/*
* Checks to see if we have ZIP64 archive, and save
* the check for later use
*/
static int
haveZIP64(Byte *p) {
jlong cenlen, cenoff, centot;
cenlen = ENDSIZ(p);
cenoff = ENDOFF(p);
centot = ENDTOT(p);
zip64_present = (cenlen == ZIP64_MAGICVAL ||
cenoff == ZIP64_MAGICVAL ||
centot == ZIP64_MAGICCOUNT);
return zip64_present;
}
static jboolean verifyEND(jzfile *zip, jlong endpos, char *endbuf) {
/* ENDSIG matched, however the size of file comment in it does not
match the real size. One "common" cause for this problem is some
"extra" bytes are padded at the end of the zipfile.
Let's do some extra verification, we don't care about the performance
in this situation.
*/
jlong cenpos = endpos - ENDSIZ(endbuf);
jlong locpos = cenpos - ENDOFF(endbuf);
char buf[4];
return (cenpos >= 0 &&
locpos >= 0 &&
readFullyAt(zip->zfd, buf, sizeof(buf), cenpos) != -1 &&
GETSIG(buf) == CENSIG &&
readFullyAt(zip->zfd, buf, sizeof(buf), locpos) != -1 &&
GETSIG(buf) == LOCSIG);
}
/*
* Reads zip file central directory. Returns the file position of first
* CEN header, otherwise returns -1 if an error occured. If zip->msg != NULL
* then the error was a zip format error and zip->msg has the error text.
* Always pass in -1 for knownTotal; it's used for a recursive call.
*/
static jlong
readCEN(jzfile *zip, jint knownTotal)
{
/* Following are unsigned 32-bit */
jlong endpos, end64pos, cenpos, cenlen, cenoff;
/* Following are unsigned 16-bit */
jint total, tablelen, i, j;
unsigned char *cenbuf = NULL;
unsigned char *cenend;
unsigned char *cp;
#ifdef USE_MMAP
static jlong pagesize;
jlong offset;
#endif
unsigned char endbuf[ENDHDR];
jint endhdrlen = ENDHDR;
jzcell *entries;
jint *table;
/* Clear previous zip error */
zip->msg = NULL;
/* Get position of END header */
if ((endpos = findEND(zip, endbuf)) == -1)
return -1; /* no END header or system error */
if (endpos == 0) return 0; /* only END header present */
freeCEN(zip);
/* Get position and length of central directory */
cenlen = ENDSIZ(endbuf);
cenoff = ENDOFF(endbuf);
total = ENDTOT(endbuf);
if (cenlen == ZIP64_MAGICVAL || cenoff == ZIP64_MAGICVAL ||
total == ZIP64_MAGICCOUNT) {
unsigned char end64buf[ZIP64_ENDHDR];
if ((end64pos = findEND64(zip, end64buf, endpos)) != -1) {
cenlen = ZIP64_ENDSIZ(end64buf);
cenoff = ZIP64_ENDOFF(end64buf);
total = (jint)ZIP64_ENDTOT(end64buf);
endpos = end64pos;
endhdrlen = ZIP64_ENDHDR;
}
}
if (cenlen > endpos)
ZIP_FORMAT_ERROR("invalid END header (bad central directory size)");
cenpos = endpos - cenlen;
/* Get position of first local file (LOC) header, taking into
* account that there may be a stub prefixed to the zip file. */
zip->locpos = cenpos - cenoff;
if (zip->locpos < 0)
ZIP_FORMAT_ERROR("invalid END header (bad central directory offset)");
#ifdef USE_MMAP
if (zip->usemmap) {
/* On Solaris & Linux prior to JDK 6, we used to mmap the whole jar file to
* read the jar file contents. However, this greatly increased the perceived
* footprint numbers because the mmap'ed pages were adding into the totals shown
* by 'ps' and 'top'. We switched to mmaping only the central directory of jar
* file while calling 'read' to read the rest of jar file. Here are a list of
* reasons apart from above of why we are doing so:
* 1. Greatly reduces mmap overhead after startup complete;
* 2. Avoids dual path code maintainance;
* 3. Greatly reduces risk of address space (not virtual memory) exhaustion.
