[PATCH] add /proc/pid/smaps
[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
18         data = mm->total_vm - mm->shared_vm - mm->stack_vm;
19         text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
20         lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
21         buffer += sprintf(buffer,
22                 "VmSize:\t%8lu kB\n"
23                 "VmLck:\t%8lu kB\n"
24                 "VmRSS:\t%8lu kB\n"
25                 "VmData:\t%8lu kB\n"
26                 "VmStk:\t%8lu kB\n"
27                 "VmExe:\t%8lu kB\n"
28                 "VmLib:\t%8lu kB\n"
29                 "VmPTE:\t%8lu kB\n",
30                 (mm->total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
31                 mm->locked_vm << (PAGE_SHIFT-10),
32                 get_mm_counter(mm, rss) << (PAGE_SHIFT-10),
33                 data << (PAGE_SHIFT-10),
34                 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
35                 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
36         return buffer;
37 }
38
39 unsigned long task_vsize(struct mm_struct *mm)
40 {
41         return PAGE_SIZE * mm->total_vm;
42 }
43
44 int task_statm(struct mm_struct *mm, int *shared, int *text,
45                int *data, int *resident)
46 {
47         int rss = get_mm_counter(mm, rss);
48
49         *shared = rss - get_mm_counter(mm, anon_rss);
50         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
51                                                                 >> PAGE_SHIFT;
52         *data = mm->total_vm - mm->shared_vm;
53         *resident = rss;
54         return mm->total_vm;
55 }
56
57 int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
58 {
59         struct vm_area_struct * vma;
60         int result = -ENOENT;
61         struct task_struct *task = proc_task(inode);
62         struct mm_struct * mm = get_task_mm(task);
63
64         if (!mm)
65                 goto out;
66         down_read(&mm->mmap_sem);
67
68         vma = mm->mmap;
69         while (vma) {
70                 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
71                         break;
72                 vma = vma->vm_next;
73         }
74
75         if (vma) {
76                 *mnt = mntget(vma->vm_file->f_vfsmnt);
77                 *dentry = dget(vma->vm_file->f_dentry);
78                 result = 0;
79         }
80
81         up_read(&mm->mmap_sem);
82         mmput(mm);
83 out:
84         return result;
85 }
86
87 static void pad_len_spaces(struct seq_file *m, int len)
88 {
89         len = 25 + sizeof(void*) * 6 - len;
90         if (len < 1)
91                 len = 1;
92         seq_printf(m, "%*c", len, ' ');
93 }
94
95 struct mem_size_stats
96 {
97         unsigned long resident;
98         unsigned long shared_clean;
99         unsigned long shared_dirty;
100         unsigned long private_clean;
101         unsigned long private_dirty;
102 };
103
104 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
105 {
106         struct task_struct *task = m->private;
107         struct vm_area_struct *vma = v;
108         struct mm_struct *mm = vma->vm_mm;
109         struct file *file = vma->vm_file;
110         int flags = vma->vm_flags;
111         unsigned long ino = 0;
112         dev_t dev = 0;
113         int len;
114
115         if (file) {
116                 struct inode *inode = vma->vm_file->f_dentry->d_inode;
117                 dev = inode->i_sb->s_dev;
118                 ino = inode->i_ino;
119         }
120
121         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
122                         vma->vm_start,
123                         vma->vm_end,
124                         flags & VM_READ ? 'r' : '-',
125                         flags & VM_WRITE ? 'w' : '-',
126                         flags & VM_EXEC ? 'x' : '-',
127                         flags & VM_MAYSHARE ? 's' : 'p',
128                         vma->vm_pgoff << PAGE_SHIFT,
129                         MAJOR(dev), MINOR(dev), ino, &len);
130
131         /*
132          * Print the dentry name for named mappings, and a
133          * special [heap] marker for the heap:
134          */
135         if (file) {
136                 pad_len_spaces(m, len);
137                 seq_path(m, file->f_vfsmnt, file->f_dentry, "\n");
138         } else {
139                 if (mm) {
