blob: ab8ccc9d14ff6df10abdd715076d870dafb32a85 [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
Matt Mackall4752c362008-02-04 22:29:02 -0800201static int show_map(struct seq_file *m, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Eric W. Biederman99f89552006-06-26 00:25:55 -0700203 struct proc_maps_private *priv = m->private;
204 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700205 struct vm_area_struct *vma = v;
206 struct mm_struct *mm = vma->vm_mm;
207 struct file *file = vma->vm_file;
208 int flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 unsigned long ino = 0;
210 dev_t dev = 0;
211 int len;
212
Kees Cook5096add2007-05-08 00:26:04 -0700213 if (maps_protect && !ptrace_may_attach(task))
214 return -EACCES;
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 if (file) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800217 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 dev = inode->i_sb->s_dev;
219 ino = inode->i_ino;
220 }
221
222 seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
Mauricio Line070ad42005-09-03 15:55:10 -0700223 vma->vm_start,
224 vma->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 flags & VM_READ ? 'r' : '-',
226 flags & VM_WRITE ? 'w' : '-',
227 flags & VM_EXEC ? 'x' : '-',
228 flags & VM_MAYSHARE ? 's' : 'p',
Mauricio Line070ad42005-09-03 15:55:10 -0700229 vma->vm_pgoff << PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 MAJOR(dev), MINOR(dev), ino, &len);
231
232 /*
233 * Print the dentry name for named mappings, and a
234 * special [heap] marker for the heap:
235 */
Mauricio Line070ad42005-09-03 15:55:10 -0700236 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 pad_len_spaces(m, len);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800238 seq_path(m, &file->f_path, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 } else {
Ingo Molnare6e54942006-06-27 02:53:50 -0700240 const char *name = arch_vma_name(vma);
241 if (!name) {
242 if (mm) {
243 if (vma->vm_start <= mm->start_brk &&
Mauricio Line070ad42005-09-03 15:55:10 -0700244 vma->vm_end >= mm->brk) {
Ingo Molnare6e54942006-06-27 02:53:50 -0700245 name = "[heap]";
246 } else if (vma->vm_start <= mm->start_stack &&
247 vma->vm_end >= mm->start_stack) {
248 name = "[stack]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700250 } else {
251 name = "[vdso]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700253 }
254 if (name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 pad_len_spaces(m, len);
Ingo Molnare6e54942006-06-27 02:53:50 -0700256 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 }
258 }
259 seq_putc(m, '\n');
Mauricio Line070ad42005-09-03 15:55:10 -0700260
Mauricio Line070ad42005-09-03 15:55:10 -0700261 if (m->count < m->size) /* vma is copied successfully */
262 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return 0;
264}
265
Jan Engelhardt03a44822008-02-08 04:21:19 -0800266static const struct seq_operations proc_pid_maps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800267 .start = m_start,
268 .next = m_next,
269 .stop = m_stop,
270 .show = show_map
271};
272
273static int maps_open(struct inode *inode, struct file *file)
274{
275 return do_maps_open(inode, file, &proc_pid_maps_op);
276}
277
278const struct file_operations proc_maps_operations = {
279 .open = maps_open,
280 .read = seq_read,
281 .llseek = seq_lseek,
282 .release = seq_release_private,
283};
284
285/*
286 * Proportional Set Size(PSS): my share of RSS.
287 *
288 * PSS of a process is the count of pages it has in memory, where each
289 * page is divided by the number of processes sharing it. So if a
290 * process has 1000 pages all to itself, and 1000 shared with one other
291 * process, its PSS will be 1500.
292 *
293 * To keep (accumulated) division errors low, we adopt a 64bit
294 * fixed-point pss counter to minimize division errors. So (pss >>
295 * PSS_SHIFT) would be the real byte count.
296 *
297 * A shift of 12 before division means (assuming 4K page size):
298 * - 1M 3-user-pages add up to 8KB errors;
299 * - supports mapcount up to 2^24, or 16M;
300 * - supports PSS up to 2^52 bytes, or 4PB.
