blob: 94063840832ad99709d4ca878d9aff8704ac6424 [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;
122
123 tail_vma = get_gate_vma(priv->task);
124 priv->tail_vma = tail_vma;
125
126 /* Start with last addr hint */
127 vma = find_vma(mm, last_addr);
128 if (last_addr && vma) {
129 vma = vma->vm_next;
130 goto out;
131 }
132
133 /*
134 * Check the vma index is within the range and do
135 * sequential scan until m_index.
136 */
137 vma = NULL;
138 if ((unsigned long)l < mm->map_count) {
139 vma = mm->mmap;
140 while (l-- && vma)
141 vma = vma->vm_next;
142 goto out;
143 }
144
145 if (l != mm->map_count)
146 tail_vma = NULL; /* After gate vma */
147
148out:
149 if (vma)
150 return vma;
151
152 /* End of vmas has been reached */
153 m->version = (tail_vma != NULL)? 0: -1UL;
154 up_read(&mm->mmap_sem);
155 mmput(mm);
156 return tail_vma;
157}
158
159static void *m_next(struct seq_file *m, void *v, loff_t *pos)
160{
161 struct proc_maps_private *priv = m->private;
162 struct vm_area_struct *vma = v;
163 struct vm_area_struct *tail_vma = priv->tail_vma;
164
165 (*pos)++;
166 if (vma && (vma != tail_vma) && vma->vm_next)
167 return vma->vm_next;
168 vma_stop(priv, vma);
169 return (vma != tail_vma)? tail_vma: NULL;
170}
171
172static void m_stop(struct seq_file *m, void *v)
173{
174 struct proc_maps_private *priv = m->private;
175 struct vm_area_struct *vma = v;
176
177 vma_stop(priv, vma);
178 if (priv->task)
179 put_task_struct(priv->task);
180}
181
182static int do_maps_open(struct inode *inode, struct file *file,
Jan Engelhardt03a44822008-02-08 04:21:19 -0800183 const struct seq_operations *ops)
Matt Mackalla6198792008-02-04 22:29:03 -0800184{
185 struct proc_maps_private *priv;
186 int ret = -ENOMEM;
187 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
188 if (priv) {
189 priv->pid = proc_pid(inode);
190 ret = seq_open(file, ops);
191 if (!ret) {
192 struct seq_file *m = file->private_data;
193 m->private = priv;
194 } else {
195 kfree(priv);
196 }
197 }
198 return ret;
199}
Mauricio Line070ad42005-09-03 15:55:10 -0700200
Joe Korty7c88db02008-10-16 15:27:09 +0400201static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Mauricio Line070ad42005-09-03 15:55:10 -0700203 struct mm_struct *mm = vma->vm_mm;
204 struct file *file = vma->vm_file;
205 int flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 unsigned long ino = 0;
207 dev_t dev = 0;
208 int len;
209
210 if (file) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800211 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 dev = inode->i_sb->s_dev;
213 ino = inode->i_ino;
214 }
215
Clement Calmels1804dc62008-08-20 14:09:00 -0700216 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
Mauricio Line070ad42005-09-03 15:55:10 -0700217 vma->vm_start,
218 vma->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 flags & VM_READ ? 'r' : '-',
220 flags & VM_WRITE ? 'w' : '-',
221 flags & VM_EXEC ? 'x' : '-',
222 flags & VM_MAYSHARE ? 's' : 'p',
Clement Calmels1804dc62008-08-20 14:09:00 -0700223 ((loff_t)vma->vm_pgoff) << PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 MAJOR(dev), MINOR(dev), ino, &len);
225
226 /*
227 * Print the dentry name for named mappings, and a
228 * special [heap] marker for the heap:
229 */
Mauricio Line070ad42005-09-03 15:55:10 -0700230 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 pad_len_spaces(m, len);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800232 seq_path(m, &file->f_path, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 } else {
Ingo Molnare6e54942006-06-27 02:53:50 -0700234 const char *name = arch_vma_name(vma);
235 if (!name) {
236 if (mm) {
237 if (vma->vm_start <= mm->start_brk &&
Mauricio Line070ad42005-09-03 15:55:10 -0700238 vma->vm_end >= mm->brk) {
Ingo Molnare6e54942006-06-27 02:53:50 -0700239 name = "[heap]";
240 } else if (vma->vm_start <= mm->start_stack &&
