X-Git-Url: http://git.ipfire.org/?a=blobdiff_plain;f=src%2Fnspawn%2Fnspawn-mount.c;h=72c007f204c657472dc153f3b247bb6f0a5cc025;hb=a4c35b6b4d8457e15caa3fe18ce8050918932d00;hp=c8e627ac78b41fba114f43ce53c969503cdf903d;hpb=b591613f2e17476ed8da29f4089c410b010c0192;p=thirdparty%2Fsystemd.git diff --git a/src/nspawn/nspawn-mount.c b/src/nspawn/nspawn-mount.c index c8e627ac78b..72c007f204c 100644 --- a/src/nspawn/nspawn-mount.c +++ b/src/nspawn/nspawn-mount.c @@ -1,5 +1,3 @@ -/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/ - /*** This file is part of systemd. @@ -23,8 +21,9 @@ #include #include "alloc-util.h" -#include "cgroup-util.h" #include "escape.h" +#include "fd-util.h" +#include "fileio.h" #include "fs-util.h" #include "label.h" #include "mkdir.h" @@ -48,7 +47,7 @@ CustomMount* custom_mount_add(CustomMount **l, unsigned *n, CustomMountType t) { assert(t >= 0); assert(t < _CUSTOM_MOUNT_TYPE_MAX); - c = realloc(*l, (*n + 1) * sizeof(CustomMount)); + c = realloc_multiply(*l, (*n + 1), sizeof(CustomMount)); if (!c) return NULL; @@ -76,13 +75,18 @@ void custom_mount_free_all(CustomMount *l, unsigned n) { free(m->work_dir); } + if (m->rm_rf_tmpdir) { + (void) rm_rf(m->rm_rf_tmpdir, REMOVE_ROOT|REMOVE_PHYSICAL); + free(m->rm_rf_tmpdir); + } + strv_free(m->lower); } free(l); } -int custom_mount_compare(const void *a, const void *b) { +static int custom_mount_compare(const void *a, const void *b) { const CustomMount *x = a, *y = b; int r; @@ -98,6 +102,109 @@ int custom_mount_compare(const void *a, const void *b) { return 0; } +static bool source_path_is_valid(const char *p) { + assert(p); + + if (*p == '+') + p++; + + return path_is_absolute(p); +} + +static char *resolve_source_path(const char *dest, const char *source) { + + if (!source) + return NULL; + + if (source[0] == '+') + return prefix_root(dest, source + 1); + + return strdup(source); +} + +int custom_mount_prepare_all(const char *dest, CustomMount *l, unsigned n) { + unsigned i; + int r; + + /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the + * parent process, so that we know the temporary directories to remove on exit before we fork off the + * children. */ + + assert(l || n == 0); + + /* Order the custom mounts, and make sure we have a working directory */ + qsort_safe(l, n, sizeof(CustomMount), custom_mount_compare); + + for (i = 0; i < n; i++) { + CustomMount *m = l + i; + + if (m->source) { + char *s; + + s = resolve_source_path(dest, m->source); + if (!s) + return log_oom(); + + free(m->source); + m->source = s; + } else { + /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */ + + m->rm_rf_tmpdir = strdup("/var/tmp/nspawn-temp-XXXXXX"); + if (!m->rm_rf_tmpdir) + return log_oom(); + + if (!mkdtemp(m->rm_rf_tmpdir)) { + m->rm_rf_tmpdir = mfree(m->rm_rf_tmpdir); + return log_error_errno(errno, "Failed to acquire temporary directory: %m"); + } + + m->source = strjoin(m->rm_rf_tmpdir, "/src"); + if (!m->source) + return log_oom(); + + if (mkdir(m->source, 0755) < 0) + return log_error_errno(errno, "Failed to create %s: %m", m->source); + } + + if (m->type == CUSTOM_MOUNT_OVERLAY) { + char **j; + + STRV_FOREACH(j, m->lower) { + char *s; + + s = resolve_source_path(dest, *j); + if (!s) + return log_oom(); + + free(*j); + *j = s; + } + + if (m->work_dir) { + char *s; + + s = resolve_source_path(dest, m->work_dir); + if (!s) + return log_oom(); + + free(m->work_dir); + m->work_dir = s; + } else { + assert(m->source); + + r = tempfn_random(m->source, NULL, &m->work_dir); + if (r < 0) + return log_error_errno(r, "Failed to acquire working directory: %m"); + } + + (void) mkdir_label(m->work_dir, 0700); + } + } + + return 0; +} + int bind_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) { _cleanup_free_ char *source = NULL, *destination = NULL, *opts = NULL; const char *p = s; @@ -112,20 +219,20 @@ int bind_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only return r; if (r == 0) return -EINVAL; - if (r == 1) { - destination = strdup(source); + destination = strdup(source[0] == '+' ? source+1 : source); if (!destination) return -ENOMEM; } - if (r == 2 && !isempty(p)) { opts = strdup(p); if (!opts) return -ENOMEM; } - if (!path_is_absolute(source)) + if (isempty(source)) + source = NULL; + else if (!source_path_is_valid(source)) return -EINVAL; if (!path_is_absolute(destination)) @@ -133,7 +240,7 @@ int bind_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only m = custom_mount_add(l, n, CUSTOM_MOUNT_BIND); if (!m) - return log_oom(); + return -ENOMEM; m->source = source; m->destination = destination; @@ -181,22 +288,87 @@ int tmpfs_mount_parse(CustomMount **l, unsigned *n, const char *s) { return 0; } +int overlay_mount_parse(CustomMount **l, unsigned *n, const char *s, bool read_only) { + _cleanup_free_ char *upper = NULL, *destination = NULL; + _cleanup_strv_free_ char **lower = NULL; + CustomMount *m; + int k; + + k = strv_split_extract(&lower, s, ":", EXTRACT_DONT_COALESCE_SEPARATORS); + if (k < 0) + return k; + if (k < 2) + return -EADDRNOTAVAIL; + if (k == 2) { + /* If two parameters are specified, the first one is the lower, the second one the upper directory. And + * we'll also define the destination mount point the same as the upper. */ + + if (!source_path_is_valid(lower[0]) || + !source_path_is_valid(lower[1])) + return -EINVAL; + + upper = lower[1]; + lower[1] = NULL; + + destination = strdup(upper[0] == '+' ? upper+1 : upper); /* take the destination without "+" prefix */ + if (!destination) + return -ENOMEM; + } else { + char **i; + + /* If more than two parameters are specified, the last one is the destination, the second to last one + * the "upper", and all before that the "lower" directories. */ + + destination = lower[k - 1]; + upper = lower[k - 2]; + lower[k - 2] = NULL; + + STRV_FOREACH(i, lower) + if (!source_path_is_valid(*i)) + return -EINVAL; + + /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory + * in /var/tmp */ + if (isempty(upper)) + upper = NULL; + else if (!source_path_is_valid(upper)) + return -EINVAL; + + if (!path_is_absolute(destination)) + return -EINVAL; + } + + m = custom_mount_add(l, n, CUSTOM_MOUNT_OVERLAY); + if (!m) + return -ENOMEM; + + m->destination = destination; + m->source = upper; + m->lower = lower; + m->read_only = read_only; + + upper = destination = NULL; + lower = NULL; + + return 0; +} + static int tmpfs_patch_options( const char *options, - bool userns, uid_t uid_shift, uid_t uid_range, + bool userns, + uid_t uid_shift, uid_t uid_range, + bool patch_ids, const char *selinux_apifs_context, char **ret) { char *buf = NULL; - if (userns && uid_shift != 0) { + if ((userns && uid_shift != 0) || patch_ids) { assert(uid_shift != UID_INVALID); - if (options) - (void) asprintf(&buf, "%s,uid=" UID_FMT ",gid=" UID_FMT, options, uid_shift, uid_shift); - else - (void) asprintf(&buf, "uid=" UID_FMT ",gid=" UID_FMT, uid_shift, uid_shift); - if (!buf) + if (asprintf(&buf, "%s%suid=" UID_FMT ",gid=" UID_FMT, + options ?: "", options ? "," : "", + uid_shift, uid_shift) < 0) return -ENOMEM; options = buf; @@ -206,27 +378,30 @@ static int tmpfs_patch_options( if (selinux_apifs_context) { char *t; - if (options) - t = strjoin(options, ",context=\"", selinux_apifs_context, "\"", NULL); - else - t = strjoin("context=\"", selinux_apifs_context, "\"", NULL); - if (!t) { - free(buf); + t = strjoin(options ?: "", options ? "," : "", + "context=\"", selinux_apifs_context, "\""); + free(buf); + if (!t) return -ENOMEM; - } - free(buf); buf = t; } #endif + if (!buf && options) { + buf = strdup(options); + if (!buf) + return -ENOMEM; + } *ret = buf; + return !!buf; } -int mount_sysfs(const char *dest) { +int mount_sysfs(const char *dest, MountSettingsMask mount_settings) { const char *full, *top, *x; int r; + unsigned long extra_flags = 0; top = prefix_roota(dest, "/sys"); r = path_check_fstype(top, SYSFS_MAGIC); @@ -243,8 +418,13 @@ int mount_sysfs(const char *dest) { (void) mkdir(full, 0755); - if (mount("sysfs", full, "sysfs", MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL) < 0) - return log_error_errno(errno, "Failed to mount sysfs to %s: %m", full); + if (mount_settings & MOUNT_APPLY_APIVFS_RO) + extra_flags |= MS_RDONLY; + + r = mount_verbose(LOG_ERR, "sysfs", full, "sysfs", + MS_NOSUID|MS_NOEXEC|MS_NODEV|extra_flags, NULL); + if (r < 0) + return r; FOREACH_STRING(x, "block", "bus", "class", "dev", "devices", "kernel") { _cleanup_free_ char *from = NULL, *to = NULL; @@ -259,31 +439,97 @@ int mount_sysfs(const char *dest) { (void) mkdir(to, 0755); - if (mount(from, to, NULL, MS_BIND, NULL) < 0) - return log_error_errno(errno, "Failed to mount /sys/%s into place: %m", x); + r = mount_verbose(LOG_ERR, from, to, NULL, MS_BIND, NULL); + if (r < 0) + return r; - if (mount(NULL, to, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL) < 0) - return log_error_errno(errno, "Failed to mount /sys/%s read-only: %m", x); + r = mount_verbose(LOG_ERR, NULL, to, NULL, + MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL); + if (r < 0) + return r; } - if (umount(full) < 0) - return log_error_errno(errno, "Failed to unmount %s: %m", full); + r = umount_verbose(full); + if (r < 0) + return r; if (rmdir(full) < 0) return log_error_errno(errno, "Failed to remove %s: %m", full); x = prefix_roota(top, "/fs/kdbus"); - (void) mkdir(x, 0755); + (void) mkdir_p(x, 0755); + + /* Create mountpoint for cgroups. Otherwise we are