4008c060f7ef1b2b9ddc51154d30e1ec865906a5
[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/pagemap.h>
7 #include <linux/mempolicy.h>
8
9 #include <asm/elf.h>
10 #include <asm/uaccess.h>
11 #include <asm/tlbflush.h>
12 #include "internal.h"
13
14 char *task_mem(struct mm_struct *mm, char *buffer)
15 {
16         unsigned long data, text, lib;
17         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
18
19         /*
20          * Note: to minimize their overhead, mm maintains hiwater_vm and
21          * hiwater_rss only when about to *lower* total_vm or rss.  Any
22          * collector of these hiwater stats must therefore get total_vm
23          * and rss too, which will usually be the higher.  Barriers? not
24          * worth the effort, such snapshots can always be inconsistent.
25          */
26         hiwater_vm = total_vm = mm->total_vm;
27         if (hiwater_vm < mm->hiwater_vm)
28                 hiwater_vm = mm->hiwater_vm;
29         hiwater_rss = total_rss = get_mm_rss(mm);
30         if (hiwater_rss < mm->hiwater_rss)
31                 hiwater_rss = mm->hiwater_rss;
32
33         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
34         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
35         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
36         buffer += sprintf(buffer,
37                 "VmPeak:\t%8lu kB\n"
38                 "VmSize:\t%8lu kB\n"
39                 "VmLck:\t%8lu kB\n"
40                 "VmHWM:\t%8lu kB\n"
41                 "VmRSS:\t%8lu kB\n"
42                 "VmData:\t%8lu kB\n"
43                 "VmStk:\t%8lu kB\n"
44                 "VmExe:\t%8lu kB\n"
45                 "VmLib:\t%8lu kB\n"
46                 "VmPTE:\t%8lu kB\n",
47                 hiwater_vm << (PAGE_SHIFT-10),
48                 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
49                 mm->locked_vm << (PAGE_SHIFT-10),
50                 hiwater_rss << (PAGE_SHIFT-10),
51                 total_rss << (PAGE_SHIFT-10),
52                 data << (PAGE_SHIFT-10),
53                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
54                 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
55         return buffer;
56 }
57
58 unsigned long task_vsize(struct mm_struct *mm)
59 {
60         return PAGE_SIZE * mm->total_vm;
61 }
62
63 int task_statm(struct mm_struct *mm, int *shared, int *text,
64                int *data, int *resident)
65 {
66         *shared = get_mm_counter(mm, file_rss);
67         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
68                                                                 >> PAGE_SHIFT;
69         *data = mm->total_vm - mm->shared_vm;
70         *resident = *shared + get_mm_counter(mm, anon_rss);
71         return mm->total_vm;
72 }
73
74 int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
75 {
76         struct vm_area_struct * vma;
77         int result = -ENOENT;
78         struct task_struct *task = get_proc_task(inode);
79         struct mm_struct * mm = NULL;
80
81         if (task) {
82                 mm = get_task_mm(task);
83                 put_task_struct(task);
84         }
85         if (!mm)
86                 goto out;
87         down_read(&mm->mmap_sem);
88
89         vma = mm->mmap;
90         while (vma) {
91                 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
92                         break;
93                 vma = vma->vm_next;
94         }
95
96         if (vma) {
97                 *mnt = mntget(vma->vm_file->f_path.mnt);
98                 *dentry = dget(vma->vm_file->f_path.dentry);
99                 result = 0;
100         }
101
102         up_read(&mm->mmap_sem);
103         mmput(mm);
104 out:
105         return result;
106 }
107
108 static void pad_len_spaces(struct seq_file *m, int len)
109 {
110         len = 25 + sizeof(void*) * 6 - len;
111         if (len < 1)
112                 len = 1;
113         seq_printf(m, "%*c", len, ' ');
114 }
115
116 struct mem_size_stats
117 {
118         unsigned long resident;
119         unsigned long shared_clean;
120         unsigned long shared_dirty;
121         unsigned long private_clean;
122         unsigned long private_dirty;
123         unsigned long referenced;
124 };
125
126 struct pmd_walker {
127         struct vm_area_struct *vma;
128         void *private;
129         void (*action)(struct vm_area_struct *, pmd_t *, unsigned long,
130                        unsigned long, void *);
131 };
132
133 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
134 {
135         struct proc_maps_private *priv = m->private;
136         struct task_struct *task = priv->task;
137         struct vm_area_struct *vma = v;
138         struct mm_struct *mm = vma->vm_mm;
139         struct file *file = vma->vm_file;
140         int flags = vma->vm_flags;
141         unsigned long ino = 0;
