2 #include <linux/hugetlb.h>
3 #include <linux/mount.h>
4 #include <linux/seq_file.h>
5 #include <linux/highmem.h>
6 #include <linux/ptrace.h>
7 #include <linux/pagemap.h>
8 #include <linux/mempolicy.h>
9 #include <linux/swap.h>
10 #include <linux/swapops.h>
13 #include <asm/uaccess.h>
14 #include <asm/tlbflush.h>
17 void task_mem(struct seq_file *m, struct mm_struct *mm)
19 unsigned long data, text, lib;
20 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
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.
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;
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;
50 hiwater_vm << (PAGE_SHIFT-10),
51 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
52 mm->locked_vm << (PAGE_SHIFT-10),
53 hiwater_rss << (PAGE_SHIFT-10),
54 total_rss << (PAGE_SHIFT-10),
55 data << (PAGE_SHIFT-10),
56 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
57 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
60 unsigned long task_vsize(struct mm_struct *mm)
62 return PAGE_SIZE * mm->total_vm;
65 int task_statm(struct mm_struct *mm, int *shared, int *text,
66 int *data, int *resident)
68 *shared = get_mm_counter(mm, file_rss);
69 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
71 *data = mm->total_vm - mm->shared_vm;
72 *resident = *shared + get_mm_counter(mm, anon_rss);
76 static void pad_len_spaces(struct seq_file *m, int len)
78 len = 25 + sizeof(void*) * 6 - len;
81 seq_printf(m, "%*c", len, ' ');
84 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
86 if (vma && vma != priv->tail_vma) {
87 struct mm_struct *mm = vma->vm_mm;
88 up_read(&mm->mmap_sem);
93 static void *m_start(struct seq_file *m, loff_t *pos)
95 struct proc_maps_private *priv = m->private;
96 unsigned long last_addr = m->version;
98 struct vm_area_struct *vma, *tail_vma = NULL;
101 /* Clear the per syscall fields in priv */
103 priv->tail_vma = NULL;
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.
112 if (last_addr == -1UL)
115 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
119 mm = mm_for_maps(priv->task);
123 tail_vma = get_gate_vma(priv->task);
124 priv->tail_vma = tail_vma;
126 /* Start with last addr hint */
127 vma = find_vma(mm, last_addr);
128 if (last_addr && vma) {
134 * Check the vma index is within the range and do
135 * sequential scan until m_index.
138 if ((unsigned long)l < mm->map_count) {
145 if (l != mm->map_count)
146 tail_vma = NULL; /* After gate vma */
152 /* End of vmas has been reached */
153 m->version = (tail_vma != NULL)? 0: -1UL;
154 up_read(&mm->mmap_sem);
159 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
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;
166 if (vma && (vma != tail_vma) && vma->vm_next)
169 return (vma != tail_vma)? tail_vma: NULL;
172 static void m_stop(struct seq_file *m, void *v)
174 struct proc_maps_private *priv = m->private;
175 struct vm_area_struct *vma = v;
179 put_task_struct(priv->task);
182 static int do_maps_open(struct inode *inode, struct file *file,
183 const struct seq_operations *ops)
185 struct proc_maps_private *priv;
187 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
189 priv->pid = proc_pid(inode);
190 ret = seq_open(file, ops);
192 struct seq_file *m = file->private_data;
201 static int show_map(struct seq_file *m, void *v)
203 struct proc_maps_private *priv = m->private;
204 struct task_struct *task = priv->task;
205 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;
209 unsigned long ino = 0;
214 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
215 dev = inode->i_sb->s_dev;
219 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
222 flags & VM_READ ? 'r' : '-',
223 flags & VM_WRITE ? 'w' : '-',
224 flags & VM_EXEC ? 'x' : '-',
225 flags & VM_MAYSHARE ? 's' : 'p',
226 ((loff_t)vma->vm_pgoff) << PAGE_SHIFT,
227 MAJOR(dev), MINOR(dev), ino, &len);
230 * Print the dentry name for named mappings, and a
231 * special [heap] marker for the heap:
234 pad_len_spaces(m, len);
235 seq_path(m, &file->f_path, "\n");
237 const char *name = arch_vma_name(vma);
240 if (vma->vm_start <= mm->start_brk &&
241 vma->vm_end >= mm->brk) {
243 } else if (vma->vm_start <= mm->start_stack &&
244 vma->vm_end >= mm->start_stack) {
252 pad_len_spaces(m, len);
258 if (m->count < m->size) /* vma is copied successfully */
259 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
263 static const struct seq_operations proc_pid_maps_op = {
270 static int maps_open(struct inode *inode, struct file *file)
272 return do_maps_open(inode, file, &proc_pid_maps_op);
275 const struct file_operations proc_maps_operations = {
279 .release = seq_release_private,
283 * Proportional Set Size(PSS): my share of RSS.
