1 //===- ScriptParser.cpp ---------------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file contains a recursive-descendent parser for linker scripts.
10 // Parsed results are stored to Config and Script global objects.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "ScriptParser.h"
15 #include "Config.h"
16 #include "Driver.h"
17 #include "InputFiles.h"
18 #include "LinkerScript.h"
19 #include "OutputSections.h"
20 #include "ScriptLexer.h"
21 #include "SymbolTable.h"
22 #include "Symbols.h"
23 #include "Target.h"
24 #include "lld/Common/CommonLinkerContext.h"
25 #include "llvm/ADT/SmallString.h"
26 #include "llvm/ADT/StringRef.h"
27 #include "llvm/ADT/StringSet.h"
28 #include "llvm/ADT/StringSwitch.h"
29 #include "llvm/BinaryFormat/ELF.h"
30 #include "llvm/Support/Casting.h"
31 #include "llvm/Support/ErrorHandling.h"
32 #include "llvm/Support/FileSystem.h"
33 #include "llvm/Support/MathExtras.h"
34 #include "llvm/Support/Path.h"
35 #include "llvm/Support/SaveAndRestore.h"
36 #include "llvm/Support/TimeProfiler.h"
37 #include <cassert>
38 #include <limits>
39 #include <optional>
40 #include <vector>
41
42 using namespace llvm;
43 using namespace llvm::ELF;
44 using namespace llvm::support::endian;
45 using namespace lld;
46 using namespace lld::elf;
47
48 namespace {
49 class ScriptParser final : ScriptLexer {
50 public:
ScriptParser(MemoryBufferRef mb)51 ScriptParser(MemoryBufferRef mb) : ScriptLexer(mb) {
52 // Initialize IsUnderSysroot
53 if (config->sysroot == "")
54 return;
55 StringRef path = mb.getBufferIdentifier();
56 for (; !path.empty(); path = sys::path::parent_path(path)) {
57 if (!sys::fs::equivalent(config->sysroot, path))
58 continue;
59 isUnderSysroot = true;
60 return;
61 }
62 }
63
64 void readLinkerScript();
65 void readVersionScript();
66 void readDynamicList();
67 void readDefsym(StringRef name);
68
69 private:
70 void addFile(StringRef path);
71
72 void readAsNeeded();
73 void readEntry();
74 void readExtern();
75 void readGroup();
76 void readInclude();
77 void readInput();
78 void readMemory();
79 void readOutput();
80 void readOutputArch();
81 void readOutputFormat();
82 void readOverwriteSections();
83 void readPhdrs();
84 void readRegionAlias();
85 void readSearchDir();
86 void readSections();
87 void readTarget();
88 void readVersion();
89 void readVersionScriptCommand();
90 void readNoCrossRefs(bool to);
91
92 SymbolAssignment *readSymbolAssignment(StringRef name);
93 ByteCommand *readByteCommand(StringRef tok);
94 std::array<uint8_t, 4> readFill();
95 bool readSectionDirective(OutputSection *cmd, StringRef tok);
96 void readSectionAddressType(OutputSection *cmd);
97 OutputDesc *readOverlaySectionDescription();
98 OutputDesc *readOutputSectionDescription(StringRef outSec);
99 SmallVector<SectionCommand *, 0> readOverlay();
100 SmallVector<StringRef, 0> readOutputSectionPhdrs();
101 std::pair<uint64_t, uint64_t> readInputSectionFlags();
102 InputSectionDescription *readInputSectionDescription(StringRef tok);
103 StringMatcher readFilePatterns();
104 SmallVector<SectionPattern, 0> readInputSectionsList();
105 InputSectionDescription *readInputSectionRules(StringRef filePattern,
106 uint64_t withFlags,
107 uint64_t withoutFlags);
108 unsigned readPhdrType();
109 SortSectionPolicy peekSortKind();
110 SortSectionPolicy readSortKind();
111 SymbolAssignment *readProvideHidden(bool provide, bool hidden);
112 SymbolAssignment *readAssignment(StringRef tok);
113 void readSort();
114 Expr readAssert();
115 Expr readConstant();
116 Expr getPageSize();
117
118 Expr readMemoryAssignment(StringRef, StringRef, StringRef);
119 void readMemoryAttributes(uint32_t &flags, uint32_t &invFlags,
120 uint32_t &negFlags, uint32_t &negInvFlags);
121
122 Expr combine(StringRef op, Expr l, Expr r);
123 Expr readExpr();
124 Expr readExpr1(Expr lhs, int minPrec);
125 StringRef readParenLiteral();
126 Expr readPrimary();
127 Expr readTernary(Expr cond);
128 Expr readParenExpr();
129
130 // For parsing version script.
131 SmallVector<SymbolVersion, 0> readVersionExtern();
132 void readAnonymousDeclaration();
133 void readVersionDeclaration(StringRef verStr);
134
135 std::pair<SmallVector<SymbolVersion, 0>, SmallVector<SymbolVersion, 0>>
136 readSymbols();
137
138 // True if a script being read is in the --sysroot directory.
139 bool isUnderSysroot = false;
140
141 // A set to detect an INCLUDE() cycle.
142 StringSet<> seen;
143
144 // If we are currently parsing a PROVIDE|PROVIDE_HIDDEN command,
145 // then this member is set to the PROVIDE symbol name.
146 std::optional<llvm::StringRef> activeProvideSym;
147 };
148 } // namespace
149
unquote(StringRef s)150 static StringRef unquote(StringRef s) {
151 if (s.starts_with("\""))
152 return s.substr(1, s.size() - 2);
153 return s;
154 }
155
156 // Some operations only support one non absolute value. Move the
157 // absolute one to the right hand side for convenience.
moveAbsRight(ExprValue & a,ExprValue & b)158 static void moveAbsRight(ExprValue &a, ExprValue &b) {
159 if (a.sec == nullptr || (a.forceAbsolute && !b.isAbsolute()))
160 std::swap(a, b);
161 if (!b.isAbsolute())
162 script->recordError(
163 a.loc + ": at least one side of the expression must be absolute");
164 }
165
add(ExprValue a,ExprValue b)166 static ExprValue add(ExprValue a, ExprValue b) {
167 moveAbsRight(a, b);
168 return {a.sec, a.forceAbsolute, a.getSectionOffset() + b.getValue(), a.loc};
169 }
170
sub(ExprValue a,ExprValue b)171 static ExprValue sub(ExprValue a, ExprValue b) {
172 // The distance between two symbols in sections is absolute.
173 if (!a.isAbsolute() && !b.isAbsolute())
174 return a.getValue() - b.getValue();
175 return {a.sec, false, a.getSectionOffset() - b.getValue(), a.loc};
176 }
177
bitAnd(ExprValue a,ExprValue b)178 static ExprValue bitAnd(ExprValue a, ExprValue b) {
179 moveAbsRight(a, b);
180 return {a.sec, a.forceAbsolute,
181 (a.getValue() & b.getValue()) - a.getSecAddr(), a.loc};
182 }
183
bitXor(ExprValue a,ExprValue b)184 static ExprValue bitXor(ExprValue a, ExprValue b) {
185 moveAbsRight(a, b);
186 return {a.sec, a.forceAbsolute,
187 (a.getValue() ^ b.getValue()) - a.getSecAddr(), a.loc};
188 }
189
bitOr(ExprValue a,ExprValue b)190 static ExprValue bitOr(ExprValue a, ExprValue b) {
191 moveAbsRight(a, b);
192 return {a.sec, a.forceAbsolute,
193 (a.getValue() | b.getValue()) - a.getSecAddr(), a.loc};
194 }
195
readDynamicList()196 void ScriptParser::readDynamicList() {
197 expect("{");
198 SmallVector<SymbolVersion, 0> locals;
199 SmallVector<SymbolVersion, 0> globals;
200 std::tie(locals, globals) = readSymbols();
201 expect(";");
202
203 if (!atEOF()) {
204 setError("EOF expected, but got " + next());
205 return;
206 }
207 if (!locals.empty()) {
208 setError("\"local:\" scope not supported in --dynamic-list");
209 return;
210 }
211
212 for (SymbolVersion v : globals)
213 config->dynamicList.push_back(v);
214 }
215
readVersionScript()216 void ScriptParser::readVersionScript() {
217 readVersionScriptCommand();
218 if (!atEOF())
219 setError("EOF expected, but got " + next());
220 }
221
readVersionScriptCommand()222 void ScriptParser::readVersionScriptCommand() {
223 if (consume("{")) {
224 readAnonymousDeclaration();
225 return;
226 }
227
228 while (!atEOF() && !errorCount() && peek() != "}") {
229 StringRef verStr = next();
230 if (verStr == "{") {
231 setError("anonymous version definition is used in "
232 "combination with other version definitions");
233 return;
234 }
235 expect("{");
236 readVersionDeclaration(verStr);
237 }
238 }
239
readVersion()240 void ScriptParser::readVersion() {
241 expect("{");
242 readVersionScriptCommand();
243 expect("}");
244 }
245
readLinkerScript()246 void ScriptParser::readLinkerScript() {
247 while (!atEOF()) {
248 StringRef tok = next();
249 if (tok == ";")
250 continue;
251
252 if (tok == "ENTRY") {
253 readEntry();
254 } else if (tok == "EXTERN") {
255 readExtern();
256 } else if (tok == "GROUP") {
257 readGroup();
258 } else if (tok == "INCLUDE") {
259 readInclude();
260 } else if (tok == "INPUT") {
261 readInput();
262 } else if (tok == "MEMORY") {
263 readMemory();
264 } else if (tok == "OUTPUT") {
265 readOutput();
266 } else if (tok == "OUTPUT_ARCH") {
267 readOutputArch();
268 } else if (tok == "OUTPUT_FORMAT") {
269 readOutputFormat();
270 } else if (tok == "OVERWRITE_SECTIONS") {
271 readOverwriteSections();
272 } else if (tok == "PHDRS") {
273 readPhdrs();
274 } else if (tok == "REGION_ALIAS") {
275 readRegionAlias();
276 } else if (tok == "SEARCH_DIR") {
277 readSearchDir();
278 } else if (tok == "SECTIONS") {
279 readSections();
280 } else if (tok == "TARGET") {
281 readTarget();
282 } else if (tok == "VERSION") {
283 readVersion();
284 } else if (tok == "NOCROSSREFS") {
285 readNoCrossRefs(/*to=*/false);
286 } else if (tok == "NOCROSSREFS_TO") {
287 readNoCrossRefs(/*to=*/true);
288 } else if (SymbolAssignment *cmd = readAssignment(tok)) {
289 script->sectionCommands.push_back(cmd);
290 } else {
291 setError("unknown directive: " + tok);
292 }
293 }
294 }
295
readDefsym(StringRef name)296 void ScriptParser::readDefsym(StringRef name) {
297 if (errorCount())
298 return;
299 Expr e = readExpr();
300 if (!atEOF())
301 setError("EOF expected, but got " + next());
302 auto *cmd = make<SymbolAssignment>(
303 name, e, 0, getCurrentMB().getBufferIdentifier().str());
304 script->sectionCommands.push_back(cmd);
305 }
306
readNoCrossRefs(bool to)307 void ScriptParser::readNoCrossRefs(bool to) {
308 expect("(");
309 NoCrossRefCommand cmd{{}, to};
310 while (!errorCount() && !consume(")"))
311 cmd.outputSections.push_back(unquote(next()));
312 if (cmd.outputSections.size() < 2)
313 warn(getCurrentLocation() + ": ignored with fewer than 2 output sections");
314 else
315 script->noCrossRefs.push_back(std::move(cmd));
316 }
317
addFile(StringRef s)318 void ScriptParser::addFile(StringRef s) {
319 if (isUnderSysroot && s.starts_with("/")) {
320 SmallString<128> pathData;
321 StringRef path = (config->sysroot + s).toStringRef(pathData);
322 if (sys::fs::exists(path))
323 ctx.driver.addFile(saver().save(path), /*withLOption=*/false);
324 else
325 setError("cannot find " + s + " inside " + config->sysroot);
326 return;
327 }
328
329 if (s.starts_with("/")) {
330 // Case 1: s is an absolute path. Just open it.
