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> |
6e21c8f1 CL |
7 | #include <linux/pagemap.h> |
8 | #include <linux/mempolicy.h> | |
85863e47 MM |
9 | #include <linux/swap.h> |
10 | #include <linux/swapops.h> | |
e070ad49 | 11 | |
1da177e4 LT |
12 | #include <asm/elf.h> |
13 | #include <asm/uaccess.h> | |
e070ad49 | 14 | #include <asm/tlbflush.h> |
1da177e4 LT |
15 | #include "internal.h" |
16 | ||
df5f8314 | 17 | void task_mem(struct seq_file *m, struct mm_struct *mm) |
1da177e4 LT |
18 | { |
19 | unsigned long data, text, lib; | |
365e9c87 HD |
20 | unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; |
21 | ||
22 | /* | |
23 | * Note: to minimize their overhead, mm maintains hiwater_vm and | |
24 | * hiwater_rss only when about to *lower* total_vm or rss. Any | |
25 | * collector of these hiwater stats must therefore get total_vm | |
26 | * and rss too, which will usually be the higher. Barriers? not | |
27 | * worth the effort, such snapshots can always be inconsistent. | |
28 | */ | |
29 | hiwater_vm = total_vm = mm->total_vm; | |
30 | if (hiwater_vm < mm->hiwater_vm) | |
31 | hiwater_vm = mm->hiwater_vm; | |
32 | hiwater_rss = total_rss = get_mm_rss(mm); | |
33 | if (hiwater_rss < mm->hiwater_rss) | |
34 | hiwater_rss = mm->hiwater_rss; | |
1da177e4 LT |
35 | |
36 | data = mm->total_vm - mm->shared_vm - mm->stack_vm; | |
37 | text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; | |
38 | lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; | |
df5f8314 | 39 | seq_printf(m, |
365e9c87 | 40 | "VmPeak:\t%8lu kB\n" |
1da177e4 LT |
41 | "VmSize:\t%8lu kB\n" |
42 | "VmLck:\t%8lu kB\n" | |
365e9c87 | 43 | "VmHWM:\t%8lu kB\n" |
1da177e4 LT |
44 | "VmRSS:\t%8lu kB\n" |
45 | "VmData:\t%8lu kB\n" | |
46 | "VmStk:\t%8lu kB\n" | |
47 | "VmExe:\t%8lu kB\n" | |
48 | "VmLib:\t%8lu kB\n" | |
49 | "VmPTE:\t%8lu kB\n", | |
365e9c87 HD |
50 | hiwater_vm << (PAGE_SHIFT-10), |
51 | (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10), | |
1da177e4 | 52 | mm->locked_vm << (PAGE_SHIFT-10), |
365e9c87 HD |
53 | hiwater_rss << (PAGE_SHIFT-10), |
54 | total_rss << (PAGE_SHIFT-10), | |
1da177e4 LT |
55 | data << (PAGE_SHIFT-10), |
56 | mm->stack_vm << (PAGE_SHIFT-10), text, lib, | |
57 | (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10); | |
1da177e4 LT |
58 | } |
59 | ||
60 | unsigned long task_vsize(struct mm_struct *mm) | |
61 | { | |
62 | return PAGE_SIZE * mm->total_vm; | |
63 | } | |
64 | ||
65 | int task_statm(struct mm_struct *mm, int *shared, int *text, | |
66 | int *data, int *resident) | |
67 | { | |
4294621f | 68 | *shared = get_mm_counter(mm, file_rss); |
1da177e4 LT |
69 | *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) |
70 | >> PAGE_SHIFT; | |
71 | *data = mm->total_vm - mm->shared_vm; | |
4294621f | 72 | *resident = *shared + get_mm_counter(mm, anon_rss); |
1da177e4 LT |
73 | return mm->total_vm; |
74 | } | |
75 | ||
1da177e4 LT |
76 | static void pad_len_spaces(struct seq_file *m, int len) |
77 | { | |
78 | len = 25 + sizeof(void*) * 6 - len; | |
79 | if (len < 1) | |
80 | len = 1; | |
81 | seq_printf(m, "%*c", len, ' '); | |
82 | } | |
83 | ||
a6198797 MM |
84 | static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma) |
85 | { | |
86 | if (vma && vma != priv->tail_vma) { | |
87 | struct mm_struct *mm = vma->vm_mm; | |
88 | up_read(&mm->mmap_sem); | |
89 | mmput(mm); | |
90 | } | |
91 | } | |
ec4dd3eb | 92 | |
a6198797 | 93 | static void *m_start(struct seq_file *m, loff_t *pos) |