241 vma->vm_end >= mm->start_stack) {
242 name = "[stack]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700244 } else {
245 name = "[vdso]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700247 }
248 if (name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 pad_len_spaces(m, len);
Ingo Molnare6e54942006-06-27 02:53:50 -0700250 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 }
252 }
253 seq_putc(m, '\n');
Joe Korty7c88db02008-10-16 15:27:09 +0400254}
255
256static int show_map(struct seq_file *m, void *v)
257{
258 struct vm_area_struct *vma = v;
259 struct proc_maps_private *priv = m->private;
260 struct task_struct *task = priv->task;
261
262 show_map_vma(m, vma);
Mauricio Line070ad42005-09-03 15:55:10 -0700263
Mauricio Line070ad42005-09-03 15:55:10 -0700264 if (m->count < m->size) /* vma is copied successfully */
265 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 return 0;
267}
268
Jan Engelhardt03a44822008-02-08 04:21:19 -0800269static const struct seq_operations proc_pid_maps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800270 .start = m_start,
271 .next = m_next,
272 .stop = m_stop,
273 .show = show_map
274};
275
276static int maps_open(struct inode *inode, struct file *file)
277{
278 return do_maps_open(inode, file, &proc_pid_maps_op);
279}
280
281const struct file_operations proc_maps_operations = {
282 .open = maps_open,
283 .read = seq_read,
284 .llseek = seq_lseek,
285 .release = seq_release_private,
286};
287
288/*
289 * Proportional Set Size(PSS): my share of RSS.
290 *
291 * PSS of a process is the count of pages it has in memory, where each
292 * page is divided by the number of processes sharing it. So if a
293 * process has 1000 pages all to itself, and 1000 shared with one other
294 * process, its PSS will be 1500.
295 *
296 * To keep (accumulated) division errors low, we adopt a 64bit
297 * fixed-point pss counter to minimize division errors. So (pss >>
298 * PSS_SHIFT) would be the real byte count.
299 *
300 * A shift of 12 before division means (assuming 4K page size):
301 * - 1M 3-user-pages add up to 8KB errors;
302 * - supports mapcount up to 2^24, or 16M;
303 * - supports PSS up to 2^52 bytes, or 4PB.
304 */
305#define PSS_SHIFT 12
306
Matt Mackall1e883282008-02-04 22:29:07 -0800307#ifdef CONFIG_PROC_PAGE_MONITOR
Peter Zijlstra214e4712008-04-28 02:12:55 -0700308struct mem_size_stats {
Matt Mackalla6198792008-02-04 22:29:03 -0800309 struct vm_area_struct *vma;
310 unsigned long resident;
311 unsigned long shared_clean;
312 unsigned long shared_dirty;
313 unsigned long private_clean;
314 unsigned long private_dirty;
315 unsigned long referenced;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700316 unsigned long swap;
Matt Mackalla6198792008-02-04 22:29:03 -0800317 u64 pss;
318};
319
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800320static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700321 struct mm_walk *walk)
Mauricio Line070ad42005-09-03 15:55:10 -0700322{
Dave Hansen21650092008-06-12 15:21:47 -0700323 struct mem_size_stats *mss = walk->private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800324 struct vm_area_struct *vma = mss->vma;
Mauricio Line070ad42005-09-03 15:55:10 -0700325 pte_t *pte, ptent;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700326 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700327 struct page *page;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800328 int mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700329
Hugh Dickins705e87c2005-10-29 18:16:27 -0700330 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
David Rientjes826fad12007-05-06 14:49:21 -0700331 for (; addr != end; pte++, addr += PAGE_SIZE) {
Mauricio Line070ad42005-09-03 15:55:10 -0700332 ptent = *pte;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700333
334 if (is_swap_pte(ptent)) {
