1 //===- OutputSections.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 #include "OutputSections.h"
10 #include "Config.h"
11 #include "LinkerScript.h"
12 #include "SymbolTable.h"
13 #include "SyntheticSections.h"
14 #include "Target.h"
15 #include "lld/Common/Memory.h"
16 #include "lld/Common/Strings.h"
17 #include "lld/Common/Threads.h"
18 #include "llvm/BinaryFormat/Dwarf.h"
19 #include "llvm/Support/Compression.h"
20 #include "llvm/Support/MD5.h"
21 #include "llvm/Support/MathExtras.h"
22 #include "llvm/Support/SHA1.h"
23 
24 using namespace llvm;
25 using namespace llvm::dwarf;
26 using namespace llvm::object;
27 using namespace llvm::support::endian;
28 using namespace llvm::ELF;
29 
30 using namespace lld;
31 using namespace lld::elf;
32 
33 uint8_t *Out::BufferStart;
34 uint8_t Out::First;
35 PhdrEntry *Out::TlsPhdr;
36 OutputSection *Out::ElfHeader;
37 OutputSection *Out::ProgramHeaders;
38 OutputSection *Out::PreinitArray;
39 OutputSection *Out::InitArray;
40 OutputSection *Out::FiniArray;
41 
42 std::vector<OutputSection *> elf::OutputSections;
43 
44 uint32_t OutputSection::getPhdrFlags() const {
45   uint32_t Ret = 0;
46   if (Config->EMachine != EM_ARM || !(Flags & SHF_ARM_PURECODE))
47     Ret |= PF_R;
48   if (Flags & SHF_WRITE)
49     Ret |= PF_W;
50   if (Flags & SHF_EXECINSTR)
51     Ret |= PF_X;
52   return Ret;
53 }
54 
55 template <class ELFT>
56 void OutputSection::writeHeaderTo(typename ELFT::Shdr *Shdr) {
57   Shdr->sh_entsize = Entsize;
58   Shdr->sh_addralign = Alignment;
59   Shdr->sh_type = Type;
60   Shdr->sh_offset = Offset;
61   Shdr->sh_flags = Flags;
62   Shdr->sh_info = Info;
63   Shdr->sh_link = Link;
64   Shdr->sh_addr = Addr;
65   Shdr->sh_size = Size;
66   Shdr->sh_name = ShName;
67 }
68 
69 OutputSection::OutputSection(StringRef Name, uint32_t Type, uint64_t Flags)
70     : BaseCommand(OutputSectionKind),
71       SectionBase(Output, Name, Flags, /*Entsize*/ 0, /*Alignment*/ 1, Type,
72                   /*Info*/ 0, /*Link*/ 0) {}
73 
74 // We allow sections of types listed below to merged into a
75 // single progbits section. This is typically done by linker
76 // scripts. Merging nobits and progbits will force disk space
77 // to be allocated for nobits sections. Other ones don't require
78 // any special treatment on top of progbits, so there doesn't
79 // seem to be a harm in merging them.
80 static bool canMergeToProgbits(unsigned Type) {
81   return Type == SHT_NOBITS || Type == SHT_PROGBITS || Type == SHT_INIT_ARRAY ||
82          Type == SHT_PREINIT_ARRAY || Type == SHT_FINI_ARRAY ||
83          Type == SHT_NOTE;
84 }
85 
86 void OutputSection::addSection(InputSection *IS) {
87   if (!HasInputSections) {
88     // If IS is the first section to be added to this section,
89     // initialize Partition, Type, Entsize and flags from IS.
90     HasInputSections = true;
91     Partition = IS->Partition;
92     Type = IS->Type;
93     Entsize = IS->Entsize;
94     Flags = IS->Flags;
95   } else {
96     // Otherwise, check if new type or flags are compatible with existing ones.
97     unsigned Mask = SHF_TLS | SHF_LINK_ORDER;
98     if ((Flags & Mask) != (IS->Flags & Mask))
99       error("incompatible section flags for " + Name + "\n>>> " + toString(IS) +
100             ": 0x" + utohexstr(IS->Flags) + "\n>>> output section " + Name +
101             ": 0x" + utohexstr(Flags));
102 
103     if (Type != IS->Type) {
104       if (!canMergeToProgbits(Type) || !canMergeToProgbits(IS->Type))
105         error("section type mismatch for " + IS->Name + "\n>>> " +
106               toString(IS) + ": " +
107               getELFSectionTypeName(Config->EMachine, IS->Type) +
108               "\n>>> output section " + Name + ": " +
109               getELFSectionTypeName(Config->EMachine, Type));
110       Type = SHT_PROGBITS;
111     }
112   }
113 
114   IS->Parent = this;
115   uint64_t AndMask =
116       Config->EMachine == EM_ARM ? (uint64_t)SHF_ARM_PURECODE : 0;
117   uint64_t OrMask = ~AndMask;
118   uint64_t AndFlags = (Flags & IS->Flags) & AndMask;
119   uint64_t OrFlags = (Flags | IS->Flags) & OrMask;
120   Flags = AndFlags | OrFlags;
121 
122   Alignment = std::max(Alignment, IS->Alignment);
123 
124   // If this section contains a table of fixed-size entries, sh_entsize
125   // holds the element size. If it contains elements of different size we
126   // set sh_entsize to 0.
