proc: maps protection
[linux-2.6.git] / fs / proc / task_mmu.c
1 #include <linux/mm.h>
2 #include <linux/hugetlb.h>
3 #include <linux/mount.h>
4 #include <linux/seq_file.h>
5 #include <linux/highmem.h>
6 #include <linux/ptrace.h>
7 #include <linux/pagemap.h>
8 #include <linux/mempolicy.h>
9
10 #include <asm/elf.h>
11 #include <asm/uaccess.h>
12 #include <asm/tlbflush.h>
13 #include "internal.h"
14
15 char *task_mem(struct mm_struct *mm, char *buffer)
16 {
17         unsigned long data, text, lib;
18         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
19
20         /*
21          * Note: to minimize their overhead, mm maintains hiwater_vm and
22          * hiwater_rss only when about to *lower* total_vm or rss.  Any
23          * collector of these hiwater stats must therefore get total_vm
24          * and rss too, which will usually be the higher.  Barriers? not
25          * worth the effort, such snapshots can always be inconsistent.
26          */
27         hiwater_vm = total_vm = mm->total_vm;
28         if (hiwater_vm < mm->hiwater_vm)
29                 hiwater_vm = mm->hiwater_vm;
30         hiwater_rss = total_rss = get_mm_rss(mm);
31         if (hiwater_rss < mm->hiwater_rss)
32                 hiwater_rss = mm->hiwater_rss;
33
34         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
35         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
36         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
37         buffer += sprintf(buffer,
38                 "VmPeak:\t%8lu kB\n"
39                 "VmSize:\t%8lu kB\n"
40                 "VmLck:\t%8lu kB\n"
41                 "VmHWM:\t%8lu kB\n"
42                 "VmRSS:\t%8lu kB\n"
43                 "VmData:\t%8lu kB\n"
44                 "VmStk:\t%8lu kB\n"
45                 "VmExe:\t%8lu kB\n"
46                 "VmLib:\t%8lu kB\n"
47                 "VmPTE:\t%8lu kB\n",
48                 hiwater_vm << (PAGE_SHIFT-10),
49                 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
50                 mm->locked_vm << (PAGE_SHIFT-10),
51                 hiwater_rss << (PAGE_SHIFT-10),
52                 total_rss << (PAGE_SHIFT-10),
53                 data << (PAGE_SHIFT-10),
54                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
55                 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
56         return buffer;
57 }
58
59 unsigned long task_vsize(struct mm_struct *mm)
60 {
61         return PAGE_SIZE * mm->total_vm;
62 }
63
64 int task_statm(struct mm_struct *mm, int *shared, int *text,
65                int *data, int *resident)
66 {
67         *shared = get_mm_counter(mm, file_rss);
68         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
69                                                                 >> PAGE_SHIFT;
70         *data = mm->total_vm - mm->shared_vm;
71         *resident = *shared + get_mm_counter(mm, anon_rss);
72         return mm->total_vm;
73 }
74
75 int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
76 {
77         struct vm_area_struct * vma;
78         int result = -ENOENT;
79         struct task_struct *task = get_proc_task(inode);
80         struct mm_struct * mm = NULL;
81
82         if (task) {
83                 mm = get_task_mm(task);
84                 put_task_struct(task);
85         }
86         if (!mm)
87                 goto out;
88         down_read(&mm->mmap_sem);
89
90         vma = mm->mmap;
91         while (vma) {
92                 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
93                         break;
94                 vma = vma->vm_next;
95         }
96
97         if (vma) {
98                 *mnt = mntget(vma->vm_file->f_path.mnt);
99                 *dentry = dget(vma->vm_file->f_path.dentry);
100                 result = 0;
101         }
102
103         up_read(&mm->mmap_sem);
104         mmput(mm);
105 out:
106         return result;
107 }
108
109 static void pad_len_spaces(struct seq_file *m, int len)
110 {
111         len = 25 + sizeof(void*) * 6 - len;
112         if (len < 1)
113                 len = 1;
114         seq_printf(m, "%*c", len, ' ');
115 }
116
117 struct mem_size_stats
118 {
119         unsigned long resident;
120         unsigned long shared_clean;
121         unsigned long shared_dirty;
122         unsigned long private_clean;
123         unsigned long private_dirty;
124         unsigned long referenced;
125 };
126
127 struct pmd_walker {
128         struct vm_area_struct *vma;
129         void *private;
130         void (*action)(struct vm_area_struct *, pmd_t *, unsigned long,
131                        unsigned long, void *);
132 };
133
134 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
135 {
136         struct proc_maps_private *priv = m->private;
137         struct task_struct *task = priv->task;
138         struct vm_area_struct *vma = v;
139         struct mm_struct *mm = vma->vm_mm;
140         struct file *file = vma->vm_file;
141         int flags = vma->vm_flags;
142         unsigned long ino = 0;
143         dev_t dev = 0;
144         int len;
145
146         if (maps_protect && !ptrace_may_attach(task))
147                 return -EACCES;
148
149         if (file) {
