verify_path: drop clever fallthrough
[gitweb.git] / sha1_file.c
index 727a9769fb7549e365e66f4ecb65f8e665898a96..dd90a785419b851adf97643db9b291ab23b01224 100644 (file)
 #include "streaming.h"
 #include "dir.h"
 #include "mru.h"
-
-#ifndef O_NOATIME
-#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
-#define O_NOATIME 01000000
-#else
-#define O_NOATIME 0
-#endif
-#endif
+#include "list.h"
+#include "mergesort.h"
+#include "quote.h"
 
 #define SZ_FMT PRIuMAX
 static inline uintmax_t sz_fmt(size_t s) { return s; }
 
 const unsigned char null_sha1[20];
 const struct object_id null_oid;
+const struct object_id empty_tree_oid = {
+       EMPTY_TREE_SHA1_BIN_LITERAL
+};
+const struct object_id empty_blob_oid = {
+       EMPTY_BLOB_SHA1_BIN_LITERAL
+};
 
 /*
  * This is meant to hold a *small* number of objects that you would
@@ -128,8 +129,10 @@ enum scld_error safe_create_leading_directories(char *path)
                *slash = '\0';
                if (!stat(path, &st)) {
                        /* path exists */
-                       if (!S_ISDIR(st.st_mode))
+                       if (!S_ISDIR(st.st_mode)) {
+                               errno = ENOTDIR;
                                ret = SCLD_EXISTS;
+                       }
                } else if (mkdir(path, 0777)) {
                        if (errno == EEXIST &&
                            !stat(path, &st) && S_ISDIR(st.st_mode))
@@ -157,75 +160,177 @@ enum scld_error safe_create_leading_directories(char *path)
 
 enum scld_error safe_create_leading_directories_const(const char *path)
 {
+       int save_errno;
        /* path points to cache entries, so xstrdup before messing with it */
        char *buf = xstrdup(path);
        enum scld_error result = safe_create_leading_directories(buf);
+
+       save_errno = errno;
        free(buf);
+       errno = save_errno;
        return result;
 }
 
-static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
+int raceproof_create_file(const char *path, create_file_fn fn, void *cb)
+{
+       /*
+        * The number of times we will try to remove empty directories
+        * in the way of path. This is only 1 because if another
+        * process is racily creating directories that conflict with
+        * us, we don't want to fight against them.
+        */
+       int remove_directories_remaining = 1;
+
+       /*
+        * The number of times that we will try to create the
+        * directories containing path. We are willing to attempt this
+        * more than once, because another process could be trying to
+        * clean up empty directories at the same time as we are
+        * trying to create them.
+        */
+       int create_directories_remaining = 3;
+
+       /* A scratch copy of path, filled lazily if we need it: */
+       struct strbuf path_copy = STRBUF_INIT;
+
+       int ret, save_errno;
+
+       /* Sanity check: */
+       assert(*path);
+
+retry_fn:
+       ret = fn(path, cb);
+       save_errno = errno;
+       if (!ret)
+               goto out;
+
+       if (errno == EISDIR && remove_directories_remaining-- > 0) {
+               /*
+                * A directory is in the way. Maybe it is empty; try
+                * to remove it:
+                */
+               if (!path_copy.len)
+                       strbuf_addstr(&path_copy, path);
+
+               if (!remove_dir_recursively(&path_copy, REMOVE_DIR_EMPTY_ONLY))
+                       goto retry_fn;
+       } else if (errno == ENOENT && create_directories_remaining-- > 0) {
+               /*
+                * Maybe the containing directory didn't exist, or
+                * maybe it was just deleted by a process that is
+                * racing with us to clean up empty directories. Try
+                * to create it:
+                */
+               enum scld_error scld_result;
+
+               if (!path_copy.len)
+                       strbuf_addstr(&path_copy, path);
+
+               do {
+                       scld_result = safe_create_leading_directories(path_copy.buf);
+                       if (scld_result == SCLD_OK)
+                               goto retry_fn;
+               } while (scld_result == SCLD_VANISHED && create_directories_remaining-- > 0);
+       }
+
+out:
+       strbuf_release(&path_copy);
+       errno = save_errno;
+       return ret;
+}
+
+static void fill_sha1_path(struct strbuf *buf, const unsigned char *sha1)
 {
        int i;
        for (i = 0; i < 20; i++) {
                static char hex[] = "0123456789abcdef";
                unsigned int val = sha1[i];
-               char *pos = pathbuf + i*2 + (i > 0);
-               *pos++ = hex[val >> 4];
-               *pos = hex[val & 0xf];
+               strbuf_addch(buf, hex[val >> 4]);
+               strbuf_addch(buf, hex[val & 0xf]);
+               if (!i)
+                       strbuf_addch(buf, '/');
        }
 }
 
 const char *sha1_file_name(const unsigned char *sha1)
 {
-       static char buf[PATH_MAX];
-       const char *objdir;
-       int len;
+       static struct strbuf buf = STRBUF_INIT;
 
-       objdir = get_object_directory();
-       len = strlen(objdir);
+       strbuf_reset(&buf);
+       strbuf_addf(&buf, "%s/", get_object_directory());
 
-       /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
-       if (len + 43 > PATH_MAX)
-               die("insanely long object directory %s", objdir);
-       memcpy(buf, objdir, len);
-       buf[len] = '/';
-       buf[len+3] = '/';
-       buf[len+42] = '\0';
-       fill_sha1_path(buf + len + 1, sha1);
-       return buf;
+       fill_sha1_path(&buf, sha1);
+       return buf.buf;
 }
 
