linux/fs/hugetlbfs/inode.c
<<
>>
Prefs
   1/*
   2 * hugetlbpage-backed filesystem.  Based on ramfs.
   3 *
   4 * William Irwin, 2002
   5 *
   6 * Copyright (C) 2002 Linus Torvalds.
   7 */
   8
   9#include <linux/module.h>
  10#include <linux/thread_info.h>
  11#include <asm/current.h>
  12#include <linux/sched.h>                /* remove ASAP */
  13#include <linux/fs.h>
  14#include <linux/mount.h>
  15#include <linux/file.h>
  16#include <linux/kernel.h>
  17#include <linux/writeback.h>
  18#include <linux/pagemap.h>
  19#include <linux/highmem.h>
  20#include <linux/init.h>
  21#include <linux/string.h>
  22#include <linux/capability.h>
  23#include <linux/ctype.h>
  24#include <linux/backing-dev.h>
  25#include <linux/hugetlb.h>
  26#include <linux/pagevec.h>
  27#include <linux/parser.h>
  28#include <linux/mman.h>
  29#include <linux/slab.h>
  30#include <linux/dnotify.h>
  31#include <linux/statfs.h>
  32#include <linux/security.h>
  33#include <linux/magic.h>
  34#include <linux/migrate.h>
  35
  36#include <asm/uaccess.h>
  37
  38static const struct super_operations hugetlbfs_ops;
  39static const struct address_space_operations hugetlbfs_aops;
  40const struct file_operations hugetlbfs_file_operations;
  41static const struct inode_operations hugetlbfs_dir_inode_operations;
  42static const struct inode_operations hugetlbfs_inode_operations;
  43
  44static struct backing_dev_info hugetlbfs_backing_dev_info = {
  45        .name           = "hugetlbfs",
  46        .ra_pages       = 0,    /* No readahead */
  47        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
  48};
  49
  50int sysctl_hugetlb_shm_group;
  51
  52enum {
  53        Opt_size, Opt_nr_inodes,
  54        Opt_mode, Opt_uid, Opt_gid,
  55        Opt_pagesize,
  56        Opt_err,
  57};
  58
  59static const match_table_t tokens = {
  60        {Opt_size,      "size=%s"},
  61        {Opt_nr_inodes, "nr_inodes=%s"},
  62        {Opt_mode,      "mode=%o"},
  63        {Opt_uid,       "uid=%u"},
  64        {Opt_gid,       "gid=%u"},
  65        {Opt_pagesize,  "pagesize=%s"},
  66        {Opt_err,       NULL},
  67};
  68
  69static void huge_pagevec_release(struct pagevec *pvec)
  70{
  71        int i;
  72
  73        for (i = 0; i < pagevec_count(pvec); ++i)
  74                put_page(pvec->pages[i]);
  75
  76        pagevec_reinit(pvec);
  77}
  78
  79static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
  80{
  81        struct inode *inode = file->f_path.dentry->d_inode;
  82        loff_t len, vma_len;
  83        int ret;
  84        struct hstate *h = hstate_file(file);
  85
  86        /*
  87         * vma address alignment (but not the pgoff alignment) has
  88         * already been checked by prepare_hugepage_range.  If you add
  89         * any error returns here, do so after setting VM_HUGETLB, so
  90         * is_vm_hugetlb_page tests below unmap_region go the right
  91         * way when do_mmap_pgoff unwinds (may be important on powerpc
  92         * and ia64).
  93         */
  94        vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
  95        vma->vm_ops = &hugetlb_vm_ops;
  96
  97        if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
  98                return -EINVAL;
  99
 100        vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 101
 102        mutex_lock(&inode->i_mutex);
 103        file_accessed(file);
 104
 105        ret = -ENOMEM;
 106        len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
 107
 108        if (hugetlb_reserve_pages(inode,
 109                                vma->vm_pgoff >> huge_page_order(h),
 110                                len >> huge_page_shift(h), vma,
 111                                vma->vm_flags))
 112                goto out;
 113
 114        ret = 0;
 115        hugetlb_prefault_arch_hook(vma->vm_mm);
 116        if (vma->vm_flags & VM_WRITE && inode->i_size < len)
 117                inode->i_size = len;
 118out:
 119        mutex_unlock(&inode->i_mutex);
 120
 121        return ret;
 122}
 123
 124/*
 125 * Called under down_write(mmap_sem).
