blob: 2c9db29ea3581d7ad79a250cea453411a0b4c9e9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#include <linux/mm.h>
2#include <linux/hugetlb.h>
Dave Hansen22e057c2011-03-22 16:33:00 -07003#include <linux/huge_mm.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -07004#include <linux/mount.h>
5#include <linux/seq_file.h>
Mauricio Line070ad42005-09-03 15:55:10 -07006#include <linux/highmem.h>
Kees Cook5096add2007-05-08 00:26:04 -07007#include <linux/ptrace.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09008#include <linux/slab.h>
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07009#include <linux/pagemap.h>
10#include <linux/mempolicy.h>
Dave Hansen22e057c2011-03-22 16:33:00 -070011#include <linux/rmap.h>
Matt Mackall85863e42008-02-04 22:29:04 -080012#include <linux/swap.h>
13#include <linux/swapops.h>
Mauricio Line070ad42005-09-03 15:55:10 -070014
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <asm/elf.h>
16#include <asm/uaccess.h>
Mauricio Line070ad42005-09-03 15:55:10 -070017#include <asm/tlbflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "internal.h"
19
Eric W. Biedermandf5f8312008-02-08 04:18:33 -080020void task_mem(struct seq_file *m, struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -070021{
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080022 unsigned long data, text, lib, swap;
Hugh Dickins365e9c872005-10-29 18:16:18 -070023 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
24
25 /*
26 * Note: to minimize their overhead, mm maintains hiwater_vm and
27 * hiwater_rss only when about to *lower* total_vm or rss. Any
28 * collector of these hiwater stats must therefore get total_vm
29 * and rss too, which will usually be the higher. Barriers? not
30 * worth the effort, such snapshots can always be inconsistent.
31 */
32 hiwater_vm = total_vm = mm->total_vm;
33 if (hiwater_vm < mm->hiwater_vm)
34 hiwater_vm = mm->hiwater_vm;
35 hiwater_rss = total_rss = get_mm_rss(mm);
36 if (hiwater_rss < mm->hiwater_rss)
37 hiwater_rss = mm->hiwater_rss;
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
40 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
41 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080042 swap = get_mm_counter(mm, MM_SWAPENTS);
Eric W. Biedermandf5f8312008-02-08 04:18:33 -080043 seq_printf(m,
Hugh Dickins365e9c872005-10-29 18:16:18 -070044 "VmPeak:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 "VmSize:\t%8lu kB\n"
46 "VmLck:\t%8lu kB\n"
Hugh Dickins365e9c872005-10-29 18:16:18 -070047 "VmHWM:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070048 "VmRSS:\t%8lu kB\n"
49 "VmData:\t%8lu kB\n"
50 "VmStk:\t%8lu kB\n"
51 "VmExe:\t%8lu kB\n"
52 "VmLib:\t%8lu kB\n"
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080053 "VmPTE:\t%8lu kB\n"
54 "VmSwap:\t%8lu kB\n",
Hugh Dickins365e9c872005-10-29 18:16:18 -070055 hiwater_vm << (PAGE_SHIFT-10),
56 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 mm->locked_vm << (PAGE_SHIFT-10),
Hugh Dickins365e9c872005-10-29 18:16:18 -070058 hiwater_rss << (PAGE_SHIFT-10),
59 total_rss << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 data << (PAGE_SHIFT-10),
61 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080062 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
63 swap << (PAGE_SHIFT-10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070064}
65
66unsigned long task_vsize(struct mm_struct *mm)
67{
68 return PAGE_SIZE * mm->total_vm;
69}
70
Alexey Dobriyana2ade7b2011-01-12 17:00:32 -080071unsigned long task_statm(struct mm_struct *mm,
72 unsigned long *shared, unsigned long *text,
73 unsigned long *data, unsigned long *resident)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080075 *shared = get_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
77 >> PAGE_SHIFT;
78 *data = mm->total_vm - mm->shared_vm;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080079 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 return mm->total_vm;
81}
82
Linus Torvalds1da177e2005-04-16 15:20:36 -070083static void pad_len_spaces(struct seq_file *m, int len)
84{
85 len = 25 + sizeof(void*) * 6 - len;
86 if (len < 1)
87 len = 1;
88 seq_printf(m, "%*c", len, ' ');
89}
90
Matt Mackalla6198792008-02-04 22:29:03 -080091static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
Mauricio Line070ad42005-09-03 15:55:10 -070092{
Matt Mackalla6198792008-02-04 22:29:03 -080093 if (vma && vma != priv->tail_vma) {
94 struct mm_struct *mm = vma->vm_mm;
95 up_read(&mm->mmap_sem);
96 mmput(mm);
97 }
98}
99
100static void *m_start(struct seq_file *m, loff_t *pos)
101{
102 struct proc_maps_private *priv = m->private;
103 unsigned long last_addr = m->version;
104 struct mm_struct *mm;
105 struct vm_area_struct *vma, *tail_vma = NULL;
106 loff_t l = *pos;
107
108 /* Clear the per syscall fields in priv */
109 priv->task = NULL;
110 priv->tail_vma = NULL;
111
112 /*
113 * We remember last_addr rather than next_addr to hit with
114 * mmap_cache most of the time. We have zero last_addr at
115 * the beginning and also after lseek. We will have -1 last_addr
116 * after the end of the vmas.
