xref: /optee_os/ldelf/ta_elf.c (revision 8bbd9b374a51a1b8617796aae8a70c271543357f)
1 // SPDX-License-Identifier: BSD-2-Clause
2 /*
3  * Copyright (c) 2019, Linaro Limited
4  */
5 
6 #include <assert.h>
7 #include <ctype.h>
8 #include <elf32.h>
9 #include <elf64.h>
10 #include <elf_common.h>
11 #include <ldelf.h>
12 #include <pta_system.h>
13 #include <stdio.h>
14 #include <stdlib.h>
15 #include <string_ext.h>
16 #include <string.h>
17 #include <tee_api_types.h>
18 #include <user_ta_header.h>
19 #include <utee_syscalls.h>
20 
21 #include "sys.h"
22 #include "ta_elf.h"
23 #include "unwind.h"
24 
25 static vaddr_t ta_stack;
26 static vaddr_t ta_stack_size;
27 
28 struct ta_elf_queue main_elf_queue = TAILQ_HEAD_INITIALIZER(main_elf_queue);
29 
30 static struct ta_elf *queue_elf(const TEE_UUID *uuid)
31 {
32 	struct ta_elf *elf = NULL;
33 
34 	TAILQ_FOREACH(elf, &main_elf_queue, link)
35 		if (!memcmp(uuid, &elf->uuid, sizeof(*uuid)))
36 			return NULL;
37 
38 	elf = calloc(1, sizeof(*elf));
39 	if (!elf)
40 		err(TEE_ERROR_OUT_OF_MEMORY, "calloc");
41 
42 	TAILQ_INIT(&elf->segs);
43 
44 	elf->uuid = *uuid;
45 	TAILQ_INSERT_TAIL(&main_elf_queue, elf, link);
46 	return elf;
47 }
48 
49 static TEE_Result e32_parse_ehdr(struct ta_elf *elf, Elf32_Ehdr *ehdr)
50 {
51 	if (ehdr->e_ident[EI_VERSION] != EV_CURRENT ||
52 	    ehdr->e_ident[EI_CLASS] != ELFCLASS32 ||
53 	    ehdr->e_ident[EI_DATA] != ELFDATA2LSB ||
54 	    ehdr->e_ident[EI_OSABI] != ELFOSABI_NONE ||
55 	    ehdr->e_type != ET_DYN || ehdr->e_machine != EM_ARM ||
56 	    (ehdr->e_flags & EF_ARM_ABIMASK) != EF_ARM_ABI_VERSION ||
57 #ifndef CFG_WITH_VFP
58 	    (ehdr->e_flags & EF_ARM_ABI_FLOAT_HARD) ||
59 #endif
60 	    ehdr->e_phentsize != sizeof(Elf32_Phdr) ||
61 	    ehdr->e_shentsize != sizeof(Elf32_Shdr))
62 		return TEE_ERROR_BAD_FORMAT;
63 
64 	elf->is_32bit = true;
65 	elf->e_entry = ehdr->e_entry;
66 	elf->e_phoff = ehdr->e_phoff;
67 	elf->e_shoff = ehdr->e_shoff;
68 	elf->e_phnum = ehdr->e_phnum;
69 	elf->e_shnum = ehdr->e_shnum;
70 	elf->e_phentsize = ehdr->e_phentsize;
71 	elf->e_shentsize = ehdr->e_shentsize;
72 
73 	return TEE_SUCCESS;
74 }
75 
76 #ifdef ARM64
77 static TEE_Result e64_parse_ehdr(struct ta_elf *elf, Elf64_Ehdr *ehdr)
78 {
79 	if (ehdr->e_ident[EI_VERSION] != EV_CURRENT ||
80 	    ehdr->e_ident[EI_CLASS] != ELFCLASS64 ||
81 	    ehdr->e_ident[EI_DATA] != ELFDATA2LSB ||
82 	    ehdr->e_ident[EI_OSABI] != ELFOSABI_NONE ||
83 	    ehdr->e_type != ET_DYN || ehdr->e_machine != EM_AARCH64 ||
84 	    ehdr->e_flags || ehdr->e_phentsize != sizeof(Elf64_Phdr) ||
85 	    ehdr->e_shentsize != sizeof(Elf64_Shdr))
86 		return TEE_ERROR_BAD_FORMAT;
87 
88 
89 	elf->is_32bit = false;
90 	elf->e_entry = ehdr->e_entry;
91 	elf->e_phoff = ehdr->e_phoff;
92 	elf->e_shoff = ehdr->e_shoff;
93 	elf->e_phnum = ehdr->e_phnum;
94 	elf->e_shnum = ehdr->e_shnum;
95 	elf->e_phentsize = ehdr->e_phentsize;
96 	elf->e_shentsize = ehdr->e_shentsize;
97 
98 	return TEE_SUCCESS;
99 }
100 #else /*ARM64*/
101 static TEE_Result e64_parse_ehdr(struct ta_elf *elf __unused,
102 				 Elf64_Ehdr *ehdr __unused)
103 {
104 	return TEE_ERROR_NOT_SUPPORTED;
105 }
106 #endif /*ARM64*/
107 
108 static void read_dyn(struct ta_elf *elf, vaddr_t addr,
109 		     size_t idx, unsigned int *tag, size_t *val)
110 {
111 	if (elf->is_32bit) {
112 		Elf32_Dyn *dyn = (Elf32_Dyn *)(addr + elf->load_addr);
113 
114 		*tag = dyn[idx].d_tag;
