to recompile their binaries at startup to change the TLS instructions.
That's not great, since it means every page in the executable needs to
be faulted. The work of rewriting TLS-related x86 opcodes, is moved to
fixupobj.com instead. This is great news for MacOS x86 users, since we
previously needed to morph the binary every time for that platform but
now that's no longer necessary. The only platforms where we need fixup
of TLS x86 opcodes at runtime are now Windows, OpenBSD, and NetBSD. On
Windows we morph TLS to point deeper into the TIB, based on a TlsAlloc
assignment, and on OpenBSD/NetBSD we morph %gs back into %fs since the
kernels do not allow us to specify a value for the %gs register.

OpenBSD users are now required to use APE Loader to run Cosmo binaries
and assimilation is no longer possible. OpenBSD kernel needs to change
to allow programs to specify a value for the %gs register, or it needs
to stop marking executable pages loaded by the kernel as mimmutable().

This release fixes __constructor__, .ctor, .init_array, and lastly the
.preinit_array so they behave the exact same way as glibc.

We no longer use hex constants to define math.h symbols like M_PI.
2024-02-20 13:27:59 -08:00

447 lines
17 KiB
C

/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
│ vi: set et ft=c ts=2 sts=2 sw=2 fenc=utf-8 :vi │
╞══════════════════════════════════════════════════════════════════════════════╡
│ Copyright 2020 Justine Alexandra Roberts Tunney │
│ │
│ Permission to use, copy, modify, and/or distribute this software for │
│ any purpose with or without fee is hereby granted, provided that the │
│ above copyright notice and this permission notice appear in all copies. │
│ │
│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL │
│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED │
│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE │
│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL │
│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR │
│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER │
│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR │
│ PERFORMANCE OF THIS SOFTWARE. │
╚─────────────────────────────────────────────────────────────────────────────*/
#include "libc/assert.h"
#include "libc/calls/blockcancel.internal.h"
#include "libc/calls/calls.h"
#include "libc/calls/landlock.h"
#include "libc/calls/struct/bpf.internal.h"
#include "libc/calls/struct/filter.internal.h"
#include "libc/calls/struct/seccomp.internal.h"
#include "libc/calls/struct/stat.h"
#include "libc/calls/struct/stat.internal.h"
#include "libc/calls/syscall-sysv.internal.h"
#include "libc/calls/syscall_support-sysv.internal.h"
#include "libc/dce.h"
#include "libc/errno.h"
#include "libc/fmt/conv.h"
#include "libc/fmt/libgen.h"
#include "libc/intrin/strace.internal.h"
#include "libc/limits.h"
#include "libc/macros.internal.h"
#include "libc/nexgen32e/vendor.internal.h"
#include "libc/runtime/internal.h"
#include "libc/runtime/runtime.h"
#include "libc/runtime/stack.h"
#include "libc/str/str.h"
#include "libc/sysv/consts/at.h"
#include "libc/sysv/consts/audit.h"
#include "libc/sysv/consts/f.h"
#include "libc/sysv/consts/fd.h"
#include "libc/sysv/consts/nrlinux.h"
#include "libc/sysv/consts/o.h"
#include "libc/sysv/consts/pr.h"
#include "libc/sysv/consts/s.h"
#include "libc/sysv/errfuns.h"
#include "libc/thread/tls.h"
#ifdef __x86_64__
#define ARCHITECTURE AUDIT_ARCH_X86_64
#elif defined(__aarch64__)
#define ARCHITECTURE AUDIT_ARCH_AARCH64
#else
#error "unsupported architecture"
#endif
#define OFF(f) offsetof(struct seccomp_data, f)
#define UNVEIL_READ \
(LANDLOCK_ACCESS_FS_READ_FILE | LANDLOCK_ACCESS_FS_READ_DIR | \
LANDLOCK_ACCESS_FS_REFER)
#define UNVEIL_WRITE \
(LANDLOCK_ACCESS_FS_WRITE_FILE | LANDLOCK_ACCESS_FS_TRUNCATE)
#define UNVEIL_EXEC (LANDLOCK_ACCESS_FS_EXECUTE)
#define UNVEIL_CREATE \
(LANDLOCK_ACCESS_FS_MAKE_CHAR | LANDLOCK_ACCESS_FS_MAKE_DIR | \
LANDLOCK_ACCESS_FS_MAKE_REG | LANDLOCK_ACCESS_FS_MAKE_SOCK | \
LANDLOCK_ACCESS_FS_MAKE_FIFO | LANDLOCK_ACCESS_FS_MAKE_BLOCK | \
