Merge branch 'rs/oidset-on-khash'
[gitweb.git] / midx.c
diff --git a/midx.c b/midx.c
index e83110ae92b4768802b6b70cb6b220ce91fc74b9..ea2f3ffe2e21700038eabf00aa975e5cc111a47f 100644 (file)
--- a/midx.c
+++ b/midx.c
@@ -1,11 +1,13 @@
 #include "cache.h"
+#include "config.h"
 #include "csum-file.h"
 #include "dir.h"
 #include "lockfile.h"
 #include "packfile.h"
 #include "object-store.h"
-#include "packfile.h"
+#include "sha1-lookup.h"
 #include "midx.h"
+#include "progress.h"
 
 #define MIDX_SIGNATURE 0x4d494458 /* "MIDX" */
 #define MIDX_VERSION 1
@@ -36,7 +38,7 @@ static char *get_midx_filename(const char *object_dir)
        return xstrfmt("%s/pack/multi-pack-index", object_dir);
 }
 
-struct multi_pack_index *load_multi_pack_index(const char *object_dir)
+struct multi_pack_index *load_multi_pack_index(const char *object_dir, int local)
 {
        struct multi_pack_index *m = NULL;
        int fd;
@@ -72,26 +74,21 @@ struct multi_pack_index *load_multi_pack_index(const char *object_dir)
        m->fd = fd;
        m->data = midx_map;
        m->data_len = midx_size;
+       m->local = local;
 
        m->signature = get_be32(m->data);
-       if (m->signature != MIDX_SIGNATURE) {
-               error(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
+       if (m->signature != MIDX_SIGNATURE)
+               die(_("multi-pack-index signature 0x%08x does not match signature 0x%08x"),
                      m->signature, MIDX_SIGNATURE);
-               goto cleanup_fail;
-       }
 
        m->version = m->data[MIDX_BYTE_FILE_VERSION];
-       if (m->version != MIDX_VERSION) {
-               error(_("multi-pack-index version %d not recognized"),
+       if (m->version != MIDX_VERSION)
+               die(_("multi-pack-index version %d not recognized"),
                      m->version);
-               goto cleanup_fail;
-       }
 
        hash_version = m->data[MIDX_BYTE_HASH_VERSION];
-       if (hash_version != MIDX_HASH_VERSION) {
-               error(_("hash version %u does not match"), hash_version);
-               goto cleanup_fail;
-       }
+       if (hash_version != MIDX_HASH_VERSION)
+               die(_("hash version %u does not match"), hash_version);
        m->hash_len = MIDX_HASH_LEN;
 
        m->num_chunks = m->data[MIDX_BYTE_NUM_CHUNKS];
@@ -104,6 +101,9 @@ struct multi_pack_index *load_multi_pack_index(const char *object_dir)
                uint64_t chunk_offset = get_be64(m->data + MIDX_HEADER_SIZE + 4 +
                                                 MIDX_CHUNKLOOKUP_WIDTH * i);
 
+               if (chunk_offset >= m->data_len)
+                       die(_("invalid chunk offset (too large)"));
+
                switch (chunk_id) {
                        case MIDX_CHUNKID_PACKNAMES:
                                m->chunk_pack_names = m->data + chunk_offset;
@@ -150,6 +150,7 @@ struct multi_pack_index *load_multi_pack_index(const char *object_dir)
        m->num_objects = ntohl(m->chunk_oid_fanout[255]);
 
        m->pack_names = xcalloc(m->num_packs, sizeof(*m->pack_names));
+       m->packs = xcalloc(m->num_packs, sizeof(*m->packs));
 
        cur_pack_name = (const char *)m->chunk_pack_names;
        for (i = 0; i < m->num_packs; i++) {
@@ -157,12 +158,10 @@ struct multi_pack_index *load_multi_pack_index(const char *object_dir)
 
                cur_pack_name += strlen(cur_pack_name) + 1;
 
-               if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0) {
-                       error(_("multi-pack-index pack names out of order: '%s' before '%s'"),
+               if (i && strcmp(m->pack_names[i], m->pack_names[i - 1]) <= 0)
+                       die(_("multi-pack-index pack names out of order: '%s' before '%s'"),
                              m->pack_names[i - 1],
                              m->pack_names[i]);
-                       goto cleanup_fail;
-               }
        }
 
        return m;
@@ -177,6 +176,181 @@ struct multi_pack_index *load_multi_pack_index(const char *object_dir)
        return NULL;
 }
 