*/
if (pagesize == 0) {
pagesize = (jlong)sysconf(_SC_PAGESIZE);
if (pagesize == 0) goto Catch;
}
if (cenpos > pagesize) {
offset = cenpos & ~(pagesize - 1);
} else {
offset = 0;
}
/* When we are not calling recursively, knownTotal is -1. */
if (knownTotal == -1) {
void* mappedAddr;
/* Mmap the CEN and END part only. We have to figure
out the page size in order to make offset to be multiples of
page size.
*/
zip->mlen = cenpos - offset + cenlen + endhdrlen;
zip->offset = offset;
mappedAddr = mmap64(0, zip->mlen, PROT_READ, MAP_SHARED, zip->zfd, (off64_t) offset);
zip->maddr = (mappedAddr == (void*) MAP_FAILED) ? NULL :
(unsigned char*)mappedAddr;
if (zip->maddr == NULL) {
jio_fprintf(stderr, "mmap failed for CEN and END part of zip file\n");
goto Catch;
}
}
cenbuf = zip->maddr + cenpos - offset;
} else
#endif
{
if ((cenbuf = malloc((size_t) cenlen)) == NULL ||
(readFullyAt(zip->zfd, cenbuf, cenlen, cenpos) == -1))
goto Catch;
}
cenend = cenbuf + cenlen;
/* Initialize zip file data structures based on the total number
* of central directory entries as stored in ENDTOT. Since this
* is a 2-byte field, but we (and other zip implementations)
* support approx. 2**31 entries, we do not trust ENDTOT, but
* treat it only as a strong hint. When we call ourselves
* recursively, knownTotal will have the "true" value.
*
* Keep this path alive even with the Zip64 END support added, just
* for zip files that have more than 0xffff entries but don't have
* the Zip64 enabled.
*/
total = (knownTotal != -1) ? knownTotal : total;
entries = zip->entries = calloc(total, sizeof(entries[0]));
tablelen = zip->tablelen = ((total/2) | 1); // Odd -> fewer collisions
table = zip->table = malloc(tablelen * sizeof(table[0]));
if (entries == NULL || table == NULL) goto Catch;
for (j = 0; j < tablelen; j++)
table[j] = ZIP_ENDCHAIN;
/* Iterate through the entries in the central directory */
for (i = 0, cp = cenbuf; cp <= cenend - CENHDR; i++, cp += CENSIZE(cp)) {
/* Following are unsigned 16-bit */
jint method, nlen;
unsigned int hsh;
if (i >= total) {
/* This will only happen if the zip file has an incorrect
* ENDTOT field, which usually means it contains more than
* 65535 entries. */
cenpos = readCEN(zip, countCENHeaders(cenbuf, cenend));
goto Finally;
}
method = CENHOW(cp);
nlen = CENNAM(cp);
if (GETSIG(cp) != CENSIG)
ZIP_FORMAT_ERROR("invalid CEN header (bad signature)");
if (CENFLG(cp) & 1)
ZIP_FORMAT_ERROR("invalid CEN header (encrypted entry)");
if (method != STORED && method != DEFLATED)
ZIP_FORMAT_ERROR("invalid CEN header (bad compression method)");
if (cp + CENHDR + nlen > cenend)
ZIP_FORMAT_ERROR("invalid CEN header (bad header size)");
/* if the entry is metadata add it to our metadata names */
if (isMetaName((char *)cp+CENHDR, nlen))
if (addMetaName(zip, (char *)cp+CENHDR, nlen) != 0)
goto Catch;
/* Record the CEN offset and the name hash in our hash cell. */
entries[i].cenpos = cenpos + (cp - cenbuf);
entries[i].hash = hashN((char *)cp+CENHDR, nlen);
/* Add the entry to the hash table */
hsh = entries[i].hash % tablelen;
entries[i].next = table[hsh];
table[hsh] = i;
}
if (cp != cenend)
ZIP_FORMAT_ERROR("invalid CEN header (bad header size)");
zip->total = i;
goto Finally;
Catch:
freeCEN(zip);
cenpos = -1;
Finally:
#ifdef USE_MMAP
if (!zip->usemmap)
#endif
free(cenbuf);
return cenpos;
}
static int
find_file(int fd, zentry *entry, const char *file_name)
{
int bytes;
int res;
int entry_size;
int read_size;
int base_offset;
Byte *p;
Byte *bp;
Byte *buffer;
Byte locbuf[LOCHDR];
if ((buffer = (Byte*)malloc(BUFSIZE)) == NULL) {
return(-1);
}
p = buffer;
bp = buffer;
/*
* Read the END Header, which is the starting point for ZIP files.
