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>
11 #include <asm/uaccess.h>
12 #include <asm/tlbflush.h>
15 char *task_mem(struct mm_struct *mm, char *buffer)
17 unsigned long data, text, lib;
18 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
21 * Note: to minimize their overhead, mm maintains hiwater_vm and
22 * hiwater_rss only when about to *lower* total_vm or rss. Any
23 * collector of these hiwater stats must therefore get total_vm
24 * and rss too, which will usually be the higher. Barriers? not
25 * worth the effort, such snapshots can always be inconsistent.
27 hiwater_vm = total_vm = mm->total_vm;
28 if (hiwater_vm < mm->hiwater_vm)
29 hiwater_vm = mm->hiwater_vm;
30 hiwater_rss = total_rss = get_mm_rss(mm);
31 if (hiwater_rss < mm->hiwater_rss)
32 hiwater_rss = mm->hiwater_rss;
34 data = mm->total_vm - mm->shared_vm - mm->stack_vm;
35 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10;
36 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text;
37 buffer += sprintf(buffer,
48 hiwater_vm << (PAGE_SHIFT-10),
49 (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10),
50 mm->locked_vm << (PAGE_SHIFT-10),
51 hiwater_rss << (PAGE_SHIFT-10),
52 total_rss << (PAGE_SHIFT-10),
53 data << (PAGE_SHIFT-10),
54 mm->stack_vm << (PAGE_SHIFT-10), text, lib,
55 (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10);
59 unsigned long task_vsize(struct mm_struct *mm)
61 return PAGE_SIZE * mm->total_vm;
64 int task_statm(struct mm_struct *mm, int *shared, int *text,
65 int *data, int *resident)
67 *shared = get_mm_counter(mm, file_rss);
68 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
70 *data = mm->total_vm - mm->shared_vm;
71 *resident = *shared + get_mm_counter(mm, anon_rss);
75 int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt)
77 struct vm_area_struct * vma;
79 struct task_struct *task = get_proc_task(inode);
80 struct mm_struct * mm = NULL;
83 mm = get_task_mm(task);
84 put_task_struct(task);
88 down_read(&mm->mmap_sem);
92 if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file)
98 *mnt = mntget(vma->vm_file->f_path.mnt);
99 *dentry = dget(vma->vm_file->f_path.dentry);
103 up_read(&mm->mmap_sem);
109 static void pad_len_spaces(struct seq_file *m, int len)
111 len = 25 + sizeof(void*) * 6 - len;
114 seq_printf(m, "%*c", len, ' ');
118 * Proportional Set Size(PSS): my share of RSS.
120 * PSS of a process is the count of pages it has in memory, where each
121 * page is divided by the number of processes sharing it. So if a
122 * process has 1000 pages all to itself, and 1000 shared with one other
123 * process, its PSS will be 1500.
125 * To keep (accumulated) division errors low, we adopt a 64bit
126 * fixed-point pss counter to minimize division errors. So (pss >>
127 * PSS_SHIFT) would be the real byte count.
129 * A shift of 12 before division means (assuming 4K page size):
130 * - 1M 3-user-pages add up to 8KB errors;
131 * - supports mapcount up to 2^24, or 16M;
132 * - supports PSS up to 2^52 bytes, or 4PB.
