linux/fs/ubifs/file.c
<<
>>
Prefs
   1/*
   2 * This file is part of UBIFS.
   3 *
   4 * Copyright (C) 2006-2008 Nokia Corporation.
   5 *
   6 * This program is free software; you can redistribute it and/or modify it
   7 * under the terms of the GNU General Public License version 2 as published by
   8 * the Free Software Foundation.
   9 *
  10 * This program is distributed in the hope that it will be useful, but WITHOUT
  11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13 * more details.
  14 *
  15 * You should have received a copy of the GNU General Public License along with
  16 * this program; if not, write to the Free Software Foundation, Inc., 51
  17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  18 *
  19 * Authors: Artem Bityutskiy (Битюцкий Артём)
  20 *          Adrian Hunter
  21 */
  22
  23/*
  24 * This file implements VFS file and inode operations for regular files, device
  25 * nodes and symlinks as well as address space operations.
  26 *
  27 * UBIFS uses 2 page flags: @PG_private and @PG_checked. @PG_private is set if
  28 * the page is dirty and is used for optimization purposes - dirty pages are
  29 * not budgeted so the flag shows that 'ubifs_write_end()' should not release
  30 * the budget for this page. The @PG_checked flag is set if full budgeting is
  31 * required for the page e.g., when it corresponds to a file hole or it is
  32 * beyond the file size. The budgeting is done in 'ubifs_write_begin()', because
  33 * it is OK to fail in this function, and the budget is released in
  34 * 'ubifs_write_end()'. So the @PG_private and @PG_checked flags carry
  35 * information about how the page was budgeted, to make it possible to release
  36 * the budget properly.
  37 *
  38 * A thing to keep in mind: inode @i_mutex is locked in most VFS operations we
  39 * implement. However, this is not true for 'ubifs_writepage()', which may be
  40 * called with @i_mutex unlocked. For example, when flusher thread is doing
  41 * background write-back, it calls 'ubifs_writepage()' with unlocked @i_mutex.
  42 * At "normal" work-paths the @i_mutex is locked in 'ubifs_writepage()', e.g.
  43 * in the "sys_write -> alloc_pages -> direct reclaim path". So, in
  44 * 'ubifs_writepage()' we are only guaranteed that the page is locked.
  45 *
  46 * Similarly, @i_mutex is not always locked in 'ubifs_readpage()', e.g., the
  47 * read-ahead path does not lock it ("sys_read -> generic_file_aio_read ->
  48 * ondemand_readahead -> readpage"). In case of readahead, @I_SYNC flag is not
  49 * set as well. However, UBIFS disables readahead.
  50 */
  51
  52#include "ubifs.h"
  53#include <linux/mount.h>
  54#include <linux/namei.h>
  55#include <linux/slab.h>
  56
  57static int read_block(struct inode *inode, void *addr, unsigned int block,
  58                      struct ubifs_data_node *dn)
  59{
  60        struct ubifs_info *c = inode->i_sb->s_fs_info;
  61        int err, len, out_len;
  62        union ubifs_key key;
  63        unsigned int dlen;
  64
  65        data_key_init(c, &key, inode->i_ino, block);
  66        err = ubifs_tnc_lookup(c, &key, dn);
  67        if (err) {
  68                if (err == -ENOENT)
  69                        /* Not found, so it must be a hole */
  70                        memset(addr, 0, UBIFS_BLOCK_SIZE);
  71                return err;
  72        }
  73
  74        ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
  75                     ubifs_inode(inode)->creat_sqnum);
  76        len = le32_to_cpu(dn->size);
  77        if (len <= 0 || len > UBIFS_BLOCK_SIZE)
  78                goto dump;
  79
  80        dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
  81        out_len = UBIFS_BLOCK_SIZE;
  82        err = ubifs_decompress(&dn->data, dlen, addr, &out_len,
  83                               le16_to_cpu(dn->compr_type));
  84        if (err || len != out_len)
  85                goto dump;
  86
  87        /*
  88         * Data length can be less than a full block, even for blocks that are
  89         * not the last in the file (e.g., as a result of making a hole and
  90         * appending data). Ensure that the remainder is zeroed out.
  91         */
  92        if (len < UBIFS_BLOCK_SIZE)
  93                memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
  94
  95        return 0;
  96
  97dump:
  98        ubifs_err("bad data node (block %u, inode %lu)",
  99                  block, inode->i_ino);
 100        ubifs_dump_node(c, dn);
 101        return -EINVAL;
 102}
 103
 104static int do_readpage(struct page *page)
 105{
 106        void *addr;
 107        int err = 0, i;
 108        unsigned int block, beyond;
 109        struct ubifs_data_node *dn;
 110        struct inode *inode = page->mapping->host;
 111        loff_t i_size = i_size_read(inode);
 112
 113        dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
 114                inode->i_ino, page->index, i_size, page->flags);
 115        ubifs_assert(!PageChecked(page));
 116        ubifs_assert(!PagePrivate(page));
 117
 118        addr = kmap(page);
 119
 120        block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
 121        beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
 122        if (block >= beyond) {
 123                /* Reading beyond inode */
 124                SetPageChecked(page);
 125                memset(addr, 0, PAGE_CACHE_SIZE);
 126                goto out;
 127        }
 128
 129        dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
 130        if (!dn) {
 131                err = -ENOMEM;
 132                goto error;
 133        }
 134
 135        i = 0;
 136        while (1) {
 137                int ret;
 138
 139                if (block >= beyond) {
 140                        /* Reading beyond inode */
 141                        err = -ENOENT;
 142                        memset(addr, 0, UBIFS_BLOCK_SIZE);
 143                } else {
 144                        ret = read_block(inode, addr, block, dn);
 145                        if (ret) {
 146                                err = ret;
 147                                if (err != -ENOENT)
 148                                        break;
 149                        } else if (block + 1 == beyond) {
 150                                int dlen = le32_to_cpu(dn->size);
 151                                int ilen = i_size & (UBIFS_BLOCK_SIZE - 1);
 152
 153                                if (ilen && ilen < dlen)
 154                                        memset(addr + ilen, 0, dlen - ilen);
 155                        }
 156                }
 157                if (++i >= UBIFS_BLOCKS_PER_PAGE)
 158                        break;
 159                block += 1;
 160                addr += UBIFS_BLOCK_SIZE;
 161        }
 162        if (err) {
 163                if (err == -ENOENT) {
 164                        /* Not found, so it must be a hole */
 165                        SetPageChecked(page);
 166                        dbg_gen("hole");
 167                        goto out_free;
 168                }
 169                ubifs_err("cannot read page %lu of inode %lu, error %d",
 170                          page->index, inode->i_ino, err);
 171                goto error;
 172        }
 173
 174out_free:
 175        kfree(dn);
 176out:
 177        SetPageUptodate(page);
 178        ClearPageError(page);
 179        flush_dcache_page(page);
 180        kunmap(page);
 181        return 0;
 182
 183error:
 184        kfree(dn);
 185        ClearPageUptodate(page);
 186        SetPageError(page);
 187        flush_dcache_page(page);
 188        kunmap(page);
 189        return err;
 190}
 191
 192/**
 193 * release_new_page_budget - release budget of a new page.
