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