summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorVicent Marti <tanoku@gmail.com>2011-03-10 01:06:24 +0200
committerVicent Marti <tanoku@gmail.com>2011-03-14 23:52:25 +0200
commit58d06cf120eb9bb9247bb807bb105981bb3482a8 (patch)
tree26cab255dc526cd39b0b6f7adf7a7bfc565ad432
parent71db842fac3ba8582255bc5b61361ddef08ef105 (diff)
downloadlibgit2-58d06cf120eb9bb9247bb807bb105981bb3482a8.tar.gz
Rewrite the Pack backend
The new pack backend is an adaptation of the original git.git code in `sha1_file.c`. It's slightly faster than the previous version and severely less memory-hungry. The call-stack of a normal pack backend query has been properly documented in the top of the header for future reference. And by properly I mean with ASCII diagrams 'n shit.
-rw-r--r--src/odb_pack.c1914
-rw-r--r--src/util.h2
2 files changed, 1058 insertions, 858 deletions
diff --git a/src/odb_pack.c b/src/odb_pack.c
index 664b00139..3067179be 100644
--- a/src/odb_pack.c
+++ b/src/odb_pack.c
@@ -33,503 +33,625 @@
#include "git2/odb_backend.h"
-/** First 4 bytes of a pack-*.idx file header.
- *
- * Note this header exists only in idx v2 and later. The idx v1
- * file format does not have a magic sequence at the front, and
- * must be detected by the first four bytes *not* being this value
- * and the first 8 bytes matching the following expression:
+#define DEFAULT_WINDOW_SIZE \
+ (sizeof(void*) >= 8 \
+ ? 1 * 1024 * 1024 * 1024 \
+ : 32 * 1024 * 1024)
+
+#define DEFAULT_MAPPED_LIMIT \
+ ((1024L * 1024L) * (sizeof(void*) >= 8 ? 8192 : 256))
+
+#define PACK_SIGNATURE 0x5041434b /* "PACK" */
+#define PACK_VERSION 2
+#define pack_version_ok(v) ((v) == htonl(2) || (v) == htonl(3))
+struct pack_header {
+ uint32_t hdr_signature;
+ uint32_t hdr_version;
+ uint32_t hdr_entries;
+};
+
+/*
+ * The first four bytes of index formats later than version 1 should
+ * start with this signature, as all older git binaries would find this
+ * value illegal and abort reading the file.
*
- * uint32_t *fanout = ... the file data at offset 0 ...
- * ntohl(fanout[0]) < ntohl(fanout[1])
+ * This is the case because the number of objects in a packfile
+ * cannot exceed 1,431,660,000 as every object would need at least
+ * 3 bytes of data and the overall packfile cannot exceed 4 GiB with
+ * version 1 of the index file due to the offsets limited to 32 bits.
+ * Clearly the signature exceeds this maximum.
*
- * The value chosen here for PACK_TOC is such that the above
- * cannot be true for an idx v1 file.
+ * Very old git binaries will also compare the first 4 bytes to the
+ * next 4 bytes in the index and abort with a "non-monotonic index"
+ * error if the second 4 byte word is smaller than the first 4
+ * byte word. This would be true in the proposed future index
+ * format as idx_signature would be greater than idx_version.
*/
-#define PACK_TOC 0xff744f63 /* -1tOc */
+#define PACK_IDX_SIGNATURE 0xff744f63 /* "\377tOc" */
-/** First 4 bytes of a pack-*.pack file header. */
-#define PACK_SIG 0x5041434b /* PACK */
+struct pack_idx_header {
+ uint32_t idx_signature;
+ uint32_t idx_version;
+};
-#define GIT_PACK_NAME_MAX (5 + 40 + 1)
+struct pack_window {
+ struct pack_window *next;
+ git_map window_map;
+ off_t offset;
+ unsigned int last_used;
+ unsigned int inuse_cnt;
+};
-struct pack_backend;
+struct pack_file {
+ struct pack_window *windows;
+ off_t pack_size;
-typedef struct {
- uint32_t n;
- unsigned char *oid;
- git_off_t offset;
- git_off_t size;
-} index_entry;
+ git_map index_map;
-typedef struct { /* '.pack' file header */
- uint32_t sig; /* PACK_SIG */
- uint32_t ver; /* pack version */
- uint32_t cnt; /* object count */
-} pack_hdr;
+ uint32_t num_objects;
+ uint32_t num_bad_objects;
+ git_oid *bad_object_sha1; /* array of git_oid */
-typedef struct git_pack {
- struct pack_backend *backend;
- git_lck lock;
-
- /** Functions to access idx_map. */
- int (*idx_search)(
- uint32_t *,
- struct git_pack *,
- const git_oid *);
- int (*idx_search_offset)(
- uint32_t *,
- struct git_pack *,
- git_off_t);
- int (*idx_get)(
- index_entry *,
- struct git_pack *,
- uint32_t n);
-
- /** The .idx file, mapped into memory. */
- git_file idx_fd;
- git_map idx_map;
- uint32_t *im_fanout;
- unsigned char *im_oid;
- uint32_t *im_crc;
- uint32_t *im_offset32;
- uint32_t *im_offset64;
- uint32_t *im_off_idx;
- uint32_t *im_off_next;
-
- /** Number of objects in this pack. */
- uint32_t obj_cnt;
-
- /** File descriptor for the .pack file. */
- git_file pack_fd;
-
- /** Memory map of the pack's contents */
- git_map pack_map;
-
- /** The size of the .pack file. */
- git_off_t pack_size;
-
- /** The mtime of the .pack file. */
- time_t pack_mtime;
-
- /** Number of git_packlist we appear in. */
- unsigned int refcnt;
-
- /** Number of active users of the idx_map data. */
- unsigned int idxcnt;
- unsigned
- invalid:1 /* the pack is unable to be read by libgit2 */
- ;
-
- /** Name of the pack file(s), without extension ("pack-abc"). */
- char pack_name[GIT_PACK_NAME_MAX];
-} git_pack;
-
-typedef struct {
- size_t n_packs;
- unsigned int refcnt;
- git_pack *packs[GIT_FLEX_ARRAY];
-} git_packlist;
-
-typedef struct pack_backend {
- git_odb_backend parent;
+ int index_version;
+ time_t mtime;
+ int pack_fd;
+ unsigned pack_local:1, pack_keep:1;
+ git_oid sha1;
- git_lck lock;
- char *objects_dir;
- git_packlist *packlist;
-} pack_backend;
+ /* something like ".git/objects/pack/xxxxx.pack" */
+ char pack_name[GIT_FLEX_ARRAY]; /* more */
+};
+struct pack_entry {
+ off_t offset;
+ git_oid sha1;
+ struct pack_file *p;
+};
-typedef struct pack_location {
- git_pack *ptr;
- uint32_t n;
-} pack_location;
+struct pack__dirent {
+ struct pack_backend *backend;
+ int is_pack_local;
+};
-static int pack_stat(git_pack *p);
-static int pack_openidx(git_pack *p);
-static void pack_decidx(git_pack *p);
-static int read_pack_hdr(pack_hdr *out, git_file fd);
-static int check_pack_hdr(git_pack *p);
-static int check_pack_sha1(git_pack *p);
-static int open_pack(git_pack *p);
+struct pack_backend {
+ git_odb_backend parent;
+ git_vector packs;
+ struct pack_file *last_found;
+ size_t window_size; /* needs default value */
-static int pack_openidx_map(git_pack *p);
-static int pack_openidx_v1(git_pack *p);
-static int pack_openidx_v2(git_pack *p);
+ size_t mapped_limit; /* needs default value */
+ size_t peak_mapped;
+ size_t mapped;
+ size_t used_ctr;
-GIT_INLINE(uint32_t) decode32(void *b)
-{
- return ntohl(*((uint32_t *)b));
-}
+ unsigned int peak_open_windows;
+ unsigned int open_windows;
-GIT_INLINE(uint64_t) decode64(void *b)
-{
- uint32_t *p = b;
- return (((uint64_t)ntohl(p[0])) << 32) | ntohl(p[1]);
-}
+ unsigned int mmap_calls;
+};
+
+/**
+ * The wonderful tale of a Packed Object lookup query
+ * ===================================================
+ * A riveting and epic story of epicness and ASCII
+ * art, presented by yours truly,
+ * Sir Vicent of Marti
+ *
+ *
+ * Chapter 1: Once upon a time...
+ * Initialization of the Pack Backend
+ * --------------------------------------------------
+ *
+ * # git_odb_backend_pack
+ * | Creates the pack backend structure, initializes the
+ * | callback pointers to our default read() and exist() methods,
+ * | and tries to preload all the known packfiles in the ODB.
+ * |
+ * |-# packfile_load_all
+ * | Tries to find the `pack` folder, if it exists. ODBs without
+ * | a pack folder are ignored altogether. If there's a `pack` folder
+ * | we run a `dirent` callback through every file in the pack folder
+ * | to find our packfiles. The packfiles are then sorted according
+ * | to a sorting callback.
+ * |
+ * |-# packfile_load__cb
+ * | | This callback is called from `dirent` with every single file
+ * | | inside the pack folder. We find the packs by actually locating
+ * | | their index (ends in ".idx"). From that index, we verify that
+ * | | the corresponding packfile exists and is valid, and if so, we
+ * | | add it to the pack list.