e070ad49 | 94 | { |
a6198797 MM |
95 | struct proc_maps_private *priv = m->private; |
96 | unsigned long last_addr = m->version; | |
97 | struct mm_struct *mm; | |
98 | struct vm_area_struct *vma, *tail_vma = NULL; | |
99 | loff_t l = *pos; | |
100 | ||
101 | /* Clear the per syscall fields in priv */ | |
102 | priv->task = NULL; | |
103 | priv->tail_vma = NULL; | |
104 | ||
105 | /* | |
106 | * We remember last_addr rather than next_addr to hit with | |
107 | * mmap_cache most of the time. We have zero last_addr at | |
108 | * the beginning and also after lseek. We will have -1 last_addr | |
109 | * after the end of the vmas. | |
110 | */ | |
111 | ||
112 | if (last_addr == -1UL) | |
113 | return NULL; | |
114 | ||
115 | priv->task = get_pid_task(priv->pid, PIDTYPE_PID); | |
116 | if (!priv->task) | |
117 | return NULL; | |
118 | ||
119 | mm = mm_for_maps(priv->task); | |
120 | if (!mm) | |
121 | return NULL; | |
00f89d21 | 122 | down_read(&mm->mmap_sem); |
a6198797 MM |
123 | |
124 | tail_vma = get_gate_vma(priv->task); | |
125 | priv->tail_vma = tail_vma; | |
126 | ||
127 | /* Start with last addr hint */ | |
128 | vma = find_vma(mm, last_addr); | |
129 | if (last_addr && vma) { | |
130 | vma = vma->vm_next; | |
131 | goto out; | |
132 | } | |
133 | ||
134 | /* | |
135 | * Check the vma index is within the range and do | |
136 | * sequential scan until m_index. | |
137 | */ | |
138 | vma = NULL; | |
139 | if ((unsigned long)l < mm->map_count) { | |
140 | vma = mm->mmap; | |
141 | while (l-- && vma) | |
142 | vma = vma->vm_next; | |
143 | goto out; | |
144 | } | |
145 | ||
146 | if (l != mm->map_count) | |
147 | tail_vma = NULL; /* After gate vma */ | |
148 | ||
149 | out: | |
150 | if (vma) | |
151 | return vma; | |
152 | ||
153 | /* End of vmas has been reached */ | |
154 | m->version = (tail_vma != NULL)? 0: -1UL; | |
155 | up_read(&mm->mmap_sem); | |
156 | mmput(mm); | |
157 | return tail_vma; | |
158 | } | |
159 | ||
160 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
161 | { | |
162 | struct proc_maps_private *priv = m->private; | |
163 | struct vm_area_struct *vma = v; | |
164 | struct vm_area_struct *tail_vma = priv->tail_vma; | |
165 | ||
166 | (*pos)++; | |
167 | if (vma && (vma != tail_vma) && vma->vm_next) | |
168 | return vma->vm_next; | |
169 | vma_stop(priv, vma); | |
170 | return (vma != tail_vma)? tail_vma: NULL; | |
171 | } | |
172 | ||
173 | static void m_stop(struct seq_file *m, void *v) | |
174 | { | |
175 | struct proc_maps_private *priv = m->private; | |
176 | struct vm_area_struct *vma = v; | |
177 | ||
178 | vma_stop(priv, vma); | |
179 | if (priv->task) | |
180 | put_task_struct(priv->task); | |
181 | } | |
182 | ||
183 | static int do_maps_open(struct inode *inode, struct file *file, | |
03a44825 | 184 | const struct seq_operations *ops) |
a6198797 MM |
185 | { |
186 | struct proc_maps_private *priv; | |
187 | int ret = -ENOMEM; | |
188 | priv = kzalloc(sizeof(*priv), GFP_KERNEL); | |
189 | if (priv) { | |
190 | priv->pid = proc_pid(inode); | |
191 | ret = seq_open(file, ops); | |
192 | if (!ret) { | |
193 | struct seq_file *m = file->private_data; | |
194 | m->private = priv; | |
195 | } else { | |
196 | kfree(priv); | |
197 | } | |
198 | } | |
199 | return ret; | |
200 | } | |
e070ad49 | 201 | |
7c88db0c | 202 | static void show_map_vma(struct seq_file *m, struct vm_area_struct *vma) |
1da177e4 | 203 | { |
e070ad49 ML |
204 | struct mm_struct *mm = vma->vm_mm; |
205 | struct file *file = vma->vm_file; | |
