repart: clamp copied timestamps to $SOURCE_DATE_EPOCH (#43545)

This greatly helps reproducibility, and we can actually achieve it with
https://github.com/dosfstools/dosfstools/commit/8da7bc93315c.

man/systemd-repart.xml

@@ -803,6 +803,36 @@
     </variablelist>
   </refsect1>
 
+  <refsect1>
+    <title>Environment</title>
+
+    <variablelist class='environment-variables'>
+      <varlistentry>
+        <term><varname>$SOURCE_DATE_EPOCH</varname></term>
+
+        <listitem><para>Takes a UNIX timestamp in seconds. Clamps access and modification times written
+        into the image. See the
+        <ulink url="https://reproducible-builds.org/specs/source-date-epoch/">reproducible builds
+        specification</ulink>.</para>
+
+        <xi:include href="version-info.xml" xpointer="v262"/></listitem>
+      </varlistentry>
+    </variablelist>
+
+    <para>Byte-for-byte reproducible images require <option>--offline=yes</option> mode. Populating a file
+    system through the kernel writes metadata that cannot be corrected from userspace afterwards.</para>
+
+    <para><varname>Format=</varname><literal>vfat</literal> images have additional noise: the
+    directory entry of the volume label is written with a wallclock timestamp by
+    <citerefentry project='man-pages'><refentrytitle>mkfs.fat</refentrytitle><manvolnum>8</manvolnum></citerefentry>
+    with dosfstools versions up to and including 4.2. A later release or a backport of the
+    <ulink url="https://github.com/dosfstools/dosfstools/commit/8da7bc93315c">upstream fix</ulink>
+    sets it to <varname>$SOURCE_DATE_EPOCH</varname>.</para>
+
+    <para>Offline and online mode also write <literal>vfat</literal> short names differently, so images
+    built in both ways never match each other.</para>
+  </refsect1>
+
   <refsect1>
     <title>Exit status</title>
 

src/repart/repart.c

@@ -6807,6 +6807,17 @@ static int file_is_denylisted(const char *source, Hashmap *denylist) {
         return 0;
 }
 
+/* Adding an entry to a directory bumps its mtime. Clamp it back to ts. */
+static int clamp_directory_mtime(int dir_fd, usec_t ts) {
+        assert(dir_fd >= 0);
+
+        if (ts == USEC_INFINITY)
+                return 0;
+
+        /* chase() hands out O_PATH descriptors. */
+        return futimens_opath(dir_fd, (const struct timespec[2]) { TIMESPEC_OMIT, *TIMESPEC_STORE(ts) });
+}
+
 static int do_copy_files(Context *context, Partition *p, const char *root) {
         _cleanup_hashmap_free_ Hashmap *subvolumes = NULL;
         int r;
@@ -6830,6 +6841,7 @@ static int do_copy_files(Context *context, Partition *p, const char *root) {
         }
 
         bool copy_ownership = fstype_can_ownership(p->format);
+        usec_t ts = parse_source_date_epoch();
 
         /* copy_tree_at() automatically copies the permissions of source directories to target directories if
          * it created them. However, the root directory is created by us, so we have to manually take care
@@ -6854,7 +6866,7 @@ static int do_copy_files(Context *context, Partition *p, const char *root) {
                 (void) copy_xattr(sfd, NULL, rfd, NULL, COPY_ALL_XATTRS);
                 if (copy_ownership)
                         (void) copy_access(sfd, rfd);
-                (void) copy_times(sfd, rfd, 0);
+                (void) copy_times_full(sfd, rfd, /* flags= */ 0, ts);
 
                 break;
         }
@@ -6866,7 +6878,6 @@ static int do_copy_files(Context *context, Partition *p, const char *root) {
                 _cleanup_hashmap_free_ Hashmap *denylist = NULL;
                 _cleanup_hashmap_free_ Hashmap *subvolumes_by_source_inode = NULL;
                 _cleanup_close_ int sfd = -EBADF, pfd = -EBADF, tfd = -EBADF;
-                usec_t ts = parse_source_date_epoch();
 
                 r = make_copy_files_denylist(context, p, line->source, line->target, &denylist);
                 if (r < 0)
@@ -6915,19 +6926,35 @@ static int do_copy_files(Context *context, Partition *p, const char *root) {
                                 if (pfd < 0)
                                         return log_error_errno(pfd, "Failed to open parent directory of target: %m");
 
-                                r = copy_tree_at(
+                                r = copy_tree_at_full(
                                                 sfd, ".",
                                                 pfd, fn,
                                                 uid, gid,
                                                 line->flags,
-                                                denylist, subvolumes_by_source_inode);
+                                                ts,
+                                                denylist, subvolumes_by_source_inode,
+                                                /* progress_path= */ NULL,
+                                                /* progress_bytes= */ NULL,
+                                                /* userdata= */ NULL);
+                                if (r >= 0) {
+                                        /* Creating the target bumped the parent directory's mtime,
+                                         * just like in the regular file case below. */
+                                        r = clamp_directory_mtime(pfd, ts);
+                                        if (r < 0)
+                                                return log_error_errno(
+                                                        r, "Failed to set timestamp of '%s': %m", dn);
+                                }
                         } else
-                                r = copy_tree_at(
+                                r = copy_tree_at_full(
                                                 sfd, ".",
                                                 tfd, ".",
                                                 uid, gid,
                                                 line->flags,
-                                                denylist, subvolumes_by_source_inode);
+                                                ts,
+                                                denylist, subvolumes_by_source_inode,
+                                                /* progress_path= */ NULL,
+                                                /* progress_bytes= */ NULL,
+                                                /* userdata= */ NULL);
                         if (r < 0)
                                 return log_error_errno(r, "Failed to copy '%s%s' to '%s%s': %m",
                                                        strempty(arg_copy_source), line->source, strempty(root), line->target);
@@ -6974,15 +7001,12 @@ static int do_copy_files(Context *context, Partition *p, const char *root) {
                         (void) copy_xattr(sfd, NULL, tfd, NULL, COPY_ALL_XATTRS);
                         if (copy_ownership)
                                 (void) copy_access(sfd, tfd);
-                        (void) copy_times(sfd, tfd, 0);
+                        (void) copy_times_full(sfd, tfd, /* flags= */ 0, ts);
 