117 */
118
119 if (last_addr == -1UL)
120 return NULL;
121
122 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
123 if (!priv->task)
Al Viroec6fd8a2011-02-15 22:22:54 -0500124 return ERR_PTR(-ESRCH);
Matt Mackalla6198792008-02-04 22:29:03 -0800125
126 mm = mm_for_maps(priv->task);
Al Viroec6fd8a2011-02-15 22:22:54 -0500127 if (!mm || IS_ERR(mm))
128 return mm;
Oleg Nesterov00f89d22009-07-10 03:27:38 +0200129 down_read(&mm->mmap_sem);
Matt Mackalla6198792008-02-04 22:29:03 -0800130
Stephen Wilson31db58b2011-03-13 15:49:15 -0400131 tail_vma = get_gate_vma(priv->task->mm);
Matt Mackalla6198792008-02-04 22:29:03 -0800132 priv->tail_vma = tail_vma;
133
134 /* Start with last addr hint */
135 vma = find_vma(mm, last_addr);
136 if (last_addr && vma) {
137 vma = vma->vm_next;
138 goto out;
139 }
140
141 /*
142 * Check the vma index is within the range and do
143 * sequential scan until m_index.
144 */
145 vma = NULL;
146 if ((unsigned long)l < mm->map_count) {
147 vma = mm->mmap;
148 while (l-- && vma)
149 vma = vma->vm_next;
150 goto out;
151 }
152
153 if (l != mm->map_count)
154 tail_vma = NULL; /* After gate vma */
155
156out:
157 if (vma)
158 return vma;
159
160 /* End of vmas has been reached */
161 m->version = (tail_vma != NULL)? 0: -1UL;
162 up_read(&mm->mmap_sem);
163 mmput(mm);
164 return tail_vma;
165}
166
167static void *m_next(struct seq_file *m, void *v, loff_t *pos)
168{
169 struct proc_maps_private *priv = m->private;
170 struct vm_area_struct *vma = v;
171 struct vm_area_struct *tail_vma = priv->tail_vma;
172
173 (*pos)++;
174 if (vma && (vma != tail_vma) && vma->vm_next)
175 return vma->vm_next;
176 vma_stop(priv, vma);
177 return (vma != tail_vma)? tail_vma: NULL;
178}
179
180static void m_stop(struct seq_file *m, void *v)
181{
182 struct proc_maps_private *priv = m->private;
183 struct vm_area_struct *vma = v;
184
Linus Torvalds76597cd2011-03-27 19:09:29 -0700185 if (!IS_ERR(vma))
186 vma_stop(priv, vma);
Matt Mackalla6198792008-02-04 22:29:03 -0800187 if (priv->task)
188 put_task_struct(priv->task);
189}
190
191static int do_maps_open(struct inode *inode, struct file *file,
Jan Engelhardt03a44822008-02-08 04:21:19 -0800192 const struct seq_operations *ops)
Matt Mackalla6198792008-02-04 22:29:03 -0800193{
194 struct proc_maps_private *priv;
195 int ret = -ENOMEM;
196 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
197 if (priv) {
198 priv->pid = proc_pid(inode);
199 ret = seq_open(file, ops);
200 if (!ret) {
201 struct seq_file *m = file->private_data;
202 m->private = priv;
203 } else {
204 kfree(priv);
205 }
206 }
207 return ret;
208}
Mauricio Line070ad42005-09-03 15:55:10 -0700209
Joe Korty7c88db02008-10-16 15:27:09 +0400210static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211{
Mauricio Line070ad42005-09-03 15:55:10 -0700212 struct mm_struct *mm = vma->vm_mm;
213 struct file *file = vma->vm_file;
214 int flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 unsigned long ino = 0;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700216 unsigned long long pgoff = 0;
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200217 unsigned long start, end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 dev_t dev = 0;
219 int len;
220
221 if (file) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800222 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 dev = inode->i_sb->s_dev;
224 ino = inode->i_ino;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700225 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 }
227
Linus Torvaldsd7824372010-08-15 11:35:52 -0700228 /* We don't show the stack guard page in /proc/maps */
229 start = vma->vm_start;
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200230 if (stack_guard_page_start(vma, start))
231 start += PAGE_SIZE;
232 end = vma->vm_end;
233 if (stack_guard_page_end(vma, end))
234 end -= PAGE_SIZE;
Linus Torvaldsd7824372010-08-15 11:35:52 -0700235
Clement Calmels1804dc62008-08-20 14:09:00 -0700236 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
Linus Torvaldsd7824372010-08-15 11:35:52 -0700237 start,
Mikulas Patockaa09a79f2011-05-09 13:01:09 +0200238 end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 flags & VM_READ ? 'r' : '-',
240 flags & VM_WRITE ? 'w' : '-',
241 flags & VM_EXEC ? 'x' : '-',
242 flags & VM_MAYSHARE ? 's' : 'p',
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700243 pgoff,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 MAJOR(dev), MINOR(dev), ino, &len);
245
246 /*
247 * Print the dentry name for named mappings, and a
248 * special [heap] marker for the heap:
249 */
Mauricio Line070ad42005-09-03 15:55:10 -0700250 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 pad_len_spaces(m, len);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800252 seq_path(m, &file->f_path, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 } else {
