fs/proc/task_mmu.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/pagewalk.h>
   3 #include <linux/vmacache.h>
   4 #include <linux/hugetlb.h>
   5 #include <linux/huge_mm.h>
   6 #include <linux/mount.h>
   7 #include <linux/seq_file.h>
   8 #include <linux/highmem.h>
   9 #include <linux/ptrace.h>
  10 #include <linux/slab.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/mempolicy.h>
  13 #include <linux/rmap.h>
  14 #include <linux/swap.h>
  15 #include <linux/sched/mm.h>
  16 #include <linux/swapops.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/page_idle.h>
  19 #include <linux/shmem_fs.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/pkeys.h>
  22
  23 #include <asm/elf.h>
  24 #include <asm/tlb.h>
  25 #include <asm/tlbflush.h>
  26 #include "internal.h"
  27
  28 #define SEQ_PUT_DEC(str, val) \
  29                 seq_put_decimal_ull_width(m, str, (val) << (PAGE_SHIFT-10), 8)
  30 void task_mem(struct seq_file *m, struct mm_struct *mm)
  31 {
  32         unsigned long text, lib, swap, anon, file, shmem;
  33         unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss;
  34
  35         anon = get_mm_counter(mm, MM_ANONPAGES);
  36         file = get_mm_counter(mm, MM_FILEPAGES);
  37         shmem = get_mm_counter(mm, MM_SHMEMPAGES);
  38
  39         /*
  40          * Note: to minimize their overhead, mm maintains hiwater_vm and
  41          * hiwater_rss only when about to *lower* total_vm or rss.  Any
  42          * collector of these hiwater stats must therefore get total_vm
  43          * and rss too, which will usually be the higher.  Barriers? not
  44          * worth the effort, such snapshots can always be inconsistent.
  45          */
  46         hiwater_vm = total_vm = mm->total_vm;
  47         if (hiwater_vm < mm->hiwater_vm)
  48                 hiwater_vm = mm->hiwater_vm;
  49         hiwater_rss = total_rss = anon + file + shmem;
  50         if (hiwater_rss < mm->hiwater_rss)
  51                 hiwater_rss = mm->hiwater_rss;
  52
  53         /* split executable areas between text and lib */
  54         text = PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK);
  55         text = min(text, mm->exec_vm << PAGE_SHIFT);
  56         lib = (mm->exec_vm << PAGE_SHIFT) - text;
  57
  58         swap = get_mm_counter(mm, MM_SWAPENTS);
  59         SEQ_PUT_DEC("VmPeak:\t", hiwater_vm);
  60         SEQ_PUT_DEC(" kB\nVmSize:\t", total_vm);
  61         SEQ_PUT_DEC(" kB\nVmLck:\t", mm->locked_vm);
  62         SEQ_PUT_DEC(" kB\nVmPin:\t", atomic64_read(&mm->pinned_vm));
  63         SEQ_PUT_DEC(" kB\nVmHWM:\t", hiwater_rss);
  64         SEQ_PUT_DEC(" kB\nVmRSS:\t", total_rss);
  65         SEQ_PUT_DEC(" kB\nRssAnon:\t", anon);
  66         SEQ_PUT_DEC(" kB\nRssFile:\t", file);
  67         SEQ_PUT_DEC(" kB\nRssShmem:\t", shmem);
  68         SEQ_PUT_DEC(" kB\nVmData:\t", mm->data_vm);
  69         SEQ_PUT_DEC(" kB\nVmStk:\t", mm->stack_vm);
  70         seq_put_decimal_ull_width(m,
  71                     " kB\nVmExe:\t", text >> 10, 8);
  72         seq_put_decimal_ull_width(m,
  73                     " kB\nVmLib:\t", lib >> 10, 8);
  74         seq_put_decimal_ull_width(m,
  75                     " kB\nVmPTE:\t", mm_pgtables_bytes(mm) >> 10, 8);
  76         SEQ_PUT_DEC(" kB\nVmSwap:\t", swap);
  77         seq_puts(m, " kB\n");
  78         hugetlb_report_usage(m, mm);
  79 }
  80 #undef SEQ_PUT_DEC
  81
  82 unsigned long task_vsize(struct mm_struct *mm)
  83 {
  84         return PAGE_SIZE * mm->total_vm;
  85 }
  86
  87 unsigned long task_statm(struct mm_struct *mm,
  88                          unsigned long *shared, unsigned long *text,
  89                          unsigned long *data, unsigned long *resident)
  90 {
  91         *shared = get_mm_counter(mm, MM_FILEPAGES) +
  92                         get_mm_counter(mm, MM_SHMEMPAGES);
  93         *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK))
  94                                                                 >> PAGE_SHIFT;
  95         *data = mm->data_vm + mm->stack_vm;
  96         *resident = *shared + get_mm_counter(mm, MM_ANONPAGES);
  97         return mm->total_vm;
  98 }
  99
 100 #ifdef CONFIG_NUMA
 101 /*
 102  * Save get_task_policy() for show_numa_map().
 103  */
 104 static void hold_task_mempolicy(struct proc_maps_private *priv)
 105 {
 106         struct task_struct *task = priv->task;
 107
 108         task_lock(task);
 109         priv->task_mempolicy = get_task_policy(task);
 110         mpol_get(priv->task_mempolicy);
 111         task_unlock(task);
 112 }
 113 static void release_task_mempolicy(struct proc_maps_private *priv)
 114 {
 115         mpol_put(priv->task_mempolicy);
 116 }
 117 #else
 118 static void hold_task_mempolicy(struct proc_maps_private *priv)
 119 {
 120 }
 121 static void release_task_mempolicy(struct proc_maps_private *priv)
 122 {
 123 }
 124 #endif
 125
 126 static void vma_stop(struct proc_maps_private *priv)
 127 {
 128         struct mm_struct *mm = priv->mm;
 129
 130         release_task_mempolicy(priv);
 131         up_read(&mm->mmap_sem);
 132         mmput(mm);
 133 }
 134
 135 static struct vm_area_struct *
 136 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma)
 137 {
 138         if (vma == priv->tail_vma)
 139                 return NULL;
 140         return vma->vm_next ?: priv->tail_vma;
 141 }
 142
 143 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma)
 144 {
 145         if (m->count < m->size) /* vma is copied successfully */
 146                 m->version = m_next_vma(m->private, vma) ? vma->vm_end : -1UL;
 147 }
 148
 149 static void *m_start(struct seq_file *m, loff_t *ppos)
 150 {
 151         struct proc_maps_private *priv = m->private;
 152         unsigned long last_addr = m->version;
 153         struct mm_struct *mm;
 154         struct vm_area_struct *vma;
 155         unsigned int pos = *ppos;
 156
 157         /* See m_cache_vma(). Zero at the start or after lseek. */
 158         if (last_addr == -1UL)
 159                 return NULL;
 160
 161         priv->task = get_proc_task(priv->inode);
 162         if (!priv->task)
 163                 return ERR_PTR(-ESRCH);
 164
 165         mm = priv->mm;
 166         if (!mm || !mmget_not_zero(mm))
 167                 return NULL;
 168
 169         if (down_read_killable(&mm->mmap_sem)) {
 170                 mmput(mm);
 171                 return ERR_PTR(-EINTR);
 172         }
 173
 174         hold_task_mempolicy(priv);
 175         priv->tail_vma = get_gate_vma(mm);
 176
 177         if (last_addr) {
 178                 vma = find_vma(mm, last_addr - 1);
 179                 if (vma && vma->vm_start <= last_addr)
 180                         vma = m_next_vma(priv, vma);
 181                 if (vma)
 182                         return vma;
 183         }
 184
 185         m->version = 0;
 186         if (pos < mm->map_count) {
 187                 for (vma = mm->mmap; pos; pos--) {
 188                         m->version = vma->vm_start;
 189                         vma = vma->vm_next;
 190                 }
 191                 return vma;
 192         }
 193
 194         /* we do not bother to update m->version in this case */