+static void close_midx(struct multi_pack_index *m)
+{
+       uint32_t i;
+       munmap((unsigned char *)m->data, m->data_len);
+       close(m->fd);
+       m->fd = -1;
+
+       for (i = 0; i < m->num_packs; i++) {
+               if (m->packs[i]) {
+                       close_pack(m->packs[i]);
+                       free(m->packs);
+               }
+       }
+       FREE_AND_NULL(m->packs);
+       FREE_AND_NULL(m->pack_names);
+}
+
+int prepare_midx_pack(struct multi_pack_index *m, uint32_t pack_int_id)
+{
+       struct strbuf pack_name = STRBUF_INIT;
+
+       if (pack_int_id >= m->num_packs)
+               die(_("bad pack-int-id: %u (%u total packs"),
+                   pack_int_id, m->num_packs);
+
+       if (m->packs[pack_int_id])
+               return 0;
+
+       strbuf_addf(&pack_name, "%s/pack/%s", m->object_dir,
+                   m->pack_names[pack_int_id]);
+
+       m->packs[pack_int_id] = add_packed_git(pack_name.buf, pack_name.len, m->local);
+       strbuf_release(&pack_name);
+       return !m->packs[pack_int_id];
+}
+
+int bsearch_midx(const struct object_id *oid, struct multi_pack_index *m, uint32_t *result)
+{
+       return bsearch_hash(oid->hash, m->chunk_oid_fanout, m->chunk_oid_lookup,
+                           MIDX_HASH_LEN, result);
+}
+
+struct object_id *nth_midxed_object_oid(struct object_id *oid,
+                                       struct multi_pack_index *m,
+                                       uint32_t n)
+{
+       if (n >= m->num_objects)
+               return NULL;
+
+       hashcpy(oid->hash, m->chunk_oid_lookup + m->hash_len * n);
+       return oid;
+}
+
+static off_t nth_midxed_offset(struct multi_pack_index *m, uint32_t pos)
+{
+       const unsigned char *offset_data;
+       uint32_t offset32;
+
+       offset_data = m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH;
+       offset32 = get_be32(offset_data + sizeof(uint32_t));
+
+       if (m->chunk_large_offsets && offset32 & MIDX_LARGE_OFFSET_NEEDED) {
+               if (sizeof(off_t) < sizeof(uint64_t))
+                       die(_("multi-pack-index stores a 64-bit offset, but off_t is too small"));
+
+               offset32 ^= MIDX_LARGE_OFFSET_NEEDED;
+               return get_be64(m->chunk_large_offsets + sizeof(uint64_t) * offset32);
+       }
+
+       return offset32;
+}
+
+static uint32_t nth_midxed_pack_int_id(struct multi_pack_index *m, uint32_t pos)
+{
+       return get_be32(m->chunk_object_offsets + pos * MIDX_CHUNK_OFFSET_WIDTH);
+}
+
+static int nth_midxed_pack_entry(struct multi_pack_index *m, struct pack_entry *e, uint32_t pos)
+{
+       uint32_t pack_int_id;
+       struct packed_git *p;
+
+       if (pos >= m->num_objects)
+               return 0;
+
+       pack_int_id = nth_midxed_pack_int_id(m, pos);
+
+       if (prepare_midx_pack(m, pack_int_id))
+               die(_("error preparing packfile from multi-pack-index"));
+       p = m->packs[pack_int_id];
+
+       /*
+       * We are about to tell the caller where they can locate the
+       * requested object.  We better make sure the packfile is
+       * still here and can be accessed before supplying that
+       * answer, as it may have been deleted since the MIDX was
+       * loaded!
+       */
+       if (!is_pack_valid(p))
+               return 0;
+
+       if (p->num_bad_objects) {
+               uint32_t i;
+               struct object_id oid;
+               nth_midxed_object_oid(&oid, m, pos);
+               for (i = 0; i < p->num_bad_objects; i++)
+                       if (hasheq(oid.hash,
+                                  p->bad_object_sha1 + the_hash_algo->rawsz * i))
+                               return 0;
+       }
+
+       e->offset = nth_midxed_offset(m, pos);
+       e->p = p;
+
+       return 1;
+}
+
+int fill_midx_entry(const struct object_id *oid, struct pack_entry *e, struct multi_pack_index *m)
+{
+       uint32_t pos;
+
+       if (!bsearch_midx(oid, m, &pos))
+               return 0;
+
+       return nth_midxed_pack_entry(m, e, pos);
+}
+
+int midx_contains_pack(struct multi_pack_index *m, const char *idx_name)
+{
+       uint32_t first = 0, last = m->num_packs;
+
+       while (first < last) {
+               uint32_t mid = first + (last - first) / 2;
+               const char *current;
+               int cmp;
+
+               current = m->pack_names[mid];
+               cmp = strcmp(idx_name, current);
+               if (!cmp)
+                       return 1;
+               if (cmp > 0) {
+                       first = mid + 1;
+                       continue;
+               }
+               last = mid;
+       }
+
+       return 0;
+}
+
+int prepare_multi_pack_index_one(struct repository *r, const char *object_dir, int local)
+{
+       struct multi_pack_index *m;
+       struct multi_pack_index *m_search;
+       int config_value;
+
+       if (repo_config_get_bool(r, "core.multipackindex", &config_value) ||
+           !config_value)
+               return 0;
+
+       for (m_search = r->objects->multi_pack_index; m_search; m_search = m_search->next)
+               if (!strcmp(object_dir, m_search->object_dir))
+                       return 1;
+
+       m = load_multi_pack_index(object_dir, local);
+
+       if (m) {
+               m->next = r->objects->multi_pack_index;
+               r->objects->multi_pack_index = m;
+               return 1;
+       }
+
+       return 0;
+}
+
 static size_t write_midx_header(struct hashfile *f,
                                unsigned char num_chunks,
                                uint32_t num_packs)
@@ -201,6 +375,7 @@ struct pack_list {
        uint32_t alloc_list;
        uint32_t alloc_names;
        size_t pack_name_concat_len;
+       struct multi_pack_index *m;
 };
 
