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);
122 down_read(&mm->mmap_sem);
124 tail_vma = get_gate_vma(priv->task);
125 priv->tail_vma = tail_vma;
127 /* Start with last addr hint */
128 vma = find_vma(mm, last_addr);
129 if (last_addr && vma) {
135 * Check the vma index is within the range and do
136 * sequential scan until m_index.
139 if ((unsigned long)l < mm->map_count) {
146 if (l != mm->map_count)
147 tail_vma = NULL; /* After gate vma */
153 /* End of vmas has been reached */
154 m->version = (tail_vma != NULL)? 0: -1UL;
155 up_read(&mm->mmap_sem);
160 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
162 struct proc_maps_private *priv = m->private;
163 struct vm_area_struct *vma = v;
164 struct vm_area_struct *tail_vma = priv->tail_vma;
167 if (vma && (vma != tail_vma) && vma->vm_next)
170 return (vma != tail_vma)? tail_vma: NULL;
173 static void m_stop(struct seq_file *m, void *v)
175 struct proc_maps_private *priv = m->private;
176 struct vm_area_struct *vma = v;
180 put_task_struct(priv->task);
183 static int do_maps_open(struct inode *inode, struct file *file,
184 const struct seq_operations *ops)
186 struct proc_maps_private *priv;
188 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
190 priv->pid = proc_pid(inode);
191 ret = seq_open(file, ops);
193 struct seq_file *m = file->private_data;
202 static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
204 struct mm_struct *mm = vma->vm_mm;
205 struct file *file = vma->vm_file;
206 int flags = vma->vm_flags;
207 unsigned long ino = 0;
208 unsigned long long pgoff = 0;
213 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
214 dev = inode->i_sb->s_dev;
216 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
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',
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) {
247 unsigned long stack_start;
248 struct proc_maps_private *pmp;
251 stack_start = pmp->task->stack_start;
253 if (vma->vm_start <= stack_start &&
254 vma->vm_end >= stack_start) {
255 pad_len_spaces(m, len);
257 "[threadstack:%08lx]",
258 #ifdef CONFIG_STACK_GROWSUP
259 vma->vm_end - stack_start
261 stack_start - vma->vm_start
271 pad_len_spaces(m, len);
278 static int show_map(struct seq_file *m, void *v)
280 struct vm_area_struct *vma = v;
281 struct proc_maps_private *priv = m->private;
282 struct task_struct *task = priv->task;
284 show_map_vma(m, vma);
286 if (m->count < m->size) /* vma is copied successfully */
287 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
291 static const struct seq_operations proc_pid_maps_op = {
298 static int maps_open(struct inode *inode, struct file *file)
300 return do_maps_open(inode, file, &proc_pid_maps_op);
303 const struct file_operations proc_maps_operations = {
307 .release = seq_release_private,
311 * Proportional Set Size(PSS): my share of RSS.
313 * PSS of a process is the count of pages it has in memory, where each
314 * page is divided by the number of processes sharing it. So if a
315 * process has 1000 pages all to itself, and 1000 shared with one other
316 * process, its PSS will be 1500.
318 * To keep (accumulated) division errors low, we adopt a 64bit
319 * fixed-point pss counter to minimize division errors. So (pss >>
320 * PSS_SHIFT) would be the real byte count.
322 * A shift of 12 before division means (assuming 4K page size):
323 * - 1M 3-user-pages add up to 8KB errors;
324 * - supports mapcount up to 2^24, or 16M;
325 * - supports PSS up to 2^52 bytes, or 4PB.
