blob: 38338ed98cc6df167febbe2fc962f55a700bfb19 [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>
Matt Mackall85863e42008-02-04 22:29:04 -08008#include <linux/ptrace.h>
Christoph Lameter6e21c8f2005-09-03 15:54:45 -07009#include <linux/mempolicy.h>
Matt Mackall85863e42008-02-04 22:29:04 -080010#include <linux/swap.h>
11#include <linux/swapops.h>
Mauricio Line070ad42005-09-03 15:55:10 -070012
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <asm/elf.h>
14#include <asm/uaccess.h>
Mauricio Line070ad42005-09-03 15:55:10 -070015#include <asm/tlbflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include "internal.h"
17
18char *task_mem(struct mm_struct *mm, char *buffer)
19{
20 unsigned long data, text, lib;
Hugh Dickins365e9c872005-10-29 18:16:18 -070021 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
22
23 /*
24 * Note: to minimize their overhead, mm maintains hiwater_vm and
25 * hiwater_rss only when about to *lower* total_vm or rss. Any
26 * collector of these hiwater stats must therefore get total_vm
27 * and rss too, which will usually be the higher. Barriers? not
28 * worth the effort, such snapshots can always be inconsistent.
29 */
30 hiwater_vm = total_vm = mm->total_vm;
31 if (hiwater_vm < mm->hiwater_vm)
32 hiwater_vm = mm->hiwater_vm;
33 hiwater_rss = total_rss = get_mm_rss(mm);
34 if (hiwater_rss < mm->hiwater_rss)
35 hiwater_rss = mm->hiwater_rss;
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
38 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
39 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
40 buffer += sprintf(buffer,
Hugh Dickins365e9c872005-10-29 18:16:18 -070041 "VmPeak:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 "VmSize:\t%8lu kB\n"
43 "VmLck:\t%8lu kB\n"
Hugh Dickins365e9c872005-10-29 18:16:18 -070044 "VmHWM:\t%8lu kB\n"
Linus Torvalds1da177e2005-04-16 15:20:36 -070045 "VmRSS:\t%8lu kB\n"
46 "VmData:\t%8lu kB\n"
47 "VmStk:\t%8lu kB\n"
48 "VmExe:\t%8lu kB\n"
49 "VmLib:\t%8lu kB\n"
50 "VmPTE:\t%8lu kB\n",
Hugh Dickins365e9c872005-10-29 18:16:18 -070051 hiwater_vm << (PAGE_SHIFT-10),
52 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070053 mm->locked_vm << (PAGE_SHIFT-10),
Hugh Dickins365e9c872005-10-29 18:16:18 -070054 hiwater_rss << (PAGE_SHIFT-10),
55 total_rss << (PAGE_SHIFT-10),
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 data << (PAGE_SHIFT-10),
57 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
58 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
59 return buffer;
60}
61
62unsigned long task_vsize(struct mm_struct *mm)
63{
64 return PAGE_SIZE * mm->total_vm;
65}
66
67int task_statm(struct mm_struct *mm, int *shared, int *text,
68 int *data, int *resident)
69{
Hugh Dickins42946212005-10-29 18:16:05 -070070 *shared = get_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
72 >> PAGE_SHIFT;
73 *data = mm->total_vm - mm->shared_vm;
Hugh Dickins42946212005-10-29 18:16:05 -070074 *resident = *shared + get_mm_counter(mm, anon_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 return mm->total_vm;
76}
77
78int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
79{
80 struct vm_area_struct * vma;
81 int result = -ENOENT;
Eric W. Biederman99f89552006-06-26 00:25:55 -070082 struct task_struct *task = get_proc_task(inode);
83 struct mm_struct * mm = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Eric W. Biederman99f89552006-06-26 00:25:55 -070085 if (task) {
86 mm = get_task_mm(task);
87 put_task_struct(task);
88 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 if (!mm)
90 goto out;
91 down_read(&mm->mmap_sem);
92
93 vma = mm->mmap;
94 while (vma) {
95 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
96 break;
97 vma = vma->vm_next;
98 }
99