-                        if (ts != USEC_INFINITY) {
-                                struct timespec tspec;
-                                timespec_store(&tspec, ts);
-
-                                if (futimens(pfd, (const struct timespec[2]) { TIMESPEC_OMIT, tspec }) < 0)
-                                        return -errno;
-                        }
+                        /* Creating the file bumped the parent directory's mtime. */
+                        r = clamp_directory_mtime(pfd, ts);
+                        if (r < 0)
+                                return log_error_errno(r, "Failed to set timestamp of '%s': %m", dn);
                 }
         }
 
@@ -7017,6 +7041,7 @@ static int do_make_directories(Partition *p, const char *root) {
 }
 
 static int do_make_symlinks(Partition *p, const char *root) {
+        usec_t ts = parse_source_date_epoch();
         int r;
 
         assert(p);
@@ -7032,6 +7057,18 @@ static int do_make_symlinks(Partition *p, const char *root) {
 
                 if (symlinkat(*target, parent_fd, f) < 0)
                         return log_error_errno(errno, "Failed to create symlink at %s to %s: %m", *path, *target);
+
+                /* The new symlink carries the wall clock, and creating it bumped the directory we put
+                 * it into. We are the last population step, so nothing overwrites this again. */
+                if (ts != USEC_INFINITY &&
+                    utimensat(parent_fd, f,
+                              (const struct timespec[2]) { *TIMESPEC_STORE(ts), *TIMESPEC_STORE(ts) },
+                              AT_SYMLINK_NOFOLLOW) < 0)
+                        return log_error_errno(errno, "Failed to set timestamp of symlink '%s': %m", *path);
+
+                r = clamp_directory_mtime(parent_fd, ts);
+                if (r < 0)
+                        return log_error_errno(r, "Failed to set timestamp of '%s': %m", *path);
         }
 
         return 0;
@@ -7218,6 +7255,21 @@ static int partition_populate_directory(Context *context, Partition *p, char **r
         if (r < 0)
                 return log_error_errno(errno, "Failed to create temporary directory: %m");
 
+        /* Nothing below touches its atime (adding an entry to a directory only moves mtime), so stamp here. */
+        usec_t ts = parse_source_date_epoch();
+        if (ts != USEC_INFINITY) {
+                _cleanup_close_ int rfd = -EBADF;
+                struct timespec tspec;
+
+                rfd = open(root, O_DIRECTORY|O_CLOEXEC|O_NOFOLLOW);
+                if (rfd < 0)
+                        return log_error_errno(errno, "Failed to open temporary directory: %m");
+
+                timespec_store(&tspec, ts);
+                if (futimens(rfd, (const struct timespec[2]) { tspec, tspec }) < 0)
+                        return log_error_errno(errno, "Failed to set timestamp of temporary directory: %m");
+        }
+
         r = do_copy_files(context, p, root);
         if (r < 0)
                 return r;

src/shared/copy.c

@@ -459,6 +459,46 @@ int copy_bytes_full(
         return max_bytes <= 0; /* return 0 if we hit EOF earlier than the size limit */
 }
 
+/* Check whether *ts is within [epoch; clamp] */
+static bool timespec_within(const struct timespec *ts, usec_t ts_clamp) {
+        assert(ts);
+
+        return ts->tv_sec < 0 ||              /* Date pre-1970, cannot clamp these */
+               timespec_load(ts) <= ts_clamp; /* Date in [epoch; clamp] */
+}
+
+/* See ts_clamp in copy.h. */
+static void clamp_inode_times(const struct stat *st, usec_t ts_clamp, struct timespec ret[static 2]) {
+        assert(st);
+        assert(ret);
+
+        ret[0] = timespec_within(&st->st_atim, ts_clamp) ? st->st_atim : *TIMESPEC_STORE(ts_clamp);
+        ret[1] = timespec_within(&st->st_mtim, ts_clamp) ? st->st_mtim : *TIMESPEC_STORE(ts_clamp);
+}
+
+/* Put *st's clamped times on the inode referred to by fdt */
+static int clamp_futimens(int fdt, const struct stat *st, usec_t ts_clamp) {
+        struct timespec times[2];
+
+        assert(fdt >= 0);
+        assert(st);
+
+        clamp_inode_times(st, ts_clamp, times);
+        return RET_NERRNO(futimens(fdt, times));
+}
+
+/* Same for (dt, to) */
+static int clamp_utimensat(int dt, const char *to, const struct stat *st, usec_t ts_clamp) {
+        struct timespec times[2];
+
+        assert(dt >= 0 || dt == AT_FDCWD);
+        assert(to);
+        assert(st);
+
+        clamp_inode_times(st, ts_clamp, times);
+        return RET_NERRNO(utimensat(dt, to, times, AT_SYMLINK_NOFOLLOW));
+}
+
 static int fd_copy_symlink(
                 int df,
                 const char *from,
@@ -467,7 +507,8 @@ static int fd_copy_symlink(
                 const char *to,
                 uid_t override_uid,
                 gid_t override_gid,
-                CopyFlags copy_flags) {
+                CopyFlags copy_flags,
+                usec_t ts_clamp) {
 
