Commit | Line | Data |
---|---|---|
1da177e4 | 1 | #include <linux/mm.h> |
615d6e87 | 2 | #include <linux/vmacache.h> |
1da177e4 | 3 | #include <linux/hugetlb.h> |
22e057c5 | 4 | #include <linux/huge_mm.h> |
1da177e4 LT |
5 | #include <linux/mount.h> |
6 | #include <linux/seq_file.h> | |
e070ad49 | 7 | #include <linux/highmem.h> |
5096add8 | 8 | #include <linux/ptrace.h> |
5a0e3ad6 | 9 | #include <linux/slab.h> |
6e21c8f1 CL |
10 | #include <linux/pagemap.h> |
11 | #include <linux/mempolicy.h> | |
22e057c5 | 12 | #include <linux/rmap.h> |
85863e47 | 13 | #include <linux/swap.h> |
6e84f315 | 14 | #include <linux/sched/mm.h> |
85863e47 | 15 | #include <linux/swapops.h> |
0f8975ec | 16 | #include <linux/mmu_notifier.h> |
33c3fc71 | 17 | #include <linux/page_idle.h> |
6a15a370 | 18 | #include <linux/shmem_fs.h> |
e070ad49 | 19 | |
1da177e4 | 20 | #include <asm/elf.h> |
7c0f6ba6 | 21 | #include <linux/uaccess.h> |
e070ad49 | 22 | #include <asm/tlbflush.h> |
1da177e4 LT |
23 | #include "internal.h" |
24 | ||
df5f8314 | 25 | void task_mem(struct seq_file *m, struct mm_struct *mm) |
1da177e4 | 26 | { |
84638335 | 27 | unsigned long text, lib, swap, ptes, pmds, anon, file, shmem; |
365e9c87 HD |
28 | unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; |
29 | ||
8cee852e JM |
30 | anon = get_mm_counter(mm, MM_ANONPAGES); |
31 | file = get_mm_counter(mm, MM_FILEPAGES); | |
32 | shmem = get_mm_counter(mm, MM_SHMEMPAGES); | |
33 | ||
365e9c87 HD |
34 | /* |
35 | * Note: to minimize their overhead, mm maintains hiwater_vm and | |
36 | * hiwater_rss only when about to *lower* total_vm or rss. Any | |
37 | * collector of these hiwater stats must therefore get total_vm | |
38 | * and rss too, which will usually be the higher. Barriers? not | |
39 | * worth the effort, such snapshots can always be inconsistent. | |
40 | */ | |
41 | hiwater_vm = total_vm = mm->total_vm; | |
42 | if (hiwater_vm < mm->hiwater_vm) | |
43 | hiwater_vm = mm->hiwater_vm; | |
8cee852e | 44 | hiwater_rss = total_rss = anon + file + shmem; |
365e9c87 HD |
45 | if (hiwater_rss < mm->hiwater_rss) |
46 | hiwater_rss = mm->hiwater_rss; | |
1da177e4 | 47 | |
1da177e4 LT |
48 | text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; |
49 | lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; | |
b084d435 | 50 | swap = get_mm_counter(mm, MM_SWAPENTS); |
dc6c9a35 KS |
51 | ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes); |
52 | pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm); | |
df5f8314 | 53 | seq_printf(m, |
365e9c87 | 54 | "VmPeak:\t%8lu kB\n" |
1da177e4 LT |
55 | "VmSize:\t%8lu kB\n" |
56 | "VmLck:\t%8lu kB\n" | |
bc3e53f6 | 57 | "VmPin:\t%8lu kB\n" |
365e9c87 | 58 | "VmHWM:\t%8lu kB\n" |
1da177e4 | 59 | "VmRSS:\t%8lu kB\n" |
8cee852e JM |
60 | "RssAnon:\t%8lu kB\n" |
61 | "RssFile:\t%8lu kB\n" | |
62 | "RssShmem:\t%8lu kB\n" | |
1da177e4 LT |
63 | "VmData:\t%8lu kB\n" |
64 | "VmStk:\t%8lu kB\n" | |
65 | "VmExe:\t%8lu kB\n" | |
66 | "VmLib:\t%8lu kB\n" | |
b084d435 | 67 | "VmPTE:\t%8lu kB\n" |
dc6c9a35 | 68 | "VmPMD:\t%8lu kB\n" |
b084d435 | 69 | "VmSwap:\t%8lu kB\n", |
365e9c87 | 70 | hiwater_vm << (PAGE_SHIFT-10), |
314e51b9 | 71 | total_vm << (PAGE_SHIFT-10), |
1da177e4 | 72 | mm->locked_vm << (PAGE_SHIFT-10), |
bc3e53f6 | 73 | mm->pinned_vm << (PAGE_SHIFT-10), |
365e9c87 HD |
74 | hiwater_rss << (PAGE_SHIFT-10), |
75 | total_rss << (PAGE_SHIFT-10), | |
8cee852e JM |
76 | anon << (PAGE_SHIFT-10), |
77 | file << (PAGE_SHIFT-10), | |
78 | shmem << (PAGE_SHIFT-10), | |
84638335 | 79 | mm->data_vm << (PAGE_SHIFT-10), |
1da177e4 | 80 | mm->stack_vm << (PAGE_SHIFT-10), text, lib, |
dc6c9a35 KS |
81 | ptes >> 10, |
82 | pmds >> 10, | |
b084d435 | 83 | swap << (PAGE_SHIFT-10)); |
5d317b2b | 84 | hugetlb_report_usage(m, mm); |
1da177e4 LT |
85 | } |
86 | ||
87 | unsigned long task_vsize(struct mm_struct *mm) | |
88 | { | |
89 | return PAGE_SIZE * mm->total_vm; | |
90 | } | |
91 | ||
a2ade7b6 AD |
92 | unsigned long task_statm(struct mm_struct *mm, |
93 | unsigned long *shared, unsigned long *text, | |
94 | unsigned long *data, unsigned long *resident) | |
1da177e4 | 95 | { |
eca56ff9 JM |
96 | *shared = get_mm_counter(mm, MM_FILEPAGES) + |
97 | get_mm_counter(mm, MM_SHMEMPAGES); | |
1da177e4 LT |
98 | *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) |
99 | >> PAGE_SHIFT; | |
84638335 | 100 | *data = mm->data_vm + mm->stack_vm; |
d559db08 | 101 | *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); |
1da177e4 LT |
102 | return mm->total_vm; |
103 | } | |
104 | ||
9e781440 KH |
105 | #ifdef CONFIG_NUMA |
106 | /* | |
498f2371 | 107 | * Save get_task_policy() for show_numa_map(). |
9e781440 KH |
108 | */ |
109 | static void hold_task_mempolicy(struct proc_maps_private *priv) | |
110 | { | |
111 | struct task_struct *task = priv->task; | |
112 | ||
113 | task_lock(task); | |
498f2371 | 114 | priv->task_mempolicy = get_task_policy(task); |
9e781440 KH |
115 | mpol_get(priv->task_mempolicy); |
116 | task_unlock(task); | |
117 | } | |
118 | static void release_task_mempolicy(struct proc_maps_private *priv) | |
119 | { | |
120 | mpol_put(priv->task_mempolicy); | |
121 | } | |
122 | #else | |
123 | static void hold_task_mempolicy(struct proc_maps_private *priv) | |
124 | { | |
125 | } | |
126 | static void release_task_mempolicy(struct proc_maps_private *priv) | |
127 | { | |
128 | } | |
129 | #endif | |
130 | ||
59b4bf12 | 131 | static void vma_stop(struct proc_maps_private *priv) |
a6198797 | 132 | { |
59b4bf12 ON |
133 | struct mm_struct *mm = priv->mm; |
134 | ||
135 | release_task_mempolicy(priv); | |
136 | up_read(&mm->mmap_sem); | |
137 | mmput(mm); | |
a6198797 | 138 | } |
ec4dd3eb | 139 | |
ad2a00e4 ON |
140 | static struct vm_area_struct * |
141 | m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma) | |
142 | { | |
143 | if (vma == priv->tail_vma) | |
144 | return NULL; | |
145 | return vma->vm_next ?: priv->tail_vma; | |
146 | } | |
147 | ||
b8c20a9b ON |
148 | static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma) |
149 | { | |
150 | if (m->count < m->size) /* vma is copied successfully */ | |
855af072 | 151 | m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL; |
b8c20a9b ON |
152 | } |
153 | ||
0c255321 | 154 | static void *m_start(struct seq_file *m, loff_t *ppos) |
e070ad49 | 155 | { |
a6198797 | 156 | struct proc_maps_private *priv = m->private; |
b8c20a9b | 157 | unsigned long last_addr = m->version; |
a6198797 | 158 | struct mm_struct *mm; |
0c255321 ON |
159 | struct vm_area_struct *vma; |
160 | unsigned int pos = *ppos; | |
a6198797 | 161 | |
b8c20a9b ON |
162 | /* See m_cache_vma(). Zero at the start or after lseek. */ |
163 | if (last_addr == -1UL) | |
164 | return NULL; | |
165 | ||
2c03376d | 166 | priv->task = get_proc_task(priv->inode); |
a6198797 | 167 | if (!priv->task) |
ec6fd8a4 | 168 | return ERR_PTR(-ESRCH); |
a6198797 | 169 | |
29a40ace | 170 | mm = priv->mm; |
388f7934 | 171 | if (!mm || !mmget_not_zero(mm)) |
29a40ace | 172 | return NULL; |
a6198797 | 173 | |
0c255321 | 174 | down_read(&mm->mmap_sem); |
9e781440 | 175 | hold_task_mempolicy(priv); |
0c255321 | 176 | priv->tail_vma = get_gate_vma(mm); |
a6198797 | 177 | |
b8c20a9b | 178 | if (last_addr) { |
855af072 RH |
179 | vma = find_vma(mm, last_addr - 1); |
180 | if (vma && vma->vm_start <= last_addr) | |
181 | vma = m_next_vma(priv, vma); | |
182 | if (vma) | |
b8c20a9b ON |
183 | return vma; |
184 | } | |
185 | ||
186 | m->version = 0; | |
0c255321 | 187 | if (pos < mm->map_count) { |
557c2d8a ON |
188 | for (vma = mm->mmap; pos; pos--) { |
189 | m->version = vma->vm_start; | |
a6198797 | 190 | vma = vma->vm_next; |
557c2d8a | 191 | } |
a6198797 | 192 | return vma; |
0c255321 | 193 | } |
a6198797 | 194 | |
557c2d8a | 195 | /* we do not bother to update m->version in this case */ |
0c255321 ON |
196 | if (pos == mm->map_count && priv->tail_vma) |
197 | return priv->tail_vma; | |
59b4bf12 ON |
198 | |
199 | vma_stop(priv); | |
200 | return NULL; | |
a6198797 MM |
201 | } |
202 | ||
203 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
204 | { | |
205 | struct proc_maps_private *priv = m->private; | |
ad2a00e4 | 206 | struct vm_area_struct *next; |
a6198797 MM |
207 | |
