arch/i386/kernel/kprobes.c

   1 /*
   2  *  Kernel Probes (KProbes)
   3  *  arch/i386/kernel/kprobes.c
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU 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., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18  *
  19  * Copyright (C) IBM Corporation, 2002, 2004
  20  *
  21  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
  22  *              Probes initial implementation ( includes contributions from
  23  *              Rusty Russell).
  24  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  25  *              interface to access function arguments.
  26  */
  27
  28 #include <linux/config.h>
  29 #include <linux/kprobes.h>
  30 #include <linux/ptrace.h>
  31 #include <linux/spinlock.h>
  32 #include <linux/preempt.h>
  33 #include <asm/kdebug.h>
  34 #include <asm/desc.h>
  35
  36 /* kprobe_status settings */
  37 #define KPROBE_HIT_ACTIVE       0x00000001
  38 #define KPROBE_HIT_SS           0x00000002
  39
  40 static struct kprobe *current_kprobe;
  41 static unsigned long kprobe_status, kprobe_old_eflags, kprobe_saved_eflags;
  42 static struct pt_regs jprobe_saved_regs;
  43 static long *jprobe_saved_esp;
  44 /* copy of the kernel stack at the probe fire time */
  45 static kprobe_opcode_t jprobes_stack[MAX_STACK_SIZE];
  46 void jprobe_return_end(void);
  47
  48 /*
  49  * returns non-zero if opcode modifies the interrupt flag.
  50  */
  51 static inline int is_IF_modifier(kprobe_opcode_t opcode)
  52 {
  53         switch (opcode) {
  54         case 0xfa:              /* cli */
  55         case 0xfb:              /* sti */
  56         case 0xcf:              /* iret/iretd */
  57         case 0x9d:              /* popf/popfd */
  58                 return 1;
  59         }
  60         return 0;
  61 }
  62
  63 int arch_prepare_kprobe(struct kprobe *p)
  64 {
  65         return 0;
  66 }
  67
  68 void arch_copy_kprobe(struct kprobe *p)
  69 {
  70         memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
  71 }
  72
  73 void arch_remove_kprobe(struct kprobe *p)
  74 {
  75 }
  76
  77 static inline void disarm_kprobe(struct kprobe *p, struct pt_regs *regs)
  78 {
  79         *p->addr = p->opcode;
  80         regs->eip = (unsigned long)p->addr;
  81 }
  82
  83 static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
  84 {
  85         regs->eflags |= TF_MASK;
  86         regs->eflags &= ~IF_MASK;
  87         /*single step inline if the instruction is an int3*/
  88         if (p->opcode == BREAKPOINT_INSTRUCTION)
  89                 regs->eip = (unsigned long)p->addr;
  90         else
  91                 regs->eip = (unsigned long)&p->ainsn.insn;
  92 }
  93
  94 /*
  95  * Interrupts are disabled on entry as trap3 is an interrupt gate and they
  96  * remain disabled thorough out this function.
  97  */
  98 static int kprobe_handler(struct pt_regs *regs)
  99 {
 100         struct kprobe *p;
 101         int ret = 0;
 102         kprobe_opcode_t *addr = NULL;
 103         unsigned long *lp;
 104
 105         /* We're in an interrupt, but this is clear and BUG()-safe. */
 106         preempt_disable();
 107         /* Check if the application is using LDT entry for its code segment and
 108          * calculate the address by reading the base address from the LDT entry.
 109          */
 110         if ((regs->xcs & 4) && (current->mm)) {
 111                 lp = (unsigned long *) ((unsigned long)((regs->xcs >> 3) * 8)
 112                                         + (char *) current->mm->context.ldt);
 113                 addr = (kprobe_opcode_t *) (get_desc_base(lp) + regs->eip -
 114                                                 sizeof(kprobe_opcode_t));
 115         } else {
 116                 addr = (kprobe_opcode_t *)(regs->eip - sizeof(kprobe_opcode_t));
 117         }
 118         /* Check we're not actually recursing */
 119         if (kprobe_running()) {
 120                 /* We *are* holding lock here, so this is safe.
 121                    Disarm the probe we just hit, and ignore it. */
 122                 p = get_kprobe(addr);
 123                 if (p) {
 124                         if (kprobe_status == KPROBE_HIT_SS) {
 125                                 regs->eflags &= ~TF_MASK;
 126                                 regs->eflags |= kprobe_saved_eflags;
 127                                 unlock_kprobes();
 128                                 goto no_kprobe;
 129                         }
 130                         disarm_kprobe(p, regs);
 131                         ret = 1;
 132                 } else {
 133                         p = current_kprobe;
 134                         if (p->break_handler && p->break_handler(p, regs)) {
 135                                 goto ss_probe;
 136                         }
 137                 }
 138                 /* If it's not ours, can't be delete race, (we hold lock). */
 139                 goto no_kprobe;
 140         }
 141
 142         lock_kprobes();
 143         p = get_kprobe(addr);
 144         if (!p) {
 145                 unlock_kprobes();
 146                 if (regs->eflags & VM_MASK) {
 147                         /* We are in virtual-8086 mode. Return 0 */
 148                         goto no_kprobe;
 149                 }
 150
 151                 if (*addr != BREAKPOINT_INSTRUCTION) {
 152                         /*
 153                          * The breakpoint instruction was removed right
 154                          * after we hit it.  Another cpu has removed
 155                          * either a probepoint or a debugger breakpoint
 156                          * at this address.  In either case, no further
 157                          * handling of this interrupt is appropriate.
