/
lx_loader.c
2121 lines (1816 loc) · 79.2 KB
1
2
3
4
5
6
7
8
9
10
#define _GNU_SOURCE 1
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <stdint.h>
#include <errno.h>
#include <sys/mman.h>
#include <assert.h>
11
#include <dlfcn.h>
12
#include <dirent.h>
13
#include <pthread.h>
14
15
#include <signal.h>
#include <ucontext.h>
16
17
18
19
20
21
// 16-bit selector kernel nonsense...
#include <sys/syscall.h>
#include <sys/types.h>
#include <asm/ldt.h>
22
23
#include "lx_loader.h"
24
25
26
static LxLoaderState _GLoaderState;
static LxLoaderState *GLoaderState = &_GLoaderState;
27
// !!! FIXME: move this into an lx_common.c file.
28
static int sanityCheckLxModule(uint8 **_exe, uint32 *_exelen)
29
30
{
if (*_exelen < 196) {
31
fprintf(stderr, "not an OS/2 LX module\n");
32
33
34
return 0;
}
const uint32 header_offset = *((uint32 *) (*_exe + 0x3C));
35
//printf("header offset is %u\n", (uint) header_offset);
36
if ((header_offset + sizeof (LxHeader)) >= *_exelen) {
37
fprintf(stderr, "not an OS/2 LX module\n");
38
39
40
41
42
43
44
45
46
return 0;
}
*_exe += header_offset; // skip the DOS stub, etc.
*_exelen -= header_offset;
const LxHeader *lx = (const LxHeader *) *_exe;
if ((lx->magic_l != 'L') || (lx->magic_x != 'X')) {
47
fprintf(stderr, "not an OS/2 LX module\n");
48
49
50
51
52
53
54
55
56
return 0;
}
if ((lx->byte_order != 0) || (lx->word_order != 0)) {
fprintf(stderr, "Program is not little-endian!\n");
return 0;
}
if (lx->lx_version != 0) {
57
fprintf(stderr, "Program is unknown LX module version (%u)\n", (uint) lx->lx_version);
58
59
60
61
return 0;
}
if (lx->cpu_type > 3) { // 1==286, 2==386, 3==486
62
fprintf(stderr, "Program needs unknown CPU type (%u)\n", (uint) lx->cpu_type);
63
64
65
66
return 0;
}
if (lx->os_type != 1) { // 1==OS/2, others: dos4, windows, win386, unknown.
67
fprintf(stderr, "Program needs unknown OS type (%u)\n", (uint) lx->os_type);
68
69
70
71
return 0;
}
if (lx->page_size != 4096) {
72
fprintf(stderr, "Program page size isn't 4096 (%u)\n", (uint) lx->page_size);
73
74
75
return 0;
}
76
77
78
79
80
if (lx->module_flags & 0x2000) {
fprintf(stderr, "Module is flagged as not-loadable\n");
return 0;
}
81
82
83
// !!! FIXME: check if EIP and ESP are non-zero vs per-process library bits, etc.
return 1;
84
} // sanityCheckLxModule
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
static char *makeOS2Path(const char *fname)
{
char *full = realpath(fname, NULL);
if (!full)
return NULL;
char *retval = (char *) malloc(strlen(full) + 3);
if (!retval)
return NULL;
retval[0] = 'C';
retval[1] = ':';
strcpy(retval + 2, full);
free(full);
for (char *ptr = retval + 2; *ptr; ptr++) {
if (*ptr == '/')
*ptr = '\\';
} // for
return retval;
} // makeOS2Path
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
// based on case-insensitive search code from PhysicsFS:
// https://icculus.org/physfs/
// It's also zlib-licensed, plus I wrote it. :) --ryan.
// !!! FIXME: this doesn't work as-is for UTF-8 case folding, since string
// !!! FIXNE: length can change!
static int locateOneElement(char *buf)
{
if (access(buf, F_OK) == 0)
return 1; // quick rejection: exists in current case.
DIR *dirp;
char *ptr = strrchr(buf, '/'); // find entry at end of path.
if (ptr == NULL) {
dirp = opendir(".");
ptr = buf;
} else if (ptr == buf) {
dirp = opendir("/");
} else {
*ptr = '\0';
dirp = opendir(buf);
*ptr = '/';
ptr++; // point past dirsep to entry itself.
} // else
for (struct dirent *dent = readdir(dirp); dent; dent = readdir(dirp)) {
if (strcasecmp(dent->d_name, ptr) == 0) {
strcpy(ptr, dent->d_name); // found a match. Overwrite with this case.
closedir(dirp);
return 1;
} // if
} // for
// no match at all...
closedir(dirp);
return 0;
} // locateOneElement
static int locatePathCaseInsensitive(char *buf)
{
int rc;
char *ptr = buf;
if (*ptr == '\0')
return 0; // Uh...I guess that's success?
if (access(buf, F_OK) == 0)
return 0; // quick rejection: exists in current case.
while ( (ptr = strchr(ptr + 1, '/')) != NULL ) {
*ptr = '\0'; // block this path section off
rc = locateOneElement(buf);
*ptr = '/'; // restore path separator
if (!rc)
return -2; // missing element in path.