140                         if (vma->vm_start <= mm->start_brk &&
141                                                 vma->vm_end >= mm->brk) {
142                                 pad_len_spaces(m, len);
143                                 seq_puts(m, "[heap]");
144                         } else {
145                                 if (vma->vm_start <= mm->start_stack &&
146                                         vma->vm_end >= mm->start_stack) {
147
148                                         pad_len_spaces(m, len);
149                                         seq_puts(m, "[stack]");
150                                 }
151                         }
152                 } else {
153                         pad_len_spaces(m, len);
154                         seq_puts(m, "[vdso]");
155                 }
156         }
157         seq_putc(m, '\n');
158
159         if (mss)
160                 seq_printf(m,
161                            "Size:          %8lu kB\n"
162                            "Rss:           %8lu kB\n"
163                            "Shared_Clean:  %8lu kB\n"
164                            "Shared_Dirty:  %8lu kB\n"
165                            "Private_Clean: %8lu kB\n"
166                            "Private_Dirty: %8lu kB\n",
167                            (vma->vm_end - vma->vm_start) >> 10,
168                            mss->resident >> 10,
169                            mss->shared_clean  >> 10,
170                            mss->shared_dirty  >> 10,
171                            mss->private_clean >> 10,
172                            mss->private_dirty >> 10);
173
174         if (m->count < m->size)  /* vma is copied successfully */
175                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
176         return 0;
177 }
178
179 static int show_map(struct seq_file *m, void *v)
180 {
181         return show_map_internal(m, v, 0);
182 }
183
184 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
185                                 unsigned long addr, unsigned long end,
186                                 struct mem_size_stats *mss)
187 {
188         pte_t *pte, ptent;
189         unsigned long pfn;
190         struct page *page;
191
192         pte = pte_offset_map(pmd, addr);
193         do {
194                 ptent = *pte;
195                 if (pte_none(ptent) || !pte_present(ptent))
196                         continue;
197
198                 mss->resident += PAGE_SIZE;
199                 pfn = pte_pfn(ptent);
200                 if (!pfn_valid(pfn))
201                         continue;
202
203                 page = pfn_to_page(pfn);
204                 if (page_count(page) >= 2) {
205                         if (pte_dirty(ptent))
206                                 mss->shared_dirty += PAGE_SIZE;
207                         else
208                                 mss->shared_clean += PAGE_SIZE;
209                 } else {
210                         if (pte_dirty(ptent))
211                                 mss->private_dirty += PAGE_SIZE;
212                         else
213                                 mss->private_clean += PAGE_SIZE;
214                 }
215         } while (pte++, addr += PAGE_SIZE, addr != end);
216         pte_unmap(pte - 1);
217         cond_resched_lock(&vma->vm_mm->page_table_lock);
218 }
219
220 static inline void smaps_pmd_range(struct vm_area_struct *vma, pud_t *pud,
221                                 unsigned long addr, unsigned long end,
222                                 struct mem_size_stats *mss)
223 {
224         pmd_t *pmd;
225         unsigned long next;
226
227         pmd = pmd_offset(pud, addr);
228         do {
229                 next = pmd_addr_end(addr, end);
230                 if (pmd_none_or_clear_bad(pmd))
231                         continue;
232                 smaps_pte_range(vma, pmd, addr, next, mss);
233         } while (pmd++, addr = next, addr != end);
234 }
235
236 static inline void smaps_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
237                                 unsigned long addr, unsigned long end,
238                                 struct mem_size_stats *mss)
239 {
240         pud_t *pud;
241         unsigned long next;
242
243         pud = pud_offset(pgd, addr);
244         do {
245                 next = pud_addr_end(addr, end);
246                 if (pud_none_or_clear_bad(pud))
