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; | |
b40d4f84 | 330 | unsigned long anonymous; |
214e471f | 331 | unsigned long swap; |
a6198797 MM |
332 | u64 pss; |
333 | }; | |
334 | ||
b3ae5acb | 335 | static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 336 | struct mm_walk *walk) |
e070ad49 | 337 | { |
2165009b | 338 | struct mem_size_stats *mss = walk->private; |
b3ae5acb | 339 | struct vm_area_struct *vma = mss->vma; |
e070ad49 | 340 | pte_t *pte, ptent; |
705e87c0 | 341 | spinlock_t *ptl; |
e070ad49 | 342 | struct page *page; |
ec4dd3eb | 343 | int mapcount; |
e070ad49 | 344 | |
705e87c0 | 345 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
826fad1b | 346 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
e070ad49 | 347 | ptent = *pte; |
214e471f PZ |
348 | |
349 | if (is_swap_pte(ptent)) { | |
350 | mss->swap += PAGE_SIZE; | |
351 | continue; | |
352 | } | |
353 | ||
705e87c0 | 354 | if (!pte_present(ptent)) |
e070ad49 ML |
355 | continue; |
356 | ||
ad820c5d NP |
357 | page = vm_normal_page(vma, addr, ptent); |
358 | if (!page) | |
e070ad49 ML |
359 | continue; |
360 | ||
b40d4f84 NK |
361 | if (PageAnon(page)) |
362 | mss->anonymous += PAGE_SIZE; | |
363 | ||
7f53a09e | 364 | mss->resident += PAGE_SIZE; |
f79f177c DR |
365 | /* Accumulate the size in pages that have been accessed. */ |
366 | if (pte_young(ptent) || PageReferenced(page)) | |
367 | mss->referenced += PAGE_SIZE; | |
ec4dd3eb FW |
368 | mapcount = page_mapcount(page); |
369 | if (mapcount >= 2) { | |
1c2499ae | 370 | if (pte_dirty(ptent) || PageDirty(page)) |
e070ad49 ML |
371 | mss->shared_dirty += PAGE_SIZE; |
372 | else | |
373 | mss->shared_clean += PAGE_SIZE; | |
ec4dd3eb | 374 | mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; |
e070ad49 | 375 | } else { |
1c2499ae | 376 | if (pte_dirty(ptent) || PageDirty(page)) |
e070ad49 ML |
377 | mss->private_dirty += PAGE_SIZE; |
378 | else | |
379 | mss->private_clean += PAGE_SIZE; | |
ec4dd3eb | 380 | mss->pss += (PAGE_SIZE << PSS_SHIFT); |
e070ad49 | 381 | } |
826fad1b | 382 | } |
705e87c0 HD |
383 | pte_unmap_unlock(pte - 1, ptl); |
384 | cond_resched(); | |
b3ae5acb | 385 | return 0; |
e070ad49 ML |
386 | } |
387 | ||
e070ad49 ML |
388 | static int show_smap(struct seq_file *m, void *v) |
389 | { | |
7c88db0c JK |
390 | struct proc_maps_private *priv = m->private; |
391 | struct task_struct *task = priv->task; | |
e070ad49 | 392 | struct vm_area_struct *vma = v; |
e070ad49 | 393 | struct mem_size_stats mss; |
2165009b DH |
394 | struct mm_walk smaps_walk = { |
395 | .pmd_entry = smaps_pte_range, | |
396 | .mm = vma->vm_mm, | |
397 | .private = &mss, | |
398 | }; | |
e070ad49 ML |
399 | |
400 | memset(&mss, 0, sizeof mss); | |
b3ae5acb | 401 | mss.vma = vma; |
d82ef020 | 402 | /* mmap_sem is held in m_start */ |
5ddfae16 | 403 | if (vma->vm_mm && !is_vm_hugetlb_page(vma)) |
2165009b | 404 | walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); |
4752c369 | 405 | |
7c88db0c | 406 | show_map_vma(m, vma); |
4752c369 MM |
407 | |
408 | seq_printf(m, | |
409 | "Size: %8lu kB\n" | |
410 | "Rss: %8lu kB\n" | |
411 | "Pss: %8lu kB\n" | |
412 | "Shared_Clean: %8lu kB\n" | |