150                 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
151                 dev = inode->i_sb->s_dev;
152                 ino = inode->i_ino;
153         }
154
155         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
156                         vma->vm_start,
157                         vma->vm_end,
158                         flags & VM_READ ? 'r' : '-',
159                         flags & VM_WRITE ? 'w' : '-',
160                         flags & VM_EXEC ? 'x' : '-',
161                         flags & VM_MAYSHARE ? 's' : 'p',
162                         vma->vm_pgoff << PAGE_SHIFT,
163                         MAJOR(dev), MINOR(dev), ino, &len);
164
165         /*
166          * Print the dentry name for named mappings, and a
167          * special [heap] marker for the heap:
168          */
169         if (file) {
170                 pad_len_spaces(m, len);
171                 seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n");
172         } else {
173                 const char *name = arch_vma_name(vma);
174                 if (!name) {
175                         if (mm) {
176                                 if (vma->vm_start <= mm->start_brk &&
177                                                 vma->vm_end >= mm->brk) {
178                                         name = "[heap]";
179                                 } else if (vma->vm_start <= mm->start_stack &&
180                                            vma->vm_end >= mm->start_stack) {
181                                         name = "[stack]";
182                                 }
183                         } else {
184                                 name = "[vdso]";
185                         }
186                 }
187                 if (name) {
188                         pad_len_spaces(m, len);
189                         seq_puts(m, name);
190                 }
191         }
192         seq_putc(m, '\n');
193
194         if (mss)
195                 seq_printf(m,
196                            "Size:           %8lu kB\n"
197                            "Rss:            %8lu kB\n"
198                            "Shared_Clean:   %8lu kB\n"
199                            "Shared_Dirty:   %8lu kB\n"
200                            "Private_Clean:  %8lu kB\n"
201                            "Private_Dirty:  %8lu kB\n"
202                            "Referenced:     %8lu kB\n",
203                            (vma->vm_end - vma->vm_start) >> 10,
204                            mss->resident >> 10,
205                            mss->shared_clean  >> 10,
206                            mss->shared_dirty  >> 10,
207                            mss->private_clean >> 10,
208                            mss->private_dirty >> 10,
209                            mss->referenced >> 10);
210
211         if (m->count < m->size)  /* vma is copied successfully */
212                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
213         return 0;
214 }
215
216 static int show_map(struct seq_file *m, void *v)
217 {
218         return show_map_internal(m, v, NULL);
219 }
220
221 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
222                             unsigned long addr, unsigned long end,
223                             void *private)
224 {
225         struct mem_size_stats *mss = private;
226         pte_t *pte, ptent;
227         spinlock_t *ptl;
228         struct page *page;
229
230         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
231         for (; addr != end; pte++, addr += PAGE_SIZE) {
232                 ptent = *pte;
233                 if (!pte_present(ptent))
234                         continue;
235
236                 mss->resident += PAGE_SIZE;
237
238                 page = vm_normal_page(vma, addr, ptent);
239                 if (!page)
240                         continue;
241
242                 /* Accumulate the size in pages that have been accessed. */
243                 if (pte_young(ptent) || PageReferenced(page))
244                         mss->referenced += PAGE_SIZE;
245                 if (page_mapcount(page) >= 2) {
246                         if (pte_dirty(ptent))
247                                 mss->shared_dirty += PAGE_SIZE;
248                         else
249                                 mss->shared_clean += PAGE_SIZE;
250                 } else {
251                         if (pte_dirty(ptent))
252                                 mss->private_dirty += PAGE_SIZE;
253                         else
254                                 mss->private_clean += PAGE_SIZE;
255                 }
256         }
257         pte_unmap_unlock(pte - 1, ptl);
258         cond_resched();
259 }
260
261 static void clear_refs_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
262                                  unsigned long addr, unsigned long end,
263                                  void *private)
264 {
265         pte_t *pte, ptent;
266         spinlock_t *ptl;
267         struct page *page;
268
269         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
270         for (; addr != end; pte++, addr += PAGE_SIZE) {
271                 ptent = *pte;
272                 if (!pte_present(ptent))