100 if (vma) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800101 *mnt = mntget(vma->vm_file->f_path.mnt);
102 *dentry = dget(vma->vm_file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 result = 0;
104 }
105
106 up_read(&mm->mmap_sem);
107 mmput(mm);
108out:
109 return result;
110}
111
112static void pad_len_spaces(struct seq_file *m, int len)
113{
114 len = 25 + sizeof(void*) * 6 - len;
115 if (len < 1)
116 len = 1;
117 seq_printf(m, "%*c", len, ' ');
118}
119
Matt Mackalla6198792008-02-04 22:29:03 -0800120static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
Mauricio Line070ad42005-09-03 15:55:10 -0700121{
Matt Mackalla6198792008-02-04 22:29:03 -0800122 if (vma && vma != priv->tail_vma) {
123 struct mm_struct *mm = vma->vm_mm;
124 up_read(&mm->mmap_sem);
125 mmput(mm);
126 }
127}
128
129static void *m_start(struct seq_file *m, loff_t *pos)
130{
131 struct proc_maps_private *priv = m->private;
132 unsigned long last_addr = m->version;
133 struct mm_struct *mm;
134 struct vm_area_struct *vma, *tail_vma = NULL;
135 loff_t l = *pos;
136
137 /* Clear the per syscall fields in priv */
138 priv->task = NULL;
139 priv->tail_vma = NULL;
140
141 /*
142 * We remember last_addr rather than next_addr to hit with
143 * mmap_cache most of the time. We have zero last_addr at
144 * the beginning and also after lseek. We will have -1 last_addr
145 * after the end of the vmas.
146 */
147
148 if (last_addr == -1UL)
149 return NULL;
150
151 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
152 if (!priv->task)
153 return NULL;
154
155 mm = mm_for_maps(priv->task);
156 if (!mm)
157 return NULL;
158
159 tail_vma = get_gate_vma(priv->task);
160 priv->tail_vma = tail_vma;
161
162 /* Start with last addr hint */
163 vma = find_vma(mm, last_addr);
164 if (last_addr && vma) {
165 vma = vma->vm_next;
166 goto out;
167 }
168
169 /*
170 * Check the vma index is within the range and do
171 * sequential scan until m_index.
172 */
173 vma = NULL;
174 if ((unsigned long)l < mm->map_count) {
175 vma = mm->mmap;
176 while (l-- && vma)
177 vma = vma->vm_next;
178 goto out;
179 }
180
181 if (l != mm->map_count)
182 tail_vma = NULL; /* After gate vma */
183
184out:
185 if (vma)
186 return vma;
187
188 /* End of vmas has been reached */
189 m->version = (tail_vma != NULL)? 0: -1UL;
190 up_read(&mm->mmap_sem);
191 mmput(mm);
192 return tail_vma;
193}
194
195static void *m_next(struct seq_file *m, void *v, loff_t *pos)
196{
197 struct proc_maps_private *priv = m->private;
198 struct vm_area_struct *vma = v;
199 struct vm_area_struct *tail_vma = priv->tail_vma;
200
201 (*pos)++;
202 if (vma && (vma != tail_vma) && vma->vm_next)
203 return vma->vm_next;
204 vma_stop(priv, vma);
205 return (vma != tail_vma)? tail_vma: NULL;
206}
207
208static void m_stop(struct seq_file *m, void *v)
209{
210 struct proc_maps_private *priv = m->private;
211 struct vm_area_struct *vma = v;
212
213 vma_stop(priv, vma);
214 if (priv->task)
215 put_task_struct(priv->task);
216}
217
218static int do_maps_open(struct inode *inode, struct file *file,
219 struct seq_operations *ops)
220{
221 struct proc_maps_private *priv;
222 int ret = -ENOMEM;
223 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
224 if (priv) {
225 priv->pid = proc_pid(inode);
226 ret = seq_open(file, ops);
227 if (!ret) {
228 struct seq_file *m = file->private_data;
229 m->private = priv;
230 } else {
231 kfree(priv);
232 }
233 }
234 return ret;
235}
Mauricio Line070ad42005-09-03 15:55:10 -0700236
Matt Mackall4752c362008-02-04 22:29:02 -0800237static int show_map(struct seq_file *m, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238{
Eric W. Biederman99f89552006-06-26 00:25:55 -0700239 struct proc_maps_private *priv = m->private;
240 struct task_struct *task = priv->task;
