1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * This file is part of UBIFS.
   4 *
   5 * Copyright (C) 2006-2008 Nokia Corporation.
   6 *
   7 * Authors: Artem Bityutskiy (Битюцкий Артём)
   8 *          Adrian Hunter
   9 */
  10
  11/*
  12 * This file implements VFS file and inode operations for regular files, device
  13 * nodes and symlinks as well as address space operations.
  14 *
  15 * UBIFS uses 2 page flags: @PG_private and @PG_checked. @PG_private is set if
  16 * the page is dirty and is used for optimization purposes - dirty pages are
  17 * not budgeted so the flag shows that 'ubifs_write_end()' should not release
  18 * the budget for this page. The @PG_checked flag is set if full budgeting is
  19 * required for the page e.g., when it corresponds to a file hole or it is
  20 * beyond the file size. The budgeting is done in 'ubifs_write_begin()', because
  21 * it is OK to fail in this function, and the budget is released in
  22 * 'ubifs_write_end()'. So the @PG_private and @PG_checked flags carry
  23 * information about how the page was budgeted, to make it possible to release
  24 * the budget properly.
  25 *
  26 * A thing to keep in mind: inode @i_mutex is locked in most VFS operations we
  27 * implement. However, this is not true for 'ubifs_writepage()', which may be
  28 * called with @i_mutex unlocked. For example, when flusher thread is doing
  29 * background write-back, it calls 'ubifs_writepage()' with unlocked @i_mutex.
  30 * At "normal" work-paths the @i_mutex is locked in 'ubifs_writepage()', e.g.
  31 * in the "sys_write -> alloc_pages -> direct reclaim path". So, in
  32 * 'ubifs_writepage()' we are only guaranteed that the page is locked.
  33 *
  34 * Similarly, @i_mutex is not always locked in 'ubifs_readpage()', e.g., the
  35 * read-ahead path does not lock it ("sys_read -> generic_file_aio_read ->
  36 * ondemand_readahead -> readpage"). In case of readahead, @I_SYNC flag is not
  37 * set as well. However, UBIFS disables readahead.
  38 */
  39
  40#include "ubifs.h"
  41#include <linux/mount.h>
  42#include <linux/slab.h>
  43#include <linux/migrate.h>
  44
  45static int read_block(struct inode *inode, void *addr, unsigned int block,
  46		      struct ubifs_data_node *dn)
  47{
  48	struct ubifs_info *c = inode->i_sb->s_fs_info;
  49	int err, len, out_len;
  50	union ubifs_key key;
  51	unsigned int dlen;
  52
  53	data_key_init(c, &key, inode->i_ino, block);
  54	err = ubifs_tnc_lookup(c, &key, dn);
  55	if (err) {
  56		if (err == -ENOENT)
  57			/* Not found, so it must be a hole */
  58			memset(addr, 0, UBIFS_BLOCK_SIZE);
  59		return err;
  60	}
  61
  62	ubifs_assert(c, le64_to_cpu(dn->ch.sqnum) >
  63		     ubifs_inode(inode)->creat_sqnum);
  64	len = le32_to_cpu(dn->size);
  65	if (len <= 0 || len > UBIFS_BLOCK_SIZE)
  66		goto dump;
  67
  68	dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
  69
  70	if (ubifs_crypt_is_encrypted(inode)) {
  71		err = ubifs_decrypt(inode, dn, &dlen, block);
  72		if (err)
  73			goto dump;
  74	}
  75
  76	out_len = UBIFS_BLOCK_SIZE;
  77	err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
  78			       le16_to_cpu(dn->compr_type));
  79	if (err || len != out_len)
  80		goto dump;
  81
  82	/*
  83	 * Data length can be less than a full block, even for blocks that are
  84	 * not the last in the file (e.g., as a result of making a hole and
  85	 * appending data). Ensure that the remainder is zeroed out.
  86	 */
  87	if (len < UBIFS_BLOCK_SIZE)
  88		memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
  89
  90	return 0;
  91
  92dump:
  93	ubifs_err(c, "bad data node (block %u, inode %lu)",
  94		  block, inode->i_ino);
  95	ubifs_dump_node(c, dn);
  96	return -EINVAL;
  97}
  98
  99static int do_readpage(struct page *page)
 100{
 101	void *addr;
 102	int err = 0, i;
 103	unsigned int block, beyond;
 104	struct ubifs_data_node *dn;
 105	struct inode *inode = page->mapping->host;
 106	struct ubifs_info *c = inode->i_sb->s_fs_info;
 107	loff_t i_size = i_size_read(inode);
 108
 109	dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
 110		inode->i_ino, page->index, i_size, page->flags);
 111	ubifs_assert(c, !PageChecked(page));
 112	ubifs_assert(c, !PagePrivate(page));
 113
 114	addr = kmap(page);
 115
 116	block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
 117	beyond = (i_size + UBIFS_BLOCK_SIZE - 1) >> UBIFS_BLOCK_SHIFT;
 118	if (block >= beyond) {
 119		/* Reading beyond inode */
 120		SetPageChecked(page);
 121		memset(addr, 0, PAGE_SIZE);
 122		goto out;
 123	}
 124
 125	dn = kmalloc(UBIFS_MAX_DATA_NODE_SZ, GFP_NOFS);
 126	if (!dn) {
 127		err = -ENOMEM;
 128		goto error;
 129	}
 130
 131	i = 0;
 132	while (1) {
 133		int ret;
 134
 135		if (block >= beyond) {
 136			/* Reading beyond inode */
 137			err = -ENOENT;
 138			memset(addr, 0, UBIFS_BLOCK_SIZE);
 139		} else {
 140			ret = read_block(inode, addr, block, dn);
 141			if (ret) {
 142				err = ret;
 143				if (err != -ENOENT)
 144					break;
 145			} else if (block + 1 == beyond) {
 146				int dlen = le32_to_cpu(dn->size);
 147				int ilen = i_size & (UBIFS_BLOCK_SIZE - 1);
 148
 149				if (ilen && ilen < dlen)
 150					memset(addr + ilen, 0, dlen - ilen);
 151			}
 152		}
 153		if (++i >= UBIFS_BLOCKS_PER_PAGE)
 154			break;
 155		block += 1;
 156		addr += UBIFS_BLOCK_SIZE;
 
 
 
 
 157	}
 
 158	if (err) {
 159		struct ubifs_info *c = inode->i_sb->s_fs_info;
 160		if (err == -ENOENT) {
 161			/* Not found, so it must be a hole */
 162			SetPageChecked(page);
 163			dbg_gen("hole");
 164			goto out_free;
 
 
 