331 ctx.driver.addFile(s, /*withLOption=*/false);
332 } else if (s.starts_with("=")) {
333 // Case 2: relative to the sysroot.
334 if (config->sysroot.empty())
335 ctx.driver.addFile(s.substr(1), /*withLOption=*/false);
336 else
337 ctx.driver.addFile(saver().save(config->sysroot + "/" + s.substr(1)),
338 /*withLOption=*/false);
339 } else if (s.starts_with("-l")) {
340 // Case 3: search in the list of library paths.
341 ctx.driver.addLibrary(s.substr(2));
342 } else {
343 // Case 4: s is a relative path. Search in the directory of the script file.
344 std::string filename = std::string(getCurrentMB().getBufferIdentifier());
345 StringRef directory = sys::path::parent_path(filename);
346 if (!directory.empty()) {
347 SmallString<0> path(directory);
348 sys::path::append(path, s);
349 if (sys::fs::exists(path)) {
350 ctx.driver.addFile(path, /*withLOption=*/false);
351 return;
352 }
353 }
354 // Then search in the current working directory.
355 if (sys::fs::exists(s)) {
356 ctx.driver.addFile(s, /*withLOption=*/false);
357 } else {
358 // Finally, search in the list of library paths.
359 if (std::optional<std::string> path = findFromSearchPaths(s))
360 ctx.driver.addFile(saver().save(*path), /*withLOption=*/true);
361 else
362 setError("unable to find " + s);
363 }
364 }
365 }
366
readAsNeeded()367 void ScriptParser::readAsNeeded() {
368 expect("(");
369 bool orig = config->asNeeded;
370 config->asNeeded = true;
371 while (!errorCount() && !consume(")"))
372 addFile(unquote(next()));
373 config->asNeeded = orig;
374 }
375
readEntry()376 void ScriptParser::readEntry() {
377 // -e <symbol> takes predecence over ENTRY(<symbol>).
378 expect("(");
379 StringRef tok = next();
380 if (config->entry.empty())
381 config->entry = unquote(tok);
382 expect(")");
383 }
384
readExtern()385 void ScriptParser::readExtern() {
386 expect("(");
387 while (!errorCount() && !consume(")"))
388 config->undefined.push_back(unquote(next()));
389 }
390
readGroup()391 void ScriptParser::readGroup() {
392 bool orig = InputFile::isInGroup;
393 InputFile::isInGroup = true;
394 readInput();
395 InputFile::isInGroup = orig;
396 if (!orig)
397 ++InputFile::nextGroupId;
398 }
399
readInclude()400 void ScriptParser::readInclude() {
401 StringRef tok = unquote(next());
402
403 if (!seen.insert(tok).second) {
404 setError("there is a cycle in linker script INCLUDEs");
405 return;
406 }
407
408 if (std::optional<std::string> path = searchScript(tok)) {
409 if (std::optional<MemoryBufferRef> mb = readFile(*path))
410 tokenize(*mb);
411 return;
412 }
413 setError("cannot find linker script " + tok);
414 }
415
readInput()416 void ScriptParser::readInput() {
417 expect("(");
418 while (!errorCount() && !consume(")")) {
419 if (consume("AS_NEEDED"))
420 readAsNeeded();
421 else
422 addFile(unquote(next()));
423 }
424 }
425
readOutput()426 void ScriptParser::readOutput() {
427 // -o <file> takes predecence over OUTPUT(<file>).
428 expect("(");
429 StringRef tok = next();
430 if (config->outputFile.empty())
431 config->outputFile = unquote(tok);
432 expect(")");
433 }
434
readOutputArch()435 void ScriptParser::readOutputArch() {
436 // OUTPUT_ARCH is ignored for now.
437 expect("(");
438 while (!errorCount() && !consume(")"))
439 skip();
440 }
441
parseBfdName(StringRef s)442 static std::pair<ELFKind, uint16_t> parseBfdName(StringRef s) {
443 return StringSwitch<std::pair<ELFKind, uint16_t>>(s)
444 .Case("elf32-i386", {ELF32LEKind, EM_386})
445 .Case("elf32-avr", {ELF32LEKind, EM_AVR})
446 .Case("elf32-iamcu", {ELF32LEKind, EM_IAMCU})
447 .Case("elf32-littlearm", {ELF32LEKind, EM_ARM})
448 .Case("elf32-bigarm", {ELF32BEKind, EM_ARM})
449 .Case("elf32-x86-64", {ELF32LEKind, EM_X86_64})
450 .Case("elf64-aarch64", {ELF64LEKind, EM_AARCH64})
451 .Case("elf64-littleaarch64", {ELF64LEKind, EM_AARCH64})
452 .Case("elf64-bigaarch64", {ELF64BEKind, EM_AARCH64})
453 .Case("elf32-powerpc", {ELF32BEKind, EM_PPC})
454 .Case("elf32-powerpcle", {ELF32LEKind, EM_PPC})
455 .Case("elf64-powerpc", {ELF64BEKind, EM_PPC64})
456 .Case("elf64-powerpcle", {ELF64LEKind, EM_PPC64})
457 .Case("elf64-x86-64", {ELF64LEKind, EM_X86_64})
458 .Cases("elf32-tradbigmips", "elf32-bigmips", {ELF32BEKind, EM_MIPS})
459 .Case("elf32-ntradbigmips", {ELF32BEKind, EM_MIPS})
460 .Case("elf32-tradlittlemips", {ELF32LEKind, EM_MIPS})
461 .Case("elf32-ntradlittlemips", {ELF32LEKind, EM_MIPS})
462 .Case("elf64-tradbigmips", {ELF64BEKind, EM_MIPS})
463 .Case("elf64-tradlittlemips", {ELF64LEKind, EM_MIPS})
464 .Case("elf32-littleriscv", {ELF32LEKind, EM_RISCV})
465 .Case("elf64-littleriscv", {ELF64LEKind, EM_RISCV})
466 .Case("elf64-sparc", {ELF64BEKind, EM_SPARCV9})
467 .Case("elf32-msp430", {ELF32LEKind, EM_MSP430})
468 .Case("elf32-loongarch", {ELF32LEKind, EM_LOONGARCH})
469 .Case("elf64-loongarch", {ELF64LEKind, EM_LOONGARCH})
470 .Case("elf64-s390", {ELF64BEKind, EM_S390})
471 .Cases("elf32-hexagon", "elf32-littlehexagon", {ELF32LEKind, EM_HEXAGON})
472 .Default({ELFNoneKind, EM_NONE});
473 }
474
475 // Parse OUTPUT_FORMAT(bfdname) or OUTPUT_FORMAT(default, big, little). Choose
476 // big if -EB is specified, little if -EL is specified, or default if neither is
477 // specified.
readOutputFormat()478 void ScriptParser::readOutputFormat() {
479 expect("(");
480
481 StringRef s = unquote(next());
482 if (!consume(")")) {
483 expect(",");
484 StringRef tmp = unquote(next());
485 if (config->optEB)
486 s = tmp;
487 expect(",");
488 tmp = unquote(next());
489 if (config->optEL)
490 s = tmp;
491 consume(")");
492 }
493 // If more than one OUTPUT_FORMAT is specified, only the first is checked.