115 		*val = dyn[idx].d_un.d_val;
116 	} else {
117 		Elf64_Dyn *dyn = (Elf64_Dyn *)(addr + elf->load_addr);
118 
119 		*tag = dyn[idx].d_tag;
120 		*val = dyn[idx].d_un.d_val;
121 	}
122 }
123 
124 static void save_hashtab_from_segment(struct ta_elf *elf, unsigned int type,
125 				      vaddr_t addr, size_t memsz)
126 {
127 	size_t dyn_entsize = 0;
128 	size_t num_dyns = 0;
129 	size_t n = 0;
130 	unsigned int tag = 0;
131 	size_t val = 0;
132 
133 	if (type != PT_DYNAMIC)
134 		return;
135 
136 	if (elf->is_32bit)
137 		dyn_entsize = sizeof(Elf32_Dyn);
138 	else
139 		dyn_entsize = sizeof(Elf64_Dyn);
140 
141 	assert(!(memsz % dyn_entsize));
142 	num_dyns = memsz / dyn_entsize;
143 
144 	for (n = 0; n < num_dyns; n++) {
145 		read_dyn(elf, addr, n, &tag, &val);
146 		if (tag == DT_HASH) {
147 			elf->hashtab = (void *)(val + elf->load_addr);
148 			break;
149 		}
150 	}
151 }
152 
153 static void save_hashtab(struct ta_elf *elf)
154 {
155 	size_t n = 0;
156 
157 	if (elf->is_32bit) {
158 		Elf32_Phdr *phdr = elf->phdr;
159 
160 		for (n = 0; n < elf->e_phnum; n++)
161 			save_hashtab_from_segment(elf, phdr[n].p_type,
162 						  phdr[n].p_vaddr,
163 						  phdr[n].p_memsz);
164 	} else {
165 		Elf64_Phdr *phdr = elf->phdr;
166 
167 		for (n = 0; n < elf->e_phnum; n++)
168 			save_hashtab_from_segment(elf, phdr[n].p_type,
169 						  phdr[n].p_vaddr,
170 						  phdr[n].p_memsz);
171 	}
172 	assert(elf->hashtab);
173 }
174 
175 static void e32_save_symtab(struct ta_elf *elf, size_t tab_idx)
176 {
177 	Elf32_Shdr *shdr = elf->shdr;
178 	size_t str_idx = shdr[tab_idx].sh_link;
179 
180 	elf->dynsymtab = (void *)(shdr[tab_idx].sh_addr + elf->load_addr);
181 	assert(!(shdr[tab_idx].sh_size % sizeof(Elf32_Sym)));
182 	elf->num_dynsyms = shdr[tab_idx].sh_size / sizeof(Elf32_Sym);
183 
184 	elf->dynstr = (void *)(shdr[str_idx].sh_addr + elf->load_addr);
185 	elf->dynstr_size = shdr[str_idx].sh_size;
186 }
187 
188 static void e64_save_symtab(struct ta_elf *elf, size_t tab_idx)
189 {
190 	Elf64_Shdr *shdr = elf->shdr;
191 	size_t str_idx = shdr[tab_idx].sh_link;
192 
193 	elf->dynsymtab = (void *)(vaddr_t)(shdr[tab_idx].sh_addr +
194 					   elf->load_addr);
195 	assert(!(shdr[tab_idx].sh_size % sizeof(Elf64_Sym)));
196 	elf->num_dynsyms = shdr[tab_idx].sh_size / sizeof(Elf64_Sym);
197 
198 	elf->dynstr = (void *)(vaddr_t)(shdr[str_idx].sh_addr + elf->load_addr);
199 	elf->dynstr_size = shdr[str_idx].sh_size;
200 }
201 
202 static void save_symtab(struct ta_elf *elf)
203 {
204 	size_t n = 0;
205 
206 	if (elf->is_32bit) {
207 		Elf32_Shdr *shdr = elf->shdr;
208 
209 		for (n = 0; n < elf->e_shnum; n++) {
210 			if (shdr[n].sh_type == SHT_DYNSYM) {
211 				e32_save_symtab(elf, n);
212 				break;
213 			}
214 		}
215 	} else {
216 		Elf64_Shdr *shdr = elf->shdr;
217 
218 		for (n = 0; n < elf->e_shnum; n++) {
219 			if (shdr[n].sh_type == SHT_DYNSYM) {
220 				e64_save_symtab(elf, n);
221 				break;
222 			}
223 		}
224 
225 	}
226 
227 	save_hashtab(elf);
228 }
229 
230 static void init_elf(struct ta_elf *elf)
231 {
232 	TEE_Result res = TEE_SUCCESS;
233 	vaddr_t va = 0;
234 	uint32_t flags = PTA_SYSTEM_MAP_FLAG_SHAREABLE;
235 
236 	res = sys_open_ta_bin(&elf->uuid, &elf->handle);
237 	if (res)
238 		err(res, "sys_open_ta_bin(%pUl)", (void *)&elf->uuid);
239 
240 	/*
241 	 * Map it read-only executable when we're loading a library where
242 	 * the ELF header is included in a load segment.