* (Clearly designed to make writing a zip file easier than reading
* one. Now isn't that precious...)
*/
if ((base_offset = find_end(fd, bp)) == -1) {
free(buffer);
return (-1);
}
/*
* There is a historical, but undocumented, ability to allow for
* additional "stuff" to be prepended to the zip/jar file. It seems
* that this has been used to prepend an actual java launcher
* executable to the jar on Windows. Although this is just another
* form of statically linking a small piece of the JVM to the
* application, we choose to continue to support it. Note that no
* guarantees have been made (or should be made) to the customer that
* this will continue to work.
*
* Therefore, calculate the base offset of the zip file (within the
* expanded file) by assuming that the central directory is followed
* immediately by the end record.
*/
base_offset = base_offset - ENDSIZ(p) - ENDOFF(p);
/*
* The END Header indicates the start of the Central Directory
* Headers. Remember that the desired Central Directory Header (CEN)
* will almost always be the second one and the first one is a small
* directory entry ("META-INF/"). Keep the code optimized for
* that case.
*
* Begin by seeking to the beginning of the Central Directory and
* reading in the first buffer full of bits.
*/
if (lseek(fd, base_offset + ENDOFF(p), SEEK_SET) < (off_t)0) {
free(buffer);
return (-1);
}
if ((bytes = read(fd, bp, MINREAD)) < 0) {
free(buffer);
return (-1);
}
/*
* Loop through the Central Directory Headers. Note that a valid zip/jar
* must have an ENDHDR (with ENDSIG) after the Central Directory.
*/
while (GETSIG(p) == CENSIG) {
/*
* If a complete header isn't in the buffer, shift the contents
* of the buffer down and refill the buffer. Note that the check
* for "bytes < CENHDR" must be made before the test for the entire
* size of the header, because if bytes is less than CENHDR, the
* actual size of the header can't be determined. The addition of
* SIGSIZ guarantees that the next signature is also in the buffer
* for proper loop termination.
*/
if (bytes < CENHDR) {
p = memmove(bp, p, bytes);
if ((res = read(fd, bp + bytes, MINREAD)) <= 0) {
free(buffer);
return (-1);
}
bytes += res;
}
entry_size = CENHDR + CENNAM(p) + CENEXT(p) + CENCOM(p);
if (bytes < entry_size + SIGSIZ) {
if (p != bp)
p = memmove(bp, p, bytes);
read_size = entry_size - bytes + SIGSIZ;
read_size = (read_size < MINREAD) ? MINREAD : read_size;
if ((res = read(fd, bp + bytes, read_size)) <= 0) {
free(buffer);
return (-1);
}
bytes += res;
}
/*
* Check if the name is the droid we are looking for; the jar file
* manifest. If so, build the entry record from the data found in
* the header located and return success.