206 | int flags = vma->vm_flags; | |
1da177e4 | 207 | unsigned long ino = 0; |
6260a4b0 | 208 | unsigned long long pgoff = 0; |
1da177e4 LT |
209 | dev_t dev = 0; |
210 | int len; | |
211 | ||
212 | if (file) { | |
2fddfeef | 213 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 LT |
214 | dev = inode->i_sb->s_dev; |
215 | ino = inode->i_ino; | |
6260a4b0 | 216 | pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; |
1da177e4 LT |
217 | } |
218 | ||
1804dc6e | 219 | seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu %n", |
e070ad49 ML |
220 | vma->vm_start, |
221 | vma->vm_end, | |
1da177e4 LT |
222 | flags & VM_READ ? 'r' : '-', |
223 | flags & VM_WRITE ? 'w' : '-', | |
224 | flags & VM_EXEC ? 'x' : '-', | |
225 | flags & VM_MAYSHARE ? 's' : 'p', | |
6260a4b0 | 226 | pgoff, |
1da177e4 LT |
227 | MAJOR(dev), MINOR(dev), ino, &len); |
228 | ||
229 | /* | |
230 | * Print the dentry name for named mappings, and a | |
231 | * special [heap] marker for the heap: | |
232 | */ | |
e070ad49 | 233 | if (file) { |
1da177e4 | 234 | pad_len_spaces(m, len); |
c32c2f63 | 235 | seq_path(m, &file->f_path, "\n"); |
1da177e4 | 236 | } else { |
e6e5494c IM |
237 | const char *name = arch_vma_name(vma); |
238 | if (!name) { | |
239 | if (mm) { | |
240 | if (vma->vm_start <= mm->start_brk && | |
e070ad49 | 241 | vma->vm_end >= mm->brk) { |
e6e5494c IM |
242 | name = "[heap]"; |
243 | } else if (vma->vm_start <= mm->start_stack && | |
244 | vma->vm_end >= mm->start_stack) { | |
245 | name = "[stack]"; | |
1da177e4 | 246 | } |
e6e5494c IM |
247 | } else { |
248 | name = "[vdso]"; | |
1da177e4 | 249 | } |
e6e5494c IM |
250 | } |
251 | if (name) { | |
1da177e4 | 252 | pad_len_spaces(m, len); |
e6e5494c | 253 | seq_puts(m, name); |
1da177e4 LT |
254 | } |
255 | } | |
256 | seq_putc(m, '\n'); | |
7c88db0c JK |
257 | } |
258 | ||
259 | static int show_map(struct seq_file *m, void *v) | |
260 | { | |
261 | struct vm_area_struct *vma = v; | |
262 | struct proc_maps_private *priv = m->private; | |
263 | struct task_struct *task = priv->task; | |
264 | ||
265 | show_map_vma(m, vma); | |
e070ad49 | 266 | |
e070ad49 ML |
267 | if (m->count < m->size) /* vma is copied successfully */ |
268 | m->version = (vma != get_gate_vma(task))? vma->vm_start: 0; | |
1da177e4 LT |
269 | return 0; |
270 | } | |
271 | ||
03a44825 | 272 | static const struct seq_operations proc_pid_maps_op = { |
a6198797 MM |
273 | .start = m_start, |
274 | .next = m_next, | |
275 | .stop = m_stop, | |
276 | .show = show_map | |
277 | }; | |
278 | ||
279 | static int maps_open(struct inode *inode, struct file *file) | |
280 | { | |
281 | return do_maps_open(inode, file, &proc_pid_maps_op); | |
282 | } | |
283 | ||
284 | const struct file_operations proc_maps_operations = { | |
285 | .open = maps_open, | |
286 | .read = seq_read, | |
287 | .llseek = seq_lseek, | |
288 | .release = seq_release_private, | |
289 | }; | |
290 | ||
291 | /* | |
292 | * Proportional Set Size(PSS): my share of RSS. | |
293 | * | |
294 | * PSS of a process is the count of pages it has in memory, where each | |
295 | * page is divided by the number of processes sharing it. So if a | |
296 | * process has 1000 pages all to itself, and 1000 shared with one other | |
297 | * process, its PSS will be 1500. | |
298 | * | |
299 | * To keep (accumulated) division errors low, we adopt a 64bit | |
300 | * fixed-point pss counter to minimize division errors. So (pss >> | |
301 | * PSS_SHIFT) would be the real byte count. | |