 194 * @c: UBIFS file-system description object
 195 *
 196 * This is a helper function which releases budget corresponding to the budget
 197 * of one new page of data.
 198 */
 199static void release_new_page_budget(struct ubifs_info *c)
 200{
 201        struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
 202
 203        ubifs_release_budget(c, &req);
 204}
 205
 206/**
 207 * release_existing_page_budget - release budget of an existing page.
 208 * @c: UBIFS file-system description object
 209 *
 210 * This is a helper function which releases budget corresponding to the budget
 211 * of changing one one page of data which already exists on the flash media.
 212 */
 213static void release_existing_page_budget(struct ubifs_info *c)
 214{
 215        struct ubifs_budget_req req = { .dd_growth = c->bi.page_budget};
 216
 217        ubifs_release_budget(c, &req);
 218}
 219
 220static int write_begin_slow(struct address_space *mapping,
 221                            loff_t pos, unsigned len, struct page **pagep,
 222                            unsigned flags)
 223{
 224        struct inode *inode = mapping->host;
 225        struct ubifs_info *c = inode->i_sb->s_fs_info;
 226        pgoff_t index = pos >> PAGE_CACHE_SHIFT;
 227        struct ubifs_budget_req req = { .new_page = 1 };
 228        int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
 229        struct page *page;
 230
 231        dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
 232                inode->i_ino, pos, len, inode->i_size);
 233
 234        /*
 235         * At the slow path we have to budget before locking the page, because
 236         * budgeting may force write-back, which would wait on locked pages and
 237         * deadlock if we had the page locked. At this point we do not know
 238         * anything about the page, so assume that this is a new page which is
 239         * written to a hole. This corresponds to largest budget. Later the
 240         * budget will be amended if this is not true.
 241         */
 242        if (appending)
 243                /* We are appending data, budget for inode change */
 244                req.dirtied_ino = 1;
 245
 246        err = ubifs_budget_space(c, &req);
 247        if (unlikely(err))
 248                return err;
 249
 250        page = grab_cache_page_write_begin(mapping, index, flags);
 251        if (unlikely(!page)) {
 252                ubifs_release_budget(c, &req);
 253                return -ENOMEM;
 254        }
 255
 256        if (!PageUptodate(page)) {
 257                if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE)
 258                        SetPageChecked(page);
 259                else {
 260                        err = do_readpage(page);
 261                        if (err) {
 262                                unlock_page(page);
 263                                page_cache_release(page);
 264                                return err;
 265                        }
 266                }
 267
 268                SetPageUptodate(page);
 269                ClearPageError(page);
 270        }
 271
 272        if (PagePrivate(page))
 273                /*
 274                 * The page is dirty, which means it was budgeted twice:
 275                 *   o first time the budget was allocated by the task which
 276                 *     made the page dirty and set the PG_private flag;
 277                 *   o and then we budgeted for it for the second time at the
 278                 *     very beginning of this function.
 279                 *
 280                 * So what we have to do is to release the page budget we
 281                 * allocated.
 282                 */
 283                release_new_page_budget(c);
 284        else if (!PageChecked(page))
 285                /*
 286                 * We are changing a page which already exists on the media.
 287                 * This means that changing the page does not make the amount
 288                 * of indexing information larger, and this part of the budget
 289                 * which we have already acquired may be released.
 290                 */
 291                ubifs_convert_page_budget(c);
 292
 293        if (appending) {
 294                struct ubifs_inode *ui = ubifs_inode(inode);
 295
 296                /*
 297                 * 'ubifs_write_end()' is optimized from the fast-path part of
 298                 * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
 299                 * if data is appended.
 300                 */
 301                mutex_lock(&ui->ui_mutex);
 302                if (ui->dirty)
 303                        /*
 304                         * The inode is dirty already, so we may free the
 305                         * budget we allocated.
 306                         */
 307                        ubifs_release_dirty_inode_budget(c, ui);
 308        }
 309
 310        *pagep = page;
 311        return 0;
 312}
 313
 314/**
 315 * allocate_budget - allocate budget for 'ubifs_write_begin()'.
 316 * @c: UBIFS file-system description object
 317 * @page: page to allocate budget for
 318 * @ui: UBIFS inode object the page belongs to
 319 * @appending: non-zero if the page is appended
 320 *
 321 * This is a helper function for 'ubifs_write_begin()' which allocates budget
 322 * for the operation. The budget is allocated differently depending on whether
 323 * this is appending, whether the page is dirty or not, and so on. This
 324 * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
 325 * in case of success and %-ENOSPC in case of failure.
 326 */
 327static int allocate_budget(struct ubifs_info *c, struct page *page,
 328                           struct ubifs_inode *ui, int appending)
 329{
 330        struct ubifs_budget_req req = { .fast = 1 };
 331
 332        if (PagePrivate(page)) {
 333                if (!appending)
 334                        /*
 335                         * The page is dirty and we are not appending, which
 336                         * means no budget is needed at all.
 337                         */
 338                        return 0;
 339
 340                mutex_lock(&ui->ui_mutex);
 341                if (ui->dirty)
 342                        /*
 343                         * The page is dirty and we are appending, so the inode
 344                         * has to be marked as dirty. However, it is already
 345                         * dirty, so we do not need any budget. We may return,
 346                         * but @ui->ui_mutex hast to be left locked because we
 347                         * should prevent write-back from flushing the inode
 348                         * and freeing the budget. The lock will be released in
 349                         * 'ubifs_write_end()'.
 350                         */
 351                        return 0;
 352
 353                /*
 354                 * The page is dirty, we are appending, the inode is clean, so
 355                 * we need to budget the inode change.
 356                 */
 357                req.dirtied_ino = 1;
 358        } else {
 359                if (PageChecked(page))
 360                        /*
 361                         * The page corresponds to a hole and does not
 362                         * exist on the media. So changing it makes
 363                         * make the amount of indexing information
 364                         * larger, and we have to budget for a new
 365                         * page.
 366                         */
 367                        req.new_page = 1;
 368                else
 369                        /*
 370                         * Not a hole, the change will not add any new
 371                         * indexing information, budget for page
 372                         * change.
 373                         */
 374                        req.dirtied_page = 1;
 375
 376                if (appending) {
 377                        mutex_lock(&ui->ui_mutex);
 378                        if (!ui->dirty)
 379                                /*
 380                                 * The inode is clean but we will have to mark
 381                                 * it as dirty because we are appending. This
 382                                 * needs a budget.
 383                                 */
 384                                req.dirtied_ino = 1;
 385                }
 386        }
 387
 388        return ubifs_budget_space(c, &req);
 389}
 390
 391/*
 392 * This function is called when a page of data is going to be written. Since
 393 * the page of data will not necessarily go to the flash straight away, UBIFS
 394 * has to reserve space on the media for it, which is done by means of
 395 * budgeting.
 396 *
 397 * This is the hot-path of the file-system and we are trying to optimize it as
 398 * much as possible. For this reasons it is split on 2 parts - slow and fast.
