uprobe/arch/x86/swap-asm/swap_uprobes.c

   1 /**
   2  * uprobe/arch/asm-x86/swap_uprobes.c
   3  * @author Alexey Gerenkov <a.gerenkov@samsung.com> User-Space Probes initial
   4  * implementation; Support x86/ARM/MIPS for both user and kernel spaces.
   5  * @author Ekaterina Gorelkina <e.gorelkina@samsung.com>: redesign module for
   6  * separating core and arch parts
   7  *
   8  * @section LICENSE
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  23  *
  24  * @section COPYRIGHT
  25  *
  26  * Copyright (C) Samsung Electronics, 2006-2010
  27  *
  28  * @section DESCRIPTION
  29  *
  30  * Arch-dependent uprobe interface implementation for x86.
  31  */
  32
  33
  34 #include <linux/kdebug.h>
  35
  36 #include <kprobe/swap_slots.h>
  37 #include <uprobe/swap_uprobes.h>
  38
  39 #include "swap_uprobes.h"
  40
  41
  42 /**
  43  * @struct uprobe_ctlblk
  44  * @brief Uprobe control block
  45  */
  46 struct uprobe_ctlblk {
  47         unsigned long flags;            /**< Flags */
  48         struct kprobe *p;               /**< Pointer to the uprobe's kprobe */
  49 };
  50
  51 static unsigned long trampoline_addr(struct uprobe *up)
  52 {
  53         return (unsigned long)(up->kp.ainsn.insn +
  54                                UPROBES_TRAMP_RET_BREAK_IDX);
  55 }
  56
  57 unsigned long arch_tramp_by_ri(struct uretprobe_instance *ri)
  58 {
  59         return trampoline_addr(&ri->rp->up);
  60 }
  61
  62 static struct uprobe_ctlblk *current_ucb(void)
  63 {
  64         /* FIXME hardcoded offset */
  65         return (struct uprobe_ctlblk *)(end_of_stack(current) + 20);
  66 }
  67
  68 static struct kprobe *get_current_probe(void)
  69 {
  70         return current_ucb()->p;
  71 }
  72
  73 static void set_current_probe(struct kprobe *p)
  74 {
  75         current_ucb()->p = p;
  76 }
  77
  78 static void save_current_flags(struct pt_regs *regs)
  79 {
  80         current_ucb()->flags = regs->flags;
  81 }
  82
  83 static void restore_current_flags(struct pt_regs *regs, unsigned long flags)
  84 {
  85         regs->flags &= ~IF_MASK;
  86         regs->flags |= flags & IF_MASK;
  87 }
  88
  89 /**
  90  * @brief Prepares uprobe for x86.
  91  *
  92  * @param up Pointer to the uprobe.
  93  * @return 0 on success,\n
  94  * -1 on error.
  95  */
  96 int arch_prepare_uprobe(struct uprobe *up)
  97 {
  98         struct kprobe *p = up2kp(up);
  99         struct task_struct *task = up->task;
 100         u8 *tramp = up->atramp.tramp;
 101         enum { call_relative_opcode = 0xe8 };
 102
 103         if (!read_proc_vm_atomic(task, (unsigned long)p->addr,
 104                                  tramp, MAX_INSN_SIZE)) {
 105                 printk(KERN_ERR "failed to read memory %p!\n", p->addr);
 106                 return -EINVAL;
 107         }
 108         /* TODO: this is a workaround */
 109         if (tramp[0] == call_relative_opcode) {
 110                 printk(KERN_INFO "cannot install probe: 1st instruction is call\n");
 111                 return -EINVAL;
 112         }
 113
 114         tramp[UPROBES_TRAMP_RET_BREAK_IDX] = BREAKPOINT_INSTRUCTION;
 115
 116         /* TODO: remove dual info */
 117         p->opcode = tramp[0];
 118
 119         p->ainsn.boostable = swap_can_boost(tramp) ? 0 : -1;
 120
 121         p->ainsn.insn = swap_slot_alloc(up->sm);
 122         if (p->ainsn.insn == NULL) {
 123                 printk(KERN_ERR "trampoline out of memory\n");
 124                 return -ENOMEM;
 125         }
 126
 127         if (!write_proc_vm_atomic(task, (unsigned long)p->ainsn.insn,
 128                                   tramp, sizeof(up->atramp.tramp))) {
 129                 swap_slot_free(up->sm, p->ainsn.insn);
 130                 printk(KERN_INFO "failed to write memory %p!\n", tramp);
 131                 return -EINVAL;
 132         }
 133
 134         /* for uretprobe */
 135         add_uprobe_table(p);
 136
 137         return 0;
 138 }
 139
 140 /**
 141  * @brief Jump pre-handler.
