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 | 7 | #include <linux/pagemap.h> |
85863e47 | 8 | #include <linux/ptrace.h> |
6e21c8f1 | 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 | ||
18 | char *task_mem(struct mm_struct *mm, char *buffer) | |
19 | { | |
20 | unsigned long data, text, lib; | |
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; | |
40 | buffer += sprintf(buffer, | |
365e9c87 | 41 | "VmPeak:\t%8lu kB\n" |
1da177e4 LT |
42 | "VmSize:\t%8lu kB\n" |
43 | "VmLck:\t%8lu kB\n" | |
365e9c87 | 44 | "VmHWM:\t%8lu kB\n" |
1da177e4 LT |
45 | "VmRSS:\t%8lu kB\n" |
46 | "VmData:\t%8lu kB\n" | |
47 | "VmStk:\t%8lu kB\n" | |
48 | "VmExe:\t%8lu kB\n" | |
49 | "VmLib:\t%8lu kB\n" | |
50 | "VmPTE:\t%8lu kB\n", | |
365e9c87 HD |
51 | hiwater_vm << (PAGE_SHIFT-10), |
52 | (total_vm - mm->reserved_vm) << (PAGE_SHIFT-10), | |
1da177e4 | 53 | mm->locked_vm << (PAGE_SHIFT-10), |
365e9c87 HD |
54 | hiwater_rss << (PAGE_SHIFT-10), |
55 | total_rss << (PAGE_SHIFT-10), | |
1da177e4 LT |
56 | data << (PAGE_SHIFT-10), |
57 | mm->stack_vm << (PAGE_SHIFT-10), text, lib, | |
58 | (PTRS_PER_PTE*sizeof(pte_t)*mm->nr_ptes) >> 10); | |
59 | return buffer; | |
60 | } | |
61 | ||
62 | unsigned long task_vsize(struct mm_struct *mm) | |
63 | { | |
64 | return PAGE_SIZE * mm->total_vm; | |
65 | } | |
66 | ||
67 | int task_statm(struct mm_struct *mm, int *shared, int *text, | |
68 | int *data, int *resident) | |
69 | { | |
4294621f | 70 | *shared = get_mm_counter(mm, file_rss); |
1da177e4 LT |
71 | *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) |
72 | >> PAGE_SHIFT; | |
73 | *data = mm->total_vm - mm->shared_vm; | |
4294621f | 74 | *resident = *shared + get_mm_counter(mm, anon_rss); |
1da177e4 LT |
75 | return mm->total_vm; |
76 | } | |
77 | ||
78 | int proc_exe_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt) | |
79 | { | |
80 | struct vm_area_struct * vma; | |
81 | int result = -ENOENT; | |
99f89551 EB |
82 | struct task_struct *task = get_proc_task(inode); |
83 | struct mm_struct * mm = NULL; | |
1da177e4 | 84 | |
99f89551 EB |
85 | if (task) { |
86 | mm = get_task_mm(task); | |
87 | put_task_struct(task); | |
88 | } | |
1da177e4 LT |
89 | if (!mm) |
90 | goto out; | |
91 | down_read(&mm->mmap_sem); | |
92 | ||
93 | vma = mm->mmap; | |
94 | while (vma) { | |
95 | if ((vma->vm_flags & VM_EXECUTABLE) && vma->vm_file) | |
96 | break; | |
97 | vma = vma->vm_next; | |
98 | } | |
99 | ||
100 | if (vma) { | |
2fddfeef JJS |
101 | *mnt = mntget(vma->vm_file->f_path.mnt); |
102 | *dentry = dget(vma->vm_file->f_path.dentry); | |
1da177e4 LT |
103 | result = 0; |
104 | } | |
105 | ||
106 | up_read(&mm->mmap_sem); | |
107 | mmput(mm); | |
108 | out: | |
109 | return result; | |
110 | } | |
111 | ||
112 | static void pad_len_spaces(struct seq_file *m, int len) | |
113 | { | |
114 | len = 25 + sizeof(void*) * 6 - len; | |
115 | if (len < 1) | |
116 | len = 1; | |
117 | seq_printf(m, "%*c", len, ' '); | |
118 | } | |
119 | ||
a6198797 MM |
120 | static void vma_stop(struct proc_maps_private *priv, struct vm_area_struct *vma) |
