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/slab.h>
8 #include <linux/pagemap.h>
9 #include <linux/mempolicy.h>
10 #include <linux/swap.h>
11 #include <linux/swapops.h>
14 #include <asm/uaccess.h>
15 #include <asm/tlbflush.h>
18 void task_mem(struct seq_file *m, struct mm_struct *mm)
20 unsigned long data, text, lib, swap;
21 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
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.
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;
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 swap = get_mm_counter(mm, MM_SWAPENTS);
53 hiwater_vm << (PAGE_SHIFT-10),
54 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
55 mm->locked_vm << (PAGE_SHIFT-10),
56 hiwater_rss << (PAGE_SHIFT-10),
57 total_rss << (PAGE_SHIFT-10),
58 data << (PAGE_SHIFT-10),
59 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
60 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10,
61 swap << (PAGE_SHIFT-10));
64 unsigned long task_vsize(struct mm_struct *mm)
66 return PAGE_SIZE * mm->total_vm;
69 unsigned long task_statm(struct mm_struct *mm,
70 unsigned long *shared, unsigned long *text,
71 unsigned long *data, unsigned long *resident)
73 *shared = get_mm_counter(mm, MM_FILEPAGES);
74 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
76 *data = mm->total_vm - mm->shared_vm;
77 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
81 static void pad_len_spaces(struct seq_file *m, int len)
83 len = 25 + sizeof(void*) * 6 - len;
86 seq_printf(m, "%*c", len, ' ');
89 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
91 if (vma && vma != priv->tail_vma) {
92 struct mm_struct *mm = vma->vm_mm;
93 up_read(&mm->mmap_sem);
98 static void *m_start(struct seq_file *m, loff_t *pos)
100 struct proc_maps_private *priv = m->private;
101 unsigned long last_addr = m->version;
102 struct mm_struct *mm;
103 struct vm_area_struct *vma, *tail_vma = NULL;
106 /* Clear the per syscall fields in priv */
108 priv->tail_vma = NULL;
111 * We remember last_addr rather than next_addr to hit with
112 * mmap_cache most of the time. We have zero last_addr at
113 * the beginning and also after lseek. We will have -1 last_addr
114 * after the end of the vmas.
117 if (last_addr == -1UL)
120 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
124 mm = mm_for_maps(priv->task);
127 down_read(&mm->mmap_sem);
129 tail_vma = get_gate_vma(priv->task);
130 priv->tail_vma = tail_vma;
132 /* Start with last addr hint */
133 vma = find_vma(mm, last_addr);
134 if (last_addr && vma) {
140 * Check the vma index is within the range and do
141 * sequential scan until m_index.
144 if ((unsigned long)l < mm->map_count) {
151 if (l != mm->map_count)
152 tail_vma = NULL; /* After gate vma */
158 /* End of vmas has been reached */
159 m->version = (tail_vma != NULL)? 0: -1UL;
160 up_read(&mm->mmap_sem);
165 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
167 struct proc_maps_private *priv = m->private;
168 struct vm_area_struct *vma = v;
169 struct vm_area_struct *tail_vma = priv->tail_vma;
172 if (vma && (vma != tail_vma) && vma->vm_next)
175 return (vma != tail_vma)? tail_vma: NULL;
178 static void m_stop(struct seq_file *m, void *v)
180 struct proc_maps_private *priv = m->private;
181 struct vm_area_struct *vma = v;
185 put_task_struct(priv->task);
188 static int do_maps_open(struct inode *inode, struct file *file,
189 const struct seq_operations *ops)
191 struct proc_maps_private *priv;
193 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
195 priv->pid = proc_pid(inode);
196 ret = seq_open(file, ops);
198 struct seq_file *m = file->private_data;
207 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
209 struct mm_struct *mm = vma->vm_mm;
210 struct file *file = vma->vm_file;
211 int flags = vma->vm_flags;
212 unsigned long ino = 0;
213 unsigned long long pgoff = 0;
219 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
220 dev = inode->i_sb->s_dev;
222 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
225 /* We don't show the stack guard page in /proc/maps */
226 start = vma->vm_start;
227 if (vma->vm_flags & VM_GROWSDOWN)
228 if (!vma_stack_continue(vma->vm_prev, vma->vm_start))
231 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n",
234 flags & VM_READ ? 'r' : '-',
235 flags & VM_WRITE ? 'w' : '-',
236 flags & VM_EXEC ? 'x' : '-',
237 flags & VM_MAYSHARE ? 's' : 'p',
239 MAJOR(dev), MINOR(dev), ino, &len);
242 * Print the dentry name for named mappings, and a
243 * special [heap] marker for the heap:
246 pad_len_spaces(m, len);
247 seq_path(m, &file->f_path, "\n");
249 const char *name = arch_vma_name(vma);
252 if (vma->vm_start <= mm->start_brk &&
253 vma->vm_end >= mm->brk) {
255 } else if (vma->vm_start <= mm->start_stack &&
256 vma->vm_end >= mm->start_stack) {
264 pad_len_spaces(m, len);
271 static int show_map(struct seq_file *m, void *v)
273 struct vm_area_struct *vma = v;
274 struct proc_maps_private *priv = m->private;
275 struct task_struct *task = priv->task;
277 show_map_vma(m, vma);
279 if (m->count < m->size) /* vma is copied successfully */
280 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
284 static const struct seq_operations proc_pid_maps_op = {
291 static int maps_open(struct inode *inode, struct file *file)
293 return do_maps_open(inode, file, &proc_pid_maps_op);
296 const struct file_operations proc_maps_operations = {
300 .release = seq_release_private,
304 * Proportional Set Size(PSS): my share of RSS.
