blob: 37558127601180a48c39a84cd9794d69a2bc08df [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>
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07007#include <linux/pagemap.h>
8#include <linux/mempolicy.h>
Matt Mackall85863e42008-02-04 22:29:04 -08009#include <linux/swap.h>
10#include <linux/swapops.h>
Mauricio Line070ad42005-09-03 15:55:10 -070011
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <asm/elf.h>
13#include <asm/uaccess.h>
Mauricio Line070ad42005-09-03 15:55:10 -070014#include <asm/tlbflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include "internal.h"
16
Eric W. Biedermandf5f8312008-02-08 04:18:33 -080017void task_mem(struct seq_file *m, struct mm_struct *mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -070018{
19 unsigned long data, text, lib;
Hugh Dickins365e9c872005-10-29 18:16:18 -070020 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
21
22 /*
23 * Note: to minimize their overhead, mm maintains hiwater_vm and
24 * hiwater_rss only when about to *lower* total_vm or rss. Any
25 * collector of these hiwater stats must therefore get total_vm
26 * and rss too, which will usually be the higher. Barriers? not
27 * worth the effort, such snapshots can always be inconsistent.
28 */
29 hiwater_vm = total_vm = mm->total_vm;
30 if (hiwater_vm < mm->hiwater_vm)
31 hiwater_vm = mm->hiwater_vm;
32 hiwater_rss = total_rss = get_mm_rss(mm);
33 if (hiwater_rss < mm->hiwater_rss)
34 hiwater_rss = mm->hiwater_rss;
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
37 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
38 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
Eric W. Biedermandf5f8312008-02-08 04:18:33 -080039 seq_printf(m,
Hugh Dickins365e9c872005-10-29 18:16:18 -070040 "VmPeak:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 "VmSize:\t%8lu kB\n"
42 "VmLck:\t%8lu kB\n"
Hugh Dickins365e9c872005-10-29 18:16:18 -070043 "VmHWM:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 "VmRSS:\t%8lu kB\n"
45 "VmData:\t%8lu kB\n"
46 "VmStk:\t%8lu kB\n"
47 "VmExe:\t%8lu kB\n"
48 "VmLib:\t%8lu kB\n"
49 "VmPTE:\t%8lu kB\n",
Hugh Dickins365e9c872005-10-29 18:16:18 -070050 hiwater_vm << (PAGE_SHIFT-10),
51 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070052 mm->locked_vm << (PAGE_SHIFT-10),
Hugh Dickins365e9c872005-10-29 18:16:18 -070053 hiwater_rss << (PAGE_SHIFT-10),
54 total_rss << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070055 data << (PAGE_SHIFT-10),
56 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
57 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058}
59
60unsigned long task_vsize(struct mm_struct *mm)
61{
62 return PAGE_SIZE * mm->total_vm;
63}
64
65int task_statm(struct mm_struct *mm, int *shared, int *text,
66 int *data, int *resident)
67{
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080068 *shared = get_mm_counter(mm, MM_FILEPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
70 >> PAGE_SHIFT;
71 *data = mm->total_vm - mm->shared_vm;
KAMEZAWA Hiroyukid559db02010-03-05 13:41:39 -080072 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 return mm->total_vm;
74}
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076static void pad_len_spaces(struct seq_file *m, int len)
77{
78 len = 25 + sizeof(void*) * 6 - len;
79 if (len < 1)
80 len = 1;
81 seq_printf(m, "%*c", len, ' ');
82}
83
Matt Mackalla6198792008-02-04 22:29:03 -080084static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
Mauricio Line070ad42005-09-03 15:55:10 -070085{
Matt Mackalla6198792008-02-04 22:29:03 -080086 if (vma && vma != priv->tail_vma) {
87 struct mm_struct *mm = vma->vm_mm;
88 up_read(&mm->mmap_sem);
89 mmput(mm);
90 }
91}
92
93static void *m_start(struct seq_file *m, loff_t *pos)
94{
95 struct proc_maps_private *priv = m->private;
96 unsigned long last_addr = m->version;
97 struct mm_struct *mm;
98 struct vm_area_struct *vma, *tail_vma = NULL;
99 loff_t l = *pos;
100
101 /* Clear the per syscall fields in priv */
102 priv->task = NULL;
103 priv->tail_vma = NULL;
104
105 /*
106 * We remember last_addr rather than next_addr to hit with
107 * mmap_cache most of the time. We have zero last_addr at
108 * the beginning and also after lseek. We will have -1 last_addr
109 * after the end of the vmas.
