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/ptrace.h>
9 #include <linux/mempolicy.h>
10 #include <linux/swap.h>
11 #include <linux/swapops.h>
12 #include <linux/seq_file.h>
15 #include <asm/uaccess.h>
16 #include <asm/tlbflush.h>
19 void task_mem(struct seq_file *m, struct mm_struct *mm)
21 unsigned long data, text, lib;
22 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
25 * Note: to minimize their overhead, mm maintains hiwater_vm and
26 * hiwater_rss only when about to *lower* total_vm or rss. Any
27 * collector of these hiwater stats must therefore get total_vm
28 * and rss too, which will usually be the higher. Barriers? not
29 * worth the effort, such snapshots can always be inconsistent.
31 hiwater_vm = total_vm = mm->total_vm;
32 if (hiwater_vm < mm->hiwater_vm)
33 hiwater_vm = mm->hiwater_vm;
34 hiwater_rss = total_rss = get_mm_rss(mm);
35 if (hiwater_rss < mm->hiwater_rss)
36 hiwater_rss = mm->hiwater_rss;
38 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
39 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
40 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
52 hiwater_vm << (PAGE_SHIFT-10),
53 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
54 mm->locked_vm << (PAGE_SHIFT-10),
55 hiwater_rss << (PAGE_SHIFT-10),
56 total_rss << (PAGE_SHIFT-10),
57 data << (PAGE_SHIFT-10),
58 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
59 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
62 unsigned long task_vsize(struct mm_struct *mm)
64 return PAGE_SIZE * mm->total_vm;
67 int task_statm(struct mm_struct *mm, int *shared, int *text,
68 int *data, int *resident)
70 *shared = get_mm_counter(mm, file_rss);
71 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
73 *data = mm->total_vm - mm->shared_vm;
74 *resident = *shared + get_mm_counter(mm, anon_rss);
78 int proc_exe_link(struct inode *inode, struct path *path)
80 struct vm_area_struct * vma;
82 struct task_struct *task = get_proc_task(inode);
83 struct mm_struct * mm = NULL;
86 mm = get_task_mm(task);
87 put_task_struct(task);
91 down_read(&mm->mmap_sem);
95 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
101 *path = vma->vm_file->f_path;
102 path_get(&vma->vm_file->f_path);
106 up_read(&mm->mmap_sem);
112 static void pad_len_spaces(struct seq_file *m, int len)
114 len = 25 + sizeof(void*) * 6 - len;
117 seq_printf(m, "%*c", len, ' ');
120 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
122 if (vma && vma != priv->tail_vma) {
123 struct mm_struct *mm = vma->vm_mm;
124 up_read(&mm->mmap_sem);
129 static void *m_start(struct seq_file *m, loff_t *pos)
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;
137 /* Clear the per syscall fields in priv */
139 priv->tail_vma = NULL;
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.
148 if (last_addr == -1UL)
151 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
155 mm = mm_for_maps(priv->task);
159 tail_vma = get_gate_vma(priv->task);
160 priv->tail_vma = tail_vma;
162 /* Start with last addr hint */
163 vma = find_vma(mm, last_addr);
164 if (last_addr && vma) {
170 * Check the vma index is within the range and do
171 * sequential scan until m_index.
