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_anon;
 425         u64 pss_file;
 426         u64 pss_shmem;
 427         u64 pss_locked;
 428         u64 swap_pss;
 429         bool check_shmem_swap;
 430 };
 431
 432 static void smaps_page_accumulate(struct mem_size_stats *mss,
 433                 struct page *page, unsigned long size, unsigned long pss,
 434                 bool dirty, bool locked, bool private)
 435 {
 436         mss->pss += pss;
 437
 438         if (PageAnon(page))
 439                 mss->pss_anon += pss;
 440         else if (PageSwapBacked(page))
 441                 mss->pss_shmem += pss;
 442         else
 443                 mss->pss_file += pss;
 444
 445         if (locked)
 446                 mss->pss_locked += pss;
 447
 448         if (dirty || PageDirty(page)) {
 449                 if (private)
 450                         mss->private_dirty += size;
 451                 else
 452                         mss->shared_dirty += size;
 453         } else {
 454                 if (private)
 455                         mss->private_clean += size;
 456                 else
 457                         mss->shared_clean += size;
 458         }
 459 }
 460
 461 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 462                 bool compound, bool young, bool dirty, bool locked)
 463 {
 464         int i, nr = compound ? 1 << compound_order(page) : 1;
 465         unsigned long size = nr * PAGE_SIZE;
 466
 467         /*
 468          * First accumulate quantities that depend only on |size| and the type
 469          * of the compound page.
 470          */
 471         if (PageAnon(page)) {
 472                 mss->anonymous += size;
 473                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 474                         mss->lazyfree += size;
 475         }
 476
 477         mss->resident += size;
 478         /* Accumulate the size in pages that have been accessed. */
 479         if (young || page_is_young(page) || PageReferenced(page))
 480                 mss->referenced += size;
 481
 482         /*
 483          * Then accumulate quantities that may depend on sharing, or that may
 484          * differ page-by-page.
 485          *
 486          * page_count(page) == 1 guarantees the page is mapped exactly once.
 487          * If any subpage of the compound page mapped with PTE it would elevate
 488          * page_count().
 489          */
 490         if (page_count(page) == 1) {
 491                 smaps_page_accumulate(mss, page, size, size << PSS_SHIFT, dirty,
 492                         locked, true);
 493                 return;
 494         }
 495         for (i = 0; i < nr; i++, page++) {
 496                 int mapcount = page_mapcount(page);
 497                 unsigned long pss = PAGE_SIZE << PSS_SHIFT;
 498                 if (mapcount >= 2)
 499                         pss /= mapcount;
 500                 smaps_page_accumulate(mss, page, PAGE_SIZE, pss, dirty, locked,
 501                                       mapcount < 2);
 502         }
 503 }
 504
 505 #ifdef CONFIG_SHMEM
 506 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 507                 struct mm_walk *walk)
 508 {
 509         struct mem_size_stats *mss = walk->private;
 510
 511         mss->swap += shmem_partial_swap_usage(
 512                         walk->vma->vm_file->f_mapping, addr, end);
 513
 514         return 0;
 515 }
 516 #endif
 517
 518 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 519                 struct mm_walk *walk)
 520 {
 521         struct mem_size_stats *mss = walk->private;
 522         struct vm_area_struct *vma = walk->vma;
 523         bool locked = !!(vma->vm_flags & VM_LOCKED);
 524         struct page *page = NULL;
 525
 526         if (pte_present(*pte)) {
 527                 page = vm_normal_page(vma, addr, *pte);
 528         } else if (is_swap_pte(*pte)) {
 529                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 530
 531                 if (!non_swap_entry(swpent)) {
 532                         int mapcount;
 533
 534                         mss->swap += PAGE_SIZE;
 535                         mapcount = swp_swapcount(swpent);
 536                         if (mapcount >= 2) {
 537                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 538
 539                                 do_div(pss_delta, mapcount);
 540                                 mss->swap_pss += pss_delta;
 541                         } else {
 542                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 543                         }
 544                 } else if (is_migration_entry(swpent))
 545                         page = migration_entry_to_page(swpent);
 546                 else if (is_device_private_entry(swpent))
 547                         page = device_private_entry_to_page(swpent);
 548         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 549                                                         && pte_none(*pte))) {
 550                 page = find_get_entry(vma->vm_file->f_mapping,
 551                                                 linear_page_index(vma, addr));
 552                 if (!page)
 553                         return;
 554
 555                 if (xa_is_value(page))
 556                         mss->swap += PAGE_SIZE;
 557                 else
 558                         put_page(page);
 559
 560                 return;
 561         }
 562
 563         if (!page)
 564                 return;
 565
 566         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
 567 }
 568
 569 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 570 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 571                 struct mm_walk *walk)
 572 {
 573         struct mem_size_stats *mss = walk->private;
 574         struct vm_area_struct *vma = walk->vma;
 575         bool locked = !!(vma->vm_flags & VM_LOCKED);
 576         struct page *page;
 577
 578         /* FOLL_DUMP will return -EFAULT on huge zero page */
 579         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 580         if (IS_ERR_OR_NULL(page))
 581                 return;
 582         if (PageAnon(page))
 583                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 584         else if (PageSwapBacked(page))
 585                 mss->shmem_thp += HPAGE_PMD_SIZE;
 586         else if (is_zone_device_page(page))
 587                 /* pass */;
 588         else
 589                 VM_BUG_ON_PAGE(1, page);
 590         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
 591 }
 592 #else
 593 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 594                 struct mm_walk *walk)
 595 {
 596 }
 597 #endif
 598
 599 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 600                            struct mm_walk *walk)
 601 {
 602         struct vm_area_struct *vma = walk->vma;
 603         pte_t *pte;
 604         spinlock_t *ptl;
 605
 606         ptl = pmd_trans_huge_lock(pmd, vma);
 607         if (ptl) {
 608                 if (pmd_present(*pmd))
 609                         smaps_pmd_entry(pmd, addr, walk);
 610                 spin_unlock(ptl);
 611                 goto out;
 612         }
 613
 614         if (pmd_trans_unstable(pmd))
 615                 goto out;
 616         /*
 617          * The mmap_sem held all the way back in m_start() is what
 618          * keeps khugepaged out of here and from collapsing things
 619          * in here.
