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", 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         down_read(&mm->mmap_sem);
 170         hold_task_mempolicy(priv);
 171         priv->tail_vma = get_gate_vma(mm);
 172
 173         if (last_addr) {
 174                 vma = find_vma(mm, last_addr - 1);
 175                 if (vma && vma->vm_start <= last_addr)
 176                         vma = m_next_vma(priv, vma);
 177                 if (vma)
 178                         return vma;
 179         }
 180
 181         m->version = 0;
 182         if (pos < mm->map_count) {
 183                 for (vma = mm->mmap; pos; pos--) {
 184                         m->version = vma->vm_start;
 185                         vma = vma->vm_next;
 186                 }
 187                 return vma;
 188         }
 189
 190         /* we do not bother to update m->version in this case */
 191         if (pos == mm->map_count && priv->tail_vma)
 192                 return priv->tail_vma;
 193
 194         vma_stop(priv);
 195         return NULL;
 196 }
 197
 198 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 199 {
 200         struct proc_maps_private *priv = m->private;
 201         struct vm_area_struct *next;
 202
 203         (*pos)++;
 204         next = m_next_vma(priv, v);
 205         if (!next)
 206                 vma_stop(priv);
 207         return next;
 208 }
 209
 210 static void m_stop(struct seq_file *m, void *v)
 211 {
 212         struct proc_maps_private *priv = m->private;
 213
 214         if (!IS_ERR_OR_NULL(v))
 215                 vma_stop(priv);
 216         if (priv->task) {
 217                 put_task_struct(priv->task);
 218                 priv->task = NULL;
 219         }
 220 }
 221
 222 static int proc_maps_open(struct inode *inode, struct file *file,
 223                         const struct seq_operations *ops, int psize)
 224 {
 225         struct proc_maps_private *priv = __seq_open_private(file, ops, psize);
 226
 227         if (!priv)
 228                 return -ENOMEM;
 229
 230         priv->inode = inode;
 231         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 232         if (IS_ERR(priv->mm)) {
 233                 int err = PTR_ERR(priv->mm);
 234
 235                 seq_release_private(inode, file);
 236                 return err;
 237         }
 238
 239         return 0;
 240 }
 241
 242 static int proc_map_release(struct inode *inode, struct file *file)
 243 {
 244         struct seq_file *seq = file->private_data;
 245         struct proc_maps_private *priv = seq->private;
 246
 247         if (priv->mm)
 248                 mmdrop(priv->mm);
 249
 250         return seq_release_private(inode, file);
 251 }
 252
 253 static int do_maps_open(struct inode *inode, struct file *file,
 254                         const struct seq_operations *ops)
 255 {
 256         return proc_maps_open(inode, file, ops,
 257                                 sizeof(struct proc_maps_private));
 258 }
 259
 260 /*
 261  * Indicate if the VMA is a stack for the given task; for
 262  * /proc/PID/maps that is the stack of the main task.
 263  */
 264 static int is_stack(struct vm_area_struct *vma)
 265 {
 266         /*
 267          * We make no effort to guess what a given thread considers to be
 268          * its "stack".  It's not even well-defined for programs written
 269          * languages like Go.
 270          */
 271         return vma->vm_start <= vma->vm_mm->start_stack &&
 272                 vma->vm_end >= vma->vm_mm->start_stack;
 273 }
 274
 275 static void show_vma_header_prefix(struct seq_file *m,
 276                                    unsigned long start, unsigned long end,
 277                                    vm_flags_t flags, unsigned long long pgoff,
 278                                    dev_t dev, unsigned long ino)
 279 {
 280         seq_setwidth(m, 25 + sizeof(void *) * 6 - 1);
 281         seq_put_hex_ll(m, NULL, start, 8);
 282         seq_put_hex_ll(m, "-", end, 8);
 283         seq_putc(m, ' ');
 284         seq_putc(m, flags & VM_READ ? 'r' : '-');
 285         seq_putc(m, flags & VM_WRITE ? 'w' : '-');
 286         seq_putc(m, flags & VM_EXEC ? 'x' : '-');
 287         seq_putc(m, flags & VM_MAYSHARE ? 's' : 'p');
 288         seq_put_hex_ll(m, " ", pgoff, 8);
 289         seq_put_hex_ll(m, " ", MAJOR(dev), 2);
 290         seq_put_hex_ll(m, ":", MINOR(dev), 2);
 291         seq_put_decimal_ull(m, " ", ino);
 292         seq_putc(m, ' ');
 293 }
 294
 295 static void
 296 show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 297 {
 298         struct mm_struct *mm = vma->vm_mm;
 299         struct file *file = vma->vm_file;
 300         vm_flags_t flags = vma->vm_flags;
 301         unsigned long ino = 0;
 302         unsigned long long pgoff = 0;
 303         unsigned long start, end;
 304         dev_t dev = 0;
 305         const char *name = NULL;
 306
 307         if (file) {
 308                 struct inode *inode = file_inode(vma->vm_file);
 309                 dev = inode->i_sb->s_dev;
 310                 ino = inode->i_ino;
 311                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 312         }
 313
 314         start = vma->vm_start;
 315         end = vma->vm_end;
 316         show_vma_header_prefix(m, start, end, flags, pgoff, dev, ino);
 317
 318         /*
 319          * Print the dentry name for named mappings, and a
 320          * special [heap] marker for the heap:
 321          */
 322         if (file) {
 323                 seq_pad(m, ' ');
 324                 seq_file_path(m, file, "\n");
 325                 goto done;
 326         }
 327
 328         if (vma->vm_ops && vma->vm_ops->name) {
 329                 name = vma->vm_ops->name(vma);
 330                 if (name)
 331                         goto done;
 332         }
 333
 334         name = arch_vma_name(vma);
 335         if (!name) {
 336                 if (!mm) {
 337                         name = "[vdso]";
 338                         goto done;
 339                 }
 340
 341                 if (vma->vm_start <= mm->brk &&
 342                     vma->vm_end >= mm->start_brk) {
 343                         name = "[heap]";
 344                         goto done;
 345                 }
 346
 347                 if (is_stack(vma))
 348                         name = "[stack]";
 349         }
 350
 351 done:
 352         if (name) {
 353                 seq_pad(m, ' ');
 354                 seq_puts(m, name);
 355         }
 356         seq_putc(m, '\n');
 357 }
 358
 359 static int show_map(struct seq_file *m, void *v)
 360 {
 361         show_map_vma(m, v);
 362         m_cache_vma(m, v);
 363         return 0;
 364 }
 365
 366 static const struct seq_operations proc_pid_maps_op = {
 367         .start  = m_start,
 368         .next   = m_next,
 369         .stop   = m_stop,
 370         .show   = show_map
 371 };
 372
 373 static int pid_maps_open(struct inode *inode, struct file *file)
 374 {
 375         return do_maps_open(inode, file, &proc_pid_maps_op);
 376 }
 377
 378 const struct file_operations proc_pid_maps_operations = {
 379         .open           = pid_maps_open,
 380         .read           = seq_read,
 381         .llseek         = seq_lseek,
 382         .release        = proc_map_release,
 383 };
 384
 385 /*
 386  * Proportional Set Size(PSS): my share of RSS.
 387  *
 388  * PSS of a process is the count of pages it has in memory, where each
 389  * page is divided by the number of processes sharing it.  So if a
 390  * process has 1000 pages all to itself, and 1000 shared with one other
 391  * process, its PSS will be 1500.
 392  *
 393  * To keep (accumulated) division errors low, we adopt a 64bit
 394  * fixed-point pss counter to minimize division errors. So (pss >>
 395  * PSS_SHIFT) would be the real byte count.
 396  *
 397  * A shift of 12 before division means (assuming 4K page size):
 398  *      - 1M 3-user-pages add up to 8KB errors;
 399  *      - supports mapcount up to 2^24, or 16M;
 400  *      - supports PSS up to 2^52 bytes, or 4PB.
