]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #include <linux/mm.h> |
2 | #include <linux/hugetlb.h> | |
22e057c5 | 3 | #include <linux/huge_mm.h> |
1da177e4 LT |
4 | #include <linux/mount.h> |
5 | #include <linux/seq_file.h> | |
e070ad49 | 6 | #include <linux/highmem.h> |
5096add8 | 7 | #include <linux/ptrace.h> |
5a0e3ad6 | 8 | #include <linux/slab.h> |
6e21c8f1 CL |
9 | #include <linux/pagemap.h> |
10 | #include <linux/mempolicy.h> | |
22e057c5 | 11 | #include <linux/rmap.h> |
85863e47 MM |
12 | #include <linux/swap.h> |
13 | #include <linux/swapops.h> | |
e070ad49 | 14 | |
1da177e4 LT |
15 | #include <asm/elf.h> |
16 | #include <asm/uaccess.h> | |
e070ad49 | 17 | #include <asm/tlbflush.h> |
1da177e4 LT |
18 | #include "internal.h" |
19 | ||
df5f8314 | 20 | void task_mem(struct seq_file *m, struct mm_struct *mm) |
1da177e4 | 21 | { |
b084d435 | 22 | unsigned long data, text, lib, swap; |
365e9c87 HD |
23 | unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; |
24 | ||
25 | /* | |
26 | * Note: to minimize their overhead, mm maintains hiwater_vm and | |
27 | * hiwater_rss only when about to *lower* total_vm or rss. Any | |
28 | * collector of these hiwater stats must therefore get total_vm | |
29 | * and rss too, which will usually be the higher. Barriers? not | |
30 | * worth the effort, such snapshots can always be inconsistent. | |
31 | */ | |
32 | hiwater_vm = total_vm = mm->total_vm; | |
33 | if (hiwater_vm < mm->hiwater_vm) | |
34 | hiwater_vm = mm->hiwater_vm; | |
35 | hiwater_rss = total_rss = get_mm_rss(mm); | |
36 | if (hiwater_rss < mm->hiwater_rss) | |
37 | hiwater_rss = mm->hiwater_rss; | |
1da177e4 LT |
38 | |
39 | data = mm->total_vm - mm->shared_vm - mm->stack_vm; | |
40 | text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; | |
41 | lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; | |
b084d435 | 42 | swap = get_mm_counter(mm, MM_SWAPENTS); |
df5f8314 | 43 | seq_printf(m, |
365e9c87 | 44 | "VmPeak:\t%8lu kB\n" |
1da177e4 LT |
45 | "VmSize:\t%8lu kB\n" |
46 | "VmLck:\t%8lu kB\n" | |
365e9c87 | 47 | "VmHWM:\t%8lu kB\n" |
1da177e4 LT |
48 | "VmRSS:\t%8lu kB\n" |
49 | "VmData:\t%8lu kB\n" | |
50 | "VmStk:\t%8lu kB\n" | |
51 | "VmExe:\t%8lu kB\n" | |
52 | "VmLib:\t%8lu kB\n" | |
b084d435 KH |
53 | "VmPTE:\t%8lu kB\n" |
54 | "VmSwap:\t%8lu kB\n", | |
365e9c87 HD |
55 | hiwater_vm << (PAGE_SHIFT-10), |
56 | (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10), | |
1da177e4 | 57 | mm->locked_vm << (PAGE_SHIFT-10), |
365e9c87 HD |
58 | hiwater_rss << (PAGE_SHIFT-10), |
59 | total_rss << (PAGE_SHIFT-10), | |
1da177e4 LT |
60 | data << (PAGE_SHIFT-10), |
61 | mm->stack_vm << (PAGE_SHIFT-10), text, lib, | |
b084d435 KH |
62 | (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10, |
63 | swap << (PAGE_SHIFT-10)); | |
1da177e4 LT |
64 | } |
65 | ||
66 | unsigned long task_vsize(struct mm_struct *mm) | |
67 | { | |
68 | return PAGE_SIZE * mm->total_vm; | |
69 | } | |
70 | ||
a2ade7b6 AD |
71 | unsigned long task_statm(struct mm_struct *mm, |
72 | unsigned long *shared, unsigned long *text, | |
73 | unsigned long *data, unsigned long *resident) | |
1da177e4 | 74 | { |
d559db08 | 75 | *shared = get_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
76 | *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) |
77 | >> PAGE_SHIFT; | |
78 | *data = mm->total_vm - mm->shared_vm; | |
d559db08 | 79 | *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); |
1da177e4 LT |
80 | return mm->total_vm; |
81 | } | |
82 | ||
1da177e4 LT |
83 | static void pad_len_spaces(struct seq_file *m, int len) |
84 | { | |
85 | len = 25 + sizeof(void*) * 6 - len; | |
86 | if (len < 1) | |
87 | len = 1; | |
88 | seq_printf(m, "%*c", len, ' '); | |
89 | } | |
90 | ||
a6198797 MM |
91 | static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma) |
92 | { | |
93 | if (vma && vma != priv->tail_vma) { | |
94 | struct mm_struct *mm = vma->vm_mm; | |
95 | up_read(&mm->mmap_sem); | |
96 | mmput(mm); | |
97 | } | |
98 | } | |
ec4dd3eb | 99 | |