*/
if (CENNAM(p) == strlen(file_name) &&
memcmp((p + CENHDR), file_name, strlen(file_name)) == 0) {
if (lseek(fd, base_offset + CENOFF(p), SEEK_SET) < (off_t)0) {
free(buffer);
return (-1);
}
if (read(fd, locbuf, LOCHDR) < 0) {
free(buffer);
return (-1);
}
if (GETSIG(locbuf) != LOCSIG) {
free(buffer);
return (-1);
}
entry->isize = CENLEN(p);
entry->csize = CENSIZ(p);
entry->offset = base_offset + CENOFF(p) + LOCHDR +
LOCNAM(locbuf) + LOCEXT(locbuf);
entry->how = CENHOW(p);
free(buffer);
return (0);
}
/*
* Point to the next entry and decrement the count of valid remaining
* bytes.
*/
bytes -= entry_size;
p += entry_size;
}
free(buffer);
return (-1); /* Fell off the end the loop without a Manifest */
}
bool JarFileParser::find_end_of_central_header() {
DECLARE_STATIC_BUFFER(unsigned char, buffer, TMPBUFFERSIZE);
BufferedFile::Raw bf = buffered_file();
/* Get the length of the file */
const jint length = (int) bf().file_size();
/* Calculate the smallest possible offset for the end header. It
* can be at most 0xFFFF + ENDHDRSIZ bytes from the end of the file, but
* the file must also have a local header and a central header
*/
jint minOffset = length - (0xFFFF + ENDHDRSIZ);
if (minOffset < LOCHDRSIZ + CENHDRSIZ) {
minOffset = LOCHDRSIZ + CENHDRSIZ;
}
/* We assume that "buffer" contains the contents
* of part of the file. currentOffset contains the offset of buffer[0].
*/
/* Read in the last ENDHDRSIZ bytes into the buffer. 99% of the time,
* the file won't have a comment, and this is the only read we'll need */
if ( (bf().seek(-ENDHDRSIZ, SEEK_END) < 0)
|| (bf().get_bytes(buffer, ENDHDRSIZ) != ENDHDRSIZ)) {
return false;
}
/* Set currentOffset to be the offset of buffer[0] */
jint currentOffset = length - ENDHDRSIZ;
/* Set bp to be the location at which to start looking */
unsigned const char* bp = buffer;
for (;;) {
/* "buffer" contains a block of data from the file, starting at
* currentOffset "position" in the file.
* We investigate whether currentOffset + (bp - buffer) is the start
* of the end header in the zip file.
*
* We use a simplified version of Knuth-Morris-Pratt search algorithm.
* The header we're looking for is 'P' 'K' 5 6
*/
switch(bp[0]) {
case '\006': /* The header must start at least 3 bytes back */
bp -= 3; break;
case '\005': /* The header must start at least 2 bytes back */
bp -= 2; break;
case 'K': /* The header must start at least 1 byte back */
bp -= 1; break;
case 'P': /* Either this is the header, or the header must
* start at least 4 back */
if (bp[1] == 'K' && bp[2] == 5 && bp[3] == 6) {
/* We have what may be a header. Let's make sure the
* implied length of the jar file matches the actual
* length.
