blob: c126c83b9a45bc838125d1ac12b44e13f89d8b26 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001#include <linux/mm.h>
2#include <linux/hugetlb.h>
3#include <linux/mount.h>
4#include <linux/seq_file.h>
Mauricio Line070ad42005-09-03 15:55:10 -07005#include <linux/highmem.h>
Kees Cook5096add2007-05-08 00:26:04 -07006#include <linux/ptrace.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09007#include <linux/slab.h>
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07008#include <linux/pagemap.h>
9#include <linux/mempolicy.h>
Matt Mackall85863e42008-02-04 22:29:04 -080010#include <linux/swap.h>
11#include <linux/swapops.h>
Mauricio Line070ad42005-09-03 15:55:10 -070012
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <asm/elf.h>
14#include <asm/uaccess.h>
Mauricio Line070ad42005-09-03 15:55:10 -070015#include <asm/tlbflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include "internal.h"
17
Eric W. Biedermandf5f8312008-02-08 04:18:33 -080018void task_mem(struct seq_file *m, struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -070019{
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080020 unsigned long data, text, lib, swap;
Hugh Dickins365e9c872005-10-29 18:16:18 -070021 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
22
23 /*
24 * Note: to minimize their overhead, mm maintains hiwater_vm and
25 * hiwater_rss only when about to *lower* total_vm or rss. Any
26 * collector of these hiwater stats must therefore get total_vm
27 * and rss too, which will usually be the higher. Barriers? not
28 * worth the effort, such snapshots can always be inconsistent.
29 */
30 hiwater_vm = total_vm = mm->total_vm;
31 if (hiwater_vm < mm->hiwater_vm)
32 hiwater_vm = mm->hiwater_vm;
33 hiwater_rss = total_rss = get_mm_rss(mm);
34 if (hiwater_rss < mm->hiwater_rss)
35 hiwater_rss = mm->hiwater_rss;
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
38 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
39 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080040 swap = get_mm_counter(mm, MM_SWAPENTS);
Eric W. Biedermandf5f8312008-02-08 04:18:33 -080041 seq_printf(m,
Hugh Dickins365e9c872005-10-29 18:16:18 -070042 "VmPeak:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 "VmSize:\t%8lu kB\n"
44 "VmLck:\t%8lu kB\n"
Hugh Dickins365e9c872005-10-29 18:16:18 -070045 "VmHWM:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046 "VmRSS:\t%8lu kB\n"
47 "VmData:\t%8lu kB\n"
48 "VmStk:\t%8lu kB\n"
49 "VmExe:\t%8lu kB\n"
50 "VmLib:\t%8lu kB\n"
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080051 "VmPTE:\t%8lu kB\n"
52 "VmSwap:\t%8lu kB\n",
Hugh Dickins365e9c872005-10-29 18:16:18 -070053 hiwater_vm << (PAGE_SHIFT-10),
54 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 mm->locked_vm << (PAGE_SHIFT-10),
Hugh Dickins365e9c872005-10-29 18:16:18 -070056 hiwater_rss << (PAGE_SHIFT-10),
57 total_rss << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 data << (PAGE_SHIFT-10),
59 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
KAMEZAWA Hiroyukib084d432010-03-05 13:41:42 -080060 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
61 swap << (PAGE_SHIFT-10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070062}
63
64unsigned long task_vsize(struct mm_struct *mm)
65{
66 return PAGE_SIZE * mm->total_vm;
67}
68
69int task_statm(struct mm_struct *mm, int *shared, int *text,
70 int *data, int *resident)
71{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080072 *shared = get_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
74 >> PAGE_SHIFT;
75 *data = mm->total_vm - mm->shared_vm;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080076 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 return mm->total_vm;
78}
79
Linus Torvalds1da177e2005-04-16 15:20:36 -070080static void pad_len_spaces(struct seq_file *m, int len)
81{
82 len = 25 + sizeof(void*) * 6 - len;
83 if (len < 1)
84 len = 1;
85 seq_printf(m, "%*c", len, ' ');
86}
87
Matt Mackalla6198792008-02-04 22:29:03 -080088static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
Mauricio Line070ad42005-09-03 15:55:10 -070089{
Matt Mackalla6198792008-02-04 22:29:03 -080090 if (vma && vma != priv->tail_vma) {
91 struct mm_struct *mm = vma->vm_mm;
92 up_read(&mm->mmap_sem);
93 mmput(mm);
94 }
95}
96
97static void *m_start(struct seq_file *m, loff_t *pos)
98{
99 struct proc_maps_private *priv = m->private;
100 unsigned long last_addr = m->version;
101 struct mm_struct *mm;
102 struct vm_area_struct *vma, *tail_vma = NULL;
103 loff_t l = *pos;
104
105 /* Clear the per syscall fields in priv */
106 priv->task = NULL;
107 priv->tail_vma = NULL;
108
109 /*
110 * We remember last_addr rather than next_addr to hit with
111 * mmap_cache most of the time. We have zero last_addr at
112 * the beginning and also after lseek. We will have -1 last_addr
113 * after the end of the vmas.
