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_read_folio()', e.g., the
  35 * read-ahead path does not lock it ("sys_read -> generic_file_aio_read ->
  36 * ondemand_readahead -> read_folio"). 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 (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, UBIFS_MAX_DATA_NODE_SZ);
  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 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{
 220	struct inode *inode = mapping->host;
 221	struct ubifs_info *c = inode->i_sb->s_fs_info;
 222	pgoff_t index = pos >> PAGE_SHIFT;
 223	struct ubifs_budget_req req = { .new_page = 1 };
 224	int err, appending = !!(pos + len > inode->i_size);
 225	struct page *page;
 226
 227	dbg_gen("ino %lu, pos %llu, len %u, i_size %lld",
 228		inode->i_ino, pos, len, inode->i_size);
 229
 230	/*
 231	 * At the slow path we have to budget before locking the page, because
 232	 * budgeting may force write-back, which would wait on locked pages and
 233	 * deadlock if we had the page locked. At this point we do not know
 234	 * anything about the page, so assume that this is a new page which is
 235	 * written to a hole. This corresponds to largest budget. Later the
 236	 * budget will be amended if this is not true.
 237	 */
 238	if (appending)
 239		/* We are appending data, budget for inode change */
 240		req.dirtied_ino = 1;
 241
 242	err = ubifs_budget_space(c, &req);
 243	if (unlikely(err))
 244		return err;
 245
 246	page = grab_cache_page_write_begin(mapping, index);
 247	if (unlikely(!page)) {
 
 248		ubifs_release_budget(c, &req);
 249		return -ENOMEM;
 250	}
 251
 252	if (!PageUptodate(page)) {
 253		if (!(pos & ~PAGE_MASK) && len == PAGE_SIZE)
 254			SetPageChecked(page);
 255		else {
 256			err = do_readpage(page);
 257			if (err) {
 258				unlock_page(page);
 259				put_page(page);
 260				ubifs_release_budget(c, &req);
 261				return err;
 262			}
 263		}
 264
 265		SetPageUptodate(page);
 266		ClearPageError(page);
 267	}
 268
 269	if (PagePrivate(page))
 270		/*
 271		 * The page is dirty, which means it was budgeted twice:
 272		 *   o first time the budget was allocated by the task which
 273		 *     made the page dirty and set the PG_private flag;
 274		 *   o and then we budgeted for it for the second time at the
 275		 *     very beginning of this function.
 276		 *
 277		 * So what we have to do is to release the page budget we
 278		 * allocated.
 279		 */
 280		release_new_page_budget(c);
 281	else if (!PageChecked(page))
 282		/*
 283		 * We are changing a page which already exists on the media.
 284		 * This means that changing the page does not make the amount
 285		 * of indexing information larger, and this part of the budget
 286		 * which we have already acquired may be released.
 287		 */
 288		ubifs_convert_page_budget(c);
 289
 290	if (appending) {
 291		struct ubifs_inode *ui = ubifs_inode(inode);
 292
 293		/*
 294		 * 'ubifs_write_end()' is optimized from the fast-path part of
 295		 * 'ubifs_write_begin()' and expects the @ui_mutex to be locked
 296		 * if data is appended.
 297		 */
 298		mutex_lock(&ui->ui_mutex);
 299		if (ui->dirty)
 300			/*
 301			 * The inode is dirty already, so we may free the
 302			 * budget we allocated.
 303			 */
 304			ubifs_release_dirty_inode_budget(c, ui);
 305	}
 306
 307	*pagep = page;
 308	return 0;
 309}
 310
 311/**
 312 * allocate_budget - allocate budget for 'ubifs_write_begin()'.
 313 * @c: UBIFS file-system description object
 314 * @page: page to allocate budget for
 315 * @ui: UBIFS inode object the page belongs to
 316 * @appending: non-zero if the page is appended
 317 *
 318 * This is a helper function for 'ubifs_write_begin()' which allocates budget
 319 * for the operation. The budget is allocated differently depending on whether
 320 * this is appending, whether the page is dirty or not, and so on. This
 321 * function leaves the @ui->ui_mutex locked in case of appending.
 322 *
 323 * Returns: %0 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,
 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 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);
 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);
 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		attach_page_private(page, (void *)1);
 
 
 
 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 * 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 (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 * 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 = pagecache_get_page(mapping, page_offset,
 790				 FGP_LOCK|FGP_ACCESSED|FGP_CREAT|FGP_NOWAIT,
 791				 ra_gfp_mask);
 792		if (!page)
 793			break;
 794		if (!PageUptodate(page))
 795			err = populate_page(c, page, bu, &n);
 796		unlock_page(page);
 797		put_page(page);
 798		if (err)
 799			break;
 800	}
 801
 802	ui->last_page_read = offset + page_idx - 1;
 803
 804out_free:
 805	if (allocate)
 806		kfree(bu->buf);
 807	return ret;
 808
 809out_warn:
 810	ubifs_warn(c, "ignoring error %d and skipping bulk-read", err);
 811	goto out_free;
 812
 813out_bu_off:
 814	ui->read_in_a_row = ui->bulk_read = 0;
 815	goto out_free;
 816}
 817
 818/**
 819 * ubifs_bulk_read - determine whether to bulk-read and, if so, do it.