 126 */
 127
 128#ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 129static unsigned long
 130hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
 131                unsigned long len, unsigned long pgoff, unsigned long flags)
 132{
 133        struct mm_struct *mm = current->mm;
 134        struct vm_area_struct *vma;
 135        unsigned long start_addr;
 136        struct hstate *h = hstate_file(file);
 137
 138        if (len & ~huge_page_mask(h))
 139                return -EINVAL;
 140        if (len > TASK_SIZE)
 141                return -ENOMEM;
 142
 143        if (flags & MAP_FIXED) {
 144                if (prepare_hugepage_range(file, addr, len))
 145                        return -EINVAL;
 146                return addr;
 147        }
 148
 149        if (addr) {
 150                addr = ALIGN(addr, huge_page_size(h));
 151                vma = find_vma(mm, addr);
 152                if (TASK_SIZE - len >= addr &&
 153                    (!vma || addr + len <= vma->vm_start))
 154                        return addr;
 155        }
 156
 157        start_addr = mm->free_area_cache;
 158
 159        if (len <= mm->cached_hole_size)
 160                start_addr = TASK_UNMAPPED_BASE;
 161
 162full_search:
 163        addr = ALIGN(start_addr, huge_page_size(h));
 164
 165        for (vma = find_vma(mm, addr); ; vma = vma->vm_next) {
 166                /* At this point:  (!vma || addr < vma->vm_end). */
 167                if (TASK_SIZE - len < addr) {
 168                        /*
 169                         * Start a new search - just in case we missed
 170                         * some holes.
 171                         */
 172                        if (start_addr != TASK_UNMAPPED_BASE) {
 173                                start_addr = TASK_UNMAPPED_BASE;
 174                                goto full_search;
 175                        }
 176                        return -ENOMEM;
 177                }
 178
 179                if (!vma || addr + len <= vma->vm_start)
 180                        return addr;
 181                addr = ALIGN(vma->vm_end, huge_page_size(h));
 182        }
 183}
 184#endif
 185
 186static int
 187hugetlbfs_read_actor(struct page *page, unsigned long offset,
 188                        char __user *buf, unsigned long count,
 189                        unsigned long size)
 190{
 191        char *kaddr;
 192        unsigned long left, copied = 0;
 193        int i, chunksize;
 194
 195        if (size > count)
 196                size = count;
 197
 198        /* Find which 4k chunk and offset with in that chunk */
 199        i = offset >> PAGE_CACHE_SHIFT;
 200        offset = offset & ~PAGE_CACHE_MASK;
 201
 202        while (size) {
 203                chunksize = PAGE_CACHE_SIZE;
 204                if (offset)
 205                        chunksize -= offset;
 206                if (chunksize > size)
 207                        chunksize = size;
 208                kaddr = kmap(&page[i]);
 209                left = __copy_to_user(buf, kaddr + offset, chunksize);
 210                kunmap(&page[i]);
 211                if (left) {
 212                        copied += (chunksize - left);
 213                        break;
 214                }
 215                offset = 0;
 216                size -= chunksize;
 217                buf += chunksize;
 218                copied += chunksize;
 219                i++;
 220        }
 221        return copied ? copied : -EFAULT;
 222}
 223
 224/*
 225 * Support for read() - Find the page attached to f_mapping and copy out the
 226 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
 227 * since it has PAGE_CACHE_SIZE assumptions.
 228 */
 229static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
 230                              size_t len, loff_t *ppos)
 231{
 232        struct hstate *h = hstate_file(filp);
 233        struct address_space *mapping = filp->f_mapping;
 234        struct inode *inode = mapping->host;
 235        unsigned long index = *ppos >> huge_page_shift(h);
 236        unsigned long offset = *ppos & ~huge_page_mask(h);
 237        unsigned long end_index;
 238        loff_t isize;
 239        ssize_t retval = 0;
 240
 241        /* validate length */
 242        if (len == 0)
 243                goto out;
 244
 245        for (;;) {
 246                struct page *page;
 247                unsigned long nr, ret;
 248                int ra;
 249
 250                /* nr is the maximum number of bytes to copy from this page */
 251                nr = huge_page_size(h);
 252                isize = i_size_read(inode);
 253                if (!isize)
 254                        goto out;
 255                end_index = (isize - 1) >> huge_page_shift(h);
 256                if (index >= end_index) {
 257                        if (index > end_index)
 258                                goto out;
 259                        nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
 260                        if (nr <= offset)
 261                                goto out;
 262                }
 263                nr = nr - offset;
 264
 265                /* Find the page */
 266                page = find_lock_page(mapping, index);
 267                if (unlikely(page == NULL)) {
 268                        /*
 269                         * We have a HOLE, zero out the user-buffer for the
 270                         * length of the hole or request.