208 | (*pos)++; | |
ad2a00e4 | 209 | next = m_next_vma(priv, v); |
59b4bf12 ON |
210 | if (!next) |
211 | vma_stop(priv); | |
212 | return next; | |
a6198797 MM |
213 | } |
214 | ||
215 | static void m_stop(struct seq_file *m, void *v) | |
216 | { | |
217 | struct proc_maps_private *priv = m->private; | |
a6198797 | 218 | |
59b4bf12 ON |
219 | if (!IS_ERR_OR_NULL(v)) |
220 | vma_stop(priv); | |
0d5f5f45 | 221 | if (priv->task) { |
a6198797 | 222 | put_task_struct(priv->task); |
0d5f5f45 ON |
223 | priv->task = NULL; |
224 | } | |
a6198797 MM |
225 | } |
226 | ||
4db7d0ee ON |
227 | static int proc_maps_open(struct inode *inode, struct file *file, |
228 | const struct seq_operations *ops, int psize) | |
229 | { | |
230 | struct proc_maps_private *priv = __seq_open_private(file, ops, psize); | |
231 | ||
232 | if (!priv) | |
233 | return -ENOMEM; | |
234 | ||
2c03376d | 235 | priv->inode = inode; |
29a40ace ON |
236 | priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); |
237 | if (IS_ERR(priv->mm)) { | |
238 | int err = PTR_ERR(priv->mm); | |
239 | ||
240 | seq_release_private(inode, file); | |
241 | return err; | |
242 | } | |
243 | ||
4db7d0ee ON |
244 | return 0; |
245 | } | |
246 | ||
29a40ace ON |
247 | static int proc_map_release(struct inode *inode, struct file *file) |
248 | { | |
249 | struct seq_file *seq = file->private_data; | |
250 | struct proc_maps_private *priv = seq->private; | |
251 | ||
252 | if (priv->mm) | |
253 | mmdrop(priv->mm); | |
254 | ||
255 | return seq_release_private(inode, file); | |
256 | } | |
257 | ||
a6198797 | 258 | static int do_maps_open(struct inode *inode, struct file *file, |
03a44825 | 259 | const struct seq_operations *ops) |
a6198797 | 260 | { |
4db7d0ee ON |
261 | return proc_maps_open(inode, file, ops, |
262 | sizeof(struct proc_maps_private)); | |
a6198797 | 263 | } |
e070ad49 | 264 | |
65376df5 JW |
265 | /* |
266 | * Indicate if the VMA is a stack for the given task; for | |
267 | * /proc/PID/maps that is the stack of the main task. | |
268 | */ | |
269 | static int is_stack(struct proc_maps_private *priv, | |
b18cb64e | 270 | struct vm_area_struct *vma) |
58cb6548 | 271 | { |
b18cb64e AL |
272 | /* |
273 | * We make no effort to guess what a given thread considers to be | |
274 | * its "stack". It's not even well-defined for programs written | |
275 | * languages like Go. | |
276 | */ | |
277 | return vma->vm_start <= vma->vm_mm->start_stack && | |
278 | vma->vm_end >= vma->vm_mm->start_stack; | |
58cb6548 ON |
279 | } |
280 | ||
b7643757 SP |
281 | static void |
282 | show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid) | |
1da177e4 | 283 | { |
e070ad49 ML |
284 | struct mm_struct *mm = vma->vm_mm; |
285 | struct file *file = vma->vm_file; | |
b7643757 | 286 | struct proc_maps_private *priv = m->private; |
ca16d140 | 287 | vm_flags_t flags = vma->vm_flags; |
1da177e4 | 288 | unsigned long ino = 0; |
6260a4b0 | 289 | unsigned long long pgoff = 0; |
a09a79f6 | 290 | unsigned long start, end; |
1da177e4 | 291 | dev_t dev = 0; |
b7643757 | 292 | const char *name = NULL; |
1da177e4 LT |
293 | |
294 | if (file) { | |
496ad9aa | 295 | struct inode *inode = file_inode(vma->vm_file); |
1da177e4 LT |
296 | dev = inode->i_sb->s_dev; |
297 | ino = inode->i_ino; | |
6260a4b0 | 298 | pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; |
1da177e4 LT |
299 | } |
300 | ||
d7824370 LT |
301 | /* We don't show the stack guard page in /proc/maps */ |
302 | start = vma->vm_start; | |
a09a79f6 MP |
303 | if (stack_guard_page_start(vma, start)) |
304 | start += PAGE_SIZE; | |
305 | end = vma->vm_end; | |
306 | if (stack_guard_page_end(vma, end)) | |
307 | end -= PAGE_SIZE; | |
d7824370 | 308 | |
652586df TH |
309 | seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); |
310 | seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ", | |
d7824370 | 311 | start, |
a09a79f6 | 312 | end, |
1da177e4 LT |
313 | flags & VM_READ ? 'r' : '-', |
314 | flags & VM_WRITE ? 'w' : '-', | |
315 | flags & VM_EXEC ? 'x' : '-', | |
316 | flags & VM_MAYSHARE ? 's' : 'p', | |
6260a4b0 | 317 | pgoff, |
652586df | 318 | MAJOR(dev), MINOR(dev), ino); |
1da177e4 LT |
319 | |
320 | /* | |
321 | * Print the dentry name for named mappings, and a | |
322 | * special [heap] marker for the heap: | |
323 | */ | |
e070ad49 | 324 | if (file) { |
652586df | 325 | seq_pad(m, ' '); |
2726d566 | 326 | seq_file_path(m, file, "\n"); |
b7643757 SP |
327 | goto done; |
328 | } | |
329 | ||
78d683e8 AL |
330 | if (vma->vm_ops && vma->vm_ops->name) { |
331 | name = vma->vm_ops->name(vma); | |
332 | if (name) | |
333 | goto done; | |
334 | } | |
335 | ||
b7643757 SP |
336 | name = arch_vma_name(vma); |
337 | if (!name) { | |
b7643757 SP |
338 | if (!mm) { |
339 | name = "[vdso]"; | |
340 | goto done; | |
341 | } | |
342 | ||
343 | if (vma->vm_start <= mm->brk && | |
344 | vma->vm_end >= mm->start_brk) { | |
345 | name = "[heap]"; | |
346 | goto done; | |
347 | } | |
348 | ||
b18cb64e | 349 | if (is_stack(priv, vma)) |
65376df5 | 350 | name = "[stack]"; |
b7643757 SP |
351 | } |
352 | ||
353 | done: | |
354 | if (name) { | |
652586df | 355 | seq_pad(m, ' '); |
b7643757 | 356 | seq_puts(m, name); |
1da177e4 LT |
357 | } |
358 | seq_putc(m, '\n'); | |
7c88db0c JK |
359 | } |
360 | ||
b7643757 | 361 | static int show_map(struct seq_file *m, void *v, int is_pid) |
7c88db0c | 362 | { |
ebb6cdde | 363 | show_map_vma(m, v, is_pid); |
b8c20a9b | 364 | m_cache_vma(m, v); |
1da177e4 LT |
365 | return 0; |
366 | } | |
367 | ||
b7643757 SP |
368 | static int show_pid_map(struct seq_file *m, void *v) |
369 | { | |
370 | return show_map(m, v, 1); | |
371 | } | |
372 | ||
373 | static int show_tid_map(struct seq_file *m, void *v) | |
374 | { | |
375 | return show_map(m, v, 0); | |
376 | } | |
377 | ||
03a44825 | 378 | static const struct seq_operations proc_pid_maps_op = { |
a6198797 MM |
379 | .start = m_start, |
380 | .next = m_next, | |
381 | .stop = m_stop, | |
b7643757 SP |
382 | .show = show_pid_map |
383 | }; | |
384 | ||
385 | static const struct seq_operations proc_tid_maps_op = { | |
386 | .start = m_start, | |
387 | .next = m_next, | |
388 | .stop = m_stop, | |
389 | .show = show_tid_map | |
a6198797 MM |
390 | }; |
391 | ||
b7643757 | 392 | static int pid_maps_open(struct inode *inode, struct file *file) |
a6198797 MM |
393 | { |
394 | return do_maps_open(inode, file, &proc_pid_maps_op); | |
395 | } | |
396 | ||
b7643757 SP |
397 | static int tid_maps_open(struct inode *inode, struct file *file) |
398 | { | |
399 | return do_maps_open(inode, file, &proc_tid_maps_op); | |
400 | } | |
401 | ||
402 | const struct file_operations proc_pid_maps_operations = { | |
403 | .open = pid_maps_open, | |
404 | .read = seq_read, | |
405 | .llseek = seq_lseek, | |
29a40ace | 406 | .release = proc_map_release, |
b7643757 SP |
407 | }; |
408 | ||
409 | const struct file_operations proc_tid_maps_operations = { | |
410 | .open = tid_maps_open, | |
a6198797 MM |
411 | .read = seq_read, |
412 | .llseek = seq_lseek, | |
29a40ace | 413 | .release = proc_map_release, |
a6198797 MM |
414 | }; |
415 | ||
416 | /* | |
417 | * Proportional Set Size(PSS): my share of RSS. | |
418 | * | |
419 | * PSS of a process is the count of pages it has in memory, where each | |
420 | * page is divided by the number of processes sharing it. So if a | |
421 | * process has 1000 pages all to itself, and 1000 shared with one other | |
422 | * process, its PSS will be 1500. | |
423 | * | |
424 | * To keep (accumulated) division errors low, we adopt a 64bit | |
425 | * fixed-point pss counter to minimize division errors. So (pss >> | |
426 | * PSS_SHIFT) would be the real byte count. | |
427 | * | |
428 | * A shift of 12 before division means (assuming 4K page size): | |
429 | * - 1M 3-user-pages add up to 8KB errors; | |
430 | * - supports mapcount up to 2^24, or 16M; | |
431 | * - supports PSS up to 2^52 bytes, or 4PB. | |
432 | */ | |
433 | #define PSS_SHIFT 12 | |
434 | ||
1e883281 | 435 | #ifdef CONFIG_PROC_PAGE_MONITOR |
214e471f | 436 | struct mem_size_stats { |
a6198797 MM |
437 | unsigned long resident; |
438 | unsigned long shared_clean; | |
439 | unsigned long shared_dirty; | |
440 | unsigned long private_clean; | |
