Mercurial > hg > truffle
view agent/src/os/linux/ps_core.c @ 14709:d3225562f0d8
guard implicit exception deopt
author | Tom Rodriguez <tom.rodriguez@oracle.com> |
---|---|
date | Thu, 20 Mar 2014 17:36:43 -0700 |
parents | 1bee3014cf2a |
children |
line wrap: on
line source
/* * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. * */ #include <jni.h> #include <unistd.h> #include <fcntl.h> #include <string.h> #include <stdlib.h> #include <stddef.h> #include <elf.h> #include <link.h> #include "libproc_impl.h" #include "salibelf.h" // This file has the libproc implementation to read core files. // For live processes, refer to ps_proc.c. Portions of this is adapted // /modelled after Solaris libproc.so (in particular Pcore.c) //---------------------------------------------------------------------- // ps_prochandle cleanup helper functions // close all file descriptors static void close_files(struct ps_prochandle* ph) { lib_info* lib = NULL; // close core file descriptor if (ph->core->core_fd >= 0) close(ph->core->core_fd); // close exec file descriptor if (ph->core->exec_fd >= 0) close(ph->core->exec_fd); // close interp file descriptor if (ph->core->interp_fd >= 0) close(ph->core->interp_fd); // close class share archive file if (ph->core->classes_jsa_fd >= 0) close(ph->core->classes_jsa_fd); // close all library file descriptors lib = ph->libs; while (lib) { int fd = lib->fd; if (fd >= 0 && fd != ph->core->exec_fd) { close(fd); } lib = lib->next; } } // clean all map_info stuff static void destroy_map_info(struct ps_prochandle* ph) { map_info* map = ph->core->maps; while (map) { map_info* next = map->next; free(map); map = next; } if (ph->core->map_array) { free(ph->core->map_array); } // Part of the class sharing workaround map = ph->core->class_share_maps; while (map) { map_info* next = map->next; free(map); map = next; } } // ps_prochandle operations static void core_release(struct ps_prochandle* ph) { if (ph->core) { close_files(ph); destroy_map_info(ph); free(ph->core); } } static map_info* allocate_init_map(int fd, off_t offset, uintptr_t vaddr, size_t memsz) { map_info* map; if ( (map = (map_info*) calloc(1, sizeof(map_info))) == NULL) { print_debug("can't allocate memory for map_info\n"); return NULL; } // initialize map map->fd = fd; map->offset = offset; map->vaddr = vaddr; map->memsz = memsz; return map; } // add map info with given fd, offset, vaddr and memsz static map_info* add_map_info(struct ps_prochandle* ph, int fd, off_t offset, uintptr_t vaddr, size_t memsz) { map_info* map; if ((map = allocate_init_map(fd, offset, vaddr, memsz)) == NULL) { return NULL; } // add this to map list map->next = ph->core->maps; ph->core->maps = map; ph->core->num_maps++; return map; } // Part of the class sharing workaround static map_info* add_class_share_map_info(struct ps_prochandle* ph, off_t offset, uintptr_t vaddr, size_t memsz) { map_info* map; if ((map = allocate_init_map(ph->core->classes_jsa_fd, offset, vaddr, memsz)) == NULL) { return NULL; } map->next = ph->core->class_share_maps; ph->core->class_share_maps = map; return map; } // Return the map_info for the given virtual address. We keep a sorted // array of pointers in ph->map_array, so we can binary search. static map_info* core_lookup(struct ps_prochandle *ph, uintptr_t addr) { int mid, lo = 0, hi = ph->core->num_maps - 1; map_info *mp; while (hi - lo > 1) { mid = (lo + hi) / 2; if (addr >= ph->core->map_array[mid]->vaddr) { lo = mid; } else { hi = mid; } } if (addr < ph->core->map_array[hi]->vaddr) { mp = ph->core->map_array[lo]; } else { mp = ph->core->map_array[hi]; } if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) { return (mp); } // Part of the