 399 *
 400 * There many budgeting cases:
 401 *     o a new page is appended - we have to budget for a new page and for
 402 *       changing the inode; however, if the inode is already dirty, there is
 403 *       no need to budget for it;
 404 *     o an existing clean page is changed - we have budget for it; if the page
 405 *       does not exist on the media (a hole), we have to budget for a new
 406 *       page; otherwise, we may budget for changing an existing page; the
 407 *       difference between these cases is that changing an existing page does
 408 *       not introduce anything new to the FS indexing information, so it does
 409 *       not grow, and smaller budget is acquired in this case;
 410 *     o an existing dirty page is changed - no need to budget at all, because
 411 *       the page budget has been acquired by earlier, when the page has been
 412 *       marked dirty.
 413 *
 414 * UBIFS budgeting sub-system may force write-back if it thinks there is no
 415 * space to reserve. This imposes some locking restrictions and makes it
 416 * impossible to take into account the above cases, and makes it impossible to
 417 * optimize budgeting.
 418 *
 419 * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
 420 * there is a plenty of flash space and the budget will be acquired quickly,
 421 * without forcing write-back. The slow path does not make this assumption.
 422 */
 423static int ubifs_write_begin(struct file *file, struct address_space *mapping,
 424                             loff_t pos, unsigned len, unsigned flags,
 425                             struct page **pagep, void **fsdata)
 426{
 427        struct inode *inode = mapping->host;
 428        struct ubifs_info *c = inode->i_sb->s_fs_info;
 429        struct ubifs_inode *ui = ubifs_inode(inode);
 430        pgoff_t index = pos >> PAGE_CACHE_SHIFT;
 431        int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
 432        int skipped_read = 0;
 433        struct page *page;
 434
 435        ubifs_assert(ubifs_inode(inode)->ui_size == inode->i_size);
 436        ubifs_assert(!c->ro_media && !c->ro_mount);
 437
 438        if (unlikely(c->ro_error))
 439                return -EROFS;
 440
 441        /* Try out the fast-path part first */
 442        page = grab_cache_page_write_begin(mapping, index, flags);
 443        if (unlikely(!page))
 444                return -ENOMEM;
 445
 446        if (!PageUptodate(page)) {
 447                /* The page is not loaded from the flash */
 448                if (!(pos & ~PAGE_CACHE_MASK) && len == PAGE_CACHE_SIZE) {
 449                        /*
 450                         * We change whole page so no need to load it. But we
 451                         * do not know whether this page exists on the media or
 452                         * not, so we assume the latter because it requires
 453                         * larger budget. The assumption is that it is better
 454                         * to budget a bit more than to read the page from the
 455                         * media. Thus, we are setting the @PG_checked flag
 456                         * here.
 457                         */
 458                        SetPageChecked(page);
 459                        skipped_read = 1;
 460                } else {
 461                        err = do_readpage(page);
 462                        if (err) {
 463                                unlock_page(page);
 464                                page_cache_release(page);
 465                                return err;
 466                        }
 467                }
 468
 469                SetPageUptodate(page);
 470                ClearPageError(page);
 471        }
 472
 473        err = allocate_budget(c, page, ui, appending);
 474        if (unlikely(err)) {
 475                ubifs_assert(err == -ENOSPC);
 476                /*
 477                 * If we skipped reading the page because we were going to
 478                 * write all of it, then it is not up to date.
 479                 */
 480                if (skipped_read) {
 481                        ClearPageChecked(page);
 482                        ClearPageUptodate(page);
 483                }
 484                /*
 485                 * Budgeting failed which means it would have to force
 486                 * write-back but didn't, because we set the @fast flag in the
 487                 * request. Write-back cannot be done now, while we have the
 488                 * page locked, because it would deadlock. Unlock and free
 489                 * everything and fall-back to slow-path.
 490                 */
 491                if (appending) {
 492                        ubifs_assert(mutex_is_locked(&ui->ui_mutex));
 493                        mutex_unlock(&ui->ui_mutex);
 494                }
 495                unlock_page(page);
 496                page_cache_release(page);
 497
 498                return write_begin_slow(mapping, pos, len, pagep, flags);
 499        }
 500
 501        /*
 502         * Whee, we acquired budgeting quickly - without involving
 503         * garbage-collection, committing or forcing write-back. We return
 504         * with @ui->ui_mutex locked if we are appending pages, and unlocked
 505         * otherwise. This is an optimization (slightly hacky though).
 506         */
 507        *pagep = page;
 508        return 0;
 509
 510}
 511
 512/**
 513 * cancel_budget - cancel budget.
 514 * @c: UBIFS file-system description object
 515 * @page: page to cancel budget for
 516 * @ui: UBIFS inode object the page belongs to
 517 * @appending: non-zero if the page is appended
 518 *
 519 * This is a helper function for a page write operation. It unlocks the
 520 * @ui->ui_mutex in case of appending.
 521 */
 522static void cancel_budget(struct ubifs_info *c, struct page *page,
 523                          struct ubifs_inode *ui, int appending)
 524{
 525        if (appending) {
 526                if (!ui->dirty)
 527                        ubifs_release_dirty_inode_budget(c, ui);
 528                mutex_unlock(&ui->ui_mutex);
 529        }
 530        if (!PagePrivate(page)) {
 531                if (PageChecked(page))
 532                        release_new_page_budget(c);
 533                else
 534                        release_existing_page_budget(c);
 535        }
 536}
 537
 538static int ubifs_write_end(struct file *file, struct address_space *mapping,
 539                           loff_t pos, unsigned len, unsigned copied,
 540                           struct page *page, void *fsdata)
 541{
 542        struct inode *inode = mapping->host;
 543        struct ubifs_inode *ui = ubifs_inode(inode);
 544        struct ubifs_info *c = inode->i_sb->s_fs_info;
 545        loff_t end_pos = pos + len;
 546        int appending = !!(end_pos > inode->i_size);
 547
 548        dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
 549                inode->i_ino, pos, page->index, len, copied, inode->i_size);
 550
 551        if (unlikely(copied < len && len == PAGE_CACHE_SIZE)) {
 552                /*
 553                 * VFS copied less data to the page that it intended and
 554                 * declared in its '->write_begin()' call via the @len
 555                 * argument. If the page was not up-to-date, and @len was
 556                 * @PAGE_CACHE_SIZE, the 'ubifs_write_begin()' function did
 557                 * not load it from the media (for optimization reasons). This
 558                 * means that part of the page contains garbage. So read the
 559                 * page now.
 560                 */
 561                dbg_gen("copied %d instead of %d, read page and repeat",
 562                        copied, len);
 563                cancel_budget(c, page, ui, appending);
 564                ClearPageChecked(page);
 565
 566                /*
 567                 * Return 0 to force VFS to repeat the whole operation, or the
 568                 * error code if 'do_readpage()' fails.
