symbolic_debuginfo/
dwarf.rs

1//! Support for DWARF debugging information, common to ELF and MachO.
2//! In rare cases, PE's may contain it as well.
3//!
4//! The central element of this module is the [`Dwarf`] trait, which is implemented by [`ElfObject`],
5//! [`MachObject`] and [`PeObject`]. The dwarf debug session object can be obtained via getters on those types.
6//!
7//! [`Dwarf`]: trait.Dwarf.html
8//! [`ElfObject`]: ../elf/struct.ElfObject.html
9//! [`MachObject`]: ../macho/struct.MachObject.html
10//! [`PeObject`]: ../pe/struct.PeObject.html
11
12use std::borrow::Cow;
13use std::collections::BTreeSet;
14use std::error::Error;
15use std::fmt;
16use std::marker::PhantomData;
17use std::ops::Deref;
18use std::sync::Arc;
19
20use fallible_iterator::FallibleIterator;
21use gimli::read::{AttributeValue, Error as GimliError, Range};
22use gimli::{constants, AbbreviationsCacheStrategy, DwarfFileType, UnitSectionOffset};
23use once_cell::sync::OnceCell;
24use thiserror::Error;
25
26use symbolic_common::{AsSelf, Language, Name, NameMangling, SelfCell};
27
28use crate::base::*;
29use crate::function_builder::FunctionBuilder;
30#[cfg(feature = "macho")]
31use crate::macho::BcSymbolMap;
32use crate::sourcebundle::SourceFileDescriptor;
33
34/// This is a fake BcSymbolMap used when macho support is turned off since they are unfortunately
35/// part of the dwarf interface
36#[cfg(not(feature = "macho"))]
37#[derive(Debug)]
38pub struct BcSymbolMap<'d> {
39    _marker: std::marker::PhantomData<&'d str>,
40}
41
42#[cfg(not(feature = "macho"))]
43impl<'d> BcSymbolMap<'d> {
44    pub(crate) fn resolve_opt(&self, _name: impl AsRef<[u8]>) -> Option<&str> {
45        None
46    }
47}
48
49#[doc(hidden)]
50pub use gimli;
51pub use gimli::RunTimeEndian as Endian;
52
53type Slice<'a> = gimli::read::EndianSlice<'a, Endian>;
54type RangeLists<'a> = gimli::read::RangeLists<Slice<'a>>;
55type Unit<'a> = gimli::read::Unit<Slice<'a>>;
56type DwarfInner<'a> = gimli::read::Dwarf<Slice<'a>>;
57
58type Die<'d, 'u> = gimli::read::DebuggingInformationEntry<'u, 'u, Slice<'d>, usize>;
59type Attribute<'a> = gimli::read::Attribute<Slice<'a>>;
60type UnitOffset = gimli::read::UnitOffset<usize>;
61type DebugInfoOffset = gimli::DebugInfoOffset<usize>;
62type EntriesRaw<'d, 'u> = gimli::EntriesRaw<'u, 'u, Slice<'d>>;
63
64type UnitHeader<'a> = gimli::read::UnitHeader<Slice<'a>>;
65type IncompleteLineNumberProgram<'a> = gimli::read::IncompleteLineProgram<Slice<'a>>;
66type LineNumberProgramHeader<'a> = gimli::read::LineProgramHeader<Slice<'a>>;
67type LineProgramFileEntry<'a> = gimli::read::FileEntry<Slice<'a>>;
68
69/// This applies the offset to the address.
70///
71/// This function does not panic but would wrap around if too large or small
72/// numbers are passed.
73fn offset(addr: u64, offset: i64) -> u64 {
74    (addr as i64).wrapping_sub(offset) as u64
75}
76
77/// The error type for [`DwarfError`].
78#[non_exhaustive]
79#[derive(Clone, Copy, Debug, PartialEq, Eq)]
80pub enum DwarfErrorKind {
81    /// A compilation unit referenced by index does not exist.
82    InvalidUnitRef(usize),
83
84    /// A file record referenced by index does not exist.
85    InvalidFileRef(u64),
86
87    /// An inline record was encountered without an inlining parent.
88    UnexpectedInline,
89
90    /// The debug_ranges of a function are invalid.
91    InvertedFunctionRange,
92
93    /// The DWARF file is corrupted. See the cause for more information.
94    CorruptedData,
95}
96
97impl fmt::Display for DwarfErrorKind {
98    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
99        match self {
100            Self::InvalidUnitRef(offset) => {
101                write!(f, "compilation unit for offset {offset} does not exist")
102            }
103            Self::InvalidFileRef(id) => write!(f, "referenced file {id} does not exist"),
104            Self::UnexpectedInline => write!(f, "unexpected inline function without parent"),
105            Self::InvertedFunctionRange => write!(f, "function with inverted address range"),
106            Self::CorruptedData => write!(f, "corrupted dwarf debug data"),
107        }
108    }
109}
110
111/// An error handling [`DWARF`](trait.Dwarf.html) debugging information.
112#[derive(Debug, Error)]
113#[error("{kind}")]
114pub struct DwarfError {
115    kind: DwarfErrorKind,
116    #[source]
117    source: Option<Box<dyn Error + Send + Sync + 'static>>,
118}
119
120impl DwarfError {
121    /// Creates a new DWARF error from a known kind of error as well as an arbitrary error
122    /// payload.
123    fn new<E>(kind: DwarfErrorKind, source: E) -> Self
124    where
125        E: Into<Box<dyn Error + Send + Sync>>,
126    {
127        let source = Some(source.into());
128        Self { kind, source }
129    }
130
131    /// Returns the corresponding [`DwarfErrorKind`] for this error.
132    pub fn kind(&self) -> DwarfErrorKind {
133        self.kind
134    }
135}
136
137impl From<DwarfErrorKind> for DwarfError {
138    fn from(kind: DwarfErrorKind) -> Self {
139        Self { kind, source: None }
140    }
141}
142
143impl From<GimliError> for DwarfError {
144    fn from(e: GimliError) -> Self {
145        Self::new(DwarfErrorKind::CorruptedData, e)
146    }
147}
148
149/// DWARF section information including its data.
150///
151/// This is returned from objects implementing the [`Dwarf`] trait.
152///
153/// [`Dwarf`]: trait.Dwarf.html
154#[derive(Clone)]
155pub struct DwarfSection<'data> {
156    /// Memory address of this section in virtual memory.
157    pub address: u64,
158
159    /// File offset of this section.
160    pub offset: u64,
161
162    /// Section address alignment (power of two).
163    pub align: u64,
164
165    /// Binary data of this section.
166    pub data: Cow<'data, [u8]>,
167}
168
169impl fmt::Debug for DwarfSection<'_> {
170    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
171        f.debug_struct("DwarfSection")
172            .field("address", &format_args!("{:#x}", self.address))
173            .field("offset", &format_args!("{:#x}", self.offset))
174            .field("align", &format_args!("{:#x}", self.align))
175            .field("len()", &self.data.len())
176            .finish()
177    }
178}
179
180/// Provides access to DWARF debugging information independent of the container file type.
181///
182/// When implementing this trait, verify whether the container file type supports compressed section
183/// data. If so, override the provided `section_data` method. Also, if there is a faster way to
184/// check for the existence of a section without loading its data, override `has_section`.
185pub trait Dwarf<'data> {
186    /// Returns whether the file was compiled for a big-endian or little-endian machine.
