1 /*
2 * Copyright (C) 2007 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 //
18 // Read-only access to Zip archives, with minimal heap allocation.
19 //
20 #define LOG_TAG "zipro"
21 //#define LOG_NDEBUG 0
22 #include <utils/Log.h>
23 #include <utils/Compat.h>
24 #include <utils/ZipFileRO.h>
25 #include <utils/misc.h>
26 #include <utils/threads.h>
27
28 #include <zlib.h>
29
30 #include <string.h>
31 #include <fcntl.h>
32 #include <errno.h>
33 #include <assert.h>
34 #include <unistd.h>
35
36 /*
37 * We must open binary files using open(path, ... | O_BINARY) under Windows.
38 * Otherwise strange read errors will happen.
39 */
40 #ifndef O_BINARY
41 # define O_BINARY 0
42 #endif
43
44 using namespace android;
45
46 /*
47 * Zip file constants.
48 */
49 #define kEOCDSignature 0x06054b50
50 #define kEOCDLen 22
51 #define kEOCDNumEntries 8 // offset to #of entries in file
52 #define kEOCDSize 12 // size of the central directory
53 #define kEOCDFileOffset 16 // offset to central directory
54
55 #define kMaxCommentLen 65535 // longest possible in ushort
56 #define kMaxEOCDSearch (kMaxCommentLen + kEOCDLen)
57
58 #define kLFHSignature 0x04034b50
59 #define kLFHLen 30 // excluding variable-len fields
60 #define kLFHNameLen 26 // offset to filename length
61 #define kLFHExtraLen 28 // offset to extra length
62
63 #define kCDESignature 0x02014b50
64 #define kCDELen 46 // excluding variable-len fields
65 #define kCDEMethod 10 // offset to compression method
66 #define kCDEModWhen 12 // offset to modification timestamp
67 #define kCDECRC 16 // offset to entry CRC
68 #define kCDECompLen 20 // offset to compressed length
69 #define kCDEUncompLen 24 // offset to uncompressed length
70 #define kCDENameLen 28 // offset to filename length
71 #define kCDEExtraLen 30 // offset to extra length
72 #define kCDECommentLen 32 // offset to comment length
73 #define kCDELocalOffset 42 // offset to local hdr
74
75 /*
76 * The values we return for ZipEntryRO use 0 as an invalid value, so we
77 * want to adjust the hash table index by a fixed amount. Using a large
78 * value helps insure that people don't mix & match arguments, e.g. to
79 * findEntryByIndex().
80 */
81 #define kZipEntryAdj 10000
82
~ZipFileRO()83 ZipFileRO::~ZipFileRO() {
84 free(mHashTable);
85 if (mDirectoryMap)
86 mDirectoryMap->release();
87 if (mFd >= 0)
88 TEMP_FAILURE_RETRY(close(mFd));
89 if (mFileName)
90 free(mFileName);
91 }
92
93 /*
94 * Convert a ZipEntryRO to a hash table index, verifying that it's in a
95 * valid range.
96 */
entryToIndex(const ZipEntryRO entry) const97 int ZipFileRO::entryToIndex(const ZipEntryRO entry) const
98 {
99 long ent = ((intptr_t) entry) - kZipEntryAdj;
100 if (ent < 0 || ent >= mHashTableSize || mHashTable[ent].name == NULL) {
101 ALOGW("Invalid ZipEntryRO %p (%ld)\n", entry, ent);
102 return -1;
103 }
104 return ent;
105 }
106
107
108 /*
109 * Open the specified file read-only. We memory-map the entire thing and
110 * close the file before returning.
111 */
open(const char * zipFileName)112 status_t ZipFileRO::open(const char* zipFileName)
113 {
114 int fd = -1;
115
116 assert(mDirectoryMap == NULL);
117
118 /*
119 * Open and map the specified file.
120 */
121 fd = TEMP_FAILURE_RETRY(::open(zipFileName, O_RDONLY | O_BINARY));
122 if (fd < 0) {
123 ALOGW("Unable to open zip '%s': %s\n", zipFileName, strerror(errno));
124 return NAME_NOT_FOUND;
125 }
126
127 mFileLength = lseek64(fd, 0, SEEK_END);
128 if (mFileLength < kEOCDLen) {
129 TEMP_FAILURE_RETRY(close(fd));
130 return UNKNOWN_ERROR;
131 }
132
133 if (mFileName != NULL) {
134 free(mFileName);
135 }
136 mFileName = strdup(zipFileName);
137
138 mFd = fd;
139
140 /*
141 * Find the Central Directory and store its size and number of entries.