 569                 */
 570                copied = do_readpage(page);
 571                goto out;
 572        }
 573
 574        if (!PagePrivate(page)) {
 575                SetPagePrivate(page);
 576                atomic_long_inc(&c->dirty_pg_cnt);
 577                __set_page_dirty_nobuffers(page);
 578        }
 579
 580        if (appending) {
 581                i_size_write(inode, end_pos);
 582                ui->ui_size = end_pos;
 583                /*
 584                 * Note, we do not set @I_DIRTY_PAGES (which means that the
 585                 * inode has dirty pages), this has been done in
 586                 * '__set_page_dirty_nobuffers()'.
 587                 */
 588                __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
 589                ubifs_assert(mutex_is_locked(&ui->ui_mutex));
 590                mutex_unlock(&ui->ui_mutex);
 591        }
 592
 593out:
 594        unlock_page(page);
 595        page_cache_release(page);
 596        return copied;
 597}
 598
 599/**
 600 * populate_page - copy data nodes into a page for bulk-read.
 601 * @c: UBIFS file-system description object
 602 * @page: page
 603 * @bu: bulk-read information
 604 * @n: next zbranch slot
 605 *
 606 * This function returns %0 on success and a negative error code on failure.
 607 */
 608static int populate_page(struct ubifs_info *c, struct page *page,
 609                         struct bu_info *bu, int *n)
 610{
 611        int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
 612        struct inode *inode = page->mapping->host;
 613        loff_t i_size = i_size_read(inode);
 614        unsigned int page_block;
 615        void *addr, *zaddr;
 616        pgoff_t end_index;
 617
 618        dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
 619                inode->i_ino, page->index, i_size, page->flags);
 620
 621        addr = zaddr = kmap(page);
 622
 623        end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
 624        if (!i_size || page->index > end_index) {
 625                hole = 1;
 626                memset(addr, 0, PAGE_CACHE_SIZE);
 627                goto out_hole;
 628        }
 629
 630        page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
 631        while (1) {
 632                int err, len, out_len, dlen;
 633
 634                if (nn >= bu->cnt) {
 635                        hole = 1;
 636                        memset(addr, 0, UBIFS_BLOCK_SIZE);
 637                } else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
 638                        struct ubifs_data_node *dn;
 639
 640                        dn = bu->buf + (bu->zbranch[nn].offs - offs);
 641
 642                        ubifs_assert(le64_to_cpu(dn->ch.sqnum) >
 643                                     ubifs_inode(inode)->creat_sqnum);
 644
 645                        len = le32_to_cpu(dn->size);
 646                        if (len <= 0 || len > UBIFS_BLOCK_SIZE)
 647                                goto out_err;
 648
 649                        dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
 650                        out_len = UBIFS_BLOCK_SIZE;
 651                        err = ubifs_decompress(&dn->data, dlen, addr, &out_len,
 652                                               le16_to_cpu(dn->compr_type));
 653                        if (err || len != out_len)
 654                                goto out_err;
 655
 656                        if (len < UBIFS_BLOCK_SIZE)
 657                                memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
 658
 659                        nn += 1;
 660                        read = (i << UBIFS_BLOCK_SHIFT) + len;
 661                } else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
 662                        nn += 1;
 663                        continue;
 664                } else {
 665                        hole = 1;
 666                        memset(addr, 0, UBIFS_BLOCK_SIZE);
 667                }
 668                if (++i >= UBIFS_BLOCKS_PER_PAGE)
 669                        break;
 670                addr += UBIFS_BLOCK_SIZE;
 671                page_block += 1;
 672        }
 673
 674        if (end_index == page->index) {
 675                int len = i_size & (PAGE_CACHE_SIZE - 1);
 676
 677                if (len && len < read)
 678                        memset(zaddr + len, 0, read - len);
 679        }
 680
 681out_hole:
 682        if (hole) {
 683                SetPageChecked(page);
 684                dbg_gen("hole");
 685        }
 686
 687        SetPageUptodate(page);
 688        ClearPageError(page);
 689        flush_dcache_page(page);
 690        kunmap(page);
 691        *n = nn;
 692        return 0;
 693
 694out_err:
 695        ClearPageUptodate(page);
 696        SetPageError(page);
 697        flush_dcache_page(page);
 698        kunmap(page);
 699        ubifs_err("bad data node (block %u, inode %lu)",
 700                  page_block, inode->i_ino);
 701        return -EINVAL;
 702}
 703
 704/**
 705 * ubifs_do_bulk_read - do bulk-read.
 706 * @c: UBIFS file-system description object
 707 * @bu: bulk-read information
 708 * @page1: first page to read
 709 *
 710 * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
 711 */
 712static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
 713                              struct page *page1)
 714{
 715        pgoff_t offset = page1->index, end_index;
 716        struct address_space *mapping = page1->mapping;
 717        struct inode *inode = mapping->host;
 718        struct ubifs_inode *ui = ubifs_inode(inode);
 719        int err, page_idx, page_cnt, ret = 0, n = 0;
 720        int allocate = bu->buf ? 0 : 1;
 721        loff_t isize;
 722
 723        err = ubifs_tnc_get_bu_keys(c, bu);
 724        if (err)
 725                goto out_warn;
 726
 727        if (bu->eof) {
 728                /* Turn off bulk-read at the end of the file */
 729                ui->read_in_a_row = 1;
 730                ui->bulk_read = 0;
 731        }
 732
 733        page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
 734        if (!page_cnt) {
 735                /*
 736                 * This happens when there are multiple blocks per page and the
 737                 * blocks for the first page we are looking for, are not
 738                 * together. If all the pages were like this, bulk-read would
 739                 * reduce performance, so we turn it off for a while.
 740                 */
 741                goto out_bu_off;
 742        }
 743
 744        if (bu->cnt) {
 745                if (allocate) {
 746                        /*
 747                         * Allocate bulk-read buffer depending on how many data
 748                         * nodes we are going to read.
 749                         */
 750                        bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
 751                                      bu->zbranch[bu->cnt - 1].len -
 752                                      bu->zbranch[0].offs;
 753                        ubifs_assert(bu->buf_len > 0);
 754                        ubifs_assert(bu->buf_len <= c->leb_size);
 755                        bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
 756                        if (!bu->buf)
 757                                goto out_bu_off;
 758                }
 759
 760                err = ubifs_tnc_bulk_read(c, bu);
 761                if (err)
 762                        goto out_warn;
 763        }
 764
 765        err = populate_page(c, page1, bu, &n);
 766        if (err)
 767                goto out_warn;
 768
 769        unlock_page(page1);
 770        ret = 1;
 771
 772        isize = i_size_read(inode);
 773        if (isize == 0)
 774                goto out_free;
 775        end_index = ((isize - 1) >> PAGE_CACHE_SHIFT);
 776
 777        for (page_idx = 1; page_idx < page_cnt; page_idx++) {
 778                pgoff_t page_offset = offset + page_idx;
 779                struct page *page;
 780
 781                if (page_offset > end_index)
 782                        break;
 783                page = find_or_create_page(mapping, page_offset,
 784                                           GFP_NOFS | __GFP_COLD);
 785                if (!page)
 786                        break;
 787                if (!PageUptodate(page))
 788                        err = populate_page(c, page, bu, &n);
 789                unlock_page(page);
 790                page_cache_release(page);
 791                if (err)
 792                        break;
 793        }
 794
 795        ui->last_page_read = offset + page_idx - 1;
 796
 797out_free:
 798        if (allocate)
 799                kfree(bu->buf);
 800        return ret;
 801
 802out_warn:
 803        ubifs_warn("ignoring error %d and skipping bulk-read", err);
 804        goto out_free;
 805
 806out_bu_off:
 807        ui->read_in_a_row = ui->bulk_read = 0;
 808        goto out_free;
 809}
 810
 811/**
 812 * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
 813 * @page: page from which to start bulk-read.
