xref: /qemu/linux-user/flatload.c (revision 513823e7521a09ed7ad1e32e6454bac3b2cbf52d)
1 /****************************************************************************/
2 /*
3  *  QEMU bFLT binary loader.  Based on linux/fs/binfmt_flat.c
4  *
5  *  This program is free software; you can redistribute it and/or modify
6  *  it under the terms of the GNU General Public License as published by
7  *  the Free Software Foundation; either version 2 of the License, or
8  *  (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  *  GNU General Public License for more details.
14  *
15  *  You should have received a copy of the GNU General Public License
16  *  along with this program; if not, see <http://www.gnu.org/licenses/>.
17  *
18  *      Copyright (C) 2006 CodeSourcery.
19  *	Copyright (C) 2000-2003 David McCullough <davidm@snapgear.com>
20  *	Copyright (C) 2002 Greg Ungerer <gerg@snapgear.com>
21  *	Copyright (C) 2002 SnapGear, by Paul Dale <pauli@snapgear.com>
22  *	Copyright (C) 2000, 2001 Lineo, by David McCullough <davidm@lineo.com>
23  *  based heavily on:
24  *
25  *  linux/fs/binfmt_aout.c:
26  *      Copyright (C) 1991, 1992, 1996  Linus Torvalds
27  *  linux/fs/binfmt_flat.c for 2.0 kernel
28  *	    Copyright (C) 1998  Kenneth Albanowski <kjahds@kjahds.com>
29  *	JAN/99 -- coded full program relocation (gerg@snapgear.com)
30  */
31 
32 /****************************************************************************/
33 
34 #include "qemu/osdep.h"
35 
36 #include "qemu.h"
37 #include "exec/page-protection.h"
38 #include "user-internals.h"
39 #include "loader.h"
40 #include "user-mmap.h"
41 #include "flat.h"
42 #include "target_flat.h"
43 
44 //#define DEBUG
45 
46 #ifdef DEBUG
47 #define	DBG_FLT(...)	printf(__VA_ARGS__)
48 #else
49 #define	DBG_FLT(...)
50 #endif
51 
52 #define RELOC_FAILED 0xff00ff01		/* Relocation incorrect somewhere */
53 #define UNLOADED_LIB 0x7ff000ff		/* Placeholder for unused library */
54 
55 struct lib_info {
56     abi_ulong start_code;       /* Start of text segment */
57     abi_ulong start_data;       /* Start of data segment */
58     abi_ulong end_data;         /* Start of bss section */
59     abi_ulong start_brk;        /* End of data segment */
60     abi_ulong text_len;	        /* Length of text segment */
61     abi_ulong entry;	        /* Start address for this module */
62     abi_ulong build_date;       /* When this one was compiled */
63     short loaded;		/* Has this library been loaded? */
64 };
65 
66 struct linux_binprm;
67 
68 /****************************************************************************/
69 /*
70  * create_flat_tables() parses the env- and arg-strings in new user
71  * memory and creates the pointer tables from them, and puts their
72  * addresses on the "stack", returning the new stack pointer value.