Ingo Molnare6e54942006-06-27 02:53:50 -0700254 const char *name = arch_vma_name(vma);
255 if (!name) {
256 if (mm) {
Aaro Koskinen0db0c012011-03-23 16:42:50 -0700257 if (vma->vm_start <= mm->brk &&
258 vma->vm_end >= mm->start_brk) {
Ingo Molnare6e54942006-06-27 02:53:50 -0700259 name = "[heap]";
260 } else if (vma->vm_start <= mm->start_stack &&
261 vma->vm_end >= mm->start_stack) {
262 name = "[stack]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700264 } else {
265 name = "[vdso]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700267 }
268 if (name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 pad_len_spaces(m, len);
Ingo Molnare6e54942006-06-27 02:53:50 -0700270 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 }
272 }
273 seq_putc(m, '\n');
Joe Korty7c88db02008-10-16 15:27:09 +0400274}
275
276static int show_map(struct seq_file *m, void *v)
277{
278 struct vm_area_struct *vma = v;
279 struct proc_maps_private *priv = m->private;
280 struct task_struct *task = priv->task;
281
282 show_map_vma(m, vma);
Mauricio Line070ad42005-09-03 15:55:10 -0700283
Mauricio Line070ad42005-09-03 15:55:10 -0700284 if (m->count < m->size) /* vma is copied successfully */
Stephen Wilson31db58b2011-03-13 15:49:15 -0400285 m->version = (vma != get_gate_vma(task->mm))
286 ? vma->vm_start : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return 0;
288}
289
Jan Engelhardt03a44822008-02-08 04:21:19 -0800290static const struct seq_operations proc_pid_maps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800291 .start = m_start,
292 .next = m_next,
293 .stop = m_stop,
294 .show = show_map
295};
296
297static int maps_open(struct inode *inode, struct file *file)
298{
299 return do_maps_open(inode, file, &proc_pid_maps_op);
300}
301
302const struct file_operations proc_maps_operations = {
303 .open = maps_open,
304 .read = seq_read,
305 .llseek = seq_lseek,
306 .release = seq_release_private,
307};
308
309/*
310 * Proportional Set Size(PSS): my share of RSS.
311 *
312 * PSS of a process is the count of pages it has in memory, where each
313 * page is divided by the number of processes sharing it. So if a
314 * process has 1000 pages all to itself, and 1000 shared with one other
315 * process, its PSS will be 1500.
316 *
317 * To keep (accumulated) division errors low, we adopt a 64bit
318 * fixed-point pss counter to minimize division errors. So (pss >>
319 * PSS_SHIFT) would be the real byte count.
320 *
321 * A shift of 12 before division means (assuming 4K page size):
322 * - 1M 3-user-pages add up to 8KB errors;
323 * - supports mapcount up to 2^24, or 16M;
324 * - supports PSS up to 2^52 bytes, or 4PB.
325 */
326#define PSS_SHIFT 12
327
Matt Mackall1e883282008-02-04 22:29:07 -0800328#ifdef CONFIG_PROC_PAGE_MONITOR
Peter Zijlstra214e4712008-04-28 02:12:55 -0700329struct mem_size_stats {
Matt Mackalla6198792008-02-04 22:29:03 -0800330 struct vm_area_struct *vma;
331 unsigned long resident;
332 unsigned long shared_clean;
333 unsigned long shared_dirty;
334 unsigned long private_clean;
335 unsigned long private_dirty;
336 unsigned long referenced;
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700337 unsigned long anonymous;
Dave Hansen4031a212011-03-22 16:33:01 -0700338 unsigned long anonymous_thp;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700339 unsigned long swap;
Matt Mackalla6198792008-02-04 22:29:03 -0800340 u64 pss;
341};
342
Dave Hansenae11c4d2011-03-22 16:32:58 -0700343
344static void smaps_pte_entry(pte_t ptent, unsigned long addr,
Dave Hansen3c9acc72011-03-22 16:32:59 -0700345 unsigned long ptent_size, struct mm_walk *walk)
Dave Hansenae11c4d2011-03-22 16:32:58 -0700346{
347 struct mem_size_stats *mss = walk->private;
348 struct vm_area_struct *vma = mss->vma;
349 struct page *page;
350 int mapcount;
351
352 if (is_swap_pte(ptent)) {
Dave Hansen3c9acc72011-03-22 16:32:59 -0700353 mss->swap += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700354 return;
355 }
356
357 if (!pte_present(ptent))
358 return;
359
360 page = vm_normal_page(vma, addr, ptent);
361 if (!page)
362 return;
363
364 if (PageAnon(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700365 mss->anonymous += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700366
Dave Hansen3c9acc72011-03-22 16:32:59 -0700367 mss->resident += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700368 /* Accumulate the size in pages that have been accessed. */
369 if (pte_young(ptent) || PageReferenced(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700370 mss->referenced += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700371 mapcount = page_mapcount(page);
372 if (mapcount >= 2) {
373 if (pte_dirty(ptent) || PageDirty(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700374 mss->shared_dirty += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700375 else