441 | unsigned long private_dirty; | |
442 | unsigned long referenced; | |
b40d4f84 | 443 | unsigned long anonymous; |
4031a219 | 444 | unsigned long anonymous_thp; |
65c45377 | 445 | unsigned long shmem_thp; |
214e471f | 446 | unsigned long swap; |
25ee01a2 NH |
447 | unsigned long shared_hugetlb; |
448 | unsigned long private_hugetlb; | |
a6198797 | 449 | u64 pss; |
8334b962 | 450 | u64 swap_pss; |
c261e7d9 | 451 | bool check_shmem_swap; |
a6198797 MM |
452 | }; |
453 | ||
c164e038 | 454 | static void smaps_account(struct mem_size_stats *mss, struct page *page, |
afd9883f | 455 | bool compound, bool young, bool dirty) |
c164e038 | 456 | { |
f4be6153 | 457 | int i, nr = compound ? 1 << compound_order(page) : 1; |
afd9883f | 458 | unsigned long size = nr * PAGE_SIZE; |
c164e038 KS |
459 | |
460 | if (PageAnon(page)) | |
461 | mss->anonymous += size; | |
462 | ||
463 | mss->resident += size; | |
464 | /* Accumulate the size in pages that have been accessed. */ | |
33c3fc71 | 465 | if (young || page_is_young(page) || PageReferenced(page)) |
c164e038 | 466 | mss->referenced += size; |
c164e038 | 467 | |
afd9883f KS |
468 | /* |
469 | * page_count(page) == 1 guarantees the page is mapped exactly once. | |
470 | * If any subpage of the compound page mapped with PTE it would elevate | |
471 | * page_count(). | |
472 | */ | |
473 | if (page_count(page) == 1) { | |
c164e038 KS |
474 | if (dirty || PageDirty(page)) |
475 | mss->private_dirty += size; | |
476 | else | |
477 | mss->private_clean += size; | |
478 | mss->pss += (u64)size << PSS_SHIFT; | |
afd9883f KS |
479 | return; |
480 | } | |
481 | ||
482 | for (i = 0; i < nr; i++, page++) { | |
483 | int mapcount = page_mapcount(page); | |
484 | ||
485 | if (mapcount >= 2) { | |
486 | if (dirty || PageDirty(page)) | |
487 | mss->shared_dirty += PAGE_SIZE; | |
488 | else | |
489 | mss->shared_clean += PAGE_SIZE; | |
490 | mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount; | |
491 | } else { | |
492 | if (dirty || PageDirty(page)) | |
493 | mss->private_dirty += PAGE_SIZE; | |
494 | else | |
495 | mss->private_clean += PAGE_SIZE; | |
496 | mss->pss += PAGE_SIZE << PSS_SHIFT; | |
497 | } | |
c164e038 KS |
498 | } |
499 | } | |
ae11c4d9 | 500 | |
c261e7d9 | 501 | #ifdef CONFIG_SHMEM |
c261e7d9 VB |
502 | static int smaps_pte_hole(unsigned long addr, unsigned long end, |
503 | struct mm_walk *walk) | |
504 | { | |
505 | struct mem_size_stats *mss = walk->private; | |
506 | ||
48131e03 VB |
507 | mss->swap += shmem_partial_swap_usage( |
508 | walk->vma->vm_file->f_mapping, addr, end); | |
c261e7d9 VB |
509 | |
510 | return 0; | |
511 | } | |
c261e7d9 VB |
512 | #endif |
513 | ||
c164e038 KS |
514 | static void smaps_pte_entry(pte_t *pte, unsigned long addr, |
515 | struct mm_walk *walk) | |
ae11c4d9 DH |
516 | { |
517 | struct mem_size_stats *mss = walk->private; | |
14eb6fdd | 518 | struct vm_area_struct *vma = walk->vma; |
b1d4d9e0 | 519 | struct page *page = NULL; |
ae11c4d9 | 520 | |
c164e038 KS |
521 | if (pte_present(*pte)) { |
522 | page = vm_normal_page(vma, addr, *pte); | |
523 | } else if (is_swap_pte(*pte)) { | |
524 | swp_entry_t swpent = pte_to_swp_entry(*pte); | |
ae11c4d9 | 525 | |
8334b962 MK |
526 | if (!non_swap_entry(swpent)) { |
527 | int mapcount; | |
528 | ||
c164e038 | 529 | mss->swap += PAGE_SIZE; |
8334b962 MK |
530 | mapcount = swp_swapcount(swpent); |
531 | if (mapcount >= 2) { | |
532 | u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT; | |
533 | ||
534 | do_div(pss_delta, mapcount); | |
535 | mss->swap_pss += pss_delta; | |
536 | } else { | |
537 | mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT; | |
538 | } | |
539 | } else if (is_migration_entry(swpent)) | |
b1d4d9e0 | 540 | page = migration_entry_to_page(swpent); |
c261e7d9 VB |
541 | } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap |
542 | && pte_none(*pte))) { | |
48131e03 VB |
543 | page = find_get_entry(vma->vm_file->f_mapping, |
544 | linear_page_index(vma, addr)); | |
545 | if (!page) | |
546 | return; | |
547 | ||
548 | if (radix_tree_exceptional_entry(page)) | |
549 | mss->swap += PAGE_SIZE; | |
550 | else | |
09cbfeaf | 551 | put_page(page); |
48131e03 VB |
552 | |
553 | return; | |
b1d4d9e0 | 554 | } |
ae11c4d9 | 555 | |
ae11c4d9 DH |
556 | if (!page) |
557 | return; | |
afd9883f KS |
558 | |
559 | smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte)); | |
ae11c4d9 DH |
560 | } |
561 | ||
c164e038 KS |
562 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
563 | static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, | |
564 | struct mm_walk *walk) | |
565 | { | |
566 | struct mem_size_stats *mss = walk->private; | |
14eb6fdd | 567 | struct vm_area_struct *vma = walk->vma; |
c164e038 KS |
568 | struct page *page; |
569 | ||
570 | /* FOLL_DUMP will return -EFAULT on huge zero page */ | |
571 | page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); | |
572 | if (IS_ERR_OR_NULL(page)) | |
573 | return; | |
65c45377 KS |
574 | if (PageAnon(page)) |
575 | mss->anonymous_thp += HPAGE_PMD_SIZE; | |
576 | else if (PageSwapBacked(page)) | |
577 | mss->shmem_thp += HPAGE_PMD_SIZE; | |
ca120cf6 DW |
578 | else if (is_zone_device_page(page)) |
579 | /* pass */; | |
65c45377 KS |
580 | else |
581 | VM_BUG_ON_PAGE(1, page); | |
afd9883f | 582 | smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd)); |
c164e038 KS |
583 | } |
584 | #else | |
585 | static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, | |
586 | struct mm_walk *walk) | |
587 | { | |
588 | } | |
589 | #endif | |
590 | ||
b3ae5acb | 591 | static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, |
2165009b | 592 | struct mm_walk *walk) |
e070ad49 | 593 | { |
14eb6fdd | 594 | struct vm_area_struct *vma = walk->vma; |
ae11c4d9 | 595 | pte_t *pte; |
705e87c0 | 596 | spinlock_t *ptl; |
e070ad49 | 597 | |
b6ec57f4 KS |
598 | ptl = pmd_trans_huge_lock(pmd, vma); |
599 | if (ptl) { | |
c164e038 | 600 | smaps_pmd_entry(pmd, addr, walk); |
bf929152 | 601 | spin_unlock(ptl); |
025c5b24 | 602 | return 0; |
22e057c5 | 603 | } |
1a5a9906 AA |
604 | |
605 | if (pmd_trans_unstable(pmd)) | |
606 | return 0; | |
22e057c5 DH |
607 | /* |
608 | * The mmap_sem held all the way back in m_start() is what | |
609 | * keeps khugepaged out of here and from collapsing things | |
610 | * in here. | |
611 | */ | |
705e87c0 | 612 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
ae11c4d9 | 613 | for (; addr != end; pte++, addr += PAGE_SIZE) |
c164e038 | 614 | smaps_pte_entry(pte, addr, walk); |
705e87c0 HD |
615 | pte_unmap_unlock(pte - 1, ptl); |
616 | cond_resched(); | |
b3ae5acb | 617 | return 0; |
e070ad49 ML |
618 | } |
619 | ||
834f82e2 CG |
620 | static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) |
621 | { | |
622 | /* | |
623 | * Don't forget to update Documentation/ on changes. | |
624 | */ | |
625 | static const char mnemonics[BITS_PER_LONG][2] = { | |
626 | /* | |
627 | * In case if we meet a flag we don't know about. | |
628 | */ | |
629 | [0 ... (BITS_PER_LONG-1)] = "??", | |
630 | ||
631 | [ilog2(VM_READ)] = "rd", | |
632 | [ilog2(VM_WRITE)] = "wr", | |
633 | [ilog2(VM_EXEC)] = "ex", | |
634 | [ilog2(VM_SHARED)] = "sh", | |
635 | [ilog2(VM_MAYREAD)] = "mr", | |
636 | [ilog2(VM_MAYWRITE)] = "mw", | |
637 | [ilog2(VM_MAYEXEC)] = "me", | |
638 | [ilog2(VM_MAYSHARE)] = "ms", | |
639 | [ilog2(VM_GROWSDOWN)] = "gd", | |
640 | [ilog2(VM_PFNMAP)] = "pf", | |
641 | [ilog2(VM_DENYWRITE)] = "dw", | |
4aae7e43 QR |
642 | #ifdef CONFIG_X86_INTEL_MPX |
643 | [ilog2(VM_MPX)] = "mp", | |
644 | #endif | |
834f82e2 CG |
645 | [ilog2(VM_LOCKED)] = "lo", |
646 | [ilog2(VM_IO)] = "io", | |
647 | [ilog2(VM_SEQ_READ)] = "sr", | |
648 | [ilog2(VM_RAND_READ)] = "rr", | |
649 | [ilog2(VM_DONTCOPY)] = "dc", | |
650 | [ilog2(VM_DONTEXPAND)] = "de", | |
651 | [ilog2(VM_ACCOUNT)] = "ac", | |
652 | [ilog2(VM_NORESERVE)] = "nr", | |
653 | [ilog2(VM_HUGETLB)] = "ht", | |
834f82e2 CG |
654 | [ilog2(VM_ARCH_1)] = "ar", |
655 | [ilog2(VM_DONTDUMP)] = "dd", | |
ec8e41ae NH |
656 | #ifdef CONFIG_MEM_SOFT_DIRTY |
657 | [ilog2(VM_SOFTDIRTY)] = "sd", | |
658 | #endif | |
834f82e2 CG |
659 | [ilog2(VM_MIXEDMAP)] = "mm", |
660 | [ilog2(VM_HUGEPAGE)] = "hg", | |
