path: root/src/mongo/util/stacktrace_posix.cpp

/*    Copyright 2009 10gen Inc.
 *
 *    Licensed under the Apache License, Version 2.0 (the "License");
 *    you may not use this file except in compliance with the License.
 *    You may obtain a copy of the License at
 *
 *    http://www.apache.org/licenses/LICENSE-2.0
 *
 *    Unless required by applicable law or agreed to in writing, software
 *    distributed under the License is distributed on an "AS IS" BASIS,
 *    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 *    See the License for the specific language governing permissions and
 *    limitations under the License.
 */

#include "mongo/platform/basic.h"

#include "mongo/util/stacktrace.h"

#include <cstdlib>
#include <dlfcn.h>
#include <iostream>
#include <string>

#include "mongo/base/init.h"
#include "mongo/db/jsobj.h"
#include "mongo/platform/backtrace.h"
#include "mongo/util/hex.h"
#include "mongo/util/log.h"
#include "mongo/util/version.h"

namespace mongo {

namespace {
    /// Maximum number of stack frames to appear in a backtrace.
    const int maxBackTraceFrames = 20;

    /// Optional string containing extra unwinding information.  Should take the form of a
    /// JSON document.
    std::string* soMapJson = NULL;

    /**
     * Returns the "basename" of a path.  The returned StringData is valid until the data referenced
     * by "path" goes out of scope or mutates.
     *
     * E.g., for "/foo/bar/my.txt", returns "my.txt".
     */
    StringData getBaseName(const StringData& path) {
        size_t lastSlash = path.rfind('/');
        if (lastSlash == std::string::npos)
            return path;
        return path.substr(lastSlash + 1);
    }

}  // namespace

    /**
     * Prints a stack backtrace for the current thread to the specified ostream.
     *
     * Does not malloc, does not throw.
     *
     * The format of the backtrace is:
     *
     * ----- BEGIN BACKTRACE -----
     * JSON backtrace
     * Human-readable backtrace
     * -----  END BACKTRACE  -----
     *
     * The JSON backtrace will be a JSON object with a "backtrace" field, and optionally others.
     * The "backtrace" field is an array, whose elements are frame objects.  A frame object has a
     * "b" field, which is the base-address of the library or executable containing the symbol, and
     * an "o" field, which is the offset into said library or executable of the symbol.
     *
     * The JSON backtrace may optionally contain additional information useful to a backtrace
     * analysis tool.  For example, on Linux it contains a subobject named "somap", describing
     * the objects referenced in the "b" fields of the "backtrace" list.
     *
     * @param os    ostream& to receive printed stack backtrace
     */
    void printStackTrace(std::ostream& os) {
        static const char unknownFileName[] = "???";
        void* addresses[maxBackTraceFrames];
        Dl_info dlinfoForFrames[maxBackTraceFrames];

        ////////////////////////////////////////////////////////////
        // Get the backtrace addresses.
        ////////////////////////////////////////////////////////////

        const int addressCount = backtrace(addresses, maxBackTraceFrames);
        if (addressCount == 0) {
            const int err = errno;
            os << "Unable to collect backtrace addresses (errno: " <<
                err << ' ' << strerror(err) << ')' << std::endl;
            return;
        }

        ////////////////////////////////////////////////////////////
        // Collect symbol information for each backtrace address.
        ////////////////////////////////////////////////////////////

        os << std::hex << std::uppercase << '\n';
        for (int i = 0; i < addressCount; ++i) {
            Dl_info& dlinfo(dlinfoForFrames[i]);
            if (!dladdr(addresses[i], &dlinfo)) {
                dlinfo.dli_fname = unknownFileName;
                dlinfo.dli_fbase = NULL;
                dlinfo.dli_sname = NULL;
                dlinfo.dli_saddr = NULL;
            }
            os << ' ' << addresses[i];
        }

        os << "\n----- BEGIN BACKTRACE -----\n";

        ////////////////////////////////////////////////////////////
        // Display the JSON backtrace
        ////////////////////////////////////////////////////////////

        os << "{\"backtrace\":[";
        for (int i = 0; i < addressCount; ++i) {
            const Dl_info& dlinfo = dlinfoForFrames[i];
            const uintptr_t fileOffset = uintptr_t(addresses[i]) - uintptr_t(dlinfo.dli_fbase);
            if (i)
                os << ',';
            os << "{\"b\":\"" << uintptr_t(dlinfo.dli_fbase) <<
                "\",\"o\":\"" << fileOffset << "\"}";
        }
        os << ']';

        if (soMapJson)
            os << ",\"processInfo\":" << *soMapJson;
        os << "}\n";