142         dev_t dev = 0;
143         int len;
144
145         if (file) {
146                 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
147                 dev = inode->i_sb->s_dev;
148                 ino = inode->i_ino;
149         }
150
151         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
152                         vma->vm_start,
153                         vma->vm_end,
154                         flags & VM_READ ? 'r' : '-',
155                         flags & VM_WRITE ? 'w' : '-',
156                         flags & VM_EXEC ? 'x' : '-',
157                         flags & VM_MAYSHARE ? 's' : 'p',
158                         vma->vm_pgoff << PAGE_SHIFT,
159                         MAJOR(dev), MINOR(dev), ino, &len);
160
161         /*
162          * Print the dentry name for named mappings, and a
163          * special [heap] marker for the heap:
164          */
165         if (file) {
166                 pad_len_spaces(m, len);
167                 seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n");
168         } else {
169                 const char *name = arch_vma_name(vma);
170                 if (!name) {
171                         if (mm) {
172                                 if (vma->vm_start <= mm->start_brk &&
173                                                 vma->vm_end >= mm->brk) {
174                                         name = "[heap]";
175                                 } else if (vma->vm_start <= mm->start_stack &&
176                                            vma->vm_end >= mm->start_stack) {
177                                         name = "[stack]";
178                                 }
179                         } else {
180                                 name = "[vdso]";
181                         }
182                 }
183                 if (name) {
184                         pad_len_spaces(m, len);
185                         seq_puts(m, name);
186                 }
187         }
188         seq_putc(m, '\n');
189
190         if (mss)
191                 seq_printf(m,
192                            "Size:           %8lu kB\n"
193                            "Rss:            %8lu kB\n"
194                            "Shared_Clean:   %8lu kB\n"
195                            "Shared_Dirty:   %8lu kB\n"
196                            "Private_Clean:  %8lu kB\n"
197                            "Private_Dirty:  %8lu kB\n"
198                            "Referenced:     %8lu kB\n",
199                            (vma->vm_end - vma->vm_start) >> 10,
200                            mss->resident >> 10,
201                            mss->shared_clean  >> 10,
202                            mss->shared_dirty  >> 10,
203                            mss->private_clean >> 10,
204                            mss->private_dirty >> 10,
205                            mss->referenced >> 10);
206
207         if (m->count < m->size)  /* vma is copied successfully */
208                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
209         return 0;
210 }
211
212 static int show_map(struct seq_file *m, void *v)
213 {
214         return show_map_internal(m, v, NULL);
215 }
216
217 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
218                             unsigned long addr, unsigned long end,
219                             void *private)
220 {
221         struct mem_size_stats *mss = private;
222         pte_t *pte, ptent;
223         spinlock_t *ptl;
224         struct page *page;
225
226         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
227         for (; addr != end; pte++, addr += PAGE_SIZE) {
228                 ptent = *pte;
229                 if (!pte_present(ptent))
230                         continue;
231
232                 mss->resident += PAGE_SIZE;
233
234                 page = vm_normal_page(vma, addr, ptent);
235                 if (!page)
236                         continue;
237
238                 /* Accumulate the size in pages that have been accessed. */
239                 if (pte_young(ptent) || PageReferenced(page))
240                         mss->referenced += PAGE_SIZE;
241                 if (page_mapcount(page) >= 2) {
242                         if (pte_dirty(ptent))
243                                 mss->shared_dirty += PAGE_SIZE;
244                         else
245                                 mss->shared_clean += PAGE_SIZE;
246                 } else {
247                         if (pte_dirty(ptent))
248                                 mss->private_dirty += PAGE_SIZE;
249                         else
250                                 mss->private_clean += PAGE_SIZE;
251                 }
252         }
253         pte_unmap_unlock(pte - 1, ptl);
254         cond_resched();
255 }
256
257 static void clear_refs_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