a6198797 | 100 | static void *m_start(struct seq_file *m, loff_t *pos) |
e070ad49 | 101 | { |
a6198797 MM |
102 | struct proc_maps_private *priv = m->private; |
103 | unsigned long last_addr = m->version; | |
104 | struct mm_struct *mm; | |
105 | struct vm_area_struct *vma, *tail_vma = NULL; | |
106 | loff_t l = *pos; | |
107 | ||
108 | /* Clear the per syscall fields in priv */ | |
109 | priv->task = NULL; | |
110 | priv->tail_vma = NULL; | |
111 | ||
112 | /* | |
113 | * We remember last_addr rather than next_addr to hit with | |
114 | * mmap_cache most of the time. We have zero last_addr at | |
115 | * the beginning and also after lseek. We will have -1 last_addr | |
116 | * after the end of the vmas. | |
117 | */ | |
118 | ||
119 | if (last_addr == -1UL) | |
120 | return NULL; | |
121 | ||
122 | priv->task = get_pid_task(priv->pid, PIDTYPE_PID); | |
123 | if (!priv->task) | |
124 | return NULL; | |
125 | ||
126 | mm = mm_for_maps(priv->task); | |
127 | if (!mm) | |
128 | return NULL; | |
00f89d21 | 129 | down_read(&mm->mmap_sem); |
a6198797 MM |
130 | |
131 | tail_vma = get_gate_vma(priv->task); | |
132 | priv->tail_vma = tail_vma; | |
133 | ||
134 | /* Start with last addr hint */ | |
135 | vma = find_vma(mm, last_addr); | |
136 | if (last_addr && vma) { | |
137 | vma = vma->vm_next; | |
138 | goto out; | |
139 | } | |
140 | ||
141 | /* | |
142 | * Check the vma index is within the range and do | |
143 | * sequential scan until m_index. | |
144 | */ | |
145 | vma = NULL; | |
146 | if ((unsigned long)l < mm->map_count) { | |
147 | vma = mm->mmap; | |
148 | while (l-- && vma) | |
149 | vma = vma->vm_next; | |
150 | goto out; | |
151 | } | |
152 | ||
153 | if (l != mm->map_count) | |
154 | tail_vma = NULL; /* After gate vma */ | |
155 | ||
156 | out: | |
157 | if (vma) | |
158 | return vma; | |
159 | ||
160 | /* End of vmas has been reached */ | |
161 | m->version = (tail_vma != NULL)? 0: -1UL; | |
162 | up_read(&mm->mmap_sem); | |
163 | mmput(mm); | |
164 | return tail_vma; | |
165 | } | |
166 | ||
167 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
168 | { | |
169 | struct proc_maps_private *priv = m->private; | |
170 | struct vm_area_struct *vma = v; | |
171 | struct vm_area_struct *tail_vma = priv->tail_vma; | |
172 | ||
173 | (*pos)++; | |
174 | if (vma && (vma != tail_vma) && vma->vm_next) | |
175 | return vma->vm_next; | |
176 | vma_stop(priv, vma); | |
177 | return (vma != tail_vma)? tail_vma: NULL; | |
178 | } | |
179 | ||
180 | static void m_stop(struct seq_file *m, void *v) | |
181 | { | |
182 | struct proc_maps_private *priv = m->private; | |
183 | struct vm_area_struct *vma = v; | |
184 | ||
185 | vma_stop(priv, vma); | |
186 | if (priv->task) | |
187 | put_task_struct(priv->task); | |
188 | } | |
189 | ||
190 | static int do_maps_open(struct inode *inode, struct file *file, | |
03a44825 | 191 | const struct seq_operations *ops) |
a6198797 MM |
192 | { |
193 | struct proc_maps_private *priv; | |
194 | int ret = -ENOMEM; | |
195 | priv = kzalloc(sizeof(*priv), GFP_KERNEL); | |
196 | if (priv) { | |
197 | priv->pid = proc_pid(inode); | |
198 | ret = seq_open(file, ops); | |
199 | if (!ret) { | |
200 | struct seq_file *m = file->private_data; | |
201 | m->private = priv; | |
202 | } else { | |
203 | kfree(priv); | |
204 | } | |
205 | } | |
206 | return ret; | |
207 | } | |
e070ad49 | 208 | |
7c88db0c | 209 | static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma) |
1da177e4 | 210 | { |
e070ad49 ML |
211 | struct mm_struct *mm = vma->vm_mm; |
212 | struct file *file = vma->vm_file; | |
213 | int flags = vma->vm_flags; | |
1da177e4 | 214 | unsigned long ino = 0; |
6260a4b0 | 215 | unsigned long long pgoff = 0; |
d7824370 | 216 | unsigned long start; |
1da177e4 LT |
217 | dev_t dev = 0; |
218 | int len; | |
219 | ||
220 | if (file) { | |
2fddfeef | 221 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 LT |
222 | dev = inode->i_sb->s_dev; |
223 | ino = inode->i_ino; | |
6260a4b0 | 224 | pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; |
1da177e4 LT |