301 */
302#define PSS_SHIFT 12
303
Matt Mackall1e883282008-02-04 22:29:07 -0800304#ifdef CONFIG_PROC_PAGE_MONITOR
Peter Zijlstra214e4712008-04-28 02:12:55 -0700305struct mem_size_stats {
Matt Mackalla6198792008-02-04 22:29:03 -0800306 struct vm_area_struct *vma;
307 unsigned long resident;
308 unsigned long shared_clean;
309 unsigned long shared_dirty;
310 unsigned long private_clean;
311 unsigned long private_dirty;
312 unsigned long referenced;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700313 unsigned long swap;
Matt Mackalla6198792008-02-04 22:29:03 -0800314 u64 pss;
315};
316
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800317static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700318 struct mm_walk *walk)
Mauricio Line070ad42005-09-03 15:55:10 -0700319{
Dave Hansen21650092008-06-12 15:21:47 -0700320 struct mem_size_stats *mss = walk->private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800321 struct vm_area_struct *vma = mss->vma;
Mauricio Line070ad42005-09-03 15:55:10 -0700322 pte_t *pte, ptent;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700323 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700324 struct page *page;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800325 int mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700326
Hugh Dickins705e87c2005-10-29 18:16:27 -0700327 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
David Rientjes826fad12007-05-06 14:49:21 -0700328 for (; addr != end; pte++, addr += PAGE_SIZE) {
Mauricio Line070ad42005-09-03 15:55:10 -0700329 ptent = *pte;
Peter Zijlstra214e4712008-04-28 02:12:55 -0700330
331 if (is_swap_pte(ptent)) {
332 mss->swap += PAGE_SIZE;
333 continue;
334 }
335
Hugh Dickins705e87c2005-10-29 18:16:27 -0700336 if (!pte_present(ptent))
Mauricio Line070ad42005-09-03 15:55:10 -0700337 continue;
338
339 mss->resident += PAGE_SIZE;
Nick Pigginad820c5d2006-03-06 15:42:58 -0800340
341 page = vm_normal_page(vma, addr, ptent);
342 if (!page)
Mauricio Line070ad42005-09-03 15:55:10 -0700343 continue;
344
David Rientjesf79f1772007-05-06 14:49:22 -0700345 /* Accumulate the size in pages that have been accessed. */
346 if (pte_young(ptent) || PageReferenced(page))
347 mss->referenced += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800348 mapcount = page_mapcount(page);
349 if (mapcount >= 2) {
Mauricio Line070ad42005-09-03 15:55:10 -0700350 if (pte_dirty(ptent))
351 mss->shared_dirty += PAGE_SIZE;
352 else
353 mss->shared_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800354 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700355 } else {
356 if (pte_dirty(ptent))
357 mss->private_dirty += PAGE_SIZE;
358 else
359 mss->private_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800360 mss->pss += (PAGE_SIZE << PSS_SHIFT);
Mauricio Line070ad42005-09-03 15:55:10 -0700361 }
David Rientjes826fad12007-05-06 14:49:21 -0700362 }
Hugh Dickins705e87c2005-10-29 18:16:27 -0700363 pte_unmap_unlock(pte - 1, ptl);
364 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800365 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700366}
367
Mauricio Line070ad42005-09-03 15:55:10 -0700368static int show_smap(struct seq_file *m, void *v)
369{
370 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700371 struct mem_size_stats mss;
Matt Mackall4752c362008-02-04 22:29:02 -0800372 int ret;
Dave Hansen21650092008-06-12 15:21:47 -0700373 struct mm_walk smaps_walk = {
374 .pmd_entry = smaps_pte_range,
375 .mm = vma->vm_mm,
376 .private = &mss,
377 };
Mauricio Line070ad42005-09-03 15:55:10 -0700378
379 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800380 mss.vma = vma;
Nick Piggin5ddfae12006-03-06 15:42:57 -0800381 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700382 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
Matt Mackall4752c362008-02-04 22:29:02 -0800383
384 ret = show_map(m, v);
385 if (ret)
386 return ret;
387
388 seq_printf(m,
389 "Size: %8lu kB\n"
