linux/fs/hugetlbfs/inode.c
<<
>>
Prefs
   1/*
   2 * hugetlbpage-backed filesystem.  Based on ramfs.
   3 *
   4 * Nadia Yvette Chambers, 2002
   5 *
   6 * Copyright (C) 2002 Linus Torvalds.
   7 * License: GPL
   8 */
   9
  10#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  11
  12#include <linux/thread_info.h>
  13#include <asm/current.h>
  14#include <linux/sched/signal.h>         /* remove ASAP */
  15#include <linux/falloc.h>
  16#include <linux/fs.h>
  17#include <linux/mount.h>
  18#include <linux/file.h>
  19#include <linux/kernel.h>
  20#include <linux/writeback.h>
  21#include <linux/pagemap.h>
  22#include <linux/highmem.h>
  23#include <linux/init.h>
  24#include <linux/string.h>
  25#include <linux/capability.h>
  26#include <linux/ctype.h>
  27#include <linux/backing-dev.h>
  28#include <linux/hugetlb.h>
  29#include <linux/pagevec.h>
  30#include <linux/fs_parser.h>
  31#include <linux/mman.h>
  32#include <linux/slab.h>
  33#include <linux/dnotify.h>
  34#include <linux/statfs.h>
  35#include <linux/security.h>
  36#include <linux/magic.h>
  37#include <linux/migrate.h>
  38#include <linux/uio.h>
  39
  40#include <linux/uaccess.h>
  41#include <linux/sched/mm.h>
  42
  43static const struct super_operations hugetlbfs_ops;
  44static const struct address_space_operations hugetlbfs_aops;
  45const struct file_operations hugetlbfs_file_operations;
  46static const struct inode_operations hugetlbfs_dir_inode_operations;
  47static const struct inode_operations hugetlbfs_inode_operations;
  48
  49enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT };
  50
  51struct hugetlbfs_fs_context {
  52        struct hstate           *hstate;
  53        unsigned long long      max_size_opt;
  54        unsigned long long      min_size_opt;
  55        long                    max_hpages;
  56        long                    nr_inodes;
  57        long                    min_hpages;
  58        enum hugetlbfs_size_type max_val_type;
  59        enum hugetlbfs_size_type min_val_type;
  60        kuid_t                  uid;
  61        kgid_t                  gid;
  62        umode_t                 mode;
  63};
  64
  65int sysctl_hugetlb_shm_group;
  66
  67enum hugetlb_param {
  68        Opt_gid,
  69        Opt_min_size,
  70        Opt_mode,
  71        Opt_nr_inodes,
  72        Opt_pagesize,
  73        Opt_size,
  74        Opt_uid,
  75};
  76
  77static const struct fs_parameter_spec hugetlb_fs_parameters[] = {
  78        fsparam_u32   ("gid",           Opt_gid),
  79        fsparam_string("min_size",      Opt_min_size),
  80        fsparam_u32oct("mode",          Opt_mode),
  81        fsparam_string("nr_inodes",     Opt_nr_inodes),
  82        fsparam_string("pagesize",      Opt_pagesize),
  83        fsparam_string("size",          Opt_size),
  84        fsparam_u32   ("uid",           Opt_uid),
  85        {}
  86};
  87
  88#ifdef CONFIG_NUMA
  89static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
  90                                        struct inode *inode, pgoff_t index)
  91{
  92        vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy,
  93                                                        index);
  94}
  95
  96static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
  97{
  98        mpol_cond_put(vma->vm_policy);
  99}
 100#else
 101static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
 102                                        struct inode *inode, pgoff_t index)
 103{
 104}
 105
 106static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
 107{
 108}
 109#endif
 110
 111static void huge_pagevec_release(struct pagevec *pvec)
 112{
 113        int i;
 114
 115        for (i = 0; i < pagevec_count(pvec); ++i)
 116                put_page(pvec->pages[i]);
 117
 118        pagevec_reinit(pvec);
 119}
 120
 121/*
 122 * Mask used when checking the page offset value passed in via system
 123 * calls.  This value will be converted to a loff_t which is signed.
 124 * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the
 125 * value.  The extra bit (- 1 in the shift value) is to take the sign
 126 * bit into account.
 127 */
 128#define PGOFF_LOFFT_MAX \
 129        (((1UL << (PAGE_SHIFT + 1)) - 1) <<  (BITS_PER_LONG - (PAGE_SHIFT + 1)))
 130
 131static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 132{
 133        struct inode *inode = file_inode(file);
 134        struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 135        loff_t len, vma_len;
 136        int ret;
 137        struct hstate *h = hstate_file(file);
 138
 139        /*
 140         * vma address alignment (but not the pgoff alignment) has
 141         * already been checked by prepare_hugepage_range.  If you add
 142         * any error returns here, do so after setting VM_HUGETLB, so
 143         * is_vm_hugetlb_page tests below unmap_region go the right
 144         * way when do_mmap unwinds (may be important on powerpc
 145         * and ia64).
 146         */
 147        vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
 148        vma->vm_ops = &hugetlb_vm_ops;
 149
 150        ret = seal_check_future_write(info->seals, vma);
 151        if (ret)
 152                return ret;
 153
 154        /*
 155         * page based offset in vm_pgoff could be sufficiently large to
 156         * overflow a loff_t when converted to byte offset.  This can
 157         * only happen on architectures where sizeof(loff_t) ==
 158         * sizeof(unsigned long).  So, only check in those instances.
 159         */
 160        if (sizeof(unsigned long) == sizeof(loff_t)) {
 161                if (vma->vm_pgoff & PGOFF_LOFFT_MAX)
 162                        return -EINVAL;
 163        }
 164
 165        /* must be huge page aligned */
 166        if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
 167                return -EINVAL;
 168
 169        vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 170        len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 171        /* check for overflow */
 172        if (len < vma_len)
 173                return -EINVAL;
 174
 175        inode_lock(inode);
 176        file_accessed(file);
 177
 178        ret = -ENOMEM;
 179        if (!hugetlb_reserve_pages(inode,
 180                                vma->vm_pgoff >> huge_page_order(h),
 181                                len >> huge_page_shift(h), vma,
 182                                vma->vm_flags))
 183                goto out;
 184
 185        ret = 0;
 186        if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 187                i_size_write(inode, len);
 188out:
 189        inode_unlock(inode);
 190
 191        return ret;
 192}
 193
 194/*
 195 * Called under mmap_write_lock(mm).