73  */
74 
75 /* Push a block of strings onto the guest stack.  */
76 static abi_ulong copy_strings(abi_ulong p, int n, char **s)
77 {
78     int len;
79 
80     while (n-- > 0) {
81         len = strlen(s[n]) + 1;
82         p -= len;
83         memcpy_to_target(p, s[n], len);
84     }
85 
86     return p;
87 }
88 
89 static int target_pread(int fd, abi_ulong ptr, abi_ulong len,
90                         abi_ulong offset)
91 {
92     void *buf;
93     int ret;
94 
95     buf = lock_user(VERIFY_WRITE, ptr, len, 0);
96     if (!buf) {
97         return -EFAULT;
98     }
99     ret = pread(fd, buf, len, offset);
100     if (ret < 0) {
101         ret = -errno;
102     }
103     unlock_user(buf, ptr, len);
104     return ret;
105 }
106 
107 /****************************************************************************/
108 
109 static abi_ulong
110 calc_reloc(abi_ulong r, struct lib_info *p, int curid, int internalp)
111 {
112     abi_ulong addr;
113     int id;
114     abi_ulong start_brk;
115     abi_ulong start_data;
116     abi_ulong text_len;
117     abi_ulong start_code;
118 
119     id = 0;
120 
121     start_brk = p[id].start_brk;
122     start_data = p[id].start_data;
123     start_code = p[id].start_code;
124     text_len = p[id].text_len;
125 
126     if (!flat_reloc_valid(r, start_brk - start_data + text_len)) {
127         fprintf(stderr, "BINFMT_FLAT: reloc outside program 0x%x "
128                 "(0 - 0x%x/0x%x)\n",
129                (int) r,(int)(start_brk-start_code),(int)text_len);
130         goto failed;
131     }
132 
133     if (r < text_len)			/* In text segment */
134         addr = r + start_code;
135     else					/* In data segment */
136         addr = r - text_len + start_data;
137 
138     /* Range checked already above so doing the range tests is redundant...*/
139     return(addr);
140 
141 failed:
142     abort();
143     return RELOC_FAILED;
144 }
145 
146 /****************************************************************************/
147 
148 /* ??? This does not handle endianness correctly.  */
149 static void old_reloc(struct lib_info *libinfo, uint32_t rl)
150 {
151 #ifdef DEBUG
152 	const char *segment[] = { "TEXT", "DATA", "BSS", "*UNKNOWN*" };
153 #endif
154 	uint32_t *ptr;
155         uint32_t offset;
156         int reloc_type;
157 
158         offset = rl & 0x3fffffff;
159         reloc_type = rl >> 30;
160         /* ??? How to handle this?  */
161 #if defined(CONFIG_COLDFIRE)
162 	ptr = (uint32_t *) ((unsigned long) libinfo->start_code + offset);
163 #else
164 	ptr = (uint32_t *) ((unsigned long) libinfo->start_data + offset);
165 #endif
166 
167 #ifdef DEBUG
168 	fprintf(stderr, "Relocation of variable at DATASEG+%x "
169 		"(address %p, currently %x) into segment %s\n",
170 		offset, ptr, (int)*ptr, segment[reloc_type]);
171 #endif
172 
173 	switch (reloc_type) {
174 	case OLD_FLAT_RELOC_TYPE_TEXT:
175 		*ptr += libinfo->start_code;
176 		break;
177 	case OLD_FLAT_RELOC_TYPE_DATA:
178 		*ptr += libinfo->start_data;
179 		break;
180 	case OLD_FLAT_RELOC_TYPE_BSS:
181 		*ptr += libinfo->end_data;
182 		break;
183 	default:
184 		fprintf(stderr, "BINFMT_FLAT: Unknown relocation type=%x\n",
185                         reloc_type);
186 		break;
187 	}
188 	DBG_FLT("Relocation became %x\n", (int)*ptr);
189 }
190 
191 /****************************************************************************/
192 
193 static int load_flat_file(struct linux_binprm * bprm,
194 		struct lib_info *libinfo, int id, abi_ulong *extra_stack)
195 {
196     struct flat_hdr * hdr;
197     abi_ulong textpos = 0, datapos = 0;
198     abi_long result;
199     abi_ulong realdatastart = 0;
200     abi_ulong text_len, data_len, bss_len, stack_len, flags;
201     abi_ulong extra;
202     abi_ulong reloc = 0, rp;
203     int i, rev, relocs = 0;
204     abi_ulong fpos;
205     abi_ulong start_code;
206     abi_ulong indx_len;
207 
208     hdr = ((struct flat_hdr *) bprm->buf);		/* exec-header */
209 
210     text_len  = ntohl(hdr->data_start);
211     data_len  = ntohl(hdr->data_end) - ntohl(hdr->data_start);
212     bss_len   = ntohl(hdr->bss_end) - ntohl(hdr->data_end);
213     stack_len = ntohl(hdr->stack_size);
214     if (extra_stack) {
215         stack_len += *extra_stack;
216         *extra_stack = stack_len;
217     }
218     relocs    = ntohl(hdr->reloc_count);
219     flags     = ntohl(hdr->flags);
220     rev       = ntohl(hdr->rev);
221 
222     DBG_FLT("BINFMT_FLAT: Loading file: %s\n", bprm->filename);
223 
224     if (rev != FLAT_VERSION && rev != OLD_FLAT_VERSION) {
225         fprintf(stderr, "BINFMT_FLAT: bad magic/rev (0x%x, need 0x%x)\n",
226                 rev, (int) FLAT_VERSION);
227         return -ENOEXEC;
228     }
229 
230     /* Don't allow old format executables to use shared libraries */
231     if (rev == OLD_FLAT_VERSION && id != 0) {
232         fprintf(stderr, "BINFMT_FLAT: shared libraries are not available\n");
233         return -ENOEXEC;
234     }
235 
236     /*
237      * fix up the flags for the older format,  there were all kinds
238      * of endian hacks,  this only works for the simple cases
239      */
240     if (rev == OLD_FLAT_VERSION && flat_old_ram_flag(flags))
241         flags = FLAT_FLAG_RAM;
242 
243     if (flags & (FLAT_FLAG_GZIP|FLAT_FLAG_GZDATA)) {
244         fprintf(stderr, "ZFLAT executables are not supported\n");
245         return -ENOEXEC;
246     }
247 
248     /*
249      * calculate the extra space we need to map in
250      */
251     extra = relocs * sizeof(abi_ulong);
252     if (extra < bss_len + stack_len)
253         extra = bss_len + stack_len;
254 
255     /* Add space for library base pointers.  Make sure this does not
256        misalign the  doesn't misalign the data segment.  */
257     indx_len = MAX_SHARED_LIBS * sizeof(abi_ulong);
258     indx_len = (indx_len + 15) & ~(abi_ulong)15;
259 
260     /*
261      * Allocate the address space.
