posix: use nanosecond accuracy when available

Programs that set mtime, such as `rsync -a`, don't work correctly on
GlusterFS, because it sets the nanoseconds to 000. This creates
problems for incremental backups, where files get accidentally copied
again and again.

For example, consider `myfile` on an ext4 system, being copied to a
GlusterFS volume, with `rsync -a` and then `cp -u` in turn. You'd
expect that after the first `rsync -a`, `cp -u` agrees that the file
need not be copied.

BUG: 1422074
Change-Id: I89c7b6a73e2e06c02851ff76b7e5cdfaa271e985
Signed-off-by: Niklas Hambüchen <mail@nh2.me>
Reviewed-on: https://review.gluster.org/16667
Smoke: Gluster Build System <jenkins@build.gluster.org>
Reviewed-by: Niels de Vos <ndevos@redhat.com>
Tested-by: Jeff Darcy <jdarcy@redhat.com>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
Reviewed-by: jiffin tony Thottan <jthottan@redhat.com>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>

6 files changed, 87 insertions, 13 deletions

diff --git a/configure.ac b/configure.ac
index 6917322bd58..28ff3e5a3f4 100644
--- a/configure.ac
+++ b/configure.ac
@@ -981,6 +981,13 @@ if test "x${have_posix_fallocate}" = "xyes"; then
    AC_DEFINE(HAVE_POSIX_FALLOCATE, 1, [define if posix_fallocate exists])
 fi
 
+BUILD_NANOSECOND_TIMESTAMPS=no
+AC_CHECK_FUNC([utimensat], [have_utimensat=yes])
+if test "x${have_utimensat}" = "xyes"; then
+   BUILD_NANOSECOND_TIMESTAMPS=yes
+   AC_DEFINE(HAVE_UTIMENSAT, 1, [define if utimensat exists])
+fi
+
 OLD_CFLAGS=${CFLAGS}
 CFLAGS="-D_GNU_SOURCE"
 AC_CHECK_DECL([SEEK_HOLE], , , [#include <unistd.h>])
@@ -1544,4 +1551,5 @@ echo "Experimental xlators : $BUILD_EXPERIMENTAL"
 echo "Events               : $BUILD_EVENTS"
 echo "EC dynamic support   : $EC_DYNAMIC_SUPPORT"
 echo "Use memory pools     : $USE_MEMPOOL"
+echo "Nanosecond m/atimes  : $BUILD_NANOSECOND_TIMESTAMPS"
 echo
diff --git a/libglusterfs/src/common-utils.c b/libglusterfs/src/common-utils.c
index 51f97442fec..e18c97f5aa1 100644
--- a/libglusterfs/src/common-utils.c
+++ b/libglusterfs/src/common-utils.c
@@ -3806,7 +3806,11 @@ int
 gf_set_timestamp  (const char *src, const char* dest)
 {
         struct stat    sb             = {0, };
-        struct timeval new_time[2]    = {{0, },{0,}};
+#if defined(HAVE_UTIMENSAT)
+        struct timespec new_time[2]   = { {0, }, {0, } };
+#else
+        struct timeval new_time[2]    = { {0, }, {0, } };
+#endif
         int            ret            = 0;
         xlator_t       *this          = NULL;
 
@@ -3821,21 +3825,35 @@ gf_set_timestamp  (const char *src, const char* dest)
                         LG_MSG_FILE_STAT_FAILED, "stat on %s", src);
                 goto out;
         }
+        /* The granularity is nano seconds if `utimensat()` is available,
+         * and micro seconds otherwise.
+         */
+#if defined(HAVE_UTIMENSAT)
+        new_time[0].tv_sec = sb.st_atime;
+        new_time[0].tv_nsec = ST_ATIM_NSEC (&sb);
+
+        new_time[1].tv_sec = sb.st_mtime;
+        new_time[1].tv_nsec = ST_MTIM_NSEC (&sb);
+
+        /* dirfd = 0 is ignored because `dest` is an absolute path. */
+        ret = sys_utimensat (AT_FDCWD, dest, new_time, AT_SYMLINK_NOFOLLOW);
+        if (ret) {
+                gf_msg (this->name, GF_LOG_ERROR, errno,
+                        LG_MSG_UTIMENSAT_FAILED, "utimensat on %s", dest);
+        }
+#else
         new_time[0].tv_sec = sb.st_atime;
         new_time[0].tv_usec = ST_ATIM_NSEC (&sb)/1000;
 
         new_time[1].tv_sec = sb.st_mtime;
         new_time[1].tv_usec = ST_MTIM_NSEC (&sb)/1000;
 
-        /* The granularity is micro seconds as per the current
-         * requiremnt. Hence using 'utimes'. This can be updated
-         * to 'utimensat' if we need timestamp in nanoseconds.
-         */
         ret = sys_utimes (dest, new_time);
         if (ret) {
                 gf_msg (this->name, GF_LOG_ERROR, errno, LG_MSG_UTIMES_FAILED,
                         "utimes on %s", dest);
         }
+#endif
 out:
         return ret;
 }
diff --git a/libglusterfs/src/libglusterfs-messages.h b/libglusterfs/src/libglusterfs-messages.h
index 29196929eb3..23ed7b727d3 100644
--- a/libglusterfs/src/libglusterfs-messages.h
+++ b/libglusterfs/src/libglusterfs-messages.h
@@ -37,7 +37,7 @@
 
 #define GLFS_LG_BASE            GLFS_MSGID_COMP_LIBGLUSTERFS
 
-#define GLFS_LG_NUM_MESSAGES    209
+#define GLFS_LG_NUM_MESSAGES    210
 
 #define GLFS_LG_MSGID_END       (GLFS_LG_BASE + GLFS_LG_NUM_MESSAGES + 1)
 /* Messaged with message IDs */
@@ -1791,6 +1791,15 @@
  * @recommendedaction
  *
  */
+
+#define LG_MSG_UTIMENSAT_FAILED                          (GLFS_LG_BASE + 210)
+
+/*!
+ * @messageid
+ * @diagnosis
+ * @recommendedaction
+ *
+ */
 /*------------*/
 