390 "Rss: %8lu kB\n"
391 "Pss: %8lu kB\n"
392 "Shared_Clean: %8lu kB\n"
393 "Shared_Dirty: %8lu kB\n"
394 "Private_Clean: %8lu kB\n"
395 "Private_Dirty: %8lu kB\n"
Peter Zijlstra214e4712008-04-28 02:12:55 -0700396 "Referenced: %8lu kB\n"
397 "Swap: %8lu kB\n",
Matt Mackall4752c362008-02-04 22:29:02 -0800398 (vma->vm_end - vma->vm_start) >> 10,
399 mss.resident >> 10,
400 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
401 mss.shared_clean >> 10,
402 mss.shared_dirty >> 10,
403 mss.private_clean >> 10,
404 mss.private_dirty >> 10,
Peter Zijlstra214e4712008-04-28 02:12:55 -0700405 mss.referenced >> 10,
406 mss.swap >> 10);
Matt Mackall4752c362008-02-04 22:29:02 -0800407
408 return ret;
Mauricio Line070ad42005-09-03 15:55:10 -0700409}
410
Jan Engelhardt03a44822008-02-08 04:21:19 -0800411static const struct seq_operations proc_pid_smaps_op = {
Matt Mackalla6198792008-02-04 22:29:03 -0800412 .start = m_start,
413 .next = m_next,
414 .stop = m_stop,
415 .show = show_smap
416};
417
418static int smaps_open(struct inode *inode, struct file *file)
419{
420 return do_maps_open(inode, file, &proc_pid_smaps_op);
421}
422
423const struct file_operations proc_smaps_operations = {
424 .open = smaps_open,
425 .read = seq_read,
426 .llseek = seq_lseek,
427 .release = seq_release_private,
428};
429
430static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
Dave Hansen21650092008-06-12 15:21:47 -0700431 unsigned long end, struct mm_walk *walk)
Matt Mackalla6198792008-02-04 22:29:03 -0800432{
Dave Hansen21650092008-06-12 15:21:47 -0700433 struct vm_area_struct *vma = walk->private;
Matt Mackalla6198792008-02-04 22:29:03 -0800434 pte_t *pte, ptent;
435 spinlock_t *ptl;
436 struct page *page;
437
438 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
439 for (; addr != end; pte++, addr += PAGE_SIZE) {
440 ptent = *pte;
441 if (!pte_present(ptent))
442 continue;
443
444 page = vm_normal_page(vma, addr, ptent);
445 if (!page)
446 continue;
447
448 /* Clear accessed and referenced bits. */
449 ptep_test_and_clear_young(vma, addr, pte);
450 ClearPageReferenced(page);
451 }
452 pte_unmap_unlock(pte - 1, ptl);
453 cond_resched();
454 return 0;
455}
456
Matt Mackallf248dcb2008-02-04 22:29:03 -0800457static ssize_t clear_refs_write(struct file *file, const char __user *buf,
458 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700459{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800460 struct task_struct *task;
461 char buffer[PROC_NUMBUF], *end;
462 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700463 struct vm_area_struct *vma;
464
Matt Mackallf248dcb2008-02-04 22:29:03 -0800465 memset(buffer, 0, sizeof(buffer));
466 if (count > sizeof(buffer) - 1)
467 count = sizeof(buffer) - 1;
468 if (copy_from_user(buffer, buf, count))
469 return -EFAULT;
470 if (!simple_strtol(buffer, &end, 0))
471 return -EINVAL;
472 if (*end == '\n')
473 end++;
474 task = get_proc_task(file->f_path.dentry->d_inode);
475 if (!task)
476 return -ESRCH;
477 mm = get_task_mm(task);
478 if (mm) {
Dave Hansen21650092008-06-12 15:21:47 -0700479 static struct mm_walk clear_refs_walk;
480 memset(&clear_refs_walk, 0, sizeof(clear_refs_walk));
481 clear_refs_walk.pmd_entry = clear_refs_pte_range;
482 clear_refs_walk.mm = mm;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800483 down_read(&mm->mmap_sem);
Dave Hansen21650092008-06-12 15:21:47 -0700484 for (vma = mm->mmap; vma; vma = vma->vm_next) {
485 clear_refs_walk.private = vma;
Matt Mackallf248dcb2008-02-04 22:29:03 -0800486 if (!is_vm_hugetlb_page(vma))
Dave Hansen21650092008-06-12 15:21:47 -0700487 walk_page_range(vma->vm_start, vma->vm_end,
488 &clear_refs_walk);
489 }
Matt Mackallf248dcb2008-02-04 22:29:03 -0800490 flush_tlb_mm(mm);
491 up_read(&mm->mmap_sem);
492 mmput(mm);
493 }
494 put_task_struct(task);
495 if (end - buffer == 0)
496 return -EIO;
497 return end - buffer;
David Rientjesb813e932007-05-06 14:49:24 -0700498}
499