 142  *
 143  * @param p Pointer to the uprobe's kprobe.
 144  * @param regs Pointer to CPU register data.
 145  * @return 0.
 146  */
 147 int setjmp_upre_handler(struct kprobe *p, struct pt_regs *regs)
 148 {
 149         struct uprobe *up = container_of(p, struct uprobe, kp);
 150         struct ujprobe *jp = container_of(up, struct ujprobe, up);
 151         kprobe_pre_entry_handler_t pre_entry =
 152                 (kprobe_pre_entry_handler_t)jp->pre_entry;
 153         entry_point_t entry = (entry_point_t)jp->entry;
 154         unsigned long args[6];
 155
 156         /* FIXME some user space apps crash if we clean interrupt bit */
 157         /* regs->EREG(flags) &= ~IF_MASK; */
 158 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2, 6, 18)
 159         trace_hardirqs_off();
 160 #endif
 161
 162         /* read first 6 args from stack */
 163         if (!read_proc_vm_atomic(current, regs->EREG(sp) + 4,
 164                                  args, sizeof(args)))
 165                 printk(KERN_WARNING
 166                        "failed to read user space func arguments %lx!\n",
 167                        regs->sp + 4);
 168
 169         if (pre_entry)
 170                 p->ss_addr[smp_processor_id()] = (kprobe_opcode_t *)
 171                                                  pre_entry(jp->priv_arg, regs);
 172
 173         if (entry)
 174                 entry(args[0], args[1], args[2], args[3], args[4], args[5]);
 175         else
 176                 arch_ujprobe_return();
 177
 178         return 0;
 179 }
 180
 181 /**
 182  * @brief Prepares uretprobe for x86.
 183  *
 184  * @param ri Pointer to the uretprobe instance.
 185  * @param regs Pointer to CPU register data.
 186  * @return Void.
 187  */
 188 int arch_prepare_uretprobe(struct uretprobe_instance *ri, struct pt_regs *regs)
 189 {
 190         /* Replace the return addr with trampoline addr */
 191         unsigned long ra = trampoline_addr(&ri->rp->up);
 192         unsigned long ret_addr;
 193         ri->sp = (kprobe_opcode_t *)regs->sp;
 194
 195         if (get_user(ret_addr, (unsigned long *)regs->sp)) {
 196                 pr_err("failed to read user space func ra %lx addr=%p!\n",
 197                        regs->sp, ri->rp->up.kp.addr);
 198                 return -EINVAL;
 199         }
 200
 201         if (put_user(ra, (unsigned long *)regs->sp)) {
 202                 pr_err("failed to write user space func ra %lx!\n", regs->sp);
 203                 return -EINVAL;
 204         }
 205
 206         ri->ret_addr = (kprobe_opcode_t *)ret_addr;
 207
 208         return 0;
 209 }
 210
 211 static bool get_long(struct task_struct *task,
 212                      unsigned long vaddr, unsigned long *val)
 213 {
 214         return task->mm == current->mm ?
 215                 !!get_user(*val, (unsigned long *)vaddr) :
 216                 sizeof(*val) != read_proc_vm_atomic(task, vaddr,
 217                                                     val, sizeof(*val));
 218 }
 219
 220 static bool put_long(struct task_struct *task,
 221                      unsigned long vaddr, unsigned long *val)
 222 {
 223         return task->mm == current->mm ?
 224                 !!put_user(*val, (unsigned long *)vaddr) :
 225                 sizeof(*val) != write_proc_vm_atomic(task, vaddr,
 226                                                      val, sizeof(*val));
 227 }
 228
 229 /**
 230  * @brief Disarms uretprobe on x86 arch.
 231  *
 232  * @param ri Pointer to the uretprobe instance.
 233  * @param task Pointer to the task for which the probe.
 234  * @return 0 on success,\n
 235  * negative error code on error.