 static void add_pack_to_midx(const char *full_path, size_t full_path_len,
@@ -209,6 +384,9 @@ static void add_pack_to_midx(const char *full_path, size_t full_path_len,
        struct pack_list *packs = (struct pack_list *)data;
 
        if (ends_with(file_name, ".idx")) {
+               if (packs->m && midx_contains_pack(packs->m, file_name))
+                       return;
+
                ALLOC_GROW(packs->list, packs->nr + 1, packs->alloc_list);
                ALLOC_GROW(packs->names, packs->nr + 1, packs->alloc_names);
 
@@ -294,6 +472,23 @@ static int midx_oid_compare(const void *_a, const void *_b)
        return a->pack_int_id - b->pack_int_id;
 }
 
+static int nth_midxed_pack_midx_entry(struct multi_pack_index *m,
+                                     uint32_t *pack_perm,
+                                     struct pack_midx_entry *e,
+                                     uint32_t pos)
+{
+       if (pos >= m->num_objects)
+               return 1;
+
+       nth_midxed_object_oid(&e->oid, m, pos);
+       e->pack_int_id = pack_perm[nth_midxed_pack_int_id(m, pos)];
+       e->offset = nth_midxed_offset(m, pos);
+
+       /* consider objects in midx to be from "old" packs */
+       e->pack_mtime = 0;
+       return 0;
+}
+
 static void fill_pack_entry(uint32_t pack_int_id,
                            struct packed_git *p,
                            uint32_t cur_object,
@@ -319,7 +514,8 @@ static void fill_pack_entry(uint32_t pack_int_id,
  * Copy only the de-duplicated entries (selected by most-recent modified time
  * of a packfile containing the object).
  */
-static struct pack_midx_entry *get_sorted_entries(struct packed_git **p,
+static struct pack_midx_entry *get_sorted_entries(struct multi_pack_index *m,
+                                                 struct packed_git **p,
                                                  uint32_t *perm,
                                                  uint32_t nr_packs,
                                                  uint32_t *nr_objects)
@@ -328,8 +524,9 @@ static struct pack_midx_entry *get_sorted_entries(struct packed_git **p,
        uint32_t alloc_fanout, alloc_objects, total_objects = 0;
        struct pack_midx_entry *entries_by_fanout = NULL;
        struct pack_midx_entry *deduplicated_entries = NULL;
+       uint32_t start_pack = m ? m->num_packs : 0;
 
