fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/vmacache.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/highmem.h>
   9 #include <linux/ptrace.h>
  10 #include <linux/slab.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/mempolicy.h>
  13 #include <linux/rmap.h>
  14 #include <linux/swap.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/swapops.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/page_idle.h>
  19 #include <linux/shmem_fs.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/pkeys.h>
  22
  23 #include <asm/elf.h>
  24 #include <asm/tlb.h>
  25 #include <asm/tlbflush.h>
  26 #include "internal.h"
  27
  28 #define SEQ_PUT_DEC(str, val) \
  29                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  30 void task_mem(struct seq_file *m, struct mm_struct *mm)
  31 {
  32         unsigned long text, lib, swap, anon, file, shmem;
  33         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  34
  35         anon = get_mm_counter(mm, MM_ANONPAGES);
  36         file = get_mm_counter(mm, MM_FILEPAGES);
  37         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  38
  39         /*
  40          * Note: to minimize their overhead, mm maintains hiwater_vm and
  41          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  42          * collector of these hiwater stats must therefore get total_vm
  43          * and rss too, which will usually be the higher.  Barriers? not
  44          * worth the effort, such snapshots can always be inconsistent.
  45          */
  46         hiwater_vm = total_vm = mm->total_vm;
  47         if (hiwater_vm < mm->hiwater_vm)
  48                 hiwater_vm = mm->hiwater_vm;
  49         hiwater_rss = total_rss = anon + file + shmem;
  50         if (hiwater_rss < mm->hiwater_rss)
  51                 hiwater_rss = mm->hiwater_rss;
  52
  53         /* split executable areas between text and lib */
  54         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  55         text = min(text, mm->exec_vm << PAGE_SHIFT);
  56         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  57
  58         swap = get_mm_counter(mm, MM_SWAPENTS);
  59         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  60         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  61         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  62         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  63         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  64         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  65         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  66         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  67         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  68         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  69         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  70         seq_put_decimal_ull_width(m,
  71                     " kB\nVmExe:\t", text >> 10, 8);
  72         seq_put_decimal_ull_width(m,
  73                     " kB\nVmLib:\t", lib >> 10, 8);
  74         seq_put_decimal_ull_width(m,
  75                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  76         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  77         seq_puts(m, " kB\n");
  78         hugetlb_report_usage(m, mm);
  79 }
  80 #undef SEQ_PUT_DEC
  81
  82 unsigned long task_vsize(struct mm_struct *mm)
  83 {
  84         return PAGE_SIZE * mm->total_vm;
  85 }
  86
  87 unsigned long task_statm(struct mm_struct *mm,
  88                          unsigned long *shared, unsigned long *text,
  89                          unsigned long *data, unsigned long *resident)
  90 {
  91         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  92                         get_mm_counter(mm, MM_SHMEMPAGES);
  93         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  94                                                                 >> PAGE_SHIFT;
  95         *data = mm->data_vm + mm->stack_vm;
  96         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  97         return mm->total_vm;
  98 }
  99
 100 #ifdef CONFIG_NUMA
 101 /*
 102  * Save get_task_policy() for show_numa_map().
 103  */
 104 static void hold_task_mempolicy(struct proc_maps_private *priv)
 105 {
 106         struct task_struct *task = priv->task;
 107
 108         task_lock(task);
 109         priv->task_mempolicy = get_task_policy(task);
 110         mpol_get(priv->task_mempolicy);
 111         task_unlock(task);
 112 }
 113 static void release_task_mempolicy(struct proc_maps_private *priv)
 114 {
 115         mpol_put(priv->task_mempolicy);
 116 }
 117 #else
 118 static void hold_task_mempolicy(struct proc_maps_private *priv)
 119 {
 120 }
 121 static void release_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 #endif
 125
 126 static void vma_stop(struct proc_maps_private *priv)
 127 {
 128         struct mm_struct *mm = priv->mm;
 129
 130         release_task_mempolicy(priv);
 131         up_read(&mm->mmap_sem);
 132         mmput(mm);
 133 }
 134
 135 static struct vm_area_struct *
 136 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 137 {
 138         if (vma == priv->tail_vma)
 139                 return NULL;
 140         return vma->vm_next ?: priv->tail_vma;
 141 }
 142
 143 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 144 {
 145         if (m->count < m->size) /* vma is copied successfully */
 146                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 147 }
 148
 149 static void *m_start(struct seq_file *m, loff_t *ppos)
 150 {
 151         struct proc_maps_private *priv = m->private;
 152         unsigned long last_addr = m->version;
 153         struct mm_struct *mm;
 154         struct vm_area_struct *vma;
 155         unsigned int pos = *ppos;
 156
 157         /* See m_cache_vma(). Zero at the start or after lseek. */
 158         if (last_addr == -1UL)
 159                 return NULL;
 160
 161         priv->task = get_proc_task(priv->inode);
 162         if (!priv->task)
 163                 return ERR_PTR(-ESRCH);
 164
 165         mm = priv->mm;
 166         if (!mm || !mmget_not_zero(mm))
 167                 return NULL;
 168
 169         if (down_read_killable(&mm->mmap_sem)) {
 170                 mmput(mm);
 171                 return ERR_PTR(-EINTR);
 172         }
 173
 174         hold_task_mempolicy(priv);
 175         priv->tail_vma = get_gate_vma(mm);
 176
 177         if (last_addr) {
 178                 vma = find_vma(mm, last_addr - 1);
 179                 if (vma && vma->vm_start <= last_addr)
 180                         vma = m_next_vma(priv, vma);
 181                 if (vma)
 182                         return vma;
 183         }
 184
 185         m->version = 0;
 186         if (pos < mm->map_count) {
 187                 for (vma = mm->mmap; pos; pos--) {
 188                         m->version = vma->vm_start;
 189                         vma = vma->vm_next;
 190                 }
 191                 return vma;
 192         }
 193
 194         /* we do not bother to update m->version in this case */
 195         if (pos == mm->map_count && priv->tail_vma)
 196                 return priv->tail_vma;
 197
 198         vma_stop(priv);
 199         return NULL;
 200 }
 201
 202 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 203 {
 204         struct proc_maps_private *priv = m->private;
 205         struct vm_area_struct *next;
 206
 207         (*pos)++;
 208         next = m_next_vma(priv, v);
 209         if (!next)
 210                 vma_stop(priv);
 211         return next;
 212 }
 213
 214 static void m_stop(struct seq_file *m, void *v)
 215 {
 216         struct proc_maps_private *priv = m->private;
 217
 218         if (!IS_ERR_OR_NULL(v))
 219                 vma_stop(priv);
 220         if (priv->task) {
 221                 put_task_struct(priv->task);
 222                 priv->task = NULL;
 223         }
 224 }
 225
 226 static int proc_maps_open(struct inode *inode, struct file *file,
 227                         const struct seq_operations *ops, int psize)
 228 {
 229         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 230
 231         if (!priv)
 232                 return -ENOMEM;
 233
 234         priv->inode = inode;
 235         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 236         if (IS_ERR(priv->mm)) {
 237                 int err = PTR_ERR(priv->mm);
 238
 239                 seq_release_private(inode, file);
 240                 return err;
 241         }
 242
 243         return 0;
 244 }
 245
 246 static int proc_map_release(struct inode *inode, struct file *file)
 247 {
 248         struct seq_file *seq = file->private_data;
 249         struct proc_maps_private *priv = seq->private;
 250
 251         if (priv->mm)
 252                 mmdrop(priv->mm);
 253
 254         return seq_release_private(inode, file);
 255 }
 256
 257 static int do_maps_open(struct inode *inode, struct file *file,
 258                         const struct seq_operations *ops)
 259 {
 260         return proc_maps_open(inode, file, ops,
 261                                 sizeof(struct proc_maps_private));
 262 }
 263
 264 /*
 265  * Indicate if the VMA is a stack for the given task; for
 266  * /proc/PID/maps that is the stack of the main task.
