github.com/bir3/gocompiler@v0.3.205/src/cmd/link/internal/loadpe/ldpe.go (about) 1 // Copyright 2010 The Go Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 // Package loadpe implements a PE/COFF file reader. 6 package loadpe 7 8 import ( 9 "bytes" 10 "github.com/bir3/gocompiler/src/cmd/internal/bio" 11 "github.com/bir3/gocompiler/src/cmd/internal/objabi" 12 "github.com/bir3/gocompiler/src/cmd/internal/sys" 13 "github.com/bir3/gocompiler/src/cmd/link/internal/loader" 14 "github.com/bir3/gocompiler/src/cmd/link/internal/sym" 15 "debug/pe" 16 "encoding/binary" 17 "errors" 18 "fmt" 19 "io" 20 "strings" 21 ) 22 23 const ( 24 // TODO: the Microsoft doco says IMAGE_SYM_DTYPE_ARRAY is 3 (same with IMAGE_SYM_DTYPE_POINTER and IMAGE_SYM_DTYPE_FUNCTION) 25 IMAGE_SYM_UNDEFINED = 0 26 IMAGE_SYM_ABSOLUTE = -1 27 IMAGE_SYM_DEBUG = -2 28 IMAGE_SYM_TYPE_NULL = 0 29 IMAGE_SYM_TYPE_VOID = 1 30 IMAGE_SYM_TYPE_CHAR = 2 31 IMAGE_SYM_TYPE_SHORT = 3 32 IMAGE_SYM_TYPE_INT = 4 33 IMAGE_SYM_TYPE_LONG = 5 34 IMAGE_SYM_TYPE_FLOAT = 6 35 IMAGE_SYM_TYPE_DOUBLE = 7 36 IMAGE_SYM_TYPE_STRUCT = 8 37 IMAGE_SYM_TYPE_UNION = 9 38 IMAGE_SYM_TYPE_ENUM = 10 39 IMAGE_SYM_TYPE_MOE = 11 40 IMAGE_SYM_TYPE_BYTE = 12 41 IMAGE_SYM_TYPE_WORD = 13 42 IMAGE_SYM_TYPE_UINT = 14 43 IMAGE_SYM_TYPE_DWORD = 15 44 IMAGE_SYM_TYPE_PCODE = 32768 45 IMAGE_SYM_DTYPE_NULL = 0 46 IMAGE_SYM_DTYPE_POINTER = 0x10 47 IMAGE_SYM_DTYPE_FUNCTION = 0x20 48 IMAGE_SYM_DTYPE_ARRAY = 0x30 49 IMAGE_SYM_CLASS_END_OF_FUNCTION = -1 50 IMAGE_SYM_CLASS_NULL = 0 51 IMAGE_SYM_CLASS_AUTOMATIC = 1 52 IMAGE_SYM_CLASS_EXTERNAL = 2 53 IMAGE_SYM_CLASS_STATIC = 3 54 IMAGE_SYM_CLASS_REGISTER = 4 55 IMAGE_SYM_CLASS_EXTERNAL_DEF = 5 56 IMAGE_SYM_CLASS_LABEL = 6 57 IMAGE_SYM_CLASS_UNDEFINED_LABEL = 7 58 IMAGE_SYM_CLASS_MEMBER_OF_STRUCT = 8 59 IMAGE_SYM_CLASS_ARGUMENT = 9 60 IMAGE_SYM_CLASS_STRUCT_TAG = 10 61 IMAGE_SYM_CLASS_MEMBER_OF_UNION = 11 62 IMAGE_SYM_CLASS_UNION_TAG = 12 63 IMAGE_SYM_CLASS_TYPE_DEFINITION = 13 64 IMAGE_SYM_CLASS_UNDEFINED_STATIC = 14 65 IMAGE_SYM_CLASS_ENUM_TAG = 15 66 IMAGE_SYM_CLASS_MEMBER_OF_ENUM = 16 67 IMAGE_SYM_CLASS_REGISTER_PARAM = 17 68 IMAGE_SYM_CLASS_BIT_FIELD = 18 69 IMAGE_SYM_CLASS_FAR_EXTERNAL = 68 /* Not in PECOFF v8 spec */ 70 IMAGE_SYM_CLASS_BLOCK = 100 71 IMAGE_SYM_CLASS_FUNCTION = 101 72 IMAGE_SYM_CLASS_END_OF_STRUCT = 102 73 IMAGE_SYM_CLASS_FILE = 103 74 IMAGE_SYM_CLASS_SECTION = 104 75 IMAGE_SYM_CLASS_WEAK_EXTERNAL = 105 76 IMAGE_SYM_CLASS_CLR_TOKEN = 107 77 IMAGE_REL_I386_ABSOLUTE = 0x0000 78 IMAGE_REL_I386_DIR16 = 0x0001 79 IMAGE_REL_I386_REL16 = 0x0002 80 IMAGE_REL_I386_DIR32 = 0x0006 81 IMAGE_REL_I386_DIR32NB = 0x0007 82 IMAGE_REL_I386_SEG12 = 0x0009 83 IMAGE_REL_I386_SECTION = 0x000A 84 IMAGE_REL_I386_SECREL = 0x000B 85 IMAGE_REL_I386_TOKEN = 0x000C 86 IMAGE_REL_I386_SECREL7 = 0x000D 87 IMAGE_REL_I386_REL32 = 0x0014 88 IMAGE_REL_AMD64_ABSOLUTE = 0x0000 89 IMAGE_REL_AMD64_ADDR64 = 0x0001 90 IMAGE_REL_AMD64_ADDR32 = 0x0002 91 IMAGE_REL_AMD64_ADDR32NB = 0x0003 92 IMAGE_REL_AMD64_REL32 = 0x0004 93 IMAGE_REL_AMD64_REL32_1 = 0x0005 94 IMAGE_REL_AMD64_REL32_2 = 0x0006 95 IMAGE_REL_AMD64_REL32_3 = 0x0007 96 IMAGE_REL_AMD64_REL32_4 = 0x0008 97 IMAGE_REL_AMD64_REL32_5 = 0x0009 98 IMAGE_REL_AMD64_SECTION = 0x000A 99 IMAGE_REL_AMD64_SECREL = 0x000B 100 IMAGE_REL_AMD64_SECREL7 = 0x000C 101 IMAGE_REL_AMD64_TOKEN = 0x000D 102 IMAGE_REL_AMD64_SREL32 = 0x000E 103 IMAGE_REL_AMD64_PAIR = 0x000F 104 IMAGE_REL_AMD64_SSPAN32 = 0x0010 105 IMAGE_REL_ARM_ABSOLUTE = 0x0000 106 IMAGE_REL_ARM_ADDR32 = 0x0001 107 IMAGE_REL_ARM_ADDR32NB = 0x0002 108 IMAGE_REL_ARM_BRANCH24 = 0x0003 109 IMAGE_REL_ARM_BRANCH11 = 0x0004 110 IMAGE_REL_ARM_SECTION = 0x000E 111 IMAGE_REL_ARM_SECREL = 0x000F 112 IMAGE_REL_ARM_MOV32 = 0x0010 113 IMAGE_REL_THUMB_MOV32 = 0x0011 114 IMAGE_REL_THUMB_BRANCH20 = 0x0012 115 IMAGE_REL_THUMB_BRANCH24 = 0x0014 116 IMAGE_REL_THUMB_BLX23 = 0x0015 117 IMAGE_REL_ARM_PAIR = 0x0016 118 IMAGE_REL_ARM64_ABSOLUTE = 0x0000 119 IMAGE_REL_ARM64_ADDR32 = 0x0001 120 IMAGE_REL_ARM64_ADDR32NB = 0x0002 121 IMAGE_REL_ARM64_BRANCH26 = 0x0003 122 IMAGE_REL_ARM64_PAGEBASE_REL21 = 0x0004 123 IMAGE_REL_ARM64_REL21 = 0x0005 124 IMAGE_REL_ARM64_PAGEOFFSET_12A = 0x0006 125 IMAGE_REL_ARM64_PAGEOFFSET_12L = 0x0007 126 IMAGE_REL_ARM64_SECREL = 0x0008 127 IMAGE_REL_ARM64_SECREL_LOW12A = 0x0009 128 IMAGE_REL_ARM64_SECREL_HIGH12A = 0x000A 129 IMAGE_REL_ARM64_SECREL_LOW12L = 0x000B 130 IMAGE_REL_ARM64_TOKEN = 0x000C 131 IMAGE_REL_ARM64_SECTION = 0x000D 132 IMAGE_REL_ARM64_ADDR64 = 0x000E 133 IMAGE_REL_ARM64_BRANCH19 = 0x000F 134 IMAGE_REL_ARM64_BRANCH14 = 0x0010 135 IMAGE_REL_ARM64_REL32 = 0x0011 136 ) 137 138 const ( 139 // When stored into the PLT value for a symbol, this token tells 140 // windynrelocsym to redirect direct references to this symbol to a stub 141 // that loads from the corresponding import symbol and then does 142 // a jump to the loaded value. 143 CreateImportStubPltToken = -2 144 145 // When stored into the GOT value for a import symbol __imp_X this 146 // token tells windynrelocsym to redirect references to the 147 // underlying DYNIMPORT symbol X. 148 RedirectToDynImportGotToken = -2 149 ) 150 151 // TODO(brainman): maybe just add ReadAt method to bio.Reader instead of creating peBiobuf 152 153 // peBiobuf makes bio.Reader look like io.ReaderAt. 154 type peBiobuf bio.Reader 155 156 func (f *peBiobuf) ReadAt(p []byte, off int64) (int, error) { 157 ret := ((*bio.Reader)(f)).MustSeek(off, 0) 158 if ret < 0 { 159 return 0, errors.New("fail to seek") 160 } 161 n, err := f.Read(p) 162 if err != nil { 163 return 0, err 164 } 165 return n, nil 166 } 167 168 // makeUpdater creates a loader.SymbolBuilder if one hasn't been created previously. 169 // We use this to lazily make SymbolBuilders as we don't always need a builder, and creating them for all symbols might be an error. 170 func makeUpdater(l *loader.Loader, bld *loader.SymbolBuilder, s loader.Sym) *loader.SymbolBuilder { 171 if bld != nil { 172 return bld 173 } 174 bld = l.MakeSymbolUpdater(s) 175 return bld 176 } 177 178 // peImportSymsState tracks the set of DLL import symbols we've seen 179 // while reading host objects. We create a singleton instance of this 180 // type, which will persist across multiple host objects. 181 type peImportSymsState struct { 182 183 // Text and non-text sections read in by the host object loader. 184 secSyms []loader.Sym 185 186 // SDYNIMPORT symbols encountered along the way 187 dynimports map[loader.Sym]struct{} 188 189 // Loader and arch, for use in postprocessing. 