-/*
- * Return the name of the pack or index file with the specified sha1
- * in its filename.  *base and *name are scratch space that must be
- * provided by the caller.  which should be "pack" or "idx".
- */
-static char *sha1_get_pack_name(const unsigned char *sha1,
-                               struct strbuf *buf,
-                               const char *which)
+struct strbuf *alt_scratch_buf(struct alternate_object_database *alt)
+{
+       strbuf_setlen(&alt->scratch, alt->base_len);
+       return &alt->scratch;
+}
+
+static const char *alt_sha1_path(struct alternate_object_database *alt,
+                                const unsigned char *sha1)
+{
+       struct strbuf *buf = alt_scratch_buf(alt);
+       fill_sha1_path(buf, sha1);
+       return buf->buf;
+}
+
+ char *odb_pack_name(struct strbuf *buf,
+                    const unsigned char *sha1,
+                    const char *ext)
 {
        strbuf_reset(buf);
        strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
-                   sha1_to_hex(sha1), which);
+                   sha1_to_hex(sha1), ext);
        return buf->buf;
 }
 
 char *sha1_pack_name(const unsigned char *sha1)
 {
        static struct strbuf buf = STRBUF_INIT;
-       return sha1_get_pack_name(sha1, &buf, "pack");
+       return odb_pack_name(&buf, sha1, "pack");
 }
 
 char *sha1_pack_index_name(const unsigned char *sha1)
 {
        static struct strbuf buf = STRBUF_INIT;
-       return sha1_get_pack_name(sha1, &buf, "idx");
+       return odb_pack_name(&buf, sha1, "idx");
 }
 
 struct alternate_object_database *alt_odb_list;
 static struct alternate_object_database **alt_odb_tail;
 
+/*
+ * Return non-zero iff the path is usable as an alternate object database.
+ */
+static int alt_odb_usable(struct strbuf *path, const char *normalized_objdir)
+{
+       struct alternate_object_database *alt;
+
+       /* Detect cases where alternate disappeared */
+       if (!is_directory(path->buf)) {
+               error("object directory %s does not exist; "
+                     "check .git/objects/info/alternates.",
+                     path->buf);
+               return 0;
+       }
+
+       /*
+        * Prevent the common mistake of listing the same
+        * thing twice, or object directory itself.
+        */
+       for (alt = alt_odb_list; alt; alt = alt->next) {
+               if (!fspathcmp(path->buf, alt->path))
+                       return 0;
+       }
+       if (!fspathcmp(path->buf, normalized_objdir))
+               return 0;
+
+       return 1;
+}
+
 /*
  * Prepare alternate object database registry.
  *
@@ -245,59 +350,34 @@ static int link_alt_odb_entry(const char *entry, const char *relative_base,
        int depth, const char *normalized_objdir)
 {
        struct alternate_object_database *ent;
-       struct alternate_object_database *alt;
-       size_t pfxlen, entlen;
        struct strbuf pathbuf = STRBUF_INIT;
 
        if (!is_absolute_path(entry) && relative_base) {
-               strbuf_addstr(&pathbuf, real_path(relative_base));
+               strbuf_realpath(&pathbuf, relative_base, 1);
                strbuf_addch(&pathbuf, '/');
        }
        strbuf_addstr(&pathbuf, entry);
 
-       normalize_path_copy(pathbuf.buf, pathbuf.buf);
-
-       pfxlen = strlen(pathbuf.buf);
+       if (strbuf_normalize_path(&pathbuf) < 0 && relative_base) {
+               error("unable to normalize alternate object path: %s",
+                     pathbuf.buf);
+               strbuf_release(&pathbuf);
+               return -1;
+       }
 
        /*
         * The trailing slash after the directory name is given by
         * this function at the end. Remove duplicates.
         */
-       while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
-               pfxlen -= 1;
+       while (pathbuf.len && pathbuf.buf[pathbuf.len - 1] == '/')
+               strbuf_setlen(&pathbuf, pathbuf.len - 1);
 
-       entlen = st_add(pfxlen, 43); /* '/' + 2 hex + '/' + 38 hex + NUL */
-       ent = xmalloc(st_add(sizeof(*ent), entlen));
-       memcpy(ent->base, pathbuf.buf, pfxlen);
-       strbuf_release(&pathbuf);
-
-       ent->name = ent->base + pfxlen + 1;
-       ent->base[pfxlen + 3] = '/';
-       ent->base[pfxlen] = ent->base[entlen-1] = 0;
-
-       /* Detect cases where alternate disappeared */
-       if (!is_directory(ent->base)) {
-               error("object directory %s does not exist; "
-                     "check .git/objects/info/alternates.",
-                     ent->base);
-               free(ent);
+       if (!alt_odb_usable(&pathbuf, normalized_objdir)) {
+               strbuf_release(&pathbuf);
                return -1;
        }
 
-       /* Prevent the common mistake of listing the same
-        * thing twice, or object directory itself.
-        */
-       for (alt = alt_odb_list; alt; alt = alt->next) {
-               if (pfxlen == alt->name - alt->base - 1 &&
-                   !memcmp(ent->base, alt->base, pfxlen)) {
-                       free(ent);
-                       return -1;
-               }
-       }
-       if (!fspathcmp(ent->base, normalized_objdir)) {
-               free(ent);
-               return -1;
-       }
+       ent = alloc_alt_odb(pathbuf.buf);
 
        /* add the alternate entry */
        *alt_odb_tail = ent;
@@ -305,20 +385,46 @@ static int link_alt_odb_entry(const char *entry, const char *relative_base,
        ent->next = NULL;
 
        /* recursively add alternates */
-       read_info_alternates(ent->base, depth + 1);
-
-       ent->base[pfxlen] = '/';
+       read_info_alternates(pathbuf.buf, depth + 1);
 