121 | { | |
122 | if (vma && vma != priv->tail_vma) { | |
123 | struct mm_struct *mm = vma->vm_mm; | |
124 | up_read(&mm->mmap_sem); | |
125 | mmput(mm); | |
126 | } | |
127 | } | |
ec4dd3eb | 128 | |
a6198797 | 129 | static void *m_start(struct seq_file *m, loff_t *pos) |
e070ad49 | 130 | { |
a6198797 MM |
131 | struct proc_maps_private *priv = m->private; |
132 | unsigned long last_addr = m->version; | |
133 | struct mm_struct *mm; | |
134 | struct vm_area_struct *vma, *tail_vma = NULL; | |
135 | loff_t l = *pos; | |
136 | ||
137 | /* Clear the per syscall fields in priv */ | |
138 | priv->task = NULL; | |
139 | priv->tail_vma = NULL; | |
140 | ||
141 | /* | |
142 | * We remember last_addr rather than next_addr to hit with | |
143 | * mmap_cache most of the time. We have zero last_addr at | |
144 | * the beginning and also after lseek. We will have -1 last_addr | |
145 | * after the end of the vmas. | |
146 | */ | |
147 | ||
148 | if (last_addr == -1UL) | |
149 | return NULL; | |
150 | ||
151 | priv->task = get_pid_task(priv->pid, PIDTYPE_PID); | |
152 | if (!priv->task) | |
153 | return NULL; | |
154 | ||
155 | mm = mm_for_maps(priv->task); | |
156 | if (!mm) | |
157 | return NULL; | |
158 | ||
159 | tail_vma = get_gate_vma(priv->task); | |
160 | priv->tail_vma = tail_vma; | |
161 | ||
162 | /* Start with last addr hint */ | |
163 | vma = find_vma(mm, last_addr); | |
164 | if (last_addr && vma) { | |
165 | vma = vma->vm_next; | |
166 | goto out; | |
167 | } | |
168 | ||
169 | /* | |
170 | * Check the vma index is within the range and do | |
171 | * sequential scan until m_index. | |
172 | */ | |
173 | vma = NULL; | |
174 | if ((unsigned long)l < mm->map_count) { | |
175 | vma = mm->mmap; | |
176 | while (l-- && vma) | |
177 | vma = vma->vm_next; | |
178 | goto out; | |
179 | } | |
180 | ||
181 | if (l != mm->map_count) | |
182 | tail_vma = NULL; /* After gate vma */ | |
183 | ||
184 | out: | |
185 | if (vma) | |
186 | return vma; | |
187 | ||
188 | /* End of vmas has been reached */ | |
189 | m->version = (tail_vma != NULL)? 0: -1UL; | |
190 | up_read(&mm->mmap_sem); | |
191 | mmput(mm); | |
192 | return tail_vma; | |
193 | } | |
194 | ||
195 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
196 | { | |
197 | struct proc_maps_private *priv = m->private; | |
198 | struct vm_area_struct *vma = v; | |
199 | struct vm_area_struct *tail_vma = priv->tail_vma; | |
200 | ||
201 | (*pos)++; | |
202 | if (vma && (vma != tail_vma) && vma->vm_next) | |
203 | return vma->vm_next; | |
204 | vma_stop(priv, vma); | |
205 | return (vma != tail_vma)? tail_vma: NULL; | |
206 | } | |
207 | ||
208 | static void m_stop(struct seq_file *m, void *v) | |
209 | { | |
210 | struct proc_maps_private *priv = m->private; | |
211 | struct vm_area_struct *vma = v; | |
212 | ||
213 | vma_stop(priv, vma); | |
214 | if (priv->task) | |
215 | put_task_struct(priv->task); | |
216 | } | |
217 | ||
218 | static int do_maps_open(struct inode *inode, struct file *file, | |