 165		}
 166		ubifs_err(c, "cannot read page %lu of inode %lu, error %d",
 167			  page->index, inode->i_ino, err);
 168		goto error;
 169	}
 170
 171out_free:
 172	kfree(dn);
 173out:
 174	SetPageUptodate(page);
 175	ClearPageError(page);
 176	flush_dcache_page(page);
 177	kunmap(page);
 178	return 0;
 179
 180error:
 181	kfree(dn);
 182	ClearPageUptodate(page);
 183	SetPageError(page);
 184	flush_dcache_page(page);
 185	kunmap(page);
 186	return err;
 187}
 188
 189/**
 190 * release_new_page_budget - release budget of a new page.
 191 * @c: UBIFS file-system description object
 192 *
 193 * This is a helper function which releases budget corresponding to the budget
 194 * of one new page of data.
 195 */
 196static void release_new_page_budget(struct ubifs_info *c)
 197{
 198	struct ubifs_budget_req req = { .recalculate = 1, .new_page = 1 };
 199
 200	ubifs_release_budget(c, &req);
 201}
 202
 203/**
 204 * release_existing_page_budget - release budget of an existing page.
 205 * @c: UBIFS file-system description object
 206 *
 207 * This is a helper function which releases budget corresponding to the budget
 208 * of changing one one page of data which already exists on the flash media.
 209 */
 210static void release_existing_page_budget(struct ubifs_info *c)
 211{
 212	struct ubifs_budget_req req = { .dd_growth = c->bi.page_budget};
 213
 214	ubifs_release_budget(c, &req);
 215}
 216
 217static int write_begin_slow(struct address_space *mapping,
 218			    loff_t pos, unsigned len, struct page **pagep,
 219			    unsigned flags)
 220{
 221	struct inode *inode = mapping->host;
 222	struct ubifs_info *c = inode->i_sb->s_fs_info;
 223	pgoff_t index = pos >> PAGE_SHIFT;
 224	struct ubifs_budget_req req = { .new_page = 1 };
 225	int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
 226	struct page *page;
 227
 228	dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
 229		inode->i_ino, pos, len, inode->i_size);
 230
 231	/*
 232	 * At the slow path we have to budget before locking the page, because
 233	 * budgeting may force write-back, which would wait on locked pages and
 234	 * deadlock if we had the page locked. At this point we do not know
 235	 * anything about the page, so assume that this is a new page which is
 236	 * written to a hole. This corresponds to largest budget. Later the
 237	 * budget will be amended if this is not true.
 238	 */
 239	if (appending)
 240		/* We are appending data, budget for inode change */
 241		req.dirtied_ino = 1;
 242
 243	err = ubifs_budget_space(c, &req);
 244	if (unlikely(err))
 245		return err;
 246
 247	page = grab_cache_page_write_begin(mapping, index, flags);
 248	if (unlikely(!page)) {
 
 249		ubifs_release_budget(c, &req);
 250		return -ENOMEM;
 251	}
 252
 253	if (!PageUptodate(page)) {
 254		if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE)
 255			SetPageChecked(page);
 256		else {
 257			err = do_readpage(page);
 258			if (err) {
 259				unlock_page(page);
 260				put_page(page);
 261				ubifs_release_budget(c, &req);
 262				return err;
 263			}
 264		}
 265
 266		SetPageUptodate(page);
 267		ClearPageError(page);
 268	}
 269
 270	if (PagePrivate(page))
 271		/*
 272		 * The page is dirty, which means it was budgeted twice:
 273		 *   o first time the budget was allocated by the task which
 274		 *     made the page dirty and set the PG_private flag;
 275		 *   o and then we budgeted for it for the second time at the
 276		 *     very beginning of this function.
 277		 *
 278		 * So what we have to do is to release the page budget we
 279		 * allocated.
 280		 */
 281		release_new_page_budget(c);
 282	else if (!PageChecked(page))
 283		/*
 284		 * We are changing a page which already exists on the media.
 285		 * This means that changing the page does not make the amount
 286		 * of indexing information larger, and this part of the budget
 287		 * which we have already acquired may be released.
 288		 */
 289		ubifs_convert_page_budget(c);
 290
 291	if (appending) {
 292		struct ubifs_inode *ui = ubifs_inode(inode);
 293
 294		/*
 295		 * 'ubifs_write_end()' is optimized from the fast-path part of
 296		 * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
 297		 * if data is appended.
 298		 */
 299		mutex_lock(&ui->ui_mutex);
 300		if (ui->dirty)
 301			/*
 302			 * The inode is dirty already, so we may free the
 303			 * budget we allocated.
 304			 */
 305			ubifs_release_dirty_inode_budget(c, ui);
 306	}
 307
 308	*pagep = page;
 309	return 0;
 310}
 311
 312/**
 313 * allocate_budget - allocate budget for 'ubifs_write_begin()'.
 314 * @c: UBIFS file-system description object
 315 * @page: page to allocate budget for
 316 * @ui: UBIFS inode object the page belongs to
 317 * @appending: non-zero if the page is appended
 318 *
 319 * This is a helper function for 'ubifs_write_begin()' which allocates budget
 320 * for the operation. The budget is allocated differently depending on whether
 321 * this is appending, whether the page is dirty or not, and so on. This
 322 * function leaves the @ui->ui_mutex locked in case of appending. Returns zero
 323 * in case of success and %-ENOSPC in case of failure.
 
 324 */
 325static int allocate_budget(struct ubifs_info *c, struct page *page,
 326			   struct ubifs_inode *ui, int appending)
 327{
 328	struct ubifs_budget_req req = { .fast = 1 };
 329
 330	if (PagePrivate(page)) {
 331		if (!appending)
 332			/*
 333			 * The page is dirty and we are not appending, which
 334			 * means no budget is needed at all.
 335			 */
 336			return 0;
 337
 338		mutex_lock(&ui->ui_mutex);
 339		if (ui->dirty)
 340			/*
 341			 * The page is dirty and we are appending, so the inode
 342			 * has to be marked as dirty. However, it is already
 343			 * dirty, so we do not need any budget. We may return,
 344			 * but @ui->ui_mutex hast to be left locked because we
 345			 * should prevent write-back from flushing the inode
 346			 * and freeing the budget. The lock will be released in
 347			 * 'ubifs_write_end()'.
 348			 */
 349			return 0;
 350
 351		/*
 352		 * The page is dirty, we are appending, the inode is clean, so
 353		 * we need to budget the inode change.
 354		 */
 355		req.dirtied_ino = 1;
 356	} else {
 357		if (PageChecked(page))
 358			/*
 359			 * The page corresponds to a hole and does not
 360			 * exist on the media. So changing it makes
 361			 * make the amount of indexing information
 362			 * larger, and we have to budget for a new
 363			 * page.
 364			 */
 365			req.new_page = 1;
 366		else
 367			/*
 368			 * Not a hole, the change will not add any new
 369			 * indexing information, budget for page
 370			 * change.
 371			 */
 372			req.dirtied_page = 1;
 373
 374		if (appending) {
 375			mutex_lock(&ui->ui_mutex);
 376			if (!ui->dirty)
 377				/*
 378				 * The inode is clean but we will have to mark
 379				 * it as dirty because we are appending. This
 380				 * needs a budget.
 381				 */
 382				req.dirtied_ino = 1;
 383		}
 384	}
 385
 386	return ubifs_budget_space(c, &req);
 387}
 388
 389/*
 390 * This function is called when a page of data is going to be written. Since
 391 * the page of data will not necessarily go to the flash straight away, UBIFS
 392 * has to reserve space on the media for it, which is done by means of
 393 * budgeting.
 394 *
 395 * This is the hot-path of the file-system and we are trying to optimize it as
 396 * much as possible. For this reasons it is split on 2 parts - slow and fast.
 397 *
 398 * There many budgeting cases:
 399 *     o a new page is appended - we have to budget for a new page and for
 400 *       changing the inode; however, if the inode is already dirty, there is
 401 *       no need to budget for it;
 402 *     o an existing clean page is changed - we have budget for it; if the page
 403 *       does not exist on the media (a hole), we have to budget for a new
 404 *       page; otherwise, we may budget for changing an existing page; the
 405 *       difference between these cases is that changing an existing page does
 406 *       not introduce anything new to the FS indexing information, so it does
 407 *       not grow, and smaller budget is acquired in this case;
 408 *     o an existing dirty page is changed - no need to budget at all, because
 409 *       the page budget has been acquired by earlier, when the page has been
 410 *       marked dirty.
 411 *
 412 * UBIFS budgeting sub-system may force write-back if it thinks there is no
 413 * space to reserve. This imposes some locking restrictions and makes it
 414 * impossible to take into account the above cases, and makes it impossible to
 415 * optimize budgeting.
 416 *
 417 * The solution for this is that the fast path of 'ubifs_write_begin()' assumes
 418 * there is a plenty of flash space and the budget will be acquired quickly,
 419 * without forcing write-back. The slow path does not make this assumption.
 420 */
 421static int ubifs_write_begin(struct file *file, struct address_space *mapping,
 422			     loff_t pos, unsigned len, unsigned flags,
 423			     struct page **pagep, void **fsdata)
 424{
 425	struct inode *inode = mapping->host;
 426	struct ubifs_info *c = inode->i_sb->s_fs_info;
 427	struct ubifs_inode *ui = ubifs_inode(inode);
 428	pgoff_t index = pos >> PAGE_SHIFT;
 429	int uninitialized_var(err), appending = !!(pos + len > inode->i_size);
 430	int skipped_read = 0;
 431	struct page *page;
 432
 433	ubifs_assert(c, ubifs_inode(inode)->ui_size == inode->i_size);
 434	ubifs_assert(c, !c->ro_media && !c->ro_mount);
 435
 436	if (unlikely(c->ro_error))
 437		return -EROFS;
 438
 439	/* Try out the fast-path part first */
 440	page = grab_cache_page_write_begin(mapping, index, flags);
 441	if (unlikely(!page))
 442		return -ENOMEM;
 