 196 */
 197
 198#ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 199static unsigned long
 200hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr,
 201                unsigned long len, unsigned long pgoff, unsigned long flags)
 202{
 203        struct hstate *h = hstate_file(file);
 204        struct vm_unmapped_area_info info;
 205
 206        info.flags = 0;
 207        info.length = len;
 208        info.low_limit = current->mm->mmap_base;
 209        info.high_limit = TASK_SIZE;
 210        info.align_mask = PAGE_MASK & ~huge_page_mask(h);
 211        info.align_offset = 0;
 212        return vm_unmapped_area(&info);
 213}
 214
 215static unsigned long
 216hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr,
 217                unsigned long len, unsigned long pgoff, unsigned long flags)
 218{
 219        struct hstate *h = hstate_file(file);
 220        struct vm_unmapped_area_info info;
 221
 222        info.flags = VM_UNMAPPED_AREA_TOPDOWN;
 223        info.length = len;
 224        info.low_limit = max(PAGE_SIZE, mmap_min_addr);
 225        info.high_limit = current->mm->mmap_base;
 226        info.align_mask = PAGE_MASK & ~huge_page_mask(h);
 227        info.align_offset = 0;
 228        addr = vm_unmapped_area(&info);
 229
 230        /*
 231         * A failed mmap() very likely causes application failure,
 232         * so fall back to the bottom-up function here. This scenario
 233         * can happen with large stack limits and large mmap()
 234         * allocations.
 235         */
 236        if (unlikely(offset_in_page(addr))) {
 237                VM_BUG_ON(addr != -ENOMEM);
 238                info.flags = 0;
 239                info.low_limit = current->mm->mmap_base;
 240                info.high_limit = TASK_SIZE;
 241                addr = vm_unmapped_area(&info);
 242        }
 243
 244        return addr;
 245}
 246
 247static unsigned long
 248hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 249                unsigned long len, unsigned long pgoff, unsigned long flags)
 250{
 251        struct mm_struct *mm = current->mm;
 252        struct vm_area_struct *vma;
 253        struct hstate *h = hstate_file(file);
 254
 255        if (len & ~huge_page_mask(h))
 256                return -EINVAL;
 257        if (len > TASK_SIZE)
 258                return -ENOMEM;
 259
 260        if (flags & MAP_FIXED) {
 261                if (prepare_hugepage_range(file, addr, len))
 262                        return -EINVAL;
 263                return addr;
 264        }
 265
 266        if (addr) {
 267                addr = ALIGN(addr, huge_page_size(h));
 268                vma = find_vma(mm, addr);
 269                if (TASK_SIZE - len >= addr &&
 270                    (!vma || addr + len <= vm_start_gap(vma)))
 271                        return addr;
 272        }
 273
 274        /*
 275         * Use mm->get_unmapped_area value as a hint to use topdown routine.
 276         * If architectures have special needs, they should define their own
 277         * version of hugetlb_get_unmapped_area.
 278         */
 279        if (mm->get_unmapped_area == arch_get_unmapped_area_topdown)
 280                return hugetlb_get_unmapped_area_topdown(file, addr, len,
 281                                pgoff, flags);
 282        return hugetlb_get_unmapped_area_bottomup(file, addr, len,
 283                        pgoff, flags);
 284}
 285#endif
 286
 287static size_t
 288hugetlbfs_read_actor(struct page *page, unsigned long offset,
 289                        struct iov_iter *to, unsigned long size)
 290{
 291        size_t copied = 0;
 292        int i, chunksize;
 293
 294        /* Find which 4k chunk and offset with in that chunk */
 295        i = offset >> PAGE_SHIFT;
 296        offset = offset & ~PAGE_MASK;
 297
 298        while (size) {
 299                size_t n;
 300                chunksize = PAGE_SIZE;
 301                if (offset)
 302                        chunksize -= offset;
 303                if (chunksize > size)
 304                        chunksize = size;
 305                n = copy_page_to_iter(&page[i], offset, chunksize, to);
 306                copied += n;
 307                if (n != chunksize)
 308                        return copied;
 309                offset = 0;
 310                size -= chunksize;
 311                i++;
 312        }
 313        return copied;
 314}
 315
 316/*
 317 * Support for read() - Find the page attached to f_mapping and copy out the
 318 * data. Its *very* similar to generic_file_buffered_read(), we can't use that
 319 * since it has PAGE_SIZE assumptions.
 320 */
 321static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
 322{
 323        struct file *file = iocb->ki_filp;
 324        struct hstate *h = hstate_file(file);
 325        struct address_space *mapping = file->f_mapping;
 326        struct inode *inode = mapping->host;
 327        unsigned long index = iocb->ki_pos >> huge_page_shift(h);
 328        unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
 329        unsigned long end_index;
 330        loff_t isize;
 331        ssize_t retval = 0;
 332
 333        while (iov_iter_count(to)) {
 334                struct page *page;
 335                size_t nr, copied;
 336
 337                /* nr is the maximum number of bytes to copy from this page */
 338                nr = huge_page_size(h);
 339                isize = i_size_read(inode);
 340                if (!isize)
 341                        break;
 342                end_index = (isize - 1) >> huge_page_shift(h);
 343                if (index > end_index)
 344                        break;
 345                if (index == end_index) {
 346                        nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 347                        if (nr <= offset)
 348                                break;
 349                }
 350                nr = nr - offset;
 351
 352                /* Find the page */
 353                page = find_lock_page(mapping, index);
 354                if (unlikely(page == NULL)) {
 355                        /*
 356                         * We have a HOLE, zero out the user-buffer for the
 357                         * length of the hole or request.
 358                         */
 359                        copied = iov_iter_zero(nr, to);
 360                } else {
 361                        unlock_page(page);
 362
 363                        /*
 364                         * We have the page, copy it to user space buffer.
 365                         */
 366                        copied = hugetlbfs_read_actor(page, offset, to, nr);
 367                        put_page(page);
 368                }
 369                offset += copied;
 370                retval += copied;
 371                if (copied != nr && iov_iter_count(to)) {
 372                        if (!retval)
 373                                retval = -EFAULT;
 374                        break;
 375                }
 376                index += offset >> huge_page_shift(h);
 377                offset &= ~huge_page_mask(h);
 378        }
 379        iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
 380        return retval;
 381}
 382
 383static int hugetlbfs_write_begin(struct file *file,
 384                        struct address_space *mapping,
 385                        loff_t pos, unsigned len, unsigned flags,
 386                        struct page **pagep, void **fsdata)
 387{
 388        return -EINVAL;
 389}
 390
 391static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 392                        loff_t pos, unsigned len, unsigned copied,
 393                        struct page *page, void *fsdata)
 394{
 395        BUG();
 396        return -EINVAL;
 397}
 398
 399static void remove_huge_page(struct page *page)
 400{
 401        ClearPageDirty(page);
 402        ClearPageUptodate(page);
 403        delete_from_page_cache(page);
 404}
 405
 406static void
 407hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end)
 408{
 409        struct vm_area_struct *vma;
 410
 411        /*
 412         * end == 0 indicates that the entire range after
 413         * start should be unmapped.