190 l *loader.Loader 191 arch *sys.Arch 192 } 193 194 var importSymsState *peImportSymsState 195 196 func createImportSymsState(l *loader.Loader, arch *sys.Arch) { 197 if importSymsState != nil { 198 return 199 } 200 importSymsState = &peImportSymsState{ 201 dynimports: make(map[loader.Sym]struct{}), 202 l: l, 203 arch: arch, 204 } 205 } 206 207 // peLoaderState holds various bits of useful state information needed 208 // while loading a single PE object file. 209 type peLoaderState struct { 210 l *loader.Loader 211 arch *sys.Arch 212 f *pe.File 213 pn string 214 sectsyms map[*pe.Section]loader.Sym 215 comdats map[uint16]int64 // key is section index, val is size 216 sectdata map[*pe.Section][]byte 217 localSymVersion int 218 } 219 220 // comdatDefinitions records the names of symbols for which we've 221 // previously seen a definition in COMDAT. Key is symbol name, value 222 // is symbol size (or -1 if we're using the "any" strategy). 223 var comdatDefinitions = make(map[string]int64) 224 225 // Load loads the PE file pn from input. 226 // Symbols from the object file are created via the loader 'l', and 227 // and a slice of the text symbols is returned. 228 // If an .rsrc section or set of .rsrc$xx sections is found, its symbols are 229 // returned as rsrc. 230 func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Reader, pkg string, length int64, pn string) (textp []loader.Sym, rsrc []loader.Sym, err error) { 231 state := &peLoaderState{ 232 l: l, 233 arch: arch, 234 sectsyms: make(map[*pe.Section]loader.Sym), 235 sectdata: make(map[*pe.Section][]byte), 236 localSymVersion: localSymVersion, 237 pn: pn, 238 } 239 createImportSymsState(state.l, state.arch) 240 241 // Some input files are archives containing multiple of 242 // object files, and pe.NewFile seeks to the start of 243 // input file and get confused. Create section reader 244 // to stop pe.NewFile looking before current position. 245 sr := io.NewSectionReader((*peBiobuf)(input), input.Offset(), 1<<63-1) 246 247 // TODO: replace pe.NewFile with pe.Load (grep for "add Load function" in debug/pe for details) 248 f, err := pe.NewFile(sr) 249 if err != nil { 250 return nil, nil, err 251 } 252 defer f.Close() 253 state.f = f 254 255 // TODO return error if found .cormeta 256 257 // create symbols for mapped sections 258 for _, sect := range f.Sections { 259 if sect.Characteristics&pe.IMAGE_SCN_MEM_DISCARDABLE != 0 { 260 continue 261 } 262 263 if sect.Characteristics&(pe.IMAGE_SCN_CNT_CODE|pe.IMAGE_SCN_CNT_INITIALIZED_DATA|pe.IMAGE_SCN_CNT_UNINITIALIZED_DATA) == 0 { 264 // This has been seen for .idata sections, which we 265 // want to ignore. See issues 5106 and 5273. 266 continue 267 } 268 269 name := fmt.Sprintf("%s(%s)", pkg, sect.Name) 270 s := state.l.LookupOrCreateCgoExport(name, localSymVersion) 271 bld := l.MakeSymbolUpdater(s) 272 273 switch sect.Characteristics & (pe.IMAGE_SCN_CNT_UNINITIALIZED_DATA | pe.IMAGE_SCN_CNT_INITIALIZED_DATA | pe.IMAGE_SCN_MEM_READ | pe.IMAGE_SCN_MEM_WRITE | pe.IMAGE_SCN_CNT_CODE | pe.IMAGE_SCN_MEM_EXECUTE) { 274 case pe.IMAGE_SCN_CNT_INITIALIZED_DATA | pe.IMAGE_SCN_MEM_READ: //.rdata 275 bld.SetType(sym.SRODATA) 276 277 case pe.IMAGE_SCN_CNT_UNINITIALIZED_DATA | pe.IMAGE_SCN_MEM_READ | pe.IMAGE_SCN_MEM_WRITE: //.bss 278 bld.SetType(sym.SNOPTRBSS) 279 280 case pe.IMAGE_SCN_CNT_INITIALIZED_DATA | pe.IMAGE_SCN_MEM_READ | pe.IMAGE_SCN_MEM_WRITE: //.data 281 bld.SetType(sym.SNOPTRDATA) 282 283 case pe.IMAGE_SCN_CNT_CODE | pe.IMAGE_SCN_MEM_EXECUTE | pe.IMAGE_SCN_MEM_READ: //.text 284 bld.SetType(sym.STEXT) 285 286 default: 287 return nil, nil, fmt.Errorf("unexpected flags %#06x for PE section %s", sect.Characteristics, sect.Name) 288 } 289 290 if bld.Type() != sym.SNOPTRBSS { 291 data, err := sect.Data() 292 if err != nil { 293 return nil, nil, err 294 } 295 state.sectdata[sect] = data 296 bld.SetData(data) 297 } 298 bld.SetSize(int64(sect.Size)) 299 state.sectsyms[sect] = s 300 if sect.Name == ".rsrc" || strings.HasPrefix(sect.Name, ".rsrc$") { 301 rsrc = append(rsrc, s) 302 } 303 } 304 305 // Make a prepass over the symbols to collect info about COMDAT symbols. 