blob: 9bd8be1d235c2b420e348d25e801bdce88efd977 [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{
Hugh Dickins42946212005-10-29 18:16:05 -070068 *shared = get_mm_counter(mm, file_rss);
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;
Hugh Dickins42946212005-10-29 18:16:05 -070072 *resident = *shared + get_mm_counter(mm, anon_rss);
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]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700247 } else {
248 name = "[vdso]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700250 }
251 if (name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 pad_len_spaces(m, len);
Ingo Molnare6e54942006-06-27 02:53:50 -0700253 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 }
255 }
256 seq_putc(m, '\n');
Joe Korty7c88db02008-10-16 15:27:09 +0400257}
258
259static int show_map(struct seq_file *m, void *v)
260{
261 struct vm_area_struct *vma = v;
262 struct proc_maps_private *priv = m->private;
263 struct task_struct *task = priv->task;
264
265 show_map_vma(m, vma);
Mauricio Line070ad42005-09-03 15:55:10 -0700266
Mauricio Line070ad42005-09-03 15:55:10 -0700267 if (m->count < m->size) /* vma is copied successfully */
268 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return 0;
270}
271
Jan Engelhardt03a44822008-02-08 04:21:19 -0800272static const struct seq_operations proc_pid_maps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800273 .start = m_start,
274 .next = m_next,
275 .stop = m_stop,
276 .show = show_map
277};
278
279static int maps_open(struct inode *inode, struct file *file)
280{
281 return do_maps_open(inode, file, &proc_pid_maps_op);
282}
283
284const struct file_operations proc_maps_operations = {
285 .open = maps_open,
286 .read = seq_read,
287 .llseek = seq_lseek,
288 .release = seq_release_private,
289};
290
291/*
292 * Proportional Set Size(PSS): my share of RSS.
293 *
294 * PSS of a process is the count of pages it has in memory, where each
295 * page is divided by the number of processes sharing it. So if a
296 * process has 1000 pages all to itself, and 1000 shared with one other
297 * process, its PSS will be 1500.
298 *
299 * To keep (accumulated) division errors low, we adopt a 64bit
300 * fixed-point pss counter to minimize division errors. So (pss >>
301 * PSS_SHIFT) would be the real byte count.
302 *
303 * A shift of 12 before division means (assuming 4K page size):
304 * - 1M 3-user-pages add up to 8KB errors;
305 * - supports mapcount up to 2^24, or 16M;
306 * - supports PSS up to 2^52 bytes, or 4PB.
307 */
308#define PSS_SHIFT 12
309
Matt Mackall1e883282008-02-04 22:29:07 -0800310#ifdef CONFIG_PROC_PAGE_MONITOR
Peter Zijlstra214e4712008-04-28 02:12:55 -0700311struct mem_size_stats {
Matt Mackalla6198792008-02-04 22:29:03 -0800312 struct vm_area_struct *vma;
313 unsigned long resident;
314 unsigned long shared_clean;
315 unsigned long shared_dirty;
316 unsigned long private_clean;
317 unsigned long private_dirty;
318 unsigned long referenced;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700319 unsigned long swap;
Matt Mackalla6198792008-02-04 22:29:03 -0800320 u64 pss;
321};
322
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800323static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700324 struct mm_walk *walk)
Mauricio Line070ad42005-09-03 15:55:10 -0700325{
Dave Hansen21650092008-06-12 15:21:47 -0700326 struct mem_size_stats *mss = walk->private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800327 struct vm_area_struct *vma = mss->vma;
Mauricio Line070ad42005-09-03 15:55:10 -0700328 pte_t *pte, ptent;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700329 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700330 struct page *page;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800331 int mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700332
Hugh Dickins705e87c2005-10-29 18:16:27 -0700333 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
David Rientjes826fad12007-05-06 14:49:21 -0700334 for (; addr != end; pte++, addr += PAGE_SIZE) {
Mauricio Line070ad42005-09-03 15:55:10 -0700335 ptent = *pte;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700336
337 if (is_swap_pte(ptent)) {
338 mss->swap += PAGE_SIZE;
339 continue;
340 }
341
Hugh Dickins705e87c2005-10-29 18:16:27 -0700342 if (!pte_present(ptent))