 158                          */
 159                         ret = 1;
 160                 }
 161                 /* Not one of ours: let kernel handle it */
 162                 goto no_kprobe;
 163         }
 164
 165         kprobe_status = KPROBE_HIT_ACTIVE;
 166         current_kprobe = p;
 167         kprobe_saved_eflags = kprobe_old_eflags
 168             = (regs->eflags & (TF_MASK | IF_MASK));
 169         if (is_IF_modifier(p->opcode))
 170                 kprobe_saved_eflags &= ~IF_MASK;
 171
 172         if (p->pre_handler && p->pre_handler(p, regs))
 173                 /* handler has already set things up, so skip ss setup */
 174                 return 1;
 175
 176 ss_probe:
 177         prepare_singlestep(p, regs);
 178         kprobe_status = KPROBE_HIT_SS;
 179         return 1;
 180
 181 no_kprobe:
 182         preempt_enable_no_resched();
 183         return ret;
 184 }
 185
 186 /*
 187  * Called after single-stepping.  p->addr is the address of the
 188  * instruction whose first byte has been replaced by the "int 3"
 189  * instruction.  To avoid the SMP problems that can occur when we
 190  * temporarily put back the original opcode to single-step, we
 191  * single-stepped a copy of the instruction.  The address of this
 192  * copy is p->ainsn.insn.
 193  *
 194  * This function prepares to return from the post-single-step
 195  * interrupt.  We have to fix up the stack as follows:
 196  *
 197  * 0) Except in the case of absolute or indirect jump or call instructions,
 198  * the new eip is relative to the copied instruction.  We need to make
 199  * it relative to the original instruction.
 200  *
 201  * 1) If the single-stepped instruction was pushfl, then the TF and IF
 202  * flags are set in the just-pushed eflags, and may need to be cleared.
 203  *
 204  * 2) If the single-stepped instruction was a call, the return address
 205  * that is atop the stack is the address following the copied instruction.
 206  * We need to make it the address following the original instruction.
 207  */
 208 static void resume_execution(struct kprobe *p, struct pt_regs *regs)
 209 {
 210         unsigned long *tos = (unsigned long *)&regs->esp;
 211         unsigned long next_eip = 0;
 212         unsigned long copy_eip = (unsigned long)&p->ainsn.insn;
 213         unsigned long orig_eip = (unsigned long)p->addr;
 214
 215         switch (p->ainsn.insn[0]) {
 216         case 0x9c:              /* pushfl */
 217                 *tos &= ~(TF_MASK | IF_MASK);
 218                 *tos |= kprobe_old_eflags;
 219                 break;
 220         case 0xc3:              /* ret/lret */
 221         case 0xcb:
 222         case 0xc2:
 223         case 0xca:
 224                 regs->eflags &= ~TF_MASK;
 225                 /* eip is already adjusted, no more changes required*/
 226                 return;
 227         case 0xe8:              /* call relative - Fix return addr */
 228                 *tos = orig_eip + (*tos - copy_eip);
 229                 break;
 230         case 0xff:
 231                 if ((p->ainsn.insn[1] & 0x30) == 0x10) {
 232                         /* call absolute, indirect */
 233                         /* Fix return addr; eip is correct. */
 234                         next_eip = regs->eip;
 235                         *tos = orig_eip + (*tos - copy_eip);
 236                 } else if (((p->ainsn.insn[1] & 0x31) == 0x20) ||       /* jmp near, absolute indirect */
 237                            ((p->ainsn.insn[1] & 0x31) == 0x21)) {       /* jmp far, absolute indirect */
 238                         /* eip is correct. */
 239                         next_eip = regs->eip;
 240                 }
 241                 break;
 242         case 0xea:              /* jmp absolute -- eip is correct */
 243                 next_eip = regs->eip;
 244                 break;
 245         default:
 246                 break;
 247         }
 248
 249         regs->eflags &= ~TF_MASK;
 250         if (next_eip) {
 251                 regs->eip = next_eip;
 252         } else {
 253                 regs->eip = orig_eip + (regs->eip - copy_eip);
 254         }
 255 }
 256
 257 /*
 258  * Interrupts are disabled on entry as trap1 is an interrupt gate and they
 259  * remain disabled thoroughout this function.  And we hold kprobe lock.