306 if err := state.preprocessSymbols(); err != nil { 307 return nil, nil, err 308 } 309 310 // load relocations 311 for _, rsect := range f.Sections { 312 if _, found := state.sectsyms[rsect]; !found { 313 continue 314 } 315 if rsect.NumberOfRelocations == 0 { 316 continue 317 } 318 if rsect.Characteristics&pe.IMAGE_SCN_MEM_DISCARDABLE != 0 { 319 continue 320 } 321 if rsect.Characteristics&(pe.IMAGE_SCN_CNT_CODE|pe.IMAGE_SCN_CNT_INITIALIZED_DATA|pe.IMAGE_SCN_CNT_UNINITIALIZED_DATA) == 0 { 322 // This has been seen for .idata sections, which we 323 // want to ignore. See issues 5106 and 5273. 324 continue 325 } 326 327 splitResources := strings.HasPrefix(rsect.Name, ".rsrc$") 328 sb := l.MakeSymbolUpdater(state.sectsyms[rsect]) 329 for j, r := range rsect.Relocs { 330 if int(r.SymbolTableIndex) >= len(f.COFFSymbols) { 331 return nil, nil, fmt.Errorf("relocation number %d symbol index idx=%d cannot be large then number of symbols %d", j, r.SymbolTableIndex, len(f.COFFSymbols)) 332 } 333 pesym := &f.COFFSymbols[r.SymbolTableIndex] 334 _, gosym, err := state.readpesym(pesym) 335 if err != nil { 336 return nil, nil, err 337 } 338 if gosym == 0 { 339 name, err := pesym.FullName(f.StringTable) 340 if err != nil { 341 name = string(pesym.Name[:]) 342 } 343 return nil, nil, fmt.Errorf("reloc of invalid sym %s idx=%d type=%d", name, r.SymbolTableIndex, pesym.Type) 344 } 345 346 rSym := gosym 347 rSize := uint8(4) 348 rOff := int32(r.VirtualAddress) 349 var rAdd int64 350 var rType objabi.RelocType 351 switch arch.Family { 352 default: 353 return nil, nil, fmt.Errorf("%s: unsupported arch %v", pn, arch.Family) 354 case sys.I386, sys.AMD64: 355 switch r.Type { 356 default: 357 return nil, nil, fmt.Errorf("%s: %v: unknown relocation type %v", pn, state.sectsyms[rsect], r.Type) 358 359 case IMAGE_REL_I386_REL32, IMAGE_REL_AMD64_REL32, 360 IMAGE_REL_AMD64_ADDR32, // R_X86_64_PC32 361 IMAGE_REL_AMD64_ADDR32NB: 362 rType = objabi.R_PCREL 363 364 rAdd = int64(int32(binary.LittleEndian.Uint32(state.sectdata[rsect][rOff:]))) 365 366 case IMAGE_REL_I386_DIR32NB, IMAGE_REL_I386_DIR32: 367 rType = objabi.R_ADDR 368 369 // load addend from image 370 rAdd = int64(int32(binary.LittleEndian.Uint32(state.sectdata[rsect][rOff:]))) 371 372 case IMAGE_REL_AMD64_ADDR64: // R_X86_64_64 373 rSize = 8 374 375 rType = objabi.R_ADDR 376 377 // load addend from image 378 rAdd = int64(binary.LittleEndian.Uint64(state.sectdata[rsect][rOff:])) 379 } 380 381 case sys.ARM: 382 switch r.Type { 383 default: 384 return nil, nil, fmt.Errorf("%s: %v: unknown ARM relocation type %v", pn, state.sectsyms[rsect], r.Type) 385 386 case IMAGE_REL_ARM_SECREL: 387 rType = objabi.R_PCREL 388 389 rAdd = int64(int32(binary.LittleEndian.Uint32(state.sectdata[rsect][rOff:]))) 390 391 case IMAGE_REL_ARM_ADDR32, IMAGE_REL_ARM_ADDR32NB: 392 rType = objabi.R_ADDR 393 394 rAdd = int64(int32(binary.LittleEndian.Uint32(state.sectdata[rsect][rOff:]))) 395 396 case IMAGE_REL_ARM_BRANCH24: 397 rType = objabi.R_CALLARM 398 399 rAdd = int64(int32(binary.LittleEndian.Uint32(state.sectdata[rsect][rOff:]))) 400 } 401 402 case sys.ARM64: 403 switch r.Type { 404 default: 405 return nil, nil, fmt.Errorf("%s: %v: unknown ARM64 relocation type %v", pn, state.sectsyms[rsect], r.Type) 406 407 case IMAGE_REL_ARM64_ADDR32, IMAGE_REL_ARM64_ADDR32NB: 408 rType = objabi.R_ADDR 409 410 rAdd = int64(int32(binary.LittleEndian.Uint32(state.sectdata[rsect][rOff:]))) 411 } 412 } 413 414 // ld -r could generate multiple section symbols for the 415 // same section but with different values, we have to take 416 // that into account, or in the case of split resources, 417 // the section and its symbols are split into two sections. 418 if issect(pesym) || splitResources { 419 rAdd += int64(pesym.Value) 420 } 421 422 rel, _ := sb.AddRel(rType) 423 rel.SetOff(rOff) 424 rel.SetSiz(rSize) 425 rel.SetSym(rSym) 426 rel.SetAdd(rAdd) 427 } 428 429 sb.SortRelocs() 430 } 431 432 // enter sub-symbols into symbol table. 433 for i, numaux := 0, 0; i < len(f.COFFSymbols); i += numaux + 1 { 434 pesym := &f.COFFSymbols[i] 435 436 numaux = int(pesym.NumberOfAuxSymbols) 437 438 name, err := pesym.FullName(f.StringTable) 439 if err != nil { 440 return nil, nil, err 441 } 442 if name == "" { 443 continue 444 } 445 if issect(pesym) { 446 continue 447 } 448 if int(pesym.SectionNumber) > len(f.Sections) { 449 continue 450 } 451 if pesym.SectionNumber == IMAGE_SYM_DEBUG { 452 continue 453 } 454 if pesym.SectionNumber == IMAGE_SYM_ABSOLUTE && bytes.Equal(pesym.Name[:], []byte("@feat.00")) { 455 // Microsoft's linker looks at whether all input objects have an empty 456 // section called @feat.00. If all of them do, then it enables SEH; 457 // otherwise it doesn't enable that feature. So, since around the Windows 458 // XP SP2 era, most tools that make PE objects just tack on that section, 459 // so that it won't gimp Microsoft's linker logic. Go doesn't support SEH, 460 // so in theory, none of this really matters to us. But actually, if the 461 // linker tries to ingest an object with @feat.00 -- which are produced by 462 // LLVM's resource compiler, for example -- it chokes because of the 463 // IMAGE_SYM_ABSOLUTE section that it doesn't know how to deal with. Since 464 // @feat.00 is just a marking anyway, skip IMAGE_SYM_ABSOLUTE sections that 465 // are called @feat.00. 466 continue 467 } 468 var sect *pe.Section 469 if pesym.SectionNumber > 0 { 470 sect = f.Sections[pesym.SectionNumber-1] 471 if _, found := state.sectsyms[sect]; !found { 472 continue 473 } 474 } 475 476 bld, s, err := state.readpesym(pesym) 477 if err != nil { 478 return nil, nil, err 479 } 480 481 if pesym.SectionNumber == 0 { // extern 482 if l.SymType(s) == sym.SXREF && pesym.Value > 0 { // global data 483 bld = makeUpdater(l, bld, s) 484 bld.SetType(sym.SNOPTRDATA) 485 bld.SetSize(int64(pesym.Value)) 486 } 487 488 continue 489 } else if pesym.SectionNumber > 0 && int(pesym.SectionNumber) <= len(f.Sections) { 490 sect = f.Sections[pesym.SectionNumber-1] 491 if _, found := state.sectsyms[sect]; !found { 492 return nil, nil, fmt.Errorf("%s: %v: missing sect.sym", pn, s) 493 } 494 } else { 495 return nil, nil, fmt.Errorf("%s: %v: sectnum < 0!", pn, s) 496 } 497 498 if sect == nil { 499 return nil, nil, nil 500 } 501 502 // Check for COMDAT symbol. 