+ * | |
+ * | |-# packfile_check
+ * | Make sure that there's a packfile to back this index, and store
+ * | some very basic information regarding the packfile itself,
+ * | such as the full path, the size, and the modification time.
+ * | We don't actually open the packfile to check for internal consistency.
+ * |
+ * |-# packfile_sort__cb
+ * Sort all the preloaded packs according to some specific criteria:
+ * we prioritize the "newer" packs because it's more likely they
+ * contain the objects we are looking for, and we prioritize local
+ * packs over remote ones.
+ *
+ *
+ *
+ * Chapter 2: To be, or not to be...
+ * A standard packed `exist` query for an OID
+ * --------------------------------------------------
+ *
+ * # pack_backend__exists
+ * | Check if the given SHA1 oid exists in any of the packs
+ * | that have been loaded for our ODB.
+ * |
+ * |-# pack_entry_find
+ * | Iterate through all the packs that have been preloaded
+ * | (starting by the pack where the latest object was found)
+ * | to try to find the OID in one of them.
+ * |
+ * |-# pack_entry_find1
+ * | Check the index of an individual pack to see if the SHA1
+ * | OID can be found. If we can find the offset to that SHA1
+ * | inside of the index, that means the object is contained
+ * | inside of the packfile and we can stop searching.
+ * | Before returning, we verify that the packfile behing the
+ * | index we are searching still exists on disk.
+ * |
+ * |-# pack_entry_find_offset
+ * | | Mmap the actual index file to disk if it hasn't been opened
+ * | | yet, and run a binary search through it to find the OID.
+ * | | See <http://book.git-scm.com/7_the_packfile.html> for specifics
+ * | | on the Packfile Index format and how do we find entries in it.
+ * | |
+ * | |-# pack_index_open
+ * | | Guess the name of the index based on the full path to the
+ * | | packfile, open it and verify its contents. Only if the index
+ * | | has not been opened already.
+ * | |
+ * | |-# pack_index_check
+ * | Mmap the index file and do a quick run through the header
+ * | to guess the index version (right now we support v1 and v2),
+ * | and to verify that the size of the index makes sense.
+ * |
+ * |-# packfile_open
+ * See `packfile_open` in Chapter 3
+ *
+ *
+ *
+ * Chapter 3: The neverending story...
+ * A standard packed `lookup` query for an OID
+ * --------------------------------------------------
+ * TODO
+ *
+ */
+
/***********************************************************
*
- * PACKFILE FUNCTIONS
- *
- * Locate, open and access the contents of a packfile
+ * FORWARD DECLARATIONS
*
***********************************************************/
-static int pack_stat(git_pack *p)
-{
- char pb[GIT_PATH_MAX];
- struct stat sb;
+static void pack_window_free_all(struct pack_backend *backend, struct pack_file *p);
+static int pack_window_contains(struct pack_window *win, off_t offset);
- if (git__fmt(pb, sizeof(pb), "%s/pack/%s.pack",
- p->backend->objects_dir,
- p->pack_name) < 0)
- return GIT_ERROR;
+static void pack_window_scan_lru(struct pack_file *p, struct pack_file **lru_p,
+ struct pack_window **lru_w, struct pack_window **lru_l);
- if (gitfo_stat(pb, &sb) || !S_ISREG(sb.st_mode))
- return GIT_ERROR;
+static int pack_window_close_lru( struct pack_backend *backend,
+ struct pack_file *current, git_file keep_fd);
- if (sb.st_size < (3 * 4 + GIT_OID_RAWSZ))
- return GIT_ERROR;
+static void pack_window_close(struct pack_window **w_cursor);
- p->pack_size = sb.st_size;
- p->pack_mtime = sb.st_mtime;
+static unsigned char *pack_window_open( struct pack_backend *backend,
+ struct pack_file *p, struct pack_window **w_cursor, off_t offset,
+ unsigned int *left);
- return GIT_SUCCESS;
-}
+static int packfile_sort__cb(const void *a_, const void *b_);
-static int pack_openidx(git_pack *p)
-{
- gitlck_lock(&p->lock);
+static void pack_index_free(struct pack_file *p);
- if (p->invalid) {
- gitlck_unlock(&p->lock);
- return GIT_ERROR;
- }
+static int pack_index_check(const char *path, struct pack_file *p);
+static int pack_index_open(struct pack_file *p);
- if (++p->idxcnt == 1 && !p->idx_search) {
- int status, version;
- uint32_t *data;
+static struct pack_file *packfile_alloc(int extra);
+static int packfile_open(struct pack_file *p);
+static int packfile_check(struct pack_file **pack_out, const char *path, int local);
+static int packfile_load__cb(void *_data, char *path);
+static int packfile_load_all(struct pack_backend *backend, const char *odb_path, int local);
- if (pack_stat(p) || pack_openidx_map(p)) {
- p->invalid = 1;
- p->idxcnt--;
- gitlck_unlock(&p->lock);
- return GIT_ERROR;
- }
- data = p->idx_map.data;
- status = GIT_SUCCESS;
- version = 1;
+static off_t nth_packed_object_offset(const struct pack_file *p, uint32_t n);
- if (decode32(&data[0]) == PACK_TOC)
- version = decode32(&data[1]);
-
- switch (version) {
- case 1:
- status = pack_openidx_v1(p);
- break;
- case 2:
- status = pack_openidx_v2(p);
- break;
- default:
- status = GIT_ERROR;
- }
+static int pack_entry_find_offset(off_t *offset_out,
+ struct pack_file *p, const git_oid *oid);
- if (status != GIT_SUCCESS) {
- gitfo_free_map(&p->idx_map);
- p->invalid = 1;
- p->idxcnt--;
- gitlck_unlock(&p->lock);
- return status;
- }
- }
+static int pack_entry_find1(struct pack_entry *e,
+ struct pack_file *p, const git_oid *oid);
- gitlck_unlock(&p->lock);
- return GIT_SUCCESS;
-}
+static int pack_entry_find(struct pack_entry *e,
+ struct pack_backend *backend, const git_oid *oid);
-static void pack_decidx(git_pack *p)
-{
- gitlck_lock(&p->lock);
- p->idxcnt--;
- gitlck_unlock(&p->lock);
-}
+static off_t get_delta_base(struct pack_backend *backend,
+ struct pack_file *p, struct pack_window **w_curs,
+ off_t *curpos, git_otype type,
+ off_t delta_obj_offset);
-static int read_pack_hdr(pack_hdr *out, git_file fd)
-{
- pack_hdr hdr;
+static unsigned long packfile_unpack_header1(
+ size_t *sizep,
+ git_otype *type,
+ const unsigned char *buf,
+ unsigned long len);
- if (gitfo_read(fd, &hdr, sizeof(hdr)))
- return GIT_ERROR;
+static int packfile_unpack_header(
+ size_t *size_p,
+ git_otype *type_p,
+ struct pack_backend *backend,
+ struct pack_file *p,
+ struct pack_window **w_curs,
+ off_t *curpos);
- out->sig = decode32(&hdr.sig);
- out->ver = decode32(&hdr.ver);
- out->cnt = decode32(&hdr.cnt);
+static int packfile_unpack_compressed(
+ git_rawobj *obj,
+ struct pack_backend *backend,
+ struct pack_file *p,
+ struct pack_window **w_curs,
+ off_t curpos,
+ size_t size,
+ git_otype type);
- return GIT_SUCCESS;
-}
+static int packfile_unpack_delta(
+ git_rawobj *obj,
+ struct pack_backend *backend,
+ struct pack_file *p,
+ struct pack_window **w_curs,
+ off_t curpos,
+ size_t delta_size,
+ git_otype delta_type,
+ off_t obj_offset);
+
+static int packfile_unpack(git_rawobj *obj, struct pack_backend *backend,
+ struct pack_file *p, off_t obj_offset);
+
+
+
+
+
+/***********************************************************
+ *
+ * PACK WINDOW MANAGEMENT
+ *
+ ***********************************************************/
-static int check_pack_hdr(git_pack *p)
+void pack_window_free_all(struct pack_backend *backend, struct pack_file *p)
{
- pack_hdr hdr;
+ while (p->windows) {
+ struct pack_window *w = p->windows;
+ assert(w->inuse_cnt == 0);
- if (read_pack_hdr(&hdr, p->pack_fd))
- return GIT_ERROR;
+ backend->mapped -= w->window_map.len;
+ backend->open_windows--;
- if (hdr.sig != PACK_SIG
- || (hdr.ver != 2 && hdr.ver != 3)
- || hdr.cnt != p->obj_cnt)
- return GIT_ERROR;
+ gitfo_free_map(&w->window_map);
- return GIT_SUCCESS;
+ p->windows = w->next;
+ free(w);
+ }
}
-static int check_pack_sha1(git_pack *p)
+GIT_INLINE(int) pack_window_contains(struct pack_window *win, off_t offset)
{
- unsigned char *data = p->idx_map.data;
- git_off_t pack_sha1_off = p->pack_size - GIT_OID_RAWSZ;
- size_t idx_pack_sha1_off = p->idx_map.len - 2 * GIT_OID_RAWSZ;
- git_oid pack_id, idx_pack_id;
+ /* We must promise at least 20 bytes (one hash) after the
+ * offset is available from this window, otherwise the offset
+ * is not actually in this window and a different window (which
+ * has that one hash excess) must be used. This is to support
+ * the object header and delta base parsing routines below.
