fsck.c:fsck_commit(): use skip_prefix() to verify and skip constant
[gitweb.git] / sha1_file.c
index 406aac45d902e0f26ea2755d4388fc6224d16a57..6e8c05d10825ee93d8248f749d4bef513d772be4 100644 (file)
@@ -105,50 +105,63 @@ int mkdir_in_gitdir(const char *path)
        return adjust_shared_perm(path);
 }
 
-int safe_create_leading_directories(char *path)
+enum scld_error safe_create_leading_directories(char *path)
 {
-       char *pos = path + offset_1st_component(path);
-       struct stat st;
+       char *next_component = path + offset_1st_component(path);
+       enum scld_error ret = SCLD_OK;
+
+       while (ret == SCLD_OK && next_component) {
+               struct stat st;
+               char *slash = next_component, slash_character;
+
+               while (*slash && !is_dir_sep(*slash))
+                       slash++;
 
-       while (pos) {
-               pos = strchr(pos, '/');
-               if (!pos)
+               if (!*slash)
                        break;
-               while (*++pos == '/')
-                       ;
-               if (!*pos)
+
+               next_component = slash + 1;
+               while (is_dir_sep(*next_component))
+                       next_component++;
+               if (!*next_component)
                        break;
-               *--pos = '\0';
+
+               slash_character = *slash;
+               *slash = '\0';
                if (!stat(path, &st)) {
                        /* path exists */
-                       if (!S_ISDIR(st.st_mode)) {
-                               *pos = '/';
-                               return -3;
-                       }
-               }
-               else if (mkdir(path, 0777)) {
+                       if (!S_ISDIR(st.st_mode))
+                               ret = SCLD_EXISTS;
+               } else if (mkdir(path, 0777)) {
                        if (errno == EEXIST &&
-                           !stat(path, &st) && S_ISDIR(st.st_mode)) {
+                           !stat(path, &st) && S_ISDIR(st.st_mode))
                                ; /* somebody created it since we checked */
-                       } else {
-                               *pos = '/';
-                               return -1;
-                       }
+                       else if (errno == ENOENT)
+                               /*
+                                * Either mkdir() failed because
+                                * somebody just pruned the containing
+                                * directory, or stat() failed because
+                                * the file that was in our way was
+                                * just removed.  Either way, inform
+                                * the caller that it might be worth
+                                * trying again:
+                                */
+                               ret = SCLD_VANISHED;
+                       else
+                               ret = SCLD_FAILED;
+               } else if (adjust_shared_perm(path)) {
+                       ret = SCLD_PERMS;
                }
-               else if (adjust_shared_perm(path)) {
-                       *pos = '/';
-                       return -2;
-               }
-               *pos++ = '/';
+               *slash = slash_character;
        }
-       return 0;
+       return ret;
 }
 
-int safe_create_leading_directories_const(const char *path)
+enum scld_error safe_create_leading_directories_const(const char *path)
 {
        /* path points to cache entries, so xstrdup before messing with it */
        char *buf = xstrdup(path);
-       int result = safe_create_leading_directories(buf);
+       enum scld_error result = safe_create_leading_directories(buf);
        free(buf);
        return result;
 }
@@ -614,7 +627,7 @@ static void scan_windows(struct packed_git *p,
        }
 }
 
-static int unuse_one_window(struct packed_git *current, int keep_fd)
+static int unuse_one_window(struct packed_git *current)
 {
        struct packed_git *p, *lru_p = NULL;
        struct pack_window *lru_w = NULL, *lru_l = NULL;
@@ -628,15 +641,8 @@ static int unuse_one_window(struct packed_git *current, int keep_fd)
                pack_mapped -= lru_w->len;
                if (lru_l)
                        lru_l->next = lru_w->next;
-               else {
+               else
                        lru_p->windows = lru_w->next;
-                       if (!lru_p->windows && lru_p->pack_fd != -1
-                               && lru_p->pack_fd != keep_fd) {
-                               close(lru_p->pack_fd);
-                               pack_open_fds--;
-                               lru_p->pack_fd = -1;
-                       }
-               }
                free(lru_w);
                pack_open_windows--;
                return 1;
@@ -644,10 +650,10 @@ static int unuse_one_window(struct packed_git *current, int keep_fd)
        return 0;
 }
 