 443
 444	if (!PageUptodate(page)) {
 445		/* The page is not loaded from the flash */
 446		if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE) {
 447			/*
 448			 * We change whole page so no need to load it. But we
 449			 * do not know whether this page exists on the media or
 450			 * not, so we assume the latter because it requires
 451			 * larger budget. The assumption is that it is better
 452			 * to budget a bit more than to read the page from the
 453			 * media. Thus, we are setting the @PG_checked flag
 454			 * here.
 455			 */
 456			SetPageChecked(page);
 457			skipped_read = 1;
 458		} else {
 459			err = do_readpage(page);
 460			if (err) {
 461				unlock_page(page);
 462				put_page(page);
 463				return err;
 464			}
 465		}
 466
 467		SetPageUptodate(page);
 468		ClearPageError(page);
 469	}
 470
 471	err = allocate_budget(c, page, ui, appending);
 472	if (unlikely(err)) {
 473		ubifs_assert(c, err == -ENOSPC);
 474		/*
 475		 * If we skipped reading the page because we were going to
 476		 * write all of it, then it is not up to date.
 477		 */
 478		if (skipped_read) {
 479			ClearPageChecked(page);
 480			ClearPageUptodate(page);
 481		}
 482		/*
 483		 * Budgeting failed which means it would have to force
 484		 * write-back but didn't, because we set the @fast flag in the
 485		 * request. Write-back cannot be done now, while we have the
 486		 * page locked, because it would deadlock. Unlock and free
 487		 * everything and fall-back to slow-path.
 488		 */
 489		if (appending) {
 490			ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
 491			mutex_unlock(&ui->ui_mutex);
 492		}
 493		unlock_page(page);
 494		put_page(page);
 495
 496		return write_begin_slow(mapping, pos, len, pagep, flags);
 497	}
 498
 499	/*
 500	 * Whee, we acquired budgeting quickly - without involving
 501	 * garbage-collection, committing or forcing write-back. We return
 502	 * with @ui->ui_mutex locked if we are appending pages, and unlocked
 503	 * otherwise. This is an optimization (slightly hacky though).
 504	 */
 505	*pagep = page;
 506	return 0;
 507
 508}
 509
 510/**
 511 * cancel_budget - cancel budget.
 512 * @c: UBIFS file-system description object
 513 * @page: page to cancel budget for
 514 * @ui: UBIFS inode object the page belongs to
 515 * @appending: non-zero if the page is appended
 516 *
 517 * This is a helper function for a page write operation. It unlocks the
 518 * @ui->ui_mutex in case of appending.
 519 */
 520static void cancel_budget(struct ubifs_info *c, struct page *page,
 521			  struct ubifs_inode *ui, int appending)
 522{
 523	if (appending) {
 524		if (!ui->dirty)
 525			ubifs_release_dirty_inode_budget(c, ui);
 526		mutex_unlock(&ui->ui_mutex);
 527	}
 528	if (!PagePrivate(page)) {
 529		if (PageChecked(page))
 530			release_new_page_budget(c);
 531		else
 532			release_existing_page_budget(c);
 533	}
 534}
 535
 536static int ubifs_write_end(struct file *file, struct address_space *mapping,
 537			   loff_t pos, unsigned len, unsigned copied,
 538			   struct page *page, void *fsdata)
 539{
 
 540	struct inode *inode = mapping->host;
 541	struct ubifs_inode *ui = ubifs_inode(inode);
 542	struct ubifs_info *c = inode->i_sb->s_fs_info;
 543	loff_t end_pos = pos + len;
 544	int appending = !!(end_pos > inode->i_size);
 545
 546	dbg_gen("ino %lu, pos %llu, pg %lu, len %u, copied %d, i_size %lld",
 547		inode->i_ino, pos, page->index, len, copied, inode->i_size);
 548
 549	if (unlikely(copied < len && len == PAGE_SIZE)) {
 550		/*
 551		 * VFS copied less data to the page that it intended and
 552		 * declared in its '->write_begin()' call via the @len
 553		 * argument. If the page was not up-to-date, and @len was
 554		 * @PAGE_SIZE, the 'ubifs_write_begin()' function did
 555		 * not load it from the media (for optimization reasons). This
 556		 * means that part of the page contains garbage. So read the
 557		 * page now.
 558		 */
 559		dbg_gen("copied %d instead of %d, read page and repeat",
 560			copied, len);
 561		cancel_budget(c, page, ui, appending);
 562		ClearPageChecked(page);
 563
 564		/*
 565		 * Return 0 to force VFS to repeat the whole operation, or the
 566		 * error code if 'do_readpage()' fails.
 567		 */
 568		copied = do_readpage(page);
 569		goto out;
 570	}
 571
 572	if (!PagePrivate(page)) {
 573		SetPagePrivate(page);
 
 
 