285 * PSS of a process is the count of pages it has in memory, where each
286 * page is divided by the number of processes sharing it. So if a
287 * process has 1000 pages all to itself, and 1000 shared with one other
288 * process, its PSS will be 1500.
290 * To keep (accumulated) division errors low, we adopt a 64bit
291 * fixed-point pss counter to minimize division errors. So (pss >>
292 * PSS_SHIFT) would be the real byte count.
294 * A shift of 12 before division means (assuming 4K page size):
295 * - 1M 3-user-pages add up to 8KB errors;
296 * - supports mapcount up to 2^24, or 16M;
297 * - supports PSS up to 2^52 bytes, or 4PB.
301 #ifdef CONFIG_PROC_PAGE_MONITOR
302 struct mem_size_stats {
303 struct vm_area_struct *vma;
304 unsigned long resident;
305 unsigned long shared_clean;
306 unsigned long shared_dirty;
307 unsigned long private_clean;
308 unsigned long private_dirty;
309 unsigned long referenced;
314 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
315 struct mm_walk *walk)
317 struct mem_size_stats *mss = walk->private;
318 struct vm_area_struct *vma = mss->vma;
324 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
325 for (; addr != end; pte++, addr += PAGE_SIZE) {
328 if (is_swap_pte(ptent)) {
329 mss->swap += PAGE_SIZE;
333 if (!pte_present(ptent))
336 mss->resident += PAGE_SIZE;
338 page = vm_normal_page(vma, addr, ptent);
342 /* Accumulate the size in pages that have been accessed. */
343 if (pte_young(ptent) || PageReferenced(page))
344 mss->referenced += PAGE_SIZE;
345 mapcount = page_mapcount(page);
347 if (pte_dirty(ptent))
348 mss->shared_dirty += PAGE_SIZE;
350 mss->shared_clean += PAGE_SIZE;
351 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
353 if (pte_dirty(ptent))
354 mss->private_dirty += PAGE_SIZE;
356 mss->private_clean += PAGE_SIZE;
357 mss->pss += (PAGE_SIZE << PSS_SHIFT);
360 pte_unmap_unlock(pte - 1, ptl);
365 static int show_smap(struct seq_file *m, void *v)
367 struct vm_area_struct *vma = v;
368 struct mem_size_stats mss;
370 struct mm_walk smaps_walk = {
371 .pmd_entry = smaps_pte_range,
376 memset(&mss, 0, sizeof mss);
378 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
379 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
381 ret = show_map(m, v);
389 "Shared_Clean: %8lu kB\n"
390 "Shared_Dirty: %8lu kB\n"
391 "Private_Clean: %8lu kB\n"
392 "Private_Dirty: %8lu kB\n"
393 "Referenced: %8lu kB\n"
395 (vma->vm_end - vma->vm_start) >> 10,
397 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
398 mss.shared_clean >> 10,
399 mss.shared_dirty >> 10,
400 mss.private_clean >> 10,
401 mss.private_dirty >> 10,
402 mss.referenced >> 10,
408 static const struct seq_operations proc_pid_smaps_op = {
415 static int smaps_open(struct inode *inode, struct file *file)
417 return do_maps_open(inode, file, &proc_pid_smaps_op);
420 const struct file_operations proc_smaps_operations = {
424 .release = seq_release_private,
427 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
428 unsigned long end, struct mm_walk *walk)
430 struct vm_area_struct *vma = walk->private;
435 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
436 for (; addr != end; pte++, addr += PAGE_SIZE) {
438 if (!pte_present(ptent))
441 page = vm_normal_page(vma, addr, ptent);
445 /* Clear accessed and referenced bits. */
446 ptep_test_and_clear_young(vma, addr, pte);
447 ClearPageReferenced(page);
449 pte_unmap_unlock(pte - 1, ptl);
454 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
455 size_t count, loff_t *ppos)
457 struct task_struct *task;
458 char buffer[PROC_NUMBUF], *end;
459 struct mm_struct *mm;
460 struct vm_area_struct *vma;
462 memset(buffer, 0, sizeof(buffer));
463 if (count > sizeof(buffer) - 1)
464 count = sizeof(buffer) - 1;
465 if (copy_from_user(buffer, buf, count))