219 | struct seq_operations *ops) | |
220 | { | |
221 | struct proc_maps_private *priv; | |
222 | int ret = -ENOMEM; | |
223 | priv = kzalloc(sizeof(*priv), GFP_KERNEL); | |
224 | if (priv) { | |
225 | priv->pid = proc_pid(inode); | |
226 | ret = seq_open(file, ops); | |
227 | if (!ret) { | |
228 | struct seq_file *m = file->private_data; | |
229 | m->private = priv; | |
230 | } else { | |
231 | kfree(priv); | |
232 | } | |
233 | } | |
234 | return ret; | |
235 | } | |
e070ad49 | 236 | |
4752c369 | 237 | static int show_map(struct seq_file *m, void *v) |
1da177e4 | 238 | { |
99f89551 EB |
239 | struct proc_maps_private *priv = m->private; |
240 | struct task_struct *task = priv->task; | |
e070ad49 ML |
241 | struct vm_area_struct *vma = v; |
242 | struct mm_struct *mm = vma->vm_mm; | |
243 | struct file *file = vma->vm_file; | |
244 | int flags = vma->vm_flags; | |
1da177e4 LT |
245 | unsigned long ino = 0; |
246 | dev_t dev = 0; | |
247 | int len; | |
248 | ||
5096add8 KC |
249 | if (maps_protect && !ptrace_may_attach(task)) |
250 | return -EACCES; | |
251 | ||
1da177e4 | 252 | if (file) { |
2fddfeef | 253 | struct inode *inode = vma->vm_file->f_path.dentry->d_inode; |
1da177e4 LT |
254 | dev = inode->i_sb->s_dev; |
255 | ino = inode->i_ino; | |
256 | } | |
257 | ||
258 | seq_printf(m, "%08lx-%08lx %c%c%c%c %08lx %02x:%02x %lu %n", | |
e070ad49 ML |
259 | vma->vm_start, |
260 | vma->vm_end, | |
1da177e4 LT |
261 | flags & VM_READ ? 'r' : '-', |
262 | flags & VM_WRITE ? 'w' : '-', | |
263 | flags & VM_EXEC ? 'x' : '-', | |
264 | flags & VM_MAYSHARE ? 's' : 'p', | |
e070ad49 | 265 | vma->vm_pgoff << PAGE_SHIFT, |
1da177e4 LT |
266 | MAJOR(dev), MINOR(dev), ino, &len); |
267 | ||
268 | /* | |
269 | * Print the dentry name for named mappings, and a | |
270 | * special [heap] marker for the heap: | |
271 | */ | |
e070ad49 | 272 | if (file) { |
1da177e4 | 273 | pad_len_spaces(m, len); |
2fddfeef | 274 | seq_path(m, file->f_path.mnt, file->f_path.dentry, "\n"); |
1da177e4 | 275 | } else { |
e6e5494c IM |
276 | const char *name = arch_vma_name(vma); |
277 | if (!name) { | |
278 | if (mm) { | |
279 | if (vma->vm_start <= mm->start_brk && | |
e070ad49 | 280 | vma->vm_end >= mm->brk) { |
e6e5494c IM |
281 | name = "[heap]"; |
282 | } else if (vma->vm_start <= mm->start_stack && | |
283 | vma->vm_end >= mm->start_stack) { | |
284 | name = "[stack]"; | |
1da177e4 | 285 | } |
e6e5494c IM |
286 | } else { |
287 | name = "[vdso]"; | |
1da177e4 | 288 | } |
e6e5494c IM |
289 | } |
290 | if (name) { | |
1da177e4 | 291 | pad_len_spaces(m, len); |
e6e5494c | 292 | seq_puts(m, name); |
1da177e4 LT |
293 | } |
294 | } | |
295 | seq_putc(m, '\n'); | |
e070ad49 | 296 | |
e070ad49 ML |
297 | if (m->count < m->size) /* vma is copied successfully */ |
298 | m->version = (vma != get_gate_vma(task))? vma->vm_start: 0; | |
1da177e4 LT |
299 | return 0; |
300 | } | |
301 | ||
a6198797 MM |
302 | static struct seq_operations proc_pid_maps_op = { |
303 | .start = m_start, | |
304 | .next = m_next, | |
305 | .stop = m_stop, | |