 260  */
 261 static inline int post_kprobe_handler(struct pt_regs *regs)
 262 {
 263         if (!kprobe_running())
 264                 return 0;
 265
 266         if (current_kprobe->post_handler)
 267                 current_kprobe->post_handler(current_kprobe, regs, 0);
 268
 269         resume_execution(current_kprobe, regs);
 270         regs->eflags |= kprobe_saved_eflags;
 271
 272         unlock_kprobes();
 273         preempt_enable_no_resched();
 274
 275         /*
 276          * if somebody else is singlestepping across a probe point, eflags
 277          * will have TF set, in which case, continue the remaining processing
 278          * of do_debug, as if this is not a probe hit.
 279          */
 280         if (regs->eflags & TF_MASK)
 281                 return 0;
 282
 283         return 1;
 284 }
 285
 286 /* Interrupts disabled, kprobe_lock held. */
 287 static inline int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
 288 {
 289         if (current_kprobe->fault_handler
 290             && current_kprobe->fault_handler(current_kprobe, regs, trapnr))
 291                 return 1;
 292
 293         if (kprobe_status & KPROBE_HIT_SS) {
 294                 resume_execution(current_kprobe, regs);
 295                 regs->eflags |= kprobe_old_eflags;
 296
 297                 unlock_kprobes();
 298                 preempt_enable_no_resched();
 299         }
 300         return 0;
 301 }
 302
 303 /*
 304  * Wrapper routine to for handling exceptions.
 305  */
 306 int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val,
 307                              void *data)
 308 {
 309         struct die_args *args = (struct die_args *)data;
 310         switch (val) {
 311         case DIE_INT3:
 312                 if (kprobe_handler(args->regs))
 313                         return NOTIFY_STOP;
 314                 break;
 315         case DIE_DEBUG:
 316                 if (post_kprobe_handler(args->regs))
 317                         return NOTIFY_STOP;
 318                 break;
 319         case DIE_GPF:
 320                 if (kprobe_running() &&
 321                     kprobe_fault_handler(args->regs, args->trapnr))
 322                         return NOTIFY_STOP;
 323                 break;
 324         case DIE_PAGE_FAULT:
 325                 if (kprobe_running() &&
 326                     kprobe_fault_handler(args->regs, args->trapnr))
 327                         return NOTIFY_STOP;
 328                 break;
 329         default:
 330                 break;
 331         }
 332         return NOTIFY_DONE;
 333 }
 334
 335 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
 336 {
 337         struct jprobe *jp = container_of(p, struct jprobe, kp);
 338         unsigned long addr;
 339
 340         jprobe_saved_regs = *regs;
 341         jprobe_saved_esp = &regs->esp;
 342         addr = (unsigned long)jprobe_saved_esp;
 343
 344         /*
 345          * TBD: As Linus pointed out, gcc assumes that the callee
 346          * owns the argument space and could overwrite it, e.g.
 347          * tailcall optimization. So, to be absolutely safe
 348          * we also save and restore enough stack bytes to cover
 349          * the argument area.
 350          */
 351         memcpy(jprobes_stack, (kprobe_opcode_t *) addr, MIN_STACK_SIZE(addr));
 352         regs->eflags &= ~IF_MASK;
 353         regs->eip = (unsigned long)(jp->entry);
 354         return 1;
 355 }
 356
 357 void jprobe_return(void)
 358 {
 359         preempt_enable_no_resched();
 360         asm volatile ("       xchgl   %%ebx,%%esp     \n"
 361                       "       int3                      \n"
 362                       "       .globl jprobe_return_end  \n"
 363                       "       jprobe_return_end:        \n"
 364                       "       nop                       \n"::"b"
 365                       (jprobe_saved_esp):"memory");
 366 }
 367
 368 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
 369 {
 370         u8 *addr = (u8 *) (regs->eip - 1);
 371         unsigned long stack_addr = (unsigned long)jprobe_saved_esp;
 372         struct jprobe *jp = container_of(p, struct jprobe, kp);
 373
 374         if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
 375                 if (&regs->esp != jprobe_saved_esp) {
 376                         struct pt_regs *saved_regs =
 377                             container_of(jprobe_saved_esp, struct pt_regs, esp);
 378                         printk("current esp %p does not match saved esp %p\n",
 379                                &regs->esp, jprobe_saved_esp);
 380                         printk("Saved registers for jprobe %p\n", jp);
 381                         show_registers(saved_regs);
 382                         printk("Current registers\n");
 383                         show_registers(regs);
 384                         BUG();
 385                 }
 386                 *regs = jprobe_saved_regs;
 387                 memcpy((kprobe_opcode_t *) stack_addr, jprobes_stack,
 388                        MIN_STACK_SIZE(stack_addr));
 389                 return 1;
 390         }
 391         return 0;
 392 }