110 */
111
112 if (last_addr == -1UL)
113 return NULL;
114
115 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
116 if (!priv->task)
117 return NULL;
118
119 mm = mm_for_maps(priv->task);
120 if (!mm)
121 return NULL;
Oleg Nesterov00f89d22009-07-10 03:27:38 +0200122 down_read(&mm->mmap_sem);
Matt Mackalla6198792008-02-04 22:29:03 -0800123
124 tail_vma = get_gate_vma(priv->task);
125 priv->tail_vma = tail_vma;
126
127 /* Start with last addr hint */
128 vma = find_vma(mm, last_addr);
129 if (last_addr && vma) {
130 vma = vma->vm_next;
131 goto out;
132 }
133
134 /*
135 * Check the vma index is within the range and do
136 * sequential scan until m_index.
137 */
138 vma = NULL;
139 if ((unsigned long)l < mm->map_count) {
140 vma = mm->mmap;
141 while (l-- && vma)
142 vma = vma->vm_next;
143 goto out;
144 }
145
146 if (l != mm->map_count)
147 tail_vma = NULL; /* After gate vma */
148
149out:
150 if (vma)
151 return vma;
152
153 /* End of vmas has been reached */
154 m->version = (tail_vma != NULL)? 0: -1UL;
155 up_read(&mm->mmap_sem);
156 mmput(mm);
157 return tail_vma;
158}
159
160static void *m_next(struct seq_file *m, void *v, loff_t *pos)
161{
162 struct proc_maps_private *priv = m->private;
163 struct vm_area_struct *vma = v;
164 struct vm_area_struct *tail_vma = priv->tail_vma;
165
166 (*pos)++;
167 if (vma && (vma != tail_vma) && vma->vm_next)
168 return vma->vm_next;
169 vma_stop(priv, vma);
170 return (vma != tail_vma)? tail_vma: NULL;
171}
172
173static void m_stop(struct seq_file *m, void *v)
174{
175 struct proc_maps_private *priv = m->private;
176 struct vm_area_struct *vma = v;
177
178 vma_stop(priv, vma);
179 if (priv->task)
180 put_task_struct(priv->task);
181}
182
183static int do_maps_open(struct inode *inode, struct file *file,
Jan Engelhardt03a44822008-02-08 04:21:19 -0800184 const struct seq_operations *ops)
Matt Mackalla6198792008-02-04 22:29:03 -0800185{
186 struct proc_maps_private *priv;
187 int ret = -ENOMEM;
188 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
189 if (priv) {
190 priv->pid = proc_pid(inode);
191 ret = seq_open(file, ops);
192 if (!ret) {
193 struct seq_file *m = file->private_data;
194 m->private = priv;
195 } else {
196 kfree(priv);
197 }
198 }
199 return ret;
200}
Mauricio Line070ad42005-09-03 15:55:10 -0700201
Joe Korty7c88db02008-10-16 15:27:09 +0400202static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
Mauricio Line070ad42005-09-03 15:55:10 -0700204 struct mm_struct *mm = vma->vm_mm;
205 struct file *file = vma->vm_file;
206 int flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 unsigned long ino = 0;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700208 unsigned long long pgoff = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 dev_t dev = 0;
210 int len;
211
212 if (file) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800213 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 dev = inode->i_sb->s_dev;
215 ino = inode->i_ino;
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700216 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 }
218
Clement Calmels1804dc62008-08-20 14:09:00 -0700219 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
Mauricio Line070ad42005-09-03 15:55:10 -0700220 vma->vm_start,
221 vma->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 flags & VM_READ ? 'r' : '-',
223 flags & VM_WRITE ? 'w' : '-',
224 flags & VM_EXEC ? 'x' : '-',
225 flags & VM_MAYSHARE ? 's' : 'p',
