backend.h revision aa3db6b1234da0e542ba7782849cf200d0d91c1c
1/* 2 * This file is part of ltrace. 3 * Copyright (C) 2012,2013 Petr Machata, Red Hat Inc. 4 * 5 * This program is free software; you can redistribute it and/or 6 * modify it under the terms of the GNU General Public License as 7 * published by the Free Software Foundation; either version 2 of the 8 * License, or (at your option) any later version. 9 * 10 * This program is distributed in the hope that it will be useful, but 11 * WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 13 * General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write to the Free Software 17 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 18 * 02110-1301 USA 19 */ 20 21#ifndef BACKEND_H 22#define BACKEND_H 23 24#include "forward.h" 25#include "sysdep.h" 26 27#include <gelf.h> 28 29enum process_status { 30 PS_INVALID, /* Failure. */ 31 PS_STOP, /* Job-control stop. */ 32 PS_TRACING_STOP, 33 PS_SLEEPING, 34 PS_ZOMBIE, 35 PS_OTHER, /* Necessary other states can be added as needed. */ 36}; 37 38/* 39 * This file contains documentation of back end interface. Some of 40 * these may be implemented on an OS level (i.e. they are the same 41 * e.g. on all Linux architectures), some may differ per architecture 42 * on the same OS (e.g. a way to insert a breakpoint into the process 43 * image is a likely candidate). 44 */ 45 46/* Convert a PID to a path to the corresponding binary. */ 47char *pid2name(pid_t pid); 48 49/* Given a PID, find a leader of thread group. */ 50pid_t process_leader(pid_t pid); 51 52/* Given a PID of leader thread, fill in PIDs of all the tasks. The 53 * function will initialize the pointer *RET_TASKS to a 54 * newly-allocated array, and will store number of elements in that 55 * array to *RET_N. You have to free that buffer when you don't need 56 * it anymore. */ 57int process_tasks(pid_t pid, pid_t **ret_tasks, size_t *ret_n); 58 59/* Answer whether the process PID is stopped. Returns 0 when not 60 * stopped, 1 when stopped, or -1 when there was an error. */ 61int process_stopped(pid_t pid); 62 63/* Answer a status of the task PID. See enum process_status. */ 64enum process_status process_status(pid_t pid); 65 66/* Wait for PID to be ready for tracing. */ 67int wait_for_proc(pid_t pid); 68 69/* Send a signal SIG to the task PID. */ 70int task_kill(pid_t pid, int sig); 71 72/* Called after PID is attached, but before it is continued. */ 73void trace_set_options(struct process *proc); 74 75/* Called after ltrace forks. Should attach the newly created child, 76 * in whose context this function is called. */ 77void trace_me(void); 78 79/* Called when ltrace needs to attach to PID, such as when it attaches 80 * to a running process, whose PID is given on the command line. */ 81int trace_pid(pid_t pid); 82 83/* Stop tracing PID. */ 84void untrace_pid(pid_t pid); 85 86/* The back end may need to store arbitrary data to a process. This 87 * is a place where it can initialize PROC->arch_dep. XXX this should 88 * be dropped in favor of arhc_process_init on pmachata/libs. */ 89void get_arch_dep(struct process *proc); 90 91/* Return current instruction pointer of PROC. 92 * 93 * XXX note that the IP must fit into an arch pointer. This prevents 94 * us to use 32-bit ltrace to trace 64-bit process, even on arches 95 * that would otherwise support this. Above we have a definition of 96 * arch_addr_t. This should be converted to an integral type and 97 * used for target addresses throughout. */ 98void *get_instruction_pointer(struct process *proc); 99 100/* Set instruction pointer of PROC to ADDR. XXX see above. */ 101void set_instruction_pointer(struct process *proc, void *addr); 102 103/* Return current stack pointer of PROC. XXX see above. */ 104void *get_stack_pointer(struct process *proc); 105 106/* Find and return caller address, i.e. the address where the current 107 * function returns. */ 108void *get_return_addr(struct process *proc, void *stack_pointer); 109 110/* Enable breakpoint SBP in process PROC. */ 111void enable_breakpoint(struct process *proc, struct breakpoint *sbp); 112 113/* Disable breakpoint SBP in process PROC. */ 114void disable_breakpoint(struct process *proc, struct breakpoint *sbp); 115 116/* Determine whether the event that we have just seen (and that is 117 * recorded in STATUS) was a syscall. If it was, return 1. If it was 118 * a return from syscall, return 2. In both cases, set *SYSNUM to the 119 * number of said syscall. If