306 | .show = show_map | |
307 | }; | |
308 | ||
309 | static int maps_open(struct inode *inode, struct file *file) | |
310 | { | |
311 | return do_maps_open(inode, file, &proc_pid_maps_op); | |
312 | } | |
313 | ||
314 | const struct file_operations proc_maps_operations = { | |
315 | .open = maps_open, | |
316 | .read = seq_read, | |
317 | .llseek = seq_lseek, | |
318 | .release = seq_release_private, | |
319 | }; | |
320 | ||
321 | /* | |
322 | * Proportional Set Size(PSS): my share of RSS. | |
323 | * | |
324 | * PSS of a process is the count of pages it has in memory, where each | |
325 | * page is divided by the number of processes sharing it. So if a | |
326 | * process has 1000 pages all to itself, and 1000 shared with one other | |
327 | * process, its PSS will be 1500. | |
328 | * | |
329 | * To keep (accumulated) division errors low, we adopt a 64bit | |
330 | * fixed-point pss counter to minimize division errors. So (pss >> | |
331 | * PSS_SHIFT) would be the real byte count. | |
332 | * | |
333 | * A shift of 12 before division means (assuming 4K page size): | |
334 | * - 1M 3-user-pages add up to 8KB errors; | |
335 | * - supports mapcount up to 2^24, or 16M; | |
336 | * - supports PSS up to 2^52 bytes, or 4PB. | |
337 | */ | |
338 | #define PSS_SHIFT 12 | |
339 | ||
340 | struct mem_size_stats | |
341 | { | |
342 | struct vm_area_struct *vma; | |
343 | unsigned long resident; | |
344 | unsigned long shared_clean; | |
345 | unsigned long shared_dirty; | |
346 | unsigned long private_clean; | |
347 | unsigned long private_dirty; | |
348 | unsigned long referenced; | |
349 | u64 pss; | |
350 | }; | |
351 | ||
b3ae5acb MM |
352 | static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
353 | void *private) | |
e070ad49 | 354 | { |
826fad1b | 355 | struct mem_size_stats *mss = private; |
b3ae5acb | 356 | struct vm_area_struct *vma = mss->vma; |
e070ad49 | 357 | pte_t *pte, ptent; |
705e87c0 | 358 | spinlock_t *ptl; |
e070ad49 | 359 | struct page *page; |
ec4dd3eb | 360 | int mapcount; |
e070ad49 | 361 | |
705e87c0 | 362 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
826fad1b | 363 | for (; addr != end; pte++, addr += PAGE_SIZE) { |
e070ad49 | 364 | ptent = *pte; |
705e87c0 | 365 | if (!pte_present(ptent)) |
e070ad49 ML |
366 | continue; |
367 | ||
368 | mss->resident += PAGE_SIZE; | |
ad820c5d NP |
369 | |
370 | page = vm_normal_page(vma, addr, ptent); | |
371 | if (!page) | |
e070ad49 ML |
372 | continue; |
373 | ||
f79f177c DR |
374 | /* Accumulate the size in pages that have been accessed. */ |
375 | if (pte_young(ptent) || PageReferenced(page)) | |
376 | mss->referenced += PAGE_SIZE; | |
ec4dd3eb FW |
377 | mapcount = page_mapcount(page); |
378 | if (mapcount >= 2) { | |
e070ad49 ML |
379 | if (pte_dirty(ptent)) |
380 | mss->shared_dirty += PAGE_SIZE; | |
381 | else | |
382 | mss->shared_clean += PAGE_SIZE; | |
ec4dd3eb | 383 | mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; |
e070ad49 ML |
384 | } else { |
385 | if (pte_dirty(ptent)) | |
386 | mss->private_dirty += PAGE_SIZE; | |