273                         continue;
274
275                 page = vm_normal_page(vma, addr, ptent);
276                 if (!page)
277                         continue;
278
279                 /* Clear accessed and referenced bits. */
280                 ptep_test_and_clear_young(vma, addr, pte);
281                 ClearPageReferenced(page);
282         }
283         pte_unmap_unlock(pte - 1, ptl);
284         cond_resched();
285 }
286
287 static inline void walk_pmd_range(struct pmd_walker *walker, pud_t *pud,
288                                   unsigned long addr, unsigned long end)
289 {
290         pmd_t *pmd;
291         unsigned long next;
292
293         for (pmd = pmd_offset(pud, addr); addr != end;
294              pmd++, addr = next) {
295                 next = pmd_addr_end(addr, end);
296                 if (pmd_none_or_clear_bad(pmd))
297                         continue;
298                 walker->action(walker->vma, pmd, addr, next, walker->private);
299         }
300 }
301
302 static inline void walk_pud_range(struct pmd_walker *walker, pgd_t *pgd,
303                                   unsigned long addr, unsigned long end)
304 {
305         pud_t *pud;
306         unsigned long next;
307
308         for (pud = pud_offset(pgd, addr); addr != end;
309              pud++, addr = next) {
310                 next = pud_addr_end(addr, end);
311                 if (pud_none_or_clear_bad(pud))
312                         continue;
313                 walk_pmd_range(walker, pud, addr, next);
314         }
315 }
316
317 /*
318  * walk_page_range - walk the page tables of a VMA with a callback
319  * @vma - VMA to walk
320  * @action - callback invoked for every bottom-level (PTE) page table
321  * @private - private data passed to the callback function
322  *
323  * Recursively walk the page table for the memory area in a VMA, calling
324  * a callback for every bottom-level (PTE) page table.
325  */
326 static inline void walk_page_range(struct vm_area_struct *vma,
327                                    void (*action)(struct vm_area_struct *,
328                                                   pmd_t *, unsigned long,
329                                                   unsigned long, void *),
330                                    void *private)
331 {
332         unsigned long addr = vma->vm_start;
333         unsigned long end = vma->vm_end;
334         struct pmd_walker walker = {
335                 .vma            = vma,
336                 .private        = private,
337                 .action         = action,
338         };
339         pgd_t *pgd;
340         unsigned long next;
341
342         for (pgd = pgd_offset(vma->vm_mm, addr); addr != end;
343              pgd++, addr = next) {
344                 next = pgd_addr_end(addr, end);
345                 if (pgd_none_or_clear_bad(pgd))
346                         continue;
347                 walk_pud_range(&walker, pgd, addr, next);
348         }
349 }
350
351 static int show_smap(struct seq_file *m, void *v)
352 {
353         struct vm_area_struct *vma = v;
354         struct mem_size_stats mss;
355
356         memset(&mss, 0, sizeof mss);
357         if (vma->vm_mm && !is_vm_hugetlb_page(vma))
358                 walk_page_range(vma, smaps_pte_range, &mss);
359         return show_map_internal(m, v, &mss);
360 }
361
362 void clear_refs_smap(struct mm_struct *mm)
363 {
364         struct vm_area_struct *vma;
365
366         down_read(&mm->mmap_sem);
367         for (vma = mm->mmap; vma; vma = vma->vm_next)
368                 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
369                         walk_page_range(vma, clear_refs_pte_range, NULL);
370         flush_tlb_mm(mm);
371         up_read(&mm->mmap_sem);
372 }
373
374 static void *m_start(struct seq_file *m, loff_t *pos)
375 {
376         struct proc_maps_private *priv = m->private;
377         unsigned long last_addr = m->version;
378         struct mm_struct *mm;
379         struct vm_area_struct *vma, *tail_vma = NULL;
380         loff_t l = *pos;
381
382         /* Clear the per syscall fields in priv */
383         priv->task = NULL;
384         priv->tail_vma = NULL;
385
386         /*
387          * We remember last_addr rather than next_addr to hit with
388          * mmap_cache most of the time. We have zero last_addr at
389          * the beginning and also after lseek. We will have -1 last_addr
390          * after the end of the vmas.
391          */
392
393         if (last_addr == -1UL)
394                 return NULL;
395
396         priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
397         if (!priv->task)
398                 return NULL;
399
400         mm = get_task_mm(priv->task);
401         if (!mm)
402                 return NULL;
403
404         priv->tail_vma = tail_vma = get_gate_vma(priv->task);
405         down_read(&mm->mmap_sem);
406
407         /* Start with last addr hint */
408         if (last_addr && (vma = find_vma(mm, last_addr))) {
409                 vma = vma->vm_next;
410                 goto out;
411         }
412
413         /*
414          * Check the vma index is within the range and do
415          * sequential scan until m_index.