        ////////////////////////////////////////////////////////////
        // Display the human-readable trace
        ////////////////////////////////////////////////////////////
        for (int i = 0; i < addressCount; ++i) {
            Dl_info& dlinfo(dlinfoForFrames[i]);
            os << ' ';
            if (dlinfo.dli_fbase) {
                os << getBaseName(dlinfo.dli_fname) << '(';
                if (dlinfo.dli_sname) {
                    const uintptr_t offset = uintptr_t(addresses[i]) - uintptr_t(dlinfo.dli_saddr);
                    os << dlinfo.dli_sname << "+0x" << offset;
                }
                else {
                    const uintptr_t offset = uintptr_t(addresses[i]) - uintptr_t(dlinfo.dli_fbase);
                    os << "+0x" << offset;
                }
                os << ')';
            }
            else {
                os << unknownFileName;
            }
            os << " [" << addresses[i] << ']' << std::endl;
        }

        os << std::dec << std::nouppercase;
        os << "-----  END BACKTRACE  -----" << std::endl;
    }
}  // namespace mongo

#if defined(__linux__)

#include <elf.h>
#include <link.h>

namespace mongo {
namespace {

    /**
     * Rounds a byte offset up to the next highest offset that is aligned with an ELF Word.
     */
    size_t roundUpToElfWordAlignment(size_t offset) {
        static const size_t elfWordSizeBytes = sizeof(ElfW(Word));
        return (offset + (elfWordSizeBytes - 1)) & ~(elfWordSizeBytes - 1);
    }

    /**
     * Returns the size in bytes of an ELF note entry with the given header.
     */
    size_t getNoteSizeBytes(const ElfW(Nhdr)& noteHeader) {
        return sizeof(noteHeader) +
            roundUpToElfWordAlignment(noteHeader.n_namesz) +
            roundUpToElfWordAlignment(noteHeader.n_descsz);
    }

    /**
     * Returns true of the given ELF program header refers to a runtime-readable segment.
     */
    bool isSegmentMappedReadable(const ElfW(Phdr)& phdr) {
        return phdr.p_flags & PF_R;
    }

    /**
     * Processes an ELF Phdr for a NOTE segment, updating "soInfo".
     *
     * Looks for the GNU Build ID NOTE, and adds a buildId field to soInfo if it finds one.
     */
    void processNoteSegment(const dl_phdr_info& info,
                            const ElfW(Phdr)& phdr,
                            BSONObjBuilder* soInfo) {
#ifdef NT_GNU_BUILD_ID
        const char* const notesBegin = reinterpret_cast<const char*>(info.dlpi_addr) + phdr.p_vaddr;
        const char* const notesEnd = notesBegin + phdr.p_memsz;
        ElfW(Nhdr) noteHeader;
        for (const char* notesCurr = notesBegin;
             (notesCurr + sizeof(noteHeader)) < notesEnd;
             notesCurr += getNoteSizeBytes(noteHeader)) {

            memcpy(&noteHeader, notesCurr, sizeof(noteHeader));
            if (noteHeader.n_type != NT_GNU_BUILD_ID)
                continue;
            const char* const noteNameBegin = notesCurr + sizeof(noteHeader);
            if (StringData(noteNameBegin, noteHeader.n_namesz - 1) !=
                StringData(ELF_NOTE_GNU, StringData::LiteralTag())) {
                continue;
            }
            const char* const noteDescBegin =
                noteNameBegin + roundUpToElfWordAlignment(noteHeader.n_namesz);
            soInfo->append("buildId", toHex(noteDescBegin, noteHeader.n_descsz));
        }
#endif
    }

    /**
     * Processes an ELF Phdr for a LOAD segment, updating "soInfo".
     *
     * The goal of this operation is to find out if the current object is an executable or a shared
     * object, by looking for the LOAD segment that maps the first several bytes of the file (the
     * ELF header).  If it's an executable, this method updates soInfo with the load address of the
     * segment
     */
    void processLoadSegment(const dl_phdr_info& info,
                            const ElfW(Phdr)& phdr,
                            BSONObjBuilder* soInfo) {
        if (phdr.p_offset)
            return;
        if (phdr.p_memsz < sizeof(ElfW(Ehdr)))
            return;