335 mss->swap += PAGE_SIZE;
336 continue;
337 }
338
Hugh Dickins705e87c2005-10-29 18:16:27 -0700339 if (!pte_present(ptent))
Mauricio Line070ad42005-09-03 15:55:10 -0700340 continue;
341
342 mss->resident += PAGE_SIZE;
Nick Pigginad820c5d2006-03-06 15:42:58 -0800343
344 page = vm_normal_page(vma, addr, ptent);
345 if (!page)
Mauricio Line070ad42005-09-03 15:55:10 -0700346 continue;
347
David Rientjesf79f1772007-05-06 14:49:22 -0700348 /* Accumulate the size in pages that have been accessed. */
349 if (pte_young(ptent) || PageReferenced(page))
350 mss->referenced += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800351 mapcount = page_mapcount(page);
352 if (mapcount >= 2) {
Mauricio Line070ad42005-09-03 15:55:10 -0700353 if (pte_dirty(ptent))
354 mss->shared_dirty += PAGE_SIZE;
355 else
356 mss->shared_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800357 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700358 } else {
359 if (pte_dirty(ptent))
360 mss->private_dirty += PAGE_SIZE;
361 else
362 mss->private_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800363 mss->pss += (PAGE_SIZE << PSS_SHIFT);
Mauricio Line070ad42005-09-03 15:55:10 -0700364 }
David Rientjes826fad12007-05-06 14:49:21 -0700365 }
Hugh Dickins705e87c2005-10-29 18:16:27 -0700366 pte_unmap_unlock(pte - 1, ptl);
367 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800368 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700369}
370
Mauricio Line070ad42005-09-03 15:55:10 -0700371static int show_smap(struct seq_file *m, void *v)
372{
Joe Korty7c88db02008-10-16 15:27:09 +0400373 struct proc_maps_private *priv = m->private;
374 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700375 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700376 struct mem_size_stats mss;
Dave Hansen21650092008-06-12 15:21:47 -0700377 struct mm_walk smaps_walk = {
378 .pmd_entry = smaps_pte_range,
379 .mm = vma->vm_mm,
380 .private = &mss,
381 };
Mauricio Line070ad42005-09-03 15:55:10 -0700382
383 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800384 mss.vma = vma;
Nick Piggin5ddfae12006-03-06 15:42:57 -0800385 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700386 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800387
Joe Korty7c88db02008-10-16 15:27:09 +0400388 show_map_vma(m, vma);
Matt Mackall4752c362008-02-04 22:29:02 -0800389
390 seq_printf(m,
391 "Size: %8lu kB\n"
392 "Rss: %8lu kB\n"
393 "Pss: %8lu kB\n"
394 "Shared_Clean: %8lu kB\n"
395 "Shared_Dirty: %8lu kB\n"
396 "Private_Clean: %8lu kB\n"
397 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700398 "Referenced: %8lu kB\n"
Mel Gorman08fba692009-01-06 14:38:53 -0800399 "Swap: %8lu kB\n"
Mel Gorman33402892009-01-06 14:38:54 -0800400 "KernelPageSize: %8lu kB\n"
401 "MMUPageSize: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800402 (vma->vm_end - vma->vm_start) >> 10,
403 mss.resident >> 10,
404 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
405 mss.shared_clean >> 10,
406 mss.shared_dirty >> 10,
407 mss.private_clean >> 10,
408 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700409 mss.referenced >> 10,
Mel Gorman08fba692009-01-06 14:38:53 -0800410 mss.swap >> 10,
Mel Gorman33402892009-01-06 14:38:54 -0800411 vma_kernel_pagesize(vma) >> 10,
412 vma_mmu_pagesize(vma) >> 10);
Matt Mackall4752c362008-02-04 22:29:02 -0800413
Joe Korty7c88db02008-10-16 15:27:09 +0400414 if (m->count < m->size) /* vma is copied successfully */
415 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
416 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700417}
418
Jan Engelhardt03a44822008-02-08 04:21:19 -0800419static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800420 .start = m_start,
421 .next = m_next,
422 .stop = m_stop,
423 .show = show_smap
424};
425
426static int smaps_open(struct inode *inode, struct file *file)
427{
