fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/mm.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 swap;
 421         unsigned long shared_hugetlb;
 422         unsigned long private_hugetlb;
 423         u64 pss;
 424         u64 pss_locked;
 425         u64 swap_pss;
 426         bool check_shmem_swap;
 427 };
 428
 429 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 430                 bool compound, bool young, bool dirty, bool locked)
 431 {
 432         int i, nr = compound ? 1 << compound_order(page) : 1;
 433         unsigned long size = nr * PAGE_SIZE;
 434
 435         if (PageAnon(page)) {
 436                 mss->anonymous += size;
 437                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 438                         mss->lazyfree += size;
 439         }
 440
 441         mss->resident += size;
 442         /* Accumulate the size in pages that have been accessed. */
 443         if (young || page_is_young(page) || PageReferenced(page))
 444                 mss->referenced += size;
 445
 446         /*
 447          * page_count(page) == 1 guarantees the page is mapped exactly once.
 448          * If any subpage of the compound page mapped with PTE it would elevate
 449          * page_count().
 450          */
 451         if (page_count(page) == 1) {
 452                 if (dirty || PageDirty(page))
 453                         mss->private_dirty += size;
 454                 else
 455                         mss->private_clean += size;
 456                 mss->pss += (u64)size << PSS_SHIFT;
 457                 if (locked)
 458                         mss->pss_locked += (u64)size << PSS_SHIFT;
 459                 return;
 460         }
 461
 462         for (i = 0; i < nr; i++, page++) {
 463                 int mapcount = page_mapcount(page);
 464                 unsigned long pss = (PAGE_SIZE << PSS_SHIFT);
 465
 466                 if (mapcount >= 2) {
 467                         if (dirty || PageDirty(page))
 468                                 mss->shared_dirty += PAGE_SIZE;
 469                         else
 470                                 mss->shared_clean += PAGE_SIZE;
 471                         mss->pss += pss / mapcount;
 472                         if (locked)
 473                                 mss->pss_locked += pss / mapcount;
 474                 } else {
 475                         if (dirty || PageDirty(page))
 476                                 mss->private_dirty += PAGE_SIZE;
 477                         else
 478                                 mss->private_clean += PAGE_SIZE;
 479                         mss->pss += pss;
 480                         if (locked)
 481                                 mss->pss_locked += pss;
 482                 }
 483         }
 484 }
 485
 486 #ifdef CONFIG_SHMEM
 487 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 488                 struct mm_walk *walk)
 489 {
 490         struct mem_size_stats *mss = walk->private;
 491
 492         mss->swap += shmem_partial_swap_usage(
 493                         walk->vma->vm_file->f_mapping, addr, end);
 494
 495         return 0;
 496 }
 497 #endif
 498
 499 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 500                 struct mm_walk *walk)
 501 {
 502         struct mem_size_stats *mss = walk->private;
 503         struct vm_area_struct *vma = walk->vma;
 504         bool locked = !!(vma->vm_flags & VM_LOCKED);
 505         struct page *page = NULL;
 506
 507         if (pte_present(*pte)) {
 508                 page = vm_normal_page(vma, addr, *pte);
 509         } else if (is_swap_pte(*pte)) {
 510                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 511
 512                 if (!non_swap_entry(swpent)) {
 513                         int mapcount;
 514
 515                         mss->swap += PAGE_SIZE;
 516                         mapcount = swp_swapcount(swpent);
 517                         if (mapcount >= 2) {
 518                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 519
 520                                 do_div(pss_delta, mapcount);
 521                                 mss->swap_pss += pss_delta;
 522                         } else {
 523                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 524                         }
 525                 } else if (is_migration_entry(swpent))
 526                         page = migration_entry_to_page(swpent);
 527                 else if (is_device_private_entry(swpent))
 528                         page = device_private_entry_to_page(swpent);
 529         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 530                                                         && pte_none(*pte))) {
 531                 page = find_get_entry(vma->vm_file->f_mapping,
 532                                                 linear_page_index(vma, addr));
 533                 if (!page)
 534                         return;
 535
 536                 if (xa_is_value(page))
 537                         mss->swap += PAGE_SIZE;
 538                 else
 539                         put_page(page);
 540
 541                 return;
 542         }
 543
 544         if (!page)
 545                 return;
 546
 547         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
 548 }
 549
 550 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 551 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 552                 struct mm_walk *walk)
 553 {
 554         struct mem_size_stats *mss = walk->private;
 555         struct vm_area_struct *vma = walk->vma;
 556         bool locked = !!(vma->vm_flags & VM_LOCKED);
 557         struct page *page;
 558
 559         /* FOLL_DUMP will return -EFAULT on huge zero page */
 560         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 561         if (IS_ERR_OR_NULL(page))
 562                 return;
 563         if (PageAnon(page))
 564                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 565         else if (PageSwapBacked(page))
 566                 mss->shmem_thp += HPAGE_PMD_SIZE;
 567         else if (is_zone_device_page(page))
 568                 /* pass */;
 569         else
 570                 VM_BUG_ON_PAGE(1, page);
 571         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
 572 }
 573 #else
 574 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 575                 struct mm_walk *walk)
 576 {
 577 }
 578 #endif
 579
 580 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 581                            struct mm_walk *walk)
 582 {
 583         struct vm_area_struct *vma = walk->vma;
 584         pte_t *pte;
 585         spinlock_t *ptl;
 586
 587         ptl = pmd_trans_huge_lock(pmd, vma);
 588         if (ptl) {
 589                 if (pmd_present(*pmd))
 590                         smaps_pmd_entry(pmd, addr, walk);
 591                 spin_unlock(ptl);
 592                 goto out;
 593         }
 594
 595         if (pmd_trans_unstable(pmd))
 596                 goto out;
 597         /*
 598          * The mmap_sem held all the way back in m_start() is what
 599          * keeps khugepaged out of here and from collapsing things
 600          * in here.