494 if (!config->bfdname.empty())
495 return;
496 config->bfdname = s;
497
498 if (s == "binary") {
499 config->oFormatBinary = true;
500 return;
501 }
502
503 if (s.consume_back("-freebsd"))
504 config->osabi = ELFOSABI_FREEBSD;
505
506 std::tie(config->ekind, config->emachine) = parseBfdName(s);
507 if (config->emachine == EM_NONE)
508 setError("unknown output format name: " + config->bfdname);
509 if (s == "elf32-ntradlittlemips" || s == "elf32-ntradbigmips")
510 config->mipsN32Abi = true;
511 if (config->emachine == EM_MSP430)
512 config->osabi = ELFOSABI_STANDALONE;
513 }
514
readPhdrs()515 void ScriptParser::readPhdrs() {
516 expect("{");
517
518 while (!errorCount() && !consume("}")) {
519 PhdrsCommand cmd;
520 cmd.name = next();
521 cmd.type = readPhdrType();
522
523 while (!errorCount() && !consume(";")) {
524 if (consume("FILEHDR"))
525 cmd.hasFilehdr = true;
526 else if (consume("PHDRS"))
527 cmd.hasPhdrs = true;
528 else if (consume("AT"))
529 cmd.lmaExpr = readParenExpr();
530 else if (consume("FLAGS"))
531 cmd.flags = readParenExpr()().getValue();
532 else
533 setError("unexpected header attribute: " + next());
534 }
535
536 script->phdrsCommands.push_back(cmd);
537 }
538 }
539
readRegionAlias()540 void ScriptParser::readRegionAlias() {
541 expect("(");
542 StringRef alias = unquote(next());
543 expect(",");
544 StringRef name = next();
545 expect(")");
546
547 if (script->memoryRegions.count(alias))
548 setError("redefinition of memory region '" + alias + "'");
549 if (!script->memoryRegions.count(name))
550 setError("memory region '" + name + "' is not defined");
551 script->memoryRegions.insert({alias, script->memoryRegions[name]});
552 }
553
readSearchDir()554 void ScriptParser::readSearchDir() {
555 expect("(");
556 StringRef tok = next();
557 if (!config->nostdlib)
558 config->searchPaths.push_back(unquote(tok));
559 expect(")");
560 }
561
562 // This reads an overlay description. Overlays are used to describe output
563 // sections that use the same virtual memory range and normally would trigger
564 // linker's sections sanity check failures.
565 // https://sourceware.org/binutils/docs/ld/Overlay-Description.html#Overlay-Description
readOverlay()566 SmallVector<SectionCommand *, 0> ScriptParser::readOverlay() {
567 Expr addrExpr;
568 if (consume(":")) {
569 addrExpr = [] { return script->getDot(); };
570 } else {
571 addrExpr = readExpr();
572 expect(":");
573 }
574 // When AT is omitted, LMA should equal VMA. script->getDot() when evaluating
575 // lmaExpr will ensure this, even if the start address is specified.
576 Expr lmaExpr =
577 consume("AT") ? readParenExpr() : [] { return script->getDot(); };
578 expect("{");
579
580 SmallVector<SectionCommand *, 0> v;
581 OutputSection *prev = nullptr;
582 while (!errorCount() && !consume("}")) {
583 // VA is the same for all sections. The LMAs are consecutive in memory
584 // starting from the base load address specified.
585 OutputDesc *osd = readOverlaySectionDescription();
586 osd->osec.addrExpr = addrExpr;
587 if (prev) {
588 osd->osec.lmaExpr = [=] { return prev->getLMA() + prev->size; };
589 } else {
590 osd->osec.lmaExpr = lmaExpr;
591 // Use first section address for subsequent sections as initial addrExpr
592 // can be DOT. Ensure the first section, even if empty, is not discarded.
593 osd->osec.usedInExpression = true;
594 addrExpr = [=]() -> ExprValue { return {&osd->osec, false, 0, ""}; };
595 }
596 v.push_back(osd);
597 prev = &osd->osec;
598 }
599
600 // According to the specification, at the end of the overlay, the location
601 // counter should be equal to the overlay base address plus size of the
602 // largest section seen in the overlay.
603 // Here we want to create the Dot assignment command to achieve that.
604 Expr moveDot = [=] {
605 uint64_t max = 0;
606 for (SectionCommand *cmd : v)
607 max = std::max(max, cast<OutputDesc>(cmd)->osec.size);
608 return addrExpr().getValue() + max;
609 };
610 v.push_back(make<SymbolAssignment>(".", moveDot, 0, getCurrentLocation()));
611 return v;
612 }
613
readOverwriteSections()614 void ScriptParser::readOverwriteSections() {
615 expect("{");
616 while (!errorCount() && !consume("}"))
617 script->overwriteSections.push_back(readOutputSectionDescription(next()));
618 }
619
readSections()620 void ScriptParser::readSections() {
621 expect("{");
622 SmallVector<SectionCommand *, 0> v;
623 while (!errorCount() && !consume("}")) {
624 StringRef tok = next();
625 if (tok == "OVERLAY") {
626 for (SectionCommand *cmd : readOverlay())
627 v.push_back(cmd);
628 continue;
629 } else if (tok == "INCLUDE") {
630 readInclude();
631 continue;
632 }
633
634 if (SectionCommand *cmd = readAssignment(tok))
635 v.push_back(cmd);
636 else
637 v.push_back(readOutputSectionDescription(tok));
638 }
639
640 // If DATA_SEGMENT_RELRO_END is absent, for sections after DATA_SEGMENT_ALIGN,
641 // the relro fields should be cleared.
642 if (!script->seenRelroEnd)
643 for (SectionCommand *cmd : v)
644 if (auto *osd = dyn_cast<OutputDesc>(cmd))
645 osd->osec.relro = false;
646
647 script->sectionCommands.insert(script->sectionCommands.end(), v.begin(),
648 v.end());
649
650 if (atEOF() || !consume("INSERT")) {
651 script->hasSectionsCommand = true;
652 return;
653 }
654
655 bool isAfter = false;
656 if (consume("AFTER"))
657 isAfter = true;
658 else if (!consume("BEFORE"))
659 setError("expected AFTER/BEFORE, but got '" + next() + "'");
660 StringRef where = next();
661 SmallVector<StringRef, 0> names;
662 for (SectionCommand *cmd : v)
663 if (auto *os = dyn_cast<OutputDesc>(cmd))
664 names.push_back(os->osec.name);
665 if (!names.empty())
666 script->insertCommands.push_back({std::move(names), isAfter, where});
667 }
668
readTarget()669 void ScriptParser::readTarget() {
670 // TARGET(foo) is an alias for "--format foo". Unlike GNU linkers,
671 // we accept only a limited set of BFD names (i.e. "elf" or "binary")
672 // for --format. We recognize only /^elf/ and "binary" in the linker
673 // script as well.
674 expect("(");
675 StringRef tok = unquote(next());
676 expect(")");
677
678 if (tok.starts_with("elf"))
679 config->formatBinary = false;
680 else if (tok == "binary")
681 config->formatBinary = true;
682 else
683 setError("unknown target: " + tok);
684 }
685
precedence(StringRef op)686 static int precedence(StringRef op) {
687 return StringSwitch<int>(op)
688 .Cases("*", "/", "%", 11)
689 .Cases("+", "-", 10)
690 .Cases("<<", ">>", 9)
691 .Cases("<", "<=", ">", ">=", 8)
692 .Cases("==", "!=", 7)
693 .Case("&", 6)
694 .Case("^", 5)
695 .Case("|", 4)
696 .Case("&&", 3)
697 .Case("||", 2)
698 .Case("?", 1)
699 .Default(-1);
700 }
701
readFilePatterns()702 StringMatcher ScriptParser::readFilePatterns() {
703 StringMatcher Matcher;
704
705 while (!errorCount() && !consume(")"))
706 Matcher.addPattern(SingleStringMatcher(next()));
707 return Matcher;
708 }
709
peekSortKind()710 SortSectionPolicy ScriptParser::peekSortKind() {
711 return StringSwitch<SortSectionPolicy>(peek())
712 .Case("REVERSE", SortSectionPolicy::Reverse)
713 .Cases("SORT", "SORT_BY_NAME", SortSectionPolicy::Name)
714 .Case("SORT_BY_ALIGNMENT", SortSectionPolicy::Alignment)
715 .Case("SORT_BY_INIT_PRIORITY", SortSectionPolicy::Priority)
716 .Case("SORT_NONE", SortSectionPolicy::None)
717 .Default(SortSectionPolicy::Default);
718 }
719
readSortKind()720 SortSectionPolicy ScriptParser::readSortKind() {
721 SortSectionPolicy ret = peekSortKind();
722 if (ret != SortSectionPolicy::Default)
723 skip();
724 return ret;
725 }
726
727 // Reads SECTIONS command contents in the following form:
728 //
729 // <contents> ::= <elem>*
730 // <elem> ::= <exclude>? <glob-pattern>
731 // <exclude> ::= "EXCLUDE_FILE" "(" <glob-pattern>+ ")"
732 //
733 // For example,
734 //
735 // *(.foo EXCLUDE_FILE (a.o) .bar EXCLUDE_FILE (b.o) .baz)
736 //
737 // is parsed as ".foo", ".bar" with "a.o", and ".baz" with "b.o".