Dave Hansen3c9acc72011-03-22 16:32:59 -0700376 mss->shared_clean += ptent_size;
377 mss->pss += (ptent_size << PSS_SHIFT) / mapcount;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700378 } else {
379 if (pte_dirty(ptent) || PageDirty(page))
Dave Hansen3c9acc72011-03-22 16:32:59 -0700380 mss->private_dirty += ptent_size;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700381 else
Dave Hansen3c9acc72011-03-22 16:32:59 -0700382 mss->private_clean += ptent_size;
383 mss->pss += (ptent_size << PSS_SHIFT);
Dave Hansenae11c4d2011-03-22 16:32:58 -0700384 }
385}
386
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800387static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700388 struct mm_walk *walk)
Mauricio Line070ad42005-09-03 15:55:10 -0700389{
Dave Hansen21650092008-06-12 15:21:47 -0700390 struct mem_size_stats *mss = walk->private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800391 struct vm_area_struct *vma = mss->vma;
Dave Hansenae11c4d2011-03-22 16:32:58 -0700392 pte_t *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700393 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700394
Dave Hansen22e057c2011-03-22 16:33:00 -0700395 spin_lock(&walk->mm->page_table_lock);
396 if (pmd_trans_huge(*pmd)) {
397 if (pmd_trans_splitting(*pmd)) {
398 spin_unlock(&walk->mm->page_table_lock);
399 wait_split_huge_page(vma->anon_vma, pmd);
400 } else {
401 smaps_pte_entry(*(pte_t *)pmd, addr,
402 HPAGE_PMD_SIZE, walk);
403 spin_unlock(&walk->mm->page_table_lock);
Dave Hansen4031a212011-03-22 16:33:01 -0700404 mss->anonymous_thp += HPAGE_PMD_SIZE;
Dave Hansen22e057c2011-03-22 16:33:00 -0700405 return 0;
406 }
407 } else {
408 spin_unlock(&walk->mm->page_table_lock);
409 }
410 /*
411 * The mmap_sem held all the way back in m_start() is what
412 * keeps khugepaged out of here and from collapsing things
413 * in here.
414 */
Hugh Dickins705e87c2005-10-29 18:16:27 -0700415 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
Dave Hansenae11c4d2011-03-22 16:32:58 -0700416 for (; addr != end; pte++, addr += PAGE_SIZE)
Dave Hansen3c9acc72011-03-22 16:32:59 -0700417 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk);
Hugh Dickins705e87c2005-10-29 18:16:27 -0700418 pte_unmap_unlock(pte - 1, ptl);
419 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800420 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700421}
422
Mauricio Line070ad42005-09-03 15:55:10 -0700423static int show_smap(struct seq_file *m, void *v)
424{
Joe Korty7c88db02008-10-16 15:27:09 +0400425 struct proc_maps_private *priv = m->private;
426 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700427 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700428 struct mem_size_stats mss;
Dave Hansen21650092008-06-12 15:21:47 -0700429 struct mm_walk smaps_walk = {
430 .pmd_entry = smaps_pte_range,
431 .mm = vma->vm_mm,
432 .private = &mss,
433 };
Mauricio Line070ad42005-09-03 15:55:10 -0700434
435 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800436 mss.vma = vma;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900437 /* mmap_sem is held in m_start */
Nick Piggin5ddfae12006-03-06 15:42:57 -0800438 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700439 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800440
Joe Korty7c88db02008-10-16 15:27:09 +0400441 show_map_vma(m, vma);
Matt Mackall4752c362008-02-04 22:29:02 -0800442
443 seq_printf(m,
444 "Size: %8lu kB\n"
445 "Rss: %8lu kB\n"
446 "Pss: %8lu kB\n"
447 "Shared_Clean: %8lu kB\n"
448 "Shared_Dirty: %8lu kB\n"
449 "Private_Clean: %8lu kB\n"
450 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700451 "Referenced: %8lu kB\n"
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700452 "Anonymous: %8lu kB\n"
Dave Hansen4031a212011-03-22 16:33:01 -0700453 "AnonHugePages: %8lu kB\n"
Mel Gorman08fba692009-01-06 14:38:53 -0800454 "Swap: %8lu kB\n"
Mel Gorman33402892009-01-06 14:38:54 -0800455 "KernelPageSize: %8lu kB\n"
Nikanth Karthikesan2d905082011-01-13 15:45:53 -0800456 "MMUPageSize: %8lu kB\n"
457 "Locked: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800458 (vma->vm_end - vma->vm_start) >> 10,
459 mss.resident >> 10,
460 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
461 mss.shared_clean >> 10,
462 mss.shared_dirty >> 10,
463 mss.private_clean >> 10,
464 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700465 mss.referenced >> 10,
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700466 mss.anonymous >> 10,
Dave Hansen4031a212011-03-22 16:33:01 -0700467 mss.anonymous_thp >> 10,
Mel Gorman08fba692009-01-06 14:38:53 -0800468 mss.swap >> 10,
Mel Gorman33402892009-01-06 14:38:54 -0800469 vma_kernel_pagesize(vma) >> 10,
Nikanth Karthikesan2d905082011-01-13 15:45:53 -0800470 vma_mmu_pagesize(vma) >> 10,
471 (vma->vm_flags & VM_LOCKED) ?