 195         if (pos == mm->map_count && priv->tail_vma)
 196                 return priv->tail_vma;
 197
 198         vma_stop(priv);
 199         return NULL;
 200 }
 201
 202 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 203 {
 204         struct proc_maps_private *priv = m->private;
 205         struct vm_area_struct *next;
 206
 207         (*pos)++;
 208         next = m_next_vma(priv, v);
 209         if (!next)
 210                 vma_stop(priv);
 211         return next;
 212 }
 213
 214 static void m_stop(struct seq_file *m, void *v)
 215 {
 216         struct proc_maps_private *priv = m->private;
 217
 218         if (!IS_ERR_OR_NULL(v))
 219                 vma_stop(priv);
 220         if (priv->task) {
 221                 put_task_struct(priv->task);
 222                 priv->task = NULL;
 223         }
 224 }
 225
 226 static int proc_maps_open(struct inode *inode, struct file *file,
 227                         const struct seq_operations *ops, int psize)
 228 {
 229         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 230
 231         if (!priv)
 232                 return -ENOMEM;
 233
 234         priv->inode = inode;
 235         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 236         if (IS_ERR(priv->mm)) {
 237                 int err = PTR_ERR(priv->mm);
 238
 239                 seq_release_private(inode, file);
 240                 return err;
 241         }
 242
 243         return 0;
 244 }
 245
 246 static int proc_map_release(struct inode *inode, struct file *file)
 247 {
 248         struct seq_file *seq = file->private_data;
 249         struct proc_maps_private *priv = seq->private;
 250
 251         if (priv->mm)
 252                 mmdrop(priv->mm);
 253
 254         return seq_release_private(inode, file);
 255 }
 256
 257 static int do_maps_open(struct inode *inode, struct file *file,
 258                         const struct seq_operations *ops)
 259 {
 260         return proc_maps_open(inode, file, ops,
 261                                 sizeof(struct proc_maps_private));
 262 }
 263
 264 /*
 265  * Indicate if the VMA is a stack for the given task; for
 266  * /proc/PID/maps that is the stack of the main task.
 267  */
 268 static int is_stack(struct vm_area_struct *vma)
 269 {
 270         /*
 271          * We make no effort to guess what a given thread considers to be
 272          * its "stack".  It's not even well-defined for programs written
 273          * languages like Go.
 274          */
 275         return vma->vm_start <= vma->vm_mm->start_stack &&
 276                 vma->vm_end >= vma->vm_mm->start_stack;
 277 }
 278
 279 static void show_vma_header_prefix(struct seq_file *m,
 280                                    unsigned long start, unsigned long end,
 281                                    vm_flags_t flags, unsigned long long pgoff,
 282                                    dev_t dev, unsigned long ino)
 283 {
 284         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 285         seq_put_hex_ll(m, NULL, start, 8);
 286         seq_put_hex_ll(m, "-", end, 8);
 287         seq_putc(m, ' ');
 288         seq_putc(m, flags & VM_READ ? 'r' : '-');
 289         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 290         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 291         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 292         seq_put_hex_ll(m, " ", pgoff, 8);
 293         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 294         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 295         seq_put_decimal_ull(m, " ", ino);
 296         seq_putc(m, ' ');
 297 }
 298
 299 static void
 300 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 301 {
 302         struct mm_struct *mm = vma->vm_mm;
 303         struct file *file = vma->vm_file;
 304         vm_flags_t flags = vma->vm_flags;
 305         unsigned long ino = 0;
 306         unsigned long long pgoff = 0;
 307         unsigned long start, end;
 308         dev_t dev = 0;
 309         const char *name = NULL;
 310
 311         if (file) {
 312                 struct inode *inode = file_inode(vma->vm_file);
 313                 dev = inode->i_sb->s_dev;
 314                 ino = inode->i_ino;
 315                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 316         }
 317
 318         start = vma->vm_start;
 319         end = vma->vm_end;
 320         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 321
 322         /*
 323          * Print the dentry name for named mappings, and a
 324          * special [heap] marker for the heap:
 325          */
 326         if (file) {
 327                 seq_pad(m, ' ');
 328                 seq_file_path(m, file, "\n");
 329                 goto done;
 330         }
 331
 332         if (vma->vm_ops && vma->vm_ops->name) {
 333                 name = vma->vm_ops->name(vma);
 334                 if (name)
 335                         goto done;
 336         }
 337
 338         name = arch_vma_name(vma);
 339         if (!name) {
 340                 if (!mm) {
 341                         name = "[vdso]";
 342                         goto done;
 343                 }
 344
 345                 if (vma->vm_start <= mm->brk &&
 346                     vma->vm_end >= mm->start_brk) {
 347                         name = "[heap]";
 348                         goto done;
 349                 }
 350
 351                 if (is_stack(vma))
 352                         name = "[stack]";
 353         }
 354
 355 done:
 356         if (name) {
 357                 seq_pad(m, ' ');
 358                 seq_puts(m, name);
 359         }
 360         seq_putc(m, '\n');
 361 }
 362
 363 static int show_map(struct seq_file *m, void *v)
 364 {
 365         show_map_vma(m, v);
 366         m_cache_vma(m, v);
 367         return 0;
 368 }
 369
 370 static const struct seq_operations proc_pid_maps_op = {
 371         .start  = m_start,
 372         .next   = m_next,
 373         .stop   = m_stop,
 374         .show   = show_map
 375 };
 376
 377 static int pid_maps_open(struct inode *inode, struct file *file)
 378 {
 379         return do_maps_open(inode, file, &proc_pid_maps_op);
 380 }
 381
 382 const struct file_operations proc_pid_maps_operations = {
 383         .open           = pid_maps_open,
 384         .read           = seq_read,
 385         .llseek         = seq_lseek,
 386         .release        = proc_map_release,
 387 };
 388
 389 /*
 390  * Proportional Set Size(PSS): my share of RSS.
 391  *
 392  * PSS of a process is the count of pages it has in memory, where each
 393  * page is divided by the number of processes sharing it.  So if a
 394  * process has 1000 pages all to itself, and 1000 shared with one other
 395  * process, its PSS will be 1500.
 396  *
 397  * To keep (accumulated) division errors low, we adopt a 64bit
 398  * fixed-point pss counter to minimize division errors. So (pss >>
 399  * PSS_SHIFT) would be the real byte count.
 400  *
 401  * A shift of 12 before division means (assuming 4K page size):
 402  *      - 1M 3-user-pages add up to 8KB errors;
 403  *      - supports mapcount up to 2^24, or 16M;
 404  *      - supports PSS up to 2^52 bytes, or 4PB.