 271                         */
 272                        ret = len < nr ? len : nr;
 273                        if (clear_user(buf, ret))
 274                                ra = -EFAULT;
 275                        else
 276                                ra = 0;
 277                } else {
 278                        unlock_page(page);
 279
 280                        /*
 281                         * We have the page, copy it to user space buffer.
 282                         */
 283                        ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
 284                        ret = ra;
 285                        page_cache_release(page);
 286                }
 287                if (ra < 0) {
 288                        if (retval == 0)
 289                                retval = ra;
 290                        goto out;
 291                }
 292
 293                offset += ret;
 294                retval += ret;
 295                len -= ret;
 296                index += offset >> huge_page_shift(h);
 297                offset &= ~huge_page_mask(h);
 298
 299                /* short read or no more work */
 300                if ((ret != nr) || (len == 0))
 301                        break;
 302        }
 303out:
 304        *ppos = ((loff_t)index << huge_page_shift(h)) + offset;
 305        return retval;
 306}
 307
 308static int hugetlbfs_write_begin(struct file *file,
 309                        struct address_space *mapping,
 310                        loff_t pos, unsigned len, unsigned flags,
 311                        struct page **pagep, void **fsdata)
 312{
 313        return -EINVAL;
 314}
 315
 316static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
 317                        loff_t pos, unsigned len, unsigned copied,
 318                        struct page *page, void *fsdata)
 319{
 320        BUG();
 321        return -EINVAL;
 322}
 323
 324static void truncate_huge_page(struct page *page)
 325{
 326        cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
 327        ClearPageUptodate(page);
 328        delete_from_page_cache(page);
 329}
 330
 331static void truncate_hugepages(struct inode *inode, loff_t lstart)
 332{
 333        struct hstate *h = hstate_inode(inode);
 334        struct address_space *mapping = &inode->i_data;
 335        const pgoff_t start = lstart >> huge_page_shift(h);
 336        struct pagevec pvec;
 337        pgoff_t next;
 338        int i, freed = 0;
 339
 340        pagevec_init(&pvec, 0);
 341        next = start;
 342        while (1) {
 343                if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 344                        if (next == start)
 345                                break;
 346                        next = start;
 347                        continue;
 348                }
 349
 350                for (i = 0; i < pagevec_count(&pvec); ++i) {
 351                        struct page *page = pvec.pages[i];
 352
 353                        lock_page(page);
 354                        if (page->index > next)
 355                                next = page->index;
 356                        ++next;
 357                        truncate_huge_page(page);
 358                        unlock_page(page);
 359                        freed++;
 360                }
 361                huge_pagevec_release(&pvec);
 362        }
 363        BUG_ON(!lstart && mapping->nrpages);
 364        hugetlb_unreserve_pages(inode, start, freed);
 365}
 366
 367static void hugetlbfs_evict_inode(struct inode *inode)
 368{
 369        truncate_hugepages(inode, 0);
 370        end_writeback(inode);
 371}
 372
 373static inline void
 374hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 375{
 376        struct vm_area_struct *vma;
 377        struct prio_tree_iter iter;
 378
 379        vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
 380                unsigned long v_offset;
 381
 382                /*
 383                 * Can the expression below overflow on 32-bit arches?
 384                 * No, because the prio_tree returns us only those vmas
 385                 * which overlap the truncated area starting at pgoff,
 386                 * and no vma on a 32-bit arch can span beyond the 4GB.
