[PATCH] zoned vm counters: conversion of nr_pagecache to per zone counter
[linux-2.6] / 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_vfsmnt);
98                 *dentry = dget(vma->vm_file->f_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 };
124
125 __attribute__((weak)) const char *arch_vma_name(struct vm_area_struct *vma)
126 {
127         return NULL;
128 }
129
130 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
131 {
132         struct proc_maps_private *priv = m->private;
133         struct task_struct *task = priv->task;
134         struct vm_area_struct *vma = v;
135         struct mm_struct *mm = vma->vm_mm;
136         struct file *file = vma->vm_file;
137         int flags = vma->vm_flags;
138         unsigned long ino = 0;
139         dev_t dev = 0;
140         int len;
141
142         if (file) {
143                 struct inode *inode = vma->vm_file->f_dentry->d_inode;
144                 dev = inode->i_sb->s_dev;
145                 ino = inode->i_ino;
146         }
147
148         seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
149                         vma->vm_start,
150                         vma->vm_end,
151                         flags & VM_READ ? 'r' : '-',
152                         flags & VM_WRITE ? 'w' : '-',
153                         flags & VM_EXEC ? 'x' : '-',
154                         flags & VM_MAYSHARE ? 's' : 'p',
155                         vma->vm_pgoff << PAGE_SHIFT,
156                         MAJOR(dev), MINOR(dev), ino, &len);
157
158         /*
159          * Print the dentry name for named mappings, and a
160          * special [heap] marker for the heap:
161          */
162         if (file) {
163                 pad_len_spaces(m, len);
164                 seq_path(m, file->f_vfsmnt, file->f_dentry, "\n");
165         } else {
166                 const char *name = arch_vma_name(vma);
167                 if (!name) {
168                         if (mm) {
169                                 if (vma->vm_start <= mm->start_brk &&
170                                                 vma->vm_end >= mm->brk) {
171                                         name = "[heap]";
172                                 } else if (vma->vm_start <= mm->start_stack &&
173                                            vma->vm_end >= mm->start_stack) {
174                                         name = "[stack]";
175                                 }
176                         } else {
177                                 name = "[vdso]";
178                         }
179                 }
180                 if (name) {
181                         pad_len_spaces(m, len);
182                         seq_puts(m, name);
183                 }
184         }
185         seq_putc(m, '\n');
186
187         if (mss)
188                 seq_printf(m,
189                            "Size:          %8lu kB\n"
190                            "Rss:           %8lu kB\n"
191                            "Shared_Clean:  %8lu kB\n"
192                            "Shared_Dirty:  %8lu kB\n"
193                            "Private_Clean: %8lu kB\n"
194                            "Private_Dirty: %8lu kB\n",
195                            (vma->vm_end - vma->vm_start) >> 10,
196                            mss->resident >> 10,
197                            mss->shared_clean  >> 10,
198                            mss->shared_dirty  >> 10,
199                            mss->private_clean >> 10,
200                            mss->private_dirty >> 10);
201
202         if (m->count < m->size)  /* vma is copied successfully */
203                 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
204         return 0;
205 }
206
207 static int show_map(struct seq_file *m, void *v)
208 {
209         return show_map_internal(m, v, NULL);
210 }
211
212 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
213                                 unsigned long addr, unsigned long end,
214                                 struct mem_size_stats *mss)
215 {
216         pte_t *pte, ptent;
217         spinlock_t *ptl;
218         struct page *page;
219
220         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
221         do {
222                 ptent = *pte;
223                 if (!pte_present(ptent))
224                         continue;
225
226                 mss->resident += PAGE_SIZE;
227
228                 page = vm_normal_page(vma, addr, ptent);