329 #ifdef CONFIG_PROC_PAGE_MONITOR
330 struct mem_size_stats {
331 struct vm_area_struct *vma;
332 unsigned long resident;
333 unsigned long shared_clean;
334 unsigned long shared_dirty;
335 unsigned long private_clean;
336 unsigned long private_dirty;
337 unsigned long referenced;
342 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
343 struct mm_walk *walk)
345 struct mem_size_stats *mss = walk->private;
346 struct vm_area_struct *vma = mss->vma;
352 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
353 for (; addr != end; pte++, addr += PAGE_SIZE) {
356 if (is_swap_pte(ptent)) {
357 mss->swap += PAGE_SIZE;
361 if (!pte_present(ptent))
364 mss->resident += PAGE_SIZE;
366 page = vm_normal_page(vma, addr, ptent);
370 /* Accumulate the size in pages that have been accessed. */
371 if (pte_young(ptent) || PageReferenced(page))
372 mss->referenced += PAGE_SIZE;
373 mapcount = page_mapcount(page);
375 if (pte_dirty(ptent))
376 mss->shared_dirty += PAGE_SIZE;
378 mss->shared_clean += PAGE_SIZE;
379 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
381 if (pte_dirty(ptent))
382 mss->private_dirty += PAGE_SIZE;
384 mss->private_clean += PAGE_SIZE;
385 mss->pss += (PAGE_SIZE << PSS_SHIFT);
388 pte_unmap_unlock(pte - 1, ptl);
393 static int show_smap(struct seq_file *m, void *v)
395 struct proc_maps_private *priv = m->private;
396 struct task_struct *task = priv->task;
397 struct vm_area_struct *vma = v;
398 struct mem_size_stats mss;
399 struct mm_walk smaps_walk = {
400 .pmd_entry = smaps_pte_range,
405 memset(&mss, 0, sizeof mss);
407 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
408 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk);
410 show_map_vma(m, vma);
416 "Shared_Clean: %8lu kB\n"
417 "Shared_Dirty: %8lu kB\n"
418 "Private_Clean: %8lu kB\n"
419 "Private_Dirty: %8lu kB\n"
420 "Referenced: %8lu kB\n"
422 "KernelPageSize: %8lu kB\n"
423 "MMUPageSize: %8lu kB\n",
424 (vma->vm_end - vma->vm_start) >> 10,
426 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
427 mss.shared_clean >> 10,
428 mss.shared_dirty >> 10,
429 mss.private_clean >> 10,
430 mss.private_dirty >> 10,
431 mss.referenced >> 10,
433 vma_kernel_pagesize(vma) >> 10,
434 vma_mmu_pagesize(vma) >> 10);
436 if (m->count < m->size) /* vma is copied successfully */
437 m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0;
441 static const struct seq_operations proc_pid_smaps_op = {
448 static int smaps_open(struct inode *inode, struct file *file)
450 return do_maps_open(inode, file, &proc_pid_smaps_op);
453 const struct file_operations proc_smaps_operations = {
457 .release = seq_release_private,
460 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
461 unsigned long end, struct mm_walk *walk)
463 struct vm_area_struct *vma = walk->private;
468 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
469 for (; addr != end; pte++, addr += PAGE_SIZE) {
471 if (!pte_present(ptent))
474 page = vm_normal_page(vma, addr, ptent);
478 /* Clear accessed and referenced bits. */
479 ptep_test_and_clear_young(vma, addr, pte);
480 ClearPageReferenced(page);
482 pte_unmap_unlock(pte - 1, ptl);
487 #define CLEAR_REFS_ALL 1
488 #define CLEAR_REFS_ANON 2
489 #define CLEAR_REFS_MAPPED 3
491 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
492 size_t count, loff_t *ppos)
494 struct task_struct *task;
495 char buffer[PROC_NUMBUF];
496 struct mm_struct *mm;
497 struct vm_area_struct *vma;
500 memset(buffer, 0, sizeof(buffer));
501 if (count > sizeof(buffer) - 1)
502 count = sizeof(buffer) - 1;
503 if (copy_from_user(buffer, buf, count))
505 if (strict_strtol(strstrip(buffer), 10, &type))
507 if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED)
509 task = get_proc_task(file->f_path.dentry->d_inode);
512 mm = get_task_mm(task);
514 struct mm_walk clear_refs_walk = {
515 .pmd_entry = clear_refs_pte_range,
518 down_read(&mm->mmap_sem);
519 for (vma = mm->mmap; vma; vma = vma->vm_next) {
520 clear_refs_walk.private = vma;
521 if (is_vm_hugetlb_page(vma))
524 * Writing 1 to /proc/pid/clear_refs affects all pages.
526 * Writing 2 to /proc/pid/clear_refs only affects
529 * Writing 3 to /proc/pid/clear_refs only affects file
532 if (type == CLEAR_REFS_ANON && vma->vm_file)
534 if (type == CLEAR_REFS_MAPPED && !vma->vm_file)
536 walk_page_range(vma->vm_start, vma->vm_end,
540 up_read(&mm->mmap_sem);
543 put_task_struct(task);
548 const struct file_operations proc_clear_refs_operations = {
549 .write = clear_refs_write,
553 u64 __user *out, *end;
556 #define PM_ENTRY_BYTES sizeof(u64)
557 #define PM_STATUS_BITS 3
558 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
559 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
560 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
561 #define PM_PSHIFT_BITS 6
562 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
563 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
564 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
565 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
566 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
568 #define PM_PRESENT PM_STATUS(4LL)
569 #define PM_SWAP PM_STATUS(2LL)
570 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
