]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | #include <linux/mm.h> |
2 | #include <linux/hugetlb.h> | |
3 | #include <linux/mount.h> | |
4 | #include <linux/seq_file.h> | |
e070ad49 | 5 | #include <linux/highmem.h> |
5096add8 | 6 | #include <linux/ptrace.h> |
5a0e3ad6 | 7 | #include <linux/slab.h> |
6e21c8f1 CL |
8 | #include <linux/pagemap.h> |
9 | #include <linux/mempolicy.h> | |
85863e47 MM |
10 | #include <linux/swap.h> |
11 | #include <linux/swapops.h> | |
e070ad49 | 12 | |
1da177e4 LT |
13 | #include <asm/elf.h> |
14 | #include <asm/uaccess.h> | |
e070ad49 | 15 | #include <asm/tlbflush.h> |
1da177e4 LT |
16 | #include "internal.h" |
17 | ||
df5f8314 | 18 | void task_mem(struct seq_file *m, struct mm_struct *mm) |
1da177e4 | 19 | { |
b084d435 | 20 | unsigned long data, text, lib, swap; |
365e9c87 HD |
21 | unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; |
22 | ||
23 | /* | |
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. | |
29 | */ | |
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; | |
1da177e4 LT |
36 | |
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; | |
b084d435 | 40 | swap = get_mm_counter(mm, MM_SWAPENTS); |
df5f8314 | 41 | seq_printf(m, |
365e9c87 | 42 | "VmPeak:\t%8lu kB\n" |
1da177e4 LT |
43 | "VmSize:\t%8lu kB\n" |
44 | "VmLck:\t%8lu kB\n" | |
365e9c87 | 45 | "VmHWM:\t%8lu kB\n" |
1da177e4 LT |
46 | "VmRSS:\t%8lu kB\n" |
47 | "VmData:\t%8lu kB\n" | |
48 | "VmStk:\t%8lu kB\n" | |
49 | "VmExe:\t%8lu kB\n" | |
50 | "VmLib:\t%8lu kB\n" | |
b084d435 KH |
51 | "VmPTE:\t%8lu kB\n" |
52 | "VmSwap:\t%8lu kB\n", | |
365e9c87 HD |
53 | hiwater_vm << (PAGE_SHIFT-10), |
54 | (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10), | |
1da177e4 | 55 | mm->locked_vm << (PAGE_SHIFT-10), |
365e9c87 HD |
56 | hiwater_rss << (PAGE_SHIFT-10), |
57 | total_rss << (PAGE_SHIFT-10), | |
1da177e4 LT |
58 | data << (PAGE_SHIFT-10), |
59 | mm->stack_vm << (PAGE_SHIFT-10), text, lib, | |
b084d435 KH |
60 | (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10, |
61 | swap << (PAGE_SHIFT-10)); | |
1da177e4 LT |
62 | } |
63 | ||
64 | unsigned long task_vsize(struct mm_struct *mm) | |
65 | { | |
66 | return PAGE_SIZE * mm->total_vm; | |
67 | } | |
68 | ||
69 | int task_statm(struct mm_struct *mm, int *shared, int *text, | |
70 | int *data, int *resident) | |
71 | { | |
d559db08 | 72 | *shared = get_mm_counter(mm, MM_FILEPAGES); |
1da177e4 LT |
73 | *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) |
74 | >> PAGE_SHIFT; | |
75 | *data = mm->total_vm - mm->shared_vm; | |
d559db08 | 76 | *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); |
1da177e4 LT |
77 | return mm->total_vm; |
78 | } | |
79 | ||
1da177e4 LT |
80 | static void pad_len_spaces(struct seq_file *m, int len) |
81 | { | |
82 | len = 25 + sizeof(void*) * 6 - len; | |
83 | if (len < 1) | |
84 | len = 1; | |
85 | seq_printf(m, "%*c", len, ' '); | |
86 | } | |
87 | ||
a6198797 MM |
88 | static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma) |
89 | { | |
90 | if (vma && vma != priv->tail_vma) { | |
91 | struct mm_struct *mm = vma->vm_mm; | |
92 | up_read(&mm->mmap_sem); | |
93 | mmput(mm); | |
94 | } | |
95 | } | |
ec4dd3eb | 96 | |
a6198797 | 97 | static void *m_start(struct seq_file *m, loff_t *pos) |
e070ad49 | 98 | { |
a6198797 MM |
99 | struct proc_maps_private *priv = m->private; |
100 | unsigned long last_addr = m->version; | |
101 | struct mm_struct *mm; | |
102 | struct vm_area_struct *vma, *tail_vma = NULL; | |
103 | loff_t l = *pos; | |
104 | ||
105 | /* Clear the per syscall fields in priv */ | |
106 | priv->task = NULL; | |
107 | priv->tail_vma = NULL; | |
108 | ||
109 | /* | |
110 | * We remember last_addr rather than next_addr to hit with | |