+       strbuf_release(&pathbuf);
        return 0;
 }
 
+static const char *parse_alt_odb_entry(const char *string,
+                                      int sep,
+                                      struct strbuf *out)
+{
+       const char *end;
+
+       strbuf_reset(out);
+
+       if (*string == '#') {
+               /* comment; consume up to next separator */
+               end = strchrnul(string, sep);
+       } else if (*string == '"' && !unquote_c_style(out, string, &end)) {
+               /*
+                * quoted path; unquote_c_style has copied the
+                * data for us and set "end". Broken quoting (e.g.,
+                * an entry that doesn't end with a quote) falls
+                * back to the unquoted case below.
+                */
+       } else {
+               /* normal, unquoted path */
+               end = strchrnul(string, sep);
+               strbuf_add(out, string, end - string);
+       }
+
+       if (*end)
+               end++;
+       return end;
+}
+
 static void link_alt_odb_entries(const char *alt, int len, int sep,
                                 const char *relative_base, int depth)
 {
-       struct string_list entries = STRING_LIST_INIT_NODUP;
-       char *alt_copy;
-       int i;
        struct strbuf objdirbuf = STRBUF_INIT;
+       struct strbuf entry = STRBUF_INIT;
 
        if (depth > 5) {
                error("%s: ignoring alternate object stores, nesting too deep.",
@@ -327,23 +433,17 @@ static void link_alt_odb_entries(const char *alt, int len, int sep,
        }
 
        strbuf_add_absolute_path(&objdirbuf, get_object_directory());
-       normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
+       if (strbuf_normalize_path(&objdirbuf) < 0)
+               die("unable to normalize object directory: %s",
+                   objdirbuf.buf);
 
-       alt_copy = xmemdupz(alt, len);
-       string_list_split_in_place(&entries, alt_copy, sep, -1);
-       for (i = 0; i < entries.nr; i++) {
-               const char *entry = entries.items[i].string;
-               if (entry[0] == '\0' || entry[0] == '#')
+       while (*alt) {
+               alt = parse_alt_odb_entry(alt, sep, &entry);
+               if (!entry.len)
                        continue;
-               if (!is_absolute_path(entry) && depth) {
-                       error("%s: ignoring relative alternate object store %s",
-                                       relative_base, entry);
-               } else {
-                       link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
-               }
+               link_alt_odb_entry(entry.buf, relative_base, depth, objdirbuf.buf);
        }
-       string_list_clear(&entries, 0);
-       free(alt_copy);
+       strbuf_release(&entry);
        strbuf_release(&objdirbuf);
 }
 
@@ -356,7 +456,7 @@ void read_info_alternates(const char * relative_base, int depth)
        int fd;
 
        path = xstrfmt("%s/info/alternates", relative_base);
-       fd = git_open_noatime(path);
+       fd = git_open(path);
        free(path);
        if (fd < 0)
                return;
@@ -373,6 +473,18 @@ void read_info_alternates(const char * relative_base, int depth)
        munmap(map, mapsz);
 }
 
+struct alternate_object_database *alloc_alt_odb(const char *dir)
+{
+       struct alternate_object_database *ent;
+
+       FLEX_ALLOC_STR(ent, path, dir);
+       strbuf_init(&ent->scratch, 0);
+       strbuf_addf(&ent->scratch, "%s/", dir);
+       ent->base_len = ent->scratch.len;
+
+       return ent;
+}
+
 void add_to_alternates_file(const char *reference)
 {
        struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
@@ -418,6 +530,93 @@ void add_to_alternates_file(const char *reference)
        free(alts);
 }
 
+void add_to_alternates_memory(const char *reference)
+{
+       /*
+        * Make sure alternates are initialized, or else our entry may be
+        * overwritten when they are.
+        */
+       prepare_alt_odb();
+
+       link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
+}
+
+/*
+ * Compute the exact path an alternate is at and returns it. In case of
+ * error NULL is returned and the human readable error is added to `err`
+ * `path` may be relative and should point to $GITDIR.
+ * `err` must not be null.
+ */
+char *compute_alternate_path(const char *path, struct strbuf *err)
+{
+       char *ref_git = NULL;
+       const char *repo, *ref_git_s;
+       int seen_error = 0;
+
+       ref_git_s = real_path_if_valid(path);
+       if (!ref_git_s) {
+               seen_error = 1;
+               strbuf_addf(err, _("path '%s' does not exist"), path);
+               goto out;
+       } else
+               /*
+                * Beware: read_gitfile(), real_path() and mkpath()
+                * return static buffer
+                */
+               ref_git = xstrdup(ref_git_s);
+
+       repo = read_gitfile(ref_git);
+       if (!repo)
+               repo = read_gitfile(mkpath("%s/.git", ref_git));
+       if (repo) {
+               free(ref_git);
+               ref_git = xstrdup(repo);
+       }
+
+       if (!repo && is_directory(mkpath("%s/.git/objects", ref_git))) {
+               char *ref_git_git = mkpathdup("%s/.git", ref_git);
+               free(ref_git);
+               ref_git = ref_git_git;
+       } else if (!is_directory(mkpath("%s/objects", ref_git))) {
+               struct strbuf sb = STRBUF_INIT;
+               seen_error = 1;
+               if (get_common_dir(&sb, ref_git)) {
+                       strbuf_addf(err,
+                                   _("reference repository '%s' as a linked "
+                                     "checkout is not supported yet."),
+                                   path);
+                       goto out;
+               }
+
+               strbuf_addf(err, _("reference repository '%s' is not a "
+                                       "local repository."), path);
+               goto out;
+       }
+
+       if (!access(mkpath("%s/shallow", ref_git), F_OK)) {
+               strbuf_addf(err, _("reference repository '%s' is shallow"),
+                           path);
+               seen_error = 1;
+               goto out;
+       }
+
+       if (!access(mkpath("%s/info/grafts", ref_git), F_OK)) {
+               strbuf_addf(err,
+                           _("reference repository '%s' is grafted"),
+                           path);
+               seen_error = 1;
+               goto out;
+       }
+
+out:
+       if (seen_error) {
+               free(ref_git);
+               ref_git = NULL;
+       }
+
+       return ref_git;
+}
+
 int foreach_alt_odb(alt_odb_fn fn, void *cb)
 {
        struct alternate_object_database *ent;
@@ -463,7 +662,7 @@ static int freshen_file(const char *fn)
  * either does not exist on disk, or has a stale mtime and may be subject to
  * pruning).
  */
-static int check_and_freshen_file(const char *fn, int freshen)
+int check_and_freshen_file(const char *fn, int freshen)
 {
        if (access(fn, F_OK))
                return 0;
@@ -482,8 +681,8 @@ static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
        struct alternate_object_database *alt;
        prepare_alt_odb();
        for (alt = alt_odb_list; alt; alt = alt->next) {
-               fill_sha1_path(alt->name, sha1);
-               if (check_and_freshen_file(alt->base, freshen))
+               const char *path = alt_sha1_path(alt, sha1);
+               if (check_and_freshen_file(path, freshen))
                        return 1;
        }
        return 0;
@@ -550,7 +749,7 @@ static int check_packed_git_idx(const char *path, struct packed_git *p)
        struct pack_idx_header *hdr;
        size_t idx_size;
        uint32_t version, nr, i, *index;
-       int fd = git_open_noatime(path);
+       int fd = git_open(path);
        struct stat st;
 
