]> asedeno.scripts.mit.edu Git - git.git/blobdiff - sha1_file.c
Fix read_mailmap to handle a caller uninterested in repo abbreviation
[git.git] / sha1_file.c
index b0b21776e79cddcdbbf93b4ee6daed2abdf804c4..32244d704e1c747780aed461bb78cf4a72199f18 100644 (file)
@@ -13,6 +13,7 @@
 #include "commit.h"
 #include "tag.h"
 #include "tree.h"
+#include "refs.h"
 
 #ifndef O_NOATIME
 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
@@ -437,7 +438,7 @@ static int check_packed_git_idx(const char *path,  struct packed_git *p)
        void *idx_map;
        struct pack_idx_header *hdr;
        size_t idx_size;
-       uint32_t nr, i, *index;
+       uint32_t version, nr, i, *index;
        int fd = open(path, O_RDONLY);
        struct stat st;
 
@@ -455,21 +456,23 @@ static int check_packed_git_idx(const char *path,  struct packed_git *p)
        idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
        close(fd);
 
-       /* a future index format would start with this, as older git
-        * binaries would fail the non-monotonic index check below.
-        * give a nicer warning to the user if we can.
-        */
        hdr = idx_map;
        if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
-               munmap(idx_map, idx_size);
-               return error("index file %s is a newer version"
-                       " and is not supported by this binary"
-                       " (try upgrading GIT to a newer version)",
-                       path);
-       }
+               version = ntohl(hdr->idx_version);
+               if (version < 2 || version > 2) {
+                       munmap(idx_map, idx_size);
+                       return error("index file %s is version %d"
+                                    " and is not supported by this binary"
+                                    " (try upgrading GIT to a newer version)",
+                                    path, version);
+               }
+       } else
+               version = 1;
 
        nr = 0;
        index = idx_map;
+       if (version > 1)
+               index += 2;  /* skip index header */
        for (i = 0; i < 256; i++) {
                uint32_t n = ntohl(index[i]);
                if (n < nr) {
@@ -479,21 +482,51 @@ static int check_packed_git_idx(const char *path,  struct packed_git *p)
                nr = n;
        }
 
-       /*
-        * 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) {
-               munmap(idx_map, idx_size);
-               return error("wrong index file size in %s", path);
+       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) {
+                       munmap(idx_map, idx_size);
+                       return error("wrong index file size in %s", path);
+               }
+       } 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;
+               if (idx_size < min_size || idx_size > min_size + (nr - 1)*8) {
+                       munmap(idx_map, idx_size);
+                       return error("wrong index file size in %s", path);
+               }
+               if (idx_size != min_size) {
+                       /* make sure we can deal with large pack offsets */
+                       off_t x = 0x7fffffffUL, y = 0xffffffffUL;
+                       if (x > (x + 1) || y > (y + 1)) {
+                               munmap(idx_map, idx_size);
+                               return error("pack too large for current definition of off_t in %s", path);
+                       }
+               }
        }
 
-       p->index_version = 1;
+       p->index_version = version;
        p->index_data = idx_map;
        p->index_size = idx_size;
+       p->num_objects = nr;
        return 0;
 }
 
@@ -516,7 +549,7 @@ static void scan_windows(struct packed_git *p,
        }
 }
 
-static int unuse_one_window(struct packed_git *current)
+static int unuse_one_window(struct packed_git *current, int keep_fd)
 {
        struct packed_git *p, *lru_p = NULL;
        struct pack_window *lru_w = NULL, *lru_l = NULL;
@@ -532,7 +565,7 @@ static int unuse_one_window(struct packed_git *current)
                        lru_l->next = lru_w->next;
                else {
                        lru_p->windows = lru_w->next;
-                       if (!lru_p->windows && lru_p != current) {
+                       if (!lru_p->windows && lru_p->pack_fd != keep_fd) {
                                close(lru_p->pack_fd);
                                lru_p->pack_fd = -1;
                        }
@@ -544,10 +577,10 @@ static int unuse_one_window(struct packed_git *current)
        return 0;
 }
 
-void release_pack_memory(size_t need)
+void release_pack_memory(size_t need, int fd)
 {
        size_t cur = pack_mapped;
-       while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
+       while (need >= (cur - pack_mapped) && unuse_one_window(NULL, fd))
                ; /* nothing */
 }
 
@@ -605,11 +638,11 @@ static int open_packed_git_1(struct packed_git *p)
                        p->pack_name, ntohl(hdr.hdr_version));
 