111 | * mmap_cache most of the time. We have zero last_addr at | |
112 | * the beginning and also after lseek. We will have -1 last_addr | |
113 | * after the end of the vmas. | |
114 | */ | |
115 | ||
116 | if (last_addr == -1UL) | |
117 | return NULL; | |
118 | ||
119 | priv->task = get_pid_task(priv->pid, PIDTYPE_PID); | |
120 | if (!priv->task) | |
121 | return NULL; | |
122 | ||
123 | mm = mm_for_maps(priv->task); | |
124 | if (!mm) | |
125 | return NULL; | |
00f89d21 | 126 | down_read(&mm->mmap_sem); |
a6198797 MM |
127 | |
128 | tail_vma = get_gate_vma(priv->task); | |
129 | priv->tail_vma = tail_vma; | |
130 | ||
131 | /* Start with last addr hint */ | |
132 | vma = find_vma(mm, last_addr); | |
133 | if (last_addr && vma) { | |
134 | vma = vma->vm_next; | |
135 | goto out; | |
136 | } | |
137 | ||
138 | /* | |
139 | * Check the vma index is within the range and do | |
140 | * sequential scan until m_index. | |
141 | */ | |
142 | vma = NULL; | |
143 | if ((unsigned long)l < mm->map_count) { | |
144 | vma = mm->mmap; | |
145 | while (l-- && vma) | |
146 | vma = vma->vm_next; | |
147 | goto out; | |
148 | } | |
149 | ||
150 | if (l != mm->map_count) | |
151 | tail_vma = NULL; /* After gate vma */ | |
152 | ||
153 | out: | |
154 | if (vma) | |
155 | return vma; | |
156 | ||
157 | /* End of vmas has been reached */ | |
158 | m->version = (tail_vma != NULL)? 0: -1UL; | |
159 | up_read(&mm->mmap_sem); | |
160 | mmput(mm); | |
161 | return tail_vma; | |
162 | } | |
163 | ||
164 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
165 | { | |
166 | struct proc_maps_private *priv = m->private; | |
167 | struct vm_area_struct *vma = v; | |
168 | struct vm_area_struct *tail_vma = priv->tail_vma; | |
169 | ||
170 | (*pos)++; | |
171 | if (vma && (vma != tail_vma) && vma->vm_next) | |
172 | return vma->vm_next; | |
173 | vma_stop(priv, vma); | |
174 | return (vma != tail_vma)? tail_vma: NULL; | |
175 | } | |
176 | ||
177 | static void m_stop(struct seq_file *m, void *v) | |
178 | { | |
179 | struct proc_maps_private *priv = m->private; | |
180 | struct vm_area_struct *vma = v; | |
181 | ||
182 | vma_stop(priv, vma); | |
183 | if (priv->task) | |
184 | put_task_struct(priv->task); | |
185 | } | |
186 | ||
187 | static int do_maps_open(struct inode *inode, struct file *file, | |
03a44825 | 188 | const struct seq_operations *ops) |
a6198797 MM |
189 | { |
190 | struct proc_maps_private *priv; | |
191 | int ret = -ENOMEM; | |
192 | priv = kzalloc(sizeof(*priv), GFP_KERNEL); | |
193 | if (priv) { | |
194 | priv->pid = proc_pid(inode); | |
195 | ret = seq_open(file, ops); | |
196 | if (!ret) { | |
197 | struct seq_file *m = file->private_data; | |
198 | m->private = priv; | |
199 | } else { | |
200 | kfree(priv); | |
201 | } | |
202 | } | |
203 | return ret; | |
204 | } | |
e070ad49 | 205 | |
7c88db0c | 206 | static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma) |
1da177e4 | 207 | { |
e070ad49 ML |
208 | struct mm_struct *mm = vma->vm_mm; |
209 | struct file *file = vma->vm_file; | |
210 | int flags = vma->vm_flags; | |
1da177e4 | 211 | unsigned long ino = 0; |
6260a4b0 | 212 | unsigned long long pgoff = 0; |
d7824370 | 213 | unsigned long start; |
1da177e4 LT |
214 | dev_t dev = 0; |
215 | int len; | |
216 | ||
217 | if (file) { | |
2fddfeef | 218 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 LT |
219 | dev = inode->i_sb->s_dev; |
220 | ino = inode->i_ino; | |
6260a4b0 | 221 | pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; |
1da177e4 LT |
222 | } |
223 | ||
d7824370 LT |
224 | /* We don't show the stack guard page in /proc/maps */ |
225 | start = vma->vm_start; | |
226 | if (vma->vm_flags & VM_GROWSDOWN) | |
39aa3cb3 SB |