 267  */
 268 static int is_stack(struct vm_area_struct *vma)
 269 {
 270         /*
 271          * We make no effort to guess what a given thread considers to be
 272          * its "stack".  It's not even well-defined for programs written
 273          * languages like Go.
 274          */
 275         return vma->vm_start <= vma->vm_mm->start_stack &&
 276                 vma->vm_end >= vma->vm_mm->start_stack;
 277 }
 278
 279 static void show_vma_header_prefix(struct seq_file *m,
 280                                    unsigned long start, unsigned long end,
 281                                    vm_flags_t flags, unsigned long long pgoff,
 282                                    dev_t dev, unsigned long ino)
 283 {
 284         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 285         seq_put_hex_ll(m, NULL, start, 8);
 286         seq_put_hex_ll(m, "-", end, 8);
 287         seq_putc(m, ' ');
 288         seq_putc(m, flags & VM_READ ? 'r' : '-');
 289         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 290         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 291         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 292         seq_put_hex_ll(m, " ", pgoff, 8);
 293         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 294         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 295         seq_put_decimal_ull(m, " ", ino);
 296         seq_putc(m, ' ');
 297 }
 298
 299 static void
 300 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 301 {
 302         struct mm_struct *mm = vma->vm_mm;
 303         struct file *file = vma->vm_file;
 304         vm_flags_t flags = vma->vm_flags;
 305         unsigned long ino = 0;
 306         unsigned long long pgoff = 0;
 307         unsigned long start, end;
 308         dev_t dev = 0;
 309         const char *name = NULL;
 310
 311         if (file) {
 312                 struct inode *inode = file_inode(vma->vm_file);
 313                 dev = inode->i_sb->s_dev;
 314                 ino = inode->i_ino;
 315                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 316         }
 317
 318         start = vma->vm_start;
 319         end = vma->vm_end;
 320         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 321
 322         /*
 323          * Print the dentry name for named mappings, and a
 324          * special [heap] marker for the heap:
 325          */
 326         if (file) {
 327                 seq_pad(m, ' ');
 328                 seq_file_path(m, file, "\n");
 329                 goto done;
 330         }
 331
 332         if (vma->vm_ops && vma->vm_ops->name) {
 333                 name = vma->vm_ops->name(vma);
 334                 if (name)
 335                         goto done;
 336         }
 337
 338         name = arch_vma_name(vma);
 339         if (!name) {
 340                 if (!mm) {
 341                         name = "[vdso]";
 342                         goto done;
 343                 }
 344
 345                 if (vma->vm_start <= mm->brk &&
 346                     vma->vm_end >= mm->start_brk) {
 347                         name = "[heap]";
 348                         goto done;
 349                 }
 350
 351                 if (is_stack(vma))
 352                         name = "[stack]";
 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)
 364 {
 365         show_map_vma(m, v);
 366         m_cache_vma(m, v);
 367         return 0;
 368 }
 369
 370 static const struct seq_operations proc_pid_maps_op = {
 371         .start  = m_start,
 372         .next   = m_next,
 373         .stop   = m_stop,
 374         .show   = show_map
 375 };
 376
 377 static int pid_maps_open(struct inode *inode, struct file *file)
 378 {
 379         return do_maps_open(inode, file, &proc_pid_maps_op);
 380 }
 381
 382 const struct file_operations proc_pid_maps_operations = {
 383         .open           = pid_maps_open,
 384         .read           = seq_read,
 385         .llseek         = seq_lseek,
 386         .release        = proc_map_release,
 387 };
 388
 389 /*
 390  * Proportional Set Size(PSS): my share of RSS.
 391  *
 392  * PSS of a process is the count of pages it has in memory, where each
 393  * page is divided by the number of processes sharing it.  So if a
 394  * process has 1000 pages all to itself, and 1000 shared with one other
 395  * process, its PSS will be 1500.
 396  *
 397  * To keep (accumulated) division errors low, we adopt a 64bit
 398  * fixed-point pss counter to minimize division errors. So (pss >>
 399  * PSS_SHIFT) would be the real byte count.
 400  *
 401  * A shift of 12 before division means (assuming 4K page size):
 402  *      - 1M 3-user-pages add up to 8KB errors;
 403  *      - supports mapcount up to 2^24, or 16M;
 404  *      - supports PSS up to 2^52 bytes, or 4PB.
 405  */
 406 #define PSS_SHIFT 12
 407
 408 #ifdef CONFIG_PROC_PAGE_MONITOR
 409 struct mem_size_stats {
 410         unsigned long resident;
 411         unsigned long shared_clean;
 412         unsigned long shared_dirty;
 413         unsigned long private_clean;
 414         unsigned long private_dirty;
 415         unsigned long referenced;
 416         unsigned long anonymous;
 417         unsigned long lazyfree;
 418         unsigned long anonymous_thp;
 419         unsigned long shmem_thp;
 420         unsigned long file_thp;
 421         unsigned long swap;
 422         unsigned long shared_hugetlb;
 423         unsigned long private_hugetlb;
 424         u64 pss;
 425         u64 pss_anon;
 426         u64 pss_file;
 427         u64 pss_shmem;
 428         u64 pss_locked;
 429         u64 swap_pss;
 430         bool check_shmem_swap;
 431 };
 432
 433 static void smaps_page_accumulate(struct mem_size_stats *mss,
 434                 struct page *page, unsigned long size, unsigned long pss,
 435                 bool dirty, bool locked, bool private)
 436 {
 437         mss->pss += pss;
 438
 439         if (PageAnon(page))
 440                 mss->pss_anon += pss;
 441         else if (PageSwapBacked(page))
 442                 mss->pss_shmem += pss;
 443         else
 444                 mss->pss_file += pss;
 445
 446         if (locked)
 447                 mss->pss_locked += pss;
 448
 449         if (dirty || PageDirty(page)) {
 450                 if (private)
 451                         mss->private_dirty += size;
 452                 else
 453                         mss->shared_dirty += size;
 454         } else {
 455                 if (private)
 456                         mss->private_clean += size;
 457                 else
 458                         mss->shared_clean += size;
 459         }
 460 }
 461
 462 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 463                 bool compound, bool young, bool dirty, bool locked)
 464 {
 465         int i, nr = compound ? compound_nr(page) : 1;
 466         unsigned long size = nr * PAGE_SIZE;
 467
 468         /*
 469          * First accumulate quantities that depend only on |size| and the type
 470          * of the compound page.
 471          */
 472         if (PageAnon(page)) {
 473                 mss->anonymous += size;
 474                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 475                         mss->lazyfree += size;
 476         }
 477
 478         mss->resident += size;
 479         /* Accumulate the size in pages that have been accessed. */
 480         if (young || page_is_young(page) || PageReferenced(page))
 481                 mss->referenced += size;
 482
 483         /*
 484          * Then accumulate quantities that may depend on sharing, or that may
 485          * differ page-by-page.
 486          *
 487          * page_count(page) == 1 guarantees the page is mapped exactly once.
 488          * If any subpage of the compound page mapped with PTE it would elevate
 489          * page_count().