302 | * | |
303 | * A shift of 12 before division means (assuming 4K page size): | |
304 | * - 1M 3-user-pages add up to 8KB errors; | |
305 | * - supports mapcount up to 2^24, or 16M; | |
306 | * - supports PSS up to 2^52 bytes, or 4PB. | |
307 | */ | |
308 | #define PSS_SHIFT 12 | |
309 | ||
1e883281 | 310 | #ifdef CONFIG_PROC_PAGE_MONITOR |
214e471f | 311 | struct mem_size_stats { |
a6198797 MM |
312 | struct vm_area_struct *vma; |
313 | unsigned long resident; | |
314 | unsigned long shared_clean; | |
315 | unsigned long shared_dirty; | |
316 | unsigned long private_clean; | |
317 | unsigned long private_dirty; | |
318 | unsigned long referenced; | |
214e471f | 319 | unsigned long swap; |
a6198797 MM |
320 | u64 pss; |
321 | }; | |
322 | ||
b3ae5acb | 323 | static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 324 | struct mm_walk *walk) |
e070ad49 | 325 | { |
2165009b | 326 | struct mem_size_stats *mss = walk->private; |
b3ae5acb | 327 | struct vm_area_struct *vma = mss->vma; |
e070ad49 | 328 | pte_t *pte, ptent; |
705e87c0 | 329 | spinlock_t *ptl; |
e070ad49 | 330 | struct page *page; |
ec4dd3eb | 331 | int mapcount; |
e070ad49 | 332 | |
705e87c0 | 333 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
826fad1b | 334 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
e070ad49 | 335 | ptent = *pte; |
214e471f PZ |
336 | |
337 | if (is_swap_pte(ptent)) { | |
338 | mss->swap += PAGE_SIZE; | |
339 | continue; | |
340 | } | |
341 | ||
705e87c0 | 342 | if (!pte_present(ptent)) |
e070ad49 ML |
343 | continue; |
344 | ||
345 | mss->resident += PAGE_SIZE; | |
ad820c5d NP |
346 | |
347 | page = vm_normal_page(vma, addr, ptent); | |
348 | if (!page) | |
e070ad49 ML |
349 | continue; |
350 | ||
f79f177c DR |
351 | /* Accumulate the size in pages that have been accessed. */ |
352 | if (pte_young(ptent) || PageReferenced(page)) | |
353 | mss->referenced += PAGE_SIZE; | |
ec4dd3eb FW |
354 | mapcount = page_mapcount(page); |
355 | if (mapcount >= 2) { | |
e070ad49 ML |
356 | if (pte_dirty(ptent)) |
357 | mss->shared_dirty += PAGE_SIZE; | |
358 | else | |
359 | mss->shared_clean += PAGE_SIZE; | |
ec4dd3eb | 360 | mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; |
e070ad49 ML |
361 | } else { |
362 | if (pte_dirty(ptent)) | |
363 | mss->private_dirty += PAGE_SIZE; | |
364 | else | |
365 | mss->private_clean += PAGE_SIZE; | |
ec4dd3eb | 366 | mss->pss += (PAGE_SIZE << PSS_SHIFT); |
e070ad49 | 367 | } |
826fad1b | 368 | } |
705e87c0 HD |
369 | pte_unmap_unlock(pte - 1, ptl); |
370 | cond_resched(); | |
b3ae5acb | 371 | return 0; |
e070ad49 ML |
372 | } |
373 | ||
e070ad49 ML |
374 | static int show_smap(struct seq_file *m, void *v) |
375 | { | |
7c88db0c JK |
376 | struct proc_maps_private *priv = m->private; |
377 | struct task_struct *task = priv->task; | |
e070ad49 | 378 | struct vm_area_struct *vma = v; |
e070ad49 | 379 | struct mem_size_stats mss; |
2165009b DH |
380 | struct mm_walk smaps_walk = { |
381 | .pmd_entry = smaps_pte_range, | |
382 | .mm = vma->vm_mm, | |
383 | .private = &mss, | |
384 | }; | |
e070ad49 ML |
385 | |
386 | memset(&mss, 0, sizeof mss); | |
b3ae5acb | 387 | mss.vma = vma; |
5ddfae16 | 388 | if (vma->vm_mm && !is_vm_hugetlb_page(vma)) |
2165009b | 389 | walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); |
4752c369 | 390 | |
7c88db0c | 391 | show_map_vma(m, vma); |
4752c369 MM |
392 | |
393 | seq_printf(m, | |
394 | "Size: %8lu kB\n" | |
395 | "Rss: %8lu kB\n" | |
396 | "Pss: %8lu kB\n" | |