229                 if (!page)
230                         continue;
231
232                 if (page_mapcount(page) >= 2) {
233                         if (pte_dirty(ptent))
234                                 mss->shared_dirty += PAGE_SIZE;
235                         else
236                                 mss->shared_clean += PAGE_SIZE;
237                 } else {
238                         if (pte_dirty(ptent))
239                                 mss->private_dirty += PAGE_SIZE;
240                         else
241                                 mss->private_clean += PAGE_SIZE;
242                 }
243         } while (pte++, addr += PAGE_SIZE, addr != end);
244         pte_unmap_unlock(pte - 1, ptl);
245         cond_resched();
246 }
247
248 static inline void smaps_pmd_range(struct vm_area_struct *vma, pud_t *pud,
249                                 unsigned long addr, unsigned long end,
250                                 struct mem_size_stats *mss)
251 {
252         pmd_t *pmd;
253         unsigned long next;
254
255         pmd = pmd_offset(pud, addr);
256         do {
257                 next = pmd_addr_end(addr, end);
258                 if (pmd_none_or_clear_bad(pmd))
259                         continue;
260                 smaps_pte_range(vma, pmd, addr, next, mss);
261         } while (pmd++, addr = next, addr != end);
262 }
263
264 static inline void smaps_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
265                                 unsigned long addr, unsigned long end,
266                                 struct mem_size_stats *mss)
267 {
268         pud_t *pud;
269         unsigned long next;
270
271         pud = pud_offset(pgd, addr);
272         do {
273                 next = pud_addr_end(addr, end);
274                 if (pud_none_or_clear_bad(pud))
275                         continue;
276                 smaps_pmd_range(vma, pud, addr, next, mss);
277         } while (pud++, addr = next, addr != end);
278 }
279
280 static inline void smaps_pgd_range(struct vm_area_struct *vma,
281                                 unsigned long addr, unsigned long end,
282                                 struct mem_size_stats *mss)
283 {
284         pgd_t *pgd;
285         unsigned long next;
286
287         pgd = pgd_offset(vma->vm_mm, addr);
288         do {
289                 next = pgd_addr_end(addr, end);
290                 if (pgd_none_or_clear_bad(pgd))
291                         continue;
292                 smaps_pud_range(vma, pgd, addr, next, mss);
293         } while (pgd++, addr = next, addr != end);
294 }
295
296 static int show_smap(struct seq_file *m, void *v)
297 {
298         struct vm_area_struct *vma = v;
299         struct mem_size_stats mss;
300
301         memset(&mss, 0, sizeof mss);
302         if (vma->vm_mm && !is_vm_hugetlb_page(vma))
303                 smaps_pgd_range(vma, vma->vm_start, vma->vm_end, &mss);
304         return show_map_internal(m, v, &mss);
305 }
306
307 static void *m_start(struct seq_file *m, loff_t *pos)
308 {
309         struct proc_maps_private *priv = m->private;
310         unsigned long last_addr = m->version;
311         struct mm_struct *mm;
312         struct vm_area_struct *vma, *tail_vma = NULL;
313         loff_t l = *pos;
314
315         /* Clear the per syscall fields in priv */
316         priv->task = NULL;
317         priv->tail_vma = NULL;
318
319         /*
320          * We remember last_addr rather than next_addr to hit with
321          * mmap_cache most of the time. We have zero last_addr at
322          * the beginning and also after lseek. We will have -1 last_addr
323          * after the end of the vmas.
324          */
325
326         if (last_addr == -1UL)
327                 return NULL;
328
329         priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
330         if (!priv->task)
331                 return NULL;
332
333         mm = get_task_mm(priv->task);
334         if (!mm)
335                 return NULL;
336
337         priv->tail_vma = tail_vma = get_gate_vma(priv->task);
338         down_read(&mm->mmap_sem);
339
340         /* Start with last addr hint */
341         if (last_addr && (vma = find_vma(mm, last_addr))) {
342                 vma = vma->vm_next;
343                 goto out;
344         }
345
346         /*
347          * Check the vma index is within the range and do
348          * sequential scan until m_index.
