/* Sets modcache->has_debug_info.
* No lock is needed as we assume the caller hasn't exposed modcache outside this
* thread yet.
*/
static bool
symcache_read_symfile(const module_data_t *mod, const char *modname,
mod_cache_t *modcache)
{
hashtable_t *symtable = &modcache->table;
bool res = false;
const char *line, *next_line;
char symbol[MAX_SYMLEN];
size_t offs;
uint64 map_size;
size_t actual_size;
bool ok;
void *map = NULL;
char symfile[MAXIMUM_PATH];
file_t f;
symcache_get_filename(modname, symfile, BUFFER_SIZE_ELEMENTS(symfile));
f = dr_open_file(symfile, DR_FILE_READ);
if (f == INVALID_FILE)
goto symcache_read_symfile_done;
LOG(2, "processing symbol cache file for %s\n", modname);
/* we avoid having to do our own buffering by just mapping the whole file */
ok = dr_file_size(f, &map_size);
if (ok) {
actual_size = (size_t) map_size;
ASSERT(actual_size == map_size, "file size too large");
map = dr_map_file(f, &actual_size, 0, NULL, DR_MEMPROT_READ, 0);
}
if (!ok || map == NULL || actual_size < map_size) {
NOTIFY_ERROR("Error mapping symcache file for %s"NL, modname);
goto symcache_read_symfile_done;
}
if (strncmp((char *)map, SYMCACHE_FILE_HEADER, strlen(SYMCACHE_FILE_HEADER)) != 0) {
WARN("WARNING: symbol cache file is corrupted\n");
goto symcache_read_symfile_done;
}
/* i#1057: We use dr_sscanf() because sscanf() from ntdll will call strlen()
* and read off the end of the mapped file if it doesn't hit a null.
*/
if (dr_sscanf((char *)map + strlen(SYMCACHE_FILE_HEADER) + 1, "%d",
(uint *)&offs) != 1 ||
/* neither forward nor backward compatible */
offs != SYMCACHE_VERSION) {
WARN("WARNING: symbol cache file has wrong version\n");
goto symcache_read_symfile_done;
}
line = strchr((char *) map, '\n');
if (line != NULL)
line++;
if (line != NULL) {
/* Module consistency checks */
uint cache_file_size;
uint64 module_file_size;
uint timestamp;
#ifdef WINDOWS
version_number_t file_version;
version_number_t product_version;
uint checksum;
size_t module_internal_size;
if (dr_sscanf(line, "%u,"UINT64_FORMAT_STRING","UINT64_FORMAT_STRING","
UINT64_FORMAT_STRING",%u,%u,%lu",
&cache_file_size, &module_file_size, &file_version.version,
&product_version.version, &checksum, ×tamp,
&module_internal_size) != 7) {
WARN("WARNING: %s symbol cache file has bad consistency header\n", modname);
goto symcache_read_symfile_done;
}
if (module_file_size != modcache->module_file_size ||
file_version.version != modcache->file_version.version ||
product_version.version != modcache->product_version.version ||
checksum != modcache->checksum ||
timestamp != modcache->timestamp ||
module_internal_size != modcache->module_internal_size) {
LOG(1, "module version mismatch: %s symbol cache file is stale\n", modname);
LOG(2, "\t"UINT64_FORMAT_STRING" vs "UINT64_FORMAT_STRING", "
UINT64_FORMAT_STRING" vs "UINT64_FORMAT_STRING", "
UINT64_FORMAT_STRING" vs "UINT64_FORMAT_STRING", "
"%u vs %u, %u vs %u, %lu vs %lu\n",
module_file_size, modcache->module_file_size,
file_version.version, modcache->file_version.version,
product_version.version, modcache->product_version.version,
checksum, modcache->checksum,
timestamp, modcache->timestamp,
module_internal_size, modcache->module_internal_size);
goto symcache_read_symfile_done;
}
#elif defined(LINUX)
if (dr_sscanf(line, "%u,"UINT64_FORMAT_STRING",%u",
&cache_file_size, &module_file_size, ×tamp) != 3) {
WARN("WARNING: %s symbol cache file has bad consistency header\n", modname);
goto symcache_read_symfile_done;
}
if (module_file_size != modcache->module_file_size ||
timestamp != modcache->timestamp) {
LOG(1, "module version mismatch: %s symbol cache file is stale\n", modname);
goto symcache_read_symfile_done;
}
#elif defined(MACOS)
uint current_version;
uint compatibility_version;
byte uuid[16];
/* XXX: if dr_sscanf supported %n maybe we could split these into
* separate scans on the same string and share code w/ Linux.
