Merge branch 'copyfile-symlinks'

Closes #23

Signed-off-by: Joachim Wiberg <troglobit@gmail.com>

ChangeLog.md

@@ -7,7 +7,16 @@ All notable changes to the project are documented in this file.
 [UNRELEASED][]
 --------------
 
+### Changes
+- Add `LITE_FOPT_RSYNC_LINKS` option to `rsync()`, recreates symlinks to
+  files in the destination instead of copying what they point to
+
 ### Fixes
+- Fix #23: `copyfile()` ignored `LITE_FOPT_COPYFILE_SYM` and always
+  copied what a symlink points to.  It now recreates the symlink, also
+  when it is dangling or points to a directory
+- Fix #22: `fcopyfile()` truncated binary data at the first NUL byte in
+  each chunk, and returned success on read and write errors
 - Fix `copyfile()` not preserving the source file mode when the process
   umask is more restrictive
 - Fix `mkpath()`, and `fmkpath()`, not applying the given mode when the

doc/API.md

@@ -244,6 +244,16 @@ Generic Functions
   
   - `LITE_FOPT_RSYNC_DELETE (0x01)`
   - `LITE_FOPT_KEEP_MTIME   (0x02)`
+  - `LITE_FOPT_RSYNC_LINKS  (0x04)`
+  
+  Symlinks in the source tree are followed by default, the destination
+  gets a copy of what they point to.  With `LITE_FOPT_RSYNC_LINKS` a
+  symlink to a file, or a dangling symlink, is recreated in the
+  destination instead.  Symlinks to directories are always followed.
+
+  With `LITE_FOPT_RSYNC_LINKS` a symlink in the destination is never
+  followed.  If one sits where a file or directory from the source
+  goes, it is removed first.
   
   In releases prior to v2.0 the argument controlling pruning was called
   `delete`, it is now called `opt`.  The APIs are 100% compatible if the

src/copyfile.c

@@ -157,7 +157,13 @@ ssize_t copyfile(const char *src, const char *dst, int len, int opt)
 	if (!buffer)
 		return 0;
 
-	if (fisdir(src)) {
+	if (sym ? lstat(src, &st) : stat(src, &st)) {
+		saved_errno = errno;
+		size = -1;
+		goto exit;
+	}
+
+	if (S_ISDIR(st.st_mode)) {
 		saved_errno = EISDIR;	/* Error: source is a directory */
 		goto exit;
 	}
@@ -165,13 +171,7 @@ ssize_t copyfile(const char *src, const char *dst, int len, int opt)
 	/* Check if target is a directory, then append src filename. */
 	isdir = adjust_target(src, &dest);
 
-	/* Check if the source file is a symlink ... */
-	if (stat(src, &st)) {
-		size = -1;
-		goto exit;
-	}
-
-	/* ... only try readlink() if symlink and @sym is set. */
+	/* Only try readlink() if symlink and @sym is set. */
 	if (sym && (st.st_mode & S_IFMT) == S_IFLNK) {
 		size = readlink(src, buffer, BUFSIZ);
 		if (size > 0) {
@@ -181,8 +181,16 @@ ssize_t copyfile(const char *src, const char *dst, int len, int opt)
 			} else {
 				buffer[size] = 0;
 				size = !symlink(buffer, dest);
+				if (!size)
+					saved_errno = errno;
+				else if (keep_mtim) {
+					struct timespec tv[2] = { st.st_atim, st.st_mtim };
+
+					utimensat(AT_FDCWD, dest, tv, AT_SYMLINK_NOFOLLOW);
+				}
 			}
-		}
+		} else
+			saved_errno = errno;
 
 		/* Don't fall back to copy, that is a potentially
 		 * dangerous race condition, see CWE-367. */

src/lite.h

@@ -67,6 +67,7 @@ extern "C"
 #define LITE_FOPT_RSYNC_DELETE  0x01	/**< Prune files from destination that are not in source */
 #define LITE_FOPT_COPYFILE_SYM  0x01	/**< Recreate symlink or follow to copy target */
 #define LITE_FOPT_KEEP_MTIME    0x02	/**< Preserve modification time */
+#define LITE_FOPT_RSYNC_LINKS   0x04	/**< Recreate symlinks to files instead of copying their target */
 