 601          */
 602         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 603         for (; addr != end; pte++, addr += PAGE_SIZE)
 604                 smaps_pte_entry(pte, addr, walk);
 605         pte_unmap_unlock(pte - 1, ptl);
 606 out:
 607         cond_resched();
 608         return 0;
 609 }
 610
 611 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 612 {
 613         /*
 614          * Don't forget to update Documentation/ on changes.
 615          */
 616         static const char mnemonics[BITS_PER_LONG][2] = {
 617                 /*
 618                  * In case if we meet a flag we don't know about.
 619                  */
 620                 [0 ... (BITS_PER_LONG-1)] = "??",
 621
 622                 [ilog2(VM_READ)]        = "rd",
 623                 [ilog2(VM_WRITE)]       = "wr",
 624                 [ilog2(VM_EXEC)]        = "ex",
 625                 [ilog2(VM_SHARED)]      = "sh",
 626                 [ilog2(VM_MAYREAD)]     = "mr",
 627                 [ilog2(VM_MAYWRITE)]    = "mw",
 628                 [ilog2(VM_MAYEXEC)]     = "me",
 629                 [ilog2(VM_MAYSHARE)]    = "ms",
 630                 [ilog2(VM_GROWSDOWN)]   = "gd",
 631                 [ilog2(VM_PFNMAP)]      = "pf",
 632                 [ilog2(VM_DENYWRITE)]   = "dw",
 633 #ifdef CONFIG_X86_INTEL_MPX
 634                 [ilog2(VM_MPX)]         = "mp",
 635 #endif
 636                 [ilog2(VM_LOCKED)]      = "lo",
 637                 [ilog2(VM_IO)]          = "io",
 638                 [ilog2(VM_SEQ_READ)]    = "sr",
 639                 [ilog2(VM_RAND_READ)]   = "rr",
 640                 [ilog2(VM_DONTCOPY)]    = "dc",
 641                 [ilog2(VM_DONTEXPAND)]  = "de",
 642                 [ilog2(VM_ACCOUNT)]     = "ac",
 643                 [ilog2(VM_NORESERVE)]   = "nr",
 644                 [ilog2(VM_HUGETLB)]     = "ht",
 645                 [ilog2(VM_SYNC)]        = "sf",
 646                 [ilog2(VM_ARCH_1)]      = "ar",
 647                 [ilog2(VM_WIPEONFORK)]  = "wf",
 648                 [ilog2(VM_DONTDUMP)]    = "dd",
 649 #ifdef CONFIG_MEM_SOFT_DIRTY
 650                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 651 #endif
 652                 [ilog2(VM_MIXEDMAP)]    = "mm",
 653                 [ilog2(VM_HUGEPAGE)]    = "hg",
 654                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 655                 [ilog2(VM_MERGEABLE)]   = "mg",
 656                 [ilog2(VM_UFFD_MISSING)]= "um",
 657                 [ilog2(VM_UFFD_WP)]     = "uw",
 658 #ifdef CONFIG_ARCH_HAS_PKEYS
 659                 /* These come out via ProtectionKey: */
 660                 [ilog2(VM_PKEY_BIT0)]   = "",
 661                 [ilog2(VM_PKEY_BIT1)]   = "",
 662                 [ilog2(VM_PKEY_BIT2)]   = "",
 663                 [ilog2(VM_PKEY_BIT3)]   = "",
 664 #if VM_PKEY_BIT4
 665                 [ilog2(VM_PKEY_BIT4)]   = "",
 666 #endif
 667 #endif /* CONFIG_ARCH_HAS_PKEYS */
 668         };
 669         size_t i;
 670
 671         seq_puts(m, "VmFlags: ");
 672         for (i = 0; i < BITS_PER_LONG; i++) {
 673                 if (!mnemonics[i][0])
 674                         continue;
 675                 if (vma->vm_flags & (1UL << i)) {
 676                         seq_putc(m, mnemonics[i][0]);
 677                         seq_putc(m, mnemonics[i][1]);
 678                         seq_putc(m, ' ');
 679                 }
 680         }
 681         seq_putc(m, '\n');
 682 }
 683
 684 #ifdef CONFIG_HUGETLB_PAGE
 685 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 686                                  unsigned long addr, unsigned long end,
 687                                  struct mm_walk *walk)
 688 {
 689         struct mem_size_stats *mss = walk->private;
 690         struct vm_area_struct *vma = walk->vma;
 691         struct page *page = NULL;
 692
 693         if (pte_present(*pte)) {
 694                 page = vm_normal_page(vma, addr, *pte);
 695         } else if (is_swap_pte(*pte)) {
 696                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 697
 698                 if (is_migration_entry(swpent))
 699                         page = migration_entry_to_page(swpent);
 700                 else if (is_device_private_entry(swpent))
 701                         page = device_private_entry_to_page(swpent);
 702         }
 703         if (page) {
 704                 int mapcount = page_mapcount(page);
 705
 706                 if (mapcount >= 2)
 707                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 708                 else
 709                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 710         }
 711         return 0;
 712 }
 713 #endif /* HUGETLB_PAGE */
 714
 715 static void smap_gather_stats(struct vm_area_struct *vma,
 716                              struct mem_size_stats *mss)
 717 {
 718         struct mm_walk smaps_walk = {
 719                 .pmd_entry = smaps_pte_range,
 720 #ifdef CONFIG_HUGETLB_PAGE
 721                 .hugetlb_entry = smaps_hugetlb_range,
 722 #endif
 723                 .mm = vma->vm_mm,
 724         };
 725
 726         smaps_walk.private = mss;
 727
 728 #ifdef CONFIG_SHMEM
 729         /* In case of smaps_rollup, reset the value from previous vma */
 730         mss->check_shmem_swap = false;
 731         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 732                 /*
 733                  * For shared or readonly shmem mappings we know that all
 734                  * swapped out pages belong to the shmem object, and we can
 735                  * obtain the swap value much more efficiently. For private
 736                  * writable mappings, we might have COW pages that are
 737                  * not affected by the parent swapped out pages of the shmem
 738                  * object, so we have to distinguish them during the page walk.