467 if (!simple_strtol(buffer, &end, 0))
471 task = get_proc_task(file->f_path.dentry->d_inode);
474 mm = get_task_mm(task);
476 struct mm_walk clear_refs_walk = {
477 .pmd_entry = clear_refs_pte_range,
480 down_read(&mm->mmap_sem);
481 for (vma = mm->mmap; vma; vma = vma->vm_next) {
482 clear_refs_walk.private = vma;
483 if (!is_vm_hugetlb_page(vma))
484 walk_page_range(vma->vm_start, vma->vm_end,
488 up_read(&mm->mmap_sem);
491 put_task_struct(task);
492 if (end - buffer == 0)
497 const struct file_operations proc_clear_refs_operations = {
498 .write = clear_refs_write,
502 u64 __user *out, *end;
505 #define PM_ENTRY_BYTES sizeof(u64)
506 #define PM_STATUS_BITS 3
507 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
508 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
509 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
510 #define PM_PSHIFT_BITS 6
511 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
512 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
513 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
514 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
515 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
517 #define PM_PRESENT PM_STATUS(4LL)
518 #define PM_SWAP PM_STATUS(2LL)
519 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
520 #define PM_END_OF_BUFFER 1
522 static int add_to_pagemap(unsigned long addr, u64 pfn,
523 struct pagemapread *pm)
525 if (put_user(pfn, pm->out))
528 if (pm->out >= pm->end)
529 return PM_END_OF_BUFFER;
533 static int pagemap_pte_hole(unsigned long start, unsigned long end,
534 struct mm_walk *walk)
536 struct pagemapread *pm = walk->private;
539 for (addr = start; addr < end; addr += PAGE_SIZE) {
540 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
547 static u64 swap_pte_to_pagemap_entry(pte_t pte)
549 swp_entry_t e = pte_to_swp_entry(pte);
550 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
553 static unsigned long pte_to_pagemap_entry(pte_t pte)
555 unsigned long pme = 0;
556 if (is_swap_pte(pte))
557 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
558 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
559 else if (pte_present(pte))
560 pme = PM_PFRAME(pte_pfn(pte))
561 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
565 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
566 struct mm_walk *walk)
568 struct vm_area_struct *vma;
569 struct pagemapread *pm = walk->private;
573 /* find the first VMA at or above 'addr' */
574 vma = find_vma(walk->mm, addr);
575 for (; addr != end; addr += PAGE_SIZE) {
576 u64 pfn = PM_NOT_PRESENT;
578 /* check to see if we've left 'vma' behind
579 * and need a new, higher one */
580 if (vma && (addr >= vma->vm_end))
581 vma = find_vma(walk->mm, addr);
583 /* check that 'vma' actually covers this address,
584 * and that it isn't a huge page vma */
585 if (vma && (vma->vm_start <= addr) &&
586 !is_vm_hugetlb_page(vma)) {
587 pte = pte_offset_map(pmd, addr);
588 pfn = pte_to_pagemap_entry(*pte);
589 /* unmap before userspace copy */
592 err = add_to_pagemap(addr, pfn, pm);
603 * /proc/pid/pagemap - an array mapping virtual pages to pfns
605 * For each page in the address space, this file contains one 64-bit entry
606 * consisting of the following:
608 * Bits 0-55 page frame number (PFN) if present
609 * Bits 0-4 swap type if swapped
610 * Bits 5-55 swap offset if swapped
611 * Bits 55-60 page shift (page size = 1<<page shift)
612 * Bit 61 reserved for future use
613 * Bit 62 page swapped
614 * Bit 63 page present
616 * If the page is not present but in swap, then the PFN contains an
617 * encoding of the swap file number and the page's offset into the
618 * swap. Unmapped pages return a null PFN. This allows determining
619 * precisely which pages are mapped (or in swap) and comparing mapped
620 * pages between processes.