 typedef struct lfile lfile_t;		/**< Opqaue context struct for lfile APIs */
 typedef struct sdbuf sdbuf_t;		/**< Opqaue context struct for telnet APIs */

src/rsync.c

@@ -24,6 +24,7 @@
 
 #include <errno.h>
 #include <fcntl.h>	/* AT_* macros */
+#include <libgen.h>	/* dirname() */
 #include <stdlib.h>	/* NULL, free() */
 #include <string.h>	/* strlen() */
 #include <strings.h>	/* rindex() */
@@ -35,7 +36,7 @@
 #include "lite.h"
 
 static int copy(char *src, char *dst, int opt);
-static int mdir(char *buf, size_t buf_len, char *dir, char *name, struct stat *st);
+static int mdir(char *buf, size_t buf_len, char *dir, char *name, struct stat *st, int opt);
 static int prune(char *dst, char **new_files, int new_num);
 static int set_mtime(char *fn, struct stat *st);
 
@@ -44,7 +45,8 @@ static int set_mtime(char *fn, struct stat *st);
  * Synchronize contents and optionally remove non-existing backups
  * @param src     Source directory
  * @param dst     Destination directory
- * @param opt     An option mask of ::LITE_FOPT_RSYNC_DELETE, ::LITE_FOPT_KEEP_MTIME
+ * @param opt     An option mask of ::LITE_FOPT_RSYNC_DELETE, ::LITE_FOPT_KEEP_MTIME,
+ *                ::LITE_FOPT_RSYNC_LINKS
  * @param filter  Optional filtering function for source directory.
  *
  * This is a miniature implementation of the famous rsync for local use
@@ -59,6 +61,17 @@ static int set_mtime(char *fn, struct stat *st);
  *
  * %LITE_FOPT_RSYNC_DELETE: Prune files from @p dst that no longer exist in @p src.
  * %LITE_FOPT_KEEP_MTIME:   Preserve modification time
+ * %LITE_FOPT_RSYNC_LINKS:  Recreate symlinks to files in @p dst
+ *
+ * Symlinks in @p src are followed by default, so @p dst gets a copy of
+ * what they point to.  With ::LITE_FOPT_RSYNC_LINKS a symlink to a file,
+ * or a dangling symlink, is recreated in @p dst with the same target
+ * text, replacing any file or symlink already there.  Symlinks to
+ * directories in @p src are always followed.
+ *
+ * With ::LITE_FOPT_RSYNC_LINKS a symlink in @p dst is never followed.
+ * If one sits where a file or directory from @p src goes, it is removed
+ * first.
  *
  * The filter callback, @p filter, if provided, is used to determine
  * what files to include from the source directory when backing up.  If
@@ -86,7 +99,7 @@ int rsync(char *src, char *dst, int opt, int (*filter)(const char *file))
 		if (!fexist(src))
 			return 1;
 
-		if (copy(src, dst, keep_mtim))
+		if (copy(src, dst, opt))
 			return 1;
 
 		return 0;
@@ -104,7 +117,7 @@ int rsync(char *src, char *dst, int opt, int (*filter)(const char *file))
 		if (stat(src, &st))
 			return 1;
 
-		if (mdir(dest, sizeof(dest), dst, ptr, &st))
+		if (mdir(dest, sizeof(dest), dst, ptr, &st, opt))
 			return 1;
 		dst = dest;
 		do_mtim = keep_mtim;
@@ -124,7 +137,7 @@ int rsync(char *src, char *dst, int opt, int (*filter)(const char *file))
 				continue;
 			}
 
-			if (mdir(dst2, sizeof(dst2), dst, files[i], &sb)) {
+			if (mdir(dst2, sizeof(dst2), dst, files[i], &sb, opt)) {
 				result++;
 				continue;
 			}
@@ -136,7 +149,7 @@ int rsync(char *src, char *dst, int opt, int (*filter)(const char *file))
 			continue;	/* Next file/dir in @src to copy... */
 		}
 
-		if (copy(source, dst, keep_mtim))
+		if (copy(source, dst, opt))
 			result++;
 	}
 
@@ -157,23 +170,91 @@ int rsync(char *src, char *dst, int opt, int (*filter)(const char *file))
 	return result;
 }
 