 405  */
 406 #define PSS_SHIFT 12
 407
 408 #ifdef CONFIG_PROC_PAGE_MONITOR
 409 struct mem_size_stats {
 410         unsigned long resident;
 411         unsigned long shared_clean;
 412         unsigned long shared_dirty;
 413         unsigned long private_clean;
 414         unsigned long private_dirty;
 415         unsigned long referenced;
 416         unsigned long anonymous;
 417         unsigned long lazyfree;
 418         unsigned long anonymous_thp;
 419         unsigned long shmem_thp;
 420         unsigned long 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 ? compound_nr(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 #else
 517 #define smaps_pte_hole          NULL
 518 #endif /* CONFIG_SHMEM */
 519
 520 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 521                 struct mm_walk *walk)
 522 {
 523         struct mem_size_stats *mss = walk->private;
 524         struct vm_area_struct *vma = walk->vma;
 525         bool locked = !!(vma->vm_flags & VM_LOCKED);
 526         struct page *page = NULL;
 527
 528         if (pte_present(*pte)) {
 529                 page = vm_normal_page(vma, addr, *pte);
 530         } else if (is_swap_pte(*pte)) {
 531                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 532
 533                 if (!non_swap_entry(swpent)) {
 534                         int mapcount;
 535
 536                         mss->swap += PAGE_SIZE;
 537                         mapcount = swp_swapcount(swpent);
 538                         if (mapcount >= 2) {
 539                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 540
 541                                 do_div(pss_delta, mapcount);
 542                                 mss->swap_pss += pss_delta;
 543                         } else {
 544                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 545                         }
 546                 } else if (is_migration_entry(swpent))
 547                         page = migration_entry_to_page(swpent);
 548                 else if (is_device_private_entry(swpent))
 549                         page = device_private_entry_to_page(swpent);
 550         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 551                                                         && pte_none(*pte))) {
 552                 page = find_get_entry(vma->vm_file->f_mapping,
 553                                                 linear_page_index(vma, addr));
 554                 if (!page)
 555                         return;
 556
 557                 if (xa_is_value(page))
 558                         mss->swap += PAGE_SIZE;
 559                 else
 560                         put_page(page);
 561
 562                 return;
 563         }
 564
 565         if (!page)
 566                 return;
 567
 568         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte), locked);
 569 }
 570
 571 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 572 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 573                 struct mm_walk *walk)
 574 {
 575         struct mem_size_stats *mss = walk->private;
 576         struct vm_area_struct *vma = walk->vma;
 577         bool locked = !!(vma->vm_flags & VM_LOCKED);
 578         struct page *page;
 579
 580         /* FOLL_DUMP will return -EFAULT on huge zero page */
 581         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 582         if (IS_ERR_OR_NULL(page))
 583                 return;
 584         if (PageAnon(page))
 585                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 586         else if (PageSwapBacked(page))
 587                 mss->shmem_thp += HPAGE_PMD_SIZE;
 588         else if (is_zone_device_page(page))
 589                 /* pass */;
 590         else
 591                 VM_BUG_ON_PAGE(1, page);
 592         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd), locked);
 593 }
 594 #else
 595 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 596                 struct mm_walk *walk)
 597 {
 598 }
 599 #endif
 600
 601 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 602                            struct mm_walk *walk)
 603 {
 604         struct vm_area_struct *vma = walk->vma;
 605         pte_t *pte;
 606         spinlock_t *ptl;
 607
 608         ptl = pmd_trans_huge_lock(pmd, vma);
 609         if (ptl) {
 610                 if (pmd_present(*pmd))
 611                         smaps_pmd_entry(pmd, addr, walk);
 612                 spin_unlock(ptl);
 613                 goto out;
 614         }
 615
 616         if (pmd_trans_unstable(pmd))
 617                 goto out;
 618         /*
 619          * The mmap_sem held all the way back in m_start() is what
 620          * keeps khugepaged out of here and from collapsing things
 621          * in here.
 622          */
 623         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 624         for (; addr != end; pte++, addr += PAGE_SIZE)
 625                 smaps_pte_entry(pte, addr, walk);
 626         pte_unmap_unlock(pte - 1, ptl);
 627 out:
 628         cond_resched();
 629         return 0;
 630 }
 631
 632 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 633 {
 634         /*
 635          * Don't forget to update Documentation/ on changes.
 636          */
 637         static const char mnemonics[BITS_PER_LONG][2] = {
 638                 /*
 639                  * In case if we meet a flag we don't know about.
 640                  */
 641                 [0 ... (BITS_PER_LONG-1)] = "??",
 642
 643                 [ilog2(VM_READ)]        = "rd",
 644                 [ilog2(VM_WRITE)]       = "wr",
 645                 [ilog2(VM_EXEC)]        = "ex",
 646                 [ilog2(VM_SHARED)]      = "sh",
 647                 [ilog2(VM_MAYREAD)]     = "mr",
 648                 [ilog2(VM_MAYWRITE)]    = "mw",
 649                 [ilog2(VM_MAYEXEC)]     = "me",
 650                 [ilog2(VM_MAYSHARE)]    = "ms",
 651                 [ilog2(VM_GROWSDOWN)]   = "gd",
 652                 [ilog2(VM_PFNMAP)]      = "pf",
 653                 [ilog2(VM_DENYWRITE)]   = "dw",
 654 #ifdef CONFIG_X86_INTEL_MPX
 655                 [ilog2(VM_MPX)]         = "mp",
 656 #endif
 657                 [ilog2(VM_LOCKED)]      = "lo",
 658                 [ilog2(VM_IO)]          = "io",
 659                 [ilog2(VM_SEQ_READ)]    = "sr",
 660                 [ilog2(VM_RAND_READ)]   = "rr",
 661                 [ilog2(VM_DONTCOPY)]    = "dc",
 662                 [ilog2(VM_DONTEXPAND)]  = "de",
 663                 [ilog2(VM_ACCOUNT)]     = "ac",
 664                 [ilog2(VM_NORESERVE)]   = "nr",
 665                 [ilog2(VM_HUGETLB)]     = "ht",
 666                 [ilog2(VM_SYNC)]        = "sf",
 667                 [ilog2(VM_ARCH_1)]      = "ar",
 668                 [ilog2(VM_WIPEONFORK)]  = "wf",
 669                 [ilog2(VM_DONTDUMP)]    = "dd",
 670 #ifdef CONFIG_MEM_SOFT_DIRTY
 671                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 672 #endif
 673                 [ilog2(VM_MIXEDMAP)]    = "mm",
 674                 [ilog2(VM_HUGEPAGE)]    = "hg",
 675                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 676                 [ilog2(VM_MERGEABLE)]   = "mg",
 677                 [ilog2(VM_UFFD_MISSING)]= "um",
 678                 [ilog2(VM_UFFD_WP)]     = "uw",
 679 #ifdef CONFIG_ARCH_HAS_PKEYS
 680                 /* These come out via ProtectionKey: */
 681                 [ilog2(VM_PKEY_BIT0)]   = "",
 682                 [ilog2(VM_PKEY_BIT1)]   = "",
 683                 [ilog2(VM_PKEY_BIT2)]   = "",
 684                 [ilog2(VM_PKEY_BIT3)]   = "",
 685 #if VM_PKEY_BIT4
 686                 [ilog2(VM_PKEY_BIT4)]   = "",
 687 #endif
 688 #endif /* CONFIG_ARCH_HAS_PKEYS */
 689         };
 690         size_t i;
 691
 692         seq_puts(m, "VmFlags: ");
 693         for (i = 0; i < BITS_PER_LONG; i++) {
 694                 if (!mnemonics[i][0])
 695                         continue;
 696                 if (vma->vm_flags & (1UL << i)) {
 697                         seq_putc(m, mnemonics[i][0]);
 698                         seq_putc(m, mnemonics[i][1]);
 699                         seq_putc(m, ' ');
 700                 }
 701         }
 702         seq_putc(m, '\n');
 703 }
 704
 705 #ifdef CONFIG_HUGETLB_PAGE
 706 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 707                                  unsigned long addr, unsigned long end,
 708                                  struct mm_walk *walk)
 709 {
 710         struct mem_size_stats *mss = walk->private;
 711         struct vm_area_struct *vma = walk->vma;
 712         struct page *page = NULL;
 713
 714         if (pte_present(*pte)) {
 715                 page = vm_normal_page(vma, addr, *pte);
 716         } else if (is_swap_pte(*pte)) {
 717                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 718
 719                 if (is_migration_entry(swpent))
 720                         page = migration_entry_to_page(swpent);
 721                 else if (is_device_private_entry(swpent))
 722                         page = device_private_entry_to_page(swpent);
 723         }
 724         if (page) {
 725                 int mapcount = page_mapcount(page);
 726
 727                 if (mapcount >= 2)
 728                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 729                 else
 730                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 731         }
 732         return 0;
 733 }
 734 #else
 735 #define smaps_hugetlb_range     NULL
 736 #endif /* HUGETLB_PAGE */
 737
 738 static const struct mm_walk_ops smaps_walk_ops = {
 739         .pmd_entry              = smaps_pte_range,
 740         .hugetlb_entry          = smaps_hugetlb_range,
 741 };
 742
 743 static const struct mm_walk_ops smaps_shmem_walk_ops = {
 744         .pmd_entry              = smaps_pte_range,
 745         .hugetlb_entry          = smaps_hugetlb_range,
 746         .pte_hole               = smaps_pte_hole,
 747 };
 748
 749 static void smap_gather_stats(struct vm_area_struct *vma,
 750                              struct mem_size_stats *mss)
 751 {
 752 #ifdef CONFIG_SHMEM
 753         /* In case of smaps_rollup, reset the value from previous vma */
 754         mss->check_shmem_swap = false;
 755         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 756                 /*
 757                  * For shared or readonly shmem mappings we know that all
 758                  * swapped out pages belong to the shmem object, and we can
 759                  * obtain the swap value much more efficiently. For private
 760                  * writable mappings, we might have COW pages that are
 761                  * not affected by the parent swapped out pages of the shmem
 762                  * object, so we have to distinguish them during the page walk.