187    ///
188    /// This can usually be determined by inspecting the file's headers. Sometimes, this is also
189    /// given by the architecture.
190    fn endianity(&self) -> Endian;
191
192    /// Returns information and raw data of a section.
193    ///
194    /// The section name is given without leading punctuation, such dots or underscores. For
195    /// instance, the name of the Debug Info section would be `"debug_info"`, which translates to
196    /// `".debug_info"` in ELF and `"__debug_info"` in MachO.
197    ///
198    /// Certain containers might allow compressing section data. In this case, this function returns
199    /// the compressed data. To get uncompressed data instead, use `section_data`.
200    fn raw_section(&self, name: &str) -> Option<DwarfSection<'data>>;
201
202    /// Returns information and data of a section.
203    ///
204    /// If the section is compressed, this decompresses on the fly and returns allocated memory.
205    /// Otherwise, this should return a slice of the raw data.
206    ///
207    /// The section name is given without leading punctuation, such dots or underscores. For
208    /// instance, the name of the Debug Info section would be `"debug_info"`, which translates to
209    /// `".debug_info"` in ELF and `"__debug_info"` in MachO.
210    fn section(&self, name: &str) -> Option<DwarfSection<'data>> {
211        self.raw_section(name)
212    }
213
214    /// Determines whether the specified section exists.
215    ///
216    /// The section name is given without leading punctuation, such dots or underscores. For
217    /// instance, the name of the Debug Info section would be `"debug_info"`, which translates to
218    /// `".debug_info"` in ELF and `"__debug_info"` in MachO.
219    fn has_section(&self, name: &str) -> bool {
220        self.raw_section(name).is_some()
221    }
222}
223
224/// A row in the DWARF line program.
225#[derive(Debug)]
226struct DwarfRow {
227    address: u64,
228    file_index: u64,
229    line: Option<u64>,
230    size: Option<u64>,
231}
232
233/// A sequence in the DWARF line program.
234#[derive(Debug)]
235struct DwarfSequence {
236    start: u64,
237    end: u64,
238    rows: Vec<DwarfRow>,
239}
240
241/// Helper that prepares a DwarfLineProgram for more efficient access.
242#[derive(Debug)]
243struct DwarfLineProgram<'d> {
244    header: LineNumberProgramHeader<'d>,
245    sequences: Vec<DwarfSequence>,
246}
247
248impl<'d> DwarfLineProgram<'d> {
249    fn prepare(program: IncompleteLineNumberProgram<'d>) -> Self {
250        let mut sequences = Vec::new();
251        let mut sequence_rows = Vec::<DwarfRow>::new();
252        let mut prev_address = 0;
253        let mut state_machine = program.rows();
254
255        while let Ok(Some((_, &program_row))) = state_machine.next_row() {
256            let address = program_row.address();
257
258            // we have seen rustc emit for WASM targets a bad sequence that spans from 0 to
259            // the end of the program.  https://github.com/rust-lang/rust/issues/79410
260            // We want to skip these bad sequences. Unfortunately, code in .o files can legitimately
261            // be located at address 0, so we incorrectly skip line sequences in that case, too.
262            // See https://github.com/getsentry/symbolic/issues/471 .
263            if address == 0 {
264                continue;
265            }
266
267            if let Some(last_row) = sequence_rows.last_mut() {
268                if address >= last_row.address {
269                    last_row.size = Some(address - last_row.address);
270                }
271            }
272
273            if program_row.end_sequence() {
274                // Theoretically, there could be multiple DW_LNE_end_sequence in a row. We're not
275                // interested in empty sequences, so we can skip them completely.
276                if !sequence_rows.is_empty() {
277                    sequences.push(DwarfSequence {
278                        start: sequence_rows[0].address,
279                        // Take a defensive approach and ensure that `high_address` always covers
280                        // the last encountered row, assuming a 1 byte instruction.
281                        end: if address < prev_address {
282                            prev_address + 1
283                        } else {
284                            address
285                        },
286                        rows: std::mem::take(&mut sequence_rows),
287                    });
288                }
289                prev_address = 0;
290            } else if address < prev_address {
291                // The standard says:
292                // "Within a sequence, addresses and operation pointers may only increase."
293                // So this row is invalid, we can ignore it.
294                //
295                // If we wanted to handle this, we could start a new sequence
296                // here, but let's wait until that is needed.
297            } else {
298                let file_index = program_row.file_index();
299                let line = program_row.line().map(|v| v.get());
300                let mut duplicate = false;
301                if let Some(last_row) = sequence_rows.last_mut() {
302                    if last_row.address == address {
303                        last_row.file_index = file_index;
304                        last_row.line = line;
305                        duplicate = true;
306                    }
307                }
308                if !duplicate {
309                    sequence_rows.push(DwarfRow {
310                        address,
311                        file_index,
312                        line,
313                        size: None,
314                    });
315                }
316                prev_address = address;
317            }
318        }
319
320        if !sequence_rows.is_empty() {
321            // A sequence without an end_sequence row.
322            // Let's assume the last row covered 1 byte.
323            let start = sequence_rows[0].address;
324            let end = prev_address + 1;
325            sequences.push(DwarfSequence {
326                start,
327                end,
328                rows: sequence_rows,
329            });
330        }
331
332        // Sequences are not guaranteed to be in order.
333        sequences.sort_by_key(|x| x.start);
334
335        DwarfLineProgram {
336            header: state_machine.header().clone(),
337            sequences,
338        }
339    }
340
341    pub fn get_rows(&self, range: &Range) -> &[DwarfRow] {
342        for seq in &self.sequences {
343            if seq.end <= range.begin || seq.start > range.end {
344                continue;
345            }
346
347            let from = match seq.rows.binary_search_by_key(&range.begin, |x| x.address) {
348                Ok(idx) => idx,
349                Err(0) => continue,
350                Err(next_idx) => next_idx - 1,
351            };
352
353            let len = seq.rows[from..]
354                .binary_search_by_key(&range.end, |x| x.address)
355                .unwrap_or_else(|e| e);
356            return &seq.rows[from..from + len];
357        }
358        &[]
359    }
360}
361
362/// A slim wrapper around a DWARF unit.
363#[derive(Clone, Copy, Debug)]
364struct UnitRef<'d, 'a> {
365    info: &'a DwarfInfo<'d>,
366    unit: &'a Unit<'d>,
367}
368
369impl<'d> UnitRef<'d, '_> {
370    /// Resolve the binary value of an attribute.
371    #[inline(always)]
372    fn slice_value(&self, value: AttributeValue<Slice<'d>>) -> Option<&'d [u8]> {
373        self.info
374            .attr_string(self.unit, value)
375            .map(|reader| reader.slice())
376            .ok()
377    }
378
379    /// Resolve the actual string value of an attribute.
380    #[inline(always)]
381    fn string_value(&self, value: AttributeValue<Slice<'d>>) -> Option<Cow<'d, str>> {
382        let slice = self.slice_value(value)?;
383        Some(String::from_utf8_lossy(slice))
384    }
385
386    /// Resolves an entry and if found invokes a function to transform it.
387    ///
388    /// As this might resolve into cached information the data borrowed from
389    /// abbrev can only be temporarily accessed in the callback.
