| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* | 
 | 2 |  * Kernel support for the ptrace() and syscall tracing interfaces. | 
 | 3 |  * | 
 | 4 |  * Copyright (C) 2000 Hewlett-Packard Co, Linuxcare Inc. | 
 | 5 |  * Copyright (C) 2000 Matthew Wilcox <matthew@wil.cx> | 
 | 6 |  * Copyright (C) 2000 David Huggins-Daines <dhd@debian.org> | 
 | 7 |  */ | 
 | 8 |  | 
 | 9 | #include <linux/kernel.h> | 
 | 10 | #include <linux/sched.h> | 
 | 11 | #include <linux/mm.h> | 
 | 12 | #include <linux/smp.h> | 
 | 13 | #include <linux/smp_lock.h> | 
 | 14 | #include <linux/errno.h> | 
 | 15 | #include <linux/ptrace.h> | 
 | 16 | #include <linux/user.h> | 
 | 17 | #include <linux/personality.h> | 
 | 18 | #include <linux/security.h> | 
 | 19 | #include <linux/compat.h> | 
| Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 20 | #include <linux/signal.h> | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 |  | 
 | 22 | #include <asm/uaccess.h> | 
 | 23 | #include <asm/pgtable.h> | 
 | 24 | #include <asm/system.h> | 
 | 25 | #include <asm/processor.h> | 
| Sam Ravnborg | 0013a85 | 2005-09-09 20:57:26 +0200 | [diff] [blame] | 26 | #include <asm/asm-offsets.h> | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 |  | 
 | 28 | /* PSW bits we allow the debugger to modify */ | 
 | 29 | #define USER_PSW_BITS	(PSW_N | PSW_V | PSW_CB) | 
 | 30 |  | 
 | 31 | #undef DEBUG_PTRACE | 
 | 32 |  | 
 | 33 | #ifdef DEBUG_PTRACE | 
 | 34 | #define DBG(x...)	printk(x) | 
 | 35 | #else | 
 | 36 | #define DBG(x...) | 
 | 37 | #endif | 
 | 38 |  | 
 | 39 | #ifdef __LP64__ | 
 | 40 |  | 
 | 41 | /* This function is needed to translate 32 bit pt_regs offsets in to | 
 | 42 |  * 64 bit pt_regs offsets.  For example, a 32 bit gdb under a 64 bit kernel | 
 | 43 |  * will request offset 12 if it wants gr3, but the lower 32 bits of | 
 | 44 |  * the 64 bit kernels view of gr3 will be at offset 28 (3*8 + 4). | 
 | 45 |  * This code relies on a 32 bit pt_regs being comprised of 32 bit values | 
 | 46 |  * except for the fp registers which (a) are 64 bits, and (b) follow | 
 | 47 |  * the gr registers at the start of pt_regs.  The 32 bit pt_regs should | 
 | 48 |  * be half the size of the 64 bit pt_regs, plus 32*4 to allow for fr[] | 
 | 49 |  * being 64 bit in both cases. | 
 | 50 |  */ | 
 | 51 |  | 
 | 52 | static long translate_usr_offset(long offset) | 
 | 53 | { | 
 | 54 | 	if (offset < 0) | 
 | 55 | 		return -1; | 
 | 56 | 	else if (offset <= 32*4)	/* gr[0..31] */ | 
 | 57 | 		return offset * 2 + 4; | 
 | 58 | 	else if (offset <= 32*4+32*8)	/* gr[0..31] + fr[0..31] */ | 
 | 59 | 		return offset + 32*4; | 
 | 60 | 	else if (offset < sizeof(struct pt_regs)/2 + 32*4) | 
 | 61 | 		return offset * 2 + 4 - 32*8; | 
 | 62 | 	else | 
 | 63 | 		return -1; | 
 | 64 | } | 
 | 65 | #endif | 
 | 66 |  | 
 | 67 | /* | 
 | 68 |  * Called by kernel/ptrace.c when detaching.. | 
 | 69 |  * | 
 | 70 |  * Make sure single step bits etc are not set. | 
 | 71 |  */ | 
 | 72 | void ptrace_disable(struct task_struct *child) | 
 | 73 | { | 
 | 74 | 	/* make sure the trap bits are not set */ | 
 | 75 | 	pa_psw(child)->r = 0; | 
 | 76 | 	pa_psw(child)->t = 0; | 
 | 77 | 	pa_psw(child)->h = 0; | 
 | 78 | 	pa_psw(child)->l = 0; | 
 | 79 | } | 
 | 80 |  | 