         _cleanup_free_ char *target = NULL;
         int r;
@@ -506,7 +547,8 @@ static int fd_copy_symlink(
                 r = -errno;
 
         (void) copy_xattr(df, from, dt, to, copy_flags);
-        (void) utimensat(dt, to, (struct timespec[]) { st->st_atim, st->st_mtim }, AT_SYMLINK_NOFOLLOW);
+
+        (void) clamp_utimensat(dt, to, st, ts_clamp);
         return r;
 }
 
@@ -798,6 +840,7 @@ static int fd_copy_tree_generic(
                 uid_t override_uid,
                 gid_t override_gid,
                 CopyFlags copy_flags,
+                usec_t ts_clamp,
                 Hashmap *denylist,
                 Hashmap *subvolumes,
                 HardlinkContext *hardlink_context,
@@ -815,6 +858,7 @@ static int fd_copy_regular(
                 uid_t override_uid,
                 gid_t override_gid,
                 CopyFlags copy_flags,
+                usec_t ts_clamp,
                 HardlinkContext *hardlink_context,
                 copy_progress_bytes_t progress,
                 void *userdata) {
@@ -862,7 +906,7 @@ static int fd_copy_regular(
         if (fchmod(fdt, st->st_mode & 07777) < 0)
                 r = -errno;
 
-        (void) futimens(fdt, (struct timespec[]) { st->st_atim, st->st_mtim });
+        (void) clamp_futimens(fdt, st, ts_clamp);
         (void) copy_xattr(fdf, NULL, fdt, NULL, copy_flags);
 
         if (FLAGS_SET(copy_flags, COPY_VERIFY_LINKED)) {
@@ -910,6 +954,7 @@ static int fd_copy_fifo(
                 uid_t override_uid,
                 gid_t override_gid,
                 CopyFlags copy_flags,
+                usec_t ts_clamp,
                 HardlinkContext *hardlink_context) {
         int r;
 
@@ -948,7 +993,7 @@ static int fd_copy_fifo(
         if (fchmodat(dt, to, st->st_mode & 07777, AT_SYMLINK_NOFOLLOW) < 0)
                 r = -errno;
 
-        (void) utimensat(dt, to, (struct timespec[]) { st->st_atim, st->st_mtim }, AT_SYMLINK_NOFOLLOW);
+        (void) clamp_utimensat(dt, to, st, ts_clamp);
 
         (void) memorize_hardlink(hardlink_context, st, dt, to);
         return r;
@@ -963,6 +1008,7 @@ static int fd_copy_node(
                 uid_t override_uid,
                 gid_t override_gid,
                 CopyFlags copy_flags,
+                usec_t ts_clamp,
                 HardlinkContext *hardlink_context) {
         int r;
 
@@ -1001,7 +1047,7 @@ static int fd_copy_node(
         if (fchmodat(dt, to, st->st_mode & 07777, AT_SYMLINK_NOFOLLOW) < 0)
                 r = -errno;
 
-        (void) utimensat(dt, to, (struct timespec[]) { st->st_atim, st->st_mtim }, AT_SYMLINK_NOFOLLOW);
+        (void) clamp_utimensat(dt, to, st, ts_clamp);
 
         (void) memorize_hardlink(hardlink_context, st, dt, to);
         return r;
@@ -1018,6 +1064,7 @@ static int fd_copy_directory(
                 uid_t override_uid,
                 gid_t override_gid,
                 CopyFlags copy_flags,
+                usec_t ts_clamp,
                 Hashmap *denylist,
                 Hashmap *subvolumes,
                 HardlinkContext *hardlink_context,
@@ -1169,7 +1216,7 @@ static int fd_copy_directory(
 
                 r = fd_copy_tree_generic(dirfd(d), de->d_name, &buf, fdt, de->d_name, original_device,
                                          depth_left-1, override_uid, override_gid, copy_flags & ~COPY_LOCK_BSD,
-                                         denylist, subvolumes, hardlink_context, child_display_path, progress_path,
+                                         ts_clamp, denylist, subvolumes, hardlink_context, child_display_path, progress_path,
                                          progress_bytes, userdata);
 
                 /* Propagate SIGINT/SIGTERM, ENOSPC, and fs-verity fails up instantly */
@@ -1193,7 +1240,8 @@ finish:
                 /* Run hardlink context cleanup now because it potentially changes timestamps */
                 hardlink_context_destroy(&our_hardlink_context);
                 (void) copy_xattr(dirfd(d), NULL, fdt, NULL, copy_flags);
-                (void) futimens(fdt, (struct timespec[]) { st->st_atim, st->st_mtim });
+
+                (void) clamp_futimens(fdt, st, ts_clamp);
         } else if (FLAGS_SET(copy_flags, COPY_RESTORE_DIRECTORY_TIMESTAMPS)) {
                 /* Run hardlink context cleanup now because it potentially changes timestamps */
                 hardlink_context_destroy(&our_hardlink_context);
@@ -1221,6 +1269,7 @@ static int fd_copy_leaf(
                 uid_t override_uid,
                 gid_t override_gid,
                 CopyFlags copy_flags,
+                usec_t ts_clamp,
                 HardlinkContext *hardlink_context,
                 const char *display_path,
                 copy_progress_bytes_t progress_bytes,
@@ -1228,13 +1277,13 @@ static int fd_copy_leaf(
         int r;
 