LANDLOCK_ACCESS_FS_MAKE_SYM)
#define FILE_BITS \
(LANDLOCK_ACCESS_FS_READ_FILE | LANDLOCK_ACCESS_FS_WRITE_FILE | \
LANDLOCK_ACCESS_FS_EXECUTE)
static const struct sock_filter kUnveilBlacklistAbiVersionBelow3[] = {
BPF_STMT(BPF_LD | BPF_W | BPF_ABS, OFF(arch)),
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, ARCHITECTURE, 1, 0),
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL_PROCESS),
BPF_STMT(BPF_LD | BPF_W | BPF_ABS, OFF(nr)),
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_linux_truncate, 1, 0),
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_linux_setxattr, 0, 1),
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ERRNO | (1 & SECCOMP_RET_DATA)),
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW),
};
static const struct sock_filter kUnveilBlacklistLatestAbi[] = {
BPF_STMT(BPF_LD | BPF_W | BPF_ABS, OFF(arch)),
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, ARCHITECTURE, 1, 0),
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_KILL_PROCESS),
BPF_STMT(BPF_LD | BPF_W | BPF_ABS, OFF(nr)),
BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, __NR_linux_setxattr, 0, 1),
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ERRNO | (1 & SECCOMP_RET_DATA)),
BPF_STMT(BPF_RET | BPF_K, SECCOMP_RET_ALLOW),
};
static int landlock_abi_version;
static int landlock_abi_errno;
__attribute__((__constructor__(40))) textstartup void init_landlock_version() {
int e = errno;
landlock_abi_version =
landlock_create_ruleset(0, 0, LANDLOCK_CREATE_RULESET_VERSION);
landlock_abi_errno = errno;
errno = e;
}
/**
* Long living state for landlock calls.
* fs_mask is set to use all the access rights from the latest landlock ABI.
* On init, the current supported abi is checked and unavailable rights are
* masked off.
*
* As of 6.2, the latest abi is v3.
*
* TODO:
* - Integrate with pledge and remove the file access?
* - Stuff state into the .protected section?
*/
_Thread_local static struct {
uint64_t fs_mask;
int fd;
} State;
static int unveil_final(void) {
int e, rc;
struct sock_fprog sandbox = {
.filter = kUnveilBlacklistLatestAbi,
.len = ARRAYLEN(kUnveilBlacklistLatestAbi),
};
if (landlock_abi_version < 3) {
sandbox = (struct sock_fprog){
.filter = kUnveilBlacklistAbiVersionBelow3,
.len = ARRAYLEN(kUnveilBlacklistAbiVersionBelow3),
};
}
e = errno;
prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
errno = e;
if ((rc = landlock_restrict_self(State.fd, 0)) != -1 &&
(rc = sys_close(State.fd)) != -1 &&
(rc = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &sandbox)) != -1) {
State.fd = 0;
}
return rc;
}
static int err_close(int rc, int fd) {
int serrno = errno;
sys_close(fd);
errno = serrno;
return rc;
}
static int unveil_init(void) {
int rc, fd;
State.fs_mask = UNVEIL_READ | UNVEIL_WRITE | UNVEIL_EXEC | UNVEIL_CREATE;
if (landlock_abi_version == -1) {
errno = landlock_abi_errno;
if (errno == EOPNOTSUPP) {
errno = ENOSYS;
}
return -1;
}
if (landlock_abi_version < 2) {
State.fs_mask &= ~LANDLOCK_ACCESS_FS_REFER;
}
if (landlock_abi_version < 3) {
State.fs_mask &= ~LANDLOCK_ACCESS_FS_TRUNCATE;
}
const struct landlock_ruleset_attr attr = {
.handled_access_fs = State.fs_mask,
};
// [undocumented] landlock_create_ruleset() always returns O_CLOEXEC
// assert(__sys_fcntl(rc, F_GETFD) == FD_CLOEXEC);
if ((rc = landlock_create_ruleset(&attr, sizeof(attr), 0)) < 0) return -1;
// grant file descriptor a higher number that's less likely to interfere
if ((fd = __sys_fcntl(rc, F_DUPFD_CLOEXEC, 100)) == -1) {
return err_close(-1, rc);
}
if (sys_close(rc) == -1) {
return err_close(-1, fd);
}
State.fd = fd;
return 0;
}
int sys_unveil_linux(const char *path, const char *permissions) {
#pragma GCC push_options
#pragma GCC diagnostic ignored "-Wframe-larger-than="
struct {
char lbuf[PATH_MAX];
char buf1[PATH_MAX];
char buf2[PATH_MAX];
char buf3[PATH_MAX];
char buf4[PATH_MAX];
} b;
CheckLargeStackAllocation(&b, sizeof(b));
#pragma GCC pop_options
int rc;
const char *dir;
const char *last;
const char *next;
if (!State.fd && (rc = unveil_init()) == -1) return rc;