-       for (cur_pack = 0; cur_pack < nr_packs; cur_pack++)
+       for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++)
                total_objects += p[cur_pack]->num_objects;
 
        /*
@@ -346,7 +543,23 @@ static struct pack_midx_entry *get_sorted_entries(struct packed_git **p,
        for (cur_fanout = 0; cur_fanout < 256; cur_fanout++) {
                uint32_t nr_fanout = 0;
 
-               for (cur_pack = 0; cur_pack < nr_packs; cur_pack++) {
+               if (m) {
+                       uint32_t start = 0, end;
+
+                       if (cur_fanout)
+                               start = ntohl(m->chunk_oid_fanout[cur_fanout - 1]);
+                       end = ntohl(m->chunk_oid_fanout[cur_fanout]);
+
+                       for (cur_object = start; cur_object < end; cur_object++) {
+                               ALLOC_GROW(entries_by_fanout, nr_fanout + 1, alloc_fanout);
+                               nth_midxed_pack_midx_entry(m, perm,
+                                                          &entries_by_fanout[nr_fanout],
+                                                          cur_object);
+                               nr_fanout++;
+                       }
+               }
+
+               for (cur_pack = start_pack; cur_pack < nr_packs; cur_pack++) {
                        uint32_t start = 0, end;
 
                        if (cur_fanout)
@@ -367,8 +580,8 @@ static struct pack_midx_entry *get_sorted_entries(struct packed_git **p,
                 * Take only the first duplicate.
                 */
                for (cur_object = 0; cur_object < nr_fanout; cur_object++) {
-                       if (cur_object && !oidcmp(&entries_by_fanout[cur_object - 1].oid,
-                                                 &entries_by_fanout[cur_object].oid))
+                       if (cur_object && oideq(&entries_by_fanout[cur_object - 1].oid,
+                                               &entries_by_fanout[cur_object].oid))
                                continue;
 
                        ALLOC_GROW(deduplicated_entries, *nr_objects + 1, alloc_objects);
@@ -542,16 +755,34 @@ int write_midx_file(const char *object_dir)
                          midx_name);
        }
 
+       packs.m = load_multi_pack_index(object_dir, 1);
+
        packs.nr = 0;
-       packs.alloc_list = 16;
-       packs.alloc_names = 16;
+       packs.alloc_list = packs.m ? packs.m->num_packs : 16;
+       packs.alloc_names = packs.alloc_list;
        packs.list = NULL;
+       packs.names = NULL;
        packs.pack_name_concat_len = 0;
        ALLOC_ARRAY(packs.list, packs.alloc_list);
        ALLOC_ARRAY(packs.names, packs.alloc_names);
 
+       if (packs.m) {
+               for (i = 0; i < packs.m->num_packs; i++) {
+                       ALLOC_GROW(packs.list, packs.nr + 1, packs.alloc_list);
+                       ALLOC_GROW(packs.names, packs.nr + 1, packs.alloc_names);
+
+                       packs.list[packs.nr] = NULL;
+                       packs.names[packs.nr] = xstrdup(packs.m->pack_names[i]);
+                       packs.pack_name_concat_len += strlen(packs.names[packs.nr]) + 1;
+                       packs.nr++;
+               }
+       }
+
        for_each_file_in_pack_dir(object_dir, add_pack_to_midx, &packs);
 
+       if (packs.m && packs.nr == packs.m->num_packs)
+               goto cleanup;
+
        if (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT)
                packs.pack_name_concat_len += MIDX_CHUNK_ALIGNMENT -
                                              (packs.pack_name_concat_len % MIDX_CHUNK_ALIGNMENT);
@@ -559,7 +790,8 @@ int write_midx_file(const char *object_dir)
        ALLOC_ARRAY(pack_perm, packs.nr);
        sort_packs_by_name(packs.names, packs.nr, pack_perm);
 
-       entries = get_sorted_entries(packs.list, pack_perm, packs.nr, &nr_entries);
+       entries = get_sorted_entries(packs.m, packs.list, pack_perm, packs.nr, &nr_entries);
+
        for (i = 0; i < nr_entries; i++) {
                if (entries[i].offset > 0x7fffffff)
                        num_large_offsets++;
@@ -571,6 +803,9 @@ int write_midx_file(const char *object_dir)
        f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
        FREE_AND_NULL(midx_name);
 
