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>
14 #include <asm/uaccess.h>
15 #include <asm/tlbflush.h>
18 char *task_mem(struct mm_struct *mm, char *buffer)
20 unsigned long data, text, lib;
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 buffer += sprintf(buffer,
51 hiwater_vm << (PAGE_SHIFT-10),
52 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
53 mm->locked_vm << (PAGE_SHIFT-10),
54 hiwater_rss << (PAGE_SHIFT-10),
55 total_rss << (PAGE_SHIFT-10),
56 data << (PAGE_SHIFT-10),
57 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
58 (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 dentry **dentry, struct vfsmount **mnt)
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 *mnt = mntget(vma->vm_file->f_path.mnt);
102 *dentry = dget(vma->vm_file->f_path.dentry);
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 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.mnt, file->f_path.dentry, "\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 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
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;
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) {
366 if (!pte_present(ptent))
369 mss->resident += PAGE_SIZE;
371 page = vm_normal_page(vma, addr, ptent);
375 /* Accumulate the size in pages that have been accessed. */
376 if (pte_young(ptent) || PageReferenced(page))
377 mss->referenced += PAGE_SIZE;
378 mapcount = page_mapcount(page);
380 if (pte_dirty(ptent))
381 mss->shared_dirty += PAGE_SIZE;
383 mss->shared_clean += PAGE_SIZE;
384 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
386 if (pte_dirty(ptent))
387 mss->private_dirty += PAGE_SIZE;
389 mss->private_clean += PAGE_SIZE;
390 mss->pss += (PAGE_SIZE << PSS_SHIFT);
393 pte_unmap_unlock(pte - 1, ptl);
398 static struct mm_walk smaps_walk = { .pmd_entry = smaps_pte_range };
400 static int show_smap(struct seq_file *m, void *v)
402 struct vm_area_struct *vma = v;
403 struct mem_size_stats mss;
406 memset(&mss, 0, sizeof mss);
408 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
409 walk_page_range(vma->vm_mm, vma->vm_start, vma->vm_end,
412 ret = show_map(m, v);
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,
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);
437 static struct seq_operations proc_pid_smaps_op = {
444 static int smaps_open(struct inode *inode, struct file *file)
446 return do_maps_open(inode, file, &proc_pid_smaps_op);
449 const struct file_operations proc_smaps_operations = {
453 .release = seq_release_private,
456 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
457 unsigned long end, void *private)
459 struct vm_area_struct *vma = private;
464 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
465 for (; addr != end; pte++, addr += PAGE_SIZE) {
467 if (!pte_present(ptent))
470 page = vm_normal_page(vma, addr, ptent);
474 /* Clear accessed and referenced bits. */
475 ptep_test_and_clear_young(vma, addr, pte);
476 ClearPageReferenced(page);
478 pte_unmap_unlock(pte - 1, ptl);
483 static struct mm_walk clear_refs_walk = { .pmd_entry = clear_refs_pte_range };
485 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
486 size_t count, loff_t *ppos)
488 struct task_struct *task;
489 char buffer[PROC_NUMBUF], *end;
490 struct mm_struct *mm;
491 struct vm_area_struct *vma;
493 memset(buffer, 0, sizeof(buffer));
494 if (count > sizeof(buffer) - 1)
495 count = sizeof(buffer) - 1;
496 if (copy_from_user(buffer, buf, count))
498 if (!simple_strtol(buffer, &end, 0))
502 task = get_proc_task(file->f_path.dentry->d_inode);
505 mm = get_task_mm(task);
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);
513 up_read(&mm->mmap_sem);
516 put_task_struct(task);
517 if (end - buffer == 0)
522 const struct file_operations proc_clear_refs_operations = {
523 .write = clear_refs_write,
527 char __user *out, *end;
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
539 static int add_to_pagemap(unsigned long addr, u64 pfn,
540 struct pagemapread *pm)
543 * Make sure there's room in the buffer for an
544 * entire entry. Otherwise, only copy part of
547 if (pm->out + PM_ENTRY_BYTES >= pm->end) {
548 if (copy_to_user(pm->out, &pfn, pm->end - pm->out))
551 return PM_END_OF_BUFFER;
554 if (put_user(pfn, pm->out))
556 pm->out += PM_ENTRY_BYTES;
560 static int pagemap_pte_hole(unsigned long start, unsigned long end,
563 struct pagemapread *pm = private;
566 for (addr = start; addr < end; addr += PAGE_SIZE) {
567 err = add_to_pagemap(addr, PM_NOT_PRESENT, pm);
574 u64 swap_pte_to_pagemap_entry(pte_t pte)
576 swp_entry_t e = pte_to_swp_entry(pte);
577 return PM_SWAP | swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT);
580 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
583 struct pagemapread *pm = private;
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))
594 /* unmap so we're not in atomic when we copy to userspace */
596 err = add_to_pagemap(addr, pfn, pm);
606 static struct mm_walk pagemap_walk = {
607 .pmd_entry = pagemap_pte_range,
608 .pte_hole = pagemap_pte_hole
612 * /proc/pid/pagemap - an array mapping virtual pages to pfns
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.
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.
627 static ssize_t pagemap_read(struct file *file, char __user *buf,
628 size_t count, loff_t *ppos)
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;
642 if (!ptrace_may_attach(task))
646 /* file position must be aligned */
647 if (*ppos % PM_ENTRY_BYTES)
651 mm = get_task_mm(task);
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);
663 down_read(¤t->mm->mmap_sem);
664 ret = get_user_pages(current, current->mm, uaddr, pagecount,
666 up_read(¤t->mm->mmap_sem);
672 pm.end = buf + count;
674 if (!ptrace_may_attach(task)) {
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);
682 /* watch out for wraparound */
683 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
684 start_vaddr = end_vaddr;
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.
692 ret = walk_page_range(mm, start_vaddr, end_vaddr,
694 if (ret == PM_END_OF_BUFFER)
696 /* don't need mmap_sem for these, but this looks cleaner */
697 *ppos += pm.out - buf;
702 for (; pagecount; pagecount--) {
703 page = pages[pagecount-1];
704 if (!PageReserved(page))
706 page_cache_release(page);
712 put_task_struct(task);
717 const struct file_operations proc_pagemap_operations = {
718 .llseek = mem_lseek, /* borrow this */
719 .read = pagemap_read,
721 #endif /* CONFIG_PROC_PAGE_MONITOR */
724 extern int show_numa_map(struct seq_file *m, void *v);
726 static int show_numa_map_checked(struct seq_file *m, void *v)
728 struct proc_maps_private *priv = m->private;
729 struct task_struct *task = priv->task;
731 if (maps_protect && !ptrace_may_attach(task))
734 return show_numa_map(m, v);
737 static struct seq_operations proc_pid_numa_maps_op = {
741 .show = show_numa_map_checked
744 static int numa_maps_open(struct inode *inode, struct file *file)
746 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
749 const struct file_operations proc_numa_maps_operations = {
750 .open = numa_maps_open,
753 .release = seq_release_private,