Matt Mackallf248dcb2008-02-04 22:29:03 -0800500const struct file_operations proc_clear_refs_operations = {
501 .write = clear_refs_write,
502};
503
Matt Mackall85863e42008-02-04 22:29:04 -0800504struct pagemapread {
Thomas Tuttleaae86792008-06-05 22:46:31 -0700505 u64 __user *out, *end;
Matt Mackall85863e42008-02-04 22:29:04 -0800506};
507
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500508#define PM_ENTRY_BYTES sizeof(u64)
509#define PM_STATUS_BITS 3
510#define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
511#define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
512#define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
513#define PM_PSHIFT_BITS 6
514#define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
515#define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
516#define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
517#define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
518#define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
519
520#define PM_PRESENT PM_STATUS(4LL)
521#define PM_SWAP PM_STATUS(2LL)
522#define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
Matt Mackall85863e42008-02-04 22:29:04 -0800523#define PM_END_OF_BUFFER 1
524
525static int add_to_pagemap(unsigned long addr, u64 pfn,
526 struct pagemapread *pm)
527{
Matt Mackall85863e42008-02-04 22:29:04 -0800528 if (put_user(pfn, pm->out))
529 return -EFAULT;
Thomas Tuttleaae86792008-06-05 22:46:31 -0700530 pm->out++;
531 if (pm->out >= pm->end)
532 return PM_END_OF_BUFFER;
Matt Mackall85863e42008-02-04 22:29:04 -0800533 return 0;
534}
535
536static int pagemap_pte_hole(unsigned long start, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700537 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800538{
Dave Hansen21650092008-06-12 15:21:47 -0700539 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800540 unsigned long addr;
541 int err = 0;
542 for (addr = start; addr < end; addr += PAGE_SIZE) {
543 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
544 if (err)
545 break;
546 }
547 return err;
548}
549
Adrian Bunk9d02dbc2008-04-28 02:12:11 -0700550static u64 swap_pte_to_pagemap_entry(pte_t pte)
Matt Mackall85863e42008-02-04 22:29:04 -0800551{
552 swp_entry_t e = pte_to_swp_entry(pte);
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500553 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
Matt Mackall85863e42008-02-04 22:29:04 -0800554}
555
Dave Hansenbcf80392008-06-12 15:21:48 -0700556static unsigned long pte_to_pagemap_entry(pte_t pte)
557{
558 unsigned long pme = 0;
559 if (is_swap_pte(pte))
560 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
561 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
562 else if (pte_present(pte))
563 pme = PM_PFRAME(pte_pfn(pte))
564 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
565 return pme;
566}
567
Matt Mackall85863e42008-02-04 22:29:04 -0800568static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
Dave Hansen21650092008-06-12 15:21:47 -0700569 struct mm_walk *walk)
Matt Mackall85863e42008-02-04 22:29:04 -0800570{
Dave Hansenbcf80392008-06-12 15:21:48 -0700571 struct vm_area_struct *vma;
Dave Hansen21650092008-06-12 15:21:47 -0700572 struct pagemapread *pm = walk->private;
Matt Mackall85863e42008-02-04 22:29:04 -0800573 pte_t *pte;
574 int err = 0;
575
Dave Hansenbcf80392008-06-12 15:21:48 -0700576 /* find the first VMA at or above 'addr' */
577 vma = find_vma(walk->mm, addr);
Matt Mackall85863e42008-02-04 22:29:04 -0800578 for (; addr != end; addr += PAGE_SIZE) {
579 u64 pfn = PM_NOT_PRESENT;
Dave Hansenbcf80392008-06-12 15:21:48 -0700580
581 /* check to see if we've left 'vma' behind
582 * and need a new, higher one */
583 if (vma && (addr >= vma->vm_end))
584 vma = find_vma(walk->mm, addr);
585
586 /* check that 'vma' actually covers this address,
587 * and that it isn't a huge page vma */
588 if (vma && (vma->vm_start <= addr) &&
589 !is_vm_hugetlb_page(vma)) {