 739                  * Unless we know that the shmem object (or the part mapped by
 740                  * our VMA) has no swapped out pages at all.
 741                  */
 742                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 743
 744                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 745                                         !(vma->vm_flags & VM_WRITE)) {
 746                         mss->swap += shmem_swapped;
 747                 } else {
 748                         mss->check_shmem_swap = true;
 749                         smaps_walk.pte_hole = smaps_pte_hole;
 750                 }
 751         }
 752 #endif
 753         /* mmap_sem is held in m_start */
 754         walk_page_vma(vma, &smaps_walk);
 755 }
 756
 757 #define SEQ_PUT_DEC(str, val) \
 758                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 759
 760 /* Show the contents common for smaps and smaps_rollup */
 761 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss)
 762 {
 763         SEQ_PUT_DEC("Rss:            ", mss->resident);
 764         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 765         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 766         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 767         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 768         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 769         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 770         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 771         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 772         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 773         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 774         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 775         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 776                                   mss->private_hugetlb >> 10, 7);
 777         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 778         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 779                                         mss->swap_pss >> PSS_SHIFT);
 780         SEQ_PUT_DEC(" kB\nLocked:         ",
 781                                         mss->pss_locked >> PSS_SHIFT);
 782         seq_puts(m, " kB\n");
 783 }
 784
 785 static int show_smap(struct seq_file *m, void *v)
 786 {
 787         struct vm_area_struct *vma = v;
 788         struct mem_size_stats mss;
 789
 790         memset(&mss, 0, sizeof(mss));
 791
 792         smap_gather_stats(vma, &mss);
 793
 794         show_map_vma(m, vma);
 795
 796         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 797         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 798         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 799         seq_puts(m, " kB\n");
 800
 801         __show_smap(m, &mss);
 802
 803         seq_printf(m, "THPeligible:    %d\n", transparent_hugepage_enabled(vma));
 804
 805         if (arch_pkeys_enabled())
 806                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 807         show_smap_vma_flags(m, vma);
 808
 809         m_cache_vma(m, vma);
 810
 811         return 0;
 812 }
 813
 814 static int show_smaps_rollup(struct seq_file *m, void *v)
 815 {
 816         struct proc_maps_private *priv = m->private;
 817         struct mem_size_stats mss;
 818         struct mm_struct *mm;
 819         struct vm_area_struct *vma;
 820         unsigned long last_vma_end = 0;
 821         int ret = 0;
 822
 823         priv->task = get_proc_task(priv->inode);
 824         if (!priv->task)
 825                 return -ESRCH;
 826
 827         mm = priv->mm;
 828         if (!mm || !mmget_not_zero(mm)) {
 829                 ret = -ESRCH;
 830                 goto out_put_task;
 831         }
 832
 833         memset(&mss, 0, sizeof(mss));
 834
 835         ret = down_read_killable(&mm->mmap_sem);
 836         if (ret)
 837                 goto out_put_mm;
 838
 839         hold_task_mempolicy(priv);
 840
 841         for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
 842                 smap_gather_stats(vma, &mss);
 843                 last_vma_end = vma->vm_end;
 844         }
 845
 846         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 847                                last_vma_end, 0, 0, 0, 0);
 848         seq_pad(m, ' ');
 849         seq_puts(m, "[rollup]\n");
 850
 851         __show_smap(m, &mss);
 852
 853         release_task_mempolicy(priv);
 854         up_read(&mm->mmap_sem);
 855
 856 out_put_mm:
 857         mmput(mm);
 858 out_put_task:
 859         put_task_struct(priv->task);
 860         priv->task = NULL;
 861
 862         return ret;
 863 }
 864 #undef SEQ_PUT_DEC
 865
 866 static const struct seq_operations proc_pid_smaps_op = {
 867         .start  = m_start,
 868         .next   = m_next,
 869         .stop   = m_stop,
 870         .show   = show_smap
 871 };
 872
 873 static int pid_smaps_open(struct inode *inode, struct file *file)
 874 {
 875         return do_maps_open(inode, file, &proc_pid_smaps_op);
 876 }
 877
 878 static int smaps_rollup_open(struct inode *inode, struct file *file)
 879 {
 880         int ret;
 881         struct proc_maps_private *priv;
 882
 883         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 884         if (!priv)
 885                 return -ENOMEM;
 886
 887         ret = single_open(file, show_smaps_rollup, priv);
 888         if (ret)
 889                 goto out_free;
 890
 891         priv->inode = inode;
 892         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 893         if (IS_ERR(priv->mm)) {
 894                 ret = PTR_ERR(priv->mm);
 895
 896                 single_release(inode, file);
 897                 goto out_free;
 898         }
 899
 900         return 0;
 901
 902 out_free:
 903         kfree(priv);
 904         return ret;
 905 }
 906
 907 static int smaps_rollup_release(struct inode *inode, struct file *file)
 908 {
 909         struct seq_file *seq = file->private_data;
 910         struct proc_maps_private *priv = seq->private;
 911
 912         if (priv->mm)
 913                 mmdrop(priv->mm);
 914
 915         kfree(priv);
 916         return single_release(inode, file);
 917 }
 918
 919 const struct file_operations proc_pid_smaps_operations = {
 920         .open           = pid_smaps_open,
 921         .read           = seq_read,
 922         .llseek         = seq_lseek,
 923         .release        = proc_map_release,
 924 };
 925
 926 const struct file_operations proc_pid_smaps_rollup_operations = {
 927         .open           = smaps_rollup_open,
 928         .read           = seq_read,
 929         .llseek         = seq_lseek,
 930         .release        = smaps_rollup_release,
 931 };
 932
 933 enum clear_refs_types {
 934         CLEAR_REFS_ALL = 1,
 935         CLEAR_REFS_ANON,
 936         CLEAR_REFS_MAPPED,
 937         CLEAR_REFS_SOFT_DIRTY,
 938         CLEAR_REFS_MM_HIWATER_RSS,
 939         CLEAR_REFS_LAST,
 940 };
 941
 942 struct clear_refs_private {
 943         enum clear_refs_types type;
 944 };
 945
 946 #ifdef CONFIG_MEM_SOFT_DIRTY
 947 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 948                 unsigned long addr, pte_t *pte)
 949 {
 950         /*
 951          * The soft-dirty tracker uses #PF-s to catch writes
 952          * to pages, so write-protect the pte as well. See the
 953          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 954          * of how soft-dirty works.