416          */
417         vma = NULL;
418         if ((unsigned long)l < mm->map_count) {
419                 vma = mm->mmap;
420                 while (l-- && vma)
421                         vma = vma->vm_next;
422                 goto out;
423         }
424
425         if (l != mm->map_count)
426                 tail_vma = NULL; /* After gate vma */
427
428 out:
429         if (vma)
430                 return vma;
431
432         /* End of vmas has been reached */
433         m->version = (tail_vma != NULL)? 0: -1UL;
434         up_read(&mm->mmap_sem);
435         mmput(mm);
436         return tail_vma;
437 }
438
439 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
440 {
441         if (vma && vma != priv->tail_vma) {
442                 struct mm_struct *mm = vma->vm_mm;
443                 up_read(&mm->mmap_sem);
444                 mmput(mm);
445         }
446 }
447
448 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
449 {
450         struct proc_maps_private *priv = m->private;
451         struct vm_area_struct *vma = v;
452         struct vm_area_struct *tail_vma = priv->tail_vma;
453
454         (*pos)++;
455         if (vma && (vma != tail_vma) && vma->vm_next)
456                 return vma->vm_next;
457         vma_stop(priv, vma);
458         return (vma != tail_vma)? tail_vma: NULL;
459 }
460
461 static void m_stop(struct seq_file *m, void *v)
462 {
463         struct proc_maps_private *priv = m->private;
464         struct vm_area_struct *vma = v;
465
466         vma_stop(priv, vma);
467         if (priv->task)
468                 put_task_struct(priv->task);
469 }
470
471 static struct seq_operations proc_pid_maps_op = {
472         .start  = m_start,
473         .next   = m_next,
474         .stop   = m_stop,
475         .show   = show_map
476 };
477
478 static struct seq_operations proc_pid_smaps_op = {
479         .start  = m_start,
480         .next   = m_next,
481         .stop   = m_stop,
482         .show   = show_smap
483 };
484
485 static int do_maps_open(struct inode *inode, struct file *file,
486                         struct seq_operations *ops)
487 {
488         struct proc_maps_private *priv;
489         int ret = -ENOMEM;
490         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
491         if (priv) {
492                 priv->pid = proc_pid(inode);
493                 ret = seq_open(file, ops);
494                 if (!ret) {
495                         struct seq_file *m = file->private_data;
496                         m->private = priv;
497                 } else {
498                         kfree(priv);
499                 }
500         }
501         return ret;
502 }
503
504 static int maps_open(struct inode *inode, struct file *file)
505 {
506         return do_maps_open(inode, file, &proc_pid_maps_op);
507 }
508
509 const struct file_operations proc_maps_operations = {
510         .open           = maps_open,
511         .read           = seq_read,
512         .llseek         = seq_lseek,
513         .release        = seq_release_private,
514 };
515
516 #ifdef CONFIG_NUMA
517 extern int show_numa_map(struct seq_file *m, void *v);
518
519 static int show_numa_map_checked(struct seq_file *m, void *v)
520 {
521         struct proc_maps_private *priv = m->private;
522         struct task_struct *task = priv->task;
523
524         if (maps_protect && !ptrace_may_attach(task))
525                 return -EACCES;
526
527         return show_numa_map(m, v);
528 }
529
530 static struct seq_operations proc_pid_numa_maps_op = {
531         .start  = m_start,
532         .next   = m_next,
533         .stop   = m_stop,
534         .show   = show_numa_map_checked
535 };
536
537 static int numa_maps_open(struct inode *inode, struct file *file)
538 {
539         return do_maps_open(inode, file, &proc_pid_numa_maps_op);
540 }
541
542 const struct file_operations proc_numa_maps_operations = {
543         .open           = numa_maps_open,
544         .read           = seq_read,
545         .llseek         = seq_lseek,
546         .release        = seq_release_private,
547 };
548 #endif
549
550 static int smaps_open(struct inode *inode, struct file *file)
551 {
552         return do_maps_open(inode, file, &proc_pid_smaps_op);
553 }
554
555 const struct file_operations proc_smaps_operations = {
556         .open           = smaps_open,
557         .read           = seq_read,
558         .llseek         = seq_lseek,
559         .release        = seq_release_private,
560 };