Mauricio Line070ad42005-09-03 15:55:10 -0700241 struct vm_area_struct *vma = v;
242 struct mm_struct *mm = vma->vm_mm;
243 struct file *file = vma->vm_file;
244 int flags = vma->vm_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 unsigned long ino = 0;
246 dev_t dev = 0;
247 int len;
248
Kees Cook5096add2007-05-08 00:26:04 -0700249 if (maps_protect && !ptrace_may_attach(task))
250 return -EACCES;
251
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (file) {
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800253 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 dev = inode->i_sb->s_dev;
255 ino = inode->i_ino;
256 }
257
258 seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
Mauricio Line070ad42005-09-03 15:55:10 -0700259 vma->vm_start,
260 vma->vm_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 flags & VM_READ ? 'r' : '-',
262 flags & VM_WRITE ? 'w' : '-',
263 flags & VM_EXEC ? 'x' : '-',
264 flags & VM_MAYSHARE ? 's' : 'p',
Mauricio Line070ad42005-09-03 15:55:10 -0700265 vma->vm_pgoff << PAGE_SHIFT,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 MAJOR(dev), MINOR(dev), ino, &len);
267
268 /*
269 * Print the dentry name for named mappings, and a
270 * special [heap] marker for the heap:
271 */
Mauricio Line070ad42005-09-03 15:55:10 -0700272 if (file) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 pad_len_spaces(m, len);
Josef "Jeff" Sipek2fddfee2006-12-08 02:36:36 -0800274 seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 } else {
Ingo Molnare6e54942006-06-27 02:53:50 -0700276 const char *name = arch_vma_name(vma);
277 if (!name) {
278 if (mm) {
279 if (vma->vm_start <= mm->start_brk &&
Mauricio Line070ad42005-09-03 15:55:10 -0700280 vma->vm_end >= mm->brk) {
Ingo Molnare6e54942006-06-27 02:53:50 -0700281 name = "[heap]";
282 } else if (vma->vm_start <= mm->start_stack &&
283 vma->vm_end >= mm->start_stack) {
284 name = "[stack]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700286 } else {
287 name = "[vdso]";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 }
Ingo Molnare6e54942006-06-27 02:53:50 -0700289 }
290 if (name) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 pad_len_spaces(m, len);
Ingo Molnare6e54942006-06-27 02:53:50 -0700292 seq_puts(m, name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 }
294 }
295 seq_putc(m, '\n');
Mauricio Line070ad42005-09-03 15:55:10 -0700296
Mauricio Line070ad42005-09-03 15:55:10 -0700297 if (m->count < m->size) /* vma is copied successfully */
298 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 return 0;
300}
301
Matt Mackalla6198792008-02-04 22:29:03 -0800302static struct seq_operations proc_pid_maps_op = {
303 .start = m_start,
304 .next = m_next,
305 .stop = m_stop,
306 .show = show_map
307};
308
309static int maps_open(struct inode *inode, struct file *file)
310{
311 return do_maps_open(inode, file, &proc_pid_maps_op);
312}
313
314const struct file_operations proc_maps_operations = {
315 .open = maps_open,
316 .read = seq_read,
317 .llseek = seq_lseek,
318 .release = seq_release_private,
319};
320
321/*
322 * Proportional Set Size(PSS): my share of RSS.
323 *
324 * PSS of a process is the count of pages it has in memory, where each
325 * page is divided by the number of processes sharing it. So if a
326 * process has 1000 pages all to itself, and 1000 shared with one other
327 * process, its PSS will be 1500.
328 *
329 * To keep (accumulated) division errors low, we adopt a 64bit
330 * fixed-point pss counter to minimize division errors. So (pss >>
331 * PSS_SHIFT) would be the real byte count.
332 *
333 * A shift of 12 before division means (assuming 4K page size):
334 * - 1M 3-user-pages add up to 8KB errors;
335 * - supports mapcount up to 2^24, or 16M;
336 * - supports PSS up to 2^52 bytes, or 4PB.