 401  */
 402 #define PSS_SHIFT 12
 403
 404 #ifdef CONFIG_PROC_PAGE_MONITOR
 405 struct mem_size_stats {
 406         unsigned long resident;
 407         unsigned long shared_clean;
 408         unsigned long shared_dirty;
 409         unsigned long private_clean;
 410         unsigned long private_dirty;
 411         unsigned long referenced;
 412         unsigned long anonymous;
 413         unsigned long lazyfree;
 414         unsigned long anonymous_thp;
 415         unsigned long shmem_thp;
 416         unsigned long swap;
 417         unsigned long shared_hugetlb;
 418         unsigned long private_hugetlb;
 419         u64 pss;
 420         u64 pss_locked;
 421         u64 swap_pss;
 422         bool check_shmem_swap;
 423 };
 424
 425 static void smaps_account(struct mem_size_stats *mss, struct page *page,
 426                 bool compound, bool young, bool dirty)
 427 {
 428         int i, nr = compound ? 1 << compound_order(page) : 1;
 429         unsigned long size = nr * PAGE_SIZE;
 430
 431         if (PageAnon(page)) {
 432                 mss->anonymous += size;
 433                 if (!PageSwapBacked(page) && !dirty && !PageDirty(page))
 434                         mss->lazyfree += size;
 435         }
 436
 437         mss->resident += size;
 438         /* Accumulate the size in pages that have been accessed. */
 439         if (young || page_is_young(page) || PageReferenced(page))
 440                 mss->referenced += size;
 441
 442         /*
 443          * page_count(page) == 1 guarantees the page is mapped exactly once.
 444          * If any subpage of the compound page mapped with PTE it would elevate
 445          * page_count().
 446          */
 447         if (page_count(page) == 1) {
 448                 if (dirty || PageDirty(page))
 449                         mss->private_dirty += size;
 450                 else
 451                         mss->private_clean += size;
 452                 mss->pss += (u64)size << PSS_SHIFT;
 453                 return;
 454         }
 455
 456         for (i = 0; i < nr; i++, page++) {
 457                 int mapcount = page_mapcount(page);
 458
 459                 if (mapcount >= 2) {
 460                         if (dirty || PageDirty(page))
 461                                 mss->shared_dirty += PAGE_SIZE;
 462                         else
 463                                 mss->shared_clean += PAGE_SIZE;
 464                         mss->pss += (PAGE_SIZE << PSS_SHIFT) / mapcount;
 465                 } else {
 466                         if (dirty || PageDirty(page))
 467                                 mss->private_dirty += PAGE_SIZE;
 468                         else
 469                                 mss->private_clean += PAGE_SIZE;
 470                         mss->pss += PAGE_SIZE << PSS_SHIFT;
 471                 }
 472         }
 473 }
 474
 475 #ifdef CONFIG_SHMEM
 476 static int smaps_pte_hole(unsigned long addr, unsigned long end,
 477                 struct mm_walk *walk)
 478 {
 479         struct mem_size_stats *mss = walk->private;
 480
 481         mss->swap += shmem_partial_swap_usage(
 482                         walk->vma->vm_file->f_mapping, addr, end);
 483
 484         return 0;
 485 }
 486 #endif
 487
 488 static void smaps_pte_entry(pte_t *pte, unsigned long addr,
 489                 struct mm_walk *walk)
 490 {
 491         struct mem_size_stats *mss = walk->private;
 492         struct vm_area_struct *vma = walk->vma;
 493         struct page *page = NULL;
 494
 495         if (pte_present(*pte)) {
 496                 page = vm_normal_page(vma, addr, *pte);
 497         } else if (is_swap_pte(*pte)) {
 498                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 499
 500                 if (!non_swap_entry(swpent)) {
 501                         int mapcount;
 502
 503                         mss->swap += PAGE_SIZE;
 504                         mapcount = swp_swapcount(swpent);
 505                         if (mapcount >= 2) {
 506                                 u64 pss_delta = (u64)PAGE_SIZE << PSS_SHIFT;
 507
 508                                 do_div(pss_delta, mapcount);
 509                                 mss->swap_pss += pss_delta;
 510                         } else {
 511                                 mss->swap_pss += (u64)PAGE_SIZE << PSS_SHIFT;
 512                         }
 513                 } else if (is_migration_entry(swpent))
 514                         page = migration_entry_to_page(swpent);
 515                 else if (is_device_private_entry(swpent))
 516                         page = device_private_entry_to_page(swpent);
 517         } else if (unlikely(IS_ENABLED(CONFIG_SHMEM) && mss->check_shmem_swap
 518                                                         && pte_none(*pte))) {
 519                 page = find_get_entry(vma->vm_file->f_mapping,
 520                                                 linear_page_index(vma, addr));
 521                 if (!page)
 522                         return;
 523
 524                 if (xa_is_value(page))
 525                         mss->swap += PAGE_SIZE;
 526                 else
 527                         put_page(page);
 528
 529                 return;
 530         }
 531
 532         if (!page)
 533                 return;
 534
 535         smaps_account(mss, page, false, pte_young(*pte), pte_dirty(*pte));
 536 }
 537
 538 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 539 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 540                 struct mm_walk *walk)
 541 {
 542         struct mem_size_stats *mss = walk->private;
 543         struct vm_area_struct *vma = walk->vma;
 544         struct page *page;
 545
 546         /* FOLL_DUMP will return -EFAULT on huge zero page */
 547         page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP);
 548         if (IS_ERR_OR_NULL(page))
 549                 return;
 550         if (PageAnon(page))
 551                 mss->anonymous_thp += HPAGE_PMD_SIZE;
 552         else if (PageSwapBacked(page))
 553                 mss->shmem_thp += HPAGE_PMD_SIZE;
 554         else if (is_zone_device_page(page))
 555                 /* pass */;
 556         else
 557                 VM_BUG_ON_PAGE(1, page);
 558         smaps_account(mss, page, true, pmd_young(*pmd), pmd_dirty(*pmd));
 559 }
 560 #else
 561 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr,
 562                 struct mm_walk *walk)
 563 {
 564 }
 565 #endif
 566
 567 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 568                            struct mm_walk *walk)
 569 {
 570         struct vm_area_struct *vma = walk->vma;
 571         pte_t *pte;
 572         spinlock_t *ptl;
 573
 574         ptl = pmd_trans_huge_lock(pmd, vma);
 575         if (ptl) {
 576                 if (pmd_present(*pmd))
 577                         smaps_pmd_entry(pmd, addr, walk);
 578                 spin_unlock(ptl);
 579                 goto out;
 580         }
 581
 582         if (pmd_trans_unstable(pmd))
 583                 goto out;
 584         /*
 585          * The mmap_sem held all the way back in m_start() is what
 586          * keeps khugepaged out of here and from collapsing things
 587          * in here.
 588          */
 589         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
 590         for (; addr != end; pte++, addr += PAGE_SIZE)
 591                 smaps_pte_entry(pte, addr, walk);
 592         pte_unmap_unlock(pte - 1, ptl);
 593 out:
 594         cond_resched();
 595         return 0;
 596 }
 597
 598 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma)
 599 {
 600         /*
 601          * Don't forget to update Documentation/ on changes.
 602          */
 603         static const char mnemonics[BITS_PER_LONG][2] = {
 604                 /*
 605                  * In case if we meet a flag we don't know about.