        /* Verify the pack matches its index. */
-       if (num_packed_objects(p) != ntohl(hdr.hdr_entries))
+       if (p->num_objects != ntohl(hdr.hdr_entries))
                return error("packfile %s claims to have %u objects"
-                       " while index size indicates %u objects",
-                       p->pack_name, ntohl(hdr.hdr_entries),
-                       num_packed_objects(p));
+                            " while index indicates %u objects",
+                            p->pack_name, ntohl(hdr.hdr_entries),
+                            p->num_objects);
        if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
                return error("end of packfile %s is unavailable", p->pack_name);
        if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
@@ -680,7 +713,7 @@ unsigned char* use_pack(struct packed_git *p,
                        win->len = (size_t)len;
                        pack_mapped += win->len;
                        while (packed_git_limit < pack_mapped
-                               && unuse_one_window(p))
+                               && unuse_one_window(p, p->pack_fd))
                                ; /* nothing */
                        win->base = xmmap(NULL, win->len,
                                PROT_READ, MAP_PRIVATE,
@@ -1030,14 +1063,27 @@ static void *unpack_sha1_rest(z_stream *stream, void *buffer, unsigned long size
                n = size;
        memcpy(buf, (char *) buffer + bytes, n);
        bytes = n;
-       if (bytes < size) {
+       if (bytes <= size) {
+               /*
+                * The above condition must be (bytes <= size), not
+                * (bytes < size).  In other words, even though we
+                * expect no more output and set avail_out to zer0,
+                * the input zlib stream may have bytes that express
+                * "this concludes the stream", and we *do* want to
+                * eat that input.
+                *
+                * Otherwise we would not be able to test that we
+                * consumed all the input to reach the expected size;
+                * we also want to check that zlib tells us that all
+                * went well with status == Z_STREAM_END at the end.
+                */
                stream->next_out = buf + bytes;
                stream->avail_out = size - bytes;
                while (status == Z_OK)
                        status = inflate(stream, Z_FINISH);
        }
        buf[size] = 0;
-       if ((status == Z_OK || status == Z_STREAM_END) && !stream->avail_in) {
+       if (status == Z_STREAM_END && !stream->avail_in) {
                inflateEnd(stream);
                return buf;
        }
@@ -1115,6 +1161,43 @@ static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type
        return unpack_sha1_rest(&stream, hdr, *size, sha1);
 }
 
+unsigned long get_size_from_delta(struct packed_git *p,
+                                 struct pack_window **w_curs,
+                                 off_t curpos)
+{
+       const unsigned char *data;
+       unsigned char delta_head[20], *in;
+       z_stream stream;
+       int st;
+
+       memset(&stream, 0, sizeof(stream));
+       stream.next_out = delta_head;
+       stream.avail_out = sizeof(delta_head);
+
+       inflateInit(&stream);
+       do {
+               in = use_pack(p, w_curs, curpos, &stream.avail_in);
+               stream.next_in = in;
+               st = inflate(&stream, Z_FINISH);
+               curpos += stream.next_in - in;
+       } while ((st == Z_OK || st == Z_BUF_ERROR) &&
+                stream.total_out < sizeof(delta_head));
+       inflateEnd(&stream);
+       if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head))
+               die("delta data unpack-initial failed");
+
+       /* Examine the initial part of the delta to figure out
+        * the result size.
+        */
+       data = delta_head;
+
+       /* ignore base size */
+       get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
+
+       /* Read the result size */
+       return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
+}
+
 static off_t get_delta_base(struct packed_git *p,
                                    struct pack_window **w_curs,
                                    off_t *curpos,
@@ -1136,7 +1219,7 @@ static off_t get_delta_base(struct packed_git *p,
                base_offset = c & 127;
                while (c & 128) {
                        base_offset += 1;
-                       if (!base_offset || base_offset & ~(~0UL >> 7))
+                       if (!base_offset || MSB(base_offset, 7))
                                die("offset value overflow for delta base object");
                        c = base_info[used++];
                        base_offset = (base_offset << 7) + (c & 127);
@@ -1178,40 +1261,8 @@ static int packed_delta_info(struct packed_git *p,
         * based on a base with a wrong size.  This saves tons of
         * inflate() calls.
         */
-       if (sizep) {
-               const unsigned char *data;
-               unsigned char delta_head[20], *in;
-               z_stream stream;
-               int st;
-
-               memset(&stream, 0, sizeof(stream));
-               stream.next_out = delta_head;
-               stream.avail_out = sizeof(delta_head);
-
-               inflateInit(&stream);
-               do {
-                       in = use_pack(p, w_curs, curpos, &stream.avail_in);
-                       stream.next_in = in;
-                       st = inflate(&stream, Z_FINISH);
-                       curpos += stream.next_in - in;
-               } while ((st == Z_OK || st == Z_BUF_ERROR)
-                       && stream.total_out < sizeof(delta_head));
-               inflateEnd(&stream);
-               if ((st != Z_STREAM_END) &&
-                   stream.total_out != sizeof(delta_head))
-                       die("delta data unpack-initial failed");
-
-               /* Examine the initial part of the delta to figure out
-                * the result size.
-                */
-               data = delta_head;
-
-               /* ignore base size */
-               get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
-
-               /* Read the result size */
-               *sizep = get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
-       }
+       if (sizep)
+               *sizep = get_size_from_delta(p, w_curs, curpos);
 