        // Segment includes beginning of file and is large enough to hold the ELF header
        ElfW(Ehdr) eHeader;
        memcpy(&eHeader,
               reinterpret_cast<const char*>(info.dlpi_addr) + phdr.p_vaddr,
               sizeof(eHeader));

        std::string quotedFileName = "\"" + escape(info.dlpi_name) + "\"";

        if (memcmp(&eHeader.e_ident[0], ELFMAG, SELFMAG)) {
            warning() << "Bad ELF magic number in image of " << quotedFileName;
            return;
        }

#define MKELFCLASS(N) _MKELFCLASS(N)
#define _MKELFCLASS(N) ELFCLASS ## N
        if (eHeader.e_ident[EI_CLASS] != MKELFCLASS(__ELF_NATIVE_CLASS)) {
            warning() << "Expected elf file class of " << quotedFileName << " to be " <<
                MKELFCLASS(__ELF_NATIVE_CLASS) << "(" << __ELF_NATIVE_CLASS <<
                "-bit), but found " << int(eHeader.e_ident[4]);
            return;
        }

        if (eHeader.e_ident[EI_VERSION] != EV_CURRENT) {
            warning() << "Wrong ELF version in " << quotedFileName << ".  Expected " <<
                EV_CURRENT << " but found " << int(eHeader.e_ident[EI_VERSION]);
            return;
        }

        soInfo->append("elfType", eHeader.e_type);

        switch (eHeader.e_type) {
        case ET_EXEC:
            break;
        case ET_DYN:
            return;
        default:
            warning() << "Surprised to find " << quotedFileName << " is ELF file of type " <<
                eHeader.e_type;
            return;
        }

        soInfo->append("b", integerToHex(phdr.p_vaddr));
    }

    /**
     * Callback that processes an ELF object linked into the current address space.
     *
     * Used by dl_iterate_phdr in ExtractSOMap, below, to build up the list of linked
     * objects.
     *
     * Each entry built by an invocation of ths function may have the following fields:
     * * "b", the base address at which an object is loaded.
     * * "path", the path on the file system to the object.
     * * "buildId", the GNU Build ID of the object.
     * * "elfType", the ELF type of the object, typically 2 or 3 (executable or SO).
     *
     * At post-processing time, the buildId field can be used to identify the file containing
     * debug symbols for objects loaded at the given "laodAddr", which in turn can be used with
     * the "backtrace" displayed in printStackTrace to get detailed unwind information.
     */
    int outputSOInfo(dl_phdr_info *info, size_t sz, void* data) {
        BSONObjBuilder soInfo(reinterpret_cast<BSONArrayBuilder*>(data)->subobjStart());
        if (info->dlpi_addr)
            soInfo.append("b", integerToHex(ElfW(Addr)(info->dlpi_addr)));
        if (info->dlpi_name && *info->dlpi_name)
            soInfo.append("path", info->dlpi_name);

        for (ElfW(Half) i = 0; i < info->dlpi_phnum; ++i) {
            const ElfW(Phdr)& phdr(info->dlpi_phdr[i]);
            if (!isSegmentMappedReadable(phdr))
                continue;
            switch (phdr.p_type) {
            case PT_NOTE:
                processNoteSegment(*info, phdr, &soInfo);
                break;
            case PT_LOAD:
                processLoadSegment(*info, phdr, &soInfo);
                break;
            default:
                break;
            }
        }
        return 0;
    }

    /**
     * Builds the "soMapJson" string for Linux, which is of the following form:
     *
     * '"somap": [<Objects described by outputSOInfo, above>]'
     */
    MONGO_INITIALIZER(ExtractSOMap)(InitializerContext*) {
        BSONObjBuilder soMap;
        soMap << "mongodbVersion" << versionString;
        soMap << "gitVersion" << gitVersion();
        BSONArrayBuilder soList(soMap.subarrayStart("somap"));
        dl_iterate_phdr(outputSOInfo, &soList);
        soList.done();
        soMapJson = new std::string(soMap.done().jsonString(Strict));
        return Status::OK();
    }
}  // namespace

}  // namespace mongo

#endif