*/
if (dr_sscanf(line, "%u,"UINT64_FORMAT_STRING",%u,%u,%u,%x,%x,%x,%x",
&cache_file_size, &module_file_size, ×tamp,
¤t_version, &compatibility_version,
(uint*)(&uuid[0]), (uint*)(&uuid[4]),
(uint*)(&uuid[8]), (uint*)(&uuid[12])) != 9) {
WARN("WARNING: %s symbol cache file has bad consistency header B\n", modname);
goto symcache_read_symfile_done;
}
if (current_version != modcache->current_version ||
compatibility_version != modcache->compatibility_version ||
memcmp(uuid, modcache->uuid, sizeof(uuid)) != 0) {
LOG(1, "module version mismatch: %s symbol cache file is stale\n", modname);
goto symcache_read_symfile_done;
}
#endif
/* We could go further w/ CRC or even MD5 but not worth it for dev tool */
if (cache_file_size != (uint)map_size) {
WARN("WARNING: %s symbol cache file is corrupted: map=%d vs file=%d\n",
modname, (uint)map_size, cache_file_size);
goto symcache_read_symfile_done;
}
}
line = strchr(line, '\n');
if (line != NULL)
line++;
if (line != NULL) {
uint has_debug_info;
if (dr_sscanf(line, "%u", &has_debug_info) != 1) {
WARN("WARNING: %s symbol cache file has bad consistency header\n", modname);
goto symcache_read_symfile_done;
}
if (has_debug_info) {
/* We assume that the current availability of debug info doesn't matter */
modcache->has_debug_info = true;
} else {
/* We delay the costly check for symbols until we've read the symcache
* b/c if its entry indicates symbols we don't need to look
*/
if (module_has_symbols(mod)) {
LOG(1, "module now has debug info: %s symbol cache is stale\n", modname);
goto symcache_read_symfile_done;
}
}
}
line = strchr(line, '\n');
if (line != NULL)
line++;
symbol[0] = '\0';
for (; line != NULL && line < ((char *)map) + map_size; line = next_line) {
const char *comma = strchr(line, ',');
const char *newline = strchr(line, '\n');
size_t symlen = (comma != NULL ? comma - line : 0);
if (newline == NULL) {
next_line = ((char *)map) + map_size + 1; /* handle EOF w/o trailing \n */
} else {
next_line = newline + 1;
}
if (symlen > 0 && symlen < MAX_SYMLEN) {
strncpy(symbol, line, symlen);
symbol[symlen] = '\0';
}
if (comma != NULL && symlen < MAX_SYMLEN && symbol[0] != '\0' &&
dr_sscanf(comma, ",0x%x", (uint *)&offs) == 1) {
#ifdef WINDOWS
/* Guard against corrupted files that cause DrMem to crash (i#1465) */
if (offs >= modcache->module_internal_size) {
/* This one we want to know about */
NOTIFY("SYMCACHE ERROR: %s file has too-large entry "PIFX" for %s"NL,
modname, offs, symbol);
goto symcache_read_symfile_done;
}
#endif
symcache_symbol_add(modname, symtable, symbol, offs);
} else {
WARN("WARNING: malformed symbol cache line \"%.*s\"\n",
next_line - line - 1, line);
/* We abort in case there were two dueling writes to the file
* and it somehow got past the self-consistency check,
* putting a header in the middle of the file, and we can't
* trust subsequent lines since they may belong to a different
* version of the module
*/
break; /* res should still be true */
}
}
res = true;
symcache_read_symfile_done:
if (map != NULL)
dr_unmap_file(map, actual_size);
if (f != INVALID_FILE)
dr_close_file(f);
if (!res)
modcache->has_debug_info = module_has_symbols(mod);
return res;
}
/* file I/O */
void md_read_from_file (module_t * head, file_t file, bool extra_info){
uint64 map_size;
size_t actual_size;
bool ok;
void * map = NULL;
char * line;
/*loop variables*/
int i;
int j;
/* linked list structure specific variables */
int no_modules;
int no_instructions;
unsigned int addr;
char module_name[MAX_STRING_LENGTH];
/* for filling up the linked list data structure */
module_t * elem;
ok = dr_file_size(file,&map_size);
if(ok){