243 	 */
244 	if (!elf->is_main)
245 		flags |= PTA_SYSTEM_MAP_FLAG_EXECUTABLE;
246 	res = sys_map_ta_bin(&va, SMALL_PAGE_SIZE, flags, elf->handle, 0, 0, 0);
247 	if (res)
248 		err(res, "sys_map_ta_bin");
249 	elf->ehdr_addr = va;
250 	if (!elf->is_main) {
251 		elf->load_addr = va;
252 		elf->max_addr = va + SMALL_PAGE_SIZE;
253 		elf->max_offs = SMALL_PAGE_SIZE;
254 	}
255 
256 	if (!IS_ELF(*(Elf32_Ehdr *)va))
257 		err(TEE_ERROR_BAD_FORMAT, "TA is not an ELF");
258 
259 	res = e32_parse_ehdr(elf, (void *)va);
260 	if (res == TEE_ERROR_BAD_FORMAT)
261 		res = e64_parse_ehdr(elf, (void *)va);
262 	if (res)
263 		err(res, "Cannot parse ELF");
264 
265 	if (elf->e_phoff + elf->e_phnum * elf->e_phentsize > SMALL_PAGE_SIZE)
266 		err(TEE_ERROR_NOT_SUPPORTED, "Cannot read program headers");
267 
268 	elf->phdr = (void *)(va + elf->e_phoff);
269 }
270 
271 static size_t roundup(size_t v)
272 {
273 	return ROUNDUP(v, SMALL_PAGE_SIZE);
274 }
275 
276 static size_t rounddown(size_t v)
277 {
278 	return ROUNDDOWN(v, SMALL_PAGE_SIZE);
279 }
280 
281 static void add_segment(struct ta_elf *elf, size_t offset, size_t vaddr,
282 			size_t filesz, size_t memsz, size_t flags, size_t align)
283 {
284 	struct segment *seg = calloc(1, sizeof(*seg));
285 
286 	if (!seg)
287 		err(TEE_ERROR_OUT_OF_MEMORY, "calloc");
288 
289 	seg->offset = offset;
290 	seg->vaddr = vaddr;
291 	seg->filesz = filesz;
292 	seg->memsz = memsz;
293 	seg->flags = flags;
294 	seg->align = align;
295 
296 	TAILQ_INSERT_TAIL(&elf->segs, seg, link);
297 }
298 
299 static void parse_load_segments(struct ta_elf *elf)
300 {
301 	size_t n = 0;
302 
303 	if (elf->is_32bit) {
304 		Elf32_Phdr *phdr = elf->phdr;
305 
306 		for (n = 0; n < elf->e_phnum; n++)
307 			if (phdr[n].p_type == PT_LOAD) {
308 				add_segment(elf, phdr[n].p_offset,
309 					    phdr[n].p_vaddr, phdr[n].p_filesz,
310 					    phdr[n].p_memsz, phdr[n].p_flags,
311 					    phdr[n].p_align);
312 			} else if (phdr[n].p_type == PT_ARM_EXIDX) {
313 				elf->exidx_start = phdr[n].p_vaddr;
314 				elf->exidx_size = phdr[n].p_filesz;
315 			}
316 	} else {
317 		Elf64_Phdr *phdr = elf->phdr;
318 
319 		for (n = 0; n < elf->e_phnum; n++)
320 			if (phdr[n].p_type == PT_LOAD)
321 				add_segment(elf, phdr[n].p_offset,
322 					    phdr[n].p_vaddr, phdr[n].p_filesz,
323 					    phdr[n].p_memsz, phdr[n].p_flags,
324 					    phdr[n].p_align);
325 	}
326 }
327 
328 static void copy_remapped_to(struct ta_elf *elf, const struct segment *seg)
329 {
330 	uint8_t *dst = (void *)(seg->vaddr + elf->load_addr);
331 	size_t n = 0;
332 	size_t offs = seg->offset;
333 	size_t num_bytes = seg->filesz;
334 
335 	if (offs < elf->max_offs) {
336 		n = MIN(elf->max_offs - offs, num_bytes);
337 		memcpy(dst, (void *)(elf->max_addr + offs - elf->max_offs), n);
338 		dst += n;
339 		offs += n;
340 		num_bytes -= n;
341 	}
342 
343 	if (num_bytes) {
344 		TEE_Result res = sys_copy_from_ta_bin(dst, num_bytes,
345 						      elf->handle, offs);
346 
347 		if (res)
348 			err(res, "sys_copy_from_ta_bin");
349 		elf->max_offs += offs;
350 	}
351 }
352 
353 static void adjust_segments(struct ta_elf *elf)
354 {
355 	struct segment *seg = NULL;
356 	struct segment *prev_seg = NULL;
357 	size_t prev_end_addr = 0;
358 	size_t align = 0;
359 	size_t mask = 0;
360 
361 	/* Sanity check */
362 	TAILQ_FOREACH(seg, &elf->segs, link) {
363 		size_t dummy __maybe_unused = 0;
364 
365 		assert(seg->align >= SMALL_PAGE_SIZE);
366 		assert(!ADD_OVERFLOW(seg->vaddr, seg->memsz, &dummy));
367 		assert(seg->filesz <= seg->memsz);
368 		assert((seg->offset & SMALL_PAGE_MASK) ==
369 		       (seg->vaddr & SMALL_PAGE_MASK));
370 
371 		prev_seg = TAILQ_PREV(seg, segment_head, link);
372 		if (prev_seg) {
373 			assert(seg->vaddr >= prev_seg->vaddr + prev_seg->memsz);
374 			assert(seg->offset >=
375 			       prev_seg->offset + prev_seg->filesz);
376 		}
377 		if (!align)
378 			align = seg->align;
379 		assert(align == seg->align);
380 	}
381 
382 	mask = align - 1;
383 
384 	seg = TAILQ_FIRST(&elf->segs);
385 	if (seg)
386 		seg = TAILQ_NEXT(seg, link);
387 	while (seg) {
388 		prev_seg = TAILQ_PREV(seg, segment_head, link);
389 		prev_end_addr = prev_seg->vaddr + prev_seg->memsz;
390 
391 		/*
392 		 * This segment may overlap with the last "page" in the
393 		 * previous segment in two different ways:
394 		 * 1. Virtual address (and offset) overlaps =>
395 		 *    Permissions needs to be merged. The offset must have
396 		 *    the SMALL_PAGE_MASK bits set as vaddr and offset must
397 		 *    add up with prevsion segment.
398 		 *
399 		 * 2. Only offset overlaps =>
400 		 *    The same page in the ELF is mapped at two different
401 		 *    virtual addresses. As a limitation this segment must
402 		 *    be mapped as writeable.