397 | "Shared_Clean: %8lu kB\n" | |
398 | "Shared_Dirty: %8lu kB\n" | |
399 | "Private_Clean: %8lu kB\n" | |
400 | "Private_Dirty: %8lu kB\n" | |
214e471f | 401 | "Referenced: %8lu kB\n" |
08fba699 | 402 | "Swap: %8lu kB\n" |
3340289d MG |
403 | "KernelPageSize: %8lu kB\n" |
404 | "MMUPageSize: %8lu kB\n", | |
4752c369 MM |
405 | (vma->vm_end - vma->vm_start) >> 10, |
406 | mss.resident >> 10, | |
407 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), | |
408 | mss.shared_clean >> 10, | |
409 | mss.shared_dirty >> 10, | |
410 | mss.private_clean >> 10, | |
411 | mss.private_dirty >> 10, | |
214e471f | 412 | mss.referenced >> 10, |
08fba699 | 413 | mss.swap >> 10, |
3340289d MG |
414 | vma_kernel_pagesize(vma) >> 10, |
415 | vma_mmu_pagesize(vma) >> 10); | |
4752c369 | 416 | |
7c88db0c JK |
417 | if (m->count < m->size) /* vma is copied successfully */ |
418 | m->version = (vma != get_gate_vma(task)) ? vma->vm_start : 0; | |
419 | return 0; | |
e070ad49 ML |
420 | } |
421 | ||
03a44825 | 422 | static const struct seq_operations proc_pid_smaps_op = { |
a6198797 MM |
423 | .start = m_start, |
424 | .next = m_next, | |
425 | .stop = m_stop, | |
426 | .show = show_smap | |
427 | }; | |
428 | ||
429 | static int smaps_open(struct inode *inode, struct file *file) | |
430 | { | |
431 | return do_maps_open(inode, file, &proc_pid_smaps_op); | |
432 | } | |
433 | ||
434 | const struct file_operations proc_smaps_operations = { | |
435 | .open = smaps_open, | |
436 | .read = seq_read, | |
437 | .llseek = seq_lseek, | |
438 | .release = seq_release_private, | |
439 | }; | |
440 | ||
441 | static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, | |
2165009b | 442 | unsigned long end, struct mm_walk *walk) |
a6198797 | 443 | { |
2165009b | 444 | struct vm_area_struct *vma = walk->private; |
a6198797 MM |
445 | pte_t *pte, ptent; |
446 | spinlock_t *ptl; | |
447 | struct page *page; | |
448 | ||
449 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
450 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
451 | ptent = *pte; | |
452 | if (!pte_present(ptent)) | |
453 | continue; | |
454 | ||
455 | page = vm_normal_page(vma, addr, ptent); | |
456 | if (!page) | |
457 | continue; | |
458 | ||
459 | /* Clear accessed and referenced bits. */ | |
460 | ptep_test_and_clear_young(vma, addr, pte); | |
461 | ClearPageReferenced(page); | |
462 | } | |
463 | pte_unmap_unlock(pte - 1, ptl); | |
464 | cond_resched(); | |
465 | return 0; | |
466 | } | |
467 | ||
f248dcb3 MM |
468 | static ssize_t clear_refs_write(struct file *file, const char __user *buf, |
469 | size_t count, loff_t *ppos) | |
b813e931 | 470 | { |
f248dcb3 MM |
471 | struct task_struct *task; |
472 | char buffer[PROC_NUMBUF], *end; | |
473 | struct mm_struct *mm; | |
b813e931 DR |
474 | struct vm_area_struct *vma; |
475 | ||
f248dcb3 MM |
476 | memset(buffer, 0, sizeof(buffer)); |
477 | if (count > sizeof(buffer) - 1) | |
478 | count = sizeof(buffer) - 1; | |
479 | if (copy_from_user(buffer, buf, count)) | |
480 | return -EFAULT; | |
481 | if (!simple_strtol(buffer, &end, 0)) | |
482 | return -EINVAL; | |
483 | if (*end == '\n') | |
484 | end++; | |
485 | task = get_proc_task(file->f_path.dentry->d_inode); | |
486 | if (!task) | |
487 | return -ESRCH; | |
488 | mm = get_task_mm(task); | |
489 | if (mm) { | |
20cbc972 AM |
490 | struct mm_walk clear_refs_walk = { |
491 | .pmd_entry = clear_refs_pte_range, | |
492 | .mm = mm, | |
493 | }; | |
f248dcb3 | 494 | down_read(&mm->mmap_sem); |