 574		atomic_long_inc(&c->dirty_pg_cnt);
 575		__set_page_dirty_nobuffers(page);
 576	}
 577
 578	if (appending) {
 579		i_size_write(inode, end_pos);
 580		ui->ui_size = end_pos;
 581		/*
 582		 * Note, we do not set @I_DIRTY_PAGES (which means that the
 583		 * inode has dirty pages), this has been done in
 584		 * '__set_page_dirty_nobuffers()'.
 585		 */
 586		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
 587		ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
 588		mutex_unlock(&ui->ui_mutex);
 589	}
 590
 591out:
 592	unlock_page(page);
 593	put_page(page);
 594	return copied;
 595}
 596
 597/**
 598 * populate_page - copy data nodes into a page for bulk-read.
 599 * @c: UBIFS file-system description object
 600 * @page: page
 601 * @bu: bulk-read information
 602 * @n: next zbranch slot
 603 *
 604 * This function returns %0 on success and a negative error code on failure.
 605 */
 606static int populate_page(struct ubifs_info *c, struct page *page,
 607			 struct bu_info *bu, int *n)
 608{
 609	int i = 0, nn = *n, offs = bu->zbranch[0].offs, hole = 0, read = 0;
 610	struct inode *inode = page->mapping->host;
 611	loff_t i_size = i_size_read(inode);
 612	unsigned int page_block;
 613	void *addr, *zaddr;
 614	pgoff_t end_index;
 615
 616	dbg_gen("ino %lu, pg %lu, i_size %lld, flags %#lx",
 617		inode->i_ino, page->index, i_size, page->flags);
 618
 619	addr = zaddr = kmap(page);
 620
 621	end_index = (i_size - 1) >> PAGE_SHIFT;
 622	if (!i_size || page->index > end_index) {
 623		hole = 1;
 624		memset(addr, 0, PAGE_SIZE);
 625		goto out_hole;
 626	}
 627
 628	page_block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
 629	while (1) {
 630		int err, len, out_len, dlen;
 631
 632		if (nn >= bu->cnt) {
 633			hole = 1;
 634			memset(addr, 0, UBIFS_BLOCK_SIZE);
 635		} else if (key_block(c, &bu->zbranch[nn].key) == page_block) {
 636			struct ubifs_data_node *dn;
 637
 638			dn = bu->buf + (bu->zbranch[nn].offs - offs);
 639
 640			ubifs_assert(c, le64_to_cpu(dn->ch.sqnum) >
 641				     ubifs_inode(inode)->creat_sqnum);
 642
 643			len = le32_to_cpu(dn->size);
 644			if (len <= 0 || len > UBIFS_BLOCK_SIZE)
 645				goto out_err;
 646
 647			dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
 648			out_len = UBIFS_BLOCK_SIZE;
 649
 650			if (ubifs_crypt_is_encrypted(inode)) {
 651				err = ubifs_decrypt(inode, dn, &dlen, page_block);
 652				if (err)
 653					goto out_err;
 654			}
 655
 656			err = ubifs_decompress(c, &dn->data, dlen, addr, &out_len,
 657					       le16_to_cpu(dn->compr_type));
 658			if (err || len != out_len)
 659				goto out_err;
 660
 661			if (len < UBIFS_BLOCK_SIZE)
 662				memset(addr + len, 0, UBIFS_BLOCK_SIZE - len);
 663
 664			nn += 1;
 665			read = (i << UBIFS_BLOCK_SHIFT) + len;
 666		} else if (key_block(c, &bu->zbranch[nn].key) < page_block) {
 667			nn += 1;
 668			continue;
 669		} else {
 670			hole = 1;
 671			memset(addr, 0, UBIFS_BLOCK_SIZE);
 672		}
 673		if (++i >= UBIFS_BLOCKS_PER_PAGE)
 674			break;
 675		addr += UBIFS_BLOCK_SIZE;
 676		page_block += 1;
 
 
 
 
 677	}
 678
 679	if (end_index == page->index) {
 680		int len = i_size & (PAGE_SIZE - 1);
 681
 682		if (len && len < read)
 683			memset(zaddr + len, 0, read - len);
 684	}
 685
 686out_hole:
 687	if (hole) {
 688		SetPageChecked(page);
 689		dbg_gen("hole");
 690	}
 691
 692	SetPageUptodate(page);
 693	ClearPageError(page);
 694	flush_dcache_page(page);
 695	kunmap(page);
 696	*n = nn;
 697	return 0;
 698
 699out_err:
 700	ClearPageUptodate(page);
 701	SetPageError(page);
 702	flush_dcache_page(page);
 703	kunmap(page);
 704	ubifs_err(c, "bad data node (block %u, inode %lu)",
 705		  page_block, inode->i_ino);
 706	return -EINVAL;
 707}
 708
 709/**
 710 * ubifs_do_bulk_read - do bulk-read.
 711 * @c: UBIFS file-system description object
 712 * @bu: bulk-read information
 713 * @page1: first page to read
 714 *
 715 * This function returns %1 if the bulk-read is done, otherwise %0 is returned.
 716 */
 717static int ubifs_do_bulk_read(struct ubifs_info *c, struct bu_info *bu,
 718			      struct page *page1)
 719{
 720	pgoff_t offset = page1->index, end_index;
 721	struct address_space *mapping = page1->mapping;
 722	struct inode *inode = mapping->host;
 723	struct ubifs_inode *ui = ubifs_inode(inode);
 724	int err, page_idx, page_cnt, ret = 0, n = 0;
 725	int allocate = bu->buf ? 0 : 1;
 726	loff_t isize;
 727	gfp_t ra_gfp_mask = readahead_gfp_mask(mapping) & ~__GFP_FS;
 728
 729	err = ubifs_tnc_get_bu_keys(c, bu);
 730	if (err)
 731		goto out_warn;
 732
 733	if (bu->eof) {
 734		/* Turn off bulk-read at the end of the file */
 735		ui->read_in_a_row = 1;
 736		ui->bulk_read = 0;
 737	}
 738
 739	page_cnt = bu->blk_cnt >> UBIFS_BLOCKS_PER_PAGE_SHIFT;
 740	if (!page_cnt) {
 741		/*
 742		 * This happens when there are multiple blocks per page and the
 743		 * blocks for the first page we are looking for, are not
 744		 * together. If all the pages were like this, bulk-read would
 745		 * reduce performance, so we turn it off for a while.
 746		 */
 747		goto out_bu_off;
 748	}
 749
 750	if (bu->cnt) {
 751		if (allocate) {
 752			/*
 753			 * Allocate bulk-read buffer depending on how many data
 754			 * nodes we are going to read.
 755			 */
 756			bu->buf_len = bu->zbranch[bu->cnt - 1].offs +
 757				      bu->zbranch[bu->cnt - 1].len -
 758				      bu->zbranch[0].offs;
 759			ubifs_assert(c, bu->buf_len > 0);
 760			ubifs_assert(c, bu->buf_len <= c->leb_size);
 761			bu->buf = kmalloc(bu->buf_len, GFP_NOFS | __GFP_NOWARN);
 762			if (!bu->buf)
 763				goto out_bu_off;
 764		}
 765
 766		err = ubifs_tnc_bulk_read(c, bu);
 767		if (err)
 768			goto out_warn;
 769	}
 770
 771	err = populate_page(c, page1, bu, &n);
 772	if (err)
 773		goto out_warn;
 774
 775	unlock_page(page1);
 776	ret = 1;
 777
 778	isize = i_size_read(inode);
 779	if (isize == 0)
 780		goto out_free;
 781	end_index = ((isize - 1) >> PAGE_SHIFT);
 782
 783	for (page_idx = 1; page_idx < page_cnt; page_idx++) {
 784		pgoff_t page_offset = offset + page_idx;
 785		struct page *page;
 786
 787		if (page_offset > end_index)
 788			break;
 789		page = find_or_create_page(mapping, page_offset, ra_gfp_mask);
 790		if (!page)
 
 
 791			break;
 792		if (!PageUptodate(page))
 793			err = populate_page(c, page, bu, &n);
 794		unlock_page(page);
 795		put_page(page);
 796		if (err)
 797			break;
 798	}
 799
 800	ui->last_page_read = offset + page_idx - 1;
 801
 802out_free:
 803	if (allocate)
 804		kfree(bu->buf);
 805	return ret;
 806
 807out_warn:
 808	ubifs_warn(c, "ignoring error %d and skipping bulk-read", err);
 809	goto out_free;
 810
 811out_bu_off:
 812	ui->read_in_a_row = ui->bulk_read = 0;
 813	goto out_free;
 814}
 815
 816/**
 817 * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
 818 * @page: page from which to start bulk-read.
 819 *
 820 * Some flash media are capable of reading sequentially at faster rates. UBIFS
 821 * bulk-read facility is designed to take advantage of that, by reading in one
 822 * go consecutive data nodes that are also located consecutively in the same
 823 * LEB. This function returns %1 if a bulk-read is done and %0 otherwise.
 