class sharing workaround // Unfortunately, we have no way of detecting -Xshare state. // Check out the share maps atlast, if we don't find anywhere. // This is done this way so to avoid reading share pages // ahead of other normal maps. For eg. with -Xshare:off we don't // want to prefer class sharing data to data from core. mp = ph->core->class_share_maps; if (mp) { print_debug("can't locate map_info at 0x%lx, trying class share maps\n", addr); } while (mp) { if (addr >= mp->vaddr && addr < mp->vaddr + mp->memsz) { print_debug("located map_info at 0x%lx from class share maps\n", addr); return (mp); } mp = mp->next; } print_debug("can't locate map_info at 0x%lx\n", addr); return (NULL); } //--------------------------------------------------------------- // Part of the class sharing workaround: // // With class sharing, pages are mapped from classes.jsa file. // The read-only class sharing pages are mapped as MAP_SHARED, // PROT_READ pages. These pages are not dumped into core dump. // With this workaround, these pages are read from classes.jsa. // FIXME: !HACK ALERT! // The format of sharing achive file header is needed to read shared heap // file mappings. For now, I am hard coding portion of FileMapHeader here. // Refer to filemap.hpp. // FileMapHeader describes the shared space data in the file to be // mapped. This structure gets written to a file. It is not a class, // so that the compilers don't add any compiler-private data to it. #define NUM_SHARED_MAPS 4 // Refer to FileMapInfo::_current_version in filemap.hpp #define CURRENT_ARCHIVE_VERSION 1 struct FileMapHeader { int _magic; // identify file type. int _version; // (from enum, above.) size_t _alignment; // how shared archive should be aligned struct space_info { int _file_offset; // sizeof(this) rounded to vm page size char* _base; // copy-on-write base address size_t _capacity; // for validity checking size_t _used; // for setting space top on read // 4991491 NOTICE These are C++ bool's in filemap.hpp and must match up with // the C type matching the C++ bool type on any given platform. // We assume the corresponding C type is char but licensees // may need to adjust the type of these fields. char _read_only; // read only space? char _allow_exec; // executable code in space? } _space[NUM_SHARED_MAPS]; // Ignore the rest of the FileMapHeader. We don't need those fields here. }; static bool read_jboolean(struct ps_prochandle* ph, uintptr_t addr, jboolean* pvalue) { jboolean i; if (ps_pdread(ph, (psaddr_t) addr, &i, sizeof(i)) == PS_OK) { *pvalue = i; return true; } else { return false; } } static bool read_pointer(struct ps_prochandle* ph, uintptr_t addr, uintptr_t* pvalue) { uintptr_t uip; if (ps_pdread(ph, (psaddr_t) addr, (char *)&uip, sizeof(uip)) == PS_OK) { *pvalue = uip; return true; } else { return false; } } // used to read strings from debuggee static bool read_string(struct ps_prochandle* ph, uintptr_t addr, char* buf, size_t size) { size_t i = 0; char c = ' '; while (c != '\0') { if (ps_pdread(ph, (psaddr_t) addr, &c, sizeof(char)) != PS_OK) { return false; } if (i < size - 1) { buf[i] = c; } else { // smaller buffer return false; } i++; addr++; } buf[i] = '\0'; return true; } #define USE_SHARED_SPACES_SYM "UseSharedSpaces" // mangled name of Arguments::SharedArchivePath #define SHARED_ARCHIVE_PATH_SYM "_ZN9Arguments17SharedArchivePathE" #define LIBJVM_NAME "/libjvm.so" static bool init_classsharing_workaround(struct ps_prochandle* ph) { lib_info* lib = ph->libs; while (lib != NULL) { // we are iterating over shared objects from the core dump. look for // libjvm.so. const char *jvm_name = 0; if ((jvm_name = strstr(lib->name, LIBJVM_NAME)) != 0) { char