337 */
338#define PSS_SHIFT 12
339
Matt Mackall1e883282008-02-04 22:29:07 -0800340#ifdef CONFIG_PROC_PAGE_MONITOR
Matt Mackalla6198792008-02-04 22:29:03 -0800341struct mem_size_stats
342{
343 struct vm_area_struct *vma;
344 unsigned long resident;
345 unsigned long shared_clean;
346 unsigned long shared_dirty;
347 unsigned long private_clean;
348 unsigned long private_dirty;
349 unsigned long referenced;
350 u64 pss;
351};
352
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800353static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
354 void *private)
Mauricio Line070ad42005-09-03 15:55:10 -0700355{
David Rientjes826fad12007-05-06 14:49:21 -0700356 struct mem_size_stats *mss = private;
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800357 struct vm_area_struct *vma = mss->vma;
Mauricio Line070ad42005-09-03 15:55:10 -0700358 pte_t *pte, ptent;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700359 spinlock_t *ptl;
Mauricio Line070ad42005-09-03 15:55:10 -0700360 struct page *page;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800361 int mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700362
Hugh Dickins705e87c2005-10-29 18:16:27 -0700363 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
David Rientjes826fad12007-05-06 14:49:21 -0700364 for (; addr != end; pte++, addr += PAGE_SIZE) {
Mauricio Line070ad42005-09-03 15:55:10 -0700365 ptent = *pte;
Hugh Dickins705e87c2005-10-29 18:16:27 -0700366 if (!pte_present(ptent))
Mauricio Line070ad42005-09-03 15:55:10 -0700367 continue;
368
369 mss->resident += PAGE_SIZE;
Nick Pigginad820c5d2006-03-06 15:42:58 -0800370
371 page = vm_normal_page(vma, addr, ptent);
372 if (!page)
Mauricio Line070ad42005-09-03 15:55:10 -0700373 continue;
374
David Rientjesf79f1772007-05-06 14:49:22 -0700375 /* Accumulate the size in pages that have been accessed. */
376 if (pte_young(ptent) || PageReferenced(page))
377 mss->referenced += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800378 mapcount = page_mapcount(page);
379 if (mapcount >= 2) {
Mauricio Line070ad42005-09-03 15:55:10 -0700380 if (pte_dirty(ptent))
381 mss->shared_dirty += PAGE_SIZE;
382 else
383 mss->shared_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800384 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
Mauricio Line070ad42005-09-03 15:55:10 -0700385 } else {
386 if (pte_dirty(ptent))
387 mss->private_dirty += PAGE_SIZE;
388 else
389 mss->private_clean += PAGE_SIZE;
Fengguang Wuec4dd3e2008-02-04 22:28:56 -0800390 mss->pss += (PAGE_SIZE << PSS_SHIFT);
Mauricio Line070ad42005-09-03 15:55:10 -0700391 }
David Rientjes826fad12007-05-06 14:49:21 -0700392 }
Hugh Dickins705e87c2005-10-29 18:16:27 -0700393 pte_unmap_unlock(pte - 1, ptl);
394 cond_resched();
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800395 return 0;
Mauricio Line070ad42005-09-03 15:55:10 -0700396}
397
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800398static struct mm_walk smaps_walk = { .pmd_entry = smaps_pte_range };
Mauricio Line070ad42005-09-03 15:55:10 -0700399
400static int show_smap(struct seq_file *m, void *v)
401{
402 struct vm_area_struct *vma = v;
Mauricio Line070ad42005-09-03 15:55:10 -0700403 struct mem_size_stats mss;
Matt Mackall4752c362008-02-04 22:29:02 -0800404 int ret;
Mauricio Line070ad42005-09-03 15:55:10 -0700405
406 memset(&mss, 0, sizeof mss);
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800407 mss.vma = vma;
Nick Piggin5ddfae12006-03-06 15:42:57 -0800408 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800409 walk_page_range(vma->vm_mm, vma->vm_start, vma->vm_end,
410 &smaps_walk, &mss);
Matt Mackall4752c362008-02-04 22:29:02 -0800411
412 ret = show_map(m, v);
413 if (ret)
414 return ret;
415
416 seq_printf(m,
417 "Size: %8lu kB\n"
418 "Rss: %8lu kB\n"
419 "Pss: %8lu kB\n"
420 "Shared_Clean: %8lu kB\n"
421 "Shared_Dirty: %8lu kB\n"