 414         */
 415        vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) {
 416                unsigned long v_offset;
 417                unsigned long v_end;
 418
 419                /*
 420                 * Can the expression below overflow on 32-bit arches?
 421                 * No, because the interval tree returns us only those vmas
 422                 * which overlap the truncated area starting at pgoff,
 423                 * and no vma on a 32-bit arch can span beyond the 4GB.
 424                 */
 425                if (vma->vm_pgoff < start)
 426                        v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT;
 427                else
 428                        v_offset = 0;
 429
 430                if (!end)
 431                        v_end = vma->vm_end;
 432                else {
 433                        v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT)
 434                                                        + vma->vm_start;
 435                        if (v_end > vma->vm_end)
 436                                v_end = vma->vm_end;
 437                }
 438
 439                unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end,
 440                                                                        NULL);
 441        }
 442}
 443
 444/*
 445 * remove_inode_hugepages handles two distinct cases: truncation and hole
 446 * punch.  There are subtle differences in operation for each case.
 447 *
 448 * truncation is indicated by end of range being LLONG_MAX
 449 *      In this case, we first scan the range and release found pages.
 450 *      After releasing pages, hugetlb_unreserve_pages cleans up region/reserve
 451 *      maps and global counts.  Page faults can not race with truncation
 452 *      in this routine.  hugetlb_no_page() holds i_mmap_rwsem and prevents
 453 *      page faults in the truncated range by checking i_size.  i_size is
 454 *      modified while holding i_mmap_rwsem.
 455 * hole punch is indicated if end is not LLONG_MAX
 456 *      In the hole punch case we scan the range and release found pages.
 457 *      Only when releasing a page is the associated region/reserve map
 458 *      deleted.  The region/reserve map for ranges without associated
 459 *      pages are not modified.  Page faults can race with hole punch.
 460 *      This is indicated if we find a mapped page.
 461 * Note: If the passed end of range value is beyond the end of file, but
 462 * not LLONG_MAX this routine still performs a hole punch operation.
 463 */
 464static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
 465                                   loff_t lend)
 466{
 467        struct hstate *h = hstate_inode(inode);
 468        struct address_space *mapping = &inode->i_data;
 469        const pgoff_t start = lstart >> huge_page_shift(h);
 470        const pgoff_t end = lend >> huge_page_shift(h);
 471        struct pagevec pvec;
 472        pgoff_t next, index;
 473        int i, freed = 0;
 474        bool truncate_op = (lend == LLONG_MAX);
 475
 476        pagevec_init(&pvec);
 477        next = start;
 478        while (next < end) {
 479                /*
 480                 * When no more pages are found, we are done.
 481                 */
 482                if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1))
 483                        break;
 484
 485                for (i = 0; i < pagevec_count(&pvec); ++i) {
 486                        struct page *page = pvec.pages[i];
 487                        u32 hash = 0;
 488
 489                        index = page->index;
 490                        if (!truncate_op) {
 491                                /*
 492                                 * Only need to hold the fault mutex in the
 493                                 * hole punch case.  This prevents races with
 494                                 * page faults.  Races are not possible in the
 495                                 * case of truncation.
 496                                 */
 497                                hash = hugetlb_fault_mutex_hash(mapping, index);
 498                                mutex_lock(&hugetlb_fault_mutex_table[hash]);
 499                        }
 500
 501                        /*
 502                         * If page is mapped, it was faulted in after being
 503                         * unmapped in caller.  Unmap (again) now after taking
 504                         * the fault mutex.  The mutex will prevent faults
 505                         * until we finish removing the page.
 506                         *
 507                         * This race can only happen in the hole punch case.
 508                         * Getting here in a truncate operation is a bug.
 509                         */
 510                        if (unlikely(page_mapped(page))) {
 511                                BUG_ON(truncate_op);
 512
 513                                mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 514                                i_mmap_lock_write(mapping);
 515                                mutex_lock(&hugetlb_fault_mutex_table[hash]);
 516                                hugetlb_vmdelete_list(&mapping->i_mmap,
 517                                        index * pages_per_huge_page(h),
 518                                        (index + 1) * pages_per_huge_page(h));
 519                                i_mmap_unlock_write(mapping);
 520                        }
 521
 522                        lock_page(page);
 523                        /*
 524                         * We must free the huge page and remove from page
 525                         * cache (remove_huge_page) BEFORE removing the
 526                         * region/reserve map (hugetlb_unreserve_pages).  In
 527                         * rare out of memory conditions, removal of the
 528                         * region/reserve map could fail. Correspondingly,
 529                         * the subpool and global reserve usage count can need
 530                         * to be adjusted.
 531                         */
 532                        VM_BUG_ON(HPageRestoreReserve(page));
 533                        remove_huge_page(page);
 534                        freed++;
 535                        if (!truncate_op) {
 536                                if (unlikely(hugetlb_unreserve_pages(inode,
 537                                                        index, index + 1, 1)))
 538                                        hugetlb_fix_reserve_counts(inode);
 539                        }
 540
 541                        unlock_page(page);
 542                        if (!truncate_op)
 543                                mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 544                }
 545                huge_pagevec_release(&pvec);
 546                cond_resched();
 547        }
 548
 549        if (truncate_op)
 550                (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
 551}
 552
 553static void hugetlbfs_evict_inode(struct inode *inode)
 554{
 555        struct resv_map *resv_map;
 556
 557        remove_inode_hugepages(inode, 0, LLONG_MAX);
 558
 559        /*
 560         * Get the resv_map from the address space embedded in the inode.
 561         * This is the address space which points to any resv_map allocated
 562         * at inode creation time.  If this is a device special inode,
 563         * i_mapping may not point to the original address space.
 564         */
 565        resv_map = (struct resv_map *)(&inode->i_data)->private_data;
 566        /* Only regular and link inodes have associated reserve maps */
 567        if (resv_map)
 568                resv_map_release(&resv_map->refs);
 569        clear_inode(inode);
 570}
 571
 572static void hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 573{
 574        pgoff_t pgoff;
 575        struct address_space *mapping = inode->i_mapping;
 576        struct hstate *h = hstate_inode(inode);
 577
 578        BUG_ON(offset & ~huge_page_mask(h));
 579        pgoff = offset >> PAGE_SHIFT;
 580
 581        i_mmap_lock_write(mapping);
 582        i_size_write(inode, offset);
 583        if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
 584                hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0);
 585        i_mmap_unlock_write(mapping);
 586        remove_inode_hugepages(inode, offset, LLONG_MAX);
 587}
 588
 589static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
 590{
 591        struct hstate *h = hstate_inode(inode);
 592        loff_t hpage_size = huge_page_size(h);
 593        loff_t hole_start, hole_end;
 594
 595        /*
 596         * For hole punch round up the beginning offset of the hole and
 597         * round down the end.