if ((path && !permissions) || (!path && permissions)) return einval();
if (!path && !permissions) return unveil_final();
struct landlock_path_beneath_attr pb = {0};
for (const char *c = permissions; *c != '\0'; c++) {
switch (*c) {
case 'r':
pb.allowed_access |= UNVEIL_READ;
break;
case 'w':
pb.allowed_access |= UNVEIL_WRITE;
break;
case 'x':
pb.allowed_access |= UNVEIL_EXEC;
break;
case 'c':
pb.allowed_access |= UNVEIL_CREATE;
break;
default:
return einval();
}
}
pb.allowed_access &= State.fs_mask;
// landlock exposes all metadata, so we only technically need to add
// realpath(path) to the ruleset. however a corner case exists where
// it isn't valid, e.g. /dev/stdin -> /proc/2834/fd/pipe:[51032], so
// we'll need to work around this, by adding the path which is valid
if (strlen(path) + 1 > PATH_MAX) return enametoolong();
last = path;
next = path;
for (int i = 0;; ++i) {
if (i == 64) {
// give up
return eloop();
}
int err = errno;
if ((rc = sys_readlinkat(AT_FDCWD, next, b.lbuf, PATH_MAX)) != -1) {
if (rc < PATH_MAX) {
// we need to nul-terminate
b.lbuf[rc] = 0;
// last = next
strcpy(b.buf1, next);
last = b.buf1;
// next = join(dirname(next), link)
strcpy(b.buf2, next);
dir = dirname(b.buf2);
if ((next = __join_paths(b.buf3, PATH_MAX, dir, b.lbuf))) {
// next now points to either: buf3, buf2, lbuf, rodata
strcpy(b.buf4, next);
next = b.buf4;
} else {
return enametoolong();
}
} else {
// symbolic link data was too long
return enametoolong();
}
} else if (errno == EINVAL) {
// next wasn't a symbolic link
errno = err;
path = next;
break;
} else if (i && (errno == ENOENT || errno == ENOTDIR)) {
// next is a broken symlink, use last
errno = err;
path = last;
break;
} else {
// readlink failed for some other reason
return -1;
}
}
// now we can open the path
BLOCK_CANCELATION;
rc = sys_openat(AT_FDCWD, path, O_PATH | O_NOFOLLOW | O_CLOEXEC, 0);
ALLOW_CANCELATION;
if (rc == -1) return rc;
pb.parent_fd = rc;
struct stat st;
if ((rc = sys_fstat(pb.parent_fd, &st)) == -1) {
return err_close(rc, pb.parent_fd);
}
if (!S_ISDIR(st.st_mode)) {
pb.allowed_access &= FILE_BITS;
}
if ((rc = landlock_add_rule(State.fd, LANDLOCK_RULE_PATH_BENEATH, &pb, 0))) {
return err_close(rc, pb.parent_fd);
}
sys_close(pb.parent_fd);
return rc;
}
/**
* Makes files accessible, e.g.
*
* unveil(".", "r"); // current directory + children are visible
* unveil("/etc", "r"); // make /etc readable too
* unveil(0, 0); // commit and lock policy
*
* Unveiling restricts a view of the filesystem to a set of allowed
* paths with specific privileges.
*
* Once you start using unveil(), the entire file system is considered
* hidden. You then specify, by repeatedly calling unveil(), which paths
* should become unhidden. When you're finished, you call `unveil(0,0)`
* which commits your policy.
*
* This function requires OpenBSD or Linux 5.13+ (2022+). If the kernel
* support isn't available (or we're in an emulator like Qemu or Blink)
* then zero is returned and nothing happens (instead of raising ENOSYS)
* because the files are still unveiled. Use `unveil("", 0)` to feature
* check the host system, which is defined as a no-op that'll fail if
* the host system doesn't have the necessary features that allow
* unveil() impose bona-fide security restrictions. Otherwise, if
* everything is good, a return value `>=0` is returned, where `0` means
* OpenBSD, and `>=1` means Linux with Landlock LSM, in which case the
* return code shall be the maximum supported Landlock ABI version.
*
* There are some differences between unveil() on Linux versus OpenBSD.
*
* 1. Build your policy and lock it in one go. On OpenBSD, policies take
* effect immediately and may evolve as you continue to call unveil()
* but only in a more restrictive direction. On Linux, nothing will
* happen until you call `unveil(0,0)` which commits and locks.
*
* 2. Try not to overlap directory trees. On OpenBSD, if directory trees
* overlap, then the most restrictive policy will be used for a given
* file. On Linux overlapping may result in a less restrictive policy
* and possibly even undefined behavior.