390    fn resolve_reference<T, F>(&self, attr: Attribute<'d>, f: F) -> Result<Option<T>, DwarfError>
391    where
392        F: FnOnce(Self, &Die<'d, '_>) -> Result<Option<T>, DwarfError>,
393    {
394        let (unit, offset) = match attr.value() {
395            AttributeValue::UnitRef(offset) => (*self, offset),
396            AttributeValue::DebugInfoRef(offset) => self.info.find_unit_offset(offset)?,
397            // TODO: There is probably more that can come back here.
398            _ => return Ok(None),
399        };
400
401        let mut entries = unit.unit.entries_at_offset(offset)?;
402        entries.next_entry()?;
403
404        if let Some(entry) = entries.current() {
405            f(unit, entry)
406        } else {
407            Ok(None)
408        }
409    }
410
411    /// Returns the offset of this unit within its section.
412    fn offset(&self) -> UnitSectionOffset {
413        self.unit.header.offset()
414    }
415
416    /// Resolves the function name of a debug entry.
417    fn resolve_function_name(
418        &self,
419        entry: &Die<'d, '_>,
420        language: Language,
421        bcsymbolmap: Option<&'d BcSymbolMap<'d>>,
422        prior_offset: Option<UnitOffset>,
423    ) -> Result<Option<Name<'d>>, DwarfError> {
424        let mut attrs = entry.attrs();
425        let mut fallback_name = None;
426        let mut reference_target = None;
427
428        while let Some(attr) = attrs.next()? {
429            match attr.name() {
430                // Prioritize these. If we get them, take them.
431                constants::DW_AT_linkage_name | constants::DW_AT_MIPS_linkage_name => {
432                    return Ok(self
433                        .string_value(attr.value())
434                        .map(|n| resolve_cow_name(bcsymbolmap, n))
435                        .map(|n| Name::new(n, NameMangling::Mangled, language)));
436                }
437                constants::DW_AT_name => {
438                    fallback_name = Some(attr);
439                }
440                constants::DW_AT_abstract_origin | constants::DW_AT_specification => {
441                    reference_target = Some(attr);
442                }
443                _ => {}
444            }
445        }
446
447        if let Some(attr) = fallback_name {
448            return Ok(self
449                .string_value(attr.value())
450                .map(|n| resolve_cow_name(bcsymbolmap, n))
451                .map(|n| Name::new(n, NameMangling::Unmangled, language)));
452        }
453
454        if let Some(attr) = reference_target {
455            return self.resolve_reference(attr, |ref_unit, ref_entry| {
456                // Self-references may have a layer of indircetion. Avoid infinite recursion
457                // in this scenario.
458                if let Some(prior) = prior_offset {
459                    if self.offset() == ref_unit.offset() && prior == ref_entry.offset() {
460                        return Ok(None);
461                    }
462                }
463
464                if self.offset() != ref_unit.offset() || entry.offset() != ref_entry.offset() {
465                    ref_unit.resolve_function_name(
466                        ref_entry,
467                        language,
468                        bcsymbolmap,
469                        Some(entry.offset()),
470                    )
471                } else {
472                    Ok(None)
473                }
474            });
475        }
476
477        Ok(None)
478    }
479}
480
481/// Wrapper around a DWARF Unit.
482#[derive(Debug)]
483struct DwarfUnit<'d, 'a> {
484    inner: UnitRef<'d, 'a>,
485    bcsymbolmap: Option<&'d BcSymbolMap<'d>>,
486    language: Language,
487    line_program: Option<DwarfLineProgram<'d>>,
488    prefer_dwarf_names: bool,
489}
490
491impl<'d, 'a> DwarfUnit<'d, 'a> {
492    /// Creates a DWARF unit from the gimli `Unit` type.
493    fn from_unit(
494        unit: &'a Unit<'d>,
495        info: &'a DwarfInfo<'d>,
496        bcsymbolmap: Option<&'d BcSymbolMap<'d>>,
497    ) -> Result<Option<Self>, DwarfError> {
498        let inner = UnitRef { info, unit };
499        let mut entries = unit.entries();
500        let entry = match entries.next_dfs()? {
501            Some((_, entry)) => entry,
502            None => return Err(gimli::read::Error::MissingUnitDie.into()),
503        };
504
505        // Clang's LLD might eliminate an entire compilation unit and simply set the low_pc to zero
506        // and remove all range entries to indicate that it is missing. Skip such a unit, as it does
507        // not contain any code that can be executed. Special case relocatable objects, as here the
508        // range information has not been written yet and all units look like this.
509        if info.kind != ObjectKind::Relocatable
510            && unit.low_pc == 0
511            && entry.attr(constants::DW_AT_ranges)?.is_none()
512        {
513            return Ok(None);
514        }
515
516        let language = match entry.attr_value(constants::DW_AT_language)? {
517            Some(AttributeValue::Language(lang)) => language_from_dwarf(lang),
518            _ => Language::Unknown,
519        };
520
521        let line_program = unit
522            .line_program
523            .as_ref()
524            .map(|program| DwarfLineProgram::prepare(program.clone()));
525
526        // The value of DW_AT_producer may be an in-place string or a
527        // reference into the debug_str section. We use `string_value`
528        // to resolve it correctly in either case.
529        let producer = entry
530            .attr_value(constants::DW_AT_producer)?
531            .and_then(|av| av.string_value(&info.inner.debug_str));
532
533        // Trust the symbol table more to contain accurate mangled names. However, since Dart's name
534        // mangling is lossy, we need to load the demangled name instead.
535        let prefer_dwarf_names = producer.as_deref() == Some(b"Dart VM");
536
537        Ok(Some(DwarfUnit {
538            inner,
539            bcsymbolmap,
540            language,
541            line_program,
542            prefer_dwarf_names,
543        }))
544    }
545
546    /// The path of the compilation directory. File names are usually relative to this path.
547    fn compilation_dir(&self) -> &'d [u8] {
548        match self.inner.unit.comp_dir {
549            Some(ref dir) => resolve_byte_name(self.bcsymbolmap, dir.slice()),
550            None => &[],
551        }
552    }
553
554    /// Parses the call site and range lists of this Debugging Information Entry.
555    ///
556    /// This method consumes the attributes of the DIE. This means that the `entries` iterator must
557    /// be placed just before the attributes of the DIE. On return, the `entries` iterator is placed
558    /// after the attributes, ready to read the next DIE's abbrev.
559    fn parse_ranges<'r>(
560        &self,
561        entries: &mut EntriesRaw<'d, '_>,
562        abbrev: &gimli::Abbreviation,
563        range_buf: &'r mut Vec<Range>,
564    ) -> Result<(&'r mut Vec<Range>, CallLocation), DwarfError> {
565        range_buf.clear();
566
567        let mut call_line = None;
568        let mut call_file = None;
569        let mut low_pc = None;
570        let mut high_pc = None;
571        let mut high_pc_rel = None;
572
573        let kind = self.inner.info.kind;
574
575        for spec in abbrev.attributes() {
576            let attr = entries.read_attribute(*spec)?;
577            match attr.name() {
578                constants::DW_AT_low_pc => match attr.value() {
579                    AttributeValue::Addr(addr) => low_pc = Some(addr),
580                    AttributeValue::DebugAddrIndex(index) => {
581                        low_pc = Some(self.inner.info.address(self.inner.unit, index)?)
582                    }
583                    _ => return Err(GimliError::UnsupportedAttributeForm.into()),
584                },
585                constants::DW_AT_high_pc => match attr.value() {
586                    AttributeValue::Addr(addr) => high_pc = Some(addr),
587                    AttributeValue::DebugAddrIndex(index) => {
588                        high_pc = Some(self.inner.info.address(self.inner.unit, index)?)
589                    }
590                    AttributeValue::Udata(size) => high_pc_rel = Some(size),
591                    _ => return Err(GimliError::UnsupportedAttributeForm.into()),
592                },
593                constants::DW_AT_call_line => match attr.value() {
594                    AttributeValue::Udata(line) => call_line = Some(line),
595                    _ => return Err(GimliError::UnsupportedAttributeForm.into()),
596                },
597                constants::DW_AT_call_file => match attr.value() {
598                    AttributeValue::FileIndex(file) => call_file = Some(file),
599                    _ => return Err(GimliError::UnsupportedAttributeForm.into()),
600                },
601                constants::DW_AT_ranges
602                | constants::DW_AT_rnglists_base
603                | constants::DW_AT_start_scope => {
604                    match self.inner.info.attr_ranges(self.inner.unit, attr.value())? {
605                        Some(mut ranges) => {
606                            while let Some(range) = match ranges.next() {
607                                Ok(range) => range,
608                                // We have seen broken ranges for some WASM debug files generated by
609                                // emscripten. They mostly manifest themselves in these errors, which
610                                // are triggered by an inverted range (going high to low).
611                                // See a few more examples of broken ranges here:
612                                // https://github.com/emscripten-core/emscripten/issues/15552
613                                Err(gimli::Error::InvalidAddressRange) => None,
614                                Err(err) => {
615                                    return Err(err.into());
616                                }
617                            } {
618                                // A range that begins at 0 indicates code that was eliminated by
619                                // the linker, see below.
620                                if range.begin > 0 || kind == ObjectKind::Relocatable {
621                                    range_buf.push(range);
622                                }
623                            }
624                        }
625                        None => continue,
626                    }
627                }
628                _ => continue,
629            }
630        }
631
632        let call_location = CallLocation {
633            call_file,
634            call_line,
635        };
636
637        if range_buf.is_empty() {
638            if let Some(range) = Self::convert_pc_range(low_pc, high_pc, high_pc_rel, kind)? {
639                range_buf.push(range);
640            }
641        }
642
643        Ok((range_buf, call_location))
644    }
645
646    fn convert_pc_range(
647        low_pc: Option<u64>,
648        high_pc: Option<u64>,
649        high_pc_rel: Option<u64>,
650        kind: ObjectKind,
651    ) -> Result<Option<Range>, DwarfError> {
652        // To go by the logic in dwarf2read, a `low_pc` of 0 can indicate an
653        // eliminated duplicate when the GNU linker is used. In relocatable
654        // objects, all functions are at `0` since they have not been placed
655        // yet, so we want to retain them.
656        let low_pc = match low_pc {
657            Some(low_pc) if low_pc != 0 || kind == ObjectKind::Relocatable => low_pc,
658            _ => return Ok(None),
659        };
660
661        let high_pc = match (high_pc, high_pc_rel) {
662            (Some(high_pc), _) => high_pc,
663            (_, Some(high_pc_rel)) => low_pc.wrapping_add(high_pc_rel),
664            _ => return Ok(None),
665        };
666
667        if low_pc == high_pc {
668            // Most likely low_pc == high_pc means the DIE should be ignored.
669            // https://sourceware.org/ml/gdb-patches/2011-03/msg00739.html
670            return Ok(None);
671        }
672
673        if low_pc == u64::MAX || low_pc == u64::MAX - 1 {
674            // Similarly, u64::MAX/u64::MAX-1 may be used to indicate deleted code.
675            // See https://reviews.llvm.org/D59553
676            return Ok(None);
677        }
678
679        if low_pc > high_pc {
680            return Err(DwarfErrorKind::InvertedFunctionRange.into());
681        }
682
683        Ok(Some(Range {
684            begin: low_pc,
685            end: high_pc,
686        }))
687    }
688
689    /// Resolves file information from a line program.
690    fn file_info(
691        &self,
692        line_program: &LineNumberProgramHeader<'d>,
693        file: &LineProgramFileEntry<'d>,
694    ) -> FileInfo<'d> {
695        FileInfo::new(
696            Cow::Borrowed(resolve_byte_name(
697                self.bcsymbolmap,
698                file.directory(line_program)
699                    .and_then(|attr| self.inner.slice_value(attr))
700                    .unwrap_or_default(),
701            )),
702            Cow::Borrowed(resolve_byte_name(
703                self.bcsymbolmap,
704                self.inner.slice_value(file.path_name()).unwrap_or_default(),
705            )),
706        )
707    }
708
709    /// Resolves a file entry by its index.
710    fn resolve_file(&self, file_id: u64) -> Option<FileInfo<'d>> {
711        let line_program = match self.line_program {
712            Some(ref program) => &program.header,
713            None => return None,
714        };
715
716        line_program
717            .file(file_id)
718            .map(|file| self.file_info(line_program, file))
719    }
720
721    /// Resolves the name of a function from the symbol table.
722    fn resolve_symbol_name(&self, address: u64) -> Option<Name<'d>> {
723        let symbol = self.inner.info.symbol_map.lookup_exact(address)?;
724        let name = resolve_cow_name(self.bcsymbolmap, symbol.name.clone()?);
725        Some(Name::new(name, NameMangling::Mangled, self.language))
726    }
727
728    /// Resolves the name of a function from DWARF debug information.
729    fn resolve_dwarf_name(&self, entry: &Die<'d, '_>) -> Option<Name<'d>> {
730        self.inner
731            .resolve_function_name(entry, self.language, self.bcsymbolmap, None)
732            .ok()
733            .flatten()
734    }
735
736    /// Parses any DW_TAG_subprogram DIEs in the DIE subtree.
737    fn parse_functions(
738        &self,
739        depth: isize,
740        entries: &mut EntriesRaw<'d, '_>,
741        output: &mut FunctionsOutput<'_, 'd>,
742    ) -> Result<(), DwarfError> {
743        while !entries.is_empty() {
744            let dw_die_offset = entries.next_offset();
745            let next_depth = entries.next_depth();
746            if next_depth <= depth {
747                return Ok(());
748            }
749            if let Some(abbrev) = entries.read_abbreviation()? {
750                if abbrev.tag() == constants::DW_TAG_subprogram {
751                    self.parse_function(dw_die_offset, next_depth, entries, abbrev, output)?;
752                } else {
753                    entries.skip_attributes(abbrev.attributes())?;
754                }
755            }
756        }
757        Ok(())
758    }
759
760    /// Parse a single function from a DWARF DIE subtree.
761    ///
762    /// The `entries` iterator must be placed after the abbrev / before the attributes of the
763    /// function DIE.
764    ///
765    /// This method can call itself recursively if another DW_TAG_subprogram entry is encountered
766    /// in the subtree.
767    ///
768    /// On return, the `entries` iterator is placed after the attributes of the last-read DIE.
769    fn parse_function(
770        &self,
771        dw_die_offset: gimli::UnitOffset<usize>,
772        depth: isize,
773        entries: &mut EntriesRaw<'d, '_>,
774        abbrev: &gimli::Abbreviation,
775        output: &mut FunctionsOutput<'_, 'd>,
776    ) -> Result<(), DwarfError> {
777        let (ranges, _) = self.parse_ranges(entries, abbrev, &mut output.range_buf)?;
778
779        let seen_ranges = &mut *output.seen_ranges;
780        ranges.retain(|range| {
781            // Filter out empty and reversed ranges.
782            if range.begin > range.end {
783                return false;
784            }
785
786            // We have seen duplicate top-level function entries being yielded from the
787            // [`DwarfFunctionIterator`], which combined with recursively walking its inlinees can
788            // blow past symcache limits.
789            // We suspect the reason is that the the same top-level functions might be defined in
790            // different compile units. We suspect this might be caused by link-time deduplication
791            // which merges templated code that is being generated multiple times in each
792            // compilation unit. We make sure to detect this here, so we can avoid creating these
793            // duplicates as early as possible.
794            let address = offset(range.begin, self.inner.info.address_offset);
795            let size = range.end - range.begin;
796
797            seen_ranges.insert((address, size))
798        });
799
800        // Ranges can be empty for three reasons: (1) the function is a no-op and does not
801        // contain any code, (2) the function did contain eliminated dead code, or (3) some
802        // tooling created garbage reversed ranges which we filtered out.
803        // In the dead code case, a surrogate DIE remains with `DW_AT_low_pc(0)` and empty ranges.
804        // That DIE might still contain inlined functions with actual ranges - these must be skipped.
805        // However, non-inlined functions may be present in this subtree, so we must still descend
806        // into it.
807        if ranges.is_empty() {
808            return self.parse_functions(depth, entries, output);
809        }
810
811        // Resolve functions in the symbol table first. Only if there is no entry, fall back
812        // to debug information only if there is no match. Sometimes, debug info contains a
813        // lesser quality of symbol names.
814        //
815        // XXX: Maybe we should actually parse the ranges in the resolve function and always
816        // look at the symbol table based on the start of the DIE range.
817        let symbol_name = if self.prefer_dwarf_names {
818            None
819        } else {
820            let first_range_begin = ranges.iter().map(|range| range.begin).min().unwrap();
821            let function_address = offset(first_range_begin, self.inner.info.address_offset);
822            self.resolve_symbol_name(function_address)
823        };
824
825        let name = symbol_name
826            .or_else(|| self.resolve_dwarf_name(&self.inner.unit.entry(dw_die_offset).unwrap()))
827            .unwrap_or_else(|| Name::new("", NameMangling::Unmangled, self.language));
828
829        // Create one function per range. In the common case there is only one range, so
830        // we usually only have one function builder here.
831        let mut builders: Vec<(Range, FunctionBuilder)> = ranges
832            .iter()
833            .map(|range| {
834                let address = offset(range.begin, self.inner.info.address_offset);
835                let size = range.end - range.begin;
836                (
837                    *range,
838                    FunctionBuilder::new(name.clone(), self.compilation_dir(), address, size),
839                )
840            })
841            .collect();
842
843        self.parse_function_children(depth, 0, entries, &mut builders, output)?;
844
845        if let Some(line_program) = &self.line_program {
846            for (range, builder) in &mut builders {
847                for row in line_program.get_rows(range) {
848                    let address = offset(row.address, self.inner.info.address_offset);
849                    let size = row.size;
850                    let file = self.resolve_file(row.file_index).unwrap_or_default();
851                    let line = row.line.unwrap_or(0);
852                    builder.add_leaf_line(address, size, file, line);
853                }
854            }
855        }
856
857        for (_range, builder) in builders {
858            output.functions.push(builder.finish());
859        }
860
861        Ok(())
862    }
863
864    /// Traverses a subtree during function parsing.
865    fn parse_function_children(
866        &self,
867        depth: isize,
868        inline_depth: u32,
869        entries: &mut EntriesRaw<'d, '_>,
870        builders: &mut [(Range, FunctionBuilder<'d>)],
871        output: &mut FunctionsOutput<'_, 'd>,
872    ) -> Result<(), DwarfError> {
873        while !entries.is_empty() {
874            let dw_die_offset = entries.next_offset();
875            let next_depth = entries.next_depth();
876            if next_depth <= depth {
877                return Ok(());
878            }
879            let abbrev = match entries.read_abbreviation()? {
880                Some(abbrev) => abbrev,
881                None => continue,
882            };
883            match abbrev.tag() {
884                constants::DW_TAG_subprogram => {
885                    self.parse_function(dw_die_offset, next_depth, entries, abbrev, output)?;
886                }
887                constants::DW_TAG_inlined_subroutine => {
888                    self.parse_inlinee(
889                        dw_die_offset,
890                        next_depth,
891                        inline_depth,
892                        entries,
893                        abbrev,
894                        builders,
895                        output,
896                    )?;
897                }
898                _ => {
899                    entries.skip_attributes(abbrev.attributes())?;
900                }
901            }
902        }
903        Ok(())
904    }
905
906    /// Recursively parse the inlinees of a function from a DWARF DIE subtree.
907    ///
908    /// The `entries` iterator must be placed just before the attributes of the inline function DIE.
909    ///
910    /// This method calls itself recursively for other DW_TAG_inlined_subroutine entries in the
911    /// subtree. It can also call `parse_function` if a `DW_TAG_subprogram` entry is encountered.
912    ///
913    /// On return, the `entries` iterator is placed after the attributes of the last-read DIE.
914    #[allow(clippy::too_many_arguments)]
915    fn parse_inlinee(
916        &self,
917        dw_die_offset: gimli::UnitOffset<usize>,
918        depth: isize,
919        inline_depth: u32,
920        entries: &mut EntriesRaw<'d, '_>,
921        abbrev: &gimli::Abbreviation,
922        builders: &mut [(Range, FunctionBuilder<'d>)],
923        output: &mut FunctionsOutput<'_, 'd>,
924    ) -> Result<(), DwarfError> {
925        let (ranges, call_location) = self.parse_ranges(entries, abbrev, &mut output.range_buf)?;
926
927        ranges.retain(|range| range.end > range.begin);
928
929        // Ranges can be empty for three reasons: (1) the function is a no-op and does not
930        // contain any code, (2) the function did contain eliminated dead code, or (3) some
931        // tooling created garbage reversed ranges which we filtered out.
932        // In the dead code case, a surrogate DIE remains with `DW_AT_low_pc(0)` and empty ranges.
933        // That DIE might still contain inlined functions with actual ranges - these must be skipped.
934        // However, non-inlined functions may be present in this subtree, so we must still descend
935        // into it.
936        if ranges.is_empty() {
937            return self.parse_functions(depth, entries, output);
938        }
939
940        let name = self
941            .resolve_dwarf_name(&self.inner.unit.entry(dw_die_offset).unwrap())
942            .unwrap_or_else(|| Name::new("", NameMangling::Unmangled, self.language));
943
944        let call_file = call_location
945            .call_file
946            .and_then(|i| self.resolve_file(i))
947            .unwrap_or_default();
948        let call_line = call_location.call_line.unwrap_or(0);
949
950        // Create a separate inlinee for each range.
951        for range in ranges.iter() {
952            // Find the builder for the outer function that covers this range. Usually there's only
953            // one outer range, so only one builder.
954            let builder = match builders.iter_mut().find(|(outer_range, _builder)| {
955                range.begin >= outer_range.begin && range.begin < outer_range.end
956            }) {
957                Some((_outer_range, builder)) => builder,
958                None => continue,
959            };
960
961            let address = offset(range.begin, self.inner.info.address_offset);
962            let size = range.end - range.begin;
963            builder.add_inlinee(
964                inline_depth,
965                name.clone(),
966                address,
967                size,
968                call_file.clone(),
969                call_line,
970            );
971        }
972
973        self.parse_function_children(depth, inline_depth + 1, entries, builders, output)
974    }
975
976    /// Collects all functions within this compilation unit.
977    fn functions(
978        &self,
979        seen_ranges: &mut BTreeSet<(u64, u64)>,
980    ) -> Result<Vec<Function<'d>>, DwarfError> {
981        let mut entries = self.inner.unit.entries_raw(None)?;
982        let mut output = FunctionsOutput::with_seen_ranges(seen_ranges);
983        self.parse_functions(-1, &mut entries, &mut output)?;
984        Ok(output.functions)
985    }
986}
987
988/// The state we pass around during function parsing.
989struct FunctionsOutput<'a, 'd> {
990    /// The list of fully-parsed outer functions. Items are appended whenever we are done
991    /// parsing an entire function.
992    pub functions: Vec<Function<'d>>,
993    /// A scratch buffer which avoids frequent allocations.
994    pub range_buf: Vec<Range>,
995    /// The set of `(address, size)` ranges of the functions we've already parsed.
996    pub seen_ranges: &'a mut BTreeSet<(u64, u64)>,
997}
998
999impl<'a> FunctionsOutput<'a, '_> {
1000    pub fn with_seen_ranges(seen_ranges: &'a mut BTreeSet<(u64, u64)>) -> Self {
1001        Self {
1002            functions: Vec::new(),
1003            range_buf: Vec::new(),
1004            seen_ranges,
1005        }
1006    }
1007}
1008
1009/// For returning (partial) call location information from `parse_ranges`.
1010#[derive(Debug, Default, Clone, Copy)]
1011struct CallLocation {
1012    pub call_file: Option<u64>,
1013    pub call_line: Option<u64>,
1014}
1015
1016/// Converts a DWARF language number into our `Language` type.
1017fn language_from_dwarf(language: gimli::DwLang) -> Language {
1018    match language {
1019        constants::DW_LANG_C => Language::C,
1020        constants::DW_LANG_C11 => Language::C,
1021        constants::DW_LANG_C89 => Language::C,
1022        constants::DW_LANG_C99 => Language::C,
1023        constants::DW_LANG_C_plus_plus => Language::Cpp,
1024        constants::DW_LANG_C_plus_plus_03 => Language::Cpp,
1025        constants::DW_LANG_C_plus_plus_11 => Language::Cpp,
1026        constants::DW_LANG_C_plus_plus_14 => Language::Cpp,
1027        constants::DW_LANG_D => Language::D,
1028        constants::DW_LANG_Go => Language::Go,
1029        constants::DW_LANG_ObjC => Language::ObjC,
1030        constants::DW_LANG_ObjC_plus_plus => Language::ObjCpp,
1031        constants::DW_LANG_Rust => Language::Rust,
1032        constants::DW_LANG_Swift => Language::Swift,
1033        _ => Language::Unknown,
1034    }
1035}
1036
1037/// Data of a specific DWARF section.
1038struct DwarfSectionData<'data, S> {
1039    data: Cow<'data, [u8]>,
1040    endianity: Endian,
1041    _ph: PhantomData<S>,
1042}
1043
1044impl<'data, S> DwarfSectionData<'data, S>
1045where
1046    S: gimli::read::Section<Slice<'data>>,
1047{
1048    /// Loads data for this section from the object file.
1049    fn load<D>(dwarf: &D) -> Self
1050    where
1051        D: Dwarf<'data>,
1052    {
1053        DwarfSectionData {
1054            data: dwarf
1055                .section(&S::section_name()[1..])
1056                .map(|section| section.data)
1057                .unwrap_or_default(),
1058            endianity: dwarf.endianity(),
1059            _ph: PhantomData,
1060        }
1061    }
1062
1063    /// Creates a gimli dwarf section object from the loaded data.
1064    fn to_gimli(&'data self) -> S {
1065        S::from(Slice::new(&self.data, self.endianity))
1066    }
1067}
1068
1069impl<'d, S> fmt::Debug for DwarfSectionData<'d, S>
1070where
1071    S: gimli::read::Section<Slice<'d>>,
1072{
1073    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1074        let owned = match self.data {
1075            Cow::Owned(_) => true,
1076            Cow::Borrowed(_) => false,
1077        };
1078
1079        f.debug_struct("DwarfSectionData")
1080            .field("type", &S::section_name())
1081            .field("endianity", &self.endianity)
1082            .field("len()", &self.data.len())
1083            .field("owned()", &owned)
1084            .finish()
1085    }
1086}
1087
1088/// All DWARF sections that are needed by `DwarfDebugSession`.
1089struct DwarfSections<'data> {
1090    debug_abbrev: DwarfSectionData<'data, gimli::read::DebugAbbrev<Slice<'data>>>,
1091    debug_addr: DwarfSectionData<'data, gimli::read::DebugAddr<Slice<'data>>>,
1092    debug_aranges: DwarfSectionData<'data, gimli::read::DebugAranges<Slice<'data>>>,
1093    debug_info: DwarfSectionData<'data, gimli::read::DebugInfo<Slice<'data>>>,
1094    debug_line: DwarfSectionData<'data, gimli::read::DebugLine<Slice<'data>>>,
1095    debug_line_str: DwarfSectionData<'data, gimli::read::DebugLineStr<Slice<'data>>>,
1096    debug_str: DwarfSectionData<'data, gimli::read::DebugStr<Slice<'data>>>,
1097    debug_str_offsets: DwarfSectionData<'data, gimli::read::DebugStrOffsets<Slice<'data>>>,
1098    debug_ranges: DwarfSectionData<'data, gimli::read::DebugRanges<Slice<'data>>>,
1099    debug_rnglists: DwarfSectionData<'data, gimli::read::DebugRngLists<Slice<'data>>>,
1100    debug_macinfo: DwarfSectionData<'data, gimli::read::DebugMacinfo<Slice<'data>>>,
1101    debug_macro: DwarfSectionData<'data, gimli::read::DebugMacro<Slice<'data>>>,
1102}
1103
1104impl<'data> DwarfSections<'data> {
1105    /// Loads all sections from a DWARF object.
1106    fn from_dwarf<D>(dwarf: &D) -> Self
1107    where
1108        D: Dwarf<'data>,
1109    {
1110        DwarfSections {
1111            debug_abbrev: DwarfSectionData::load(dwarf),
1112            debug_addr: DwarfSectionData::load(dwarf),
1113            debug_aranges: DwarfSectionData::load(dwarf),
1114            debug_info: DwarfSectionData::load(dwarf),
1115            debug_line: DwarfSectionData::load(dwarf),
1116            debug_line_str: DwarfSectionData::load(dwarf),
1117            debug_str: DwarfSectionData::load(dwarf),
1118            debug_str_offsets: DwarfSectionData::load(dwarf),
1119            debug_ranges: DwarfSectionData::load(dwarf),
1120            debug_rnglists: DwarfSectionData::load(dwarf),
1121            debug_macinfo: DwarfSectionData::load(dwarf),
1122            debug_macro: DwarfSectionData::load(dwarf),
1123        }
1124    }
1125}
1126
1127struct DwarfInfo<'data> {
1128    inner: DwarfInner<'data>,
1129    headers: Vec<UnitHeader<'data>>,
1130    units: Vec<OnceCell<Option<Unit<'data>>>>,
1131    symbol_map: SymbolMap<'data>,
1132    address_offset: i64,
1133    kind: ObjectKind,
1134}
1135
1136impl<'d> Deref for DwarfInfo<'d> {
1137    type Target = DwarfInner<'d>;
1138
1139    fn deref(&self) -> &Self::Target {
1140        &self.inner
1141    }
1142}
1143
1144impl<'d> DwarfInfo<'d> {
1145    /// Parses DWARF information from its raw section data.
1146    pub fn parse(
1147        sections: &'d DwarfSections<'d>,
1148        symbol_map: SymbolMap<'d>,
1149        address_offset: i64,
1150        kind: ObjectKind,
1151    ) -> Result<Self, DwarfError> {
1152        let mut inner = gimli::read::Dwarf {
1153            abbreviations_cache: Default::default(),
1154            debug_abbrev: sections.debug_abbrev.to_gimli(),
1155            debug_addr: sections.debug_addr.to_gimli(),
1156            debug_aranges: sections.debug_aranges.to_gimli(),
1157            debug_info: sections.debug_info.to_gimli(),
1158            debug_line: sections.debug_line.to_gimli(),
1159            debug_line_str: sections.debug_line_str.to_gimli(),
1160            debug_str: sections.debug_str.to_gimli(),
1161            debug_str_offsets: sections.debug_str_offsets.to_gimli(),
1162            debug_types: Default::default(),
1163            debug_macinfo: sections.debug_macinfo.to_gimli(),
1164            debug_macro: sections.debug_macro.to_gimli(),
1165            locations: Default::default(),
1166            ranges: RangeLists::new(
1167                sections.debug_ranges.to_gimli(),
1168                sections.debug_rnglists.to_gimli(),
1169            ),
1170            file_type: DwarfFileType::Main,
1171            sup: Default::default(),
1172        };
1173        inner.populate_abbreviations_cache(AbbreviationsCacheStrategy::Duplicates);
1174
1175        // Prepare random access to unit headers.
1176        let headers = inner.units().collect::<Vec<_>>()?;
1177        let units = headers.iter().map(|_| OnceCell::new()).collect();
1178
1179        Ok(DwarfInfo {
1180            inner,
1181            headers,
1182            units,
1183            symbol_map,
1184            address_offset,
1185            kind,
1186        })
1187    }
1188
1189    /// Loads a compilation unit.
1190    fn get_unit(&self, index: usize) -> Result<Option<&Unit<'d>>, DwarfError> {
1191        // Silently ignore unit references out-of-bound
1192        let cell = match self.units.get(index) {
1193            Some(cell) => cell,
1194            None => return Ok(None),
1195        };
1196
1197        let unit_opt = cell.get_or_try_init(|| {
1198            // Parse the compilation unit from the header. This requires a top-level DIE that
1199            // describes the unit itself. For some older DWARF files, this DIE might be missing
1200            // which causes gimli to error out. We prefer to skip them silently as this simply marks
1201            // an empty unit for us.
1202            let header = self.headers[index];
1203            match self.inner.unit(header) {
1204                Ok(unit) => Ok(Some(unit)),
1205                Err(gimli::read::Error::MissingUnitDie) => Ok(None),
1206                Err(error) => Err(DwarfError::from(error)),
1207            }
1208        })?;
1209
1210        Ok(unit_opt.as_ref())
1211    }
1212
1213    /// Resolves an offset into a different compilation unit.
1214    fn find_unit_offset(
1215        &self,
1216        offset: DebugInfoOffset,
1217    ) -> Result<(UnitRef<'d, '_>, UnitOffset), DwarfError> {
1218        let section_offset = UnitSectionOffset::DebugInfoOffset(offset);
1219        let search_result = self
1220            .headers
1221            .binary_search_by_key(&section_offset, UnitHeader::offset);
1222
1223        let index = match search_result {
1224            Ok(index) => index,
1225            Err(0) => return Err(DwarfErrorKind::InvalidUnitRef(offset.0).into()),
1226            Err(next_index) => next_index - 1,
1227        };
1228
1229        if let Some(unit) = self.get_unit(index)? {
1230            if let Some(unit_offset) = section_offset.to_unit_offset(unit) {
1231                return Ok((UnitRef { unit, info: self }, unit_offset));
1232            }
1233        }
1234
1235        Err(DwarfErrorKind::InvalidUnitRef(offset.0).into())
1236    }
1237
1238    /// Returns an iterator over all compilation units.
1239    fn units(&'d self, bcsymbolmap: Option<&'d BcSymbolMap<'d>>) -> DwarfUnitIterator<'d> {
1240        DwarfUnitIterator {
1241            info: self,
1242            bcsymbolmap,
1243            index: 0,
1244        }
1245    }
1246}
1247
1248impl<'slf, 'd: 'slf> AsSelf<'slf> for DwarfInfo<'d> {
1249    type Ref = DwarfInfo<'slf>;
1250
1251    fn as_self(&'slf self) -> &'slf Self::Ref {
1252        unsafe { std::mem::transmute(self) }
1253    }
1254}
1255
1256impl fmt::Debug for DwarfInfo<'_> {
1257    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1258        f.debug_struct("DwarfInfo")
1259            .field("headers", &self.headers)
1260            .field("symbol_map", &self.symbol_map)
1261            .field("address_offset", &self.address_offset)
1262            .finish()
1263    }
1264}
1265
1266/// An iterator over compilation units in a DWARF object.
1267struct DwarfUnitIterator<'s> {
1268    info: &'s DwarfInfo<'s>,
1269    bcsymbolmap: Option<&'s BcSymbolMap<'s>>,
1270    index: usize,
1271}
1272
1273impl<'s> Iterator for DwarfUnitIterator<'s> {
1274    type Item = Result<DwarfUnit<'s, 's>, DwarfError>;
1275
1276    fn next(&mut self) -> Option<Self::Item> {
1277        while self.index < self.info.headers.len() {
1278            let result = self.info.get_unit(self.index);
1279            self.index += 1;
1280
1281            let unit = match result {
1282                Ok(Some(unit)) => unit,
1283                Ok(None) => continue,
1284                Err(error) => return Some(Err(error)),
1285            };
1286
1287            match DwarfUnit::from_unit(unit, self.info, self.bcsymbolmap) {
1288                Ok(Some(unit)) => return Some(Ok(unit)),
1289                Ok(None) => continue,
1290                Err(error) => return Some(Err(error)),
1291            }
1292        }
1293
1294        None
1295    }
1296}
1297
1298impl std::iter::FusedIterator for DwarfUnitIterator<'_> {}
1299
1300/// A debugging session for DWARF debugging information.
1301pub struct DwarfDebugSession<'data> {
1302    cell: SelfCell<Box<DwarfSections<'data>>, DwarfInfo<'data>>,
1303    bcsymbolmap: Option<Arc<BcSymbolMap<'data>>>,
1304}
1305
1306impl<'data> DwarfDebugSession<'data> {
1307    /// Parses a dwarf debugging information from the given DWARF file.
1308    pub fn parse<D>(
1309        dwarf: &D,
1310        symbol_map: SymbolMap<'data>,
1311        address_offset: i64,
1312        kind: ObjectKind,
1313    ) -> Result<Self, DwarfError>
1314    where
1315        D: Dwarf<'data>,
1316    {
1317        let sections = DwarfSections::from_dwarf(dwarf);
1318        let cell = SelfCell::try_new(Box::new(sections), |sections| {
1319            DwarfInfo::parse(unsafe { &*sections }, symbol_map, address_offset, kind)
1320        })?;
1321
1322        Ok(DwarfDebugSession {
1323            cell,
1324            bcsymbolmap: None,
1325        })
1326    }
1327
1328    /// Loads the [`BcSymbolMap`] into this debug session.
1329    ///
1330    /// All the file and function names yielded by this debug session will be resolved using
1331    /// the provided symbol map.
1332    #[cfg(feature = "macho")]
1333    pub(crate) fn load_symbolmap(&mut self, symbolmap: Option<Arc<BcSymbolMap<'data>>>) {
1334        self.bcsymbolmap = symbolmap;
1335    }
1336
1337    /// Returns an iterator over all source files in this debug file.
1338    pub fn files(&self) -> DwarfFileIterator<'_> {
1339        DwarfFileIterator {
1340            units: self.cell.get().units(self.bcsymbolmap.as_deref()),
1341            files: DwarfUnitFileIterator::default(),
1342            finished: false,
1343        }
1344    }
1345
1346    /// Returns an iterator over all functions in this debug file.
1347    pub fn functions(&self) -> DwarfFunctionIterator<'_> {
1348        DwarfFunctionIterator {
1349            units: self.cell.get().units(self.bcsymbolmap.as_deref()),
1350            functions: Vec::new().into_iter(),
1351            seen_ranges: BTreeSet::new(),
1352            finished: false,
1353        }
1354    }
1355
1356    /// See [DebugSession::source_by_path] for more information.
1357    pub fn source_by_path(
1358        &self,
1359        _path: &str,
1360    ) -> Result<Option<SourceFileDescriptor<'_>>, DwarfError> {
1361        Ok(None)
1362    }
1363}
1364
1365impl<'session> DebugSession<'session> for DwarfDebugSession<'_> {
1366    type Error = DwarfError;
1367    type FunctionIterator = DwarfFunctionIterator<'session>;
1368    type FileIterator = DwarfFileIterator<'session>;
1369
1370    fn functions(&'session self) -> Self::FunctionIterator {
1371        self.functions()
1372    }
1373
1374    fn files(&'session self) -> Self::FileIterator {
1375        self.files()
1376    }
1377
1378    fn source_by_path(&self, path: &str) -> Result<Option<SourceFileDescriptor<'_>>, Self::Error> {
1379        self.source_by_path(path)
1380    }
1381}
1382
1383#[derive(Debug, Default)]
1384struct DwarfUnitFileIterator<'s> {
1385    unit: Option<DwarfUnit<'s, 's>>,
1386    index: usize,
1387}
1388
1389impl<'s> Iterator for DwarfUnitFileIterator<'s> {
1390    type Item = FileEntry<'s>;
1391
1392    fn next(&mut self) -> Option<Self::Item> {
1393        let unit = self.unit.as_ref()?;
1394        let line_program = unit.line_program.as_ref().map(|p| &p.header)?;
1395        let file = line_program.file_names().get(self.index)?;
1396
1397        self.index += 1;
1398
1399        Some(FileEntry::new(
1400            Cow::Borrowed(unit.compilation_dir()),
1401            unit.file_info(line_program, file),
1402        ))
1403    }
1404}
1405
1406fn resolve_byte_name<'s>(bcsymbolmap: Option<&'s BcSymbolMap<'s>>, s: &'s [u8]) -> &'s [u8] {
1407    bcsymbolmap
1408        .and_then(|b| b.resolve_opt(s))
1409        .map(AsRef::as_ref)
1410        .unwrap_or(s)
1411}
1412
1413fn resolve_cow_name<'s>(bcsymbolmap: Option<&'s BcSymbolMap<'s>>, s: Cow<'s, str>) -> Cow<'s, str> {
1414    bcsymbolmap
1415        .and_then(|b| b.resolve_opt(s.as_bytes()))
1416        .map(Cow::Borrowed)
1417        .unwrap_or(s)
1418}
1419
1420/// An iterator over source files in a DWARF file.
1421pub struct DwarfFileIterator<'s> {
1422    units: DwarfUnitIterator<'s>,
1423    files: DwarfUnitFileIterator<'s>,
1424    finished: bool,
1425}
1426
1427impl<'s> Iterator for DwarfFileIterator<'s> {
1428    type Item = Result<FileEntry<'s>, DwarfError>;
1429
1430    fn next(&mut self) -> Option<Self::Item> {
1431        if self.finished {
1432            return None;
1433        }
1434
1435        loop {
1436            if let Some(file_entry) = self.files.next() {
1437                return Some(Ok(file_entry));
1438            }
1439
1440            let unit = match self.units.next() {
1441                Some(Ok(unit)) => unit,
1442                Some(Err(error)) => return Some(Err(error)),
1443                None => break,
1444            };
1445
1446            self.files = DwarfUnitFileIterator {
1447                unit: Some(unit),
1448                index: 0,
1449            };
1450        }
1451
1452        self.finished = true;
1453        None
1454    }
1455}
1456
1457/// An iterator over functions in a DWARF file.
1458pub struct DwarfFunctionIterator<'s> {
1459    units: DwarfUnitIterator<'s>,
1460    functions: std::vec::IntoIter<Function<'s>>,
1461    seen_ranges: BTreeSet<(u64, u64)>,
1462    finished: bool,
1463}
1464
1465impl<'s> Iterator for DwarfFunctionIterator<'s> {
1466    type Item = Result<Function<'s>, DwarfError>;
1467
1468    fn next(&mut self) -> Option<Self::Item> {
1469        if self.finished {
1470            return None;
1471        }
1472
1473        loop {
1474            if let Some(func) = self.functions.next() {
1475                return Some(Ok(func));
1476            }
1477
1478            let unit = match self.units.next() {
1479                Some(Ok(unit)) => unit,
1480                Some(Err(error)) => return Some(Err(error)),
1481                None => break,
1482            };
1483
1484            self.functions = match unit.functions(&mut self.seen_ranges) {
1485                Ok(functions) => functions.into_iter(),
1486                Err(error) => return Some(Err(error)),
1487            };
1488        }
1489
1490        self.finished = true;
1491        None
1492    }
1493}
1494
1495impl std::iter::FusedIterator for DwarfFunctionIterator<'_> {}
1496
1497#[cfg(test)]
1498mod tests {
1499    use super::*;
1500
1501    use crate::macho::MachObject;
1502
1503    #[cfg(feature = "macho")]
1504    #[test]
1505    fn test_loads_debug_str_offsets() {
1506        // File generated using dsymutil
1507
1508        let data = std::fs::read("tests/fixtures/helloworld").unwrap();
1509
1510        let obj = MachObject::parse(&data).unwrap();
1511
1512        let sections = DwarfSections::from_dwarf(&obj);
1513        assert_eq!(sections.debug_str_offsets.data.len(), 48);
1514    }
1515}