KAMEZAWA Hiroyuki6260a4b2009-04-06 19:00:30 -0700226 pgoff,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 MAJOR(dev), MINOR(dev), ino, &len);
228
229 /*
230 * Print the dentry name for named mappings, and a
231 * special [heap] marker for the heap:
232 */
Mauricio Line070ad42005-09-03 15:55:10 -0700233 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 pad_len_spaces(m, len);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800235 seq_path(m, &file->f_path, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 } else {
Ingo Molnare6e54942006-06-27 02:53:50 -0700237 const char *name = arch_vma_name(vma);
238 if (!name) {
239 if (mm) {
240 if (vma->vm_start <= mm->start_brk &&
Mauricio Line070ad42005-09-03 15:55:10 -0700241 vma->vm_end >= mm->brk) {
Ingo Molnare6e54942006-06-27 02:53:50 -0700242 name = "[heap]";
243 } else if (vma->vm_start <= mm->start_stack &&
244 vma->vm_end >= mm->start_stack) {
245 name = "[stack]";
Stefani Seiboldd899bf72009-09-22 16:45:40 -0700246 } else {
247 unsigned long stack_start;
248 struct proc_maps_private *pmp;
249
250 pmp = m->private;
251 stack_start = pmp->task->stack_start;
252
253 if (vma->vm_start <= stack_start &&
254 vma->vm_end >= stack_start) {
255 pad_len_spaces(m, len);
256 seq_printf(m,
257 "[threadstack:%08lx]",
258#ifdef CONFIG_STACK_GROWSUP
259 vma->vm_end - stack_start
260#else
261 stack_start - vma->vm_start
262#endif
263 );
264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700266 } else {
267 name = "[vdso]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700269 }
270 if (name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 pad_len_spaces(m, len);
Ingo Molnare6e54942006-06-27 02:53:50 -0700272 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 }
274 }
275 seq_putc(m, '\n');
Joe Korty7c88db02008-10-16 15:27:09 +0400276}
277
278static int show_map(struct seq_file *m, void *v)
279{
280 struct vm_area_struct *vma = v;
281 struct proc_maps_private *priv = m->private;
282 struct task_struct *task = priv->task;
283
284 show_map_vma(m, vma);
Mauricio Line070ad42005-09-03 15:55:10 -0700285
Mauricio Line070ad42005-09-03 15:55:10 -0700286 if (m->count < m->size) /* vma is copied successfully */
287 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 return 0;
289}
290
Jan Engelhardt03a44822008-02-08 04:21:19 -0800291static const struct seq_operations proc_pid_maps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800292 .start = m_start,
293 .next = m_next,
294 .stop = m_stop,
295 .show = show_map
296};
297
298static int maps_open(struct inode *inode, struct file *file)
299{
300 return do_maps_open(inode, file, &proc_pid_maps_op);
301}
302
303const struct file_operations proc_maps_operations = {
304 .open = maps_open,
305 .read = seq_read,
306 .llseek = seq_lseek,
307 .release = seq_release_private,
308};
309
310/*
311 * Proportional Set Size(PSS): my share of RSS.
312 *
313 * PSS of a process is the count of pages it has in memory, where each
314 * page is divided by the number of processes sharing it. So if a
315 * process has 1000 pages all to itself, and 1000 shared with one other
316 * process, its PSS will be 1500.
317 *
318 * To keep (accumulated) division errors low, we adopt a 64bit
319 * fixed-point pss counter to minimize division errors. So (pss >>
320 * PSS_SHIFT) would be the real byte count.
321 *
322 * A shift of 12 before division means (assuming 4K page size):
323 * - 1M 3-user-pages add up to 8KB errors;
324 * - supports mapcount up to 2^24, or 16M;
325 * - supports PSS up to 2^52 bytes, or 4PB.