 387                 */
 388                if (vma->vm_pgoff < pgoff)
 389                        v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
 390                else
 391                        v_offset = 0;
 392
 393                __unmap_hugepage_range(vma,
 394                                vma->vm_start + v_offset, vma->vm_end, NULL);
 395        }
 396}
 397
 398static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 399{
 400        pgoff_t pgoff;
 401        struct address_space *mapping = inode->i_mapping;
 402        struct hstate *h = hstate_inode(inode);
 403
 404        BUG_ON(offset & ~huge_page_mask(h));
 405        pgoff = offset >> PAGE_SHIFT;
 406
 407        i_size_write(inode, offset);
 408        mutex_lock(&mapping->i_mmap_mutex);
 409        if (!prio_tree_empty(&mapping->i_mmap))
 410                hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff);
 411        mutex_unlock(&mapping->i_mmap_mutex);
 412        truncate_hugepages(inode, offset);
 413        return 0;
 414}
 415
 416static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
 417{
 418        struct inode *inode = dentry->d_inode;
 419        struct hstate *h = hstate_inode(inode);
 420        int error;
 421        unsigned int ia_valid = attr->ia_valid;
 422
 423        BUG_ON(!inode);
 424
 425        error = inode_change_ok(inode, attr);
 426        if (error)
 427                return error;
 428
 429        if (ia_valid & ATTR_SIZE) {
 430                error = -EINVAL;
 431                if (attr->ia_size & ~huge_page_mask(h))
 432                        return -EINVAL;
 433                error = hugetlb_vmtruncate(inode, attr->ia_size);
 434                if (error)
 435                        return error;
 436        }
 437
 438        setattr_copy(inode, attr);
 439        mark_inode_dirty(inode);
 440        return 0;
 441}
 442
 443static struct inode *hugetlbfs_get_root(struct super_block *sb,
 444                                        struct hugetlbfs_config *config)
 445{
 446        struct inode *inode;
 447
 448        inode = new_inode(sb);
 449        if (inode) {
 450                struct hugetlbfs_inode_info *info;
 451                inode->i_ino = get_next_ino();
 452                inode->i_mode = S_IFDIR | config->mode;
 453                inode->i_uid = config->uid;
 454                inode->i_gid = config->gid;
 455                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 456                info = HUGETLBFS_I(inode);
 457                mpol_shared_policy_init(&info->policy, NULL);
 458                inode->i_op = &hugetlbfs_dir_inode_operations;
 459                inode->i_fop = &simple_dir_operations;
 460                /* directory inodes start off with i_nlink == 2 (for "." entry) */
 461                inc_nlink(inode);
 462        }
 463        return inode;
 464}
 465
 466static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 467                                        struct inode *dir,
 468                                        umode_t mode, dev_t dev)
 469{
 470        struct inode *inode;
 471
 472        inode = new_inode(sb);
 473        if (inode) {
 474                struct hugetlbfs_inode_info *info;
 475                inode->i_ino = get_next_ino();
 476                inode_init_owner(inode, dir, mode);
 477                inode->i_mapping->a_ops = &hugetlbfs_aops;
 478                inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
 479                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 480                INIT_LIST_HEAD(&inode->i_mapping->private_list);
 481                info = HUGETLBFS_I(inode);
 482                /*
 483                 * The policy is initialized here even if we are creating a
 484                 * private inode because initialization simply creates an
 485                 * an empty rb tree and calls spin_lock_init(), later when we
 486                 * call mpol_free_shared_policy() it will just return because
 487                 * the rb tree will still be empty.
 488                 */
 489                mpol_shared_policy_init(&info->policy, NULL);
 490                switch (mode & S_IFMT) {
 491                default:
 492                        init_special_inode(inode, mode, dev);
 493                        break;
 494                case S_IFREG:
 495                        inode->i_op = &hugetlbfs_inode_operations;
 496                        inode->i_fop = &hugetlbfs_file_operations;
 497                        break;
 498                case S_IFDIR:
 499                        inode->i_op = &hugetlbfs_dir_inode_operations;
 500                        inode->i_fop = &simple_dir_operations;
 501
 502                        /* directory inodes start off with i_nlink == 2 (for "." entry) */
 503                        inc_nlink(inode);
 504                        break;
 505                case S_IFLNK:
 506                        inode->i_op = &page_symlink_inode_operations;
 507                        break;
 508                }
 509                lockdep_annotate_inode_mutex_key(inode);
 510        }
 511        return inode;
 512}
 513
 514/*
 515 * File creation. Allocate an inode, and we're done..