 763                  * Unless we know that the shmem object (or the part mapped by
 764                  * our VMA) has no swapped out pages at all.
 765                  */
 766                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 767
 768                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 769                                         !(vma->vm_flags & VM_WRITE)) {
 770                         mss->swap += shmem_swapped;
 771                 } else {
 772                         mss->check_shmem_swap = true;
 773                         walk_page_vma(vma, &smaps_shmem_walk_ops, mss);
 774                         return;
 775                 }
 776         }
 777 #endif
 778         /* mmap_sem is held in m_start */
 779         walk_page_vma(vma, &smaps_walk_ops, mss);
 780 }
 781
 782 #define SEQ_PUT_DEC(str, val) \
 783                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 784
 785 /* Show the contents common for smaps and smaps_rollup */
 786 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss,
 787         bool rollup_mode)
 788 {
 789         SEQ_PUT_DEC("Rss:            ", mss->resident);
 790         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 791         if (rollup_mode) {
 792                 /*
 793                  * These are meaningful only for smaps_rollup, otherwise two of
 794                  * them are zero, and the other one is the same as Pss.
 795                  */
 796                 SEQ_PUT_DEC(" kB\nPss_Anon:       ",
 797                         mss->pss_anon >> PSS_SHIFT);
 798                 SEQ_PUT_DEC(" kB\nPss_File:       ",
 799                         mss->pss_file >> PSS_SHIFT);
 800                 SEQ_PUT_DEC(" kB\nPss_Shmem:      ",
 801                         mss->pss_shmem >> PSS_SHIFT);
 802         }
 803         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 804         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 805         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 806         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 807         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 808         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 809         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 810         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 811         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 812         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 813         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 814                                   mss->private_hugetlb >> 10, 7);
 815         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 816         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 817                                         mss->swap_pss >> PSS_SHIFT);
 818         SEQ_PUT_DEC(" kB\nLocked:         ",
 819                                         mss->pss_locked >> PSS_SHIFT);
 820         seq_puts(m, " kB\n");
 821 }
 822
 823 static int show_smap(struct seq_file *m, void *v)
 824 {
 825         struct vm_area_struct *vma = v;
 826         struct mem_size_stats mss;
 827
 828         memset(&mss, 0, sizeof(mss));
 829
 830         smap_gather_stats(vma, &mss);
 831
 832         show_map_vma(m, vma);
 833
 834         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 835         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 836         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 837         seq_puts(m, " kB\n");
 838
 839         __show_smap(m, &mss, false);
 840
 841         seq_printf(m, "THPeligible:             %d\n",
 842                    transparent_hugepage_enabled(vma));
 843
 844         if (arch_pkeys_enabled())
 845                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 846         show_smap_vma_flags(m, vma);
 847
 848         m_cache_vma(m, vma);
 849
 850         return 0;
 851 }
 852
 853 static int show_smaps_rollup(struct seq_file *m, void *v)
 854 {
 855         struct proc_maps_private *priv = m->private;
 856         struct mem_size_stats mss;
 857         struct mm_struct *mm;
 858         struct vm_area_struct *vma;
 859         unsigned long last_vma_end = 0;
 860         int ret = 0;
 861
 862         priv->task = get_proc_task(priv->inode);
 863         if (!priv->task)
 864                 return -ESRCH;
 865
 866         mm = priv->mm;
 867         if (!mm || !mmget_not_zero(mm)) {
 868                 ret = -ESRCH;
 869                 goto out_put_task;
 870         }
 871
 872         memset(&mss, 0, sizeof(mss));
 873
 874         ret = down_read_killable(&mm->mmap_sem);
 875         if (ret)
 876                 goto out_put_mm;
 877
 878         hold_task_mempolicy(priv);
 879
 880         for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
 881                 smap_gather_stats(vma, &mss);
 882                 last_vma_end = vma->vm_end;
 883         }
 884
 885         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 886                                last_vma_end, 0, 0, 0, 0);
 887         seq_pad(m, ' ');
 888         seq_puts(m, "[rollup]\n");
 889
 890         __show_smap(m, &mss, true);
 891
 892         release_task_mempolicy(priv);
 893         up_read(&mm->mmap_sem);
 894
 895 out_put_mm:
 896         mmput(mm);
 897 out_put_task:
 898         put_task_struct(priv->task);
 899         priv->task = NULL;
 900
 901         return ret;
 902 }
 903 #undef SEQ_PUT_DEC
 904
 905 static const struct seq_operations proc_pid_smaps_op = {
 906         .start  = m_start,
 907         .next   = m_next,
 908         .stop   = m_stop,
 909         .show   = show_smap
 910 };
 911
 912 static int pid_smaps_open(struct inode *inode, struct file *file)
 913 {
 914         return do_maps_open(inode, file, &proc_pid_smaps_op);
 915 }
 916
 917 static int smaps_rollup_open(struct inode *inode, struct file *file)
 918 {
 919         int ret;
 920         struct proc_maps_private *priv;
 921
 922         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 923         if (!priv)
 924                 return -ENOMEM;
 925
 926         ret = single_open(file, show_smaps_rollup, priv);
 927         if (ret)
 928                 goto out_free;
 929
 930         priv->inode = inode;
 931         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 932         if (IS_ERR(priv->mm)) {
 933                 ret = PTR_ERR(priv->mm);
 934
 935                 single_release(inode, file);
 936                 goto out_free;
 937         }
 938
 939         return 0;
 940
 941 out_free:
 942         kfree(priv);
 943         return ret;
 944 }
 945
 946 static int smaps_rollup_release(struct inode *inode, struct file *file)
 947 {
 948         struct seq_file *seq = file->private_data;
 949         struct proc_maps_private *priv = seq->private;
 950
 951         if (priv->mm)
 952                 mmdrop(priv->mm);
 953
 954         kfree(priv);
 955         return single_release(inode, file);
 956 }
 957
 958 const struct file_operations proc_pid_smaps_operations = {
 959         .open           = pid_smaps_open,
 960         .read           = seq_read,
 961         .llseek         = seq_lseek,
 962         .release        = proc_map_release,
 963 };
 964
 965 const struct file_operations proc_pid_smaps_rollup_operations = {
 966         .open           = smaps_rollup_open,
 967         .read           = seq_read,
 968         .llseek         = seq_lseek,
 969         .release        = smaps_rollup_release,
 970 };
 971
 972 enum clear_refs_types {
 973         CLEAR_REFS_ALL = 1,
 974         CLEAR_REFS_ANON,
 975         CLEAR_REFS_MAPPED,
 976         CLEAR_REFS_SOFT_DIRTY,
 977         CLEAR_REFS_MM_HIWATER_RSS,
 978         CLEAR_REFS_LAST,
 979 };
 980
 981 struct clear_refs_private {
 982         enum clear_refs_types type;
 983 };
 984
 985 #ifdef CONFIG_MEM_SOFT_DIRTY
 986 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 987                 unsigned long addr, pte_t *pte)
 988 {
 989         /*
 990          * The soft-dirty tracker uses #PF-s to catch writes
 991          * to pages, so write-protect the pte as well. See the
 992          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 993          * of how soft-dirty works.