503 if sz, ok1 := state.comdats[uint16(pesym.SectionNumber-1)]; ok1 { 504 if psz, ok2 := comdatDefinitions[l.SymName(s)]; ok2 { 505 if sz == psz { 506 // OK to discard, we've seen an instance 507 // already. 508 continue 509 } 510 } 511 } 512 if l.OuterSym(s) != 0 { 513 if l.AttrDuplicateOK(s) { 514 continue 515 } 516 outerName := l.SymName(l.OuterSym(s)) 517 sectName := l.SymName(state.sectsyms[sect]) 518 return nil, nil, fmt.Errorf("%s: duplicate symbol reference: %s in both %s and %s", pn, l.SymName(s), outerName, sectName) 519 } 520 521 bld = makeUpdater(l, bld, s) 522 sectsym := state.sectsyms[sect] 523 bld.SetType(l.SymType(sectsym)) 524 l.AddInteriorSym(sectsym, s) 525 bld.SetValue(int64(pesym.Value)) 526 bld.SetSize(4) 527 if l.SymType(sectsym) == sym.STEXT { 528 if bld.External() && !bld.DuplicateOK() { 529 return nil, nil, fmt.Errorf("%s: duplicate symbol definition", l.SymName(s)) 530 } 531 bld.SetExternal(true) 532 } 533 if sz, ok := state.comdats[uint16(pesym.SectionNumber-1)]; ok { 534 // This is a COMDAT definition. Record that we're picking 535 // this instance so that we can ignore future defs. 536 if _, ok := comdatDefinitions[l.SymName(s)]; ok { 537 return nil, nil, fmt.Errorf("internal error: preexisting COMDAT definition for %q", name) 538 } 539 comdatDefinitions[l.SymName(s)] = sz 540 } 541 } 542 543 // Sort outer lists by address, adding to textp. 544 // This keeps textp in increasing address order. 545 for _, sect := range f.Sections { 546 s := state.sectsyms[sect] 547 if s == 0 { 548 continue 549 } 550 l.SortSub(s) 551 importSymsState.secSyms = append(importSymsState.secSyms, s) 552 if l.SymType(s) == sym.STEXT { 553 for ; s != 0; s = l.SubSym(s) { 554 if l.AttrOnList(s) { 555 return nil, nil, fmt.Errorf("symbol %s listed multiple times", l.SymName(s)) 556 } 557 l.SetAttrOnList(s, true) 558 textp = append(textp, s) 559 } 560 } 561 } 562 563 return textp, rsrc, nil 564 } 565 566 // PostProcessImports works to resolve inconsistencies with DLL import 567 // symbols; it is needed when building with more "modern" C compilers 568 // with internal linkage. 569 // 570 // Background: DLL import symbols are data (SNOPTRDATA) symbols whose 571 // name is of the form "__imp_XXX", which contain a pointer/reference 572 // to symbol XXX. It's possible to have import symbols for both data 573 // symbols ("__imp__fmode") and text symbols ("__imp_CreateEventA"). 574 // In some case import symbols are just references to some external 575 // thing, and in other cases we see actual definitions of import 576 // symbols when reading host objects. 577 // 578 // Previous versions of the linker would in most cases immediately 579 // "forward" import symbol references, e.g. treat a references to 580 // "__imp_XXX" a references to "XXX", however this doesn't work well 581 // with more modern compilers, where you can sometimes see import 582 // symbols that are defs (as opposed to external refs). 