114 */
115
116 if (last_addr == -1UL)
117 return NULL;
118
119 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
120 if (!priv->task)
121 return NULL;
122
123 mm = mm_for_maps(priv->task);
124 if (!mm)
125 return NULL;
Oleg Nesterov00f89d22009-07-10 03:27:38 +0200126 down_read(&mm->mmap_sem);
Matt Mackalla6198792008-02-04 22:29:03 -0800127
128 tail_vma = get_gate_vma(priv->task);
129 priv->tail_vma = tail_vma;
130
131 /* Start with last addr hint */
132 vma = find_vma(mm, last_addr);
133 if (last_addr && vma) {
134 vma = vma->vm_next;
135 goto out;
136 }
137
138 /*
139 * Check the vma index is within the range and do
140 * sequential scan until m_index.
141 */
142 vma = NULL;
143 if ((unsigned long)l < mm->map_count) {
144 vma = mm->mmap;
145 while (l-- && vma)
146 vma = vma->vm_next;
147 goto out;
148 }
149
150 if (l != mm->map_count)
151 tail_vma = NULL; /* After gate vma */
152
153out:
154 if (vma)
155 return vma;
156
157 /* End of vmas has been reached */
158 m->version = (tail_vma != NULL)? 0: -1UL;
159 up_read(&mm->mmap_sem);
160 mmput(mm);
161 return tail_vma;
162}
163
164static void *m_next(struct seq_file *m, void *v, loff_t *pos)
165{
166 struct proc_maps_private *priv = m->private;
167 struct vm_area_struct *vma = v;
168 struct vm_area_struct *tail_vma = priv->tail_vma;
169
170 (*pos)++;
171 if (vma && (vma != tail_vma) && vma->vm_next)
172 return vma->vm_next;
173 vma_stop(priv, vma);
174 return (vma != tail_vma)? tail_vma: NULL;
175}
176
177static void m_stop(struct seq_file *m, void *v)
178{
179 struct proc_maps_private *priv = m->private;
180 struct vm_area_struct *vma = v;
181
182 vma_stop(priv, vma);
183 if (priv->task)
184 put_task_struct(priv->task);
185}
186
187static int do_maps_open(struct inode *inode, struct file *file,
Jan Engelhardt03a44822008-02-08 04:21:19 -0800188 const struct seq_operations *ops)
Matt Mackalla6198792008-02-04 22:29:03 -0800189{
190 struct proc_maps_private *priv;
191 int ret = -ENOMEM;
192 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
193 if (priv) {
194 priv->pid = proc_pid(inode);
195 ret = seq_open(file, ops);
196 if (!ret) {
197 struct seq_file *m = file->private_data;
198 m->private = priv;
199 } else {
200 kfree(priv);
201 }
202 }
203 return ret;
204}
Mauricio Line070ad42005-09-03 15:55:10 -0700205
Joe Korty7c88db02008-10-16 15:27:09 +0400206static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207{
Mauricio Line070ad42005-09-03 15:55:10 -0700208 struct mm_struct *mm = vma->vm_mm;
209 struct file *file = vma->vm_file;
210 int flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 unsigned long ino = 0;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700212 unsigned long long pgoff = 0;
Linus Torvaldsd7824372010-08-15 11:35:52 -0700213 unsigned long start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 dev_t dev = 0;
215 int len;
216
217 if (file) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800218 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 dev = inode->i_sb->s_dev;
220 ino = inode->i_ino;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700221 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 }
223
Linus Torvaldsd7824372010-08-15 11:35:52 -0700224 /* We don't show the stack guard page in /proc/maps */
225 start = vma->vm_start;
226 if (vma->vm_flags & VM_GROWSDOWN)
Stefan Bader39aa3cb2010-08-31 15:52:27 +0200227 if (!vma_stack_continue(vma->vm_prev, vma->vm_start))
228 start += PAGE_SIZE;
Linus Torvaldsd7824372010-08-15 11:35:52 -0700229