142 */
143 if (!mapCentralDirectory()) {
144 goto bail;
145 }
146
147 /*
148 * Verify Central Directory and create data structures for fast access.
149 */
150 if (!parseZipArchive()) {
151 goto bail;
152 }
153
154 return OK;
155
156 bail:
157 free(mFileName);
158 mFileName = NULL;
159 TEMP_FAILURE_RETRY(close(fd));
160 return UNKNOWN_ERROR;
161 }
162
163 /*
164 * Parse the Zip archive, verifying its contents and initializing internal
165 * data structures.
166 */
mapCentralDirectory(void)167 bool ZipFileRO::mapCentralDirectory(void)
168 {
169 ssize_t readAmount = kMaxEOCDSearch;
170 if (readAmount > (ssize_t) mFileLength)
171 readAmount = mFileLength;
172
173 unsigned char* scanBuf = (unsigned char*) malloc(readAmount);
174 if (scanBuf == NULL) {
175 ALOGW("couldn't allocate scanBuf: %s", strerror(errno));
176 free(scanBuf);
177 return false;
178 }
179
180 /*
181 * Make sure this is a Zip archive.
182 */
183 if (lseek64(mFd, 0, SEEK_SET) != 0) {
184 ALOGW("seek to start failed: %s", strerror(errno));
185 free(scanBuf);
186 return false;
187 }
188
189 ssize_t actual = TEMP_FAILURE_RETRY(read(mFd, scanBuf, sizeof(int32_t)));
190 if (actual != (ssize_t) sizeof(int32_t)) {
191 ALOGI("couldn't read first signature from zip archive: %s", strerror(errno));
192 free(scanBuf);
193 return false;
194 }
195
196 {
197 unsigned int header = get4LE(scanBuf);
198 if (header == kEOCDSignature) {
199 ALOGI("Found Zip archive, but it looks empty\n");
200 free(scanBuf);
201 return false;
202 } else if (header != kLFHSignature) {
203 ALOGV("Not a Zip archive (found 0x%08x)\n", header);
204 free(scanBuf);
205 return false;
206 }
207 }
208
209 /*
210 * Perform the traditional EOCD snipe hunt.
211 *
212 * We're searching for the End of Central Directory magic number,
213 * which appears at the start of the EOCD block. It's followed by
214 * 18 bytes of EOCD stuff and up to 64KB of archive comment. We
215 * need to read the last part of the file into a buffer, dig through
216 * it to find the magic number, parse some values out, and use those
217 * to determine the extent of the CD.
218 *
219 * We start by pulling in the last part of the file.
220 */
221 off64_t searchStart = mFileLength - readAmount;
222
223 if (lseek64(mFd, searchStart, SEEK_SET) != searchStart) {
224 ALOGW("seek %ld failed: %s\n", (long) searchStart, strerror(errno));
225 free(scanBuf);
226 return false;
227 }
228 actual = TEMP_FAILURE_RETRY(read(mFd, scanBuf, readAmount));
229 if (actual != (ssize_t) readAmount) {
230 ALOGW("Zip: read " ZD ", expected " ZD ". Failed: %s\n",
231 (ZD_TYPE) actual, (ZD_TYPE) readAmount, strerror(errno));
232 free(scanBuf);
233 return false;
234 }
235
236 /*
237 * Scan backward for the EOCD magic. In an archive without a trailing
238 * comment, we'll find it on the first try. (We may want to consider
239 * doing an initial minimal read; if we don't find it, retry with a
240 * second read as above.)