} // while
// check final element...
return locateOneElement(buf) ? 0 : -1;
} // locatePathCaseInsensitive
static char *makeUnixPath(const char *os2path, uint32 *err)
{
if ((strcasecmp(os2path, "NUL") == 0) || (strcasecmp(os2path, "\\DEV\\NUL") == 0))
os2path = "/dev/null";
// !!! FIXME: emulate other OS/2 device names (CON, etc).
//else if (strcasecmp(os2path, "CON") == 0)
else {
char drive = os2path[0];
if ((drive >= 'a') && (drive <= 'z'))
drive += 'A' - 'a';
if ((drive >= 'A') && (drive <= 'Z')) {
if (os2path[1] == ':') { // it's a drive letter.
if (drive == 'C')
os2path += 2; // skip "C:" if it's there.
else {
*err = 26; //ERROR_NOT_DOS_DISK;
return NULL;
} // else
} // if
} // if
} // else
const size_t len = strlen(os2path);
char *retval = (char *) malloc(len + 1);
if (!retval) {
*err = 8; //ERROR_NOT_ENOUGH_MEMORY;
return NULL;
} // else
strcpy(retval, os2path);
for (char *ptr = strchr(retval, '\\'); ptr; ptr = strchr(ptr + 1, '\\'))
*ptr = '/'; // convert to Unix-style path separators.
locatePathCaseInsensitive(retval);
return retval;
} // makeUnixPath
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
/* this algorithm is from lxlite 138u. */
static int decompressExePack2(uint8 *dst, const uint32 dstlen, const uint8 *src, const uint32 srclen)
{
uint32 sOf = 0;
uint32 dOf = 0;
uint32 bOf = 0;
uint32 b1 = 0;
uint32 b2 = 0;
#define SRCAVAIL(n) ((sOf + (n)) <= srclen)
#define DSTAVAIL(n) ((dOf + (n)) <= dstlen)
do {
if (!SRCAVAIL(1))
break;
b1 = src[sOf];
switch (b1 & 3) {
case 0:
if (b1 == 0) {
if (SRCAVAIL(2)) {
if (src[sOf + 1] == 0) {
sOf += 2;
break;
} else if (SRCAVAIL(3) && DSTAVAIL(src[sOf + 1])) {
memset(dst + dOf, src[sOf + 2], src[sOf + 1]);
sOf += 3;
dOf += src[sOf - 2];
} else {
return 0;
}
} else {
return 0;
}
} else if (SRCAVAIL((b1 >> 2) + 1) && DSTAVAIL(b1 >> 2)) {
memcpy(dst + dOf, src + (sOf + 1), b1 >> 2);
dOf += b1 >> 2;
sOf += (b1 >> 2) + 1;
} else {
return 0;
}
break;
case 1:
if (!SRCAVAIL(2))
return 0;
bOf = (*((const uint16 *) (src + sOf))) >> 7;
b2 = ((b1 >> 4) & 7) + 3;
b1 = ((b1 >> 2) & 3);
sOf += 2;
if (SRCAVAIL(b1) && DSTAVAIL(b1 + b2) && (dOf + b1 - bOf >= 0)) {
memcpy(dst + dOf, src + sOf, b1);
dOf += b1;
sOf += b1;
memmove(dst + dOf, dst + (dOf - bOf), b2);
dOf += b2;
} else {
return 0;
} // else
break;
case 2:
if (!SRCAVAIL(2))
return 0;
bOf = (*((const uint16 *) (src + sOf))) >> 4;
b1 = ((b1 >> 2) & 3) + 3;
if (DSTAVAIL(b1) && (dOf - bOf >= 0)) {
memmove(dst + dOf, dst + (dOf - bOf), b1);
dOf += b1;
sOf += 2;
} else {
return 0;
} // else
break;
case 3:
if (!SRCAVAIL(3))
return 0;
b2 = ((*((const uint16 *) (src + sOf))) >> 6) & 0x3F;
b1 = (src[sOf] >> 2) & 0x0F;
bOf = (*((const uint16 *) (src + (sOf + 1)))) >> 4;
sOf += 3;
if (SRCAVAIL(b1) && DSTAVAIL(b1 + b2) && (dOf + b1 - bOf >= 0))
{
memcpy(dst + dOf, src + sOf, b1);
dOf += b1;
sOf += b1;
memmove(dst + dOf, dst + (dOf - bOf), b2);
dOf += b2;
} else {
return 0;
} // else
break;
} // switch
} while (dOf < dstlen);
#undef SRCAVAIL
#undef DSTAVAIL
// pad out the rest of the page with zeroes.