it wasn't a syscall, return 0. If 120 * there was an error, return -1. */ 121int syscall_p(struct process *proc, int status, int *sysnum); 122 123/* Continue execution of the process with given PID. */ 124void continue_process(pid_t pid); 125 126/* Called after we received a signal SIGNUM. Should do whatever 127 * book-keeping is necessary and continue the process if 128 * necessary. */ 129void continue_after_signal(pid_t pid, int signum); 130 131/* Called after we received a system call SYSNUM. RET_P is 0 if this 132 * is system call, otherwise it's return from a system call. The 133 * callback should do whatever book-keeping is necessary and continue 134 * the process if necessary. */ 135void continue_after_syscall(struct process *proc, int sysnum, int ret_p); 136 137/* Called after we hit a breakpoint SBP. Should do whatever 138 * book-keeping is necessary and then continue the process. */ 139void continue_after_breakpoint(struct process *proc, struct breakpoint *sbp); 140 141/* Called after we received a vfork. Should do whatever book-keeping 142 * is necessary and continue the process if necessary. N.B. right 143 * now, with Linux/GNU the only back end, this is not necessary. I 144 * imagine other systems may be different. */ 145void continue_after_vfork(struct process *proc); 146 147/* Called after the process exec's. Should do whatever book-keeping 148 * is necessary and then continue the process. */ 149void continue_after_exec(struct process *proc); 150 151/* Called when trace_me or primary trace_pid fail. This may plug in 152 * any platform-specific knowledge of why it could be so. */ 153void trace_fail_warning(pid_t pid); 154 155/* A pair of functions called to initiate a detachment request when 156 * ltrace is about to exit. Their job is to undo any effects that 157 * tracing had and eventually detach process, perhaps by way of 158 * installing a process handler. 159 * 160 * OS_LTRACE_EXITING_SIGHANDLER is called from a signal handler 161 * context right after the signal was captured. It returns 1 if the 162 * request was handled or 0 if it wasn't. 163 * 164 * If the call to OS_LTRACE_EXITING_SIGHANDLER didn't handle the 165 * request, OS_LTRACE_EXITING is called when the next event is 166 * generated. Therefore it's called in "safe" context, without 167 * re-entrancy concerns, but it's only called after an even is 168 * generated. */ 169int os_ltrace_exiting_sighandler(void); 170void os_ltrace_exiting(void); 171 172/* Should copy COUNT bytes from address ADDR of process PROC to local 173 * buffer BUF. */ 174size_t umovebytes(struct process *proc, void *addr, void *buf, size_t count); 175 176/* Find out an address of symbol SYM in process PROC, and return. 177 * Returning NULL delays breakpoint insertion and enables heaps of 178 * arch-specific black magic that we should clean up some day. 179 * 180 * XXX the same points as for get_instruction_pointer apply. */ 181void *sym2addr(struct process *proc, struct library_symbol *sym); 182 183/* Obtain address of PLT entry corresponding to relocation RELA in 184 * file LTE. This is NDX-th PLT entry in the file. 185 * 186 * XXX should this return arch_addr_t? */ 187GElf_Addr arch_plt_sym_val(struct ltelf *lte, size_t ndx, GElf_Rela *rela); 188 189/* Called at some point after we have attached to PROC. This callback 190 * should insert an introspection breakpoint for handling dynamic 191 * linker library loads. */ 192int linkmap_init(struct process *proc, arch_addr_t dyn_addr); 193 194/* This should produce and return the next event of one of the traced 195 * processes. The returned pointer will not be freed by the core and 196 * should be either statically allocated, or the management should be 197 * done some other way. */ 198struct Event *next_event(void); 199 200/* Called when process PROC was removed. */ 201void process_removed(struct process *proc); 202 203/* This should extract entry point address and interpreter (dynamic 204 * linker) bias if possible. Returns 0 if there were no errors, -1 205 * otherwise. Sets *ENTRYP and *INTERP_BIASP to non-zero values if 206 * the corresponding value is known, or zero otherwise; this is not 207 * done for pointers that are NULL. */ 208int process_get_entry(struct process *proc, 209 arch_addr_t *entryp, 210 arch_addr_t *interp_biasp); 211 212 213/* Optional callbacks 214 * 215 * Some callbacks are only available if backend (arch.h) has a certain 216 * define. If such a define is not present, default implementation 