136 struct mem_size_stats
138 unsigned long resident;
139 unsigned long shared_clean;
140 unsigned long shared_dirty;
141 unsigned long private_clean;
142 unsigned long private_dirty;
143 unsigned long referenced;
148 struct vm_area_struct *vma;
150 void (*action)(struct vm_area_struct *, pmd_t *, unsigned long,
151 unsigned long, void *);
154 static int show_map_internal(struct seq_file *m, void *v, struct mem_size_stats *mss)
156 struct proc_maps_private *priv = m->private;
157 struct task_struct *task = priv->task;
158 struct vm_area_struct *vma = v;
159 struct mm_struct *mm = vma->vm_mm;
160 struct file *file = vma->vm_file;
161 int flags = vma->vm_flags;
162 unsigned long ino = 0;
166 if (maps_protect && !ptrace_may_attach(task))
170 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
171 dev = inode->i_sb->s_dev;
175 seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n",
178 flags & VM_READ ? 'r' : '-',
179 flags & VM_WRITE ? 'w' : '-',
180 flags & VM_EXEC ? 'x' : '-',
181 flags & VM_MAYSHARE ? 's' : 'p',
182 vma->vm_pgoff << PAGE_SHIFT,
183 MAJOR(dev), MINOR(dev), ino, &len);
186 * Print the dentry name for named mappings, and a
187 * special [heap] marker for the heap:
190 pad_len_spaces(m, len);
191 seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n");
193 const char *name = arch_vma_name(vma);
196 if (vma->vm_start <= mm->start_brk &&
197 vma->vm_end >= mm->brk) {
199 } else if (vma->vm_start <= mm->start_stack &&
200 vma->vm_end >= mm->start_stack) {
208 pad_len_spaces(m, len);
219 "Shared_Clean: %8lu kB\n"
220 "Shared_Dirty: %8lu kB\n"
221 "Private_Clean: %8lu kB\n"
222 "Private_Dirty: %8lu kB\n"
223 "Referenced: %8lu kB\n",
224 (vma->vm_end - vma->vm_start) >> 10,
226 (unsigned long)(mss->pss >> (10 + PSS_SHIFT)),
227 mss->shared_clean >> 10,
228 mss->shared_dirty >> 10,
229 mss->private_clean >> 10,
230 mss->private_dirty >> 10,
231 mss->referenced >> 10);
233 if (m->count < m->size) /* vma is copied successfully */
234 m->version = (vma != get_gate_vma(task))? vma->vm_start: 0;
238 static int show_map(struct seq_file *m, void *v)
240 return show_map_internal(m, v, NULL);
243 static void smaps_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
244 unsigned long addr, unsigned long end,
247 struct mem_size_stats *mss = private;
253 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
254 for (; addr != end; pte++, addr += PAGE_SIZE) {
256 if (!pte_present(ptent))
259 mss->resident += PAGE_SIZE;
261 page = vm_normal_page(vma, addr, ptent);
265 /* Accumulate the size in pages that have been accessed. */
266 if (pte_young(ptent) || PageReferenced(page))
267 mss->referenced += PAGE_SIZE;
268 mapcount = page_mapcount(page);
270 if (pte_dirty(ptent))
271 mss->shared_dirty += PAGE_SIZE;
273 mss->shared_clean += PAGE_SIZE;
274 mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
276 if (pte_dirty(ptent))
277 mss->private_dirty += PAGE_SIZE;
279 mss->private_clean += PAGE_SIZE;
280 mss->pss += (PAGE_SIZE << PSS_SHIFT);
283 pte_unmap_unlock(pte - 1, ptl);
287 static void clear_refs_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
288 unsigned long addr, unsigned long end,
295 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
296 for (; addr != end; pte++, addr += PAGE_SIZE) {
298 if (!pte_present(ptent))
301 page = vm_normal_page(vma, addr, ptent);
305 /* Clear accessed and referenced bits. */
306 ptep_test_and_clear_young(vma, addr, pte);
307 ClearPageReferenced(page);
309 pte_unmap_unlock(pte - 1, ptl);
313 static inline void walk_pmd_range(struct pmd_walker *walker, pud_t *pud,
314 unsigned long addr, unsigned long end)
319 for (pmd = pmd_offset(pud, addr); addr != end;
320 pmd++, addr = next) {
321 next = pmd_addr_end(addr, end);
322 if (pmd_none_or_clear_bad(pmd))
324 walker->action(walker->vma, pmd, addr, next, walker->private);
328 static inline void walk_pud_range(struct pmd_walker *walker, pgd_t *pgd,
329 unsigned long addr, unsigned long end)
334 for (pud = pud_offset(pgd, addr); addr != end;
335 pud++, addr = next) {
336 next = pud_addr_end(addr, end);
337 if (pud_none_or_clear_bad(pud))
339 walk_pmd_range(walker, pud, addr, next);
344 * walk_page_range - walk the page tables of a VMA with a callback
346 * @action - callback invoked for every bottom-level (PTE) page table
347 * @private - private data passed to the callback function
349 * Recursively walk the page table for the memory area in a VMA, calling
350 * a callback for every bottom-level (PTE) page table.