472 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0);
Matt Mackall4752c362008-02-04 22:29:02 -0800473
Joe Korty7c88db02008-10-16 15:27:09 +0400474 if (m->count < m->size) /* vma is copied successfully */
Stephen Wilson31db58b2011-03-13 15:49:15 -0400475 m->version = (vma != get_gate_vma(task->mm))
476 ? vma->vm_start : 0;
Joe Korty7c88db02008-10-16 15:27:09 +0400477 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700478}
479
Jan Engelhardt03a44822008-02-08 04:21:19 -0800480static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800481 .start = m_start,
482 .next = m_next,
483 .stop = m_stop,
484 .show = show_smap
485};
486
487static int smaps_open(struct inode *inode, struct file *file)
488{
489 return do_maps_open(inode, file, &proc_pid_smaps_op);
490}
491
492const struct file_operations proc_smaps_operations = {
493 .open = smaps_open,
494 .read = seq_read,
495 .llseek = seq_lseek,
496 .release = seq_release_private,
497};
498
499static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700500 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800501{
Dave Hansen21650092008-06-12 15:21:47 -0700502 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800503 pte_t *pte, ptent;
504 spinlock_t *ptl;
505 struct page *page;
506
Dave Hansen03319322011-03-22 16:32:56 -0700507 split_huge_page_pmd(walk->mm, pmd);
508
Matt Mackalla6198792008-02-04 22:29:03 -0800509 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
510 for (; addr != end; pte++, addr += PAGE_SIZE) {
511 ptent = *pte;
512 if (!pte_present(ptent))
513 continue;
514
515 page = vm_normal_page(vma, addr, ptent);
516 if (!page)
517 continue;
518
519 /* Clear accessed and referenced bits. */
520 ptep_test_and_clear_young(vma, addr, pte);
521 ClearPageReferenced(page);
522 }
523 pte_unmap_unlock(pte - 1, ptl);
524 cond_resched();
525 return 0;
526}
527
Moussa A. Ba398499d2009-09-21 17:02:29 -0700528#define CLEAR_REFS_ALL 1
529#define CLEAR_REFS_ANON 2
530#define CLEAR_REFS_MAPPED 3
531
Matt Mackallf248dcb2008-02-04 22:29:03 -0800532static ssize_t clear_refs_write(struct file *file, const char __user *buf,
533 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700534{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800535 struct task_struct *task;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700536 char buffer[PROC_NUMBUF];
Matt Mackallf248dcb2008-02-04 22:29:03 -0800537 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700538 struct vm_area_struct *vma;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700539 long type;
David Rientjesb813e932007-05-06 14:49:24 -0700540
Matt Mackallf248dcb2008-02-04 22:29:03 -0800541 memset(buffer, 0, sizeof(buffer));
542 if (count > sizeof(buffer) - 1)
543 count = sizeof(buffer) - 1;
544 if (copy_from_user(buffer, buf, count))
545 return -EFAULT;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700546 if (strict_strtol(strstrip(buffer), 10, &type))
547 return -EINVAL;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700548 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
Matt Mackallf248dcb2008-02-04 22:29:03 -0800549 return -EINVAL;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800550 task = get_proc_task(file->f_path.dentry->d_inode);
551 if (!task)
552 return -ESRCH;
553 mm = get_task_mm(task);
554 if (mm) {
Andrew Morton20cbc972008-07-05 12:29:05 -0700555 struct mm_walk clear_refs_walk = {
556 .pmd_entry = clear_refs_pte_range,
557 .mm = mm,
558 };
Matt Mackallf248dcb2008-02-04 22:29:03 -0800559 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700560 for (vma = mm->mmap; vma; vma = vma->vm_next) {
561 clear_refs_walk.private = vma;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700562 if (is_vm_hugetlb_page(vma))
563 continue;
564 /*
565 * Writing 1 to /proc/pid/clear_refs affects all pages.
566 *
567 * Writing 2 to /proc/pid/clear_refs only affects
568 * Anonymous pages.
569 *
570 * Writing 3 to /proc/pid/clear_refs only affects file
571 * mapped pages.
572 */
573 if (type == CLEAR_REFS_ANON && vma->vm_file)
574 continue;
575 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
576 continue;
577 walk_page_range(vma->vm_start, vma->vm_end,
578 &clear_refs_walk);
Dave Hansen21650092008-06-12 15:21:47 -0700579 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800580 flush_tlb_mm(mm);
581 up_read(&mm->mmap_sem);
582 mmput(mm);
583 }
584 put_task_struct(task);
Vincent Lifb92a4b2009-09-22 16:45:36 -0700585
586 return count;
David Rientjesb813e932007-05-06 14:49:24 -0700587}
588
Matt Mackallf248dcb2008-02-04 22:29:03 -0800589const struct file_operations proc_clear_refs_operations = {
590 .write = clear_refs_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200591 .llseek = noop_llseek,
Matt Mackallf248dcb2008-02-04 22:29:03 -0800592};
593
Matt Mackall85863e42008-02-04 22:29:04 -0800594struct pagemapread {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900595 int pos, len;
596 u64 *buffer;
Matt Mackall85863e42008-02-04 22:29:04 -0800597};
598
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500599#define PM_ENTRY_BYTES sizeof(u64)
600#define PM_STATUS_BITS 3
601#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
602#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
603#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
604#define PM_PSHIFT_BITS 6
605#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
606#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
607#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
608#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
609#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
610
611#define PM_PRESENT PM_STATUS(4LL)
612#define PM_SWAP PM_STATUS(2LL)