 994          */
 995         pte_t ptent = *pte;
 996
 997         if (pte_present(ptent)) {
 998                 pte_t old_pte;
 999
1000                 old_pte = ptep_modify_prot_start(vma, addr, pte);
1001                 ptent = pte_wrprotect(old_pte);
1002                 ptent = pte_clear_soft_dirty(ptent);
1003                 ptep_modify_prot_commit(vma, addr, pte, old_pte, ptent);
1004         } else if (is_swap_pte(ptent)) {
1005                 ptent = pte_swp_clear_soft_dirty(ptent);
1006                 set_pte_at(vma->vm_mm, addr, pte, ptent);
1007         }
1008 }
1009 #else
1010 static inline void clear_soft_dirty(struct vm_area_struct *vma,
1011                 unsigned long addr, pte_t *pte)
1012 {
1013 }
1014 #endif
1015
1016 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
1017 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1018                 unsigned long addr, pmd_t *pmdp)
1019 {
1020         pmd_t old, pmd = *pmdp;
1021
1022         if (pmd_present(pmd)) {
1023                 /* See comment in change_huge_pmd() */
1024                 old = pmdp_invalidate(vma, addr, pmdp);
1025                 if (pmd_dirty(old))
1026                         pmd = pmd_mkdirty(pmd);
1027                 if (pmd_young(old))
1028                         pmd = pmd_mkyoung(pmd);
1029
1030                 pmd = pmd_wrprotect(pmd);
1031                 pmd = pmd_clear_soft_dirty(pmd);
1032
1033                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1034         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
1035                 pmd = pmd_swp_clear_soft_dirty(pmd);
1036                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
1037         }
1038 }
1039 #else
1040 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
1041                 unsigned long addr, pmd_t *pmdp)
1042 {
1043 }
1044 #endif
1045
1046 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
1047                                 unsigned long end, struct mm_walk *walk)
1048 {
1049         struct clear_refs_private *cp = walk->private;
1050         struct vm_area_struct *vma = walk->vma;
1051         pte_t *pte, ptent;
1052         spinlock_t *ptl;
1053         struct page *page;
1054
1055         ptl = pmd_trans_huge_lock(pmd, vma);
1056         if (ptl) {
1057                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1058                         clear_soft_dirty_pmd(vma, addr, pmd);
1059                         goto out;
1060                 }
1061
1062                 if (!pmd_present(*pmd))
1063                         goto out;
1064
1065                 page = pmd_page(*pmd);
1066
1067                 /* Clear accessed and referenced bits. */
1068                 pmdp_test_and_clear_young(vma, addr, pmd);
1069                 test_and_clear_page_young(page);
1070                 ClearPageReferenced(page);
1071 out:
1072                 spin_unlock(ptl);
1073                 return 0;
1074         }
1075
1076         if (pmd_trans_unstable(pmd))
1077                 return 0;
1078
1079         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1080         for (; addr != end; pte++, addr += PAGE_SIZE) {
1081                 ptent = *pte;
1082
1083                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1084                         clear_soft_dirty(vma, addr, pte);
1085                         continue;
1086                 }
1087
1088                 if (!pte_present(ptent))
1089                         continue;
1090
1091                 page = vm_normal_page(vma, addr, ptent);
1092                 if (!page)
1093                         continue;
1094
1095                 /* Clear accessed and referenced bits. */
1096                 ptep_test_and_clear_young(vma, addr, pte);
1097                 test_and_clear_page_young(page);
1098                 ClearPageReferenced(page);
1099         }
1100         pte_unmap_unlock(pte - 1, ptl);
1101         cond_resched();
1102         return 0;
1103 }
1104
1105 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1106                                 struct mm_walk *walk)
1107 {
1108         struct clear_refs_private *cp = walk->private;
1109         struct vm_area_struct *vma = walk->vma;
1110
1111         if (vma->vm_flags & VM_PFNMAP)
1112                 return 1;
1113
1114         /*
1115          * Writing 1 to /proc/pid/clear_refs affects all pages.
1116          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1117          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1118          * Writing 4 to /proc/pid/clear_refs affects all pages.
1119          */
1120         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1121                 return 1;
1122         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1123                 return 1;
1124         return 0;
1125 }
1126
1127 static const struct mm_walk_ops clear_refs_walk_ops = {
1128         .pmd_entry              = clear_refs_pte_range,
1129         .test_walk              = clear_refs_test_walk,
1130 };
1131
1132 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1133                                 size_t count, loff_t *ppos)
1134 {
1135         struct task_struct *task;
1136         char buffer[PROC_NUMBUF];
1137         struct mm_struct *mm;
1138         struct vm_area_struct *vma;
1139         enum clear_refs_types type;
1140         struct mmu_gather tlb;
1141         int itype;
1142         int rv;
1143
1144         memset(buffer, 0, sizeof(buffer));
1145         if (count > sizeof(buffer) - 1)
1146                 count = sizeof(buffer) - 1;
1147         if (copy_from_user(buffer, buf, count))
1148                 return -EFAULT;
1149         rv = kstrtoint(strstrip(buffer), 10, &itype);
1150         if (rv < 0)
1151                 return rv;
1152         type = (enum clear_refs_types)itype;
1153         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1154                 return -EINVAL;
1155
1156         task = get_proc_task(file_inode(file));
1157         if (!task)
1158                 return -ESRCH;
1159         mm = get_task_mm(task);
1160         if (mm) {
1161                 struct mmu_notifier_range range;
1162                 struct clear_refs_private cp = {
1163                         .type = type,
1164                 };
1165
1166                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1167                         if (down_write_killable(&mm->mmap_sem)) {
1168                                 count = -EINTR;
1169                                 goto out_mm;
1170                         }
1171
1172                         /*
1173                          * Writing 5 to /proc/pid/clear_refs resets the peak
1174                          * resident set size to this mm's current rss value.