422 "Private_Clean: %8lu kB\n"
423 "Private_Dirty: %8lu kB\n"
424 "Referenced: %8lu kB\n",
425 (vma->vm_end - vma->vm_start) >> 10,
426 mss.resident >> 10,
427 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
428 mss.shared_clean >> 10,
429 mss.shared_dirty >> 10,
430 mss.private_clean >> 10,
431 mss.private_dirty >> 10,
432 mss.referenced >> 10);
433
434 return ret;
Mauricio Line070ad42005-09-03 15:55:10 -0700435}
436
Matt Mackalla6198792008-02-04 22:29:03 -0800437static struct seq_operations proc_pid_smaps_op = {
438 .start = m_start,
439 .next = m_next,
440 .stop = m_stop,
441 .show = show_smap
442};
443
444static int smaps_open(struct inode *inode, struct file *file)
445{
446 return do_maps_open(inode, file, &proc_pid_smaps_op);
447}
448
449const struct file_operations proc_smaps_operations = {
450 .open = smaps_open,
451 .read = seq_read,
452 .llseek = seq_lseek,
453 .release = seq_release_private,
454};
455
456static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
457 unsigned long end, void *private)
458{
459 struct vm_area_struct *vma = private;
460 pte_t *pte, ptent;
461 spinlock_t *ptl;
462 struct page *page;
463
464 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
465 for (; addr != end; pte++, addr += PAGE_SIZE) {
466 ptent = *pte;
467 if (!pte_present(ptent))
468 continue;
469
470 page = vm_normal_page(vma, addr, ptent);
471 if (!page)
472 continue;
473
474 /* Clear accessed and referenced bits. */
475 ptep_test_and_clear_young(vma, addr, pte);
476 ClearPageReferenced(page);
477 }
478 pte_unmap_unlock(pte - 1, ptl);
479 cond_resched();
480 return 0;
481}
482
Matt Mackallb3ae5ac2008-02-04 22:29:01 -0800483static struct mm_walk clear_refs_walk = { .pmd_entry = clear_refs_pte_range };
484
Matt Mackallf248dcb2008-02-04 22:29:03 -0800485static ssize_t clear_refs_write(struct file *file, const char __user *buf,
486 size_t count, loff_t *ppos)
David Rientjesb813e932007-05-06 14:49:24 -0700487{
Matt Mackallf248dcb2008-02-04 22:29:03 -0800488 struct task_struct *task;
489 char buffer[PROC_NUMBUF], *end;
490 struct mm_struct *mm;
David Rientjesb813e932007-05-06 14:49:24 -0700491 struct vm_area_struct *vma;
492
Matt Mackallf248dcb2008-02-04 22:29:03 -0800493 memset(buffer, 0, sizeof(buffer));
494 if (count > sizeof(buffer) - 1)
495 count = sizeof(buffer) - 1;
496 if (copy_from_user(buffer, buf, count))
497 return -EFAULT;
498 if (!simple_strtol(buffer, &end, 0))
499 return -EINVAL;
500 if (*end == '\n')
501 end++;
502 task = get_proc_task(file->f_path.dentry->d_inode);
503 if (!task)
504 return -ESRCH;
505 mm = get_task_mm(task);
506 if (mm) {
507 down_read(&mm->mmap_sem);
508 for (vma = mm->mmap; vma; vma = vma->vm_next)
509 if (!is_vm_hugetlb_page(vma))
510 walk_page_range(mm, vma->vm_start, vma->vm_end,
511 &clear_refs_walk, vma);
512 flush_tlb_mm(mm);
513 up_read(&mm->mmap_sem);
514 mmput(mm);
515 }
516 put_task_struct(task);
517 if (end - buffer == 0)
518 return -EIO;
519 return end - buffer;
David Rientjesb813e932007-05-06 14:49:24 -0700520}
521
Matt Mackallf248dcb2008-02-04 22:29:03 -0800522const struct file_operations proc_clear_refs_operations = {
523 .write = clear_refs_write,
524};
525
Matt Mackall85863e42008-02-04 22:29:04 -0800526struct pagemapread {
527 char __user *out, *end;
528};
529
530#define PM_ENTRY_BYTES sizeof(u64)
531#define PM_RESERVED_BITS 3
532#define PM_RESERVED_OFFSET (64 - PM_RESERVED_BITS)
533#define PM_RESERVED_MASK (((1LL<<PM_RESERVED_BITS)-1) << PM_RESERVED_OFFSET)
534#define PM_SPECIAL(nr) (((nr) << PM_RESERVED_OFFSET) | PM_RESERVED_MASK)
535#define PM_NOT_PRESENT PM_SPECIAL(1LL)
536#define PM_SWAP PM_SPECIAL(2LL)
537#define PM_END_OF_BUFFER 1
538
539static int add_to_pagemap(unsigned long addr, u64 pfn,
540 struct pagemapread *pm)
541{
542 /*
543 * Make sure there's room in the buffer for an
544 * entire entry. Otherwise, only copy part of
545 * the pfn.