127   if (Entsize != IS->Entsize)
128     Entsize = 0;
129 
130   if (!IS->Assigned) {
131     IS->Assigned = true;
132     if (SectionCommands.empty() ||
133         !isa<InputSectionDescription>(SectionCommands.back()))
134       SectionCommands.push_back(make<InputSectionDescription>(""));
135     auto *ISD = cast<InputSectionDescription>(SectionCommands.back());
136     ISD->Sections.push_back(IS);
137   }
138 }
139 
140 static void sortByOrder(MutableArrayRef<InputSection *> In,
141                         llvm::function_ref<int(InputSectionBase *S)> Order) {
142   std::vector<std::pair<int, InputSection *>> V;
143   for (InputSection *S : In)
144     V.push_back({Order(S), S});
145   llvm::stable_sort(V, less_first());
146 
147   for (size_t I = 0; I < V.size(); ++I)
148     In[I] = V[I].second;
149 }
150 
151 uint64_t elf::getHeaderSize() {
152   if (Config->OFormatBinary)
153     return 0;
154   return Out::ElfHeader->Size + Out::ProgramHeaders->Size;
155 }
156 
157 bool OutputSection::classof(const BaseCommand *C) {
158   return C->Kind == OutputSectionKind;
159 }
160 
161 void OutputSection::sort(llvm::function_ref<int(InputSectionBase *S)> Order) {
162   assert(isLive());
163   for (BaseCommand *B : SectionCommands)
164     if (auto *ISD = dyn_cast<InputSectionDescription>(B))
165       sortByOrder(ISD->Sections, Order);
166 }
167 
168 // Fill [Buf, Buf + Size) with Filler.
169 // This is used for linker script "=fillexp" command.
170 static void fill(uint8_t *Buf, size_t Size,
171                  const std::array<uint8_t, 4> &Filler) {
172   size_t I = 0;
173   for (; I + 4 < Size; I += 4)
174     memcpy(Buf + I, Filler.data(), 4);
175   memcpy(Buf + I, Filler.data(), Size - I);
176 }
177 
178 // Compress section contents if this section contains debug info.
179 template <class ELFT> void OutputSection::maybeCompress() {
180   using Elf_Chdr = typename ELFT::Chdr;
181 
182   // Compress only DWARF debug sections.
183   if (!Config->CompressDebugSections || (Flags & SHF_ALLOC) ||
184       !Name.startswith(".debug_"))
185     return;
186 
187   // Create a section header.
188   ZDebugHeader.resize(sizeof(Elf_Chdr));
189   auto *Hdr = reinterpret_cast<Elf_Chdr *>(ZDebugHeader.data());
190   Hdr->ch_type = ELFCOMPRESS_ZLIB;
191   Hdr->ch_size = Size;
192   Hdr->ch_addralign = Alignment;
193 
194   // Write section contents to a temporary buffer and compress it.
195   std::vector<uint8_t> Buf(Size);
196   writeTo<ELFT>(Buf.data());
197   if (Error E = zlib::compress(toStringRef(Buf), CompressedData))
198     fatal("compress failed: " + llvm::toString(std::move(E)));
199 
200   // Update section headers.
201   Size = sizeof(Elf_Chdr) + CompressedData.size();
202   Flags |= SHF_COMPRESSED;
203 }
204 
205 static void writeInt(uint8_t *Buf, uint64_t Data, uint64_t Size) {
206   if (Size == 1)
207     *Buf = Data;
208   else if (Size == 2)
209     write16(Buf, Data);
210   else if (Size == 4)
211     write32(Buf, Data);
212   else if (Size == 8)
213     write64(Buf, Data);
214   else
215     llvm_unreachable("unsupported Size argument");
216 }
217 
218 template <class ELFT> void OutputSection::writeTo(uint8_t *Buf) {
219   if (Type == SHT_NOBITS)
220     return;
221 
222   // If -compress-debug-section is specified and if this is a debug seciton,
223   // we've already compressed section contents. If that's the case,
224   // just write it down.