classes_jsa[PATH_MAX]; struct FileMapHeader header; int fd = -1; int m = 0; size_t n = 0; uintptr_t base = 0, useSharedSpacesAddr = 0; uintptr_t sharedArchivePathAddrAddr = 0, sharedArchivePathAddr = 0; jboolean useSharedSpaces = 0; map_info* mi = 0; memset(classes_jsa, 0, sizeof(classes_jsa)); jvm_name = lib->name; useSharedSpacesAddr = lookup_symbol(ph, jvm_name, USE_SHARED_SPACES_SYM); if (useSharedSpacesAddr == 0) { print_debug("can't lookup 'UseSharedSpaces' flag\n"); return false; } // Hotspot vm types are not exported to build this library. So // using equivalent type jboolean to read the value of // UseSharedSpaces which is same as hotspot type "bool". if (read_jboolean(ph, useSharedSpacesAddr, &useSharedSpaces) != true) { print_debug("can't read the value of 'UseSharedSpaces' flag\n"); return false; } if ((int)useSharedSpaces == 0) { print_debug("UseSharedSpaces is false, assuming -Xshare:off!\n"); return true; } sharedArchivePathAddrAddr = lookup_symbol(ph, jvm_name, SHARED_ARCHIVE_PATH_SYM); if (sharedArchivePathAddrAddr == 0) { print_debug("can't lookup shared archive path symbol\n"); return false; } if (read_pointer(ph, sharedArchivePathAddrAddr, &sharedArchivePathAddr) != true) { print_debug("can't read shared archive path pointer\n"); return false; } if (read_string(ph, sharedArchivePathAddr, classes_jsa, sizeof(classes_jsa)) != true) { print_debug("can't read shared archive path value\n"); return false; } print_debug("looking for %s\n", classes_jsa); // open the class sharing archive file fd = pathmap_open(classes_jsa); if (fd < 0) { print_debug("can't open %s!\n", classes_jsa); ph->core->classes_jsa_fd = -1; return false; } else { print_debug("opened %s\n", classes_jsa); } // read FileMapHeader from the file memset(&header, 0, sizeof(struct FileMapHeader)); if ((n = read(fd, &header, sizeof(struct FileMapHeader))) != sizeof(struct FileMapHeader)) { print_debug("can't read shared archive file map header from %s\n", classes_jsa); close(fd); return false; } // check file magic if (header._magic != 0xf00baba2) { print_debug("%s has bad shared archive file magic number 0x%x, expecing 0xf00baba2\n", classes_jsa, header._magic); close(fd); return false; } // check version if (header._version != CURRENT_ARCHIVE_VERSION) { print_debug("%s has wrong shared archive file version %d, expecting %d\n", classes_jsa, header._version, CURRENT_ARCHIVE_VERSION); close(fd); return false; } ph->core->classes_jsa_fd = fd; // add read-only maps from classes.jsa to the list of maps for (m = 0; m < NUM_SHARED_MAPS; m++) { if (header._space[m]._read_only) { base = (uintptr_t) header._space[m]._base; // no need to worry about the fractional pages at-the-end. // possible fractional pages are handled by core_read_data. add_class_share_map_info(ph, (off_t) header._space[m]._file_offset, base, (size_t) header._space[m]._used); print_debug("added a share archive map at 0x%lx\n", base); } } return true; } lib = lib->next; } return true; } //--------------------------------------------------------------------------- // functions to handle map_info // Order mappings based on virtual address. We use this function as the // callback for sorting the array of map_info pointers. static int core_cmp_mapping(const void *lhsp, const void *rhsp) { const map_info *lhs = *((const map_info **)lhsp); const map_info *rhs = *((const map_info **)rhsp); if (lhs->vaddr == rhs->vaddr) { return (0); } return (lhs->vaddr < rhs->vaddr ? -1 : 1); } // we sort map_info by starting virtual address so that we can do // binary search to read from an address. static bool sort_map_array(struct ps_prochandle* ph) { size_t num_maps = ph->core->num_maps; map_info* map = ph->core->maps; int i = 0; // allocate map_array map_info** array; if ( (array = (map_info**) malloc(sizeof(map_info*) * num_maps)) == NULL) { print_debug("can't allocate memory for map array\n"); return false; } // add maps to array while (map) { array[i] = map; i++; map = map->next; } // sort is called twice. If this is second time, clear map array if (ph->core->map_array) { free(ph->core->map_array); } ph->core->map_array = array; // sort the map_info array by base virtual address. qsort(ph->core->map_array, ph->core->num_maps, sizeof (map_info*), core_cmp_mapping); // print map if (is_debug()) { int j = 0; print_debug("---- sorted virtual address map ----\n"); for (j = 0; j < ph->core->num_maps; j++) { print_debug("base = 0x%lx\tsize = %zu\n", ph->core->map_array[j]->vaddr, ph->core->map_array[j]->memsz); } } return true; } #ifndef MIN #define MIN(x, y) (((x) < (y))? (x): (y)) #endif static bool core_read_data(struct ps_prochandle* ph, uintptr_t addr, char *buf, size_t size) { ssize_t resid = size; int page_size=sysconf(_SC_PAGE_SIZE); while (resid != 0) { map_info *mp = core_lookup(ph, addr); uintptr_t mapoff; ssize_t len, rem; off_t off; int fd; if (mp == NULL) { break; /* No mapping for this address */ } fd = mp->fd; mapoff = addr - mp->vaddr; len = MIN(resid, mp->memsz - mapoff); off = mp->offset + mapoff; if ((len = pread(fd, buf, len, off)) <= 0) { break; } resid -= len; addr += len; buf = (char *)buf + len; // mappings always start at page boundary. But, may end in fractional // page. fill zeros for possible fractional page at the end of a mapping. rem = mp->memsz % page_size; if (rem > 0) { rem = page_size - rem; len = MIN(resid, rem); resid -= len; addr += len; // we are not assuming 'buf' to be zero initialized. memset(buf, 0, len); buf += len; } } if (resid) { print_debug("core read failed for %d byte(s) @ 0x%lx (%d more bytes)\n", size, addr, resid); return false; } else { return true; } } // null implementation for write static bool core_write_data(struct ps_prochandle* ph, uintptr_t addr, const char *buf , size_t size) { return false; } static bool core_get_lwp_regs(struct ps_prochandle* ph, lwpid_t lwp_id, struct user_regs_struct* regs) { // for core we have cached the lwp regs from NOTE section thread_info* thr = ph->threads; while (thr) { if (thr->lwp_id == lwp_id) { memcpy(regs, &thr->regs, sizeof(struct user_regs_struct)); return true; } thr = thr->next; } return false; } static ps_prochandle_ops core_ops = { .release= core_release, .p_pread= core_read_data, .p_pwrite= core_write_data, .get_lwp_regs= core_get_lwp_regs }; // read regs and create thread from NT_PRSTATUS entries from core file static bool core_handle_prstatus(struct ps_prochandle* ph, const char* buf, size_t nbytes) { // we have to read prstatus_t from buf // assert(nbytes == sizeof(prstaus_t), "size mismatch on prstatus_t"); prstatus_t* prstat = (prstatus_t*) buf; thread_info* newthr; print_debug("got integer regset for lwp %d\n", prstat->pr_pid); // we set pthread_t to -1 for core dump if((newthr = add_thread_info(ph, (pthread_t) -1, prstat->pr_pid)) == NULL) return false; // copy regs memcpy(&newthr->regs, prstat->pr_reg, sizeof(struct user_regs_struct)); if (is_debug()) { print_debug("integer regset\n"); #ifdef i386 // print the regset print_debug("\teax = 0x%x\n", newthr->regs.eax); print_debug("\tebx = 0x%x\n", newthr->regs.ebx); print_debug("\tecx = 0x%x\n", newthr->regs.ecx); print_debug("\tedx = 0x%x\n", newthr->regs.edx); print_debug("\tesp = 0x%x\n", newthr->regs.esp); print_debug("\tebp = 0x%x\n", newthr->regs.ebp); print_debug("\tesi = 0x%x\n", newthr->regs.esi); print_debug("\tedi = 0x%x\n", newthr->regs.edi); print_debug("\teip = 