[NETLINK]: Add properly module refcounting for kernel netlink sockets.
[linux-2.6.git] / kernel / ptrace.c
1 /*
2  * linux/kernel/ptrace.c
3  *
4  * (C) Copyright 1999 Linus Torvalds
5  *
6  * Common interfaces for "ptrace()" which we do not want
7  * to continually duplicate across every architecture.
8  */
9
10 #include <linux/module.h>
11 #include <linux/sched.h>
12 #include <linux/errno.h>
13 #include <linux/mm.h>
14 #include <linux/highmem.h>
15 #include <linux/pagemap.h>
16 #include <linux/smp_lock.h>
17 #include <linux/ptrace.h>
18 #include <linux/security.h>
19 #include <linux/signal.h>
20
21 #include <asm/pgtable.h>
22 #include <asm/uaccess.h>
23
24 /*
25  * ptrace a task: make the debugger its new parent and
26  * move it to the ptrace list.
27  *
28  * Must be called with the tasklist lock write-held.
29  */
30 void __ptrace_link(task_t *child, task_t *new_parent)
31 {
32         if (!list_empty(&child->ptrace_list))
33                 BUG();
34         if (child->parent == new_parent)
35                 return;
36         list_add(&child->ptrace_list, &child->parent->ptrace_children);
37         REMOVE_LINKS(child);
38         child->parent = new_parent;
39         SET_LINKS(child);
40 }
41  
42 /*
43  * Turn a tracing stop into a normal stop now, since with no tracer there
44  * would be no way to wake it up with SIGCONT or SIGKILL.  If there was a
45  * signal sent that would resume the child, but didn't because it was in
46  * TASK_TRACED, resume it now.
47  * Requires that irqs be disabled.
48  */
49 void ptrace_untrace(task_t *child)
50 {
51         spin_lock(&child->sighand->siglock);
52         if (child->state == TASK_TRACED) {
53                 if (child->signal->flags & SIGNAL_STOP_STOPPED) {
54                         child->state = TASK_STOPPED;
55                 } else {
56                         signal_wake_up(child, 1);
57                 }
58         }
59         spin_unlock(&child->sighand->siglock);
60 }
61
62 /*
63  * unptrace a task: move it back to its original parent and
64  * remove it from the ptrace list.
65  *
66  * Must be called with the tasklist lock write-held.
67  */
68 void __ptrace_unlink(task_t *child)
69 {
70         if (!child->ptrace)
71                 BUG();
72         child->ptrace = 0;
73         if (!list_empty(&child->ptrace_list)) {
74                 list_del_init(&child->ptrace_list);
75                 REMOVE_LINKS(child);
76                 child->parent = child->real_parent;
77                 SET_LINKS(child);
78         }
79
80         if (child->state == TASK_TRACED)
81                 ptrace_untrace(child);
82 }
83
84 /*
85  * Check that we have indeed attached to the thing..
86  */
87 int ptrace_check_attach(struct task_struct *child, int kill)
88 {
89         int ret = -ESRCH;
90
91         /*
92          * We take the read lock around doing both checks to close a
93          * possible race where someone else was tracing our child and
94          * detached between these two checks.  After this locked check,
95          * we are sure that this is our traced child and that can only
96          * be changed by us so it's not changing right after this.