306 * PSS of a process is the count of pages it has in memory, where each
307 * page is divided by the number of processes sharing it. So if a
308 * process has 1000 pages all to itself, and 1000 shared with one other
309 * process, its PSS will be 1500.
311 * To keep (accumulated) division errors low, we adopt a 64bit
312 * fixed-point pss counter to minimize division errors. So (pss >>
313 * PSS_SHIFT) would be the real byte count.
315 * A shift of 12 before division means (assuming 4K page size):
316 * - 1M 3-user-pages add up to 8KB errors;
317 * - supports mapcount up to 2^24, or 16M;
318 * - supports PSS up to 2^52 bytes, or 4PB.
322 #ifdef CONFIG_PROC_PAGE_MONITOR
323 struct mem_size_stats {
324 struct vm_area_struct *vma;
325 unsigned long resident;
326 unsigned long shared_clean;
327 unsigned long shared_dirty;
328 unsigned long private_clean;
329 unsigned long private_dirty;
330 unsigned long referenced;
331 unsigned long anonymous;
336 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
337 struct mm_walk *walk)
339 struct mem_size_stats *mss = walk->private;
340 struct vm_area_struct *vma = mss->vma;
346 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
347 for (; addr != end; pte++, addr += PAGE_SIZE) {
350 if (is_swap_pte(ptent)) {
351 mss->swap += PAGE_SIZE;
355 if (!pte_present(ptent))
358 page = vm_normal_page(vma, addr, ptent);
363 mss->anonymous += PAGE_SIZE;
365 mss->resident += PAGE_SIZE;
366 /* Accumulate the size in pages that have been accessed. */
367 if (pte_young(ptent) || PageReferenced(page))
368 mss->referenced += PAGE_SIZE;
369 mapcount = page_mapcount(page);
371 if (pte_dirty(ptent) || PageDirty(page))
372 mss->shared_dirty += PAGE_SIZE;
374 mss->shared_clean += PAGE_SIZE;
375 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
377 if (pte_dirty(ptent) || PageDirty(page))
378 mss->private_dirty += PAGE_SIZE;
380 mss->private_clean += PAGE_SIZE;
381 mss->pss += (PAGE_SIZE << PSS_SHIFT);
384 pte_unmap_unlock(pte - 1, ptl);
389 static int show_smap(struct seq_file *m, void *v)
391 struct proc_maps_private *priv = m->private;
392 struct task_struct *task = priv->task;
393 struct vm_area_struct *vma = v;
394 struct mem_size_stats mss;
395 struct mm_walk smaps_walk = {
396 .pmd_entry = smaps_pte_range,
401 memset(&mss, 0, sizeof mss);
403 /* mmap_sem is held in m_start */
404 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
405 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
407 show_map_vma(m, vma);
413 "Shared_Clean: %8lu kB\n"
414 "Shared_Dirty: %8lu kB\n"
415 "Private_Clean: %8lu kB\n"
416 "Private_Dirty: %8lu kB\n"
417 "Referenced: %8lu kB\n"
418 "Anonymous: %8lu kB\n"
420 "KernelPageSize: %8lu kB\n"
421 "MMUPageSize: %8lu kB\n",
422 (vma->vm_end - vma->vm_start) >> 10,
424 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
425 mss.shared_clean >> 10,
426 mss.shared_dirty >> 10,
427 mss.private_clean >> 10,
428 mss.private_dirty >> 10,
429 mss.referenced >> 10,
432 vma_kernel_pagesize(vma) >> 10,
433 vma_mmu_pagesize(vma) >> 10);
435 if (m->count < m->size) /* vma is copied successfully */
436 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
440 static const struct seq_operations proc_pid_smaps_op = {
447 static int smaps_open(struct inode *inode, struct file *file)
449 return do_maps_open(inode, file, &proc_pid_smaps_op);
452 const struct file_operations proc_smaps_operations = {
456 .release = seq_release_private,
459 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
460 unsigned long end, struct mm_walk *walk)
462 struct vm_area_struct *vma = walk->private;
467 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
468 for (; addr != end; pte++, addr += PAGE_SIZE) {
470 if (!pte_present(ptent))
473 page = vm_normal_page(vma, addr, ptent);
477 /* Clear accessed and referenced bits. */
478 ptep_test_and_clear_young(vma, addr, pte);
479 ClearPageReferenced(page);