227 | if (!vma_stack_continue(vma->vm_prev, vma->vm_start)) |
228 | start += PAGE_SIZE; | |
d7824370 | 229 | |
1804dc6e | 230 | seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n", |
d7824370 | 231 | start, |
e070ad49 | 232 | vma->vm_end, |
1da177e4 LT |
233 | flags & VM_READ ? 'r' : '-', |
234 | flags & VM_WRITE ? 'w' : '-', | |
235 | flags & VM_EXEC ? 'x' : '-', | |
236 | flags & VM_MAYSHARE ? 's' : 'p', | |
6260a4b0 | 237 | pgoff, |
1da177e4 LT |
238 | MAJOR(dev), MINOR(dev), ino, &len); |
239 | ||
240 | /* | |
241 | * Print the dentry name for named mappings, and a | |
242 | * special [heap] marker for the heap: | |
243 | */ | |
e070ad49 | 244 | if (file) { |
1da177e4 | 245 | pad_len_spaces(m, len); |
c32c2f63 | 246 | seq_path(m, &file->f_path, "\n"); |
1da177e4 | 247 | } else { |
e6e5494c IM |
248 | const char *name = arch_vma_name(vma); |
249 | if (!name) { | |
250 | if (mm) { | |
251 | if (vma->vm_start <= mm->start_brk && | |
e070ad49 | 252 | vma->vm_end >= mm->brk) { |
e6e5494c IM |
253 | name = "[heap]"; |
254 | } else if (vma->vm_start <= mm->start_stack && | |
255 | vma->vm_end >= mm->start_stack) { | |
256 | name = "[stack]"; | |
1da177e4 | 257 | } |
e6e5494c IM |
258 | } else { |
259 | name = "[vdso]"; | |
1da177e4 | 260 | } |
e6e5494c IM |
261 | } |
262 | if (name) { | |
1da177e4 | 263 | pad_len_spaces(m, len); |
e6e5494c | 264 | seq_puts(m, name); |
1da177e4 LT |
265 | } |
266 | } | |
267 | seq_putc(m, '\n'); | |
7c88db0c JK |
268 | } |
269 | ||
270 | static int show_map(struct seq_file *m, void *v) | |
271 | { | |
272 | struct vm_area_struct *vma = v; | |
273 | struct proc_maps_private *priv = m->private; | |
274 | struct task_struct *task = priv->task; | |
275 | ||
276 | show_map_vma(m, vma); | |
e070ad49 | 277 | |
e070ad49 ML |
278 | if (m->count < m->size) /* vma is copied successfully */ |
279 | m->version = (vma != get_gate_vma(task))? vma->vm_start: 0; | |
1da177e4 LT |
280 | return 0; |
281 | } | |
282 | ||
03a44825 | 283 | static const struct seq_operations proc_pid_maps_op = { |
a6198797 MM |
284 | .start = m_start, |
285 | .next = m_next, | |
286 | .stop = m_stop, | |
287 | .show = show_map | |
288 | }; | |
289 | ||
290 | static int maps_open(struct inode *inode, struct file *file) | |
291 | { | |
292 | return do_maps_open(inode, file, &proc_pid_maps_op); | |
293 | } | |
294 | ||
295 | const struct file_operations proc_maps_operations = { | |
296 | .open = maps_open, | |
297 | .read = seq_read, | |
298 | .llseek = seq_lseek, | |
299 | .release = seq_release_private, | |
300 | }; | |
301 | ||
302 | /* | |
303 | * Proportional Set Size(PSS): my share of RSS. | |
304 | * | |
305 | * PSS of a process is the count of pages it has in memory, where each | |
306 | * page is divided by the number of processes sharing it. So if a | |
307 | * process has 1000 pages all to itself, and 1000 shared with one other | |
308 | * process, its PSS will be 1500. | |
309 | * | |
310 | * To keep (accumulated) division errors low, we adopt a 64bit | |
311 | * fixed-point pss counter to minimize division errors. So (pss >> | |
312 | * PSS_SHIFT) would be the real byte count. | |
313 | * | |
314 | * A shift of 12 before division means (assuming 4K page size): | |
315 | * - 1M 3-user-pages add up to 8KB errors; | |
316 | * - supports mapcount up to 2^24, or 16M; | |
317 | * - supports PSS up to 2^52 bytes, or 4PB. | |
318 | */ | |
319 | #define PSS_SHIFT 12 | |
320 | ||
1e883281 | 321 | #ifdef CONFIG_PROC_PAGE_MONITOR |
214e471f | 322 | struct mem_size_stats { |
a6198797 MM |
323 | struct vm_area_struct *vma; |
324 | unsigned long resident; | |
325 | unsigned long shared_clean; | |
326 | unsigned long shared_dirty; | |
327 | unsigned long private_clean; | |