 620          */
 621         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 622         for (; addr != end; pte++, addr += PAGE_SIZE)
 623                 smaps_pte_entry(pte, addr, walk);
 624         pte_unmap_unlock(pte - 1, ptl);
 625 out:
 626         cond_resched();
 627         return 0;
 628 }
 629
 630 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 631 {
 632         /*
 633          * Don't forget to update Documentation/ on changes.
 634          */
 635         static const char mnemonics[BITS_PER_LONG][2] = {
 636                 /*
 637                  * In case if we meet a flag we don't know about.
 638                  */
 639                 [0 ... (BITS_PER_LONG-1)] = "??",
 640
 641                 [ilog2(VM_READ)]        = "rd",
 642                 [ilog2(VM_WRITE)]       = "wr",
 643                 [ilog2(VM_EXEC)]        = "ex",
 644                 [ilog2(VM_SHARED)]      = "sh",
 645                 [ilog2(VM_MAYREAD)]     = "mr",
 646                 [ilog2(VM_MAYWRITE)]    = "mw",
 647                 [ilog2(VM_MAYEXEC)]     = "me",
 648                 [ilog2(VM_MAYSHARE)]    = "ms",
 649                 [ilog2(VM_GROWSDOWN)]   = "gd",
 650                 [ilog2(VM_PFNMAP)]      = "pf",
 651                 [ilog2(VM_DENYWRITE)]   = "dw",
 652 #ifdef CONFIG_X86_INTEL_MPX
 653                 [ilog2(VM_MPX)]         = "mp",
 654 #endif
 655                 [ilog2(VM_LOCKED)]      = "lo",
 656                 [ilog2(VM_IO)]          = "io",
 657                 [ilog2(VM_SEQ_READ)]    = "sr",
 658                 [ilog2(VM_RAND_READ)]   = "rr",
 659                 [ilog2(VM_DONTCOPY)]    = "dc",
 660                 [ilog2(VM_DONTEXPAND)]  = "de",
 661                 [ilog2(VM_ACCOUNT)]     = "ac",
 662                 [ilog2(VM_NORESERVE)]   = "nr",
 663                 [ilog2(VM_HUGETLB)]     = "ht",
 664                 [ilog2(VM_SYNC)]        = "sf",
 665                 [ilog2(VM_ARCH_1)]      = "ar",
 666                 [ilog2(VM_WIPEONFORK)]  = "wf",
 667                 [ilog2(VM_DONTDUMP)]    = "dd",
 668 #ifdef CONFIG_MEM_SOFT_DIRTY
 669                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 670 #endif
 671                 [ilog2(VM_MIXEDMAP)]    = "mm",
 672                 [ilog2(VM_HUGEPAGE)]    = "hg",
 673                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 674                 [ilog2(VM_MERGEABLE)]   = "mg",
 675                 [ilog2(VM_UFFD_MISSING)]= "um",
 676                 [ilog2(VM_UFFD_WP)]     = "uw",
 677 #ifdef CONFIG_ARCH_HAS_PKEYS
 678                 /* These come out via ProtectionKey: */
 679                 [ilog2(VM_PKEY_BIT0)]   = "",
 680                 [ilog2(VM_PKEY_BIT1)]   = "",
 681                 [ilog2(VM_PKEY_BIT2)]   = "",
 682                 [ilog2(VM_PKEY_BIT3)]   = "",
 683 #if VM_PKEY_BIT4
 684                 [ilog2(VM_PKEY_BIT4)]   = "",
 685 #endif
 686 #endif /* CONFIG_ARCH_HAS_PKEYS */
 687         };
 688         size_t i;
 689
 690         seq_puts(m, "VmFlags: ");
 691         for (i = 0; i < BITS_PER_LONG; i++) {
 692                 if (!mnemonics[i][0])
 693                         continue;
 694                 if (vma->vm_flags & (1UL << i)) {
 695                         seq_putc(m, mnemonics[i][0]);
 696                         seq_putc(m, mnemonics[i][1]);
 697                         seq_putc(m, ' ');
 698                 }
 699         }
 700         seq_putc(m, '\n');
 701 }
 702
 703 #ifdef CONFIG_HUGETLB_PAGE
 704 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 705                                  unsigned long addr, unsigned long end,
 706                                  struct mm_walk *walk)
 707 {
 708         struct mem_size_stats *mss = walk->private;
 709         struct vm_area_struct *vma = walk->vma;
 710         struct page *page = NULL;
 711
 712         if (pte_present(*pte)) {
 713                 page = vm_normal_page(vma, addr, *pte);
 714         } else if (is_swap_pte(*pte)) {
 715                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 716
 717                 if (is_migration_entry(swpent))
 718                         page = migration_entry_to_page(swpent);
 719                 else if (is_device_private_entry(swpent))
 720                         page = device_private_entry_to_page(swpent);
 721         }
 722         if (page) {
 723                 int mapcount = page_mapcount(page);
 724
 725                 if (mapcount >= 2)
 726                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 727                 else
 728                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 729         }
 730         return 0;
 731 }
 732 #endif /* HUGETLB_PAGE */
 733
 734 static void smap_gather_stats(struct vm_area_struct *vma,
 735                              struct mem_size_stats *mss)
 736 {
 737         struct mm_walk smaps_walk = {
 738                 .pmd_entry = smaps_pte_range,
 739 #ifdef CONFIG_HUGETLB_PAGE
 740                 .hugetlb_entry = smaps_hugetlb_range,
 741 #endif
 742                 .mm = vma->vm_mm,
 743         };
 744
 745         smaps_walk.private = mss;
 746
 747 #ifdef CONFIG_SHMEM
 748         /* In case of smaps_rollup, reset the value from previous vma */
 749         mss->check_shmem_swap = false;
 750         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 751                 /*
 752                  * For shared or readonly shmem mappings we know that all
 753                  * swapped out pages belong to the shmem object, and we can
 754                  * obtain the swap value much more efficiently. For private
 755                  * writable mappings, we might have COW pages that are
 756                  * not affected by the parent swapped out pages of the shmem
 757                  * object, so we have to distinguish them during the page walk.
