1 | // SPDX-FileCopyrightText: 2023 UnionTech Software Technology Co., Ltd. |
2 | // |
3 | // SPDX-License-Identifier: GPL-3.0-or-later |
4 | |
5 | static const char kLinuxGateLibraryName[] = "linux-gate.so" ; |
6 | static const char kMappedFileUnsafePrefix[] = "/dev/" ; |
7 | static const char kDeletedSuffix[] = " (deleted)" ; |
8 | |
9 | // Traits classes so consumers can write templatized code to deal |
10 | // with specific ELF bits. |
11 | struct ElfClass32 { |
12 | typedef Elf32_Addr Addr; |
13 | typedef Elf32_Ehdr Ehdr; |
14 | typedef Elf32_Nhdr Nhdr; |
15 | typedef Elf32_Phdr Phdr; |
16 | typedef Elf32_Shdr Shdr; |
17 | typedef Elf32_Half Half; |
18 | typedef Elf32_Off Off; |
19 | typedef Elf32_Sym Sym; |
20 | typedef Elf32_Word Word; |
21 | typedef struct { |
22 | int32_t d_tag; |
23 | uint32_t d_val; |
24 | } ElfDyn; |
25 | |
26 | static const int kClass = ELFCLASS32; |
27 | static const uint16_t kMachine = EM_386; |
28 | static const size_t kAddrSize = sizeof(Elf32_Addr); |
29 | static constexpr const char* kMachineName = "x86" ; |
30 | }; |
31 | |
32 | struct ElfClass64 { |
33 | typedef Elf64_Addr Addr; |
34 | typedef Elf64_Ehdr Ehdr; |
35 | typedef Elf64_Nhdr Nhdr; |
36 | typedef Elf64_Phdr Phdr; |
37 | typedef Elf64_Shdr Shdr; |
38 | typedef Elf64_Half Half; |
39 | typedef Elf64_Off Off; |
40 | typedef Elf64_Sym Sym; |
41 | typedef Elf64_Word Word; |
42 | typedef struct { |
43 | uint64_t d_tag; |
44 | uint64_t d_val; |
45 | } ElfDyn; |
46 | |
47 | static const int kClass = ELFCLASS64; |
48 | static const uint16_t kMachine = EM_X86_64; |
49 | static const size_t kAddrSize = sizeof(Elf64_Addr); |
50 | static constexpr const char* kMachineName = "x86_64" ; |
51 | }; |
52 | |
53 | struct ElfSegment { |
54 | const void* start; |
55 | size_t size; |
56 | }; |
57 | |
58 | #define ELFMAG "\177ELF" |
59 | #define SELFMAG 4 |
60 | bool IsValidElf(const void* elf_base) |
61 | { |
62 | return strncmp(reinterpret_cast<const char*>(elf_base), |
63 | ELFMAG, SELFMAG) == 0; |
64 | } |
65 | |
66 | int ElfClass(const void* elf_base) |
67 | { |
68 | const ElfW(Ehdr)* = |
69 | reinterpret_cast<const ElfW(Ehdr)*>(elf_base); |
70 | |
71 | return elf_header->e_ident[EI_CLASS]; |
72 | } |
73 | |
74 | template<typename ElfClass, typename T> |
75 | const T* GetOffset(const typename ElfClass::Ehdr* , |
76 | typename ElfClass::Off offset) { |
77 | return reinterpret_cast<const T*>(reinterpret_cast<uintptr_t>(elf_header) + |
78 | offset); |
79 | } |
80 | |
81 | template<typename ElfClass> |
82 | void FindElfClassSegment(const char *elf_base, |
83 | typename ElfClass::Word segment_type, |
84 | vector<ElfSegment> *segments) |
85 | { |
86 | typedef typename ElfClass::Ehdr Ehdr; |
87 | typedef typename ElfClass::Phdr Phdr; |
88 | |
89 | assert(elf_base); |
90 | assert(segments); |
91 | |
92 | assert(strncmp(elf_base, ELFMAG, SELFMAG) == 0); |
93 | |
94 | const Ehdr* = reinterpret_cast<const Ehdr*>(elf_base); |
95 | assert(elf_header->e_ident[EI_CLASS] == ElfClass::kClass); |
96 | |
97 | const Phdr* phdrs = |
98 | GetOffset<ElfClass, Phdr>(elf_header, elf_header->e_phoff); |
99 | |
100 | for (int i = 0; i < elf_header->e_phnum; ++i) { |
101 | if (phdrs[i].p_type == segment_type) { |
102 | ElfSegment seg = {0,0}; |
103 | seg.start = elf_base + phdrs[i].p_offset; |
104 | seg.size = phdrs[i].p_filesz; |
105 | segments->push_back(seg); |
106 | } |
107 | } |
108 | } |
109 | |
110 | template<typename ElfClass> |
111 | const typename ElfClass::Shdr* FindElfSectionByName( |
112 | const char* name, |
113 | typename ElfClass::Word section_type, |