+ */
+ off_t win_off = win->offset;
+ return win_off <= offset
+ && (offset + 20) <= (off_t)(win_off + win->window_map.len);
+}
- if (gitfo_lseek(p->pack_fd, pack_sha1_off, SEEK_SET) == -1)
- return GIT_ERROR;
+static void pack_window_scan_lru(
+ struct pack_file *p,
+ struct pack_file **lru_p,
+ struct pack_window **lru_w,
+ struct pack_window **lru_l)
+{
+ struct pack_window *w, *w_l;
+
+ for (w_l = NULL, w = p->windows; w; w = w->next) {
+ if (!w->inuse_cnt) {
+ if (!*lru_w || w->last_used < (*lru_w)->last_used) {
+ *lru_p = p;
+ *lru_w = w;
+ *lru_l = w_l;
+ }
+ }
+ w_l = w;
+ }
+}
- if (gitfo_read(p->pack_fd, pack_id.id, sizeof(pack_id.id)))
- return GIT_ERROR;
+static int pack_window_close_lru(
+ struct pack_backend *backend,
+ struct pack_file *current,
+ git_file keep_fd)
+{
+ struct pack_file *lru_p = NULL;
+ struct pack_window *lru_w = NULL, *lru_l = NULL;
+ size_t i;
+
+ if (current)
+ pack_window_scan_lru(current, &lru_p, &lru_w, &lru_l);
+
+ for (i = 0; i < backend->packs.length; ++i)
+ pack_window_scan_lru(git_vector_get(&backend->packs, i), &lru_p, &lru_w, &lru_l);
+
+ if (lru_p) {
+ backend->mapped -= lru_w->window_map.len;
+ gitfo_free_map(&lru_w->window_map);
+
+ if (lru_l)
+ lru_l->next = lru_w->next;
+ else {
+ lru_p->windows = lru_w->next;
+ if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
+ gitfo_close(lru_p->pack_fd);
+ lru_p->pack_fd = -1;
+ }
+ }
- git_oid_mkraw(&idx_pack_id, data + idx_pack_sha1_off);
+ free(lru_w);
+ backend->open_windows--;
+ return GIT_SUCCESS;
+ }
- if (git_oid_cmp(&pack_id, &idx_pack_id))
- return GIT_ERROR;
+ return GIT_ERROR;
+}
- return GIT_SUCCESS;
+static void pack_window_close(struct pack_window **w_cursor)
+{
+ struct pack_window *w = *w_cursor;
+ if (w) {
+ w->inuse_cnt--;
+ *w_cursor = NULL;
+ }
}
-static int open_pack(git_pack *p)
+static unsigned char *pack_window_open(
+ struct pack_backend *backend,
+ struct pack_file *p,
+ struct pack_window **w_cursor,
+ off_t offset,
+ unsigned int *left)
{
- char pb[GIT_PATH_MAX];
- struct stat sb;
+ struct pack_window *win = *w_cursor;
- if (p->pack_fd != -1)
- return GIT_SUCCESS;
+ if (p->pack_fd == -1 && packfile_open(p) < GIT_SUCCESS)
+ return NULL;
- if (git__fmt(pb, sizeof(pb), "%s/pack/%s.pack",
- p->backend->objects_dir,
- p->pack_name) < 0)
- return GIT_ERROR;
+ /* Since packfiles end in a hash of their content and it's
+ * pointless to ask for an offset into the middle of that
+ * hash, and the pack_window_contains function above wouldn't match
+ * don't allow an offset too close to the end of the file.
+ */
+ if (offset > (p->pack_size - 20))
+ return NULL;
- if (pack_openidx(p))
- return GIT_ERROR;
+ if (!win || !pack_window_contains(win, offset)) {
- if ((p->pack_fd = gitfo_open(pb, O_RDONLY)) < 0)
- goto error_cleanup;
+ if (win)
+ win->inuse_cnt--;
- if (gitfo_fstat(p->pack_fd, &sb)
- || !S_ISREG(sb.st_mode) || p->pack_size != sb.st_size
- || check_pack_hdr(p) || check_pack_sha1(p))
- goto error_cleanup;
+ for (win = p->windows; win; win = win->next) {
+ if (pack_window_contains(win, offset))
+ break;
+ }
- if (!git__is_sizet(p->pack_size) ||
- gitfo_map_ro(&p->pack_map, p->pack_fd, 0, (size_t)p->pack_size) < 0)
- goto error_cleanup;
+ if (!win) {
+ size_t window_align = backend->window_size / 2;
+ size_t len;
- pack_decidx(p);
- return GIT_SUCCESS;
+ win = git__calloc(1, sizeof(*win));
+ win->offset = (offset / window_align) * window_align;
-error_cleanup:
- gitfo_close(p->pack_fd);
- p->pack_fd = -1;
- pack_decidx(p);
- return GIT_ERROR;
-}
+ len = (size_t)(p->pack_size - win->offset);
+ if (len > backend->window_size)
+ len = backend->window_size;
-static void pack_dec(git_pack *p)
-{
- int need_free;
-
- gitlck_lock(&p->lock);
- need_free = !--p->refcnt;
- gitlck_unlock(&p->lock);
-
- if (need_free) {
- if (p->idx_search) {
- gitfo_free_map(&p->idx_map);
- gitfo_close(p->idx_fd);
- free(p->im_fanout);
- free(p->im_off_idx);
- free(p->im_off_next);
- if (p->pack_fd != -1) {
- gitfo_close(p->pack_fd);
- gitfo_free_map(&p->pack_map);
- }
+ backend->mapped += len;
+
+ while (backend->mapped_limit < backend->mapped &&
+ pack_window_close_lru(backend, p, p->pack_fd) == GIT_SUCCESS) {}
+
+ if (gitfo_map_ro(&win->window_map, p->pack_fd,
+ win->offset, len) < GIT_SUCCESS)
+ return NULL;
+
+ backend->mmap_calls++;
+ backend->open_windows++;
+
+ if (backend->mapped > backend->peak_mapped)
+ backend->peak_mapped = backend->mapped;
+
+ if (backend->open_windows > backend->peak_open_windows)
+ backend->peak_open_windows = backend->open_windows;
+
+ win->next = p->windows;
+ p->windows = win;
}
+ }
- gitlck_free(&p->lock);
- free(p);
+ if (win != *w_cursor) {
+ win->last_used = backend->used_ctr++;
+ win->inuse_cnt++;
+ *w_cursor = win;
}
+
+ offset -= win->offset;
+ assert(git__is_sizet(offset));
+
+ if (left)
+ *left = win->window_map.len - (size_t)offset;
+
+ return (unsigned char *)win->window_map.data + offset;
}
-static void packlist_dec(pack_backend *backend, git_packlist *pl)
-{
- int need_free;
- assert(backend && pl);
- gitlck_lock(&backend->lock);
- need_free = !--pl->refcnt;
- gitlck_unlock(&backend->lock);
- if (need_free) {
- size_t j;
- for (j = 0; j < pl->n_packs; j++)
- pack_dec(pl->packs[j]);
- free(pl);
+
+
+
+/***********************************************************
+ *
+ * PACK INDEX METHODS
+ *
+ ***********************************************************/
+
+static void pack_index_free(struct pack_file *p)
+{
+ if (p->index_map.data) {
+ gitfo_free_map(&p->index_map);
+ p->index_map.data = NULL;
}
}
-static git_pack *alloc_pack(const char *pack_name)
+static int pack_index_check(const char *path, struct pack_file *p)
{
- git_pack *p = git__calloc(1, sizeof(*p));
- if (!p)
- return NULL;
+ struct pack_idx_header *hdr;
+ uint32_t version, nr, i, *index;
- gitlck_init(&p->lock);
- strcpy(p->pack_name, pack_name);
- p->refcnt = 1;
- p->pack_fd = -1;
- return p;
-}
+ void *idx_map;
+ size_t idx_size;
-struct scanned_pack {
- struct scanned_pack *next;
- git_pack *pack;
-};
+ struct stat st;
-static int scan_one_pack(void *state, char *name)
-{
- struct scanned_pack **ret = state, *r;
- char *s = strrchr(name, '/'), *d;
+ /* TODO: properly open the file without access time */
+ git_file fd = gitfo_open(path, O_RDONLY /*| O_NOATIME */);
- if (git__prefixcmp(s + 1, "pack-")
- || git__suffixcmp(s, ".pack")
- || strlen(s + 1) != GIT_PACK_NAME_MAX + 4)
- return 0;
-
- d = strrchr(s + 1, '.');
- strcpy(d + 1, "idx"); /* "pack-abc.pack" -> "pack-abc.idx" */
- if (gitfo_exists(name))
- return 0;
+ int error;
- if ((r = git__malloc(sizeof(*r))) == NULL)
- return GIT_ERROR;
+ if (fd < 0)
+ return GIT_EOSERR;
- *d = '\0'; /* "pack-abc.pack" -_> "pack-abc" */
- if ((r->pack = alloc_pack(s + 1)) == NULL) {
- free(r);
- return GIT_ERROR;
+ if (gitfo_fstat(fd, &st) < GIT_SUCCESS) {
+ gitfo_close(fd);
+ return GIT_EOSERR;
}
- r->next = *ret;
- *ret = r;
- return 0;
-}
+ if (!git__is_sizet(st.st_size))
+ return GIT_ENOMEM;
-static git_packlist *scan_packs(pack_backend *backend)
-{
- char pb[GIT_PATH_MAX];
- struct scanned_pack *state = NULL, *c;
- size_t cnt;
- git_packlist *new_list;
+ idx_size = (size_t)st.st_size;
- if (git__fmt(pb, sizeof(pb), "%s/pack", backend->objects_dir) < 0)
- return NULL;
- gitfo_dirent(pb, sizeof(pb), scan_one_pack, &state);