1175                          */
1176                         reset_mm_hiwater_rss(mm);
1177                         up_write(&mm->mmap_sem);
1178                         goto out_mm;
1179                 }
1180
1181                 if (down_read_killable(&mm->mmap_sem)) {
1182                         count = -EINTR;
1183                         goto out_mm;
1184                 }
1185                 tlb_gather_mmu(&tlb, mm, 0, -1);
1186                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1187                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1188                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1189                                         continue;
1190                                 up_read(&mm->mmap_sem);
1191                                 if (down_write_killable(&mm->mmap_sem)) {
1192                                         count = -EINTR;
1193                                         goto out_mm;
1194                                 }
1195                                 /*
1196                                  * Avoid to modify vma->vm_flags
1197                                  * without locked ops while the
1198                                  * coredump reads the vm_flags.
1199                                  */
1200                                 if (!mmget_still_valid(mm)) {
1201                                         /*
1202                                          * Silently return "count"
1203                                          * like if get_task_mm()
1204                                          * failed. FIXME: should this
1205                                          * function have returned
1206                                          * -ESRCH if get_task_mm()
1207                                          * failed like if
1208                                          * get_proc_task() fails?
1209                                          */
1210                                         up_write(&mm->mmap_sem);
1211                                         goto out_mm;
1212                                 }
1213                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1214                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1215                                         vma_set_page_prot(vma);
1216                                 }
1217                                 downgrade_write(&mm->mmap_sem);
1218                                 break;
1219                         }
1220
1221                         mmu_notifier_range_init(&range, MMU_NOTIFY_SOFT_DIRTY,
1222                                                 0, NULL, mm, 0, -1UL);
1223                         mmu_notifier_invalidate_range_start(&range);
1224                 }
1225                 walk_page_range(mm, 0, mm->highest_vm_end, &clear_refs_walk_ops,
1226                                 &cp);
1227                 if (type == CLEAR_REFS_SOFT_DIRTY)
1228                         mmu_notifier_invalidate_range_end(&range);
1229                 tlb_finish_mmu(&tlb, 0, -1);
1230                 up_read(&mm->mmap_sem);
1231 out_mm:
1232                 mmput(mm);
1233         }
1234         put_task_struct(task);
1235
1236         return count;
1237 }
1238
1239 const struct file_operations proc_clear_refs_operations = {
1240         .write          = clear_refs_write,
1241         .llseek         = noop_llseek,
1242 };
1243
1244 typedef struct {
1245         u64 pme;
1246 } pagemap_entry_t;
1247
1248 struct pagemapread {
1249         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1250         pagemap_entry_t *buffer;
1251         bool show_pfn;
1252 };
1253
1254 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1255 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1256
1257 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1258 #define PM_PFRAME_BITS          55
1259 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1260 #define PM_SOFT_DIRTY           BIT_ULL(55)
1261 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1262 #define PM_FILE                 BIT_ULL(61)
1263 #define PM_SWAP                 BIT_ULL(62)
1264 #define PM_PRESENT              BIT_ULL(63)
1265
1266 #define PM_END_OF_BUFFER    1
1267
1268 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1269 {
1270         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1271 }
1272
1273 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1274                           struct pagemapread *pm)
1275 {
1276         pm->buffer[pm->pos++] = *pme;
1277         if (pm->pos >= pm->len)
1278                 return PM_END_OF_BUFFER;
1279         return 0;
1280 }
1281
1282 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1283                                 struct mm_walk *walk)
1284 {
1285         struct pagemapread *pm = walk->private;
1286         unsigned long addr = start;
1287         int err = 0;
1288
1289         while (addr < end) {
1290                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1291                 pagemap_entry_t pme = make_pme(0, 0);
1292                 /* End of address space hole, which we mark as non-present. */
1293                 unsigned long hole_end;
1294
1295                 if (vma)
1296                         hole_end = min(end, vma->vm_start);
1297                 else
1298                         hole_end = end;
1299
1300                 for (; addr < hole_end; addr += PAGE_SIZE) {
1301                         err = add_to_pagemap(addr, &pme, pm);
1302                         if (err)
1303                                 goto out;
1304                 }
1305
1306                 if (!vma)
1307                         break;
1308
1309                 /* Addresses in the VMA. */
1310                 if (vma->vm_flags & VM_SOFTDIRTY)
1311                         pme = make_pme(0, PM_SOFT_DIRTY);
1312                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1313                         err = add_to_pagemap(addr, &pme, pm);
1314                         if (err)
1315                                 goto out;
1316                 }
1317         }
1318 out:
1319         return err;
1320 }
1321
1322 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1323                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1324 {
1325         u64 frame = 0, flags = 0;
1326         struct page *page = NULL;
1327
1328         if (pte_present(pte)) {
1329                 if (pm->show_pfn)
1330                         frame = pte_pfn(pte);
1331                 flags |= PM_PRESENT;
1332                 page = vm_normal_page(vma, addr, pte);
1333                 if (pte_soft_dirty(pte))
1334                         flags |= PM_SOFT_DIRTY;
1335         } else if (is_swap_pte(pte)) {
1336                 swp_entry_t entry;
1337                 if (pte_swp_soft_dirty(pte))
1338                         flags |= PM_SOFT_DIRTY;
1339                 entry = pte_to_swp_entry(pte);
1340                 if (pm->show_pfn)
1341                         frame = swp_type(entry) |
1342                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1343                 flags |= PM_SWAP;
1344                 if (is_migration_entry(entry))
1345                         page = migration_entry_to_page(entry);
1346
1347                 if (is_device_private_entry(entry))
1348                         page = device_private_entry_to_page(entry);
1349         }
1350
1351         if (page && !PageAnon(page))
1352                 flags |= PM_FILE;
1353         if (page && page_mapcount(page) == 1)
1354                 flags |= PM_MMAP_EXCLUSIVE;
1355         if (vma->vm_flags & VM_SOFTDIRTY)
1356                 flags |= PM_SOFT_DIRTY;
1357
1358         return make_pme(frame, flags);
1359 }
1360
1361 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1362                              struct mm_walk *walk)
1363 {
1364         struct vm_area_struct *vma = walk->vma;
1365         struct pagemapread *pm = walk->private;
1366         spinlock_t *ptl;
1367         pte_t *pte, *orig_pte;
1368         int err = 0;
1369
1370 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1371         ptl = pmd_trans_huge_lock(pmdp, vma);
1372         if (ptl) {
1373                 u64 flags = 0, frame = 0;
1374                 pmd_t pmd = *pmdp;
1375                 struct page *page = NULL;
1376
1377                 if (vma->vm_flags & VM_SOFTDIRTY)
1378                         flags |= PM_SOFT_DIRTY;
1379
1380                 if (pmd_present(pmd)) {
1381                         page = pmd_page(pmd);
1382
1383                         flags |= PM_PRESENT;
1384                         if (pmd_soft_dirty(pmd))
1385                                 flags |= PM_SOFT_DIRTY;
1386                         if (pm->show_pfn)
1387                                 frame = pmd_pfn(pmd) +
1388                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1389                 }
1390 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1391                 else if (is_swap_pmd(pmd)) {
1392                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1393                         unsigned long offset;
1394
1395                         if (pm->show_pfn) {
1396                                 offset = swp_offset(entry) +
1397                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1398                                 frame = swp_type(entry) |
1399                                         (offset << MAX_SWAPFILES_SHIFT);
1400                         }
1401                         flags |= PM_SWAP;
1402                         if (pmd_swp_soft_dirty(pmd))
1403                                 flags |= PM_SOFT_DIRTY;
1404                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1405                         page = migration_entry_to_page(entry);
1406                 }
1407 #endif
1408
1409                 if (page && page_mapcount(page) == 1)
1410                         flags |= PM_MMAP_EXCLUSIVE;
1411
1412                 for (; addr != end; addr += PAGE_SIZE) {
1413                         pagemap_entry_t pme = make_pme(frame, flags);
1414
1415                         err = add_to_pagemap(addr, &pme, pm);
1416                         if (err)
1417                                 break;
1418                         if (pm->show_pfn) {
1419                                 if (flags & PM_PRESENT)
1420                                         frame++;
1421                                 else if (flags & PM_SWAP)
1422                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1423                         }
1424                 }
1425                 spin_unlock(ptl);
1426                 return err;
1427         }
1428
1429         if (pmd_trans_unstable(pmdp))
1430                 return 0;
1431 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1432
1433         /*
1434          * We can assume that @vma always points to a valid one and @end never
1435          * goes beyond vma->vm_end.