2165009b DH |
495 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
496 | clear_refs_walk.private = vma; | |
f248dcb3 | 497 | if (!is_vm_hugetlb_page(vma)) |
2165009b DH |
498 | walk_page_range(vma->vm_start, vma->vm_end, |
499 | &clear_refs_walk); | |
500 | } | |
f248dcb3 MM |
501 | flush_tlb_mm(mm); |
502 | up_read(&mm->mmap_sem); | |
503 | mmput(mm); | |
504 | } | |
505 | put_task_struct(task); | |
506 | if (end - buffer == 0) | |
507 | return -EIO; | |
508 | return end - buffer; | |
b813e931 DR |
509 | } |
510 | ||
f248dcb3 MM |
511 | const struct file_operations proc_clear_refs_operations = { |
512 | .write = clear_refs_write, | |
513 | }; | |
514 | ||
85863e47 | 515 | struct pagemapread { |
aae8679b | 516 | u64 __user *out, *end; |
85863e47 MM |
517 | }; |
518 | ||
f16278c6 HR |
519 | #define PM_ENTRY_BYTES sizeof(u64) |
520 | #define PM_STATUS_BITS 3 | |
521 | #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) | |
522 | #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) | |
523 | #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) | |
524 | #define PM_PSHIFT_BITS 6 | |
525 | #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) | |
526 | #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) | |
527 | #define PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) | |
528 | #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) | |
529 | #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) | |
530 | ||
531 | #define PM_PRESENT PM_STATUS(4LL) | |
532 | #define PM_SWAP PM_STATUS(2LL) | |
533 | #define PM_NOT_PRESENT PM_PSHIFT(PAGE_SHIFT) | |
85863e47 MM |
534 | #define PM_END_OF_BUFFER 1 |
535 | ||
536 | static int add_to_pagemap(unsigned long addr, u64 pfn, | |
537 | struct pagemapread *pm) | |
538 | { | |
85863e47 MM |
539 | if (put_user(pfn, pm->out)) |
540 | return -EFAULT; | |
aae8679b TT |
541 | pm->out++; |
542 | if (pm->out >= pm->end) | |
543 | return PM_END_OF_BUFFER; | |
85863e47 MM |
544 | return 0; |
545 | } | |
546 | ||
547 | static int pagemap_pte_hole(unsigned long start, unsigned long end, | |
2165009b | 548 | struct mm_walk *walk) |
85863e47 | 549 | { |
2165009b | 550 | struct pagemapread *pm = walk->private; |
85863e47 MM |
551 | unsigned long addr; |
552 | int err = 0; | |
553 | for (addr = start; addr < end; addr += PAGE_SIZE) { | |
554 | err = add_to_pagemap(addr, PM_NOT_PRESENT, pm); | |
555 | if (err) | |
556 | break; | |
557 | } | |
558 | return err; | |
559 | } | |
560 | ||
9d02dbc8 | 561 | static u64 swap_pte_to_pagemap_entry(pte_t pte) |
85863e47 MM |
562 | { |
563 | swp_entry_t e = pte_to_swp_entry(pte); | |
f16278c6 | 564 | return swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT); |
85863e47 MM |
565 | } |
566 | ||
49c50342 | 567 | static u64 pte_to_pagemap_entry(pte_t pte) |
bcf8039e | 568 | { |
49c50342 | 569 | u64 pme = 0; |
bcf8039e DH |
570 | if (is_swap_pte(pte)) |
571 | pme = PM_PFRAME(swap_pte_to_pagemap_entry(pte)) | |
572 | | PM_PSHIFT(PAGE_SHIFT) | PM_SWAP; | |
573 | else if (pte_present(pte)) | |
574 | pme = PM_PFRAME(pte_pfn(pte)) | |
575 | | PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT; | |
576 | return pme; | |
577 | } | |
578 | ||
85863e47 | 579 | static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 580 | struct mm_walk *walk) |
85863e47 | 581 | { |
bcf8039e | 582 | struct vm_area_struct *vma; |
2165009b | 583 | struct pagemapread *pm = walk->private; |
85863e47 MM |
584 | pte_t *pte; |
585 | int err = 0; | |