258                                  unsigned long addr, unsigned long end,
259                                  void *private)
260 {
261         pte_t *pte, ptent;
262         spinlock_t *ptl;
263         struct page *page;
264
265         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
266         for (; addr != end; pte++, addr += PAGE_SIZE) {
267                 ptent = *pte;
268                 if (!pte_present(ptent))
269                         continue;
270
271                 page = vm_normal_page(vma, addr, ptent);
272                 if (!page)
273                         continue;
274
275                 /* Clear accessed and referenced bits. */
276                 ptep_test_and_clear_young(vma, addr, pte);
277                 ClearPageReferenced(page);
278         }
279         pte_unmap_unlock(pte - 1, ptl);
280         cond_resched();
281 }
282
283 static inline void walk_pmd_range(struct pmd_walker *walker, pud_t *pud,
284                                   unsigned long addr, unsigned long end)
285 {
286         pmd_t *pmd;
287         unsigned long next;
288
289         for (pmd = pmd_offset(pud, addr); addr != end;
290              pmd++, addr = next) {
291                 next = pmd_addr_end(addr, end);
292                 if (pmd_none_or_clear_bad(pmd))
293                         continue;
294                 walker->action(walker->vma, pmd, addr, next, walker->private);
295         }
296 }
297
298 static inline void walk_pud_range(struct pmd_walker *walker, pgd_t *pgd,
299                                   unsigned long addr, unsigned long end)
300 {
301         pud_t *pud;
302         unsigned long next;
303
304         for (pud = pud_offset(pgd, addr); addr != end;
305              pud++, addr = next) {
306                 next = pud_addr_end(addr, end);
307                 if (pud_none_or_clear_bad(pud))
308                         continue;
309                 walk_pmd_range(walker, pud, addr, next);
310         }
311 }
312
313 /*
314  * walk_page_range - walk the page tables of a VMA with a callback
315  * @vma - VMA to walk
316  * @action - callback invoked for every bottom-level (PTE) page table
317  * @private - private data passed to the callback function
318  *
319  * Recursively walk the page table for the memory area in a VMA, calling
320  * a callback for every bottom-level (PTE) page table.
321  */
322 static inline void walk_page_range(struct vm_area_struct *vma,
323                                    void (*action)(struct vm_area_struct *,
324                                                   pmd_t *, unsigned long,
325                                                   unsigned long, void *),
326                                    void *private)
327 {
328         unsigned long addr = vma->vm_start;
329         unsigned long end = vma->vm_end;
330         struct pmd_walker walker = {
331                 .vma            = vma,
332                 .private        = private,
333                 .action         = action,
334         };
335         pgd_t *pgd;
336         unsigned long next;
337
338         for (pgd = pgd_offset(vma->vm_mm, addr); addr != end;
339              pgd++, addr = next) {
340                 next = pgd_addr_end(addr, end);
341                 if (pgd_none_or_clear_bad(pgd))
342                         continue;
343                 walk_pud_range(&walker, pgd, addr, next);
344         }
345 }
346
347 static int show_smap(struct seq_file *m, void *v)
348 {
349         struct vm_area_struct *vma = v;
350         struct mem_size_stats mss;
351
352         memset(&mss, 0, sizeof mss);
353         if (vma->vm_mm && !is_vm_hugetlb_page(vma))
354                 walk_page_range(vma, smaps_pte_range, &mss);
355         return show_map_internal(m, v, &mss);
356 }
357
358 void clear_refs_smap(struct mm_struct *mm)
359 {
360         struct vm_area_struct *vma;
361
362         down_read(&mm->mmap_sem);
363         for (vma = mm->mmap; vma; vma = vma->vm_next)
364                 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
365                         walk_page_range(vma, clear_refs_pte_range, NULL);
366         flush_tlb_mm(mm);
367         up_read(&mm->mmap_sem);
368 }
369
370 static void *m_start(struct seq_file *m, loff_t *pos)
371 {
372         struct proc_maps_private *priv = m->private;
373         unsigned long last_addr = m->version;
374         struct mm_struct *mm;
375         struct vm_area_struct *vma, *tail_vma = NULL;
376         loff_t l = *pos;
377
378         /* Clear the per syscall fields in priv */
379         priv->task = NULL;
380         priv->tail_vma = NULL;
381
382         /*
383          * We remember last_addr rather than next_addr to hit with
384          * mmap_cache most of the time. We have zero last_addr at
385          * the beginning and also after lseek. We will have -1 last_addr
386          * after the end of the vmas.