 814 *
 815 * Some flash media are capable of reading sequentially at faster rates. UBIFS
 816 * bulk-read facility is designed to take advantage of that, by reading in one
 817 * go consecutive data nodes that are also located consecutively in the same
 818 * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
 819 */
 820static int ubifs_bulk_read(struct page *page)
 821{
 822        struct inode *inode = page->mapping->host;
 823        struct ubifs_info *c = inode->i_sb->s_fs_info;
 824        struct ubifs_inode *ui = ubifs_inode(inode);
 825        pgoff_t index = page->index, last_page_read = ui->last_page_read;
 826        struct bu_info *bu;
 827        int err = 0, allocated = 0;
 828
 829        ui->last_page_read = index;
 830        if (!c->bulk_read)
 831                return 0;
 832
 833        /*
 834         * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
 835         * so don't bother if we cannot lock the mutex.
 836         */
 837        if (!mutex_trylock(&ui->ui_mutex))
 838                return 0;
 839
 840        if (index != last_page_read + 1) {
 841                /* Turn off bulk-read if we stop reading sequentially */
 842                ui->read_in_a_row = 1;
 843                if (ui->bulk_read)
 844                        ui->bulk_read = 0;
 845                goto out_unlock;
 846        }
 847
 848        if (!ui->bulk_read) {
 849                ui->read_in_a_row += 1;
 850                if (ui->read_in_a_row < 3)
 851                        goto out_unlock;
 852                /* Three reads in a row, so switch on bulk-read */
 853                ui->bulk_read = 1;
 854        }
 855
 856        /*
 857         * If possible, try to use pre-allocated bulk-read information, which
 858         * is protected by @c->bu_mutex.
 859         */
 860        if (mutex_trylock(&c->bu_mutex))
 861                bu = &c->bu;
 862        else {
 863                bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
 864                if (!bu)
 865                        goto out_unlock;
 866
 867                bu->buf = NULL;
 868                allocated = 1;
 869        }
 870
 871        bu->buf_len = c->max_bu_buf_len;
 872        data_key_init(c, &bu->key, inode->i_ino,
 873                      page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
 874        err = ubifs_do_bulk_read(c, bu, page);
 875
 876        if (!allocated)
 877                mutex_unlock(&c->bu_mutex);
 878        else
 879                kfree(bu);
 880
 881out_unlock:
 882        mutex_unlock(&ui->ui_mutex);
 883        return err;
 884}
 885
 886static int ubifs_readpage(struct file *file, struct page *page)
 887{
 888        if (ubifs_bulk_read(page))
 889                return 0;
 890        do_readpage(page);
 891        unlock_page(page);
 892        return 0;
 893}
 894
 895static int do_writepage(struct page *page, int len)
 896{
 897        int err = 0, i, blen;
 898        unsigned int block;
 899        void *addr;
 900        union ubifs_key key;
 901        struct inode *inode = page->mapping->host;
 902        struct ubifs_info *c = inode->i_sb->s_fs_info;
 903
 904#ifdef UBIFS_DEBUG
 905        spin_lock(&ui->ui_lock);
 906        ubifs_assert(page->index <= ui->synced_i_size << PAGE_CACHE_SIZE);
 907        spin_unlock(&ui->ui_lock);
 908#endif
 909
 910        /* Update radix tree tags */
 911        set_page_writeback(page);
 912
 913        addr = kmap(page);
 914        block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
 915        i = 0;
 916        while (len) {
 917                blen = min_t(int, len, UBIFS_BLOCK_SIZE);
 918                data_key_init(c, &key, inode->i_ino, block);
 919                err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
 920                if (err)
 921                        break;
 922                if (++i >= UBIFS_BLOCKS_PER_PAGE)
 923                        break;
 924                block += 1;
 925                addr += blen;
 926                len -= blen;
 927        }
 928        if (err) {
 929                SetPageError(page);
 930                ubifs_err("cannot write page %lu of inode %lu, error %d",
 931                          page->index, inode->i_ino, err);
 932                ubifs_ro_mode(c, err);
 933        }
 934
 935        ubifs_assert(PagePrivate(page));
 936        if (PageChecked(page))
 937                release_new_page_budget(c);
 938        else
 939                release_existing_page_budget(c);
 940
 941        atomic_long_dec(&c->dirty_pg_cnt);
 942        ClearPagePrivate(page);
 943        ClearPageChecked(page);
 944
 945        kunmap(page);
 946        unlock_page(page);
 947        end_page_writeback(page);
 948        return err;
 949}
 950
 951/*
 952 * When writing-back dirty inodes, VFS first writes-back pages belonging to the
 953 * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
 954 * situation when a we have an inode with size 0, then a megabyte of data is
 955 * appended to the inode, then write-back starts and flushes some amount of the
 956 * dirty pages, the journal becomes full, commit happens and finishes, and then
 957 * an unclean reboot happens. When the file system is mounted next time, the
 958 * inode size would still be 0, but there would be many pages which are beyond
 959 * the inode size, they would be indexed and consume flash space. Because the
 960 * journal has been committed, the replay would not be able to detect this
 961 * situation and correct the inode size. This means UBIFS would have to scan
 962 * whole index and correct all inode sizes, which is long an unacceptable.
 963 *
 964 * To prevent situations like this, UBIFS writes pages back only if they are
 965 * within the last synchronized inode size, i.e. the size which has been
 966 * written to the flash media last time. Otherwise, UBIFS forces inode
 967 * write-back, thus making sure the on-flash inode contains current inode size,
 968 * and then keeps writing pages back.
 969 *
 970 * Some locking issues explanation. 'ubifs_writepage()' first is called with
 971 * the page locked, and it locks @ui_mutex. However, write-back does take inode
 972 * @i_mutex, which means other VFS operations may be run on this inode at the
 973 * same time. And the problematic one is truncation to smaller size, from where
 974 * we have to call 'truncate_setsize()', which first changes @inode->i_size,
 975 * then drops the truncated pages. And while dropping the pages, it takes the
 976 * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
 977 * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
 978 * This means that @inode->i_size is changed while @ui_mutex is unlocked.
 979 *
 980 * XXX(truncate): with the new truncate sequence this is not true anymore,
 981 * and the calls to truncate_setsize can be move around freely.  They should
 982 * be moved to the very end of the truncate sequence.
 983 *
 984 * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
 985 * inode size. How do we do this if @inode->i_size may became smaller while we
 986 * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
 987 * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
 988 * internally and updates it under @ui_mutex.
 989 *
 990 * Q: why we do not worry that if we race with truncation, we may end up with a
 991 * situation when the inode is truncated while we are in the middle of
 992 * 'do_writepage()', so we do write beyond inode size?