403 		 */
404 
405 		/* Case 1. */
406 		if (rounddown(seg->vaddr) < prev_end_addr) {
407 			assert((seg->vaddr & mask) == (seg->offset & mask));
408 			assert(prev_seg->memsz == prev_seg->filesz);
409 
410 			/*
411 			 * Merge the segments and their permissions.
412 			 * Note that the may be a small hole between the
413 			 * two sections.
414 			 */
415 			prev_seg->filesz = seg->vaddr + seg->filesz -
416 					   prev_seg->vaddr;
417 			prev_seg->memsz = seg->vaddr + seg->memsz -
418 					   prev_seg->vaddr;
419 			prev_seg->flags |= seg->flags;
420 
421 			TAILQ_REMOVE(&elf->segs, seg, link);
422 			free(seg);
423 			seg = TAILQ_NEXT(prev_seg, link);
424 			continue;
425 		}
426 
427 		/* Case 2. */
428 		if ((seg->offset & mask) &&
429 		    rounddown(seg->offset) <
430 		    (prev_seg->offset + prev_seg->filesz)) {
431 
432 			assert(seg->flags & PF_W);
433 			seg->remapped_writeable = true;
434 		}
435 
436 		/*
437 		 * No overlap, but we may need to align address, offset and
438 		 * size.
439 		 */
440 		seg->filesz += seg->vaddr - rounddown(seg->vaddr);
441 		seg->memsz += seg->vaddr - rounddown(seg->vaddr);
442 		seg->vaddr = rounddown(seg->vaddr);
443 		seg->offset = rounddown(seg->offset);
444 		seg = TAILQ_NEXT(seg, link);
445 	}
446 
447 }
448 
449 static void populate_segments_legacy(struct ta_elf *elf)
450 {
451 	TEE_Result res = TEE_SUCCESS;
452 	struct segment *seg = NULL;
453 	vaddr_t va = 0;
454 
455 	TAILQ_FOREACH(seg, &elf->segs, link) {
456 		struct segment *last_seg = TAILQ_LAST(&elf->segs, segment_head);
457 		size_t pad_end = roundup(last_seg->vaddr + last_seg->memsz -
458 					 seg->vaddr - seg->memsz);
459 		size_t num_bytes = roundup(seg->memsz);
460 
461 		if (!elf->load_addr)
462 			va = 0;
463 		else
464 			va = seg->vaddr + elf->load_addr;
465 
466 
467 		if (!(seg->flags & PF_R))
468 			err(TEE_ERROR_NOT_SUPPORTED,
469 			    "Segment must be readable");
470 
471 		res = sys_map_zi(num_bytes, 0, &va, 0, pad_end);
472 		if (res)
473 			err(res, "sys_map_zi");
474 		res = sys_copy_from_ta_bin((void *)va, seg->filesz,
475 					   elf->handle, seg->offset);
476 		if (res)
477 			err(res, "sys_copy_from_ta_bin");
478 
479 		if (!elf->load_addr)
480 			elf->load_addr = va;
481 		elf->max_addr = va + num_bytes;
482 		elf->max_offs = seg->offset + seg->filesz;
483 	}
484 }
485 
486 static size_t get_pad_begin(void)
487 {
488 #ifdef CFG_TA_ASLR
489 	size_t min = CFG_TA_ASLR_MIN_OFFSET_PAGES;
490 	size_t max = CFG_TA_ASLR_MAX_OFFSET_PAGES;
491 	TEE_Result res = TEE_SUCCESS;
492 	uint32_t rnd32 = 0;
493 	size_t rnd = 0;
494 
495 	COMPILE_TIME_ASSERT(CFG_TA_ASLR_MIN_OFFSET_PAGES <
496 			    CFG_TA_ASLR_MAX_OFFSET_PAGES);
497 	if (max > min) {
498 		res = utee_cryp_random_number_generate(&rnd32, sizeof(rnd32));
499 		if (res) {
500 			DMSG("Random read failed: %#"PRIx32, res);
501 			return min * SMALL_PAGE_SIZE;
502 		}
503 		rnd = rnd32 % (max - min);
504 	}
505 
506 	return (min + rnd) * SMALL_PAGE_SIZE;
507 #else /*!CFG_TA_ASLR*/
508 	return 0;
509 #endif /*!CFG_TA_ASLR*/
510 }
511 
512 static void populate_segments(struct ta_elf *elf)
513 {
514 	TEE_Result res = TEE_SUCCESS;
515 	struct segment *seg = NULL;
516 	vaddr_t va = 0;
517 	size_t pad_begin = 0;
518 
519 	TAILQ_FOREACH(seg, &elf->segs, link) {
520 		struct segment *last_seg = TAILQ_LAST(&elf->segs, segment_head);
521 		size_t pad_end = roundup(last_seg->vaddr + last_seg->memsz -
522 					 seg->vaddr - seg->memsz);
523 
524 		if (seg->remapped_writeable) {
525 			size_t num_bytes = roundup(seg->vaddr + seg->memsz) -
526 					   rounddown(seg->vaddr);
527 
528 			assert(elf->load_addr);
529 			va = rounddown(elf->load_addr + seg->vaddr);
530 			assert(va >= elf->max_addr);
531 			res = sys_map_zi(num_bytes, 0, &va, 0, pad_end);
532 			if (res)
533 				err(res, "sys_map_zi");
534 
535 			copy_remapped_to(elf, seg);
536 			elf->max_addr = va + num_bytes;
537 		} else {
538 			uint32_t flags =  0;
539 			size_t filesz = seg->filesz;
540 			size_t memsz = seg->memsz;
541 			size_t offset = seg->offset;
542 			size_t vaddr = seg->vaddr;
543 
544 			if (offset < elf->max_offs) {
545 				/*
546 				 * We're in a load segment which overlaps
547 				 * with (or is covered by) the first page
548 				 * of a shared library.