225   if (!CompressedData.empty()) {
226     memcpy(Buf, ZDebugHeader.data(), ZDebugHeader.size());
227     memcpy(Buf + ZDebugHeader.size(), CompressedData.data(),
228            CompressedData.size());
229     return;
230   }
231 
232   // Write leading padding.
233   std::vector<InputSection *> Sections = getInputSections(this);
234   std::array<uint8_t, 4> Filler = getFiller();
235   bool NonZeroFiller = read32(Filler.data()) != 0;
236   if (NonZeroFiller)
237     fill(Buf, Sections.empty() ? Size : Sections[0]->OutSecOff, Filler);
238 
239   parallelForEachN(0, Sections.size(), [&](size_t I) {
240     InputSection *IS = Sections[I];
241     IS->writeTo<ELFT>(Buf);
242 
243     // Fill gaps between sections.
244     if (NonZeroFiller) {
245       uint8_t *Start = Buf + IS->OutSecOff + IS->getSize();
246       uint8_t *End;
247       if (I + 1 == Sections.size())
248         End = Buf + Size;
249       else
250         End = Buf + Sections[I + 1]->OutSecOff;
251       fill(Start, End - Start, Filler);
252     }
253   });
254 
255   // Linker scripts may have BYTE()-family commands with which you
256   // can write arbitrary bytes to the output. Process them if any.
257   for (BaseCommand *Base : SectionCommands)
258     if (auto *Data = dyn_cast<ByteCommand>(Base))
259       writeInt(Buf + Data->Offset, Data->Expression().getValue(), Data->Size);
260 }
261 
262 static void finalizeShtGroup(OutputSection *OS,
263                              InputSection *Section) {
264   assert(Config->Relocatable);
265 
266   // sh_link field for SHT_GROUP sections should contain the section index of
267   // the symbol table.
268   OS->Link = In.SymTab->getParent()->SectionIndex;
269 
270   // sh_info then contain index of an entry in symbol table section which
271   // provides signature of the section group.
272   ArrayRef<Symbol *> Symbols = Section->File->getSymbols();
273   OS->Info = In.SymTab->getSymbolIndex(Symbols[Section->Info]);
274 }
275 
276 void OutputSection::finalize() {
277   std::vector<InputSection *> V = getInputSections(this);
278   InputSection *First = V.empty() ? nullptr : V[0];
279 
280   if (Flags & SHF_LINK_ORDER) {
281     // We must preserve the link order dependency of sections with the
282     // SHF_LINK_ORDER flag. The dependency is indicated by the sh_link field. We
283     // need to translate the InputSection sh_link to the OutputSection sh_link,
284     // all InputSections in the OutputSection have the same dependency.
285     if (auto *EX = dyn_cast<ARMExidxSyntheticSection>(First))
286       Link = EX->getLinkOrderDep()->getParent()->SectionIndex;
287     else if (auto *D = First->getLinkOrderDep())
288       Link = D->getParent()->SectionIndex;
289   }
290 
291   if (Type == SHT_GROUP) {
292     finalizeShtGroup(this, First);
293     return;
294   }
295 
296   if (!Config->CopyRelocs || (Type != SHT_RELA && Type != SHT_REL))
297     return;
298 
299   if (isa<SyntheticSection>(First))
300     return;
301 
302   Link = In.SymTab->getParent()->SectionIndex;
303   // sh_info for SHT_REL[A] sections should contain the section header index of
304   // the section to which the relocation applies.
305   InputSectionBase *S = First->getRelocatedSection();
306   Info = S->getOutputSection()->SectionIndex;
307   Flags |= SHF_INFO_LINK;
308 }
309 
310 // Returns true if S matches /Filename.?\.o$/.
311 static bool isCrtBeginEnd(StringRef S, StringRef Filename) {
312   if (!S.endswith(".o"))
313     return false;
314   S = S.drop_back(2);
315   if (S.endswith(Filename))
316     return true;
317   return !S.empty() && S.drop_back().endswith(Filename);
318 }
319 
320 static bool isCrtbegin(StringRef S) { return isCrtBeginEnd(S, "crtbegin"); }
321 static bool isCrtend(StringRef S) { return isCrtBeginEnd(S, "crtend"); }
322 
323 // .ctors and .dtors are sorted by this priority from highest to lowest.
324 //
325 //  1. The section was contained in crtbegin (crtbegin contains
326 //     some sentinel value in its .ctors and .dtors so that the runtime
327 //     can find the beginning of the sections.)