738 // The semantics of that is section .foo in any file, section .bar in
739 // any file but a.o, and section .baz in any file but b.o.
readInputSectionsList()740 SmallVector<SectionPattern, 0> ScriptParser::readInputSectionsList() {
741 SmallVector<SectionPattern, 0> ret;
742 while (!errorCount() && peek() != ")") {
743 StringMatcher excludeFilePat;
744 if (consume("EXCLUDE_FILE")) {
745 expect("(");
746 excludeFilePat = readFilePatterns();
747 }
748
749 StringMatcher SectionMatcher;
750 // Break if the next token is ), EXCLUDE_FILE, or SORT*.
751 while (!errorCount() && peekSortKind() == SortSectionPolicy::Default) {
752 StringRef s = peek();
753 if (s == ")" || s == "EXCLUDE_FILE")
754 break;
755 // Detect common mistakes when certain non-wildcard meta characters are
756 // used without a closing ')'.
757 if (!s.empty() && strchr("(){}", s[0])) {
758 skip();
759 setError("section pattern is expected");
760 break;
761 }
762 SectionMatcher.addPattern(unquote(next()));
763 }
764
765 if (!SectionMatcher.empty())
766 ret.push_back({std::move(excludeFilePat), std::move(SectionMatcher)});
767 else if (excludeFilePat.empty())
768 break;
769 else
770 setError("section pattern is expected");
771 }
772 return ret;
773 }
774
775 // Reads contents of "SECTIONS" directive. That directive contains a
776 // list of glob patterns for input sections. The grammar is as follows.
777 //
778 // <patterns> ::= <section-list>
779 // | <sort> "(" <section-list> ")"
780 // | <sort> "(" <sort> "(" <section-list> ")" ")"
781 //
782 // <sort> ::= "SORT" | "SORT_BY_NAME" | "SORT_BY_ALIGNMENT"
783 // | "SORT_BY_INIT_PRIORITY" | "SORT_NONE"
784 //
785 // <section-list> is parsed by readInputSectionsList().
786 InputSectionDescription *
readInputSectionRules(StringRef filePattern,uint64_t withFlags,uint64_t withoutFlags)787 ScriptParser::readInputSectionRules(StringRef filePattern, uint64_t withFlags,
788 uint64_t withoutFlags) {
789 auto *cmd =
790 make<InputSectionDescription>(filePattern, withFlags, withoutFlags);
791 expect("(");
792
793 while (!errorCount() && !consume(")")) {
794 SortSectionPolicy outer = readSortKind();
795 SortSectionPolicy inner = SortSectionPolicy::Default;
796 SmallVector<SectionPattern, 0> v;
797 if (outer != SortSectionPolicy::Default) {
798 expect("(");
799 inner = readSortKind();
800 if (inner != SortSectionPolicy::Default) {
801 expect("(");
802 v = readInputSectionsList();
803 expect(")");
804 } else {
805 v = readInputSectionsList();
806 }
807 expect(")");
808 } else {
809 v = readInputSectionsList();
810 }
811
812 for (SectionPattern &pat : v) {
813 pat.sortInner = inner;
814 pat.sortOuter = outer;
815 }
816
817 std::move(v.begin(), v.end(), std::back_inserter(cmd->sectionPatterns));
818 }
819 return cmd;
820 }
821
822 InputSectionDescription *
readInputSectionDescription(StringRef tok)823 ScriptParser::readInputSectionDescription(StringRef tok) {
824 // Input section wildcard can be surrounded by KEEP.
825 // https://sourceware.org/binutils/docs/ld/Input-Section-Keep.html#Input-Section-Keep
826 uint64_t withFlags = 0;
827 uint64_t withoutFlags = 0;
828 if (tok == "KEEP") {
829 expect("(");
830 if (consume("INPUT_SECTION_FLAGS"))
831 std::tie(withFlags, withoutFlags) = readInputSectionFlags();
832 InputSectionDescription *cmd =
833 readInputSectionRules(next(), withFlags, withoutFlags);
834 expect(")");
835 script->keptSections.push_back(cmd);
836 return cmd;
837 }
838 if (tok == "INPUT_SECTION_FLAGS") {
839 std::tie(withFlags, withoutFlags) = readInputSectionFlags();
840 tok = next();
841 }
842 return readInputSectionRules(tok, withFlags, withoutFlags);
843 }
844
readSort()845 void ScriptParser::readSort() {
846 expect("(");
847 expect("CONSTRUCTORS");
848 expect(")");
849 }
850
readAssert()851 Expr ScriptParser::readAssert() {
852 expect("(");
853 Expr e = readExpr();
854 expect(",");
855 StringRef msg = unquote(next());
856 expect(")");
857
858 return [=] {
859 if (!e().getValue())
860 errorOrWarn(msg);
861 return script->getDot();
862 };
863 }
864
865 #define ECase(X) \
866 { #X, X }
867 constexpr std::pair<const char *, unsigned> typeMap[] = {
868 ECase(SHT_PROGBITS), ECase(SHT_NOTE), ECase(SHT_NOBITS),
869 ECase(SHT_INIT_ARRAY), ECase(SHT_FINI_ARRAY), ECase(SHT_PREINIT_ARRAY),
870 };
871 #undef ECase
872
873 // Tries to read the special directive for an output section definition which
874 // can be one of following: "(NOLOAD)", "(COPY)", "(INFO)", "(OVERLAY)", and
875 // "(TYPE=<value>)".
readSectionDirective(OutputSection * cmd,StringRef tok)876 bool ScriptParser::readSectionDirective(OutputSection *cmd, StringRef tok) {
877 if (tok != "NOLOAD" && tok != "COPY" && tok != "INFO" && tok != "OVERLAY" &&
878 tok != "TYPE")
879 return false;
880
881 if (consume("NOLOAD")) {
882 cmd->type = SHT_NOBITS;
883 cmd->typeIsSet = true;
884 } else if (consume("TYPE")) {
885 expect("=");
886 StringRef value = peek();
887 auto it = llvm::find_if(typeMap, [=](auto e) { return e.first == value; });
888 if (it != std::end(typeMap)) {
889 // The value is a recognized literal SHT_*.
890 cmd->type = it->second;
891 skip();
892 } else if (value.starts_with("SHT_")) {
893 setError("unknown section type " + value);
894 } else {
895 // Otherwise, read an expression.
896 cmd->type = readExpr()().getValue();
897 }
898 cmd->typeIsSet = true;
899 } else {
900 skip(); // This is "COPY", "INFO" or "OVERLAY".
901 cmd->nonAlloc = true;
902 }
903 expect(")");
904 return true;
905 }
906
907 // Reads an expression and/or the special directive for an output
908 // section definition. Directive is one of following: "(NOLOAD)",
909 // "(COPY)", "(INFO)" or "(OVERLAY)".
910 //
911 // An output section name can be followed by an address expression
912 // and/or directive. This grammar is not LL(1) because "(" can be
913 // interpreted as either the beginning of some expression or beginning
914 // of directive.
915 //
916 // https://sourceware.org/binutils/docs/ld/Output-Section-Address.html
917 // https://sourceware.org/binutils/docs/ld/Output-Section-Type.html
readSectionAddressType(OutputSection * cmd)918 void ScriptParser::readSectionAddressType(OutputSection *cmd) {
919 if (consume("(")) {
920 // Temporarily set inExpr to support TYPE=<value> without spaces.
921 SaveAndRestore saved(inExpr, true);
922 if (readSectionDirective(cmd, peek()))
923 return;
924 cmd->addrExpr = readExpr();
925 expect(")");
926 } else {
927 cmd->addrExpr = readExpr();
928 }
929
930 if (consume("(")) {
931 SaveAndRestore saved(inExpr, true);
932 StringRef tok = peek();
933 if (!readSectionDirective(cmd, tok))
934 setError("unknown section directive: " + tok);
935 }
936 }
937
checkAlignment(Expr e,std::string & loc)938 static Expr checkAlignment(Expr e, std::string &loc) {
939 return [=] {
940 uint64_t alignment = std::max((uint64_t)1, e().getValue());
941 if (!isPowerOf2_64(alignment)) {
942 error(loc + ": alignment must be power of 2");
943 return (uint64_t)1; // Return a dummy value.
944 }
945 return alignment;
946 };
947 }
948
readOverlaySectionDescription()949 OutputDesc *ScriptParser::readOverlaySectionDescription() {
950 OutputDesc *osd = script->createOutputSection(next(), getCurrentLocation());
951 osd->osec.inOverlay = true;
952 expect("{");
953 while (!errorCount() && !consume("}")) {
954 uint64_t withFlags = 0;
955 uint64_t withoutFlags = 0;
956 if (consume("INPUT_SECTION_FLAGS"))
957 std::tie(withFlags, withoutFlags) = readInputSectionFlags();
958 osd->osec.commands.push_back(
959 readInputSectionRules(next(), withFlags, withoutFlags));
960 }
961 osd->osec.phdrs = readOutputSectionPhdrs();
962 return osd;
963 }
964
readOutputSectionDescription(StringRef outSec)965 OutputDesc *ScriptParser::readOutputSectionDescription(StringRef outSec) {
966 OutputDesc *cmd =
967 script->createOutputSection(unquote(outSec), getCurrentLocation());
968 OutputSection *osec = &cmd->osec;
969 // Maybe relro. Will reset to false if DATA_SEGMENT_RELRO_END is absent.