not allowed since we + * remount /sys read-only. + */ + if (cg_ns_supported()) { + x = prefix_roota(top, "/fs/cgroup"); + (void) mkdir_p(x, 0755); + } + + return mount_verbose(LOG_ERR, NULL, top, NULL, + MS_BIND|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT|extra_flags, NULL); +} + +static int mkdir_userns(const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) { + int r; + + assert(path); + + r = mkdir(path, mode); + if (r < 0 && errno != EEXIST) + return -errno; + + if ((mask & MOUNT_USE_USERNS) == 0) + return 0; + + if (mask & MOUNT_IN_USERNS) + return 0; - if (mount(NULL, top, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, NULL) < 0) - return log_error_errno(errno, "Failed to make %s read-only: %m", top); + r = lchown(path, uid_shift, uid_shift); + if (r < 0) + return -errno; return 0; } +static int mkdir_userns_p(const char *prefix, const char *path, mode_t mode, MountSettingsMask mask, uid_t uid_shift) { + const char *p, *e; + int r; + + assert(path); + + if (prefix && !path_startswith(path, prefix)) + return -ENOTDIR; + + /* create every parent directory in the path, except the last component */ + p = path + strspn(path, "/"); + for (;;) { + char t[strlen(path) + 1]; + + e = p + strcspn(p, "/"); + p = e + strspn(e, "/"); + + /* Is this the last component? If so, then we're done */ + if (*p == 0) + break; + + memcpy(t, path, e - path); + t[e-path] = 0; + + if (prefix && path_startswith(prefix, t)) + continue; + + r = mkdir_userns(t, mode, mask, uid_shift); + if (r < 0) + return r; + } + + return mkdir_userns(path, mode, mask, uid_shift); +} + int mount_all(const char *dest, - bool use_userns, bool in_userns, - bool use_netns, + MountSettingsMask mount_settings, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context) { @@ -293,45 +539,59 @@ int mount_all(const char *dest, const char *type; const char *options; unsigned long flags; - bool fatal; - bool in_userns; - bool use_netns; + MountSettingsMask mount_settings; } MountPoint; static const MountPoint mount_table[] = { - { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, true, true, false }, - { "/proc/sys", "/proc/sys", NULL, NULL, MS_BIND, true, true, false }, /* Bind mount first */ - { NULL, "/proc/sys", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, true, true, false }, /* Then, make it r/o */ - { "tmpfs", "/sys", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV, true, false, true }, - { "sysfs", "/sys", "sysfs", NULL, MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, true, false, false }, - { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME, true, false, false }, - { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true, false, false }, - { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, true, false, false }, - { "tmpfs", "/tmp", "tmpfs", "mode=1777", MS_STRICTATIME, true, false, false }, + /* inner child mounts */ + { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, MOUNT_FATAL|MOUNT_IN_USERNS }, + { "/proc/sys", "/proc/sys", NULL, NULL, MS_BIND, MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ...*/ + { "/proc/sys/net", "/proc/sys/net", NULL, NULL, MS_BIND, MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO|MOUNT_APPLY_APIVFS_NETNS }, /* (except for this) */ + { NULL, "/proc/sys", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, MOUNT_FATAL|MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* ... then, make it r/o */ + { "/proc/sysrq-trigger", "/proc/sysrq-trigger", NULL, NULL, MS_BIND, MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ...*/ + { NULL, "/proc/sysrq-trigger", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* ... then, make it r/o */ + + /* outer child mounts */ + { "tmpfs", "/tmp", "tmpfs", "mode=1777", MS_STRICTATIME, MOUNT_FATAL }, + { "tmpfs", "/sys", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV, MOUNT_FATAL|MOUNT_APPLY_APIVFS_NETNS }, + { "sysfs", "/sys", "sysfs", NULL, MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV, MOUNT_FATAL|MOUNT_APPLY_APIVFS_RO }, /* skipped if above was mounted */ + { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV, MOUNT_FATAL }, /* skipped if above was mounted */ + + { "tmpfs", "/dev", "tmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME, MOUNT_FATAL }, + { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME, MOUNT_FATAL }, + { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME, MOUNT_FATAL }, #ifdef HAVE_SELINUX - { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND, false, false, false }, /* Bind mount first */ - { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, false, false, false }, /* Then, make it r/o */ + { "/sys/fs/selinux", "/sys/fs/selinux", NULL, NULL, MS_BIND, 