241 */
242 int i;
243 for (i = readAmount - kEOCDLen; i >= 0; i--) {
244 if (scanBuf[i] == 0x50 && get4LE(&scanBuf[i]) == kEOCDSignature) {
245 ALOGV("+++ Found EOCD at buf+%d\n", i);
246 break;
247 }
248 }
249 if (i < 0) {
250 ALOGD("Zip: EOCD not found, %s is not zip\n", mFileName);
251 free(scanBuf);
252 return false;
253 }
254
255 off64_t eocdOffset = searchStart + i;
256 const unsigned char* eocdPtr = scanBuf + i;
257
258 assert(eocdOffset < mFileLength);
259
260 /*
261 * Grab the CD offset and size, and the number of entries in the
262 * archive. After that, we can release our EOCD hunt buffer.
263 */
264 unsigned int numEntries = get2LE(eocdPtr + kEOCDNumEntries);
265 unsigned int dirSize = get4LE(eocdPtr + kEOCDSize);
266 unsigned int dirOffset = get4LE(eocdPtr + kEOCDFileOffset);
267 free(scanBuf);
268
269 // Verify that they look reasonable.
270 if ((long long) dirOffset + (long long) dirSize > (long long) eocdOffset) {
271 ALOGW("bad offsets (dir %ld, size %u, eocd %ld)\n",
272 (long) dirOffset, dirSize, (long) eocdOffset);
273 return false;
274 }
275 if (numEntries == 0) {
276 ALOGW("empty archive?\n");
277 return false;
278 }
279
280 ALOGV("+++ numEntries=%d dirSize=%d dirOffset=%d\n",
281 numEntries, dirSize, dirOffset);
282
283 mDirectoryMap = new FileMap();
284 if (mDirectoryMap == NULL) {
285 ALOGW("Unable to create directory map: %s", strerror(errno));
286 return false;
287 }
288
289 if (!mDirectoryMap->create(mFileName, mFd, dirOffset, dirSize, true)) {
290 ALOGW("Unable to map '%s' (" ZD " to " ZD "): %s\n", mFileName,
291 (ZD_TYPE) dirOffset, (ZD_TYPE) (dirOffset + dirSize), strerror(errno));
292 return false;
293 }
294
295 mNumEntries = numEntries;
296 mDirectoryOffset = dirOffset;
297
298 return true;
299 }
300
301
302 /*
303 * Round up to the next highest power of 2.
304 *
305 * Found on http://graphics.stanford.edu/~seander/bithacks.html.
306 */
roundUpPower2(unsigned int val)307 static unsigned int roundUpPower2(unsigned int val)
308 {
309 val--;
310 val |= val >> 1;
311 val |= val >> 2;
312 val |= val >> 4;
313 val |= val >> 8;
314 val |= val >> 16;
315 val++;
316
317 return val;
318 }
319
parseZipArchive(void)320 bool ZipFileRO::parseZipArchive(void)
321 {
322 bool result = false;
323 const unsigned char* cdPtr = (const unsigned char*) mDirectoryMap->getDataPtr();
324 size_t cdLength = mDirectoryMap->getDataLength();
325 int numEntries = mNumEntries;
326
327 /*
328 * Create hash table. We have a minimum 75% load factor, possibly as
329 * low as 50% after we round off to a power of 2.
330 */
331 mHashTableSize = roundUpPower2(1 + (numEntries * 4) / 3);
332 mHashTable = (HashEntry*) calloc(mHashTableSize, sizeof(HashEntry));
333
334 /*
335 * Walk through the central directory, adding entries to the hash
336 * table.
337 */
338 const unsigned char* ptr = cdPtr;
339 for (int i = 0; i < numEntries; i++) {
340 if (get4LE(ptr) != kCDESignature) {
341 ALOGW("Missed a central dir sig (at %d)\n", i);
342 goto bail;
343 }
344 if (ptr + kCDELen > cdPtr + cdLength) {
345 ALOGW("Ran off the end (at %d)\n", i);
346 goto bail;
347 }
348
349 long localHdrOffset = (long) get4LE(ptr + kCDELocalOffset);
350 if (localHdrOffset >= mDirectoryOffset) {
351 ALOGW("bad LFH offset %ld at entry %d\n", localHdrOffset, i);
352 goto bail;
353 }
354
355 unsigned int fileNameLen, extraLen, commentLen, hash;
356
357 fileNameLen = get2LE(ptr + kCDENameLen);
358 extraLen = get2LE(ptr + kCDEExtraLen);
359 commentLen = get2LE(ptr + kCDECommentLen);
360
361 /* add the CDE filename to the hash table */
362 hash = computeHash((const char*)ptr + kCDELen, fileNameLen);
363 addToHash((const char*)ptr + kCDELen, fileNameLen, hash);
364
365 ptr += kCDELen + fileNameLen + extraLen + commentLen;
366 if ((size_t)(ptr - cdPtr) > cdLength) {
367 ALOGW("bad CD advance (%d vs " ZD ") at entry %d\n",
368 (int) (ptr - cdPtr), (ZD_TYPE) cdLength, i);
369 goto bail;
370 }
371 }
372 ALOGV("+++ zip good scan %d entries\n", numEntries);
373 result = true;
374
375 bail:
376 return result;
377 }
378
379 /*
380 * Simple string hash function for non-null-terminated strings.