 820 * @page: page from which to start bulk-read.
 821 *
 822 * Some flash media are capable of reading sequentially at faster rates. UBIFS
 823 * bulk-read facility is designed to take advantage of that, by reading in one
 824 * go consecutive data nodes that are also located consecutively in the same
 825 * LEB.
 826 *
 827 * Returns: %1 if a bulk-read is done and %0 otherwise.
 828 */
 829static int ubifs_bulk_read(struct page *page)
 830{
 831	struct inode *inode = page->mapping->host;
 832	struct ubifs_info *c = inode->i_sb->s_fs_info;
 833	struct ubifs_inode *ui = ubifs_inode(inode);
 834	pgoff_t index = page->index, last_page_read = ui->last_page_read;
 835	struct bu_info *bu;
 836	int err = 0, allocated = 0;
 837
 838	ui->last_page_read = index;
 839	if (!c->bulk_read)
 840		return 0;
 841
 842	/*
 843	 * Bulk-read is protected by @ui->ui_mutex, but it is an optimization,
 844	 * so don't bother if we cannot lock the mutex.
 845	 */
 846	if (!mutex_trylock(&ui->ui_mutex))
 847		return 0;
 848
 849	if (index != last_page_read + 1) {
 850		/* Turn off bulk-read if we stop reading sequentially */
 851		ui->read_in_a_row = 1;
 852		if (ui->bulk_read)
 853			ui->bulk_read = 0;
 854		goto out_unlock;
 855	}
 856
 857	if (!ui->bulk_read) {
 858		ui->read_in_a_row += 1;
 859		if (ui->read_in_a_row < 3)
 860			goto out_unlock;
 861		/* Three reads in a row, so switch on bulk-read */
 862		ui->bulk_read = 1;
 863	}
 864
 865	/*
 866	 * If possible, try to use pre-allocated bulk-read information, which
 867	 * is protected by @c->bu_mutex.
 868	 */
 869	if (mutex_trylock(&c->bu_mutex))
 870		bu = &c->bu;
 871	else {
 872		bu = kmalloc(sizeof(struct bu_info), GFP_NOFS | __GFP_NOWARN);
 873		if (!bu)
 874			goto out_unlock;
 875
 876		bu->buf = NULL;
 877		allocated = 1;
 878	}
 879
 880	bu->buf_len = c->max_bu_buf_len;
 881	data_key_init(c, &bu->key, inode->i_ino,
 882		      page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT);
 883	err = ubifs_do_bulk_read(c, bu, page);
 884
 885	if (!allocated)
 886		mutex_unlock(&c->bu_mutex);
 887	else
 888		kfree(bu);
 889
 890out_unlock:
 891	mutex_unlock(&ui->ui_mutex);
 892	return err;
 893}
 894
 895static int ubifs_read_folio(struct file *file, struct folio *folio)
 896{
 897	struct page *page = &folio->page;
 898
 899	if (ubifs_bulk_read(page))
 900		return 0;
 901	do_readpage(page);
 902	folio_unlock(folio);
 903	return 0;
 904}
 905
 906static int do_writepage(struct page *page, int len)
 907{
 908	int err = 0, i, blen;
 909	unsigned int block;
 910	void *addr;
 
 911	union ubifs_key key;
 912	struct inode *inode = page->mapping->host;
 913	struct ubifs_info *c = inode->i_sb->s_fs_info;
 914
 915#ifdef UBIFS_DEBUG
 916	struct ubifs_inode *ui = ubifs_inode(inode);
 917	spin_lock(&ui->ui_lock);
 918	ubifs_assert(c, page->index <= ui->synced_i_size >> PAGE_SHIFT);
 919	spin_unlock(&ui->ui_lock);
 920#endif
 921
 922	/* Update radix tree tags */
 923	set_page_writeback(page);
 924
 925	addr = kmap(page);
 926	block = page->index << UBIFS_BLOCKS_PER_PAGE_SHIFT;
 927	i = 0;
 928	while (len) {
 929		blen = min_t(int, len, UBIFS_BLOCK_SIZE);
 930		data_key_init(c, &key, inode->i_ino, block);
 931		err = ubifs_jnl_write_data(c, inode, &key, addr, blen);
 932		if (err)
 933			break;
 934		if (++i >= UBIFS_BLOCKS_PER_PAGE)
 
 935			break;
 936		block += 1;
 937		addr += blen;
 938		len -= blen;
 
 
 
 
 939	}
 
 940	if (err) {
 941		SetPageError(page);
 942		ubifs_err(c, "cannot write page %lu of inode %lu, error %d",
 943			  page->index, inode->i_ino, err);
 944		ubifs_ro_mode(c, err);
 945	}
 946
 947	ubifs_assert(c, PagePrivate(page));
 948	if (PageChecked(page))
 949		release_new_page_budget(c);
 950	else
 951		release_existing_page_budget(c);
 952
 953	atomic_long_dec(&c->dirty_pg_cnt);
 954	detach_page_private(page);
 955	ClearPageChecked(page);
 956
 957	kunmap(page);
 958	unlock_page(page);
 959	end_page_writeback(page);
 960	return err;
 961}
 962
 963/*
 964 * When writing-back dirty inodes, VFS first writes-back pages belonging to the
 965 * inode, then the inode itself. For UBIFS this may cause a problem. Consider a
 966 * situation when a we have an inode with size 0, then a megabyte of data is
 967 * appended to the inode, then write-back starts and flushes some amount of the
 968 * dirty pages, the journal becomes full, commit happens and finishes, and then
 969 * an unclean reboot happens. When the file system is mounted next time, the
 970 * inode size would still be 0, but there would be many pages which are beyond
 971 * the inode size, they would be indexed and consume flash space. Because the
 972 * journal has been committed, the replay would not be able to detect this
 973 * situation and correct the inode size. This means UBIFS would have to scan
 974 * whole index and correct all inode sizes, which is long an unacceptable.