428 return do_maps_open(inode, file, &proc_pid_smaps_op);
429}
430
431const struct file_operations proc_smaps_operations = {
432 .open = smaps_open,
433 .read = seq_read,
434 .llseek = seq_lseek,
435 .release = seq_release_private,
436};
437
438static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700439 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800440{
Dave Hansen21650092008-06-12 15:21:47 -0700441 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800442 pte_t *pte, ptent;
443 spinlock_t *ptl;
444 struct page *page;
445
446 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
447 for (; addr != end; pte++, addr += PAGE_SIZE) {
448 ptent = *pte;
449 if (!pte_present(ptent))
450 continue;
451
452 page = vm_normal_page(vma, addr, ptent);
453 if (!page)
454 continue;
455
456 /* Clear accessed and referenced bits. */
457 ptep_test_and_clear_young(vma, addr, pte);
458 ClearPageReferenced(page);
459 }
460 pte_unmap_unlock(pte - 1, ptl);
461 cond_resched();
462 return 0;
463}
464
Matt Mackallf248dcb2008-02-04 22:29:03 -0800465static ssize_t clear_refs_write(struct file *file, const char __user *buf,
466 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700467{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800468 struct task_struct *task;
469 char buffer[PROC_NUMBUF], *end;
470 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700471 struct vm_area_struct *vma;
472
Matt Mackallf248dcb2008-02-04 22:29:03 -0800473 memset(buffer, 0, sizeof(buffer));
474 if (count > sizeof(buffer) - 1)
475 count = sizeof(buffer) - 1;
476 if (copy_from_user(buffer, buf, count))
477 return -EFAULT;
478 if (!simple_strtol(buffer, &end, 0))
479 return -EINVAL;
480 if (*end == '\n')
481 end++;
482 task = get_proc_task(file->f_path.dentry->d_inode);
483 if (!task)
484 return -ESRCH;
485 mm = get_task_mm(task);
486 if (mm) {
Andrew Morton20cbc972008-07-05 12:29:05 -0700487 struct mm_walk clear_refs_walk = {
488 .pmd_entry = clear_refs_pte_range,
489 .mm = mm,
490 };
Matt Mackallf248dcb2008-02-04 22:29:03 -0800491 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700492 for (vma = mm->mmap; vma; vma = vma->vm_next) {
493 clear_refs_walk.private = vma;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800494 if (!is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700495 walk_page_range(vma->vm_start, vma->vm_end,
496 &clear_refs_walk);
497 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800498 flush_tlb_mm(mm);
499 up_read(&mm->mmap_sem);
500 mmput(mm);
501 }
502 put_task_struct(task);
503 if (end - buffer == 0)
504 return -EIO;
505 return end - buffer;
David Rientjesb813e932007-05-06 14:49:24 -0700506}
507
Matt Mackallf248dcb2008-02-04 22:29:03 -0800508const struct file_operations proc_clear_refs_operations = {
509 .write = clear_refs_write,
510};
511
Matt Mackall85863e42008-02-04 22:29:04 -0800512struct pagemapread {
Thomas Tuttleaae86792008-06-05 22:46:31 -0700513 u64 __user *out, *end;
Matt Mackall85863e42008-02-04 22:29:04 -0800514};
515
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500516#define PM_ENTRY_BYTES sizeof(u64)
517#define PM_STATUS_BITS 3
518#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
519#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
520#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
521#define PM_PSHIFT_BITS 6
522#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
523#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
524#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
525#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
526#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
527
528#define PM_PRESENT PM_STATUS(4LL)
529#define PM_SWAP PM_STATUS(2LL)
530#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800531#define PM_END_OF_BUFFER 1
532
533static int add_to_pagemap(unsigned long addr, u64 pfn,
534 struct pagemapread *pm)
535{