 758                  * Unless we know that the shmem object (or the part mapped by
 759                  * our VMA) has no swapped out pages at all.
 760                  */
 761                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 762
 763                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 764                                         !(vma->vm_flags & VM_WRITE)) {
 765                         mss->swap += shmem_swapped;
 766                 } else {
 767                         mss->check_shmem_swap = true;
 768                         smaps_walk.pte_hole = smaps_pte_hole;
 769                 }
 770         }
 771 #endif
 772         /* mmap_sem is held in m_start */
 773         walk_page_vma(vma, &smaps_walk);
 774 }
 775
 776 #define SEQ_PUT_DEC(str, val) \
 777                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 778
 779 /* Show the contents common for smaps and smaps_rollup */
 780 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 781         bool rollup_mode)
 782 {
 783         SEQ_PUT_DEC("Rss:            ", mss->resident);
 784         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 785         if (rollup_mode) {
 786                 /*
 787                  * These are meaningful only for smaps_rollup, otherwise two of
 788                  * them are zero, and the other one is the same as Pss.
 789                  */
 790                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 791                         mss->pss_anon >> PSS_SHIFT);
 792                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 793                         mss->pss_file >> PSS_SHIFT);
 794                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 795                         mss->pss_shmem >> PSS_SHIFT);
 796         }
 797         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 798         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 799         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 800         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 801         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 802         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 803         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 804         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 805         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 806         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 807         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 808                                   mss->private_hugetlb >> 10, 7);
 809         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 810         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 811                                         mss->swap_pss >> PSS_SHIFT);
 812         SEQ_PUT_DEC(" kB\nLocked:         ",
 813                                         mss->pss_locked >> PSS_SHIFT);
 814         seq_puts(m, " kB\n");
 815 }
 816
 817 static int show_smap(struct seq_file *m, void *v)
 818 {
 819         struct vm_area_struct *vma = v;
 820         struct mem_size_stats mss;
 821
 822         memset(&mss, 0, sizeof(mss));
 823
 824         smap_gather_stats(vma, &mss);
 825
 826         show_map_vma(m, vma);
 827
 828         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 829         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 830         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 831         seq_puts(m, " kB\n");
 832
 833         __show_smap(m, &mss, false);
 834
 835         seq_printf(m, "THPeligible:             %d\n",
 836                    transparent_hugepage_enabled(vma));
 837
 838         if (arch_pkeys_enabled())
 839                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 840         show_smap_vma_flags(m, vma);
 841
 842         m_cache_vma(m, vma);
 843
 844         return 0;
 845 }
 846
 847 static int show_smaps_rollup(struct seq_file *m, void *v)
 848 {
 849         struct proc_maps_private *priv = m->private;
 850         struct mem_size_stats mss;
 851         struct mm_struct *mm;
 852         struct vm_area_struct *vma;
 853         unsigned long last_vma_end = 0;
 854         int ret = 0;
 855
 856         priv->task = get_proc_task(priv->inode);
 857         if (!priv->task)
 858                 return -ESRCH;
 859
 860         mm = priv->mm;
 861         if (!mm || !mmget_not_zero(mm)) {
 862                 ret = -ESRCH;
 863                 goto out_put_task;
 864         }
 865
 866         memset(&mss, 0, sizeof(mss));
 867
 868         ret = down_read_killable(&mm->mmap_sem);
 869         if (ret)
 870                 goto out_put_mm;
 871
 872         hold_task_mempolicy(priv);
 873
 874         for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
 875                 smap_gather_stats(vma, &mss);
 876                 last_vma_end = vma->vm_end;
 877         }
 878
 879         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 880                                last_vma_end, 0, 0, 0, 0);
 881         seq_pad(m, ' ');
 882         seq_puts(m, "[rollup]\n");
 883
 884         __show_smap(m, &mss, true);
 885
 886         release_task_mempolicy(priv);
 887         up_read(&mm->mmap_sem);
 888
 889 out_put_mm:
 890         mmput(mm);
 891 out_put_task:
 892         put_task_struct(priv->task);
 893         priv->task = NULL;
 894
 895         return ret;
 896 }
 897 #undef SEQ_PUT_DEC
 898
 899 static const struct seq_operations proc_pid_smaps_op = {
 900         .start  = m_start,
 901         .next   = m_next,
 902         .stop   = m_stop,
 903         .show   = show_smap
 904 };
 905
 906 static int pid_smaps_open(struct inode *inode, struct file *file)
 907 {
 908         return do_maps_open(inode, file, &proc_pid_smaps_op);
 909 }
 910
 911 static int smaps_rollup_open(struct inode *inode, struct file *file)
 912 {
 913         int ret;
 914         struct proc_maps_private *priv;
 915
 916         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 917         if (!priv)
 918                 return -ENOMEM;
 919
 920         ret = single_open(file, show_smaps_rollup, priv);
 921         if (ret)
 922                 goto out_free;
 923
 924         priv->inode = inode;
 925         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 926         if (IS_ERR(priv->mm)) {
 927                 ret = PTR_ERR(priv->mm);
 928
 929                 single_release(inode, file);
 930                 goto out_free;
 931         }
 932
 933         return 0;
 934
 935 out_free:
 936         kfree(priv);
 937         return ret;
 938 }
 939
 940 static int smaps_rollup_release(struct inode *inode, struct file *file)
 941 {
 942         struct seq_file *seq = file->private_data;
 943         struct proc_maps_private *priv = seq->private;
 944
 945         if (priv->mm)
 946                 mmdrop(priv->mm);
 947
 948         kfree(priv);
 949         return single_release(inode, file);
 950 }
 951
 952 const struct file_operations proc_pid_smaps_operations = {
 953         .open           = pid_smaps_open,
 954         .read           = seq_read,
 955         .llseek         = seq_lseek,
 956         .release        = proc_map_release,
 957 };
 958
 959 const struct file_operations proc_pid_smaps_rollup_operations = {
 960         .open           = smaps_rollup_open,
 961         .read           = seq_read,
 962         .llseek         = seq_lseek,
 963         .release        = smaps_rollup_release,
 964 };
 965
 966 enum clear_refs_types {
 967         CLEAR_REFS_ALL = 1,
 968         CLEAR_REFS_ANON,
 969         CLEAR_REFS_MAPPED,
 970         CLEAR_REFS_SOFT_DIRTY,
 971         CLEAR_REFS_MM_HIWATER_RSS,
 972         CLEAR_REFS_LAST,
 973 };
 974
 975 struct clear_refs_private {
 976         enum clear_refs_types type;
 977 };
 978
 979 #ifdef CONFIG_MEM_SOFT_DIRTY
 980 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 981                 unsigned long addr, pte_t *pte)
 982 {
 983         /*
 984          * The soft-dirty tracker uses #PF-s to catch writes
 985          * to pages, so write-protect the pte as well. See the
 986          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 987          * of how soft-dirty works.