 955          */
 956         pte_t ptent = *pte;
 957
 958         if (pte_present(ptent)) {
 959                 pte_t old_pte;
 960
 961                 old_pte = ptep_modify_prot_start(vma, addr, pte);
 962                 ptent = pte_wrprotect(old_pte);
 963                 ptent = pte_clear_soft_dirty(ptent);
 964                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
 965         } else if (is_swap_pte(ptent)) {
 966                 ptent = pte_swp_clear_soft_dirty(ptent);
 967                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 968         }
 969 }
 970 #else
 971 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 972                 unsigned long addr, pte_t *pte)
 973 {
 974 }
 975 #endif
 976
 977 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 978 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 979                 unsigned long addr, pmd_t *pmdp)
 980 {
 981         pmd_t old, pmd = *pmdp;
 982
 983         if (pmd_present(pmd)) {
 984                 /* See comment in change_huge_pmd() */
 985                 old = pmdp_invalidate(vma, addr, pmdp);
 986                 if (pmd_dirty(old))
 987                         pmd = pmd_mkdirty(pmd);
 988                 if (pmd_young(old))
 989                         pmd = pmd_mkyoung(pmd);
 990
 991                 pmd = pmd_wrprotect(pmd);
 992                 pmd = pmd_clear_soft_dirty(pmd);
 993
 994                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 995         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
 996                 pmd = pmd_swp_clear_soft_dirty(pmd);
 997                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 998         }
 999 }
1000 #else
1001 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1002                 unsigned long addr, pmd_t *pmdp)
1003 {
1004 }
1005 #endif
1006
1007 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1008                                 unsigned long end, struct mm_walk *walk)
1009 {
1010         struct clear_refs_private *cp = walk->private;
1011         struct vm_area_struct *vma = walk->vma;
1012         pte_t *pte, ptent;
1013         spinlock_t *ptl;
1014         struct page *page;
1015
1016         ptl = pmd_trans_huge_lock(pmd, vma);
1017         if (ptl) {
1018                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1019                         clear_soft_dirty_pmd(vma, addr, pmd);
1020                         goto out;
1021                 }
1022
1023                 if (!pmd_present(*pmd))
1024                         goto out;
1025
1026                 page = pmd_page(*pmd);
1027
1028                 /* Clear accessed and referenced bits. */
1029                 pmdp_test_and_clear_young(vma, addr, pmd);
1030                 test_and_clear_page_young(page);
1031                 ClearPageReferenced(page);
1032 out:
1033                 spin_unlock(ptl);
1034                 return 0;
1035         }
1036
1037         if (pmd_trans_unstable(pmd))
1038                 return 0;
1039
1040         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1041         for (; addr != end; pte++, addr += PAGE_SIZE) {
1042                 ptent = *pte;
1043
1044                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1045                         clear_soft_dirty(vma, addr, pte);
1046                         continue;
1047                 }
1048
1049                 if (!pte_present(ptent))
1050                         continue;
1051
1052                 page = vm_normal_page(vma, addr, ptent);
1053                 if (!page)
1054                         continue;
1055
1056                 /* Clear accessed and referenced bits. */
1057                 ptep_test_and_clear_young(vma, addr, pte);
1058                 test_and_clear_page_young(page);
1059                 ClearPageReferenced(page);
1060         }
1061         pte_unmap_unlock(pte - 1, ptl);
1062         cond_resched();
1063         return 0;
1064 }
1065
1066 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1067                                 struct mm_walk *walk)
1068 {
1069         struct clear_refs_private *cp = walk->private;
1070         struct vm_area_struct *vma = walk->vma;
1071
1072         if (vma->vm_flags & VM_PFNMAP)
1073                 return 1;
1074
1075         /*
1076          * Writing 1 to /proc/pid/clear_refs affects all pages.
1077          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1078          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1079          * Writing 4 to /proc/pid/clear_refs affects all pages.
1080          */
1081         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1082                 return 1;
1083         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1084                 return 1;
1085         return 0;
1086 }
1087
1088 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1089                                 size_t count, loff_t *ppos)
1090 {
1091         struct task_struct *task;
1092         char buffer[PROC_NUMBUF];
1093         struct mm_struct *mm;
1094         struct vm_area_struct *vma;
1095         enum clear_refs_types type;
1096         struct mmu_gather tlb;
1097         int itype;
1098         int rv;
1099
1100         memset(buffer, 0, sizeof(buffer));
1101         if (count > sizeof(buffer) - 1)
1102                 count = sizeof(buffer) - 1;
1103         if (copy_from_user(buffer, buf, count))
1104                 return -EFAULT;
1105         rv = kstrtoint(strstrip(buffer), 10, &itype);
1106         if (rv < 0)
1107                 return rv;
1108         type = (enum clear_refs_types)itype;
1109         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1110                 return -EINVAL;
1111
1112         task = get_proc_task(file_inode(file));
1113         if (!task)
1114                 return -ESRCH;
1115         mm = get_task_mm(task);
1116         if (mm) {
1117                 struct mmu_notifier_range range;
1118                 struct clear_refs_private cp = {
1119                         .type = type,
1120                 };
1121                 struct mm_walk clear_refs_walk = {
1122                         .pmd_entry = clear_refs_pte_range,
1123                         .test_walk = clear_refs_test_walk,
1124                         .mm = mm,
1125                         .private = &cp,
1126                 };
1127
1128                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1129                         if (down_write_killable(&mm->mmap_sem)) {
1130                                 count = -EINTR;
1131                                 goto out_mm;
1132                         }
1133
1134                         /*
1135                          * Writing 5 to /proc/pid/clear_refs resets the peak
1136                          * resident set size to this mm's current rss value.