 236  */
 237 int arch_disarm_urp_inst(struct uretprobe_instance *ri,
 238                          struct task_struct *task, unsigned long tr)
 239 {
 240         unsigned long ret_addr;
 241         unsigned long sp = (unsigned long)ri->sp;
 242         unsigned long tramp_addr;
 243
 244         if (tr == 0)
 245                 tramp_addr = arch_tramp_by_ri(ri);
 246         else
 247                 tramp_addr = tr; /* ri - invalid */
 248
 249         if (get_long(task, sp, &ret_addr)) {
 250                 printk(KERN_INFO "---> %s (%d/%d): failed to read stack from %08lx\n",
 251                        task->comm, task->tgid, task->pid, sp);
 252                 return -EFAULT;
 253         }
 254
 255         if (tramp_addr == ret_addr) {
 256                 if (put_long(task, sp, (unsigned long *)&ri->ret_addr)) {
 257                         printk(KERN_INFO "---> %s (%d/%d): failed to write "
 258                                "orig_ret_addr to %08lx",
 259                                task->comm, task->tgid, task->pid, sp);
 260                         return -EFAULT;
 261                 }
 262         } else {
 263                 printk(KERN_INFO "---> %s (%d/%d): trampoline NOT found at sp = %08lx\n",
 264                        task->comm, task->tgid, task->pid, sp);
 265                 return -ENOENT;
 266         }
 267
 268         return 0;
 269 }
 270
 271 /**
 272  * @brief Gets trampoline address.
 273  *
 274  * @param p Pointer to the uprobe's kprobe.
 275  * @param regs Pointer to CPU register data.
 276  * @return Trampoline address.
 277  */
 278 unsigned long arch_get_trampoline_addr(struct kprobe *p, struct pt_regs *regs)
 279 {
 280         return trampoline_addr(kp2up(p));
 281 }
 282
 283 /**
 284  * @brief Restores return address.
 285  *
 286  * @param orig_ret_addr Original return address.
 287  * @param regs Pointer to CPU register data.
 288  * @return Void.
 289  */
 290 void arch_set_orig_ret_addr(unsigned long orig_ret_addr, struct pt_regs *regs)
 291 {
 292         regs->EREG(ip) = orig_ret_addr;
 293 }
 294
 295 /**
 296  * @brief Removes uprobe.
 297  *
 298  * @param up Pointer to the target uprobe.
 299  * @return Void.
 300  */
 301 void arch_remove_uprobe(struct uprobe *up)
 302 {
 303         struct kprobe *p = up2kp(up);
 304
 305         swap_slot_free(up->sm, p->ainsn.insn);
 306 }
 307
 308 int arch_arm_uprobe(struct uprobe *p)
 309 {
 310         int ret;
 311         kprobe_opcode_t insn = BREAKPOINT_INSTRUCTION;
 312         unsigned long vaddr = (unsigned long)p->kp.addr;
 313
 314         ret = write_proc_vm_atomic(p->task, vaddr, &insn, sizeof(insn));
 315         if (!ret) {
 316                 pr_err("arch_arm_uprobe: failed to write memory tgid=%u vaddr=%08lx\n",
 317                        p->task->tgid, vaddr);
 318
 319                 return -EACCES;
 320         }
 321
 322         return 0;
 323 }
 324
 325 void arch_disarm_uprobe(struct kprobe *p, struct task_struct *task)
 326 {
 327         int ret;
 328         unsigned long vaddr = (unsigned long)p->addr;
 329
 330         ret = write_proc_vm_atomic(task, vaddr, &p->opcode, sizeof(p->opcode));
 331         if (!ret) {
 332                 pr_err("arch_disarm_uprobe: failed to write memory tgid=%u, vaddr=%08lx\n",
 333                        task->tgid, vaddr);
 334         }
 335 }
 336
 337 static void set_user_jmp_op(void *from, void *to)
 338 {
 339         struct __arch_jmp_op {
 340                 char op;
 341                 long raddr;
 342         } __packed jop;
 343
 344         jop.raddr = (long)(to) - ((long)(from) + 5);
 345         jop.op = RELATIVEJUMP_INSTRUCTION;
 346
 347         if (put_user(jop.op, (char *)from) ||
 348             put_user(jop.raddr, (long *)(from + 1)))
 349                 pr_err("failed to write jump opcode to user space %p\n", from);
 350 }
 351
 352 static void resume_execution(struct kprobe *p,
