Rewrite ZipOS

This reduces the virtual memory usage of Emacs for me by 30%. We now
have a simpler implementation that uses read(), rather mmap()ing the
whole executable.
This commit is contained in:
Justine Tunney 2023-10-03 07:27:25 -07:00
parent ff77f2a6af
commit b01282e23e
No known key found for this signature in database
GPG key ID: BE714B4575D6E328
21 changed files with 408 additions and 421 deletions

View file

@ -16,127 +16,89 @@
TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
PERFORMANCE OF THIS SOFTWARE.
*/
#include "libc/assert.h"
#include "libc/intrin/bits.h"
#include "libc/atomic.h"
#include "libc/calls/calls.h"
#include "libc/cosmo.h"
#include "libc/dce.h"
#include "libc/errno.h"
#include "libc/intrin/kprintf.h"
#include "libc/intrin/promises.internal.h"
#include "libc/intrin/strace.internal.h"
#include "libc/intrin/weaken.h"
#include "libc/macros.internal.h"
#include "libc/runtime/internal.h"
#include "libc/runtime/runtime.h"
#include "libc/runtime/symbols.internal.h"
#include "libc/runtime/zipos.internal.h"
#include "libc/str/str.h"
#include "libc/thread/thread.h"
#include "libc/x/x.h"
#include "libc/zip.internal.h"
#include "third_party/puff/puff.h"
#include "libc/sysv/consts/map.h"
#include "libc/sysv/consts/o.h"
#include "libc/sysv/consts/prot.h"
__static_yoink("__get_symbol");
static pthread_spinlock_t g_lock;
struct SymbolTable *__symtab; // for kprintf
struct SymbolTableLoader __symtab;
static ssize_t GetZipFile(struct Zipos *zipos, const char *name) {
size_t i, n, c, z;
z = strlen(name);
c = GetZipCdirOffset(zipos->cdir);
n = GetZipCdirRecords(zipos->cdir);
for (i = 0; i < n; ++i, c += ZIP_CFILE_HDRSIZE(zipos->map + c)) {
if (ZIP_CFILE_NAMESIZE(zipos->map + c) == z &&
!memcmp(ZIP_CFILE_NAME(zipos->map + c), name, z)) {
return c;
}
}
return -1;
}
/**
* Reads symbol table from zip directory.
* @note This code can't depend on dlmalloc()
*/
static struct SymbolTable *GetSymbolTableFromZip(struct Zipos *zipos) {
ssize_t cf, lf;
size_t size, size2;
static struct SymbolTable *GetSymbolTableFromZip(void) {
int fd;
struct SymbolTable *res = 0;
if ((cf = GetZipFile(zipos, ".symtab." _ARCH_NAME)) != -1 ||
(cf = GetZipFile(zipos, ".symtab")) != -1) {
lf = GetZipCfileOffset(zipos->map + cf);
size = GetZipLfileUncompressedSize(zipos->map + lf);
size2 = ROUNDUP(size, FRAMESIZE);
if ((res = _mapanon(size2))) {
switch (ZIP_LFILE_COMPRESSIONMETHOD(zipos->map + lf)) {
case kZipCompressionNone:
memcpy(res, (void *)ZIP_LFILE_CONTENT(zipos->map + lf), size);
break;
case kZipCompressionDeflate:
if (__inflate((void *)res, size,
(void *)ZIP_LFILE_CONTENT(zipos->map + lf),
GetZipLfileCompressedSize(zipos->map + lf))) {
munmap(res, size2);
res = 0;
}
break;
default:
munmap(res, size2);
res = 0;
break;
}
if ((fd = open("/zip/.symtab." _ARCH_NAME, O_RDONLY)) != -1 ||
(fd = open("/zip/.symtab", O_RDONLY)) != -1) {
void *map;
ssize_t size;
if ((size = lseek(fd, 0, SEEK_END)) != -1 &&
(map = mmap(0, size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0)) !=
MAP_FAILED) {
res = map;
}
close(fd);
}
STRACE("GetSymbolTableFromZip() → %p", res);
return res;
}
/**
* Reads symbol table from .com.dbg file.
* @note This code can't depend on dlmalloc()
*/
static struct SymbolTable *GetSymbolTableFromElf(void) {
const char *s;
if (PLEDGED(RPATH) && (s = FindDebugBinary())) {
return OpenSymbolTable(s);
const char *path;
if ((path = FindDebugBinary())) {
return OpenSymbolTable(path);
} else {
return 0;
}
}
static void GetSymbolTableInit(void) {
if (!PLEDGED(RPATH)) return;
int e = errno;
if ((__symtab.st = GetSymbolTableFromZip())) {
__symtab.st->names =
(uint32_t *)((char *)__symtab.st + __symtab.st->names_offset);
__symtab.st->name_base =
(char *)((char *)__symtab.st + __symtab.st->name_base_offset);
}
if (!__symtab.st) {
__symtab.st = GetSymbolTableFromElf();
}
errno = e;
}
/**
* Returns symbol table singleton.
*
* This uses multiple strategies to find the symbol table. The first
* strategy, depends on whether or not the following is linked:
*
* __static_yoink("__zipos_get");
* __static_yoink("zipos");
*
* In that case, the symbol table may be read from `/zip/.symtab` which
* is generated by `o//tool/build/symtab.com`. The second strategy is to
* look for the concomitant `.com.dbg` executable, which may very well
* be the one currently executing, or it could be placed in the same
* folder as your `.com` binary, or lastly, it could be explicitly
* specified via the `COMDBG` environment variable.
* In that case, the symbol table may be read from `/zip/.symtab.ARCH`
* or `/zip/.symtab` which are generated by `o//tool/build/symtab.com`
* or `o//tool/build/apelink.com`.
*
* Function tracing is disabled throughout the duration of this call.
* Backtraces and other core runtime functionality depend on this.
* The second strategy is to look for the ELF executable for the current
* program. If you're running a .com binary, it'll look for the .com.dbg
* file. If it's running the .com.dbg or .elf file, then it'll just read
* the symbols from itself. In other cases, you can explicitly specify a
* debug symbol binary via the `COMDBG` environment variable.
*
* When using pledge() security, it's recommended that this function get
* called *before* calling pledge() if it lacks the rpath promise, so it
* can load the symbols into memory before the filesystem goes away.
*
* @return symbol table, or NULL if not found
*/
struct SymbolTable *GetSymbolTable(void) {
struct Zipos *z;
if (pthread_spin_trylock(&g_lock)) return 0;
if (!__symtab && !__isworker) {
if (_weaken(__zipos_get) && (z = _weaken(__zipos_get)())) {
if ((__symtab = GetSymbolTableFromZip(z))) {
__symtab->names =
(uint32_t *)((char *)__symtab + __symtab->names_offset);
__symtab->name_base =
(char *)((char *)__symtab + __symtab->name_base_offset);
}
}
if (!__symtab) {
__symtab = GetSymbolTableFromElf();
}
}
pthread_spin_unlock(&g_lock);
return __symtab;
cosmo_once(&__symtab.once, GetSymbolTableInit);
return __symtab.st;
}