-void release_pack_memory(size_t need, int fd)
+void release_pack_memory(size_t need)
 {
        size_t cur = pack_mapped;
-       while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
+       while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
                ; /* nothing */
 }
 
@@ -658,7 +664,7 @@ void *xmmap(void *start, size_t length,
        if (ret == MAP_FAILED) {
                if (!length)
                        return NULL;
-               release_pack_memory(length, fd);
+               release_pack_memory(length);
                ret = mmap(start, length, prot, flags, fd, offset);
                if (ret == MAP_FAILED)
                        die_errno("Out of memory? mmap failed");
@@ -682,6 +688,83 @@ void close_pack_windows(struct packed_git *p)
        }
 }
 
+/*
+ * The LRU pack is the one with the oldest MRU window, preferring packs
+ * with no used windows, or the oldest mtime if it has no windows allocated.
+ */
+static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
+{
+       struct pack_window *w, *this_mru_w;
+       int has_windows_inuse = 0;
+
+       /*
+        * Reject this pack if it has windows and the previously selected
+        * one does not.  If this pack does not have windows, reject
+        * it if the pack file is newer than the previously selected one.
+        */
+       if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
+               return;
+
+       for (w = this_mru_w = p->windows; w; w = w->next) {
+               /*
+                * Reject this pack if any of its windows are in use,
+                * but the previously selected pack did not have any
+                * inuse windows.  Otherwise, record that this pack
+                * has windows in use.
+                */
+               if (w->inuse_cnt) {
+                       if (*accept_windows_inuse)
+                               has_windows_inuse = 1;
+                       else
+                               return;
+               }
+
+               if (w->last_used > this_mru_w->last_used)
+                       this_mru_w = w;
+
+               /*
+                * Reject this pack if it has windows that have been
+                * used more recently than the previously selected pack.
+                * If the previously selected pack had windows inuse and
+                * we have not encountered a window in this pack that is
+                * inuse, skip this check since we prefer a pack with no
+                * inuse windows to one that has inuse windows.
+                */
+               if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
+                   this_mru_w->last_used > (*mru_w)->last_used)
+                       return;
+       }
+
+       /*
+        * Select this pack.
+        */
+       *mru_w = this_mru_w;
+       *lru_p = p;
+       *accept_windows_inuse = has_windows_inuse;
+}
+
+static int close_one_pack(void)
+{
+       struct packed_git *p, *lru_p = NULL;
+       struct pack_window *mru_w = NULL;
+       int accept_windows_inuse = 1;
+
+       for (p = packed_git; p; p = p->next) {
+               if (p->pack_fd == -1)
+                       continue;
+               find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
+       }
+
+       if (lru_p) {
+               close(lru_p->pack_fd);
+               pack_open_fds--;
+               lru_p->pack_fd = -1;
+               return 1;
+       }
+
+       return 0;
+}
+
 void unuse_pack(struct pack_window **w_cursor)
 {
        struct pack_window *w = *w_cursor;
@@ -737,15 +820,38 @@ void free_pack_by_name(const char *pack_name)
 static unsigned int get_max_fd_limit(void)
 {
 #ifdef RLIMIT_NOFILE
-       struct rlimit lim;
+       {
+               struct rlimit lim;
 
-       if (getrlimit(RLIMIT_NOFILE, &lim))
-               die_errno("cannot get RLIMIT_NOFILE");
+               if (!getrlimit(RLIMIT_NOFILE, &lim))
+                       return lim.rlim_cur;
+       }
+#endif
 
-       return lim.rlim_cur;
-#elif defined(_SC_OPEN_MAX)
-       return sysconf(_SC_OPEN_MAX);
-#elif defined(OPEN_MAX)
+#ifdef _SC_OPEN_MAX
+       {
+               long open_max = sysconf(_SC_OPEN_MAX);
+               if (0 < open_max)
+                       return open_max;
+               /*
+                * Otherwise, we got -1 for one of the two
+                * reasons:
+                *
+                * (1) sysconf() did not understand _SC_OPEN_MAX
+                *     and signaled an error with -1; or
+                * (2) sysconf() said there is no limit.
+                *
+                * We _could_ clear errno before calling sysconf() to
+                * tell these two cases apart and return a huge number
+                * in the latter case to let the caller cap it to a
+                * value that is not so selfish, but letting the
+                * fallback OPEN_MAX codepath take care of these cases
+                * is a lot simpler.
+                */
+       }
+#endif
+
+#ifdef OPEN_MAX
        return OPEN_MAX;
 #else
        return 1; /* see the caller ;-) */
@@ -777,7 +883,7 @@ static int open_packed_git_1(struct packed_git *p)
                        pack_max_fds = 1;
        }
 