549 				 */
550 				if (vaddr + filesz < SMALL_PAGE_SIZE) {
551 					size_t num_bytes = 0;
552 
553 					/*
554 					 * If this segment is completely
555 					 * covered, take next.
556 					 */
557 					if (vaddr + memsz <= SMALL_PAGE_SIZE)
558 						continue;
559 
560 					/*
561 					 * All data of the segment is
562 					 * loaded, but we need to zero
563 					 * extend it.
564 					 */
565 					va = elf->max_addr;
566 					num_bytes = roundup(vaddr + memsz) -
567 						    roundup(vaddr) -
568 						    SMALL_PAGE_SIZE;
569 					assert(num_bytes);
570 					res = sys_map_zi(num_bytes, 0, &va, 0,
571 							 0);
572 					if (res)
573 						err(res, "sys_map_zi");
574 					elf->max_addr = roundup(va + num_bytes);
575 					continue;
576 				}
577 
578 				/* Partial overlap, remove the first page. */
579 				vaddr += SMALL_PAGE_SIZE;
580 				filesz -= SMALL_PAGE_SIZE;
581 				memsz -= SMALL_PAGE_SIZE;
582 				offset += SMALL_PAGE_SIZE;
583 			}
584 
585 			if (!elf->load_addr) {
586 				va = 0;
587 				pad_begin = get_pad_begin();
588 				/*
589 				 * If mapping with pad_begin fails we'll
590 				 * retry without pad_begin, effectively
591 				 * disabling ASLR for the current ELF file.
592 				 */
593 			} else {
594 				va = vaddr + elf->load_addr;
595 				pad_begin = 0;
596 			}
597 
598 			if (seg->flags & PF_W)
599 				flags |= PTA_SYSTEM_MAP_FLAG_WRITEABLE;
600 			else
601 				flags |= PTA_SYSTEM_MAP_FLAG_SHAREABLE;
602 			if (seg->flags & PF_X)
603 				flags |= PTA_SYSTEM_MAP_FLAG_EXECUTABLE;
604 			if (!(seg->flags & PF_R))
605 				err(TEE_ERROR_NOT_SUPPORTED,
606 				    "Segment must be readable");
607 			if (flags & PTA_SYSTEM_MAP_FLAG_WRITEABLE) {
608 				res = sys_map_zi(memsz, 0, &va, pad_begin,
609 						 pad_end);
610 				if (pad_begin && res == TEE_ERROR_OUT_OF_MEMORY)
611 					res = sys_map_zi(memsz, 0, &va, 0,
612 							 pad_end);
613 				if (res)
614 					err(res, "sys_map_zi");
615 				res = sys_copy_from_ta_bin((void *)va, filesz,
616 							   elf->handle, offset);
617 				if (res)
618 					err(res, "sys_copy_from_ta_bin");
619 			} else {
620 				res = sys_map_ta_bin(&va, filesz, flags,
621 						     elf->handle, offset,
622 						     pad_begin, pad_end);
623 				if (pad_begin && res == TEE_ERROR_OUT_OF_MEMORY)
624 					res = sys_map_ta_bin(&va, filesz, flags,
625 							     elf->handle,
626 							     offset, 0,
627 							     pad_end);
628 				if (res)
629 					err(res, "sys_map_ta_bin");
630 			}
631 
632 			if (!elf->load_addr)
633 				elf->load_addr = va;
634 			elf->max_addr = roundup(va + filesz);
635 			elf->max_offs += filesz;
636 		}
637 	}
638 }
639 
640 static void map_segments(struct ta_elf *elf)
641 {
642 	TEE_Result res = TEE_SUCCESS;
643 
644 	parse_load_segments(elf);
645 	adjust_segments(elf);
646 	if (TAILQ_FIRST(&elf->segs)->offset < SMALL_PAGE_SIZE) {
647 		vaddr_t va = 0;
648 		size_t sz = elf->max_addr - elf->load_addr;
649 		struct segment *seg = TAILQ_LAST(&elf->segs, segment_head);
650 		size_t pad_begin = get_pad_begin();
651 
652 		/*
653 		 * We're loading a library, if not other parts of the code
654 		 * need to be updated too.
655 		 */
656 		assert(!elf->is_main);
657 
658 		/*
659 		 * Now that we know how much virtual memory is needed move
660 		 * the already mapped part to a location which can
661 		 * accommodate us.