387          */
388
389         if (last_addr == -1UL)
390                 return NULL;
391
392         priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
393         if (!priv->task)
394                 return NULL;
395
396         mm = get_task_mm(priv->task);
397         if (!mm)
398                 return NULL;
399
400         priv->tail_vma = tail_vma = get_gate_vma(priv->task);
401         down_read(&mm->mmap_sem);
402
403         /* Start with last addr hint */
404         if (last_addr && (vma = find_vma(mm, last_addr))) {
405                 vma = vma->vm_next;
406                 goto out;
407         }
408
409         /*
410          * Check the vma index is within the range and do
411          * sequential scan until m_index.
412          */
413         vma = NULL;
414         if ((unsigned long)l < mm->map_count) {
415                 vma = mm->mmap;
416                 while (l-- && vma)
417                         vma = vma->vm_next;
418                 goto out;
419         }
420
421         if (l != mm->map_count)
422                 tail_vma = NULL; /* After gate vma */
423
424 out:
425         if (vma)
426                 return vma;
427
428         /* End of vmas has been reached */
429         m->version = (tail_vma != NULL)? 0: -1UL;
430         up_read(&mm->mmap_sem);
431         mmput(mm);
432         return tail_vma;
433 }
434
435 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
436 {
437         if (vma && vma != priv->tail_vma) {
438                 struct mm_struct *mm = vma->vm_mm;
439                 up_read(&mm->mmap_sem);
440                 mmput(mm);
441         }
442 }
443
444 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
445 {
446         struct proc_maps_private *priv = m->private;
447         struct vm_area_struct *vma = v;
448         struct vm_area_struct *tail_vma = priv->tail_vma;
449
450         (*pos)++;
451         if (vma && (vma != tail_vma) && vma->vm_next)
452                 return vma->vm_next;
453         vma_stop(priv, vma);
454         return (vma != tail_vma)? tail_vma: NULL;
455 }
456
457 static void m_stop(struct seq_file *m, void *v)
458 {
459         struct proc_maps_private *priv = m->private;
460         struct vm_area_struct *vma = v;
461
462         vma_stop(priv, vma);
463         if (priv->task)
464                 put_task_struct(priv->task);
465 }
466
467 static struct seq_operations proc_pid_maps_op = {
468         .start  = m_start,
469         .next   = m_next,
470         .stop   = m_stop,
471         .show   = show_map
472 };
473
474 static struct seq_operations proc_pid_smaps_op = {
475         .start  = m_start,
476         .next   = m_next,
477         .stop   = m_stop,
478         .show   = show_smap
479 };
480
481 static int do_maps_open(struct inode *inode, struct file *file,
482                         struct seq_operations *ops)
483 {
484         struct proc_maps_private *priv;
485         int ret = -ENOMEM;
486         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
487         if (priv) {
488                 priv->pid = proc_pid(inode);
489                 ret = seq_open(file, ops);
490                 if (!ret) {
491                         struct seq_file *m = file->private_data;
492                         m->private = priv;
493                 } else {
494                         kfree(priv);
495                 }
496         }
497         return ret;
498 }
499
500 static int maps_open(struct inode *inode, struct file *file)
501 {
502         return do_maps_open(inode, file, &proc_pid_maps_op);
503 }
504
505 const struct file_operations proc_maps_operations = {
506         .open           = maps_open,
507         .read           = seq_read,
508         .llseek         = seq_lseek,
509         .release        = seq_release_private,
510 };
511
512 #ifdef CONFIG_NUMA
513 extern int show_numa_map(struct seq_file *m, void *v);
514
515 static struct seq_operations proc_pid_numa_maps_op = {
516         .start  = m_start,
517         .next   = m_next,
518         .stop   = m_stop,
519         .show   = show_numa_map
520 };
521
522 static int numa_maps_open(struct inode *inode, struct file *file)
523 {
524         return do_maps_open(inode, file, &proc_pid_numa_maps_op);
525 }
526
527 const struct file_operations proc_numa_maps_operations = {
528         .open           = numa_maps_open,
529         .read           = seq_read,
530         .llseek         = seq_lseek,
531         .release        = seq_release_private,
532 };
533 #endif
534
535 static int smaps_open(struct inode *inode, struct file *file)
536 {
537         return do_maps_open(inode, file, &proc_pid_smaps_op);
538 }
539
540 const struct file_operations proc_smaps_operations = {
541         .open           = smaps_open,
542         .read           = seq_read,
543         .llseek         = seq_lseek,
544         .release        = seq_release_private,
545 };