 606                  */
 607                 [0 ... (BITS_PER_LONG-1)] = "??",
 608
 609                 [ilog2(VM_READ)]        = "rd",
 610                 [ilog2(VM_WRITE)]       = "wr",
 611                 [ilog2(VM_EXEC)]        = "ex",
 612                 [ilog2(VM_SHARED)]      = "sh",
 613                 [ilog2(VM_MAYREAD)]     = "mr",
 614                 [ilog2(VM_MAYWRITE)]    = "mw",
 615                 [ilog2(VM_MAYEXEC)]     = "me",
 616                 [ilog2(VM_MAYSHARE)]    = "ms",
 617                 [ilog2(VM_GROWSDOWN)]   = "gd",
 618                 [ilog2(VM_PFNMAP)]      = "pf",
 619                 [ilog2(VM_DENYWRITE)]   = "dw",
 620 #ifdef CONFIG_X86_INTEL_MPX
 621                 [ilog2(VM_MPX)]         = "mp",
 622 #endif
 623                 [ilog2(VM_LOCKED)]      = "lo",
 624                 [ilog2(VM_IO)]          = "io",
 625                 [ilog2(VM_SEQ_READ)]    = "sr",
 626                 [ilog2(VM_RAND_READ)]   = "rr",
 627                 [ilog2(VM_DONTCOPY)]    = "dc",
 628                 [ilog2(VM_DONTEXPAND)]  = "de",
 629                 [ilog2(VM_ACCOUNT)]     = "ac",
 630                 [ilog2(VM_NORESERVE)]   = "nr",
 631                 [ilog2(VM_HUGETLB)]     = "ht",
 632                 [ilog2(VM_SYNC)]        = "sf",
 633                 [ilog2(VM_ARCH_1)]      = "ar",
 634                 [ilog2(VM_WIPEONFORK)]  = "wf",
 635                 [ilog2(VM_DONTDUMP)]    = "dd",
 636 #ifdef CONFIG_MEM_SOFT_DIRTY
 637                 [ilog2(VM_SOFTDIRTY)]   = "sd",
 638 #endif
 639                 [ilog2(VM_MIXEDMAP)]    = "mm",
 640                 [ilog2(VM_HUGEPAGE)]    = "hg",
 641                 [ilog2(VM_NOHUGEPAGE)]  = "nh",
 642                 [ilog2(VM_MERGEABLE)]   = "mg",
 643                 [ilog2(VM_UFFD_MISSING)]= "um",
 644                 [ilog2(VM_UFFD_WP)]     = "uw",
 645 #ifdef CONFIG_ARCH_HAS_PKEYS
 646                 /* These come out via ProtectionKey: */
 647                 [ilog2(VM_PKEY_BIT0)]   = "",
 648                 [ilog2(VM_PKEY_BIT1)]   = "",
 649                 [ilog2(VM_PKEY_BIT2)]   = "",
 650                 [ilog2(VM_PKEY_BIT3)]   = "",
 651 #if VM_PKEY_BIT4
 652                 [ilog2(VM_PKEY_BIT4)]   = "",
 653 #endif
 654 #endif /* CONFIG_ARCH_HAS_PKEYS */
 655         };
 656         size_t i;
 657
 658         seq_puts(m, "VmFlags: ");
 659         for (i = 0; i < BITS_PER_LONG; i++) {
 660                 if (!mnemonics[i][0])
 661                         continue;
 662                 if (vma->vm_flags & (1UL << i)) {
 663                         seq_putc(m, mnemonics[i][0]);
 664                         seq_putc(m, mnemonics[i][1]);
 665                         seq_putc(m, ' ');
 666                 }
 667         }
 668         seq_putc(m, '\n');
 669 }
 670
 671 #ifdef CONFIG_HUGETLB_PAGE
 672 static int smaps_hugetlb_range(pte_t *pte, unsigned long hmask,
 673                                  unsigned long addr, unsigned long end,
 674                                  struct mm_walk *walk)
 675 {
 676         struct mem_size_stats *mss = walk->private;
 677         struct vm_area_struct *vma = walk->vma;
 678         struct page *page = NULL;
 679
 680         if (pte_present(*pte)) {
 681                 page = vm_normal_page(vma, addr, *pte);
 682         } else if (is_swap_pte(*pte)) {
 683                 swp_entry_t swpent = pte_to_swp_entry(*pte);
 684
 685                 if (is_migration_entry(swpent))
 686                         page = migration_entry_to_page(swpent);
 687                 else if (is_device_private_entry(swpent))
 688                         page = device_private_entry_to_page(swpent);
 689         }
 690         if (page) {
 691                 int mapcount = page_mapcount(page);
 692
 693                 if (mapcount >= 2)
 694                         mss->shared_hugetlb += huge_page_size(hstate_vma(vma));
 695                 else
 696                         mss->private_hugetlb += huge_page_size(hstate_vma(vma));
 697         }
 698         return 0;
 699 }
 700 #endif /* HUGETLB_PAGE */
 701
 702 static void smap_gather_stats(struct vm_area_struct *vma,
 703                              struct mem_size_stats *mss)
 704 {
 705         struct mm_walk smaps_walk = {
 706                 .pmd_entry = smaps_pte_range,
 707 #ifdef CONFIG_HUGETLB_PAGE
 708                 .hugetlb_entry = smaps_hugetlb_range,
 709 #endif
 710                 .mm = vma->vm_mm,
 711         };
 712
 713         smaps_walk.private = mss;
 714
 715 #ifdef CONFIG_SHMEM
 716         /* In case of smaps_rollup, reset the value from previous vma */
 717         mss->check_shmem_swap = false;
 718         if (vma->vm_file && shmem_mapping(vma->vm_file->f_mapping)) {
 719                 /*
 720                  * For shared or readonly shmem mappings we know that all
 721                  * swapped out pages belong to the shmem object, and we can
 722                  * obtain the swap value much more efficiently. For private
 723                  * writable mappings, we might have COW pages that are
 724                  * not affected by the parent swapped out pages of the shmem
 725                  * object, so we have to distinguish them during the page walk.
 726                  * Unless we know that the shmem object (or the part mapped by
 727                  * our VMA) has no swapped out pages at all.