Clement Calmels1804dc62008-08-20 14:09:00 -0700230 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
Linus Torvaldsd7824372010-08-15 11:35:52 -0700231 start,
Mauricio Line070ad42005-09-03 15:55:10 -0700232 vma->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 flags & VM_READ ? 'r' : '-',
234 flags & VM_WRITE ? 'w' : '-',
235 flags & VM_EXEC ? 'x' : '-',
236 flags & VM_MAYSHARE ? 's' : 'p',
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700237 pgoff,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 MAJOR(dev), MINOR(dev), ino, &len);
239
240 /*
241 * Print the dentry name for named mappings, and a
242 * special [heap] marker for the heap:
243 */
Mauricio Line070ad42005-09-03 15:55:10 -0700244 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 pad_len_spaces(m, len);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800246 seq_path(m, &file->f_path, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 } else {
Ingo Molnare6e54942006-06-27 02:53:50 -0700248 const char *name = arch_vma_name(vma);
249 if (!name) {
250 if (mm) {
251 if (vma->vm_start <= mm->start_brk &&
Mauricio Line070ad42005-09-03 15:55:10 -0700252 vma->vm_end >= mm->brk) {
Ingo Molnare6e54942006-06-27 02:53:50 -0700253 name = "[heap]";
254 } else if (vma->vm_start <= mm->start_stack &&
255 vma->vm_end >= mm->start_stack) {
256 name = "[stack]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700258 } else {
259 name = "[vdso]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700261 }
262 if (name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 pad_len_spaces(m, len);
Ingo Molnare6e54942006-06-27 02:53:50 -0700264 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
266 }
267 seq_putc(m, '\n');
Joe Korty7c88db02008-10-16 15:27:09 +0400268}
269
270static int show_map(struct seq_file *m, void *v)
271{
272 struct vm_area_struct *vma = v;
273 struct proc_maps_private *priv = m->private;
274 struct task_struct *task = priv->task;
275
276 show_map_vma(m, vma);
Mauricio Line070ad42005-09-03 15:55:10 -0700277
Mauricio Line070ad42005-09-03 15:55:10 -0700278 if (m->count < m->size) /* vma is copied successfully */
279 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 return 0;
281}
282
Jan Engelhardt03a44822008-02-08 04:21:19 -0800283static const struct seq_operations proc_pid_maps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800284 .start = m_start,
285 .next = m_next,
286 .stop = m_stop,
287 .show = show_map
288};
289
290static int maps_open(struct inode *inode, struct file *file)
291{
292 return do_maps_open(inode, file, &proc_pid_maps_op);
293}
294
295const struct file_operations proc_maps_operations = {
296 .open = maps_open,
297 .read = seq_read,
298 .llseek = seq_lseek,
299 .release = seq_release_private,
300};
301
302/*
303 * Proportional Set Size(PSS): my share of RSS.
304 *
305 * PSS of a process is the count of pages it has in memory, where each
306 * page is divided by the number of processes sharing it. So if a
307 * process has 1000 pages all to itself, and 1000 shared with one other
308 * process, its PSS will be 1500.
309 *
310 * To keep (accumulated) division errors low, we adopt a 64bit
311 * fixed-point pss counter to minimize division errors. So (pss >>
312 * PSS_SHIFT) would be the real byte count.
313 *
314 * A shift of 12 before division means (assuming 4K page size):
315 * - 1M 3-user-pages add up to 8KB errors;
316 * - supports mapcount up to 2^24, or 16M;
317 * - supports PSS up to 2^52 bytes, or 4PB.