0 }, /* Bind mount first */ + { NULL, "/sys/fs/selinux", NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, 0 }, /* Then, make it r/o */ #endif }; unsigned k; int r; + bool use_userns = (mount_settings & MOUNT_USE_USERNS); + bool netns = (mount_settings & MOUNT_APPLY_APIVFS_NETNS); + bool ro = (mount_settings & MOUNT_APPLY_APIVFS_RO); + bool in_userns = (mount_settings & MOUNT_IN_USERNS); for (k = 0; k < ELEMENTSOF(mount_table); k++) { _cleanup_free_ char *where = NULL, *options = NULL; const char *o; + bool fatal = (mount_table[k].mount_settings & MOUNT_FATAL); - if (in_userns != mount_table[k].in_userns) + if (in_userns != (bool)(mount_table[k].mount_settings & MOUNT_IN_USERNS)) continue; - if (!use_netns && mount_table[k].use_netns) + if (!netns && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_NETNS)) continue; - where = prefix_root(dest, mount_table[k].where); - if (!where) - return log_oom(); + if (!ro && (bool)(mount_table[k].mount_settings & MOUNT_APPLY_APIVFS_RO)) + continue; - r = path_is_mount_point(where, AT_SYMLINK_FOLLOW); + r = chase_symlinks(mount_table[k].where, dest, CHASE_NONEXISTENT|CHASE_PREFIX_ROOT, &where); + if (r < 0) + return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, mount_table[k].where); + + r = path_is_mount_point(where, NULL, 0); if (r < 0 && r != -ENOENT) return log_error_errno(r, "Failed to detect whether %s is a mount point: %m", where); @@ -339,35 +599,35 @@ int mount_all(const char *dest, if (mount_table[k].what && r > 0) continue; - r = mkdir_p(where, 0755); - if (r < 0) { - if (mount_table[k].fatal) + r = mkdir_userns_p(dest, where, 0755, mount_settings, uid_shift); + if (r < 0 && r != -EEXIST) { + if (fatal) return log_error_errno(r, "Failed to create directory %s: %m", where); - log_warning_errno(r, "Failed to create directory %s: %m", where); + log_debug_errno(r, "Failed to create directory %s: %m", where); continue; } o = mount_table[k].options; if (streq_ptr(mount_table[k].type, "tmpfs")) { - r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, selinux_apifs_context, &options); + if (in_userns) + r = tmpfs_patch_options(o, use_userns, 0, uid_range, true, selinux_apifs_context, &options); + else + r = tmpfs_patch_options(o, use_userns, uid_shift, uid_range, false, selinux_apifs_context, &options); if (r < 0) return log_oom(); if (r > 0) o = options; } - if (mount(mount_table[k].what, - where, - mount_table[k].type, - mount_table[k].flags, - o) < 0) { - - if (mount_table[k].fatal) - return log_error_errno(errno, "mount(%s) failed: %m", where); - - log_warning_errno(errno, "mount(%s) failed, ignoring: %m", where); - } + r = mount_verbose(fatal ? LOG_ERR : LOG_DEBUG, + mount_table[k].what, + where, + mount_table[k].type, + mount_table[k].flags, + o); + if (r < 0 && fatal) + return r; } return 0; @@ -377,12 +637,14 @@ static int parse_mount_bind_options(const char *options, unsigned long *mount_fl const char *p = options; unsigned long flags = *mount_flags; char *opts = NULL; + int r; assert(options); for (;;) { _cleanup_free_ char *word = NULL; - int r = extract_first_word(&p, &word, ",", 0); + + r = extract_first_word(&p, &word, ",", 0); if (r < 0) return log_error_errno(r, "Failed to extract mount option: %m"); if (r == 0) @@ -406,12 +668,13 @@ static int parse_mount_bind_options(const char *options, unsigned long *mount_fl } static int mount_bind(const char *dest, CustomMount *m) { - struct stat source_st, dest_st; - const char *where; + + _cleanup_free_ char *mount_opts = NULL, *where = NULL; unsigned long mount_flags = MS_BIND | MS_REC; - _cleanup_free_ char *mount_opts = NULL; + struct stat source_st, dest_st; int r; + assert(dest); assert(m); if (m->options) { @@ -423,9 +686,14 @@ static int mount_bind(const char *dest, CustomMount *m) { if (stat(m->source, &source_st) < 0) return log_error_errno(errno, "Failed to stat %s: %m", m->source); - where = prefix_roota(dest, m->destination); + r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where); + if (r < 0) + return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination); + if (r > 0) { /* Path exists already? */ + + if (stat(where, &dest_st) < 0) + return log_error_errno(errno, "Failed to stat %s: %m", where); - if (stat(where, &dest_st) >= 0) { if (S_ISDIR(source_st.st_mode) && !S_ISDIR(dest_st.st_mode)) { log_error("Cannot bind mount directory %s on file %s.", m->source, where); return -EINVAL; @@ -436,30 +704,30 @@ static int mount_bind(const char *dest, CustomMount *m) { return -EINVAL; } - } else if (errno == ENOENT) { + } else { /* Path doesn't exist yet? */ r = mkdir_parents_label(where, 0755); if (r < 0) return log_error_errno(r, "Failed to make parents of %s: %m", where); - } else { - return log_error_errno(errno, "Failed to stat %s: %m", where); - } - /* Create the mount point. Any non-directory file can be - * mounted on any non-directory file (regular, fifo, socket, - * char, block). - */ - if (S_ISDIR(source_st.st_mode)) - r = mkdir_label(where, 0755); - else - r = touch(where); - if (r < 0 && r != -EEXIST) - return log_error_errno(r, "Failed to create mount point %s: %m", where); + /* Create the mount point. Any non-directory file can be + * mounted on any non-directory file (regular, fifo, socket, + * char, block). + */ + if (S_ISDIR(source_st.st_mode)) + r = mkdir_label(where, 0755); + else + r = touch(where); + if (r < 0) + return log_error_errno(r, "Failed to create mount point %s: %m", where); + + } - if (mount(m->source, where, NULL, mount_flags, mount_opts) < 0) - return log_error_errno(errno, "mount(%s) failed: %m", where); + r = mount_verbose(LOG_ERR, m->source, where, NULL, mount_flags, mount_opts); + if (r < 0) + return r; if (m->read_only) { - r = bind_remount_recursive(where, true); + r = bind_remount_recursive(where, true, NULL); if (r < 0) return log_error_errno(r, "Read-only bind mount failed: %m"); } @@ -473,31 +741,31 @@ static int mount_tmpfs( bool userns, uid_t uid_shift, uid_t uid_range, const char *selinux_apifs_context) { - const char *where, *options; - _cleanup_free_ char *buf = NULL; + const char *options; + _cleanup_free_ char *buf = NULL, *where = NULL; int r; assert(dest); assert(m); - where = prefix_roota(dest, m->destination); - - r = mkdir_p_label(where, 0755); - if (r < 0 && r != -EEXIST) - return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where); + r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where); + if (r < 0) + return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination); + if (r == 0) { /* Doesn't exist yet? */ + r = mkdir_p_label(where, 0755); + if (r < 0) + return log_error_errno(r, "Creating mount point for tmpfs %s failed: %m", where); + } - r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, selinux_apifs_context, &buf); + r = tmpfs_patch_options(m->options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf); if (r < 0) return log_oom(); options = r > 0 ? buf : m->options; - if (mount("tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options) < 0) - return log_error_errno(errno, "tmpfs mount to %s failed: %m", where); - - return 0; + return mount_verbose(LOG_ERR, "tmpfs", where, "tmpfs", MS_NODEV|MS_STRICTATIME, options); } -static char *joined_and_escaped_lower_dirs(char * const *lower) { +static char *joined_and_escaped_lower_dirs(char **lower) { _cleanup_strv_free_ char **sv = NULL; sv = strv_copy(lower); @@ -513,18 +781,22 @@ static char *joined_and_escaped_lower_dirs(char * const *lower) { } static int mount_overlay(const char *dest, CustomMount *m) { - _cleanup_free_ char *lower = NULL; - const char *where, *options; + + _cleanup_free_ char *lower = NULL, *where = NULL, *escaped_source = NULL; + const char *options; int r; assert(dest); assert(m); - where = prefix_roota(dest, m->destination); - - r = mkdir_label(where, 0755); - if (r < 0 && r != -EEXIST) - return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where); + r = chase_symlinks(m->destination, dest, CHASE_PREFIX_ROOT|CHASE_NONEXISTENT, &where); + if (r < 0) + return log_error_errno(r, "Failed to resolve %s/%s: %m", dest, m->destination); + if (r == 0) { /* Doesn't exist yet? */ + r = mkdir_label(where, 0755); + if (r < 0) + return log_error_errno(r, "Creating mount point for overlay %s failed: %m", where); + } (void) mkdir_p_label(m->source, 0755); @@ -532,23 +804,15 @@ static int mount_overlay(const char *dest, CustomMount *m) { if (!lower) return log_oom(); - if (m->read_only) { - _cleanup_free_ char *escaped_source = NULL; - - escaped_source = shell_escape(m->source, ",:"); - if (!escaped_source) - return log_oom(); + escaped_source = shell_escape(m->source, ",:"); + if (!escaped_source) + return log_oom(); + if (m->read_only) options = strjoina("lowerdir=", escaped_source, ":", lower); - } else { - _cleanup_free_ char *escaped_source = NULL, *escaped_work_dir = NULL; - - assert(m->work_dir); - (void) mkdir_label(m->work_dir, 0700); + else { + _cleanup_free_ char *escaped_work_dir = NULL; - escaped_source = shell_escape(m->source, ",:"); - if (!escaped_source) - return log_oom(); escaped_work_dir = shell_escape(m->work_dir, ",:"); if (!escaped_work_dir) return log_oom(); @@ -556,10 +820,7 @@ static int mount_overlay(const char *dest, CustomMount *m) { options = strjoina("lowerdir=", lower, ",upperdir=", escaped_source, ",workdir=", escaped_work_dir); } - if (mount("overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options) < 