Matt Mackall85863e42008-02-04 22:29:04 -0800536 if (put_user(pfn, pm->out))
537 return -EFAULT;
Thomas Tuttleaae86792008-06-05 22:46:31 -0700538 pm->out++;
539 if (pm->out >= pm->end)
540 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800541 return 0;
542}
543
544static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700545 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800546{
Dave Hansen21650092008-06-12 15:21:47 -0700547 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800548 unsigned long addr;
549 int err = 0;
550 for (addr = start; addr < end; addr += PAGE_SIZE) {
551 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
552 if (err)
553 break;
554 }
555 return err;
556}
557
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700558static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800559{
560 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500561 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800562}
563
Matt Mackall49c50342008-12-09 13:14:21 -0800564static u64 pte_to_pagemap_entry(pte_t pte)
Dave Hansenbcf80392008-06-12 15:21:48 -0700565{
Matt Mackall49c50342008-12-09 13:14:21 -0800566 u64 pme = 0;
Dave Hansenbcf80392008-06-12 15:21:48 -0700567 if (is_swap_pte(pte))
568 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
569 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
570 else if (pte_present(pte))
571 pme = PM_PFRAME(pte_pfn(pte))
572 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
573 return pme;
574}
575
Matt Mackall85863e42008-02-04 22:29:04 -0800576static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700577 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800578{
Dave Hansenbcf80392008-06-12 15:21:48 -0700579 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700580 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800581 pte_t *pte;
582 int err = 0;
583
Dave Hansenbcf80392008-06-12 15:21:48 -0700584 /* find the first VMA at or above 'addr' */
585 vma = find_vma(walk->mm, addr);
Matt Mackall85863e42008-02-04 22:29:04 -0800586 for (; addr != end; addr += PAGE_SIZE) {
587 u64 pfn = PM_NOT_PRESENT;
Dave Hansenbcf80392008-06-12 15:21:48 -0700588
589 /* check to see if we've left 'vma' behind
590 * and need a new, higher one */
591 if (vma && (addr >= vma->vm_end))
592 vma = find_vma(walk->mm, addr);
593
594 /* check that 'vma' actually covers this address,
595 * and that it isn't a huge page vma */
596 if (vma && (vma->vm_start <= addr) &&
597 !is_vm_hugetlb_page(vma)) {
598 pte = pte_offset_map(pmd, addr);
599 pfn = pte_to_pagemap_entry(*pte);
600 /* unmap before userspace copy */
601 pte_unmap(pte);
602 }
Matt Mackall85863e42008-02-04 22:29:04 -0800603 err = add_to_pagemap(addr, pfn, pm);
604 if (err)
605 return err;
606 }
607
608 cond_resched();
609
610 return err;
611}
612
Matt Mackall85863e42008-02-04 22:29:04 -0800613/*
614 * /proc/pid/pagemap - an array mapping virtual pages to pfns
615 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500616 * For each page in the address space, this file contains one 64-bit entry
617 * consisting of the following:
618 *
619 * Bits 0-55 page frame number (PFN) if present
620 * Bits 0-4 swap type if swapped
621 * Bits 5-55 swap offset if swapped
622 * Bits 55-60 page shift (page size = 1<<page shift)
623 * Bit 61 reserved for future use
624 * Bit 62 page swapped
625 * Bit 63 page present
626 *
627 * If the page is not present but in swap, then the PFN contains an
628 * encoding of the swap file number and the page's offset into the
629 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800630 * precisely which pages are mapped (or in swap) and comparing mapped
631 * pages between processes.
632 *
633 * Efficient users of this interface will use /proc/pid/maps to
634 * determine which areas of memory are actually mapped and llseek to
635 * skip over unmapped regions.