583 // 584 // The main actions taken below are to search for references to 585 // SDYNIMPORT symbols in host object text/data sections and flag the 586 // symbols for later fixup. When we see a reference to an import 587 // symbol __imp_XYZ where XYZ corresponds to some SDYNIMPORT symbol, 588 // we flag the symbol (via GOT setting) so that it can be redirected 589 // to XYZ later in windynrelocsym. When we see a direct reference to 590 // an SDYNIMPORT symbol XYZ, we also flag the symbol (via PLT setting) 591 // to indicated that the reference will need to be redirected to a 592 // stub. 593 func PostProcessImports() error { 594 ldr := importSymsState.l 595 arch := importSymsState.arch 596 keeprelocneeded := make(map[loader.Sym]loader.Sym) 597 for _, s := range importSymsState.secSyms { 598 isText := ldr.SymType(s) == sym.STEXT 599 relocs := ldr.Relocs(s) 600 for i := 0; i < relocs.Count(); i++ { 601 r := relocs.At(i) 602 rs := r.Sym() 603 if ldr.SymType(rs) == sym.SDYNIMPORT { 604 // Tag the symbol for later stub generation. 605 ldr.SetPlt(rs, CreateImportStubPltToken) 606 continue 607 } 608 isym, err := LookupBaseFromImport(rs, ldr, arch) 609 if err != nil { 610 return err 611 } 612 if isym == 0 { 613 continue 614 } 615 if ldr.SymType(isym) != sym.SDYNIMPORT { 616 continue 617 } 618 // For non-text symbols, forward the reference from __imp_X to 619 // X immediately. 620 if !isText { 621 r.SetSym(isym) 622 continue 623 } 624 // Flag this imp symbol to be processed later in windynrelocsym. 625 ldr.SetGot(rs, RedirectToDynImportGotToken) 626 // Consistency check: should be no PLT token here. 627 splt := ldr.SymPlt(rs) 628 if splt != -1 { 629 return fmt.Errorf("internal error: import symbol %q has invalid PLT setting %d", ldr.SymName(rs), splt) 630 } 631 // Flag for dummy relocation. 632 keeprelocneeded[rs] = isym 633 } 634 } 635 for k, v := range keeprelocneeded { 636 sb := ldr.MakeSymbolUpdater(k) 637 r, _ := sb.AddRel(objabi.R_KEEP) 638 r.SetSym(v) 639 } 640 importSymsState = nil 641 return nil 642 } 643 644 func issect(s *pe.COFFSymbol) bool { 645 return s.StorageClass == IMAGE_SYM_CLASS_STATIC && s.Type == 0 && s.Name[0] == '.' 646 } 647 648 func (state *peLoaderState) readpesym(pesym *pe.COFFSymbol) (*loader.SymbolBuilder, loader.Sym, error) { 649 symname, err := pesym.FullName(state.f.StringTable) 650 if err != nil { 651 return nil, 0, err 652 } 653 var name string 654 if issect(pesym) { 655 name = state.l.SymName(state.sectsyms[state.f.Sections[pesym.SectionNumber-1]]) 656 } else { 657 name = symname 658 // A note on the "_main" exclusion below: the main routine 659 // defined by the Go runtime is named "_main", not "main", so 660 // when reading references to _main from a host object we want 661 // to avoid rewriting "_main" to "main" in this specific 662 // instance. See #issuecomment-1143698749 on #35006 for more 663 // details on this problem. 664 if state.arch.Family == sys.I386 && name[0] == '_' && name != "_main" && !strings.HasPrefix(name, "__imp_") { 665 name = name[1:] // _Name => Name 666 } 667 } 668 669 // remove last @XXX 670 if i := strings.LastIndex(name, "@"); i >= 0 { 671 name = name[:i] 672 } 673 674 var s loader.Sym 675 var bld *loader.SymbolBuilder 676 switch pesym.Type { 677 default: 678 return nil, 0, fmt.Errorf("%s: invalid symbol type %d", symname, pesym.Type) 679 680 case IMAGE_SYM_DTYPE_FUNCTION, IMAGE_SYM_DTYPE_NULL: 681 switch pesym.StorageClass { 682 case IMAGE_SYM_CLASS_EXTERNAL: //global 683 s = state.l.LookupOrCreateCgoExport(name, 0) 684 685 case IMAGE_SYM_CLASS_NULL, IMAGE_SYM_CLASS_STATIC, IMAGE_SYM_CLASS_LABEL: 686 s = state.l.LookupOrCreateCgoExport(name, state.localSymVersion) 687 bld = makeUpdater(state.l, bld, s) 688 bld.SetDuplicateOK(true) 689 690 default: 