 598         */
 599        hole_start = round_up(offset, hpage_size);
 600        hole_end = round_down(offset + len, hpage_size);
 601
 602        if (hole_end > hole_start) {
 603                struct address_space *mapping = inode->i_mapping;
 604                struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 605
 606                inode_lock(inode);
 607
 608                /* protected by i_rwsem */
 609                if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
 610                        inode_unlock(inode);
 611                        return -EPERM;
 612                }
 613
 614                i_mmap_lock_write(mapping);
 615                if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
 616                        hugetlb_vmdelete_list(&mapping->i_mmap,
 617                                                hole_start >> PAGE_SHIFT,
 618                                                hole_end  >> PAGE_SHIFT);
 619                i_mmap_unlock_write(mapping);
 620                remove_inode_hugepages(inode, hole_start, hole_end);
 621                inode_unlock(inode);
 622        }
 623
 624        return 0;
 625}
 626
 627static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
 628                                loff_t len)
 629{
 630        struct inode *inode = file_inode(file);
 631        struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 632        struct address_space *mapping = inode->i_mapping;
 633        struct hstate *h = hstate_inode(inode);
 634        struct vm_area_struct pseudo_vma;
 635        struct mm_struct *mm = current->mm;
 636        loff_t hpage_size = huge_page_size(h);
 637        unsigned long hpage_shift = huge_page_shift(h);
 638        pgoff_t start, index, end;
 639        int error;
 640        u32 hash;
 641
 642        if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 643                return -EOPNOTSUPP;
 644
 645        if (mode & FALLOC_FL_PUNCH_HOLE)
 646                return hugetlbfs_punch_hole(inode, offset, len);
 647
 648        /*
 649         * Default preallocate case.
 650         * For this range, start is rounded down and end is rounded up
 651         * as well as being converted to page offsets.
 652         */
 653        start = offset >> hpage_shift;
 654        end = (offset + len + hpage_size - 1) >> hpage_shift;
 655
 656        inode_lock(inode);
 657
 658        /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
 659        error = inode_newsize_ok(inode, offset + len);
 660        if (error)
 661                goto out;
 662
 663        if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
 664                error = -EPERM;
 665                goto out;
 666        }
 667
 668        /*
 669         * Initialize a pseudo vma as this is required by the huge page
 670         * allocation routines.  If NUMA is configured, use page index
 671         * as input to create an allocation policy.
 672         */
 673        vma_init(&pseudo_vma, mm);
 674        pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
 675        pseudo_vma.vm_file = file;
 676
 677        for (index = start; index < end; index++) {
 678                /*
 679                 * This is supposed to be the vaddr where the page is being
 680                 * faulted in, but we have no vaddr here.
 681                 */
 682                struct page *page;
 683                unsigned long addr;
 684
 685                cond_resched();
 686
 687                /*
 688                 * fallocate(2) manpage permits EINTR; we may have been
 689                 * interrupted because we are using up too much memory.
 690                 */
 691                if (signal_pending(current)) {
 692                        error = -EINTR;
 693                        break;
 694                }
 695
 696                /* Set numa allocation policy based on index */
 697                hugetlb_set_vma_policy(&pseudo_vma, inode, index);
 698
 699                /* addr is the offset within the file (zero based) */
 700                addr = index * hpage_size;
 701
 702                /*
 703                 * fault mutex taken here, protects against fault path
 704                 * and hole punch.  inode_lock previously taken protects
 705                 * against truncation.
 706                 */
 707                hash = hugetlb_fault_mutex_hash(mapping, index);
 708                mutex_lock(&hugetlb_fault_mutex_table[hash]);
 709
 710                /* See if already present in mapping to avoid alloc/free */
 711                page = find_get_page(mapping, index);
 712                if (page) {
 713                        put_page(page);
 714                        mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 715                        hugetlb_drop_vma_policy(&pseudo_vma);
 716                        continue;
 717                }
 718
 719                /*
 720                 * Allocate page without setting the avoid_reserve argument.
 721                 * There certainly are no reserves associated with the
 722                 * pseudo_vma.  However, there could be shared mappings with
 723                 * reserves for the file at the inode level.  If we fallocate
 724                 * pages in these areas, we need to consume the reserves
 725                 * to keep reservation accounting consistent.
 726                 */
 727                page = alloc_huge_page(&pseudo_vma, addr, 0);
 728                hugetlb_drop_vma_policy(&pseudo_vma);
 729                if (IS_ERR(page)) {
 730                        mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 731                        error = PTR_ERR(page);
 732                        goto out;
 733                }
 734                clear_huge_page(page, addr, pages_per_huge_page(h));
 735                __SetPageUptodate(page);
 736                error = huge_add_to_page_cache(page, mapping, index);
 737                if (unlikely(error)) {
 738                        restore_reserve_on_error(h, &pseudo_vma, addr, page);
 739                        put_page(page);
 740                        mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 741                        goto out;
 742                }
 743
 744                mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 745
 746                SetHPageMigratable(page);
 747                /*
 748                 * unlock_page because locked by add_to_page_cache()
 749                 * put_page() due to reference from alloc_huge_page()
 750                 */
 751                unlock_page(page);
 752                put_page(page);
 753        }
 754
 755        if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
 756                i_size_write(inode, offset + len);
 757        inode->i_ctime = current_time(inode);
 758out:
 759        inode_unlock(inode);
 760        return error;
 761}
 762
 763static int hugetlbfs_setattr(struct user_namespace *mnt_userns,
 764                             struct dentry *dentry, struct iattr *attr)
 765{
 766        struct inode *inode = d_inode(dentry);
 767        struct hstate *h = hstate_inode(inode);
 768        int error;
 769        unsigned int ia_valid = attr->ia_valid;
 770        struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 771
 772        error = setattr_prepare(&init_user_ns, dentry, attr);
 773        if (error)
 774                return error;
 775
 776        if (ia_valid & ATTR_SIZE) {
 777                loff_t oldsize = inode->i_size;
 778                loff_t newsize = attr->ia_size;
 779
 780                if (newsize & ~huge_page_mask(h))
 781                        return -EINVAL;
 782                /* protected by i_rwsem */
 783                if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
 784                    (newsize > oldsize && (info->seals & F_SEAL_GROW)))
 785                        return -EPERM;
 786                hugetlb_vmtruncate(inode, newsize);
 787        }
 788
 789        setattr_copy(&init_user_ns, inode, attr);
 790        mark_inode_dirty(inode);
 791        return 0;
 792}
 793
 794static struct inode *hugetlbfs_get_root(struct super_block *sb,
 795                                        struct hugetlbfs_fs_context *ctx)
 796{
 797        struct inode *inode;
 798
 799        inode = new_inode(sb);
 800        if (inode) {
 801                inode->i_ino = get_next_ino();
 802                inode->i_mode = S_IFDIR | ctx->mode;
 803                inode->i_uid = ctx->uid;
 804                inode->i_gid = ctx->gid;
 805                inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 806                inode->i_op = &hugetlbfs_dir_inode_operations;
 807                inode->i_fop = &simple_dir_operations;
 808                /* directory inodes start off with i_nlink == 2 (for "." entry) */
 809                inc_nlink(inode);
 810                lockdep_annotate_inode_mutex_key(inode);
 811        }
 812        return inode;
 813}
 814
 815/*
 816 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
 817 * be taken from reclaim -- unlike regular filesystems. This needs an
 818 * annotation because huge_pmd_share() does an allocation under hugetlb's
 819 * i_mmap_rwsem.