970 osec->relro = script->seenDataAlign && !script->seenRelroEnd;
971
972 size_t symbolsReferenced = script->referencedSymbols.size();
973
974 if (peek() != ":")
975 readSectionAddressType(osec);
976 expect(":");
977
978 std::string location = getCurrentLocation();
979 if (consume("AT"))
980 osec->lmaExpr = readParenExpr();
981 if (consume("ALIGN"))
982 osec->alignExpr = checkAlignment(readParenExpr(), location);
983 if (consume("SUBALIGN"))
984 osec->subalignExpr = checkAlignment(readParenExpr(), location);
985
986 // Parse constraints.
987 if (consume("ONLY_IF_RO"))
988 osec->constraint = ConstraintKind::ReadOnly;
989 if (consume("ONLY_IF_RW"))
990 osec->constraint = ConstraintKind::ReadWrite;
991 expect("{");
992
993 while (!errorCount() && !consume("}")) {
994 StringRef tok = next();
995 if (tok == ";") {
996 // Empty commands are allowed. Do nothing here.
997 } else if (SymbolAssignment *assign = readAssignment(tok)) {
998 osec->commands.push_back(assign);
999 } else if (ByteCommand *data = readByteCommand(tok)) {
1000 osec->commands.push_back(data);
1001 } else if (tok == "CONSTRUCTORS") {
1002 // CONSTRUCTORS is a keyword to make the linker recognize C++ ctors/dtors
1003 // by name. This is for very old file formats such as ECOFF/XCOFF.
1004 // For ELF, we should ignore.
1005 } else if (tok == "FILL") {
1006 // We handle the FILL command as an alias for =fillexp section attribute,
1007 // which is different from what GNU linkers do.
1008 // https://sourceware.org/binutils/docs/ld/Output-Section-Data.html
1009 if (peek() != "(")
1010 setError("( expected, but got " + peek());
1011 osec->filler = readFill();
1012 } else if (tok == "SORT") {
1013 readSort();
1014 } else if (tok == "INCLUDE") {
1015 readInclude();
1016 } else if (tok == "(" || tok == ")") {
1017 setError("expected filename pattern");
1018 } else if (peek() == "(") {
1019 osec->commands.push_back(readInputSectionDescription(tok));
1020 } else {
1021 // We have a file name and no input sections description. It is not a
1022 // commonly used syntax, but still acceptable. In that case, all sections
1023 // from the file will be included.
1024 // FIXME: GNU ld permits INPUT_SECTION_FLAGS to be used here. We do not
1025 // handle this case here as it will already have been matched by the
1026 // case above.
1027 auto *isd = make<InputSectionDescription>(tok);
1028 isd->sectionPatterns.push_back({{}, StringMatcher("*")});
1029 osec->commands.push_back(isd);
1030 }
1031 }
1032
1033 if (consume(">"))
1034 osec->memoryRegionName = std::string(next());
1035
1036 if (consume("AT")) {
1037 expect(">");
1038 osec->lmaRegionName = std::string(next());
1039 }
1040
1041 if (osec->lmaExpr && !osec->lmaRegionName.empty())
1042 error("section can't have both LMA and a load region");
1043
1044 osec->phdrs = readOutputSectionPhdrs();
1045
1046 if (peek() == "=" || peek().starts_with("=")) {
1047 inExpr = true;
1048 consume("=");
1049 osec->filler = readFill();
1050 inExpr = false;
1051 }
1052
1053 // Consume optional comma following output section command.
1054 consume(",");
1055
1056 if (script->referencedSymbols.size() > symbolsReferenced)
1057 osec->expressionsUseSymbols = true;
1058 return cmd;
1059 }
1060
1061 // Reads a `=<fillexp>` expression and returns its value as a big-endian number.
1062 // https://sourceware.org/binutils/docs/ld/Output-Section-Fill.html
1063 // We do not support using symbols in such expressions.
1064 //
1065 // When reading a hexstring, ld.bfd handles it as a blob of arbitrary
1066 // size, while ld.gold always handles it as a 32-bit big-endian number.
1067 // We are compatible with ld.gold because it's easier to implement.
1068 // Also, we require that expressions with operators must be wrapped into
1069 // round brackets. We did it to resolve the ambiguity when parsing scripts like:
1070 // SECTIONS { .foo : { ... } =120+3 /DISCARD/ : { ... } }
readFill()1071 std::array<uint8_t, 4> ScriptParser::readFill() {
1072 uint64_t value = readPrimary()().val;
1073 if (value > UINT32_MAX)
1074 setError("filler expression result does not fit 32-bit: 0x" +
1075 Twine::utohexstr(value));
1076
1077 std::array<uint8_t, 4> buf;
1078 write32be(buf.data(), (uint32_t)value);
1079 return buf;
1080 }
1081
readProvideHidden(bool provide,bool hidden)1082 SymbolAssignment *ScriptParser::readProvideHidden(bool provide, bool hidden) {
1083 expect("(");
1084 StringRef name = next(), eq = peek();
1085 if (eq != "=") {
1086 setError("= expected, but got " + next());
1087 while (!atEOF() && next() != ")")
1088 ;
1089 return nullptr;
1090 }
1091 llvm::SaveAndRestore saveActiveProvideSym(activeProvideSym);
1092 if (provide)
1093 activeProvideSym = name;
1094 SymbolAssignment *cmd = readSymbolAssignment(name);
1095 cmd->provide = provide;
1096 cmd->hidden = hidden;
1097 expect(")");
1098 return cmd;
1099 }
1100
readAssignment(StringRef tok)1101 SymbolAssignment *ScriptParser::readAssignment(StringRef tok) {
1102 // Assert expression returns Dot, so this is equal to ".=."
1103 if (tok == "ASSERT")
1104 return make<SymbolAssignment>(".", readAssert(), 0, getCurrentLocation());
1105
1106 size_t oldPos = pos;
1107 SymbolAssignment *cmd = nullptr;
1108 bool savedSeenRelroEnd = script->seenRelroEnd;
1109 const StringRef op = peek();
1110 if (op.starts_with("=")) {
1111 // Support = followed by an expression without whitespace.
1112 SaveAndRestore saved(inExpr, true);
1113 cmd = readSymbolAssignment(tok);
1114 } else if ((op.size() == 2 && op[1] == '=' && strchr("*/+-&^|", op[0])) ||
1115 op == "<<=" || op == ">>=") {
1116 cmd = readSymbolAssignment(tok);
1117 } else if (tok == "PROVIDE") {
1118 SaveAndRestore saved(inExpr, true);
1119 cmd = readProvideHidden(true, false);
1120 } else if (tok == "HIDDEN") {
1121 SaveAndRestore saved(inExpr, true);
1122 cmd = readProvideHidden(false, true);
1123 } else if (tok == "PROVIDE_HIDDEN") {
1124 SaveAndRestore saved(inExpr, true);
1125 cmd = readProvideHidden(true, true);
1126 }
1127
1128 if (cmd) {
1129 cmd->dataSegmentRelroEnd = !savedSeenRelroEnd && script->seenRelroEnd;
1130 cmd->commandString =
1131 tok.str() + " " +
1132 llvm::join(tokens.begin() + oldPos, tokens.begin() + pos, " ");
1133 expect(";");
1134 }
1135 return cmd;
1136 }
1137
readSymbolAssignment(StringRef name)1138 SymbolAssignment *ScriptParser::readSymbolAssignment(StringRef name) {
1139 name = unquote(name);
1140 StringRef op = next();
1141 assert(op == "=" || op == "*=" || op == "/=" || op == "+=" || op == "-=" ||
1142 op == "&=" || op == "^=" || op == "|=" || op == "<<=" || op == ">>=");
1143 // Note: GNU ld does not support %=.
1144 Expr e = readExpr();
1145 if (op != "=") {
1146 std::string loc = getCurrentLocation();
1147 e = [=, c = op[0]]() -> ExprValue {
1148 ExprValue lhs = script->getSymbolValue(name, loc);
1149 switch (c) {
1150 case '*':
1151 return lhs.getValue() * e().getValue();
1152 case '/':
1153 if (uint64_t rv = e().getValue())
1154 return lhs.getValue() / rv;
1155 error(loc + ": division by zero");
1156 return 0;
1157 case '+':
1158 return add(lhs, e());
1159 case '-':
1160 return sub(lhs, e());
1161 case '<':
1162 return lhs.getValue() << e().getValue() % 64;
1163 case '>':
1164 return lhs.getValue() >> e().getValue() % 64;
1165 case '&':
1166 return lhs.getValue() & e().getValue();
1167 case '^':
1168 return lhs.getValue() ^ e().getValue();
1169 case '|':
1170 return lhs.getValue() | e().getValue();
1171 default:
1172 llvm_unreachable("");
1173 }
1174 };
1175 }
1176 return make<SymbolAssignment>(name, e, ctx.scriptSymOrderCounter++,
1177 getCurrentLocation());
1178 }
1179
1180 // This is an operator-precedence parser to parse a linker
1181 // script expression.
readExpr()1182 Expr ScriptParser::readExpr() {
1183 // Our lexer is context-aware. Set the in-expression bit so that
1184 // they apply different tokenization rules.