-
- /* TODO - merge old entries into the new array */
- for (cnt = 0, c = state; c; c = c->next)
- cnt++;
- new_list = git__malloc(sizeof(*new_list)
- + (sizeof(new_list->packs[0]) * cnt));
- if (!new_list)
- goto fail;
-
- for (cnt = 0, c = state; c; ) {
- struct scanned_pack *n = c->next;
- c->pack->backend = backend;
- new_list->packs[cnt++] = c->pack;
- free(c);
- c = n;
- }
- new_list->n_packs = cnt;
- new_list->refcnt = 2;
- backend->packlist = new_list;
- return new_list;
-
-fail:
- while (state) {
- struct scanned_pack *n = state->next;
- pack_dec(state->pack);
- free(state);
- state = n;
+ if (idx_size < 4 * 256 + 20 + 20) {
+ gitfo_close(fd);
+ return GIT_EOBJCORRUPTED;
}
- return NULL;
-}
-static git_packlist *packlist_get(pack_backend *backend)
-{
- git_packlist *pl;
-
- gitlck_lock(&backend->lock);
- if ((pl = backend->packlist) != NULL)
- pl->refcnt++;
- else
- pl = scan_packs(backend);
- gitlck_unlock(&backend->lock);
- return pl;
-}
+ error = gitfo_map_ro(&p->index_map, fd, 0, idx_size);
+ gitfo_close(fd);
-static int locate_packfile(pack_location *location, pack_backend *backend, const git_oid *id)
-{
- git_packlist *pl = packlist_get(backend);
- size_t j;
+ if (error < GIT_SUCCESS)
+ return error;
- if (!pl)
- return GIT_ENOTFOUND;
+ hdr = idx_map = p->index_map.data;
- for (j = 0; j < pl->n_packs; j++) {
+ if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
+ version = ntohl(hdr->idx_version);
- git_pack *pack = pl->packs[j];
- uint32_t pos;
- int res;
+ if (version < 2 || version > 2) {
+ gitfo_free_map(&p->index_map);
+ return GIT_EOBJCORRUPTED; /* unsupported index version */
+ }
- if (pack_openidx(pack))
- continue;
+ } else
+ version = 1;
- res = pack->idx_search(&pos, pack, id);
- pack_decidx(pack);
+ nr = 0;
+ index = idx_map;
- if (!res) {
- packlist_dec(backend, pl);
+ if (version > 1)
+ index += 2; /* skip index header */
- location->ptr = pack;
- location->n = pos;
+ for (i = 0; i < 256; i++) {
+ uint32_t n = ntohl(index[i]);
+ if (n < nr) {
+ gitfo_free_map(&p->index_map);
+ return GIT_EOBJCORRUPTED; /* non-monotonic index */
+ }
+ nr = n;
+ }
- return GIT_SUCCESS;
+ if (version == 1) {
+ /*
+ * Total size:
+ * - 256 index entries 4 bytes each
+ * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
+ * - 20-byte SHA1 of the packfile
+ * - 20-byte SHA1 file checksum
+ */
+ if (idx_size != 4*256 + nr * 24 + 20 + 20) {
+ gitfo_free_map(&p->index_map);
+ return GIT_EOBJCORRUPTED;
+ }
+ } else if (version == 2) {
+ /*
+ * Minimum size:
+ * - 8 bytes of header
+ * - 256 index entries 4 bytes each
+ * - 20-byte sha1 entry * nr
+ * - 4-byte crc entry * nr
+ * - 4-byte offset entry * nr
+ * - 20-byte SHA1 of the packfile
+ * - 20-byte SHA1 file checksum
+ * And after the 4-byte offset table might be a
+ * variable sized table containing 8-byte entries
+ * for offsets larger than 2^31.
+ */
+ unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
+ unsigned long max_size = min_size;
+
+ if (nr)
+ max_size += (nr - 1)*8;
+
+ if (idx_size < min_size || idx_size > max_size) {
+ gitfo_free_map(&p->index_map);
+ return GIT_EOBJCORRUPTED;
}
+ /* Make sure that off_t is big enough to access the whole pack...
+ * Is this an issue in libgit2? It shouldn't. */
+ if (idx_size != min_size && (sizeof(off_t) <= 4)) {
+ gitfo_free_map(&p->index_map);
+ return GIT_EOSERR;
+ }
}
- packlist_dec(backend, pl);
- return GIT_ENOTFOUND;
+ p->index_version = version;
+ p->num_objects = nr;
+ return GIT_SUCCESS;
}
+static int pack_index_open(struct pack_file *p)
+{
+ char *idx_name;
+ int error;
+
+ if (p->index_map.data)
+ return GIT_SUCCESS;
+ idx_name = git__strdup(p->pack_name);
+ strcpy(idx_name + strlen(idx_name) - STRLEN(".pack"), ".idx");
+ error = pack_index_check(idx_name, p);
+ free(idx_name);
+ return error;
+}
@@ -541,350 +663,225 @@ static int locate_packfile(pack_location *location, pack_backend *backend, const
/***********************************************************
*
- * PACKFILE INDEX FUNCTIONS
- *
- * Get index formation for packfile indexes v1 and v2
+ * PACKFILE METHODS
*
***********************************************************/
-static int pack_openidx_map(git_pack *p)
+static int packfile_sort__cb(const void *a_, const void *b_)
{
- char pb[GIT_PATH_MAX];
- git_off_t len;
-
- if (git__fmt(pb, sizeof(pb), "%s/pack/%s.idx",
- p->backend->objects_dir,
- p->pack_name) < 0)
- return GIT_ERROR;
-
- if ((p->idx_fd = gitfo_open(pb, O_RDONLY)) < 0)
- return GIT_ERROR;
-
- if ((len = gitfo_size(p->idx_fd)) < 0
- || !git__is_sizet(len)
- || gitfo_map_ro(&p->idx_map, p->idx_fd, 0, (size_t)len)) {
- gitfo_close(p->idx_fd);
- return GIT_ERROR;
- }
+ struct pack_file *a = *((struct pack_file **)a_);
+ struct pack_file *b = *((struct pack_file **)b_);
+ int st;
- return GIT_SUCCESS;
+ /*
+ * Local packs tend to contain objects specific to our
+ * variant of the project than remote ones. In addition,
+ * remote ones could be on a network mounted filesystem.
+ * Favor local ones for these reasons.
+ */
+ st = a->pack_local - b->pack_local;
+ if (st)
+ return -st;
+
+ /*
+ * Younger packs tend to contain more recent objects,
+ * and more recent objects tend to get accessed more
+ * often.
+ */
+ if (a->mtime < b->mtime)
+ return 1;
+ else if (a->mtime == b->mtime)
+ return 0;
+
+ return -1;
}
-typedef struct {
- git_off_t offset;
- uint32_t n;
-} offset_idx_info;
+static struct pack_file *packfile_alloc(int extra)
+{
+ struct pack_file *p = git__malloc(sizeof(*p) + extra);
+ memset(p, 0, sizeof(*p));
+ p->pack_fd = -1;
+ return p;
+}
-static int cmp_offset_idx_info(const void *lhs, const void *rhs)
+
+static void packfile_free(struct pack_backend *backend, struct pack_file *p)
{
- const offset_idx_info *a = lhs;
- const offset_idx_info *b = rhs;
- return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
+ assert(p);
+
+ /* clear_delta_base_cache(); */
+ pack_window_free_all(backend, p);
+
+ if (p->pack_fd != -1)
+ gitfo_close(p->pack_fd);
+
+ pack_index_free(p);
+
+ free(p->bad_object_sha1);
+ free(p);
}
-static int make_offset_index(git_pack *p, offset_idx_info *data)
+static int packfile_open(struct pack_file *p)
{
- git_off_t min_off = 3 * 4, max_off = p->pack_size - GIT_OID_RAWSZ;
- uint32_t *idx, *next;
- uint32_t j;
+ struct stat st;
+ struct pack_header hdr;
+ git_oid sha1;
+ unsigned char *idx_sha1;
- qsort(data, p->obj_cnt, sizeof(*data), cmp_offset_idx_info);
+ if (!p->index_map.data && pack_index_open(p) < GIT_SUCCESS)
+ return GIT_ENOTFOUND;
- if (data[0].offset < min_off || data[p->obj_cnt].offset > max_off)
- return GIT_ERROR;
+ /* TODO: open with noatime */
+ p->pack_fd = gitfo_open(p->pack_name, O_RDONLY);
+ if (p->pack_fd < 0 || gitfo_fstat(p->pack_fd, &st) < GIT_SUCCESS)
+ return GIT_EOSERR;
+
+ /* If we created the struct before we had the pack we lack size. */
+ if (!p->pack_size) {
+ if (!S_ISREG(st.st_mode))
+ goto cleanup;
+ p->pack_size = (off_t)st.st_size;
+ } else if (p->pack_size != st.st_size)
+ goto cleanup;
- if ((idx = git__malloc(sizeof(*idx) * (p->obj_cnt+1))) == NULL)
- return GIT_ERROR;
- if ((next = git__malloc(sizeof(*next) * p->obj_cnt)) == NULL) {
- free(idx);
- return GIT_ERROR;
- }
+#if 0
+ /* We leave these file descriptors open with sliding mmap;
+ * there is no point keeping them open across exec(), though.