 820 */
 821static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
 822
 823static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 824                                        struct inode *dir,
 825                                        umode_t mode, dev_t dev)
 826{
 827        struct inode *inode;
 828        struct resv_map *resv_map = NULL;
 829
 830        /*
 831         * Reserve maps are only needed for inodes that can have associated
 832         * page allocations.
 833         */
 834        if (S_ISREG(mode) || S_ISLNK(mode)) {
 835                resv_map = resv_map_alloc();
 836                if (!resv_map)
 837                        return NULL;
 838        }
 839
 840        inode = new_inode(sb);
 841        if (inode) {
 842                struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
 843
 844                inode->i_ino = get_next_ino();
 845                inode_init_owner(&init_user_ns, inode, dir, mode);
 846                lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
 847                                &hugetlbfs_i_mmap_rwsem_key);
 848                inode->i_mapping->a_ops = &hugetlbfs_aops;
 849                inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
 850                inode->i_mapping->private_data = resv_map;
 851                info->seals = F_SEAL_SEAL;
 852                switch (mode & S_IFMT) {
 853                default:
 854                        init_special_inode(inode, mode, dev);
 855                        break;
 856                case S_IFREG:
 857                        inode->i_op = &hugetlbfs_inode_operations;
 858                        inode->i_fop = &hugetlbfs_file_operations;
 859                        break;
 860                case S_IFDIR:
 861                        inode->i_op = &hugetlbfs_dir_inode_operations;
 862                        inode->i_fop = &simple_dir_operations;
 863
 864                        /* directory inodes start off with i_nlink == 2 (for "." entry) */
 865                        inc_nlink(inode);
 866                        break;
 867                case S_IFLNK:
 868                        inode->i_op = &page_symlink_inode_operations;
 869                        inode_nohighmem(inode);
 870                        break;
 871                }
 872                lockdep_annotate_inode_mutex_key(inode);
 873        } else {
 874                if (resv_map)
 875                        kref_put(&resv_map->refs, resv_map_release);
 876        }
 877
 878        return inode;
 879}
 880
 881/*
 882 * File creation. Allocate an inode, and we're done..
 883 */
 884static int do_hugetlbfs_mknod(struct inode *dir,
 885                        struct dentry *dentry,
 886                        umode_t mode,
 887                        dev_t dev,
 888                        bool tmpfile)
 889{
 890        struct inode *inode;
 891        int error = -ENOSPC;
 892
 893        inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
 894        if (inode) {
 895                dir->i_ctime = dir->i_mtime = current_time(dir);
 896                if (tmpfile) {
 897                        d_tmpfile(dentry, inode);
 898                } else {
 899                        d_instantiate(dentry, inode);
 900                        dget(dentry);/* Extra count - pin the dentry in core */
 901                }
 902                error = 0;
 903        }
 904        return error;
 905}
 906
 907static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
 908                           struct dentry *dentry, umode_t mode, dev_t dev)
 909{
 910        return do_hugetlbfs_mknod(dir, dentry, mode, dev, false);
 911}
 912
 913static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
 914                           struct dentry *dentry, umode_t mode)
 915{
 916        int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry,
 917                                     mode | S_IFDIR, 0);
 918        if (!retval)
 919                inc_nlink(dir);
 920        return retval;
 921}
 922
 923static int hugetlbfs_create(struct user_namespace *mnt_userns,
 924                            struct inode *dir, struct dentry *dentry,
 925                            umode_t mode, bool excl)
 926{
 927        return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
 928}
 929
 930static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns,
 931                             struct inode *dir, struct dentry *dentry,
 932                             umode_t mode)
 933{
 934        return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true);
 935}
 936
 937static int hugetlbfs_symlink(struct user_namespace *mnt_userns,
 938                             struct inode *dir, struct dentry *dentry,
 939                             const char *symname)
 940{
 941        struct inode *inode;
 942        int error = -ENOSPC;
 943
 944        inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
 945        if (inode) {
 946                int l = strlen(symname)+1;
 947                error = page_symlink(inode, symname, l);
 948                if (!error) {
 949                        d_instantiate(dentry, inode);
 950                        dget(dentry);
 951                } else
 952                        iput(inode);
 953        }
 954        dir->i_ctime = dir->i_mtime = current_time(dir);
 955
 956        return error;
 957}
 958
 959static int hugetlbfs_migrate_page(struct address_space *mapping,
 960                                struct page *newpage, struct page *page,
 961                                enum migrate_mode mode)
 962{
 963        int rc;
 964
 965        rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 966        if (rc != MIGRATEPAGE_SUCCESS)
 967                return rc;
 968
 969        if (hugetlb_page_subpool(page)) {
 970                hugetlb_set_page_subpool(newpage, hugetlb_page_subpool(page));
 971                hugetlb_set_page_subpool(page, NULL);
 972        }
 973
 974        if (mode != MIGRATE_SYNC_NO_COPY)
 975                migrate_page_copy(newpage, page);
 976        else
 977                migrate_page_states(newpage, page);
 978
 979        return MIGRATEPAGE_SUCCESS;
 980}
 981
 982static int hugetlbfs_error_remove_page(struct address_space *mapping,
 983                                struct page *page)
 984{
 985        struct inode *inode = mapping->host;
 986        pgoff_t index = page->index;
 987
 988        remove_huge_page(page);
 989        if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
 990                hugetlb_fix_reserve_counts(inode);
 991
 992        return 0;
 993}
 994
 995/*
 996 * Display the mount options in /proc/mounts.