 516 */
 517static int hugetlbfs_mknod(struct inode *dir,
 518                        struct dentry *dentry, umode_t mode, dev_t dev)
 519{
 520        struct inode *inode;
 521        int error = -ENOSPC;
 522
 523        inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
 524        if (inode) {
 525                dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 526                d_instantiate(dentry, inode);
 527                dget(dentry);   /* Extra count - pin the dentry in core */
 528                error = 0;
 529        }
 530        return error;
 531}
 532
 533static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
 534{
 535        int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
 536        if (!retval)
 537                inc_nlink(dir);
 538        return retval;
 539}
 540
 541static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, struct nameidata *nd)
 542{
 543        return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
 544}
 545
 546static int hugetlbfs_symlink(struct inode *dir,
 547                        struct dentry *dentry, const char *symname)
 548{
 549        struct inode *inode;
 550        int error = -ENOSPC;
 551
 552        inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
 553        if (inode) {
 554                int l = strlen(symname)+1;
 555                error = page_symlink(inode, symname, l);
 556                if (!error) {
 557                        d_instantiate(dentry, inode);
 558                        dget(dentry);
 559                } else
 560                        iput(inode);
 561        }
 562        dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 563
 564        return error;
 565}
 566
 567/*
 568 * mark the head page dirty
 569 */
 570static int hugetlbfs_set_page_dirty(struct page *page)
 571{
 572        struct page *head = compound_head(page);
 573
 574        SetPageDirty(head);
 575        return 0;
 576}
 577
 578static int hugetlbfs_migrate_page(struct address_space *mapping,
 579                                struct page *newpage, struct page *page,
 580                                enum migrate_mode mode)
 581{
 582        int rc;
 583
 584        rc = migrate_huge_page_move_mapping(mapping, newpage, page);
 585        if (rc)
 586                return rc;
 587        migrate_page_copy(newpage, page);
 588
 589        return 0;
 590}
 591
 592static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
 593{
 594        struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
 595        struct hstate *h = hstate_inode(dentry->d_inode);
 596
 597        buf->f_type = HUGETLBFS_MAGIC;
 598        buf->f_bsize = huge_page_size(h);
 599        if (sbinfo) {
 600                spin_lock(&sbinfo->stat_lock);
 601                /* If no limits set, just report 0 for max/free/used
 602                 * blocks, like simple_statfs() */
 603                if (sbinfo->max_blocks >= 0) {
 604                        buf->f_blocks = sbinfo->max_blocks;
 605                        buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
 606                        buf->f_files = sbinfo->max_inodes;
 607                        buf->f_ffree = sbinfo->free_inodes;
 608                }
 609                spin_unlock(&sbinfo->stat_lock);
 610        }
 611        buf->f_namelen = NAME_MAX;
 612        return 0;
 613}
 614
 615static void hugetlbfs_put_super(struct super_block *sb)
 616{
 617        struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
 618
 619        if (sbi) {
 620                sb->s_fs_info = NULL;
 621                kfree(sbi);
 622        }
 623}
 624
 625static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 626{
 627        if (sbinfo->free_inodes >= 0) {
 628                spin_lock(&sbinfo->stat_lock);
 629                if (unlikely(!sbinfo->free_inodes)) {
 630                        spin_unlock(&sbinfo->stat_lock);
 631                        return 0;
 632                }
 633                sbinfo->free_inodes--;
 634                spin_unlock(&sbinfo->stat_lock);
 635        }
 636
 637        return 1;
 638}
 639
 640static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 641{
 642        if (sbinfo->free_inodes >= 0) {
 643                spin_lock(&sbinfo->stat_lock);
 644                sbinfo->free_inodes++;
 645                spin_unlock(&sbinfo->stat_lock);
 646        }
 647}
 648
 649
 650static struct kmem_cache *hugetlbfs_inode_cachep;
 651
 652static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 653{
 654        struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