| Christoph Hellwig | 481bed4 | 2005-11-07 00:59:47 -0800 | [diff] [blame] | 81 | long arch_ptrace(struct task_struct *child, long request, long addr, long data) | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | { | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 83 | 	long ret; | 
 | 84 | #ifdef DEBUG_PTRACE | 
 | 85 | 	long oaddr=addr, odata=data; | 
 | 86 | #endif | 
 | 87 |  | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | 	switch (request) { | 
 | 89 | 	case PTRACE_PEEKTEXT: /* read word at location addr. */  | 
 | 90 | 	case PTRACE_PEEKDATA: { | 
 | 91 | 		int copied; | 
 | 92 |  | 
 | 93 | #ifdef __LP64__ | 
 | 94 | 		if (is_compat_task(child)) { | 
 | 95 | 			unsigned int tmp; | 
 | 96 |  | 
 | 97 | 			addr &= 0xffffffffL; | 
 | 98 | 			copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0); | 
 | 99 | 			ret = -EIO; | 
 | 100 | 			if (copied != sizeof(tmp)) | 
 | 101 | 				goto out_tsk; | 
 | 102 | 			ret = put_user(tmp,(unsigned int *) data); | 
 | 103 | 			DBG("sys_ptrace(PEEK%s, %d, %lx, %lx) returning %ld, data %x\n", | 
 | 104 | 				request == PTRACE_PEEKTEXT ? "TEXT" : "DATA", | 
 | 105 | 				pid, oaddr, odata, ret, tmp); | 
 | 106 | 		} | 
 | 107 | 		else | 
 | 108 | #endif | 
 | 109 | 		{ | 
 | 110 | 			unsigned long tmp; | 
 | 111 |  | 
 | 112 | 			copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0); | 
 | 113 | 			ret = -EIO; | 
 | 114 | 			if (copied != sizeof(tmp)) | 
 | 115 | 				goto out_tsk; | 
 | 116 | 			ret = put_user(tmp,(unsigned long *) data); | 
 | 117 | 		} | 
 | 118 | 		goto out_tsk; | 
 | 119 | 	} | 
 | 120 |  | 
 | 121 | 	/* when I and D space are separate, this will have to be fixed. */ | 
 | 122 | 	case PTRACE_POKETEXT: /* write the word at location addr. */ | 
 | 123 | 	case PTRACE_POKEDATA: | 
 | 124 | 		ret = 0; | 
 | 125 | #ifdef __LP64__ | 
 | 126 | 		if (is_compat_task(child)) { | 
 | 127 | 			unsigned int tmp = (unsigned int)data; | 
 | 128 | 			DBG("sys_ptrace(POKE%s, %d, %lx, %lx)\n", | 
 | 129 | 				request == PTRACE_POKETEXT ? "TEXT" : "DATA", | 
 | 130 | 				pid, oaddr, odata); | 
 | 131 | 			addr &= 0xffffffffL; | 
 | 132 | 			if (access_process_vm(child, addr, &tmp, sizeof(tmp), 1) == sizeof(tmp)) | 
 | 133 | 				goto out_tsk; | 
 | 134 | 		} | 
 | 135 | 		else | 
 | 136 | #endif | 
 | 137 | 		{ | 
 | 138 | 			if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data)) | 
 | 139 | 				goto out_tsk; | 
 | 140 | 		} | 
 | 141 | 		ret = -EIO; | 
 | 142 | 		goto out_tsk; | 
 | 143 |  | 
 | 144 | 	/* Read the word at location addr in the USER area.  For ptraced | 
 | 145 | 	   processes, the kernel saves all regs on a syscall. */ | 
 | 146 | 	case PTRACE_PEEKUSR: { | 
 | 147 | 		ret = -EIO; | 
 | 148 | #ifdef __LP64__ | 
 | 149 | 		if (is_compat_task(child)) { | 
 | 150 | 			unsigned int tmp; | 
 | 151 |  | 
 | 152 | 			if (addr & (sizeof(int)-1)) | 
 | 153 | 				goto out_tsk; | 
 | 154 | 			if ((addr = translate_usr_offset(addr)) < 0) | 
 | 155 | 				goto out_tsk; | 
 | 156 |  | 
 | 157 | 			tmp = *(unsigned int *) ((char *) task_regs(child) + addr); | 
 | 158 | 			ret = put_user(tmp, (unsigned int *) data); | 
 | 159 | 			DBG("sys_ptrace(PEEKUSR, %d, %lx, %lx) returning %ld, addr %lx, data %x\n", | 
 | 160 | 				pid, oaddr, odata, ret, addr, tmp); | 
 | 161 | 		} | 