387 | else | |
388 | mss->private_clean += PAGE_SIZE; | |
ec4dd3eb | 389 | mss->pss += (PAGE_SIZE << PSS_SHIFT); |
e070ad49 | 390 | } |
826fad1b | 391 | } |
705e87c0 HD |
392 | pte_unmap_unlock(pte - 1, ptl); |
393 | cond_resched(); | |
b3ae5acb | 394 | return 0; |
e070ad49 ML |
395 | } |
396 | ||
b3ae5acb | 397 | static struct mm_walk smaps_walk = { .pmd_entry = smaps_pte_range }; |
e070ad49 ML |
398 | |
399 | static int show_smap(struct seq_file *m, void *v) | |
400 | { | |
401 | struct vm_area_struct *vma = v; | |
e070ad49 | 402 | struct mem_size_stats mss; |
4752c369 | 403 | int ret; |
e070ad49 ML |
404 | |
405 | memset(&mss, 0, sizeof mss); | |
b3ae5acb | 406 | mss.vma = vma; |
5ddfae16 | 407 | if (vma->vm_mm && !is_vm_hugetlb_page(vma)) |
b3ae5acb MM |
408 | walk_page_range(vma->vm_mm, vma->vm_start, vma->vm_end, |
409 | &smaps_walk, &mss); | |
4752c369 MM |
410 | |
411 | ret = show_map(m, v); | |
412 | if (ret) | |
413 | return ret; | |
414 | ||
415 | seq_printf(m, | |
416 | "Size: %8lu kB\n" | |
417 | "Rss: %8lu kB\n" | |
418 | "Pss: %8lu kB\n" | |
419 | "Shared_Clean: %8lu kB\n" | |
420 | "Shared_Dirty: %8lu kB\n" | |
421 | "Private_Clean: %8lu kB\n" | |
422 | "Private_Dirty: %8lu kB\n" | |
423 | "Referenced: %8lu kB\n", | |
424 | (vma->vm_end - vma->vm_start) >> 10, | |
425 | mss.resident >> 10, | |
426 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), | |
427 | mss.shared_clean >> 10, | |
428 | mss.shared_dirty >> 10, | |
429 | mss.private_clean >> 10, | |
430 | mss.private_dirty >> 10, | |
431 | mss.referenced >> 10); | |
432 | ||
433 | return ret; | |
e070ad49 ML |
434 | } |
435 | ||
a6198797 MM |
436 | static struct seq_operations proc_pid_smaps_op = { |
437 | .start = m_start, | |
438 | .next = m_next, | |
439 | .stop = m_stop, | |
440 | .show = show_smap | |
441 | }; | |
442 | ||
443 | static int smaps_open(struct inode *inode, struct file *file) | |
444 | { | |
445 | return do_maps_open(inode, file, &proc_pid_smaps_op); | |
446 | } | |
447 | ||
448 | const struct file_operations proc_smaps_operations = { | |
449 | .open = smaps_open, | |
450 | .read = seq_read, | |
451 | .llseek = seq_lseek, | |
452 | .release = seq_release_private, | |
453 | }; | |
454 | ||
455 | static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, | |
456 | unsigned long end, void *private) | |
457 | { | |
458 | struct vm_area_struct *vma = private; | |
459 | pte_t *pte, ptent; | |
460 | spinlock_t *ptl; | |
461 | struct page *page; | |
462 | ||
463 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
464 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
465 | ptent = *pte; | |
466 | if (!pte_present(ptent)) | |
467 | continue; | |
468 | ||
469 | page = vm_normal_page(vma, addr, ptent); | |
470 | if (!page) | |
471 | continue; | |
472 | ||
473 | /* Clear accessed and referenced bits. */ | |
474 | ptep_test_and_clear_young(vma, addr, pte); | |
475 | ClearPageReferenced(page); | |
476 | } | |
477 | pte_unmap_unlock(pte - 1, ptl); | |
478 | cond_resched(); | |
479 | return 0; | |