Mauricio Line070ad42005-09-03 15:55:10 -0700343 continue;
344
345 mss->resident += PAGE_SIZE;
Nick Pigginad820c5d2006-03-06 15:42:58 -0800346
347 page = vm_normal_page(vma, addr, ptent);
348 if (!page)
Mauricio Line070ad42005-09-03 15:55:10 -0700349 continue;
350
David Rientjesf79f1772007-05-06 14:49:22 -0700351 /* Accumulate the size in pages that have been accessed. */
352 if (pte_young(ptent) || PageReferenced(page))
353 mss->referenced += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800354 mapcount = page_mapcount(page);
355 if (mapcount >= 2) {
Mauricio Line070ad42005-09-03 15:55:10 -0700356 if (pte_dirty(ptent))
357 mss->shared_dirty += PAGE_SIZE;
358 else
359 mss->shared_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800360 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700361 } else {
362 if (pte_dirty(ptent))
363 mss->private_dirty += PAGE_SIZE;
364 else
365 mss->private_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800366 mss->pss += (PAGE_SIZE << PSS_SHIFT);
Mauricio Line070ad42005-09-03 15:55:10 -0700367 }
David Rientjes826fad12007-05-06 14:49:21 -0700368 }
Hugh Dickins705e87c2005-10-29 18:16:27 -0700369 pte_unmap_unlock(pte - 1, ptl);
370 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800371 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700372}
373
Mauricio Line070ad42005-09-03 15:55:10 -0700374static int show_smap(struct seq_file *m, void *v)
375{
Joe Korty7c88db02008-10-16 15:27:09 +0400376 struct proc_maps_private *priv = m->private;
377 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700378 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700379 struct mem_size_stats mss;
Dave Hansen21650092008-06-12 15:21:47 -0700380 struct mm_walk smaps_walk = {
381 .pmd_entry = smaps_pte_range,
382 .mm = vma->vm_mm,
383 .private = &mss,
384 };
Mauricio Line070ad42005-09-03 15:55:10 -0700385
386 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800387 mss.vma = vma;
Nick Piggin5ddfae12006-03-06 15:42:57 -0800388 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700389 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800390
Joe Korty7c88db02008-10-16 15:27:09 +0400391 show_map_vma(m, vma);
Matt Mackall4752c362008-02-04 22:29:02 -0800392
393 seq_printf(m,
394 "Size: %8lu kB\n"
395 "Rss: %8lu kB\n"
396 "Pss: %8lu kB\n"
397 "Shared_Clean: %8lu kB\n"
398 "Shared_Dirty: %8lu kB\n"
399 "Private_Clean: %8lu kB\n"
400 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700401 "Referenced: %8lu kB\n"
Mel Gorman08fba692009-01-06 14:38:53 -0800402 "Swap: %8lu kB\n"
Mel Gorman33402892009-01-06 14:38:54 -0800403 "KernelPageSize: %8lu kB\n"
404 "MMUPageSize: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800405 (vma->vm_end - vma->vm_start) >> 10,
406 mss.resident >> 10,
407 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
408 mss.shared_clean >> 10,
409 mss.shared_dirty >> 10,
410 mss.private_clean >> 10,
411 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700412 mss.referenced >> 10,
Mel Gorman08fba692009-01-06 14:38:53 -0800413 mss.swap >> 10,
Mel Gorman33402892009-01-06 14:38:54 -0800414 vma_kernel_pagesize(vma) >> 10,
415 vma_mmu_pagesize(vma) >> 10);
Matt Mackall4752c362008-02-04 22:29:02 -0800416
Joe Korty7c88db02008-10-16 15:27:09 +0400417 if (m->count < m->size) /* vma is copied successfully */
418 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
419 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700420}
421
Jan Engelhardt03a44822008-02-08 04:21:19 -0800422static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800423 .start = m_start,
424 .next = m_next,
425 .stop = m_stop,
426 .show = show_smap
427};
428
429static int smaps_open(struct inode *inode, struct file *file)
430{
431 return do_maps_open(inode, file, &proc_pid_smaps_op);
432}
433
434const struct file_operations proc_smaps_operations = {
435 .open = smaps_open,
436 .read = seq_read,
437 .llseek = seq_lseek,
438 .release = seq_release_private,
439};
440
441static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700442 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800443{
Dave Hansen21650092008-06-12 15:21:47 -0700444 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800445 pte_t *pte, ptent;
446 spinlock_t *ptl;
447 struct page *page;
448