 
 824 */
 825static int ubifs_bulk_read(struct page *page)
 826{
 827	struct inode *inode = page->mapping->host;
 828	struct ubifs_info *c = inode->i_sb->s_fs_info;
 829	struct ubifs_inode *ui = ubifs_inode(inode);
 830	pgoff_t index = page->index, last_page_read = ui->last_page_read;
 831	struct bu_info *bu;
 832	int err = 0, allocated = 0;
 833
 834	ui->last_page_read = index;
 835	if (!c->bulk_read)
 836		return 0;
 837
 838	/*
 839	 * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
 840	 * so don't bother if we cannot lock the mutex.
 841	 */
 842	if (!mutex_trylock(&ui->ui_mutex))
 843		return 0;
 844
 845	if (index != last_page_read + 1) {
 846		/* Turn off bulk-read if we stop reading sequentially */
 847		ui->read_in_a_row = 1;
 848		if (ui->bulk_read)
 849			ui->bulk_read = 0;
 850		goto out_unlock;
 851	}
 852
 853	if (!ui->bulk_read) {
 854		ui->read_in_a_row += 1;
 855		if (ui->read_in_a_row < 3)
 856			goto out_unlock;
 857		/* Three reads in a row, so switch on bulk-read */
 858		ui->bulk_read = 1;
 859	}
 860
 861	/*
 862	 * If possible, try to use pre-allocated bulk-read information, which
 863	 * is protected by @c->bu_mutex.
 864	 */
 865	if (mutex_trylock(&c->bu_mutex))
 866		bu = &c->bu;
 867	else {
 868		bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
 869		if (!bu)
 870			goto out_unlock;
 871
 872		bu->buf = NULL;
 873		allocated = 1;
 874	}
 875
 876	bu->buf_len = c->max_bu_buf_len;
 877	data_key_init(c, &bu->key, inode->i_ino,
 878		      page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
 879	err = ubifs_do_bulk_read(c, bu, page);
 880
 881	if (!allocated)
 882		mutex_unlock(&c->bu_mutex);
 883	else
 884		kfree(bu);
 885
 886out_unlock:
 887	mutex_unlock(&ui->ui_mutex);
 888	return err;
 889}
 890
 891static int ubifs_readpage(struct file *file, struct page *page)
 892{
 893	if (ubifs_bulk_read(page))
 894		return 0;
 895	do_readpage(page);
 896	unlock_page(page);
 897	return 0;
 898}
 899
 900static int do_writepage(struct page *page, int len)
 901{
 902	int err = 0, i, blen;
 903	unsigned int block;
 904	void *addr;
 
 905	union ubifs_key key;
 906	struct inode *inode = page->mapping->host;
 907	struct ubifs_info *c = inode->i_sb->s_fs_info;
 908
 909#ifdef UBIFS_DEBUG
 910	struct ubifs_inode *ui = ubifs_inode(inode);
 911	spin_lock(&ui->ui_lock);
 912	ubifs_assert(c, page->index <= ui->synced_i_size >> PAGE_SHIFT);
 913	spin_unlock(&ui->ui_lock);
 914#endif
 915
 916	/* Update radix tree tags */
 917	set_page_writeback(page);
 918
 919	addr = kmap(page);
 920	block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
 921	i = 0;
 922	while (len) {
 923		blen = min_t(int, len, UBIFS_BLOCK_SIZE);
 924		data_key_init(c, &key, inode->i_ino, block);
 925		err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
 926		if (err)
 927			break;
 928		if (++i >= UBIFS_BLOCKS_PER_PAGE)
 
 929			break;
 930		block += 1;
 931		addr += blen;
 932		len -= blen;
 
 
 
 
 933	}
 
 934	if (err) {
 935		SetPageError(page);
 936		ubifs_err(c, "cannot write page %lu of inode %lu, error %d",
 937			  page->index, inode->i_ino, err);
 938		ubifs_ro_mode(c, err);
 939	}
 940
 941	ubifs_assert(c, PagePrivate(page));
 942	if (PageChecked(page))
 943		release_new_page_budget(c);
 944	else
 945		release_existing_page_budget(c);
 946
 947	atomic_long_dec(&c->dirty_pg_cnt);
 948	ClearPagePrivate(page);
 949	ClearPageChecked(page);
 950
 951	kunmap(page);
 952	unlock_page(page);
 953	end_page_writeback(page);
 954	return err;
 955}
 956
 957/*
 958 * When writing-back dirty inodes, VFS first writes-back pages belonging to the
 959 * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
 960 * situation when a we have an inode with size 0, then a megabyte of data is
 961 * appended to the inode, then write-back starts and flushes some amount of the
 962 * dirty pages, the journal becomes full, commit happens and finishes, and then
 963 * an unclean reboot happens. When the file system is mounted next time, the
 964 * inode size would still be 0, but there would be many pages which are beyond
 965 * the inode size, they would be indexed and consume flash space. Because the
 966 * journal has been committed, the replay would not be able to detect this
 967 * situation and correct the inode size. This means UBIFS would have to scan
 968 * whole index and correct all inode sizes, which is long an unacceptable.
 969 *
 970 * To prevent situations like this, UBIFS writes pages back only if they are
 971 * within the last synchronized inode size, i.e. the size which has been
 972 * written to the flash media last time. Otherwise, UBIFS forces inode
 973 * write-back, thus making sure the on-flash inode contains current inode size,
 974 * and then keeps writing pages back.
 975 *
 976 * Some locking issues explanation. 'ubifs_writepage()' first is called with
 977 * the page locked, and it locks @ui_mutex. However, write-back does take inode
 978 * @i_mutex, which means other VFS operations may be run on this inode at the
 979 * same time. And the problematic one is truncation to smaller size, from where
 980 * we have to call 'truncate_setsize()', which first changes @inode->i_size,
 981 * then drops the truncated pages. And while dropping the pages, it takes the
 982 * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
 983 * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
 984 * This means that @inode->i_size is changed while @ui_mutex is unlocked.
 985 *
 986 * XXX(truncate): with the new truncate sequence this is not true anymore,
 987 * and the calls to truncate_setsize can be move around freely.  They should
 988 * be moved to the very end of the truncate sequence.
 989 *
 990 * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
 991 * inode size. How do we do this if @inode->i_size may became smaller while we
 992 * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
 993 * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
 994 * internally and updates it under @ui_mutex.
 995 *
 996 * Q: why we do not worry that if we race with truncation, we may end up with a
 997 * situation when the inode is truncated while we are in the middle of
 998 * 'do_writepage()', so we do write beyond inode size?
 999 * A: If we are in the middle of 'do_writepage()', truncation would be locked
1000 * on the page lock and it would not write the truncated inode node to the
1001 * journal before we have finished.
1002 */
1003static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
 
1004{
1005	struct inode *inode = page->mapping->host;
1006	struct ubifs_info *c = inode->i_sb->s_fs_info;
1007	struct ubifs_inode *ui = ubifs_inode(inode);
1008	loff_t i_size =  i_size_read(inode), synced_i_size;
1009	pgoff_t end_index = i_size >> PAGE_SHIFT;
1010	int err, len = i_size & (PAGE_SIZE - 1);
1011	void *kaddr;
1012
1013	dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1014		inode->i_ino, page->index, page->flags);
1015	ubifs_assert(c, PagePrivate(page));
1016
1017	/* Is the page fully outside @i_size? (truncate in progress) */
1018	if (page->index > end_index || (page->index == end_index && !len)) {
1019		err = 0;
1020		goto out_unlock;
1021	}
1022
1023	spin_lock(&ui->ui_lock);
1024	synced_i_size = ui->synced_i_size;
1025	spin_unlock(&ui->ui_lock);
1026
1027	/* Is the page fully inside @i_size? */
1028	if (page->index < end_index) {
1029		if (page->index >= synced_i_size >> PAGE_SHIFT) {
1030			err = inode->i_sb->s_op->write_inode(inode, NULL);
1031			if (err)
1032				goto out_unlock;
1033			/*
1034			 * The inode has been written, but the write-buffer has
1035			 * not been synchronized, so in case of an unclean
1036			 * reboot we may end up with some pages beyond inode
1037			 * size, but they would be in the journal (because
1038			 * commit flushes write buffers) and recovery would deal
1039			 * with this.
1040			 */
1041		}
1042		return do_writepage(page, PAGE_SIZE);
1043	}
1044
1045	/*
1046	 * The page straddles @i_size. It must be zeroed out on each and every
1047	 * writepage invocation because it may be mmapped. "A file is mapped
1048	 * in multiples of the page size. For a file that is not a multiple of
1049	 * the page size, the remaining memory is zeroed when mapped, and
1050	 * writes to that region are not written out to the file."
1051	 */
1052	kaddr = kmap_atomic(page);
1053	memset(kaddr + len, 0, PAGE_SIZE - len);
1054	flush_dcache_page(page);
1055	kunmap_atomic(kaddr);
1056
1057	if (i_size > synced_i_size) {
1058		err = inode->i_sb->s_op->write_inode(inode, NULL);
1059		if (err)
1060			goto out_unlock;
1061	}
1062
1063	return do_writepage(page, len);
1064
 