225 | } |
226 | ||
d7824370 LT |
227 | /* We don't show the stack guard page in /proc/maps */ |
228 | start = vma->vm_start; | |
229 | if (vma->vm_flags & VM_GROWSDOWN) | |
39aa3cb3 SB |
230 | if (!vma_stack_continue(vma->vm_prev, vma->vm_start)) |
231 | start += PAGE_SIZE; | |
d7824370 | 232 | |
1804dc6e | 233 | seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n", |
d7824370 | 234 | start, |
e070ad49 | 235 | vma->vm_end, |
1da177e4 LT |
236 | flags & VM_READ ? 'r' : '-', |
237 | flags & VM_WRITE ? 'w' : '-', | |
238 | flags & VM_EXEC ? 'x' : '-', | |
239 | flags & VM_MAYSHARE ? 's' : 'p', | |
6260a4b0 | 240 | pgoff, |
1da177e4 LT |
241 | MAJOR(dev), MINOR(dev), ino, &len); |
242 | ||
243 | /* | |
244 | * Print the dentry name for named mappings, and a | |
245 | * special [heap] marker for the heap: | |
246 | */ | |
e070ad49 | 247 | if (file) { |
1da177e4 | 248 | pad_len_spaces(m, len); |
c32c2f63 | 249 | seq_path(m, &file->f_path, "\n"); |
1da177e4 | 250 | } else { |
e6e5494c IM |
251 | const char *name = arch_vma_name(vma); |
252 | if (!name) { | |
253 | if (mm) { | |
254 | if (vma->vm_start <= mm->start_brk && | |
e070ad49 | 255 | vma->vm_end >= mm->brk) { |
e6e5494c IM |
256 | name = "[heap]"; |
257 | } else if (vma->vm_start <= mm->start_stack && | |
258 | vma->vm_end >= mm->start_stack) { | |
259 | name = "[stack]"; | |
1da177e4 | 260 | } |
e6e5494c IM |
261 | } else { |
262 | name = "[vdso]"; | |
1da177e4 | 263 | } |
e6e5494c IM |
264 | } |
265 | if (name) { | |
1da177e4 | 266 | pad_len_spaces(m, len); |
e6e5494c | 267 | seq_puts(m, name); |
1da177e4 LT |
268 | } |
269 | } | |
270 | seq_putc(m, '\n'); | |
7c88db0c JK |
271 | } |
272 | ||
273 | static int show_map(struct seq_file *m, void *v) | |
274 | { | |
275 | struct vm_area_struct *vma = v; | |
276 | struct proc_maps_private *priv = m->private; | |
277 | struct task_struct *task = priv->task; | |
278 | ||
279 | show_map_vma(m, vma); | |
e070ad49 | 280 | |
e070ad49 ML |
281 | if (m->count < m->size) /* vma is copied successfully */ |
282 | m->version = (vma != get_gate_vma(task))? vma->vm_start: 0; | |
1da177e4 LT |
283 | return 0; |
284 | } | |
285 | ||
03a44825 | 286 | static const struct seq_operations proc_pid_maps_op = { |
a6198797 MM |
287 | .start = m_start, |
288 | .next = m_next, | |
289 | .stop = m_stop, | |
290 | .show = show_map | |
291 | }; | |
292 | ||
293 | static int maps_open(struct inode *inode, struct file *file) | |
294 | { | |
295 | return do_maps_open(inode, file, &proc_pid_maps_op); | |
296 | } | |
297 | ||
298 | const struct file_operations proc_maps_operations = { | |
299 | .open = maps_open, | |
300 | .read = seq_read, | |
301 | .llseek = seq_lseek, | |
302 | .release = seq_release_private, | |
303 | }; | |
304 | ||
305 | /* | |
306 | * Proportional Set Size(PSS): my share of RSS. | |
307 | * | |
308 | * PSS of a process is the count of pages it has in memory, where each | |
309 | * page is divided by the number of processes sharing it. So if a | |
310 | * process has 1000 pages all to itself, and 1000 shared with one other | |
311 | * process, its PSS will be 1500. | |
312 | * | |
313 | * To keep (accumulated) division errors low, we adopt a 64bit | |
314 | * fixed-point pss counter to minimize division errors. So (pss >> | |
315 | * PSS_SHIFT) would be the real byte count. | |
316 | * | |
317 | * A shift of 12 before division means (assuming 4K page size): | |
318 | * - 1M 3-user-pages add up to 8KB errors; | |
319 | * - supports mapcount up to 2^24, or 16M; | |
320 | * - supports PSS up to 2^52 bytes, or 4PB. | |
321 | */ | |
322 | #define PSS_SHIFT 12 | |
323 | ||
1e883281 | 324 | #ifdef CONFIG_PROC_PAGE_MONITOR |
214e471f | 325 | struct mem_size_stats { |
a6198797 MM |
326 | struct vm_area_struct *vma; |
327 | unsigned long resident; | |
328 | unsigned long shared_clean; | |
329 | unsigned long shared_dirty; | |
330 | unsigned long private_clean; | |
331 | unsigned long private_dirty; | |
332 | unsigned long referenced; | |
b40d4f84 | 333 | unsigned long anonymous; |