+ */
+ fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
+ if (fd_flag < 0)
+ return error("cannot determine file descriptor flags");
- for (j = 0; j < p->obj_cnt+1; j++)
- idx[j] = data[j].n;
+ fd_flag |= FD_CLOEXEC;
+ if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
+ return GIT_EOSERR;
+#endif
- for (j = 0; j < p->obj_cnt; j++) {
- assert(idx[j] < p->obj_cnt);
- assert(idx[j+1] < p->obj_cnt+1);
+ /* Verify we recognize this pack file format. */
+ if (gitfo_read(p->pack_fd, &hdr, sizeof(hdr)) < GIT_SUCCESS)
+ goto cleanup;
- next[idx[j]] = idx[j+1];
- }
+ if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
+ goto cleanup;
- p->im_off_idx = idx;
- p->im_off_next = next;
- return GIT_SUCCESS;
-}
+ if (!pack_version_ok(hdr.hdr_version))
+ goto cleanup;
-static int idxv1_search(uint32_t *out, git_pack *p, const git_oid *id)
-{
- unsigned char *data = p->im_oid;
- uint32_t lo = id->id[0] ? p->im_fanout[id->id[0] - 1] : 0;
- uint32_t hi = p->im_fanout[id->id[0]];
+ /* Verify the pack matches its index. */
+ if (p->num_objects != ntohl(hdr.hdr_entries))
+ goto cleanup;
- do {
- uint32_t mid = (lo + hi) >> 1;
- uint32_t pos = 24 * mid;
- int cmp = memcmp(id->id, data + pos + 4, 20);
- if (cmp < 0)
- hi = mid;
- else if (!cmp) {
- *out = mid;
- return GIT_SUCCESS;
- } else
- lo = mid + 1;
- } while (lo < hi);
- return GIT_ENOTFOUND;
-}
+ if (gitfo_lseek(p->pack_fd, p->pack_size - GIT_OID_RAWSZ, SEEK_SET) == -1)
+ goto cleanup;
-static int idxv1_search_offset(uint32_t *out, git_pack *p, git_off_t offset)
-{
- if (offset > 0 && offset < (p->pack_size - GIT_OID_RAWSZ)) {
- uint32_t lo = 0, hi = p->obj_cnt+1;
- unsigned char *data = p->im_oid;
- uint32_t *idx = p->im_off_idx;
- do {
- uint32_t mid = (lo + hi) >> 1;
- uint32_t n = idx[mid];
- uint32_t pos = n * (GIT_OID_RAWSZ + 4);
- git_off_t here = decode32(data + pos);
- if (offset < here)
- hi = mid;
- else if (offset == here) {
- *out = n;
- return GIT_SUCCESS;
- } else
- lo = mid + 1;
- } while (lo < hi);
- }
- return GIT_ENOTFOUND;
-}
+ if (gitfo_read(p->pack_fd, sha1.id, GIT_OID_RAWSZ) < GIT_SUCCESS)
+ goto cleanup;
-static int idxv1_get(index_entry *e, git_pack *p, uint32_t n)
-{
- unsigned char *data = p->im_oid;
- uint32_t *next = p->im_off_next;
-
- if (n < p->obj_cnt) {
- uint32_t pos = n * (GIT_OID_RAWSZ + 4);
- git_off_t next_off = p->pack_size - GIT_OID_RAWSZ;
- e->n = n;
- e->oid = data + pos + 4;
- e->offset = decode32(data + pos);
- if (next[n] < p->obj_cnt) {
- pos = next[n] * (GIT_OID_RAWSZ + 4);
- next_off = decode32(data + pos);
- }
- e->size = next_off - e->offset;
- return GIT_SUCCESS;
- }
- return GIT_ENOTFOUND;
+ idx_sha1 = ((unsigned char *)p->index_map.data) + p->index_map.len - 40;
+
+ if (git_oid_cmp(&sha1, (git_oid *)idx_sha1) != 0)
+ goto cleanup;
+
+ return GIT_SUCCESS;
+
+cleanup:
+ gitfo_close(p->pack_fd);
+ p->pack_fd = -1;
+ return GIT_EPACKCORRUPTED;
}
-static int pack_openidx_v1(git_pack *p)
+static int packfile_check(struct pack_file **pack_out, const char *path, int local)
{
- uint32_t *src_fanout = p->idx_map.data;
- uint32_t *im_fanout;
- offset_idx_info *info;
- size_t expsz;
- uint32_t j;
-
-
- if ((im_fanout = git__malloc(sizeof(*im_fanout) * 256)) == NULL)
- return GIT_ERROR;
-
- im_fanout[0] = decode32(&src_fanout[0]);
- for (j = 1; j < 256; j++) {
- im_fanout[j] = decode32(&src_fanout[j]);
- if (im_fanout[j] < im_fanout[j - 1]) {
- free(im_fanout);
- return GIT_ERROR;
- }
- }
- p->obj_cnt = im_fanout[255];
+ struct stat st;
+ struct pack_file *p;
+ size_t path_len;
+
+ *pack_out = NULL;
+ path_len = strlen(path);
+ p = packfile_alloc(path_len + 2);
- expsz = 4 * 256 + 24 * p->obj_cnt + 2 * 20;
- if (expsz != p->idx_map.len) {
- free(im_fanout);
- return GIT_ERROR;
+ /*
+ * Make sure a corresponding .pack file exists and that
+ * the index looks sane.
+ */
+ path_len -= STRLEN(".idx");
+ if (path_len < 1) {
+ free(p);
+ return GIT_ENOTFOUND;
}
- p->idx_search = idxv1_search;
- p->idx_search_offset = idxv1_search_offset;
- p->idx_get = idxv1_get;
- p->im_fanout = im_fanout;
- p->im_oid = (unsigned char *)(src_fanout + 256);
+ memcpy(p->pack_name, path, path_len);
- if ((info = git__malloc(sizeof(*info) * (p->obj_cnt+1))) == NULL) {
- free(im_fanout);
- return GIT_ERROR;
- }
+ strcpy(p->pack_name + path_len, ".keep");
+ if (gitfo_exists(p->pack_name) == GIT_SUCCESS)
+ p->pack_keep = 1;
- for (j = 0; j < p->obj_cnt; j++) {
- uint32_t pos = j * (GIT_OID_RAWSZ + 4);
- info[j].offset = decode32(p->im_oid + pos);
- info[j].n = j;
+ strcpy(p->pack_name + path_len, ".pack");
+ if (gitfo_stat(p->pack_name, &st) < GIT_SUCCESS || !S_ISREG(st.st_mode)) {
+ free(p);
+ return GIT_ENOTFOUND;
}
- info[p->obj_cnt].offset = p->pack_size - GIT_OID_RAWSZ;
- info[p->obj_cnt].n = p->obj_cnt;
- if (make_offset_index(p, info)) {
- free(im_fanout);
- free(info);
- return GIT_ERROR;
- }
- free(info);
+ /* ok, it looks sane as far as we can check without
+ * actually mapping the pack file.
+ */
+ p->pack_size = (off_t)st.st_size;
+ p->pack_local = local;
+ p->mtime = st.st_mtime;
+ /* see if we can parse the sha1 oid in the packfile name */
+ if (path_len < 40 ||
+ git_oid_mkstr(&p->sha1, path + path_len - GIT_OID_HEXSZ) < GIT_SUCCESS)
+ memset(&p->sha1, 0x0, GIT_OID_RAWSZ);
+
+ *pack_out = p;
return GIT_SUCCESS;
}
-static int idxv2_search(uint32_t *out, git_pack *p, const git_oid *id)
+static int packfile_load__cb(void *_data, char *path)
{
- unsigned char *data = p->im_oid;
- uint32_t lo = id->id[0] ? p->im_fanout[id->id[0] - 1] : 0;
- uint32_t hi = p->im_fanout[id->id[0]];
+ struct pack__dirent *data = (struct pack__dirent *)_data;
+ struct pack_file *pack;
+ int error;
- do {
- uint32_t mid = (lo + hi) >> 1;
- uint32_t pos = 20 * mid;
- int cmp = memcmp(id->id, data + pos, 20);
- if (cmp < 0)
- hi = mid;
- else if (!cmp) {
- *out = mid;
- return GIT_SUCCESS;
- } else
- lo = mid + 1;
- } while (lo < hi);
- return GIT_ENOTFOUND;
-}
+ if (git__suffixcmp(path, ".idx") != 0)
+ return GIT_SUCCESS; /* not an index */
-static int idxv2_search_offset(uint32_t *out, git_pack *p, git_off_t offset)
-{
- if (offset > 0 && offset < (p->pack_size - GIT_OID_RAWSZ)) {
- uint32_t lo = 0, hi = p->obj_cnt+1;
- uint32_t *idx = p->im_off_idx;
- do {
- uint32_t mid = (lo + hi) >> 1;
- uint32_t n = idx[mid];
- uint32_t o32 = decode32(p->im_offset32 + n);
- git_off_t here = o32;
-
- if (o32 & 0x80000000) {
- uint32_t o64_idx = (o32 & ~0x80000000);
- here = decode64(p->im_offset64 + 2*o64_idx);
- }
+ /* FIXME: git.git checks for duplicate packs.