622 * Efficient users of this interface will use /proc/pid/maps to
623 * determine which areas of memory are actually mapped and llseek to
624 * skip over unmapped regions.
626 static ssize_t pagemap_read(struct file *file, char __user *buf,
627 size_t count, loff_t *ppos)
629 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
630 struct page **pages, *page;
631 unsigned long uaddr, uend;
632 struct mm_struct *mm;
633 struct pagemapread pm;
636 struct mm_walk pagemap_walk = {};
639 unsigned long start_vaddr;
640 unsigned long end_vaddr;
646 if (!ptrace_may_access(task, PTRACE_MODE_READ))
650 /* file position must be aligned */
651 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
655 mm = get_task_mm(task);
660 uaddr = (unsigned long)buf & PAGE_MASK;
661 uend = (unsigned long)(buf + count);
662 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
666 pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
671 down_read(¤t->mm->mmap_sem);
672 ret = get_user_pages(current, current->mm, uaddr, pagecount,
674 up_read(¤t->mm->mmap_sem);
679 if (ret != pagecount) {
686 pm.end = (u64 *)(buf + count);
688 pagemap_walk.pmd_entry = pagemap_pte_range;
689 pagemap_walk.pte_hole = pagemap_pte_hole;
690 pagemap_walk.mm = mm;
691 pagemap_walk.private = ±
694 svpfn = src / PM_ENTRY_BYTES;
695 start_vaddr = svpfn << PAGE_SHIFT;
696 end_vaddr = TASK_SIZE_OF(task);
698 /* watch out for wraparound */
699 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
700 start_vaddr = end_vaddr;
703 * The odds are that this will stop walking way
704 * before end_vaddr, because the length of the
705 * user buffer is tracked in "pm", and the walk
706 * will stop when we hit the end of the buffer.
708 ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
709 if (ret == PM_END_OF_BUFFER)
711 /* don't need mmap_sem for these, but this looks cleaner */
712 *ppos += (char *)pm.out - buf;
714 ret = (char *)pm.out - buf;
717 for (; pagecount; pagecount--) {
718 page = pages[pagecount-1];
719 if (!PageReserved(page))
721 page_cache_release(page);
728 put_task_struct(task);
733 const struct file_operations proc_pagemap_operations = {
734 .llseek = mem_lseek, /* borrow this */
735 .read = pagemap_read,
737 #endif /* CONFIG_PROC_PAGE_MONITOR */
740 extern int show_numa_map(struct seq_file *m, void *v);
742 static const struct seq_operations proc_pid_numa_maps_op = {
746 .show = show_numa_map,
749 static int numa_maps_open(struct inode *inode, struct file *file)
751 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
754 const struct file_operations proc_numa_maps_operations = {
755 .open = numa_maps_open,
758 .release = seq_release_private,