        return type;
 }
@@ -1355,11 +1406,18 @@ static void *unpack_compressed_entry(struct packed_git *p,
 #define MAX_DELTA_CACHE (256)
 
 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 struct delta_base_cache_entry {
+       struct delta_base_cache_lru_list lru;
+       void *data;
        struct packed_git *p;
        off_t base_offset;
        unsigned long size;
-       void *data;
        enum object_type type;
 } delta_base_cache[MAX_DELTA_CACHE];
 
@@ -1369,7 +1427,7 @@ static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
 
        hash = (unsigned long)p + (unsigned long)base_offset;
        hash += (hash >> 8) + (hash >> 16);
-       return hash & 0xff;
+       return hash % MAX_DELTA_CACHE;
 }
 
 static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
@@ -1387,6 +1445,8 @@ static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
 found_cache_entry:
        if (!keep_cache) {
                ent->data = NULL;
+               ent->lru.next->prev = ent->lru.prev;
+               ent->lru.prev->next = ent->lru.next;
                delta_base_cached -= ent->size;
        }
        else {
@@ -1404,6 +1464,8 @@ 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;
        }
 }
@@ -1411,26 +1473,38 @@ static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
 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 i, hash = pack_entry_hash(p, base_offset);
+       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;
 
        release_delta_base_cache(ent);
        delta_base_cached += base_size;
-       for (i = 0; delta_base_cached > delta_base_cache_limit
-               && i < ARRAY_SIZE(delta_base_cache); i++) {
-               struct delta_base_cache_entry *f = delta_base_cache + i;
+
+       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 (i = 0; delta_base_cached > delta_base_cache_limit
-               && i < ARRAY_SIZE(delta_base_cache); i++)
-               release_delta_base_cache(delta_base_cache + i);
+       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;
+               release_delta_base_cache(f);
+       }
 
        ent->p = p;
        ent->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;
 }
 
 static void *unpack_delta_entry(struct packed_git *p,
@@ -1490,38 +1564,60 @@ void *unpack_entry(struct packed_git *p, off_t obj_offset,
        return data;
 }
 
-uint32_t num_packed_objects(const struct packed_git *p)
+const unsigned char *nth_packed_object_sha1(const struct packed_git *p,
+                                           uint32_t n)
 {
-       /* See check_packed_git_idx() */
-       return (uint32_t)((p->index_size - 20 - 20 - 4*256) / 24);
+       const unsigned char *index = p->index_data;
+       if (n >= p->num_objects)
+               return NULL;
+       index += 4 * 256;
+       if (p->index_version == 1) {
+               return index + 24 * n + 4;
+       } else {
+               index += 8;
+               return index + 20 * n;
+       }
 }
 
-int nth_packed_object_sha1(const struct packed_git *p, uint32_t n,
-                          unsigned char* sha1)
+static off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
 {
        const unsigned char *index = p->index_data;
        index += 4 * 256;
-       if (num_packed_objects(p) <= n)
-               return -1;
-       hashcpy(sha1, index + 24 * n + 4);
-       return 0;
+       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)));
+       }
 }
 
 off_t find_pack_entry_one(const unsigned char *sha1,
                                  struct packed_git *p)
 {
        const uint32_t *level1_ofs = p->index_data;
-       int hi = ntohl(level1_ofs[*sha1]);
-       int lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
        const unsigned char *index = p->index_data;
+       unsigned hi, lo;
 