662 		 */
663 		res = sys_remap(elf->load_addr, &va, sz, pad_begin,
664 				roundup(seg->vaddr + seg->memsz));
665 		if (res == TEE_ERROR_OUT_OF_MEMORY)
666 			res = sys_remap(elf->load_addr, &va, sz, 0,
667 					roundup(seg->vaddr + seg->memsz));
668 		if (res)
669 			err(res, "sys_remap");
670 		elf->ehdr_addr = va;
671 		elf->load_addr = va;
672 		elf->max_addr = va + sz;
673 		elf->phdr = (void *)(va + elf->e_phoff);
674 	}
675 	if (elf->is_legacy)
676 		populate_segments_legacy(elf);
677 	else
678 		populate_segments(elf);
679 }
680 
681 static int hex(char c)
682 {
683 	char lc = tolower(c);
684 
685 	if (isdigit(lc))
686 		return lc - '0';
687 	if (isxdigit(lc))
688 		return lc - 'a' + 10;
689 	return -1;
690 }
691 
692 static uint32_t parse_hex(const char *s, size_t nchars, uint32_t *res)
693 {
694 	uint32_t v = 0;
695 	size_t n;
696 	int c;
697 
698 	for (n = 0; n < nchars; n++) {
699 		c = hex(s[n]);
700 		if (c == (char)-1) {
701 			*res = TEE_ERROR_BAD_FORMAT;
702 			goto out;
703 		}
704 		v = (v << 4) + c;
705 	}
706 	*res = TEE_SUCCESS;
707 out:
708 	return v;
709 }
710 
711 /*
712  * Convert a UUID string @s into a TEE_UUID @uuid
713  * Expected format for @s is: xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx
714  * 'x' being any hexadecimal digit (0-9a-fA-F)
715  */
716 static TEE_Result parse_uuid(const char *s, TEE_UUID *uuid)
717 {
718 	TEE_Result res = TEE_SUCCESS;
719 	TEE_UUID u = { 0 };
720 	const char *p = s;
721 	size_t i;
722 
723 	if (strlen(p) != 36)
724 		return TEE_ERROR_BAD_FORMAT;
725 	if (p[8] != '-' || p[13] != '-' || p[18] != '-' || p[23] != '-')
726 		return TEE_ERROR_BAD_FORMAT;
727 
728 	u.timeLow = parse_hex(p, 8, &res);
729 	if (res)
730 		goto out;
731 	p += 9;
732 	u.timeMid = parse_hex(p, 4, &res);
733 	if (res)
734 		goto out;
735 	p += 5;
736 	u.timeHiAndVersion = parse_hex(p, 4, &res);
737 	if (res)
738 		goto out;
739 	p += 5;
740 	for (i = 0; i < 8; i++) {
741 		u.clockSeqAndNode[i] = parse_hex(p, 2, &res);
742 		if (res)
743 			goto out;
744 		if (i == 1)
745 			p += 3;
746 		else
747 			p += 2;
748 	}
749 	*uuid = u;
750 out:
751 	return res;
752 }
753 
754 static void add_deps_from_segment(struct ta_elf *elf, unsigned int type,
755 				  vaddr_t addr, size_t memsz)
756 {
757 	size_t dyn_entsize = 0;
758 	size_t num_dyns = 0;
759 	size_t n = 0;
760 	unsigned int tag = 0;
761 	size_t val = 0;
762 	TEE_UUID uuid = { };
763 	char *str_tab = NULL;
764 
765 	if (type != PT_DYNAMIC)
766 		return;
767 
768 	if (elf->is_32bit)
769 		dyn_entsize = sizeof(Elf32_Dyn);
770 	else
771 		dyn_entsize = sizeof(Elf64_Dyn);
772 
773 	assert(!(memsz % dyn_entsize));
774 	num_dyns = memsz / dyn_entsize;
775 
776 	for (n = 0; n < num_dyns; n++) {
777 		read_dyn(elf, addr, n, &tag, &val);
778 		if (tag == DT_STRTAB) {
779 			str_tab = (char *)(val + elf->load_addr);
780 			break;
781 		}
782 	}
783 
784 	for (n = 0; n < num_dyns; n++) {
785 		read_dyn(elf, addr, n, &tag, &val);
786 		if (tag != DT_NEEDED)
787 			continue;
788 		parse_uuid(str_tab + val, &uuid);
789 		queue_elf(&uuid);
790 	}
791 }
792 
793 static void add_dependencies(struct ta_elf *elf)
794 {
795 	size_t n = 0;
796 
797 	if (elf->is_32bit) {
798 		Elf32_Phdr *phdr = elf->phdr;
799 
800 		for (n = 0; n < elf->e_phnum; n++)
801 			add_deps_from_segment(elf, phdr[n].p_type,
802 					      phdr[n].p_vaddr, phdr[n].p_memsz);
803 	} else {
804 		Elf64_Phdr *phdr = elf->phdr;
805 
806 		for (n = 0; n < elf->e_phnum; n++)
807 			add_deps_from_segment(elf, phdr[n].p_type,
808 					      phdr[n].p_vaddr, phdr[n].p_memsz);
809 	}
810 }
811 
812 static void copy_section_headers(struct ta_elf *elf)
813 {
814 	TEE_Result res = TEE_SUCCESS;
815 	size_t sz = elf->e_shnum * elf->e_shentsize;
816 	size_t offs = 0;
817 
818 	elf->shdr = malloc(sz);
819 	if (!elf->shdr)
820 		err(TEE_ERROR_OUT_OF_MEMORY, "malloc");
821 
822 	/*
823 	 * We're assuming that section headers comes after the load segments,
824 	 * but if it's a very small dynamically linked library the section
825 	 * headers can still end up (partially?) in the first mapped page.
826 	 */
827 	if (elf->e_shoff < SMALL_PAGE_SIZE) {
828 		assert(!elf->is_main);
829 		offs = MIN(SMALL_PAGE_SIZE - elf->e_shoff, sz);
830 		memcpy(elf->shdr, (void *)(elf->load_addr + elf->e_shoff),
831 		       offs);
832 	}
833 
834 	if (offs < sz) {
835 		res = sys_copy_from_ta_bin((uint8_t *)elf->shdr + offs,
836 					   sz - offs, elf->handle,
837 					   elf->e_shoff + offs);
838 		if (res)
839 			err(res, "sys_copy_from_ta_bin");
840 	}
841 }
842 
843 static void close_handle(struct ta_elf *elf)
844 {
845 	TEE_Result res = sys_close_ta_bin(elf->handle);
846 
847 	if (res)
848 		err(res, "sys_close_ta_bin");
849 	elf->handle = -1;
850 }
851 
852 void ta_elf_load_main(const TEE_UUID *uuid, uint32_t *is_32bit,
853 		      uint64_t *entry, uint64_t *sp, uint32_t *ta_flags)
854 {
855 	struct ta_elf *elf = queue_elf(uuid);
856 	struct ta_head *head;
857 	vaddr_t va = 0;
858 	TEE_Result res = TEE_SUCCESS;
859 
860 	assert(elf);
861 	elf->is_main = true;
862 
863 	init_elf(elf);
864 
865 	/*
866 	 * Legacy TAs doesn't set entry point, instead it's set in ta_head.