308
309
310
if ((dstlen - dOf) > 0)
memset(dst + dOf, '\0', dstlen - dOf);
311
312
313
return 1;
} // decompressExePack2
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
static int decompressIterated(uint8 *dst, uint32 dstlen, const uint8 *src, uint32 srclen)
{
while (srclen) {
if (srclen < 4)
return 0;
const uint16 iterations = *((uint16 *) src); src += 2;
const uint16 len = *((uint16 *) src); src += 2;
srclen -= 4;
if (dstlen < (iterations * len))
return 0;
else if (srclen < len)
return 0;
for (uint16 i = 0; i < iterations; i++) {
memcpy(dst, src, len);
dst += len;
dstlen -= len;
} // for
src += len;
srclen -= len;
} // while
// pad out the rest of the page with zeroes.
if (dstlen > 0)
memset(dst, '\0', dstlen);
return 1;
} // decompressIterated
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
// !!! FIXME: mutex this
static int allocateSelector(const uint16 selector, const uint32 addr, const unsigned int contents, const int is32bit)
{
assert(selector < (sizeof (GLoaderState->ldt) / sizeof (GLoaderState->ldt[0])));
if (GLoaderState->ldt[selector])
return 0; // already in use.
//const int expand_down = (contents == MODIFY_LDT_CONTENTS_STACK);
struct user_desc entry;
entry.entry_number = (unsigned int) selector;
entry.base_addr = (unsigned int) addr; //(expand_down ? (addr + 0x10000) : addr);
entry.limit = 16; //expand_down ? 0 : 16;
entry.seg_32bit = is32bit;
entry.contents = contents;
entry.read_exec_only = 0;
entry.limit_in_pages = 1;
entry.seg_not_present = 0;
entry.useable = 1;
if (syscall(SYS_modify_ldt, 1, &entry, sizeof (entry)) != 0)
return 0;
GLoaderState->ldt[selector] = addr;
return 1;
} // allocateSelector
// !!! FIXME: mutex this
static int findSelector(const uint32 _addr, uint16 *outselector, uint16 *outoffset)
{
uint32 addr = _addr;
if (addr < 4096)
return 0; // we won't map the NULL page.
const uint32 *ldt = GLoaderState->ldt;
int available = -1;
int preferred = -1;
// optimize for the case where we need a selector that happenes to be in tiled memory,
// since it's fast to look up.
if (addr < (1024 * 1024 * 512)) {
const uint16 idx = (uint16) (addr >> 16);
const uint32 tile = ldt[idx];
if (tile == 0) {
preferred = available = idx; // we can use this piece.
} else if ((tile <= addr) && ((tile + 0x10000) > addr)) {
*outselector = idx;
*outoffset = (uint16) (addr - tile);
//printf("SELECTOR: found tiled selector 0x%X for address %p\n", (uint) idx, (void *) addr);
return 1; // already allocated to this address.
} // if
} // if
for (int i = 0; i < (sizeof (GLoaderState->ldt) / sizeof (GLoaderState->ldt[0])); i++) {
const uint32 tile = ldt[i];
if (tile == 0)
available = i;
else if ((tile <= addr) && ((tile + 0x10000) > addr)) {
*outselector = (uint16) i;
*outoffset = (uint16) (addr - tile);
//printf("SELECTOR: found existing selector 0x%X for address %p\n", (uint) i, (void *) addr);
return 1; // already allocated to this address.
} // else if
} // for
// nothing allocated to this address so far. Try to allocate something.
if (available == -1) {
fprintf(stderr, "Uhoh, we've run out of LDT selectors! Probably about to crash...\n"); fflush(stderr);
return 0; // uh oh, out of selectors!
} // if
const uint32 diff = addr % 0x10000;
addr -= diff; // make sure we start on a 64k border.
const uint16 selector = (uint16) (preferred != -1) ? preferred : available;
if (!allocateSelector(selector, addr, MODIFY_LDT_CONTENTS_CODE, 0)) {
fprintf(stderr, "Uhoh, we've failed to allocate LDT selector %u! Probably about to crash...\n", (uint) selector); fflush(stderr);
return 0;
} // if
//printf("SELECTOR: allocated selector 0x%X for address %p\n", (uint) selector, (void *) addr);
*outselector = selector;
*outoffset = (uint16) diff;
return 1;
} // findSelector
// !!! FIXME: mutex this
static void freeSelector(const uint16 selector)
{
assert(selector < (sizeof (GLoaderState->ldt) / sizeof (GLoaderState->ldt[0])));
if (!GLoaderState->ldt[selector])
return; // already free.
struct user_desc entry;
memset(&entry, '\0', sizeof (entry));
entry.entry_number = (unsigned int) selector;
entry.read_exec_only = 1;
entry.seg_not_present = 1;
if (syscall(SYS_modify_ldt, 1, &entry, sizeof (entry)) != 0)
return; // oh well.