318 */
319#define PSS_SHIFT 12
320
Matt Mackall1e883282008-02-04 22:29:07 -0800321#ifdef CONFIG_PROC_PAGE_MONITOR
Peter Zijlstra214e4712008-04-28 02:12:55 -0700322struct mem_size_stats {
Matt Mackalla6198792008-02-04 22:29:03 -0800323 struct vm_area_struct *vma;
324 unsigned long resident;
325 unsigned long shared_clean;
326 unsigned long shared_dirty;
327 unsigned long private_clean;
328 unsigned long private_dirty;
329 unsigned long referenced;
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700330 unsigned long anonymous;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700331 unsigned long swap;
Matt Mackalla6198792008-02-04 22:29:03 -0800332 u64 pss;
333};
334
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800335static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700336 struct mm_walk *walk)
Mauricio Line070ad42005-09-03 15:55:10 -0700337{
Dave Hansen21650092008-06-12 15:21:47 -0700338 struct mem_size_stats *mss = walk->private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800339 struct vm_area_struct *vma = mss->vma;
Mauricio Line070ad42005-09-03 15:55:10 -0700340 pte_t *pte, ptent;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700341 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700342 struct page *page;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800343 int mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700344
Hugh Dickins705e87c2005-10-29 18:16:27 -0700345 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
David Rientjes826fad12007-05-06 14:49:21 -0700346 for (; addr != end; pte++, addr += PAGE_SIZE) {
Mauricio Line070ad42005-09-03 15:55:10 -0700347 ptent = *pte;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700348
349 if (is_swap_pte(ptent)) {
350 mss->swap += PAGE_SIZE;
351 continue;
352 }
353
Hugh Dickins705e87c2005-10-29 18:16:27 -0700354 if (!pte_present(ptent))
Mauricio Line070ad42005-09-03 15:55:10 -0700355 continue;
356
Nick Pigginad820c52006-03-06 15:42:58 -0800357 page = vm_normal_page(vma, addr, ptent);
358 if (!page)
Mauricio Line070ad42005-09-03 15:55:10 -0700359 continue;
360
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700361 if (PageAnon(page))
362 mss->anonymous += PAGE_SIZE;
363
Minchan Kim7f53a092010-01-08 14:43:10 -0800364 mss->resident += PAGE_SIZE;
David Rientjesf79f1772007-05-06 14:49:22 -0700365 /* Accumulate the size in pages that have been accessed. */
366 if (pte_young(ptent) || PageReferenced(page))
367 mss->referenced += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800368 mapcount = page_mapcount(page);
369 if (mapcount >= 2) {
KOSAKI Motohiro1c2499a2010-09-22 13:05:06 -0700370 if (pte_dirty(ptent) || PageDirty(page))
Mauricio Line070ad42005-09-03 15:55:10 -0700371 mss->shared_dirty += PAGE_SIZE;
372 else
373 mss->shared_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800374 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700375 } else {
KOSAKI Motohiro1c2499a2010-09-22 13:05:06 -0700376 if (pte_dirty(ptent) || PageDirty(page))
Mauricio Line070ad42005-09-03 15:55:10 -0700377 mss->private_dirty += PAGE_SIZE;
378 else
379 mss->private_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800380 mss->pss += (PAGE_SIZE << PSS_SHIFT);
Mauricio Line070ad42005-09-03 15:55:10 -0700381 }
David Rientjes826fad12007-05-06 14:49:21 -0700382 }
Hugh Dickins705e87c2005-10-29 18:16:27 -0700383 pte_unmap_unlock(pte - 1, ptl);
384 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800385 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700386}
387
Mauricio Line070ad42005-09-03 15:55:10 -0700388static int show_smap(struct seq_file *m, void *v)
389{
Joe Korty7c88db02008-10-16 15:27:09 +0400390 struct proc_maps_private *priv = m->private;
391 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700392 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700393 struct mem_size_stats mss;
Dave Hansen21650092008-06-12 15:21:47 -0700394 struct mm_walk smaps_walk = {
395 .pmd_entry = smaps_pte_range,
396 .mm = vma->vm_mm,
397 .private = &mss,
398 };
Mauricio Line070ad42005-09-03 15:55:10 -0700399
400 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800401 mss.vma = vma;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900402 /* mmap_sem is held in m_start */
Nick Piggin5ddfae12006-03-06 15:42:57 -0800403 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700404 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800405
Joe Korty7c88db02008-10-16 15:27:09 +0400406 show_map_vma(m, vma);
Matt Mackall4752c362008-02-04 22:29:02 -0800407
408 seq_printf(m,
409 "Size: %8lu kB\n"