-       while (pack_max_fds <= pack_open_fds && unuse_one_window(NULL, -1))
+       while (pack_max_fds <= pack_open_fds && close_one_pack())
                ; /* nothing */
 
        p->pack_fd = git_open_noatime(p->pack_name);
@@ -893,7 +999,7 @@ unsigned char *use_pack(struct packed_git *p,
                        win->len = (size_t)len;
                        pack_mapped += win->len;
                        while (packed_git_limit < pack_mapped
-                               && unuse_one_window(p, p->pack_fd))
+                               && unuse_one_window(p))
                                ; /* nothing */
                        win->base = xmmap(NULL, win->len,
                                PROT_READ, MAP_PRIVATE,
@@ -939,7 +1045,7 @@ static struct packed_git *alloc_packed_git(int extra)
 
 static void try_to_free_pack_memory(size_t size)
 {
-       release_pack_memory(size, -1);
+       release_pack_memory(size);
 }
 
 struct packed_git *add_packed_git(const char *path, int path_len, int local)
@@ -1597,6 +1703,38 @@ static off_t get_delta_base(struct packed_git *p,
        return base_offset;
 }
 
+/*
+ * Like get_delta_base above, but we return the sha1 instead of the pack
+ * offset. This means it is cheaper for REF deltas (we do not have to do
+ * the final object lookup), but more expensive for OFS deltas (we
+ * have to load the revidx to convert the offset back into a sha1).
+ */
+static const unsigned char *get_delta_base_sha1(struct packed_git *p,
+                                               struct pack_window **w_curs,
+                                               off_t curpos,
+                                               enum object_type type,
+                                               off_t delta_obj_offset)
+{
+       if (type == OBJ_REF_DELTA) {
+               unsigned char *base = use_pack(p, w_curs, curpos, NULL);
+               return base;
+       } else if (type == OBJ_OFS_DELTA) {
+               struct revindex_entry *revidx;
+               off_t base_offset = get_delta_base(p, w_curs, &curpos,
+                                                  type, delta_obj_offset);
+
+               if (!base_offset)
+                       return NULL;
+
+               revidx = find_pack_revindex(p, base_offset);
+               if (!revidx)
+                       return NULL;
+
+               return nth_packed_object_sha1(p, revidx->nr);
+       } else
+               return NULL;
+}
+
 int unpack_object_header(struct packed_git *p,
                         struct pack_window **w_curs,
                         off_t *curpos,
@@ -1754,6 +1892,22 @@ static int packed_object_info(struct packed_git *p, off_t obj_offset,
                }
        }
 
+       if (oi->delta_base_sha1) {
+               if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
+                       const unsigned char *base;
+
+                       base = get_delta_base_sha1(p, &w_curs, curpos,
+                                                  type, obj_offset);
+                       if (!base) {
+                               type = OBJ_BAD;
+                               goto out;
+                       }
+
+                       hashcpy(oi->delta_base_sha1, base);
+               } else
+                       hashclr(oi->delta_base_sha1);
+       }
+
 out:
        unuse_pack(&w_curs);
        return type;
@@ -1982,6 +2136,16 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
                int i;
                struct delta_base_cache_entry *ent;
 
+               ent = get_delta_base_cache_entry(p, curpos);
+               if (eq_delta_base_cache_entry(ent, p, curpos)) {
+                       type = ent->type;
+                       data = ent->data;
+                       size = ent->size;
+                       clear_delta_base_cache_entry(ent);
+                       base_from_cache = 1;
+                       break;
+               }
+
                if (do_check_packed_object_crc && p->index_version > 1) {
                        struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
                        unsigned long len = revidx[1].offset - obj_offset;
@@ -1996,16 +2160,6 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
                        }
                }
 