actual_size = (size_t)map_size;
DR_ASSERT(actual_size == map_size);
map = dr_map_file(file, &actual_size, 0, NULL, DR_MEMPROT_READ, 0);
}
dr_sscanf((char *)map,"%d\n",&no_modules);
//dr_printf("%d\n",no_modules); //debug
line = (char *)map;
for(i=0;i<no_modules;i++){
line = strchr(line,'\n');
line++; //start of the next line
//dr_sscanf(line,"%[^\t\n]\n",module_name);
dr_get_token(line,module_name,MAX_STRING_LENGTH);
//getFinalName(module_name);
//dr_printf("%s\n",module_name); //debug
line = strchr(line,'\n');
line++; //start of the next line
dr_sscanf(line,"%d\n",&no_instructions);
//dr_printf("%d\n",no_instructions); //debug
//create a new element
elem = new_elem(module_name,no_instructions+2);
head->next = elem;
head = elem;
for(j=0;j<no_instructions;j++){
line = strchr(line,'\n');
line++; //start of the next line
dr_sscanf(line,"%u\n",&addr);
//dr_printf(line,"%x\n",addr); //debug
add_bb_to_list(head->bbs,addr, extra_info, head->size_bbs);
}
}
}
drcovlib_status_t
drmodtrack_offline_read(file_t file, const char **map,
OUT void **handle, OUT uint *num_mods)
{
module_read_info_t *info = NULL;
uint i;
uint64 file_size;
size_t map_size = 0;
const char *buf, *map_start;
uint version;
if (handle == NULL || num_mods == NULL)
return DRCOVLIB_ERROR_INVALID_PARAMETER;
if (file == INVALID_FILE) {
if (map == NULL)
return DRCOVLIB_ERROR_INVALID_PARAMETER;
map_start = *map;
} else {
if (!dr_file_size(file, &file_size))
return DRCOVLIB_ERROR_INVALID_PARAMETER;
map_size = (size_t)file_size;
map_start = (char *) dr_map_file(file, &map_size, 0, NULL, DR_MEMPROT_READ, 0);
if (map_start == NULL || map_size < file_size)
return DRCOVLIB_ERROR_INVALID_PARAMETER; /* assume bad perms or sthg */
}
if (map_start == NULL)
return DRCOVLIB_ERROR_INVALID_PARAMETER;
buf = map_start;
/* Module table header, handling the pre-versioning legacy format. */
if (dr_sscanf(buf, "Module Table: %u\n", num_mods) == 1)
version = 1;
else if (dr_sscanf(buf, "Module Table: version %u, count %u\n", &version,
num_mods) != 2 ||
version != MODULE_FILE_VERSION)
goto read_error;
buf = move_to_next_line(buf);
if (version > 1) {
// Skip header line
buf = move_to_next_line(buf);
}
info = (module_read_info_t *)dr_global_alloc(sizeof(*info));
if (file != INVALID_FILE) {
info->map = map_start;
info->map_size = map_size;
} else
info->map = NULL;
info->num_mods = *num_mods;
info->mod = (module_read_entry_t *)dr_global_alloc(*num_mods * sizeof(*info->mod));
/* module lists */
for (i = 0; i < *num_mods; i++) {
uint mod_id;
if (version == 1) {
if (dr_sscanf(buf, " %u, %" INT64_FORMAT"u, %[^\n\r]",
&mod_id, &info->mod[i].size, info->mod[i].path) != 3 ||
mod_id != i)
goto read_error;
} else {
app_pc end, entry;
#ifdef WINDOWS
uint checksum, timestamp;
if (dr_sscanf(buf, " %u, "PIFX", "PIFX", "PIFX", 0x%x, 0x%x, %[^\n\r]",
&mod_id, &info->mod[i].base, &end, &entry,
&checksum, ×tamp,
info->mod[i].path) != 7 ||
mod_id != i)
goto read_error;
#else
if (dr_sscanf(buf, " %u, "PIFX", "PIFX", "PIFX", %[^\n\r]",
&mod_id, &info->mod[i].base, &end, &entry,
info->mod[i].path) != 5 ||
mod_id != i)
goto read_error;
#endif
info->mod[i].size = end - info->mod[i].base;
}
buf = move_to_next_line(buf);
}
if (file == INVALID_FILE)
*map = buf;
*handle = (void *)info;
return DRCOVLIB_SUCCESS;
read_error:
if (info != NULL) {
dr_global_free(info->mod, *num_mods * sizeof(*info->mod));
dr_global_free(info, sizeof(*info));
}
if (file != INVALID_FILE)
dr_unmap_file((char *)map_start, map_size);
return DRCOVLIB_ERROR;
}
void
symcache_module_load(void *drcontext, const module_data_t *mod, bool loaded)
{
/* look for cache file for this module.