*
* 3. OpenBSD and Linux disagree on error codes. On OpenBSD, accessing
* paths outside of the allowed set raises ENOENT, and accessing ones
* with incorrect permissions raises EACCES. On Linux, both these
* cases raise EACCES.
*
* 4. Unlike OpenBSD, Linux does nothing to conceal the existence of
* paths. Even with an unveil() policy in place, it's still possible
* to access the metadata of all files using functions like stat()
* and open(O_PATH), provided you know the path. A sandboxed process
* can always, for example, determine how many bytes of data are in
* /etc/passwd, even if the file isn't readable. But it's still not
* possible to use opendir() and go fishing for paths which weren't
* previously known.
*
* 5. Use ftruncate() rather than truncate() if you wish for portability
* to Linux kernels versions released before February 2022. One issue
* Landlock hadn't addressed as of ABI version 2 was restrictions
* over truncate() and setxattr() which could permit certain kinds of
* modifications to files outside the sandbox. When your policy is
* committed, we install a SECCOMP BPF filter to disable those calls,
* however similar trickery may be possible through other unaddressed
* calls like ioctl(). Using the pledge() function in addition to
* unveil() will solve this, since it installs a strong system call
* access policy. Linux 6.2 has improved this situation with Landlock
* ABI v3, which added the ability to control truncation operations -
* this means the SECCOMP BPF filter will only disable truncate() on
* Linux 6.1 or older.
*
* 6. Set your process-wide policy at startup from the main thread. On
* OpenBSD unveil() will apply process-wide even when called from a
* child thread; whereas with Linux, calling unveil() from a thread
* will cause your ruleset to only apply to that thread in addition
* to any descendent threads it creates.
*
* 7. Always specify at least one path. OpenBSD has unclear semantics
* when `unveil(0,0)` is used without any previous calls.
*
* 8. On OpenBSD calling `unveil(0,0)` will prevent unveil() from being
* used again. On Linux this is allowed, because Landlock is able to
* do that securely, i.e. the second ruleset can only be a subset of
* the previous ones.
*
* This system call is supported natively on OpenBSD and polyfilled on
* Linux using the Landlock LSM[1].
*
* @param path is the file or directory to unveil
* @param permissions is a string consisting of zero or more of the
* following characters:
*
* - 'r' makes `path` available for read-only path operations,
* corresponding to the pledge promise "rpath".
*
* - `w` makes `path` available for write operations, corresponding
* to the pledge promise "wpath".
*
* - `x` makes `path` available for execute operations,
* corresponding to the pledge promises "exec" and "execnative".
*
* - `c` allows `path` to be created and removed, corresponding to
* the pledge promise "cpath".
*
* @return 0 on success, or -1 w/ errno; note: if `unveil("",0)` is used
* to perform a feature check, then on Linux a value greater than 0
* shall be returned which is the supported Landlock ABI version
* @raise EPERM if unveil() is called after locking
* @raise EINVAL if one argument is set and the other is not
* @raise EINVAL if an invalid character in `permissions` was found
* @raise ENOSYS if `unveil("",0)` was used and security isn't possible
* @raise EOPNOTSUPP if `unveil("",0)` was used and Landlock LSM is disabled
* @note on Linux this function requires Linux Kernel 5.13+ and version 6.2+
* to properly support truncation operations
* @see [1] https://docs.kernel.org/userspace-api/landlock.html
*/
int unveil(const char *path, const char *permissions) {
int e, rc;
e = errno;
if (path && !*path) {
// OpenBSD will always fail on both unveil("",0) and unveil("",""),
// since an empty `path` is invalid and `permissions` is mandatory.
// Cosmopolitan Libc uses it as a feature check convention, to test
// if the host environment enables unveil() to impose true security
// restrictions because the default behavior is to silently succeed
// so that programs will err on the side of working if distributed.
if (permissions) return einval();
if (IsOpenbsd()) return 0;
if (landlock_abi_version != -1) {
unassert(landlock_abi_version >= 1);
return landlock_abi_version;
} else {
unassert(landlock_abi_errno);
errno = landlock_abi_errno;
return -1;
}
} else if (!IsTiny() && IsGenuineBlink()) {
rc = 0; // blink doesn't support landlock; avoid noisy log warnings
} else if (IsLinux()) {
rc = sys_unveil_linux(path, permissions);
} else {
rc = sys_unveil(path, permissions);
}
if (rc == -1 && errno == ENOSYS) {
errno = e;
rc = 0;
}
STRACE("unveil(%#s, %#s) → %d% m", path, permissions, rc);
return rc;
}