+ * But that makes no fucking sense. Our dirent is not
+ * going to generate dupicate entries */
- if (offset < here)
- hi = mid;
- else if (offset == here) {
- *out = n;
- return GIT_SUCCESS;
- } else
- lo = mid + 1;
- } while (lo < hi);
- }
- return GIT_ENOTFOUND;
-}
+ error = packfile_check(&pack, path, data->is_pack_local);
+ if (error < GIT_SUCCESS)
+ return error;
-static int idxv2_get(index_entry *e, git_pack *p, uint32_t n)
-{
- unsigned char *data = p->im_oid;
- uint32_t *next = p->im_off_next;
-
- if (n < p->obj_cnt) {
- uint32_t o32 = decode32(p->im_offset32 + n);
- git_off_t next_off = p->pack_size - GIT_OID_RAWSZ;
- e->n = n;
- e->oid = data + n * GIT_OID_RAWSZ;
- e->offset = o32;
- if (o32 & 0x80000000) {
- uint32_t o64_idx = (o32 & ~0x80000000);
- e->offset = decode64(p->im_offset64 + 2*o64_idx);
- }
- if (next[n] < p->obj_cnt) {
- o32 = decode32(p->im_offset32 + next[n]);
- next_off = o32;
- if (o32 & 0x80000000) {
- uint32_t o64_idx = (o32 & ~0x80000000);
- next_off = decode64(p->im_offset64 + 2*o64_idx);
- }
- }
- e->size = next_off - e->offset;
- return GIT_SUCCESS;
+ if (git_vector_insert(&data->backend->packs, pack) < GIT_SUCCESS) {
+ free(pack);
+ return GIT_ENOMEM;
}
- return GIT_ENOTFOUND;
+
+ return GIT_SUCCESS;
}
-static int pack_openidx_v2(git_pack *p)
+static int packfile_load_all(struct pack_backend *backend, const char *odb_path, int local)
{
- unsigned char *data = p->idx_map.data;
- uint32_t *src_fanout = (uint32_t *)(data + 8);
- uint32_t *im_fanout;
- offset_idx_info *info;
- size_t sz, o64_sz, o64_len;
- uint32_t j;
-
- if ((im_fanout = git__malloc(sizeof(*im_fanout) * 256)) == NULL)
- return GIT_ERROR;
-
- im_fanout[0] = decode32(&src_fanout[0]);
- for (j = 1; j < 256; j++) {
- im_fanout[j] = decode32(&src_fanout[j]);
- if (im_fanout[j] < im_fanout[j - 1]) {
- free(im_fanout);
- return GIT_ERROR;
- }
- }
- p->obj_cnt = im_fanout[255];
+ int error;
+ char path[GIT_PATH_MAX];
+ struct pack__dirent data;
- /* minimum size of .idx file (with empty 64-bit offsets table): */
- sz = 4 + 4 + 256 * 4 + p->obj_cnt * (20 + 4 + 4) + 2 * 20;
- if (p->idx_map.len < sz) {
- free(im_fanout);
- return GIT_ERROR;
- }
+ data.backend = backend;
+ data.is_pack_local = local;
- p->idx_search = idxv2_search;
- p->idx_search_offset = idxv2_search_offset;
- p->idx_get = idxv2_get;
- p->im_fanout = im_fanout;
- p->im_oid = (unsigned char *)(src_fanout + 256);
- p->im_crc = (uint32_t *)(p->im_oid + 20 * p->obj_cnt);
- p->im_offset32 = p->im_crc + p->obj_cnt;
- p->im_offset64 = p->im_offset32 + p->obj_cnt;
-
- if ((info = git__malloc(sizeof(*info) * (p->obj_cnt+1))) == NULL) {
- free(im_fanout);
- return GIT_ERROR;
- }
+ git__joinpath(path, odb_path, "pack");
+ if (gitfo_isdir(path) < GIT_SUCCESS)
+ return GIT_SUCCESS;
- /* check 64-bit offset table index values are within bounds */
- o64_sz = p->idx_map.len - sz;
- o64_len = o64_sz / 8;
- for (j = 0; j < p->obj_cnt; j++) {
- uint32_t o32 = decode32(p->im_offset32 + j);
- git_off_t offset = o32;
- if (o32 & 0x80000000) {
- uint32_t o64_idx = (o32 & ~0x80000000);
- if (o64_idx >= o64_len) {
- free(im_fanout);
- free(info);
- return GIT_ERROR;
- }
- offset = decode64(p->im_offset64 + 2*o64_idx);
- }
- info[j].offset = offset;
- info[j].n = j;
- }
- info[p->obj_cnt].offset = p->pack_size - GIT_OID_RAWSZ;
- info[p->obj_cnt].n = p->obj_cnt;
+ error = gitfo_dirent(path, GIT_PATH_MAX, packfile_load__cb, (void *)&data);
+ if (error < GIT_SUCCESS)
+ return error;
- if (make_offset_index(p, info)) {
- free(im_fanout);
- free(info);
- return GIT_ERROR;
- }
- free(info);
+ git_vector_sort(&backend->packs);
+ backend->last_found = git_vector_get(&backend->packs, 0);
return GIT_SUCCESS;
}
@@ -898,221 +895,421 @@ static int pack_openidx_v2(git_pack *p)
/***********************************************************
*
- * PACKFILE READING FUNCTIONS
- *
- * Read the contents of a packfile
+ * PACKFILE ENTRY SEARCH INTERNALS
*
***********************************************************/
+static off_t nth_packed_object_offset(const struct pack_file *p, uint32_t n)
+{
+ const unsigned char *index = p->index_map.data;
+ index += 4 * 256;
+ if (p->index_version == 1) {
+ return ntohl(*((uint32_t *)(index + 24 * n)));
+ } else {
+ uint32_t off;
+ index += 8 + p->num_objects * (20 + 4);
+ off = ntohl(*((uint32_t *)(index + 4 * n)));
+ if (!(off & 0x80000000))
+ return off;
+ index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
+ return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
+ ntohl(*((uint32_t *)(index + 4)));
+ }
+}
-static int unpack_object(git_rawobj *out, git_pack *p, index_entry *e);
-
-static int unpack_object_delta(git_rawobj *out, git_pack *p,
- index_entry *base_entry,
- uint8_t *delta_buffer,
- size_t delta_deflated_size,
- size_t delta_inflated_size)
+static int pack_entry_find_offset(
+ off_t *offset_out,
+ struct pack_file *p,
+ const git_oid *oid)
{
- int res = 0;
- uint8_t *delta = NULL;
- git_rawobj base_obj;
+ const uint32_t *level1_ofs = p->index_map.data;
+ const unsigned char *index = p->index_map.data;
+ unsigned hi, lo, stride;
- base_obj.data = NULL;
- base_obj.type = GIT_OBJ_BAD;
- base_obj.len = 0;
+ *offset_out = 0;
- if ((res = unpack_object(&base_obj, p, base_entry)) < 0)
- goto cleanup;
+ if (index == NULL) {
+ int error;
- delta = git__malloc(delta_inflated_size + 1);
+ if ((error = pack_index_open(p)) < GIT_SUCCESS)
+ return error;
- if ((res = git_odb__inflate_buffer(delta_buffer, delta_deflated_size,
- delta, delta_inflated_size)) < 0)
- goto cleanup;
+ assert(p->index_map.data);
- res = git__delta_apply(out, base_obj.data, base_obj.len, delta, delta_inflated_size);
+ index = p->index_map.data;
+ level1_ofs = p->index_map.data;
+ }
- out->type = base_obj.type;
+ if (p->index_version > 1) {
+ level1_ofs += 2;
+ index += 8;
+ }
-cleanup:
- free(delta);
- git_rawobj_close(&base_obj);
- return res;
-}
+ index += 4 * 256;
+ hi = ntohl(level1_ofs[(int)oid->id[0]]);
+ lo = ((oid->id[0] == 0x0) ? 0 : ntohl(level1_ofs[(int)oid->id[0] - 1]));
-static int unpack_object(git_rawobj *out, git_pack *p, index_entry *e)
-{
- git_otype object_type;
- size_t inflated_size, deflated_size, shift;
- uint8_t *buffer, byte;
+ if (p->index_version > 1) {
+ stride = 20;
+ } else {
+ stride = 24;
+ index += 4;
+ }
- assert(out && p && e && git__is_sizet(e->size));
+#ifdef INDEX_DEBUG_LOOKUP
+ printf("%02x%02x%02x... lo %u hi %u nr %d\n",
+ oid->id[0], oid->id[1], oid->id[2], lo, hi, p->num_objects);
+#endif
- if (open_pack(p))
- return GIT_ERROR;
+#ifdef GIT2_INDEX_LOOKUP /* TODO: use the advanced lookup method from git.git */
- buffer = (uint8_t *)p->pack_map.data + e->offset;
- deflated_size = (size_t)e->size;