326 */
327#define PSS_SHIFT 12
328
Matt Mackall1e883282008-02-04 22:29:07 -0800329#ifdef CONFIG_PROC_PAGE_MONITOR
Peter Zijlstra214e4712008-04-28 02:12:55 -0700330struct mem_size_stats {
Matt Mackalla6198792008-02-04 22:29:03 -0800331 struct vm_area_struct *vma;
332 unsigned long resident;
333 unsigned long shared_clean;
334 unsigned long shared_dirty;
335 unsigned long private_clean;
336 unsigned long private_dirty;
337 unsigned long referenced;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700338 unsigned long swap;
Matt Mackalla6198792008-02-04 22:29:03 -0800339 u64 pss;
340};
341
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800342static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700343 struct mm_walk *walk)
Mauricio Line070ad42005-09-03 15:55:10 -0700344{
Dave Hansen21650092008-06-12 15:21:47 -0700345 struct mem_size_stats *mss = walk->private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800346 struct vm_area_struct *vma = mss->vma;
Mauricio Line070ad42005-09-03 15:55:10 -0700347 pte_t *pte, ptent;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700348 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700349 struct page *page;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800350 int mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700351
Hugh Dickins705e87c2005-10-29 18:16:27 -0700352 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
David Rientjes826fad12007-05-06 14:49:21 -0700353 for (; addr != end; pte++, addr += PAGE_SIZE) {
Mauricio Line070ad42005-09-03 15:55:10 -0700354 ptent = *pte;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700355
356 if (is_swap_pte(ptent)) {
357 mss->swap += PAGE_SIZE;
358 continue;
359 }
360
Hugh Dickins705e87c2005-10-29 18:16:27 -0700361 if (!pte_present(ptent))
Mauricio Line070ad42005-09-03 15:55:10 -0700362 continue;
363
Nick Pigginad820c5d2006-03-06 15:42:58 -0800364 page = vm_normal_page(vma, addr, ptent);
365 if (!page)
Mauricio Line070ad42005-09-03 15:55:10 -0700366 continue;
367
Minchan Kim7f53a092010-01-08 14:43:10 -0800368 mss->resident += PAGE_SIZE;
David Rientjesf79f1772007-05-06 14:49:22 -0700369 /* Accumulate the size in pages that have been accessed. */
370 if (pte_young(ptent) || PageReferenced(page))
371 mss->referenced += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800372 mapcount = page_mapcount(page);
373 if (mapcount >= 2) {
Mauricio Line070ad42005-09-03 15:55:10 -0700374 if (pte_dirty(ptent))
375 mss->shared_dirty += PAGE_SIZE;
376 else
377 mss->shared_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800378 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700379 } else {
380 if (pte_dirty(ptent))
381 mss->private_dirty += PAGE_SIZE;
382 else
383 mss->private_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800384 mss->pss += (PAGE_SIZE << PSS_SHIFT);
Mauricio Line070ad42005-09-03 15:55:10 -0700385 }
David Rientjes826fad12007-05-06 14:49:21 -0700386 }
Hugh Dickins705e87c2005-10-29 18:16:27 -0700387 pte_unmap_unlock(pte - 1, ptl);
388 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800389 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700390}
391
Mauricio Line070ad42005-09-03 15:55:10 -0700392static int show_smap(struct seq_file *m, void *v)
393{
Joe Korty7c88db02008-10-16 15:27:09 +0400394 struct proc_maps_private *priv = m->private;
395 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700396 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700397 struct mem_size_stats mss;
Dave Hansen21650092008-06-12 15:21:47 -0700398 struct mm_walk smaps_walk = {
399 .pmd_entry = smaps_pte_range,
400 .mm = vma->vm_mm,
401 .private = &mss,
402 };
Mauricio Line070ad42005-09-03 15:55:10 -0700403
404 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800405 mss.vma = vma;
Nick Piggin5ddfae12006-03-06 15:42:57 -0800406 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700407 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800408
Joe Korty7c88db02008-10-16 15:27:09 +0400409 show_map_vma(m, vma);
Matt Mackall4752c362008-02-04 22:29:02 -0800410
411 seq_printf(m,
412 "Size: %8lu kB\n"
413 "Rss: %8lu kB\n"
414 "Pss: %8lu kB\n"
415 "Shared_Clean: %8lu kB\n"
416 "Shared_Dirty: %8lu kB\n"
417 "Private_Clean: %8lu kB\n"