 #define glfs_msg_end_lg GLFS_LG_MSGID_END, "Invalid: End of messages"
diff --git a/libglusterfs/src/syscall.c b/libglusterfs/src/syscall.c
index 4e4c6a728da..a7d4402808d 100644
--- a/libglusterfs/src/syscall.c
+++ b/libglusterfs/src/syscall.c
@@ -272,6 +272,16 @@ sys_utimes (const char *filename, const struct timeval times[2])
 }
 
 
+#if defined(HAVE_UTIMENSAT)
+int
+sys_utimensat (int dirfd, const char *filename, const struct timespec times[2],
+               int flags)
+{
+        return utimensat (dirfd, filename, times, flags);
+}
+#endif
+
+
 int
 sys_creat (const char *pathname, mode_t mode)
 {
diff --git a/libglusterfs/src/syscall.h b/libglusterfs/src/syscall.h
index b1bcad138c5..6bb374822ee 100644
--- a/libglusterfs/src/syscall.h
+++ b/libglusterfs/src/syscall.h
@@ -129,6 +129,12 @@ sys_ftruncate (int fd, off_t length);
 int
 sys_utimes (const char *filename, const struct timeval times[2]);
 
+#if defined(HAVE_UTIMENSAT)
+int
+sys_utimensat (int dirfd, const char *filename, const struct timespec times[2],
+               int flags);
+#endif
+
 int
 sys_creat (const char *pathname, mode_t mode);
 
diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix.c
index f3fca45492b..04d141f41b2 100644
--- a/xlators/storage/posix/src/posix.c
+++ b/xlators/storage/posix/src/posix.c
@@ -84,6 +84,25 @@ extern char *marker_xattrs[];
 
 #endif
 
+/* Setting microseconds or nanoseconds depending on what's supported:
+   The passed in `tv` can be
+       struct timespec
+   if supported (better, because it supports nanosecond resolution) or
+       struct timeval
+   otherwise. */
+#if HAVE_UTIMENSAT
+#define SET_TIMESPEC_NSEC_OR_TIMEVAL_USEC(tv, nanosecs) \
+        tv.tv_nsec = nanosecs
+#define PATH_SET_TIMESPEC_OR_TIMEVAL(path, tv) \
+        (sys_utimensat (AT_FDCWD, path, tv, AT_SYMLINK_NOFOLLOW))
+#else
+#define SET_TIMESPEC_NSEC_OR_TIMEVAL_USEC(tv, nanosecs) \
+        tv.tv_usec = nanosecs / 1000
+#define PATH_SET_TIMESPEC_OR_TIMEVAL(path, tv) \
+        (lutimes (path, tv))
+#endif
+
+
 dict_t*
 posix_dict_set_nlink (dict_t *req, dict_t *res, int32_t nlink)
 {
@@ -384,7 +403,11 @@ posix_do_utimes (xlator_t *this,
                  int valid)
 {
         int32_t ret = -1;
-        struct timeval tv[2]     = {{0,},{0,}};
+#if defined(HAVE_UTIMENSAT)
+        struct timespec tv[2]    = { {0,}, {0,} };
+#else
+        struct timeval tv[2]     = { {0,}, {0,} };
+#endif
         struct stat stat;
         int    is_symlink = 0;
 
@@ -400,23 +423,23 @@ posix_do_utimes (xlator_t *this,
 
         if ((valid & GF_SET_ATTR_ATIME) == GF_SET_ATTR_ATIME) {
                 tv[0].tv_sec  = stbuf->ia_atime;
-                tv[0].tv_usec = stbuf->ia_atime_nsec / 1000;
+                SET_TIMESPEC_NSEC_OR_TIMEVAL_USEC(tv[0], stbuf->ia_atime_nsec);
         } else {
                 /* atime is not given, use current values */
                 tv[0].tv_sec  = ST_ATIM_SEC (&stat);
-                tv[0].tv_usec = ST_ATIM_NSEC (&stat) / 1000;
+                SET_TIMESPEC_NSEC_OR_TIMEVAL_USEC(tv[0], ST_ATIM_NSEC (&stat));
         }
 
         if ((valid & GF_SET_ATTR_MTIME) == GF_SET_ATTR_MTIME) {
                 tv[1].tv_sec  = stbuf->ia_mtime;
-                tv[1].tv_usec = stbuf->ia_mtime_nsec / 1000;
+                SET_TIMESPEC_NSEC_OR_TIMEVAL_USEC(tv[1], stbuf->ia_mtime_nsec);
         } else {
                 /* mtime is not given, use current values */
                 tv[1].tv_sec  = ST_MTIM_SEC (&stat);
-                tv[1].tv_usec = ST_MTIM_NSEC (&stat) / 1000;
+                SET_TIMESPEC_NSEC_OR_TIMEVAL_USEC(tv[1], ST_MTIM_NSEC (&stat));
         }
 
-        ret = lutimes (path, tv);
+        ret = PATH_SET_TIMESPEC_OR_TIMEVAL(path, tv);
         if ((ret == -1) && (errno == ENOSYS)) {
                 gf_msg_debug (this->name, 0, "%s (%s)",
                         path, strerror (errno));
@@ -425,7 +448,7 @@ posix_do_utimes (xlator_t *this,
                         goto out;
                 }
 
-                ret = sys_utimes (path, tv);
+                ret = PATH_SET_TIMESPEC_OR_TIMEVAL(path, tv);
         }
 
 out: