Auto merge of #114534 - niluxv:strict_prov_unwind, r=cuviper,workingjubilee

Strict provenance unwind

1. Turned many `usize`s in the personality/unwind code that are actually pointers into `*const u8`.
2. Rewrote `read_encoded_pointer` to conform to strict-provenance, along the lines as described by `@eddyb` [in zulip some time ago](https://rust-lang.zulipchat.com/#narrow/stream/136281-t-opsem/topic/strict.20provenance.20in.20dwarf.3A.3Aeh/near/276197290).

This should make supporting CHERI in the future easier (but they use a [slightly modified format in the callsite table](https://cheri-compiler-explorer.cl.cam.ac.uk/z/n6GhhW), which requires a CHERI specific modification to `find_eh_action`).

2 files changed, 94 insertions, 46 deletions

diff --git a/library/std/src/sys/personality/dwarf/eh.rs b/library/std/src/sys/personality/dwarf/eh.rs
index 79624703a4c..a78084de0fa 100644
--- a/library/std/src/sys/personality/dwarf/eh.rs
+++ b/library/std/src/sys/personality/dwarf/eh.rs
@@ -1,6 +1,7 @@
 //! Parsing of GCC-style Language-Specific Data Area (LSDA)
 //! For details see:
 //!  * <https://refspecs.linuxfoundation.org/LSB_3.0.0/LSB-PDA/LSB-PDA/ehframechpt.html>
+//!  * <https://refspecs.linuxfoundation.org/LSB_5.0.0/LSB-Core-generic/LSB-Core-generic/dwarfext.html>
 //!  * <https://itanium-cxx-abi.github.io/cxx-abi/exceptions.pdf>
 //!  * <https://www.airs.com/blog/archives/460>
 //!  * <https://www.airs.com/blog/archives/464>
@@ -37,17 +38,19 @@ pub const DW_EH_PE_indirect: u8 = 0x80;
 
 #[derive(Copy, Clone)]
 pub struct EHContext<'a> {
-    pub ip: usize,                             // Current instruction pointer
-    pub func_start: usize,                     // Address of the current function
-    pub get_text_start: &'a dyn Fn() -> usize, // Get address of the code section
-    pub get_data_start: &'a dyn Fn() -> usize, // Get address of the data section
+    pub ip: *const u8,                             // Current instruction pointer
+    pub func_start: *const u8,                     // Pointer to the current function
+    pub get_text_start: &'a dyn Fn() -> *const u8, // Get pointer to the code section
+    pub get_data_start: &'a dyn Fn() -> *const u8, // Get pointer to the data section
 }
 
+/// Landing pad.
+type LPad = *const u8;
 pub enum EHAction {
     None,
-    Cleanup(usize),
-    Catch(usize),
-    Filter(usize),
+    Cleanup(LPad),
+    Catch(LPad),
+    Filter(LPad),
     Terminate,
 }
 
@@ -81,22 +84,24 @@ pub unsafe fn find_eh_action(lsda: *const u8, context: &EHContext<'_>) -> Result
     let ip = context.ip;
 
     if !USING_SJLJ_EXCEPTIONS {
+        // read the callsite table
         while reader.ptr < action_table {
-            let cs_start = read_encoded_pointer(&mut reader, context, call_site_encoding)?;
-            let cs_len = read_encoded_pointer(&mut reader, context, call_site_encoding)?;
-            let cs_lpad = read_encoded_pointer(&mut reader, context, call_site_encoding)?;
+            // these are offsets rather than pointers;
+            let cs_start = read_encoded_offset(&mut reader, call_site_encoding)?;
+            let cs_len = read_encoded_offset(&mut reader, call_site_encoding)?;
+            let cs_lpad = read_encoded_offset(&mut reader, call_site_encoding)?;
             let cs_action_entry = reader.read_uleb128();
             // Callsite table is sorted by cs_start, so if we've passed the ip, we
             // may stop searching.
-            if ip < func_start + cs_start {
+            if ip < func_start.wrapping_add(cs_start) {
                 break;
             }
-            if ip < func_start + cs_start + cs_len {
+            if ip < func_start.wrapping_add(cs_start + cs_len) {
                 if cs_lpad == 0 {
                     return Ok(EHAction::None);
                 } else {
-                    let lpad = lpad_base + cs_lpad;
-                    return Ok(interpret_cs_action(action_table as *mut u8, cs_action_entry, lpad));
+                    let lpad = lpad_base.wrapping_add(cs_lpad);
+                    return Ok(interpret_cs_action(action_table, cs_action_entry, lpad));
                 }
             }
         }
@@ -106,12 +111,12 @@ pub unsafe fn find_eh_action(lsda: *const u8, context: &EHContext<'_>) -> Result
         // SjLj version:
         // The "IP" is an index into the call-site table, with two exceptions:
         // -1 means 'no-action', and 0 means 'terminate'.
-        match ip as isize {
+        match ip.addr() as isize {
             -1 => return Ok(EHAction::None),
             0 => return Ok(EHAction::Terminate),
             _ => (),
         }
-        let mut idx = ip;
+        let mut idx = ip.addr();
         loop {
             let cs_lpad = reader.read_uleb128();
             let cs_action_entry = reader.read_uleb128();
@@ -119,17 +124,18 @@ pub unsafe fn find_eh_action(lsda: *const u8, context: &EHContext<'_>) -> Result
             if idx == 0 {
                 // Can never have null landing pad for sjlj -- that would have
                 // been indicated by a -1 call site index.
-                let lpad = (cs_lpad + 1) as usize;
-                return Ok(interpret_cs_action(action_table as *mut u8, cs_action_entry, lpad));
+                // FIXME(strict provenance)
+                let lpad = ptr::from_exposed_addr((cs_lpad + 1) as usize);
+                return Ok(interpret_cs_action(action_table, cs_action_entry, lpad));
             }
         }
     }
 }
 