247                         continue;
248                 smaps_pmd_range(vma, pud, addr, next, mss);
249         } while (pud++, addr = next, addr != end);
250 }
251
252 static inline void smaps_pgd_range(struct vm_area_struct *vma,
253                                 unsigned long addr, unsigned long end,
254                                 struct mem_size_stats *mss)
255 {
256         pgd_t *pgd;
257         unsigned long next;
258
259         pgd = pgd_offset(vma->vm_mm, addr);
260         do {
261                 next = pgd_addr_end(addr, end);
262                 if (pgd_none_or_clear_bad(pgd))
263                         continue;
264                 smaps_pud_range(vma, pgd, addr, next, mss);
265         } while (pgd++, addr = next, addr != end);
266 }
267
268 static int show_smap(struct seq_file *m, void *v)
269 {
270         struct vm_area_struct *vma = v;
271         struct mm_struct *mm = vma->vm_mm;
272         struct mem_size_stats mss;
273
274         memset(&mss, 0, sizeof mss);
275
276         if (mm) {
277                 spin_lock(&mm->page_table_lock);
278                 smaps_pgd_range(vma, vma->vm_start, vma->vm_end, &mss);
279                 spin_unlock(&mm->page_table_lock);
280         }
281
282         return show_map_internal(m, v, &mss);
283 }
284
285 static void *m_start(struct seq_file *m, loff_t *pos)
286 {
287         struct task_struct *task = m->private;
288         unsigned long last_addr = m->version;
289         struct mm_struct *mm;
290         struct vm_area_struct *vma, *tail_vma;
291         loff_t l = *pos;
292
293         /*
294          * We remember last_addr rather than next_addr to hit with
295          * mmap_cache most of the time. We have zero last_addr at
296          * the beginning and also after lseek. We will have -1 last_addr
297          * after the end of the vmas.
298          */
299
300         if (last_addr == -1UL)
301                 return NULL;
302
303         mm = get_task_mm(task);
304         if (!mm)
305                 return NULL;
306
307         tail_vma = get_gate_vma(task);
308         down_read(&mm->mmap_sem);
309
310         /* Start with last addr hint */
311         if (last_addr && (vma = find_vma(mm, last_addr))) {
312                 vma = vma->vm_next;
313                 goto out;
314         }
315
316         /*
317          * Check the vma index is within the range and do
318          * sequential scan until m_index.
319          */
320         vma = NULL;
321         if ((unsigned long)l < mm->map_count) {
322                 vma = mm->mmap;
323                 while (l-- && vma)
324                         vma = vma->vm_next;
325                 goto out;
326         }
327
328         if (l != mm->map_count)
329                 tail_vma = NULL; /* After gate vma */
330
331 out:
332         if (vma)
333                 return vma;
334
335         /* End of vmas has been reached */
336         m->version = (tail_vma != NULL)? 0: -1UL;
337         up_read(&mm->mmap_sem);
338         mmput(mm);
339         return tail_vma;
340 }
341
342 static void m_stop(struct seq_file *m, void *v)
343 {
344         struct task_struct *task = m->private;
345         struct vm_area_struct *vma = v;
346         if (vma && vma != get_gate_vma(task)) {
347                 struct mm_struct *mm = vma->vm_mm;
348                 up_read(&mm->mmap_sem);
349                 mmput(mm);
350         }
351 }
352
353 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
354 {
355         struct task_struct *task = m->private;
356         struct vm_area_struct *vma = v;
357         struct vm_area_struct *tail_vma = get_gate_vma(task);
358
359         (*pos)++;
360         if (vma && (vma != tail_vma) && vma->vm_next)
361                 return vma->vm_next;
362         m_stop(m, v);
363         return (vma != tail_vma)? tail_vma: NULL;
364 }
365
366 struct seq_operations proc_pid_maps_op = {
367         .start  = m_start,
368         .next   = m_next,
369         .stop   = m_stop,
370         .show   = show_map
371 };
372
373 struct seq_operations proc_pid_smaps_op = {
374         .start  = m_start,
375         .next   = m_next,
376         .stop   = m_stop,
377         .show   = show_smap
378 };
379
380 #ifdef CONFIG_NUMA
381
382 struct numa_maps {
383         unsigned long pages;
384         unsigned long anon;