661 | [ilog2(VM_NOHUGEPAGE)] = "nh", | |
662 | [ilog2(VM_MERGEABLE)] = "mg", | |
16ba6f81 AA |
663 | [ilog2(VM_UFFD_MISSING)]= "um", |
664 | [ilog2(VM_UFFD_WP)] = "uw", | |
c1192f84 DH |
665 | #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS |
666 | /* These come out via ProtectionKey: */ | |
667 | [ilog2(VM_PKEY_BIT0)] = "", | |
668 | [ilog2(VM_PKEY_BIT1)] = "", | |
669 | [ilog2(VM_PKEY_BIT2)] = "", | |
670 | [ilog2(VM_PKEY_BIT3)] = "", | |
671 | #endif | |
834f82e2 CG |
672 | }; |
673 | size_t i; | |
674 | ||
675 | seq_puts(m, "VmFlags: "); | |
676 | for (i = 0; i < BITS_PER_LONG; i++) { | |
c1192f84 DH |
677 | if (!mnemonics[i][0]) |
678 | continue; | |
834f82e2 CG |
679 | if (vma->vm_flags & (1UL << i)) { |
680 | seq_printf(m, "%c%c ", | |
681 | mnemonics[i][0], mnemonics[i][1]); | |
682 | } | |
683 | } | |
684 | seq_putc(m, '\n'); | |
685 | } | |
686 | ||
25ee01a2 NH |
687 | #ifdef CONFIG_HUGETLB_PAGE |
688 | static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask, | |
689 | unsigned long addr, unsigned long end, | |
690 | struct mm_walk *walk) | |
691 | { | |
692 | struct mem_size_stats *mss = walk->private; | |
693 | struct vm_area_struct *vma = walk->vma; | |
694 | struct page *page = NULL; | |
695 | ||
696 | if (pte_present(*pte)) { | |
697 | page = vm_normal_page(vma, addr, *pte); | |
698 | } else if (is_swap_pte(*pte)) { | |
699 | swp_entry_t swpent = pte_to_swp_entry(*pte); | |
700 | ||
701 | if (is_migration_entry(swpent)) | |
702 | page = migration_entry_to_page(swpent); | |
703 | } | |
704 | if (page) { | |
705 | int mapcount = page_mapcount(page); | |
706 | ||
707 | if (mapcount >= 2) | |
708 | mss->shared_hugetlb += huge_page_size(hstate_vma(vma)); | |
709 | else | |
710 | mss->private_hugetlb += huge_page_size(hstate_vma(vma)); | |
711 | } | |
712 | return 0; | |
713 | } | |
714 | #endif /* HUGETLB_PAGE */ | |
715 | ||
c1192f84 DH |
716 | void __weak arch_show_smap(struct seq_file *m, struct vm_area_struct *vma) |
717 | { | |
718 | } | |
719 | ||
b7643757 | 720 | static int show_smap(struct seq_file *m, void *v, int is_pid) |
e070ad49 ML |
721 | { |
722 | struct vm_area_struct *vma = v; | |
e070ad49 | 723 | struct mem_size_stats mss; |
2165009b DH |
724 | struct mm_walk smaps_walk = { |
725 | .pmd_entry = smaps_pte_range, | |
25ee01a2 NH |
726 | #ifdef CONFIG_HUGETLB_PAGE |
727 | .hugetlb_entry = smaps_hugetlb_range, | |
728 | #endif | |
2165009b DH |
729 | .mm = vma->vm_mm, |
730 | .private = &mss, | |
731 | }; | |
e070ad49 ML |
732 | |
733 | memset(&mss, 0, sizeof mss); | |
c261e7d9 VB |
734 | |
735 | #ifdef CONFIG_SHMEM | |
736 | if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) { | |
6a15a370 VB |
737 | /* |
738 | * For shared or readonly shmem mappings we know that all | |
739 | * swapped out pages belong to the shmem object, and we can | |
740 | * obtain the swap value much more efficiently. For private | |
741 | * writable mappings, we might have COW pages that are | |
742 | * not affected by the parent swapped out pages of the shmem | |
743 | * object, so we have to distinguish them during the page walk. | |
744 | * Unless we know that the shmem object (or the part mapped by | |
745 | * our VMA) has no swapped out pages at all. | |
746 | */ | |
747 | unsigned long shmem_swapped = shmem_swap_usage(vma); | |
748 | ||
749 | if (!shmem_swapped || (vma->vm_flags & VM_SHARED) || | |
750 | !(vma->vm_flags & VM_WRITE)) { | |
751 | mss.swap = shmem_swapped; | |
752 | } else { | |
753 | mss.check_shmem_swap = true; | |
754 | smaps_walk.pte_hole = smaps_pte_hole; | |
755 | } | |
c261e7d9 VB |
756 | } |
757 | #endif | |
758 | ||
d82ef020 | 759 | /* mmap_sem is held in m_start */ |
14eb6fdd | 760 | walk_page_vma(vma, &smaps_walk); |
4752c369 | 761 | |
b7643757 | 762 | show_map_vma(m, vma, is_pid); |
4752c369 MM |
763 | |
764 | seq_printf(m, | |
765 | "Size: %8lu kB\n" | |
766 | "Rss: %8lu kB\n" | |
767 | "Pss: %8lu kB\n" | |
768 | "Shared_Clean: %8lu kB\n" | |
769 | "Shared_Dirty: %8lu kB\n" | |
770 | "Private_Clean: %8lu kB\n" | |
771 | "Private_Dirty: %8lu kB\n" | |
214e471f | 772 | "Referenced: %8lu kB\n" |
b40d4f84 | 773 | "Anonymous: %8lu kB\n" |
4031a219 | 774 | "AnonHugePages: %8lu kB\n" |
65c45377 | 775 | "ShmemPmdMapped: %8lu kB\n" |
25ee01a2 NH |
776 | "Shared_Hugetlb: %8lu kB\n" |
777 | "Private_Hugetlb: %7lu kB\n" | |
08fba699 | 778 | "Swap: %8lu kB\n" |
8334b962 | 779 | "SwapPss: %8lu kB\n" |
3340289d | 780 | "KernelPageSize: %8lu kB\n" |
2d90508f NK |
781 | "MMUPageSize: %8lu kB\n" |
782 | "Locked: %8lu kB\n", | |
4752c369 MM |
783 | (vma->vm_end - vma->vm_start) >> 10, |
784 | mss.resident >> 10, | |
785 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), | |
786 | mss.shared_clean >> 10, | |
787 | mss.shared_dirty >> 10, | |
788 | mss.private_clean >> 10, | |
789 | mss.private_dirty >> 10, | |
214e471f | 790 | mss.referenced >> 10, |
b40d4f84 | 791 | mss.anonymous >> 10, |
4031a219 | 792 | mss.anonymous_thp >> 10, |
65c45377 | 793 | mss.shmem_thp >> 10, |
25ee01a2 NH |
794 | mss.shared_hugetlb >> 10, |
795 | mss.private_hugetlb >> 10, | |
08fba699 | 796 | mss.swap >> 10, |
8334b962 | 797 | (unsigned long)(mss.swap_pss >> (10 + PSS_SHIFT)), |
3340289d | 798 | vma_kernel_pagesize(vma) >> 10, |
2d90508f NK |
799 | vma_mmu_pagesize(vma) >> 10, |
800 | (vma->vm_flags & VM_LOCKED) ? | |
801 | (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); | |
4752c369 | 802 | |
c1192f84 | 803 | arch_show_smap(m, vma); |
834f82e2 | 804 | show_smap_vma_flags(m, vma); |
b8c20a9b | 805 | m_cache_vma(m, vma); |
7c88db0c | 806 | return 0; |
e070ad49 ML |
807 | } |
808 | ||
b7643757 SP |
809 | static int show_pid_smap(struct seq_file *m, void *v) |
810 | { | |
811 | return show_smap(m, v, 1); | |
812 | } | |
813 | ||
814 | static int show_tid_smap(struct seq_file *m, void *v) | |
815 | { | |
816 | return show_smap(m, v, 0); | |
817 | } | |
818 | ||
03a44825 | 819 | static const struct seq_operations proc_pid_smaps_op = { |
a6198797 MM |
820 | .start = m_start, |
821 | .next = m_next, | |
822 | .stop = m_stop, | |
b7643757 SP |
823 | .show = show_pid_smap |
824 | }; | |
825 | ||
826 | static const struct seq_operations proc_tid_smaps_op = { | |
827 | .start = m_start, | |
828 | .next = m_next, | |
829 | .stop = m_stop, | |
830 | .show = show_tid_smap | |
a6198797 MM |
831 | }; |
832 | ||
b7643757 | 833 | static int pid_smaps_open(struct inode *inode, struct file *file) |
a6198797 MM |
834 | { |
835 | return do_maps_open(inode, file, &proc_pid_smaps_op); | |
836 | } | |
837 | ||
b7643757 SP |
838 | static int tid_smaps_open(struct inode *inode, struct file *file) |
839 | { | |
840 | return do_maps_open(inode, file, &proc_tid_smaps_op); | |
841 | } | |
842 | ||
843 | const struct file_operations proc_pid_smaps_operations = { | |
844 | .open = pid_smaps_open, | |
845 | .read = seq_read, | |
846 | .llseek = seq_lseek, | |
29a40ace | 847 | .release = proc_map_release, |
b7643757 SP |
848 | }; |
849 | ||
850 | const struct file_operations proc_tid_smaps_operations = { | |
851 | .open = tid_smaps_open, | |
a6198797 MM |
852 | .read = seq_read, |
853 | .llseek = seq_lseek, | |
29a40ace | 854 | .release = proc_map_release, |
a6198797 MM |
855 | }; |
856 | ||
040fa020 PE |
857 | enum clear_refs_types { |
858 | CLEAR_REFS_ALL = 1, | |
859 | CLEAR_REFS_ANON, | |
860 | CLEAR_REFS_MAPPED, | |
0f8975ec | 861 | CLEAR_REFS_SOFT_DIRTY, |
695f0559 | 862 | CLEAR_REFS_MM_HIWATER_RSS, |
040fa020 PE |
863 | CLEAR_REFS_LAST, |
864 | }; | |
865 | ||
af9de7eb | 866 | struct clear_refs_private { |
0f8975ec | 867 | enum clear_refs_types type; |
af9de7eb PE |
868 | }; |
869 | ||
7d5b3bfa | 870 | #ifdef CONFIG_MEM_SOFT_DIRTY |
0f8975ec PE |
871 | static inline void clear_soft_dirty(struct vm_area_struct *vma, |
872 | unsigned long addr, pte_t *pte) | |
873 | { | |
0f8975ec PE |
874 | /* |
875 | * The soft-dirty tracker uses #PF-s to catch writes | |
876 | * to pages, so write-protect the pte as well. See the | |
877 | * Documentation/vm/soft-dirty.txt for full description | |
878 | * of how soft-dirty works. | |
879 | */ | |
880 | pte_t ptent = *pte; | |
179ef71c CG |
881 | |
882 | if (pte_present(ptent)) { | |
326c2597 | 883 | ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte); |