413 | "Shared_Dirty: %8lu kB\n" | |
414 | "Private_Clean: %8lu kB\n" | |
415 | "Private_Dirty: %8lu kB\n" | |
214e471f | 416 | "Referenced: %8lu kB\n" |
b40d4f84 | 417 | "Anonymous: %8lu kB\n" |
08fba699 | 418 | "Swap: %8lu kB\n" |
3340289d MG |
419 | "KernelPageSize: %8lu kB\n" |
420 | "MMUPageSize: %8lu kB\n", | |
4752c369 MM |
421 | (vma->vm_end - vma->vm_start) >> 10, |
422 | mss.resident >> 10, | |
423 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), | |
424 | mss.shared_clean >> 10, | |
425 | mss.shared_dirty >> 10, | |
426 | mss.private_clean >> 10, | |
427 | mss.private_dirty >> 10, | |
214e471f | 428 | mss.referenced >> 10, |
b40d4f84 | 429 | mss.anonymous >> 10, |
08fba699 | 430 | mss.swap >> 10, |
3340289d MG |
431 | vma_kernel_pagesize(vma) >> 10, |
432 | vma_mmu_pagesize(vma) >> 10); | |
4752c369 | 433 | |
7c88db0c JK |
434 | if (m->count < m->size) /* vma is copied successfully */ |
435 | m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0; | |
436 | return 0; | |
e070ad49 ML |
437 | } |
438 | ||
03a44825 | 439 | static const struct seq_operations proc_pid_smaps_op = { |
a6198797 MM |
440 | .start = m_start, |
441 | .next = m_next, | |
442 | .stop = m_stop, | |
443 | .show = show_smap | |
444 | }; | |
445 | ||
446 | static int smaps_open(struct inode *inode, struct file *file) | |
447 | { | |
448 | return do_maps_open(inode, file, &proc_pid_smaps_op); | |
449 | } | |
450 | ||
451 | const struct file_operations proc_smaps_operations = { | |
452 | .open = smaps_open, | |
453 | .read = seq_read, | |
454 | .llseek = seq_lseek, | |
455 | .release = seq_release_private, | |
456 | }; | |
457 | ||
458 | static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, | |
2165009b | 459 | unsigned long end, struct mm_walk *walk) |
a6198797 | 460 | { |
2165009b | 461 | struct vm_area_struct *vma = walk->private; |
a6198797 MM |
462 | pte_t *pte, ptent; |
463 | spinlock_t *ptl; | |
464 | struct page *page; | |
465 | ||
466 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
467 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
468 | ptent = *pte; | |
469 | if (!pte_present(ptent)) | |
470 | continue; | |
471 | ||
472 | page = vm_normal_page(vma, addr, ptent); | |
473 | if (!page) | |
474 | continue; | |
475 | ||
476 | /* Clear accessed and referenced bits. */ | |
477 | ptep_test_and_clear_young(vma, addr, pte); | |
478 | ClearPageReferenced(page); | |
479 | } | |
480 | pte_unmap_unlock(pte - 1, ptl); | |
481 | cond_resched(); | |
482 | return 0; | |
483 | } | |
484 | ||
398499d5 MB |
485 | #define CLEAR_REFS_ALL 1 |
486 | #define CLEAR_REFS_ANON 2 | |
487 | #define CLEAR_REFS_MAPPED 3 | |
488 | ||
f248dcb3 MM |
489 | static ssize_t clear_refs_write(struct file *file, const char __user *buf, |
490 | size_t count, loff_t *ppos) | |
b813e931 | 491 | { |
f248dcb3 | 492 | struct task_struct *task; |
fb92a4b0 | 493 | char buffer[PROC_NUMBUF]; |
f248dcb3 | 494 | struct mm_struct *mm; |
b813e931 | 495 | struct vm_area_struct *vma; |
fb92a4b0 | 496 | long type; |
b813e931 | 497 | |
f248dcb3 MM |
498 | memset(buffer, 0, sizeof(buffer)); |
499 | if (count > sizeof(buffer) - 1) | |
500 | count = sizeof(buffer) - 1; | |