385         unsigned long mapped;
386         unsigned long mapcount_max;
387         unsigned long node[MAX_NUMNODES];
388 };
389
390 /*
391  * Calculate numa node maps for a vma
392  */
393 static struct numa_maps *get_numa_maps(const struct vm_area_struct *vma)
394 {
395         struct page *page;
396         unsigned long vaddr;
397         struct mm_struct *mm = vma->vm_mm;
398         int i;
399         struct numa_maps *md = kmalloc(sizeof(struct numa_maps), GFP_KERNEL);
400
401         if (!md)
402                 return NULL;
403         md->pages = 0;
404         md->anon = 0;
405         md->mapped = 0;
406         md->mapcount_max = 0;
407         for_each_node(i)
408                 md->node[i] =0;
409
410         spin_lock(&mm->page_table_lock);
411         for (vaddr = vma->vm_start; vaddr < vma->vm_end; vaddr += PAGE_SIZE) {
412                 page = follow_page(mm, vaddr, 0);
413                 if (page) {
414                         int count = page_mapcount(page);
415
416                         if (count)
417                                 md->mapped++;
418                         if (count > md->mapcount_max)
419                                 md->mapcount_max = count;
420                         md->pages++;
421                         if (PageAnon(page))
422                                 md->anon++;
423                         md->node[page_to_nid(page)]++;
424                 }
425         }
426         spin_unlock(&mm->page_table_lock);
427         return md;
428 }
429
430 static int show_numa_map(struct seq_file *m, void *v)
431 {
432         struct task_struct *task = m->private;
433         struct vm_area_struct *vma = v;
434         struct mempolicy *pol;
435         struct numa_maps *md;
436         struct zone **z;
437         int n;
438         int first;
439
440         if (!vma->vm_mm)
441                 return 0;
442
443         md = get_numa_maps(vma);
444         if (!md)
445                 return 0;
446
447         seq_printf(m, "%08lx", vma->vm_start);
448         pol = get_vma_policy(task, vma, vma->vm_start);
449         /* Print policy */
450         switch (pol->policy) {
451         case MPOL_PREFERRED:
452                 seq_printf(m, " prefer=%d", pol->v.preferred_node);
453                 break;
454         case MPOL_BIND:
455                 seq_printf(m, " bind={");
456                 first = 1;
457                 for (z = pol->v.zonelist->zones; *z; z++) {
458
459                         if (!first)
460                                 seq_putc(m, ',');
461                         else
462                                 first = 0;
463                         seq_printf(m, "%d/%s", (*z)->zone_pgdat->node_id,
464                                         (*z)->name);
465                 }
466                 seq_putc(m, '}');
467                 break;
468         case MPOL_INTERLEAVE:
469                 seq_printf(m, " interleave={");
470                 first = 1;
471                 for_each_node(n) {
472                         if (test_bit(n, pol->v.nodes)) {
473                                 if (!first)
474                                         seq_putc(m,',');
475                                 else
476                                         first = 0;
477                                 seq_printf(m, "%d",n);
478                         }
479                 }
480                 seq_putc(m, '}');
481                 break;
482         default:
483                 seq_printf(m," default");
484                 break;
485         }
486         seq_printf(m, " MaxRef=%lu Pages=%lu Mapped=%lu",
487                         md->mapcount_max, md->pages, md->mapped);
488         if (md->anon)
489                 seq_printf(m," Anon=%lu",md->anon);
490
491         for_each_online_node(n) {
492                 if (md->node[n])
493                         seq_printf(m, " N%d=%lu", n, md->node[n]);
494         }
495         seq_putc(m, '\n');
496         kfree(md);
497         if (m->count < m->size)  /* vma is copied successfully */
498                 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
499         return 0;
500 }
501
502 struct seq_operations proc_pid_numa_maps_op = {
503         .start  = m_start,
504         .next   = m_next,
505         .stop   = m_stop,
506         .show   = show_numa_map
507 };
508 #endif