 | 162 | 		else | 
 | 163 | #endif | 
 | 164 | 		{ | 
 | 165 | 			unsigned long tmp; | 
 | 166 |  | 
 | 167 | 			if ((addr & (sizeof(long)-1)) || (unsigned long) addr >= sizeof(struct pt_regs)) | 
 | 168 | 				goto out_tsk; | 
 | 169 | 			tmp = *(unsigned long *) ((char *) task_regs(child) + addr); | 
 | 170 | 			ret = put_user(tmp, (unsigned long *) data); | 
 | 171 | 		} | 
 | 172 | 		goto out_tsk; | 
 | 173 | 	} | 
 | 174 |  | 
 | 175 | 	/* Write the word at location addr in the USER area.  This will need | 
 | 176 | 	   to change when the kernel no longer saves all regs on a syscall. | 
 | 177 | 	   FIXME.  There is a problem at the moment in that r3-r18 are only | 
 | 178 | 	   saved if the process is ptraced on syscall entry, and even then | 
 | 179 | 	   those values are overwritten by actual register values on syscall | 
 | 180 | 	   exit. */ | 
 | 181 | 	case PTRACE_POKEUSR: | 
 | 182 | 		ret = -EIO; | 
 | 183 | 		/* Some register values written here may be ignored in | 
 | 184 | 		 * entry.S:syscall_restore_rfi; e.g. iaoq is written with | 
 | 185 | 		 * r31/r31+4, and not with the values in pt_regs. | 
 | 186 | 		 */ | 
 | 187 | 		 /* PT_PSW=0, so this is valid for 32 bit processes under 64 | 
 | 188 | 		 * bit kernels. | 
 | 189 | 		 */ | 
 | 190 | 		if (addr == PT_PSW) { | 
 | 191 | 			/* PT_PSW=0, so this is valid for 32 bit processes | 
 | 192 | 			 * under 64 bit kernels. | 
 | 193 | 			 * | 
 | 194 | 			 * Allow writing to Nullify, Divide-step-correction, | 
 | 195 | 			 * and carry/borrow bits. | 
 | 196 | 			 * BEWARE, if you set N, and then single step, it won't | 
 | 197 | 			 * stop on the nullified instruction. | 
 | 198 | 			 */ | 
 | 199 | 			DBG("sys_ptrace(POKEUSR, %d, %lx, %lx)\n", | 
 | 200 | 				pid, oaddr, odata); | 
 | 201 | 			data &= USER_PSW_BITS; | 
 | 202 | 			task_regs(child)->gr[0] &= ~USER_PSW_BITS; | 
 | 203 | 			task_regs(child)->gr[0] |= data; | 
 | 204 | 			ret = 0; | 
 | 205 | 			goto out_tsk; | 
 | 206 | 		} | 
 | 207 | #ifdef __LP64__ | 
 | 208 | 		if (is_compat_task(child)) { | 
 | 209 | 			if (addr & (sizeof(int)-1)) | 
 | 210 | 				goto out_tsk; | 
 | 211 | 			if ((addr = translate_usr_offset(addr)) < 0) | 
 | 212 | 				goto out_tsk; | 
 | 213 | 			DBG("sys_ptrace(POKEUSR, %d, %lx, %lx) addr %lx\n", | 
 | 214 | 				pid, oaddr, odata, addr); | 
 | 215 | 			if (addr >= PT_FR0 && addr <= PT_FR31 + 4) { | 
 | 216 | 				/* Special case, fp regs are 64 bits anyway */ | 
 | 217 | 				*(unsigned int *) ((char *) task_regs(child) + addr) = data; | 
 | 218 | 				ret = 0; | 
 | 219 | 			} | 
 | 220 | 			else if ((addr >= PT_GR1+4 && addr <= PT_GR31+4) || | 
 | 221 | 					addr == PT_IAOQ0+4 || addr == PT_IAOQ1+4 || | 
 | 222 | 					addr == PT_SAR+4) { | 
 | 223 | 				/* Zero the top 32 bits */ | 
 | 224 | 				*(unsigned int *) ((char *) task_regs(child) + addr - 4) = 0; | 
 | 225 | 				*(unsigned int *) ((char *) task_regs(child) + addr) = data; | 
 | 226 | 				ret = 0; | 
 | 227 | 			} | 
 | 228 | 			goto out_tsk; | 
 | 229 | 		} | 
 | 230 | 		else | 
 | 231 | #endif | 
 | 232 | 		{ | 
 | 233 | 			if ((addr & (sizeof(long)-1)) || (unsigned long) addr >= sizeof(struct pt_regs)) | 
 | 234 | 				goto out_tsk; | 
 | 235 | 			if ((addr >= PT_GR1 && addr <= PT_GR31) || | 
 | 236 | 					addr == PT_IAOQ0 || addr == PT_IAOQ1 || | 