 490          */
 491         if (page_count(page) == 1) {
 492                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 493                         locked, true);
 494                 return;
 495         }
 496         for (i = 0; i < nr; i++, page++) {
 497                 int mapcount = page_mapcount(page);
 498                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 499                 if (mapcount >= 2)
 500                         pss /= mapcount;
 501                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 502                                       mapcount < 2);
 503         }
 504 }
 505
 506 #ifdef CONFIG_SHMEM
 507 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 508                 struct mm_walk *walk)
 509 {
 510         struct mem_size_stats *mss = walk->private;
 511
 512         mss->swap += shmem_partial_swap_usage(
 513                         walk->vma->vm_file->f_mapping, addr, end);
 514
 515         return 0;
 516 }
 517 #else
 518 #define smaps_pte_hole          NULL
 519 #endif /* CONFIG_SHMEM */
 520
 521 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 522                 struct mm_walk *walk)
 523 {
 524         struct mem_size_stats *mss = walk->private;
 525         struct vm_area_struct *vma = walk->vma;
 526         bool locked = !!(vma->vm_flags & VM_LOCKED);
 527         struct page *page = NULL;
 528
 529         if (pte_present(*pte)) {
 530                 page = vm_normal_page(vma, addr, *pte);
 531         } else if (is_swap_pte(*pte)) {
 532                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 533
 534                 if (!non_swap_entry(swpent)) {
 535                         int mapcount;
 536
 537                         mss->swap += PAGE_SIZE;
 538                         mapcount = swp_swapcount(swpent);
 539                         if (mapcount >= 2) {
 540                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 541
 542                                 do_div(pss_delta, mapcount);
 543                                 mss->swap_pss += pss_delta;
 544                         } else {
 545                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 546                         }
 547                 } else if (is_migration_entry(swpent))
 548                         page = migration_entry_to_page(swpent);
 549                 else if (is_device_private_entry(swpent))
 550                         page = device_private_entry_to_page(swpent);
 551         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 552                                                         && pte_none(*pte))) {
 553                 page = find_get_entry(vma->vm_file->f_mapping,
 554                                                 linear_page_index(vma, addr));
 555                 if (!page)
 556                         return;
 557
 558                 if (xa_is_value(page))
 559                         mss->swap += PAGE_SIZE;
 560                 else
 561                         put_page(page);
 562
 563                 return;
 564         }
 565
 566         if (!page)
 567                 return;
 568
 569         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
 570 }
 571
 572 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 573 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 574                 struct mm_walk *walk)
 575 {
 576         struct mem_size_stats *mss = walk->private;
 577         struct vm_area_struct *vma = walk->vma;
 578         bool locked = !!(vma->vm_flags & VM_LOCKED);
 579         struct page *page;
 580
 581         /* FOLL_DUMP will return -EFAULT on huge zero page */
 582         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 583         if (IS_ERR_OR_NULL(page))
 584                 return;
 585         if (PageAnon(page))
 586                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 587         else if (PageSwapBacked(page))
 588                 mss->shmem_thp += HPAGE_PMD_SIZE;
 589         else if (is_zone_device_page(page))
 590                 /* pass */;
 591         else
 592                 mss->file_thp += HPAGE_PMD_SIZE;
 593         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
 594 }
 595 #else
 596 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 597                 struct mm_walk *walk)
 598 {
 599 }
 600 #endif
 601
 602 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 603                            struct mm_walk *walk)
 604 {
 605         struct vm_area_struct *vma = walk->vma;
 606         pte_t *pte;
 607         spinlock_t *ptl;
 608
 609         ptl = pmd_trans_huge_lock(pmd, vma);
 610         if (ptl) {
 611                 if (pmd_present(*pmd))
 612                         smaps_pmd_entry(pmd, addr, walk);
 613                 spin_unlock(ptl);
 614                 goto out;
 615         }
 616
 617         if (pmd_trans_unstable(pmd))
 618                 goto out;
 619         /*
 620          * The mmap_sem held all the way back in m_start() is what
 621          * keeps khugepaged out of here and from collapsing things
 622          * in here.
 623          */
 624         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 625         for (; addr != end; pte++, addr += PAGE_SIZE)
 626                 smaps_pte_entry(pte, addr, walk);
 627         pte_unmap_unlock(pte - 1, ptl);
 628 out:
 629         cond_resched();
 630         return 0;
 631 }
 632
 633 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 634 {
 635         /*
 636          * Don't forget to update Documentation/ on changes.
 637          */
 638         static const char mnemonics[BITS_PER_LONG][2] = {
 639                 /*
 640                  * In case if we meet a flag we don't know about.
 641                  */
 642                 [0 ... (BITS_PER_LONG-1)] = "??",
 643
 644                 [ilog2(VM_READ)]        = "rd",
 645                 [ilog2(VM_WRITE)]       = "wr",
 646                 [ilog2(VM_EXEC)]        = "ex",
 647                 [ilog2(VM_SHARED)]      = "sh",
 648                 [ilog2(VM_MAYREAD)]     = "mr",
 649                 [ilog2(VM_MAYWRITE)]    = "mw",
 650                 [ilog2(VM_MAYEXEC)]     = "me",
 651                 [ilog2(VM_MAYSHARE)]    = "ms",
 652                 [ilog2(VM_GROWSDOWN)]   = "gd",
 653                 [ilog2(VM_PFNMAP)]      = "pf",
 654                 [ilog2(VM_DENYWRITE)]   = "dw",
 655 #ifdef CONFIG_X86_INTEL_MPX
 656                 [ilog2(VM_MPX)]         = "mp",
 657 #endif
 658                 [ilog2(VM_LOCKED)]      = "lo",
 659                 [ilog2(VM_IO)]          = "io",
 660                 [ilog2(VM_SEQ_READ)]    = "sr",
 661                 [ilog2(VM_RAND_READ)]   = "rr",
 662                 [ilog2(VM_DONTCOPY)]    = "dc",
 663                 [ilog2(VM_DONTEXPAND)]  = "de",
 664                 [ilog2(VM_ACCOUNT)]     = "ac",
 665                 [ilog2(VM_NORESERVE)]   = "nr",
 666                 [ilog2(VM_HUGETLB)]     = "ht",
 667                 [ilog2(VM_SYNC)]        = "sf",
 668                 [ilog2(VM_ARCH_1)]      = "ar",
 669                 [ilog2(VM_WIPEONFORK)]  = "wf",
 670                 [ilog2(VM_DONTDUMP)]    = "dd",
 671 #ifdef CONFIG_MEM_SOFT_DIRTY
 672                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 673 #endif
 674                 [ilog2(VM_MIXEDMAP)]    = "mm",
 675                 [ilog2(VM_HUGEPAGE)]    = "hg",
 676                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 677                 [ilog2(VM_MERGEABLE)]   = "mg",
 678                 [ilog2(VM_UFFD_MISSING)]= "um",
 679                 [ilog2(VM_UFFD_WP)]     = "uw",
 680 #ifdef CONFIG_ARCH_HAS_PKEYS
 681                 /* These come out via ProtectionKey: */
 682                 [ilog2(VM_PKEY_BIT0)]   = "",
 683                 [ilog2(VM_PKEY_BIT1)]   = "",
 684                 [ilog2(VM_PKEY_BIT2)]   = "",
 685                 [ilog2(VM_PKEY_BIT3)]   = "",
 686 #if VM_PKEY_BIT4
 687                 [ilog2(VM_PKEY_BIT4)]   = "",
 688 #endif
 689 #endif /* CONFIG_ARCH_HAS_PKEYS */
 690         };
 691         size_t i;
 692
 693         seq_puts(m, "VmFlags: ");
 694         for (i = 0; i < BITS_PER_LONG; i++) {
 695                 if (!mnemonics[i][0])
 696                         continue;
 697                 if (vma->vm_flags & (1UL << i)) {
 698                         seq_putc(m, mnemonics[i][0]);
 699                         seq_putc(m, mnemonics[i][1]);
 700                         seq_putc(m, ' ');
 701                 }
 702         }
 703         seq_putc(m, '\n');
 704 }
 705
 706 #ifdef CONFIG_HUGETLB_PAGE
 707 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 708                                  unsigned long addr, unsigned long end,
 709                                  struct mm_walk *walk)
 710 {
 711         struct mem_size_stats *mss = walk->private;
 712         struct vm_area_struct *vma = walk->vma;
 713         struct page *page = NULL;
 714
 715         if (pte_present(*pte)) {
 716                 page = vm_normal_page(vma, addr, *pte);
 717         } else if (is_swap_pte(*pte)) {
 718                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 719
 720                 if (is_migration_entry(swpent))
 721                         page = migration_entry_to_page(swpent);
 722                 else if (is_device_private_entry(swpent))
 723                         page = device_private_entry_to_page(swpent);
 724         }
 725         if (page) {
 726                 int mapcount = page_mapcount(page);
 727
 728                 if (mapcount >= 2)
 729                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 730                 else
 731                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 732         }