179ef71c | 884 | ptent = pte_wrprotect(ptent); |
a7b76174 | 885 | ptent = pte_clear_soft_dirty(ptent); |
326c2597 | 886 | ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent); |
179ef71c CG |
887 | } else if (is_swap_pte(ptent)) { |
888 | ptent = pte_swp_clear_soft_dirty(ptent); | |
326c2597 | 889 | set_pte_at(vma->vm_mm, addr, pte, ptent); |
179ef71c | 890 | } |
0f8975ec | 891 | } |
5d3875a0 LD |
892 | #else |
893 | static inline void clear_soft_dirty(struct vm_area_struct *vma, | |
894 | unsigned long addr, pte_t *pte) | |
895 | { | |
896 | } | |
897 | #endif | |
0f8975ec | 898 | |
5d3875a0 | 899 | #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
7d5b3bfa KS |
900 | static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, |
901 | unsigned long addr, pmd_t *pmdp) | |
902 | { | |
326c2597 | 903 | pmd_t pmd = pmdp_huge_get_and_clear(vma->vm_mm, addr, pmdp); |
7d5b3bfa KS |
904 | |
905 | pmd = pmd_wrprotect(pmd); | |
a7b76174 | 906 | pmd = pmd_clear_soft_dirty(pmd); |
7d5b3bfa | 907 | |
7d5b3bfa KS |
908 | set_pmd_at(vma->vm_mm, addr, pmdp, pmd); |
909 | } | |
7d5b3bfa | 910 | #else |
7d5b3bfa KS |
911 | static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma, |
912 | unsigned long addr, pmd_t *pmdp) | |
913 | { | |
914 | } | |
915 | #endif | |
916 | ||
a6198797 | 917 | static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, |
2165009b | 918 | unsigned long end, struct mm_walk *walk) |
a6198797 | 919 | { |
af9de7eb | 920 | struct clear_refs_private *cp = walk->private; |
5c64f52a | 921 | struct vm_area_struct *vma = walk->vma; |
a6198797 MM |
922 | pte_t *pte, ptent; |
923 | spinlock_t *ptl; | |
924 | struct page *page; | |
925 | ||
b6ec57f4 KS |
926 | ptl = pmd_trans_huge_lock(pmd, vma); |
927 | if (ptl) { | |
7d5b3bfa KS |
928 | if (cp->type == CLEAR_REFS_SOFT_DIRTY) { |
929 | clear_soft_dirty_pmd(vma, addr, pmd); | |
930 | goto out; | |
931 | } | |
932 | ||
933 | page = pmd_page(*pmd); | |
934 | ||
935 | /* Clear accessed and referenced bits. */ | |
936 | pmdp_test_and_clear_young(vma, addr, pmd); | |
33c3fc71 | 937 | test_and_clear_page_young(page); |
7d5b3bfa KS |
938 | ClearPageReferenced(page); |
939 | out: | |
940 | spin_unlock(ptl); | |
941 | return 0; | |
942 | } | |
943 | ||
1a5a9906 AA |
944 | if (pmd_trans_unstable(pmd)) |
945 | return 0; | |
03319327 | 946 | |
a6198797 MM |
947 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
948 | for (; addr != end; pte++, addr += PAGE_SIZE) { | |
949 | ptent = *pte; | |
a6198797 | 950 | |
0f8975ec PE |
951 | if (cp->type == CLEAR_REFS_SOFT_DIRTY) { |
952 | clear_soft_dirty(vma, addr, pte); | |
953 | continue; | |
954 | } | |
955 | ||
179ef71c CG |
956 | if (!pte_present(ptent)) |
957 | continue; | |
958 | ||
a6198797 MM |
959 | page = vm_normal_page(vma, addr, ptent); |
960 | if (!page) | |
961 | continue; | |
962 | ||
963 | /* Clear accessed and referenced bits. */ | |
964 | ptep_test_and_clear_young(vma, addr, pte); | |
33c3fc71 | 965 | test_and_clear_page_young(page); |
a6198797 MM |
966 | ClearPageReferenced(page); |
967 | } | |
968 | pte_unmap_unlock(pte - 1, ptl); | |
969 | cond_resched(); | |
970 | return 0; | |
971 | } | |
972 | ||
5c64f52a NH |
973 | static int clear_refs_test_walk(unsigned long start, unsigned long end, |
974 | struct mm_walk *walk) | |
975 | { | |
976 | struct clear_refs_private *cp = walk->private; | |
977 | struct vm_area_struct *vma = walk->vma; | |
978 | ||
48684a65 NH |
979 | if (vma->vm_flags & VM_PFNMAP) |
980 | return 1; | |
981 | ||
5c64f52a NH |
982 | /* |
983 | * Writing 1 to /proc/pid/clear_refs affects all pages. | |
984 | * Writing 2 to /proc/pid/clear_refs only affects anonymous pages. | |
985 | * Writing 3 to /proc/pid/clear_refs only affects file mapped pages. | |
986 | * Writing 4 to /proc/pid/clear_refs affects all pages. | |
987 | */ | |
988 | if (cp->type == CLEAR_REFS_ANON && vma->vm_file) | |
989 | return 1; | |
990 | if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file) | |
991 | return 1; | |
992 | return 0; | |
993 | } | |
994 | ||
f248dcb3 MM |
995 | static ssize_t clear_refs_write(struct file *file, const char __user *buf, |
996 | size_t count, loff_t *ppos) | |
b813e931 | 997 | { |
f248dcb3 | 998 | struct task_struct *task; |
fb92a4b0 | 999 | char buffer[PROC_NUMBUF]; |
f248dcb3 | 1000 | struct mm_struct *mm; |
b813e931 | 1001 | struct vm_area_struct *vma; |
040fa020 PE |
1002 | enum clear_refs_types type; |
1003 | int itype; | |
0a8cb8e3 | 1004 | int rv; |
b813e931 | 1005 | |
f248dcb3 MM |
1006 | memset(buffer, 0, sizeof(buffer)); |
1007 | if (count > sizeof(buffer) - 1) | |
1008 | count = sizeof(buffer) - 1; | |
1009 | if (copy_from_user(buffer, buf, count)) | |
1010 | return -EFAULT; | |
040fa020 | 1011 | rv = kstrtoint(strstrip(buffer), 10, &itype); |
0a8cb8e3 AD |
1012 | if (rv < 0) |
1013 | return rv; | |
040fa020 PE |
1014 | type = (enum clear_refs_types)itype; |
1015 | if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) | |
f248dcb3 | 1016 | return -EINVAL; |
541c237c | 1017 | |
496ad9aa | 1018 | task = get_proc_task(file_inode(file)); |
f248dcb3 MM |
1019 | if (!task) |
1020 | return -ESRCH; | |
1021 | mm = get_task_mm(task); | |
1022 | if (mm) { | |
af9de7eb | 1023 | struct clear_refs_private cp = { |
0f8975ec | 1024 | .type = type, |
af9de7eb | 1025 | }; |
20cbc972 AM |
1026 | struct mm_walk clear_refs_walk = { |
1027 | .pmd_entry = clear_refs_pte_range, | |
5c64f52a | 1028 | .test_walk = clear_refs_test_walk, |
20cbc972 | 1029 | .mm = mm, |
af9de7eb | 1030 | .private = &cp, |
20cbc972 | 1031 | }; |
695f0559 PC |
1032 | |
1033 | if (type == CLEAR_REFS_MM_HIWATER_RSS) { | |
4e80153a MH |
1034 | if (down_write_killable(&mm->mmap_sem)) { |
1035 | count = -EINTR; | |
1036 | goto out_mm; | |
1037 | } | |
1038 | ||
695f0559 PC |
1039 | /* |
1040 | * Writing 5 to /proc/pid/clear_refs resets the peak | |
1041 | * resident set size to this mm's current rss value. | |
1042 | */ | |
695f0559 PC |
1043 | reset_mm_hiwater_rss(mm); |
1044 | up_write(&mm->mmap_sem); | |
1045 | goto out_mm; | |
1046 | } | |
1047 | ||
f248dcb3 | 1048 | down_read(&mm->mmap_sem); |
64e45507 PF |
1049 | if (type == CLEAR_REFS_SOFT_DIRTY) { |
1050 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
1051 | if (!(vma->vm_flags & VM_SOFTDIRTY)) | |
1052 | continue; | |
1053 | up_read(&mm->mmap_sem); | |
4e80153a MH |
1054 | if (down_write_killable(&mm->mmap_sem)) { |
1055 | count = -EINTR; | |
1056 | goto out_mm; | |
1057 | } | |
64e45507 PF |
1058 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
1059 | vma->vm_flags &= ~VM_SOFTDIRTY; | |
1060 | vma_set_page_prot(vma); | |
1061 | } | |
1062 | downgrade_write(&mm->mmap_sem); | |
1063 | break; | |
1064 | } | |
0f8975ec | 1065 | mmu_notifier_invalidate_range_start(mm, 0, -1); |
64e45507 | 1066 | } |
0f30206b | 1067 | walk_page_range(0, mm->highest_vm_end, &clear_refs_walk); |
0f8975ec PE |
1068 | if (type == CLEAR_REFS_SOFT_DIRTY) |
1069 | mmu_notifier_invalidate_range_end(mm, 0, -1); | |
f248dcb3 MM |
1070 | flush_tlb_mm(mm); |
1071 | up_read(&mm->mmap_sem); | |
695f0559 | 1072 | out_mm: |
f248dcb3 MM |
1073 | mmput(mm); |
1074 | } | |
1075 | put_task_struct(task); | |
fb92a4b0 VL |
1076 | |
1077 | return count; | |
b813e931 DR |
1078 | } |
1079 | ||
f248dcb3 MM |
1080 | const struct file_operations proc_clear_refs_operations = { |
1081 | .write = clear_refs_write, | |
6038f373 | 1082 | .llseek = noop_llseek, |
f248dcb3 MM |
1083 | }; |
1084 | ||
092b50ba NH |
1085 | typedef struct { |
1086 | u64 pme; | |
1087 | } pagemap_entry_t; | |
1088 | ||
85863e47 | 1089 | struct pagemapread { |
8c829622 | 1090 | int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ |
092b50ba | 1091 | pagemap_entry_t *buffer; |
1c90308e | 1092 | bool show_pfn; |
85863e47 MM |
1093 | }; |
1094 | ||
5aaabe83 NH |
1095 | #define PAGEMAP_WALK_SIZE (PMD_SIZE) |
1096 | #define PAGEMAP_WALK_MASK (PMD_MASK) | |
1097 | ||
deb94544 KK |
1098 | #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) |
1099 | #define PM_PFRAME_BITS 55 | |
1100 | #define PM_PFRAME_MASK GENMASK_ULL(PM_PFRAME_BITS - 1, 0) | |
1101 | #define PM_SOFT_DIRTY BIT_ULL(55) | |
77bb499b | 1102 | #define PM_MMAP_EXCLUSIVE BIT_ULL(56) |