 975 *
 976 * To prevent situations like this, UBIFS writes pages back only if they are
 977 * within the last synchronized inode size, i.e. the size which has been
 978 * written to the flash media last time. Otherwise, UBIFS forces inode
 979 * write-back, thus making sure the on-flash inode contains current inode size,
 980 * and then keeps writing pages back.
 981 *
 982 * Some locking issues explanation. 'ubifs_writepage()' first is called with
 983 * the page locked, and it locks @ui_mutex. However, write-back does take inode
 984 * @i_mutex, which means other VFS operations may be run on this inode at the
 985 * same time. And the problematic one is truncation to smaller size, from where
 986 * we have to call 'truncate_setsize()', which first changes @inode->i_size,
 987 * then drops the truncated pages. And while dropping the pages, it takes the
 988 * page lock. This means that 'do_truncation()' cannot call 'truncate_setsize()'
 989 * with @ui_mutex locked, because it would deadlock with 'ubifs_writepage()'.
 990 * This means that @inode->i_size is changed while @ui_mutex is unlocked.
 991 *
 992 * XXX(truncate): with the new truncate sequence this is not true anymore,
 993 * and the calls to truncate_setsize can be move around freely.  They should
 994 * be moved to the very end of the truncate sequence.
 995 *
 996 * But in 'ubifs_writepage()' we have to guarantee that we do not write beyond
 997 * inode size. How do we do this if @inode->i_size may became smaller while we
 998 * are in the middle of 'ubifs_writepage()'? The UBIFS solution is the
 999 * @ui->ui_isize "shadow" field which UBIFS uses instead of @inode->i_size
1000 * internally and updates it under @ui_mutex.
1001 *
1002 * Q: why we do not worry that if we race with truncation, we may end up with a
1003 * situation when the inode is truncated while we are in the middle of
1004 * 'do_writepage()', so we do write beyond inode size?
1005 * A: If we are in the middle of 'do_writepage()', truncation would be locked
1006 * on the page lock and it would not write the truncated inode node to the
1007 * journal before we have finished.
1008 */
1009static int ubifs_writepage(struct page *page, struct writeback_control *wbc)
 
1010{
1011	struct inode *inode = page->mapping->host;
1012	struct ubifs_info *c = inode->i_sb->s_fs_info;
1013	struct ubifs_inode *ui = ubifs_inode(inode);
1014	loff_t i_size =  i_size_read(inode), synced_i_size;
1015	pgoff_t end_index = i_size >> PAGE_SHIFT;
1016	int err, len = i_size & (PAGE_SIZE - 1);
1017	void *kaddr;
1018
1019	dbg_gen("ino %lu, pg %lu, pg flags %#lx",
1020		inode->i_ino, page->index, page->flags);
1021	ubifs_assert(c, PagePrivate(page));
1022
1023	/* Is the page fully outside @i_size? (truncate in progress) */
1024	if (page->index > end_index || (page->index == end_index && !len)) {
1025		err = 0;
1026		goto out_unlock;
1027	}
1028
1029	spin_lock(&ui->ui_lock);
1030	synced_i_size = ui->synced_i_size;
1031	spin_unlock(&ui->ui_lock);
1032
1033	/* Is the page fully inside @i_size? */
1034	if (page->index < end_index) {
1035		if (page->index >= synced_i_size >> PAGE_SHIFT) {
1036			err = inode->i_sb->s_op->write_inode(inode, NULL);
1037			if (err)
1038				goto out_redirty;
1039			/*
1040			 * The inode has been written, but the write-buffer has
1041			 * not been synchronized, so in case of an unclean
1042			 * reboot we may end up with some pages beyond inode
1043			 * size, but they would be in the journal (because
1044			 * commit flushes write buffers) and recovery would deal
1045			 * with this.