+       if (p->index_version > 1) {
+               level1_ofs += 2;
+               index += 8;
+       }
        index += 4 * 256;
+       hi = ntohl(level1_ofs[*sha1]);
+       lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
 
        do {
-               int mi = (lo + hi) / 2;
-               int cmp = hashcmp(index + 24 * mi + 4, sha1);
+               unsigned mi = (lo + hi) / 2;
+               unsigned x = (p->index_version > 1) ? (mi * 20) : (mi * 24 + 4);
+               int cmp = hashcmp(index + x, sha1);
                if (!cmp)
-                       return ntohl(*((uint32_t *)((char *)index + (24 * mi))));
+                       return nth_packed_object_offset(p, mi);
                if (cmp > 0)
                        hi = mi;
                else
@@ -1772,7 +1868,7 @@ void *read_object_with_reference(const unsigned char *sha1,
        }
 }
 
-static void write_sha1_file_prepare(void *buf, unsigned long len,
+static void write_sha1_file_prepare(const void *buf, unsigned long len,
                                     const char *type, unsigned char *sha1,
                                     char *hdr, int *hdrlen)
 {
@@ -1900,7 +1996,7 @@ static void setup_object_header(z_stream *stream, const char *type, unsigned lon
        stream->avail_out -= hdrlen;
 }
 
-int hash_sha1_file(void *buf, unsigned long len, const char *type,
+int hash_sha1_file(const void *buf, unsigned long len, const char *type,
                    unsigned char *sha1)
 {
        char hdr[32];
@@ -1911,7 +2007,7 @@ int hash_sha1_file(void *buf, unsigned long len, const char *type,
 
 int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned char *returnsha1)
 {
-       int size;
+       int size, ret;
        unsigned char *compressed;
        z_stream stream;
        unsigned char sha1[20];
@@ -1943,7 +2039,7 @@ int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned cha
                return error("sha1 file %s: %s\n", filename, strerror(errno));
        }
 
-       snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
+       snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
 
        fd = mkstemp(tmpfile);
        if (fd < 0) {
@@ -1971,15 +2067,21 @@ int write_sha1_file(void *buf, unsigned long len, const char *type, unsigned cha
        /* Then the data itself.. */
        stream.next_in = buf;
        stream.avail_in = len;
-       while (deflate(&stream, Z_FINISH) == Z_OK)
-               /* nothing */;
-       deflateEnd(&stream);
+       ret = deflate(&stream, Z_FINISH);
+       if (ret != Z_STREAM_END)
+               die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
+
+       ret = deflateEnd(&stream);
+       if (ret != Z_OK)
+               die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
+
        size = stream.total_out;
 
        if (write_buffer(fd, compressed, size) < 0)
                die("unable to write sha1 file");
        fchmod(fd, 0444);
-       close(fd);
+       if (close(fd))
+               die("unable to write sha1 file");
        free(compressed);
 
        return move_temp_to_file(tmpfile, filename);
@@ -2064,7 +2166,7 @@ int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
        int ret;
        SHA_CTX c;
 
-       snprintf(tmpfile, sizeof(tmpfile), "%s/obj_XXXXXX", get_object_directory());
+       snprintf(tmpfile, sizeof(tmpfile), "%s/tmp_obj_XXXXXX", get_object_directory());
 
        local = mkstemp(tmpfile);
        if (local < 0) {
@@ -2113,7 +2215,9 @@ int write_sha1_from_fd(const unsigned char *sha1, int fd, char *buffer,
        } while (1);
        inflateEnd(&stream);
 
-       close(local);
+       fchmod(local, 0444);
+       if (close(local) != 0)
+               die("unable to write sha1 file");
        SHA1_Final(real_sha1, &c);
        if (ret != Z_STREAM_END) {
                unlink(tmpfile);
@@ -2234,10 +2338,9 @@ int index_fd(unsigned char *sha1, int fd, struct stat *st, int write_object,
         */
        if ((type == OBJ_BLOB) && S_ISREG(st->st_mode)) {
                unsigned long nsize = size;
-               char *nbuf = buf;
-               if (convert_to_git(path, &nbuf, &nsize)) {
-                       if (size)
-                               munmap(buf, size);
+               char *nbuf = convert_to_git(path, buf, &nsize);
+               if (nbuf) {
+                       munmap(buf, size);
                        size = nsize;
                        buf = nbuf;
                        re_allocated = 1;
@@ -2289,6 +2392,8 @@ int index_path(unsigned char *sha1, const char *path, struct stat *st, int write
                                     path);
                free(target);
                break;
+       case S_IFDIR:
+               return resolve_gitlink_ref(path, "HEAD", sha1);
        default:
                return error("%s: unsupported file type", path);
        }