 728                  */
 729                 unsigned long shmem_swapped = shmem_swap_usage(vma);
 730
 731                 if (!shmem_swapped || (vma->vm_flags & VM_SHARED) ||
 732                                         !(vma->vm_flags & VM_WRITE)) {
 733                         mss->swap += shmem_swapped;
 734                 } else {
 735                         mss->check_shmem_swap = true;
 736                         smaps_walk.pte_hole = smaps_pte_hole;
 737                 }
 738         }
 739 #endif
 740
 741         /* mmap_sem is held in m_start */
 742         walk_page_vma(vma, &smaps_walk);
 743         if (vma->vm_flags & VM_LOCKED)
 744                 mss->pss_locked += mss->pss;
 745 }
 746
 747 #define SEQ_PUT_DEC(str, val) \
 748                 seq_put_decimal_ull_width(m, str, (val) >> 10, 8)
 749
 750 /* Show the contents common for smaps and smaps_rollup */
 751 static void __show_smap(struct seq_file *m, const struct mem_size_stats *mss)
 752 {
 753         SEQ_PUT_DEC("Rss:            ", mss->resident);
 754         SEQ_PUT_DEC(" kB\nPss:            ", mss->pss >> PSS_SHIFT);
 755         SEQ_PUT_DEC(" kB\nShared_Clean:   ", mss->shared_clean);
 756         SEQ_PUT_DEC(" kB\nShared_Dirty:   ", mss->shared_dirty);
 757         SEQ_PUT_DEC(" kB\nPrivate_Clean:  ", mss->private_clean);
 758         SEQ_PUT_DEC(" kB\nPrivate_Dirty:  ", mss->private_dirty);
 759         SEQ_PUT_DEC(" kB\nReferenced:     ", mss->referenced);
 760         SEQ_PUT_DEC(" kB\nAnonymous:      ", mss->anonymous);
 761         SEQ_PUT_DEC(" kB\nLazyFree:       ", mss->lazyfree);
 762         SEQ_PUT_DEC(" kB\nAnonHugePages:  ", mss->anonymous_thp);
 763         SEQ_PUT_DEC(" kB\nShmemPmdMapped: ", mss->shmem_thp);
 764         SEQ_PUT_DEC(" kB\nShared_Hugetlb: ", mss->shared_hugetlb);
 765         seq_put_decimal_ull_width(m, " kB\nPrivate_Hugetlb: ",
 766                                   mss->private_hugetlb >> 10, 7);
 767         SEQ_PUT_DEC(" kB\nSwap:           ", mss->swap);
 768         SEQ_PUT_DEC(" kB\nSwapPss:        ",
 769                                         mss->swap_pss >> PSS_SHIFT);
 770         SEQ_PUT_DEC(" kB\nLocked:         ",
 771                                         mss->pss_locked >> PSS_SHIFT);
 772         seq_puts(m, " kB\n");
 773 }
 774
 775 static int show_smap(struct seq_file *m, void *v)
 776 {
 777         struct vm_area_struct *vma = v;
 778         struct mem_size_stats mss;
 779
 780         memset(&mss, 0, sizeof(mss));
 781
 782         smap_gather_stats(vma, &mss);
 783
 784         show_map_vma(m, vma);
 785
 786         SEQ_PUT_DEC("Size:           ", vma->vm_end - vma->vm_start);
 787         SEQ_PUT_DEC(" kB\nKernelPageSize: ", vma_kernel_pagesize(vma));
 788         SEQ_PUT_DEC(" kB\nMMUPageSize:    ", vma_mmu_pagesize(vma));
 789         seq_puts(m, " kB\n");
 790
 791         __show_smap(m, &mss);
 792
 793         seq_printf(m, "THPeligible:    %d\n", transparent_hugepage_enabled(vma));
 794
 795         if (arch_pkeys_enabled())
 796                 seq_printf(m, "ProtectionKey:  %8u\n", vma_pkey(vma));
 797         show_smap_vma_flags(m, vma);
 798
 799         m_cache_vma(m, vma);
 800
 801         return 0;
 802 }
 803
 804 static int show_smaps_rollup(struct seq_file *m, void *v)
 805 {
 806         struct proc_maps_private *priv = m->private;
 807         struct mem_size_stats mss;
 808         struct mm_struct *mm;
 809         struct vm_area_struct *vma;
 810         unsigned long last_vma_end = 0;
 811         int ret = 0;
 812
 813         priv->task = get_proc_task(priv->inode);
 814         if (!priv->task)
 815                 return -ESRCH;
 816
 817         mm = priv->mm;
 818         if (!mm || !mmget_not_zero(mm)) {
 819                 ret = -ESRCH;
 820                 goto out_put_task;
 821         }
 822
 823         memset(&mss, 0, sizeof(mss));
 824
 825         down_read(&mm->mmap_sem);
 826         hold_task_mempolicy(priv);
 827
 828         for (vma = priv->mm->mmap; vma; vma = vma->vm_next) {
 829                 smap_gather_stats(vma, &mss);
 830                 last_vma_end = vma->vm_end;
 831         }
 832
 833         show_vma_header_prefix(m, priv->mm->mmap->vm_start,
 834                                last_vma_end, 0, 0, 0, 0);
 835         seq_pad(m, ' ');
 836         seq_puts(m, "[rollup]\n");
 837
 838         __show_smap(m, &mss);
 839
 840         release_task_mempolicy(priv);
 841         up_read(&mm->mmap_sem);
 842         mmput(mm);
 843
 844 out_put_task:
 845         put_task_struct(priv->task);
 846         priv->task = NULL;
 847
 848         return ret;
 849 }
 850 #undef SEQ_PUT_DEC
 851
 852 static const struct seq_operations proc_pid_smaps_op = {
 853         .start  = m_start,
 854         .next   = m_next,
 855         .stop   = m_stop,
 856         .show   = show_smap
 857 };
 858
 859 static int pid_smaps_open(struct inode *inode, struct file *file)
 860 {
 861         return do_maps_open(inode, file, &proc_pid_smaps_op);
 862 }
 863
 864 static int smaps_rollup_open(struct inode *inode, struct file *file)
 865 {
 866         int ret;
 867         struct proc_maps_private *priv;
 868
 869         priv = kzalloc(sizeof(*priv), GFP_KERNEL_ACCOUNT);
 870         if (!priv)
 871                 return -ENOMEM;
 872
 873         ret = single_open(file, show_smaps_rollup, priv);
 874         if (ret)
 875                 goto out_free;
 876
 877         priv->inode = inode;
 878         priv->mm = proc_mem_open(inode, PTRACE_MODE_READ);
 879         if (IS_ERR(priv->mm)) {
 880                 ret = PTR_ERR(priv->mm);
 881
 882                 single_release(inode, file);
 883                 goto out_free;
 884         }
 885
 886         return 0;
 887
 888 out_free:
 889         kfree(priv);
 890         return ret;
 891 }
 892
 893 static int smaps_rollup_release(struct inode *inode, struct file *file)
 894 {
 895         struct seq_file *seq = file->private_data;
 896         struct proc_maps_private *priv = seq->private;
 897
 898         if (priv->mm)
 899                 mmdrop(priv->mm);
 900
 901         kfree(priv);
 902         return single_release(inode, file);
 903 }
 904
 905 const struct file_operations proc_pid_smaps_operations = {
 906         .open           = pid_smaps_open,
 907         .read           = seq_read,
 908         .llseek         = seq_lseek,
 909         .release        = proc_map_release,
 910 };
 911
 912 const struct file_operations proc_pid_smaps_rollup_operations = {
 913         .open           = smaps_rollup_open,
 914         .read           = seq_read,
 915         .llseek         = seq_lseek,
 916         .release        = smaps_rollup_release,
 917 };
 918
 919 enum clear_refs_types {
 920         CLEAR_REFS_ALL = 1,
 921         CLEAR_REFS_ANON,
 922         CLEAR_REFS_MAPPED,
 923         CLEAR_REFS_SOFT_DIRTY,
 924         CLEAR_REFS_MM_HIWATER_RSS,
 925         CLEAR_REFS_LAST,
 926 };
 927
 928 struct clear_refs_private {
 929         enum clear_refs_types type;
 930 };
 931
 932 #ifdef CONFIG_MEM_SOFT_DIRTY
 933 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 934                 unsigned long addr, pte_t *pte)
 935 {
 936         /*
 937          * The soft-dirty tracker uses #PF-s to catch writes
 938          * to pages, so write-protect the pte as well. See the
 939          * Documentation/admin-guide/mm/soft-dirty.rst for full description
 940          * of how soft-dirty works.