1046			 */
1047		}
1048		return do_writepage(page, PAGE_SIZE);
1049	}
1050
1051	/*
1052	 * The page straddles @i_size. It must be zeroed out on each and every
1053	 * writepage invocation because it may be mmapped. "A file is mapped
1054	 * in multiples of the page size. For a file that is not a multiple of
1055	 * the page size, the remaining memory is zeroed when mapped, and
1056	 * writes to that region are not written out to the file."
1057	 */
1058	kaddr = kmap_atomic(page);
1059	memset(kaddr + len, 0, PAGE_SIZE - len);
1060	flush_dcache_page(page);
1061	kunmap_atomic(kaddr);
1062
1063	if (i_size > synced_i_size) {
1064		err = inode->i_sb->s_op->write_inode(inode, NULL);
1065		if (err)
1066			goto out_redirty;
1067	}
1068
1069	return do_writepage(page, len);
1070out_redirty:
1071	/*
1072	 * redirty_page_for_writepage() won't call ubifs_dirty_inode() because
1073	 * it passes I_DIRTY_PAGES flag while calling __mark_inode_dirty(), so
1074	 * there is no need to do space budget for dirty inode.
1075	 */
1076	redirty_page_for_writepage(wbc, page);
1077out_unlock:
1078	unlock_page(page);
1079	return err;
1080}
1081
 
 
 
 
 
 
1082/**
1083 * do_attr_changes - change inode attributes.
1084 * @inode: inode to change attributes for
1085 * @attr: describes attributes to change
1086 */
1087static void do_attr_changes(struct inode *inode, const struct iattr *attr)
1088{
1089	if (attr->ia_valid & ATTR_UID)
1090		inode->i_uid = attr->ia_uid;
1091	if (attr->ia_valid & ATTR_GID)
1092		inode->i_gid = attr->ia_gid;
1093	if (attr->ia_valid & ATTR_ATIME)
1094		inode_set_atime_to_ts(inode, attr->ia_atime);
1095	if (attr->ia_valid & ATTR_MTIME)
1096		inode_set_mtime_to_ts(inode, attr->ia_mtime);
1097	if (attr->ia_valid & ATTR_CTIME)
1098		inode_set_ctime_to_ts(inode, attr->ia_ctime);
1099	if (attr->ia_valid & ATTR_MODE) {
1100		umode_t mode = attr->ia_mode;
1101
1102		if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
1103			mode &= ~S_ISGID;
1104		inode->i_mode = mode;
1105	}
1106}
1107
1108/**
1109 * do_truncation - truncate an inode.
1110 * @c: UBIFS file-system description object
1111 * @inode: inode to truncate
1112 * @attr: inode attribute changes description
1113 *
1114 * This function implements VFS '->setattr()' call when the inode is truncated
1115 * to a smaller size.
1116 *
1117 * Returns: %0 in case of success and a negative error code
1118 * in case of failure.
1119 */
1120static int do_truncation(struct ubifs_info *c, struct inode *inode,
1121			 const struct iattr *attr)
1122{
1123	int err;
1124	struct ubifs_budget_req req;
1125	loff_t old_size = inode->i_size, new_size = attr->ia_size;
1126	int offset = new_size & (UBIFS_BLOCK_SIZE - 1), budgeted = 1;
1127	struct ubifs_inode *ui = ubifs_inode(inode);
1128
1129	dbg_gen("ino %lu, size %lld -> %lld", inode->i_ino, old_size, new_size);
1130	memset(&req, 0, sizeof(struct ubifs_budget_req));
1131
1132	/*
1133	 * If this is truncation to a smaller size, and we do not truncate on a
1134	 * block boundary, budget for changing one data block, because the last
1135	 * block will be re-written.
1136	 */
1137	if (new_size & (UBIFS_BLOCK_SIZE - 1))
1138		req.dirtied_page = 1;
1139
1140	req.dirtied_ino = 1;
1141	/* A funny way to budget for truncation node */
1142	req.dirtied_ino_d = UBIFS_TRUN_NODE_SZ;
1143	err = ubifs_budget_space(c, &req);
1144	if (err) {
1145		/*
1146		 * Treat truncations to zero as deletion and always allow them,
1147		 * just like we do for '->unlink()'.
1148		 */
1149		if (new_size || err != -ENOSPC)
1150			return err;
1151		budgeted = 0;
1152	}
1153
1154	truncate_setsize(inode, new_size);
1155
1156	if (offset) {
1157		pgoff_t index = new_size >> PAGE_SHIFT;
1158		struct page *page;
1159
1160		page = find_lock_page(inode->i_mapping, index);
1161		if (page) {
1162			if (PageDirty(page)) {
1163				/*
1164				 * 'ubifs_jnl_truncate()' will try to truncate
1165				 * the last data node, but it contains
1166				 * out-of-date data because the page is dirty.