418 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700419 "Referenced: %8lu kB\n"
Mel Gorman08fba692009-01-06 14:38:53 -0800420 "Swap: %8lu kB\n"
Mel Gorman33402892009-01-06 14:38:54 -0800421 "KernelPageSize: %8lu kB\n"
422 "MMUPageSize: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800423 (vma->vm_end - vma->vm_start) >> 10,
424 mss.resident >> 10,
425 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
426 mss.shared_clean >> 10,
427 mss.shared_dirty >> 10,
428 mss.private_clean >> 10,
429 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700430 mss.referenced >> 10,
Mel Gorman08fba692009-01-06 14:38:53 -0800431 mss.swap >> 10,
Mel Gorman33402892009-01-06 14:38:54 -0800432 vma_kernel_pagesize(vma) >> 10,
433 vma_mmu_pagesize(vma) >> 10);
Matt Mackall4752c362008-02-04 22:29:02 -0800434
Joe Korty7c88db02008-10-16 15:27:09 +0400435 if (m->count < m->size) /* vma is copied successfully */
436 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
437 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700438}
439
Jan Engelhardt03a44822008-02-08 04:21:19 -0800440static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800441 .start = m_start,
442 .next = m_next,
443 .stop = m_stop,
444 .show = show_smap
445};
446
447static int smaps_open(struct inode *inode, struct file *file)
448{
449 return do_maps_open(inode, file, &proc_pid_smaps_op);
450}
451
452const struct file_operations proc_smaps_operations = {
453 .open = smaps_open,
454 .read = seq_read,
455 .llseek = seq_lseek,
456 .release = seq_release_private,
457};
458
459static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700460 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800461{
Dave Hansen21650092008-06-12 15:21:47 -0700462 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800463 pte_t *pte, ptent;
464 spinlock_t *ptl;
465 struct page *page;
466
467 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
468 for (; addr != end; pte++, addr += PAGE_SIZE) {
469 ptent = *pte;
470 if (!pte_present(ptent))
471 continue;
472
473 page = vm_normal_page(vma, addr, ptent);
474 if (!page)
475 continue;
476
477 /* Clear accessed and referenced bits. */
478 ptep_test_and_clear_young(vma, addr, pte);
479 ClearPageReferenced(page);
480 }
481 pte_unmap_unlock(pte - 1, ptl);
482 cond_resched();
483 return 0;
484}
485
Moussa A. Ba398499d2009-09-21 17:02:29 -0700486#define CLEAR_REFS_ALL 1
487#define CLEAR_REFS_ANON 2
488#define CLEAR_REFS_MAPPED 3
489
Matt Mackallf248dcb2008-02-04 22:29:03 -0800490static ssize_t clear_refs_write(struct file *file, const char __user *buf,
491 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700492{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800493 struct task_struct *task;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700494 char buffer[PROC_NUMBUF];
Matt Mackallf248dcb2008-02-04 22:29:03 -0800495 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700496 struct vm_area_struct *vma;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700497 long type;
David Rientjesb813e932007-05-06 14:49:24 -0700498
Matt Mackallf248dcb2008-02-04 22:29:03 -0800499 memset(buffer, 0, sizeof(buffer));
500 if (count > sizeof(buffer) - 1)
501 count = sizeof(buffer) - 1;
502 if (copy_from_user(buffer, buf, count))
503 return -EFAULT;
Vincent Lifb92a4b2009-09-22 16:45:36 -0700504 if (strict_strtol(strstrip(buffer), 10, &type))
505 return -EINVAL;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700506 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
Matt Mackallf248dcb2008-02-04 22:29:03 -0800507 return -EINVAL;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800508 task = get_proc_task(file->f_path.dentry->d_inode);
509 if (!task)
510 return -ESRCH;
511 mm = get_task_mm(task);
512 if (mm) {
Andrew Morton20cbc972008-07-05 12:29:05 -0700513 struct mm_walk clear_refs_walk = {
514 .pmd_entry = clear_refs_pte_range,
515 .mm = mm,
516 };
Matt Mackallf248dcb2008-02-04 22:29:03 -0800517 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700518 for (vma = mm->mmap; vma; vma = vma->vm_next) {
519 clear_refs_walk.private = vma;
Moussa A. Ba398499d2009-09-21 17:02:29 -0700520 if (is_vm_hugetlb_page(vma))
521 continue;
522 /*
523 * Writing 1 to /proc/pid/clear_refs affects all pages.