97          */
98         read_lock(&tasklist_lock);
99         if ((child->ptrace & PT_PTRACED) && child->parent == current &&
100             (!(child->ptrace & PT_ATTACHED) || child->real_parent != current)
101             && child->signal != NULL) {
102                 ret = 0;
103                 spin_lock_irq(&child->sighand->siglock);
104                 if (child->state == TASK_STOPPED) {
105                         child->state = TASK_TRACED;
106                 } else if (child->state != TASK_TRACED && !kill) {
107                         ret = -ESRCH;
108                 }
109                 spin_unlock_irq(&child->sighand->siglock);
110         }
111         read_unlock(&tasklist_lock);
112
113         if (!ret && !kill) {
114                 wait_task_inactive(child);
115         }
116
117         /* All systems go.. */
118         return ret;
119 }
120
121 int ptrace_attach(struct task_struct *task)
122 {
123         int retval;
124         task_lock(task);
125         retval = -EPERM;
126         if (task->pid <= 1)
127                 goto bad;
128         if (task == current)
129                 goto bad;
130         if (!task->mm)
131                 goto bad;
132         if(((current->uid != task->euid) ||
133             (current->uid != task->suid) ||
134             (current->uid != task->uid) ||
135             (current->gid != task->egid) ||
136             (current->gid != task->sgid) ||
137             (current->gid != task->gid)) && !capable(CAP_SYS_PTRACE))
138                 goto bad;
139         smp_rmb();
140         if (!task->mm->dumpable && !capable(CAP_SYS_PTRACE))
141                 goto bad;
142         /* the same process cannot be attached many times */
143         if (task->ptrace & PT_PTRACED)
144                 goto bad;
145         retval = security_ptrace(current, task);
146         if (retval)
147                 goto bad;
148
149         /* Go */
150         task->ptrace |= PT_PTRACED | ((task->real_parent != current)
151                                       ? PT_ATTACHED : 0);
152         if (capable(CAP_SYS_PTRACE))
153                 task->ptrace |= PT_PTRACE_CAP;
154         task_unlock(task);
155
156         write_lock_irq(&tasklist_lock);
157         __ptrace_link(task, current);
158         write_unlock_irq(&tasklist_lock);
159
160         force_sig_specific(SIGSTOP, task);
161         return 0;
162
163 bad:
164         task_unlock(task);
165         return retval;
166 }
167
168 int ptrace_detach(struct task_struct *child, unsigned int data)
169 {
170         if (!valid_signal(data))
171                 return  -EIO;
172
173         /* Architecture-specific hardware disable .. */
174         ptrace_disable(child);
175
176         /* .. re-parent .. */
177         child->exit_code = data;
178
179         write_lock_irq(&tasklist_lock);
180         __ptrace_unlink(child);
181         /* .. and wake it up. */
182         if (child->exit_state != EXIT_ZOMBIE)
183                 wake_up_process(child);
184         write_unlock_irq(&tasklist_lock);
185
186         return 0;
187 }
188
189 /*
190  * Access another process' address space.
191  * Source/target buffer must be kernel space, 
192  * Do not walk the page table directly, use get_user_pages
193  */
194
195 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
196 {
197         struct mm_struct *mm;
198         struct vm_area_struct *vma;
199         struct page *page;
200         void *old_buf = buf;
201
202         mm = get_task_mm(tsk);
203         if (!mm)
204                 return 0;
205
206         down_read(&mm->mmap_sem);
207         /* ignore errors, just check how much was sucessfully transfered */
208         while (len) {
209                 int bytes, ret, offset;
210                 void *maddr;
211
212                 ret = get_user_pages(tsk, mm, addr, 1,
213                                 write, 1, &page, &vma);
214                 if (ret <= 0)
215                         break;
216
217                 bytes = len;
218                 offset = addr & (PAGE_SIZE-1);
219                 if (bytes > PAGE_SIZE-offset)
220                         bytes = PAGE_SIZE-offset;
221
222                 maddr = kmap(page);
223                 if (write) {
224                         copy_to_user_page(vma, page, addr,
225                                           maddr + offset, buf, bytes);
226                         set_page_dirty_lock(page);
227                 } else {
228                         copy_from_user_page(vma, page, addr,
229                                             buf, maddr + offset, bytes);
230                 }
231                 kunmap(page);
232                 page_cache_release(page);
233                 len -= bytes;
234                 buf += bytes;
235                 addr += bytes;
236         }
237         up_read(&mm->mmap_sem);
238         mmput(mm);
239         
240         return buf - old_buf;
241 }
242
243 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
244 {