        if (fd < 0)
@@ -956,7 +1155,7 @@ static int open_packed_git_1(struct packed_git *p)
        while (pack_max_fds <= pack_open_fds && close_one_pack())
                ; /* nothing */
 
-       p->pack_fd = git_open_noatime(p->pack_name);
+       p->pack_fd = git_open(p->pack_name);
        if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
                return -1;
        pack_open_fds++;
@@ -1297,10 +1496,46 @@ static void prepare_packed_git_one(char *objdir, int local)
        strbuf_release(&path);
 }
 
+static int approximate_object_count_valid;
+
+/*
+ * Give a fast, rough count of the number of objects in the repository. This
+ * ignores loose objects completely. If you have a lot of them, then either
+ * you should repack because your performance will be awful, or they are
+ * all unreachable objects about to be pruned, in which case they're not really
+ * interesting as a measure of repo size in the first place.
+ */
+unsigned long approximate_object_count(void)
+{
+       static unsigned long count;
+       if (!approximate_object_count_valid) {
+               struct packed_git *p;
+
+               prepare_packed_git();
+               count = 0;
+               for (p = packed_git; p; p = p->next) {
+                       if (open_pack_index(p))
+                               continue;
+                       count += p->num_objects;
+               }
+       }
+       return count;
+}
+
+static void *get_next_packed_git(const void *p)
+{
+       return ((const struct packed_git *)p)->next;
+}
+
+static void set_next_packed_git(void *p, void *next)
+{
+       ((struct packed_git *)p)->next = next;
+}
+
 static int sort_pack(const void *a_, const void *b_)
 {
-       struct packed_git *a = *((struct packed_git **)a_);
-       struct packed_git *b = *((struct packed_git **)b_);
+       const struct packed_git *a = a_;
+       const struct packed_git *b = b_;
        int st;
 
        /*
@@ -1327,28 +1562,8 @@ static int sort_pack(const void *a_, const void *b_)
 
 static void rearrange_packed_git(void)
 {
-       struct packed_git **ary, *p;
-       int i, n;
-
-       for (n = 0, p = packed_git; p; p = p->next)
-               n++;
-       if (n < 2)
-               return;
-
-       /* prepare an array of packed_git for easier sorting */
-       ary = xcalloc(n, sizeof(struct packed_git *));
-       for (n = 0, p = packed_git; p; p = p->next)
-               ary[n++] = p;
-
-       qsort(ary, n, sizeof(struct packed_git *), sort_pack);
-
-       /* link them back again */
-       for (i = 0; i < n - 1; i++)
-               ary[i]->next = ary[i + 1];
-       ary[n - 1]->next = NULL;
-       packed_git = ary[0];
-
-       free(ary);
+       packed_git = llist_mergesort(packed_git, get_next_packed_git,
+                                    set_next_packed_git, sort_pack);
 }
 
 static void prepare_packed_git_mru(void)
@@ -1369,11 +1584,8 @@ void prepare_packed_git(void)
                return;
        prepare_packed_git_one(get_object_directory(), 1);
        prepare_alt_odb();
-       for (alt = alt_odb_list; alt; alt = alt->next) {
-               alt->name[-1] = 0;
-               prepare_packed_git_one(alt->base, 0);
-               alt->name[-1] = '/';
-       }
+       for (alt = alt_odb_list; alt; alt = alt->next)
+               prepare_packed_git_one(alt->path, 0);
        rearrange_packed_git();
        prepare_packed_git_mru();
        prepare_packed_git_run_once = 1;
@@ -1381,6 +1593,7 @@ void prepare_packed_git(void)
 
 void reprepare_packed_git(void)
 {
+       approximate_object_count_valid = 0;
        prepare_packed_git_run_once = 0;
        prepare_packed_git();
 }
@@ -1393,7 +1606,7 @@ static void mark_bad_packed_object(struct packed_git *p,
                if (!hashcmp(sha1, p->bad_object_sha1 + GIT_SHA1_RAWSZ * i))
                        return;
        p->bad_object_sha1 = xrealloc(p->bad_object_sha1,
-                                     st_mult(GIT_SHA1_RAWSZ,
+                                     st_mult(GIT_MAX_RAWSZ,
                                              st_add(p->num_bad_objects, 1)));
        hashcpy(p->bad_object_sha1 + GIT_SHA1_RAWSZ * p->num_bad_objects, sha1);
        p->num_bad_objects++;
@@ -1459,61 +1672,81 @@ int check_sha1_signature(const unsigned char *sha1, void *map,
        return hashcmp(sha1, real_sha1) ? -1 : 0;
 }
 