449 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
450 for (; addr != end; pte++, addr += PAGE_SIZE) {
451 ptent = *pte;
452 if (!pte_present(ptent))
453 continue;
454
455 page = vm_normal_page(vma, addr, ptent);
456 if (!page)
457 continue;
458
459 /* Clear accessed and referenced bits. */
460 ptep_test_and_clear_young(vma, addr, pte);
461 ClearPageReferenced(page);
462 }
463 pte_unmap_unlock(pte - 1, ptl);
464 cond_resched();
465 return 0;
466}
467
Matt Mackallf248dcb2008-02-04 22:29:03 -0800468static ssize_t clear_refs_write(struct file *file, const char __user *buf,
469 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700470{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800471 struct task_struct *task;
472 char buffer[PROC_NUMBUF], *end;
473 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700474 struct vm_area_struct *vma;
475
Matt Mackallf248dcb2008-02-04 22:29:03 -0800476 memset(buffer, 0, sizeof(buffer));
477 if (count > sizeof(buffer) - 1)
478 count = sizeof(buffer) - 1;
479 if (copy_from_user(buffer, buf, count))
480 return -EFAULT;
481 if (!simple_strtol(buffer, &end, 0))
482 return -EINVAL;
483 if (*end == '\n')
484 end++;
485 task = get_proc_task(file->f_path.dentry->d_inode);
486 if (!task)
487 return -ESRCH;
488 mm = get_task_mm(task);
489 if (mm) {
Andrew Morton20cbc972008-07-05 12:29:05 -0700490 struct mm_walk clear_refs_walk = {
491 .pmd_entry = clear_refs_pte_range,
492 .mm = mm,
493 };
Matt Mackallf248dcb2008-02-04 22:29:03 -0800494 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700495 for (vma = mm->mmap; vma; vma = vma->vm_next) {
496 clear_refs_walk.private = vma;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800497 if (!is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700498 walk_page_range(vma->vm_start, vma->vm_end,
499 &clear_refs_walk);
500 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800501 flush_tlb_mm(mm);
502 up_read(&mm->mmap_sem);
503 mmput(mm);
504 }
505 put_task_struct(task);
506 if (end - buffer == 0)
507 return -EIO;
508 return end - buffer;
David Rientjesb813e932007-05-06 14:49:24 -0700509}
510
Matt Mackallf248dcb2008-02-04 22:29:03 -0800511const struct file_operations proc_clear_refs_operations = {
512 .write = clear_refs_write,
513};
514
Matt Mackall85863e42008-02-04 22:29:04 -0800515struct pagemapread {
Thomas Tuttleaae86792008-06-05 22:46:31 -0700516 u64 __user *out, *end;
Matt Mackall85863e42008-02-04 22:29:04 -0800517};
518
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500519#define PM_ENTRY_BYTES sizeof(u64)
520#define PM_STATUS_BITS 3
521#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
522#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
523#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
524#define PM_PSHIFT_BITS 6
525#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
526#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
527#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
528#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
529#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
530
531#define PM_PRESENT PM_STATUS(4LL)
532#define PM_SWAP PM_STATUS(2LL)
533#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800534#define PM_END_OF_BUFFER 1
535
536static int add_to_pagemap(unsigned long addr, u64 pfn,
537 struct pagemapread *pm)
538{
Matt Mackall85863e42008-02-04 22:29:04 -0800539 if (put_user(pfn, pm->out))
540 return -EFAULT;
Thomas Tuttleaae86792008-06-05 22:46:31 -0700541 pm->out++;
542 if (pm->out >= pm->end)
543 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800544 return 0;
545}
546
547static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700548 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800549{
Dave Hansen21650092008-06-12 15:21:47 -0700550 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800551 unsigned long addr;
552 int err = 0;
553 for (addr = start; addr < end; addr += PAGE_SIZE) {
554 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
555 if (err)
556 break;
557 }
558 return err;
559}
560
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700561static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800562{
563 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500564 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800565}
566