691 return nil, 0, fmt.Errorf("%s: invalid symbol binding %d", symname, pesym.StorageClass) 692 } 693 } 694 695 if s != 0 && state.l.SymType(s) == 0 && (pesym.StorageClass != IMAGE_SYM_CLASS_STATIC || pesym.Value != 0) { 696 bld = makeUpdater(state.l, bld, s) 697 bld.SetType(sym.SXREF) 698 } 699 700 return bld, s, nil 701 } 702 703 // preprocessSymbols walks the COFF symbols for the PE file we're 704 // reading and looks for cases where we have both a symbol definition 705 // for "XXX" and an "__imp_XXX" symbol, recording these cases in a map 706 // in the state struct. This information will be used in readpesym() 707 // above to give such symbols special treatment. This function also 708 // gathers information about COMDAT sections/symbols for later use 709 // in readpesym(). 710 func (state *peLoaderState) preprocessSymbols() error { 711 712 // Locate comdat sections. 713 state.comdats = make(map[uint16]int64) 714 for i, s := range state.f.Sections { 715 if s.Characteristics&uint32(pe.IMAGE_SCN_LNK_COMDAT) != 0 { 716 state.comdats[uint16(i)] = int64(s.Size) 717 } 718 } 719 720 // Examine symbol defs. 721 for i, numaux := 0, 0; i < len(state.f.COFFSymbols); i += numaux + 1 { 722 pesym := &state.f.COFFSymbols[i] 723 numaux = int(pesym.NumberOfAuxSymbols) 724 if pesym.SectionNumber == 0 { // extern 725 continue 726 } 727 symname, err := pesym.FullName(state.f.StringTable) 728 if err != nil { 729 return err 730 } 731 if _, isc := state.comdats[uint16(pesym.SectionNumber-1)]; !isc { 732 continue 733 } 734 if pesym.StorageClass != uint8(IMAGE_SYM_CLASS_STATIC) { 735 continue 736 } 737 // This symbol corresponds to a COMDAT section. Read the 738 // aux data for it. 739 auxsymp, err := state.f.COFFSymbolReadSectionDefAux(i) 740 if err != nil { 741 return fmt.Errorf("unable to read aux info for section def symbol %d %s: pe.COFFSymbolReadComdatInfo returns %v", i, symname, err) 742 } 743 if auxsymp.Selection == pe.IMAGE_COMDAT_SELECT_SAME_SIZE { 744 // This is supported. 745 } else if auxsymp.Selection == pe.IMAGE_COMDAT_SELECT_ANY { 746 // Also supported. 747 state.comdats[uint16(pesym.SectionNumber-1)] = int64(-1) 748 } else { 749 // We don't support any of the other strategies at the 750 // moment. I suspect that we may need to also support 751 // "associative", we'll see. 752 return fmt.Errorf("internal error: unsupported COMDAT selection strategy found in path=%s sec=%d strategy=%d idx=%d, please file a bug", state.pn, auxsymp.SecNum, auxsymp.Selection, i) 753 } 754 } 755 return nil 756 } 757 758 // LookupBaseFromImport examines the symbol "s" to see if it 759 // corresponds to an import symbol (name of the form "__imp_XYZ") and 760 // if so, it looks up the underlying target of the import symbol and 761 // returns it. An error is returned if the symbol is of the form 762 // "__imp_XYZ" but no XYZ can be found. 763 func LookupBaseFromImport(s loader.Sym, ldr *loader.Loader, arch *sys.Arch) (loader.Sym, error) { 764 sname := ldr.SymName(s) 765 if !strings.HasPrefix(sname, "__imp_") { 766 return 0, nil 767 } 768 basename := sname[len("__imp_"):] 769 if arch.Family == sys.I386 && basename[0] == '_' { 770 basename = basename[1:] // _Name => Name 771 } 772 isym := ldr.Lookup(basename, 0) 773 if isym == 0 { 774 return 0, fmt.Errorf("internal error: import symbol %q with no underlying sym", sname) 775 } 776 return isym, nil 777 }