-int git_open_noatime(const char *name)
+int git_open_cloexec(const char *name, int flags)
 {
-       static int sha1_file_open_flag = O_NOATIME;
+       int fd;
+       static int o_cloexec = O_CLOEXEC;
 
-       for (;;) {
-               int fd;
+       fd = open(name, flags | o_cloexec);
+       if ((o_cloexec & O_CLOEXEC) && fd < 0 && errno == EINVAL) {
+               /* Try again w/o O_CLOEXEC: the kernel might not support it */
+               o_cloexec &= ~O_CLOEXEC;
+               fd = open(name, flags | o_cloexec);
+       }
 
-               errno = 0;
-               fd = open(name, O_RDONLY | sha1_file_open_flag);
-               if (fd >= 0)
-                       return fd;
+#if defined(F_GETFD) && defined(F_SETFD) && defined(FD_CLOEXEC)
+       {
+               static int fd_cloexec = FD_CLOEXEC;
 
-               /* Might the failure be due to O_NOATIME? */
-               if (errno != ENOENT && sha1_file_open_flag) {
-                       sha1_file_open_flag = 0;
-                       continue;
+               if (!o_cloexec && 0 <= fd && fd_cloexec) {
+                       /* Opened w/o O_CLOEXEC?  try with fcntl(2) to add it */
+                       int flags = fcntl(fd, F_GETFD);
+                       if (fcntl(fd, F_SETFD, flags | fd_cloexec))
+                               fd_cloexec = 0;
                }
-
-               return -1;
        }
+#endif
+       return fd;
 }
 
-static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
+/*
+ * Find "sha1" as a loose object in the local repository or in an alternate.
+ * Returns 0 on success, negative on failure.
+ *
+ * The "path" out-parameter will give the path of the object we found (if any).
+ * Note that it may point to static storage and is only valid until another
+ * call to sha1_file_name(), etc.
+ */
+static int stat_sha1_file(const unsigned char *sha1, struct stat *st,
+                         const char **path)
 {
        struct alternate_object_database *alt;
 
-       if (!lstat(sha1_file_name(sha1), st))
+       *path = sha1_file_name(sha1);
+       if (!lstat(*path, st))
                return 0;
 
        prepare_alt_odb();
        errno = ENOENT;
        for (alt = alt_odb_list; alt; alt = alt->next) {
-               fill_sha1_path(alt->name, sha1);
-               if (!lstat(alt->base, st))
+               *path = alt_sha1_path(alt, sha1);
+               if (!lstat(*path, st))
                        return 0;
        }
 
        return -1;
 }
 
-static int open_sha1_file(const unsigned char *sha1)
+/*
+ * Like stat_sha1_file(), but actually open the object and return the
+ * descriptor. See the caveats on the "path" parameter above.
+ */
+static int open_sha1_file(const unsigned char *sha1, const char **path)
 {
        int fd;
        struct alternate_object_database *alt;
        int most_interesting_errno;
 
-       fd = git_open_noatime(sha1_file_name(sha1));
+       *path = sha1_file_name(sha1);
+       fd = git_open(*path);
        if (fd >= 0)
                return fd;
        most_interesting_errno = errno;
 
        prepare_alt_odb();
        for (alt = alt_odb_list; alt; alt = alt->next) {
-               fill_sha1_path(alt->name, sha1);
-               fd = git_open_noatime(alt->base);
+               *path = alt_sha1_path(alt, sha1);
+               fd = git_open(*path);
                if (fd >= 0)
                        return fd;
                if (most_interesting_errno == ENOENT)
@@ -1523,12 +1756,21 @@ static int open_sha1_file(const unsigned char *sha1)
        return -1;
 }
 
-void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
+/*
+ * Map the loose object at "path" if it is not NULL, or the path found by
+ * searching for a loose object named "sha1".
+ */
+static void *map_sha1_file_1(const char *path,
+                            const unsigned char *sha1,
+                            unsigned long *size)
 {
        void *map;
        int fd;
 
-       fd = open_sha1_file(sha1);
+       if (path)
+               fd = git_open(path);
+       else
+               fd = open_sha1_file(sha1, &path);
        map = NULL;
        if (fd >= 0) {
                struct stat st;
@@ -1537,7 +1779,7 @@ void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
                        *size = xsize_t(st.st_size);
                        if (!*size) {
                                /* mmap() is forbidden on empty files */
-                               error("object file %s is empty", sha1_file_name(sha1));
+                               error("object file %s is empty", path);
                                return NULL;
                        }
                        map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
@@ -1547,6 +1789,11 @@ void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
        return map;
 }
 
+void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
+{
+       return map_sha1_file_1(NULL, sha1, size);
+}
+
 unsigned long unpack_object_header_buffer(const unsigned char *buf,
                unsigned long len, enum object_type *type, unsigned long *sizep)
 {
@@ -1752,11 +1999,9 @@ static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
 
 int parse_sha1_header(const char *hdr, unsigned long *sizep)
 {
-       struct object_info oi;
+       struct object_info oi = OBJECT_INFO_INIT;
 
        oi.sizep = sizep;
-       oi.typename = NULL;
-       oi.typep = NULL;
        return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
 }
 
@@ -1994,8 +2239,8 @@ static enum object_type packed_to_object_type(struct packed_git *p,
        goto out;
 }
 
-static int packed_object_info(struct packed_git *p, off_t obj_offset,
-                             struct object_info *oi)
+int packed_object_info(struct packed_git *p, off_t obj_offset,
+                      struct object_info *oi)
 {
        struct pack_window *w_curs = NULL;
        unsigned long size;
@@ -2095,136 +2340,141 @@ static void *unpack_compressed_entry(struct packed_git *p,
        return buffer;
 }
 
-#define MAX_DELTA_CACHE (256)
-
+static struct hashmap delta_base_cache;
 static size_t delta_base_cached;
 