328 | unsigned long private_dirty; | |
329 | unsigned long referenced; | |
214e471f | 330 | unsigned long swap; |
a6198797 MM |
331 | u64 pss; |
332 | }; | |
333 | ||
b3ae5acb | 334 | static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 335 | struct mm_walk *walk) |
e070ad49 | 336 | { |
2165009b | 337 | struct mem_size_stats *mss = walk->private; |
b3ae5acb | 338 | struct vm_area_struct *vma = mss->vma; |
e070ad49 | 339 | pte_t *pte, ptent; |
705e87c0 | 340 | spinlock_t *ptl; |
e070ad49 | 341 | struct page *page; |
ec4dd3eb | 342 | int mapcount; |
e070ad49 | 343 | |
705e87c0 | 344 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
826fad1b | 345 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
e070ad49 | 346 | ptent = *pte; |
214e471f PZ |
347 | |
348 | if (is_swap_pte(ptent)) { | |
349 | mss->swap += PAGE_SIZE; | |
350 | continue; | |
351 | } | |
352 | ||
705e87c0 | 353 | if (!pte_present(ptent)) |
e070ad49 ML |
354 | continue; |
355 | ||
ad820c5d NP |
356 | page = vm_normal_page(vma, addr, ptent); |
357 | if (!page) | |
e070ad49 ML |
358 | continue; |
359 | ||
7f53a09e | 360 | mss->resident += PAGE_SIZE; |
f79f177c DR |
361 | /* Accumulate the size in pages that have been accessed. */ |
362 | if (pte_young(ptent) || PageReferenced(page)) | |
363 | mss->referenced += PAGE_SIZE; | |
ec4dd3eb FW |
364 | mapcount = page_mapcount(page); |
365 | if (mapcount >= 2) { | |
e070ad49 ML |
366 | if (pte_dirty(ptent)) |
367 | mss->shared_dirty += PAGE_SIZE; | |
368 | else | |
369 | mss->shared_clean += PAGE_SIZE; | |
ec4dd3eb | 370 | mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; |
e070ad49 ML |
371 | } else { |
372 | if (pte_dirty(ptent)) | |
373 | mss->private_dirty += PAGE_SIZE; | |
374 | else | |
375 | mss->private_clean += PAGE_SIZE; | |
ec4dd3eb | 376 | mss->pss += (PAGE_SIZE << PSS_SHIFT); |
e070ad49 | 377 | } |
826fad1b | 378 | } |
705e87c0 HD |
379 | pte_unmap_unlock(pte - 1, ptl); |
380 | cond_resched(); | |
b3ae5acb | 381 | return 0; |
e070ad49 ML |
382 | } |
383 | ||
e070ad49 ML |
384 | static int show_smap(struct seq_file *m, void *v) |
385 | { | |
7c88db0c JK |
386 | struct proc_maps_private *priv = m->private; |
387 | struct task_struct *task = priv->task; | |
e070ad49 | 388 | struct vm_area_struct *vma = v; |
e070ad49 | 389 | struct mem_size_stats mss; |
2165009b DH |
390 | struct mm_walk smaps_walk = { |
391 | .pmd_entry = smaps_pte_range, | |
392 | .mm = vma->vm_mm, | |
393 | .private = &mss, | |
394 | }; | |
e070ad49 ML |
395 | |
396 | memset(&mss, 0, sizeof mss); | |
b3ae5acb | 397 | mss.vma = vma; |
d82ef020 | 398 | /* mmap_sem is held in m_start */ |
5ddfae16 | 399 | if (vma->vm_mm && !is_vm_hugetlb_page(vma)) |
2165009b | 400 | walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); |
4752c369 | 401 | |
7c88db0c | 402 | show_map_vma(m, vma); |
4752c369 MM |
403 | |
404 | seq_printf(m, | |
405 | "Size: %8lu kB\n" | |
406 | "Rss: %8lu kB\n" | |
407 | "Pss: %8lu kB\n" | |
408 | "Shared_Clean: %8lu kB\n" | |
409 | "Shared_Dirty: %8lu kB\n" | |
410 | "Private_Clean: %8lu kB\n" | |
411 | "Private_Dirty: %8lu kB\n" | |
214e471f | 412 | "Referenced: %8lu kB\n" |
08fba699 | 413 | "Swap: %8lu kB\n" |
3340289d MG |
414 | "KernelPageSize: %8lu kB\n" |
415 | "MMUPageSize: %8lu kB\n", | |
4752c369 MM |
416 | (vma->vm_end - vma->vm_start) >> 10, |
417 | mss.resident >> 10, | |
418 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), | |
419 | mss.shared_clean >> 10, | |
420 | mss.shared_dirty >> 10, | |
421 | mss.private_clean >> 10, | |
422 | mss.private_dirty >> 10, | |
214e471f | 423 | mss.referenced >> 10, |
08fba699 | 424 | mss.swap >> 10, |
3340289d MG |
425 | vma_kernel_pagesize(vma) >> 10, |
426 | vma_mmu_pagesize(vma) >> 10); | |