349          */
350         vma = NULL;
351         if ((unsigned long)l < mm->map_count) {
352                 vma = mm->mmap;
353                 while (l-- && vma)
354                         vma = vma->vm_next;
355                 goto out;
356         }
357
358         if (l != mm->map_count)
359                 tail_vma = NULL; /* After gate vma */
360
361 out:
362         if (vma)
363                 return vma;
364
365         /* End of vmas has been reached */
366         m->version = (tail_vma != NULL)? 0: -1UL;
367         up_read(&mm->mmap_sem);
368         mmput(mm);
369         return tail_vma;
370 }
371
372 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
373 {
374         if (vma && vma != priv->tail_vma) {
375                 struct mm_struct *mm = vma->vm_mm;
376                 up_read(&mm->mmap_sem);
377                 mmput(mm);
378         }
379 }
380
381 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
382 {
383         struct proc_maps_private *priv = m->private;
384         struct vm_area_struct *vma = v;
385         struct vm_area_struct *tail_vma = priv->tail_vma;
386
387         (*pos)++;
388         if (vma && (vma != tail_vma) && vma->vm_next)
389                 return vma->vm_next;
390         vma_stop(priv, vma);
391         return (vma != tail_vma)? tail_vma: NULL;
392 }
393
394 static void m_stop(struct seq_file *m, void *v)
395 {
396         struct proc_maps_private *priv = m->private;
397         struct vm_area_struct *vma = v;
398
399         vma_stop(priv, vma);
400         if (priv->task)
401                 put_task_struct(priv->task);
402 }
403
404 static struct seq_operations proc_pid_maps_op = {
405         .start  = m_start,
406         .next   = m_next,
407         .stop   = m_stop,
408         .show   = show_map
409 };
410
411 static struct seq_operations proc_pid_smaps_op = {
412         .start  = m_start,
413         .next   = m_next,
414         .stop   = m_stop,
415         .show   = show_smap
416 };
417
418 static int do_maps_open(struct inode *inode, struct file *file,
419                         struct seq_operations *ops)
420 {
421         struct proc_maps_private *priv;
422         int ret = -ENOMEM;
423         priv = kzalloc(sizeof(*priv), GFP_KERNEL);
424         if (priv) {
425                 priv->pid = proc_pid(inode);
426                 ret = seq_open(file, ops);
427                 if (!ret) {
428                         struct seq_file *m = file->private_data;
429                         m->private = priv;
430                 } else {
431                         kfree(priv);
432                 }
433         }
434         return ret;
435 }
436
437 static int maps_open(struct inode *inode, struct file *file)
438 {
439         return do_maps_open(inode, file, &proc_pid_maps_op);
440 }
441
442 struct file_operations proc_maps_operations = {
443         .open           = maps_open,
444         .read           = seq_read,
445         .llseek         = seq_lseek,
446         .release        = seq_release_private,
447 };
448
449 #ifdef CONFIG_NUMA
450 extern int show_numa_map(struct seq_file *m, void *v);
451
452 static struct seq_operations proc_pid_numa_maps_op = {
453         .start  = m_start,
454         .next   = m_next,
455         .stop   = m_stop,
456         .show   = show_numa_map
457 };
458
459 static int numa_maps_open(struct inode *inode, struct file *file)
460 {
461         return do_maps_open(inode, file, &proc_pid_numa_maps_op);
462 }
463
464 struct file_operations proc_numa_maps_operations = {
465         .open           = numa_maps_open,
466         .read           = seq_read,
467         .llseek         = seq_lseek,
468         .release        = seq_release_private,
469 };
470 #endif
471
472 static int smaps_open(struct inode *inode, struct file *file)
473 {
474         return do_maps_open(inode, file, &proc_pid_smaps_op);
475 }
476
477 struct file_operations proc_smaps_operations = {
478         .open           = smaps_open,
479         .read           = seq_read,
480         .llseek         = seq_lseek,
481         .release        = seq_release_private,
482 };