 | 237 | 					(addr >= PT_FR0 && addr <= PT_FR31 + 4) || | 
 | 238 | 					addr == PT_SAR) { | 
 | 239 | 				*(unsigned long *) ((char *) task_regs(child) + addr) = data; | 
 | 240 | 				ret = 0; | 
 | 241 | 			} | 
 | 242 | 			goto out_tsk; | 
 | 243 | 		} | 
 | 244 |  | 
 | 245 | 	case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */ | 
 | 246 | 	case PTRACE_CONT: | 
 | 247 | 		ret = -EIO; | 
 | 248 | 		DBG("sys_ptrace(%s)\n", | 
 | 249 | 			request == PTRACE_SYSCALL ? "SYSCALL" : "CONT"); | 
| Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 250 | 		if (!valid_signal(data)) | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 251 | 			goto out_tsk; | 
 | 252 | 		child->ptrace &= ~(PT_SINGLESTEP|PT_BLOCKSTEP); | 
 | 253 | 		if (request == PTRACE_SYSCALL) { | 
 | 254 | 			set_tsk_thread_flag(child, TIF_SYSCALL_TRACE); | 
 | 255 | 		} else { | 
 | 256 | 			clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); | 
 | 257 | 		}		 | 
 | 258 | 		child->exit_code = data; | 
 | 259 | 		goto out_wake_notrap; | 
 | 260 |  | 
 | 261 | 	case PTRACE_KILL: | 
 | 262 | 		/* | 
 | 263 | 		 * make the child exit.  Best I can do is send it a | 
 | 264 | 		 * sigkill.  perhaps it should be put in the status | 
 | 265 | 		 * that it wants to exit. | 
 | 266 | 		 */ | 
| Matthew Wilcox | 83aceb5 | 2005-11-17 16:37:24 -0500 | [diff] [blame] | 267 | 		ret = 0; | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 268 | 		DBG("sys_ptrace(KILL)\n"); | 
 | 269 | 		if (child->exit_state == EXIT_ZOMBIE)	/* already dead */ | 
 | 270 | 			goto out_tsk; | 
 | 271 | 		child->exit_code = SIGKILL; | 
 | 272 | 		goto out_wake_notrap; | 
 | 273 |  | 
 | 274 | 	case PTRACE_SINGLEBLOCK: | 
 | 275 | 		DBG("sys_ptrace(SINGLEBLOCK)\n"); | 
 | 276 | 		ret = -EIO; | 
| Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 277 | 		if (!valid_signal(data)) | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | 			goto out_tsk; | 
 | 279 | 		clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); | 
 | 280 | 		child->ptrace &= ~PT_SINGLESTEP; | 
 | 281 | 		child->ptrace |= PT_BLOCKSTEP; | 
 | 282 | 		child->exit_code = data; | 
 | 283 |  | 
 | 284 | 		/* Enable taken branch trap. */ | 
 | 285 | 		pa_psw(child)->r = 0; | 
 | 286 | 		pa_psw(child)->t = 1; | 
 | 287 | 		pa_psw(child)->h = 0; | 
 | 288 | 		pa_psw(child)->l = 0; | 
 | 289 | 		goto out_wake; | 
 | 290 |  | 
 | 291 | 	case PTRACE_SINGLESTEP: | 
 | 292 | 		DBG("sys_ptrace(SINGLESTEP)\n"); | 
 | 293 | 		ret = -EIO; | 
| Jesper Juhl | 7ed20e1 | 2005-05-01 08:59:14 -0700 | [diff] [blame] | 294 | 		if (!valid_signal(data)) | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | 			goto out_tsk; | 
 | 296 |  | 
 | 297 | 		clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); | 
 | 298 | 		child->ptrace &= ~PT_BLOCKSTEP; | 
 | 299 | 		child->ptrace |= PT_SINGLESTEP; | 
 | 300 | 		child->exit_code = data; | 
 | 301 |  | 
 | 302 | 		if (pa_psw(child)->n) { | 
 | 303 | 			struct siginfo si; | 
 | 304 |  | 
 | 305 | 			/* Nullified, just crank over the queue. */ | 
 | 306 | 			task_regs(child)->iaoq[0] = task_regs(child)->iaoq[1]; | 
 | 307 | 			task_regs(child)->iasq[0] = task_regs(child)->iasq[1]; | 
 | 308 | 			task_regs(child)->iaoq[1] = task_regs(child)->iaoq[0] + 4; | 