 988          */
 989         pte_t ptent = *pte;
 990
 991         if (pte_present(ptent)) {
 992                 pte_t old_pte;
 993
 994                 old_pte = ptep_modify_prot_start(vma, addr, pte);
 995                 ptent = pte_wrprotect(old_pte);
 996                 ptent = pte_clear_soft_dirty(ptent);
 997                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
 998         } else if (is_swap_pte(ptent)) {
 999                 ptent = pte_swp_clear_soft_dirty(ptent);
1000                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1001         }
1002 }
1003 #else
1004 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1005                 unsigned long addr, pte_t *pte)
1006 {
1007 }
1008 #endif
1009
1010 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1011 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1012                 unsigned long addr, pmd_t *pmdp)
1013 {
1014         pmd_t old, pmd = *pmdp;
1015
1016         if (pmd_present(pmd)) {
1017                 /* See comment in change_huge_pmd() */
1018                 old = pmdp_invalidate(vma, addr, pmdp);
1019                 if (pmd_dirty(old))
1020                         pmd = pmd_mkdirty(pmd);
1021                 if (pmd_young(old))
1022                         pmd = pmd_mkyoung(pmd);
1023
1024                 pmd = pmd_wrprotect(pmd);
1025                 pmd = pmd_clear_soft_dirty(pmd);
1026
1027                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1028         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1029                 pmd = pmd_swp_clear_soft_dirty(pmd);
1030                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1031         }
1032 }
1033 #else
1034 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1035                 unsigned long addr, pmd_t *pmdp)
1036 {
1037 }
1038 #endif
1039
1040 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1041                                 unsigned long end, struct mm_walk *walk)
1042 {
1043         struct clear_refs_private *cp = walk->private;
1044         struct vm_area_struct *vma = walk->vma;
1045         pte_t *pte, ptent;
1046         spinlock_t *ptl;
1047         struct page *page;
1048
1049         ptl = pmd_trans_huge_lock(pmd, vma);
1050         if (ptl) {
1051                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1052                         clear_soft_dirty_pmd(vma, addr, pmd);
1053                         goto out;
1054                 }
1055
1056                 if (!pmd_present(*pmd))
1057                         goto out;
1058
1059                 page = pmd_page(*pmd);
1060
1061                 /* Clear accessed and referenced bits. */
1062                 pmdp_test_and_clear_young(vma, addr, pmd);
1063                 test_and_clear_page_young(page);
1064                 ClearPageReferenced(page);
1065 out:
1066                 spin_unlock(ptl);
1067                 return 0;
1068         }
1069
1070         if (pmd_trans_unstable(pmd))
1071                 return 0;
1072
1073         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1074         for (; addr != end; pte++, addr += PAGE_SIZE) {
1075                 ptent = *pte;
1076
1077                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1078                         clear_soft_dirty(vma, addr, pte);
1079                         continue;
1080                 }
1081
1082                 if (!pte_present(ptent))
1083                         continue;
1084
1085                 page = vm_normal_page(vma, addr, ptent);
1086                 if (!page)
1087                         continue;
1088
1089                 /* Clear accessed and referenced bits. */
1090                 ptep_test_and_clear_young(vma, addr, pte);
1091                 test_and_clear_page_young(page);
1092                 ClearPageReferenced(page);
1093         }
1094         pte_unmap_unlock(pte - 1, ptl);
1095         cond_resched();
1096         return 0;
1097 }
1098
1099 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1100                                 struct mm_walk *walk)
1101 {
1102         struct clear_refs_private *cp = walk->private;
1103         struct vm_area_struct *vma = walk->vma;
1104
1105         if (vma->vm_flags & VM_PFNMAP)
1106                 return 1;
1107
1108         /*
1109          * Writing 1 to /proc/pid/clear_refs affects all pages.
1110          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1111          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1112          * Writing 4 to /proc/pid/clear_refs affects all pages.