GLoaderState->ldt[selector] = 0;
} // freeSelector
static void *convert1616to32(const uint32 addr1616)
{
452
453
454
if (addr1616 == 0)
return NULL;
455
456
457
const uint16 selector = (uint16) (addr1616 >> 19); // slide segment down, and shift out control bits.
const uint16 offset = (uint16) (addr1616 % 0x10000); // all our LDT segments start at 64k boundaries (at the moment!).
assert(GLoaderState->ldt[selector] != 0);
458
//printf("convert1616to32: 0x%X -> %p\n", (uint) addr1616, (void *) (size_t) (GLoaderState->ldt[selector] + offset));
459
460
461
return (void *) (size_t) (GLoaderState->ldt[selector] + offset);
} // convert1616to32
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
static uint32 convert32to1616(void *addr32)
{
if (addr32 == NULL)
return 0;
uint16 selector = 0;
uint16 offset = 0;
if (!findSelector((uint32) addr32, &selector, &offset)) {
fprintf(stderr, "Uhoh, ran out of LDT entries?!\n");
return 0; // oh well, crash, probably.
} // if
//printf("selector: 0x%X\n", (uint) selector);
selector = (selector << 3) | 7;
//printf("shifted selector: 0x%X\n", (uint) selector);
return (((uint32)selector) << 16) | ((uint32) offset);
} // convert32to1616
481
482
483
484
485
486
487
// EMX (and probably many other things) occasionally has to call a 16-bit
// system API, and assumes its stack is tiled in the LDT; it'll just shift
// the stack pointer and use it as a stack segment for the 16-bit call
// without calling DosFlatToSeg(). So we tile the main thread's stack and
// pray that covers it. If we have to tile _every_ thread's stack, we can do
// that later.
// !!! FIXME: if we do this for secondary thread stacks, we'll need to mutex this.
488
static void initOs2StackSegments(uint32 addr, uint32 stacklen, const int deinit)
489
490
491
492
493
494
495
496
497
498
499
500
501
{
//printf("base == %p, stacklen == %u\n", (void*)addr, (uint) stacklen);
const uint32 diff = addr % 0x10000;
addr -= diff; // make sure we start on a 64k border.
stacklen += diff; // make sure we start on a 64k border.
// We fill in LDT tiles for the entire stack (EMX, etc, assume this will work).
if (stacklen % 0x10000) // pad this out to 64k
stacklen += 0x10000 - (stacklen % 0x10000);
// !!! FIXME: do we have to allocate these backwards? (stack grows down).
while (stacklen) {
//printf("Allocating selector 0x%X for stack %p ... \n", (uint) (addr >> 16), (void *) addr);
502
503
504
505
506
507
508
if (deinit) {
freeSelector((uint16) (addr >> 16));
} else {
if (!allocateSelector((uint16) (addr >> 16), addr, MODIFY_LDT_CONTENTS_DATA/*STACK*/, 0)) {
FIXME("uhoh, couldn't set up an LDT entry for a stack segment! Might crash later!");
} // if
} // else
509
510
511
512
513
stacklen -= 0x10000;
addr += 0x10000;
} // while
} // initOs2StackSegments
514
515
516
517
518
519
520
521
// OS/2 threads keep their Thread Information Block at FS:0x0000, so we have
// to ask the Linux kernel to screw around with 16-bit selectors on our
// behalf so we don't crash out when apps try to access it directly.
// OS/2 provides a C-callable API to obtain the (32-bit linear!) TIB address
// without going directly to the FS register, but lots of programs (including
// the EMX runtime) touch the register directly, so we have to deal with it.
// You must call this once for each thread that will go into LX land, from
// that thread, as soon as possible after starting.
522
static uint16 initOs2Tib(uint8 *tibspace, void *_topOfStack, const size_t stacklen, const uint32 tid)
523
524
525
{
uint8 *topOfStack = (uint8 *) _topOfStack;
526
LxTIB *tib = (LxTIB *) tibspace;
527
528
529
530
LxTIB2 *tib2 = (LxTIB2 *) (tib + 1);
FIXME("This is probably 50% wrong");
tib->tib_pexchain = NULL;
531
532
tib->tib_pstack = topOfStack - stacklen;
tib->tib_pstacklimit = (void *) topOfStack;
533
534
535
536
tib->tib_ptib2 = tib2;
tib->tib_version = 20; // !!! FIXME
tib->tib_ordinal = 79; // !!! FIXME
537
tib2->tib2_ultid = tid;
538
539
540
541
542
543
544
545
546
tib2->tib2_ulpri = 512;
tib2->tib2_version = 20;
tib2->tib2_usMCCount = 0;
tib2->tib2_fMCForceFlag = 0;
// !!! FIXME: I barely know what I'm doing here, this could all be wrong.
struct user_desc entry;
entry.entry_number = -1;
entry.base_addr = (unsigned int) ((size_t)tib);
547
entry.limit = LXTIBSIZE;
548
549
550
551
552
553
554
555
556
entry.seg_32bit = 1;
entry.contents = MODIFY_LDT_CONTENTS_DATA;
entry.read_exec_only = 0;
entry.limit_in_pages = 0;
entry.seg_not_present = 0;
entry.useable = 1;
const long rc = syscall(SYS_set_thread_area, &entry);
assert(rc == 0); FIXME("this can legit fail, though!");
557
558
559
// The "<< 3 | 3" makes this a GDT selector at ring 3 permissions.
