libdrgn: use libdwfl

libdwfl is the elfutils "DWARF frontend library". It has high-level
functionality for looking up symbols, walking stack traces, etc. In
order to use this functionality, we need to report our debugging
information through libdwfl. For userspace programs, libdwfl has a much
better implementation than drgn for automatically finding debug
information from a core dump or PID. However, for the kernel, libdwfl
has a few issues:

- It only supports finding debug information for the running kernel, not
  vmcores.
- It determines the vmlinux address range by reading /proc/kallsyms,
  which is slow (~70ms on my machine).
- If separate debug information isn't available for a kernel module, it
  finds it by walking /lib/modules/$(uname -r)/kernel; this is repeated
  for every module.
- It doesn't find kernel modules with names containing both dashes and
  underscores (e.g., aes-x86_64).

Luckily, drgn already solved all of these problems, and with some
effort, we can keep doing it ourselves and report it to libdwfl.

The conversion replaces a bunch of code for dealing with userspace core
dump notes, /proc/$pid/maps, and relocations.

libdrgn/dwarf_index.h

@@ -12,14 +12,15 @@
 #ifndef DRGN_DWARF_INDEX_H
 #define DRGN_DWARF_INDEX_H
 
-#include <elfutils/libdw.h>
+#include <elfutils/libdwfl.h>
 #include <libelf.h>
 #include <omp.h>
 #include <stddef.h>
 #include <stdint.h>
 
 #include "drgn.h"
 #include "hash_table.h"
+#include "string_builder.h"
 #include "vector.h"
 
 /**
@@ -44,38 +45,53 @@
  * @{
  */
 
-enum {
-	SECTION_SYMTAB,
-	SECTION_DEBUG_ABBREV,
-	SECTION_DEBUG_INFO,
-	SECTION_DEBUG_LINE,
-	SECTION_DEBUG_STR,
-	DRGN_DWARF_INDEX_NUM_SECTIONS,
-};
-
-struct drgn_dwarf_index_file {
-	Elf_Data *sections[DRGN_DWARF_INDEX_NUM_SECTIONS];
-	bool failed;
+/**
+ * drgn-specific data for the @c Dwfl_Module userdata pointer.
+ *
+ * For a newly created userdata, @c indexed is @c false and @c err is @c NULL.
+ * They are updated by @ref drgn_dwarf_index_update(). @c err may be set with
+ * @ref drgn_dwfl_module_userdata_set_error() before @ref
+ * drgn_dwarf_index_update() is called to skip indexing for that module; the
+ * error message will be added to the @ref DRGN_ERROR_MISSING_DEBUG_INFO error.
+ *
+ * @sa drgn_dwfl_find_elf(), drgn_dwfl_section_address()
+ */
+struct drgn_dwfl_module_userdata {
+	/** Whether the module is indexed in a @ref drgn_dwarf_index. */
+	bool indexed;
+	/** File descriptor of @ref drgn_dwfl_module_userdata::elf. */
 	int fd;
-	/*
-	 * If this is NULL, then we didn't open the file and don't own the Elf
-	 * handle.
-	 */
-	const char *path;
+	/** Error encountered while indexing. */
+	struct drgn_error *err;
+	/** Path of @ref drgn_dwfl_module_userdata::elf. */
+	char *path;
+	/** ELF handle to use. */
 	Elf *elf;
-	Dwarf *dwarf;
-	Elf_Data *rela_sections[DRGN_DWARF_INDEX_NUM_SECTIONS];
-	struct drgn_dwarf_index_file *next;
 };
 
-static inline const char *
-drgn_dwarf_index_file_to_key(struct drgn_dwarf_index_file * const *entry)
+struct drgn_dwfl_module_userdata *drgn_dwfl_module_userdata_create(void);
+
+void
+drgn_dwfl_module_userdata_destroy(struct drgn_dwfl_module_userdata *userdata);
+
+/* This takes ownership of err. */
+void
+drgn_dwfl_module_userdata_set_error(struct drgn_dwfl_module_userdata *userdata,
+				    const char *message,
+				    struct drgn_error *err);
+
+extern const Dwfl_Callbacks drgn_dwfl_callbacks;
+
+/** Get the @ref drgn_dwfl_module_userdata for a @c Dwfl_Module. */
+static inline struct drgn_dwfl_module_userdata *
+drgn_dwfl_module_userdata(Dwfl_Module *module)
 {
-	return (*entry)->path;
+	void **userdatap;
+
+	dwfl_module_info(module, &userdatap, NULL, NULL, NULL, NULL, NULL,
+			 NULL);
+	return *userdatap;
 }
-DEFINE_HASH_TABLE_TYPE(drgn_dwarf_index_file_table,
-		       struct drgn_dwarf_index_file *,
-		       drgn_dwarf_index_file_to_key)
 