 655        struct hugetlbfs_inode_info *p;
 656
 657        if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
 658                return NULL;
 659        p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
 660        if (unlikely(!p)) {
 661                hugetlbfs_inc_free_inodes(sbinfo);
 662                return NULL;
 663        }
 664        return &p->vfs_inode;
 665}
 666
 667static void hugetlbfs_i_callback(struct rcu_head *head)
 668{
 669        struct inode *inode = container_of(head, struct inode, i_rcu);
 670        kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 671}
 672
 673static void hugetlbfs_destroy_inode(struct inode *inode)
 674{
 675        hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
 676        mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
 677        call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
 678}
 679
 680static const struct address_space_operations hugetlbfs_aops = {
 681        .write_begin    = hugetlbfs_write_begin,
 682        .write_end      = hugetlbfs_write_end,
 683        .set_page_dirty = hugetlbfs_set_page_dirty,
 684        .migratepage    = hugetlbfs_migrate_page,
 685};
 686
 687
 688static void init_once(void *foo)
 689{
 690        struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 691
 692        inode_init_once(&ei->vfs_inode);
 693}
 694
 695const struct file_operations hugetlbfs_file_operations = {
 696        .read                   = hugetlbfs_read,
 697        .mmap                   = hugetlbfs_file_mmap,
 698        .fsync                  = noop_fsync,
 699        .get_unmapped_area      = hugetlb_get_unmapped_area,
 700        .llseek         = default_llseek,
 701};
 702
 703static const struct inode_operations hugetlbfs_dir_inode_operations = {
 704        .create         = hugetlbfs_create,
 705        .lookup         = simple_lookup,
 706        .link           = simple_link,
 707        .unlink         = simple_unlink,
 708        .symlink        = hugetlbfs_symlink,
 709        .mkdir          = hugetlbfs_mkdir,
 710        .rmdir          = simple_rmdir,
 711        .mknod          = hugetlbfs_mknod,
 712        .rename         = simple_rename,
 713        .setattr        = hugetlbfs_setattr,
 714};
 715
 716static const struct inode_operations hugetlbfs_inode_operations = {
 717        .setattr        = hugetlbfs_setattr,
 718};
 719
 720static const struct super_operations hugetlbfs_ops = {
 721        .alloc_inode    = hugetlbfs_alloc_inode,
 722        .destroy_inode  = hugetlbfs_destroy_inode,
 723        .evict_inode    = hugetlbfs_evict_inode,
 724        .statfs         = hugetlbfs_statfs,
 725        .put_super      = hugetlbfs_put_super,
 726        .show_options   = generic_show_options,
 727};
 728
 729static int
 730hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 731{
 732        char *p, *rest;
 733        substring_t args[MAX_OPT_ARGS];
 734        int option;
 735        unsigned long long size = 0;
 736        enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE;
 737
 738        if (!options)
 739                return 0;
 740
 741        while ((p = strsep(&options, ",")) != NULL) {
 742                int token;
 743                if (!*p)
 744                        continue;
 745
 746                token = match_token(p, tokens, args);
 747                switch (token) {
 748                case Opt_uid:
 749                        if (match_int(&args[0], &option))
 750                                goto bad_val;
 751                        pconfig->uid = option;
 752                        break;
 753
 754                case Opt_gid:
 755                        if (match_int(&args[0], &option))
 756                                goto bad_val;
 757                        pconfig->gid = option;
 758                        break;
 759
 760                case Opt_mode:
 761                        if (match_octal(&args[0], &option))
 762                                goto bad_val;
 763                        pconfig->mode = option & 01777U;
 764                        break;
 765
 766                case Opt_size: {
 767                        /* memparse() will accept a K/M/G without a digit */
 768                        if (!isdigit(*args[0].from))
 769                                goto bad_val;
 770                        size = memparse(args[0].from, &rest);
 771                        setsize = SIZE_STD;
 772                        if (*rest == '%')
 773                                setsize = SIZE_PERCENT;
 774                        break;
 775                }
 776