1167				 * Write the page now, so that
1168				 * 'ubifs_jnl_truncate()' will see an already
1169				 * truncated (and up to date) data node.
1170				 */
1171				ubifs_assert(c, PagePrivate(page));
1172
1173				clear_page_dirty_for_io(page);
1174				if (UBIFS_BLOCKS_PER_PAGE_SHIFT)
1175					offset = new_size &
1176						 (PAGE_SIZE - 1);
1177				err = do_writepage(page, offset);
1178				put_page(page);
1179				if (err)
1180					goto out_budg;
1181				/*
1182				 * We could now tell 'ubifs_jnl_truncate()' not
1183				 * to read the last block.
1184				 */
1185			} else {
1186				/*
1187				 * We could 'kmap()' the page and pass the data
1188				 * to 'ubifs_jnl_truncate()' to save it from
1189				 * having to read it.
1190				 */
1191				unlock_page(page);
1192				put_page(page);
1193			}
1194		}
1195	}
1196
1197	mutex_lock(&ui->ui_mutex);
1198	ui->ui_size = inode->i_size;
1199	/* Truncation changes inode [mc]time */
1200	inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1201	/* Other attributes may be changed at the same time as well */
1202	do_attr_changes(inode, attr);
1203	err = ubifs_jnl_truncate(c, inode, old_size, new_size);
1204	mutex_unlock(&ui->ui_mutex);
1205
1206out_budg:
1207	if (budgeted)
1208		ubifs_release_budget(c, &req);
1209	else {
1210		c->bi.nospace = c->bi.nospace_rp = 0;
1211		smp_wmb();
1212	}
1213	return err;
1214}
1215
1216/**
1217 * do_setattr - change inode attributes.
1218 * @c: UBIFS file-system description object
1219 * @inode: inode to change attributes for
1220 * @attr: inode attribute changes description
1221 *
1222 * This function implements VFS '->setattr()' call for all cases except
1223 * truncations to smaller size.
1224 *
1225 * Returns: %0 in case of success and a negative
1226 * error code in case of failure.
1227 */
1228static int do_setattr(struct ubifs_info *c, struct inode *inode,
1229		      const struct iattr *attr)
1230{
1231	int err, release;
1232	loff_t new_size = attr->ia_size;
1233	struct ubifs_inode *ui = ubifs_inode(inode);
1234	struct ubifs_budget_req req = { .dirtied_ino = 1,
1235				.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1236
1237	err = ubifs_budget_space(c, &req);
1238	if (err)
1239		return err;
1240
1241	if (attr->ia_valid & ATTR_SIZE) {
1242		dbg_gen("size %lld -> %lld", inode->i_size, new_size);
1243		truncate_setsize(inode, new_size);
1244	}
1245
1246	mutex_lock(&ui->ui_mutex);
1247	if (attr->ia_valid & ATTR_SIZE) {
1248		/* Truncation changes inode [mc]time */
1249		inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1250		/* 'truncate_setsize()' changed @i_size, update @ui_size */
1251		ui->ui_size = inode->i_size;
1252	}
1253
1254	do_attr_changes(inode, attr);
1255
1256	release = ui->dirty;
1257	if (attr->ia_valid & ATTR_SIZE)
1258		/*
1259		 * Inode length changed, so we have to make sure
1260		 * @I_DIRTY_DATASYNC is set.
