make USE_NSEC work as expected
authorKjetil Barvik <barvik@broadpark.no>
Thu, 19 Feb 2009 20:08:29 +0000 (21:08 +0100)
committerJunio C Hamano <gitster@pobox.com>
Fri, 20 Feb 2009 05:39:48 +0000 (21:39 -0800)
Since the filesystem ext4 is now defined as stable in Linux v2.6.28,
and ext4 supports nanonsecond resolution timestamps natively, it is
time to make USE_NSEC work as expected.

This will make racy git situations less likely to happen.  For 'git
checkout' this means it will be less likely that we have to open, read
the contents of the file into RAM, and check if file is really
modified or not.  The result sould be a litle less used CPU time, less
pagefaults and a litle faster program, at least for 'git checkout'.

Since the number of possible racy git situations would increase when
disks gets faster, this patch would be more and more helpfull as times
go by.  For a fast Solid State Disk, this patch should be helpfull.

Note that, when file operations starts to take less than 1 nanosecond,
one would again start to get more racy git situations.

For more info on racy git, see Documentation/technical/racy-git.txt
For more info on ext4, see http://kernelnewbies.org/Ext4

Signed-off-by: Kjetil Barvik <barvik@broadpark.no>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
cache.h
read-cache.c
unpack-trees.c

diff --git a/cache.h b/cache.h
index 770d8bc950557a72b449ffa0e7841e226a924832..2badbfedc47ebf58a979b9e2d160f223ef05206a 100644 (file)
--- a/cache.h
+++ b/cache.h
@@ -140,8 +140,8 @@ struct ondisk_cache_entry_extended {
 };
 
 struct cache_entry {
-       unsigned int ce_ctime;
-       unsigned int ce_mtime;
+       struct cache_time ce_ctime;
+       struct cache_time ce_mtime;
        unsigned int ce_dev;
        unsigned int ce_ino;
        unsigned int ce_mode;
@@ -282,7 +282,7 @@ struct index_state {
        struct cache_entry **cache;
        unsigned int cache_nr, cache_alloc, cache_changed;
        struct cache_tree *cache_tree;
-       time_t timestamp;
+       struct cache_time timestamp;
        void *alloc;
        unsigned name_hash_initialized : 1,
                 initialized : 1;
index 59a274b464754bc5e79949cf294111ccc6bc0e8b..bb07371597970c6343194e3e3a4e660e5d9201f6 100644 (file)
@@ -67,8 +67,15 @@ void rename_index_entry_at(struct index_state *istate, int nr, const char *new_n
  */
 void fill_stat_cache_info(struct cache_entry *ce, struct stat *st)
 {
-       ce->ce_ctime = st->st_ctime;
-       ce->ce_mtime = st->st_mtime;
+       ce->ce_ctime.sec = (unsigned int)st->st_ctime;
+       ce->ce_mtime.sec = (unsigned int)st->st_mtime;
+#ifdef USE_NSEC
+       ce->ce_ctime.nsec = (unsigned int)st->st_ctim.tv_nsec;
+       ce->ce_mtime.nsec = (unsigned int)st->st_mtim.tv_nsec;
+#else
+       ce->ce_ctime.nsec = 0;
+       ce->ce_mtime.nsec = 0;
+#endif
        ce->ce_dev = st->st_dev;
        ce->ce_ino = st->st_ino;
        ce->ce_uid = st->st_uid;
@@ -196,11 +203,18 @@ static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
        default:
                die("internal error: ce_mode is %o", ce->ce_mode);
        }
-       if (ce->ce_mtime != (unsigned int) st->st_mtime)
+       if (ce->ce_mtime.sec != (unsigned int)st->st_mtime)
                changed |= MTIME_CHANGED;
-       if (trust_ctime && ce->ce_ctime != (unsigned int) st->st_ctime)
+       if (trust_ctime && ce->ce_ctime.sec != (unsigned int)st->st_ctime)
                changed |= CTIME_CHANGED;
 
+#ifdef USE_NSEC
+       if (ce->ce_mtime.nsec != (unsigned int)st->st_mtim.tv_nsec)
+               changed |= MTIME_CHANGED;
+       if (trust_ctime && ce->ce_ctime.nsec != (unsigned int)st->st_ctim.tv_nsec)
+               changed |= CTIME_CHANGED;
+#endif
+
        if (ce->ce_uid != (unsigned int) st->st_uid ||
            ce->ce_gid != (unsigned int) st->st_gid)
                changed |= OWNER_CHANGED;
@@ -232,8 +246,16 @@ static int ce_match_stat_basic(struct cache_entry *ce, struct stat *st)
 static int is_racy_timestamp(const struct index_state *istate, struct cache_entry *ce)
 {
        return (!S_ISGITLINK(ce->ce_mode) &&
-               istate->timestamp &&
-               ((unsigned int)istate->timestamp) <= ce->ce_mtime);
+               istate->timestamp.sec &&
+#ifdef USE_NSEC
+                /* nanosecond timestamped files can also be racy! */
+               (istate->timestamp.sec < ce->ce_mtime.sec ||
+                (istate->timestamp.sec == ce->ce_mtime.sec &&
+                 istate->timestamp.nsec <= ce->ce_mtime.nsec))
+#else
+               istate->timestamp.sec <= ce->ce_mtime.sec
+#endif
+                );
 }
 
 int ie_match_stat(const struct index_state *istate,
@@ -1159,8 +1181,15 @@ static void convert_from_disk(struct ondisk_cache_entry *ondisk, struct cache_en
        size_t len;
        const char *name;
 