480 | } | |
481 | ||
b3ae5acb MM |
482 | static struct mm_walk clear_refs_walk = { .pmd_entry = clear_refs_pte_range }; |
483 | ||
f248dcb3 MM |
484 | static ssize_t clear_refs_write(struct file *file, const char __user *buf, |
485 | size_t count, loff_t *ppos) | |
b813e931 | 486 | { |
f248dcb3 MM |
487 | struct task_struct *task; |
488 | char buffer[PROC_NUMBUF], *end; | |
489 | struct mm_struct *mm; | |
b813e931 DR |
490 | struct vm_area_struct *vma; |
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; | |
497 | if (!simple_strtol(buffer, &end, 0)) | |
498 | return -EINVAL; | |
499 | if (*end == '\n') | |
500 | end++; | |
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) { | |
506 | down_read(&mm->mmap_sem); | |
507 | for (vma = mm->mmap; vma; vma = vma->vm_next) | |
508 | if (!is_vm_hugetlb_page(vma)) | |
509 | walk_page_range(mm, vma->vm_start, vma->vm_end, | |
510 | &clear_refs_walk, vma); | |
511 | flush_tlb_mm(mm); | |
512 | up_read(&mm->mmap_sem); | |
513 | mmput(mm); | |
514 | } | |
515 | put_task_struct(task); | |
516 | if (end - buffer == 0) | |
517 | return -EIO; | |
518 | return end - buffer; | |
b813e931 DR |
519 | } |
520 | ||
f248dcb3 MM |
521 | const struct file_operations proc_clear_refs_operations = { |
522 | .write = clear_refs_write, | |
523 | }; | |
524 | ||
85863e47 MM |
525 | struct pagemapread { |
526 | char __user *out, *end; | |
527 | }; | |
528 | ||
529 | #define PM_ENTRY_BYTES sizeof(u64) | |
530 | #define PM_RESERVED_BITS 3 | |
531 | #define PM_RESERVED_OFFSET (64 - PM_RESERVED_BITS) | |
532 | #define PM_RESERVED_MASK (((1LL<<PM_RESERVED_BITS)-1) << PM_RESERVED_OFFSET) | |
533 | #define PM_SPECIAL(nr) (((nr) << PM_RESERVED_OFFSET) | PM_RESERVED_MASK) | |
534 | #define PM_NOT_PRESENT PM_SPECIAL(1LL) | |
535 | #define PM_SWAP PM_SPECIAL(2LL) | |
536 | #define PM_END_OF_BUFFER 1 | |
537 | ||
538 | static int add_to_pagemap(unsigned long addr, u64 pfn, | |
539 | struct pagemapread *pm) | |
540 | { | |
541 | /* | |
542 | * Make sure there's room in the buffer for an | |
543 | * entire entry. Otherwise, only copy part of | |
544 | * the pfn. | |
545 | */ | |
546 | if (pm->out + PM_ENTRY_BYTES >= pm->end) { | |
547 | if (copy_to_user(pm->out, &pfn, pm->end - pm->out)) | |
548 | return -EFAULT; | |
549 | pm->out = pm->end; | |
550 | return PM_END_OF_BUFFER; | |
551 | } | |
552 | ||
553 | if (put_user(pfn, pm->out)) | |
554 | return -EFAULT; | |
555 | pm->out += PM_ENTRY_BYTES; | |
556 | return 0; | |
557 | } | |
558 | ||
559 | static int pagemap_pte_hole(unsigned long start, unsigned long end, | |
560 | void *private) | |
561 | { | |
562 | struct pagemapread *pm = private; | |
563 | unsigned long addr; | |
564 | int err = 0; | |
565 | for (addr = start; addr < end; addr += PAGE_SIZE) { | |
566 | err = add_to_pagemap(addr, PM_NOT_PRESENT, pm); | |
567 | if (err) | |
568 | break; | |
569 | } | |
570 | return err; | |
571 | } | |
572 | ||
573 | u64 swap_pte_to_pagemap_entry(pte_t pte) | |
574 | { | |
575 | swp_entry_t e = pte_to_swp_entry(pte); | |