-static struct delta_base_cache_lru_list {
-       struct delta_base_cache_lru_list *prev;
-       struct delta_base_cache_lru_list *next;
-} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
+static LIST_HEAD(delta_base_cache_lru);
 
-static struct delta_base_cache_entry {
-       struct delta_base_cache_lru_list lru;
-       void *data;
+struct delta_base_cache_key {
        struct packed_git *p;
        off_t base_offset;
+};
+
+struct delta_base_cache_entry {
+       struct hashmap hash;
+       struct delta_base_cache_key key;
+       struct list_head lru;
+       void *data;
        unsigned long size;
        enum object_type type;
-} delta_base_cache[MAX_DELTA_CACHE];
+};
 
-static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
+static unsigned int pack_entry_hash(struct packed_git *p, off_t base_offset)
 {
-       unsigned long hash;
+       unsigned int hash;
 
-       hash = (unsigned long)(intptr_t)p + (unsigned long)base_offset;
+       hash = (unsigned int)(intptr_t)p + (unsigned int)base_offset;
        hash += (hash >> 8) + (hash >> 16);
-       return hash % MAX_DELTA_CACHE;
+       return hash;
 }
 
 static struct delta_base_cache_entry *
 get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
 {
-       unsigned long hash = pack_entry_hash(p, base_offset);
-       return delta_base_cache + hash;
+       struct hashmap_entry entry;
+       struct delta_base_cache_key key;
+
+       if (!delta_base_cache.cmpfn)
+               return NULL;
+
+       hashmap_entry_init(&entry, pack_entry_hash(p, base_offset));
+       key.p = p;
+       key.base_offset = base_offset;
+       return hashmap_get(&delta_base_cache, &entry, &key);
+}
+
+static int delta_base_cache_key_eq(const struct delta_base_cache_key *a,
+                                  const struct delta_base_cache_key *b)
+{
+       return a->p == b->p && a->base_offset == b->base_offset;
 }
 
-static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
-                                    struct packed_git *p, off_t base_offset)
+static int delta_base_cache_hash_cmp(const void *va, const void *vb,
+                                    const void *vkey)
 {
-       return (ent->data && ent->p == p && ent->base_offset == base_offset);
+       const struct delta_base_cache_entry *a = va, *b = vb;
+       const struct delta_base_cache_key *key = vkey;
+       if (key)
+               return !delta_base_cache_key_eq(&a->key, key);
+       else
+               return !delta_base_cache_key_eq(&a->key, &b->key);
 }
 
 static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
 {
-       struct delta_base_cache_entry *ent;
-       ent = get_delta_base_cache_entry(p, base_offset);
-       return eq_delta_base_cache_entry(ent, p, base_offset);
+       return !!get_delta_base_cache_entry(p, base_offset);
 }
 
-static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
+/*
+ * Remove the entry from the cache, but do _not_ free the associated
+ * entry data. The caller takes ownership of the "data" buffer, and
+ * should copy out any fields it wants before detaching.
+ */
+static void detach_delta_base_cache_entry(struct delta_base_cache_entry *ent)
 {
-       ent->data = NULL;
-       ent->lru.next->prev = ent->lru.prev;
-       ent->lru.prev->next = ent->lru.next;
+       hashmap_remove(&delta_base_cache, ent, &ent->key);
+       list_del(&ent->lru);
        delta_base_cached -= ent->size;
+       free(ent);
 }
 
 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
-       unsigned long *base_size, enum object_type *type, int keep_cache)
+       unsigned long *base_size, enum object_type *type)
 {
        struct delta_base_cache_entry *ent;
-       void *ret;
 
        ent = get_delta_base_cache_entry(p, base_offset);
-
-       if (!eq_delta_base_cache_entry(ent, p, base_offset))
+       if (!ent)
                return unpack_entry(p, base_offset, type, base_size);
 
-       ret = ent->data;
-
-       if (!keep_cache)
-               clear_delta_base_cache_entry(ent);
-       else
-               ret = xmemdupz(ent->data, ent->size);
        *type = ent->type;
        *base_size = ent->size;
-       return ret;
+       return xmemdupz(ent->data, ent->size);
 }
 
 static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
 {
-       if (ent->data) {
-               free(ent->data);
-               ent->data = NULL;
-               ent->lru.next->prev = ent->lru.prev;
-               ent->lru.prev->next = ent->lru.next;
-               delta_base_cached -= ent->size;
-       }
+       free(ent->data);
+       detach_delta_base_cache_entry(ent);
 }
 
 void clear_delta_base_cache(void)
 {
-       unsigned long p;
-       for (p = 0; p < MAX_DELTA_CACHE; p++)
-               release_delta_base_cache(&delta_base_cache[p]);
+       struct list_head *lru, *tmp;
+       list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
+               struct delta_base_cache_entry *entry =
+                       list_entry(lru, struct delta_base_cache_entry, lru);
+               release_delta_base_cache(entry);
+       }
 }
 
 static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
        void *base, unsigned long base_size, enum object_type type)
 {
-       unsigned long hash = pack_entry_hash(p, base_offset);
-       struct delta_base_cache_entry *ent = delta_base_cache + hash;
-       struct delta_base_cache_lru_list *lru;
+       struct delta_base_cache_entry *ent = xmalloc(sizeof(*ent));
+       struct list_head *lru, *tmp;
 
-       release_delta_base_cache(ent);
        delta_base_cached += base_size;
 