 353                              struct pt_regs *regs,
 354                              unsigned long flags)
 355 {
 356         unsigned long *tos, tos_dword = 0;
 357         unsigned long copy_eip = (unsigned long)p->ainsn.insn;
 358         unsigned long orig_eip = (unsigned long)p->addr;
 359         kprobe_opcode_t insns[2];
 360
 361         regs->EREG(flags) &= ~TF_MASK;
 362
 363         tos = (unsigned long *)&tos_dword;
 364         if (get_user(tos_dword, (unsigned long *)regs->sp)) {
 365                 pr_err("failed to read from user space sp=%lx!\n", regs->sp);
 366                 return;
 367         }
 368
 369         if (get_user(*(unsigned short *)insns, (unsigned short *)p->ainsn.insn)) {
 370                 pr_err("failed to read first 2 opcodes %p!\n", p->ainsn.insn);
 371                 return;
 372         }
 373
 374         switch (insns[0]) {
 375         case 0x9c: /* pushfl */
 376                 *tos &= ~(TF_MASK | IF_MASK);
 377                 *tos |= flags & (TF_MASK | IF_MASK);
 378                 break;
 379         case 0xc2: /* iret/ret/lret */
 380         case 0xc3:
 381         case 0xca:
 382         case 0xcb:
 383         case 0xcf:
 384         case 0xea: /* jmp absolute -- eip is correct */
 385                 /* eip is already adjusted, no more changes required */
 386                 p->ainsn.boostable = 1;
 387                 goto no_change;
 388         case 0xe8: /* call relative - Fix return addr */
 389                 *tos = orig_eip + (*tos - copy_eip);
 390                 break;
 391         case 0x9a: /* call absolute -- same as call absolute, indirect */
 392                 *tos = orig_eip + (*tos - copy_eip);
 393
 394                 if (put_user(tos_dword, (unsigned long *)regs->sp)) {
 395                         pr_err("failed to write dword to sp=%lx\n", regs->sp);
 396                         return;
 397                 }
 398
 399                 goto no_change;
 400         case 0xff:
 401                 if ((insns[1] & 0x30) == 0x10) {
 402                         /*
 403                          * call absolute, indirect
 404                          * Fix return addr; eip is correct.
 405                          * But this is not boostable
 406                          */
 407                         *tos = orig_eip + (*tos - copy_eip);
 408
 409                         if (put_user(tos_dword, (unsigned long *)regs->sp)) {
 410                                 pr_err("failed to write dword to sp=%lx\n", regs->sp);
 411                                 return;
 412                         }
 413
 414                         goto no_change;
 415                 } else if (((insns[1] & 0x31) == 0x20) || /* jmp near, absolute
 416                                                            * indirect */
 417                            ((insns[1] & 0x31) == 0x21)) {
 418                         /* jmp far, absolute indirect */
 419                         /* eip is correct. And this is boostable */
 420                         p->ainsn.boostable = 1;
 421                         goto no_change;
 422                 }
 423         case 0xf3:
 424                 if (insns[1] == 0xc3)
 425                         /* repz ret special handling: no more changes */
 426                         goto no_change;
 427                 break;
 428         default:
 429                 break;
 430         }
 431
 432         if (put_user(tos_dword, (unsigned long *)regs->sp)) {
 433                 pr_err("failed to write dword to sp=%lx\n", regs->sp);
 434                 return;
 435         }
 436
 437         if (p->ainsn.boostable == 0) {
 438                 if ((regs->EREG(ip) > copy_eip) && (regs->EREG(ip) - copy_eip) +
 439                     5 < MAX_INSN_SIZE) {
 440                         /*
 441                          * These instructions can be executed directly if it
 442                          * jumps back to correct address.
 443                          */
 444                         set_user_jmp_op((void *) regs->EREG(ip),
 445                                         (void *)orig_eip +
 446                                         (regs->EREG(ip) - copy_eip));
 447                         p->ainsn.boostable = 1;
 448                 } else {
 449                         p->ainsn.boostable = -1;
 450                 }
 451         }
 452
 453         regs->EREG(ip) = orig_eip + (regs->EREG(ip) - copy_eip);
 454
 455 no_change:
 456         return;
 457 }
 458
 459 static bool prepare_ss_addr(struct kprobe *p, struct pt_regs *regs)
 460 {
 461         unsigned long *ss_addr = (long *)&p->ss_addr[smp_processor_id()];
 462
 463         if (*ss_addr) {
 464                 regs->ip = *ss_addr;
 465                 *ss_addr = 0;
 466                 return true;
 467         } else {
 468                 regs->ip = (unsigned long)p->ainsn.insn;
 469                 return false;
 470         }
 471 }
 472
 473 static void prepare_ss(struct pt_regs *regs)
 474 {
 475         /* set single step mode */
 476         regs->flags |= TF_MASK;
 477         regs->flags &= ~IF_MASK;
 478 }
 479
 480 static int uprobe_handler(struct pt_regs *regs)
 481 {
 482         struct kprobe *p;
 483         kprobe_opcode_t *addr;
 484         struct task_struct *task = current;
 485         pid_t tgid = task->tgid;
 486
 487         save_current_flags(regs);
 488
 489         addr = (kprobe_opcode_t *)(regs->EREG(ip) - sizeof(kprobe_opcode_t));
 490         p = get_ukprobe(addr, tgid);
 491
 492         if (p == NULL) {
 493                 void *tramp_addr = (void *)addr - UPROBES_TRAMP_RET_BREAK_IDX;
 494
 495                 p = get_ukprobe_by_insn_slot(tramp_addr, tgid, regs);
 496                 if (p == NULL) {
 497                         printk(KERN_INFO "no_uprobe\n");
 498                         return 0;
 499                 }
 500
 501                 trampoline_uprobe_handler(p, regs);
 502                 return 1;
 503         } else {
 504                 if (!p->pre_handler || !p->pre_handler(p, regs)) {
 505                         if (p->ainsn.boostable == 1 && !p->post_handler) {
 506                                 prepare_ss_addr(p, regs);
 507                                 return 1;
 508                         }
 509
 510                         if (prepare_ss_addr(p, regs) == false) {
 511                                 set_current_probe(p);
 512                                 prepare_ss(regs);
 513                         }
 514                 }
 515         }
 516
 517         return 1;
 518 }
 519
 520 static int post_uprobe_handler(struct pt_regs *regs)
 521 {
 522         struct kprobe *p = get_current_probe();
 523         unsigned long flags = current_ucb()->flags;
 524
 525         if (p == NULL) {
 526                 printk("task[%u %u %s] current uprobe is not found\n",
 527                        current->tgid, current->pid, current->comm);
 528                 return 0;
 529         }
 530
 531         resume_execution(p, regs, flags);
 532         restore_current_flags(regs, flags);
 533
 534         /* clean stack */
 535         current_ucb()->p = 0;
 536         current_ucb()->flags = 0;
 537
 538         return 1;
 539 }
 540
 541 static int uprobe_exceptions_notify(struct notifier_block *self,
 542                                     unsigned long val, void *data)
 543 {
 544         struct die_args *args = (struct die_args *)data;
 545         int ret = NOTIFY_DONE;
 546
 547         if (args->regs == NULL || !user_mode_vm(args->regs))
 548                 return ret;
 549
 550         switch (val) {
 551 #ifdef CONFIG_KPROBES
 552         case DIE_INT3:
 553 #else
 554         case DIE_TRAP:
 555 #endif
 556                 if (uprobe_handler(args->regs))
 557                         ret = NOTIFY_STOP;
 558                 break;
 559         case DIE_DEBUG:
 560                 if (post_uprobe_handler(args->regs))
 561                         ret = NOTIFY_STOP;
 562                 break;
 563         default:
 564                 break;
 565         }
 566
 567         return ret;
 568 }
 569
 570 static struct notifier_block uprobe_exceptions_nb = {
 571         .notifier_call = uprobe_exceptions_notify,
 572         .priority = INT_MAX
 573 };
 574
 575 /**
 576  * @brief Registers notify.
 577  *
 578  * @return register_die_notifier result.
 579  */
 580 int swap_arch_init_uprobes(void)
 581 {
 582         return register_die_notifier(&uprobe_exceptions_nb);
 583 }
 584
 585 /**
 586  * @brief Unregisters notify.
 587  *
 588  * @return Void.
 589  */
 590 void swap_arch_exit_uprobes(void)
 591 {
 592         unregister_die_notifier(&uprobe_exceptions_nb);
 593 }
 594