         if (S_ISREG(st->st_mode))
-                r = fd_copy_regular(df, from, st, dt, to, override_uid, override_gid, copy_flags, hardlink_context, progress_bytes, userdata);
+                r = fd_copy_regular(df, from, st, dt, to, override_uid, override_gid, copy_flags, ts_clamp, hardlink_context, progress_bytes, userdata);
         else if (S_ISLNK(st->st_mode))
-                r = fd_copy_symlink(df, from, st, dt, to, override_uid, override_gid, copy_flags);
+                r = fd_copy_symlink(df, from, st, dt, to, override_uid, override_gid, copy_flags, ts_clamp);
         else if (S_ISFIFO(st->st_mode))
-                r = fd_copy_fifo(df, from, st, dt, to, override_uid, override_gid, copy_flags, hardlink_context);
+                r = fd_copy_fifo(df, from, st, dt, to, override_uid, override_gid, copy_flags, ts_clamp, hardlink_context);
         else if (S_ISBLK(st->st_mode) || S_ISCHR(st->st_mode) || S_ISSOCK(st->st_mode))
-                r = fd_copy_node(df, from, st, dt, to, override_uid, override_gid, copy_flags, hardlink_context);
+                r = fd_copy_node(df, from, st, dt, to, override_uid, override_gid, copy_flags, ts_clamp, hardlink_context);
         else
                 r = -EOPNOTSUPP;
 
@@ -1252,6 +1301,7 @@ static int fd_copy_tree_generic(
                 uid_t override_uid,
                 gid_t override_gid,
                 CopyFlags copy_flags,
+                usec_t ts_clamp,
                 Hashmap *denylist,
                 Hashmap *subvolumes,
                 HardlinkContext *hardlink_context,
@@ -1266,7 +1316,7 @@ static int fd_copy_tree_generic(
 
         if (S_ISDIR(st->st_mode))
                 return fd_copy_directory(df, from, st, dt, to, original_device, depth_left-1, override_uid,
-                                         override_gid, copy_flags, denylist, subvolumes, hardlink_context,
+                                         override_gid, copy_flags, ts_clamp, denylist, subvolumes, hardlink_context,
                                          display_path, progress_path, progress_bytes, userdata);
 
         /* Only if we are copying a directory we are fine if the target dir is referenced by fd only */
@@ -1280,14 +1330,14 @@ static int fd_copy_tree_generic(
         } else if (t == DENY_CONTENTS)
                 log_debug("%s is configured to have its contents excluded, but is not a directory", from ?: "file to copy");
 
-        r = fd_copy_leaf(df, from, st, dt, to, override_uid, override_gid, copy_flags, hardlink_context, display_path, progress_bytes, userdata);
+        r = fd_copy_leaf(df, from, st, dt, to, override_uid, override_gid, copy_flags, ts_clamp, hardlink_context, display_path, progress_bytes, userdata);
         /* We just tried to copy a leaf node of the tree. If it failed because the node already exists *and* the COPY_REPLACE flag has been provided, we should unlink the node and re-copy. */
         if (r == -EEXIST && (copy_flags & COPY_REPLACE)) {
                 /* This codepath is us trying to address an error to copy, if the unlink fails, lets just return the original error. */
                 if (unlinkat(dt, to, 0) < 0)
                         return r;
 
-                r = fd_copy_leaf(df, from, st, dt, to, override_uid, override_gid, copy_flags, hardlink_context, display_path, progress_bytes, userdata);
+                r = fd_copy_leaf(df, from, st, dt, to, override_uid, override_gid, copy_flags, ts_clamp, hardlink_context, display_path, progress_bytes, userdata);
         }
 
         return r;
@@ -1301,6 +1351,7 @@ int copy_tree_at_full(
                 uid_t override_uid,
                 gid_t override_gid,
                 CopyFlags copy_flags,
+                usec_t ts_clamp,
                 Hashmap *denylist,
                 Hashmap *subvolumes,
                 copy_progress_path_t progress_path,
@@ -1316,7 +1367,7 @@ int copy_tree_at_full(
                 return -errno;
 
         r = fd_copy_tree_generic(fdf, from, &st, fdt, to, st.st_dev, COPY_DEPTH_MAX, override_uid,
-                                 override_gid, copy_flags, denylist, subvolumes, NULL, NULL, progress_path,
+                                 override_gid, copy_flags, ts_clamp, denylist, subvolumes, NULL, NULL, progress_path,
                                  progress_bytes, userdata);
         if (r < 0)
                 return r;
@@ -1388,10 +1439,11 @@ int copy_directory_at_full(
                         override_uid,
                         override_gid,
                         copy_flags,
+                        /* ts_clamp= */ USEC_INFINITY,
                         /* denylist= */ NULL,
                         /* subvolumes= */ NULL,
-                        /* progress_path= */ NULL,
-                        /* progress_bytes= */ NULL,
+                        /* hardlink_context= */ NULL,
+                        /* display_path= */ NULL,
                         progress_path,
                         progress_bytes,
                         userdata);
@@ -1639,8 +1691,9 @@ fail:
         return r;
 }
 