410 "Rss: %8lu kB\n"
411 "Pss: %8lu kB\n"
412 "Shared_Clean: %8lu kB\n"
413 "Shared_Dirty: %8lu kB\n"
414 "Private_Clean: %8lu kB\n"
415 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700416 "Referenced: %8lu kB\n"
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700417 "Anonymous: %8lu kB\n"
Mel Gorman08fba692009-01-06 14:38:53 -0800418 "Swap: %8lu kB\n"
Mel Gorman33402892009-01-06 14:38:54 -0800419 "KernelPageSize: %8lu kB\n"
420 "MMUPageSize: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800421 (vma->vm_end - vma->vm_start) >> 10,
422 mss.resident >> 10,
423 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
424 mss.shared_clean >> 10,
425 mss.shared_dirty >> 10,
426 mss.private_clean >> 10,
427 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700428 mss.referenced >> 10,
Nikanth Karthikesanb40d4f82010-10-27 15:34:10 -0700429 mss.anonymous >> 10,
Mel Gorman08fba692009-01-06 14:38:53 -0800430 mss.swap >> 10,
Mel Gorman33402892009-01-06 14:38:54 -0800431 vma_kernel_pagesize(vma) >> 10,
432 vma_mmu_pagesize(vma) >> 10);
Matt Mackall4752c362008-02-04 22:29:02 -0800433
Joe Korty7c88db02008-10-16 15:27:09 +0400434 if (m->count < m->size) /* vma is copied successfully */
435 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
436 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700437}
438
Jan Engelhardt03a44822008-02-08 04:21:19 -0800439static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800440 .start = m_start,
441 .next = m_next,
442 .stop = m_stop,
443 .show = show_smap
444};
445
446static int smaps_open(struct inode *inode, struct file *file)
447{
448 return do_maps_open(inode, file, &proc_pid_smaps_op);
449}
450
451const struct file_operations proc_smaps_operations = {
452 .open = smaps_open,
453 .read = seq_read,
454 .llseek = seq_lseek,
455 .release = seq_release_private,
456};
457
458static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700459 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800460{
Dave Hansen21650092008-06-12 15:21:47 -0700461 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800462 pte_t *pte, ptent;
463 spinlock_t *ptl;
464 struct page *page;
465
466 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
467 for (; addr != end; pte++, addr += PAGE_SIZE) {
468 ptent = *pte;
469 if (!pte_present(ptent))
470 continue;
471
472 page = vm_normal_page(vma, addr, ptent);
473 if (!page)
474 continue;
475
476 /* Clear accessed and referenced bits. */
477 ptep_test_and_clear_young(vma, addr, pte);
478 ClearPageReferenced(page);
479 }
480 pte_unmap_unlock(pte - 1, ptl);
481 cond_resched();
482 return 0;
483}
484
Moussa A. Ba398499d2009-09-21 17:02:29 -0700485#define CLEAR_REFS_ALL 1
486#define CLEAR_REFS_ANON 2
487#define CLEAR_REFS_MAPPED 3
488
Matt Mackallf248dcb2008-02-04 22:29:03 -0800489static ssize_t clear_refs_write(struct file *file, const char __user *buf,
490 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700491{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800492 struct task_struct *task;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700493 char buffer[PROC_NUMBUF];
Matt Mackallf248dcb2008-02-04 22:29:03 -0800494 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700495 struct vm_area_struct *vma;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700496 long type;
David Rientjesb813e932007-05-06 14:49:24 -0700497
Matt Mackallf248dcb2008-02-04 22:29:03 -0800498 memset(buffer, 0, sizeof(buffer));
499 if (count > sizeof(buffer) - 1)
500 count = sizeof(buffer) - 1;
501 if (copy_from_user(buffer, buf, count))
502 return -EFAULT;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700503 if (strict_strtol(strstrip(buffer), 10, &type))
504 return -EINVAL;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700505 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
Matt Mackallf248dcb2008-02-04 22:29:03 -0800506 return -EINVAL;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800507 task = get_proc_task(file->f_path.dentry->d_inode);
508 if (!task)
509 return -ESRCH;
510 mm = get_task_mm(task);
511 if (mm) {
Andrew Morton20cbc972008-07-05 12:29:05 -0700512 struct mm_walk clear_refs_walk = {
513 .pmd_entry = clear_refs_pte_range,
514 .mm = mm,
515 };
Matt Mackallf248dcb2008-02-04 22:29:03 -0800516 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700517 for (vma = mm->mmap; vma; vma = vma->vm_next) {
518 clear_refs_walk.private = vma;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700519 if (is_vm_hugetlb_page(vma))
520 continue;
521 /*
522 * Writing 1 to /proc/pid/clear_refs affects all pages.