217 * (most often doing nothing at all) us used instead. This is used 218 * for gradual extensions of ltrace, so that backends that are not 219 * fully up to date, or that don't need certain functionality, keep 220 * working, while other backends take advantage of the optional 221 * features. */ 222 223/* The following callbacks have to be implemented in backend if arch.h 224 * defines ARCH_HAVE_LTELF_DATA. Those are used to init and destroy 225 * LTE->arch. arch_elf_init returns 0 on success or a negative value 226 * on failure. */ 227int arch_elf_init(struct ltelf *lte, struct library *lib); 228void arch_elf_destroy(struct ltelf *lte); 229 230/* The following callbacks have to be implemented in backend if arch.h 231 * defines ARCH_HAVE_BREAKPOINT_DATA. Those are used to init, 232 * destroy, and clone SBP->arch. arch_breakpoint_init and 233 * arch_breakpoint_clone return 0 on success or a negative value on 234 * failure. */ 235int arch_breakpoint_init(struct process *proc, struct breakpoint *sbp); 236void arch_breakpoint_destroy(struct breakpoint *sbp); 237int arch_breakpoint_clone(struct breakpoint *retp, struct breakpoint *sbp); 238 239/* The following callbacks have to be implemented in backend if arch.h 240 * defines ARCH_HAVE_LIBRARY_DATA. Those are used to init, destroy 241 * and clone LIB->arch. arch_library_init and arch_library_clone 242 * return 0 on success or a negative value on failure. */ 243int arch_library_init(struct library *lib); 244void arch_library_destroy(struct library *lib); 245int arch_library_clone(struct library *retp, struct library *lib); 246 247/* The following callbacks have to be implemented in backend if arch.h 248 * defines ARCH_HAVE_LIBRARY_SYMBOL_DATA. Those are used to init, 249 * destroy and clone LIBSYM->arch. arch_library_symbol_init and 250 * arch_library_symbol_clone return 0 on success or a negative value 251 * on failure. */ 252int arch_library_symbol_init(struct library_symbol *libsym); 253void arch_library_symbol_destroy(struct library_symbol *libsym); 254int arch_library_symbol_clone(struct library_symbol *retp, 255 struct library_symbol *libsym); 256 257/* The following callbacks have to be implemented in backend if arch.h 258 * defines ARCH_HAVE_PROCESS_DATA. Those are used to init, destroy 259 * and clone PROC->arch. arch_process_exec is called to update 260 * PROC->arch in case that PROC underwent an exec. See notes at 261 * process_init, process_destroy, process_clone and process_exec in 262 * proc.h. */ 263int arch_process_init(struct process *proc); 264void arch_process_destroy(struct process *proc); 265int arch_process_clone(struct process *retp, struct process *proc); 266int arch_process_exec(struct process *proc); 267 268/* The following callbacks have to be implemented in OS backend if 269 * os.h defines OS_HAVE_PROCESS_DATA. The protocol is same as for, 270 * respectively, arch_process_init, arch_process_destroy, 271 * arch_process_clone and arch_process_exec. */ 272int os_process_init(struct process *proc); 273void os_process_destroy(struct process *proc); 274int os_process_clone(struct process *retp, struct process *proc); 275int os_process_exec(struct process *proc); 276 277/* The following callback has to be implemented in backend if arch.h 278 * defines ARCH_HAVE_GET_SYM_INFO. 279 * 280 * This is called for every PLT relocation RELA in ELF file LTE (which 281 * is named FILENAME), that ltrace is about to add. The corresponding 282 * PLT entry is for SYM_INDEX-th relocation in the file. This call is 283 * supposed to initialize SYM and RELA. It returns 0 if there were no 284 * errors and given symbol should be used, 1 if the symbol should not 285 * be used, or a negative value if there were errors. 286 * 287 * The backend implementation can delegate some of the work to default 288 * implementation in elf_get_sym_info. */ 289int arch_get_sym_info(struct ltelf *lte, const char *filename, size_t sym_index, 290 GElf_Rela *rela, GElf_Sym *sym); 291 292enum plt_status { 293 PLT_FAIL, 294 PLT_OK, 295 PLT_DEFAULT, 296}; 297 298/* The following callback has to be implemented in backend if arch.h 299 * defines ARCH_HAVE_ADD_PLT_ENTRY. 