1113          */
1114         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1115                 return 1;
1116         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1117                 return 1;
1118         return 0;
1119 }
1120
1121 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1122                                 size_t count, loff_t *ppos)
1123 {
1124         struct task_struct *task;
1125         char buffer[PROC_NUMBUF];
1126         struct mm_struct *mm;
1127         struct vm_area_struct *vma;
1128         enum clear_refs_types type;
1129         struct mmu_gather tlb;
1130         int itype;
1131         int rv;
1132
1133         memset(buffer, 0, sizeof(buffer));
1134         if (count > sizeof(buffer) - 1)
1135                 count = sizeof(buffer) - 1;
1136         if (copy_from_user(buffer, buf, count))
1137                 return -EFAULT;
1138         rv = kstrtoint(strstrip(buffer), 10, &itype);
1139         if (rv < 0)
1140                 return rv;
1141         type = (enum clear_refs_types)itype;
1142         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1143                 return -EINVAL;
1144
1145         task = get_proc_task(file_inode(file));
1146         if (!task)
1147                 return -ESRCH;
1148         mm = get_task_mm(task);
1149         if (mm) {
1150                 struct mmu_notifier_range range;
1151                 struct clear_refs_private cp = {
1152                         .type = type,
1153                 };
1154                 struct mm_walk clear_refs_walk = {
1155                         .pmd_entry = clear_refs_pte_range,
1156                         .test_walk = clear_refs_test_walk,
1157                         .mm = mm,
1158                         .private = &cp,
1159                 };
1160
1161                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1162                         if (down_write_killable(&mm->mmap_sem)) {
1163                                 count = -EINTR;
1164                                 goto out_mm;
1165                         }
1166
1167                         /*
1168                          * Writing 5 to /proc/pid/clear_refs resets the peak
1169                          * resident set size to this mm's current rss value.
1170                          */
1171                         reset_mm_hiwater_rss(mm);
1172                         up_write(&mm->mmap_sem);
1173                         goto out_mm;
1174                 }
1175
1176                 if (down_read_killable(&mm->mmap_sem)) {
1177                         count = -EINTR;
1178                         goto out_mm;
1179                 }
1180                 tlb_gather_mmu(&tlb, mm, 0, -1);
1181                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1182                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1183                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1184                                         continue;
1185                                 up_read(&mm->mmap_sem);
1186                                 if (down_write_killable(&mm->mmap_sem)) {
1187                                         count = -EINTR;
1188                                         goto out_mm;
1189                                 }
1190                                 /*
1191                                  * Avoid to modify vma->vm_flags
1192                                  * without locked ops while the
1193                                  * coredump reads the vm_flags.
1194                                  */
1195                                 if (!mmget_still_valid(mm)) {
1196                                         /*
1197                                          * Silently return "count"
1198                                          * like if get_task_mm()
1199                                          * failed. FIXME: should this
1200                                          * function have returned
1201                                          * -ESRCH if get_task_mm()
1202                                          * failed like if
1203                                          * get_proc_task() fails?
1204                                          */
1205                                         up_write(&mm->mmap_sem);
1206                                         goto out_mm;
1207                                 }
1208                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1209                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1210                                         vma_set_page_prot(vma);
1211                                 }
1212                                 downgrade_write(&mm->mmap_sem);
1213                                 break;
1214                         }
1215
1216                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1217                                                 0, NULL, mm, 0, -1UL);
1218                         mmu_notifier_invalidate_range_start(&range);
1219                 }
1220                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1221                 if (type == CLEAR_REFS_SOFT_DIRTY)
1222                         mmu_notifier_invalidate_range_end(&range);
1223                 tlb_finish_mmu(&tlb, 0, -1);
1224                 up_read(&mm->mmap_sem);
1225 out_mm:
1226                 mmput(mm);
1227         }
1228         put_task_struct(task);
1229
1230         return count;
1231 }
1232
1233 const struct file_operations proc_clear_refs_operations = {
1234         .write          = clear_refs_write,
1235         .llseek         = noop_llseek,
1236 };
1237
1238 typedef struct {
1239         u64 pme;
1240 } pagemap_entry_t;
1241
1242 struct pagemapread {
1243         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1244         pagemap_entry_t *buffer;
1245         bool show_pfn;
1246 };
1247
1248 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1249 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1250
1251 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1252 #define PM_PFRAME_BITS          55
1253 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1254 #define PM_SOFT_DIRTY           BIT_ULL(55)
1255 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1256 #define PM_FILE                 BIT_ULL(61)
1257 #define PM_SWAP                 BIT_ULL(62)
1258 #define PM_PRESENT              BIT_ULL(63)
1259
1260 #define PM_END_OF_BUFFER    1
1261
1262 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1263 {
1264         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1265 }
1266
1267 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1268                           struct pagemapread *pm)
1269 {
1270         pm->buffer[pm->pos++] = *pme;
1271         if (pm->pos >= pm->len)
1272                 return PM_END_OF_BUFFER;
1273         return 0;
1274 }
1275
1276 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1277                                 struct mm_walk *walk)
1278 {
1279         struct pagemapread *pm = walk->private;
1280         unsigned long addr = start;
1281         int err = 0;
1282
1283         while (addr < end) {
1284                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1285                 pagemap_entry_t pme = make_pme(0, 0);
1286                 /* End of address space hole, which we mark as non-present. */
1287                 unsigned long hole_end;
1288
1289                 if (vma)
1290                         hole_end = min(end, vma->vm_start);
1291                 else
1292                         hole_end = end;
1293
1294                 for (; addr < hole_end; addr += PAGE_SIZE) {
1295                         err = add_to_pagemap(addr, &pme, pm);
1296                         if (err)
1297                                 goto out;
1298                 }
1299
1300                 if (!vma)
1301                         break;
1302
1303                 /* Addresses in the VMA. */
1304                 if (vma->vm_flags & VM_SOFTDIRTY)
1305                         pme = make_pme(0, PM_SOFT_DIRTY);
1306                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1307                         err = add_to_pagemap(addr, &pme, pm);