 997 */
 998static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
 999{
1000        struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb);
1001        struct hugepage_subpool *spool = sbinfo->spool;
1002        unsigned long hpage_size = huge_page_size(sbinfo->hstate);
1003        unsigned hpage_shift = huge_page_shift(sbinfo->hstate);
1004        char mod;
1005
1006        if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
1007                seq_printf(m, ",uid=%u",
1008                           from_kuid_munged(&init_user_ns, sbinfo->uid));
1009        if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
1010                seq_printf(m, ",gid=%u",
1011                           from_kgid_munged(&init_user_ns, sbinfo->gid));
1012        if (sbinfo->mode != 0755)
1013                seq_printf(m, ",mode=%o", sbinfo->mode);
1014        if (sbinfo->max_inodes != -1)
1015                seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
1016
1017        hpage_size /= 1024;
1018        mod = 'K';
1019        if (hpage_size >= 1024) {
1020                hpage_size /= 1024;
1021                mod = 'M';
1022        }
1023        seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
1024        if (spool) {
1025                if (spool->max_hpages != -1)
1026                        seq_printf(m, ",size=%llu",
1027                                   (unsigned long long)spool->max_hpages << hpage_shift);
1028                if (spool->min_hpages != -1)
1029                        seq_printf(m, ",min_size=%llu",
1030                                   (unsigned long long)spool->min_hpages << hpage_shift);
1031        }
1032        return 0;
1033}
1034
1035static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1036{
1037        struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
1038        struct hstate *h = hstate_inode(d_inode(dentry));
1039
1040        buf->f_type = HUGETLBFS_MAGIC;
1041        buf->f_bsize = huge_page_size(h);
1042        if (sbinfo) {
1043                spin_lock(&sbinfo->stat_lock);
1044                /* If no limits set, just report 0 for max/free/used
1045                 * blocks, like simple_statfs() */
1046                if (sbinfo->spool) {
1047                        long free_pages;
1048
1049                        spin_lock(&sbinfo->spool->lock);
1050                        buf->f_blocks = sbinfo->spool->max_hpages;
1051                        free_pages = sbinfo->spool->max_hpages
1052                                - sbinfo->spool->used_hpages;
1053                        buf->f_bavail = buf->f_bfree = free_pages;
1054                        spin_unlock(&sbinfo->spool->lock);
1055                        buf->f_files = sbinfo->max_inodes;
1056                        buf->f_ffree = sbinfo->free_inodes;
1057                }
1058                spin_unlock(&sbinfo->stat_lock);
1059        }
1060        buf->f_namelen = NAME_MAX;
1061        return 0;
1062}
1063
1064static void hugetlbfs_put_super(struct super_block *sb)
1065{
1066        struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
1067
1068        if (sbi) {
1069                sb->s_fs_info = NULL;
1070
1071                if (sbi->spool)
1072                        hugepage_put_subpool(sbi->spool);
1073
1074                kfree(sbi);
1075        }
1076}
1077
1078static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1079{
1080        if (sbinfo->free_inodes >= 0) {
1081                spin_lock(&sbinfo->stat_lock);
1082                if (unlikely(!sbinfo->free_inodes)) {
1083                        spin_unlock(&sbinfo->stat_lock);
1084                        return 0;
1085                }
1086                sbinfo->free_inodes--;
1087                spin_unlock(&sbinfo->stat_lock);
1088        }
1089
1090        return 1;
1091}
1092
1093static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1094{
1095        if (sbinfo->free_inodes >= 0) {
1096                spin_lock(&sbinfo->stat_lock);
1097                sbinfo->free_inodes++;
1098                spin_unlock(&sbinfo->stat_lock);
1099        }
1100}
1101
1102
1103static struct kmem_cache *hugetlbfs_inode_cachep;
1104
1105static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
1106{
1107        struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
1108        struct hugetlbfs_inode_info *p;
1109
1110        if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
1111                return NULL;
1112        p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
1113        if (unlikely(!p)) {
1114                hugetlbfs_inc_free_inodes(sbinfo);
1115                return NULL;
1116        }
1117
1118        /*
1119         * Any time after allocation, hugetlbfs_destroy_inode can be called
1120         * for the inode.  mpol_free_shared_policy is unconditionally called
1121         * as part of hugetlbfs_destroy_inode.  So, initialize policy here
1122         * in case of a quick call to destroy.
1123         *
1124         * Note that the policy is initialized even if we are creating a
1125         * private inode.  This simplifies hugetlbfs_destroy_inode.
1126         */
1127        mpol_shared_policy_init(&p->policy, NULL);
1128
1129        return &p->vfs_inode;
1130}
1131
1132static void hugetlbfs_free_inode(struct inode *inode)
1133{
1134        kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
1135}
1136
1137static void hugetlbfs_destroy_inode(struct inode *inode)
1138{
1139        hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
1140        mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
1141}
1142
1143static const struct address_space_operations hugetlbfs_aops = {
1144        .write_begin    = hugetlbfs_write_begin,
1145        .write_end      = hugetlbfs_write_end,
1146        .set_page_dirty =  __set_page_dirty_no_writeback,
1147        .migratepage    = hugetlbfs_migrate_page,
1148        .error_remove_page      = hugetlbfs_error_remove_page,
1149};
1150
1151
1152static void init_once(void *foo)
1153{
1154        struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
1155
1156        inode_init_once(&ei->vfs_inode);
1157}
1158
1159const struct file_operations hugetlbfs_file_operations = {
1160        .read_iter              = hugetlbfs_read_iter,
1161        .mmap                   = hugetlbfs_file_mmap,
1162        .fsync                  = noop_fsync,
1163        .get_unmapped_area      = hugetlb_get_unmapped_area,
1164        .llseek                 = default_llseek,
1165        .fallocate              = hugetlbfs_fallocate,
1166};
1167
1168static const struct inode_operations hugetlbfs_dir_inode_operations = {
1169        .create         = hugetlbfs_create,
1170        .lookup         = simple_lookup,
1171        .link           = simple_link,
1172        .unlink         = simple_unlink,
1173        .symlink        = hugetlbfs_symlink,
1174        .mkdir          = hugetlbfs_mkdir,
1175        .rmdir          = simple_rmdir,
1176        .mknod          = hugetlbfs_mknod,
1177        .rename         = simple_rename,
1178        .setattr        = hugetlbfs_setattr,
1179        .tmpfile        = hugetlbfs_tmpfile,
1180};
1181
1182static const struct inode_operations hugetlbfs_inode_operations = {
1183        .setattr        = hugetlbfs_setattr,
1184};
1185
1186static const struct super_operations hugetlbfs_ops = {
1187        .alloc_inode    = hugetlbfs_alloc_inode,
1188        .free_inode     = hugetlbfs_free_inode,
1189        .destroy_inode  = hugetlbfs_destroy_inode,
1190        .evict_inode    = hugetlbfs_evict_inode,
1191        .statfs         = hugetlbfs_statfs,
1192        .put_super      = hugetlbfs_put_super,
1193        .show_options   = hugetlbfs_show_options,
1194};
1195
1196/*
1197 * Convert size option passed from command line to number of huge pages
1198 * in the pool specified by hstate.  Size option could be in bytes
1199 * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1200 */
1201static long
1202hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1203                         enum hugetlbfs_size_type val_type)
1204{
1205        if (val_type == NO_SIZE)
1206                return -1;
1207
1208        if (val_type == SIZE_PERCENT) {
1209                size_opt <<= huge_page_shift(h);
1210                size_opt *= h->max_huge_pages;
1211                do_div(size_opt, 100);
1212        }
1213
1214        size_opt >>= huge_page_shift(h);
1215        return size_opt;
1216}
1217
1218/*
1219 * Parse one mount parameter.