+       if (packs.m)
+               close_midx(packs.m);
+
        cur_chunk = 0;
        num_chunks = large_offsets_needed ? 5 : 4;
 
@@ -662,6 +897,7 @@ int write_midx_file(const char *object_dir)
        finalize_hashfile(f, NULL, CSUM_FSYNC | CSUM_HASH_IN_STREAM);
        commit_lock_file(&lk);
 
+cleanup:
        for (i = 0; i < packs.nr; i++) {
                if (packs.list[i]) {
                        close_pack(packs.list[i]);
@@ -673,5 +909,99 @@ int write_midx_file(const char *object_dir)
        free(packs.list);
        free(packs.names);
        free(entries);
+       free(pack_perm);
+       free(midx_name);
        return 0;
 }
+
+void clear_midx_file(const char *object_dir)
+{
+       char *midx = get_midx_filename(object_dir);
+
+       if (remove_path(midx)) {
+               UNLEAK(midx);
+               die(_("failed to clear multi-pack-index at %s"), midx);
+       }
+
+       free(midx);
+}
+
+static int verify_midx_error;
+
+static void midx_report(const char *fmt, ...)
+{
+       va_list ap;
+       verify_midx_error = 1;
+       va_start(ap, fmt);
+       vfprintf(stderr, fmt, ap);
+       fprintf(stderr, "\n");
+       va_end(ap);
+}
+
+int verify_midx_file(const char *object_dir)
+{
+       uint32_t i;
+       struct progress *progress = NULL;
+       struct multi_pack_index *m = load_multi_pack_index(object_dir, 1);
+       verify_midx_error = 0;
+
+       if (!m)
+               return 0;
+
+       for (i = 0; i < m->num_packs; i++) {
+               if (prepare_midx_pack(m, i))
+                       midx_report("failed to load pack in position %d", i);
+       }
+
+       for (i = 0; i < 255; i++) {
+               uint32_t oid_fanout1 = ntohl(m->chunk_oid_fanout[i]);
+               uint32_t oid_fanout2 = ntohl(m->chunk_oid_fanout[i + 1]);
+
+               if (oid_fanout1 > oid_fanout2)
+                       midx_report(_("oid fanout out of order: fanout[%d] = %"PRIx32" > %"PRIx32" = fanout[%d]"),
+                                   i, oid_fanout1, oid_fanout2, i + 1);
+       }
+
+       for (i = 0; i < m->num_objects - 1; i++) {
+               struct object_id oid1, oid2;
+
+               nth_midxed_object_oid(&oid1, m, i);
+               nth_midxed_object_oid(&oid2, m, i + 1);
+
+               if (oidcmp(&oid1, &oid2) >= 0)
+                       midx_report(_("oid lookup out of order: oid[%d] = %s >= %s = oid[%d]"),
+                                   i, oid_to_hex(&oid1), oid_to_hex(&oid2), i + 1);
+       }
+
+       progress = start_progress(_("Verifying object offsets"), m->num_objects);
+       for (i = 0; i < m->num_objects; i++) {
+               struct object_id oid;
+               struct pack_entry e;
+               off_t m_offset, p_offset;
+
+               nth_midxed_object_oid(&oid, m, i);
+               if (!fill_midx_entry(&oid, &e, m)) {
+                       midx_report(_("failed to load pack entry for oid[%d] = %s"),
+                                   i, oid_to_hex(&oid));
+                       continue;
+               }
+
+               if (open_pack_index(e.p)) {
+                       midx_report(_("failed to load pack-index for packfile %s"),
+                                   e.p->pack_name);
+                       break;
+               }
+
+               m_offset = e.offset;
+               p_offset = find_pack_entry_one(oid.hash, e.p);
+
+               if (m_offset != p_offset)
+                       midx_report(_("incorrect object offset for oid[%d] = %s: %"PRIx64" != %"PRIx64),
+                                   i, oid_to_hex(&oid), m_offset, p_offset);
+
+               display_progress(progress, i + 1);
+       }
+       stop_progress(&progress);
+
+       return verify_midx_error;
+}