1137                          */
1138                         reset_mm_hiwater_rss(mm);
1139                         up_write(&mm->mmap_sem);
1140                         goto out_mm;
1141                 }
1142
1143                 down_read(&mm->mmap_sem);
1144                 tlb_gather_mmu(&tlb, mm, 0, -1);
1145                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1146                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1147                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1148                                         continue;
1149                                 up_read(&mm->mmap_sem);
1150                                 if (down_write_killable(&mm->mmap_sem)) {
1151                                         count = -EINTR;
1152                                         goto out_mm;
1153                                 }
1154                                 /*
1155                                  * Avoid to modify vma->vm_flags
1156                                  * without locked ops while the
1157                                  * coredump reads the vm_flags.
1158                                  */
1159                                 if (!mmget_still_valid(mm)) {
1160                                         /*
1161                                          * Silently return "count"
1162                                          * like if get_task_mm()
1163                                          * failed. FIXME: should this
1164                                          * function have returned
1165                                          * -ESRCH if get_task_mm()
1166                                          * failed like if
1167                                          * get_proc_task() fails?
1168                                          */
1169                                         up_write(&mm->mmap_sem);
1170                                         goto out_mm;
1171                                 }
1172                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1173                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1174                                         vma_set_page_prot(vma);
1175                                 }
1176                                 downgrade_write(&mm->mmap_sem);
1177                                 break;
1178                         }
1179
1180                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1181                                                 0, NULL, mm, 0, -1UL);
1182                         mmu_notifier_invalidate_range_start(&range);
1183                 }
1184                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1185                 if (type == CLEAR_REFS_SOFT_DIRTY)
1186                         mmu_notifier_invalidate_range_end(&range);
1187                 tlb_finish_mmu(&tlb, 0, -1);
1188                 up_read(&mm->mmap_sem);
1189 out_mm:
1190                 mmput(mm);
1191         }
1192         put_task_struct(task);
1193
1194         return count;
1195 }
1196
1197 const struct file_operations proc_clear_refs_operations = {
1198         .write          = clear_refs_write,
1199         .llseek         = noop_llseek,
1200 };
1201
1202 typedef struct {
1203         u64 pme;
1204 } pagemap_entry_t;
1205
1206 struct pagemapread {
1207         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1208         pagemap_entry_t *buffer;
1209         bool show_pfn;
1210 };
1211
1212 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1213 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1214
1215 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1216 #define PM_PFRAME_BITS          55
1217 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1218 #define PM_SOFT_DIRTY           BIT_ULL(55)
1219 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1220 #define PM_FILE                 BIT_ULL(61)
1221 #define PM_SWAP                 BIT_ULL(62)
1222 #define PM_PRESENT              BIT_ULL(63)
1223
1224 #define PM_END_OF_BUFFER    1
1225
1226 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1227 {
1228         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1229 }
1230
1231 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1232                           struct pagemapread *pm)
1233 {
1234         pm->buffer[pm->pos++] = *pme;
1235         if (pm->pos >= pm->len)
1236                 return PM_END_OF_BUFFER;
1237         return 0;
1238 }
1239
1240 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1241                                 struct mm_walk *walk)
1242 {
1243         struct pagemapread *pm = walk->private;
1244         unsigned long addr = start;
1245         int err = 0;
1246
1247         while (addr < end) {
1248                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1249                 pagemap_entry_t pme = make_pme(0, 0);
1250                 /* End of address space hole, which we mark as non-present. */
1251                 unsigned long hole_end;
1252
1253                 if (vma)
1254                         hole_end = min(end, vma->vm_start);
1255                 else
1256                         hole_end = end;
1257
1258                 for (; addr < hole_end; addr += PAGE_SIZE) {
1259                         err = add_to_pagemap(addr, &pme, pm);
1260                         if (err)
1261                                 goto out;
1262                 }
1263
1264                 if (!vma)
1265                         break;
1266
1267                 /* Addresses in the VMA. */
1268                 if (vma->vm_flags & VM_SOFTDIRTY)
1269                         pme = make_pme(0, PM_SOFT_DIRTY);
1270                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1271                         err = add_to_pagemap(addr, &pme, pm);
1272                         if (err)
1273                                 goto out;
1274                 }
1275         }
1276 out:
1277         return err;
1278 }
1279
1280 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1281                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1282 {
1283         u64 frame = 0, flags = 0;
1284         struct page *page = NULL;
1285
1286         if (pte_present(pte)) {
1287                 if (pm->show_pfn)
1288                         frame = pte_pfn(pte);
1289                 flags |= PM_PRESENT;
1290                 page = _vm_normal_page(vma, addr, pte, true);
1291                 if (pte_soft_dirty(pte))
1292                         flags |= PM_SOFT_DIRTY;
1293         } else if (is_swap_pte(pte)) {
1294                 swp_entry_t entry;
1295                 if (pte_swp_soft_dirty(pte))
1296                         flags |= PM_SOFT_DIRTY;
1297                 entry = pte_to_swp_entry(pte);
1298                 if (pm->show_pfn)
1299                         frame = swp_type(entry) |
1300                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1301                 flags |= PM_SWAP;