1185 SaveAndRestore saved(inExpr, true);
1186 Expr e = readExpr1(readPrimary(), 0);
1187 return e;
1188 }
1189
combine(StringRef op,Expr l,Expr r)1190 Expr ScriptParser::combine(StringRef op, Expr l, Expr r) {
1191 if (op == "+")
1192 return [=] { return add(l(), r()); };
1193 if (op == "-")
1194 return [=] { return sub(l(), r()); };
1195 if (op == "*")
1196 return [=] { return l().getValue() * r().getValue(); };
1197 if (op == "/") {
1198 std::string loc = getCurrentLocation();
1199 return [=]() -> uint64_t {
1200 if (uint64_t rv = r().getValue())
1201 return l().getValue() / rv;
1202 error(loc + ": division by zero");
1203 return 0;
1204 };
1205 }
1206 if (op == "%") {
1207 std::string loc = getCurrentLocation();
1208 return [=]() -> uint64_t {
1209 if (uint64_t rv = r().getValue())
1210 return l().getValue() % rv;
1211 error(loc + ": modulo by zero");
1212 return 0;
1213 };
1214 }
1215 if (op == "<<")
1216 return [=] { return l().getValue() << r().getValue() % 64; };
1217 if (op == ">>")
1218 return [=] { return l().getValue() >> r().getValue() % 64; };
1219 if (op == "<")
1220 return [=] { return l().getValue() < r().getValue(); };
1221 if (op == ">")
1222 return [=] { return l().getValue() > r().getValue(); };
1223 if (op == ">=")
1224 return [=] { return l().getValue() >= r().getValue(); };
1225 if (op == "<=")
1226 return [=] { return l().getValue() <= r().getValue(); };
1227 if (op == "==")
1228 return [=] { return l().getValue() == r().getValue(); };
1229 if (op == "!=")
1230 return [=] { return l().getValue() != r().getValue(); };
1231 if (op == "||")
1232 return [=] { return l().getValue() || r().getValue(); };
1233 if (op == "&&")
1234 return [=] { return l().getValue() && r().getValue(); };
1235 if (op == "&")
1236 return [=] { return bitAnd(l(), r()); };
1237 if (op == "^")
1238 return [=] { return bitXor(l(), r()); };
1239 if (op == "|")
1240 return [=] { return bitOr(l(), r()); };
1241 llvm_unreachable("invalid operator");
1242 }
1243
1244 // This is a part of the operator-precedence parser. This function
1245 // assumes that the remaining token stream starts with an operator.
readExpr1(Expr lhs,int minPrec)1246 Expr ScriptParser::readExpr1(Expr lhs, int minPrec) {
1247 while (!atEOF() && !errorCount()) {
1248 // Read an operator and an expression.
1249 StringRef op1 = peek();
1250 if (precedence(op1) < minPrec)
1251 break;
1252 skip();
1253 if (op1 == "?")
1254 return readTernary(lhs);
1255 Expr rhs = readPrimary();
1256
1257 // Evaluate the remaining part of the expression first if the
1258 // next operator has greater precedence than the previous one.
1259 // For example, if we have read "+" and "3", and if the next
1260 // operator is "*", then we'll evaluate 3 * ... part first.
1261 while (!atEOF()) {
1262 StringRef op2 = peek();
1263 if (precedence(op2) <= precedence(op1))
1264 break;
1265 rhs = readExpr1(rhs, precedence(op2));
1266 }
1267
1268 lhs = combine(op1, lhs, rhs);
1269 }
1270 return lhs;
1271 }
1272
getPageSize()1273 Expr ScriptParser::getPageSize() {
1274 std::string location = getCurrentLocation();
1275 return [=]() -> uint64_t {
1276 if (target)
1277 return config->commonPageSize;
1278 error(location + ": unable to calculate page size");
1279 return 4096; // Return a dummy value.
1280 };
1281 }
1282
readConstant()1283 Expr ScriptParser::readConstant() {
1284 StringRef s = readParenLiteral();
1285 if (s == "COMMONPAGESIZE")
1286 return getPageSize();
1287 if (s == "MAXPAGESIZE")
1288 return [] { return config->maxPageSize; };
1289 setError("unknown constant: " + s);
1290 return [] { return 0; };
1291 }
1292
1293 // Parses Tok as an integer. It recognizes hexadecimal (prefixed with
1294 // "0x" or suffixed with "H") and decimal numbers. Decimal numbers may
1295 // have "K" (Ki) or "M" (Mi) suffixes.
parseInt(StringRef tok)1296 static std::optional<uint64_t> parseInt(StringRef tok) {
1297 // Hexadecimal
1298 uint64_t val;
1299 if (tok.starts_with_insensitive("0x")) {
1300 if (!to_integer(tok.substr(2), val, 16))
1301 return std::nullopt;
1302 return val;
1303 }
1304 if (tok.ends_with_insensitive("H")) {
1305 if (!to_integer(tok.drop_back(), val, 16))
1306 return std::nullopt;
1307 return val;
1308 }
1309
1310 // Decimal
1311 if (tok.ends_with_insensitive("K")) {
1312 if (!to_integer(tok.drop_back(), val, 10))
1313 return std::nullopt;
1314 return val * 1024;
1315 }
1316 if (tok.ends_with_insensitive("M")) {
1317 if (!to_integer(tok.drop_back(), val, 10))
1318 return std::nullopt;
1319 return val * 1024 * 1024;
1320 }
1321 if (!to_integer(tok, val, 10))
1322 return std::nullopt;
1323 return val;
1324 }
1325
readByteCommand(StringRef tok)1326 ByteCommand *ScriptParser::readByteCommand(StringRef tok) {
1327 int size = StringSwitch<int>(tok)
1328 .Case("BYTE", 1)
1329 .Case("SHORT", 2)
1330 .Case("LONG", 4)
1331 .Case("QUAD", 8)
1332 .Default(-1);
1333 if (size == -1)
1334 return nullptr;
1335
1336 size_t oldPos = pos;
1337 Expr e = readParenExpr();
1338 std::string commandString =
1339 tok.str() + " " +
1340 llvm::join(tokens.begin() + oldPos, tokens.begin() + pos, " ");
1341 return make<ByteCommand>(e, size, commandString);
1342 }
1343
parseFlag(StringRef tok)1344 static std::optional<uint64_t> parseFlag(StringRef tok) {
1345 if (std::optional<uint64_t> asInt = parseInt(tok))
1346 return asInt;
1347 #define CASE_ENT(enum) #enum, ELF::enum
1348 return StringSwitch<std::optional<uint64_t>>(tok)
1349 .Case(CASE_ENT(SHF_WRITE))
1350 .Case(CASE_ENT(SHF_ALLOC))
1351 .Case(CASE_ENT(SHF_EXECINSTR))
1352 .Case(CASE_ENT(SHF_MERGE))
1353 .Case(CASE_ENT(SHF_STRINGS))
1354 .Case(CASE_ENT(SHF_INFO_LINK))
1355 .Case(CASE_ENT(SHF_LINK_ORDER))
1356 .Case(CASE_ENT(SHF_OS_NONCONFORMING))
1357 .Case(CASE_ENT(SHF_GROUP))
1358 .Case(CASE_ENT(SHF_TLS))
1359 .Case(CASE_ENT(SHF_COMPRESSED))
1360 .Case(CASE_ENT(SHF_EXCLUDE))
1361 .Case(CASE_ENT(SHF_ARM_PURECODE))
1362 .Default(std::nullopt);
1363 #undef CASE_ENT
1364 }
1365
1366 // Reads the '(' <flags> ')' list of section flags in
1367 // INPUT_SECTION_FLAGS '(' <flags> ')' in the
1368 // following form:
1369 // <flags> ::= <flag>
1370 // | <flags> & flag
1371 // <flag> ::= Recognized Flag Name, or Integer value of flag.
1372 // If the first character of <flag> is a ! then this means without flag,
1373 // otherwise with flag.
1374 // Example: SHF_EXECINSTR & !SHF_WRITE means with flag SHF_EXECINSTR and
1375 // without flag SHF_WRITE.
readInputSectionFlags()1376 std::pair<uint64_t, uint64_t> ScriptParser::readInputSectionFlags() {
1377 uint64_t withFlags = 0;
1378 uint64_t withoutFlags = 0;
1379 expect("(");
1380 while (!errorCount()) {
1381 StringRef tok = unquote(next());
1382 bool without = tok.consume_front("!");
1383 if (std::optional<uint64_t> flag = parseFlag(tok)) {
1384 if (without)
1385 withoutFlags |= *flag;
1386 else
1387 withFlags |= *flag;
1388 } else {
1389 setError("unrecognised flag: " + tok);
1390 }
1391 if (consume(")"))
1392 break;
1393 if (!consume("&")) {
1394 next();
1395 setError("expected & or )");
1396 }
1397 }
1398 return std::make_pair(withFlags, withoutFlags);
1399 }
1400
readParenLiteral()1401 StringRef ScriptParser::readParenLiteral() {
1402 expect("(");
1403 bool orig = inExpr;
1404 inExpr = false;
1405 StringRef tok = next();
1406 inExpr = orig;
1407 expect(")");
1408 return tok;
1409 }
1410
checkIfExists(const OutputSection & osec,StringRef location)1411 static void checkIfExists(const OutputSection &osec, StringRef location) {
1412 if (osec.location.empty() && script->errorOnMissingSection)
1413 script->recordError(location + ": undefined section " + osec.name);
1414 }
1415
isValidSymbolName(StringRef s)1416 static bool isValidSymbolName(StringRef s) {
1417 auto valid = [](char c) {
1418 return isAlnum(c) || c == '$' || c == '.' || c == '_';
1419 };
1420 return !s.empty() && !isDigit(s[0]) && llvm::all_of(s, valid);
1421 }
1422
readPrimary()1423 Expr ScriptParser::readPrimary() {
1424 if (peek() == "(")
1425 return readParenExpr();
1426
1427 if (consume("~")) {
1428 Expr e = readPrimary();
1429 return [=] { return ~e().getValue(); };
1430 }
1431 if (consume("!")) {
1432 Expr e = readPrimary();
1433 return [=] { return !e().getValue(); };
1434 }
1435 if (consume("-")) {
1436 Expr e = readPrimary();
1437 return [=] { return -e().getValue(); };
1438 }
1439
1440 StringRef tok = next();
1441 std::string location = getCurrentLocation();
1442
1443 // Built-in functions are parsed here.