501 | if (copy_from_user(buffer, buf, count)) | |
502 | return -EFAULT; | |
fb92a4b0 VL |
503 | if (strict_strtol(strstrip(buffer), 10, &type)) |
504 | return -EINVAL; | |
398499d5 | 505 | if (type < CLEAR_REFS_ALL || type > CLEAR_REFS_MAPPED) |
f248dcb3 | 506 | return -EINVAL; |
f248dcb3 MM |
507 | task = get_proc_task(file->f_path.dentry->d_inode); |
508 | if (!task) | |
509 | return -ESRCH; | |
510 | mm = get_task_mm(task); | |
511 | if (mm) { | |
20cbc972 AM |
512 | struct mm_walk clear_refs_walk = { |
513 | .pmd_entry = clear_refs_pte_range, | |
514 | .mm = mm, | |
515 | }; | |
f248dcb3 | 516 | down_read(&mm->mmap_sem); |
2165009b DH |
517 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
518 | clear_refs_walk.private = vma; | |
398499d5 MB |
519 | if (is_vm_hugetlb_page(vma)) |
520 | continue; | |
521 | /* | |
522 | * Writing 1 to /proc/pid/clear_refs affects all pages. | |
523 | * | |
524 | * Writing 2 to /proc/pid/clear_refs only affects | |
525 | * Anonymous pages. | |
526 | * | |
527 | * Writing 3 to /proc/pid/clear_refs only affects file | |
528 | * mapped pages. | |
529 | */ | |
530 | if (type == CLEAR_REFS_ANON && vma->vm_file) | |
531 | continue; | |
532 | if (type == CLEAR_REFS_MAPPED && !vma->vm_file) | |
533 | continue; | |
534 | walk_page_range(vma->vm_start, vma->vm_end, | |
535 | &clear_refs_walk); | |
2165009b | 536 | } |
f248dcb3 MM |
537 | flush_tlb_mm(mm); |
538 | up_read(&mm->mmap_sem); | |
539 | mmput(mm); | |
540 | } | |
541 | put_task_struct(task); | |
fb92a4b0 VL |
542 | |
543 | return count; | |
b813e931 DR |
544 | } |
545 | ||
f248dcb3 MM |
546 | const struct file_operations proc_clear_refs_operations = { |
547 | .write = clear_refs_write, | |
6038f373 | 548 | .llseek = noop_llseek, |
f248dcb3 MM |
549 | }; |
550 | ||
85863e47 | 551 | struct pagemapread { |
d82ef020 KH |
552 | int pos, len; |
553 | u64 *buffer; | |
85863e47 MM |
554 | }; |
555 | ||
f16278c6 HR |
556 | #define PM_ENTRY_BYTES sizeof(u64) |
557 | #define PM_STATUS_BITS 3 | |
558 | #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) | |
559 | #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) | |
560 | #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) | |
561 | #define PM_PSHIFT_BITS 6 | |
562 | #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) | |
563 | #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) | |
564 | #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) | |
565 | #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) | |
566 | #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) | |
567 | ||
568 | #define PM_PRESENT PM_STATUS(4LL) | |
569 | #define PM_SWAP PM_STATUS(2LL) | |
570 | #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT) | |
85863e47 MM |
571 | #define PM_END_OF_BUFFER 1 |
572 | ||
573 | static int add_to_pagemap(unsigned long addr, u64 pfn, | |
574 | struct pagemapread *pm) | |
575 | { | |
d82ef020 KH |
576 | pm->buffer[pm->pos++] = pfn; |
577 | if (pm->pos >= pm->len) | |
aae8679b | 578 | return PM_END_OF_BUFFER; |
85863e47 MM |
579 | return 0; |
580 | } | |
581 | ||
582 | static int pagemap_pte_hole(unsigned long start, unsigned long end, | |