523 *
524 * Writing 2 to /proc/pid/clear_refs only affects
525 * Anonymous pages.
526 *
527 * Writing 3 to /proc/pid/clear_refs only affects file
528 * mapped pages.
529 */
530 if (type == CLEAR_REFS_ANON && vma->vm_file)
531 continue;
532 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
533 continue;
534 walk_page_range(vma->vm_start, vma->vm_end,
535 &clear_refs_walk);
Dave Hansen21650092008-06-12 15:21:47 -0700536 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800537 flush_tlb_mm(mm);
538 up_read(&mm->mmap_sem);
539 mmput(mm);
540 }
541 put_task_struct(task);
Vincent Lifb92a4b2009-09-22 16:45:36 -0700542
543 return count;
David Rientjesb813e932007-05-06 14:49:24 -0700544}
545
Matt Mackallf248dcb2008-02-04 22:29:03 -0800546const struct file_operations proc_clear_refs_operations = {
547 .write = clear_refs_write,
Arnd Bergmann6038f372010-08-15 18:52:59 +0200548 .llseek = noop_llseek,
Matt Mackallf248dcb2008-02-04 22:29:03 -0800549};
550
Matt Mackall85863e42008-02-04 22:29:04 -0800551struct pagemapread {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900552 int pos, len;
553 u64 *buffer;
Matt Mackall85863e42008-02-04 22:29:04 -0800554};
555
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500556#define PM_ENTRY_BYTES sizeof(u64)
557#define PM_STATUS_BITS 3
558#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
559#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
560#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
561#define PM_PSHIFT_BITS 6
562#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
563#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
564#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
565#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
566#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
567
568#define PM_PRESENT PM_STATUS(4LL)
569#define PM_SWAP PM_STATUS(2LL)
570#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800571#define PM_END_OF_BUFFER 1
572
573static int add_to_pagemap(unsigned long addr, u64 pfn,
574 struct pagemapread *pm)
575{
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900576 pm->buffer[pm->pos++] = pfn;
577 if (pm->pos >= pm->len)
Thomas Tuttleaae86792008-06-05 22:46:31 -0700578 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800579 return 0;
580}
581
582static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700583 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800584{
Dave Hansen21650092008-06-12 15:21:47 -0700585 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800586 unsigned long addr;
587 int err = 0;
588 for (addr = start; addr < end; addr += PAGE_SIZE) {
589 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
590 if (err)
591 break;
592 }
593 return err;
594}
595
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700596static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800597{
598 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500599 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800600}
601
Matt Mackall49c50342008-12-09 13:14:21 -0800602static u64 pte_to_pagemap_entry(pte_t pte)
Dave Hansenbcf80392008-06-12 15:21:48 -0700603{
Matt Mackall49c50342008-12-09 13:14:21 -0800604 u64 pme = 0;
Dave Hansenbcf80392008-06-12 15:21:48 -0700605 if (is_swap_pte(pte))
606 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
607 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
608 else if (pte_present(pte))
609 pme = PM_PFRAME(pte_pfn(pte))
610 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
611 return pme;
612}
613
Matt Mackall85863e42008-02-04 22:29:04 -0800614static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700615 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800616{
