Merge branch 'ds/blame-on-bloom'
[git] / pack-check.c
1 #include "cache.h"
2 #include "repository.h"
3 #include "pack.h"
4 #include "pack-revindex.h"
5 #include "progress.h"
6 #include "packfile.h"
7 #include "object-store.h"
8
9 struct idx_entry {
10         off_t                offset;
11         unsigned int nr;
12 };
13
14 static int compare_entries(const void *e1, const void *e2)
15 {
16         const struct idx_entry *entry1 = e1;
17         const struct idx_entry *entry2 = e2;
18         if (entry1->offset < entry2->offset)
19                 return -1;
20         if (entry1->offset > entry2->offset)
21                 return 1;
22         return 0;
23 }
24
25 int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
26                    off_t offset, off_t len, unsigned int nr)
27 {
28         const uint32_t *index_crc;
29         uint32_t data_crc = crc32(0, NULL, 0);
30
31         do {
32                 unsigned long avail;
33                 void *data = use_pack(p, w_curs, offset, &avail);
34                 if (avail > len)
35                         avail = len;
36                 data_crc = crc32(data_crc, data, avail);
37                 offset += avail;
38                 len -= avail;
39         } while (len);
40
41         index_crc = p->index_data;
42         index_crc += 2 + 256 + p->num_objects * (the_hash_algo->rawsz/4) + nr;
43
44         return data_crc != ntohl(*index_crc);
45 }
46
47 static int verify_packfile(struct repository *r,
48                            struct packed_git *p,
49                            struct pack_window **w_curs,
50                            verify_fn fn,
51                            struct progress *progress, uint32_t base_count)
52
53 {
54         off_t index_size = p->index_size;
55         const unsigned char *index_base = p->index_data;
56         git_hash_ctx ctx;
57         unsigned char hash[GIT_MAX_RAWSZ], *pack_sig;
58         off_t offset = 0, pack_sig_ofs = 0;
59         uint32_t nr_objects, i;
60         int err = 0;
61         struct idx_entry *entries;
62
63         if (!is_pack_valid(p))
64                 return error("packfile %s cannot be accessed", p->pack_name);
65
66         r->hash_algo->init_fn(&ctx);
67         do {
68                 unsigned long remaining;
69                 unsigned char *in = use_pack(p, w_curs, offset, &remaining);
70                 offset += remaining;
71                 if (!pack_sig_ofs)
72                         pack_sig_ofs = p->pack_size - r->hash_algo->rawsz;
73                 if (offset > pack_sig_ofs)
74                         remaining -= (unsigned int)(offset - pack_sig_ofs);
75                 r->hash_algo->update_fn(&ctx, in, remaining);
76         } while (offset < pack_sig_ofs);
77         r->hash_algo->final_fn(hash, &ctx);
78         pack_sig = use_pack(p, w_curs, pack_sig_ofs, NULL);
79         if (!hasheq(hash, pack_sig))
80                 err = error("%s pack checksum mismatch",
81                             p->pack_name);
82         if (!hasheq(index_base + index_size - r->hash_algo->hexsz, pack_sig))
83                 err = error("%s pack checksum does not match its index",
84                             p->pack_name);
85         unuse_pack(w_curs);
86
87         /* Make sure everything reachable from idx is valid.  Since we
88          * have verified that nr_objects matches between idx and pack,
89          * we do not do scan-streaming check on the pack file.
90          */
91         nr_objects = p->num_objects;
92         ALLOC_ARRAY(entries, nr_objects + 1);
93         entries[nr_objects].offset = pack_sig_ofs;
94         /* first sort entries by pack offset, since unpacking them is more efficient that way */
95         for (i = 0; i < nr_objects; i++) {
96                 entries[i].offset = nth_packed_object_offset(p, i);
97                 entries[i].nr = i;
98         }
99         QSORT(entries, nr_objects, compare_entries);
100
101         for (i = 0; i < nr_objects; i++) {
102                 void *data;
103                 struct object_id oid;
104                 enum object_type type;
105                 unsigned long size;
106                 off_t curpos;
107                 int data_valid;
108
109                 if (nth_packed_object_id(&oid, p, entries[i].nr) < 0)
110                         BUG("unable to get oid of object %lu from %s",
111                             (unsigned long)entries[i].nr, p->pack_name);
112
113                 if (p->index_version > 1) {
114                         off_t offset = entries[i].offset;
115                         off_t len = entries[i+1].offset - offset;
116                         unsigned int nr = entries[i].nr;
117                         if (check_pack_crc(p, w_curs, offset, len, nr))
118                                 err = error("index CRC mismatch for object %s "
119                                             "from %s at offset %"PRIuMAX"",
120                                             oid_to_hex(&oid),
121                                             p->pack_name, (uintmax_t)offset);
122                 }
123
124                 curpos = entries[i].offset;
125                 type = unpack_object_header(p, w_curs, &curpos, &size);
126                 unuse_pack(w_curs);
127
128                 if (type == OBJ_BLOB && big_file_threshold <= size) {
129                         /*
130                          * Let check_object_signature() check it with
131                          * the streaming interface; no point slurping
132                          * the data in-core only to discard.
133                          */
134                         data = NULL;
135                         data_valid = 0;
136                 } else {
137                         data = unpack_entry(r, p, entries[i].offset, &type, &size);
138                         data_valid = 1;
139                 }
140
141                 if (data_valid && !data)
142                         err = error("cannot unpack %s from %s at offset %"PRIuMAX"",
143                                     oid_to_hex(&oid), p->pack_name,
144                                     (uintmax_t)entries[i].offset);
145                 else if (check_object_signature(r, &oid, data, size, type_name(type)))
146                         err = error("packed %s from %s is corrupt",
147                                     oid_to_hex(&oid), p->pack_name);
148                 else if (fn) {
149                         int eaten = 0;
150                         err |= fn(&oid, type, size, data, &eaten);
151                         if (eaten)
152                                 data = NULL;
153                 }
154                 if (((base_count + i) & 1023) == 0)
155                         display_progress(progress, base_count + i);
156                 free(data);
157
158         }
159         display_progress(progress, base_count + i);
160         free(entries);
161
162         return err;
163 }
164
165 int verify_pack_index(struct packed_git *p)
166 {
167         off_t index_size;
168         const unsigned char *index_base;
169         git_hash_ctx ctx;
170         unsigned char hash[GIT_MAX_RAWSZ];
171         int err = 0;
172
173         if (open_pack_index(p))
174                 return error("packfile %s index not opened", p->pack_name);
175         index_size = p->index_size;
176         index_base = p->index_data;
177
178         /* Verify SHA1 sum of the index file */
179         the_hash_algo->init_fn(&ctx);
180         the_hash_algo->update_fn(&ctx, index_base, (unsigned int)(index_size - the_hash_algo->rawsz));
181         the_hash_algo->final_fn(hash, &ctx);
182         if (!hasheq(hash, index_base + index_size - the_hash_algo->rawsz))
183                 err = error("Packfile index for %s hash mismatch",
184                             p->pack_name);
185         return err;
186 }
187
188 int verify_pack(struct repository *r, struct packed_git *p, verify_fn fn,
189                 struct progress *progress, uint32_t base_count)
190 {
191         int err = 0;
192         struct pack_window *w_curs = NULL;
193
194         err |= verify_pack_index(p);
195         if (!p->index_data)
196                 return -1;
197
198         err |= verify_packfile(r, p, &w_curs, fn, progress, base_count);
199         unuse_pack(&w_curs);
200
201         return err;
202 }