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