571 #define PM_END_OF_BUFFER 1
573 static int add_to_pagemap(unsigned long addr, u64 pfn,
574 struct pagemapread *pm)
576 if (put_user(pfn, pm->out))
579 if (pm->out >= pm->end)
580 return PM_END_OF_BUFFER;
584 static int pagemap_pte_hole(unsigned long start, unsigned long end,
585 struct mm_walk *walk)
587 struct pagemapread *pm = walk->private;
590 for (addr = start; addr < end; addr += PAGE_SIZE) {
591 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
598 static u64 swap_pte_to_pagemap_entry(pte_t pte)
600 swp_entry_t e = pte_to_swp_entry(pte);
601 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
604 static u64 pte_to_pagemap_entry(pte_t pte)
607 if (is_swap_pte(pte))
608 pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte))
609 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
610 else if (pte_present(pte))
611 pme = PM_PFRAME(pte_pfn(pte))
612 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
616 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
617 struct mm_walk *walk)
619 struct vm_area_struct *vma;
620 struct pagemapread *pm = walk->private;
624 /* find the first VMA at or above 'addr' */
625 vma = find_vma(walk->mm, addr);
626 for (; addr != end; addr += PAGE_SIZE) {
627 u64 pfn = PM_NOT_PRESENT;
629 /* check to see if we've left 'vma' behind
630 * and need a new, higher one */
631 if (vma && (addr >= vma->vm_end))
632 vma = find_vma(walk->mm, addr);
634 /* check that 'vma' actually covers this address,
635 * and that it isn't a huge page vma */
636 if (vma && (vma->vm_start <= addr) &&
637 !is_vm_hugetlb_page(vma)) {
638 pte = pte_offset_map(pmd, addr);
639 pfn = pte_to_pagemap_entry(*pte);
640 /* unmap before userspace copy */
643 err = add_to_pagemap(addr, pfn, pm);
654 * /proc/pid/pagemap - an array mapping virtual pages to pfns
656 * For each page in the address space, this file contains one 64-bit entry
657 * consisting of the following:
659 * Bits 0-55 page frame number (PFN) if present
660 * Bits 0-4 swap type if swapped
661 * Bits 5-55 swap offset if swapped
662 * Bits 55-60 page shift (page size = 1<<page shift)
663 * Bit 61 reserved for future use
664 * Bit 62 page swapped
665 * Bit 63 page present
667 * If the page is not present but in swap, then the PFN contains an
668 * encoding of the swap file number and the page's offset into the
669 * swap. Unmapped pages return a null PFN. This allows determining
670 * precisely which pages are mapped (or in swap) and comparing mapped
671 * pages between processes.
673 * Efficient users of this interface will use /proc/pid/maps to
674 * determine which areas of memory are actually mapped and llseek to
675 * skip over unmapped regions.
677 static ssize_t pagemap_read(struct file *file, char __user *buf,
678 size_t count, loff_t *ppos)
680 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
681 struct page **pages, *page;
682 unsigned long uaddr, uend;
683 struct mm_struct *mm;
684 struct pagemapread pm;
687 struct mm_walk pagemap_walk = {};
690 unsigned long start_vaddr;
691 unsigned long end_vaddr;
697 if (!ptrace_may_access(task, PTRACE_MODE_READ))
701 /* file position must be aligned */
702 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
710 mm = get_task_mm(task);
715 uaddr = (unsigned long)buf & PAGE_MASK;
716 uend = (unsigned long)(buf + count);
717 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
721 pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
726 down_read(¤t->mm->mmap_sem);
727 ret = get_user_pages(current, current->mm, uaddr, pagecount,
729 up_read(¤t->mm->mmap_sem);
734 if (ret != pagecount) {
740 pm.out = (u64 __user *)buf;
741 pm.end = (u64 __user *)(buf + count);
743 pagemap_walk.pmd_entry = pagemap_pte_range;
744 pagemap_walk.pte_hole = pagemap_pte_hole;
745 pagemap_walk.mm = mm;
746 pagemap_walk.private = ±
749 svpfn = src / PM_ENTRY_BYTES;
750 start_vaddr = svpfn << PAGE_SHIFT;
751 end_vaddr = TASK_SIZE_OF(task);
753 /* watch out for wraparound */
754 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
755 start_vaddr = end_vaddr;
758 * The odds are that this will stop walking way
759 * before end_vaddr, because the length of the
760 * user buffer is tracked in "pm", and the walk
761 * will stop when we hit the end of the buffer.
763 ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
764 if (ret == PM_END_OF_BUFFER)
766 /* don't need mmap_sem for these, but this looks cleaner */
767 *ppos += (char __user *)pm.out - buf;
769 ret = (char __user *)pm.out - buf;
772 for (; pagecount; pagecount--) {
773 page = pages[pagecount-1];
774 if (!PageReserved(page))
776 page_cache_release(page);
783 put_task_struct(task);
788 const struct file_operations proc_pagemap_operations = {
789 .llseek = mem_lseek, /* borrow this */
790 .read = pagemap_read,
792 #endif /* CONFIG_PROC_PAGE_MONITOR */
795 extern int show_numa_map(struct seq_file *m, void *v);
797 static const struct seq_operations proc_pid_numa_maps_op = {
801 .show = show_numa_map,
804 static int numa_maps_open(struct inode *inode, struct file *file)
806 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
809 const struct file_operations proc_numa_maps_operations = {
810 .open = numa_maps_open,
813 .release = seq_release_private,