-int copy_times(int fdf, int fdt, CopyFlags flags) {
+int copy_times_full(int fdf, int fdt, CopyFlags flags, usec_t ts_clamp) {
         struct stat st;
+        int r;
 
         assert(fdf >= 0);
         assert(fdt >= 0);
@@ -1648,14 +1701,15 @@ int copy_times(int fdf, int fdt, CopyFlags flags) {
         if (fstat(fdf, &st) < 0)
                 return -errno;
 
-        if (futimens(fdt, (struct timespec[2]) { st.st_atim, st.st_mtim }) < 0)
-                return -errno;
+        r = clamp_futimens(fdt, &st, ts_clamp);
+        if (r < 0)
+                return r;
 
         if (FLAGS_SET(flags, COPY_CRTIME)) {
                 usec_t crtime;
 
                 if (fd_getcrtime(fdf, &crtime) >= 0)
-                        (void) fd_setcrtime(fdt, crtime);
+                        (void) fd_setcrtime(fdt, MIN(crtime, ts_clamp));
         }
 
         return 0;

src/shared/copy.h

@@ -81,12 +81,15 @@ static inline int copy_file_atomic(const char *from, const char *to, mode_t mode
         return copy_file_atomic_full(from, to, mode, 0, 0, copy_flags, NULL, NULL);
 }
 
-int copy_tree_at_full(int fdf, const char *from, int fdt, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags, Hashmap *denylist, Hashmap *subvolumes, copy_progress_path_t progress_path, copy_progress_bytes_t progress_bytes, void *userdata);
+/* ts_clamp: upper bound for the timestamps of created inodes. $SOURCE_DATE_EPOCH semantic, see
+ * https://reproducible-builds.org/specs/source-date-epoch/. USEC_INFINITY clamps nothing, i.e. keeps
+ * the source's own timestamps. */
+int copy_tree_at_full(int fdf, const char *from, int fdt, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags, usec_t ts_clamp, Hashmap *denylist, Hashmap *subvolumes, copy_progress_path_t progress_path, copy_progress_bytes_t progress_bytes, void *userdata);
 static inline int copy_tree_at(int fdf, const char *from, int fdt, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags, Hashmap *denylist, Hashmap *subvolumes) {
-        return copy_tree_at_full(fdf, from, fdt, to, override_uid, override_gid, copy_flags, denylist, subvolumes, NULL, NULL, NULL);
+        return copy_tree_at_full(fdf, from, fdt, to, override_uid, override_gid, copy_flags, USEC_INFINITY, denylist, subvolumes, NULL, NULL, NULL);
 }
 static inline int copy_tree(const char *from, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags, Hashmap *denylist, Hashmap *subvolumes) {
-        return copy_tree_at_full(AT_FDCWD, from, AT_FDCWD, to, override_uid, override_gid, copy_flags, denylist, subvolumes, NULL, NULL, NULL);
+        return copy_tree_at_full(AT_FDCWD, from, AT_FDCWD, to, override_uid, override_gid, copy_flags, USEC_INFINITY, denylist, subvolumes, NULL, NULL, NULL);
 }
 
 int copy_directory_at_full(int dir_fdf, const char *from, int dir_fdt, const char *to, uid_t override_uid, gid_t override_gid, CopyFlags copy_flags, copy_progress_path_t progress_path, copy_progress_bytes_t progress_bytes, void *userdata);
@@ -99,7 +102,11 @@ static inline int copy_bytes(int fdf, int fdt, uint64_t max_bytes, CopyFlags cop
         return copy_bytes_full(fdf, fdt, max_bytes, copy_flags, NULL, NULL, NULL, NULL);
 }
 
-int copy_times(int fdf, int fdt, CopyFlags flags);
+/* See ts_clamp above. Covers atime, mtime, and with COPY_CRTIME the birth time. */
+int copy_times_full(int fdf, int fdt, CopyFlags flags, usec_t ts_clamp);
+static inline int copy_times(int fdf, int fdt, CopyFlags flags) {
+        return copy_times_full(fdf, fdt, flags, USEC_INFINITY);
+}
 int copy_access(int fdf, int fdt);
 int copy_owner(int fdf, int fdt);
 int copy_rights_with_fallback(int fdf, int fdt, const char *patht);

src/shared/mkfs-util.c

@@ -5,6 +5,7 @@
 #include <unistd.h>
 
 #include "escape.h"
+#include "install-file.h"
 #include "log.h"
 #include "memory-util.h"
 #include "mkfs-util.h"
@@ -17,6 +18,7 @@
 #include "stdio-util.h"
 #include "string-util.h"
 #include "strv.h"
+#include "time-util.h"
 #include "utf8.h"
 
 int mkfs_exists(const char *fstype) {
@@ -150,6 +152,26 @@ static int mangle_fat_label(const char *s, char **ret) {
         return 0;
 }
 
+/* $SOURCE_DATE_EPOCH in seconds as *we* read it, or NULL if we have none. The tools we call parse the
+ * variable themselves, so without this they would act on values we reject (unparsable, overflowing),
+ * read differently (octal 017), or that secure_getenv() declines to look at under AT_SECURE. */
+static int source_date_epoch_seconds(char **ret) {
+        usec_t epoch;
+
+        assert(ret);
+
+        epoch = parse_source_date_epoch();
+        if (epoch == USEC_INFINITY) {
+                *ret = NULL;
+                return 0;
+        }
+
+        if (asprintf(ret, USEC_FMT, epoch / USEC_PER_SEC) < 0)
+                return -ENOMEM;
+
+        return 1;
+}
+
 static int mtools_exec(char *const *argv, ForkFlags extra_fork_flags) {
         int r;
 