381 */
computeHash(const char * str,int len)382 /*static*/ unsigned int ZipFileRO::computeHash(const char* str, int len)
383 {
384 unsigned int hash = 0;
385
386 while (len--)
387 hash = hash * 31 + *str++;
388
389 return hash;
390 }
391
392 /*
393 * Add a new entry to the hash table.
394 */
addToHash(const char * str,int strLen,unsigned int hash)395 void ZipFileRO::addToHash(const char* str, int strLen, unsigned int hash)
396 {
397 int ent = hash & (mHashTableSize-1);
398
399 /*
400 * We over-allocate the table, so we're guaranteed to find an empty slot.
401 */
402 while (mHashTable[ent].name != NULL)
403 ent = (ent + 1) & (mHashTableSize-1);
404
405 mHashTable[ent].name = str;
406 mHashTable[ent].nameLen = strLen;
407 }
408
409 /*
410 * Find a matching entry.
411 *
412 * Returns NULL if not found.
413 */
findEntryByName(const char * fileName) const414 ZipEntryRO ZipFileRO::findEntryByName(const char* fileName) const
415 {
416 /*
417 * If the ZipFileRO instance is not initialized, the entry number will
418 * end up being garbage since mHashTableSize is -1.
419 */
420 if (mHashTableSize <= 0) {
421 return NULL;
422 }
423
424 int nameLen = strlen(fileName);
425 unsigned int hash = computeHash(fileName, nameLen);
426 int ent = hash & (mHashTableSize-1);
427
428 while (mHashTable[ent].name != NULL) {
429 if (mHashTable[ent].nameLen == nameLen &&
430 memcmp(mHashTable[ent].name, fileName, nameLen) == 0)
431 {
432 /* match */
433 return (ZipEntryRO)(long)(ent + kZipEntryAdj);
434 }
435
436 ent = (ent + 1) & (mHashTableSize-1);
437 }
438
439 return NULL;
440 }
441
442 /*
443 * Find the Nth entry.
444 *
445 * This currently involves walking through the sparse hash table, counting
446 * non-empty entries. If we need to speed this up we can either allocate
447 * a parallel lookup table or (perhaps better) provide an iterator interface.
448 */
findEntryByIndex(int idx) const449 ZipEntryRO ZipFileRO::findEntryByIndex(int idx) const
450 {
451 if (idx < 0 || idx >= mNumEntries) {
452 ALOGW("Invalid index %d\n", idx);
453 return NULL;
454 }
455
456 for (int ent = 0; ent < mHashTableSize; ent++) {
457 if (mHashTable[ent].name != NULL) {
458 if (idx-- == 0)
459 return (ZipEntryRO) (intptr_t)(ent + kZipEntryAdj);
460 }
461 }
462
463 return NULL;
464 }
465
466 /*
467 * Get the useful fields from the zip entry.
468 *
469 * Returns "false" if the offsets to the fields or the contents of the fields
470 * appear to be bogus.
471 */
getEntryInfo(ZipEntryRO entry,int * pMethod,size_t * pUncompLen,size_t * pCompLen,off64_t * pOffset,long * pModWhen,long * pCrc32) const472 bool ZipFileRO::getEntryInfo(ZipEntryRO entry, int* pMethod, size_t* pUncompLen,
473 size_t* pCompLen, off64_t* pOffset, long* pModWhen, long* pCrc32) const
474 {
475 bool ret = false;
476
477 const int ent = entryToIndex(entry);
478 if (ent < 0)
479 return false;
480
481 HashEntry hashEntry = mHashTable[ent];
482
483 /*
484 * Recover the start of the central directory entry from the filename
485 * pointer. The filename is the first entry past the fixed-size data,
486 * so we can just subtract back from that.