352 static inline void walk_page_range(struct vm_area_struct *vma,
353 void (*action)(struct vm_area_struct *,
354 pmd_t *, unsigned long,
355 unsigned long, void *),
358 unsigned long addr = vma->vm_start;
359 unsigned long end = vma->vm_end;
360 struct pmd_walker walker = {
368 for (pgd = pgd_offset(vma->vm_mm, addr); addr != end;
369 pgd++, addr = next) {
370 next = pgd_addr_end(addr, end);
371 if (pgd_none_or_clear_bad(pgd))
373 walk_pud_range(&walker, pgd, addr, next);
377 static int show_smap(struct seq_file *m, void *v)
379 struct vm_area_struct *vma = v;
380 struct mem_size_stats mss;
382 memset(&mss, 0, sizeof mss);
383 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
384 walk_page_range(vma, smaps_pte_range, &mss);
385 return show_map_internal(m, v, &mss);
388 void clear_refs_smap(struct mm_struct *mm)
390 struct vm_area_struct *vma;
392 down_read(&mm->mmap_sem);
393 for (vma = mm->mmap; vma; vma = vma->vm_next)
394 if (vma->vm_mm && !is_vm_hugetlb_page(vma))
395 walk_page_range(vma, clear_refs_pte_range, NULL);
397 up_read(&mm->mmap_sem);
400 static void *m_start(struct seq_file *m, loff_t *pos)
402 struct proc_maps_private *priv = m->private;
403 unsigned long last_addr = m->version;
404 struct mm_struct *mm;
405 struct vm_area_struct *vma, *tail_vma = NULL;
408 /* Clear the per syscall fields in priv */
410 priv->tail_vma = NULL;
413 * We remember last_addr rather than next_addr to hit with
414 * mmap_cache most of the time. We have zero last_addr at
415 * the beginning and also after lseek. We will have -1 last_addr
416 * after the end of the vmas.
419 if (last_addr == -1UL)
422 priv->task = get_pid_task(priv->pid, PIDTYPE_PID);
426 mm = mm_for_maps(priv->task);
430 priv->tail_vma = tail_vma = get_gate_vma(priv->task);
432 /* Start with last addr hint */
433 if (last_addr && (vma = find_vma(mm, last_addr))) {
439 * Check the vma index is within the range and do
440 * sequential scan until m_index.
443 if ((unsigned long)l < mm->map_count) {
450 if (l != mm->map_count)
451 tail_vma = NULL; /* After gate vma */
457 /* End of vmas has been reached */
458 m->version = (tail_vma != NULL)? 0: -1UL;
459 up_read(&mm->mmap_sem);
464 static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma)
466 if (vma && vma != priv->tail_vma) {
467 struct mm_struct *mm = vma->vm_mm;
468 up_read(&mm->mmap_sem);
473 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
475 struct proc_maps_private *priv = m->private;
476 struct vm_area_struct *vma = v;
477 struct vm_area_struct *tail_vma = priv->tail_vma;
480 if (vma && (vma != tail_vma) && vma->vm_next)
483 return (vma != tail_vma)? tail_vma: NULL;
486 static void m_stop(struct seq_file *m, void *v)
488 struct proc_maps_private *priv = m->private;
489 struct vm_area_struct *vma = v;
493 put_task_struct(priv->task);
496 static struct seq_operations proc_pid_maps_op = {
503 static struct seq_operations proc_pid_smaps_op = {
510 static int do_maps_open(struct inode *inode, struct file *file,
511 struct seq_operations *ops)
513 struct proc_maps_private *priv;
515 priv = kzalloc(sizeof(*priv), GFP_KERNEL);
517 priv->pid = proc_pid(inode);
518 ret = seq_open(file, ops);
520 struct seq_file *m = file->private_data;
529 static int maps_open(struct inode *inode, struct file *file)
531 return do_maps_open(inode, file, &proc_pid_maps_op);
534 const struct file_operations proc_maps_operations = {
538 .release = seq_release_private,
542 extern int show_numa_map(struct seq_file *m, void *v);
544 static int show_numa_map_checked(struct seq_file *m, void *v)
546 struct proc_maps_private *priv = m->private;
547 struct task_struct *task = priv->task;
549 if (maps_protect && !ptrace_may_attach(task))
552 return show_numa_map(m, v);
555 static struct seq_operations proc_pid_numa_maps_op = {
559 .show = show_numa_map_checked
562 static int numa_maps_open(struct inode *inode, struct file *file)
564 return do_maps_open(inode, file, &proc_pid_numa_maps_op);
567 const struct file_operations proc_numa_maps_operations = {
568 .open = numa_maps_open,
571 .release = seq_release_private,
575 static int smaps_open(struct inode *inode, struct file *file)
577 return do_maps_open(inode, file, &proc_pid_smaps_op);
580 const struct file_operations proc_smaps_operations = {
584 .release = seq_release_private,