Dave Hansenbcf80392008-06-12 15:21:48 -0700617 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700618 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800619 pte_t *pte;
620 int err = 0;
621
Dave Hansenbcf80392008-06-12 15:21:48 -0700622 /* find the first VMA at or above 'addr' */
623 vma = find_vma(walk->mm, addr);
Matt Mackall85863e42008-02-04 22:29:04 -0800624 for (; addr != end; addr += PAGE_SIZE) {
625 u64 pfn = PM_NOT_PRESENT;
Dave Hansenbcf80392008-06-12 15:21:48 -0700626
627 /* check to see if we've left 'vma' behind
628 * and need a new, higher one */
629 if (vma && (addr >= vma->vm_end))
630 vma = find_vma(walk->mm, addr);
631
632 /* check that 'vma' actually covers this address,
633 * and that it isn't a huge page vma */
634 if (vma && (vma->vm_start <= addr) &&
635 !is_vm_hugetlb_page(vma)) {
636 pte = pte_offset_map(pmd, addr);
637 pfn = pte_to_pagemap_entry(*pte);
638 /* unmap before userspace copy */
639 pte_unmap(pte);
640 }
Matt Mackall85863e42008-02-04 22:29:04 -0800641 err = add_to_pagemap(addr, pfn, pm);
642 if (err)
643 return err;
644 }
645
646 cond_resched();
647
648 return err;
649}
650
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700651#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800652static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
653{
654 u64 pme = 0;
655 if (pte_present(pte))
656 pme = PM_PFRAME(pte_pfn(pte) + offset)
657 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
658 return pme;
659}
660
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700661/* This function walks within one hugetlb entry in the single call */
662static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
663 unsigned long addr, unsigned long end,
664 struct mm_walk *walk)
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800665{
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800666 struct pagemapread *pm = walk->private;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800667 int err = 0;
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700668 u64 pfn;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800669
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800670 for (; addr != end; addr += PAGE_SIZE) {
Naoya Horiguchi116354d2010-04-06 14:35:04 -0700671 int offset = (addr & ~hmask) >> PAGE_SHIFT;
672 pfn = huge_pte_to_pagemap_entry(*pte, offset);
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800673 err = add_to_pagemap(addr, pfn, pm);
674 if (err)
675 return err;
676 }
677
678 cond_resched();
679
680 return err;
681}
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700682#endif /* HUGETLB_PAGE */
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800683
Matt Mackall85863e42008-02-04 22:29:04 -0800684/*
685 * /proc/pid/pagemap - an array mapping virtual pages to pfns
686 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500687 * For each page in the address space, this file contains one 64-bit entry
688 * consisting of the following:
689 *
690 * Bits 0-55 page frame number (PFN) if present
691 * Bits 0-4 swap type if swapped
692 * Bits 5-55 swap offset if swapped
693 * Bits 55-60 page shift (page size = 1<<page shift)
694 * Bit 61 reserved for future use
695 * Bit 62 page swapped
696 * Bit 63 page present
697 *
698 * If the page is not present but in swap, then the PFN contains an
699 * encoding of the swap file number and the page's offset into the
700 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800701 * precisely which pages are mapped (or in swap) and comparing mapped
702 * pages between processes.
703 *
704 * Efficient users of this interface will use /proc/pid/maps to
705 * determine which areas of memory are actually mapped and llseek to
706 * skip over unmapped regions.