2165009b | 583 | struct mm_walk *walk) |
85863e47 | 584 | { |
2165009b | 585 | struct pagemapread *pm = walk->private; |
85863e47 MM |
586 | unsigned long addr; |
587 | int err = 0; | |
588 | for (addr = start; addr < end; addr += PAGE_SIZE) { | |
589 | err = add_to_pagemap(addr, PM_NOT_PRESENT, pm); | |
590 | if (err) | |
591 | break; | |
592 | } | |
593 | return err; | |
594 | } | |
595 | ||
9d02dbc8 | 596 | static u64 swap_pte_to_pagemap_entry(pte_t pte) |
85863e47 MM |
597 | { |
598 | swp_entry_t e = pte_to_swp_entry(pte); | |
f16278c6 | 599 | return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT); |
85863e47 MM |
600 | } |
601 | ||
49c50342 | 602 | static u64 pte_to_pagemap_entry(pte_t pte) |
bcf8039e | 603 | { |
49c50342 | 604 | u64 pme = 0; |
bcf8039e DH |
605 | if (is_swap_pte(pte)) |
606 | pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte)) | |
607 | | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP; | |
608 | else if (pte_present(pte)) | |
609 | pme = PM_PFRAME(pte_pfn(pte)) | |
610 | | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; | |
611 | return pme; | |
612 | } | |
613 | ||
85863e47 | 614 | static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 615 | struct mm_walk *walk) |
85863e47 | 616 | { |
bcf8039e | 617 | struct vm_area_struct *vma; |
2165009b | 618 | struct pagemapread *pm = walk->private; |
85863e47 MM |
619 | pte_t *pte; |
620 | int err = 0; | |
621 | ||
bcf8039e DH |
622 | /* find the first VMA at or above 'addr' */ |
623 | vma = find_vma(walk->mm, addr); | |
85863e47 MM |
624 | for (; addr != end; addr += PAGE_SIZE) { |
625 | u64 pfn = PM_NOT_PRESENT; | |
bcf8039e DH |
626 | |
627 | /* check to see if we've left 'vma' behind | |
628 | * and need a new, higher one */ | |
629 | if (vma && (addr >= vma->vm_end)) | |
630 | vma = find_vma(walk->mm, addr); | |
631 | ||
632 | /* check that 'vma' actually covers this address, | |
633 | * and that it isn't a huge page vma */ | |
634 | if (vma && (vma->vm_start <= addr) && | |
635 | !is_vm_hugetlb_page(vma)) { | |
636 | pte = pte_offset_map(pmd, addr); | |
637 | pfn = pte_to_pagemap_entry(*pte); | |
638 | /* unmap before userspace copy */ | |
639 | pte_unmap(pte); | |
640 | } | |
85863e47 MM |
641 | err = add_to_pagemap(addr, pfn, pm); |
642 | if (err) | |
643 | return err; | |
644 | } | |
645 | ||
646 | cond_resched(); | |
647 | ||
648 | return err; | |
649 | } | |
650 | ||
1a5cb814 | 651 | #ifdef CONFIG_HUGETLB_PAGE |
5dc37642 NH |
652 | static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset) |
653 | { | |
654 | u64 pme = 0; | |
655 | if (pte_present(pte)) | |
656 | pme = PM_PFRAME(pte_pfn(pte) + offset) | |
657 | | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; | |
658 | return pme; | |
659 | } | |
660 | ||
116354d1 NH |
661 | /* This function walks within one hugetlb entry in the single call */ |
662 | static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, | |
663 | unsigned long addr, unsigned long end, | |
664 | struct mm_walk *walk) | |
5dc37642 | 665 | { |
5dc37642 | 666 | struct pagemapread *pm = walk->private; |
5dc37642 | 667 | int err = 0; |
116354d1 | 668 | u64 pfn; |
5dc37642 | 669 | |