524 *
525 * Writing 2 to /proc/pid/clear_refs only affects
526 * Anonymous pages.
527 *
528 * Writing 3 to /proc/pid/clear_refs only affects file
529 * mapped pages.
530 */
531 if (type == CLEAR_REFS_ANON && vma->vm_file)
532 continue;
533 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
534 continue;
535 walk_page_range(vma->vm_start, vma->vm_end,
536 &clear_refs_walk);
Dave Hansen21650092008-06-12 15:21:47 -0700537 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800538 flush_tlb_mm(mm);
539 up_read(&mm->mmap_sem);
540 mmput(mm);
541 }
542 put_task_struct(task);
Vincent Lifb92a4b2009-09-22 16:45:36 -0700543
544 return count;
David Rientjesb813e932007-05-06 14:49:24 -0700545}
546
Matt Mackallf248dcb2008-02-04 22:29:03 -0800547const struct file_operations proc_clear_refs_operations = {
548 .write = clear_refs_write,
549};
550
Matt Mackall85863e42008-02-04 22:29:04 -0800551struct pagemapread {
Thomas Tuttleaae86792008-06-05 22:46:31 -0700552 u64 __user *out, *end;
Matt Mackall85863e42008-02-04 22:29:04 -0800553};
554
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500555#define PM_ENTRY_BYTES sizeof(u64)
556#define PM_STATUS_BITS 3
557#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
558#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
559#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
560#define PM_PSHIFT_BITS 6
561#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
562#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
563#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
564#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
565#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
566
567#define PM_PRESENT PM_STATUS(4LL)
568#define PM_SWAP PM_STATUS(2LL)
569#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800570#define PM_END_OF_BUFFER 1
571
572static int add_to_pagemap(unsigned long addr, u64 pfn,
573 struct pagemapread *pm)
574{
Matt Mackall85863e42008-02-04 22:29:04 -0800575 if (put_user(pfn, pm->out))
576 return -EFAULT;
Thomas Tuttleaae86792008-06-05 22:46:31 -0700577 pm->out++;
578 if (pm->out >= pm->end)
579 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800580 return 0;
581}
582
583static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700584 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800585{
Dave Hansen21650092008-06-12 15:21:47 -0700586 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800587 unsigned long addr;
588 int err = 0;
589 for (addr = start; addr < end; addr += PAGE_SIZE) {
590 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
591 if (err)
592 break;
593 }
594 return err;
595}
596
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700597static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800598{
599 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500600 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800601}
602
Matt Mackall49c50342008-12-09 13:14:21 -0800603static u64 pte_to_pagemap_entry(pte_t pte)
Dave Hansenbcf80392008-06-12 15:21:48 -0700604{
Matt Mackall49c50342008-12-09 13:14:21 -0800605 u64 pme = 0;
Dave Hansenbcf80392008-06-12 15:21:48 -0700606 if (is_swap_pte(pte))
607 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
608 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
609 else if (pte_present(pte))
610 pme = PM_PFRAME(pte_pfn(pte))
611 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
612 return pme;
613}
614
Matt Mackall85863e42008-02-04 22:29:04 -0800615static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700616 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800617{
Dave Hansenbcf80392008-06-12 15:21:48 -0700618 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700619 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800620 pte_t *pte;
621 int err = 0;
622
Dave Hansenbcf80392008-06-12 15:21:48 -0700623 /* find the first VMA at or above 'addr' */
624 vma = find_vma(walk->mm, addr);
Matt Mackall85863e42008-02-04 22:29:04 -0800625 for (; addr != end; addr += PAGE_SIZE) {
626 u64 pfn = PM_NOT_PRESENT;
Dave Hansenbcf80392008-06-12 15:21:48 -0700627
628 /* check to see if we've left 'vma' behind
629 * and need a new, higher one */
630 if (vma && (addr >= vma->vm_end))