deb94544 KK |
1103 | #define PM_FILE BIT_ULL(61) |
1104 | #define PM_SWAP BIT_ULL(62) | |
1105 | #define PM_PRESENT BIT_ULL(63) | |
1106 | ||
85863e47 MM |
1107 | #define PM_END_OF_BUFFER 1 |
1108 | ||
deb94544 | 1109 | static inline pagemap_entry_t make_pme(u64 frame, u64 flags) |
092b50ba | 1110 | { |
deb94544 | 1111 | return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags }; |
092b50ba NH |
1112 | } |
1113 | ||
1114 | static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, | |
85863e47 MM |
1115 | struct pagemapread *pm) |
1116 | { | |
092b50ba | 1117 | pm->buffer[pm->pos++] = *pme; |
d82ef020 | 1118 | if (pm->pos >= pm->len) |
aae8679b | 1119 | return PM_END_OF_BUFFER; |
85863e47 MM |
1120 | return 0; |
1121 | } | |
1122 | ||
1123 | static int pagemap_pte_hole(unsigned long start, unsigned long end, | |
2165009b | 1124 | struct mm_walk *walk) |
85863e47 | 1125 | { |
2165009b | 1126 | struct pagemapread *pm = walk->private; |
68b5a652 | 1127 | unsigned long addr = start; |
85863e47 | 1128 | int err = 0; |
092b50ba | 1129 | |
68b5a652 PF |
1130 | while (addr < end) { |
1131 | struct vm_area_struct *vma = find_vma(walk->mm, addr); | |
deb94544 | 1132 | pagemap_entry_t pme = make_pme(0, 0); |
87e6d49a PF |
1133 | /* End of address space hole, which we mark as non-present. */ |
1134 | unsigned long hole_end; | |
68b5a652 | 1135 | |
87e6d49a PF |
1136 | if (vma) |
1137 | hole_end = min(end, vma->vm_start); | |
1138 | else | |
1139 | hole_end = end; | |
1140 | ||
1141 | for (; addr < hole_end; addr += PAGE_SIZE) { | |
1142 | err = add_to_pagemap(addr, &pme, pm); | |
1143 | if (err) | |
1144 | goto out; | |
68b5a652 PF |
1145 | } |
1146 | ||
87e6d49a PF |
1147 | if (!vma) |
1148 | break; | |
1149 | ||
1150 | /* Addresses in the VMA. */ | |
1151 | if (vma->vm_flags & VM_SOFTDIRTY) | |
deb94544 | 1152 | pme = make_pme(0, PM_SOFT_DIRTY); |
87e6d49a | 1153 | for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { |
68b5a652 PF |
1154 | err = add_to_pagemap(addr, &pme, pm); |
1155 | if (err) | |
1156 | goto out; | |
1157 | } | |
85863e47 | 1158 | } |
68b5a652 | 1159 | out: |
85863e47 MM |
1160 | return err; |
1161 | } | |
1162 | ||
deb94544 | 1163 | static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm, |
052fb0d6 | 1164 | struct vm_area_struct *vma, unsigned long addr, pte_t pte) |
85863e47 | 1165 | { |
deb94544 | 1166 | u64 frame = 0, flags = 0; |
052fb0d6 | 1167 | struct page *page = NULL; |
85863e47 | 1168 | |
052fb0d6 | 1169 | if (pte_present(pte)) { |
1c90308e KK |
1170 | if (pm->show_pfn) |
1171 | frame = pte_pfn(pte); | |
deb94544 | 1172 | flags |= PM_PRESENT; |
052fb0d6 | 1173 | page = vm_normal_page(vma, addr, pte); |
e9cdd6e7 | 1174 | if (pte_soft_dirty(pte)) |
deb94544 | 1175 | flags |= PM_SOFT_DIRTY; |
052fb0d6 | 1176 | } else if (is_swap_pte(pte)) { |
179ef71c CG |
1177 | swp_entry_t entry; |
1178 | if (pte_swp_soft_dirty(pte)) | |
deb94544 | 1179 | flags |= PM_SOFT_DIRTY; |
179ef71c | 1180 | entry = pte_to_swp_entry(pte); |
052fb0d6 KK |
1181 | frame = swp_type(entry) | |
1182 | (swp_offset(entry) << MAX_SWAPFILES_SHIFT); | |
deb94544 | 1183 | flags |= PM_SWAP; |
052fb0d6 KK |
1184 | if (is_migration_entry(entry)) |
1185 | page = migration_entry_to_page(entry); | |
052fb0d6 KK |
1186 | } |
1187 | ||
1188 | if (page && !PageAnon(page)) | |
1189 | flags |= PM_FILE; | |
77bb499b KK |
1190 | if (page && page_mapcount(page) == 1) |
1191 | flags |= PM_MMAP_EXCLUSIVE; | |
deb94544 KK |
1192 | if (vma->vm_flags & VM_SOFTDIRTY) |
1193 | flags |= PM_SOFT_DIRTY; | |
052fb0d6 | 1194 | |
deb94544 | 1195 | return make_pme(frame, flags); |
bcf8039e DH |
1196 | } |
1197 | ||
356515e7 | 1198 | static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end, |
2165009b | 1199 | struct mm_walk *walk) |
85863e47 | 1200 | { |
f995ece2 | 1201 | struct vm_area_struct *vma = walk->vma; |
2165009b | 1202 | struct pagemapread *pm = walk->private; |
bf929152 | 1203 | spinlock_t *ptl; |
05fbf357 | 1204 | pte_t *pte, *orig_pte; |
85863e47 MM |
1205 | int err = 0; |
1206 | ||
356515e7 | 1207 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
b6ec57f4 KS |
1208 | ptl = pmd_trans_huge_lock(pmdp, vma); |
1209 | if (ptl) { | |
356515e7 KK |
1210 | u64 flags = 0, frame = 0; |
1211 | pmd_t pmd = *pmdp; | |
0f8975ec | 1212 | |
356515e7 | 1213 | if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(pmd)) |
deb94544 | 1214 | flags |= PM_SOFT_DIRTY; |
d9104d1c | 1215 | |
356515e7 KK |
1216 | /* |
1217 | * Currently pmd for thp is always present because thp | |
1218 | * can not be swapped-out, migrated, or HWPOISONed | |
1219 | * (split in such cases instead.) | |
1220 | * This if-check is just to prepare for future implementation. | |
1221 | */ | |
1222 | if (pmd_present(pmd)) { | |
77bb499b KK |
1223 | struct page *page = pmd_page(pmd); |
1224 | ||
1225 | if (page_mapcount(page) == 1) | |
1226 | flags |= PM_MMAP_EXCLUSIVE; | |
1227 | ||
356515e7 | 1228 | flags |= PM_PRESENT; |
1c90308e KK |
1229 | if (pm->show_pfn) |
1230 | frame = pmd_pfn(pmd) + | |
1231 | ((addr & ~PMD_MASK) >> PAGE_SHIFT); | |
356515e7 KK |
1232 | } |
1233 | ||
025c5b24 | 1234 | for (; addr != end; addr += PAGE_SIZE) { |
356515e7 | 1235 | pagemap_entry_t pme = make_pme(frame, flags); |
025c5b24 | 1236 | |
092b50ba | 1237 | err = add_to_pagemap(addr, &pme, pm); |
025c5b24 NH |
1238 | if (err) |
1239 | break; | |
1c90308e | 1240 | if (pm->show_pfn && (flags & PM_PRESENT)) |
356515e7 | 1241 | frame++; |
5aaabe83 | 1242 | } |
bf929152 | 1243 | spin_unlock(ptl); |
025c5b24 | 1244 | return err; |
5aaabe83 NH |
1245 | } |
1246 | ||
356515e7 | 1247 | if (pmd_trans_unstable(pmdp)) |
45f83cef | 1248 | return 0; |
356515e7 | 1249 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
81d0fa62 | 1250 | |
f995ece2 NH |
1251 | /* |
1252 | * We can assume that @vma always points to a valid one and @end never | |
1253 | * goes beyond vma->vm_end. | |
1254 | */ | |
356515e7 | 1255 | orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl); |
f995ece2 NH |
1256 | for (; addr < end; pte++, addr += PAGE_SIZE) { |
1257 | pagemap_entry_t pme; | |
05fbf357 | 1258 | |
deb94544 | 1259 | pme = pte_to_pagemap_entry(pm, vma, addr, *pte); |
f995ece2 | 1260 | err = add_to_pagemap(addr, &pme, pm); |
05fbf357 | 1261 | if (err) |
81d0fa62 | 1262 | break; |
85863e47 | 1263 | } |
f995ece2 | 1264 | pte_unmap_unlock(orig_pte, ptl); |
85863e47 MM |
1265 | |
1266 | cond_resched(); | |
1267 | ||
1268 | return err; | |
1269 | } | |
1270 | ||
1a5cb814 | 1271 | #ifdef CONFIG_HUGETLB_PAGE |
116354d1 | 1272 | /* This function walks within one hugetlb entry in the single call */ |
356515e7 | 1273 | static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask, |
116354d1 NH |
1274 | unsigned long addr, unsigned long end, |
1275 | struct mm_walk *walk) | |
5dc37642 | 1276 | { |
5dc37642 | 1277 | struct pagemapread *pm = walk->private; |
f995ece2 | 1278 | struct vm_area_struct *vma = walk->vma; |
356515e7 | 1279 | u64 flags = 0, frame = 0; |
5dc37642 | 1280 | int err = 0; |
356515e7 | 1281 | pte_t pte; |
5dc37642 | 1282 | |
f995ece2 | 1283 | if (vma->vm_flags & VM_SOFTDIRTY) |
deb94544 | 1284 | flags |= PM_SOFT_DIRTY; |
d9104d1c | 1285 | |
356515e7 KK |
1286 | pte = huge_ptep_get(ptep); |
1287 | if (pte_present(pte)) { | |
1288 | struct page *page = pte_page(pte); | |
1289 | ||
1290 | if (!PageAnon(page)) | |
1291 | flags |= PM_FILE; | |
1292 | ||
77bb499b KK |
1293 | if (page_mapcount(page) == 1) |
1294 | flags |= PM_MMAP_EXCLUSIVE; | |
1295 | ||
356515e7 | 1296 | flags |= PM_PRESENT; |
1c90308e KK |
1297 | if (pm->show_pfn) |
1298 | frame = pte_pfn(pte) + | |
1299 | ((addr & ~hmask) >> PAGE_SHIFT); | |
356515e7 KK |
1300 | } |
1301 | ||
5dc37642 | 1302 | for (; addr != end; addr += PAGE_SIZE) { |
356515e7 KK |
1303 | pagemap_entry_t pme = make_pme(frame, flags); |
1304 | ||
092b50ba | 1305 | err = add_to_pagemap(addr, &pme, pm); |
5dc37642 NH |
1306 | if (err) |
1307 | return err; | |
1c90308e | 1308 | if (pm->show_pfn && (flags & PM_PRESENT)) |
356515e7 | 1309 | frame++; |
5dc37642 NH |
1310 | } |
1311 | ||
1312 | cond_resched(); | |
1313 | ||
1314 | return err; | |
1315 | } | |
1a5cb814 | 1316 | #endif /* HUGETLB_PAGE */ |
5dc37642 | 1317 | |