* fill in hashtable: key is string, value is list of offsets
*/
mod_cache_t *modcache;
const char *modname = dr_module_preferred_name(mod);
file_t f;
if (modname == NULL)
return; /* don't support caching */
/* if smaller than threshold, not worth caching */
if (mod->end - mod->start < op_modsize_cache_threshold) {
LOG(1, "%s: module %s too small to cache\n", __FUNCTION__, modname);
return;
}
ASSERT(initialized, "symcache was not initialized");
/* support initializing prior to module events => called twice */
dr_mutex_lock(symcache_lock);
modcache = (mod_cache_t *) hashtable_lookup(&symcache_table,
(void *)mod->full_path);
dr_mutex_unlock(symcache_lock);
if (modcache != NULL)
return;
modcache = (mod_cache_t *) global_alloc(sizeof(*modcache), HEAPSTAT_HASHTABLE);
memset(modcache, 0, sizeof(*modcache));
hashtable_init_ex(&modcache->table, SYMCACHE_MODULE_TABLE_HASH_BITS,
HASH_STRING, true/*strdup*/, true/*synch*/,
symcache_free_list, NULL, NULL);
/* store consistency fields */
f = dr_open_file(mod->full_path, DR_FILE_READ);
if (f != INVALID_FILE) {
bool ok = dr_file_size(f, &modcache->module_file_size);
if (!ok)
WARN("WARNING: unable to determine size of %s\n", mod->full_path);
dr_close_file(f);
} else
WARN("WARNING: unable to open %s\n", mod->full_path);
#ifdef WINDOWS
modcache->file_version = mod->file_version;
modcache->product_version = mod->product_version;
modcache->checksum = mod->checksum;
modcache->timestamp = mod->timestamp;
modcache->module_internal_size = mod->module_internal_size;
#endif
modcache->modname = drmem_strdup(modname, HEAPSTAT_HASHTABLE);
modcache->from_file = symcache_read_symfile(mod, modname, modcache);
dr_mutex_lock(symcache_lock);
if (!hashtable_add(&symcache_table, (void *)mod->full_path, (void *)modcache)) {
/* this should be really rare to have dup paths (xref i#729) -- and
* actually we now have a lookup up above so we should only get here
* on a race while we let go of the lock
*/
WARN("WARNING: duplicate module paths: only caching symbols from first\n");
hashtable_delete(&modcache->table);
global_free(modcache, sizeof(*modcache), HEAPSTAT_HASHTABLE);
}
dr_mutex_unlock(symcache_lock);
}
static bool
symcache_read_symfile(const module_data_t *mod, const char *modname, mod_cache_t *modcache)
{
hashtable_t *symtable = &modcache->table;
bool res = false;
const char *line, *next_line;
char symbol[MAX_SYMLEN];
size_t offs;
uint64 map_size;
size_t actual_size;
bool ok;
void *map = NULL;
char symfile[MAXIMUM_PATH];
file_t f;
symcache_get_filename(modname, symfile, BUFFER_SIZE_ELEMENTS(symfile));
f = dr_open_file(symfile, DR_FILE_READ);
if (f == INVALID_FILE)
return res;
LOG(2, "processing symbol cache file for %s\n", modname);
/* we avoid having to do our own buffering by just mapping the whole file */
ok = dr_file_size(f, &map_size);
if (ok) {
actual_size = (size_t) map_size;
ASSERT(actual_size == map_size, "file size too large");
map = dr_map_file(f, &actual_size, 0, NULL, DR_MEMPROT_READ, 0);
}
if (!ok || map == NULL || actual_size < map_size) {
NOTIFY_ERROR("Error mapping symcache file for %s"NL, modname);
goto symcache_read_symfile_done;
}
if (strncmp((char *)map, SYMCACHE_FILE_HEADER, strlen(SYMCACHE_FILE_HEADER)) != 0) {
WARN("WARNING: symbol cache file is corrupted\n");
goto symcache_read_symfile_done;
}
if (sscanf((char *)map + strlen(SYMCACHE_FILE_HEADER) + 1, "%d", &offs) != 1 ||
/* neither forward nor backward compatible */
offs != SYMCACHE_VERSION) {