590 pte = pte_offset_map(pmd, addr);
591 pfn = pte_to_pagemap_entry(*pte);
592 /* unmap before userspace copy */
593 pte_unmap(pte);
594 }
Matt Mackall85863e42008-02-04 22:29:04 -0800595 err = add_to_pagemap(addr, pfn, pm);
596 if (err)
597 return err;
598 }
599
600 cond_resched();
601
602 return err;
603}
604
605static struct mm_walk pagemap_walk = {
606 .pmd_entry = pagemap_pte_range,
607 .pte_hole = pagemap_pte_hole
608};
609
610/*
611 * /proc/pid/pagemap - an array mapping virtual pages to pfns
612 *
Hans Rosenfeldf16278c2008-03-21 18:46:59 -0500613 * For each page in the address space, this file contains one 64-bit entry
614 * consisting of the following:
615 *
616 * Bits 0-55 page frame number (PFN) if present
617 * Bits 0-4 swap type if swapped
618 * Bits 5-55 swap offset if swapped
619 * Bits 55-60 page shift (page size = 1<<page shift)
620 * Bit 61 reserved for future use
621 * Bit 62 page swapped
622 * Bit 63 page present
623 *
624 * If the page is not present but in swap, then the PFN contains an
625 * encoding of the swap file number and the page's offset into the
626 * swap. Unmapped pages return a null PFN. This allows determining
Matt Mackall85863e42008-02-04 22:29:04 -0800627 * precisely which pages are mapped (or in swap) and comparing mapped
628 * pages between processes.
629 *
630 * Efficient users of this interface will use /proc/pid/maps to
631 * determine which areas of memory are actually mapped and llseek to
632 * skip over unmapped regions.
633 */
634static ssize_t pagemap_read(struct file *file, char __user *buf,
635 size_t count, loff_t *ppos)
636{
637 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
638 struct page **pages, *page;
639 unsigned long uaddr, uend;
640 struct mm_struct *mm;
641 struct pagemapread pm;
642 int pagecount;
643 int ret = -ESRCH;
644
645 if (!task)
646 goto out;
647
648 ret = -EACCES;
649 if (!ptrace_may_attach(task))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700650 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800651
652 ret = -EINVAL;
653 /* file position must be aligned */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700654 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
Marcelo Tosattifb393802008-03-13 12:32:35 -0700655 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800656
657 ret = 0;
658 mm = get_task_mm(task);
659 if (!mm)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700660 goto out_task;
Matt Mackall85863e42008-02-04 22:29:04 -0800661
662 ret = -ENOMEM;
663 uaddr = (unsigned long)buf & PAGE_MASK;
664 uend = (unsigned long)(buf + count);
665 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
666 pages = kmalloc(pagecount * sizeof(struct page *), GFP_KERNEL);
667 if (!pages)
Marcelo Tosattifb393802008-03-13 12:32:35 -0700668 goto out_mm;
Matt Mackall85863e42008-02-04 22:29:04 -0800669
670 down_read(&current->mm->mmap_sem);
671 ret = get_user_pages(current, current->mm, uaddr, pagecount,
672 1, 0, pages, NULL);
673 up_read(&current->mm->mmap_sem);
674
675 if (ret < 0)
676 goto out_free;
677
Marcelo Tosattifb393802008-03-13 12:32:35 -0700678 if (ret != pagecount) {
679 pagecount = ret;
680 ret = -EFAULT;
681 goto out_pages;
682 }
683
Thomas Tuttleaae86792008-06-05 22:46:31 -0700684 pm.out = (u64 *)buf;
685 pm.end = (u64 *)(buf + count);
Matt Mackall85863e42008-02-04 22:29:04 -0800686
687 if (!ptrace_may_attach(task)) {
688 ret = -EIO;
689 } else {
690 unsigned long src = *ppos;
691 unsigned long svpfn = src / PM_ENTRY_BYTES;
692 unsigned long start_vaddr = svpfn << PAGE_SHIFT;
693 unsigned long end_vaddr = TASK_SIZE_OF(task);
694
695 /* watch out for wraparound */
696 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
697 start_vaddr = end_vaddr;
698
699 /*
700 * The odds are that this will stop walking way
701 * before end_vaddr, because the length of the
702 * user buffer is tracked in "pm", and the walk
703 * will stop when we hit the end of the buffer.