85863e47 MM |
1318 | /* |
1319 | * /proc/pid/pagemap - an array mapping virtual pages to pfns | |
1320 | * | |
f16278c6 HR |
1321 | * For each page in the address space, this file contains one 64-bit entry |
1322 | * consisting of the following: | |
1323 | * | |
052fb0d6 | 1324 | * Bits 0-54 page frame number (PFN) if present |
f16278c6 | 1325 | * Bits 0-4 swap type if swapped |
052fb0d6 | 1326 | * Bits 5-54 swap offset if swapped |
deb94544 | 1327 | * Bit 55 pte is soft-dirty (see Documentation/vm/soft-dirty.txt) |
77bb499b KK |
1328 | * Bit 56 page exclusively mapped |
1329 | * Bits 57-60 zero | |
052fb0d6 | 1330 | * Bit 61 page is file-page or shared-anon |
f16278c6 HR |
1331 | * Bit 62 page swapped |
1332 | * Bit 63 page present | |
1333 | * | |
1334 | * If the page is not present but in swap, then the PFN contains an | |
1335 | * encoding of the swap file number and the page's offset into the | |
1336 | * swap. Unmapped pages return a null PFN. This allows determining | |
85863e47 MM |
1337 | * precisely which pages are mapped (or in swap) and comparing mapped |
1338 | * pages between processes. | |
1339 | * | |
1340 | * Efficient users of this interface will use /proc/pid/maps to | |
1341 | * determine which areas of memory are actually mapped and llseek to | |
1342 | * skip over unmapped regions. | |
1343 | */ | |
1344 | static ssize_t pagemap_read(struct file *file, char __user *buf, | |
1345 | size_t count, loff_t *ppos) | |
1346 | { | |
a06db751 | 1347 | struct mm_struct *mm = file->private_data; |
85863e47 | 1348 | struct pagemapread pm; |
ee1e6ab6 | 1349 | struct mm_walk pagemap_walk = {}; |
5d7e0d2b AM |
1350 | unsigned long src; |
1351 | unsigned long svpfn; | |
1352 | unsigned long start_vaddr; | |
1353 | unsigned long end_vaddr; | |
a06db751 | 1354 | int ret = 0, copied = 0; |
85863e47 | 1355 | |
388f7934 | 1356 | if (!mm || !mmget_not_zero(mm)) |
85863e47 MM |
1357 | goto out; |
1358 | ||
85863e47 MM |
1359 | ret = -EINVAL; |
1360 | /* file position must be aligned */ | |
aae8679b | 1361 | if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) |
a06db751 | 1362 | goto out_mm; |
85863e47 MM |
1363 | |
1364 | ret = 0; | |
08161786 | 1365 | if (!count) |
a06db751 | 1366 | goto out_mm; |
08161786 | 1367 | |
1c90308e KK |
1368 | /* do not disclose physical addresses: attack vector */ |
1369 | pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN); | |
1370 | ||
8c829622 | 1371 | pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); |
1372 | pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY); | |
5d7e0d2b | 1373 | ret = -ENOMEM; |
d82ef020 | 1374 | if (!pm.buffer) |
a06db751 | 1375 | goto out_mm; |
85863e47 | 1376 | |
356515e7 | 1377 | pagemap_walk.pmd_entry = pagemap_pmd_range; |
5d7e0d2b | 1378 | pagemap_walk.pte_hole = pagemap_pte_hole; |
1a5cb814 | 1379 | #ifdef CONFIG_HUGETLB_PAGE |
5dc37642 | 1380 | pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; |
1a5cb814 | 1381 | #endif |
5d7e0d2b AM |
1382 | pagemap_walk.mm = mm; |
1383 | pagemap_walk.private = ± | |
1384 | ||
1385 | src = *ppos; | |
1386 | svpfn = src / PM_ENTRY_BYTES; | |
1387 | start_vaddr = svpfn << PAGE_SHIFT; | |
a06db751 | 1388 | end_vaddr = mm->task_size; |
5d7e0d2b AM |
1389 | |
1390 | /* watch out for wraparound */ | |
a06db751 | 1391 | if (svpfn > mm->task_size >> PAGE_SHIFT) |
5d7e0d2b AM |
1392 | start_vaddr = end_vaddr; |
1393 | ||
1394 | /* | |
1395 | * The odds are that this will stop walking way | |
1396 | * before end_vaddr, because the length of the | |
1397 | * user buffer is tracked in "pm", and the walk | |
1398 | * will stop when we hit the end of the buffer. | |
1399 | */ | |
d82ef020 KH |
1400 | ret = 0; |
1401 | while (count && (start_vaddr < end_vaddr)) { | |
1402 | int len; | |
1403 | unsigned long end; | |
1404 | ||
1405 | pm.pos = 0; | |
ea251c1d | 1406 | end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; |
d82ef020 KH |
1407 | /* overflow ? */ |
1408 | if (end < start_vaddr || end > end_vaddr) | |
1409 | end = end_vaddr; | |
1410 | down_read(&mm->mmap_sem); | |
1411 | ret = walk_page_range(start_vaddr, end, &pagemap_walk); | |
1412 | up_read(&mm->mmap_sem); | |
1413 | start_vaddr = end; | |
1414 | ||
1415 | len = min(count, PM_ENTRY_BYTES * pm.pos); | |
309361e0 | 1416 | if (copy_to_user(buf, pm.buffer, len)) { |
d82ef020 | 1417 | ret = -EFAULT; |
a06db751 | 1418 | goto out_free; |
d82ef020 KH |
1419 | } |
1420 | copied += len; | |
1421 | buf += len; | |
1422 | count -= len; | |
85863e47 | 1423 | } |
d82ef020 KH |
1424 | *ppos += copied; |
1425 | if (!ret || ret == PM_END_OF_BUFFER) | |
1426 | ret = copied; | |
1427 | ||
98bc93e5 KM |
1428 | out_free: |
1429 | kfree(pm.buffer); | |
a06db751 KK |
1430 | out_mm: |
1431 | mmput(mm); | |
85863e47 MM |
1432 | out: |
1433 | return ret; | |
1434 | } | |
1435 | ||
541c237c PE |
1436 | static int pagemap_open(struct inode *inode, struct file *file) |
1437 | { | |
a06db751 KK |
1438 | struct mm_struct *mm; |
1439 | ||
a06db751 KK |
1440 | mm = proc_mem_open(inode, PTRACE_MODE_READ); |
1441 | if (IS_ERR(mm)) | |
1442 | return PTR_ERR(mm); | |
1443 | file->private_data = mm; | |
1444 | return 0; | |
1445 | } | |
1446 | ||
1447 | static int pagemap_release(struct inode *inode, struct file *file) | |
1448 | { | |
1449 | struct mm_struct *mm = file->private_data; | |
1450 | ||
1451 | if (mm) | |
1452 | mmdrop(mm); | |
541c237c PE |
1453 | return 0; |
1454 | } | |
1455 | ||
85863e47 MM |
1456 | const struct file_operations proc_pagemap_operations = { |
1457 | .llseek = mem_lseek, /* borrow this */ | |
1458 | .read = pagemap_read, | |
541c237c | 1459 | .open = pagemap_open, |
a06db751 | 1460 | .release = pagemap_release, |
85863e47 | 1461 | }; |
1e883281 | 1462 | #endif /* CONFIG_PROC_PAGE_MONITOR */ |
85863e47 | 1463 | |
6e21c8f1 | 1464 | #ifdef CONFIG_NUMA |
6e21c8f1 | 1465 | |
f69ff943 | 1466 | struct numa_maps { |
f69ff943 SW |
1467 | unsigned long pages; |
1468 | unsigned long anon; | |
1469 | unsigned long active; | |
1470 | unsigned long writeback; | |
1471 | unsigned long mapcount_max; | |
1472 | unsigned long dirty; | |
1473 | unsigned long swapcache; | |
1474 | unsigned long node[MAX_NUMNODES]; | |
1475 | }; | |
1476 | ||
5b52fc89 SW |
1477 | struct numa_maps_private { |
1478 | struct proc_maps_private proc_maps; | |
1479 | struct numa_maps md; | |
1480 | }; | |
1481 | ||
eb4866d0 DH |
1482 | static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, |
1483 | unsigned long nr_pages) | |
f69ff943 SW |
1484 | { |
1485 | int count = page_mapcount(page); | |
1486 | ||
eb4866d0 | 1487 | md->pages += nr_pages; |
f69ff943 | 1488 | if (pte_dirty || PageDirty(page)) |
eb4866d0 | 1489 | md->dirty += nr_pages; |
f69ff943 SW |
1490 | |
1491 | if (PageSwapCache(page)) | |
eb4866d0 | 1492 | md->swapcache += nr_pages; |
f69ff943 SW |
1493 | |
1494 | if (PageActive(page) || PageUnevictable(page)) | |
eb4866d0 | 1495 | md->active += nr_pages; |
f69ff943 SW |
1496 | |
1497 | if (PageWriteback(page)) | |
eb4866d0 | 1498 | md->writeback += nr_pages; |
f69ff943 SW |
1499 | |
1500 | if (PageAnon(page)) | |
eb4866d0 | 1501 | md->anon += nr_pages; |
f69ff943 SW |
1502 | |
1503 | if (count > md->mapcount_max) | |
1504 | md->mapcount_max = count; | |
1505 | ||
eb4866d0 | 1506 | md->node[page_to_nid(page)] += nr_pages; |
f69ff943 SW |
1507 | } |
1508 | ||
3200a8aa DH |
1509 | static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, |
1510 | unsigned long addr) | |
1511 | { | |
1512 | struct page *page; | |
1513 | int nid; | |
1514 | ||
1515 | if (!pte_present(pte)) | |
1516 | return NULL; | |
1517 | ||
1518 | page = vm_normal_page(vma, addr, pte); | |
1519 | if (!page) | |
1520 | return NULL; | |
1521 | ||
1522 | if (PageReserved(page)) | |
1523 | return NULL; | |
1524 | ||
1525 | nid = page_to_nid(page); | |
4ff1b2c2 | 1526 | if (!node_isset(nid, node_states[N_MEMORY])) |
3200a8aa DH |
1527 | return NULL; |
1528 | ||
1529 | return page; | |
1530 | } | |
1531 | ||
28093f9f GS |
1532 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
1533 | static struct page *can_gather_numa_stats_pmd(pmd_t pmd, | |
1534 | struct vm_area_struct *vma, | |
1535 | unsigned long addr) | |
1536 | { | |
1537 | struct page *page; | |