546 */
547 if (pm->out + PM_ENTRY_BYTES >= pm->end) {
548 if (copy_to_user(pm->out, &pfn, pm->end - pm->out))
549 return -EFAULT;
550 pm->out = pm->end;
551 return PM_END_OF_BUFFER;
552 }
553
554 if (put_user(pfn, pm->out))
555 return -EFAULT;
556 pm->out += PM_ENTRY_BYTES;
557 return 0;
558}
559
560static int pagemap_pte_hole(unsigned long start, unsigned long end,
561 void *private)
562{
563 struct pagemapread *pm = private;
564 unsigned long addr;
565 int err = 0;
566 for (addr = start; addr < end; addr += PAGE_SIZE) {
567 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
568 if (err)
569 break;
570 }
571 return err;
572}
573
574u64 swap_pte_to_pagemap_entry(pte_t pte)
575{
576 swp_entry_t e = pte_to_swp_entry(pte);
577 return PM_SWAP | swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
578}
579
580static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
581 void *private)
582{
583 struct pagemapread *pm = private;
584 pte_t *pte;
585 int err = 0;
586
587 for (; addr != end; addr += PAGE_SIZE) {
588 u64 pfn = PM_NOT_PRESENT;
589 pte = pte_offset_map(pmd, addr);
590 if (is_swap_pte(*pte))
591 pfn = swap_pte_to_pagemap_entry(*pte);
592 else if (pte_present(*pte))
593 pfn = pte_pfn(*pte);
594 /* unmap so we're not in atomic when we copy to userspace */
595 pte_unmap(pte);
596 err = add_to_pagemap(addr, pfn, pm);
597 if (err)
598 return err;
599 }
600
601 cond_resched();
602
603 return err;
604}
605
606static struct mm_walk pagemap_walk = {
607 .pmd_entry = pagemap_pte_range,
608 .pte_hole = pagemap_pte_hole
609};
610
611/*
612 * /proc/pid/pagemap - an array mapping virtual pages to pfns
613 *
614 * For each page in the address space, this file contains one 64-bit
615 * entry representing the corresponding physical page frame number
616 * (PFN) if the page is present. If there is a swap entry for the
617 * physical page, then an encoding of the swap file number and the
618 * page's offset into the swap file are returned. If no page is
619 * present at all, PM_NOT_PRESENT is returned. This allows determining
620 * precisely which pages are mapped (or in swap) and comparing mapped
621 * pages between processes.
622 *
623 * Efficient users of this interface will use /proc/pid/maps to
624 * determine which areas of memory are actually mapped and llseek to
625 * skip over unmapped regions.