867 	 * If entry point isn't set explicitly, set to the start of the
868 	 * first executable section by the linker. Since ta_head also
869 	 * always comes first in legacy TA it means that the entry point
870 	 * will be set to 0x20.
871 	 *
872 	 * NB, everything before the commit a73b5878c89d ("Replace
873 	 * ta_head.entry with elf entry") is considered legacy TAs for
874 	 * ldelf.
875 	 */
876 	if (elf->e_entry == sizeof(*head))
877 		elf->is_legacy = true;
878 
879 	map_segments(elf);
880 	add_dependencies(elf);
881 	copy_section_headers(elf);
882 	save_symtab(elf);
883 	close_handle(elf);
884 
885 	head = (struct ta_head *)elf->load_addr;
886 
887 	*is_32bit = elf->is_32bit;
888 	if (elf->is_legacy) {
889 		assert(head->depr_entry != UINT64_MAX);
890 		*entry = head->depr_entry + elf->load_addr;
891 	} else {
892 		assert(head->depr_entry == UINT64_MAX);
893 		*entry = elf->e_entry + elf->load_addr;
894 	}
895 
896 	res = sys_map_zi(head->stack_size, 0, &va, 0, 0);
897 	if (res)
898 		err(res, "sys_map_zi stack");
899 
900 	if (head->flags & ~TA_FLAGS_MASK)
901 		err(TEE_ERROR_BAD_FORMAT, "Invalid TA flags(s) %#"PRIx32,
902 		    head->flags & ~TA_FLAGS_MASK);
903 
904 	*ta_flags = head->flags;
905 	*sp = va + head->stack_size;
906 	ta_stack = va;
907 	ta_stack_size = head->stack_size;
908 }
909 
910 void ta_elf_load_dependency(struct ta_elf *elf, bool is_32bit)
911 {
912 	if (elf->is_main)
913 		return;
914 
915 	init_elf(elf);
916 	if (elf->is_32bit != is_32bit)
917 		err(TEE_ERROR_BAD_FORMAT, "ELF %pUl is %sbit (expected %sbit)",
918 		    (void *)&elf->uuid, elf->is_32bit ? "32" : "64",
919 		    is_32bit ? "32" : "64");
920 
921 	map_segments(elf);
922 	add_dependencies(elf);
923 	copy_section_headers(elf);
924 	save_symtab(elf);
925 	close_handle(elf);
926 }
927 
928 void ta_elf_finalize_mappings(struct ta_elf *elf)
929 {
930 	TEE_Result res = TEE_SUCCESS;
931 	struct segment *seg = NULL;
932 
933 	if (!elf->is_legacy)
934 		return;
935 
936 	TAILQ_FOREACH(seg, &elf->segs, link) {
937 		vaddr_t va = elf->load_addr + seg->vaddr;
938 		uint32_t flags =  0;
939 
940 		if (seg->flags & PF_W)
941 			flags |= PTA_SYSTEM_MAP_FLAG_WRITEABLE;
942 		if (seg->flags & PF_X)
943 			flags |= PTA_SYSTEM_MAP_FLAG_EXECUTABLE;
944 
945 		res = sys_set_prot(va, seg->memsz, flags);
946 		if (res)
947 			err(res, "sys_set_prot");
948 	}
949 }
950 
951 static void __printf(3, 4) print_wrapper(void *pctx, print_func_t print_func,
952 					 const char *fmt, ...)
953 {
954 	va_list ap;
955 
956 	va_start(ap, fmt);
957 	print_func(pctx, fmt, ap);
958 	va_end(ap);
959 }
960 
961 static void print_seg(void *pctx, print_func_t print_func,
962 		      size_t idx __maybe_unused, int elf_idx __maybe_unused,
963 		      vaddr_t va __maybe_unused, paddr_t pa __maybe_unused,
964 		      size_t sz __maybe_unused, uint32_t flags)
965 {
966 	int width __maybe_unused = 8;
967 	char desc[14] __maybe_unused = "";
968 	char flags_str[] __maybe_unused = "----";
969 
970 	if (elf_idx > -1) {
971 		snprintf(desc, sizeof(desc), " [%d]", elf_idx);
972 	} else {
973 		if (flags & DUMP_MAP_EPHEM)
974 			snprintf(desc, sizeof(desc), " (param)");
975 		if (flags & DUMP_MAP_LDELF)
976 			snprintf(desc, sizeof(desc), " (ldelf)");
977 		if (va == ta_stack)
978 			snprintf(desc, sizeof(desc), " (stack)");
979 	}
980 
981 	if (flags & DUMP_MAP_READ)
982 		flags_str[0] = 'r';
983 	if (flags & DUMP_MAP_WRITE)
984 		flags_str[1] = 'w';
985 	if (flags & DUMP_MAP_EXEC)
986 		flags_str[2] = 'x';
987 	if (flags & DUMP_MAP_SECURE)
988 		flags_str[3] = 's';
989 
990 	print_wrapper(pctx, print_func,
991 		      "region %2zu: va 0x%0*"PRIxVA" pa 0x%0*"PRIxPA" size 0x%06zx flags %s%s\n",
992 		      idx, width, va, width, pa, sz, flags_str, desc);
993 }
994 
995 static bool get_next_in_order(struct ta_elf_queue *elf_queue,
996 			      struct ta_elf **elf, struct segment **seg,
997 			      size_t *elf_idx)
998 {
999 	struct ta_elf *e = NULL;
1000 	struct segment *s = NULL;
1001 	size_t idx = 0;
1002 	vaddr_t va = 0;
1003 	struct ta_elf *e2 = NULL;
1004 	size_t i2 = 0;
1005 
1006 	assert(elf && seg && elf_idx);