214e471f | 334 | unsigned long swap; |
a6198797 MM |
335 | u64 pss; |
336 | }; | |
337 | ||
ae11c4d9 DH |
338 | |
339 | static void smaps_pte_entry(pte_t ptent, unsigned long addr, | |
3c9acc78 | 340 | unsigned long ptent_size, struct mm_walk *walk) |
ae11c4d9 DH |
341 | { |
342 | struct mem_size_stats *mss = walk->private; | |
343 | struct vm_area_struct *vma = mss->vma; | |
344 | struct page *page; | |
345 | int mapcount; | |
346 | ||
347 | if (is_swap_pte(ptent)) { | |
3c9acc78 | 348 | mss->swap += ptent_size; |
ae11c4d9 DH |
349 | return; |
350 | } | |
351 | ||
352 | if (!pte_present(ptent)) | |
353 | return; | |
354 | ||
355 | page = vm_normal_page(vma, addr, ptent); | |
356 | if (!page) | |
357 | return; | |
358 | ||
359 | if (PageAnon(page)) | |
3c9acc78 | 360 | mss->anonymous += ptent_size; |
ae11c4d9 | 361 | |
3c9acc78 | 362 | mss->resident += ptent_size; |
ae11c4d9 DH |
363 | /* Accumulate the size in pages that have been accessed. */ |
364 | if (pte_young(ptent) || PageReferenced(page)) | |
3c9acc78 | 365 | mss->referenced += ptent_size; |
ae11c4d9 DH |
366 | mapcount = page_mapcount(page); |
367 | if (mapcount >= 2) { | |
368 | if (pte_dirty(ptent) || PageDirty(page)) | |
3c9acc78 | 369 | mss->shared_dirty += ptent_size; |
ae11c4d9 | 370 | else |
3c9acc78 DH |
371 | mss->shared_clean += ptent_size; |
372 | mss->pss += (ptent_size << PSS_SHIFT) / mapcount; | |
ae11c4d9 DH |
373 | } else { |
374 | if (pte_dirty(ptent) || PageDirty(page)) | |
3c9acc78 | 375 | mss->private_dirty += ptent_size; |
ae11c4d9 | 376 | else |
3c9acc78 DH |
377 | mss->private_clean += ptent_size; |
378 | mss->pss += (ptent_size << PSS_SHIFT); | |
ae11c4d9 DH |
379 | } |
380 | } | |
381 | ||
b3ae5acb | 382 | static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 383 | struct mm_walk *walk) |
e070ad49 | 384 | { |
2165009b | 385 | struct mem_size_stats *mss = walk->private; |
b3ae5acb | 386 | struct vm_area_struct *vma = mss->vma; |
ae11c4d9 | 387 | pte_t *pte; |
705e87c0 | 388 | spinlock_t *ptl; |
e070ad49 | 389 | |
22e057c5 DH |
390 | spin_lock(&walk->mm->page_table_lock); |
391 | if (pmd_trans_huge(*pmd)) { | |
392 | if (pmd_trans_splitting(*pmd)) { | |
393 | spin_unlock(&walk->mm->page_table_lock); | |
394 | wait_split_huge_page(vma->anon_vma, pmd); | |
395 | } else { | |
396 | smaps_pte_entry(*(pte_t *)pmd, addr, | |
397 | HPAGE_PMD_SIZE, walk); | |
398 | spin_unlock(&walk->mm->page_table_lock); | |
399 | return 0; | |
400 | } | |
401 | } else { | |
402 | spin_unlock(&walk->mm->page_table_lock); | |
403 | } | |
404 | /* | |
405 | * The mmap_sem held all the way back in m_start() is what | |
406 | * keeps khugepaged out of here and from collapsing things | |
407 | * in here. | |
408 | */ | |
705e87c0 | 409 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
ae11c4d9 | 410 | for (; addr != end; pte++, addr += PAGE_SIZE) |
3c9acc78 | 411 | smaps_pte_entry(*pte, addr, PAGE_SIZE, walk); |
705e87c0 HD |
412 | pte_unmap_unlock(pte - 1, ptl); |
413 | cond_resched(); | |
b3ae5acb | 414 | return 0; |
e070ad49 ML |
415 | } |
416 | ||
e070ad49 ML |
417 | static int show_smap(struct seq_file *m, void *v) |
418 | { | |
7c88db0c JK |
419 | struct proc_maps_private *priv = m->private; |
420 | struct task_struct *task = priv->task; | |
e070ad49 | 421 | struct vm_area_struct *vma = v; |
e070ad49 | 422 | struct mem_size_stats mss; |
2165009b DH |
423 | struct mm_walk smaps_walk = { |
424 | .pmd_entry = smaps_pte_range, | |
425 | .mm = vma->vm_mm, | |
426 | .private = &mss, | |
427 | }; | |
e070ad49 ML |
428 | |
429 | memset(&mss, 0, sizeof mss); | |
b3ae5acb | 430 | mss.vma = vma; |
d82ef020 | 431 | /* mmap_sem is held in m_start */ |
5ddfae16 | 432 | if (vma->vm_mm && !is_vm_hugetlb_page(vma)) |
2165009b | 433 | walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); |
4752c369 | 434 | |
7c88db0c | 435 | show_map_vma(m, vma); |
4752c369 MM |
436 | |
437 | seq_printf(m, | |
438 | "Size: %8lu kB\n" | |
439 | "Rss: %8lu kB\n" | |
440 | "Pss: %8lu kB\n" | |
441 | "Shared_Clean: %8lu kB\n" | |