 
 
 
 
 
1065out_unlock:
1066	unlock_page(page);
1067	return err;
1068}
1069
 
 
 
 
 
 
1070/**
1071 * do_attr_changes - change inode attributes.
1072 * @inode: inode to change attributes for
1073 * @attr: describes attributes to change
1074 */
1075static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1076{
1077	if (attr->ia_valid & ATTR_UID)
1078		inode->i_uid = attr->ia_uid;
1079	if (attr->ia_valid & ATTR_GID)
1080		inode->i_gid = attr->ia_gid;
1081	if (attr->ia_valid & ATTR_ATIME) {
1082		inode->i_atime = timestamp_truncate(attr->ia_atime,
1083						  inode);
1084	}
1085	if (attr->ia_valid & ATTR_MTIME) {
1086		inode->i_mtime = timestamp_truncate(attr->ia_mtime,
1087						  inode);
1088	}
1089	if (attr->ia_valid & ATTR_CTIME) {
1090		inode->i_ctime = timestamp_truncate(attr->ia_ctime,
1091						  inode);
1092	}
1093	if (attr->ia_valid & ATTR_MODE) {
1094		umode_t mode = attr->ia_mode;
1095
1096		if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1097			mode &= ~S_ISGID;
1098		inode->i_mode = mode;
1099	}
1100}
1101
1102/**
1103 * do_truncation - truncate an inode.
1104 * @c: UBIFS file-system description object
1105 * @inode: inode to truncate
1106 * @attr: inode attribute changes description
1107 *
1108 * This function implements VFS '->setattr()' call when the inode is truncated
1109 * to a smaller size. Returns zero in case of success and a negative error code
 
 
1110 * in case of failure.
1111 */
1112static int do_truncation(struct ubifs_info *c, struct inode *inode,
1113			 const struct iattr *attr)
1114{
1115	int err;
1116	struct ubifs_budget_req req;
1117	loff_t old_size = inode->i_size, new_size = attr->ia_size;
1118	int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1119	struct ubifs_inode *ui = ubifs_inode(inode);
1120
1121	dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1122	memset(&req, 0, sizeof(struct ubifs_budget_req));
1123
1124	/*
1125	 * If this is truncation to a smaller size, and we do not truncate on a
1126	 * block boundary, budget for changing one data block, because the last
1127	 * block will be re-written.
1128	 */
1129	if (new_size & (UBIFS_BLOCK_SIZE - 1))
1130		req.dirtied_page = 1;
1131
1132	req.dirtied_ino = 1;
1133	/* A funny way to budget for truncation node */
1134	req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1135	err = ubifs_budget_space(c, &req);
1136	if (err) {
1137		/*
1138		 * Treat truncations to zero as deletion and always allow them,
1139		 * just like we do for '->unlink()'.
1140		 */
1141		if (new_size || err != -ENOSPC)
1142			return err;
1143		budgeted = 0;
1144	}
1145
1146	truncate_setsize(inode, new_size);
1147
1148	if (offset) {
1149		pgoff_t index = new_size >> PAGE_SHIFT;
1150		struct page *page;
1151
1152		page = find_lock_page(inode->i_mapping, index);
1153		if (page) {
1154			if (PageDirty(page)) {
1155				/*
1156				 * 'ubifs_jnl_truncate()' will try to truncate
1157				 * the last data node, but it contains
1158				 * out-of-date data because the page is dirty.
1159				 * Write the page now, so that
1160				 * 'ubifs_jnl_truncate()' will see an already
1161				 * truncated (and up to date) data node.
1162				 */
1163				ubifs_assert(c, PagePrivate(page));
1164
1165				clear_page_dirty_for_io(page);
1166				if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1167					offset = new_size &
1168						 (PAGE_SIZE - 1);
1169				err = do_writepage(page, offset);
1170				put_page(page);
1171				if (err)
1172					goto out_budg;
1173				/*
1174				 * We could now tell 'ubifs_jnl_truncate()' not
1175				 * to read the last block.
1176				 */
1177			} else {
1178				/*
1179				 * We could 'kmap()' the page and pass the data
1180				 * to 'ubifs_jnl_truncate()' to save it from
1181				 * having to read it.
1182				 */
1183				unlock_page(page);
1184				put_page(page);
1185			}
1186		}
1187	}
1188
1189	mutex_lock(&ui->ui_mutex);
1190	ui->ui_size = inode->i_size;
1191	/* Truncation changes inode [mc]time */
1192	inode->i_mtime = inode->i_ctime = current_time(inode);
1193	/* Other attributes may be changed at the same time as well */
1194	do_attr_changes(inode, attr);
1195	err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1196	mutex_unlock(&ui->ui_mutex);
1197
1198out_budg:
1199	if (budgeted)
1200		ubifs_release_budget(c, &req);
1201	else {
1202		c->bi.nospace = c->bi.nospace_rp = 0;
1203		smp_wmb();
1204	}
1205	return err;
1206}
1207
1208/**
1209 * do_setattr - change inode attributes.
1210 * @c: UBIFS file-system description object
1211 * @inode: inode to change attributes for
1212 * @attr: inode attribute changes description
1213 *
1214 * This function implements VFS '->setattr()' call for all cases except
1215 * truncations to smaller size. Returns zero in case of success and a negative
 
 
1216 * error code in case of failure.
1217 */
1218static int do_setattr(struct ubifs_info *c, struct inode *inode,
1219		      const struct iattr *attr)
1220{
1221	int err, release;
1222	loff_t new_size = attr->ia_size;
1223	struct ubifs_inode *ui = ubifs_inode(inode);
1224	struct ubifs_budget_req req = { .dirtied_ino = 1,
1225				.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1226
1227	err = ubifs_budget_space(c, &req);
1228	if (err)
1229		return err;
1230
1231	if (attr->ia_valid & ATTR_SIZE) {
1232		dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1233		truncate_setsize(inode, new_size);
1234	}
1235
1236	mutex_lock(&ui->ui_mutex);
1237	if (attr->ia_valid & ATTR_SIZE) {
1238		/* Truncation changes inode [mc]time */
1239		inode->i_mtime = inode->i_ctime = current_time(inode);
1240		/* 'truncate_setsize()' changed @i_size, update @ui_size */
1241		ui->ui_size = inode->i_size;
1242	}
1243
1244	do_attr_changes(inode, attr);
1245
1246	release = ui->dirty;
1247	if (attr->ia_valid & ATTR_SIZE)
1248		/*
1249		 * Inode length changed, so we have to make sure
1250		 * @I_DIRTY_DATASYNC is set.
1251		 */
1252		 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1253	else
1254		mark_inode_dirty_sync(inode);
1255	mutex_unlock(&ui->ui_mutex);
1256
1257	if (release)
1258		ubifs_release_budget(c, &req);
1259	if (IS_SYNC(inode))
1260		err = inode->i_sb->s_op->write_inode(inode, NULL);
1261	return err;
1262}
1263
1264int ubifs_setattr(struct dentry *dentry, struct iattr *attr)
 