// If this did "| 7" instead of "| 3", it'd be an LDT selector.
// Use findSelector() or allocateSelector() for LDT entries, though!
560
561
const unsigned int segment = (entry.entry_number << 3) | 3;
__asm__ __volatile__ ( "movw %%ax, %%fs \n\t" : : "a" (segment) );
562
return (uint16) entry.entry_number;
563
564
} // initOs2Tib
565
566
567
568
569
570
571
572
573
574
575
576
static void deinitOs2Tib(const uint16 selector)
{
// !!! FIXME: I barely know what I'm doing here, this could all be wrong.
struct user_desc entry;
memset(&entry, '\0', sizeof (entry));
entry.entry_number = selector;
entry.read_exec_only = 1;
entry.seg_not_present = 1;
const long rc = syscall(SYS_set_thread_area, &entry);
assert(rc == 0); FIXME("this can legit fail, though!");
} // deinitOs2Tib
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
static void freeLxModule(LxModule *lxmod);
static __attribute__((noreturn)) void terminate(const uint32 exitcode)
{
freeLxModule(GLoaderState->main_module);
// clear out anything that is still loaded...
while (GLoaderState->loaded_modules) {
LxModule *lowest_mod = GLoaderState->loaded_modules;
for (LxModule *i = GLoaderState->loaded_modules; i; i = i->next) {
if (i->refcount < lowest_mod->refcount)
lowest_mod = i;
} // for
freeLxModule(lowest_mod);
} // while
// OS/2's docs say this only keeps the lower 16 bits of exitcode.
// !!! FIXME: ...but Unix only keeps the lowest 8 bits. Will have to
// !!! FIXME: tapdance to pass larger values back to OS/2 parent processes.
if (exitcode > 255)
FIXME("deal with process exit codes > 255. We clamped this one!");
_exit((int) (exitcode & 0xFF));
} // terminate
static __attribute__((noreturn)) void endLxProcess(const uint32 exitcode)
{
if (GLoaderState->dosExit)
GLoaderState->dosExit(1, exitcode); // let exit lists run. Should call terminate!
terminate(exitcode); // just in case.
} // endLxProcess
609
610
611
612
613
static void missingEntryPointCalled(const char *module, const char *entry)
{
fflush(stdout);
fflush(stderr);
614
fprintf(stderr, "\n\nMissing entry point '%s' in module '%s'! Aborting.\n\n\n", entry, module);
615
616
//STUBBED("output backtrace");
fflush(stderr);
617
618
terminate(1);
} // missingEntryPointCalled
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
static void *generateMissingTrampoline(const char *_module, const char *_entry)
{
static void *page = NULL;
static uint32 pageused = 0;
static uint32 pagesize = 0;
if (pagesize == 0)
pagesize = getpagesize();
if ((!page) || ((pagesize - pageused) < 32))
{
if (page)
mprotect(page, pagesize, PROT_READ | PROT_EXEC);
page = mmap(NULL, pagesize, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
pageused = 0;
} // if
void *trampoline = page + pageused;
char *ptr = (char *) trampoline;
char *module = strdup(_module);
char *entry = strdup(_entry);
*(ptr++) = 0x55; // pushl %ebp
*(ptr++) = 0x89; // movl %esp,%ebp
*(ptr++) = 0xE5; // ...movl %esp,%ebp
*(ptr++) = 0x68; // pushl immediate
memcpy(ptr, &entry, sizeof (char *));
ptr += sizeof (uint32);
*(ptr++) = 0x68; // pushl immediate
memcpy(ptr, &module, sizeof (char *));
ptr += sizeof (uint32);
*(ptr++) = 0xB8; // movl immediate to %eax
const void *fn = missingEntryPointCalled;
memcpy(ptr, &fn, sizeof (void *));
ptr += sizeof (void *);
*(ptr++) = 0xFF; // call absolute in %eax.
*(ptr++) = 0xD0; // ...call absolute in %eax.
const uint32 trampoline_len = (uint32) (ptr - ((char *) trampoline));
assert(trampoline_len <= 32);
pageused += trampoline_len;
if (pageused % 4) // keep these aligned to 32 bits.
pageused += (4 - (pageused % 4));
//printf("Generated trampoline %p for module '%s' export '%s'\n", trampoline, module, entry);
return trampoline;
} // generateMissingTrampoline
static void *generateMissingTrampoline16(const char *_module, const char *_entry, const LxExport **_lxexp)
{
static void *page = NULL;
static uint32 pageused = 0;
const uint32 pagesize = 0x10000;
static uint16 selector = 0xFFFF;
if ((!page) || ((pagesize - pageused) < 64))
{
if (page)
mprotect(page, pagesize, PROT_READ | PROT_EXEC);
page = mmap(NULL, pagesize * 2, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
const size_t diff = (((size_t)page) % 0x10000);
if (diff) { // align to 64k, unmap the difference.
void *origpage = page;
page = ((uint8 *) page) + (0x10000 - diff);
munmap(origpage, 0x10000 - diff);
} else {
munmap(((uint8 *)page) + 0x10000, 0x10000); // unmap the extra half we didn't need.