1308                         if (err)
1309                                 goto out;
1310                 }
1311         }
1312 out:
1313         return err;
1314 }
1315
1316 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1317                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1318 {
1319         u64 frame = 0, flags = 0;
1320         struct page *page = NULL;
1321
1322         if (pte_present(pte)) {
1323                 if (pm->show_pfn)
1324                         frame = pte_pfn(pte);
1325                 flags |= PM_PRESENT;
1326                 page = vm_normal_page(vma, addr, pte);
1327                 if (pte_soft_dirty(pte))
1328                         flags |= PM_SOFT_DIRTY;
1329         } else if (is_swap_pte(pte)) {
1330                 swp_entry_t entry;
1331                 if (pte_swp_soft_dirty(pte))
1332                         flags |= PM_SOFT_DIRTY;
1333                 entry = pte_to_swp_entry(pte);
1334                 if (pm->show_pfn)
1335                         frame = swp_type(entry) |
1336                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1337                 flags |= PM_SWAP;
1338                 if (is_migration_entry(entry))
1339                         page = migration_entry_to_page(entry);
1340
1341                 if (is_device_private_entry(entry))
1342                         page = device_private_entry_to_page(entry);
1343         }
1344
1345         if (page && !PageAnon(page))
1346                 flags |= PM_FILE;
1347         if (page && page_mapcount(page) == 1)
1348                 flags |= PM_MMAP_EXCLUSIVE;
1349         if (vma->vm_flags & VM_SOFTDIRTY)
1350                 flags |= PM_SOFT_DIRTY;
1351
1352         return make_pme(frame, flags);
1353 }
1354
1355 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1356                              struct mm_walk *walk)
1357 {
1358         struct vm_area_struct *vma = walk->vma;
1359         struct pagemapread *pm = walk->private;
1360         spinlock_t *ptl;
1361         pte_t *pte, *orig_pte;
1362         int err = 0;
1363
1364 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1365         ptl = pmd_trans_huge_lock(pmdp, vma);
1366         if (ptl) {
1367                 u64 flags = 0, frame = 0;
1368                 pmd_t pmd = *pmdp;
1369                 struct page *page = NULL;
1370
1371                 if (vma->vm_flags & VM_SOFTDIRTY)
1372                         flags |= PM_SOFT_DIRTY;
1373
1374                 if (pmd_present(pmd)) {
1375                         page = pmd_page(pmd);
1376
1377                         flags |= PM_PRESENT;
1378                         if (pmd_soft_dirty(pmd))
1379                                 flags |= PM_SOFT_DIRTY;
1380                         if (pm->show_pfn)
1381                                 frame = pmd_pfn(pmd) +
1382                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1383                 }
1384 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1385                 else if (is_swap_pmd(pmd)) {
1386                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1387                         unsigned long offset;
1388
1389                         if (pm->show_pfn) {
1390                                 offset = swp_offset(entry) +
1391                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1392                                 frame = swp_type(entry) |
1393                                         (offset << MAX_SWAPFILES_SHIFT);
1394                         }
1395                         flags |= PM_SWAP;
1396                         if (pmd_swp_soft_dirty(pmd))
1397                                 flags |= PM_SOFT_DIRTY;
1398                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1399                         page = migration_entry_to_page(entry);
1400                 }
1401 #endif
1402
1403                 if (page && page_mapcount(page) == 1)
1404                         flags |= PM_MMAP_EXCLUSIVE;
1405
1406                 for (; addr != end; addr += PAGE_SIZE) {
1407                         pagemap_entry_t pme = make_pme(frame, flags);
1408
1409                         err = add_to_pagemap(addr, &pme, pm);
1410                         if (err)
1411                                 break;
1412                         if (pm->show_pfn) {
1413                                 if (flags & PM_PRESENT)
1414                                         frame++;
1415                                 else if (flags & PM_SWAP)
1416                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1417                         }
1418                 }
1419                 spin_unlock(ptl);
1420                 return err;
1421         }
1422
1423         if (pmd_trans_unstable(pmdp))
1424                 return 0;
1425 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1426
1427         /*
1428          * We can assume that @vma always points to a valid one and @end never
1429          * goes beyond vma->vm_end.
1430          */
1431         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1432         for (; addr < end; pte++, addr += PAGE_SIZE) {
1433                 pagemap_entry_t pme;
1434
1435                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1436                 err = add_to_pagemap(addr, &pme, pm);
1437                 if (err)
1438                         break;
1439         }
1440         pte_unmap_unlock(orig_pte, ptl);
1441
1442         cond_resched();
1443
1444         return err;
1445 }
1446
1447 #ifdef CONFIG_HUGETLB_PAGE
1448 /* This function walks within one hugetlb entry in the single call */
1449 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1450                                  unsigned long addr, unsigned long end,
1451                                  struct mm_walk *walk)
1452 {
1453         struct pagemapread *pm = walk->private;
1454         struct vm_area_struct *vma = walk->vma;
1455         u64 flags = 0, frame = 0;
1456         int err = 0;
1457         pte_t pte;
1458
1459         if (vma->vm_flags & VM_SOFTDIRTY)
1460                 flags |= PM_SOFT_DIRTY;
1461
1462         pte = huge_ptep_get(ptep);
1463         if (pte_present(pte)) {
1464                 struct page *page = pte_page(pte);
1465
1466                 if (!PageAnon(page))
1467                         flags |= PM_FILE;
1468
1469                 if (page_mapcount(page) == 1)
1470                         flags |= PM_MMAP_EXCLUSIVE;
1471
1472                 flags |= PM_PRESENT;
1473                 if (pm->show_pfn)
1474                         frame = pte_pfn(pte) +
1475                                 ((addr & ~hmask) >> PAGE_SHIFT);
1476         }
1477
1478         for (; addr != end; addr += PAGE_SIZE) {
1479                 pagemap_entry_t pme = make_pme(frame, flags);
1480
1481                 err = add_to_pagemap(addr, &pme, pm);
1482                 if (err)
1483                         return err;
1484                 if (pm->show_pfn && (flags & PM_PRESENT))
1485                         frame++;
1486         }
1487
1488         cond_resched();
1489
1490         return err;
1491 }
1492 #endif /* HUGETLB_PAGE */
1493
1494 /*
1495  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1496  *
1497  * For each page in the address space, this file contains one 64-bit entry
1498  * consisting of the following:
1499  *
1500  * Bits 0-54  page frame number (PFN) if present
1501  * Bits 0-4   swap type if swapped
1502  * Bits 5-54  swap offset if swapped
1503  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1504  * Bit  56    page exclusively mapped
1505  * Bits 57-60 zero
1506  * Bit  61    page is file-page or shared-anon
1507  * Bit  62    page swapped
1508  * Bit  63    page present
1509  *
1510  * If the page is not present but in swap, then the PFN contains an
1511  * encoding of the swap file number and the page's offset into the
1512  * swap. Unmapped pages return a null PFN. This allows determining
1513  * precisely which pages are mapped (or in swap) and comparing mapped
1514  * pages between processes.