 struct drgn_dwarf_index_die;
 DEFINE_HASH_MAP_TYPE(drgn_dwarf_index_die_map, struct string, size_t)
@@ -102,23 +118,14 @@ struct drgn_dwarf_index_shard {
  * files. It is much faster for this task than other generic DWARF parsing
  * libraries.
  *
- * A new DWARF index is created by @ref drgn_dwarf_index_create(). It is freed
- * by @ref drgn_dwarf_index_destroy().
- *
- * Indexing happens in two steps: the files to index are opened using @ref
- * drgn_dwarf_index_open(), then they all are parsed and indexed by @ref
- * drgn_dwarf_index_update(). The update step is parallelized across CPUs, so it
- * is most efficient to open as many files as possible before indexing them all
- * at once in parallel.
- *
  * Searches in the index are done with a @ref drgn_dwarf_index_iterator.
  */
 struct drgn_dwarf_index {
-	/** @privatesection */
-	struct drgn_dwarf_index_file_table files;
-	struct drgn_dwarf_index_file *opened_first, *opened_last;
-	struct drgn_dwarf_index_file *indexed_first, *indexed_last;
-	/* The index is sharded to reduce lock contention. */
+	/**
+	 * Index shards.
+	 *
+	 * This is sharded to reduce lock contention.
+	 */
 	struct drgn_dwarf_index_shard shards[1 << DRGN_DWARF_INDEX_SHARD_BITS];
 };
 
@@ -134,44 +141,36 @@ void drgn_dwarf_index_init(struct drgn_dwarf_index *dindex);
 void drgn_dwarf_index_deinit(struct drgn_dwarf_index *dindex);
 
 /**
- * Open a file and add it to a DWARF index.
+ * Index new DWARF information.
  *
- * This function does the first part of indexing a file: it opens the file,
- * reads or maps it, and checks that it contains the required debugging
- * information. However, it does not actually parse the debugging information.
- * To do so, call drgn_dwarf_index_update() once all of the files to index have
- * been opened.
+ * This parses and indexes the debugging information for all modules in @p dwfl
+ * that have not yet been indexed.
  *
- * If this fails, the file is not opened, but previously opened files are not
- * affected.
+ * On success, @ref drgn_dwfl_module_userdata::indexed is set to @c true for all
+ * modules that we were able to index, and @ref drgn_dwfl_module_userdata::err
+ * is set to non-@c NULL for all other modules.
+ *
+ * If debug information was not available for one or more modules, a @ref
+ * DRGN_ERROR_MISSING_DEBUG_INFO error is returned.
+ *
+ * On any other error, no new debugging information is indexed.
  *
- * @param[in] dindex DWARF index.
- * @param[in] path Path to open.
- * @param[out] elf If not @c NULL, the opened ELF file. It is valid until @ref
- * drgn_dwarf_index_destroy() is called.
  * @return @c NULL on success, non-@c NULL on error.
  */
-struct drgn_error *drgn_dwarf_index_open(struct drgn_dwarf_index *dindex,
-					 const char *path, Elf **elf);
-
-/** Close any files which haven't been indexed yet. */
-void drgn_dwarf_index_close_unindexed(struct drgn_dwarf_index *dindex);
+struct drgn_error *drgn_dwarf_index_update(struct drgn_dwarf_index *dindex,
+					   Dwfl *dwfl);
 
 /**
- * Index newly opened files.
- *
- * This function does the second part of indexing a file: it applies ELF
- * relocations, then parses and indexes the debugging information in all of the
- * files opened by @ref drgn_dwarf_index_open() since the last call to @ref
- * drgn_dwarf_index_update() or @ref drgn_dwarf_index_create().
- *
- * If this fails, no new debugging information is indexed and all opened files
- * which were not already indexed are closed.
+ * Remove all @c Dwfl_Modules that aren't indexed (see @ref
+ * drgn_dwfl_module_userdata::indexed) from @p dwfl.
  *
- * @param[in] dindex DWARF index.
- * @return @c NULL on success, non-@c NULL on error.
+ * This should be called if @ref drgn_dwarf_index_update() returned an error or
+ * if modules were reported and @ref drgn_dwarf_index_update() was not called.
  */
-struct drgn_error *drgn_dwarf_index_update(struct drgn_dwarf_index *dindex);
+void drgn_remove_unindexed_dwfl_modules(Dwfl *dwfl);
+
+/** Remove all @Dwfl_Modules from @p dwfl. */
+void drgn_remove_all_dwfl_modules(Dwfl *dwfl);
 