} // else
pageused = 0;
uint16 offset = 0xFFFF;
selector = 0xFFFF;
findSelector((uint32) page, &selector, &offset);
assert(selector != 0xFFFF);
assert(offset == 0);
} // if
void *trampoline = page + pageused;
char *ptr = (char *) trampoline;
char *module = strdup(_module);
char *entry = strdup(_entry);
// convert stack to linear address...
*(ptr++) = 0x8C; /* mov ax,ss... */
*(ptr++) = 0xD0; /* ...mov ax,ss */
*(ptr++) = 0xC1; /* shr ax,byte 0x3... */
*(ptr++) = 0xE8; /* ...shr ax,byte 0x3 */
*(ptr++) = 0x03; /* ...shr ax,byte 0x3 */
*(ptr++) = 0x66; /* shl eax,byte 0x10... */
*(ptr++) = 0xC1; /* ...shl eax,byte 0x10 */
*(ptr++) = 0xE0; /* ...shl eax,byte 0x10 */
*(ptr++) = 0x10; /* ...shl eax,byte 0x10 */
*(ptr++) = 0x89; /* mov ax,sp... */
*(ptr++) = 0xE0; /* ...mov ax,sp */
// jmp to 32-bit code...
*(ptr++) = 0x66; /* jmp dword 0x7788:0x33332222... */
*(ptr++) = 0xEA; /* ...jmp dword 0x7788:0x33332222 */
const uint32 jmp32addr = (uint32) (ptr + 6);
memcpy(ptr, &jmp32addr, 4); ptr += 4;
memcpy(ptr, &GLoaderState->original_cs, 2); ptr += 2;
// now we're in 32-bit land again.
*(ptr++) = 0x66; /* mov cx,0xabcd... */
*(ptr++) = 0xB9; /* ...mov cx,0xabcd */
memcpy(ptr, &GLoaderState->original_ss, 2); ptr += 2;
*(ptr++) = 0x8E; /* mov ss,ecx... */
*(ptr++) = 0xD1; /* ...mov ss,ecx */
*(ptr++) = 0x89; /* mov esp,eax... */
*(ptr++) = 0xC4; /* ...mov esp,eax */
*(ptr++) = 0x66; /* mov cx,0x8888... */
*(ptr++) = 0xB9; /* ...mov cx,0x8888 */
memcpy(ptr, &GLoaderState->original_ds, 2); ptr += 2;
733
734
735
736
737
*(ptr++) = 0x66; /* mov cx,0x8888... */
*(ptr++) = 0xB9; /* ...mov cx,0x8888 */
memcpy(ptr, &GLoaderState->original_es, 2); ptr += 2;
*(ptr++) = 0x8E; /* mov es,ecx... */ \
*(ptr++) = 0xC1; /* ...mov es,ecx */ \
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
// okay, CPU is in a sane state again, call the trampoline. Don't bother cleaning up.
*(ptr++) = 0x68; // pushl immediate
memcpy(ptr, &entry, sizeof (char *));
ptr += sizeof (uint32);
*(ptr++) = 0x68; // pushl immediate
memcpy(ptr, &module, sizeof (char *));
ptr += sizeof (uint32);
*(ptr++) = 0xB8; // movl immediate to %eax
const void *fn = missingEntryPointCalled;
memcpy(ptr, &fn, sizeof (void *));
ptr += sizeof (void *);
*(ptr++) = 0xFF; // call absolute in %eax.
*(ptr++) = 0xD0; // ...call absolute in %eax.
// (and never return.)
const uint32 trampoline_len = (uint32) (ptr - ((char *) trampoline));
assert(trampoline_len <= 64);
pageused += trampoline_len;
if (pageused % 4) // keep these aligned to 32 bits.
pageused += (4 - (pageused % 4));
//printf("Generated trampoline %p for module '%s' 16-bit export '%s'\n", trampoline, module, entry);
// this hack is not thread safe and only works at all because we don't store this long-term.
static LxExport lxexp;
static LxMmaps lxmmap;
lxexp.addr = trampoline;
lxexp.object = &lxmmap;
lxmmap.mapped = lxmmap.addr = page;
lxmmap.size = 0x10000;
lxmmap.alias = selector;
*_lxexp = &lxexp;
return trampoline;
} // generateMissingTrampoline16
776
static __attribute__((noreturn)) void runLxModule(LxModule *lxmod, const int argc, char **argv, char **envp)
777
{
778
779
uint8 *stack = (uint8 *) ((size_t) lxmod->esp);
780
// ...and you pass it the pointer to argv0. This is (at least as far as the docs suggest) appended to the environment table.
781
//fprintf(stderr, "jumping into LX land for exe '%s'...! eip=%p esp=%p\n", lxmod->name, (void *) lxmod->eip, stack); fflush(stderr);
782
783
784
GLoaderState->running = 1;
785
786
787
788
789
__asm__ __volatile__ (
"movl %%esi,%%esp \n\t" // use the OS/2 process's stack.