+/* Leaves errno alone, a missing @path is not an error to rsync() */
+static int islink(const char *path)
+{
+	int rc, saved_errno = errno;
+	struct stat st;
+
+	rc = !lstat(path, &st) && S_ISLNK(st.st_mode);
+	errno = saved_errno;
+
+	return rc;
+}
+
 static int copy(char *src, char *dst, int opt)
 {
-	copyfile(src, dst, 0, opt | LITE_FOPT_COPYFILE_SYM);
-	if (errno) {
-		if (errno != EEXIST)
-			return 1;
+	char *dest = NULL, *temp = NULL, *file = NULL;
+	int result = 1, saved_errno, made_temp = 0;
+	int copt = opt & LITE_FOPT_KEEP_MTIME;
 
-		errno = 0;
+	if (!(opt & LITE_FOPT_RSYNC_LINKS)) {
+		copyfile(src, dst, 0, copt);
+		return errno ? 1 : 0;
+	}
+	copt |= LITE_FOPT_COPYFILE_SYM;
+
+	if (fisdir(dst)) {
+		char *base = strrchr(src, '/');
+
+		if (asprintf(&dest, "%s%s%s", dst, fisslashdir(dst) ? "" : "/", base ? base + 1 : src) < 0)
+			return 1;
+	} else {
+		dest = strdup(dst);
+		if (!dest)
+			return 1;
 	}
 
-	return 0;
+	/*
+	 * A link in the destination, e.g. from a previous run, may point
+	 * anywhere.  Remove it so copyfile() cannot write through it.
+	 */
+	if (islink(dest) && !islink(src))
+		unlink(dest);
+
+	copyfile(src, dst, 0, copt);
+	if (errno != EEXIST) {
+		result = errno ? 1 : 0;
+		goto exit;
+	}
+
+	/* Replace an existing destination only after the new link is ready. */
+	if (asprintf(&temp, "%s/.rsync-XXXXXX", dirname(strdupa(dest))) < 0)
+		goto exit;
+	if (!mkdtemp(temp))
+		goto exit;
+	made_temp = 1;
+	if (asprintf(&file, "%s/link", temp) < 0)
+		goto exit;
+
+	copyfile(src, file, 0, copt);
+	if (errno)
+		goto exit;
+	result = rename(file, dest) != 0;
+
+exit:
+	saved_errno = errno;
+	if (file)
+		unlink(file);
+	if (made_temp)
+		rmdir(temp);
+	free(file);
+	free(temp);
+	free(dest);
+	errno = saved_errno;
+
+	return result;
 }
 
 /* Creates dir/name @mode ... skipping / if dir already ends so. */
-static int mdir(char *buf, size_t buf_len, char *dir, char *name, struct stat *st)
+static int mdir(char *buf, size_t buf_len, char *dir, char *name, struct stat *st, int opt)
 {
 	snprintf(buf, buf_len, "%s%s%s", dir, fisslashdir(dir) ? "" : "/", name);
+
+	/* Same as in copy(), do not descend into a stale link. */
+	if ((opt & LITE_FOPT_RSYNC_LINKS) && islink(buf))
+		unlink(buf);
+
 	if (mkdir(buf, st->st_mode)) {
 		if (EEXIST != errno)
 			return 1;

test/.gitignore

@@ -5,6 +5,7 @@
 /chomp
 /copyfile
 /fcopyfile
+/copyfile-symlink
 /dir
 /fexist
 /fisdir

test/Makefile.am

@@ -20,6 +20,7 @@ TESTS           =
 TESTS          += chomp
 TESTS          += copyfile
 TESTS          += fcopyfile
+TESTS          += copyfile-symlink
 TESTS          += dir
 TESTS          += fexist
 TESTS          += fisdir
@@ -41,3 +42,4 @@ TESTS          += which
 TESTS          += yorn
 