262      */
263     probe_guest_base(bprm->filename, 0,
264                      text_len + data_len + extra + indx_len - 1);
265 
266     /*
267      * there are a couple of cases here,  the separate code/data
268      * case,  and then the fully copied to RAM case which lumps
269      * it all together.
270      */
271     if ((flags & (FLAT_FLAG_RAM|FLAT_FLAG_GZIP)) == 0) {
272         /*
273          * this should give us a ROM ptr,  but if it doesn't we don't
274          * really care
275          */
276         DBG_FLT("BINFMT_FLAT: ROM mapping of file (we hope)\n");
277 
278         textpos = target_mmap(0, text_len, PROT_READ|PROT_EXEC,
279                               MAP_PRIVATE, bprm->src.fd, 0);
280         if (textpos == -1) {
281             fprintf(stderr, "Unable to mmap process text\n");
282             return -1;
283         }
284 
285         realdatastart = target_mmap(0, data_len + extra + indx_len,
286                                     PROT_READ|PROT_WRITE|PROT_EXEC,
287                                     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
288 
289         if (realdatastart == -1) {
290             fprintf(stderr, "Unable to allocate RAM for process data\n");
291             return realdatastart;
292         }
293         datapos = realdatastart + indx_len;
294 
295         DBG_FLT("BINFMT_FLAT: Allocated data+bss+stack (%d bytes): %x\n",
296                         (int)(data_len + bss_len + stack_len), (int)datapos);
297 
298         fpos = ntohl(hdr->data_start);
299         result = target_pread(bprm->src.fd, datapos,
300                               data_len + (relocs * sizeof(abi_ulong)),
301                               fpos);
302         if (result < 0) {
303             fprintf(stderr, "Unable to read data+bss\n");
304             return result;
305         }
306 
307         reloc = datapos + (ntohl(hdr->reloc_start) - text_len);
308 
309     } else {
310 
311         textpos = target_mmap(0, text_len + data_len + extra + indx_len,
312                               PROT_READ | PROT_EXEC | PROT_WRITE,
313                               MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
314         if (textpos == -1 ) {
315             fprintf(stderr, "Unable to allocate RAM for process text/data\n");
316             return -1;
317         }
318 
319         realdatastart = textpos + ntohl(hdr->data_start);
320         datapos = realdatastart + indx_len;
321         reloc = (textpos + ntohl(hdr->reloc_start) + indx_len);
322 
323         result = target_pread(bprm->src.fd, textpos,
324                               text_len, 0);
325         if (result >= 0) {
326             result = target_pread(bprm->src.fd, datapos,
327                                   data_len + (relocs * sizeof(abi_ulong)),
328                                   ntohl(hdr->data_start));
329         }
330         if (result < 0) {
331             fprintf(stderr, "Unable to read code+data+bss\n");
332             return result;
333         }
334     }
335 
336     DBG_FLT("Mapping is 0x%x, Entry point is 0x%x, data_start is 0x%x\n",
337             (int)textpos, 0x00ffffff&ntohl(hdr->entry),
338             ntohl(hdr->data_start));
339 
340     /* The main program needs a little extra setup in the task structure */
341     start_code = textpos + sizeof (struct flat_hdr);
342 
343     DBG_FLT("%s %s: TEXT=%x-%x DATA=%x-%x BSS=%x-%x\n",
344             id ? "Lib" : "Load", bprm->filename,
345             (int) start_code, (int) (textpos + text_len),
346             (int) datapos,
347             (int) (datapos + data_len),
348             (int) (datapos + data_len),
349             (int) (((datapos + data_len + bss_len) + 3) & ~3));
350 
351     text_len -= sizeof(struct flat_hdr); /* the real code len */
352 
353     /* Store the current module values into the global library structure */
354     libinfo[id].start_code = start_code;
355     libinfo[id].start_data = datapos;
356     libinfo[id].end_data = datapos + data_len;
357     libinfo[id].start_brk = datapos + data_len + bss_len;
358     libinfo[id].text_len = text_len;
359     libinfo[id].loaded = 1;
360     libinfo[id].entry = (0x00ffffff & ntohl(hdr->entry)) + textpos;
361     libinfo[id].build_date = ntohl(hdr->build_date);
362 
363     /*
364      * We just load the allocations into some temporary memory to
365      * help simplify all this mumbo jumbo
366      *
367      * We've got two different sections of relocation entries.