174 if ((unsigned long)l < mm->map_count) {
181 if (l != mm->map_count)
182 tail_vma = NULL; /* After gate vma */
188 /* End of vmas has been reached */
189 m->version = (tail_vma != NULL)? 0: -1UL;
190 up_read(&mm->mmap_sem);
195 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
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;
202 if (vma && (vma != tail_vma) && vma->vm_next)
205 return (vma != tail_vma)? tail_vma: NULL;
208 static void m_stop(struct seq_file *m, void *v)
210 struct proc_maps_private *priv = m->private;
211 struct vm_area_struct *vma = v;
215 put_task_struct(priv->task);
218 static int do_maps_open(struct inode *inode, struct file *file,
219 const struct seq_operations *ops)
221 struct proc_maps_private *priv;
223 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
225 priv->pid = proc_pid(inode);
226 ret = seq_open(file, ops);
228 struct seq_file *m = file->private_data;
237 static int show_map(struct seq_file *m, void *v)
239 struct proc_maps_private *priv = m->private;
240 struct task_struct *task = priv->task;
241 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;
245 unsigned long ino = 0;
249 if (maps_protect && !ptrace_may_attach(task))
253 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
254 dev = inode->i_sb->s_dev;
258 seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
261 flags & VM_READ ? 'r' : '-',
262 flags & VM_WRITE ? 'w' : '-',
263 flags & VM_EXEC ? 'x' : '-',
264 flags & VM_MAYSHARE ? 's' : 'p',
265 vma->vm_pgoff << PAGE_SHIFT,
266 MAJOR(dev), MINOR(dev), ino, &len);
269 * Print the dentry name for named mappings, and a
270 * special [heap] marker for the heap:
273 pad_len_spaces(m, len);
274 seq_path(m, &file->f_path, "\n");
276 const char *name = arch_vma_name(vma);
279 if (vma->vm_start <= mm->start_brk &&
280 vma->vm_end >= mm->brk) {
282 } else if (vma->vm_start <= mm->start_stack &&
283 vma->vm_end >= mm->start_stack) {
291 pad_len_spaces(m, len);
297 if (m->count < m->size) /* vma is copied successfully */
298 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
302 static const struct seq_operations proc_pid_maps_op = {
309 static int maps_open(struct inode *inode, struct file *file)
311 return do_maps_open(inode, file, &proc_pid_maps_op);
314 const struct file_operations proc_maps_operations = {
318 .release = seq_release_private,
322 * Proportional Set Size(PSS): my share of RSS.
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.
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.
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.
340 #ifdef CONFIG_PROC_PAGE_MONITOR
341 struct mem_size_stats {
342 struct vm_area_struct *vma;
343 unsigned long resident;
344 unsigned long shared_clean;
345 unsigned long shared_dirty;
346 unsigned long private_clean;
347 unsigned long private_dirty;
348 unsigned long referenced;
353 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
356 struct mem_size_stats *mss = private;
357 struct vm_area_struct *vma = mss->vma;
363 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
364 for (; addr != end; pte++, addr += PAGE_SIZE) {
367 if (is_swap_pte(ptent)) {
368 mss->swap += PAGE_SIZE;
372 if (!pte_present(ptent))
375 mss->resident += PAGE_SIZE;
377 page = vm_normal_page(vma, addr, ptent);
381 /* Accumulate the size in pages that have been accessed. */
382 if (pte_young(ptent) || PageReferenced(page))
383 mss->referenced += PAGE_SIZE;
384 mapcount = page_mapcount(page);
386 if (pte_dirty(ptent))
387 mss->shared_dirty += PAGE_SIZE;
389 mss->shared_clean += PAGE_SIZE;
390 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
392 if (pte_dirty(ptent))
393 mss->private_dirty += PAGE_SIZE;
395 mss->private_clean += PAGE_SIZE;
396 mss->pss += (PAGE_SIZE << PSS_SHIFT);
399 pte_unmap_unlock(pte - 1, ptl);
404 static struct mm_walk smaps_walk = { .pmd_entry = smaps_pte_range };
406 static int show_smap(struct seq_file *m, void *v)
408 struct vm_area_struct *vma = v;
409 struct mem_size_stats mss;
412 memset(&mss, 0, sizeof mss);
414 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
415 walk_page_range(vma->vm_mm, vma->vm_start, vma->vm_end,
418 ret = show_map(m, v);
426 "Shared_Clean: %8lu kB\n"
427 "Shared_Dirty: %8lu kB\n"
428 "Private_Clean: %8lu kB\n"
429 "Private_Dirty: %8lu kB\n"
430 "Referenced: %8lu kB\n"
432 (vma->vm_end - vma->vm_start) >> 10,
434 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)),
435 mss.shared_clean >> 10,
436 mss.shared_dirty >> 10,
437 mss.private_clean >> 10,
438 mss.private_dirty >> 10,
439 mss.referenced >> 10,
445 static const struct seq_operations proc_pid_smaps_op = {