@@ -161,6 +183,16 @@ static int mtools_exec(char *const *argv, ForkFlags extra_fork_flags) {
                 log_debug("Invoking mtools command: %s", strna(j));
         }
 
+        _cleanup_free_ char *seconds = NULL, *source_date_epoch = NULL;
+        r = source_date_epoch_seconds(&seconds);
+        if (r < 0)
+                return log_oom();
+        if (seconds) {
+                source_date_epoch = strjoin("SOURCE_DATE_EPOCH=", seconds);
+                if (!source_date_epoch)
+                        return log_oom();
+        }
+
         r = pidref_safe_fork(
                         "(mtools)",
                         FORK_RESET_SIGNALS|FORK_RLIMIT_NOFILE_SAFE|FORK_DEATHSIG_SIGTERM|FORK_WAIT|FORK_STDOUT_TO_STDERR|FORK_CLOSE_ALL_FDS|extra_fork_flags,
@@ -174,7 +206,7 @@ static int mtools_exec(char *const *argv, ForkFlags extra_fork_flags) {
                 execve(argv[0], argv,
                        STRV_MAKE("MTOOLS_SKIP_CHECK=1",
                                  "TZ=UTC",
-                                 strv_find_prefix(environ, "SOURCE_DATE_EPOCH=")));
+                                 source_date_epoch));
 
                 log_full_errno(FLAGS_SET(extra_fork_flags, FORK_LOG) ? LOG_ERR : LOG_DEBUG, errno, "Failed to execute %s: %m", argv[0]);
                 _exit(EXIT_FAILURE);
@@ -430,8 +462,11 @@ int make_filesystem(
                  * 0 value handling, where $E2FSPROGS_FAKE_TIME=0 is ignored and the current time is used,
                  * but $SOURCE_DATE_EPOCH=0 sets 1970-01-01 as the timestamp. */
                 if (!secure_getenv("E2FSPROGS_FAKE_TIME")) { /* honor $E2FSPROGS_FAKE_TIME if already set */
-                        const char *e = secure_getenv("SOURCE_DATE_EPOCH");
-                        if (e && strv_extend_strv(&env, STRV_MAKE("E2FSPROGS_FAKE_TIME", e), /* filter_duplicates= */ false) < 0)
+                        _cleanup_free_ char *seconds = NULL;
+
+                        if (source_date_epoch_seconds(&seconds) < 0)
+                                return log_oom();
+                        if (seconds && strv_extend_strv(&env, STRV_MAKE("E2FSPROGS_FAKE_TIME", seconds), /* filter_duplicates= */ false) < 0)
                                 return log_oom();
                 }
 
@@ -650,6 +685,10 @@ int make_filesystem(
                         fork_flags |= FORK_NEW_MOUNTNS;
         }
 
+        _cleanup_free_ char *source_date_epoch = NULL;
+        if (source_date_epoch_seconds(&source_date_epoch) < 0)
+                return log_oom();
+
         log_info("Formatting %s as %s", node, fstype);
 
         if (DEBUG_LOGGING) {
@@ -671,14 +710,21 @@ int make_filesystem(
 
                 STRV_FOREACH_PAIR(k, v, env)
                         if (setenv(*k, *v, /* replace= */ true) < 0) {
-                                log_error_errno(r, "Failed to set %s=%s environment variable: %m", *k, *v);
+                                log_error_errno(errno, "Failed to set %s=%s environment variable: %m", *k, *v);
                                 _exit(EXIT_FAILURE);
                         }
 
+                /* Same as in mtools_exec(): the mkfs tool gets our reading of the variable, or none. */
+                if (source_date_epoch ? setenv("SOURCE_DATE_EPOCH", source_date_epoch, /* replace= */ true) < 0
+                                      : unsetenv("SOURCE_DATE_EPOCH") < 0) {
+                        log_error_errno(errno, "Failed to adjust $SOURCE_DATE_EPOCH: %m");
+                        _exit(EXIT_FAILURE);
+                }
+
                 /* mkfs.btrfs refuses to operate on block devices with mounted partitions, even if operating
                  * on unformatted free space, so let's trick it and other mkfs tools into thinking no
                  * partitions are mounted. See https://github.com/kdave/btrfs-progs/issues/640 for more
-                 ° information. */
+                 * information. */
                  if (fork_flags & FORK_NEW_MOUNTNS)
                         (void) mount_nofollow_verbose(LOG_DEBUG, "/dev/null", "/proc/self/mounts", NULL, MS_BIND, NULL);
 

src/test/test-copy.c

@@ -26,6 +26,7 @@
 #include "string-util.h"
 #include "strv.h"
 #include "tests.h"
+#include "time-util.h"
 #include "tmpfile-util.h"
 #include "xattr-util.h"
 