 733         return 0;
 734 }
 735 #else
 736 #define smaps_hugetlb_range     NULL
 737 #endif /* HUGETLB_PAGE */
 738
 739 static const struct mm_walk_ops smaps_walk_ops = {
 740         .pmd_entry              = smaps_pte_range,
 741         .hugetlb_entry          = smaps_hugetlb_range,
 742 };
 743
 744 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 745         .pmd_entry              = smaps_pte_range,
 746         .hugetlb_entry          = smaps_hugetlb_range,
 747         .pte_hole               = smaps_pte_hole,
 748 };
 749
 750 static void smap_gather_stats(struct vm_area_struct *vma,
 751                              struct mem_size_stats *mss)
 752 {
 753 #ifdef CONFIG_SHMEM
 754         /* In case of smaps_rollup, reset the value from previous vma */
 755         mss->check_shmem_swap = false;
 756         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 757                 /*
 758                  * For shared or readonly shmem mappings we know that all
 759                  * swapped out pages belong to the shmem object, and we can
 760                  * obtain the swap value much more efficiently. For private
 761                  * writable mappings, we might have COW pages that are
 762                  * not affected by the parent swapped out pages of the shmem
 763                  * object, so we have to distinguish them during the page walk.
 764                  * Unless we know that the shmem object (or the part mapped by
 765                  * our VMA) has no swapped out pages at all.
 766                  */
 767                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 768
 769                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 770                                         !(vma->vm_flags & VM_WRITE)) {
 771                         mss->swap += shmem_swapped;
 772                 } else {
 773                         mss->check_shmem_swap = true;
 774                         walk_page_vma(vma, &smaps_shmem_walk_ops, mss);
 775                         return;
 776                 }
 777         }
 778 #endif
 779         /* mmap_sem is held in m_start */
 780         walk_page_vma(vma, &smaps_walk_ops, mss);
 781 }
 782
 783 #define SEQ_PUT_DEC(str, val) \
 784                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 785
 786 /* Show the contents common for smaps and smaps_rollup */
 787 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 788         bool rollup_mode)
 789 {
 790         SEQ_PUT_DEC("Rss:            ", mss->resident);
 791         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 792         if (rollup_mode) {
 793                 /*
 794                  * These are meaningful only for smaps_rollup, otherwise two of
 795                  * them are zero, and the other one is the same as Pss.
 796                  */
 797                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 798                         mss->pss_anon >> PSS_SHIFT);
 799                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 800                         mss->pss_file >> PSS_SHIFT);
 801                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 802                         mss->pss_shmem >> PSS_SHIFT);
 803         }
 804         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 805         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 806         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 807         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 808         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 809         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 810         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 811         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 812         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 813         SEQ_PUT_DEC(" kB\nFilePmdMapped: ", mss->file_thp);
 814         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 815         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 816                                   mss->private_hugetlb >> 10, 7);
 817         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 818         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 819                                         mss->swap_pss >> PSS_SHIFT);
 820         SEQ_PUT_DEC(" kB\nLocked:         ",
 821                                         mss->pss_locked >> PSS_SHIFT);
 822         seq_puts(m, " kB\n");
 823 }
 824
 825 static int show_smap(struct seq_file *m, void *v)
 826 {
 827         struct vm_area_struct *vma = v;
 828         struct mem_size_stats mss;
 829
 830         memset(&mss, 0, sizeof(mss));
 831
 832         smap_gather_stats(vma, &mss);
 833
 834         show_map_vma(m, vma);
 835
 836         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 837         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 838         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 839         seq_puts(m, " kB\n");
 840
 841         __show_smap(m, &mss, false);
 842
 843         seq_printf(m, "THPeligible:             %d\n",
 844                    transparent_hugepage_enabled(vma));
 845
 846         if (arch_pkeys_enabled())
 847                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 848         show_smap_vma_flags(m, vma);
 849
 850         m_cache_vma(m, vma);
 851
 852         return 0;
 853 }
 854
 855 static int show_smaps_rollup(struct seq_file *m, void *v)
 856 {
 857         struct proc_maps_private *priv = m->private;
 858         struct mem_size_stats mss;
 859         struct mm_struct *mm;
 860         struct vm_area_struct *vma;
 861         unsigned long last_vma_end = 0;
 862         int ret = 0;
 863
 864         priv->task = get_proc_task(priv->inode);
 865         if (!priv->task)
 866                 return -ESRCH;
 867
 868         mm = priv->mm;
 869         if (!mm || !mmget_not_zero(mm)) {
 870                 ret = -ESRCH;
 871                 goto out_put_task;
 872         }
 873
 874         memset(&mss, 0, sizeof(mss));
 875
 876         ret = down_read_killable(&mm->mmap_sem);
 877         if (ret)
 878                 goto out_put_mm;
 879
 880         hold_task_mempolicy(priv);
 881
 882         for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
 883                 smap_gather_stats(vma, &mss);
 884                 last_vma_end = vma->vm_end;
 885         }
 886
 887         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 888                                last_vma_end, 0, 0, 0, 0);
 889         seq_pad(m, ' ');
 890         seq_puts(m, "[rollup]\n");
 891
 892         __show_smap(m, &mss, true);
 893
 894         release_task_mempolicy(priv);
 895         up_read(&mm->mmap_sem);
 896
 897 out_put_mm:
 898         mmput(mm);
 899 out_put_task:
 900         put_task_struct(priv->task);
 901         priv->task = NULL;
 902
 903         return ret;
 904 }
 905 #undef SEQ_PUT_DEC
 906
 907 static const struct seq_operations proc_pid_smaps_op = {
 908         .start  = m_start,
 909         .next   = m_next,
 910         .stop   = m_stop,
 911         .show   = show_smap
 912 };
 913
 914 static int pid_smaps_open(struct inode *inode, struct file *file)
 915 {
 916         return do_maps_open(inode, file, &proc_pid_smaps_op);
 917 }
 918
 919 static int smaps_rollup_open(struct inode *inode, struct file *file)
 920 {
 921         int ret;
 922         struct proc_maps_private *priv;
 923
 924         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 925         if (!priv)
 926                 return -ENOMEM;
 927
 928         ret = single_open(file, show_smaps_rollup, priv);
 929         if (ret)
 930                 goto out_free;
 931
 932         priv->inode = inode;
 933         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 934         if (IS_ERR(priv->mm)) {
 935                 ret = PTR_ERR(priv->mm);
 936
 937                 single_release(inode, file);
 938                 goto out_free;
 939         }
 940
 941         return 0;
 942
 943 out_free:
 944         kfree(priv);
 945         return ret;
 946 }
 947
 948 static int smaps_rollup_release(struct inode *inode, struct file *file)
 949 {
 950         struct seq_file *seq = file->private_data;
 951         struct proc_maps_private *priv = seq->private;
 952
 953         if (priv->mm)
 954                 mmdrop(priv->mm);
 955
 956         kfree(priv);
 957         return single_release(inode, file);
 958 }
 959
 960 const struct file_operations proc_pid_smaps_operations = {
 961         .open           = pid_smaps_open,
 962         .read           = seq_read,
 963         .llseek         = seq_lseek,
 964         .release        = proc_map_release,
 965 };
 966
 967 const struct file_operations proc_pid_smaps_rollup_operations = {
 968         .open           = smaps_rollup_open,
 969         .read           = seq_read,
 970         .llseek         = seq_lseek,
 971         .release        = smaps_rollup_release,
 972 };
 973
 974 enum clear_refs_types {
 975         CLEAR_REFS_ALL = 1,
 976         CLEAR_REFS_ANON,
 977         CLEAR_REFS_MAPPED,
 978         CLEAR_REFS_SOFT_DIRTY,
 979         CLEAR_REFS_MM_HIWATER_RSS,
 980         CLEAR_REFS_LAST,
 981 };
 982
 983 struct clear_refs_private {
 984         enum clear_refs_types type;
 985 };
 986
 987 #ifdef CONFIG_MEM_SOFT_DIRTY
 988 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 989                 unsigned long addr, pte_t *pte)
 990 {
 991         /*
 992          * The soft-dirty tracker uses #PF-s to catch writes
 993          * to pages, so write-protect the pte as well. See the
 994          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 995          * of how soft-dirty works.