*/
int endpos = (int) currentOffset + (bp - buffer);
if (endpos + ENDHDRSIZ + ENDCOM(bp) == length) {
juint cenOffset = endpos - ENDSIZ(bp);
juint locOffset = cenOffset - ENDOFF(bp);
unsigned char sig[4];
if (bf().seek(locOffset, SEEK_SET) >= 0 &&
bf().get_bytes(sig, 4) == 4 &&
sig[0] == (unsigned char)'P' &&
sig[1] == (unsigned char)'K' &&
sig[2] == (unsigned char) 3 &&
sig[3] == (unsigned char) 4) {
raw_current_entry()->cenOffset = cenOffset;
raw_current_entry()->nextCenOffset = cenOffset;
raw_current_entry()->locOffset = locOffset;
#if ENABLE_ROM_GENERATOR
raw_current_entry()->totalEntryCount = ENDTOT(bp);
#endif
}
return true; // Found central header
}
}
/* FALL THROUGH */
default:
/* The header must start at least four characters back, since
* the current character isn't in the header */
bp -= 4;
}
if (bp < buffer) {
/* We've moved outside our window into the file. We must
* move the window backwards */
size_t count = (size_t) (currentOffset - minOffset); /* Bytes left in file */
if (((jint)count) <= 0) {
/* Nothing left to read. Time to give up */
return false;
} else {
/* up to ((bp - buffer) + ENDHDRSIZ) bytes in the buffer might
* still be part of the end header, so the most bytes we can
* actually read are
* TMPBUFFERSIZE - ((bp - buffer) + ENDHDRSIZE).
*/
size_t available = (TMPBUFFERSIZE - ENDHDRSIZ) + (buffer - bp);
if (count > available) {
count = available;
}
}
/* Back up, while keeping our virtual currentOffset the same */
currentOffset -= count;
bp += count;
jvm_memmove(buffer + count, buffer, TMPBUFFERSIZE - count);
if ( bf().seek(currentOffset, SEEK_SET) < 0 ||
bf().get_bytes(buffer, count) != size_t(count) ) {
return false;
}
}
} /* end of for loop */
}
/*
* Computes and positions at the start of the CEN header, ie. the central
* directory, this will also return the offset if there is a zip file comment
* at the end of the archive, for most cases this would be 0.
*/
static jlong
compute_cen(int fd, Byte *bp)
{
int bytes;
Byte *p;
jlong base_offset;
jlong offset;
char buffer[MINREAD];
p = buffer;
/*
* Read the END Header, which is the starting point for ZIP files.
* (Clearly designed to make writing a zip file easier than reading
* one. Now isn't that precious...)
*/
if ((base_offset = find_end(fd, bp)) == -1) {
return (-1);
}
p = bp;
/*
* There is a historical, but undocumented, ability to allow for
* additional "stuff" to be prepended to the zip/jar file. It seems
* that this has been used to prepend an actual java launcher
* executable to the jar on Windows. Although this is just another
* form of statically linking a small piece of the JVM to the
* application, we choose to continue to support it. Note that no
* guarantees have been made (or should be made) to the customer that
* this will continue to work.
*
* Therefore, calculate the base offset of the zip file (within the
* expanded file) by assuming that the central directory is followed
* immediately by the end record.
*/
if (zip64_present) {
if ((offset = ZIP64_LOCOFF(p)) < (jlong)0) {
return -1;
}
if (JLI_Lseek(fd, offset, SEEK_SET) < (jlong) 0) {
return (-1);
}
if ((bytes = read(fd, buffer, MINREAD)) < 0) {
return (-1);
}
if (GETSIG(buffer) != ZIP64_ENDSIG) {
return -1;
}
if ((offset = ZIP64_ENDOFF(buffer)) < (jlong)0) {
return -1;
}
if (JLI_Lseek(fd, offset, SEEK_SET) < (jlong)0) {
return (-1);
}
p = buffer;
base_offset = base_offset - ZIP64_ENDSIZ(p) - ZIP64_ENDOFF(p) - ZIP64_ENDHDR;
} else {
base_offset = base_offset - ENDSIZ(p) - ENDOFF(p);
/*
* The END Header indicates the start of the Central Directory
* Headers. Remember that the desired Central Directory Header (CEN)
* will almost always be the second one and the first one is a small
* directory entry ("META-INF/"). Keep the code optimized for
* that case.
*
* Seek to the beginning of the Central Directory.
*/
if (JLI_Lseek(fd, base_offset + ENDOFF(p), SEEK_SET) < (jlong) 0) {
return (-1);
}
}
return base_offset;
}