613#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800614#define PM_END_OF_BUFFER 1
615
616static int add_to_pagemap(unsigned long addr, u64 pfn,
617 struct pagemapread *pm)
618{
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900619 pm->buffer[pm->pos++] = pfn;
620 if (pm->pos >= pm->len)
Thomas Tuttleaae86792008-06-05 22:46:31 -0700621 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800622 return 0;
623}
624
625static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700626 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800627{
Dave Hansen21650092008-06-12 15:21:47 -0700628 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800629 unsigned long addr;
630 int err = 0;
631 for (addr = start; addr < end; addr += PAGE_SIZE) {
632 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
633 if (err)
634 break;
635 }
636 return err;
637}
638
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700639static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800640{
641 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500642 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800643}
644
Matt Mackall49c50342008-12-09 13:14:21 -0800645static u64 pte_to_pagemap_entry(pte_t pte)
Dave Hansenbcf80392008-06-12 15:21:48 -0700646{
Matt Mackall49c50342008-12-09 13:14:21 -0800647 u64 pme = 0;
Dave Hansenbcf80392008-06-12 15:21:48 -0700648 if (is_swap_pte(pte))
649 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
650 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
651 else if (pte_present(pte))
652 pme = PM_PFRAME(pte_pfn(pte))
653 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
654 return pme;
655}
656
Matt Mackall85863e42008-02-04 22:29:04 -0800657static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700658 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800659{
Dave Hansenbcf80392008-06-12 15:21:48 -0700660 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700661 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800662 pte_t *pte;
663 int err = 0;
664
Dave Hansen03319322011-03-22 16:32:56 -0700665 split_huge_page_pmd(walk->mm, pmd);
666
Dave Hansenbcf80392008-06-12 15:21:48 -0700667 /* find the first VMA at or above 'addr' */
668 vma = find_vma(walk->mm, addr);
Matt Mackall85863e42008-02-04 22:29:04 -0800669 for (; addr != end; addr += PAGE_SIZE) {
670 u64 pfn = PM_NOT_PRESENT;
Dave Hansenbcf80392008-06-12 15:21:48 -0700671
672 /* check to see if we've left 'vma' behind
673 * and need a new, higher one */
674 if (vma && (addr >= vma->vm_end))
675 vma = find_vma(walk->mm, addr);
676
677 /* check that 'vma' actually covers this address,
678 * and that it isn't a huge page vma */
679 if (vma && (vma->vm_start <= addr) &&
680 !is_vm_hugetlb_page(vma)) {
681 pte = pte_offset_map(pmd, addr);
682 pfn = pte_to_pagemap_entry(*pte);
683 /* unmap before userspace copy */
684 pte_unmap(pte);
685 }
Matt Mackall85863e42008-02-04 22:29:04 -0800686 err = add_to_pagemap(addr, pfn, pm);
687 if (err)
688 return err;
689 }
690
691 cond_resched();
692
693 return err;
694}
695
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700696#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800697static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
698{
699 u64 pme = 0;
700 if (pte_present(pte))
701 pme = PM_PFRAME(pte_pfn(pte) + offset)
702 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
703 return pme;
704}
705
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700706/* This function walks within one hugetlb entry in the single call */
707static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
708 unsigned long addr, unsigned long end,
709 struct mm_walk *walk)
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800710{
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800711 struct pagemapread *pm = walk->private;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800712 int err = 0;
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700713 u64 pfn;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800714
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800715 for (; addr != end; addr += PAGE_SIZE) {
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700716 int offset = (addr & ~hmask) >> PAGE_SHIFT;
717 pfn = huge_pte_to_pagemap_entry(*pte, offset);
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800718 err = add_to_pagemap(addr, pfn, pm);
719 if (err)
720 return err;
721 }
722
723 cond_resched();
724
725 return err;
726}
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700727#endif /* HUGETLB_PAGE */
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800728
Matt Mackall85863e42008-02-04 22:29:04 -0800729/*
730 * /proc/pid/pagemap - an array mapping virtual pages to pfns
731 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500732 * For each page in the address space, this file contains one 64-bit entry
733 * consisting of the following:
734 *
735 * Bits 0-55 page frame number (PFN) if present
736 * Bits 0-4 swap type if swapped
737 * Bits 5-55 swap offset if swapped
738 * Bits 55-60 page shift (page size = 1<<page shift)
739 * Bit 61 reserved for future use
740 * Bit 62 page swapped
741 * Bit 63 page present
742 *
743 * If the page is not present but in swap, then the PFN contains an
744 * encoding of the swap file number and the page's offset into the
745 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800746 * precisely which pages are mapped (or in swap) and comparing mapped
747 * pages between processes.
748 *
749 * Efficient users of this interface will use /proc/pid/maps to
750 * determine which areas of memory are actually mapped and llseek to
751 * skip over unmapped regions.