 996          */
 997         pte_t ptent = *pte;
 998
 999         if (pte_present(ptent)) {
1000                 pte_t old_pte;
1001
1002                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1003                 ptent = pte_wrprotect(old_pte);
1004                 ptent = pte_clear_soft_dirty(ptent);
1005                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1006         } else if (is_swap_pte(ptent)) {
1007                 ptent = pte_swp_clear_soft_dirty(ptent);
1008                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1009         }
1010 }
1011 #else
1012 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1013                 unsigned long addr, pte_t *pte)
1014 {
1015 }
1016 #endif
1017
1018 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1019 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1020                 unsigned long addr, pmd_t *pmdp)
1021 {
1022         pmd_t old, pmd = *pmdp;
1023
1024         if (pmd_present(pmd)) {
1025                 /* See comment in change_huge_pmd() */
1026                 old = pmdp_invalidate(vma, addr, pmdp);
1027                 if (pmd_dirty(old))
1028                         pmd = pmd_mkdirty(pmd);
1029                 if (pmd_young(old))
1030                         pmd = pmd_mkyoung(pmd);
1031
1032                 pmd = pmd_wrprotect(pmd);
1033                 pmd = pmd_clear_soft_dirty(pmd);
1034
1035                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1036         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1037                 pmd = pmd_swp_clear_soft_dirty(pmd);
1038                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1039         }
1040 }
1041 #else
1042 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1043                 unsigned long addr, pmd_t *pmdp)
1044 {
1045 }
1046 #endif
1047
1048 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1049                                 unsigned long end, struct mm_walk *walk)
1050 {
1051         struct clear_refs_private *cp = walk->private;
1052         struct vm_area_struct *vma = walk->vma;
1053         pte_t *pte, ptent;
1054         spinlock_t *ptl;
1055         struct page *page;
1056
1057         ptl = pmd_trans_huge_lock(pmd, vma);
1058         if (ptl) {
1059                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1060                         clear_soft_dirty_pmd(vma, addr, pmd);
1061                         goto out;
1062                 }
1063
1064                 if (!pmd_present(*pmd))
1065                         goto out;
1066
1067                 page = pmd_page(*pmd);
1068
1069                 /* Clear accessed and referenced bits. */
1070                 pmdp_test_and_clear_young(vma, addr, pmd);
1071                 test_and_clear_page_young(page);
1072                 ClearPageReferenced(page);
1073 out:
1074                 spin_unlock(ptl);
1075                 return 0;
1076         }
1077
1078         if (pmd_trans_unstable(pmd))
1079                 return 0;
1080
1081         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1082         for (; addr != end; pte++, addr += PAGE_SIZE) {
1083                 ptent = *pte;
1084
1085                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1086                         clear_soft_dirty(vma, addr, pte);
1087                         continue;
1088                 }
1089
1090                 if (!pte_present(ptent))
1091                         continue;
1092
1093                 page = vm_normal_page(vma, addr, ptent);
1094                 if (!page)
1095                         continue;
1096
1097                 /* Clear accessed and referenced bits. */
1098                 ptep_test_and_clear_young(vma, addr, pte);
1099                 test_and_clear_page_young(page);
1100                 ClearPageReferenced(page);
1101         }
1102         pte_unmap_unlock(pte - 1, ptl);
1103         cond_resched();
1104         return 0;
1105 }
1106
1107 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1108                                 struct mm_walk *walk)
1109 {
1110         struct clear_refs_private *cp = walk->private;
1111         struct vm_area_struct *vma = walk->vma;
1112
1113         if (vma->vm_flags & VM_PFNMAP)
1114                 return 1;
1115
1116         /*
1117          * Writing 1 to /proc/pid/clear_refs affects all pages.
1118          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1119          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1120          * Writing 4 to /proc/pid/clear_refs affects all pages.
1121          */
1122         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1123                 return 1;
1124         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1125                 return 1;
1126         return 0;
1127 }
1128
1129 static const struct mm_walk_ops clear_refs_walk_ops = {
1130         .pmd_entry              = clear_refs_pte_range,
1131         .test_walk              = clear_refs_test_walk,
1132 };
1133
1134 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1135                                 size_t count, loff_t *ppos)
1136 {
1137         struct task_struct *task;
1138         char buffer[PROC_NUMBUF];
1139         struct mm_struct *mm;
1140         struct vm_area_struct *vma;
1141         enum clear_refs_types type;
1142         struct mmu_gather tlb;
1143         int itype;
1144         int rv;
1145
1146         memset(buffer, 0, sizeof(buffer));
1147         if (count > sizeof(buffer) - 1)
1148                 count = sizeof(buffer) - 1;
1149         if (copy_from_user(buffer, buf, count))
1150                 return -EFAULT;
1151         rv = kstrtoint(strstrip(buffer), 10, &itype);
1152         if (rv < 0)
1153                 return rv;
1154         type = (enum clear_refs_types)itype;
1155         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1156                 return -EINVAL;
1157
1158         task = get_proc_task(file_inode(file));
1159         if (!task)
1160                 return -ESRCH;
1161         mm = get_task_mm(task);
1162         if (mm) {
1163                 struct mmu_notifier_range range;
1164                 struct clear_refs_private cp = {
1165                         .type = type,
1166                 };
1167
1168                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1169                         if (down_write_killable(&mm->mmap_sem)) {
1170                                 count = -EINTR;
1171                                 goto out_mm;
1172                         }
1173
1174                         /*
1175                          * Writing 5 to /proc/pid/clear_refs resets the peak
1176                          * resident set size to this mm's current rss value.