487 */
488 const unsigned char* ptr = (const unsigned char*) hashEntry.name;
489 off64_t cdOffset = mDirectoryOffset;
490
491 ptr -= kCDELen;
492
493 int method = get2LE(ptr + kCDEMethod);
494 if (pMethod != NULL)
495 *pMethod = method;
496
497 if (pModWhen != NULL)
498 *pModWhen = get4LE(ptr + kCDEModWhen);
499 if (pCrc32 != NULL)
500 *pCrc32 = get4LE(ptr + kCDECRC);
501
502 size_t compLen = get4LE(ptr + kCDECompLen);
503 if (pCompLen != NULL)
504 *pCompLen = compLen;
505 size_t uncompLen = get4LE(ptr + kCDEUncompLen);
506 if (pUncompLen != NULL)
507 *pUncompLen = uncompLen;
508
509 /*
510 * If requested, determine the offset of the start of the data. All we
511 * have is the offset to the Local File Header, which is variable size,
512 * so we have to read the contents of the struct to figure out where
513 * the actual data starts.
514 *
515 * We also need to make sure that the lengths are not so large that
516 * somebody trying to map the compressed or uncompressed data runs
517 * off the end of the mapped region.
518 *
519 * Note we don't verify compLen/uncompLen if they don't request the
520 * dataOffset, because dataOffset is expensive to determine. However,
521 * if they don't have the file offset, they're not likely to be doing
522 * anything with the contents.
523 */
524 if (pOffset != NULL) {
525 long localHdrOffset = get4LE(ptr + kCDELocalOffset);
526 if (localHdrOffset + kLFHLen >= cdOffset) {
527 ALOGE("ERROR: bad local hdr offset in zip\n");
528 return false;
529 }
530
531 unsigned char lfhBuf[kLFHLen];
532
533 #ifdef HAVE_PREAD
534 /*
535 * This file descriptor might be from zygote's preloaded assets,
536 * so we need to do an pread64() instead of a lseek64() + read() to
537 * guarantee atomicity across the processes with the shared file
538 * descriptors.
539 */
540 ssize_t actual =
541 TEMP_FAILURE_RETRY(pread64(mFd, lfhBuf, sizeof(lfhBuf), localHdrOffset));
542
543 if (actual != sizeof(lfhBuf)) {
544 ALOGW("failed reading lfh from offset %ld\n", localHdrOffset);
545 return false;
546 }
547
548 if (get4LE(lfhBuf) != kLFHSignature) {
549 ALOGW("didn't find signature at start of lfh; wanted: offset=%ld data=0x%08x; "
550 "got: data=0x%08lx\n",
551 localHdrOffset, kLFHSignature, get4LE(lfhBuf));
552 return false;
553 }
554 #else /* HAVE_PREAD */
555 /*
556 * For hosts don't have pread64() we cannot guarantee atomic reads from
557 * an offset in a file. Android should never run on those platforms.
558 * File descriptors inherited from a fork() share file offsets and
559 * there would be nothing to protect from two different processes
560 * calling lseek64() concurrently.