626 */
627static ssize_t pagemap_read(struct file *file, char __user *buf,
628 size_t count, loff_t *ppos)
629{
630 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
631 struct page **pages, *page;
632 unsigned long uaddr, uend;
633 struct mm_struct *mm;
634 struct pagemapread pm;
635 int pagecount;
636 int ret = -ESRCH;
637
638 if (!task)
639 goto out;
640
641 ret = -EACCES;
642 if (!ptrace_may_attach(task))
643 goto out;
644
645 ret = -EINVAL;
646 /* file position must be aligned */
647 if (*ppos % PM_ENTRY_BYTES)
648 goto out;
649
650 ret = 0;
651 mm = get_task_mm(task);
652 if (!mm)
653 goto out;
654
655 ret = -ENOMEM;
656 uaddr = (unsigned long)buf & PAGE_MASK;
657 uend = (unsigned long)(buf + count);
658 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
659 pages = kmalloc(pagecount * sizeof(struct page *), GFP_KERNEL);
660 if (!pages)
661 goto out_task;
662
663 down_read(&current->mm->mmap_sem);
664 ret = get_user_pages(current, current->mm, uaddr, pagecount,
665 1, 0, pages, NULL);
666 up_read(&current->mm->mmap_sem);
667
668 if (ret < 0)
669 goto out_free;
670
671 pm.out = buf;
672 pm.end = buf + count;
673
674 if (!ptrace_may_attach(task)) {
675 ret = -EIO;
676 } else {
677 unsigned long src = *ppos;
678 unsigned long svpfn = src / PM_ENTRY_BYTES;
679 unsigned long start_vaddr = svpfn << PAGE_SHIFT;
680 unsigned long end_vaddr = TASK_SIZE_OF(task);
681
682 /* watch out for wraparound */
683 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
684 start_vaddr = end_vaddr;
685
686 /*
687 * The odds are that this will stop walking way
688 * before end_vaddr, because the length of the
689 * user buffer is tracked in "pm", and the walk
690 * will stop when we hit the end of the buffer.
691 */
692 ret = walk_page_range(mm, start_vaddr, end_vaddr,
693 &pagemap_walk, &pm);
694 if (ret == PM_END_OF_BUFFER)
695 ret = 0;
696 /* don't need mmap_sem for these, but this looks cleaner */
697 *ppos += pm.out - buf;
698 if (!ret)
699 ret = pm.out - buf;
700 }
701
702 for (; pagecount; pagecount--) {
703 page = pages[pagecount-1];
704 if (!PageReserved(page))
705 SetPageDirty(page);
706 page_cache_release(page);
707 }
708 mmput(mm);
709out_free:
710 kfree(pages);
711out_task:
712 put_task_struct(task);
713out:
714 return ret;
715}
716
717const struct file_operations proc_pagemap_operations = {
718 .llseek = mem_lseek, /* borrow this */
719 .read = pagemap_read,
720};
Matt Mackall1e883282008-02-04 22:29:07 -0800721#endif /* CONFIG_PROC_PAGE_MONITOR */
Matt Mackall85863e42008-02-04 22:29:04 -0800722
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700723#ifdef CONFIG_NUMA
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800724extern int show_numa_map(struct seq_file *m, void *v);
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700725
Kees Cook5096add2007-05-08 00:26:04 -0700726static int show_numa_map_checked(struct seq_file *m, void *v)
727{
728 struct proc_maps_private *priv = m->private;
729 struct task_struct *task = priv->task;
730
731 if (maps_protect && !ptrace_may_attach(task))
732 return -EACCES;
733
734 return show_numa_map(m, v);
735}
736
Eric W. Biederman662795d2006-06-26 00:25:48 -0700737static struct seq_operations proc_pid_numa_maps_op = {
Christoph Lameter1a75a6c2006-01-08 01:01:02 -0800738 .start = m_start,
739 .next = m_next,
740 .stop = m_stop,
Kees Cook5096add2007-05-08 00:26:04 -0700741 .show = show_numa_map_checked
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700742};
Eric W. Biederman662795d2006-06-26 00:25:48 -0700743
744static int numa_maps_open(struct inode *inode, struct file *file)
745{
746 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
747}
748
Arjan van de Ven00977a52007-02-12 00:55:34 -0800749const struct file_operations proc_numa_maps_operations = {
Eric W. Biederman662795d2006-06-26 00:25:48 -0700750 .open = numa_maps_open,
751 .read = seq_read,
752 .llseek = seq_lseek,
Eric W. Biederman99f89552006-06-26 00:25:55 -0700753 .release = seq_release_private,
Eric W. Biederman662795d2006-06-26 00:25:48 -0700754};
Christoph Lameter6e21c8f2005-09-03 15:54:45 -0700755#endif