-       for (lru = delta_base_cache_lru.next;
-            delta_base_cached > delta_base_cache_limit
-            && lru != &delta_base_cache_lru;
-            lru = lru->next) {
-               struct delta_base_cache_entry *f = (void *)lru;
-               if (f->type == OBJ_BLOB)
-                       release_delta_base_cache(f);
-       }
-       for (lru = delta_base_cache_lru.next;
-            delta_base_cached > delta_base_cache_limit
-            && lru != &delta_base_cache_lru;
-            lru = lru->next) {
-               struct delta_base_cache_entry *f = (void *)lru;
+       list_for_each_safe(lru, tmp, &delta_base_cache_lru) {
+               struct delta_base_cache_entry *f =
+                       list_entry(lru, struct delta_base_cache_entry, lru);
+               if (delta_base_cached <= delta_base_cache_limit)
+                       break;
                release_delta_base_cache(f);
        }
 
-       ent->p = p;
-       ent->base_offset = base_offset;
+       ent->key.p = p;
+       ent->key.base_offset = base_offset;
        ent->type = type;
        ent->data = base;
        ent->size = base_size;
-       ent->lru.next = &delta_base_cache_lru;
-       ent->lru.prev = delta_base_cache_lru.prev;
-       delta_base_cache_lru.prev->next = &ent->lru;
-       delta_base_cache_lru.prev = &ent->lru;
+       list_add_tail(&ent->lru, &delta_base_cache_lru);
+
+       if (!delta_base_cache.cmpfn)
+               hashmap_init(&delta_base_cache, delta_base_cache_hash_cmp, 0);
+       hashmap_entry_init(ent, pack_entry_hash(p, base_offset));
+       hashmap_add(&delta_base_cache, ent);
 }
 
 static void *read_object(const unsigned char *sha1, enum object_type *type,
@@ -2268,11 +2518,11 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
                struct delta_base_cache_entry *ent;
 
                ent = get_delta_base_cache_entry(p, curpos);
-               if (eq_delta_base_cache_entry(ent, p, curpos)) {
+               if (ent) {
                        type = ent->type;
                        data = ent->data;
                        size = ent->size;
-                       clear_delta_base_cache_entry(ent);
+                       detach_delta_base_cache_entry(ent);
                        base_from_cache = 1;
                        break;
                }
@@ -2351,6 +2601,7 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
        while (delta_stack_nr) {
                void *delta_data;
                void *base = data;
+               void *external_base = NULL;
                unsigned long delta_size, base_size = size;
                int i;
 
@@ -2377,6 +2628,7 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
                                      p->pack_name);
                                mark_bad_packed_object(p, base_sha1);
                                base = read_object(base_sha1, &type, &base_size);
+                               external_base = base;
                        }
                }
 
@@ -2395,6 +2647,7 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
                              "at offset %"PRIuMAX" from %s",
                              (uintmax_t)curpos, p->pack_name);
                        data = NULL;
+                       free(external_base);
                        continue;
                }
 
@@ -2414,6 +2667,7 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
                        error("failed to apply delta");
 
                free(delta_data);
+               free(external_base);
        }
 
        *final_type = type;
@@ -2447,6 +2701,17 @@ const unsigned char *nth_packed_object_sha1(struct packed_git *p,
        }
 }
 
+const struct object_id *nth_packed_object_oid(struct object_id *oid,
+                                             struct packed_git *p,
+                                             uint32_t n)
+{
+       const unsigned char *hash = nth_packed_object_sha1(p, n);
+       if (!hash)
+               return NULL;
+       hashcpy(oid->hash, hash);
+       return oid;
+}
+
 void check_pack_index_ptr(const struct packed_git *p, const void *vptr)
 {
        const unsigned char *ptr = vptr;
@@ -2653,8 +2918,9 @@ static int sha1_loose_object_info(const unsigned char *sha1,
         * object even exists.
         */
        if (!oi->typep && !oi->typename && !oi->sizep) {
+               const char *path;
                struct stat st;
-               if (stat_sha1_file(sha1, &st) < 0)
+               if (stat_sha1_file(sha1, &st, &path) < 0)
                        return -1;
                if (oi->disk_sizep)
                        *oi->disk_sizep = st.st_size;
@@ -2686,7 +2952,7 @@ static int sha1_loose_object_info(const unsigned char *sha1,
        if (status && oi->typep)
                *oi->typep = status;
        strbuf_release(&hdrbuf);
-       return 0;
+       return (status < 0) ? status : 0;
 }
 
 int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
@@ -2760,7 +3026,7 @@ int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi,
 int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
 {
        enum object_type type;
-       struct object_info oi = {NULL};
+       struct object_info oi = OBJECT_INFO_INIT;
 
        oi.typep = &type;
        oi.sizep = sizep;
@@ -2777,7 +3043,7 @@ static void *read_packed_sha1(const unsigned char *sha1,
 
        if (!find_pack_entry(sha1, &e))
                return NULL;
-       data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
+       data = cache_or_unpack_entry(e.p, e.offset, size, type);
        if (!data) {
                /*
                 * We're probably in deep shit, but let's try to fetch
@@ -2850,6 +3116,8 @@ void *read_sha1_file_extended(const unsigned char *sha1,
 {
        void *data;
        const struct packed_git *p;
+       const char *path;
+       struct stat st;
        const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
 
        errno = 0;
@@ -2865,12 +3133,9 @@ void *read_sha1_file_extended(const unsigned char *sha1,
                die("replacement %s not found for %s",
                    sha1_to_hex(repl), sha1_to_hex(sha1));
 
-       if (has_loose_object(repl)) {
-               const char *path = sha1_file_name(sha1);
-
+       if (!stat_sha1_file(repl, &st, &path))
                die("loose object %s (stored in %s) is corrupt",
                    sha1_to_hex(repl), path);
-       }
 
        if ((p = has_packed_and_bad(repl)) != NULL)
                die("packed object %s (stored in %s) is corrupt",
@@ -3216,6 +3481,8 @@ int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
 {
        struct pack_entry e;
 
+       if (!startup_info->have_repository)
+               return 0;
        if (find_pack_entry(sha1, &e))
                return 1;
        if (has_loose_object(sha1))
@@ -3493,15 +3760,15 @@ static int for_each_file_in_obj_subdir(int subdir_nr,
                strbuf_setlen(path, baselen);
                strbuf_addf(path, "/%s", de->d_name);
 