WARN("WARNING: symbol cache file has wrong version\n");
goto symcache_read_symfile_done;
}
line = strchr((char *) map, '\n');
if (line != NULL)
line++;
if (line != NULL) {
/* Module consistency checks */
uint cache_file_size;
uint64 module_file_size;
#ifdef WINDOWS
version_number_t file_version;
version_number_t product_version;
uint checksum;
uint timestamp;
size_t module_internal_size;
if (sscanf(line, " %u,"UINT64_FORMAT_STRING","UINT64_FORMAT_STRING","
UINT64_FORMAT_STRING",%u,%u,%lu",
&cache_file_size, &module_file_size, &file_version.version,
&product_version.version, &checksum, ×tamp,
&module_internal_size) != 7) {
WARN("WARNING: %s symbol cache file has bad consistency header\n", modname);
goto symcache_read_symfile_done;
}
if (module_file_size != modcache->module_file_size ||
file_version.version != modcache->file_version.version ||
product_version.version != modcache->product_version.version ||
checksum != modcache->checksum ||
timestamp != modcache->timestamp ||
module_internal_size != modcache->module_internal_size) {
LOG(1, "module version mismatch: %s symbol cache file is stale\n", modname);
LOG(2, "\t"UINT64_FORMAT_STRING" vs "UINT64_FORMAT_STRING", "
UINT64_FORMAT_STRING" vs "UINT64_FORMAT_STRING", "
UINT64_FORMAT_STRING" vs "UINT64_FORMAT_STRING", "
"%u vs %u, %u vs %u, %lu vs %lu\n",
module_file_size, modcache->module_file_size,
file_version.version, modcache->file_version.version,
product_version.version, modcache->product_version.version,
checksum, modcache->checksum,
timestamp, modcache->timestamp,
module_internal_size, modcache->module_internal_size);
goto symcache_read_symfile_done;
}
#else
if (sscanf(line, "%u,"UINT64_FORMAT_STRING,
&cache_file_size, &module_file_size) != 2) {
WARN("WARNING: %s symbol cache file has bad consistency header\n", modname);
goto symcache_read_symfile_done;
}
if (module_file_size != modcache->module_file_size) {
LOG(1, "module version mismatch: %s symbol cache file is stale\n", modname);
goto symcache_read_symfile_done;
}
#endif
/* We could go further w/ CRC or even MD5 but not worth it for dev tool */
if (cache_file_size != (uint)map_size) {
WARN("WARNING: %s symbol cache file is corrupted\n", modname);
goto symcache_read_symfile_done;
}
}
line = strchr(line, '\n');
if (line != NULL)
line++;
symbol[0] = '\0';
for (; line != NULL && line < ((char *)map) + map_size; line = next_line) {
const char *newline = strchr(line, '\n');
if (newline == NULL) {
next_line = ((char *)map) + map_size + 1; /* handle EOF w/o trailing \n */
} else {
next_line = newline + 1;
}
if (sscanf(line, "%"MAX_SYMLEN_MINUS_1_STR"[^,],0x%x", symbol, &offs) == 2) {
symcache_symbol_add(modname, symtable, symbol, offs);
} else if (symbol[0] != '\0' && sscanf(line, ",0x%x", &offs) == 1) {
/* duplicate entries are allowed to not list the symbol, to save
* space in the file (mainly for post-call caching i#669)
*/
symcache_symbol_add(modname, symtable, symbol, offs);
} else {
WARN("WARNING: malformed symbol cache line \"%.*s\"\n",
next_line - line - 1, line);
/* We abort in case there were two dueling writes to the file
* and it somehow got past the self-consistency check,
* putting a header in the middle of the file, and we can't
* trust subsequent lines since they may belong to a different
* version of the module
*/
break; /* res should still be true */
}
}
res = true;
symcache_read_symfile_done:
if (map != NULL)
dr_unmap_file(map, actual_size);
if (f != INVALID_FILE)
dr_close_file(f);
return res;
}