328 //
329 //  2. The section has an optional priority value in the form of ".ctors.N"
330 //     or ".dtors.N" where N is a number. Unlike .{init,fini}_array,
331 //     they are compared as string rather than number.
332 //
333 //  3. The section is just ".ctors" or ".dtors".
334 //
335 //  4. The section was contained in crtend, which contains an end marker.
336 //
337 // In an ideal world, we don't need this function because .init_array and
338 // .ctors are duplicate features (and .init_array is newer.) However, there
339 // are too many real-world use cases of .ctors, so we had no choice to
340 // support that with this rather ad-hoc semantics.
341 static bool compCtors(const InputSection *A, const InputSection *B) {
342   bool BeginA = isCrtbegin(A->File->getName());
343   bool BeginB = isCrtbegin(B->File->getName());
344   if (BeginA != BeginB)
345     return BeginA;
346   bool EndA = isCrtend(A->File->getName());
347   bool EndB = isCrtend(B->File->getName());
348   if (EndA != EndB)
349     return EndB;
350   StringRef X = A->Name;
351   StringRef Y = B->Name;
352   assert(X.startswith(".ctors") || X.startswith(".dtors"));
353   assert(Y.startswith(".ctors") || Y.startswith(".dtors"));
354   X = X.substr(6);
355   Y = Y.substr(6);
356   return X < Y;
357 }
358 
359 // Sorts input sections by the special rules for .ctors and .dtors.
360 // Unfortunately, the rules are different from the one for .{init,fini}_array.
361 // Read the comment above.
362 void OutputSection::sortCtorsDtors() {
363   assert(SectionCommands.size() == 1);
364   auto *ISD = cast<InputSectionDescription>(SectionCommands[0]);
365   llvm::stable_sort(ISD->Sections, compCtors);
366 }
367 
368 // If an input string is in the form of "foo.N" where N is a number,
369 // return N. Otherwise, returns 65536, which is one greater than the
370 // lowest priority.
371 int elf::getPriority(StringRef S) {
372   size_t Pos = S.rfind('.');
373   if (Pos == StringRef::npos)
374     return 65536;
375   int V;
376   if (!to_integer(S.substr(Pos + 1), V, 10))
377     return 65536;
378   return V;
379 }
380 
381 std::vector<InputSection *> elf::getInputSections(OutputSection *OS) {
382   std::vector<InputSection *> Ret;
383   for (BaseCommand *Base : OS->SectionCommands)
384     if (auto *ISD = dyn_cast<InputSectionDescription>(Base))
385       Ret.insert(Ret.end(), ISD->Sections.begin(), ISD->Sections.end());
386   return Ret;
387 }
388 
389 // Sorts input sections by section name suffixes, so that .foo.N comes
390 // before .foo.M if N < M. Used to sort .{init,fini}_array.N sections.
391 // We want to keep the original order if the priorities are the same
392 // because the compiler keeps the original initialization order in a
393 // translation unit and we need to respect that.
394 // For more detail, read the section of the GCC's manual about init_priority.
395 void OutputSection::sortInitFini() {
396   // Sort sections by priority.
397   sort([](InputSectionBase *S) { return getPriority(S->Name); });
398 }
399 
400 std::array<uint8_t, 4> OutputSection::getFiller() {
401   if (Filler)
402     return *Filler;
403   if (Flags & SHF_EXECINSTR)
404     return Target->TrapInstr;
405   return {0, 0, 0, 0};
406 }
407 
408 template void OutputSection::writeHeaderTo<ELF32LE>(ELF32LE::Shdr *Shdr);
409 template void OutputSection::writeHeaderTo<ELF32BE>(ELF32BE::Shdr *Shdr);
410 template void OutputSection::writeHeaderTo<ELF64LE>(ELF64LE::Shdr *Shdr);
411 template void OutputSection::writeHeaderTo<ELF64BE>(ELF64BE::Shdr *Shdr);
412 
413 template void OutputSection::writeTo<ELF32LE>(uint8_t *Buf);
414 template void OutputSection::writeTo<ELF32BE>(uint8_t *Buf);
415 template void OutputSection::writeTo<ELF64LE>(uint8_t *Buf);
416 template void OutputSection::writeTo<ELF64BE>(uint8_t *Buf);
417 
418 template void OutputSection::maybeCompress<ELF32LE>();
419 template void OutputSection::maybeCompress<ELF32BE>();
420 template void OutputSection::maybeCompress<ELF64LE>();
421 template void OutputSection::maybeCompress<ELF64BE>();
422