1436          */
1437         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1438         for (; addr < end; pte++, addr += PAGE_SIZE) {
1439                 pagemap_entry_t pme;
1440
1441                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1442                 err = add_to_pagemap(addr, &pme, pm);
1443                 if (err)
1444                         break;
1445         }
1446         pte_unmap_unlock(orig_pte, ptl);
1447
1448         cond_resched();
1449
1450         return err;
1451 }
1452
1453 #ifdef CONFIG_HUGETLB_PAGE
1454 /* This function walks within one hugetlb entry in the single call */
1455 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1456                                  unsigned long addr, unsigned long end,
1457                                  struct mm_walk *walk)
1458 {
1459         struct pagemapread *pm = walk->private;
1460         struct vm_area_struct *vma = walk->vma;
1461         u64 flags = 0, frame = 0;
1462         int err = 0;
1463         pte_t pte;
1464
1465         if (vma->vm_flags & VM_SOFTDIRTY)
1466                 flags |= PM_SOFT_DIRTY;
1467
1468         pte = huge_ptep_get(ptep);
1469         if (pte_present(pte)) {
1470                 struct page *page = pte_page(pte);
1471
1472                 if (!PageAnon(page))
1473                         flags |= PM_FILE;
1474
1475                 if (page_mapcount(page) == 1)
1476                         flags |= PM_MMAP_EXCLUSIVE;
1477
1478                 flags |= PM_PRESENT;
1479                 if (pm->show_pfn)
1480                         frame = pte_pfn(pte) +
1481                                 ((addr & ~hmask) >> PAGE_SHIFT);
1482         }
1483
1484         for (; addr != end; addr += PAGE_SIZE) {
1485                 pagemap_entry_t pme = make_pme(frame, flags);
1486
1487                 err = add_to_pagemap(addr, &pme, pm);
1488                 if (err)
1489                         return err;
1490                 if (pm->show_pfn && (flags & PM_PRESENT))
1491                         frame++;
1492         }
1493
1494         cond_resched();
1495
1496         return err;
1497 }
1498 #else
1499 #define pagemap_hugetlb_range   NULL
1500 #endif /* HUGETLB_PAGE */
1501
1502 static const struct mm_walk_ops pagemap_ops = {
1503         .pmd_entry      = pagemap_pmd_range,
1504         .pte_hole       = pagemap_pte_hole,
1505         .hugetlb_entry  = pagemap_hugetlb_range,
1506 };
1507
1508 /*
1509  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1510  *
1511  * For each page in the address space, this file contains one 64-bit entry
1512  * consisting of the following:
1513  *
1514  * Bits 0-54  page frame number (PFN) if present
1515  * Bits 0-4   swap type if swapped
1516  * Bits 5-54  swap offset if swapped
1517  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1518  * Bit  56    page exclusively mapped
1519  * Bits 57-60 zero
1520  * Bit  61    page is file-page or shared-anon
1521  * Bit  62    page swapped
1522  * Bit  63    page present
1523  *
1524  * If the page is not present but in swap, then the PFN contains an
1525  * encoding of the swap file number and the page's offset into the
1526  * swap. Unmapped pages return a null PFN. This allows determining
1527  * precisely which pages are mapped (or in swap) and comparing mapped
1528  * pages between processes.
1529  *
1530  * Efficient users of this interface will use /proc/pid/maps to
1531  * determine which areas of memory are actually mapped and llseek to
1532  * skip over unmapped regions.
1533  */
1534 static ssize_t pagemap_read(struct file *file, char __user *buf,
1535                             size_t count, loff_t *ppos)
1536 {
1537         struct mm_struct *mm = file->private_data;
1538         struct pagemapread pm;
1539         unsigned long src;
1540         unsigned long svpfn;
1541         unsigned long start_vaddr;
1542         unsigned long end_vaddr;
1543         int ret = 0, copied = 0;
1544
1545         if (!mm || !mmget_not_zero(mm))
1546                 goto out;
1547
1548         ret = -EINVAL;
1549         /* file position must be aligned */
1550         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1551                 goto out_mm;
1552
1553         ret = 0;
1554         if (!count)
1555                 goto out_mm;
1556
1557         /* do not disclose physical addresses: attack vector */
1558         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1559
1560         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1561         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1562         ret = -ENOMEM;
1563         if (!pm.buffer)
1564                 goto out_mm;
1565
1566         src = *ppos;
1567         svpfn = src / PM_ENTRY_BYTES;
1568         start_vaddr = svpfn << PAGE_SHIFT;
1569         end_vaddr = mm->task_size;
1570
1571         /* watch out for wraparound */
1572         if (svpfn > mm->task_size >> PAGE_SHIFT)
1573                 start_vaddr = end_vaddr;
1574
1575         /*
1576          * The odds are that this will stop walking way
1577          * before end_vaddr, because the length of the
1578          * user buffer is tracked in "pm", and the walk
1579          * will stop when we hit the end of the buffer.