 unsafe fn interpret_cs_action(
-    action_table: *mut u8,
+    action_table: *const u8,
     cs_action_entry: u64,
-    lpad: usize,
+    lpad: LPad,
 ) -> EHAction {
     if cs_action_entry == 0 {
         // If cs_action_entry is 0 then this is a cleanup (Drop::drop). We run these
@@ -138,7 +144,7 @@ unsafe fn interpret_cs_action(
     } else {
         // If lpad != 0 and cs_action_entry != 0, we have to check ttype_index.
         // If ttype_index == 0 under the condition, we take cleanup action.
-        let action_record = (action_table as *mut u8).offset(cs_action_entry as isize - 1);
+        let action_record = action_table.offset(cs_action_entry as isize - 1);
         let mut action_reader = DwarfReader::new(action_record);
         let ttype_index = action_reader.read_sleb128();
         if ttype_index == 0 {
@@ -157,22 +163,24 @@ fn round_up(unrounded: usize, align: usize) -> Result<usize, ()> {
     if align.is_power_of_two() { Ok((unrounded + align - 1) & !(align - 1)) } else { Err(()) }
 }
 
-unsafe fn read_encoded_pointer(
-    reader: &mut DwarfReader,
-    context: &EHContext<'_>,
-    encoding: u8,
-) -> Result<usize, ()> {
-    if encoding == DW_EH_PE_omit {
+/// Read a offset (`usize`) from `reader` whose encoding is described by `encoding`.
+///
+/// `encoding` must be a [DWARF Exception Header Encoding as described by the LSB spec][LSB-dwarf-ext].
+/// In addition the upper ("application") part must be zero.
+///
+/// # Errors
+/// Returns `Err` if `encoding`
+/// * is not a valid DWARF Exception Header Encoding,
+/// * is `DW_EH_PE_omit`, or
+/// * has a non-zero application part.
+///
+/// [LSB-dwarf-ext]: https://refspecs.linuxfoundation.org/LSB_5.0.0/LSB-Core-generic/LSB-Core-generic/dwarfext.html
+unsafe fn read_encoded_offset(reader: &mut DwarfReader, encoding: u8) -> Result<usize, ()> {
+    if encoding == DW_EH_PE_omit || encoding & 0xF0 != 0 {
         return Err(());
     }
-
-    // DW_EH_PE_aligned implies it's an absolute pointer value
-    if encoding == DW_EH_PE_aligned {
-        reader.ptr = reader.ptr.with_addr(round_up(reader.ptr.addr(), mem::size_of::<usize>())?);
-        return Ok(reader.read::<usize>());
-    }
-
-    let mut result = match encoding & 0x0F {
+    let result = match encoding & 0x0F {
+        // despite the name, LLVM also uses absptr for offsets instead of pointers
         DW_EH_PE_absptr => reader.read::<usize>(),
         DW_EH_PE_uleb128 => reader.read_uleb128() as usize,
         DW_EH_PE_udata2 => reader.read::<u16>() as usize,
@@ -184,25 +192,66 @@ unsafe fn read_encoded_pointer(
         DW_EH_PE_sdata8 => reader.read::<i64>() as usize,
         _ => return Err(()),
     };
+    Ok(result)
+}
+
+/// Read a pointer from `reader` whose encoding is described by `encoding`.
+///
+/// `encoding` must be a [DWARF Exception Header Encoding as described by the LSB spec][LSB-dwarf-ext].
+///
+/// # Errors
+/// Returns `Err` if `encoding`
+/// * is not a valid DWARF Exception Header Encoding,
+/// * is `DW_EH_PE_omit`, or
+/// * combines `DW_EH_PE_absptr` or `DW_EH_PE_aligned` application part with an integer encoding
+///   (not `DW_EH_PE_absptr`) in the value format part.
+///
+/// [LSB-dwarf-ext]: https://refspecs.linuxfoundation.org/LSB_5.0.0/LSB-Core-generic/LSB-Core-generic/dwarfext.html
+unsafe fn read_encoded_pointer(
+    reader: &mut DwarfReader,
+    context: &EHContext<'_>,
+    encoding: u8,
+) -> Result<*const u8, ()> {
+    if encoding == DW_EH_PE_omit {
+        return Err(());
+    }
 