1515  *
1516  * Efficient users of this interface will use /proc/pid/maps to
1517  * determine which areas of memory are actually mapped and llseek to
1518  * skip over unmapped regions.
1519  */
1520 static ssize_t pagemap_read(struct file *file, char __user *buf,
1521                             size_t count, loff_t *ppos)
1522 {
1523         struct mm_struct *mm = file->private_data;
1524         struct pagemapread pm;
1525         struct mm_walk pagemap_walk = {};
1526         unsigned long src;
1527         unsigned long svpfn;
1528         unsigned long start_vaddr;
1529         unsigned long end_vaddr;
1530         int ret = 0, copied = 0;
1531
1532         if (!mm || !mmget_not_zero(mm))
1533                 goto out;
1534
1535         ret = -EINVAL;
1536         /* file position must be aligned */
1537         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1538                 goto out_mm;
1539
1540         ret = 0;
1541         if (!count)
1542                 goto out_mm;
1543
1544         /* do not disclose physical addresses: attack vector */
1545         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1546
1547         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1548         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1549         ret = -ENOMEM;
1550         if (!pm.buffer)
1551                 goto out_mm;
1552
1553         pagemap_walk.pmd_entry = pagemap_pmd_range;
1554         pagemap_walk.pte_hole = pagemap_pte_hole;
1555 #ifdef CONFIG_HUGETLB_PAGE
1556         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1557 #endif
1558         pagemap_walk.mm = mm;
1559         pagemap_walk.private = &pm;
1560
1561         src = *ppos;
1562         svpfn = src / PM_ENTRY_BYTES;
1563         start_vaddr = svpfn << PAGE_SHIFT;
1564         end_vaddr = mm->task_size;
1565
1566         /* watch out for wraparound */
1567         if (svpfn > mm->task_size >> PAGE_SHIFT)
1568                 start_vaddr = end_vaddr;
1569
1570         /*
1571          * The odds are that this will stop walking way
1572          * before end_vaddr, because the length of the
1573          * user buffer is tracked in "pm", and the walk
1574          * will stop when we hit the end of the buffer.
1575          */
1576         ret = 0;
1577         while (count && (start_vaddr < end_vaddr)) {
1578                 int len;
1579                 unsigned long end;
1580
1581                 pm.pos = 0;
1582                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1583                 /* overflow ? */
1584                 if (end < start_vaddr || end > end_vaddr)
1585                         end = end_vaddr;
1586                 ret = down_read_killable(&mm->mmap_sem);
1587                 if (ret)
1588                         goto out_free;
1589                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1590                 up_read(&mm->mmap_sem);
1591                 start_vaddr = end;
1592
1593                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1594                 if (copy_to_user(buf, pm.buffer, len)) {
1595                         ret = -EFAULT;
1596                         goto out_free;
1597                 }
1598                 copied += len;
1599                 buf += len;
1600                 count -= len;
1601         }
1602         *ppos += copied;
1603         if (!ret || ret == PM_END_OF_BUFFER)
1604                 ret = copied;
1605
1606 out_free:
1607         kfree(pm.buffer);
1608 out_mm:
1609         mmput(mm);
1610 out:
1611         return ret;
1612 }
1613
1614 static int pagemap_open(struct inode *inode, struct file *file)
1615 {
1616         struct mm_struct *mm;
1617
1618         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1619         if (IS_ERR(mm))
1620                 return PTR_ERR(mm);
1621         file->private_data = mm;
1622         return 0;
1623 }
1624
1625 static int pagemap_release(struct inode *inode, struct file *file)
1626 {
1627         struct mm_struct *mm = file->private_data;
1628
1629         if (mm)
1630                 mmdrop(mm);
1631         return 0;
1632 }
1633
1634 const struct file_operations proc_pagemap_operations = {
1635         .llseek         = mem_lseek, /* borrow this */
1636         .read           = pagemap_read,
1637         .open           = pagemap_open,
1638         .release        = pagemap_release,
1639 };
1640 #endif /* CONFIG_PROC_PAGE_MONITOR */
1641
1642 #ifdef CONFIG_NUMA
1643
1644 struct numa_maps {
1645         unsigned long pages;
1646         unsigned long anon;
1647         unsigned long active;
1648         unsigned long writeback;
1649         unsigned long mapcount_max;
1650         unsigned long dirty;
1651         unsigned long swapcache;
1652         unsigned long node[MAX_NUMNODES];
1653 };
1654
1655 struct numa_maps_private {
1656         struct proc_maps_private proc_maps;
1657         struct numa_maps md;
1658 };
1659
1660 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1661                         unsigned long nr_pages)
1662 {
1663         int count = page_mapcount(page);
1664
1665         md->pages += nr_pages;
1666         if (pte_dirty || PageDirty(page))
1667                 md->dirty += nr_pages;
1668
1669         if (PageSwapCache(page))
1670                 md->swapcache += nr_pages;
1671
1672         if (PageActive(page) || PageUnevictable(page))
1673                 md->active += nr_pages;
1674
1675         if (PageWriteback(page))
1676                 md->writeback += nr_pages;
1677
1678         if (PageAnon(page))
1679                 md->anon += nr_pages;
1680
1681         if (count > md->mapcount_max)
1682                 md->mapcount_max = count;
1683
1684         md->node[page_to_nid(page)] += nr_pages;
1685 }
1686
1687 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1688                 unsigned long addr)
1689 {
1690         struct page *page;
1691         int nid;
1692
1693         if (!pte_present(pte))
1694                 return NULL;
1695
1696         page = vm_normal_page(vma, addr, pte);
1697         if (!page)
1698                 return NULL;
1699
1700         if (PageReserved(page))
1701                 return NULL;
1702
1703         nid = page_to_nid(page);
1704         if (!node_isset(nid, node_states[N_MEMORY]))
1705                 return NULL;
1706
1707         return page;
1708 }
1709
1710 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1711 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1712                                               struct vm_area_struct *vma,