561 */
562
563 {
564 AutoMutex _l(mFdLock);
565
566 if (lseek64(mFd, localHdrOffset, SEEK_SET) != localHdrOffset) {
567 ALOGW("failed seeking to lfh at offset %ld\n", localHdrOffset);
568 return false;
569 }
570
571 ssize_t actual =
572 TEMP_FAILURE_RETRY(read(mFd, lfhBuf, sizeof(lfhBuf)));
573 if (actual != sizeof(lfhBuf)) {
574 ALOGW("failed reading lfh from offset %ld\n", localHdrOffset);
575 return false;
576 }
577
578 if (get4LE(lfhBuf) != kLFHSignature) {
579 off64_t actualOffset = lseek64(mFd, 0, SEEK_CUR);
580 ALOGW("didn't find signature at start of lfh; wanted: offset=%ld data=0x%08x; "
581 "got: offset=" ZD " data=0x%08lx\n",
582 localHdrOffset, kLFHSignature, (ZD_TYPE) actualOffset, get4LE(lfhBuf));
583 return false;
584 }
585 }
586 #endif /* HAVE_PREAD */
587
588 off64_t dataOffset = localHdrOffset + kLFHLen
589 + get2LE(lfhBuf + kLFHNameLen) + get2LE(lfhBuf + kLFHExtraLen);
590 if (dataOffset >= cdOffset) {
591 ALOGW("bad data offset %ld in zip\n", (long) dataOffset);
592 return false;
593 }
594
595 /* check lengths */
596 if ((off64_t)(dataOffset + compLen) > cdOffset) {
597 ALOGW("bad compressed length in zip (%ld + " ZD " > %ld)\n",
598 (long) dataOffset, (ZD_TYPE) compLen, (long) cdOffset);
599 return false;
600 }
601
602 if (method == kCompressStored &&
603 (off64_t)(dataOffset + uncompLen) > cdOffset)
604 {
605 ALOGE("ERROR: bad uncompressed length in zip (%ld + " ZD " > %ld)\n",
606 (long) dataOffset, (ZD_TYPE) uncompLen, (long) cdOffset);
607 return false;
608 }
609
610 *pOffset = dataOffset;
611 }
612
613 return true;
614 }
615
616 /*
617 * Copy the entry's filename to the buffer.
618 */
getEntryFileName(ZipEntryRO entry,char * buffer,int bufLen) const619 int ZipFileRO::getEntryFileName(ZipEntryRO entry, char* buffer, int bufLen)
620 const
621 {
622 int ent = entryToIndex(entry);
623 if (ent < 0)
624 return -1;
625
626 int nameLen = mHashTable[ent].nameLen;
627 if (bufLen < nameLen+1)
628 return nameLen+1;
629
630 memcpy(buffer, mHashTable[ent].name, nameLen);
631 buffer[nameLen] = '\0';
632 return 0;
633 }
634
635 /*
636 * Create a new FileMap object that spans the data in "entry".
637 */
createEntryFileMap(ZipEntryRO entry) const638 FileMap* ZipFileRO::createEntryFileMap(ZipEntryRO entry) const
639 {
640 /*
641 * TODO: the efficient way to do this is to modify FileMap to allow
642 * sub-regions of a file to be mapped. A reference-counting scheme
643 * can manage the base memory mapping. For now, we just create a brand
644 * new mapping off of the Zip archive file descriptor.
645 */
646
647 FileMap* newMap;
648 size_t compLen;
649 off64_t offset;
650
651 if (!getEntryInfo(entry, NULL, NULL, &compLen, &offset, NULL, NULL))
652 return NULL;
653
654 newMap = new FileMap();
655 if (!newMap->create(mFileName, mFd, offset, compLen, true)) {
656 newMap->release();
657 return NULL;
658 }
659
660 return newMap;
661 }
662
663 /*
664 * Uncompress an entry, in its entirety, into the provided output buffer.
665 *
666 * This doesn't verify the data's CRC, which might be useful for
667 * uncompressed data. The caller should be able to manage it.
668 */
uncompressEntry(ZipEntryRO entry,void * buffer) const669 bool ZipFileRO::uncompressEntry(ZipEntryRO entry, void* buffer) const
670 {
671 const size_t kSequentialMin = 32768;
672 bool result = false;
673 int ent = entryToIndex(entry);
674 if (ent < 0)
675 return -1;
676
677 int method;
678 size_t uncompLen, compLen;
679 off64_t offset;
680 const unsigned char* ptr;
681
682 getEntryInfo(entry, &method, &uncompLen, &compLen, &offset, NULL, NULL);
683
684 FileMap* file = createEntryFileMap(entry);
685 if (file == NULL) {
686 goto bail;
687 }
688
689 ptr = (const unsigned char*) file->getDataPtr();
690
691 /*
692 * Experiment with madvise hint. When we want to uncompress a file,
693 * we pull some stuff out of the central dir entry and then hit a
694 * bunch of compressed or uncompressed data sequentially. The CDE
695 * visit will cause a limited amount of read-ahead because it's at
696 * the end of the file. We could end up doing lots of extra disk
697 * access if the file we're prying open is small. Bottom line is we
698 * probably don't want to turn MADV_SEQUENTIAL on and leave it on.