368      * The first is the GOT which resides at the beginning of the data segment
369      * and is terminated with a -1.  This one can be relocated in place.
370      * The second is the extra relocation entries tacked after the image's
371      * data segment. These require a little more processing as the entry is
372      * really an offset into the image which contains an offset into the
373      * image.
374      */
375     if (flags & FLAT_FLAG_GOTPIC) {
376         rp = datapos;
377         while (1) {
378             abi_ulong addr;
379             if (get_user_ual(addr, rp))
380                 return -EFAULT;
381             if (addr == -1)
382                 break;
383             if (addr) {
384                 addr = calc_reloc(addr, libinfo, id, 0);
385                 if (addr == RELOC_FAILED)
386                     return -ENOEXEC;
387                 if (put_user_ual(addr, rp))
388                     return -EFAULT;
389             }
390             rp += sizeof(abi_ulong);
391         }
392     }
393 
394     /*
395      * Now run through the relocation entries.
396      * We've got to be careful here as C++ produces relocatable zero
397      * entries in the constructor and destructor tables which are then
398      * tested for being not zero (which will always occur unless we're
399      * based from address zero).  This causes an endless loop as __start
400      * is at zero.  The solution used is to not relocate zero addresses.
401      * This has the negative side effect of not allowing a global data
402      * reference to be statically initialised to _stext (I've moved
403      * __start to address 4 so that is okay).
404      */
405     if (rev > OLD_FLAT_VERSION) {
406         abi_ulong persistent = 0;
407         for (i = 0; i < relocs; i++) {
408             abi_ulong addr, relval;
409 
410             /* Get the address of the pointer to be
411                relocated (of course, the address has to be
412                relocated first).  */
413             if (get_user_ual(relval, reloc + i * sizeof(abi_ulong)))
414                 return -EFAULT;
415             relval = ntohl(relval);
416             if (flat_set_persistent(relval, &persistent))
417                 continue;
418             addr = flat_get_relocate_addr(relval);
419             rp = calc_reloc(addr, libinfo, id, 1);
420             if (rp == RELOC_FAILED)
421                 return -ENOEXEC;
422 
423             /* Get the pointer's value.  */
424             if (get_user_ual(addr, rp))
425                 return -EFAULT;
426             addr = flat_get_addr_from_rp(addr, relval, flags, &persistent);
427             if (addr != 0) {
428                 /*
429                  * Do the relocation.  PIC relocs in the data section are
430                  * already in target order
431                  */
432                 if ((flags & FLAT_FLAG_GOTPIC) == 0)
433                     addr = ntohl(addr);
434                 addr = calc_reloc(addr, libinfo, id, 0);
435                 if (addr == RELOC_FAILED)
436                     return -ENOEXEC;
437 
438                 /* Write back the relocated pointer.  */
439                 if (flat_put_addr_at_rp(rp, addr, relval))
440                     return -EFAULT;
441             }
442         }
443     } else {
444         for (i = 0; i < relocs; i++) {
445             abi_ulong relval;
446             if (get_user_ual(relval, reloc + i * sizeof(abi_ulong)))
447                 return -EFAULT;
448             old_reloc(&libinfo[0], relval);
449         }
450     }
451 
452     /* zero the BSS.  */
453     memset(g2h_untagged(datapos + data_len), 0, bss_len);
454 
455     return 0;
456 }
457 
458 
459 /****************************************************************************/
460 int load_flt_binary(struct linux_binprm *bprm, struct image_info *info)
461 {
462     struct lib_info libinfo[MAX_SHARED_LIBS];
463     abi_ulong p;
464     abi_ulong stack_len;
465     abi_ulong start_addr;
466     abi_ulong sp;
467     int res;
468     int i, j;
469 
470     memset(libinfo, 0, sizeof(libinfo));
471     /*
472      * We have to add the size of our arguments to our stack size
473      * otherwise it's too easy for users to create stack overflows
474      * by passing in a huge argument list.  And yes,  we have to be
475      * pedantic and include space for the argv/envp array as it may have
476      * a lot of entries.