442 | "Shared_Dirty: %8lu kB\n" | |
443 | "Private_Clean: %8lu kB\n" | |
444 | "Private_Dirty: %8lu kB\n" | |
214e471f | 445 | "Referenced: %8lu kB\n" |
b40d4f84 | 446 | "Anonymous: %8lu kB\n" |
08fba699 | 447 | "Swap: %8lu kB\n" |
3340289d | 448 | "KernelPageSize: %8lu kB\n" |
2d90508f NK |
449 | "MMUPageSize: %8lu kB\n" |
450 | "Locked: %8lu kB\n", | |
4752c369 MM |
451 | (vma->vm_end - vma->vm_start) >> 10, |
452 | mss.resident >> 10, | |
453 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), | |
454 | mss.shared_clean >> 10, | |
455 | mss.shared_dirty >> 10, | |
456 | mss.private_clean >> 10, | |
457 | mss.private_dirty >> 10, | |
214e471f | 458 | mss.referenced >> 10, |
b40d4f84 | 459 | mss.anonymous >> 10, |
08fba699 | 460 | mss.swap >> 10, |
3340289d | 461 | vma_kernel_pagesize(vma) >> 10, |
2d90508f NK |
462 | vma_mmu_pagesize(vma) >> 10, |
463 | (vma->vm_flags & VM_LOCKED) ? | |
464 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); | |
4752c369 | 465 | |
7c88db0c JK |
466 | if (m->count < m->size) /* vma is copied successfully */ |
467 | m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0; | |
468 | return 0; | |
e070ad49 ML |
469 | } |
470 | ||
03a44825 | 471 | static const struct seq_operations proc_pid_smaps_op = { |
a6198797 MM |
472 | .start = m_start, |
473 | .next = m_next, | |
474 | .stop = m_stop, | |
475 | .show = show_smap | |
476 | }; | |
477 | ||
478 | static int smaps_open(struct inode *inode, struct file *file) | |
479 | { | |
480 | return do_maps_open(inode, file, &proc_pid_smaps_op); | |
481 | } | |
482 | ||
483 | const struct file_operations proc_smaps_operations = { | |
484 | .open = smaps_open, | |
485 | .read = seq_read, | |
486 | .llseek = seq_lseek, | |
487 | .release = seq_release_private, | |
488 | }; | |
489 | ||
490 | static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, | |
2165009b | 491 | unsigned long end, struct mm_walk *walk) |
a6198797 | 492 | { |
2165009b | 493 | struct vm_area_struct *vma = walk->private; |
a6198797 MM |
494 | pte_t *pte, ptent; |
495 | spinlock_t *ptl; | |
496 | struct page *page; | |
497 | ||
03319327 DH |
498 | split_huge_page_pmd(walk->mm, pmd); |
499 | ||
a6198797 MM |
500 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
501 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
502 | ptent = *pte; | |
503 | if (!pte_present(ptent)) | |
504 | continue; | |
505 | ||
506 | page = vm_normal_page(vma, addr, ptent); | |
507 | if (!page) | |
508 | continue; | |
509 | ||
510 | /* Clear accessed and referenced bits. */ | |
511 | ptep_test_and_clear_young(vma, addr, pte); | |
512 | ClearPageReferenced(page); | |
513 | } | |
514 | pte_unmap_unlock(pte - 1, ptl); | |
515 | cond_resched(); | |
516 | return 0; | |
517 | } | |
518 | ||
398499d5 MB |
519 | #define CLEAR_REFS_ALL 1 |
520 | #define CLEAR_REFS_ANON 2 | |
521 | #define CLEAR_REFS_MAPPED 3 | |
522 | ||
f248dcb3 MM |
523 | static ssize_t clear_refs_write(struct file *file, const char __user *buf, |
524 | size_t count, loff_t *ppos) | |
b813e931 | 525 | { |
f248dcb3 | 526 | struct task_struct *task; |
fb92a4b0 | 527 | char buffer[PROC_NUMBUF]; |
f248dcb3 | 528 | struct mm_struct *mm; |
b813e931 | 529 | struct vm_area_struct *vma; |
fb92a4b0 | 530 | long type; |
b813e931 | 531 | |
f248dcb3 MM |
532 | memset(buffer, 0, sizeof(buffer)); |
533 | if (count > sizeof(buffer) - 1) | |
534 | count = sizeof(buffer) - 1; | |
535 | if (copy_from_user(buffer, buf, count)) | |
536 | return -EFAULT; | |
fb92a4b0 VL |
537 | if (strict_strtol(strstrip(buffer), 10, &type)) |
538 | return -EINVAL; | |
398499d5 | 539 | if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED) |
f248dcb3 | 540 | return -EINVAL; |
f248dcb3 MM |
541 | task = get_proc_task(file->f_path.dentry->d_inode); |
542 | if (!task) | |
543 | return -ESRCH; | |
544 | mm = get_task_mm(task); | |
545 | if (mm) { | |
20cbc972 AM |
546 | struct mm_walk clear_refs_walk = { |
547 | .pmd_entry = clear_refs_pte_range, | |
548 | .mm = mm, | |
549 | }; | |