481 pte_unmap_unlock(pte - 1, ptl);
486 #define CLEAR_REFS_ALL 1
487 #define CLEAR_REFS_ANON 2
488 #define CLEAR_REFS_MAPPED 3
490 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
491 size_t count, loff_t *ppos)
493 struct task_struct *task;
494 char buffer[PROC_NUMBUF];
495 struct mm_struct *mm;
496 struct vm_area_struct *vma;
499 memset(buffer, 0, sizeof(buffer));
500 if (count > sizeof(buffer) - 1)
501 count = sizeof(buffer) - 1;
502 if (copy_from_user(buffer, buf, count))
504 if (strict_strtol(strstrip(buffer), 10, &type))
506 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
508 task = get_proc_task(file->f_path.dentry->d_inode);
511 mm = get_task_mm(task);
513 struct mm_walk clear_refs_walk = {
514 .pmd_entry = clear_refs_pte_range,
517 down_read(&mm->mmap_sem);
518 for (vma = mm->mmap; vma; vma = vma->vm_next) {
519 clear_refs_walk.private = vma;
520 if (is_vm_hugetlb_page(vma))
523 * Writing 1 to /proc/pid/clear_refs affects all pages.
525 * Writing 2 to /proc/pid/clear_refs only affects
528 * Writing 3 to /proc/pid/clear_refs only affects file
531 if (type == CLEAR_REFS_ANON && vma->vm_file)
533 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
535 walk_page_range(vma->vm_start, vma->vm_end,
539 up_read(&mm->mmap_sem);
542 put_task_struct(task);
547 const struct file_operations proc_clear_refs_operations = {
548 .write = clear_refs_write,
549 .llseek = noop_llseek,
557 #define PM_ENTRY_BYTES sizeof(u64)
558 #define PM_STATUS_BITS 3
559 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
560 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
561 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
562 #define PM_PSHIFT_BITS 6
563 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
564 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
565 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
566 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
567 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
569 #define PM_PRESENT PM_STATUS(4LL)
570 #define PM_SWAP PM_STATUS(2LL)
571 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
572 #define PM_END_OF_BUFFER 1
574 static int add_to_pagemap(unsigned long addr, u64 pfn,
575 struct pagemapread *pm)
577 pm->buffer[pm->pos++] = pfn;
578 if (pm->pos >= pm->len)
579 return PM_END_OF_BUFFER;
583 static int pagemap_pte_hole(unsigned long start, unsigned long end,
584 struct mm_walk *walk)
586 struct pagemapread *pm = walk->private;
589 for (addr = start; addr < end; addr += PAGE_SIZE) {
590 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
597 static u64 swap_pte_to_pagemap_entry(pte_t pte)
599 swp_entry_t e = pte_to_swp_entry(pte);
600 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
603 static u64 pte_to_pagemap_entry(pte_t pte)
606 if (is_swap_pte(pte))
607 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
608 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
609 else if (pte_present(pte))
610 pme = PM_PFRAME(pte_pfn(pte))
611 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
615 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
616 struct mm_walk *walk)
618 struct vm_area_struct *vma;
619 struct pagemapread *pm = walk->private;
623 /* find the first VMA at or above 'addr' */
624 vma = find_vma(walk->mm, addr);
625 for (; addr != end; addr += PAGE_SIZE) {
626 u64 pfn = PM_NOT_PRESENT;
628 /* check to see if we've left 'vma' behind
629 * and need a new, higher one */
630 if (vma && (addr >= vma->vm_end))
631 vma = find_vma(walk->mm, addr);
633 /* check that 'vma' actually covers this address,
634 * and that it isn't a huge page vma */
635 if (vma && (vma->vm_start <= addr) &&
636 !is_vm_hugetlb_page(vma)) {
637 pte = pte_offset_map(pmd, addr);
638 pfn = pte_to_pagemap_entry(*pte);
639 /* unmap before userspace copy */
642 err = add_to_pagemap(addr, pfn, pm);
652 #ifdef CONFIG_HUGETLB_PAGE
653 static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