752 */
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900753#define PAGEMAP_WALK_SIZE (PMD_SIZE)
Naoya Horiguchiea251c12010-11-24 12:57:13 -0800754#define PAGEMAP_WALK_MASK (PMD_MASK)
Matt Mackall85863e42008-02-04 22:29:04 -0800755static ssize_t pagemap_read(struct file *file, char __user *buf,
756 size_t count, loff_t *ppos)
757{
758 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
Matt Mackall85863e42008-02-04 22:29:04 -0800759 struct mm_struct *mm;
760 struct pagemapread pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800761 int ret = -ESRCH;
Alexey Dobriyanee1e6ab2008-07-21 14:21:36 -0700762 struct mm_walk pagemap_walk = {};
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700763 unsigned long src;
764 unsigned long svpfn;
765 unsigned long start_vaddr;
766 unsigned long end_vaddr;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900767 int copied = 0;
Matt Mackall85863e42008-02-04 22:29:04 -0800768
769 if (!task)
770 goto out;
771
Al Viroca6b0bf2011-02-15 22:04:37 -0500772 mm = mm_for_maps(task);
Al Viroec6fd8a2011-02-15 22:22:54 -0500773 ret = PTR_ERR(mm);
774 if (!mm || IS_ERR(mm))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700775 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800776
777 ret = -EINVAL;
778 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700779 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700780 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800781
782 ret = 0;
Vitaly Mayatskikh08161782009-04-30 15:08:18 -0700783
784 if (!count)
785 goto out_task;
786
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900787 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
788 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700789 ret = -ENOMEM;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900790 if (!pm.buffer)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700791 goto out_mm;
Matt Mackall85863e42008-02-04 22:29:04 -0800792
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700793 pagemap_walk.pmd_entry = pagemap_pte_range;
794 pagemap_walk.pte_hole = pagemap_pte_hole;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700795#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800796 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700797#endif
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700798 pagemap_walk.mm = mm;
799 pagemap_walk.private = &pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800800
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700801 src = *ppos;
802 svpfn = src / PM_ENTRY_BYTES;
803 start_vaddr = svpfn << PAGE_SHIFT;
804 end_vaddr = TASK_SIZE_OF(task);
Matt Mackall85863e42008-02-04 22:29:04 -0800805
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700806 /* watch out for wraparound */
807 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
808 start_vaddr = end_vaddr;
809
810 /*
811 * The odds are that this will stop walking way
812 * before end_vaddr, because the length of the
813 * user buffer is tracked in "pm", and the walk
814 * will stop when we hit the end of the buffer.
815 */
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900816 ret = 0;
817 while (count && (start_vaddr < end_vaddr)) {
818 int len;
819 unsigned long end;
Matt Mackall85863e42008-02-04 22:29:04 -0800820
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900821 pm.pos = 0;
Naoya Horiguchiea251c12010-11-24 12:57:13 -0800822 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900823 /* overflow ? */
824 if (end < start_vaddr || end > end_vaddr)
825 end = end_vaddr;
826 down_read(&mm->mmap_sem);
827 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
828 up_read(&mm->mmap_sem);
829 start_vaddr = end;
830
831 len = min(count, PM_ENTRY_BYTES * pm.pos);
Dan Carpenter309361e02010-04-06 13:45:39 +0300832 if (copy_to_user(buf, pm.buffer, len)) {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900833 ret = -EFAULT;
834 goto out_free;
835 }
836 copied += len;
837 buf += len;
838 count -= len;
Matt Mackall85863e42008-02-04 22:29:04 -0800839 }
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900840 *ppos += copied;
841 if (!ret || ret == PM_END_OF_BUFFER)
842 ret = copied;
843
Matt Mackall85863e42008-02-04 22:29:04 -0800844out_free:
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900845 kfree(pm.buffer);
Marcelo Tosattifb393802008-03-13 12:32:35 -0700846out_mm:
847 mmput(mm);
Matt Mackall85863e42008-02-04 22:29:04 -0800848out_task:
849 put_task_struct(task);
850out:
851 return ret;
852}
853
854const struct file_operations proc_pagemap_operations = {
855 .llseek = mem_lseek, /* borrow this */
856 .read = pagemap_read,
857};
Matt Mackall1e883282008-02-04 22:29:07 -0800858#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800859
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700860#ifdef CONFIG_NUMA
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700861
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700862struct numa_maps {
863 struct vm_area_struct *vma;
864 unsigned long pages;
865 unsigned long anon;
866 unsigned long active;
867 unsigned long writeback;
868 unsigned long mapcount_max;
869 unsigned long dirty;
870 unsigned long swapcache;
871 unsigned long node[MAX_NUMNODES];
872};
873
Stephen Wilson5b52fc82011-05-24 17:12:49 -0700874struct numa_maps_private {
875 struct proc_maps_private proc_maps;
876 struct numa_maps md;
877};
878
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700879static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty)
880{
881 int count = page_mapcount(page);
882
883 md->pages++;
884 if (pte_dirty || PageDirty(page))
885 md->dirty++;
886
887 if (PageSwapCache(page))
888 md->swapcache++;
889
890 if (PageActive(page) || PageUnevictable(page))
891 md->active++;
892
893 if (PageWriteback(page))
894 md->writeback++;
895
896 if (PageAnon(page))
897 md->anon++;
898
899 if (count > md->mapcount_max)
900 md->mapcount_max = count;
901
902 md->node[page_to_nid(page)]++;
903}
904
905static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
906 unsigned long end, struct mm_walk *walk)
907{
908 struct numa_maps *md;
909 spinlock_t *ptl;
910 pte_t *orig_pte;
911 pte_t *pte;
912
913 md = walk->private;
914 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
915 do {
916 struct page *page;
917 int nid;
918
919 if (!pte_present(*pte))
920 continue;
921
922 page = vm_normal_page(md->vma, addr, *pte);
923 if (!page)
924 continue;
925
926 if (PageReserved(page))
927 continue;
928
929 nid = page_to_nid(page);
930 if (!node_isset(nid, node_states[N_HIGH_MEMORY]))
931 continue;
932
933 gather_stats(page, md, pte_dirty(*pte));
934
935 } while (pte++, addr += PAGE_SIZE, addr != end);
936 pte_unmap_unlock(orig_pte, ptl);
937 return 0;
938}
939#ifdef CONFIG_HUGETLB_PAGE
940static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
941 unsigned long addr, unsigned long end, struct mm_walk *walk)
942{
943 struct numa_maps *md;
944 struct page *page;
945
946 if (pte_none(*pte))
947 return 0;
948
949 page = pte_page(*pte);
950 if (!page)
951 return 0;
952
953 md = walk->private;
954 gather_stats(page, md, pte_dirty(*pte));
955 return 0;
956}
957
958#else
959static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask,
960 unsigned long addr, unsigned long end, struct mm_walk *walk)
961{
962 return 0;
963}
964#endif
965
966/*
967 * Display pages allocated per node and memory policy via /proc.