1444 // https://sourceware.org/binutils/docs/ld/Builtin-Functions.html.
1445 if (tok == "ABSOLUTE") {
1446 Expr inner = readParenExpr();
1447 return [=] {
1448 ExprValue i = inner();
1449 i.forceAbsolute = true;
1450 return i;
1451 };
1452 }
1453 if (tok == "ADDR") {
1454 StringRef name = unquote(readParenLiteral());
1455 OutputSection *osec = &script->getOrCreateOutputSection(name)->osec;
1456 osec->usedInExpression = true;
1457 return [=]() -> ExprValue {
1458 checkIfExists(*osec, location);
1459 return {osec, false, 0, location};
1460 };
1461 }
1462 if (tok == "ALIGN") {
1463 expect("(");
1464 Expr e = readExpr();
1465 if (consume(")")) {
1466 e = checkAlignment(e, location);
1467 return [=] { return alignToPowerOf2(script->getDot(), e().getValue()); };
1468 }
1469 expect(",");
1470 Expr e2 = checkAlignment(readExpr(), location);
1471 expect(")");
1472 return [=] {
1473 ExprValue v = e();
1474 v.alignment = e2().getValue();
1475 return v;
1476 };
1477 }
1478 if (tok == "ALIGNOF") {
1479 StringRef name = unquote(readParenLiteral());
1480 OutputSection *osec = &script->getOrCreateOutputSection(name)->osec;
1481 return [=] {
1482 checkIfExists(*osec, location);
1483 return osec->addralign;
1484 };
1485 }
1486 if (tok == "ASSERT")
1487 return readAssert();
1488 if (tok == "CONSTANT")
1489 return readConstant();
1490 if (tok == "DATA_SEGMENT_ALIGN") {
1491 expect("(");
1492 Expr e = readExpr();
1493 expect(",");
1494 readExpr();
1495 expect(")");
1496 script->seenDataAlign = true;
1497 return [=] {
1498 uint64_t align = std::max(uint64_t(1), e().getValue());
1499 return (script->getDot() + align - 1) & -align;
1500 };
1501 }
1502 if (tok == "DATA_SEGMENT_END") {
1503 expect("(");
1504 expect(".");
1505 expect(")");
1506 return [] { return script->getDot(); };
1507 }
1508 if (tok == "DATA_SEGMENT_RELRO_END") {
1509 // GNU linkers implements more complicated logic to handle
1510 // DATA_SEGMENT_RELRO_END. We instead ignore the arguments and
1511 // just align to the next page boundary for simplicity.
1512 expect("(");
1513 readExpr();
1514 expect(",");
1515 readExpr();
1516 expect(")");
1517 script->seenRelroEnd = true;
1518 return [=] { return alignToPowerOf2(script->getDot(), config->maxPageSize); };
1519 }
1520 if (tok == "DEFINED") {
1521 StringRef name = unquote(readParenLiteral());
1522 // Return 1 if s is defined. If the definition is only found in a linker
1523 // script, it must happen before this DEFINED.
1524 auto order = ctx.scriptSymOrderCounter++;
1525 return [=] {
1526 Symbol *s = symtab.find(name);
1527 return s && s->isDefined() && ctx.scriptSymOrder.lookup(s) < order ? 1
1528 : 0;
1529 };
1530 }
1531 if (tok == "LENGTH") {
1532 StringRef name = readParenLiteral();
1533 if (script->memoryRegions.count(name) == 0) {
1534 setError("memory region not defined: " + name);
1535 return [] { return 0; };
1536 }
1537 return script->memoryRegions[name]->length;
1538 }
1539 if (tok == "LOADADDR") {
1540 StringRef name = unquote(readParenLiteral());
1541 OutputSection *osec = &script->getOrCreateOutputSection(name)->osec;
1542 osec->usedInExpression = true;
1543 return [=] {
1544 checkIfExists(*osec, location);
1545 return osec->getLMA();
1546 };
1547 }
1548 if (tok == "LOG2CEIL") {
1549 expect("(");
1550 Expr a = readExpr();
1551 expect(")");
1552 return [=] {
1553 // LOG2CEIL(0) is defined to be 0.
1554 return llvm::Log2_64_Ceil(std::max(a().getValue(), UINT64_C(1)));
1555 };
1556 }
1557 if (tok == "MAX" || tok == "MIN") {
1558 expect("(");
1559 Expr a = readExpr();
1560 expect(",");
1561 Expr b = readExpr();
1562 expect(")");
1563 if (tok == "MIN")
1564 return [=] { return std::min(a().getValue(), b().getValue()); };
1565 return [=] { return std::max(a().getValue(), b().getValue()); };
1566 }
1567 if (tok == "ORIGIN") {
1568 StringRef name = readParenLiteral();
1569 if (script->memoryRegions.count(name) == 0) {
1570 setError("memory region not defined: " + name);
1571 return [] { return 0; };
1572 }
1573 return script->memoryRegions[name]->origin;
1574 }
1575 if (tok == "SEGMENT_START") {
1576 expect("(");
1577 skip();
1578 expect(",");
1579 Expr e = readExpr();
1580 expect(")");
1581 return [=] { return e(); };
1582 }
1583 if (tok == "SIZEOF") {
1584 StringRef name = unquote(readParenLiteral());
1585 OutputSection *cmd = &script->getOrCreateOutputSection(name)->osec;
1586 // Linker script does not create an output section if its content is empty.
1587 // We want to allow SIZEOF(.foo) where .foo is a section which happened to
1588 // be empty.
1589 return [=] { return cmd->size; };
1590 }
1591 if (tok == "SIZEOF_HEADERS")
1592 return [=] { return elf::getHeaderSize(); };
1593
1594 // Tok is the dot.
1595 if (tok == ".")
1596 return [=] { return script->getSymbolValue(tok, location); };
1597
1598 // Tok is a literal number.
1599 if (std::optional<uint64_t> val = parseInt(tok))
1600 return [=] { return *val; };
1601
1602 // Tok is a symbol name.
1603 if (tok.starts_with("\""))
1604 tok = unquote(tok);
1605 else if (!isValidSymbolName(tok))
1606 setError("malformed number: " + tok);
1607 if (activeProvideSym)
1608 script->provideMap[*activeProvideSym].push_back(tok);
1609 else
1610 script->referencedSymbols.push_back(tok);
1611 return [=] { return script->getSymbolValue(tok, location); };
1612 }
1613
readTernary(Expr cond)1614 Expr ScriptParser::readTernary(Expr cond) {
1615 Expr l = readExpr();
1616 expect(":");
1617 Expr r = readExpr();
1618 return [=] { return cond().getValue() ? l() : r(); };
1619 }
1620
readParenExpr()1621 Expr ScriptParser::readParenExpr() {
1622 expect("(");
1623 Expr e = readExpr();
1624 expect(")");
1625 return e;
1626 }
1627
readOutputSectionPhdrs()1628 SmallVector<StringRef, 0> ScriptParser::readOutputSectionPhdrs() {
1629 SmallVector<StringRef, 0> phdrs;
1630 while (!errorCount() && peek().starts_with(":")) {
1631 StringRef tok = next();
1632 phdrs.push_back((tok.size() == 1) ? next() : tok.substr(1));
1633 }
1634 return phdrs;
1635 }
1636
1637 // Read a program header type name. The next token must be a
1638 // name of a program header type or a constant (e.g. "0x3").