1302                 if (is_migration_entry(entry))
1303                         page = migration_entry_to_page(entry);
1304
1305                 if (is_device_private_entry(entry))
1306                         page = device_private_entry_to_page(entry);
1307         }
1308
1309         if (page && !PageAnon(page))
1310                 flags |= PM_FILE;
1311         if (page && page_mapcount(page) == 1)
1312                 flags |= PM_MMAP_EXCLUSIVE;
1313         if (vma->vm_flags & VM_SOFTDIRTY)
1314                 flags |= PM_SOFT_DIRTY;
1315
1316         return make_pme(frame, flags);
1317 }
1318
1319 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1320                              struct mm_walk *walk)
1321 {
1322         struct vm_area_struct *vma = walk->vma;
1323         struct pagemapread *pm = walk->private;
1324         spinlock_t *ptl;
1325         pte_t *pte, *orig_pte;
1326         int err = 0;
1327
1328 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1329         ptl = pmd_trans_huge_lock(pmdp, vma);
1330         if (ptl) {
1331                 u64 flags = 0, frame = 0;
1332                 pmd_t pmd = *pmdp;
1333                 struct page *page = NULL;
1334
1335                 if (vma->vm_flags & VM_SOFTDIRTY)
1336                         flags |= PM_SOFT_DIRTY;
1337
1338                 if (pmd_present(pmd)) {
1339                         page = pmd_page(pmd);
1340
1341                         flags |= PM_PRESENT;
1342                         if (pmd_soft_dirty(pmd))
1343                                 flags |= PM_SOFT_DIRTY;
1344                         if (pm->show_pfn)
1345                                 frame = pmd_pfn(pmd) +
1346                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1347                 }
1348 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1349                 else if (is_swap_pmd(pmd)) {
1350                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1351                         unsigned long offset;
1352
1353                         if (pm->show_pfn) {
1354                                 offset = swp_offset(entry) +
1355                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1356                                 frame = swp_type(entry) |
1357                                         (offset << MAX_SWAPFILES_SHIFT);
1358                         }
1359                         flags |= PM_SWAP;
1360                         if (pmd_swp_soft_dirty(pmd))
1361                                 flags |= PM_SOFT_DIRTY;
1362                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1363                         page = migration_entry_to_page(entry);
1364                 }
1365 #endif
1366
1367                 if (page && page_mapcount(page) == 1)
1368                         flags |= PM_MMAP_EXCLUSIVE;
1369
1370                 for (; addr != end; addr += PAGE_SIZE) {
1371                         pagemap_entry_t pme = make_pme(frame, flags);
1372
1373                         err = add_to_pagemap(addr, &pme, pm);
1374                         if (err)
1375                                 break;
1376                         if (pm->show_pfn) {
1377                                 if (flags & PM_PRESENT)
1378                                         frame++;
1379                                 else if (flags & PM_SWAP)
1380                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1381                         }
1382                 }
1383                 spin_unlock(ptl);
1384                 return err;
1385         }
1386
1387         if (pmd_trans_unstable(pmdp))
1388                 return 0;
1389 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1390
1391         /*
1392          * We can assume that @vma always points to a valid one and @end never
1393          * goes beyond vma->vm_end.
1394          */
1395         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1396         for (; addr < end; pte++, addr += PAGE_SIZE) {
1397                 pagemap_entry_t pme;
1398
1399                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1400                 err = add_to_pagemap(addr, &pme, pm);
1401                 if (err)
1402                         break;
1403         }
1404         pte_unmap_unlock(orig_pte, ptl);
1405
1406         cond_resched();
1407
1408         return err;
1409 }
1410
1411 #ifdef CONFIG_HUGETLB_PAGE
1412 /* This function walks within one hugetlb entry in the single call */
1413 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1414                                  unsigned long addr, unsigned long end,
1415                                  struct mm_walk *walk)
1416 {
1417         struct pagemapread *pm = walk->private;
1418         struct vm_area_struct *vma = walk->vma;
1419         u64 flags = 0, frame = 0;
1420         int err = 0;
1421         pte_t pte;
1422
1423         if (vma->vm_flags & VM_SOFTDIRTY)
1424                 flags |= PM_SOFT_DIRTY;
1425
1426         pte = huge_ptep_get(ptep);
1427         if (pte_present(pte)) {
1428                 struct page *page = pte_page(pte);
1429
1430                 if (!PageAnon(page))
1431                         flags |= PM_FILE;
1432
1433                 if (page_mapcount(page) == 1)
1434                         flags |= PM_MMAP_EXCLUSIVE;
1435
1436                 flags |= PM_PRESENT;
1437                 if (pm->show_pfn)
1438                         frame = pte_pfn(pte) +
1439                                 ((addr & ~hmask) >> PAGE_SHIFT);
1440         }
1441
1442         for (; addr != end; addr += PAGE_SIZE) {
1443                 pagemap_entry_t pme = make_pme(frame, flags);
1444
1445                 err = add_to_pagemap(addr, &pme, pm);
1446                 if (err)
1447                         return err;
1448                 if (pm->show_pfn && (flags & PM_PRESENT))
1449                         frame++;
1450         }
1451
1452         cond_resched();
1453
1454         return err;
1455 }
1456 #endif /* HUGETLB_PAGE */
1457
1458 /*
1459  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1460  *
1461  * For each page in the address space, this file contains one 64-bit entry
1462  * consisting of the following:
1463  *
1464  * Bits 0-54  page frame number (PFN) if present
1465  * Bits 0-4   swap type if swapped
1466  * Bits 5-54  swap offset if swapped
1467  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1468  * Bit  56    page exclusively mapped
1469  * Bits 57-60 zero
1470  * Bit  61    page is file-page or shared-anon
1471  * Bit  62    page swapped
1472  * Bit  63    page present
1473  *
1474  * If the page is not present but in swap, then the PFN contains an
1475  * encoding of the swap file number and the page's offset into the
1476  * swap. Unmapped pages return a null PFN. This allows determining
1477  * precisely which pages are mapped (or in swap) and comparing mapped
1478  * pages between processes.