707 */
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900708#define PAGEMAP_WALK_SIZE (PMD_SIZE)
Naoya Horiguchiea251c12010-11-24 12:57:13 -0800709#define PAGEMAP_WALK_MASK (PMD_MASK)
Matt Mackall85863e42008-02-04 22:29:04 -0800710static ssize_t pagemap_read(struct file *file, char __user *buf,
711 size_t count, loff_t *ppos)
712{
713 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
Matt Mackall85863e42008-02-04 22:29:04 -0800714 struct mm_struct *mm;
715 struct pagemapread pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800716 int ret = -ESRCH;
Alexey Dobriyanee1e6ab2008-07-21 14:21:36 -0700717 struct mm_walk pagemap_walk = {};
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700718 unsigned long src;
719 unsigned long svpfn;
720 unsigned long start_vaddr;
721 unsigned long end_vaddr;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900722 int copied = 0;
Matt Mackall85863e42008-02-04 22:29:04 -0800723
724 if (!task)
725 goto out;
726
727 ret = -EACCES;
Stephen Smalley006ebb42008-05-19 08:32:49 -0400728 if (!ptrace_may_access(task, PTRACE_MODE_READ))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700729 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800730
731 ret = -EINVAL;
732 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700733 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700734 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800735
736 ret = 0;
Vitaly Mayatskikh08161782009-04-30 15:08:18 -0700737
738 if (!count)
739 goto out_task;
740
Matt Mackall85863e42008-02-04 22:29:04 -0800741 mm = get_task_mm(task);
742 if (!mm)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700743 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800744
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900745 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
746 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700747 ret = -ENOMEM;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900748 if (!pm.buffer)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700749 goto out_mm;
Matt Mackall85863e42008-02-04 22:29:04 -0800750
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700751 pagemap_walk.pmd_entry = pagemap_pte_range;
752 pagemap_walk.pte_hole = pagemap_pte_hole;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700753#ifdef CONFIG_HUGETLB_PAGE
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800754 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
Naoya Horiguchi1a5cb812010-05-24 14:32:12 -0700755#endif
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700756 pagemap_walk.mm = mm;
757 pagemap_walk.private = &pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800758
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700759 src = *ppos;
760 svpfn = src / PM_ENTRY_BYTES;
761 start_vaddr = svpfn << PAGE_SHIFT;
762 end_vaddr = TASK_SIZE_OF(task);
Matt Mackall85863e42008-02-04 22:29:04 -0800763
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700764 /* watch out for wraparound */
765 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
766 start_vaddr = end_vaddr;
767
768 /*
769 * The odds are that this will stop walking way
770 * before end_vaddr, because the length of the
771 * user buffer is tracked in "pm", and the walk
772 * will stop when we hit the end of the buffer.
773 */
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900774 ret = 0;
775 while (count && (start_vaddr < end_vaddr)) {
776 int len;
777 unsigned long end;
Matt Mackall85863e42008-02-04 22:29:04 -0800778
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900779 pm.pos = 0;
Naoya Horiguchiea251c12010-11-24 12:57:13 -0800780 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900781 /* overflow ? */
782 if (end < start_vaddr || end > end_vaddr)
783 end = end_vaddr;
784 down_read(&mm->mmap_sem);
785 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
786 up_read(&mm->mmap_sem);
787 start_vaddr = end;
788
789 len = min(count, PM_ENTRY_BYTES * pm.pos);
Dan Carpenter309361e2010-04-06 13:45:39 +0300790 if (copy_to_user(buf, pm.buffer, len)) {
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900791 ret = -EFAULT;
792 goto out_free;
793 }
794 copied += len;
795 buf += len;
796 count -= len;
Matt Mackall85863e42008-02-04 22:29:04 -0800797 }
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900798 *ppos += copied;
799 if (!ret || ret == PM_END_OF_BUFFER)
800 ret = copied;
801
Matt Mackall85863e42008-02-04 22:29:04 -0800802out_free:
KAMEZAWA Hiroyukid82ef022010-04-02 09:11:29 +0900803 kfree(pm.buffer);
Marcelo Tosattifb393802008-03-13 12:32:35 -0700804out_mm:
805 mmput(mm);
Matt Mackall85863e42008-02-04 22:29:04 -0800806out_task:
807 put_task_struct(task);
808out:
809 return ret;
810}
811
812const struct file_operations proc_pagemap_operations = {
813 .llseek = mem_lseek, /* borrow this */
814 .read = pagemap_read,
815};
Matt Mackall1e883282008-02-04 22:29:07 -0800816#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800817
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700818#ifdef CONFIG_NUMA
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800819extern int show_numa_map(struct seq_file *m, void *v);
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700820
Jan Engelhardt03a44822008-02-08 04:21:19 -0800821static const struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800822 .start = m_start,
823 .next = m_next,
824 .stop = m_stop,
Alexey Dobriyan3bbfe052008-10-10 03:27:16 +0400825 .show = show_numa_map,
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700826};
Eric W. Biederman662795d2006-06-26 00:25:48 -0700827
828static int numa_maps_open(struct inode *inode, struct file *file)
829{
830 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
831}
832
Arjan van de Ven00977a52007-02-12 00:55:34 -0800833const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -0700834 .open = numa_maps_open,
835 .read = seq_read,
836 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -0700837 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -0700838};
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700839#endif