699 *
700 * So, if the compressed size of the file is above a certain minimum
701 * size, temporarily boost the read-ahead in the hope that the extra
702 * pair of system calls are negated by a reduction in page faults.
703 */
704 if (compLen > kSequentialMin)
705 file->advise(FileMap::SEQUENTIAL);
706
707 if (method == kCompressStored) {
708 memcpy(buffer, ptr, uncompLen);
709 } else {
710 if (!inflateBuffer(buffer, ptr, uncompLen, compLen))
711 goto unmap;
712 }
713
714 if (compLen > kSequentialMin)
715 file->advise(FileMap::NORMAL);
716
717 result = true;
718
719 unmap:
720 file->release();
721 bail:
722 return result;
723 }
724
725 /*
726 * Uncompress an entry, in its entirety, to an open file descriptor.
727 *
728 * This doesn't verify the data's CRC, but probably should.
729 */
uncompressEntry(ZipEntryRO entry,int fd) const730 bool ZipFileRO::uncompressEntry(ZipEntryRO entry, int fd) const
731 {
732 bool result = false;
733 int ent = entryToIndex(entry);
734 if (ent < 0)
735 return -1;
736
737 int method;
738 size_t uncompLen, compLen;
739 off64_t offset;
740 const unsigned char* ptr;
741
742 getEntryInfo(entry, &method, &uncompLen, &compLen, &offset, NULL, NULL);
743
744 FileMap* file = createEntryFileMap(entry);
745 if (file == NULL) {
746 goto bail;
747 }
748
749 ptr = (const unsigned char*) file->getDataPtr();
750
751 if (method == kCompressStored) {
752 ssize_t actual = TEMP_FAILURE_RETRY(write(fd, ptr, uncompLen));
753 if (actual < 0) {
754 ALOGE("Write failed: %s\n", strerror(errno));
755 goto unmap;
756 } else if ((size_t) actual != uncompLen) {
757 ALOGE("Partial write during uncompress (" ZD " of " ZD ")\n",
758 (ZD_TYPE) actual, (ZD_TYPE) uncompLen);
759 goto unmap;
760 } else {
761 ALOGI("+++ successful write\n");
762 }
763 } else {
764 if (!inflateBuffer(fd, ptr, uncompLen, compLen))
765 goto unmap;
766 }
767
768 result = true;
769
770 unmap:
771 file->release();
772 bail:
773 return result;
774 }
775
776 /*
777 * Uncompress "deflate" data from one buffer to another.
778 */
inflateBuffer(void * outBuf,const void * inBuf,size_t uncompLen,size_t compLen)779 /*static*/ bool ZipFileRO::inflateBuffer(void* outBuf, const void* inBuf,
780 size_t uncompLen, size_t compLen)
781 {
782 bool result = false;
783 z_stream zstream;
784 int zerr;
785
786 /*
787 * Initialize the zlib stream struct.
788 */
789 memset(&zstream, 0, sizeof(zstream));
790 zstream.zalloc = Z_NULL;
791 zstream.zfree = Z_NULL;
792 zstream.opaque = Z_NULL;
793 zstream.next_in = (Bytef*)inBuf;
794 zstream.avail_in = compLen;
795 zstream.next_out = (Bytef*) outBuf;
796 zstream.avail_out = uncompLen;
797 zstream.data_type = Z_UNKNOWN;
798
799 /*
800 * Use the undocumented "negative window bits" feature to tell zlib
801 * that there's no zlib header waiting for it.
802 */
803 zerr = inflateInit2(&zstream, -MAX_WBITS);
804 if (zerr != Z_OK) {
805 if (zerr == Z_VERSION_ERROR) {
806 ALOGE("Installed zlib is not compatible with linked version (%s)\n",
807 ZLIB_VERSION);
808 } else {
809 ALOGE("Call to inflateInit2 failed (zerr=%d)\n", zerr);
810 }
811 goto bail;
812 }
813
814 /*
815 * Expand data.