4752c369 | 427 | |
7c88db0c JK |
428 | if (m->count < m->size) /* vma is copied successfully */ |
429 | m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0; | |
430 | return 0; | |
e070ad49 ML |
431 | } |
432 | ||
03a44825 | 433 | static const struct seq_operations proc_pid_smaps_op = { |
a6198797 MM |
434 | .start = m_start, |
435 | .next = m_next, | |
436 | .stop = m_stop, | |
437 | .show = show_smap | |
438 | }; | |
439 | ||
440 | static int smaps_open(struct inode *inode, struct file *file) | |
441 | { | |
442 | return do_maps_open(inode, file, &proc_pid_smaps_op); | |
443 | } | |
444 | ||
445 | const struct file_operations proc_smaps_operations = { | |
446 | .open = smaps_open, | |
447 | .read = seq_read, | |
448 | .llseek = seq_lseek, | |
449 | .release = seq_release_private, | |
450 | }; | |
451 | ||
452 | static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, | |
2165009b | 453 | unsigned long end, struct mm_walk *walk) |
a6198797 | 454 | { |
2165009b | 455 | struct vm_area_struct *vma = walk->private; |
a6198797 MM |
456 | pte_t *pte, ptent; |
457 | spinlock_t *ptl; | |
458 | struct page *page; | |
459 | ||
460 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
461 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
462 | ptent = *pte; | |
463 | if (!pte_present(ptent)) | |
464 | continue; | |
465 | ||
466 | page = vm_normal_page(vma, addr, ptent); | |
467 | if (!page) | |
468 | continue; | |
469 | ||
470 | /* Clear accessed and referenced bits. */ | |
471 | ptep_test_and_clear_young(vma, addr, pte); | |
472 | ClearPageReferenced(page); | |
473 | } | |
474 | pte_unmap_unlock(pte - 1, ptl); | |
475 | cond_resched(); | |
476 | return 0; | |
477 | } | |
478 | ||
398499d5 MB |
479 | #define CLEAR_REFS_ALL 1 |
480 | #define CLEAR_REFS_ANON 2 | |
481 | #define CLEAR_REFS_MAPPED 3 | |
482 | ||
f248dcb3 MM |
483 | static ssize_t clear_refs_write(struct file *file, const char __user *buf, |
484 | size_t count, loff_t *ppos) | |
b813e931 | 485 | { |
f248dcb3 | 486 | struct task_struct *task; |
fb92a4b0 | 487 | char buffer[PROC_NUMBUF]; |
f248dcb3 | 488 | struct mm_struct *mm; |
b813e931 | 489 | struct vm_area_struct *vma; |
fb92a4b0 | 490 | long type; |
b813e931 | 491 | |
f248dcb3 MM |
492 | memset(buffer, 0, sizeof(buffer)); |
493 | if (count > sizeof(buffer) - 1) | |
494 | count = sizeof(buffer) - 1; | |
495 | if (copy_from_user(buffer, buf, count)) | |
496 | return -EFAULT; | |
fb92a4b0 VL |
497 | if (strict_strtol(strstrip(buffer), 10, &type)) |
498 | return -EINVAL; | |
398499d5 | 499 | if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED) |
f248dcb3 | 500 | return -EINVAL; |
f248dcb3 MM |
501 | task = get_proc_task(file->f_path.dentry->d_inode); |
502 | if (!task) | |
503 | return -ESRCH; | |
504 | mm = get_task_mm(task); | |
505 | if (mm) { | |
20cbc972 AM |
506 | struct mm_walk clear_refs_walk = { |
507 | .pmd_entry = clear_refs_pte_range, | |
508 | .mm = mm, | |
509 | }; | |
f248dcb3 | 510 | down_read(&mm->mmap_sem); |
2165009b DH |
511 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
512 | clear_refs_walk.private = vma; | |
398499d5 MB |
513 | if (is_vm_hugetlb_page(vma)) |
514 | continue; | |
515 | /* | |
516 | * Writing 1 to /proc/pid/clear_refs affects all pages. | |
517 | * | |
518 | * Writing 2 to /proc/pid/clear_refs only affects | |
519 | * Anonymous pages. | |
520 | * | |
521 | * Writing 3 to /proc/pid/clear_refs only affects file | |
522 | * mapped pages. | |
523 | */ | |
524 | if (type == CLEAR_REFS_ANON && vma->vm_file) | |
525 | continue; | |
526 | if (type == CLEAR_REFS_MAPPED && !vma->vm_file) | |
527 | continue; | |
528 | walk_page_range(vma->vm_start, vma->vm_end, | |
529 | &clear_refs_walk); | |