@@ -798,4 +799,62 @@ TEST_RET(copy_with_verity) {
         return 0;
 }
 
+static void test_copy_times_one(
+                const struct timespec source[static 2],
+                usec_t ts_clamp,
+                const struct timespec expect[static 2]) {
+        _cleanup_(unlink_tempfilep) char in_fn[] = "/tmp/test-copy-times-in-XXXXXX";
+        _cleanup_(unlink_tempfilep) char out_fn[] = "/tmp/test-copy-times-out-XXXXXX";
+        _cleanup_close_ int in_fd = -EBADF, out_fd = -EBADF;
+        struct stat st;
+
+        in_fd = mkostemp_safe(in_fn);
+        assert_se(in_fd >= 0);
+        out_fd = mkostemp_safe(out_fn);
+        assert_se(out_fd >= 0);
+
+        assert_se(futimens(in_fd, source) >= 0);
+
+        ASSERT_OK(copy_times_full(in_fd, out_fd, /* flags= */ 0, ts_clamp));
+
+        assert_se(fstat(out_fd, &st) >= 0);
+        ASSERT_EQ(st.st_atim.tv_sec, expect[0].tv_sec);
+        ASSERT_EQ(st.st_atim.tv_nsec, expect[0].tv_nsec);
+        ASSERT_EQ(st.st_mtim.tv_sec, expect[1].tv_sec);
+        ASSERT_EQ(st.st_mtim.tv_nsec, expect[1].tv_nsec);
+}
+
+#define TIMESPEC_PAIR(a, m) ((const struct timespec[2]) { { .tv_sec = (a) }, { .tv_sec = (m) } })
+
+TEST(copy_times_clamp) {
+        const time_t before = 1600000000, epoch = 1700000000, after = 1750000000;
+        const usec_t clamp = epoch * USEC_PER_SEC;
+
+        /* Older than the clamp: kept. */
+        test_copy_times_one(TIMESPEC_PAIR(before, before), clamp, TIMESPEC_PAIR(before, before));
+
+        /* Newer: pulled back onto the clamp. */
+        test_copy_times_one(TIMESPEC_PAIR(after, after), clamp, TIMESPEC_PAIR(epoch, epoch));
+
+        /* Exactly on it: kept, the clamp is an inclusive upper bound. The comparison runs at microsecond
+         * resolution, so a nanosecond past the clamp still counts as "on it". */
+        test_copy_times_one(
+                        (const struct timespec[2]) { { .tv_sec = epoch, .tv_nsec = 1 },
+                                                     { .tv_sec = epoch, .tv_nsec = 1 } },
+                        clamp,
+                        (const struct timespec[2]) { { .tv_sec = epoch, .tv_nsec = 1 },
+                                                     { .tv_sec = epoch, .tv_nsec = 1 } });
+
+        /* atime and mtime are decided independently: a file that was merely read after the epoch
+         * loses its atime and keeps its mtime. */
+        test_copy_times_one(TIMESPEC_PAIR(after, before), clamp, TIMESPEC_PAIR(epoch, before));
+        test_copy_times_one(TIMESPEC_PAIR(before, after), clamp, TIMESPEC_PAIR(before, epoch));
+
+        /* Before 1970: older than any epoch, so kept rather than pushed forward. */
+        test_copy_times_one(TIMESPEC_PAIR(-1, -1), clamp, TIMESPEC_PAIR(-1, -1));
+
+        /* No clamp requested: the source's own timestamps. */
+        test_copy_times_one(TIMESPEC_PAIR(after, after), USEC_INFINITY, TIMESPEC_PAIR(after, after));
+}
+
 DEFINE_TEST_MAIN(LOG_DEBUG);

test/units/TEST-58-REPART.sh

@@ -2166,7 +2166,7 @@ EOF
 }
 
 testcase_make_symlinks() {
-    local defs imgs output
+    local defs imgs output epoch
 
     if systemd-detect-virt --quiet --container; then
         echo "Skipping MakeSymlinks= test in container."
@@ -2191,7 +2191,13 @@ MakeSymlinks=/dir/foo-%a:/bar-%a
 MakeSymlinks=/dir/bar-%a:../bar-%a
 EOF
 
-    systemd-repart --offline="$OFFLINE" \
+    # Build with an epoch, so that this also covers do_make_symlinks()' time stamping. It has to
+    # stamp the symlink itself and repair the directory it lands in, because creating an entry there
+    # bumps that directory back to the wall clock.
+    epoch=1700000000
+
+    env SOURCE_DATE_EPOCH="$epoch" \
+        systemd-repart --offline="$OFFLINE" \
                    --definitions="$defs" \
                    --empty=create \
                    --size=1G \
@@ -2205,6 +2211,12 @@ EOF
     assert_eq "$(readlink "$imgs/mnt/dir/foo")" "/bar"
     assert_eq "$(readlink "$imgs/mnt/dir/foo-${architecture}")" "/bar-${architecture}"
     assert_eq "$(readlink "$imgs/mnt/dir/bar-${architecture}")" "../bar-${architecture}"
+
+    # the symlink's own mtime (no -L)
+    assert_eq "$(stat -c %Y "$imgs/mnt/dir/foo")" "$epoch"
+    # a MakeDirectories= dir, stamped with the epoch and then bumped by the symlinks below it.
+    assert_eq "$(stat -c %Y "$imgs/mnt/dir")" "$epoch"
+
     systemd-dissect -U "$imgs/mnt"
 }
 
