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