Matt Mackall49c50342008-12-09 13:14:21 -0800567static u64 pte_to_pagemap_entry(pte_t pte)
Dave Hansenbcf80392008-06-12 15:21:48 -0700568{
Matt Mackall49c50342008-12-09 13:14:21 -0800569 u64 pme = 0;
Dave Hansenbcf80392008-06-12 15:21:48 -0700570 if (is_swap_pte(pte))
571 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
572 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
573 else if (pte_present(pte))
574 pme = PM_PFRAME(pte_pfn(pte))
575 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
576 return pme;
577}
578
Matt Mackall85863e42008-02-04 22:29:04 -0800579static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700580 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800581{
Dave Hansenbcf80392008-06-12 15:21:48 -0700582 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700583 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800584 pte_t *pte;
585 int err = 0;
586
Dave Hansenbcf80392008-06-12 15:21:48 -0700587 /* find the first VMA at or above 'addr' */
588 vma = find_vma(walk->mm, addr);
Matt Mackall85863e42008-02-04 22:29:04 -0800589 for (; addr != end; addr += PAGE_SIZE) {
590 u64 pfn = PM_NOT_PRESENT;
Dave Hansenbcf80392008-06-12 15:21:48 -0700591
592 /* check to see if we've left 'vma' behind
593 * and need a new, higher one */
594 if (vma && (addr >= vma->vm_end))
595 vma = find_vma(walk->mm, addr);
596
597 /* check that 'vma' actually covers this address,
598 * and that it isn't a huge page vma */
599 if (vma && (vma->vm_start <= addr) &&
600 !is_vm_hugetlb_page(vma)) {
601 pte = pte_offset_map(pmd, addr);
602 pfn = pte_to_pagemap_entry(*pte);
603 /* unmap before userspace copy */
604 pte_unmap(pte);
605 }
Matt Mackall85863e42008-02-04 22:29:04 -0800606 err = add_to_pagemap(addr, pfn, pm);
607 if (err)
608 return err;
609 }
610
611 cond_resched();
612
613 return err;
614}
615
Matt Mackall85863e42008-02-04 22:29:04 -0800616/*
617 * /proc/pid/pagemap - an array mapping virtual pages to pfns
618 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500619 * For each page in the address space, this file contains one 64-bit entry
620 * consisting of the following:
621 *
622 * Bits 0-55 page frame number (PFN) if present
623 * Bits 0-4 swap type if swapped
624 * Bits 5-55 swap offset if swapped
625 * Bits 55-60 page shift (page size = 1<<page shift)
626 * Bit 61 reserved for future use
627 * Bit 62 page swapped
628 * Bit 63 page present
629 *
630 * If the page is not present but in swap, then the PFN contains an
631 * encoding of the swap file number and the page's offset into the
632 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800633 * precisely which pages are mapped (or in swap) and comparing mapped
634 * pages between processes.
635 *
636 * Efficient users of this interface will use /proc/pid/maps to
637 * determine which areas of memory are actually mapped and llseek to
638 * skip over unmapped regions.
639 */
640static ssize_t pagemap_read(struct file *file, char __user *buf,
641 size_t count, loff_t *ppos)
642{
643 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
644 struct page **pages, *page;
645 unsigned long uaddr, uend;
646 struct mm_struct *mm;
647 struct pagemapread pm;
648 int pagecount;
649 int ret = -ESRCH;
Alexey Dobriyanee1e6ab2008-07-21 14:21:36 -0700650 struct mm_walk pagemap_walk = {};
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700651 unsigned long src;
652 unsigned long svpfn;
653 unsigned long start_vaddr;
654 unsigned long end_vaddr;
Matt Mackall85863e42008-02-04 22:29:04 -0800655
656 if (!task)
657 goto out;
658
659 ret = -EACCES;
Stephen Smalley006ebb42008-05-19 08:32:49 -0400660 if (!ptrace_may_access(task, PTRACE_MODE_READ))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700661 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800662
663 ret = -EINVAL;
664 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700665 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700666 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800667
668 ret = 0;
Vitaly Mayatskikh08161782009-04-30 15:08:18 -0700669
670 if (!count)
671 goto out_task;
672
Matt Mackall85863e42008-02-04 22:29:04 -0800673 mm = get_task_mm(task);
674 if (!mm)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700675 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800676
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700677