"pushl %%eax \n\t" // cmd
"pushl %%ecx \n\t" // env
"pushl $0 \n\t" // reserved
790
"pushl %%edx \n\t" // module handle
791
792
793
794
795
796
797
798
799
800
801
802
"leal 1f,%%eax \n\t" // address that entry point should return to.
"pushl %%eax \n\t"
"pushl %%edi \n\t" // the OS/2 process entry point (we'll "ret" to it instead of jmp, so stack and registers are all correct).
"xorl %%eax,%%eax \n\t"
"xorl %%ebx,%%ebx \n\t"
"xorl %%ecx,%%ecx \n\t"
"xorl %%edx,%%edx \n\t"
"xorl %%esi,%%esi \n\t"
"xorl %%edi,%%edi \n\t"
"xorl %%ebp,%%ebp \n\t"
"ret \n\t" // go to OS/2 land!
"1: \n\t" // ...and return here.
803
804
805
806
"pushl %%eax \n\t" // push exit code from OS/2 app.
"call endLxProcess \n\t" // never returns.
// If we returned here, %eax has the exit code from the app.
: // no outputs.
807
808
809
: "a" (GLoaderState->pib.pib_pchcmd),
"c" (GLoaderState->pib.pib_pchenv),
"d" (lxmod), "S" (stack), "D" (lxmod->eip)
810
811
812
813
814
815
: "memory"
);
__builtin_unreachable();
} // runLxModule
816
817
static void runLxLibraryInitOrTerm(LxModule *lxmod, const int isTermination)
{
818
819
820
uint8 *stack = NULL;
// force us over to OS/2 main module's stack if we aren't already on it.
821
if (!GLoaderState->running)
822
stack = (uint8 *) ((size_t) GLoaderState->main_module->esp);
823
824
//fprintf(stderr, "jumping into LX land to %s library '%s'...! eip=%p esp=%p\n", isTermination ? "terminate" : "initialize", lxmod->name, (void *) lxmod->eip, stack); fflush(stderr);
825
826
827
828
829
__asm__ __volatile__ (
"pushal \n\t" // save all the current registers.
"pushfl \n\t" // save all the current flags.
"movl %%esp,%%ecx \n\t" // save our stack to a temporary register.
830
831
"testl %%esi,%%esi \n\t" // force the OS/2 process's stack?
"je 1f \n\t" // if 0, nope, we're already good.
832
"movl %%esi,%%esp \n\t" // use the OS/2 process's stack.
833
"1: \n\t" //
834
835
"pushl %%ecx \n\t" // save original stack pointer for real.
"pushl %%eax \n\t" // isTermination
836
"pushl %%edx \n\t" // library module handle
837
"leal 2f,%%eax \n\t" // address that entry point should return to.
838
839
840
841
842
843
844
845
846
847
"pushl %%eax \n\t"
"pushl %%edi \n\t" // the OS/2 library entry point (we'll "ret" to it instead of jmp, so stack and registers are all correct).
"xorl %%eax,%%eax \n\t"
"xorl %%ebx,%%ebx \n\t"
"xorl %%ecx,%%ecx \n\t"
"xorl %%edx,%%edx \n\t"
"xorl %%esi,%%esi \n\t"
"xorl %%edi,%%edi \n\t"
"xorl %%ebp,%%ebp \n\t"
"ret \n\t" // go to OS/2 land!
848
849
"2: \n\t" // ...and return here.
"addl $8,%%esp \n\t" // drop arguments to entry point.
850
851
852
853
"popl %%esp \n\t" // restore native process stack now.
"popfl \n\t" // restore our original flags.
"popal \n\t" // restore our original registers.
: // no outputs
854
: "a" (isTermination), "d" (lxmod), "S" (stack), "D" (lxmod->eip)
855
856
857
: "memory"
);
858
//fprintf(stderr, "...survived time in LX land!\n"); fflush(stderr);
859
860
// !!! FIXME: this entry point returns a result...do we abort if it reports error?
861
// !!! FIXME: (actually, DosLoadModule() can report that failure. Abort if (GLoaderState->running == 0), though!)
862
863
} // runLxLibraryInitOrTerm
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
static void runLxLibraryInit(LxModule *lxmod)
{
// we don't check the LIBINIT flag, because if the EIP fields are valid
// but this bit is missing, we're supposed to do "global" initialization,
// which I guess means it runs once for all processes on the system
// instead of once per process...but we aren't really a full OS/2 system
// so maybe it's okay to do global init in isolation...? We'll cross that
// bridge when we come to it, I guess. Either way: this is running here,
// flag or not.
if (1) { //if (lxmod->module_flags & 0x4) { // LIBINIT flag
if (lxmod->lx.eip_object != 0)
runLxLibraryInitOrTerm(lxmod, 0);
} // if
} // runLxLibraryInit
static void runLxLibraryTerm(LxModule *lxmod)
{
// See notes about global initialization in runLxLibraryInit(); it
// applies to deinit here, too.
if (1) { //if (lxmod->module_flags & 0x40000000) { // LIBTERM flag
if (lxmod->lx.eip_object != 0)
runLxLibraryInitOrTerm(lxmod, 1);
} // if
} // runLxLibraryTerm
888
889
890
891
892
893
894
895
static void freeLxModule(LxModule *lxmod)
{
if (!lxmod)
return;
// !!! FIXME: mutex from here
lxmod->refcount--;
896
//fprintf(stderr, "unref'd module '%s' to %u\n", lxmod->name, (uint) lxmod->refcount);
897
898
899
if (lxmod->refcount > 0)
return; // something is still using it.