114 | const typename ElfClass::Shdr* sections, |
115 | const char* section_names, |
116 | const char* names_end, |
117 | int nsection) |
118 | { |
119 | assert(name != NULL); |
120 | assert(sections != NULL); |
121 | assert(nsection > 0); |
122 | |
123 | int name_len = strlen(name); |
124 | if (name_len == 0) |
125 | return NULL; |
126 | |
127 | for (int i = 0; i < nsection; ++i) { |
128 | const char* section_name = section_names + sections[i].sh_name; |
129 | if (sections[i].sh_type == section_type && |
130 | names_end - section_name >= name_len + 1 && |
131 | strcmp(name, section_name) == 0) { |
132 | return sections + i; |
133 | } |
134 | } |
135 | return NULL; |
136 | } |
137 | |
138 | template<typename ElfClass> |
139 | void FindElfClassSection(const char *elf_base, |
140 | const char *section_name, |
141 | typename ElfClass::Word section_type, |
142 | const void **section_start, |
143 | size_t *section_size) { |
144 | typedef typename ElfClass::Ehdr Ehdr; |
145 | typedef typename ElfClass::Shdr Shdr; |
146 | |
147 | assert(elf_base); |
148 | assert(section_start); |
149 | assert(section_size); |
150 | |
151 | assert(strncmp(elf_base, ELFMAG, SELFMAG) == 0); |
152 | |
153 | const Ehdr* = reinterpret_cast<const Ehdr*>(elf_base); |
154 | assert(elf_header->e_ident[EI_CLASS] == ElfClass::kClass); |
155 | |
156 | const Shdr* sections = |
157 | GetOffset<ElfClass, Shdr>(elf_header, elf_header->e_shoff); |
158 | const Shdr* section_names = sections + elf_header->e_shstrndx; |
159 | const char* names = |
160 | GetOffset<ElfClass, char>(elf_header, section_names->sh_offset); |
161 | const char *names_end = names + section_names->sh_size; |
162 | |
163 | const Shdr* section = |
164 | FindElfSectionByName<ElfClass>(section_name, section_type, |
165 | sections, names, names_end, |
166 | elf_header->e_shnum); |
167 | |
168 | if (section != NULL && section->sh_size > 0) { |
169 | *section_start = elf_base + section->sh_offset; |
170 | *section_size = section->sh_size; |
171 | } |
172 | } |
173 | |
174 | template<typename ElfClass> |
175 | const char* FindLibClassStartWith(const void *elf_base, |
176 | size_t size, const char* prefix) |
177 | { |
178 | typedef typename ElfClass::Shdr Shdr; |
179 | typedef typename ElfClass::ElfDyn ElfDyn; |
180 | |
181 | assert(elf_base); |
182 | |
183 | Shdr* dynamic = nullptr; |
184 | size_t dynamic_size = 0; |
185 | FindElfClassSection<ElfClass>((const char*)elf_base, ".dynamic" , |
186 | SHT_PROGBITS, (const void**)&dynamic, &dynamic_size); |
187 | if (nullptr == dynamic) { |
188 | return nullptr; |
189 | } |
190 | |
191 | Shdr* dynstr = nullptr; |
192 | size_t dynstr_size = 0; |
193 | FindElfClassSection<ElfClass>((const char*)elf_base, ".dynstr" , |
194 | SHT_PROGBITS, (const void**)&dynstr, &dynstr_size); |
195 | if (nullptr == dynstr) { |
196 | return nullptr; |
197 | } |
198 | |
199 | if (dynamic->sh_entsize != sizeof(ElfDyn)) { |
200 | LOG(DEBUG) << "Invalid ELF file: incorrect .dynamic size " |
201 | << dynamic->sh_entsize; |
202 | return nullptr; |
203 | } |
204 | if (!dynamic->sh_size) { |
205 | return nullptr; |
206 | } |
207 | if (dynamic->sh_size % dynamic->sh_entsize) { |
208 | LOG(DEBUG) << "Invalid ELF file: incorrect .dynamic section size " |
209 | << dynamic->sh_size; |
210 | return nullptr; |
211 | } |
212 | if (dynstr->sh_size == 0) { |
213 | LOG(DEBUG) << "Invalid ELF file: empty string table" ; |
214 | return nullptr; |
215 | } |
216 | |
217 | if (dynamic->sh_offset + dynamic->sh_size >= size) { |