586 | ||
bcf8039e DH |
587 | /* find the first VMA at or above 'addr' */ |
588 | vma = find_vma(walk->mm, addr); | |
85863e47 MM |
589 | for (; addr != end; addr += PAGE_SIZE) { |
590 | u64 pfn = PM_NOT_PRESENT; | |
bcf8039e DH |
591 | |
592 | /* check to see if we've left 'vma' behind | |
593 | * and need a new, higher one */ | |
594 | if (vma && (addr >= vma->vm_end)) | |
595 | vma = find_vma(walk->mm, addr); | |
596 | ||
597 | /* check that 'vma' actually covers this address, | |
598 | * and that it isn't a huge page vma */ | |
599 | if (vma && (vma->vm_start <= addr) && | |
600 | !is_vm_hugetlb_page(vma)) { | |
601 | pte = pte_offset_map(pmd, addr); | |
602 | pfn = pte_to_pagemap_entry(*pte); | |
603 | /* unmap before userspace copy */ | |
604 | pte_unmap(pte); | |
605 | } | |
85863e47 MM |
606 | err = add_to_pagemap(addr, pfn, pm); |
607 | if (err) | |
608 | return err; | |
609 | } | |
610 | ||
611 | cond_resched(); | |
612 | ||
613 | return err; | |
614 | } | |
615 | ||
85863e47 MM |
616 | /* |
617 | * /proc/pid/pagemap - an array mapping virtual pages to pfns | |
618 | * | |
f16278c6 HR |
619 | * For each page in the address space, this file contains one 64-bit entry |
620 | * consisting of the following: | |
621 | * | |
622 | * Bits 0-55 page frame number (PFN) if present | |
623 | * Bits 0-4 swap type if swapped | |
624 | * Bits 5-55 swap offset if swapped | |
625 | * Bits 55-60 page shift (page size = 1<<page shift) | |
626 | * Bit 61 reserved for future use | |
627 | * Bit 62 page swapped | |
628 | * Bit 63 page present | |
629 | * | |
630 | * If the page is not present but in swap, then the PFN contains an | |
631 | * encoding of the swap file number and the page's offset into the | |
632 | * swap. Unmapped pages return a null PFN. This allows determining | |
85863e47 MM |
633 | * precisely which pages are mapped (or in swap) and comparing mapped |
634 | * pages between processes. | |
635 | * | |
636 | * Efficient users of this interface will use /proc/pid/maps to | |
637 | * determine which areas of memory are actually mapped and llseek to | |
638 | * skip over unmapped regions. | |
639 | */ | |
640 | static ssize_t pagemap_read(struct file *file, char __user *buf, | |
641 | size_t count, loff_t *ppos) | |
642 | { | |
643 | struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); | |
644 | struct page **pages, *page; | |
645 | unsigned long uaddr, uend; | |
646 | struct mm_struct *mm; | |
647 | struct pagemapread pm; | |
648 | int pagecount; | |
649 | int ret = -ESRCH; | |
ee1e6ab6 | 650 | struct mm_walk pagemap_walk = {}; |
5d7e0d2b AM |
651 | unsigned long src; |
652 | unsigned long svpfn; | |
653 | unsigned long start_vaddr; | |
654 | unsigned long end_vaddr; | |
85863e47 MM |
655 | |
656 | if (!task) | |
657 | goto out; | |
658 | ||
659 | ret = -EACCES; | |
006ebb40 | 660 | if (!ptrace_may_access(task, PTRACE_MODE_READ)) |
fb39380b | 661 | goto out_task; |
85863e47 MM |
662 | |
663 | ret = -EINVAL; | |
664 | /* file position must be aligned */ | |
aae8679b | 665 | if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) |
fb39380b | 666 | goto out_task; |
85863e47 MM |
667 | |
668 | ret = 0; | |
08161786 VM |
669 | |
670 | if (!count) | |
671 | goto out_task; | |
672 | ||
85863e47 MM |
673 | mm = get_task_mm(task); |
674 | if (!mm) | |
fb39380b | 675 | goto out_task; |
85863e47 | 676 | |
5d7e0d2b | 677 | |
85863e47 MM |
678 | uaddr = (unsigned long)buf & PAGE_MASK; |
679 | uend = (unsigned long)(buf + count); | |