1177                          */
1178                         reset_mm_hiwater_rss(mm);
1179                         up_write(&mm->mmap_sem);
1180                         goto out_mm;
1181                 }
1182
1183                 if (down_read_killable(&mm->mmap_sem)) {
1184                         count = -EINTR;
1185                         goto out_mm;
1186                 }
1187                 tlb_gather_mmu(&tlb, mm, 0, -1);
1188                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1189                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1190                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1191                                         continue;
1192                                 up_read(&mm->mmap_sem);
1193                                 if (down_write_killable(&mm->mmap_sem)) {
1194                                         count = -EINTR;
1195                                         goto out_mm;
1196                                 }
1197                                 /*
1198                                  * Avoid to modify vma->vm_flags
1199                                  * without locked ops while the
1200                                  * coredump reads the vm_flags.
1201                                  */
1202                                 if (!mmget_still_valid(mm)) {
1203                                         /*
1204                                          * Silently return "count"
1205                                          * like if get_task_mm()
1206                                          * failed. FIXME: should this
1207                                          * function have returned
1208                                          * -ESRCH if get_task_mm()
1209                                          * failed like if
1210                                          * get_proc_task() fails?
1211                                          */
1212                                         up_write(&mm->mmap_sem);
1213                                         goto out_mm;
1214                                 }
1215                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1216                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1217                                         vma_set_page_prot(vma);
1218                                 }
1219                                 downgrade_write(&mm->mmap_sem);
1220                                 break;
1221                         }
1222
1223                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1224                                                 0, NULL, mm, 0, -1UL);
1225                         mmu_notifier_invalidate_range_start(&range);
1226                 }
1227                 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1228                                 &cp);
1229                 if (type == CLEAR_REFS_SOFT_DIRTY)
1230                         mmu_notifier_invalidate_range_end(&range);
1231                 tlb_finish_mmu(&tlb, 0, -1);
1232                 up_read(&mm->mmap_sem);
1233 out_mm:
1234                 mmput(mm);
1235         }
1236         put_task_struct(task);
1237
1238         return count;
1239 }
1240
1241 const struct file_operations proc_clear_refs_operations = {
1242         .write          = clear_refs_write,
1243         .llseek         = noop_llseek,
1244 };
1245
1246 typedef struct {
1247         u64 pme;
1248 } pagemap_entry_t;
1249
1250 struct pagemapread {
1251         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1252         pagemap_entry_t *buffer;
1253         bool show_pfn;
1254 };
1255
1256 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1257 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1258
1259 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1260 #define PM_PFRAME_BITS          55
1261 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1262 #define PM_SOFT_DIRTY           BIT_ULL(55)
1263 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1264 #define PM_FILE                 BIT_ULL(61)
1265 #define PM_SWAP                 BIT_ULL(62)
1266 #define PM_PRESENT              BIT_ULL(63)
1267
1268 #define PM_END_OF_BUFFER    1
1269
1270 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1271 {
1272         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1273 }
1274
1275 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1276                           struct pagemapread *pm)
1277 {
1278         pm->buffer[pm->pos++] = *pme;
1279         if (pm->pos >= pm->len)
1280                 return PM_END_OF_BUFFER;
1281         return 0;
1282 }
1283
1284 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1285                                 struct mm_walk *walk)
1286 {
1287         struct pagemapread *pm = walk->private;
1288         unsigned long addr = start;
1289         int err = 0;
1290
1291         while (addr < end) {
1292                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1293                 pagemap_entry_t pme = make_pme(0, 0);
1294                 /* End of address space hole, which we mark as non-present. */
1295                 unsigned long hole_end;
1296
1297                 if (vma)
1298                         hole_end = min(end, vma->vm_start);
1299                 else
1300                         hole_end = end;
1301
1302                 for (; addr < hole_end; addr += PAGE_SIZE) {
1303                         err = add_to_pagemap(addr, &pme, pm);
1304                         if (err)
1305                                 goto out;
1306                 }
1307
1308                 if (!vma)
1309                         break;
1310
1311                 /* Addresses in the VMA. */
1312                 if (vma->vm_flags & VM_SOFTDIRTY)
1313                         pme = make_pme(0, PM_SOFT_DIRTY);
1314                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1315                         err = add_to_pagemap(addr, &pme, pm);
1316                         if (err)
1317                                 goto out;
1318                 }
1319         }
1320 out:
1321         return err;
1322 }
1323
1324 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1325                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1326 {
1327         u64 frame = 0, flags = 0;
1328         struct page *page = NULL;
1329
1330         if (pte_present(pte)) {
1331                 if (pm->show_pfn)
1332                         frame = pte_pfn(pte);
1333                 flags |= PM_PRESENT;
1334                 page = vm_normal_page(vma, addr, pte);
1335                 if (pte_soft_dirty(pte))
1336                         flags |= PM_SOFT_DIRTY;
1337         } else if (is_swap_pte(pte)) {
1338                 swp_entry_t entry;
1339                 if (pte_swp_soft_dirty(pte))
1340                         flags |= PM_SOFT_DIRTY;
1341                 entry = pte_to_swp_entry(pte);
1342                 if (pm->show_pfn)
1343                         frame = swp_type(entry) |
1344                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1345                 flags |= PM_SWAP;
1346                 if (is_migration_entry(entry))
1347                         page = migration_entry_to_page(entry);
1348
1349                 if (is_device_private_entry(entry))
1350                         page = device_private_entry_to_page(entry);
1351         }
1352
1353         if (page && !PageAnon(page))
1354                 flags |= PM_FILE;
1355         if (page && page_mapcount(page) == 1)
1356                 flags |= PM_MMAP_EXCLUSIVE;
1357         if (vma->vm_flags & VM_SOFTDIRTY)
1358                 flags |= PM_SOFT_DIRTY;
1359
1360         return make_pme(frame, flags);
1361 }
1362
1363 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1364                              struct mm_walk *walk)
1365 {
1366         struct vm_area_struct *vma = walk->vma;
1367         struct pagemapread *pm = walk->private;
1368         spinlock_t *ptl;
1369         pte_t *pte, *orig_pte;
1370         int err = 0;
1371
1372 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1373         ptl = pmd_trans_huge_lock(pmdp, vma);
1374         if (ptl) {
1375                 u64 flags = 0, frame = 0;
1376                 pmd_t pmd = *pmdp;
1377                 struct page *page = NULL;
1378
1379                 if (vma->vm_flags & VM_SOFTDIRTY)
1380                         flags |= PM_SOFT_DIRTY;
1381
1382                 if (pmd_present(pmd)) {
1383                         page = pmd_page(pmd);
1384
1385                         flags |= PM_PRESENT;
1386                         if (pmd_soft_dirty(pmd))
1387                                 flags |= PM_SOFT_DIRTY;
1388                         if (pm->show_pfn)
1389                                 frame = pmd_pfn(pmd) +
1390                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1391                 }
1392 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1393                 else if (is_swap_pmd(pmd)) {
1394                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1395                         unsigned long offset;
1396
1397                         if (pm->show_pfn) {
1398                                 offset = swp_offset(entry) +
1399                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1400                                 frame = swp_type(entry) |
1401                                         (offset << MAX_SWAPFILES_SHIFT);
1402                         }
1403                         flags |= PM_SWAP;
1404                         if (pmd_swp_soft_dirty(pmd))
1405                                 flags |= PM_SOFT_DIRTY;
1406                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1407                         page = migration_entry_to_page(entry);
1408                 }
1409 #endif
1410
1411                 if (page && page_mapcount(page) == 1)
1412                         flags |= PM_MMAP_EXCLUSIVE;
1413
1414                 for (; addr != end; addr += PAGE_SIZE) {
1415                         pagemap_entry_t pme = make_pme(frame, flags);
1416
1417                         err = add_to_pagemap(addr, &pme, pm);
1418                         if (err)
1419                                 break;
1420                         if (pm->show_pfn) {
1421                                 if (flags & PM_PRESENT)
1422                                         frame++;
1423                                 else if (flags & PM_SWAP)
1424                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1425                         }
1426                 }
1427                 spin_unlock(ptl);
1428                 return err;
1429         }
1430
1431         if (pmd_trans_unstable(pmdp))
1432                 return 0;
1433 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1434
1435         /*
1436          * We can assume that @vma always points to a valid one and @end never
1437          * goes beyond vma->vm_end.