636 */
637static ssize_t pagemap_read(struct file *file, char __user *buf,
638 size_t count, loff_t *ppos)
639{
640 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
641 struct page **pages, *page;
642 unsigned long uaddr, uend;
643 struct mm_struct *mm;
644 struct pagemapread pm;
645 int pagecount;
646 int ret = -ESRCH;
Alexey Dobriyanee1e6ab2008-07-21 14:21:36 -0700647 struct mm_walk pagemap_walk = {};
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700648 unsigned long src;
649 unsigned long svpfn;
650 unsigned long start_vaddr;
651 unsigned long end_vaddr;
Matt Mackall85863e42008-02-04 22:29:04 -0800652
653 if (!task)
654 goto out;
655
656 ret = -EACCES;
Stephen Smalley006ebb42008-05-19 08:32:49 -0400657 if (!ptrace_may_access(task, PTRACE_MODE_READ))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700658 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800659
660 ret = -EINVAL;
661 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700662 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700663 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800664
665 ret = 0;
666 mm = get_task_mm(task);
667 if (!mm)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700668 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800669
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700670
Matt Mackall85863e42008-02-04 22:29:04 -0800671 uaddr = (unsigned long)buf & PAGE_MASK;
672 uend = (unsigned long)(buf + count);
673 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700674 ret = 0;
675 if (pagecount == 0)
676 goto out_mm;
677 pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
678 ret = -ENOMEM;
Matt Mackall85863e42008-02-04 22:29:04 -0800679 if (!pages)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700680 goto out_mm;
Matt Mackall85863e42008-02-04 22:29:04 -0800681
682 down_read(&current->mm->mmap_sem);
683 ret = get_user_pages(current, current->mm, uaddr, pagecount,
684 1, 0, pages, NULL);
685 up_read(&current->mm->mmap_sem);
686
687 if (ret < 0)
688 goto out_free;
689
Marcelo Tosattifb393802008-03-13 12:32:35 -0700690 if (ret != pagecount) {
691 pagecount = ret;
692 ret = -EFAULT;
693 goto out_pages;
694 }
695
Thomas Tuttleaae86792008-06-05 22:46:31 -0700696 pm.out = (u64 *)buf;
697 pm.end = (u64 *)(buf + count);
Matt Mackall85863e42008-02-04 22:29:04 -0800698
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700699 pagemap_walk.pmd_entry = pagemap_pte_range;
700 pagemap_walk.pte_hole = pagemap_pte_hole;
701 pagemap_walk.mm = mm;
702 pagemap_walk.private = &pm;
Matt Mackall85863e42008-02-04 22:29:04 -0800703
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700704 src = *ppos;
705 svpfn = src / PM_ENTRY_BYTES;
706 start_vaddr = svpfn << PAGE_SHIFT;
707 end_vaddr = TASK_SIZE_OF(task);
Matt Mackall85863e42008-02-04 22:29:04 -0800708
Andrew Morton5d7e0d22008-07-05 01:02:01 -0700709 /* watch out for wraparound */
710 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
711 start_vaddr = end_vaddr;
712
713 /*
714 * The odds are that this will stop walking way
715 * before end_vaddr, because the length of the
716 * user buffer is tracked in "pm", and the walk
717 * will stop when we hit the end of the buffer.
718 */
719 ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
720 if (ret == PM_END_OF_BUFFER)
721 ret = 0;
722 /* don't need mmap_sem for these, but this looks cleaner */
723 *ppos += (char *)pm.out - buf;
724 if (!ret)
725 ret = (char *)pm.out - buf;
Matt Mackall85863e42008-02-04 22:29:04 -0800726
Marcelo Tosattifb393802008-03-13 12:32:35 -0700727out_pages:
Matt Mackall85863e42008-02-04 22:29:04 -0800728 for (; pagecount; pagecount--) {
729 page = pages[pagecount-1];
730 if (!PageReserved(page))
731 SetPageDirty(page);
732 page_cache_release(page);
733 }
Matt Mackall85863e42008-02-04 22:29:04 -0800734out_free:
735 kfree(pages);
Marcelo Tosattifb393802008-03-13 12:32:35 -0700736out_mm:
737 mmput(mm);
Matt Mackall85863e42008-02-04 22:29:04 -0800738out_task:
739 put_task_struct(task);
740out:
741 return ret;
742}
743
744const struct file_operations proc_pagemap_operations = {
745 .llseek = mem_lseek, /* borrow this */
746 .read = pagemap_read,
747};
Matt Mackall1e883282008-02-04 22:29:07 -0800748#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800749
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700750#ifdef CONFIG_NUMA
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800751extern int show_numa_map(struct seq_file *m, void *v);
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700752
Jan Engelhardt03a44822008-02-08 04:21:19 -0800753static const struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800754 .start = m_start,
755 .next = m_next,
756 .stop = m_stop,
Alexey Dobriyan3bbfe052008-10-10 03:27:16 +0400757 .show = show_numa_map,
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700758};
Eric W. Biederman662795d2006-06-26 00:25:48 -0700759
760static int numa_maps_open(struct inode *inode, struct file *file)
761{
762 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
763}
764
Arjan van de Ven00977a52007-02-12 00:55:34 -0800765const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -0700766 .open = numa_maps_open,
767 .read = seq_read,
768 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -0700769 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -0700770};
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700771#endif