f248dcb3 | 550 | down_read(&mm->mmap_sem); |
2165009b DH |
551 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
552 | clear_refs_walk.private = vma; | |
398499d5 MB |
553 | if (is_vm_hugetlb_page(vma)) |
554 | continue; | |
555 | /* | |
556 | * Writing 1 to /proc/pid/clear_refs affects all pages. | |
557 | * | |
558 | * Writing 2 to /proc/pid/clear_refs only affects | |
559 | * Anonymous pages. | |
560 | * | |
561 | * Writing 3 to /proc/pid/clear_refs only affects file | |
562 | * mapped pages. | |
563 | */ | |
564 | if (type == CLEAR_REFS_ANON && vma->vm_file) | |
565 | continue; | |
566 | if (type == CLEAR_REFS_MAPPED && !vma->vm_file) | |
567 | continue; | |
568 | walk_page_range(vma->vm_start, vma->vm_end, | |
569 | &clear_refs_walk); | |
2165009b | 570 | } |
f248dcb3 MM |
571 | flush_tlb_mm(mm); |
572 | up_read(&mm->mmap_sem); | |
573 | mmput(mm); | |
574 | } | |
575 | put_task_struct(task); | |
fb92a4b0 VL |
576 | |
577 | return count; | |
b813e931 DR |
578 | } |
579 | ||
f248dcb3 MM |
580 | const struct file_operations proc_clear_refs_operations = { |
581 | .write = clear_refs_write, | |
6038f373 | 582 | .llseek = noop_llseek, |
f248dcb3 MM |
583 | }; |
584 | ||
85863e47 | 585 | struct pagemapread { |
d82ef020 KH |
586 | int pos, len; |
587 | u64 *buffer; | |
85863e47 MM |
588 | }; |
589 | ||
f16278c6 HR |
590 | #define PM_ENTRY_BYTES sizeof(u64) |
591 | #define PM_STATUS_BITS 3 | |
592 | #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) | |
593 | #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) | |
594 | #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) | |
595 | #define PM_PSHIFT_BITS 6 | |
596 | #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) | |
597 | #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) | |
598 | #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) | |
599 | #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) | |
600 | #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) | |
601 | ||
602 | #define PM_PRESENT PM_STATUS(4LL) | |
603 | #define PM_SWAP PM_STATUS(2LL) | |
604 | #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT) | |
85863e47 MM |
605 | #define PM_END_OF_BUFFER 1 |
606 | ||
607 | static int add_to_pagemap(unsigned long addr, u64 pfn, | |
608 | struct pagemapread *pm) | |
609 | { | |
d82ef020 KH |
610 | pm->buffer[pm->pos++] = pfn; |
611 | if (pm->pos >= pm->len) | |
aae8679b | 612 | return PM_END_OF_BUFFER; |
85863e47 MM |
613 | return 0; |
614 | } | |
615 | ||
616 | static int pagemap_pte_hole(unsigned long start, unsigned long end, | |
2165009b | 617 | struct mm_walk *walk) |
85863e47 | 618 | { |
2165009b | 619 | struct pagemapread *pm = walk->private; |
85863e47 MM |
620 | unsigned long addr; |
621 | int err = 0; | |
622 | for (addr = start; addr < end; addr += PAGE_SIZE) { | |
623 | err = add_to_pagemap(addr, PM_NOT_PRESENT, pm); | |
624 | if (err) | |
625 | break; | |
626 | } | |
627 | return err; | |
628 | } | |
629 | ||
9d02dbc8 | 630 | static u64 swap_pte_to_pagemap_entry(pte_t pte) |
85863e47 MM |
631 | { |
632 | swp_entry_t e = pte_to_swp_entry(pte); | |
f16278c6 | 633 | return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT); |
85863e47 MM |
634 | } |
635 | ||
49c50342 | 636 | static u64 pte_to_pagemap_entry(pte_t pte) |
bcf8039e | 637 | { |
49c50342 | 638 | u64 pme = 0; |
bcf8039e DH |
639 | if (is_swap_pte(pte)) |
640 | pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte)) | |
641 | | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP; | |
642 | else if (pte_present(pte)) | |
643 | pme = PM_PFRAME(pte_pfn(pte)) | |
644 | | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; | |
645 | return pme; | |
646 | } | |
647 | ||
85863e47 | 648 | static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 649 | struct mm_walk *walk) |
85863e47 | 650 | { |
bcf8039e | 651 | struct vm_area_struct *vma; |
2165009b | 652 | struct pagemapread *pm = walk->private; |
85863e47 MM |
653 | pte_t *pte; |
654 | int err = 0; | |
655 | ||
03319327 DH |
656 | split_huge_page_pmd(walk->mm, pmd); |