 | 309 | 			pa_psw(child)->n = 0; | 
 | 310 | 			pa_psw(child)->x = 0; | 
 | 311 | 			pa_psw(child)->y = 0; | 
 | 312 | 			pa_psw(child)->z = 0; | 
 | 313 | 			pa_psw(child)->b = 0; | 
 | 314 | 			ptrace_disable(child); | 
 | 315 | 			/* Don't wake up the child, but let the | 
 | 316 | 			   parent know something happened. */ | 
 | 317 | 			si.si_code = TRAP_TRACE; | 
 | 318 | 			si.si_addr = (void __user *) (task_regs(child)->iaoq[0] & ~3); | 
 | 319 | 			si.si_signo = SIGTRAP; | 
 | 320 | 			si.si_errno = 0; | 
 | 321 | 			force_sig_info(SIGTRAP, &si, child); | 
 | 322 | 			//notify_parent(child, SIGCHLD); | 
 | 323 | 			//ret = 0; | 
 | 324 | 			goto out_wake; | 
 | 325 | 		} | 
 | 326 |  | 
 | 327 | 		/* Enable recovery counter traps.  The recovery counter | 
 | 328 | 		 * itself will be set to zero on a task switch.  If the | 
 | 329 | 		 * task is suspended on a syscall then the syscall return | 
 | 330 | 		 * path will overwrite the recovery counter with a suitable | 
 | 331 | 		 * value such that it traps once back in user space.  We | 
 | 332 | 		 * disable interrupts in the childs PSW here also, to avoid | 
 | 333 | 		 * interrupts while the recovery counter is decrementing. | 
 | 334 | 		 */ | 
 | 335 | 		pa_psw(child)->r = 1; | 
 | 336 | 		pa_psw(child)->t = 0; | 
 | 337 | 		pa_psw(child)->h = 0; | 
 | 338 | 		pa_psw(child)->l = 0; | 
 | 339 | 		/* give it a chance to run. */ | 
 | 340 | 		goto out_wake; | 
 | 341 |  | 
 | 342 | 	case PTRACE_DETACH: | 
 | 343 | 		ret = ptrace_detach(child, data); | 
 | 344 | 		goto out_tsk; | 
 | 345 |  | 
 | 346 | 	case PTRACE_GETEVENTMSG: | 
 | 347 |                 ret = put_user(child->ptrace_message, (unsigned int __user *) data); | 
| Matthew Wilcox | 83aceb5 | 2005-11-17 16:37:24 -0500 | [diff] [blame] | 348 | 		goto out_tsk; | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 |  | 
 | 350 | 	default: | 
 | 351 | 		ret = ptrace_request(child, request, addr, data); | 
| Matthew Wilcox | 83aceb5 | 2005-11-17 16:37:24 -0500 | [diff] [blame] | 352 | 		goto out_tsk; | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 353 | 	} | 
 | 354 |  | 
 | 355 | out_wake_notrap: | 
 | 356 | 	ptrace_disable(child); | 
 | 357 | out_wake: | 
 | 358 | 	wake_up_process(child); | 
 | 359 | 	ret = 0; | 
 | 360 | out_tsk: | 
| Christoph Hellwig | 481bed4 | 2005-11-07 00:59:47 -0800 | [diff] [blame] | 361 | 	DBG("arch_ptrace(%ld, %d, %lx, %lx) returning %ld\n", | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | 		request, pid, oaddr, odata, ret); | 
 | 363 | 	return ret; | 
 | 364 | } | 
 | 365 |  | 
 | 366 | void syscall_trace(void) | 
 | 367 | { | 
 | 368 | 	if (!test_thread_flag(TIF_SYSCALL_TRACE)) | 
 | 369 | 		return; | 
 | 370 | 	if (!(current->ptrace & PT_PTRACED)) | 
 | 371 | 		return; | 
 | 372 | 	ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) | 
 | 373 | 				 ? 0x80 : 0)); | 
 | 374 | 	/* | 
 | 375 | 	 * this isn't the same as continuing with a signal, but it will do | 
 | 376 | 	 * for normal use.  strace only continues with a signal if the | 
 | 377 | 	 * stopping signal is not SIGTRAP.  -brl | 
 | 378 | 	 */ | 
 | 379 | 	if (current->exit_code) { | 
 | 380 | 		send_sig(current->exit_code, current, 1); | 
 | 381 | 		current->exit_code = 0; | 
 | 382 | 	} | 
 | 383 | } |