 993 * A: If we are in the middle of 'do_writepage()', truncation would be locked
 994 * on the page lock and it would not write the truncated inode node to the
 995 * journal before we have finished.
 996 */
 997static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
 998{
 999        struct inode *inode = page->mapping->host;
1000        struct ubifs_inode *ui = ubifs_inode(inode);
1001        loff_t i_size =  i_size_read(inode), synced_i_size;
1002        pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1003        int err, len = i_size & (PAGE_CACHE_SIZE - 1);
1004        void *kaddr;
1005
1006        dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1007                inode->i_ino, page->index, page->flags);
1008        ubifs_assert(PagePrivate(page));
1009
1010        /* Is the page fully outside @i_size? (truncate in progress) */
1011        if (page->index > end_index || (page->index == end_index && !len)) {
1012                err = 0;
1013                goto out_unlock;
1014        }
1015
1016        spin_lock(&ui->ui_lock);
1017        synced_i_size = ui->synced_i_size;
1018        spin_unlock(&ui->ui_lock);
1019
1020        /* Is the page fully inside @i_size? */
1021        if (page->index < end_index) {
1022                if (page->index >= synced_i_size >> PAGE_CACHE_SHIFT) {
1023                        err = inode->i_sb->s_op->write_inode(inode, NULL);
1024                        if (err)
1025                                goto out_unlock;
1026                        /*
1027                         * The inode has been written, but the write-buffer has
1028                         * not been synchronized, so in case of an unclean
1029                         * reboot we may end up with some pages beyond inode
1030                         * size, but they would be in the journal (because
1031                         * commit flushes write buffers) and recovery would deal
1032                         * with this.
1033                         */
1034                }
1035                return do_writepage(page, PAGE_CACHE_SIZE);
1036        }
1037
1038        /*
1039         * The page straddles @i_size. It must be zeroed out on each and every
1040         * writepage invocation because it may be mmapped. "A file is mapped
1041         * in multiples of the page size. For a file that is not a multiple of
1042         * the page size, the remaining memory is zeroed when mapped, and
1043         * writes to that region are not written out to the file."
1044         */
1045        kaddr = kmap_atomic(page);
1046        memset(kaddr + len, 0, PAGE_CACHE_SIZE - len);
1047        flush_dcache_page(page);
1048        kunmap_atomic(kaddr);
1049
1050        if (i_size > synced_i_size) {
1051                err = inode->i_sb->s_op->write_inode(inode, NULL);
1052                if (err)
1053                        goto out_unlock;
1054        }
1055
1056        return do_writepage(page, len);
1057
1058out_unlock:
1059        unlock_page(page);
1060        return err;
1061}
1062
1063/**
1064 * do_attr_changes - change inode attributes.
1065 * @inode: inode to change attributes for
1066 * @attr: describes attributes to change
1067 */
1068static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1069{
1070        if (attr->ia_valid & ATTR_UID)
1071                inode->i_uid = attr->ia_uid;
1072        if (attr->ia_valid & ATTR_GID)
1073                inode->i_gid = attr->ia_gid;
1074        if (attr->ia_valid & ATTR_ATIME)
1075                inode->i_atime = timespec_trunc(attr->ia_atime,
1076                                                inode->i_sb->s_time_gran);
1077        if (attr->ia_valid & ATTR_MTIME)
1078                inode->i_mtime = timespec_trunc(attr->ia_mtime,
1079                                                inode->i_sb->s_time_gran);
1080        if (attr->ia_valid & ATTR_CTIME)
1081                inode->i_ctime = timespec_trunc(attr->ia_ctime,
1082                                                inode->i_sb->s_time_gran);
1083        if (attr->ia_valid & ATTR_MODE) {
1084                umode_t mode = attr->ia_mode;
1085
1086                if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1087                        mode &= ~S_ISGID;
1088                inode->i_mode = mode;
1089        }
1090}
1091
1092/**
1093 * do_truncation - truncate an inode.
1094 * @c: UBIFS file-system description object
1095 * @inode: inode to truncate
1096 * @attr: inode attribute changes description
1097 *
1098 * This function implements VFS '->setattr()' call when the inode is truncated
1099 * to a smaller size. Returns zero in case of success and a negative error code
1100 * in case of failure.
1101 */
1102static int do_truncation(struct ubifs_info *c, struct inode *inode,
1103                         const struct iattr *attr)
1104{
1105        int err;
1106        struct ubifs_budget_req req;
1107        loff_t old_size = inode->i_size, new_size = attr->ia_size;
1108        int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1109        struct ubifs_inode *ui = ubifs_inode(inode);
1110
1111        dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1112        memset(&req, 0, sizeof(struct ubifs_budget_req));
1113
1114        /*
1115         * If this is truncation to a smaller size, and we do not truncate on a
1116         * block boundary, budget for changing one data block, because the last
1117         * block will be re-written.
1118         */
1119        if (new_size & (UBIFS_BLOCK_SIZE - 1))
1120                req.dirtied_page = 1;
1121
1122        req.dirtied_ino = 1;
1123        /* A funny way to budget for truncation node */
1124        req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1125        err = ubifs_budget_space(c, &req);
1126        if (err) {
1127                /*
1128                 * Treat truncations to zero as deletion and always allow them,
1129                 * just like we do for '->unlink()'.
1130                 */
1131                if (new_size || err != -ENOSPC)
1132                        return err;
1133                budgeted = 0;
1134        }
1135
1136        truncate_setsize(inode, new_size);
1137
1138        if (offset) {
1139                pgoff_t index = new_size >> PAGE_CACHE_SHIFT;
1140                struct page *page;
1141
1142                page = find_lock_page(inode->i_mapping, index);
1143                if (page) {
1144                        if (PageDirty(page)) {
1145                                /*
1146                                 * 'ubifs_jnl_truncate()' will try to truncate
1147                                 * the last data node, but it contains
1148                                 * out-of-date data because the page is dirty.
1149                                 * Write the page now, so that
1150                                 * 'ubifs_jnl_truncate()' will see an already
1151                                 * truncated (and up to date) data node.
1152                                 */
1153                                ubifs_assert(PagePrivate(page));
1154
1155                                clear_page_dirty_for_io(page);
1156                                if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1157                                        offset = new_size &
1158                                                 (PAGE_CACHE_SIZE - 1);
1159                                err = do_writepage(page, offset);
1160                                page_cache_release(page);
1161                                if (err)
1162                                        goto out_budg;
1163                                /*
1164                                 * We could now tell 'ubifs_jnl_truncate()' not
1165                                 * to read the last block.
1166                                 */
1167                        } else {
1168                                /*
1169                                 * We could 'kmap()' the page and pass the data
1170                                 * to 'ubifs_jnl_truncate()' to save it from
1171                                 * having to read it.