1265{
1266	int err;
1267	struct inode *inode = d_inode(dentry);
1268	struct ubifs_info *c = inode->i_sb->s_fs_info;
1269
1270	dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1271		inode->i_ino, inode->i_mode, attr->ia_valid);
1272	err = setattr_prepare(dentry, attr);
1273	if (err)
1274		return err;
1275
1276	err = dbg_check_synced_i_size(c, inode);
1277	if (err)
1278		return err;
1279
1280	err = fscrypt_prepare_setattr(dentry, attr);
1281	if (err)
1282		return err;
1283
1284	if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1285		/* Truncation to a smaller size */
1286		err = do_truncation(c, inode, attr);
1287	else
1288		err = do_setattr(c, inode, attr);
1289
1290	return err;
1291}
1292
1293static void ubifs_invalidatepage(struct page *page, unsigned int offset,
1294				 unsigned int length)
1295{
1296	struct inode *inode = page->mapping->host;
1297	struct ubifs_info *c = inode->i_sb->s_fs_info;
1298
1299	ubifs_assert(c, PagePrivate(page));
1300	if (offset || length < PAGE_SIZE)
1301		/* Partial page remains dirty */
1302		return;
1303
1304	if (PageChecked(page))
1305		release_new_page_budget(c);
1306	else
1307		release_existing_page_budget(c);
1308
1309	atomic_long_dec(&c->dirty_pg_cnt);
1310	ClearPagePrivate(page);
1311	ClearPageChecked(page);
1312}
1313
1314int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1315{
1316	struct inode *inode = file->f_mapping->host;
1317	struct ubifs_info *c = inode->i_sb->s_fs_info;
1318	int err;
1319
1320	dbg_gen("syncing inode %lu", inode->i_ino);
1321
1322	if (c->ro_mount)
1323		/*
1324		 * For some really strange reasons VFS does not filter out
1325		 * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1326		 */
1327		return 0;
1328
1329	err = file_write_and_wait_range(file, start, end);
1330	if (err)
1331		return err;
1332	inode_lock(inode);
1333
1334	/* Synchronize the inode unless this is a 'datasync()' call. */
1335	if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1336		err = inode->i_sb->s_op->write_inode(inode, NULL);
1337		if (err)
1338			goto out;
1339	}
1340
1341	/*
1342	 * Nodes related to this inode may still sit in a write-buffer. Flush
1343	 * them.
1344	 */
1345	err = ubifs_sync_wbufs_by_inode(c, inode);
1346out:
1347	inode_unlock(inode);
1348	return err;
1349}
1350
1351/**
1352 * mctime_update_needed - check if mtime or ctime update is needed.
1353 * @inode: the inode to do the check for
1354 * @now: current time
1355 *
1356 * This helper function checks if the inode mtime/ctime should be updated or
1357 * not. If current values of the time-stamps are within the UBIFS inode time
1358 * granularity, they are not updated. This is an optimization.
 
 
1359 */
1360static inline int mctime_update_needed(const struct inode *inode,
1361				       const struct timespec64 *now)
1362{
1363	if (!timespec64_equal(&inode->i_mtime, now) ||
1364	    !timespec64_equal(&inode->i_ctime, now))
 
 
1365		return 1;
1366	return 0;
1367}
1368
1369/**
1370 * ubifs_update_time - update time of inode.
1371 * @inode: inode to update
 
 
1372 *
1373 * This function updates time of the inode.
 
 
1374 */
1375int ubifs_update_time(struct inode *inode, struct timespec64 *time,
1376			     int flags)
1377{
1378	struct ubifs_inode *ui = ubifs_inode(inode);
1379	struct ubifs_info *c = inode->i_sb->s_fs_info;
1380	struct ubifs_budget_req req = { .dirtied_ino = 1,
1381			.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1382	int iflags = I_DIRTY_TIME;
1383	int err, release;
1384
1385	if (!IS_ENABLED(CONFIG_UBIFS_ATIME_SUPPORT))
1386		return generic_update_time(inode, time, flags);
 
 
1387
1388	err = ubifs_budget_space(c, &req);
1389	if (err)
1390		return err;
1391
1392	mutex_lock(&ui->ui_mutex);
1393	if (flags & S_ATIME)
1394		inode->i_atime = *time;
1395	if (flags & S_CTIME)
1396		inode->i_ctime = *time;
1397	if (flags & S_MTIME)
1398		inode->i_mtime = *time;
1399
1400	if (!(inode->i_sb->s_flags & SB_LAZYTIME))
1401		iflags |= I_DIRTY_SYNC;
1402
1403	release = ui->dirty;
1404	__mark_inode_dirty(inode, iflags);
1405	mutex_unlock(&ui->ui_mutex);
1406	if (release)
1407		ubifs_release_budget(c, &req);
1408	return 0;
1409}
1410
1411/**
1412 * update_mctime - update mtime and ctime of an inode.
1413 * @inode: inode to update
1414 *
1415 * This function updates mtime and ctime of the inode if it is not equivalent to
1416 * current time. Returns zero in case of success and a negative error code in
 
 
1417 * case of failure.
1418 */
1419static int update_mctime(struct inode *inode)
1420{
1421	struct timespec64 now = current_time(inode);
1422	struct ubifs_inode *ui = ubifs_inode(inode);
1423	struct ubifs_info *c = inode->i_sb->s_fs_info;
1424
1425	if (mctime_update_needed(inode, &now)) {
1426		int err, release;
1427		struct ubifs_budget_req req = { .dirtied_ino = 1,
1428				.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1429
1430		err = ubifs_budget_space(c, &req);
1431		if (err)
1432			return err;
1433
1434		mutex_lock(&ui->ui_mutex);
1435		inode->i_mtime = inode->i_ctime = current_time(inode);
1436		release = ui->dirty;
1437		mark_inode_dirty_sync(inode);
1438		mutex_unlock(&ui->ui_mutex);
1439		if (release)
1440			ubifs_release_budget(c, &req);
1441	}
1442
1443	return 0;
1444}
1445
1446static ssize_t ubifs_write_iter(struct kiocb *iocb, struct iov_iter *from)
1447{
1448	int err = update_mctime(file_inode(iocb->ki_filp));
1449	if (err)
1450		return err;
1451
1452	return generic_file_write_iter(iocb, from);
1453}
1454
1455static int ubifs_set_page_dirty(struct page *page)
 
1456{
1457	int ret;
1458	struct inode *inode = page->mapping->host;
1459	struct ubifs_info *c = inode->i_sb->s_fs_info;
1460
1461	ret = __set_page_dirty_nobuffers(page);
1462	/*
1463	 * An attempt to dirty a page without budgeting for it - should not
1464	 * happen.
1465	 */
1466	ubifs_assert(c, ret == 0);
1467	return ret;
1468}
1469
1470#ifdef CONFIG_MIGRATION
1471static int ubifs_migrate_page(struct address_space *mapping,
1472		struct page *newpage, struct page *page, enum migrate_mode mode)
1473{
1474	int rc;
1475
1476	rc = migrate_page_move_mapping(mapping, newpage, page, 0);
1477	if (rc != MIGRATEPAGE_SUCCESS)
1478		return rc;
1479
1480	if (PagePrivate(page)) {
1481		ClearPagePrivate(page);
1482		SetPagePrivate(newpage);
1483	}
1484
1485	if (mode != MIGRATE_SYNC_NO_COPY)
1486		migrate_page_copy(newpage, page);
1487	else
1488		migrate_page_states(newpage, page);
1489	return MIGRATEPAGE_SUCCESS;
1490}
1491#endif
1492
1493static int ubifs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1494{
1495	struct inode *inode = page->mapping->host;
1496	struct ubifs_info *c = inode->i_sb->s_fs_info;
1497
 
 
 
1498	/*
1499	 * An attempt to release a dirty page without budgeting for it - should
1500	 * not happen.
 