 941          */
 942         pte_t ptent = *pte;
 943
 944         if (pte_present(ptent)) {
 945                 ptent = ptep_modify_prot_start(vma->vm_mm, addr, pte);
 946                 ptent = pte_wrprotect(ptent);
 947                 ptent = pte_clear_soft_dirty(ptent);
 948                 ptep_modify_prot_commit(vma->vm_mm, addr, pte, ptent);
 949         } else if (is_swap_pte(ptent)) {
 950                 ptent = pte_swp_clear_soft_dirty(ptent);
 951                 set_pte_at(vma->vm_mm, addr, pte, ptent);
 952         }
 953 }
 954 #else
 955 static inline void clear_soft_dirty(struct vm_area_struct *vma,
 956                 unsigned long addr, pte_t *pte)
 957 {
 958 }
 959 #endif
 960
 961 #if defined(CONFIG_MEM_SOFT_DIRTY) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
 962 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 963                 unsigned long addr, pmd_t *pmdp)
 964 {
 965         pmd_t old, pmd = *pmdp;
 966
 967         if (pmd_present(pmd)) {
 968                 /* See comment in change_huge_pmd() */
 969                 old = pmdp_invalidate(vma, addr, pmdp);
 970                 if (pmd_dirty(old))
 971                         pmd = pmd_mkdirty(pmd);
 972                 if (pmd_young(old))
 973                         pmd = pmd_mkyoung(pmd);
 974
 975                 pmd = pmd_wrprotect(pmd);
 976                 pmd = pmd_clear_soft_dirty(pmd);
 977
 978                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 979         } else if (is_migration_entry(pmd_to_swp_entry(pmd))) {
 980                 pmd = pmd_swp_clear_soft_dirty(pmd);
 981                 set_pmd_at(vma->vm_mm, addr, pmdp, pmd);
 982         }
 983 }
 984 #else
 985 static inline void clear_soft_dirty_pmd(struct vm_area_struct *vma,
 986                 unsigned long addr, pmd_t *pmdp)
 987 {
 988 }
 989 #endif
 990
 991 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr,
 992                                 unsigned long end, struct mm_walk *walk)
 993 {
 994         struct clear_refs_private *cp = walk->private;
 995         struct vm_area_struct *vma = walk->vma;
 996         pte_t *pte, ptent;
 997         spinlock_t *ptl;
 998         struct page *page;
 999
1000         ptl = pmd_trans_huge_lock(pmd, vma);
1001         if (ptl) {
1002                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1003                         clear_soft_dirty_pmd(vma, addr, pmd);
1004                         goto out;
1005                 }
1006
1007                 if (!pmd_present(*pmd))
1008                         goto out;
1009
1010                 page = pmd_page(*pmd);
1011
1012                 /* Clear accessed and referenced bits. */
1013                 pmdp_test_and_clear_young(vma, addr, pmd);
1014                 test_and_clear_page_young(page);
1015                 ClearPageReferenced(page);
1016 out:
1017                 spin_unlock(ptl);
1018                 return 0;
1019         }
1020
1021         if (pmd_trans_unstable(pmd))
1022                 return 0;
1023
1024         pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
1025         for (; addr != end; pte++, addr += PAGE_SIZE) {
1026                 ptent = *pte;
1027
1028                 if (cp->type == CLEAR_REFS_SOFT_DIRTY) {
1029                         clear_soft_dirty(vma, addr, pte);
1030                         continue;
1031                 }
1032
1033                 if (!pte_present(ptent))
1034                         continue;
1035
1036                 page = vm_normal_page(vma, addr, ptent);
1037                 if (!page)
1038                         continue;
1039
1040                 /* Clear accessed and referenced bits. */
1041                 ptep_test_and_clear_young(vma, addr, pte);
1042                 test_and_clear_page_young(page);
1043                 ClearPageReferenced(page);
1044         }
1045         pte_unmap_unlock(pte - 1, ptl);
1046         cond_resched();
1047         return 0;
1048 }
1049
1050 static int clear_refs_test_walk(unsigned long start, unsigned long end,
1051                                 struct mm_walk *walk)
1052 {
1053         struct clear_refs_private *cp = walk->private;
1054         struct vm_area_struct *vma = walk->vma;
1055
1056         if (vma->vm_flags & VM_PFNMAP)
1057                 return 1;
1058
1059         /*
1060          * Writing 1 to /proc/pid/clear_refs affects all pages.
1061          * Writing 2 to /proc/pid/clear_refs only affects anonymous pages.
1062          * Writing 3 to /proc/pid/clear_refs only affects file mapped pages.
1063          * Writing 4 to /proc/pid/clear_refs affects all pages.
1064          */
1065         if (cp->type == CLEAR_REFS_ANON && vma->vm_file)
1066                 return 1;
1067         if (cp->type == CLEAR_REFS_MAPPED && !vma->vm_file)
1068                 return 1;
1069         return 0;
1070 }
1071
1072 static ssize_t clear_refs_write(struct file *file, const char __user *buf,
1073                                 size_t count, loff_t *ppos)
1074 {
1075         struct task_struct *task;
1076         char buffer[PROC_NUMBUF];
1077         struct mm_struct *mm;
1078         struct vm_area_struct *vma;
1079         enum clear_refs_types type;
1080         struct mmu_gather tlb;
1081         int itype;
1082         int rv;
1083
1084         memset(buffer, 0, sizeof(buffer));
1085         if (count > sizeof(buffer) - 1)
1086                 count = sizeof(buffer) - 1;
1087         if (copy_from_user(buffer, buf, count))
1088                 return -EFAULT;
1089         rv = kstrtoint(strstrip(buffer), 10, &itype);
1090         if (rv < 0)
1091                 return rv;
1092         type = (enum clear_refs_types)itype;
1093         if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST)
1094                 return -EINVAL;
1095
1096         task = get_proc_task(file_inode(file));
1097         if (!task)
1098                 return -ESRCH;
1099         mm = get_task_mm(task);
1100         if (mm) {
1101                 struct mmu_notifier_range range;
1102                 struct clear_refs_private cp = {
1103                         .type = type,
1104                 };
1105                 struct mm_walk clear_refs_walk = {
1106                         .pmd_entry = clear_refs_pte_range,
1107                         .test_walk = clear_refs_test_walk,
1108                         .mm = mm,
1109                         .private = &cp,
1110                 };
1111
1112                 if (type == CLEAR_REFS_MM_HIWATER_RSS) {
1113                         if (down_write_killable(&mm->mmap_sem)) {
1114                                 count = -EINTR;
1115                                 goto out_mm;
1116                         }
1117
1118                         /*
1119                          * Writing 5 to /proc/pid/clear_refs resets the peak
1120                          * resident set size to this mm's current rss value.