0) - return log_error_errno(errno, "overlay mount to %s failed: %m", where); - - return 0; + return mount_verbose(LOG_ERR, "overlay", where, "overlay", m->read_only ? MS_RDONLY : 0, options); } int mount_custom( @@ -601,13 +862,62 @@ int mount_custom( return 0; } -static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controller, const char *hierarchy, bool read_only) { - char *to; +/* Retrieve existing subsystems. This function is called in a new cgroup + * namespace. + */ +static int get_controllers(Set *subsystems) { + _cleanup_fclose_ FILE *f = NULL; + char line[LINE_MAX]; + + assert(subsystems); + + f = fopen("/proc/self/cgroup", "re"); + if (!f) + return errno == ENOENT ? -ESRCH : -errno; + + FOREACH_LINE(line, f, return -errno) { + int r; + char *e, *l, *p; + + l = strchr(line, ':'); + if (!l) + continue; + + l++; + e = strchr(l, ':'); + if (!e) + continue; + + *e = 0; + + if (STR_IN_SET(l, "", "name=systemd")) + continue; + + p = strdup(l); + if (!p) + return -ENOMEM; + + r = set_consume(subsystems, p); + if (r < 0) + return r; + } + + return 0; +} + +static int mount_legacy_cgroup_hierarchy( + const char *dest, + const char *controller, + const char *hierarchy, + CGroupUnified unified_requested, + bool read_only) { + + const char *to, *fstype, *opts; int r; to = strjoina(strempty(dest), "/sys/fs/cgroup/", hierarchy); - r = path_is_mount_point(to, 0); + r = path_is_mount_point(to, dest, 0); if (r < 0 && r != -ENOENT) return log_error_errno(r, "Failed to determine if %s is mounted already: %m", to); if (r > 0) @@ -617,21 +927,143 @@ static int mount_legacy_cgroup_hierarchy(const char *dest, const char *controlle /* The superblock mount options of the mount point need to be * identical to the hosts', and hence writable... */ - if (mount("cgroup", to, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, controller) < 0) - return log_error_errno(errno, "Failed to mount to %s: %m", to); + if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) { + if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) { + fstype = "cgroup2"; + opts = NULL; + } else { + fstype = "cgroup"; + opts = "none,name=systemd,xattr"; + } + } else { + fstype = "cgroup"; + opts = controller; + } + + r = mount_verbose(LOG_ERR, "cgroup", to, fstype, MS_NOSUID|MS_NOEXEC|MS_NODEV, opts); + if (r < 0) + return r; - /* ... hence let's only make the bind mount read-only, not the - * superblock. */ + /* ... hence let's only make the bind mount read-only, not the superblock. */ if (read_only) { - if (mount(NULL, to, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0) - return log_error_errno(errno, "Failed to remount %s read-only: %m", to); + r = mount_verbose(LOG_ERR, NULL, to, NULL, + MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL); + if (r < 0) + return r; } + return 1; } -static int mount_legacy_cgroups( +/* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */ +static int mount_legacy_cgns_supported( const char *dest, - bool userns, uid_t uid_shift, uid_t uid_range, + CGroupUnified unified_requested, + bool userns, + uid_t uid_shift, + uid_t uid_range, + const char *selinux_apifs_context) { + + _cleanup_set_free_free_ Set *controllers = NULL; + const char *cgroup_root = "/sys/fs/cgroup", *c; + int r; + + (void) mkdir_p(cgroup_root, 0755); + + /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */ + r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW); + if (r < 0) + return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m"); + if (r == 0) { + _cleanup_free_ char *options = NULL; + + /* When cgroup namespaces are enabled and user namespaces are + * used then the mount of the cgroupfs is done *inside* the new + * user namespace. We're root in the new user namespace and the + * kernel will happily translate our uid/gid to the correct + * uid/gid as seen from e.g. /proc/1/mountinfo. So we simply + * pass uid 0 and not uid_shift to tmpfs_patch_options(). + */ + r = tmpfs_patch_options("mode=755", userns, 0, uid_range, true, selinux_apifs_context, &options); + if (r < 0) + return log_oom(); + + r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs", + MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options); + if (r < 0) + return r; + } + + if (cg_all_unified() > 0) + goto skip_controllers; + + controllers = set_new(&string_hash_ops); + if (!controllers) + return log_oom(); + + r = get_controllers(controllers); + if (r < 0) + return log_error_errno(r, "Failed to determine cgroup controllers: %m"); + + for (;;) { + _cleanup_free_ const char *controller = NULL; + + controller = set_steal_first(controllers); + if (!controller) + break; + + r = mount_legacy_cgroup_hierarchy("", controller, controller, unified_requested, !userns); + if (r < 0) + return r; + + /* When multiple hierarchies are co-mounted, make their + * constituting individual hierarchies a symlink to the + * co-mount. + */ + c = controller; + for (;;) { + _cleanup_free_ char *target = NULL, *tok = NULL; + + r = extract_first_word(&c, &tok, ",", 0); + if (r < 0) + return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m"); + if (r == 0) + break; + + target = prefix_root("/sys/fs/cgroup", tok); + if (!target) + return log_oom(); + + if (streq(controller, tok)) + break; + + r = symlink_idempotent(controller, target); + if (r == -EINVAL) + return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m"); + if (r < 0) + return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m"); + } + } + +skip_controllers: + r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false); + if (r < 0) + return r; + + if (!userns) + return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL, + MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755"); + + return 0; +} + +/* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */ +static int mount_legacy_cgns_unsupported( + const char *dest, + CGroupUnified unified_requested, + bool userns, + uid_t uid_shift, + uid_t uid_range, const char *selinux_apifs_context) { _cleanup_set_free_free_ Set *controllers = NULL; @@ -643,21 +1075,23 @@ static int mount_legacy_cgroups( (void) mkdir_p(cgroup_root, 0755); /* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */ - r = path_is_mount_point(cgroup_root, AT_SYMLINK_FOLLOW); + r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW); if (r < 0) return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m"); if (r == 0) { _cleanup_free_ char *options = NULL; - r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, selinux_apifs_context, &options); + r = tmpfs_patch_options("mode=755", userns, uid_shift, uid_range, false, selinux_apifs_context, &options); if (r < 0) return log_oom(); - if (mount("tmpfs", cgroup_root, "tmpfs", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options) < 0) - return log_error_errno(errno, "Failed to mount /sys/fs/cgroup: %m"); + r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs", + MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options); + if (r < 0) + return r; } - if (cg_unified() > 0) + if (cg_all_unified() > 0) goto skip_controllers; controllers = set_new(&string_hash_ops); @@ -683,7 +1117,7 @@ static int mount_legacy_cgroups( if (r == -EINVAL) { /* Not a symbolic link, but directly a single cgroup hierarchy */ - r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true); + r = mount_legacy_cgroup_hierarchy(dest, controller, controller, unified_requested, true); if (r < 0) return r; @@ -703,29 +1137,25 @@ static int mount_legacy_cgroups( continue; } - r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true); + r = mount_legacy_cgroup_hierarchy(dest, combined, combined, unified_requested, true); if (r < 0) return r; r = symlink_idempotent(combined, target); - if (r == -EINVAL) { - log_error("Invalid existing symlink for combined hierarchy"); - return r; - } + if (r == -EINVAL) + return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m"); if (r < 0) return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m"); } } skip_controllers: - r = mount_legacy_cgroup_hierarchy(dest, "none,name=systemd,xattr", "systemd", false); + r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER, "systemd", unified_requested, false); if (r < 0) return r; - if (mount(NULL, cgroup_root, NULL, MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755") < 0) - return log_error_errno(errno, "Failed to remount %s read-only: %m", cgroup_root); - - return 0; + return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL, + MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755"); } static int mount_unified_cgroups(const char *dest) { @@ -738,7 +1168,7 @@ static int mount_unified_cgroups(const char *dest) { (void) mkdir_p(p, 0755); - r = path_is_mount_point(p, AT_SYMLINK_FOLLOW); + r = path_is_mount_point(p, dest, AT_SYMLINK_FOLLOW); if (r < 0) return log_error_errno(r, "Failed to determine if %s is mounted already: %m", p); if (r > 0) { @@ -752,27 +1182,29 @@ static int mount_unified_cgroups(const char *dest) { return -EINVAL; } - if (mount("cgroup", p, "cgroup", MS_NOSUID|MS_NOEXEC|MS_NODEV, "__DEVEL__sane_behavior") < 0) - return log_error_errno(errno, "Failed to mount unified cgroup hierarchy to %s: %m", p); - - return 0; + return mount_verbose(LOG_ERR, "cgroup", p, "cgroup2", MS_NOSUID|MS_NOEXEC|MS_NODEV, NULL); } int mount_cgroups( const char *dest, - bool unified_requested, - bool userns, uid_t uid_shift, uid_t uid_range, - const char *selinux_apifs_context) { + CGroupUnified