 
 
1501	 */
1502	if (PageWriteback(page))
1503		return 0;
1504	ubifs_assert(c, PagePrivate(page));
1505	ubifs_assert(c, 0);
1506	ClearPagePrivate(page);
1507	ClearPageChecked(page);
1508	return 1;
 
 
 
1509}
1510
1511/*
1512 * mmap()d file has taken write protection fault and is being made writable.
1513 * UBIFS must ensure page is budgeted for.
1514 */
1515static vm_fault_t ubifs_vm_page_mkwrite(struct vm_fault *vmf)
1516{
1517	struct page *page = vmf->page;
1518	struct inode *inode = file_inode(vmf->vma->vm_file);
1519	struct ubifs_info *c = inode->i_sb->s_fs_info;
1520	struct timespec64 now = current_time(inode);
1521	struct ubifs_budget_req req = { .new_page = 1 };
1522	int err, update_time;
1523
1524	dbg_gen("ino %lu, pg %lu, i_size %lld",	inode->i_ino, page->index,
1525		i_size_read(inode));
1526	ubifs_assert(c, !c->ro_media && !c->ro_mount);
1527
1528	if (unlikely(c->ro_error))
1529		return VM_FAULT_SIGBUS; /* -EROFS */
1530
1531	/*
1532	 * We have not locked @page so far so we may budget for changing the
1533	 * page. Note, we cannot do this after we locked the page, because
1534	 * budgeting may cause write-back which would cause deadlock.
1535	 *
1536	 * At the moment we do not know whether the page is dirty or not, so we
1537	 * assume that it is not and budget for a new page. We could look at
1538	 * the @PG_private flag and figure this out, but we may race with write
1539	 * back and the page state may change by the time we lock it, so this
1540	 * would need additional care. We do not bother with this at the
1541	 * moment, although it might be good idea to do. Instead, we allocate
1542	 * budget for a new page and amend it later on if the page was in fact
1543	 * dirty.
1544	 *
1545	 * The budgeting-related logic of this function is similar to what we
1546	 * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1547	 * for more comments.
1548	 */
1549	update_time = mctime_update_needed(inode, &now);
1550	if (update_time)
1551		/*
1552		 * We have to change inode time stamp which requires extra
1553		 * budgeting.
1554		 */
1555		req.dirtied_ino = 1;
1556
1557	err = ubifs_budget_space(c, &req);
1558	if (unlikely(err)) {
1559		if (err == -ENOSPC)
1560			ubifs_warn(c, "out of space for mmapped file (inode number %lu)",
1561				   inode->i_ino);
1562		return VM_FAULT_SIGBUS;
1563	}
1564
1565	lock_page(page);
1566	if (unlikely(page->mapping != inode->i_mapping ||
1567		     page_offset(page) > i_size_read(inode))) {
1568		/* Page got truncated out from underneath us */
1569		goto sigbus;
1570	}
1571
1572	if (PagePrivate(page))
1573		release_new_page_budget(c);
1574	else {
1575		if (!PageChecked(page))
1576			ubifs_convert_page_budget(c);
1577		SetPagePrivate(page);
1578		atomic_long_inc(&c->dirty_pg_cnt);
1579		__set_page_dirty_nobuffers(page);
1580	}
1581
1582	if (update_time) {
1583		int release;
1584		struct ubifs_inode *ui = ubifs_inode(inode);
1585
1586		mutex_lock(&ui->ui_mutex);
1587		inode->i_mtime = inode->i_ctime = current_time(inode);
1588		release = ui->dirty;
1589		mark_inode_dirty_sync(inode);
1590		mutex_unlock(&ui->ui_mutex);
1591		if (release)
1592			ubifs_release_dirty_inode_budget(c, ui);
1593	}
1594
1595	wait_for_stable_page(page);
1596	return VM_FAULT_LOCKED;
1597
1598sigbus:
1599	unlock_page(page);
1600	ubifs_release_budget(c, &req);
1601	return VM_FAULT_SIGBUS;
1602}
1603
1604static const struct vm_operations_struct ubifs_file_vm_ops = {
1605	.fault        = filemap_fault,
1606	.map_pages = filemap_map_pages,
1607	.page_mkwrite = ubifs_vm_page_mkwrite,
1608};
1609
1610static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1611{
1612	int err;
1613
1614	err = generic_file_mmap(file, vma);
1615	if (err)
1616		return err;
1617	vma->vm_ops = &ubifs_file_vm_ops;
1618
1619	if (IS_ENABLED(CONFIG_UBIFS_ATIME_SUPPORT))
1620		file_accessed(file);
1621
1622	return 0;
1623}
1624
1625static const char *ubifs_get_link(struct dentry *dentry,
1626					    struct inode *inode,
1627					    struct delayed_call *done)
1628{
1629	struct ubifs_inode *ui = ubifs_inode(inode);
1630
1631	if (!IS_ENCRYPTED(inode))
1632		return ui->data;
1633
1634	if (!dentry)
1635		return ERR_PTR(-ECHILD);
1636
1637	return fscrypt_get_symlink(inode, ui->data, ui->data_len, done);
1638}
1639
 
 
 
 
 
 
 
 
 
 
 
1640const struct address_space_operations ubifs_file_address_operations = {
1641	.readpage       = ubifs_readpage,
1642	.writepage      = ubifs_writepage,
1643	.write_begin    = ubifs_write_begin,
1644	.write_end      = ubifs_write_end,
1645	.invalidatepage = ubifs_invalidatepage,
1646	.set_page_dirty = ubifs_set_page_dirty,
1647#ifdef CONFIG_MIGRATION
1648	.migratepage	= ubifs_migrate_page,
1649#endif
1650	.releasepage    = ubifs_releasepage,
1651};
1652
1653const struct inode_operations ubifs_file_inode_operations = {
1654	.setattr     = ubifs_setattr,
1655	.getattr     = ubifs_getattr,
1656#ifdef CONFIG_UBIFS_FS_XATTR
1657	.listxattr   = ubifs_listxattr,
1658#endif
1659	.update_time = ubifs_update_time,
 
 
1660};
1661
1662const struct inode_operations ubifs_symlink_inode_operations = {
1663	.get_link    = ubifs_get_link,
1664	.setattr     = ubifs_setattr,
1665	.getattr     = ubifs_getattr,
1666#ifdef CONFIG_UBIFS_FS_XATTR
1667	.listxattr   = ubifs_listxattr,
1668#endif
1669	.update_time = ubifs_update_time,
1670};
1671
1672const struct file_operations ubifs_file_operations = {
1673	.llseek         = generic_file_llseek,
1674	.read_iter      = generic_file_read_iter,
1675	.write_iter     = ubifs_write_iter,
1676	.mmap           = ubifs_file_mmap,
1677	.fsync          = ubifs_fsync,
1678	.unlocked_ioctl = ubifs_ioctl,
1679	.splice_read	= generic_file_splice_read,
1680	.splice_write	= iter_file_splice_write,
1681	.open		= fscrypt_file_open,
1682#ifdef CONFIG_COMPAT
1683	.compat_ioctl   = ubifs_compat_ioctl,
1684#endif
1685};