+ int pos = sha1_entry_pos(index, stride, 0, lo, hi, p->num_objects, oid);
+ if (pos < 0)
+ return GIT_ENOTFOUND;
- if (deflated_size == 0)
- deflated_size = (size_t)(p->pack_size - e->offset);
+ *offset_out = nth_packed_object_offset(p, pos);
+ return GIT_SUCCESS;
- byte = *buffer++ & 0xFF;
- deflated_size--;
- object_type = (byte >> 4) & 0x7;
- inflated_size = byte & 0xF;
- shift = 4;
+#else /* use an old and boring binary search */
- while (byte & 0x80) {
- byte = *buffer++ & 0xFF;
- deflated_size--;
- inflated_size += (byte & 0x7F) << shift;
- shift += 7;
- }
+ do {
+ unsigned mi = (lo + hi) / 2;
+ int cmp = memcmp(index + mi * stride, oid->id, GIT_OID_RAWSZ);
- switch (object_type) {
- case GIT_OBJ_COMMIT:
- case GIT_OBJ_TREE:
- case GIT_OBJ_BLOB:
- case GIT_OBJ_TAG: {
+ if (!cmp) {
+ *offset_out = nth_packed_object_offset(p, mi);
+ return GIT_SUCCESS;
+ }
- /* Handle a normal zlib stream */
- out->len = inflated_size;
- out->type = object_type;
- out->data = git__malloc(inflated_size + 1);
+ if (cmp > 0)
+ hi = mi;
+ else
+ lo = mi+1;
- if (git_odb__inflate_buffer(buffer, deflated_size, out->data, out->len) < 0) {
- free(out->data);
- out->data = NULL;
+ } while (lo < hi);
+
+ return GIT_ENOTFOUND;
+#endif
+}
+
+static int pack_entry_find1(
+ struct pack_entry *e,
+ struct pack_file *p,
+ const git_oid *oid)
+{
+ off_t offset;
+
+ assert(p);
+
+ if (p->num_bad_objects) {
+ unsigned i;
+ for (i = 0; i < p->num_bad_objects; i++)
+ if (git_oid_cmp(oid, &p->bad_object_sha1[i]) == 0)
return GIT_ERROR;
- }
+ }
- return GIT_SUCCESS;
- }
+ if (pack_entry_find_offset(&offset, p, oid) < GIT_SUCCESS)
+ return GIT_ENOTFOUND;
+
+ /* we found an entry in the index;
+ * make sure the packfile backing the index
+ * still exists on disk */
+ if (p->pack_fd == -1 && packfile_open(p) < GIT_SUCCESS)
+ return GIT_EOSERR;
- case GIT_OBJ_OFS_DELTA: {
+ e->offset = offset;
+ e->p = p;
- git_off_t delta_offset;
- index_entry entry;
+ git_oid_cpy(&e->sha1, oid);
+ return GIT_SUCCESS;
+}
- byte = *buffer++ & 0xFF;
- delta_offset = byte & 0x7F;
+static int pack_entry_find(struct pack_entry *e, struct pack_backend *backend, const git_oid *oid)
+{
+ size_t i;
- while (byte & 0x80) {
- delta_offset += 1;
- byte = *buffer++ & 0xFF;
- delta_offset <<= 7;
- delta_offset += (byte & 0x7F);
- }
+ if (backend->last_found &&
+ pack_entry_find1(e, backend->last_found, oid) == GIT_SUCCESS)
+ return GIT_SUCCESS;
- entry.n = 0;
- entry.oid = NULL;
- entry.offset = e->offset - delta_offset;
- entry.size = 0;
+ for (i = 0; i < backend->packs.length; ++i) {
+ struct pack_file *p;
- if (unpack_object_delta(out, p, &entry,
- buffer, deflated_size, inflated_size) < 0)
- return GIT_ERROR;
+ p = git_vector_get(&backend->packs, i);
+ if (p == backend->last_found)
+ continue;
+ if (pack_entry_find1(e, p, oid) == GIT_SUCCESS) {
+ backend->last_found = p;
return GIT_SUCCESS;
}
+ }
+
+ return GIT_ENOTFOUND;
+}
+
+
- case GIT_OBJ_REF_DELTA: {
- git_oid base_id;
- uint32_t n;
- index_entry entry;
- int res = GIT_ERROR;
- git_oid_mkraw(&base_id, buffer);
- if (!p->idx_search(&n, p, &base_id) &&
- !p->idx_get(&entry, p, n)) {
- res = unpack_object_delta(out, p, &entry,
- buffer + GIT_OID_RAWSZ, deflated_size, inflated_size);
- }
- return res;
- }
- default:
- return GIT_EOBJCORRUPTED;
- }
-}
-static int read_packed(git_rawobj *out, const pack_location *loc)
-{
- index_entry e;
- int res;
- assert(out && loc);
- if (pack_openidx(loc->ptr) < 0)
- return GIT_EPACKCORRUPTED;
+/***********************************************************
+ *
+ * PACKFILE ENTRY UNPACK INTERNALS
+ *
+ ***********************************************************/
- res = loc->ptr->idx_get(&e, loc->ptr, loc->n);
+static unsigned long packfile_unpack_header1(
+ size_t *sizep,
+ git_otype *type,
+ const unsigned char *buf,
+ unsigned long len)
+{
+ unsigned shift;
+ unsigned long size, c;
+ unsigned long used = 0;
- if (!res)
- res = unpack_object(out, loc->ptr, &e);
+ c = buf[used++];
+ *type = (c >> 4) & 7;
+ size = c & 15;
+ shift = 4;
+ while (c & 0x80) {
+ if (len <= used || bitsizeof(long) <= shift)
+ return 0;
- pack_decidx(loc->ptr);
+ c = buf[used++];
+ size += (c & 0x7f) << shift;
+ shift += 7;
+ }
- return res;
+ *sizep = (size_t)size;
+ return used;
}
-static int read_header_packed(git_rawobj *out, const pack_location *loc)
+static int packfile_unpack_header(
+ size_t *size_p,
+ git_otype *type_p,
+ struct pack_backend *backend,
+ struct pack_file *p,
+ struct pack_window **w_curs,
+ off_t *curpos)
{
- git_pack *pack;
- index_entry e;
- int error = GIT_SUCCESS, shift;
- uint8_t *buffer, byte;
+ unsigned char *base;
+ unsigned int left;
+ unsigned long used;
+
+ /* pack_window_open() assures us we have [base, base + 20) available
+ * as a range that we can look at at. (Its actually the hash
+ * size that is assured.) With our object header encoding
+ * the maximum deflated object size is 2^137, which is just
+ * insane, so we know won't exceed what we have been given.
+ */
+ base = pack_window_open(backend, p, w_curs, *curpos, &left);
+ if (base == NULL)
+ return GIT_ENOMEM;
- assert(out && loc);
+ used = packfile_unpack_header1(size_p, type_p, base, left);
- pack = loc->ptr;
+ if (used == 0)
+ return GIT_EOBJCORRUPTED;
- if (pack_openidx(pack))
- return GIT_EPACKCORRUPTED;
+ *curpos += used;
+ return GIT_SUCCESS;
+}
- if (pack->idx_get(&e, pack, loc->n) < 0 ||
- open_pack(pack) < 0) {
- error = GIT_ENOTFOUND;
- goto cleanup;
+static int packfile_unpack_compressed(
+ git_rawobj *obj,
+ struct pack_backend *backend,
+ struct pack_file *p,
+ struct pack_window **w_curs,
+ off_t curpos,
+ size_t size,
+ git_otype type)
+{
+ int st;
+ z_stream stream;
+ unsigned char *buffer, *in;
+
+ buffer = git__malloc(size);
+
+ memset(&stream, 0, sizeof(stream));
+ stream.next_out = buffer;
+ stream.avail_out = size + 1;
+
+ st = inflateInit(&stream);
+ if (st != Z_OK) {
+ free(buffer);
+ return GIT_EZLIB;
}
- buffer = (uint8_t *)pack->pack_map.data + e.offset;
+ do {
+ in = pack_window_open(backend, p, w_curs, curpos, &stream.avail_in);
+ stream.next_in = in;
+ st = inflate(&stream, Z_FINISH);
- byte = *buffer++ & 0xFF;
- out->type = (byte >> 4) & 0x7;
- out->len = byte & 0xF;
- shift = 4;
+ if (!stream.avail_out)
+ break; /* the payload is larger than it should be */
- while (byte & 0x80) {
- byte = *buffer++ & 0xFF;
- out->len += (byte & 0x7F) << shift;
- shift += 7;
+ curpos += stream.next_in - in;
+ } while (st == Z_OK || st == Z_BUF_ERROR);
+
+ inflateEnd(&stream);
+
+ if ((st != Z_STREAM_END) || stream.total_out != size) {
+ free(buffer);
+ return GIT_EZLIB;
}
- /*
- * FIXME: if the object is not packed as a whole,
- * we need to do a full load and apply the deltas before
- * being able to read the header.