1172                                 */
1173                                unlock_page(page);
1174                                page_cache_release(page);
1175                        }
1176                }
1177        }
1178
1179        mutex_lock(&ui->ui_mutex);
1180        ui->ui_size = inode->i_size;
1181        /* Truncation changes inode [mc]time */
1182        inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1183        /* Other attributes may be changed at the same time as well */
1184        do_attr_changes(inode, attr);
1185        err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1186        mutex_unlock(&ui->ui_mutex);
1187
1188out_budg:
1189        if (budgeted)
1190                ubifs_release_budget(c, &req);
1191        else {
1192                c->bi.nospace = c->bi.nospace_rp = 0;
1193                smp_wmb();
1194        }
1195        return err;
1196}
1197
1198/**
1199 * do_setattr - change inode attributes.
1200 * @c: UBIFS file-system description object
1201 * @inode: inode to change attributes for
1202 * @attr: inode attribute changes description
1203 *
1204 * This function implements VFS '->setattr()' call for all cases except
1205 * truncations to smaller size. Returns zero in case of success and a negative
1206 * error code in case of failure.
1207 */
1208static int do_setattr(struct ubifs_info *c, struct inode *inode,
1209                      const struct iattr *attr)
1210{
1211        int err, release;
1212        loff_t new_size = attr->ia_size;
1213        struct ubifs_inode *ui = ubifs_inode(inode);
1214        struct ubifs_budget_req req = { .dirtied_ino = 1,
1215                                .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1216
1217        err = ubifs_budget_space(c, &req);
1218        if (err)
1219                return err;
1220
1221        if (attr->ia_valid & ATTR_SIZE) {
1222                dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1223                truncate_setsize(inode, new_size);
1224        }
1225
1226        mutex_lock(&ui->ui_mutex);
1227        if (attr->ia_valid & ATTR_SIZE) {
1228                /* Truncation changes inode [mc]time */
1229                inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1230                /* 'truncate_setsize()' changed @i_size, update @ui_size */
1231                ui->ui_size = inode->i_size;
1232        }
1233
1234        do_attr_changes(inode, attr);
1235
1236        release = ui->dirty;
1237        if (attr->ia_valid & ATTR_SIZE)
1238                /*
1239                 * Inode length changed, so we have to make sure
1240                 * @I_DIRTY_DATASYNC is set.
1241                 */
1242                 __mark_inode_dirty(inode, I_DIRTY_SYNC | I_DIRTY_DATASYNC);
1243        else
1244                mark_inode_dirty_sync(inode);
1245        mutex_unlock(&ui->ui_mutex);
1246
1247        if (release)
1248                ubifs_release_budget(c, &req);
1249        if (IS_SYNC(inode))
1250                err = inode->i_sb->s_op->write_inode(inode, NULL);
1251        return err;
1252}
1253
1254int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
1255{
1256        int err;
1257        struct inode *inode = dentry->d_inode;
1258        struct ubifs_info *c = inode->i_sb->s_fs_info;
1259
1260        dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1261                inode->i_ino, inode->i_mode, attr->ia_valid);
1262        err = inode_change_ok(inode, attr);
1263        if (err)
1264                return err;
1265
1266        err = dbg_check_synced_i_size(c, inode);
1267        if (err)
1268                return err;
1269
1270        if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1271                /* Truncation to a smaller size */
1272                err = do_truncation(c, inode, attr);
1273        else
1274                err = do_setattr(c, inode, attr);
1275
1276        return err;
1277}
1278
1279static void ubifs_invalidatepage(struct page *page, unsigned long offset)
1280{
1281        struct inode *inode = page->mapping->host;
1282        struct ubifs_info *c = inode->i_sb->s_fs_info;
1283
1284        ubifs_assert(PagePrivate(page));
1285        if (offset)
1286                /* Partial page remains dirty */
1287                return;
1288
1289        if (PageChecked(page))
1290                release_new_page_budget(c);
1291        else
1292                release_existing_page_budget(c);
1293
1294        atomic_long_dec(&c->dirty_pg_cnt);
1295        ClearPagePrivate(page);
1296        ClearPageChecked(page);
1297}
1298
1299static void *ubifs_follow_link(struct dentry *dentry, struct nameidata *nd)
1300{
1301        struct ubifs_inode *ui = ubifs_inode(dentry->d_inode);
1302
1303        nd_set_link(nd, ui->data);
1304        return NULL;
1305}
1306
1307int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1308{
1309        struct inode *inode = file->f_mapping->host;
1310        struct ubifs_info *c = inode->i_sb->s_fs_info;
1311        int err;
1312
1313        dbg_gen("syncing inode %lu", inode->i_ino);
1314
1315        if (c->ro_mount)
1316                /*
1317                 * For some really strange reasons VFS does not filter out
1318                 * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1319                 */
1320                return 0;
1321
1322        err = filemap_write_and_wait_range(inode->i_mapping, start, end);
1323        if (err)
1324                return err;
1325        mutex_lock(&inode->i_mutex);
1326
1327        /* Synchronize the inode unless this is a 'datasync()' call. */
1328        if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1329                err = inode->i_sb->s_op->write_inode(inode, NULL);
1330                if (err)
1331                        goto out;
1332        }
1333
1334        /*
1335         * Nodes related to this inode may still sit in a write-buffer. Flush
1336         * them.
1337         */
1338        err = ubifs_sync_wbufs_by_inode(c, inode);
1339out:
1340        mutex_unlock(&inode->i_mutex);
1341        return err;
1342}
1343
1344/**
1345 * mctime_update_needed - check if mtime or ctime update is needed.
1346 * @inode: the inode to do the check for
1347 * @now: current time
1348 *
1349 * This helper function checks if the inode mtime/ctime should be updated or
1350 * not. If current values of the time-stamps are within the UBIFS inode time
1351 * granularity, they are not updated. This is an optimization.
1352 */
1353static inline int mctime_update_needed(const struct inode *inode,
1354                                       const struct timespec *now)
1355{
1356        if (!timespec_equal(&inode->i_mtime, now) ||
1357            !timespec_equal(&inode->i_ctime, now))
1358                return 1;
1359        return 0;
1360}
1361
1362/**
1363 * update_ctime - update mtime and ctime of an inode.
1364 * @c: UBIFS file-system description object
1365 * @inode: inode to update
1366 *
1367 * This function updates mtime and ctime of the inode if it is not equivalent to
1368 * current time. Returns zero in case of success and a negative error code in
1369 * case of failure.
1370 */
1371static int update_mctime(struct ubifs_info *c, struct inode *inode)
1372{
1373        struct timespec now = ubifs_current_time(inode);
1374        struct ubifs_inode *ui = ubifs_inode(inode);
1375
1376        if (mctime_update_needed(inode, &now)) {
1377                int err, release;
1378                struct ubifs_budget_req req = { .dirtied_ino = 1,
1379                                .dirtied_ino_d = ALIGN(ui->data_len, 8) };
1380
1381                err = ubifs_budget_space(c, &req);
1382                if (err)
1383                        return err;
1384
1385                mutex_lock(&ui->ui_mutex);
1386                inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1387                release = ui->dirty;
1388                mark_inode_dirty_sync(inode);
1389                mutex_unlock(&ui->ui_mutex);
1390                if (release)
1391                        ubifs_release_budget(c, &req);
1392        }
1393
1394        return 0;
1395}
1396
1397static ssize_t ubifs_aio_write(struct kiocb *iocb, const struct iovec *iov,
1398                               unsigned long nr_segs, loff_t pos)
1399{
1400        int err;
1401        struct inode *inode = iocb->ki_filp->f_mapping->host;
1402        struct ubifs_info *c = inode->i_sb->s_fs_info;
1403
1404        err = update_mctime(c, inode);
1405        if (err)
1406                return err;
1407
1408        return generic_file_aio_write(iocb, iov, nr_segs, pos);
1409}
1410
1411static int ubifs_set_page_dirty(struct page *page)
1412{
1413        int ret;
1414
1415        ret = __set_page_dirty_nobuffers(page);
1416        /*
1417         * An attempt to dirty a page without budgeting for it - should not
1418         * happen.