5dc37642 | 670 | for (; addr != end; addr += PAGE_SIZE) { |
116354d1 NH |
671 | int offset = (addr & ~hmask) >> PAGE_SHIFT; |
672 | pfn = huge_pte_to_pagemap_entry(*pte, offset); | |
5dc37642 NH |
673 | err = add_to_pagemap(addr, pfn, pm); |
674 | if (err) | |
675 | return err; | |
676 | } | |
677 | ||
678 | cond_resched(); | |
679 | ||
680 | return err; | |
681 | } | |
1a5cb814 | 682 | #endif /* HUGETLB_PAGE */ |
5dc37642 | 683 | |
85863e47 MM |
684 | /* |
685 | * /proc/pid/pagemap - an array mapping virtual pages to pfns | |
686 | * | |
f16278c6 HR |
687 | * For each page in the address space, this file contains one 64-bit entry |
688 | * consisting of the following: | |
689 | * | |
690 | * Bits 0-55 page frame number (PFN) if present | |
691 | * Bits 0-4 swap type if swapped | |
692 | * Bits 5-55 swap offset if swapped | |
693 | * Bits 55-60 page shift (page size = 1<<page shift) | |
694 | * Bit 61 reserved for future use | |
695 | * Bit 62 page swapped | |
696 | * Bit 63 page present | |
697 | * | |
698 | * If the page is not present but in swap, then the PFN contains an | |
699 | * encoding of the swap file number and the page's offset into the | |
700 | * swap. Unmapped pages return a null PFN. This allows determining | |
85863e47 MM |
701 | * precisely which pages are mapped (or in swap) and comparing mapped |
702 | * pages between processes. | |
703 | * | |
704 | * Efficient users of this interface will use /proc/pid/maps to | |
705 | * determine which areas of memory are actually mapped and llseek to | |
706 | * skip over unmapped regions. | |
707 | */ | |
d82ef020 | 708 | #define PAGEMAP_WALK_SIZE (PMD_SIZE) |
ea251c1d | 709 | #define PAGEMAP_WALK_MASK (PMD_MASK) |
85863e47 MM |
710 | static ssize_t pagemap_read(struct file *file, char __user *buf, |
711 | size_t count, loff_t *ppos) | |
712 | { | |
713 | struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); | |
85863e47 MM |
714 | struct mm_struct *mm; |
715 | struct pagemapread pm; | |
85863e47 | 716 | int ret = -ESRCH; |
ee1e6ab6 | 717 | struct mm_walk pagemap_walk = {}; |
5d7e0d2b AM |
718 | unsigned long src; |
719 | unsigned long svpfn; | |
720 | unsigned long start_vaddr; | |
721 | unsigned long end_vaddr; | |
d82ef020 | 722 | int copied = 0; |
85863e47 MM |
723 | |
724 | if (!task) | |
725 | goto out; | |
726 | ||
727 | ret = -EACCES; | |
006ebb40 | 728 | if (!ptrace_may_access(task, PTRACE_MODE_READ)) |
fb39380b | 729 | goto out_task; |
85863e47 MM |
730 | |
731 | ret = -EINVAL; | |
732 | /* file position must be aligned */ | |
aae8679b | 733 | if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) |
fb39380b | 734 | goto out_task; |
85863e47 MM |
735 | |
736 | ret = 0; | |
08161786 VM |
737 | |
738 | if (!count) | |
739 | goto out_task; | |
740 | ||
85863e47 MM |
741 | mm = get_task_mm(task); |
742 | if (!mm) | |
fb39380b | 743 | goto out_task; |
85863e47 | 744 | |
d82ef020 KH |
745 | pm.len = PM_ENTRY_BYTES * (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); |
746 | pm.buffer = kmalloc(pm.len, GFP_TEMPORARY); | |
5d7e0d2b | 747 | ret = -ENOMEM; |
d82ef020 | 748 | if (!pm.buffer) |
fb39380b | 749 | goto out_mm; |
85863e47 | 750 | |