452 static int smaps_open(struct inode *inode, struct file *file)
454 return do_maps_open(inode, file, &proc_pid_smaps_op);
457 const struct file_operations proc_smaps_operations = {
461 .release = seq_release_private,
464 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
465 unsigned long end, void *private)
467 struct vm_area_struct *vma = private;
472 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
473 for (; addr != end; pte++, addr += PAGE_SIZE) {
475 if (!pte_present(ptent))
478 page = vm_normal_page(vma, addr, ptent);
482 /* Clear accessed and referenced bits. */
483 ptep_test_and_clear_young(vma, addr, pte);
484 ClearPageReferenced(page);
486 pte_unmap_unlock(pte - 1, ptl);
491 static struct mm_walk clear_refs_walk = { .pmd_entry = clear_refs_pte_range };
493 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
494 size_t count, loff_t *ppos)
496 struct task_struct *task;
497 char buffer[PROC_NUMBUF], *end;
498 struct mm_struct *mm;
499 struct vm_area_struct *vma;
501 memset(buffer, 0, sizeof(buffer));
502 if (count > sizeof(buffer) - 1)
503 count = sizeof(buffer) - 1;
504 if (copy_from_user(buffer, buf, count))
506 if (!simple_strtol(buffer, &end, 0))
510 task = get_proc_task(file->f_path.dentry->d_inode);
513 mm = get_task_mm(task);
515 down_read(&mm->mmap_sem);
516 for (vma = mm->mmap; vma; vma = vma->vm_next)
517 if (!is_vm_hugetlb_page(vma))
518 walk_page_range(mm, vma->vm_start, vma->vm_end,
519 &clear_refs_walk, vma);
521 up_read(&mm->mmap_sem);
524 put_task_struct(task);
525 if (end - buffer == 0)
530 const struct file_operations proc_clear_refs_operations = {
531 .write = clear_refs_write,
535 char __user *out, *end;
538 #define PM_ENTRY_BYTES sizeof(u64)
539 #define PM_STATUS_BITS 3
540 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS)
541 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET)
542 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK)
543 #define PM_PSHIFT_BITS 6
544 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS)
545 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET)
546 #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK)
547 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1)
548 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK)
550 #define PM_PRESENT PM_STATUS(4LL)
551 #define PM_SWAP PM_STATUS(2LL)
552 #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT)
553 #define PM_END_OF_BUFFER 1
555 static int add_to_pagemap(unsigned long addr, u64 pfn,
556 struct pagemapread *pm)
559 * Make sure there's room in the buffer for an
560 * entire entry. Otherwise, only copy part of
563 if (pm->out + PM_ENTRY_BYTES >= pm->end) {
564 if (copy_to_user(pm->out, &pfn, pm->end - pm->out))
567 return PM_END_OF_BUFFER;
570 if (put_user(pfn, pm->out))
572 pm->out += PM_ENTRY_BYTES;
576 static int pagemap_pte_hole(unsigned long start, unsigned long end,
579 struct pagemapread *pm = private;
582 for (addr = start; addr < end; addr += PAGE_SIZE) {
583 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
590 static u64 swap_pte_to_pagemap_entry(pte_t pte)
592 swp_entry_t e = pte_to_swp_entry(pte);
593 return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
596 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
599 struct pagemapread *pm = private;
603 for (; addr != end; addr += PAGE_SIZE) {
604 u64 pfn = PM_NOT_PRESENT;
605 pte = pte_offset_map(pmd, addr);
606 if (is_swap_pte(*pte))
607 pfn = PM_PFRAME(swap_pte_to_pagemap_entry(*pte))
608 | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP;
609 else if (pte_present(*pte))
610 pfn = PM_PFRAME(pte_pfn(*pte))
611 | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
612 /* unmap so we're not in atomic when we copy to userspace */
614 err = add_to_pagemap(addr, pfn, pm);
624 static struct mm_walk pagemap_walk = {
625 .pmd_entry = pagemap_pte_range,
626 .pte_hole = pagemap_pte_hole
630 * /proc/pid/pagemap - an array mapping virtual pages to pfns
632 * For each page in the address space, this file contains one 64-bit entry
633 * consisting of the following:
635 * Bits 0-55 page frame number (PFN) if present
636 * Bits 0-4 swap type if swapped
637 * Bits 5-55 swap offset if swapped
638 * Bits 55-60 page shift (page size = 1<<page shift)
639 * Bit 61 reserved for future use
640 * Bit 62 page swapped
641 * Bit 63 page present
643 * If the page is not present but in swap, then the PFN contains an
644 * encoding of the swap file number and the page's offset into the
645 * swap. Unmapped pages return a null PFN. This allows determining
646 * precisely which pages are mapped (or in swap) and comparing mapped
647 * pages between processes.