218 | LOG(DEBUG) << "Invalid ELF file: can't read .dynamic" ; |
219 | return nullptr; |
220 | } |
221 | |
222 | ElfDyn* dyn_list = (ElfDyn*)((char *)elf_base + dynamic->sh_offset); |
223 | const char* strtable = (const char*)((char *)elf_base + dynstr->sh_offset); |
224 | |
225 | for (size_t i = 0; i < dynamic->sh_size / dynamic->sh_entsize; ++i) { |
226 | if (dyn_list[i].d_tag == DT_NEEDED && |
227 | dyn_list[i].d_val < dynstr->sh_size) { |
228 | const char* name = strtable + dyn_list[i].d_val; |
229 | if (!strcmp(name, prefix)) { |
230 | return name; |
231 | } |
232 | } |
233 | } |
234 | |
235 | return nullptr; |
236 | } |
237 | |
238 | bool FindElfSegments(const void* elf_mapped_base, |
239 | uint32_t segment_type, |
240 | vector<ElfSegment>* segments) |
241 | { |
242 | assert(elf_mapped_base); |
243 | assert(segments); |
244 | |
245 | if (!IsValidElf(elf_mapped_base)) |
246 | return false; |
247 | |
248 | int cls = ElfClass(elf_mapped_base); |
249 | const char* elf_base = |
250 | static_cast<const char*>(elf_mapped_base); |
251 | |
252 | if (cls == ELFCLASS32) { |
253 | FindElfClassSegment<ElfClass32>(elf_base, segment_type, segments); |
254 | return true; |
255 | } else if (cls == ELFCLASS64) { |
256 | FindElfClassSegment<ElfClass64>(elf_base, segment_type, segments); |
257 | return true; |
258 | } |
259 | |
260 | return false; |
261 | } |
262 | |
263 | // ELF note name and desc are 32-bits word padded. |
264 | #define NOTE_PADDING(a) ((a + 3) & ~3) |
265 | |
266 | // These functions are also used inside the crashed process, so be safe |
267 | // and use the syscall/libc wrappers instead of direct syscalls or libc. |
268 | |
269 | static bool ElfClassBuildIDNoteIdentifier(const void *section, size_t length, |
270 | uint8_t* identifier) { |
271 | static_assert(sizeof(ElfClass32::Nhdr) == sizeof(ElfClass64::Nhdr), |
272 | "Elf32_Nhdr and Elf64_Nhdr should be the same" ); |
273 | typedef typename ElfClass32::Nhdr Nhdr; |
274 | |
275 | const void* section_end = reinterpret_cast<const char*>(section) + length; |
276 | const Nhdr* = reinterpret_cast<const Nhdr*>(section); |
277 | while (reinterpret_cast<const void *>(note_header) < section_end) { |
278 | if (note_header->n_type == NT_GNU_BUILD_ID) |
279 | break; |
280 | note_header = reinterpret_cast<const Nhdr*>( |
281 | reinterpret_cast<const char*>(note_header) + sizeof(Nhdr) + |
282 | NOTE_PADDING(note_header->n_namesz) + |
283 | NOTE_PADDING(note_header->n_descsz)); |
284 | } |
285 | if (reinterpret_cast<const void *>(note_header) >= section_end || |
286 | note_header->n_descsz == 0) { |
287 | return false; |
288 | } |
289 | |
290 | const uint8_t* build_id = reinterpret_cast<const uint8_t*>(note_header) + |
291 | sizeof(Nhdr) + NOTE_PADDING(note_header->n_namesz); |
292 | memcpy(identifier, build_id, note_header->n_descsz); |
293 | |
294 | return true; |
295 | } |
296 | |
297 | bool FindElfSection(const void *elf_mapped_base, |
298 | const char *section_name, |
299 | uint32_t section_type, |
300 | const void **section_start, |
301 | size_t *section_size) |
302 | { |
303 | assert(elf_mapped_base); |
304 | assert(section_start); |
305 | assert(section_size); |
306 | |
307 | *section_start = NULL; |
308 | *section_size = 0; |
309 | |
310 | if (!IsValidElf(elf_mapped_base)) |
311 | return false; |
312 | |
313 | int cls = ElfClass(elf_mapped_base); |
314 | const char* elf_base = |
315 | static_cast<const char*>(elf_mapped_base); |
316 | |
317 | if (cls == ELFCLASS32) { |
318 | FindElfClassSection<ElfClass32>(elf_base, section_name, section_type, |
319 | section_start, section_size); |