816 */
817 zerr = inflate(&zstream, Z_FINISH);
818 if (zerr != Z_STREAM_END) {
819 ALOGW("Zip inflate failed, zerr=%d (nIn=%p aIn=%u nOut=%p aOut=%u)\n",
820 zerr, zstream.next_in, zstream.avail_in,
821 zstream.next_out, zstream.avail_out);
822 goto z_bail;
823 }
824
825 /* paranoia */
826 if (zstream.total_out != uncompLen) {
827 ALOGW("Size mismatch on inflated file (%ld vs " ZD ")\n",
828 zstream.total_out, (ZD_TYPE) uncompLen);
829 goto z_bail;
830 }
831
832 result = true;
833
834 z_bail:
835 inflateEnd(&zstream); /* free up any allocated structures */
836
837 bail:
838 return result;
839 }
840
841 /*
842 * Uncompress "deflate" data from one buffer to an open file descriptor.
843 */
inflateBuffer(int fd,const void * inBuf,size_t uncompLen,size_t compLen)844 /*static*/ bool ZipFileRO::inflateBuffer(int fd, const void* inBuf,
845 size_t uncompLen, size_t compLen)
846 {
847 bool result = false;
848 const size_t kWriteBufSize = 32768;
849 unsigned char writeBuf[kWriteBufSize];
850 z_stream zstream;
851 int zerr;
852
853 /*
854 * Initialize the zlib stream struct.
855 */
856 memset(&zstream, 0, sizeof(zstream));
857 zstream.zalloc = Z_NULL;
858 zstream.zfree = Z_NULL;
859 zstream.opaque = Z_NULL;
860 zstream.next_in = (Bytef*)inBuf;
861 zstream.avail_in = compLen;
862 zstream.next_out = (Bytef*) writeBuf;
863 zstream.avail_out = sizeof(writeBuf);
864 zstream.data_type = Z_UNKNOWN;
865
866 /*
867 * Use the undocumented "negative window bits" feature to tell zlib
868 * that there's no zlib header waiting for it.
869 */
870 zerr = inflateInit2(&zstream, -MAX_WBITS);
871 if (zerr != Z_OK) {
872 if (zerr == Z_VERSION_ERROR) {
873 ALOGE("Installed zlib is not compatible with linked version (%s)\n",
874 ZLIB_VERSION);
875 } else {
876 ALOGE("Call to inflateInit2 failed (zerr=%d)\n", zerr);
877 }
878 goto bail;
879 }
880
881 /*
882 * Loop while we have more to do.
883 */
884 do {
885 /*
886 * Expand data.
887 */
888 zerr = inflate(&zstream, Z_NO_FLUSH);
889 if (zerr != Z_OK && zerr != Z_STREAM_END) {
890 ALOGW("zlib inflate: zerr=%d (nIn=%p aIn=%u nOut=%p aOut=%u)\n",
891 zerr, zstream.next_in, zstream.avail_in,
892 zstream.next_out, zstream.avail_out);
893 goto z_bail;
894 }
895
896 /* write when we're full or when we're done */
897 if (zstream.avail_out == 0 ||
898 (zerr == Z_STREAM_END && zstream.avail_out != sizeof(writeBuf)))
899 {
900 long writeSize = zstream.next_out - writeBuf;
901 int cc = TEMP_FAILURE_RETRY(write(fd, writeBuf, writeSize));
902 if (cc < 0) {
903 ALOGW("write failed in inflate: %s", strerror(errno));
904 goto z_bail;
905 } else if (cc != (int) writeSize) {
906 ALOGW("write failed in inflate (%d vs %ld)", cc, writeSize);
907 goto z_bail;
908 }
909
910 zstream.next_out = writeBuf;
911 zstream.avail_out = sizeof(writeBuf);
912 }
913 } while (zerr == Z_OK);
914
915 assert(zerr == Z_STREAM_END); /* other errors should've been caught */
916
917 /* paranoia */
918 if (zstream.total_out != uncompLen) {
919 ALOGW("Size mismatch on inflated file (%ld vs " ZD ")\n",
920 zstream.total_out, (ZD_TYPE) uncompLen);
921 goto z_bail;
922 }
923
924 result = true;
925
926 z_bail:
927 inflateEnd(&zstream); /* free up any allocated structures */
928
929 bail:
930 return result;
931 }
932