1438          */
1439         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1440         for (; addr < end; pte++, addr += PAGE_SIZE) {
1441                 pagemap_entry_t pme;
1442
1443                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1444                 err = add_to_pagemap(addr, &pme, pm);
1445                 if (err)
1446                         break;
1447         }
1448         pte_unmap_unlock(orig_pte, ptl);
1449
1450         cond_resched();
1451
1452         return err;
1453 }
1454
1455 #ifdef CONFIG_HUGETLB_PAGE
1456 /* This function walks within one hugetlb entry in the single call */
1457 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1458                                  unsigned long addr, unsigned long end,
1459                                  struct mm_walk *walk)
1460 {
1461         struct pagemapread *pm = walk->private;
1462         struct vm_area_struct *vma = walk->vma;
1463         u64 flags = 0, frame = 0;
1464         int err = 0;
1465         pte_t pte;
1466
1467         if (vma->vm_flags & VM_SOFTDIRTY)
1468                 flags |= PM_SOFT_DIRTY;
1469
1470         pte = huge_ptep_get(ptep);
1471         if (pte_present(pte)) {
1472                 struct page *page = pte_page(pte);
1473
1474                 if (!PageAnon(page))
1475                         flags |= PM_FILE;
1476
1477                 if (page_mapcount(page) == 1)
1478                         flags |= PM_MMAP_EXCLUSIVE;
1479
1480                 flags |= PM_PRESENT;
1481                 if (pm->show_pfn)
1482                         frame = pte_pfn(pte) +
1483                                 ((addr & ~hmask) >> PAGE_SHIFT);
1484         }
1485
1486         for (; addr != end; addr += PAGE_SIZE) {
1487                 pagemap_entry_t pme = make_pme(frame, flags);
1488
1489                 err = add_to_pagemap(addr, &pme, pm);
1490                 if (err)
1491                         return err;
1492                 if (pm->show_pfn && (flags & PM_PRESENT))
1493                         frame++;
1494         }
1495
1496         cond_resched();
1497
1498         return err;
1499 }
1500 #else
1501 #define pagemap_hugetlb_range   NULL
1502 #endif /* HUGETLB_PAGE */
1503
1504 static const struct mm_walk_ops pagemap_ops = {
1505         .pmd_entry      = pagemap_pmd_range,
1506         .pte_hole       = pagemap_pte_hole,
1507         .hugetlb_entry  = pagemap_hugetlb_range,
1508 };
1509
1510 /*
1511  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1512  *
1513  * For each page in the address space, this file contains one 64-bit entry
1514  * consisting of the following:
1515  *
1516  * Bits 0-54  page frame number (PFN) if present
1517  * Bits 0-4   swap type if swapped
1518  * Bits 5-54  swap offset if swapped
1519  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1520  * Bit  56    page exclusively mapped
1521  * Bits 57-60 zero
1522  * Bit  61    page is file-page or shared-anon
1523  * Bit  62    page swapped
1524  * Bit  63    page present
1525  *
1526  * If the page is not present but in swap, then the PFN contains an
1527  * encoding of the swap file number and the page's offset into the
1528  * swap. Unmapped pages return a null PFN. This allows determining
1529  * precisely which pages are mapped (or in swap) and comparing mapped
1530  * pages between processes.
1531  *
1532  * Efficient users of this interface will use /proc/pid/maps to
1533  * determine which areas of memory are actually mapped and llseek to
1534  * skip over unmapped regions.
1535  */
1536 static ssize_t pagemap_read(struct file *file, char __user *buf,
1537                             size_t count, loff_t *ppos)
1538 {
1539         struct mm_struct *mm = file->private_data;
1540         struct pagemapread pm;
1541         unsigned long src;
1542         unsigned long svpfn;
1543         unsigned long start_vaddr;
1544         unsigned long end_vaddr;
1545         int ret = 0, copied = 0;
1546
1547         if (!mm || !mmget_not_zero(mm))
1548                 goto out;
1549
1550         ret = -EINVAL;
1551         /* file position must be aligned */
1552         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1553                 goto out_mm;
1554
1555         ret = 0;
1556         if (!count)
1557                 goto out_mm;
1558
1559         /* do not disclose physical addresses: attack vector */
1560         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1561
1562         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1563         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1564         ret = -ENOMEM;
1565         if (!pm.buffer)
1566                 goto out_mm;
1567
1568         src = *ppos;
1569         svpfn = src / PM_ENTRY_BYTES;
1570         start_vaddr = svpfn << PAGE_SHIFT;
1571         end_vaddr = mm->task_size;
1572
1573         /* watch out for wraparound */
1574         if (svpfn > mm->task_size >> PAGE_SHIFT)
1575                 start_vaddr = end_vaddr;
1576
1577         /*
1578          * The odds are that this will stop walking way
1579          * before end_vaddr, because the length of the
1580          * user buffer is tracked in "pm", and the walk
1581          * will stop when we hit the end of the buffer.