 check_PROGRAMS  = $(TESTS)
+copyfile_symlink_SOURCES = copyfile-symlink.c

test/copyfile-symlink.c

@@ -0,0 +1,152 @@
+#include <errno.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include "check.h"
+
+static void check_link(const char *path, const char *target)
+{
+	char buf[256];
+	struct stat st;
+	ssize_t len;
+
+	fail_unless(lstat(path, &st) == 0 && S_ISLNK(st.st_mode));
+	len = readlink(path, buf, sizeof(buf));
+	fail_unless(len == (ssize_t)strlen(target));
+	fail_unless(memcmp(buf, target, len) == 0);
+}
+
+/* A link left in the destination by an earlier run must never be written through. */
+static void rsync_stale_links(void)
+{
+	struct stat st;
+	FILE *fp;
+
+	fail_unless(mkdir("src", 0700) == 0 && mkdir("dst", 0700) == 0);
+	fail_unless(touch("victim") == 0);
+	fail_unless(symlink("../victim", "src/a") == 0);
+	fail_unless(symlink("../victims", "src/d") == 0);
+	fail_unless(rsync("src/", "dst/", LITE_FOPT_RSYNC_LINKS, NULL) == 0);
+	check_link("dst/a", "../victim");
+	check_link("dst/d", "../victims");
+
+	/* The source links become a file and a directory, and dst/d stops dangling. */
+	fail_unless(unlink("src/a") == 0 && unlink("src/d") == 0);
+	fp = fopen("src/a", "w");
+	fail_unless(fp && fputs("data", fp) >= 0 && fclose(fp) == 0);
+	fail_unless(mkdir("src/d", 0700) == 0 && touch("src/d/f") == 0);
+	fail_unless(mkdir("victims", 0700) == 0);
+	fail_unless(rsync("src/", "dst/", LITE_FOPT_RSYNC_LINKS, NULL) == 0);
+
+	fail_unless(lstat("dst/a", &st) == 0 && S_ISREG(st.st_mode) && st.st_size == 4);
+	fail_unless(stat("victim", &st) == 0 && st.st_size == 0);
+	fail_unless(lstat("dst/d", &st) == 0 && S_ISDIR(st.st_mode));
+	fail_unless(fexist("dst/d/f") && !fexist("victims/f"));
+
+	fail_unless(unlink("src/a") == 0 && unlink("src/d/f") == 0 && rmdir("src/d") == 0);
+	fail_unless(unlink("dst/a") == 0 && unlink("dst/d/f") == 0 && rmdir("dst/d") == 0);
+	fail_unless(rmdir("src") == 0 && rmdir("dst") == 0);
+	fail_unless(unlink("victim") == 0 && rmdir("victims") == 0);
+}
+
+int main(void)
+{
+	char tmp[] = "/tmp/libite-copylink-XXXXXX";
+	const char payload[] = "target contents";
+	char buf[sizeof(payload)];
+	const char *targets[] = { "file", "dir", "missing", ".", "link" };
+	struct timespec times[2] = { { 1234567800, 0 }, { 1234567801, 0 } };
+	struct stat src_st, dst_st;
+	char cwd[4096];
+	size_t i;
+	FILE *fp;
+
+	fail_unless(getcwd(cwd, sizeof(cwd)) && mkdtemp(tmp));
+	fail_unless(chdir(tmp) == 0);
+	fail_unless(mkdir("dir", 0700) == 0);
+	fail_unless(mkdir("out", 0700) == 0);
+	fp = fopen("file", "w");
+	fail_unless(fp && fwrite(payload, 1, sizeof(payload), fp) == sizeof(payload));
+	fail_unless(fclose(fp) == 0);
+
+	for (i = 0; i < NELEMS(targets); i++) {
+		fail_unless(symlink(targets[i], "link") == 0);
+		fail_unless(copyfile("link", "copied", 0, LITE_FOPT_COPYFILE_SYM) == 1);
+		check_link("copied", targets[i]);
+		fail_unless(copyfile("link", "out", 0, LITE_FOPT_COPYFILE_SYM) == 1);
+		check_link("out/link", targets[i]);
+		/* Existing targets must survive a failed symlink creation. */
+		errno = 0;
+		fail_unless(copyfile("link", "file", 0, LITE_FOPT_COPYFILE_SYM) == 0);
+		fail_unless(errno == EEXIST);
+		check_link("copied", targets[i]);
+		fail_unless(unlink("copied") == 0 && unlink("out/link") == 0);
+		fail_unless(unlink("link") == 0);
+	}
+
+	fail_unless(symlink("file", "link") == 0);