320 | return *section_start != NULL; |
321 | } else if (cls == ELFCLASS64) { |
322 | FindElfClassSection<ElfClass64>(elf_base, section_name, section_type, |
323 | section_start, section_size); |
324 | return *section_start != NULL; |
325 | } |
326 | |
327 | return false; |
328 | } |
329 | |
330 | /////////////////////////////////////////////////////////////////////////////// |
331 | // |
332 | // |
333 | // A utility class for mapping a file into memory for read-only access of |
334 | // the file content. Its implementation avoids calling into libc functions |
335 | // by directly making system calls for open, close, mmap, and munmap. |
336 | class MemoryMappedFile { |
337 | public: |
338 | MemoryMappedFile(); |
339 | |
340 | // Constructor that calls Map() to map a file at |path| into memory. |
341 | // If Map() fails, the object behaves as if it is default constructed. |
342 | MemoryMappedFile(const char* path, size_t offset); |
343 | |
344 | ~MemoryMappedFile(); |
345 | |
346 | // Maps a file at |path| into memory, which can then be accessed via |
347 | // content() as a MemoryRange object or via data(), and returns true on |
348 | // success. Mapping an empty file will succeed but with data() and size() |
349 | // returning nullptr and 0, respectively. An existing mapping is unmapped |
350 | // before a new mapping is created. |
351 | bool Map(const char* path, size_t offset); |
352 | |
353 | // Unmaps the memory for the mapped file. It's a no-op if no file is |
354 | // mapped. |
355 | void Unmap(); |
356 | |
357 | // Returns a MemoryRange object that covers the memory for the mapped |
358 | // file. The MemoryRange object is empty if no file is mapped. |
359 | const MemoryRange& content() const { return content_; } |
360 | |
361 | // Returns a pointer to the beginning of the memory for the mapped file. |
362 | // or nullptr if no file is mapped or the mapped file is empty. |
363 | const void* data() const { return content_.data(); } |
364 | |
365 | // Returns the size in bytes of the mapped file, or zero if no file |
366 | // is mapped. |
367 | size_t size() const { return content_.length(); } |
368 | |
369 | private: |
370 | // Mapped file content as a MemoryRange object. |
371 | MemoryRange content_; |
372 | }; |
373 | |
374 | MemoryMappedFile::MemoryMappedFile() {} |
375 | |
376 | MemoryMappedFile::MemoryMappedFile(const char* path, size_t offset) { |
377 | Map(path, offset); |
378 | } |
379 | |
380 | MemoryMappedFile::~MemoryMappedFile() { |
381 | Unmap(); |
382 | } |
383 | |
384 | bool MemoryMappedFile::Map(const char* path, size_t offset) { |
385 | Unmap(); |
386 | |
387 | int fd = open(path, O_RDONLY, 0); |
388 | if (fd == -1) { |
389 | return false; |
390 | } |
391 | |
392 | #if defined(__x86_64__) || defined(__aarch64__) || defined(__sw_64) || \ |
393 | (defined(__mips__) && _MIPS_SIM == _ABI64) |
394 | |
395 | struct stat st; |
396 | if (fstat(fd, &st) == -1 || st.st_size < 0) { |
397 | #else |
398 | struct kernel_stat64 st; |
399 | if (fstat64(fd, &st) == -1 || st.st_size < 0) { |
400 | #endif |
401 | close(fd); |
402 | return false; |
403 | } |
404 | |
405 | // Strangely file size can be negative, but we check above that it is not. |
406 | size_t file_len = static_cast<size_t>(st.st_size); |
407 | if (0 == file_len) { |
408 | void* data = mmap(nullptr, 16*1024, |
409 | PROT_READ|PROT_WRITE, |
410 | MAP_ANONYMOUS|MAP_PRIVATE, 0, 0); |
411 | if (data == MAP_FAILED) { |
412 | close(fd); |
413 | return false; |
414 | } |
415 | |
416 | file_len = read(fd, data, 16*1024); |