-               ent = get_delta_base_cache_entry(p, curpos);
-               if (eq_delta_base_cache_entry(ent, p, curpos)) {
-                       type = ent->type;
-                       data = ent->data;
-                       size = ent->size;
-                       clear_delta_base_cache_entry(ent);
-                       base_from_cache = 1;
-                       break;
-               }
-
                type = unpack_object_header(p, &w_curs, &curpos, &size);
                if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
                        break;
@@ -2337,17 +2491,23 @@ static int sha1_loose_object_info(const unsigned char *sha1,
        git_zstream stream;
        char hdr[32];
 
+       if (oi->delta_base_sha1)
+               hashclr(oi->delta_base_sha1);
+
        /*
         * If we don't care about type or size, then we don't
-        * need to look inside the object at all.
+        * need to look inside the object at all. Note that we
+        * do not optimize out the stat call, even if the
+        * caller doesn't care about the disk-size, since our
+        * return value implicitly indicates whether the
+        * object even exists.
         */
        if (!oi->typep && !oi->sizep) {
-               if (oi->disk_sizep) {
-                       struct stat st;
-                       if (stat_sha1_file(sha1, &st) < 0)
-                               return -1;
+               struct stat st;
+               if (stat_sha1_file(sha1, &st) < 0)
+                       return -1;
+               if (oi->disk_sizep)
                        *oi->disk_sizep = st.st_size;
-               }
                return 0;
        }
 
@@ -2370,13 +2530,14 @@ static int sha1_loose_object_info(const unsigned char *sha1,
        return 0;
 }
 
-int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
+int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
 {
        struct cached_object *co;
        struct pack_entry e;
        int rtype;
+       const unsigned char *real = lookup_replace_object_extended(sha1, flags);
 
-       co = find_cached_object(sha1);
+       co = find_cached_object(real);
        if (co) {
                if (oi->typep)
                        *(oi->typep) = co->type;
@@ -2384,27 +2545,29 @@ int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi)
                        *(oi->sizep) = co->size;
                if (oi->disk_sizep)
                        *(oi->disk_sizep) = 0;
+               if (oi->delta_base_sha1)
+                       hashclr(oi->delta_base_sha1);
                oi->whence = OI_CACHED;
                return 0;
        }
 
-       if (!find_pack_entry(sha1, &e)) {
+       if (!find_pack_entry(real, &e)) {
                /* Most likely it's a loose object. */
-               if (!sha1_loose_object_info(sha1, oi)) {
+               if (!sha1_loose_object_info(real, oi)) {
                        oi->whence = OI_LOOSE;
                        return 0;
                }
 
                /* Not a loose object; someone else may have just packed it. */
                reprepare_packed_git();
-               if (!find_pack_entry(sha1, &e))
+               if (!find_pack_entry(real, &e))
                        return -1;
        }
 
        rtype = packed_object_info(e.p, e.offset, oi);
        if (rtype < 0) {
-               mark_bad_packed_object(e.p, sha1);
-               return sha1_object_info_extended(sha1, oi);
+               mark_bad_packed_object(e.p, real);
+               return sha1_object_info_extended(real, oi, 0);
        } else if (in_delta_base_cache(e.p, e.offset)) {
                oi->whence = OI_DBCACHED;
        } else {
@@ -2426,7 +2589,7 @@ int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
 
        oi.typep = &type;
        oi.sizep = sizep;
-       if (sha1_object_info_extended(sha1, &oi) < 0)
+       if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
                return -1;
        return type;
 }
@@ -2518,8 +2681,7 @@ void *read_sha1_file_extended(const unsigned char *sha1,
        void *data;
        char *path;
        const struct packed_git *p;
-       const unsigned char *repl = (flag & READ_SHA1_FILE_REPLACE)
-               ? lookup_replace_object(sha1) : sha1;
+       const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
 
        errno = 0;
        data = read_object(repl, type, size);
@@ -2713,7 +2875,9 @@ static int create_tmpfile(char *buffer, size_t bufsiz, const char *filename)
                /* Make sure the directory exists */
                memcpy(buffer, filename, dirlen);
                buffer[dirlen-1] = 0;
-               if (mkdir(buffer, 0777) || adjust_shared_perm(buffer))
+               if (mkdir(buffer, 0777) && errno != EEXIST)
+                       return -1;
+               if (adjust_shared_perm(buffer))
                        return -1;
 
                /* Try again */