+	/* Without SYM, continue to copy target data and preserve regular-file metadata. */
+	fail_unless(copyfile("link", "copied", 0, LITE_FOPT_KEEP_MTIME) == sizeof(payload));
+	fail_unless(lstat("copied", &dst_st) == 0 && S_ISREG(dst_st.st_mode));
+	fail_unless(stat("file", &src_st) == 0 && src_st.st_mtime == dst_st.st_mtime);
+	fp = fopen("copied", "r");
+	fail_unless(fp && fread(buf, 1, sizeof(buf), fp) == sizeof(payload));
+	fail_unless(memcmp(buf, payload, sizeof(payload)) == 0 && fclose(fp) == 0);
+	fail_unless(unlink("copied") == 0);
+
+	fail_unless(utimensat(AT_FDCWD, "link", times, AT_SYMLINK_NOFOLLOW) == 0);
+	fail_unless(copyfile("link", "copied", 0, LITE_FOPT_COPYFILE_SYM | LITE_FOPT_KEEP_MTIME) == 1);
+	check_link("copied", "file");
+	fail_unless(lstat("copied", &dst_st) == 0 && dst_st.st_mtime == times[1].tv_sec);
+	fail_unless(unlink("copied") == 0);
+	/* rsync() follows links unless asked to recreate them. */
+	fail_unless(rsync("link", "copied", 0, NULL) == 0);
+	fail_unless(lstat("copied", &dst_st) == 0 && S_ISREG(dst_st.st_mode));
+	fail_unless(dst_st.st_size == sizeof(payload));
+	fail_unless(unlink("copied") == 0);
+	fail_unless(rsync("link", "copied", LITE_FOPT_RSYNC_LINKS, NULL) == 0);
+	check_link("copied", "file");
+	/* Reuse the destination after changing the source link. */
+	fail_unless(unlink("link") == 0 && symlink("file", "link") == 0);
+	fail_unless(rsync("link", "out", LITE_FOPT_RSYNC_LINKS, NULL) == 0);
+	check_link("out/link", "file");
+	fail_unless(unlink("link") == 0 && symlink("copied", "link") == 0);
+	fail_unless(rsync("link", "out", LITE_FOPT_RSYNC_LINKS, NULL) == 0);
+	check_link("out/link", "copied");
+	/* A same-path copy must not unlink its source. */
+	fail_unless(rsync("link", "link", LITE_FOPT_RSYNC_LINKS, NULL) == 0);
+	check_link("link", "copied");
+	fail_unless(unlink("out/link") == 0);
+	/* Replacing a regular destination must also retain the source link. */
+	fail_unless(touch("out/link") == 0);
+	fail_unless(rsync("link", "out", LITE_FOPT_RSYNC_LINKS, NULL) == 0);
+	check_link("out/link", "copied");
+	fail_unless(unlink("out/link") == 0);
+	/* A failed replacement must keep both the destination and source. */
+	fail_unless(mkdir("out/link", 0700) == 0);
+	fail_unless(touch("out/link/keep") == 0);
+	fail_unless(rsync("link", "out", LITE_FOPT_RSYNC_LINKS, NULL) != 0);
+	fail_unless(fexist("out/link/keep"));
+	check_link("link", "copied");
+	fail_unless(unlink("out/link/keep") == 0 && rmdir("out/link") == 0);
+	if (getuid() != 0) {
+		fail_unless(mkdir("protected", 0700) == 0);
+		fail_unless(symlink("original", "protected/link") == 0);
+		fail_unless(chmod("protected", 0500) == 0);
+		fail_unless(rsync("link", "protected", LITE_FOPT_RSYNC_LINKS, NULL) != 0);
+		check_link("protected/link", "original");
+		check_link("link", "copied");
+		fail_unless(chmod("protected", 0700) == 0);
+		fail_unless(unlink("protected/link") == 0 && rmdir("protected") == 0);
+	}
+	fail_unless(unlink("copied") == 0 && unlink("link") == 0);
+	rsync_stale_links();
+
+	errno = 0;
+	fail_unless(copyfile("dir", "copied", 0, LITE_FOPT_COPYFILE_SYM) == 0 && errno == EISDIR);
+	errno = 0;
+	fail_unless(copyfile("missing", "copied", 0, LITE_FOPT_COPYFILE_SYM) == -1 && errno == ENOENT);
+
+	fail_unless(unlink("file") == 0 && rmdir("dir") == 0 && rmdir("out") == 0);
+	fail_unless(chdir(cwd) == 0 && rmdir(tmp) == 0);
+	return 0;
+}