 /**
  * Iterator over DWARF debugging information.
@@ -215,14 +214,17 @@ void drgn_dwarf_index_iterator_init(struct drgn_dwarf_index_iterator *it,
  *
  * @param[in] it DWARF index iterator.
  * @param[out] die_ret Returned DIE.
+ * @param[out] bias_ret Returned difference between addresses in the loaded
+ * module and addresses in the debugging information. This may be @c NULL if it
+ * is not needed.
  * @return @c NULL on success, non-@c NULL on error. In particular, when there
  * are no more matching DIEs, @p die_ret is not modified and an error with code
  * @ref DRGN_ERROR_STOP is returned; this @ref DRGN_ERROR_STOP error does not
  * have to be passed to @ref drgn_error_destroy().
  */
 struct drgn_error *
 drgn_dwarf_index_iterator_next(struct drgn_dwarf_index_iterator *it,
-			       Dwarf_Die *die_ret);
+			       Dwarf_Die *die_ret, uint64_t *bias_ret);
 
 /** @} */
 

libdrgn/dwarf_info_cache.c

@@ -369,14 +369,14 @@ drgn_dwarf_info_cache_find_complete(struct drgn_dwarf_info_cache *dicache,
 	 * Find a matching DIE. Note that drgn_dwarf_index does not contain DIEs
 	 * with DW_AT_declaration, so this will always be a complete type.
 	 */
-	err = drgn_dwarf_index_iterator_next(&it, &die);
+	err = drgn_dwarf_index_iterator_next(&it, &die, NULL);
 	if (err)
 		return err;
 	/*
 	 * Look for another matching DIE. If there is one, then we can't be sure
 	 * which type this is, so leave it incomplete rather than guessing.
 	 */
-	err = drgn_dwarf_index_iterator_next(&it, &die);
+	err = drgn_dwarf_index_iterator_next(&it, &die, NULL);
 	if (!err)
 		return &drgn_stop;
 	else if (err->code != DRGN_ERROR_STOP)
@@ -1387,7 +1387,7 @@ struct drgn_error *drgn_dwarf_type_find(enum drgn_type_kind kind,
 
 	drgn_dwarf_index_iterator_init(&it, &dicache->dindex, name, name_len,
 				       &tag, 1);
-	while (!(err = drgn_dwarf_index_iterator_next(&it, &die))) {
+	while (!(err = drgn_dwarf_index_iterator_next(&it, &die, NULL))) {
 		if (die_matches_filename(&die, filename)) {
 			err = drgn_type_from_dwarf(dicache, &die, ret);
 			if (err)
@@ -1408,8 +1408,8 @@ struct drgn_error *drgn_dwarf_type_find(enum drgn_type_kind kind,
 
 static struct drgn_error *
 drgn_symbol_from_dwarf_subprogram(struct drgn_dwarf_info_cache *dicache,
-				  Dwarf_Die *die, const char *name,
-				  struct drgn_symbol *ret)
+				  Dwarf_Die *die, uint64_t bias,
+				  const char *name, struct drgn_symbol *ret)
 {
 	struct drgn_error *err;
 	struct drgn_qualified_type qualified_type;
@@ -1427,20 +1427,14 @@ drgn_symbol_from_dwarf_subprogram(struct drgn_dwarf_info_cache *dicache,
 					 "could not find address of '%s'",
 					 name);
 	}
-	ret->address = low_pc;
+	ret->address = low_pc + bias;
 	ret->little_endian = dwarf_die_is_little_endian(die);
-	if (dicache->relocation_hook) {
-		err = dicache->relocation_hook(name, die, ret,
-					       dicache->relocation_arg);
-		if (err)
-			return err;
-	}
 	return NULL;
 }
 
 static struct drgn_error *
 drgn_symbol_from_dwarf_variable(struct drgn_dwarf_info_cache *dicache,
-				Dwarf_Die *die, const char *name,
+				Dwarf_Die *die, uint64_t bias, const char *name,
 				struct drgn_symbol *ret)
 {
 	struct drgn_error *err;
@@ -1470,14 +1464,8 @@ drgn_symbol_from_dwarf_variable(struct drgn_dwarf_info_cache *dicache,
 		return drgn_error_create(DRGN_ERROR_DWARF_FORMAT,
 					 "DW_AT_location has unimplemented operation");
 	}
-	ret->address = loc[0].number;
+	ret->address = loc[0].number + bias;
 	ret->little_endian = dwarf_die_is_little_endian(die);
-	if (dicache->relocation_hook) {
-		err = dicache->relocation_hook(name, die, ret,
-					       dicache->relocation_arg);
-		if (err)
-			return err;
-	}
 	return NULL;
 }
 