2165009b | 530 | } |
f248dcb3 MM |
531 | flush_tlb_mm(mm); |
532 | up_read(&mm->mmap_sem); | |
533 | mmput(mm); | |
534 | } | |
535 | put_task_struct(task); | |
fb92a4b0 VL |
536 | |
537 | return count; | |
b813e931 DR |
538 | } |
539 | ||
f248dcb3 MM |
540 | const struct file_operations proc_clear_refs_operations = { |
541 | .write = clear_refs_write, | |
542 | }; | |
543 | ||
85863e47 | 544 | struct pagemapread { |
d82ef020 KH |
545 | int pos, len; |
546 | u64 *buffer; | |
85863e47 MM |
547 | }; |
548 | ||
f16278c6 HR |
549 | #define PM_ENTRY_BYTES sizeof(u64) |
550 | #define PM_STATUS_BITS 3 | |
551 | #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) | |
552 | #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) | |
553 | #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) | |
554 | #define PM_PSHIFT_BITS 6 | |
555 | #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) | |
556 | #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) | |
557 | #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) | |
558 | #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) | |
559 | #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) | |
560 | ||
561 | #define PM_PRESENT PM_STATUS(4LL) | |
562 | #define PM_SWAP PM_STATUS(2LL) | |
563 | #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT) | |
85863e47 MM |
564 | #define PM_END_OF_BUFFER 1 |
565 | ||
566 | static int add_to_pagemap(unsigned long addr, u64 pfn, | |
567 | struct pagemapread *pm) | |
568 | { | |
d82ef020 KH |
569 | pm->buffer[pm->pos++] = pfn; |
570 | if (pm->pos >= pm->len) | |
aae8679b | 571 | return PM_END_OF_BUFFER; |
85863e47 MM |
572 | return 0; |
573 | } | |
574 | ||
575 | static int pagemap_pte_hole(unsigned long start, unsigned long end, | |
2165009b | 576 | struct mm_walk *walk) |
85863e47 | 577 | { |
2165009b | 578 | struct pagemapread *pm = walk->private; |
85863e47 MM |
579 | unsigned long addr; |
580 | int err = 0; | |
581 | for (addr = start; addr < end; addr += PAGE_SIZE) { | |
582 | err = add_to_pagemap(addr, PM_NOT_PRESENT, pm); | |
583 | if (err) | |
584 | break; | |
585 | } | |
586 | return err; | |
587 | } | |
588 | ||
9d02dbc8 | 589 | static u64 swap_pte_to_pagemap_entry(pte_t pte) |
85863e47 MM |
590 | { |
591 | swp_entry_t e = pte_to_swp_entry(pte); | |
f16278c6 | 592 | return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT); |
85863e47 MM |
593 | } |
594 | ||
49c50342 | 595 | static u64 pte_to_pagemap_entry(pte_t pte) |
bcf8039e | 596 | { |
49c50342 | 597 | u64 pme = 0; |
bcf8039e DH |
598 | if (is_swap_pte(pte)) |
599 | pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte)) | |
600 | | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP; | |
601 | else if (pte_present(pte)) | |
602 | pme = PM_PFRAME(pte_pfn(pte)) | |
603 | | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; | |
604 | return pme; | |
605 | } | |
606 | ||
85863e47 | 607 | static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 608 | struct mm_walk *walk) |
85863e47 | 609 | { |
bcf8039e | 610 | struct vm_area_struct *vma; |
2165009b | 611 | struct pagemapread *pm = walk->private; |
85863e47 MM |
612 | pte_t *pte; |
613 | int err = 0; | |
614 | ||
bcf8039e DH |
615 | /* find the first VMA at or above 'addr' */ |
616 | vma = find_vma(walk->mm, addr); | |
85863e47 MM |
617 | for (; addr != end; addr += PAGE_SIZE) { |
618 | u64 pfn = PM_NOT_PRESENT; | |
bcf8039e DH |
619 | |
620 | /* check to see if we've left 'vma' behind | |
621 | * and need a new, higher one */ | |
622 | if (vma && (addr >= vma->vm_end)) | |
623 | vma = find_vma(walk->mm, addr); | |
624 | ||
625 | /* check that 'vma' actually covers this address, | |
626 | * and that it isn't a huge page vma */ | |