1580          */
1581         ret = 0;
1582         while (count && (start_vaddr < end_vaddr)) {
1583                 int len;
1584                 unsigned long end;
1585
1586                 pm.pos = 0;
1587                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1588                 /* overflow ? */
1589                 if (end < start_vaddr || end > end_vaddr)
1590                         end = end_vaddr;
1591                 ret = down_read_killable(&mm->mmap_sem);
1592                 if (ret)
1593                         goto out_free;
1594                 ret = walk_page_range(mm, start_vaddr, end, &pagemap_ops, &pm);
1595                 up_read(&mm->mmap_sem);
1596                 start_vaddr = end;
1597
1598                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1599                 if (copy_to_user(buf, pm.buffer, len)) {
1600                         ret = -EFAULT;
1601                         goto out_free;
1602                 }
1603                 copied += len;
1604                 buf += len;
1605                 count -= len;
1606         }
1607         *ppos += copied;
1608         if (!ret || ret == PM_END_OF_BUFFER)
1609                 ret = copied;
1610
1611 out_free:
1612         kfree(pm.buffer);
1613 out_mm:
1614         mmput(mm);
1615 out:
1616         return ret;
1617 }
1618
1619 static int pagemap_open(struct inode *inode, struct file *file)
1620 {
1621         struct mm_struct *mm;
1622
1623         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1624         if (IS_ERR(mm))
1625                 return PTR_ERR(mm);
1626         file->private_data = mm;
1627         return 0;
1628 }
1629
1630 static int pagemap_release(struct inode *inode, struct file *file)
1631 {
1632         struct mm_struct *mm = file->private_data;
1633
1634         if (mm)
1635                 mmdrop(mm);
1636         return 0;
1637 }
1638
1639 const struct file_operations proc_pagemap_operations = {
1640         .llseek         = mem_lseek, /* borrow this */
1641         .read           = pagemap_read,
1642         .open           = pagemap_open,
1643         .release        = pagemap_release,
1644 };
1645 #endif /* CONFIG_PROC_PAGE_MONITOR */
1646
1647 #ifdef CONFIG_NUMA
1648
1649 struct numa_maps {
1650         unsigned long pages;
1651         unsigned long anon;
1652         unsigned long active;
1653         unsigned long writeback;
1654         unsigned long mapcount_max;
1655         unsigned long dirty;
1656         unsigned long swapcache;
1657         unsigned long node[MAX_NUMNODES];
1658 };
1659
1660 struct numa_maps_private {
1661         struct proc_maps_private proc_maps;
1662         struct numa_maps md;
1663 };
1664
1665 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1666                         unsigned long nr_pages)
1667 {
1668         int count = page_mapcount(page);
1669
1670         md->pages += nr_pages;
1671         if (pte_dirty || PageDirty(page))
1672                 md->dirty += nr_pages;
1673
1674         if (PageSwapCache(page))
1675                 md->swapcache += nr_pages;
1676
1677         if (PageActive(page) || PageUnevictable(page))
1678                 md->active += nr_pages;
1679
1680         if (PageWriteback(page))
1681                 md->writeback += nr_pages;
1682
1683         if (PageAnon(page))
1684                 md->anon += nr_pages;
1685
1686         if (count > md->mapcount_max)
1687                 md->mapcount_max = count;
1688
1689         md->node[page_to_nid(page)] += nr_pages;
1690 }
1691
1692 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1693                 unsigned long addr)
1694 {
1695         struct page *page;
1696         int nid;
1697
1698         if (!pte_present(pte))
1699                 return NULL;
1700
1701         page = vm_normal_page(vma, addr, pte);
1702         if (!page)
1703                 return NULL;
1704
1705         if (PageReserved(page))
1706                 return NULL;
1707
1708         nid = page_to_nid(page);
1709         if (!node_isset(nid, node_states[N_MEMORY]))
1710                 return NULL;
1711
1712         return page;
1713 }
1714
1715 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1716 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1717                                               struct vm_area_struct *vma,
1718                                               unsigned long addr)
1719 {
1720         struct page *page;
1721         int nid;
1722
1723         if (!pmd_present(pmd))
1724                 return NULL;
1725
1726         page = vm_normal_page_pmd(vma, addr, pmd);
1727         if (!page)
1728                 return NULL;
1729
1730         if (PageReserved(page))
1731                 return NULL;
1732
1733         nid = page_to_nid(page);
1734         if (!node_isset(nid, node_states[N_MEMORY]))
1735                 return NULL;
1736
1737         return page;
1738 }
1739 #endif
1740
1741 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1742                 unsigned long end, struct mm_walk *walk)
1743 {
1744         struct numa_maps *md = walk->private;
1745         struct vm_area_struct *vma = walk->vma;
1746         spinlock_t *ptl;
1747         pte_t *orig_pte;
1748         pte_t *pte;
1749
1750 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1751         ptl = pmd_trans_huge_lock(pmd, vma);
1752         if (ptl) {
1753                 struct page *page;
1754
1755                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1756                 if (page)
1757                         gather_stats(page, md, pmd_dirty(*pmd),
1758                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1759                 spin_unlock(ptl);
1760                 return 0;
1761         }
1762
1763         if (pmd_trans_unstable(pmd))
1764                 return 0;
1765 #endif
1766         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1767         do {
1768                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1769                 if (!page)
1770                         continue;
1771                 gather_stats(page, md, pte_dirty(*pte), 1);
1772
1773         } while (pte++, addr += PAGE_SIZE, addr != end);
1774         pte_unmap_unlock(orig_pte, ptl);
1775         cond_resched();
1776         return 0;
1777 }
1778 #ifdef CONFIG_HUGETLB_PAGE
1779 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1780                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1781 {
1782         pte_t huge_pte = huge_ptep_get(pte);
1783         struct numa_maps *md;
1784         struct page *page;
1785
1786         if (!pte_present(huge_pte))
1787                 return 0;
1788
1789         page = pte_page(huge_pte);
1790         if (!page)
1791                 return 0;
1792
1793         md = walk->private;
1794         gather_stats(page, md, pte_dirty(huge_pte), 1);
1795         return 0;
1796 }
1797
1798 #else
1799 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1800                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1801 {
1802         return 0;
1803 }
1804 #endif
1805
1806 static const struct mm_walk_ops show_numa_ops = {
1807         .hugetlb_entry = gather_hugetlb_stats,
1808         .pmd_entry = gather_pte_stats,
1809 };
1810
1811 /*
1812  * Display pages allocated per node and memory policy via /proc.
1813  */
1814 static int show_numa_map(struct seq_file *m, void *v)
1815 {
1816         struct numa_maps_private *numa_priv = m->private;
1817         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1818         struct vm_area_struct *vma = v;
1819         struct numa_maps *md = &numa_priv->md;
1820         struct file *file = vma->vm_file;
1821         struct mm_struct *mm = vma->vm_mm;
1822         struct mempolicy *pol;
1823         char buffer[64];
1824         int nid;
1825
1826         if (!mm)
1827                 return 0;
1828
1829         /* Ensure we start with an empty set of numa_maps statistics. */
1830         memset(md, 0, sizeof(*md));
1831
1832         pol = __get_vma_policy(vma, vma->vm_start);
1833         if (pol) {
1834                 mpol_to_str(buffer, sizeof(buffer), pol);
1835                 mpol_cond_put(pol);
1836         } else {
1837                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1838         }
1839
1840         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1841
1842         if (file) {
1843                 seq_puts(m, " file=");
1844                 seq_file_path(m, file, "\n\t= ");
1845         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1846                 seq_puts(m, " heap");
1847         } else if (is_stack(vma)) {
1848                 seq_puts(m, " stack");
1849         }
1850
1851         if (is_vm_hugetlb_page(vma))
1852                 seq_puts(m, " huge");
1853
1854         /* mmap_sem is held by m_start */
1855         walk_page_vma(vma, &show_numa_ops, md);
1856
1857         if (!md->pages)
1858                 goto out;
1859
1860         if (md->anon)
1861                 seq_printf(m, " anon=%lu", md->anon);
1862
1863         if (md->dirty)
1864                 seq_printf(m, " dirty=%lu", md->dirty);
1865
1866         if (md->pages != md->anon && md->pages != md->dirty)
1867                 seq_printf(m, " mapped=%lu", md->pages);
1868
1869         if (md->mapcount_max > 1)
1870                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1871
1872         if (md->swapcache)
1873                 seq_printf(m, " swapcache=%lu", md->swapcache);
1874
1875         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1876                 seq_printf(m, " active=%lu", md->active);
1877
1878         if (md->writeback)
1879                 seq_printf(m, " writeback=%lu", md->writeback);
1880
1881         for_each_node_state(nid, N_MEMORY)
1882                 if (md->node[nid])
1883                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1884
1885         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1886 out:
1887         seq_putc(m, '\n');
1888         m_cache_vma(m, vma);
1889         return 0;
1890 }
1891
1892 static const struct seq_operations proc_pid_numa_maps_op = {
1893         .start  = m_start,
1894         .next   = m_next,
1895         .stop   = m_stop,
1896         .show   = show_numa_map,
1897 };
1898
1899 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1900 {
1901         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1902                                 sizeof(struct numa_maps_private));
1903 }
1904
1905 const struct file_operations proc_pid_numa_maps_operations = {
1906         .open           = pid_numa_maps_open,
1907         .read           = seq_read,
1908         .llseek         = seq_lseek,
1909         .release        = proc_map_release,
1910 };
1911
1912 #endif /* CONFIG_NUMA */