576 | return PM_SWAP | swp_type(e) | (swp_offset(e) << MAX_SWAPFILES_SHIFT); | |
577 | } | |
578 | ||
579 | static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, | |
580 | void *private) | |
581 | { | |
582 | struct pagemapread *pm = private; | |
583 | pte_t *pte; | |
584 | int err = 0; | |
585 | ||
586 | for (; addr != end; addr += PAGE_SIZE) { | |
587 | u64 pfn = PM_NOT_PRESENT; | |
588 | pte = pte_offset_map(pmd, addr); | |
589 | if (is_swap_pte(*pte)) | |
590 | pfn = swap_pte_to_pagemap_entry(*pte); | |
591 | else if (pte_present(*pte)) | |
592 | pfn = pte_pfn(*pte); | |
593 | /* unmap so we're not in atomic when we copy to userspace */ | |
594 | pte_unmap(pte); | |
595 | err = add_to_pagemap(addr, pfn, pm); | |
596 | if (err) | |
597 | return err; | |
598 | } | |
599 | ||
600 | cond_resched(); | |
601 | ||
602 | return err; | |
603 | } | |
604 | ||
605 | static struct mm_walk pagemap_walk = { | |
606 | .pmd_entry = pagemap_pte_range, | |
607 | .pte_hole = pagemap_pte_hole | |
608 | }; | |
609 | ||
610 | /* | |
611 | * /proc/pid/pagemap - an array mapping virtual pages to pfns | |
612 | * | |
613 | * For each page in the address space, this file contains one 64-bit | |
614 | * entry representing the corresponding physical page frame number | |
615 | * (PFN) if the page is present. If there is a swap entry for the | |
616 | * physical page, then an encoding of the swap file number and the | |
617 | * page's offset into the swap file are returned. If no page is | |
618 | * present at all, PM_NOT_PRESENT is returned. This allows determining | |
619 | * precisely which pages are mapped (or in swap) and comparing mapped | |
620 | * pages between processes. | |
621 | * | |
622 | * Efficient users of this interface will use /proc/pid/maps to | |
623 | * determine which areas of memory are actually mapped and llseek to | |
624 | * skip over unmapped regions. | |
625 | */ | |
626 | static ssize_t pagemap_read(struct file *file, char __user *buf, | |
627 | size_t count, loff_t *ppos) | |
628 | { | |
629 | struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); | |
630 | struct page **pages, *page; | |
631 | unsigned long uaddr, uend; | |
632 | struct mm_struct *mm; | |
633 | struct pagemapread pm; | |
634 | int pagecount; | |
635 | int ret = -ESRCH; | |
636 | ||
637 | if (!task) | |
638 | goto out; | |
639 | ||
640 | ret = -EACCES; | |
641 | if (!ptrace_may_attach(task)) | |
642 | goto out; | |
643 | ||
644 | ret = -EINVAL; | |
645 | /* file position must be aligned */ | |
646 | if (*ppos % PM_ENTRY_BYTES) | |
647 | goto out; | |
648 | ||
649 | ret = 0; | |
650 | mm = get_task_mm(task); | |
651 | if (!mm) | |
652 | goto out; | |
653 | ||
654 | ret = -ENOMEM; | |
655 | uaddr = (unsigned long)buf & PAGE_MASK; | |
656 | uend = (unsigned long)(buf + count); | |
657 | pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE; | |
658 | pages = kmalloc(pagecount * sizeof(struct page *), GFP_KERNEL); | |
659 | if (!pages) | |
660 | goto out_task; | |
661 | ||
662 | down_read(¤t->mm->mmap_sem); | |