-    result += match encoding & 0x70 {
-        DW_EH_PE_absptr => 0,
+    let base_ptr = match encoding & 0x70 {
+        DW_EH_PE_absptr => core::ptr::null(),
         // relative to address of the encoded value, despite the name
-        DW_EH_PE_pcrel => reader.ptr.expose_addr(),
+        DW_EH_PE_pcrel => reader.ptr,
         DW_EH_PE_funcrel => {
-            if context.func_start == 0 {
+            if context.func_start.is_null() {
                 return Err(());
             }
             context.func_start
         }
         DW_EH_PE_textrel => (*context.get_text_start)(),
         DW_EH_PE_datarel => (*context.get_data_start)(),
+        // aligned means the value is aligned to the size of a pointer
+        DW_EH_PE_aligned => {
+            reader.ptr =
+                reader.ptr.with_addr(round_up(reader.ptr.addr(), mem::size_of::<*const u8>())?);
+            core::ptr::null()
+        }
         _ => return Err(()),
     };
 
+    let mut ptr = if base_ptr.is_null() {
+        // any value encoding other than absptr would be nonsensical here;
+        // there would be no source of pointer provenance
+        if encoding & 0x0F != DW_EH_PE_absptr {
+            return Err(());
+        }
+        reader.read::<*const u8>()
+    } else {
+        let offset = read_encoded_offset(reader, encoding & 0x0F)?;
+        base_ptr.wrapping_add(offset)
+    };
+
     if encoding & DW_EH_PE_indirect != 0 {
-        result = *ptr::from_exposed_addr::<usize>(result);
+        ptr = *(ptr.cast::<*const u8>());
     }
 
-    Ok(result)
+    Ok(ptr)
 }
diff --git a/library/std/src/sys/personality/gcc.rs b/library/std/src/sys/personality/gcc.rs
index 559d2c7db47..6f317131145 100644
--- a/library/std/src/sys/personality/gcc.rs
+++ b/library/std/src/sys/personality/gcc.rs
@@ -38,7 +38,6 @@
 
 use super::dwarf::eh::{self, EHAction, EHContext};
 use crate::ffi::c_int;
-use libc::uintptr_t;
 use unwind as uw;
 
 // Register ids were lifted from LLVM's TargetLowering::getExceptionPointerRegister()
@@ -160,9 +159,9 @@ cfg_if::cfg_if! {
                         uw::_Unwind_SetGR(
                             context,
                             UNWIND_DATA_REG.0,
-                            exception_object as uintptr_t,
+                            exception_object as uw::_Unwind_Ptr,
                         );
-                        uw::_Unwind_SetGR(context, UNWIND_DATA_REG.1, 0);
+                        uw::_Unwind_SetGR(context, UNWIND_DATA_REG.1, core::ptr::null());
                         uw::_Unwind_SetIP(context, lpad);
                         return uw::_URC_INSTALL_CONTEXT;
                     }
@@ -222,9 +221,9 @@ cfg_if::cfg_if! {
                         uw::_Unwind_SetGR(
                             context,
                             UNWIND_DATA_REG.0,
-                            exception_object as uintptr_t,
+                            exception_object.cast(),
                         );
-                        uw::_Unwind_SetGR(context, UNWIND_DATA_REG.1, 0);
+                        uw::_Unwind_SetGR(context, UNWIND_DATA_REG.1, core::ptr::null());
                         uw::_Unwind_SetIP(context, lpad);
                         uw::_URC_INSTALL_CONTEXT
                     }