1479  *
1480  * Efficient users of this interface will use /proc/pid/maps to
1481  * determine which areas of memory are actually mapped and llseek to
1482  * skip over unmapped regions.
1483  */
1484 static ssize_t pagemap_read(struct file *file, char __user *buf,
1485                             size_t count, loff_t *ppos)
1486 {
1487         struct mm_struct *mm = file->private_data;
1488         struct pagemapread pm;
1489         struct mm_walk pagemap_walk = {};
1490         unsigned long src;
1491         unsigned long svpfn;
1492         unsigned long start_vaddr;
1493         unsigned long end_vaddr;
1494         int ret = 0, copied = 0;
1495
1496         if (!mm || !mmget_not_zero(mm))
1497                 goto out;
1498
1499         ret = -EINVAL;
1500         /* file position must be aligned */
1501         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1502                 goto out_mm;
1503
1504         ret = 0;
1505         if (!count)
1506                 goto out_mm;
1507
1508         /* do not disclose physical addresses: attack vector */
1509         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1510
1511         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1512         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1513         ret = -ENOMEM;
1514         if (!pm.buffer)
1515                 goto out_mm;
1516
1517         pagemap_walk.pmd_entry = pagemap_pmd_range;
1518         pagemap_walk.pte_hole = pagemap_pte_hole;
1519 #ifdef CONFIG_HUGETLB_PAGE
1520         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1521 #endif
1522         pagemap_walk.mm = mm;
1523         pagemap_walk.private = &pm;
1524
1525         src = *ppos;
1526         svpfn = src / PM_ENTRY_BYTES;
1527         start_vaddr = svpfn << PAGE_SHIFT;
1528         end_vaddr = mm->task_size;
1529
1530         /* watch out for wraparound */
1531         if (svpfn > mm->task_size >> PAGE_SHIFT)
1532                 start_vaddr = end_vaddr;
1533
1534         /*
1535          * The odds are that this will stop walking way
1536          * before end_vaddr, because the length of the
1537          * user buffer is tracked in "pm", and the walk
1538          * will stop when we hit the end of the buffer.
1539          */
1540         ret = 0;
1541         while (count && (start_vaddr < end_vaddr)) {
1542                 int len;
1543                 unsigned long end;
1544
1545                 pm.pos = 0;
1546                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1547                 /* overflow ? */
1548                 if (end < start_vaddr || end > end_vaddr)
1549                         end = end_vaddr;
1550                 ret = down_read_killable(&mm->mmap_sem);
1551                 if (ret)
1552                         goto out_free;
1553                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1554                 up_read(&mm->mmap_sem);
1555                 start_vaddr = end;
1556
1557                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1558                 if (copy_to_user(buf, pm.buffer, len)) {
1559                         ret = -EFAULT;
1560                         goto out_free;
1561                 }
1562                 copied += len;
1563                 buf += len;
1564                 count -= len;
1565         }
1566         *ppos += copied;
1567         if (!ret || ret == PM_END_OF_BUFFER)
1568                 ret = copied;
1569
1570 out_free:
1571         kfree(pm.buffer);
1572 out_mm:
1573         mmput(mm);
1574 out:
1575         return ret;
1576 }
1577
1578 static int pagemap_open(struct inode *inode, struct file *file)
1579 {
1580         struct mm_struct *mm;
1581
1582         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1583         if (IS_ERR(mm))
1584                 return PTR_ERR(mm);
1585         file->private_data = mm;
1586         return 0;
1587 }
1588
1589 static int pagemap_release(struct inode *inode, struct file *file)
1590 {
1591         struct mm_struct *mm = file->private_data;
1592
1593         if (mm)
1594                 mmdrop(mm);
1595         return 0;
1596 }
1597
1598 const struct file_operations proc_pagemap_operations = {
1599         .llseek         = mem_lseek, /* borrow this */
1600         .read           = pagemap_read,
1601         .open           = pagemap_open,
1602         .release        = pagemap_release,
1603 };
1604 #endif /* CONFIG_PROC_PAGE_MONITOR */
1605
1606 #ifdef CONFIG_NUMA
1607
1608 struct numa_maps {
1609         unsigned long pages;
1610         unsigned long anon;
1611         unsigned long active;
1612         unsigned long writeback;
1613         unsigned long mapcount_max;
1614         unsigned long dirty;
1615         unsigned long swapcache;
1616         unsigned long node[MAX_NUMNODES];
1617 };
1618
1619 struct numa_maps_private {
1620         struct proc_maps_private proc_maps;
1621         struct numa_maps md;
1622 };
1623
1624 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1625                         unsigned long nr_pages)
1626 {
1627         int count = page_mapcount(page);
1628
1629         md->pages += nr_pages;
1630         if (pte_dirty || PageDirty(page))
1631                 md->dirty += nr_pages;
1632
1633         if (PageSwapCache(page))
1634                 md->swapcache += nr_pages;
1635
1636         if (PageActive(page) || PageUnevictable(page))
1637                 md->active += nr_pages;
1638
1639         if (PageWriteback(page))
1640                 md->writeback += nr_pages;
1641
1642         if (PageAnon(page))
1643                 md->anon += nr_pages;
1644
1645         if (count > md->mapcount_max)
1646                 md->mapcount_max = count;
1647
1648         md->node[page_to_nid(page)] += nr_pages;
1649 }
1650
1651 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1652                 unsigned long addr)
1653 {
1654         struct page *page;
1655         int nid;
1656
1657         if (!pte_present(pte))
1658                 return NULL;
1659
1660         page = vm_normal_page(vma, addr, pte);
1661         if (!page)
1662                 return NULL;
1663
1664         if (PageReserved(page))
1665                 return NULL;
1666
1667         nid = page_to_nid(page);
1668         if (!node_isset(nid, node_states[N_MEMORY]))
1669                 return NULL;
1670
1671         return page;
1672 }
1673
1674 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1675 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1676                                               struct vm_area_struct *vma,