1007 	e = *elf;
1008 	s = *seg;
1009 	assert((e == NULL && s == NULL) || (e != NULL && s != NULL));
1010 
1011 	if (s) {
1012 		s = TAILQ_NEXT(s, link);
1013 		if (s) {
1014 			*seg = s;
1015 			return true;
1016 		}
1017 	}
1018 
1019 	if (e)
1020 		va = e->load_addr;
1021 
1022 	/* Find the ELF with next load address */
1023 	e = NULL;
1024 	TAILQ_FOREACH(e2, elf_queue, link) {
1025 		if (e2->load_addr > va) {
1026 			if (!e || e2->load_addr < e->load_addr) {
1027 				e = e2;
1028 				idx = i2;
1029 			}
1030 		}
1031 		i2++;
1032 	}
1033 	if (!e)
1034 		return false;
1035 
1036 	*elf = e;
1037 	*seg = TAILQ_FIRST(&e->segs);
1038 	*elf_idx = idx;
1039 	return true;
1040 }
1041 
1042 void ta_elf_print_mappings(void *pctx, print_func_t print_func,
1043 			   struct ta_elf_queue *elf_queue, size_t num_maps,
1044 			   struct dump_map *maps, vaddr_t mpool_base)
1045 {
1046 	struct segment *seg = NULL;
1047 	struct ta_elf *elf = NULL;
1048 	size_t elf_idx = 0;
1049 	size_t idx = 0;
1050 	size_t map_idx = 0;
1051 
1052 	/*
1053 	 * Loop over all segments and maps, printing virtual address in
1054 	 * order. Segment has priority if the virtual address is present
1055 	 * in both map and segment.
1056 	 */
1057 	get_next_in_order(elf_queue, &elf, &seg, &elf_idx);
1058 	while (true) {
1059 		vaddr_t va = -1;
1060 		size_t sz = 0;
1061 		uint32_t flags = DUMP_MAP_SECURE;
1062 		size_t offs = 0;
1063 
1064 		if (seg) {
1065 			va = rounddown(seg->vaddr + elf->load_addr);
1066 			sz = roundup(seg->vaddr + seg->memsz) -
1067 				     rounddown(seg->vaddr);
1068 		}
1069 
1070 		while (map_idx < num_maps && maps[map_idx].va <= va) {
1071 			uint32_t f = 0;
1072 
1073 			/* If there's a match, it should be the same map */
1074 			if (maps[map_idx].va == va) {
1075 				/*
1076 				 * In shared libraries the first page is
1077 				 * mapped separately with the rest of that
1078 				 * segment following back to back in a
1079 				 * separate entry.
1080 				 */
1081 				if (map_idx + 1 < num_maps &&
1082 				    maps[map_idx].sz == SMALL_PAGE_SIZE) {
1083 					vaddr_t next_va = maps[map_idx].va +
1084 							  maps[map_idx].sz;
1085 					size_t comb_sz = maps[map_idx].sz +
1086 							 maps[map_idx + 1].sz;
1087 
1088 					if (next_va == maps[map_idx + 1].va &&
1089 					    comb_sz == sz &&
1090 					    maps[map_idx].flags ==
1091 					    maps[map_idx + 1].flags) {
1092 						/* Skip this and next entry */
1093 						map_idx += 2;
1094 						continue;
1095 					}
1096 				}
1097 				assert(maps[map_idx].sz == sz);
1098 			} else if (maps[map_idx].va < va) {
1099 				if (maps[map_idx].va == mpool_base)
1100 					f |= DUMP_MAP_LDELF;
1101 				print_seg(pctx, print_func, idx, -1,
1102 					  maps[map_idx].va, maps[map_idx].pa,
1103 					  maps[map_idx].sz,
1104 					  maps[map_idx].flags | f);
1105 				idx++;
1106 			}
1107 			map_idx++;
1108 		}
1109 
1110 		if (!seg)
1111 			break;
1112 
1113 		offs = rounddown(seg->offset);
1114 		if (seg->flags & PF_R)
1115 			flags |= DUMP_MAP_READ;
1116 		if (seg->flags & PF_W)
1117 			flags |= DUMP_MAP_WRITE;
1118 		if (seg->flags & PF_X)
1119 			flags |= DUMP_MAP_EXEC;
1120 
1121 		print_seg(pctx, print_func, idx, elf_idx, va, offs, sz, flags);
1122 		idx++;
1123 
1124 		if (!get_next_in_order(elf_queue, &elf, &seg, &elf_idx))
1125 			seg = NULL;
1126 	}
1127 
1128 	elf_idx = 0;
1129 	TAILQ_FOREACH(elf, elf_queue, link) {
1130 		print_wrapper(pctx, print_func,
1131 			      " [%zu] %pUl @ 0x%0*"PRIxVA"\n",
1132 			      elf_idx, (void *)&elf->uuid, 8, elf->load_addr);
1133 		elf_idx++;
1134 	}
1135 }
1136 
1137 #ifdef CFG_UNWIND
1138 void ta_elf_stack_trace_a32(uint32_t regs[16])
1139 {
1140 	struct unwind_state_arm32 state = { };
1141 
1142 	memcpy(state.registers, regs, sizeof(state.registers));
1143 	print_stack_arm32(&state, ta_stack, ta_stack_size);
1144 }
1145 
1146 void ta_elf_stack_trace_a64(uint64_t fp, uint64_t sp, uint64_t pc)
1147 {
1148 	struct unwind_state_arm64 state = { .fp = fp, .sp = sp, .pc = pc };
1149 
1150 	print_stack_arm64(&state, ta_stack, ta_stack_size);
1151 }
1152 #endif
1153