627 | if (vma && (vma->vm_start <= addr) && | |
628 | !is_vm_hugetlb_page(vma)) { | |
629 | pte = pte_offset_map(pmd, addr); | |
630 | pfn = pte_to_pagemap_entry(*pte); | |
631 | /* unmap before userspace copy */ | |
632 | pte_unmap(pte); | |
633 | } | |
85863e47 MM |
634 | err = add_to_pagemap(addr, pfn, pm); |
635 | if (err) | |
636 | return err; | |
637 | } | |
638 | ||
639 | cond_resched(); | |
640 | ||
641 | return err; | |
642 | } | |
643 | ||
1a5cb814 | 644 | #ifdef CONFIG_HUGETLB_PAGE |
5dc37642 NH |
645 | static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset) |
646 | { | |
647 | u64 pme = 0; | |
648 | if (pte_present(pte)) | |
649 | pme = PM_PFRAME(pte_pfn(pte) + offset) | |
650 | | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; | |
651 | return pme; | |
652 | } | |
653 | ||
116354d1 NH |
654 | /* This function walks within one hugetlb entry in the single call */ |
655 | static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, | |
656 | unsigned long addr, unsigned long end, | |
657 | struct mm_walk *walk) | |
5dc37642 | 658 | { |
5dc37642 | 659 | struct pagemapread *pm = walk->private; |
5dc37642 | 660 | int err = 0; |
116354d1 | 661 | u64 pfn; |
5dc37642 | 662 | |
5dc37642 | 663 | for (; addr != end; addr += PAGE_SIZE) { |
116354d1 NH |
664 | int offset = (addr & ~hmask) >> PAGE_SHIFT; |
665 | pfn = huge_pte_to_pagemap_entry(*pte, offset); | |
5dc37642 NH |
666 | err = add_to_pagemap(addr, pfn, pm); |
667 | if (err) | |
668 | return err; | |
669 | } | |
670 | ||
671 | cond_resched(); | |
672 | ||
673 | return err; | |
674 | } | |
1a5cb814 | 675 | #endif /* HUGETLB_PAGE */ |
5dc37642 | 676 | |
85863e47 MM |
677 | /* |
678 | * /proc/pid/pagemap - an array mapping virtual pages to pfns | |
679 | * | |
f16278c6 HR |
680 | * For each page in the address space, this file contains one 64-bit entry |
681 | * consisting of the following: | |
682 | * | |
683 | * Bits 0-55 page frame number (PFN) if present | |
684 | * Bits 0-4 swap type if swapped | |
685 | * Bits 5-55 swap offset if swapped | |
686 | * Bits 55-60 page shift (page size = 1<<page shift) | |
687 | * Bit 61 reserved for future use | |
688 | * Bit 62 page swapped | |
689 | * Bit 63 page present | |
690 | * | |
691 | * If the page is not present but in swap, then the PFN contains an | |
692 | * encoding of the swap file number and the page's offset into the | |
693 | * swap. Unmapped pages return a null PFN. This allows determining | |
85863e47 MM |
694 | * precisely which pages are mapped (or in swap) and comparing mapped |
695 | * pages between processes. | |
696 | * | |
697 | * Efficient users of this interface will use /proc/pid/maps to | |
698 | * determine which areas of memory are actually mapped and llseek to | |
699 | * skip over unmapped regions. | |
700 | */ | |
d82ef020 | 701 | #define PAGEMAP_WALK_SIZE (PMD_SIZE) |
85863e47 MM |
702 | static ssize_t pagemap_read(struct file *file, char __user *buf, |
703 | size_t count, loff_t *ppos) | |
704 | { | |
705 | struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); | |
85863e47 MM |
706 | struct mm_struct *mm; |
707 | struct pagemapread pm; | |
85863e47 | 708 | int ret = -ESRCH; |
ee1e6ab6 | 709 | struct mm_walk pagemap_walk = {}; |
5d7e0d2b AM |
710 | unsigned long src; |
711 | unsigned long svpfn; | |
712 | unsigned long start_vaddr; | |
713 | unsigned long end_vaddr; | |
d82ef020 | 714 | int copied = 0; |
85863e47 MM |
715 | |
716 | if (!task) | |
717 | goto out; | |
718 | ||
719 | ret = -EACCES; | |
006ebb40 | 720 | if (!ptrace_may_access(task, PTRACE_MODE_READ)) |
fb39380b | 721 | goto out_task; |
85863e47 MM |
722 | |
723 | ret = -EINVAL; | |
724 | /* file position must be aligned */ | |