1121                          */
1122                         reset_mm_hiwater_rss(mm);
1123                         up_write(&mm->mmap_sem);
1124                         goto out_mm;
1125                 }
1126
1127                 down_read(&mm->mmap_sem);
1128                 tlb_gather_mmu(&tlb, mm, 0, -1);
1129                 if (type == CLEAR_REFS_SOFT_DIRTY) {
1130                         for (vma = mm->mmap; vma; vma = vma->vm_next) {
1131                                 if (!(vma->vm_flags & VM_SOFTDIRTY))
1132                                         continue;
1133                                 up_read(&mm->mmap_sem);
1134                                 if (down_write_killable(&mm->mmap_sem)) {
1135                                         count = -EINTR;
1136                                         goto out_mm;
1137                                 }
1138                                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1139                                         vma->vm_flags &= ~VM_SOFTDIRTY;
1140                                         vma_set_page_prot(vma);
1141                                 }
1142                                 downgrade_write(&mm->mmap_sem);
1143                                 break;
1144                         }
1145
1146                         mmu_notifier_range_init(&range, mm, 0, -1UL);
1147                         mmu_notifier_invalidate_range_start(&range);
1148                 }
1149                 walk_page_range(0, mm->highest_vm_end, &clear_refs_walk);
1150                 if (type == CLEAR_REFS_SOFT_DIRTY)
1151                         mmu_notifier_invalidate_range_end(&range);
1152                 tlb_finish_mmu(&tlb, 0, -1);
1153                 up_read(&mm->mmap_sem);
1154 out_mm:
1155                 mmput(mm);
1156         }
1157         put_task_struct(task);
1158
1159         return count;
1160 }
1161
1162 const struct file_operations proc_clear_refs_operations = {
1163         .write          = clear_refs_write,
1164         .llseek         = noop_llseek,
1165 };
1166
1167 typedef struct {
1168         u64 pme;
1169 } pagemap_entry_t;
1170
1171 struct pagemapread {
1172         int pos, len;           /* units: PM_ENTRY_BYTES, not bytes */
1173         pagemap_entry_t *buffer;
1174         bool show_pfn;
1175 };
1176
1177 #define PAGEMAP_WALK_SIZE       (PMD_SIZE)
1178 #define PAGEMAP_WALK_MASK       (PMD_MASK)
1179
1180 #define PM_ENTRY_BYTES          sizeof(pagemap_entry_t)
1181 #define PM_PFRAME_BITS          55
1182 #define PM_PFRAME_MASK          GENMASK_ULL(PM_PFRAME_BITS - 1, 0)
1183 #define PM_SOFT_DIRTY           BIT_ULL(55)
1184 #define PM_MMAP_EXCLUSIVE       BIT_ULL(56)
1185 #define PM_FILE                 BIT_ULL(61)
1186 #define PM_SWAP                 BIT_ULL(62)
1187 #define PM_PRESENT              BIT_ULL(63)
1188
1189 #define PM_END_OF_BUFFER    1
1190
1191 static inline pagemap_entry_t make_pme(u64 frame, u64 flags)
1192 {
1193         return (pagemap_entry_t) { .pme = (frame & PM_PFRAME_MASK) | flags };
1194 }
1195
1196 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme,
1197                           struct pagemapread *pm)
1198 {
1199         pm->buffer[pm->pos++] = *pme;
1200         if (pm->pos >= pm->len)
1201                 return PM_END_OF_BUFFER;
1202         return 0;
1203 }
1204
1205 static int pagemap_pte_hole(unsigned long start, unsigned long end,
1206                                 struct mm_walk *walk)
1207 {
1208         struct pagemapread *pm = walk->private;
1209         unsigned long addr = start;
1210         int err = 0;
1211
1212         while (addr < end) {
1213                 struct vm_area_struct *vma = find_vma(walk->mm, addr);
1214                 pagemap_entry_t pme = make_pme(0, 0);
1215                 /* End of address space hole, which we mark as non-present. */
1216                 unsigned long hole_end;
1217
1218                 if (vma)
1219                         hole_end = min(end, vma->vm_start);
1220                 else
1221                         hole_end = end;
1222
1223                 for (; addr < hole_end; addr += PAGE_SIZE) {
1224                         err = add_to_pagemap(addr, &pme, pm);
1225                         if (err)
1226                                 goto out;
1227                 }
1228
1229                 if (!vma)
1230                         break;
1231
1232                 /* Addresses in the VMA. */
1233                 if (vma->vm_flags & VM_SOFTDIRTY)
1234                         pme = make_pme(0, PM_SOFT_DIRTY);
1235                 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) {
1236                         err = add_to_pagemap(addr, &pme, pm);
1237                         if (err)
1238                                 goto out;
1239                 }
1240         }
1241 out:
1242         return err;
1243 }
1244
1245 static pagemap_entry_t pte_to_pagemap_entry(struct pagemapread *pm,
1246                 struct vm_area_struct *vma, unsigned long addr, pte_t pte)
1247 {
1248         u64 frame = 0, flags = 0;
1249         struct page *page = NULL;
1250
1251         if (pte_present(pte)) {
1252                 if (pm->show_pfn)
1253                         frame = pte_pfn(pte);
1254                 flags |= PM_PRESENT;
1255                 page = _vm_normal_page(vma, addr, pte, true);
1256                 if (pte_soft_dirty(pte))
1257                         flags |= PM_SOFT_DIRTY;
1258         } else if (is_swap_pte(pte)) {
1259                 swp_entry_t entry;
1260                 if (pte_swp_soft_dirty(pte))
1261                         flags |= PM_SOFT_DIRTY;
1262                 entry = pte_to_swp_entry(pte);
1263                 if (pm->show_pfn)
1264                         frame = swp_type(entry) |
1265                                 (swp_offset(entry) << MAX_SWAPFILES_SHIFT);
1266                 flags |= PM_SWAP;
1267                 if (is_migration_entry(entry))
1268                         page = migration_entry_to_page(entry);
1269
1270                 if (is_device_private_entry(entry))
1271                         page = device_private_entry_to_page(entry);
1272         }
1273
1274         if (page && !PageAnon(page))
1275                 flags |= PM_FILE;
1276         if (page && page_mapcount(page) == 1)
1277                 flags |= PM_MMAP_EXCLUSIVE;
1278         if (vma->vm_flags & VM_SOFTDIRTY)
1279                 flags |= PM_SOFT_DIRTY;
1280
1281         return make_pme(frame, flags);
1282 }
1283
1284 static int pagemap_pmd_range(pmd_t *pmdp, unsigned long addr, unsigned long end,
1285                              struct mm_walk *walk)
1286 {
1287         struct vm_area_struct *vma = walk->vma;
1288         struct pagemapread *pm = walk->private;
1289         spinlock_t *ptl;
1290         pte_t *pte, *orig_pte;
1291         int err = 0;
1292
1293 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1294         ptl = pmd_trans_huge_lock(pmdp, vma);
1295         if (ptl) {
1296                 u64 flags = 0, frame = 0;
1297                 pmd_t pmd = *pmdp;
1298                 struct page *page = NULL;
1299
1300                 if (vma->vm_flags & VM_SOFTDIRTY)
1301                         flags |= PM_SOFT_DIRTY;
1302
1303                 if (pmd_present(pmd)) {
1304                         page = pmd_page(pmd);
1305
1306                         flags |= PM_PRESENT;
1307                         if (pmd_soft_dirty(pmd))
1308                                 flags |= PM_SOFT_DIRTY;
1309                         if (pm->show_pfn)
1310                                 frame = pmd_pfn(pmd) +
1311                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1312                 }
1313 #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION
1314                 else if (is_swap_pmd(pmd)) {
1315                         swp_entry_t entry = pmd_to_swp_entry(pmd);
1316                         unsigned long offset;
1317
1318                         if (pm->show_pfn) {
1319                                 offset = swp_offset(entry) +
1320                                         ((addr & ~PMD_MASK) >> PAGE_SHIFT);
1321                                 frame = swp_type(entry) |
1322                                         (offset << MAX_SWAPFILES_SHIFT);
1323                         }
1324                         flags |= PM_SWAP;
1325                         if (pmd_swp_soft_dirty(pmd))
1326                                 flags |= PM_SOFT_DIRTY;
1327                         VM_BUG_ON(!is_pmd_migration_entry(pmd));
1328                         page = migration_entry_to_page(entry);
1329                 }
1330 #endif
1331
1332                 if (page && page_mapcount(page) == 1)
1333                         flags |= PM_MMAP_EXCLUSIVE;
1334
1335                 for (; addr != end; addr += PAGE_SIZE) {
1336                         pagemap_entry_t pme = make_pme(frame, flags);
1337
1338                         err = add_to_pagemap(addr, &pme, pm);
1339                         if (err)
1340                                 break;
1341                         if (pm->show_pfn) {
1342                                 if (flags & PM_PRESENT)
1343                                         frame++;
1344                                 else if (flags & PM_SWAP)
1345                                         frame += (1 << MAX_SWAPFILES_SHIFT);
1346                         }
1347                 }
1348                 spin_unlock(ptl);
1349                 return err;
1350         }
1351
1352         if (pmd_trans_unstable(pmdp))
1353                 return 0;
1354 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1355
1356         /*
1357          * We can assume that @vma always points to a valid one and @end never
1358          * goes beyond vma->vm_end.