656 if (pte_present(pte))
657 pme = PM_PFRAME(pte_pfn(pte) + offset)
658 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
662 /* This function walks within one hugetlb entry in the single call */
663 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask,
664 unsigned long addr, unsigned long end,
665 struct mm_walk *walk)
667 struct pagemapread *pm = walk->private;
671 for (; addr != end; addr += PAGE_SIZE) {
672 int offset = (addr & ~hmask) >> PAGE_SHIFT;
673 pfn = huge_pte_to_pagemap_entry(*pte, offset);
674 err = add_to_pagemap(addr, pfn, pm);
683 #endif /* HUGETLB_PAGE */
686 * /proc/pid/pagemap - an array mapping virtual pages to pfns
688 * For each page in the address space, this file contains one 64-bit entry
689 * consisting of the following:
691 * Bits 0-55 page frame number (PFN) if present
692 * Bits 0-4 swap type if swapped
693 * Bits 5-55 swap offset if swapped
694 * Bits 55-60 page shift (page size = 1<<page shift)
695 * Bit 61 reserved for future use
696 * Bit 62 page swapped
697 * Bit 63 page present
699 * If the page is not present but in swap, then the PFN contains an
700 * encoding of the swap file number and the page's offset into the
701 * swap. Unmapped pages return a null PFN. This allows determining
702 * precisely which pages are mapped (or in swap) and comparing mapped
703 * pages between processes.
705 * Efficient users of this interface will use /proc/pid/maps to
706 * determine which areas of memory are actually mapped and llseek to
707 * skip over unmapped regions.
709 #define PAGEMAP_WALK_SIZE (PMD_SIZE)
710 #define PAGEMAP_WALK_MASK (PMD_MASK)
711 static ssize_t pagemap_read(struct file *file, char __user *buf,
712 size_t count, loff_t *ppos)
714 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
715 struct mm_struct *mm;
716 struct pagemapread pm;
718 struct mm_walk pagemap_walk = {};
721 unsigned long start_vaddr;
722 unsigned long end_vaddr;
729 if (!ptrace_may_access(task, PTRACE_MODE_READ))
733 /* file position must be aligned */
734 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
742 mm = get_task_mm(task);
746 pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
747 pm.buffer = kmalloc(pm.len, GFP_TEMPORARY);
752 pagemap_walk.pmd_entry = pagemap_pte_range;
753 pagemap_walk.pte_hole = pagemap_pte_hole;
754 #ifdef CONFIG_HUGETLB_PAGE
755 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
757 pagemap_walk.mm = mm;
758 pagemap_walk.private = ±
761 svpfn = src / PM_ENTRY_BYTES;
762 start_vaddr = svpfn << PAGE_SHIFT;
763 end_vaddr = TASK_SIZE_OF(task);
765 /* watch out for wraparound */
766 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
767 start_vaddr = end_vaddr;
770 * The odds are that this will stop walking way
771 * before end_vaddr, because the length of the
772 * user buffer is tracked in "pm", and the walk
773 * will stop when we hit the end of the buffer.
776 while (count && (start_vaddr < end_vaddr)) {
781 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
783 if (end < start_vaddr || end > end_vaddr)
785 down_read(&mm->mmap_sem);
786 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
787 up_read(&mm->mmap_sem);
790 len = min(count, PM_ENTRY_BYTES * pm.pos);
791 if (copy_to_user(buf, pm.buffer, len)) {
800 if (!ret || ret == PM_END_OF_BUFFER)
808 put_task_struct(task);
813 const struct file_operations proc_pagemap_operations = {
814 .llseek = mem_lseek, /* borrow this */
815 .read = pagemap_read,
817 #endif /* CONFIG_PROC_PAGE_MONITOR */
820 extern int show_numa_map(struct seq_file *m, void *v);
822 static const struct seq_operations proc_pid_numa_maps_op = {
826 .show = show_numa_map,
829 static int numa_maps_open(struct inode *inode, struct file *file)
831 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
834 const struct file_operations proc_numa_maps_operations = {
835 .open = numa_maps_open,
838 .release = seq_release_private,