Matt Mackall85863e42008-02-04 22:29:04 -0800678 uaddr = (unsigned long)buf & PAGE_MASK;
679 uend = (unsigned long)(buf + count);
680 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700681 ret = 0;
682 if (pagecount == 0)
683 goto out_mm;
684 pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
685 ret = -ENOMEM;
Matt Mackall85863e42008-02-04 22:29:04 -0800686 if (!pages)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700687 goto out_mm;
Matt Mackall85863e42008-02-04 22:29:04 -0800688
689 down_read(&current->mm->mmap_sem);
690 ret = get_user_pages(current, current->mm, uaddr, pagecount,
691 1, 0, pages, NULL);
692 up_read(&current->mm->mmap_sem);
693
694 if (ret < 0)
695 goto out_free;
696
Marcelo Tosattifb393802008-03-13 12:32:35 -0700697 if (ret != pagecount) {
698 pagecount = ret;
699 ret = -EFAULT;
700 goto out_pages;
701 }
702
Milind Arun Choudhary09729a92009-02-20 16:56:45 +0300703 pm.out = (u64 __user *)buf;
704 pm.end = (u64 __user *)(buf + count);
Matt Mackall85863e42008-02-04 22:29:04 -0800705
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700706 pagemap_walk.pmd_entry = pagemap_pte_range;
707 pagemap_walk.pte_hole = pagemap_pte_hole;
708 pagemap_walk.mm = mm;
709 pagemap_walk.private = &pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800710
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700711 src = *ppos;
712 svpfn = src / PM_ENTRY_BYTES;
713 start_vaddr = svpfn << PAGE_SHIFT;
714 end_vaddr = TASK_SIZE_OF(task);
Matt Mackall85863e42008-02-04 22:29:04 -0800715
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700716 /* watch out for wraparound */
717 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
718 start_vaddr = end_vaddr;
719
720 /*
721 * The odds are that this will stop walking way
722 * before end_vaddr, because the length of the
723 * user buffer is tracked in "pm", and the walk
724 * will stop when we hit the end of the buffer.
725 */
726 ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
727 if (ret == PM_END_OF_BUFFER)
728 ret = 0;
729 /* don't need mmap_sem for these, but this looks cleaner */
Milind Arun Choudhary09729a92009-02-20 16:56:45 +0300730 *ppos += (char __user *)pm.out - buf;
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700731 if (!ret)
Milind Arun Choudhary09729a92009-02-20 16:56:45 +0300732 ret = (char __user *)pm.out - buf;
Matt Mackall85863e42008-02-04 22:29:04 -0800733
Marcelo Tosattifb393802008-03-13 12:32:35 -0700734out_pages:
Matt Mackall85863e42008-02-04 22:29:04 -0800735 for (; pagecount; pagecount--) {
736 page = pages[pagecount-1];
737 if (!PageReserved(page))
738 SetPageDirty(page);
739 page_cache_release(page);
740 }
Matt Mackall85863e42008-02-04 22:29:04 -0800741out_free:
742 kfree(pages);
Marcelo Tosattifb393802008-03-13 12:32:35 -0700743out_mm:
744 mmput(mm);
Matt Mackall85863e42008-02-04 22:29:04 -0800745out_task:
746 put_task_struct(task);
747out:
748 return ret;
749}
750
751const struct file_operations proc_pagemap_operations = {
752 .llseek = mem_lseek, /* borrow this */
753 .read = pagemap_read,
754};
Matt Mackall1e883282008-02-04 22:29:07 -0800755#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800756
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700757#ifdef CONFIG_NUMA
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800758extern int show_numa_map(struct seq_file *m, void *v);
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700759
Jan Engelhardt03a44822008-02-08 04:21:19 -0800760static const struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800761 .start = m_start,
762 .next = m_next,
763 .stop = m_stop,
Alexey Dobriyan3bbfe052008-10-10 03:27:16 +0400764 .show = show_numa_map,
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700765};
Eric W. Biederman662795d2006-06-26 00:25:48 -0700766
767static int numa_maps_open(struct inode *inode, struct file *file)
768{
769 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
770}
771
Arjan van de Ven00977a52007-02-12 00:55:34 -0800772const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -0700773 .open = numa_maps_open,
774 .read = seq_read,
775 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -0700776 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -0700777};
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700778#endif