1359          */
1360         orig_pte = pte = pte_offset_map_lock(walk->mm, pmdp, addr, &ptl);
1361         for (; addr < end; pte++, addr += PAGE_SIZE) {
1362                 pagemap_entry_t pme;
1363
1364                 pme = pte_to_pagemap_entry(pm, vma, addr, *pte);
1365                 err = add_to_pagemap(addr, &pme, pm);
1366                 if (err)
1367                         break;
1368         }
1369         pte_unmap_unlock(orig_pte, ptl);
1370
1371         cond_resched();
1372
1373         return err;
1374 }
1375
1376 #ifdef CONFIG_HUGETLB_PAGE
1377 /* This function walks within one hugetlb entry in the single call */
1378 static int pagemap_hugetlb_range(pte_t *ptep, unsigned long hmask,
1379                                  unsigned long addr, unsigned long end,
1380                                  struct mm_walk *walk)
1381 {
1382         struct pagemapread *pm = walk->private;
1383         struct vm_area_struct *vma = walk->vma;
1384         u64 flags = 0, frame = 0;
1385         int err = 0;
1386         pte_t pte;
1387
1388         if (vma->vm_flags & VM_SOFTDIRTY)
1389                 flags |= PM_SOFT_DIRTY;
1390
1391         pte = huge_ptep_get(ptep);
1392         if (pte_present(pte)) {
1393                 struct page *page = pte_page(pte);
1394
1395                 if (!PageAnon(page))
1396                         flags |= PM_FILE;
1397
1398                 if (page_mapcount(page) == 1)
1399                         flags |= PM_MMAP_EXCLUSIVE;
1400
1401                 flags |= PM_PRESENT;
1402                 if (pm->show_pfn)
1403                         frame = pte_pfn(pte) +
1404                                 ((addr & ~hmask) >> PAGE_SHIFT);
1405         }
1406
1407         for (; addr != end; addr += PAGE_SIZE) {
1408                 pagemap_entry_t pme = make_pme(frame, flags);
1409
1410                 err = add_to_pagemap(addr, &pme, pm);
1411                 if (err)
1412                         return err;
1413                 if (pm->show_pfn && (flags & PM_PRESENT))
1414                         frame++;
1415         }
1416
1417         cond_resched();
1418
1419         return err;
1420 }
1421 #endif /* HUGETLB_PAGE */
1422
1423 /*
1424  * /proc/pid/pagemap - an array mapping virtual pages to pfns
1425  *
1426  * For each page in the address space, this file contains one 64-bit entry
1427  * consisting of the following:
1428  *
1429  * Bits 0-54  page frame number (PFN) if present
1430  * Bits 0-4   swap type if swapped
1431  * Bits 5-54  swap offset if swapped
1432  * Bit  55    pte is soft-dirty (see Documentation/admin-guide/mm/soft-dirty.rst)
1433  * Bit  56    page exclusively mapped
1434  * Bits 57-60 zero
1435  * Bit  61    page is file-page or shared-anon
1436  * Bit  62    page swapped
1437  * Bit  63    page present
1438  *
1439  * If the page is not present but in swap, then the PFN contains an
1440  * encoding of the swap file number and the page's offset into the
1441  * swap. Unmapped pages return a null PFN. This allows determining
1442  * precisely which pages are mapped (or in swap) and comparing mapped
1443  * pages between processes.
1444  *
1445  * Efficient users of this interface will use /proc/pid/maps to
1446  * determine which areas of memory are actually mapped and llseek to
1447  * skip over unmapped regions.
1448  */
1449 static ssize_t pagemap_read(struct file *file, char __user *buf,
1450                             size_t count, loff_t *ppos)
1451 {
1452         struct mm_struct *mm = file->private_data;
1453         struct pagemapread pm;
1454         struct mm_walk pagemap_walk = {};
1455         unsigned long src;
1456         unsigned long svpfn;
1457         unsigned long start_vaddr;
1458         unsigned long end_vaddr;
1459         int ret = 0, copied = 0;
1460
1461         if (!mm || !mmget_not_zero(mm))
1462                 goto out;
1463
1464         ret = -EINVAL;
1465         /* file position must be aligned */
1466         if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES))
1467                 goto out_mm;
1468
1469         ret = 0;
1470         if (!count)
1471                 goto out_mm;
1472
1473         /* do not disclose physical addresses: attack vector */
1474         pm.show_pfn = file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN);
1475
1476         pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT);
1477         pm.buffer = kmalloc_array(pm.len, PM_ENTRY_BYTES, GFP_KERNEL);
1478         ret = -ENOMEM;
1479         if (!pm.buffer)
1480                 goto out_mm;
1481
1482         pagemap_walk.pmd_entry = pagemap_pmd_range;
1483         pagemap_walk.pte_hole = pagemap_pte_hole;
1484 #ifdef CONFIG_HUGETLB_PAGE
1485         pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
1486 #endif
1487         pagemap_walk.mm = mm;
1488         pagemap_walk.private = &pm;
1489
1490         src = *ppos;
1491         svpfn = src / PM_ENTRY_BYTES;
1492         start_vaddr = svpfn << PAGE_SHIFT;
1493         end_vaddr = mm->task_size;
1494
1495         /* watch out for wraparound */
1496         if (svpfn > mm->task_size >> PAGE_SHIFT)
1497                 start_vaddr = end_vaddr;
1498
1499         /*
1500          * The odds are that this will stop walking way
1501          * before end_vaddr, because the length of the
1502          * user buffer is tracked in "pm", and the walk
1503          * will stop when we hit the end of the buffer.
1504          */
1505         ret = 0;
1506         while (count && (start_vaddr < end_vaddr)) {
1507                 int len;
1508                 unsigned long end;
1509
1510                 pm.pos = 0;
1511                 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK;
1512                 /* overflow ? */
1513                 if (end < start_vaddr || end > end_vaddr)
1514                         end = end_vaddr;
1515                 down_read(&mm->mmap_sem);
1516                 ret = walk_page_range(start_vaddr, end, &pagemap_walk);
1517                 up_read(&mm->mmap_sem);
1518                 start_vaddr = end;
1519
1520                 len = min(count, PM_ENTRY_BYTES * pm.pos);
1521                 if (copy_to_user(buf, pm.buffer, len)) {
1522                         ret = -EFAULT;
1523                         goto out_free;
1524                 }
1525                 copied += len;
1526                 buf += len;
1527                 count -= len;
1528         }
1529         *ppos += copied;
1530         if (!ret || ret == PM_END_OF_BUFFER)
1531                 ret = copied;
1532
1533 out_free:
1534         kfree(pm.buffer);
1535 out_mm:
1536         mmput(mm);
1537 out:
1538         return ret;
1539 }
1540
1541 static int pagemap_open(struct inode *inode, struct file *file)
1542 {
1543         struct mm_struct *mm;
1544
1545         mm = proc_mem_open(inode, PTRACE_MODE_READ);
1546         if (IS_ERR(mm))
1547                 return PTR_ERR(mm);
1548         file->private_data = mm;
1549         return 0;
1550 }
1551
1552 static int pagemap_release(struct inode *inode, struct file *file)
1553 {
1554         struct mm_struct *mm = file->private_data;
1555
1556         if (mm)
1557                 mmdrop(mm);
1558         return 0;
1559 }
1560
1561 const struct file_operations proc_pagemap_operations = {
1562         .llseek         = mem_lseek, /* borrow this */
1563         .read           = pagemap_read,
1564         .open           = pagemap_open,
1565         .release        = pagemap_release,
1566 };
1567 #endif /* CONFIG_PROC_PAGE_MONITOR */
1568
1569 #ifdef CONFIG_NUMA
1570
1571 struct numa_maps {
1572         unsigned long pages;
1573         unsigned long anon;
1574         unsigned long active;
1575         unsigned long writeback;
1576         unsigned long mapcount_max;
1577         unsigned long dirty;
1578         unsigned long swapcache;
1579         unsigned long node[MAX_NUMNODES];
1580 };
1581
1582 struct numa_maps_private {
1583         struct proc_maps_private proc_maps;
1584         struct numa_maps md;
1585 };
1586
1587 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty,
1588                         unsigned long nr_pages)