1582          */
1583         ret = 0;
1584         while (count && (start_vaddr < end_vaddr)) {
1585                 int len;
1586                 unsigned long end;
1587
1588                 pm.pos = 0;
1589                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1590                 /* overflow ? */
1591                 if (end < start_vaddr || end > end_vaddr)
1592                         end = end_vaddr;
1593                 ret = down_read_killable(&mm->mmap_sem);
1594                 if (ret)
1595                         goto out_free;
1596                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1597                 up_read(&mm->mmap_sem);
1598                 start_vaddr = end;
1599
1600                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1601                 if (copy_to_user(buf, pm.buffer, len)) {
1602                         ret = -EFAULT;
1603                         goto out_free;
1604                 }
1605                 copied += len;
1606                 buf += len;
1607                 count -= len;
1608         }
1609         *ppos += copied;
1610         if (!ret || ret == PM_END_OF_BUFFER)
1611                 ret = copied;
1612
1613 out_free:
1614         kfree(pm.buffer);
1615 out_mm:
1616         mmput(mm);
1617 out:
1618         return ret;
1619 }
1620
1621 static int pagemap_open(struct inode *inode, struct file *file)
1622 {
1623         struct mm_struct *mm;
1624
1625         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1626         if (IS_ERR(mm))
1627                 return PTR_ERR(mm);
1628         file->private_data = mm;
1629         return 0;
1630 }
1631
1632 static int pagemap_release(struct inode *inode, struct file *file)
1633 {
1634         struct mm_struct *mm = file->private_data;
1635
1636         if (mm)
1637                 mmdrop(mm);
1638         return 0;
1639 }
1640
1641 const struct file_operations proc_pagemap_operations = {
1642         .llseek         = mem_lseek, /* borrow this */
1643         .read           = pagemap_read,
1644         .open           = pagemap_open,
1645         .release        = pagemap_release,
1646 };
1647 #endif /* CONFIG_PROC_PAGE_MONITOR */
1648
1649 #ifdef CONFIG_NUMA
1650
1651 struct numa_maps {
1652         unsigned long pages;
1653         unsigned long anon;
1654         unsigned long active;
1655         unsigned long writeback;
1656         unsigned long mapcount_max;
1657         unsigned long dirty;
1658         unsigned long swapcache;
1659         unsigned long node[MAX_NUMNODES];
1660 };
1661
1662 struct numa_maps_private {
1663         struct proc_maps_private proc_maps;
1664         struct numa_maps md;
1665 };
1666
1667 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1668                         unsigned long nr_pages)
1669 {
1670         int count = page_mapcount(page);
1671
1672         md->pages += nr_pages;
1673         if (pte_dirty || PageDirty(page))
1674                 md->dirty += nr_pages;
1675
1676         if (PageSwapCache(page))
1677                 md->swapcache += nr_pages;
1678
1679         if (PageActive(page) || PageUnevictable(page))
1680                 md->active += nr_pages;
1681
1682         if (PageWriteback(page))
1683                 md->writeback += nr_pages;
1684
1685         if (PageAnon(page))
1686                 md->anon += nr_pages;
1687
1688         if (count > md->mapcount_max)
1689                 md->mapcount_max = count;
1690
1691         md->node[page_to_nid(page)] += nr_pages;
1692 }
1693
1694 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1695                 unsigned long addr)
1696 {
1697         struct page *page;
1698         int nid;
1699
1700         if (!pte_present(pte))
1701                 return NULL;
1702
1703         page = vm_normal_page(vma, addr, pte);
1704         if (!page)
1705                 return NULL;
1706
1707         if (PageReserved(page))
1708                 return NULL;
1709
1710         nid = page_to_nid(page);
1711         if (!node_isset(nid, node_states[N_MEMORY]))
1712                 return NULL;
1713
1714         return page;
1715 }
1716
1717 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1718 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1719                                               struct vm_area_struct *vma,
1720                                               unsigned long addr)
1721 {
1722         struct page *page;
1723         int nid;
1724
1725         if (!pmd_present(pmd))
1726                 return NULL;
1727
1728         page = vm_normal_page_pmd(vma, addr, pmd);
1729         if (!page)
1730                 return NULL;
1731
1732         if (PageReserved(page))
1733                 return NULL;
1734
1735         nid = page_to_nid(page);
1736         if (!node_isset(nid, node_states[N_MEMORY]))
1737                 return NULL;
1738
1739         return page;
1740 }
1741 #endif
1742
1743 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1744                 unsigned long end, struct mm_walk *walk)
1745 {
1746         struct numa_maps *md = walk->private;
1747         struct vm_area_struct *vma = walk->vma;
1748         spinlock_t *ptl;
1749         pte_t *orig_pte;
1750         pte_t *pte;
1751
1752 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1753         ptl = pmd_trans_huge_lock(pmd, vma);
1754         if (ptl) {
1755                 struct page *page;
1756
1757                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1758                 if (page)
1759                         gather_stats(page, md, pmd_dirty(*pmd),
1760                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1761                 spin_unlock(ptl);
1762                 return 0;
1763         }
1764
1765         if (pmd_trans_unstable(pmd))
1766                 return 0;
1767 #endif
1768         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1769         do {
1770                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1771                 if (!page)
1772                         continue;
1773                 gather_stats(page, md, pte_dirty(*pte), 1);
1774
1775         } while (pte++, addr += PAGE_SIZE, addr != end);
1776         pte_unmap_unlock(orig_pte, ptl);
1777         cond_resched();
1778         return 0;
1779 }
1780 #ifdef CONFIG_HUGETLB_PAGE
1781 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1782                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1783 {
1784         pte_t huge_pte = huge_ptep_get(pte);
1785         struct numa_maps *md;
1786         struct page *page;
1787
1788         if (!pte_present(huge_pte))
1789                 return 0;
1790
1791         page = pte_page(huge_pte);
1792         if (!page)
1793                 return 0;
1794
1795         md = walk->private;
1796         gather_stats(page, md, pte_dirty(huge_pte), 1);
1797         return 0;
1798 }
1799
1800 #else
1801 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1802                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1803 {
1804         return 0;
1805 }
1806 #endif
1807
1808 static const struct mm_walk_ops show_numa_ops = {
1809         .hugetlb_entry = gather_hugetlb_stats,
1810         .pmd_entry = gather_pte_stats,
1811 };
1812
1813 /*
1814  * Display pages allocated per node and memory policy via /proc.
1815  */
1816 static int show_numa_map(struct seq_file *m, void *v)
1817 {
1818         struct numa_maps_private *numa_priv = m->private;
1819         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1820         struct vm_area_struct *vma = v;
1821         struct numa_maps *md = &numa_priv->md;
1822         struct file *file = vma->vm_file;
1823         struct mm_struct *mm = vma->vm_mm;
1824         struct mempolicy *pol;
1825         char buffer[64];
1826         int nid;
1827
1828         if (!mm)
1829                 return 0;
1830
1831         /* Ensure we start with an empty set of numa_maps statistics. */
1832         memset(md, 0, sizeof(*md));
1833
1834         pol = __get_vma_policy(vma, vma->vm_start);
1835         if (pol) {
1836                 mpol_to_str(buffer, sizeof(buffer), pol);
1837                 mpol_cond_put(pol);
1838         } else {
1839                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1840         }
1841
1842         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1843
1844         if (file) {
1845                 seq_puts(m, " file=");
1846                 seq_file_path(m, file, "\n\t= ");
1847         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1848                 seq_puts(m, " heap");
1849         } else if (is_stack(vma)) {
1850                 seq_puts(m, " stack");
1851         }
1852
1853         if (is_vm_hugetlb_page(vma))
1854                 seq_puts(m, " huge");
1855
1856         /* mmap_sem is held by m_start */
1857         walk_page_vma(vma, &show_numa_ops, md);
1858
1859         if (!md->pages)
1860                 goto out;
1861
1862         if (md->anon)
1863                 seq_printf(m, " anon=%lu", md->anon);
1864
1865         if (md->dirty)
1866                 seq_printf(m, " dirty=%lu", md->dirty);
1867
1868         if (md->pages != md->anon && md->pages != md->dirty)
1869                 seq_printf(m, " mapped=%lu", md->pages);
1870
1871         if (md->mapcount_max > 1)
1872                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1873
1874         if (md->swapcache)
1875                 seq_printf(m, " swapcache=%lu", md->swapcache);
1876
1877         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1878                 seq_printf(m, " active=%lu", md->active);
1879
1880         if (md->writeback)
1881                 seq_printf(m, " writeback=%lu", md->writeback);
1882
1883         for_each_node_state(nid, N_MEMORY)
1884                 if (md->node[nid])
1885                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1886
1887         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1888 out:
1889         seq_putc(m, '\n');
1890         m_cache_vma(m, vma);
1891         return 0;
1892 }
1893
1894 static const struct seq_operations proc_pid_numa_maps_op = {
1895         .start  = m_start,
1896         .next   = m_next,
1897         .stop   = m_stop,
1898         .show   = show_numa_map,
1899 };
1900
1901 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1902 {
1903         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1904                                 sizeof(struct numa_maps_private));
1905 }
1906
1907 const struct file_operations proc_pid_numa_maps_operations = {
1908         .open           = pid_numa_maps_open,
1909         .read           = seq_read,
1910         .llseek         = seq_lseek,
1911         .release        = proc_map_release,
1912 };
1913
1914 #endif /* CONFIG_NUMA */