300 * 301 * This is called for every PLT relocation R in ELF file LTE, that 302 * ltrace is about to add to a library constructed in process PROC. 303 * The corresponding PLT entry is for symbol called NAME, and it's 304 * I-th relocation in the file. 305 * 306 * If this function returns PLT_DEFAULT, PLT address is obtained by 307 * calling arch_plt_sym_val, and symbol is allocated. If PLT_OK or 308 * PLT_DEFAULT are returned, the chain of symbols passed back in RET 309 * is added to library under construction. */ 310enum plt_status arch_elf_add_plt_entry(struct process *proc, struct ltelf *lte, 311 const char *name, GElf_Rela *rela, 312 size_t i, struct library_symbol **ret); 313 314/* This callback needs to be implemented if arch.h defines 315 * ARCH_HAVE_DYNLINK_DONE. It is called after the dynamic linker is 316 * done with the process start-up. */ 317void arch_dynlink_done(struct process *proc); 318 319/* This callback needs to be implemented if arch.h defines 320 * ARCH_HAVE_SYMBOL_RET. It is called after a traced call returns. */ 321void arch_symbol_ret(struct process *proc, struct library_symbol *libsym); 322 323 324/* This callback needs to be implemented if arch.h defines 325 * ARCH_HAVE_FIND_DL_DEBUG. 326 * It is called by generic code to find the address of the dynamic 327 * linkers debug structure. 328 * DYN_ADDR holds the address of the dynamic section. 329 * If the debug area is found, return 0 and fill in the address in *RET. 330 * If the debug area is not found, return a negative value. */ 331int arch_find_dl_debug(struct process *proc, arch_addr_t dyn_addr, 332 arch_addr_t *ret); 333 334/* This is called to obtain a list of directories to search when 335 * loading config files. The callback sets *RETP to a pointer to the 336 * first element of a NULL-terminated array of directory names. It's 337 * legitimate to set *RETP to NULL to indicate there are no 338 * directories. The function returns 0 on success or a negative value 339 * on a failure. 340 * 341 * If PRIVATE is set, the list in *RETP should contain only user's own 342 * directories (presumably under HOME if there's any such thing on the 343 * given OS). Otherwise only system directories should be reported. 344 * 345 * The directories don't have to exist. Directories passed in -F are 346 * handled separately by the caller and this callback shouldn't 347 * concern itself with it. */ 348int os_get_config_dirs(int private, const char ***retp); 349 350/* This is called to obtain list of legacy config files to import, if 351 * any. A reference to initialized vector of char* is passed in. 352 * 353 * This returns 0 on success, in which case strings from *RETP (if 354 * any) are interpreted as files names. These files belong to the 355 * caller and will eventually be freed. 356 * 357 * Returns a negative value for failure, in which case *RETP contents 358 * are not consulted in any way. */ 359int os_get_ltrace_conf_filenames(struct vect *retp); 360 361/* If arch.h defines ARCH_HAVE_FETCH_ARG, the following callbacks have 362 * to be implemented: arch_fetch_arg_init, arch_fetch_arg_clone, 363 * arch_fetch_arg_done, arch_fetch_arg_next and arch_fetch_retval. 364 * See fetch.h for details. */ 365 366/* If arch.h defines both ARCH_HAVE_FETCH_ARG and 367 * ARCH_HAVE_FETCH_PACK, the following callbacks have to be 368 * implemented: arch_fetch_param_pack_start, 369 * arch_fetch_param_pack_end. See fetch.h for details. */ 370 371enum sw_singlestep_status { 372 SWS_FAIL, 373 SWS_OK, 374 SWS_HW, 375}; 376struct sw_singlestep_data; 377 378/* The following callback has to be implemented in backend if arch.h 379 * defines ARCH_HAVE_SW_SINGLESTEP. 380 * 381 * This is called before the OS backend requests hardware singlestep. 382 * arch_sw_singlestep should consider whether a singlestep needs to be 383 * done in software. If not, it returns SWS_HW. Otherwise it needs 384 * to add one or several breakpoints by calling ADD_CB. When it is 385 * done, it continues the process as appropriate, and answers either 386 * SWS_OK, or SWS_FAIL, depending on how it went. 387 * 388 * PROC is the process that should perform the singlestep, BP the 389 * breakpoint that we are singlestepping over. ADD_CB is a callback 390 * to request adding breakpoints that should trap the process after 391 * it's continued. The arguments to ADD_CB are the address where the 392 * breakpoint should be added, and DATA. ADD_CB returns 0 on success 393 * or a negative value on failure. It is expected that 394 * arch_sw_singlestep returns SWS_FAIL if ADD_CB returns error. */ 395enum sw_singlestep_status arch_sw_singlestep(struct process *proc, 396 struct breakpoint *bp, 397 int (*add_cb)(arch_addr_t addr, 398 struct sw_singlestep_data *), 399 struct sw_singlestep_data *data); 400 401#endif /* BACKEND_H */ 402