1589 {
1590         int count = page_mapcount(page);
1591
1592         md->pages += nr_pages;
1593         if (pte_dirty || PageDirty(page))
1594                 md->dirty += nr_pages;
1595
1596         if (PageSwapCache(page))
1597                 md->swapcache += nr_pages;
1598
1599         if (PageActive(page) || PageUnevictable(page))
1600                 md->active += nr_pages;
1601
1602         if (PageWriteback(page))
1603                 md->writeback += nr_pages;
1604
1605         if (PageAnon(page))
1606                 md->anon += nr_pages;
1607
1608         if (count > md->mapcount_max)
1609                 md->mapcount_max = count;
1610
1611         md->node[page_to_nid(page)] += nr_pages;
1612 }
1613
1614 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma,
1615                 unsigned long addr)
1616 {
1617         struct page *page;
1618         int nid;
1619
1620         if (!pte_present(pte))
1621                 return NULL;
1622
1623         page = vm_normal_page(vma, addr, pte);
1624         if (!page)
1625                 return NULL;
1626
1627         if (PageReserved(page))
1628                 return NULL;
1629
1630         nid = page_to_nid(page);
1631         if (!node_isset(nid, node_states[N_MEMORY]))
1632                 return NULL;
1633
1634         return page;
1635 }
1636
1637 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1638 static struct page *can_gather_numa_stats_pmd(pmd_t pmd,
1639                                               struct vm_area_struct *vma,
1640                                               unsigned long addr)
1641 {
1642         struct page *page;
1643         int nid;
1644
1645         if (!pmd_present(pmd))
1646                 return NULL;
1647
1648         page = vm_normal_page_pmd(vma, addr, pmd);
1649         if (!page)
1650                 return NULL;
1651
1652         if (PageReserved(page))
1653                 return NULL;
1654
1655         nid = page_to_nid(page);
1656         if (!node_isset(nid, node_states[N_MEMORY]))
1657                 return NULL;
1658
1659         return page;
1660 }
1661 #endif
1662
1663 static int gather_pte_stats(pmd_t *pmd, unsigned long addr,
1664                 unsigned long end, struct mm_walk *walk)
1665 {
1666         struct numa_maps *md = walk->private;
1667         struct vm_area_struct *vma = walk->vma;
1668         spinlock_t *ptl;
1669         pte_t *orig_pte;
1670         pte_t *pte;
1671
1672 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1673         ptl = pmd_trans_huge_lock(pmd, vma);
1674         if (ptl) {
1675                 struct page *page;
1676
1677                 page = can_gather_numa_stats_pmd(*pmd, vma, addr);
1678                 if (page)
1679                         gather_stats(page, md, pmd_dirty(*pmd),
1680                                      HPAGE_PMD_SIZE/PAGE_SIZE);
1681                 spin_unlock(ptl);
1682                 return 0;
1683         }
1684
1685         if (pmd_trans_unstable(pmd))
1686                 return 0;
1687 #endif
1688         orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl);
1689         do {
1690                 struct page *page = can_gather_numa_stats(*pte, vma, addr);
1691                 if (!page)
1692                         continue;
1693                 gather_stats(page, md, pte_dirty(*pte), 1);
1694
1695         } while (pte++, addr += PAGE_SIZE, addr != end);
1696         pte_unmap_unlock(orig_pte, ptl);
1697         cond_resched();
1698         return 0;
1699 }
1700 #ifdef CONFIG_HUGETLB_PAGE
1701 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1702                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1703 {
1704         pte_t huge_pte = huge_ptep_get(pte);
1705         struct numa_maps *md;
1706         struct page *page;
1707
1708         if (!pte_present(huge_pte))
1709                 return 0;
1710
1711         page = pte_page(huge_pte);
1712         if (!page)
1713                 return 0;
1714
1715         md = walk->private;
1716         gather_stats(page, md, pte_dirty(huge_pte), 1);
1717         return 0;
1718 }
1719
1720 #else
1721 static int gather_hugetlb_stats(pte_t *pte, unsigned long hmask,
1722                 unsigned long addr, unsigned long end, struct mm_walk *walk)
1723 {
1724         return 0;
1725 }
1726 #endif
1727
1728 /*
1729  * Display pages allocated per node and memory policy via /proc.
1730  */
1731 static int show_numa_map(struct seq_file *m, void *v)
1732 {
1733         struct numa_maps_private *numa_priv = m->private;
1734         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
1735         struct vm_area_struct *vma = v;
1736         struct numa_maps *md = &numa_priv->md;
1737         struct file *file = vma->vm_file;
1738         struct mm_struct *mm = vma->vm_mm;
1739         struct mm_walk walk = {
1740                 .hugetlb_entry = gather_hugetlb_stats,
1741                 .pmd_entry = gather_pte_stats,
1742                 .private = md,
1743                 .mm = mm,
1744         };
1745         struct mempolicy *pol;
1746         char buffer[64];
1747         int nid;
1748
1749         if (!mm)
1750                 return 0;
1751
1752         /* Ensure we start with an empty set of numa_maps statistics. */
1753         memset(md, 0, sizeof(*md));
1754
1755         pol = __get_vma_policy(vma, vma->vm_start);
1756         if (pol) {
1757                 mpol_to_str(buffer, sizeof(buffer), pol);
1758                 mpol_cond_put(pol);
1759         } else {
1760                 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy);
1761         }
1762
1763         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
1764
1765         if (file) {
1766                 seq_puts(m, " file=");
1767                 seq_file_path(m, file, "\n\t= ");
1768         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
1769                 seq_puts(m, " heap");
1770         } else if (is_stack(vma)) {
1771                 seq_puts(m, " stack");
1772         }
1773
1774         if (is_vm_hugetlb_page(vma))
1775                 seq_puts(m, " huge");
1776
1777         /* mmap_sem is held by m_start */
1778         walk_page_vma(vma, &walk);
1779
1780         if (!md->pages)
1781                 goto out;
1782
1783         if (md->anon)
1784                 seq_printf(m, " anon=%lu", md->anon);
1785
1786         if (md->dirty)
1787                 seq_printf(m, " dirty=%lu", md->dirty);
1788
1789         if (md->pages != md->anon && md->pages != md->dirty)
1790                 seq_printf(m, " mapped=%lu", md->pages);
1791
1792         if (md->mapcount_max > 1)
1793                 seq_printf(m, " mapmax=%lu", md->mapcount_max);
1794
1795         if (md->swapcache)
1796                 seq_printf(m, " swapcache=%lu", md->swapcache);
1797
1798         if (md->active < md->pages && !is_vm_hugetlb_page(vma))
1799                 seq_printf(m, " active=%lu", md->active);
1800
1801         if (md->writeback)
1802                 seq_printf(m, " writeback=%lu", md->writeback);
1803
1804         for_each_node_state(nid, N_MEMORY)
1805                 if (md->node[nid])
1806                         seq_printf(m, " N%d=%lu", nid, md->node[nid]);
1807
1808         seq_printf(m, " kernelpagesize_kB=%lu", vma_kernel_pagesize(vma) >> 10);
1809 out:
1810         seq_putc(m, '\n');
1811         m_cache_vma(m, vma);
1812         return 0;
1813 }
1814
1815 static const struct seq_operations proc_pid_numa_maps_op = {
1816         .start  = m_start,
1817         .next   = m_next,
1818         .stop   = m_stop,
1819         .show   = show_numa_map,
1820 };
1821
1822 static int pid_numa_maps_open(struct inode *inode, struct file *file)
1823 {
1824         return proc_maps_open(inode, file, &proc_pid_numa_maps_op,
1825                                 sizeof(struct numa_maps_private));
1826 }
1827
1828 const struct file_operations proc_pid_numa_maps_operations = {
1829         .open           = pid_numa_maps_open,
1830         .read           = seq_read,
1831         .llseek         = seq_lseek,
1832         .release        = proc_map_release,
1833 };
1834
1835 #endif /* CONFIG_NUMA */