1220 */
1221static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
1222{
1223        struct hugetlbfs_fs_context *ctx = fc->fs_private;
1224        struct fs_parse_result result;
1225        char *rest;
1226        unsigned long ps;
1227        int opt;
1228
1229        opt = fs_parse(fc, hugetlb_fs_parameters, param, &result);
1230        if (opt < 0)
1231                return opt;
1232
1233        switch (opt) {
1234        case Opt_uid:
1235                ctx->uid = make_kuid(current_user_ns(), result.uint_32);
1236                if (!uid_valid(ctx->uid))
1237                        goto bad_val;
1238                return 0;
1239
1240        case Opt_gid:
1241                ctx->gid = make_kgid(current_user_ns(), result.uint_32);
1242                if (!gid_valid(ctx->gid))
1243                        goto bad_val;
1244                return 0;
1245
1246        case Opt_mode:
1247                ctx->mode = result.uint_32 & 01777U;
1248                return 0;
1249
1250        case Opt_size:
1251                /* memparse() will accept a K/M/G without a digit */
1252                if (!isdigit(param->string[0]))
1253                        goto bad_val;
1254                ctx->max_size_opt = memparse(param->string, &rest);
1255                ctx->max_val_type = SIZE_STD;
1256                if (*rest == '%')
1257                        ctx->max_val_type = SIZE_PERCENT;
1258                return 0;
1259
1260        case Opt_nr_inodes:
1261                /* memparse() will accept a K/M/G without a digit */
1262                if (!isdigit(param->string[0]))
1263                        goto bad_val;
1264                ctx->nr_inodes = memparse(param->string, &rest);
1265                return 0;
1266
1267        case Opt_pagesize:
1268                ps = memparse(param->string, &rest);
1269                ctx->hstate = size_to_hstate(ps);
1270                if (!ctx->hstate) {
1271                        pr_err("Unsupported page size %lu MB\n", ps >> 20);
1272                        return -EINVAL;
1273                }
1274                return 0;
1275
1276        case Opt_min_size:
1277                /* memparse() will accept a K/M/G without a digit */
1278                if (!isdigit(param->string[0]))
1279                        goto bad_val;
1280                ctx->min_size_opt = memparse(param->string, &rest);
1281                ctx->min_val_type = SIZE_STD;
1282                if (*rest == '%')
1283                        ctx->min_val_type = SIZE_PERCENT;
1284                return 0;
1285
1286        default:
1287                return -EINVAL;
1288        }
1289
1290bad_val:
1291        return invalfc(fc, "Bad value '%s' for mount option '%s'\n",
1292                      param->string, param->key);
1293}
1294
1295/*
1296 * Validate the parsed options.
1297 */
1298static int hugetlbfs_validate(struct fs_context *fc)
1299{
1300        struct hugetlbfs_fs_context *ctx = fc->fs_private;
1301
1302        /*
1303         * Use huge page pool size (in hstate) to convert the size
1304         * options to number of huge pages.  If NO_SIZE, -1 is returned.
1305         */
1306        ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1307                                                   ctx->max_size_opt,
1308                                                   ctx->max_val_type);
1309        ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1310                                                   ctx->min_size_opt,
1311                                                   ctx->min_val_type);
1312
1313        /*
1314         * If max_size was specified, then min_size must be smaller
1315         */
1316        if (ctx->max_val_type > NO_SIZE &&
1317            ctx->min_hpages > ctx->max_hpages) {
1318                pr_err("Minimum size can not be greater than maximum size\n");
1319                return -EINVAL;
1320        }
1321
1322        return 0;
1323}
1324
1325static int
1326hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc)
1327{
1328        struct hugetlbfs_fs_context *ctx = fc->fs_private;
1329        struct hugetlbfs_sb_info *sbinfo;
1330
1331        sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1332        if (!sbinfo)
1333                return -ENOMEM;
1334        sb->s_fs_info = sbinfo;
1335        spin_lock_init(&sbinfo->stat_lock);
1336        sbinfo->hstate          = ctx->hstate;
1337        sbinfo->max_inodes      = ctx->nr_inodes;
1338        sbinfo->free_inodes     = ctx->nr_inodes;
1339        sbinfo->spool           = NULL;
1340        sbinfo->uid             = ctx->uid;
1341        sbinfo->gid             = ctx->gid;
1342        sbinfo->mode            = ctx->mode;
1343
1344        /*
1345         * Allocate and initialize subpool if maximum or minimum size is
1346         * specified.  Any needed reservations (for minimum size) are taken
1347         * taken when the subpool is created.
1348         */
1349        if (ctx->max_hpages != -1 || ctx->min_hpages != -1) {
1350                sbinfo->spool = hugepage_new_subpool(ctx->hstate,
1351                                                     ctx->max_hpages,
1352                                                     ctx->min_hpages);
1353                if (!sbinfo->spool)
1354                        goto out_free;
1355        }
1356        sb->s_maxbytes = MAX_LFS_FILESIZE;
1357        sb->s_blocksize = huge_page_size(ctx->hstate);
1358        sb->s_blocksize_bits = huge_page_shift(ctx->hstate);
1359        sb->s_magic = HUGETLBFS_MAGIC;
1360        sb->s_op = &hugetlbfs_ops;
1361        sb->s_time_gran = 1;
1362
1363        /*
1364         * Due to the special and limited functionality of hugetlbfs, it does
1365         * not work well as a stacking filesystem.