649 * Efficient users of this interface will use /proc/pid/maps to
650 * determine which areas of memory are actually mapped and llseek to
651 * skip over unmapped regions.
653 static ssize_t pagemap_read(struct file *file, char __user *buf,
654 size_t count, loff_t *ppos)
656 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
657 struct page **pages, *page;
658 unsigned long uaddr, uend;
659 struct mm_struct *mm;
660 struct pagemapread pm;
668 if (!ptrace_may_attach(task))
672 /* file position must be aligned */
673 if (*ppos % PM_ENTRY_BYTES)
677 mm = get_task_mm(task);
682 uaddr = (unsigned long)buf & PAGE_MASK;
683 uend = (unsigned long)(buf + count);
684 pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
685 pages = kmalloc(pagecount * sizeof(struct page *), GFP_KERNEL);
689 down_read(¤t->mm->mmap_sem);
690 ret = get_user_pages(current, current->mm, uaddr, pagecount,
692 up_read(¤t->mm->mmap_sem);
697 if (ret != pagecount) {
704 pm.end = buf + count;
706 if (!ptrace_may_attach(task)) {
709 unsigned long src = *ppos;
710 unsigned long svpfn = src / PM_ENTRY_BYTES;
711 unsigned long start_vaddr = svpfn << PAGE_SHIFT;
712 unsigned long end_vaddr = TASK_SIZE_OF(task);
714 /* watch out for wraparound */
715 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
716 start_vaddr = end_vaddr;
719 * The odds are that this will stop walking way
720 * before end_vaddr, because the length of the
721 * user buffer is tracked in "pm", and the walk
722 * will stop when we hit the end of the buffer.
724 ret = walk_page_range(mm, start_vaddr, end_vaddr,
726 if (ret == PM_END_OF_BUFFER)
728 /* don't need mmap_sem for these, but this looks cleaner */
729 *ppos += pm.out - buf;
735 for (; pagecount; pagecount--) {
736 page = pages[pagecount-1];
737 if (!PageReserved(page))
739 page_cache_release(page);
746 put_task_struct(task);
751 const struct file_operations proc_pagemap_operations = {
752 .llseek = mem_lseek, /* borrow this */
753 .read = pagemap_read,
755 #endif /* CONFIG_PROC_PAGE_MONITOR */
758 extern int show_numa_map(struct seq_file *m, void *v);
760 static int show_numa_map_checked(struct seq_file *m, void *v)
762 struct proc_maps_private *priv = m->private;
763 struct task_struct *task = priv->task;
765 if (maps_protect && !ptrace_may_attach(task))
768 return show_numa_map(m, v);
771 static const struct seq_operations proc_pid_numa_maps_op = {
775 .show = show_numa_map_checked
778 static int numa_maps_open(struct inode *inode, struct file *file)
780 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
783 const struct file_operations proc_numa_maps_operations = {
784 .open = numa_maps_open,
787 .release = seq_release_private,