704 */
Dave Hansen21650092008-06-12 15:21:47 -0700705 ret = walk_page_range(start_vaddr, end_vaddr,
706 &pagemap_walk);
Matt Mackall85863e42008-02-04 22:29:04 -0800707 if (ret == PM_END_OF_BUFFER)
708 ret = 0;
709 /* don't need mmap_sem for these, but this looks cleaner */
Thomas Tuttleaae86792008-06-05 22:46:31 -0700710 *ppos += (char *)pm.out - buf;
Matt Mackall85863e42008-02-04 22:29:04 -0800711 if (!ret)
Thomas Tuttleaae86792008-06-05 22:46:31 -0700712 ret = (char *)pm.out - buf;
Matt Mackall85863e42008-02-04 22:29:04 -0800713 }
714
Marcelo Tosattifb393802008-03-13 12:32:35 -0700715out_pages:
Matt Mackall85863e42008-02-04 22:29:04 -0800716 for (; pagecount; pagecount--) {
717 page = pages[pagecount-1];
718 if (!PageReserved(page))
719 SetPageDirty(page);
720 page_cache_release(page);
721 }
Matt Mackall85863e42008-02-04 22:29:04 -0800722out_free:
723 kfree(pages);
Marcelo Tosattifb393802008-03-13 12:32:35 -0700724out_mm:
725 mmput(mm);
Matt Mackall85863e42008-02-04 22:29:04 -0800726out_task:
727 put_task_struct(task);
728out:
729 return ret;
730}
731
732const struct file_operations proc_pagemap_operations = {
733 .llseek = mem_lseek, /* borrow this */
734 .read = pagemap_read,
735};
Matt Mackall1e883282008-02-04 22:29:07 -0800736#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800737
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700738#ifdef CONFIG_NUMA
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800739extern int show_numa_map(struct seq_file *m, void *v);
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700740
Kees Cook5096add2007-05-08 00:26:04 -0700741static int show_numa_map_checked(struct seq_file *m, void *v)
742{
743 struct proc_maps_private *priv = m->private;
744 struct task_struct *task = priv->task;
745
746 if (maps_protect && !ptrace_may_attach(task))
747 return -EACCES;
748
749 return show_numa_map(m, v);
750}
751
Jan Engelhardt03a44822008-02-08 04:21:19 -0800752static const struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800753 .start = m_start,
754 .next = m_next,
755 .stop = m_stop,
Kees Cook5096add2007-05-08 00:26:04 -0700756 .show = show_numa_map_checked
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700757};
Eric W. Biederman662795d2006-06-26 00:25:48 -0700758
759static int numa_maps_open(struct inode *inode, struct file *file)
760{
761 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
762}
763
Arjan van de Ven00977a52007-02-12 00:55:34 -0800764const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -0700765 .open = numa_maps_open,
766 .read = seq_read,
767 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -0700768 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -0700769};
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700770#endif