657 | ||
bcf8039e DH |
658 | /* find the first VMA at or above 'addr' */ |
659 | vma = find_vma(walk->mm, addr); | |
85863e47 MM |
660 | for (; addr != end; addr += PAGE_SIZE) { |
661 | u64 pfn = PM_NOT_PRESENT; | |
bcf8039e DH |
662 | |
663 | /* check to see if we've left 'vma' behind | |
664 | * and need a new, higher one */ | |
665 | if (vma && (addr >= vma->vm_end)) | |
666 | vma = find_vma(walk->mm, addr); | |
667 | ||
668 | /* check that 'vma' actually covers this address, | |
669 | * and that it isn't a huge page vma */ | |
670 | if (vma && (vma->vm_start <= addr) && | |
671 | !is_vm_hugetlb_page(vma)) { | |
672 | pte = pte_offset_map(pmd, addr); | |
673 | pfn = pte_to_pagemap_entry(*pte); | |
674 | /* unmap before userspace copy */ | |
675 | pte_unmap(pte); | |
676 | } | |
85863e47 MM |
677 | err = add_to_pagemap(addr, pfn, pm); |
678 | if (err) | |
679 | return err; | |
680 | } | |
681 | ||
682 | cond_resched(); | |
683 | ||
684 | return err; | |
685 | } | |
686 | ||
1a5cb814 | 687 | #ifdef CONFIG_HUGETLB_PAGE |
5dc37642 NH |
688 | static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset) |
689 | { | |
690 | u64 pme = 0; | |
691 | if (pte_present(pte)) | |
692 | pme = PM_PFRAME(pte_pfn(pte) + offset) | |
693 | | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; | |
694 | return pme; | |
695 | } | |
696 | ||
116354d1 NH |
697 | /* This function walks within one hugetlb entry in the single call */ |
698 | static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, | |
699 | unsigned long addr, unsigned long end, | |
700 | struct mm_walk *walk) | |
5dc37642 | 701 | { |
5dc37642 | 702 | struct pagemapread *pm = walk->private; |
5dc37642 | 703 | int err = 0; |
116354d1 | 704 | u64 pfn; |
5dc37642 | 705 | |
5dc37642 | 706 | for (; addr != end; addr += PAGE_SIZE) { |
116354d1 NH |
707 | int offset = (addr & ~hmask) >> PAGE_SHIFT; |
708 | pfn = huge_pte_to_pagemap_entry(*pte, offset); | |
5dc37642 NH |
709 | err = add_to_pagemap(addr, pfn, pm); |
710 | if (err) | |
711 | return err; | |
712 | } | |
713 | ||
714 | cond_resched(); | |
715 | ||
716 | return err; | |
717 | } | |
1a5cb814 | 718 | #endif /* HUGETLB_PAGE */ |
5dc37642 | 719 | |
85863e47 MM |
720 | /* |
721 | * /proc/pid/pagemap - an array mapping virtual pages to pfns | |
722 | * | |
f16278c6 HR |
723 | * For each page in the address space, this file contains one 64-bit entry |
724 | * consisting of the following: | |
725 | * | |
726 | * Bits 0-55 page frame number (PFN) if present | |
727 | * Bits 0-4 swap type if swapped | |
728 | * Bits 5-55 swap offset if swapped | |
729 | * Bits 55-60 page shift (page size = 1<<page shift) | |
730 | * Bit 61 reserved for future use | |
731 | * Bit 62 page swapped | |
732 | * Bit 63 page present | |
733 | * | |
734 | * If the page is not present but in swap, then the PFN contains an | |
735 | * encoding of the swap file number and the page's offset into the | |
736 | * swap. Unmapped pages return a null PFN. This allows determining | |
85863e47 MM |
737 | * precisely which pages are mapped (or in swap) and comparing mapped |
738 | * pages between processes. | |
739 | * | |
740 | * Efficient users of this interface will use /proc/pid/maps to | |
741 | * determine which areas of memory are actually mapped and llseek to | |
742 | * skip over unmapped regions. | |
743 | */ | |
d82ef020 | 744 | #define PAGEMAP_WALK_SIZE (PMD_SIZE) |
ea251c1d | 745 | #define PAGEMAP_WALK_MASK (PMD_MASK) |
85863e47 MM |
746 | static ssize_t pagemap_read(struct file *file, char __user *buf, |
747 | size_t count, loff_t *ppos) | |
748 | { | |
749 | struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); | |
85863e47 MM |
750 | struct mm_struct *mm; |
751 | struct pagemapread pm; | |
85863e47 | 752 | int ret = -ESRCH; |
ee1e6ab6 | 753 | struct mm_walk pagemap_walk = {}; |
5d7e0d2b AM |
754 | unsigned long src; |
755 | unsigned long svpfn; | |
756 | unsigned long start_vaddr; | |
757 | unsigned long end_vaddr; | |
d82ef020 | 758 | int copied = 0; |
85863e47 MM |
759 | |