0x%x\n", newthr->regs.eip); #endif #if defined(amd64) || defined(x86_64) // print the regset print_debug("\tr15 = 0x%lx\n", newthr->regs.r15); print_debug("\tr14 = 0x%lx\n", newthr->regs.r14); print_debug("\tr13 = 0x%lx\n", newthr->regs.r13); print_debug("\tr12 = 0x%lx\n", newthr->regs.r12); print_debug("\trbp = 0x%lx\n", newthr->regs.rbp); print_debug("\trbx = 0x%lx\n", newthr->regs.rbx); print_debug("\tr11 = 0x%lx\n", newthr->regs.r11); print_debug("\tr10 = 0x%lx\n", newthr->regs.r10); print_debug("\tr9 = 0x%lx\n", newthr->regs.r9); print_debug("\tr8 = 0x%lx\n", newthr->regs.r8); print_debug("\trax = 0x%lx\n", newthr->regs.rax); print_debug("\trcx = 0x%lx\n", newthr->regs.rcx); print_debug("\trdx = 0x%lx\n", newthr->regs.rdx); print_debug("\trsi = 0x%lx\n", newthr->regs.rsi); print_debug("\trdi = 0x%lx\n", newthr->regs.rdi); print_debug("\torig_rax = 0x%lx\n", newthr->regs.orig_rax); print_debug("\trip = 0x%lx\n", newthr->regs.rip); print_debug("\tcs = 0x%lx\n", newthr->regs.cs); print_debug("\teflags = 0x%lx\n", newthr->regs.eflags); print_debug("\trsp = 0x%lx\n", newthr->regs.rsp); print_debug("\tss = 0x%lx\n", newthr->regs.ss); print_debug("\tfs_base = 0x%lx\n", newthr->regs.fs_base); print_debug("\tgs_base = 0x%lx\n", newthr->regs.gs_base); print_debug("\tds = 0x%lx\n", newthr->regs.ds); print_debug("\tes = 0x%lx\n", newthr->regs.es); print_debug("\tfs = 0x%lx\n", newthr->regs.fs); print_debug("\tgs = 0x%lx\n", newthr->regs.gs); #endif } return true; } #define ROUNDUP(x, y) ((((x)+((y)-1))/(y))*(y)) // read NT_PRSTATUS entries from core NOTE segment static bool core_handle_note(struct ps_prochandle* ph, ELF_PHDR* note_phdr) { char* buf = NULL; char* p = NULL; size_t size = note_phdr->p_filesz; // we are interested in just prstatus entries. we will ignore the rest. // Advance the seek pointer to the start of the PT_NOTE data if (lseek(ph->core->core_fd, note_phdr->p_offset, SEEK_SET) == (off_t)-1) { print_debug("failed to lseek to PT_NOTE data\n"); return false; } // Now process the PT_NOTE structures. Each one is preceded by // an Elf{32/64}_Nhdr structure describing its type and size. if ( (buf = (char*) malloc(size)) == NULL) { print_debug("can't allocate memory for reading core notes\n"); goto err; } // read notes into buffer if (read(ph->core->core_fd, buf, size) != size) { print_debug("failed to read notes, core file must have been truncated\n"); goto err; } p = buf; while (p < buf + size) { ELF_NHDR* notep = (ELF_NHDR*) p; char* descdata = p + sizeof(ELF_NHDR) + ROUNDUP(notep->n_namesz, 4); print_debug("Note header with n_type = %d and n_descsz = %u\n", notep->n_type, notep->n_descsz); if (notep->n_type == NT_PRSTATUS) { if (core_handle_prstatus(ph, descdata, notep->n_descsz) != true) { return false; } } p = descdata + ROUNDUP(notep->n_descsz, 4); } free(buf); return true; err: if (buf) free(buf); return false; } // read all segments from core file static bool read_core_segments(struct ps_prochandle* ph, ELF_EHDR* core_ehdr) { int i = 0; ELF_PHDR* phbuf = NULL; ELF_PHDR* core_php = NULL; if ((phbuf = read_program_header_table(ph->core->core_fd, core_ehdr)) == NULL) return false; /* * Now iterate through the program headers in the core file. * We're interested in two types of Phdrs: PT_NOTE (which * contains a set of saved /proc structures), and PT_LOAD (which * represents a memory mapping from the process's address space). * * Difference b/w Solaris PT_NOTE and Linux/BSD PT_NOTE: * * In Solaris there are two PT_NOTE segments the first PT_NOTE (if present) * contains /proc structs in the pre-2.6 unstructured /proc format. the last * PT_NOTE has data in new /proc format. * * In Solaris, there is only one pstatus (process status). pstatus contains * integer register set among other stuff. For each LWP, we have one lwpstatus * entry that has integer regset for that LWP. * * Linux threads are actually 'clone'd processes. To support core analysis * of "multithreaded" process, Linux creates more than one pstatus (called * "prstatus") entry in PT_NOTE. Each prstatus entry has integer regset for one * "thread". Please refer to Linux kernel src file 'fs/binfmt_elf.c', in particular * function "elf_core_dump". */ for (core_php = phbuf, i = 0; i < core_ehdr->e_phnum; i++) { switch (core_php->p_type) { case PT_NOTE: if (core_handle_note(ph, core_php) != true) { goto err; } break; case PT_LOAD: { if (core_php->p_filesz != 0) { if (add_map_info(ph, ph->core->core_fd, core_php->p_offset, core_php->p_vaddr, core_php->p_filesz) == NULL) goto err; } break; } } core_php++; } free(phbuf); return true; err: free(phbuf); return false; } // read segments of a shared object static bool read_lib_segments(struct ps_prochandle* ph, int lib_fd, ELF_EHDR* lib_ehdr, uintptr_t lib_base) { int i = 0; ELF_PHDR* phbuf; ELF_PHDR* lib_php = NULL; int page_size = sysconf(_SC_PAGE_SIZE); if ((phbuf = read_program_header_table(lib_fd, lib_ehdr)) == NULL) { return false; } // we want to process only PT_LOAD segments that are not writable. // i.e., text segments. The read/write/exec (data) segments would // have been already added from core file segments. for (lib_php = phbuf, i = 0; i < lib_ehdr->e_phnum; i++) { if ((lib_php->p_type == PT_LOAD) && !(lib_php->p_flags & PF_W) && (lib_php->p_filesz != 0)) { uintptr_t target_vaddr = lib_php->p_vaddr + lib_base; map_info *existing_map = core_lookup(ph, target_vaddr); if (existing_map == NULL){ if (add_map_info(ph, lib_fd, lib_php->p_offset, target_vaddr, lib_php->p_memsz) == NULL) { goto err; } } else { // Coredump stores value of p_memsz elf field // rounded up to page boundary. if ((existing_map->memsz != page_size) && (existing_map->fd != lib_fd) && (ROUNDUP(existing_map->memsz, page_size) != ROUNDUP(lib_php->p_memsz, page_size))) { print_debug("address conflict @ 0x%lx (existing map size = %ld, size = %ld, flags = %d)\n", target_vaddr, existing_map->memsz, lib_php->p_memsz, lib_php->p_flags); goto err; } /* replace PT_LOAD segment with library segment */ print_debug("overwrote with new address mapping (memsz %ld -> %ld)\n", existing_map->memsz, ROUNDUP(lib_php->p_memsz, page_size)); existing_map->fd = lib_fd; existing_map->offset = lib_php->p_offset; existing_map->memsz = ROUNDUP(lib_php->p_memsz, page_size); } } lib_php++; } free(phbuf); return true; err: free(phbuf); return false; } // process segments from interpreter (ld.so or ld-linux.so) static bool read_interp_segments(struct ps_prochandle* ph) { ELF_EHDR interp_ehdr; if (read_elf_header(ph->core->interp_fd, &interp_ehdr) != true) { print_debug("interpreter is not a valid ELF file\n"); return false; } if (read_lib_segments(ph, ph->core->interp_fd, &interp_ehdr, ph->core->ld_base_addr) != true) { print_debug("can't read segments of interpreter\n"); return false; } return true; } // process segments of a a.out static bool read_exec_segments(struct ps_prochandle* ph, ELF_EHDR* exec_ehdr) { int i = 0; ELF_PHDR* phbuf = NULL; ELF_PHDR* exec_php = NULL; if ((phbuf = read_program_header_table(ph->core->exec_fd, exec_ehdr)) == NULL) return false; for (exec_php = phbuf, i = 0; i < exec_ehdr->e_phnum; i++) { switch (exec_php->p_type) { // add mappings for PT_LOAD segments case PT_LOAD: { // add only non-writable segments of non-zero filesz if (!