@@ -2756,6 +2768,123 @@ EOF
     cmp "$imgs/test1.img" "$imgs/test2.img"
 }
 
+testcase_vfat_reproducibility() {
+    local defs imgs img dot_efi dot_linux
+
+    if ! command -v mdir >/dev/null; then
+        echo "Skipping vfat reproducibility test, mtools is not installed."
+        return 0
+    fi
+
+    defs="$(mktemp --directory "/tmp/test-repart.defs.XXXXXXXXXX")"
+    imgs="$(mktemp --directory "/var/tmp/test-repart.imgs.XXXXXXXXXX")"
+    # shellcheck disable=SC2064
+    trap "rm -rf '$defs' '$imgs'" RETURN
+
+    tee "$defs/esp.conf" <<EOF
+[Partition]
+Type=esp
+Format=vfat
+CopyFiles=/:/
+EOF
+
+    # $SOURCE_DATE_EPOCH is a clamp, not an override, so put one file on either side of it to cover
+    # both directions.
+    mkdir -p "$imgs/tree/EFI/Linux"
+    echo old >"$imgs/tree/EFI/Linux/old.efi"
+    echo new >"$imgs/tree/EFI/Linux/new.efi"
+    touch --date=@1600000000 "$imgs/tree/EFI/Linux/old.efi"
+    touch --date=@1750000000 "$imgs/tree/EFI/Linux/new.efi"
+
+    # The timestamps as mdir(1) reports them.
+    local -r time_epoch="2023-11-14  22:13"   # $SOURCE_DATE_EPOCH below, i.e. @1700000000
+    local -r time_old="2020-09-13  12:26"     # old.efi's mtime, before the epoch
+    local -r time_new="2025-06-15  15:06"     # new.efi's mtime, after the epoch
+
+    # Build $1 from the tree and set $img to it, in the "file@@offset" form mdir(1) wants. The
+    # remaining arguments are passed to env(1), to control $SOURCE_DATE_EPOCH.
+    build_image() {
+        local name="$1" output offset
+        shift
+
+        output=$(env "$@" \
+            systemd-repart \
+            --offline="$OFFLINE" \
+            --definitions="$defs" \
+            --empty=create \
+            --size=auto \
+            --seed="$seed" \
+            --dry-run=no \
+            --root="$imgs/tree" \
+            --json=pretty \
+            "$imgs/$name")
+
+        offset=$(jq -r '.[0].offset' <<<"$output")
+        img="$imgs/$name@@$offset"
+
+        # dump listing for debugging
+        fat_entry -/ ::/
+    }
+
+    # Print $img's mdir(1) entry for the given path. mdir lists a directory's contents; a trailing `*`
+    # gets the directory's own entry instead. Insensitive path matching, thus works with mtools' 8.3
+    # lower case and kernel vfat's upper case names.
+    fat_entry() {
+        env MTOOLS_SKIP_CHECK=1 TZ=UTC mdir -i "$img" "$@"
+    }
+
+    # Same, for the "." entry of directory $1, which cannot be addressed directly.
+    fat_dot_entry() {
+        fat_entry "$1" | grep -E '^\. +<DIR>'
+    }
+
+    build_image epoch.img SOURCE_DATE_EPOCH=1700000000
+
+    # The directories and the file newer than the epoch are clamped down to it, ...
+    assert_in "$time_epoch" "$(fat_entry '::/EFI*')"
+    assert_in "$time_epoch" "$(fat_entry '::/EFI/Linux*')"
+    assert_in "$time_epoch" "$(fat_entry '::/EFI/Linux/new.efi')"
+    # ... while the older one keeps its own mtime.
+    assert_in "$time_old" "$(fat_entry '::/EFI/Linux/old.efi')"
+
+    # "." and ".." behave differently
+    dot_efi=$(fat_dot_entry '::/EFI')
+    dot_linux=$(fat_dot_entry '::/EFI/Linux')
+    if [[ "$OFFLINE" == "yes" ]]; then
+        # mmd creates them and respects SOURCE_DATE_EPOCH
+        assert_in "$time_epoch" "$dot_efi"
+        assert_in "$time_epoch" "$dot_linux"
+    else
+        # The kernel's vfat driver creates them with wallclock mtime. Our utimensat()
+        # afterwards only rewrites the directory's entry in its *parent*, never this pair.
+        # Out of reach from userspace, remains unreproducible. Just log them.
+        echo "'.' entries carry the wall clock, as expected online: $dot_efi / $dot_linux"
+    fi
+
+    # Without an epoch, both files keep their own mtime.
+    build_image wallclock.img -u SOURCE_DATE_EPOCH
+
+    assert_in "$time_new" "$(fat_entry '::/EFI/Linux/new.efi')"
+    assert_in "$time_old" "$(fat_entry '::/EFI/Linux/old.efi')"
+
+    # A value we cannot parse is dropped rather than passed on
+    build_image bogus.img SOURCE_DATE_EPOCH=99999999999999999999
+
+    assert_in "$time_new" "$(fat_entry '::/EFI/Linux/new.efi')"
+    assert_in "$time_old" "$(fat_entry '::/EFI/Linux/old.efi')"
+
+    # SOURCE_DATE_EPOCH in hex - we do parse that, but mtools doesn't
+    build_image hex.img SOURCE_DATE_EPOCH=0x6553F100
+
+    assert_in "$time_epoch" "$(fat_entry '::/EFI*')"
+
+    # Note the images still are not reproducible byte for byte: the volume label entry carries
+    # mkfs.fat's wall clock, in *both* modes, and that needs a dosfstools with $SOURCE_DATE_EPOCH
+    # support (https://github.com/dosfstools/dosfstools/commit/8da7bc93315c, release > 4.2). Once
+    # that lands, offline mode can move to a `cmp` like in testcase_ext_reproducibility; online
+    # mode cannot, because of the "." and ".." entries above.
+}
+
 testcase_luks2_keyhash() {
     local defs imgs output root