968 */
969static int show_numa_map(struct seq_file *m, void *v)
970{
Stephen Wilson5b52fc82011-05-24 17:12:49 -0700971 struct numa_maps_private *numa_priv = m->private;
972 struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700973 struct vm_area_struct *vma = v;
Stephen Wilson5b52fc82011-05-24 17:12:49 -0700974 struct numa_maps *md = &numa_priv->md;
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700975 struct file *file = vma->vm_file;
976 struct mm_struct *mm = vma->vm_mm;
977 struct mm_walk walk = {};
978 struct mempolicy *pol;
979 int n;
980 char buffer[50];
981
982 if (!mm)
983 return 0;
984
Stephen Wilson5b52fc82011-05-24 17:12:49 -0700985 /* Ensure we start with an empty set of numa_maps statistics. */
986 memset(md, 0, sizeof(*md));
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700987
988 md->vma = vma;
989
990 walk.hugetlb_entry = gather_hugetbl_stats;
991 walk.pmd_entry = gather_pte_stats;
992 walk.private = md;
993 walk.mm = mm;
994
Stephen Wilson5b52fc82011-05-24 17:12:49 -0700995 pol = get_vma_policy(proc_priv->task, vma, vma->vm_start);
Stephen Wilsonf69ff942011-05-24 17:12:47 -0700996 mpol_to_str(buffer, sizeof(buffer), pol, 0);
997 mpol_cond_put(pol);
998
999 seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1000
1001 if (file) {
1002 seq_printf(m, " file=");
1003 seq_path(m, &file->f_path, "\n\t= ");
1004 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1005 seq_printf(m, " heap");
1006 } else if (vma->vm_start <= mm->start_stack &&
1007 vma->vm_end >= mm->start_stack) {
1008 seq_printf(m, " stack");
1009 }
1010
1011 walk_page_range(vma->vm_start, vma->vm_end, &walk);
1012
1013 if (!md->pages)
1014 goto out;
1015
1016 if (md->anon)
1017 seq_printf(m, " anon=%lu", md->anon);
1018
1019 if (md->dirty)
1020 seq_printf(m, " dirty=%lu", md->dirty);
1021
1022 if (md->pages != md->anon && md->pages != md->dirty)
1023 seq_printf(m, " mapped=%lu", md->pages);
1024
1025 if (md->mapcount_max > 1)
1026 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1027
1028 if (md->swapcache)
1029 seq_printf(m, " swapcache=%lu", md->swapcache);
1030
1031 if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1032 seq_printf(m, " active=%lu", md->active);
1033
1034 if (md->writeback)
1035 seq_printf(m, " writeback=%lu", md->writeback);
1036
1037 for_each_node_state(n, N_HIGH_MEMORY)
1038 if (md->node[n])
1039 seq_printf(m, " N%d=%lu", n, md->node[n]);
1040out:
1041 seq_putc(m, '\n');
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001042
1043 if (m->count < m->size)
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001044 m->version = (vma != proc_priv->tail_vma) ? vma->vm_start : 0;
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001045 return 0;
1046}
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001047
Jan Engelhardt03a44822008-02-08 04:21:19 -08001048static const struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -08001049 .start = m_start,
1050 .next = m_next,
1051 .stop = m_stop,
Alexey Dobriyan3bbfe052008-10-10 03:27:16 +04001052 .show = show_numa_map,
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07001053};
Eric W. Biederman662795d2006-06-26 00:25:48 -07001054
1055static int numa_maps_open(struct inode *inode, struct file *file)
1056{
Stephen Wilson5b52fc82011-05-24 17:12:49 -07001057 struct numa_maps_private *priv;
1058 int ret = -ENOMEM;
1059 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
1060 if (priv) {
1061 priv->proc_maps.pid = proc_pid(inode);
1062 ret = seq_open(file, &proc_pid_numa_maps_op);
1063 if (!ret) {
1064 struct seq_file *m = file->private_data;
1065 m->private = priv;
1066 } else {
1067 kfree(priv);
1068 }
1069 }
1070 return ret;
Eric W. Biederman662795d2006-06-26 00:25:48 -07001071}
1072
Arjan van de Ven00977a52007-02-12 00:55:34 -08001073const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -07001074 .open = numa_maps_open,
1075 .read = seq_read,
1076 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -07001077 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -07001078};
Stephen Wilsonf69ff942011-05-24 17:12:47 -07001079#endif /* CONFIG_NUMA */