-               if (strlen(de->d_name) == 38)  {
-                       char hex[41];
-                       unsigned char sha1[20];
+               if (strlen(de->d_name) == GIT_SHA1_HEXSZ - 2)  {
+                       char hex[GIT_MAX_HEXSZ+1];
+                       struct object_id oid;
 
-                       snprintf(hex, sizeof(hex), "%02x%s",
-                                subdir_nr, de->d_name);
-                       if (!get_sha1_hex(hex, sha1)) {
+                       xsnprintf(hex, sizeof(hex), "%02x%s",
+                                 subdir_nr, de->d_name);
+                       if (!get_oid_hex(hex, &oid)) {
                                if (obj_cb) {
-                                       r = obj_cb(sha1, path->buf, data);
+                                       r = obj_cb(&oid, path->buf, data);
                                        if (r)
                                                break;
                                }
@@ -3575,8 +3842,7 @@ static int loose_from_alt_odb(struct alternate_object_database *alt,
        struct strbuf buf = STRBUF_INIT;
        int r;
 
-       /* copy base not including trailing '/' */
-       strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
+       strbuf_addstr(&buf, alt->path);
        r = for_each_loose_file_in_objdir_buf(&buf,
                                              data->cb, NULL, NULL,
                                              data->data);
@@ -3608,13 +3874,13 @@ static int for_each_object_in_pack(struct packed_git *p, each_packed_object_fn c
        int r = 0;
 
        for (i = 0; i < p->num_objects; i++) {
-               const unsigned char *sha1 = nth_packed_object_sha1(p, i);
+               struct object_id oid;
 
-               if (!sha1)
+               if (!nth_packed_object_oid(&oid, p, i))
                        return error("unable to get sha1 of object %u in %s",
                                     i, p->pack_name);
 
-               r = cb(sha1, p, i, data);
+               r = cb(&oid, p, i, data);
                if (r)
                        break;
        }
@@ -3641,3 +3907,119 @@ int for_each_packed_object(each_packed_object_fn cb, void *data, unsigned flags)
        }
        return r ? r : pack_errors;
 }
+
+static int check_stream_sha1(git_zstream *stream,
+                            const char *hdr,
+                            unsigned long size,
+                            const char *path,
+                            const unsigned char *expected_sha1)
+{
+       git_SHA_CTX c;
+       unsigned char real_sha1[GIT_MAX_RAWSZ];
+       unsigned char buf[4096];
+       unsigned long total_read;
+       int status = Z_OK;
+
+       git_SHA1_Init(&c);
+       git_SHA1_Update(&c, hdr, stream->total_out);
+
+       /*
+        * We already read some bytes into hdr, but the ones up to the NUL
+        * do not count against the object's content size.
+        */
+       total_read = stream->total_out - strlen(hdr) - 1;
+
+       /*
+        * This size comparison must be "<=" to read the final zlib packets;
+        * see the comment in unpack_sha1_rest for details.
+        */
+       while (total_read <= size &&
+              (status == Z_OK || status == Z_BUF_ERROR)) {
+               stream->next_out = buf;
+               stream->avail_out = sizeof(buf);
+               if (size - total_read < stream->avail_out)
+                       stream->avail_out = size - total_read;
+               status = git_inflate(stream, Z_FINISH);
+               git_SHA1_Update(&c, buf, stream->next_out - buf);
+               total_read += stream->next_out - buf;
+       }
+       git_inflate_end(stream);
+
+       if (status != Z_STREAM_END) {
+               error("corrupt loose object '%s'", sha1_to_hex(expected_sha1));
+               return -1;
+       }
+       if (stream->avail_in) {
+               error("garbage at end of loose object '%s'",
+                     sha1_to_hex(expected_sha1));
+               return -1;
+       }
+
+       git_SHA1_Final(real_sha1, &c);
+       if (hashcmp(expected_sha1, real_sha1)) {
+               error("sha1 mismatch for %s (expected %s)", path,
+                     sha1_to_hex(expected_sha1));
+               return -1;
+       }
+
+       return 0;
+}
+
+int read_loose_object(const char *path,
+                     const unsigned char *expected_sha1,
+                     enum object_type *type,
+                     unsigned long *size,
+                     void **contents)
+{
+       int ret = -1;
+       void *map = NULL;
+       unsigned long mapsize;
+       git_zstream stream;
+       char hdr[32];
+
+       *contents = NULL;
+
+       map = map_sha1_file_1(path, NULL, &mapsize);
+       if (!map) {
+               error_errno("unable to mmap %s", path);
+               goto out;
+       }
+
+       if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0) {
+               error("unable to unpack header of %s", path);
+               goto out;
+       }
+
+       *type = parse_sha1_header(hdr, size);
+       if (*type < 0) {
+               error("unable to parse header of %s", path);
+               git_inflate_end(&stream);
+               goto out;
+       }
+
+       if (*type == OBJ_BLOB) {
+               if (check_stream_sha1(&stream, hdr, *size, path, expected_sha1) < 0)
+                       goto out;
+       } else {
+               *contents = unpack_sha1_rest(&stream, hdr, *size, expected_sha1);
+               if (!*contents) {
+                       error("unable to unpack contents of %s", path);
+                       git_inflate_end(&stream);
+                       goto out;
+               }
+               if (check_sha1_signature(expected_sha1, *contents,
+                                        *size, typename(*type))) {
+                       error("sha1 mismatch for %s (expected %s)", path,
+                             sha1_to_hex(expected_sha1));
+                       free(*contents);
+                       goto out;
+               }
+       }
+
+       ret = 0; /* everything checks out */
+
+out:
+       if (map)
+               munmap(map, mapsize);
+       return ret;
+}