1419         */
1420        ubifs_assert(ret == 0);
1421        return ret;
1422}
1423
1424static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1425{
1426        /*
1427         * An attempt to release a dirty page without budgeting for it - should
1428         * not happen.
1429         */
1430        if (PageWriteback(page))
1431                return 0;
1432        ubifs_assert(PagePrivate(page));
1433        ubifs_assert(0);
1434        ClearPagePrivate(page);
1435        ClearPageChecked(page);
1436        return 1;
1437}
1438
1439/*
1440 * mmap()d file has taken write protection fault and is being made writable.
1441 * UBIFS must ensure page is budgeted for.
1442 */
1443static int ubifs_vm_page_mkwrite(struct vm_area_struct *vma,
1444                                 struct vm_fault *vmf)
1445{
1446        struct page *page = vmf->page;
1447        struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1448        struct ubifs_info *c = inode->i_sb->s_fs_info;
1449        struct timespec now = ubifs_current_time(inode);
1450        struct ubifs_budget_req req = { .new_page = 1 };
1451        int err, update_time;
1452
1453        dbg_gen("ino %lu, pg %lu, i_size %lld", inode->i_ino, page->index,
1454                i_size_read(inode));
1455        ubifs_assert(!c->ro_media && !c->ro_mount);
1456
1457        if (unlikely(c->ro_error))
1458                return VM_FAULT_SIGBUS; /* -EROFS */
1459
1460        /*
1461         * We have not locked @page so far so we may budget for changing the
1462         * page. Note, we cannot do this after we locked the page, because
1463         * budgeting may cause write-back which would cause deadlock.
1464         *
1465         * At the moment we do not know whether the page is dirty or not, so we
1466         * assume that it is not and budget for a new page. We could look at
1467         * the @PG_private flag and figure this out, but we may race with write
1468         * back and the page state may change by the time we lock it, so this
1469         * would need additional care. We do not bother with this at the
1470         * moment, although it might be good idea to do. Instead, we allocate
1471         * budget for a new page and amend it later on if the page was in fact
1472         * dirty.
1473         *
1474         * The budgeting-related logic of this function is similar to what we
1475         * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1476         * for more comments.
1477         */
1478        update_time = mctime_update_needed(inode, &now);
1479        if (update_time)
1480                /*
1481                 * We have to change inode time stamp which requires extra
1482                 * budgeting.
1483                 */
1484                req.dirtied_ino = 1;
1485
1486        err = ubifs_budget_space(c, &req);
1487        if (unlikely(err)) {
1488                if (err == -ENOSPC)
1489                        ubifs_warn("out of space for mmapped file "
1490                                   "(inode number %lu)", inode->i_ino);
1491                return VM_FAULT_SIGBUS;
1492        }
1493
1494        lock_page(page);
1495        if (unlikely(page->mapping != inode->i_mapping ||
1496                     page_offset(page) > i_size_read(inode))) {
1497                /* Page got truncated out from underneath us */
1498                err = -EINVAL;
1499                goto out_unlock;
1500        }
1501
1502        if (PagePrivate(page))
1503                release_new_page_budget(c);
1504        else {
1505                if (!PageChecked(page))
1506                        ubifs_convert_page_budget(c);
1507                SetPagePrivate(page);
1508                atomic_long_inc(&c->dirty_pg_cnt);
1509                __set_page_dirty_nobuffers(page);
1510        }
1511
1512        if (update_time) {
1513                int release;
1514                struct ubifs_inode *ui = ubifs_inode(inode);
1515
1516                mutex_lock(&ui->ui_mutex);
1517                inode->i_mtime = inode->i_ctime = ubifs_current_time(inode);
1518                release = ui->dirty;
1519                mark_inode_dirty_sync(inode);
1520                mutex_unlock(&ui->ui_mutex);
1521                if (release)
1522                        ubifs_release_dirty_inode_budget(c, ui);
1523        }
1524
1525        unlock_page(page);
1526        return 0;
1527
1528out_unlock:
1529        unlock_page(page);
1530        ubifs_release_budget(c, &req);
1531        if (err)
1532                err = VM_FAULT_SIGBUS;
1533        return err;
1534}
1535
1536static const struct vm_operations_struct ubifs_file_vm_ops = {
1537        .fault        = filemap_fault,
1538        .page_mkwrite = ubifs_vm_page_mkwrite,
1539};
1540
1541static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1542{
1543        int err;
1544
1545        err = generic_file_mmap(file, vma);
1546        if (err)
1547                return err;
1548        vma->vm_ops = &ubifs_file_vm_ops;
1549        return 0;
1550}
1551
1552const struct address_space_operations ubifs_file_address_operations = {
1553        .readpage       = ubifs_readpage,
1554        .writepage      = ubifs_writepage,
1555        .write_begin    = ubifs_write_begin,
1556        .write_end      = ubifs_write_end,
1557        .invalidatepage = ubifs_invalidatepage,
1558        .set_page_dirty = ubifs_set_page_dirty,
1559        .releasepage    = ubifs_releasepage,
1560};
1561
1562const struct inode_operations ubifs_file_inode_operations = {
1563        .setattr     = ubifs_setattr,
1564        .getattr     = ubifs_getattr,
1565        .setxattr    = ubifs_setxattr,
1566        .getxattr    = ubifs_getxattr,
1567        .listxattr   = ubifs_listxattr,
1568        .removexattr = ubifs_removexattr,
1569};
1570
1571const struct inode_operations ubifs_symlink_inode_operations = {
1572        .readlink    = generic_readlink,
1573        .follow_link = ubifs_follow_link,
1574        .setattr     = ubifs_setattr,
1575        .getattr     = ubifs_getattr,
1576};
1577
1578const struct file_operations ubifs_file_operations = {
1579        .llseek         = generic_file_llseek,
1580        .read           = do_sync_read,
1581        .write          = do_sync_write,
1582        .aio_read       = generic_file_aio_read,
1583        .aio_write      = ubifs_aio_write,
1584        .mmap           = ubifs_file_mmap,
1585        .fsync          = ubifs_fsync,
1586        .unlocked_ioctl = ubifs_ioctl,
1587        .splice_read    = generic_file_splice_read,
1588        .splice_write   = generic_file_splice_write,
1589#ifdef CONFIG_COMPAT
1590        .compat_ioctl   = ubifs_compat_ioctl,
1591#endif
1592};
1593
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.