417 | close(fd); |
418 | content_.Set(data, file_len); |
419 | return true; |
420 | } |
421 | |
422 | // If the file does not extend beyond the offset, simply use an empty |
423 | // MemoryRange and return true. Don't bother to call mmap() |
424 | // even though mmap() can handle an empty file on some platforms. |
425 | if (offset >= file_len) { |
426 | close(fd); |
427 | return true; |
428 | } |
429 | |
430 | void* data = mmap(nullptr, file_len, PROT_READ, MAP_PRIVATE, fd, offset); |
431 | close(fd); |
432 | if (data == MAP_FAILED) { |
433 | return false; |
434 | } |
435 | |
436 | content_.Set(data, file_len - offset); |
437 | return true; |
438 | } |
439 | |
440 | void MemoryMappedFile::Unmap() { |
441 | if (content_.data()) { |
442 | munmap(const_cast<uint8_t*>(content_.data()), content_.length()); |
443 | content_.Set(nullptr, 0); |
444 | } |
445 | } |
446 | |
447 | // A class for reading a file, line by line, without using fopen/fgets or other |
448 | // functions which may allocate memory. |
449 | class LineReader { |
450 | public: |
451 | LineReader(int fd) |
452 | : fd_(fd), |
453 | hit_eof_(false), |
454 | buf_used_(0) { |
455 | } |
456 | |
457 | // The maximum length of a line. |
458 | static const size_t kMaxLineLen = 1024*4; |
459 | |
460 | // Return the next line from the file. |
461 | // line: (output) a pointer to the start of the line. The line is NUL |
462 | // terminated. |
463 | // len: (output) the length of the line (not inc the NUL byte) |
464 | // |
465 | // Returns true iff successful (false on EOF). |
466 | // |
467 | // One must call |PopLine| after this function, otherwise you'll continue to |
468 | // get the same line over and over. |
469 | bool GetNextLine(const char **line, unsigned int *len) { |
470 | for (;;) { |
471 | if (buf_used_ == 0 && hit_eof_) |
472 | return false; |
473 | |
474 | for (unsigned int i = 0; i < buf_used_; ++i) { |
475 | if (buf_[i] == '\n') { |
476 | buf_[i] = 0; |
477 | *len = i; |
478 | *line = buf_; |
479 | return true; |
480 | } |
481 | } |
482 | |
483 | if (buf_used_ == sizeof(buf_)) { |
484 | // we scanned the whole buffer and didn't find an end-of-line marker. |
485 | // This line is too long to process. |
486 | return false; |
487 | } |
488 | |
489 | // We didn't find any end-of-line terminators in the buffer. However, if |
490 | // this is the last line in the file it might not have one: |
491 | if (hit_eof_) { |
492 | assert(buf_used_); |
493 | // There's room for the NUL because of the buf_used_ == sizeof(buf_) |
494 | // check above. |
495 | buf_[buf_used_] = 0; |
496 | *len = buf_used_; |
497 | buf_used_ += 1; // since we appended the NUL. |
498 | *line = buf_; |
499 | return true; |
500 | } |
501 | |
502 | // Otherwise, we should pull in more data from the file |
503 | const ssize_t n = read(fd_, buf_ + buf_used_, |
504 | sizeof(buf_) - buf_used_); |
505 | if (n < 0) { |
506 | return false; |
507 | } else if (n == 0) { |
508 | hit_eof_ = true; |
509 | } else { |
510 | buf_used_ += n; |
511 | } |
512 | |
513 | // At this point, we have either set the hit_eof_ flag, or we have more |
514 | // data to process... |
515 | } |
516 | } |
517 | |
518 | void PopLine(unsigned int len) { |
519 | // len doesn't include the NUL byte at the end. |
520 | |
521 | assert(buf_used_ >= len + 1); |
522 | buf_used_ -= len + 1; |
523 | memmove(buf_, buf_ + len + 1, buf_used_); |
524 | } |
525 | |
526 | private: |
527 | const int fd_; |
528 | |
529 | bool hit_eof_; |
530 | unsigned int buf_used_; |
531 | char buf_[kMaxLineLen]; |
532 | }; |
533 | |