1677                                               unsigned long addr)
1678 {
1679         struct page *page;
1680         int nid;
1681
1682         if (!pmd_present(pmd))
1683                 return NULL;
1684
1685         page = vm_normal_page_pmd(vma, addr, pmd);
1686         if (!page)
1687                 return NULL;
1688
1689         if (PageReserved(page))
1690                 return NULL;
1691
1692         nid = page_to_nid(page);
1693         if (!node_isset(nid, node_states[N_MEMORY]))
1694                 return NULL;
1695
1696         return page;
1697 }
1698 #endif
1699
1700 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1701                 unsigned long end, struct mm_walk *walk)
1702 {
1703         struct numa_maps *md = walk->private;
1704         struct vm_area_struct *vma = walk->vma;
1705         spinlock_t *ptl;
1706         pte_t *orig_pte;
1707         pte_t *pte;
1708
1709 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1710         ptl = pmd_trans_huge_lock(pmd, vma);
1711         if (ptl) {
1712                 struct page *page;
1713
1714                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1715                 if (page)
1716                         gather_stats(page, md, pmd_dirty(*pmd),
1717                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1718                 spin_unlock(ptl);
1719                 return 0;
1720         }
1721
1722         if (pmd_trans_unstable(pmd))
1723                 return 0;
1724 #endif
1725         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1726         do {
1727                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1728                 if (!page)
1729                         continue;
1730                 gather_stats(page, md, pte_dirty(*pte), 1);
1731
1732         } while (pte++, addr += PAGE_SIZE, addr != end);
1733         pte_unmap_unlock(orig_pte, ptl);
1734         cond_resched();
1735         return 0;
1736 }
1737 #ifdef CONFIG_HUGETLB_PAGE
1738 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1739                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1740 {
1741         pte_t huge_pte = huge_ptep_get(pte);
1742         struct numa_maps *md;
1743         struct page *page;
1744
1745         if (!pte_present(huge_pte))
1746                 return 0;
1747
1748         page = pte_page(huge_pte);
1749         if (!page)
1750                 return 0;
1751
1752         md = walk->private;
1753         gather_stats(page, md, pte_dirty(huge_pte), 1);
1754         return 0;
1755 }
1756
1757 #else
1758 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1759                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1760 {
1761         return 0;
1762 }
1763 #endif
1764
1765 /*
1766  * Display pages allocated per node and memory policy via /proc.
1767  */
1768 static int show_numa_map(struct seq_file *m, void *v)
1769 {
1770         struct numa_maps_private *numa_priv = m->private;
1771         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1772         struct vm_area_struct *vma = v;
1773         struct numa_maps *md = &numa_priv->md;
1774         struct file *file = vma->vm_file;
1775         struct mm_struct *mm = vma->vm_mm;
1776         struct mm_walk walk = {
1777                 .hugetlb_entry = gather_hugetlb_stats,
1778                 .pmd_entry = gather_pte_stats,
1779                 .private = md,
1780                 .mm = mm,
1781         };
1782         struct mempolicy *pol;
1783         char buffer[64];
1784         int nid;
1785
1786         if (!mm)
1787                 return 0;
1788
1789         /* Ensure we start with an empty set of numa_maps statistics. */
1790         memset(md, 0, sizeof(*md));
1791
1792         pol = __get_vma_policy(vma, vma->vm_start);
1793         if (pol) {
1794                 mpol_to_str(buffer, sizeof(buffer), pol);
1795                 mpol_cond_put(pol);
1796         } else {
1797                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1798         }
1799
1800         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1801
1802         if (file) {
1803                 seq_puts(m, " file=");
1804                 seq_file_path(m, file, "\n\t= ");
1805         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1806                 seq_puts(m, " heap");
1807         } else if (is_stack(vma)) {
1808                 seq_puts(m, " stack");
1809         }
1810
1811         if (is_vm_hugetlb_page(vma))
1812                 seq_puts(m, " huge");
1813
1814         /* mmap_sem is held by m_start */
1815         walk_page_vma(vma, &walk);
1816
1817         if (!md->pages)
1818                 goto out;
1819
1820         if (md->anon)
1821                 seq_printf(m, " anon=%lu", md->anon);
1822
1823         if (md->dirty)
1824                 seq_printf(m, " dirty=%lu", md->dirty);
1825
1826         if (md->pages != md->anon && md->pages != md->dirty)
1827                 seq_printf(m, " mapped=%lu", md->pages);
1828
1829         if (md->mapcount_max > 1)
1830                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1831
1832         if (md->swapcache)
1833                 seq_printf(m, " swapcache=%lu", md->swapcache);
1834
1835         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1836                 seq_printf(m, " active=%lu", md->active);
1837
1838         if (md->writeback)
1839                 seq_printf(m, " writeback=%lu", md->writeback);
1840
1841         for_each_node_state(nid, N_MEMORY)
1842                 if (md->node[nid])
1843                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1844
1845         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1846 out:
1847         seq_putc(m, '\n');
1848         m_cache_vma(m, vma);
1849         return 0;
1850 }
1851
1852 static const struct seq_operations proc_pid_numa_maps_op = {
1853         .start  = m_start,
1854         .next   = m_next,
1855         .stop   = m_stop,
1856         .show   = show_numa_map,
1857 };
1858
1859 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1860 {
1861         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1862                                 sizeof(struct numa_maps_private));
1863 }
1864
1865 const struct file_operations proc_pid_numa_maps_operations = {
1866         .open           = pid_numa_maps_open,
1867         .read           = seq_read,
1868         .llseek         = seq_lseek,
1869         .release        = proc_map_release,
1870 };
1871
1872 #endif /* CONFIG_NUMA */