631 vma = find_vma(walk->mm, addr);
632
633 /* check that 'vma' actually covers this address,
634 * and that it isn't a huge page vma */
635 if (vma && (vma->vm_start <= addr) &&
636 !is_vm_hugetlb_page(vma)) {
637 pte = pte_offset_map(pmd, addr);
638 pfn = pte_to_pagemap_entry(*pte);
639 /* unmap before userspace copy */
640 pte_unmap(pte);
641 }
Matt Mackall85863e42008-02-04 22:29:04 -0800642 err = add_to_pagemap(addr, pfn, pm);
643 if (err)
644 return err;
645 }
646
647 cond_resched();
648
649 return err;
650}
651
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
661static int pagemap_hugetlb_range(pte_t *pte, unsigned long addr,
662 unsigned long end, struct mm_walk *walk)
663{
664 struct vm_area_struct *vma;
665 struct pagemapread *pm = walk->private;
666 struct hstate *hs = NULL;
667 int err = 0;
668
669 vma = find_vma(walk->mm, addr);
670 if (vma)
671 hs = hstate_vma(vma);
672 for (; addr != end; addr += PAGE_SIZE) {
673 u64 pfn = PM_NOT_PRESENT;
674
675 if (vma && (addr >= vma->vm_end)) {
676 vma = find_vma(walk->mm, addr);
677 if (vma)
678 hs = hstate_vma(vma);
679 }
680
681 if (vma && (vma->vm_start <= addr) && is_vm_hugetlb_page(vma)) {
682 /* calculate pfn of the "raw" page in the hugepage. */
683 int offset = (addr & ~huge_page_mask(hs)) >> PAGE_SHIFT;
684 pfn = huge_pte_to_pagemap_entry(*pte, offset);
685 }
686 err = add_to_pagemap(addr, pfn, pm);
687 if (err)
688 return err;
689 }
690
691 cond_resched();
692
693 return err;
694}
695
Matt Mackall85863e42008-02-04 22:29:04 -0800696/*
697 * /proc/pid/pagemap - an array mapping virtual pages to pfns
698 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500699 * For each page in the address space, this file contains one 64-bit entry
700 * consisting of the following:
701 *
702 * Bits 0-55 page frame number (PFN) if present
703 * Bits 0-4 swap type if swapped
704 * Bits 5-55 swap offset if swapped
705 * Bits 55-60 page shift (page size = 1<<page shift)
706 * Bit 61 reserved for future use
707 * Bit 62 page swapped
708 * Bit 63 page present
709 *
710 * If the page is not present but in swap, then the PFN contains an
711 * encoding of the swap file number and the page's offset into the
712 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800713 * precisely which pages are mapped (or in swap) and comparing mapped
714 * pages between processes.
715 *
716 * Efficient users of this interface will use /proc/pid/maps to
717 * determine which areas of memory are actually mapped and llseek to
718 * skip over unmapped regions.
719 */
720static ssize_t pagemap_read(struct file *file, char __user *buf,
721 size_t count, loff_t *ppos)
722{
723 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
724 struct page **pages, *page;
725 unsigned long uaddr, uend;
726 struct mm_struct *mm;
727 struct pagemapread pm;
728 int pagecount;
729 int ret = -ESRCH;
Alexey Dobriyanee1e6ab2008-07-21 14:21:36 -0700730 struct mm_walk pagemap_walk = {};
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700731 unsigned long src;
732 unsigned long svpfn;
733 unsigned long start_vaddr;
734 unsigned long end_vaddr;
Matt Mackall85863e42008-02-04 22:29:04 -0800735
736 if (!task)
737 goto out;
738
739 ret = -EACCES;
Stephen Smalley006ebb42008-05-19 08:32:49 -0400740 if (!ptrace_may_access(task, PTRACE_MODE_READ))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700741 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800742
743 ret = -EINVAL;
744 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700745 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700746 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800747
748 ret = 0;
Vitaly Mayatskikh08161782009-04-30 15:08:18 -0700749
750 if (!count)
751 goto out_task;
752
Matt Mackall85863e42008-02-04 22:29:04 -0800753 mm = get_task_mm(task);
754 if (!mm)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700755 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800756
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700757
Matt Mackall85863e42008-02-04 22:29:04 -0800758 uaddr = (unsigned long)buf & PAGE_MASK;
759 uend = (unsigned long)(buf + count);
760 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700761 ret = 0;