1261		 */
1262		 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1263	else
1264		mark_inode_dirty_sync(inode);
1265	mutex_unlock(&ui->ui_mutex);
1266
1267	if (release)
1268		ubifs_release_budget(c, &req);
1269	if (IS_SYNC(inode))
1270		err = inode->i_sb->s_op->write_inode(inode, NULL);
1271	return err;
1272}
1273
1274int ubifs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
1275		  struct iattr *attr)
1276{
1277	int err;
1278	struct inode *inode = d_inode(dentry);
1279	struct ubifs_info *c = inode->i_sb->s_fs_info;
1280
1281	dbg_gen("ino %lu, mode %#x, ia_valid %#x",
1282		inode->i_ino, inode->i_mode, attr->ia_valid);
1283	err = setattr_prepare(&nop_mnt_idmap, dentry, attr);
1284	if (err)
1285		return err;
1286
1287	err = dbg_check_synced_i_size(c, inode);
1288	if (err)
1289		return err;
1290
1291	err = fscrypt_prepare_setattr(dentry, attr);
1292	if (err)
1293		return err;
1294
1295	if ((attr->ia_valid & ATTR_SIZE) && attr->ia_size < inode->i_size)
1296		/* Truncation to a smaller size */
1297		err = do_truncation(c, inode, attr);
1298	else
1299		err = do_setattr(c, inode, attr);
1300
1301	return err;
1302}
1303
1304static void ubifs_invalidate_folio(struct folio *folio, size_t offset,
1305				 size_t length)
1306{
1307	struct inode *inode = folio->mapping->host;
1308	struct ubifs_info *c = inode->i_sb->s_fs_info;
1309
1310	ubifs_assert(c, folio_test_private(folio));
1311	if (offset || length < folio_size(folio))
1312		/* Partial folio remains dirty */
1313		return;
1314
1315	if (folio_test_checked(folio))
1316		release_new_page_budget(c);
1317	else
1318		release_existing_page_budget(c);
1319
1320	atomic_long_dec(&c->dirty_pg_cnt);
1321	folio_detach_private(folio);
1322	folio_clear_checked(folio);
1323}
1324
1325int ubifs_fsync(struct file *file, loff_t start, loff_t end, int datasync)
1326{
1327	struct inode *inode = file->f_mapping->host;
1328	struct ubifs_info *c = inode->i_sb->s_fs_info;
1329	int err;
1330
1331	dbg_gen("syncing inode %lu", inode->i_ino);
1332
1333	if (c->ro_mount)
1334		/*
1335		 * For some really strange reasons VFS does not filter out
1336		 * 'fsync()' for R/O mounted file-systems as per 2.6.39.
1337		 */
1338		return 0;
1339
1340	err = file_write_and_wait_range(file, start, end);
1341	if (err)
1342		return err;
1343	inode_lock(inode);
1344
1345	/* Synchronize the inode unless this is a 'datasync()' call. */
1346	if (!datasync || (inode->i_state & I_DIRTY_DATASYNC)) {
1347		err = inode->i_sb->s_op->write_inode(inode, NULL);
1348		if (err)
1349			goto out;
1350	}
1351
1352	/*
1353	 * Nodes related to this inode may still sit in a write-buffer. Flush
1354	 * them.
1355	 */
1356	err = ubifs_sync_wbufs_by_inode(c, inode);
1357out:
1358	inode_unlock(inode);
1359	return err;
1360}
1361
1362/**
1363 * mctime_update_needed - check if mtime or ctime update is needed.
1364 * @inode: the inode to do the check for
1365 * @now: current time
1366 *
1367 * This helper function checks if the inode mtime/ctime should be updated or
1368 * not. If current values of the time-stamps are within the UBIFS inode time
1369 * granularity, they are not updated. This is an optimization.
1370 *
1371 * Returns: %1 if time update is needed, %0 if not
1372 */
1373static inline int mctime_update_needed(const struct inode *inode,
1374				       const struct timespec64 *now)
1375{
1376	struct timespec64 ctime = inode_get_ctime(inode);
1377	struct timespec64 mtime = inode_get_mtime(inode);
1378
1379	if (!timespec64_equal(&mtime, now) || !timespec64_equal(&ctime, now))
1380		return 1;
1381	return 0;
1382}
1383
1384/**
1385 * ubifs_update_time - update time of inode.
1386 * @inode: inode to update
1387 * @flags: time updating control flag determines updating
1388 *	    which time fields of @inode
1389 *
1390 * This function updates time of the inode.
1391 *
1392 * Returns: %0 for success or a negative error code otherwise.
1393 */
1394int ubifs_update_time(struct inode *inode, int flags)
1395{
1396	struct ubifs_inode *ui = ubifs_inode(inode);
1397	struct ubifs_info *c = inode->i_sb->s_fs_info;
1398	struct ubifs_budget_req req = { .dirtied_ino = 1,
1399			.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1400	int err, release;
1401
1402	if (!IS_ENABLED(CONFIG_UBIFS_ATIME_SUPPORT)) {
1403		generic_update_time(inode, flags);
1404		return 0;
1405	}
1406
1407	err = ubifs_budget_space(c, &req);
1408	if (err)
1409		return err;
1410
1411	mutex_lock(&ui->ui_mutex);
1412	inode_update_timestamps(inode, flags);
1413	release = ui->dirty;
1414	__mark_inode_dirty(inode, I_DIRTY_SYNC);
1415	mutex_unlock(&ui->ui_mutex);
1416	if (release)
1417		ubifs_release_budget(c, &req);
1418	return 0;
1419}
1420
1421/**
1422 * update_mctime - update mtime and ctime of an inode.
1423 * @inode: inode to update
1424 *
1425 * This function updates mtime and ctime of the inode if it is not equivalent to
1426 * current time.
1427 *
1428 * Returns: %0 in case of success and a negative error code in
1429 * case of failure.