900
if ((lxmod->initialized) && (lxmod != GLoaderState->main_module)) {
901
if (!lxmod->nativelib) {
902
runLxLibraryTerm(lxmod);
903
904
905
906
907
908
} else {
LxNativeModuleDeinitEntryPoint fn = (LxNativeModuleDeinitEntryPoint) dlsym(lxmod->nativelib, "lxNativeModuleDeinit");
if (fn)
fn();
} // else
} // if
909
910
911
912
913
914
915
if (lxmod->next)
lxmod->next->prev = lxmod->prev;
if (lxmod->prev)
lxmod->prev->next = lxmod->next;
916
917
if (lxmod == GLoaderState->loaded_modules)
GLoaderState->loaded_modules = lxmod->next;
918
919
if (GLoaderState->main_module == lxmod) {
920
GLoaderState->main_module = NULL;
921
922
923
924
925
LxMmaps *lxmmap = &lxmod->mmaps[lxmod->lx.esp_object - 1];
const uint32 stackbase = (uint32) ((size_t)lxmmap->addr);
initOs2StackSegments(stackbase, lxmmap->size, 1);
lxmmap->mapped = NULL; // don't unmap the stack we're probably using!
} // if
926
927
// !!! FIXME: mutex to here
928
929
for (uint32 i = lxmod->lx.num_import_mod_entries; i > 0; i--)
freeLxModule(lxmod->dependencies[i-1]);
930
931
932
free(lxmod->dependencies);
for (uint32 i = 0; i < lxmod->lx.module_num_objects; i++) {
933
934
if (lxmod->mmaps[i].alias != 0xFFFF)
freeSelector(lxmod->mmaps[i].alias);
935
if (lxmod->mmaps[i].mapped)
936
munmap(lxmod->mmaps[i].mapped, lxmod->mmaps[i].size);
937
} // for
938
free(lxmod->mmaps);
939
940
941
if (lxmod->nativelib)
dlclose(lxmod->nativelib);
942
943
944
945
946
else {
for (uint32 i = 0; i < lxmod->num_exports; i++)
free((void *) lxmod->exports[i].name);
free((void *) lxmod->exports);
} // else
947
948
949
950
951
952
free(lxmod);
} // freeLxModule
static LxModule *loadLxModuleByModuleNameInternal(const char *modname, const int dependency_tree_depth);
953
static void *getModuleProcAddrByOrdinal(const LxModule *module, const uint32 ordinal, const LxExport **_lxexp, const int want16bit)
954
{
955
//printf("lookup module == '%s', ordinal == %u\n", module->name, (uint) ordinal);
956
957
958
const LxExport *lxexp = module->exports;
for (uint32 i = 0; i < module->num_exports; i++, lxexp++) {
959
960
961
962
963
964
965
966
if (lxexp->ordinal == ordinal) {
if (_lxexp)
*_lxexp = lxexp;
void *retval = lxexp->addr;
if (lxexp->object && lxexp->object->alias != 0xFFFF) // 16-bit bridge? this is actually a void**, due to some macro salsa. :/
retval = *((void**) retval);
return retval;
} // if
967
968
} // for
969
970
971
if (_lxexp)
*_lxexp = NULL;
972
973
#if 1
char entry[128];
974
snprintf(entry, sizeof (entry), "ordinal #%u", (uint) ordinal);
975
return want16bit ? generateMissingTrampoline16(module->name, entry, _lxexp) : generateMissingTrampoline(module->name, entry);
976
#else
977
return NULL;
978
#endif
979
980
} // getModuleProcAddrByOrdinal
981
static void *getModuleProcAddrByName(const LxModule *module, const char *name, const LxExport **_lxexp, const int want16bit)
982
{
983
//printf("lookup module == '%s', name == '%s'\n", module->name, name);
984
985
986
const LxExport *lxexp = module->exports;
for (uint32 i = 0; i < module->num_exports; i++, lxexp++) {
987
988
989
990
991
992
993
994
if (strcmp(lxexp->name, name) == 0) {
if (_lxexp)
*_lxexp = lxexp;
void *retval = lxexp->addr;
if (lxexp->object && lxexp->object->alias != 0xFFFF) // 16-bit bridge? this is actually a void**, due to some macro salsa. :/
retval = *((void**) retval);
return retval;
} // if
995
996
} // for
997
998
999
if (_lxexp)
*_lxexp = NULL;
1000
#if 1