-       ce->ce_ctime = ntohl(ondisk->ctime.sec);
-       ce->ce_mtime = ntohl(ondisk->mtime.sec);
+       ce->ce_ctime.sec = ntohl(ondisk->ctime.sec);
+       ce->ce_mtime.sec = ntohl(ondisk->mtime.sec);
+#ifdef USE_NSEC
+       ce->ce_ctime.nsec = ntohl(ondisk->ctime.nsec);
+       ce->ce_mtime.nsec = ntohl(ondisk->mtime.nsec);
+#else
+       ce->ce_ctime.nsec = 0;
+       ce->ce_mtime.nsec = 0;
+#endif
        ce->ce_dev   = ntohl(ondisk->dev);
        ce->ce_ino   = ntohl(ondisk->ino);
        ce->ce_mode  = ntohl(ondisk->mode);
@@ -1226,7 +1255,8 @@ int read_index_from(struct index_state *istate, const char *path)
                return istate->cache_nr;
 
        errno = ENOENT;
-       istate->timestamp = 0;
+       istate->timestamp.sec = 0;
+       istate->timestamp.nsec = 0;
        fd = open(path, O_RDONLY);
        if (fd < 0) {
                if (errno == ENOENT)
@@ -1278,7 +1308,13 @@ int read_index_from(struct index_state *istate, const char *path)
                src_offset += ondisk_ce_size(ce);
                dst_offset += ce_size(ce);
        }
-       istate->timestamp = st.st_mtime;
+       istate->timestamp.sec = st.st_mtime;
+#ifdef USE_NSEC
+       istate->timestamp.nsec = (unsigned int)st.st_mtim.tv_nsec;
+#else
+       istate->timestamp.nsec = 0;
+#endif
+
        while (src_offset <= mmap_size - 20 - 8) {
                /* After an array of active_nr index entries,
                 * there can be arbitrary number of extended
@@ -1308,14 +1344,15 @@ unmap:
 
 int is_index_unborn(struct index_state *istate)
 {
-       return (!istate->cache_nr && !istate->alloc && !istate->timestamp);
+       return (!istate->cache_nr && !istate->alloc && !istate->timestamp.sec);
 }
 
 int discard_index(struct index_state *istate)
 {
        istate->cache_nr = 0;
        istate->cache_changed = 0;
-       istate->timestamp = 0;
+       istate->timestamp.sec = 0;
+       istate->timestamp.nsec = 0;
        istate->name_hash_initialized = 0;
        free_hash(&istate->name_hash);
        cache_tree_free(&(istate->cache_tree));
@@ -1461,10 +1498,15 @@ static int ce_write_entry(git_SHA_CTX *c, int fd, struct cache_entry *ce)
        struct ondisk_cache_entry *ondisk = xcalloc(1, size);
        char *name;
 
-       ondisk->ctime.sec = htonl(ce->ce_ctime);
+       ondisk->ctime.sec = htonl(ce->ce_ctime.sec);
+       ondisk->mtime.sec = htonl(ce->ce_mtime.sec);
+#ifdef USE_NSEC
+       ondisk->ctime.nsec = htonl(ce->ce_ctime.nsec);
+       ondisk->mtime.nsec = htonl(ce->ce_mtime.nsec);
+#else
        ondisk->ctime.nsec = 0;
-       ondisk->mtime.sec = htonl(ce->ce_mtime);
        ondisk->mtime.nsec = 0;
+#endif
        ondisk->dev  = htonl(ce->ce_dev);
        ondisk->ino  = htonl(ce->ce_ino);
        ondisk->mode = htonl(ce->ce_mode);
index 273b5da0a256337c3d3973b475e7cfa567cee35a..11902cd51da9d83952df03edc1aadffc41e3f05b 100644 (file)
@@ -360,8 +360,12 @@ int unpack_trees(unsigned len, struct tree_desc *t, struct unpack_trees_options
 
        memset(&o->result, 0, sizeof(o->result));
        o->result.initialized = 1;
-       if (o->src_index)
-               o->result.timestamp = o->src_index->timestamp;
+       if (o->src_index) {
+               o->result.timestamp.sec = o->src_index->timestamp.sec;
+#ifdef USE_NSEC
+               o->result.timestamp.nsec = o->src_index->timestamp.nsec;
+#endif
+       }
        o->merge_size = len;
 
        if (!dfc)