245         int copied = 0;
246
247         while (len > 0) {
248                 char buf[128];
249                 int this_len, retval;
250
251                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
252                 retval = access_process_vm(tsk, src, buf, this_len, 0);
253                 if (!retval) {
254                         if (copied)
255                                 break;
256                         return -EIO;
257                 }
258                 if (copy_to_user(dst, buf, retval))
259                         return -EFAULT;
260                 copied += retval;
261                 src += retval;
262                 dst += retval;
263                 len -= retval;                  
264         }
265         return copied;
266 }
267
268 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
269 {
270         int copied = 0;
271
272         while (len > 0) {
273                 char buf[128];
274                 int this_len, retval;
275
276                 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
277                 if (copy_from_user(buf, src, this_len))
278                         return -EFAULT;
279                 retval = access_process_vm(tsk, dst, buf, this_len, 1);
280                 if (!retval) {
281                         if (copied)
282                                 break;
283                         return -EIO;
284                 }
285                 copied += retval;
286                 src += retval;
287                 dst += retval;
288                 len -= retval;                  
289         }
290         return copied;
291 }
292
293 static int ptrace_setoptions(struct task_struct *child, long data)
294 {
295         child->ptrace &= ~PT_TRACE_MASK;
296
297         if (data & PTRACE_O_TRACESYSGOOD)
298                 child->ptrace |= PT_TRACESYSGOOD;
299
300         if (data & PTRACE_O_TRACEFORK)
301                 child->ptrace |= PT_TRACE_FORK;
302
303         if (data & PTRACE_O_TRACEVFORK)
304                 child->ptrace |= PT_TRACE_VFORK;
305
306         if (data & PTRACE_O_TRACECLONE)
307                 child->ptrace |= PT_TRACE_CLONE;
308
309         if (data & PTRACE_O_TRACEEXEC)
310                 child->ptrace |= PT_TRACE_EXEC;
311
312         if (data & PTRACE_O_TRACEVFORKDONE)
313                 child->ptrace |= PT_TRACE_VFORK_DONE;
314
315         if (data & PTRACE_O_TRACEEXIT)
316                 child->ptrace |= PT_TRACE_EXIT;
317
318         return (data & ~PTRACE_O_MASK) ? -EINVAL : 0;
319 }
320
321 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t __user * data)
322 {
323         siginfo_t lastinfo;
324         int error = -ESRCH;
325
326         read_lock(&tasklist_lock);
327         if (likely(child->sighand != NULL)) {
328                 error = -EINVAL;
329                 spin_lock_irq(&child->sighand->siglock);
330                 if (likely(child->last_siginfo != NULL)) {
331                         lastinfo = *child->last_siginfo;
332                         error = 0;
333                 }
334                 spin_unlock_irq(&child->sighand->siglock);
335         }
336         read_unlock(&tasklist_lock);
337         if (!error)
338                 return copy_siginfo_to_user(data, &lastinfo);
339         return error;
340 }
341
342 static int ptrace_setsiginfo(struct task_struct *child, siginfo_t __user * data)
343 {
344         siginfo_t newinfo;
345         int error = -ESRCH;
346
347         if (copy_from_user(&newinfo, data, sizeof (siginfo_t)))
348                 return -EFAULT;
349
350         read_lock(&tasklist_lock);
351         if (likely(child->sighand != NULL)) {
352                 error = -EINVAL;
353                 spin_lock_irq(&child->sighand->siglock);
354                 if (likely(child->last_siginfo != NULL)) {
355                         *child->last_siginfo = newinfo;
356                         error = 0;
357                 }
358                 spin_unlock_irq(&child->sighand->siglock);
359         }
360         read_unlock(&tasklist_lock);
361         return error;
362 }
363
364 int ptrace_request(struct task_struct *child, long request,
365                    long addr, long data)
366 {
367         int ret = -EIO;
368
369         switch (request) {
370 #ifdef PTRACE_OLDSETOPTIONS
371         case PTRACE_OLDSETOPTIONS:
372 #endif
373         case PTRACE_SETOPTIONS:
374                 ret = ptrace_setoptions(child, data);
375                 break;
376         case PTRACE_GETEVENTMSG:
377                 ret = put_user(child->ptrace_message, (unsigned long __user *) data);
378                 break;
379         case PTRACE_GETSIGINFO:
380                 ret = ptrace_getsiginfo(child, (siginfo_t __user *) data);
381                 break;
382         case PTRACE_SETSIGINFO:
383                 ret = ptrace_setsiginfo(child, (siginfo_t __user *) data);
384                 break;
385         default:
386                 break;
387         }
388
389         return ret;
390 }