 777                case Opt_nr_inodes:
 778                        /* memparse() will accept a K/M/G without a digit */
 779                        if (!isdigit(*args[0].from))
 780                                goto bad_val;
 781                        pconfig->nr_inodes = memparse(args[0].from, &rest);
 782                        break;
 783
 784                case Opt_pagesize: {
 785                        unsigned long ps;
 786                        ps = memparse(args[0].from, &rest);
 787                        pconfig->hstate = size_to_hstate(ps);
 788                        if (!pconfig->hstate) {
 789                                printk(KERN_ERR
 790                                "hugetlbfs: Unsupported page size %lu MB\n",
 791                                        ps >> 20);
 792                                return -EINVAL;
 793                        }
 794                        break;
 795                }
 796
 797                default:
 798                        printk(KERN_ERR "hugetlbfs: Bad mount option: \"%s\"\n",
 799                                 p);
 800                        return -EINVAL;
 801                        break;
 802                }
 803        }
 804
 805        /* Do size after hstate is set up */
 806        if (setsize > NO_SIZE) {
 807                struct hstate *h = pconfig->hstate;
 808                if (setsize == SIZE_PERCENT) {
 809                        size <<= huge_page_shift(h);
 810                        size *= h->max_huge_pages;
 811                        do_div(size, 100);
 812                }
 813                pconfig->nr_blocks = (size >> huge_page_shift(h));
 814        }
 815
 816        return 0;
 817
 818bad_val:
 819        printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
 820               args[0].from, p);
 821        return -EINVAL;
 822}
 823
 824static int
 825hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 826{
 827        struct inode * inode;
 828        struct dentry * root;
 829        int ret;
 830        struct hugetlbfs_config config;
 831        struct hugetlbfs_sb_info *sbinfo;
 832
 833        save_mount_options(sb, data);
 834
 835        config.nr_blocks = -1; /* No limit on size by default */
 836        config.nr_inodes = -1; /* No limit on number of inodes by default */
 837        config.uid = current_fsuid();
 838        config.gid = current_fsgid();
 839        config.mode = 0755;
 840        config.hstate = &default_hstate;
 841        ret = hugetlbfs_parse_options(data, &config);
 842        if (ret)
 843                return ret;
 844
 845        sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
 846        if (!sbinfo)
 847                return -ENOMEM;
 848        sb->s_fs_info = sbinfo;
 849        sbinfo->hstate = config.hstate;
 850        spin_lock_init(&sbinfo->stat_lock);
 851        sbinfo->max_blocks = config.nr_blocks;
 852        sbinfo->free_blocks = config.nr_blocks;
 853        sbinfo->max_inodes = config.nr_inodes;
 854        sbinfo->free_inodes = config.nr_inodes;
 855        sb->s_maxbytes = MAX_LFS_FILESIZE;
 856        sb->s_blocksize = huge_page_size(config.hstate);
 857        sb->s_blocksize_bits = huge_page_shift(config.hstate);
 858        sb->s_magic = HUGETLBFS_MAGIC;
 859        sb->s_op = &hugetlbfs_ops;
 860        sb->s_time_gran = 1;
 861        inode = hugetlbfs_get_root(sb, &config);
 862        if (!inode)
 863                goto out_free;
 864
 865        root = d_alloc_root(inode);
 866        if (!root) {
 867                iput(inode);
 868                goto out_free;
 869        }
 870        sb->s_root = root;
 871        return 0;
 872out_free:
 873        kfree(sbinfo);
 874        return -ENOMEM;
 875}
 876
 877int hugetlb_get_quota(struct address_space *mapping, long delta)
 878{
 879        int ret = 0;
 880        struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 881
 882        if (sbinfo->free_blocks > -1) {
 883                spin_lock(&sbinfo->stat_lock);
 884                if (sbinfo->free_blocks - delta >= 0)
 885                        sbinfo->free_blocks -= delta;
 886                else
 887                        ret = -ENOMEM;
 888                spin_unlock(&sbinfo->stat_lock);
 889        }
 890
 891        return ret;
 892}
 893
 894void hugetlb_put_quota(struct address_space *mapping, long delta)
 895{
 896        struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(mapping->host->i_sb);
 897
 898        if (sbinfo->free_blocks > -1) {
 899                spin_lock(&sbinfo->stat_lock);
 900                sbinfo->free_blocks += delta;