1538 | int nid; | |
1539 | ||
1540 | if (!pmd_present(pmd)) | |
1541 | return NULL; | |
1542 | ||
1543 | page = vm_normal_page_pmd(vma, addr, pmd); | |
1544 | if (!page) | |
1545 | return NULL; | |
1546 | ||
1547 | if (PageReserved(page)) | |
1548 | return NULL; | |
1549 | ||
1550 | nid = page_to_nid(page); | |
1551 | if (!node_isset(nid, node_states[N_MEMORY])) | |
1552 | return NULL; | |
1553 | ||
1554 | return page; | |
1555 | } | |
1556 | #endif | |
1557 | ||
f69ff943 SW |
1558 | static int gather_pte_stats(pmd_t *pmd, unsigned long addr, |
1559 | unsigned long end, struct mm_walk *walk) | |
1560 | { | |
d85f4d6d NH |
1561 | struct numa_maps *md = walk->private; |
1562 | struct vm_area_struct *vma = walk->vma; | |
f69ff943 SW |
1563 | spinlock_t *ptl; |
1564 | pte_t *orig_pte; | |
1565 | pte_t *pte; | |
1566 | ||
28093f9f | 1567 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
b6ec57f4 KS |
1568 | ptl = pmd_trans_huge_lock(pmd, vma); |
1569 | if (ptl) { | |
025c5b24 NH |
1570 | struct page *page; |
1571 | ||
28093f9f | 1572 | page = can_gather_numa_stats_pmd(*pmd, vma, addr); |
025c5b24 | 1573 | if (page) |
28093f9f | 1574 | gather_stats(page, md, pmd_dirty(*pmd), |
025c5b24 | 1575 | HPAGE_PMD_SIZE/PAGE_SIZE); |
bf929152 | 1576 | spin_unlock(ptl); |
025c5b24 | 1577 | return 0; |
32ef4384 DH |
1578 | } |
1579 | ||
1a5a9906 AA |
1580 | if (pmd_trans_unstable(pmd)) |
1581 | return 0; | |
28093f9f | 1582 | #endif |
f69ff943 SW |
1583 | orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); |
1584 | do { | |
d85f4d6d | 1585 | struct page *page = can_gather_numa_stats(*pte, vma, addr); |
f69ff943 SW |
1586 | if (!page) |
1587 | continue; | |
eb4866d0 | 1588 | gather_stats(page, md, pte_dirty(*pte), 1); |
f69ff943 SW |
1589 | |
1590 | } while (pte++, addr += PAGE_SIZE, addr != end); | |
1591 | pte_unmap_unlock(orig_pte, ptl); | |
a66c0410 | 1592 | cond_resched(); |
f69ff943 SW |
1593 | return 0; |
1594 | } | |
1595 | #ifdef CONFIG_HUGETLB_PAGE | |
632fd60f | 1596 | static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, |
f69ff943 SW |
1597 | unsigned long addr, unsigned long end, struct mm_walk *walk) |
1598 | { | |
5c2ff95e | 1599 | pte_t huge_pte = huge_ptep_get(pte); |
f69ff943 SW |
1600 | struct numa_maps *md; |
1601 | struct page *page; | |
1602 | ||
5c2ff95e | 1603 | if (!pte_present(huge_pte)) |
f69ff943 SW |
1604 | return 0; |
1605 | ||
5c2ff95e | 1606 | page = pte_page(huge_pte); |
f69ff943 SW |
1607 | if (!page) |
1608 | return 0; | |
1609 | ||
1610 | md = walk->private; | |
5c2ff95e | 1611 | gather_stats(page, md, pte_dirty(huge_pte), 1); |
f69ff943 SW |
1612 | return 0; |
1613 | } | |
1614 | ||
1615 | #else | |
632fd60f | 1616 | static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask, |
f69ff943 SW |
1617 | unsigned long addr, unsigned long end, struct mm_walk *walk) |
1618 | { | |
1619 | return 0; | |
1620 | } | |
1621 | #endif | |
1622 | ||
1623 | /* | |
1624 | * Display pages allocated per node and memory policy via /proc. | |
1625 | */ | |
b7643757 | 1626 | static int show_numa_map(struct seq_file *m, void *v, int is_pid) |
f69ff943 | 1627 | { |
5b52fc89 SW |
1628 | struct numa_maps_private *numa_priv = m->private; |
1629 | struct proc_maps_private *proc_priv = &numa_priv->proc_maps; | |
f69ff943 | 1630 | struct vm_area_struct *vma = v; |
5b52fc89 | 1631 | struct numa_maps *md = &numa_priv->md; |
f69ff943 SW |
1632 | struct file *file = vma->vm_file; |
1633 | struct mm_struct *mm = vma->vm_mm; | |
d85f4d6d NH |
1634 | struct mm_walk walk = { |
1635 | .hugetlb_entry = gather_hugetlb_stats, | |
1636 | .pmd_entry = gather_pte_stats, | |
1637 | .private = md, | |
1638 | .mm = mm, | |
1639 | }; | |
f69ff943 | 1640 | struct mempolicy *pol; |
948927ee DR |
1641 | char buffer[64]; |
1642 | int nid; | |
f69ff943 SW |
1643 | |
1644 | if (!mm) | |
1645 | return 0; | |
1646 | ||
5b52fc89 SW |
1647 | /* Ensure we start with an empty set of numa_maps statistics. */ |
1648 | memset(md, 0, sizeof(*md)); | |
f69ff943 | 1649 | |
498f2371 ON |
1650 | pol = __get_vma_policy(vma, vma->vm_start); |
1651 | if (pol) { | |
1652 | mpol_to_str(buffer, sizeof(buffer), pol); | |
1653 | mpol_cond_put(pol); | |
1654 | } else { | |
1655 | mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); | |
1656 | } | |
f69ff943 SW |
1657 | |
1658 | seq_printf(m, "%08lx %s", vma->vm_start, buffer); | |
1659 | ||
1660 | if (file) { | |
17c2b4ee | 1661 | seq_puts(m, " file="); |
2726d566 | 1662 | seq_file_path(m, file, "\n\t= "); |
f69ff943 | 1663 | } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { |
17c2b4ee | 1664 | seq_puts(m, " heap"); |
b18cb64e | 1665 | } else if (is_stack(proc_priv, vma)) { |
65376df5 | 1666 | seq_puts(m, " stack"); |
f69ff943 SW |
1667 | } |
1668 | ||
fc360bd9 | 1669 | if (is_vm_hugetlb_page(vma)) |
17c2b4ee | 1670 | seq_puts(m, " huge"); |
fc360bd9 | 1671 | |
d85f4d6d NH |
1672 | /* mmap_sem is held by m_start */ |
1673 | walk_page_vma(vma, &walk); | |
f69ff943 SW |
1674 | |
1675 | if (!md->pages) | |
1676 | goto out; | |
1677 | ||
1678 | if (md->anon) | |
1679 | seq_printf(m, " anon=%lu", md->anon); | |
1680 | ||
1681 | if (md->dirty) | |
1682 | seq_printf(m, " dirty=%lu", md->dirty); | |
1683 | ||
1684 | if (md->pages != md->anon && md->pages != md->dirty) | |
1685 | seq_printf(m, " mapped=%lu", md->pages); | |
1686 | ||
1687 | if (md->mapcount_max > 1) | |
1688 | seq_printf(m, " mapmax=%lu", md->mapcount_max); | |
1689 | ||
1690 | if (md->swapcache) | |
1691 | seq_printf(m, " swapcache=%lu", md->swapcache); | |
1692 | ||
1693 | if (md->active < md->pages && !is_vm_hugetlb_page(vma)) | |
1694 | seq_printf(m, " active=%lu", md->active); | |
1695 | ||
1696 | if (md->writeback) | |
1697 | seq_printf(m, " writeback=%lu", md->writeback); | |
1698 | ||
948927ee DR |
1699 | for_each_node_state(nid, N_MEMORY) |
1700 | if (md->node[nid]) | |
1701 | seq_printf(m, " N%d=%lu", nid, md->node[nid]); | |
198d1597 RA |
1702 | |
1703 | seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10); | |
f69ff943 SW |
1704 | out: |
1705 | seq_putc(m, '\n'); | |
b8c20a9b | 1706 | m_cache_vma(m, vma); |
f69ff943 SW |
1707 | return 0; |
1708 | } | |
5b52fc89 | 1709 | |
b7643757 SP |
1710 | static int show_pid_numa_map(struct seq_file *m, void *v) |
1711 | { | |
1712 | return show_numa_map(m, v, 1); | |
1713 | } | |
1714 | ||
1715 | static int show_tid_numa_map(struct seq_file *m, void *v) | |
1716 | { | |
1717 | return show_numa_map(m, v, 0); | |
1718 | } | |
1719 | ||
03a44825 | 1720 | static const struct seq_operations proc_pid_numa_maps_op = { |
b7643757 SP |
1721 | .start = m_start, |
1722 | .next = m_next, | |
1723 | .stop = m_stop, | |
1724 | .show = show_pid_numa_map, | |
6e21c8f1 | 1725 | }; |
662795de | 1726 | |
b7643757 SP |
1727 | static const struct seq_operations proc_tid_numa_maps_op = { |
1728 | .start = m_start, | |
1729 | .next = m_next, | |
1730 | .stop = m_stop, | |
1731 | .show = show_tid_numa_map, | |
1732 | }; | |
1733 | ||
1734 | static int numa_maps_open(struct inode *inode, struct file *file, | |
1735 | const struct seq_operations *ops) | |
662795de | 1736 | { |
4db7d0ee ON |
1737 | return proc_maps_open(inode, file, ops, |
1738 | sizeof(struct numa_maps_private)); | |
662795de EB |
1739 | } |
1740 | ||
b7643757 SP |
1741 | static int pid_numa_maps_open(struct inode *inode, struct file *file) |
1742 | { | |
1743 | return numa_maps_open(inode, file, &proc_pid_numa_maps_op); | |
1744 | } | |
1745 | ||
1746 | static int tid_numa_maps_open(struct inode *inode, struct file *file) | |
1747 | { | |
1748 | return numa_maps_open(inode, file, &proc_tid_numa_maps_op); | |
1749 | } | |
1750 | ||
1751 | const struct file_operations proc_pid_numa_maps_operations = { | |
1752 | .open = pid_numa_maps_open, | |
1753 | .read = seq_read, | |
1754 | .llseek = seq_lseek, | |
29a40ace | 1755 | .release = proc_map_release, |
b7643757 SP |
1756 | }; |
1757 | ||
1758 | const struct file_operations proc_tid_numa_maps_operations = { | |
1759 | .open = tid_numa_maps_open, | |
662795de EB |
1760 | .read = seq_read, |
1761 | .llseek = seq_lseek, | |
29a40ace | 1762 | .release = proc_map_release, |
662795de | 1763 | }; |
f69ff943 | 1764 | #endif /* CONFIG_NUMA */ |