680 | pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE; | |
5d7e0d2b AM |
681 | ret = 0; |
682 | if (pagecount == 0) | |
683 | goto out_mm; | |
684 | pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL); | |
685 | ret = -ENOMEM; | |
85863e47 | 686 | if (!pages) |
fb39380b | 687 | goto out_mm; |
85863e47 MM |
688 | |
689 | down_read(¤t->mm->mmap_sem); | |
690 | ret = get_user_pages(current, current->mm, uaddr, pagecount, | |
691 | 1, 0, pages, NULL); | |
692 | up_read(¤t->mm->mmap_sem); | |
693 | ||
694 | if (ret < 0) | |
695 | goto out_free; | |
696 | ||
fb39380b MT |
697 | if (ret != pagecount) { |
698 | pagecount = ret; | |
699 | ret = -EFAULT; | |
700 | goto out_pages; | |
701 | } | |
702 | ||
09729a99 MAC |
703 | pm.out = (u64 __user *)buf; |
704 | pm.end = (u64 __user *)(buf + count); | |
85863e47 | 705 | |
5d7e0d2b AM |
706 | pagemap_walk.pmd_entry = pagemap_pte_range; |
707 | pagemap_walk.pte_hole = pagemap_pte_hole; | |
708 | pagemap_walk.mm = mm; | |
709 | pagemap_walk.private = ± | |
710 | ||
711 | src = *ppos; | |
712 | svpfn = src / PM_ENTRY_BYTES; | |
713 | start_vaddr = svpfn << PAGE_SHIFT; | |
714 | end_vaddr = TASK_SIZE_OF(task); | |
715 | ||
716 | /* watch out for wraparound */ | |
717 | if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) | |
718 | start_vaddr = end_vaddr; | |
719 | ||
720 | /* | |
721 | * The odds are that this will stop walking way | |
722 | * before end_vaddr, because the length of the | |
723 | * user buffer is tracked in "pm", and the walk | |
724 | * will stop when we hit the end of the buffer. | |
725 | */ | |
726 | ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk); | |
727 | if (ret == PM_END_OF_BUFFER) | |
728 | ret = 0; | |
729 | /* don't need mmap_sem for these, but this looks cleaner */ | |
09729a99 | 730 | *ppos += (char __user *)pm.out - buf; |
5d7e0d2b | 731 | if (!ret) |
09729a99 | 732 | ret = (char __user *)pm.out - buf; |
85863e47 | 733 | |
fb39380b | 734 | out_pages: |
85863e47 MM |
735 | for (; pagecount; pagecount--) { |
736 | page = pages[pagecount-1]; | |
737 | if (!PageReserved(page)) | |
738 | SetPageDirty(page); | |
739 | page_cache_release(page); | |
740 | } | |
85863e47 MM |
741 | out_free: |
742 | kfree(pages); | |
fb39380b MT |
743 | out_mm: |
744 | mmput(mm); | |
85863e47 MM |
745 | out_task: |
746 | put_task_struct(task); | |
747 | out: | |
748 | return ret; | |
749 | } | |
750 | ||
751 | const struct file_operations proc_pagemap_operations = { | |
752 | .llseek = mem_lseek, /* borrow this */ | |
753 | .read = pagemap_read, | |
754 | }; | |
1e883281 | 755 | #endif /* CONFIG_PROC_PAGE_MONITOR */ |
85863e47 | 756 | |
6e21c8f1 | 757 | #ifdef CONFIG_NUMA |
1a75a6c8 | 758 | extern int show_numa_map(struct seq_file *m, void *v); |
6e21c8f1 | 759 | |
03a44825 | 760 | static const struct seq_operations proc_pid_numa_maps_op = { |
1a75a6c8 CL |
761 | .start = m_start, |
762 | .next = m_next, | |
763 | .stop = m_stop, | |
3bbfe059 | 764 | .show = show_numa_map, |
6e21c8f1 | 765 | }; |
662795de EB |
766 | |
767 | static int numa_maps_open(struct inode *inode, struct file *file) | |
768 | { | |
769 | return do_maps_open(inode, file, &proc_pid_numa_maps_op); | |
770 | } | |
771 | ||
00977a59 | 772 | const struct file_operations proc_numa_maps_operations = { |
662795de EB |
773 | .open = numa_maps_open, |
774 | .read = seq_read, | |
775 | .llseek = seq_lseek, | |
99f89551 | 776 | .release = seq_release_private, |
662795de | 777 | }; |
6e21c8f1 | 778 | #endif |