unified_requested, + bool userns, + uid_t uid_shift, + uid_t uid_range, + const char *selinux_apifs_context, + bool use_cgns) { - if (unified_requested) + if (unified_requested >= CGROUP_UNIFIED_ALL) return mount_unified_cgroups(dest); - else - return mount_legacy_cgroups(dest, userns, uid_shift, uid_range, selinux_apifs_context); + else if (use_cgns) + return mount_legacy_cgns_supported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context); + + return mount_legacy_cgns_unsupported(dest, unified_requested, userns, uid_shift, uid_range, selinux_apifs_context); } int mount_systemd_cgroup_writable( const char *dest, - bool unified_requested) { + CGroupUnified unified_requested) { _cleanup_free_ char *own_cgroup_path = NULL; const char *systemd_root, *systemd_own; @@ -788,7 +1220,7 @@ int mount_systemd_cgroup_writable( if (path_equal(own_cgroup_path, "/")) return 0; - if (unified_requested) { + if (unified_requested >= CGROUP_UNIFIED_ALL) { systemd_own = strjoina(dest, "/sys/fs/cgroup", own_cgroup_path); systemd_root = prefix_roota(dest, "/sys/fs/cgroup"); } else { @@ -797,14 +1229,13 @@ int mount_systemd_cgroup_writable( } /* Make our own cgroup a (writable) bind mount */ - if (mount(systemd_own, systemd_own, NULL, MS_BIND, NULL) < 0) - return log_error_errno(errno, "Failed to turn %s into a bind mount: %m", own_cgroup_path); + r = mount_verbose(LOG_ERR, systemd_own, systemd_own, NULL, MS_BIND, NULL); + if (r < 0) + return r; /* And then remount the systemd cgroup root read-only */ - if (mount(NULL, systemd_root, NULL, MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL) < 0) - return log_error_errno(errno, "Failed to mount cgroup root read-only: %m"); - - return 0; + return mount_verbose(LOG_ERR, NULL, systemd_root, NULL, + MS_BIND|MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_RDONLY, NULL); } int setup_volatile_state( @@ -825,7 +1256,7 @@ int setup_volatile_state( /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */ - r = bind_remount_recursive(directory, true); + r = bind_remount_recursive(directory, true, NULL); if (r < 0) return log_error_errno(r, "Failed to remount %s read-only: %m", directory); @@ -835,16 +1266,13 @@ int setup_volatile_state( return log_error_errno(errno, "Failed to create %s: %m", directory); options = "mode=755"; - r = tmpfs_patch_options(options, userns, uid_shift, uid_range, selinux_apifs_context, &buf); + r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf); if (r < 0) return log_oom(); if (r > 0) options = buf; - if (mount("tmpfs", p, "tmpfs", MS_STRICTATIME, options) < 0) - return log_error_errno(errno, "Failed to mount tmpfs to /var: %m"); - - return 0; + return mount_verbose(LOG_ERR, "tmpfs", p, "tmpfs", MS_STRICTATIME, options); } int setup_volatile( @@ -871,16 +1299,15 @@ int setup_volatile( return log_error_errno(errno, "Failed to create temporary directory: %m"); options = "mode=755"; - r = tmpfs_patch_options(options, userns, uid_shift, uid_range, selinux_apifs_context, &buf); + r = tmpfs_patch_options(options, userns, uid_shift, uid_range, false, selinux_apifs_context, &buf); if (r < 0) return log_oom(); if (r > 0) options = buf; - if (mount("tmpfs", template, "tmpfs", MS_STRICTATIME, options) < 0) { - r = log_error_errno(errno, "Failed to mount tmpfs for root directory: %m"); + r = mount_verbose(LOG_ERR, "tmpfs", template, "tmpfs", MS_STRICTATIME, options); + if (r < 0) goto fail; - } tmpfs_mounted = true; @@ -893,23 +1320,21 @@ int setup_volatile( goto fail; } - if (mount(f, t, NULL, MS_BIND|MS_REC, NULL) < 0) { - r = log_error_errno(errno, "Failed to create /usr bind mount: %m"); + r = mount_verbose(LOG_ERR, f, t, NULL, MS_BIND|MS_REC, NULL); + if (r < 0) goto fail; - } bind_mounted = true; - r = bind_remount_recursive(t, true); + r = bind_remount_recursive(t, true, NULL); if (r < 0) { log_error_errno(r, "Failed to remount %s read-only: %m", t); goto fail; } - if (mount(template, directory, NULL, MS_MOVE, NULL) < 0) { - r = log_error_errno(errno, "Failed to move root mount: %m"); + r = mount_verbose(LOG_ERR, template, directory, NULL, MS_MOVE, NULL); + if (r < 0) goto fail; - } (void) rmdir(template); @@ -917,28 +1342,10 @@ int setup_volatile( fail: if (bind_mounted) - (void) umount(t); + (void) umount_verbose(t); if (tmpfs_mounted) - (void) umount(template); + (void) umount_verbose(template); (void) rmdir(template); return r; } - -VolatileMode volatile_mode_from_string(const char *s) { - int b; - - if (isempty(s)) - return _VOLATILE_MODE_INVALID; - - b = parse_boolean(s); - if (b > 0) - return VOLATILE_YES; - if (b == 0) - return VOLATILE_NO; - - if (streq(s, "state")) - return VOLATILE_STATE; - - return _VOLATILE_MODE_INVALID; -}