1430 */
1431static int update_mctime(struct inode *inode)
1432{
1433	struct timespec64 now = current_time(inode);
1434	struct ubifs_inode *ui = ubifs_inode(inode);
1435	struct ubifs_info *c = inode->i_sb->s_fs_info;
1436
1437	if (mctime_update_needed(inode, &now)) {
1438		int err, release;
1439		struct ubifs_budget_req req = { .dirtied_ino = 1,
1440				.dirtied_ino_d = ALIGN(ui->data_len, 8) };
1441
1442		err = ubifs_budget_space(c, &req);
1443		if (err)
1444			return err;
1445
1446		mutex_lock(&ui->ui_mutex);
1447		inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1448		release = ui->dirty;
1449		mark_inode_dirty_sync(inode);
1450		mutex_unlock(&ui->ui_mutex);
1451		if (release)
1452			ubifs_release_budget(c, &req);
1453	}
1454
1455	return 0;
1456}
1457
1458static ssize_t ubifs_write_iter(struct kiocb *iocb, struct iov_iter *from)
1459{
1460	int err = update_mctime(file_inode(iocb->ki_filp));
1461	if (err)
1462		return err;
1463
1464	return generic_file_write_iter(iocb, from);
1465}
1466
1467static bool ubifs_dirty_folio(struct address_space *mapping,
1468		struct folio *folio)
1469{
1470	bool ret;
1471	struct ubifs_info *c = mapping->host->i_sb->s_fs_info;
1472
1473	ret = filemap_dirty_folio(mapping, folio);
1474	/*
1475	 * An attempt to dirty a page without budgeting for it - should not
1476	 * happen.
1477	 */
1478	ubifs_assert(c, ret == false);
1479	return ret;
1480}
1481
1482static bool ubifs_release_folio(struct folio *folio, gfp_t unused_gfp_flags)
1483{
1484	struct inode *inode = folio->mapping->host;
1485	struct ubifs_info *c = inode->i_sb->s_fs_info;
1486
1487	if (folio_test_writeback(folio))
1488		return false;
1489
1490	/*
1491	 * Page is private but not dirty, weird? There is one condition
1492	 * making it happened. ubifs_writepage skipped the page because
1493	 * page index beyonds isize (for example. truncated by other
1494	 * process named A), then the page is invalidated by fadvise64
1495	 * syscall before being truncated by process A.
1496	 */
1497	ubifs_assert(c, folio_test_private(folio));
1498	if (folio_test_checked(folio))
1499		release_new_page_budget(c);
1500	else
1501		release_existing_page_budget(c);
1502
1503	atomic_long_dec(&c->dirty_pg_cnt);
1504	folio_detach_private(folio);
1505	folio_clear_checked(folio);
1506	return true;
1507}
1508
1509/*
1510 * mmap()d file has taken write protection fault and is being made writable.
1511 * UBIFS must ensure page is budgeted for.
1512 */
1513static vm_fault_t ubifs_vm_page_mkwrite(struct vm_fault *vmf)
1514{
1515	struct page *page = vmf->page;
1516	struct inode *inode = file_inode(vmf->vma->vm_file);
1517	struct ubifs_info *c = inode->i_sb->s_fs_info;
1518	struct timespec64 now = current_time(inode);
1519	struct ubifs_budget_req req = { .new_page = 1 };
1520	int err, update_time;
1521
1522	dbg_gen("ino %lu, pg %lu, i_size %lld",	inode->i_ino, page->index,
1523		i_size_read(inode));
1524	ubifs_assert(c, !c->ro_media && !c->ro_mount);
1525
1526	if (unlikely(c->ro_error))
1527		return VM_FAULT_SIGBUS; /* -EROFS */
1528
1529	/*
1530	 * We have not locked @page so far so we may budget for changing the
1531	 * page. Note, we cannot do this after we locked the page, because
1532	 * budgeting may cause write-back which would cause deadlock.
1533	 *
1534	 * At the moment we do not know whether the page is dirty or not, so we
1535	 * assume that it is not and budget for a new page. We could look at
1536	 * the @PG_private flag and figure this out, but we may race with write
1537	 * back and the page state may change by the time we lock it, so this
1538	 * would need additional care. We do not bother with this at the
1539	 * moment, although it might be good idea to do. Instead, we allocate
1540	 * budget for a new page and amend it later on if the page was in fact
1541	 * dirty.
1542	 *
1543	 * The budgeting-related logic of this function is similar to what we
1544	 * do in 'ubifs_write_begin()' and 'ubifs_write_end()'. Glance there
1545	 * for more comments.
1546	 */
1547	update_time = mctime_update_needed(inode, &now);
1548	if (update_time)
1549		/*
1550		 * We have to change inode time stamp which requires extra
1551		 * budgeting.
