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