readPhdrType()1639 unsigned ScriptParser::readPhdrType() {
1640 StringRef tok = next();
1641 if (std::optional<uint64_t> val = parseInt(tok))
1642 return *val;
1643
1644 unsigned ret = StringSwitch<unsigned>(tok)
1645 .Case("PT_NULL", PT_NULL)
1646 .Case("PT_LOAD", PT_LOAD)
1647 .Case("PT_DYNAMIC", PT_DYNAMIC)
1648 .Case("PT_INTERP", PT_INTERP)
1649 .Case("PT_NOTE", PT_NOTE)
1650 .Case("PT_SHLIB", PT_SHLIB)
1651 .Case("PT_PHDR", PT_PHDR)
1652 .Case("PT_TLS", PT_TLS)
1653 .Case("PT_GNU_EH_FRAME", PT_GNU_EH_FRAME)
1654 .Case("PT_GNU_STACK", PT_GNU_STACK)
1655 .Case("PT_GNU_RELRO", PT_GNU_RELRO)
1656 .Case("PT_OPENBSD_MUTABLE", PT_OPENBSD_MUTABLE)
1657 .Case("PT_OPENBSD_RANDOMIZE", PT_OPENBSD_RANDOMIZE)
1658 .Case("PT_OPENBSD_SYSCALLS", PT_OPENBSD_SYSCALLS)
1659 .Case("PT_OPENBSD_WXNEEDED", PT_OPENBSD_WXNEEDED)
1660 .Case("PT_OPENBSD_BOOTDATA", PT_OPENBSD_BOOTDATA)
1661 .Default(-1);
1662
1663 if (ret == (unsigned)-1) {
1664 setError("invalid program header type: " + tok);
1665 return PT_NULL;
1666 }
1667 return ret;
1668 }
1669
1670 // Reads an anonymous version declaration.
readAnonymousDeclaration()1671 void ScriptParser::readAnonymousDeclaration() {
1672 SmallVector<SymbolVersion, 0> locals;
1673 SmallVector<SymbolVersion, 0> globals;
1674 std::tie(locals, globals) = readSymbols();
1675 for (const SymbolVersion &pat : locals)
1676 config->versionDefinitions[VER_NDX_LOCAL].localPatterns.push_back(pat);
1677 for (const SymbolVersion &pat : globals)
1678 config->versionDefinitions[VER_NDX_GLOBAL].nonLocalPatterns.push_back(pat);
1679
1680 expect(";");
1681 }
1682
1683 // Reads a non-anonymous version definition,
1684 // e.g. "VerStr { global: foo; bar; local: *; };".
readVersionDeclaration(StringRef verStr)1685 void ScriptParser::readVersionDeclaration(StringRef verStr) {
1686 // Read a symbol list.
1687 SmallVector<SymbolVersion, 0> locals;
1688 SmallVector<SymbolVersion, 0> globals;
1689 std::tie(locals, globals) = readSymbols();
1690
1691 // Create a new version definition and add that to the global symbols.
1692 VersionDefinition ver;
1693 ver.name = verStr;
1694 ver.nonLocalPatterns = std::move(globals);
1695 ver.localPatterns = std::move(locals);
1696 ver.id = config->versionDefinitions.size();
1697 config->versionDefinitions.push_back(ver);
1698
1699 // Each version may have a parent version. For example, "Ver2"
1700 // defined as "Ver2 { global: foo; local: *; } Ver1;" has "Ver1"
1701 // as a parent. This version hierarchy is, probably against your
1702 // instinct, purely for hint; the runtime doesn't care about it
1703 // at all. In LLD, we simply ignore it.
1704 if (next() != ";")
1705 expect(";");
1706 }
1707
hasWildcard(StringRef s)1708 bool elf::hasWildcard(StringRef s) {
1709 return s.find_first_of("?*[") != StringRef::npos;
1710 }
1711
1712 // Reads a list of symbols, e.g. "{ global: foo; bar; local: *; };".
1713 std::pair<SmallVector<SymbolVersion, 0>, SmallVector<SymbolVersion, 0>>
readSymbols()1714 ScriptParser::readSymbols() {
1715 SmallVector<SymbolVersion, 0> locals;
1716 SmallVector<SymbolVersion, 0> globals;
1717 SmallVector<SymbolVersion, 0> *v = &globals;
1718
1719 while (!errorCount()) {
1720 if (consume("}"))
1721 break;
1722 if (consumeLabel("local")) {
1723 v = &locals;
1724 continue;
1725 }
1726 if (consumeLabel("global")) {
1727 v = &globals;
1728 continue;
1729 }
1730
1731 if (consume("extern")) {
1732 SmallVector<SymbolVersion, 0> ext = readVersionExtern();
1733 v->insert(v->end(), ext.begin(), ext.end());
1734 } else {
1735 StringRef tok = next();
1736 v->push_back({unquote(tok), false, hasWildcard(tok)});
1737 }
1738 expect(";");
1739 }
1740 return {locals, globals};
1741 }
1742
1743 // Reads an "extern C++" directive, e.g.,
1744 // "extern "C++" { ns::*; "f(int, double)"; };"
1745 //
1746 // The last semicolon is optional. E.g. this is OK:
1747 // "extern "C++" { ns::*; "f(int, double)" };"
readVersionExtern()1748 SmallVector<SymbolVersion, 0> ScriptParser::readVersionExtern() {
1749 StringRef tok = next();
1750 bool isCXX = tok == "\"C++\"";
1751 if (!isCXX && tok != "\"C\"")
1752 setError("Unknown language");
1753 expect("{");
1754
1755 SmallVector<SymbolVersion, 0> ret;
1756 while (!errorCount() && peek() != "}") {
1757 StringRef tok = next();
1758 ret.push_back(
1759 {unquote(tok), isCXX, !tok.starts_with("\"") && hasWildcard(tok)});
1760 if (consume("}"))
1761 return ret;
1762 expect(";");
1763 }
1764
1765 expect("}");
1766 return ret;
1767 }
1768
readMemoryAssignment(StringRef s1,StringRef s2,StringRef s3)1769 Expr ScriptParser::readMemoryAssignment(StringRef s1, StringRef s2,
1770 StringRef s3) {
1771 if (!consume(s1) && !consume(s2) && !consume(s3)) {
1772 setError("expected one of: " + s1 + ", " + s2 + ", or " + s3);
1773 return [] { return 0; };
1774 }
1775 expect("=");
1776 return readExpr();
1777 }
1778
1779 // Parse the MEMORY command as specified in:
1780 // https://sourceware.org/binutils/docs/ld/MEMORY.html
1781 //
1782 // MEMORY { name [(attr)] : ORIGIN = origin, LENGTH = len ... }
readMemory()1783 void ScriptParser::readMemory() {
1784 expect("{");
1785 while (!errorCount() && !consume("}")) {
1786 StringRef tok = next();
1787 if (tok == "INCLUDE") {
1788 readInclude();
1789 continue;
1790 }
1791
1792 uint32_t flags = 0;
1793 uint32_t invFlags = 0;
1794 uint32_t negFlags = 0;
1795 uint32_t negInvFlags = 0;
1796 if (consume("(")) {
1797 readMemoryAttributes(flags, invFlags, negFlags, negInvFlags);
1798 expect(")");
1799 }
1800 expect(":");
1801
1802 Expr origin = readMemoryAssignment("ORIGIN", "org", "o");
1803 expect(",");
1804 Expr length = readMemoryAssignment("LENGTH", "len", "l");
1805
1806 // Add the memory region to the region map.
1807 MemoryRegion *mr = make<MemoryRegion>(tok, origin, length, flags, invFlags,
1808 negFlags, negInvFlags);
1809 if (!script->memoryRegions.insert({tok, mr}).second)
1810 setError("region '" + tok + "' already defined");
1811 }
1812 }
1813
1814 // This function parses the attributes used to match against section
1815 // flags when placing output sections in a memory region. These flags
1816 // are only used when an explicit memory region name is not used.
readMemoryAttributes(uint32_t & flags,uint32_t & invFlags,uint32_t & negFlags,uint32_t & negInvFlags)1817 void ScriptParser::readMemoryAttributes(uint32_t &flags, uint32_t &invFlags,
1818 uint32_t &negFlags,
1819 uint32_t &negInvFlags) {
1820 bool invert = false;
1821
1822 for (char c : next().lower()) {
1823 if (c == '!') {
1824 invert = !invert;
1825 std::swap(flags, negFlags);
1826 std::swap(invFlags, negInvFlags);
1827 continue;
1828 }
1829 if (c == 'w')
1830 flags |= SHF_WRITE;
1831 else if (c == 'x')
1832 flags |= SHF_EXECINSTR;
1833 else if (c == 'a')
1834 flags |= SHF_ALLOC;
1835 else if (c == 'r')
1836 invFlags |= SHF_WRITE;
1837 else
1838 setError("invalid memory region attribute");
1839 }
1840
1841 if (invert) {
1842 std::swap(flags, negFlags);
1843 std::swap(invFlags, negInvFlags);
1844 }
1845 }
1846
readLinkerScript(MemoryBufferRef mb)1847 void elf::readLinkerScript(MemoryBufferRef mb) {
1848 llvm::TimeTraceScope timeScope("Read linker script",
1849 mb.getBufferIdentifier());
1850 ScriptParser(mb).readLinkerScript();
1851 }
1852
readVersionScript(MemoryBufferRef mb)1853 void elf::readVersionScript(MemoryBufferRef mb) {
1854 llvm::TimeTraceScope timeScope("Read version script",
1855 mb.getBufferIdentifier());
1856 ScriptParser(mb).readVersionScript();
1857 }
1858
readDynamicList(MemoryBufferRef mb)1859 void elf::readDynamicList(MemoryBufferRef mb) {
1860 llvm::TimeTraceScope timeScope("Read dynamic list", mb.getBufferIdentifier());
1861 ScriptParser(mb).readDynamicList();
1862 }
1863
readDefsym(StringRef name,MemoryBufferRef mb)1864 void elf::readDefsym(StringRef name, MemoryBufferRef mb) {
1865 llvm::TimeTraceScope timeScope("Read defsym input", name);
1866 ScriptParser(mb).readDefsym(name);
1867 }
1868