aae8679b | 725 | if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) |
fb39380b | 726 | goto out_task; |
85863e47 MM |
727 | |
728 | ret = 0; | |
08161786 VM |
729 | |
730 | if (!count) | |
731 | goto out_task; | |
732 | ||
85863e47 MM |
733 | mm = get_task_mm(task); |
734 | if (!mm) | |
fb39380b | 735 | goto out_task; |
85863e47 | 736 | |
d82ef020 KH |
737 | pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); |
738 | pm.buffer = kmalloc(pm.len, GFP_TEMPORARY); | |
5d7e0d2b | 739 | ret = -ENOMEM; |
d82ef020 | 740 | if (!pm.buffer) |
fb39380b | 741 | goto out_mm; |
85863e47 | 742 | |
5d7e0d2b AM |
743 | pagemap_walk.pmd_entry = pagemap_pte_range; |
744 | pagemap_walk.pte_hole = pagemap_pte_hole; | |
1a5cb814 | 745 | #ifdef CONFIG_HUGETLB_PAGE |
5dc37642 | 746 | pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; |
1a5cb814 | 747 | #endif |
5d7e0d2b AM |
748 | pagemap_walk.mm = mm; |
749 | pagemap_walk.private = ± | |
750 | ||
751 | src = *ppos; | |
752 | svpfn = src / PM_ENTRY_BYTES; | |
753 | start_vaddr = svpfn << PAGE_SHIFT; | |
754 | end_vaddr = TASK_SIZE_OF(task); | |
755 | ||
756 | /* watch out for wraparound */ | |
757 | if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) | |
758 | start_vaddr = end_vaddr; | |
759 | ||
760 | /* | |
761 | * The odds are that this will stop walking way | |
762 | * before end_vaddr, because the length of the | |
763 | * user buffer is tracked in "pm", and the walk | |
764 | * will stop when we hit the end of the buffer. | |
765 | */ | |
d82ef020 KH |
766 | ret = 0; |
767 | while (count && (start_vaddr < end_vaddr)) { | |
768 | int len; | |
769 | unsigned long end; | |
770 | ||
771 | pm.pos = 0; | |
772 | end = start_vaddr + PAGEMAP_WALK_SIZE; | |
773 | /* overflow ? */ | |
774 | if (end < start_vaddr || end > end_vaddr) | |
775 | end = end_vaddr; | |
776 | down_read(&mm->mmap_sem); | |
777 | ret = walk_page_range(start_vaddr, end, &pagemap_walk); | |
778 | up_read(&mm->mmap_sem); | |
779 | start_vaddr = end; | |
780 | ||
781 | len = min(count, PM_ENTRY_BYTES * pm.pos); | |
309361e0 | 782 | if (copy_to_user(buf, pm.buffer, len)) { |
d82ef020 KH |
783 | ret = -EFAULT; |
784 | goto out_free; | |
785 | } | |
786 | copied += len; | |
787 | buf += len; | |
788 | count -= len; | |
85863e47 | 789 | } |
d82ef020 KH |
790 | *ppos += copied; |
791 | if (!ret || ret == PM_END_OF_BUFFER) | |
792 | ret = copied; | |
793 | ||
85863e47 | 794 | out_free: |
d82ef020 | 795 | kfree(pm.buffer); |
fb39380b MT |
796 | out_mm: |
797 | mmput(mm); | |
85863e47 MM |
798 | out_task: |
799 | put_task_struct(task); | |
800 | out: | |
801 | return ret; | |
802 | } | |
803 | ||
804 | const struct file_operations proc_pagemap_operations = { | |
805 | .llseek = mem_lseek, /* borrow this */ | |
806 | .read = pagemap_read, | |
807 | }; | |
1e883281 | 808 | #endif /* CONFIG_PROC_PAGE_MONITOR */ |
85863e47 | 809 | |
6e21c8f1 | 810 | #ifdef CONFIG_NUMA |
1a75a6c8 | 811 | extern int show_numa_map(struct seq_file *m, void *v); |
6e21c8f1 | 812 | |
03a44825 | 813 | static const struct seq_operations proc_pid_numa_maps_op = { |
1a75a6c8 CL |
814 | .start = m_start, |
815 | .next = m_next, | |
816 | .stop = m_stop, | |
3bbfe059 | 817 | .show = show_numa_map, |
6e21c8f1 | 818 | }; |
662795de EB |
819 | |
820 | static int numa_maps_open(struct inode *inode, struct file *file) | |
821 | { | |
822 | return do_maps_open(inode, file, &proc_pid_numa_maps_op); | |
823 | } | |
824 | ||
00977a59 | 825 | const struct file_operations proc_numa_maps_operations = { |
662795de EB |
826 | .open = numa_maps_open, |
827 | .read = seq_read, | |
828 | .llseek = seq_lseek, | |
99f89551 | 829 | .release = seq_release_private, |
662795de | 830 | }; |
6e21c8f1 | 831 | #endif |