1552		 */
1553		req.dirtied_ino = 1;
1554
1555	err = ubifs_budget_space(c, &req);
1556	if (unlikely(err)) {
1557		if (err == -ENOSPC)
1558			ubifs_warn(c, "out of space for mmapped file (inode number %lu)",
1559				   inode->i_ino);
1560		return VM_FAULT_SIGBUS;
1561	}
1562
1563	lock_page(page);
1564	if (unlikely(page->mapping != inode->i_mapping ||
1565		     page_offset(page) > i_size_read(inode))) {
1566		/* Page got truncated out from underneath us */
1567		goto sigbus;
1568	}
1569
1570	if (PagePrivate(page))
1571		release_new_page_budget(c);
1572	else {
1573		if (!PageChecked(page))
1574			ubifs_convert_page_budget(c);
1575		attach_page_private(page, (void *)1);
1576		atomic_long_inc(&c->dirty_pg_cnt);
1577		__set_page_dirty_nobuffers(page);
1578	}
1579
1580	if (update_time) {
1581		int release;
1582		struct ubifs_inode *ui = ubifs_inode(inode);
1583
1584		mutex_lock(&ui->ui_mutex);
1585		inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
1586		release = ui->dirty;
1587		mark_inode_dirty_sync(inode);
1588		mutex_unlock(&ui->ui_mutex);
1589		if (release)
1590			ubifs_release_dirty_inode_budget(c, ui);
1591	}
1592
1593	wait_for_stable_page(page);
1594	return VM_FAULT_LOCKED;
1595
1596sigbus:
1597	unlock_page(page);
1598	ubifs_release_budget(c, &req);
1599	return VM_FAULT_SIGBUS;
1600}
1601
1602static const struct vm_operations_struct ubifs_file_vm_ops = {
1603	.fault        = filemap_fault,
1604	.map_pages = filemap_map_pages,
1605	.page_mkwrite = ubifs_vm_page_mkwrite,
1606};
1607
1608static int ubifs_file_mmap(struct file *file, struct vm_area_struct *vma)
1609{
1610	int err;
1611
1612	err = generic_file_mmap(file, vma);
1613	if (err)
1614		return err;
1615	vma->vm_ops = &ubifs_file_vm_ops;
1616
1617	if (IS_ENABLED(CONFIG_UBIFS_ATIME_SUPPORT))
1618		file_accessed(file);
1619
1620	return 0;
1621}
1622
1623static const char *ubifs_get_link(struct dentry *dentry,
1624					    struct inode *inode,
1625					    struct delayed_call *done)
1626{
1627	struct ubifs_inode *ui = ubifs_inode(inode);
1628
1629	if (!IS_ENCRYPTED(inode))
1630		return ui->data;
1631
1632	if (!dentry)
1633		return ERR_PTR(-ECHILD);
1634
1635	return fscrypt_get_symlink(inode, ui->data, ui->data_len, done);
1636}
1637
1638static int ubifs_symlink_getattr(struct mnt_idmap *idmap,
1639				 const struct path *path, struct kstat *stat,
1640				 u32 request_mask, unsigned int query_flags)
1641{
1642	ubifs_getattr(idmap, path, stat, request_mask, query_flags);
1643
1644	if (IS_ENCRYPTED(d_inode(path->dentry)))
1645		return fscrypt_symlink_getattr(path, stat);
1646	return 0;
1647}
1648
1649const struct address_space_operations ubifs_file_address_operations = {
1650	.read_folio     = ubifs_read_folio,
1651	.writepage      = ubifs_writepage,
1652	.write_begin    = ubifs_write_begin,
1653	.write_end      = ubifs_write_end,
1654	.invalidate_folio = ubifs_invalidate_folio,
1655	.dirty_folio	= ubifs_dirty_folio,
1656	.migrate_folio	= filemap_migrate_folio,
1657	.release_folio	= ubifs_release_folio,
1658};
1659
1660const struct inode_operations ubifs_file_inode_operations = {
1661	.setattr     = ubifs_setattr,
1662	.getattr     = ubifs_getattr,
1663	.listxattr   = ubifs_listxattr,
1664	.update_time = ubifs_update_time,
1665	.fileattr_get = ubifs_fileattr_get,
1666	.fileattr_set = ubifs_fileattr_set,
1667};
1668
1669const struct inode_operations ubifs_symlink_inode_operations = {
1670	.get_link    = ubifs_get_link,
1671	.setattr     = ubifs_setattr,
1672	.getattr     = ubifs_symlink_getattr,
1673	.listxattr   = ubifs_listxattr,
1674	.update_time = ubifs_update_time,
1675};
1676
1677const struct file_operations ubifs_file_operations = {
1678	.llseek         = generic_file_llseek,
1679	.read_iter      = generic_file_read_iter,
1680	.write_iter     = ubifs_write_iter,
1681	.mmap           = ubifs_file_mmap,
1682	.fsync          = ubifs_fsync,
1683	.unlocked_ioctl = ubifs_ioctl,
1684	.splice_read	= filemap_splice_read,
1685	.splice_write	= iter_file_splice_write,
1686	.open		= fscrypt_file_open,
1687#ifdef CONFIG_COMPAT
1688	.compat_ioctl   = ubifs_compat_ioctl,
1689#endif
1690};