5d7e0d2b AM |
751 | pagemap_walk.pmd_entry = pagemap_pte_range; |
752 | pagemap_walk.pte_hole = pagemap_pte_hole; | |
1a5cb814 | 753 | #ifdef CONFIG_HUGETLB_PAGE |
5dc37642 | 754 | pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; |
1a5cb814 | 755 | #endif |
5d7e0d2b AM |
756 | pagemap_walk.mm = mm; |
757 | pagemap_walk.private = ± | |
758 | ||
759 | src = *ppos; | |
760 | svpfn = src / PM_ENTRY_BYTES; | |
761 | start_vaddr = svpfn << PAGE_SHIFT; | |
762 | end_vaddr = TASK_SIZE_OF(task); | |
763 | ||
764 | /* watch out for wraparound */ | |
765 | if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) | |
766 | start_vaddr = end_vaddr; | |
767 | ||
768 | /* | |
769 | * The odds are that this will stop walking way | |
770 | * before end_vaddr, because the length of the | |
771 | * user buffer is tracked in "pm", and the walk | |
772 | * will stop when we hit the end of the buffer. | |
773 | */ | |
d82ef020 KH |
774 | ret = 0; |
775 | while (count && (start_vaddr < end_vaddr)) { | |
776 | int len; | |
777 | unsigned long end; | |
778 | ||
779 | pm.pos = 0; | |
ea251c1d | 780 | end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; |
d82ef020 KH |
781 | /* overflow ? */ |
782 | if (end < start_vaddr || end > end_vaddr) | |
783 | end = end_vaddr; | |
784 | down_read(&mm->mmap_sem); | |
785 | ret = walk_page_range(start_vaddr, end, &pagemap_walk); | |
786 | up_read(&mm->mmap_sem); | |
787 | start_vaddr = end; | |
788 | ||
789 | len = min(count, PM_ENTRY_BYTES * pm.pos); | |
309361e0 | 790 | if (copy_to_user(buf, pm.buffer, len)) { |
d82ef020 KH |
791 | ret = -EFAULT; |
792 | goto out_free; | |
793 | } | |
794 | copied += len; | |
795 | buf += len; | |
796 | count -= len; | |
85863e47 | 797 | } |
d82ef020 KH |
798 | *ppos += copied; |
799 | if (!ret || ret == PM_END_OF_BUFFER) | |
800 | ret = copied; | |
801 | ||
85863e47 | 802 | out_free: |
d82ef020 | 803 | kfree(pm.buffer); |
fb39380b MT |
804 | out_mm: |
805 | mmput(mm); | |
85863e47 MM |
806 | out_task: |
807 | put_task_struct(task); | |
808 | out: | |
809 | return ret; | |
810 | } | |
811 | ||
812 | const struct file_operations proc_pagemap_operations = { | |
813 | .llseek = mem_lseek, /* borrow this */ | |
814 | .read = pagemap_read, | |
815 | }; | |
1e883281 | 816 | #endif /* CONFIG_PROC_PAGE_MONITOR */ |
85863e47 | 817 | |
6e21c8f1 | 818 | #ifdef CONFIG_NUMA |
1a75a6c8 | 819 | extern int show_numa_map(struct seq_file *m, void *v); |
6e21c8f1 | 820 | |
03a44825 | 821 | static const struct seq_operations proc_pid_numa_maps_op = { |
1a75a6c8 CL |
822 | .start = m_start, |
823 | .next = m_next, | |
824 | .stop = m_stop, | |
3bbfe059 | 825 | .show = show_numa_map, |
6e21c8f1 | 826 | }; |
662795de EB |
827 | |
828 | static int numa_maps_open(struct inode *inode, struct file *file) | |
829 | { | |
830 | return do_maps_open(inode, file, &proc_pid_numa_maps_op); | |
831 | } | |
832 | ||
00977a59 | 833 | const struct file_operations proc_numa_maps_operations = { |
662795de EB |
834 | .open = numa_maps_open, |
835 | .read = seq_read, | |
836 | .llseek = seq_lseek, | |
99f89551 | 837 | .release = seq_release_private, |
662795de | 838 | }; |
6e21c8f1 | 839 | #endif |