1713                                               unsigned long addr)
1714 {
1715         struct page *page;
1716         int nid;
1717
1718         if (!pmd_present(pmd))
1719                 return NULL;
1720
1721         page = vm_normal_page_pmd(vma, addr, pmd);
1722         if (!page)
1723                 return NULL;
1724
1725         if (PageReserved(page))
1726                 return NULL;
1727
1728         nid = page_to_nid(page);
1729         if (!node_isset(nid, node_states[N_MEMORY]))
1730                 return NULL;
1731
1732         return page;
1733 }
1734 #endif
1735
1736 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1737                 unsigned long end, struct mm_walk *walk)
1738 {
1739         struct numa_maps *md = walk->private;
1740         struct vm_area_struct *vma = walk->vma;
1741         spinlock_t *ptl;
1742         pte_t *orig_pte;
1743         pte_t *pte;
1744
1745 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1746         ptl = pmd_trans_huge_lock(pmd, vma);
1747         if (ptl) {
1748                 struct page *page;
1749
1750                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1751                 if (page)
1752                         gather_stats(page, md, pmd_dirty(*pmd),
1753                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1754                 spin_unlock(ptl);
1755                 return 0;
1756         }
1757
1758         if (pmd_trans_unstable(pmd))
1759                 return 0;
1760 #endif
1761         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1762         do {
1763                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1764                 if (!page)
1765                         continue;
1766                 gather_stats(page, md, pte_dirty(*pte), 1);
1767
1768         } while (pte++, addr += PAGE_SIZE, addr != end);
1769         pte_unmap_unlock(orig_pte, ptl);
1770         cond_resched();
1771         return 0;
1772 }
1773 #ifdef CONFIG_HUGETLB_PAGE
1774 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1775                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1776 {
1777         pte_t huge_pte = huge_ptep_get(pte);
1778         struct numa_maps *md;
1779         struct page *page;
1780
1781         if (!pte_present(huge_pte))
1782                 return 0;
1783
1784         page = pte_page(huge_pte);
1785         if (!page)
1786                 return 0;
1787
1788         md = walk->private;
1789         gather_stats(page, md, pte_dirty(huge_pte), 1);
1790         return 0;
1791 }
1792
1793 #else
1794 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1795                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1796 {
1797         return 0;
1798 }
1799 #endif
1800
1801 /*
1802  * Display pages allocated per node and memory policy via /proc.
1803  */
1804 static int show_numa_map(struct seq_file *m, void *v)
1805 {
1806         struct numa_maps_private *numa_priv = m->private;
1807         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1808         struct vm_area_struct *vma = v;
1809         struct numa_maps *md = &numa_priv->md;
1810         struct file *file = vma->vm_file;
1811         struct mm_struct *mm = vma->vm_mm;
1812         struct mm_walk walk = {
1813                 .hugetlb_entry = gather_hugetlb_stats,
1814                 .pmd_entry = gather_pte_stats,
1815                 .private = md,
1816                 .mm = mm,
1817         };
1818         struct mempolicy *pol;
1819         char buffer[64];
1820         int nid;
1821
1822         if (!mm)
1823                 return 0;
1824
1825         /* Ensure we start with an empty set of numa_maps statistics. */
1826         memset(md, 0, sizeof(*md));
1827
1828         pol = __get_vma_policy(vma, vma->vm_start);
1829         if (pol) {
1830                 mpol_to_str(buffer, sizeof(buffer), pol);
1831                 mpol_cond_put(pol);
1832         } else {
1833                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1834         }
1835
1836         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1837
1838         if (file) {
1839                 seq_puts(m, " file=");
1840                 seq_file_path(m, file, "\n\t= ");
1841         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1842                 seq_puts(m, " heap");
1843         } else if (is_stack(vma)) {
1844                 seq_puts(m, " stack");
1845         }
1846
1847         if (is_vm_hugetlb_page(vma))
1848                 seq_puts(m, " huge");
1849
1850         /* mmap_sem is held by m_start */
1851         walk_page_vma(vma, &walk);
1852
1853         if (!md->pages)
1854                 goto out;
1855
1856         if (md->anon)
1857                 seq_printf(m, " anon=%lu", md->anon);
1858
1859         if (md->dirty)
1860                 seq_printf(m, " dirty=%lu", md->dirty);
1861
1862         if (md->pages != md->anon && md->pages != md->dirty)
1863                 seq_printf(m, " mapped=%lu", md->pages);
1864
1865         if (md->mapcount_max > 1)
1866                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1867
1868         if (md->swapcache)
1869                 seq_printf(m, " swapcache=%lu", md->swapcache);
1870
1871         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1872                 seq_printf(m, " active=%lu", md->active);
1873
1874         if (md->writeback)
1875                 seq_printf(m, " writeback=%lu", md->writeback);
1876
1877         for_each_node_state(nid, N_MEMORY)
1878                 if (md->node[nid])
1879                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1880
1881         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1882 out:
1883         seq_putc(m, '\n');
1884         m_cache_vma(m, vma);
1885         return 0;
1886 }
1887
1888 static const struct seq_operations proc_pid_numa_maps_op = {
1889         .start  = m_start,
1890         .next   = m_next,
1891         .stop   = m_stop,
1892         .show   = show_numa_map,
1893 };
1894
1895 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1896 {
1897         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1898                                 sizeof(struct numa_maps_private));
1899 }
1900
1901 const struct file_operations proc_pid_numa_maps_operations = {
1902         .open           = pid_numa_maps_open,
1903         .read           = seq_read,
1904         .llseek         = seq_lseek,
1905         .release        = proc_map_release,
1906 };
1907
1908 #endif /* CONFIG_NUMA */