- *
- * I don't think there are any workarounds for this.'
+ obj->type = type;
+ obj->len = size;
+ obj->data = buffer;
+ return GIT_SUCCESS;
+}
+
+static off_t get_delta_base(
+ struct pack_backend *backend,
+ struct pack_file *p,
+ struct pack_window **w_curs,
+ off_t *curpos,
+ git_otype type,
+ off_t delta_obj_offset)
+{
+ unsigned char *base_info = pack_window_open(backend, p, w_curs, *curpos, NULL);
+ off_t base_offset;
+
+ /* pack_window_open() assured us we have [base_info, base_info + 20)
+ * as a range that we can look at without walking off the
+ * end of the mapped window. Its actually the hash size
+ * that is assured. An OFS_DELTA longer than the hash size
+ * is stupid, as then a REF_DELTA would be smaller to store.
*/
+ if (type == GIT_OBJ_OFS_DELTA) {
+ unsigned used = 0;
+ unsigned char c = base_info[used++];
+ base_offset = c & 127;
+ while (c & 128) {
+ base_offset += 1;
+ if (!base_offset || MSB(base_offset, 7))
+ return 0; /* overflow */
+ c = base_info[used++];
+ base_offset = (base_offset << 7) + (c & 127);
+ }
+ base_offset = delta_obj_offset - base_offset;
+ if (base_offset <= 0 || base_offset >= delta_obj_offset)
+ return 0; /* out of bound */
+ *curpos += used;
+ } else if (type == GIT_OBJ_REF_DELTA) {
+ /* The base entry _must_ be in the same pack */
+ if (pack_entry_find_offset(&base_offset, p, (git_oid *)base_info) < GIT_SUCCESS)
+ return GIT_EPACKCORRUPTED;
+ *curpos += 20;
+ } else
+ return 0;
- if (out->type == GIT_OBJ_OFS_DELTA || out->type == GIT_OBJ_REF_DELTA) {
- error = unpack_object(out, pack, &e);
- git_rawobj_close(out);
+ return base_offset;
+}
+
+static int packfile_unpack_delta(
+ git_rawobj *obj,
+ struct pack_backend *backend,
+ struct pack_file *p,
+ struct pack_window **w_curs,
+ off_t curpos,
+ size_t delta_size,
+ git_otype delta_type,
+ off_t obj_offset)
+{
+ off_t base_offset;
+ git_rawobj base, delta;
+ int error;
+
+ base_offset = get_delta_base(backend, p, w_curs, &curpos, delta_type, obj_offset);
+ if (base_offset == 0)
+ return GIT_EOBJCORRUPTED;
+
+ pack_window_close(w_curs);
+ error = packfile_unpack(&base, backend, p, base_offset);
+
+ /* TODO: git.git tries to load the base from other packfiles
+ * or loose objects */
+ if (error < GIT_SUCCESS)
+ return error;
+
+ error = packfile_unpack_compressed(&delta, backend, p, w_curs, curpos, delta_size, delta_type);
+ if (error < GIT_SUCCESS) {
+ git_rawobj_close(&base);
+ return error;
}
-cleanup:
- pack_decidx(loc->ptr);
+ obj->type = base.type;
+ error = git__delta_apply(obj,
+ base.data, base.len,
+ delta.data, delta.len);
+
+ git_rawobj_close(&base);
+ git_rawobj_close(&delta);
+
+ /* TODO: we might want to cache this shit. eventually */
+ //add_delta_base_cache(p, base_offset, base, base_size, *type);
return error;
}
+static int packfile_unpack(
+ git_rawobj *obj,
+ struct pack_backend *backend,
+ struct pack_file *p,
+ off_t obj_offset)
+{
+ struct pack_window *w_curs = NULL;
+ off_t curpos = obj_offset;
+ int error;
+
+ size_t size;
+ git_otype type;
+ /*
+ * TODO: optionally check the CRC on the packfile
+ */
+
+ obj->data = NULL;
+ obj->len = 0;
+ obj->type = GIT_OBJ_BAD;
+
+ error = packfile_unpack_header(&size, &type, backend, p, &w_curs, &curpos);
+ if (error < GIT_SUCCESS)
+ return error;
+
+ switch (type) {
+ case GIT_OBJ_OFS_DELTA:
+ case GIT_OBJ_REF_DELTA:
+ error = packfile_unpack_delta(
+ obj, backend, p, &w_curs, curpos,
+ size, type, obj_offset);
+ break;
+
+ case GIT_OBJ_COMMIT:
+ case GIT_OBJ_TREE:
+ case GIT_OBJ_BLOB:
+ case GIT_OBJ_TAG:
+ error = packfile_unpack_compressed(
+ obj, backend, p, &w_curs, curpos,
+ size, type);
+ break;
+
+ default:
+ error = GIT_EOBJCORRUPTED;
+ break;
+ }
+
+ pack_window_close(&w_curs);
+ return error;
+}
@@ -1126,79 +1323,80 @@ cleanup:
*
***********************************************************/
+/*
int pack_backend__read_header(git_rawobj *obj, git_odb_backend *backend, const git_oid *oid)
{
pack_location location;
assert(obj && backend && oid);
- if (locate_packfile(&location, (pack_backend *)backend, oid) < 0)
+ if (locate_packfile(&location, (struct pack_backend *)backend, oid) < 0)
return GIT_ENOTFOUND;
return read_header_packed(obj, &location);
}
+*/
int pack_backend__read(git_rawobj *obj, git_odb_backend *backend, const git_oid *oid)
{
- pack_location location;
-
- assert(obj && backend && oid);
+ struct pack_entry e;
+ int error;
- if (locate_packfile(&location, (pack_backend *)backend, oid) < 0)
- return GIT_ENOTFOUND;
+ if ((error = pack_entry_find(&e, (struct pack_backend *)backend, oid)) < GIT_SUCCESS)
+ return error;
- return read_packed(obj, &location);
+ return packfile_unpack(obj, (struct pack_backend *)backend, e.p, e.offset);
}
int pack_backend__exists(git_odb_backend *backend, const git_oid *oid)
{
- pack_location location;
- assert(backend && oid);
- return locate_packfile(&location, (pack_backend *)backend, oid) == GIT_SUCCESS;
+ struct pack_entry e;
+ return pack_entry_find(&e, (struct pack_backend *)backend, oid) == GIT_SUCCESS;
}
void pack_backend__free(git_odb_backend *_backend)
{
- pack_backend *backend;
- git_packlist *pl;
+ struct pack_backend *backend;
+ size_t i;
assert(_backend);
- backend = (pack_backend *)_backend;
-
- gitlck_lock(&backend->lock);
-
- pl = backend->packlist;
- backend->packlist = NULL;
+ backend = (struct pack_backend *)_backend;
- gitlck_unlock(&backend->lock);
- if (pl)
- packlist_dec(backend, pl);
-
- gitlck_free(&backend->lock);
+ for (i = 0; i < backend->packs.length; ++i) {
+ struct pack_file *p = git_vector_get(&backend->packs, i);
+ packfile_free(backend, p);
+ }
- free(backend->objects_dir);
+ git_vector_free(&backend->packs);
free(backend);
}
int git_odb_backend_pack(git_odb_backend **backend_out, const char *objects_dir)
{
- pack_backend *backend;
+ int error;
+ struct pack_backend *backend;
- backend = git__calloc(1, sizeof(pack_backend));
+ backend = git__calloc(1, sizeof(struct pack_backend));
if (backend == NULL)
return GIT_ENOMEM;
- backend->objects_dir = git__strdup(objects_dir);
- if (backend->objects_dir == NULL) {
+ if (git_vector_init(&backend->packs, 8, packfile_sort__cb) < GIT_SUCCESS) {
free(backend);
return GIT_ENOMEM;
}
- gitlck_init(&backend->lock);
+ backend->window_size = DEFAULT_WINDOW_SIZE;
+ backend->mapped_limit = DEFAULT_MAPPED_LIMIT;
+
+ error = packfile_load_all(backend, objects_dir, 1);
+ if (error < GIT_SUCCESS) {
+ pack_backend__free((git_odb_backend *)backend);
+ return error;
+ }
backend->parent.read = &pack_backend__read;
- backend->parent.read_header = &pack_backend__read_header;
+ backend->parent.read_header = NULL;
backend->parent.write = NULL;
backend->parent.exists = &pack_backend__exists;
backend->parent.free = &pack_backend__free;
diff --git a/src/util.h b/src/util.h
index fab93c1c2..e0dfd7b20 100644
--- a/src/util.h
+++ b/src/util.h
@@ -2,6 +2,8 @@
#define INCLUDE_util_h__
#define ARRAY_SIZE(x) (sizeof(x)/sizeof(x[0]))
+#define bitsizeof(x) (CHAR_BIT * sizeof(x))
+#define MSB(x, bits) ((x) & (~0ULL << (bitsizeof(x) - (bits))))
/*
* Don't wrap malloc/calloc.