@@ -1492,6 +1480,7 @@ drgn_dwarf_symbol_find(const char *name, size_t name_len, const char *filename,
 	size_t num_tags;
 	struct drgn_dwarf_index_iterator it;
 	Dwarf_Die die;
+	uint64_t bias;
 
 	num_tags = 0;
 	if (flags & DRGN_FIND_OBJECT_CONSTANT)
@@ -1503,7 +1492,7 @@ drgn_dwarf_symbol_find(const char *name, size_t name_len, const char *filename,
 
 	drgn_dwarf_index_iterator_init(&it, &dicache->dindex, name,
 				       strlen(name), tags, num_tags);
-	while (!(err = drgn_dwarf_index_iterator_next(&it, &die))) {
+	while (!(err = drgn_dwarf_index_iterator_next(&it, &die, &bias))) {
 		if (!die_matches_filename(&die, filename))
 			continue;
 		switch (dwarf_tag(&die)) {
@@ -1521,10 +1510,11 @@ drgn_dwarf_symbol_find(const char *name, size_t name_len, const char *filename,
 		}
 		case DW_TAG_subprogram:
 			return drgn_symbol_from_dwarf_subprogram(dicache, &die,
-								 name, ret);
+								 bias, name,
+								 ret);
 		case DW_TAG_variable:
 			return drgn_symbol_from_dwarf_variable(dicache, &die,
-							       name, ret);
+							       bias, name, ret);
 		default:
 			DRGN_UNREACHABLE();
 		}
@@ -1549,8 +1539,6 @@ drgn_dwarf_info_cache_create(struct drgn_type_index *tindex,
 	dwarf_type_map_init(&dicache->cant_be_incomplete_array_map);
 	dicache->depth = 0;
 	dicache->tindex = tindex;
-	dicache->relocation_hook = NULL;
-	dicache->relocation_arg = NULL;
 	*ret = dicache;
 	return NULL;
 }

libdrgn/dwarf_info_cache.h

@@ -77,28 +77,6 @@ struct drgn_dwarf_info_cache {
 	int depth;
 	/** Type index. */
 	struct drgn_type_index *tindex;
-	/**
-	 * Relocation callback.
-	 *
-	 * Objects in an ELF file are often relocated when they are loaded into
-	 * a program (e.g., shared libraries or position-independent
-	 * executables). This callback can be used to adjust the address which
-	 * was found in the DWARF debugging information entry for the symbol.
-	 *
-	 * On entry, @p sym is fully initialized and not a constant. This
-	 * should look at @c sym->address and modify it as appropriate.
-	 *
-	 * @param[in] prog @ref drgn_dwarf_symbol_index::prog.
-	 * @param[in] name Name of the symbol.
-	 * @param[in] die DWARF DIE of the symbol.
-	 * @param[in,out] sym Symbol to relocate.
-	 * @return @c NULL on success, non-@c NULL on error.
-	 */
-	struct drgn_error *(*relocation_hook)(const char *name, Dwarf_Die *die,
-					      struct drgn_symbol *sym,
-					      void *arg);
-	/* Argument to pass to @ref drgn_dwarf_info_cache::relocation_hook. */
-	void *relocation_arg;
 };
 
 /** Create a @ref drgn_dwarf_info_cache. */

libdrgn/error.c

@@ -2,7 +2,7 @@
 // SPDX-License-Identifier: GPL-3.0+
 
 #include <errno.h>
-#include <elfutils/libdw.h>
+#include <elfutils/libdwfl.h>
 #include <libelf.h>
 #include <stdarg.h>
 #include <stdio.h>
@@ -22,11 +22,6 @@ struct drgn_error drgn_stop = {
 	.message = "stop iteration",
 };
 
-struct drgn_error drgn_not_elf = {
-	.code = DRGN_ERROR_ELF_FORMAT,
-	.message = "not an ELF file",
-};
-
 static struct drgn_error *drgn_error_create_nodup(enum drgn_error_code code,
 						  char *message)
 {
@@ -194,3 +189,9 @@ struct drgn_error *drgn_error_libdw(void)
 	return drgn_error_format(DRGN_ERROR_DWARF_FORMAT, "libdw error: %s",
 				 dwarf_errmsg(-1));
 }
+
+struct drgn_error *drgn_error_libdwfl(void)
+{
+	return drgn_error_format(DRGN_ERROR_DWARF_FORMAT, "libdwfl error: %s",
+				 dwfl_errmsg(-1));
+}