762 if (pagecount == 0)
763 goto out_mm;
764 pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
765 ret = -ENOMEM;
Matt Mackall85863e42008-02-04 22:29:04 -0800766 if (!pages)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700767 goto out_mm;
Matt Mackall85863e42008-02-04 22:29:04 -0800768
769 down_read(&current->mm->mmap_sem);
770 ret = get_user_pages(current, current->mm, uaddr, pagecount,
771 1, 0, pages, NULL);
772 up_read(&current->mm->mmap_sem);
773
774 if (ret < 0)
775 goto out_free;
776
Marcelo Tosattifb393802008-03-13 12:32:35 -0700777 if (ret != pagecount) {
778 pagecount = ret;
779 ret = -EFAULT;
780 goto out_pages;
781 }
782
Milind Arun Choudhary09729a92009-02-20 16:56:45 +0300783 pm.out = (u64 __user *)buf;
784 pm.end = (u64 __user *)(buf + count);
Matt Mackall85863e42008-02-04 22:29:04 -0800785
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700786 pagemap_walk.pmd_entry = pagemap_pte_range;
787 pagemap_walk.pte_hole = pagemap_pte_hole;
Naoya Horiguchi5dc37642009-12-14 18:00:01 -0800788 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700789 pagemap_walk.mm = mm;
790 pagemap_walk.private = &pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800791
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700792 src = *ppos;
793 svpfn = src / PM_ENTRY_BYTES;
794 start_vaddr = svpfn << PAGE_SHIFT;
795 end_vaddr = TASK_SIZE_OF(task);
Matt Mackall85863e42008-02-04 22:29:04 -0800796
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700797 /* watch out for wraparound */
798 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
799 start_vaddr = end_vaddr;
800
801 /*
802 * The odds are that this will stop walking way
803 * before end_vaddr, because the length of the
804 * user buffer is tracked in "pm", and the walk
805 * will stop when we hit the end of the buffer.
806 */
807 ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
808 if (ret == PM_END_OF_BUFFER)
809 ret = 0;
810 /* don't need mmap_sem for these, but this looks cleaner */
Milind Arun Choudhary09729a92009-02-20 16:56:45 +0300811 *ppos += (char __user *)pm.out - buf;
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700812 if (!ret)
Milind Arun Choudhary09729a92009-02-20 16:56:45 +0300813 ret = (char __user *)pm.out - buf;
Matt Mackall85863e42008-02-04 22:29:04 -0800814
Marcelo Tosattifb393802008-03-13 12:32:35 -0700815out_pages:
Matt Mackall85863e42008-02-04 22:29:04 -0800816 for (; pagecount; pagecount--) {
817 page = pages[pagecount-1];
818 if (!PageReserved(page))
819 SetPageDirty(page);
820 page_cache_release(page);
821 }
Matt Mackall85863e42008-02-04 22:29:04 -0800822out_free:
823 kfree(pages);
Marcelo Tosattifb393802008-03-13 12:32:35 -0700824out_mm:
825 mmput(mm);
Matt Mackall85863e42008-02-04 22:29:04 -0800826out_task:
827 put_task_struct(task);
828out:
829 return ret;
830}
831
832const struct file_operations proc_pagemap_operations = {
833 .llseek = mem_lseek, /* borrow this */
834 .read = pagemap_read,
835};
Matt Mackall1e883282008-02-04 22:29:07 -0800836#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800837
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700838#ifdef CONFIG_NUMA
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800839extern int show_numa_map(struct seq_file *m, void *v);
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700840
Jan Engelhardt03a44822008-02-08 04:21:19 -0800841static const struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800842 .start = m_start,
843 .next = m_next,
844 .stop = m_stop,
Alexey Dobriyan3bbfe052008-10-10 03:27:16 +0400845 .show = show_numa_map,
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700846};
Eric W. Biederman662795d2006-06-26 00:25:48 -0700847
848static int numa_maps_open(struct inode *inode, struct file *file)
849{
850 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
851}
852
Arjan van de Ven00977a52007-02-12 00:55:34 -0800853const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -0700854 .open = numa_maps_open,
855 .read = seq_read,
856 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -0700857 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -0700858};
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700859#endif