1366         */
1367        sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
1368        sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx));
1369        if (!sb->s_root)
1370                goto out_free;
1371        return 0;
1372out_free:
1373        kfree(sbinfo->spool);
1374        kfree(sbinfo);
1375        return -ENOMEM;
1376}
1377
1378static int hugetlbfs_get_tree(struct fs_context *fc)
1379{
1380        int err = hugetlbfs_validate(fc);
1381        if (err)
1382                return err;
1383        return get_tree_nodev(fc, hugetlbfs_fill_super);
1384}
1385
1386static void hugetlbfs_fs_context_free(struct fs_context *fc)
1387{
1388        kfree(fc->fs_private);
1389}
1390
1391static const struct fs_context_operations hugetlbfs_fs_context_ops = {
1392        .free           = hugetlbfs_fs_context_free,
1393        .parse_param    = hugetlbfs_parse_param,
1394        .get_tree       = hugetlbfs_get_tree,
1395};
1396
1397static int hugetlbfs_init_fs_context(struct fs_context *fc)
1398{
1399        struct hugetlbfs_fs_context *ctx;
1400
1401        ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL);
1402        if (!ctx)
1403                return -ENOMEM;
1404
1405        ctx->max_hpages = -1; /* No limit on size by default */
1406        ctx->nr_inodes  = -1; /* No limit on number of inodes by default */
1407        ctx->uid        = current_fsuid();
1408        ctx->gid        = current_fsgid();
1409        ctx->mode       = 0755;
1410        ctx->hstate     = &default_hstate;
1411        ctx->min_hpages = -1; /* No default minimum size */
1412        ctx->max_val_type = NO_SIZE;
1413        ctx->min_val_type = NO_SIZE;
1414        fc->fs_private = ctx;
1415        fc->ops = &hugetlbfs_fs_context_ops;
1416        return 0;
1417}
1418
1419static struct file_system_type hugetlbfs_fs_type = {
1420        .name                   = "hugetlbfs",
1421        .init_fs_context        = hugetlbfs_init_fs_context,
1422        .parameters             = hugetlb_fs_parameters,
1423        .kill_sb                = kill_litter_super,
1424};
1425
1426static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1427
1428static int can_do_hugetlb_shm(void)
1429{
1430        kgid_t shm_group;
1431        shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1432        return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1433}
1434
1435static int get_hstate_idx(int page_size_log)
1436{
1437        struct hstate *h = hstate_sizelog(page_size_log);
1438
1439        if (!h)
1440                return -1;
1441        return hstate_index(h);
1442}
1443
1444/*
1445 * Note that size should be aligned to proper hugepage size in caller side,
1446 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1447 */
1448struct file *hugetlb_file_setup(const char *name, size_t size,
1449                                vm_flags_t acctflag, int creat_flags,
1450                                int page_size_log)
1451{
1452        struct inode *inode;
1453        struct vfsmount *mnt;
1454        int hstate_idx;
1455        struct file *file;
1456
1457        hstate_idx = get_hstate_idx(page_size_log);
1458        if (hstate_idx < 0)
1459                return ERR_PTR(-ENODEV);
1460
1461        mnt = hugetlbfs_vfsmount[hstate_idx];
1462        if (!mnt)
1463                return ERR_PTR(-ENOENT);
1464
1465        if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1466                struct ucounts *ucounts = current_ucounts();
1467
1468                if (user_shm_lock(size, ucounts)) {
1469                        pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n",
1470                                current->comm, current->pid);
1471                        user_shm_unlock(size, ucounts);
1472                }
1473                return ERR_PTR(-EPERM);
1474        }
1475
1476        file = ERR_PTR(-ENOSPC);
1477        inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0);
1478        if (!inode)
1479                goto out;
1480        if (creat_flags == HUGETLB_SHMFS_INODE)
1481                inode->i_flags |= S_PRIVATE;
1482
1483        inode->i_size = size;
1484        clear_nlink(inode);
1485
1486        if (!hugetlb_reserve_pages(inode, 0,
1487                        size >> huge_page_shift(hstate_inode(inode)), NULL,
1488                        acctflag))
1489                file = ERR_PTR(-ENOMEM);
1490        else
1491                file = alloc_file_pseudo(inode, mnt, name, O_RDWR,
1492                                        &hugetlbfs_file_operations);
1493        if (!IS_ERR(file))
1494                return file;
1495
1496        iput(inode);
1497out:
1498        return file;
1499}
1500
1501static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h)
1502{
1503        struct fs_context *fc;
1504        struct vfsmount *mnt;
1505
1506        fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT);
1507        if (IS_ERR(fc)) {
1508                mnt = ERR_CAST(fc);
1509        } else {
1510                struct hugetlbfs_fs_context *ctx = fc->fs_private;
1511                ctx->hstate = h;
1512                mnt = fc_mount(fc);
1513                put_fs_context(fc);
1514        }
1515        if (IS_ERR(mnt))
1516                pr_err("Cannot mount internal hugetlbfs for page size %luK",
1517                       huge_page_size(h) >> 10);
1518        return mnt;
1519}
1520
1521static int __init init_hugetlbfs_fs(void)
1522{
1523        struct vfsmount *mnt;
1524        struct hstate *h;
1525        int error;
1526        int i;
1527
1528        if (!hugepages_supported()) {
1529                pr_info("disabling because there are no supported hugepage sizes\n");
1530                return -ENOTSUPP;
1531        }
1532
1533        error = -ENOMEM;
1534        hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1535                                        sizeof(struct hugetlbfs_inode_info),
1536                                        0, SLAB_ACCOUNT, init_once);
1537        if (hugetlbfs_inode_cachep == NULL)
1538                goto out;
1539
1540        error = register_filesystem(&hugetlbfs_fs_type);
1541        if (error)
1542                goto out_free;
1543
1544        /* default hstate mount is required */
1545        mnt = mount_one_hugetlbfs(&default_hstate);
1546        if (IS_ERR(mnt)) {
1547                error = PTR_ERR(mnt);
1548                goto out_unreg;
1549        }
1550        hugetlbfs_vfsmount[default_hstate_idx] = mnt;
1551
1552        /* other hstates are optional */
1553        i = 0;
1554        for_each_hstate(h) {
1555                if (i == default_hstate_idx) {
1556                        i++;
1557                        continue;
1558                }
1559
1560                mnt = mount_one_hugetlbfs(h);
1561                if (IS_ERR(mnt))
1562                        hugetlbfs_vfsmount[i] = NULL;
1563                else
1564                        hugetlbfs_vfsmount[i] = mnt;
1565                i++;
1566        }
1567
1568        return 0;
1569
1570 out_unreg:
1571        (void)unregister_filesystem(&hugetlbfs_fs_type);
1572 out_free:
1573        kmem_cache_destroy(hugetlbfs_inode_cachep);
1574 out:
1575        return error;
1576}
1577fs_initcall(init_hugetlbfs_fs)
1578