 901                spin_unlock(&sbinfo->stat_lock);
 902        }
 903}
 904
 905static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
 906        int flags, const char *dev_name, void *data)
 907{
 908        return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
 909}
 910
 911static struct file_system_type hugetlbfs_fs_type = {
 912        .name           = "hugetlbfs",
 913        .mount          = hugetlbfs_mount,
 914        .kill_sb        = kill_litter_super,
 915};
 916
 917static struct vfsmount *hugetlbfs_vfsmount;
 918
 919static int can_do_hugetlb_shm(void)
 920{
 921        return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 922}
 923
 924struct file *hugetlb_file_setup(const char *name, size_t size,
 925                                vm_flags_t acctflag,
 926                                struct user_struct **user, int creat_flags)
 927{
 928        int error = -ENOMEM;
 929        struct file *file;
 930        struct inode *inode;
 931        struct path path;
 932        struct dentry *root;
 933        struct qstr quick_string;
 934
 935        *user = NULL;
 936        if (!hugetlbfs_vfsmount)
 937                return ERR_PTR(-ENOENT);
 938
 939        if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
 940                *user = current_user();
 941                if (user_shm_lock(size, *user)) {
 942                        printk_once(KERN_WARNING "Using mlock ulimits for SHM_HUGETLB is deprecated\n");
 943                } else {
 944                        *user = NULL;
 945                        return ERR_PTR(-EPERM);
 946                }
 947        }
 948
 949        root = hugetlbfs_vfsmount->mnt_root;
 950        quick_string.name = name;
 951        quick_string.len = strlen(quick_string.name);
 952        quick_string.hash = 0;
 953        path.dentry = d_alloc(root, &quick_string);
 954        if (!path.dentry)
 955                goto out_shm_unlock;
 956
 957        path.mnt = mntget(hugetlbfs_vfsmount);
 958        error = -ENOSPC;
 959        inode = hugetlbfs_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0);
 960        if (!inode)
 961                goto out_dentry;
 962
 963        error = -ENOMEM;
 964        if (hugetlb_reserve_pages(inode, 0,
 965                        size >> huge_page_shift(hstate_inode(inode)), NULL,
 966                        acctflag))
 967                goto out_inode;
 968
 969        d_instantiate(path.dentry, inode);
 970        inode->i_size = size;
 971        clear_nlink(inode);
 972
 973        error = -ENFILE;
 974        file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
 975                        &hugetlbfs_file_operations);
 976        if (!file)
 977                goto out_dentry; /* inode is already attached */
 978
 979        return file;
 980
 981out_inode:
 982        iput(inode);
 983out_dentry:
 984        path_put(&path);
 985out_shm_unlock:
 986        if (*user) {
 987                user_shm_unlock(size, *user);
 988                *user = NULL;
 989        }
 990        return ERR_PTR(error);
 991}
 992
 993static int __init init_hugetlbfs_fs(void)
 994{
 995        int error;
 996        struct vfsmount *vfsmount;
 997
 998        error = bdi_init(&hugetlbfs_backing_dev_info);
 999        if (error)
1000                return error;
1001
1002        hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1003                                        sizeof(struct hugetlbfs_inode_info),
1004                                        0, 0, init_once);
1005        if (hugetlbfs_inode_cachep == NULL)
1006                goto out2;
1007
1008        error = register_filesystem(&hugetlbfs_fs_type);
1009        if (error)
1010                goto out;
1011
1012        vfsmount = kern_mount(&hugetlbfs_fs_type);
1013
1014        if (!IS_ERR(vfsmount)) {
1015                hugetlbfs_vfsmount = vfsmount;
1016                return 0;
1017        }
1018
1019        error = PTR_ERR(vfsmount);
1020
1021 out:
1022        if (error)
1023                kmem_cache_destroy(hugetlbfs_inode_cachep);
1024 out2:
1025        bdi_destroy(&hugetlbfs_backing_dev_info);
1026        return error;
1027}
1028
1029static void __exit exit_hugetlbfs_fs(void)
1030{
1031        kmem_cache_destroy(hugetlbfs_inode_cachep);
1032        kern_unmount(hugetlbfs_vfsmount);
1033        unregister_filesystem(&hugetlbfs_fs_type);
1034        bdi_destroy(&hugetlbfs_backing_dev_info);
1035}
1036
1037module_init(init_hugetlbfs_fs)
1038module_exit(exit_hugetlbfs_fs)
1039
1040MODULE_LICENSE("GPL");
1041
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.