(exec_php->p_flags & PF_W) && exec_php->p_filesz != 0) { if (add_map_info(ph, ph->core->exec_fd, exec_php->p_offset, exec_php->p_vaddr, exec_php->p_filesz) == NULL) goto err; } break; } // read the interpreter and it's segments case PT_INTERP: { char interp_name[BUF_SIZE]; pread(ph->core->exec_fd, interp_name, MIN(exec_php->p_filesz, BUF_SIZE), exec_php->p_offset); print_debug("ELF interpreter %s\n", interp_name); // read interpreter segments as well if ((ph->core->interp_fd = pathmap_open(interp_name)) < 0) { print_debug("can't open runtime loader\n"); goto err; } break; } // from PT_DYNAMIC we want to read address of first link_map addr case PT_DYNAMIC: { ph->core->dynamic_addr = exec_php->p_vaddr; print_debug("address of _DYNAMIC is 0x%lx\n", ph->core->dynamic_addr); break; } } // switch exec_php++; } // for free(phbuf); return true; err: free(phbuf); return false; } #define FIRST_LINK_MAP_OFFSET offsetof(struct r_debug, r_map) #define LD_BASE_OFFSET offsetof(struct r_debug, r_ldbase) #define LINK_MAP_ADDR_OFFSET offsetof(struct link_map, l_addr) #define LINK_MAP_NAME_OFFSET offsetof(struct link_map, l_name) #define LINK_MAP_NEXT_OFFSET offsetof(struct link_map, l_next) // read shared library info from runtime linker's data structures. // This work is done by librtlb_db in Solaris static bool read_shared_lib_info(struct ps_prochandle* ph) { uintptr_t addr = ph->core->dynamic_addr; uintptr_t debug_base; uintptr_t first_link_map_addr; uintptr_t ld_base_addr; uintptr_t link_map_addr; uintptr_t lib_base_diff; uintptr_t lib_base; uintptr_t lib_name_addr; char lib_name[BUF_SIZE]; ELF_DYN dyn; ELF_EHDR elf_ehdr; int lib_fd; // _DYNAMIC has information of the form // [tag] [data] [tag] [data] ..... // Both tag and data are pointer sized. // We look for dynamic info with DT_DEBUG. This has shared object info. // refer to struct r_debug in link.h dyn.d_tag = DT_NULL; while (dyn.d_tag != DT_DEBUG) { if (ps_pdread(ph, (psaddr_t) addr, &dyn, sizeof(ELF_DYN)) != PS_OK) { print_debug("can't read debug info from _DYNAMIC\n"); return false; } addr += sizeof(ELF_DYN); } // we have got Dyn entry with DT_DEBUG debug_base = dyn.d_un.d_ptr; // at debug_base we have struct r_debug. This has first link map in r_map field if (ps_pdread(ph, (psaddr_t) debug_base + FIRST_LINK_MAP_OFFSET, &first_link_map_addr, sizeof(uintptr_t)) != PS_OK) { print_debug("can't read first link map address\n"); return false; } // read ld_base address from struct r_debug if (ps_pdread(ph, (psaddr_t) debug_base + LD_BASE_OFFSET, &ld_base_addr, sizeof(uintptr_t)) != PS_OK) { print_debug("can't read ld base address\n"); return false; } ph->core->ld_base_addr = ld_base_addr; print_debug("interpreter base address is 0x%lx\n", ld_base_addr); // now read segments from interp (i.e ld.so or ld-linux.so or ld-elf.so) if (read_interp_segments(ph) != true) { return false; } // after adding interpreter (ld.so) mappings sort again if (sort_map_array(ph) != true) { return false; } print_debug("first link map is at 0x%lx\n", first_link_map_addr); link_map_addr = first_link_map_addr; while (link_map_addr != 0) { // read library base address of the .so. Note that even though <sys/link.h> calls // link_map->l_addr as "base address", this is * not * really base virtual // address of the shared object. This is actually the difference b/w the virtual // address mentioned in shared object and the actual virtual base where runtime // linker loaded it. We use "base diff" in read_lib_segments call below. if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_ADDR_OFFSET, &lib_base_diff, sizeof(uintptr_t)) != PS_OK) { print_debug("can't read shared object base address diff\n"); return false; } // read address of the name if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_NAME_OFFSET, &lib_name_addr, sizeof(uintptr_t)) != PS_OK) { print_debug("can't