663 | ret = get_user_pages(current, current->mm, uaddr, pagecount, | |
664 | 1, 0, pages, NULL); | |
665 | up_read(¤t->mm->mmap_sem); | |
666 | ||
667 | if (ret < 0) | |
668 | goto out_free; | |
669 | ||
670 | pm.out = buf; | |
671 | pm.end = buf + count; | |
672 | ||
673 | if (!ptrace_may_attach(task)) { | |
674 | ret = -EIO; | |
675 | } else { | |
676 | unsigned long src = *ppos; | |
677 | unsigned long svpfn = src / PM_ENTRY_BYTES; | |
678 | unsigned long start_vaddr = svpfn << PAGE_SHIFT; | |
679 | unsigned long end_vaddr = TASK_SIZE_OF(task); | |
680 | ||
681 | /* watch out for wraparound */ | |
682 | if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) | |
683 | start_vaddr = end_vaddr; | |
684 | ||
685 | /* | |
686 | * The odds are that this will stop walking way | |
687 | * before end_vaddr, because the length of the | |
688 | * user buffer is tracked in "pm", and the walk | |
689 | * will stop when we hit the end of the buffer. | |
690 | */ | |
691 | ret = walk_page_range(mm, start_vaddr, end_vaddr, | |
692 | &pagemap_walk, &pm); | |
693 | if (ret == PM_END_OF_BUFFER) | |
694 | ret = 0; | |
695 | /* don't need mmap_sem for these, but this looks cleaner */ | |
696 | *ppos += pm.out - buf; | |
697 | if (!ret) | |
698 | ret = pm.out - buf; | |
699 | } | |
700 | ||
701 | for (; pagecount; pagecount--) { | |
702 | page = pages[pagecount-1]; | |
703 | if (!PageReserved(page)) | |
704 | SetPageDirty(page); | |
705 | page_cache_release(page); | |
706 | } | |
707 | mmput(mm); | |
708 | out_free: | |
709 | kfree(pages); | |
710 | out_task: | |
711 | put_task_struct(task); | |
712 | out: | |
713 | return ret; | |
714 | } | |
715 | ||
716 | const struct file_operations proc_pagemap_operations = { | |
717 | .llseek = mem_lseek, /* borrow this */ | |
718 | .read = pagemap_read, | |
719 | }; | |
720 | ||
6e21c8f1 | 721 | #ifdef CONFIG_NUMA |
1a75a6c8 | 722 | extern int show_numa_map(struct seq_file *m, void *v); |
6e21c8f1 | 723 | |
5096add8 KC |
724 | static int show_numa_map_checked(struct seq_file *m, void *v) |
725 | { | |
726 | struct proc_maps_private *priv = m->private; | |
727 | struct task_struct *task = priv->task; | |
728 | ||
729 | if (maps_protect && !ptrace_may_attach(task)) | |
730 | return -EACCES; | |
731 | ||
732 | return show_numa_map(m, v); | |
733 | } | |
734 | ||
662795de | 735 | static struct seq_operations proc_pid_numa_maps_op = { |
1a75a6c8 CL |
736 | .start = m_start, |
737 | .next = m_next, | |
738 | .stop = m_stop, | |
5096add8 | 739 | .show = show_numa_map_checked |
6e21c8f1 | 740 | }; |
662795de EB |
741 | |
742 | static int numa_maps_open(struct inode *inode, struct file *file) | |
743 | { | |
744 | return do_maps_open(inode, file, &proc_pid_numa_maps_op); | |
745 | } | |
746 | ||
00977a59 | 747 | const struct file_operations proc_numa_maps_operations = { |
662795de EB |
748 | .open = numa_maps_open, |
749 | .read = seq_read, | |
750 | .llseek = seq_lseek, | |
99f89551 | 751 | .release = seq_release_private, |
662795de | 752 | }; |
6e21c8f1 | 753 | #endif |