760 | if (!task) | |
761 | goto out; | |
762 | ||
763 | ret = -EACCES; | |
006ebb40 | 764 | if (!ptrace_may_access(task, PTRACE_MODE_READ)) |
fb39380b | 765 | goto out_task; |
85863e47 MM |
766 | |
767 | ret = -EINVAL; | |
768 | /* file position must be aligned */ | |
aae8679b | 769 | if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) |
fb39380b | 770 | goto out_task; |
85863e47 MM |
771 | |
772 | ret = 0; | |
08161786 VM |
773 | |
774 | if (!count) | |
775 | goto out_task; | |
776 | ||
85863e47 MM |
777 | mm = get_task_mm(task); |
778 | if (!mm) | |
fb39380b | 779 | goto out_task; |
85863e47 | 780 | |
d82ef020 KH |
781 | pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); |
782 | pm.buffer = kmalloc(pm.len, GFP_TEMPORARY); | |
5d7e0d2b | 783 | ret = -ENOMEM; |
d82ef020 | 784 | if (!pm.buffer) |
fb39380b | 785 | goto out_mm; |
85863e47 | 786 | |
5d7e0d2b AM |
787 | pagemap_walk.pmd_entry = pagemap_pte_range; |
788 | pagemap_walk.pte_hole = pagemap_pte_hole; | |
1a5cb814 | 789 | #ifdef CONFIG_HUGETLB_PAGE |
5dc37642 | 790 | pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; |
1a5cb814 | 791 | #endif |
5d7e0d2b AM |
792 | pagemap_walk.mm = mm; |
793 | pagemap_walk.private = ± | |
794 | ||
795 | src = *ppos; | |
796 | svpfn = src / PM_ENTRY_BYTES; | |
797 | start_vaddr = svpfn << PAGE_SHIFT; | |
798 | end_vaddr = TASK_SIZE_OF(task); | |
799 | ||
800 | /* watch out for wraparound */ | |
801 | if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) | |
802 | start_vaddr = end_vaddr; | |
803 | ||
804 | /* | |
805 | * The odds are that this will stop walking way | |
806 | * before end_vaddr, because the length of the | |
807 | * user buffer is tracked in "pm", and the walk | |
808 | * will stop when we hit the end of the buffer. | |
809 | */ | |
d82ef020 KH |
810 | ret = 0; |
811 | while (count && (start_vaddr < end_vaddr)) { | |
812 | int len; | |
813 | unsigned long end; | |
814 | ||
815 | pm.pos = 0; | |
ea251c1d | 816 | end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; |
d82ef020 KH |
817 | /* overflow ? */ |
818 | if (end < start_vaddr || end > end_vaddr) | |
819 | end = end_vaddr; | |
820 | down_read(&mm->mmap_sem); | |
821 | ret = walk_page_range(start_vaddr, end, &pagemap_walk); | |
822 | up_read(&mm->mmap_sem); | |
823 | start_vaddr = end; | |
824 | ||
825 | len = min(count, PM_ENTRY_BYTES * pm.pos); | |
309361e0 | 826 | if (copy_to_user(buf, pm.buffer, len)) { |
d82ef020 KH |
827 | ret = -EFAULT; |
828 | goto out_free; | |
829 | } | |
830 | copied += len; | |
831 | buf += len; | |
832 | count -= len; | |
85863e47 | 833 | } |
d82ef020 KH |
834 | *ppos += copied; |
835 | if (!ret || ret == PM_END_OF_BUFFER) | |
836 | ret = copied; | |
837 | ||
85863e47 | 838 | out_free: |
d82ef020 | 839 | kfree(pm.buffer); |
fb39380b MT |
840 | out_mm: |
841 | mmput(mm); | |
85863e47 MM |
842 | out_task: |
843 | put_task_struct(task); | |
844 | out: | |
845 | return ret; | |
846 | } | |
847 | ||
848 | const struct file_operations proc_pagemap_operations = { | |
849 | .llseek = mem_lseek, /* borrow this */ | |
850 | .read = pagemap_read, | |
851 | }; | |
1e883281 | 852 | #endif /* CONFIG_PROC_PAGE_MONITOR */ |
85863e47 | 853 | |
6e21c8f1 | 854 | #ifdef CONFIG_NUMA |
1a75a6c8 | 855 | extern int show_numa_map(struct seq_file *m, void *v); |
6e21c8f1 | 856 | |
03a44825 | 857 | static const struct seq_operations proc_pid_numa_maps_op = { |
1a75a6c8 CL |
858 | .start = m_start, |
859 | .next = m_next, | |
860 | .stop = m_stop, | |
3bbfe059 | 861 | .show = show_numa_map, |
6e21c8f1 | 862 | }; |
662795de EB |
863 | |
864 | static int numa_maps_open(struct inode *inode, struct file *file) | |
865 | { | |
866 | return do_maps_open(inode, file, &proc_pid_numa_maps_op); | |
867 | } | |
868 | ||
00977a59 | 869 | const struct file_operations proc_numa_maps_operations = { |
662795de EB |
870 | .open = numa_maps_open, |
871 | .read = seq_read, | |
872 | .llseek = seq_lseek, | |
99f89551 | 873 | .release = seq_release_private, |
662795de | 874 | }; |
6e21c8f1 | 875 | #endif |