read address of shared object name\n"); return false; } // read name of the shared object lib_name[0] = '\0'; if (lib_name_addr != 0 && read_string(ph, (uintptr_t) lib_name_addr, lib_name, sizeof(lib_name)) != true) { print_debug("can't read shared object name\n"); // don't let failure to read the name stop opening the file. If something is really wrong // it will fail later. } if (lib_name[0] != '\0') { // ignore empty lib names lib_fd = pathmap_open(lib_name); if (lib_fd < 0) { print_debug("can't open shared object %s\n", lib_name); // continue with other libraries... } else { if (read_elf_header(lib_fd, &elf_ehdr)) { lib_base = lib_base_diff + find_base_address(lib_fd, &elf_ehdr); print_debug("reading library %s @ 0x%lx [ 0x%lx ]\n", lib_name, lib_base, lib_base_diff); // while adding library mappings we need to use "base difference". if (! read_lib_segments(ph, lib_fd, &elf_ehdr, lib_base_diff)) { print_debug("can't read shared object's segments\n"); close(lib_fd); return false; } add_lib_info_fd(ph, lib_name, lib_fd, lib_base); // Map info is added for the library (lib_name) so // we need to re-sort it before calling the p_pdread. if (sort_map_array(ph) != true) return false; } else { print_debug("can't read ELF header for shared object %s\n", lib_name); close(lib_fd); // continue with other libraries... } } } // read next link_map address if (ps_pdread(ph, (psaddr_t) link_map_addr + LINK_MAP_NEXT_OFFSET, &link_map_addr, sizeof(uintptr_t)) != PS_OK) { print_debug("can't read next link in link_map\n"); return false; } } return true; } // the one and only one exposed stuff from this file struct ps_prochandle* Pgrab_core(const char* exec_file, const char* core_file) { ELF_EHDR core_ehdr; ELF_EHDR exec_ehdr; ELF_EHDR lib_ehdr; struct ps_prochandle* ph = (struct ps_prochandle*) calloc(1, sizeof(struct ps_prochandle)); if (ph == NULL) { print_debug("can't allocate ps_prochandle\n"); return NULL; } if ((ph->core = (struct core_data*) calloc(1, sizeof(struct core_data))) == NULL) { free(ph); print_debug("can't allocate ps_prochandle\n"); return NULL; } // initialize ph ph->ops = &core_ops; ph->core->core_fd = -1; ph->core->exec_fd = -1; ph->core->interp_fd = -1; // open the core file if ((ph->core->core_fd = open(core_file, O_RDONLY)) < 0) { print_debug("can't open core file\n"); goto err; } // read core file ELF header if (read_elf_header(ph->core->core_fd, &core_ehdr) != true || core_ehdr.e_type != ET_CORE) { print_debug("core file is not a valid ELF ET_CORE file\n"); goto err; } if ((ph->core->exec_fd = open(exec_file, O_RDONLY)) < 0) { print_debug("can't open executable file\n"); goto err; } if (read_elf_header(ph->core->exec_fd, &exec_ehdr) != true || exec_ehdr.e_type != ET_EXEC) { print_debug("executable file is not a valid ELF ET_EXEC file\n"); goto err; } // process core file segments if (read_core_segments(ph, &core_ehdr) != true) { goto err; } // process exec file segments if (read_exec_segments(ph, &exec_ehdr) != true) { goto err; } // exec file is also treated like a shared object for symbol search if (add_lib_info_fd(ph, exec_file, ph->core->exec_fd, (uintptr_t)0 + find_base_address(ph->core->exec_fd, &exec_ehdr)) == NULL) { goto err; } // allocate and sort maps into map_array, we need to do this // here because read_shared_lib_info needs to read from debuggee // address space if (sort_map_array(ph) != true) { goto err; } if (read_shared_lib_info(ph) != true) { goto err; } // sort again because we have added more mappings from shared objects if (sort_map_array(ph) != true) { goto err; } if (init_classsharing_workaround(ph) != true) { goto err; } return ph; err: Prelease(ph); return NULL; }