477      */
478     stack_len = 0;
479     for (i = 0; i < bprm->argc; ++i) {
480         /* the argv strings */
481         stack_len += strlen(bprm->argv[i]);
482     }
483     for (i = 0; i < bprm->envc; ++i) {
484         /* the envp strings */
485         stack_len += strlen(bprm->envp[i]);
486     }
487     stack_len += (bprm->argc + 1) * 4; /* the argv array */
488     stack_len += (bprm->envc + 1) * 4; /* the envp array */
489 
490 
491     mmap_lock();
492     res = load_flat_file(bprm, libinfo, 0, &stack_len);
493     mmap_unlock();
494 
495     if (is_error(res)) {
496             return res;
497     }
498 
499     /* Update data segment pointers for all libraries */
500     for (i=0; i<MAX_SHARED_LIBS; i++) {
501         if (libinfo[i].loaded) {
502             abi_ulong seg;
503             seg = libinfo[i].start_data;
504             for (j=0; j<MAX_SHARED_LIBS; j++) {
505                 seg -= 4;
506                 /* FIXME - handle put_user() failures */
507                 if (put_user_ual(libinfo[j].loaded
508                                  ? libinfo[j].start_data
509                                  : UNLOADED_LIB,
510                                  seg))
511                     return -EFAULT;
512             }
513         }
514     }
515 
516     p = ((libinfo[0].start_brk + stack_len + 3) & ~3) - 4;
517     DBG_FLT("p=%x\n", (int)p);
518 
519     /* Copy argv/envp.  */
520     p = copy_strings(p, bprm->envc, bprm->envp);
521     p = copy_strings(p, bprm->argc, bprm->argv);
522     /* Align stack.  */
523     sp = p & ~(abi_ulong)(sizeof(abi_ulong) - 1);
524     /* Enforce final stack alignment of 16 bytes.  This is sufficient
525        for all current targets, and excess alignment is harmless.  */
526     stack_len = bprm->envc + bprm->argc + 2;
527     stack_len += flat_argvp_envp_on_stack() ? 2 : 0; /* argv, argp */
528     stack_len += 1; /* argc */
529     stack_len *= sizeof(abi_ulong);
530     sp -= (sp - stack_len) & 15;
531     sp = loader_build_argptr(bprm->envc, bprm->argc, sp, p,
532                              flat_argvp_envp_on_stack());
533 
534     /* Fake some return addresses to ensure the call chain will
535      * initialise library in order for us.  We are required to call
536      * lib 1 first, then 2, ... and finally the main program (id 0).
537      */
538     start_addr = libinfo[0].entry;
539 
540     /* Stash our initial stack pointer into the mm structure */
541     info->start_code = libinfo[0].start_code;
542     info->end_code = libinfo[0].start_code + libinfo[0].text_len;
543     info->start_data = libinfo[0].start_data;
544     info->end_data = libinfo[0].end_data;
545     info->brk = libinfo[0].start_brk;
546     info->start_stack = sp;
547     info->stack_limit = libinfo[0].start_brk;
548     info->entry = start_addr;
549     info->code_offset = info->start_code;
550     info->data_offset = info->start_data - libinfo[0].text_len;
551 
552     DBG_FLT("start_thread(entry=0x%x, start_stack=0x%x)\n",
553             (int)info->entry, (int)info->start_stack);
554 
555     return 0;
556 }
557